iv introduction to the first issue editor-in-chief, prof. wen-hsiang hsieh on behalf of the editorial board, i am delighted to announce the publication of the inaugural issue of advances in technology innovation (aiti), published by the taiwan association of engineering and technology innovation (taeti). aiti is an international, multidiscipline, peer-reviewed scholarly journal, published quarterly for researchers, developers, technical managers, and educators in the field of engineering and technology innovation. the journal is intended as an outlet for scientists and academicians all over the world to share, promote, and discuss the newest finding, issues and developments with the community. aiti is now accepting submissions from perspective authors to be considered for publication and included in the future issues of the journal. articles of original research, reports, reviews, and commentaries are welcomed by aiti. for more information about the journal or the submission process, please visit the journal’s website. it is a great pleasure to include in the first issue of aiti, six papers written by leading researchers in the international community. i would like to thank all these authors for their wonderful papers, and we also believe that they will make this into a valuable source for researchers and practitioners. the efforts of many people have made this journal possible. first, special thanks must be given to the editorial board. they have helped to define the role of the journal, the review process, the format, and the content of the journal. next, i would like to acknowledge the reviewers for their timely and comprehensive reviews. finally, i must be grateful to the staffs of the editorial office, who have worked for so many months to compile this first issue and complete the website.  advances in technology innovation , vol. 1, no. 2, 2016, pp. 50 52 50 copyright © taeti tubular steel arch stabilized by textile membranes ondrej svoboda, josef machacek * czech technical university in prague, faculty of civil engineering, prague, czech republic. received 17 february 2016; received in revised form 01 april 2016; accept ed 06 april 2016 abstract tubular steel arch supporting textile me mbrane roofing is investigated experimentally and numerically. the stabilizat ion effects of the text ile me mbrane on in-p lane and out-of-plane behavior of the arch is of prima ry interest. first a model of a large me mbrane structure tested in laboratory is described. prestressed me mbranes of pvc coated polyester fabric fe rrari ® précontraint 702s were used as a currently standard and excellent material. the test arrangement, loading and resulting load/deflection values are presented. the supporting structure consisted of two steel a rch tubes, outer at edge of the me mbrane and inner supporting interior of the me mb rane roofing. the stability and strengt h behavior of the inner tube under both symmetrica l and asymmetrica l loading was monitored and is shown in some details. second the sofist ik software was employed to analyze the structural behavior in 3d, using geometrica lly nonlinear analysis with imperfections (gnia). the nu me rica l ana lysis , fe mesh sensitivity, the me mb rane prestressing and common boundary conditions are validated by test results. finally a para metrica l study concerning stability of mid a rch with various geometries in a me mb rane structure with several supporting arches is presented, with recommendations for a practical design. ke ywor ds : te xt ile me mbranes, prestressing, steel arch, arch stabilizat ion, gnia, tests 1. introduction design of steel structures cooperating with tsf (tensioned fabric structures) requires geometrica lly and materia lly non -linear analysis with imperfect ions (gmnia). the essential in such analysis is an appropriate input of the me mb rane materia l behavior; see [1], [2], [3], [4], etc. while me mb rane surface is e xc lusively tensioned, supporting steelwork is most often e xposed to a compression and/or bending. this type of loading, in co mbination with slender steel ele ments, results into stability proble ms . usually the me mb rane represents a spring support for the steel structure and the co mple x structure need to be designed using proper software package allowing integrated modelling, e.g. easy [5], sofistik [6], etc., wh ile a separated modelling of the me mb rane and steelwork is rather limited [7]. th is paper demonstrates significant stabilizing effects of me mb ranes to the respective supporting steelwork, based on numerical para metrica l studies validated by tests. 2. validation by tests the tested me mbrane structure supported by two steel tube arches is shown in fig. 1. the me mb rane is pvc coated polyester fabric ferra ri ® précontraint 702s (with bra king loadings in both warp and fill direct ions sult ≈ 56 kn/ m, wo rking loading smax = sult/5 ≈ 11.2 kn/ m and suitable prestressing up to pmax = smax/5 ≈ 2.24 kn/ m). princ ipal dimensions of the vertical inner tubular arch of ø 26.9x3.2 [mm] are lxh = 4500x1200 [mm], while outer tubular arch has inclination of 60° in respect to horizontal. fig. 1 the layout of the tested model * corresponding aut hor, email: machacek@cvut .fsv.cz advances in technology innovation , vol. 1, no. 2, 2016, pp. 50 52 51 copyright © taeti the shape and cut of the me mbranes resulted fro m formfinder software [ 8] and the me mb ranes were prestressed roughly with p ≈ 0.2 kn/ m. the investigation concerned e xclusively the inner steel a rch to find stabilizing effect of the me mbrane to its nonlinear behavior. first the inner arch alone (without fastening the me mbrane) was loaded and second, after the me mbrane assembly, the complete me mb rane structure. for loading calibrated pouches with steel pellets we re used and suspended from seven points of the arch (for the symmetrica l loading see fig. 2) or fro m four points in case of asymmetrical loading. fig. 2 sy mmetrical loading of the tested model during the tests the deflections and stresses were carefu lly monitored in 9 locations (with no. 4 at midspan). in this paper the symmetrica l test and deflections only are described due to a space limit . the deflections (vertical in fig. 3 and transverse in fig. 4) de monstrate that t he arch without me mbrane buckled out-of-plane at total loading of f0 = 5.5 kn, with vertica l deflection along all span down, while the test of the arch stabilized by the me mbrane was terminated under total load of fm = 8.3 kn, showing the enormous stabilizing effect of the membrane. fig. 3 symmetrical loading vertical deflections fig. 4 symmetrical loading – transverse deflections sofistik software [6] was used to perform gnia (using n-r iteration) both for the inner arch alone and the complete me mbrane structure, emp loying orthotropic model with 5 para meters according to [3]. various meshing of the me mb rane was analysed (square sizes of 25, 50, 100 and 200 [mm]) with diffe rences ≤ 0.2 % and optimu m size o f 50 mm was used in the ana lysis. the equilibriu m state and final unloaded shape of the me mbrane structure was found under initia l sofistik software ca lculations. both arch alone analysis and membrane with the two arches were performed and results compared with tests showing e xcellent agree ment. the results of gnia for the symmet rica l loading under various prestress of the me mbrane p [kn/ m] are shown in fig. 8 and with the prestress of 0.2 kn/ m (corresponding to test) justify use of the model for fo llo wing para metric studies. fig. 5 comparison of the test and gnia vertical deflections under various prestressing 3. parametrical studies in-plane and out-of-p lane stability of 132 central arches in the 5 arches assembly (see fig. 5), where the edge a rches were continuously transversely supported, have been studied under various geometries, loadings and me mbrane prestressing. more details and results are ready for publication, but due to limited space not shown here. advances in technology innovation , vol. 1, no. 2, 2016, pp. 50 52 52 copyright © taeti fig. 5 mid-arch with out-of-plane buckling 4. conclusions (1) the effect of te xt ile me mb ranes on both in-plane and out-of-plane supporting arch stability and strength is enormous. (2) gnia (by sofistik) proved to be adequate, provided the right value of the me mb rane prestressing is used. (3) large parametric studies of barrel me mb rane structures supported by a row of steel a rches show enormous increase of both in -plane and particularly out-of-plane buckling loads in comparison to the ones of an arch alone. provided the outer arches are transversely supported, the out-of plane buckling of the mid arches due to me mbrane support may always be neglected. acknowledgement this work was supported by the czech grant agency; grant gacr no. 105/13/25781s. references [1] s. kato, t. yoshino, and h. m ina mi, “formulat ion of constitutive equatio ns for fabric me mb ranes based on concept of fabric lattice model,” engineering structures, vol. 21, pp. 691-708, 1999. [2] p. gosling, “ basic philosophy and calling notice,” tensinet analysis & materia l working group, tensinews, vol. 13, pp. 12-15, 2007. [3] c. ga lliot and r. h. luchsinger, “a simp le model describing the non -linear b ia xia l tensile behaviour of pvc/coated polyester fabrics for use in finite e le ment analysis,” co mp. structures, vol. 90, pp. 438-447, 2009. [4] j. b. pargana and w. m . a. le itao, “a simp lified stress -strain model for coated plain-weave fabrics used in tensioned fabric structures,” engineering structures, vol. 84, pp. 439-450, 2015. [5] “technet gmbh berlin-stuttgart,” http://www.technet-gmbh.com, 2016. [6] “sofistik 2014,” http://www.sofistik.de/, 2015. [7] d. jermo ljev and j. macháček, “ imp le ment ation of non-meta llic me mbranes into steel supporting structures ,” proc. recent advances in mechanics and materials in design, ponta delgada, pp. 907-908, july 2015. [8] “formfinder software gmb h, wien,” http://www.formfinder.at/ ma in/software/, 2015.  advances in technology innovation , vol. 1, no. 1, 2016, pp. 13 15 13 copyright © taeti composite elements for biomimetic aerospace structures with progressive shape variation capabilities alessandro airoldi * , paolo bettini, matteo boiocchi, giuseppe sala department of aerospace science and technologies , politecnico di milano, italy received 01 february 2016; received in revised form 11 march 2016; accept ed 02 april 2016 abstract the paper presents some engineering solutions for the development of innovative aerodynamic surfaces with the capability of progressive shape variation. a brief introduction of the most significant issues related to the design of such morphing structures is provided. thereafter, two types of structural solutions are presented for the design of interna l co mp liant structures and fle xible e xterna l skins. the proposed solutions explo it the properties and the manufacturing techniques of long fibre reinforced plastic in order to fulfil the severe and contradictory require ments related to the trade-off between mo rphing performance and load carrying capabilities . keywor ds: morphing structures, composite structures, chiral topologies, corrugated laminates 1. introduction morphing structures have been intensively studied in the last decades in the aerospace field, with the objective of developing innovative, more fle xib le and effic ient method s to change the shape of aerodynamic surfaces. imitation of nature plays an important role in conceiving such type of structures, since organisms have solved the problems related to flight control and adaptation to different flight phases without the use of rigid moveable surfaces , which a re currently used in aircra ft. for instance, a fle xib le wing with the capability of shape variation can increase the curvature, when higher lift is required at lo w veloc ity, whereas, at high speed, curvature can be reduced to decrease drag (1). another concept, called “chira l sail” is proposed in (2) and is based on wing with a centra l morph ing part that increases its camber when angle of attitude is changed (fig. 1). th is can lead to noticeable advantages for the surfaces that generate the forces for the stabilization of a vehicle, like the tail e mpenn ages of aircra ft. indeed, these surfaces could be reduced with overall we ight saving and drag reduction. however, a lthough morphing is an appealing concept, there are critica l engineering issues to be solved for the development of such type of structures, which are hereby summarized in the following point: a) co mpliance of s tructures must be finely tuned to accomplish shape variations induced by aerodynamic loads (passive morph ing) or of actuators (active mo rphing). b) shape can vary but must retain the aerodynamic effic iency, without angular points , surface waviness, and anomalous modification of profiles . c) aerodynamic loads acting in morphing directions must be transmitted, so that morph ing structures must exhib it fle xib ility and strength at the same time (passive morph ing), or reacted by load bearing actuators (active morphing). d) stiffness and strength in non -morphing directions must be ma ximized to avoid the need of additional structural parts that would increase structural weight, thus reducing or eliminating the advantages of morphing concepts. fig. 1 variable ca mber wing with centra l morphing part * corresponding aut hor, email: alessandro.airoldi@polimi.it advances in technology innovation , vol. 1, no. 1, 2016, pp. 13 15 14 copyright © taeti in the following sections , two concepts will be presented to fulfill the afore mentioned severe and contradictory requirements. 2. composite chiral structures chiral topologies are special non centre-symmetric geometries that consist of circula r e le ments, called nodes, connected by straight liga ments (fig. 2-a). the ir defo rmation mechanis m leads to a transverse expansion, when a tensile load is applied, and a contraction, under the action of a co mpressive load (fig. 2-b). hence, if the chiral tessellation is considered as a meta-materia l, it turns out to be characterized by a negative poisson’s ratio (au xet ic behavior). such response avoids the development of localized displacements and weak points, and allo ws the achieve ment of controlled shape variations, as it is shown in fig. 2 -c, refe rred to deformation modes of the chira l sail depicted in fig. 1. for such reasons, chiral topologies were proposed to develop the internal structure of morph ing airfoils [ 1], but manufacturing of chiral honeycombs represent a critical proble m for application to rea l world structures. the process devised at politecnico di milano (4), allo ws the production of composite chira l ele ments by means of a procedure based on the bonding of composite units , which are produced in a prev ious step and then uniformly pressed together during bonding by means of elastomeric inserts (4,5). the resulting composite chira l e le ments can be produced with very thin liga ments and different types of composite materia ls, thus enhancing the design fle xib ility o f the concept. so me e xa mp les of manufactured chiral ele ments are provided in fig. 3. fig. 2 chira l topologies (a) defo rmation mechanis m (b) and deformation mode of a chiral airfoil (c) fig. 3 composite chiral honeycomb (a)and composite chiral rib for the chiral sail concept (b) 3. composite corrugated laminates the development of morphing surfaces always requires the development of a fle xib le skin to collect ae rodynamic forces . the require ments presented in the introduction are valid also for the skin, which has to undergo large recoverable strains, carrying and transmitting aerodynamic pressures to the internal structure. moreover, in traditional aeronautical constructions , skin also provides a valuable contribution to structural stiffness, so that optima l solution s should present adequate structural response in non-morphing directions. co mposite corrugated la minates have been proposed (6) to develop skins thanks to their inherent anisotropy that allo w high compliance and strains at failure in corrugation directions , and noticeable stiffness and strength in non-morphing directions (fig. 4-a). fig. 4 corrugated laminate (a) composite chiral rib for the chiral sail (b) unfortunately, the usage of co mposite corrugated la minates as morphing skin has a ma jor dra wback represented by the surface irregularities that increase ae rodynamic drag and reduce the generated lift (7). for such a reason, a solution was developed at politecnico d i m ilano (8) to integrate in the corrugated laminate an elastomeric layer, supported by honeycomb inserts (figs. 4-b and 4-c). the developed skin system was tested up to elongation of 20% (fig. a b advances in technology innovation , vol. 1, no. 1, 2016, pp. 13 15 15 copyright © taeti 4-d), proving that the elastomeric layer provides a smooth efficient aerodynamic surface without interfering with the mechanica l response of corrugated laminates. the composite corrugated la minates can be designed with appropriate lay-up and guarantee very high bending stiffness in non-morphing directions and even a not negligible shear stiffness, which can exceed 50% of the shear stiffness obtained by a conventional panel of the same we ight and lay-up (8). 4. concluding remarks the development of innovative solutions for internal structure and skin of morphing aerodynamic surface has been accomplished by jointly e xplo iting special structural geo metries and the properties of composite materials. the presented solutions fulfil the peculia r require ments of mo rphing structures and have been applied to develop a demonstrator of the chiral s ail concept, which is shown in fig. 5, in order to assess technological feasib ility, functional aspects and structural response. fig. 5 chiral sail demonstrator references [1] d. bo rnengo, f. scarpa, and c. re millat, “evaluation o f he xagonal chiral structu re for moprh ing airfoil concept,” journal of aerospace engineering, vol. 219, pp. 185-192, 2005. [2] a. airo ldi, m. crespi, g. quaranta, and g. sala, “design of a morphing a irfo il with compos ite chira l structure,” journal of aircra ft, vol. 49, no. 4, pp. 1008-1019, 2012. [3] p. pan ichelli, a. gilarde lli, a. airo ldi, g. quaranta, and g. sala, “morphing composite structures for adaptative high lift devices ,” proc. 6th int. conference mechanics and materials in design, ponta delgada, azores, pp. 26-30 july 2015 [4] p. bettin i, a. airoldi, g. sala , l. di landro, m. ruzzene, and a . spadoni, “ co mposite chiral structures for morphing a irfoils: numerical ana lyses and development of a manufacturing process ,” co mposites part b engineering, vol. 41, no. 2, pp. 133-147, 2010. [5] a. airoldi, p. bettini, p. panichelli, and g. sala, “ chira l topologies for co mposite moprh ing structures – part ii: novel configurations and technological process”, physica statu solidi (b), vol. 252, pp. 1446-1454, july 2015 [6] t. yo koze ki, s. takeda, t. ogasawara, and t. ishikawa , “ properties of corrugated composites for candidate fle xible wing structures ,” composites part a: applied science manufacturing, vol. 37, no. 4, pp. 1578-1586, 2006. [7] y. xia, o. bilgren, and m. i. friswe ll, “the effects of corrugated skins on aerodynamic performances ,” journal of intelligent material systems and structures, vol. 25, no. 7, pp. 786-794, 2014. [8] s. fournie r, a. a irold i, e. borlandelli, and g. sa la, “fle xib le co mposite supports for morph ing skins ,” proc. xxii aidaa conference, naples, italy, pp. 9-12, 2013.  advances in technology innovation, vol. 2, no. 1, 2017, pp. 25 28 25 ewma controller with concurrent adjustment for a high-mixed production process shui-pin lee department of industrial management, chien hsin university of science and technology, taoyuan, taiwan. received 22 february 2016; received in revised form 15 april 2016; accepted 17 april 2016 abstract the exponentially weighted moving average (ewma) feedback controller is a very popular run-to-run (rtr) process control scheme in the semiconductor industry. traditionally, the manufacturing environment was simplified as a single product and single tool process. in the feedback control, the adjustment of the recipe for the next run is related to the deviation of the current output against the desired target. however, in a commercial foundry, every tool always works for many products. it is called a multiple products and single tool (mpst) process. the challenge of this process is how to adjust the recipes among different products in a production row. in this study, a modified threaded ewma feedback controller, the ewma with concurrent adjustment, is proposed to deal with the issue of multiple products in a tool. when the process disturbances follow an ima(1,1) time series model, the stability of the proposed method will be proven. the optimum discount factors of the proposed ewma controller will be investigated by several simulations in terms of the number of multiple products, their distribution and the scheduling of the process. moreover, according to results of the performance comparisons with threaded ewma, the proposed controller is advantage in the large number of multiple products and in low-frequency products. keywords: ewma, run-to-run, feedback controller, high-mixed production process, concurrent adjustment 1. introduction run-to-run (r2r) process controller has been a conversional quality control technique in semiconductor manufacturing process. it was purposed by integrating statistical process control (spc) and engineering process control (epc) for overcoming the shift or gradually drift in the complex manufacturing process [1-2]. in most semiconductor manufacturing process, the run-to-run process controller for adjusting the input recipe is based on a known prediction model. assume that the i/o relationship of the siso process is linear, the process outputs ( 𝑦𝑡 ) can be expressed as follows: 0 0 1t t ty x     (1) where 𝑥𝑡−1 denotes the process input at run t that has been adjusted after its previous output 𝑦𝑡−1 obtained, 𝛼0 and 𝛽0 are the intercept and slope parameters, respectively, and 𝜖𝑡 denotes the process disturbances. the next process input at run 𝑡 + 1 will be adjusted by: t t a x b    (2) where τ is the process target, 𝑏 is the estimate of the slope 𝛽 and 𝑎𝑡 is the new estimate of 𝛼0 by using the following ewma formula:    1 11t t t ta y bx a      (3) typically, the researches related to the topic of how to enhance the performances of r2r controller are very restricted in a single product and single tool (spst) production environment. however, in many real commercial production processes, a tool will produce several different products. the same type of products might not be produced in-a-row. the kind of manufacturing * corresponding author, email: shuipin@uch.edu.tw advances in technology innovation, vol. 2, no. 1, 2017, pp. 25 28 26 copyright © taeti mode is often called a multiple-productmultiple-tool (mpmt) or a high-mixed production process. the foundries in taiwan are the typical examples of the high-mixed manufacturing mode. the manufacturing environment consists of multiple products passing through a sequence of batch processing steps being that are out by multiple parallel tools. recently, r2r control implementation in a mpmt environment has been discussed by many authors [3-5]. most of these works have attempted to identify parameters that can characterize the product and tool states. however, information is never shared between products or tools in such an approach. thus, epc or spc are ineffective for low frequency products. lee et al. [6] proposed a cumulative sum-type statistical process control for a mpmt process to detect substantial changes in tool and product effects. the estimate of gain parameter will be updated after signals. the recipes of inactive products will be adjusted if spc emits a signal. for statistical significance, the signal is related to the number of different products. in this short paper, the ewma with concurrent adjustment algorithm for a multiple products and single tool (mpst) process is proposed for inactive products. such that, when those products become active, their corresponding recipes can reflect the effect of the tool change. the organization of the paper is as follows. the proposed ewma with concurrent adjustment algorithm is introduced in section 2. in section 3, the stability condition of the proposed method and the evaluation of its performance will be shown. conclusions were then drawn in section 4. 2. ewma controller with concurrent adjustment if we assume the initial estimates of parameters α and β of (1) are correct, the input recipe will be set �̃� = 𝜏−𝛼 𝛽0 for all runs to make the expected output meet the process target. without loss generality, we can simplify (1) by the following expression: 0 1t t ty x   (4) assume that there are j products will be produced in a tool. let 𝑥𝑗,𝑡−1and 𝑦𝑗,𝑡denote the process input and output at run 𝑡. since a tool just can produce one product at each run, the product index 𝑗 is a function of 𝑡. the product 𝑗 is active at run 𝑡, but other 𝐽 − 1 products are inactive. in this paper, the i/o relationship of the mpst environment is , , 1 ,j t j j t j ty x   (5) where 𝛽𝑗 denotes the slope parameter with respect to product 𝑗 and 𝜖𝑗,𝑡 denotes the process disturbance. denote 𝛽𝑗 = 𝛽0𝑓𝐽 , where 𝛽0 denotes the nominal tool effect parameter and 𝑓𝑗 denotes the nominal product effect for product 𝑗 about the reference product. moreover, we take the contrast 𝑓1 = 1 for identifiability. denote 𝑏𝑗,0 as the initial estimate of 𝛽𝑗 , 𝑗 = 1, ⋯ , j then 𝑏1,0 is also the estimate of 𝛽0 and the estimates of the nominal product effects are 𝑓𝑗 = 𝑏𝑗,0 𝑏1,0 . hence, the initial prediction model can be expressed as follows:  , , 1 0 , 1ˆ|j t j t j j te y x b f x  (6) for the current active product 𝑗, its recipe can be adjusted by , , 0 , ˆ t j t j t j a x b f    (7) where 𝑎𝑗,𝑡 = 𝑎𝑗,𝑡−1 + 𝜔 (𝑦𝑗,𝑡 − 𝜏𝑡 ) is the formal ewma formula, but for the inactive products 𝑗′ ≠ 𝑗. we propose the concurrent adjustment ', ', 1j t j t ta a c  (8) where 𝑐𝑡 = 𝜔 (𝑦𝑗,𝑡 − 𝜏𝑡 ) and 𝜗 denotes the concurrent factor. 3. results and discussion the offset of the process output can be expressed by       1 1 0,2 , , 1, 2 0, 0, , , 0, , 1 1 1 j j l l j l j l j l j j t j j lj t h h j j h j l j t j h j t h y y c b c b b b                        (9) where 𝑙 denotes the series index of product 𝑗 , advances in technology innovation, vol. 2, no. 1, 2017, pp. 25 28 27 copyright © taeti ∅𝑗 = (1 − 𝜔𝛽0,𝑗 𝑏0,𝑗 ) denote the stable factors, 𝐶𝑗,𝑙 = ∑ 𝑐𝑖 𝑡−1 𝑖=𝑡−ℎ 𝑙 𝑗 denotes the cumulative deviation from target and 𝐵 denotes the backward operator. when the process disturbances is an ima (1, 1), or a arma (p, q), the variance of the process output is bounded if |∅𝑗 | is smaller than 1 for all 𝑗. according to several simulations in different conditions of the production environment, 𝜔 = 0.2 and 𝜗 = 1.0 are suggested. the performance of the proposed method was evaluated based on the criteria of the relative efficiency, the ratio of mean square errors of the proposed method to that of the threaded ewma controller under the same conditions. fig. 1 shows the results of the relative efficiency comparisons no matter what product scheduling (random or concentration) is, what the number of product types (2 or 4) is and what the magnitude of slope shift is, fig. 1 explicitly shows the proposed method is better than the threaded ewma controller. fig. 1 the relative efficiency of the ewma controller with concurrent adjustment with respect to the threaded ewma controller 4. conclusions in this paper, the ewma with concurrent adjustment algorithm control was proposed for a multiple products process. the exact expression of the process output of the proposed control algorithm is derived. hence, its stability conditions can be obtained. based on several numerical simulations,𝜔 = 0.2 and 𝜗 = 1.0 are suggested for the discount factor of ewma controller and concurrent factor when the process disturbances are a white noise series. compare to the threaded ewma controller in different production scheduling, the advantage of the ewma with concurrent adjustment is increasing in terms of the magnitude of the slope shift. moreover, the larger the number of product types, the bigger the advantage. acknowledgement the support of the minister of science and technology (taiwan), under grant most 104-2221-e-231-005 is gratefully acknowledged. references [1] a. ingolfsson a and e. sachs, “stability and sensitivity of an ewma controller,” journal of quality technology, vol. 25, pp. 271-287, 1993. [2] e. sachs, a. hu, and e. ingolfsson, “run by run process control: combining spc and feedback control,” ieee transactions semiconductor manufacturing, vol. 8, pp. 26-43, 1995. [3] a. v. prabhu and t. f. edgar, “a new state estimation method for high-mix semiconductor manufacturing processes,” journal of process control, vol. 19, pp. 1149-1161, 2009. [4] m. d. ma, c. c. chang, d. s. h. wong, and s. s. jang, “threaded ewma controller tuning and performance evaluation in a high-mixed system,” ieee transactions semiconductor manufacturing, vol. 22, pp. 1-5, 2009. [5] a. j. pasadyn and t. f. edga, “observability and state estimation for multiple product control in semiconductor manufacturing,” ieee transactions semiconductor manufacturing, vol. 18, pp. 592-604, 2005. [6] s. p. lee, d. s. h. wong, c. i. sun, w. h. chen, and s. s. jang, “integrated statistical process control and engineering process advances in technology innovation, vol. 2, no. 1, 2017, pp. 25 28 28 copyright © taeti control for a manufacturing process with multiple tools and multiple products,” journal of industrial and production engineering, vol. 32, no. 3, pp. 174-185, 2015.  advances in technology innovation, vol. 2, no. 4, 2017, pp. 130 132 130 oxidation of titanium and ti/ (tib+tic) composite poonam yadav , dong bok lee * school of advanced materials science and engineering , sungkyunkwan university, suwon 16419, south korea. received 09 oct ober 2016; received in revised form 29 november 2016; accept ed 30 november 2016 abstract titanium and titanium matrix composite reinforced with 10 vol. % (tib+tic) (10tmc) were oxidized at 700, 800 and 900 o c for 30, 50 and 70 h, and their oxidation behavior was studied. oxidation of ti and 10tmc in air led to the formation of rutile-tio2 without any other oxides. the oxide scale thickened with an increase in the oxidation time and temperature. significant growth of the rutile oxide caused thermal stresses which generated during oxidation and voids formed at the scale/matrix interface, and the oxide scales were susceptible to cracking. ke ywor ds : titaniu m, titaniu m mat rix co mposite, oxidation, oxide scale 1. introduction titanium and titanium alloys are light in weight, have high tensile strength, toughness, and excellent corrosion resistance. they are widely used for highly demanding structural parts such as aircrafts, chemical plants, automotive, ocean engineering, oil refinery, biomaterials, highly stressed components, and military applications [1-3]. recently, the introduction of particulates or continuous fibers with low density and high elastic modulus into titanium alloys has further improved the specific modulus, specific strength, creep resistance, wear resistance, and the service temperatures [4-8]. in order to utilize the titanium matrix composites (tmc) for industrial pu rpose, the oxidation study is imperative. in this study, the oxidation behavior of titanium and titanium matrix composite reinforced with 10 vol.% (tib+tic) particulates (hereafter, designated as 10tmc) were investigated after isothermally oxidizing at 700-900 o c for 30-70 h in air. particulate reinforced titanium matrix composites have attracted considerable attentions. 2. experimental titanium and the mold for tmcs were prepared by a common casting technique [9]. prepared mold was mounted in the vacuum induction melting furnace. after that, b4c (99% purity, 1500 mm, 150 mm and 0.5 mm) was added to pure ti (99% purity, grade 2), following which the synthesis of the tmcs was carried out using vacuum induction melting. the pressure of the furnace atmosphere was held at 1.33 x 10 -1 pa and charged with inert argon gas at a pressure of 4.9 x 10 3 pa. the oxide mold was super-heated sufficiently to overcome the problem of low viscosity of tmcs, and then 1.88 mass% b4c was added to the pure ti in the graphite crucible in order to form the 10 vol% of (tib+tic) reinforcement [9]. ti and 10tmc samples were cut to 10x10x2 mm 3 in size, mechanically polished, and ultrasonically cleaned with ethanol solution. research materials were oxidized at 700, 800 and 900 o c for 30, 50 and 70 h in air. following oxidation process, matrix phases and oxide scales were investigated using scanning electron microscopy (sem) equipped with energy dispersive spectroscopy (eds), x-ray diffraction (xrd) using cu-kα radiation operated at 40 kv and 100 ma. the matrix grain sizes were measured through the conventional mean-linear-intercept method using sem. the vickers microhardness was measured with a load of 100 n for 5 sec for more than 5 points on each sample at ambient temperature. 3. results and discussion fig. 1 shows the x-ray diffraction patterns of the oxide scales formed after isothermal oxidation at 800 o c for 70 h. fig. 1 x-ray diffraction patterns after oxidation at 800 o c for 70 h. (a) titanium, (b) titanium matrix composite rein-forced with 10 vol.% (tib+tic) fig. 2 sem image after oxidation at 900 °c. titanium for (a) 30 h, (b) 50 h, and (c) 70 h. titanium matrix composite reinforced with 10 vol.% (tib+tic) for (d) 30 h, (e) 50 h, and (f) 70 h *corresponding aut hor, email: dlee@skku.ac.kr advances in technology innovation, vol. 2, no. 4, 2017, pp. 130 132 131 cop y right © taeti 100µm (b) scale epoxy matrixvoids cracks 100µm (c) scale epoxy matrixvoids cracks 100µm (a) scale epoxy matrix voids cracks 20µm (d) scale epoxy matrix voids cracks 20µm (e) scale epoxy matrix voids cracks 20µm (f) scale epoxy matrix voids cracks 50 h 30 h 70 h 10%tmc pure ti for titanium and 10tmc, the oxide scales consisted of rutile-tio2 without any other oxides . apparently, boron and carbon in (tib+tic) evaporated during oxidation. sem micrographs are shown in fig. 2. the oxide grain size of 10tmc increased with an increase of an oxidation time more rapidly than that of pure titanium. particularly, the coarse rutile-tio2 grains formed on 10tmc after oxidation at 900 °c for 70 h were noticeable. fig. 3 shows the grain evolution of oxide grains that formed on 10tmc as a function of the oxidation temperature. the grain size has increased proportionally to the oxidation temperature, owing to the increased reaction and diffusion rates at high temperature. fig. 3 sem image of oxide scales formed on titanium ma-trix composite reinforced with 10 vol.% (tib+tic) after oxidation for 70 h at (a) 700 °c, (b) 800 °c, and (c) 900 °c fig. 4 sem cross-section image after oxidation at 900 °c. titanium for (a) 30 h, (b) 50 h, and (c) 70 h. titanium matrix composite reinforced with 10 vol.% (tib+tic) for (d) 30 h, (e) 50 h, and (f) 70 h fig. 4 shows the cross -section of titanium and 10tmc after oxidation at 900 o c for 30, 50, and 70 h. as the oxidation reaction proceeded, oxide layer formation has generated stresses, which induced cracking and delamination of the oxide scale. these effects in voids formation at scale/metal interface that could act as stress concentration sites. voids apparently formed owing to the significant outward diffusion of ti ions to form thick tio2 oxide scale. the average thickness of the oxide scale was 340 μm (fig. 4(a)), 350 μm (fig. 4(b)), 400 μm (fig. 4(c)), 125 μm (fig. 4(d)), 170 μm (fig. 4(e)), and 190 μm (fig. 4(f)). the addition of 10 vol. % (tib+tic) beneficially improved the oxidation resistance of the ti metal. during the initial stage of oxidation, oxygen is absorbed on the surface of the metal by physisorption. the oxygen molecules dissociate creating oxygen anions through a chemisorption process until a monolayer of oxide is formed. fig. 5 shows the sem/eds analysis results of 10tmc after oxidation at 700 °c for 70 h. the oxide scale was 7 μm-thicks (fig. 5(a)). unlike in the fig. 4, the oxide scale was dense, adherent, and crack-free owing to the slower growth rate and less thermal stresses generated during oxidation. fig. 5 titanium matrix composite reinforced with 10 vol.% (tib+tic) after oxidation at 700 °c for 70 h. sem cross-section, (b) eds line profiles along a-b vickers microhardness of 10tmc was measured at the surface of the formed oxides as a function of the oxidation time and temperature, and presented in fig. 6. during the early stage of oxidation, the microhardness increased at each temperature owing to the formation of thin hard oxide scale. the microhardness increased with higher oxidation temperature. ho wever at the later stage of oxidation, the microhardness d ecreased to a certain extent. fig. 6 microhardness of titanium matrix composite rein-forced with 10 vol.% (tib+tic) [top of the surface] as a function of oxidation time and temperature advances in technology innovation, vol. 2, no. 4, 2017, pp. 130 132 132 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti it is clear that, the hardness increases with temperature as well as with the oxidation time, up to 30 hrs. further oxidation causes the hardness to decrease; this may be attributed to stress relieving of the oxide grains arisen during heating stage of the oxidation process which induce cracking and delamination of the oxide scale. an evident matching with the results obtained from density; as the density increases, the hardness increases. 4. conclusions the oxidation behavior of titanium and 10tmc was investigated after isothermal oxidation at 700, 800 and 900 o c up to 70 h. the xrd analysis revealed that the oxide scale consisted of rutile-tio2.the scale thickness, growth and thermal stress increased with the increase in the oxidation time and temperature. the oxidation at 900 o c for 70 h led to the cracking and void formation in the oxide scale as well as around the oxide scale/matrix interface. microhardness of the oxide scale increased for the short stage of oxidation (30 h) and decreased for at the longer oxidation times (70 h) due to the stress-relieving arisen by the oxide grain growth. acknowledgement this work was supported under the framework of international cooperation program managed by national research foundation of korea (2016k2a9a1a01952060). references [1] m. yamada, “an overview on the development of tit anium alloys for non-aerospace application in japan,” material science and engineering a, vol. 213, pp. 8-15, 1996. [2] m. peters, j. hemptenmacher, j. kumpfert, and c. leyens, titanium and titanium alloys: fundamentals and applications, germany: wiley-vch, pp. 1-36, 2003. [3] s. r. seagle, “the state of the usa titanium industry in 1995,” material science and engineering a, vol. 213, pp. 1-7, 1996. [4] s. abkowitz, p. f. weihrauch, s. m. abkowitz, and h. l. heussi, “the commercial application of low-cost titanium composites,” journal of metals, vol. 47, pp. 40-41, 1995. [5] s. ranganath, “a review on particulate-reinforced titanium matrix composites,” journal of materials science, vol. 32, pp. 1-16, 1997. [6] f. h. froes, “recent advances in titanium metal matrix composites,” usa, tms, warrendale, pa, 1995. [7] t. yamamoto, a. otsuki, and k. ishihara, “synthesis of near net shape high density tib/ti composite,” material science engineering a, vol. 239, pp. 239-240, 1997. [8] d. b. lee, y. c. lee, and d. j. kim, “the oxidation of tib2 ceramics containing cr and fe,” oxidation of metals , vol. 56, pp. 177-189, 2001. [9] b. j. choi and y. j. kim, “ effect of b4c size on tensile property of (tib+tic) particulate rein-forced titanium matrix composites by investment casting,” materials transactions, vol. 52, pp. 1926-1930, 2011. http://link.springer.com/journal/11085 http://link.springer.com/journal/11085  advances in technology innovation, vol. 1, no. 1, 2016, pp. 21 24 21 copyright © taeti participatory communication referred to meta-design approach through the flexpeaker™ application of innovative material in exhibition design pei-hsuan su department of visual communication design, national taiwan university of arts , new taipei city, taiwan. received 22 february 2016; received in revised form 10 may 2016; accept ed 15 may 2016 abstract modelling a co mmun ication system in materia l culture today always involves with objects, people, organizations, activit ies and interre lationships among them. the research er suggests bringing together stakeholders engaged to exchange ideas, which the interactions relate to mult iple pro fessions and disciplines in a participatory scope of commun ication system. owing to the invention of dig ital media , the status quo of images and sounds has revolutionized and caused changes of the mode of art e xhib itions that produce activities and aesthetic concepts in terms of numerica l representation, modularity, auto mation, visual variability and transcoding. underly ing a participatory-design approach, the research emphasizes a co-c reative meta-interpretation of museum’s visitors . in addition, the research delves further into the use of new med ia-fle xpeaker™ [it ri], as the carrier. co mbin ing art and design with innovative technology, the research focuses on exa mining design objects and innovative material which a re applied in new media a rt and e xh ibition, in the hope to find new angles of partic ipatory interpretation of the “integrated innovation” in curating an exhibition. keywor ds : communication system, participatory design, exhibit design, flexible speaker 1. introduction intertwined within a participatory communication system nowadays , the invention of digital med ia has revolutionized the quality of images and sounds . it has also changed the way of images and sounds in which they are contemplated, and the mode of a rt e xh ibit ions that produce activities and esthetics concepts, because of the applications of “nu merical representation, modularity, auto mation, v isual variability and transcoding” in this new media age [1]. in fact, the reproductability of digital technology leads to the hyper-real scenes where images and sounds are produced, copied and stimulated in a way that mimet ic cannot be told apart from the original. co mbin ing art and design with innovative technology, my research focuses on e xa mining design objects and materia l innovation which a re applied in ne w media art and exhib ition, in the hope to find new angles of participatory interpretation of the “ integrated innovation” in curating an exhibition. the researcher e mphasizes how to bring in museum’s visitors’ co-creation and meta-interpretation on the e xhib ition site. the research delves further into the application of certain a new med ia-fle xpeaker™ [it ri], as the carrier. for instance, the social context and historical data are visualized and transcoded into specific images and life photography printed upon both sides of the surface of flexpeake r™, as we ll as with sound effects to stimulate visitors' imagination of the events, the time and the underlying environment at the e xh ib ition spaces in sun yun-suan me morial museum in taipe i [2]. therefo re, the e xhib ition devices and design objects do help visitors to realize the significance of diverse ele ments of sounds, images and historical-docu mentary data in relation to its cultural-heritage representation. * corresponding aut hor, email: sherriesutaipei@gmail.com advances in technology innovation , vol. 1, no. 1, 2016, pp. 21 24 22 copyright © taeti 2. method this participatory commun ication study is based upon how to employ partic ipatory design in the art exh ibitions and museum spaces. as the matter of fact, much e mphasis has been placed on design methods such as co-creation but even that has been largely limited to the involvement of visitors as participants. in addition, most of us get used to know the idea of hardware-oriented or professionally dominated des ign, so that we do think of visitors must adapt to the technology and the expertise must be shared with same background knowledge. ho wever, the researcher would like to notify that museum’s curators get to employ so-called “partic ipatory design” during curating time which focuses on systematic e xhib it development to envision context within design thinking and practices . fundamentally different fro m creating complete settled–up e xhib it systems, the curators need to consider a defining activity for empowe ring part ic ipation, a imed at creating certain interactive design devices as the exhibits , for other visitors. in addition, based on the idea of “partic ipatory design,” through focus ing on general e xhib ited structures interacted with visiting processes rather than on fixed e xh ibited objects and contents , the museum visitors may get deeper involved and achieve to the “world-as -e xpe rienced situated action ” later engaged in the e xh ib it design system. here we anticipate the partic ipatory nature applied in e xhibit design and the transition to more autonomous action for discovery by museum visitors . this method releases more power to the visitors to enhance their engagement while building up an exhibit system. 3. innovative electronic material of flexible speaker there are nume rous technologies and products related to fle xible e lectronics. fle xib le electronics is sort of general term for using organic materia l, printing manufacturing process, electron ic circuit, optoelectronic co mponents, or the technology of setting on fle xib le substrate with low cost and the characteristics of being fle xible [3]. in specific , the researcher introduces here the flexpeake r™ as one kind of significant innovative electronic materia ls. it developed by fle xib le electronics pilot lab of it ri, ta iwan; fle xpeake r™ is one of fle xib le electronics applications , which helps itri received the wall street journal’s 2009 technology innovation award [4] [5]. the technology utilizes paper and metal layers as the materia l with a thickness of less than 0.1 c m and uses standard printing for large-size paper-thin fle xib le speaker mass production. the great sound quality covers a range of 20 to 200 khz. it is especially good for high -frequency sounds such as the chirps of birds and in sects, where fidelity equals or e xceeds that of conventional speakers [6]. in addition, the fle xpeaker™ uses only 10% as much power of conventional speakers, ma king it environmentally friendly. the new technology will bring the acoustic speaker industry into a brand-new era, and help create revolutionary consumer products such as me mo ry cards with voice capabilit ies and ultra-thin m p3 p layers. it could even be incorporated into other products that are integrated into exhib it designs, green buildings, electric vehic les, entertain ment and medica l devices. the technology of fle xpeaker™ will help create new lifestyles and cater to the pursuit of personalized, social and cultura l, as we ll as humanized applications [7]. 4. results and discussion for the pulse o f the t imes and trends, the researcher would like to discuss how the curators has emp loyed integrated innovation in e xhibit ion design, since 2014, and delving further into the new media as the carrier in sun yun-suan me morial museum, taipe i. ne w shapes of exh ibition objects have served as communicat ion media so that the images and sounds can be mixed used on the creations, and formed by projecting, fra ming, in laying, attaching, hanging, and erecting (fig.1). there are mixing of innovative technologies and devices installed in the exh ibition roo m. for instances, the curator uses one ultra-short throw projector, hidden rightly in the rea r of one piece of the glass -board, to deliver information that represents a virtual television screen on the glass (fig. 2 & 3) . in fact, the device o f the “ultra -short throw pro jector” was developed by delta electronics, inc., taiwan. advances in technology innovation, vol. 1, no. 1, 2016, pp. 21 24 23 copyright © taeti fig. 1 the diversity of e xhib ition sites at sun yun-suan memorial museum in taipei fig. 2 using one “ultra -short throw projector” to deliver info rmation wh ich was hided rightly in the rear of the glass -board fig. 3 representing a virtual television screen in the front of the glass -boards in the exhib it room fig. 4 life photography printed upon both sides of the surface of paper-thin fle xpeaker™ to dedicate that sun leaded taiwan power co mpany in restoring the power network in taiwan in addition, the curator introduces the economic-industrial contribution of the e x-pre mier of r.o.c., sun, yun-suan, who managed a staff of several hundred at taipan power co mpany, and was able to get 80% of the power network in ta iwan (destroyed during the world war ii) restored in five months in the year of 1946 [8]. his biographical social-conte xtual data are visualized and transcoded into specific images and life photography printed upon both sides of the surface of paper-thin flexpeake r™ (fig. 4) as we ll as with sound effects to stimulate visitors’ imag ination of the events, the time and the underlying environment. therefo re, sun yun-suan me morial museum ca me out the integration of arts -design-expertise in creating new exhibit experiences with technology. 5. conclusions refe rred to the idea of pa rtic ipatory e xhibit-design, the sun yun-suan me moria l advances in technology innovation , vol. 1, no. 1, 2016, pp. 21 24 24 copyright © taeti museum helps its visitors to take part in the e xperience of converging sound, images and documentary data by mixing a series of contemporary innovative technologies and devices. the reproduction of contemporary sounds and images leads to the hyperreal scenes at the exh ibit ion spaces in the museum where sounds and images are produced, copied and stimulated in a way that mimetic cannot be told apart from the orig inal data. just as “simu lations” of jean baudrillard [ 9], the digita l visual communicat ion media and digital interface changed into various ways and forms. however, it shows the hybrid of contemporary images and sounds. the creations are within the ran ge of the post-modernism, wh ich stress on the interdisciplinary and inter-te xtua l of se miotic translations. with a collaborative mu lti-disciplinary schema, we have the co mprehensive knowledge and means address to problems. he re the curators need to work with historians, visual designers, web developers, user expe rience designers, and information designers in o rder to enhance the exhibit design can be extre me ly powerful in creating the kinds of tools or vehicles that visitors will not only need but will be able to effectively use. there is critical to emp loying a meta-design fra mework, ensuring that the exh ibit environ ment enables end-users to engage in in formed part icipation. the communicat ion system and devices developed in the exhib ition should be integrated in consultation with those end-users , as museum’s visitors . acknowledgement the support of the 2014-2015 it ri’s curating team working with the sun yun-suan memorial museum is gratefully acknowledged. references [1] l. manovich, the language of the ne w media, new york: mit press , 2001. [2] p. h. su, “the aesthetics on meta-interpretation and hyperreal: an e xperimental converging practice of sounds, images and data applied in sun yun-suan me morial museum, taipei,” sid 2015 confe rence of sounds, images and data, new york unive rsity, steinhardt school, u. s., july 2015. [3] j. c. yeh, “the applicat ions of fle xib le electronics go everywhere in the future,” co mpotech china, pp. 39-41, oct. 2011. (in chinese) [4] itri, fle xib le electronics pilot lab. retrieved from https://www.itri.org.t w/eng/content/msgpi c/contents.aspx?siteid=1&m mmid=6177 54414641431112&msid=6177544376336 12573 (2016/ 02). [5] itri, “2012 it ri fo ru m: ne w technologies and venture capital.” no.69, 2nd quarter 2012. retrieved from: https://www.itri.org.t w/eng/content/public ations/contents.aspx?&siteid=1&m mmid =617731525164776565&msid=61776236 2453772447 (2016/ 02). [6] itri, innovations and applications_smart living_smart endpoints_ flexpeake r™. retrieved from: https://www.itri.org.t w/eng/content/msgpi c01/ contents.aspx?siteid=1&m mmid=62 0651706136357202&msid=62102402525 7654754 (2016/ 02). [7] itri, “ kevin kelly: ‘i was ama zed by itri’s innovations.’” no.69, 2nd quarter 2012. retrieved from https://www.itri.org.t w/eng/content/public ations/contents.aspx?&siteid=1&m mmid =617731525164776565&msid=61776236 3015606130 (2016/ 02). [8] e. l. yang, the biography of sun yun-suan, taipei: common wealth magazine press, 1989. (in chinese) [9] j. baudrillard, simu lacra and simulat ion (sheila glaser t rans), u. s.: un iversity of michigan press, 1994. (original work published in 1981).  advances in technology innovation, vol. 4, no. 1, 2019, pp. 30 36 design and testing of a remote deployable water purification system powered by solar energy amber e. keith, jesse j. french * department of mechanical engineering, letourneau university, longview, texas, usa received 20 july 2017; received in revised form 05 september 2017; accepted 15 january 2018 abstract the design of an all-inclusive, self-sufficient, sustainable water purification system for application in developing regions of the world can improve living conditions for people world-wide, especially in regions where access to clean drinking water is limited or unavailable. according to the world health organization, the 2015 global census estimated that 663 million people worldwide live without access to safe drinking water sources [1]. an estimated 315,000 children die each year from diarrheal diseases caused by lack of clean water and poor sanitation [2]. based on these statistics, an all-inclusive, self-sufficient, remote-deployable water purification system has been designed, constructed, and tested to validate the concept of a renewable energy system. the system is integrated into a standard 20-foot shipping container for ease of deployment worldwide. once situated in the operating area, the shipping container is used as the system shelter and solar panels are mounted to the roof at a location-dependent fixed angle. the solar panels are connected to a battery bank which operates the system. the water purification process utilizes a five-step progression which filters the contaminated freshwater, removing suspended particles and bacteria from the water and purifying the water to the standards of the epa safe drinking act [3]. testing verifies the capability of the solar panels to generate enough electricity to power the system and recharge the battery bank. the solar panel array has the rated power output of 2,320 watts. the water purification operates on a maximum of 97.9 watts of available power. with the fully charged battery bank, the water purification system can operate for 24-hours without additional solar input. with a freshwater source, the purification system can yield up to 440 liters of water per hour. keywords: renewable energy, solar power, clean water, water purification, self-sufficient 1. introduction without water, a person will die of dehydration within three days [4]. without water, crops cannot be grown and starvation can result. though over 70% of the earth’s surface is covered in this vital source of life, clean drinking water is scares [2]. freshwater, water that is not full of dissolved salts, makes up only about 2.5% of the water source on the entire planet [5]. of that 2.5%, only about half of that is accessible for use [5]. according to the world population clock, there is currently 7.5 billion people on the plant with a projected population reaching 8 billion by the year 2023 and 10 billion by t he year 2056 [6]. those 7.5 billion people must share the one percent of available freshwater on the world for drinking, cooking, and cleaning. unfortunately, the distribution of drinking water is not proportional to the distribution of the population [7]. some areas enjoy a surplus of water, whereas others have significant shortages [7]. as can be seen in fig. 1, the water * corresponding author. e-mail address: jessefrench@letu.edu advances in technology innovation, vol. 4, no. 1, 2019, pp. 30 36 31 scarcity is very drastic, especially for desert regions such as central africa. according to many sources, including lifewater, a non-profit working to end the global water and sanitation crisis, water and poverty are mutually dependent [8]. in areas where access to clean drinking water has been introduced, the quality of life of the people also increases [8]. fig. 1 global water scarcity [9] an estimated six to eight million people die from water-related diseases each year [10]. inventor of the segway, dean kamen, once said, “we could empty half the hospital beds in the world by just giving people clean water” [11]. dean kamen has gone on to design the slingshot water purification system which uses a vapor distillation method to bring about this goal to reduce the number of people in hospital beds, simply by improving access to clean water [11]. water shortages are not the only shortage affecting the developing world. the international energy agency has estimated that 1.2 billion people worldwide live without access to electricity [12]. seventeen percent of the global population lives without electricity, even for simple household illumination, cellphone charging, or computer use. by proving a system that generates excess electricity after purifying water, two global issues can be solved with a single system. 2. system design the sustainable purification system (sps) is powered solely by renewable resources to provide access to clean drinking water limited electricity to underdeveloped regions around the world. for simplicity of deployment and maintenance, the water purification system has been designed to be chemical-free, since access to chemicals in remote locations would be nearly impossible and not cost-effective. the minimum desired purified water output is 150 liters per day. this is a sufficient water output for a 50-person community with the assumption that each resident receives 3-liters of purified water per day. further testing has been conducted to determine the exact potential output for the system based on optimum solar conditions and power supplied from the battery bank. the water purification system is powered by eight 290 watt solar panels mounted on the roof of the shipping container. the solar panels are mounted at a fixed angle of 30.0°. this angle was chosen based on the latitudinal position of the testing location. electricity generated by the solar panels is directed through a charge controller, which will either utilize the power for pumping and purifying water or for charging the on-board battery bank. excess electricity is available for system, lighting and installed outlets for powering electronics such as cellphones. the sps container can be seen in fig. 2 with the solar array mounted on the roof at the fixed angle. the system is monitored by the charge controller which is designed to keep the battery bank from dropping below a 50% charge value in order to preserve the lifespan of the deep-cycle batteries. an arduino-based controller monitors the battery voltage and water levels in order to automatically produce water, as needed. if the battery bank drops below 50% capacity, the arduino controller will shut off the system to prevent over-discharge. the charge controller and arduino for the system can be seen in fig. 3 below. advances in technology innovation, vol. 4, no. 1, 2019, pp. 30 36 32 fig. 2 sustainable purification system shipping container enclosure fig. 3 arduino and charge controller setup for the sustainable purification system fig. 4 water purification system with labeling the system design for purification is broken into three main sub-systems: pumping, purification, and storage. the five stages of filters can be seen in fig. 4. first, the water is pumped from the source; whether a deep or shallow well, river, lake, or sea. the pump is a fully submerged pump that is coupled with a flow switch to turn off the pump when insignificant water is available. next, the water proceeds through the three-stage purification system: filtration, sterilization, and purification. the entire design is constructed into a custom workbench to contain the components. the filtration stage includes three filters types: sand, spin-down, and two cartridge filters. the water flows through the sand filter to remove large particulates, such as dirt and leaves. this filter is designed to process up to 30,000 liters of water at up to 140 liters per minute. the spin-down filter is a 15-micron screen to remove particles left behind after the sand filter. the final stage is a set of cartridge filters remove down to five microns and then one-micron sized particles. any particulate remaining after the filtration stage is less than one micron in size. for comparison, a human hair is about 75 microns in size [13]. next is the sterilization phase. a uv lamp provides fluence to the water to kill bacteria and viruses. the bacteria and viruses are not removed in this stage, just eradicated. in the final purification phase, the water passes through a rapidpure filter system. this stage filters particles down to 1.75 microns and then implements a positive charge to remove additional particles. advances in technology innovation, vol. 4, no. 1, 2019, pp. 30 36 33 this final phase removes 99.9% of bacteria, 99.9% of viruses, and 99.8% of cyst. the rapidpure filter also reduces the concentration of bromine, chlorine, iodine, lead, penicillin, and other various heavy metals. at this time, the water is safe to drink and flows into the storage tank. 3. experimental procedure testing was conducted to determine the rate of purification to determine how much water the system could purify in a full eight-hour cycle of operation. the power requirements for the individual components of the system were also characterized. to determine the system power requirements, the charge controller was set in the float state with the batteries fully charged. this means that all power generated would be used for powering the system and not for additional charging of the battery bank. to ascertain how much power was used by the system, the purification system cycled ten liters of water. the charge controller displays the amperage draw of the system and the overall voltage. by knowing the amperage and voltage of the system, the power draw can be calculated with ohm’s law: 𝑐𝑢𝑟𝑟𝑒𝑛𝑡 [𝐴] × 𝑣𝑜𝑙𝑡𝑎𝑔𝑒 [𝑉] = 𝑝𝑜𝑤𝑒𝑟 [𝑊]. fig. 5 digital flow meter installed in system between the purification system and storage tank fig. 6 k24 digital flow meter with 3d printed adapters the pump is rated for a maximum flow rate of 7.0 liters per minute when the pump is at a head of 20 feet. this means that if the pump is operating at peak performance, then in a single hour, the system should theoretically cycle 420 liters of wate r. this provides three liters of water per resident for up to 140 people per hour. for flow rate testing, the k24 turbine digital flow meter was placed in-line with the system between the rapidpure filters and the storage tank as seen in fig. 5. the k24 digital flow meter has one-inch british standard pipe (bsp) threads on both ends of the display for the inlet and outlet. the hose between the final stage of the purification system and the storage tank is ¾ -inch clear tubing. the solution to convert the 1-inch bst to 3/4-inch barb hose fitting was a series of three transition adapters. since bsp is not a standard size available in the united states, the adapters were 3d printed and threaded together with sealing tape to prevent leaking between the adapters. advances in technology innovation, vol. 4, no. 1, 2019, pp. 30 36 34 the flowmeter displays the flow rate and number of liters that have passed through the system. during testing, the flow rate was recorded periodically during a continuous twenty-minute system operation. the k24 digital flowmeter can be seen fig. 6 in with the three types of transition adapters labeled. 4. results and discussion 4.1 water purification rates during testing for system operations, data was recorded for elapsed time and liters cycled. the digital flowmeter displayed a consistent 7.1 liters per minute flow rate during the entire period. after an elapsed time of thirteen and a half minutes, a full 100-liters had been processed through the system. at the conclusion of testing, 147.4 liters of water had been purified. this is slightly higher than the predicted 140 liters from the expected flow rate. this testing was conducted under optimum conditions since the overall head of the pump is less than 20 feet. extrapolating the testing results, it is estimated that the system can purify up to 440 liters per hour. if the system operates for 8 hours a day, up to 3,520-liters water can successfully be purified in a single day of operations. since the design parameter is to provide 3-liters of clean drinking water to each resident, the current system is properly sized for a 1,000-person community. the predicted verse the actual volume of purified water compared to elapsed time can be seen in fig. 7. as shown in the graph, the flow rate of the water purification is linear. fig. 7 volume of purified water vs. elapsed time 4.2 power consumption table 1 current and power draw for system components control system uv filter pump whole system current [a] 0.7 1.4 2.4 3.6 voltage [v] 27.2 27.2 27.2 27.2 power draw [w] 19.04 38.08 65.25 97.92 the power system requires 24v of direct-current [dc] electricity. the charge controller displays a constant system voltage of 27.2v, which is the charging voltage of the storage batteries. table 1 shows a summary of the amperage and power draw of different combinations of system components. the arduino controller and charge controller are continuously operated, so the current draw is constant at 0.7a. the 19w power draw from the control system is required for system monitoring and operations. most of the purification system is passive and does not require electricity; the uv filter being the only component that requires power to function. with the uv filter operation and the pump turned off, the system requires 38w of power to 0 20 40 60 80 100 120 140 160 0 5 10 15 20 w a te r v o lu m e [ l ] elapsed time [min] flowrate predicted vs. actual volume predicted volume [l] actual volume [l] advances in technology innovation, vol. 4, no. 1, 2019, pp. 30 36 35 operate. this scenario, however, is impractical since without the pump, no water is actively passing the uv filter. the next evaluation was with the pump functioning and the uv filter turned off. the pump combined with the control system draws 2.4a for a power draw of 65w. finally, with all components of the system functioning, the current draw raises to 3.6a and requires 97.9w of power. the graph in fig. 8 displays the power draw of the various components with the power draw from the control system being constant. fig. 8 power requirements as additional components are activated; cumulative effect on the overall power draw the system power draw is significantly less than the overall power potential of the solar array. excess power generated can be utilized for charging the battery bank for use when the solar insolation is minimal. in the case where the battery bank is already fully charged and the sun is out to continue generating power, excess electricity. potentially, surplus power generated could be utilized for charging cell phones or powering household lighting; however, first priority of the system is to keep the battery bank at full charge in case of insufficient sun for power generation for operating the system. 5. conclusion with the self-contained system integrated into the shipping container, the water purification system can be deployed, set up, and operated wherever in the world there is a need for clean water and access to solar energy. the current water purification system is operational and sufficient for providing clean drinking water to a small community anywhere in the world provided there is an accessible source of fresh water for filtration. through testing, it has been demonstrated that the solar array provides more than sufficient power to operate the water purification system, with the excess being used to charge the battery bank. with a fully charged battery bank, the system can be operated full 24-hour period without solar input. this system can be designed and optimized for any application and for a variety of locations. the solar array can be expanded and the fixed mounted angle of the panels is customized to the final destination, based on the latitude of the location. through the research and development of the sustainable purification system, a practical system is built and tested which is relevant to the global water crisis. references [1] who library cataloguing-in-publication data, “progress on sanitation and drinking water-2015 update and mdg assessment,” pp. 17-35, 2015. [2] “statistics-the hard facts behind the crisis,” http://www.wateraid.org/what-we-do/the-crisis/statistics. february 24,2017. 0 10 20 30 40 50 60 70 80 90 control system uv filter pump whole system p o w e r d ra w [ w ] power draw vs. system components pump uv filter control system advances in technology innovation, vol. 4, no. 1, 2019, pp. 30 36 36 [3] “safe drinking water act,” https://www.epa.gov/sdwa, 2002. [4] c. binns, “how long can a person survive without water? ” http://www.livescience.com/32320-how-long-can-a-person survive-without-water.html, november 30, 2012. [5] h. perlman, “how much water is there on, in, and above the earth ” http://water.usgs.gov/edu/earthhowmuch.html, december 2, 2016. [6] “current world population,” http://www.worldometers.info/world-population december 11, 2016. [7] b. chaouchi, a. zrelli, and g. slimane, “desalination of brackish water by means of a parabolic solar concentrator,” desalination, vol. 217, no. 1, pp. 118-126, 2007. [8] “water and poverty,” https://lifewater.org/blog/water-poverty/, december 26, 2017. [9] managing water under uncertainty and risk, the united nations world water depvelopment report 4, paris, unesco, 2012. [10] “water facts and figures,” http://www.unwater.org/water-cooperation-2013/water-cooperation/facts-and-figures/en/, 2013. [11] s. nasr, “how the slingshot water purifier works,” http://science.howstuffworks.com/environmental/green-tech/ remediation/slingshot-water-purifier.htm, july 27, 2009. [12] “energy poverty,” http:// www.iea.org/topics/energypoverty/, january 10, 2017. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation, vol. 2, no. 1, 2017, pp. 18 21 18 optimal warranty length and selling price to maximize the profit yu-hung chien 1,* , chung-piao chiang 2 1 department of applied statistics, national taichung university of science and technology, taichung, taiwan. 2 graduate institute of physical education, national taiwan sport university, taoyuan, taiwan. received 22 february 2016; received in revised form 15 april 2016; accepted 18 april 2016 abstract this study focuses on the problem of determining the optimal coverage period and selling price of warranted products from the manufacturer’s perspective. we first consider how to maximize the profit per unit under the assumption that the product can be sold or the demand is independent of the warranty policy. then we try to maximize the total profit for a planning period for the case where the demand for the product depends on the warranty coverage period and selling price. since the warranty period and the selling price should be positively correlated, we first solve the profit maximization problem with the warranty depended demand under the constraint that the selling price is a linear function of the warranty coverage period (warranty based pricing). furthermore, we investigate the case when such a constraint is removed (non-warranty-based pricing). optimizing on two independent decision variables, the coverage period and the selling price, certainly improves the total profit. under the two variable optimal conditions, it is observed that while the positive relationship between the optimal coverage period and the optimal price is confirmed, it is more complex than a linear one. we also find that the profit advantage for the non-warranty-based pricing over the warranty-based pricing is more significant for shorter coverage period. however, when the coverage period exceeds a threshold value, such a profit advantage becomes insignificant. the results of this study provide practitioners with useful insights in designing the profit optimal product warranty in highly competitive market. keywords: warranty, profit maximization, repair, replacement, renewable, optimization, price 1. introduction in today’s highly competitive market of consumer products, product warranties become an important criterion that distinguishes good and poor quality products. a warranty satisfying customer well can enhance customer willingness of purchasing, hence stimulate customer demand. this is because that a comprehensive warranty that covers a longer period implies a higher quality of the product. therefore, manufacturers try to offer a variety of warranties to expand the market share as consumers consider warranties as one of most important aspects in deciding which product to be purchased. however, a more attractive (or better customer service) warranty also implies higher service costs for manufacturers. as the product becomes more complex and expensive, the coverage of the warranty is required longer. an obvious trade-off between the benefit and cost of a warranty for a manufacturer is a critical issue for both academic researchers and practitioners. in the past decade, most of studies in the literature focused on the issue of how to minimize the warranty cost via preventive maintenance. shafiee and chukova (2013) provided a survey in this area from 2001 to 2011. there are relatively fewer works addressing both benefit and cost aspects of a product warranty. in fact, the marginal benefit can well exceed the marginal cost when the warranty coverage period is not long enough. for example, murthy et al. (2004) pointed out that on average the marginal profit for product with a warranty is about 30% compared with 10% without a warranty. in addition, * corresponding author, email: yhchien@nutc.edu.tw advances in technology innovation, vol. 2, no. 1, 2017, pp. 18 21 19 copyright © taeti a warranty with more comprehensive and longer coverage period helps build up the company and product image that may have the same positive effects of advertisement on future demand. thus, a good warranty benefits both consumer and manufacturers. as mentioned before, there are not many studies on determining product warranty parameters with consideration of its effects on market demand. glickman and berger (1976) is one the earliest papers on the effects on consumer demand of warranty parameters. they addressed the problem of determining the optimal price and the warranty period and considered demand function of market statistic in which the demand decreases exponentially with price and increases exponentially with the warranty length (coverage period). applications of the demand function of this form can be found in the literature such as mitra and jayprakash (1997). ladany and shore (2007) also consider a cobb-douglas-type demand function in determining the optimal warranty period, they presented a review on manufacturer’s pricing strategies for supply chain with warranty period-dependent demand. from the consumer’s perspective, the longer the warranty period is, the better the after-sales service is. however, a longer warranty period implies a higher expected warranty cost to the manufacturer. this is particular true for the renewing free-replacement warranty (rfew) under which the product must be replaced with a new one and a new full warranty period if the product fails within the warranty period offered. since the purpose of refw is to ensure a product to function for the entire warranty period, the length of the warranty period can be viewed as a measure of the product’s quality and reliability. in addition, due to its cost implication to a manufacturer, we can assume that there is a positive correlation between the selling price and the warranty period. in this paper, we first find the optimal warranty period or selling price to maximize the profit of a sold product. another issue we address is the impact of the rfew on the customer demand. it is reasonable to assume that the warranty period and the customer demand can be positively correlated for a certain range of the selling price. to model such a relation, we utilize the cobb-douglas-type demand function of selling price and warranty period to determine the optimal or profit maximization warranty period and/or selling price for the manufacturer 2. mathematical formulation consider a product with a random life-time of x. let f(x), f(x) (=p(x  x)), )(xf (=1 f(x)), and )(xr ( )()( xfxf ) be the probability density function (pdf), cumulative distribution function (cdf), survival function (sf), and failure rate function (fr) of x, respectively. let w be the period of the rfew and p c be the selling price of the product. we first assume that there is a positive linear relation between the selling price and w or  p c a+bw. such an assumption has been made in ladany and shore (2007). under this assumption, we can find the optimal w to maximize the profit of a sold product. 2.1. unit profit maximization for a product sold denote by y the number of failures (replacements) of a sold product within the rfrw. according to the scheme of rfrw, the probability mass function of y can be expressed as           .1 if ),( ,0 if ),( ywfwf ywf yyp y (1) let cr be the replacement cost when the product fails within the rfrw period. based on (1), the expected cost of rfrw is )( )( )()( wf wf cyecwc rr  (2) thus the expected profit for the product sold is )( )( )()( wf wf cbwawccw rp  (3) let * 0w be the maximize of (3) or  *0w  ww 0max  . if the failure rate is an increasing function (ifr), we can establish the following properties. property 1. for a product sold with the rfrw, if the selling price is a linear function of the warranty period, the profit maximization * 0w has the following properties: advances in technology innovation, vol. 2, no. 1, 2017, pp. 18 21 20 copyright © taeti (i) if rcbr )0( , then 0 * 0 w and  *0w 0)0(  a ; (ii) if rcbr )0( ， then  * 00 w exists and unique, and   aw * 0  . this proposition indicates that there exists a finite warranty period that maximizes the profit and the profit of selling one product is at least a. whenever the initial failure rate exceeds a threshold (or rcbr )0( ), the rfrw should not be provided or 0 * 0 w and the unit profit is still at least a. 2.2. total profit maximization for products sold in a planning period now we solve the profit maximization problem when the impact of the rfrw on the customer demand is taken into account. let )(wq denote the quantity sold in a planning period for products sold under rfrw. it is assumed that the demand depends on the warranty period w and is modeled as cobb-douglas-type function:   )()()( wcwq p (4) where 0 , 0 , 0 . denoting the total profit for the period by )(w , we have )()()( wwqw   (5) denote the profit maximal warranty period by * w or )(max)( 0 * ww w   . we can obtain the following result. property 2. for the product with an rfrw, if the selling price is a linear function of w and the demand function is given in (4), then there exists a unique and finite w that maximizes the total profit for a planning period. 2.3. total profit maximization without warranty-based pricing now we assume that both the price and the warranty period are decision variables for maximizing the total profit for a planning period. the total profit function ) ,( wc p can be written as ) ,( wcp ) ,() ,( wcwcq pp         )( )( )()( wf wf ccwc rpp   (6) where 0 , 0 and 0 . we can show that there exists an optimal solution to (6). property 3. for the profit function ) ,( wc p given in (6), under the condition 01  , if there exist ( * pc , * w ) satisfying )( )( 1 * ** wf wrw             and )( )( 1 * * * wf wfc c r p     , then ( * pc , * w ) maximize ) ,( wc p  or ) ,( max) ,( 0 ,0 ** wcwc pwcp p   . an interesting finding here is that when both the price and warranty period are decision variables, the optimal warranty period * w only depends on the demand function parameters of  and . it does not even depend on the replacement cost. specifically, as  increases both * w and * pc increase; as  decreases both * w and * pc decrease. these relations are intuitive. now we solve the profit maximization problem when the impact of the rfrw on the customer demand is taken into account. let )(wq denote the quantity sold in a planning period for products sold under rfrw. it is assumed that the demand depends on the warranty period w and is modeled as cobb-douglas-type function: 3. numerical illustration and sensitivity analysis assume that lifetime of the product x has a weibull distributed survival function   xxf  exp)( (7) where scale parameter and shape parameter are 1 and 2 , respectively. clearly, such a distribution has the ifr property. advances in technology innovation, vol. 2, no. 1, 2017, pp. 18 21 21 copyright © taeti our first numerical example is the one with the warranty-based pricing and is based on the set of parameters as 0.1rc , 0.1a , 2.0b , 10 , 5.0 , and 5.0 . the unit profit function and its derivative can be shown. the decreasing first-order derivative of the unit profit function shows the concavity of the function. thus the first-order derivative condition will give the global optimal point. the total profit )(w and its derivative are shown below respectively, which indicate that a unique optimal warranty period exists although the profit function has both convex and concave sections. next we consider the case where both the selling price and the warranty period are decision variables for maximizing the total profit. the optimal * pc , * w and ) ,( ** wc p are computed under various  and  , presented in table 1. table 1 the optimal w * , c * p and π (𝑐𝑝 ∗ , 𝑊 ∗) under various 𝛽 and 𝛾   1.1 1.2 1.3 1.4 1.2 2.446 2175.7 12.330 1.680 47.451 7.176 1.262 8.4862 5.3544 0.935 2.4447 4.6083 1.4 2.645 6002.0 14.865 1.839 85.281 8.0364 1.423 14.247 5.6810 1.117 4.3440 4.6343 1.6 3.000 44561. 22.517 2.109 253.33 10.553 1.680 34.270 6.7785 1.386 10.198 5.0772 4. conclusions in this paper, we have studied the problem of joint determination of the optimal selling price of a product and the optimal warranty period offered at no extra charge with the product as an incentive to the consumer. expected demand in our model was taken to be the cobb-douglas type demand function. structural and insightful properties of the optimal results are obtained and compared analytically. our study provides a useful quantitative tool for sellers to decide price and warranty length for products sold. acknowledgement the support of the ministry of science and technology, under grant nsc 102 2221 e 025 004 my3 is gratefully acknowledged. references [1] t. s. glickman and p. d. berger, “optimal price and protection period decisions for a product under warranty,” management science, vol. 22, pp.1381-1389, 1976. [2] s. p. ladany and h. shore, “profit maximizing warranty period with sales expressed by a demand function,” quality and reliability engineering international, vol. 23, pp. 291-301, 2007. [3] a. mitra and g. p. jayprakash, “market share and warranty costs for renewable warranty programs,” international journal of production economics, vol. 50, pp. 155-168, 1997. [4] d. n. p. murthy, o. solem and t. roren, “product warranty logistics: issues and challenges,” european journal of operational research, vol. 156, pp. 110-126, 2004. [5] m. shafiee and s. chukova, “product warranty logistics: issues and challenges,” european journal of operational research, vol. 229, pp. 561-572, 2013.  advances in technology innovation , vol. 1, no. 2, 2016, pp. 41 45 41 copyright © taeti experimental investigation into mechanical properties of nanomaterial-reinforced table tennis rubber yu-fen chen 1,* , jian-hong wu 2 , chen-chih huang 3 1 office of physical education, national formosa university, yunlin, taiwan. 2 taiwan semiconductor manufacturing company limited, hsinchu, taiwan. 3 department of sport, health & leisure, wufeng university, chiayi county, taiwan . received 02 february 2016; received in revised form 28 march 2016; accept ed 02 april 2016 abstract a new table tennis rubber is prepared consisting of carbon nanotubes, zinc o xide and titanium o xide added to a mixture of natural and synthesized rubber. the nano-re inforced rubber is attached to wooden table tennis blades and patterned with four diffe rent surface structures, name ly flat, long pimp les, short pimp les and med iu m pimp les. the results show that of the five rubbers, the nano-reinforced rubber with a flat surface offe rs a significantly imp roved elastic and mechanical performance. keywor ds : table tennis rubber, surface modification, carbon nanotube, zinc oxide, titanium oxide 1. introduction poly mer co mpound materials have many advantages over traditional engineering metals and alloys, including a high strength, a low weight, good resilience, a low cost, and superior chemical resistance. consequently, the synthesis and characterization of poly mer co mposites has attracted significant attention in the literature [1-3]. furthermore, with the advancement of nanotechnology, nanometer-scale materia ls are now used widely throughout the text iles, biomed ical, agricultural, industrial, e lectronics and energy generation fields. many studies have shown that nanoparticle addition provides an effective means of altering the mechanica l properties of compound materia ls, thereby improving the performance of e xisting products or paving the way for the develop ment of ne w ones [4-5]. poly mer co mpound mate rials have found e xtensive use in the sports equip ment fie ld. fo r e xa mp le , the rac kets used by table tennis players were orig inally made simp ly o f wood, and hence games we re played at slo w speed with a lac k of spin. in the 1920s, h o wever, eu ropean manufacturers attached a rubber skin to the bat; thereby enabling players to strike the ball with a far greater velocity and to e xert a higher degree of control over the ball trajectory [6-7]. in later years, japanese manufacturers replaced the rubber skin with innovative polymer co mpound materia ls; lead ing to a further significant imp rove ment in p layer performance [8-10]. the literature conta ins many investigat ions into poly mer co mpound materia ls and rubber mod ificat ion. ho wever, the modif ication o f poly mer co mposite materia ls fo r sporting applicat ions has thus far attracted relat ive ly little attention. accord ingly , the present study develops a new rubber materia l for table tennis rackets consisting of a mixture of carbon nanotubes (cnts), zinc oxide (zno) and titanium o xide (tio2) added to natural and synthesized rubber. the nano-re inforced rubber is attached to wooden table tennis paddles and patterned with four different surface structures, na me ly flat , long pimples, short pimp les and med iu m pimp les. the restitut ion coeffic ient and mechanica l prope rties (y ie ld stress, elastic modu lus and shear modulus) o f the four nano-reinforced rubbers are then investigated and compared with those of a flat non-reinforced rubber skin. * corresponding aut hor, email: yvonne@nfu.edu.tw advances in technology innovation , vol. 1, no. 2, 2016, pp. 41 45 42 copyright © taeti 2. experimental process 0.035 g cnts, 0.105 g tio2 and 0.175 g zno were added to a 35-g mixture of natural and synthesized rubber. the nano-re inforced rubber was glued to wooden table tennis blades and patterned with four diffe rent surface structures, name ly flat, short pimples, long pimp les and med iu m pimp les, as shown in figs. 1(a )~(d), respectively. (a) flat (b) short pimples (c) long pimples (d) medium pimples fig. 1 the different surface structures of nano reinforced rubber 2.1. restitution coefficient the restitution coefficients of the nano-rein forced rubber skins were evaluated in a wind-less environ ment using the e xperimental setup shown in fig. 2. in each test, a table tennis ball was placed at a height of 300 mm above the racket and was then dropped vertically onto the racket surface. the rebound height of the ball was recorded using a high-speed camera and the restitution coeffic ient of the rubber was then computed as height drop height rebound e . for each rubber, the restitution coeffic ient was calculated in three separate tests and then averaged to obtain a final representative value. fig. 2 experimental setup used for restitution coefficient testing 2.2. mechanical properties the mechanica l prope rties of the nano-rein forced rubber skins we re eva luated using the material testing system (m ts 810) shown in fig. 3. test specimens with dimensions of 150 mm x 3 mm x 1 mm (length x width x th ickness) we re p repared . each specimen was e xtended at a constant rate of 10 -1 s -1 until the point of fracture . the load and d isp lac e me nt v a lues we re me as u re d continuously during the test, and were then used to compute the elastic modulus and shear modulus of the rubber in accordance with basic engineering theory. the mt s system co mprised three components, namely: power unit: a hydraulic power system used to actuate the system. load unit: a stand-alone testing unit consisting of a load fra me , c rosshead lifts and locks, actuators, servo-valves, transducers and grip controls. advances in technology innovation , vol. 1, no. 2, 2016, pp. 41 45 43 copyright © taeti contr ol uni t: a control system used to coordinate and control the power unit and load unit. fig. 3 mt s system used for mechanica l p roperty testing 3. results and discussion 3.1. restitution coefficient fig. 4 shows the restitution coeffic ients of the four nano-re inforced rubbers. as e xpected, the restitution coefficient has a value of less than 1 for all five skins; indicating a non -fully -elastic collision between the ball and the racket. notably, the nano-re inforced rubbers all yie ld a slightly higher reinstitution coeffic ient than the non-reinforced skin. in other words, all four rubbers have a higher elasticity than the original skin. the performance imp rovement is particularly apparent for the three re inforced rubbers with pimples -out surface patterns. fig. 4 restitution coefficients of nano reinforced rubbers and non-reinforced rubber 3.2. stress-bearing capability fig. 5 shows the yield stress values of the nano-rein forced and non-reinforced rubbers. (note that the stress values indicate the ma ximu m stress recorded in the tensile tests, i.e., the stress at which specimen fa ilure occurred.) as shown, the flat nano-reinforced rubber has a ma ximu m stress of approximately 8.03 mpa. by contrast, the non-reinforced rubber has a ma ximu m stress of around 7.04mpa. in other words, the re inforced rubber has an imp roved stress -bearing capability, and thus provides a better wear resistance. notably, however, the pimples -out rubbers all have a lowe r stress -bearing capability than the non -reinforced rubber. the loss in strength is particularly apparent in the rubber with a long -pimp le structure. fig. 5 stress -bearing capabilities of nano reinforced rubbers and non-reinforced rubber 3.3. elastic modulus for each rubber, the elastic modulus was computed as ε = slope = ∆𝜎 ∆𝜖 = (𝜎2 − 𝜎1) (𝜖2 𝜖1⁄ )⁄⁄ . the corresponding results are shown in fig . 6. it is seen that the flat nano-re inforced rubber has an elastic modulus of approximate ly 1.8. for the non-reinforced flat rubber, the elastic modulus is equal to approximately 1.09. in other words, the addition of cnts, zno and tio2 is beneficia l in improving the stiffness of the rubber skin. however, the use of a pimp les -out surface pattern greatly reduces the rubber stiffness. for e xa mple , the nano-reinfo rced rubber with a long-pimple structure has an elastic modulus of just 0.13, i.e., a round 8 times lower than that of the non-reinforced flat rubber skin. advances in technology innovation , vol. 1, no. 2, 2016, pp. 41 45 44 copyright © taeti fig. 6 elastic modulus values of nano-reinforced rubbers and non-reinforced rubber 3.4. shear modulus for each rubber, the shear modulus was computed as g=e/2(1+v), where v is the poisson ratio (the values of the poisson ratio for the present rubbers is 0.45). as shown in fig. 7, the shear modulus of the flat nano-re inforced rubber (0.6) is around 50% higher than that of the flat non-reinforced rubber (0.4). in other words, the reinforced rubber has a significantly improved shear resistance. however, for all the pimples -out rubbers, the shear modulus is lowe r than that of the non-reinforced rubber. consequently, these skins are more prone to shear damage, and therefore fa il at a lowe r maximu m stress (see fig. 4). fig. 7 shear modulus values of nanoreinforced rubbers and non-reinforced rubber 4. conclusions this study has synthesized a new table tennis rubber consisting of natural and synthesized rubber reinforced with a mixture of ca rbon nanotubes (cnts), zinc o xide (zno) and titanium o xide (t io2). re inforced rubber skins have been attached to wooden table tennis paddles and patterned with four different surface structures, namely flat, long-pimple, short-pimple and medium-pimple. the restitution performance and mechanical properties of the various rubbers have been evaluated and compared with those of a flat non-re inforced rubber skin. the e xperimental results have shown th at the flat nano-rein forced rubber outperforms the non-reinforced rubber in terms of a higher restitution coeffic ient, a superior stress -bearing capability, and an improved stiffness. as a result, it provides several impo rtant practical advantages over the non-reinforced rubber, including a superior e lasticity and an imp roved wear resistance (i.e., a longer service life). the pimples -out reinforced rubbers provide a slightly higher elasticity than either of the two flat rubbers. however, the elasticity improve ment is obtained at the e xpense of significantly lowe r mechanica l properties. as a result, the pimp le-based coatings are less practical for real-world table tennis applications. acknowledgement the authors gratefully ac knowledge the e xperimental assistance provided to this study by professor s.c. lin of the depart ment of power mechanical engineering at nat ional formosa university, ta iwan. furthermo re, the preparation of the rubbers used in the present study by training co. ltd, taiwan, is also greatly appreciated. references [1] n. r. park, i. y. ko, j. m. doh, w. y. kong, j. k. yoon, and i.j. shon, “rapid consolidation of nanocrystalline 3ni-a l2o3 composite from mechanically synthesized powders by high freguency inductiion sintering,” materials characterization, vol. 61, no. 3, pp. 227-282, march 2010. [2] r. rit asa lo, x. w. liu, o. sod e rbe rg . a . ke s ki-ho n ko la , v. pit ka n e n , “the mic rost ructu ra l e ffects on the mech an ic a l and the rma l p rop e rt ies o f pu lsed e lect ric curren t. s inte red cu -a l2 o3 co mpos ites,” proced ia eng ine ering, vo l. 10, pp . 124-129, 2011. [3] g. z. zhao, l. s. d. s. zhang, x. feng, s. yuan, and j. zhou, “synergistic effect of nanobarite and carbon black fillers in natural rubber matrix,” materials and design, vol. 35, pp. 847-853, march 2012. advances in technology innovation , vol. 1, no. 2, 2016, pp. 41 45 45 copyright © taeti [4] m. jaya la kshmi, n. venugopal, k. phan i ra ja, and m. mohan rao, “ nano sno2 al2o3 mixed o xide and sno2al2 o3-carbon composite o xides as new and nove l electrodes fo r supercapac itor app lications,” journal of powe r sources , vo l. 158, pp. 1538-1543, august 2006. [5] c. l. arqu imedes, v. c. ad ilon, j. r. isaias, m. lilia, b. carrillo, and z. m . elv ira , “ synthesis of γ-al2o3 nanopowde r by the sol-gel method: effect o f diffe rent acid p recursors on the superfic ia l, mo rpholog ica l and structural p ropert ies,” journal of ce ra mic processing research , vol. 9, no. 5, pp. 474-477, 2008. [6] x. p. zhang and h. wu, “an e xpe rimenta l investigation into the in fluence of the speed and spin by ball of diffe rent dia mete rs and we ight,” science and rac ket sports, london, pp. 206-208, 1998. [7] y. féry and l. crognie r, “ on the tact ica l significance of game s ituations in anticipating ball tra jecto ries in tennis ,” research quarterly for e xe rc ise and sport, vol. 72, no. 2, 2001. [8] m. dic ks, k. dav ids, and c. button, “rep resentative task designs for the study of perception and action in sport. international journal of sport psychology,” vol. 40, no. 4, pp. 506-524, 2009. [9] r. a . pinder, i. renshaw, k. dav ids, and h. kerhe rve, “ princ ip les for the use of ball project ion mach ines in e lite and develop mental sport progra ms ,” sports medicine, vol. 41, no. 10, pp. 793, 2011. [10] g. t enenbaum, t . sa r-el, and m. ba r-eli, “anticipat ion of ball locat ion in lo w and high-skill performers : a deve lop menta l perspective,” psychology of sport and exe rc ise, vo l. 1, no. 2, pp. 117-128, october 2000.  advances in technology innovation , vol. 3, no. 2, 2018, pp. 9499 ball nut preload diagnosis of the hollow ball screw through support vector machine yi-cheng huang * , jing-hong zheng department of mechatronics engineering, national changhua university of education, changhua, taiwan . received 21 july 2017; received in revised form 17 sept ember 2017; accept ed 18 november 2017 abstract this paper studies the diagnostic results of hollow ball screws with different ball nut preload through the support vector machine (svm) process. the method is testified by considering the us e of ball screw pretension and different ball nut preload. svm was used to dis criminate the hollow ball screw preload status through the vibration signals and servo motor current signals. maximum dynamic preloads of 2%, 4%, and 6% ball screws were predesig ned, manufactured, and conducted experimentally. signal patterns with different preload features are separated by svm. the irregularity development of the ball screw driving motion current and rolling balls vibration of the ball screw can be discriminated via svm based on complexity perception. the experimental results successfully show that the prognostic status of ball nut preload can be envisaged by the proposed methodology. the smart reasoning for the health of the ball screw is available based on class ification of svm. this diagnostic method satisfies the purposes of prognostic effectiveness on knowing the ball nut preload status . keywords: ball nut preload, ball screw, support vector machine 1. introduction precision cnc machines are widely used in mod ern industry for mass production. recently, many strategies have been proposed to diagnose machine status, providing operators with important information to extending the machine’s useful lifespan. chen and lee [1] established a prognostic system to acquire and analyze vibration sign als corresponding to various ball screw states. calculated results are saved in a database, and a training mode l is established using several classification methods . since ball screws are widely used in linear actuators for various types of machinery and equipment. preloading is an effective means to eliminate the backlash and increase the stiffness of the ball screw for precision motion, thus maximizing efficiency [2]. preload loss leads to a lower natural frequency, lower stiffness, oscillatory positioning, and chance of rapid downtime in the manufacturing process. some methods proposed for tuning preload values are time consuming and incur increased downtime, thus raising the need to predict the ball screw nut preload status during machine operation. fault diagnosis with most acquired signals requires the assistance of conventional fourier transform or discrete wavelet transform in the frequency and time domains [3-4]. however, for many real applications, the signals are multi-components and are often corrupted by noise. as mentioned, loss of ball screw preload not only decreases the bandwidth of the frequency response spectrum but also reduces positioning accuracy. accordingly, industrial mass production applications would benefit from the lifetime prediction for ball screw preload loss status , but few studies have focused on this topic. the complexity of a ball screw with * corresponding author. e-m ail y chuang@ cc.ncue.edu.tw advances in technology innovation , vol. 3, no. 2, 2018, pp. 9499 copyright © taeti 95 preload in operation is highly nonlinear and non -stationary. traditional entropy measurements quantify only the regularity (predictability) of a time series on a single scale. no straightforward correlation exists, however, between regularity and complexity. this s tudy applies support vector machine (svm) [5-8] to classify the complexity of finite length time series. this computational tool has been applied successfully both to physical data sets, and can be used with a variety of measures of different patterns. the feature patterns of preload status are determined and abstracted via the function of svm. svm analysis is used to discriminate the prognostic preload status of ball screws in this preliminary study 2. dynamic model of the ball screw drive system to model the feed drive system, this paper set s different ball nut stiffness for different preload between the ball screw shaft and the ball nut. the presetting preload value can be deployed by inserting different ball size for single ball nut design or using disk spring that applied to the ball screw when double ball nut is the preference. fig. 1 shows the picture of the in-lab single-axis feed drive platform. to analyze the dyn amic characteristic of the ball screw system under different preload and varying table mass, the feed drive system is modeled by a lumped parameter system shown in fig . 2. fig. 2 is the schematic illustration for the single-axis ball screw feed drive system. in general, mechanical systems have three passive linear components. the spring and the mass are energy-storage elements, while the viscous damper is the dissipated energy. both of the rotational and translation mechanical system modeled below are actuated by the servo motor torque, indicated as t. the overall stiffness of a ball screw feed drive system can be determined by the stiffness of the ball screw itself, which is comprised of the ball screw shaft, th e ball nut, supporting bearings of the ball screw, and the stiffness between the ball screw and the working table [9]. xxrkxbxm tbbntttt )()0(   (1) 0)()0()0(  tbbnbebbbb xxrkxkx bxm  (2) ) θ(θk)] xxθ(r[krθqθj bmgtbbnbbbb   (3) tkqj bmgmmmm  )(   (4) rearranging eqs. (1)-(4), we have                                                                                                            t rr r e r m b b t gg gngn nnn nnn m b b t m b b t m b b t m b b t r 0 0 0 00 0 0 000 000 000 000 000 000 000 000 x x kk kkkkk kkkk kkk x x q q b b x x j j m m 2 n               (5) (a) the in-house single axis platform (b) the picture of the acceleration sensor attached to the ball nut fig. 1 photo of the in-lab single axis servo drive system and the acceleration sensor attached to the ball nut advances in technology innovation , vol. 3, no. 2, 2018, pp. 9499 copyright © taeti 96 fig. 2 illustration for the schematic diagram of the single-axis lumped parameters ball screw drive system 3. support vector machine conventionally, there are many possible linear classifiers that can be divided into two different featured data. svm provides a method that can maximize the margin in between two separated data. one of linear or nonlinear classifiers is termed the optimal separating hyperplane, as shown in fig. 3. intuitively, one would expect margin boundary to generalize well as opposed to the other possible boundaries . fig. 3 illustration for the seperating hyperplane for linear (left) and nonlinear (right) svm 4. experiment results fig. 4 linear svm kernel function for vibration signals by 5 minutes running with 2 % preload ball nut by pretension of 5μ and 20μ of the ball screw fig. 5 linear svm kernel function for vibration signals by 30 minutes running with 2 % preload ball nut by pretension of 5μ and 20μ of the ball screw advances in technology innovation , vol. 3, no. 2, 2018, pp. 9499 copyright © taeti 97 figs. 4-5 show the hyperplane divides pretension of 5μ (with + mark in red color) and 20μ(with * mark in green color) into two half-spaces when the table with 2 % ball nut preload was operated by 5 and 30 minutes, respectively. though using one sensor signals, the classification rate is not very high. nevertheless, it can be used as the initial decision boundary of a binary classifier when the cnc table does the repetitive backward and forward motion consecutively under 30 minutes. fig. 6 shows that deploying such linear decision boundary can’t classify the two pretensions by 60 minutes using svm’s linear kernel function. the average classification rate was only 79 %, 69 % and 48 % by running cnc table with 5, 30 and 60 minutes , respectively. since 2 % preload ball screw was treated as the preload loss condition while 4 % one was in the standard condition. after consecutive 30 minutes operation, the increasing temperature distribution of the ball screw causes the pretension effect loss and less mechanical complexities sensed by vibration sensor naturally. from figs. 4-6, the experimental results illustrate the thermo expansion around the ball nut and ball screw renders the diagnosis effect on the classification of the ball pretension vanished . fig. 6 linear svm kernel function for vibration signals by 5 minutes running with 2 % preload ball nut by pretension of 5μ and 20μ of the ball screw fig. 7 linear svm kernel function from vibration signals by 5 minutes running with 4 % preload ball nut by pretension of 5μ and 20μ of the ball screw research study here is aim to produce a classifier that can work well on different ball nut preload situations when they are applied by different ball screw pretensions. figs. 7-8 show the hyperplane divides pretension of 5μ (with + mark in red color) and 20μ(with * mark in green color) into two half-spaces when the table with 4 % ball nut preload was operated by 5 and 30 minutes, respectively. the linear classifier is termed the optimal separating hyperplane when the machine was operated by 30 minutes. the maximum margin (maximizes the distance between the 5μand 20μ) is in fig. 8 with 100 % average precision classification. experimental results show based on standard preload, the conventional warm-up procedure in industry can do the machine diagnosis with two different pretension on the ball screws by vibration signals. since different pretension creates different ball screw stiffness in axial direction, the svm can generate very good classification by standard operations. nevertheless, in fig. 9, after 60 minutes, the separating hyperplane generated by linear kernel function of svm can’t determine the pretension features by classification. the average precision classification was only 48 % and 54 % by running cnc table with 5 and 60 minutes , respectively. the reason was that due to the initial 5 minutes , the 4 % ball nut preload dominated the total stiffness of the ball screw. after 60 minutes operation, the increasing temperature of the ball screw renders the ball screw pretension effect diminished and the stiffness of the ball nut and the blurred vibration signals were loss thereof. fig. 10 shows using linear svm kernel function can’t separate the 2 % and 6 % ball nut preload features when the cnc table wa s running after 30 minutes for each ball screw pretension fixed at 5μ and 20μ. nevertheless, using nonlinear radial basis function (rbf) svm kernel function, the features patterns in fig. 11 can be separated. such promising results were based on the nonlinear kernel function. heuristically, one or more sensors can facilitate the svm classification and toward successfully. advances in technology innovation , vol. 3, no. 2, 2018, pp. 9499 copyright © taeti 98 fig. 8 linear svm kernel function from vibration signals by 30 minutes running with 4 % preload ball nut by pretension of 5μ and 20μ of the ball screw fig. 9 linear svm kernel function from vibration signals by 60 minutes running with 4 % preload ball nut by pretension of 5μ and 20μ of the ball screw (a) for separating 2 % and 6 % preload ball nut with pretension of 5μ of the ball screw (b) for separating 2 % and 6 % preload ball nut with pretension of 20μof the ball screw fig. 10 linear svm kernel function from vibration signals by 30 minutes running (a) for separating 2 % and 6 % preload ball nut with pretension of 5μ of the ball screw (b) for separating 2 % and 6 % preload ball nut with pretension of 20μ of the ball screw fig. 11 nonlinear rbf svm kernel function from vibration signals by 30 minutes running advances in technology innovation , vol. 3, no. 2, 2018, pp. 9499 copyright © taeti 99 5. concluding remarks this research uses the s ignal analysis techniques by svm to detect the preload status of hollow ball screw nuts. different ball screw pretensions and ball nut preload in a one axis machine tool feed drive table were studied. both the highly nonlinear processing svm methods and linear hyper plane were used to diagnos e the machinery health status . vibration measure was used to dynamically quantify the hollow ball screw’s complexity and identify the irregular development of the ball nut preload. experimental results show the sensed vibration signals of the preload features were extracted clearly with standard 4 % preload when the cnc table was operated by 30 minutes. compen sated positioning technique by ball screw pretension with 5μ and 20μ was not significantly contaminated by cnc 30 minutes operation time by linear svm prognostic diagnosis when the ball nut preload was 4 %. nonlinear rbf svm was able to separate the 2 %ball nut preload 6 % ball nut preload when the cnc table was operated by 30 minutes. heuristically, with more sensors, the svm can facilitate the bettering classification and toward high precision classification. future work will combine the vibration signals and motor current signals as new training data through the support vector machine process . acknowledgement this work was supported by most grant 106-2218-e-018-001 for which the authors are very much grateful. references [1] y. chen and j. lee, “methodology for ball screw component health assessment and failure analysis ,” proc. the asme international science and engineering conference, vol. 2, pp. v002t02a031-1v002t02a031-9, 2013. [2] h. shimoda, “stiffness analysis of ball screws -influence of load distribution and manufacturing error,” international journal of the japan society for precision engineering, vol. 33, no. 3, pp. 168-172, november 1999. [3] p. c. tsai, c. c. cheng, and y. c. hwang, “ball screw preload loss detection using ball pass frequency ,” mechanical systems and signal processing, vol. 48, no. 1-2, pp. 77-91, october 2014. [4] c. g. zhou, h. t. feng, z. t. chen, and y. ou, “correlation between preload and no -load drag torque of ball screws,” international journal of machine tools & manufacture, vol. 102, pp. 35-40, march 2016. [5] s. r. gunn, “support vector machines for classification and regression,” univ. of southampton, faculty of engineering, science and mathematics school of electronics and computer science , technical report, may 10, 1998. [6] a. smola and s. v. n. vishwanathan, introduction to machine learning, cambridge university press , 2008. [7] c. cortes and v. m. vapnik, “support-vector networks ,” machine learning, vol. 20, no. 3, pp. 273-297, september 1995. [8] s. j. wang, a. mathew, y. chen, l. f. xi, l. ma, and j. lee, “empirical analysis of support vector machine ensemble classifiers,” expert system with applications vol. 36, no. 3, pp. 6466-6476, april 2009. [9] y. c. huang and x. y. chen, “investigation of a ball screw feed drive system based on dynamic modeling for motion control, ” advances in technology innovation, vol. 2, no. 2, pp. 29-33, 2017.  advances in technology innovation , vol. 2, no. 2, 2017, pp. 46 50 46 assessment of arteriovenous shunt pathway function and hypervolemia for hemodialysis patients by using integrated rapid screening system wei-ling chen 1 , chia-hung lin 1 , chung-dann kan 2,* 1 department of engineering and ma intenance, kaohsiung veterans genera l hospital, kaohsiung, taiwan. 2 division of cardiovascular surgery, department of surgery, national cheng kung university hospital, college of medicine, national cheng kung university, tainan, taiwan. received 06 january 2016; received in revised form 25 march 2016; accept ed 01 april 2016 abstract currently, the hemodialysis patients received body weight measurement by themselves, vita l sign checking by nursing staffs before dia lysis. whenever, the arteriovenous routes with problems doubted, the patients needed to be referred to surgeon for vascular echography checking and then to be corrected. how to integrate these three tasks in one time is a very important issue. the project proposes to combine our previous study of audio-phono angiographic technology in detecting vascular stenosis with rapid screening system to evaluate dialysis patients’ arteriovenous routes function and their status of e xcess body fluids: inspecting and integrating the blood pressure, body weight, and fistula function work into a rapid screening system, and using the quantization of fistula phono angiography pitch to achieve assessing arteriovenous routes. future hoping is developed a complete integrated intelligence system by combining the arteriovenous fistula signal processing with feature extraction with wireless sensor network technology. keywor ds : arteriovenous shunt, screening system, hypervolemia, dual-core embedded system 1. introduction chronic kidney disease is a global public health proble m with high mo rbidity and mo rtality. treat ment of end-stage renal disease (esrd) typically involves a kidney transplant or dialysis, which assists damaged kidneys in removing waste and excess fluid fro m the body. in taiwan, the incidence and prevalent rates of esrd we re the highest in the wo rld [1], he modia lysis is the most common choice for esrd patients. ho wever, to perform a he modia lysis, surgeons must create pathological fistulas to provide vascular access routes for treating esrd. arteriovenous access (ava) stenosis is regarded as the prima ry cause of a va dysfunction in he modia lysis patients and a co mmon occurrence in patients undergoing extended hemodia lysis therapy. according to nkf-doqi guide lines [2], regula r monitoring and surveillance of a va function is mandatory. basically, a va is a continuous circuit that starts and ends at the heart; it is not simp ly an anastomosis. moreover, clin ical assessments are based on a physical e xa mination of ava. th ree physical exa mination steps are look, feel, and listen. clin ically, a va flow rates must reach 600– 1000 ml/ min for he modia lysis treatment to be considered efficient. this high-flo w volu me causes vibration of the vascular wa ll, which then transmitted to the skin surface, manifesting as a palpable thrill o r aud ible bruit [3]. the state of body water is an important factor in the routine hemodialysis (hd) treatment and post-dialysis healthcare. hypervolemia is a medica l condition as excess body water in the blood, leading to increases in body sodium (na) content and a consequent increase in e xt ra-ce llular body water. when dia lysis patients suffer fro m hypertension will increase in we ight and peripheral ede ma in the legs and arms [4-6]. the purpose of this paper is to combine our previous study of audio-phonoangiographic technology in detecting vas cular stenosis with rapid screening system to evaluate dialysis * corresponding aut hor, email: kcd56@mail.ncku.edu.tw advances in technology innovation , vol. 2, no. 2, 2017, pp. 46 50 47 copyright © taeti fig. 1 block diagram of the rapid screening system patients’ arteriovenous routes function and their status of excess body fluids: inspecting and integrating the blood pressure, body weight, and fistula function work into a rapid screening system, and using the quant izat ion o f fistu la phonoangiography pitch to achieve assessing arteriovenous routes. the feasibility of the integrated rapid screening system will be provide an easy operational instrument for efficacious and real time monitoring arteriovenous routes (fig.1). first, the screening system employed national instruments (ni) my rio dual-core embedded microprocessor system integrated with system sensors, software, wireless/wired communications, and output display unit in developing a system which is suitable for rapid screening on hemodialysis units. second, the screening system is to use audio signal feature extraction algorithms, fuzzy algorithms, and excess body fluid assessment methods on starting begin human status checking. using vascular echography results as of the narrow criterion to establish the judgment database, and create an alert threshold as indicators of future development, therefore, to achieve the best diagnostic work. future hoping is developed a complete integrated intelligence system by combining the arteriovenous fistula signal processing with feature e xtraction with wire less sensor network technology. 2. method 2.1. preliminary diagnosis and classification in clinica l research, the degree of narro wing of norma l vessels has been used as an index for the degree of ava in patients . exa mination results have been used as a reference to confirm the dos according to x-ray or sono images . for the measurement segment a site-e site (fig. 2), the dos is defined as follows [7-9]. (a) the detection site (b) sono image of an abnormal blood vessel (c) spectrum of short-time fourier transform (d) spectrum of fourier transform fig. 2 sono angiography and spectrum analy sis in clinical reasearch advances in technology innovation , vol. 2, no. 2, 2017, pp. 46 50 48 copyright © taeti %100)1(% 2 2  d d dos (1) %)%(% postpre dosdosdos  (2) 2.2. design of fuzzy petri nets (fpns)screening system based on fractional-order dynamic error regarding the surgical outcomes of the 42 p atients, the overall pre-pta dos% was above 80%, which was regarded as the reference level, whereas the post-pta dos% was distributed across three groups (i.e., > 50%, 30%–50%,<30%). no statistical significance was observed between the groups at the a sites and v sites (p > 0.05). therefore, the correlation between dos% and index  is stronger at the v sites than at the a sites [7]. exponent regression was used to model the relationship between dos% and index  and between dos% and index . the prediction model fit a nonlinear curve passing directly through all of the experimental data, as shown in fig. 3(a). the correlation between dos and index  at the v site can be expressed as: dos% = 0.1914  exp(0.2333), r 2 = 0.3802 (3) (a) dos % versus index  at the v site (b) evaluation of the severity of ava stenosis fig. 3 evaluation of the severity of ava stenosis at the v site with dos % versus index exponential regression was used to perform a least squares curve fit, which min imized the sum of the squares of the deviations of the e xperimental data values e xtracted fro m the prediction model, thereby obtaining the e xa min ation criteria for the predict ion. fig. 3(b) depicts the range of pre-pta dos residuals ( class i:  < 3, class ii: 3 <  < 5, and class iii:  > 5), which were used as a baseline for eva luating the post-pta dos residuals . we also compa red the preand post-pta  indices monthly and determined the ideal ranges for evaluating the severity of a va stenosis in the order of class iii ( < 3), class ii (3 <  < 5), and (class i:  > 5). (1) a gaussian me mbe rship function can be parameterized by mean (mean = 0– 6) and standard deviation (1 = 2 = 3 = … = r = 0.3802) by applying equation (4). ] )( exp[ 2 2 r r r mean      (4) (2) fig. 4 depicts the gaussian membership functions for the three classes. through this approach, we obtained seven me mbership functions r, r = 1, 2, 3, …, 7, with specific ranges denoted as r. the cf of each input in the various ranges is within the range of [0,1]. the fpn can perform fuzzy inference calculations to evaluate the dos for each propos ition specified by a clinical physician. assume that for the degree of proposition cm (class es i–iii), m = 1, 2, 3, and place pm is associated with the proposition dm = m(pm), m = 1, 2, 3.. fig. 4 gaussian me mbership functions for the three classes of dos advances in technology innovation , vol. 2, no. 2, 2017, pp. 46 50 49 copyright © taeti 3. results and discussion 3.1 feasibility tests with the proposed screening system fig. 5(a) and (b ) show the overall test results ; compared with the dos% results , the accuracy is 85.71% with six failures. the measurement sites, quantification errors, and undetected stenosis could affect the effic iency of the proposed method. the study used at least two 8-second records from the measurement site of 21 patients (i.e ., pre and post-pa). we determined three dos classes. the terms ul, l, and vl corre spond to monotonically decreasing curves that define the degree of ce rtainty. the output of function m may be an arb itrary curve that can be defined as a function that must vary between 0.3679 and 3.000. the place pm can determine the dos level, and more likely c loses to value 1, 2, or 3 for the goal proposition. if p lace pm is only partially simila r, its value is less than 1 and it gradually decays to 0. fro m these results, we determined the function of a vs, wh ich can be evaluated using the proposed diagnosis system. (a) feasibility diagnosis results for the 42 tests (b) degree of stenosis (dos) versus pm fig. 5 feasibility diagnosis results for the 42 tests, and the degree of stenosis (dos) versus pm due to variations in the frequency spectra of the various classes and differences a mong the patients, the characteristic frequencies occupy a range of frequency bands, with some characteristic frequencies overlapping or crossing bands. phys icians determined the final degree of certainty as a function of the variance in frequency and magn itude. however, obtaining the diagnosis results required an off-line analysis. traditional petri nets (pns) require constant transitions and weighted parameters ( and w), and they encounter difficulty in handling variance in the frequency spectra. pns were considered appropriate for processing binary data in t rue–false decisions, on–off switching, and automatic control applications [10]. co mparing fpns with pns, the inference rules in the knowledge base of t he rule-based system are both modeled. fpns use the cf of the fuzzy infe rence rules and weights of the propositions by using binary data and can automatically perform we ighted fuzzy reasoning calculations for analyzing spectral variance. thus, the proposed rule-based diagnosis system can perform fu zzy infe rence calculat ions in a flexible and intelligent manner. 3.2. long-term examination using frequency parameter-based fuzzy petri net a 54-year-old fe male patient undergoing hemodia lysis treatment with an a vg (right forearm loop) agreed to part icipate in a long-term e xa mination. data were collected between june 25, 2011, and july 11, 2012. in a routine monitoring cycle, monthly e xa minations were perfo rmed to evaluate avg function. over 3 months of observation, the first characteristic frequency gradually increased fro m 68 to 170 hz (fig. 6). on september 6, 2011, the patient presented with a severe avg occlusion, and a physician confirmed class ii stenosis . ultrasonic e xa mination indicated a dos% of 66% at the measurement site, and the patient received pta treatment. the three characteristic frequencies were 170, 425, and 681 hz. fig. 6 long-term examination at v site from june 25, 2011, to september 6, 2011 advances in technology innovation , vol. 2, no. 2, 2017, pp. 46 50 50 copyright © taeti 4. conclusion in an embedded system development env ironment, we used four fundamental arithmet ic and logical-reasoning operations to configure the combinational logics or asics, and then embedded the intelligent algorithms into a compact chip. the proposed diagnosis system allo ws rule-based configuration and automatic weighted fuzzy reasoning calculations. fle xib le and intelligent algorith ms require no iteration for updating system we ights. therefore, this system can handle comple x configuration designs and is appropriate for the short design cycle of prot otype imp le mentation, testing, debugging, and modification. currently, graphical user interfaces of windows-based applications enable reprogramming and have flexible architectures that facilitate the rapid development of customized asics. the proposed fpn algorithm can be implemented using four fundamental operations with specific data structures. therefore, it can also be implemented with the four fundamental operations. by combining an fpn and logical operation functions, the proposed diagnosis system provides a promising means for implementing a portable monitor for avs evaluation in home care. acknowledgement this study is supported in part by the research grant of kaohsiung veterans genera l hospital (vghks105-070) and the ministry of science and technology, taiwan (most), under contract number: most 105-2218-e-075b -001. references [1] a. j. collins et al., “ usrds 2012 annual data report: atlas of chronic kidney disease and end-stage renal disease in the united states ,” american j. kindeny diseases, vol. 59, pp. 342, 2012. [2] n. k. foundation, “kdoqi clinical practice guidelines and clinical practice reco mmendations for 2006 updates: he modialysis adequacy, peritoneal dia lysis adequacy and vascular access,” am. j. kidney dis., vol. 48, pp. 1-322, 2006. [3] f. loth, p. f. fischer, and h. s. bassiouny, “blood flow in end-to-side anastomoses,” annu. rev. fluid mech., vol. 40, pp. 367-393, 2008. [4] p. w. cha mney, k. matthias, c. rode, w. kle inekofort, and v. w ize mann, “a new technique for establishing dry we ight in hemodia lysis patients via whole body bioimpedance,” kidney international, vol. 61, pp. 2250-2258, 2002. [5] a. h. t za malou kas, g. h. murata, d. j. vanderjagt, k. s. serv illa, and r. h. glew, “body composition evaluation in peritoneal dia lysis pat ients us ing ant hro po metric formulas estimat ing body water,” advances in pe ritoneal dialysis, vol. 19, pp. 212-216, 2003. [6] r. agarwal, “hypervolemia is associated with increased mortality among hemodialysis patients,” hypertension, vol. 56, pp. 512-517, 2010. [7] y. c. du, w. l. chen, c. h. lin, c. d. kan, and m. j. wu, “residual stenosis estimation of arteriovenous grafts using a dual channel honoangiography with fract iona l-o rder features,” ieee journal of bio medica l and health informat ics, vol. 18, no. 2, pp. 703-713, 2014. [8] w. l. chen, c. h. lin, t. chen, p. j. chen, and c. d. kan, “phono angiography with a fractional order chaotic system a new and easy algorithm in analyzing residual arteriovenous,” medical & biological eng & computing, vol. 51, pp. 1011-1019, 2013. [9] w. l. chen, c. h. lin, t. chen, p. j. chen, and c. d. kan, “stenosis detection using burg method with autoregressive model for hemodialysis patients,” journal of medical and bio medical engineering, vol. 33, no. 4, pp. 356-362, 2013. [10] r. s. lees and c. f. de wey, “phono angiography: a new noninvasive diagnostic method for studying arterial d isease,” proceedings of the national academy of sciences, vol. 67, no. 2, pp. 935-942, 1970.  advances in technology innovation , vol. 3, no. 3, 2018, pp. 141 146 a novel stretch sensor to measure venous hemodynamics syrpailyne wankhar 1,* , albert a. kota 2 , dheepak selvaraj 2 1 department of bioengineering, christian medical college and hospital, vellore, india . 2 department of vascular surgery, christian medical college and hospital, vellore, india. received 21 july 2017; received in revised form 10 oct ober 2017; accept ed 23 oct ober 2017 abstract chronic venous insufficiency is a debilitating condition causing varicose veins and venous ulcers. the pathophysiology includes reflux and venous obstruction. the diagnosis is often made by clinical examination and confirmed by venous doppler studies. plethysmography helps to quantitatively examine the reflux and diagnose the burden of deep venous pathology to better underst and venous hemodynamics, which is not elicited by venous duplex examination alone. however, most of these tests are qualitative, expensive , and not easily available. in this paper, we demonstrate the potential use of a novel stretch sensor in the assessmen t of venous hemodynamics during different maneuvers by measuring the change in calf circumference. we designed the stretch sensor by using semiconductor strain gauges pasted onto a small metal bar to form a load cell. the elastic and velcro material attach ed to the load cell form a belt. it converts the change in limb circumference to a proportional tension (force of distension) when placed around the calf muscle. we recorded the change in limb circumference from arrays of stretch sensors by using an in-house data acquisition system. we calculated the venous volume (vv), venous filling index (vfi), ejection fraction (ef) and residual venous volume (rvv) on two normal subjects and on two patients to assess venous hemodynamics. the values (vv > 60 ml, vfi < 2ml/s, ef > 60%, rvv < 35%) in normal subjects and (vv < 60 ml, vfi > 2ml/s, ef < 60%, rvv > 35%) in patients were comparable to those reported in the literature . keywords: stretch sensor, venous hemodynamic, plethysmography, venous duplex 1. introduction the pathophysiology of reflux and venous obstruction due to chronic venous insufficiency is a debilitating condition causing varicose veins and venous ulcers. the diagnosis is often made by clinical examination and confirmed by venous doppler studies [1-2]. there are three systems in the venous circulation of the lower extremity namely superficial, deep veins , and perforators. normal physiology allows blood to flow from superficial to deep veins via perforators and then returns the blood to the heart in unidirectional flow. when there is damage to the valves, blood can flow bidirectionally, causing the volume overload in the lower extremities leading to the pathological changes. the goal of further evaluation is to characterize the misdirect ion of venous blood volume [2-3]. it is , therefore, crucial to study the venous hemodynamic, which is not elicited by venous duplex examination alone. plethysmography helps us to quantitatively examine the reflux and diagnose the burden of deep venous pathology (degree of venous obstruction, ejection capacity of calf muscle pump). different plethysmographic measurement of venous reflux includes strain -gauge plethysmography (sgp) using silastic conductor tube which is stretched by a change in calf circumference. this , in turn, increases the resistance resulting in a voltage * corresponding author. e-m ail address: sy rpaily ne@ cm cvellore.ac.in advances in technology innovation , vol. 3, no. 3, 2018, pp. 141 146 copyright © taeti 142 change which is acquired for further analysis. the disadvantage of this technique is that it is susceptible to degrade over time resulting in the change in conductivity of the silastic conductor. impedance plethysmography (ipg), on the other hand, results in a change in resistance due to change in limb circumference which is directly proportional to relative volume change [4]. the disadvantage of this method is that it is cumbersome as it is sensitive to sen sor placement. photoplethysmography (ppg) is another method which uses a light source in the infrared wavelength to emit light into the tissue. the light is absorbed by t he blood and whatever is reflected or transmitted is measured by a reflective/transmis sive based photo detector, and net absorption is displayed as a line tracing [4]. the disadvantage of this method is that it is also sensitive to sensor placement. dielect ric elastomer described in [5], uses a flexible capacitor with stretchable dielectric attached to 2 parallel plate electrodes. when the electrodes are pulled uniaxially or biaxially, it changes the area and thickness of the sensor (stretch mode) resulting in the change in capacitance. one of the major drawbacks of this sensor is that the a ccuracy is affected by speed and, therefore, the capacitive based stretch sensor requires good circuit design to achieve better accuracy [5]. the ‘stretchsense’, based on this principle is commercially available. however, it is expensive. air plethysmograp hy (apg) is considered as the gold standard. it uses a 30-40 cm length air cuff that is attached to the lower leg from knee to ankle. the low-pressure cuff provides a precise quantitative evaluation of volume changes of the entire lower leg [4]. although apg is considered the gold standard, it is not readily available because it is expensive. as discussed , there are different techniques available such as ppg, ipg, apg etc. however, most of these tests are qualitative, expensive, and not easily available [4-9]. in this paper, we demonstrate the measurement of venous volume using arrays of the novel stretch sensor and thereby enable us in developing a novel prototype instrument for measuring venou s hemodynamic. 2. design methodology end-supported-centre loaded load cell was constructed using semiconductor strain gauges (bcm strain g auges b2-120-p-3) and an aluminum bar (31 mm x 5 mm x 3 mm) as described by wankhar et al. [9]. a pair of strain gauges pasted on the top face and bottom face of the beam will experience tension and compression respectively. these strain gauges along with fixed resistors form a wheatstone bridge to give an output voltage proportional to the bending of the beam. the output of the strain gauge bridge connects to a dc bridge amplifier for further processing. eq. (1) is the standard form of the strain on the surface of the beam as measured by the strain gauges for centre-loaded end supported beam: 2 4 l fl l e bh   (1) 3.1. static calibration of the stretch sensor static calibration of the overall stretch sensor consisting of the load cell attached to the velcro (rigid) and elastic (spring) material was done. offset correction was done before measurements were taken. the change in length due to stretching of the elastic band was recorded by using an in-house data acquisition system ‘cmcdaq’. the setup for measuring venous volume by measuring the change in calf circumference using arrays of 3 stretch sensors is shown in fig. 1. this is a novel prototype instrument for quantifying venous function. the total volume measured by the stretch sensor is calculated using eq . (2): 2 2 2 1 1 1 2 2 2 3 3 3 4 (2 2 2 ) h v c c c c c c c c c               (2) where, h = height of the stretch sensor belt, c = change in circumference measured by the stretch sensor advances in technology innovation , vol. 3, no. 3, 2018, pp. 141 146 copyright © taeti 143 fig. 1 setup for measuring venous volume by measuring change in calf circumference using arrays of stretch sensors 3.2. measurement of venous volume using the novel stretch sensor for measuring venous volume, the subject performed different maneuvers as described in [1, 10]. at the beginning of the experiment, the subject was asked to lie down in a supine position. the subject’s leg to be assessed was lifted to 45° angle for about a minute so that the veins are emptied and then the leg was placed back in the initial position. this was done to lower the venous pressure in the legs to a zero line (pressure) of the right atrium (≈ 0 mmhg) which was recorded as baseline volume us ing an in-house data acquisition system [10]. then the subject was asked to stand slowly and load on the non -assessed leg. the assessed leg was in non-weight bearing position so that the leg venous pressure increases due to the hydrostatic column of blood extending from the right atrium to the assessed leg [10]. the blood volume increases as the veins are elastic and this change in volume was recorded. fig . 2 illustrates the different parameters as measured by the novel stretch sensor during different maneuvers. the measurement between baseline volume in the supine position (‘1’) and the erect volume plateau is known as venous volume (vv'). we also determined the venous refilling time (vft90), which is the time from which the volume increases from baseline (‘2’) to 90% vv and venous filling index (vfi) was calculated using eq . 3(a). the volume then increases till it plateaus (‘3’). during this time, the subject was in a standing position and supported only on the non -assessed leg. then, the subject was asked to weight bear both legs and performed a single plantar flexion (‘4’). the subject then unloads the assessed leg (non-weight bearing stance, ‘5’) again. this produced a change in the volume called the ejection volume (ev). ejection fraction (ef) was then calculated using eq . 3(b). now the subject was asked to perform the same maneuver (plantar flexion) 10 times (‘6’), which caused a larger volume reduction. the residual volume (rv) was measured as the difference between the volume after 10 plantar flexions and the baseline v olume. the residual volume fraction (rvf) was calculated by dividing rv by vv as given by eq. 3(c). fig. 2 change in venous volume during different maneuvers measured by the stretch sensors advances in technology innovation , vol. 3, no. 3, 2018, pp. 141 146 copyright © taeti 144 the equations for calculating vfi, ef, and rvf are as follows : 90 90% vv vfi vft  (3a) 100 ev ef vv  (3b) 100 rv rvf vv  (3c) 3. results 3.1. static calibration of the stretch sensor the overall stretch sensor is non-linear in its characteristics because of the properties of elastic band attached to the belt as shown by the static calibration curve in fig . 3. fitting a linear equation to the data by linear regression from a single sensor is shown in fig. 3 (a). eqs . (4) and (5) provide the relationship between force and change in length of the stretch sensor belt. 0.915 0.1065y x  (4) with r 2 = 0.9652 3.1445 7.0357y x  (5) with r 2 = 0.9972, where y is force and x is change in length. similarly, linear equation to the data by linear regression from all other sensors shown in fig. 3 (b) is given by eqs . (6), (7), (8), and (9) for sensors 2 and 3 respectively. 0.9142 0.1155y x  (6) with r 2 = 0.9418 3.1928 6.9242y x  (7) with r 2 = 0.9982 0.9274 0.1118y x  (8) with r 2 = 0.9555 3.0342 6.4521y x  (9) with r 2 = 0.9976, where y is force and x is change in length. for all applications, the stretch sensor is pre-loaded/stretched to a length change of 2 cm or more, indicating the region of operation is in the second part of the curve with r 2  0.9972 (fig. 3(a)). (a) static calibration curve of one stretch sensor (b) static calibration curve of all 3 stretch sensors fig. 3 static calibration curve of one stretch sensor and all 3 stretch sens ors advances in technology innovation , vol. 3, no. 3, 2018, pp. 141 146 copyright © taeti 145 3.2. measurement of venous volume using the novel stretch sensor we calculated venous volume (vv), venous filling index (vfi), ejection fraction (ef) and residual venous volume (rvf) on two normal subjects (1male, 1female) and two patients (2 males). fig. 4 shows the changes in venous volume during different maneuvers measured by the stretch sensors for venous hemodynamic assessment on one normal subject. the calculated parameters were vv = 100 ml, vfi = 1 ml/s, ef = 84.4%, and rvf = 14.9% in a male subject, and vv = 65 ml, vfi = 0.5 ml/s, ef = 61%, and rvf = 29.7% in a female subject. similarly, fig . 5 shows the changes in venous volume during different maneuvers measured by the stretch sensors on one patient. the calculated values were vv = 51 ml, vfi = 4.6 ml/s, ef = 40%, and rvf = 41%, and vv = 55 ml, vfi = 5.6 ml/s, ef = 58%, and rvf = 38% in 2 male patients respectively. the measured parameters that quantify venou s function on 2 healthy subjects (vv > 60 ml, vfi < 2 ml/s, ef > 60 %, rvf< 35%) and 2 patients (vv < 60 ml, vfi > 2 ml/s, ef < 60%, rvf> 35%) were comparable to those reported in the literature [3, 7-8, 11]. fig. 4 changes in venous volume during different maneuvers measured by the stretch sensors on a normal subject (vv=100 ml, vfi=1 ml/s, ef=84.4%, rvf = 14.9%) fig. 5 changes in venous volume during different maneuvers measured by the stretch sensor on a patient (vv=51 ml, vfi=4.6 ml/s, ef=40%, rvf=41%) 4. discussion fig. 4 and fig. 5 show noticeable differences in the waveform which is evident from the calculated values of vv, vfi, ef, and rvf. however, the very distinctive region in the waveform is also the regular peak seen with each tip toe during repeated pla ntar flexion in a normal subject and random peaks in a patient. the ra ndom peaks with each tip toe may be due to the inability of the calf muscle to contract normally because of the pathological changes in the venous system of the lower extremities. this also means that there is an indirect effect of the deep veins which are entirely encased within the muscle and bone. since the calf muscle (pump) is not capable of functioning normally, this would result in the pooling of blood as evident from the data (rvf > 35%) shown in fig. 4. in the normal situation, when the calf muscle contracts, the pressure in the deep veins increases to nearly 250 mmhg allowing the ejection of blood to the heart uni-directionally [12]. this is possible since the ejection fraction of the calf muscle pump is 65 % as compared to thigh pump which is only 15 % [12]. in our study, the venous duplex was also performed on patients and it showed (i) increased superficial reflux, (ii) no deep venous reflux, (iii) no deep venous thrombosis and (iv) increased perforator reflux. the increased perforator reflux cou ld possibly be attributed to valvular incompetence in the perforators. this would further result in increased superficial reflux due to retrograde flow from the perforators to the sup erficial veins when the pressure in the deep veins increases during muscle contraction (vfi = 4.6 ml/s, minor reflux) in one of the patient [11]. in summary, we demonstrated the assessment of venous hemodynamic by using the stretch sensors, which has not been reported in the literature. the belt can be used with great ease besides; it can provide continuous information about the dynamic movement or changes in limb circumference over time. this is desirable in a clinical setting that requires simple and fast means of assessment as well as meaningful information. advances in technology innovation , vol. 3, no. 3, 2018, pp. 141 146 copyright © taeti 146 5. conclusions commercially available air plethysmograph (apg) which is used as a standard tool for assessment of venous hemodynamic is very expensive for a low income and developing country like india. we have shown the possibility of measuring venous volume and therefore, assessment of venous hemodynamic by using arrays of the novel stretch sensors along the calf muscle. although the result looks promising, more data collection is required for validating the novel device for its applications in clinical settings . references [1] r. a. oliveira, n. barros jr., and f. miranda jr., “variability of venous hemodynamics detected by air plethysmography in ceap clinical classes,” jornal vascular brasileiro, vol. 6, no. 4, pp. 359-365, december 2007. [2] c. wittens, a. h. davies, n. bæ kgaard, et al., “editor’s choice management of chronic venous disease: clinical practice guidelines of the european society for vascular surgery (esvs),” european journal of vascular and endovascular surgery g, vol. 49, no. 6, pp. 678-737, june 2015. [3] v. ibegbuna, k. t. delis , and a. n. nicolaides, “haemodynamic and clinical impact of superficial, deep and perforator vein incompetence,” european journal of vascular and endovascular surgery g, vol. 31, no. 5, pp. 535-541, may 2006. [4] a. n. nicolaides and d. s. sumner, investigations of patients with deep vein thrombosis and chronic venous insufficiency , la: med-orion publishing, january1991. [5] b. o’brien, t. gisby, and i. anderson, “stretch sensors for human body motion,” proc. spie, vol. 9056, pp. 905618-1-905618-9, 2014. [6] s. raju, “new aproaches to the diagnosis and treatment of venous obstruction,” journal of vascular surgery, vol. 4, no. 1, pp. 42-54, july 1986. [7] e. criado, m. a. farber, w. a. marston, et al., “the role of air plethysmography in the diagnosis of chronic venous insufficiency,” journal of vascular surgery, vol. 27, no. 4, pp. 660-670, april 1998. [8] b. b. lee, a. n. nicolaides, k. myers, et al., “venous hemodynamic changes in lower limb venous disease: the uip consensus according to scientific evidence,” international angiology, vol. 35, no. 3, pp. 236-352, june 2016. [9] s. wankhar, a. a. kota, and d. selvaraj, “a versatile stretch sensor for measuring physiological movement using a centre loaded, end supported load cell,” journal of medical engineering and tec hnology, vol. 41, no. 5, july 2017. [10] j. j. bergan and n. bunke-paquette, the vein book, 2nd ed., usa: oxford university press , 2014. [11] m. l. katz, a. j. comerota, and r. kerr, “air plethysmography (apg): a new technique to evaluate patients with chronic venous insufficiency,” the journal of vascular technology, vol. 15, pp. 23-27, 1991. [12] m. h. meissner, “lower extremity venous anatomy,” seminars in interventional radiology, vol. 22, no. 3, pp. 147-156, 2005.  advances in technology innovation , vol. 2, no. 2, 2017, pp. 56 60 56 the research of infrared thermal image diagnostic model for building external wall tiles li-wei chiang * , sy-jye guo department of civil engineering, national taiwan university, taipei 106, taiwan, roc. received 22 february 2016; received in revised form 29 april 2016; accept ed 04 may 2016 abstract this study focuses on a deterioration diagnostic standard and inspection time for e xterna l wall t iles. the study was carried out using a tap tone test, followed by fast fourier transform and pattern recognition to analyse the tapping results. based on the test results, the study recommends that for frequencies between 200 and 800 hz on the spectrogram, it can be deduced that a cavity is present in the tiles. the study was carried out using infrared thermal imag ing record to det eriorat ion and norma l t ile position. this study observed temperature changes per hour, it is obvious temperature diffe rence with deteriora tion tiles and norma l tiles between 09: 00-11: 00 (east), 10:00-14:00 (west), 10:00-12:00 (south) and 11:00-13:00 (north), the study was calculated that the average temperature difference of 1.0 degrees. this study suggests that the best e xperimental inspection time is 09: 00-14:00. exclude the impact of e xte rnal factors, if the temperature d ifference more than 1.0 degrees, it can be deduced that a cavity is present in th e t iles, and this can be used as basis for determining tile deterioration. keywor ds : external wall tiles, infrared thermal image inspection, tap tone inspection, temperature difference, inspection period 1. introduction in taiwan, there has experienced frequent safety incidents involving the e xternal wa lls of buildings recent years. the most common incidents include falling tiles, falling advertising signage and falling e xposed pipelines, all seriously impacting the surrounding environment and public safety. on march 14, 2015, a tile falling inc ident in taipei city, taiwan, resulted in a fatality. according to the taipei city go vernment statistics, there were 1188 buildings listed for control as important cases with spalling wa ll tiles. the govern ment of kaohsiung city, taiwan investigated the 6492 apart ment buildings with six or more floors in the city. there we re 859 buildings with t iles spalling. kaohsiung city government listed 367 buildings as important cases for control. according to research data (chang, 2008), construction in taiwan peaked in the years 1981 and 1994. according to this research, taiwanese buildings e xperience significant deteriorat ion after 30 years. safety check require ments for taiwanese buildings should thus peak in 2011 and 2024. integrating news relating to fatalities in tile-fa lling incidents in taiwan, this research sees an immed iate need for a co mp lete diagno stic mechanism and preventive method for the further occurrences of such public safety incidents. to effectively drive external wall replacement safety checks and studies on evaluation systems, the industry requires relevant emp irica l results and evaluation criteria. th is study aims to e mploy data collection and e mp irical research to provide a basis of reference and an evaluation method for use in external wall tile diagnosis. 2. method 2.1. tap tone method this study used s ounding diagnosis to comple ment the visual diagnosis method. the tap tone diagnostic method is the most convenient, economical and wide ly used evaluation method for inspecting surface defects. tile inspection is norma lly co mbined with other inspection methods (such as the visual inspection method) as a basis for inspection of the condition of wall tile deterioration. * corresponding aut hor, email: d9952 1007@ntu.edu.tw advances in technology innovation , vol. 2, no. 2, 2017, pp. 56 60 57 copyright © taeti the theory behind the tap tone diagnostic method is associated with determinations made through sound frequency. besides pitch, sound also has intensity and tone. genera lly speaking, pitch, intensity (loudness) and timbre are the three main ele ments of sound. the pitch of a sound is determined by its frequency of vibration. a higher frequency of vibration indicates a higher pitch, and a lo wer frequency of vibration indicates a lower p itch. the intensity of sound is determined by the vibration magnitude (amp litude) of the sound wave. higher a mp litudes equate to more powerful sound waves, and thus louder sounds. observing the diffe rent sounds generated from tapping on norma l and dete riorated tiles with a tap tone diagnosis stick, this study uses a fast fourier transform to determine a spectrogram. frequency (hz, x-coordinate) and amplitude (v, y-coordinate) a re used to form the basis for determin ing the degree of tile deterioration. spectrogra ms fro m the tap tone diagnosis are shown in fig. 3 and fig. 4. the spectrogram's peak and trough for norma l tiles show stable frequency (fig. 1), while deteriorated tiles show apparent instability (fig. 2). fig. 1 transformed spectrum for a tap tone diagnosis on normal tile fig. 2 transformed spectrum for a tap tone diagnosis on deteriorated tiles fig. 3 test procedure 2.2. infrared thermal image method this study uses the infrared therma l imaging to detect building e xternal wall tiles. infrared therma l imaging method is use natural sunlight, which a llo ws the building wa ll tile te mperature changes . in general, if there is no concrete pos ition defect (bulging phenomenon), infrared therma l imaging d isplays a consistent surface color. since the defective portion and the norma l portion of different therma l conductivity, this study used infrared therma l imaging device to measure ments the temperature diffe rence to determine peeling. selected object confirm the external wall tiles surface moist condition tap tone diagnosis and consistent strength tap tone record results marking on the external wall, to records the deterioration and normal tiles adjust the machine parameters temperatures every 30 minutes, record the normal and deterioration position of the infrared thermal imaging photos flir tools soft to analysis tile temperature advances in technology innovation , vol. 2, no. 2, 2017, pp. 56 60 58 copyright © taeti in this study, every 30 minutes to record the e xternal wall t iles temperature of the building, fro m 09:00 to 17:00, then through the analysis of temperature-t ime curve to discuss the best measure ment time. when detecting the position of internal de fects, the temperature deterioration of the position will be higher than other norma l area in the day heating period (08:00-11:00), on the contrary, when the cooling time in the day (16:00-19:00), the degradation position will lower te mperature than the norma l position. and then, this study used tap tone diagnostic method to confirm the deterioration of the range. 3. results and discussion 3.1. tap tone diagnosis results the study carried out tap tone diagnosis on 2 buildings . there are 63 a reas of deterioration out of 126 tap tone diagnostic tests. the study deems the first peak of the spectrogram as noise, the beginning of the second wave as the start point of deteriorat ion sound, and its end point as the end point of the deterioration sound. the spectrogram is then segmented by lattice points. further statistical analysis of the 126 deteriorated areas shows that the average start point of deteriorated tiles is 239hz, while that for the end point is 761hz. the deterioration e xtent is indicated by the shaded portion. the study suggests that upon tap tone diagnosis, if the fast fourier transform graph shows a frequency fro m 200 to 800 hz, it can be speculated that there is a cavity in the tiles. if tap tone diagnosis is required, it is suggested to tap at least 5 spots on the facade to increase overall detection accuracy. the study tentatively recommends that if the tile deterioration ratio e xceeds 50%, immediate overall facade renovation of the building should be carried out. if the tile deterioration ratio is less than 50%, then safety measures to prevent surrounding objects from falling should be immediately imposed. this study investigated 2 building cases in the national taiwan un iversity. this study used lattice segmentation method for analyzing 126 tiles deterioration image spectrums. it is found that the deterioration of the tiles starts at fre quency average about 240hz and ends at about 760hz. deterio ration hatched range shown in table 1 and table 2. table 1 case a of tap tone diagnosis results case a table 2 case b of tap tone diagnosis results case b advances in technology innovation , vol. 2, no. 2, 2017, pp. 56 60 59 copyright © taeti the study conso lid ates resu lts an d recommendations of visual inspection and taps tone diagnosis, and proposes a diagnostic criteria and standard operating procedures for e xternal wa ll tile deterioration diagnosis of taiwan buildings. 3.2. infrared thermal image diagnosis results fig. 4 case a external wa ll tiles infra red images (sp1: deterioration, sp2: normal) according to statistics , the study found that the east tiles ma ximu m te mperature is 2.0 degrees, occurred at 9: 00-10: 00. the west tiles ma ximu m te mperature is 1.3 degrees, occurred at 12: 00-13: 00. the south tiles ma ximu m temperature is 1.3 degrees also, occurs at 11: 00-12: 00. the north tiles ma ximu m te mperature is 1.2 degrees, occurred at 12: 00-13: 00. in this study, the statistics of the external wall te mperature, an average temperature d iffe rence of 1 degree results. this study suggests that the threshold can be used as observation fig. 5 case a temperature difference of east external wall tiles fig. 6 case a te mperature diffe rence of west external wall tiles fig. 7 case a temperature diffe rence of south external wall tiles fig. 8 case a temperature difference of north external wall tiles this study analyzed the average of daily temperature difference, the temperature difference if more than one degree, and in the best time of observation of this study suggests, it can be speculated that there is deterioration of tile production. aggregated data are shown in table 3. table 3 the investigation time and threshold of external wall tiles east west sout h nort h maximum t emperat ure difference 2.0 1.3 1.3 1.2 average t emperat ure difference 1 0.9 1.1 0.9 recommendation inspect ion t ime 09-11 12-14 10-12 11-13 advances in technology innovation , vol. 2, no. 2, 2017, pp. 56 60 60 copyright © taeti 4. conclusions since the construction boom in the 1970s in taiwan, many build ing e xterna l wa ll have su ffered natural deterioration, and many inc idents of e xterna l wall tiles falling have occurred, with some even resulting in fatalities. as the nu mber of such cases has increased, a comp lete e xterna l wall tile d iagnostic system is required. this study proposes public safety visual diagnostic models for e xternal wa ll tiles of build ings, including tap tone and infrared the rma l image diagnostic concepts, diagnostic method and diagnostic evaluation reference datum. it a lso offers a infrared thermal images method for e xternal wa ll tiles, as we ll as suggesting the adoption of tap tone diagnosis on tiles that p otentially have significant impact on public safety. it also verifies the developed evaluation model, and proposes countermeasures and recommendations for deterioration conditions. based on the test results, the study reco mmends that for frequencies between 200 and 800 hz on the spectrogram, it can be deduced that a cavity is present in the tiles, and this can be used as basis for determining tile deterioration. the study was carried out using infrared therma l imaging record to deteriorat ion and norma l t ile position. th is study observed temperature changes per 30 mins, it is obvious temperature d iffe rence with deteriorat ion tiles and norma l tiles between 09: 00 -11: 00 (east), 10:00-14:00 (west), 10:00-12:00 (south) and 11:00-13:00 (north), the study was calculated that the average temperature d ifference of 1.0 degrees. this study suggests that the best experimental inspection time is 09: 00 -14:00. exclude the impact of e xte rnal factors, if the temperature d ifference more than 1.0 degrees, it can be deduced that a cavity is present in the t iles, and this can be used as basis for determining tile deterioration. acknowledgement this research was supported by the ministry of science and technology (most 103-2221-e -002-171), and sinotech foundation for research and development of engineering sciences and technologies. references [1] l. w. chiang, s. j. guo, c. y. chang, and t. p. lo, “the develop ment of a diagnostic model for the deterioration of exte rnal wa ll tiles of aged buildings in taiwan,” journal of asian architecture and bu ild ing eng ineering, vol. 15, no.1, pp. 111-118, january 2016. [2] l. w. chiang, s. j. guo, and c. y. chang, “the model of visual inspection for building external wall deterioration tiles,” journal of architecture, vol. 87, pp. 49-66, 2014. (in chinese) [3] e. d. km, “service life prediction for buildings exposed to severe weather,” journal of asian architecture and building engineering, vol. 10, no.1, pp. 211-215, 2011. [4] f. tong, s. k. tso, and x. m. xua, “tile-wall bonding integrity inspection based on time-domain features of impact acoustics,” journal of sensors and actuators a 132, pp. 557-566, 2006. [5] u. taketo, k. kato, and s. hirono, non-destructive of the concrete structure, tokyo: morikita publishing co., ltd, 2000.  advances in technology innovation, vol. 2, no. 1, 2017, pp. 08 12 8 the tangent medial circles inside the region defined by hermite curve tangent to unit circle ching-shoei chiang computer science and information management, soochow university, taipei, taiwan. received 19 february 2016; received in revised form 08 april 2016; accepted 10 april 2016 abstract the design of curves, surfaces, and solids are important in computer aided geometric design (cagd). images, surround by boundary curves, are also investigated by many researchers. one way to describe an image is using the medial axis transform. under this consideration, the properties of the boundary curves tangent to circles become important to design for 2d images. in this paper, we want to find the medial axis transform (mat) for a special class of region, which is bounded by unit circle and the curves whose end points is on a the circle, and endpoints tangent vectors are parallel to the tangent of circle at the end points. during the process, we want find the medial circle tangent to other medial circle, until we reach the medial circle whose center is the center of the osculating circle for the point with local maximum curvature. there are 4 cases, symmetric/non-symmetric region with singular point/local maximum curvature point, and proposed algorithm for these 4 cases. we introduced algorithm for this 4 cases in this paper. keywords: medial axis transform, hermite curve, computer geometric modeling 1. introduction the 2d image can be stored by many different ways, including its boundary curves, a medial axis curves with radius function, union of many primitive figures, and so on. there are many researchers use different curves to simulate different images. for example, cinque, levialdi and malizia [1] uses cubic bezier curve to do the shape description. yang, lu and lee[2] use bezier curve to approach the shape description for chinese calligraphy characters. chang and yan[3] derived an algorithm to approach the hand-drawn image by using cubic bezier curve. cao and kot[4] derived an algorithm to do data embedding in electronic inks without losing data. the boundary curves can be also used to derived the offset curves and medial axis of an images[5], and also to simulate the nature objects, such as flowers[6]. in this paper, we would like to investigate the cubic hermite curves, where the end points are on a unit circle, and its end points tangent line is parallel to the tangent line of circles. using these properties, with more constraints on the singular point or maximum curvature at specified parameter, we survey the couture of the curves, so that the design of the curve may be simplify by giving constraint. 2. definition and theorem let introduce the hermite curve first. given two points p0 and p1 , with two associated tangent vectors and v1 , the hermite curve c(t) defined as (see fig. 1): fig. 1 hermite curve to simplify our problem, we assume that the end points p0 and p1 are on the unit circle, the first point p0 =(-1,0), and the second point p1 =(-cosθ , sinθ). it’s associated tangent vectors parallel to (0,1) and (sinθ , cosθ ). we have the following definition for this curve. * corresponding author, email: chiang@scu.edu.tw advances in technology innovation, vol. 2, no. 1, 2017, pp. 08 12 9 copyright © taeti definition 1: two points p0 =(-1, 0), p1 =(-cosθ , sinθ ) on the unit circle with its associated tangent vector v0=α0(0,1), v1=α1(sinθ , cosθ ) produced a hermite curve, we call it unit circle hermite curve, denoted h1(t;θ,α0,α1), where α0>0, α1>0, 0<t<1, 0<θ<2π. notice that for the general case, we can always convert the design problem into the design problem for h1(t;θ,α0,α1). we can always translate the center of current medial axis circle to the origin, scale the radius into one and rotate the circle so that the first boundary point is on (-1,0). we call this process the standardization of the problem. after we solve the problem, produce the curve we want, we can always inverse the rotate, scale, and translate process, to see the designed curves for the original design problem. in [7], we have the following two theorems: theorm 1: let c(t)=h1(t;θ,α0,α1), then c'(t)=(x′(t),y′(t)) =(0, 0) ↔ α0b(t)=α1d(t) =a(t) tan ( θ 2 ) , where a(t) =6t(t-1), b(t)=(3t-1) (t-1), d(t)=t(3t-2). we would like to find the ma and mat for the region bounded by the hermite curve with the circular arc from (-cosθ, sinθ) to the point (-1,0) rotate about the origin clockwise, we call this region r(t;θ,α). theorem 2: let the hermite curve c(t)=(x(t),y(t))= h1(t;θ,α0,α1), if the local maximum curvature is at t0, then (x'(t0)y'''(t0)-x'''(t0)y'(t0))(x'(t0)2+y'(t0)2)-3(x'(t0)y ''(t0)-x''(t0)y'(t0))(x'(t0)x''(t0)+y'(t0)y''(t0))=0. notice that this equation has degree 5 in t, which means there are 5 local maximum/minimum curvatures on the curve. the ma is very sensitive to its boundary, so many researcher use approximated ma instead. in pattern recognition, the important information for the region is easier to capture by using approximated ma. although it could have more than one local maximum curvature in our region, which means more than one end point of the mat in the region, we would like to find one endpoint in the region, so that the ma of the region has only two end points (one is on the singular point, or the center of the osculating circle associated with the boundary curve with local maximum curvature, the other one is on the origin). in this paper, we use ma to represent the approximated ma. we start simple case here. let’s start from the case α0=α1=α. let’s also assume that all points on the curve is outside the unit circle and the curve has no self-intersection. the ma of a region is easy to find if we know the region is symmetric to a line, so we have the following theorem: theorem 3: the curve c(t)=h1(t;θ,α,α) is symmetric to the line y= (−tan θ 2 )x. furthermore, c(t) and c(1-t) are symmetric points. proof: let t = tan θ 2 , we have c(t)+c(1−t) 2 = αtt(t−1)−1 1+t2 . obviously, this point is on the line y=tx. notice that this line y=tx passing through c(1/2). notice further that because c(t) and c(1-t) are symmetric points, so the point c(1/2) is a singular point, or c(t) is a vertex of the curve, which means c(1/2) is a singular point, or has constant curvature, or has local maximum/minimum curvature. with the symmetric property, we know the ma of the region is on the line (−tan θ 2 )x. we would like to find the ma point from its footpoint c(t) or c(1-t), we have the following theorem: theorem 4: the medial axis (xm, ym) associated with c(t)=(x(t),y(t)), where 0<t<1/2 is (to simplify the representation of the equation, we eliminate the parameter t for x, y, x′, y′, xm, ym and rm): (xm, ym) = (x,y)∙(x′,y′ ) (1,t)∙(x′,y′) (1,t) and rm = | (1,t)∙(−y,x) (1,t)∙(x′,y′) | ‖(x′, y′)‖ where ∙ is the inner product of two vectors, |a| is the absolute value of a, and ||(x,y)|| is the norm of the vector, t=−tan θ 2 . with the ma point and its distance to its footpoint, we find mat of the region. we also call mat the medial axis circles in this paper. notice that (xm(0), ym(0)) = (xm(1), ym(1))=(0,0). 3. algorithm and experimental result in this paper, we want find the medial axis circles inside the region, and all medial axis advances in technology innovation, vol. 2, no. 1, 2017, pp. 08 12 10 copyright © taeti circles tangent with its neighbors. in our assumption, we can always find one end points of the mat. in some case, there is singular point exist, the singular point is the end point of the mat branch (the other one is on the origin). on the other case, we can always find a point with local maximum curvature; we use the center of its osculating circle as the end point of the mat branch. how to find normal point of the mat? consider the end ma point we find, either singular point or points with local maximum curvature. assume it has parameter tend, 0<tend<1, c(t)=(x(t),y(t)) and we want find t1 and t2, 0<t0<tend and tend<t1<1, so that c(t1) and c(t2) are footpoints of a ma point (x,y), from the property of ma point, we have: (x-x(t0),y-y(t0))․(x′(t0),y′(t0))=0 (x-x(t1),y-y(t1)) ․ (x′(t1),y′(t1))=0 ((x-x(t0)) 2 +(y-y(t0)) 2 =(x-x(t1)) 2 +(y-y(t1)) 2 the first two equations indicates that the line through ma point and its footpoints perpendicular to tangent of the footpoint on the curve. the third equation indicates that the distances from ma point to its two footpoints are equal. now we have 3 equations, and 4 variables, which are x, y, t0 and t1. given t0, we can find the ma point (x, y) and t1. the idea to solve the system of equation is: algorithm 1: finding one ma point. 1. solve (x,y) by cramer’s rule from the first 2 equations, and x, y has two variables t0 and t1. 2. substitute the value x, y into the third equations, and leave one equation with 2 variables t0 and t1. 3. given t0, we have one equation with one variable t1, so we can find solutions for t1. with careful selection of t1, we find the ma points with its two footpoints. the above idea solves the mat (ma with radius function) for one specified point. as long as we know the parameter for the end point, we can always find all mat for parameter from 0 to tend. we would like to find not only one ma point, but also its ma circle tangent to current ma circles. we proposed two methods to find next ma circle. one method solve the problem algebraically, and the other method using an algorithm with binary search to approach the result. the first method only suitable for the symmetric case, which means the case with α0=α1. on the case that α0=α1=α, we want find the next circle tangent to the first circle, which centered at (0,0) with radius 1 in the very beginning. from theorem 4, we know the curve whose parameter t associate a mat circle (xm, ym, rm), we assume the next circle whose associated boundary parameter is t, so we want to find the circle tangent to the unit circle centered at the origin. the constraints we give is the distance between the centers of two circle is equal to the sum of the radius of two circles. that is, ‖ (x𝑚, y𝑚)‖ = rm + 1. to find the medial axis circle (xm, ym, rm) associated with c(t)=(x(t),y(t)), where 0<t<1/2, tangents to the unit circle centered at the origin, we have to solve a one parameter equation with degree 10. because the solutions for this equation are sensitive to its coefficient, we did not implement this method. we use the binary search strategy instead. algorithm 2: finding the next ma point (eps is a small number): # current ma circle is (x,y,r) with parameter t. the return value is next ma circle (x1, y1, r1) and t1 1. set the tmin=t, the current parameter value, and tmax is the parameter value for the end ma point. 2. let dist (distance between two centers) be 2, and srad (the sum of the radius of two circles) be 1. 3. while |dist – srad| > eps and |tmax-tmin| < eps : tt = (tmin+tmax)/2 find the mat (x1, y1, r1) for the parameter tt # (from algorithm 1). find dist, srad from these two circles. if dist srad > eps, then tmax = tt if dist srad < -eps, then tmin = tt 4. the value x1, y1, r1 is the ma circle we find. from the above algorithm, we can find the next ma circles for the parameter t. however, when the parameter t value close to the parameter associate with the local maximum curvature, it can only produce circles intersect with previadvances in technology innovation, vol. 2, no. 1, 2017, pp. 08 12 11 copyright © taeti ous circle, which means it cannot find more circle tangent to current circle, and "inside" the hermite curve. from the theorem describes in the last section, with the idea mentioned above, we give 4 examples in this section. the first two examples is for the symmetric boundary curve where α0=α1=α, for boundary curve with singular point and with local maximum curvature respectively. the third and fourth cases are for the non-symmetric boundary curve, also for boundary curve with singular point and with local maximum curvature respectively. example 1: (symmetric curve with singular point): given θ=5/9, t=0.5, we can find α0=α1=7.15025. the associated mat is shown in fig. 2. fig. 2 symmetric region with singular point example 2: (symmetric curve with local maximum curvature): given θ=5/6, α0=α1=8, the associated mat is shown in fig. 3. fig. 3 symm. region with localmaximum curvature in fig. 3(a), we find one ma circle tangent to the original one. we can find the next circle tangent to the red curve and the hermite cuve, however, this circle is not totally “inside” the hermit curve. on this case, we draw the osculating circle associated with the hermite curve at local maximum curvature (see figure 3(b)). notice that the above two examples has the properties that the ma points are on one straight line.the design requirements and design constraints are summarized based on the characteristics of the mechanism. example 3: (non-symmetric curve with singular point): given θ =5/6, t=0.4, we find α0=44.78461, α1=16.79423. the associated mat is shown in fig. 4. fig. 4 non-symm. region with singular point example 4: (non-symmetric with local maximum curvature): given θ=7/6, α0=18, α1=6, the associated mat is shown in fig. 5. fig. 5 non-symm. region with maximum curvature in fig. 5, the same as fig 3, the last circle is associated with the hermite curve at local maximum curvature. 4. conclusions the unit circle hermite curve, h1(t;θ,α0,α1), can be used to design curves and images. on many cases, the singular point may be needed to be considered. for example, the chinese characters may have cusp on the boundary. before the design of the character, the relationship between the parameter value and the contour of the boundary curve is important. when we give two values of the parameters for h1(t;θ,α0,α1) curve, the can find the other two values with simple computation. when we define a curve, the only memory we need is 7 locations to store these 4 parameters, and other 3 parameters for the standardization process. so, the design process advances in technology innovation, vol. 2, no. 1, 2017, pp. 08 12 12 copyright © taeti saved not only the time, but also the memory. when we find the ma circles inside the region bounded by unit circle and h1(t;θ,α0,α1), it is possible that the last circle intersects the osculating circle at the local maximum curvature. the osculating circle associated with end mat point of the region, so it is better to show this circle, so that the mat of the region also contains the end mat point. there are more constraints we can used to design the curves. for example, the maximum curvature happened at t0, the curve passing through a point p0 , the curve tangent to a line l0 , and so on. we believe the result will as simple as the case we introduced here. we can also consider the inverse process, that is, from circles tangents to each other, and find its boundary hermite curve. we will leave this for further research. acknowledgement this work was supported in part by the national science council in taiwan under grants most 104-2221-e-031-001. references [1] l. cinque, s. levialdi, and a. malizia, “shape description using cubic polynomial bezier curves,” pattern recognition letters, vol. 19, pp. 821-828, 1998. [2] h. m. yang, j. j. lu, and h. j. lee, “a bezier curve-based approach to shape description for chinese calligraphy characters,” proceedings of the sixth international conference on document analysis and recognition, pp. 276-280, 2001. [3] h. h. chang and h. yan, “vectorization of hand-drawn image using piecewise cubic bezier curves fitting,” pattern recognition, vol. 31, no. 11, pp. 1747-1755, 1998. [4] h. cao and a. c. kot, “lossless data embedding in electronic inks,” ieee transactions on information forensics and security, vol. 5, no. 2, pp. 314-323, 2010. [5] l. cao, z. jia, and j. liu, “computation of medial axis and offset curves of curved boundaries in planar domains based on the cesaro’s approach,” computer aided geometric design, vol. 26, no. 4, pp. 444-454, 2009. [6] p. qin, and c. chen, “simulation model of flower using the integration of l-systems with bezier surfaces,” international journal of computer science and network security, vol. 6, no. 2, pp. 65-68, 2006. [7] c. s. chiang and l. y. hsu, “describing the edge contour of chinese calligraphy with circle and cubic hermit curve,” computer graphics workshop, july 2013. [8] c. s. chiang, “the medial axis transform of the region defined by circles and hermit curve,” international conference on computer science and engineering (iccse), pp. 22-24, july, 2015.  advances in technology innovation, vol. 3, no. 4, 2018, pp. 157 165 reliability study of gan-on-sic hemt rf power amplifiers m. bakowski 4,* , j. lang 1 , j-k. lim 4 , j. hellén 3 , t. m. j. nilsson 3 , b. schodt 5 , r. poder 5 , i. belov 2 , p. leisner 1 1 sp technical research institute of sweden, box 857, 501 15 borås, sweden. 2 jönköping university, school of engineering, box 1026, 551 11 jönköping, sweden. 3 saab ab, solhusgatan 10, 412 76 gothenburg, sweden. 4 rise acreo ab, box 1070, 164 25 kista, sweden. 5 sp technical research institute of sweden, a.c. meyers v. 15, 2450 copenhagen, denmark. received 21 july 2017; received in revised form 13 september 2017; accepted 19 september 2017 abstract the rf power amplifier demonstrators containing each one gan-on-sic, hemt, chz015a-qeg, from ums in smd quad-flat no-leads package (qfn) were subjected to thermal cycles (tc) and power cycles (pc) and evaluated electrically, thermally and structurally. two types of solders, sn63pb36ag2 and lead-free snagcu (sac305), and two types of tim materials, nanotim and tgon tm 805, for pcb attachment to the liquid cold plate were tested for thermo-mechanical reliability. changes in the electrical performance of the devices, namely the reduction of the current saturation value, threshold voltage shift, increase of the leakage current and degradation of the hf performance were observed as a result of an accumulated current stress during pc. no significant changes in the investigated solder or tim materials were observed. keywords: gan-on sic, hemt, rf power amplifier, thermo-mechanical and electrical reliability 1. introduction radio frequency (rf) power amplifiers are used to convert a low-power signal into a higher power signal that drives the antenna of a transmitter. there are two trends in the rf amplifier applications. the first is replacing silicon-based transistors, such as ldmos fets, with gan high electron-mobility power transistors (hemt). this trend is observed, for example, in telecom applications. the second trend is a transition from gaas devices to gan-on-sic hemts, which is observed in the applications with higher demands on output power and reliable high temperature operation, such as military applications. the motivation for this study is the replacement of gaas transistors with gan-on-sic devices in rf power amplifiers in order to satisfy demands for higher output power, higher operating temperature (<200c) and higher system efficiency [1]. reliability of wbg semiconductors and especially of gan hemts is of primary importance and a subject of increasing number of investigations [2-6]. the subject of investigations is most often devices themselves. the objectives of this investigation are to experimentally investigate thermo-mechanical robustness of rf power amplifiers including packaged hemt devices assembled on pcb boards and exposed to power cycling (pc) and thermal cycling (tc) and to evaluate prospective thermal management methods including thermal interface materials (tim) by experiment and simulation. to satisfy environmental demands, there is also need for investigation of thermo-mechanical behavior of modern lead-free solder alloys in harsh environments. * corresponding author. e-mail address: mietek.bakowski@ri.se tel.: +46-70-7817760; fax: +46-8-7517230 advances in technology innovation, vol. 3, no. 4, 2018, pp. 157 165 copyright © taeti 158 2. experimental in total 50 rf amplifier demonstrators were assembled for the study. they consisted each of an 8 cu-layer pcb board with a simplified hf design containing one gan-on-sic, hemt, chz015a-qeg, from ums in smd quad-flat no-leads package (qfn). two types of solders, sn63pb36ag2 and lead-free snagcu (sac305), and two types of tim materials, nanotim and tgon tm 805, for pcb attachment to the liquid cold plate were tested for thermo-mechanical reliability. in 8 cu-layer pcb (25x35 mm) layers 2 and 7 contain solid cu while layers 3 to 6 contain a 25% symmetrical cu-layers with daisy chains. the outer material is ro4350 and material in-between cu-layers is fr4. the transistor area (3.5x3.5 mm) is covered by 0.3 wide cu-plated vias distributed with 0.5 mm spacing center to center (fig. 1). 24 demonstrators were subjected to thermal stress by 2300 thermal cycles (tc) between -20°c and 80°c and remaining 26 were subjected to the electrical and thermal stress by power cycling (pc) with drain current of 100 ma at a drain voltage of 45 v and cycle time of 2 min. hf characterization of all the boards was done before subjecting them to the thermal and electrical stresses. all the devices were also subjected to static electrical characterization by measuring threshold voltage and output and voltage blocking characteristics. the static electrical and hf characterizations were performed again after 2300 cycles of tc stress and after 1100, 4700 and 14500 cycles of pc stress. in addition, failure analysis was performed on the tc and pc stressed demonstrators by using optical microscopy and 2d x-ray microscopy. (a) cross-section (b) top view fig. 1 rf amplifier demonstrators the main reason for the tc tests was to test the reliability of the solder joints. the suggested tests were based on standard ipc 785, treating hemt devices and solder. one thermal cycle took 113 min. one month of the accelerated test is equal to about one year of use in field conditions. the temperature in the chamber and the temperature on at least one board were recorded during the whole test. fig. 2 pc arrangement on a liquid cold plate and tc arrangement in climatic chamber advances in technology innovation, vol. 3, no. 4, 2018, pp. 157 165 copyright © taeti 159 the main reason for the pc tests was to investigate degradation of the transistor package, including the die, solder joints and tim during cyclic heating and cooling. the temperature distribution on the hemt package and pcb was measured several times during the test by thermal imaging and by thermocouples. power cycling was performed at room temperature with demonstrator boards mounted on a liquid cold plate. the cycle time was set to 2 min and total number of cycles corresponds to 20 years of use. the boards were id labeled and underwent a screening procedure under which some were visually inspected, powered up and investigated with x-ray. after finishing tc and pc tests selected boards were visually inspected in white light, x-rayed and a couple of boards were subjected to cross-section investigations for failure analysis. 3. results the results of electrical and hf characterization show clearly changes and degradation in device performance as a function of the number of the pc cycles and accumulated current stress time. an increase of the leakage current is observed also after tc tests. no significant differences between the boards equipped with different solder and tim materials were observed which means that the observed changes in performance are due to the degradation of the hemt devices. 3.1. pc tests 26 boards were mounted on a cold plate (water cooling with a flow of 0.015 l/s) with thermal paste between the al-plate of the demonstrator and the cold plate. the transistors were exposed to drain voltage of 45 v and a varying gate voltage of -1.9 v to – 3.0 v. at -1.9v the gate was opened yielding a drain current of ~100 ma. an equal on/off sequence of total 2 min was repeated in total 14 500 times to correspond to 20 years of use. the 26 boards were divided into two groups connected in parallel and run with one power supply per group. the experimental set-up is shown in fig. 2. 3.1.1. output characteristics typical output characteristics of the investigated devices are shown in fig. 3 after various numbers of pc cycles. the summary of values of saturated drain current before and after 14500 cycles of pc is shown in fig. 4 for three groups of boards having different solder and tim materials (sn63pb36ag2&tgon805, sac305&tgon805 and sac305&nanotim). the saturated drain current values were taken for therain voltage vds=10 v and gate voltage vgs=0.8 v. a reduction in value of the saturation current is about 16 %, regardless of the type of the solder or tim material used. the observed reduction in the value of saturation current (vds=10 v, vgs=0.8 v) is about 5, 12 and 16 % after 1100, 4700 and 14500 cycles of pc stress, respectively. voltage, v 0 2 4 6 8 10 12 14 c u rr e n t, m a 0 100 200 300 400 500 pre-stress 1100 power cycles 4700 power cycles 14500 power cylces v gs = 0.8 v v gs = 0.6 v v gs = 0.4 v v gs = 0.2 v i sat at v gs = 0.8 v, v ds = 10 v device type s a tu ra te d d ra in c u rr e n t, m a 340 360 380 400 420 440 460 480 500 520 sn 63 pb 36 ag 2 snagcu (sac305) nano tim before pc after pc fig. 3 typical output characteristics measured on the same device after various numbers of power cycles fig. 4 saturation current values measured before and after 14500 cycles of pc stress advances in technology innovation, vol. 3, no. 4, 2018, pp. 157 165 copyright © taeti 160 the time dependence of the saturation current values is attributed to the degradation of the hemt characteristics and is discussed further in the discussion section below. 3.1.2. threshold voltage shift the threshold voltage was measured at vds=45 v by sweeping gate voltage, vgs, from -7 v to 0 v. the threshold voltage value was taken as a gate voltage corresponding to drain current ids=10 ma. there is a clear tendency of the threshold voltage shift towards more negative values showed by devices from different groups (sn63pb36ag2&tgon805, sac305&tgon805 and sac305&nanotim) as shown in fig. 5. the shift is by -0.1 to -0.15 v from the typical value of about -2.25 v. v th at 10 ma device type t h re s h o ld v o lt a g e , v -2,8 -2,6 -2,4 -2,2 -2,0 before pc after pc sn 63 pb 36 ag 2 snagcu (sac305) nano tim fig. 5 threshold voltage values measured before and after 14500 cycles of pc stress 3.1.2. leakage current the leakage current is less well behaved. there are devices with relatively small change in leakage current but also devices which show an increase in leakage current by a factor larger than 2, as a result of the pc stress. the typical forward blocking characteristics of the devices after varying numbers of power cycles are shown in fig. 6. the measurements were done sweeping drain voltage, vds, up to 100 v with gate voltage vgs=-7 v. the devices display high values of leakage current with the relatively strong dependence on the drain voltage. fig. 6 typical forward blocking characteristics of the investigated hemt devices after consecutive series of power cycles (vgs=-7 v) the summary of the leakage current data as a function of a number of power cycles is discussed in the discussion section below. 3.2. tc tests 24 boards were exposed to 2300 temperature cycles of ∆t=100 °c between -20 °c to +80 °c in a temperature chamber. the ramping up/down slope was around 2-3 °c/min and the dwell time was 15 min. 12 boards had lead solder material and 12 advances in technology innovation, vol. 3, no. 4, 2018, pp. 157 165 copyright © taeti 161 boards had lead free solder material. the two kinds of boards (lead & lead free) were evenly mixed over the grid in the chamber as shown in fig. 2. the boards were electrically characterized before and after the thermal cycles. 3.2.1. saturation current and threshold voltage figure 7 shows summary of saturated drain current and threshold voltage values for the boards with lead free solder. results show only minor changes after tc stress. the data for the reference boards with lead solder are very similar. these changes could be possibly related to the deterioration of the solder joints, however, no change in the joints was observed when performing material analysis. visual inspection as well as 2d x-ray imaging was performed on all the thermally cycled demonstrators. no failures related to the soldering were revealed during both tc and pc experiments. fig. 7 saturated drain current value and threshold voltage value before and after tc stress 3.2.2. leakage current a significant increase of leakage current was observed in almost all the devices after the tc test as shown in figs 8 and 9. the mechanism behind this behavior is not clear. fig. 8 typical blocking characteristics of the investigated hemt devices before and after 2300 thermal cycles of tc fig. 9 leakage current before and after 2300 cycles of tc stress 3.3. hf performance the rf measurements were performed using a network analyzer and registering s parameter values corresponding to the frequency of 1.3 ghz as shown in fig. 10. results of hf measurements show a slight decrease in s11 (return loss) and s21 (gain) parameters from -10.91±0.23 to -10.0±0.21 and from 16.04±0.24 to 15.51±0.28, respectively, after pc stress (measurement conditions vds=45 v, ids=100 ma, power level 5 dbm, 1.3 ghz). the summary of forward gain results (s21) is presented in fig. 11 and the summary of forward gain and return loss for sac305 boards is shown in fig. 12. overall, the high frequency measurements show a slight performance degradation of the power cycled boards and the same trend is observed for the demonstrators with lead free and lead solder. advances in technology innovation, vol. 3, no. 4, 2018, pp. 157 165 copyright © taeti 162 fig. 10 measurement of s21 (upper curve) and s11 (lower curve) parameters in the frequency range 30 khz to 3 ghz fig. 11 forward gain (s21) of boards with standard and lead-free solder before and after 14500 cycles of pc stress fig. 12 forward gain (s21) and return loss (s11) of all boards with led-free solder before and after pc stress 3.4. thermal modeling fig. 13 ir measurement and cfd simulation advances in technology innovation, vol. 3, no. 4, 2018, pp. 157 165 copyright © taeti 163 temperature distribution measured with ir camera and result of cfd calculation are shown in fig. 13 for dissipated power of 7.2 w (steady state) with tcoolant=11 °c, tamb=21 °c. a good qualitative agreement has been obtained between measurements and simulations. the coolant flow of 0.15-0.19 m/s is optimal for the set-up. other simulation results are that influence of using different solder materials is less than 1°c and that the newly developed nanotim results in 7°c lower temperature compared to commercially available tgon 805. validation of the transient cfd model is demonstrated in fig. 14 and fig. 15. fig. 14 test board with attached thermocouples fig. 15 simulated and measured temperature on chip and on pcb board (b) under tc pulse the significant finding shown in fig. 16 is that temperatures at different soldering locations of qfn hemt package are very different. it is important to take that into consideration when establishing and modeling reliability of soldered joints. fig. 16 temperature at the qfn package and at pcb obtained from ir measurements 3.5. material analysis to reveal possible failures in solder joints, the packages were inspected with sem and 2d x-ray, fig. 17 and fig. 18. no obvious visual damage to the test pcbs was identified and no significant failure modes for the solder joints or component cases were revealed after pc and tc tests. advances in technology innovation, vol. 3, no. 4, 2018, pp. 157 165 copyright © taeti 164 fig. 17 sem images of a cross section of failed power cycled and thermally cycled devices with sac 305 fig. 18 2dx-ray images of a cross section of failed power cycled and thermally cycled devices with sac 305 4. discussion the electrical parameters that are clearly influenced by the pc are the saturation value of the drain current and leakage current. the drain saturation and leakage currents of all the devices show logarithmic dependence on the stress time (number of pulses), as can be seen in figs 19 and 20. this is an indication that the changes are most probably related to charge trapping at the near interfacial trapping sites located in the algan layer or in the passivation layer. assuming trapping sites are distributed in distance from the specific interface and that charging is governed by the tunneling mechanism leads to the logarithmic dependence of the accumulated charge on charging time [6]. this is because the charge transfer to the trapping sites becomes less efficient, as trapping sites closest to the interface become occupied, due to exponentially decaying tunneling probability with distance. the same time dependence is then reproduced by drain saturation current and leakage current given relatively linear dependence of these two parameters on the interface charge. fig. 19 saturated drain current value versus number of power cycles fig. 20 leakage current value versus number of power cycles advances in technology innovation, vol. 3, no. 4, 2018, pp. 157 165 copyright © taeti 165 also tc has influence on leakage current as can be seen in figs. 8 and 9. however no significant influence of the tc on the remaining parameters was registered (fig. 7). investigated devices show high leakage currents and soft blocking characteristics that seem to be easily influenced by thermal and electrical stress. some of the experienced device failures are most probably related to the device voltage blocking properties. in total 11 boards failed during the tests, 10 out of 26 tested boards failed during the pc runs and 1 out of 24 tested boards failed during the tc run. 5. conclusions changes in electrical performance of the devices namely reduction of the drain current saturation value, threshold voltage shift, increase of the leakage current and degradation of the hf performance were observed as result of an accumulated current stress during pc tests. the most significant changes were observed in the drain saturation current and in the leakage current of the devices. the changes in these two parameters seem to be logarithmic in time and indicate that the mechanism behind them is charging of near interface states either in the algan layer or in the passivation by tunneling. a high rate of failures (40%) was observed in the pc tests. the failures are predominantly related to the same mechanism that governs the increase of the leakage current. also thermo-mechanical stress due to tc resulted in the increase of the leakage current of the devices. no significant visual damages in the investigated solder or tim materials were visually observed in electron microscopy and 2d x-ray microscopy after pc and tc tests. acknowledgments the support of vinnova, sweden ś innovation agency, and sweden energy authority of this project with number 2014-05667, is acknowledged. references [1] u. k. mishra, l. shen, t. e. kazior, and y. f. wu, “gan-based rf power devices and amplifiers,” proc. the ieee, vol. 96, no. 2, pp. 287-305, february 2008. [2] h. kim, v. tilak, b. m. green, j. a. smart, w. j. schaff, j. r. shealy, and l. f. eastman, “reliability evaluation of high power algan/gan hemts on sic substrate,” physica status solidi (a), vol. 188, no. 1, pp. 203-206, november 2001. [3] y. c. chou, d. leung, i. smorchkova, m. wojtowicz, r. grunbacher, l. callejo, q. kan, r. lai, p. h. liu, d. eng, and a. oki, “degradation of algan/gan hemts under elevated temperature lifetesting,” microelectronics reliability, vol. 44, no. 7, pp. 1033-1038, july 2004. [4] e. zanoni, g. meneghesso, m. meneghini, a. stocco, f. rampazzo, r. silvestri, i. rossetto, and n. ronchi, “electric-field and thermally-activated failure mechanisms of algan/gan high electron mobility transistors,” ecs transactions, vol. 41, no. 8, pp. 237-249, 2011. [5] g. meneghesso, m. meneghini, d. bisi, r. silvestri, a. zanandrea, o. hilt, e. bahat-treidel, f. brunner, a. knauer, j. wuerfl, and e. zanoni, “gan-based power hemts: parasitic, reliability and high field issues,” ecs transactions, vol. 58, no. 4, pp. 187-198, 2013. [6] a. barnes, esccon 2013, 12-14 march 2013, esa/estec, holland. [7] f. b. mclean, h. e. boesch, j. m. mcgarrity, and r. b. oswald, “rapid annealing and charge injection in al2o3 mis capacitors,” ieee transactions on nuclear science, vol. 21, no. 6, pp. 47-55, december 1974.  advances in technology innovation , vol. 3, no. 2, 2018, pp. 78 85 efficacy of nanocutting fluids in machiningan experimental investigation vamsi krishna pasam1,*, padmini rapeti2, surendra babu battula2 1department of mechanical engineering, national institute of technology, warangal, telangana state, india. 2department of industrial engineering, gitam institute of technology, gitam university, visakhapatnam, andhra pradesh, india. received 21 july 2017; received in revised form 10 oct ober 2017; accept ed 23 oct ober 2017 abstract this paper presents the experimental investigations on the performance of eco -friendly vegetable oil based nanofluids in turning operation. cutting temperatures, cutting forces, tool wear and surface roughness under constant cutting conditions are measured during machining. the influence of nanofluids prepared from nanoboric acid (nba) and carbon nanotubes (cnt) mixed separately with coconut oil (cc), on machining performance is examined. comparative analysis of the results obtained is done under dry, soluble oil (sl) and lubricant environments at 0.25% nano particle inclusions (npi). to understand the influence of npi, the experiments are conducted using ccnba and cccnt at varying npi also. application of cccnt resulted in improved machining performance compared to ccnba . reduction in cutting temperatures, main cutting force, tool wear and surface roughness is approximately 13%, 37.5%, 44% and 40% respectively by the application of cccnt compared to dry machining. thus application of ccnba and cccnt at 0.5% npi is more effective in improving machining performance. keywords: nanocutting fluids, vegetable oil, mql, surface roughness 1. introduction cutting fluids have been most frequently used from the perspective of cooling and lubricating actions during machining in manufacturing industry. the number of investigations revealed that application of conventional cutting fluids hinders the ecological balance and negatively affects the safety of operators [1-2]. these investigations led to the search for environmentally benign cutting fluids. this res ulted different alternatives to conventional cutting fluids in terms of cryogenic cooling, vegetable oils, solid lubricant, nanofluids, and minimum quantity lubrication (mql) technique. being biodegradable, operator friendly, abundantly available and affo rdable, vegetable oils have been thoroughly experimented to examine their applicability in machining operations. rapeseed oil, coconut oil, sunflower oil etc., are some of the vegetable oils which caught the sight of researchers giving rise to interesting results. investigations on the effect of new formulations of vegetable oils on surface integrity and part accuracy in reaming and tapping operations with aisi 316l stainl ess steel were done. vegetable oil based cutting fluids exhibited better performance t han mineral oils by way of tool life, tool wear, cutting forces and chip formation [3].the applicability of coconut oil as a base fluid in industrial applications was examined and machining performance was improved and compared to conventional cutting fluid [4]. solid lubricant assisted machining has evolved as one of the novel techniques. graphite, boric acid, and mos2 are some of the solid lubricants which have improved * corresponding author. e-m ail address: vam sikrishna@ nitw.ac.in advances in technology innovation , vol. 3, no. 2, 2018, pp. 78 85 copyright © taeti 79 tribological properties that have grabbed the attention of tribologists. owing to thes e properties, solid lubricants aid in wear reduction. these materials are characterized by weak atomic interactions between their layered structures, allowing low shear strength. improvement in process using graphite, caf2, baf2 and mos2 in grinding operation was reported [5-6]. an attempt was made to use graphite as a solid lubricant to reduce the heat generation at the milling zone [7], resulting in remarkable reduction of cutting force, specific energy and surface roughness. comparative analysis of machining with a cutting fluid in terms of cutting forces, surface quality and specific energy using graphite and mos2 during end milling process revealed encouraging results with solid lubricants [8]. experiments were conducted to assess the performance of solid lubricants during hard turning while machining bearing steel with mixed ceramic inserts at different cutting conditions and tool geometry [9]. it was reported that surface finish improved from 8 to 15% with the use of solid lubricants compared to dry ha rd turning. investigations on the performance of boric acid in machining revealed that, process performance improved with reduced particle size while using dry solid lubricants. cutting forces, cutting temperatures and tool flank wear were reduced and surface finish improved [10]. in view of enhanced thermo physical properties of nanofluids, various studies to evaluate these properties of nanofluids like specific heat and thermal conductivity were conducted and sufficient improvement in thermal and physical properties of nanofluids was observed [11-13]. in this context, cnts (carbon nanotubes) with enhanced mechanical, optical and chemical characteristics enable their application in machining operations. cnts have a high surface to volume ratio, good electri cal conductivity and more over their linear geometry makes their surface highly accessible to the electrolyte. the strength to we ight ratio of cnt is 500 times greater than aluminum. analysis of surface characteristics of the tool steel material using multiwall carbon nanotubes to improve the surface finish of material to nano level was done [14]. carbon nanotubes are special interes ts to researchers because of the novel properties like extraordinary strength, unique electrical properties and efficient con duction of heat. depending on the available literature, it has been found that nanofluids have a much higher and strongly temperature-dependent thermal conductivity at very low particle concentrations than conventional fluids. mql is a promising technique of applying lubricants during machining, which minimizes the usage of cutting fluid and reducing the cost of manufacturing. research works conducted to examine the nature of the mql technique during turning operation revealed that mql reduces induced thermal shock and helps to increase the workpiece surface integrity in situations of high tool pressure [15]. extensive research was carried out to analyze the performance of different types of nanofluids prepared from al 2o3, zno, nano diamond particles in mql grinding [16]. it was concluded by using nanofluids as lubricants basic properties and machining parameters showed considerable improvement. experimental investigation on the performance of nano boric acid suspensions in sae-40 and coconut oil showed that nano solid lubricant suspensions in vegetable oil improved the performance parameters during turning of aisi 1040 steel [17]. the affirmative aspects of nano solid lubricants, vegetable oils through mql technique are being explored by researchers worldwide in order to bring out environmentally safe and user friendly cutting fluids to overcome the disadvantages of conventional ones. in this context, the present work is an attempt to investigate the affect of vegetable oil based nanofluid s on machining parameters. nano boric acid and cnt are used as suspensions (0.25% by weight) in coconut oil and machining is carried out at constant cutting conditions. cutting forces, cutting temperatures and surface roughness are measured and compared in dry machining environment; using soluble oil, and two nanofluids prepared . 2. materials and methods the present work deals with the application of two nanofluids namely ccnba (coconut oil + nanoboric acid) and cccnt (coconut oil + cnt) which are used as lubricants during ma chining. other lubricant environments considered include dry advances in technology innovation , vol. 3, no. 2, 2018, pp. 78 85 copyright © taeti 80 machining, cc (coconut oil) and conventional soluble oil (sl) to perform comparative analysis of the performance of nanofluid s. nano boric acid suspensions are added to coconut oil on weight percentage basis (0.25% to 1%) and mixed manually. to enable proper dispersion of nanoparticles in coconut oil, sonication is done for a period of one hour . thus , two nanofluids namely ccnba and cccnt are prepared and applied as cutting fluids during turning . all the machining experiments are conducted on psg-124 lathe with a carbide tool (nc6110); heat treated aisi 1040 steel of 30±2 hrc is used as work piece(30mm diameter and 100mm length). cutting temperatures during turning are sensed by embedded thermocouple (k-type shielded) placed at the bottom of the tool insert in the tool holder. details of machining are consolidated in table 1. cutting force (fz) is tracked and recorded online using kistler 5070 dynamometer. tool flank wear is measured with tool makers microscope, cutting temperatures are measured with embedded thermocouple, surftest – sj301 with stylus radius 0.0025 mm and cut-off length 0.8 µm was employed for measuring average surface roughness (ra). the values are recorded thrice and average is considered to avoid discrepancies in measurement . machining is initially conducted under dry condition, followed by application of sl, cc, ccnba, cccnt as cutting fluids one after the other. cutting fluids are supplied using the mql technique with a flow rate of 10 ml / minute. experiments are conducted at constant cutting conditions (speed: 560 rpm; feed: 0.14mm; depth of cut: 0.5 mm) and 0.25% npi followed by varying npi. machining conditions are chosen referring to previous literature [17, 21]. cutting temperatures, cutting forces, tool flank wear and surface roughness are measured during all the cases and comparative analysis reflecting the performance of nanofluids with other cases is done. schematic representation of experimentation is given in fig. 1. table 1 machining environment cutting conditions speed : 60m/min ; feed : 0.14mm/rev ; depth of cut :0.5 mm npi 0.25%, 0.5%, 0.75%, 1% mql flow rate 10ml/min lubricant environment a. dry, sl, cc, ccnba, cccnt at 0.25% npi b. ccnba & cccnt at varying npi tool pslnr 2020 k12 cutting tool cnmg120408 nc6110 (coated carbide) cutting tool geometry -60, -60, 00, 60, 60, 60, 0.8mm fig. 1 schematic representation of experimental work advances in technology innovation , vol. 3, no. 2, 2018, pp. 78 85 copyright © taeti 81 3. results and discussion machining is carried out in dry environment and using sl, cc,ccnba,cccnt as cutting fluids. the results reflecting the variation of cutting temperature, cutting forces (fz) and surface roughness (ra) with respect to time during dry and other environments are presented and discussed in the following sect ions . 3.1. cutting temperatures variation of cutting temperatures with machining time is presented in fig. 2. it is observed that cutting temperatures increased with increase in time for all the lubricant environments. cc, cccnt and ccnba are effective in reducing temperatures when compared to dry machining and sl assisted machining. the affirmative behavior of cc can be attributed to the presence of saturated fatty acids to an extent of 90% and length of carbon chains of cc which belongs to medium chain category (6-12 carbons). the chemical composition and fatty acid profile of cc are the primary reasons behind its performance as cutting fluid. molecular structures of vegetable oils like cc are termed as triacylglicerides. they contain medium chain of fa tty acids (mcfa 6 to 12 carbons length) which are joined at the hydroxyl groups (oh) through ester linkages. cc with high % of sfa and mcfa, at higher temperatures enables formation of a stable film which separates the surfaces in contact [18, 19] exhibits good wettability. thus, its interfacial properties are enhanced, which plays vital role in lubricant application. [20].the phenomenon of format ion of cc film on the metallic surface is more significant in case of vegetable oils like cc. this behavior of cc is accelerated by the additions of nanoparticles (ccnba and cccnt) [21]. the reasons behind cooling property of nanofluids is the increase in thermal conductivity and heat transfer coefficient which improve the heat dissipation capacity of nanofluids, thus reducing cutting temperatures at machining zone. owing to these reasons, nanoparticles in cc tend to reduce cutting temperatures. the performance of cccnt is found to be slightly better by way of reduction in cutting temperatures due to its thermal p roperties [22]. the percentage improvement in the reduction of cutting temperatures by using vegetable oil based nanocutting fluids is found to be 13% compared to dry machining. thus it can be inferred that vegetable oil based nanofluids are more efficient in reducing cutting temperatures . fig.2 variation of cutting temperatures with machining time (speed: 560 rpm; feed: 0.14mm/rev; depth of cut : 0.5mm; npi : 0.25%) 3.2. cutting forces cutting force (fz) is measured at constant cutting conditions an d the variation with machining time is shown in fig. 3. it is noticed that, performance of cc, ccnba and cccnt is effective in reducing fz compared to dry and sl assisted machining. cutting forces are reduced because of the reduction in coefficient of fric tion between rubbing surfaces by the application of cc and cc based nanofluids. cc has strong intermolecular interactions get stuck to metallic surfaces resulting in a consistent lubricant layer which persists longer even at high temperatures. the film formation ability is found to be more pronouncing with advances in technology innovation , vol. 3, no. 2, 2018, pp. 78 85 copyright © taeti 82 cc which comprise of mcfa and which have high amounts of sfa [23]. another reason which influences reduction in cutting force is viscosity of nanofluids. when viscosity is low heat dissipation capacity of nanofluid increases and when viscosity is high the ability of a nanofluid to form a consistent film separating both the surfaces in contact during machining gets enhan ced [24-25]. thus, the coefficient of friction between the surfaces in contact reduces which in turn leads to reduction in cutting forces. in the present case, it is found that this phenomenon is more pronouncing by the application of cccnt compared to all the other lubricant environments. the percentage improvement in reduction of cutting fo rce with cccnt is approximately 37.5% compared to dry machining. fig. 3 variation of cutting force with machining time (speed : 560 rpm; feed : 0.14mm/rev; depth of cut : 0.5mm; npi : 0.25%) 3.3. tool wear tool wear measured corresponds to tool flank wear which is measured after every pass. the readings are recorded by using olympus analysis software. the tool is set on the adjustable table of the sc30 optical microscope such that the cutting edge of the tool is just above the focus of 5x and 10x zoom lenses. tool wear is marked at different zones to get the precise value of wear. the measured tool flank wear is tracked after every turn and average tool wear is taken and graphs are plotted. it is seen from fig. 4 that tool wear has decreased from dry machining to cccnt assisted machining. machining is influenced by high amounts of heat generated at the two deformation zones, which are termed as primary and secondary. this heat induces high temperatures in tool and workpiece. endowed with better thermal conductivity, enhanced heat transfer rate, cc based nanofluids lead to tool wear reduction [25]. this is accelerated by sfa and mcfa of cc due to which stable film of nano -lubricant persists as discussed in section 3.2. this in turn reduces the shear res istance along the interface and the nano -lubricant film separates the tool-work interfaces, which reduces plastic contacts tending to lower the tool wear [26-27]. compared to dry machining cccnt is observed to reduce tool wear to an extent of 44%. fig. 4 variation of tool wear with cutting fluid environment (speed : 560 rpm; feed : 0.14mm/rev; depth of cut : 0.5mm; npi : 0.25%) advances in technology innovation , vol. 3, no. 2, 2018, pp. 78 85 copyright © taeti 83 3.4. surface roughness results of surface roughness (ra) measurement are shown in fig. 5. it is observed that the surface roughness is very high during dry machining, followed by sl assisted machining. cc performed midway between ccnba and cccnt in improving surface quality. the application of ccnba and cccnt resulted in reduced surface roughness compared to other environments. significant reduction in ra is obtained by the application of cccnt. lubricity of cc, high chemical affinity of nanoparticles towards ferrous surfaces results in covering the valleys between asperities and pores on the surface of workpiece during machining [27]. improved cooling and lubrication of nanofluids, impart better surface quality to the workpiece by reducing cutting temperature, cutting forces and tool wear. thus , the application of coconut oil based nanofluids imparts good surface quality to machined surface. an improvement of 40% in the reduction of surface roughness is observed by the application of cccnt compared to dry machining. fig. 5 comparasion of surface roughness in different lubricant environments(speed : 560 rpm; feed : 0.14mm/rev; depth of cut : 0.5mm; npi : 0.25%) 3.5 influence of nanoparticle inclusions fig. 6 variation in machining performance with npi (speed : 560 rpm; feed : 0.14mm/rev; depth of cut : 0.5mm ) advances in technology innovation , vol. 3, no. 2, 2018, pp. 78 85 copyright © taeti 84 the variation of cutting temperatures, cutting forces, tool wear and surface roughness with varying npi is presented in fig. 6(a) to fig. 6(d). it can be visualized that up to 0.5% npi substantial improvement in machining is possible by reduction in temperatures, forces, tool wear and surface roughness by applying b oth types of nanofluids. cccnt at 0.5% has exhibited better machining performance compared to that of ccnba at 0.5% npi. this can be attributed to the phenomenon of agglomeration which affects the uniform dispersions in base oil with increase in npi [26-27]. thus , formation of a consistent nano lubricant separating the surfaces in contact is hindered. thus , beyond 0.5% npi reduction in cutting temperatures, cutting forces, tool wear and surface roughness is not evident . 4. conclusions the effectiveness of vegetable oil based nanocutting fluids ccnba, cccnt is examined in this work. comparative analysis of machining performance of these nanofluids using pure vegetable oil, soluble oil and dry states is done and the following conclusions are drawn: (1) vegetable oil based nanocutting fluids are found to be effective in reducing cutting temperatures, cutting forces, tool wear and surface roughness compared to dry machining and soluble oil assisted machining . (2) application of ccnba and cccnt as cutting fluids during machining resulted in better machining performance than that of cc based machining. (3) on the total cccnt with 0.5% npi is found to exhibit better effectiveness compared to all the cutting fluid environments . (4) percentage improvement in machining performance by the application of cccnt in reducing cutting temperatures, main cutting force, tool wear and surface roughness is observed to be 13%, 37.5%, 44% and 40% respectively compared to dry machining. references [1] e. o. bennett and d. l. bennett, “occupational airways diseases in the metal working indu stries ,” tribology international, vol. 18, no. 3, pp. 169-176, june 1985. [2] w. bartz, “ecological and environmental aspects of cutting fluids,” lubrication engineering, vol. 57, pp. 13-16, march 2001. [3] w. belluco, and l. de chiffre, “surface integrity and part accuracy in reaming and t apping stainless steel with new vegetable based cutting oils,” tribology international, vol. 35, no. 12, pp. 865-870, december 2002. [4] n. h. jayadas and k. prabhakaran nair, “coconut oil as base oil for industrial lubricants evaluation and modification of thermal, oxidative and low temperature properties ,” tribology international, vol. 39, no. 9, pp. 873-878, september 2006. [5] s. shaji and v. radhakrishnan, “analysis of process parameters in surface grinding with graphite as lubricant based on taguchi method,” journal of materials processing technology, vol. 141, no. 1, pp. 51-59, october 2003. [6] a. venugopal and p. v. rao, “performance improvement of grinding of sic using graphite as a solid lubricant,” materials and manufacturing processes, vol. 19, no. 2, pp. 177-186, february 2004. [7] n. suresh kumar reddy and p. v. venkateswara rao, “performance improvement of end milling using graphite as a solid lubricant,” materials and manufacturing processes , vol. 20, no. 4, pp. 673-686, february 2005. [8] n. s. kumar reddy and p. v . rao, “experimental investigation to study effect of solid lubricants on cutting forces & surface quality in end milling,” international journal of machine tools & manufacture, vol. 46, no. 2, pp. 189-198, february 2006. [9] d. singh and p. v. rao, “performance improvement of hard turning with solid lubricants ,” international journal of advanced manufacturing technology, vol. 38, no. 5-6, pp. 529-535, august 2008. [10] p. vamsi krishna and d. n. rao, “the influence of solid lubricant particle size on machining parameters in turning,” international journal of machine tools and manufacture, vol. 48, no. 1, pp. 107-111, january 2008. [11] a. r. moghadassi, s. m. hosseini, d. henneke, and a. elkamel, “a model of nanofluid effective thermal conductivity based on dimensionless groups,” journal of thermal analysis and calorimetry, vol. 96 , no. 1, pp. 81-84, april 2001. advances in technology innovation , vol. 3, no. 2, 2018, pp. 78 85 copyright © taeti 85 [12] y. hwang, h. s. park, j. k. lee, and w. h. jung, “thermal conductivity and lubrication characteristics of nanofluids ,” current applied physics, vol. 6, no. 1, pp. 67-71, august 2006. [13] p. k. devdatta, k. praveen namburu, and k. debendra das, “comparison of heat transfer rates of different nanofluids on the basis of the mouromtseff number,” electronics cooling, vol. 13,no. 3, pp. 28-32, august 2007. [14] s. prabhu and b. k. vinayagam, “afm investigation in grinding process with nanofluids using taguchi analysis ,” international journal of advanced manufacturing technology , vol. 60, no. 1-4, pp. 149-160, april 2012. [15] a. attanasio, m. gelfi, c. giardini, and c. remino, “minimum quantity lubrication in turnin g,” wear, vol. 260, pp. 333-338, february 2006. [16] b. shen, a. shih, and s. c. tung, “application of nanofluids in minimum quantity lubrication grinding ,” tribology & lubrication technology, vol. 51, pp. 730-737, october 2008. [17] p. vamsi krishna, r. r. srikant, and d. nageswara rao, “experimental investigation on the performance of nanoboric acid suspensions in sae-40 and coconut oil during turning of aisi1040 steel,” international journal of machine tools & manufacture, vol. 50, no. 10, pp. 911-916, october 2010. [18] sapna johnson and nirmali saikia, “fatty acid profiles of edible oils and fats in india,” centre for science and environment, cse/pml/pr-32/, pp. 1-42, 2009. [19] l. horary and e. richler, “bio-based lubricants and greases : technology and products,” john wiley and sons, vol. 17: tribology in practice series. [20] h. thomas and w. apple, “proceedings: american oil chemists society,” monograph, aocs, p. 91, 1991. [21] g. krishna mohana rao, r. padmini, and p. vamsi krishna, “performance evaluation of eco-friendly nanofluids in machining,” proc. 1st international conf. aeronautical and mechanical engineering, pp. 221-229, 2013. [22] s. shaikh, k. lafdi, and r. ponnappan, “thermal conductivity improvement in carbon nanoparticle doped pao oil: anexperimental study,” journal of applied phys ics, vol. 101, no. 6, pp. 064302(1-7), march 2007. [23] m. h. cetin, b. ozcelik, e. kuram, and e. demirbas, e, “evaluation of vegetable based cutting fluids with extreme pressure and cutting parameters in turning of aisi 304l by taguchi method ,” journal of cleaner production, vol. 19, no. 17-18, pp. 2049-2056, november-december 2011. [24] p. sivashanmugam, “application of nanofluids in heat transfer applications,” an overview of heat transfer phenomena, october 2012. [25] s. a. lawal, i. a. choudary, and y. nukman, “application of vegetable oil-based metal working fluids in machining ferrous metals a review,” international journal of machine tools & manu facture, vol. 52, no. 1, pp. 1-12, january 2012. [26] w. yu, and h. xie, “a review on nanofluids: preparation, stability mechanisms and applications ,” journal of nanomaterials, vol. 2012, pp.1-17, 2012. [27] a. erdemir, “solid lubricants and self lubricating films,” modern tribology handbook, argon national library, 2001.  advances in technology innovation, vol. 4, no. 2, 2019, pp. 84-93 a study on heading and attitude estimation of underwater track vehicle dae-hyeong ji, hyeung-sik cho * , sang-ki jeong, ji-youn oh, seo-kang kim and sam-sang you division of mechanical engineering, korea maritime and ocean university, busan, korea received 03 may 2018; received in revised form 05 august 2018; accepted 24 october 2018 abstract in this paper, we studied the designing and manufacturing of an underwater track vehicle (utv) that can be operated in the underwater environment. we designed the electrical and control system for precise operation of the utv and conduct test operation by using the utv. we developed an attitude reference system (ars) that was composed of the ring laser gyro (rlg) sensor, a geomagnetic sensor, and usbl sensor to estimate precise heading and attitude of the utv. an inertial navigation system (ins) is developed to be combined with the developed ars. also, the ins navigation is configured by supplementing the usbl sensor information. we design a controller for a precise trajectory and attitude tracking by installing the ars on the utv. in this paper, we used an extended kalman filter in the ins to estimate the position and attitude of the utv. the ars is studied to obtain more precise sensor information in an uncertain environment underwater. performance tests of the developed ins using the utv are conducted and the results show that the system has the best performance. keywords: ring laser gyro(rlg), ultra-short baseline(usbl), attitude reference system(ars), inertial navigation system(ins), extended kalman filter(ekf) 1. introduction many kinds of research have been conducted to develop an underwater platform and are capable of observing and moving the ocean with abundant resources and energy [1-2]. fig. 1 is an example of an underwater platform being developed from observation and operation in the water. motion modeling of such an underwater tracked vehicle has been carried out [3]. one of the critical technologies of this marine observation and work platforms is to measure their direction and location information accurately. fig. 1 underwater work platform * corresponding author. e-mail address: hchoi@kmou.ac.kr advances in technology innovation, vol. 4, no. 2, 2019, pp. 84-93 85 it is very important to obtain the heading information of the navigation sensor to measure its position and the travel direction. among the navigation sensors, ring laser gyro (rlg) is the key sensor to obtain the heading information. attitude reference system (ars) is a sensor system that obtains the heading and position information by using a gyro sensor and geomagnetic sensor [4-6]. [7] implemented mems inertial sensors, the error correction, and algorithms for high performance as the sensor needed to develop low cost and high-performance ars. also, to calculate the posture by integrating the gyro sensor value has a disadvantage is that it drifts due to sensor drift. so, [8] recently implemented ars by using an acceleration sensor and gyro sensor to overcome this drawback of the combination of various sensors. [9] estimated the posture of marg (magnetic, angular rate, and gravity) sensor information of ekf by using quaternion. [10] conducted a study to follow the position of rov in usv by combining imu and usbl. in this paper, we presented the underwater navigation of the underwater track vehicle (utv) which can move on the sea floor. the ars of the hull, which is an important factor of the underwater navigation, were constructed using rlg and a magnetometer. also, a small utv was designed and fabricated to verify the actual sea area of the constructed underwater navigation system. the utv is designed to be able to move on land and sea and to mount various navigational sensors such as gps, depth, imu, usbl on the hull. this paper is organized as follows. section 2 of this paper describes the outline and specifications of the designed utv. section 3 describes the implementation of ars using rlg and magnetometer information. in section 4, the implemented ars was attached to the test-bed and the experiment was performed through the correct step. in section 5, we concluded this paper. 2. underwater track vehicle a small utv was designed to verify the excellent performance of the developed underwater navigation system through experiments in the water tank and the sea area. the designed utv is shown in fig. 2(a). it is a structure equipped with an infinite orbit for movement on the land and seafloor. the external shape is designed to have a length of 1200mm, the endpoint distance of both endless tracks is 700mm and the distance from the bottom to the top is 800mm. depends on the control method, a total of two motors control each caterpillar individually and can be turned forward / backward and left / right. the utv also has various navigational sensors such as gps, depth, imu, and usbl. the picture of the utv is shown in fig. 2(b) and the total weight is about 80kg. the utv is transported to the experimental sea by the ship in the sea area test and is designed to be operated by the power supply and communicate by using the tether cable by descending to the sea floor through the crane in the target sea area. route control and tracking operation are carried on by applying the designed underwater navigation system to utv. (a) design utv (b) produced utv fig. 2 underwater track vehicle advances in technology innovation, vol. 4, no. 2, 2019, pp. 84-93 86 3. attitude reference system ars, which is an essential sensor in the underwater navigation of utv, was constructed using rlg and a magnetometer. the development of the ars is as follows. 3.1. configuration of the ars the ars is composed of rlg and magnetometer and the addition of the navigation algorithm. the ars using rlg has a problem of integration error of data due to time increase. to reduce the accumulation of integration error, we constructed an extended kalman filter algorithm using gyro data, acceleration data, and geomagnetism data. the block diagram that corrects the integration error accumulation of the sensor data using the extended kalman filter is shown in fig. 3. fig. 3 structure of ars algorithm fig. 3 is a method of calibrating using acceleration and external geomagnetism data. the angular velocity of rlg is integrated by using the relation between euler angle and angular velocity, and the data of roll, pitch, and yaw derived are corrected by the extended kalman filter. the reason for using a similar sensor, gyro, and geomagnetism is to utilize the advantages of each sensor [10-11]. yaw data from the geomagnetic sensor are sensitive to the surrounding magnetic field and has a large error. also, it is greatly affected by the surrounding environment. to grasp this characteristic, we experimented with the characteristics of the geomagnetic sensor and constructed the experimental apparatus using the gyro and acceleration sensor to compare with the experimental data. 3.2. ars testbed the testbed created for the performance test of the ars consists of rlg, mems type attitude, heading reference system (ahrs) that provides information on acceleration data and sensor module including the geomagnetic sensor. the storage and computation of the test device data were handled using an embedded linux processor. fig. 4 shows a diagram of the configuration of the ars. fig. 4 ars diagram advances in technology innovation, vol. 4, no. 2, 2019, pp. 84-93 87 in fig. 4, the ahrs, gps, and imu sensor data is used for the algorithm that is ported to the embedded linux. the computed data and the stored data are sent to the computer that stores the data in rf wireless communication. to extract the data from the sensors, a device that provides motion was created in the form of a rotating table. the developed equipment is shown in fig. 5. the rotating table was constructed by applying a servo motor with a resolution of 0.088° and arm-m4 series micro-controller was used to configure the system to control the rotation speed and the rotation angle of the motor precisely. the rotating table is shown in fig. 5. the control system of the rotating table is shown in fig. 6. fig. 5 ars test-bed(left) & rotating table(right) fig. 6 composition of turntable control system (a) result of stop test (b) result of exercise test fig. 7 analyze the geomagnetic sensor the performance of the ars placed on the table using the rotating table was tested and the data of the geomagnetic sensor were analyzed. the data of the geomagnetic sensor are shown in fig. 7 as a result of repeated rotation test at the stopping state and the constant angle. as shown in fig. 7, the heading error in the static state of the geomagnetic sensor is shown in fig. 7(a), advances in technology innovation, vol. 4, no. 2, 2019, pp. 84-93 88 the minimum angle is 0.1° and the maximum angle is 0.3°. as shown in fig. 7(b), the result of each repeated experiment (0°-20°) showed an error of 2.4° or more. also, the overall error in 10 times independent geomagnetic sensor tests did not deviate significantly from the above graph, but the mean range of errors varied finely depending on the surrounding environment. therefore, geo magnetic data in the absence of motion does not show large error, but the operation in the system cannot guarantee data in the system. in an environment that exerts a magnetic force on the periphery, the error value is not linear, so it is found difficult to use by itself. the rlg has a significant advantage that it is not affected by the surrounding magnetic environment. however, since the angular velocity information is provided, integration is required to obtain each information. at this time, the cumulative error occurs with time due to integration. in this experiment, the euler angular data obtained by converting the rlg rotation experiment results into the euler angles and integrating them are shown in fig. 8. it shows the integration error of about 1.5 degrees due to the accumulation of the integration error when the angular velocity of the ring laser gyro is converted into the euler angle and the result is tested for about 10 minutes. fig. 8 result of rlg sensor test magnetic force profoundly influences the geomagnetic sensor. therefore, we designed the system that uses only the sensor providing initial yaw angle information, constructs the ars using the rlg sensor, and processes the sensor data with the extended kalman filter (ekf). the ekf algorithm is constructed as follows. in this system, the state variable of the ekf is denoted by x = [𝜙 𝜃 𝜓]. the system model shows the relationship between the gyro acceleration and the euler angle as shown in eq. (1) [12].   1 sin tan cos tan sin tan cos tan 0 cos sin cos sin 0 sin sec cos sec sin sec cos sec p p q r q w q r w f x w r q r                                                                        (1) since the nonlinear function 𝑓(𝑥) cannot be applied to the ekf, the 𝑓(𝑥) of eq. (1) is partially differentiated for each state variable to obtain the jacobian as eq. (2). 1 1 1 2 2 cos tan sin tan sin sec cos sec 0 2 2 2 sin cos 0 0 cos sec sin sec sin sec tan cos sec tan 0 3 3 3 f f f q r q r f f f a q r q r q r f f f                                                                              (2) the ekf is an algorithm for discrete-time. therefore, when the eq. (2) is discretized, the system matrix a is expressed by eq. (3). advances in technology innovation, vol. 4, no. 2, 2019, pp. 84-93 89 a i dt a   (3) in eq. (3), dt is the sample time and i is the identity matrix. the acceleration and geomagnetism data are used as the correction values and are calculated as shown in eq. (4). 0 sin 0 cos sin 0 cos cos fx u w v p g fy v w u q g fz w v u r g                                                             (4) about 1 1 sin sin cos fy fx mag kg g                    ， ， 1 mag mag mag k k   (5) in eq. (4), ϕ and θ are calculated approximately by acceleration and the calculated ϕ and θ have large errors. the correction data ψ is obtained by using the geomagnetism data as shown in eq. (5). by subtracting the first measured data from the measured geomagnetism data, a change in angle can be obtained. the data ψ including the integration error of the rlg is corrected using the angular variation of the geomagnetism data. next, the measured model equation (z) can be calculated from the acceleration data and the geomagnetism data as shown in eq. (6) to measure all three state variables. therefore, it can be expressed by eq. (6). 1 0 0 0 1 0 0 0 1 z v hx v                           (6) eq. (6) does not need to find a jacobian in a linear form with measured values. finally, the system noise covariance and the measurement noise covariance q, r matrix can be measured close to the real one by having equipment that can accurately measure signal characteristics and rlg characteristics. however, since the measurement pieces of equipment are costly, the q and r matrices are considered as design factors and the performance trends are observed and determined. the system design and matrices used for the ekf are obtained through eqs. (1)-(6) and the calculation is performed by applying the ekf according to calculation order. 4. experiments and considerations of ars 4.1. calibration of the ars the corrected data using the correction algorithm was verified by comparing the experimental data. the data was obtained at a sampling time of 100 hz and data of 12, 6400 pieces in about 21 minutes. the measured data are shown in fig. 9 as a graph. in fig. 9, the three graphs located on the left side are rlg data alone and euler angles are converted and integrated. in this case, an integration error of about 2.4° to 3.7° occurred for about 20 minutes. however, after applying the ekf, the integral error hardly occurred. the offsets are 0.3° and 1.2°, respectively, at the roll and pitch angles, are the error with the mechanism that fixes the rlg. next, an experiment was conducted to evaluate the error of posture by rotating forward and backward by repeating the attitude constantly. experiments were carried out in the range of -30° to 0° for a period using the experimental setup of fig. 5. advances in technology innovation, vol. 4, no. 2, 2019, pp. 84-93 90 about 500,000 data were received for about 83 minutes with a sampling time of 100hz. the measured data are shown in figs. 10-11 are shown. fig. 9 attitude correction in the stationary state in fig. 10, the uncorrected yaw data exhibited an error of more than about 12° during rotation of about -30° to 0° for about 83 minutes. fig. 11 is a graph of geomagnetism data used as correction data. since there is no integration as shown in the graph, there is no integration error. however, it shows a measurement error of 6° to 7° by repeatedly rotating about -30° to 0° for about 83 minutes. fig. 10 yaw angle of raw gyro fig. 11 yaw angle of raw magnetometer the results are shown in fig. 12 as a result of fusing data of figs. 10-11 with an extended kalman filter. in fig. 12, it is shown that there is a slight error in the starting position and the final position due to the mechanical error of the rotating table advances in technology innovation, vol. 4, no. 2, 2019, pp. 84-93 91 system. therefore, as shown in fig. 12, there is a slight error in the position of the start point and the end position. however, the results of the whole experiment are shown in figs. 10-11 and it can be confirmed that the proposed algorithm does not diverge for about 83 minutes. fig. 12 yaw angle of calibrated ars 4.2. test of platform the calibrated ars was applied to the actual moving platform (utv) to check the performance of the ars under study. the turntable test showed satisfactory performance for the ars under study. however, in application to an actual platform, the performance is affected by the actuating mechanism that generates many vibrations during movement is shown in fig 13; it might cause a little more positioning and attitude errors. fig. 13 test-bed for ars fig. 14 outdoor driving test fig. 15 result of outdoor driving test advances in technology innovation, vol. 4, no. 2, 2019, pp. 84-93 92 the configuration of the utv consisted of the rlg constituting the ars and the correction geomagnetic sensor tcm-3. the ars was installed in the waterproof housing of the utv. also, a similar mems type imu was installed and all the obtained data is recorded using rf wireless communication system in real time. the ars has been developed for underwater usage. however, it is difficult to verify the performance under the water. therefore, the land test was carried out to confirm the performance. the scene of the outdoor test is shown in fig. 14 and the results are shown in fig. 15. fig. 15 shows the result of applying the algorithm and the result of non-using the algorithm. the graph without the algorithm (red color) diverges from the heading angle. during the 10-minute test time, the cumulative error occurred at about 2 °. since the error due to integration increases rapidly with time, data that do not apply the algorithm will quickly diverge in the future. therefore, the current test results show approximately 2° difference between using and non-using the algorithms. if the test is performed for a long time, the error due to integration will increase. despite a lot of vibration while driving on uneven surfaces, the heading data of the ars showed an error of about 0.02° to 0.5°, which is quite small. 5. conclusions in this paper, ars which can acquire the attitude of the underwater vehicle is an essential factor in underwater navigation and is designed by using rlg and magnetometer. also, a small utv was designed and fabricated to verify the performance of the designed underwater navigation system through a water tank and sea area experiment. in the study, the extended kalman filter was used to reduce the integration error of data over time, which is the biggest problem of ars using rlg. then, by using this, an algorithm that efficiently removes the integration error of the yaw angle by correcting it through the fusion of the gyro, acceleration, and geomagnetism data was studied and was implemented in the system. through the turntable experiment, it was confirmed that the cumulative error accumulation amount before the correction was more than 12° in the result of about 83 minutes. the geomagnetism data to correct this error also showed measurement error more than 6° to 8°. as a result of calibrating the gyro data and geomagnetism data with the extended kalman filter, the postural error of about 0.2° to 0.4° in the experiment over 1 hour occurred. this error is also assumed to be caused by the mechanical part of the experimental equipment. the algorithm of the ars was verified by the ars applied to the test-bed. it can be seen that only a slight error of about 0.02° to 0.4° occurred in the straight running of approximately 100m. however, due to the nature of the integral error that rapidly diverges instantaneously, many platform application tests in various situations are required. acknowledgments this research is a part of the project national research foundation of korea (nrf-2016r1a2b4011875) and a part of project titled “r&d center for underwater construction robotics”, south korea (pjt200539), funded by the ministry of oceans and fisheries (mof). conflicts of interest the authors declare no conflict of interest. references [1] t. s. kim, i. s. jang, c. j. shin, and m. k. lee, “underwater construction robot for rubble leveling on the seabed for port construction,” 14th international conference on control, automation and systems, 2014, pp. 1657-1661. [2] c. h. kim, t. s. kim, and m. k. lee, “study on the estimation of the cylinder displacement of an underwater robot for harbor construction using a pressure sensor,” journal of navigation and port research, vol. 36, no. 10, pp. 865-871, 2012. advances in technology innovation, vol. 4, no. 2, 2019, pp. 84-93 93 [3] d. h. choi, y. j. lee, s. m. hong, h. s. choi, and j. y. kim, “a study on dynamic modeling for underwater tracked vehicle,” journal of ocean engineering and technology, vol. 29, no. 5, pp. 386-391, 2015. [4] j. h. kim, h. g. min, j. d. cho, j. h. jang, s. h. kwon, and e. t. jeung, “design of angular estimator of inertial sensor using the least square method, ” world academy of science, engineering and technology, vol. 60, pp. 1541-1544, 2009. [5] d. jung and p. tsiotras, “inertial attitude and position reference system development for a small uav,” in aiaa infotech@ aerospace 2007 conference and exhibit, may 2007, pp. 2763. [6] y. c. lai, s. s. jan, and f. b. hsiao, “development of a low-cost attitude and heading reference system using a three-axis rotating platform,” sensors, vol. 10, no. 4, pp. 2472-2491, 2010. [7] d. gebre-egziabher, r. c. hayward, and j. d. powell, “a low-cost gps/inertial attitude heading reference system (ahrs) for general aviation applications,” proc. of the 1998 ieee position location and navigation symposium, january 1998, pp. 518-525. [8] w. qin, w. yuan, h. chang, l. xue, and g. yuan, “fuzzy adaptive extended kalman filter for miniature attitude and heading reference system,” in proc. of the 2009 4th ieee international conference on nano/micro engineered and molecular systems, january 2009, pp. 1026-1030. [9] j. l. marins, x. yun, e. r. bachmann, r. b. mcghee, and m. j. zyda, “an extended kalman filter for quaternion-based orientation estimation using marg sensors,” proceedings of the 2001 ieee/rsj international conference on intelligent robots and systems, november 2001, vol. 4, pp. 2003-2011. [10] d. w. jung, s. m. hong, j. h. lee, h. j. cho, h. s. choi, and m. t. vu, “a study on unmanned surface vehicle combined with remotely operated vehicle system,” proceedings of engineering and technology innovation (2018), vol. 9, pp. 17-24. [11] h. s. yu, c. j. kim, c. k. song, and h. w. park, “an application of large heading attitude model for initial alignment of inertial navigation system,” the korean society for aeronautical and space sciences, pp. 1983-1988, 2012. [12] h. s. yu, s. w. choi, and s. j. lee, “nonlinear filtering approaches to in-flight alignment of sdins with large initial attitude error,” journal of institute of control robotics and systems, pp. 468-473, 2014. [13] t. h. chung, k. w. song, and b. j. chang, “design of the kalman filter for transfer alignment of strapdown inertial navigation system,” journal of institute of control, robotics and systems, 1991, pp. 142-146. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation , vol. 1, no. 2, 2016, pp. 33 37 33 copyright © taeti innovation management of stock trading system yi-ping lee * department of business administration national chung hsing university , taichung, taiwan. received 01 april 2016; received in revised form 08 may 2016; accept ed 12 may 2016 abstract stock trading ma rket fluctuated wild ly, which affect a country's economic and business operation. one of the important factor in stock trading system is price change limit, in the case of taiwan is 7 percent ups and downs per day. nowadays stock price change limit is set by the government, and that is based on the economic development of the country, but this price change limit is not the same with business’s view point. basically business is a private owned property, and the authority of business have to responsible for the performance of operation. re lease part control right of the price change limit to business by an innovation manage ment way, wh ich method allo w companies decide 1 to 2 percent ups and downs price change limit, but this rate is still under the control of govern ment. innovation stock trading system will activate the trading of stock market. keywor ds : innovation management, per day stock trading price change limit, business operation, business management. 1. introduction it is easy to achieve the goal of the national wild regulatory comp liance of stock trade system, but the d isadvantage of this way is like ly to ignore the changing factors of economic environment and mission of stock trading participants. there are too many d iversity in the stock trading partic ipants, such as government focus on control and manage ment, but business focus on the survive of competit ion, e xcept this, there are different industry characteristics, size of firm, purpose of raising funds for each f irm, diffe rent way of enterprise in management of control enterprise and investors may have diffe rent goal of investment. for the better development of stock trade market, we need mu ltip le stock change range controls to satisfy multiple stock trading conditions. the common re ference of stock trading system is per day’s price change limit, the other is buy and sale a stock in the same day, most government will e xpand or shrink the price change limit rate to active or tighten the stock ma rket trading volu me, on the other hand, government will tighten the trading system by limit trad ing a stock in the same day, that is partial way of government’s control to the stock trading market. every country have its own different stock trading range limits, stock trading market may be a country's economic showcase, china’s stock change range usually keep as ten percent, but they tried a system na med circu it breake r system(cbs) in recent years to stop the trade in stock market. traders of chinese stock trade with technica l analysis, they will make a strategy decision by observe and analyze transactions before they choose to buy or sale stocks, (hui qu, xindan li, 2014). there is no daily stock trade range limit in the u.s. stock market, but in japan has taken level d istance formula for daily trade range. individual traders of stock market are still be the ma jority of investor of stock ma rket, those investor’s values perception also affect the stock ma rket trading patterns and fra me, (ja mdee sutthisit, sheng xiong wu, bing yu , 2012). stock market will be affected by e xternal events, and the stock da ily change range limit (sdcrl) of stock also affect the manner and degree of reaction to those events (gregory j. kuserk, peter r. locke, 1996), utility of sdcrl in stock market is an stop point of event reaction in trading, with which to forbid event over reaction in a transaction day, this is investors point of vie w in stock trading, the price change range limit will fit the issue of risk preference in a stock ma rket, investors of risk preference will make trading decision connect with the sdcrl, then the situation of a stock * corresponding aut hor, email: yiping727@yahoo.com.tw advances in technology innovation , vol. 1, no. 2, 2016, pp. 33 37 34 copyright © taeti ma rket will face the fa milia r group me mber in a short time, and that is an sudden impact to stock ma rket, and this impact will lo w down the stability of stock trading, most governments and investors dislike this situation, especially in a down stage stock market. usually, sdcrl is a mandatory regulation of the government, but the rapid and p luralis m changes in the market a re unpredictable and turmo il. there should have many different purposes in raise capital, business may ra ise capital for scale reason in expand market, some are ra ise capital for cut into overseas markets, the same point to most business are for continuing operations (guy, john, 1995). investors choose favor stock market and business to invest, thus, establish a stock trading system with sdcrl partia l control by business will fit the mu ltip le need of stock trading ma rket, and which is consistence with government, business owner and investors. (eugene f. fa ma, ja mes d. macbeth). profitability usually associated with investor’s reward and business growth (merton h. m ille r, myron s. scholes , 1982), business and investors plays an important role of the stock trading process, but the change in stock price limits have no room for the m to e xpress their opinion, if they can show what they want to fra me a d iversified stock trad ing system, may them will have their idea, such as the high ranking manager can set his operation project into relevant thinking (guy, john, 1995). there are a lot of prior research focus on the range of sdcrl, some of them are re move of the sdcrl in stock trading, but we d id not find a research discuss about the sdcrl partia l decide by the business, this research first pion eer made this cla im for business owner have the right to claim his idea of sdcrl. which will create diversity and vibrant stock trade market. 2. different approach design-related factors of business partial controlled sdcrl operation step of sdcrl is pre file by the business first, and government will rev iew the application, if it is permitted, govern can send a set account and password, business have to public the informat ion the same time as the government before a setting period of time, a fter that, then can directly get into the account to change the sdcrl to what they want, and this change of sdcrl can’t be revise before due time, for the reason of it is a public information, any change should have prepare time for participants. if the pre file of sdcrl be re jected, which should have an explanation of for what reason, business have to follow and revise as the order of re jection, until it is permitted. if the business do not want to ma ke any change of the sdcrl, this decis ion will be respect and kindly treated. the business’s sdcrl is licensed by the government, which is set for a period of time , and this could be one year or half year, it is up to the application and government’s permission, and the range of sdcrl could be 1 to 2 percent, in the case of taiwan sdcrl is 7 percent, which can be decided by business highest upward to 9 percent or lowest down to 5 percent, the adjustment of sdcrl is up to the business’s strategy and need of operation. 2.1. government’s point of views to sdcrl the responsibility of govern ment is provide an effic ient economic environment to business and investors, and that is good to the whole national economic partic ipants, include the benefit of non-stock trading part ies, e xcept that, this environment will let business easy in ra ise necessary fund to expend scale and investors have multip le choice in buy and sell trade in stock market. nowadays information engineering technology is fully developed, it is easy to set prior filing license in government stock e xchange. the business can set a sdcrl as his will before a government allowed t ime and range, and that sdcrl will work as setting at due time. the advantage of designed new sdcrl to government are c reate an activate and mu ltip le stock market, wh ich let the business have more space to create comme rcia l chances, sprit of economic freedo m let business have space to thinking, creating and chances to win, if the government accept the idea of sdcrl, the cost most are to control co mputer setting, but the benefit of new sdcrl will created millions of chances. if a business did not set such a sdcrl at permitted time, that will be treat as the business advances in technology innovation , vol. 1, no. 2, 2016, pp. 33 37 35 copyright © taeti give up the right of business setting sdcrl, this business’ sdcrl will be the original value of government set sdcrl. business partial controlled sdcrl was authorized and allowed by the government, which is still under the controlled of the government, when stock market is upward , if trend of upward will hurt the national economic seriously, the government can withdraw granted the rights and order the business or direct amend the business set sdcrl, then the government can change it to an acceptable sdcrl in that period of time, until the economic recovered. 2.2. business’s point of views to sdcrl business was born to win benefit and avoid risk, and they will take best care of themselves, new sdcrl let the business have chance to do and choose risk avoidance, if business operate an e xpand strategy, he can choose highest range of sdcrl, or business can operate an traditional way by choose lowest range of sdcrl. if permitted, business will ma ke some distance of adjustments of its sdcrl, the adjustment can show the business character and style in the stock trading ma rket, this can be a kind of corporate governance, and the pre -filing adjustment is a lready pre filed to the government. the effect of business controlled sdcrl is when an upward stock ma rket, the b iggest sdcrl will strengthen buy power, if the business want to e xpand the business scale, he can choose and strategy higher sdcrl, this sdcrl re flect the idea of business owner and his logica l mode l, that is a kind of business operation strategy, which activate the business world, there are a lot business in the world, not only different in size , policy, but also in culture background and industry, we can mix a ll diffe rent factors to create an colorful business world. basically, business owned stock are open trade in the ma rket, thus, the price of a stock should be some distance protect by the business, if the price is abnormally upward or down ward for reasons very often, which will be a sign of under value or over va lue of the business stock, and the other condition is the opportunist try to control and gain benefit fro m this business stock, thus, business need to develop its trading strategy in keep an amount of stock and capital to buy and sale its stock, that is the trade strategy to that business. short term trading is a co mmon phenomenon to a stock with e xtraord inary profit, but this kind of short term trad ing just like double edge sword, which co mpany with both benefit and hurt, thus, business have some d istance of responsibility to protect shareholder from bubble crash. usually, stock price shot up for the announcement of dividends, business can make a strategy of dividends announcement to achieve the goal of raise capital, on the other hand, release of dividends will slow down the business’ growth. the advantage of new designed sdcrl to business is it can be used to persuade customers, stockholder and investors by diversities of strategy in sdcrl, if this strategy been testified and last for a long period of time , which will be treated as a re liab le sign by business related parties, after that, no matter loan to the bank or new international investment will be treat the same as reliable. 2.3. investors’ point of views to sdcrl the effect of ne w designed sdcrl in investors are in buy and sale decision, if in an upward stock market, ne w designed sdcrl will strength the power of buy action, and if in downward stock market, the biggest of sdcrl will strengthen the power of investor in sale action. if investors are risk preference, they will choose the business with high sdcrl, especially in an upward stock market, if investors are risk avoidance, they will choose a lower sdcrl, for the new designed sdcrl let business have chance to create investors’ need space in risk return. 3. discussion this research discuss the adjustment of sdcrl, wh ich can distinguish the business's economic condition and demand, as an exa mple , if both filing and audit system are set to a period of one year, then, near ne xt year will be a ne w cycle for both business and government to rethink about the decision of setting sdcrl, if it is really good to the participants and government, why not set it be a longer t ime, the best point is this stock system testified by most business’s thinking. advances in technology innovation , vol. 1, no. 2, 2016, pp. 33 37 36 copyright © taeti 3.1. different mission of stock trading participants all upper factors are fulfilled depend on the condition of the country and companies , for both of these two group have different mission to e xerc ise, and theses missions have big diffe rence between them. business want to raise funds to invest in the industry and survive in the ma rket co mpetit ion, or business may declare dividends higher in a setting time to attract investors (merton h. miller, kevin rock, 1985), may that is a way of interaction between companies and investors, and these upper actions are focus on increase influence of business (merton h. m ille r, myron s. scholes, 1982). mission of government is based on whole national economic and regulation setting, this is not for a single industry or company, but diffe rent industry may be huge difference, such as bank want higher interest rate, but foreign trade industry is totally different, they need lower interest rate to strengthen the power of e xport ability, on the other hand, about the individual investors, they also have different preferences and needs on stock investment (eugene f. fa ma ,1971), may some people tend to risk pre ference, e xcept that, there still e xist diffe rent way of measure risk to gain pro fit fro m stock ma rket, thus, it is hard to have a qualitative rules or principles of sdcrl in stock trade ma rket, which can inc lude and cover all situation, one size of sdcrl for stock trade is a single standard of sdcrl, wh ich is hard to fits so many different groups and cover there necessities. 3.2. innovation of sdcrl our innovated approach, allowing the original sdcrl in stock market be mo re live ly, diverse thinking achieve simu ltaneous sound be organized in stock trad ing market by corporate governance, with which to active capita l transactions purposes. if the business have a strong self-confidence, the business can choose the biggest sdcrl in stock trading market, or some business hope stable environment to grow, then he can choose a minimu m sdcrl to reduce business fluctuations. the upper description is match the spirit of our idea, then business can choose a stable environment without too strong vibration to them in operation, (michael c. jensen, willia m h. mec kling, 1976), lowest sdcrl will be best choice. on the other side, the government still can control the sdcrl fro m policy and economic way to satisfy what is necessary for the country, three win is the goal of this research. after the atlas of new designs is obtained, a detailed design can be carried out by selecting one from the atlas. 4. conclusions there are positive implication of ana lysis in new innovative stock trade system, which is good not only the national economic but a lso to business and individual investors. integration of all ideas to establish a new stock trading system, co mbine all opinions of all stock trade partic ipants, it is possible to form a consensus and generate mutual trust of all parties to create biggest wealth, (ca m ca ldwe ll, mark h. hansen (2010). 4.1. stock trading efficiency for the goal of improve the existing rig idities stock trad ing policy, especially in nowadays’ hard environment, investors need to face waves of stock trade decline, not only the government or investors are helpless, new method can improve so me o f the missing, we wish all part ies of the stock trading market are action in accordance with the wishes of themselves without prejudice to national economic interests, most important is the way to achieve effective of stock trading . further, the government still can increase or decrease the proportion of control in sdcrl when necessary. 4.2. government’s control right diffe rent stock have its best sdcrl in adjustments, in the case of financial stocks are large a mount transaction in daily stock trade, we can set a smaller proportion of the amount sdcrl for them to adjust, one percent is better for huge enterprises in setting sdcrl, and perspective traditional industry would be better to have substantial authorizat ions, such as 1.5 percent of sdcrl. the govern ment can always open or tighten the magnitude authorization. business can base on the economic environment to set timely favorable ad justment to create the advances in technology innovation , vol. 1, no. 2, 2016, pp. 33 37 37 copyright © taeti best living environ ment, name ly with the method to achieve win -win situation to a ll trade member. 4.3. suggestion of sdcrl every business have its own way to measure performance (stephen j. brown, jero ld b. warner, 1980), consider about the business ’s need in raise fund and individual investor’s goal, e xcept that, investors have it’s different goal, thus, we need a d iversify stock trade mechanism. fro m business’s point of view in governance have to consider about the best and least risk way to raise fund (robert s. ha mada, 1971), firm have to win profit fro m the market, investors have different preference of his goal, combine three part ies necessary with suggested sdcrl stock trade system, include government regulation, business and individual investor’s goal, this article suggest a new stock trade system, the strength of the new system distinguished diffe rent trade me mbe r and it’s goal. th is paper suggested that the sdcrl principally decided by the government, but allo w businesses adjust according to their own need. possible future research design for this issue may focus on the scope of e mpirical study of sdcrl t rading system. in this paper, the result has shown that the new designs can produce a more wide range sdcrl of non-uniform stock trading system motion than the existing stock trading system. references [1] c. cald well and m. h. hansen, “trustwor thiness, governance, and wealth creation,” journal of business ethics, vol. 97, no. 2, pp. 173-188, 2010. [2] e. f. fa ma, “the behavior of stock-ma rket prices,” the journal of business, vol. 38, no. 1, pp. 34-105, 1965. [3] e. f. fa ma , “ risk, return, and equilibriu m,” journal of politica l economy, vol. 79, no. 1, pp. 30-55, 1971. [4] e. f. fa ma and j. d. mac beth, “risk, return, and equilibriu m: e mpirica l tests,” the journal of politica l economy, vol. 81, no. 3, pp. 607-636, 1973. [5] g. j. kuserk and p. r. locke, “ma rket ma king with price limits,” the journal of future market, vol. 16, no. 6, pp. 677-696, 1996. [6] j. guy, “making: the role of governance,” journal of financial planning, vol. 8, no. 1, pp. 34, 1995. [7] h. m. markowitz, “foundations of portfolio theory,” the journal of finance, vol. 46, no. 2, pp. 469-477, 1991. [8] h. qu and x. li, “ building technical trading system with genetic programming: a new method to test the efficiency of chinese stock markets,” co mputational econo mics , vol. 43, pp. 301-311, 2014. [9] j. sutthisit, s. wu, and b. yu , “positive feedback trading in chinese stock ma rkets: empirical evidence from shanghai, shenzhen, and hong kong stock e xchanges,” journal of financia l and economic pract ice, vol. 12, no. 1, pp. 34-57, 2012. [10] m. e. ce kirdekc i, and v. iliev, “trad ing system development: trad ing the opening range breakouts,” degree of bachelor of science, worcester polytechnic institute (wpi), 2010.  advances in technology innovation , vol. 1, no. 1, 2016, pp. 07 12 7 copyright © taeti benefit analysis of emergency standby system promoted to cogeneration system shyi-wen wang department of electrical engineering, chien kuo technology university, changhua, taiwan. received 03 march 2016; received in revised form 25 april 2016; accept ed 30 april 2016 abstract benefit analysis of emergency standby system comb ined with absorption chille r promoted to cogeneration system is introduced. economic evaluations of such upgraded projects play a majo r part in the decisions made by investors. t ime-o f-use rate structure, fue l cost and system constraints are taken into account in the evaluation. therefore, the proble m is formulated as a mixed-integer progra mming problem. using two-stage methodology and modified mixed-integer progra mming technique, a novel algorithm is developed and introduced here to solve the nonlinear optimization proble m. the net present value (npv) method is used to evaluate the annual benefits and years of payback for the cogeneration system. the results indicate that upgrading standby generators to cogeneration systems is profitable and should be encouraged, especially fo r those utilities with insufficient spinning reserves, and moreover, for those having difficu lty constructing new p ower plants. keywor ds : emergency standby system, cogeneration system, time-of-use rate structure, mixed-integer programming, nonlinear optimization 1. introduction a marked increase in the cost of constructing a generation, transmission and distribution system have also resulted in higher de mand charges to customers in the past few years. to reduce the system's peak load and therefore reduce the system's idle stand -by capacity, seasonal and time -of-day (or peak-load pric ing) rate structures are usually applied by the utilities. this has increased the benefit of peak shaving by using dispersed-storage-and-generation (dsg) systems [1]. when made a part of an energy management system (ems)[2] or distribution dispatch center (ddc) of a utility system, dsg may provide benefits to utilities by reducing system peak load, improving reliability, and increasing operation efficiency, which implies that the opportunity cost of generation can thus be reduced by the dsg. cost reductions or avoided costs result fro m saving of both capital carrying charges and operation e xpenses. however, additional capita l investment will be required fo r the dsg, so the benefit obtained by the utility should be shared with partic ipating customers to promote the dsg. t wo customer incentives for dsg a re that it a llo ws a time -of-use (t ou) rate structure based on the peak-load pric ing theory and reduced fuel costs. considering the tre mendous quantities of waste heat generated in the production of electric ity, it is apparent that there is an opportunity to save fuel by a cogeneration system (cgs). cgs can be described as the simu ltaneous generation of electrica l or mechanica l power and usable energy by a sin gle energy conversion system [3]. it has long been common here and abroad. currently, there is renewed interest in cgs because the overall energy efficienc ies are claimed to be as high as 70 to 85%. reducing the cost of electricity for industrial and comme rcia l users, re liev ing e xcessive demand on utilit ies, and using fuel effic iently a re certainly worthwhile goals. however, carefu l planning in design and operation is necessary to meet these goals. therefore, in this paper, a novel optima l operation scheme for sma ll cogeneration systems that are upgraded from standby generators is introduced. * corresponding aut hor, email: wshin@ctu.edu.tw advances in technology innovation , vol. 1, no. 1, 2016, pp. 07 12 8 copyright © taeti 2. method fig. 1 illustrates the structure of a small cgs investigated in this paper. the sa mple cgs system is upgraded from a 100 kw gas -engine-driven standby generator by adding an absorption refrigerator and other necessary accessories. the hot-water capacity of this absorption refrigerator is 50 rt . the electricity demands of the system a re usually supplied fro m a public electric utility. however, the cogenerator may operate in a para lle l manner to optimize the energy supplies of both the electric ity and cooling de mands for some suitable time period. in order to optimize the operation procedure, the parameters of the system are obtained by fie ld testing the sample cgs under fu ll and some partia l load conditions. on the basis of the data of the fie ld tests, the fuel cost curve of this system can be obtained by the curve-fitting techniques as shown in fig. 2 by the dotted line when the fuel cost is 7.84 nt$/m 3 . in th is system the heat recov ered by the absorption chiller is transformed to equivalent p o we r and so is de duc te d fro m th e po we r d e ma nd o f t he c en t rifug a l ch ille r. on t h e gas engine absorption chiller exhaust boiler centrifigual chiller gcity gas generator purchased electricity mixing tank pump water steam exhaust air exh aust air space cooling demand exhaust air electricity demand fig. 1 structure of a gas-engine cogeneration system fig. 2 fuel-cost curve of the sample cgs system contrary, for the conventional standby generator the o n ly output is electric power. its power output curve is also shown in fig. 2 for comparison with that of the cgs. 2.1. problem formulation for customers with a cgs that is upgraded fro m a standby generator, the total operating cost can be e xpressed simply as the sum of the payments for electric ity to the electric utility and for gas consumption to the gas company. the monthly electricity cost that is the cost function of the optima l operation scheduling of a cgs is proposed as follows:     ),0(++ + +),( 30 24 1 cddmaxrcdr hppsd tphpocpsf mpcd gigiii i igigii      (1) where i: time interval psi: pseudo switch used to indicate the on/off (1/0) status of cgs during time interval i oc(pgi, hgi): operation cost of the cgs during time interval i (nt$/h) pgi: output power of the cgs during time interval i (kwh) hgi: equivalent power of recovered heat of the cgs during time interval i (kwh) tpi: tou rate structure during time interval i (nt$/kwh) di: demand during time interval i (kwh) rcd: rate when demand is under the contract capacity rp : rate if demand exceeds contract capacity cd: contract capacity dm: monthly maximu m demand on the basis of the formulat ion of the cost function, the optima l operation scheme of the sample cgs and the optima l use of power fro m the utility can therefore be determined, ultimately, so as to minimize the monthly electric ity cost. control variables of this optimization proble m include binary and continuous variables that correspond to the on/off status and the load level of cgs, respectively. thus, this problem is formulated as a mixed-integer programming problem. 2.2. optimal operation scheme using a two-stage methodology and a modified mixed-integer progra mming technique, 0 50 100 150 200 250 300 20 40 60 80 100 120 140 c o st ( n t /h ) power output (kw/h) sta n… advances in technology innovation , vol. 1, no. 1, 2016, pp. 07 12 9 copyright © taeti a novel algorithm is developed and introduced here to solve the nonlinear optimization proble m. in the first stage, the reduction of the capacity charge for power due to the decrease of contract capacity caused by installation of the cgs is not take into consideration. therefore, at this stage a linear progra mming algorithm can be applied. however, in the second stage the neglected factor is taken into account and the decomposition and alternative policy method is adopted [4]. the total operating cost of cgs during each time interval can be expressed as :   )p(dtppocf giiigii  1,2...24i  (2) where pgi is the sum of the electric power and heat output of the cgs. the heat should be converted to its equivalent electric power before the summat ion can be made. the fuel cost curve shown in fig. 2 can therefore be formulated as : oc(p ) k p k p k gi 1 gi 2 gi 3      2 1,2...24i  (3) the coefficients k1, k2, and k3 depend on the type, capacity, and maker of the cgs. substituting (3) into (2), and then minimizing the obtained equation by linear programming techniques, the optima l operating capacity of the cgs for each time interval can be obtained as follows: p (k tp ) / (2k ) giom 2 i 1   i 1,2...24 (4) fro m (4), it is clear that fuel cost (k1,k2) and tou rate structure (tpi) are two important factors in determining the optimal operation capacity. the constraint of the generator output of the cgs is p p p gmin giom gmax   i 1,2...24 (5) where p gmin ： minimum generator output of the cgs p gmax ： maximum generator output of the cgs the minimu m and ma ximu m outputs of the sample cgs are 30 kw/h and 150 kw/h, respectively. the time interval is set to one hour in this paper. the optima l operation benefit during each time interval can therefore be expressed as : be tp p oc(p ) iom i giom giom    i 1,2...24 (6) equation (6) shows that the optima l operating benefit is gained from the difference between the savings gained by reducing the electric ity charge and the cost of operating the cgs. the following steps describe the optima l operation scheme in which the benefit gained by the reduction of contracted electricity capacity is considered: step 1: eva luate the optima l operating capacity using (4). in this step, the reduction of the electric ity capacity charge due to the decrease of contract capacity made possible by installation of the cgs is neglected. step 2: sort the hourly loads so that they are in descending order, that is d d j j+1  j 1,2,...23 (7) and record the corresponding hour of each hourly load. the demand diffe rences between ma ximu m hourly load and every hourly load can be calculated as : ed d d j 1 j 1    j 1,2,...23 (8) in which d1 is the maximu m hourly load. therefore, if the reduction of the electricity demand charge due to the decrease of contract capacity made possible by installation o f the cgs is neglected, the optimal operating capacity of the cgs during each pseudo time period (j) is : p max (p , ed ) gj gjom j  j 1,2,...23 (9) step 3: the optima l operating benefits for various operating hours are advances in technology innovation , vol. 1, no. 1, 2016, pp. 07 12 10 copyright © taeti be tp p oc( p ) + r ed / 30 k j gj gj j 1 k cd k       k 1,2,...24 (10) step 4: select the ma ximu m operating benefit from step 3. be max (be ) max k  k 1,2,...24 (11) therefore, the optimal operation scheme of the cgs for each pseudo time is : 1,...24+kk,=j 1,2,...k=j 0 p p gj gj     (12) step 5: reorder (12) by the actual t ime order using the record built at step 2. the daily maximu m demand can be decreased as : )pmax(dd giimd  i 1,2,...24 (13) step 6: print out the optima l operation schedule, operation benefit, daily ma ximu m de mand, and then stop the process. 3. results and discussion on the basis of the proposed algorithm, a program has been developed. three samp le customers: a hotel, a hospital and an office building are used to demonstrate the proposed algorith m. the sample cgs is assigned to operate six months per year, fro m june through september in a ll cases. tou rates of taipowe r are applied for a ll samp le customers. furthermore, the cgss of the hotel and hospital are assumed to operate 30 days a month and 24 days for that of the office building. 3.1 the demand pattern of sample customers fig. 3 shows the daily demand patterns for air conditioning, electricity, and total demand of the three sample customers in the summer. relatively larger hourly demands for electricity occur in the sample hotel in the period between 11:30 and 23:00, and the total demand holds almost constant between 11:00 and 21:00. on the other hand, the total demand of the sample hospital is almost constant between 9:00 and 21:00. there are lunch hours between 12:00 and 14:00. the d emands of the sample office building are more concentrated, and result in a lower load factor. (a) hotal (b) hospital (c) office building fig. 3 daily demand patterns on the summer 3.2 optimal operation capacity of the sample cgs fig. 4 shows the optima l operation capacity and operation benefit for various fuel costs. for e xa mple , fig. 4-(a ) indicates that if the fue l cost is 6.272 nt$/ m 3 the optima l operation capacity is 53 kw/h and operation benefit is negative (-72). there fore, it is not worthwhile to operate the cgs in off-peak periods. fig. 4-(b) shows that the operation benefit is positive for peak periods. hence it is worthwhile to operate the sample cgs when the fuel cost is lower than 8 nt$/m 3 . for example, if fuel cost is 6.272 nt$/m 3 , the optimal operation capacity is 150 kw/h, that is, the ma ximu m capacity of the sample cgs. in this case the operation benefit is 62 nt$/h. advances in technology innovation , vol. 1, no. 1, 2016, pp. 07 12 11 copyright © taeti (a) off-peak period (t ou rate structure is 0.77 nt$/kwh) (b) peak period (tou rate structure is 1.89 nt$/kwh) fig. 4 optimal operation capacity and operation benefit for tou rate structure of taiwan 3.3 optimal operation scheme of sample customers figs. 5 sho w th e opt ima l op erat ion sche me fo r fue l costs o f 9.408 nt $/ m 3 . fo r each hou rly lo ad, op t ima l gen erat ing c apac ity of cgs, and th e a mount o f po we r be ing purchased are shown. fig . 5-(a) sho ws th at the ma xi mu m d a ily op e ra t ion be ne fit o f th e sa mp le hot e l is p ro duc ed by ope rat in g th e cgs e le ven hou rs (fro m 12:00 t o 22:00) pe r d ay . th e g ene rat in g c apa c it ies at t ime inte rva ls 14, 15, 17, an d 18 a re lo we r th an o pt ima l cap ac ity . he nc e , the ge ne rat ing cap ac it ies o f the cgs a re set a t the o pt ima l ca pa c ity , th at is , 117.99 kw / h . 3.4 economic analysis the net present value (npv) method is used to evaluate the annual benefits and years of payback [5]. fig. 6 shows the annual benefits of using the cgs for the samp le customers at various fuel costs. in general, the more the load coincides in t ime is, the greater the benefit obtained. the numerica l results of the payback analysis for the sample customers are sho wn in fig. 7. (a) hotel (b) hospital (c) office building fig. 5 optimal operation scheme at fuel cost 9.408 nt$/m 3 fig. 6 annual benefit advances in technology innovation , vol. 1, no. 1, 2016, pp. 07 12 12 copyright © taeti fig. 7 return of investment for various fuel cost 4. conclusions a novel optima l operation scheme for sma ll cogeneration systems upgraded fro m standby generators has been introduced in this paper. on the basis of the proposed algorithm, a p rogra m has been developed. then three sample customers are used to demonstrate the proposed algorith m. the benefits and cos ts of these sample cases are exa mined. the fuel cost, time-of-use rate structure and system constraints are taken into account in the evaluation. the results indicate that upgrading standby generators to cogeneration systems is profitable and should be encouraged, especially for those utilit ies with insuffic ient spinning reserves, and moreover, for utilit ies having difficulty constructing new power plants. acknowledgement the support of the ministry of science and technology (taiwan), under grant most 104-2221-e-270-001 is gratefully acknowledged. references [1] t. gonen, electric powe r distribution system engineering, ne w yo rk: mcgraw-hill, 2008. [2] r. c. dugan, m. f. mc granaghan, s. santoso, and h. w. beaty, electrical po wer systems quality, ne w york: mc graw-hill, 2004. [3] public utility regulatory po lic ies act, report no. 95-1750, confe rence report, oct. 10, 2001. [4] s. ashok and r. banerjee, “ optima l operation of industrial cogeneration for load manage ment,” ieee trans. powe r syst., vol. 18, no. 2, pp. 931-937, may 2003. [5] k. methaprayoon, w. j. lee, s. rasmiddatta, j. r. liao, and r. j. rose, “multistage artific ial neural network short-term load fo recasting engine with front-end weather forecast,” ieee t rans. ind. appl., vol. 43, no. 6, pp. 1410– 1416, nov./dec. 2007. [6] c. t. hsu, c. s. chen, and c. h. lin, “electric powe r system analysis and design of an expanding steel cogeneration plant,” ieee trans. ind. appl. , vol. 47, no. 4, pp. 1527-1535, jul./aug. 2011. [7] c. h. hsu, “a graph representation for the structural synthesis of geared kine matic chains ,” journal of the franklin institute, vol. 330, pp. 131-143, jan. 1992.  advances in technology innovation , vol. 2, no. 2, 2017, pp. 34 39 34 synchronized injection molding machine with servomotors sheng-liang chen, hoai-nam dinh * , van-thanh nguyen institute of manufacturing information and systems, national cheng kung university, tainan, taiwan received 30 march 2016; received in revised form 25 april 2016; accept ed 01 may 2016 abstract injection mold ing machine (imm) is one of the most important equipment in plastic industry. as a cyclic process, injection mo lding can be divided into three steps includes filling process, packing-holding process and cooling process, among wh ich filling process and packing process are both most important phases for the quality of part, and the corresponding crucial process variables are injection velocity and packing pressure in filling and packing phases. moreover the determining a suitable injection time, screw position and cavity pressure for transfer from injection v elocity control to packing pressure control which is commonly called filling to packing switchover point is also critical for high quality part. this study is concerned with two research aspects: double servomotors synchronization control for injection unit, and filling to packing switchover methods. the simulation result of switching method based on injection time and ball screw position those are similar, and the result of switching method based on the cavity pressure that is better. keywor ds : s ervo motors, synchronization, v elocity to pressure switchover, pressure control, injection molding machine. 1. introduction injection mold ing mach ine used servo motor to save energy and suit for prec ise products. plastic in jection mo lding an e xtensive range of modern injection mo lding machines means that most custom mold ing require ments can be met. plastic inject ion mo ldings will be manufactured using the most cost effective production methods . the injection process included heating process and injecting the material into the mold [1]. the process produces are very quickly with great accuracy. it is widely accepted that all-electric injection mo lding machines are the most energy efficient of the three technologies [2]. in [3], there analysis and give some solution about cla mping system. hydrau lica lly driven injection translates the movement offering high force combined with high speed or electric servomotors has created the groundwork for realizing linear electromechanical screw movement with stronger force and more accurate. in this research, we focus only 2 phases, filling phase and packing phase, each of them has distinct control require ment and how to switchover fro m one to other. in the filling phase, the position and velocity of the in jection screw are controlled ensure the correct me lt front v elocity. at the beginning of the packing phase, the cavity pressure is controlled at a higher boost pressure to ensure correct part weight. a fter that, the cavity pressure is controlled at a lower pressure to maintain the part quality and avoid over packing. it ends after the gate freezes off. 2. methodologies and system modelling design 2.1. methodologies control of an in jection mold ing machine consists of many aspects. however, in this research we focus on designing controller for injection unit and techniques in filling phase and packing phase. the required injection force is so large but s ervo motor is e mployed that means designing control system for inject ion unit is e xtre me ly important. even an injection mo lding mach ine with perfect in jection unit control system but that does not mean the product quality is assured. as discussed previously, to achieve high quality product, the process variable, cavity pressure is controlled. and the most difficult but most important in controlling cavity pressure is how to switchover fro m filling phase to packing phase. when the required inject ion force o r pressure is so large but because of energy saving, cost effective, environ ment pollution and fle xi* corresponding aut hor, email: dinhhoainam52@gmail.com advances in technology innovation , vol. 2, no. 2, 2017, pp. 34 39 35 copyright © taeti ble, the servo drive is a good choice for the injection unit drive system. however, the servo drive has power limitation, if the capacity of the injection mold ing machine is huge so one servo motor cannot supply enough power and it costs a lot. that is why double servo motor is emp loyed in the inject ion unit drive system. ho w to ma ke them go smoothly with the same mot ion is one of the research objectives and the other is in vestigat ing so me methods to switchove r between filling phase and packing for our applica tion. the specific objectives of the present work are:  design a control system to control two servo motors move with the same motion. double servo motors drive the in jection unit and they are controlled in order to achieve a high st able large injection force or pressure.  investigate some methods of switchover filling controlling and packing controller. 2.2. system modelling design fig. 1 the schematic of the injection molding control system [4] the present work uses pc-based control system. the servo drives communicate with the controller via ethernet. the controller sends command to the servo drives and receives the encoder feedback signals from the servo drives. the cavity pressure is measured by a load ce ll and it co mmunicates with the controlle r v ia rs232. the data are monitored on the screen and can be collected by input output programming. the control algorith ms and the human machine interface are written in visual c # 2005, as shown the overall in jection molding control system in fig. 1. servo motor system identification and characterizat ion: the first step in the system identification stage is collecting data that are input and output in operat ing system. then system identification is done by using the system identification toolbox in matlab. the transfer functions in z do main a re obtained. the systems here consist of servo drive and servo motor, and the velocity controller is integrated in the driver. so, the transfer functions are the velocity closed loops. system identificat ion verificat ion: verify the system identification result by comparing simu lation and e xperiment. that means with the same co mmands one passes by the p hysical system the others passes by the transfer function, and then compare the outputs . control system design: the transfer functions of two servo motors are obtained; control theory is applied to design the control system. more specific, first, design individual motor controlle r and then design controller for double servomotors control system. all steps are simu lated in matlab and simu lin k. after that, the controlle r a lgorith ms are written in co mputer progra ms, the performance of the controllers are tested, and the controller para meters can be tuned and adjusted. with acceptable synchronous error, two servomotors are connected to a timing belt and this system will drive a ball screw. 3. double servomotors control system 3.1. control system design according to the different para mete rs of system are input, are calculated and converted fro m control signals to drive motors, then applied to machine. four servo motors control moving p laten slip on ma in t ie bar. servo motor as drive in jection machine, servo motor speed stability and high prec ision position control, in addition to control effect of a llowing high re producibility can also fine-tune the speed control and position control. fig. 2 structure of motors block control in injection machine advances in technology innovation , vol. 2, no. 2, 2017, pp. 34 39 36 copyright © taeti the purpose of motion control system is used to calculate the random variable consists of slide motion, velocity. deve loping state-space form, the governing equations can be written by using the rotational speed and electric current are the state variables. u l i l r l k i k j b idt d m mm m                                   1 0 (1) 𝑦 = [1 0] [ 𝜔𝑚 𝑖 ] (2) where l is the winding inductance, r is the winding resistance, b is damp ing coeffic ient, i is the current, u is the armature voltage, k is the torque gain, 𝜔𝑚 is the rotational ve locity, j is the armature (rotating part of motor) mo ment of inertia. some e xperiences have done, we find the transfer function of two servo motors with 2000rp m, and position outputs are measured. the transfer functions from veloc ity command to velocity output of two a c servo motors can be expressed in laplace variable s. 243.002812.0 06412.062.14 21    ss s g (3) 4279.06198.0 708.123.14 22    ss s g (4) the cross coupled control system min imizes synchronous errors in mu lt i-a xis mot ion control system [5]. it is co mmon used in cnc motion control system. technology of cross couple control is developed for mu lti a xis [6]. the paralle l technique control in mult i paralle l system was proposed [7]. the technique used feedback signal as positional signal and veloc ity signal to modify the command signal. this control has simp le structure, and easy to imp le ment. ho wever, some possible effects can decrease the performance of this control method, for e xa mp le, unmatched sinusoidal disturbances; reduction degrades the stability, unmatched model. utilize the advantages of the control methods above, combination between them was made. the new synchronous controller is mentioned in [8]. th is controller can overcome the difficulty of the master slave control method and utilize the advantages of the synchronous controller and the relative dynamic stiffness motion control. we have realized that cross -coupled control is used to increase the coupling relationship between axes. it a lso ma kes the capacity of disturbance better, decrease the error model. however, the measuring performance cannot improve in each axis. enconder 1/s controller1 c1(s) filter1 motor 1 g1(s) angular velocity angular position+ enconder 1/s controller1 c2(s) filter2 motor 2 g2(s) angular velocity angular position + position average fig. 3 block diagram of cross -coupling controller of two motor the errors of coupling axis are determined by the difference between the angular position and the average angular position. to trac king the same tra jectory input, where appropriate filter s . in this case the control la ws adopted for a ll system 4. pressure control and velocity to pressure switch control 4.1. velocity to pressure after two servomotors are synchronized , we connect them to a ba ll screw by a timing belt in fig. 4 [9]. assume that, the gear ratio is unity, the pitch radius of the ball screw is equal to the shaft radius of the motor, and the thread lead of the ball screw is 5mm. so now, the motion command of the ball screw can be computed fro m the motion co mmand of the motor. the ball screw position x and the ball screw linear ve loc ity v are expressed by the following equations . fig. 4 schematic of the injection unit drive system [9]   360 h x mm (5) advances in technology innovation , vol. 2, no. 2, 2017, pp. 34 39 37 copyright © taeti   / dx v mm s dt  (6) in which, θ is the motor position in degree, and h is the thread lead of the ball screw in mm. when ball screw move, the cavity pressure will be changed when the mold is closed. assume we can measure the cavity press by a load ce ll. but the difficulty is how to ca ll ca lculate the pressure command. in this study we don’t jump into the theoretical way, for e xa mple , with a given ma terial, mo ld, and para meters of the mechanisms we have to calculate the pressure command via mathe matics way. but we will use a practica l approach. because the servomotors are stet-upped in the velocity control mode that means the final co mmand to the driver is ve lo city co mmand, so the pressure command has to be transferred into the velocity co mmand. the relationship between them is found by finding the relationship between the pressure feed back and the velocity feedback. the velocity of the ball screw is computed fro m its position by taking a discrete derivative. the procedure to connect pressure to velocity is shown in fig. 5. fig. 5 conversion of pressure into velocity procedure 4.2. pressure control cavity pressure measurement is crucia l for process parameter. the flo w o f the plastics is increase in high pressure injection. in lo w pressure case, the flo w of the p lastic also reduces . and the cavity pressure during the packing phase is the key variab le to ensure the co mplete filling and the we ight of the final product. but we can see the performance of the pressure controller in a whole cycle. a pressure controller is designed to make sure the output cavity pressure follo w the co mmand cavity pressure command. the closed loop pressure is show in fig. 6. the pressure controller is a simp le p roportional controller and the proportional gain is tuned by try-and-error method. fig. 6 the closed loop pressure 4.3. switchover from velocity to pressure filling to pac king switchover is known as velocity to pressure switchover, where velocity refers to injection velocity and pressure to packing pressure. in the filling phase, the v elocity of the inject ion screw is controlled ensure the correct me lt front velocity. at the beginning of the packing phase, the cavity pressure is controlled at a higher boost pressure to ensure correct part we ight. after that, the cavity pressure is controlled at a lowe r pressure to maintain the part quality and avoid over packing. it ends after the gate free zes off. the ve locity to pre ssure switchover structure is shown in fig. 7. the most important in velocity to pressure switchover is how to locate the switching point. if switchover occurs too late, cause an over-packed cavity, characterized by a pressure peak in the packing phase. the pressure peak will not reduce to the lower holding pressure until the switchover because the high injection pressure is still applied after volu met ric filling. switching over early may g enerate an under-packed cavity, characterized by a pressure drop in the packing phase. load cell function change unit from pressure to velocity kp driver position velocity pressure fig. 7 flow chart of inject processing machine as we know that the reference input co mmand is velocity, but output is pressure. therefor we must find the relation between velocity and pressure. following [6], re lation between displacement, a xia l motor input differentia l pressure and flow rate is founded as 1 2 1 ( ) [ ( ( )) ( ) 2 ( ( )) ( ) ] q t g y t ps p t g y t ps p t     (7) where q(t ), p(t) a re flow rate and d ifferentia l pressure of motor input, ps is source pressure, g1(y(t)), g2(y(t)) are conductance of all edges of value calculates by advances in technology innovation , vol. 2, no. 2, 2017, pp. 34 39 38 copyright © taeti   1 2 ( ( )) ( ( )) 0 ( ) g k y t b if y t b g y t if y t b         (8) 2 2 ( ( )) ( ) ( ( )) 0 ( ) g k y t b if y t b g y t if y t b        (9) 5. simulation and experiment results the present work uses pc-based control system. the servo drives communicate with the controller via ethernet. the controller sends command to the servo drives and receives the encoder feedback signals from the servo drives. hardwa re is use in this study is provided by fo xnum co mpany. it includes controller, servomotors and servo drivers . servo driver can connect to controller via a co mmun ication card comm3. the simu lation stage follows the control system design stage. we show the s imulation results step-by-step, the error of master and slave motors in fig. 8. w ithout synchronous controller, the error will increase. in fact, even we select the same motors, but it will happen. with synchronous controller, the error is co mpensated. that makes the error reduce so much. the performance of the closed loop pressure with the proportional gain kp = 2.0 is show in fig. 8. at the beginning of the cycle , the output pressure cannot track the pressure command but when time goes on nearly 2seconds, t he output can track the co mmand. it is easy to recognize that in the injection phase the system is a non-linear system that is why only a simp le proportional controller cannot deal with. ho wever, our purpose is to control the cavity pressure during the packing phase, and at this interval, time is la rger than 2 second, the out pressure can follow the pressure command. if the system is non-linear in any phase, especially in packing phase, we must take its properties into account. until now, we can just only ma ke sure two motors can tracking the command. how is about the synchronous position error between the m? we cannot predict it in the physical system, because when some controlle rs are e mployed, they will take action every t ime to co mpensate the errors. even the feedback controlle r can eliminate the steady sate error, the feed forward controller can reduce the tracking error, but at the transient state, two motor response very diffe rently. using synchronous controller to decrease the synchronous error of motors fig. 8 performance of the closed loop pressure fig. 9 synchronous error with synchronous controller and without synchronous controller fig. 10 synchronous position error, without synchronous controller and with synchronous controller fro m fig. 10 we get the co mparison the maximu m absolute of synchronous error between using with synchronous controller and using without synchronous controller. fig. 11 trac king position error of motor 1 and motor 2 advances in technology innovation , vol. 2, no. 2, 2017, pp. 34 39 39 copyright © taeti we may consider now, when the synchronous controller is added into the system, it has effect on the performance of the overall control system or not? so, we have to study about some pro perties of the overall system. two important properties of a control system are time response and tracking erro r. so, we can see time response of motor 1 and motor 2 in fig . 10 and fig. 11. the position errors of two motors have a little suddenly values. the overall control is better than using open loop controller. table 1 the value of synchronous position error system without synchronous controller with synchronous controller maximum absolute of synchronous error 2.0304 0 1.2300 0 6. conclusions th is research focuses on two subjects: (1) design and imple ment control a lgorith m for double servo motors and their application to the injection unit control system. in particu lar, the present work utilizes these advanced algorithms on the injection unit that is driven by double servo motors; and (2) investigate some methods to switchover fro m filling phase to packing phase for our application. the acco mplishments in this research involve modelling, control a lgorith m i mp le ment at ion , s imu la t ing , and e xperimental imp le mentation. control system can be made more intelligently with artific ia l intelligence techniques. the advent of low cost very large scale integrated microprocessor and the proficiency of hard ware, control algorithms can be implemented in the hardware to get high response and performance. references [1] “injection molding” http://www.custompar tnet.com, 2009. [2] a. kanungo and e. swan, “a ll electric injection mo lding machines: how much energy can you save?” proceedings fro m the thirtieth industrial energy technology conference, ne w orleans, la, usa, may 6-9, 2008. [3] f. johannaber, “injection mo lding machines: a user’s guide,” 3rd ed. hanser munich vienna: new york, 1994. [4] “f70i injection mo lding mach ine control sy stem” http://www.foxnum.com/product021_en.aspx [5] y. ko ren,, “cross -coupled bia xia l co mputer control for manufacturing systems ,” journal of dyna mic systems, measurement, and control, vol. 102, no. 4, pp. 265-272, 1980. [6] marvin h.cheng, cheng-yi chen and aniruddha mitra, “synchronization controller synthesis of mult i-a xis mot ion system,” international journal of innovative co mputing information and control, vol.7, pp. 918-921, july 2011. [7] o. s. kwon, s. h. choe, and h. heo, “a study on the dual-servo system using improved cross -coupling control method,” 10th international confe rence on environment and electrical engineering (eeeic’11), may 2011, pp. 1-4. [8] m. f. hsieh, c. j. tung, w. s. yao, m. c. wu and y. s. liao, “servo design of a vertical a xis drive using dual linear motors for high speed electric discharge machining,” international journal of machine tools and manufacture, vol. 47, no. 3-4, pp. 546-554, 2007. [9] n. a kasaka, “a synchronous position co ntrol method at pressure control between mu lti-ac servomotors driven in inject ion mo lding machine,” in sice 2003 annual conference, vol. 3, pp. 2712 -2719, aug. 2003.  advances in technology innovation, vol. 4, no. 1, 2019, pp. 21 29 performance evaluation of mqtt as a communication protocol for iot and prototyping yuya sasaki 1,* , tetsuya yokotani 2 1 graduate school of electrical engineering and electronics,kanazawa institute of technology, ishikawa, japan 2 department of electronics, information and communication engineering, college of engineering, kanazawa institute of technology, ishikawa, japan received 03 march 2018; received in revised form 05 august 2018; accepted 14 september 2018 abstract the hypertext transfer protocol (http) has been widely used as a communication protocol for internet access. however, for internet of things (iot) communication, which is expected to grow in the future, http requires a large overhead and cannot provide efficiency. in order to solve this problem, lightweight communication protocols for iot have been discussed. in this paper, we clarify some problems of http for iot and propose mq telemetry transport (mqtt), which is a promising candidate for the iot protocol, after conducting a performance comparison with http. keywords: iot, http, mqtt, lightweight protocol, performance evaluation 1. introduction in recent times, there have been numerous discussions on the internet of things (iot) worldwide [1-2]. a large number of iot devices are connected to networks, and iot devices collect data from various sensors. though the data size of sensor devices is very small, it is communicated in large quantities through the network. currently, internet access requires tcp/udp/ip and application protocols over these protocols. amongst the application protocols, hypertext transfer protocol (http) [3] is common. it is applied for general internet access. in iot communication, a large number of devices communicate using very small data packets. when http is applied to iot communication, protocol overhead causes serious degradation of performance. moreover, the ip address in iot is thought to depend on the physical location and causes complexity in network control. in order to solve these problems, data aware networking (dan), [4] e.g., information centric networking (icn) [5-6], named data networking (ndn) [7], and content centric networking (ccn) [8], is being considered and discussed to be used as the architecture of iot. mq telemetry transport (mqtt) [8] is categorized as dan. mqtt has been developed for iot. protocol overhead is reduced, and information is transferred in a name-based system referred to as topical. it is not required that the address in mqtt depends on the physical location as the ip address. offered traffic in the network is reduced to transfer information. this paper describes, performance evaluation of iot communication based on mqtt, and compares the performance of http and mqtt protocol. this paper reports prototyping of iot devices based on mqtt and http. finally, it proposes using a combination of mqtt and conventional ip. 2. related technologies and standardization trends in order to realize iot communication, active discussions have been held between the industry, academia, and government. discussions on various topics such as technology development, international standardization strategy, and * corresponding author. e-mail address: b6700656@planet.kanazawa-it.ac.jp advances in technology innovation, vol. 4, no. 1, 2019, pp. 21 29 22 promotion of demonstration experiments are held by the iot promotion consortium [10]. in this section, taking these arguments into consideration, we summarize the discussion on environmental in this paper. 2.1. communication network for iot the communication network scenario for iot is reported in [11]. fig. 1 shows the system configuration required for realizing iot communication. in fig. 1, the area network is introduced for iot communication. however, wide area network infrastructure coexists with various services of legacy and iot. an outline of the communication protocol is shown in fig. 2. in the wide area network infrastructure, ip communication is widely spread. presently, iot communication cannot be differentiated from ip communication. however, http has been used for wide internet access, and every time information is accessed, a need for a three-way handshake with the tcp is required. in case of iot communication, a large number of small packets are generated because they handle traffic from the sensors. in the case of http, a large communication capacity is required to attach a header to each of them. for this reason, as shown in fig. 2, lightweight protocol for iot communication is necessary. fig. 1 system configuration of iot fig. 2 outline of communication protocol including for iot 2.2. international standardization trend the discussion on international standardization of lightweight protocol for iot is summarized in this section. the iot standardization can be represented as shown in fig. 3. discussions have been made on the architecture and requirements of communication systems in iso/iec, jtc1/sc41, and itu-t sg13. among them, there is a need for a lightweight protocol with low overhead. specifically, instead of using iot devices information processing infrastructure application gateway sensor etc… optical communication wireless communication area network (lan, han, pan, ban, …) optical communication wireless communication wide area network infrastructure optical communication legacy application protocol (http, ftp, smtp, etc...) lightweight protocol (mqtt, etc…) tcp/ip, udp/ip physical layer, data link layer wireless transmission protocol 6lowpan/udp, zigbee, etc… legacy communication communication for iot wide area network infrastructure area network advances in technology innovation, vol. 4, no. 1, 2019, pp. 21 29 23 location-dependent information such as ip address, “name-based access method” that directly accesses information is regarded to be more efficient. these are classified as icn. for example, [4] defines a framework for it. discussions have been made on the detailed methods, such as irtf and onem2m forum standards [12-13]. mqtt discussed in this paper is also regarded as an effective method. fig. 3 standardization overview on iot 3. overview of http and mqtt in this section, http and mqtt are summarized with respect to communication sequences. 3.1. overview of http the http protocol is used for the communication of information written in hypertext markup language (html). as a feature, in order to obtain information, it is necessary to specify the information location. uniform resource locator (url) is used to specify the address of that information. in principle, it conducts stateless communication. connection/disconnection of a communication session is performed every time information is accessed. however, when http is operated over tcp/ip, reliable communication is provided. http communicates with two responses, http request and http response. the http communication sequence is shown in fig. 4. fig. 4 communication sequences on http 3.2. overview of mqtt mqtt is a communication protocol of the publish /subscribe type, and the client communicates via mqtt broker [14]. as a feature of mqtt, the fixed header of the mqtt packet is a minimum of two bytes. mqtt uses proprietary name-based methods called topics to transfer information. a topic does not depend on the physical location of the ip address. the network load can be reduced by routing. from fig. 5, topic has a hierarchical structure. there are three types of topic iec iso de jure itu-t sg13, sg20 jtc 1 / sc41 vertical architecture platform type architecture forum ietf, irtf, w3c, ieee, aioti, ….. onem2m architecture · requirements detailed provision · protocol specification syn syn ack device server ack http request http response fin ack ack fin ack ack dns advances in technology innovation, vol. 4, no. 1, 2019, pp. 21 29 24 designation methods such as a perfect match, forward match, and partial match. moreover, mqtt can set three levels of delivery guarantee. therefore, it can be set according to the importance of communication. from the sequence diagram of fig. 6, mqtt does not connect/disconnect the session for each communication; however, it keeps a communication session connected. therefore, the communication session is simplified and communication wastage is reduced. from these features, mqtt applies to iot communication, small fixed header size, and it is thought that the communication becomes lightweight from retention of communication sessions. fig. 5 structural example of topic fig. 6 communication sequences on mqtt 4. comparison of http and mqtt traffic in iot communication, when a large number of devices are connected, overhead is caused. it causes problems such as network delay and server load. therefore, we evaluate their performance from the viewpoint of traffic and examine the suitable protocol for iot [15-16]. 4.1. definition of the network model fig. 7 shows the network model for the comparison with http and mqtt for communication traffic. in fig. 7, it is assumed that some devices such as sensors are connected to the server. it is a model that receives data from some sensor device via server. fig. 7 network model 4.2. http traffic http is applied to the network model of fig. 7, and traffic in data send/receive is calculated as follows. from the communication sequence of fig. 4, each communication packet is summed, and the obtained traffic is taken as traffic in one communication. here, the size of http request/response packets is set to 300 bytes each. it is assumed that 100 and 1000 sensor devices are connected to the transmitting devices. its payload size is 0 bytes. sensor temperaturecamera floor1 room1 *writing example of topic sensor/temperature/room1 room2 room3 etc… publish broker connect connack publish puback subscribe connect connack publish puback subscribe subback device1 device2 device n server user device advances in technology innovation, vol. 4, no. 1, 2019, pp. 21 29 25 4.3. mqtt traffic mqtt is applied to the network model of fig. 7, and the traffic is calculated [17]. from the communication sequence of mqtt in fig. 6, the sum of packet sizes of mqtt packets is taken as traffic of one communication. consider the case of operating on tcp/ip, one communication is from the publish devices to reception of data on the subscribe devices. table 1 shows each header size of mqtt. table 1 mqtt packet type message type bytes connect 14 connack 4 publish 6 + topic(1chara/byte) puback 4 subscribe 7 + topic(1chara/byte) suback 5 4.4. comparison result of http and mqtt traffic fig. 8 traffic volume of http and mqtt the traffic of http and mqtt as discussed in section 4-2 and 4-3 are as shown in fig. 8. from this, compare the traffic on http and mqtt, as the communication increases, difference in traffic between http and mqtt increases, and when communication occurs 30 times, mqtt becomes about 1/5 of the traffic compared to http. as can be seen from fig. 8, in mqtt, the traffic of both 500 and 1000 devices connected to the server is less than half of the traffic connected with 500 devices by http. therefore, it can be understood from this that traffic can be reduced using mqtt in iot communication. 5. comparison of network resources 5.1. definition of network resources by comparing the network resources of http and mqtt, we present the definition of network resources. it is defined as the occupation time in server from the start of communication to the end of device (fig.9). fig. 9 definition of network resources 0 5000 10000 15000 20000 25000 30000 0 5 10 15 20 25 30 t ra ff ic [ k b ] number of communications http 500devices http 1000devices mqtt 500devices mqtt 1000devices communication start communication end information access occupy resources advances in technology innovation, vol. 4, no. 1, 2019, pp. 21 29 26 5.2. measurement of access delay time in order to calculate network resources, we measure the access delay in actual communication [15] [18]. for verification, we used raspberry pi as a device and wireshark [19] of packet capture software to measure the communication access delay time (fig. 10). fig. 10 network configuration of access delay measurement in the case of http communication, the access delay time is measured when the server is accessed from the pc to acquire data. in the case of mqtt communication, the access delay time is measured from the publish device to the subscribe device via the mqtt broker. fig. 11 http access delay time fig. 12 mqtt access delay time pc (wireshark) http server http communication mqtt broker (wireshark) subscribe devicepublish device mqtt communication http request http response publish publish 4.909 4.91 4.911 4.912 4.913 4.914 4.915 4.916 4.917 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 t im e [s ] number of communications ave： 4.9115[s] 0 0.0005 0.001 0.0015 0.002 0.0025 0.003 0.0035 0.004 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 t im e [s ] number of communications ave： 0.00106[s] advances in technology innovation, vol. 4, no. 1, 2019, pp. 21 29 27 the results of the access delay time measurement of http and mqtt are shown in fig. 11 and 12. communication was carried out 20 times each, and the average was calculated. in http (fig. 11), the access delay was about 4.91 s on an average. in mqtt (fig. 12), the average was about 0.00106 s. in the first instance of communication for the mqtt, access delay is large in order to establish a communication session. because of looking at each sequence, in http, a delay of about 4.8 to 4.9 s occurs when a finack packet of a session disconnection request from a server is sent. this delay is considered to be a processing delay or keep time by using the server software. that does not mean that such delays will occur in all http communications. however, even if we subtract about 4.9 seconds from the measured http access delay time, we found that there is about 10 times short access delay time of mqtt communication. there are differences in the size of each protocol, in the case of http, it seems that there was a big difference in the access delay time owing to differences such as connection/disconnection of communication sessions. 5.3. required network resources next, network resources shown in fig. 9 are evaluated. we assume that access to information follows poisson distribution and time required for access follows an exponential distribution of average access time obtained in section 5-2. in addition, capacity of network resources is assumed to be sufficiently large. in this case, if the device generating information is k, it is represented by m/m/∞//k (finite population model). equilibrium state probability in this model is pk, the arrival rate of the information is λ, assuming that the average processing time is h, necessary network resource n is expressed by eq (1). 𝑁 = ∑ 𝑘 ⋅ 𝑝𝑘 = ∑ 𝑘 ⋅ (𝜆ℎ)𝑘 ( 𝑘 𝑘 )𝑘𝑘=0 (1 + 𝜆ℎ)𝑘 𝑘 𝑘=0 = 𝑘 ⋅ 𝜆ℎ 1 + 𝜆ℎ (1) fig. 13 shows the numerical calculation using arrival rate λ as a parameter with respect to equation (1). as can be seen from fig. 13, the mqtt requires less network resources for http. fig. 13 comparison of network resources 6. affinity with ip the internet is currently based on ip communication. however, the movement of devices frequently occurs in iot communication. assuming this, since ip communication depends on the location, smooth communication cannot be obtained. therefore, network architectures such as dan attract attention [4] [20]. we consider the relationship of mqtt with ip when applied to a wide area network. it is assumed that mqtt is applied to a wide area network, as shown in fig. 14. it is assumed that there are multiple mqtt brokers in many devices. 0 2000 4000 6000 8000 10000 12000 5 30 55 80 105 130 155 180 r e s o u rc e s [ s ] arrival rate http(k=5000) http(k=10000) mqtt(k=5000) mqtt(k=10000) advances in technology innovation, vol. 4, no. 1, 2019, pp. 21 29 28 in this scenario, there are two possible routing problems. case 1, the publish devices connected to the mqtt broker and the subscribe devices communicate. if devices that want to obtain data are connected to another mqtt broker, the connection needs to be changed. however, publish devices do not know to which mqtt broker the subscribe device is connected. case 2, the publish device has multiple sensors. for each sensor data, the publish device sends data to the mqtt broker. at this time, it is unknown where the subscribe device is divided for each sensor data is connected. fig. 14 wide-area network configuration of mqtt as a solution to these problems, there is a method of connecting the mqtt broker in a ring shape. alternatively, data may be sent to all adjacent mqtt brokers. in this case a large overhead will occur. therefore, it may be possible to link topic and ip and perform routing using a mechanism like dns. therefore, when mqtt is used assuming a wide area network, it is difficult to communicate with only name-based routing that is independent of ip [21] [22]. it is similar to the above routing mechanism, and mqtt has another version for wireless sensor network mqtt -sn (mqtt for sensor networks). for routing, a predefined topic id and short topic are introduced. therefore, there is a feature where mapping is done automatically by informing the topic id and topic to the client, gateway, and broker in advance without setting it. it has already been thought that when a name and address are attached independently, it can be further simplified or linked to another id or the likes. this idea is not limited to mqtt; it can be said that it is common to name-based protocols. as a more sophisticated mechanism, it is conceivable that a name is automatically allocated in name-based routing. 7. conclusions tu-t y. 3033 defines four goals and twelve design goals that reflect the new requirements of future network, which specifies the framework of dan [4]. the essence of dan is name-based communication that routes data objects in the network by name or identifier, and the name-based communication allows dan to locate data objects irrespective of the location. name-based communication has received considerable attention as it guarantees the continuation of communication without being interrupted by the change of data object position [23]. in this paper, we evaluate the performance from a viewpoint of traffic volume, access delay, and network resources on overhead with both the http and the mqtt protocols applied to the iot platform and compare them. from this, it is found that mqtt is superior to http when applied to iot. however, for this network model, as shown in fig. 7, and in wide area iot networks, as discussed in section 6, it is necessary to study the problem caused by transfer of name-based information future works. case1 (publish) case1 (subscribe) × case2 sensor1,2,,n (publish) ?? ? ? case2 sensor1 (subscribe) case2 sensor2 (subscribe) connect brokers to each other ： mqtt broker ： device advances in technology innovation, vol. 4, no. 1, 2019, pp. 21 29 29 references [1] d. miorandi, s. sicari, f. d. pellegrini, and i. chlamtac, “internet of things: vision, applications and research challenges,” ad hoc networks, vol. 10, no. 7, pp. 1497-1516, september 2012. [2] a. al-fuqaha, m. guizani, m. mohammadi, m. aledhari, and m. ayyash, “internet of things: a survey on enabling technologies, protocols, and applications,” ieee communications surveys & tutorials, vol. 17, no. 4, pp. 2347-2376, june 2015. [3] r. fielding and j. reschke, “hypertext transfer protocol (http/1.1),” ietf, rfc 7230-7235, 2014. [4] itu-t, “framework of data aware networking for future networks,” itu-t recommendation, y.3033, january 2014. [5] g. xylomenos, c. ververdis, v. a. siris, n. fotiou, c. tsilopoulos, x. vasilakos, k. katsaros, and g. c polyzos, “a survey of information-centric networking research,” ieee communications surveys & tutorials, vol. 16, no. 2, pp. 1024-1049, 2014. [6] m. amadeo, c. campolo, j. quevedo, d. corujo, a. molinaro, a. iera, r. aguiar, and a. v. vasilakos, “information-centric networking for the internet of things: challenges and opportunities,” ieee network, vol. 30, no. 2, pp. 92-100, march 2016. [7] m. m. s. soniya and k. kumar, “a survey on named data networking,” proc. electronics and communication systems 2nd international conf., ieee press, june 2015, pp. 1520-1522. [8] o. waltari and j. kangasharju, “content-centric networking in the internet of things,” 2016 13th ieee annual ccnc, pp. 73-78, march 2016. [9] “mqtt.org,” http://mqtt.org/. [10] “iot acceleration consortium,” http://www.iotac.jp/. [11] t. yokotani, “tutorial: m2m/iot technical overview and its standardization,” proc. the 18th asia-pacific network operation and management symposium, october 2016. [12] t. kinosita and t. hirata, “technical standardization of iot / m2m, industry alliance trend,” proc. ieice general conf. march 2017. [13] t. sato and k. yu, “introduction of standards activities for information-centric networking,” ieice general conference bt-2-3, 2017. [14] m. särelä, t. rinta-aho, and s. tarkoma, “rtfm: publish/subscribe internetworking architecture,” ist mobile summit, june 2008. [15] z. babovic, j. protic, and v. milutinovic, “web performance evaluation for internet of things applications,” ieee access, vol. 4, pp. 6974-6992. october 2016. [16] d. thangavel, x. ma, a. valera, h. tan, and c. tan, “performance evaluation of mqtt and coap via a common middleware,” 2014 ieee ninth international conf. on issnip, ieee press, april 2014, pp. 1-6. [17] t. yokotani and y. sasaki, “comparison with http and mqtt on required network resources for iot,” proc. control, electronics, renewable energy and communications, ieee press, september 2016, pp. 1-6. [18] r. nakamura and h. ohsaki, “performance evaluation and improvement of large-scale content-centric networking,” proc. information networking (icoin), ieee press, april 2017, pp. 103-108. [19] “wireshark,” https://www.wireshark.org/. [20] j. chen, s. li, h. yu, y. zhang, d. raychaudhuri, r. ravindran, h. gao, l. dong, g. wang, and h. liu, “exploiting icn for realizing service-oriented communication in iot,” ieee communications magazine, vol. 54, no. 12, pp. 24-30, december 2016. [21] antitharanukul, k. osathanunkul, k. hantrakul, p. pramokchon, and p. khoenkaw, “mqtt-topic naming criteria of open data for smart cities,” proc. computer science and engineering conf. (icsec), ieee press, december 2016, pp. 1-6. [22] n. tantitharanukul, k. osathanunkul, k. hantrakul, p. pramokchon, and p. khoenkaw, “mqtt-topics management system for sharing of open data,” proc. digital arts, media and technology (icdamt), ieee press, march 2017, pp. 62-65. [23] j. lópez, m. arifuzzaman, l. zhu, z. wen, and s. takuro, “seamless mobility in data aware networking,” proc. 2015 itu kaleidoscope: trust in the information society, ieee press, december 2015, pp. 1-7. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). https://ieeexplore.ieee.org/xpl/mostrecentissue.jsp?punumber=7112413 https://ieeexplore.ieee.org/xpl/mostrecentissue.jsp?punumber=7112413 6___aiti#10525___ advances in technology innovation, vol. 7, no. 4, 2022, pp. 295-302 an experimental study on the mechanical properties of low-aluminum and rich-iron-calcium fly ash-based geopolymer concrete jack widjajakusuma 1,* , ika bali 2 , gino pranata ng 1 , kevin aprilio wibowo 1 1 department of civil engineering, pelita harapan university, tangerang, indonesia 2 department of civil engineering, president university, cikarang, indonesia received 10 december 2021; received in revised form 24 may 2022; accepted 25 may 2022 doi: https://doi.org/10.46604/aiti.2022.10525 abstract limited studies have been conducted on low-aluminum and rich-iron-calcium fly ash (laricfa)-based geopolymer concrete with increased strength. this study aims to investigate the mechanical characteristics of laricfa-based geopolymer concrete, including its compressive strength, split tensile strength, and ultimate moment. the steps of this study include material preparation and testing, concrete mix design and casting, specimen curing and testing, and the analysis of testing results. furthermore, the specimen tests consist of the bending, compressive, and split tensile strength tests. the results show that the average compressive strength and the ultimate moment of the geopolymer concrete are 38.20 mpa and 22.90 kn· m, respectively, while the average ratio between the split tensile and compressive strengths is around 0.09. therefore, the fly ash-based geopolymer concrete can be used in structural components. keywords: geopolymer concrete, fly ash, rich iron, low aluminum, mechanical characteristics 1. introduction geopolymer concrete with volcanic ash (class n in the astm c 618-19 classification) was used during ancient roman times as a building material. it is seawater resistant with durability that reaches thousands of years [1]. many studies showed that the geopolymer concrete from fly ash has better resistance to seawater and chloride in comparison to normal concrete [2-4]. however, the chemical processes behind the formation of geopolymers are not clearly understood. these processes can be simplified into three stages [5-7]. the first is the dissolution of silicate and aluminum elements from fly ash dust in an alkaline solution to produce aluminate and silicate species. commonly used alkaline solutions include naoh, koh, and na2sio3. meanwhile, the second is the process of forming aluminosilicate oxide gels, and the third is the polycondensation process which is a gel network arrangement that produces three-dimensional aluminosilicate networks. furthermore, reactive aluminum plays an important role in the structure and strength of fly ash geopolymers [6, 8]. chemical compounds such as calcium and iron have other effects during polymerization processes. calcium will react with silicon and aluminum to form various phases of calcium silicate and aluminate hydrates due to the contribution of water. this chemical reaction is accelerated by the presence of aluminate and silicate types dissolved in the geopolymerization process. similar chemical reactions also occur in portland and calcium aluminate cement [9]. the presence of calcium plays an important role in accelerating the geopolymer pavement process [5, 10] due to its ability to harden at room temperature [11-12]. currently, knowledge about the role and location of calcium in geopolymer structures is still very limited [6]. * corresponding author. e-mail address: jack.widjajakusuma@uph.edu tel.: +62(0)21 5460901; fax: +62(0)21 5460910 advances in technology innovation, vol. 7, no. 4, 2022, pp. 295-302 several recent studies stated that iron oxide has an important role in the formation of geopolymers [13-14]. venyite et al. [15] stated that limited aluminate leads to the replacement of aluminum (al) by iron (fe) atoms to form ferro-silicate-aluminate. furthermore, studies on the role of iron oxide in the polymer formation process are still very limited. this is due to limited methods for analyzing geopolymer structures. the method most often used in geopolymer analysis is nuclear magnetic resonance spectroscopy (nmr), which will be disturbed in the analysis if there is a high iron element [15-16]. the study from gomes et al. [17] stated that iron oxide decreases the strength of geopolymer concrete, although venyite et al. [15] had a different result. apart from the chemical content of fly ash, several parameters that also determine the strength of geopolymer concrete are dust grains fineness, temperature and duration of curing, type and molarity of alkali activator, and ph [5, 10, 18]. this study is motivated to investigate the fly ash-based geopolymer concrete, due to the limited studies on rich-iron and low-aluminum fly ash-based geopolymer concrete with increased strength. in addition, this study is motivated by the need of using local waste in the form of fly ash as a substitute for cement in indonesia. according to a new regulation enacted by the indonesian government, i.e., government regulation no. 22 of 2021 on the implementation of environmental protection and management, the classification of the fly ash resulting from combustion at steam power plants has been revised from toxic waste to non-toxic waste. through this regulation, the indonesian government encourages the use of fly ash as much as possible. one possible application is to use fly ash as construction materials. since 2015, the indonesian government has launched a program to build a million houses per year for the people of indonesia. in particular, since the covid-19 pandemic occurred in 2019, the need for “fit-for-purpose” housing has been one of the needs that must be met because almost all activities, including work, study, and worship, are carried out within homes. this study aims to investigate the mechanical characteristics (e.g., the compressive strength, split tensile strength, and ultimate moment) of low-aluminum and rich-iron-calcium fly ash (laricfa)-based geopolymer concrete. the results are expected to aid the indonesian government in substituting normal concrete with laricfa-based geopolymer concrete and building residential houses that are environmentally friendly and cheaper than those made from portland cement. in this study, the specimens made are treated at room temperature and meet the requirements for compressive strength and ultimate moment. furthermore, the fly ash used has low al2o3 (< 10%), which is equivalent to the content in portland cement. it also has a very high fe2o3 (nearly 50%) and calcium content (>10%). to carry out the investigation, the study steps are divided as shown in fig. 1. \ material preparation: fly ash, naoh, na2sio3, cement, water, aggregate material testing: xray fluorescence (xrf) test for fly ash, specific gravity, water content, sludge level, sieve analysis mix design calculation for normal concrete geopolymer concrete casting for specimens specimens curing specimens testing: compressive strength, split tensile strength, bending feasibility of substitution normal concrete through laricfa geopolymer concrete mix design calculation for geopolymer concrete normal concrete casting for specimens fig. 1 study methodology 296 advances in technology innovation, vol. 7, no. 4, 2022, pp. 295-302 2. material and method the main study materials for geopolymer concrete formation are fly ash, alkaline solution (as an activator), and coarse and fine aggregates, which are described in this section. the fly ash used is obtained from the steam power plant of suralaya, banten, indonesia. furthermore, table 1 provides chemical compositions based on the results from the x-ray fluorescence (xrf) test. the content of sio2 + al2o3 + fe2o3 is equal to 76.14% which is greater than 70%. based on the astm c 618-19 standard, the fly ash belongs to class f, with high iron oxide impurities and low alumina content. its cao content is also high (18.24%), which causes geopolymers to quickly harden at room temperature [19]. in polymer concrete, the alkaline solution acts as an activator that dissolves and binds silica and alumina contained in fly ash so that a polymerization reaction occurs. the alkaline solutions used in this study are natrium hydroxide (naoh) and natrium silicate (na2sio3). furthermore, the coarse aggregate used for both the polymer and normal concrete is screened and has a maximum size of 1.50 cm. the fine aggregate used in this study is silica sand passing sieve no. 30 (600 µ m). both the coarse and fine aggregates are tested according to astm c127, astm c128, astm c33, and sni 1964-2008. therefore, the specific gravity test results under saturated surface dry (ssd) conditions for coarse and fine aggregates are 2.38 and 2.62, respectively, with a 0.55% water content by the aggregate weight and a 4.92% sludge content by the aggregate weight. cement such as portland composite cement is used as a binder in normal concrete. for the ultimate moment testing of concrete beams, the reinforcing steel used is bjtp-24 with diameters of 8 and 12 mm and the average yield stress (��) of 392.85 mpa. the mix design of the geopolymer (8 molarity/m naoh) and normal concrete (target compressive strength 30+10 mpa) in this study can be seen in table 2. furthermore, the naoh content prepared is 8 m and placed at room temperature for 24 hours before use. na2sio3, commonly called water glass, is one of the materials that make up an alkaline solution which can be in the form of a liquid or a solid. in this study, the water glass used is a liquid with 55% natrium silicate concentration and 45% water. natrium silicate is made by mixing sio2 with natrium (na2sio3) or potassium carbonate (k2co3) dissolved with high-pressure steam leading to a thick (semi-viscous) liquid nature [20]. table 1 chemical composition of fly ash compound name concentration (%) fe2o3 48.51 sio2 21.06 cao 18.24 al2o3 6.58 k2o 1.42 p2o5 1.02 so3 0.87 bao 0.69 mno 0.55 sro 0.47 mgo 0.30 zno 0.10 zro2 0.08 na2o 0.05 rb2o 0.03 cl 0.02 br 0.02 y2o3 0.01 297 advances in technology innovation, vol. 7, no. 4, 2022, pp. 295-302 table 2 mix design of the geopolymer and normal concrete material (kg/m 3 ) geopolymer concrete normal concrete coarse aggregate 853.95 1033.64 fine aggregate 727.44 497.68 fly ash (type f) 470.51 water glass 179.97 naoh 15.30 naoh water 44.69 water 205.00 cement 508.69 fig. 2 reinforcement steel for geopolymer concrete-1 and normal concrete-1 beams fig. 3 compressive strength test fig. 4 tensile strength test fig. 5 bending test on the concrete beam the concrete specimens used in this study are cylinders with a diameter and height of 100 and 200 mm, respectively, for determining the compressive and split tensile strength. meanwhile, the beam specimens for conducting the bending test have dimensions of 1600 mm × 125 mm × 250 mm. the geopolymer concrete beam dimensions are selected to analyze the casting process and test the object characteristics so that the geopolymer concrete beams can be compared with typical beams used in the structure of standard residential houses. all beam specimens use an upper reinforcement of 2 ø8, while for lower reinforcement there are two variations, namely 2 ø12 and 3 ø12. the lower reinforcement 2 ø12 is used for the geopolymer and normal concrete-1 specimens (fig. 2). meanwhile, for the specimens of geopolymer and normal concrete-2 beams, the lower reinforcement used is 3 ø12. in this study, the curing process for geopolymer concrete specimens in the form of cylinders and blocks are carried out by the placement at room temperature (± 25°c) until the day of testing. the curing period for cylindrical and beam specimens for the geopolymer concrete-1 and concrete-2 is 65 days. the curing process for normal concrete is carried out by keeping the concrete wet to enable optimality and water availability for the cement hydration process. furthermore, normal cylindrical concrete is placed in a container filled with water, while normal beam concrete is covered with fabric and watered every day. the duration of treatment for the normal concrete-1 and concrete-2 is 69 and 68 days, respectively. in this study, the mechanical characteristics of geopolymer concrete are obtained using the compressive and split tensile strength tests based on astm c39/c39m-0 (fig. 3) and astm c496/c496m-17 (fig. 4), respectively. furthermore, the bending test based on astm c78/c78m-2 is used to determine the ultimate moment of the concrete beam (fig. 5). the testing results of the mechanical characteristics of polymer and normal concrete are then compared. 298 advances in technology innovation, vol. 7, no. 4, 2022, pp. 295-302 3. results and discussion the test results of compressive strength (f'c), split tensile strength ����), and ultimate moment (mu) on the geopolymer and normal concrete specimens are shown in table 3. from the data in table 3, the compressive strength (�� �) of these two concrete specimens is almost the same. the average compressive strength of geopolymer concrete reaches 38.2 mpa, which is 13% lower than normal concrete. this value indicates that it can be rationally accepted as an alternative material to normal concrete. the split tensile and compressive strengths of geopolymer concrete are 9.27% and 8.54%, while the split tensile and compressive strengths of normal concrete are 10.87% and 13.77%, respectively (table 3). normal concrete has a bigger ratio than geopolymer concrete, but this is not a problem because the tensile strength is not the primary function of concrete (the reinforcement can provide the tensile strength). a bending test (fig. 6) is carried out to obtain the ultimate moment of the concrete beam (in the middle of the beam), which is calculated as: ( ) 2 2 2 2 3 2 4 u o u o u sw sw p l p ll l l m q q= + × × − × − × × (1) where pu is the force from the bending test (kn), � is the concrete self-weight (kn/m), l is the concrete length (m), and lo is the support-to-support length (m). likewise, for the ultimate moment (�� , those of geopolymer and normal concrete are close to each other. the average ultimate moment of geopolymer concrete reaches 22.90 kn· m in this study, which is relatively slightly better than normal concrete (table 3). this shows that the bonding between the plain rebar and geopolymer concrete is relatively better than normal concrete (fig. 7-8). the condition of the plain rebar which supports the occurrence of this strong bond with geopolymer concrete is the absence of rust. table 3 testing results of the geopolymer and normal concrete concrete type cylinder specimen ø × h (mm) �� � average (mpa) ��� average (mpa) beam specimen l × b × h (mm) �� average (mpa) �� (kn· m) geopolymer concrete-1 100 × 200 36.08 3.34 1600 × 125 × 250 392.85 17.15 geopolymer concrete-2 100 × 200 38.20 3.26 1600 × 125 × 250 392.85 22.90 normal concrete-1 100 × 200 43.93 4.77 1600 × 125 × 250 392.85 17.02 normal concrete-2 100 × 200 35.01 4.82 1600 × 125 × 250 392.85 22.65 fig. 7 bonding of the geopolymer concrete beam to the plain rebar fig. 6 configuration of the bending test on the concrete beam fig. 8 bonding of the normal concrete beam to the plain rebar 299 advances in technology innovation, vol. 7, no. 4, 2022, pp. 295-302 based on the testing results of mechanical properties, the compressive strength (�� �), split tensile strength (���), and ultimate moment (��) between the geopolymer and normal concrete are almost the same. due to the relatively high calcium content of fly ash in geopolymer concrete (18.24%), the designed strength can be achieved with curing at room temperature as shown in the results from other studies [22-23]. even though the al2o3 content in fly ash is very low (6.58%), the relatively high fe2o3 content (48.51%) enables iron atoms to replace ferro-silicate-aluminate aluminum atoms [6], which allows the strength of geopolymer concrete to reach above 30 mpa with the naoh activator that has relatively low molarity (8 m). the bending test shows that the deflection of geopolymer concrete-1 and 2 are 39 and 22 mm, while the deflection of normal concrete-1 and 2 are 16.6 and 12.4 mm, respectively. this shows that the geopolymer concrete beam and its modulus of elasticity are more flexible and smaller than normal concrete, respectively. all specimens of geopolymer concrete and normal concrete experience flexural cracks and crack patterns that are almost the same (fig. 9-12). the specimens of geopolymer concrete reach the ultimate moment and show dominant flexural cracks. meanwhile, for normal concrete beam specimens, the dominant flexural and shear cracks occur in normal concrete-1 and normal concrete-2, respectively. the flexural crack width shown in geopolymer concrete is larger than in normal concrete. this is due to the lower tensile strength and modulus of elasticity of geopolymer concrete compared to normal concrete. furthermore, the cracks are wider and more evenly distributed in the pure flexural region (between the two loading points) for the geopolymer concrete beams. this phenomenon can be seen in geopolymer concrete-1 beam (fig. 8). thus, geopolymer concrete beams provide greater deformation opportunities before failure. in the casting process, the difference between the geopolymer and normal concrete is the duration of the setting time. geopolymer concrete has a setting time of about 30-60 minutes, while for normal concrete it is between 1-2 hours. the casting and molding of fresh geopolymer concrete are carried out very quickly and require more energy. furthermore, the workability of geopolymer concrete is lower than normal concrete. the viscosity of geopolymer concrete is higher than that of normal concrete, and it is more difficult to compact or pound geopolymer concrete than normal concrete. the compaction process in this study uses a rubber hammer and a vibrator. although the workability of geopolymer concrete is lower, the specimen results have only a few pores which are the same as the case of normal concrete (fig. 13-14). this is due to the compaction being carried out properly, despite its high energy requirements. fig. 9 flexural crack of geopolymer concrete-1 fig. 10 flexural crack of geopolymer concrete-2 fig. 11 flexural crack of normal concrete-1 fig. 12 shear crack of normal concrete-2 300 advances in technology innovation, vol. 7, no. 4, 2022, pp. 295-302 fig. 13 visible pore holes in the geopolymer concrete beam fig. 14 visible pore holes in the normal concrete beam 4. conclusions the mechanical characteristics testing of the laricfa-based geopolymer concrete is carried out for determining the compressive strength, split tensile strength, and ultimate moment. according to the results, the following conclusions can be obtained: (1) the use of laricfa has great practical advantages with its characteristics of low al2o3 (6.58%), high fe2o3 (48.51%), and cao (18.24%) contents. one of the advantages is that the geopolymer concrete with laricfa can reach an average compressive strength of 38.2 mpa only through treatment at room temperature. (2) the ratio between the split tensile and compressive strengths of geopolymer concrete is almost the same as that of normal concrete. (3) furthermore, the average ultimate moment of geopolymer concrete reaches 22.9 kn· m, which is relatively better than that of normal concrete. this indicates better bonding between geopolymer concrete and plain rebar than with normal concrete. (4) geopolymer concrete can be recommended for use as a structural component in simple house construction because it has mechanical characteristics that are almost the same as normal concrete. acknowledgments this study was partly supported by the directorate for research and community service, directorate general of research and development strengthening, ministry of research, technology and higher education of indonesia no. 100.add/ll3/pg/2020, centre for research and community development, and the pelita harapan university through grant p-031-fast/i/2019. conflicts of interest the authors declare no conflicts of interest. references [1] m. d. jackson, et al., “phillipsite and al-tobermorite mineral cements produced through low-temperature water-rock reactions in roman marine concrete,” american mineralogist, vol. 102, no. 7, pp. 1435-1450, july 2017. [2] d. reddy, et al., “durability of fly ash-based geopolymer structural concrete in the marine environment,” journal of material in civil engineering, vol. 25, no. 6, pp. 781-787, 2012. [3] p. chindaprasirt, et al., “effect of sodium hydroxide concentration on chloride penetration and steel corrosion of fly ash-based geopolymer concrete under marine site,” construction and building materials, vol. 63, pp. 303-310, july 2014. [4] m. s. darmawan, et al., “shear strength of geopolymer concrete beams using high calcium content fly ash in a marine environment,” buildings, vol. 9, no. 4, article no. 98, april 2019. [5] z. g. ralli, et al., “state of the art on geopolymer concrete,” international journal of structural integrity, vol. 12, no. 4, pp. 511-533, 2020. 301 advances in technology innovation, vol. 7, no. 4, 2022, pp. 295-302 [6] p. duxson, et al., “geopolymer technology: the current state of the art,” journal of materials science, vol. 42, no. 9, pp. 2917-2933, 2007. [7] m. łach, et al., “development and characterization of thermal insulation geopolymer foams based on fly ash,” proceedings of engineering and technology innovation, vol. 16, pp. 23-29, august 2020. [8] n. zhang, et al., “on the incorporation of class f fly-ash to enhance the geopolymerization effects and splitting tensile strength of the gold mine tailings-based geopolymer,” construction and building materials, vol. 308, article no. 125112, november 2021. [9] a. fernández-jiménez, et al., “new cementitious materials based on alkali-activated fly ash: performance at high temperatures,” journal of the american ceramic society, vol. 91, no. 10, pp. 3308-3314, october 2008. [10] m. sambucci, et al., “recent advances in geopolymer technology. a potential eco-friendly solution in the construction materials industry: a review,” journal of composites science, vol. 5, no. 4, article no. 109, april 2021. [11] w. kurdowski, cement and concrete chemistry, netherlands: springer, 2014. [12] j. temuujin, et al., “influence of calcium compounds on the mechanical properties of fly ash geopolymer pastes,” journal of hazardous materials, vol. 167, no. 1-3, pp. 82-88, august 2009. [13] t. tho-in, et al., “pervious high-calcium fly ash geopolymer concrete,” construction and building materials, vol. 30, pp. 366-371, may 2012. [14] t. nongnuang, et al., “characteristics of waste iron powder as a fine filler in a high-calcium fly ash geopolymer,” materials, vol. 14, no. 10, article no. 2515, may 2021. [15] p. venyite, et al., “effect of combined metakaolin and basalt powder additions to laterite-based geopolymers activated by rice husk ash (rha)/naoh solution,” silicon, vol. 14, no. 4, pp. 1643-1662, 2021. [16] j. davidovits, et al., “ferro-sialate geopolymers (-fe-o-si-o-al-o-),” https://www.geopolymer.org/news/27-ferro-sialate-geopolymers/, 2020. [17] k. c. gomes, et al., “iron distribution in geopolymer with ferromagnetic rich precursor,” materials science forum, vol. 643, pp. 131-138, march 2010. [18] n. essaidi, et al., “the role of hematite in aluminosilicate gels based on metakaolin,” ceramics silikati, vol. 58, no. 1, pp. 1-11, july 2014. [19] j. c. petermann, et al., “alkali-activated geopolymers: a literature review,” technical report afrl-rx-ty-tr-2010-0097, air force research laboratory, july 20, 2010. [20] p. chindaprasirt, et al., “effect of calcium-rich compounds on setting time and strength development of alkali-activated fly ash cured at ambient temperature,” case studies in construction materials, vol. 9, article no. e00198, december 2018. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). 302 microsoft word 6-v8n1(2023)-aiti#10034(73-80).docx advances in technology innovation, vol. 8, no. 1, 2023, pp. 73-80 calculation of temperature-dependent thermal expansion coefficient of metal crystals based on anharmonic correlated debye model tong sy tien*, nguyen thi minh thuy, vu thi kim lien, nguyen thi ngoc anh, do ngọc bich, le quang thanh department of basic sciences, university of fire prevention and fighting, hanoi, vietnam received 09 may 2022; received in revised form 07 august 2022; accepted 17 august 2022 doi: https://doi.org/10.46604/aiti.2023.10034 abstract this study aims to calculate the anharmonic thermal expansion (te) coefficient of metal crystals in the temperature dependence. the calculation model is derived from the anharmonic correlated debye (acd) model that is developed using the many-body perturbation approach and correlated debye model based on the anharmonic effective potential. this potential has taken into account the influence on the absorbing and backscattering atoms of all their nearest neighbors in the crystal lattice. the numerical results for the crystalline zinc (zn) and crystalline copper (cu) are in agreement with those obtained by the other theoretical model and experiments at several temperatures. the analytical results show that the acd model is useful and efficient in analyzing the te of coefficient of metal crystals. keywords: thermal expansion coefficient, metal crystals, anharmonic correlated debye model 1. introduction in recent years, the anharmonic thermal expansion (te) coefficient has been widely used to determine many dynamic properties of materials [1]. like the compressibility and heat capacity, the te coefficient is important because it is one of the independent thermodynamic properties which can be measured experimentally with high precision [2]. accurate information on the te coefficient in the temperature dependence is required for the metallurgical industry as well as in engineering physics [3]. the te coefficient is often used to determine the matching between different metal components in the alloy [4]. a large difference in the te coefficients between the metal components can lead to adverse deformation in the alloy when the temperature changes [5]. moreover, the position of atoms and their interatomic distance always changes and are not stationary under the influence of thermal vibrations [6]. this influence causes thermal disorder and anharmonic effects in the crystal lattice, so the te coefficient is sensitive to the temperature change and can be varied with increasing temperature [7]. in materials science, high precision measurements of lattice parameters using the x-ray method based on the development of the theory of the crystal structures have become increasingly important. moreover, the accurate determination of lattice parameters enables one to investigate the te coefficient of various crystalline substances, even if the weight of the substances available for measurement is only a few milligrams [2, 4]. besides, the extended x-ray absorption fine structure (exafs) technique is a widely employed probe of the dynamical behaviors and structural parameters in disordered systems [8]. it is because exafs spectroscopy can contain information on local structures around x-ray absorbing atoms and gives interatomic distances and coordination numbers in crystal lattices [9]. * corresponding author. e-mail address: tongsytien@yahoo.com tel.: +849-1-2439564; fax: +849-8-1439564 advances in technology innovation, vol. 8, no. 1, 2023, pp. 73-80 74 this resulted in the exafs technique being developed and expanded greatly based on the rapid development of synchrotron radiation facilities worldwide [10]. in reality, thermal disorders can disturb the exafs oscillation [6, 9], so the te coefficient is also sensitive to exafs oscillation under the influence of temperature. nowadays, metals and advances in manufacturing processes have brought us the industrial revolutions, so it becomes irreplaceable materials in the growth of human civilization. they are used extensively in manufacturing machines for industries, agriculture or farming, and automobiles, including road vehicles, railways, airplanes, rockets, etc. [2-4]. the te of coefficient of metal crystals has also been investigated using the anharmonic correlated einstein (ace) model [11-12] and experiments [11-13]. still, this model only uses a unique correlated einstein frequency to describe the atomic vibrations, so it cannot mimic the acoustic phonon branches presenting in lattice crystals. recently, an anharmonic correlated debye (acd) model [14] can treat even the acoustic phonon branches presenting in lattice crystals because it describes the atomic vibrations using the frequencies varying from 0 to the correlated debye frequency. it has also been efficiently used to investigate the anharmonic thermodynamic properties of many materials [9, 15-16]. still, it has not yet been used to analyze the anharmonic te coefficient of metal crystals. therefore, the calculation and analysis of the temperature-dependent te coefficient of metal crystals based on extending the acd model will be a necessary addition to experimental data analysis in the advanced material technique. 2. calculation of the anharmonic te coefficient usually, the te coefficient �� ��� characterizes the net thermal expansion (nte) caused by thermal vibrations in the crystal lattice [17], as shown in fig. 1. fig. 1 thermal expansion of metal with a change in temperature the te coefficient can be determined [11-12] by: ( ) ( ) ( ) ( ) 1 0 0 0 t dr t d t t t r dt r dt σ α ∆ = = ∆ ℓ ≃ ℓ (1) where � is the absolute temperature, ���� is the instantaneous distance between the backscattering and absorbing atoms, �� is the equilibrium distance between the backscattering and absorbing atoms, and � � ���� is the first exafs cumulant and can be expressed in terms of the power moments of the radial pair distribution (rd) function [18-19]: ( ) ( ) ( ) 1 0t r t rσ = − (2) normally, the morse potential can validly determine the pair interaction (pi) potential of the crystals [20-21]. if this potential is expanded up to the third-order around its minimum position, it can be written as ( ) ( )2 2 2 3 3 0( ) e 2 , x x x d e d d x d x x r t r α α ϕ α α − − = − ≅ − + − = − (3) where � is the dissociation energy, � is the width of the potential, and x is the deviation distance between the backscattering and absorbing atoms. to determine the thermodynamic parameters of a system, it is necessary to specify its anharmonic effective (ae) potential and force constants [8, 22]. one considers a monatomic system with an ae potential (ignoring the constant contribution) is extended up to the third-order: advances in technology innovation, vol. 8, no. 1, 2023, pp. 73-80 75 2 3 3 1 ( ) 2 eff eff v x k x k x= − (4) where � �� is the effective force constant, �� is the local force constant giving asymmetry of potential due to the inclusion of anharmonicity, and these force constants are considered in the temperature independence. in the relative vibrations of absorbing eq. (1) and backscattering eq. (2) atoms, including the effect of correlation and taking into account only the nearest-neighbor interactions, the ae potential [23-24] is given using the pi potential, i.e. � �( )12 ij 1,2 1,2 ( ) , eff i i i j i v v x v xr r m µ ε ε = ≠ = + =  (5) where �� ( 1i = , 2) is the mass of the ith atom and equals m in the monatomic crystals, respectively, � = � ��/�� + ��� is the reduced mass of the absorber and backscatter, ��� is a unit vector, the sum i is the over absorbers ( 1i = ) and backscatters ( 2i = ), the sum j is over the nearest neighbors, and. the acd model is derived from the dualism of an elementary particle in quantum theory and is perfected based on the correlated debye model using the ae potential and many-body perturbation approach [9, 14]. it is often used effectively in treating monatomic systems with less complex phonon density of states and multiple acoustic phonons [15-16]. in this model, the atomic vibrations can be quantized and treated as a system consisting of many phonons, in which each atomic vibration corresponds to a wave that has a frequency ���� and is described via the dispersion relation ( ) 0sin , 2 , 2 d d kqa q q m a π ω ω ω   = = ≤    (6) where � is the phonon wavenumber in the first brillouin (fb) zone, a is the lattice constant, and �� is the correlated debye frequency and characterizes the atomic thermal vibrations. the general expression of the first exafs cumulant in the acd model was calculated in the temperature dependence by hung et al. [14]. still, these obtained expressions are not optimized yet because they still depend on the lattice constant a. in this investigation, the previous acd model has been extended to calculate the temperature-dependent te coefficient of metal crystals. after using the general expression of the first exafs cumulant and converting from variable � to variable � in the formula p=��/2, the temperature-dependent first cumulant of metal crystals are obtained as ( ) ( ) ( ) ( ) ( ) ( ) ( ) / 2 / 2 1 3 3 2 2 0 0 1 exp3 3 coth 2 1 expeff eff b b b p k t p k t p k k t p dp p dp k k kt π π σ ω ω ω ω ωπ π + −   − −  =  =       ℏ ℏℏ ℏ ℏ (7) substituting this cumulant into eq. (1) to calculate the temperature-dependent te coefficient of metal crystals, it yields ( ) ( ) ( ) / 2 3 2 0 0 3 cot 2h eff t b t p k k d p dp r t k dt π ω π α ω          =    ℏ ℏ (8) using an approximation exp"−ℏ����/�% �& ≈ 0, the te coefficient of metal crystals in the low-temperature (lt) limit �� → 0� can be calculated from eq. (8), i.e. ( ) 2 3 2 0 2 b d eff t k k t r k t π ω α ℏ ≃ (9) using an approximation exp"−ℏ����/�% �& ≈ 1 − ℏ����/�% � , the te coefficient of metal crystals in the high-temperature (ht) limit �t → ∞� can be calculated from eq. (8), that is advances in technology innovation, vol. 8, no. 1, 2023, pp. 73-80 76 ( ) 3 2 0 3 b t eff k r k t k α ≃ (10) thus, an extended acd model has been perfected to efficiently calculate the temperature dependence of the te coefficient of metal crystals. the obtained expressions using this model can satisfy all their fundamental properties in the temperature dependence. these expressions have also been optimized to not depend on the lattice constant a as in the previous acd model [14-15]. 3. numerical result and discussion in this section, the obtained expressions using the acd model in sec. 2 are applied to the numerical calculations of crystalline zinc (zn) and crystalline copper (cu) that have the hexagonal close-packed (hcp) and face-centered cubic (fcc) structures, respectively. herein, the local force constants � �� and �� , the correlated einstein frequency �� , and the temperature-dependent first exafs cumulant � � ���� and te coefficient �� ��� are the quantities to be performed in numerical calculations. these obtained numerical results are compared with those obtained using the ace [11-12, 22-23, 25] and experiments [11-13, 22, 26-28]. from these obtained comparisons, the development and effectiveness of the acd model are analyzed and discussed in investigating the temperature-dependent thermal expansion coefficient of metal crystals. the following is the presentation of these numerical results: the ae potential of zn and cu can be calculated using eq. (5) based on the morse potential in eq. (3) and the crystal structure properties of these metals. after ignoring the overall constant in the obtained result and comparing it with eq. (4), the local force constants � �� and �� are deduced in the expressions of the morse potential parameters d and �, while the correlated debye frequency �� is calculated using eq. (6) via the effective force constant � �� . the values of the local force constants � �� and �� and the correlated debye frequency �� of zn and cu are given in table 1. herein, the obtained results of the correlated debye frequency �� using the ace [22, 25] are inferred from the correlated einstein frequency � and related expression �� = √2� [9, 23-24, 28]. meanwhile, the obtained results using the experimental exafs data are derived from the measured morse potential parameters and eq. (6). it can be seen that the obtained results using the acd model fit with those obtained using the ace [22-23, 25] and experimental data [22, 27], especially for the correlated debye frequency �� and effective force constant � �� . table 1 the thermodynamic parameters ��, ��, and �� of zn and cu obtained using the acd and ace models and experiments metal method mass d (ev) α (å-1) r0 (å) � �� (evå -2) �� (evå -3) �� (×10 13hz) zn acd model 65.377b 0.1698c 1.7054c 2.7931c 2.3887a 1.0528a 3.7552a ace model 65.377b 0.1698c 1.7054c 2.7931c 2.4692d 1.0528d 3.8066d experiment 0.1685d 1.7000d 2.7650d 2.4348d 1.0348d 3.7801d cu acd model 63,546b 0.3429c 1.3588c 2.8860c 3.1655a 1.0753a 4.3848a ace model 63,546b 0.3429c 1.3588c 2.8860c 3.1655e 1.0753e 4.3848e experiment 0.3300f 1.3800f 3.2000f 1.3000f 4.4086f athis work, breference [29], creference [21], dreference [22], ereference [23, 25], freference [27] the dispersion relation ���� of zn and cu in the fb zone is calculated by eq. (6) and is represented in fig. 1. it can be shown that the obtained results using the acd are a symmetric function of a linear chain of q, and its maximum value is �� at the bounds of the fb zone with � = ±0/2, as seen in fig. 2. these characteristics of the dispersion relation ���� are completely consistent with similar obtained results of other crystals in previous works, such as the crystalline iron (fe) [15], crystalline molybdenum (mo) [15], and crystalline wolfram (w) [15], and crystalline germanium (ge) [16]. advances in technology innovation, vol. 8, no. 1, 2023, pp. 73-80 77 the temperature dependence of the first exafs cumulant � � ���� of (a) zn and (b) cu in a range from 0 to 700 k is represented in fig. 3, in which the obtained results using the acd model are calculated by eq. (7). herein, the values of cu are smaller than those of zn, which is because the local force constants �� of these metals are roughly equivalent, but cu has the effective force constant � �� larger than those of zn, and this cumulant decreases rapidly as the effective force constant increases, as seen in table 1. it can be seen that the obtained results using the acd model are in good agreement with those obtained using the ace [11-12] model and experiments [11-12, 26, 28]. also, in comparison with the experimental values, the obtained results using acd are better in agreement with those obtained using the ace model, especially in the lt region. for example, the obtained results of zn using the acd model, ace model, and an experiment at � ≃ 77 k are � � � ≃ 5.29×102�å, �( ) ≃5.34×102�å [11], and �( ) ≃ 5.26×102�å [28], respectively. meanwhile, the obtained results of cu using the acd model, ace model, and an experiment at � ≃ 80 k are �( ) ≃ 3.26×102�å, �( ) ≃ 3.34×102�å [12], and �( ) ≃ 3.00×102�å [26], respectively. moreover, both the acd and ace [11-12] models show the contribution of quantum effects in the lt region, but the obtained results using the ace model [11-12] are slightly greater than those obtained using the acd model. the minor difference lies in that the ace model [11-12] uses only one effective frequency to describe the atomic thermal vibrations, as depicted in fig. 3. (a) the obtained result of zn (b) the obtained result of cu fig. 2 the dispersion relation of metals obtained from the acd model (a) the obtained results of zn (b) the obtained results of cu fig. 3 the temperature-dependent first exafs cumulant of metals obtained using the acd and ace models and experiments the temperature dependence of the te coefficient ��(�) of (a) zn and (b) cu in a range from 0 to 700 k is represented in fig. 4, in which the obtained result using the acd model is calculated by eq. (8). herein, the values of zn are bigger than those of cu because the temperature-dependent first exafs cumulant of cu changes more slowly than those of zn, as seen in fig. 3. it can be seen that the obtained results using the acd model agree with those obtained using the ace [11-12] model and advances in technology innovation, vol. 8, no. 1, 2023, pp. 73-80 78 experiments [11-13]. for example, the obtained results of zn using the acd model, ace model, and an experiment at � ≃ 300 k are �� ≃ 1.57×10 23k2 , �� ≃ 1.56×10 23k2 [11], and �� ≃ 1.58×10 23k2 [11], respectively. meanwhile, the obtained results of cu using the acd model, ace model, and an experiment at � ≃ 100 k are �� ≃ 0.76×10 23k2 , �� ≃ 0.71×1023k2 [12], and �� ≃ 0.80×10 23k2 [13], respectively. the quantum effects in the acd model [14] can be shown by the obtained results in the lt limit low temperatures. as shown in the temperature range from above 0 k and below about 10 k, the obtained results of the first exafs cumulant are a region of an almost unchanged value, while the obtained results of the te coefficient correspond to a region of values greater than zero. moreover, the obtained results using the acd model are not destroyed quickly in the lt limit as those obtained using the ace model [11-12], which fits perfectly with eq. (9) and shows that the acd model can fully describe the quantum effects. also, the obtained results using the acd model increase with the rise of temperature t and approach the constants in the ht limit, which fits perfectly with eq. (10) and shows that the acd model can efficiently describe the anharmonic effects, as seen in fig. 4. these results are consistent with those obtained by using the quantum methods [11-12, 23]. (a) the obtained results of zn (b) the obtained results of cu fig. 4 the temperature-dependent te coefficient of metals obtained using the acd and ace models and experiments 4. conclusions in this investigation, the expansion and development of an efficient model have been performed to calculate and analyze the temperature-dependent te coefficient of metal crystals. the calculated results of the anharmonic te coefficient using the acd model satisfied all of their fundamental properties. the te coefficient increases with increasing temperature t, which means that the crystal lattice expands strongly at higher temperatures. these results can also describe the influence of anharmonic effects at high temperatures and the influence of the quantum effects at low temperatures on the te coefficient. the good agreement of the obtained numerical results for zn and cu with those obtained using the ace model and experiments at various temperatures shows the effectiveness of the present model in investigating the temperature-dependent te coefficient of metal crystals. this model can be applied to calculate and analyze the anharmonic te coefficient of other crystals at a temperature ranging from absolute zero degree to their melting points. acknowledgments this work is supported by the university of fire prevention and fighting, 243 khuat duy tien, thanh xuan, hanoi, vietnam. conflicts of interest the authors declare no conflict of interest. advances in technology innovation, vol. 8, no. 1, 2023, pp. 73-80 79 references [1] h. liu, w. sun, z. zhang, l. lovings, and c. lind, “thermal expansion behavior in the a2m3o12 family of materials,” solids, vol. 2, no. 1, pp. 87-107, february 2021. [2] j. w. hwang, “thermal expansion of nickel and iron, and the influence of nitrogen on the lattice parameter of iron at the curie temperature,” masters theses, department of metallurgical engineering, university of missouri–rolla, rolla, missouri, 1972. [3] v. a. drebushchak, “thermal expansion of solids: review on theories,” journal of thermal analysis and calorimetry, vol. 142, pp. 1097-1113, january 2020. [4] t. dengg, v. razumovskiy, l. romaner, g. kresse, p. puschnig, and j. spitaler, “thermal expansion coefficient of wre alloys from first principles,” physical review b, vol. 96, no. 3, pp. 035148, july 2017. [5] g. laplanche, p. gadaud, o. horst, f. otto, g. eggeler, and e. p. george, “temperature dependencies of the elastic moduli and thermal expansion coefficient of an equiatomic, single-phase cocrfemnni high-entropy alloy,” journal of alloys and compounds, vol. 623, pp. 348-353, february 2015. [6] p. a. lee, p. h. citrin, p. eisenberger, and b. m. kincaid, “extended x-ray absorption fine structure its strengths and limitations as a structural tool,” reviews of modern physics, vol. 53, no. 4, pp. 769-806, october 1981. [7] z. k. liu, y. wang, and s. shang, “thermal expansion anomaly regulated by entropy,” scientific reports, vol. 4, pp. 7043, november 2014. [8] z. k. liu, y. wang, and s. shang, “temperature-dependent exafs debye-waller factor of distorted hcp crystals,” journal of the physical society of japan, vo. 91, no. 5, pp.054703, april 2022. [9] t. yokoyama, k. kobayashi, t. ohta, and a. ugawa, “anharmonic interatomic potentials of diatomic and linear triatomic molecules studied by extended x-ray absorption fine structure,” physical review b, vol. 53, no. 10, pp. 6111-6122, march 1996. [10] t. s. tien, “effect of the non-ideal axial ratio c/a on anharmonic exafs oscillation of h.c.p. crystals,” journal of synchrotron radiation, vol. 28, pp. 1544-1557, september 2021. [11] n. v. hung, c. s. thang, n. b. duc, d. q. vuong, and t. s. tien, “temperature dependence of theoretical and experimental debye-waller factors, thermal expansion and xafs of metallic zinc,” physica b: condensed matter, vol. 521, pp. 198-203, september 2017. [12] n. v. hung, c. s. thang, n. b. duc, d. q. vuong, and t. s. tien, “advances in theoretical and experimental xafs studies of thermodynamic properties, anharmonic effects and structural determination of fcc crystals,” the european physical journal b, vol. 90, no.12, pp. 256, december 2017. [13] y. s. toukian, r. k. kirby, r. e. taylor, and p. d. desai, thermophysical properties of matter–thermal expansion: metallic elements and alloys, new york: ifi/plenum, 1975. [14] n. v. hung, n. b. trung, and b. kirchner, “anharmonic correlated debye model debye-waller factors,” physica b: condensed matter, vol. 405, no. 11, pp. 2519-2525, june 2010. [15] n. v. hung, t. t. hue, h. d. khoa, and d. q. vuong, “anharmonic correlated debye model high-order expanded interatomic effective potential and debye-waller factors of bcc crystals,” physica b: condensed matter, vol. 503, pp. 174-178, december 2016. [16] t. s. tien, “analysis of exafs oscillation of monocrystalline diamond-semiconductors using anharmonic correlated debye model,” european physical journal plus, vol. 136, no.5, pp. 539, may 2021. [17] c. y. ho and r. e. taylor, thermal expansion of solids, ohio: asm international, 1998. [18] j. m. tranquada, and r. ingalls, “extended x-ray absorption fine-structure study of anharmonicity in cubr,” physical review b, vol. 28, no. 6, pp. 3520-3528, september 1983. [19] l. tröger, t. yokoyama, d. arvanitis, t. lederer, m. tischer, and k. baberschke, “determination of bond lengths, atomic mean-square relative displacements, and local thermal expansion by means of soft-x-ray photoabsorption,” physical review b, vol. 49, no. 2, pp. 888-903, january 1994. [20] p. m. morse, “diatomic molecules according to the wave mechanics. ii. vibrational levels,” physical review, vol. 34, pp. 57-64, july 1929. [21] l. a. girifalco and v. g. weizer, “application of the morse potential function to cubic metals,” physical review, vol. 114, no. 3, pp. 687-690, may 1959. [22] n. v. hung, l. h. hung, t. s. tien, and r. r. frahm, “anharmonic effective potential, effective local force constant and exafs of hcp crystals: theory and comparison to experiment,” international journal of modern physics b, vol. 22, no. 29, pp. 5155-5166, november 2008. advances in technology innovation, vol. 8, no. 1, 2023, pp. 73-80 80 [23] n. v. hung and j. j. rehr, “anharmonic correlated einstein-model debye-waller factors,” physical review b, vol. 56, no. 1, pp. 43-46, july 1997. [24] t. s. tien, “advances in studies of the temperature dependence of the exafs amplitude and phase of fcc crystals,” journal of physics d: applied physics, vol. 53, no. 31, pp. 315303, june 2020. [25] n. v. hung and p. fornasini, “anharmonic effective potential, correlation effects, and exafs cumulants calculated from a morse interaction potential for fcc metals,” journal of the physical society of japan, vol. 76, no. 8, article no. 084601, august 2007. [26] t. yokoyama, t. satsukawa, and t. ohta, “anharmonic interatomic potentials of metals and metal bromides determined by exafs,” japanese journal of applied physics, vol. 28, no. 10r, pp. 1905-1908, october 1989. [27] i. v. pirog, t. i. nedoseikina, i. a. zarubin, and a. t. shuvaev, “anharmonic pair potential study in face-centered-cubic structure metals,” journal of physics: condensed matter, vol. 14, no. 8, pp. 1825-1832, february 2002. [28] n. v. hung, t. s. tien, n. b. duc, and d. q. vuong, “high-order expanded xafs debye-waller factors of hcp crystals based on classical anharmonic correlated einstein model,” modern physics letters b, vol. 28, no. 21, pp. 1450174, august 2014. [29] j. meija, t. b. coplen, m. berglund, w. a. brand, p. d. bièvre, m. gröning, et al, “atomic weights of the elements 2013 (iupac technical report),” pure and applied chemistry, vol. 88, no. 3, pp. 265-291, february 2016. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation , vol. 3, no. 3, 2018, pp. 101 108 design and analysis of a water channel for characterization of low reynolds number flows judah d. rutledge, jesse j. french * department of mechanical engineering, letourneau university, longview, texas , usa. received 21 july 2017; received in revised form 05 sept ember 2017; accept ed 13 december 2017 abstact a water channel for performing flow v isualization and stud ying scale models in flu id mechanics was designed , analyzed, and fabricated with comme rcia lly availab le co mponents . the materia l cost of the channel is 10% of the leading educational units , and the fabrication processes required for channel construction are basic and typical of local craftsmen in developing countries. both structural analysis and flo w rate ca lculations we re performed to verify the functionality of the tunnel. hand calculations and finite ele ment analysis were used to model stress and deflection in the channel floor under hydrostatic loads. these were used to select a polycarbonate panel thickness that will withstand the hydrostatic and hydrodynamic loads on the channel floor and walls for a projected useful lifespan of 40 years. the water channel has a test section area that is 30 c m by 42 c m and up to 1 m long. the system pump is capable of generating incident flows of up to 7.1 c m/sec in the test section. the channel is also designed to be upgraded with a tow ca rriage, a llowing for flow v isualization as well as fully subme rged and partia lly submerged models with reynolds or froudes number dependent studies. keywor ds: fluid mechanics, aerodynamics, water tunnel, laminar flow, scale model, education 1. introduction in order to perform scale model tests that can be correlated with the behavior of prototypes in a rea l world environ ment, engineers ma ke use of controlled environ ments to simulate or duplicate the conditions e xperienced by a co mponent in operation. since in-situ testing can often be prohibitive in terms of both cost and time, significant time and attention are given to the scale model testing of prototype. additionally, scale models can be used to test and observe phenomena that are unmanageable or that happen at rates humans cannot typically observe in real time, be it a phenomenon that occurs in fractions of a second, such as vortices traveling down the length of a swept wing, or that takes months to develop , as in erosion case studies [1]. finally, scale models are invaluable for gaining deeper understanding of the physical phenomenon being explored. this application, in particular, is invaluable for higher education institutions. for products finding application in the realm of fluid mec hanics, wind and water tunnels are the primary means for achieving the repeatability and control required for such tests. water tunnels, specifically, are used when testing phenomena related to ma rine or aquatic applications ; they are also used in aerodynamics when detailed flo w visualizat ion is necessary. there are several co mpanies that have designed excellent water tunnels for research and education [2]. however, these water tunnels can be expensive and difficult to acquire in developing nations . because of this, a water tunnel was designed using comme rcia lly availab le co mponents with the end goal of manufacturing an adequate but affordable solution for studying fluid mechanics. such a tunnel design is affordable, can offer opportunities to smalle r instit utions in first world countries and to * corresponding author. e-mail address: jessefrench@letu.edu advances in technology innovation , vol. 3, no. 3, 2018, pp. 101 108 copyright © taeti 102 institutions in developing countries worldwide, and benefits the local economy in the country of construction. this paper describes the economic considerations, the construction of this device, and the structural and flow analysis for the design. 2. manufacturing and cost a ca d mode l h ighlighting standard water channel co mponents of the channel assembly is shown in fig . 1 [3]. the channel structure is constructed in two separate components, both of which are welded out of a36 steel and powder-coated for corrosion resistance. the first component is the channel fra me . the fra me is constructed out of angle iron in the corners and reinforced with square tubing every 40 c m along the length of the channel to support the sidewa lls and to mitigate deformation of the tunnel. these supports also serve to attach fixtures to the cha nnel for model specific tests, for mounting photography equipment, or for attaching rails for a towing carriage or for wave generation. the second compone nt is the work bench. the work bench provides a stable work p lace for tests and supports the channel tank, which we ighs 370 kilogra ms when filled to capacity. the bench has four adjustable feet which a llo w for accurate levelling of the channel assembly. th e channel itself is 2.4 meters long with a test section that is 100 c m long and a 30 cm x 40 c m. cross -section. the rectangular channel design simplifies construction, though some flow conditioning is lost without a contraction. fig. 1 cad model of the water channel assembly only mate ria ls and manufacturing methods commonly available in developing nations were used for constructing the channel. the work bench and fra me were constructed with a36 rectangular steel tubing and powder coated to mitigate corrosion due to contact with water, though the appropriate primer and paint coat would a lso be adequate. the channel walls and floor were constructed from polycarbonate panels. the ben ch and frame we re we lded using gas metal arc weld ing, though the thickness of steel used can also be stick-we lded or oxy-fuel welded if other forms of arc weld ing are not available. in order to resist distortion and leaking, the enclosure was bonded with scigrip 16, and sealed along the inner seams with a masterseal np1, a polyurethane sealant rated for continuous water immersion. the channel was plumbed with standard pvc pipes and fittings, and a centrifugal pu mp was sourced to power the flo w loop . the total construction time of the channel fro m was 2.5 months. fig. 2 shows the finished channel with the flow conditioners in place. a cost analysis shows that the channel is highly affordable when co mpared to the typical cost of an educational water tunnel, which is in the $20,000 range. because labor for the construction was a portion of this project, only the material purchases are used to determine the baseline cost of producing a channel unit. the breakdown of materia l costs is shown in table 1. the total cost of producing the water channel is 1609.09 usd, which is roughly 10% of the cost for purchasing a commercially produced tunnel for education and research [4]. advances in technology innovation , vol. 3, no. 3, 2018, pp. 101 108 copyright © taeti 103 fig. 2 completed water channel assembly table 1 material purchase costs for the water channel c ompon e n t un i t c ost q ty c ompon e nt c ost .375x48x96" lexan sheet 567.00 1 417.00 .375x24x48 sheet 178.21 1 137.24 mast erseal np 1 t ube 5.36 3 16.08 scigrip 16 can 12.51 1 12.51 t eflon t ape roll 1.48 1 1.48 silicone gasket unit 2.4 2 4.80 2 in x 10 ft p vc pipe unit 8.37 2 16.74 2 in socket female x np t male unit 1.17 2 2.34 2 in socket femal x npt female unit 1.2 2 2.40 2 in 90 deg elbow unit 0.98 6 5.88 p vc p rimer & cement p ack 8.81 1 8.81 2 in t hreaded adapt or unit 1.32 2 2.64 p vc p rimer & cement p ack 8.81 1 8.81 garboard drain p lug unit 10.17 1 10.17 rubber gasket unit 12.91 2 25.82 p ent air 011515 whisper flow unit 664.75 1 664.75 wiring cable feet 2.32 15 34.80 wiring p lug unit 19.97 1 19.97 double p ole t oggle swit ch unit 5.98 1 5.98 swit ch junct ion box unit 5.95 1 5.95 8x32 st ainless st eel screws bag/24 6.48 1 6.48 3/8 inch bolt unit 1.61 4 6.44 1"x2"x14 gauge st eel t ubing 24' st ick 31.00 3 93.00 .75"x.75"x11 gauge st eel angle 24' st ick 18.00 2 36.00 1"x1"x14 gauge st eel t ubing 20' st ick 21.00 3 63.00 total c ost 1609.09 3. structural analysis both hand calculations and finite ele ment analysis (fea) we re used for the structural analysis of the channel. aspects of the channel that were analyzed for fa ilu re include crit ical co mponents such as the table legs, the channel ribs, and the enclosure. material selection was performed based prima rily on availability and cos t, and mathemat ical ana lyses were performed to determine if the selection was adequate. in order to guarantee that the tank geometry was not comp ro mised, stress and distortion calculations were performed on the frame, the channel walls, the channel floor, and the work bench. advances in technology innovation , vol. 3, no. 3, 2018, pp. 101 108 copyright © taeti 104 fig. 3 plate boundary conditions for floor and ends of channel fig. 4 hydrostatic pressure on water channel panel in most of the channel design, the limit ing factor was not structural integrity, but was instead deflection. even if the ma ximu m stresses in a test fixture are re latively s mall, deflections in the structure can be sufficiently large to affect the geometry of the test section and introduce error in the test data. a secondary concern with deflection is the visual detect ion by the operator. visible deflection can detract attention fro m the test and reflect poorly on the quality of construction of the testing device. because of the magnitude of the deflections caused by the hydrostatic loads in the channel, visual deflect i on was the most pro minent concern for all o f the ite ms evaluated. the ma ximu m deflection a llo wed for any channel co mponent was 0.1 inches (2.5 mm), which is detectable with measure ment devices but is not a misalign ment typically visib le upon simple observation. the stress and deflection calculations for the floor channel are presented here as a case study of the methods used. in order to choose the thickness of polycarbonate used for the channel floor and size, the floor panel between two support beams was modeled as a rectangular p late for ma ximu m stress and ma ximu m deflection ca lculations. both hand calculations were performed using roark ’s formulas for stress and strain [5], and various fea models we re created using com sol multiphysics 5.1. based on the frame geo metry, the plate length and width for one section of the channel are 16 inches and 12 inches respectively. thicknesses of 1/4, 3/8, and 1/2 inch we re c onsidered for deflection and stress . since the channel walls were origina lly only going to be sealed, not bonded to the flo or, they were not considered fixed to the floor, resulting in these sides being simply supported. where the panel is supported by a frame rib is considered fixed since the loading opposite of the span is nearly symmetrica l. these boundary conditions, then, represent the most severe loading condition, that which is at the ends of the channel where three sides of the plate are simply supported, and only the side that is supported by a span of the tunnel is assumed to be cantilevered (fig. 3). for the hand calculations, table 11.4.3 was refe rred to in roark ’s formulas for stress and strain . this describes the ma ximu m stress and ma ximu m deflection as a function of plate geometry, the materia l properties, and an applied, constant pressure. because the ma ximu m deflection and stress in plates with straight boundaries are determined nu merica lly, no e xpression for deflect ion and stress as a function of position is derived. instead, the dimensionless ratio, 𝑎/𝑏 is used to characterize plates through a range of aspect ratios and derive empirical constants used in conjunction with the plate thickn ess and material properties to estimate maximu m stres s and deflection according to eq s. 4.1 and 4.2. 2 2max qb t    (1) 4 3 max qb y et   (2) here 𝑞 is a constant hydrostatic pressure equal to the depth of the water times its specific gravity (fig. 4), 𝐸 is young’s modulus, and 𝑏 and 𝑡 are the width and thickness of the plate, respectively. the final variab le s in these equations, 𝛽 and 𝛼, are functions of the aspect ratio, 𝑎/𝑏 , and are determined by interpolating values fro m table 11.4.3 in roark ’s. the provided tables are fo r a poisson’s ratio of 0.3. according to the handbook, the returned deflect ion will be accurate to with in 8%, and the maximu m calculated stress to within 15% [5]. advances in technology innovation , vol. 3, no. 3, 2018, pp. 101 108 copyright © taeti 105 fig. 5 fea results for stress distribution on for the channel floor (psi) the same boundary conditions and geometry we re used fo r the fea analysis using comsol. since roark’s analysis is based on a poisson’s ratio of 0.30, and the actual poisson’s ratio for polycarbonate is 0.37, fea mode ls we re run for both poisson’s ratios of 0.3 and 0.37 in order to co mpare the effects of poisson’s ratio on the stress and deflection and determine if roark’s can still be us ed as an accurate approximation (fig. 5). the hand calculations and the fea models showed a lower corre lation for the ma ximu m displace ment (table 2), but high correlation for the ma ximu m stress (table 3). the ma ximu m e rror between roark’s and comsol for a po isson’s ratio of 0.30 was 3% fo r stress and 18% for displace ment. modifying po isson’s ratio did not significantly affect the stress or displacement, changing the stress and displacement values by about 1% and 5% respectively. based on this, the tables in roark’s handbook for stress and strain of rectangular plates can be used for similar designs where an fea package is not available. the most important result fro m this analysis, however, is that the displacement of both the 3/8 inch and the 1/ 2 inch panel is within the allo wable limit of 2.5 mm. since the 3/8 panel fulfilled the strength and deflection requirements, and the cost difference between the 3/8 inch and the 1/2 inch panels was significant, the 3/8 inch panel was selected for the channel enclosure. table 2 maximum deflection for channel floor [mm] roark’s comsol (nu = 0.30) comsol (nu = 0.37) percent difference [%] cm vs. r (nu=0.30) nu=0.30 vs. nu=0.37 1 /4 inch 3.83 4.47 4.24 14.3 5.4 3 /8 inch 1.13 1.34 1.27 15.7 5.5 1 /2 inch 0.47 0.572 0.542 17.8 5.5 table 3 maximum stress for the channel floor [kpa] roark’s comsol (nu = 0.30) comsol (nu = 0.37) percent difference [%] cm vs. r (nu=0.30) nu=0.30 vs. nu=0.37 1 /4 inch 5615 5587 5559 0.5 0.5 3 /8 inch 2495 2424 2450 2.9 1.1 1 /2 inch 1413 1401 1414 0.9 0.9 because the prolonged loading nature of a hydrostatic pressure vessel ma kes such structures susceptible to creep when constructed fro m poly mers, a creep rupture stress analysis was also performed on the panel [ 6]. a literature revie w for the creep rupture properties of polycarbonate showed that available plot s only predicted the c reep rupture strength out to 45,000 hours, or approximately 5 years [7]. however, since the creep rupture stress curve for polycarbonate showed a highly linear trend on a logarith mic scale, an e xt rapolation out to one more order of magnitude was a lso performed. these estimat ions predicted that the panel has a very high factor of safety with respect to creep rupture over the channel’s design life. the creep rupture strength at five years is 49,000 kpa, and the estimated the rupture strength decreases to 43,000 kpa at 40 years of continuous loading. still, the panel stress at 40 years is only 6% of the estimated rupture strength for this loading period. this estimation produces a factor of safety of 16.7, ind icating that the channel floor and walls are not at risk of failing due to creep. advances in technology innovation , vol. 3, no. 3, 2018, pp. 101 108 copyright © taeti 106 4. flow rate analysis when determining the channel flow rate, two systems were considered: a pu mp providing constant flow rate and a surge tank wh ich could potentially provide higher flow rates for a short period of time . the flow rate ca lculat ions for both systems were perfo rmed based on the dynamic head source and pipe and fitting frict ion factors determined fro m the literature. because the pump and hardware used were specified in eng lish units, all head a nd flow rate ca lculations were performed in the same, and the final flow velocity in the channel test section was then converted into centimeters per second. 4.1. surge tank flowrate the surge tank flo w rate ca lculations were performed for a surge tank located nominally 10 feet (3 m) above the pipe entering the channel. this head was assumed to be constant, regardless of the level of water in the 200 liter dru m chosen as the surge tank. in order to calculate flow rate, an energy balance was performed bet ween the barrel outlet and the channel inlet using the dynamic head, elevation change, and frictional loss coefficients for each component in the system (eq. 3). here, 𝑃 /𝛾 is the dynamic head, 𝑙 and 𝐷 are the length and diameter of the pipe, and 𝑓 and 𝐾𝐿 are frictional loss coefficients (fig. 6). the loss coefficients for pipe fittings, ∑𝐾𝐿 , are reynolds number independent and obtained from tables in fundamentals of fluid mechanics by munson, et. al. [8]. the loss coeffic ient in the pipe, 𝑓, is reynolds number dependent and was determined using the method outlined by le wis. f. moody for pipe flo w friction factors [ 9]. as recommended by munson, et. al., the pipe was assumed to be smooth, therefore the ratio of the equivalent roughness pipe dia meter, 𝜖/𝐷 , was equal to zero. 22 2 2 1 1 2 2 1 2 2 2 2 2 l k vp v p v l v z z f g g d g g          (3) velocity at the entrance and exit of the pipe is constant, and the equation simplifies to: 22 1 2 2 l k vl v f d g g z    (4) from this, the velocity of the water in the pipe can be solved for as: 1 2 2 l p z g l f k d v   (5) fig. 7 shows the iterat ive process used for ca lculating water velocity in the channel pipe. first, a frictional loss coeffic ient, 𝑓 (𝑅𝑒 ), is assumed. fro m this, the velocity in the pipe is calculated using eq. 5, and the flow reynolds number is calculated for this velocity and pipe diameter: ( ) p dv re v    (6) fig. 6 surge tank geometry and loss coefficients fig. 7 logic diagram for determining velocity from constant head advances in technology innovation , vol. 3, no. 3, 2018, pp. 101 108 copyright © taeti 107 by refe rring to a moody flow chart, this reynolds number is then used to determine a new frictional loss coefficient. this process is repeated until the loss coefficient converges to a value. the last calculated velocity is then the velocity in the p ipe and is used to determine the flowrate in the water channel. using this approach, the pipe flow velocity was determined to be 233 in/s in the channel pipes , which corresponds to a speed of 8.3 cm per second in the water tunnel. 4.2. pump flowrate the pentair whisperflo 011515, rated at 1.5 kw of power, was selected as the second candidate for running the water channel. the flow rate generated by the pump in the water tunnel is calculated with a procedure simila r to that used for the surge tank calculations, the primary d iffe rence being that a pump does not provide a constant head. therefore, the pump head, 𝐻(𝑄), had to be incorporated into the iterative solution for channel flow . the dynamic head generated by a pump is dependent on flow rate, which introduces another step in the iterative process used to determine the speed of the water channel. since a dynamic head, 𝑃1 /𝛾, is present because of the pump, but there is no net change in elevation by the water across the flow loop (fig. 8), eq. 3 takes on the form: 22 1 2 2 l k vp l v h f d g g     (7) in order to determine the flo w rate generated by the whisperflo 011515, the dynamic head was assumed to be equal to the head drop across the piping of the water tunnel, and the pressures at the entrance and exit of the channel we re assumed to be equal to atmospheric pressure. because flow rate, head, and viscous friction are interrelated, an iterat ive solving approach was once again imp le mented (fig. 9), where the flow rate was determined fro m the pu mp performance curve fo r the whisperflo 011515 (fig. 10). in order to solve for the flowrate, a dynamic head of 20 ft was assumed, and the flowrate was determined fro m the pump performance curve. the pipe frict ion factor was calcu la ted fro m the moody chart for this reynolds number, and the actual head was calculated fro m 𝑉𝑝 and 𝑓. flow rate was determined for this new ca lculated head, and the process was iterated until the calcu lated head converged to 10.9 ft o f water. fro m this h ead, the velocity in the pipes was determined to be 208 in/s , and the calculated speed in the water channel was 7.4 cm/s . according to these calculations, the surge tank and the whisperflo generate water speeds in the channel that vary by only 0.9 c m/second. therefore, there is no significant advantage to constructing a surge tank for attaining greater flow rates. additionally, imp le menting a surge tank requires the design, construction, and space allocation of a tower to support the wat er, and a surge tank will not generate a truly steady flow, since the level in the surge tank changes the actual head by 40% as it drains. finally, this design would not be capable of continuous operation, would require significant modificat ion to change the functionality fro m a flow channel to a tow channel and would still require the purchase of a pump to move water fro m the bottom reservoir up to the surge tank. since all of these factors are resolved by using a pump to continuously power the syst em, and the flow rate gain fro m using a surge tank is negligib le, a pu mp was selected as the head source for generating flow in the water channel. these calculations were va lidated after the channel was co mpleted. the flo w veloc ity distribution had a maximu m speed of 7.1 cm/second in the center and 5.6 cm/second near the walls within the boundary layer. fig. 8 channel pump head and loss coefficients fig. 9 logic d iagra m to calcu late velocity generated by pump advances in technology innovation , vol. 3, no. 3, 2018, pp. 101 108 copyright © taeti 108 fig. 10 performance curve for the pentair whisperflo line of pool pumps. curve i corresponds to the model 011515 [10] 5. conclusions a water channel for flow visualizat ion and scale model testing was successfully designed and manufactured using components and manufacturing methods that are globally availab le. structural ana lysis and flow rate calculations were used to predict tunnel performance based on available materia ls. a high corre lation was found between fea models and the hand calculations using standard texts. flow estimat ions using pipe flow and pipe fitt ing friction factors were used to predict the dynamic head source for the water channel. ma ximu m flow rate ca lculations corre lated closely to measured velocities in the test section. the fra me and work bench are welded fro m mild steel, and the channel t est section is constructed from clea r polycarbonate. the channel workbench footprint is one meter wide and 2.4 meters long, and the test section is 100 c m long with a 30 c m x 40 cross -section. the total cost of materia ls for manufacturing the channel was 1 609.09 usd. this water channel provides an economic alternative for education or basic research of low reynolds number flows. references [1] r. i. emo ri and d. j. schuring, sca le models in engineering: fundamentals and applications , great britain : a. wheaton & co. ltd., 1977. [2] rolling hills research, “water tunnel model 0710,” www.rollinghillsresearch.com/water_tunnels/model_0710.ht ml [3] engineering laboratory design, inc., “water tunnels,” http://www.eldinc.com/pages/0246;watertunnels/ [4] y. farsiani and b. r. elb ing “characterization of the custom-designed, high reynolds number water tunn el,” proc. of the asme fluids engineering division meeting, 2016. [5] w. c. young and r. g. budynas, flat plates, 7th ed. new york: mcgraw hill, 2002. [6] j. jansen, “understanding creep failure of plastics,” plastics engineering, society of plastics engineers, 2015. [7] p. j. gramann, j. cruz, and j. a. jansen, “lifetime prediction of plastic parts,” proc. in annual technical conference, 2012, pp. 1313-1318. [8] b. r. munson, d. f. young, and t. h. okiishi, “fundamentals of fluid mechanics dimensional analysis of pipe flow,” 6th ed. new jersey: john wiley & sons, inc., 2002. [9] l. f. moody, “friction factors for pipe flow,” transactions of the american society of mechanical engineers, pp. 671-682, 1944 [10] whisperflo high performance pump, “whisperflo curves ,” http://www.pentairpool.com/products/index.html. microsoft word aiti#11957 20230811 (in press 1).docx advances in technology innovation, vol. x, no. x, 20xx, pp. xx-xx english language proofreader: chih-wei chang virtual modeling of an industrial robotic arm for energy consumption estimation jin-siang shaw1,*, yi-hua huang2 1department of mechanical engineering, national taipei university of technology, taiwan, roc 2institute of mechatronic engineering, national taipei university of technology, taiwan, roc received 13 april 2023; received in revised form 03 july 2023; accepted 04 july 2023 doi: https://doi.org/10.46604/aiti.2023.11957 abstract this study aims to improve the traditional control methods of industrial robotic arms for path planning in line with efforts to conserve energy and reduce carbon emissions. the digital twin of a six-axis industrial robotic arm with an energy consumption model is innovatively designed. by directly dragging the end effector of a digital twin model, the robotic arm can be controlled for path planning, allowing path tuning to be easily made. in addition, the dynamic equation of the industrial robotic arm is derived, and the energy consumption of the corresponding path can be estimated. four cases are designed to test the validity of the digital twin. experimental results show that the physical robotic arm follows its digital twin model with the corresponding energy consumption computed. the estimated energy consumptions agree quite well with each designed case scenario. keywords: digital twin, robotic arm, unity, euler-lagrange equation, energy consumption 1. introduction in the advanced 21st century, technologies, such as the internet of things (iot), cloud computing, big data analysis, and intelligent robot systems, will cause changes in the global manufacturing industry and lead to an era of intelligent manufacturing in industry 4.0. with the advent of industry 4.0, there has been remarkable growth in the synergistic integration of existing and emerging technologies, for example, the digital twin (dt) in robotics [1], the predictive maintenance based on dt [2], and the dt-driven machining [3]. moreover, gartner predicted that more than 20 billion us of equipment (mostly from the manufacturing industry) would be applied to iot technology [4]. in industry 4.0, engineers need to establish iot and cyber-physical systems for coordinating smart factories, with digital twins playing a crucial role [5]. through computer-integrated manufacturing, digital information technology, data analysis, and data-driven services, appropriate and practical production integration can improve the efficiency, productivity, and flexibility of the production process [6-8]. a virtual model in a dt can use this information to monitor, optimize, and predict real objects by accessing their physical properties, behaviors, and specifications [9]. therefore, the dt concept is considered a promising and innovative research field. however, owing to the complexity of constructing a digital model corresponding to physical space in a virtual space, its application in the field of industrial manufacturing is rarely observed. although digital twinning has been widely recognized as a futuristic and innovative research field, its applications in industrial manufacturing are uncommon due to the complexity of developing digital models in a virtual space that corresponds to and replicates a physical space. currently, most of the research papers and applications related to robotic arms and digital twinning are focused on establishing teaching and demonstration scenarios [10-12]. * corresponding author. e-mail address: jshaw@ntut.edu.tw advances in technology innovation, vol. x, no. x, 20xx, pp. xx-xx 2 the windows operating system is widely used worldwide, as shown in fig. 1 [13]. the most significant difference between ros and windows is the complexity of ros operations. while most software today is designed based on the windows operating system architecture. its scalability is far more convenient than that of ros, and the resources are more abundant. since most people have been in contact with computers set up under the windows operating system architecture since childhood, they are familiar with the operating interface and use computers with much greater proficiency than those under ros architecture. as a result, there is no need to invest additional time learning the two operating systems, ubuntu and ros, and only one operating system can be used to complete the robot research. fig. 1 global desktop operating system market share [13] according to the world robotics 2022 industrial robots report released by the international federation of robotics, there are currently 3.5 million industrial robots used in global factories, and approximately 510,000 units were delivered globally in 2021. the applications include handling, welding, assembly, cleaning, distribution, etc. [14]. however, many industries have not invested in robot-related technologies, and the main problem lies in the limitations of traditional robot control methods. considerable knowledge and technological intervention are required to make it operate normally, which seriously constrains the development of the robotics industry. furthermore, revenue from industrial robots continues to grow globally, and there is a huge energy cost to drive these automated systems with multiple robots operating [15]. typical energy consumption modeling methods for robots rely on kinematic and dynamic characteristics to calculate energy consumption. heredia et al. [16] established mathematical models of robotic arms for energy consumption calculation. these models were parameterized and trained through collected experimental data to evaluate their accuracy. another approach involved using a trained multilayer perceptron (mlp) or resnet model to predict the relevant energy consumption of the robotic arm [17-18]. however, the aforementioned data-driven modeling required a large amount of training data collected from various sensors. in this study, the corresponding energy consumption can be computed solely by the derived energy consumption model with the provided angular speeds of each joint from the dt, and no need to collect training data. this study aims to create a dt and energy consumption model in the windows framework, in which the physical robotic arm moves by dragging the end effector of the virtual robotic arm. in other words, the physical robotic arm can be collaborative in this manner, and the energy consumption of the corresponding path can be computed. with the energy consumption evaluated, the planning path with the lowest energy consumption value can be selected for the physical robotic arm to operate, resulting in significant energy savings. advances in technology innovation, vol. x, no. x, 20xx, pp. xx-xx 3 2. methodology although the simulation of a robotic arm in ros using moveit, gazebo, and rviz is quite common, no cases have been found that utilize the ros framework in constructing a digital twin model capable of estimating energy consumption for robotic arms. in fact, to the best of the authors’ knowledge, no research has been reported on using the dt model that can predict the corresponding energy consumption for robotic arms under either the windows or ros system. since windows is the most popular operating system in the world, the development of a dt of an industrial robotic arm under the windows architecture is planned. the system architecture is illustrated in fig. 2. model preprocessing of the industrial robotic arm is primarily performed because the initial industrial robotic arm model is not similar to that under the ros architecture. instead, similar to the unified robot description format (urdf) file, the urdf file includes all relevant information regarding the industrial robotic arm, such as the pivot point, rotation range, and hue of each joint. therefore, it is necessary to preprocess the original model using computer graphics software. after processing, the model is exported to unity, a software used as a dt. to allow data transmission between the industrial robotic arm and dt software, it is necessary to establish network communication between the two, enabling the realtime transfer of important data required to simulate the state of the physical robot arm in the dt, and vice versa. this study used the “socket” developed in the tcp/ip environment to transfer data between the two. this approach differs from previous control methods of transmitting the world coordinate position of an industrial robot arm and the current rotation angle of each joint. this decision allows users to understand the movement patterns of an industrial robot arm better when it is being controlled for the first time. in addition, a human-machine monitoring interface is established for allowing users quickly understand the system status and control the industrial robotic arm in real-time. to meet the requirements of remote connection control, this interface integrates functions, such as robot arm communication and the real-time angles of each joint. finally, the dynamic equation of the six-axis industrial robotic arm is calculated using the euler-lagrange equation. hence, the required torque of each motor can be estimated for any specific path. fig. 2 system architecture diagram 2.1. building digital twins – unity in this study, the unity software was used to build the dt, which is a cross-platform 2d/3d game engine developed by unity technologies, a game software development company. generally, unity supports the following aspects. first, unity can be used to develop stand-alone games for windows, macos, linux, and other operating systems. second, it’s also available for games on mobile devices such as ios and android. in addition to game development, unity is widely used in the production of real-time 3d animations, architectural visualizations, and other types of projects. lastly, it supports the physx physics engine and particle system and provides network multiplayer connection functionality without requiring the user to learn a complex programming language. given this advantage, it meets the various requirements of game production. thus, the launch of unity has lowered the threshold for game development and made game creation for individuals and small teams realistic. 2.2. industrial robotic arm model preprocessing before exporting the model to the unity game engine, the industrial robotic arm model must undergo preprocessing. meanwhile, the relationship between the joints and connecting links must be defined. the staubli tx60l industrial robotic advances in technology innovation, vol. x, no. x, 20xx, pp. xx-xx 4 arm used in this study has a cad model in the standard triangulated language (stl) format, which is the output from the stp format provided by the manufacturer staubli through solidworks. this format can be used in blender computer graphics software. the parent-child relationship must be defined in blender computer graphics software. therefore, by defining the parent-child relationship of each link of the industrial robotic arm, the concept of this relationship can be controlled effectively, as shown in fig. 3. for example, when turning “axis-1,” all the links that have a parent-child relationship will follow because “axis-1” is defined as the parent of “axis-2” to “axis-6.” it does not affect the base, so the base will not move. however, if the base is moved, the entire model will move along with it because the base is the parent from “axis-1” to “axis-6,” and it indicates the relative relationship of the six axes (axis-1, axis-2, axis-3, axis-4, axis-5, and axis-6) as shown in fig. 4 [19]. fig. 3 parent-child relationship fig. 4 3d model and the relationship of each axis in addition to the above settings, the pivotal point of each link must be specified. the pivot point, also known as the joint center point, defines the motion of the links [20]. the parent-child relationship specifies the motion between the links and the pivot point specifies the motion of the joints, as shown in fig. 5. fig. 5 axis-1 pivot point setup 2.3. kinematics of the industrial robotic arm there are several methods to achieve the movement of a robotic arm by dragging the end effector in a virtual environment. one of the most commonly used methods is the cyclic coordinate descent (ccd) algorithm [21-22]. a popular approach for implementation is to use bio ik [23]. bio ik is the solution of inverse kinematics problems by utilizing a ccd algorithm for calculations. this study used the ccd algorithm for solving inverse kinematics problems and the traditional denavithartenberg (dh) parameter method for solving forward kinematics. by using the fast and less computationally intensive ccd algorithm and traditional dh parameter method, this study aimed to solve forward and inverse kinematics problems. advances in technology innovation, vol. x, no. x, 20xx, pp. xx-xx 5 the ccd algorithm is currently one of the simplest and most popular methods for solving inverse kinematics, and it has been widely used in the computer game industry. the ccd algorithm calculates from one joint to another joint angle such that the endpoint is as close as possible to the target. after each iteration update, the algorithm measures the distance between the endpoint and the target to ensure that the endpoint is far enough from the target or already required to stop computing. additionally, to avoid infinite recursive loops caused by unreachable and conflicting target points, the algorithm must be limited by setting a maximum number of iterations. essentially, the base of an inverse kinematic-driven mechanism is immovable, and the axis and angle are calculated using the formula, respectively, as shown in fig. 6. 1cos e c t c e c t c p p p p p p p p θ −         − − = • − − (1) e c t c e c t c p p p p p p p p r − − × − − = � (2) this method is also applicable to 2d and 3d coordinates by aligning a joint (pc) with the endpoint (pe) and target, and the endpoint is finally approached by reciprocating the above steps towards the target. to solve the problem of forward kinematics [24], this study used the dh parameter method. the dh parameter method designs a coordinate system for each connecting rod according to set rules, and then this coordinate system can be used to describe the transformation relationship from one connecting rod coordinate system to the adjacent connecting rod coordinate system. the transformation of the adjacent coordinate system is decomposed into several steps with only one parameter respectively. the combination of the corresponding transformations completes the transformation of the adjacent coordinate systems. fig. 6 ccd algorithm flow 2.4. digital twin user interface for the industrial robotic arm two scenarios were considered in this study. the first scenario used forward kinematics to control the virtual robotic arm and synchronously moved the physical robotic arm. when the physical robotic arm was controlled by the robotic arm user interface, the virtual robotic arm moved synchronously. as shown in fig. 7, after pressing the start received button (button 1), the arm begins to receive the six joint information read from the cs8c controller. this information was used to move the virtual robotic arm into a physical robotic arm. each joint of the virtual robotic arm can be rotated by controlling the plus/minus advances in technology innovation, vol. x, no. x, 20xx, pp. xx-xx 6 buttons (button 2) of each joint. the rotation angle information was transmitted to the physical robotic arm which rotated accordingly. after achieving the goal of mutual control between the two parties, press the quit game button (button 3) to end the currently executing scene. fig. 7 unity digital twin scene 1 the second scene primarily involves moving the virtual robotic arm through inverse kinematics, as shown in fig. 8. in this scene, a red sphere was designed as the target object (target). the end effector moved with the target object. the movement of the virtual robotic arm was controlled by dragging the target object. when it moved to a position, the dragging was stopped, and the angle of each joint was calculated using the ccd script designed in unity. this angle information was sent to the physical robot arm such that it could move like a virtual robotic arm. the goal was to control the physical robot arm by dragging the dt end effector. the point correction of a traditional industrial robotic arm was significantly reduced by dragging the dt end effector to the designated target point for the corresponding work settings, which makes it more flexible and can be used in complex environments. such a dt with a user interface can also be used to establish a fully realistic workplace and perform path planning and other operations on the robotic arm through a remote control, thereby enabling task completion without entering the factory. fig. 8 unity digital twin scene 2 2.5. matlab energy consumption calculation with the gradual advancement of science and technology and the development of digital avatar-related technologies, factories are moving towards smart manufacturing, and smart factories are becoming a major trend. through these technologies, factory personnel can monitor the internal conditions of a factory in real-time using computers and mobile phones. with the advances in technology innovation, vol. x, no. x, 20xx, pp. xx-xx 7 rise of international oil prices and the improvement of green energy-related technologies, taiwan officially announced the “taiwan 2050 net zero emissions path and strategy general description” in march 2022 [25]. energy conversion is a critical issue for achieving the goal of net-zero emissions. in the current situation, where green energy and other related technologies are not yet mature, effective management of equipment energy consumption in factories is crucial. this study aimed to address this issue using the dt of the robot arm in unity and inputting the simulated robotic arm path into matlab for energy consumption estimation. therefore, a dynamic model of a robot arm, which is a complicated and nonlinear system, is required. this study used a dynamic model of the robotic arm to calculate the torque of each joint motor through the moving path of the robotic arm dt environment established above and computed the total energy required for this moving path. the purpose was to allow users to use unity to first perform path planning with the dt of the robotic arm and then obtain the total energy consumption of the path through matlab calculations. this method enabled users to carry out various path planning through the digital avatar of the robotic arm and the realistic environment of the actual scene. by analyzing different paths, users can select and execute the path that consumes the least amount of energy. after obtaining the path that consumed the lowest energy, the physical robotic arm could operate according to the path. in robotics, the euler-lagrange equation can be used to calculate torque [26]. , 1, ,i i i d l l i m dt q q τ ∂ ∂ − = = ∂ ∂ … ɺ (3) the lagrangian, l, is defined as the kinetic energy minus the potential energy. l k p= − (4) the kinetic and potential energy equations of the n-link robotic arm are expressed in: ( ) ( ) ( ) ( ) ( ) ( ) 1 1 2 n t t tt i vi vi wi i i i wi i q m j q j q j q r q i r q j q qk =      += ɺ ɺ (5) 1 1 n n t i ci i i i p p g r m = = = =  (6) where �� and q denote vectors of joint angular velocity and position, respectively, n denotes the total number of joints, �� denotes the torque of the ith motor, �� denotes the mass of the i th link, �� � � and �� � � represent the jacobians of the i th link speed and angular velocity, respectively, �� � denotes the orientation matrix of the i th link, �� denotes the inertia tensor of the ith link, g denotes the gravity vector, and finally ��� denotes the mass center vector of the i th link. after obtaining the equations for kinetic and potential energies, eqs. (5) and (6) are substituted into eq. (4), and eq. (4) is substituted into eq. (3), which can be used to obtain the dynamic equation of the robotic arm in a vector-matrix form. ( ) ( ) ( ),q q c q q q g qm τ+ + =ɺɺ ɺ ɺ (7) where �� represents the vector of joint angular acceleration. m represents the n × n inertia matrix, c represents the n × n matrix of the coriolis force and centrifugal force, and g represents the gravity of the n × 1 vector. parameters in m, c, and g in eq. (7) will be identified by the cad model provided by the manufacturer staubli, such as the geometry and property of each link, the link’s moment of inertia, etc. the joint variable q can be obtained by applying the ccd algorithm for the inverse problem. therefore, the torque of each link executing the path �� , �� , and q can be computed in eq. (7) by summing the three terms on the left-hand side. the total power required for the robotic arm is then estimated by 1 τ θ η ==  ɺ n i i i i p (8) advances in technology innovation, vol. x, no. x, 20xx, pp. xx-xx 8 where �� is the transfer efficiency of the i th motor. the resulting energy consumption of the robotic arm for this path is determined by integrating the power over time, 0 ft t w pdt=  (9) where �� is the final time; �� is the initial time. it is noted that the dynamic equation in eq. (7), an often-seen model for a robotic arm in the literature, does not consider friction between the robots’ joints. this is because friction in robot joints including coulomb friction and viscous friction is difficult to exactly identify the power it consumes constitutes only a small part of the total power. here �� is used to represent the transfer efficiency of the i th motor to reflect this friction power loss. by performing these calculations in eqs. (7)-(9), it can determine the energy consumption required for different paths and hence select the path with the lowest energy consumption to perform the work. 3. exmental and result this study moved the end effector in four different scenarios, as shown in fig. 9: case 1 involved moving in a straight line, case 2 involved moving twice the distance of case 1, case 3 involved moving in a small circle, and case 4 involved moving in a large circle. these scenarios allowed us to compare the energy consumption of cases 1 and 2, where the moving path of case 2 was twice that of case 1, resulting in higher energy consumption. similarly, the energy consumption in case 3 was expected to be lower than that in case 4. (a) case 1 (b) case 2 (c) case 3 (d) case 4 fig. 9 different cases of the simulation for energy consumption advances in technology innovation, vol. x, no. x, 20xx, pp. xx-xx 9 in case 1, the distance between a and b was 0.5 m. in case 2, it was 1.0 m. in the case of circular motion, the large circle had a radius of 0.5 m, and the small circle had a radius of 0.25 m. the movement speed was set to 4 cm/sec for case 1, where case 2 had twice the speed of case 1. the movement speed was set to 11 cm/sec for case 3, where case 4 had twice the speed of case 3. the reason for choosing low speed in experiments was to facilitate the integration of the virtual and physical environments, which made it easier to control the robotic arm and avoid collisions and other factors. as shown in fig. 10, the total power diagram was calculated using eqs. (7) and (8), while eq. (9) was used to compute the resulting energy consumption. table 1 listed the final energy consumption for each case, which agreed with the predictions. specifically, when the robotic arm’s travel distance and travel speed were doubled, the corresponding energy consumption was also more than doubled (nonlinear), as seen in case 2 over case 1 and case 4 over case 3. moreover, the ratio of energy consumption of case 2 over case 1 (2.9) was slightly larger than that of case 4 over case 3 (2.4). this can be attributed to the fact that a straight-line trajectory (cases 1 and 2) requires more joints to execute compared to a circular trajectory (cases 3 and 4) for an articulated robotic arm. additionally, in eq. (9), it was evident that the energy consumption was highly nonlinear concerning the operational conditions. to verify and validate the accuracy of the energy consumption model, appropriate sensors such as torque and angular speed sensors can be installed on the robotic arm to compute the true energy consumption values. however, it remains a task for future work. (a) case 1 (b) case 2 (c) case 3 (d) case 4 fig. 10 power response for each case table 1 energy consumption of each case case energy consumption (j) 1 38 2 91 3 95 4 195 advances in technology innovation, vol. x, no. x, 20xx, pp. xx-xx 10 4. conclusions and future work the research utilizes the unity game engine and the robotic principle of the euler-lagrange equation to establish a digital twin and energy consumption model for an industrial robot arm. the final energy performance is described through four case studies. it was observed that the motion length also influenced the calculated energy expenditure accordingly. by directly manipulating the end effector of the dt model, the industrial arm can be easily controlled remotely for task scheduling without the need to physically enter the factory, introducing a new form of control. additionally, the digital twins and energy consumption models have demonstrated a reasonable level of accuracy. in future studies, dt models of robotic arms capable of handling various loads in real-world applications will be considered. to enhance the accuracy of the energy consumption model, installing torque, comparing the results with the model’s estimated energy consumption, and the angular velocity sensors on the physical industrial robotic arm to measure its energy consumption are required. conflicts of interest the authors declare no conflict of interest. references [1] a. mazumder, m. f. sahed, z. tasneem, p. das, f. r. badal, m. f. ali, et al., “towards next generation digital twin in robotics: trends, scopes, challenges, and future,” heliyon, vol. 9, no. 2, article no. e13359, february 2023. [2] d. zhong, z. xia, y. zhu, and j. duan, “overview of predictive maintenance based on digital twin technology,” heliyon, vol. 9, no. 4, article no. e14534, april 2023. [3] s. liu, j. bao, and p. zheng, “a review of digital twin-driven machining: from digitization to intellectualization,” journal of manufacturing systems, vol. 67, pp. 361-378, april 2023. [4] y. cui, s. kara, and k. c. chan, “manufacturing big data ecosystem: a systematic literature review,” robotics and computer-integrated manufacturing, vol. 62, article no. 101861, april 2020. [5] m. a. m. s. lemstra and m. a. de mesquita, “industry 4.0: a tertiary literature review,” technological forecasting and social change, vol. 186, no. part b, article no. 122204, january 2023. [6] j. cheng, h. zhang, f. tao, and c. f. juang, “dt-ii: digital twin enhanced industrial internet reference framework towards smart manufacturing,” robotics and computer-integrated manufacturing, vol. 62, article no. 101881, april 2020. [7] m. liu, s. fang, h. dong, and c. xu, “review of digital twin about concepts, technologies, and industrial applications,” journal of manufacturing systems, vol. 58, no. part b, pp. 346-361, january 2021. [8] y. lu, c. liu, k. i. k. wang, h. huang, and x. xu, “digital twin-driven smart manufacturing: connotation, reference model, applications and research issues,” robotics and computer-integrated manufacturing, vol. 61, article no. 101837, february 2020. [9] a. v. barenji, x. liu, h. guo, and z. li “a digital twin-driven approach towards smart manufacturing: reduced energy consumption for a robotic cell,” international journal of computer integrated manufacturing, vol. 34, no. 7-8, pp. 844-859, 2021. [10] c. constantinescu, s. giosan, r. matei, and d. wohlfeld, “a holistic methodology for development of real-time digital twins,” procedia cirp, vol. 88, pp. 163-166, 2020. [11] a. gallala, a. a. kumar, b. hichri, and p. plapper, “digital twin for human–robot interactions by means of industry 4.0 enabling technologies,” sensors, vol. 22, no. 13, article no. 4950, july 2022. [12] y. palazhchenko, v. shendryk, and s. shendryk, “digital twins data visualization methods. problems of human interaction: a review,” in international conference “new technologies, development and applications”, pp. 478-485, may 2023. [13] statcounter globalstats, “desktop operating system market share worldwide,” https://gs.statcounter.com/os-marketshare/desktop/worldwide, september 30, 2022. advances in technology innovation, vol. x, no. x, 20xx, pp. xx-xx 11 [14] j. oyekan, m. farnsworth, w. hutabarat, d. miller, and a. tiwari, “applying a 6 dof robotic arm and digital twin to automate fan-blade reconditioning for aerospace maintenance, repair, and overhaul,” sensors, vol. 20, no. 16, article no. 4637, august 2020. [15] m. gadaleta, g. berselli, m. pellicciari, and f. grassia, “extensive experimental investigation for the optimization of the energy consumption of a high payload industrial robot with open research dataset,” robotics and computerintegrated manufacturing, vol. 68, article no. 102046, april 2021. [16] j. heredia, c. schlette, and m. b. kjærgaard, “data-driven energy estimation of individual instructions in userdefined robot programs for collaborative robots,” ieee robotics and automation letters, vol. 6, no. 4, pp. 68366843, october 2021. [17] j. yan and m. zhang, “a transfer-learning based energy consumption modeling method for industrial robots,” journal of cleaner production, vol. 325, article no. 129299, november 2021. [18] m. yao, q. zhao, z. shao, and y. zhao, “research on power modeling of the industrial robot based on resnet,” 7th international conference on automation, control and robotics engineering, pp. 87-92, july 2022. [19] g. garg, “digital twin for industrial robotics,” m.s. thesis, department of robotics and computer engineering, university of tartu, tartu, 2021. [20] g. garg, v. kuts, and g. anbarjafari, “digital twin for fanuc robots: industrial robot programming and simulation using virtual reality,” sustainability, vol. 13, no. 18, article no. 10336, september 2021. [21] p. yotchon and y. jewajinda, “combining a differential evolution algorithm with cyclic coordinate descent for inverse kinematics of manipulator robot,” 3rd international conference on electronics representation and algorithm pp. 35-40, july 2021. [22] c. jacob, f. espinosa, a. luxenburger, d. merkel, j. mohr, t. schwartz, et al., “digital twins for distributed collaborative work in shared production,” ieee international conference on artificial intelligence and virtual reality, pp. 210-212, december 2022. [23] g. gheorghe, “mathematical modeling of the walking hexapod robot using denavit-hartenberg parameters,” annals of ‘constantin brancusi’ university of targu-jiu. engineering series, no. 1, pp. 64-69, 2021. [24] a. vatankhah barenji, x. liu, h. guo, and z. li, “a digital twin-driven approach towards smart manufacturing: reduced energy consumption for a robotic cell,” international journal of computer integrated manufacturing, vol. 34, no. 7-8, pp. 844-859, 2021. [25] national development council, “taiwan’s 2050 net-zero emission,” https://www.ndc.gov.tw/content_list.aspx?n=fd76ecbae77d9811, march 30, 2022 [26] m. w. spong, s. hutchinson, and m. vidyasagar, robot modeling and control, 1st ed., new jersey: wiley, 2005. copyright© by the authors. licensee taeti, taiwan. this article is an open-access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). microsoft word 1-v8n1(2023)-aiti#10658(01-11).docx advances in technology innovation, vol. 8, no. 1, 2023, pp. 01-11 optimization of weld parameters in wire and arc-based directed energy deposition of high strength low alloy steels van thao le1,*, dinh si mai1, van thuc dang1, duc manh dinh1, thi hong cao2, van anh nguyen3 1advanced technology center, le quy don technical university, hanoi, vietnam 2institute for tropical technology, vietnam academy of science and technology, hanoi, vietnam 3welding engineering and laser processing centre, cranfield university, bedford, uk received 10 august 2022; received in revised form 27 september 2022; accepted 30 september 2022 doi: https://doi.org/10.46604/aiti.2023.10658 abstract this paper aims to investigate the fabrication of high strength low alloy (hsla) steels by wire and arc-based directed energy deposition (waded). firstly, the relationship between the process variables (including the travel speed-v, the current-c, and the voltage-u) and the geometrical characteristics of weld beads (including the bead height (bh), bead width (bw), and melting pool length (mpl)) was investigated. secondly, the optimal process variables were identified using the desirability approach. the results indicate that voltage-u has the highest impact on bw and mpl, meanwhile the travel speed-v is the most impacting factor on bh. the optimal variables for the waded process of hsal steels are v = 0.3 m/min, c = 160 a, and u = 19 v. the component fabricated with the optimal variables is fully dense without spatters and defects, confirming the efficiency of the waded process for hsla steels. keywords: waded, hsla steel, weld bead, optimal variables 1. introduction emerged since the 1980s, additive manufacturing (am) technologies, especially metallic am, are strongly developed, and they are becoming the key technology in the industry 4.0 era [1]. because of the layer-by-layer manufacturing principle, am technologies can fabricate very complex structures from various materials, including metallic alloys that are very difficult to machine by conventional processes such as milling and turning [2]. the metallic am technologies can be categorized based on the energy source and feedstock form or the fabrication methods. according to the feedstock form and the fabrication methods, there are two main groups of metallic am, i.e., powder bed fusion (pbf) and directed energy deposition (ded) [3]. compared to pbf-am processes, ded-am can produce metallic components with larger dimensions and can be applied effectively for repairing and remanufacturing applications [4-7]. the ded-am processes include two technologies: (i) laser and powder-based ded (lpded) using a laser source to melt metal powder (fig. 1(a)), and (ii) wire and arc-based ded (waded) utilizing an arc source to melt the metal wire (fig. 1(b)). the nozzle in the waded is a welding torch, and the wire feeding method depends on the used welding source, for example, gas tungsten arc welding (gtaw), plasma arc welding (paw), and gas metal arc welding (gmaw) [8]. compared to other metal am technologies, waded has a superior rate of material deposition (from 3 to 8 kg/h), high efficiency of material utilization, low costs of investing systems and devices, and easy implementation [9]. * corresponding author. e-mail address: vtle@lqdtu.edu.vn advances in technology innovation, vol. 8, no. 1, 2023, pp. 01-11 2 the metal wire available in the welding market can be used for waded processes. they are also much cheaper than metal powder used in pbf-am and lpded. therefore, waded is becoming a potential solution for manufacturing components with large and wide dimensions [10]. (a) the lpded process [11] (b) the waded process fig. 1 schemas of lpded and waded processes among metals used in waded, including steels, aluminum, titanium, and nickel-based alloys, steels are the most investigated materials because of their relatively low costs, wide applications, and availability in the welding market. in the literature, many steel grades have been investigated and fabricated with the waded processes, for example, low-carbon steels (e.g., er70s-6) [12-14], austenite stainless steels (e.g., 304, 308l, 308lsi, 309l, and 316l) [15], and highstrength low alloy (hsla) steels (e.g., er110s-g and er120s-g) [16-18]. the hsla steels are largely utilized for manufacturing lightweight structures and high strength components in many sectors, for example, automotive, shipbuilding, tools and die industries. these steel grades have high strength, toughness, weldability, and low costs [19]. recently, many researchers have investigated and fabricated components from hsla steels by the waded processes. sun et al. [20] produced thin-wall hsla steel components by waded and examined the anisotropy in mechanical characteristics. they stated that the strengths of the as-fabricated component in the horizontal direction were lower than those in the vertical direction. rodrigues et al. [16] studied the impact of the heat input on the microstructures and mechanical properties of the wadeded thin-wall hsla steel part. they observed that there were no significant differences in microstructure between the specimens fabricated with low and high levels of heat input. dirisu et al. [19] analyzed the toughness properties of hsla steels produced by waded. they found that the refinement in grains and the increase in density of grain boundaries resulted in high resistance to failure of the samples in the vertical direction. fang et al. [21] also fabricated hsla steel parts by waded. the researcher observed that the part had a superior balance in strength and ductility. the variation in microhardness was due to the difference of thermal histories in different regions of the component. although certain authors have investigated hsla steels fabricated by the waded processes, as mentioned above, they mainly analyzed microstructures and mechanical characterization of the as-built material, as well as the influence of process variables on the quality of the part. until now, the studies on predicting and optimizing process variables in waded of hsla steel to obtain the expected geometrical properties of weld beads are still limited. most of the previously mentioned studies have selected the process variables (e.g., the arc voltage, welding current, and wire-feed speed) based on the suggestion of the wire manufacturers for conventional welding or based on several tests, while the waded process is very different from welding. in waded processes, the quality and geometry of weld beads (e.g., stable and smooth shape and less spatter) notably affect the process stability and the appearance of as-fabricated components [22-25]. therefore, this study aims to explore the relationship between the process variables {c, u, and v} and the geometrical characteristics of weld beads {bw, bh, and mpl} and identify optimal process parameters that produce weld beads with expected geometrical properties. advances in technology innovation, vol. 8, no. 1, 2023, pp. 01-11 3 2. materials and methods 2.1. materials and the waded system in this investigation, the hsla steel wire (sm-110) with a 1.2-mm diameter supplied by hyundai welding and the low-carbon steel substrates with a size of 200 × 200 × 10 mm had been utilized. the chemical elements of the wire are 1.95%ni, 1.90%mo, 0.34%cr, 0.58%mo, 0.80%si, 0.089%c, 0.010%p, 0.004%s, and %fe in balance (wt.%). a waded system (fig. 2(a)) consists of a gmaw source, a welding torch, a welding wire feeder, a 6-axis robot, and a shielding gas feeder was used to fabricate the samples (e.g., single weld tracks, fig. 2(b) and fig. 2(c)). the relationships between the wire feed speed and welding current in this gmaw-am system is approximately increasing linearly. a mixed gas of argon (80%) and co2 (20%) with 16 l/min in flow speed was applied during the waded process. (a) the welding robot (b) weld bead samples (c) wb’s characteristics fig. 2 the waded system, the wb samples with the measurement area, and the response description 2.2. experimental procedure to achieve the goal of the study, the research procedure shown in fig. 3 was performed. the research procedure includes the following steps: (i) identification of the process variables and their ranges, (ii) design of experiment, (iii) fabrication of single weld beads by the waded process, (iv) measurement of the responses (i.e., bh, bw, and mpl), development of regression models for the responses and determining the impact of process variables on the responses through the analysis of variance (anova), and (v) the optimization of process variables. to observe the relationship between the process variables and the responses, the taguchi l9 orthogonal array was adopted to design the experimental plan. three input variables, including the current-c, the travel speed of the weld torch-v, and the voltage-u were selected for the investigation because they directly influence the shape and dimensions of the weld beads. three levels for each variable were used (table 1). as a result, there were nine experiment runs. table 1 input variables and their levels for the doe item levels voltage-u (v) 18 20 22 current-c (a) 120 140 160 travel speed-v (m/min) 0.3 0.4 0.5 advances in technology innovation, vol. 8, no. 1, 2023, pp. 01-11 4 fig. 3 the study flowchart after the fabrication of weld bead samples, the responses of weld beads, including bh, bw, and mpl were measured. these characteristics of weld beads play important roles in the waded process. bh and bw are related to the layer width and height, whereas the mpl is related to the planning of deposition paths. the measurements were executed in the stable regions of welding beads (fig. 2(b)) utilizing a digital mitutoyo caliper with 0.01 mm in resolution and ± 0.02 mm in accuracy. the bw and bh were measured five times at five locations in the middle zone of weld beads (fig. 2(b)). finally, the average value of the five measurements of bw and bh was taken for analysis. on the other hand, the measurement of mpl was repeated three times to ensure the reliability of the measurement results. the experiment runs and measured results were presented in table 2. table 2 experiment runs and the response measurement experiment run input variables responses current-c (a) voltage-u (v) travel speed-v (m/min) bw (mm) bh (mm) mpl (mm) 1 120 18 0.3 5.39 3.09 11.26 2 120 20 0.4 5.86 2.54 11.64 3 120 22 0.5 5.82 1.82 12.39 4 140 18 0.4 4.58 2.92 10.49 5 140 20 0.5 5.41 2.33 11.34 6 140 22 0.3 7.51 2.81 13.52 7 160 18 0.5 4.18 2.77 9.79 8 160 20 0.3 6.87 3.00 11.95 9 160 22 0.4 7.82 2.37 13.24 2.3. analysis and optimization methods to evaluate the impact of input variables on the responses and identify the impact contribution of each input, the anova and the minitab software was adopted. the anova was executed with a 5% significance level and a 95% confidence level. in the waded, bh, and bw are expected to be maximal while the mpl is minimal to enhance productivity. as a result, the optimization problem was formulated as eq. (1). { } { } { } { } , , , 120 160 , 18 22 , 0.3 0.5 / find c u v to maximize bh bw and minimize mpl subject to c a u v v m min≤ ≤ ≤ ≤ ≤ ≤      (1) advances in technology innovation, vol. 8, no. 1, 2023, pp. 01-11 5 the optimal input variables were estimated by the desirability function (df) method [26]. moreover, the weight for each response (i.e., bw, bh, and mpl) was identified by the critic method [27]. the steps of the critic method are described as follows: step 1: construct the decision matrix (dm) of k experimental runs and p evaluation attributes: dm = [��� ] ×� . step 2: normalize the dm using eq. (2): ˆ worst ij j ij best worst j j m m m m m − = − (2) where �� �� is the normalized value of the �� alternative for ��� attribute, �� ���� and �� ����� are the best and worst values of ��� attribute. step 3: calculate the standard deviation of each normalized attribute using eq. (3): 2 1 ˆ( ) k ij j i j m m k σ = − =  (3) where k is the number of experimental runs and ��� is the average value of � �� normalized attribute. step 4: construct the symmetric matrix [���� ] × with the linear-correlation coefficient ���� between the attributes. ���� is calculated by eq. (4): 1 2 2 1 1 ˆ ˆ( )( ) ˆ ˆ( ) ( ) k li i lj j l ij k k li i lj j l l m m m m cc m m m m = = = − − = − −    (4) step 5: calculate the attribute information ��� by eq. (5): 1 (1 ) k j j jl l ai ccσ = = − (5) step 6: calculate the weight �� for each attribute (i.e., response): 1 j j p j j ai w ai = =  (6) 3. results and discussion 3.1. regression models the regressive models of the responses bw, bh, and mpl are shown in tables 3, 4, and 5, respectively. they were developed with the help of minitab software. in the case of bw, the p-values of u and v in table 3, are smaller than 0.05, while the p-value of c is bigger than 0.05, meaning that u and v are the significant terms of the bw model. the values of r-sq, r-sq(adj), and r-sq(pred) are 95.67%, 93.07%, and 83.83%, respectively. it indicates that the bw model has acceptable accuracy and can terms be used to predict the response in the entire design space. for the bh model, all the p-values of c, u, and v in table 4 are smaller than 0.05. therefore, all the model terms c, u, and v are significant. the values of the determination coefficients r-sq, r-sq(adj), and r-sq(pred) are 96.51%, 94.72%, and advances in technology innovation, vol. 8, no. 1, 2023, pp. 01-11 6 86.18%, respectively, indicating a reasonable accuracy of the bh model. this model can also be used to predict the response in the entire design space. for the developed model of mpl, the p-values of u and v in table 5 are smaller than 0.05, while the p-value of c is bigger than 0.05, indicating that u and v are the significant term of the mpl model. the values of r-sq, r-sq(adj), and r-sq(pred) are 97.73%, 96.36%, and 91.15%, respectively. therefore, the mpl model has an acceptable accuracy, and it can be used to predict the responses (i.e., bw, bh, and mpl) in the whole design space. table 3 anova of bw source df seq ss contribution adj ss adj ms f-value p-value regression 3 11.8791 95.67% 11.8791 3.9597 36.81 0.001 c 1 0.5424 4.37% 0.5424 0.5424 5.04 0.075 u 1 8.1713 65.81% 8.1713 8.1713 75.97 0.000 v 1 3.1654 25.49% 3.1654 3.1654 29.43 0.003 error 5 0.5378 4.33% 0.5378 0.1076 total 8 12.4169 100.00% regressive model �� = �4.93 # 0.01503 × ' # 0.5835 × ) � 7.26 × r-sq = 95.67% r-sq(adj) = 93.07% r-sq(pred) = 83.83% table 4 anova of bh source df seq ss contribution adj ss adj ms f-value p-value regression 3 1.26082 96.51% 1.26082 0.420272 46.14 0.000 c 1 0.07935 6.07% 0.07935 0.079350 8.71 0.032 u 1 0.52807 40.42% 0.52807 0.528067 57.98 0.001 v 1 0.65340 50.02% 0.65340 0.653400 71.74 0.000 error 5 0.04554 3.49% 0.04554 0.009108 total 8 1.30636 100.00% regressive model �. = 6.109 # 0.00575 × ' � 0.1483 × ) � 3.300 × r-sq = 96.51% r-sq(adj) = 94.42% r-sq(pred) = 86.18% table 5 anova of mpl source df seq ss contribution adj ss adj ms f-value p-value regression 3 11.3854 97.73% 11.3854 3.79513 71.65 0.000 c 1 0.0160 0.14% 0.0160 0.01602 0.30 0.606 u 1 9.6520 82.85% 9.6520 9.65202 182.22 0.000 v 1 1.7173 14.74% 1.7173 1.71735 32.42 0.002 error 5 0.2648 2.27% 0.2648 0.05297 total 8 11.6502 100.00% regressive model /01 = 1.55 � 0.00258 × ' # 0.6342 × ) � 5.350 × r-sq = 97.73% r-sq(adj) = 96.36% r-sq(pred) = 91.15% 3.2. relationship between the input variables and the responses as shown in fig. 4(a), it is indicated that the bw increases when u and c increment. meanwhile, v shows an opposite impact tendency. the bw decreases with the augmentation in v. based on the anova results (table 3), the voltage-u has the most impact contribution to the bw with 65.81%, followed by the travel speed (25.49%), and the current (4.37%), respectively. the influence of the input variables on the bw can be explained as follows. an increase in voltage also makes an increase in the length and spreading of the arc. therefore, bw becomes larger with a higher level of voltage [28]. an increase in welding current leads to an increase in wire feed speed and material deposition, resulting in an augmentation in melting pool size and in advances in technology innovation, vol. 8, no. 1, 2023, pp. 01-11 7 the width of weld beads (bw) [23]. oppositely, an increase in travel speed causes a reduction in material deposition quantity per length unit. therefore, the bw becomes narrow as the travel speed increases [28]. the influence of the input variables on the bh is shown in fig. 4(b). it is shown that the increase in both the voltage-u and the travel speed-v causes a decrease in the bh. on the other hand, the bh increases when the current-c increase. however, the bh sightly increases when c increases from 140 a to 160 a. in this case, the travel speed-v reveals the highest impact on the bh with a contribution of 50.02%, followed by the voltage-u (40.42%), and the current-c (6.02%), respectively (table 4). indeed, when increasing the travel speed, the quantity of deposited materials per length unit is reduced. hence, bh is reduced [28-29]. the spreading area of the arc is larger when the voltage increases, leading to flatter weld beads [30]. as a result, bh reveals a reducing trend with an increment in voltage. as c increases, the wire feeding speed augments. hence, the volume of materials deposited increases, leading to an increase in bh [28]. for the mpl, as depicted in fig. 4(c), it strongly increases when the voltage increases, meanwhile the mpl slightly increases with the increase in the current-c. the mpl, on the other hand, increases as v augments from 0.3 m/min to 0.4 m/min, after that it decreases. these findings can be explained by the anova results (table 5). it is shown that the voltage-c has the most influential contribution of 82.85% to mpl, while the current has the lowest impact contribution of 0.14% to mpl. (a) effects of process variables on bw (b) effects of process variables on bh (c) effects of process variables on mpl fig. 4 influences of the input variables on the responses 3.3. optimization results as mentioned previously, the weights for the responses were calculated by the critic method. the weight values of bw, bh, and mpl were 0.42, 0.23, and 0.34, respectively. the solution of the multi-attribute optimization problem (eq. (1)) was shown in fig. 5. it indicated that the optimal input variables are {c = 160 a, v = 0.3 m/min, and u = 19 v}. this set of input parameters is corresponding to the df value of 0.8651 and the predicted responses bw = 6.39 mm, bh = 3.22 mm, and mpl = 11.59 mm. compared to the worst case, where the bh was the smallest (the experiment run #3), the optimal input parameters allow enhancing the bh and bw by 77% and 10%, respectively, while reducing the mpl by 6%. advances in technology innovation, vol. 8, no. 1, 2023, pp. 01-11 8 fig. 5 optimization solution to confirm the efficiency of the optimal input parameters, they were tested to fabricate a three-bead multi-layer-cylindric wall (fig. 6). it is shown that the optimal parameters enable the production of smooth weld beads without spatters and major defects. there is only a minor defect in shape on the top surface at the stopping point of the weld path. in this investigation, as the starting point and the ending point of the weld path were not identical, there was an overlap between the arc-striking and the arc-extinguishing regions of the arc (fig. 7(a)). therefore, the difference in height between the arc-striking and the arc-extinguishing regions was compensated. on the other hand, when the starting and the ending points are identical, there is a space in the circle weld bead between the starting and the ending points (fig. 7(b). as the number of layers increases, the gap depth increases, resulting in major defects in the shape of the part. fig. 6 multi-bead multi-layer cylindric part (a) not identical (b) identical fig. 7 programming of a circle weld path: the starting point and the ending point advances in technology innovation, vol. 8, no. 1, 2023, pp. 01-11 9 moreover, the as-built material is fully dense without defects such as pores, cracks, and lack of fusion. the microstructure mainly consists of α-ferrite phases in acicular and granular morphologies (fig. 8(a)). the edx analysis results also indicate that the as-built material has similar chemical elements as the wire material (fig. 8(b)). the melted metal was perfectly protected by the shielding gas during the waded process. (a) microstructure (b) chemical elements fig. 8 microstructure and chemical elements of the as-built material 4. conclusions in this study, hsla steel was utilized as the raw material in the waded process. the study aims to predict the connection between the main process variables {c, u, and v} and the geometrical characteristics of weld beads {bw, bh, and mpl}. furthermore, finding the optimal process variables is also the study’s aim. the main findings of this investigation are highlighted as follows. (1) the voltage-u has the highest impact on bw and mpl, meanwhile the travel speed-v is the most impacting factor on bh. an increase in u leads to an increase in bw and mpl and a decrease in bh. on the other hand, an increase in v causes a decrease in both bw and bh. (2) all the predictive models of bw, bh, and mpl have an acceptable accuracy with the determination values r-sq = 95.67%, 96.51%, and 97.73%, respectively. they can be applied to predict the responses in the entire design space. (3) the optimal process variables for the waded process of sm-110 hsla steel are v = 0.3 m/min, c = 160 a, and u = 19 v. the component fabricated with the optimal variables has a smooth top surface. moreover, spatters and defects don’t present in the fabrication. the as-built material is fully dense without defects such as pores, cracks, and lack of fusion. (4) the microstructure of the as-built material is mainly composed of ferrite phases with acicular and granular morphologies. its chemical elements also have the comparable weight percentage to those of the wire material. conflicts of interest the authors declare no conflict of interest. acknowledgment this investigation is funded by le quy don technical university under grant number đh.03.2022. the authors would like to acknowledge great supports from weldtech company for our project. references [1] u. m. dilberoglu, b. gharehpapagh, u. yaman, and m. dolen, “the role of additive manufacturing in the era of industry 4.0,” procedia manufacturing, vol. 11, pp. 545-554, 2017. advances in technology innovation, vol. 8, no. 1, 2023, pp. 01-11 10 [2] s. m. thompson, l. bian, n. shamsaei, and a. yadollahi, “an overview of direct laser deposition for additive manufacturing; part i: transport phenomena, modeling and diagnostics,” additive manufacturing, vol. 8, pp. 36-62, october 2015. [3] v. t. le, h. paris, and g. mandil, “the development of a strategy for direct part reuse using additive and subtractive manufacturing technologies,” additive manufacturing, vol. 22, pp. 687-699, august 2018. [4] v. t. le, h. paris, and g. mandil, “process planning for combined additive and subtractive manufacturing technologies in a remanufacturing context,” journal of manufacturing systems, vol. 44, no. 1, pp. 243-254, july 2017. [5] a. ramalho, t. g. santos, b. bevans, z. smoqi, p. rao, and j. p. oliveira, “effect of contaminations on the acoustic emissions during wire and arc additive manufacturing of 316l stainless steel,” additive manufacturing, vol. 51, article no. 102585, march 2022. [6] s. li, j. y. li, z. w. jiang, y. cheng, y. z. li, s. tang, et al., “controlling the columnar-to-equiaxed transition during directed energy deposition of inconel 625,” additive manufacturing, vol. 57, article no. 102958, september 2022. [7] t. a. rodrigues, n. bairrão, f. w. c. farias, a. shamsolhodaei, j. shen, n. zhou, et al., “steel-copper functionally graded material produced by twin-wire and arc additive manufacturing (t-waam),” materials & design, vol. 213, article no. 110270, january 2022. [8] v. t. le, d. s. mai, m. c. bui, k. wasmer, v. a. nguyen, d. m. dinh, et al., “influences of the process parameter and thermal cycles on the quality of 308l stainless steel walls produced by additive manufacturing utilizing an arc welding source,” welding in the world, vol. 66, no. 8, pp. 1565-1580, august 2022. [9] d. jafari, t. h. j. vaneker, and i. gibson, “wire and arc additive manufacturing: opportunities and challenges to control the quality and accuracy of manufactured parts,” materials & design, vol. 202, article no. 109471, april 2021. [10] s. w. williams, f. martina, a. c. addison, j. ding, g. pardal, and p. colegrove, “wire + arc additive manufacturing,” materials science and technology, vol. 32, no. 7, pp. 641-647, 2016. [11] w. e. frazier, “metal additive manufacturing: a review,” journal of materials engineering and performance, vol. 23, no. 6, pp. 1917-1928, june 2014. [12] j. xiong, y. li, r. li, and z. yin, “influences of process parameters on surface roughness of multi-layer single-pass thin-walled parts in gmaw-based additive manufacturing,” journal of materials processing technology, vol. 252, pp. 128-136, february 2018. [13] v. t. le, “a preliminary study on gas metal arc welding-based additive manufacturing of metal parts,” vnuhcm journal of science and technology development, vol. 23, no. 1, pp. 422-429, february 2020. [14] v. t. le, q. h. hoang, v. c. tran, d. s. mai, d. m. dinh, and t. k. doan, “effects of welding current on the shape and microstructure formation of thin-walled low-carbon parts built by wire arc additive manufacturing,” vietnam journal of science and technology, vol. 58, no. 4, pp. 461-472, july 2020. [15] w. jin, c. zhang, s. jin, y. tian, d. wellmann, and w. liu, “wire arc additive manufacturing of stainless steels: a review,” applied sciences, vol. 10, no. 5, article no. 1563, march 2020. [16] t. a. rodrigues, v. duarte, j. a. avila, t. g. santos, r. m. miranda, and j. p. oliveira, “wire and arc additive manufacturing of hsla steel: effect of thermal cycles on microstructure and mechanical properties,” additive manufacturing, vol. 27, pp. 440-450, may 2019. [17] j. g. lopes, c. m. machado, v. r. duarte, t. a. rodrigues, t. g. santos, and j. p. oliveira, “effect of milling parameters on hsla steel parts produced by wire and arc additive manufacturing (waam),” journal of manufacturing processes, vol. 59, pp. 739-749, november 2020. [18] a. v. nemani, m. ghaffari, and a. nasiri, “comparison of microstructural characteristics and mechanical properties of shipbuilding steel plates fabricated by conventional rolling versus wire arc additive manufacturing,” additive manufacturing, vol. 32, article no. 101086, march 2020. [19] p. dirisu, s. ganguly, a. mehmanparast, f. martina, and s. williams, “analysis of fracture toughness properties of wire + arc additive manufactured high strength low alloy structural steel components,” materials science and engineering: a, vol. 765, article no. 138285, september 2019. [20] l. sun, f. jiang, r. huang, d. yuan, c. guo, and j. wang, “anisotropic mechanical properties and deformation behavior of low-carbon high-strength steel component fabricated by wire and arc additive manufacturing,” materials science and engineering: a, vol. 787, article no. 139514, june 2020. [21] q. fang, l. zhao, b. liu, c. chen, y. peng, z. tian, et al., “microstructure and mechanical properties of 800-mpa-class high-strength low-alloy steel part fabricated by wire arc additive manufacturing,” journal of materials engineering and performance, 2022. https://doi.org/10.1007/s11665-022-06784-7 advances in technology innovation, vol. 8, no. 1, 2023, pp. 01-11 11 [22] v. t. le, d. s. mai, and q. h. hoang, “a study on wire and arc additive manufacturing of low-carbon steel components: process stability, microstructural and mechanical properties,” journal of the brazilian society of mechanical sciences and engineering, vol. 42, no. 9, article no. 480, september 2020. [23] v. t. le, d. s. mai, t. k. doan, and q. h. hoang, “prediction of welding bead geometry for wire arc additive manufacturing of ss308l walls using response surface methodology,” transport and communications science journal, vol. 71, no. 4, pp. 431-443, may 2020. [24] z. wang, s. zimmer-chevret, f. léonard, and g. abba, “prediction of bead geometry with consideration of interlayer temperature effect for cmt-based wire-arc additive manufacturing,” welding in the world, vol. 65, no. 12, pp. 2255-2266, december 2021. [25] j. xiong, g. zhang, j. hu, and l. wu, “bead geometry prediction for robotic gmaw-based rapid manufacturing through a neural network and a second-order regression analysis,” journal of intelligent manufacturing, vol. 25, no. 1, pp. 157-163, february 2014. [26] v. c. nguyen, t. d. nguyen, and v. t. le, “optimization of sustainable milling of skd11 steel under minimum quantity lubrication,” proceedings of the institution of mechanical engineers, part e: journal of process mechanical engineering, 2022. https://doi.org/10.1177/09544089221110978 [27] g. c. m. patel and jagadish, “experimental modeling and optimization of surface quality and thrust forces in drilling of high-strength al 7075 alloy: critic and meta-heuristic algorithms,” journal of the brazilian society of mechanical sciences and engineering, vol. 43, no. 5, article no. 244, may 2021. [28] d. t. sarathchandra, m. j. davidson, and g. visvanathan, “parameters effect on ss304 beads deposited by wire arc additive manufacturing,” materials and manufacturing processes, vol. 35, no. 7, pp. 852-858, 2020. [29] d. s. nagesh and g. l. datta, “prediction of weld bead geometry and penetration in shielded metal-arc welding using artificial neural networks,” journal of materials processing technology, vol. 123, no. 2, pp. 303-312, april 2002. [30] s. jindal, r. chhibber, and n. p. mehta, “effect of welding parameters on bead profile, microhardness and h2 content in submerged arc welding of high-strength low-alloy steel,” proceedings of the institution of mechanical engineers, part b: journal of engineering manufacture, vol. 228, no. 1, pp. 82-94, january 2014. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation, vol. 4, no. 1, 2019, pp. 11 20 a controlled fermentation environment for producing quality soya sauce sophia ferng 1 , wei-hua huang 1 , chien-ping wu 2 , yung-tsong lu 2 , cheng-kuang hsu 1 , robin yih-yuan chiou 1 , ching-hua ting 2,* 1 department of food science, national chiayi university, chiayi 600, taiwan, roc 2 department of mechanical and energy engineering, national chiayi university, chiayi 600, taiwan, roc received 04 may 2018; received in revised form 10 july 2018; accepted 15 august 2018 abstract soy sauce fermentation under controlled temperature is a way to shorten the fermentation time. an energy-saving fermentation system was developed to power a heat pump for maintaining the temperature of sauce moromi at 37±1°c during fermentation. the chemical properties of the sauce moromi and the sensory properties of the soy sauce produced using the controlled fermentation system were evaluated and compared to those of the sauce moromi fermented outdoors without temperature control. the sauce moromi processed using the controlled fermentation system had significantly higher total nitrogen, formal nitrogen, amino nitrogen, reducing sugar and organic acid contents than the moromi fermented outdoor. however, no significant difference was found in overall liking score between two soy sauces. the soy sauce fermented under the control temperature showed higher brix and salt concentration, but lower ph value than the sauce fermented outdoor. it was possible that the beneficial effects of reducing sugar and organic acid contents were rebuffed by the disadvantage of salt concentration. it was concluded that a controlled fermentation environment deserves the potential to produce a high quality soy sauce. keywords: soy sauce, moromi fermentation, quality promotion, controlled environment 1. introduction soy sauce has been the most important seasoning in the orient [1]. it is prepared from a mixture of steam-cooked soybean and roasted wheat with koji mold aspergillus oryzae or a. solace. in sauce moromi fermentation, the proteins and polysaccharides in the moromi are hydrolyzed by the fungal protease and amylase to form different types of small nitrogen compounds and sugars, such as amino acids and reducing sugars. the maillard reaction between amino acids and reducing sugars during moromi fermentation results in the unique flavor of soy sauce [2, 3]. cooking raw soy sauce with sugar improves the maillard reaction and further enriches the sensory properties of the soy sauce. the fermentation of soy sauce using the traditional method generally takes from six months to one year [4, 5]. the fermentation time can be shortened if the fermentation temperature is control artificially. young and wood (1974) reported that if the fermentation temperature was controlled at 35-40°c, it took only 3-4 months to ferment the sauce moromi. reference [2] stated that 45°c was an important temperature for sauce moromi fermentation. soy sauce making under controlled temperature required energy input. hence under the consideration of energy saving, we desii was fermented in a pottery tank to preserve the flavor of the sauce, (2) heat pump was 3-4 times more efficient in its use of elecgned an energy-saving fermentation system to transfer solar energy to electric power for a heat pump and used hot water to maintain the temperature of sauce moromi at 37±1°c during fermentation. the advantages of this system were: (1) the moromtric power than a simple electrical resistance heater, and (3) heat transfer was more efficient by using water as the medium. * corresponding author. e-mail address: cting@mail.ncyu.edu.tw advances in technology innovation, vol. 4, no. 1, 2019, pp. 11 20 12 the changes of chemical properties of the sauce moromi in the solar energy system were determined and the sensory properties of the soy sauce made from the system were also assessed to evaluate the potential of suing solar energy to produce a high quality soy sauce. 2. materials and methods 2.1. raw materials and chemicals soybean and wheat were purchased from a local farmer in chiayi city, taiwan and immediately stored in a refrigerator at 4°c prior to use. the koji mold (aspergillus oryzae) was purchased from a supplier (chuan feng microbe co., taichung city, taiwan). all of the chemicals used in the analysis were of analytical grade. 2.2. koji fermentation for koji production, raw soybeans were first washed and soaked in water for 24 h at 4°c and then steamed at 121°c for 20 min. wheat was roasted for 30 min at 140°c and then cracked into 4 or 5 pieces per kernel accompanied by smaller particles of wheat flour. roasted wheat and steamed soybeans were mixed at a ratio of 1:2 (w/w). the mixture was then inoculated with 0.1% (w/w) of a. oryzae (approximately 10 8 spores g -1 ), and subsequently dispersed onto perforated stainless steel trays (68 x 44 x 3 cm). each tray was loaded with the mixture to around 3 cm thickness and incubated at 30 ± 3°c. the preparation of koji was completed in 60–72 h when the culture began turning into greenish yellow in color. the koji was stored at 4°c until use. 2.3. moromi fermentation fig. 1 setup of the energy-saving soy sauce fermentation system. fig. 1 illustrates the structure of the temperature-controlled fermentation system. a 160 l pottery tank that accommodates soy mash is bathed in a stainless tank (diameter 84 cm and height 65 cm) full of circulating water for temperature control of the mash. temperatures inside and outside of the pottery tank are to be measured for on-line fermentation temperature control and off-line analysis. the hot, circulating water was prepared using a heat pump powered by a solar power generator and energy management system developed in house. the energy system trades with the electricity grid to assure a safe and yet economical power supply for the process. hot water is to be prepared in the hottest hours of a day for the interest of optimum heat pump efficiency. a programmable logic controller (plc) performs controls and an industrial personal computer (ipc) executes system monitoring and data acquisition. finished koji was put in the pottery tank and then mixed with 18% saline at a weight ratio of 1: 2. a stainless steel cover was made for the mash tank with one hole in the center to the insert thermocouple for measuring the central temperature of the mash, another annulus cover was used to prevent the evaporation of water out of the stainless steel tank. the system was instructed to maintain the central temperature of the mash at 37c+1°c. during the first week of fermentation, the mash was stirred for 3 min once every day and then once a week for the remaining period. the total fermentation period was 13 weeks. advances in technology innovation, vol. 4, no. 1, 2019, pp. 11 20 13 2.4. microbe analytical methods determination of microorganism changes: the 10-fold-concentrated cells were serially diluted (10 0 to 10 -6 ) in 0.85% sterile saline solution, and 100 µ l of properly diluted samples were spread by pca agar media and incubated at 30~35°c for 1~2 days for the total viable count. diluted sample (100 µ l) were spread by ym agar plate media (bd difco, franklin lakes, nj, usa), and then incubated for 1 day at 30°c for the counting of yeast. whereas mrs agar plate media (bd difco, franklin lakes, nj, usa) for the counting of lactic acid bacteria, which were incubated for 2–3 days at 30°c in an anaerobic condition (anaeropack, mitusubishi kagaku, kogyo, tokyo, japan). the colonies formed (25–250) were calculated as log cfu per milliliter of soy sauce [7]. 2.5. chemical analytical methods determination of total nitrogen: total nitrogen was determined according to the aoac standard [8] with a little modification. three milliliters of the sample, 1g of catalyst (k2so4/cuso4.5h2o = 10:1), and 20 ml of concentrated h2so4 were added into a digestion tube and then heated for 6 h at 450°c for digestion. when the digestion was completed, 25 ml of diluted sample was mixed with 25 ml of 30% naoh and subjected to distillation. the distillate was absorbed with 25 ml of 0.1 n h2so4. the content of total nitrogen was determined by back-titrating h2so4 with 0.1 n naoh. determination of formal nitrogen and ammonium nitrogen: formal nitrogen and ammonium nitrogen were determined according to the chinese national standard [9]. to determine formal nitrogen, diluted sample and formalin solution were adjusted to ph 8.1, then 20 ml of formalin solution was mixed with 25 ml of diluted samples for 3 min, and titrated to ph 8.1 with 0.01 n naoh. amino nitrogen was determined by subtracting ammonium nitrogen from formal nitrogen according to the cns [9]. determination of ph: the ph values of the soy sauce samples were measured with a pb-10 ph meter (sartorius, göttingen, germany). determination of organic acid content: organic acid content as lactic acid was determined by alkali titration. to determine total acidity as lactic acid, 25 ml of properly diluted sample was titrated with 0.1 n naoh to ph 8.1. determination of reducing sugar: the reducing sugar (rs) in the samples was determined by the 3,5-dinitrosalicylic acid (dns) method [10] with a little modification. dns reagent, which contained (w/v) 1% dns, 2% naoh and 30% sodium potassium tartrate, was prepared by dissolving each component in distilled water. one milliliter of the sample diluents were mixed with 1 ml of dns reagent and then kept in a boiling water bath for 5 min to colorize. after cooling with cold water, 5 ml of distilled water was added and the od540 nm was measured by using distilled water as a blank control. the standard glucose solutions were used to make a calibrated curve for quantitative analysis. determination of browning: the level of browning was determined using the method of reference [11]. the absorbance of moromi samples was measured at od555 nm. determination of sodium chloride concentration: sodium chloride content was determined by volumetric titration with agno3 using the mohr’s method (hamilton & simpson, 1964). a 1 ml of 0.25 m k2cro4 was added into 250 ml flask containing 10 ml of properly diluted sample, then titrated with 0.01 n agno3 until the appearance of brownish color. the standard curve of nacl was prepared with 10–25% (w/v) of nacl solutions. 2.6. sensory evaluation there was a total of 4 different soy sauces under sensory evaluation. the moromi samples from temperature-controlled brewing mash (tcm) and traditional brewing mash (tbm) fermented for 3 months were pressed, and then cooked with sugar (sauce moromi: sugar = 100 g: 10 g) and water (sauce moromi : water =100 g: 40 g) till the added water evaporated out. after cooling down the mixture, ethanol (sauce moromi : ethanol = 100 g: 1 g) was added, filtered with cheese cloth and clarified the advances in technology innovation, vol. 4, no. 1, 2019, pp. 11 20 14 mixture additionally by sedimentation. after bottling, the soy sauces were autoclaved at 85°c for 30 min. these soy sauces were labeled as samples of tcm-3 and tbm-3. a soy sauce made from tcm without the addition of sugar during cooking and without the addition of ethanol was labeled “tcm-3-raw” as a raw soy sauce. finally, a commercial soy sauce was served as a standard soy sauce which was selected from a very popular brand in taiwan. the color, flavor, taste, savor and overall palatability of soy sauce samples was evaluated by a panel of 60 untrained volunteers. soy sauce samples (10 ml) were prepared in small white plastic discs bearing 3-digit random numbers and presented them to the panelists in random order. the program started with observation of the color, smell of the flavor and taste of the relish by eating some cool noodle with the samples then scored their preferences based on 9-point hedonic scale. 2.7. statistical analysis the data were analyzed by analysis of variance (anova), and significant differences in mean values among data were determined at p < 0.05 by duncan’s multiple-range tests using spss 20 (spss inc., chicago, il). 3. results and discussion 3.1. the changes of microbe counts in the mash during moromi fermentation (a) viable bacteria (b) yeast (c) lactic acid bacteria fig. 2 changes in total (a), (b), and (c) of traditional brewing mash (tbm) and temperature control brewing mash (tcm) during moromi fermentation. ○: tbm; ●: tcm. data are presented as mean±standard deviation (n=3). *,** a significant difference between tbm and tcm at the level of p<0.05 and p<0.01, respectively. the unique flavor of moromi is attributed to lactic acid bacteria in the first stage of moromi fermentation, resulting in decreased ph in moromi to 6.0 or below, at which yeast starts to propagate [13]. fig.1 shows the changes of total viable bacteria, yeast, and lactic acid bacteria of traditional brewing mash (tbm) and temperature control brewing mash (t cm) during moromi fermentation. the total viable bacteria and yeast counts in both tbm and tcm went down and then went up, and then decreased gradually with fermentation time (fig. 2a, fig. 2b). the lowest total viable bacteria and yeast counts in tbm occurred at 5 weeks, while tcm had the lowest counts at 3 weeks. similar changes in total viable bacteria and yeast counts in black soybean moromi were reported [14]. the decrease of total viable bacteria and yeast counts in the initial stage of moromi fermentation was due to the inhibitory effects of high salt concentration in the moromi and low ph induced by lactic acid bacteria [15]. advances in technology innovation, vol. 4, no. 1, 2019, pp. 11 20 15 no significant changes in the count of lactic acid bacteria in both tbm and tcm were found from 1 week to 9 weeks, and then the count decreased slightly by fermentation time (fig. 2c). in general, there were no significant differences in the yeast and lactic acid bacteria counts between tbm and tcm. the results indicated that the moromi fermentation under control temperature at 37°c did not influence the growths of yeast and lactic acid bacteria compared to the traditional fermentation method that took place at outdoor without temperature control. 3.2. the changes of chemical properties of the mash during moromi fermentation during moromi fermentation process, the proteins of soybean and wheat were hydrolyzed by protease of microbes into small fragments of peptides, amino acids and ammonia. therefore, total nitrogen, formal nitrogen and amino nitrogen contents were often used as quality indexes for sauce moromi [14, 16]. changes in total nitrogen, formal nitrogen, and amino nitrogen contents of tbm and tcm during moromi fermentation are shown in fig. 3a it was found that total nitrogen content increased during moromi fermentation in both tcm and tbm, and their marked increases were noted within 3 weeks fermentation (fig. 3a). tcm showing a higher increase rate compared to tbm indicates that fermentation at 37°c could promote the protease activities and enhance the rate of protein hydrolysis than traditional fermentation method. both formal nitrogen and amino nitrogen contents in tcm increased almost linear with fermentation time up to 7 weeks and then leveled off as the fermentation process continue (fig. 3b, fig. 3c). like the total nitrogen content, both formal nitrogen and amino nitrogen contents were higher in tcm than those in tbm. the results suggested that moromi fermentation at the elevated controlled temperature (37°c) would promote the proteases activities to hydrolyze soybean and wheat proteins into different types of small nitrogen compounds. our data agreed with reference [2], where they stated that high temperature, peculiarly the temperature of 45°c, was a very useful method to promote the proteases in the koji to solubilize and hydrolyze soy proteins efficiently. reference [17] reported that black soybean sauce fermented at 50°c had higher total nitrogen content but lower formal nitrogen content compared to the sauce fermented at 30°c, outdoor or indoor. however, reference [18] compared moromi fermentation at 25, 35, 45 and room temperature, and found that fermentation temperature did not affect total nitrogen content. (a) total nitrogen (b) formal nitrogen (c) amino nitrogen fig. 3 changes in (a), (b), and (c) of traditional brewing mash (tbm) andtemperature control brewing mash (tcm) during moromi fermentation. ○: tbm, ●: tcm. data are presented as mean±standard deviation (n=3). advances in technology innovation, vol. 4, no. 1, 2019, pp. 11 20 16 fig. 4 shows the changes of ph and organic acid in tbm and tcm. the ph values decreased gradually in both tbm and tcm (fig. 4a). the rate of decrease in tcm was higher than that in tbm. the mash with a low ph value had the advantage of preventing the contamination of unwanted microbes and maintaining the condition well-suited for the action of yeast. in general, the accumulation of free fatty acids, amino acids, and peptides containing carbolylic side chains as the results of hydrolysis of raw materials, the autolysis of microbial cells, as well as the microbial fermentation of carbohydrates, which may account for the decline of ph during moromi fermentation [3, 11, 19]. the accumulation of organic acid in tcm was greater than that in tbm, and the results agreed with more rapid decrease of ph in tcm than in tbm (fig. 4b). (a) ph (b) organic acid fig. 4 changes in (a) and (b) of traditional brewing mash (tbm) and temperature control brewing mash (tcm) during moromi fermentation. ○: tbm, ●: tcm. data are presented as mean ± standard deviation (n=3). fig. 5 changes in reducing sugar content of traditional brewing mash (tbm) and temperature control brewing mash (tcm) during moromi fermentation. ○: tbm, ●: tcm. data are presented as mean ± standard deviation (n=3). fig. 6 change in browning of traditional brewing mash (tbm) and temperature control brewing mash (tcm) during moromi fermentation.the higher the od555 nm value, the higher the browning level. ○: tbm, ●: tcm. data are presented as mean ± standard deviation (n=3). during moromi fermentation, amylases from the microbes hydrolyze the starch in soybean and wheat moromi to form advances in technology innovation, vol. 4, no. 1, 2019, pp. 11 20 17 small sugars [2, 17]. therefore, reducing sugar can be used as a quality indicator for sauce moromi. reducing sugar content is also a very important attribute regarding to the sensory properties of soy sauce. it can react with amino acids to form maillard’s compounds and produce unique flavor of soy sauce. the reducing sugar content in tcm increased almost linearly during fermentation; however, the content in tbm increased up to 7 weeks and then leveled off (fig. 5). tcm had higher reducing sugar content than tbm; therefore, tcm also showed greater browning reaction than tbm after 3 weeks fermentation (fig. 6). our data showed that tcm had higher total nitrogen, formal nitrogen, amino nitrogen, organic acid and reducing sugar contents than tbm. this was because the proteases and amylases showed higher activity at the elevated controlled temperature (37°c) and hydrolyzed the proteins and starch in soybean and wheat moromi to form different types of small nitrogen compounds and sugars. and based on these chemical indexes, it was clear that tcm had higher quality properties than tbm. 3.3. sensory properties of tcm and tbm soy sauces the sensory scores of four different soy sauces are shown in table 1. the color, flavor, texture, savor and overall liking scores in the tested soy sauces showed very high agreement; therefore, the overall liking score alone could be used to describe the overall sensory preference of the tested soy sauces. the overall liking score could be classified into three groups: the commercial soy sauce showing the highest overall liking scores, followed by tbm-3 and tcm-3, and tcm-3-raw. the commercial product is a very popular blend in taiwan, and the formulation may include certain ingredients specifically for improving the flavor properties. therefore, it was not a surprise to see that the commercial product received the highest score in all the sensory properties. for other soy sauces, only additional 10% sugar was added before cooking and then 1% ethanol was added after cooking, while only water was added during the cooking stage and no ethanol was added for tcm-3-raw. table 1. hedonic rating scores by sensory evaluation sample colour flavour texture savour overall tbm-3 5.72 b 5.32 b 5.25 b 5.29 b 5.31 b tcm-3-raw 5.49 b 4.58 c 4.55 c 4.71 bc 4.63 c tcm-3 5.65 b 5.45 b 5.22 b 5.29 b 5.43 b commercial 6.45 a 6.46 a 6.51 a 6.35 a 6.57 a a, b, c, different letters indicate significant difference in a column (p<0.05). table 2. correlation coefficients between chemical properties and the overall sensory score total nitrogen formal nitrogen amino nitrogen salinity brix reducing sugar ph organic acid overall favourite total nitrogen 1 formal nitrogen 0.09 1 amino nitrogen 0.09 1.00 1 salinity -0.26 0.84 0.84 1 brix 0.49 -0.51 -0.49 -0.43 1 reducing sugar 0.31 -0.90 -0.90 -0.81 0.76 1 ph -0.15 0.61 0.64 0.30 -0.59 -0.78 1 organic acid 0.67 -0.55 -0.56 -0.56 0.75 0.86 -0.81 1 overall favourite 0.32 -0.91 -0.91 -0.86 0.74 0.99 -0.71 0.82 1 tcm-3 and tbm-3 were produced from 3 month-fermented tcm and tbm, respectively, with the addition of 10% sugar to the sauce prior to the cooking process and then 1% ethanol were added after cooling down. based on the results obtained from the moromi fermentation, it was clear that tcm showed better chemical properties than tbm; however, the overall liking scores between tcm-3 and tbm-3 did not differ significantly. it was possible that different chemical properties contributed differently to human mouth’s perception, and the contributions from certain chemical properties were evened by other chemical properties. it was also possible that the addition of sugar prior to cooking and also the addition of ethanol concealed the differences in the chemical properties and resulted in the same level of the overall liking score. advances in technology innovation, vol. 4, no. 1, 2019, pp. 11 20 18 tcm-3-raw was produced from 3 month-fermented tcm with the addition of only water during cooking. without extra sugar to promote the maillard reaction, tcm-3-raw received a lower overall liking score than tcm-3 the chemical properties of four soy sauces shown in table 2, and the correlation coefficients of the chemical properties and the overall liking scores are shown in table 3. compared to the chemical properties of other soy sauces, the commercial soy sauce had lower formal nitrogen, amino nitrogen and salt concentration but high brix, reducing sugar and organic acid contents. this agreed with the correlation coefficients between the overall liking score and chemical properties, significant negative coefficients for formal nitrogen content (r = -0.88), amino nitrogen content (r = -0.87) and salt concentration (r = -0.88), but significant positive coefficients for brix (r = 0.81), reducing sugar content (r = 0.996) and organic acid content (r = 0.76). tcm-3 had a higher salt concentration, reducing sugar and organic acid contents than tbm-3; however, there was no significant difference in all the sensory scores. it was possible that the beneficial effects of reducing sugar and organic acid contents were rebuffed by the disadvantage of high salt concentration. with the addition of sugar and ethanol, tcm-3 showed higher brix, reducing sugar and organic acid contents than tcm-3-raw. thus, tcm-3 was expected to gain higher sensory scores than tcm-3-raw. during soy sauce making, it is a general practice to add sugar during cooking to promote the maillard reaction and to enhance the flavor properties of soy sauce. it is well know that free amino acids are the main sources contributing to the flavor of soy sauce [20]. free amino acid contents in the four different soy sauces were determined using hplc and the results are shown in fig. 7. in all soy sauces, glu had the highest content (about 14-20%) followed by asp and leu (fig. 7a). the free amino acid profile was similar to the data reported reported data [21]. it was noted that the commercial soy sauce had the lowest glu, asp and leu, while tcm-3-raw showed the highest contents. the sum of 17 measured amino acid content was calculated and the data indicated that the commercial soy sauce had the lowest total amino acid content, while tcm-3-raw had high total amino acid content. (a) total free amino acid contents (b) individual free amino acid contents fig. 7 (a) and (b) in five different soy sauces. tbm-3: tbm at 3 months; tcm-3: tcm at 3 months; tcm-3-raw: raw soy sauce made from tcm at 3 months without the addition of sugar and ethanol; tcm-3: tcm at 3 months; commercial soy sauce: a very popular soy sauce in taiwan. data are presented as mean + standard deviation (n=2). total nitrogen, formal nitrogen and amino nitrogen contents are often served as quality indexes of soy sauces and the decision making in the selection of ending time for the moromi fermentation process. during fermentation, amino acid reacted by reducing sugar to form maillard compounds, such as pentosidine, carboxymethyllysine and furosine, and these compounds further enriched the flavor of soy sauce [22, 23]. in this study, we found a strong correlation (r = 0.996) between the overall advances in technology innovation, vol. 4, no. 1, 2019, pp. 11 20 19 liking score of soy sauces and reducing sugar; however, the correlation was insignificant for total nitrogen content (table 3). nevertheless, even negative correlation coefficient was found between the overall liking score and formal nitrogen content (r = -0.88), as well as amino nitrogen content (r = -0.87). moreover, the total free amino acid content and individual glu, asp and leu all correlated negatively with the overall liking score of the soy sauces. the question need to be addressed was why formal nitrogen and amino nitrogen contents are considered as quality indexes for soy sauce, but these contents in the tested soy sauce were negative correlated with the sensory scores. it was noted that all 17 measured free amino acid contents in tcm-3 were lower than tcm-3-raw indicating that it must have a significant factor resulted in such differences in all different types of free amino acids. since the difference in tcm-3 and tcm-3-raw was with or without the addition of extra sugar during cooking to promote the maillard reaction, it was suggested that free amino acids in tcm-3 interacted with sugars to form maillard compounds and thus the levels of measurable free amino acids decreased in tcm-3 but it also resulted in better sensory properties in tcm-3. based on the results, it was concluded that the maillard reaction took place during the cooking process is a critical attribute to sensory properties of soy sauces. 4. conclusions under the considerations of energy saving and hygienic improving, a solar energy system was designed for soy sauce making. electricity converted from the solar energy was used to power a heat pump that supplies hot circulating water for fermentation temperature control. hot water was prepared at noon and then stored and circulated (normally at night time) for temperature control. this scheme assures an energy-saving fermentation process. the sauce moromi fermented at the system showed better chemical properties than the traditional brewing moromi. therefore, the use of solar energy system has the potential to produce a high quality soy sauce under an energy-saving stance. acknowledgement this work was sponsored by the ministry of science and technology of the roc (taiwan) government under the grant most 103-2221-e-415 -016. references [1] d. fukushima, “soy proteins for foods centering around soy sauce and tofu,” journal of the american oil chemists’ society, vol. 58, no. 3, pp. 346-354, march 1981. [2] n. w. su, m. l. wang, k. f. kwok, and m. h. lee, “effects of temperature and sodium chloride concentration on the activities of proteases and amylases in soy sauce koji,” journal of agricultural and food chemistry, vol. 53, no. 5, pp. 1521-1525, february 2005. [3] t. yokotsuka, “soy sauce biochemistry,” advances in food research, vol. 30, pp. 195-329, 1986. [4] j. s. lee, s. j. rho, y. w. kim, k. w. lee, and h. g. lee, “evaluation of biological activities of the short-term fermented soybean extract,” food science and biotechnology, vol. 22, no. 4, pp. 973-978, august 2013. [5] y. zhu and j. tramper, “koji where east meets west in fermentation,” biotechnology advances, vol. 31, no. 8, pp. 1448-1457, december 2013. [6] f. m. yong and b. j. b. wood, “microbiology and biochemistry of soy sauce fermentation,” advances in applied microbiology, vol. 17, pp. 157-194, 1974. [7] m. h. kang, j. s. kim, s. j. park, and t.shibamoto, “a study on biological and chemical changes in fermented soybeans, grown under organic vs non-organic conditions, during storage,” journal of agriculture and biological sciences, vol. 3, no. 2, pp. 278-286, march 2012. [8] official methods of analysis, 17th ed., maryland: association of official analytical chemists, 2000. [9] chinese national standard (cns), taiwan, 2002. [10] g. l. miller, “use of dinitrosalicylic acid reagent for determination of reducing sugar,” analytical chemistry, vol 31, no. 3, pp. 426-428, march 1959. advances in technology innovation, vol. 4, no. 1, 2019, pp. 11 20 20 [11] c. c. chou and m. y. ling, “biochemical changes in soy sauce prepared with extruded and traditional raw materials,” food research international, vol. 31, no. 6-7, pp. 487-492, august 1998. [12] l. f. hamilton and s. g. simpson, “precipitation methods,” quantitative chemical analysis, macmillan, new york, 1964. [13] r. y. cui, j. zheng, c. d. wu, and r. q. zhou, “effect of different halophilic microbial fermentation patterns on the volatile compound profiles and sensory properties of sauce moromi,” european food research and technology, vol. 239, no. 2, pp. 321-331, august 2014. [14] z. y. chen, y. z. feng, c. cui, h. f. zhao, and m. m. zhao, “effects of koji-making with mixed strains on physicochemical and sensory properties of chinese-type soy sauce,” journal of the science of food and agriculture, vol. 95, no. 10, pp. 2145-2154, august 2015. [15] y. z. yan, y. l. qian, f. d. ji, j. y. chen, b. z. han, “microbial composition during chinese soy sauce koji-making based on culture dependent and independent methods,” food microbiology, vol. 34, no. 1, pp. 189-195, may 2013. [16] j. feng, x. b. zhan, z. y. zheng, d. wang, l. m. zhang, and c. c. lin, “a two-step inoculation of candida etchellsii to enhance soy sauce flavour and quality,” international journal of food science and technology, vol. 47, no. 10, pp. 2072-2078, october 2012. [17] t. r. chen and r. y. y. chiou, “microbial and compositional changes of inyu (black soybean sauce) broth during fermentation under various temperature conditions,” journal of the chinese agricultural chemical society, vol. 34, no. 2, pp. 157-164, 1995. [18] t. y. wu, m. s. kan, l. f. siow, and l. k. palniandy, “effect of temperature on moromi fermentation of soy sauce with intermittent aeration,” african journal of biotechnology, vol. 9, no. 5, pp. 702-706, 2010. [19] c. van der sluis, j. tramper, and r. h. wijffels, “enhancing and accelerating flavour formation by salt-tolerant yeasts in japanese soy-sauce processes,” trends in food science and technology, vol. 12, no. 9, pp. 322-327, september 2001. [20] x. gao, c. cui, j. ren, h. zhao, q. zhao, and m. zhao, “changes in the chemical composition of traditional chinese-type soy sauce at different stages of manufacture and its relation to taste,” international journal of food science and technology, vol. 46, no. 2, pp. 243-249, january 2011. [21] x. gao, p. sun, j. lu, and z. jin, “characterization and formation mechanism of proteins in the secondary precipitate of soy sauce,” european food research and technology, vol. 237, no. 4, pp. 647-654, october 2013. [22] p. c. chao, c. c. hsu, and m. m. yin, “analysis of glycative products in sauces and sauce-treated foods,” food chemistry, vol. 113, no. 1, pp. 262-266, 2009. [23] p. c. chao, c. c. hsu, and m. m. yin, “analysis of glycative products in sauces and sauce-treated foods,” food chemistry, vol. 113, no. 1, pp. 262-266, 2009. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). microsoft word 5-v8n3(2023)-aiti#11508(219-228).docx advances in technology innovation, vol. 8, no. 3, 2023, pp. 219-228 english language proofreader: jing-heng yen assessing the performance of melted plastic as a replacement for sand in paving block noor a’fiana desyani*, arief sabdo yuwono, heriansyah putra department of civil and environmental engineering, ipb university, bogor, indonesia received 06 february 2023; received in revised form 06 may 2023; accepted 10 may 2023 doi: https://doi.org/10.46604/aiti.2023.11508 abstract plastic waste generates numerous environmental problems, such as garbage accumulation and plastic waste pollution in the oceans. this study aims to evaluate the effectiveness of melted plastic waste as a substitute material in paving blocks. the melted low-density polyethylene (ldpe) plastic is used as the cemented agent in the paving block. after melting, the melted ldpe plastic is mixed thoroughly with sand immediately and forms a paving block mold. the effectiveness of melted plastic as a bonding agent is evaluated based on the parameters of compressive strength, water absorption, and wear resistance. the results show that paving blocks with a melted plastic of 10% reach the required level of 9.39 mpa for the park. hence, using melted plastic in paving blocks can be an alternative strategy to reduce plastic waste. keywords: compressive strength, composite paving block, plastic melter, water absorption, wear resistance 1. introduction one of the environmental issues is the increasing amount of plastic waste. according to the data from indonesian ministry of environment and forestry, plastic waste in indonesia saw a 19% increase in 2020 compared to 2019. the leading causes of the growing volume of plastic waste each year are urbanization and changes in lifestyle [1]. unlike organic waste which can be decomposed by bacteria, plastic waste takes a relatively long time, causing accumulation in the final processing site. other harmful effects of plastic waste include solid plastic waste pollution in the oceans, as well as the introduction of microplastics and nanoplastics [2]. moreover, plastic waste can cause an explosion if it comes into contact with coal waste in the ocean [3]. according to jambeck et al. [4], plastic waste is one of the solid wastes created in significant amounts. world plastic production reached 396 million tons annually in 2018 [5], equivalent to one megaton per day. this number is predicted to double over the next 10-20 years [6-7]. indonesia is in second position as the country that produces the most plastic waste, with the amount of plastic waste in the sea reaching 1.29 million tons annually [8]. in order to prevent additional pollution of the environment and the nearby living beings, plastic waste issues need to be treated seriously. applying the 3r principles, namely reduce, reuse, and recycle, in daily life can help lessen the adverse effects of plastic waste. recycling is the process of converting waste into reusable materials. the benefits of the recycling process include saving energy, reducing pollution, and mitigating land damage and greenhouse gas emissions. to reduce plastic production from new raw materials, it is essential to follow recycling after plastic consumption. singh et al. [9] stated that 90% of plastic waste can be recycled. however, 80% of plastic waste currently ends up in landfills. additionally, 8% is burned, and only 7% is recycled. stockpiled plastic waste emits only 2% of co2 emissions [2], while longterm effects include the need for extensive land areas, as well as soil and groundwater pollution. therefore, a solution is needed to overcome these problems, including recycling plastic waste. * corresponding author. e-mail address: afianadesyani@apps.ipb.ac.id advances in technology innovation, vol. 8, no. 3, 2023, pp. 219-228 220 heat quickly destroys plastic because it has a low boiling point. with such plastic properties, the melting process can be a simple alternative for plastic recycling in the first step. a plastic melter is a device used to melt raw plastic materials, and its design is kept as simple as possible to ensure easy operation. furthermore, a plastic melter is a tool that melts certain types of plastic waste and reshapes it into new products, such as plastic bricks and concrete blocks. in this study, a plastic melter was developed. the melted plastic was utilized as a substitute material for paving blocks. based on the waste generation study of ipb waste management 2018 [10], the type of plastic used was low-density polyethylene (ldpe). ldpe was the most significant fraction of plastic waste in the research location. this initiative aims to help reduce the environmental burden of plastic waste. in construction, melted plastic is used as a substitute for concrete or paving block aggregate. additionally, various forms of plastic are also used as a substitute for materials, such as plastic pellets, chopped plastic, and melted plastic. mozumder et al. [11] used plastic pellets as a substitute for sand with variations of 0%, 2%, 4%, 6%, 8%, and 10% by weight of sand. the highest compressive strength was achieved when adding 4% plastic, and the compressive strength decreased significantly when adding 8% and 10% plastic. sofyan et al. [12] performed sand substitution with plastic pellets with 10%, 20%, 30%, and 40% variations. the results showed that the compressive strength of concrete with plastic mixtures decreased by 56.06% compared to the control sample (0% plastic). in addition to plastic pellets, chopped plastic is also used as a substitute for aggregate in concrete. nurmaidah and pradana [13] used plastic fiber (chopped plastic) as a substitute for sand with variations of 0%, 3%, 6%, and 9% by weight of sand. with the same w/c value, adding up to 6% plastic can improve the compressive strength of concrete. however, the increase is not significantly different from the control sample (0% plastic). qaidi et al. [14] used chopped plastic as a substitute for sand with variations of 0%, 25%, and 50%. the test results showed that adding plastic decreases compressive strength. previous studies also used melted plastic as a substitute for paving blocks. agyeman et al. [15] researched paving blocks with melted plastic as an alternative binder. the results showed that paving blocks with a plastic mixture increase the compressive strength by approximately 40% more than paving block control at 21 days. sudarno et al. [16] created a paving block using combinations of melted plastic and gravel. the results showed that paving blocks with 50% melted plastic and 50% gravel composition had the highest compressive strength, namely 50.97 mpa. many research studies have shown that using melted plastic as a cement substitute enhances compressive strength. the potential of melted plastic as a sand substitute in paving blocks was investigated in this study. qaidi et al. [14] stated that the compressive strength of concrete with plastic mixtures tends to decrease due to the smooth surface of plastic, which results in a weaker bond. therefore, in this study, melted plastic was mixed thoroughly with sand to produce a stronger bond with the concrete. the plastic was melted to significantly improve the adhesion between the plastic and the cement matrix. the melted plastic was used at 5%, 10%, and 15% of the weight of the sand. 2. material and methods 2.1. material the materials used to make the plastic melter included black hollow iron measuring 3 × 3 cm and 4 × 4 cm, an iron tube with 10 inches diameter and 2 inches diameter, an iron plate with 1 mm thickness, and an analog bimetal thermometer with 3 inches diameter. the materials used to make paving blocks with a mixture of plastic were sand from cimangkok village (sukabumi regency, indonesia), portland composite cement (pcc), water, and plastic waste. the type of plastic waste was ldpe collected from margajaya urban village, bogor city, indonesia. advances in technology innovation, vol. 8, no. 3, 2023, pp. 219-228 221 2.2. design and construction of plastic melter the plastic melter has a table frame and a closed tubular melting chamber. the capacity of the plastic melting chamber is easily calculated by considering the following factors: the capacity of the melting device is determined at 12 kg, the specific gravity of the melted plastic is 0.94 g/cm3, the multiplier of 110% is used as additional space, and the multiplier of 2 is used as an assumption of the melting time. thus, the melting chamber capacity is 28 l. the melting chamber is determined as a closed tube with a diameter of 25.4 cm and a height of 55 cm. fig. 1 depicts the 3d design of the plastic melter created with sketchup software. (a) isometric view (b) top view fig. 1 3d view of plastic melter plastic melter must be heat-resistant or high-temperature materials to ensure proper melting of the plastic. iron and stainless steel are examples of high-temperature-resistant materials, which are durable and easy to clean. in this study, iron was used to make the plastic melter because it is generally more economical than stainless steel. in operation, the plastic melter is connected to a gas stove and equipped with a thermometer to monitor the melting temperature of the plastic. an output pipe measuring 2 inches with an open-close regulating valve is at the bottom of the melting chamber. based on preliminary research in this study, the plastic melted at a temperature of 180 ℃, as recorded on the top of the plastic melter. according to hasaya etal., plastic melts at 148 ℃ and has a boiling point of 234 ℃ [17]. the plastic melting process was carried out during the performance test to achieve the expected temperature. the performance test results demonstrated that the plastic melter performed well, reaching the planned plastic melting temperature. after heating for 90 minutes and reaching its melting temperature, the valve at the bottom of the melting chamber was opened. finally, the melted plastic can be used as a substitute for sand in paving blocks. 2.3. sample mixture and evaluation table 1 mix variations of paving block sample cement sand plastic water (g/cm3) control 0,34 1,36 0,00 0,17 5% plastic 0,34 1,29 0,07 0,17 10% plastic 0,34 1,22 0,14 0,17 15% plastic 0,34 1,16 0,20 0,17 paving blocks were made with 1 portland cement and 4 sand aggregates [18]. it is the best composition that produces optimal strength. the amount of water in the paving block was determined at 10% of the weight of the sand and cement [19]. advances in technology innovation, vol. 8, no. 3, 2023, pp. 219-228 222 water helps chemical reactions occur during the binding process in paving blocks. mix variations were made as shown in table 1. by varying the plastic mixture’s percentages to 5%, 10%, and 15% of the weight of sand, plastic melting can replace the need for sand. the procedures for making paving blocks with the addition of melted plastic can be seen in fig. 2. first, the sand is heated using a pan and stirred to distribute the heat evenly. melted plastic is combined with heated sand in a pan during the sample mixture process to prevent the hardening of plastic. after mixing well, the pan is moved from the stove and the cement is added. then, water is added after the melted plastic, heated sand, and cement are thoroughly mixed. the paving block dough is printed and pressed using a manual press. mixing until the pressing process should be completed quickly to prevent the plastic from hardening rapidly. fig. 2 procedures for making paving blocks with a plastic paving blocks were cured for 28 days following concrete curing standards on sni 03-0691-1996 [20]. the curing method was finished by immersing the paving block in water. after the curing period, the paving block evaluation was carried out on compressive strength, water absorption, and wear resistance, referring to the indonesian standard of sni 03-0691-1996 concerning concrete brick (paving block) [20]. compressive strength testing requires a 7.5× 7.5× 7.5 cm cube-shaped sample test, or the sample can be adjusted along its shortest side. based on sni 03-0691-1996, which refers to sni 03-1974-1990 concerning the concrete compressive strength test method, the sample test must be a cube to ensure that the strength is spread evenly across the three sides of the paving block [20]. the compressive strength test was carried out with three samples. the compressive strength test used the universal testing machine (utm). the value of the compressive strength of the paving block can be calculated by: compressive strength p a = (1) where p is the compressive load (n), and a is the sample area (mm2). the water absorption test on paving blocks was carried out by soaking them in water for 24 hours and then weighing them. after that, the paving blocks were dried in the oven for 24 hours at a temperature of ±105℃. water absorption of paving blocks can be obtained by: water absorption 100% m d d − = × (2) where m is the saturated mass (kg), and d is the dry mass (kg). advances in technology innovation, vol. 8, no. 3, 2023, pp. 219-228 223 the wear resistance test on paving blocks was carried out using a test object measuring 5 × 5 × 2 cm. the wear resistance test procedure refers to sni 03-0028 1987 concerning the test method for cement tiles. wear resistance testing on the paving block can be found by: wear resistance 1.26 0.0246g= + (3) where g is the mass loss (g/min). paving blocks must meet the indonesian standards as presented in table 2. in this study, paving blocks are planned to have d quality. related to the quality of paving blocks, sni 03-0691-1996 regulates the physical properties of paving blocks in each quality, as shown in table 2. table 2 physical properties of paving blocks (sni 03-0691-1996) quality compressive strenght (mpa) wear resistance (mm/min) water absoption (%) usage average min. average min. a 40 35 0.090 0.103 3 pavement b 20 17 0.130 0.149 6 parking areas c 15 12.5 0.160 0.184 8 pedestrian d 10 8.5 0.219 0.251 10 parks 3. result and discussion 3.1. properties test properties testing was carried out on the parameters of specific gravity, water absorption, organic content, silt content, and the distribution of grains of sand aggregate. the results of the sand properties test can be seen in table 3, while the distribution of sand aggregates can be seen in fig. 3. the organic test chart can determine the amount of organic sand. based on astm c40/c40m-11, sand can be used if the organic test results are below the third color on the test chart in fig. 4. according to the results of the properties test in table 3, it is known that the aggregate meets the standards of most properties test parameters so that it can be used. on melted plastic, properties tests were also performed. fig. 5 shows a plastic melter that is used to melt plastic waste. as shown in table 4, the specific gravity of melted plastic is 0.94 g/cm3 based on measurements. fig. 3 distribution of sand aggregate grains table 3 the result of the sand properties test parameter score standard standard specific gravity (ssd) (-) 1.95 1.6 3.2 sni 1970-2008 water absorption (%) 5.65 < 3 sni 1970-2008 organic content (-) 2 < 3 astm c40/c40m-11 sludge levels (%) 0.6 < 5 sk sni s-04-1989-f 0 10 20 30 40 50 60 70 80 90 100 0.01 0.1 1 10 p as si n g (% ) sieve size (mm) advances in technology innovation, vol. 8, no. 3, 2023, pp. 219-228 224 fig. 4 organic impurities color chart (astm c40/c40m-11) fig. 5 plastic melter table 4 specific gravity of the sand parameter water sand weight (g) 81 76 volume (cm³) 81 81 specific gravity (g/cm³) 1.00 0.94 3.2. density of paving blocks as shown in fig. 6, a comparison of the density of each sample tested is presented. the density of each paving block is observed to decrease as the number of plastic melts increases. it can be attributed to the lower specific gravity of the plastic compared to that of the sand, resulting in a decrease in the volume weight (density) [21]. moreover, the concrete’s pores in paving blocks with a plastic mixture can also lead to a reduction in density between particles. [22]. fig. 6 the density of the paving block 3.3. compressive strength fig. 7 depicts that paving blocks with plastic mixtures have various compressive strengths, namely 7.63-9.39 mpa. the result shows that paving blocks without plastic have higher compressive strength than those with a plastic mixture. according to sni 03-0691-1996, the compressive strength standard for d quality is 8.5-12.5 mpa [20]. the d quality is the lowest quality paving block and can be used for parks. in this study, paving blocks with a 10% plastic mixture satisfy the d quality. however, paving blocks with 5% and 15% plastic mixture do not even meet the compressive strength standard for d quality paving blocks. the results of this study are similar to several previous studies. awodiji et al. [23] used hdpe and pet plastic as a substitute for coarse aggregate with a ratio of 1:1, 1:1.5, and 1:2. the results showed that the sand-hdpe combination had higher compressive strength than regular paving blocks (sand-cement). however, the sand-pet combination had a low 0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 5 10 15 d en si ty ( kg /m ³) percent of plastic (%) wet density dry density advances in technology innovation, vol. 8, no. 3, 2023, pp. 219-228 225 compressive strength. krasna et al. [24] used plastic with 0%, 25%, 50%, 75%, and 100% of sand. compressive strength decreased with the addition of plastic. the compressive strength decreased by 7% to 59% from the compressive strength of paving blocks without plastic. mercante et al. [25] reported that adding plastic to a mortar reduced its compressive strength. this reduction in compressive strength can be caused by decreased density in the paving block mixture [21]. the smooth and slick plastic surface can weaken the aggregates’ bond. several factors affect the decrease in compressive strength: the lower strength of plastic compared to natural aggregate, the weak bond between cement and plastic, and the increase in air content in concrete [26]. in addition, the low compressive strength value can also influence by the paving blocks made with a hand-operated manual press. fig. 7 compressive strength of a paving block 3.4. water absorption fig. 8 depicts the water absorption test results. it shows how paving blocks without a plastic mixture absorb more water than those with a plastic mixture. according to sni 03-0691-1996, the maximum water absorption is 10% which is included in the d quality of paving blocks [20]. water absorption in paving blocks shows a decreasing trend with the addition of melted plastic. in fig. 8, all paving block samples have more than 10% water absorption, indicating that it does not reach the d quality. awodiji et al. [23] produced paving blocks using different combinations of sand-hdpe, sand-pet, and sand-cement with a ratio of 1:1, 1:1.5, and 1:2. the obtained results indicated that hdpe plastic is better than pet owing to its lower water absorption. the more plastic is used, the lower the water absorption, reaching 0%. paving blocks with a plastic mixture has lower water absorption due to plastic’s nature, which cannot absorb water [27]. plastics tend to pass through the water rather than absorb water [22]. the amount of water absorption also indicates susceptibility to damage [25]. high water absorption causes the pores of the paving block to fill with much water. this condition reduces compressive strength, making it prone to cracking or destruction. paving blocks with low water absorption last longer because they are less susceptible to chemical reactions, physical stress, and mechanical damage [27]. fig. 8 water absorption of a paving block 12.85 7.74 9.39 7.63 0.00 2.00 4.00 6.00 8.00 10.00 12.00 14.00 0 5 10 15 c o m p re ss iv e st re n gt h ( m p a) percent of plastic (%) 11.40 11.60 11.80 12.00 12.20 12.40 12.60 12.80 13.00 13.20 13.40 0 5 10 15 w at er a b so rp ti o n ( % ) percent of plastic (%) advances in technology innovation, vol. 8, no. 3, 2023, pp. 219-228 226 3.5. wear resistance the results of the wear resistance test showed that paving blocks without a plastic mixture (the control sample) outperform those with a plastic mixture. as shown in fig. 9, the addition of plastic increases the wear value. suardiana et al. [28] stated that adding plastic can reduce wear, which contradicts this study’s results. one of the factors that affect wear is surface roughness. the plastic mixed in the paving block has a rough texture, causing the surface to wear out quickly. the high wear value on paving blocks can also be caused by cement that does not react (bind) to the plastic. however, as shown in table 2, the minimum wear resistance for quality d paving blocks is less than 0.219. the value of wear resistance in this study was under the quality standard required by sni 03-0691-1996 for concrete brick (paving blocks), classifying it as good [20]. fig. 9 wear resistance of the paving block 3.6. discussion analysis was performed on the compressive strength, water absorption, and wear resistance of paving block quality based on sni 03-0691-1996, revealing that adding plastic to paving blocks tends to reduce the quality of paving blocks. paving blocks with 10% plastic have a compressive strength of 9.39 mpa and qualify as d quality, which can be used in parks. however, paving blocks with 5% and 15% of the plastic mixture do not meet the quality required by sni 03-0691-1996 [20]. regarding water absorption parameters, paving blocks with the plastic mixture in all different proportions result in exceeding 12% absorption. it also did not reach the d quality. compared to the control sample (0% plastic), the paving block with plastic mixture has lower water absorption. on the other hand, the wear resistance of paving blocks with the addition of plastic achieves the expected quality of d. based on the results, paving blocks with the addition of plastic can only fulfill one of the three parameters required for d quality, namely, wear resistance. however, paving blocks with a mixture of plastic have the potential to be used at a 10% variation. paving blocks with a 10% plastic mixture have a higher compressive strength than those of 5% and 15% plastic. melted plastic has a better ability than other forms of plastic to bond with other aggregates in paving blocks. paving blocks with 10% melted plastic meet quality d for compressive strength. furthermore, paving blocks with a melted plastic mixture also have a low water absorption rate. the lower the water absorption capacity, the better quality of the paving block. it means paving blocks will be more durable. further research and testing such as a scanning electron microscope (sem) test is needed to prove the bonding effect of melted plastic aggregate. manually pressed paving can impact the quality of the resulting paving blocks. the density of different paving blocks can be affected by differences in human power when making them, thereby influencing their durability. hence, additional research into the effect of plastic on paving blocks made with an automatic press machine is required. adding plastic to construction materials like paving blocks has good potential. it is a step toward recycling plastic waste and reducing the environmental burden of plastic. yin et al. [29] stated that adding plastic can increase concrete’s tensile strength. it is due to the ability of plastic to resist crack initiation in concrete. furthermore, concrete with a melted plastic mixture also has a smaller decrease in flexural strength than that with the addition of plastic in large size [25]. 0.041 0.05 0.066 0.067 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0 5 10 15 w ea r re si st an ce ( m m /m in ) percent of plastic (%) advances in technology innovation, vol. 8, no. 3, 2023, pp. 219-228 227 4. conclusion this study investigated ldpe plastic waste in the form of melted plastic. in paving blocks, melted plastic substitutes sand at 5%, 10%, and 15% of the sand weight. the compressive strength, water absorption, and wear resistance of conventional and melted plastic forms of paving blocks were analyzed based on sni 03-0691-1996. the following conclusions have been drawn: 1. paving blocks with a plastic mixture have lower compressive strength than the control sample (0% plastic). the compressive strength of paving blocks with a mixture of 10% plastic meets d quality, while the 5% and 15% variations do not meet d quality standards. 2. the water absorption of the paving blocks with the plastic mixture is superior to the control sample (0% plastic). however, it does not meet the d quality standard. as the plastic increases, the water absorption decreases. 3. the wear resistance of paving blocks with a plastic mixture is advantageous. the more plastic, the more wear-resistant paving blocks. according to sni 03-0691-1996, paving block with melted plastic tends to decrease in quality. however, paving blocks with a plastic mixture at a 10% variation have the potential to be used. recycling plastic into substitute materials for paving blocks can help reduce the negative environmental impact of plastic. finally, the results indicate that concrete containing a melted plastic mixture can be used as non-structural concrete. conflicts of interest the authors declare no conflict of interest. references [1] p. o. awoyera and a. adesina, “plastic wastes to construction products: status, limitations and future perspective,” case studies in construction materials, vol. 12, article no. e00330, june 2020. [2] i. vollmer, m. j. f. jenks, m. c. p. roelands, r. j. white, t. van harmelen, p. de wild, et al., “beyond mechanical recycling: giving new life to plastic waste,” angewandte chemie, vol. 59, no. 36, pp. 15402-15423, september 2020. [3] g. bidegain and i. paul-pont, “commentary: plastic waste associated with disease on coral reefs,” frontiers in marine science, vol. 5, article no. 237, july 2018. [4] j. jambeck, b. d. hardesty, a. l. brooks, t. friend, k. teleki, j. fabres, et al., “challenges and emerging solutions to the land-based plastic waste issue in africa,” marine policy, vol. 96, pp. 256-263, october 2018. [5] m. shams, i. alam, and m. s. mahbub, “plastic pollution during covid-19: plastic waste directives and its longterm impact on the environment,” environmental advances, vol. 5, article no. 100119, october 2021. [6] k. boyle and b. örmeci, “microplastics and nanoplastics in the freshwater and terrestrial environment: a review,” water, vol. 12, no. 9, article no. 2633, september 2020. [7] t. liu, a. nafees, s. khan, m. f. javed, f. aslam, h. alabduljabbar, et al., “comparative study of mechanical properties between irradiated and regular plastic waste as a replacement of cement and fine aggregate for manufacturing of green concrete,” ain shams engineering journal, vol. 13, no. 2, article no. 101563, march 2022. [8] f. pietra, “chapter 7 the oceans,” tetrahedron organic chemistry series, vol. 21, pp. 35-60, 2002. [9] n. singh, d. hui, r. singh, i. p. s. ahuja, l. feo, and f. fraternali, “recycling of plastic solid waste: a state of art review and future applications,” composites part b: engineering, vol. 115, pp. 409-422, april 2017. [10] ipb, “ipb waste management: waste generation study,” unpublished. (in indonesia) [11] r. s. mozumder, m. m. abedin, and m. t. islam, “experimental study on light weight concrete using plastic waste as a partial replacement of fine aggregate,” unpublished. [12] m. sofyan, h. parung, m. w. tjaronge, and a. a. amiruddin, “selected mechanical and physical properties concrete with polypropylene plastic granule aggregate,” iop conference series: earth and environmental science, vol. 1117, artichle no. 012014, 2022. advances in technology innovation, vol. 8, no. 3, 2023, pp. 219-228 228 [13] n. nurmaidah and y. t. pradana, “the effect of the mixture of plastic waste as a lightweight concrete material,” budapest international research in exact sciences (birex) journal, vol. 1, no. 2, pp. 65-75, april 2019. [14] s. qaidi, y. al-kamaki, i. hakeem, a. f. dulaimi, y. özkılıç, m. sabri, et al., “investigation of the physicalmechanical properties and durability of high-strength concrete with recycled pet as a partial replacement for fine aggregates,” frontiers in materials, vol. 10, article no. 1101146, 2023. [15] s. agyeman, n. k. obeng-ahenkora, s. assiamah, and g. twumasi, “exploiting recycled plastic waste as an alternative binder for paving blocks production,” case studies in construction materials, vol. 11, article no. e00246, december 2019. [16] s. sudarno, s. nicolaas, and v. assa, “pemanfaatan limbah plastik untuk pembuatan paving block,” jurnal teknik sipil terapan, vol. 3, no. 2, pp. 101-110, september 2021. (in indonesia) [17] h. hasaya, r. masrida, and d. firmansyah, “potensi pemanfaatan ulang sampah plastik menjadi eco-paving block,” jurnal jaring saintek, vol. 3, no. 1, pp. 25-31, april 2021. (in indonesia) [18] a. e. saputra, “peningkatan uji kuat tekan paving block dengan bahan limbah,” jurnal ilmiah teknik pertanian tektan, vol. 11, no. 3, pp. 165-172, december 2019. (in indonesia) [19] a. e. saputra, i. raharjo, and s. suprapto, “uji eksperimental kuat tekan mortar paving block dengan bahan limbah substitusi agregat halus dan semen,” prosiding seminar nasional pengembangan teknologi pertanian, vol. 2018, pp. 366-372, 2018. (in indonesia) [20] indonesian standard for concrete brick (paving block), sni 03-0691-1996, 1996. (in indonesia) [21] m. h. dermawan, “model kuat tekan, porositas dan ketahanan aus proporsi limbah peleburan besi dan semen untuk bahan dasar paving block,” jurnal teknik sipil dan perencanaan, vol. 13, no. 1, pp. 41-50, 2011. (in indonesia) [22] e. e. putri, ismeddiyanto, and r. suryanita, “sifat mekanik paving block komposit sebagai lapis perkerasan bebas genangan air (permeable pavement),” jurnal teknik, vol. 13, no. 1, pp. 9-16, april 2019. (in indonesia) [23] c. t. g. awodiji, s. sule, and c. v. oguguo, “comparative study on the strength properties of paving blocks produced from municipal plastic waste,” nigerian journal of technology, vol. 40, no. 5, pp. 762-770, september 2021. [24] w. a. krasna, r. noor, and d. d. ramadani, “utilization of plastic waste polyethylene terephthalate (pet) as a coarse aggregate alternative in paving block,” matec web of conferences, vol. 280, article no. 04007, 2019. [25] i. mercante, c. alejandrino, j. p. ojeda, j. chini, c. maroto, and n. fajardo, “mortar and concrete composites with recycled plastic: a review,” science and technology of materials, vol. 30, no. supplement 1, pp. 69-79, december 2018. [26] a. j. babafemi, b. šavija, s. c. paul, and v. anggraini, “engineering properties of concrete with waste recycled plastic: a review,” sustainability, vol. 10, no. 11, article no. 3875, november 2018. [27] y. yusrianti, n. noverma, and o. e. hapsari, “analisis sifat fisis penyerapan air pada paving block dengan campuran variasi limbah abu ketel dan limbah botol plastik,” al ard jurnal teknik lingkungan, vol. 5, no. september, pp. 1-8, 2019. (in indonesia) [28] i. w. suardiana, n. p. g. suardana, and c. i. p. k. kencanawati, “pengaruh waktu perendaman terhadap daya serap air dan keausan pada paving block plastik-pasir,” seminar nasional teknoka, vol. 5, pp. 266-273, january 2021. (in indonesia) [29] s. yin, r. tuladhar, f. shi, m. combe, t. collister, and n. sivakugan, “use of macro plastic fibres in concrete: a review,” construction and building materials, vol. 93, pp. 180-188, september 2015. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). microsoft word 6-v8n3(2023)-aiti#11638(229-239).docx advances in technology innovation, vol. 8, no. 3, 2023, pp. 229-239 english language proofreader: hsin-te hsieh design of an adiabatic calorimeter for cementitious mixtures by multi-objective optimization jhonatan a. becerra-duitama1,*, mauricio mauledoux2, óscar f. avilés2 1faculty of engineering and basic sciences, juan de castellanos university, tunja, colombia 2faculty of engineering, military university nueva granada, bogotá, colombia received 22 february 2023; received in revised form 17 june 2023; accepted 19 june 2023 doi: https://doi.org/10.46604/aiti.2023.11638 abstract this study aims to design an adiabatic calorimeter for cementitious mixtures using nsga-ii and the pareto optimal solution set. in this multi-objective optimization, the controller effort and heating time are selected as objective functions. likewise, the volume and the material to be heated were chosen as decision variables. the optimal solution was selected using nash bargaining methods. after implementing the optimal solution, the wilcoxon test was applied to statistically validate the developed work. the measurements performed were compared with other research and it was observed an improvement in the measurement of heat of hydration in cementitious mixtures. also, it was noted a decrease in the error in the temperature measurement. keywords: calorimeter, multi-objective problem, nsga-ii, optimization, pareto front 1. introduction nowadays, the measurements of the heat of hydration in cementitious mixtures are required. for this purpose, some equipment has been developed over the years to perform these tests. the most commonly used devices to determine the heat of hydration in cement mixtures are isothermal, semi-adiabatic, and adiabatic calorimeters. the latter is the least used because it is complicated and costly to manufacture, and its correct operation requires high sensitivity [1]. nevertheless, it is the most accurate method and provides the best measurements [2]. for this reason, methods based on concurrent engineering were applied, to develop a low-cost and easy-to-acquire calorimeter. concurrent design can be applied to develop any project in a cross-functional manner bearing in mind all stages of the product life cycle, from manufacturing to final disposal [3-5]. additionally, the concurrent design reduces production costs and time to market while increasing product quality. deshpande [6] mentions that various companies such as hewlett-packard, texas instruments, general motors, and motorola have successfully implemented concurrent engineering projects. among the benefits obtained, can be noted: a 55 % reduction in time to market, a 70 % increase in performance, and a 350 % increase in quality. in addition, kügler et al. [7] found a reduction in production time from 30 to 18 months in the development of a new minicomputer. wang and feng [8] indicated that implementing concurrent design procedures in companies reduced the number of design changes by 50 %, shortened product design time by 40 % to 70 %, and decreased waste generated by 75 %. previously this study, this type of concurrent design methodology had not been used to develop adiabatic calorimeters for cement mixtures, even considering the existing problems with these devices. the first problem is the control system that has a low response time, prasath and santhanam [9] described how the developed calorimeter took about one hour to reach the * corresponding author. e-mail address: jalexanderbecerra@jdc.edu.co advances in technology innovation, vol. 8, no. 3, 2023, pp. 229-239 230 established reference. the second problem is a high working sensitivity and good insulating material are necessary for the heat transfer to be close to zero, allowing for higher accuracy in the measured data [10]. the third problem is the high production cost, making it a commercially nonviable and quite exclusive device [11]. for the above-mentioned reasons, methods based on concurrent design were implemented, to design a low-cost and easyto-obtain calorimeter. therefore, it was used a multi-objective optimization algorithm, which took into account all aspects of the system’s operation. to achieve the optimal design, the non-dominated sorting genetic algorithm ii (nsga-ii) was used to determine the parameters, dimensions, and materials required for the design of the adiabatic calorimeter [12]. likewise, it was implemented the calorimeter designed to perform heat of hydration measurements on cement paste samples. the novelty of this work is to display that the calorimeter was developed using concurrent design (optimization algorithms) unlike existing calorimeters implemented [13-16]. it is expected that this work will be one of the first to develop optimized calorimeters for cement mixtures [17]. 2. methods this section describes the development of the project, which was divided into three parts. the first part is the implementation of the optimization algorithm, for which the non-dominant sorting genetic algorithm (nsga-ii) was used. it is important to point out that, before implementing the nsga-ii algorithm, two evolutionary algorithms were tested. the evolutionary algorithms used were: the multiobjective differential evolution (mode) algorithm [18] and the differential evolution multiobjective optimization (demo) algorithm [19]. however, due to the nature of the calorimeter, the two algorithms could not be implemented, because the optimization problem had mixed variables, since one objective function had to choose the insulating material (something tangible), and the other objective function had to choose the physical dimensions of the calorimeter (numbers). also, implementing either of the two evolutionary algorithms would have required transforming the objective functions so that both were under the same conditions. this would have introduced imprecision and inaccuracy in the results obtained. for this reason, the nsga-ii algorithm was chosen, since it did allow working with mixed variables. the objective functions to take into account were the environment to be heated and the cumulative error of the measurement. by applying the algorithm and the two objective functions, the pareto front was determined, by which the solution to be implemented was chosen. hence, it was needed to apply bargaining methods. the methods used were the nash method and the egalitarian method. the second part of the project involved the design and implementation of the control system from simulation to its commissioning. finally, the culmination of the third part of the project was focused on the construction of the adiabatic calorimeter, bearing in mind the design parameters calculated in the previous parts. in addition, some tests were carried out with cement paste specimens, verifying the correct operation by checking with other similar studies. fig. 1 shows the schematic diagram where the design steps are represented. fig. 1 schematic diagram of the design process 2.1. algorithm optimization to obtain the best solution for the problem, it was adopted the non-dominant sorting genetic algorithm (nsga-ii) taken from kalami [20]. the parameters of the algorithm employed are shown in table 1. before implementing the optimization advances in technology innovation, vol. 8, no. 3, 2023, pp. 229-239 231 algorithm, it was established the objective functions that governed the selection of the optimal calorimeter design. it is worth noting that the first function chosen depends directly on the thermal process. the parameters and materials included in this function were selected based on the physical nature of the process. moreover, this function was established to minimize the heating time of the environment to be controlled. table 1 nsga-ii parameters p1 p2 p3 p4 p5 100 100 0.7 0.4 0.02 where p1 is the maximum number of iterations, p2 is the population size, p3 is the crossover percentage, p4 is the mutation percentage, and p5 is the mutation rate. additionally, the second objective function was defined as the minimization of the cumulative error in the measurement. as can be seen, the behavior of the two functions is conflicting, since to achieve a short warm-up time, the controller must act quickly. therefore, finding the right balance between both functions was required to obtain the optimal design. the followings are the objective functions and the decision variables proposed for the development of the project. objective function 1 (f1) is intended to minimize water heating time. the heating time in terms of heat generated and power delivered to the system is expressed as 1 cespecifi f t c t q p v pρ = = × × ∆= × (1) where t is time (s), q is generated heat (j), p is heating element power (w), � is the density of the chosen substance (kg/m3), v is the volume of the vessel containing the substance (m3), cspecific is the specific heat of the substance (j/kg°c), and δ� is temperature variation (°c). decision variables: the calorimeter material and dimensions were setting it up as decision variables. table 2 shows the materials proposed as insulation for the calorimeter and their characteristic values. table 2 proposed materials for calorimeter insulation material specific heat ce (j/kg°c) density (kg/m3) m1 4186 1000 m2 3730 1035 m3 3300 1053 m4 1 1.239 where m1 is water, m2 is water and coolant at 50%, m3 is water and coolant at 30%, and m4 is air. as the refrigerant ethylene glycol was chosen. the proposed dimensions are set out in table 3: table 3 proposed calorimeter dimensions volume (l) width (cm) length (cm) height (cm) 15 24 22 28 25 30 26 32 30 30 31 32 35 30 32 36 objective function 2 (f2) is intended to minimize cumulative measurement error. to ensure minimal error, it was called to simulate the response of the controller, to establish the best parameters to be implemented. the simulation was performed using matlab software. first, the mathematical model of the thermal system to be controlled was set up. the differential equation representing the thermal system is expressed as: advances in technology innovation, vol. 8, no. 3, 2023, pp. 229-239 232 dt p t dt v cpρ × = × × (2) afterward, the type of controller to be implemented could be designed and simulated. the input variable is voltage and the output variable is temperature. the controller was determined using the sliding mode method. the sliding surface proposed for the controller utilized is represented by 1 0 1 dx k x xσ = + − (3) the equivalent controller and the attractive controller are respectively defined as ( )0 1d v cp ueq k x x i t ρ × × = × − × (4) ( )sgnn v cp u l i t ρ σ × × = × × × (5) after that, it was determined the cumulative error of the controller using simulink and matlab. for this purpose, it was established the value of the constant k0 present in eqs. (3)-(4), based on the previously defined decision variables. fig. 2 shows the method to identify the cumulative error of the controller. it can be seen that the study signal is the input signal, which is the temperature in the cement sample. likewise, it can be seen in the sliding mode representation of the controller. the orange color represents the attractive control, while the fuchsia color represents the equivalent control. sliding mode control provides a systematic approach to the problem of maintaining accuracy, stability, adjustability, and performance in the face of model inaccuracy. it is also an efficient tool for designing robust controllers for complex nonlinear and multivariable plants. there are two main advantages of this type of control. first, the dynamic behavior of the system can be adapted by the particular choice of a slip function. secondly, the closed-loop response becomes insensitive to some particular uncertainties. fig. 2 computation of the cumulative error of the controller when the algorithm finishes its process, it displays the set of optimal solutions, entitled the pareto front. because all the values obtained are possible answers, it is necessary to apply bargaining methods to obtain the choice criterion. the negotiation methods used to find the selected solution to manufacture the calorimeter were the egalitarian method and the nash method [21]. 2.2. construction of the calorimeter after applying the optimization algorithm, the control system was designed and implemented. as mentioned previously, the controller design was selected using sliding modes. to fully understand the behavior of the control system, it is important advances in technology innovation, vol. 8, no. 3, 2023, pp. 229-239 233 to mention the various parts and the operation of the calorimeter. the part where the measurement was carried out is a cubic acrylic container with a side length of 35 cm (fig. 3(a)). the cement sample was contained inside a 10 cm box (fig. 3(b)). this box has an extended lid that allows the sensor to enter the sample. in turn, the sample is housed inside the small box and contained inside the large container (fig. 3(c)). (a) cubic acrylic container (b) sample recipient (c) containers used fig. 3 vessels used for calorimeter measurement moreover, to ensure minimal heat loss, the water temperature was controlled by three separate systems. the first is a temperature homogenization system, which mixed the water so that the temperature was approximately the same throughout the volume. this system has two mixing paddles operated by electric motors. the functioning of this system is independent of the variation of the temperature in the calorimeter, so that, as soon as it is started, the paddles are activated. the second is the heating system, which consists of two heating resistors (fig. 4). the actuators of these two systems are located on the cover of the main box. in turn, the calorimeter has 3 temperature sensors, 1 for the sample and 2 for water temperature measurement. fig. 4 mixing paddles and heaters fig. 5 cooling system the third and last system was the cooling system. this has a pump that allows the flow of water to an external container filled with cooling liquid, to reduce the calorimeter’s ambient temperature through heat exchange, fig. 5. it can be observed that the heating and cooling systems depend on the temperature of the sample (reference temperature). when the sample temperature is below the ambient temperature, the cooling system is activated; consequently, when the sample temperature is above the ambient temperature, the heating system must be activated. to measure temperature changes, submersible ntc 10k sensors were used. in addition to the three systems mentioned above, the calorimeter has power and a control element. fig. 6 shows these components. the power element consists of relays (a) that allow the activation or deactivation of the heating and cooling system actuators. also, the control element includes a microcontroller (arduino mega) (b), an electronic conditioner for the sensors (c), a bluetooth module for data acquisition (d), and a module for data storage using micro-sd memory (e). fig. 7 shows the calorimeter modeled in solidworks, where the different components and systems that are part of the developed device can be seen more clearly. advances in technology innovation, vol. 8, no. 3, 2023, pp. 229-239 234 fig. 6 power and control part fig. 7 calorimeter developed 3. results and discussion as a result of using the optimization algorithm (nsga-ii), it was obtained a set of solutions or pareto front. to achieve this, the algorithm was implemented with a population of 100 individuals, fig. 8. the behavior of the pareto front is consistent with the definition of the two objective functions since the system is a minimization problem and its shape must be convex. this generates a conflict in the possible solution; because, for the heating time to be minimal, the cumulative error of the controller must be large, and, by contrast, for the cumulative error of the controller to be reduced, the heating time must be increased. convex pareto fronts are obtained with a set of acceptable multi-objective optimization solutions and satisfying the objective functions [22-24]. fig. 8 pareto front fig. 9 nash and egalitarian negotiation methods advances in technology innovation, vol. 8, no. 3, 2023, pp. 229-239 235 for a better analysis of the results obtained, it was needed to normalize the pareto front, reduce the number of decimal places and obtain simpler values to analyze. it should be noted that the points appear in this way in the graph because the data were filtered and only normalized data (values between 0 and 1) appear. it is important to note that the data processed by the nsga-ii algorithm has the solutions to the proposed problem or the decision variables, that is, the genotype. as a result of collecting this data, the response of the objective functions of the problem, represented by the pareto front, meaning the phenotype, was found. to obtain the best design (the best solution), it was required to establish the genotype, so bargaining methods were used, which are aimed to find a better decision. fig. 9 shows the pareto front with the two negotiation methods used. by applying the bargaining methods, it can be evidenced they pass over 3 coordinate points. the egalitarian bargaining method goes through the coordinate point (0.2, 0.2), while the nash bargaining method goes through the coordinate points (0.1, 0.4), (0.2, 0.2), and (0.4, 0.1). the optimal values of the phenotype and genotype are shown in table 4. as previously mentioned, the decision variables selected were the volume and type of material to be heated. the optimal material to heat is water; with the volume varying between 18 and 30.5 liters. it is highlighted these values were the result of the data processing performed by the nsga-ii algorithm. table 4 decision variables obtained by the negotiation methods. phenotype negotiation methods decision variables (genotype) volume (l) materiales to be heated (0.1, 0.4) nash 18 water (0.2, 0.2) equal/nash 30.5 water (0.4, 0.1) nash 27.4 water despite the fact all three solutions were optimal, a water volume of 18 liters was finally chosen. this choice was made to minimize the costs in the manufacture of the calorimeter, since the smaller the dimensions, the lower the manufacturing cost. it is clarified the choice of the final design will depend on the designer’s criteria and on some factors not considered in the algorithm, such as costs, physical space, portability, and sample size, among others. the calorimeter’s dimensions were made slightly larger than the volume of water to be heated so that all the components could be positioned comfortably and practically. as mentioned before, the control system was designed using sliding modes. this type of control is composed of two parts. the first part is the equivalent control; which has the function of restricting the reference temperature to the chosen non-stick surface. the second method is the attractive control, aimed at preventing the reference temperature from deviating from the desired trajectory, thus attracting it to the calculated non-stick surface. (a) error obtained with concurrent design (b) error obtained within lin [25] fig. 10 error obtained in the tests after verifying the design of the controller was correct, it was then implemented. fig. 10(a) shows the reference temperature variation (orange curve) and the water temperature (black curve). it is observed that the difference between the reference temperature and the water temperature did not exceed 0.2 °c/h. compared with lin [25], the relationship in the advances in technology innovation, vol. 8, no. 3, 2023, pp. 229-239 236 change of temperatures is similar, as shown in fig. 10(b), and the temperature loss has less variation; this is a result of the concurrent design used. it is noted the system to be controlled is a thermal system and the response of these systems is slow due to the nature of the system itself. to assure the calorimeter’s proper functioning, three tests were carried out using identical cement paste mixtures. the main characteristics of the mixture employed were; 50 g of cement, 35 ml of water, a water/cement ratio of 0.7, and a total volume of 24.74 cm3. fig. 11 shows the temperatures obtained for each test. it is seen that the difference between the ambient temperature and the temperature of the sample was quite minimal (about 0.2 °c). however, the temperature of the sample increases as time goes by, because of the chemical nature of the cement. the temperature values obtained are related to the size of the sample; it is important to keep in mind that using a smaller mass will result in a smaller temperature variation. fig. 11 measured temperatures to verify that the variation in ambient temperature versus sample temperature is statistically acceptable, the wilcoxon test was used to verify the results obtained. for this purpose, it was defined the null hypothesis and the alternative hypothesis as: null hypothesis (h0): the curves obtained are not acceptable. alternative hypothesis (h1): the curves obtained are acceptable. to obtain the values of the parameters p and h, the data used to obtain the curves in fig. 11 and matlab were used. table 5 shows the data obtained by applying the wilcoxon test: table 5 values obtained from wilcoxon test temperature p h δt1 vs δt2 0.00024 1 δt1 vs δt3 0.0166 1 δt2 vs δt3 0.00006 1 the δt values correspond to the difference between the sample temperature and the ambient temperature for each measurement taken (t1, t2, and t3). in other words, the value considered was the error in the measurement, since the main objective of the calorimeter is that the difference between temperatures is as close to zero as possible. it can be seen that the p < 0.05 and the value of h = 1 for all the measurements made, therefore the null hypothesis is rejected and the alternative hypothesis is accepted. fig. 12 shows the heat of hydration curves obtained for each measurement. the behavior of the three curves was similar, with a slight difference between the maximum values. furthermore, the maximum heat peak occurred approximately 2 hours after the test. the curves obtained are based on the temperatures measured in the calorimeter, the mathematical model that governs the behavior of these curves is part of another study, however, it was important in establishing the behavior of the advances in technology innovation, vol. 8, no. 3, 2023, pp. 229-239 237 hydration heat. it is important to clarify that, even though the sample conditions were the same for each measurement, cement hydration is a chemical process that is not exact and is approximated by mathematical models; therefore, the results obtained in fig. 12 are not the same, but its performance is in line with what is reported in the literature. the resulting curves were compared to some similar investigations. in chidiac and shafikhani [26], the heat of the hydration curve was obtained for different types of cement mortar mixtures, which means the number of samples used was higher than in the present work. the maximum peak was achieved approximately 9 hours after the start of the measurement and heat of hydration values between 0.5 and 4.0 w/g were obtained. the measurement time was 26 hours. fig. 12 hydration heat curves obtained additionally, sanderson et al. [1] evaluated the behavior of the hydration heat in cement mortars with a sample volume of 125 cm3. the values of the heat of hydration oscillate between 0.0005 w/g and 0.004 w/g, and the maximum peak is reached almost 12 hours after testing. the measurement time was 24 hours. besides, in lin and chen [27] cubic concrete specimens of 5.8 m3 volume were made. the measurement time was 144 hours. the maximum heat peak was 3.7 w/kg and the time to reach it was approximately 9 hours. additionally, in prasath and santhanam [9], the heat of the hydration curve was obtained for concrete samples with a volume of 5300 cm3 approximately, and the measurement time was approximately 40 hours. the maximum heat peak was 4.8 w/kg and was reached 10 hours after the start of the test. on the other hand, chen et al. [28] measured the heat of hydration in cement pastes with limestone-calcined and clay slag. due to the nature of the samples, the peak values vary, however, the maximum peak heat was obtained between 9 and 15 hours, and the measurement time was approximately 48 hours. the maximum heat value was 1.1 mw/g. it is observed that the values and times in the graphs vary since all the mixtures do not have the same volumes, the same mixture designs, or the same measurement time. nevertheless, it can be said that the results obtained by the calorimeter meet the expectations, considering that the behavior of the heat curves obtained corresponds to what has been reported in the literature. 4. conclusions concurrent engineering was used to obtain the optimum design of an adiabatic calorimeter for cementitious mixtures. nsga-ii was used as the optimization algorithm. with the development of this work, it was found that: (1) one benefit of using nsga-ii to design and construct the adiabatic calorimeter for cement mixtures was the algorithm's ability to work with mixed variables, allowing for the optimization of both the material to be heated and the device's size. the evolutionary algorithms demo and mode, which were tested prior to selecting the optimization algorithm to be employed, were unable to achieve this. (2) another benefit was that, unlike the evolutionary algorithms demo and mode, the second objective function permitted the use of design variables with decimal numbers, enabling the convex pareto front typical of a minimization issue. advances in technology innovation, vol. 8, no. 3, 2023, pp. 229-239 238 (3) it was possible to produce a pareto front with a convex form, good distribution, and dispersion; this prevented the algorithm from becoming stuck in regionally optimal spots. the diversity of the sample was well preserved across the many simulations thanks to the excellent performance of the agglomeration distance operator for nsga-ii during selection. (4) the measurement error was lower compared to studies where concurrent engineering was not applied, thus demonstrating the advantage of using optimization algorithms for the design. (5) the wilcoxon test was applied, and a value of p < 0.05 and a value of h=1 were found, indicating the statistical validity of the tests performed. (6) among the limitations found when implementing nsga-ii, one of them was its long simulation time, since it took 2 to 3 hours to obtain the pareto front, increasing the computational cost of using this algorithm. (7) also, due to the nature of the process to be optimized, it was not possible to use nsga-ii without modification since it was necessary to create a specific function that included the chosen objective functions. (8) the novelty of this work is that the calorimeter was developed using concurrent design (optimization algorithms) unlike existing calorimeters. this work is expected to pioneer the development of optimized calorimeters for cement mixtures. acknowledgments the authors acknowledge juan de castellanos university for the funding given for the development of the project. the authors would also like to thank professor manuel alberto montaña suárez for the complete translation of the article. conflicts of interest the authors declare no conflict of interest. references [1] r. a. sanderson, g. m. cann, and j. l. provis, “comparison of calorimetric methods for the assessment of slag cement hydration,” advances in applied ceramics, vol. 116, no. 4, pp. 186-192, 2017. [2] w. wongkeo, p. thongsanitgarn, c. s. poon, and a. chaipanich, “heat of hydration of cement pastes containing high-volume fly ash and silica fume,” journal of thermal analysis and calorimetry, vol. 138, no. 3, pp. 2065-2075, november 2019. [3] t. s. gan, m. steffan, m. grunow, and r. akkerman, “concurrent design of product and supply chain architectures for modularity and flexibility: process, methods, and application,” international journal of production research, vol. 60, no. 7, pp. 2292-2311, 2021. [4] m. h. esfe, m. hajian, d. toghraie, m. khaje khabaz, a. rahmanian, m. pirmoradian, et al., “prediction the dynamic viscosity of mwcnt-al2o3 (30:70)/ oil 5w50 hybrid nano-lubricant using principal component analysis (pca) with artificial neural network (ann),” egyptian informatics journal, vol. 23, no. 3, pp. 427-436, september 2022. [5] j. s. xia, m. k. khabaz, i. patra, i. khalid, j. r. n. alvarez, a. rahmanian, et al., “using feed-forward perceptron artificial neural network (ann) model to determine the rolling force, power and slip of the tandem cold rolling,” isa transactions, vol. 132, pp. 353-363, january 2023. [6] a. deshpande, “concurrent engineering, knowledge management, and product innovation,” journal of operations and strategic planning, vol. 1, no. 2, pp. 204-231, december 2018. [7] p. kügler, f. dworschak, b. schleich, and s. wartzack, “the evolution of knowledge-based engineering from a design research perspective: literature review 2012–2021,” advanced engineering informatics, vol. 55, article no. 101892, january 2023. [8] t. wang and j. feng, “optimal review strategy for concurrent execution of design–construction tasks in db mode,” journal of construction engineering and management, vol. 149, no. 1, article no. 04022139, january 2023. [9] g. prasath and m. santhanam, “experimental investigations on heat evolution during hydration of cementitious materials in concrete using adiabatic calorimeter,” indian concrete journal, vol. 87, no. 5, pp. 19-28, 2013. advances in technology innovation, vol. 8, no. 3, 2023, pp. 229-239 239 [10] k. a. riding, j. vosahlik, k. bartojay, c. lucero, a. sedaghat, a. zayed, et al., “methodology comparison for concrete adiabatic temperature rise,” aci materials journal, vol. 116, no. 2, pp. 45-53, 2019. [11] b. klemczak and m. batog, “heat of hydration of low-clinker cements,” journal of thermal analysis and calorimetry, vol. 123, no. 2, pp. 1351-1360, february 2016. [12] d. zou, d. gong, and h. ouyang, “a non-dominated sorting genetic approach using elite crossover for the combined cooling, heating, and power system with three energy storages,” applied energy, vol. 329, article no. 120227, january 2023. [13] g. de schutter and l. taerwe, “general hydration model for portland cement and blast furnace slag cement,” cement and concrete research, vol. 25, no. 3, pp. 593-604, april 1995. [14] y. k. ramu, i. akhtar, and m. santhanam, “use of adiabatic calorimetry for performance assessment of concretes,” advances in cement research, vol. 28, no. 8, pp. 485-493, september 2016. [15] e. r. graeff, “elevação de temperatura de concretos com baixo consumo de cimento e adição de cinza volante,” ms. dissertation, department civil engineering, universidade federal de santa catarina, florianópolis, sc, 2017. (in português) [16] j. wang, j. zhang, and j. zhang, “cement hydration rate of ordinarily and internally cured concretes,” journal of advanced concrete technology, vol. 16, no. 7, pp. 306-316, 2018. [17] s. verma, m. pant, and v. snasel, “a comprehensive review on nsga-ii for multi-objective combinatorial optimization problems,” ieee access, vol. 9, pp. 57757-57791, 2021. [18] h. a. al-jamimi, g. m. binmakhashen, k. deb, and t. a. saleh, “multiobjective optimization and analysis of petroleum refinery catalytic processes: a review,” fuel, vol. 288, article no. 119678, march 2021. [19] m. ayaz, a. panwar, and m. pant, “a brief review on multi-objective differential evolution,” soft computing: theories and applications: proceedings of socta 2018, pp. 1027-1040, 2020. [20] m. kalami, “nsga-ii in matlab,” https://www.yarpiz.com, january 24, 2023. [21] y. wang, z. liu, c. cai, l. xue, y. ma, h. shen, et al., “research on the optimization method of integrated energy system operation with multi-subject game,” energy, vol. 245, article no. 123305, april 2022. [22] y. zhou, s. cao, r. kosonen, and m. hamdy, “multi-objective optimisation of an interactive buildings-vehicles energy sharing network with high energy flexibility using the pareto archive nsga-ii algorithm”, energy conversion and management, vol. 218, article no. 113017, august 2020. [23] a. vukadinović, j. radosavljević, a. đorđević, m. protić, and n. petrović, “multi-objective optimization of energy performance for a detached residential building with a sunspace using the nsga-ii genetic algorithm,” solar energy, vol. 224, pp. 1426-1444, august 2021. [24] z. babajamali, m. khaje khabaz, f. aghadavoudi, f. farhatnia, s. a. eftekhari, and d. toghraie, “pareto multiobjective optimization of tandem cold rolling settings for reductions and inter stand tensions using nsga-ii,” isa transactions, vol. 130, pp. 399-408, november 2022. [25] y. lin, “thermal stress analysis for early age mass concrete members,” ph.d. dissertation, department of civil and environmental engineering, west virginia university, morgantown, wv, 2015. [26] s. e. chidiac and m. shafikhani, “cement degree of hydration in mortar and concrete,” journal of thermal analysis and calorimetry, vol. 138, no. 3, pp. 2305-2313, november 2019. [27] y. lin and h. l. chen, “thermal analysis and adiabatic calorimetry for early-age concrete members,” journal of thermal analysis and calorimetry, vol. 122, no. 2, pp. 937-945, november 2015. [28] y. chen, y. zhang, b. šavija, and o. çopuroğlu, “fresh properties of limestone-calcined clay-slag cement pastes,” cement and concrete composites, vol. 138, article no. 104962, april 2023. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation, vol. 2, no. 4, 2017, pp. 119 125 119 cop y right © taeti a cnn based approach for garments texture design classification s. m. sofiqul islam, emon kumar dey * , md. nurul ahad tawhid, b. m. mainul hossain institute of information technology, university of dhaka, bangladesh. received 03 oct ober 2016; received in revised form 04 february 2017; accept ed 08 february 2017 abstract identifying garments texture design automatically for recommending the fashion trends is important nowadays because of the rapid growth of online shopping. by learning the properties of images efficiently, a machine can give better accuracy of classification. several hand-engineered feature coding exists for identifying garments design classes. recently, deep convolutional neural networks (cnns) have shown better performances for different object recognition. deep cnn uses multiple levels of representation and abstraction that helps a machine to understand the types of data more accurately. in this paper, a cnn model for identifying garments design classes has been proposed. experimental results on two different datasets show better results than existing two well-known cnn models (alexnet and vggnet) and some state-of-the-art hand-engineered feature extraction methods. ke ywor ds : cnn, deep learn ing, a le xnet, vggnet, texture descriptor, garment categories, garment trend identification, design classification for garments 1. introduction online shopping is popular nowadays. customer select products from the web pages according to their choice and that can help to predict the direction of trends. if a retailer knows popular design styles of clothing products, it can increase the production of those styles to achieve more profit. therefore, if a system can classify the garments products according to different style, texture, size etc., it can automatically suggest different products to the customers based on their choices. the system proposed in this paper, can classify clothes according to textures . effective design classification based on textures, local spatial variations of intensity or colour in images has been an important topic of interest in the past decades. a successful classification, detection or segmentation requires an efficient description of image textures. to fulfil this purpose, lots of well-known hand-engineered feature extraction methods such as census transform histogram (centrist) [1], local binary pattern (lbp) [2], histogram of oriented gradient (hog) [3] etc., are exist. lbp gains popularity because of their computational simplicities and better accuracies. but, it is very sensitive to uniform and near uniform region. ltp [4], completed local binary pattern (clbp) [5] can handle this issue more accurately. between these two methods, clbp is better choice because this method is rotation invariant. cent rist [1] has gain popularity by incorporating spatial pyramid (sp) structure. but, most recently completed cent rist (ccentrist) and ternary centrist (tcentrist) [6] gained high accuracies for garments design classification. although several hand-engineered feature extraction approaches exist for garments design classification, deep learning is rarely used in this field. our goal is to apply appropriate deep learning model to measure the performance of garments design identification based on textures. in recent year, deep learning has become popular in the field of machine learning and computer vision. using large architectures with numerous features, many deep learning models achieve high performance in the field of object detection, text classification, image classification, face verification, gender classification, scene-classification, digits and traffic signs recognition, etc. some of the available deep learning models are alexnet [7], vggnet [9], berkeley-trained models [10], places-cnn model [8], places -cnds models on scene recognition [11], models for age and gender classification [12], googlenet model [13], etc. these methods have achieved dramatic improvements and attracted considerable interest in both the academic and industrial communities. in general, deep learning algorithms attempt to learn hierarchical features, corresponding to different levels of abstraction. each of these models concerned about some specific issues: preventing over-fitting, connection of nodes between adjacent layers, large learning capacity, etc. several factors need to be considered for working with deep learning network such as availability of large training set, powerful gpu for training and testing, better model regularization strategies, the amount of training time that one can tolerate, etc. the major contributions of this paper are as follows. (1) in this paper, a b rie f revie w on e xisting we ll known hand-eng ineered feature e xt raction methods for garments design class identification has been conducted. (2) this research has applied some e xisting deep convolutional neura l net work models for classifying the clothing products on some datasets and compared the results with several state-of-the-art hand-engineered feature extraction methods. (3) a new deep convolutional neural network mode l has been proposed for classifying garments design class. this proposed model is applied on two different datasets and has found a remarkable output. the rest of the paper is structured as follows. section 2 and section 3 describe the background studies and the methodology respectively. section 4 has presented the experimental results and finally section 5 concluded the overall work with necessary explanation. 2. background studies in this section, some e xisting garments clothing segmentation and classification strategies have been described. some existing deep learning models; that have *corresponding aut hor. email: emonkd@iit.du.ac.bd advances in technology innovation, vol. 2, no. 4, 2017, pp. 119 125 120 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti been used for several applications in computer vision are also narrated in this section. 2.1. garment product segmentation and identification ya maguchi et al. [14] proposed a method for clothing parsing. for this work, they created fashionista dataset consisting of 158,235 images. from this dataset, they selected 685 images for training and testing their system. they identified 14 different parts of a body and different clothing regions. in [15], they deal with clothing parsing problem using retrieval based approach. their proposed approach focused on pre-trained global clothing models, local clothing models, and transferred parse. authors found that their proposed final parse achieve 84.68% parsing accuracy. menfredi et al. [16] proposed a new approach for automatic garments segmentation and classification. they classified garments into nine different classes such as skirts, shirt, dresses, etc. for this work, authors used a projection histogram for extracting few specific garments. they divided the whole image into 117 cells and group them into 3*3 cells. they computed hog features [17] from each cell and the orientations are grouped into nine bins. they used multiclass linear support vector for training. serra et al. [18] did similar type of work, where authors used conditional random field (crf) for divided outfits. vittayakorn et al. [19] used five different features such as color, texture, shape, parse and style descriptor to identify three different visual trends, namely floral print, pastel color and neon color from runway to street fashion. however, using more color and design classes would be more beneficial in this field. kalantidis et al. [20] proposed a system to identify the relevant product where they firstly estimated the pose of a person from an input image and then segmented the clothing area such as shirt, tops, jeans, etc. finally, they applied an image retrieval technique which is 50 times faster than [14] for identifying similar clothes for each class . gallagher et al. [21] used grab cut algorithm for identifying a person by segmenting the clothing parts. bourdev et al. [22] proposed a new method for detecting some attributes and type of cloths from an input image. here attributes are gender, hair style and types of clothes such as t-shirts, pants, jeans, and shorts etc. for this work, they created a dataset consisting of 8000 people images with annotation. 2.2. texture based classification nowadays , garments design classification based on texture has become more popular and there are several existing well known methods such as histogram o f oriented gradients (hog), local binary pattern (lbp), features wavelets transform, noise adaptive binary pattern (nabp), gabor filters, scale-invariant feature transform (sift) etc. recently, lbp has become popular because of its computational simplicity. lbp was proposed for describing the local structure of an image and it has been used in several areas such as facial image analysis, including face detection, face recognition and facial expression analysis, demographic (gender, race, age, etc.) classification, moving object detection, etc. however, lbp is very sensitive in uniform and near uniform regions. in the last few years, lots of researches have been done by modification of lbp to improve the performance. such as derivative-based lbp, dominant lbp, rotation invariant, center-symmetric lbp, etc. tan and triggs [4] proposed a new texture based method local ternary patterns (ltp), which can tolerate noises up to a certain level. they used a fixed threshold (±5), for making ltp more discriminant and less sensitive to noise in a uniform region. there are also several other methods that can handle noises in different application areas, such as the methods described by jun et al. [25]. they proposed local gradient pattern (lgp) for texture based face detection. this method is a variant of lbp and uses adaptive threshold for code generation. guo et al. [5] proposed completed local binary pattern (clbp), which incorporates sign, magnitude and center pixel information. this method is rotation invariant and capable of handling the fluctuation of intensity. wu et al. [1] proposed census transform histogram (cent rist) which is very similar to lbp and mainly work as a visual descriptor for recognizing scene categories. centrist proposes a spatial representation based on a spatial pyramid matching scheme (spm) [26] to capture global structure from images. centrist uses total 31 blocks to avoid the artefacts. dey et al. [6] proposed two new descriptors for garments design class identification namely completed centrist (ccentrist) and ternary centrist (tcentrist). these descriptors are based on completed local binary pattern (clbp), local ternary pattern (ltp) and census tranformed histogram (cent rist). authors applied these two descriptors on two different publically available databases and achieve nearly about 3% more accuracy than the existing state-of-the art methods . 2.3. deep learning this sub-section, will describe some deep learning techniques for garments design classification. deep network learn features automatically from large number of unlabelled data, hence more useful hidden discriminative features are extracted. it has achieved popularity in classic problems, such as speech recognition, object recognition and detection, natural language processing, etc. convo lutional neural networks (cnn) is now being used in several image, pattern and signal processing researches. liu et al. [33] introduced au-aware deep networks (audn) by constructing a deep architecture for facial expression recognition. for extracting high level features from each au-aware receptive fields (aurf), they used restricted boltzmann machine (rbms). later, this technique was applied on three expression database namely ck+, mmi and sfew. results achieved from this technique were better or at least competitive. however, this method fails when several kinds of challenging images (e.g., the subjects have higher expression non-uniformity, most of them have moustache and wear accessories such as glasses) are appeared. krizhevsky et al. [7] proposed a new cnn architecture which achieved top-1 and top-5 error rates of 37.5% and 17.0% on the test data. however, there is still an open issue that, if a single convolutional layer is removed, network’s performance is degraded. here, authors did not use any unsupervised pretraining data to simplify this work but it could be more helpful if the computational power and size of the network were increased. dey et al. [6] used deep learning model in texture based garments design classification. in their experiment, using berkeley-trained model [10], they obtained advances in technology innovation, vol. 2, no. 4, 2017, pp. 119 125 121 cop y right © taeti 73.54% accuracy in clothing attribute dataset. however, they claimed that accuracy might be improved by changing layers and other related issues. zhoub et al. [8] proposed a technique which extracted the difference between the density and diversity of image datasets. here, authors used cnn to learn deep features for scene recognition tasks. for their dataset vgg s-16 models achieved 88.8% accuracy in top-5 val/test. however, there exist some difficulties such as the variability in camera poses, decoration styles or the objects that appear in the scene. lao et al. [27] used convolutional neural network for fashion class identification. authors divided their work into four parts those are multiclass classification of clothing type; clothing attribute classification; clothing retrieval of nearest neighbours; and clothing object detection. for this work they used apparel classification with style (acs), clothing attribute (ca) and colourful-fashion (cf) datasets and found 50.2% and 74.5% accuracy for clothing style classification and clothing attribute datasets. hu et al. [28] used deep convolutional neural networks for high-resolution remote sensing (hrrs) scene classification. for this work, they proposed two models for extracting cnn features from different layers. authors also used convolutional feature coding scheme for aggregating the dense convolutional features into a global representation. their proposed two models achieved remarkable performance and improved the state-of-the-art by a significant margin. for garments design class identification many approaches have been proposed. but, there are only a few works that have been conducted based on deep learning. this research has experimented different deep learning methods for identifying different garments design class based on textures. 3. methodology this section describes the methodology for identifying the garments design classes. basic steps of the procedure are shown in fig. 1. input images are firstly segmented and classified into several classes based on their texture design. after that, these images are separated for training, validation and testing from each of the class. proposed model is then applied alongside with two well-known deep convolutional neural network (cnn) models alexnet and vgg_s in two different garment datasets for the purpose of training and testing. finally, the accuracy of proposed system is compared with the existing models. we have also compared the results with traditional state-of-the-arts hand-engineered feature extraction method. alexnet and vgg_s have been chosen in this work because of their computational simplicity and better performance in several areas. they work well on unsupervised dataset. these two models can handle over-fitting problem when working with large dataset by using data augmentation technique. besides, these two models use a recently-developed regularization method called "dropout" that is proven to be very effective. these two models gained significant results in challenging benchmarks on image recognition and object detection. brief descriptions about these two models alongside our proposed model are described in the following sub-sections. fig. 1 basic steps of our working procedure fig. 2 the full architecture of alexnet model 3.1. alexnet model alexnet model was proposed by krizhevsky et al. [7]. there are three types of layer in a deep convolution neural network; such as convolution layer, pooling layer and fully-connected (fc) layers. full architecture of alexnet model was created by combining these three layers. in this architecture, there are total eight learned layers: five convolutional layers and three fully connected layers. convolution layer is the core building block and each of those convolution layer consists of some learnable filters. filters size are different from one another. full alexnet archit ectural model is shown in fig. 2. first convolution layer takes the input images by resizing each of the images into 224×224 with 96 kernels. the second layer takes the input from first convolution layer with 256 kernels after passing through a pooling layer. pooling layer operates independently and reduce the amount of parameters and computation in the network. hence, control the over-fitting problems. in this architecture, the third, fourth and fifth layers are connected to one another without any connection of pooling layers. the third layer consists of 384 kernels which takes input from the o utput of second layer. the fourth layer has 384 and fifth layer contains 256 kernels. each of last three fully connected layers contains 4096 neurons. the output of the last fully connected layer is sent as input to a 1000 way softmax layer which produces a distribution over the 1000 class labels. here, multinomial logistic regression is also used for maximizing the training cases. 3.2. vggnet model chatfield et al. [9], based on caffe toolkit proposed three different architectures of deep cnn models: vgg_f, vgg_m and vgg_s; each of which explores a different speed/accuracy trade-off: (1) vgg_f: th is cnn arch itecture is almost simila r to ale xnet. but vgg_ f contains sma lle r nu mber of filters and small stride in some convolutional layers. (2) vgg_m: it is a mediu m size cnn which is very similar proposed by zeiler et al. [30]. the 1 st convolution layer of this network has s maller stride and pooling layer. 4th convolution layer use smalle r numbers of filters for balancing the computational speed. (3) vgg_s: th is architecture is relat ively slow than vgg_f and vgg_m and it is a simp lified version of accurate model in the over-feat fra me work wh ich has six convolutional layers. fig. 3 shows the full architecture of vgg_ s mode l. it has taken the first five layers fro m the original model and has a sma lle r number o f filters in 5th layer. it has large pooling size in 1st and 5th convolutional layer than vgg_m. this model has been used to evaluate the garments design advances in technology innovation, vol. 2, no. 4, 2017, pp. 119 125 122 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti class identification. as depicted in fig. 3, this vgg_ s model contains five convolution layers with s ma lle r number of filters in the 5th layer and three fully connected layers. there are another two models based on vggnet na me ly vgg-vd16 and vgg-vd19. between a le xnet and vgg_s mode ls, the ma in diffe rence is that vgg_s model has sma ll stride in some convolutional layers and pooling size is large attached with the 1st and 5th convolutional layer. here , fully-connected layers 6 and 7 are regularized using dropout and the last layer acts as a mult i-way s oft-ma x classifier. fig. 3 the full architecture of vgg_s model fig. 4 the full architecture of our proposed model 3.3. proposed transferred cnn for classifying garments design class, a new scenario has been proposed in this paper based on alexnet, to observe the performance and effectiveness of deep features by total nine learned layers. among these layers five of these layers are convolutional layer and remaining four are fully connected layers. like alexnet, first convolution layer of proposed model takes the input images by filtering each of the images into 224×224 size with 96 kernels. the second layer takes the input from first convolution layer after passing through a pooling layer. pooling layers are added after first, second and fifth convolution layer like alexnet. a new fully connected layer (fc3) which takes input from the output of second fully connected layer (fc2) has been added in this proposed model. output of the last layer (fc4) is connected to a softmax layer for classifying the categories. the proposed model used data augmentation technique to reduce overfitting in the training stage. because, recent works show that data augmentation also helps to improve classification performance [7]. the full architecture of the proposed model is shown in fig. 4. 3.4. datasets fig. 5 exa mp le of clothing attribute dataset: column 1 to 6 represents e xa mple of flora l, graphics, pla id, solid color, spotted and stripe respectively fig. 6 exa mp le images fro m fashion dataset: each of the row represents jeans, leather, print, single co lor and stripe category respectively two publicly available datasets : fashion [31] and clothing attribute datasets (cad) [32] that was originally created for garment product recogn ition have been considered for this research. from fashion dataset, 5400 images are manually selected and categorized into five design classes, namely “single color” (2440 images), “print” (1141 images), “stripe” (565 images), “jeans” (614 images) and “leather” (640 images). again from clothing attribute dataset; 1575 images and manually selected and categorized into six different categories; after segmenting garments area from the original dataset. the categories are “floral” (69 images), “graphics” (110 images), “plaid” (105 images), “spotted” (100 images), “striped” (140 images) and “solid” advances in technology innovation, vol. 2, no. 4, 2017, pp. 119 125 123 cop y right © taeti pattern (1051images). original cad contain 1856 different images with 26 ground truth clothing attributes such as necktie, color, pattern etc. fig. 5 and fig. 6 show some sample images from “fashion” and “clothing attribute” datasets used in our work. table 1 describes proper training and validation samples about clothing attribute and fashion datasets. for clothing attribute dataset, different training and validation samples has been used; such as 10, 20, 30 images per class for training and validation, and rest of the images for testing to identify the classification results. in fashion dataset, 60, 100, 200 and 300 images are used for training and 10, 10, 20 and 30 images for validation and rest of the images for testing respectively . table 1 dataset used for e xperiments sample with different training and validation samples databases clothing attribute dataset (cad) fashion dataset classes 6 5 total samples 1575 5400 training sample/class i)10 ii) 20 iii) 30 i) 60 ii) 100 iii) 200 iv) 300 validation sample/class i) 10 ii) 20 iii) 30 i) 10 ii) 10 iii) 20 iv) 30 4. experimental result this section describes the experimental detail and divided into two sub-sections. first sub-section discusses about the implementation environment and next one describes the results. 4.1. implementation environment experimentation environment for this research has been set by following a straightforward process. we fine-tuned the caffenet [29] model and use ubuntu 12.4 operating system. this research considered high speed gpu for making the computation faster. because cpu is nearly ten times slower than gpu for working with large datasets and complex cnn. nvidia geforce gtx 950 4gb gpu and intel core i7 processor has been used for faster training and testing. 4.2. experimental result and discussion this research mainly experimented on two existing deep convolutional neural network models alongside with the proposed model on fashion dataset and clothing attribute dataset. performance of the proposed deep learning model has been compared with the existing models and also with some existing well-known hand-engineering feature extraction approaches for garment design class identification. different training, validation and testing sample from two different datasets have been used and shown in table 1. the training and testing results of alexnet, vgg_s and proposed model are provided in table 2, table 3, table 4, and table 5. these accuracies are calculated based on the training, validation samples/class used for each dataset. from table 3 and fig. 7 it can be found that in most of the cases vgg_s performs better than alexnet model. table 2 recognition rate (%) in training phase of cad dataset models training s ample validation s ample results cad with 6 classes alexnet 10 10 75.1 20 20 75.6 30 30 75.5 vgg_s 10 10 76.2 20 20 76.5 30 30 76.6 prop osed m odel 10 10 77.2 20 20 77.3 30 30 77.7 table 3 recognition rate (%) in testing phase of cad dataset models training s ample validation s ample results cad with 6 classes alexnet 10 10 75.3 20 20 75.5 30 30 75.6 vgg_s 10 10 76.5 20 20 76.4 30 30 76.8 prop osed m odel 10 10 77.1 20 20 77.4 30 30 77.8 table 4 recognition rate (%) in training phase of fashion dataset dataset models training s ample validation s ample results fashion dataset with 5 classes alexnet 60 10 74.3 100 10 75.9 200 20 78.1 300 30 81.5 vgg_s 60 10 75.3 100 10 76.8 200 20 78.6 300 30 82.7 prop osed m odel 60 10 76.6 100 10 78.1 200 20 81.1 300 30 84.1 table 5 recognition rate (%) in testing phase of fashion dataset dataset models training s ample validation s ample results fashion dataset with 5 classes alexnet 60 10 74.8 100 10 76.6 200 20 79.1 300 30 81.8 vgg_s 60 10 76.1 100 10 77.3 200 20 80.8 300 30 82.9 prop osed m odel 60 10 76.7 100 10 78.1 200 20 82.7 300 30 84.5 using clothing attribute dataset, alexnet and vgg_s model of cnn shows maximu m 75.6% and 76.8% accuracies respectively while our proposed model of cnn achieved 77.8% accuracy. on the other hand, using fashion dataset with 5 different classes, 81.8% accuracy has been achieved using alexnet and 82.9% accuracy using vgg_s respectively and advances in technology innovation, vol. 2, no. 4, 2017, pp. 119 125 124 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti our proposed model achieved 84.5% accuracy. from table 3 and table 5, it is clear that more training sample increase the accuracy. table 6 and table 7 describe the experimental results using seven different hand-engineered feature extraction methods which are hog, gist, lgp, centrist, tcentrist, ccent rist, and nabp on clothing attribute dataset and fashion dataset. for these methods support vector machine (svm) was used for classification purpose. table 6 experimental results of different methods for clothing attribute dataset method accuracy hog 63.76% gist 72.31% lgp 65.55% centrist 71.97% tcentrist 74.48% ccentrist 74.97% nabp 74.18% berkeley 73.54% alexnet (30) 75.6% vgg_s (30) 76.8% prop osed m odel 77.8% table 7 experimental results of different methods for fashion dataset method accuracy hog 79.15% gist 81.67% lgp 79.79% centrist 79.72% tcentrist 84.07% ccentrist 84.23% nabp 83.22% alexnet 81.8% vgg_s 82.9% prop osed m odel 84.5% table 6 also shows the result of three deep learning models berkeley, ale xnet, vgg_s along with our proposed model for clothing attribute dataset. from this table, it is c lear that performance of different deep learning models are better than any hand -engineering feature extraction method for clothing attribute dataset . table 7 shows that, for fashion dataset our proposed method performs better. though alexnet and vgg_s show slightly less accuracy than tcentrist, ccentrist and nabp. fig. 7 co mparison between ale xnet, vgg s and our proposed models for clothing attribute dataset fig. 8 co mparison between ale xnet, vgg s and our proposed models for fashion dataset 5. conclusion in this paper, some deep cnn models for identifying garments design class along with our proposed cnn have been used and also the results are compared with several hand-engineered feature extraction methods. using two different datasets , this proposed deep convolutional neural network with five convolutional layers and four fully connected layers shows better performance than some existing deep convolutional model as well as several hand-engineered feature extraction methods. fc layers and convolutional layers used in a deep cnn represent the features more elaborately, which are stronger than any of hand-engineered feature extraction techniques. 77.8% accuracy has been achieved on clothing attribute dataset with 6 different classes and 84.5% accuracy on fashion dataset containing 5 texture design categories using the proposed model. when a database contains more generic properties for every class, then a deep network can extract the generic features easily and accurately. it is mentioned earlier that the used datasets were manually categorized in different clothing product classes and used only a few numbers of classes. for this reason, the classes contain less generic properties most of the time. additional fc layer used in the proposed model helps the model to understand the features from these datasets more accurately. this research work will help other future researchers for choosing appropriate deep learning model for garments texture design classification. in future, we will try to improve the results by adopting more sophisticated strategies. acknowledgement this research is funded by bangladesh university grants commission, no: reg/prosha-3/2016/46889. references [1] j. wu and j. m. rehg, “centrist: a visual descriptor for scene categorization,” ieee transactions on pattern analysis and machine intelligence, vol. 33, no. 8, pp. 1489-1501, august 2011. [2] t. oja la, m . piet ika inen and t . maenp aa , “mu lt ireso lut ion g ray -sc a le and rot at ion inva riant te xtu re c lass ificat io n with loca l b in ary p atte rns, ” ieee t ransac t ions on patte rn ana lys is and mach in e intelligen ce , vol. 24, no. 7, pp. 971-987, 2002. advances in technology innovation, vol. 2, no. 4, 2017, pp. 119 125 125 cop y right © taeti [3] o. l. junior, d. delgado, v. gonçalves and u. nunes, “trainable classifier-fusion schemes: an application to pedestrian detection,” proc. 12th international ieee conference on intelligent transportation systems (itcs 09), ieee press, pp. 1-6, 2009. [4] t. x. yang and b. triggs, “enhanced local texture feature sets for face recognition under difficult lighting conditions,” ieee transactions on image processing, vol. 19, no. 6, pp. 1635-1650, 2010. [5] z. guo, l. zhang and d. zhang, “a completed modeling of local binary pattern operator for texture classification,” ieee transactions on image processing, vol. 19, no. 6, pp. 1657-1663, 2010. [6] e. k. dey, m. n. a. tawhid and m. shoyaib, “an automated system for garment texture design class identification,” computers, vol. 4, no. 3, pp. 265-282, 2015. [7] a. krizhevsky, i. sutskever and g. e. hinton, “imagenet classification with deep convolutional neural networks,” proc. advances in neural information processing systems, pp. 1097-1105, 2012. [8] b. zhou, a. lapedriza, j. xiao, a. torralba and a. oliva, “learning deep features for scene recognition using places database,” proc. advances in neural information processing systems, pp. 487-495, 2014. [9] k. chatfield, k. simonyan, a. vedaldi and a. zisserman, “return of the devil in the details: delving deep into convolutional nets,” arxiv preprint arxiv: 1405.3531, 2014. [10] p. heit, “the berkeley model,” health education, vol. 8, no. 1, pp. 2-3, 1977. [11] l. wang, c. y. lee, z. tu and s. lazebnik, “training deeper convolutional networks with deep supervision,” arxiv preprint arxiv: 1505.02496, 2015. [12] g. levi and t. hassner, “age and gender classification using convolutional neural networks,” proc. ieee conference on computer vision and pattern recognition workshops, pp. 34-42, 2015. [13] z. ge, c. mccool and p. corke, “content specific feature learning for fine-grained plant classification,” proc. clef (working notes), 2015. [14] k. yamaguchi, m. h. kiapour, l. e. ortiz and t. l. berg, “parsing clothing in fashion photographs,” proc. computer vision and pattern recognition (cvpr), pp. 3570-3577, 2012. [15] k. yamaguchi, m. h. kiapour and t. l. berg, “paper doll parsing: retrieving similar styles to parse clothing items,” proc. of the ieee international conference on computer vision, pp. 3519-3526, 2013. [16] c. shan, s. gong and p. w. mcowan, “facial expression recognition based on local binary patterns: a comprehensive study,” image and vision computing, vol. 27, no. 6, pp. 803-816, 2009. [17] x. feng, a. hadid and m. pietikäinen, “a coarse-to-fine classification scheme for facial e xpression recognition,” proc. international conference image analysis and recognition, springer berlin heidelberg, pp. 668-675, 2004. [18] e. simo-serra, s. fidler, f. moreno-noguer and r. urtasun, “a high performance crf model for clothes parsing,” proc. asian conference on computer vision, springer international publishing, pp. 64-81, 2014. [19] s. vittayakorn, k. ya maguchi, a. c. berg and t. l. berg, “runway to realway: visual analysis of fashion,” proc. ieee winter conference on applications of computer vision, ieee press, pp. 951-958, 2015. [20] y. kalantidis, l. kennedy and l. j. li, “ getting the look: clothing recognition and segmentation for automatic product suggestions in everyday photos,” proc. of the 3rd acm conference on international conference on multimedia retrieval, acm, pp. 105-112, 2013. [21] a. c. gallagher and t. chen, “clothing cosegmentation for recognizing people,” proc. computer vision and pattern recognition (cvpr 2008), ieee press, pp. 1-8, 2008. [22] l. bourdev, s. maji and j. malik, “describing people: a poselet-based approach to attribute classification,” proc. international conference on computer vision, ieee press, pp. 1543-1550, 2011. [23] s. arivazhagan and l. ganesan, “texture classification using wavelet transform,” pattern recognition letters, vol. 24, no. 9, pp. 1513-1521, 2003. [24] m. m. rahman, s. rahman, m. kamal, e. k. dey, m. a. a. wadud and m. shoyaib, “noise adaptive binary pattern for face image analysis,” proc. 18th international conference on computer and information technology (iccit), ieee press, pp. 390-395, 2015. [25] b. jun, i. choi and d. kim, “local transform features and hybridization for accurate face and human detection,” ieee transactions on pattern analysis and machine intelligence, vol. 35, no. 6, pp. 1423-1436, 2013. [26] s. la zebnik, c. schmid and j. ponce, “beyond bags of features: s patial pyramid matching for recognizing natural scene categories,” proc. ieee co mputer society conference on computer vision and pattern recognition (cvpr'06), ieee press, pp. 2169-2178, 2006. [27] b. lao and k. jagadeesh, “convolutional neural networks for fashion classification and object detection,” http://cs231n.stanford.edu/reports/blao_kjag_cs23 1n_finalpaperfashionclassification.pdf, june 26, 2016. [28] f. hu, g. s. xia, j. hu and l. zhang, “transferring deep convolutional neural networks for the scene classification of high-resolution remote sensing imagery,” remote sensing, vol. 7, no. 11, pp. 14680-14707, 2015. [29] y. jia, e. shelhamer, j. donahue, s. karayev, j. long, r. girshick, s. guadarrama and t. darrell, “caffe: convolutional architecture for fast feature embedding,” proc. 22nd acm international conference on multimedia, acm, pp. 675-678, 2014. [30] m. d. zeiler and r. fergus, “visualizing and understanding convolutional networks,” proc. european conference on computer vision, springer international publishing, pp. 818-833, 2014. [31] m. manfredi, c. grana, s. calderara and r. cucchiara, “a complete system for garment segmentation and color classification,” machine vision and applications, vol. 25, no. 4, pp. 955-969, 2014. [32] h. chen, a. gallagher and b. girod, “describing clothing by semantic attributes,” proc. european conference on computer vision, springer berlin heidelberg, pp. 609-623, 2012. [33] m. liu, s. li, s. shan and x. chen, “au-aware deep networks for facial expression recognition,” proc. 10th automatic face and gesture recognition (fg), ieee press, pp. 1-6, 2013.  advances in technology innovation, vol. 2, no. 4, 2017, pp. 105 112 105 numerical study of the operation of motorcycles covering the urban dynamometer driving schedule albert boretti * department of mechanical and aerospace engineering, benjamin m. statler college of engineering and mineral resources, west virginia university, morgantown, wv 26506, usa received 02 sept ember 2016; received in revised form 24 oct ober 2016; accept ed 25 oct ober 2016 abstract it is shown, for the most challenging case of a cruiser motorcycle of low weight-specific and displacement-specific power and torque, that the tuning for better top end perfo rmances is irrelevant for the operation over the driving schedule used for certification. during the certification test, the engine only operates in the low speeds and loads portion of the map. it is concluded that any statement about motorcycles’ pollution and fuel consumption should be only based on the measurement of their regulated emissions through proper chassis d ynamometer tests, possibly redefining the driving schedule to better represent real driving conditions. ke ywor ds : motorcycles, pollutant e missions, driv ing cycles , afterma rket tuners , real driving conditions 1. introduction the actual operation of motorcycles covering emission and fuel econo my cert ification cycles has been brought back to the attention of lawma kers, original equip ment manufacturers and the general public by the recent ban of harley-dav idson (hd) aftermarket electronic control units (ecu) tuners in the un ited states of a merica (us). the use of afterma rket ecu tuners does not necessarily translate in worse regulated pollutant e missions as othe rwise a lleged by the us environ mental protection agency (epa). actually, these devices are mo re likely not re levant to any claim concerning emissions. the ecu tuners are simple corrections of the fuel injection para meters to deliver air -to-fuel (afr) ratio that may increase throttle response, power/torque output and overall ride ability. their target is mostly the steady wide open throttle (wot) operation of the engine, i.e. the high load operation, as well as the high speed operation, plus the sharp accelerations. 5 years. they are not supposed to satisfy the emission rules of the time they were certified after the 5 years. therefore, it does not ma ke any sense to discuss the po llutant emissions of motorcycles older than 5 years with or without ecu tuners fitted. ho wever, the retuning of an old engine may in principle offe r the opportunity to introduce some improve ments rather than declines in performances, power and torque, as well as fuel econo my and pollutant emissions, as the old factory calibration does not necessarily represent the best choice of engine controlling p arameter on a specific motorcycle more than 5 years old. according to epa rules, new motorcycles are tested on a driving cycle , where the engine delivers the power needed for the motorcycle to fo llo w a low velocity schedule with everything but sharp accelerations and everything but high speeds. hence, engines able to deliver much larger power and torque outputs operate significantly far fro m high load and high speed during the cert ification cycle , changing their load and speed much slower than what they could do, and reaching top power and torque outputs very far from their theoretical maximu m. for a gasoline fue led motorcycle having a three -way catalytic converter (tw c), port fuel injection and an oxygen sensor feed back to the ecu, any change of the controlling para meters returning a c loser to stoichiometric air-fuel-ratio is not e xpected to translate in any worsening of the e missions. only operating the engine richer for increased power and torque output at higher loads and speeds may have pollutant emissions and fuel economy downfalls in these operating points. fig. 1 p resents the typical e fficiency map of a catalytic converter. the emission reduction of a typical port fue l injected, ho mogeneous charge, and gasoline engine is based on the effic ient operation of the twc that require a close to stoichiometry air-to-fuel ratio. th is is obtained by operating the fuel injectors to deliver a stoichio metric mixtu re as monitored by the e xhaust oxygen sen sor feed-back. around the stoichiometric point (a/f=14.63), all the three pollutants (hc, co and no) are a lmost totally re moved (>95 %).a slightly richer mixture translates in more co and hc but not no. a slightly leaner mixture translates in more no but not co and hc. the engine operation in super sport, touring and cruiser motorcycles covering the epa urban dyna mo meter driving schedule (udds, 40 cfr part 86, appendix i to part 86 dynamo meter schedules) will be considered in the paper. cruiser motorcycles are specific models designed with engines having low end specific perfo rmances, i.e . sma ll d isplacement specific torque and power, small we ight specific torque and power, lo w speed, if compared to touring and obviously super sport bikes. cru isers have large torques only because of the large displacement. *corresponding aut hor, email: a.a.boretti@gmail.com advances in technology innovation, vol. 2, no. 4, 2017, pp. 105 112 106 copy right © taeti copy right © taeti copy right © taeti copy right © taeti cop y right © taeti fig . 1 conversion curves for hc, co and no as a funct ion of the a ir/ fuel ratio , for a port fuel injected gasoline engine fitted with a twc removed (>95 % ) 2. epa motorcycles’ emission rules street motorcycles’ emissions are regulated under section 202 of the clean air act. bac kground information on emission rules for motorcycles sold in the us may be found in [1, 2]. table 1 (fro m [1]) su mmarizes the emission limits to be satisfied during chassis dynamo meter testing of the motorcycle. st reet motorcycles’ emissions were regulated by a single unchanging set of standards for all mode l years from 1978-2005. in 2004, epa established 2 tiers of conventional pollutant exhaust emissions standards. tier 1 ca me into effect in 2006. in 2010, standards for class iii motorcycles were updated to tier 2 standards. only class iii motorcycles having a displacement in e xcess of 279 c m 3 are considered here, as the street motorcycle market is mostly made by super sport and touring bikes. scooters are not considered. highway motorcycles exhaust emission standards only apply since 1978. befo re 1978 there we re no e mission standards a motorcycle was requested to comply with. the standards applied first to new gasoline fueled moto rcycles (since december 31, 1977). then, later on, the standards were also applied to new, methanol-fueled mo torcycles (since dece mber 31, 1989), to new, natural gas-fueled and liquefied petroleu m gas -fueled motorcyc les (since dece mber 31, 1996) and finally new motorcycles regardless of fuel (since 2006). the table also includes useful life and warranty period. they are e xp ressed in years and kilo meters, and whichever comes first limits the need of co mpliance. the term “useful life” [3] does not mean that a motorcycle must be scrapped or turned over to the government after ce rtain mileage limits are reached. it does not mean that a vehicle is no longer useful or that the vehicle must be scrapped once these limits are reached. the term has no effect on the owners’ ability to ride or keep their motorcycles for as long as they want. the cu rrent useful life for motorcycles with engines over 279 c m 3 is 5 years or 30,000 kilo meters (about 18,640 miles), whichever first occurs. the test procedures for motorc ycles fro m m y 1978 and later are detailed in 40 cfr part 86 subpart f. fig. 2 presents the cycle. this cycle is characterized by low speeds. fig. 2 udds velocity schedule http://www.intechopen.com/books/diesel-engine-combustion-emissions-and-condition-monitoring/nox-storage-and-reduction-for-diesel-engine-exhaust-aftertreatment#f1 advances in technology innovation, vol. 2, no. 4, 2017, pp. 105 112 107 cop y right © taeti table 1 emission standards in the us (from [1]) year class engine size (cm 3 ) hc (g/km) hc + nox (g/km) co (g/km) useful life warranty 1978-2005 i 50-169 5.0 12.0 5 / 12,000 5 / 12,000 ii 170-279 5 / 18,000 5 / 18,000 iii 280+ 5 / 30,000 5 / 30,000 2006+ i-a < 50 1.0 1.4 12.0 5 / 6,000 5 / 6,000 i-b 50-169 1.0 1.4 12.0 5 / 12,000 5 / 12,000 ii 170-279 1.0 1.4 12.0 5 / 18,000 5 / 18,000 2006-2009 iii (t ier 1) 280+ 1.4 12.0 5 / 30,000 5 / 30,000 2010+ iii (t ier 2) 280+ 0.8 12.0 5 / 30,000 5 / 30,000 3. hd clean air act settlement the u.s. epa and the u.s. depart ment of justice (doj) announced on august 18, 2016 a settle ment with hd co mpanies, that required the companies to stop selling and to buy back and destroy “illegal tuning devices that in-crease air pollution fro m their motorcycles”, and to sell only tuning devices that are cert ified to meet clean air act emissions standards. hd was also requested to pay a $12 million civ il penalty and spend $3 million on a project to mitigate a ir pollution through a project to replace conve ntional woodstoves with cleaner -burning stoves in loca l communities. epa alleges that hd vio lated the clean air act by manufacturing and selling about 340,000 devices, known as tuners that “allow users to change how a motorcycle’s engine functions”. according to epa “these changes can cause the motorcycles to e mit higher a mounts of certain air pollutants than they would in the original configuration that hd certified with epa”. according to epa, since january 2008, hd man ufac-tured and sold tuners that allow users to modify “certain aspects of a motorcycles’ e missions control system”. according to epa, these modified settings increase power and performance, but a lso increase the motorcycles’ emissions of hydrocarbons and nitrogen oxides (nox). the cla im o f vio lations is not based on any chassis dy-namo meter measure ments of the performances of motorcycles not having exceeded the useful life of 5 years or 30,000 km tested first without, and then with the kit fitted, to prove that a specific motorcycle model was not compliant because of the fitting of a specific kit. 4. street performance tuners the screa min' eag le street performance tuner is a performance engine manage ment system for e lectronic fuel inject ion (efi) equipped ha rley davidson models [5-7]. the kit utilizes a wide-band o xygen sensor feedback to provide continuous air-to-fuel rat io (afr) tuning corrections based upon riding conditions. the kit is a imed to deliver increased throttle response and torque, improved overall ride ability and performance, as well as a smoother and cooler running engine. in many cases, the kit helps improving fuel econo my, depending upon the bike’s configuration and the set -up of the afr targets. afr targets set to richer values than the stock levels to gain performance may result in moderate decrease in fuel econo my. the street tuner permits li mited tunability within the emissions range to optimize drivability without compro mising e mission, but it is o bviously intended to work outside the closed loop portion of the engine map where the afr is ensured to be about stoichiometric for the best operation of the three-way-catalytic converter. fig. 3 typical afr map of a large hd cruiser with a big v-twins engine advances in technology innovation, vol. 2, no. 4, 2017, pp. 105 112 108 copy right © taeti copy right © taeti copy right © taeti copy right © taeti cop y right © taeti a typical tuners fuel map of a large hd c ruiser is provided in [7] and reproduced in fig. 3. the engine is a big v-t wins engine (twin ca m 96, 96.96 cubic inch or 1,584 c m 3 ).these engines are characterized by much smaller specific torque and power density than the average super sport and touring bikes. ma ximu m power (but at the wheels, where it is typically 10-15% sma ller than at the crank) is only 68 hp @ 5,000 rp m, while ma ximu m torque (also at the wheel) is 110 n m @ 3,000 rp m. this engine powers motorcycles of 307.5 kg wet we ight including oil and gas. only above 80% map (roughly 50% throttle) and 4,000 rp m, where the engine does not operate during ty pical driving cyc les inc luding the cert ification cycle , the afr is made rich. diffe rent “stages” of tuning are considered in [7]. the stage 2 is an upgrade that includes new cams. the stage 1 is upgrade also requiring e xhaust and air c leaner. both these upgrades are valid 50-state legal modification. the air fue l map of [7] has the cells in red with the stoichiometric 14.6 afr in them over the low speed low load portion of the map that is re levant to the emission certification. both stage 1 and stage 2 tunings do not affect this area. the ecu runs in c losed loop mode looking at the oxygen sensor to satisfy the optima l co mposition of the e xhaust gases for the twc to reduce the tail p ipe e missions. in [7], the engine works c losed loop to 3,750 rp m and to 80 kpa of manifo ld absolute pressure (map). the ecu uses manifo ld pressure from the map sensor to determine the actual engine load rather than the throttle. throttle position does not relate linearly to the map sensor reading. 80 kpa map is typically a round 40% throttle. ref. [7] assumes the oem afr table is s ame or very similar of fig. 3, but may obviously differs outside the 3,750 rp m and 80 kpa area. the refore, tuners are possibly delivering same afr vs. map and speed of the oem in the low map and low speed area of emissions’ control, and then they differ. it is worth to mention that usually steady state afr maps do not need fuel rich conditions except than a pproaching wot conditions, i.e. close to the ma ximu m loads for any speed. the fuel rich mixture at speed e xceeding 3,750 rpm any load seems quite questionable. the tuner operates rich everywhere out of th e closed loop area very like ly because the injection system is ev erything but effective in delive ring the amount of fuel needed when the throttle opens sharply. even racing engines these days go rich only approaching wot cond itions at any speed, as even these extre me engines run slightly lean part load to reduce unnecessary fuel co nsumption. in addition to the afr map, ref. [7] a lso provides the bias tables and the ignition advance map for the hd stage 1 and stage 2 bikes. it is not the object of the paper to enter more in details of the specific tunings, only to show in the next section how the operation of a motorcycle over a driving schedule for e mission certification never utilizes the high loads or high speeds parts of the map that are the ultimat e goal of tuning an engine for mostly imp roving power and torque output. 5. method map based computer models are used to investigate the operation of an engine when the motorcycle is covering a driving schedule. vehicle driv ing cycle simulat ions have been around for many years. basic solutions of the new-ton’s equation of motion fo r a vehicle following a pre-scribed velocity schedule returns the instantaneous power requested to the engine with a simp lified modelling of transmission losses, aerodynamic and rolling resistance, and vehicle and engine inertia. trans mission ratios then also return the speed requested to the engine. interpolating the steady state maps of brake specific fuel consumption or specific e missions, it is then possible to evaluate the fuel consumption and the pollutant emissions on a driving cycle. for cold start, correction curves are needed. for the interested reader, these simulat ions are presented in [12-24]. to simplify, a driving cycle simulator solves the newton’s equation of motion. if fp,e is the engine propulsive force and fb,f is the friction brake force, it is: fp,e-fb,f-fa-fr= m∙a (1) with m the mass, a the acceleration, =dv/dt, with v velocity of the motorcycle and t the time , fa the ae rodynamic drag force, =½∙ρ∙v 2 ∙cd∙a, with ρ a ir density, cd drag coefficient (a lways positive for a retard ing force) and a re fe rence area, fr the rolling resistance force, an e mp irica l function of the speed of the motorcycle. in terms of powers, by multiplying for the speed of the motorcycle, it is then pp,e -pb,f = m∙v∙dv/dt +½∙ρ∙v 3 ∙cd∙a+pr (2) the above propulsive power is computed at the wheel. the power of the engine at the crankshaft pb is larger than the power at the whee l pp,e to include the transmission efficiency η. the speed of rotation of the engine is then obtained by the speed of the motorcycle by considering tire radius, gear and gear ratios. the gear is determined by an upshift/downshift strategy. fro m a ve locity schedule v(t), it is thus possible to compute the instantaneous power pp,e and pb,f, and fro m pp,e, then the power pb and the speed n that the engine must provide. when m∙v∙dv/dt +½∙ρ∙v 3 ∙cd∙a+pr ≥0 , equation (2) returns pp,e with pb,f=0. when m∙v∙dv/dt +½∙ρ∙v 3 ∙cd∙a+pr <0, equation (2) returns pb,f with pp,e=0. pb,f represents in this case not only the actual power dissipated in the friction brakes p * b,f , but also the negative power requested to motor the engine at the given speed n (engine brake). engine performances are typically defined in terms o f power pb, torque tb and brake mean effect ive pressure bm ep. the power pb is proportional to the product of torque tb and speed n. the bm ep is proportional to the ratio of torque tb and total displaced volume vd. engine data are provided as the wide open throttle torque output tb vs. speed n, plus the maps of specific fue l consumption and pollutant emissions vs. bm ep and n. th is way the driving cycle simu lator returns the fuel economy and the pollutant emissions during warmed -up cycles, with e mpirical penalty functions needed for cold -start cycles. advances in technology innovation, vol. 2, no. 4, 2017, pp. 105 112 109 cop y right © taeti the model simu lates a motorcycle perfo rming a test cycle. the udds is considered. the cycle is everything but aggressive, and it is characterized by mostly low speed. in the udds cyc le, fig. 2, only in one of the acceleration, cruise and deceleration schedules it is requested a bike velocity of 90 km/h, and in only 3 other areas the bike reaches a speed above 50 km/h but less than 60 km/h. figs. 4, 5 and 6 present reference data of bm ep, torque and power vs. engine speed and throttle opening % for the typica l large cruiser considered here, having a lo w displacement specific powe r and torque, lo w ma ximu m speed. the engine is 1,300 c m 3 and it is fitted on a heavy motorcycle of weight 380 kg inc luding the driver during the simulated chassis dynamometer test. fig. 4 typical brake mean effective pressure map of a large cruiser motorcycle fig. 5 typical torque map of a large cruiser motorcycle fig. 6 typical power output map of a large cruiser motorcycle advances in technology innovation, vol. 2, no. 4, 2017, pp. 105 112 110 copy right © taeti copy right © taeti copy right © taeti copy right © taeti cop y right © taeti table 2 model parameters rated engine speed 6,500 rpm upshift 4,000 rpm downshift 2,000 rpm ratio of 1 st gear 9.312 ratio of gear 2 6.421 ratio of gear 3 4.774 ratio of gear 4 3.926 ratio of gear 5 3.279 ratio of gear 6 2.79 motorcycle weight 377 kg engine power at rated speed 55 kw tire rolling radius 457.2 mm tire rolling resistance factor 0.0122 engine displacement 1,304 cm 3 engine inertia 0.05 kg-m 2 frontal area 0.2 m 2 coefficient of drag 0.6 wheelbase 2 m initial engine speed 1,500 rpm initial gear number 1 table 2 presents the relevant model para meters, id le speed, rated engine speed and engine power at rated speed, upshift and downshift speed, that may differ at every gear, ratios of 1 st to 6 th gear (if a 6 gear transmission is considered as it is in this case), motorcycle we ight, tire ro lling radius, tire ro lling resistance factor, engine displacement, engine inertia, frontal a rea, coeffic ient of drag, whee lbase, initial engine speed and gear number. 6. results and discussion the engine map bm ep (bra ke mean effect ive pressure) vs. engine speed at different loads is the one of fig. 4, where the load is expressed in terms of acceleration pos ition (ap). fig. 7 presents the computed operating points, while fig. 8 presents the computed time d istribution on engine map of the operating points of a large cruiser motorcycle covering the udds cyc le. the engine operates below 2.5 bar bm ep and belo w 4,000 rp m over the cycle. every map point above these values has time d istribution ze ro, i.e . whatever could be the e mission in these points, and this has no effect on the regulated emissions. the most part of the time the engine is idling. then, when delive ring an output, the engine is always operating well be low 2.5 bar bmep and 3,750 rpm. fig. 7 typical operating points of a large cruiser motorcycle covering the udds cycle fig. 8 typica l t ime distribution on engine map of the operating points of a large cru iser motorcycle covering the udds cycle advances in technology innovation, vol. 2, no. 4, 2017, pp. 105 112 111 cop y right © taeti in terms of performances, today’s super sport, touring and cruiser bikes may have very high specific power and torque densities. as hd does not provide online info rmat ion about power and torque figures, typical perfo rmance parameters are proposed for other manufacturers. the 998 c m 3 ya maha yzf r-1 [9], one of the most powerful super sport bikes, has for e xa mp le 200 hp/ liter revving 13,500 rp m. the specific torque is less e xceptional, as the result of the tuning for high speeds , but still 112 n m/ lite r revving 11,500 rp m. the wet we ight including full oil and fue l tank is 199 kg. th is bike has a top speed of 300 km/h. as an e xa mp le of touring bikes, the 1,298 c m 3 ya maha fjr1300a [10] has 112 hp/liter revving 8,000 rpm and 106 n m/ liter revving at 7,000 rp m. the wet weight (inc luding full oil and fuel tank) is 289 kg. this bike has a top speed of 245 km/h. finally, as a typica l cruiser, the 1,304 c m 3 ya maha xvs1300 custom [11] has 56 hp/ liter revving 5,500 rp m and 79 n m/ liter rev ving 3,000 rp m. the wet we ight including full oil and fuel tank is 293 kg. this b ike has a top speed of 175 km/h. there fore, in norma l driving correctly accounted for e mission reg ulations, motorcycles work very far from their potentials. the most part of the motorcycles in the super sport and touring classes are usually mo re performant than the cruisers. they have much larger power and torque to weight ratio, as they are much lighter, and also have much larger displace ment specific powe r and torque. the most part of the super sport and touring motorcycles are therefore working even fa rther away fro m their h ighest speed and highest load points where they may operate off-stoichio metry during typical driving cycles including the udds e mission cycle. the resu lts proposed in the previous section are therefore a worst case scenario. 7. conclusions it is pure speculation to cla im that ecu tuners can cause the motorcycles to e mit higher a mounts of certain air pollutants than they would in the orig inal cert ified co nfiguration without even mentioning the specific motorc ycle where the tuners are fitted. in princip le, ecu tuners are not expected to affect any regulated emission. if fitted to motorcycles having e xceeded the useful life , presently defined as 5 years or 30 ,000 kilo meters (about 18,640 miles) whichever first occurs, as these motorcycles are not presently expected to comply with any emission rule, having or no the tuners makes no difference. for new motorcycles, the ecu tuners are expected to modify the afr only at the higher loads and speeds that are very far fro m the area of operating points that are d esigned closed loop stoichiometric, to comply with the emission rules properly using the twc. old and new motorcycles cannot be claimed a-priori not compliant without providing any evidence of failure to perform as required by regulation, and obviously they cannot be claimed not compliant if there is no rule to comply with. any statement about motorcycles’ pollution and fuel consumption should be only based on the measurement of their regulated emissions through proper chassis dynamometer tests. the results emphasize the importance of rea l world driving in motorcycles. the paper shows that the ecu tuners have no effect on the presently regulated pollutants emission, even if modifying the afr certa inly lead to change in e mission performance of the vehicles. while the ecu tuners may not affect the pollutants emission under well-constraint laboratory certification tests, they certainly change the emissions over real world driving. the paper therefore e mphasizes the importance of the inclusion of real world driving in emission certification tests. the introduction of better e mission certification tests will ultimately translate in superior fuel conversion efficiencies of the internal co mbustion engine over the full range of loads and speeds, for e xa mp le also simply adopting jet ignition and direct injection [25], plus the hybridizat ion of the powe r tra in, for e xa mp le with a flywheel or a li-ion battery based kinetic energy recovery system [26]. references [1] “us epa light -duty veh ic les a nd t ruc ks e miss ion standa rds,” h ttps :// www.ep a .gov/ e miss ion -stan da r ds-re fe ren ce -gu ide/ light -duty -v eh ic les -and -t ruc ks emission-standa rds , retrieved august 31, 2016. [2] “t ranspo rtpo lic y.n et us: moto rc yc les : emissions ,” htt p :// t ranspo rtpo licy .n et/ inde x.p hp? t it le =us:_ m otorcy c les :_ emiss ions , ret riev ed august 31, 2016. [3] “us epa in fo rmat ion fo r moto rcy c le o wn e rs ,” https :// www3.epa .go v/ot aq/ regs/ ro adb ike/ 420f030 45.pdf, retrieved august 31, 2016. [4] “us epa ha rley -dav idson c le an a ir act sett le ment ,” htt ps:// www.epa .go v/en fo rce me nt/ ha rley -da v idso n-c lean -a ir-a ct -sett le ment , retrieved august 31, 2016. [5] “ha rley -dav idson ca nada sc re a min ' eag le ® st re et pe rfo rmanc e tun e r kit,” ht tp ://a ccessories.ha rley -d av idson.ca/p rodu ct/sc re a min -eag lesup/sup-st re et -p erfo rmanc e -tune r-kit/ 41000008b, re trieved august 31, 2016. [6] “pro supe r tune r, ha rley da v idson supe r tun er wbt course ,” http ://p rosupe rtune r.ha rle y-d av idson.co m/ training/enu/ inde x.ht ml, retrieved august 31, 2016. [7] “tune you r h a rley moto rcy c le pe rfo rmanc e gu ide , ha rley -dav idson c losed loop ecm fu e l map reve a l ed,” http ://t uneyou rha rley.co m/b iket ech/ content/h a rley -d av idson-c losed -loop -e c m-fu e l-map -revea led , retrieved august 31, 2016. [8] b. pe reda -ayo , and j. r. go n zále z-ve lasco, “ nox storage and redu ct ion fo r d iese l eng in e e xh aust a ft er t re at ment ,” diese l eng in e co mbust ion , emiss i ons and cond it ion mon ito ring, dr. s. bari ed., 2013. [9] “ ya maha yz f -r1 2016,” https :// www.y a maha -mo tor.eu/u k/p roducts/ moto rcy c les/supe rsport /y zf -r1.a sp x? v ie w=featu restec hspecs# 2a d ejc o vkxka j v4 4.99, retrieved august 31, 2016. [10] “ ya maha fjr1300s 2016,” https :// www.y a maha -m otor.e u/u k/p rodu cts/ moto rcy c les/spo rt -tou rin g/ fjr1 300a.asp x, retrieved august 31, 2016. [11] “ ya maha xvs1300 c usto m 2016,” ht tps:/ / www.y a ma ha -moto r.eu/u k/p rod ucts/ mo to rcy c les/c ru ise r/ x vs1300cu .asp x? v ie w=fe atu rest echspe cs#vb 9p o fy 3 z43xy20q.99, retrieved august 31, 2016. https://www3.epa.gov/otaq/regs/roadbike/420f03045.pdf https://www3.epa.gov/otaq/regs/roadbike/420f03045.pdf https://www.epa.gov/enforcement/harley-davidson-clean-air-act-settlement https://www.epa.gov/enforcement/harley-davidson-clean-air-act-settlement http://prosupertuner.harley-davidson.com/training/enu/index.html http://prosupertuner.harley-davidson.com/training/enu/index.html advances in technology innovation, vol. 2, no. 4, 2017, pp. 105 112 112 copy right © taeti copy right © taeti copy right © taeti copy right © taeti cop y right © taeti [12] b. jacobson , “ on veh ic le d riv ing cyc le simu lat ion ,” soc iet y o f auto mot ive eng in ee ring (sa e) tec hn i cal paper 950031, 1995. [13] b. banda ru , l. rao, p. babu , k. va rathan , j. ba laji, “ rea l ro ad t rans ient d riv in g cyc le simu lat ions in eng ine testbed fo r fu e l econ o my p red ict ion ,” so ciety o f auto mot ive eng inee rin g (sa e) te chn ic a l paper 2014-01-2716, 2014. [14] n. de mbski, g. rizzon i, a . so lima n, j. frave rt , k. kelly, “ dev e lop ment of re fuse ve h ic le d riv ing and duty cy c les ,” so c iety o f auto mot ive eng ine ering (sae) technical paper 2005-01-1165, 2005. [15] m. m onta ze righ, h. va rasteh and m . nagh izadeh , “driv ing cyc le s imu lat ion fo r h eady du ty eng in e e mission eva luat ion and test ing ,” soc iety o f au to mot iv e eng in ee ring (sa e) te chn ic a l pape r 2005-01-3796, 2005. [16] s. t ra jkov ic , p. tunesta l and b. joh ansson, “ ve hic le d riv ing cyc le simu lat ion o f a pneu mat ic h y brid b us based on e xpe rime nta l eng ine measu re ments ,” soc ie ty o f a uto mot ive eng ine ering (sa e) technical paper 2010-01-0825, 2010. [17] j. dabad ie , p. mene ga zzi, r. t rigu i and b. jean neret, “a ne w too l fo r adv anced veh ic le s imu l a tions ,” soc iety o f auto mot iv e eng inee rin g (sa e) technical paper 2005-24-044, 2005. [18] a. bo re tt i, a. os man and i. a ris , “ des ign o f ran kin e cyc le syste ms to d e live r fue l econo my b ene fits over co ld sta rt d riv ing c yc les ,” soc iety o f auto mot ive eng inee rin g (sa e) t echn ic a l pape r 2012-01-1713, 2012. [19] r. a llen and t . rosentha l, “mee t ing impo rt ant cuing requ ire ments with modest, re a l-t ime , inte ract ive d riv ing simu lat ions, ” soc iety of auto mot iv e engineering (sae) technical paper 940228, 1994. [20] l. d. rag ione and g. m ecc arie llo , “th e eva luat ion of a ne w kin e mat ic e missio ns mode l on rea l and simu lated driv ing cyc les ,” sa e int . j. fue ls lub r. , vol. 3, no. 2, pp.521-531, 2010. [21] f. an, m . ba rth and g. sco ra , “ impa cts o f d ive rse driv ing cy c les o n e lect ric and hyb rid e lec tric v e hic le pe rforma nce ,” soc iety o f auto mot iv e eng ineering (sae) technical paper 972646, 1997. [22] y. jo, l. bro mbe rg and j. hey wood , “ octan e re quire me nt o f a turbocha rg ed spa rk ign it ion eng in e in va rious d riv ing cy c les ,” soc iety o f auto mot iv e eng ine ering (sa e) te chn ica l pape r 2016-01-0831, 2016. [23] h. wa ng, x. li, g. zh eng, y. fang , “study o f a hybrid refuse t ruc k wit h c ity driv ing cy c les ,” so ciety o f auto mot ive eng inee rin g (sa e) te chn ic a l paper 2014-01-1800, 2014. [24] c. hong and c. yen , “ driv in g cyc le t est s imu lat ion fo r passenge r ca rs and moto rcyc les ,” soc iety o f auto mot iv e eng ine e ring (sa e) t echn ic a l pape r 970274, 1997. [25] a. bo rett i, “ nu me rica l mod e ling o f a jet ign it ion dire ct in ject ion (jidi) lpg eng ine ,” int e rnat ion a l jou rna l o f eng inee ring and te chno logy innov at ion , vol. 7, no.1, pp. 24-38, 2017. [26] a. bo rett i, “ co mpa rison of rege ne rat iv e b ra king effic ienc ies o f m y2012 a nd m y2013 nissan lea f,” inte rn at iona l jou rn a l o f eng ine ering and t ec h nology inn ovat ion , vo l. 6, no . 3, pp . 214-224, 2016.  advances in technology innovation, vol. 4, no. 1, 2019, pp. 44 57 new ventures, internationalization, and asymmetric grin curve: analysis of taiwan’s big data jwu-rong lin 1 , chen-jui huang 2,* , ching-yu chen 1 , huey-ling shiau 2 , yan-chen yeh 1 1 department of international business, tunghai university, taichung, taiwan, roc 2 department of finance, tunghai university, taichung, taiwan, roc received 02 march 2018; received in revised form 18 july 2018; accepted 04 september 2018 abstract under globalization, small and medium enterprises (smes) that predominate taiwan’s economy have been primarily original equipment manufacturers (oems) continuing to adjust operating strategies in order to extend supply chains and enhance competitiveness. this paper adopts the big data composed of 104,377 taiwanese manufacturers from the 2011 industry, commerce, and service census to assess impact of the business life cycle, brand revenue, r&d spending, and internationalization on value creation. major findings are as follows. first, the link of the firm’s operating years with value creation is characterized by a quadratic u-shaped curve where the minimum point corresponds to 15 years of operation, suggesting a cost of lower value added for new ventures at the early stage of development. second, a reversed u-shaped curve of value creation is found as regards brand revenue and r&d spending, with the greater impact of the latter. third, the impact of overseas investment and export expansion is also captured by a reversed u-shaped curve, with greater impact for the former. fourth, an asymmetric grin curve rather than a smile curve is found in taiwan’s manufacturers, whose value creation can be strengthened by strategies that focus on the learning curve, internationalization, internet, operating scale, and capital intensity. keywords: new venture, internationalization, grin curve, value creation 1. introduction the emergence of the regional comprehensive economic partnership (rcep), which enlarges the trade link of the ten asean members and six asian countries, marks a new challenge as regards the cost advantage that most original equipment manufacturers (oems) in taiwan have benefited over past decades. these firms are now forced to develop strategies that transform the current cost-based industry into the one with high value added through new ventures, brand benefit, r&d spending, and internationalization. however, the 2011 industry, commerce, and service census conducted by taiwan’s directorate-general of budget, accounting, and statistics shows that the business operating years ranges from 1 to 100 and averages at 18 among the 104,377 firms surveyed. this seems to imply significant discrepancy in terms of the business life cycle. in addition, the number of firms that have been engaged in brand, r&d, overseas investment, and export activities are only 12,440, 10,062, 14,326, and 17,206, respectively accounting for 11.92%, 9.64%, 13.73%, and 16.49% of the total sample and leaving the ratio of the value added to total revenue at 36% only. this paper intends to deepen the issue on insufficient value creation observed in most small and medium enterprises (smes) that dominate taiwan’s manufacturing sectors and discuss relevant strategies for transformation in the business model. * corresponding author. e-mail address: cjhuang@thu.edu.tw. advances in technology innovation, vol. 4, no. 1, 2019, pp. 44 57 45 the remainder of this paper is structured as follows. section 2 succinctly reviews theoretic foundations and relevant empirical evidence. section 3 proposes the empirical model and hypotheses to be tested. section 4 analyzes primary regression results and checks robustness of the empirical model. section 5 concludes with discussion on research limitations and suggestion for future studies. 2. theoretic foundations and literature review the smile curve characterizes the relation between value creation and supply chains. the value added is expressed on the vertical axis, whereas the supply chains from the upstream to the downstream are expressed from the left to the right on the horizontal axis. over the supply chains, upstream, midstream, and downstream firms respectively play the role of original design manufacturers (odms), original equipment manufacturers (oems), and own branding manufacturers (obms). in less developing countries (ldcs), the smile curve appears reversed, called the forced smile curve. in between, the grin curve raises the two ends of the smile curve and appears flatter to reflect the evolution in industry competition. the typical smile curve, forced smile curve, and grin curve are illustrated in fig. 1. fig. 1 smile curve, forced smile curve, and grin curve new ventures have grown rapidly over the past decades. lussier [1] defines the new ventures as businesses established within 10 years. in taiwan, the start-up and incubation center of the small and medium enterprise administration of the ministry of economic affairs adopts instead a definition of 5 years, also applied to associated start-up loans open to the youth. but its business start-up award targets businesses of an age of not more than 3 years. across relevant studies, businesses established within 3 to 14 years exhibit essential characteristics of new ventures. as this research employs data from the industry, commerce, and service census conducted every 5 years by taiwan’s directorate-general of budget, accounting, and statistics, subsequent analysis will define new ventures by an age of not more than five years. businesses that have continuously operated over more than 5 years are regarded as survivors. marco et al. [2] examine a sample of italian companies between 1982 and 1992. firm characteristics before and after the public initial offerings (ipo) are compared in order to find determinants for listing on exchanges. the authors adopt the competitive theory to distinguish cost-side and benefit-side determinants and classify new ventures and survivors by firm age. the empirical evidence indicates lower capital demand reflected by a lower debt ratio for new ventures as these businesses have shorter operations and/or funds provided by establishing shareholders are able to meet the demand. erel et al. [3] turn to a larger sample of european companies involved with mergers and acquisitions from 2001 to 2008 and investigate the sensitivity of cash holdings and investment to cash flows. they find that it is the survivor rather than the new venture which significantly increase cash holdings one year after the ipo. in addition, positive sensitivity of investment to cash flows is found among new ventures, which implies that investment by financially constrained firms mainly relies on own funds as post-ipo firm characteristics cannot not be changed immediately. odm oem obm smile curve value added grin curve forced smile curve advances in technology innovation, vol. 4, no. 1, 2019, pp. 44 57 46 relevant studies such as maine et al. [4] discuss value creation by startupand science-based businesses such as biotechnology firms and deepen the specifics inherent in value creation by companies producing high-end materials. the authors apply hierarchical clustering to differentiate various types of material technology and observe differences in value creation driven by decision incorporating uncertainty, commercialization of products, target-market-tailored r&d, and value chain integration across the sample firms. chemmanur et al. [5] advance by connecting the venture capital firm with corporate innovation. the authors substantiate a higher degree of innovation in venture capital firms established and supported by parent firms than in those which operate independently, suggesting the key role played by industrial knowledge and greater room for failure in value creation achieved by external venture capital. yang et al. [6] continue the studies in corporate growth and find a u-shaped relation between diversification by venture capital businesses and value creation of new ventures. quentier [7] deepens with analysis of self-employment by the unemployed through creation of new ventures and find that government subsidies in the form of loans better increase quality and selection of new ventures than direct subsidies. the loans also serve to solve financial constraint faced by these new ventures. fig. 2 synthesizes the theoretical foundations inherent in the technology gap theory suggested by posner [8] with illustration of strategies by the oem, odm, and obm. the firm is not established over the period of t0~t1 for lack of competitiveness and starts to imitate other firms by operation in the form of oem regarding conditions such as the economies of scale, production standardization, and initial capacity over the period of t1~t2. then the firm gradually raises operation performance (op) along the path of t1a under the imitation lag. however, the firm which keeps the oem strategy may face stagnant or decreasing op and enjoy the marginal profit only under the agency problem caused by high concentration of buyers or an increase in competitors as suggested by the transaction cost theory. to avoid such a strategic problem for oem, the firm should adopt the odm strategy by an increase in spending on research and development to gain patents at the initial expense of the design lag and failure in the early state of innovation, making op fall along the path of ab over the period of t2~t3. beyond t3, the firm’s op gradually rises along the path of bc with advantages from technological improvement and creation of new product. if other oems also experience imitation and design lags to catch the firm’s odm strategy, the firm’s op may become stagnant or declining beyond t4 under the love for varieties by consumers. as suggested in the product differentiation theory by krugman [9] and attribute differential theory by lancaster [10], the odm’s gross margin will decrease with mass customization. to increase op, the obm strategy should be adopted, but the firm has to experience the demand lag and substantial branding investment over the period t4~t5 where op falls along the path of cd. only as the capability for product differentiation, operational digitalization, investment in intangible assets, and internationalization reach a stable level, the firm’s op grows along the path of de. fig. 2 strategies by oem, odm, and obm fig. 3 business life cycle, internationalization, and value creation in literature, dunning [11], wagner [12], bhide [13], zahra et al. [14], kuo and li [15], head and ries [16], and hmieleski and baron [17] discuss from various dimensions the strategies for business evolution and expansion in the context of international competition. kumar and siddharthan [18] place focus on the indian enterprises, whereas chen [19], lundquist [20], lin [21], and lin et al. [22] examine businesses in taiwan. this paper attempts to evaluate whether the firm’s operating advances in technology innovation, vol. 4, no. 1, 2019, pp. 44 57 47 years (business life cycle) and internationalization (overseas investment and export) shift the smile curve up to strengthen value creation, illustrated by fig. 3. subsequent analysis is to construct a regression model of value creation and assess the impact of the operating years, brand, r&d, internationalization, and other control variables of management on the value added for taiwan’s manufacturers in addition to proposing appropriate strategies. 3. empirical model and hypotheses 3.1. data the empirical analysis adopts the 2011 industry, commerce, and service census conducted by taiwan’s directorate-general of budget, accounting, and statistics. the original sample covers 167,840 firms. with the deletion of firms whose variable values are erroneous or omitted, the final sample contains 104,377 firms regrouped into ten classes. table 1 recapitulates variable definitions. table 1 variable definitions variable definition remark panel a: value creation lva value added in logarithm panel b: business life cycle age operating years age 2 squared age panel c: asymmetric grin curve br brand revenue to total revenue in percentage rd r&d spending to total spending in percentage fi overseas investment to net fixed assets in percentage ex export revenue to total revenue in percentage panel d: operation digitalization ec1 information disclosure on internet dummy ec21 purchases on internet dummy ec22 internet purchases to total spending in percentage ec31 sales on internet dummy ec32 internet sales to total revenue in percentage panel e: control variable opr manufacturing as main business dummy lsize net assets in logarithm klr net fixed assets to number of employees capital-labor ratio panel f: industry class inda food and drink dummy indb textile, clothing, and leather dummy indc paper and printing dummy indd oil, coal, and chemistry dummy inde rubber and plastics dummy indf metal dummy indg electronics and computer dummy indh machinery dummy indi car and other vehicle dummy indj furniture dummy 3.2. model and hypotheses to measure the impact of innovation, brand revenue, r&d spending, and internationalization on value creation for taiwan’s manufacturers, the model of eq. (1) is constructed by the following regression equation. advances in technology innovation, vol. 4, no. 1, 2019, pp. 44 57 48 lva = a0 + a1age + a2age 2 + a3br + a4br 2 + a5rd + a6rd 2 + a7fi + a8fi 2 + a9ex + a10ex 2 + a11ec1 + a12ec21+ a13ec22+ a14ec31+ a15ec32 + a16opr + a17lsize + a18klr + a19klr 2 + a20indb + a21indc + a22indd + a23inde + a24indf + a25indg+ a26indh + a27indi + a28indj + e (1) lva is taken in logarithm and measure value creation. the parameters a1 and a2 serve to assess the role played by the business life cycle in the firm’s value added, whereas the parameters from a3 to a6 respectively focus on the impact of obms (a3 and a4) and odms (a5 and a6). the parameters from a7 to a10 serve to assess the role played by internationalization in terms of overseas investment (a7 and a8) and export (a9 and a10). the parameters from a11 to a15 serve to evaluate the contribution of operation digitalization through internet to value creation. finally, the influence of the control variables and industry class dummies are captured by the parameters from a16 and a28. the last term e represents the error term. the model of eq. (1) is preliminarily estimated by the ordinary least squares method. the white test suggests significant heteroscedasticity in residuals, with a chi-squared value of 58,305. hence, subsequent analysis substitutes the robust standard errors for standard errors estimated by the ordinary least squares method. besides, results from parameter estimation of the model of eq. (1) are used as foundations for testing the seven core hypotheses summarized from eqs. on (2) to (8). h1: value creation is a u-shaped quadratic function of business life cycle, implying increasing marginal benefit, or a1 < 0 and a2 > 0. (2) h2: value creation is a reversed u-shaped quadratic function of brand revenue, implying decreasing marginal benefit, or a3 > 0 and a4 < 0. (3) h3: value creation is a reversed u-shaped quadratic function of r&d spending, implying decreasing marginal benefit, or a5 > 0 and a6 < 0. (4) h4: the impact of brand revenue and r&d spending on value creation is asymmetric, or a3 + 2a4br  a5 + 2a6rd. (5) h5: value creation is a reversed u-shaped quadratic function of overseas investment, implying decreasing marginal benefit, or a7 > 0 and a8 < 0. (6) h6: value creation is a reversed u-shaped quadratic function of export, implying decreasing marginal benefit, or a9 < 0 and a10 < 0. (7) h7: the impact of overseas investment and export on value creation is asymmetric, or a7 + 2a8fi  a9 + 2a10ex. (8) 4. empirical results 4.1. descriptive statistics table 2 summarizes descriptive statistics for the value creation, business life cycle, brand, r&d, internationalization, operation digitalization, and other control variables of management across 104,377 manufacturers in taiwan in 2011. for value creation (va), the mean for the original values is at $39,043 thousand new taiwan dollars across surveyed manufactures in taiwan. the gap between the maximum value and the minimum value appears substantial, confirmed by a high level of its standard deviation. a similar pattern is observed in the business life cycle (age) ranging from new ventures (1 year) to sustainable operating firms (100 years). the average of age is close to 18 years. advances in technology innovation, vol. 4, no. 1, 2019, pp. 44 57 49 the ratios of brand revenue to total revenue (br), r&d spending to total spending (rd), overseas investment to net fixed asset (fi), and export to total revenue (ex) are respectively 8.8812%, 0.3251%, 0.0511%, and 8.3778%, showing strong bias to oems and insufficient internationalization for most manufacturers in taiwan. as regards operation digitalization, the ratios of internet purchases to total spending (ec22) and internet sales to total revenue (ec32) are 1.3210% and 1.3337% only, suggesting that purchases and sales through internet remain at its early stage of development in taiwan’s manufacturing industry. in terms of control variables, the dispersion for the value of net assets (lsize) that reflects the firm’s operating scale remains significant in logarithm. the capital-to-labor ratio (klr) measured by net fixed assets over the number of employees is averaged at 1.5335. table 2 descriptive statistics variable mean maximum minimum std. dev. panel a: value creation lva 8.5416 19.4878 5.0626 1.3276 panel b: business life cycle age 17.5632 100 1 0.2784 panel c: asymmetric grin curve br 8.8812 100 0 25.6504 rd 0.3251 74.5755 0 2253.051 fi 0.0511 224.4135 0 2.0569 ex 8.3778 100 0 80.3330 panel d: operation digitalization ec1 0.6122 1 0 1.3542 ec21 0.1136 1 0 227.8531 ec22 1.3209 99.5421 0 22.1068 ec31 0.0847 1 0 1774.022 ec32 1.3337 99.9995 0 0.4872 panel e: control variable opr 0.8308 1 0 0.3173 size 248,994 1.53e+09 169 6.4962 klr 1.5335 63.8248 0.0016 0.3749 panel f: industry class inda 0.0382 1 0 0.1917 indb 0.0681 1 0 0.2519 indc 0.0964 1 0 0.2951 indd 0.0338 1 0 0.1806 inde 0.0882 1 0 0.2836 indf 0.2933 1 0 0.4553 indg 0.0755 1 0 0.2642 indh 0.2135 1 0 0.4098 indi 0.0464 1 0 0.2103 indj 0.0466 1 0 0.2107 4.2. estimation results table 3 below summarizes major results from estimation of equation (1) with adjustment in heteroscedasticity. with additional regression for independent variables, the variance inflation factor (vif) appears low except for age, age2, br, br2, ex, and ex2 whose vif exceeds 10. therefore, there is, overall, no serious problem of collinearity. the adjusted r2 is around 0.6533, substantiating sufficient explanatory power of the model specified. the chi-squared value for the white test is at 58,305, implying significant heteroscedasticity in residuals. therefore, the t-values reported in table 3 are adjusted by robust standard errors. advances in technology innovation, vol. 4, no. 1, 2019, pp. 44 57 50 table 3 estimation results variable lva t-value vif constant 3.2936*** 104.3353 na age -0.0029*** -3.4095 11.8816 age 2 1.01e-0.4*** 4.7657 11.7598 br 0.0041*** 6.0753 40.9094 br 2 -3.08e-05*** -4.0718 39.8211 rd 0.0805*** 22.0287 3.8738 rd 2 -0.0016*** -12.6025 3.4283 fi 0.0255*** 3.5577 4.3578 fi 2 -1.28e-04*** -3.4249 4.2128 ex 0.0119*** 26.6641 13.6124 ex 2 -9.36e-05*** -17.0069 12.4328 ec1 0.0393*** 7.5535 1.1698 ec21 0.1310*** 10.1139 2.6106 ec22 -4.72e-05 -0.0358 10.4591 ec222 -3.83e-05* -1.8291 7.8524 ec31 0.0175 1.1186 2.8294 ec32 -7.47e-04 -0.6356 11.0646 ec322 8.26e-06 0.5897 7.9876 opr -0.0441*** -6.4857 1.0724 klr -0.0942*** -16.0990 4.8555 klr 2 1.72e-04*** 5.5128 1.4425 lsize 0.5627*** 139.6625 5.6422 indb 0.0771*** 4.4743 3.0369 indc 0.1214*** 7.5717 4.5435 indd 0.1636*** 7.5286 1.7445 inde 0.1141*** 6.9995 3.9985 indf 0.1983*** 13.2852 8.7908 indg 0.1771*** 9.7818 2.8244 indh 0.0903*** 5.9271 6.8611 indi 0.2896*** 15.6108 2.5161 indj -0.0217 -1.2003 2.5268 adjusted r 2 0.6533 white test 58,394*** note: ***, **, * for significance at the 1%, 5%, 10% level. fig. 4 marginal impact of business life cycle fig. 5 marginal impact of brand revenue advances in technology innovation, vol. 4, no. 1, 2019, pp. 44 57 51 next, the seven hypotheses summarized from (2.1) to (2.7) are each examined. for h1, the estimated coefficient for age is significantly lower than zero and that for age2 is significantly greater than zero. this confirms that the relation betwee n value creation (lva) and the business life cycle can be portrayed by a u-shaped curve which decreases first before increasing. this link can be illustrated by fig. 4. in fig. 4, the minimum point of the lva curve corresponds to around 15 years of the business life cycle. hence, new ventures have to pay a cost of lower value creation at the early stage of corporate development, supporting h1 (2.1). for h2, the marginal impact of br on lva is observed from a reversed u-shaped curve that rises first before falling, illustrated by fig. 5. the maximum point of the lva curve is at 66.5584%, far above the sample mean (8.8812%). therefore, the role of branding in value creation is captured by a grin curve and there seems great room for improvement in taiwan’s current obms, supporting h2 (2.2). as regards h3, the coefficient for rd is positive whereas that for rd2 is negative. this can be illustrated by fig. 6, where the relation between r&d spending and value creation is presented by a reversed u-shaped curve. the maximum point of the lva curve is at 25.1563%, far above the mean for rd (0.3251%). hence, on the left side of supply chains, the strategies for odms are subject to a rising-then-falling grin curve, supporting h3 (2.3). comparing the marginal impact of brand (fig. 5) and the marginal impact of r&d (fig. 6), it is observed that the former impact is smaller than the latter impact. in other words, we observe an asymmetric grin curve whose left-side peak is higher than the right-side peak, supporting h4 (2.4). with respect to h5, the estimated coefficients for fi and fi2 also supports h5 (2.5), which implies a reversed u-shaped lva curve illustrated by fig. 7. the maximum point of this curve corresponds to a ratio of overseas investment to net fixed assets at 99.6094%, far above the industry mean (0.0511%), suggesting current underinvestment overseas by most smes in taiwan’s manufacturing industry. the finding for h6 is analogous to that for h5. as illustrated by fig. 8, the optimal export ratio is at 63.5684% and also far above the industry mean (8.3788%). hence, taiwan’s manufacturers which aim to strengthen the value creation need to substantially raise the weight of export in total revenue. comparing fig. 7 and fig. 8, the marginal impact of overseas investment is also higher than that of the export, implying an asymmetric grin curve and substantiating h7 (2.7). fig. 6 marginal impact of r&d spending fig. 7 marginal impact of overseas investment fig. 8 marginal impact of export advances in technology innovation, vol. 4, no. 1, 2019, pp. 44 57 52 in terms of operation digitalization, the positive coefficients for the two dummies, ec1 and ec21, suggest that information disclosure and purchases on the internet serve to improve value creation by significant cost cut. however, the coefficients for ec22 significant negative and the coefficient for ec 31 and ec 32 are found insignificant. therefore, at least at the current stage, internet activities may not effectively contribute to value creation in taiwan’s businesses in the manufacturing sector. finally, the empirical findings with respect to control variables can be analyzed from five aspects. first, the coefficient for opr is significantly negative, implying that whether the firm’s main activity involves manufacturing or not cannot effectively increase the level of value creation. second, the positive sign for the firm’s size (lsize) confirms that value creation rises with corporate expansion. third, the capital-to-labor ratio (klr) exerts a u-shaped impact on value creation. capital intensity hence plays an important role in the firm’s value added. four, as regards the dummies for the various classes in taiwan’s manufacturing industry, the value added appears higher among indb～indi as we adopt food and drink (inda) as the benchmark industry class for comparison. fig. 9 integrates previous findings and analysis illustrated by fig. 5 to fig. 8 and presents an overall grin curve for taiwan’s manufacturing industry. figure 9 suggests that the grin curve shifts down as the firm’s operating years are less than 15 years. only as the firm’s life cycle exceeds 15 years will value creation be raised. fig. 9 asymmetric grin curve moreover, the marginal impact of r&d on value creation is higher than that of the brand, leading to an asymmetric grin curve where the left-side peak is higher than the right-side peak. the marginal impact of overseas investment is greater than that of exports, too. overall, the seven hypotheses stated in (2.1) to (2.7) are in line with fig. 9, which highlights graphic asymmetry in the grin curve for taiwan’s manufacturers. 4.3. robustness check to check the robustness of the previous analysis, we conduct three additional tests. table 4 summarizes results of testing the differences in means for major variables between new venture and survivors on the basis of two definitions for new ventures: defined by 12 years against definition by 15 years. overall, all variables except for ec31, ec32, indf, and indi exhibit significant differences between new ventures and survivors regardless of the definition for new ventures adopted, substantiating robustness of our analysis framework and results. table 5 compares coefficients respectively estimated with linear, quadratic, and cubic forms for selected variables for regression of lva. the estimation results appear consistent across the three regression models, further confirming the advances in technology innovation, vol. 4, no. 1, 2019, pp. 44 57 53 robustness of our model specification. more specifically, the adjusted r2 remains around a similar level across the three models, implying that the respectively effect of key variables included in table 3 on value creation (lva) all holds regardless of the functional form adopted. for variables such as age, br, rd, fi, and ex, the impact on lva is consistently significant. in contract, the role for dummies such as e31 and e32 keeps absent. table 4 differences in means between new ventures and survivors new venture defined by 12 years new venture defined by 15 years variable age≤12 age≥12 age≤15 age≥15 lva 8.3191 8.6497 8.3599 8.6698 (1443.683***) (1397.573***) age 6.3419 23.0156 7.7371 24.5002 (134293.5***) (175364***) br 7.2344 9.6813 23.6322 26.9382 (209.4597***) (233.1375***) rd 0.3677 0.3045 0.3831 0.28426 (21.6962***) (58.4488***) fi 0.0210 0.0657 0.0244 0.0700 (25.0707***) (28.7915***) ex 6.4475 9.3157 6.7972 9.4936 (388.1113***) (378.0316***) opr 0.7816 0.8547 0.7869 0.8619 (880.1822***) (1022.868***) lsize 8.8140 9.3984 8.8743 9.4423 (2377.543***) (2477.660***) klr 1.4360 1.5809 1.4419 1.5982 (32.2502***) (41.4143***) ec1 0.5523 0.6414 0.5705 0.6417 (773.5952***) (543.8418***) ec21 0.1175 0.1117 0.1191 0.1097 (7.5348***) (22.3491***) ec22 1.3690 1.2976 1.3943 1.2691 (2.7802*) (9.3998***) ec31 0.0853 0.0843 0.0864 0.0834 (0.2861) (2.8233*) ec32 7.6027 7.4995 0.0864 1.3224 (0.1274) (0.1645) inda 0.2150 0.1791 0.0459 0.0328 (148.7985***) (119.8543***) indb 0.0588 0.0726 0.0609 0.0732 (68.5690***) (60.6506***) indc 0.0882 0.1003 0.0898 0.1010 (38.5585***) (36.6144***) indd 0.0319 0.0344 0.0305 0.0361 (5.5848**) (23.6399***) inde 0.0712 0.0965 0.0725 0.0993 (183.2514***) (227.4697***) indf 0.4542 0.4558 0.2954 0.2918 (1.3431) (1.5731) indg 0.0942 0.0664 0.0936 0.0627 (254.5745***) (347.7844***) indh 0.2275 0.2067 0.2228 0.2070 (58.7679***) (37.4631***) indi 0.0465 0.0463 0.0463 0.0465 (0.0275) (0.0216) indj 0.0420 0.0487 0.0423 0.0496 (23.2437***) (30.7220***) observations 34,132 70,245 43,194 61,183 note: ***, **, * for significance of the f-value at the 1%, 5%, 10% level. advances in technology innovation, vol. 4, no. 1, 2019, pp. 44 57 54 table 5 robustness test by function form linear quadratic cubic variable lva lva lva constant 3.3533*** (53.0204) 3.2936*** (104.3353) 3.1809*** (108.8993) age 0.0014*** (5.0475) -0.0029*** (-3.4095) 0.0019 (1.3231) age 2 na 0.0001*** (4.7657) -0.0001** (-1.9651) age 3 na na 2.48e-06*** (3.0442) br 0.0019*** (14.3199) 0.0041*** (6.0753) 2.00e-05 (0.0090) br 2 na -3.08e-05*** (-4.0718) 5.57e-05 (0.9852) br 3 na na -4.84e-07 (-1.3533) rd 0.0296*** (10.5057) 0.0805*** (22.0287) 0.1573*** (33.8342) rd 2 na -0.0016*** (-12.6025) -0.0077*** (-23.9350) rd 3 na na 7.84e-05*** (18.7086) fi 0.0081** (2.3060) 0.0255*** (3.5577) 0.0594*** (6.1528) fi 2 na -0.0001*** (-3.4249) -0.0012*** (-5.3520) fi 3 na na 4.38e-06*** (5.0506) ex 0.0052*** (29.7733) 0.0119*** (26.6641) 0.0159*** (13.3506) ex 2 na -9.36e-05*** (-17.0069) -0.0003*** (-6.9366) ex 3 na na 1.17e-06*** (4.5717) ec1 0.0496*** (8.6773) 0.0393*** (7.5535) 0.0306*** (6.0460) ec21 0.1578*** (13.3077) 0.1310*** (10.1139) 0.1059*** (7.3308) ec22 -0.0024*** (-3.9812) -4.72e-05 (-0.0358) 0.0064*** (2.7088) ec22 2 na -3.83e-05* (-1.8291) -0.0003*** (-3.7794) ec22 3 na na 2.29e-06*** (3.5345) ec31 0.0137 (0.9719) 0.0175 (1.1186) 0.0178 (1.0281) ec32 0.0005 (0.9216) -0.0007 (-0.6356) -0.0003 (-0.1436) ec32 2 na 8.26e-06 (7.9876) -1.58e-05 (-0.2124) ec32 3 na na 2.38e-07 (0.4314) opr -0.0362*** (-0.0588) -0.0441*** (1.0724) -0.0438*** (-6.5869) klr -0.0497*** (-3.4393) -0.0942*** (4.8555) -0.1414*** (-21.8759) klr 2 na 0.0002*** (1.4425) 0.0009*** (6.3332) klr 3 na na -1.13e-06*** (-5.2410) lsize 0.5423*** (52.5408) 0.5627*** (5.6422) 0.5823*** (161.4039) advances in technology innovation, vol. 4, no. 1, 2019, pp. 44 57 55 table 5 robustness test by function form (continued) linear quadratic cubic variable lva lva lva indb 0.0965*** (4.9269) 0.0771*** (3.0369) 0.0556*** (3.2904) indc 0.1321*** (7.7247) 0.1214*** (4.5435) 0.1075*** (6.8314) indd 0.1857*** (8.6562) 0.1636*** (1.7445) 0.1520*** (7.0760) inde 0.1360*** (7.6700) 0.1141*** (3.9985) 0.0943*** (5.9018) indf 0.2177*** (13.2065) 0.1983*** (8.7908) 0.1779*** (12.1368) indg 0.2373*** (11.3780) 0.1771*** (2.8244) 0.1301*** (7.3822) indh 0.1175*** (6.9732) 0.0903*** (6.8611) 0.0637*** (4.2661) indi 0.3201*** (15.6212) 0.2896*** (2.5161) 0.2529*** (13.9078) indj -0.0113*** (-0.5899) -0.0217 (2.5268) -0.0414** (-2.3481) adj. r 2 0.6332 0.6533 0.6696 white test 90341.73*** 58,394*** 34,574*** note: ***, **, * for significance at the 1%, 5%, 10% level. 5. conclusions this paper empirically adopts the big data obtained from the 2011 industry, commerce, and service census conducted by taiwan’s directorate-general of budget, accounting, and statistics and examines the link between the business life cycle, brand, r&d, internationalization, and value creation for taiwan’s manufacturers. regression results based on 104,377 observations in the prescreened sample can be recapitulated in seven points. (1) a relatively low level of the value added, brand revenue, r&d spending, and internationalization suggests that taiwan’s manufacturing industry is currently subject to a business environment dominated by oems mainly oriented to the local market. (2) new ventures whose business life is less than 15 years have to face a low level of value creation. only beyond 15 years will value creation be strengthened by greater efficiency in operation. (3) the marginal impact of brand revenue and r&d spending on value creation is captured by a reversed u-shaped curve. the decreasing marginal effect is higher for r&d than for brand revenue. (4) the marginal impact of the two gauges for internationalization (overseas investment and export) on value creation is captured by a reversed u-shaped curve, too. the decreasing marginal effect is higher for overseas investment than for export. (5) operation digitalization remains at a low level for taiwan’s manufacturers and its impact on value creation appears insignificant as of 2011. (6) the average business life cycle is around 18 years in the sample. under intensifying competition from globalization and e-business, the business life cycle is anticipated to be further shortened, creating more challenges to taiwan’s manufacturing industry. advances in technology innovation, vol. 4, no. 1, 2019, pp. 44 57 56 (7) businesses in the manufacturing industry in taiwan face an asymmetric grin curve rather than a smile curve. therefore, value creation can be strengthened through enhancement in the business life cycle, internationalization, internet activity, operating scale, and capital intensity. this study is conditioned on a few limitations, though. the big data obtained from the government census are based on five-year surveys. the problem associated with potentially lagged information appears unavoidable. besides, discontinuity in our data makes it difficult to analyze the stock value and lagged effect for activities associated with the brand, r&d, and overseas investment. availability of a more complete dataset will benefit future research, which can also extend analysis to issues such as income distribution and equality under intensified globalization. acknowledgement the authors would like to acknowledge a research grant (most 103-2632-h-029-002-my2) from the ministry science and technology in taiwan, roc. references [1] r. n. lussier, “a nonfinancial business success versus failure prediction model for young firms,” journal of small business management, vol. 33, no. 1, pp. 8-20, 1995. [2] p. marco, p. fabio, and z. luigi, “why do companies go public,” the journal of finance, vol. 53, no. 1, pp. 27-64, february 1998. [3] i. erel, y. jang, and m. s. weisbach, “do acquisitions relieve target firm’s financial constraints,” journal of finance, vol. 70, no. 1, pp. 289-328, 2015. [4] e. maine, s. lubik, and e. garnsey, “value creation strategies for science-based business: a study of advanced materials ventures,” innovation: management, vol. 15, no. 1, pp. 35-51, december 2014. [5] t. j. chemmanur, e. loutskina, and x. tian, “corporate venture capital, value creation, and innovation,” the review of financial studies, vol. 27, no. 8, pp. 2434-2473, august 2014. [6] y. yang, v. narayanan, and d. de carolis, “the relationship between portfolio diversification and firm value: the evidence from corporate venture capital activity,” strategic management journal, vol. 35, no. 13, pp. 1993-2011, december 2014. [7] j. m. quentier, “self-employment start-ups and value creation: an empirical analysis of german micro data,” advances in competitiveness research, vol. 20, no. 1-2, pp. 37-57, 2012. [8] m. v. posner, “international trade and technical change,” oxford economic papers, vol. 13, no. 3, pp. 323-341, 1961. [9] p. r. krugman, “increasing returns, imperfect competition, and international trade,” journal of international economics, vol. 9, no. 4, pp. 467-479, 1979. [10] k. lancaster, “intra-industry trade under perfect monopolistic competition,” journal of international economics, vol. 10, no. 2, pp. 151-175, may 1980. [11] j. h. dunning, “toward an eclectic theory of international production: some empirical test,” journal of international business studies, vol. 11, no. 1, pp. 9-31, 1980. [12] j. wagner, “exports, firm size and firm dynamics,” small business economics, vol. 7, no. 1, pp. 29-39, february 1995. [13] a. bhide, the origin and evolution of new business. new york: oxford university press, 2000. [14] s. a. zahra, r. d. ireland, and m. a. hill, “international expansion by performance,” academy of management journal, vol. 43, no. 5, pp. 925-950, 2000. [15] h. c. kuo and y. li, “a dynamic decision model of sme’s fdi,” small business economics, vol. 20, no. 3, pp. 219-231, may 2003. [16] k. head and j. ries, “fdi as an outcome of the market for corporate control: theory and evidence,” journal of international economics, vol. 74, no. 1, pp. 2-20, january 2008. [17] k. m. hmieleski and r. a. baron, “entrepreneurs’ optimism and new venture performance: a social cognitive perspective,” the academy of management review, vol. 52, no. 3, pp. 473-488, 2009. [18] n. kumar and n. s. siddharthan, “technology, firm size and export behavior in developing countries: the case of indian enterprises,” journal of development studies, vol. 31, no. 2, pp. 289-309, december 1994. https://econpapers.repec.org/article/blastratm/ advances in technology innovation, vol. 4, no. 1, 2019, pp. 44 57 57 [19] a. chen, “taiwan’s paradigm shift: industrial engineers are shaping a nation,” industrial engineer, vol. 40, no. 10, pp. 31-34, october 2008. [20] e. lundquist, “shih’s curve can bring smiles,” eweek, vol. 56, september 2007. [21] f. j. lin, “the determinants of foreign direct investment in china: the case of taiwanese firms in the it industry,” journal of business research, vol. 63, no. 5, pp. 479-485, may 2010. [22] j. r. lin, c. j. huang, and c. h. hsieh, “internationalization, domestic employment, and overstatement of export contribution,” taipei economic inquiry, vol. 51, no. 1, pp. 135-169, 2015. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation, vol. 2, no. 1, 2017, pp. 22 24 22 experimental study of leaching and penetration of nitrite ions in nitrite-type repair materials on the surface of concrete masumi inoue 1,* , heesup choi 1 , yuhji sudoh 2 , and koichi ayuta 1 1 department of civil and environmental engineering, kitami institute of technology, hokkaido, japan. 2 chemicals division. basic chemicals department, nissan chemical industries, ltd., tokyo, japan. received 29 january 2016; received in revised form 25 march 2016; accepted 28 march 2016 abstract this study aimed to clarify the leaching properties of nitrite ions in nitrite-type repair materials exposed to rainfall. repaired concrete specimens were prepared for leaching tests using a lithium nitrite solution, and the amounts of leaching and penetration of nitrite ions were measured under simulated rainfall. the results demonstrated that the amount of leaching could be controlled by using polymer cement paste and mortar surface coatings containing lithium nitrite solution, and by using polymer cement mortar surface coatings following direct lithium nitrite solution coatings. furthermore, the amount of nitrite ion leaching in all cases was lower than the discharge standard value established by the water pollution control law. keywords: nitrite-type repair material, nitrite ion, leaching, penetration, polymer cement mortar, polymer cement paste 1. introduction in general, nitrite-type repair materials are commonly used for the purpose of minimizing of corrosion of reinforcements in concrete structures. it is widely known that nitrite ions effectively regenerate passive films on reinforcement surfaces of concrete as they penetrate the surrounding areas [1-3]. however, since nitrite ions easily dissolve in water, there is concern that they can be leached from the surfaces of repair materials upon exposure to rainfall. the aim of this study was to elucidate the leaching properties of nitrite ions in nitrite-type repair materials exposed to rainfall. repaired concrete specimens were prepared for leaching tests using a lithium nitrite solution, and the amounts of leaching and penetration of nitrite ions were measured under simulated rainfall conditions. 2. method the repair methods, which used a 40% water solution composed of lithium nitrite (ln40), are shown in table 1. these methods are widely used in repair work sites. the repaired concrete specimen is shown in fig. 1. the water-cement ratio was 60%. the specimen was demolded one day after casting, cured in water (20±1 °c) until the age of seven days, and then stored in room-temperature conditions (20±1 °c and 50±5% relative humidity). after curing, the test surface was polished with a sand paper, and cleaned via air-cleaning, and the specimens were repaired using ln40. after repairing, the repaired specimens were cured in room-temperature conditions (20±1 °c and 50±5% relative humidity) for seven additional days. the specimen surfaces excluding test surfaces were coated with an epoxy resin in order to prevent nitrite ion penetration. the installation conditions of the specimens are shown in fig. 2. specimens were installed at a 35° angle and attached to an acrylic plate on the side. in the leaching test, distilled water simulating rainfall was sprayed to the specimen test surfaces. the amount of water sprayed each day was 1.86 g/mm 2 , a value reflecting the av* corresponding author, email: m-inoue@mail.kitami-it.ac.jp advances in technology innovation, vol. 2, no. 1, 2017, pp. 22 24 23 copyright © taeti erage annual rainfall in japan (about 1,700 mm). a wet-dry cycle consisted of one day of spraying and three days of drying, a pattern reflecting the average annual rainfall in japan (about 128 days). the wet-dry cycle was repeated for one year (91 cycles). distilled water was sprayed every two hours from 10:00 am to 4:00 pm on a spraying day. the amount of water sprayed each time was 0.465 g/mm 2 (a quarter of 1.860 g/mm 2 ). distilled water flowing down the test surface was collected after every spraying, and the leached nitrite ion (no2 ) concentration was measured using ion chromatography. in the penetration test, distilled water simulating rainfall was sprayed using the same method as the leaching test. however, the no2 concentration in the repair material and repaired concrete was measured at 6 months (45 cycles) and 1 year (91 cycles). table 1 repaired concrete specimens name of specimen repair method no2 solid amount (g) leaching penetration n non repair 0 ln coating of ln40 6.26 0.70 lnpcp surface coating by pcp*1 using ln40 (thickness:2mm) 12.4 1.34 lnpcm surface coating by pcm*2 using ln40 (thickness:5mm) 22.9 2.54 ln+pcm surface coating by pcm*2 after coating of ln40 (thickness:5mm) 6.26 *1: polymer cement paste, *2: polymer cement mortar table 2 leaching ratio of no2 amount of leaching to no2 solid amount after 91cycles repair method ln ln pcp ln pcm ln +pcm no2 solid amount in repair material (g) 6.3 12.4 22.9 6.3 after 91 cycles no2 leaching (g) 1.39 0.52 0.93 0.49 no2 leaching ratio (%) 22.2 4.2 4.1 7.8 (a) leaching test (b) penetration test fig. 1 specimen overview fig. 2 installation conditions of specimens 3. results and discussion the no2 concentration changes found in the leaching test are shown in fig. 3, and the ratio of no2 amount of leaching to no2 solid amount after 91cycles is shown in table 2. the no2 leaching amount of ln was largest after every cycle, and the leaching ratio of no2 solid amount after 91 cycles was about 22%. the no2 leaching amount of ln+pcm after one cycle decreased by about 70% compared with that of ln. this is because the no2 leaching was controlled by the pcm coating. on the other hand, the no2 leaching amount of lnpcm was larger than that of lnpcp. however, the leaching ratio of no2 solid amount of lnpcm was almost same as that of lnpcp, as shown in table 2. this implies that there is no actual difference between the leaching properties of lnpcp and lnpcm. it was confirmed that the no2 concentration in all the repair methods changed at roughly 10 cycles. afterwards, the change in no2 concentration was small. although small amounts of no2 were detected, the concentrations of all repair methods after 10 cycles were almost same as that of n (non-repair). therefore, it is thought that no2 hardly leached after 10 cycles for this experimental range of conditions. considering the discharge standard of the water pollution control law, the no2 leaching amount of all cases in advances in technology innovation, vol. 2, no. 1, 2017, pp. 22 24 24 copyright © taeti every cycle was smaller than that of the discharge standard value (329 ppm). the ratio of no2 amount of leaching and penetration to no2 solid amount in repair materials after 91 cycles is shown in fig. 4. the no2 penetration and leaching ratio of ln were about 74% and 22%, respectively. on the other hand, in the case of lnpcp and lnpcm, both addition ratios of the amount of ion penetration in concrete and residual ions in the repair materials were about 90%. furthermore, both leaching ratios were equal to about 4%. therefore, it is thought that no2 leaching can be controlled using pcp or pcm containing ln40. the total ratios of no2 amounts of leaching and penetration including residual ions in repair materials were about 93-96%, and there was little difference between no2 solid amount of repair material and total no2 solid amount after 91 cycles. however, the difference was within the range of measurement error of ion chromatography. therefore, it is thought that the behavior internal and external of no2 in nitrite-type repair materials was mostly evaluated. fig. 3 change of no2 concentration in leaching test 4. conclusions the aim of this study was to clarify the leaching properties of nitrite ions in nitrite-type repair materials upon exposure to rainfall. repaired concrete specimens were prepared using lithium nitrite solution for leaching tests, and the amounts of leaching and penetration of nitrite ions were measured under simulated rainfall. the following conclusions were drawn from the investigation. 1) it was confirmed that the no2 concentration in all repair methods changed at approximately cycle 10. subsequent changes in no2 concentration were small. 2) the no2 leaching amounts of all repair methods in all cycles were smaller than that of the discharge standard value. 3) the no2 leaching can be controlled using pcp and pcm surface coatings with the addition of ln40 and surface coating by pcm after direct coating with ln40. fig. 4 ratio of no2 amount of leaching and penetration to no2 solid amount in repair materials after 91 cycles references [1] a. m. rosenberg, et al., “a corrosion inhibitor formulated with calcium nitrite for use in reinforced concrete,” astm, stp 629, pp. 89-99, 1977. [2] a. kobayashi, s. ushijima, i. kamuro, and m. koshikawa, “a study on the protection of steel in concrete by penetrative corrosion inhibitor,” journal of japan society of civil engineers, vol. 1990, no. 420, pp. 51-60, 1990. [3] t. hori, s. yamasaki, and y. masuda, “a study on the corrosion inhibiting effect of mortar with high nitrite content,” concrete research and technology, vol. 5, no. 1, pp. 89-98, 1994.  advances in technology innovation , vol. 1, no. 2, 2016, pp. 29 32 29 copyright © taeti reduction of residual stresses in sapphire cover glass induced by mechanical polishing and laser chamfering through etching shih-jeh wu 1,* , hsiang-chen hsu 1,2 and wen-fei lin 3 1 department of mechanical and automation engineering, i-shou university, kaohsiung, taiwan. 2 department of industrial management, i-shou university, kaohsiung, taiwan . 3 department of material science and engineering, i-shou university, kaohsiung, taiwan. received 02 february 2016; received in revised form 30 march 2016; accept ed 02 april 2016 abstract sapphire is a hard and anti-scratch materia l commonly used as cover glass of mobile devices such as watches and mobile phones. a mechanica l polishing using diamond slurry is usually necessary to create mirror surface. additional cha mfe ring at the edge is sometimes needed by mechanical grinding. these processes induce residual stresses and the mechanica l strength of the sapphire work p iece is impaired. in this study wet etching by phosphate acid process is applied to relief the induced stress in a 1” d ia meter sapphire cover glass. the sapphire is polished before the edge is chamfe red by a picosecond laser. residual stresses are measured by laser curvature method at different stages of mach ining. the results show that the wet etching process effectively relie f the stress and the laser mach ining does not incur serious residual stress. ke ywor ds : picosecond laser, laser curvature method, residual stress, stress relief, wet etching, sapphire 1. introduction sapphire is the material to cover glasses that are scratch and impact-resistant, yet fle xib le despite the introduction of tempered glass such as corning ® gorilla . in addition to the use for watch glasses or as cover or filter glasses for came ra lenses, now it is being tried for cell phone displays as well. sapphire glass is made of colorless plates of synthetic corundum, i.e . minera ls produced with mo lten alu minu m o xide (al2o3). in fact, sapphire glass does not have a glass -like a morphous structure, but rather a crystalline structure. with a hardness of 9 on the mohs scale, sapphire is one of the hardest transparent materials next to diamond [1, 2]. sapphire has very wide optical trans mission band from uv to near-infrared, (0.15-5.5 µ m)[2]. with its particu lar properties, sapphire glass offers advantages over the chemically hardened glass often used in the display industry. traditional methods for machining sapphire glass are generally mechanical grinding and polishing. optica l fabrication processes of these operations lead to the creation of su rface and sub-surface defect and residual stresses [3, 4]. maybe in mic ro-size, these defects degrade the strength and the performance of functional materia ls. another new development of sapphire mach ining is laser cutting. laser mach ining is superior to conventional mechanica l methods in terms of fle xib ility and edge quality when right laser type and proper beam processing is applied. mechanical saw creates mic ro cracks at edges as the speed is high which in turn deteriorates the bending strength. however, laser thermal e ffect can also induce stress which is not desirable either. a few methods can be applied to e liminate the above consequential effect fro m polishing process, name ly, trad it ional loose-abrasive polishing, wet etching, and dry plas ma etching [5-7]. the first method typically integrates a polishing step into the grinder itself, which offers the advantage of integrating the damage re moval into the grinder tool and builds upon * corresponding aut hor, email: wsj007@isu.edu.t w https://en.wikipedia.org/wiki/uv https://en.wikipedia.org/wiki/near-infrared advances in technology innovation , vol. 1, no. 2, 2016, pp. 29 32 30 copyright © taeti traditional che mica l mechanica l polishing (cm p) technology. however, it has the disadvantage of low re moval rates and perpetuates the surface profile. the second method uses familiar wet-etching processes to remove surface damage. wet chemical etching is one of the most common thinning techniques. to etch one side o f the work piece, the workpiece is immersed in etching solution e.g. dilute hydrofluoric ac id (dhf). the other side is protected either by additional layers, or by applying special chucks allo wing the p rocessing of work piece . the th ird method uses atmospheric dry plasma etching to re move surface damage. this method has the advantage such as the surface damage is re moved, the edges are imp roved by rounding the sharp edge, and the surface roughness can be controlled where needed for adhesion [7]. in this paper we de monstrate a production mach ining and compare the residual stresses after diffe rent re liev ing process. the sapphire is polished before the edge is chamfe red by a picosecond laser. residual stresses are measured by laser curvature method [8] at different stages of mach ining. the results show that the wet etching process effectively relief the stress and the laser machining does not incur serious residual stress. 2. method the schematic of the sapphire glass samples under process is as shown in fig. 1. the sapphire samples are in 1” dia meter. the sapphire glass samples we re po lished by cm p before the edge is chamfered by a picosecond laser and then they under wet etching different processes i.e., wet etching by dhf and dry plasma etching. residual stress es are measured by laser curvature method at different stages of mach ining. the schematic of the optical setup is shown in fig. 2. the line deflection (bow) was fig. 1 schematic of the sapphire glass samples under process first measured across the center of the sample (initially flat before machining) as shown in fig. 3. the stress can be then estimated by the curvature. a diagrammatic sketch is shown in fig. 4. fig. 2 optical setup of the laser curvature method fig. 3 line deflection (bow) measured across the center of the sample 3d stress map fig. 4 a diagrammatic sketch of the 3d stress map by the laser curvature method advances in technology innovation , vol. 1, no. 2, 2016, pp. 29 32 31 copyright © taeti 3. results and discussion the average surface roughness by different methods of burnish is shown in fig. 5. fro m lo w to high the roughness can be listed sequentially : mechanica l polishing, dry etching, wet etching and thermal annealing. however, the difference within acceptable range. the line deflections of the mechanically polished sapphire glass sample (1.44mm thic kness) before and after laser chamfe ring are as shown in fig. 6. there is a tensile stress residing the measured surface. interestingly, the laser machining does relief the residual stress induced. this mat be d ue to the instantaneous temperature rise during laser irradiation which leads an annealing effect. fig. 5 average roughness by different methods of burnish fig. 6 bow of the mechanically polished sapphire glass sample (1.44mm thickness) before and after laser chamfering the bow of the mechanically polished and laser mach ined sapphire glass sample (0.4mm thickness) after dry plasma etching is as shown in fig. 7. the influence of residual stress on surface deflection is more pro minent in thinner sapphire glass. befo re the etching the deflection is not symmetrical and irregular while it converts to a symmetrica l shape and the ma xima l deflect ion at the center increases slightly fro m 0.7 to 0.8 μ m a fter 300 seconds as a results of stress rela xation. the deflection at the center then starts to drop to 0.5 μ m until 360 seconds has passed. at this stage the stress has been relieved to the full extent. fig. 7 bow of the mechanically polished and laser machined sapphire glass sample (0.4mm thickness) after plasma dry etching bow of the mechanically polished and laser mach ined sapphire glass sample (0.4mm thickness) after wet etching by dhf is as shown in fig. 8. the deflect ion is irregular before the wet etching similar to dry etching. after the wet etching starts the bow turns to the other direction i.e., fro m concave to convex. as the etching process progresses the deflection drops until after 180 seconds the ma ximal deflection re mains the same at 0.57 μ m. it is believed the residual stress is relieved to the full e xtent. the reason for the change of concavity is that before the polishing process the sapphire glass is not evenly cut or ground and the work p iece retains its original shape after stress is relieved. fig. 8 bow of the mechanically polished and laser machined sapphire glass sample (0.4mm thickness) after wet etching by dhf advances in technology innovation , vol. 1, no. 2, 2016, pp. 29 32 32 copyright © taeti the optical t ransmission and reflection ratio before and after wet etching are as shown in fig. 9. there is not much difference in reflection, however, the transmission is improved. fig. 9 the optical transmission and reflection ratio before and after wet etching 4. conclusions in this paper, we demonstrate a production mach ining and compare the residual stresses after different re liev ing process. the laser chamfe ring does not incur further stress, on the contrary, it somehow reduces the stress. both dry and wet etchings are effective in re lief of residual stress induced by mechanical polishing. also the etching process did not impa ir the transparency of the sapphire glass. ball-on-ring test is proposed to confirm the effect of residual stresses on the strength of the sapphire work pieces before and after wet etching. acknowledgement the authors would like to express their appreciation to ministry of science and technology, taiwan, roc, for financial supports under project no. most103-2221-e-214-018, most104-2221-e-214-051 and most104-2632 e-214-002. special appreciation is also extended to e&r engineering corp. for carrying out all the experiments. references [1] “gorilla glass success, what is sapphire glass?” http://www.corning.com/news_center/features/ gorillaglasssuccess.aspx, corning incorporated. [2] “everything you wanted to know about sapphire glass, but were afra id to ask, ” http://www.cultofmac.co m/267068/everyth ing-wanted-know-sapphire-glass -afraid-ask -qa/. [3] d. wang, j. lee , k. ho lland, t. bibby, s. beaudoin, and t. ca le, “ von mises stress in chemical‐mechanical polishing processes,” j. electroche m. soc. , vol. 144, no. 3, pp. 1121-1127, 1997. [4] g. kermouche, j. rech, h. ha md i, and j. m. bergheau, “on the residual stress field induced by a scratching round abrasive grain,” wear, vol. 269, no. 1-2, pp. 86-92, may 2010. [5] c. landesberger, c. paschke, and k. bock, “influence of wafe r grinding and etching techniques on the fracture strength of thin silicon substrates ,” advanced materials research, vol. 325, pp. 659-665, 2011. [6] k. gu rnett and t. adams, “ult ra-thin semiconductor wafe r applications and processes,” iii-vs rev iew, vo l. 19, pp. 38– 40, 2006. [7] z. j. pei, g. r. fisher, and j. liu, “ grinding of silicon wa fers: a rev iew fro m historica l perspectives,” international journal of machine tools and manufacture, vol. 48, pp. 1297-1307, 2008. [8] j. wang, p. shrotriya, and k. s. kim, “surface residual stress measure ment using curvature interfero metry,” expe rimental mechanics, vol. 46, pp. 39-46, 2006.  advances in technology innovation , vol. 1, no. 1, 2016, pp. 25 27 25 copyright © taeti buckling experiment on anisotropic long and short cylinde rs atsushi takano department of mechanical engineering, kanagawa university , yokohama, japan. received 02 february 2016; received in revised form 22 march 2016; accept ed 25 march 2016 abstract a buckling e xpe riment was performed on anisotropic, long and short cylinders with various radius -to-thickness ratios . the 13 cylinders had symmetric and anti-sy mmetric layups, were between 2 and 6 in terms of the length-to-radius ratio, between 154 and 647 in radius -to-thickness ratio, and made of t wo kinds of carbon fiber re inforced plastic (cfrp) prepreg with high or low fiber modulus. the theoretical buckling loads for the cylinders were calculated fro m the previously published solution by using linear bifurcation theory considering layup anisotropy and transverse shear deformat ion and by using deep shell theory to account for the e ffect o f length and compared with the test results. the theoretical buckling loads for the cy linders were calculated fro m the previously published solution by using linear bifurcation theory considering layup anisotropy and transverse shear deformation and by using deep shell theory to account for the effect of length. the knockdown factor, defined as the ratio of the e xperimental value to the theoretical value, was found to be between 0.451 and 0.877. the test results indicated that a large length-to-radius ratio reduces the knockdown factor, but the radius -to-thickness ratio and other factors do not affect it. keywor ds : buckling, cylinders, anisotropy 1. introduction many buckling tests have been performed on orthotropic and anisotropic cylinders , which a re often made fro m cfrp. as summarized by the author [1] and as shown in fig. 1, the values of the knockdown factor have been calculated from these tests are scattered between 50% and 100% . fig. 1 knockdown factor and r/t [1] note that the theoretical buckling loads were calculated by using the solution with the least amount of simplificat ion in the linear b ifurcation theory from a mong the previously published solutions [2]. the previous experiments and evaluations , however, main ly concentrated on the effect of the radius to thickness ratio (r/t). recently, long cfrp cy linders have started to be used in large satellites [3], but no information is yet availab le on the effect of length on the knockdown factor. thus, buckling tests were performed on long and short cfrp cy linders, and the results were used to calculate the knockdown factors. 2. method 2.1. mak ing specimens the cfrp prepregs used to ma ke the cylinders were tr350j075s, el = 114.7gpa (hereafter ca lled “t r”) and hsx350c075s, el = 260.3gpa (hereafter called “hsx” ); both were made by m itsubishi rayon co., ltd. here , el is the longitudinal young’s modulus measured in a tensile test, and the transverse young’s modulus et and shear modulus glt we re assumed to be 6 gpa and 4gpa, respectively. * corresponding aut hor, email: at akano@kanagawa-u.ac.jp advances in technology innovation , vol. 1, no. 1, 2016, pp. 25 27 26 copyright © taeti the test specimens (cfrp cylinders) we re made in-house by hand-layup. cfrp prepregs were layered manually on an alu min iu m mandre l (150 mm in dia meter), and heat shrinkable tape was wound around them. the layup sequence is shown in table 1. the cfrp prepregs on the mandrel were therma lly cured in an oven kept at 130°c for 2 hours. the cured cylinders were cut by grinders, and both ends were bonded to steel rings by epoxy adhesive. to prevent therma l residual stress on the cylinders, the curing of the epoxy adhesive was conducted at room temperature. twelve sheets of three-axis strain gages were bonded on the top and bottom (5 mm fro m the steel end rings) and middle (the center of the length) of the cylinders in the circumferentia l direction, each separated by 90°. 2.2. test method a typical compression test configuration is shown in fig. 2. section paper was la id under the specimen to align the centers of the cylinder and the universal testing instrument (shimad zu a g-i 100kn). to make the load uniform, a rubber sheet and a silicone rubber sheet were laid on the top end of the cylinder, a 20 mm thic k stainless plate was put over the sheets, and a rubber sheet was la id on the bottom end of the cy linder. to minimize the offset of the load, a sma ll compression load was applied, and the strain outputs in the middle were checked. when a significant difference between the strains was observed, the location of the cylinder on the universal testing instrument was adjusted. when the difference between the strains no longer changed or a la rge offset was observed between the centers of the cylinder and the universal testing instrument, however, the difference between the strains was ignored and the test proceeded to the next step. fig. 2 photograph of buckling test the half leve l co mpression test (applying half the e xpected buckling load, including 0.5 of the knockdown factor) was performed to check for anoma lies in the data. after that, the full level co mpression test was performed to buckle the cylinder. load was applied until the cross -head displacement reached 1.5 of the buckling displacement. 3. results and discussion typical load and displace ment results are shown in fig. 3 and 4. fig. 3 tr 6 ply, l/d=1, gap allowed. fig. 4 tr 6 ply, l/d=3, gap allowed. table 1 summa rizes the test results. note that an e xtre me ly low knockdown factor 0.397 was caused by 8mm of offset load , and the modified knockdown factor considering the offset is 0.481. accordingly, the knockdown factors are scattered between 0.451 and 0.877. the knockdown factors of the longer cylinders are lowe r than those of the shorter cylinders, while other factors (symmetric or anti-symmetric, ply gap, fiber modulus and l/r) seem to have no effect on the trend. thus , a regression analysis with categorical variab les was conducted to check the effects . the results are shown in table 2. a regression analysis can be used instead of an analysis of variance (anova) when the samp le size is unbalanced like in this case. he re, r/t (thickness) and the symmetric or anti-symmet ric factor a re advances in technology innovation , vol. 1, no. 1, 2016, pp. 25 27 27 copyright © taeti mu lticollinear; thus , the symmetric or anti-symmetric factor should be e xc luded fro m the regression analysis. table 1 summary of the test results table 2 regression tables the mean of the knockdown factors is 0.611 with 0.02% of p-value, and it is smalle r than 5% (standard statistical criteria); hence, the value is statistically significant. other effects , however, are not significant because their p-values are higher than 5%. only the p-va lue of l/r, which is 19.23%, is smaller than the others . the mean of the knockdown factors is 0.627 for l/r=2 and 0.537 for l/r=6. the s ma lle r va lue is 14.4% smaller than the larger va lue. thus , the length may affect the knockdown factor. to c larify its effect, more buckling tests are required. 4. conclusions thirteen buckling tests were performed on symmetric and anti-sy mmetric, long and short cylinders to investigate effect of varying the length. the diffe rence in the mean knockdown factor between lengths was found to be 14.4% . a regression analysis , however, indicated that the diffe rence was not statistically significant. more buckling tests considering other factors should be conducted in order to find the cause of the scatter in the knockdown factor. references [1] a. takano, “statistical knockdown factors of buckling anisotropic cylinders under axial compression,” journal of applied mechanics, vol. 79, 051004, pp 1-17, 2012. [2] a. takano, “improvement of flügge’s equations for buckling of moderately thick anisotropic cylindrical shells,” aiaa journal, vol. 46, no. 4, pp. 903-911, 2008. [3] y. takano, t. masai, h. seko, a. takano, and m. miura, “development of the lightweight large composite-honeycomb-sandwich central cylinder for next-generation satellites,” aerospace technology japan, vol. 10, pp. 11-16, 2012. theory p a [n] test p a [n] tr 6 ply l/r=2 (-70/70/0/0/70/-70) 0.488 136 21983 13197 0.600 gap tr 6 ply l/r=4 (-70/70/0/0/70/-70) 0.488 287 21986 11870 0.540 gap tr 6 ply l/r=6 (-70/70/0/0/70/-70) 0.488 436 21987 11537 0.525 gap hsx 6 ply l/r=2 (-70/70/0/0/70/-70) 0.349 136 21123 12997 0.615 gap hsx 6 ply l/r=6 (-70/70/0/0/70/-70) 0.349 443 21873 12647 0.578 gap hsx 3 ply l/r=2 (-70/0/70) 0.175 148 2060 1806 0.877 overlap hsx 3 ply l/r=6 (-70/0/70) 0.175 443 2061 1437 0.697 overlap tr 6 ply l/r=2 (-70/70/0/0/70/-70) 0.488 136 21983 12665 0.576 overlap tr 6 ply l/r=6 (-70/70/0/0/70/-70) 0.488 436 21987 8738 0.397 * overlap hsx 6 ply l/r=2 (-70/70/0/0/70/-70) 0.349 136 21123 10366 0.491 overlap hsx 6 ply l/r=6 (-70/70/0/0/70/-70) 0.349 436 21873 9875 0.451 overlap hsx 2 ply l/r=2 (-50/50) 0.116 136 964 583 0.605 overlap hsx 2 ply l/r=6 (-50/50) 0.116 436 944 464 0.492 overlap *not e: 8mm of load offset was observed and modified knockdown fact or considering t he offset load is 0.481. specimen name layip sequence thickness t [mm] length l [mm] buckling load ply gap or overlap knockdown factor coefficients std error t p-value lower 95% upper 95% intersept 0.611 0.095 6.459 0.02% 0.393 0.829 gap/overlap -0.052 0.080 -0.646 53.62% -0.236 0.133 tr/hsx 0.047 0.087 0.542 60.28% -0.154 0.249 l /r -0.025 0.017 -1.424 19.23% -0.065 0.015 r /t 0.000 0.000 0.782 45.70% 0.000 0.001  advances in technology innovation , vol. 1, no. 2, 2016, pp. 46 49 46 copyright © taeti study of injection molding warpage using analytic hierarchy process and taguchi method dyi-cheng chen * , chen-kun huang department of industrial education and technology , national changhua university of education, changhua, taiwan. received 22 february 2016; received in revised form 09 april 2016; accept ed 12 april 2016 abstract this study integrated analytic hierarchy process and taguchi method to investigate into injection mold ing warpage. the warpage important factor will be elected by analytic hie rarchy process (ahp), the ahp h ierarchy analysis factor fro m docu ments collected and aggregate out data, then through the expert questionnaire delete lo w weight factor. finally, we used taguchi quality engineering method to decide injection molding optimized co mbination factors. furthermo re, the paper used injection pressure, holding pressure, holding time, mo ld temperature to analyze four factors, three levels taguchi design data. moreov er, the paper discussed the reaction of each factor on the s / n ratio and analysis of variance to obtain the best combination of minimal warpage. keywor ds : injection molding, analytic hierarchy process (ahp), taguchi method 1. introduction plastic molding methods are injection molding, extrusion molding, blow molding, co -injection molding method, gas -assisted molding method, of which the injection molding method is the most widely used plastic molding technology . kamaruddin [1] used taguchi to improve mixed plastic products . the analysis of the results shows that the optimal combination for low shrinkage are low melting temperature, high injection pressure, low holding pressure, long holding time and long cooling time. shuaib [2] performed to determine the factors that contribute to warpage for a thin shallow injection -molded part. the process used taguchi and anova technique. the result shows that by s/n response and percentage contribution in anova, packing time has been identified to be the most significan t factors on affecting the warpage on thin shallow part. radhwan et al. [3] applied taguchi method for the optimization of selected process parameters such as the mold te mperature, me lt te mperature, packing pressure, packing time, and cooling time. the s/n ratio and analysis of variance were utilized to see the most significant factors contributing to shrinkage. nasir et al. [4] designed mold in single and dual type of gate in order to investigate the deflection of warpage for thick component in injection molding process. opasanon and lertsanti [5] imple mented the analytic hierarchy process (ahp) to evaluate and rank the importance of the logistics issues according to the needs and requirements of the company’s policy makers. four criteria considered in the ahp include cost, responsiveness, reliability, and utilization. kil et al. [6] study to identify the ma jor variables identified as important for considering the stabilization of slope revegetation based on hydro seeding applications and evaluate weights of each variable using the analytic hie rarchy process (ahp). this study integrated analytic hierarchy process and taguchi method to investigate into injection molding warpage. 2. results and discussion of ahp and taguchi method 2.1. analytic hierarchy process (ahp) in this study, injection mold ing gather relevant informat ion, collate and analyze the relevant factors. as shown in the present study hierarchica l structure shown in fig. 1. in this study, interviews the way interviews professors fro m this and related industry contains several interviews with scholars states to carry out * corresponding aut hor, email: dcchen@cc.ncue.edu.tw http://xueshu.baidu.com/s?wd=author%3a%28nasir%2c%20s.m.%29%20&tn=se_baiduxueshu_c1gjeupa&ie=utf-8&sc_f_para=sc_hilight%3dperson advances in technology innovation , vol. 1, no. 2, 2016, pp. 46 49 47 copyright © taeti private visits to the volume. fig. 1 hierarchy architecture diagram using microsoft excel software to analyse all questionnaires, one of the factors to estimate the impact of the overall configuration of the inner surface of the weight value, you can understand the factors within a ll facet degree of importance the key factors for overall warpage of inject ion mo lding, as a result as shown in table 1. table 1 the overall weight table main weight secondary eight overall weight a. pressure a1 0.52 0.25 (1) 0.48 a2 0.34 0.16 (2) a3 0.12 0.06 (8) b. time b1 0.44 0.13 (3) 0.29 b2 0.19 0.05 (9) b3 0.22 0.06 (7) b4 0.13 0.03 (10) c. temperature c1 0.29 0.06 (6) 0.22 c2 0.34 0.07 (5) c3 0.35 0.05 (4) 2.2. taguchi design of experiment (1) choose quality characteristics in order to measure the output quality and characteristics fro m desired value, taguchi has utilized the signal-to-noise ratio; s/ n. s/n ratio also used to classify the results and evaluates them to determine the optimu m para meters. there are three s/n ratio’s characteristics; the nomina l the better, the smaller the better and the higher the better. since this research is carried to reduce warpage, the smaller the better characteristic has been chosen and it is expressed as :          n i i y n ns 1 21 log10/ (1) yi represents the observation, n is the nu mber of tests in one trial. (2) choose control factor as shown table 2, there are four factors identified to be the para meters in this research. they are the injection pressure (a), packing pressure (b), packing time (c),and the mold temperature (d).taguchi method is used to analyze these four injection mo lding process parameters based on three-level design of e xperiments and orthogonal array l9(3 4 ) is created . the levels, factors and orthogonal array variance and the combination are shown in table 3 respectively. table 2 selected factors and levels factor level 1 level 2 level 3 a. injection pressure 100 110 120 b. packing pressure 65 75 85 c. packing time 7 9 11 d. mold temperature 80 90 100 table 3 combination of parameters in orthogonal array variance a b c d 1 100 65 7 80 2 100 75 9 90 3 100 85 11 100 4 110 65 9 100 5 110 75 11 80 6 110 85 7 90 7 120 65 11 90 8 120 75 7 100 9 120 85 9 80 (3) experimental data analysis after molde x3d analysis, the results of the e xperiment to measure out the a mount of warpage calcu lated s/n ratio, calcu lated by equation (1) s/ n rat io of each group, as shown in table 4. in table 4 can be obtained by injection mo lding of each factor on the table and the amount of warpage of the reaction the reaction diagra m, as shown in table 5 and figure 2. in smaller quality characteristics s / n rat io greater the better quality characteristics, according to tables and graphs can identify the best factor level co mbination a3b2c3d1, injection pressure 120mpa, packing pressure 75mpa , packing time 11sec, mold temperature 80℃. table 4 results of s/n and warpage of results a b c d warpage s/n advances in technology innovation , vol. 1, no. 2, 2016, pp. 46 49 48 copyright © taeti (mm) ratio 1 1 1 1 1 0.0684 23.2989 2 1 2 2 2 0.0688 23.2482 3 1 3 3 3 0.0722 22.8293 4 2 1 2 3 0.0788 22.0695 5 2 2 3 1 0.0555 25.1141 6 2 3 1 2 0.0764 22.3381 7 3 1 3 2 0.0621 24.1382 8 3 2 1 3 0.0830 21.6184 9 3 3 2 1 0.0602 24.4081 ave. 0.0694 23.2292 table 5 the response table of s/n ratio a b c d level 1 23.13 23.17 22.42 24.27 level 2 23.17 23.33 23.24 23.24 level 3 23.39 23.19 24.03 22.17 effect 0.26 0.16 1.61 2.10 rank 3 4 2 1 combination a3 b2 c3 d1 fig. 2 s / n ratio reaction (4) analysis of variance (anova) analysis of variance (anova, analysis of variance) is ma inly determined change of each factor on the quality characteristics variation effect, which is another way to find the most influential factor for the entire e xpe riment, in order to assess the experimental error. table 6 initia l variance analysis results of the present experiment. table 6 the first analysis of variance factor ss dof variance a 0.1173 2 0.05866 b 0.0438 2 0.02189 c 3.8826 2 1.94131 d 6.6239 2 3.31196 other 0.0000 0 0.000 total 10.6676 8 1.33345 in table 6 that the variance b factor holding pressure variation co mpared to the nu mber of other factors to low, so the integration of this factor to the error vector for a second analysis of variance. table 7 shows d factor for the ent ire injection mo ld ing mo ld te mpe rature have a significant impact, accounting for 62.1% of the overall e xperiment, followed by c packing time and a. injection pressure. in table 5 choose the best combination a3b2c3d1 for mo ld flow analysis again to verify that the best combination of para meters, the optimu m a mount of warpage results about 0.0549 mm are shown in table 8. fig. 3 shows the simulation of best combination a3b2c3d1. table 7 the second analysis of variance factor ss dof var. f-ratio confi dence ρ% a 0.117 2 0.0586 2.68 72.81% 1.09 b pooled c 3.882 2 1.9413 88.69 99.99% 36.4 d 6.623 2 3.3119 151.3 99.99% 62.1 error 0.043 2 0.0218 *at least 99% confidence total 10.66 8 table 8 best combination of parameters a b c d warpage combination 120 mpa 75 mpa 11 sec 80 ℃ 0.0549 fig. 3 simulation of combination a3b2c3d1 3. conclusions the paper used taguchi quality engineering method to decide injection mo lding optimized combination factors. the results have shown that: (1) the best factor level co mbination a3b2c3d1, inject ion pressure 120mpa, packing pressure 75mpa, packing time 11sec, mo ld temperature 80 ℃; (2) the entire injection mo lding mold te mperature have a significant impact, accounting for 62.1% of the overall e xperiment; and (3) the optimu m a mount of advances in technology innovation , vol. 1, no. 2, 2016, pp. 46 49 49 copyright © taeti warpage results about 0.0549 mm. references [1] s. ka ma ruddin, “application of taguchi method in the optimization of inject ion mould ing para meters for manufacturing products from plastic blend,” international journal of engineering & technology, vol. 2, no. 6, dec. 2010. [2] n. a. shuaib, “warpage factors effectiveness of a thin shallow injection-molded part using taguchi method,” international journal of engineering & technology ijet-ijens, vol. 11, no. 1, feb. 2011. [3] h. radhwan, m. t . mustaffa, a. f. annuar, h. azmi, and m. z. zakaria, “an optimization of shrinkage in injection mo lding parts by using taguchi method,” journal of advanced research in applied mechanics, vol. 10, no. 1, pp. 1-8, 2015. [4] s. m . nasir, k. a. is mail, z. shayfull, and n. a. shuaib, “ co mparison between single and mu lti gates for min imization of wa rpage using taguchi method in injection mold ing process for abs material,” key engineering materials, vol. 594-595, pp. 842-851, 2013. [5] s. opasanon and p. le rtsanti, “impact analysis of logistics facility re location using the analytic hierarchy process (ahp),” international transactions in operational research, pp. 325–339, 2013. [6] s. h. kil, k. l. dong, j. h. kim, m. h. li, and g. neman “utilizing the analytic hierarchy process to establish weighted values for evaluating the stability of slope revegetation based on hydroseeding applications in south korea,” sustainability, vol. 8, p. 58, 2016.  advances in technology innovation , vol. 1, no. 1, 2016, pp. 16 20 16 copyright © taeti innovation management of patent commercialization yi-ping lee * department of business administration national chung hsing university , taichung, taiwan. received 01 april 2016; received in revised form 10 may 2016; accept ed 13 may 2016 abstract patent commerc ialization is an important issue to most country, for only 0.3 percent of patent have comme rc ialized in the world. researchers try hard to enhance the rate of patent comme rcia lization. this paper focus on innovation of organization and management to enhance the rate of patent comme rcia lization. which is based on team wo rk and plan to promote step by step, for most patent owner develop patent but do not equally put effort on patent commerc ialization, and that is waste precious time and resource. this research directly analysis government published patent secondary data and the other part is questionnaires survey to patentees , with which to find best way of patent commercialization. keywor ds : patent commercializat ion, patent management, strategy of commercialization 1. introduction corresponding to the number of patents increased rapidly in the whole world during recent years , this phenomenon imply that the world people are pay attention to the influence of intellectual property rights . most country build a good national patent environment to provide creative and co mpetitive resource, the less barriers of patent commerc ialization re moved, and the higher rate of patent comme rcia lization will be. patent system protect inventor’s interest from patent infringe ment, which means one who produce or sale patent product without authorization, he will be punished by pay loss of patentee’s damage or more seriously treble fined for h is behavior of patent right infringement. another way of patent use is the patent owner can e xerc ise his patent right in attack and defense in a co mpete war, he can use the patent right to beat competitors in the ma rket, the value of patents value may up to millions dollars, or much more than that, upper concepts highlight the reason and importance of patent commercialization. patent number keep on increase but the rate of patent commerc ia lizat ion still low, average rate of patent commerc ialization in the whole world are only near 0.3 percent. innovation is a global trend for most inventor, patentees pass the patent exa mination by long time research and continuously improve ment to inventions , but after a lot challenge and then gain the patent right. the process of patent application need about 3 more years, if the inventor is lucky enough, he may pass patent exa mination in first time, or he have to file roundtrip several times . patentee have to pay an annual fee annually to ma intain his patent right is another barrier of patent commercialization. why patentee shelved commerc ialization action after owned the patent? most inventors are top cleaver in their do ma in of industry, they can solve very comple x technic proble m, because patent should satisfy factors of useful, novelty and non-obvious, all these three factors have to testified by the patent exa miner through precedent exa mination, there can’t have any precedent be find in any other patent database of the whole world, or e lse the patent exa miner will turn down the patent application. every cases of patent application are need a lot of resources from government and individuals , all front end process before patentee own his patent, regardless technic development or patent application are need to pay money, furthermore, patent ma intenance fees pay annually increasing is money consuming. since patent product’s life cycle need continuously pay money fro m the beginning of patent invention to the patent right perished. an obscure question need to find the answer and why? the question is the patent pay compared with the outcome of va lue is e xtre me ly inequivalent, wh ich imply that the patentee gain * corresponding aut hor, email: yiping727@yahoo.com.tw advances in technology innovation , vol. 1, no. 1, 2016, pp. 16 20 17 copyright © taeti nothing in the end of patent life cyc le, are they really want to pay price and without gain any benefit? we can trace back to the meaning of patent right to find the answer, here the answer should be very clear, the patent right originally mean the patentee have the right to e xc lusive others to infringe his patent right by the government authority, e xcept that, the patentee have the right to sale the patent right, produce the product, if the patentee can do one of upper things, he will not be gain nothing fro m patent, this identify that most patentees are want gain benefit fro m patents but inability to do something have his patent commercialized. most patentee was numb about keep on pay fees annually after ga in patent right, outcome of patents is zero year after year, in particu lar, they still have no active action of patent comme rcia lization, which is not match the rule of investment, what a re they work for is the question, how to solve the proble m is the goal of this paper. this paper focus on the secondary data of government intellectual property to analysis and find partia l factors of patentee’s inability of patent commercialization. 2. literature review and hypotheses china is in response to the global trends, they establish a an modern patent system in recently years, which is pave with international standards , with which to imp ly patent protection (lulin gao, 2008), it is difficu lt to have patents comme rcia lized is a long term phenomenon in the world, may the reason of inadequate equipment or lac k of incentive for comme rcia lization (avimanyu datta, 2012), comme rcia lization of university inventions (yonghong wu, 2013) basically patent inventors almost are e xcellent and cleaver, but in the link of patent comme rcia lization are transfixed and s ilently endure no recovery and obtained repeatedly in long-term of non-use patent investment, this situation is not only a country’s problem, there are almost the same in the world, wh ich waste a large a mount of national resources of e xa mination and indiv idual’s monetary or time resources, this is really an pract ical and academic issue, for the benefit of all country and patentee, unravel the patent commerc ialization problem and do good to society and patentee in the same time , this artic le as the results prevail for understanding patent commerc ialization of patentees. first, patent commerc ialization factors clarify, second, succeeded experience highlighting. 3. research frame and hypothesis this paper have the operational definition on research facets as follo w, wh ich include patent number, patent of comme rcia lization, organization, p ro motion action and investment of fund and time a re co mputable fro m data or questionnaire. all patent data are collect fro m intellectual property office, m inistry of economic affairs (moea), fo r some of the private informat ion should collect fro m individuals, such as investment of fund or time , and both of upper two ma in factors are set to computable number in different facets. patent comme rcia lization was composed of several factors, which inc lude human resource of promotion, action of pro motion, investment of fund and time etc. all factors are affect the patent comme rcia lization rate, thus, hypothesis as follow, fo r patent comme rcia lizat ion was calculated on patent and comme rcia lized patent, thus, the quantity of patent in equation’s denominator will affect the result of patent comme rcia lization rate. for patent comme rcia lization is so difficult, thus it should be promoted by organized tea m, basically, the more e mp loyee me mbe r work together, the outcome of that team will be better. h1: the more quantity of patent number in patent data, the lowe r o f patent commercialization rate. h2a: the more quantity of e mp loyee number in organization of patent commerc ialization, the higher of patent commerc ia lizat ion rate. h2b: the more quantity of action in patent comme rcia lization, the higher of patent commercialization rate. h3a: the more quantity of fund in investment of patent comme rcia lization, the higher of patent commercialization rate. h3b: the more quantity of time in investment of patent comme rcia lization, the higher of patent commercialization rate. advances in technology innovation , vol. 1, no. 1, 2016, pp. 16 20 18 copyright © taeti 4. method due to research the individual patents comme rcia lization, this research focus on the process and case of patent comme rcia lization, because the rate of 0.3 percent patent comme rcia lization is now e xist the same in most country, thus, every successful e xperience will be very precious, we visit the patentee to find detail and factors of these patent commercialization(nicole ziegler, 2013). this research focus on patent owner, up to now, all cases are from ta iwan patentee, and the patentee have the level of own about 100 patents, and the rate of patent comme rcia lizat ion over 5 percent, compare with the common patent comme rcia lization rate with 0.3 rate, 5 percent could be a very successful and precious e xperience, fo r these case much better than common cases about ten times of comme rcia lization rate. the method of analysis was adopt statistical package for the socia l sciences (spss) in analysis, secondary data and questionnaire was use in research re liability and validity, we e xclude the invalid samp le. va riab le define and evaluation in this research was use patent number of patentee owned, individual patent commerc ia lizat ion rate was set and calculated fro m government published patent data number (pdn) divided by number of patentee comme rcia lized patent (pcn), the sa me as total number of patent comme rc ialization rate was averaged upper two factors to gain comple x commercialization rate. pcr=pn/pcn (1) the patent commerc ialization rate was affected by patent promotion (pp), and which was composed by two factors, first, promotion organization, second, promotion action, and promotion organizat ion and management, promotion organizat ion was define as the have tax pay for salary and number of patent promotion e mp loyee (ppe), action of promotion(ap) was define as have record of application of govern ment subsidiary (a gs), records of patent t transaction (pt) in platfor m or produce of patent product(ppp), pro motion time consuming(pt c), invest in capital(ic) etc. patent strategy was adapt from questionnaire to the patentee, such as how much fund put to comme rcia lization (fpc) his patent, and how many patent is plan to commerc ialization (ppc), ppe=sa*ne (2) ap=ags+pt+ppp+pt c (3) pcs=ic+it (4) 4.1. secondary data analysis all cases are selected by the level of the patentee have higher rate of patent comme rcia lization, if a case can achieve this level, this kind of case have ten times of pcr than usually patentee have, and far e xceed most country’s pcr. thus, for two reason of this research design, first, near 100 patents owner and 5 percent of pcr a re strictly level for most patent owner, most patent owner in the world a re in the situation of pcr lo wer than 0.3 percent, thus, the level of 5 percent pcr should be reliable fo r the test and study, first of all, in the first section, we choose the r&d tea m me mbe r less than 10 e mp loyee, for most patentee are individual inventor, we will limited our range for the study be exactly right to most patentee, thus, in the first section the participant will not be a firm level r&d operation. 4.2. primary questionnaire data analysis through phone call and e-ma il to confirm the visit date and time , then have face to face questionnaire discussion with the patentee, furthermore, we visit the patent product factory and discuss with the manager and sales, questionnaire was design as likert 5.0 scale, this process can be part of further internal empirical. 4.3. cases study case 1 have a research and development team, which is organized by about 5 people and led by a pro fessor/dr. who is a pro fessor of national university, and serve in depart ment of brain neurology, through over 26 years of work and research from day to night, the professor/patentee own and partially involved 93 patents, most patents are medical equip ment, he chaired his research team, focus on invention and patent design. case 2 have a research and development team, wh ich is organized usually by about 2 to 5 people and led by a professor who is a professor of national university too, and serve in depart ment of material science, through about 11 year research, the patentee own and part ially involved 172 patents. advances in technology innovation , vol. 1, no. 1, 2016, pp. 16 20 19 copyright © taeti case 3 do not have a research and development team, who usually innovation by himself, through over 12 years of research, the patentee own and partially involved 72 patents, all patent are through long term and continuous work and study, keep on try and error to win the patent, after this stage, patentee develop an partner owned product manufactory. table 1 patent case analysis patentee a b c patent domain medical equipment material science multiple domain patent quantity 93 172 72 new invented 85 172 24 new design 8 0 25 new type 0 0 1 patentee research duration 1999-2015 2004-2015 2004-2016 patentee research year 26 11 12 commercial quantity 5 7 32 commercial rate 18.50% 24% 44% highest invention quantity in a year 15 45 22 patent applied by university 76 150 0 patent applied by group 7 15 23 patent applied by individual 10 7 49 average year to own patent 3.2year 4.5year 7.2month 5. results and discussion faith and determination be possible o obstacle to patentee in attempt to promote patent, plan and execute are main affect factors, patentee’s faith of patent commercialization, faith and commitment of a person's attempt to complete plan (ajzen1985). possible future research design for this is sue, the research design may focus on case study and questionnaire, include government officer, business manager, stock investors and experts, for all these people are important member patent commercialization. after the research of data analysis, a successful patent comme rcia lizat ion are have a team to support the action of patent comme rcia lization, which include research and development, promotion me mbe r of comme rcia lization, financia l support and ma rket ing. if with a tea m operation continuously, especially on the promotion work, then, patent would be possible be commerc ialized, else, most patentee focus on research and development, there almost have no resource and promotion action have put in, thus, lack of pro motion action in commerc ialization is the ma in reason exist low rate of patent commercialization . 6. conclusions in this paper, three new way of the patent comme rcia lization have been generated in this research. the feasibility of the new method is verified by. the result has shown that the new designs can produce a more wide range of non-uniform output motion than the existing design. therefore, they are better alternat ives for driving a variable speed input mechanism. references [1] a. datta, r. reed, and l. jessup, “factors affecting the governance of innovation comme rcia lization: a theoretica l mode l,” journal of business and management, vol. 18, no. 1, 2012. [2] gera ld, ude ll, m. hignite, “ne w product comme rcia lization: needs and strategies ,” the journal of applied management and entrepreneurship, vol. 12, pp. 75-92, 2007. [3] l. gao, research technology management, china’s patent system and globalization, 51.6, pp. 34-37, 2008. advances in technology innovation , vol. 1, no. 1, 2016, pp. 16 20 20 copyright © taeti [4] n. l. vanderford, l. t. weiss, “he idi l. weiss ic-based cancer research co mme rcia lization,” public libra ry of science (plos), vol. 8, no. 8, p. e 72268, 2013. [5] n. mac ias, c. knowles, f. ka mke , and a. kutnar, “co mmerc ialization potential of viscoelastic therma l co mpressed wood: insights from the us forest products industry,” forest products journal, vol. 61, no. 7, pp. 500-509, 2011. [6] n. zieg ler, f. ruether, m. a. bader, and o. gassman, “creating value through external intellectual property comme rc ialization: a desorptive capacity view,” the journal of technology transfer, vol. 38, no. 6, pp. 930-949, 2013. [7] y. wu, e. w. welch, and w. l. huang, “co mme rcia lizat ion of university inventions: individual and institutional factors affecting licensing of university patents ,” technovation, vol. 36-37, pp. 12-25, 2013. [8] r. ghafe le and b. gibe rt, “ip comme rcia lization tactics in developing country contexts,” journal of management and strategy, vol. 5, no. 2, pp. 1-15, 2014. [9] h. s. yan and w. r. chen, “on the output motion characteristics of variable input speed servo-controlled slider-c rank mechanis ms,” mechanism and machine theory, vol. 35, pp. 541-561, 2000. [10] h. s. yan, m. c. tsai, and m. h. hsu, “an e xperimental study of the effects of ca m speeds on cam-followe r systems,” mechanism and machine theory, vol. 31, pp. 397-412, 1996.  advances in technology innovation, vol. 2, no. 1, 2017, pp. 13 17 13 the study of the microbes degraded polystyrene zhi-long tang, ting-an kuo, and hsiao-han liu * department of biological science & technology, i-shou university, kaohsiung, taiwan, roc. received 30 january 2016; received in revised form 28 march 2016; accepted 30 march 2016 abstract under the observation that tenebrio molitor and zophobas morio could eat polystyrene (ps), we setup the platform to screen the gut microbes of these two worms. to take advantage of that tenebrio molitor and zophobas morio can eat and digest polystyrene as its diet, we analyzed these special microbes with ps plate and ps turbidity system with time courses. there were two strains tm1 and zm1 which isolated from tenebrio molitor and zophobas morio, and were identified by 16s rdna sequencing. the results showed that tm1 and zm1 were cocci-like and short rod shape gram-negative bacteria under microscope. the ps plate and turbidity assay showed that tm1 and zm1 could utilize polystyrene as their carbon sources. the further study of ps degraded enzyme and cloning warrants our attention that this platform will be an excellent tools to explore and solve this problem. keywords: polystyrene, tenebrio molitor, zophobas morio, 16s rdna sequencing 1. introduction currently, the most popular method to decompose polystyrene was the thermal decomposition method, but this method would produce large amounts of dioxin and cause serious pollution to the environment [1, 2]. all these thermal methods faced an interaction with the polymer breakdown products which could lead to the formation of well-known dioxin precursors such as halogenated phenols should also be taken into account when evaluating the impact of bfr-treated materials on the environment [1]. however, the recent research directions of polystyrene degradation will be that use the way of biological decomposition, and these methods would reduce the waste polystyrene and dioxin. the approach of biodegradation of polystyrene was initiated in 1979, all bio-species show a very low degraded ability with c 14 -labelled plastic[3]. the best result from sludge microbes degraded the polystyrene for 0.57% after 11 weeks. the other species degraded less than 0 to 0.3% after 35 days. most organisms were not able to degrade ps. to take advantage of ps-eating mealworms, yang et.al. test the consumption of the ps[4]. furthermore, they explored the gut of mealworms, and identified of novel bacteria yt2 which belongs to exiguobacterium sp. and other 12 isolates based on its 16s rrna sequences[5, 6]. however, their methods for further identification and characterization of ps degrading enzyme of these microbes were not easy. our motivation of this study is to establish a more efficient and easy platform to study the ps degraded microbes within these worms. we utilize the mealworms (tenebrio molitor) as well as superworms (zophobas morio) for screening the ps eating microbes. furthermore, the microbes in this study were decomposing more quickly than previous studies. the setup of this novel screening platform will be an excellent tool to explore and solve the further study of ps degraded enzyme. the aims of this study is to isolate the polystyrene degrading microbes from the gut of t. molitor and z. morio, and the following methods will be used to exam ps-degrading ability of these microbes[7]. 2. method 2.1. preparation of ps emulsion and ps plate to prepare the ps emulsion, we dissolve the 7.5g ps powder by 50 ml chloroform. then transferred this solution to a 250 ml glass bottle with 100 ml basal medium (1.6g k2hpo4,1g (nh4)2so4,0.2g kh2po4,0.2g mgso4·7h2o,0.1g nacl,0.015g cacl2,0.01g feso4·7h2o/1000 ml) and biodegradable detergent[7]. next transferred the ps emulsion to 4℃ for 2 days. * corresponding author, email: hansliu@isu.edu.tw advances in technology innovation, vol. 2, no. 1, 2017, pp. 13 17 14 copyright © taeti after 2 days, we put the ps emulsion in hood to volatilize the organic solvents. the preparation of ps agar plate is to transfer 50 ml ps emulsion to a 100 ml glass bottle with 50 ml basal media and 0.75 g agar, sterile for 15 min at 121℃ and pour plate as general agar plates[7]. 2.2. isolation of ps degrading microbes a group of 20 mealworms and superworms separated fed with polystyrene as a sole diet for 3 weeks was collected, and washed out as a gut microbes’ suspension with basal medium. at first these gut microbes’ suspension was plating on the basal agar plate incubating for 24 h at 37℃ in aerobic and anaerobic conditions, and the microbes on the basal agar plate are collected. and then transferred the microbes on the basal agar plate to ps plate with yeast extract incubating for 24 h at 37℃ as above conditions. finally, the microbes on the ps agar plate are collected and the pure colonies were preserved for the following experiments. 2.3. bacteria strain characterization and 16s rdna sequencing there were two strains tm1 and zm1 which isolated from tenebrio molitor and zophobas morio, and were identified by gram staining, tsi agar test and 16s rdna sequencing. the genomic dna of the bacteria strains for 16s rdna sequencing were extracted by genomic dna extraction kit (thermo fisher scientific). the 16s rdna gene was amplifying by 27f-degl (5′-agrgttygatymtggctcag-3′) and 1492r (5′-ggttaccttgttacgactt-3′)[8]. the sequences were compared with 16s ribosomal rna sequences on the 16s rrna database using the basic local alignment search tool (blast) in ncbi. 2.4. turbidity assay turbidity assay was a method following the measurement that chua et.al. used[7]. this method could measure the ps-degrading ability of these microbes, and it was a quantitative indicator of the ps degradation as well. to quantify ps degrading activity, 1 ml of an overnight liquid culture of tm1 and zm1 in ps medium was mixed with 20 ml of basal medium containing 0.5 ml 0.75 % ps emulsion and incubated at 37℃ for following time courses. the ps degrading activity was measured spectrophotometrically at 600 nm. to observe the utilization of ps emulsion and ps degrading microbes’ suspension will be at 0.2, 4, 6, 24, 48 h respectively. 3. results and discussion 3.1. isolation and enrichment of ps degrading microbes after 3 weeks, the tenebrio molitor and zophobas morio were dissected and then the guts of the worms were prepared for the gut microbes’ suspension. the gut microbes’ suspension was transferred to the basal agar plate incubating for 24 h at 37℃ with different yeast extract concentration. the result showed that all of the agar plates have many colonies on basal agar (fig. 1). fig. 1 gut microbes plated on the basal agar plate (a) microbes from the gut of tenebrio molitor incubate with yeast extract (0.05g/l). (b) microbes from the gut of tenebrio molitor incubate with yeast extract (0.1g/l). (c) microbes from the gut of zophobas morio incubate with yeast extract (0.05g/l). (d) microbes from the gut of zophobas morio incubate with yeast extract (0.1g/l) advances in technology innovation, vol. 2, no. 1, 2017, pp. 13 17 15 copyright © taeti all of the colonies on the basal agar plate were rinsed with saline and collected. the suspension was plated on the ps plate incubating for 24 h at 37°c with different yeast extract concentration. the result showed that all of the agars have white colonies (fig. 2). the pure colonies were isolated by streaking plate method until the pure colonies were obtained (fig. 3). fig. 2 gut microbes plated on the ps agar plate (a) microbes from the gut of tenebrio molitor incubate with yeast extract (0.1g/l). (b) microbes from the gut of tenebrio molitor incubate with yeast extract (0.05g/l). (c) microbes from the gut of zophobas morio incubate with yeast extract (0.1g/l). (d) microbes from the gut of zophobas morio incubate with yeast extract (0.05g/l) fig. 3 pure colonies were isolated by streak plate method (a) pure colonies of tm1 incubate at anaerobic condition. (b) pure colonies of tm1 incubate at aerobic condition. (c) pure colonies of zm1 incubate at anaerobic condition. (d) pure colonies of zm1 incubate at aerobic condition to confirm that our ps plate could grow only the ps degrading microbes, we cultured different bacteria on ps plate and found that only the microbes which have the ability of ps degradation could grow on the ps plate (fig. 4). the final result showed that only the tm1 and zm1 can grow on the ps plate. fig. 4 the ps plate culture with different bacteria: zm1, tm1, s. aureus, dh5  (e. coli), lactobacillus 3.2. turbidity assay for the ps-degrading ability of the tm1 and zm1 the colonies from two strains tm1 and zm1 were added to the ps emulsion to confirm their ability of ps degrading. (fig. 5). the result showed that the degrading ability of tm1 and zm1 with yeast extract had more degrading activities than which that without yeast extract. fig. 5 turbidity assay (a) turbidity assay of tm1, zm1 at aerobic condition with or without yeast extract (0.5g/l) (b) turbidity assay of tm1, zm1 at anaerobic condition with or without yeast extract (0.5g/l). advances in technology innovation, vol. 2, no. 1, 2017, pp. 13 17 16 copyright © taeti 3.3. bacteria strain characterization and 16s rdna sequencing by gram staining, the tm1 and zm1 are showed both gram negative bacteria. and in the result of tsi agar test, we can confirm that the tm1 and zm1 are not the same bacteria strain (table 1). from tsi agar result, the tm1 could be alcaligenes sp., pseudomonas sp., or acinetobacter sp., and zm1 could be klebsiella pneumoniae. howerver, from 16s rdna data, the tm1 could be aeromonas sp., and zm1 could be klebsiella pneumoniae (fig. 6). this final confirmation of tm1 strain needs more accurate identification of other method. table 1 the results of tsi agar test tm-1 zm-1 slant alkaline acid butt alkaline acid h2s no no gas production no yes possible species alcaligenes pseudomonas acinetobacter klebsiella (a) description max score total score query cover e value ident accession aeromonas taiwanensis strain a2-50 16s ribosomal rna gene, partial sequence 928 928 92% 0.0 84% nr 116585.1 aeromonas sanarellii strain a2-67 16s ribosomal rna gene, partial sequence 922 922 92% 0.0 84% nr 116584.1 aeromonas dhakensis strain p21 16s ribosomal rna gene, complete sequence 922 922 92% 0.0 84% nr 042155.1 aeromonas hydrophila strain dsm 30187 16s ribosomal rna gene, complete sequence 922 922 92% 0.0 84% nr 119190.1 (b) description max score total score query cover e value ident accession klebsiella pneumoniae strain dsm 30104 16s ribosomal rna gene, partial sequence 1857 1857 99% 0.0 99% nr 117683.1 fig. 6 the blast results of 16s rdna sequencing.(a) the blast result of 16s rdna sequencing of tm1(b) the blast result of 16s rdna sequencing of zm1 4. conclusions in this paper, we setup the ps degrading screening platform, which successfully select the tm1, zm1 from worms’ gut. the characteristic of tm1 and zm1 have been identified through gram staining, tsi agar test, 16s rdna sequencing. this will be the first reported strain from the mealworm and superworm. in the turbidity assay, we can first time show that yeast extract will be a very important co-factor for the tm1 and zm1 with more efficient ps degrading ability. the further study of ps degraded enzyme and cloning from tm1 and zm1 warrants our attention that this platform will be an excellent tools to explore and solve this problem. acknowledgement the support of the minister of science and technology, roc (taiwan), under college student research scholarship 104-2815-c214-010-b is gratefully acknowledged. references [1] k. desmet, m. schelfaut, and p. sandra, “determination of bromophenols as dioxin precursors in combustion gases of fire retarded extruded polystyrene by sorptive sampling-capillary gas chromatography–mass spectrometry,” journal of chromato-graphy a, vol. 1071, pp. 125-129, april 4, 2005. [2] h. nishizaki, k. yoshida, and j. wang, “comparative study of various methods for thermogravimetric analysis of polystyrene degradation,” journal of applied polymer science, vol. 25, pp. 2869-2877, 1980. [3] d. kaplan, r. hartenstein, and j. sutter, “biodegradation of polystyrene, poly (metnyl methacrylate), and phenol formaldehyde,” applied and environmental microbiology, vol. 38, pp. 551-553, 1979. [4] y. yang, j. yang, w. m. wu, j. zhao, y. song, l. gao, r. yang, and l. jiang, “biodegradation and mineralization of polystyrene by plastic-eating mealworms: part 1. chemical and physical characterization and isotopic advances in technology innovation, vol. 2, no. 1, 2017, pp. 13 17 17 copyright © taeti tests,” environmental science & technology, vol. 49, pp. 12080-12086, 2015. [5] j. yang, y. yang, w. m. wu, j. zhao, y. song, l. gao, r. yang, and l. jang, “biodegradation and mineralization of polystyrene by plastic-eating mealworms. 2. role of gut microorganisms (supporting information),” environmental science and technology, vol. 49, no. 20, october, 2015. [6] t. k. chua, m. tseng, and m. k. yang, “degradation of poly (ε-caprolactone) by thermophilic streptomyces thermoviolaceus subsp. thermoviolaceus 76t-2,” amb express, vol. 3, p. 8, 2013. [7] b. van den bogert, w. m. de vos, e. g. zoetendal, and m. kleerebezem, “microarray analysis and barcoded pyrosequencing provide consistent microbial profiles depending on the source of human intestinal samples,” applied and environmental microbiology, vol. 77, pp. 2071-2080, 2011.  advances in technology innovation, vol. 4, no. 1, 2019, pp. 37 43 application of rotating arms type permanent magnet motor chih-chiang hong 1,* , deng-maw tsai 2 department of mechanical engineering, hsiuping university of science and technology, taichung, taiwan, roc received 17 april 2018; received in revised form 15 june 2018; accepted 14 july 2018 abstract the present application related to a rotating arm type permanent magnet motor, in particular to a permanent magnet motor that was actuated by the action of the magnetic forces of a permanent magnet stator of a stator module and two thin ring permanent magnets of a rotor module. the purpose of present permanent magnet motor was to provide the torque produced by a magnetic force of the magnetic energy of a stator to drive the rotation of the rotor. the rotating arm type permanent magnet motor comprised an external support frame base, a stator module and a rotor module. the external support frame base included an upper frame, a lower frame and two side frames. the stator module included an elastic metal plate cantilever arm and a permanent magnet stator. the rotor module included a rotating shaft, a double arc shaped rotating arms type support frame, and two thin ring permanent magnets. magnet material n45h (sintered nd–fe–b) was used for the permanent magnets of stator and rotor. the values of magnetic forces (attraction force and exclusion force) were inverse proportional to the position distances of poles (n pole and s pole) above the top level of thin ring permanent magnets, e.g. c= 12mm. the greater value of magnetic forces got the faster rotation speed. the values of magnetic forces could produce the good enough torque for the thin ring permanent magnets to rotate the shaft when position b= 8mm better than that when position b= 17mm. in the future, with the help of using an external swing motion of electrical controlled swing-type device to produce complete rotation, the present permanent magnet motor might save electricity and power sources. some preliminary rotation speed data of rotating shaft in a permanent magnet motor were obtained and presented with manually controlled. keywords: rotating arms type, permanent magnet motor, stator, rotor 1. introduction some kinds of power sources including electric power, water power, wind energy and solar power are used and applied in many fields. the electric power is generated in power plant that usually operated and supplied by fossil fuel. electricity is used as a power in the electromagnetic motor to transfer electrical energy into mechanical energy and drive the movement of mechanism into the forms of rotation, vibration and linear motions. in 2018, woodford [1] presented the movement rule, rotational work and motor type for the electromagnetic motor. in 2018, liu et al. [2] presented the motor's efficiency and working capacity of electromagnetic-fluid-thermal simulations for the permanent magnet linear motor (pmlm). water power is used to generate electricity in water storage reservoir, water usually used and operated through the dam were built in the valley and water flows all year round in the rivers. wind energy and solar power are usually used in the fields of renewable energy sources (res) for better friendly life of the natural environment. in 2016, boie et al. [3] presented an integration analysis for the developments of res and electricity market in north african (na). also, improving electric motor efficiency * corresponding author. e-mail address: cchong@mail.hust.edu.tw tel.: +886-919037599; fax: +886-4-24961187 advances in technology innovation, vol. 4, no. 1, 2019, pp. 37 43 38 is more interested in a permanent magnet motor to save energy and reduce energy consumption. in 2015, sun and zhang [4] introduced a novel bi-directional energy conversion system to provide high efficiency of voltage on permanent magnet direct current (dc) motor by using a super-capacitor rather than a battery. there are some control applications in the type of permanent magnet synchronous motor (pmsm), permanent magnet brushless direct current (pmbldc) motor, linear motor, stepper motor and alternating current motor. in 2013, ramírez-leyva et al. [5] used a power electronics texas instruments experimental system to validate the passivity strategy of a pmsm speed control. higher performance of the controller can be reached by using the pmsm. in 2012, demirtas and karaoglan [6] used the response surface methodology (rsm) to obtain the proportional integral controller coefficients in a pmbldc motor. some optimal parameter values are found by using the rsm tuning. in 2009, hassanpour isfahani and vaez-zadeh [7] presented the high efficiency, high power factor and high power density of line start permanent magnet electric synchronous motors to reduce the energy consumption. for the fans, compressors and pump applications which move a fluid, the load torque is proportional to the square of the rotational speed. in 2008, ganssle et al. [8] described the rotor with alternating north and south poles in permanent magnet stepper motor. as the coils are energized by electric, the rotor is pulled around. in 2007, bolund et al. [9] gave an overview of flywheel technolo gy and application, and studied the flywheel rotor stored the kinetic energy in high voltage motor/generators for renewable energy generation. in 2006, matsuura [10] reviewed the nd–fe–b (neodymium-iron-boron) sintered magnets for the applications of industrial motor, automobile and electric appliances. some improvements of permanent magnet properties still undergoing invented. in 2004, compter [11] presented the six degrees of freedom of movement control for the electro -dynamic planar motor with moving coils to generate two directional movements. in 2002, coey [12] reviewed the permanent magnet applications of the magnetic force effects in three types of magnetic fields, for actuators, motors, generators and sensors i n uniform magnetic field, for controlling beams, bearings and mineral separations in non-uniform magnetic field, and for magnetometers, switchable clamps and metal separation in time-varying magnetic field. in 2013, chowdhury [13] introduced the kinetics and multidisciplinary enterprise future in molecular motor. it would be a challenge intended to drive a motor without using external electric power or gas power for the better living environment in the future. the author presented and published the us patent for the manuscripts. in 2016, hong [14] presented the rotating arm type permanent magnet motor in the patent us 2016/0094095 a1. the author also presented some papers about the investigations of permanent magnet and magnetostrictive materials. in 2014, hong [15] completed the simple design and application of a bending type segment permanent magnet actuator with n45h (sintered nd–fe–b) material. in 2013, hong [16] used the terfenol-d material to design and construct the application of a magnetostrictive actuator. in 2013, hong [17] studied the transient response of magnetostrictive functionally graded material (fgm) square plates under rapid heating with the generalized differential quadrature (gdq) method. in 2012, hong [18] used the gdq method to investigate the rapid heating induced vibration of magnetostrictive fgm plates. it is interesting to investigate and develop a new rotating arm type permanent magnet motor that comprising an external support frame base, a stator module and a rotor module. the purpose of this paper is to investigate the possible rotational motion of permanent magnet motor. in the future, with the help of using an external swing motion of electrical controlled swing-type device to produce complete rotation, the present permanent magnet motor might save electricity and power sources. 2. design and construction new design and application of permanent magnet motors were developed and constructed as shown in fig. 1. the rotating arm type permanent magnet motor comprising an external support frame base, a stator module and a rotor module. the external support frame base included an upper frame, a lower frame and two side frames. each of the upper and lower frames had a through hole formed at a position near a center position, and the bearing was installed in the center position. the rectangular block made in aluminum material with outer dimensions 220 mm  100 mm  20 mm was used as the frame advances in technology innovation, vol. 4, no. 1, 2019, pp. 37 43 39 base, and used the copper wire cutoff manufacturing machine to cut off the rectangular inner block with dimensions 180 mm  80 mm  20 mm. drilled the through holes 𝜙8 mm in the central position for rolling bearings and rotating shaft to be installed into the positions. the rolling bearings with outer diameter 𝜙8 mm, inner diameter 𝜙5 mm and length 2mm were used to fix rotated shaft. the rotating shaft made in aluminum material with diameter 𝜙5 mm, length 150mm was used to perform the rotational motion. the rotor module included a rotating shaft, a double arc shaped rotating arms type support frame, and two thin ring permanent magnets, wherein a left and right bracket was installed onto the left and right side of the double arc shaped rotating arms type support frame, respectively, and a center hole was formed at the center of the double arc shaped rotating arms type support frame, and the thin ring permanent magnet was an n45h permanent magnet. in each side of the double arc shaped rotating arms type support frame with outer diameter 𝜙76 mm, inner diameter 𝜙66 mm, arc angle 240°, made in aluminum material was shown in fig. 2. two thin (2mm thickness) ring permanent magnets were fixed onto each side surface of the double arc shaped rotating arms type support frame, respectively. the stator module included an elastic metal plate cantilever arm and a permanent magnet stator, wherein the elastic metal plate cantilever arm was a sus304 stainless steel plate, and the permanent magnet stator was also an n45h permanent magnet. permanent magnet n45h also called “neo” or “nd-fe-b” magnets were used for the magnet material. the thin ring permanent magnet n45h with outer diameter 𝜙76 mm, inner diameter 72mm, height 10mm as shown in fig. 3 was provided by mag city co., ltd. company located in taiwan. the permanent magnet stator with 11mm arc length, 10mm height and 2 mm thickness thin permanent magnet was shown in fig. 4 and placed in the proper position away from the top of the rotating shaft. currently to interpret the matter of designs, the permanent magnet stator was placed simply at the tip of rectangular 63 mm  12 mm  1 mm stainless steel sus304 strip that was fixed simply together with the cantilever two-layer strip bounded by rectangular 320 mm  23 mm  1 mm stainless steel sus304 was shown in fig. 5. fig. 1 simple construction of rotating arms type permanent magnet motor fig. 2 double arc shaped rotating arms type support frame fig. 3 thin ring permanent magnet n45h advances in technology innovation, vol. 4, no. 1, 2019, pp. 37 43 40 fig. 4 top view of two thin ring permanent magnets and permanent magnet stator fig. 5 front view of two thin ring permanent magnets and permanent magnet stator declaration of the test methods was stated as follows; some of international standard test methods were usually used such as astm and iso standards that define the precision way by liu et al. [19] in 2011. in this paper did not use any international standard tests for the convenient, only personal investigation view of doe method was used to find the preliminary and fundamental results. locations of permanent magnets in rotor and stator modules were stated as follows, generally, the permanent magnet material n45h was very brittle and working in the general temperature 25°c of environment. for the installation of rotor module, two thin ring permanent magnets were simply fixed at left and right surfaces respectively onto the double arc shaped rotating arms type support frame with the locations that were defined in the fig. 5 (wherein a was the distance between an edge of the left thin ring permanent magnet and rotating shaft center, b was the distance between an edge of the right thin ring permanent magnet and rotating shaft center, and c was the distance between the permanent magnet stator and the upper surface of the double arc shaped rotating arms type support frame). the local cartesian coordinate z-axis was in the direction of rotating shaft, x -axis and y-axis were perpendicular to the rotating shaft. usually, thin ring permanent magnet 1 of rotor module was horizontally located at a position from its right side of ring to the left side of x -axis. thin ring permanent magnet 2 of rotor module was horizontally located at b position from its left side of ring to the right side of x-axis. the center of each one thin ring permanent magnets was located in the y-axis. the permanent magnet stator was placed vertically above the rotating shaft and located at c position to the top surface of the double arc shaped rotating arms type support frame. the magnetic force direction of thin ring permanent magnets of rotor module was in the z -axis, e.g. the upper half of thin ring permanent magnet was s pole, thus the lower half of thin ring permanent magnet was n pole. the placements of two thin ring permanent magnets were installed with an opposite polarity with respect to each other, e.g. the s pole of thin ring permanent magnet 1 was directed upward, thus the n pole of thin ring permanent magnet 2 was directed upward. the magnetic force direction of permanent magnet stator was also in the z -axis, e.g. the upper half of permanent magnet stator was n pole, thus the lower half of permanent magnet stator was s pole. advances in technology innovation, vol. 4, no. 1, 2019, pp. 37 43 41 3. results and discussion permanent magnet rotating test was stated as follows, when the permanent magnet stator properly approached to above of the rotating shaft and located at c position, the exclusion force came from the s pole of permanent magnet stator and the s pole of thin ring permanent magnet 1, in the same time in the opposite side of the double arc shaped rotating arms type support frame, the attraction force came from the s pole of permanent magnet stator and the n pole of thin ring permanent magnet 2, the torque produced from the exclusion force, attraction force and ring positions a, b with respect to z -axis, thus the shaft began to rotate. the magnetic forces came from the thickness sides of permanent magnet stator (2mm) could be neglected by properly using the storage energy effect of fly wheel devices for thin ring permanent magnets. after half rotation of shaft, the attraction force came from the s pole of permanent magnet stator and the n pole of thin ring permanent magnet 2, in the same time in the opposite side of the double arc shaped rotating arms type support frame, the exclusion force came from the s pole of permanent magnet stator and the s pole of thin ring permanent magnet 1, the torque produced again from the attraction force, exclusion force and thin ring permanent magnets positions a, b with respect to z -axis, thus the shaft continued to rotate and completed a whole rotation. the two thin ring permanent magnets rotated as a flywheel to store its energy. with the help of flywheel energy storage, magnetic torque came from the magnetic energy and flexible function came from the stainless steel sus304 strips, the shaft continued easily and possibly to complete the next rotation. fig. 6 simply geometry of nd-fe-b magnets with meshes in the quickfield software fig. 7 torque and magnet field of nd-fe-b magnets in the quickfield software it would be better to have simulation data to validate the experimental results, the simulation of magnetic forces for the permanent magnet would be found in the commercial programs, e.g. quickfield, comsol, maya and ansys. it was available to use the quickfield student edition 6.3 sp1 software for the magnetostatics study to simulate the torque in the dominated-simply geometry of nd-fe-b magnets with meshes as shown in fig. 6. the torque and magnet field of nd-fe-b magnets in the quickfield software were shown in fig. 7, the mechanical torque t= -0.0031619 nm was found for the nd-fe-b magnets with coercive force hc= 653000a/m. the permanent magnet motor rotation test successively rotated with manually controlled the permanent magnet stator to provide the forward and backward of swings and had the following preliminary data. to investigate the effect of positions of permanent magnet stator on the rotation speed of rotating shaft, considering the positions a= 8mm, b= 13mm, c= 12-50mm, the variation of rotating shaft rotation speed (rpm) vs. c (mm) was shown in fig. 8 with manually controlled. the rotation speed values of rotating shaft were increasing with c values decreasing. because the values of magnetic forces (attraction force and exclusion force) were inverse proportional to the position distances of poles (n pole and s pole) above the top level of thin ring advances in technology innovation, vol. 4, no. 1, 2019, pp. 37 43 42 permanent magnets, e.g. c= 12mm. the greater value of magnetic forces got the faster rotation speed. the rotation speed values of rotating shaft would be zero when c= 0mm, because the values of magnetic forces could not produce enough torque for the thin ring permanent magnets to rotate the shaft. when the permanent magnet stator was in the position c=12mm, the rotation speed value of rotating shaft was 24 rpm. to investigate the effect of positions of two thin ring permanent magnets on the rotation speed of rotating shaft, considering the positions a= 8mm, b= 8-17mm, c= 12mm, the variation of rotating shaft rotation speed (rpm) vs. b(mm) was shown in fig. 9 with manually controlled. the rotation speed values of rotating shaft were also increased from 10 rpm to 36 rpm with b values decreasing from 17mm to 8 mm. when the thin ring permanent magnet 2 was in the position b=8mm, the rotation speed value of rotating shaft was 36 rpm. the positions in equal distance of two thin ring permanent magnets a=b=8mm with respect to z-axis were in good position to provide the rotating shaft more easily to rotate. because the values of magnetic forces could produce the good enough torque for the thin ring permanent magnets to rotate the shaft when position b= 8mm better than that when position b= 17mm. fig. 8 rotating shaft rotation speed (rpm) vs. c (mm) with manually controlled fig. 9 rotating shaft rotation speed (rpm) vs. b (mm) with manually controlled in conventional, a motor was a mechanical device that converts energy (created by air, electricity or liquid) into motion and torque. the simple rotating arm type permanent magnet motor was a new, different design of the conventional motor. the present permanent magnet motor provided the torque produced by a magnetic force of the magnetic energy of a stator to drive the rotation of the rotor. this rotating arm type design and construction of the permanent magnet motor would be used as a basic method to form some usual rotating arm type permanent magnet engine in the future. the rotating arm type permanent magnet motor might be developed to produce a suitable shaft rotation speed and also had the load torque capability at the forthcoming time. 4. conclusions a new successful and simple rotating arms type design and application is introduced for the permanent magnet motor that is constructed by the action of the magnetic forces of a permanent magnet stator of a stator module and two thin ring permanent magnets of a rotor module. the permanent magnet motor rotation test successively rotated with manually controlled the permanent magnet stator to provide the forward and backward of swings and had the preliminary data. when the two thin ring permanent magnets locate in the equal distance with respect to rotate shaft axis, it is in good position to provide the rotating shaft more easily to rotate. the importance of this application might be contributed to the pollution reduction of mechanical part innovation designs in the green energy. the rotating arm type permanent magnet motor can produce complete rotation with a suitable motion controller in the future. for example, electrical controlled swing-type device would be applied as the kinetic energy input system to produce completely rotation with its own magnetic energy for the permanent magnet driving 0 6 12 18 24 10 15 20 25 30 35 40 45 50 r o ta ti o n s p e e d ( r p m ) (mm) 0 6 12 18 24 30 36 42 8 10 12 14 16 18 r o ta ti o n s p e e d ( r p m ) (mm) advances in technology innovation, vol. 4, no. 1, 2019, pp. 37 43 43 motor. also the swing-type device would be used and applied to the input force forms of natural wave produced from ocean, lake and river. the electrical controlled swing-type device would be used to provide the forward and backward of swings for the permanent magnet stator. the main contributions of this paper might be presented an approach method of advanced designs in the energy saving for permanent magnet motor. acknowledgement the completion of this application was made possible by a grant nsc 102-2221-e-164-003 from ministry of science and technology (most), taiwan, roc. references [1] lectric motors,” https://www.explainthatstuff.com/electricmotors.html, april 4, 2018. [2] x. liu, h. yu, z. shi, t. xia and m. hu, “electromagnetic-fluid-thermal field calculation and analysis of a permanent magnet linear motor, ” applied thermal engineering, vol. 129, pp. 802-811, 2018. [3] i. boie, c. kost, s. bohn, m. agsten, p. bretschneider, o. snigovyi, m. pudlik, m. ragwitz, t. schlegl and d. westermann, “opportunities and challenges of high renewable energy deployment and electricity exchange for north africa and europe scenarios for power sector and transmission infrastructure in 2030 and 2050,” renewable energy, vol. 87, pp. 130–144, 2016. [4] l. sun and n. zhang, “design, implementation and characterization of a novel bi-directional energy conversion system on dc motor drive using super-capacitors,” applied energy, vol. 153, pp. 101–111, 2015. [5] f. h. ramírez-leyva, e. peralta-sánchez, j. j. vásquez-sanjuan and f. trujillo-romero, “passivity-based speed control for permanent magnet motors,” procedia technology, vol. 7, pp. 215–222, 2013. [6] m. demirtas and a. d. karaoglan, “optimization of pi parameters for dsp-based permanent magnet brushless motor drive using response surface methodology,” energy conversion and management, vol. 56, pp. 104–111, 2012. [7] a. hassanpour isfahani and s. vaez-zadeh, “line start permanent magnet synchronous motors: challenges and opportunities,” energy, vol. 34, pp. 1755–1763, 2009. [8] j. ganssle, s. ball, a. s. berger, k. e. curtis, l. a. r. w. edwards, r. gentile, m. gomez, j. m. holland, d. j. katz, c. keydel, j. labrosse, o. meding, r. oshana and p. wilson, embedded systems: world class designs. uk: elsevier, 2008. [9] b. bolund, h. bernhoff and m. leijon, “flywheel energy and power storage systems,” renewable and sustainable energy reviews, vol. 11, pp. 235–258, 2007. [10] y. matsuura, “recent development of nd–fe–b sintered magnets and their applications,” journal of magnetism and magnetic materials, vol. 303, pp. 344–347, 2006. [11] i. j. c. compter, “electro-dynamic planar motor,” precision engineering, vol.28, no. 2, pp. 171–180, 2004. [12] j. m. d. coey, “permanent magnet applications,” journal of magnetism and magnetic materials, vol. 248, pp. 441–456, 2002. [13] d. chowdhury, “stochastic mechano-chemical kinetics of molecular motors: a multidisciplinary enterprise from a physicist’s perspective,” physics reports, vol. 529, pp. 1–197, 2013. [14] c. c. hong, rotating arms type permanent magnet motor. united states patent application publication hong, us 20160094095a1, mar. 2016. [15] c. c. hong, “application of a bending type segment permanent magnet actuator,” materials science: an indian journal, vol. 11, no. 9, pp. 311–315, 2014. [16] c. c. hong, “application of a magnetostrictive actuator,” materials & design, vol. 46, pp. 617–621, 2013. [17] c. c. hong, “transient response of magnetostrictive functionally graded material square plates under rapid heating,” journal of mechanics, vol. 29, no. 1, pp. 135–142, 2013. [18] c. c. hong, “rapid heating induced vibration of magnetostrictive functionally graded material plates,” asme, journal of vibration and acoustics, vol. 134, 021019, pp. 1–11, 2012. [19] p. liu, r.l. browning, h.j. sue, j. li and s. jones, “quantitative scratch visibility assessment of polymers based on erichsen and astm/iso scratch testing methodologies,” polymer test, vol. 30, pp. 633–640, 2011. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation , vol. 1, no. 2, 2016, pp. 38 40 38 copyright © taeti predictive adaptive control of an activated sludge wastewater treatment process ioana nascu 1,2 , ioan nascu 3,* , grigore vlad 4 1 department of che mica l engineering, centre for process systems engineering (cpse), impe ria l college london, u.k. 2 artie mcferrin department of chemical engineering, texas a&m, college station tx, usa . 3 department of automation, technical university of cluj-napoca, romania. 4 icpe bistrita, romania. received 22 february 2016; received in revised form 28 march 2016; accept ed 30 march 2016 abstract this paper presents an application regarding a model based predictive adaptive controlle r used to improve the effluent quality of a conventional activated sludge wastewater treat ment process. the adaptive control scheme consists of two modules: a robust parameter estimator and a predictive controlle r. the controlle r design is based on the process model obtained by recursive estimation. the performances of the adaptive control algorithm are investigated and compared to the non-adaptive one. both the set point tracking and the regulatory performances have been tested. the results show that this control strategy will help overcome the challenge for ma intaining the discharged water quality to meet the regulations. keywor ds : wastewater treatment, predictive control, adaptive control, parameter estimation 1. introduction wastewater treatment plants (wwtps) are key infrastructures for ensuring a proper protection of our environment. the biologica l treatment is an important part of any wwtp and the activated sludge process is the most common biotreatment process used to treat sewage and industrial wastewaters. conventional activated sludge systems are focused on the removal of carbonaceous organic matter. these biologica l processes are nonlinear and comple x, representing a challenge fro m the control p oint of view due to the enhanced environmental regulations related to the effluent quality and the large variat ions in the influent flow rates and concentrations [1]. an overview of the activated sludge wastewater treat ment process mathe matica l modeling is presented in [2]. a variety of control strategies for wwtp were proposed in the available literature : conventional pid control, fuzzy control, predict ive and optima l control [3,4], all of them presenting good performances in a certa in operating point. the control strategy proposed in this paper takes into consideration the controller adaptation to the process parameter changes caused by high variations in the influent flow rate or concentration. the purpose of this paper is to investigate the performance of an adaptive control algorith m (a gpc) based on the generalized pred ictive control (gpc) method [5]. the performances of the agpc algorithm a re investigated on an activated s ludge wastewater treatment process . therefore, the activated sludge process was first modeled and the model was calibrated and validated based on a comb ination of laboratory tests and plant operating measured data, available from romanofir wwtp. 2. process description and modeling the model of the process was developed based on data obtained from an operational wwtp. the biological treat ment process of this wwtp is a conventional activated sludge system with two components: a bioreactor operating under aerobic conditions and a settler (fig. 1). a widely used model to describe the dynamics of bio logical treat ment processes is * corresponding aut hor, email: ioan.nascu@aut.utcluj.ro advances in technology innovation , vol. 1, no. 2, 2016, pp. 38 40 39 copyright © taeti the activated sludge model nr.1 (asm1) [6]. since this model is comp le x, containing a large number of state variables and parameters, it is necessary to simplify it into a simp ler model, more suited for control purposes. considering only two materia l co mponents , a soluble substrate and a particulate bio mass component , the mathemat ical mode l of the biologica l treatment process for the re moval of organic matter will be composed of a set of four non-linear differentia l equations, three equations for the aerated bioreactor and one for the settler. the process model and the sensitivity analys is are described in greater detail in [7]. fig. 1 biological treatment process schematic 3. adaptive control algorithm the adaptive control scheme consists on two modules: a robust parameter estimator and a model based predictive controller. the controller design is based on certainty equivalence princip le, the estimates of the process model para meters are used instead of the unknown true values in the control law design. the controller design is based on the process model obtained by recursive estimation. 3.1. controller design the gpc design procedure is we ll described in the literature [5]. the key idea is to min imize the following cost function:       un j n nj r jtuj jtyjtyj 1 2 2 )]1()[([ )]()([ 2 1  (1) where: δ is the differenc ing operator 1-q -1 , yr is the future re ference sequence and n1 the minimu m costing horizon, n2 the ma ximu m costing horizon, nu the control horizon and ρ(j) the control-weighting sequence are the controller design parameters. 3.2. parameter estimation to estimate the para meters of the plant model, a version of the standard recursive least square algorith m (rlsa) was used. the use of the basic identification scheme may lead to unstable adaptive process control. for practica l imple mentations of adaptive control strategies, this scheme must be modified in order to provide a robust parameter estimator. the para meter estimation algorithm includes data normalization, a dead zone, a forgetting factor and data prefiltering [8]. 4. simulation results the simu lations were carried out in the matlab environ ment and the closed loop controllers performances for the concentration of organic matter (ss), considered as controlled process output, were investigated. the aeration flow (w) was considered the manipulated input. the nonlinear model o f the b iologica l treat ment process was used to simu late the process dynamics. to obtain the transfer function fro m w to ss the model was linearized around the nomina l operating point (the influent organic matter concentration ssin=765mg/ l). the second operating point was considered for a lowe r influent organic matter load, ssin=300 mg/l. performances for effluent organic matter concentration ssef which is more relevant we re determined and plotted. fig. 2 setpoint tracking performances. agpc-continuous line, gpc-dotted line in fig. 2 setpoint tracking performances of the adaptive predictive controller (a gpc) and predictive controller (gpc) for the second operating point can be compared. performances are appreciably improved in the agpc case (continuous line) because the estimator will correct the model para meter values and will adapt the controller to this situation. 45 50 55 60 65 70 75 80 120 121 122 123 124 125 t [ h ] s s e f [m g /l ] advances in technology innovation , vol. 1, no. 2, 2016, pp. 38 40 40 copyright © taeti fig. 3 presents the regulatory performances during simu lation tests when the disturbances are applied on the influent organic matter concentration ssin. the ss setpoint is fixed such as the regulations for the organic matter concentration in discharged water are met most of the time (ss effluent (cco-cr) <125 mg/ l). also in this case agpc has slightly improved performances. fig. 3 regulatory performances . agpc continuous line, gpc dotted line 5. conclusions in this paper, the performances of an adaptive model based predictive control strategies for the organic matter concentration in the effluent of the b iologica l t reat ment process of a wwtp have been evaluated. simulation studies were based on the non -linear model, obtained from mass balance equations and they have indicated that the proposed control method performs we ll and can be easily used. both the setpoint tracking and the regulatory performances have been investigated. the regulations for the organic matter concentration in discharged water are satisfied in a high percentage. acknowledgement the support of the ro manian nat ional authority for scientific research, uefiscdi, under grant caseau 274/2014, is gratefully acknowledged. references [1] r. katebi, m. a. johnson, and j. wilkie, “control and instrumentation for wastewater treatment plants,” springer, london, 2012. [2] u. jeppsson, “modelling aspects of wastewater treatment processes,” ph.d. thesis, dept. of industrial electrical eng. and automation, lund university, sweden, 1996. [3] m. a. brdys, m. grochowski, t. gminski, k. konarc za k, and m. dre wa, “hiera rchica l predictive control of integrated wastewater treatment systems,” control eng. pract. , vol. 16, no. 6, pp. 751-767, june 2008. [4] b. holenda, e. do mokos, a. rédey, and j. faza kas, “ dissolved oxygen control of the activated sludge wastewater treatment process using model predictive control,” co mputers and chemica l engineering, vol. 32, pp. 1278-1286, 2008. [5] d. w. clarke, c. mohtadi, and p. tuffs, “generalized predictive control part 1. the basic algorithm,” automatica, vol. 23, no. 2, pp. 137-148, 1987. [6] m. henze, w. gu je r, t. m ino, and m. van loosdrecht, “activated sludge models asm1, asm2, asm 2d and asm3,” iwa publishing, london, 2000. [7] s. cristescu, i. naşcu, and i. naşcu, “sensitivity analysis of an activated sludge model for a qastewater treatment plant,” proc. international conference on system theory, control and computing, pp. 595-600, oct. 2015. [8] i. nascu, adaptive control, media mira , cluj-napoca, 2002. 200 250 300 350 400 118 119 120 121 122 t [ h ] s s e f [m g /l ] https://www.google.ro/search?newwindow=1&sa=n&hl=ro&biw=1476&bih=770&tbm=bks&tbm=bks&q=inauthor:%22reza+katebi%22&ved=0ahukewijnzr89rxjahwh_q4khcbbcreq9agiajaj https://www.google.ro/search?newwindow=1&sa=n&hl=ro&biw=1476&bih=770&tbm=bks&tbm=bks&q=inauthor:%22jacqueline+wilkie%22&ved=0ahukewijnzr89rxjahwh_q4khcbbcreq9agibdaj  advances in technology innovation, vol. 2, no. 4, 2017, pp. 126 129 126 high-temperature corrosion of t22 steel in n2/h2s-mixed gas min jung kim, dong bok lee * school of advanced materials science and engineering, sungkyunkwan university, suwon 16419, south korea. received 06 oct ober 2016; received in revised form 29 december 2016; accept ed 16 january 2017 abstract astm t22 steel (fe-2.25cr-1mo in wt.%) was corroded at 600 and 700 o c for 5-70 h under an atmospheric pressure that consisted of n2-(0.5. 2.5)%h2s-mixed gas. t22 steel corroded rapidly, forming outer fes scales and inner (fes, fecr2o4)-mixed scale. the formation of the outer fes scale facilitated the oxidation of cr to fecr2o4 in the inner scales. since the nonprotective fes scale was present over the whole scale, t22 steel displayed poor corrosion resistance. ke ywor ds : fe-cr-mo alloy, t22 steel, corrosion, h2s gas, sulfidation 1. introduction the integrated gasification combined cycle (igcc) power plants are operating in u.s., japan, germany, and netherlands. it is a new technology that turns coal into synthesis gas (syngas) and produces the electricity. it promises low emissions and improved efficiency compared to conventional coal-fired power plants that produce the electricity directly by burning coals [1]. however, one of the main problems in igcc is the corrosion occurring by the syngas in a gasification unit, because the syngas consisted primarily of the extremely corrosive h2s gas. this limits the operating temperature and the process efficiency of the igcc power plants. it is noted that the h2s gas has been a major concern in oil refinery plants, high-temperature gas turbines, and petrochemical units. h2s gas dissociates into sulfur and hydrogen ions, and reacts with the steel according to the reaction; h2s+fe → fes+h2 [2-4]. generally, most sulphides are highly nonstoichiometric, and ionic diffusion in the scales is hence quite fast [5,6]. sulfidation is therefore a quite serious problem. hydrogen also significantly decreases the corrosion resistance and mechanical properties of the steel [7-10]. in this study, t22 steel was corroded at 600 and 700 oc for up to 70 h in n2-(0.5, 2.5)%h2s-mixed gas in order to understand its corrosion behavior in the h2s-mixed gas. this is important in igcc power plants, oil refinery plants, high-temperature gas turbines, and petrochemical units. although the oxidation behavior of t22 steel was extensively studied [11-12], little is reported about the high-temperature corrosion behavior of t22 steel in h2s-mixed gas. the purpose of this study is to investigate the corrosion behavio r of t22 steel in n2/h2s-mixed gas . 2. method t22 steel plate with a nominal composition of fe-2.25cr-1.0mo-0.45mn-0.3si-0.12c in wt% were cut into a size of 2x10x15 mm 3 , ground up to a 1000-grit finish with sic paper, ultrasonically cleaned in acetone, corroded, and inspected to examine its corrosion behavior. each sample was suspended by a pt wire in a quartz reaction tube positioned vertically inside the hot zone of the vertical electrical furnace, and corroded at 600 and 700 o c for up to 70 h in n2-(0.5, 2.5)%h2s-mixed gas maintained at 1 atm. the employed n2 gas was 99.999% pure, and h2s gas was 99.5% pure. the corroded samples were characterized by a scanning electron microscope (sem; jeol jem-2100f operated at 200 kev), an x-ray diffractometer (xrd) with cu-kα radiation operating at 40 kv and 300 ma in θ/2θ configuration, and an electron probe microanalyzer (epma). 3. results and discussion the corrosion kinetics of t22 steel in n2-(0.5, 2.5)%h2s gas are depicted in fig. 1. weight gains were the sum of weight gain due to scaling and weight loss due to scale spallation. they increased with an increase in the temperature and the h2s concentration. the fastest corrosion rate was observed in the sample corroded at 700 o c in n2-2.5%h2s-mixed gas. the almost linear, large weight gains depicted in fig. 1 indicate vastly fast corrosion kinetics for all the samples. it is noted that local cracking, partial spallation and void formation in the formed scales were unavoidable for all the samples, including at 600 o c in n2-2.5%h2s-mixed gas for 70 h. such scale failure became more serious as corrosion progressed. although t22 steel displayed reasonable oxidation resistance in the oxidizing atmospheres, it was non-protective in the harsh h2s-mixed corrosion environment. fig. 1 weight gains of t22 steel at 600 and 700 o c in n2-(0.5, 2.5)%h2s-mixed gas fig. 2(a) indicates that t22 steel consisted mainly of α-fe. the other minor phase, perlite, was not detected fig. 2(a) due to its small amount. in this study, the corrosion at 600 and 700 o c for 5-70 h in n2-(0.5, 2.5) %h2s-mixed gas inevitably led to the formation of the outer fes scale and the inner (fes, fecr2o4)-mixed s cale. fig. 2(b) indicates the outer fes scale that formed after corrosion at 700 o c for 20 h in n2-2.5%h2s gas. the outer, non-adherent fes scale was detached off by slightly hitting the sample, and the inner scale was x-rayed as shown in fig. 2(c). this revealed the inner (fes, * corresponding aut hor. email: dlee@skku.ac.kr http://en.wikipedia.org/wiki/coal http://en.wikipedia.org/wiki/syngas http://en.wikipedia.org/wiki/syngas http://en.wikipedia.org/wiki/gasification advances in technology innovation, vol. 2, no. 4, 2017, pp. 126 129 127 cop y right © taeti fecr2o4)-mixed scale, along with the α-fe matrix phase. the amount of cr in t22 steel was not large enough to completely cover the matrix surface with the protective cr2o3 scale. t22 steel reacted with the h2s gas to form fes, releasing hydrogen according to the equation; fe(s) +h2s (g) → fes(s) +h2 (g). since fes has a very high concentration of cation vacancies, it grew fast to form the outer scale through the outward diffusion of fe 2+ ions [9]. the formation of fes decreased the sulfur potential, and thereby the oxygen potential underneath, facilitating the formation of the oxides in the inner scale, as shown in fig. 2(c). the minor alloying elements such as mo and mn in t22 steel tended to be expelled from the inner scale, due to their small amount or activity. the distribution of alloying elements depends on the thermodynamic stability of corresponding oxides or sulfides and activity of concerning elements. the impurity oxygen in n2-(0.5, 2.5) %h2s-mixed gas reacted with t22 steel according to eqs. (1) and (2). fe(s)+1/2 o2 (g) → feo(s), (1) 2cr(s)+3/2 o2 (g) → cr2o3(s). (2) thermodynamica lly, the o xides are generally more stable than the corresponding sulfides. the spinel is formed by diffusion of fe 2+ from feo to cr2o3 oxides through spinel accompanied by diffusion of hole and evolution of oxygen gas at feo/spinel interface. the fe 2+ reacts with cr2o3 to produce spinel at the spinel/cr2o3 interface, and cr 3+ d iffuses with hole to the cr2o3/gas interface in order to keep electro-neutrality [13]. the formed fe o and cr2o3 o xides particles, the solid-state reaction occurred to form the more stable fe cr2o4 spinel scale, wh ich gradually dispersed in external o xide layer. fe cr2o4 spinel reacted with feo and cr2o3 according to the equation. feo(s)+ cr2o3(s) → fecr2o4(s) fig. 2 xrd patterns of t22 steel. (a) before corrosion, (b) the outer scale, and (c) the inner scale that formed after corrosion at 700 o c for 20 h in n2-2.5%h2s gas the morphology of surface scales that formed on t22 steel after corrosion in n2-0.5%h2s gas is shown in fig. 3. from the early corrosion stage at 600 o c, fes platelets progressively protruded through the ensuing outward diffusion of fe 2+ ions over the smooth underlying scale (figs. 3(a) and (b)). they spalled off easily due to their fast growth rate and incorporation of hydrogen released from the h2s gas. the formed scales were quite fragile so that cracks were seen in fig. 3(a). at 700 o c, the fes platelets grew to coarse, protruded fes grains, as shown in fig. 3(c). as corrosion progressed, fes grains grew bigger, leading to the generation of cracks in the surface fes scale (fig. 3(d)). the eds analysis indicated that the outer scale and the inner scale consisted primarily of fes (figs. 3(e) and (f)), respectively. fig. 3 sem top view of the scales that formed on t22 steel after corrosion in n2-0.5% h2s gas. (a) at 600 o c fo r 5 h, (b ) at 600 o c fo r 40 h, (c) at 700 o c for 5 h, (d) at 700 o c for 40 h, (e) eds spectrum of spot ①, (f) eds spectrum of ② the morphology of surface scales that formed on t22 steel after corrosion in n2-2.5%h2s gas is shown in fig. 4. from the early corrosion stage at 600 o c, coarse, facetted fes grains covered the whole surface (fig. 4(a)). they continuously grew bigger as the corrosion progressed (figs. 4(b)-(d)). in fig. 4(d), cracks propagated interand trans-granularly. with the increase of concentration of the h2s gas from 0.5 to 2.5 %, the grains at the surface of the scale became much coarser as shown in figs. 3 and 4, indicating that the h2s gas accelerated corrosion. advances in technology innovation, vol. 2, no. 4, 2017, pp. 126 129 128 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti fig. 4 sem top view of the scales that formed on t22 steel after corrosion in n2-2.5% h2s gas. (a) at 600 o c fo r 5 h, (b ) at 600 o c fo r 40 h, (c) at 700 o c for 5 h, (d) at 700 o c for 40 h. fig. 5 shows sem/eds analytica l results of t 22 steel after corrosion in n2-0.5%h2s gas at 600 o c for 70 h. the scale consisted of about 120 μ m-thic k outer scale, and about 70 μ m-thic k inner scale. the outer scale was d etached from the inner scale, and vertica l cracks were seen in the inner scale, owing to the la rge stress arisen by (1) the mis match in the therma l e xpansion coeffic ients among the outer scale, inner scale, and the matrix, (2) the difference in the growth rates of various oxides and sulfides, and (3) the hydrogen dissolution in the scale. the eds analysis indicated that the outer scale and the inner scale consisted prima rily o f fes (fig. 5(b)), and (fes, fecr2o4) (fig. 5(c)), respectively. this was consistent of the xrd results depicted in fig. 2. fes platelets protruded over the outer fes scale (fig. 5(a)). fig. 5 sem/ eds analytica l results of t22 steel after corrosion in n2-0.5% h2s gas at 600 oc fo r 70 h. (a) cross -sectional image, (b) eds spectrum of spot ①, (c) eds spectrum of ②. 4. conclusions t22 steel was corroded at 600 and 700 o c for up to 70 h in n2/h2s-mixed gas under total pressure of 1 atm. the corrosion occurred almost linearly through the sulfidation, together with oxidation to a less extent. the outer scale consisted primarily of fes that formed by the outward diffusion of fe 2+ ions. the inner (fes, fecr2o4)-mixed scale formed by the inward diffusion of predominantly sulfur and a small amount of oxygen. the outer scale kept growing outwards during corrosion. the formed scales were non-adherent, and susceptible to cracking. since the nonprotective fes scale was present over the whole scale, t22 steel displayed poor corrosion resistance in n2/h2s-mixed environments at high temperatu res. hence, pack-cementation, hot dipping, plasma evaporation, plating techniques were developed in order to improve its corrosion resistance. acknowledgement this research was supported by basic science research program through the national research foundation of korea (2016r1a2b1013169) funded by the ministry of education. references [1] r. a. meyers, encyclopedia of physical science and technology, 3rd ed., usa; academic press, 2001. [2] n. birks, g. h. meier, and f. s. pettit, introduction to high temperature oxidation of metals, 2nd ed., usa: cambridge university, 2006. [3] d. young, high temperature oxidation and corrosion of metals, uk: elsevier, 2008. [4] a. m. mebel, and d. y. hwang, “theoretical study of the reaction mechanism of fe atoms with h2o, h2s, o2 and h + ,” the journal of physical chemistry a, vol. 105, pp. 7460-7467, 2001. [5] s. mrowec and k. przybyls ki, “transport properties of sulfide scales and sulfidation of metals and alloys,” oxidation of metals, vol. 23, pp. 107-139, 1985. [6] s. mrowec, t. walec, t. werber, “high-temperature sulfur corrosion of iron-chromium alloys ,” oxidation of metals, vol. 1, pp. 93-120, 1969. [7] s. mrowec and m. wedrychowska, “kinetics and mechanism of high-temperature sulfur corrosion of fe-cr-al alloys ,” oxidation of metals, vol. 13, pp. 481-504, 1979. [8] t. liu, c. wang, h. shen, w. chou, n. y. iwata, and a. kimura, “the effect of cr and al concentration on the oxidation behavior of oxide dispersion strengthened ferritic alloys ,” corrosion science, vol. 76, pp. 310-316, 2013. [9] m. danielewski, s. mrowec, and a. stołosa, “sulfidation of iron at high temperatures and diffusion kinetics in ferrous sulfide,” oxidation of metals, vol. 17, pp. 77-97, 1982. [10] d. b. lee, m. a. abro, p. yadav, s. h. bak, y. shi, and m. j. kim, “corrosion of fe-9%cr-1%mo steel at 600 and 700 o c in n2/(0.5, 0.25)%h2s-mixed gas ,” journal of the korea institute of surface engineering, vol. 47, pp. 147-151, 2016. [11] j. purbolaksono, j. ahmad, a. khinani, a. ali, and a. z. rashid, “failure case studies of sa213-t22 steel tubes of boiler through computer simulations ,” journal http://pubs.acs.org/journal/jpcafh http://www.springerlink.com/content/?author=s.+mrowec http://www.springerlink.com/content/?author=k.+przybylski http://pubs.acs.org/journal/jpcafh http://pubs.acs.org/journal/jpcafh http://www.sciencedirect.com/science/article/pii/s0950423009000916 http://www.sciencedirect.com/science/article/pii/s0950423009000916 advances in technology innovation, vol. 2, no. 4, 2017, pp. 126 129 129 cop y right © taeti of loss prevention in the process industries, vol. 23, pp. 98-105, 2010. [12] b. s. sidhu and s. prakash, “high-temperature oxidation behavior of nicra1y bond coats and stellite-6 plasma-sprayed coatings ,” oxidation of metals, vol. 63, pp. 241-259, 2005. [13] k. nagata, r. nishiwaki, y. nakamura, and t. maruyama, “kinetic mechanisms of the formations of mgcr2o4 and fecr2o4 spinels from their metal oxides metals oxides ,” solid state ionics, vol. 49, pp. 161-166, 1991.  advances in technology innovation , vol. 3, no. 3, 2018, pp. 126 132 international flipped class for chinese honors bachelor students in the frame of multidisciplinary fields: reliability and microelectronics olivier bonnaud 1,2,3,* , yves danto 3,4 , yinghui kuang 5 , li yuan 5 1 department of sensors and microelectronics, ietr, university of rennes 1, rennes, france. 2 gip-cnfm (national coordination for education in microelectronics and nanotechnologies), grenoble, france . 3 department of reliability, ims, university of bordeaux, bordeaux, france . 4 south-east university (seu), nanjing, jiangsu, china. 5 chien-shiung wu college (honors), south -east university (seu), nanjing, jiangsu, china. received 21 july 2017; received in revised form 19 sept ember 2017; accept ed 28 december 2017 abstract this paper reports an innovative pedagogic experience performed at south-east university (seu) with electrical engineering bachelor honors students (computer science, mechanics, and electronics). the purpose was to develop their motivation and to make them aware of the strategic importance of two a spects of electronic engineering i.e. integrated technologies and reliability assessment of devices and systems. the pedagogical approach was based on a flipped class and learning by project that consisted to involve the students in the two topics. after s everal lectures on the fundamentals of microelectronics and reliability of electronics components performed by foreign professors, twelve groups of five students were built. each group had to develop one topic, chosen for its strategic importance. thus, from a given set of main literature references, the students prepared during three days a twelve pages report and an oral presentation, both in english language. results were generally very good. most of the students succeeded in addressing issues that were completely new for them. they clearly built by themselves the skills allowing understanding of all the important aspects of the topics they had to approach. this paper gives details on the organization, the content and the final evaluation . keywords: pedagogical innovation, learning by project, flipped classes, microelectronics, reliability, transdisciplinary approach, international teaching 1. introduction this paper reports an innovative pedagogic experience performed at south -east university (seu) of nanjing (jiangsu, china) with engineering bachelor honors students (computer science, mechanics, bio -medical engineering and electronics). the purpose was to develop their motivation and to make them aware of the strategic importance of two aspects of electro nics engineering i.e. integrated technologies and reliability assessment of devices and systems. indeed, while electronics is driv ing most of human activities sectors, it becomes very important to highlight the major issues of advanced technologies and ver y high reliability levels of devices and systems. in addition, the evolution of the technologies allows an increasingly spreading of the applications. there are today many fields of application of the electronics in the new concepts of smart connected obje cts and internet of things (iot). the main well-known application domains are health, transport, communications, security, energy and environment with the microelectronics at the heart of the systems [1]. this means as well an increasingly multidisciplinar y * corresponding author. e-m ail address: olivier.bonnaud@ univ-rennes1.fr advances in technology innovation , vol. 3, no. 3, 2018, pp. 126 132 copyright © taeti 127 approach in the pedagogical strategy and an innovative behavior of the future graduate and post -graduate [2]. the evolution of the content of the engineering formations must answer progressively to this new trend of the technics. with the same goal, a transverse discipline that manages crucially the use of a product, is the reliability, discipline that becomes a major objectiv e of all the products that are becoming more and more complex. reliability is applied at multiple levels of complexity. the simples t case corresponds to a discrete object, for example an elementary transistor. higher complexity appears in integrated circuits that can contain several billion elementary transistors. the worst situation is related to huge systems, for example data center s that can contain millions of complex integrated circuits, each of which is built of billions of elementary transistors! these two subjects are therefore part of the objectives for students and the challenge is to choose the best pedagogical way to acquir e these skill. the pedagogical approach was based on a flipped class and learning by project that consisted to involve the students in the two topics with a major part of their personal investment. after several lectures on the fundamentals of microelectro nics and reliability of electronics components performed by two foreign professors in english language, twelve groups of five students were built. each group had to develop one topic, chosen for its strategic importance, such as the reliability of the spac e electronics, the data centers, the challenge of very high density packaging, but also the huge development of connecting objects, the ulsi and the large area electronics. thus, from a given set of main literature references, the students have to prepare during five days a twelve pages report and an oral presentation, both in english language. after a short presentation of the context, the paper deals with the organization of these international flipped classes, the content of the personal works in the both specialties, an analysis of the main results, and a description of the evaluation of the questionnaire filled by the students . 2. honors bachelor the honors bachelor students are highly selected students that are supposed to produce an efficient work and t o continue in post graduate studies. they are rigorously selected at the entrance and are enrolled in an institute, in this case, the chien-shiung wu college. all the students are selected on the basis of a good capability to work in english language. they are thus expected to attend to lectures and seminars performed by foreign teachers. the students are also able to write documents in english, even if with scientific purpose they have to learn new vocabulary adapted to the domain. there are several promot ions spread in four main disciplines: computer science, electrical engineering, mechanics and bio -medical engineering. the managers of the bachelor degree have the possibility to organize specific modules, with the goal to make aware the students in high technologies with a new pedagogical approach. thus the concept of flipped classes was adopted. two groups of thirty students were built, the first one having the objective to study the new field of microelectronics, the second the new field of reliab ility. as it is well known, many applications in engineering are increasingly involving smart connected objects based on microelectronics systems [3]. two aspects are thus very important: the technics and technologies that are at the core of the systems [4], and the reliability [5] that is one of the main parameters that characterize good products adapted for their mission profiles. these two fields are today a significant component of the background of engineers. they are also at the heart of th e future challenge of the technological innovation. that is the reason of the choice of these topics for the students at the level of the third year of the bachelor:  microelectronics and nanotechnologies: engineering sciences at the heart of the smart connected objects and of their applications,  the reliability challenge for advanced technologies: a critical concern for research, production, economy and modern society . advances in technology innovation , vol. 3, no. 3, 2018, pp. 126 132 copyright © taeti 128 these two topics are selected by sixty students of the institute. in order to maintain a high motivation of the students, they build by themselves six groups of five students with the objective to work on one of the proposed topics . 3. awareness to microelectronics and to the reliability the evolution of the fields of microelectronics and nanotechnologies has pre sently two main orientations: on the one hand, the ultra large scale integration following the famous moore’s law [6] with finfet or fdsoi technologies [7], and on the othe r hand, the heterogeneous technology approach that inserts the new devices in systems, mainly the smart connected objects based on integrated technologies but also on large area electronics, on thin film technologies, on sensors and actuators that can be linked to the applications. sensors may concern the physical, chemical, spatial and b iological detections, while the actuators can drive mechanical, fluidic, optical, thermal mechanisms that are involved in the monitoring of many security or medical equipment. the evolution of these technologies corresponds to a new law of evolution, named “more than moore” and includes system in package, systems on chip involving the third dimension and even lab -on-chip. fig. 1 shows the classical representation of the moore’s law issue of the ipc report [8]. fig. 1 moore’s law representation: the exponential increasing of the integration was verified during more than fifty years, even if the slope was a little bit smaller than the predicted one by g. moore in 1965! (after ipc report [8]) this evolution is also a consequence o f the fabulous improvement of the design tools, in the frame of the computer aided design approach. these tools are able to design and simulate very complex circuits with billions of elementary devices (mainl y transistors) but also imply the integration of many new functions in heterogeneous technologies. these last are combining electronics functions with either mechanical elements and optical devices, or functionalized surfaces able to detect chemical or biological species. besides, a multidisciplinary ap proach is increasingly needed in order to develop in the frame of research and development team’s new products in many domains of applications such as health, transport, communications, security, energy and environment with the microelectronics in the heart of the systems [1]. this combination of these knowledge and multidisciplinary competences is the engine of the innovation. thus, it is a real challenge to form the young to this new app roach. in parallel, the increasing complexity generates new problemat ic that is linked to the mission profiles of the products, the ageing, the security of the systems that are all gathered in the reliability domain. it is more and more important to evaluat e the meantime to failure of each product (mtbf), and more especially the failure in time parameter (fit) that must be very low even in a reliable complex system. this parameter is strongly function of the environment that can be harsh in many applications. this behavior corresponds to the presence or the variation of the parameters such as, temperature, humidity, electromagnetic advances in technology innovation , vol. 3, no. 3, 2018, pp. 126 132 copyright © taeti 129 radiation, particle bombardment, electrical stress (high voltage or high current), mechanical vibrations, mechanical stress, chemical or aggressive ambiances, etc. fig . 2(1) shows the classical variation of the failure rate in function of the time. the analysis of the reliability consists to understand the three main stages of the evolution, the early infant failure rate, the normal failure and the end of life failure. however, for so complex integrated electronic devices, the question of obtaining very high yield production levels has imposed strategies of total quality management in the devices processing, with a drastic reduction of latent defects and infant failures at the first stage of the devices life time. furthermore, as electronics systems are now driving almost all sectors of human activities, their reliability is now one of the main factors conditioning their life safety. thus , the classical “bath” shaped failure rate curve is being target ed in an almost constant “zero” failure rate, up to the end of life time, as shown fig. 2(2). the solutions to reach such results are difficult to derive and need to develop particular skills for engineers . fig. 2 failure analysis: 1) classical bathtub failure rate evolution (blue); 2) almost constant zero failure rate. the second curve is the targeted behaviour of the innovative circuits and systems (red) these two topics are selected by sixty students of the institute. in order to maintain a high motivation of the student, they build by themselves six groups of five students with the objective to work on one proposed topic . 4. organization of the flipped classes the modules are organized on two weeks. the students are shared into two classes of thirty students, divided into six groups of five students each. the face-to-face duration between students and teachers is about twelve hours. the personal work of each student is approximately five times higher. it includes bibliography research, analysis of the given documents, compilation of the research by the group, repartition of the work, organization of a report, redaction of the manuscript, preparation of the oral presentation involving classical tools such as powerpoint diaporama. more or less, the presentation is half an hour that means a very precise duration and a clear sharing of the intervention by the five speakers . the students did not yet have solid experience in both areas. as a result, the beginning of both modules was devoted to several lectures that explain to students the context of the disciplines and their main principles. from this approach, the s tudents are selecting their subject after the second conference. each group had to develop one topic, chosen for its s trategic importance, such as the reliability of the space electronics, the data centers, the challenge of very high density packaging, but also the huge development of connecting objects, the ulsi and the very large area electronics. thus, from a given set of main literature references, the students prepared during three days a twelve pages report and an oral presentation, both in english language . advances in technology innovation , vol. 3, no. 3, 2018, pp. 126 132 copyright © taeti 130 5. effective works of the students the students had to prepare their presentations. the subjects of the presenta tion and of the short report cover the most important aspects of the field, i.e., the evolution of the integration, the development of new technologies, the extension of the field of applications, more especially in the evolution of the connecting objects and of internet of things (iot) on the following topics for the microelectronics field:  new technologies for the future integration ,  evolution of mems,  large area electronics ,  microelectronics for telecommunications,  connecting object for environment application  the substrates for integrated microelectronics and solar cells . for the reliability, the subjects cover the main types of applications on the devices and on the systems :  the reliability of the space electronics,  the harsh environments: reliability impact on electronics devices ,  the reliability of data centers,  the challenge of the very high density packaging,  the dielectric degradation and breakdown.  the radiation effects. 6. survey based on a questionnaire: main results in order to have a good appreciation on the pedagogic efficiency of the approach, the students were asked on the several points of the experience. ten questions were directly focused on the organization and on the content of both modules :  adequacy of the topics to the curricula of the bachelor students ,  diversity of the subjects with the curricula (engineering),  duration of the modules that were concentrated on two weeks only ,  organization (lectures, schedule, presentations ),  quality of the evaluation of the work (marks obtained by t he students),  originality of the pedagogical approach ,  contribution to the awareness of the technological challenges ,  improvement of the general skills and knowledge,  relevance of an oral presentation,  impact on english language practice. the answers were shared in five types: very bad, bad, good, very good and excellent. the results are anonymous in order to avoid any limitation on the student side. in addition, a case was reserved to free comments. very few students have filled this case. among the sixty students, fifty eight have answered. thus, the return percentage is 96.7% that represents a real significant sampling of appreciations. fig. 3 shows the histograms of the answers for the two modules . advances in technology innovation , vol. 3, no. 3, 2018, pp. 126 132 copyright © taeti 131 fig. 3 cumulated answers of the survey. on the ten questions (300 answers), for both specialities, the students found the approach very good in average. a large majority has considered that the foreign language was not a drawback. only two students had some bad feeling on the modules (organization and interest) for both specialties, the students found the approach very good in average; more than 70% have considered the experience very good or excellent. in the free comments, no remark on the amount of personal work they had to do. they well accept the approach mainly due to a good motivation and probably for its novelty, for them. a large majority has considered that the for eign language was not a drawback. only two students among the fifty -eight had some bad feeling of the modules (organization and interest). fig. 4 results of the survey. general appreciation of the students on the two modules. only two students among fifty -eight were not satisfied of the modules and of the flipped class approach fig. 5 result of the survey on the opportunity to reproduce this experience for the next promotion. only three students among fifty-eight have considered that the modules could be cancelled next year fig. 4 shows the general evaluation of the modules. a large majority was very satisfied and more (a majority of excellent!). its represents more than 80%. this is a proof that the two modules have answered to a need that was very appreciated by the students. on the basis of these results, a specific question about the reproduction of the same modules for the next promotion was asked. a large majority of the answers was “yes”. fig . 5 shows the histogram of these answers. advances in technology innovation , vol. 3, no. 3, 2018, pp. 126 132 copyright © taeti 132 among fifty eight students, only 3 estimated that the experience is not interesting for the new students. in other words, 94% of the students found that this experience can be reproduced for the future honors . 7. conclusions results were generally very good. most of the students succeeded in addressing issues that were completely new for them. they clearly built by themselves the skills allowing understanding of all the important aspects of the topics they had to app roach. they were able to highlight the main problems to solve, to improve the current solutions, and to perfo rm a presentation in english. in conclusion, such a self-made intermediate project by the students has been very beneficial to introduce new teaching sessions related to the new concepts applied to advanced microelectronic technologies and high reliabilit y devices, within the framework of international studies. this approach is included in the updated strategy for higher education oriented towards innovation [9]. the french national network of microelectronics, gip-cnfm [10-11], has adopted this strategy for many years. more recently, the chinese government [12] has also followed this evolution . acknowledgment the authors want to thank the colleagues with south -east university that helped them for the pedagogic and administrative organization of this work. special thanks to lorraine chagoya-garzon, executive assistant of the gip-cnfm, for the technical support in the redaction of this paper. references [1] o. bonnaud and l. fesquet, “multidisciplinary topics for the innovative education in microelectronics and its applications,” proc. of 14th international conf. information technology based higher education and training , 2015, pp. 1-5. [2] o. bonnaud and l. fesquet, “innovating projects as a pedagogical strategy for the french network for education in microelectronics and nanotechnologies,” proc. of ieee international conf. microelectronic systems education, 2013, pp. 5-8. [3] o. bonnaud and l. fesquet, “communicating and smart objects: multidisciplinary topics for the innovative education in microelectronics and its applications,” proc. of international conf. information technology based higher education and training, june 2015, pp. 1-5. [4] g. matheron, keynote, microelectronics evolution, european, microelectronics summit, paris, 2014. [5] f. jensen, electronic component reliability: fundamentals, modelling, evaluation, and assurance, 1st ed., wiley, 1996. [6] g. e. moore, “cramming more components onto integrated circuits,” electronics magazine, vol. 38, no. 8, pp. 114-117, 1965. [7] o. bonnaud and l. fesquet, “trends in nanoelectronic education from fdsoi and finfet technologies to circuit design specifications,” proc. 10th european workshop on microelectronics education, may 2014. [8] m. swaminathan and j. m. pettit, 3rd system integration workshop , 2011. [9] o. bonnaud and l. fesquet, “the new strategy based on innovative projects in microelectronics and nanotechnologies,” ecs journal of solid state science and technology , vol. 2, no. 11, pp. 1-7, 2013. [10] “cnfm: coordination nationale pour la formation en microélectronique et en nanot echnologies,” http://www.cnfm.fr. [11] o. bonnaud and p. gentil, “gip-cnfm: a potential model for micro and nanoelectronics education, invited communication, design and technology of integrated systems in nanoscale era,” proc. design and technology of integrated systems , march 2008. [12] o. bonnaud and l. wei, “a way to introduce innovative approach in the field of microelectronics and nanotechnologies in the chinese education system,” proc. of engineering and technology innovation, vol. 4, pp. 19-21, 2016.  advances in technology innovation , vol. 2, no. 2, 2017, pp. 51 55 51 simulation of desiccant cooling kamaruddin a. 1,* , aep s. uyun 1 , alie bamahry 2 and rino imanda 2 1 the graduate school/renewable energy, darma persada universityjl. indonesia 2 graduate student, the graduate school/renewable energy, darma persada university, indonesia received 09 march 2016; received in revised form 30 april 2016; accept ed 05 may 2016 abstract desiccant cooling system has been an attractive topic for study lately, due to its env ironmentally friendly nature. it also consume less electric ity and capable to be operated without refrigerant. a simu lation s tudy was conducted using 1.5 m long ducting equipped with one desiccant wheel, one sensible heat e xchanger wheel, one evaporative cooling chamber and two blowers and one electric heater. the simu lation study used 8.16 m/s prima ry air, the drying coefficient fro m desiccant wheel, k1=2.1 (1/s), mass transfer coeffic ient in evaporative cooling, k2=1.2 kg vapor/s, heat transfer coefficient in desiccant wheel, h 1=4.5 w/ m 2 o c, and heat transfer coeffic ient in sensible heat e xchanger wheel h2= 4.5 w/ m 2 o c. the simulat ion results show that the final te mperature before entering into the a ir conditioning roo m was 25 o c and rh o f 65 % , we re in accordance with the indonesian comfort index. ke ywor ds: desiccant wheel, desiccant cooling, evaporative cooling, sensible heat exchanger wheel, silica gel 1. introduction indonesia lies in the tropic whe re the average air te mperature is around 30 o c and rh around 80% a ll year round. under this weather cond ition it is not fit for people to work in the office or stay at home. therefore, there is an urgent need for air conditioning fac ilities in order to be able to live in a better condition. there is also need for industry to increase their productivity by creating better working environ ment. as the current air conditioning system requires high electric ity consumption which requires high fossil fuel input, there is a need to find altern ative for the conventional air conditioning system. the best option would be a desiccan t cooling system wh ich does not need refrigerant for o perating the system. through the manipu lation of air condition using silica ge l it is possible to create a comfort condition of a room. research on this type of a ir conditioning system is new in indonesia and very rare if any attempt to apply the system in indonesia. re search by chadi maalouf. at al in france (2006) indicated that by using solar energy they were capable to construct adsorption cooling system for application in several city in france. daou et al. (2004) and jurinak (1982) have conducted research to determine the perfo rmance of an adsorption cooling machine using silica gel the pennington cycle. rajat subhra das et al. (1995) study the application of solar energy for liquid desiccant cooling system in india t wo dimensionless parameters enthalpy and moisture effect iveness are taken as performance indices of the absorber. the performance of the overall system is presented in terms of its cooling capacity, moisture re moval rate and cop (coefficient of performance). davangere et al. (1999) had applied a desiccant cooling system with capacity 10 kw (2.85 ton refrigeration) assisted by vapor compression machine. the resulting room te mperature 26.7 o c with humidity ratio of w=0.01183 kg/kg dry a ir for the condition florida wh ich have outside air of 36 o c. they conducted analysis using psychrometric chart and the result of their simu lation works were applied to four c ities in the usa. be llia , et al (2000) had studied several hybrids cooling system using various desiccant wheels and using desica lc tm co mputer progra m and applied to four cities in italy. they concluded that the ma ximu m saving in cost was 22% , and for the theater the saving were greater fro m 23% to 38% with electricity saving of up to 55%. the purpose of the study is to obtain mathe matica l mode l for the purpose of simula tion of desiccant cooling system. * corresponding aut hor, email: kamaruddinabd@gmail.com advances in technology innovation , vol. 2, no. 2, 2017, pp. 51 55 52 copyright © taeti 2. the working principle of a desiccant cooling system fig. 1 shows the ma jor co mponent of a desiccant cooling system which comp rises of a desiccant wheel containing silica-gel, sensible heat exchanger whee l, a hot water heater su pplied fro m solar co llector, b lowers and evaporative cooler (pons and kodama, 2014). outside air is introduced through point (1) passing the hot desiccant wheel where the hu mid ity is re duced to point (2). the air will further passed through the sensible heat exchanger (point 3) where its temperature will be reduced while keeping its rh constant. fro m the sensible heat e xchanger the air will be introduced into the evaporative cooling where its temperature will be reduced by its rh will be increased (point 4). when entering the room the temperature and rh will reach 26 o c and 55 % , respectively, a comfo rtable condition for a ir conditioning. the air condition in the desiccant cooling system can also be traced using the psychometric ch art in fig.2. fro m the room under condition of point (5) the air will be passed again through the evaporative cooling unit which will reduced its temperature and increase its rh. after passing through the sensible heat e xchanger its temperature will inc rease while its rh is kept constant as in point (6). after passing through the heater, the air temperature inc rease again heating the desiccant wheel to the condition as point (8). after passing the desiccant wheel the a ir will gain moisture due to evaporation from the de siccant wheel and it temperature will drop. fig. 1 main co mponent of desiccant cooling system (cnrs-limsi and ku ma moto university, 2014) fig. 2 air condition in a psychometric chart (cnrs-limsi and kumamoto university, 2014) 3. mathematical modelling if m is the total mass of air in the duct and x is the humid ity ratio then the humidity ratio change along the z a xis of the total length of 1.5 m of the duct can be calculated using the fo llowing mass balance equations m 0.5z0 f/)(1  orvmemk dz dx 　 (1) m 1.5z 1.1for /)(  vmsxxc dz dx (2) to calculate the change in temperature along the duct from the in let to the duct out let and energy balance will be used. m 0.5z0.0for )1(11   ttxaxh dt dz x dz dt mcp (3) m 1.1z0.5for )2(22   ttxaxh dt dz x dz dt mcp (4) 1.5mz1.1for )(   sttevh dt dz x dz dt mcp (5) for the condition of the return a ir fro m the air condition room, the following mass balance equation will be used. advances in technology innovation, vol. 2, no. 2, 2017, pp. 51 55 53 copyright © taeti m 1.1z 1.5for /)(  vmsxxc dz dx (6) m 0.0z0.5 f /)(1   or vmemk dz dx (7) fro m the energy balance the following rela tion can be obtained 1.5mz1.1for )(   sttevh dt dz x dz dt mcp (8) m 0.8z1.1for )2(22   ttxaxh dt dz x dz dt mcp (9) )2(22 2 22 ttxaxh dt dtx xcpxm  (10) mz t hx t hx a hx h dz dt mcp 5.08.0for )(   (11) m 0.0z0.5for )1(11   ttxaxh dt dz x dz dt mcp (12) )1(11 1 11 txtxah dt xdt xcpxm  (13) the a mount of heat supplied fro m solar collector can be ca lculated using the fo llo wing equations. )( tatcacluacrad iqu  (14) )( hxttcwcpwmqu   (15) 4. system simulation with the use of parameters listed in table 1, a simulation study was conducted. the results are as shown in fig. 4 for hu midity rat io change along the duct and in fig. 5 showing the temperature change. table 2 shows simu lation data for solar collector hot water supply. table 1 simulation data quantity quantity quantity m=0.06kg ts=23 o c hev=3.7 (w) v= 8.16 m/s ax1=4.5 m 2 mx2=1.25 kg k1=0.5 (1/s) ax2=1.5 m 2 cpx2=0.897 kj/kg o c k2= 0.5 (kg vapor/s) h1=4.5 w/m 2 c ahx=1.5 m 2 xs=0.007 h2=4.5 w/m 2 c hhx=3.5 w/m 2 o c me=7(%db) tx3=68 o c thx=68 o c mx1=2.3 kg cpx1=0.921 kj/kg o c table 2 data for solar collector heating mw (kg/s) cpw(kj/kg o c) irad (w/m 2 c) 15 4.19 600 ac (m 2 ) ul (w/m 2 c) 1.64 4.5 change in absolute humid ity of inco ming outside air is presented in fig. 3 while the change in the incoming air te mperature is shown in fig. 5 below. fig. 3 change of entering a ir hu midity ratio across the duct fig. 4 change of entering air te mperature fro m duct inlet 0 0.005 0.01 0.015 0.02 0 0.20.40.60.8 1 1.21.4 h u m id it y r a ti o ( k g v a p o r/ k g d ry a ir ) distance from duct inlet(m) 0 20 40 60 0 0.2 0.4 0.6 0.8 1 1.2 1.4a ir t e m p e ra tu re (o c ) distance from duct inlet (m) advances in technology innovation , vol. 2, no. 2, 2017, pp. 51 55 54 copyright © taeti for the returning air fro m the room its h umidity ratio and te mperature change are as shown in fig. 6 and 7. fig. 5 change in returning humidity ratio along the duct fig. 6 change in air te mperature leav ing the conditioned room to achieve solar collector te mperature of 75 o c there is a need to supply 651.9 watt of energy and if this heat is supplied to the heater in the form of heat e xchanger so that heat exchanger temperature can reach 65 o c the rate of water flow should be kept at 15 kg/s this te mperature will be used to heat the desiccant wheel to drive the moisture out from the desiccant. if the results of humidity ratio change an d the air te mperature change a long the duct are plotted in the psychrometric chart the results is shown as in fig. 8. fig. 7 change in a ir condition a long the duct as plotted in the psychrometric chart (da vanagere dkk, 1999) as shown here the air condition entering the room has achieved the comfort condition of 25 o c and rh of 65% (see point 4). 5. conclusions 1) it was possible to develop mathe matica l model for desiccant cooling. 2) simu lation results using 0.35 x 0.35 m c ross section and length of 1.5 m, with air flow rate of 0.5kg/ m 2 s, and with heater te mperature of 68 o c and using desiccant wheel, a sensible heat exchanger wheel and evaporative coo ling it was possible to create comfo rt air condition of 25 o c and rh 65%. 3) it was necessary to supply heat from solar collector having area of 1.64 m 2 with average solar irradiation of 600 w/ m 2 and water flo w rate 15 kg/s in order to produce the necessary heating temperature of 68 o c. acknowledgement the authors wish to e xtend th eir g ratitude to the directorate genera l of higher education for providing a research grant under contract no.104/k3/km/2015, february 23, 2015 and to darma pe rsada university research institute and public empowerment and cooperation through contract no : 022/ sp3 / lp2mk / unsada/ii/2015 february 23, 2015. nomenclature ac = area of solar collector (m 2 ) ax1 = heat transfer surface of the desiccant (m 2 ) ax2 = surface heat transfer of the sensible heat exchanger (m 2 ) c = mass transfer coefficient in the evaporative cooler (kg/s.) cp = specific heat of the air (kj/kg o c) cpx1 = specific heat of desiccant (kj/kg o c) cpx2 = specific heat of sensible heat exchanger (kj/kg o c) h1 = heat transfer coefficient of the desiccant wheel (w/m 2 o c) h2 = heat transfer coefficient of the sensible heat exchanger (w/m 2 o c) hhx = heat transfer coeffic ient of the heater (w/m 2 o c) 0 0.01 0.02 0.03 1.5 1.3 1.1 0.9 0.7 0.5 0.3 0.1 h u m id it y r a ti o (k g v a p o r/ k g d ry a ir ) distance from duct inlet (m) 0 20 40 60 80 1.5 1.3 1.1 0.9 0.7 0.5 0.3 0.1 a ir t e m p e ra tu re (o c ) distance from duct inlet(m) advances in technology innovation, vol. 2, no. 2, 2017, pp. 51 55 55 copyright © taeti hev = rate of heat transfer with the evaporative cooling (w) irad = solar irradiation (w/m 2 ) k1 = desorption constant (1/det.) m = mass of air (kg) mx1 = mass of desiccant wheel (kg) mx2 = mass of sensible heat exchanger wheel (kg) m = moisture content of desiccant (%db) me = equilib riu m mo isture content of desiccant (%db) qu = usefull energy from the sun (watt) ta = ambient temperature ( o c) tc = collector temperature ( o c) thx = heat exchanger temperature ( o c) ts = temperature of evaporative cooling ( o c) tx1 = desiccant temperature ( o c) t x2 = te mperature of sensible heat exchanger ( o c) ul = overall loss coeffic ient of the collector (w/m 2 c) v = air flow rate (m/det.) references [1] l. be llia, p. ma zzei, f. min ichie llo, and d. palma, “air conditioning systems with desiccant wheel foritalian climates ,” international journal on architectural sc ience, vol. 1, no. 4, pp. 193-213, 2000. [2] c. maalouf, e. wurtz, and f. a lla rd, “evaluation of the cooling potential of a dessicant evaporative cooling system using the simspa rk env iro n ment ,” uta p, university of reims, ca mpus du moulin de la housse, reims, france, chadi. maa louf@ univ-reims.fr. [3] b. s. davangere, s. a. she rie f, and d. y. goswa mi, “feasibility s tudy of solar des icant air-conditioning system–part 1. psychrometric and analysis of the condition -ed zone,” international journal of energy research, vol. 23, no. 1, pp. 7-21, 1999. [4] k. daou, r. z. wang, and z. z. xia, “desiccant cooling air conditioning: a review ,” renewable & sustainable energy rev iews , vol. 10, no. 2, pp. 55-77, 2016. [5] m. m. s. de zfouli, s. mat, g. pirasteh, k. s. m. sahari, k. sopian, and m. h. ruslan, “simu lation analysis of the four configure tions of solar desiccant cooling system using evaporative cooling in tropical weather in malaysia,” international journal of photoenergy, vol. 2014, http://dx.doi.org/10.1155/2014/ 843617 [6] s. p. halliday, c. b. beggs , and p. a. sleigh, “the use of solar desiccant cooling in the uk: a feasibility study,” applied therma l engineering, vol. 22, no. 12, pp.1327-1338, 2002. [7] l. he rnandez, j. heywood, a. ku mar, and y. ro man, sola r desiccant air conditioner, http://solar.sdsu.edu/final docs/solar air conditioning final report 490b.pdf , visited april, 2014. [8] j. j. jurinak, “open cycle desiccant cooling component models and s ystem simulat ions,” ph. d. thesis, dept. mech. eng., university of wisconsin-madison. [9] s. ja in , p. l. dha r, “ev a luat ion o f solid-desiccant-based evaporative cooling cycles for typical hot and humid c limates ,” international journal of refrigerat ion , vol. 18, no. 5, pp. 287–296, 1995. [10] m. pons, and a. kodama, “second law analysis of the open cycles for solid desiccant air-conditioners,” c.n.r.s.l.i.m.s.i., france a nd ku ma mot o un ive rs it y, jap an , https://perso.limsi.fr/mpons/dsopen.htm, visited april 2014  advances in technology innovation, vol. 3, no. 4, 2018, pp. 179 184 evaluation of infill effect on mechanical properties of consumer 3d printing materials gabriel a. johnson, jesse j. french * department of mechanical engineering, letourneau university, longview, texas, usa received 20 july 2017; received in revised form 04 december 2017; accepted 05 december 2017 abstract during the additive manufacturing “boom” of the last decade, consumer level 3d printers have kept pace with commercial/industrial printers, both in numbers and features. however, in material characterization data, the access to date for the consumer has significantly lagged behind. consumer level 3d printers provide a significant asset to entrepreneurs, small businesses, universities, college students, and hobbyists due to the low initial capital cost and relatively low operational costs. commercial grade 3d printers and the associated filaments sold for their use typically have well documented material properties and print parameters. consumer 3d printers, however, typically have limited or no access to mechanical test data for their materials. this paper describes the work of the authors to fill the existing knowledge gap in the mechanical properties of consumer level 3d printer filament. astm tensile (d638) tests were performed on samples produced by two commercially available 3d printers. the materials tested include pla, abs, petg, various nylons, polycarbonate/abs, and asa filaments. samples were printed with infill percentages ranging from 15% to 100% to test for tensile properties. keywords: additive manufacturing, polymers, 3d printing, tensile testing, thermoplastics 1. introduction additive manufacturing, often referred to as 3d printing, gives companies and individuals the ability to rapidly design, test, and improve concepts, as well as the ability to mass-produce components. commercial machines, while expensive to own and operate, produce consistent and reliable printed components due to the mechanical testing and process optimization performed along with the development of the printer. thus far, the testing and process optimization for consumer-level machines has lagged. since some consumer 3d printing is employed by hobbyists, the strength of available printing materials has been considered irrelevant. however, for those producing load-bearing components with consumer-level 3d printers, the strength of printing-compatible materials are important. fig. 1 infill comparison for printed components * corresponding author. e-mail address: jessefrench@letu.edu advances in technology innovation, vol. 3, no. 4, 2018, pp. 179 184 copyright © taeti 180 astm (american society of testing and materials) standard d638, standard test method for tensile properties of plastics, was used as a guide for the testing procedure [1]. for this work, the infill percentages shown in fig. 1 were tested. the 15%-90% samples were a triangular infill, with infill lines at 0°, +60°, and -60° relative to the principle axis of the tensile specimen. due to the nature of the slicing software, the 100% sample had to be printed with solid layers in a ±45° orientation, relative to the principle axis of the specimen. although each material is printed at different temperatures with different travel speeds according to the needs of the material, the final geometry allows for direct comparison between materials. 2. materials tested in total, seven materials were tested for a total of three companies. polylactic acid (pla) is a common material for 3d printing [2]. it is relatively easy to print, but is somewhat brittle and has a low service temperature. acrylonitrile butadiene styrene (abs), often used for injection molded components, is likely the most common thermoplastic in use today [2]. abs can be more difficult to print due to layer bonding and warping issues, but has improved mechanical properties over the pla. polyethylene terephthalate glycol (petg), a thermoplastic co-polyester, is quickly becoming a more common material for 3d printing [3]. acrylonitrile styrene acrylate (asa) is a uv-stable thermoplastic, often used in the marine, auto, and rv industries due to its weathering resistance [4]. asa is similar to abs in terms of mechanical properties and printing parameters. polycarbonate/abs (pc/abs) is a blended thermoplastic, designed for high-heat applications [5]. pc/abs also has a high tendency to warp, though an enclosed build volume reduces the tendency to warp. petg produces parts with little to no warping, but can have a poor surface finish if improper printer settings are used. pla, abs, and petg were all purchased from the company makergeeks. the asa and pc/abs were purchased from another company. the nylon materials tested were all produced by the company taulman3d. nylon materials are known for their strength, chemical, and thermal resistance. nylon 645 is a specially designed nylon 6/9, optimized for improved tensile strength and improved optical clarity [6]. nylon 910 is a special form of nylon developed by taulman3d expressly for the purpose of 3d printing. it is extremely tough, strong, and resistant to high temperatures [6]. table 1 shows a summary of the published material properties for the materials tested in this work. σy is the yied stress, or the stress at which the material begins to rupture, leading to failure. table 1 approximate material properties of printed materials, taken from various sources material density [g/cm3] σy [mpa] modulus [gpa] failure strain [%] pla [12] 1.29 44.8 3.8 23.1 abs [13] 1.07 43.3 2.3 24.8 asa [4] 1.08 44.6 2.3 34.1 pc/abs [5] 1.17 56.0 2.7 63.3 petg [3] 1.27 65 3.0 109.0 nylon 6/6 [9] 1.14 75 2.8 50 3. specimen production the specimens tested in this work were produced using two commercially available printers, shown in fig. 2. table 2 shows a summary of the printing parameters used for each material and indicates which machine printed each sample. the pla, abs, petg, and nylon materials were printed from nominal 2.85mm printer filament, while the asa and pc/abs were printed from nominal 1.75mm printer filament. the dimensions of the samples were taken from the type 1 sample defined by the astm standard [1]. this specimen type was chosen since it is large enough for the printing infill to become a predominate component of the specimen strength. in fig. 2, the left image is of a commercially available, consumer-grade 3d printer, while the right image is of a printer constructed by the author from a kit of parts. simplify3d was used as the slicing software and all the printer gcode files were produced through it. samples were produced in groups of three according to infill percentage, labeled, and organized by sample type. samples with apparent flaws in the test section or radius, as well as samples that were warped, were reprinted. in order to improve print quality, all materials were dried for at least 24 hours in a desiccant-filled box advances in technology innovation, vol. 3, no. 4, 2018, pp. 179 184 copyright © taeti 181 heated to 45°c. filament drying is optional for some materials, but abs, asa, petg, nylon, and pc/abs materials are hydroscopic and must be dried prior to printing [7]. some materials with a high possibility of warping, such as the nylons or abs, were printed with a brim; a single-layer feature added by the slicing software to increase the bed contact area, thereby increasing adhesion and reducing warping. aside from the printing parameters mentioned in table 2, all samples were printed with three top/bottom layers, three perimeters, and a triangular infill pattern. the 100% infill samples were printed solid by forcing the slicing software to make top/bottom layers through the whole thickness of the part. the print cooling fan was used for pla, abs, and petg in order to improve the surface finish. previous work by lanzotti indicates that the number of external perimeters has a much larger effect on tensile strength than layer thickness, so layer thickness was chosen according to material-specific best practice [8]. (a) abs 15, 30, and 50% specimens mid-print on printer 1 (b) pc/abs 90% specimens min-print on printer 2 fig. 2 print parameters for all materials table 2 print parameters for all materials material printer temperatures [℃] (hot end/ bed) layer height [mm] print speed [mm/s] pla 1 215/70 0.21 45 abs 1 250/110 0.22 40 petg 1 245/70 0.25 45 nylon 910 1 235/100 0.25 30 nylon 645 1 235/110 0.25 20 asa 2 245/110 0.20 35 pc/abs 2 285/115 0.25 30 4. test method fig. 3 specimen loaded in tensile testing machine, prepared for test fig. 4 crazing marks visible on outside of abs 50% infill sample advances in technology innovation, vol. 3, no. 4, 2018, pp. 179 184 copyright © taeti 182 after printing, samples were removed from the print bed, post-processed to remove the brim (if present), labeled according to material, infill percentage, and specimen number, measured with digital calipers, and then tested. testing was performed using a mti 5k bench-top universal testing system fitted with serrated-jaw tensile grips and an epsilon strain extensometer rated for 10% strain. samples were loaded into the jaws, checked for vertical alignment, and then tested at a crosshead travel speed of 50 mm/min, as specified by the astm standard for rigid polymers [1]. fig. 3 shows an example of a specimen loaded in the tensile testing machine, prior to testing. after specimen failure, the extensometer was removed from the specimen and the specimen removed from the tensile jaws. failed specimens were retained for photography and analysis. due to supply limitations, only three samples each material was tested unless the sample needed to be retested due to a test failure. 5. results fig. 7 shows a summary of these results, comparing average tensile data from all materials tested. as expected, the tensile yield strength of the samples was directly affected by infill percentage. at 15% infill, all seven materials had a yield stress less than 25 mpa. the pc/abs was the strongest at 15% infill, while asa, abs, and nylon 910 tied for the lowest yield strength at 15% infill. as the infill percentage increased, the range of yield strengths also increased. at 30% infill pla had a yield strength of 30.1 mpa, while the nylon 910 had a yield strength of 13.1 mpa. the 50% infill samples had a tighter range of tensile strengths, varying from 30mpa (pc/abs) to 15.6 mpa (nylon 910). above 50% infill, the yield strengths began to increase significantly. at 75% infill, the petg had an average tensile yield strength of 36.1 mpa while the asa had an average tensile yield strength of 19.1 mpa. finally, at 100% infill, the nylon 910 had an average yield stress of 69.9 mpa, making it the strongest material tested. pla, which was the weakest material at 100% infill, had an average yield stress of 32.98 mpa. on an individual-material level, nylon 910 was the most affected by infill percentage; 12.7 mpa at 15% infill to 69.9 mpa at 100% infill, a difference of 57.2 mpa. pla, while weaker overall, varied considerably less than the nylon 910; 18.4 mpa at 15% infill to 32.9 mpa at 100% infill, a difference of 14.5 mpa. at 30% infill, the pla had the highest tensile strength of 30.1 mpa, a value higher than the 50% and 75% infill tensile strengths for the same material. petg varied from 16.4 mpa at 15% infill to 51.8 mpa at 100% infill, a variance of 35.4 mpa. fig. 5 comparison of ductile (top) and brittle (bottom) failure modes for two petg specimens. the top specimen is 100% infill while the bottom sample is 30% infill the modulus of elasticity is also dependent on infill percentage. as the infill percentage increases, so does the modulus. this effect can be seen in fig. 6, which show a comparison of different infill percentages within the petg material. the slope of the elastic region of the data line is the modulus of elasticity, which increases as the infill percentage increases. for the petg, the modulus increased from 0.69 gpa to 1.85 gpa as the infill percentage increased from 15% to 100%. fig. 5 shows how the 30% petg sample failed with almost no deformation, while the 100% sample exhibited necking prior to failure. nylon 910 varied from 0.34 gpa to 2.44 gpa as the infill increases from 15% to 100%. the remaining materials also indicate modulus dependence on infill percentage. advances in technology innovation, vol. 3, no. 4, 2018, pp. 179 184 copyright © taeti 183 in some materials, the infill percentage affected the failure mode of the specimens. for example, in fig. 5, the petg 100% infill specimen failed in a ductile manner, as indicated by the severe necking in the center of the reduced section. the bottom specimen in fig. 5, which was a petg 30% infill specimen, failed in a more brittle manner. these failure modes are supported by the data from petg tests shown in fig. 6; the curve for the 30% specimen fails without much plasticity (failure at 3.22% strain) while the 100% curve indicates plastic deformation (failure at 11.3% strain). the nylon 910 failed in a similar manner, though all of the 910 samples had some degree of plasticity prior to failure. all infill percentages of the pla samples failed in a brittle manner, with no visible necking or deformation prior to failure. the abs samples showed evidence of crazing, a change in color due to mechanical deformation, prior to failure [9]. fig. 4 shows an example of the crazing on an abs sample, as well as a cross section of the failed component. of particular interest in fig. 4 is the alignment of the crazing marks with the infill lines. the top of fig. 4 shows a cross section of the same sample shown in the center of fig. 4. many of the specimen failures occurred along an infill line, indicating that the infill pattern may cause a stress concentration. fig. 6 comparison of tensile data for petg samples. legend format: petg[% infill]-[sample number] fig. 7 average tensile strengths for all materials tested at a variety of infill percentages fig. 8 comparison of required infill percentages for same component strength while the 100% infill specimens are consistently the strongest, solid components take more time and material to produce. there exists a “sweet-spot,” a point at which material and time costs are minimized while component strength is maximized. for evaluation of this sweet-spot, consider a cube with 60mm sides (216 cm 3 volume), that needs to have a tensile strength of at least 30 mpa. if any of the materials tested in this work were available, each would require a different infill percentage to achieve this strength. since the volume of polymer filament required directly affects the cost of printed components, required polymer volume is an important design consideration. fig. 8 represents this concept graphically, showing that in order to reach the desired tensile strength, different infill percentages need to be used depending on the material used. not represented in fig. 8, but still of importance to anyone designing a printed component, are the ultimate strain of the plastic, chemical resistance, and thermal-mechanical properties. evaluating the effect of infill on these mechanical properties is beyond the scope of this work. advances in technology innovation, vol. 3, no. 4, 2018, pp. 179 184 copyright © taeti 184 6. conclusions the data presented from this research provides a design basis for the designer seeking to create a component with a specific loading requirement. this project tested seven materials from three manufacturers of polymer filament. two consumer-grade, open-sourced 3d printers were used to fabricate the tensile specimens. at 100% infill, nylon 910 had the highest tensile strength at 69 mpa, while pla was the weakest at 33 mpa. as the infill percentage decreased, the range of tensile strengths for all materials was more tightly grouped, indicating that infill percentage has a significant impact on tensile strength. infill percentage was also found to affect modulus, elongation, and failure mode. references [1] astm, astm d638-14: standard test method for tensile properties of plastics, 2014. [2] lulzbot, “filament guide,” https://devel.lulzbot.com/filament/archive/lulzbot_3d_printing_filament_guide.pdf. [3] o. olabisi and k. adewale, handbook of thermoplastics, 2nd ed. boca raton: crc press, taylor & francis group, 2016. [4] 3dxtech, “3dxmax asa 3d filament,” http://www.3dxtech.com/content/asa_filament_v2.1.pdf. [5] 3dxtech, “3dxmax pc/abs 3d filament,” http://www.3dxtech.com/content/pc-abs_filament_v2.1.pdf. [6] taulman3d, “main features,” http://taulman3d.com/main-features.html. [7] t. landry, “beat moisture before it kills your 3d printing filament,” https://www.matterhackers.com/news/filament-and water. [8] a. lanzotti, m. grasso, g. staiano, and m. martorelli, “the impact of process parameters on mechanical properties of parts fabricated in pla with an open-source 3-d printer,” rapid prototyping journal, vol. 21, no. 5, pp. 604-617, 2015. [9] f. p. beer, e. r. johnston, j. t. dewolf, and d. f. mazurek, mechanics of materials, new york: mcgraw-hill, 2012. [10] matweb, “overview of materials for polylactic acid (pla) biopolymer,” http://www.matweb.com. [11] matweb, “overview of materials for acrylonitrile butadiene styrene (abs), molded,” www.matweb.com.  advances in technology innovation , vol. 1, no. 1, 2016, pp. 01 06 1 copyright © taeti the effect of the curvature-rate on the response of local sharp-notched sus304 stainless steel tubes under cyclic bending kuo-long lee 1 , shih-bin chien 2 , and wen-fung pan 3,* 1 department of innovative design and entrepreneurship management, far east university , tainan, taiwan. 2,3 department of engineering science, national cheng kung university, tainan, taiwan . received 20 january 2016; received in revised form 24 march 2016; accept ed 28 march 2016 abstract in this study, the response of local sharp-notched sus304 stainless steel tubes with diffe rent notch depths of 0.2, 0.4, 0.6, 0.8 and 1.0 mm subjected to cyclic bending at different curvature-rates of 0.0035, 0.035 and 0.35 m -1 s -1 were e xpe rimentally investigated. the tube bending machine and curvature-ovalization measure ment apparatus, which was designed by pan et a l. [1], we re used for conducting the curvature-controlled cyclic bending. for a constant curvature-rate, the mo ment-curvature curve revealed that the cyclic hardening and became a steady loop after a few bending cycles; the notch depth had almost no influence on the curves. moreover, the ovalizat ion-curvature curve increased in an increasing and ratcheting manner with the nu mber of bending cycles. la rge notch depths resulted in larger ovalization of the tube cross -section. in addition, for a constant notch depth, higher curvature-rates led to larger cyclic hardening and faster increasing of ovalization. keywor ds : local sharp notch, sus304 stainless s teel tubes, notch depth, curvature-rate, cyclic bending, moment, curvature, ovalization 1. introduction it is we ll known that the bending of c ircula r tubes results in the ovalization (change in the outer dia meter div ided by the origina l outer dia meter) of the tube cross -section. this ovalization increases slowly during reverse bending and continuous cyclic b ending and, in turn, results in the deterioration of the circular tube, which buckles when the ovalization reaches some critica l value. the circu lar tube is severely damaged during buckling and cannot bear the load, which ultimate ly results in obstruction and leakage of the materia l being transported. as such, a complete understanding of the response of the circular tube to cyclic bending is essential for industrial applications. in 1998, pan et al. [1] designed and set up a new measure ment apparatus. it was used with the cyclic bending machine to study various kinds of tubes under diffe rent cyclic bending conditions. for instance, pan and her [2] investigated the response and stability of 304 stainless steel tubes that were subjected to cyclic bending with diffe rent curvature-rates, lee et al. [3] studied the influence of the do/t ratio on the response and stability of c ircula r tubes that were subjected to symmetrical cyc lic bending, lee et al. [4] e xperimentally e xp lored the effect of the do/t ratio and curvature-rate on the response and stability of circula r tubes subjected to cyclic bending, and chang and pan [5] discussed the buckling life estimat ion of c ircu lar tubes subjected to cyclic bending. expe rimental investigations have shown that some engineering materials, such as 304 stainless steel, 316 stainless steel and high-strength titanium a lloy, change mechanica l properties (yield strength, hardening, ductility … etc.) under different strain-rates or stress -rates. therefore, once a tube wh ich is fabricat ed by afore mentioned materials is manipulated under cyclic bending at diffe rent curvature-rates, the response and collapse of tubes for each * corresponding aut hor, email: z7808034@email.ncku.edu.tw advances in technology innovation, vol. 1, no. 1, 2016, pp. 01 06 2 copyright © taeti curvature-rate are expected to be generated diffe rently. pan and his co-worke rs have investigated the influence of curvature-rate on the response and collapse of sus304 stainless steel tubes (pan and her [2]), titaniu m a lloy tubes (lee and pan [6]) and 316l stainless steel tubes (chang et al. [7]) subjected to cyclic bending. however, a ll of their investigations considered tubes with a smooth surface. if a tube with a notch is considered, the response should be different from a tube with a smooth surface. in this study, the response for local sharp notched sus304 stainless steel tubes subjected to cyclic bending at different curvature-rates is discussed. a four-point bending mach ine (sha w and kyria kides [8], lee et a l. [3]) was used to conduct the cyclic bending test. a curvature ovalization measurement apparatus (coma) designed and reported previously by pan et al. [1] was used to control and measure the curvature. for local sharp-notched tubes, five different notch depths, 0.2, 0.4, 0.6, 0.8 and 1.0 mm, we re considered in this study. in addition, three diffe rent curvature-rates, 0.0035, 0.035 and 0.35 m -1 s -1 , were controlled. the magnitude of the bending mo ment was measured by two load cells mounted in the bending device, and the magnitudes of the curvature and ovalization of the tube cross -section were measured by coma. 2. experiment local s harp-notched sus304 stainless steel tubes with five diffe rent notch depths were subjected to cyclic bending at three different curvature-rates by using a tubebending device and a curvature-ovalization measurement apparatus in this study. detailed descriptions of the device, apparatus, materia ls, specimens and test procedures are given as follows . 2.1. bending device fig. 1 shows a picture of the bending device. it is designed as a four-point bending machine, capable of apply ing bending and reverse bending. the device consists of two rotating sprockets resting on two support beams. heavy chains run around the sprockets and are connected to two hydraulic cylinders and load cells forming a c losed loop. each tube is tested and fitted with solid rod e xtension. the contact between the tube and the rollers is free to move along axia l direct ion during bending. the load transfer to the test specimen is in the form of a couple formed by concentrated loads fro m two of the rollers. once e ither the top or bottom cylinder is contracted, the sprockets are rotated, and pure bending of the test specimen is achieved. reverse bending can be achieved by reversing the direction of the flow in the hydraulic c ircuit. detailed description of the bending device can be found in shaw and kyriakides [8] and lee et al. [3]. fig. 1 a picture of the bending device fig. 2 a picture of the coma 2.2. curvature-ovalization measurement apparatus (coma) the coma, shown in fig. 2, is an instrument used to measure the tube curvature and ovalizat ion of a tube cross -section. it is a lightwe ight instrument, wh ich is mounted close to the tube mid-span. there a re three inclinometers in the coma. t wo inclinometers are fixed on two holders, which a re denoted side-inclinometers . these holders are fixed on the circular tube before the test begins. from the fixed distance between the two side inclinometers and the angle change detected by the two side-inclino meters, the tube curvature can be derived. in addition, a magnetic detector advances in technology innovation , vol. 1, no. 1, 2016, pp. 01 06 3 copyright © taeti in the middle part of the coma is used to measure the change of the outside dia meter. a more detailed description of the bending device and the coma is given in pan et al. [1]. 2.3. material and specimens the circu lar tubes used in this study were made of sus304 stainless steel. the tubes’ chemical co mposition is cr (18.36%), ni (8.43%), mn (1.81%), si (0.39% ), …., and a few other trace elements, with the re mainder being fe. the ultimate stress, 0.2% strain offset the yield stress and the percent elongation are 626 mpa, 296 mpa and 35%, respectively. the ra w smooth sus 304 stainless steel tube had an outside diameter do of 36.6 mm and wa ll thickness t of 1.5 mm. the raw tubes were mach ined on the outside surface to obtain the desired local notch depth a of 0.2, 0.4, 0.6, 0.8 and 1.0 mm. fig. 3 shows a schematic dra wing of the local sharp-notched tube. according to the drill of the machine, the corresponding surface dia meters b were 0.6, 1.2, 1.8, 2.4 and 3.0 mm, respectively. fig. 3 a schematic drawing of the local sharp notched tube 2.4. test procedure the test involved a curvature-controlled cyclic bending. the controlled -curvature ranges were fro m  0.015 to  0.45 m -1 and three diffe rent curvature-rates of the cyclic bending test were 0.0035, 0.035 and 0.35 m -1 s -1 . the magnitude of the bending mo ment was measured by two load cells mounted in the bending device. the magnitudes of the curvature and ovalizat ion of the tube cross -section were measured by the coma. 3. results and discussion fig. 4 shows a typical set of experimentally determined mo ment (m) curvature () curve for local sharp-notched sus304 stainless steel tubes , with notch depth of a = 0.2 mm, subjected to cyclic bending under the curvature-rate of 0.0035 m -1 s -1 . the tubes were cycled between  =  0.3 m -1 . however, the tube exh ibits cyclic hardening and becomes stable after a few cycles. since the notch is small and local, the notch depth has almost no influence on the m- curve. therefore, the m- curves for different values of a are not shown in this paper. fig. 4 experimentally determined mo ment (m) curvature () curve for local sharp notched sus304 stainless steel tube, with notch depth of a = 0.2 mm, subjected to cyclic bending under the curvature-rate o f 0.035 m -1 s -1 figs. 5(a)-(b) present experimentally determined mo ment (m) curvature () curve for local sharp-notched sus304 stainless steel tubes , with notch depth of a = 0.2 mm, subjected to cyclic bending under the curvature-rate of 0.0035 and 0.35 m -1 s -1 , respectively. it is evident that the m- curves shown in figs. 4, 5(a) and 5(b) are very simila r. ho wever, higher curvature-rates lead to higher magnitude of the ma ximu m mo ment at the ma ximu m curvature. the ma ximu m mo ments of 303, 316 and 325 n-m correspond to the curvature-rates of 0.0035, 0.035 and 0.35 m -1 s -1 , respectively. the highest and lowest curvature-rates have 100 times diffe rence. but, the ma ximu m mo ment only increases 7.3%. due to simila r phenomenon, the e xperiment results of the m- response for loca l sharp-notched sus304 stainless steel tubes with a = 0.4, 0.6, 0.8 and 1.0 mm under cyclic bending at the curvature-rates of 0.0035, 0.035 and 0.35 m -1 s -1 are omitted in this paper. advances in technology innovation, vol. 1, no. 1, 2016, pp. 01 06 4 copyright © taeti (a) (b) fig. 5 experimentally determined mo ment (m) curvature () curve for local sharp notched sus304 stainless steel tubes , with notch depth of a = 0.2 mm, subjected to cyclic bending under the curvature rates o f (a) 0.0035 and (b) 0.35 m -1 s -1 figs . 6(a)-(e) depict the e xperimentally determined ovalizat ion of the tube cross -section (δdo/do) versus the applied curvature () for local sharp-notched sus304 stainless steel tubes , with notch depths a of 0.2, 0.4, 0.6, 0.8 and 1.0 mm, respectively, subjected to cyclic bending at the curvature-rate of 0.035 m -1 s -1 . the ovalization is defined as δdo/do where do is the outside diameter and δdo is the change in the outside diameter. it can be seen that the ovalization increases in a ratcheting manner with the number of bending cycles. higher a of the notch tube leads to a more severe unsymmetrica l trend of the δdo/do- curve. in addition, higher a of the notch tube causes greater ovalizat ion of the tube cross -section. the ma ximu m ovalizations of 0.0023, 0.0025, 0.0026, 0.0027 and 0.0028 fo r the curvature of -0.3 m -1 at the 6 th cycle correspond to notch depths a of 0.2, 0.4, 0.6, 0.8 and 1.0 mm, respectively. (a) (b) (c) fig. 6 experimentally determined ovalization of the tube cross -section (δdo/do) versus the applied curvature () for local sharp notched sus304 stainless steel tubes , with notch depths of a = (a) 0.2, (b) 0.4, (c) 0.6, (d) 0.8 and (e) 1.0 mm, subjected to cyclic bending under the curvature rate of 0.035 m -1 s -1 (continued) advances in technology innovation , vol. 1, no. 1, 2016, pp. 01 06 5 copyright © taeti (d) (e) fig. 6 experimentally determined ovalization of the tube cross -section (δdo/do) versus the applied curvature () for local sharp notched sus304 stainless steel tubes , with notch depths of a = (a) 0.2, (b) 0.4, (c) 0.6, (d) 0.8 and (e) 1.0 mm, subjected to cyclic bending under the curvature rate of 0.035 m -1 s -1 figs . 7(a)-(b) depict the e xperimentally determined ovalizat ion of the tube’s cross -section (δdo/do) versus the applied curvature () for local sharp-notched sus304 stainless steel tubes , with notch depth of a = 0.2 mm, subjected to cyclic bending under the curvature-rates of 0.035 and 0.35 m -1 s -1 , respectively. it can be noted that a higher degree of ovalizat ion can be noticed under higher curvature-rates. the ma ximu m ovalizations of 0.0018, 0.0023 and 0.0028 for the curvature of -0.3 m -1 at the 6 th cycle correspond to the curvature-rates of 0.0035, 0.035 and 0.35 m -1 s -1 , respectively. the highest and lowest curvature-rates have 100 t imes diffe rence. but, the ma ximu m mo ment increases 55.6 %. it is concluded that the curvature-rate has a strong influence on the δdo/do- curve. again, due to similar results, the e xperimental results of the δdo/do- response for loca l sharp-notched sus304 stainless steel tubes with notch depths of 0.4, 0.6, 0.8 and 1.0 mm under the curvature-rates of 0.0035 and 0.35 m -1 s -1 are omitted in this paper. (a) (b) fig. 7 experimentally determined ovalization of the tube cross -section (δdo/do) versus the applied curvature () for loca l sharp-notched sus304 stainless steel tubes , with notch depths of a = 0.2 mm, subjected to cyclic bending under the curvature-rate of (a) 0.0035 and (b) 0.35 m -1 s -1 4. conclusions the response of local sharp-notched sus304 stainless steel tubes with d ifferent notch depths subjected to cyclic bending at different curvature-rates was experimentally inves tigated in this study. based on the e xperimental results, the following important conclusions can be drawn: advances in technology innovation, vol. 1, no. 1, 2016, pp. 01 06 6 copyright © taeti (1) it is found from the m- curves that the local sharp-notched sus304 stainless steel tubes with any notch depth at any curvature-rate exh ibits cyclical hardening and gradually steady after a few cycles under symmetrica l curvature-controlled cyclic bending. (2) it can be seen that a higher curvature-rate leads to a higher magnitude of the mo ment. in addition, the curvature-rate has a slight influence on the m- curves (figs. (4), 5(a ) and 5(b)). (3) it is observed from the δdo/do- curves that the ovalization of the tube cross -section increases in an unsymmetrical and ratcheting manner with the nu mber of cycles. higher a leads to more severe unsymmetrica l trend of the δdo/do- curve. (4) it can be seen that a higher curvature-rate leads to a greater ovalization of the tube cross -section. in addition, the curvature-rate has a strong influence on the δdo/do- curves (figs. 6(a), 7(a) and 7(b)). acknowledgement the support of the national sc ience council (taiwan), under grant nsc 102-2221 e-006-037 is gratefully acknowledged. references [1] w. f. pan, t . r. wang, and c. m. hsu, “a curvatureovalization measurement apparatus for circula r tubes under cyclic bending,” experimental mechanics , vol. 38, no. 2, pp. 99-102, 1998. [2] w. f. pan and y. s. her, “ viscoplastic collapse of thinwa lled tubes under cyclic bending,” journal of engineering materials and technology, vol. 120, no. 4, pp. 287-290, 1998. [3] k. l. lee, w. f. pan, and j. n. kuo, “the influence of the diameter -to-thickness ratio on the stability of c ircular tubes under cyclic bending,” international journal of solids and structures , vol. 38, no. 14, pp. 2401-2413, 2001. [4] k. l. lee, w. f. pan, and c. m . hsu, “expe rimental and theoretical evaluations of the effect between dia meter-tothic kness ratio and curvature-rate on the stability of circula r tubes under cyclic bending,” jsm e international journal, series a, vol. 47, no. 2, pp. 212-222, 2004. [5] k. h. chang and w. f. pan, “buckling life estimation of c ircula r tubes under cyclic bending,” international journal of solids and structures , vol. 46, no. 2, pp. 254-270, 2009. [6] k. l. lee and w. f. pan, “ viscoplastic collapse of titanium a lloy tubes under cyclic bending,” structural eng ineering and mechanics, vol. 11, no. 3, pp. 315-324, 2001. [7] k. l. lee, c. m. hsu, s. r. sheu, and w. f. pan, “ viscoplastic response and collapse of 316l stainless steel tubes under cyclic bending,” steel and composite structures, vol. 5, no. 5, pp. 359-374, 2005. [8] p. k. shaw and s. kyria kides, “inelastic analysis of thin-wa lled tubes under cyclic bending,” international journal of solids and structures , vol. 21, no. 11, pp. 1073-1110, 1985.  advances in technology innovation , vol. 2, no. 2, 2017, pp. 40 45 40 position control and novel application of scara robot with vision system hsiang-chen hsu 1,2,* , li-ming chu 3 , zong-kun wang 2 and shu-chi tsao 2 1 department of industrial management, i-shou university, kaohsiung, taiwan. 2 department of mechanical and automation engineering, i-shou university, kaohsiung, taiwan . 3 department of mechanical engineering, southern taiwan university of science and technology, tainan, taiwan. received 02 february 2016; received in revised form 27 april 2016 accept ed 02 may 2016 abstract in this paper, a scara robot arm with v ision system has been developed to improve the accuracy of pick-and-place the surface mount device (sm d) on pcb during surface mount process. position of the sca ra robot can be controlled by using coordinate auto compens ation technique. robotic move ment and position control are auto-calculated based on forward and inverse kine mat ics with enhanced the intelligent image vision system. the determined x-y position and rotation angle can then be applied to the desired pick & p lace location for the scara robot. a series of experiments has been conducted to improve the accuracy of pick-and-place smds on pcb. ke ywor ds : scara, pic k-and-place , forward and inverse kine matics , vision system 1. introduction industrial heavy duty manipulator such as selective co mp liance assembly robot arm (scara) is an automatic device which is capable to carry and move co mponents /devices/ parts in the manufacturing process. the first sca ra robot was invented by professor hiroshi makino fro m university o f ya manashi, japan in 1978 [1]. fig.1 de monstrates the basic structure includes kinetic 4-a xis and 4-dof, translation on x, y, z and rotation about vertical z-a xis. two-link co mp liant arms with rotated wrist behave somewhat like the hu man a rm that joints allo w the arm to move vertica lly and horizontally in a limited space. scara was specially designed for precision devices fig. 1 the first scara robot by hiroshi makino [1] assembly, especially place a pinned component in a hole. a typical scara robot is a stationary robot arm including base, elbow, vertica l e xtension and tool roll and comp rising b oth rotary and prismat ic joints . sca ra robots may vary in s ize and shape but they all are consistent in a unique 4-a xis motion [2]. w ith this distinguished feature, scara particularly fits the pick-and-place surface mount devices on pcb (printing circuit board) and move-to-take delicate silicon wafers or glass panels on magazine. scara robot was introduced in seiko watch assembling lines in 1981 and since then industrial scara robot has been widely used in electronic, semiconductor, automobile, electronic, plastic, food and pharmaceutica l factories [3] a ll over the world. sca ra robot is the principle use of robotics fie ld and most of the do ma in of robotics field is on industry and academia. the control system of industrial sca ra robot is a highly non-linear, * corresponding aut hor, email: hchsu@isu.edu.t w advances in technology innovation , vol. 2, no. 2, 2017, pp. 40 45 41 copyright © taeti strong coupling and time-vary ing systems [4]. kine mat ics modeling is one of the key technologies to verify the model. many researchers and engineers [5-6] have presented the manipulating ability of robotic mechanis ms in pos itioning and orienting end-effectors and propose a measure of manipulability. the motion tra jectory of a robot arm is calcu lated using the geometric analysis. the pid control techniques [7] have been proposed to solve the nonlinearity issue with acceptable results to control the move ment of robot arms . the motion trajectory of a robot arms is calcu lated using the geometric analysis with matlab software. the performance of industrial heavy duty robots working in unstructured environments can be improved using visual perception and learning techniques [7-10]. the object recognition is acco mplished using an artific ia l neura l network (ann) arch itecture. a novel technique used in the assembly lines integrates computer vision to capture the shape of the objects, online grasp determination based on that shape, and image-bas ed control for grasp e xecution. visual servoing system [8] consists of high-speed image processing, kine matics, dynamics, control theory, and real-time co mputing to control the position and orientation of a robot with respect to an object. furthermore, the color of objects could also be considered and included in the image-base system. the developed image processing and workpiece recognition algorithm is based on lab view vision development module [9]. later, like two eyes on human, two came ras are developed to detect the robotic arm move ment in 3-d space. in this system, the robotic arm is controlled and moved, and a fter mathe matic calcu lations the precise position of the motors is calculated to reach the designated position [10]. the automation in surface mount precision assembly lines often consists of sca ra robots equipped with grippers, image v ision system and lin ked by motorized conveyances. in this system, the combination of high performance motion control with integrated vision guidance and conveyor tracking are demonstrated. the robotic arms are used for the pick-and-place of the smt components . the place ment system on printed circuit board (pcb) is ma inly influenced by the surface mount device (sm d) robots and the production environment in assembly line. temperature on the reflow process is often above 250 o c wh ich easily distorts the tray. the deformed tray would result in misalignment of the placement system. the yield rate is cons equently descended and the cost would be enormous . the purpose of this paper is to improve the accuracy of the placement system on pcb du ring surface mount process . coordinate auto-compensation technique on deformed t ray is developed to control the position of sca ra robots equipped with image vision system. robotic move ment and position control are ca lculated based on forward and inverse kinematics . 2. theoretical development fig. 2 illustrates the coordinate system of random point located on pcb (x, y) with respect to robotic arm. fig. 2 random point (x, y) on pcb 2.1. forward kinematics location of end effector can be determined by the length of robotic a rm and rotation angle of each axis 𝑋 = 𝐿1 𝑐𝑜𝑠 𝜃1 + 𝐿2 𝑐𝑜𝑠 (𝜃1 + 𝜃2 ) (1) 𝑌 = 𝐿1 𝑠𝑖𝑛𝜃1 + 𝐿2 𝑠𝑖𝑛(𝜃1 + 𝜃2 ) (2) where l1, l2 a re robotic a rm length, and 1, 2 are rotation angle of each axis advances in technology innovation , vol. 2, no. 2, 2017, pp. 40 45 42 copyright © taeti 2.2. inverse kinematics rotation angle of each axis can be determined by the coordinate system of end point. 2 2 2 2 1 1 2 2 1 2 cos 2 x y l l l l       (3)     1 2 21 1 1 2 2 tan tan l siny x l l cos                       (4) 2.3. compensation for image and distance coordinate fig. 3 de monstrates the captured image of pixe l with respect to distance (1 pixe l equals to 0.01 mm). the origin o f d istance coordinate is located in the center of image coordinate and then pixe l (320, 240) would be the same as distance (3.2mm, 2.4mm). fig. 3 image/distance coordinate system 2.4. compensation for deformed pcb tray coordinate fig. 4 presents the compensation for deformed pcb tray coordinate. assuming the pcb tray offset 1 mm due to thermal induced warpage, the coordinate of random point in image capture system is determined. 𝑋2 = 𝑃𝐶𝐵 𝑋 + 𝑃𝑖𝑥𝑒𝑙 𝑋 (5) 𝑌2 = 𝑃𝐶𝐵 𝑌 + 𝑃𝑖𝑥𝑒𝑙 𝑌 (6) where x1=28.2, y1=52.4 are system default parameters. after mathemat ical ca lculation, the placement coordinates are 𝑋 = 𝑋2 − 𝑋1 (7) 𝑌 = 𝑌2 − 𝑌1 (8) fig. 4 schematic illustration of pcb coordinate system 2.5. image processing grayscale dig ital image is a range of shades of gray without apparent color. the reason for diffe rentiating gray images is that only specify a single intensity value for each pixel, i.e. less informat ion for each pixel. in order to reduce the comple xity of post-processing grayscale image scheme is applied to capture the characteristic image. the hough transform is a technique which can be used to isolate features of a particula r shape, such as line, circ le, e llipse, etc., within an image. based on hough transform, c ircula r detection on labview vision assistant is applied to determine the p recise position on d eformed pcb. 3. structure of improved placement system 3.1. hardware description a toshiba scara robot (#sr-424shp) with control card (#piso-ps400) shown in fig.5 is emp loyed in this study. fig. 6 illustrates smd components placed on pcb with tray (carrier). the placement system on trial run conveyer is presented in fig. 7. advances in technology innovation , vol. 2, no. 2, 2017, pp. 40 45 43 copyright © taeti fig. 5 toshiba scara robot (#sr-424shp) fig. 6 smd components on pcb with tray (carrier) fig. 7 place ment system on trial run conveyor the overall place ment system also consists of trial run conveyor for pcb tray, sm d co mponents feeder, buzze r, 3 cylinders (baffle, push and charge-in). 2 ca me ras (coordinate auto-compensation and poka-yoke), 3 sensors with arduino uno and a lighting system (top and broadside). all smd co mponents are held in specific trays which a re loaded in upstream vibratory parts feeding stations. 3.2. software development the control software us ed in this study is lab view 2012 and lab view vision assistant 2012. 4. results and discussion coordinate co mpensation system for sm d place ment and starved feeding (queliao) a uto-detection system have been developed in this paper. in the first, anchor point on pcb was shot by camera and the offset amount on pcb was then calculated by compensation of the image and distance coordinate. the precise location on pcb for co mponents placement was then d etermined. for queliao and those parts did not place in the desired position, the developed system can a lso sound a buzzer signal on the control annunciator panel and warn the upstream feeding stations to load smd components. fig. 8 flow chart of coordinate compensation system for smd placement fig. 8 and fig. 9 present the flow chart for coordinate compensation system and queliao auto-detection system, respectively. fig. 10 demonstrates the accuracy of smd co mponents pick-and-place improved by the developed algorithm. the accuracy on auto assembly (99.73%) has been dramatica lly imp roved after teaching mode (78.66%). start capture smd image grayscale digital image hough transform calculation pixel position for point (x, y) stop advances in technology innovation , vol. 2, no. 2, 2017, pp. 40 45 44 copyright © taeti fig. 9 flow chart of queliao auto-detectionsystem fig. 10 accuracy of smd components pick-and-place 5. conclusions in this paper, an improved pick -and-place smd on pcb system has been accomplished by using sca ra robot and mach ine v ision. the result has shown that 97.33% yie ld rate was achieved and more than 2% of erro r could be eliminated by improving the hardware of system. besides top lighting camera, anchor point on the skewed pcb tray can be calibrated and revised by a broadside webcam. re mote operation using wireless network is feasible. acknowledgement the authors would like to e xp ress their appreciation to m inistry of science and technology, taiwan, roc, for financia l supports under project no. most103-2221-e-214-018 and most104-2221-e-214-051. appreciation is also extended to oriental semiconductor electronic, ltd. for carrying out all the experiments. references [1] sca ra, the robot hall of fa me, powe r by carnegie mellon, http://www.robothalloffame.org/inductees/ 06inductees/scara.html [2] a. burisch, j. wrege, a. raatz, j. hesselbach, and r. degen, “parvus – miniaturised robot for improved flexibility in micro production,” assembly automation, vol. 27, no. 1, 1980. [3] s. k. dwivedy and p. eberhard, “dynamic analysis of flexible manipulators, a literature review,” journal of mechanism and machine theory, vol. 41, no. 7, pp. 749-777, 2006. [4] a. visioli and g. legnani, “on the trajectory tracking control of industrial scara robot manipulators,” ieee transactions on industrial electronics, vol. 49, no. 1, pp. 224-232, 2002. [5] g. s. huang, c. k. tung, h. c. lin, and s. h. hsiao, “inverse kinematics analysis trajectory planning for a robot arm,” proceedings of 8th asian control conference (ascc 2011), kaohsiung, taiwan, pp. 965-970, may 2011. [6] j. fang and w. li, “four degrees of freedom scara robot kine matics modelling and simulation analysis,” international journal of computer, consumer and control, vol. 2, no. 4, pp. 20-27, 2013. [7] f. escobar, s. díaz, c. gutiérrez, y. ledeneva, c. hernández, d. rodríguez, and r. lemus, “simulation of control of a scara robot actuated by pneumatic artificial muscles using rnapm,” journal of applied research and technology, vol. 12, no. 5, pp. 939-946, 2014. [8] s. h. han, w. h. see , j. lee, m. h. lee , and h. hashimoto, “image-based visual servoing control of a sca ra type dual-arm robot,” ieee international sy mposium on indu strial electronics, cholula , puebla , me xico, vol. 2, pp. 517-522, dec. 2000. [9] h. zhu, j. xu, d. he, k. xing, and z. chen, “design and implementation of the moving workpiece sorting system based on labview,” 26th chinese control and decision conference start image captured image mask advanced morphology stop part analysis http://ieeexplore.ieee.org/xpl/tocresult.jsp?isnumber=21159 advances in technology innovation , vol. 2, no. 2, 2017, pp. 40 45 45 copyright © taeti (2014 ccdc), changsha, china, pp. 5034-5038, may 2014. [10] r. szabo and a. gontean, “robotic arm control with stereo vision made in lab wind ows/cvi,” 38th international conference on telecommunications and signal processing (tsp), prague, czech, pp. 1-5, july 2015.  advances in technology innovation , vol. 2, no. 2, 2017, pp. 29 33 29 investigation of a ball screw feed drive system based on dynamic modeling for motion control yi-cheng huang * , xiang-yuan chen department of mechatronics engineering, national changhua university of education, changhua, taiwan. received 01 february 2016; received in revised form 28 april 2016; accept ed 02 may 2016 abstract this paper e xa mines the frequency response relationship between the ball screw nut preload, ball screw torsional stiffness variations and table mass effect for a single-a xis feed drive system. identificat ion for the frequency response of an industrial ball screw drive system is very i mportant for the precision motion when the v ibration modes of the system are critica l for controller design. in this study, there is translation and rotation modes of a ball screw feed drive system when positioning table is actuated by a servo motor. a lu mped dynamic model to study the ball nut preload variation and torsional stiffness of the ball screw drive system is d erived first. the mathe matica l mode ling and numerical simulat ion provide the informat ion of peak frequency response as the diffe rent levels of ball nut preload, ball screw torsional stiffness and table mass . the trend of increasing preload will indicate the abrupt peak change in fre quency response spectrum ana lysis in some mode shapes . this study provides an approach to investigate the dynamic frequency response of a ball screw drive system, which provides significant information fo r better control performance when precise motion control is concerned. keywor ds : ball nut preload, ball screw drive system, dynamic modeling 1. introduction precision co mputer nu merica l control (cnc) mach ines are widely used in modern industry for mass production. the ball screws a re widely applied in the linear actuators of machinery and equipment because of the high effic iency, less backlash, easy lubrication, and easy maintenance. since ba ll screw play a significant role in co nverting rotary motion into linear mot ion pre loading is effective to eliminate backlash and increase the stiffness of ball screw for precision motion concerns. the dynamic frequency responses of the feed drive system depends on the stiffness comb inations of the ball screw, ball nut, fixed support bearings, the fle xible coupler and the stiffness between the ball screw and the working table. such frequency response results fro m the a xial mode shapes and torsional mode shape of the ball screw drive system when it is actuated by servo motor. each mode shape affects and determines the motion control fre quency response bandwidth when the control speed is limited and becomes a critica l issue. the working table mass and the bolt stiffness between the machine bed base and attached ground floor also plays an important role for the frequency response when the vibration mode of the cnc machine is concerned with precision accuracy. as in intelligent control fie ld, the control signals that fed into the controlled plant are based on the feedback control error that should be learnable. there fore, so me control efforts [1-2] are focused the design of the bandwidth of the filter that can filter out the un -learnable errors contents and can be get back to control system for bettering control h istory. since the bandwidth of controlled system determines the motion speed response and its performance. the lu mped dynamic model derivation is significant in determin ing the bandwidth for motion control law that can be used for controller design. since such frequency contents of compensated error are suggested to be within the bandwidth when control signals are actuating. th is paper will derive the dynamic model of the ball screw feed drive system first and e xa mine the re lationship between the ball screw preload variation , ball screw torsional stiffness and effect of table mass. * corresponding aut hor, email: ychuang@cc.ncue.edu.t w advances in technology innovation , vol. 2, no. 2, 2017, pp. 29 33 30 copyright © taeti simu lation results will unveil the frequency response of the translational and torsional peak modes. nu me rical simulat ion shows stable convergence by using hybrid partic le swarm optimization of iterative lea rning control [2] on this developed ball screw drive system. 2. mathematical model and numerical simulation 2.1. dynamic model of the ball screw drive system to model the feed drive system, th is paper set differnt stiffness of the ball nut stiffness for diffe rent preload between the ba ll screw shaft and the ball nut. the presetting preload value can be deployed by inserting different ball size for single ball nut design or using disk spring that applied to the ball screw when double ball nut is the preference. fig. 1 shows the picture of the in-lab single-a xis feed drive p latform. to analyze the dynamic characteristic of the ball screw system under different preload and varying table mass, the feed drive system is modeled by a lumped parameter system shown in fig . 2. fig. 2 is the schematic illustration for the single-a xis ball screw feed drive system. in general, mechanica l systems have three passive linear components. the spring and the mass are energy-storage elements, while the viscous damper is the dissipated energy. both of the rotational and translation mechanical system modeled below are actuated by the servo motor torque, indicated as t. as the same derivation in [3], the overall stiffness of a ball screw feed drive system can be determined by the stiffness of the ball screw itself, which is comprised of the ball screw shaft, the ball nut, supporting bearings of the ball screw, and the stiffness between the ball screw and the working table. )( t bm gkmmqmmj      (1) 𝐽𝑏 × �̈�𝑚 + 𝑄𝑚 × �̇�𝑏 + 𝑅 × [𝐾𝑛 × (𝑅 × 𝜃𝑏 + 𝑋𝑏 − 𝑋𝑡)] = 𝐾𝑔 × (𝜃𝑚 − 𝜃𝑏 ) (2) 𝑀𝑏 × �̈�𝑏 + 𝐵𝑏 × (�̇�𝑏 − 0) + 𝐾𝑒 × (𝑋𝑏 − 0) + 𝐾𝑛 × (𝑅 × 𝜃𝑏 + 𝑋𝑏 − 𝑋𝑡 ) = 0 (3) txb x b rnk txtbtxtm )( )0(     (4) rearranging eqs (1)-(4), we have [ 𝑀𝑡 0 0 0 0 𝑀𝑏 0 0 0 0 𝐽𝑏 0 0 0 0 𝐽𝑚 ] × [ �̈�𝑡 �̈�𝑏 �̈�𝑏 �̈�𝑚] + [ 𝐵𝑡 0 0 0 0 𝐵𝑏 0 0 0 0 𝑄𝑏 0 0 0 0 𝑄𝑚 ] × [ �̇�𝑡 �̇�𝑏 �̇�𝑏 �̇�𝑚] + [ 𝐾𝑛 −𝐾𝑛 −𝐾𝑛𝑅 0 −𝐾𝑛 𝐾𝑒 + 𝐾𝑛 𝐾𝑛 𝑅 0 −𝐾𝑛 𝑅 𝐾𝑛 𝑅 𝐾𝑔 + 𝐾𝑛 𝑅 2 −𝐾𝑔 0 0 −𝐾𝑔 𝐾𝑔 ] × [ 𝑋𝑡 𝑋𝑏 𝜃𝑏 𝜃𝑚 ] = [ 0 0 0 𝑇 ] (5) where the stiffness matrix is different from [3]. fig. 1 the in-house single axis platform fig. 2 illustration for the schematic d iagra m of the single-a xis lu mped para meters ball screw drive system advances in technology innovation , vol. 2, no. 2, 2017, pp. 29 33 31 copyright © taeti 2.2. numerical simulation table 1 list the simu lated para meters and associated values used in eqs (1) -(4). the dynamic equation of the single-a xis feed drive system model with varied p reload and table mass can be expressed in a compact form: [𝑀]{�̈�} + [𝐶]{�̇�} + [𝐾]{𝑢} = 𝑓 (6) where [m], [c], and [k] a re the 4x4 square ma trices, refe rred as the mass (or the mo ment of inertia ), the viscous damping, and the stiffness matrices, respectively. {u} represents a four degree of freedom model. it consists with the xt , xb , θb , and θm for the displace ment of the working table, a xia l displace ment of the ball screw, rotation angle of the ball screw, and the rotation angle of the motor, respectively. table 1 important parameters of ball screw drive system parameters value working table mass (m t) 47.09kg ball screw mass (m b) 9kg inertia moment of the motor (jm) 4.45× 10−4kgm2 inertia moment of the ball screw (jb) 1.3× 10−3kgm2 equivalent axial stiffness of ball screw shaft (ke) 1.8663× 107n/m stiffness of the ball nut (kn) 2.3345× 108n/m torsional stiffness of the ball screw (kg) 3.49× 108n/m viscous damp ing coefficient of the guide way of the working table (bt) 10n s/m viscous damp ing coefficient of the supp orting bearin g of the ball screw (bb) 10n s/m rotational viscous damp ing coefficient of the motor (qm) 0n ms rotational viscous damp ing coefficient of the sup p ort bearing of ball screw (qb) 0n ms angle conversion axial disp lacement of the constant (r) 0.0025 motor torque (t) 1.8× 10−3nm/𝑠2 disp lacement of the working table (xt) state variable (m) axial disp lacement of the ball screw (xb) state variable (m) rotation angle of the motor (θm) state variable (rad) rotation angle of the ball screw (θb) state variable (rad) the homogeneous solution of eq s (5) represents the transient response of the lumped system whereas the forc ing function of applied motor torque renders the table positioning. ho mogeneous solution of the equation (2.1.5) results in four e igenvectors v1, v2, v3, and v4 associated with each e igenvalue ( λ = ω2 ) of 3.2124 × 104 , 3.3322 × 106 , 1.0699 × 1011 ,3.421 × 107(rad2 /s 2). v1={ −0.7299 −0.6836 0 0 }v2={ 0.1762 −0.9844 0 0 } v3 ={ 0 0 0.9482 −0.3176 }v4={ 0 0 0.7067 0.7076 } (7) the four eigenvectors corresponding to the eigen frequencies of 28.52 hz, 290.52 hz, 52059 hz, 930.9 hz w is calculated by 2% of the rated dynamic load. these eigenvectors represent the mode shapes of the ball screw feed drive system. the first three significant eigen fre quencies are re lated to the three resonant frequencies. the first and second modes are fro m the a xia l vibration. the third mode is fro m the torsional vibration. in eqs (7), the first mode of the working table and the ball scre w is moving in phase while the second mode is moving out of phase. fig. 3 shows the bode plot of the dynamic system based on diffe rent kn values. the preload variation is simulated fro m 2% , 4%, 6% , 8% to 10% of the rated dynamic loading of the ball screw. fig. 3 bode plot of the ball screw drive system based on the preload value of 2% (b lue), 4%(green), 6%(red), 8%(cyan) and 10%(purple) of the rated dynamic loading advances in technology innovation , vol. 2, no. 2, 2017, pp. 29 33 32 copyright © taeti en large ment of the first, second and third modes of the bode plot, the three frequencies are ranging fro m 30.5 hz to 30.8 hz, 294 hz to 380 hz and 52000 hz to 55000 hz respectively. it is obvious that the second translational mode will be affected more than the first mode when the ball nut preload is vary ing. the solution of the fourth mode is about 930 hz, the contribution of this mode is not significant even though the preload is varied. fig. 4 the enla rge ment of the first mode of the bode plot in fig. 3 based on different ball nut stiffness fig. 5 the enlarge ment o f the second mode of the bode plot in fig. 3 based on diffe rent ball but stiffness fig. 6 the en large ment of the third mode of bode plot in fig. 3 based on different ba ll nut stiffness as stated, the third eigenvector indicates the torsional vibration. since the servo motor drives the ball screw through a coupler providing damping and stiffness. the torsional stiffness of the ball screw is investigated by calculating kg by πdr 4g/32l . fig. 7 shows the variations of kg in the range of ± 10%. as shown, the third mode de monstrates large frequency shift with increasing the torsional stiffness of the ball screw. fig. 8 indicates the frequency shift is range fro m 50700 hz to 58000 hz, wh ile the first and second modes are not changed noticeably. fig. 7 bode plot of the ball screw torsional stiffness 100% ( blue), 105% (green), 110%(red), 95%(cyan), 90%( purple) fig. 8 the enla rge ment of the torsional mode of the bode plot in fig. 7 based on different ball screw torsional stiffness figs. 9-11 detail the effect of the table mass. the characteristic frequency shifts from the 24hz to 30.5hz when the table mass increases fro m 47.09 kg to 94.18 kg. as predicted, the table mass preserves the effect on the first bandwidth in the translational mode and some effect on the second mode. increasing table mass does not affect the bandwidth of the third to rsional mode. fig. 12 shows the numerical si mulation plot of the convergence error when the ball scre w drive system is controlled by hybrid particle swa rm optimization for the dynamic bandwidth tuning of an iterative learning control. advances in technology innovation , vol. 2, no. 2, 2017, pp. 29 33 33 copyright © taeti fig. 9 bode plot of the working table mass 47.09kg(blue), 70.63kg(green), 94.18kg(red) fig. 10 the en large ment of the first mode of the bode plot in fig. 9 based on different working table mass fig. 11 the enla rge ment of the second mode of the bode plot in fig. 9 based on different working table mass fig. 12 plot of a convergence error of the nume ric a l s imu lat ion by using hpso-ilc [2] for ball screw drive system 3. conclusions a lu mped dynamic model fo r describing diffe rent ball nut preload level, ba ll screw torsional stiffnesss and the table mass effects of the ball screw feed drive system is derived and numerically simulated. based on the different percent of the preload and table mass , the frequency spectrum analysis of the nu merical si mulation provides the limits and constraints of bandwidth tuning for motion control applications. the preload variation can be diagnosed by the peak frequency change and the magnitude of the peak frequency in a specific frequency range when the a xial mode or the torsional mode is excited. the derivation and numerical simulation results of the lumped dynamic mode l provides significant informat ion in determining the zero phase bandwidth tuning. application of a hybrid particle swarm optimization iterat ive learning control law deploys successfully when the frequency contents of the compensated error was constrained in every control actuation. acknowledgement this work was supported by most grant 104-2221-e-018-015 for which the authors are very much grateful. references [1] m. s. tsai, c. l. yen, and h. t . yau, “integration of an empirica l mode decompos ition algorith m with iterative learn ing control for h igh-precision machining,” ieee/asm e transaction on mechatronic, vol. 18, no. 3, pp. 878-886, 2013. [2] y. c. huang, y. w. su, and p. c. chuo, “iterative learn ing control bandwidth tuning using th e pa rt ic le s wa rm opt imiza t ion techn ique fo r h igh p rec is ion mot ion,” microsystem technologies. doi: 10.1007/ s00542-015-2649-6, 2015. [3] g. h. feng and y. l. pan, “investigation of ball screw p reload variation based on d ynamic modeling of a preload adjustable feed-drive system and spectrum analysis of b a ll-n uts sense d v ib rat ion s igna l s,” international journal of machine too ls & manufacture, vol. 52, no. 1, pp. 85-96, 2012.  advances in technology innovation, vol. 3, no. 1, 2018, pp. 36 42 projecting the mental model of social networking site usage chun-hui wu 1,* , yih-her yan 2 , kuen-ming shu 3 1 department of information management, national formosa university, yunlin, taiwan, roc. 2 department of electrical engineering, national formosa university, yunlin, taiwan, roc. 3 department of mechanical and computer-aided engineering, national formosa university, yunlin, taiwan, roc. received 02 june 2017; received in revised form 27 july 2017; accept ed 11 august 2017 abstract the growth of online social networking sites (sns) has created a new world of connection and communication for online users. sns usage has become an important part of people’s daily lives. this study aims to obtain new insights towards sns usage behaviour. based on participants’ mental models, it is hoped to make more clear exposition about their perceptions and experiences as well as to explore what factors affect their behaviour for using social networking sites. a blend of qualitative methodologies was adopted for data collectio n and analysis, including the zaltman metaphor elicitation technique (zmet) method, the laddering technique, and the means -end chain theory. the results of this study show that the most important values of using sns include its convenience, maintaining rel ationship, gaining relaxation, as well as reaching coherence. additionally, participants pointed out they cared about their online privacy issues very much and had found some potential dangers; however, they continued to use these sites because of the great benefits and enjoyment. keywords : social networking sites, projective technique, zaltman metaphor elicitation technique, mental model, means -end chain theory 1. introduction with the advancement and attractiveness of web 2.0 technology, online social networking sites (snss) such as facebook, google+, instagram, twitter, and line, have emerged as rapidly growing mechanisms that allow users to communicate with each other for sharing information, posting videos, pictures, comments, and messages at any time a nd from any places around the world. snss usage has become increasingly more influential on most internet users’ daily lives and radically have changed how they spend their time online. as snss continue to evolve at a breakneck speed, so does the usage gro wth on the respective platforms. according to the 2016 nielsen company’s report [1], the global average time spent per person on social networking sites is now nearly five and half hours per month. in addition, researchers from the pew research center’s in ternet found that nearly 80 percent of online adults used social networking sites, and almost 88% of teens and young adults were a member of at least one social network [2-3]. social networking sites support individuals to present themselves, articulate t heir social networks, and establish or maintain connections with others not only in forming but also maintaining relationships. in recent years, these sites have been the * corresponding author. e-m ail address: m elody @ nfu.edu.tw tel.: +886-5-6315741; fax: +886-5-6364127 advances in technology innovation, vol. 3, no. 1, 2018, pp. 36 42 copyright © taeti 37 fastest-expanding websites [4]. the growth of online social networks has created a new world of connection and communication for online users. however, it is not only changing how online users interact with one another, but also changing how business es touch with their consumers. the popularity of social networking is increasing very rap idly, and, therefore, businesses are turning to these platforms in order to promote their products and to reach potential consumers. this is new opportunities for busines ses to communicate with consumers for gaining much more transparency and ultimately to build profitable relationships with their consumers. it is predicted that us advertisers will reach $2.6 billion to place advertises on social networking sites by the end of 2012, up more than $1.2 billion from 2008 [5]. this progress illustrates the increasing influence of social networking sites in modern business environment. these phenomena have attracted much attention of practitioners and researchers to question what factors may influence the usage behavior on snss [6-7]. yet despite this interest, there seems to be very limited understanding of users’ mind map. essentially, a clear understanding of users’ thoughts and feelings could help businesses better allocate funds appropriately to effective factors and redesign or eliminate non -effective factors. by being aware of how mental models impact users’ understanding of snss can help businesses better apply marketing strategies that are consistent with users’ expectations and behaviour in social media [8]. this study, therefore, sets out to explore the following research questions: what are the mental models of online users toward the value of social networking sites? this study aimed to obtain new insights towards social networking sites usage behavior. based on participants’ mental models, it is hoped to make more clear exposition about their perceptions and experiences as well as to explore what factors may has an impact on their behavior for using social networkin g sites. an interpretative approach by means of the zaltman metaphor elicitation techniqu e (zmet), a projective technique, was conducted to a university context. 2. the projective technique the projective technique is a blend of three qualitative methodologies including the zmet method, the laddering technique, and the means -end chain theory. 2.1. the zaltman metaphor elicitation technique for understanding users’ actual thoughts toward social networking sites, the projective technique, zmet, was selected. the rationale for using zmet to collect and analyze data is that the zmet process of th inking about and searching for images is able to bring hidden, unconscious thoughts to the surface [9]. the zmet method provides opportunity for researchers to lo ok at the phenomena in more varied and deeper ways than is possible through other traditional qualitative methods. most qualitative research techniques, such as case study and focus groups, depend on verbal communication as a data-collection method. however, for more than 80% of people communication being nonverbal and non -linear [10], verbal-based interactions with subjects may result in an incomplete communication. additionally, people decision -making and behavior are guided by largely hidden experiences because probably 95% of all cognition is unconscious [11]. thus, the way in which thoughts occur may be very different from the way in which they are communicated. cognitive scientists claims that people think in images, not words; however lots of today’s qualitative research techniques rely on verbal-centric communication (i.e., literal, verbal language) as a data collection method [10, 12]. as subjects are better able to transmit their thoughts and feels in nonverbal terms, combining nonverbal images with verbal communication is able to generate a more meaningful message than totally relying on verbal communication. in this regard, zmet was designed to be a more precise research tool which can make up the deficiencies in current qualitative research methods . advances in technology innovation, vol. 3, no. 1, 2018, pp. 36 42 copyright © taeti 38 zmet is a hybrid methodology, which integrates the visual projection technique, in -depth personal interview, and a range of qualitative data-coding method, such as categorization, abstraction of categories, and comparison of instances within data to elicit the metaphors, constructs, and mental models that drive consumers’ thinking and beha vior. for improving qualitative research, zmet uses multidisciplinary ideas such as cognitive neuroscience, neurobiology, art critique, literary criticism, v isual anthropology, visual sociology, the philosophy of mind, art therapy, and psycholinguistics to combine knowledge from the social sciences, biological sciences, and the humanities. metaphors, photo analysis, and narrating are key concepts used in zmet, and each adds value to this research process. this technique achieves high validity since issues a nd structures emerge from the data collected by the respondents themselves. therefore, zmet has been employed in numerous academic studies. in addition, it has gained considerable interest and been used in over 20 countries around the world by the world's leading companies, such as at&t, coca-cola, motorola, american express, to explore consumer and organizational issues [9]. 2.2. the means-end chain theory and the laddering technique the means -end theory sustains that the way participants relate to topics can be represented by a hierarchical model of three interconnected levels: attributes, consequences, and values [13-14]. the zmet approach can offer a deeper and richer understanding of the important personal constructs elicited through laddering proces s [15]. the analysis of the means -end data comprised of three stages. at the first stage, the coding of sequences of attributes, consequences and values (the ladders) takes place in the results of zmet process. the second stage involves the development of meaningful categories by grouping together phrases with identical meanings. the identification of categories is by a-c-v phrases and key words that participants used during the zmet interviews and from concepts derived from the literature review. the s tudy will follow an iterative process of recoding data, splitting, combining categories and generating new or dropping existing categories, followed by an aggregation of codes for individual means –end chains across participants. finally, two hierarchical value maps (hvm) of teachers and students will be generated. the map consists of nodes, which stand for the most important attributes/consequences/values (conceptual meanings) and lines, which represent the linkages between the concepts. the map graphically sums up the information collected during the zmet interviews . 3. research methodology to address the research question, a qualitative research framework is presented in fig. 1. in order to capture the mental of online users who were using social networking sites, a blend of qualitative methodologies including the zmet method, means -end approach, and laddering process were used for data collection and analysis . online users were interviewed with the zmet approach and laddering process individual mental models consensus maps zmet laddering zmet laddering value consequence attribute value consequence attribute mec theory mec theory select participants using rpii questionnaire fig. 1 research framework advances in technology innovation, vol. 3, no. 1, 2018, pp. 36 42 copyright © taeti 39 this study had the zmet approach as theory basis, combining means -end approach and laddering process to analyze the high involvement people thoughts and feelings for snss usage. according to this method, we could realize the correlation about snss attributes, usage consequences, and ultimate va lues. qualitative-based procedures were selected for this study for three reasons. first, this study seeks to deeply better understand what is actually happening within social networking sites. secon d, it is hoped to obtain more in-depth information that may be difficult to convey quantitatively. third, zmet method provides opportunity for researchers to look at the phenomena in more varied and deeper ways than is possible through other traditiona l qualitative method. 3.1. selecting participants in this s tudy, the revised personal involvement inventory (rpii) questionnaire was utilized to help find a representative sample of facebook users, who meet the requirements for the research objective and have a high relevance. the rpii is illustrated in table 1. after qualifying for participation using rpii, a total of 12 facebook users are recruited to participate in our study. for understanding users’ actual thoughts toward social networking sites, the projective technique, zmet, was selected. the rationale for using zmet to collect and analyze data is that the zmet process of thinking about and searching for images is able to bring hidden, unconscious thoughts to the surface [9]. the zmet method provides opportunity for researchers to look at th e phenomena in more varied and deeper ways than is possible through other traditional qualitative methods. twelfth qualitative interviews based around the zmet approach were used for this study including 12 users who have highly participated in social networking sites were conducted. table 1 rpiirevised personal involvement inventory 1 unimportant : : : : : : important 2 boring : : : : : : interesting 3 irrelevant : : : : : : relevant 4 unexciting : : : : : : exciting 5 means nothing : : : : : : means a lot to me 6 unappealing : : : : : : appealing 7 mundane : : : : : : fascinating 8 worthless : : : : : : valuable 9 uninvolving : : : : : : involving 10 need : : : : : : not needed participants will be instructed to gather eight to ten pictures that represent their t houghts and feelings about social networking sites. these pictures could come from any source, including photographs, magazines, books, newspapers, or catalogues, and the instructions will be given seven to ten days prior to the interview. participants wer e contacted one or two days before their interview to confirm their understanding of the task and the whole processes, and each interview took approximately two hours in a quiet room and were recorded . storytelling 1 missed issues and images 2 sorting task 3 construct elicitation 4 opposite images 5 consensus maps 10 the vignette 9 the summary image 8 sensory images 7 most representative picture 6 fig. 2 the ten core steps in implementing zmet advances in technology innovation, vol. 3, no. 1, 2018, pp. 36 42 copyright © taeti 40 during the two-hour zmet interview, ten steps in the zmet method are summarized as follow and shown as fig. 2. (1) storytelling: provides participants with an opportunity to tell th eir stories(as shown in table 2) because human memory and communication is story-based. (2) missed issues and images: participants describe any issues for which they are unable to find a picture to obtain and explain their relevance. (3) sorting task: participants are asked to sort their pictures into meaningful pile s . (4) construct elicitation: the laddering technique is used to elicit basic constructs and their relationships. participants’ pictures serve as stimuli. (5) opposite images: respondent describes pictures that represent the opposite of the task. (6) most representative picture: participants indicate the picture that is most representative. (7) sensory images: descriptions are elicited of what does and does not describe the taste, touch, smell, sound, color, and emotion of the concept being explored. (8) the summary image: respondent, with scissors used to cut photos, creates a summary image. (9) the vignette: participant is asked to create a story or short imaginary video that communicates important issues related to the topic under consideration. (10) consensus maps: all the res pondents’ mental maps were merged to a consensus map. table 2 storytelling images briefly description image description privacy i do care about online privacy issues very much. there are various threats that people may encounter on social media. for avoiding dangers, great skills and caution are required for using social networking sites. otherwise, it is easy to observe or monitor other users’ emotional state and interactive statuses. addiction similar to drug addiction, many users are addicted or obsessed with facebook and have difficulty logging off even after they have been on for hours. convenience sns is like a convenience store that is available 24 hours a day and offers everything from hot meals to package delivery. for example, facebook is convenient because we can log into it anywhere, on a pc or on a mobile phone. it is helpful and convenient in ways that it helps users find friends, lets our current friends suggest some people we may know but have not already added. 3.2. data analysis participants ’ feelings, thoughts, and behaviors were analyzed based on means -end chain theory. all interviews were recorded and later transcribed to enable empirical content analysis for achieving means -end chains . each means -end chain links the relationships between attributes, consequences, and values (a -c-v), it can also analyze the emotion product attributes brings to the consumers, what overall value thinking [16]. the link is shown in fig. 3. advances in technology innovation, vol. 3, no. 1, 2018, pp. 36 42 copyright © taeti 41 attributes (means) consequence (means)(end) values (end) fig. 3 m eans-end chain connects relations between attributes/consequence/values a means -end chain links attributes of the social network sites, consequences of these attributes to participants , and the personal values that the consequences reinforce. next, each participant’s transcripts were read to discover the relationships between attributes, consequences, and values. a participant’s sequence of attributes, consequences, and values (a -c-v) is called a means -end chain and represents a perceptual orientation of decision criteria. finally, a serious of a-c-v chains were merged together to derive individual mental models and the aggregate consensus maps of all participants . a consensus map represents the main concepts identified by all participants and the linkages between the concepts as reflected in their interviews. the consensus map (see fig. 4) in this study resulted from the linkages between system attributes, usage consequences, and personal values. this consensus map presented over 80% of all constructs mentioned by each one participant. sharing consequences values attributes newsfeed maintaining relationship perceived easy of use recording and storing reconnecting with friends interactive with others gaining relaxation convenience reaching coherence the wall fb platform blog games chatroom activity groupssharing fig. 4 the mental model towards social networking sites 4. findings and conclusions the zmet method provides a good way for understanding consumers’ cognition and projecting consumers’ behaviour. therefor, research data were collected with the zmet method to understand users ’ insights toward using social networking sites more thoroughly and more deeply. the analytical results show that the majority of respondents in this study indicated the most important value of social networking sites includes its convenience , maintaining relationship, gaining relaxation, as well as reaching coherence. this study’s findings were consistent with p revious research of of social networking sites [17-19]. additionally, respondents pointed out they cared about their online privacy issues very much and had found some potential dangers. however, they still continued to use social networking sites. the main reason is the perceived benefit and enjoyment providing by social networking sites are greater than perceived potential dangers for most respondents. the benefits of social networking service usage include formal educational outcomes, informal education and learning, creativity, individual identit y and self-expression, strengthening social relationships, belonging and collective identity, building and strengthening communities, civic and political participation, self-efficacy and wellbeing. advances in technology innovation, vol. 3, no. 1, 2018, pp. 36 42 copyright © taeti 42 references [1] sean casey, “the 2016 nielsen social media report,” http://www.nielsen.com/us/en/insights/reports/2017/2016-nielsen-so cial-media-report.html, 2017. [2] pew research center, “79% of online adults (68% of all americans) use facebook,” http://www.pewinternet.org/2016/11/1 1/social-media-update-2016/pi_2016-11-11_social-media-update_0-02/. [3] s. jones and s. fox, “generations online in 2009,” http://www.pewinternet.org/’/media//files/reports/2009/pip_generatio ns_2009.pdf, 2009. [4] k. c. laudon and j. p. laudon, management information systems: managing the digital firm, new jersey: prentice-hall, 2012. [5] a. lenhart, k. purcell, a. smith, and k. zickuhr, “social media and young adults ,” washington, dc: pew internet and american life project, http://pewinternet.org/reports/2010/social-media-and-young-adults.aspx, 2010. [6] the 2010 nielsen company report, “global audience spends two hours more a month on social networks than last yea r,” http://blog.nielsen.com/nielsenwire/global/global-audience-spends -two-hours -more-a-month-on-social-networks -than-la st-year/>, 2010. [7] j. hahn, “social networking ad spending to reach $2.6 billion,” http://www.emarketer.com/article.aspx?id=1006278/. [8] p. rydén, t. ringberg, and r. wilke, “an investigation of the mental models of social media in the minds of managers ,” the 43rd emac annual conf. 2014, 2014, pp. 1-9. [9] g. zaltman and r. coulter, “seeing the voice of the customer: metaphor based advertising research ,” journal of the marketing research society, vol. 35, no. 4, pp. 35-51, 1995. [10] g. catchings -castello, “the zmet alternative,” marketing research, vol. 12, no. 2, p. 6, 2000. [11] g. zaltman, “eliciting mental models through imagery 22,” the languages of the brain, vol. 6, pp. 363, 2002. [12] a. g. woodside, “advancing means end chains by incorporating heider’s balance theory and fournier's consumer-brand relationship typology,” psychology & marketing, vol. 21, no. 4, pp. 279-294, march 2004. [13] m. vriens and f. t. hofstede, “linking attributes, benefits, and consumer values,” marketing research, vol. 12, no. 3, pp. 4-10, 2000. [14] r. b. woodruff and s. gardial, know your customer: new approaches to understanding customer value and satisfaction. wiley, 1996. [15] g. l. christensen and j. c. olson, “mapping consumers’ mental models with zmet,” psychology and marketing, vol. 19, no. 6, pp. 477-502, may 2002. [16] t. j. reynolds and j. gutman, “laddering theory, method, analysis, and interpretation ,” journal of advertising research, vol. 28, no. 1, pp. 11-31, march 1988. [17] d. boyd and n. ellison, “social network sites: definition, history, and scholarship,” journal of computer-mediated communication, vol. 13, no. 4, pp. 210-230, 2007. [18] r. junco, “the relationship between frequency of facebook use, participation in facebook activities, and student engagement,” computers & education, vol. 58, no. 1, pp. 162-171, january 2012. [19] o. kwon and y. wen, “an empirical study of the factors affecting social network service use ,” computers in human behavior, vol. 26, no. 2, pp. 254-263, march 2010.  advances in technology innovation, vol. 3, no. 3, 2018, pp. 118 125 a gi proposal to display ecg digital signals wirelessly real-time transmitted onto a remote pc marius-corneliu rosu 1,* , hamed yassin-kassab 1 , adriana iliesiu 2 , serban georgica obreja 3 1 department of applied electronics and information engineering, univers ity politehnica of bucharest, romania. 2 department of cardiology, university carol davil of bucharest, romania. 3 department of telecommunications , university politehnica of bucharest, romania. received 21 july 2017; received in revised form 26 sept ember 2017; accept ed 17 december 2017 abstract the sensors, as wireless communication s ystem, comply the 7-layer model open systems interconnection (osi). in this paper, a point-to-point transmission model was used. the ecg signal is transmitted from the router se nsor (rs) to an end coordinator node (cn) plugged -in to the laptop via usb port; rs acquires ecg signal in analogical mode, and is also responsible with sampling, quantization and sending it wirelessly direct to cn. the distance between rs and cn is a single-hop transmission, and does not exceed the range of the xbees2pro transceivers . the communication protocol is zigbee. remote viewing of the transmitted signal is performed on a graphical interface (gi) written under matlab, after the signal has been digitized; the choice of matlab was motivated by future developments. particular aspects will be highlighted, so that the reader to be edified about the results obtained during laboratory experiments. recording demonstrate that the purpose exposed in title has been reached: direct link in real-time was established, and the digital ecg signal received is reconstituted accurately on matlab gi; signal received on laptop is compared with the analog signal displayed on oscilloscope . keywords: gi, real-time ecg, wireless -transmission, zigbee, matlab 1. introduction remote viewing of a digital signal is performed on a graphical interface (gi), written in a programming language, enabling the entire signal overview and detailed views. visualization by gi, it showed to be faster than standard methods. different levels of visualization and fast navigation in bio-signals are essential conditions for a gi. many expectations are demanded from specialists when is used a software for more tasks . the truth sometimes looks quite different: the software does not work as is expected, and because of some bugs unpleasant consequences and unwanted effects will occur . gi is usually the main interface; the only one can say between the user and the given system. traditionally testing with gi is done manual involving money and time; that's why the results are often unreliable. so better for testing are specialized software which can avoid these problems and time period to enter on market can be shorted. thus, scripting language for gi should allow rapid testing and debugging as well as the possibility to check gi with dedicated programs . * corresponding author. e-m ail address: m arius-corneliu.rosu@ sdettib.pub.ro advances in technology innovation , vol. 3, no. 3, 2018, pp. 118 125 copyright © taeti 119 1.1. graphical interface design issues engineering for gi becomes a subject in the software industry starting with 90s. gi had replaced the text -based interface (tbi) with one based on graphics, and for users it becomes more intuitive and easier to work with . in [1], it is shown the need to transpose the tbi to the gi platform in an automatic way. this must include the gi builder and compiler code for event -driven language c/c++. applications that use gi are created based on processes requirements, and during development these may change. typically, gi are modeled as state machines defined by windows, widgets, widgets parameters and values; usually shaping gi starts from implementing requirements which a system must meet. long-term bio-signals' acquisitions are the main sources of informations about the patient's condition . hence, a broad set of issues into a gi design must be harmonized. modeling was also an important obstacle in the tests performed on the basis of patterns because they were not used more intensively in the industry. it has been shown, however, that in well -defined applications, models generation methods can be used with good results . consequently, the automatic generation of models has become widely used in designers thinking . various techniques used to generate models of gi are synthesized in table1. for a more thorough documentation, the reader is guided to consult the reference [2]. in table1, most techniques require human intervention; the effective implemented solution largely depends on the degree of human intervention. stages of drafting, in short, are the following: (1) design and modeling of the main components for optimal response of the monitored system (ms). (2) establishing of minimum parameters required to cover the ms. (3) generate automatically if possible with a predetermined method (e.g. supervised learning) the pattern of algorithms and test parameters. (4) testing the monitored system by running the generated patterns and parameters. to explore the events, gi inputs must be stimulated while outputs return the results; also, the created gi for the ms, permit to check the implementation (software or hardware) for the system requirements . 2. signal consideration various technologies are needed to develop an effective gi as discussed in the previous chapter; design and development require a multidisciplinary approach. long -term acquisitions of bio-signals are important; they constitute the main source of information on the patients' condition. visualization tools are also important: they allow the inspection of complex bio -signals in a fast manner and friendly use. in terms of scope, this paper aims to depict the possibility of using t he programs dedicated mainly to signal processing (matlab this case). in terms of variety, bio -signs are impossible to classify rigorously: the only certainty is that these are with very low values (the useful ecg signal, between -2 and +2mv) , while interferences, artifacts, and noises, distort them, having a major influence (e.g. electrods offset level between -300 and +300mv). a bio-signals classification is not universally valid. in fig. 1, we drew a classification attempt, and fig. 2 illustrates once again that it is practically unachievable for a strict and unique classification. from these, ecg signal is graph transposition of heart electrical activity, and the mechanical activity of the heart is somewhat regular with the exception of being patients with cardiovascular disease (cvd), so ecg may be considered a quasi-periodic waveforms . particularities can be summarized as follows: advances in technology innovation , vol. 3, no. 3, 2018, pp. 118 125 copyright © taeti 120 (1) ecg is ideally for electronic reporting (2) ecg signals are weak; (3) bad influences have interferences, artifacts, an d noises; (4) bandwidth is considered depending on the intended purpose, monitoring or diagnosis (5) some areas of the ecg signal frequencies must be preserved ; fig. 1 bio-signals wave shapes attempt classification fig. 2 second approach for bio-signal classification (after [3]) note: in figures are indicated, heart rate fc, respiratory rate fr, and other additional information advances in technology innovation , vol. 3, no. 3, 2018, pp. 118 125 copyright © taeti 121 table 1 techniques for generating gi models t ask. gi mode l s technique name description remarks 1 manual modeling m odeling the test sy stem; defining the required test area; control of entry , while on exit is observed the events using test scrip ts; the model rep resents the real situation and test scrip ts that generates him; sy stem testing by the so-called comp liance testing requirement sp ecifications, interface, and so on, should accurately describe the behavior of the sy stem; requirements are shown in natural language by texts and diagrams; just p ersons with analy tical abilities can create a formal and structured model 2 automatic modeling the initial objectives of model are reduced; enables imp lementation of the comp liance testing requirements; the model is basically a reverse engineering task; imp lementation is the source for redesign of the work; can reveal bugs that cause accidents in imp lementations 3 semi-automatic modeling m odeling techniques, p erformed most p art automatically are assisted manually by a p erson during the p rocess; revised, corrected and extended manually after generating the initial model. 4 reverse engineering extract design artifacts, and derive abstractions less dep endent of imp lementation. there are two ap p roaches: static, analy zing source code or binary rep resentation, without running the sy stem, while dynamic, examines the external behavior of the sy stem by running it . 5 static analy sis built for imp ortant elements of source code figure behavioral and structural; several issues make the static analy sis a comp licated task: inheritance, p oly morp hism, and dy namic binding, make difficult to know which methods are going to be executed; that's why for this, the method was not successful in generating p atterns for gi testing p urp oses. 6 dy namic source code instrumentation analy sis during the execution of a sy stem, observations are made by instrumentation instructions imp lemented as source code, which can be on gi controlled and observed source code is the best way to get information, because the level of abstraction and sources of information are controlled by reverse engineering; 7 dy namic by te code analy sis it is used to avoid rep eated op erations less invasive than source code can be used in java 8 dy namic gi apis the technique is used to collect information on gi status, and to control the gi from one state to another the model was created for understanding the behavior of the original gi, in aiming to build the correct shell; it is used an indep endent source code for interaction. 9 m anual gi exp loration techniques exp lore manually by the user the imp lemented gi, while the background p rocess record user actions. all cap ture or p lay back gi tests tools, utilizes manual methods of exp loration; m odel is used to generate test cases able to find errors. 10 fully automated gi exp loration techniques generation of gi testing models; gi models are obtained as well as runtime behavior automatically . the p rocess works in rip p ing mode, starting with first gi window and continuing by op ening all other sub-windows; the behavior of the structure and execution of gi are extracted. 11 user-assisted gi automated exp loration techniques it is a mix between automatic exp loration and user-assisted manual exp loration. it consists of three p hases: 1. extracting structural information about each gi window; 2. interaction with gi; 3. takeover of behavioral information through automatic p rocess. the model is used for abstract cases. 12 feedback gi refinement all the p rocesses described above can be imp roved by this technique; this technique requires that an initial test case or suite to be created, manually or automatically , and to be run on software. various sources of feedback are p ossible, e.g. the code coverage rep ort. feedback serves to enhance the gi model, allowing for better test coverage and test generation with lower costs of execution 3. motivation and previous achievements the increasing of hardware applications in medicine had generated the issues of visualization into a manner that can be understood by the human brain; signal analysis in medicine has now become a method of interpretation and research since it can offer feedback on the patient's condition or state of the experiment by processing it . so, viewing and processing techniques advances in technology innovation , vol. 3, no. 3, 2018, pp. 118 125 copyright © taeti 122 involve researchers or clinicians direct in selecting interesting parts of analyzed bio -signal. the acquisition on a longer period of time is a method that allows continuous monitoring, and was proposed in [4]; the method has, however, barriers consisting of vast amounts of end point with double column. it becomes important for remote monitoring continuous real-time view. design of visualization tools is related to a correct choice of data information storage structure; signals must be recorded in a format that gives possibility for posteriori data accessing. several standards for storing biological data and inter-communication have been developed over time. a standard format that allows saving the data on multiple channels is european data format (edf), and it is described in [5, 6]. division of health science and technology from massachusetts institute of technology (mit) developed research resources for an online investigation that can be applied to complex biological signals [7]; it consists of the following main components: (1) physiobank is the online database of physiologic signals; (2) physiotoolk it, is the bio-signal analysis tools: the ecg data stored in edf format can be viewed in the edf browser, an open source tool that allows signal processing, such as filtering, power spectrum, or heart rhythm; (3) physionet the web forum [8]; the gi's requirments for viewing and process ing of the long term bio-signals are not easy to be achieved. other ideas in visualizing large data sets, but not for bio -signals, is web mapping services consistin g in a multilevel visualization of an very large datasets; another area requiring long -term visualization are algorithms for data mining [9]. displayed solutions described above refer to portions of signal finite in time. for bio -signs with virtually infinite duration, ecg in this paper, visualization imposes a new concept of gi' s that have to display in real-time the monitored signal wirelessly transmitted. 4. the concept biological signals can be acquired through various methods. as was mentioned, visualization, analysis , and processing, are essential in long term monitoring. in this paper, we intend to see a new possibility for building a gi by using other resources than the classic ones . 4.1. data file formats it should be recalled once again that in long term monitoring, the amount of received information is enormous. also, the data format in a standardized mode will facilitate cooperation and exchange of data between centers; a brief list of formats is described in further:  *.txt: text, known by users as computer files, with a wide use, those are basic file system; can be accessed and modifie d thru text editors. disadvantage: slow access in data;  *.edf: edf is just a basic format to archive and exchange bio -signals; sampling frequencies and physical sizes are not imposed. disadvantage: implementation of edf unit is time consuming task ;  *.dat: physiobank is characterized by the archive of physiologic signals recorded digital. each physio -bank is possible to have more than one record divided in three files: *.hea file header information; name or url short text file describing signals; *.dat – binary signal file containing digital samples;  *.h5: hdf5 hierarchical data format 5 for storing and managing data with simple structure that include: datasets and groups;  *.mat: matlab format, the most known in biomedical engineering area; data are sav ed in binary. even though is not open source, it is a powerful tool; since ecg acquisition we made is under ma tlab format, the idea was to try to realize a gi also under matlab tool. advances in technology innovation , vol. 3, no. 3, 2018, pp. 118 125 copyright © taeti 123 4.2. basis idea as the ‘tmtools’ command in matlab is used to establish the communication via usb port between an oscilloscope and a pc, we began to wonder if a similar issue can be applied to a cn plugged by usb onto a remote pc, which acts as server and workstation for the received signal. the only requirement self-imposed is the ability to view in real-time the signal received which is transmitted in digital domain between a router and a coordinator node; the last one is connected to the remote pc via a usb port. it is known that matlab provides programs and basic commands that can be modified for the aimed purpose. thus, if the ecg samples from oscilloscope acquisition were made possible via the usb port in matlab, the same thing was wanted and in reverse operation, to display the signal wirelessly received by the coordinator, in a gi written under matlab; the choice was motivated by flexibility and adaptability to future developments. for this, however, we are required the following checks : (1) if the serial device send data in "lines", through a line terminator character; (2) how many bytes is expected to be read each time by the serial port; verification of the above can be done by following the instructions set forth in [10]. since most usb drivers include the ability to act as a "virtual serial port" or "virtual com port" (vcp), this feature can be used to establish communication with the matlab interface. in windows, drivers for vcp are sometimes a separate feature or require additional installation steps ; once installed, serialdatastream function is responsible for receiving and processing a continuous stream of data from a device connected to a usb. some aspects must be specified so that the reader to be cleared up about how the connectivity can be established between coordinator and pc via usb interface: (1) first, it should be remembered that due to micro -processor adc module in sensor node, wireless transmission it will be made in the digital domain; (2) in matlab must be connected an object with fopen(obj) command, so that serial communication with coordinator via usb to be possible; (3) obj from command fopen(obj) is defined at least by the parameters: '#com po rt', 'baudrate', and 'databits'; (4) before running the script should be checked which ‘#com port' was allocated to coordinator, otherwise matlab will not know how to communicate, displaying an error in command window; further, writing in matlab of gi is given by the instructions of reference [11]. 4.3. wireless test results setting and configuration of sensors are important; sensor dedicated to ecg signal was designed for zigbee protocols communications . the channel access technology method was preferred to connect the wireless devices “point to point”; mac protocols thru data-link layer are responsible to allocate resources for this task (osi level 2). fig. 3 shows that used transceivers were wireless ly connected, and fig. 4 shows the coordinator hardware connected directly to the laptop via usb port; the appropriate mac address also can be readable. settings were made also in the xctu configuration software . a direct link has been established in real time between sensors ; fig.5 represents snapshots after recording during laboratory experiments , and the received signal is reconstituted accurately on gi written in matlab. on router at mode setting was made, while for coordinator api mode setting was chosen ; additional information and a quick reference guide for xbees2pro transceivers used can be downloaded from [12]. advances in technology innovation , vol. 3, no. 3, 2018, pp. 118 125 copyright © taeti 124 on the other hand, energy consumption must be well controlled; controllers and radios devices ar e the main consumers while memory and sensors are in lower position . those demands come because of the generic implemented solutions limited efficiency, and without taking into account the requirements of limited energy resources, wearable systems cannot b e developed. fig. 3 xbees2pro link connection fig. 4 coordinator node plugged into laptop fig. 5 photos of laptop screen; ecg signal can be distinguished 5. conclusions and future work considering the problems with the viewing and processing bio -signals in the long term, and the importance of ecg signal in healthcare. this paper has presented a gi proposal to display ecg digital signal wirelessly transmitted onto a remote pc using matlab toolbox. some requirements should be considered so that the communication to be established successfully via usb; they were detailed in chapter 4. advances in technology innovation , vol. 3, no. 3, 2018, pp. 118 125 copyright © taeti 125 further, writing and developing of gi will be made depending on the complexity and number of determinations that are intended to be made in real time. so, viewing the transmitted digital signal must be performed on a gi which must be friendly and easy to use. at the same time, the end user must have the opportunity of immediate determinations of the main characteristics of the signal (e.g. rr distance in ecg). also, future gi improvements should enable the designer to carry out immediately at minimal cost. channel access technology method was preferred for point to point wireless connection; data -link layer thru mac protocols has the responsibility to allocate resources for this task (osi level 2). thanks to flexibility and adaptability, programs and functions for an automatic processing routine in parameters of interest in ecg signal will be added. this paper discusses traditional methods for gi and proposes a new approach of a program dedicated more to processing and less for interfaces. the challenges in this direction remain undetermined, the rate of renewal of wireless technologies b eing very dynamics; tiny devices are able to improve the daily life and to ensure a bette r monitoring of patients . references [1] g. antoniol, r. fiutem, e. merlo, and p. tonella, “application and user interface migration from basic to visual c++,” proc. 11th international conference on software maintenance, 1995, pp. 76-85. [2] a. kull, “automatic gui model generation: state of the art,” proc. ieee 23rd international symp. software reliability engineering workshops, november 2012, pp. 207-212. [3] e. kaniusas, “biomedical signals and s ensors i,” chp. 1 biological and medical physics, biomedical engineerin g, berlin: springer-verlag, 2012. [4] h. kayyali, s. weimer, c. frederick, c. martin, d. basa, j. juguilon, and f. jugilioni, “remotely attended home monitoring of sleep disorders ,” telemedicine and e-health, vol. 14, no. 4, pp. 371-374, may 2008. [5] “edf-european data format,” http://www.edfplus.info/. [6] b. kemp and j. olivan, “european data format “plus” (edf+), an edf alike standard format for the exchange of physiological data,” clinical neurophysiology, vol. 114, no. 9, pp. 1755-1761, september 2003. [7] g. b. moody, r. g. mark, and a. l. goldberger, “physionet: a web-based resource for the study of physiologic signals ,” ieee engineering in medicine and biology magazine, vol. 20, no. 3, pp. 70-75, may 2002. [8] “physionet the research resource for complex physiologic signals ,” http://www.physionet.org/. [9] n. kumar, n. lolla, e. keogh, s. lonardi, and c. a. ratanamahatana. “time-series bitmaps: a practical visualization tool for working with large time series databases ,” proc. of siam international conference on data mining, april 2005, pp. 531-535. [10] “warning: unsuccessful read: matching failure in format ,” http://stackoverflow.com/questions/35671418/warning -unsucce ssful-read-matching-failure-in-format, february 29, 2016. [11] m. a. hopcroft, “serialdatastream,” https://www.mathworks.com/matlabcentral/fileexchange/31958-serialdatastream/cont ent/serialdatastream.m, march 20, 2012. [12] “xbee s2 quick reference guide,” https://www.tunnelsup.com/xbee-guide/, november 30, 2012.  advances in technology innovation, vol. 3, no. 3, 2018, pp. 133 140 flowfield analysis of a pneumatic solenoid valve sheam-chyun lin 1,* , yu-ming lin 1 , yu-song haung 1 , cheng-liang yao 2 , bo-syuan jian 2 1 department of mechanical engineering, national taiwan university of science and technology, taipei, taiwan 2 metal industries research & development centre, kaohsiung, taiwan , r.o.c. received 21 july 2017; received in revised from 24 july 2017; accept ed 17 sept ember 2017 abstract pneumatic solenoid valve has been widely used in the vehicle control systems for meeting the rapid -reaction demand triggered by the dynamic conditions encountered during the driv ing course of vehic le. for ensuring the safety of human being, the reliable and effect ive solenoid valve is in great de mand to shorten the reaction time and thus becomes the topic of this research. th is nume rica l study chooses a comme rcia l 3/ 2-way solenoid valve as the reference valve for analys ing its performance. at first, cfd software fluent is adopted to simu late the flow field associated with the valve configurat ion. then, the co mprehensive flo w v isualization is imp le mented to identify the locations of adverse flow patterns. accordingly, it is found that a high-pressure region exists in the zone between the nozzle e xit and the top of the iron core. thereafter, the no zzle dia meter and the distance between nozzle and spool are identified as the important design parameters for improving the pressure response characteristics of valve. in conclusion, this work establishes a rigorous and systematic cfd sche me to evaluate the performance of pneu matic solenoid valve. ke ywor ds: pneumatic solenoid valve, co mpressible nume rica l simulat ion, transient characteristics, pressure-rising process 1. introduction pneumatic system has been used extensively in many areas of industrial applicat ions, such as automation control, medica l instruments, and control unit of the vehicle. it is essential to choose an appropriate valve as the interface to electronic controls for performing the required adjustments or actions in accordance to its function design. among the automobile safety system, an accurate and reliab le pneumatic solenoid valve with a short response time is crit ical in the anti-loc k braking system (known as abs), which offers imp roved vehicle control and decreases stopping distance. therefore, the understanding on the flow patterns inside the solenoid valve is in great demand to shorten the reaction time , and thus becomes the goal of this research. usually, servo valve and on–off valve are two types of electro -pneumatic valves used in controlling the pneumatic actuator. the expensive servo valves with comple x structure are used to achieve the high linear control accuracy. on the other hand, due to the low cost, compact size , and simple structure, the fast-switching on–off valves have received considerable attention in vehicle industry and researchers [1-4]. in 2006, topçu et al. [5] develops the simple, ine xpensive fast-switching valve for applications of pneumatic position control. four prototype valves have been built and the basic mode of operation confirmed. in addit ion, the switching characteristics of the on–off valve with 2/2-way function has been investigated both theoretically and experimentally . simu lated results of the valves dynamics were in agree ment with the experimental results, and thus the validity of the proposed mathematical model was confirmed. * corresponding author. e-mail address: sclynn@mail.ntust.edu.tw tel.: +886-2-7376453; fax: +886-2-27376460 advances in technology innovation, vol. 3, no. 3, 2018, pp. 133 140 copyright © taeti 134 as expected, the magnetic fie ld has a dominant influence on the response characteristics of a solenoid valve. thus, an optima l design of the magnetic fie ld of a high -speed response solenoid valve is executed by tao et al. [6]. they used the finite ele ment method to optimize the solenoid va lve for achiev ing la rger magnetic force and low power v ia the changes on parameters and materials. later, wang et al. [7] investigated influences of cross -sectional area of the iron core and ampere turn on the static electro magnetic characteristics through numerical simu lation. they found that the ampere turn has great effect on electro magnetic force for the magnetic saturation phenomenon. besides , the simu lation method is validated by the experiment. as regards the flow field analysis, several cfd reports [8-12] are focused on analyzing the flow fie ld inside the valve. peng et al. [8] adopted the comme rcia l cfd software fluent to establish cfd mode l for simulat ing the inner flo w fie ld of a servo valve when the valve spool is located in certain positions. also, several improve ments in the core shape of valve are raised and evaluated via the established numerical mode l. later, ma and sun [9] used cfd software fluent to simulate the static and dynamic flow fields of an e lectro magnetic valve. the function between the mass flow rate and the drop of pressure through the electromagnetic valve was obtained from the results of static flow fie ld simulat ion. fro m the nume rica l simu lation of the unsteady flow, the valve closed procedure was calculated. the results indicate that the inner flow field nu merical simulation of the valve by fluent can reflect its working procedure. more recently, in 2016, liu et al. [10] conducted a research on a solenoid valve used in the hydraulic control system. based on the conditions occurring in the operation of the hydraulic drive system, the therma l field o f the head is analy zed b y ansys. it is illustrated that the solenoid valve has a good performance under high te mperature condition. they presented a method to monitor the performance of the valve while the reactor is working. fro m the previous papers, it is demonstrated that numerical simu lation can be adopted as a reliable and useful tool in the valve design. later, liu et al. [11] p resented a nonlinear dynamic model of a la rge flo w solenoid with the mult i-physics dynamic simulat ion software called simulat ionx. the dynamic characteristics of this solenoid valve are analyzed and validated by comparing the test and cfd results. in fact, co mputational flu id dynamics (cfd) is increasingly being used as a reliable method for determin ing performance characteristics of other valve. fa rre ll et al. [12] e xecuted a series of cfd investigation on characterizing the opening and c losing of check valves. they adopted cfx which is a part o f the ansys suite of finite e le ment progra ms, to predict and characterize the performances of swing check and lift chec k valves. also, the good agreements are found via co mparing the available test data of the modeled valves with the numerical results . fig. 1 methodology of this numerical simulation over a pneumatic solenoid valve advances in technology innovation, vol. 3, no. 3, 2018, pp. 133 140 copyright © taeti 135 therefore, th is computational flu id dynamics (cfd) study chooses a commerc ial 3/2-way solenoid valve, wh ich is used e xtensively in vehicle control system, to e xa mine its dynamic performance. at first, flo w-fie ld simu lation associated with the valve construction is e xecuted by using the commerc ial cfd code ansys fluent. then, the co mprehensive flow visualization is imple mented to identify the locations of adverse flow patterns, which a re critica l for proposing the improving alternatives. also, the flowchart for this valve research is illustrated in fig. 1 2. working principle of charging process and description of physical model 2.1. work ing principle of charging process fig. 2 shows the overall valve system, which is co mposed of the solenoid valve, p iston connector, connecting duct, and the outlet vessel. it is necessary to describe the working principle for the pressure -rise process of solenoid valve in brief. for increasing pressure to its setting value for activating the abs system, a h igh-pressure (10.1 bar) a ir source is connected to the nozzle inside the top portion of valve (see fig. 3). thus, this big pressure difference generates a chocking situation (sonic speed at the nozzle e xit ) and an in flo w with the constant mass flow rate at the beginning of this filling process. however, after the vessel pressure reaches a fixed value (near 52.8 % of the source pressure), the flow rate of this inlet airstream beco mes smaller with a rising vessel pressure. finally, this process ends when vessel pressure is equivalent to the pressure source . fig. 2 overall system of the pneumatic solenoid valve fig. 3 open mode of the pneumatic solenoid valve clearly, this charging process is an unsteady flow undergoing a significant pressure variation; thus the transient simu lation and compressible assumption are needed to realize the comp licate physical phenomena. ho wever, it is known that the unsteady cfd s imu lation not only needs a high -performance server with huge me mory resource, but also takes a much longer cpu time to obtain the result. thus, the steady simu lation is carried out on the complete valve system with an opened vessel end in this study for evaluating the flow patterns inside the geometry. therea fter, a co mprehensive analysis o f the flow pattern inside of the valve system is executed via the simulation outcomes for finding out the modification possibilities. advances in technology innovation, vol. 3, no. 3, 2018, pp. 133 140 copyright © taeti 136 2.2. description of physical model the actual valve configuration is quiet co mple x and difficult to establish a nume rical model fo r cfd simu lation. thus, proper simplifications on the cad file a re needed to attain an effective nu merica l mode l, wh ich is div ided into several portions with d ifferent grid densities as indicated in fig. 4(a ). genera lly, to capture the actual physical phenomenon precisely, the intense grid distribution is placed on regions with an abrupt property variat ion on veloc ity, pressure, or direct ion. fo r the va lve cons idered here, as illustrated in fig. 4(b ), these locations include the nozzle, the s mall c learance between nozzle e xit and the armature, e xpansion part in the connector, and junction between the connecting pipe and the vessel. the total grid nu mber of this numerical model is 8.6 million. (a) overall system of the valve (b) connecting duct to the vessel fig. 4 grid system of the pressure-increasing process for a pneumatic solenoid valve 3. numerical scheme this study simulates the comple x flow patterns inside the electromagnetic valve by utilizing the commerc ia l computational fluid dyna mics (cfd) software fluent [ 13] to solve the fully three-dimensional co mpressible navier -stokes equations with the standard k-ε turbulence model. also, the se mi-imp lic it method for pressure-linked equations (sim ple) [14] is imple mented to solve the velocity and pressure coupling calculation for steady cases. hence, the flo w v isualization inside the valve can be performed and observed carefully to locate the reversed flow p atterns. in this work, several appropriate assumptions and boundary conditions were made to simulate the actual flo w patterns inside a ceiling fan. they are described as : (1) inlet boundary condition: the inlet boundary condition of valve is set as pabs=11 bar for serving as an extremely high-pressure input. (2) outlet boundary condition the outlet boundary condition at the right wall of vessel is set as the atmospheric pressure. (3) wall boundary condition this numerical model sets the no-slip boundary condition on the solid surfaces of the solenoid valve system. all kinds of flowing flu id proble ms are determined by physical principles, which a re e xp ressed in conservative form for mathe matica l description. they are mass equation (continuity equation) and mo mentum equ ation. moreover, as the fluid is under the turbulent condition, the additional turbulent equation is needed to incorporate with the governing equations. the continuity and momentum equations in conservation form are expressed as follows: (1) continuity conservative equation m i i s x u t       )( (1) here iu is the velocity,  is the density, and m s is the source term. advances in technology innovation, vol. 3, no. 3, 2018, pp. 133 140 copyright © taeti 137 (2) momentum conservative equation iiiii fgpuuu t     )()()(  (2) where p is the static pressure, 𝜏𝑖𝑗 is the stress tensor, and ρ𝑔𝑖 and 𝐹𝑖 are gravitational and external body forces, respectively. also, the k-ε turbulence model is utilized to solve the nav ier-stokes equations. with respect to the incompressible flo w and no source condition under the steady -state, the momentum equation is :       ji jl l ij i j j i ji ji j i uu xx u x u x u xx p uu x u t                                         3 2 (3) note that eq. (3) is ca lled the reynolds -averaged navie r-stokes (rans) equation, where the reynolds stress -ρui ′ uj ′ should be appropriately modeled by the boussinesq hypothesis [15] for relat ing to the mean velocity gradients . the advantage of this approach is the relatively lo w co mputational cost associated with the computation of the turbulent viscosity. the k-ε model computes the turbulent viscosity as a function of turbulence kinetic energy k and turbulence dissipation rate ε: (4) (5) (6) where 𝐺𝑘 = 𝑢𝑖 ( ∂𝑢𝑖 ∂𝑥𝑗 + ∂𝑢𝑗 ∂𝑥𝑖 ) ∂𝑢𝑖 ∂𝑥𝑗 is the turbulent kinetic energy generated by the mean veloc ity gradients. 𝐶1𝜀 , 𝐶2𝜀 , 𝐶𝜇 , 𝜎𝐾 , and 𝜎𝜀 are mode l constants with the fo llo wing e mp irica lly derived values: 𝐶1𝜀= 1.44, 𝐶2𝜀= 1.92, 𝐶𝜇 = 0.09, 𝜎𝐾= 1.0 and 𝜎𝜀= 1.3, respectively [16]. 4. numerical simulations and discussions                                     k jk t j i i g x k x k x k t     k c k gg xxxt k j t j i i 2 21                                         2 c t  (a) overall velocity distribution (b) region associated with nozzle exit (c) expansion part of the valve connector (d) connecting duct to the storage vessel fig. 5 velocity distribution for the pressure-rising process inside a pneumatic solenoid valve advances in technology innovation, vol. 3, no. 3, 2018, pp. 133 140 copyright © taeti 138 this numerical study chooses a comme rcia l 3/2 -way solenoid valve as the reference valve for analy zing its performance. firstly, cfd software fluent is adopted to simu late the steady flow field associated with the valve configuration. later, with the aids of analyzing numerical results, the comprehensive flow visualization is imple mented to identify the locations of adverse flow patterns, which are c ritica l for proposing the improving a lternatives. hence, the thorough realization on performance features of this valve is attained. fig. 5(a) shows the overall velocity distribution inside this pneumatic solenoid valve. the high-pressure incoming air stream flows into the valve through the nozzle and undergoes an accelerating and expanding process. then, this high-speed stream at the nozzle e xit enters the small c learance between the nozzle and the armature. certa inly, the comp ressed air hits the armature strongly and directly before it flo ws into the inner space of valve. as indicated in fig. 5(b), there are several significant circu lations occurred in the right portion while a much wea ker c ircu lation e xists in the right part , which is due to an air outlet provided by the connector. later, owing to the stepwise geometry inside the connector, expansion and circulation are observed in these area-enlarging locations (see fig. 5c). finally, the compressed air reaches the 1-liter vessel via the connecting duct with a sma ll c ross section. certain ly, as de monstrated in fig. 5(d ), t wo circulat ions generate on both sides of the incoming air flow. in addition, the pressure distribution in the overa ll system of this pneu matic solenoid valve is illustrated in fig. 6(a). obviously, the pressure trend decreases along the flow path from the nozzle, the connector, the connecting duct, and the storage vessel as expected. certain ly, the most dra matic pressure variation occurs inside the nozzle and region near the clearance between nozzle and a rmature as indicated in fig. 6(b). as a result, the comp rehensive flow visualizat ion yie lds the locations of adverse flow patterns, also, circulation and reserved flows are observed at region near the nozzle e xit, e xpansion part of valve connector, and junction of the connecting duct to vessel. the above informat ion is c rit ical for proposing the improving a lternatives. accordingly, the no zzle dia meter and the d istance between nozzle and spool top are identified as t he important design parameters to enhance the pressure response characteristics of valve. (a) overall system (b) region near the nozzle exit inside the valve fig. 6 pressure distribution for the pressure-rising process inside a pneumatic solenoid valve advances in technology innovation, vol. 3, no. 3, 2018, pp. 133 140 copyright © taeti 139 5. conclusions the flo w patterns and response characteristics of a 3/2-way solenoid valve under the charging mode are analy zed in this numerical investigation. with the aids of comprehensive flo w visualizat ion, the locations of adverse flow mechanis ms are realized and identified as the foundation for further modifications on its reaction performance. it fo llo ws that the locations of circulat ion and reserved flo ws are observed at region near the nozzle e xit, e xpansion part of valve connect or, and junction of the connecting duct to the storage vessel. also, it is found that a high-pressure region e xists in the reg ion between the nozzle e xit and the top of iron core. accordingly, the nozzle dia meter and the distance between nozzle and spool top are recognized as the important design parameters for improving the pressure response characteristics of solenoid valve. clearly, the reaction time can be reduced by increasing the nozzle dia meter with an appropriate d istance between nozzle and spool top. moreover, the charging time for th is valve is estimated successfully via a transient cfd ca lculation in an acceptable deviation fro m test result. in conclusion, this work de monstrates a rigorous and systematic cfd scheme to evaluate the perfo rmance characteristics and to provide important information on key design parameters of the pneumatic solenoid valve nomenclature cμ、c1ε、c2ε constants of standard turbulent k-ɛ mode l ρ fluid density gk turbulent kinetic energy generated by the mean velocity gradients ρa air density t time 𝜏𝑖𝑗 shear stress tensor u absolute velocity tensor σk prandtl constant of turbulent kinetic equation μ turbulent viscosity σε prandtl constant of turbulent dissipation equation references [1] r. b. van varseveld and g. m. bone, “accurate position control of a pneumatic actuator using on/off solenoid valves ,” ieee/asme transactions on mechatronics , pp. 195-204, september 1997. [2] a. gentile, n. i. giannoccaro, and g. re ina, “ experimental tests on position control of a pneumatic actuator using on/off solenoid valves,” industrial technology, pp. 555-559, december 2002. [3] g. be lforte, s. mauro, and g. mattiazzo, “a method for increasing the dynamic performance of pneumat ic servo systems with digital valves ,” mechatronics , vol. 14, no. 10, pp. 1105-1120, december 2004. [4] t. a. parlikar, w. s. chang, y. h. qiu, m. d. seeman, and d. j. perreault, “design and experimental implementation of an electromagnetic engine valve drive,” ieee/asme transactions on mechatronics, pp. 482-494, october 2005. [5] e. e. topçu, i̇. yüksel, and z. kamış, “development of electro-pneumatic fast switching valve and investigation of its characteristics,” mechatronics, vol. 16, no 6, pp. 365-378, july 2006. [6] g. tao, h. y. chen, y. y. j, and z . b. he, “opt ima l design of the magnetic field of a h igh -speed response solenoid valve,” journal of materials processing techn ology, vol. 129, no. 1-3, pp. 555-558, october 2002. [7] l. wang, g. x. li, c. l. xu, x. xi, x. j. wu, and s. p. sun, “ effect o f characteristic para meters on the magnetic properties of solenoid valve for high-pressure common rail diesel engine,” energy conversion and management, vol. 127, pp. 656-666, november 2016. [8] z. f. peng, c. g. sun, r. b. yuan, and p. zhang, “the cfd analysis of main valve flo w fie ld and structural optimizat ion for double-nozzle flapper servo valve,” procedia engineering, vol. 31, pp. 115-121, 2012. [9] y. x. ma and d. c. sun, “the numerical simulation of the flow field in an electromagnetic valve ,” machine tool & hydraulics, vol. 36, no. 1, pp. 111-113, january. 2008. [10] q. f. liu, f. l. zhao, and h. l. bo, “numerical simulation of the head of the direct action solenoid valve under the high temperature condition,” 24th international conf. on nuclear engineering, asme press, june 26-30, 2016. [11] z. liu, x. han, and y. f. liu, “ dynamic simu lation of large flo w solenoid valve,” international mechanica l engineering congress and exposition, asme press, november 11-17, 2016. [12] r. farrell, l. i. ezekoye, and m. rain, “check valve flow and disk lift simulation using cfd ,” 2017 pressure vessels and piping conf. paper, asme press, july 16-20, 2017. [13] ansys fluent user’s guide-14.5, ansys inc, 2012. [14] b. e. launder and d. b. spalding, lectures in mathematical & models of turbulence, london, england, july 1972. advances in technology innovation, vol. 3, no. 3, 2018, pp. 133 140 copyright © taeti 140 [15] j. o. hinze, turbulence, mcgraw-hill publishing co, 1975. [16] s. v. patankar and d. b. spalding, “a calculation procedure for heat mass and momentum transfer in three -dimensional parabolic flows,” international journal of heat mass transfer, vol. 15, no. 10, pp. 1787-1806, october 1972.  advances in technology innovation, vol. 3, no. 2, 2018, pp. 51 58 a study on aluminum pad large deformation during copper wirebonding for high power ic package hsiang-chen hsu 1,2,* , shaw-yuan wang 1 , and li-ming chu 3,4 1 department of mechanical and automation engineering, i-shou university, kaohsiung, taiwan , roc. 2 department of industrial management, i-shou university, kaohsiung, taiwan, roc. 3 interdisciplinary program of green and information technology, national taitung university, taitung, taiwan, roc. 4 department of applied science, national taitung university, taitun g, taiwan, roc. received 21 july 2017; received in revised form 09 sept ember 2017; accept ed 13 sept ember 2017 abstract in this paper, a 3-d finite element prediction on aluminum pad squeeze during the copper wire bonding process for high power ic package is presented. ansys parametric design language (apdl) has been implemented on modelling, mesh density, boundary condition (bc), impact stage and contact mode for the first bond process. the ansys/ls-dyna solver is applied to solve dynamics and ls-prepost is used to observe the predicted large plastic deformation on bond pad and stress on microstructure under pad. in view of high power ic package, larger diameter of copper wire is required for electric loading for its low cost. in this research, a large diameter of 2 mil (50 um) uncoated pure copper (4n) wire is applied to simulate first bond impact -contact process. as the scale double enlarged, the problems encountered in simulation are usually evident. preliminary results on impact stage demonstrate that ne gative volume/hourglass on large distortion can be solved by tune -up inertial contact settings and mesh density. however, sever hourglass defect would occur on ultrasonic stage and remain a pending problem. a series of prediction has been conducted on the first bond process during impact stage and the results can then be applied to the dynamic wirebonding assembly process . keywords: copper wirebonding, large plastic deformation, impact -contact, hourglass defect 1. introduction although several advanced interconnection techniques have been developed recently [1-3], the wire bonding process is still widely used in semiconductor packaging industries for its easy application and low cost. thermosonic bonding (t/s bonding) technique [4] has shown better reliability and good interconnection among all wirebonding processes. the 4n (99.99%) thin gold wire (1mil diameter) is commonly used in wirebonding process because it has mechanically good elongation and electrically low resistivity [5]. bonding wire usually divid ed into three different zones, namely free air ball (fab), haz (heat affected zone) and as -drawn wire due to the effect of efo (electric flame-off). the material behavior of three zones in gold wire has been fully investigated by many earlier works [5-6]. however, elastic modulus (e) and poisson’s ratio (υ) at leveled working temperature for fab and haz on gold wire as well as cu wire are still scarcely known. the micro -vickers indentation test was applied to obtain the vickers hardness (hv) and then transfer to ultimate tensile stress (uts). while the surface tensile strength can be determined by nano-indentation test. the thermo-tensile mechanical attributes were measured by our self-designed pull test fixture. * corresponding author. e-m ail address: hchsu@ isu.edu.tw advances in technology innovation , vol. 3, no. 2, 2018, pp. 51 58 copyright © taeti 52 as the price of gold (au) increases, the subs titute material was focused on pd-coated cu wire or bare cu wire just for cost down. the aluminum (al) pad has been replaced by al-cu pad in accordance with cu material property in the wire bonding process. it is also an excellent replacement for au wire d ue to its similar electrical properties. self inductance and self capacitance are nearly the same for au and cu wire. however, the higher rigidity and more hardness [6-7] results in easily squeeze out the bond pad around the smashed ball, as shown in fig. 1. another issue is more bonding power and higher working temperature required in cu wire bonding process. the dielectric material in circuit under pad (cup) for wafer technology node (<60 nm) is low-k material and for wafer technology node (<28 nm) is ultra low-k material. because the low-k/ultra low-k material (nano-porous silicon oxide) is brittle, copper wirebonding process easily damaged the inter-metal dielectric (imd) layer, which is demonstrated in fig. 2. fig. 1 large deformation and squeeze out bond pad near ball bond [8] fig. 2 micro cracks occurred in imd layer beneath the bond pad [8] another advantage of cu wire is lower resistivity. hence, cu wire is a very common interconnected media used for high power transmission. although ultra-thin diameter of wire (0.8 mil (20 m)) has been applied to asic ic or memory ic, large diameter of copper wire (2 mil (50 m)) is still used on high power ic. size effects for larger diameter are visible. reliability on bond pad squeeze and cup micro-crack are much more overt. these attract many researchers and scholar attention on copper wire high power ic package. because the complete mechanism of the wire bonding process includes z-motion, impact and ultrasonic vibration stages [9-10], many material properties of bond wire were scarcely realized. the wirebonding is therefore essentially difficult to simulate by numerical approach. few papers [11-12] published the reliability of wirebonding process using finite element analysis (fea) software ansys/ls-dyna. however, some of the bonding material data in these papers were numerical assumption. with sufficient experimental material properties obtained in this research, both 2d and 3d fea models were developed to predict the dynamic response of the wire bonding process . 2. experimental works tensile mechanical properties of thin 4n au wire and cu wire before/after efo (electrical flame-off) have been investigated by self-design pull test fixture. microstructure characteristics of fab and ha z are also carefully examined. the mechanical tensile properties for such ultra-thin wire are difficultly evaluated in traditional pull test. hence, microtensile tests have been carried out by employing the instron-3365 universal test system with 5n±0.5% load cell and self-designed pull test fixtures. thermal effects (25, 125 and 200 o c) on material properties are taken into account. the chamber needs to be specially processed with nitrogen gas to avoid oxidization for cu wire. it has been reported that the plastic behavior and fracture will occur in the haz area when external loading is applied. it is , therefore, clear that the breakage sites of efo wire are in the neck somewhere between haz and fab. fig. 3 schematically illustrates the self-designed pull test fixtures for the haz. advances in technology innovation , vol. 3, no. 2, 2018, pp. 51 58 copyright © taeti 53 fig. 3 self-designed tensile test fixtures for the neck (haz) 2.1. micro mechanical tensile test fig. 4 and fig. 5 show the temperature effects on stress -strain curves for as -drawn and efo bare cu wire. fig. 4 micro-tensile mechanical properties at different temperature for as -drawn bare cu wire fig. 5 micro-tensile mechanical properties at different temperature for efo bare cu wire true stress -strain ( t  t  ) relationships from plastic deformation to necking can be evaluated from engineering stress -strain curves ( - ). by power law in eq. (1), n tt k  (1) where n is the hardening index and k is the coefficient of strength. the value of hardening index is 10  n which is shown to be a key factor for fea simulation. taking logarithm on both sides tt nk  logloglog  (2)         d d d d n ij ij     )log()log( )log()log( )(log )(log (3) for strain-rate effect, m t k    (4)                                            12 12 12 12 ,, log log loglog loglog log log ln ln              ttttttm (5) advances in technology innovation , vol. 3, no. 2, 2018, pp. 51 58 copyright © taeti 54 from the above figures, true stress -strain mechanical properties in this study is listed in table 1, where e is young’s modulus , y  is yield strength, ts is tens ile strength and f  is failure strain. from equation (2), hardening index n and coefficient of strength k can be quickly determined as 408 mpa. table 1 true stress -strain mechanical properties material density (g/cm 3 ) poisson’s ratio young’s modulus(gpa) yield stress (mpa) tangent modulus (gpa) elongation (%) n  copper ball (200℃) 19.3 0.43 30 110 0.30 12 al-pad 2.71 0.33 69 400 1.38 13 passivation 1.31 0.24 32 350 3.20 3 low-k imd 2.00 0.30 18 80 1.80 3 copper via 8.91 0.38 121 330 0.60 3 oxide 2.64 0.32 66 430 6.60 3 usg 2.00 0.23 80 380 8.00 3 al-cu pad 2.85 0.31 100 400 610 0.054 die 2.33 0.23 161 2.2. finite element prediction both 2-d and 3-d finite element models based on fea software ansys/ls-dyna codes are developed to simulate the wirebonding process. the geometry of the overall structure was first built to create models. since fea prediction also focus es on the strain/stress beneath the bond pad, the entire microstructure of cu/ low-k layer should be well defined. fig. 6 and fig. 7 pres ent the 2d fea geometry, microstructure of cu/low-k imd and 3d fea solid model, respectively. because the accurate material properties should be reflected as inputs for precise finite element analysis, the above experimental measured data can then be directly applied to the existing fea model. a fine mesh scheme is required to evaluate large plastic deformations on mashed fab with a sufficient accuracy. it should be noted that the small contact region between capillary and fab is always collapsed, which resulted in the so-called hourglass mesh or zig-zag patch during iteration approach. this happens on hexahedral 3d solid reduced integration elements in fea model and needed to be re-meshed very frequently. in addition, the precise dimension for fea model sh ould be carefully measured. table 2 lists the modal cart of capillary (tool) and table 3 specifies the dimension of cu/low-k imd microstructure beneath the bond pad . the diameter of cu wire is 50 m and 85 m of fab (free air ball). fig. 6 finite element 2d copper wirebonding model. fig. 7 finite element 3d copper wirebonding model. advances in technology innovation , vol. 3, no. 2, 2018, pp. 51 58 copyright © taeti 55 table 2 detail charts for capillary descriptions dimensions diagram of capillary tip diameter (t) 188um hole diameter (h) 58um chamfer diameter (b) 81um inside chamfer (ic) 11um inside chamfer angle (ca) 90° face angle (fa) 22° outside radius (or) 38um table 3 detail dimension for microstructure beneath the bond pad length(um) width(um) height(um) pad 146 73 5.2 passivation 146 73 4.58 low-k imd 146 73 4 copper via 14 14 4 oxide 146 73 1.8 die 146 73 1.8 2.3. physical mechanism the bonding process is simulated in three steps. in the first step, the capillary (also refer to the “tool”) push fab downward 10 m within 0.7 ms to touch the pad. second, the tool is continuous ly pushing fab impact pad and the contact face/length between ball and pad became welded. the third step provides a slightly downward force and ultrasonic vibration, which refers to 120 khz frequency, 1 μm amplitude within 4 ms vibration time. loading for the tool is: (1) y-travel and impact time is 0.3 ms, (2) the vertical displacement is 36 m, (3) the horizontal displacement is 2 m. fig. 8 demonstrates the key physical mechanism in wirebonding process fig. 8 capillary displacement physical mechanism 2.4. boundary conditions all the boundary conditions based on the wirebonding physical process for 3d predicted fea model are illustrated in fig. 9. the bottom surface of fea model is assumed to be fixed. advances in technology innovation , vol. 3, no. 2, 2018, pp. 51 58 copyright © taeti 56 fig. 9 boundary conditions for the predicted 3d fea model 3. result and discussion 3.1. squeeze-out bond pad fig. 10 fea predicted wirebonding model larger plastic deformation occurred in the area under smashed fab. fig. 10 displays the predicted results for fea wirebonding model. as can be seen, the largest deformation took place in 2 regions: (1) contact area between capillary and fab, and (2) contact area between fab and bond pad. negative volume, zig -zag patch and hourglass elements are always happening and producing meaningless results during numerical iterations. mesh density and element shape in the largest deformation regions are particularly addressed to avoid numerical iteration errors. squeeze -out bond pad can be controlled by giving the optimal conditions in fea model. fig. 11 presents the predicted maximum squeeze-out bond pad. fig. 11 predicted maximum squeeze-out bond pad advances in technology innovation , vol. 3, no. 2, 2018, pp. 51 58 copyright © taeti 57 3.2. predicted effective stress once the bond pad squeeze has been optimized, special attentions are focused on the dynamic response beneath t he bond pad. fig. 12 shows the predicted effective stress vs. time history in low-k imd layers and the predicted effective stress contours is shown in fig. 13. fig. 12 predicted maximum effective stress vs. time history in low-k microstructure fig. 13 predicted maximum effective stress at the end of impact motion 3.3. parametric studies an increase in the bond pad would result in a decrease both in the maximum effective stress in the bond pad and the squeeze-out bond pad. on the contrary, an increase in the bond pad would result in an increase in the maximum effective stress in the low-k imd microstructure. table 4 lists predicted results as the bond pad height is increased . table 4 predicted maximum effective stress for different bond heights. bond pad height (m) 5.2 6.2 7.2 maximum effective stress (mpa) on bond pad 131 111 109 squeeze-out bond pad (m) 2.22 1.44 1.34 maximum effective stress (mpa) in low-k imd 96.69 96.76 98.88 4. conclusions in this paper, the predicted copper wirebonding proces s for high power ic based on fea model has been developed. the insight of the physical mechanism of wirebonding p rocess has also been explored. the experiment works on micro mechanical tensile test and true stress -strain relationships for cu wire have been determined. with these material properties, the fea predicted wirebonding model becomes feasible and numerical simulation errors have been fixed in this research. parametric stu dy reveals an increase in the bond pad height would reduce the large plastic d eformation as well as the maximum effective stress on the bond pad. the predicted results can be directly applied to the practical assembly process . advances in technology innovation , vol. 3, no. 2, 2018, pp. 51 58 copyright © taeti 58 references [1] y. j. lin, c. kang, l. chua, w. k. choi, and s. w. yoon, “advanced 3d ewlb-pop (embedded wafer level ball grid array-package on package) technology,” proc. 66 th electronic components and technology conference, may 2016, pp. 1772-1777. [2] m. c. hsieh, k. t. kang, h. c. choi, and y. c. kim, “thin profile flip chip package-on-package development,” proc. 11 th international microsystems, packaging, assembly and circuits technology conference, oct ober 2016, pp. 143-147. [3] n. srikanth, s. murali, y. m. wong, and c. j. vath iii, “critical study of thermosonic copper ball bonding,” thin solid films, vol. 462-463, pp.339-345, september 2004. [4] d. s. liu, y. c. chao, and c. h. wang, “study of wire bonding looping formation in the electric packaging process using the three-dimensional finite element method,” finite element in analysis and design, vol. 40, no. 3, pp. 263-286, january 2004. [5] y. liu, s. irving, and t. luk, “thermosonic wire bonding process simulation and bond pad over active stress analysis,” ieee transactions on electronics packaging manufacturing, vol. 31, no. 1, pp. 61-71, january 2008. [6] f. y. hung, t. s. lui, l. h. chen, and y. t. wang, “recrystallization and fracture characteristics of thin copper wire,” journal of materials science, vol. 42, no. 14, pp. 5476-5482, july 2007. [7] c. hanga, c. wang, m. shi, x. wu, and h. wang, “study of copper free air ball in thermosonic copper ball bonding,” proc. 6 th international electric packaging technology conference , august-september, 2005, pp. 414-418. [8] w. y. chang, h. c. hsu, s. l. fu, y. s. lai, and c. l. yeh, “an investigation on heat affected zone for au wire/cu wire and advanced finite element wirebonding model,” proc. 3 rd microsystems, packaging, assembly & circuits technology conference, october 2008, pp. 419-423. [9] c. l. yeh, y. s. lai, and c. l. kao, “transient simulation of wire pull test on cu/low-k wafers,” ieee transactions on advanced packaging, vol. 29, no. 3, pp. 631-638, august 2006. [10] h. c. hsu, c. y. hu, w. y. chang, c. l. yeh, and y. s. lai, “dynamic finite element analysis on underlay microstructure of cu/low-k wafer during wirebonding,” finite element analysis , pp. 453-478, august 2010. [11] s. k. prasad, advanced wirebond interconnection technology, kluwer academic publishers , pp. 3-55, 2004. [12] r. c. j. wang, c. c. lee, l. d. chen, k. wu, and k. s. chang-liao, “a study of cu/low-k stress -induced voiding at via bottom and its microstructure effect,” microelectronic reliability, vol. 46, no. 9-11, pp.1673-1678, september-november 2006. http://www.springerlink.com/content/?author=yuan-tin+wang microsoft word 1-v8n1(2023)-aiti#10548(29-37).docx advances in technology innovation, vol. 8, no. 1, 2023, pp. 29-37 on the estimation of the mission performance index of unmanned surface vehicles based on the mission coverage area jae-yong lee, nam-sun son* department of autonomous & intelligent maritime systems research division, korea research institute of ships & ocean engineering, daejeon, korea received november 12, 2021; received in revised form 07 may 2022; accepted 08 may 2022 doi: https://doi.org/10.46604/aiti.2023.10548 abstract for mission planning and replanning of multiple unmanned surface vehicles (usvs), it is important to estimate each usv’s mission performance in terms of sea surveillance (e.g., illegal ship control). in this study, a mission performance index (mpi) is proposed based on the mission coverage area for estimating the usvs’ mission performance of illegal ship control. the penalty value is considered in the mpi calculation procedure owing to the track-off of the usv. in addition, the usv simulation is conducted under illegal ship control, and the mpi is calculated based on changing the mission coverage area. the results show that the mpi increases with the path width of the mission coverage area. keywords: unmanned surface vehicle (usv), mission coverage area, mission performance index (mpi), illegal ship control, usv simulation 1. introduction generally, an unmanned surface vehicle (usv) is a small ship (1.5 to 15 m and 0.5 to 9 t) that can be controlled remotely or autonomously to perform missions under unfavorable weather conditions [1]. although usvs are typically developed for military use, they have also been applied to marine transportation, marine surveys, response to trespassing ships in sea farms, sea rescue, and fusion with underwater remotely operated vehicles [2-5]. furthermore, the international maritime organization has been reviewing international agreements on maritime safety and security regarding maritime autonomous surface ships [6]. currently, asian countries (e.g., korea) are conducting government-supported projects in maritime industries, such as shipping, ports, shipbuilding, and offshore [7]. recently, owing to the fourth industrial revolution, unmanned, automated, and online fields have been highlighted. accordingly, usv-related studies have been conducted to support or replace the missions performed by manned ships [8-9]. in particular, illegal ship control is important for monitoring usvs for mission planning and replanning [10]. in this study, the mission performance index (mpi) is defined to estimate the mission performance of multiple usvs using the mission coverage area for sea surveillance, such as illegal ship control. the mpi is estimated using the resulting trajectory of the usv, which is the ratio of the individual mission area to the total mission coverage area. to evaluate the mpi, a usv simulation is conducted under illegal ship control. in this study, the features of the mpi for the usv and the simulation results are described. * corresponding author. e-mail address: nnso@kriso.re.kr tel.: +82-42-3636; fax: +82-42-3624 advances in technology innovation, vol. 8, no. 1, 2023, pp. 29-37 30 previous studies pertaining to path following and target tracking are explained in section 2. the proposed mpi-based estimation method for usvs is presented in section 3. the verification of the mpi via simulation is described in section 4, and the conclusions and future studies are provided in section 5. 2. previous path-following and target-tracking investigations in the korea research institute of ships and ocean engineering, the aragon series of usvs (aragon1, aragon2, and aragon3) was developed through a project entitled “development of multipurpose intelligent unmanned surface vehicle” [10]. recently, usv swarms have been investigated using the aragon series of usvs through a project entitled “development of situation awareness and autonomous navigation technology of usv based on artificial intelligence” [11]. in this study, a usv (aragon1) was used as an illegal ship to simulate illegal ship control, where the vehicle steers away from a patrol ship (i.e., another usv (aragon3)) via path following. therefore, a path-following algorithm is required when aragon3 trails aragon1’s escape path. previously, a path-following algorithm was developed using the line-of-sight (los) [12]. as shown in fig. 1, the heading angle must be controlled for path-following control. in this case, wpk is the waypoint of k (xk, yk), pt is the current position (xt, yt) of the usv, ψt is the heading angle, and δt is the nozzle angle of the waterjet. the target heading angle (ψtc) was calculated using the waypoint of k, and the position of aragon1 was calculated using the los for path following [12]. in other words, aragon3 was used for illegal ship control. a target-tracking algorithm is required for a patrol ship (i.e., aragon3). yun et al. [13] developed a target-tracking algorithm using the concept of virtual points, as illustrated in fig. 2. their proposed usv was used as a patrol ship to track the target ship (illegal ship) in the starboard and port or stern. when tracking a target ship, the usv must maintain a certain direction with a separation distance at a specific speed. as shown in fig. 2, the usv tracks the target ship via virtual points “a” and “b.” in this case, virtual point “a” is generated by the heading angle of the target ship, and virtual point “b” is generated by the separation distance from the target ship. the target-tracking algorithm is based on tracking a target ship with the los angle. in addition, various international studies related to path following and target tracking have been conducted [14]. an improved los guidance algorithm that can be adjusted based on the path-following error was investigated [15]. additionally, a deep reinforcement learning method for solving the path-following problem of the usv has been investigated based on the decision-making network [16-17]. the straight-line target control of usvs has been investigated based on maneuverability and agility using the straight-line concept with high speed [18]. the purpose of this study is to estimate the mpi using the resulting trajectory of an illegal ship (aragon1) and a patrol ship (aragon 3). in this study, the path-following algorithm is applied to an illegal ship (aragon1) using the los, and the target-tracking algorithm is applied to the patrol ship (aragon3) using the concept of virtual points. fig. 1 path-following algorithm [12] fig. 2 target-tracking algorithm [13] advances in technology innovation, vol. 8, no. 1, 2023, pp. 29-37 31 3. estimation of mpi 3.1. concept of mpi the performance of usvs for mission planning and replanning is to be estimated. the mpi is calculated using the resulting trajectory of the usv, i.e., the ratio of the individual mission area to the mission coverage area. as shown in fig. 3, the mission coverage area is in the sea environment of illegal ship control, where the fairways of ships are extremely narrow owing to fish farms, tides, and reefs. the mpi is to be analyzed while the path width of the mission coverage area is changed from “2l” to “6l.” in addition, if the usv is outside the mission coverage area, then mpi* is applied to estimate the mpi while considering the penalty, as shown in fig. 4. fig. 3 concept of mpi and cases of mission coverage area with various path widths fig. 4 concept of mpi and penalty area 3.2. procedure of mpi estimation in this study, the mpi was estimated to analyze the mission performance of the usvs. the overall procedure is summarized in table 1. first, the usv information was input, including the name, length, breadth, and speed of the usv. the trajectory information of the usv was selected from the simulation results, and the mission coverage area was set using various path widths. next, the penalty area was calculated by considering the track-off from the mission coverage area. finally, the mpi was estimated using the mission coverage area and penalty area. table 1 procedure for estimating mpi* step description 1 input information of usv 2 select trajectory information of usv 3 set mission coverage area (path width: 2l to 6l) of usv 4 calculate usv trajectory area 6 calculate penalty area 7 estimate mpi advances in technology innovation, vol. 8, no. 1, 2023, pp. 29-37 32 fig. 5 flowchart for estimating mpi fig. 6 example of mpi estimation a flowchart of mpi estimation is shown in fig. 5. the path width was changed from 2l to 6l to verify the change in the estimated mpi. furthermore, the total area of the usv trajectory (cat(i)), total mission coverage area (tmc(i)), and penalty area (pa(i)) were calculated using the equations shown in fig. 5. mpi#(i) is the ratio of the relative mpi*(i) of the usv to the mpi in the case of the resulting trajectory without track-off. an example of mpi estimation is shown in fig. 6. the notations used in mpi estimation are presented in the nomenclature section. here, information regarding aragon1 and aragon3 is shown, such as the latitude, longitude, waypoints, speed, and heading. the mission coverage area (eq. (1)), usv trajectory area (eq. (2)), and penalty areas (eq. (3)) are calculated as follows: advances in technology innovation, vol. 8, no. 1, 2023, pp. 29-37 33 1 ( ) ( ( ) ) 2 4 6) n jj tmc i wd i li or or = = × ( × (1) 1 ( ) ( ( ) ) ( ) n jj cat i d i b i = = × (2) 1 ( ) ( ( ) ) ( ) n jj pa i gd i b i = = × (3) 4. simulation and results 4.1. simulation scenario in this study, a simulation pertaining to the illegal control of a usv was conducted in gwangam port, korea to verify the proposed mpi-based estimation method. as illegal ships frequently appear on sea farms, a sea farm was selected as the game area for the simulation. aragon1 was used as an illegal ship, and aragon3 was used as a patrol ship for illegal ship control. the escape path of aragon1 was planned in advance by considering the zone of the sea farm, as shown in fig. 7, and aragon1 trailed this path using the path-following algorithm based on the los [12]. aragon3 chased the illegal ship (aragon1) for illegal ship control using the target control algorithm based on the concept of virtual points [13]. table 2 summarizes the information regarding aragon1 and aragon3. the usv simulation was conducted via dynamic simulation based on the nomoto model [19]. the resulting trajectories of aragon1 and aragon3 using the path-following and target-tracking algorithms are shown in fig. 8. fig. 9 shows the histories of the speed and heading angle of aragon1 and aragon3 based on the simulation. as shown in figs. 8 and 9, the simulation was conducted with the path widths ranging from “2l” to “6l” to verify the change in the estimated mpi. the red and blue lines represent the resulting trajectories of aragon1 and aragon3, respectively. the path-following algorithm of aragon1 was successfully conducted at a speed of 10 knots. furthermore, the target control algorithm of aragon3 was successfully implemented at a high speed in the simulation. the heading error between aragon1 and aragon3 can be determined because aragon3 tracks aragon1 via the target-tracking algorithm on the stern side. fig. 7 path of aragon1 and location of sea farm table 2 information regarding aragon1 and aragon3 no. ship length (m) breadth (m) maximum speed (knot) 1 illegal ship (aragon1) 8 2.5 10 2 patrol ship (aragon3) 8 2.5 max 20 advances in technology innovation, vol. 8, no. 1, 2023, pp. 29-37 34 fig. 8 resulting trajectories of aragon1 and aragon3 (a) speed of aragon1 and aragon3 (b) heading angle of aragon1 and aragon3 fig. 9 time histories of the speed and heading angle of aragon1 and aragon3 4.2. analysis of mpi based on simulation results as shown in tables 3 and 4, the mpi* exceeds 95% for the illegal ship (aragon1) and patrol ship (aragon3). this implies that aragon1 successfully conducted path following as an illegal ship, and aragon3 successfully conducted target tracking as a patrol ship. the mpi# of aragon1 was 4.9% higher than that of aragon3, on average. this is because aragon1 adhered to the designated path, whereas aragon3 tracked aragon1. therefore, aragon3 exhibited track-off in target control owing to the heading error during target tracking. furthermore, as the path width of the mission coverage area widened (from 2l to 6l), mpi# increased because the penalty area in both aragon1 and aragon3 reduced. table 3 mpi results mission coverage area index units aragon1 aragon3 2l cat(i) (m2) 4,611.62 4,847.85 np(i) (ea) 84 1,399 pa(i) (m2) 210 3,497.5 tmc(i) (m2) 29,948.91 31,591.37 advances in technology innovation, vol. 8, no. 1, 2023, pp. 29-37 35 table 3 mpi results (continued) mission coverage area index units aragon1 aragon3 4l cat(i) (m2) 4,611.62 4,847.85 np(i) (ea) 25 858 pa(i) (m2) 62.5 2,145 tmc(i) (m2) 60,084.80 63,563.23 6l cat(i) (m2) 4611.62 4,847.85 np(i) (ea) 0 299 pa(i) (m2) 0 747.5 tmc(i)) (m2) 89,869.70 95,389.62 table 4 comparative analysis of relative mpi* without track-off path width usv mpi* mpi# (%) = cat(i) ÷ (tmc(i) + pa(i)) = mpi * ÷ [(b(i)) ÷ (path widths)] × 100 2l aragon1 0.1529 97.86 aragon3 0.1381 88.42 4l aragon1 0.0766 98.14 aragon3 0.0738 94.44 6l aragon1 0.0513 98.52 aragon3 0.0504 96.82 5. conclusions and future studies in this study, an mpi was defined to estimate the mission performance of multiple usvs using the mission coverage area. in particular, the penalty value was considered in the mpi calculation procedure owing to the track-off of the usv. to verify the proposed mpi-based estimation method, illegal ship control was simulated using an illegal ship (aragon1) and a patrol ship (aragon3). the illegal ship (aragon1) and patrol ship (aragon3) successfully conducted path following and target tracking, respectively. therefore, the mpi exceeded 4.9%. the mpi of the illegal ship (aragon1) was higher than that of the patrol usv (aragon3) because aragon1 adhered to the designated path, whereas aragon3 indicated track-off during target tracking. furthermore, as the path width of the mission coverage area increased (from 2l to 6l), the mpi increased owing to a decrease in the penalty area. in the future, the authors will conduct an experiment pertaining to illegal ship control in an actual sea using multiple usvs; subsequently, the experimentally obtained mpi will be compared with simulation results. nomenclature notation units definition i number of ith usvs (aragon1 or aragon3) n(i) name of ith usv (aragon1 or aragon3) l(i) m length of ith usv b(i) m breadth of ith usv v(i) knot speed of ith usv ψ(i) ° heading of ith usv lg(i)j ° j th trajectory point of longitude for ith usv lt(i)j ° j th trajectory point of latitude for ith usv p(i)xj m x value converted from lgij of j th trajectory point in ith usv p(i)yj m y value converted from ltij of j th trajectory point in ith usv d(i)j m distance between usv trajectories; ����� = ������ − ���� � �� + ������ − ����� � �� td(i)j m total distance of usv trajectory; ������ = ∑ ���� �� j cat(i) m2 total usv trajectory area; ������ = ������ × ���� w(i)xj m x of j th waypoint for ith usv w(i)yj m y of j th waypoint for ith usv advances in technology innovation, vol. 8, no. 1, 2023, pp. 29-37 36 wd(i)j m distance between waypoints; ������ = ������ − ���� � �� + ������ − ����� � �� tmc(i) m2 total mission coverage area; ������ = ∑ ������� × ����� × path width �� � g(i)xj m x of j th waypoint for ith usv outside mission coverage area g(i)yj m y of j th waypoint for ith usv outside mission coverage area gd(i)j m distance between waypoints outside mission coverage area; ������ − &��� �� + ������ − &���� �� np(i) number of points deviating from mission coverage area for ith usv (ea) pa(i) m2 penalty area; ����� = ∑ �'����� � × ����� �� mpi(i) mission performance index; ��(��� = ������ ÷ *��� + �����+ mpi#(i) % ratio of relative mpi* of usv to mpi for case involving resulting trajectory without track-off; ��(#��� = ��(∗ ÷ ������ ÷ �path width� × 100 acknowledgments this study was supported by the project titled “development of situation awareness and autonomous navigation technology of unmanned surface vehicle based on artificial intelligence (pes3880, pes4270),” which was funded by the korea research institute of ships and ocean engineering. conflicts of interest the authors declare no conflicts of interest regarding the publication of this study. references [1] v. bertram, “unmanned surface vehicles—a survey,” skibsteknisk selskab, vol. 1, pp. 1-14, january 2008. [2] j. e. manley, “unmanned surface vehicles, 15 years of development,” oceans, pp. 1-4, september 2008. [3] d. w. jung, et al., “a study on unmanned surface vehicle combined with remotely operated vehicle system,” proceedings of engineering and technology innovation, vol. 9, pp. 17-24, july 2018. [4] j. ansary, et al., “swarms of aquatic unmanned surface vehicles (usv), a review from simulation to field implementation,” proceedings of international design engineering technical conferences and computers and information in engineering conference, vol. 83914, pp. 436-528, august 2020. [5] n. s. son, et al., “on the sea trial test for the validation of an autonomous collision avoidance system of unmanned surface vehicle, aragon,” proceedings of the oceans 2018 mts/ieee charleston, pp. 1-5, october 2018. [6] c. j. chae, et al., “a study on identification of development status of mass technologies and directions of improvement,” applied sciences, vol. 10, no. 13, article no. 4564, july 2020. [7] b. s. hwang, et al., “the fourth industrial revolution and future innovation: policy fields,” proceedings of the korea technology innovation society conference, pp. 436-528, august 2017. [8] r. murphy, et al., “an analysis of international use of robots for covid-19,” robotics and autonomous systems, vol. 148, article no. 103922, february 2021. [9] s. y. kim, “research trends of usv,” journal of the society of naval architects of korea, vol. 51, no. 2, p. 2, june 2014. [10] s. y. kim, et al., “project: development of multi-purpose intelligent unmanned surface vehicle,” korea institute of ocean science and technology, final report pjt20040, august 15, 2019. [11] n. s. son, “project: development of situation awareness and navigation technology of unmanned surface vehicle based on the artificial intelligence,” korea research institute of ships and ocean engineering, report pes3880, january 2022. [12] n. s. son, et al., “study on a waypoint tracking algorithm for unmanned surface vehicle (usv),” journal of navigation and port research, vol. 33, no. 1, pp. 35-41, march 2009. [13] k. h. yun, et al., “a study on the algorithm of target tracking system for usv based on simulation method,” journal of ships and ocean engineering, vol. 56, pp. 19-25, december 2015. [14] m. a. bin mansor, et al., “motion control algorithm for path following and trajectory tracking for unmanned surface vehicle: a review paper,” proceedings of the 3rd international conference on control, robotics, and cybernetics, pp. 73-77, september 2018. advances in technology innovation, vol. 8, no. 1, 2023, pp. 29-37 37 [15] t. liu, et al., “path following control of the underactuated usv based on the improved line-of-sight guidance algorithm,” journal of polish maritime research, vol. 24, no. 1, pp. 3-11, april 2017. [16] y. zhao, et al., “path following optimization for an underactuated usv using smoothly-convergent deep reinforcement learning,” ieee transactions on intelligent transportation systems, vol. 22, no. 10, pp. 6208-6220, october 2021. [17] j. h. woo, et al., “deep reinforcement learning-based controller for path following of an unmanned surface vehicle,” ocean engineering, vol. 183, pp. 155-166, july 2019. [18] m. breivik, et al., “straight-line target tracking for unmanned surface vehicles,” modeling, identification, and control, vol. 29, no. 4, pp. 131-149, april 2008. [19] t. i. fossen, guidance and control of ocean vehicles, new york: john wiley and sons, 1994. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation, vol. 1, no. 2, 2016, pp. 53 57 53 cop y right © taeti improved svm classifier incorporating adaptive condensed instances based on hybrid continuous-discrete particle swarm optimization chun-liang lu1,*, tsun-chen lin2 1department of applied information and multimedia, ching kuo institute of management and hea lth, keelung, taiwan. 2department of computer and communication engineering, dahan institute of technology, hualien, taiwan. received 20 sep 2016; received in revised form 24 sep 2016; accept ed 28 sep 2016 abstract in recent years, support vector machine (svm) based on empirical risk minimization is supervised learning model which has been successfully used in the classification and regression. the standard soft-margin svm trains a classifier by solving an optimization problem to decide which instances of the training data set are support vectors. however, in many real applications , it is imperative to perform feature selection to detect which features are actually relevant. in order to further improve the performance, we propose the adaptive condensed instances (aci) strategy based on the hybrid particle swarm optimization (hpso) algorithm for the svm classifier design. the basic idea of the proposed method is to adopt hpso to simultaneously optimize the aci and svm kernel parameters for the classification accuracy enhancement. the numerical experiments on several uci benchmark datasets are conducted to find the optimal parameters for building the svm model. experiment results show that the proposed framework can achieve better performance than other published methods in literature and provide a simple but subtle strategy to effectively improve the classification accuracy for svm classifier. ke ywor ds : hybrid partic le swarm optimizat ion (hpso), adaptive condensed instances (aci), support vector machine (svm) 1. introduction support vector machine (svm) was originally proposed by vapnik [1] which is a powerful classification method with state-of-the-art performance in machine learning theory, has drawn considerable attentions due to its high generalization ability for a wide range of applications including speaker recognition [2], bioinformatics [3] and text categorization [4]. in many pattern classification tasks , we are confronted with the problem that the input space is high dimensional and to find out the combination of original input features which contribute most to the classification is crucial. the computational cost of classification grows heavily with data dimension size, making feature selection an important issue for the svm. the feature selection mechanism falls into three categories: filtering, wrapper, and embedded methods [5]. filters generally involve a non-iterative computation on the original features, which can execute very fast, but not usually optimal since the learning algorithm are not taken into account. wrapper methods usually achieve better results than filters since they are tuned to the specific interactions between the classifier with original feature set and very computationally intensive. finally, unlike filters and wrappers, the embedded techniques simultaneously determine features and classifier during the training process but the computational time is smaller than wrapper methods. feature selection problem is a challenging task because there can be complex interaction among features . therefore, an exhaustive search is practically impossible, and the efficient global search technique is needed. evolutionary computation (ec) are well known heuristic approaches global search ability such as simulated annealing (sa) [6], genetic algorithm (ga) [7], and particle swarm optimization (pso) [8], have gained a lot of attention from researchers in the area. compared with other ec algorithms such as sa and ga, pso is comput ationally less expensive and can converge more quickly. a ga-based feature selection method, which optimized both the feature selection and parameters for svm, was proposed by huang [9], and the authors pointed out that the algorithm may work superior to the conventional grid search method. however, the treatment of these redundant or irrelevant instances is not taken into account in the classification procedure. in this paper, the effectively adaptive condensed instances (aci) strategy that we previously published [7] is applied to decide which instances of the training data set are support vectors for coping with the problem mentioned above. short communications of the early stages of this work have appeared in [7]. here we significantly extend our approach to account for the relevant instances selection during the svm training process. in order to further improve the classification performance, the aci strategy based on the hybrid particle swarm optimization (hpso) algorithm is proposed for the svm classifier design. several uci benchmark datasets are conducted to validate the effectiveness and the experiment results show that the proposed framework can achieve better performance than other ga-based existing methods in literature. the remainder of this paper is organized as follows. section 2 describes the related work including the basic pso and svm classifier. section 3 illustrates particle representation, hybrid pso with disturbance operation, aci scheme and the proposed framework for the svm classifier. section 4 provides the experiment results, and conclusions are made in section 5. 2. related work 2.1. particle swarm optimization (pso) algorithm the pso algorithm, which is originally developed by kennedy and eberhart [10], is a search algorithm modeling the social behavior of birds within a flock. in the pso algorithm, individuals referred to as particles, are flown through hyper dimensional search space. pso is easy to imp le ment, fe w parameters to ad just, and usually faster convergence rates *corresponding aut hor, email: leucl@ems.cku.edu.t w advances in technology innovation, vol. 1, no. 2, 2016, pp. 53 57 54 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti than other evolutionary algorithms. during the optimiza tion procedure, particles communicate good positions to each other and adjust position according to their history e xperience and the ne ighboring particles. the basic concept of the pso algorithm is illustrated as follow: 1 1 1 2 2 ( ) ( ) k k k k id id id id k k id id v w v c r pb x c r gb x            (1) 1 1k k k id id id x v x     (2) where 1, 2,...,d d , 1, 2,...,i n , and d is the dimension of the search space, n is the population size, k is the iterative times; k id v is the i-th particle velocity, k id x is the current particle solution, the i-th particle position is updated by equation (2). k id pb is the i-th particle best ( best p ) solution achieved so far; k id gb is the global best ( best g ) solution obtained by any particle in the population; 1r and 2r are random values in the range [0,1] for denoting remembrance ability. both of 1c and 2c are learning factors, w is inertia factor. a large inertia weight facilitates global exploration, while a small one tends to local exploration. generally, the value of each component in v can be clamped to the range [− maxv , maxv ] for controlling excessive roaming of the particle outside the search space. the pso procedure is organized in the following sequence of steps. step 1: initialize: randomly generating initial particles. step 2: fitness evaluation: calculate the fitness values of each particle in the population. step 3: update: co mpare fitness values of each particle to update the velocity and position by using equation (1) and (2). step 4: termination criterion: repeat the step 2 to step 3 until the number of iteration reaches the pre-defined maximu m number or a termination criterion is satisfied, and the best solution best g is displayed. 2.2. support vector machine (svm) classifier support vector machine (svm) have drawn much attention due to their good performance and solid theoretical fou ndations [11]. the main concepts of svm are to first transform input data into a higher dimensional space by means of a kernel function and then to find an optimal separating hyper-plane between the two data sets. a practical difficulty of using svm is the selection of parameters such as the penalty parameter c of the error term and the kernel parameter  in rbf kernel function. the appropriate choice of parameters is to get the better generalization performance. the description of svm is as follows. given a set of training data 1( , , )px x with corresponding class labels 1( , , )py y and { 1, 1}iy    . the svm attempts to find a decision surface ( )f x , to jointly ma ximize the margin between the two classes and minimize the classification error on the training set. 1 ( ) ( , ) p i i ii f x y k x x b    (3) where k is a kernel function, i  is the lagrange multiplier corresponding to the i-th training data i x and b is the bias. the simple kernel is the inner product function ˆ ˆ( , ) ,k x x x x which produces the linear decision boundaries. nonlinear kernel function maps data points to a high-dimensional feature space as linear decision spaces . two commonly used kernels are the polynomial kernel ˆ( , ) p k x x and the radial basis function (rbf) kernel ˆ( , ) r k x x . the integer polynomial order  in p k and the width factor  in r k are hyper-parameters which are tuned to a specific classification problem.  ˆ ˆ( , ) 1 ,pk x x x x    (4) 2ˆ ˆ( , ) x x r k x x e    (5) 3. method since the optimal hyper-plane obtained by the svm depends on only a small part of the data points (support vectors), it may become sensitive to noises or outliers in the training set. in this section, the aci scheme based on hybrid pso (hpso) is proposed to tackle feature selection, condensed instances extraction and parameters setting simultaneously for svm. the particle representation, fitness definition, disturbance strategy for pso operation, aci scheme and the proposed hybrid framework for svm are described as follows. 3.1. particle representation this study used the rbf kernel function for the svm classifier to implement our proposed method. the rbf kernel function requires two parameters c and  should be set. using the adaptively condensed instance rate for aci scheme and the rbf kernel for svm, these three parameters cond  , c , and features used as input attributes must be optimized simu ltaneously for our proposed hybrid system. the particle, therefore, is comprised of four parts, cond  , c ,  are the continuous variables and the features mask are the discrete variables. table 1 shows the particle representation of our design. table 1 the hybrid particle representation as shown in table 1, the representation of particle i with dimension of 3 f n  , where f n is the number of features that varies from different datasets, ,1 ,~ fi i nx x are the features mask, , 1fi n x  indicates the parameter value cond , , 2fi nx  represents the parameter value c and , 3fi nx  denotes the parameter value . fitness function f is the guide of hpso operation to ma ximize the classification accuracy and minimize the number of selected features. cc a is the svm classification accuracy, i f denotes the feature mask which 1 represents the feature i is selected and 0 indicates that feature i is not selected. thus, the particles with high classification accuracy and a small number of features produce a high fitness value and affect particle’s positions on the next iteration. considering the tradeoff b etween the classification accuracy and selected feature number, these two weight 1w and 2w values can be adjusted according to the preference for svm classifier design. advances in technology innovation, vol. 1, no. 2, 2016, pp. 53 57 55 cop y right © taeti 1 1 2 1 fn i i f w acc w f              (6) 3.2. pso with disturbance operation in the discrete pso, the particle’s personal best and global best is updated as in continuous value. the major different between discrete pso with continuous version is that velocities of the particles are rather defined in terms of probabilities that a bit whether change to one. by this definition, a velocity must be restricted within the range min max[ , ]v v . this can be accomplished by a sigmoid function ( )s v , and the new particle pos ition is calculated using the following rule: if 1 min max( , ) k id v v v   then 1 1max minmax(min( , ), ) k k id id v v v v    , where 1 1 1( ) 1 k id k id v s v e      (7) if 1( ) ( )k id rand s v   , then 1 1k id x   ; else 1 0k id x   . the function ( ) id s v is a sigmoid limiting transformation and ( )rand is a random number selected from a uniform distribution in [0, 1]. note that the discrete pso is susceptible to a sigmoid function ( )s v saturation which occurs when velocity values are either too large or too small. for a velocity of zero, it is a probability of 50% for the bit to flip. according to the searching behavior of pso, the best g value will be an important clue in leading particles to the global optimal solution. it is unavoidable for the solution to fall into the local minimum while particles try to find better solutions. in order to allow the solution exploration in the area to produce more potential solutions, a mutation-like disturbance operation is inserted between eq. (1) and eq. (2). the disturbance operation random selects k dimensions (1 k  problem dimensions) of m particles (1  m  particle numbers) to put gaussian noise into their moving vectors (velocities). the disturbance operation will affect particles moving toward to unexpected direction in selected dimensions but not previous experience. it will lead particle jump out from local search and further can explore more diversity of searching space. 3.3. adaptive condensed instances (aci) scheme the effectively aci scheme that we previously published in [7] is extended to decide which instances of the training data set are support vectors. the adaptively condensed instances coefficient in the data reduced process is flexible to edit out noisy samples, reduce the superfluous data points and make the svm less sensitive to noises and outliners. the adaptively condensed instances coefficient [0,1] cond   is defined as the ratio of the selected number of condensed instances to overall dataset. the aci scheme is described as follows. step 1: initialize: randomly generating initial instance set. step 2: condense: to decide whether all samples have been achieved the user defined threshold cond  . if so, terminate the process; otherwise, go to step 3. step 3: extend: if there are any un-condensed instances , using the nearest neighbor voting to renew the condensed instances; otherwise, no more new data is joined and go to step 2. 3.4. the proposed framework for svm classifier based on the particle representation, fitness definition, pso with disturbance operation and aci scheme mentioned above, details of the proposed hybrid framework for svm procedure from step 1 to step 9 are described as follows. step 1: data preparation given a dataset d is considered using the 10 fold cross validation to split the data into 10 groups. each group contains training and testing sets. the training and testing sets are represented as train d and test d , respectively. step 2: hybrid pso initialization and parameters setting set the pso parameters including the number of iterations, number of particles, velocity, particle dimension, disturbance rate, and weight for fitness. generate initial hybrid particles comprised of the features mask, cond  , c and  . step 3: condensed instances selection via the aci scheme according to the condensed instances coefficient cond  represented in the particle and calculated from step 2, when all the condensed instances are computed, the nearest neighbor voting is used to renew the condensed instances set. step 4: feature scaling feature scaling is to properly reveal the interactions between features and to avoid attributes in greater numeric ranges dominating those in smaller numeric ranges. normalization by eq. (8) can be linearly scaled to range [-1, +1] or [0, 1], where ( ) j i a x is the original attribute value of feature i x , ' ( ) j i a x is scaled value, max j and min j correspond to the maximum and minimum values for j a over all samples. ' ( ) min( ) , max min j i j j i j j a x a x i     (8) step 5: feature selection according to the feature mask, which is represented in the particle from step 2, is to select input features for training set train d and testing set test d . the selected features subset can be denoted as _f traind and _f testd , respectively. step 6: to train and test for svm classifier for the parameters cond  , c and  which are represented in the particle, to train the svm classifier on the training dataset _f traind , then the classification accuracy cca for svm on the testing dataset _f testd can be evaluated. step 7: fitness evaluation for each particle, the fitness value is to be calculated by the eq. (6). the optimal fitness value can be stored on the evolution process of pso to search for the better fitness of particle in the next particles evolution procedure. step 8: termination criteria when the number of iteration reaches the pre-defined ma ximum number or a termination criterion is satisfied, the best solution best g is obtained, and the program ends; otherwise, go to the next step. advances in technology innovation, vol. 1, no. 2, 2016, pp. 53 57 56 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti step 9: hybrid pso operation in the evolution process, discrete-valued and continuous-valued dimension of hpso with the disturbance operator continued to be applied for searching better particle solutions. 4. experiment results 4.1. dataset and system description to measure the performance of the developed hybrid framework, several real benchmark datasets in table 2 are conducted to verify the effectiveness of performance. these table 2 the uci benchmark datasets [7] datasets are partitioned using the 10-fold cross validation. our implementation platform was implemented on matlab 2013, by extending the libsvm which is originally designed by chang and lin [12]. through initial experiment, the parameter values of the pso were set as follows. the swarm size is set to 100 particles. the searching ranges of continuous type dimension parameters are: [0,1] cond   , 2 4[10 ,10 ]c  and 4 4[10 ,10 ]  . the discrete type particle for features mask, we set min max[ , ] [ 6,6]v v   , which yields a range of [0.9975,0.0025] using the sigmoid limiting transformation by eq. (7). both the cognition learning factor 1c and the social learning factor 2c are set to 2. the disturbance rate is 0.05, and the number of generation is 500. the inertia weight factor min 0.4w  and max 0.9w  . the linearly decreasing inertia weight is set as eq. (9), where now i is the current iteration and maxi is the pre-defined maximum iteration. according to the fitness function defined by eq. (6), set the accuracy’s weight 1 0.8w  and the feature’s weight 2 0.2w  . max max min max ( )now i w w w w i    (9) table 3 the parameters setting for pso and ga in [9], the authors presented the ga-based method without aci mechanism for searching the best c ,  , and features subset. the existing ga-svm method without aci scheme deals solely with feature selection and parameters optimization by means of genetic algorithm, and the treatment of these redundant or noisy instances in a classification process did not be taken into account. our proposed hybrid framework has been tested fairly extensively and compared with these approaches including both ga-svm and pso-svm approaches without aci mechanism. furthermore, the comparison of pso and ga technique with the aci scheme for svm is also presented. while applying ga algorithm, a number of parameters are required to be specified. the two algorithms (both pso and ga) are run for the same number of fitness function evalu ations. table 3 summarized the parameters used for pso and ga technique. the empirical results are reported in section 4.2. 4.2. result and comparison the results obtained by the developed hpso-aci-svm approach are compared with those of ga -svm proposed by huang et al. [9] without aci scheme. taking the heart disease dataset, for example, the classification accuracy cc a , number of selected features f n , and the best parameters cond  , c ,  for each fold are shown in table 4. for the hpso-aci-svm method, average classification accuracy rate is 96.87%, and average number of features is 4.9. for the ga-svm without aci approach, its average classification accuracy rate is only 94.81%, and average number of features is 5.6. table 4 comparison for hpso-aci-svm and ga-svm table 5 comparison for hpso-aci-svm, pso-svm and ga-svm advances in technology innovation, vol. 1, no. 2, 2016, pp. 53 57 57 cop y right © taeti table 5 shows the summary results for the average class ification accuracy rate of the hpso-aci-svm hybrid framework and pso-svm, ga-svm without aci method on six uci datasets. in table 5, the classification accuracy rate is represented as the form of ‘average  standard deviation’. to highlight the advantage, we used the non-parametric wilcoxon signed rank test for all of the datasets. in table 5, the p-values of hpso-aci-svm versus pso-svm and ga-svm are smaller than the statistical significance level of 0.05 except iris dataset. that is to say, the developed hpso-aci-svm yields higher classification accuracy rate across different datasets to enhance the performance for the svm. further, under our proposed hybrid framework with aci mechanism, the performance of both pso and ga optimization techniques in terms of the average classification accuracy rate _ cc avg a and the average number of selected features _ f avg n is compared. in table 6, the hpso exhibits slightly higher classification accuracy and fewer selected features than ga. it is observed that, from an evolutionary point of view, the performance of the hpso is better than ga. however, the results indicate that both pso and ga algorithms can be used in optimizing the parameters under our developed hybrid framework to effectively improve the classification accuracy for svm classifier design. table 6 comparison for hpso-aci-svm and ga-aci-svm 5. conclusion for svm classifier, it is imperative to perform feature selection to detect which features are actually relevant. in this study, the effectively adaptive condensed instances (aci) strategy is applied to decide which instances of the training data set are support vectors. furthermore, we propose the a ci scheme based on the hybrid particle swarm optimization (hpso) algorithm to simultaneously optimize the condensed instances and svm kernel parameters for the classification accuracy enhancement. several uci benchmark datasets are conducted to validate the effectiveness of the proposed method, and the experiment results show that the proposed hybrid framework can achieve better performance than other existing methods in literature. investigating more large scale dataset as well as combining other heuristic algorithm for hybrid system may be interesting future work. references [1] v. n. vapnik, "an overview of statistical learning theory," ieee transactions on neural networks, vol. 10, pp. 988-999, 1999. [2] i. j. ding, c. t. yen, and d. c. ou "a method to integrate gmm, svm and dtw for speaker recognition," international journal of engineering and technology innovation, vol. 4, pp. 38-47, 2014. [3] r. kothandan and s. biswas , "identifying micrornas involved in cancer pathway using support vector machines ," computational biology and chemistry, vol. 55, pp. 31-36, apr. 2015. [4] b. ra mesh and j. g. r. sathiaseelan, "an advanced multi class instance selection based support vector machine for text classification," procedia computer science, vol. 57, pp. 1124-1130, 2015. [5] x. zhang, g. wu, z. dong and c. crawford, "embedded feature-selection support vector machine for driving pattern recognition," journal of the franklin institute, vol. 352, pp. 669-685, 2015. [6] s.w. lin, z. j. lee, s. c. chen and t. y. tseng, "parameter determination of support vector machine and feature selection using simulated annealing approach ," applied soft computing, vol. 8, pp. 1505-1512, 2008. [7] c. l. lu, i. f. chung and t. c. lin, "the hybrid dynamic prototype construction and parameter optimization with genetic algorithm for support vector machine," international journal of engineering and technology innovation, vol. 5, pp. 220-232, 2015. [8] b. sahu and d. mishra, "a novel feature selection algorithm using particle swarm optimization for cancer microarray data," procedia engineering, vol. 38, pp. 27-31, 2012. [9] c. l. huang and c. j. wang, "a ga-based feature s election and parameters optimization for support vector machines ," expert systems with applications, vol. 31, pp. 231-240, 2006 [10] j. kennedy and r. c. eberhart, "a discrete binary version of the particle swarm algorithm," in proceedings of the world on systematics, cybernetics and informatics, vol. 8, pp. 4104-4109, 1997. [11] v. vapnik, the nature of statistical learning theory, new york: springer-verlag, 1995. [12] c. c. chang, and c. j. lin, "training nu-support vector regression: theory and algorithms ," neural computation, vol. 14, pp. 1959-1977, 2002.  advances in technology innovation, vol. 4, no. 2, 2019, pp. 94-104 implementation of adaptive embedded controller for a temperature process j. satheesh kumar 1,* , deepu sankar 2 1 department of electronics and instrumentation, dayananda sagar college of engineering, bangalore, india 2 department of electronics and instrumentation, karunya university, india received 08 december 2017; received in revised from 11 april 2018; accepted 18 august 2018 abstract the paper proposed and carried out an adaptive embedded control strategy with the help of arduino open hardware platform. the proposed control strategy is to carry out a cost-effective interface between the simulation software and a real-time process. the data acquisition and control is done with the help of arduino uno which has been interfaced with matlab simulink the control algorithms developed in simulink model can be downloaded into the arduino uno, working as a standalone controller. in this paper, various control algorithms are used to control the temperature process, including embedded modified model reference adaptive control (mmrac). its performance is compared to other control algorithms. the result shows that the mmrac scheme improves the transient performance of the temperature control system. keywords: embedded, control, matlab, simulink, mrac 1. introduction nowadays, the automation and control of most of the real-time systems are achieved by using a computer connected through the expensive data acquisition system (das). the specification and requirement of das depend on the software being used. the proposed methodology uses a low-cost das and an embedded system to develop a controller that can give better performance. the development of das and the controller in the same hardware improves the compactness of the proposed system. in this paper, the arduino uno micro controller is used for the development of das and controller. the control algorithm is developed using matlab simulink tool. tawanda [1] presented similar algorithm developments using simulink. the matlab simulink provides a platform to interface simulink with arduino uno through arduino io library and support package. the development of the control algorithm in the embedded system has become simpler by this approach. the early versions of the embedded target devices for simulink tool are expensive. the real-time interface (rti) developed by arduino uno, is cost effective. five different control algorithms, namely proportional control (p), proportional-integral control (pi), proportional integral derivative control (pid), model reference adaptive control (mrac) and mmrac were developed to control a temperature process using simulink. there are two ways to test the developed algorithm in real-time. one method is to use the arduino union, along with the matlab simulink running on the pc. another way is to use arduino uno as a standalone controller in real time. in the second case, after downloading the control algorithms in arduino uno microcontroller, the computer can be disconnected from it. the second method provides the flexibility of using the arduino uno as a standalone embedded controller. this embedded controller is used to control the temperature process in real-time environment. * corresponding author. e-mail address: jsatheeshngl@gmail.com tel: +91-9003368217 advances in technology innovation, vol. 4, no. 2, 2019, pp. 94-104 95 the pid controller is used in process industries to control various processes. based on the necessity and nature of the process, the adaptive controller is considered as a suitable controller for many processes. mrac is one such adaptive controller. several adaptive control structures are discussed by [2] and [3]. the robustness of the control system are improved by combining both pid and adaptive control action. the result presented by [4] shows the improved performance by the addition of pid in adaptive control. a modified control scheme of mrac is discussed in this paper. in the modified mrac, pid controller action is included. the error given to the pid controller is derived from the model output and the process output. the combined action of the pid controller and mrac drives the final control element in the process. this adaptive controller will give better settling time and efficiency compared to mrac. therefore an arduino aided mmrac is another main aspect of the paper. the most significant property of a real-time interface in matlab simulink and arduino uno helps in the development of standalone controller. hence this work is identified as an adaptive embedded controller for a temperature process. 2. simulink-arduino real time interface matlab and simulink have been associated with engineering and academic world as a design and development tool. the method of linking, design, and development into the real-time process in an optimized way is a challenging task. standard data acquisition cards are available to facilitate the real-time interface. data acquisition systems for matlab are discussed in [5]. software specific data acquisition systems are used in many cases; however, these das can only be used in a specific software application and may not be compatible with other applications, and such systems are presented in the article [6]. matlab provides the platform to interface the simulink with the real-time process through the arduino io package or arduino support package. the simulink model can be developed using the arduino io library blocks. the required control algorithms are developed using simulink and arduino support package. the arduino support package converts the simulink model to code, which runs directly on arduino uno. therefore the arduino uno can be disconnected from the host computer, and it controls the real-time process independently. 3. experimental analysis the experiment is conducted in a thermal process analyzer, it includes a thyristor controlled electric furnace and an air blower. the air blower acts as a load variable in the process. the blower speed can be varied manually to introduce the disturbance. the controlled variable is the furnace temperature and the manipulated variable is the control voltage, to the thyristor-based power control system. fig. 1 block diagram of thermal process control systems the above mentioned experimental setup is the prototype of a thermal process. the control system block diagram for the thermal system is shown in fig. 1. according to the desired temperature, the control voltage is determined from the control advances in technology innovation, vol. 4, no. 2, 2019, pp. 94-104 96 algorithm, which is present in the microcontroller. the temperature measurement is made with the help of lm35 precision integrated-circuit temperature sensor. the output voltage of this sensor is linearly proportional to the centigrade temperature and the scale factor is 10mv per degree centigrade. the temperature range of this sensor is -55°c to 150°c. the measured temperature is acquired with arduino uno which is a microcontroller based on the atmega328. a thyristor-based power control circuit is used to regulate the temperature of the furnace. the control voltage for the thyristor-based power control circuit determines the temperature of the furnace by regulating the ac supply voltage using a pulse width modulation method. the control voltage is obtained from the arduino uno. typically the control voltage is an analog signal. since arduino uno supports only digital outputs an external digital to analog convertor (dac) is used. dac circuit is designed using dac0808 which is an 8-bit monolithic digital-to-analog converter. the control voltage from dac is given to a thyristor-based power control for proper temperature regulation. 4. embedded adaptive controller the possibilities of embedded controller implementation using simulink are summarized in this section. the real-time implementation of an adaptive controller using simulink requires a target device. the various kinds of target devices compatible with simulink are micro controllers, digital signal processors (dsp) and xpc target. most of the target devices are costly and unbearable for a smaller level projects. the real-time control by xpc target using simulink is developed as mentioned in [7] and [8]. the xpc target devices are compatible with high-performance industrial computers. the dsp based embedded platform provides the facility of model conversion from simulink to real-time environment. it gives the flexibility of an interfacing more number of peripheral devices. among these target devices, microcontrollers are the cheapest device used in an embedded system. in microcontroller category also various processors are available, some of the processors are atmega, arm processors, etc. the arm processor-based embedded adaptive controller is developed as presented in [9]. arduino is one of the microcontrollers used for the development of embedded controllers. control algorithm development by arduino integrated development environment (ide) is discussed in [10]. through ide, arduino can be programmed using embedded programming. the ide based control algorithm development is a bit complex and time-consuming. the simulink model-based program is not possible with ide. the arduino support package facilitates the simulink model-based programming in arduino. in this paper, a control algorithm is implemented in arduino uno using the arduino support package. the process temperature is read by ic-based temperature sensor lm35. arduino reads the temperature of the thermal system through its pin number 1. since the arduino is having 10-bit analog to digital convertor (adc), the maximum adc value is 1024 and the reference is given 5v. the scaling has to be done properly to get the process temperature. as per the data sheet, the output of lm35 is 10mv/ºc. the actual temperature is found by eq. (1). 5 ( c) 0.483 1024 0.01 out out v t v      (1) where 𝑉𝑜𝑢𝑡 is the output of lm35 in decimal equivalent. the equivalent temperature value is given to the control algorithm to compute the required control signal. the control signal is sent to its digital output pins as shown in fig. 2. the control signal generated is converted into 8-bit data and is given to 8 digits write blocks. integer to bit converter block demands only integer values, hence a rounding function is used before it. the 8-bit conversion requires the multiplication of the control signal with constant so that the 0-5v will be converted to 0-255. the multiplication constant 51 is chosen to convert 0-5 range to 0-255 range. the 8-bit digital outputs are sent to digital pins 4, 5, 6, 7, 10, 11, 12 and 13 of arduino. the thyristor-based control circuit requires an analog voltage. hence these digital signals are converted into analog signal by a digital to analog convertor (dac). an external dac is used for this conversion. advances in technology innovation, vol. 4, no. 2, 2019, pp. 94-104 97 fig. 2 data acquisition system 4.1. implementation of p, pi, and pid controllers fig. 3 p, pi & pid controllers using simulink arduino support package the real-time implementation of p, pi and pid controllers using simulink arduino support package is shown in fig. 3. the developed control algorithm is downloaded into arduino to work as a standalone controller. all the three control functions are achieved using the same matlab subfunction by properly enabling the proportional, integral and derivative gains. the setpoint is given to pin number 2 of the arduino uno using a potential divider circuit. this circuit provides the output in t he range of 0-5v to pin number 2 of arduino uno. the provision for adjusting the setpoint is also properly scaled based on eq (1). the integral controller may experience the integral windup in real-time process control. the pid control algorithm includes the anti-reset windup algorithm to reduce integral windup. the data acquisition system for the thermal process shown in fig. 3 is illustrated in fig. 2. advances in technology innovation, vol. 4, no. 2, 2019, pp. 94-104 98 4.2. development of mathematical model for the temperature process the laboratory-based temperature process analyzer consists of a smaller size furnace. the operating region is in between 25°c to 80°c. process reaction curve method is used to find the mathematical model of the process. the feedback in fig. 3 is removed. the initial steady state is noted. a step input of known magnitude is given to pin number 2 of the arduino at time t1. the time at which the process variable begins to change is noted as t2. the process variable is allowed to reach the new steady state. the difference between t1 and t2 is the dead time of the process. the change in the process variable is computed by subtracting the new steady state of the initial steady state. time taken by the response to reach 63.2% of process variable change is noted as t3. the time constant of the process is the difference between t3 and t2. the process gain is calculated by dividing the magnitude of the final steady state by the magnitude of the change in step input. the derived model is a first-order system with the dead time process. the model of the system is given in eq. (2). 10 1.2 ( ) 55 1 s e g s s    (2) 4.3. implementation of model reference adaptive control adaptive control is one of the widely used control strategies to design advanced control systems for better performance and accuracy [11]. model reference adaptive control (mrac) is a direct adaption strategy with an adjustable mechanism to adjust controller parameters. the implementation of mrac for the first order process is carried out as discussed in [12]. the mrac works on the principle of adjusting the control parameters so that the output of the actual plant tracks the output of a reference model having the same reference input [13]. the implementation of mrac and the design of adjustment mechanisms are referred to from [14]. the reference model is used to give an ideal response of the adaptive control system to the reference input. the controller is usually described by a set of adjustable parameters. in this paper two adjustable parameters, θ1 and θ2 are used to describe the control law. adaptive mechanism is used to alter the parameters of the controller so that the actual plant could track the reference model. mathematical approaches like mit rule, lyapunov theory, and theory of augmented error can be used to develop the adjusting mechanism. since the experimental thermal process is a first order system, mit based mrac for a first order system is followed for the development of controllers. the model of the temperature process and the reference model are given in eqs. (3) and (4). ( ) g( ) ( ) y s b s u s s a    (3) where y is the output from the process and u is the control input to the process. ( ) ( ) ( ) m m m c m y s b g s u s s a    (4) where m y is the reference model output and the 𝑢𝑐 is the command signal to the reference model. the controller equation is given in eq. (5). 1 2 cu u y  (5) where 1  and 2  are the controller parameters, y is the measured temperature and u is the controller output. advances in technology innovation, vol. 4, no. 2, 2019, pp. 94-104 99 fig. 4 mrac using simulink arduino support package as per the mit rule, the controller parameter has to be changed in the direction of the negative gradient of cost function j. the cost function is chosen in this application is given in eq. (6) and the rate of change of controller parameter is described in eq. (7) e    is known as sensitivity derivative. 21 2 j e (6) ' ' d j e e dt            (7) where e is the error given in eq. (8). m e y y (8) where y is the process output and it is described by eq. (9). from eqs. (3) and (5), y is derived. 1 2 c b y u s a b      (9) the controller parameters are obtained from eqs. (8) and (9). the denominators of sensitivity derivatives are approximated as per eq. (12). 1 2 c e b u s a b       (10) advances in technology innovation, vol. 4, no. 2, 2019, pp. 94-104 100 2 2 e b y s a b       (11) 2 m s a b s a    (12) the mit rule is applied to determine the controller parameters. the sensitivity parameters given in eqs. (10) and (11) are substituted in eq. (7) to find out the controller parameters. the parameters b and m a are combined with adaptation gain '  . the equations for updating the controller parameter 1  are given in eqs. (13) and (14). similarly, the other controller parameter 2  is updated using eqs. (15) and (16). '1 mc c m m d ab u e u e dt s a s a                       (13) 1 m c m a u edt s a             (14) '2 m m m d ab y e y e dt s a s a                       (15) 2 m m a y edt s a             (16) where  is the adaptation gain, c u is command signal, e is an error and y is the measured process variable. practical implementation leads to integral windup due to the integrator present in the equations. hence anti-reset windup is also included in the control algorithm. the reason for reset windup is the accumulation of the input to the integrator. the accumulated value leads to the saturation of the controller output. the anti-reset algorithm helps to remove the integral windup. the anti-reset windup procedure includes a saturation block, which limits the controller output. the controller output before the saturation block is subtracted from the controller output after the saturation block. the subtracted output is added to the input of the integrator. this procedure will not allow the integrator to accumulate its output. this procedure doesn’t have a ny effect till the controller output is below the saturation value because the subtracted value before and after the saturation block is zero. once the controller output begins to increase from the maximum value, the anti-reset windup procedure will start functioning. the same anti reset procedure is used in all the controller implementations described in this paper. the controller implementation is shown in fig. 4. the control signal from the algorithm is limited to 5v using the saturation block in the simulink. as shown in fig. 2, this output from the saturation block is converted into corresponding digital output. using a dac the digital output is converted to analog and given to the thyristor-based power control circuit to control the furnace temperature. 4.4. implementation of modified model reference adaptive control in order to improve the transient performance of the control system, the mrac scheme is modified [15]. in the modified mmrac, classical pid control action is also included. the modified controller output u is derived as per the eq. (17). the controller parameters 1  and 2  have the same purpose similar to the mrac. the mmrac is developed by subtracting the effect of pid controller action from the mrac [16]. 1 2 c p i d de u u y k e k edt k dt            (17) advances in technology innovation, vol. 4, no. 2, 2019, pp. 94-104 101 where p k , i k and d k are the proportional, integral and derivative gains for the pid controller. the optimal values of pid controller parameters are appropriately chosen to provide the best process regulations. fig. 5 mmrac using simulink arduino support package even though many tuning procedures are available, choosing pid controller parameters is a challenging task for control engineers, hence for a real-time process, control engineers prefer trial and error method [17]. the derivative action of pid control improves the predictive nature of the control system and improves the speed of response. it predicts future controlled output by computing the rate of change of error and accordingly takes the control action to minimize the error. the integral action removes the offset in the controlled variable. the integral windup due to the integral action is removed by the anti-reset windup procedure mentioned in the previous section. hence the modified control scheme includes all the advantages of pid control and mrac. the integrator's used in the computation of 1  and 2  also produces reset windup; hence the same anti-reset windup procedure is followed to remove the integral windup. the real-time implementation of the modified model reference adaptive controller is shown in fig. 5. the control algorithm is developed using simulink arduino support package. the real-time interface is developed by the help of analog input block and the digital output block. the data acquisition system for the thermal process analyzer is developed in the subsystem as shown in fig. 2. the setpoint of the standalone mmrac is set with the help of the knob present in the hardware. the value set by the user is read by the controller and it is properly converted into engineering units. the control algorithm compares both the reference model output and the actual process output. the computed controller parameters force the process to follow the reference model. the computation is based on eq. (17) so that the change in the controller parameter minimizes the error. according to the control algorithm, the desired control signal is sent to the thyristor-based power control circuit to control the temperature. advances in technology innovation, vol. 4, no. 2, 2019, pp. 94-104 102 5. results and discussion the classical pid controller algorithm and model-based algorithm was developed in this research. real-time implementation using the microcontroller is simplified by this approach. the experimental setup shows the actual electric furnace with a blower. the thyristor-based power control circuit, controller implementation using simulink and the arduino uno with the peripheral circuits are also depicted in fig. 6. the control algorithms developed in this work are p controller, pi controller, pid controller, mrac and mmrac. the required control algorithm was downloaded into the arduino uno hardware. the electric furnace was allowed to settle at two different temperatures 40°c and 60°c. the controller performance was observed by plotting the response of the temperature control system with respect to time. the transient performance of the system is analyzed for all the controllers. fig. 6 experimental setup 5.1. controller observation and response the following control algorithms p, pi, pid, mrac, and mmrac were implemented in real time. a comparative analysis has been made using the available data. the temperature of the furnace is plotted in graphs with respect to time. the response of the furnace for the setpoint of 40°c is shown in fig. 7. since the setpoint is 40°c and it is closer to the room temperature, the process takes around 20 seconds to reach the setpoint. the rise time of the process varies with respect to the controllers used. the response of the furnace for the setpoint of 60°c is shown in fig. 8. since the process lag for the electric furnace is considerably high, it takes around 200 seconds to reach the setpoint. the time from 200 seconds to 380 seconds is used for the comparative analysis. fig. 7 comparative response of furnace for 40°c setpoint fig. 8 comparative response of furnace for 60°c setpoint 5.2. performance analysis the performance of the controller is analyzed using the settling time. the tolerance limit for the settling time is 1%. the comparative analysis is listed in table 1. the mmrac in real time shows better performance when compared with other advances in technology innovation, vol. 4, no. 2, 2019, pp. 94-104 103 controllers. the other performance metrics are not considered, because except for pi and pid controllers, other controllers have very negligible oscillations or zero oscillations around the setpoint. so there is no overshoot or negligible overshoot for proportional control, mrac and mmrac. the settling time of the process using mrac and mmrac, for the setpoint of 40°c is less than the rise time. the settling time and rise time are the same for the process using mrac and mmrac, for the setpoint of 60°c. in all operating regions, mrac and mmrac respond quickly to the process parameter changes and take the corrective action. table 1 performance analysis controller set point = 40°c set point = 60°c settling time (sec) rise time (sec) settling time (sec) rise time (sec) p 100 100 320 370 pi 80 30 230 240 pid 60 20 220 230 mrac 40 50 220 220 mmrac 20 30 210 210 6. conclusions the arduino open hardware platform is used as a low-cost data acquisition system and controller. p, pi, pid, mrac, and mmrac are implemented in the arduino microcontroller using matlab simulink. the performance analysis is made in terms of settling time. the developed embedded controller provides the platform to implement any type of complex control algorithms. this will help in the field of process control to implement complex algorithms with low-cost devices. the embedded based mmrac controller shows better performance over the other controllers mentioned in this paper. hence, the system under study is best suited for mmrac. this work may be extended by developing intelligent controller using the arduino microcontroller. implementing complicated algorithms can be simplified by using the real-time interface developed in this paper. hence the proposed methodology will be beneficial to academic institutions and research organizations in testing and implementing various sophisticated algorithms. conflicts of interest the authors declare no conflict of interest. references [1] t. mushiri, a. mahachi, and c. mbohwa, “a model reference adaptive control system for the pneumatic valve of the bottle washer in beverages using simulink,” proc. international conference on sustainable materials processing and manufacturing (smpm 17), elsevier, january 2017, pp. 364-373. [2] b. m. mirkin and per-olof gutman, “output feedback model reference adaptive control for multi-input-multi-output plants with state delay,” systems & control letters, vol. 54, pp. 961-972, february 2005. [3] r. a. fahmy, r. i. badr, and farouk a. rahman, “adaptive pid controller using rls for siso stable and unstable systems,” advances in power electronics, vol. 2014, pp. 1-5, october 2014. [4] r. a. fahmy, r. i. badr, and farouk a. rahman, “adaptive pid controller using rls for siso stable and unstable systems,” advances in power electronics, vol. 2014, pp. 1-5, october 2014. [5] k. k. mckee, gareth l. forbes, ilyas mazhar, rodney entwistle, and ian howard, “low cost remote data acquisition system, ”curtin university, department of mechanical engineering, technical note, december 2013. [6] s. humayun, maria mehmood, and faran mahmood, “developing a labview and matlab-based test bed for data acquisition, analysis and calibration of frequency generators over gpib,” international journal of computer applications, vol. 40, pp. 11-15, february 2012. [7] handbook of networked and embedded control systems. birkhauser boston, university of maryland, 2008. [8] d. hercog, a. rojko, m. curkovic, b. gergic, and k. jezernik, “embedded platform for rapid implementation of local and remote motion control experiments,” przeglad elektrotechniczny, vol. 87, pp. 73-76, march 2011. [9] m. engin, “design and control applications of mechatronic systems in engineering,” 1st ed. london: intec, 2017. advances in technology innovation, vol. 4, no. 2, 2019, pp. 94-104 104 [10] a. m. el-nagar, “embedded intelligent adaptive pi controller for an electromechanical system,” isa transactions, vol. 64, pp. 314-327, september 2016. [11] ioana nascu, ioan nascu, and g.vlad, “predictive adaptive control of an activated sludge wastewater treatment process,” advances in technology innovation, vol. 1, pp. 38-40, march 2016. [12] m. p. r. v. rao, “new design of model reference adaptive control systems,” journal of applied mechanical engineering, vol. 3, pp. 1-3, january 2014. [13] s.h. rajani, b. m. krishna, and u. nair, “adaptive and modified adaptive control for pressure regulation in a hypersonic wind tunnel,” international journal of modelling identification and control, vol. 29, pp. 78-87, march 2018. [14] k. j. astrom and b. wittenmark, “adaptive control,” 2nd ed. new york: dover publications inc., 2013. [15] s. zhang and s. dian, “controller design for compound pendulum with pid and mrac switch control,” proc. international conference on automation control and robotics engineering (cacre 2018), iop publishing, july 2018, pp. 1-5. [16] r.j. pawar and b.j. parvat, “mrac and modified mrac controller design for level process control,” proc. indian control conference (icc 18), ieee explore, january 2018, pp. 217-222. [17] l. guessas and k. benmahammed, “adaptive backstepping and pid optimized by genetic algorithm in control of chaotic,” international journal of innovative computing information and control, vol. 7, pp. 5299-5312, september 2011. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation, vol. 4, no. 2, 2019, pp. 73-83 a cloud information platform for 3d printing rehabilitation devices ta-cheng chen 1,2,* , yi-wen chen 2 , yen-shan chen 2 , yun-tzu kuo 1 1 3d printing research center, asia university, taichung, taiwan, roc 2 department of information management, national formosa university, yunlin, taiwan, roc received 04 may 2018; received in revised form 13 august 2018; accepted 24 december 2018 abstract due to the problems of current population aging, occupational injuries and traffic accidents have led to an increasing number of elder people who are physically disabled in taiwan. these persons, mostly seek medical treatment from medical institutions to help restore them to their premorbid level. therefore the number of patients who go to medical institutions is gradually increasing and the splinting is largely applied in the medical cares. 3d printing technology has been widely used in the medical industry in recent years, especially in the development of rehabilitation devices. up until now, to produce the customized splint requires either face to face communication or messaging software by relevant parties. due to the complexity of the medical process, it is often time-consuming for occupational therapists to discuss the medical records with the splint design engineers via the above mentioned means. the other difficulty is that data management becomes a real problem. medical communication and information management are thus the most urgent issues that need to be investigated. in this study, we applied the information technology and cloud-based technology to design a simple and user-friendly web-based interface for making 3d printing splint. this web-based interface utilizes cloud-based technology to provide an information platform for communication and co-management between the relevant stakeholders. the aim of this study is to make system management, retrieving patients’ information and browsing 3d graphics be more convenient for users, and thus to improve the efficiency and effectiveness of producing splints based on the proposed system. keywords: 3d printing, rehabilitation devices, splint, cloud, information platform 1. introduction the 3d printing technology has been applied to the medical field in recent years. this technology has partially solved several problems raised from stuffiness, clothing, and uncomfortable level of wearing the traditional medical assistive devices, such as splints. moreover, 3d printing technology has achieved a further breakthrough of advocating the effective of reducing patients’ feelings of inferiority when wearing an assistive splint. thus, 3d printing has become a feasible solution in solving the aforementioned problems through the production of improved splints. although there are many various medical information systems in medical applications, most of them are the kind of information sharing systems to share the patients’ medical records. however, there are very few information system platforms developed for the purpose of medical device manufacturing and little information about functional treatment units. therefore, doctors and occupational therapist must rely on paper works, telephones, face to face or additional use of communication software to communicate with each other so that such communication process is time consuming. with the 3d printing technology, the splint production method has been transformed from hand-made to digital production. without the information, communication platform, to check the blue print of the splint on the platform by all the related medical care parties becomes * corresponding author. e-mail address: tchen@nfu.edu.tw advances in technology innovation, vol. 4, no. 2, 2019, pp. 73-83 74 impossible. it makes everyone have to communicate with each other by using traditional ways. the process for completing a customized splint by using 3d printing technology is complicated. at first, after diagnosis and medical treatment, doctor keys in all the diagnostic information on the window of patient rehabilitation treatment card through the computer. next, based on the doctor’s diagnosis, the 3d model of the splint has to be designed by the designers and then assessed by the therapist. while the 3d model is approved, the 3d printing process of the splint is then activated. however, during the process, the related personnel must communicate with each other, check the best parameter settings of the 3d graphic in details and discuss and amend repeatedly. the communication methods of the existing medical clinic are using the telephone, messaging software or face to face to communicate. since the medical personnel are too busy to discuss freely in the scene. using the current communication ways as described above without seeing 3d graphics, it will cause many difficulties in expression and make the related data is hard to be managed and stored properly. therefore, this study is to turn the workflow of the 3d printed rehabilitation products into the clouding computerized processes. the information sharing platform enables various parties to cooperate in making a more perfect splint for patients. based on the proposed system, a multi-directional communication and information transmission pipeline can be built so that it makes all the operations with better efficiency and effectiveness. the purposes of this research are illustrated as follows: (1) the electronic system has been adopted instead of using paper files to reduce the unnecessary waste of resources. (2) to explore how to use the information platform to improve the original work situation in the hospital. (3) design the 3d printing assistive device splint cloud information platform to solve the previous problems such as management and communication inconveniences to enhance the efficiency of personnel. 2. literature review a rehabilitation assistive device refers to any device or product used to maintain or improve the quality of life and movement ability of people with disabilities [1]. among the several health care aids, splints have been widely used in medical departments, such as rehabilitation and orthopedics, due to time constraints [2]. a splint is a kind of rehabilitation equipment, used outside of the human body, mainly for the fixation, or protection limbs [3]. splints are used not only to improve the discomfort of limbs, but also to help people to use their limbs normally. compared to the splint made of plaster, we need a lighter, cleaner, and air-permeable splint. so, we can say that the plaster should be taken place by the good ones. creating splints are one of the professional skills of occupational therapists. a splint can be made from a wide variety of materials. a traditional splint is usually made of low-temperature thermoplastic medical materials [4] and can be tailored to individual limb curves. this feature solves the gypsum breathable problem, reduces the burden of patients wearing splint, and makes wound dressing or cleaning of the skin around the wound convenient [5]. however, the traditional splint materials are not waterproof, prone to allergies, difficult to shape on the limbs, and cannot hollow out any position on the splint surface [6]. with 3d printing technology, all kinds of graphics can be generated directly through 3d graphics software design, without restrictions on size, shape, and color. it has a high degree of freedom and complexity in design, so it is more convenient for the production of small-scale limb splint, and can also improve the patient's fit and comfort when using a splint. with the advantage of 3d printing, the pre-opening hole position and the whole mesh pattern are corrected directly in the splint design, as shown in fig. 1. this helps to improve splint's breathable and the waterproof function so that it will reduce the skin irritation or ulceration of the affected body area. using 3d printing technology, it can avoid unnecessary waste of materials and reduce the production costs. after forming 3d printing splints, these splints do not need special maintenances [7]. so, 3d printing technology has become a significant issue of producing splint to an occupational therapist. the differences between the traditional splints and the 3d printing splints are illustrated in table 1. advances in technology innovation, vol. 4, no. 2, 2019, pp. 73-83 75 fig. 1 the production process of 3d printing splint sequence diagram [7] table 1 comparison of traditional splints and 3d printing splints traditional splints 3d printed splints production methods manual 3d printer time to use shorter longer breathable medium height customized low height waterproof no yes suitability to smaller limbs low height complexity low height the introduction of 3d printing has brought many possibilities to the medical treatment. in recent years, various 3d printing medical devices have been developed in medical clinics, such as teeth, tracheal scaffold, splint and so on. these splints are the harness and used in the external body. the purposes of these devices are for treating, immobilizing, or protecting the function of damaged body parts [3]. during the manufacturing of 3d printing splints, all the assigned personnel need to discuss the 3d graphics with one another. however, the current method that uses considerable paperwork delivery fails to meet the demand to check the 3d graphics of splints. the process requires the use of messaging software, telephone call, or face-to-face discussion among personnel. these inefficient methods will cause communication inconvenience, management problems, and 3d graphics browsing difficulty. in the research on medical information transmission, yang and liu imported a device for the traditional manual work mode of nurses and developed a nursing information system [8]. this device assists in daily works by using a computer or network. the results showed that the device could reduce overtime rate and unnecessary waste of materials, make the patient information inquiry convenient, and improve work efficiency. tian et al. showed that the use of electronics to improve medical image processing operations could reduce patient waiting time and accurately manage the medical graphics data to promote high-quality service [9]. doel et al. combined medical image processing operations with cloud technology to assist the data storage, management, and analytical tasks of medical images on the web [10]. this process allows health care personnel and researchers to share the medical image data conveniently. as previously mentioned, electronics and cloud technology will be the necessary tools for future development in the medical industry. recently, umair and kim proposed an online 3d printing service platform and cooperated with medical clinics who are interested in 3d printing [11]. through this platform, the information about 3d printing medical-related data and medical clinics can be provided. furthermore, the platform can provide knowledge and production of 3d printing medical material information to patients and medical clinics. however, the purpose of the present study is focused on the medical assistive device splint production in health care clinics without outsourcing. therefore, we integrated the entire operation data of medical doctors, occupational therapists, and 3d printing engineers. these professionals could comfortably and easily use the platform for job processing and information sharing by using related technologies. in addition, we designed the splint cloud information platform of a 3d printing assistive device to improve the splint manufacturing efficiency. designing a web-based interface that allows the users to get started quickly and promote the usage intention must be investigated. if the system interface is poorly designed, it will reduce the usage intention of the users. therefore, laurel and mountford proposed that human–computer design principles are mainly divided into: (1) user-oriented design. (2) interfaces consistency. (3) graphical representations. (4) multiple views [12]. advances in technology innovation, vol. 4, no. 2, 2019, pp. 73-83 76 li proposed that the expression clearly can enhance the usage intention of the users. besides, before developing the system, the requirements should be clarified [13]. booch proposed that unified modeling language(uml) which is the visualized language for developing software and uses different diagrams to describe the system requirement through different viewpoints [14]. the diagrams of uml can be divided into: (1) use case diagram. (2) class diagram. (3) component diagram. (4) activity diagram. (5) collaboration diagram. (6) sequence diagram. (7) statechart diagram. (8) deployment diagram. uml is using the standard diagrams to simplify the lengthy text statements and allows the system developers, designers and users to understand the cognition of system framework, flow path, and all aspects quickly [15]. 3. research approach fig. 2 the production process of 3d printing splint sequence diagram the present day, there was no jointed managed pipeline for the transmission of the splints production information, so this study cooperated with the medical clinics and learned the current processes and methods of producing 3d printing splint in person in a medical clinic. we used the information technology and cloud technology to solve the previous paper waste, both sides communication, browsing 3d graphics and managing medical data difficultly and other problems. we explored the workflow of producing splints in the hospital in this study. the basic functional items in the workflow for producing the 3d printing rehabilitation splints are: (1) patient data records (2) medical records (3) information about the splint creation (4) upload 3d splint drawing file (5) execution of 3d printing (6) query 3d printing splint information (7) advances in technology innovation, vol. 4, no. 2, 2019, pp. 73-83 77 account management (8) splint category (9) message bulletin (10) permission control. based on the above functions are defined for each item, the users of the system are then clarified with the corresponding various authorities. they are as follows: (1) medical unit doctor. (2) medical unitnurse. (3) medical unit occupational therapist. (4) splint design units3d graphics engine. (5) 3d printing unit 3d printing engineer. (6) platform system manager administrator. in this study, the system demand analysis and design use the uml to explain the platform design, as described as follows. during the initial visit of the patient, the doctor must collect the patient’s basic data. then, the doctor creates a patient treatment record and selects the suitable splint. subsequently, the occupational therapists establish the splint production contents in accordance with the splint specification, which is established by the doctor, and assist in splint production. next, the splint designer starts to draw the 3d graphics of the customization splint based on the splint production contents. following design completion, the 3d graphics are uploaded into the splint version management in the platform. the occupational therapists will check whether the design is a 3d graphics; otherwise, the splint designer must upload again the data after correction until it passes the verification stage. after passing the verification, the splint designer sends the 3d graphics to the 3d printing engineers. the engineers will receive the printing details and produce it. afterward, the generated splint will be checked by the occupational therapists. if the splint is not in accordance with the specifications, it must be returned to the splint designers, who will upload the 3d graphics files until the finished product passes verification. finally, the case is terminated by the occupational therapists. the 3d printing splint production process is shown in fig. 2. 4. platform design fig. 3 forestage platform framework diagram the proposed system platform in this study is divided into two major parts, namely, the forestage system and backstage management. when entering the proposed system, the platform will lead the users to the forestage or backstage screen on the basis of the users’ status. the users of the cloud information platform forestage of the 3d printing rehabilitation device are advances in technology innovation, vol. 4, no. 2, 2019, pp. 73-83 78 doctors, nurses, physical therapists, and 3d graphics and printing engineers. the process structure of the system forestage is shown in fig. 3. there are four functional sections in the forestage, they are “patient list”, “splint production”, “3d printing splint”, and “about the system” respectively. the corresponding web interface is shown as fig. 5. moreover, the system platform administrator has the highest login authority of the backstage. the backstage structure is illustrated in fig. 4, and there are four main functional sections including “splint kind setting”, “account management”, “web page function setting” and “about the system”. the two major functions (forestage and backstage) describe the structure of the proposed cloud information platform, for which all the different roles of users can use the platform to produce the customized splint for each of the patients. fig. 4 backstage framework diagram the system requirements in this study are presented on the web platform. in general, the planning of the platform includes the following five major parts which are described in the forestage and backstage of the cloud information platform as follows: (1) basic data about patient management: all the medical care data of the patient, including the following four functional pages as shown in fig. 5. (a) the basic data of patients: the basic data for recording the patients. (b) treatment record: the treatment records of the patients, including medical information, diagnostic logs out and treated areas for the occupational therapist to know the treatment demands of patients. (c) splint production: records of splint specification and 3d graphics file management for every patient. this information also provides 2d and 3d graphics uploading, browsing, and leaving a message. the details are illustrated in fig. 7. (d) 3d printing splint: 3d printing splint for the 3d printing engineer to receive the 3d model printing information. advances in technology innovation, vol. 4, no. 2, 2019, pp. 73-83 79 (2) categories of splint setting: manage the information of various 3d printing splint. see fig. 6. (3) account management: it is the basic users’ data and the authority in the platform as shown in fig. 8. (4) permission settings of pages and function: for setting the permission of all users as shown in fig. 9. (5) about the system: it is the announcement pages of the internal information and its declaration of the internal information and its records the usage condition of the platform. in the process of making a splint for a patient, all the data are divided into four major items, one of which is the patient's basic information, the second is the patient's medical record, the third is the splint production menu, and the fourth is the 3d printing detailed information. the medical personnel can view the contents of the patient’s case as needed. in terms of interface design, the button conforms to the "representation of content in graphics" and "content must be clear" as mentioned in the literature, as shown in fig. 5. fig. 5 basic data of patient management fig. 6 categories of splint setting this content in fig. 6 is designed as a backstage administrator to manage the information of the 3d printed splint category. from this page, the administrator can edit and add the type, location, laterality, coverage, knuckle range and whether the secondary wood is still in use to provide the front-end medical staff to select the appropriate splint for the patient. when making a secondary wood, it needs to be revised after discussion. a secondary wood production requires several versions. advances in technology innovation, vol. 4, no. 2, 2019, pp. 73-83 80 therefore, this study divides the splint production menu into two parts: the secondary wood production content and the secondary splint production version management, so as to reduce the complexity of the user browsing. the contents of the splint production include the name of the physical therapist responsible for the case, the functional therapist number, the medical institution, the functional therapist's diagnosis, the type and side of the splint, the printing materials of the splint, and the uploaded image of the injured part of the patient, delete, download, and provide image zoom browsing, physical therapist notes, and status display. and we also need to consider the possibility that the patient temporarily changed his/her mind not to use 3d printed splint during the production process. therefore, the “void button” is given to stop the subsequent actions, as shown in fig. 7. fig. 7 production of splint fig. 8 account management advances in technology innovation, vol. 4, no. 2, 2019, pp. 73-83 81 this function page in fig. 8 is a page for the background manager to manage employee data, which is divided into two parts: the employee's basic data and permission settings. employee basic information includes user account information and employee basic information. the account information includes the employee account number, password setting, and consideration of personnel transfer. therefore, the user status field is provided to control whether the user has the right to use the platform, and the corresponding authority is given to the user. the way to query personnel is to query the name, account number, job title, subordinate unit and status, so that the user can quickly search for the required information. since this platform has many patient privacy information, it is necessary to provide permission settings for the administrator to control. therefore, we make the platform users use the functionality of the page to be restricted by permissions. each person's authority is given the corresponding authority according to the set role. the page and function permissions are based on the content of the function, such as basic patient information, medical records, splint production, 3d printed splint, the backstage of the about system, the splint settings, personnel data management, role permissions settings. if there are special functions, the extended permission function is given to the user for additional settings, as shown in fig. 9. fig. 9 permission settings of pages and function 5. conclusions and future researches three-dimensional printing is a digitized manufacturing technology; thus, manufacturing using such technology is made by computer drawing software and then produced by using the 3d printing machine. the assigned personnel need to browse the 3d graphics files in the manufacturing process; however, the personnel must face several unpleasant situations, such as unnecessary resource waste, 3d graphics management, information transmission, and communication without the use of the information platform. it will be very easy to cause the mistakes during the communication and file transmission. therefore, in this study, we introduce the flow of utilizing 3d printing device in hospital, and utilize the cloud technologies and uml to design the information platform and then constructed it. in this study, the cloud information platform for 3d printing assistive device splint can provide browsing functions of 3d graphics, feedback mechanism, and automatic managing function efficiently. this platform can prevent un-necessary wastes of the paperwork in delivery of documents, mistakes caused by browsing 3d graphics with wrong patient, and the difficulties in communication and document management. the proposed platform can also improve the splint production efficiency (see table 2). from table 2, we can see the comparison of before and after while using the proposed information platform. advances in technology innovation, vol. 4, no. 2, 2019, pp. 73-83 82 table 2 comparison table of before and after use the information platform before implementing of information platform after implementing of information platform information transfer method paper network browse 3d graphics method face to face communication / other web-based browse in this study platform communication channel face to face communication / telephone/ messaging software in this study platform data management the manual way of management an automatic way of management in this study, the cloud information platform of the 3d printing rehabilitation device is designed and constructed with the cooperation of many doctors in the medical clinics. in literature, many scholars have confirmed that the use of electronic and cloud technology in medical operations can effectively help medical personnel to file, find, save and operate their daily operations. however, the effectiveness of using the platform in medical clinics should be investigated in the future. and it needs to verify whether the process of introducing 3d printed rehabilitation device in medical care is suitable or not. thus, we aspire that the use of the system can be used in medical clinics in future research and perform further verification and improve the operating efficiency of the splint production. 6. acknowledgments the authors would like to thank all the colleagues and students who contributed to this study. this research is supported by the ministry of science and technology, taiwan, r.o.c. under grant no. most 106-2632-e-468-001. conflicts of interest the authors declare no conflict of interest. references [1] b. r. bryant and p. c. seay, “the technology-related assistance to individuals with disabilities act: relevance to individuals with learning disabilities and their advocates,” journal of learning disabilities, vol. 31, no. 1, pp. 4-15, january 1998. [2] e. e. fess, “a history of splinting: to understand the present, view the past,” journal of hand therapy, vol. 15, no. 2, pp. 97-132, april 2002. [3] b. m. coppard and h. lohman, introduction to splinting-e-book, elsevier health sciences, 2013. [4] t. h. huang, c. k. feng, y. w. gung, m. w. tsai, c. s. chen, and c. l. liu, “optimization analysis of thumbspica splint design,” master’s thesis, department of rehabilitation science and technology, national yang-ming university, 2005. [5] f. s. koopman, m. edelaar, r. slikker, k. reynders, l. h. v. van der woude, and m. j. m. hoozemans, “splinting the hand and upper extremity: principles and process,” lippincott williams & wilkins, baltimore, md, 2003. [6] c. l. chen, y. l. teng, s. z. lou, c. h. lin, f. f. chen, and k. t. yeung, “user satisfaction with orthotic devices and service in taiwan,” plos one, vol. 9, no. 10, october 2014. [7] 3d printing industry (3dpi), “wasp: saving the world with 3d printed splints & cranial implants,” https://3dprintingindustry.com/news/wasp-saving-the-world-with-3d-printed-splints-cranial-implants-49019/, may 14, 2015. [8] c. w. yang and m. t. liu, “mobile nursing information system with learning-on-demand services at taichung veterans general hospital,” proc. computer, consumer and control (is3c), 2012 international symposium on ieee press, june 2012, pp. 618-621. [9] j. tian, j. xue, y. dai, j. chen, and j. zheng, “a novel software platform for medical image processing and analyzing,” ieee transactions on information technology in biomedicine, vol. 12, no. 6, pp. 800-812, may 2008. [10] t. doel, d. i. shakir, r. pratt, m. aertsen, j. moggridge, e. bellon, and s. ourselin, “gift-cloud: a data sharing and collaboration platform for medical imaging research,” computer methods and programs in biomedicine, vol. 139, pp. 181-190, february 2017. [11] m. umair and w. s. kim, “an online 3d printing portal for general and medical fields,” proc. computational intelligence and communication networks (cicn), 2015 international conference, ieee press, march 2015, pp. 278-282. advances in technology innovation, vol. 4, no. 2, 2019, pp. 73-83 83 [12] b. laurel and s. j. mountford, “the art of human-computer interface design,” addison-wesley longman publishing co., inc., 1990. [13] j. w. li, “a study on the user interface for hand-held products: pda as example,” master’s thesis, department of design, dy university, 2003. [14] g. booch, “the unified modeling language user guide, pearson education india,” 2005. [15] w. d. shen, “the campus property management system-analysis, design and implementation using uml,” master’s thesis, department of computer science and information engineering, tamkang university, 2008. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation, vol. 4, no. 1, 2019, pp. 1 10 prestressed stainless steel stayed columns with two crossarms josef machacek * , radek pichal faculty of civil engineering, czech technical university in prague, prague, czech republic received 03 may 2018; received in revised form 09 july 2018; accepted 11 august 2018 abstract the efficiency of prestressed stayed elements when designing very slender steel columns was proved for both stability and strength capacity of the columns. with an increasing number of crossarms placed along the length of the column the effectiveness is further growing in comparison to a stayed column with just one crossarm. first the stability of an “ideal” (perfect) prestressed stayed column with two crossarms is investigated analytically. similar principal behavior as in the case of prestressed stayed columns with just one crossarm is confirmed and the three zones depending on the value of prestressing are revealed. expressions for minimal, optimal and maximal prestress are derived analytically. after receiving critical buckling value of the stayed column with the two crossarms without any prestressing by linear buckling analysis (lba), the maximal critical buckling loading of the column under optimal prestressing is derived. the results are fully demonstrated on a practical example. second the strength capacity of such column but with relevant initial deflections is investigated by the geometrically and materially nonlinear analysis with imperfections (gmnia) using ansys software. comparisons of critical and strength values under various prestressing are analyzed with respect to a practical design. finally some recommendations for following studies and practical use are given. keywords: stayed columns, two crossarms, stainless steel, prestressing, nonlinear buckling 1. introduction extremely slender columns suffer with a low strength capacity due to buckling. a smart solution of the problem provide prestressed stayed steel elements. the crossarms connected by prestressed cables or rod stays with the column ends rapidly increase both the critical column load and its collapse capacity, see fig. 1. (a) grande arche, paris (b) parc del centre del poblenou,barcelona (b) estádio algarve, faro fig. 1 examples of structures using stayed columns * corresponding author. e-mail address: machacek@fsv.cvut.cz tel.: +420776562811 http://www.google.co.uk/url?sa=i&rct=j&q=&esrc=s&source=images&cd=&cad=rja&uact=8&ved=0ahukewjbrq7x--xmahxjarqkhvkoa0uqjrwibw&url=http://www.panoramio.com/photo/23176377&psig=afqjcngtsigt8tuugzpmiuutij4dvvqfvw&ust=1463741361896792 advances in technology innovation, vol. 4, no. 1, 2019, pp. 1 10 2 the prestressed stayed columns with just one crossarm were deeply investigated analytically, numerically and experimentally within several last decades. among others, the milestones were achieved by smith et al. [1] and hafez et al. [2], who discovered the three zones of behavior depending on the prestressing level of the stays and wadee et al. [3], who investigated the maximal load capacity. these results are roughly demonstrated in fig. 2, where the “zones” may be explained as follows: zone 1 (up to tmin), where the prestressing in the stays disappears when the applied load is less or equal to the euler load (ncr = ne); zone 2 (up to an optimal prestressing topt), where the stays remain effective until the applied load triggers a buckling; zone 3 (above topt), where all the stays remain active (in tension) even after the buckling. higher prestressing than topt increases the column loading and, therefore, decreases the critical column load ncr. n n stays a , e crossarm a , i , e s s a a a column a , i , ec c c x z y t t t t a aa a l approx. m aximum c apacity n n [kn] t t 3toptmin n cr,max n = n cr,min e opt t ~ 0 ,4 t o p t approx. m aximum c apacity n z o n e 1 z o n e 2 z o n e 3a b cr critical load n by hafez et al.cr approx. m aximum c apacity n by wadee et al. approx. m aximum c apacity nmax (a) planar (b) spatial (c) relationship of critical load vs prestressing fig. 2 stayed columns with one central crossarm deep investigations covering critical values, initial imperfections or maximal capacity of the stayed columns with just one crossarm were provided by wong and temple [4], chan et al. [5], saito and wadee [6], [7], experimental and numerical investigations by araujo et al. [8], servitova and machacek [9], lima et al. [10], osofero et al. [11], ribeiro et al. [12], serra et al. [12] and were mostly commented also by the authors in [14], [15], [16], [17]. 1 6 4 m a a a a stays a , e crossarms a , i , e s s a a a column a , i , e c c c l a a  n n a a a a a a ti ti titi ti ti ti ti (a) mast according to vojevodin (b) double stayed prestressed column (c) geometry of the column fig. 3 stayed columns with two crossarms nevertheless, both researchers and designers were aware of a greater capacity of the prestressed stayed columns with more than one crossarm since the origin of the studies (see several masts designed by vojevodin [18], fig. 3). while this knowledge was supported by numerical analysis and some tests (e.g. khosla [19], jemah and williams [20]), the more deep investigation was performed in the last years only. martins et al. [21] tested double crossarm stayed tube columns with a total length of 18 m, two different column cross-section diameters and various prestressing. the tests provided valuable experimental data for axial shortening and lateral deflections under loading. yu and wadee [22] investigated numerically triple crossarm stayed columns using abaqus software. apart from varying the entry data concerning ratio of the column length to crossarms, diameter of cable stays and value of prestressing, the “efficiency indicators” were used to optimize the total design advances in technology innovation, vol. 4, no. 1, 2019, pp. 1 10 3 and optimal prestressing. for the same columns these authors [23] further developed a nonlinear analytical model verified by abaqus nonlinear analysis, studied parametrically buckling modes and drew attention to the possibility of a mode jumping. lapira et al. [24] analyzed triple crossarm prestressed stayed columns and also columns with additional stay system located in the middle quarters of the system. they derived analytical formulas in 3d for maximal critical values, corresponding optimal and any other prestressing, which verified by fem using abaqus software. in this paper the authors follow-up the mentioned previous investigations concerning stayed columns with just one crossarm and analyze 3d prestressed stainless steel stayed columns with two crossarms in accordance with fig. 3, both analytically for “ideal” (perfect) column and numerically for “ideal” or imperfect column using ansys software. 2. stability behavior of an “ideal” prestressed stayed columns with two crossarms the three approaches leading to the determination of critical loads are presented. first the geometrical analysis based on a geometry and equilibrium conditions, second linear buckling analysis (which however, as shown later, can be used for unprestressed columns only) and third the finite element method (fem) using geometrically nonlinear analysis with imperfections (gnia). 2.1. geometrical analysis first the “ideal” (perfect) column is analyzed in the similar way as was done for just one single crossarm by smith et al. [1] and hafez et al. [2]. the geometrical analysis of a stayed column in accordance with fig. 3 is based on a change of element lengths due to axial deformation at the instant of buckling. the fundamental assumptions of the derivation are: the column is perfectly straight and concentrically loaded. the connections between column and crossarms are assumed to be rigid and between the stays and column/crossarms are assumed as ideal hinges. the maximal buckling load of the central column is assumed to be obtained by linear buckling analysis (lba) for the column without any prestressing. this analysis assumes the stays active both in tension and compression, while neglecting buckling of stays. the axial deformations of the column and crossarms are not considered in lba, however, need to be considered for analytical derivation of the magnitude of tension in the stays. the changes of element lengths at the instant of buckling are shown in fig. 4. the shortening of the outside and central stays gives: 1 2 6 6 3 c c c se ca c cacos ( tg ) sin cos sin cos                    and 3 c sm    (1) a ti ti sina cosa ti (a) axial changes (b) outside stay element (c) force resolution fig. 4 axial compression changes of the column and force resolution in the crossarm advances in technology innovation, vol. 4, no. 1, 2019, pp. 1 10 4 the spatial column has 4 stays (n = 4), the planar one 2 stays (n = 2). the initial axial force in the column, ni, induced by the initial stay pretension, ti, and final one, nf, after the application of the external load, na, are: i in nt cos and 4f a f a fn n nt cos n t cos     (2) the stiffness of the column, kc, of the crossarm member, kca, and of the stay, ks, are: l ea k cc c  ; ca caac ca l ea k  ; s ss s l ea k  (3) after evaluation of the shortening of the column, elongation of the crossarm member and shortening of the stay, the decrease of the tension in the stays and substitution of these values in eq. (1) gives: 12 2 1 3 3 a i f a ,n c s c ca n cos t t n c n cos sin k k k k               (4) after some substitutions the final tension in the stays, nf, may be written as:   2 1 3 f i f i c s ca n nt cos cos t t sin k k k              (5) for the applied (external) load, na, the substitution into eq. (2) yields:     2 22 1 1 3 a f i f i ,n c s ca n cos n n nt cos n nt cos c sin k k k                          (6) using these formulas, the behavior of the stayed column with the two crossarms may be described similarly as behavior of the column with just one crossarm. in fig. 5 the resulting three zones of behavior with respect to value of prestressing are shown. also, for a column with two crossarms analyzed in following paragraphs (see the entry data given in paragraph 2.2) the buckling modes and critical loads of unprestressed column are shown. n [kn] t t 3toptmin n cr,max n = n cr,min e opt z o n e 1 zone 2 zone 3 cr n cr, t=0 (a) initial pretension (b) ncr,1,t=0 = 44.43 kn ncr,2,t=0 = 72.12 kn ncr,3,t=0 = 91.90 kn fig. 5 initial pretension vs buckling load and example of the first three buckling modes corresponding to the later analyzed column using lba zone 1: the initial prestressing is small (tf ≤ tmin) and after external loading disappears. the column behaves as unstayed and buckles at the euler load, ne. minimal prestressing, tmin, corresponding to this behavior, follows from eq. (4): advances in technology innovation, vol. 4, no. 1, 2019, pp. 1 10 5 2 2 l ie nn cc ecr   ; n cc nenai c l ie cncntt ,12 2 ,1,1min   (7) zone 2: the prestressing is larger than the minimal one but smaller than the optimal one, tmin < tf < topt. after triggering of buckling the prestressing in the stays disappears (tf = 0), but the stays at convex side will become active immediately. the resulting critical load, ncr, will be higher than euler’s one and according to eq. (4) will correspond to prestressing ti: n i acr c t nn ,1  (8) maximal critical load, ncr,max, follows from eq. (6), after substituting for na = ncr,n,t=0, where the latter is the value of critical load received from lba for fully active stays without any prestressing (ti = 0), see fig. 5: n, t,n,cr fmax,cr c n nn 2 0  (9) corresponding optimal prestressing is given by: ncropt cnt ,1max, (10) zone 3: the prestressing is larger than the optimal one. after buckling the stays remain in a tension and the critical load falls off due to respective force components in the stays. the residual tension in the stays follows from eqs. (5) and (9) as:    ,max ,max 32 cos cos cos 1 sin 3 r i cr i i cr i c s ca t t n nt t n nt c k k k                  (11) the critical load in zone 3 can be obtained from eq. (6) after substituting for na = ncr and nf = ncr,max:   nicrcr ctnnn ,2max, cos (12) 2.2. numerical stability analysis using 3d gnia the analytical investigation of the critical load based on geometrical analysis of an “ideal” (perfect) central column may be verified by using fem. in the field of prestressed elements the stability behavior in the full range of prestressing can’t be solved by linear buckling analysis (lba) but due to the sudden change of inner energy at the instant of buckling the geometrically nonlinear analysis with imperfections (gnia) is necessary (see, saito and wadee [5] or previous articles of the authors, e.g. [15], [17]). under a small prestressing the stays at buckling become slacked and the ones at convex side are immediately activated (zones 1 and 2). if all the stays, both at convex and concave sides are activated, the column behavior corresponds to the zone 3. the actual behavior of the stayed column with the two crossarms was therefore investigated by using ansys software in 3d and geometrically nonlinear analysis with imperfections (gnia). the results are demonstrated on the same example as analyzed by the authors in [16], however, now with the two crossarms instead of one crossarm. the entry data (length, area, second moment of area, modulus of elasticity) are as follows, see fig. 3: central stainless steel tube ø 50x2 [mm]: l = 5000 mm, ac = 302 mm2, ic = 87009 mm4, ec = 200 gpa. crossarm stainless steel tubes ø 25x1.5 [mm]: a = 250 mm, aa = 111 mm2, ia = 7676 mm4, ea = 200 gpa. advances in technology innovation, vol. 4, no. 1, 2019, pp. 1 10 6 stays as macalloy cables 1x19 stainless steel ø 4 mm: ls = 2513 mm, as = 12.6 mm2, es = 200 gpa. substituting these values into the formulas presented in chapter 2.1 yields: the invariables: c1,4 = 0.0351, c2,4 = 1.1612. critical loads and prestressing (using ncr,1,t=0 = 44.43 kn from lba, see fig. 5): ne = 6870 n; tmin = 241 n; ncr,max = 38262 n; topt = 1343 n; ncr,3topt = 25925 n. the ansys model involved beam188 elements for the central and crossarm tubes and link180 elements for the cable stays, with the same boundary conditions as mentioned in the chapter 2.1 (connections between column and crossarms are assumed to be rigid, between stays and column/crossarms are ideal hinges). the meshing study resulted into division of l/250 and a/25 as satisfactory one. first, the required initial deflections were introduced, followed by the relevant prestressing of stays through their thermal change (i.e. by cooling). finally, axial deflections to the central column (giving an external load) up to collapse were imposed. a standard newton-raphson iteration was used. to verify the analytical values, first gnia was used, with elastic material behavior, followed by gmnia for stainless steel material. for the stability analysis the “ideal” column was analyzed, the initial imperfections need to be negligible, therefore infinitesimal deflections were employed with the amplitude of w0 = l/500000 = 0.01 mm (in the 3d as w0y = w0z = w0/√2) for both symmetric and antisymmetric modes, see fig. 6. n n x z y l w 0 w /2 0 w /2 0 (a) stayed column (b) initial deflections (c) gnia results fig. 6 initial deflections w0 = l/500000 for “ideal” column and gnia results (identical for both initial deflection modes) the resulting load-prestressing relationship received from the gnia is presented in fig. 6. the curves for critical and maximal loading were received from rather lengthy numerical calculations of 26 prestressing values (in reality cooling of the stays), each with 1000 of compression steps. in low prestressing, roughly up to 1.6 kn, when at the instant of attaining the critical load (i.e. when the tension in all stays become zero) the both stays at the convex side become active in tension (while at concave side being slacked) and the maximal (capacity) load of the “ideal” column becomes higher than the critical one. with higher value of prestressing the all four stays remain in some tension (zone 3) and both critical and maximal values are identical (show also the explanation in fig. 8 for gmnia). comparison of analytical and numerical (gnia) values is rather difficult (see fig. 6) but the analytical values, i.e. ncr.analyt = 38262 n at prestressing of topt,analyt = 1343 n, may be compared with the gnia ones when the stays on the concave side at the buckling don‘t slacken and critical and maximal loads merge together, i.e. ncr,gnia = 38200 n at prestressing of topt,gnia = 1599 n, being acceptable. nevertheless, gnia maximal load for “ideal” column of roughly 52618 n arises at prestressing of 9808 n. it should be noted, that deflected shape for both modes of negligible initial deflection are identical, after ncr = 47852 n first interactive (in combination of symmetric and antisymmetric mode), after prestressing of t = 9808 n becomes antisymmetrical. advances in technology innovation, vol. 4, no. 1, 2019, pp. 1 10 7 3. maximal loading of an imperfect prestressed stayed columns with two crossarms the investigation of a maximal loading (strength capacity) of a compression element requires the modelling of a real, imperfect column. the initial imperfections in such stayed columns were introduced in accordance with eurocode en 1993-1-1 as equivalent initial deflections with amplitude w0 = l/200 = 25 mm (valid for cold-formed thin-walled tubes and elastic analysis). the deflection was considered again in the spatial direction, i.e. with w0y = w0z = w0/√2. the first two modes of initial deflections were considered only, in accordance with figure 5: antisymmetrical with the amplitude w0,l/2 = l/400 and symmetrical with the amplitude w0 = l/200. l l (a) antisymmetrical initial deflection (b) symmetrical initial deflection fig. 7 gnia critical and maximal capacity for the initial deflection w0 = l/200 the gnia for the above analyzed example with the same stainless steel elastic modulus (according to en 1993-1-4) e = 200 gpa resulted in various prestressing into values presented in fig. 7. as in the lba the decisive mode proved to be the antisymmetric one, giving maximal loading nmax = 31988 n at prestressing of ti = 8353 n. another study employed gmnia (geometrically and materially nonlinear analysis with imperfections), differing just in using the nonlinear material behavior as tested by the authors in [16], resulting in the stress-strain relationship acc. to fig. 8. (a) stress-strain relationship of the stainless steel material (b) load-axial deflection curve for prestressing of ti = 1278 n (c) corresponding tension fig. 8 example of forces in the stays 0.002 0.004 0.006 0.008 200 300 400 500 0 0 100 test [mpa]  = 434.1 mpa ansys 0.2 e = 184.0 gpa in e e1 2 e n e 3 e 4 e 5 advances in technology innovation, vol. 4, no. 1, 2019, pp. 1 10 8 again both the “ideal” (perfect) central columns with infinitesimal amplitude of initial deflections w0 = l/500000 = 0.01 mm were analyzed (an example of one analysis under initial prestressing of ti = 1278 n is shown for illustration in fig. 8). the results for the critical loads (concerning “ideal” stayed columns) for antisymmetrical initial deflections now partly differ from results in the same column but with symmetrical initial deflections, show in fig. 9. nevertheless, the important value of ncr,gmnia = 34429 n at prestressing of topt,gmnia = 1583 n corresponds well with the former gnia ncr,gnia = 38000 n at prestressing topt,gnia = 1599 n, considering the ratio of the elastic moduli in gnia/gmnia as ein/e = 184/210 = 0.92. a simple reduction using this factor gives ncr,gnia,e=184 mpa = 38000·0.92 = 34960 n which differ from ncr,gmnia = 34429 n by 1.5 %. (a) antisymmetrical initial deflection (b) symmetrical initial deflection fig. 9 gmnia results for the “ideal” column with initial deflections w0 = l/500000 the gmnia results of the real column with an amplitude of initial deflection of w0 = l/200 = 25 mm are shown in fig. 10. in comparison with gnia (fig. 7) the maximal (capacity) loadings are significantly lower due to nonlinear stress-strain relationship of the stainless steel material. the decrease of the maximal loading for decisive antisymmetrical initial deflection mode from 31988 n to 25040 n gives 27.7 %, while for symmetric initial deflection mode the drop from 41544 n to 34000 n is 22.2 %. l l (a) antisymmetrical initial deflection (b) symmetrical initial deflection fig. 10 gmnia critical and maximal capacity for the initial deflection w0 = l/200 4. conclusions the paper investigates prestressed stainless steel stayed columns with two crossarms, located at the thirds of the central column length. in the first part the analytical stability of an “ideal” column is investigated. based on 2d geometrical analysis the behavior of the column is described for an arbitrary prestressing and general geometry. similarly as in the case of the stayed columns with just one central crossarm the three zones according to the extent of prestressing are revealed and formulas for minimal and maximal critical loadings in 2d and 3d together with “optimal” prestressing are derived. the resulting formulas are applied to a practical example, which has all the geometric and material characteristics the same as for the stayed column which was investigated both experimentally and theoretically by the authors in the past, presented e.g. in [17]. the only difference consists in the two crossarms in the thirds of the column length instead of just one central crossarm. 0 5000 10000 15000 20000 25000 30000 35000 40000 45000 50000 0 2000 4000 6000 8000 10000 critical load (w0 = l/500000) maximal load (w0 = l/500000) stay prestressing [n] e xt e rn a l l o a d [ n ] 34429 n 1 5 8 3 n 0 5000 10000 15000 20000 25000 30000 35000 40000 45000 50000 0 2000 4000 6000 8000 10000 critical load (w0 = l/500000) maximal load (w0 = l/500000) stay prestressing [n]stay prestressing [n] 1 6 6 1 n e xt e rn a l l o a d [ n ] 35000 n advances in technology innovation, vol. 4, no. 1, 2019, pp. 1 10 9 numerical analysis started with a linear buckling analysis (lba) under zero prestressing and stays active in compression, giving critical loadings and deflection modes required for the use in the above analytical solution. subsequently a 3d model in ansys software was prepared to analyze both the stability of “ideal” (perfect) column and maximal loading (column capacity) of the “real” (imperfect) stayed column. the “no compression option” for the stays was adopted to simulate any slackening of these elements and respective initial deflections were introduced: infinitesimal ones for critical loadings in the case of “ideal” column (l/500000) and required ones according to eurocode 1993-1-1 for imperfect column (l/200 for thin-walled cold-formed members). the initial deflection modes were introduced in accord with lba either as antisymmetrical one with two half sine waves or symmetrical one with the full sine wave. the main results of the investigation may be summarized as follows: (a) the analytical formulas for the double crossarm prestressed stayed columns concerning minimal and optimal prestressing giving maximal critical analytical loading were derived. (b) numerical modelling (using ansys software) of the column with a practical size was presented for a verification of the analytical approach. it was shown, that the stability behavior of an “ideal” prestressed column is more complex in comparison to the simple analytical approach and a postcritical behavior even for “ideal” column need to be evaluated. nevertheless, the analytical value of the maximal critical loading was well comparable with the numerical one for the prestressing when the stays on the concave side at buckling do not slacken. (c) numerical analysis using gnia with the initial deflections of the main column corresponding to thin-walled cold-formed tubes (with amplitude l/200) gives significantly lower maximal (capacity) loading in comparison with the critical loading. this decrease in the comparison with analytical critical loading in the specified shown example is at 84 % and in comparison to maximal loading of “ideal” column is roughly at 61 %. (d) considering gmnia with the stainless steel material nonlinearity leads to even greater reduction in comparison to the former values. the decrease of maximal (capacity) loading due to initial lower young’s modulus and due to material nonlinearity is another 27.7 %. (e) finally the comparison of effectivity between the stayed column with just one crossarm (see [17]) and the column with the two cross arms (with the above specified geometry otherwise unchanged) in the gnia shows due to adding the second crossarm increase of both critical loading of “ideal” column and maximal loading of imperfect column. the increase of the critical loading is 21.0 % (from 31580 n to 38200 n) and increase of the maximal loading is 63.5 % (from 19570 n to 31988 n). although the presented percentage values are perfectly valid just for the specific geometry only, the higher efficiency of the two crossarms is undisputable and may provide designers with a guidance for more economical and reliable design. acknowledgement support of the czech grant agency grant gacr no. 17-24769s is gratefully acknowledged. references [1] r. j. smith, g. t. mccaffrey, and j. s. ellis, “buckling of a single-crossarm stayed column,” journal of the structural division asce, pp. 249-268, january 1975. [2] h. h. hafez, m. c. temple, and j. s. ellis, “pretensioning of single-crossarm stayed columns,” journal of the structural division asce, pp. 359-375, february 1979. [3] m. a. wadee, l. gardner, and a. i. osofero, “design of prestressed stayed columns,” journal of constructional steel research, vol. 80, pp. 287-298, janury 2013. advances in technology innovation, vol. 4, no. 1, 2019, pp. 1 10 10 [4] k. s. wong and m. c. temple, “stayed columns with initial imperfections,” journal of the structural division asce, pp. 1623-1640, 1982. [5] s. l. chan, g. p. shu, and z. t. lü, “stability analysis and parametric study of pre-stressed stayed columns,” engineering structures, vol. 24, no. 1, pp. 115-124, january 2002. [6] d. saito and m. a. wadee, “post-buckling behaviour of prestressed steel stayed columns,” engineering structures, vol. 30, no. 5, pp. 1224-1239, may 2008. [7] d. saito and m. a. wadee, “numerical studies of interactive buckling in prestressed steel stayed columns,” engineering structures, vol. 31, no. 2, pp. 432-443, february 2009. [8] r. r. araujo, s. a. l. andrade, p. d. s. vellasco, j. g. s. silva, and l. r. o. lima, “experimental and numerical assessment of stayed steel columns,” journal of constructional steel research, vol. 64, no. 9, pp. 1020-1029, september 2008. [9] k. servitova and j. machacek, “analysis of stainless steel stayed columns,” proc. 6th intern. symp. steel structures, korean society of steel construction, seoul, 2011, pp. 874-881. [10] l. r. o. lima, p. c. g. vellasco, and j. g. s. silva, “numerical modelling of prestressed stayed stainless steel columns,” tubular structures xiv, taylor and francis, september 2012, london, pp. 377-382. [11] a. i. osofero, m. a. wadee, and l. gardner, “experimental study of critical and post-buckling behaviour of prestressed stayed steel columns,” journal of constructional steel research, vol. 79, pp. 226-241, december 2012. [12] d. m. ribeiro, p. c. g. s. vellasco, r. r. araujo, l. r. o. lima, and a. t. silva, “numerical assessment of stayed austenitic steel columns,” proc. of eurosteel 2017, copenhagen, september 2017. [13] m. serra, a. shahbazian, l. s. silva, l. marques, c. rebelo, and p. c. g. s. vellasco, “a full scale experimental study of prestressed stayed columns,” engineering structures, vol. 100, pp. 490-510, october 2015. [14] r. pichal and j. machacek, “stability of stainless steel prestressed stayed columns,” proc. 21 st international. conference engineering mechanics 2015, svratka, czech republic may 2015, pp. 230-231. [15] r. pichal and j. machacek, “3d stability of prestressed stayed columns,” proc. 22 nd intern. conf. engineering mechanics, may 2016, pp. 462-465. [16] r. pichal and j. machacek, “buckling and post-buckling of prestressed stainless steel stayed columns,” engineering structures and technologies, vol. 9, no. 2, pp. 63-69, june 2017. [17] r. pichal and j. machacek, “single-crossarm stainless steel stayed columns,” advances in technology innovation, vol. 3, no.1, pp. 9-16, 2018. [18] a. a. vojevodin, “experience from prestressed mast structures (article in russian: opyt iz sooruzenija shprengelnych macht),” vestnik svjazi, no. 5, 1956. [19] c. m. khosla, “buckling loads of stayed columns using the finite element method,” electronic theses and dissertations, dept. civil and env. eng., university of windsor, windsor, on, 1975. [20] a. k. jemah and f. w. williams, “parametric experiments and stayed columns with slender bipods,” international. journal of mechanical sciences, vol. 32, no. 2, pp. 83-100, 1990. [21] j. p. martins, a. shahbazian, l. s. silva, c. rebelo, and r. simoes, “structural behavior of prestressed stayed columns with single and double cross-arms using normal and high strength steel,” archives of civil and mechanical engineering, vol. 16, no. 4, pp. 618-633, september 2016. [22] j. yu and m. a. wadee, “optimal prestressing of triple-bay prestressed stayed columns,” structures, vol. 12, pp. 132-144, november 2017. [23] j. yu and m. a. wadee, “mode interaction in triple-bay prestressed stayed columns,” international journal of non-linear mechanics, vol. 88, pp. 47-66, january 2017. [24] l. lapira, m. a. wadee, and l. gardner, “stability of multiple-crossarm prestressed stayed columns with additional stay systems,” structures, vol. 12, pp. 227-241, november 2017. [25] r. pichal and j. machacek, “buckling and postbuckling behaviour of stainless steel stayed double crossarm prestressed compression elements,” proc. 24th internatioal. conf. engineering mechanics 2018, svratka, may 2018, pp. 681-684. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation, vol. 4, no. 2, 2019, pp. 105-115 analysis of 2017 gartner’s three megatrends to thrive the disruptive business, technology trends 2008-2016, dynamic capabilities of vuca and foresight leadership tools kaivo-oja jari 1 , theresa lauraéus 2,* 1 ffrc, tse, department of future, university of turku, rehtorinpellonkatu 3, 20014 turun yliopisto, turku, finland 2 department of information and services, aalto university, runeberginkatu 22-24, 00076 aalto, helsinki, finland received 03 may 2018; received in revised form 05 august 2018; accepted 22 september 2018 abstract nowadays, digitalization is the key element of business competition. this paper analyzes the concept of dynamic capabilities in the context of technological and business digitalization. we investigate the dynamic competence needed to create and manage a new digital business, which is emerging from the technological transformation. firstly, in this paper, we analyze the data of gartner hype cycles 2008-2017. thus, we present a comparative analysis of the changes in the gartner hype cycle. secondly, our aim is to present gartner ś three distinct megatrends. we will present a summary of these gartner evaluations and discuss key tendencies and trends of technological changes. thirdly, the special focus of this article is the challenges of orchestration of dynamic capabilities in the special conditions of vuca business disruptive business competition. further, we define the role of competence gap identification inside a firm. finally, we are presenting some useful tools to manage dynamic capabilities. keywords: digitalization, three digitalization megatrends, vuca business environment, digital business management, dynamic capabilities, business leadership tools, gartner ś hype curve and digital technology trends, the foresight management tools 1. introduction we will present the foresight tools for innovation, knowledge and corporate management. the management tools for corporate innovation, foresight, which help leaders to manage vuca into consideration, especially in the conditions of hyper-competition and technological digitalization. in this paper, the authors link this discussion to the vuca approach. the authors note that if corporations create corporate strategies with old-fashioned approaches without the vuca approach, most “strategic plans” are having low value-added. in many cases, corporate foresight is ineffective and unsatisfactory for leaders. in this paper, the author presents a new corporate foresight framework, which is more relevant for corporations and takes current technology transformation more seriously. they also present some management tools of foresight, which help leaders manage volatility, uncertainty, complexity, and ambiguity into consideration, especially in the conditions of hyper-competition and technological disruption. key issues in modern vuca management are agility (a response to volatility), information knowledge management (a response to uncertainty), restructuring (a response to complexity), and experimentation (a response to ambiguity). 2. theory 2.1. disruptive technologies the importance of new technologies for society arises from the discovery that ideas and their implementation generate growth and well-being [1]. how can managers know if the technology will disrupt their organization and firm? the definition * corresponding author. e-mail address: theresa.lauraeus@aalto.fi advances in technology innovation, vol. 4, no. 2, 2019, pp. 105-115 106 of disruptive technology relates closely to the disruptive innovation concept of christensen (1997) [2]. bower and christensen, (1995) [3] described a new idea that has long affected the considerations of business sustainability: the notion that new technologies can create new markets, radically change, or disrupt, the status quo in existing markets. useful foresight tools are: challenging tools, decision-making tools, aligning tools, learning tools, and the ability to combine these management tools in the practices of corporate management and leadership. the general conclusion of the paper is that corporate leaders must reinvent strategic planning, framing it to the vuca conditions and simply-be more strategic. especially the technological transformation with artificial intelligence (ai) and robotics is changing many basic assumptions of business management and strategic planning. we have known for a while that disruptive technologies [3, 4, 5] fundamentally change the ways people live and work and how businesses operate, and disruptive technological changes ultimately affect the global economy. existing disruptive innovation theory [2, 6] focuses on key issues like market characteristics, new markets, and low-end innovations. disruptive innovations [6] are new services and products that initially gain a market share at the bottom end of the market by making a product or service available to a new group of “low-end” consumers. they are less wealthy or skilled than consumers of such products have been historically. one definition of a disruptive innovation [4, 6, 7-11] focuses on the functional quality and cost of innovation. this definition defines disruptive innovations as an innovation with a “good enough” functionality that has a low cost [4, 6, 7-11]. theoretically, the lower quality and lower priced innovation incrementally [3, 6] improve until eventually, the innovation competes with market-leading products, thus strongly disrupting the market status quo [3, 6]. the other definition of disruptive innovations does not focus on an innovation's cost or quality, but on market characteristics. 2.2. gartner ś hype cycle phases the hype cycle for emerging technologies report is the longest-running annual gartner hype cycle [12-21] providing a cross-industry perspective on the technologies and trends that business strategists, chief innovation officers, r&d leaders, entrepreneurs, global market develops, and emerging-technology teams should consider in developing emerging-technology portfolios. the theories behind the hype-cycle, fenn and raskino [22] argue that three human nature phenomena are responsible for the curve’s shape: attraction to novelty, social contagion, and heuristic attitude in decision making. adamuthe, tomke & thampi [23] study among the others, the description of hype-cycle phases [24] given by j. fenn [24]: 1. technology/innovation trigger phase: a breakthrough, public demonstration, product launch or other event generates significant press and industry interest. 2. peak of inflated expectations phase: during this phase of over-enthusiasm and unrealistic projections, a flurry of well-publicized activity by technology leaders results in some successes but more failures as the technology is pushed to its limits. the only enterprises making money are conference organizers and magazine publishers. 3. trough of disillusionment phase: because the technology does not live up to its over inflated expectations, it rapidly becomes unfashionable and the press abandons the topic. 4. slope of enlightenment phase: focused experimentation and solid hard work by an increasingly diverse range of organizations lead to a true understanding of the technology’s applicability, risks, and benefits. commercial off-the-shelf methodologies and tools become available to ease the development process. 5. plateau of productivity phase: the real-world benefits of the technology are demonstrated and accepted. tools and methodologies are increasingly stable as they enter their second and third generation. the final height of the plateau varies whether the technology is broadly applicable or benefits only a niche market. http://www.gartner.com/technology/research/methodologies/hype-cycle.jsp advances in technology innovation, vol. 4, no. 2, 2019, pp. 105-115 107 fig. 1 hype cycle for emerging technologies 2016 [20] 2.3. the use of life cycle measures how should organizations and individuals use the hype cycle and deal with the expectations of novel disruptive innovations? [22] first of all, a company can find individual product opportunities. secondly, through segmentation analysis [22], it can build a deeper understanding of its customers that revealed significant opportunities. a company can know the special “golden” shopper segments. thirdly, organizations can respond effectively to competition [22, 25]. and finally, a company is able to identify and introduce new businesses. the hype cycle helps explain why people adopt, abandon or ignore innovations. too many organizations are tempted to jump into innovation prematurely, at the phase of peak of inflated expectations [22, 25], while others wait too long. adopting innovation without understanding the hype cycle [22, 26] can not lead to good decisions and waste of r&d money. 3. results 3.1. understanding the dynamics of technological disruption: comparing the gartner ś hype cycle between years 2008 and 2016 in year 2008, the hype cycle includes: erasable paper printing system, context delivery architecture, behavioral economics, mobile robotics, augmented reality, surface, computers, cloud computing, 3d printing, microblogging, green it, social computing platforms, solid-state drives, public virtual world, web 2.0, service-oriented business applications, virtual assistance, rfid, corporate blogging, idea management, social network analysis, electronic paper, tablet pc, soa, location-aware applications, basic web applications. in the year 2016: the 16 new technologies included in the hype cycle for the first time. these technologies include 4d printing, general-purpose machine intelligence, 802.11ax, context brokering, neuromorphic hardware, data broker paas (dbrpaas), personal analytics, smart workspace, smart data discovery, commercial uavs (drones), machine learning, nanotube electronics, software-defined anything (sdx), enterprise taxonomy and ontology management, blockchain, connected home. in addition, some earlier items stay: the volumetric displays, brain-computer interface, visual personal assistants, affective computers, iot platforms, gestore control devices, micro data centers, smart robotics, machine learning, autonomous vehicles, natural-language question answering, augmented reality, virtual reality. the gartner [27]. sees leading revolution is the smart machine technologies, which will revolutionize manufacturing and its related industries include the following: smart dust, machine learning, virtual personal assistants, cognitive expert advances in technology innovation, vol. 4, no. 2, 2019, pp. 105-115 108 advisors, smart data discovery, smart workspace, conversational user interfaces, smart robots, commercial uavs (drones), autonomous vehicles, natural-language question answering, personal analytics, enterprise taxonomy and ontology management, data broker paas (dbrpaas), and context brokering. emerging technologies are enabling entirely new business models, driving a platform revolution. platform-enabling technologies making new business models possible, include neuromorphic hardware, quantum computing, blockchain, iot platform, software-defined security and software-defined anything (sdx). summary: some technologies of the year 2008 hype cycle were foresighted very well and they are nowadays very popular: tablet pcs, internet of things, cloud services, 3d printers even in the libraries of finland, solid state drive is well foresight, and also augmented reality is coming fast. in the year 2016: fourteen technologies were taken off the hype cycle including hybrid cloud computing, consumer 3d printing, enterprise 3d printing, and speech-to-speech translation because they are not hyped anymore. additional technologies removed from the hype cycle include 3d bioprinting systems for organ transplant, advanced analytics with self-service delivery, bioacoustic sensing, citizen data science, digital dexterity, digital security, internet of things, neuro business, people-literate technology. summary: some technologies of the year 2008 hype cycle were foresighted very well and they are nowadays very popular: tablet pcs, internet of things, cloud services, 3d printers, even in the libraries of finland, solid state drive is well foresight, and also augmented reality is coming fast. in the year 2016: fourteen technologies were taken off the hype cycle including hybrid cloud computing, consumer 3d printing, and enterprise 3d printing, and speech-to-speech translation because they are not hyped anymore. additional technologies removed from the hype cycle include 3d bioprinting systems for organ transplant, advanced analytics with self-service delivery, bioacoustic sensing, citizen data science, digital dexterity, digital security, internet of things, neuro business, people-literate technology. 3.2. gartner [27] identifies three megatrends that will drive digital business into the next decade fig. 2 hype cycle for emerging technologies, 2017 [21]. note: paas = platform as a service; uavs = unmanned aerial vehicles the first megatrend is artificial intelligence (ai) everywhere. it means transparently immersive experiences and digital platforms are the trends that will provide unrivaled intelligence, create profoundly new experiences and offer platforms that allow organizations to connect with new digital business environments. http://www.gartner.com/smarterwithgartner/the-disruptive-power-of-artificial-intelligence/ http://www.gartner.com/smarterwithgartner/transform-business-outcomes-with-immersive-technology/ http://www.gartner.com/smarterwithgartner/enterprise-architects-define-digital-platforms/ http://www.gartner.com/smarterwithgartner/enterprise-architects-define-digital-platforms/ http://www.gartner.com/smarterwithgartner/8-dimensions-of-business-ecosystems/ advances in technology innovation, vol. 4, no. 2, 2019, pp. 105-115 109 the emerging technologies of the hype cycle is the longest-running annual gartner hype cycle [21]. it provides an important cross-industry perspective on the digital environment and disruptive technologies [28]. the hype cycle also shows the trends for global market developers, chief innovation officers, and r&d leaders. it is important for digital business strategists and digital entrepreneurs to develop digital – and emerging-technology portfolios. the emerging technologies hype cycle is unique among most gartner hype cycles [21]. there are insights at least 2,000 technologies and they are seeking emerging technologies and digital trends. the emerging technologies hype cycle focuses on the technologies, which promise a high degree of competitive advantage over the next 5 to 10 years (fig. 1). when we focus on technology innovation, it is significantly important to evaluate these high-level trends and the featured technologies, as well as the potential impact on their businesses. in addition to the potential impact on businesses, these trends provide a significant opportunity for enterprise architecture leaders to help senior business and it leaders respond to digital business opportunities and threats by creating signature-ready actionable and diagnostic deliverables that guide investment decisions. 3.2.1. ai everywhere ai everywhere: artificial intelligence technologies will be the first technology megatrends. ai will be the most disruptive class of technologies because ai will include radical computational power, unprecedented advances, near endless amounts of data in the digital networks. thus, organizations with ai technologies will have harnessed data in order to adapt new situations and solve digital problems that no one has ever encountered previously. in this ai theme, the companies that are seeking advance, consider should following the disruption technologies: deep learning, artificial general intelligence, autonomous vehicles, deep reinforcement learning, commercial uavs (drones), cognitive computing, smart robots, conversational user interfaces, machine learning, enterprise taxonomy, ontology management, smart dust, and smart workspace. 3.2.2. transparently immersive experiences transparently immersive experiences: the digital technology is coming more and more human-centric to the point where it will introduce transparency between people, businesses and things. this relationship will become significantly important when the evolution of digital technology becomes more contextual and adaptability in the workplace or at home. the most important thing is interacting with people and new businesses. the critical technologies of transparently immersive experiences includes: connected home, human augmentation, 4d printing, augmented reality (ar), computer-brain interface, nanotube electronics, virtual reality (vr), and volumetric displays. 3.2.3. digital platforms digital platforms: the emerging technologies require enabling foundations, which will provide characteristics like: enough data volume, advanced computing power, and ubiquity-enabling ecosystems. the ecosystem-enabling platforms allow entirely new business models between humans and digital technology. the innovative digital platform technologies include: iot platform, serverless paas, neuromorphic hardware, quantum computing, 5g, digital twin, edge computing, blockchain, and software-defined security. these three digital megatrends include the human-centric enabling technologies within transparently immersive experiences: most important are smart workspace, augmented reality, virtual reality, connected home, and the growing brain-computer interface. then, we are becoming the edge of digital technologies that are pulling the other trends along the hype cycle. http://www.gartner.com/technology/research/methodologies/hype-cycle.jsp http://www.gartner.com/technology/research/hype-cycles/?cm_sp=sr-_-hc-_-btn http://www.gartner.com/it-glossary/digital-business/ http://www.gartner.com/it-glossary/digital-business/ http://www.gartner.com/smarterwithgartner/artificial-intelligence-and-the-enterprise/ http://www.gartner.com/smarterwithgartner/the-road-to-connected-autonomous-cars/ http://www.gartner.com/smarterwithgartner/let-machine-learning-boost-your-business-intelligence/ http://www.gartner.com/smarterwithgartner/control-the-connected-home-with-virtual-personal-assistants/ http://www.gartner.com/it-glossary/human-augmentation/ http://www.gartner.com/smarterwithgartner/exploring-augmented-reality-for-business-and-consumers/ http://www.gartner.com/it-glossary/computer-brain-interface/ http://www.gartner.com/it-glossary/nanotube/ http://www.gartner.com/it-glossary/vr-virtual-reality/ http://www.gartner.com/it-glossary/volumetric-displays/ http://www.gartner.com/it-glossary/volumetric-displays/ http://www.gartner.com/it-glossary/quantum-computing/ http://www.gartner.com/it-glossary/quantum-computing/ http://www.gartner.com/smarterwithgartner/5g-whos-in-the-drivers-seat/ advances in technology innovation, vol. 4, no. 2, 2019, pp. 105-115 110 the emerging technologies of ai everywhere are moving fast through the hype cycle. those technologies are key enablers to create transparent and immersive experiences such as deep learning, autonomous learning, and cognitive computing. finally, digital platforms are rapidly moving up the hype cycle. the new innovative it realities provide the underlining platforms that will fuel the future. this future set of technologies includes quantum computing and blockchain, which will create the most transformative and dramatic impacts in the next 5 to 10 years. these three megatrends show that the more companies are able to make technology an integral part of employees', partners' and customers' experiences, the more they will be able to connect in new and dynamic ways to employees', partners' and customers' ecosystems and platforms. for digitalization management and technology leadership, it is important to understand the opportunities and threats affecting digital business. further, it is significant to take the lead in technology-enabled business innovations and help organizations define an effective digital business strategy. 3.3. corporate foresight, management and leadership tools since the late 1980s, the term “foresight” has been used to describe activities which inform decision-makers by improving the inputs about the long-term future of an organization [29-31]. the term foresight and strategic or corporate foresight need to be briefly discussed. the term foresight has been used since the late 1980s to describe an inherently human activity [32]. the term strategic, organizational or corporate foresight has been used to describe future research activities in corporations [32] or organizations. martin [33] and coates [34] emphasize that foresight deals with the long-term future and vecchiato [35] use strategic foresight deliberately to emphasize the tight relationship between foresight and strategy formulation (rohrback and gemüden) [36]. academic studies have generated knowledge on the need for corporate foresight systems (rohrback and gemüden, ruff) [36, 37] and the value contribution of strategic foresight [38] (e.g. vecchiato and roveda, burt and van der heiden). fig. 3 framework showing relationships in the vuca environment. modification of [40] in fig. 3, we have figured out key elements of the competitive landscape which are relevant to the vuca environment. globalization, hyper-competition and rapid technological change create key pre-conditions in the vuca environment. the key challenge for leaders is to change, the threats of the competitive landscape to opportunities. the role of mindsets is very important in this respect. a global mindset or the ability to view the world using a broad perspective converts globalization threats into growth opportunities by thinking beyond geographic boundaries, valuing integration beyond across borders and advances in technology innovation, vol. 4, no. 2, 2019, pp. 105-115 111 appreciating regional and cultural diversity. an innovation mindset is a mental framework that fosters the development and implementation of new ideas. a virtual mindset or the ability of managers to hand over their companies ́activities to hyper competition [39] and external providers, turns hyper-competition into prospects for growth by facilitating flexibility and responsiveness. finally, a collaborative mindset means willingness allowing companies or corporations to engage in business partnerships. collaboration mindset integrates all the other mindsets, which can lead to synergy by business complementarities [40]. we can conclude that these four critical mindsets help corporations to manage disruptive technological innovations. the ability to change threats into opportunities is a critical asset in the vuca conditions. 3.4. the novel vuca challenges for corporate leadership and management in fig. 4, we present a novel synthesis about the vuca challenges and key solutions. the volatility of the environment requires agility on organizational culture. the uncertainty of the environment requires updated information and knowledge management. the complexity of the environment requires active restructuring of corporate organization. the ambiguity of the environment requires experimentation of management activities in the corporations. 3.5. foresight tools important for the vuca environment fig. 4 the vuca challenges and key solution concepts the term vuca has gained currency in the military during the late 1990s to describe an environment of volatility, uncertainty, complexity, and ambiguity. it reflects a shift from traditional cold war military conflicts to asymmetric warfare with agile dispersed opponents fighting under different rules for causes we do not fully understand. business condition is increasingly encountering vuca conditions as well and this poses deep new challenges. after short-run pressures are in conflict with long-run challenges. peter drucker [41] was one of the first to emphasize that management is doing things right and that leadership is about doing the right things [41]. however, in the vuca conditions, it is not easy to define, what are the right things and how to do things in the right ways. in recent manage and leadership literature, krupp and schoemaker [42] have presented a comprehensive answer, the sig discipline model to meet the vuca challenge. in fig. 5 the model is presented. fig. 5 the six disciplines [42] advances in technology innovation, vol. 4, no. 2, 2019, pp. 105-115 112 table 1 the tools relevant to the vuca environment, relevant foresight tools and the key functions of foresight tools inside corporations tools relevant to the vuca environment relevant foresight tools the key functions inside corporations anticipating tools statistical forecasting tools, especially based on probability analysis identify risks and emerging new mark-markets interpreting tools statistical forecasting tools, risk analysis, especially based on probability analysis expert and crowdsourcing methods (delphi methodology and crowdsourcing techniques) analytical reflection of the results of anticipating tools creation of “big picture” of markets and corporate stakeholders challenging tools weak signal and wild card analyses, creativity tools, the analyses of desirability and feasibility, mirroring and benchmarking tools, technology roadmaps, trend and scenario analyses, competitor analyses identify alternatives and uncertainties in the environments eliminate the conceptional problems of group thinking amplify weak and strong signals decision-making tools priority setting tools, multi-objective decision-making tools, and models dr. z methodology and analysis: 1. don t́ rock the boat, 2. joining forces, 3. go it alone, 4. look for a friend and 5. fight the good fight. help decision-makers to be future-oriented decision-makers enable decision making with identifying options and comparing alternative options relevant for corporations pre-condition to the use of decision-making tools is to link challenging tools to decision-making tools aligning tools stakeholder analysis tools action planning deep dialogue tools bridging differences and understanding stakeholders learning tools organization of simultaneous experiments experimental fast learning tools (“valid experiments” and “robust experimental designs”) fast learning organization tools (“easy and quick experiment set-up” and “experimental data available quickly and automatically”) deep learning tools based on ai. create strong passions for experimentation and learning inside a corporation combination tools transcendent leadership tools transcendent leadership combines (1) leadership of self, (2) leadership of others and (3) leadership of the organization in this paper, we are not discussing all the elements of the six disciplines. we focus on the foresight aspects of corporate leadership. however, from fig. 5 we can learn the following messages. leaders and managers can develop their ability and capabilities [42]. the practical recommendations: (1) anticipate changes in the market environment by staying closely connected with customers, partners and competitors, rather than becoming disconnected and reactive. (2) interpret a wide array of data and viewpoints rather than looking only for evidence that confirms their prior beliefs. (3) challenge assumptions and the status quo by surrounding themselves with people who think outside the box and open to new ideas. (4) decide what to do after examining their options and then the courage to get it done than waffling or belaboring the decision-making process. (5) align the interests and incentives of stakeholders based on understanding different views, rather than relying on their power or position. (6) learn from success and failure by experimenting, making small bets, and mining the lessons from both the good and the bad outcomes to create quick learning cycles. in table 1, we have reported the tools relevant to the vuca environment, relevant foresight tools and the key functions of tools inside corporations. this table 1 summarizes the insights of krupp and schoemaker [41], but includes some additional remarks of the authors. especially we have clarified and defined the issue of key foresight tool and key functions of foresight tools inside corporations in this table. advances in technology innovation, vol. 4, no. 2, 2019, pp. 105-115 113 these methodologies are discussed widely in the fields of futures studies and foresight [29, 31-32, 41-45]. the table above refers to the use of these futures oriented methodologies. 4. conclusions this paper contributes academically and practically to the discussion of digitalization and disruptive technologies. digital and disruptive technologies drive most economic growth and productivity. further, we practically analyzed and dis-empirical demonstration of the difference between gartner ś hype curve years 2008 and 2016. some technologies of the year 2008 hype cycle were foresighted very well, and they are nowadays very popular: tablet pcs, internet of things, cloud services, 3d printers, solid state drive is well foresight, and the augmented reality is coming fast. compared to the year 2016: fourteen technologies were taken off the hype cycle, and the 16 new technologies included in the hype cycle for the first time in the year 2016. gartner ś hype curve helps the leaders to understand the dynamics of technological disruption, which is extremely important for corporate leaders to be able to foresight the future. the three megatrends of digital business are ai everywhere, transparently immersive experiences, and digital platforms. the digital platforms are rapidly moving up on the hype cycle. the new innovative it realities provide the underlining platforms that will fuel the future. this future set of technologies includes quantum computing and blockchain, which will create the most transformative and dramatic impacts in the next 5 to 10 years. these three megatrends show that the more companies are able to make technology an integral part of employees', partners', and customers' experiences, the more they will be able to connect in new and dynamic ways to employees', partners', customers' ecosystems, and platforms. this paper combines the discussions of technological disruption, gartner ś hype curve, and the vuca environment and leadership. in the vuca conditions, leaders and managers need a new arsenal of foresight and management tools and methods. this paper elaborates some key theoretical approaches and practical solutions to the corporations facing turbulent vuca conditions. these tools can be classified to anticipation tools, interpreting tools, challenging tools, decision-making tools, aligning tools, learning tools, and combination tools. in this paper, we focus on the foresight aspects of corporate leadership. the practical recommendations for leaders: a) stay closely connected with customers, partners and competitors, b) interpret a wide array of data and viewpoints, c) challenge assumptions, d) decide what to do and then encourage personal, e) align the interests and incentives of stakeholders, f) learn from success and failure to create quick learning cycles. with the systematic application of these tools, corporate leaders and managers can face the vuca tests of surviving in the markets, where globalization, hyper-competition, fast turbulent technological changes test corporations and create increasing volatility, uncertainty, complexity, and ambiguity. already awareness of these vuca conditions and possible tools are important issues. many corporate leaders and managers need an updated understanding of these issues. global mindset, virtual mindset, innovation mindset, and collaboration mindset are key issues in the vuca environment. this paper helps corporate leaders and managers to understand key issues relevant to these mindsets, especially for an innovation mindset. conflicts of interest the authors declare no conflict of interest. advances in technology innovation, vol. 4, no. 2, 2019, pp. 105-115 114 references [1] c. jones, growth and ideas, ch. 16 in handbook of economic growth 1b, aghion p. and s. durlauf, internet article published in elsevier b.v, 2005. [2] c. m. christensen, “the innovator's dilemma: when new technologies cause great firms to fail,” boston: harvard business school press, 1997. [3] j. l. bower and c. m. christensen, “disruptive technologies: catching the wave. harvard business review video,” https://hbr.org/1995/01/disruptive-technologies-catching-the-wave, 1995. [4] c. m. christensen, h. baumann, r. ruggles, and t. m. sadtler. “disruptive innovation for social change,” harvard business review, vol. 84, no. 12, 94, 2006. [5] h. koski, h. melkas, m. mäntylä, r. pieters, r. svento, t. särkikoski, h. talja, j. hyyppä, h. kaartinen, h. hyyppä, and l. matikainen, “technology disruptions as enablers of organizational and social innovation in digitalized environment,” etla working papers no. 45, 2016. [6] d. nagy, j. schuessler, and a. dubinsky, “defining and identifying disruptive innovations,” industrial marketing management, vol. 57, pp. 119-126, 2016. [7] c. m. christensen, r. bohmer, and j. kenagy, “will disruptive innovations cure health care? ” harvard business review, vol. 78, no. 5, pp. 102-112, 2000. [8] c. m. christensen, m. raynor, and mcdonald r, “what is disruptive innovation? ” journal of harvard business review, december 2015. [9] c. m. christensen, m. b horn, and c. w. johnson, “disrupting class. how disruptive innovation will change the way the world learns,” (city): mc grow hill, 2008. [10] j. paap and r. katz, “anticipating disruptive innovation,” research and technology management, vol. 47, no. 5, pp. 13-22, 2004. [11] p. thomond and f. lettice, “disruptive innovation explored,” cranfield university, cranfield, england. conference paper presented at: 9th ipse international conference on concurrent engineering: research and applications (ce2002), july 2002. [12] gartner hype cycle 2008. figure in the internet article: gartner ltd, 2008. [13] gartner hype cycle 2009. figure in the internet article: gartner ltd, 2009. [14] gartner hype cycle 2010. figure in the internet article: gartner ltd, 2010. [15] gartner hype cycle 2011. figure in the internet article: gartner ltd, 2011. [16] gartner hype cycle 2012. figure in the internet article: gartner ltd, 2012. [17] gartner hype cycle 2013. figure in the internet article: gartner ltd, 2013. [18] gartner hype cycle 2014. figure in the internet article: gartner ltd, 2014. [19] gartner hype cycle 2015. figure in the internet article: gartner ltd, 2015. [20] gartner hype cycle 2016. figure in the internet article: gartner ltd, 2016. [21] gartner hype cycle 2017. figure in the internet article: gartner ltd, 2017. [22] j. fenn and m. raskino, “mastering the hype cycle: how to choose the right innovation at the right time,” journal of harvard business press, 2008. [23] w. j. abernathy and m. u. james, “patterns of industrial innovation,” technology review, vol. 64, pp. 254-228, 1978. [24] j. fenn, when to leap on the hype cycle: research note, stamford, ct: gartner group, 1999. [25] k. b. dahlin and d. m. behrens, “when is an invention really radical?: defining and measuring technological radicalness,” research policy, vol. 34, no. 5, pp. 717-737, 2005. [26] a. c. adamuthe, j.v. tomke, and g. t. thampi, “an empirical analysis of hype-cycle: a case study of cloud computing technologies,” international journal of advanced research in computer and communication engineering, vol. 4, no. 10, october 2015. [27] gartner inc., “gartner's 2016 hype cycle for emerging technologies identifies three key trends that organizations must track to gain competitive advantage,” http://www.gartner.com/newsroom/id/3412017; 2016. [28] l. columus, gartner hype cycle for emerging technologies. adds blockchain & machine learning for first time, www.forbes.com, 2016. [29] j. kaivo-oja and t. santonen, “futures of innovation systems and innovation management: open innovation paradigm analysed from futures perspectives,” in open innovation: a multifaceted perspective: part i, ed: world scientific, 2016, pp. 111-158. [30] m. keenan, d. loveridge, i. miles, and j. kaivo-oja, handbook of knowledge society foresight, european foundation for the improvement of living and working conditions, dublin, 2003. advances in technology innovation, vol. 4, no. 2, 2019, pp. 105-115 115 [31] d. loveridge, “foresight,” the art and science of anticipating the future. london: routledge, vol. 11, no. 5, pp. 80-86, 2009. [32] h. a. von der gracht, c. r. vennemann, and i. l. darkow, “corporate foresight and innovation management: a portfolio-approach in evaluating organizational development,” futures, vol. 42, no.4, pp. 380-393, 2010. [33] b. r. martin, “the origins of the concept of 'foresight' in science and technology: an insider's perspective,” technological forecasting and social change, vol. 77 no. 9, pp. 1438-1447, november 2010. [34] j. f. coates, “the future of foresight-a us perspective,” journal of technological forecasting and social change, vol. 77 no. 9, pp. 1428-1437, 2010. [35] r. vecchiato, “strategic planning and organizational flexibility in turbulent environments,” foresight, vol. 17, no. 3, pp. 257-273, 2015. [36] r. rohrbeck and h. g. gemünden, “corporate foresight: its three roles in enhancing the innovation capacity of a firm,” technological forecasting and social chang, vol 78, no 2, 231-243, february 2011. [37] f. ruff, “corporate foresight: integrating the future business environment into innovation and strategy,” international journal of technology management, vol 34, no 3-4, pp. 278–295, 2006. [38] r. vecchiato and c. roveda, “strategic foresight in corporate organizations: handling the effect and response uncertainty of technology and social drivers of change,” technological forecasting and social change, vol. 77, no. 9, pp. 1527-1539, november 2010. [39] r. a. d’aveni, “hypercompetition: managing the dynamics of strategic maneuvering.” free press, 1994. [40] s. lahiri, p. n. l, and r. w. renn, “will the new competitive landscape cause your firm’s decline? it depends on your mindset,” business horizons, vol. 51, no. 4, pp. 311-320, 2008. [41] p. f. drucker, “essential drucker: the best of sixty years of peter drucker's essential writings on management,” new york: harper business, 2008. [42] s. krupp and p. j. h. schoemaker, “winning the long game. how strategic leaders shape the future? ” new york: public affairs, us, 2014. [43] j. s. armstrong, “principles of forecasting,” a handbook for researchers and practitioners, new york: springer science + business media, 2001. [44] k. borch, s. m. dingli, and m.s. jörgensen. “participation and integration in foresight. dialogue, dssemination, and vsions.” edward elgar, city: publisher cheltenham, uk, 2013. [45] k. haegman, e. marinelli, f. scapalo, a. ricci, and a. sokolov, “quantitative and qualitative approaches in future-oriented technology analysis (fta): from combination to integration?” technological forecasting & social change, vol. 80, no. 3, pp. 386-397, 2013. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (ccby) license (https://creativecommons.org/licenses/by-nc/4.0/). microsoft word 6-v8n2(2023)-aiti#11232(150-161).docx advances in technology innovation, vol. 8, no. 2, 2023, pp. 150-161 state estimation and optimal control of four-tank system with stochastic approximation approach xian wen sim1, sie long kek1,*, sy yi sim2, jiao li3 1department of mathematics and statistics, universiti tun hussein onn malaysia, johor, malaysia 2department of electrical engineering, universiti tun hussein onn malaysia, johor, malaysia 3school of mathematics and statistics, changsha university of science and technology, hunan, china received 28 november 2022; received in revised form 20 february 2023; accepted 26 february 2023 doi: https://doi.org/10.46604/aiti.2023.11232 abstract this study aims to optimally control the level of a four-tank system at the steady state in the random disturbance environment using the stochastic approximation (sa) approach. firstly, the stochastic optimal control problem of an equivalent discrete-time is introduced, where the voltages to the pumps are the control inputs. by minimizing the sum of squared errors, the liquid levels are estimated. then, first-order necessary conditions are derived by defining the hamiltonian function. thus, the optimal voltages are calculated based on the estimated liquid levels to update the gradient of the cost function. finally, for illustration, parameters in the system are considered and a simulation is conducted. the simulation results show that the state estimation and control law design can perform well, and the liquid levels are addressed along the steady state. in conclusion, the applicability of the sa approach for handling a four-tank system with random disturbances is demonstrated. keywords: four-tank system, nonlinear optimal control, stochastic approximation, state estimation 1. introduction water tank systems have been widely studied in engineering, such as mechanical and chemical systems. in particular, a four-tank system, which is a standard nonlinear dynamical system, has a more complex liquid-level control problem than two and three-tank systems. considering the interconnected structure between the inputs and outputs in these four tanks, changing a single input to control the system output is complex, and the stabilization of the system changes once a single input is changed. the four-tank system with coupling effects between the input and output causes complex nonlinear behavior, and this nonlinear process is troublesome to manage. thus, it is challenging to stabilize the response output from a four-tank system that is a multiple-input multiple-output (mimo) process [1-2]. moreover, most industrial processes have the problem of controlling the liquid level in the tanks. typically, liquids used in chemical plants and in mixing treatments in tanks must be controlled at a steady-state height, and the flow between tanks should be regulated [3] at the desired level. the various applications of liquid-level control, including nuclear power plants [4], food processing [5], and beverages and pharmaceuticals [6], have been well-presented in the literature. for simplicity, a four-tank system consisting of four interconnected tanks and two pumps are considered. this system aims to control the liquid level in the two lower tanks using two pumps. the inputs of the process are the voltage of the pumps, and the outputs are the liquid levels in the lower two tanks [7-8]. these two pumps convey liquid from a basin to four overhead * corresponding author. e-mail address: slkek@uthm.edu.my advances in technology innovation, vol. 8, no. 2, 2023, pp. 150-161 151 tanks. the liquid is freely drained from the two upper-level tanks into the two bottom-level tanks. then, the liquid levels in the two bottom tanks are measured. in such a piping system, each pump affects the liquid levels in both the measured tanks. a portion of the flow from the pumps is directed to one of the tanks at the bottom level, while the rest of the flow from the pumps is directed to the overhead tank, which drains freely into the other tanks at the bottom level. the amount of flow between the inputs and the outputs can vary by adjusting the bypass valves of the system [9-10]. several practical studies have been conducted on the control of the four-tank systems. proportional-integral-derivative (pid), model predictive control (mpc), and fuzzy modified model reference adaptive control (fmmrac) were implemented to control a four-tank system [11]. the system’s performance was analyzed using conventional proportional-integral (pi) and hybrid fuzzy pi controllers. linear quadratic regulator (lqr), linear quadratic gaussian regulator (lqgr), and h2 and h∞ controllers [12] were applied to manage the four-tank system separately. then, compare their performances in terms of disturbance rejection to study the effects of these control systems on the four-tank system. moreover, a distributed control and estimation method [13] was designed for a multivariate four-tank process, whereas active disturbance rejection techniques [14] and disturbance observers [15] were proposed for studying four-tank systems in the presence of random disturbances. furthermore, some studies have been conducted on tank systems using the simultaneous perturbation stochastic approximation (spsa) method. for example, the water cooling of sulfuric acid in a two-tank system was studied using the spsa technique, where a neural model and a model-based predictive neural pid controller were developed [16]. a novel optimization method based on the spsa approach was proposed to maximize the control performance of steam-generator level control [17]. the spsa method was also developed and implemented on a dual-tank liquid-level control system to improve the performance of a pid control system for parameter optimization in the tank model and actual equipment [18]. in addition, the spsa method was applied to optimize the trajectories of good placement in the reservoir management optimization problem [19]. however, few studies used the sa technique to manage four-tank systems. therefore, this study aims to explore the application of the sa approach [20] for controlling a four-tank system with random disturbances, which might occur from measurement and human errors. since the actual state of the system, which is the water level in the tank, is uncertain in the presence of random disturbances, estimating the water level in the tank will be difficult. in this situation, researchers are motivated to propose a computational approach based on the sa approach to handle the state estimation of the system and control the water level at a steady state as the main contribution to this study. this computational algorithm is known as the stochastic approximation for state-control (sasc) algorithm. firstly, a loss function is defined, where the differences between the actual and estimated output are minimized. then, the sa updating rule is derived for obtaining state estimates after the first-order necessary condition is satisfied. in addition, a stochastic optimization problem is introduced to minimize the cost function when a sequence of control inputs can be determined to stabilize the system. thus, the hamiltonian function is defined, and the optimality conditions are derived. on this basis, the control law is designed through the sa updating rule based on state estimates. in this study, the performance of the sasc algorithm is given by mean squared errors (mse) for state estimation and the cost function for the control effort. for illustration, the values of parameters in the system are considered, and the four-tank system’s discrete-time stochastic optimal control problem is solved iteratively using the sasc algorithm. finally, the simulation results are presented and discussed after being compared with the results from the extended kalman filter (ekf). the rest of this paper is organized as follows. section 2 describes the mathematical model of a four-tank system and defines the system’s discrete-time nonlinear stochastic optimal control problem. section 3 explains state estimation and control law design using the sa algorithm, and the iterative procedure with the sa updating scheme is summarized as the sasc algorithm. section 4 presents and discusses simulation results obtained using the sasc algorithm. finally, section 5 presents a concluding remark. advances in technology innovation, vol. 8, no. 2, 2023, pp. 150-161 152 2. problem description consider a four-tank system [2-3] consisting of four connected tanks and two pumps, as shown in fig. 1. pump 1 extracts liquid from the basin below and pumps it to tanks 1 and 4, whereas pump 2 pumps liquid from the basin below to tanks 2 and 3. denote ��, ��, ��, and �� as the liquid level of tanks 1, 2, 3, and 4, respectively, while �� and �� are the flow parameters of pumps 1 and 2. fig. 1 four tanks system suppose �� �, �� �, �� �, and �� � are the flows into tanks 1, 2, 3, and 4, respectively, while � �� � and � �� � represent the flow out of the electrical pumps 1 and 2. the flows into tanks 1 and 4 are equal to the flow out of electrical pump 1, whereas those into tanks 2 and 3 are equal to electrical pump 2. the relations between the flows at each outlet pipe and the total flows from pumps 1 and 2 depend on the flow parameters �� and �� are given as follows [6], 1 1 1pump q k v= (1) 2 2 2pump q k v= (2) 1 1 1 1in q k v γ= (3) 2 2 2 2in q k v γ= (4) ( )3 2 2 21inq k v γ= − (5) ( )4 1 1 11inq k v γ= − (6) where � and � are the pump coefficients, while �� and �� are the input voltages to the pumps. the flow out of tanks is given by torricelli’s principle by using bernoulli’s equation [7], 1 1 1 2 out q a gl= (7) advances in technology innovation, vol. 8, no. 2, 2023, pp. 150-161 153 2 2 2 2 out q a gl= (8) 3 3 3 2 out q a gl= (9) 4 4 4 2 out q a gl= (10) where ��, ��, ��, and �� are the cross-section of the outlet pipe of tanks 1, 2, 3, and 4, while g is the acceleration due to gravity. the mathematical model of the four-tank system is described by the mass balance equation [11-12] given by tank 1: 1 1 1 3 1in out out dl a q q q dt = + − (11) tank 2: 2 2 2 4 2in out out dl a q q q dt = + − (12) tank 3: 3 3 3 3in out dl a q q dt = − (13) tank 4: 4 4 4 4in out dl a q q dt = − (14) where ��, ��, ��, and �� are the area of the cross-section of tanks 1, 2, 3, and 4, respectively. eqs. (11)-(14) show that the net change in the volume in a tank is equal to the difference between the volume entering and leaving the tank. by substituting eqs. (3)-(10) into eqs. (11)-(14), the dynamics of the four-tank system are given as follows, 31 1 1 1 1 3 1 1 1 1 2 2 adl a k gl gl v dt a a a γ = − + + (15) 2 2 4 2 2 2 4 2 2 2 2 2 2 dl a a k gl gl v dt a a a γ = − + + (16) ( )2 23 3 3 2 3 3 1 2 kdl a gl v dt a a γ− = − + (17) ( )1 14 4 4 1 4 4 1 2 kdl a gl v dt a a γ− = − + (18) define the state variable � = ���, ��, ��, ��� � with �� = ��, �� = �� , �� = ��, and �� = �� , and let the control variable � = ���, ��� � with �� = �� and �� = ��, the dynamics of the four-tank system in eqs. (15)-(18) can be formulated in its equivalent discrete-time state equation [21], ( ) ( ) ( )1 ,x k f x k u k+ =    (19) where � = ���, ��, ��, ��� � : ℜ� × ℜ� → ℜ� is the system dynamics given by 31 1 1 1 1 1 3 1 1 1 1 2 2 aa k f x gx gx u a a a γ τ   = + − + +    (20) advances in technology innovation, vol. 8, no. 2, 2023, pp. 150-161 154 2 4 2 2 2 2 2 4 2 2 2 2 2 2 a a k f x gx gx u a a a γ τ   = + − + +    (21) ( )2 23 3 3 3 2 3 3 12 ka f x gx u a a γ τ   = + − +    (22) ( )1 14 4 4 4 1 4 4 12 ka f x gx u a a γ τ   = + − +    (23) with � as the sampling time. the output variable � = ���, ��� � is defined by ( ) ( )y k h x k=    (24) where the output measurement channel ℎ = �ℎ�, ℎ�� � : ℜ� → ℜ� is 1 1 h x= (25) 2 2 h x= (26) which represents the solution for the liquid level of tank 1 and 2 as ��� � and ��� �, respectively. therefore, the discrete-time system [22] consists of eqs. (19) and (24), which are disturbed by random noises, are denoted as ( ) ( ) ( ) ( )1 ,x k f x k u k g kω+ = +   (27) ( ) ( ) ( )y k h x k kη= +   (28) where g is a 4 × 4 coefficient matrix, whereas !� � ∈ ℛ�, = 0,1, ⋯ , ' − 1, and )� � ∈ ℛ�, = 0,1, ⋯ , ', are the additive gaussian white noises with zero mean, and their respective covariance matrices are given by *+ ∈ ℛ �×� and ,∈ ℛ �×�. ( ) 00x x= (29) the initial state eq. (29) is a random vector with the mean and state error covariance matrix are, respectively, given by ( )0 0e x x= (30) ( ) ( )0 0 0 0 0 t e x x x x m − − =   (31) where ./ ∈ ℛ �×� is a positive definite matrix and 01 2 is the expectation operator. it is assumed that the initial state, process noise, and measurement noise are statistically independent. here, the aim is to determine a set of admissible control sequences �� � ∈ ℛ�, = 0,1, ⋯ , ' − 1, such that the following expected cost function ( ) ( ) ( ) ( ) 1 0 , n k j u e x n l x k u kϕ − =   = +           (32) is minimized over the dynamical system given in eqs. (27) and (28), where 3: ℜ� → ℛ is the terminal cost and 4: ℜ� × ℜ� → ℛ is the cost under summation. this problem is known as the discrete-time nonlinear stochastic optimal control problem for the four-tank system and is referred to as problem (p). 3. stochastic approximation (sa) approach now, consider the following recursive equation, 1i i i i a gθ θ+ = − × (33) advances in technology innovation, vol. 8, no. 2, 2023, pp. 150-161 155 where 5� is the set of parameters to be estimated, 6� = 6�5� � is the stochastic gradient, and �� is the gain sequence. this equation is known as the sa approach [20]. thus, the state estimation and the optimal control design based on the sa approach will be further discussed. 3.1. state estimation consider the state mean propagation [23-24] for eqs. (27) and (28), given by ( ) ( ) ( )1 ,x k f x k u k+ =    (34) ( ) ( )y k h x k=    (35) where �̅� � and �8� � are the expected state sequence and the expected output sequence, respectively. to find the optimal state estimate, introduce the following weighted least-squares problem [21-22], ( ) ( ) ( ) ( ) ( ) ( ){ } ( ) ( ){ } ( ) ( ) ( ){ } 1 0 1 1 min 2 1 2 t sse x t j x x k x k m x k x k y k h x k r y k h x kη − − = − −       + − −       (36) where 9::; is the sum of squares error, ./ is the initial state error covariance, and , is the output noise covariance. by taking the first-order derivative, the gradient of the sum squares of errors is defined by ( ) ( ) ( ) ( ) ( ) ( ) ( ){ } 11 0 t x sse j x m x k x k c r y k h x kη −− ∇ = − − −       (37) where < = =ℎ =�⁄ . thus, by using the sa approach [20] in eq. (33), the optimal state estimate is obtained from ( ) ( ) ( ) 1 1, ˆ ˆ i i i i x sse x k x k a j x k +  = − ×∇   (38) where �?,� > 0 is the learning rate and the optimal output estimate is determined by ( ) ( )ˆ ˆ i i y k h x k =   (39) remarks: the state estimate �a is the optimal state estimate, which minimizes the sum of squares of error 9::; . the norm of the state estimate �a and the state mean �̅ is relatively close within a small tolerance. using the sa approach for state estimation does not need the state error covariance matrix equation as derived in the kalman filtering approach [21-22]. 3.2. optimality conditions refer to the problem (p), the expected cost function [24] in eq. (32) can be defined by ( ) ( ) ( ) ( ) 1 0 , n k j u x n l x k u kϕ − = = +       (40) and the state propagation from eq. (34) is considered. define the hamiltonian function [19], ( ) ( ) ( ) ( ) ( ) ( )ˆ, 1 , t h u l x k u k p k f x k u k= + +       (41) where b� � ∈ ℛ�, = 0,1, ⋯ , ' is the costate sequence to be determined later. thus, the augmented cost function becomes ( ) ( ) ( ) ( ) ( ) 1 0 1 1 n t k j u x n h k p k x kϕ − = ′ = + − + +    (42) advances in technology innovation, vol. 8, no. 2, 2023, pp. 150-161 156 examining the increment in the augmented cost function 9′ due to increments in all variables, which are �̅� �, b� �, and �� � according to the lagrange multiplier theory, this increment d9′ should be zero at a constrained minimum [21-22]. thus, taking the first-order derivative of the augmented cost function and hamiltonian function, the following optimality conditions are derived. (a) stationary condition ( ) ( ) ( ) ( ) ( ) ( ) ( )ˆ, , 1 0 t u k u k l x k u k f x k u k p k∇ + ∇ + =       (43) (b) state equation ( ) ( ) ( )ˆ1 ,x k f x k u k+ =    (44) (c) costate equation ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )ˆ, , 1 t x k x k p k l x k u k f x k u k p k= ∇ + ∇ +       (45) (d) output equation ( ) ( )y k h x k=    (46) (e) boundary conditions ( ) 0ˆ 0x x= (47) ( ) ( ) ( )x np n x nϕ= ∇    (48) remark: for the sake of convenience, the following standard cost function in quadratic criterion [23-24] could be calculated ( ) ( ) ( ) ( ) 1 2 t x n x n s n x nφ =   (49) ( ) ( ) ( ) ( ) ( ) ( ) 1 , 2 t t l x k u k x k qx k u k ru k = +     (50) when a proper cost function is not provided. thus, the necessary conditions are simple. 3.3. optimal control design define an equivalent stochastic optimization problem [20] to problem (p), and this problem is regarded as the problem (q), given by ( )minimize j u′→ (51) where the necessary conditions in eqs. (44) and (45) are satisfied. hence, solving the problem (q) would allow the design of the control law. by this, the gradient of the objective function in eq. (42) is expressed by ( ) ( )u uj u h k′∇ = ∇ (52) where ( ) ( ) ( ) ( ) ( )ˆ, , 1 t u u u h l x k u k f x k u k p k∇ = ∇ + ∇ +       (53) and the necessary condition for the problem (q) is given by eq. (43). hence, the control law is updated from ( ) ( ) ( ) 1 2, i i i i u u k u k a j u k +  ′= − ×∇   (54) advances in technology innovation, vol. 8, no. 2, 2023, pp. 150-161 157 where ��,� > 0 is the learning rate. here, the principle of separation is assumed to be satisfied when applying the sa approach [20] for state estimation and optimal control design. this statement is true for solving stochastic optimal control problems. 3.4. sa for state-control algorithm from the discussion above, the computational procedure for applying the sa approach to state estimation and control law design is summarized as an iterative algorithm named the sasc algorithm. the steps of the sasc algorithm are as follows: data: given �, ℎ, f, 3, 4, ', ./, g+, ,-, ��, ��, and �. step 0: determine the initial control �� �/ = �/ for = 0,1, ⋯ , ' − 1 and the initial state �a� � / = �/ for = 0,1, ⋯ , ' set the tolerance h and the iteration i = 0. step 1: calculate the sum of squares error 9::; 1�a� � � 2 from eq. (36), and the stochastic gradient ∇k 9::; 1�a� � � 2 from eq. (37), respectively. step 2: update the state estimate �a� ��l� from eq. (38). step 3: compute the output estimate �a� �� from eq. (39). step 4: solve the state equation forward in time from eq. (44) with the given initial state �̅/ to obtain the state solution �̅� � � . step 5: solve the costate equation backward in time from eq. (45) with the given final costate b�'� to provide the costate solution b� �� . step 6: compute the output measurement �8� �� from eq. (46). step 7: calculate the cost function 91�� �2� from eq. (42) and calculate the stochastic gradient ∇�91�� �2 � from eq. (53). step 8: update the control law �� ��l� from eq. (54). step 9: test the convergence. if �a� ��l� = �a� �� and �� ��l� = �� �� within a given tolerance h, stop, else set the iteration i = i + 1 and go to step 1. remarks: from the steps of the sasc algorithm, the initial value of the state and control can be set to a zero vector in step 0. the state estimation procedure is implemented from step 1 to step 3, and the two-point boundary-value problem is solved from step 4 to step 5. while the optimal control law is designed from step 7 to step 8, and the appropriate stopping criteria for the iteration can be set in step 9. 4. simulation results table 1 parameters in problem (p) system parameters values the cross-sectional area of the outlet hole for the tanks �� = 0.071 pq �, �� = 0.057 pq �, �� = 0.071 pq �, �� = 0.057 pq � the cross-sectional area of the tanks �� = 28 pq �, �� = 32 pq �, �� = 28 pq �, �� = 32 pq � pump proportionality constants � = 3.14 pq �/x: , � = 6.29 pq �/x: flow coefficient of the pumps �� = 0.35, �� = 1.35 gravitational acceleration 6 = 981 pq/[ � initial liquid level ���0� = 10.43, ���0� = 15.98, ���0� = 6.6, ���0� = 9.57 estimation and control parameters values sampling time � = 0.1 [ final time step ' = 80 weighting matrices \] = �̂×�, g = di�6�10, 10, 10 _, 10_�, , = 100`�×� covariance matrices ./ = 0.2`�×�, g+ = 0.01`�×�, ,= 0.01`�×� advances in technology innovation, vol. 8, no. 2, 2023, pp. 150-161 158 consider the system parameters for the four-tank system [3, 9], and the estimation and control parameters defined for the problem (p) listed in table 1. using these parameters, an illustrative example of the simulation of the four-tank system is studied to demonstrate the practicality of the sasc algorithm. the simulation of this study is conducted in the environment of the gnu octave 7.2.0. table 2 presents the simulation results for controlling the four-tank system with random disturbances. the implementation of the sasc algorithm required 1,254 iterations to achieve convergence. for benchmark purposes, the results of the sasc algorithm were compared with results from the ekf algorithm since the ekf algorithm is the standard technique for nonlinear filtering and estimation. the optimal cost of 3.4959 × 107 units obtained using the sasc algorithm was 1.15 % less than the optimal cost given by the ekf algorithm. this reduction demonstrated the practical application of the sasc algorithm in minimizing the cost function of the system. however, the performance of the sasc algorithm for state estimation, measured through the sum of squares error (sse) and the mse, was 98.7 % more accurate than that of the ekf algorithm. therefore, the efficiency and accuracy of the sasc algorithm for solving the discrete-time nonlinear stochastic optimal control of the four-tank system were demonstrated. table 2 simulation result algorithm optimal cost sse mse ekf 3.5366 × 107 2.307386 × 10–1 2.884233 × 10–3 sasc 3.4959 × 107 3.091891 × 10–3 3.864864 × 10–5 fig. 2 shows the final output trajectories when the sasc algorithm achieved convergence. the liquid level in tank 1, in which y� represents the actual liquid level and yb� is the estimated liquid level in tank 1, which was reduced from 10.43 cm and reached about 4.5 cm after 2 seconds at the final time of the iteration. while the liquid level in tank 2, in which y� represents the actual liquid level and yb� is the estimated liquid level in tank 2, which reached about 37.5 cm after 2 seconds at the final time of the iteration after increasing from 15.98 cm. in the random disturbance situation, it was challenging to maintain the steady state of the liquid level in tanks 1 and 2. however, the final liquid levels in both tanks 1 and 2 were approximately determined in a satisfactory form by using the sasc algorithm. (a) output trajectory y� – the liquid level in tank 1 (b) output trajectory y� – the liquid level in tank 2 fig. 2 final output trajectories and real output trajectories the final state trajectories are shown in fig. 3. the liquid levels in tanks 1 and 2 exhibited fluctuation behaviors disturbed by random noise and were not easy to measure smoothly. using the sasc algorithm, these trajectories were estimated acceptedly, and their trajectories were approximately measured. conversely, the liquid levels in tanks 3 and 4 were unaffected by random disturbances. this is because their trajectories were smoothly predicted at their respective steady states approximately along with zero. advances in technology innovation, vol. 8, no. 2, 2023, pp. 150-161 159 fig. 4 shows the final control trajectories for regulating the liquid levels in the tanks. the optimal input voltage to pump 1 increased from -270 v to 0 v, and the optimal input voltage of pump 2 decreased from 182 v to 0 v. these control efforts effectively maintained liquid levels at approximately zero after 2 seconds. therefore, the optimal solution to the four-tank problem was obtained satisfactorily when stationary conditions were satisfied, as shown in fig. 5. (a) state trajectory �� – the liquid level in tank 1 (b) state trajectory �� – the liquid level in tank 2 (c) state trajectory �� – the liquid level in tank 3 (d) state trajectory �� – the liquid level in tank 4 fig. 3 final state trajectory and real state trajectory fig. 4 final control trajectories fig. 5 stationary conditions 5. concluding remarks optimizing and controlling the four-tank system with random disturbances through the sa approach were discussed in this study. firstly, the discrete-time stochastic optimal control problem for the four-tank system was described by considering the presence of random disturbances. subsequently, by applying the sa approach, the iterative algorithm, namely the sasc approach, was proposed to estimate the state dynamics and to design the optimal control law. therefore, the state estimation advances in technology innovation, vol. 8, no. 2, 2023, pp. 150-161 160 of the system was satisfactorily handled, and the optimal control law, which was thoroughly designed based on the sa updating rule, was applied to minimize the performance index of the system. for illustration, researchers studied the control problem of a four-tank system with given parameters. the simulation results showed that the system was stabilized and controlled in a stochastic environment after using the sasc algorithm proposed. these results were also compared with results from the ekf technique, and a discussion was given. in conclusion, the efficiency and accuracy of the sasc algorithm are demonstrated. for future research, it is recommended to apply recent variants of the sa approach, like the adam algorithm, for solving the stochastic optimal control problem of the four-tank system so that more acceptable results can be determined. in this way, the water level at the steady state will be identified in fewer iteration numbers, where the algorithm provides an iterative solution that can converge faster. hence, the practicality and usefulness of the algorithm will be recommended. acknowledgment this research was supported by the universiti tun hussein onn malaysia (uthm) through tier 1 (vot q121). conflicts of interest the authors declare no conflicts of interest. references [1] j. k. pradhan, a. ghosh, and c. n. bhende, “two-degree-of-freedom multi-input multi-output proportional–integral– derivative control design: application to quadruple-tank system,” proceedings of the institution of mechanical engineers, part i: journal of systems and control engineering, vol. 233, no. 3, pp. 303-319, march 2019. [2] b. a. kumar, r. jeyabharathi, s. surendhar, s. senthilrani, and s. gayathri, “control of four tank system using model predictive controller,” ieee international conference on system, computation, automation and networking, pp. 1-5, march 2019. [3] g. rigatos, g. cuccurullo, p. siano, k. busawon, and m. abbaszadeh, “nonlinear optimal control for the quadruple water tank system,” aip conference proceedings, vol. 2293, no. 1, article no. 310005, november 2020. [4] l. zhang, h. xie, q. duan, c. lu, j. li, and z. lv, “power level control of nuclear power plant based on asymptotical state observer under neutron sensor fault,” science and technology of nuclear installations, vol. 2021, article no. 8833729, 2021. [5] n. a. al-awad, “optimal control of quadruple tank system using genetic algorithm,” international journal of computing and digital systems, vol. 8, no. 1, pp. 51-59, january 2019. [6] r. kashyap, n. jaggi, and b. pratap, “robust controller design and performance analysis of four-tank coupled system,” 1st international conference on signal processing, vlsi and communication engineering, pp. 1-6, march 2019. [7] s. a. panahi and s. b. m. noor, “model predictive control design for a nonlinear four-tank system,” ifac proceedings volumes, vol. 43, no. 8, pp. 382-387, 2010. [8] r. hedjar and m. bounkhel, “wireless model predictive control: application to water-level system,” advances in mechanical engineering, vol. 8, no. 4, pp. 1-13, 2016. [9] p. prasad, “the performance analysis of four tank system for conventional controller and hybrid controller,” international research journal of engineering and technology, vol. 3, no. 6, pp. 2182-2187, june 2016. [10] k. h. johansson, “the quadruple-tank process: a multivariable laboratory process with an adjustable zero,” ieee transactions on control systems technology, vol. 8, no. 3, pp. 456-465, may 2000. [11] s. b. prusty, u. c. pati, and k. k. mahapatra, “model based predictive control of the four tank system,” advances in system optimization and control, vol. 509, pp. 249-260, 2019. [12] w. a. altabey, “model optimal control of the four tank system,” international journal of systems science and applied mathematics, vol. 1, no. 4, pp. 30-41, november 2016. [13] a. k. ali and m. s. mahmoud, “distributed estimation and control of a four-tank process,” global journal of engineering and technology advances, vol. 9, no. 3, pp. 133-142, december 2021. advances in technology innovation, vol. 8, no. 2, 2023, pp. 150-161 161 [14] x. meng, h. yu, j. zhang, t. xu, and h. wu, “liquid level control of four-tank system based on active disturbance rejection technology,” measurement, vol. 175, article no. 109146, april 2021. [15] x. meng, h. yu, t. xu, and h. wu, “disturbance observer and l2-gain-based state error feedback linearization control for the quadruple-tank liquid-level system,” energies, vol. 13, no. 20, article no. 5500, october 2020. [16] c. renotte, a. v. wouwer, and m. remy, “neural modeling and control of a heat exchanger based on spsa techniques,” proceedings of the 2000 american control conference, vol. 5, pp. 3299-3303, june 2000. [17] x. kong, x. wang, and z. ma, “a novel method for controller parameters optimization of steam generator level control,” 21st international conference on nuclear engineering, article no. 16294, july-august 2013. [18] i. kong, i. qian, and i. wang, “spsa-based pid parameters optimization for a dual-tank liquid-level control system,” ieee international conference on industrial engineering and engineering management, pp. 1463-1467, december 2016. [19] m. jesmani, b. jafarpour, m. c. bellout, and b. foss, “a reduced random sampling strategy for fast robust well placement optimization,” journal of petroleum science and engineering, vol. 184, article no. 106414, january 2020. [20] j. c. spall, introduction to stochastic search and optimization: estimation, simulation, and control, 1st ed., hoboken: wiley-interscience, 2003. [21] a. e. bryson and y. c. ho, applied optimal control: optimization, estimation, and control, 1st ed., new york: taylor and francis, 1975. [22] f. l. lewis, d. l. vrabie, and v. l. syrmos, optimal control, 3rd ed., hoboken: john wiley & sons, 2012. [23] d. e. kirk, optimal control theory: an introduction, new york: dover publications mineola, 2004. [24] k. l. teo, b. li, c. yu, and v. rehbock, applied and computational optimal control: a control parametrization approach, 1st ed., switzerland: springer cham, 2021. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation, vol. 3, no. 3, 2018, pp. 109 117 effects of unsteady aerodynamic pressure load in the thermal environment of fgm plates chih-chiang hong * department of mechanical engineering, hsiuping university of science and technology , taichung, taiwan, roc. received 13 august 2017; received in revised form 14 oct ober 2017; accept ed 20 oct ober 2017 abstract the effects of unsteady aerodynamic pressure load with varied shear correction coeffic ient on the functionally graded material (fgm) plates are investigated. therma l vibrat ion is studied by using the first -order shear deformation theory (fsdt) and the generalized diffe rential quadrature (gdq) method. usually, in the fgm analyses, the computed and varied values of shear correction coeffic ient are the function of the total thickness of plates, fgm power law inde x, and environ ment te mperature. the e ffects of environ ment te mperature and fgm power la w inde x on the thermal stress and center deflection of a irflow over the upper surface of fgm p lates are obtained and investigated. in addition, the effects , with and without the fluid flow over the upper surface of fgm plates , on the center deflection and normal stress are also in vestigated. keywor ds: aerodynamic pressure, varied shear correction coefficient, fgm, thermal vibration, gdq 1. introduction there are some investigations of aerodynamic pressure load on the functionally graded material (fgm) p lates and shells. in 2015, fa ze lzadeh et al. [1] studied the effects of vo lu me fraction, aspect ratio and non-dimensional in-plane forces on the nanocomposite fgm p lates under the action of supersonic aerodynamic pressure. in 2015, lee and kim [2] investigated the structure characteristics of fgm panels under supersonic aerodyna mic force. in 2013, ra fiee et a l. [3] ca lculated and studied the nonlinear vib ration of pie zoelectric fgm shells under combined e lectrica l, therma l, mechanical and aerodynamic loading. in 2012, ghad imi et al. [4] investigated and studied the therma l flutte r characteristics of cantilever fgm plates under supersonic aerodynamic loads. in 2012, prakash et al. [5] co mputed and investigated the large amplitude fle xura l v ibration characteristics of fgm plates with supersonic airflow by using the finite e le ment me thod (fem ). in 2008, sohn and kim [6] presented and exa mined the therma l buckling and flutter characteristics of fgm panels under aero -therma l loads. in 2007, faze lzadeh and hosseini [7] studied and investigated a turbo -machinery fgm rotating blades beam u nder supersonic aero-thermo-elastic loading. in 2007, wu et a l. [8] calcu lated and analyzed the dynamic stability of fgm p lates subjected to aero-thermo-mechanical loads by using the moving least squares differentia l quadrature method. in 2007, navazi and haddadpour [9] investigated the aero -thermo-e lastic stability margins of fgm p lates and panels in the supersonic flow. in 2006, prakash and ganapathi [10] investigated supersonic flutter behavior of fgm plates and flat panels by using the fem . there are some computational investigations of generalized differentia l quadrature (gdq) in the composited fgm plates and shells. in 2017, hong [11] investigated the effects of varied shear correction on flutter va lue of the center deflection and the therma l v ibration of fgm shells in an unsteady supersonic flow. in 2015, tornabene et al. [12] presented a survey of strong formulat ion fem focused on the numerica l investigation of differentia l quadrature method. in 2014, hong [13] studied the therma l v ibration and transient response of terfenol-d fgm p lates by using the gdq method and considering the first -order * corresponding author. e-mail address: cchong@mail.hust.edu.tw advances in technology innovation, vol. 3, no. 3, 2018, pp. 109 117 cop y right © taeti 110 shear deformat ion theory (fsdt ) model and the varied modified shear correct ion factor e ffects. in 2014, hong [14] investigated the rapid heating induced vibration of terfenol-d fgm circula r cylindrica l shells by using the gdq method and without considering the effects of shear deformation. in 2013, hong [15] presented the therma l vib ration of te rfenol -d fgm shells by using the gdq method and also without considerin g the effects of shear deformat ion. in 2012, hong [16] studied the therma l vibrat ion of rapid heating for terfenol-d fgm plates by using the gdq method and considering the fsdt e ffects. in 2009, tornabene and viola [17] presented the free vibration analysis of fgm panels and shells by using the gdq method and considering the fsdt model. it is interesting to study and investigate the therma l stresses and center deflection of gdq computations by considering the fsdt and the varied effects of the shear correct ion coefficient of airflow over the upper surface of fgm plates with four edges in simp ly supported boundary conditions. environ ment te mperature and fgm power law inde x t wo para metric effects on the therma l stress and center deflection of a irflow over the upper surface of fgm p lates are also obtained and investigated. the effects of with and without the fluid flow over the upper surface of fgm p lates on the center deflection and normal stress are also calculated . fig. 1 fluid flow over the upper surface of two-material fgm plates 2. formulation for fluid flo w over the upper surface of two-materia l fgm plates is shown in fig. 1 with thickness h1 of fgm materia l 1 and thickness h2 of fgm materia l 2. the materia l properties are considered in the most dominated property young’s modulus efgm of fgm with standard variation fo rm o f powe r law inde x rn, the others of materia l properties are assumed in the simple average form [18]. the properties of indiv idual constituent material of fgm a re functions of enviro nment te mperature t. the time-dependent, linear fsdt equations of displacements u, vand w\ of fgm plates are assumed in the following [19]. ),,(),,( 0 tyxztyxuu x  (1) ),,(),,( 0 tyxztyxvv y  ),,( tyxww  where u 0 and v 0 a re d isplacements in the x and y a xes direction, respectively, w is t ransverse displacement in the z a xis direction of the middle-plane of plates,  x and  y are the shear rotations, t is time. the norma l stresses (σ𝑥 and σ𝑦 ) and the shear stress es (σ𝑥𝑦 , σ𝑦𝑧 andσ𝑥 𝑧 ) in the fgm p late under temperature d ifference ∆𝑇 for the k th layer can be obtained as follows [20-21]. )()( 662616 262212 161211 )( k xyxy yy xx kk xy y x t t t qqq qqq qqq                                         (2) )()(5545 4544 )( kxz yz kkxz yz qq qq                        advances in technology innovation, vol. 3, no. 3, 2018, pp. 109 117 cop y right © taeti 111 where α𝑥 and α𝑦 are the coeffic ients of thermal e xpansion , α𝑥 𝑦 is the coeffic ient of therma l shear, �̅�𝑖𝑗 is the stiffness of fgm plates. ε𝑥, ε𝑦 and ε𝑥𝑦 are in-p lane strains , not negligible ε𝑦𝑧 and ε𝑥𝑧 are shear strains . the temperature difference between the fgm plate and curing area is given by the following equation. ),,(),,( 1*0 tyxt h z tyxtt  (3) in which t0 and t1 are temperature parameters in functions of x, y, and t, h * is the total thickness of plates. the dynamic equilibriu m d ifferentia l equations of flu id flo w over the upper surface of fgm p lates can be represented and obtained in matrix fo rm [15, 21]. in the matrix e le ments, there are some coefficients (aij, bij, dij), (i,j = 1,2,6), ai*j*,(i * , j * = 4, 5) with partia l derivatives of displace ments and shear rotations . the e xterna l loads are subjected to f1,..., f5 with partia l derivatives of therma l loads (𝑁, 𝑀 ), mechanical loads (p1, p2, q) and inertia terms (𝜌, h, i). in which (aij, bij, dij), ai*j*, f1,..., f5 and (𝜌, h, i) are in the following expressions . 11 p y n x n f xyx        (4) 22 p y n x n f yxy        22 2 3 ** hzhz t w m u x w m u qf              y m x m f xyx       4 y m x m f yxy       5 , dzztqqqmn xyy h h xxx ),1()(),( 1612 2 2 11 * *     (5) dzztqqqmn xyyx h hyy ),1()(),( 2622 2 2 12 * *    dzztqqqmn xyyx h hxyxy ),1()(),( 6626 2 2 16 * *    * * 22 2 ( , , ) (1, , ) , ( , 1, 2, 6) h ij ij ij ijh a b d q z z dz i j    * ** * * * * *2 2 , ( , 4, 5) h hi j i j a k q dz i j    (6) dzzzih h h ),,1(),,( 22 2 0 * *  (7) in wh ich 𝑘𝛼 is the shear correction coeffic ient, 𝜌0 is the density of ply. the values of 𝑘𝛼 are usually functions of h * , t, and rn. q is the aerodynamic pressure load for the unsteady, in viscid fluid flow over the upper surface of fgm plate with free stream density 𝜌∞ , velocity 𝑈∞ , and mach number 𝑀∞ . advances in technology innovation, vol. 3, no. 3, 2018, pp. 109 117 cop y right © taeti 112 the simp le forms of �̅� 𝑖𝑗 and �̅�𝑖∗𝑗 ∗ for fgm p lates were introduced by shen [22] in 2007 and used to calculate norma l, shear stresses and aij. the modified shear correction factor 𝑘𝛼 can be derived and obtained directly fro m the total strain energy principle derivation as follows for the fgm plates [13] by considering the varied value effect of 𝑘𝛼 on the plates . * 1 fgmzsv k h fgmziv   (8) in which the e xp ressions of fgm zsv and f gmziv are functions of h * , rn, the poisson’s ratios vfgm, the young’s modulus of the fgm constituent materials e1 and e2 of the fgm plates, respectively. the dynamic gdq d iscrete equations in matrix notation can be derived and obtained for the dynamic equilibriu m diffe rential equations by considering four sides simply supported, fluid flo w over the upper surface of fgm plates. the gdq method was presented by shu and richards in 1990 [13, 23-25]. 3. computational results to study and obtain the gdq results of varied shear correct ion coeffic ient calcu lations with plates layers in the stacking sequence (0°/0°), under four sides simply supported boundary condition, no in-plane distributed forces (p1 = p2 = 0) and under the external aerodynamic pressure load (q) of airflow over the upper surface of fgm plates with 𝜌∞ = 0.00000678𝑙 𝑏 /𝑖𝑛 3 , 𝑈∞ = 23304𝑖𝑛 /𝑠 and 𝑀∞ = 2 at altitude 50,000ft. the following coordinates xi and yi for the grid points numbers n and m of fgm plates are used 1 0.5 [1 cos( )] , 1, 2, ..., 1 i i x a i n n       (9) 1 0.5 [1 cos( )] , 1, 2, ..., 1 j j y b j m m       the displacement and temperature of thermal vibrat ions are used in time sinusoidal form as follows for a simple case study. )sin()],(),([ 0 tyxzyxuu mnx  (10) )sin()],(),([ 0 tyxzyxvv mny  )sin(),( tyxww mn  * * 2 2 2 ( , ) sin( ) ( , ) cos( ) mn mn mnz h z h u uw x y q t w x y t m x m                 (11) 0 1* [ ( , ) ( , )]sin( ) z t t x y t x y t h    (12) and with the temperature parameter in the following simple vibration 0),( 0 yxt (13) )/sin()/sin(),( 11 byaxtyxt  in which 𝜔𝑚𝑛 is the natural frequency in mode shape numbers m and n of the plates, γ is the frequency of applied heat flu x, �̅�1 is the amplitude of temperature. advances in technology innovation, vol. 3, no. 3, 2018, pp. 109 117 cop y right © taeti 113 the sus304 (stainless steel) for fgm material 1 and the si3n4 (silicon nitride) for fgm material 2 are used in the numerical gdq computations. firstly, the dynamic convergence study of center deflection amplitude w (a/2, b/2) (unit mm) in airflow over the upper surface of fgm plates are obtained in table 1 by considering the varied effects of shear correction coefficient and with h * = 1.2 mm, h1 = h2 = 0.6 mm, m = n = 1, rn =1, 𝑘𝛼 = 0.149001, t = 100k, 1 100t k , t = 6s. the error accuracy is 7.2e-05 for the center deflection amplitude of a / b = 1, a / h * =10. the 17×17 grid point can be considered in the good convergence result and treated in the following gdq computations of time responses for deflection and stress of fgm plates. it might be mentioned that the deformations of plates usually increase with the increasing a / h * for the cases of non thermal loads . however, in the table1 shows the absolute values of center deflection for thin a / h * =100 are much smaller than that for thick a / h * =10 and 5 due to the phenomenon effect of thermal loads. in the fgm plates (𝐵𝑖𝑗 ≠ 0), varied values of 𝑘𝛼 are usually functions of h * , rn and t. for a / h * =10, a / b = 1, h * from 0.12mm to 2.4 mm, h1 = h2, calculated values of 𝑘𝛼 under t =100k are shown in table 2, used for the gdq and shear calculations. for h * = 0.12mm, values of 𝑘𝛼 (from 0.109359 to 1.08902) are increasing with rn (from 0.1 to 2). for h * = 1.2mm, values of 𝑘𝛼 (from 0.891024e-01 to 0.508881) are increasing with rn (from 0.1 to 10). for h * = 2.4mm, values of 𝑘𝛼 (from 0.836925e-01 to 0.118029e-02) are small decreasing with rn (from 0.1 to 10). usually, the values of 𝑘𝛼 are dominantly in the inverse proportion to h * at a given values of rn and t, e.g. values of 𝑘𝛼 (firstly 0.109359, then 0.891024e-01, finally 0.836925e-01) are decreasing with h * (from 0.12mm, then 1.2mm to 2.4mm) at rn = 0.1 and t =100k. table 1 dynamic convergence of airflow over the upper surface of fgm plates a / h * gdq method deflection w (a/2, b/2) at t = 6s 𝑁 × 𝑀 a / b = 0.5 a / b = 1 a / b = 2 100 13 × 13 0.166916e-18 -0.864896e-16 -0.113379e-14 15 × 15 0.166916e-18 -0.864895e-16 -0.113379e-14 17 × 17 0.166914e-18 -0.864888e-16 -0.113378e-14 14 13 × 13 -0.270625e-14 -0.436330e-12 -0.948251e-10 15 × 15 -0.270623e-14 -0.436326e-12 -0.947482e-10 17 × 17 -0.270623e-14 -0.436324e-12 -0.944192e-10 10 13 × 13 -0.190895e-13 -0.318221e-11 -0.674404e-09 15 × 15 -0.190895e-13 -0.318221e-11 -0.673224e-09 17 × 17 -0.190893e-13 -0.318244e-11 -0.674404e-09 8 13 × 13 -0.703716e-13 -0.118106e-10 -0.323377e-08 15 × 15 -0.703718e-13 -0.118357e-10 -0.466268e-08 17 × 17 -0.703708e-13 -0.118450e-10 -0.595615e-08 5 13 × 13 -0.108713e-11 -0.170468e-09 0.259696e-06 15 × 15 -0.108830e-11 -0.167507e-09 0.271164e-06 17 × 17 -0.108713e-11 -0.174309e-09 0.278226e-06 table 2 varied shear correction coefficient 𝑘𝛼 vs. rn under t =100k h * (mm) 𝑘𝛼 rn = 0.1 rn = 0.2 rn = 0.5 rn = 1 rn = 2 rn = 5 rn = 10 0.12 0.109359 0.140739 0.288792 0.687883 1.08902 0.989219 0.878104 1.2 0.891024e-01 0.939102e-01 0.111874 0.149001 0.231364 0.415802 0.508881 2.4 0.836925e-01 0.828082e-01 0.815941e-01 0.796610e-01 0.632624e-01 0.219191e-01 0.118029e-02 secondly, the a mplitude of center deflection w (a/2, b/2) (unit mm) for the a irflo w over the upper surface of fgm plates is calculated. fig. 2 shows the response values of center deflection a mplitude w (a/2, b/2) (unit mm) versus time t in fgm p late for a / h * =10, 14 and th in a / h * = 100, respectively, a / b = 1, h * = 1.2 mm, h1= h2= 0.6mm, rn = 1, 𝑘𝛼 = 0.117077, t =653k , 𝑇1 =100k, starting time t = 0.001s and t =0.1s-3.0s with time step is 0.1s . the absolute value of center deflection a mp litude is 7.04e-12mm occurs at t = 0.2s, the steady state value of center deflection is -4.54e-12mm for a / h * =10. the absolute value of center deflection a mplitude is 1.31e-12mm occurs at t = 0.1s, the steady state value of center deflection is -6.21e-13mm for a / h * =14. the absolute values of center deflection for thin a / h * =100 are much smaller than that for a / h * =10 and 14. advances in technology innovation, vol. 3, no. 3, 2018, pp. 109 117 cop y right © taeti 114 (a) w (a/2, b/2)versus t for a / h * = 10 (b) w (a/2, b/2)versus t for a / h * = 14 (c) w (a/2, b/2) versus t for a / h * = 100 fig. 2 w (a/2, b/2) versus t for a / h * = 10, 14 and 100 (a) x  versus z for a / h* = 10 (b) xz  versus z for a / h* = 10 (c) x  versus t for a / h* = 10 (d) x  versus t for a / h* = 14 fig. 3 stresses versus z and t for a / h * = 10, 14 and 100 (continued) -1.0e-11 -5.0e-12 0.0e+00 5.0e-12 0 0.5 1 1.5 2 2.5 3 w (a/2, b/2) t -2.5e-12 -1.3e-12 0.0e+00 1.3e-12 2.5e-12 0 0.5 1 1.5 2 2.5 3 w (a/2, b/2) t -2.5e-12 -1.3e-12 0.0e+00 1.3e-12 2.5e-12 0 0.5 1 1.5 2 2.5 3 w (a/2, b/2) t -1.2e-03 -8.0e-04 -4.0e-04 1.0e-19 4.0e-04 8.0e-04 -0.5 -0.25 0 0.25 0.5 𝜎𝑥 z / h* -1.6e-12 -1.2e-12 -8.0e-13 -4.0e-13 -0.5 -0.25 0 0.25 0.5 𝜎𝑥z z / h* -9.20e-04 -9.15e-04 -9.10e-04 -9.05e-04 -9.00e-04 0 0.5 1 1.5 2 2.5 3 𝜎𝑥 t -9.140e -04 -9.135e -04 -9.130e -04 -9.125e -04 -9.120e -04 0 0.5 1 1.5 2 2.5 3 𝜎𝑥 t advances in technology innovation, vol. 3, no. 3, 2018, pp. 109 117 cop y right © taeti 115 (e) x  versus t for a / h* = 100 fig. 3 stresses versus z and t for a / h * = 10, 14 and 100 normal stress 𝜎𝑥 and shear stress 𝜎𝑥𝑧 are three-dimensional components and usually in functions of x, y, and z. typically their values vary through the plate thickness for the airflow over the upper surface of fgm plates. fig. 3(a) shows the normal stress 𝜎𝑥 (unit gpa) versus z and fig. 3 (b) shows the shear stress 𝜎𝑥𝑧 (unit gpa) versus z at center position (x = a / 2, y = b / 2) of plates, respectively at t = 0.1s, a / h * =10 and a / b = 1. the absolute value (9.15e-04 gpa) of normal stress 𝜎𝑥 at z = 0.5h * is found in the much greater value than the value (1.3e-12 gpa) of shear stress 𝜎𝑥𝑧 at z = 0.5h * , thus the normal stress 𝜎𝑥 can be treated as the dominated stress for the airflow over the upper surface of fgm plates. figs. 3(c)-3(e) shows the time responses of the dominated stresses 𝜎𝑥 (unit gpa) at the center position of outer surface z = 0.5h * as the analyses of deflection case in fig. 2 for l / h * =10, 14 and thin l / h * =100, respectively. the maximum absolute value of stresses 𝜎𝑥 is 9.15e-04 gpa occurs constantly in the periods t =0.2s-3s for l / h * =10. fig. 4 shows the center deflection amplitude ( / 2, / 2)w a b (unit mm) versus t for all different values rn (from 0.1 to 10) of fgm plates calculated and varied values of k  , for the airflow over the upper surface of fgm plates * / 10l h  , / 1a b  , * h = 1.2 mm, 1 h = 2 h = 0.6 mm, 1 100t k , at 3t  s. the maximu m value of center deflection amplitude is 1.22e-11mm occurs at 653t k for rn = 10. the center deflection amplitude values are all small, decreasing versus t from 653t k to 1000t k , for rn = 2, 5 and 10, they can withstand for higher temperature ( 1000t k ) of environment. the center deflection amplitude values are all small, increasing versus t from 653t k to 1000t k , for rn = 0.1, 0.2 and 0.5. fig. 4 w (a/2, b/2) versus t fig. 5 shows the dominated stresses x  (unit gpa ) at the center position of outer surface * 0.5z h versus t for all diffe rent values rn of fgm p lates as the analyses of deflection case in fig. 4. the absolute values of do minated stresses x versus t are increasing (fro m 100t k to 653t k ) and then decreasing (fro m 653t k to 1000t k ) for rn = 1, a ll decreasing (fro m 100t k to 1000t k ) for rn = 10, all increasing (fro m 100t k to 1000t k ) for rn = 0.1, 0.2, 0.5 and 2. -9.1250e-04 -9.1225e-04 -9.1200e-04 0 0.5 1 1.5 2 2.5 3 𝜎𝑥 t -1.5e-11 -1.0e-11 -5.0e-12 0.0e+00 5.0e-12 1.0e-11 1.5e-11 0 250 500 750 1000 rn=10 rn=5 rn=2 rn=1 rn=0.5 rn=0.2 rn=0.1 w (a/2, b/2) t advances in technology innovation, vol. 3, no. 3, 2018, pp. 109 117 cop y right © taeti 116 fig. 5 x  versus t the effects of with and without the fluid flow over the upper surface of fgm plates on the center deflection and normal stress for / 1a b  , * h = 1.2 mm, 1 h = 2 h = 0.6 mm, 1 n r  , k  = 0.117077, 653t k , 1 100t k are also considered as follows. the fig. 6 (a) shows the response values of center deflection amplitude ( / 2, / 2)w a b (unit mm) versus time t for * / 10a h  with and without airflow over the upper surface of fgm plates. the absolute values of center deflection amplitude for without airflow (0.111079 mm) are much greater than that with airflow (5.26e-14 mm) at t = 0.001s for * / 10a h  . fig. 6 (b) shows the normal stress x  (unit gpa) versus z at the center position ( / 2, / 2x a y b  ) of plates at t = 0.001s for * / 10a h  with and without airflow over the upper surface of fgm plates. the normal stresses are almost in the same values for with ( x  =-9.12e-04 gpa at * 0.5z h ) and without ( x  =-9.25e-04 gpa at * 0.5z h ) airflow cases. (a) w (a/2, b/2) versus t for a / h * = 10 (b) w (a/2, b/2) versus t for a / h * = 10 fig. 6 w (a/2, b/2) versus t and x  versus z for a / h* = 10 with and without airflow 4. conclusions in this study, the gdq solutions have been obtained and investigated for the deflections and stresses in the thermal vibration of fgm plates by considering the varied effects of shear correction coefficient and the airflo w over the upper surface of fgm plates. the gdq res ults have shown that varied values of k  are usually functions of * h , rn and t. the absolute value of center deflection a mplitude is 7.04e-12 mm occurs at t = 0.2s for * / 10a h  at 653t k . the center deflect ion amp litude values are a ll sma ll and decreasing versus t fro m 653t k to 1000t k , for rn = 2, 5 and 10, the fgm plates also can withstand for higher temperature environment. references [1] s. a. fazelzadeh, s. pouresmaeeli, and e. ghavanloo, “aeroelastic characteristics of functionally graded carbon nanotube-reinforced composite plates under a supersonic flow,” computer methods in applied mechanics and engineering vol. 285, pp. 714-729, march 2015. -1.1e-03 -1.0e-03 -9.0e-04 -8.0e-04 -7.0e-04 -6.0e-04 -5.0e-04 0 250 500 750 1000 rn=10 rn=5 rn=2 rn=1 rn=0.5 rn=0.2 rn=0.1 𝜎𝑥 t -0.20 -0.15 -0.10 -0.05 0.00 0.05 0 0.5 1 1.5 2 2.5 3 without air flow with air flow w (a/2, b/2) t -1.2e-03 -8.0e-04 -4.0e-04 1.0e-19 4.0e-04 8.0e-04 -0.5 -0.25 0 0.25 0.5 without air flow with air flow 𝜎𝑥 z / h* advances in technology innovation, vol. 3, no. 3, 2018, pp. 109 117 cop y right © taeti 117 [2] c. y. lee and j. h. kim, “ eva luation of ho mogenized effective p roperties for fgm panels in ae ro -therma l environments,” composite structures, vol. 120, pp. 442-450, february 2015. [3] m. rafiee, m. mohammadi, b. s. aragh, and h. yaghoobi, “nonlinear free and forced thermo-electro-aero-elastic vibration and dynamic response of piezoelectric functionally graded laminated composite shells: part ii: numerical results,” composite structures, vol. 103, pp. 188-196, september 2013. [4] m. ghadimi, m. dardel, m. h. pashaei, and m. m. barzegari, “effects of geometric imperfections on the aeroelastic behavior of functionally graded wings in supersonic flow,” aerospace science and technology , vol. 23, no. 1, pp. 492-504, december 2012. [5] t. prakash, m. k. singha, and m. ganapathi, “a finite element study on the large amplitude flexural vibration characteristics of fgm plates under aerodynamic load,” international journal of non-linear mechanics , vol. 47, no. 5, pp. 439-447, june 2012. [6] k. j. sohn and j. h. kim, “structural stability of functionally graded panels subjected to aero -thermal loads,” composite structures, vol. 82, no. 3, pp. 317-325, february 2008. [7] s. a. fazelzadeh and m. hosseini, “aerothermoelastic behavior of supersonic rotating thin -walled beams made of functionally graded materials,” journal of fluids and structures , vol. 23, no. 8, pp. 1251-1264, november 2007. [8] l. wu, h. wang, and d. wang, “dynamic stability analysis of fgm plates by the moving least squares differential quadrature method,” composite structures, vol. 77, no. 3, pp. 383-394, february 2007. [9] h. m. nava zi and h. haddadpour, “aero-thermoelastic stability of functionally graded plates,” co mposite structures, vol. 80, no. 4, pp. 580-587, october 2007. [10] t. prakash and m. ganapathi, “supersonic flutter characteristics of functionally graded flat panels including thermal effects,” composite structures, vol. 72, no. 1, pp. 10-18, january 2006. [11] c. c. hong, “effects of varied shear correction on the thermal vibration of functionally-graded material shells in an unsteady supersonic flow,” aerospace, vol. 4, no. 1, pp. 1-15, december 2017. [12] f. tornabene, n. fantuzzi, f. ubertini, and e. viola, “strong formulation finite element method based on differential quadrature: a survey,” applied mechanics reviews , vol. 67, no. 2, pp. 1-55, march 2015. [13] c. c. hong, “thermal v ibration and transient response of magnetostrictive functionally graded materia l plates, ” european journal of mechanics a/solids , vol. 43, pp. 78-88, january-february 2014. [14] c. c. hong, “rapid heating induced vibration of circular cylindrical shells with magnetostrictive functionally graded material,” archives of civil and mechanical engineering , vol. 14, no. 4, pp. 710-720, august 2014. [15] c. c. hong, “thermal v ibration of magnetostrictive functionally graded materia l shells,” european journal of mechanics a/solids , vol. 40, pp. 114-122, july-august 2013. [16] c. c. hong, “rapid heating induced vibration of magnetostrictive functionally graded material plates, ” transactions of the asme, journal of vibration and acoustics , vol. 134, no. 2, pp. 1-11, january 2012. [17] f. tornabene and e. viola , “ free vibrat ion analysis of functionally graded panels and shells of revolution,” meccanica vol. 44, no. 3, pp. 255-281, june 2009. [18] s. h. chi and y. l. chung, “mechanical behavior of functionally graded material plates under transverse load, part i: analysis,” international journal of solids and structures , vol. 43, no. 13, pp. 3657-3674, june 2006. [19] m. s. qatu, r. w. sullivan, and w. wang, “recent research advances on the dynamic analysis of composite shells: 2000-2009,” composite structures, vol. 93, no. 1, pp. 14-31, december 2010. [20] s. j. lee and j. n. reddy, “non-linear response of laminated composite plates under thermomecha nical loading,” international journal of non-linear mechanics , vol. 40, no. 7, pp. 971-985, september 2005. [21] j. m. whitney, structural analysis of la minated anisotropic plates, lancaster: pennsylvania, usa, technomic pub lishing company, inc., 1987. [22] h. s. shen, “ nonlinear thermal bending response of fgm plates due to heat condition,” co mposites part b: engineering, vol. 38, no. 2, pp. 201-215, march 2007. [23] c. shu and b. e. richards, “high resolution of natural convection in a s quare cavity by generalized differential quadrature,” proc. 3rd international conf. advanced in numerical methods in engineering: theory and applications, swansea, vol. 2, pp. 978-985, 1990. [24] c. w. bert, s. k. jang, and a. g. striz, “nonlinear bending analysis of orthotropic rectangular plates by the method of differential quadrature,” computational mechanics , vol. 5, no. 2-3, pp. 217-226, march 1989. [25] c. shu and h. du, “implementation of clamped and simply supported boundary conditions in the gdq free vibration analyses of beams and plates,” international journal of solids and structures , vol. 34, no. 7, pp. 819-835, march 1997. microsoft word aiti#11678 20230809 (in press 2).docx advances in technology innovation, vol. x, no. x, 20xx, pp. xx-xx english language proofreader: chih-wei chang a hidden semi-markov model for predicting production cycle time using bluetooth low energy data karishma agrawal*, supachai vorapojpisut department of electrical and computer engineering, faculty of engineering, thammasat school of engineering, thammasat university, thailand received 02 march 2023; received in revised form 09 june 2023; accepted 10 june 2023 doi: https://doi.org/10.46604/aiti.2023.11678 abstract this study proposes a statistical model to characterize the temporal characteristics of an entire production process. the model utilizes received signal strength indicator (rssi) data obtained from a bluetooth low energy (ble) network. a hidden semi-markov model (hsmm) is formulated based on the characteristics of the production process, and the forward-backward algorithm is employed to re-estimate the probability distribution of state durations. the proposed method is validated through numerical, simulation, and real-world experiments, yielding promising results. the results show that the kullback-leibler divergence (kld) score of 0.1843, while the simulation achieves an average vector distance score of 0.9740. the real-time experiment also shows a reasonable accuracy, with an average hsmm estimated throughput time of 30.48 epochs, compared to the average real throughput time of 33.99 epochs. overall, the model serves as a valuable tool for predicting the cycle time and throughput time of a production line. keywords: bluetooth low energy, received signal strength indicator, hidden semi-markov model, learning problem 1. introduction production is a systematic and organized process that involves a series of interconnected working areas or stages. as shown in fig. 1, the production process includes a specific set of tasks or operations that transform raw materials into finished products. to ensure that the production process is completed on time, it is critical to have accurate and up-to-date product status as well as location information. many industries now use barcodes or rfid [1-2] to track products in production areas. however, the main disadvantage of employing rfid or barcodes is the possibility of human errors. for example, data loss or inaccuracy may occur if the rfid tag/barcode is not read correctly or if a worker scans the wrong rfid tag/barcode. this work uses bluetooth low energy (ble) technology to collect received signal strength indicator (rssi) and timestamp data, which provides the proximity of products being manufactured. even though the scope of ble detection is larger than that of barcode/rfid, the collected data from ble devices usually suffers from signal strength fluctuation and noisy environments. consequently, the application of ble tracking is much more limited compared to barcode/rfid. fig. 1 general example of a production process * corresponding author. e-mail address: karishma.agra@dome.tu.ac.th advances in technology innovation, vol. x, no. x, 20xx, pp. xx-xx 2 the key performance indicators (kpis) of a production process can be defined by using temporal behaviors [3-4]. examples of kpis for production processes are cycle time and throughput time. cycle time [3] refers to the time it takes to finish a cycle in a particular operation. in the meantime, throughput time [4] is the time it takes to process a certain material from raw material to a final product. based on the kpis, manufacturers can identify areas for improvement, reduce costs, and optimize productivity. therefore, this information is of great significance for manufacturers who attempt to produce highquality products in the shortest timeframe to remain competitive. the traditional method of manually monitoring products to calculate cycle time had various limitations, such as the potential for human errors and limited data collection. at present, manufacturers have adopted a precise and efficient approach to analyzing historical data for cycle time calculation, that is, to use barcode/rfid [1-2] and statistics. the queuing theory is a widely used and traditional approach to analyzing and predicting the temporal properties of production processes. this theory states that production tasks arrive, await maintenance, and then exit once the process is complete. however, queuing models [5] usually consider the timing of each station independently and do not capture the dynamic nature of the whole production process. the main objective of this study is to predict cycle time and throughput time in a production line using rssi values. to achieve this, hidden semi-markov models (hsmm) are used to analyze the rssi and timestamp data. the paper presents an algorithm to determine the model parameters. the study contributes to estimating both cycle and throughput time by using rssi values. furthermore, it is a new approach compared to previous studies [6] that mainly focused on tracking and monitoring activities. this paper provides a more specific and comprehensive version of the work [7] previously presented, including detailed explanations and additional experiments. the paper is structured as follows: section 2 provides a background on ble technology and introduces the concept of hsmm. while section 3 elaborates on the problem and presents mathematical solutions based on rssi-based data. in section 4, three experiments are conducted in numerical, simulation, and real-world settings. section 5 analyzes three case studies to demonstrate the practical application of the proposed method. finally, section 6 discusses and concludes how the cycle time of a product can be computed with the mathematical solutions presented in this paper. 2. backgrounds this section provides an overview of ble technology and introduces the concept of hsmm (hidden semi-markov model). it presents the fundamental characteristics of ble and its practical applications. furthermore, it discusses the potential use of hsmm as a solution to enhance the capabilities of ble networks in comprehending the attributes of the production process. 2.1. bluetooth low energy (ble) technology ble [6-7] technology offers wireless connectivity with low cost, low power consumption, and simple installation. hence, it’s a popular choice for tracking applications. a ble network consists of two types of devices: ble scanners and ble tags. typically, ble tags are battery-powered devices that periodically broadcast their ids, packet-embedded data, and rssi data [8]. ble scanners detect and collect these signals along with a timestamp and send the information to the server. rssi is a measurement that indicates the strength of the signal between a ble tag and a scanner. additionally, higher values represent a stronger and more dependable connection. decibels (dbm) is the standard unit of measurement for rssi. there are two major problems in ble technology to monitor products during production. firstly, the rssi can fluctuate significantly, even if the distance between the ble scanner and tag remains unchanged, due to interference, random noise, and multipath effects [8]. secondly, the inconsistent receiver sensitivity of each device [8] limits the communication range and generates random and insufficient rssi data from the scanner. to illustrate these problems, fig. 2 demonstrates the correlation between the rssi values and distance obtained from an experiment conducted with wemos r32 (esp32) boards. advances in technology innovation, vol. x, no. x, 20xx, pp. xx-xx 3 various studies have been conducted to resolve the issue of rssi fluctuation in object-tracking applications. two popular methods for indoor localization are trilateration and fingerprinting, which utilize rssi values obtained from multiple ble scanners to estimate the 2d coordinates of a ble tag [8]. however, these methods may not be suitable for production environments due to their large coverage area. proximity detection presents a simpler approach, which estimates an object's location by detecting the presence of a ble tag near a ble scanner [8]. nonetheless, it may have limitations when multiple ble scanners give conflicting results. fig. 2 rssi vs distance of ble devices recent studies have shown that the hidden markov model (hmm) has the potential to explain indoor trajectory with rssi data. for example, han et al. [9] used the ble signal from the smartwatch and implemented an hmm algorithm to determine the user’s indoor location and the most likely sequence based on previous rssi data. similarly, arslan et al. [10] utilized realtime ble rssi data from the construction site and hmm to identify semantic trajectories for extracting worker movement awareness to improve safety. however, conventional hmm approaches do not specify the dwell time of the object in each state, which is not likely to happen in practical scenarios [11-12]. the hsmm has been proposed as an extension of the conventional hmm, including the modeling of the duration of each state. as a result, hsmm has been successfully applied in indoor tracking [11-13] and various other domains, including human-robot collaboration (hrc) [14] in assembly scenarios and machine condition recognition [15]. despite these advancements, there is insufficient literature specifically focusing on understanding and predicting temporal characteristics in production settings. 2.2. hidden semi-markov model (hsmm) the hmm [9-11] is a double-layer stochastic process that uses the markov property to model a sequence of hidden states and observed outputs. the first layer represents the underlying and hidden states of the process, which are not directly observable. the second layer symbolizes the emission states, which are related to the observable and measurable states. the concept of hmm comprises five components, including state (s), initial state probability (π), transition probability matrix (a), observation state (o), and emission probability matrix (b). the emission probability matrix (b) defines the probability of an observation �� at any given time instance t to determine a state � ∈ �1,2, ⋯ , �. the transition probability matrix a represents the probability of moving from one hidden state to another. these transition probabilities are to calculate the probability of states given a sequence of observations. the markov property means that the probability of transitioning from one state to another only depends on the current state. and the hmm uses the markov property as the prediction of the next state based on the current state. since rssi information being used as observation o may not imply the exact location of a device, this paper considers various stages of production, such as material preparation, process, and completion as the set of hidden states s. advances in technology innovation, vol. x, no. x, 20xx, pp. xx-xx 4 the transition probability matrix a in an hmm is time-independent, which is not suitable for modeling temporal behaviors. the hsmm [14-17] addresses this issue by introducing the variable duration d, which signifies the time in each state. fig. 3 illustrates two primary characteristics of an hsmm that distinguish it from an hmm [13]: (1) the probability of self-transition is assumed to be zero, i.e., � � 0 no return to the same state. (2) the transition from a single state depends upon the sojourn duration or the time spent in that state. these properties of the hsmm allow for the representation of variable state durations, making it possible to estimate the duration or cycle time of each stage in the production process which conventional hmm fails to achieve. (a) hmm (b) hsmm fig. 3 graphical representation of hmm and hsmm in this study, the ble scanning pattern is analyzed by using a discrete time approach, where an epoch is defined as � ∈ �1,2, ⋯ , ∞�. to simplify computations, the duration of each state is treated as a discrete random variable and bounded by an integer value � � 1. the probability of duration for state i have given d is represented by � ���, where � ∈ �1, � and � ∈ �1, ��. ( ) ( )i r tp d p d s i= = (1) the constraint for � ��� is that the sum of � ��� from � � 1 to d is equal to 1 for all states � [7]. there are three basic problems with the hsmm framework used in academic works [7, 13]: (1) likelihood problem: this involves computing the likelihood of an observed sequence o given an hsmm parameter �. (2) decoding problem: this refers to the problem of finding the optimal state sequence given an observed sequence o and an hsmm parameter �. (3) learning problem: this involves estimating an hsmm parameter � from an observed sequence o. it re-estimates the distribution of transitions, emission probabilities, and state durations. the primary objective of this paper is to address the learning problem associated with determining the probability distribution for state durations, which is formulated based on the characteristics of the production process. as for the forward-backward algorithm, it is employed to re-estimate the probability distribution of state durations using a learning problem. advances in technology innovation, vol. x, no. x, 20xx, pp. xx-xx 5 3. problem statement this study considers a ble system with a ble tag on each product and a ble scanner in each working area to collect data on the movement of products. the main objective of this study is to use hsmm concepts to create a mathematical model that can predict the cycle time of production based on the duration characteristics of each working area. this study formulates the product tracking problem as the hsmm model � � ��, �, �, ��, where a represents the transition matrix with duration probability matrix p for each state, b signifies the emission matrix, and � is the initial state probability. the following kpis are studied in the paper: (1) cycle time: cycle time is the time it takes to complete a single task in a specific working area during the production process. the calculation of the cycle time can be expressed as: i i ict et st= − (2) where �� represents the average cycle time for a product to complete its operations in the working area �. �� is the time when the operation begins, and �� is the time when it ends. (2) throughput time: the throughput time is the total time required for a product to move through the production process, and it represents: 1 throughput time n i i ct = =  (3) where �� represents the cycle time for a product to complete its operations in the working area �. the following assumptions are made for a production process based on how raw materials move in the production line to produce finished products [7]. (1) the working areas (stations) are represented as the hidden states � ∈ �1,2, ⋯ , �. each manufactured product is equipped with a specific tag id, and a ble tag is attached to it. ble scanners scan these ble tags and capture rssi values. these rssi values at time � are collected by multiple scanners and used as an observation �� � !��"#,� , !��"$,� , ⋯ , !��"%,� &. (2) the emission probability b consists of ' ��� � � (��� |�� � �, * , ∑ � are assumed to be fixed but may be updated periodically based on the collected rssi data [11-12]: ( ) ( ) 1 , m m m i t t i i m b o o µ = = n (4) where * , and ∑ , are the mean and covariance for the -�. gaussian component for observed rssi values at the state �, respectively. the gaussian distribution is denoted by / . to simplify the calculation, the assumption is made that the observations are independent across time [12-13]. therefore, ' ���#:�$� can be expressed as the product of the probabilities of observing each rssi data point from epoch �1 to �2, and it can be expressed as: ( ) ( )21: 2 1 t i i tt t t t b o b o== ∏ (5) (3) define ���, 1� to represent the probability of moving from area � to area 1 as the ble tag moves through areas sequentially from 1 to n. the calculation of ���, 1� can be simplified as: ( ) 1 ; 1 , 0 ; j i a i j elsewhere    = + = (6) (4) the initial state probability � is set to 1, indicating that every sequence will always begin at state 1. advances in technology innovation, vol. x, no. x, 20xx, pp. xx-xx 6 this study focuses on the learning problem of the hsmm framework to capture the cycle time of the production process. to re-estimate the duration probability of each state, it is necessary to solve likelihood problems using a forward-backward procedure (��|�� with two subdivided variables: forward and backward. in hsmm, the forward variable 2� ��� represents the likelihood of being in state � given the partial observations �#:� up to time �, for a given hsmm parameter � [13]. ( ) ( ) ( ) ( ) ( ) ( )min , * 1: 1:1 , t d t i i it t d t d td i p o s ends at t i p d b oα λ α − − +== =  (7) ( ) ( ) ( )* ,1: 1, 1 n t i i j it i i p o s begins at t iα λ α α = = + =  (8) where � � 1, ⋯ , �, � � 1, ⋯ , , 2� ∗��� is the joint probability of obtaining the observation sequence up to time �, and entry to state � begins at the next time � + 1, given the model �. the calculation of the proximity likelihood for a location is performed using the following formula: � ( ) 1 t t i n s arg max iα < < = (9) the definition of the hsmm backward variable 5� ∗��� can be found in formula [13]: ( ) ( ) ( ) ( ) ( ) , * : : 11 , t t d t i i it t t d t t dd i p o s begins at t i p d b oβ λ β − + + −= = =  (10) ( ) ( ) ( )* ,: 1 1, n t i i j tt t j i p o s ends at t jβ λ α β = = − =  (11) where � � �, ⋯ ,1, � � 1, ⋯ , , 5� ∗��� is the probability of the partial observation sequence from time � to the end, given that the system leaves the state � at the previous time � − 1 and with a model �. the re-estimation of the duration probability �̂ ��� matrix, based on the forward and backward variables, is expressed as [13]: ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) 1 * * 1 : 11 1 * * 1 : 11 1 ˆ t d i it t t d t dt i d t d i it t t d t dd t i p d b o i p d i p d b o i α β α β − + − + − += − + − + − += = =    (12) the outline of the learning problem used in the paper is as follows [7]. algorithm 1: re-estimation of � ��� input: a, b, coupled sequences �� 8 , �8 �; 9 � 1,2, ⋯ , : training: set the values of � ��� with uniform distribution initialize the (; ��� matrix loop for 9 � 1, ⋯ , : loop for � � 1, ⋯ , � calculate 2� ���, 2� ∗���, 5� ���, 5� ∗��� compute the (; ��� (; ��� � (; ��� + �̂ ��� � ��� � (; ���/� output: � ��� the � ��� parameter in the hsmm represents the probability of the system staying in the state � for a duration of d time units. in other words, the � ��� distribution of state � governs the cycle time for that specific stage of the production process. it is important to note that the �̂ ��� eq. (12) is computed from the entire sequence. to predict temporal properties, a sequence is advances in technology innovation, vol. x, no. x, 20xx, pp. xx-xx 7 constructed by sampling durations from the transition probabilities and �̂ ���. this allows for the estimation of cycle times and throughput time. overall, the hsmm can be modeled from collected rssi and timestamp data in a production process, and the �̂ ��� parameter plays a key role in estimating these kpis. 4. experiments three studies were conducted to evaluate the performance of the proposed algorithm based on numerical, simulated, and real-world experiments. furthermore, the evaluation aimed to validate the effectiveness of the algorithm across various scenarios and provide empirical evidence of its capabilities. 4.1. numerical experiment in this experiment, 1000 state/observation sequences were based on given hsmm parameters (�, a, b, p). each coupled sequence contains two types of data: observed (rssi) signals and state sequences. hsmm parameters were defined as follows: (1) the number of working areas n is 7. (2) the maximum duration � � 8 epochs. (3) probability distribution matrix ( � �� ���� for the system to leave the state � after d duration is given by. 0 0.6 0.4 0 0 0 0 0 0 0 0 0 0.3 0.1 0.3 0.3 0 0 0 0 0 0.7 0.1 0.2 0 0 0.3 0.4 0.3 0 0 0 0 0 0 0.4 0.2 0.2 0.2 0 0 0.4 0.4 0.2 0 0 0 0 1 0 0 0 0 0 0 0 p           =             (13) (4) the values of rssi are computed using a gaussian mixture model with mean and variance: 50 75 100 100 100 100 100 78 55 80 100 100 100 100 95 78 56 87 100 100 100 100 100 70 50 79 100 100 100 100 100 78 54 80 100 100 100 100 100 79 50 80 100 100 100 100 100 100 53 µ − − − − − − −    − − − − − − −    − − − − − − −   = − − − − − − −   − − − − − − −   − − − − − − −  − − − − − − −   (14) 5 2 1 1 1 1 1 1 3 2 1 1 1 1 1 2 4 2 1 1 1 1 1 3 5 2 1 1 1 1 1 2 3 2 1 1 1 1 1 2 4 2 1 1 1 1 2 2 3 σ           =             (15) the sequence generation is outlined using algorithm 2. then, algorithm 1 is used with the generated data to estimate the probability distribution matrix �̂. algorithm 2: sequence generation input: � � �, *, ?, �, (, is the number of states and m is rssi observations advances in technology innovation, vol. x, no. x, 20xx, pp. xx-xx 8 output: hidden state s and rssi observations o for � � 1 �@ if � �� 1 �abcbd�be � select a state based on the given � initial probabilities else �abcbd�be � select a state using the previous state and a probability end �fg���@habcbd�be � based on the �abcbd�be and p probabilities, choose the duration for 9 � 1 �@ �fg���@habcbd�be for � 1 �@ i ��-, 9� � generate random value from * and ? with �abcbd�be end end �ejkl� mn �1: �fg���@habcbd�be � � �abcbd�be � � o�, �ejkl� mn p � � o�, @p end 4.2. simulated experiment the matlab bluetooth toolbox enables users to create a simulation environment of ble devices. fig. 4 shows the 2d coordinates of the ble scanners with moving ble tags. this study uses matlab bluetooth toolbox to simulate ble transmission to find rssi values when tags are moved along the given trajectory. one thousand state/observation sequences are generated from the movement of ble tags in the simulated environment. the ble tags are advanced randomly according to the selected duration. fig. 5 provides an example of the observed rssi values and state in a simulated sequence. fig. 4 layout with ble 2-d scanner coordinates fig. 5 example of simulated data 4.3. real-world experiment in a real-world scenario, different stations were prepared to collect rssi data from ble devices using the components: wemos r32 boards as ble tags and scanners, wi-fi access points, and the firebase database. the data collection was performed in a u-shaped area of 7 × 4 m with approximately 2 meters between each scanner as shown in fig. 6. advances in technology innovation, vol. x, no. x, 20xx, pp. xx-xx 9 (1) wemos r32 boards are programmed as ble scanners that connect with the wifi access point and report id/rssi/timestamp to the firebase database. (2) wemos r32 boards are programmed as ble tags were moved from one working area to another according to a predetermined sequence. (3) each ble tag is reset at the endpoint to renew the tag id. fig. 6 ble scanner allocation in a real-time experiment the scanning period for all ble scanners is set at 5 seconds. 25 sets of data were collected by moving the ble tags through a pre-determined sequence from one working area to the next. as illustrated in fig. 7, the results show that not all scanners can receive the rssi signal. the occupation of the tag in each ble scanner area is determined using the relationship between distance and rssi in fig. 2. fig. 7 rssi measurements in real-world environments 5. discussion this section presents the results of three studies conducted to assess the effectiveness of the algorithm in different scenarios. the objective is to validate its capability in comprehending the characteristics of the production process through empirical evidence. advances in technology innovation, vol. x, no. x, 20xx, pp. xx-xx 10 5.1. analysis of numerical experiment in this study, the kullback-leibler divergence (kld) [13] is utilized as the distance metric to determine the similarity between the original p and the re-estimated �̂. ( ) �( )( ) ( ) ( ) �( )1 d i i i ikl d i p d d p d p d p d ln p d = =  (16) the average kld value for the entire system approximately stands at 0.1843. this value is close to zero, indicating that the estimated �qr ��� is nearly equivalent to the actual � ���. to show the similarities, fig. 8 displays a comparison between the predicted duration using hsmm and the actual duration with a range represented by minimum and maximum values. the average throughput time based on numerical data is 28.51 epochs, while the hsmm estimated average throughput time is 27.69 epochs. fig. 9 shows the trend of the average kld values concerning the number of coupled sequences used in the training phase. it demonstrates that the proposed algorithm converges quickly with a reasonable amount of data. as the number of samples increased from 100 to 1000, the kld value decreased from 1.7281 to 0.1843. moreover, the evidence supports that algorithm 1 is a reliable and accurate method for estimating the duration probability. fig. 8 predicted vs actual duration comparison fig. 9 trend of average kld value in the numerical experiment 5.2. analysis of the simulation experiment in the simulated experiment, 1000 sequences were generated, and then algorithm 1 was employed to parameterize an hsmm model. the average throughput time for the simulated sequences was 20.53 epochs, while the estimated throughput time using the hsmm model was 19.89 epochs. this result is promising because it shows that the estimated throughput time is close to the simulated sequences. the study was continued by running simulations and estimations for another 500 state sequences as part of the forecast set. the similarity between the forecasted and simulated sequences is evaluated with run-length encoding (rle) and vector distance. rle is a simple data compression technique that works well for sequences where the same value occurs in many consecutive manners. it encodes the sequence by replacing consecutive repeating values with a single value and its count. for example, the rle input presents in “aaabbcccc” and the output is “3a2b4c”. the vector distance is used to measure the similarity of rle repetitive counts as a vector. the vector distance is calculated by: ( ) 2 x yvector distance = − (17) advances in technology innovation, vol. x, no. x, 20xx, pp. xx-xx 11 where x represents the simulated data, and y is the forecasted data. for example, the rle of the simulated data x becomes “3a2b4c” and the rle of the forecasted data y becomes “3a2b4c”. in this case, the vector distance between the rle encoded sequences can be calculated using the following formula. ( ) ( ) ( ) 2 2 2 2 3 3 + 2 2 + 4 3 = 0+0+1 =1− − − (18) the lower the vector distance it becomes, the greater similarity it results in. a value of zero indicates that two sequences are identical. the comparison of vector distances between one sample to 500 samples is depicted in fig. 10. the average vector distance decreased from 3.8531 for one sample to 0.9740 for 500 samples. thus, it indicates that the forecasted and simulated sequences had become more similar. this result shows that the estimated duration becomes more precise as the number of samples increases. fig. 10 vector distance for simulated and forecasted sequences 5.3. analysis of real-world experiment the objective of the real experiment is to demonstrate the effectiveness of algorithm 1 with realistic data. to achieve this, data were collected from 7 ble scanners for 25 ble tags. the dataset was filtered based on tag id to retain the rssi values associated with the scanner number. the filtered data was input into algorithm 1, which is used to re-estimate parameters �̂. fig. 11 compares the state sequences generated by the trained hsmm model to the actual data collected during the experiments. fig. 11 hidden state alignment example advances in technology innovation, vol. x, no. x, 20xx, pp. xx-xx 12 table 1 state durations (unit of 5 secs.) area (state) average cycle time (in epochs) real-world hsmm #1 3.28 2.88 #2 6.68 5.88 #3 5.84 6.16 #4 8.76 8.56 #5 3.36 1.52 #6 4.60 4.44 #7 1.48 1.04 throughput time (in epochs) 33.99 30.48 table 1 presents a comparison of the average state duration values between the collected data and those generated by the proposed hsmm model. the average throughput time for the realistic data is 33.99 epochs (169.95 seconds), while the average throughput time for the hsmm model is 30.48 epochs (152.4 seconds), with a difference of only 3.51 epochs. the difference highlights the accuracy of the hsmm model in capturing the dynamics of the system. the close match between the realistic data and the hsmm model suggests that the model is capable of accurately predicting throughput time and state sequences based on the rssi data. 6. conclusions this study aims to overcome the limitations of previous research by uncovering temporal behavior and identifying potential correlations in the hsmm with rssi data from the production line. to achieve this, the proposed method utilized two key steps. first, the method involved updating duration probabilities using forward and backward procedures. it allows probable determination of observed sequences in an hsmm, which is a crucial aspect of understanding temporal behavior. second, learning procedures were employed to update the duration parameter, enabling the model to adapt and improve its predictions based on the collected data. the proximity between the actual and predicted values in experiments highlights the reasonable accuracy of the hsmm model. it indicates that the proposed method effectively predicted temporal behavior. future work could involve further refinement of the method to include the condition when there is no rssi available. additionally, it is important to note that the limitations of this work include the dependence on accurate and consistent rssi data and the assumption of stationary behavior in the production process. despite this limitation, this study has shown the potential of rssi and timestamp data for predicting production metrics in a manufacturing setting. conflicts of interest the authors declare no conflict of interest. references [1] w. c. tan and m. s. sidhu, “review of rfid and iot integration in supply chain management,” operations research perspectives, vol. 9, article no. 100229, january 2022. [2] w. c. chen, m. h. nguyen, and p. h. tai, “an intelligent manufacturing system for injection molding,” proceedings of engineering and technology innovation, vol. 8, pp. 09-14, april 2018. [3] j. v. tébar-rubio, f. j. ramírez, and m. j. ruiz-ortega, “conducting action research to improve operational efficiency in manufacturing: the case of a first-tier automotive supplier,” systemic practice and action research, vol. 36, no. 3, pp. 427-459, june 2023. [4] r. joppen, s. von enzberg, j. gundlach, a. kühn, and r. dumitrescu, “key performance indicators in the production of the future,” procedia cirp, vol. 81, pp. 759-764, 2019. [5] s. gao, j. i. u. rubrico, t. higashi, t. kobayashi, k. taneda, and j. ota, “efficient throughput analysis of production lines based on modular queues,” ieee access, vol. 7, pp. 95314-95326, july 2019. advances in technology innovation, vol. x, no. x, 20xx, pp. xx-xx 13 [6] p. bencak, d. hercog, and t. lerher, “indoor positioning system based on bluetooth low energy technology and a nature-inspired optimization algorithm,” electronics, vol. 11, no. 3, article no. 308, february 2022. [7] s. vorapojpisut and k. agrawal, “modeling of manufacturing processes using hidden semi-markov model and rssi data,” 17th international joint symposium on artificial intelligence and natural language processing, november 2022. [8] t. kluge, c. groba, and t. springer, “trilateration, fingerprinting, and centroid: taking indoor positioning with bluetooth le to the wild,” 21st international symposium on “a world of wireless, mobile and multimedia networks”, august-september 2020. [9] d. han, h. rho, and s. lim, “hmm-based indoor localization using smart watches’ ble signals,” 6th international conference on future internet of things and cloud, pp. 296-302, august 2018. [10] m. arslan, c. cruz, and d. ginhac, “semantic trajectory insights for worker safety in dynamic environments,” automation in construction, vol. 106, article no. 102854, october 2019. [11] s. sun, y. li, w. s. t. rowe, x. wang, a. kealy, and b. moran, “practical evaluation of a crowdsourcing indoor localization system using hidden markov models,” ieee sensors journal, vol. 19, no. 20, pp. 9332-9340, october 2019. [12] s. sun, x. wang, b. moran, a. al-hourani, and w. s. t. rowe, “radio source localization using received signal strength in a multipath environment,” 22nd international conference on information fusion, pp. 1-6, july 2019. [13] s. sun, x. wang, b. moran, and w. s. t. rowe, “a hidden semi-markov model for indoor radio source localization using received signal strength,” signal processing, vol. 166, article no. 107230, january 2020. [14] c. h. lin, k. j. wang, a. a. tadesse, and b. h. woldegiorgis, “human-robot collaboration empowered by hidden semi-markov model for operator behaviour prediction in a smart assembly system,” journal of manufacturing systems, vol. 62, pp. 317-333, january 2022. [15] w. yang and l. chen, “machine condition recognition via hidden semi-markov model,” computers & industrial engineering, vol. 158, article no. 107430, august 2021. [16] b. mor, s. garhwal, and a. kumar, “a systematic review of hidden markov models and their applications,” archives of computational methods in engineering, vol. 28, no. 3, pp. 1429-1448, may 2021. [17] k. li, c. qiu, x. zhou, m. chen, y. lin, x. jia, et al., “modeling and tagging of time sequence signals in the milling process based on an improved hidden semi-markov model,” expert systems with applications, vol. 205, article no. 117758, november 2022. copyright© by the authors. licensee taeti, taiwan. this article is an open-access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation, vol. 2, no. 1, 2017, pp. 01 07 1 assessment of a charge transport model for ldpe through conduction current measurement anggie chandra kusumasembada 1,* , gilbert teyssedre 2 , severine le roy 2 , laurent boudou 2 , ngapuli irmea sinisuka 1 1 department of electrical engineering, bandung institute of technology, bandung, indonesia. 2 laplace (laboratoire plasma et conversion d’energie), université de toulouse, cnrs, ups, inpt; 118 route de narbonne, f-31062 toulouse cedex 9, france. received 31 march 2016; received in revised form 01 may 2016; accepted 02 may 2016 abstract conduction current measurements have been widely used to characterize charge transport behavior in insulating materials. however, the interpretation of transport mechanisms and more generally of non-linear processes from current measurements alone is not straightforward. for this reason, space charge measurements, on the one hand, and models of charge transport encompassing charge generation, trapping and transport have been developed. the completeness and accuracy of a model can be assessed only if a substantial range of stress conditions, being field and temperature for the current topics, is available. the purpose of this communication is to enrich the investigation of low density polyethylene ldpe insulation material characteristic using conduction current measurement. measurements were conducted on 250 μm thick ldpe samples, for dc fields in the range 2 to 50 kv/mm and for temperatures from 20 to 70°c. experimental data, i.e. transient current in charge/discharge and quasi-steady state currents are compared to the prediction of a bipolar transport model that has been developed over the last years and fitted to the case of ldpe. the deviation of model results is substantial, with essentially an overestimation of the non-linearity of the current-field dependence. these differences are discussed along with prospects from improving the model. aside from these modelling approaches, we show that thermal preconditioning of samples appears to be influential in the measured apparent conductivity. keywords: ldpe, conduction current, charge transport 1. introduction investigation on polyethylene material as electrical insulating material receives significant attention as its demand increases, especially since polyethylene is more and more used in high voltage dc cables application. current understandings regarding charge transport and mechanisms related to space charge will benefit for reaching better performance and reliable hvdc insulation systems. low density polyethylene (ldpe) as part of polyethylene group is the main concern in this paper. ldpe charge transport characterization by conduction current has been conducted in various researches. charging mechanism characteristics by means of threshold representation [1, 2] on space charge features provides one way to describe its character. comparison between polymers was also conducted, as ldpe vs hdpe – i.e. high density pe [2], ldpe vs. ldpe + antioxidant vs. xlpe – i.e. crosslinked pe [3], and xlpe vs epdm, i.e. rubber with ethylene-propylene-diene monomer [4]. the purpose of this paper is to enrich study regarding ldpe charge transport by presenting measurement results on charging and discharging currents and comparing results with an already available model of conduction based on bipolar charge generation and transport. the model has been parameterized and refined over the years and encompass charge injection, charge transport and charge recombination [5, 6, 7]. its optimization is based on experimental results relevant to charging/discharging current, space charge measurements and electroluminescence. * corresponding author, email: anggi.kusumasembada@gmail.com advances in technology innovation, vol. 2, no. 1, 2017, pp. 01 07 2 copyright © taeti along these objectives, preconditioning factors that influence the measurement and how modelling reacts to it are also investigated. indeed, variations in preconditioning is considered as time elapsed before measurement once the sample is set to a given temperature, or previously applied electrical stress in the course of measurements. this could explain variations observed in output results. the model that has been developed can indeed integrate to some extent this thermo-electrical history. 2. experimental procedure 2.1. conduction current measurements ldpe material was considered for the conducted investigation. ldpe without antioxidant, provided by borealis, was chosen. for the measurement, ldpe pellets were first press-moulded to be prepared as plaque specimen. plaque sample was processed at 140 °c under a pressure of 3 bars for 20 minutes. completed samples are disks of 8 cm in diameter with 250 ± 10 μm in thickness. kapton was used as template and pressing layer during press moulding, the template was arranged to create plaques of 250 μm thickness. for ensuring measurement contact, each sample was provided with gold electrodes by sputtering, the gold layer has 5 cm in diameter and 30nm in thickness. a silicone layer was laid at the periphery of the electrode to avoid edge effects. several samples were prepared to be tested in different thermal preconditioning procedure: no thermal preconditioning, 1-hour, <52 hours, and >52 hours thermal conditioning. conduction current measurements were registered in air at 3 different temperatures (30, 50, 70° c) and 13 values of the applied electric fields (2, 4, 6, 8, 10, 13, 16, 19, 22, 25, 30, 40, 50 kv/mm). the sample was clamped between two brass electrodes with polished surface. the current was recorded through a keithley 617 ammeter with a 2 s dwelling time under charging state for 3 hours, and discharging state for 1 hour. extracted quantities mainly are transient currents and quasi steady state current, which will be derived as current density and conductivity. current density from transient current measurement provides information related to conduction mechanism, current density is deduced by the following equation: 𝐽(𝑡) = 𝐼(𝑡) 𝐴 (1) where i(t) is the measured current and a the area of electrode (20 cm²). conductivity value of the insulation is calculated using the following equation: 𝜎 = 𝐽∞ 𝐸0 (2) where j∞ is the steady state current density, e0 the applied field. in this work, the current values utilized in current density equation are quasi steady state current values which were taken during the last 800 s of the 10800 s measurement time of charging current measurement. 2.2. model the model features bipolar transport and trapping of electrons and holes. the model was created to fit experimental measurement of current, space charge, thermos-stimulated currents, electroluminescence, etc. [6, 8]. fig. 1 below illustrates the schematic representation of the model for ldpe [5]. it is a two levels model for each kind of carriers, defining so 4 kinds of species: mobile and trapped electrons and same for holes. fig. 1 physical model schematic the set of equations constituting the model is common to transport models in dielectric media, being liquids, solids or gases: transport of electrons and holes, neglecting diffusion: 𝑗𝑒(𝑥) = 𝜇𝑒𝑛𝑒𝜇(𝑥)𝐸(𝑥) (3) 𝑗ℎ(𝑥) = 𝜇ℎ𝑛ℎ𝜇(𝑥)𝐸(𝑥) (4) where μe is the electron mobility, μh the hole mobility, neμ the mobile electron density, nhμ the mobile hole density, e the electric field, and x the spatial coordinate. advances in technology innovation, vol. 2, no. 1, 2017, pp. 01 07 3 copyright © taeti poisson’s equation: 𝜕𝐸(𝑥) 𝜕𝑥 = 𝜌(𝑥) 𝜀 (5) where ε is the dielectric permittivity, ρ the net charge density. the conservation equation, meaning that local variations of density of given specie are due to transport or to variation as a source: 𝜕𝑛𝑖(𝑡) 𝜕𝑡 + 𝜕𝑗𝑖(𝑥) 𝜕𝑥 = 𝑠𝑖 (6) where s encompasses the source terms (i.e. trapping, detrapping, and recombination process). those source terms have for example the following form for mobile electrons: 𝑠1 = −𝑆1𝑛ℎ𝑡𝑛𝑒𝜇 − 𝑆3𝑛ℎ𝜇𝑛𝑒𝜇 − 𝐵𝑒𝑛𝑒𝜇 (1 − 𝑛𝑒𝑡 𝑛0𝑒𝑡 ) + 𝑣. 𝑒𝑥𝑝 ( 𝑤𝑡𝑟𝑒 𝑘𝑏𝑇 )𝑛𝑒𝑡 (7) si is the recombination coefficient, be the trapping coefficient for electrons and bh the trapping coefficient for holes. densities of trapped holes and electrons are stated with net and nht, while maximal trap densities of electrons and holes are stated by n0et and n0ht. w𝑡𝑟𝑒 is the detrapping barrier height. modeling of charge injection during applied voltage at each electrode is expressed with the following equation, for electrons as an example: 𝐽𝑒(0) = 𝐴𝑇 2𝑒𝑥𝑝 (− 𝑤𝑒 𝑘𝑇 ) ⌊𝑒𝑥𝑝 ( 𝑒 𝑘𝑇 √ 𝑒𝐸(0) 4𝜋𝜀 ) − 1⌋ (8) equation for charge extraction at the other side is written as follows: 𝐽𝑒(𝑑) = 𝜇𝑒𝑛𝑒𝜇(𝑑)𝐸(𝑑) (9) the total current density through the material which incorporates the electrons and holes current density follows: 𝐽(𝑡) = 1 𝐷 ∫ (𝐽𝑒(𝑥,𝑡) + 𝐽ℎ(𝑥,𝑡) 𝐷 0 )𝑑𝑥 (10) latest refinements incorporated into the model concern the use of langevin-type recombination, where the recombination coefficients are function of the mobility of the carriers [8]. the mobility is a constant effective mobility that already takes into account the possible trapping and detrapping of charges into shallow traps. 3. results and discussion 3.1. transient current measurements 0 3600 7200 10800 0 50 100 150 200 c u rr e n t (p a ) time (s) 2 to 50kv/mm fig. 2 charging current transient in 250 μm thick ldpe plaque measured at 30°c for 13 different values of the applied field ranging from 2 to 30kv/mm (cf. §2.1) 10000 10400 10800 0 10 20 30 q u a s i s te a d y s ta te c u rr e n t (p a ) time (s) 2 kv/mm 4 kv/mm 6 kv/mm 8 kv/mm 10 kv/mm 13 kv/mm 16 kv/mm 19 kv/mm 22 kv/mm 25 kv/mm 30 kv/mm 40 kv/mm 50 kv/mm fig. 3 quasi steady-state charging current in 250 μm thick ldpe plaque measured at 30°c (long time data of fig. 2) transient current measurements were realized on ldpe and examples of the results obtained at 30°c are shown in fig. 2 and fig. 3. fig. 2 depicts a quickly reducing current magnitude toward a steady state. fig. 3 focuses on the longer time region in which current have been averaged for plotting the characteristics. the current appears indeed steady at this scale. in fig. advances in technology innovation, vol. 2, no. 1, 2017, pp. 01 07 4 copyright © taeti 2 and 3, some noise is detected for high field steps possibly due to some micro-discharges in the high voltage range (all measurements were realized in air at atmospheric pressure). however this noise does not have substantial impact on estimated conductivity. we shall see later on that the present charge time configuration (3 hours) is not sufficient to achieve steady state. 3.2. precondition effect on measurement substantial change in time in the conductivity has been reported recently depending on pre-annealing time of ldpe and crosslinked polyethylene (xlpe) by h. ghorbani [9]. indeed, aside from the apparent decrease in conductivity as a function of stressing time measured at 50°c, there was also a decrease in conductivity with the pre-storage time at 50°c before the measurements. similarly, montanari et al. [3] reported on a decrease in the transient currents measured at room temperature when ldpe or xlpe samples have been previously thermally annealed for 90h at 50°c [3]. the current decrease was all over the measurement time of 3h. as measurements realized here are relatively long (50h per temperature step) when realizing consecutively the all set of polarization/depolarization steps, there can be an evolution of the conductivity due to this conditioning. 30 40 50 60 70 10 -16 10 -15 c o n d u c ti v it y (  -1 .m -1 ) temperature (°c) set a set b 1 0 t o 5 0 k v /m m fig. 4 comparison of conductivity with varying preconditioning of ldpe plaque sample under various applied field. set a was conducted continuously for the 3 temperatures with same sample (sample 1), while set b was conducted with a different fresh sample for each temperature value. quasi steady current values plotted in fig. 4 were obtained for the following cases: (a) set a: same sample stressed successively in the different field steps and different temperatures (30, then 50 and 70°c); (b) set b: one different sample for the different temperature levels. in all the results, the apparent conductivity for the fresh sample is higher than the one for the previously stressed. quantitatively the difference between the two steps is a drop of the conductivity by about 30 to 50% after pre-stressing. the variation is about the same for 50 and 70°c. the trends are consistent with the previously reported results. however, the situation is a bit more complicated in the present case compared to results of h. ghorbani as here the history concerns both the thermal and electrical conditioning: both are likely to decrease conductivity for different reasons: (a) electrical pre-stressing may generate space charge into the insulation, e.g. close to the injecting electrode: this will act as counter-field for further charge injection and is a process that can explain the decay in time of the current. if trapped charges are stable, the memory effect will be generated owing to the pre-existing charge. a transport model might anticipate such features. (b) thermal pre-stressing can induce drying/outgassing of sample if some residues are present, and/or change of the morphology as crystallinity. substantial changes of crystallinity were reported h. ghorbani [9] for the long term testing at 50°c on ldpe and xlpe. crystallinity of ldpe increases as heat treatment time lengthened. this can in turn alter the electrical response of the material. one way to distinguish morphological vs. residue effects would be to probe again samples one exposed to ambient conditions. 3.3. current vs. field characteristics several works have reported on the threshold of current vs. field for various specimens such as: xlpe, rubber, hdpe, and ldpe [2-4]. the current density is plotted as a function of field following this 'threshold' representation – i.e. log-log plot in fig. 5. samples did not undergo thermal preconditioning before measurements apart from the stabilization time at the set temadvances in technology innovation, vol. 2, no. 1, 2017, pp. 01 07 5 copyright © taeti perature. it can be seen that the j-e curves of sample 2 (50° c) and sample 3 (70° c) starts unlike temperature-field characteristics of polyethylene. indeed, the current tend to drop while increasing the applied field, which is an unexpected behavior. the effect is not observed for the measurement at 30°c. the explanation for the effect is most probably the one described above, i.e. a conditioning effect at the measurement temperature: as measurement at each voltage level requires 4 h time, these conditioning effects can be significant. beyond a field of 10kv/mm, the curves for the different temperatures have similar shapes. 1 10 100 10 -10 10 -9 10 -8 10 -7 2.8 2.3 2.3 c u r r e n t d e n s it y ( a /m ²) applied field (kv/mm) 30°c 50°c 70°c (1) fig. 5 current density log-log plot for ldpe plaque specimens. electrical threshold are defined by applying fitting lines the slope of j-e plot as in fig. 5 should define whether ohmic conduction or ionic or space charge limited conduction [10] take place in the transport processes. current density plot in this work shows variation of the slope as field increases. notably for 30°c data, the characteristic changes from a nearly ohmic regime (slope close to 1) to highly non-linear regime with a slope estimated to 2.3. the threshold takes place at about 13 kv/mm where charge transport behavior of ldpe changes. for higher temperature, it is difficult to decide if the threshold varies owing to the evolution in time of the response of the material. 4. comparison to model outputs 4.1. model results fig. 6 shows a comparison between experimental and simulated current transients obtained at 20°c. it must be stressed here that the model has been applied with the currently available data set of coefficients, see [5], best fitted to measurements at room temperature, but without any attempt of later optimization of the parameters. 0 3600 7200 10800 10 -10 10 -9 10 -8 10 -7 2 6 10 16 22 30 c u r r e n t d e n s it y ( a /m ² ) time (s) fig. 6 transient current density vs. time: comparison between actual measurements symbols and simulation (solid lines) at 20° c for fields of 2, 6, 10, 16, 22 and 30 kv/mm as shown in the legend results for the lower voltage are relatively noisy owing to the fact that the average current is small, of the order of 0.2 pa in average at long time. it can be seen from fig. 6 that at measurement time longer than 2000 s, charge transport characteristics appear differently between lower field and higher field. simulation result at 20 °c shows increasing transient current as applied voltage rises. results from the model reveal a steep transient in the first minutes followed by a slower transient over 1 h for the step at 2kv/mm and this step is not so pronounced for higher fields. the steeper decay is due to the fact that an initial density of charges is supposed to be present in the material and this was to cope with experimental electroluminescence measurements [5]. these pre-existing charges move under the effect of the applied field. the slower decay in current results from the injection at the electrode, followed by transport and trapping of the both type of carriers. for the other steps in field, the preexisting charge is that computed along the depolarization stages following the previous steps. at this stage of the model, orientation polarization processes are not included: they could be present and at the origin of the decay in the experimental current. advances in technology innovation, vol. 2, no. 1, 2017, pp. 01 07 6 copyright © taeti 4.2. discussion as stated above, the transient part of the charging current has several reasons for being not reproduced, notably the fact that the orientation polarization contributions are not included in the model. this has been done recently in the case of poly (ethylene naphthalate), a polymer known for having strong dipolar response. impedance spectroscopy data available in the frequency domain have been fitted to known relaxation functions according to identified relaxation processes. then it was converted to the time domain and this orientation polarization contribution could be treated separately from the transport aspects. the translation to the case of ldpe is not straightforward, as it is a non-polar material and therefore polarization if any has to be related to polar residues as oxidized groups for example. second, the weak magnitude of such processes would make it tricky to analyze in the frequency domain for conversion in the time domain. so, we are currently not in position to explicitly dissociate orientation polarization from space charge processes in ldpe. this, all the more that efforts in parameterizing the model at short time have been put more on electroluminescence features – reflecting charge recombination processes than on transient current. the behavior at long time should in principle fit more directly to the experimental. however, comparing between experiment and simulation, the difference almost reaches one decade in quasi steady state part for 2 kv/mm. with increasing field, the difference tends to be less, but still is by a factor 2 for 30kv/mm. so, on the all, the model tend to over-estimate the non-linearity of the response of the material. this is so while the rough material for making films is the same as that used for preparing samples on which the model is based. one could argue on the necessity of refining the model such as integrating polarization and using the latest developments in the physical hypotheses in it [8]. however, we would like to make the point on the experimental features. the main differences, regarding current measurement results are that previous experiments [5] were achieved in dry atmosphere instead of air. although polarization was 3 h, the selected field values for long polarization protocols were much coarser with data at 10, 40, 60 and 80kv/mm. presently, the first source of inconsistency to be fixed is the difference in experimental results regarding conductivity data, which were an order of magnitude higher in [7-5]. one possible route is the method of preparation of the films: ghorbani [9] showed that the nature of the films used as cover layer may indeed have a great influence on conductivity values. these results point on the carefulness to be given in the preparation of samples and measurements on insulating materials, and more generally on the definition of the system that we intend to probe and model. electrode nature and processing conditions constitute full field of potential discrepancy between experimental results. 5. conclusions our purpose in this paper was to assess the robustness of the outputs of a charge transport model by comparing the model predictions to a set of experimental data obtained at various fields and temperatures on ldpe. current-voltage characteristics reveal once more that a threshold at around 10kv/mm define a change in conduction mechanisms beyond which conductivity is clearly non-linear. experimental data obtained at 20 °c have revealed a substantial deviation from results expected from the transport model. perhaps one of the first conclusions is that one to be extremely careful in defining the system, i.e. material, processing, electroding, conditioning and measurement conditions as they may greatly impact the results. second, there is interesting memory or pre-conditioning effects to control and understand. part of it is of pure electrical nature, as previous charging effects on a given characteristic. in principle, if the model is complete, it should predict charge storage and subsequent impact on transport. a more difficult case to handle is thermal preconditioning effects, which are revealed here through a decrease of the measured current, but that would demand further investigation as its origin can be multiple, resorting to physical evolution of the structure or to moieties evacuation. acknowledgement one of the author, a.c kusumasembada gratefully acknowledges support from pt. sanggadelima nusantara. advances in technology innovation, vol. 2, no. 1, 2017, pp. 01 07 7 copyright © taeti references [1] l. a. dissado, c. laurent, g. c. montanari, and p. h. f. morshuis, “demonstrating a threshold for trapped space charge accumulation in solid dielectrics under dc field,” ieee trans. dielectr. electr. insul., vol. 12, pp. 612-620, 2005. [2] g. c. montanari, g. mazzanti, f. palmieri, and a. motori, “investigation of charge transport and trapping in ldpe and hdpe through space charge and conduction current measurement,” ieee 7th international conference on solid dielectric, pp. 240-244, june 2001. [3] g. c. montanari, c. laurent, g. teyssedre, a. campus, and u. h. nilsson, “from ldpe to xlpe: investigating the change of electrical properties. part i: space charge, conduction, and lifetime,” ieee trans. dielectr. electr. insul., vol. 12, pp. 438-446, 2005. [4] t. t. n. vu, g. teyssedre, b. vissouvanadin, s. le roy, and c. laurent, “correlating conductivity and space charge measurement in multi-dielectrics under various electrical and thermal stresses,” ieee trans. dielectr. electr. insul., vol. 22, pp. 117-127, 2015. [5] s. le roy, g. teyssedre, c. laurent, g. c. montanari, and f. palmieri, “description of charge transport in polyethylene using a fluid model with a constant mobility: fitting model and experiments,” j. phys. d: appl. phys., vol. 39, pp. 1427-1435, 2006. [6] s. le roy, f. baudoin, v. griseri, c. laurent, and g. teyssèdre, “space charge modeling in electron-beam irradiated polyethylene: fitting model and experiments,” j. appl. phys. 112, 023704, 2012. [7] m. q. hoang, l. boudou, s. le roy, and g. teyssèdre, “electrical characterization of ldpe films using thermo-stimulated depolarization currents method: measurement and simulation based on a transport model,” proc. 8ème conférence société française d’electrostatique, proc. sfe-8, pp. 1-5, 2012. [8] s. le roy, g. teyssèdre, and c. laurent, “modelling space charge in a cable geometry,” ieee trans. dielectr. electr. insul., in press. [9] h. ghorbani, “characterization of conduction and polarization properties of hvdc cable xlpe insulation materials,” licentiate thesis, school of electrical engineering, kth royal institute of technology, stockholm, 2015. [10] g. teyssedre and c. laurent, “charge transport modeling in insulating polymers: from molecular to macroscopic scale,” ieee trans. dielectr. electr. insul., vol. 12, pp. 857-875, 2005.  advances in technology innovation, vol. 4, no. 2, 2019, pp. 59-72 complex independent laboratory tests to signify fuel economy and emissions under real world driving albert boretti 1,* , petros lappas 2 1 independent scientist, bundoora, australia 2 rmit, bundoora, australia received 27 july 2018; received in revised form 21 august 2018; accepted 13 september 2018 abstract we discuss the shortcomings of the current european union emission rules, which are not likely to produce any improvement of fuel economy and emissions during real-world driving vs. the pre-diesel gate rules, because of flaws in the prescribed laboratory and on-the-road tests. the present chassis dynamometer emissions certification test is oversimplified, not representative, and not properly performed. the simple addition of generic on-the-road real-world driving tests does not help, because these tests have poor accuracy and repeatability. therefore, we suggest the use of a more realistic, complicated, suite of driving cycles, to be rigorously performed with chassis dynamometer equipment by independent bodies, rather than the original equipment manufacturers (oems). additionally, we suggest that emission limits be revised, to permit the optimal mix between different engine technologies and different fuels. keywords: internal combustion engine, emissions, fuel consumption, certification 1. introduction in the aftermath of the volkswagen group (vw) emissions scandal of 2015, boretti [1], the move to real driving emissions (rde) tests has been drastically accelerated. the new emission rules will have to be closer to real-world driving and designed to deliver the best global results in terms of environment, economy, and sustainability. certainly, the emission tests will have to be much more complicated and performed by independent bodies. the new emissions tests must be carefully designed to not simply phase out the internal combustion engine (ice) for the sake of electric mobility, which can in many cases have serious environmental and economic disadvantages. for a historical perspective, vw was accused of having programmed their diesel engines to “activate certain emissions controls only during laboratory emissions testing” causing “the vehicles' nox output to meet us standards during regulatory testing but emit up to 40 times more nox in real-world driving”. the accusation was technically flawed, as the emission standards solely required compliance during the prescribed procedure, the ftp, in full laboratory tests. there is no emission standard that requires a vehicle to comply with certain emissions or fuel consumption limits for any other vehicle use or test equipment. the vw cars homologated for the us market were compliant with the ftp. however, these vehicles were then emitting different amounts of pollutant and using different amounts of fuel when operated differently. this was no surprise. there is no reason to expect the same limits to be satisfied over every other possible operation of the vehicle under every possible condition. the measurements collected by the consultants of the icct were performed without any scientific method by using inappropriate schedules vaguely defined “real world driving”. simply, the cars were tested on the road. they discovered what * corresponding author. e-mail address: a.a.boretti@gmail.com advances in technology innovation, vol. 4, no. 2, 2019, pp. 59-72 60 many automotive engineers already knew: the oems tune their products to perform well on the prescribed certification tests. apart from the specific “defeat device” (en.wikipedia.org/wiki/defeat_device) used to turn-off the nox emissions control when not covering the regulatory cycle (so yielding better fuel economy and engine performance), every oem uses techniques to deliver good performance on the homologation test without penalizing the real-world driving operation. while vw certainly used defeat devices, it is not known if other oems manipulated their controls to identify testing conditions, intentionally turning on and off features during testing and not testing. it is however shared knowledge that a specific calibration is performed for the test cycle of the country of homologation. the calibration is not advanced, nor mature enough, to reduce emissions during off cycle as well as it does during the specific cycle of homologation. until new emission standards are defined, a vehicle remains compliant if the emissions fall below the limits during the prescribed certification tests. compliance is not determined in any other way. as discussed in boretti [1], the united states environmental protection agency (epa) and the united states prosecutes, initiated by the action of the international council on clean transportation (icct) have forced the oem towards a rapid transition to battery based electric mobility. the latest decision of the eu to ban the internal combustion engine (ice) by 2030, which does not leave much opportunity to improve the ice, is a logical consequence of the vw scandal, as predicted in boretti [1]. however, electric mobility is very far from becoming a real substitute, boretti [2]. hence, there is a need to properly improve emission standards. as europe was the driving force behind the rise of the diesel for passenger car applications, it was not a surprise that europe was the first to react to the scandal with different measures, including the definition of new emission standards. the procedure based on the anachronistic new european driving cycle (nedc), a cycle designed by using a ruler, is being replaced by a procedure based on the worldwide harmonized light vehicles test cycle (wltc), with the transition from nedc to wltc occurring over the period 2017-2019. passenger cars must still satisfy pollutant emissions below the euro 6 thresholds and be labelled with their co2 emissions. while there are several wltc test cycles applicable to vehicle categories of different power-to-mass ratio, a given passenger car must perform within the standards only over a single test cycle. a single test cycle was also only required by the nedc. however, in addition, rde supplementary tests have been introduced. the feedback from independent chassis dynamometer and rde tests is still missing, as it is premature to expect such a feedback from new rules not even being completely introduced. however, these two measures – the nedc replaced by the wltc, and the introduction of rde supplements are not expected to ensure that the real-world emissions are really within the agreed euro 6 emission standards. within australia, the current minimum standard for new light vehicles is adr 79/04, based on the euro 5 standards, australian government department of infrastructure, regional development and cities [3]. this rule is still being reviewed to consider whether australia should adopt the euro 6 standards for new light vehicles. it is still unclear if, and for how long, australia will continue to accept vehicles declared compliant by the original equipment manufacturers (oem) with euro 5 and euro 6 standards. it is also unclear whether emissions compliance will be determined over the chassis dynamometer with the nedc or wltc, and/or over laboratory tests, and/or supplemented with rde tests. the requirement in euro 6 regulations to further lower the nox limit of diesel-powered vehicles down to the same limit as gasoline vehicles have significantly raised the cost of the diesel. at the same time, the regulations reduce the competitiveness of the diesel regarding fuel economy. the euro 6 regulations drastically increased the cost of diesel, making them a very difficult commercial proposition for subcompacts and minicars, while still providing advantages for long highway drives in large sedans. the novel emission standards will also have to carefully weigh the minimum amount permitted for all the regulated pollutants, to permit the best mix of different solutions in the different segments, and for different uses, to achieve the best global outcomes. advances in technology innovation, vol. 4, no. 2, 2019, pp. 59-72 61 fig. 1 presents an example of traffic congestion, in melbourne, victoria, australia. the drivers of passenger cars and light-duty vehicles experience real-world driving conditions that have very little in common with the short, highly stylized, simplified, driving schedule that are used for emission certification. melbourne’s suburban and city roads are struggling to meet the demands of the fast-growing population. peak hour has stretched to six hours a day across melbourne’s congested road network, 6:30am to 9 am and 3 pm to 6:30 pm each day. this translates in longer, more frequent stops, and driving at a small fraction of the speed limit during the lengthened peak-congestion periods. according to www.tomtom.com, in 2016 the increase in the morning peak travel time for melbourne was 55%, while the increase in the evening peak travel time was 58%. the global congestion level was 33% more time if compared to free flow uncongested conditions, or 34 minutes extra travel time per day, for two average rides of 103 minutes per day. therefore, during peak hours, commuters may spend more than one hour and one half for every section of their return trip, experiencing driving conditions that have nothing to do with the new european driving cycle (nedc) or the worldwide harmonized light vehicles test cycle (wltc) used in the certification test. fig. 1 traffic congestion in melbourne, australia the aim of the paper is to discuss the changes needed in the european union emission rules to ensure that the compliant vehicles will then meet the expectations of emissions and fuel economy when driven on the road, while accounting for the use of all the available fuels and different engine technologies (e.g., stoichiometric, lean-burning homogeneous or stratified, premixed or diffusion combustion, spark or compression ignition, see boretti and watson [4], boretti [5], boretti and watson [6], boretti [7], boretti, paudel and tempia [8]). diversification of engine technologies and combustion fuel supply, not only traditional fossil fuels, but also alternative fossil fuels and synthetic fuels and biofuels, see dhinesh et al. [9], annamalai et al. [10], subramani, parthasarathy, balasubramanian, and ramalingam [11], dhinesh and annamalai [12], dhinesh, raj, kalaiselvan and krishna moorthy [13], should also be considered in determining the emission levels. this is because emissions cannot be the same for different fuels and different engine technologies, as every fuel and engine technology has its own emission advantages and disadvantages. the paper first examines the flaws of the current laboratory procedures, then the pitfalls of the current additional on-road tests. a suite-of-cycles laboratory test is then proposed to address the issues of repeatability, representability, and reliability. a discussion and conclusion section then close the paper. http://www.tomtom.com/ advances in technology innovation, vol. 4, no. 2, 2019, pp. 59-72 62 2. flaws of current laboratory procedures fig. 2 nedc cycle. velocity vs. time fig. 3 nedc cycle. velocity vs. distance fig. 4 wltc cycle. velocity vs. time advances in technology innovation, vol. 4, no. 2, 2019, pp. 59-72 63 fig. 5 wltc cycle. velocity vs. distance figs. 2-3 and figs. 4-5 present the nedc and wltc schedules, while table 1 and table 2 summarizes the main parameters of the two cycles. all the euro rules, including the latest euro 6 rules, have been based for many years on the unrealistic, simplistic, short, cold-start, new european driving cycle (nedc). the nedc is a stylized driving speed pattern with low accelerations, constant speed cruises, and many idling events. this cycle was designed by using a ruler, with 4 very simplistic city driving sections, followed by 1 extra urban driving section that is concluded with a sharp deceleration from 120 km/h to rest, as shown in fig. 2. the nedc cycle, and the procedure that was using this cycle had many critics, as it is reported in the literature: table 1 nedc parameters phase duration [s] stop duration [s] distance [m] percentage stop vmax [km/h] vave w/o stops [km/h] vave w/ stops [km/h] amin [m/s 2 ] amax [m/s 2 ] ece15-1 195 57 994.1 29.23% 50 25.93 18.35 -0.93 1.042 ece15-2 195 57 994.1 29.23% 50 25.93 18.35 -0.93 1.042 ece15-3 195 57 994.1 29.23% 50 25.93 18.35 -0.93 1.042 ece15-4 195 57 994.1 29.23% 50 25.93 18.35 -0.93 1.042 eudc 400 39 6954.9 9.75% 120 69.36 62.59 -1.39 0.833 total 1180 267 10931.3 table 2 wltc parameters phase duration [s] stop duration [s] distance [m] percentage stop vmax [km/h] vave w/o stops [km/h] vave w/ stops [km/h] amin [m/s 2 ] amax [m/s 2 ] low 3 589 156 3095 26.50% 56.5 25.7 18.9 -1.47 1.47 medium 3-2 433 48 4756 11.10% 76.6 44.5 39.5 -1.49 1.57 high 3-2 455 31 7162 6.80% 97.4 60.8 56.7 -1.49 1.58 extra-high 3 323 7 8254 2.20% 131.3 94 92 -1.21 1.03 total 1800 242 23267 i. the nedc cycle is out-of-date, it does not represent real-world driving conditions (parker [14], mock, german, bandivadekar, and riemersma [15], plotkin [16], kågeson [17], steen [18], shaw [19], peker [20]). ii. the procedure is not accurate (parker [14]). iii. the fixed speeds, gear shift points and accelerations of the nedc offer possibilities for manufacturers to engage in “cycle beating” to optimize engine emission performance to the corresponding operating points of the test cycle, (mock, german, bandivadekar, and riemersma [15]). advances in technology innovation, vol. 4, no. 2, 2019, pp. 59-72 64 iv. emissions from typical driving conditions would be much higher (mock, german, bandivadekar, and riemersma [15]). v. it is also claimed that the tests can be conducted at 2 km/h below the required speed for better fuel economy (mock, german, bandivadekar, and riemersma [15]). vi. roof-rails and passenger door-mirror can be removed for the test (mock, german, bandivadekar, and riemersma [15]). vii. tire-pressures for the test can be set above the recommended values thus reducing rolling-resistance (mock, german, bandivadekar, and riemersma [15]). viii. the cycle allows large emission differences between test and reality (kågeson [17]). ix. the cycle has numerous loopholes (steen [18], shaw [19], peker [20]). x. the nedc tests are not necessarily repeatable and comparable (steen [18]). xi. the nedc tests can be performed using optional economy settings which will not typically be selected by drivers (steen [8]). xii. the nedc test-cycle is performed with air-conditioning switched off (steen [18]). xiii. no official-body polices the tests (parker [14]). xiv. compliance is declared by the oem without independent verification (parker [14]). xv. the latest sharp deceleration from 120 km/h to zero never occurs in real-world driving (parker [14]). xvi. this deceleration ending the cycle especially rewards hybrid vehicles (parker [14]). xvii. the last high-speed driving section in the nedc reduces the fuel consumption over the cycle (parker [14]). xviii. the cycle does not include traffic congestion (parker [14]). xix. for diesel cars no significant nox reductions have been achieved after 13 years of stricter standards (european federation for transport and environment [21]). xx. similarly, to the nox emissions, even if not claimed so far, it is also questionable whether there has been real improvement in co2 emissions and fuel economy over the last 13 years. about these two latter statements, it is a matter of fact (us environmental protection agency [22]) that the epa has resolved a civil enforcement case against vw alleging violations of the clean air act by approximately 590,000 diesel vehicles sold in the united states. “specifically, the u.s. complaint alleges that each of these vehicles contains, as part of the engine control module, certain computer algorithms and calibrations that cause the emissions control system of those vehicles to perform differently during normal vehicle operation and use than during emissions testing. the u.s. complaint alleges that these computer algorithms and calibrations are prohibited defeat devices under the caa, and that during normal vehicle operation and use, the cars emit levels of nox significantly in excess of the epa compliant levels.” it is a fact admitted by vw in a court-of-law settlement, that the emissions of their diesel cars under real-world driving were largely exceeding the present emission standards in the united states. the actual on-the-road nox emissions of these 590,000 diesel vehicles were well above the thresholds, not only of the latest emission standards, but also on the prior emission standards, franco, sánchez, german, and mock [23], thompson, carder, besch, thiruvengadam, and kappanna [24]. the lack of any improvement of the diesel in real-world driving has been the result of flawed procedures and lack of any independent control on the compliance and co2 performances that are declared by the oem. this is not addressed by simply replacing the nedc cycle of fig. 2 with the wltc of fig. 4 and introducing inaccurate and irreproducible rde supplementary tests. advances in technology innovation, vol. 4, no. 2, 2019, pp. 59-72 65 despite being longer (1,800 s vs. 1,180 s, or 23,262 m vs. 11,013 m), the wltc is still a single cycle, hence the engine and vehicle control can still be highly optimized to perform well over this cycle, leaving other vehicle operation condition unassessed. the wltc still closes with the unrealistic final deceleration from 130 km/h to rest, that disproportionately rewards hybrids and electric vehicles. the wltc still has a significant highway driving component resulting in much lower fuel consumption than during the real-world driving in congested city traffic. the wltc does not have the many stops of normal city driving. the reasons why the nedc test procedure failed was not only because the driving schedule was simplistic and unrealistic. it was also because the results of the fuel economy and emissions tests were declared by the oem and not measured by independent organizations. this yielded claims of fuel economy and co2 emissions that were inconsistent, parker [14]. this meant that the consumption of one litre of nominally equivalent petrol or diesel fuel produced very different co2 emission results. in addition, real-world driving fuel economy and emissions were largely underrated. a major flaw was that the tests were performed without any detailed specification of the actual fuel properties, and the fuel economy is not directly measured, but rather, it is indirectly computed from inaccurate formulae that use the emitted pollutant levels. in the unece r101, the fuel consumption is calculated from the measured emissions of hydrocarbons, carbon monoxide, and carbon dioxide. the fuel consumption (fc), expressed in liters per 100 km is calculated by means of the following simplistic formula for petrol engines fc = (0.118/ρ) ∙[(0.848·hc) +(0.429·co) +(0.273∙co2)] (1) and the following formula for diesel engines fc=(0.116/ρ) ∙[(0.861∙hc) +(0.429∙co) +(0.273∙co2)] (2) in the previous equations hc, co and co2 are the measured emission of hydrocarbons, carbon monoxide and carbon dioxide in g/km, while ρ is the (specific) density in kg/liter of the test fuel. as explained in parker [14], the above simplistic procedure is significantly inaccurate. pump gasoline and diesel fuels are refined products containing mixtures of many different hydrocarbons. in the case of gasoline, they may also contain a significant amount of oxygenates. the fuel properties, including the density, carbon and energy content strongly vary from one fuel pump to another. the above equations are theoretically valid only for a precise c/h ratio for gasoline or diesel. diesel fuel may have a carbon content that varies from c3 to c25, or a mass percentage of carbon in the range 84-87% with a hydrogen range of 13-16%. the specific density may range from 0.81 to 0.89. the lower heating value may range from 41.87 to 44.19 mj/kg. boiling temperatures, cetane number, and viscosity also vary significantly. a gasoline fuel may vary even more, even before accounting for mixing with methanol ch 3oh and ethanol c2h5oh. a gasoline fuel may have a carbon content ranging from c4 to c12, or a percentage of carbon in weight 85-88% and hydrogen 12-15%. the specific density may range from 0.72 to 0.78. the lower heating value may also range from 41.87 to 44.19 mj/kg. boiling temperatures, reid vapor pressure, octane number, and viscosity also vary significantly. the unece r101 procedure simply does not work with real petrol and diesel fuels available at commercial pumps, because of their significantly variable properties. as an additional remark, the concentrations of hc, co, and co2 are difficult to measure with the same accuracy as the mass of fuel consumed, the density of the fuel and the lower heating value of the fuel. it is no surprise that the actual co2 emission and fuel economy figures may differ largely from the certified values. the analysis of the 2014 uk government data for 2,358 types of diesel and 2,103 petrol vehicles of parker [14] showed that same advances in technology innovation, vol. 4, no. 2, 2019, pp. 59-72 66 volume of pump fuels used during the certification test by the oem, unfortunately, produced very different co2 emissions. very likely, these pump fuels also had very different energy contents in addition to very different carbon contents. the co2 emission per liter of diesel fuel was shown in parker [14] to range from a maximum of 3,049 g for a minimum of 2,125 g, with an average of 2,625 g. the co2 emission per liter of petrol fuel was shown in parker [14] to vary even more, from a maximum of 3,735 g to a minimum of 1,767 g with an average of 2,327 g. the analysis in parker [14] was based on the uk government data (uk vehicle certification agency [25]). because of fuel price (fuel consumption) and taxation (co2 emission) both contribute to the vehicle cost, co2 and fuel economy data were supposed to be consistent. apart from parker [14], the scientific literature does not report any technical discussion of the european certification test regarding the reliability of the co2 and the fuel economy figures declared by the oems. this lack of a healthy, technical discussion is what then generated the vw emissions scandal. fig. 6 relative co2 per litre of theoretically same diesel fuel as accepted by the uk govt. data august 2017 as retrieved from uk vehicle certification agency [25] figs. 6 7 present the latest data collected by the uk government for 2,560 diesels and 2,379 petrol vehicles satisfying euro 6 emission standards that were available for sale in the uk during august 2017. these are the data declared by the manufacturers and not the result of independent tests. fig. 6 shows the carbon dioxide emissions in g co2 per litre of diesel fuel consumed. the average is 2,617 g of co2 per litre. however, apart from outliers which are very likely errors in the transcription, there is a large spreading between different cars of several percentage points that is illogical, as a litre of diesel should produce the same co2 emission, if one assumes that all pump fuels are equivalent for emissions purposes. fig. 7 shows the carbon dioxide emission in g co2 per litre of petrol fuel consumed. the average is 2,310 g of co2 per litre. however, apart from outliers which are very likely errors in the transcription, there is also here a noticeable spreading between different cars of several percentage points that is illogical, as the same litre of petrol should produce the same co2 emission. it stands to reason that one standard liter of diesel produces the same amount of co2, and a similar argument can be made for a standard liter of gasoline, only when the combustion of the fuel and the after treatment of the exhaust gases is complete, converting all the hydrocarbons to co2 and h2o. as evidenced by exhaust measurements, this complete conversion is indeed the case for modern vehicles. in carefully designed experiments where the fuel economy and the carbon dioxide emissions are both measured with accuracy, and the same fuel is used, the spreading as found in the uk data is unlikely to occur. in addition to verifying that vehicles pass newly defined emission limits over a suite of more realistic and complicated schedules, there is also the need to make the measurements more rigorous and to perform these measurements in independent laboratories. advances in technology innovation, vol. 4, no. 2, 2019, pp. 59-72 67 fig. 7 relative co2 per litre of theoretically same petrol fuel as accepted by the uk govt. data august 2017 as retrieved from uk vehicle certification agency [25] 3. flaws of current real-world-driving tests the controversy over dynamometer vs. real-world driving testing, and the criticisms of emission standards, has increased after the vw emission scandal, costlow [26]. the discussion generated various solutions aimed at closing the gaps, such as the use of portable emissions measurement systems (pems) for real-world mobile emissions testing, birch [27], or birch [28]. on the road measurements with pems are becoming more and more popular. the eu has been the first to adopt rde tests to measure the pollutants emitted on the road, association des constructeurs européens d’automobiles [29], with pems. these rde tests do not replace the chassis dyno full laboratory test, but they complement the standard test. it is very questionable to claim that compliance with the standard flawed procedure, plus compliance of the rde tests, ensure that cars will now deliver low emissions for on-road conditions. every rde test is different from another. there are major factors (e.g., traffic, weather, stops) that are beyond control and the measurements are inaccurate. apart from road and weather conditions, there is also no “standard” pems equipment defined, and equipment manufactured by different suppliers may deliver different results. pems measurements cannot provide the same accuracy as a full laboratory experiments on a chassis dynamometer. rde step 1 (with a nox conformity factor of 2.1) has applied since 1 september 2017 to new car types. it will apply to all types as of september 2019. rde step 2 (with a nox conformity factor of 1.0 plus an error margin of 0.5) will apply in january 2020 for new types and then from january 2021 for all types. as association des constructeurs européens d’automobiles [29] admits, conformity factors are defined to account for errors, as the pems tests are not as accurate as a full laboratory system, so they will not measure to the same level of repeatable accuracy. while association des constructeurs européens d’automobiles [29] states that the oems must set their design objectives well below the legal limit to be certain about complying, it is still unclear which legal limit they are referring to. so far, there is no legislation prescribing threshold pollutant emissions ot her than the full laboratory cold start nedc. as an example of rde tests on australian roads, abmarc [30] presents a summary of the emissions and fuel consumption test results from seventeen different passenger and light-duty vehicles measured by pems. the number of vehicles powered by petrol, diesel, and lpg were 10, 6 and 1 respectively. each vehicle was tested twice, with one cold and one warm. the testing was conducted on melbourne roads between may 2016 and march 2017, i.e. winter to summer, under strong variable weather conditions. the vehicles tested were 2014-year models or newer taken from the general service fleet and they have driven at least 2,000 km but no more than 85,000 km. the tests were conducted by driving each vehicle around a route consisting of urban, extra-urban and freeway driving, with approximately one-third of the test being driven in each segment. each test is driven in normal traffic conditions. the covered distance is more than seven times the distance covered during the nedc. advances in technology innovation, vol. 4, no. 2, 2019, pp. 59-72 68 the actual velocity schedule is not provided in the report abmarc [30]. it would have been interesting to compare the nedc velocity schedule with the rde velocity schedule, and possibly compute the energy requirements and the stop timings of both. similarly, relevant, especially for nox, could have been a comparison of the temperature traces of air, metal, and media, also not provided in abmarc [30]. on average, the real-world fuel consumption across all vehicles tested was 25% higher than the official fuel results based on the cold start nedc tests. this is not a surprise. the fuel economy always changes to the driving schedule. the average fuel consumption of vehicles improved when tested with a warm engine. the real-world average fuel consumption for warm engines was 22.9% higher than the official results on the nedc tests. the few percentages penalties due to cold start are also not a surprise. one interesting result was that fuel economy penalties were larger for diesel than petrol vehicles. on average, diesel vehicles have the highest variation between real-world fuel consumption and the laboratory results, with 30.4% higher consumption during the cold test. abmarc [30] suggests that the real-world fuel consumers of diesel vehicles is harder to predict based on the nedc laboratory results. we add those individual drivers may provide different fuel economies covering the same path. regarding the exceedance of specific laboratory pollutant limits during real-world driving, all the nox limit failures were by diesel vehicles. only 1 of the 6 diesel vehicles satisfied the limit when tested on the road. diesel vehicles have been otherwise usually compliant for pm emissions with 1 exception of the cold test only. pm emissions were comfortably met for the cold test, but not during the hot test. two of the 10 petrol vehicles exceeded the co limits both hot and cold. in the summary table of abmarc [30], which only showed data for cold conditions, the 3 diesel euro 4 vehicles exceeded the nedc nox and hc+nox limits, while they complied with the co limit. one of the three failed the pm limit. the 2 petrol (gasoline) euro 4 vehicles passed the nox, co and nhmc tests. the only one euro 5 diesel vehicle failed the nox and hc+nox tests, but passed the co and pm. of the 7 euro 5 petrol vehicles, 2 failed the co test, and 1 the nhmc test while passing all the other tests. finally, of the 2 diesel euro 6 vehicles, 1 failed the nox and hc+nox limits and the only one euro 6 petrol vehicle passed all the tests. the additional vehicle tested was an lpg fueled euro 4 compliantly and passed all the tests. it is not a surprise that the diesel struggled with the nox test. the nox exceedance may be the result of higher temperatures during real-world driving. higher ambient air temperatures, as well as higher temperatures of metal and media in the longer rde, may contribute to the higher nox. these tests only demonstrate that different driving schedules may yield different fuel economies and that the lean burning diesel engines struggle with nox emissions. there is a significant in-cylinder production of nox and an inadequate reduction of it by the after treatment relative to the certification test emission requirements. the co from 2 petrol vehicles and the pm of 1 diesel were probably the result of lack of service. 4. proposed suite-of-cycles laboratory tests fig. 8 and 9 are examples of results from this independent laboratory testing of a diesel 2014 chevrolet cruze diesel, and of a gasoline mazda 3 with ieloop. these data are from the downloadable dynamometer database and were generated at the advanced powertrain research facility (aprf) at argonne national laboratory under the funding and guidance of the united states department of energy (doe), argonne national laboratory [31]. the figures summarize the fuel economy results simplistic and anachronistic driving cycles such as the nedc, but also a single more complex cycle, such as the wltc, must be replaced with a suite of more difficult cycles. an example of the use of a limited suite of cycles, but only for fue l advances in technology innovation, vol. 4, no. 2, 2019, pp. 59-72 69 economy tests, is from argonne national laboratory [31], which offers in their the downloadable dynamometer database (d3) publicly available testing data regarding advanced technology vehicles including a few conventional vehicles. fig. 8 fuel economy of a 2014 chevrolet cruze diesel. test fuel is type 2007 cert diesel hf0582, fuel density 0.851 kg/litre, fuel net lhv 42.796 mj/kg fig. 9 fuel economy of a 2014 mazda 3 with ieloop. fuel type epa tier ii eee hf0437, fuel density 0.742 kg/litre, fuel net lhv 42.68 mj/kg a solution to the flaws of the current procedures is to perform the tests independently from the manufacturers, by using the same certification fuel of the same properties, starting from the c/h ratio and the lower heating value, and to measure the fuel consumption and the emissions with accuracy. 5. discussion and conclusions even without using defeat devices, the oems have found their way to achieve fuel consumption and emissions results during the cold start nedc that drastically differ from the real-world driving emissions and fuel economy that the customers will then experience in their every-day use of the vehicle, abmarc [30]. the new emission rules will have to address this major pitfall, to make mass mobility more fuel efficient and less polluting. to prevent the sale of vehicles whose real-world driving emissions are far from the current certification levels, it is not acceptable to have uncertain measurements by the oem on a single certification cycle, followed by independent, poorly defined, inaccurate and irreproducible rde tests on the road. the new rules will have to address other major failures of the regulations, such as the one concerning plug-in hybrid vehicles. instead of the current protocol of testing plug-in hybrids over one drive cycle discharging the battery, the revised rules should add a test without the battery connected. the added test would compensate for the unrealistic zero fuel consumption and emissions during the first cycle. similarly, the new rules will have to prevent the recharging of the traction battery of the hybrid electric vehicles during the final sharp braking event 120 km/h (nedc) or 130 km/h (wltc) to rest that is nowhere to be experienced during real-world driving in europe but helps hybrid electric vehicles. also relevant is a different weight of city and highway driving, as hybrids have advantages in stop-and-go traffic, but shortcomings in highway driving, and the length of the travel. advances in technology innovation, vol. 4, no. 2, 2019, pp. 59-72 70 properly designed new real-world driving emissions tests will completely change the powertrain mix from the present regulations. in addition to lower cost ices that can run on gasoline, diesel and alternative fuels, such as ethanol, methanol, lpg, cng, lng, dme and ammonia, the new emission rules should promote a hybridization of powertrains. a major hurdle to better cars powered by ices is the proposed ban of the ice in favor of electric cars in europe, oltermann [32], castle [33], freedland [34], adepetu and keshav [35]. as the electric vehicle mass mobility is still far from being demonstrated to be superior under the economic and environmental criteria of ice based mobility, this ban will be hopefully revised. otherwise, the car makers will not invest in improving the ice technology. as discussed in boretti [2], it is not feasible in a growing world to completely phase out the ice in favor of electric cars recharged by renewable energy sources. hence, better emission standards for ices powered vehicles are needed more than ever. the use of pems on vehicles tested on the road is not a reliable solution to the present unrealistic certification tests. consistent test conditions cannot be enforced, and the results will be inaccurate. on the road, real-world driving tests cannot produce fair assessments. the only positive aspect of these tests is that they are performed by independent bodies and not the oem, and they use longer and more realistic driving conditions than the nedc. the use of pems on vehicles tested on the road is not a reliable solution to the present unrealistic certification tests. consistent test conditions cannot be enforced, and the results will be inaccurate. on the road, real-world driving tests cannot produce fair assessments. the only positive aspect of these tests is that they are performed by independent bodies and not the oem, and they use longer and more realistic driving conditions than the nedc. there is certainly a need for standardized and repeatable procedures, but still using rigorous chassis dynamometer tests, that are based on more complex and realistic driving conditions. hence, what is needed is a chassis dynamometer full laboratory procedure that comprises multiple driving schedules, this time performed by independent bodies. this novel testing procedure will reduce the ability of the manufacturers to implement “defeat devices”, but also limit the use of emissions reduction technologies that are only effective on a single, test cycle such as the nedc or the wltc. it is proposed that the pollutant thresholds should be dynamically adjusted to permit the best tradeoff. for example, the larger nox emissions of the diesel engine (relative to the gasoline engine) should be compensated by its much larger fuel conversion efficiencies over most of the load and speed map. apart from combined cycle gas turbine (ccgt) power plants and large stationary diesel power generators, most fossil fuel powered electrical generators in the world operate at a fraction of the thermal efficiency of diesel engines for transport. this means that the diesel engine is still needed, and more stringent emission standards can be applied to diesel generators to compensate for possible laxer regulations in transport diesel. we suggest the definition of new emission rules be used as an opportunity to revise current policies that fail to produce balanced societal outcomes for the environment and a sustainable economy. new emission rules should be universally applicable to all alternative vehicles ranging from electric cars to vehicles powered by hydrogen carrier fuels. we have discussed the pitfalls of the current european union emission rules, that will not produce any improvement of fuel economy and emissions during real-world driving vs. the pre-diesel gate condition. this is because of the flaws in the prescribed laboratory and on-the-road tests that have been evidenced in the paper. present laboratory tests are based on a given, simplistic velocity schedule. however, they may be accurate and reproducible, and thus reliable, if performed by independent bodies. on the other hand, real-world driving tests are not very well defined and not repeatable, while emissions and fuel economy measure with poor accuracy. to ensure that emission compliant vehicles will meet expectations in terms of emissions and fuel economy when driven in the real world, it is necessary to introduce accurate, repeatable, representative, complex, laboratory tests independently performed by the certification bodies. advances in technology innovation, vol. 4, no. 2, 2019, pp. 59-72 71 as a further measure, to ensure the optimal mix between different engine technologies (e.g., stoichiometric, lean-burning homogeneous or stratified, premixed or diffusion combustion, spark or compression ignition) and different fuels, such as diesel, gasoline, lpg, cng, lng, ethanol, methanol, dme, ammonia, there is a need to dynamically adjust the pollutant emissions thresholds to compensate fuel economy and carbon dioxide emission. it is to be noted that although lean-burn, compression ignition, diffusion combustion engines may fail to meet extremely stringent emissions of nitrogen oxides, they provide the opportunity to use combustion fuels more efficiently. they also can drastically clean-up the air of extremely polluted areas where some pollutants in the air such as particulate matter, are an order of magnitude more concentrated than at the tail-pipe after the particle trap. this is not only the case for many cities in china, but also the case of some cities in europe, especially during the winter season. hence, policy makers should keep in mind all the issues in designing certification tests if targeting the best outcome for real-world consumers. acknowledgement the authors received no funding and declare no conflict of interests. nomenclature ccgt combined cycle gas turbine epa environmental protection agency (us) icct international council on clean transportation ice internal combustion engines lhv lower heating value nedc new european driving cycle oem original equipment manufacturers pems portable emissions measurement systems rde real driving emissions ftp federal test procedure (us) udds urban dynamometer driving schedule sftp supplemental federal test procedures (us) wltc worldwide harmonised light vehicles test cycle wltp worldwide harmonised light vehicles test procedure vw volkswagen group conflicts of interest the authors declare no conflict of interest. references [1] a. boretti, “the future of the internal combustion engine after diesel-gate,” sae p. 2017-28-1933, 2017. [2] a. boretti, “life cycle analysis comparison of electric and internal combustion engine-based mobility,” sae p. 2018-28-0037, 2018. [3] australian government department of infrastructure, regional development and cities, “australian design rules (adrs) for safety and the environment,” infrastructure.gov.au, 2018. [4] a. boretti and h. c. watson, “the lean burn direct injection jet ignition gas engine,” international journal of hydrogen energy, vol. 34, no. 18, pp. 7835-7841, 2009. [5] a. boretti, “advantages of converting diesel engines to run as dual fuel ethanol-diesel,” applied thermal engineering, vol. 47, pp. 1-9, 2012. [6] a. a. boretti and h. c. watson, “development of a direct injection high efficiency liquid phase lpg spark ignition engine,” sae international journal of engines, vol. 2, no. 1, pp. 1639-1649, 2009. [7] a. boretti, “towards 40% efficiency with bmep exceeding 30 bar in directly injected, turbocharged, spark ignition ethanol engines,” energy conversion and management, vol. 57, pp. 154-166, 2012. advances in technology innovation, vol. 4, no. 2, 2019, pp. 59-72 72 [8] a. boretti, r. paudel and a. tempia, “experimental and computational analysis of the combustion evolution in direct-injection spark-controlled jet ignition engines fuelled with gaseous fuels,” proceedings of the institution of mechanical engineers, part d: journal of automobile engineering, vol. 224, no. 9, pp. 1241-1261, 2010. [9] b. dhinesh, r. niruban bharathi, j. isaac joshua ramesh lalvani, m. parthasarathy and k. annamalai, “an experimental analysis on the influence of fuel borne additives on the single cylinder diesel engine powered by cymbopogon flexuosus biofuel,” journal of the energy institute, vol. 90, no. 4, pp. 634-645, 2017. [10] m. annamalai, b. dhinesh, k. nanthagopal, p. sivarama krishnan, j. isaac joshua ramesh lalvani, m. parthasarathy and k. annamalai, “an assessment on performance, combustion and emission behavior of a diesel engine powered by ceria nanoparticle blended emulsified biofuel,” energy conversion and management, vol. 123, pp. 372-380, 2016. [11] l. subramani, m. parthasarathy, d. balasubramanian, and k. ramalingam, “novel garcinia gummi-gutta methyl ester (ggme) as a potential alternative feedstock for existing unmodified di diesel engine,” renewable energy, vol. 125, pp. 568-577, 2018. [12] b. dhinesh and m. annamalai, “a study on performance, combustion and emission behaviour of diesel engine powered by novel nano nerium oleander biofuel,” journal of cleaner production, vol. 196, pp. 74-83, 2018. [13] b. dhinesh, y. m. a. raj, c. kalaiselvan and r. krishna moorthy, “a numerical and experimental assessment of a coated diesel engine powered by high-performance nano biofuel,” energy conversion and management, vol. 171, pp. 815-824, 2018. [14] a. parker, “influence of test fuel properties and composition on unece r101 co2 and fuel economy valuation,” nonlinear engineering, vol. 4, no. 4, pp. 261-270, 2015. [15] p. mock, j. german, a. bandivadekar, and i. riemersma, “discrepancies between type-approval and “real-world” fuel-consumption and co,” the international council on clean transportation, issue 13, paper no. 2012-2, 2012. [16] s.e. plotkin, “examining fuel economy and carbon standards for light vehicles,” energy policy, vol. 37, no. 10, pp. 3843-3853, 2009. [17] p. kågeson, “cycle-beating and the eu test cycle for cars,” european federation for transport and environment (t&e), brussels, 1998. [18] p. steen, “major car makers respond to our fuel claims campaign,” conversation.which.co.uk, 2015. [19] n. shaw, “are misleading fuel consumption claims costing you money?” conversation.which.co.uk, 2015. [20] e. peker, “auto makers could face big fines under new european rules,” wsj.com, 2017. [21] european federation for transport and environment, “who adds pressure for stricter euro-5 standards,” news from the european federation for transport and environment, vol. 206, no. 146, 2006. [22] us environmental protection agency, “volkswagen clean air act civil settlement,” us epa, 2017. [23] v. franco, f.p. sánchez, j. german, and p. mock, “real-world exhaust emissions from modern diesel cars: a meta-analysis of pems emissions data from eu (euro 6) and us (tier 2 bin 5/ulev ii) diesel passenger cars: part 1: aggregated results,” the international council of clean transportation, 2014. [24] g.j. thompson, d.k. carder, m.c. besch, a. thiruvengadam, and h.k. kappanna, “in-use emissions testing of light-duty diesel vehicles in the united states, final report prepared for the international council on clean transportation (icct),” eenews.net, 2014. [25] uk vehicle certification agency, “car fuel data, co2 and vehicle tax tools,” dft.gov.uk, 2017. [26] t. costlow, “vw emissions scandal will impact future engine controls, testing,” sae.org, article 14384, 2015. [27] s. birch, “millbrook introduces pems for real-world mobile emissions testing,” sae.org, article 12610, 2013. [28] s. birch, “testing for real,” sae.org, article 7094, 2009. [29] association des constructeurs européens d’automobiles, “the real driving emissions (rde) test,” caremissionstestingfacts.eu, 2018. [30] abmarc, “real-world driving. fuel efficiency & emissions testing,” aaa.asn.au, 2017. [31] argonne national laboratory, “downloadable dynamometer database,” anl.gov, 2017. [32] p. oltermann, “german court delays ruling on city bans for heavily polluting diesel cars,” theguardian.com, 2018. [33] s. castle, “britain to ban new diesel and gas cars by 2040,” nytimes.com, 2017. [34] j. freedland, “if this is the end of the car as we know it, we have the eu to thank,” theguardian.com, 2017. [35] a. adepetu and s. keshav, “the relative importance of price and driving range on electric vehicle adoption: los angeles case study,” transportation, vol. 44, no. 2, pp. 353-373, 2017. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation , vol. 2, no. 3, 2016, pp. 85 88 85 the innovative bike conceptual design by using modified functional element design method nien-te liu 1,* , chang-tzuoh wu 2 1 department of product design, shu-te university, kaohsiung, taiwan. 2 department of industrial design, kaohsiung normal university, kaohsiung, taiwan. received 29 january 2016; received in revised form 27 april 2016; accept ed 02 may 2016 abstract the purpose of the study is to propose a new design process by modifying functional e le ment design approach which can commence a large amount of innovative concepts within a short period of time. firstly, the origina l creative functional ele ments design method is analyzed and the drawbacks are discussed. then, the modified is proposed and is divided into 6 steps. the creative functional ele ment representations, generalization, specialization, and particularizatio n are used in this method. every step is described clearly, and users could design by follo wing the process easily. in this paper, a clear and accurate design process is proposed based on the creative functional ele ment design method. by fo llo wing this method, a lot of innovative bicycles will be created quickly. keywor ds : conceptual design, functional element design method, generalizat ion, specialization, particularization 1. introduction design in b lack bo x is the most co mmonly used in practical design. but, it is not popular for study and learning [1]. in this study, we analy ze functional ele ments and try to solve the black box proble m of creating ne w concepts for innovative bicycles. every innovative bicycle has one or more creat ive functional ele ments . if we choose several suitable creative function elements in our design, then we can come up with a new and creative bicycle. liu and wu have proposed the representations for bicycle characteristics, but the process for bicycle design is not mentioned in the study [3]. this paper proposes representations for bicycles. by using symbols representations, bicycles can be represented simple and quickly for innovation design. ke considers bicycle to be a module prod uct. t h us , in t e rms o f d iffe re n t d es ig n a n d combination in spare parts and components, there are four factors that can directly impact on the difference of the class level and price of a bicycle: function, material, appearance and manufacturing quality. in his study, he also mentions that in defin ing the position of a new product, three indexes as new materia l, ne w function and new purpose can serve as the course of research and development [4]. there fore, in the creative design of b icycles, function and material are very important factors. in the s tudy of ma, it has proposed that module design is a concept of introducing part/component module into the design process a n d to s imp li fy a n d o rg an ize v a rio u s components/parts by application of systematic method. in so doing, it can raise the functional performance and adaptability of bicycles so as to satisfy the application require ments [5]. in the study, it disintegrates a bicycle into various e le me n t c o mp o n e n t mo d u le s wit h e a c h c o mpon ent rep resent ing on e fu nct ion . accordingly, a component can also be regarded as one functional element. in hung’s study, he divides a bicycle into five co mponents as fra me structure, front, rear suspension mechanisms, steering handler, and seat for engaging conceptual design. in the final stage, he conducts a concrete integral design of integrating various systems by establishing a n in t e g rat io n p ro cess th ro ugh a pp ly in g morphologica l matrix [6]. in terms of shortfall, since his design divides a bicycle into five components, wh ich constrains the definition of bicycles into a narrow one and easy to limits the * corresponding aut hor, email: ntliu@stu.edu.tw advances in technology innovation , vol. 2, no. 3, 2016, pp. 85 88 86 copyright © taeti design result, rendering an unfavorable impact on the creative design. in his study, he has proposed that the future development of bicyc les can be divided into five categories: 1) use of new materia ls, 2) addition of new functions, 3) new rid ing ways, 4) introduction of new elec tronic products, 5) dedicated developed products depending on respective market require ments. his research has proposed that in the future creative design of bicycles, addition of new functions is an indispensable lin k. thus, it is very important to more focus on choice and integration of new functional ele ments in the design process. ji has categorized the spare parts/ components of a bicycle according to function into the fo llowing six majo r systems: 1) transmission system, 2) steering system, 3) wheel system, 4) bra king system, 5) veh icle fra me structure system, 6) gear system, [7]. chang categories on bicycle composition, he divides the functions of a bicycle into seven ma jor systems: 1) trans mission system, 2) steering system, 3) wheel system, 4) bra king system, 5) structure system, 6) fitting system, 7) accessory system [2]. at present, the traditional methods used for designing bicycles lack analysis and induction, it much limits the design development with crea tive concept. this study would continue the former research about the design process for innovative bikes [2]. the purpose of this paper is to help designers to produce their creat ive design concepts by applying massive support of fun ctional e le ments in a short time and modified the proposed design process. 2. functional elements representation this research collects 242 award works of bicycle products between 1996 and 2006 fro m the bicycle design competit ions. the bicycle design co mpetit ion is sponsored by the department of industrial technology, min istry of economic affa irs, r. o. c. and managed and produced by the cycling & hea lth tech industry r&d center. the first such competition was the "1996 ta iwan creative bicycle design co mpetition." the study sort out these award works into a co mparison table on functional e le ments according to year sequence order and serial number. the award works were preliminarily categorized into “standard” and “specific” categories. “standard category” means a norma l bicycle category and is simila r to the market -sale rid ing bikes in structure which signifies a bike with simple change of appearance and style. “specific category” is comparatively different to the riding way, structure and functions of market-sale bikes, such as contractible, different way of storage, or addition/subtraction of certain functions. cite bike for e xa mp le, we d o a functional nature analysis. before such functional ele ment analysis, we define various ele ments as shown in the symbol figure. each functional ele ment has its own represent symbol. the batch number for addition of function p can be known in the categories of standard and specific types. addition of function p shows the addition and subtraction change for many bike designs. therefore, we give these serial nu mbers as p01, p02, p03 … in our analysis process [2]. for transform a bicycle into a symbolized representation, a process is suggested here: 1. take a picture of a bicycle; 2. find functional ele ments of the bicycle; 3. mark sy mbols on every functional e le ments of the bicycle; 4. check and note the connecting relationship b etween every two functional elements. by the process, every bicycle has its own symbolized representations by the functional elements. take a norma l bicyc le as an exa mple, we ma rk various functional ele ment symbols on the bike, and then connect and define the relationship between various functional elements. under normal circumstance, when various functional ele ments are inseparable, we use solid line to show the connection between various functions, which indicates their d irect connection; when it is indirectly driven by chain-like object, we then use a solid line with an arro w. the arrow points at functional ele ments fro m hu man -driven end. accordingly, we can have a functional e le ment representation as shown in fig. 1 [2].where means seat, means control, means single wheel, and means driven by human power. fig. 1 functional element representation of a normal bicycle advances in technology innovation, vol. 2, no. 3, 2016, pp. 85 88 87 copyright © taeti existing bicycles symbolized bicycles atlas of generalized bicycles atlas of feasible specialized bicycles creative functional element representations combinations generalized bicycles generalization atlas of feasible bicycle innovative bicycles design requirements and constraints specialization design characteristics particularization existing bicycles 3. original design process the orig inal innovational design process for bicycle is proposed in liu’s research [2], but the design process is not clear enough to operate step by step. the design process is proposed and shown as fig. 2. fig. 2 the bicycle design process using fun ctional element representation in the first step, the designers’ descriptions for design theme are necessary. the theme is decided by personal preferences or design co nstraints. the description is more c learly, the selecting of functional ele ments is more easy and fast. in the second step, functional ele ments are chosen in accordance with the contents d escribed. the constraints of combination of a bicycle need to be considered. so, some elements a re necessary. in the third step, functional ele ment symbols a re a rra n g e d fo r n e w d e s ig n c o n c e pts . according to designers ’ need for creative functional ele ments, relative positions are arranged. here designers need understand the ra t ion a l ru le s o f b ic yc le s fo r a rra ng in g functional elements. finally, according bicycle’s creative functional ele ment symbol table , the representations obtained fro m previous step are transformed into new design concept of bicycles. 4. modified design process for the design process mentioned above, there are so me p laces interpret unclea r thus designer will be hard to work following the process step by step. therefore, the modified design process is proposed and shown as fig. 3. fig. 3 the modified design process description of design theme choice of functional elements arrangement of space position of functional elements realization design using creative functional elements advances in technology innovation , vol. 2, no. 3, 2016, pp. 85 88 88 copyright © taeti in step 1, a designer needs choose an e xisting bicycle for developing. by analyzing the e xis ting bicycles, design characteristics are collected. by c reative functional ele ments representations, bicycles are symbolized. in s t ep 2, s y mb o li ze d b ic y c le s a re generalized. every symbo l is mean ingful for symbolized bicycles. by generalization, every symbol would be changed to a non -mean ingful symbol. in st ep 3, fro m g ene ra lized b icyc les a nd combination, different topology could be used for deigning. thus, atlas of generalized bicyc les is obtained. in s t ep 4, by sp e c ia liza t io n , e ve ry n on mean ingful symbol in atlas of generalized b icycles is assigned a creative functional ele ment one by one. during assigning process, des ig n re qu ire ments and const ra in ts a re considered at the same time. atlas of feasible specialization bicycles are obtained in this step. in step 5, feasible specializat ion bicycles are particularized. the form of a ll bicycles can be observed easily. in step 6, by deleting the existing bicycle in atlas of feasible bicycle in step 5, innovative bicycles are obtained. 5. conclusions the approach proposed by this research is to modified functional e le ments design method for bicycle design. the advantage of this approach is the possibility of creating more concepts easily, but the original process, designer is hard to follow step by step for designing. in this paper, a modified process is proposed and divided into 6 steps clearly. the c reative functional e le ment representations, generalization, specialization, and particularizat ion are used in this method. when designing innovative bicycles, designers just need finish the process and innovative b icycles will be obtained. acknowledgement the support of the minister of education (taiwan), under grant most 104-2221-e 366-009 is gratefully acknowledged. references [1] n. f. m. roo zenburg, j. ee kels, product design: fundamentals and methods, wiley, 1995. [2] n. t. liu, c. t. wu, and y. c. lin, “the study on the innovation bicycle design using functional ele ment characteristics,” applied mechanics and material, vol. 764-765, pp. 383-387, 2015. [3] n. t. liu and c. t. wu, “ on the symbolized representation of innovative bicycle with functional ele ments characteristics for creative design,” transaction of the canadian society for mechanical engineering, vol. 37, no. 3, pp. 949-957, 2013. [4] c. m . ke , “ vie wing bicycle design by product status,” bicycle industry, vol. 33, pp. 38-43, 2001. [5] c. h. ma, t. h. lin, and y. c. huang, “theory and application on module design model – case study on bicycle,” journal of industrial design, vol. 109, pp. 127-134, 2003. [6] t. d. hung, “ creative design on two-whee l bike s ystem,” institute of mechanical and mechatronics engineering, nat ional sun yat-sen university, 2005. [7] y. h. ji, “ creative design on bike’s spe ed -c han g ing dev ice ,” inst itut e o f mechanical and mechatronics engineering, national sun yat-sen university, 2002.  advances in technology innovation , vol. 2, no. 3, 2016, pp. 68 72 68 spatial and spectral nonparametric linear feature extraction method for hyperspectral image classification jinn-min yang * , shih-hsuan wei department of mathematics education, national taichung university of education, taichung, taiwan . received 02 may 2016; received in revised form 21 june 2016; accept ed 21 june 2016 abstract feature extraction (fe) or dimensionality reduction (dr) plays quite an important role in the field of pattern recognition. feature extraction aims to reduce the d imens iona lity of the high-d imens iona l dataset to enhanc e the classification accuracy and foster the classification speed, particularly when the training sample size is small, namely the small sample size (sss) problem. remotely sensed hyperspectral images (hsis) are often with hundreds of measured features (bands) which potentially provides more accurate and detailed information for classification, but it generally needs more samples to estimate parameters to achieve a satisfactory result. the cost of collect ing ground -t ruth o f re mote ly sensed hyperspectral scene can be considerably difficult and expensive. therefore, fe techniques have been an important part for hyperspectral image classification. unlike lots of feature extraction methods are based only on the spectral (band) information of the training samples, some feature extraction methods integrating both spatial and spectral information of training samples show more effective results in recent years. spatial contexture information has been proven to be useful to improve the hsi data representation and to increase classification accuracy. in this paper, we propose a spatial and spectral nonparametric linear feature extraction method for hyperspectral image classification. the spatial and spectral information is extracted for each training sample and used to design the within-class and between-class scatter matrices for constructing the feature extraction model. the experimental results on one benchmark hyperspectral image demonstrate that the proposed method obtains stable and satisfactory results than some existing spectral-based feature extraction. keywords : feature extraction, hyperspectral image, dimensionality reduction, classification, small sample size problem 1. introduction re motely sensed hyperspectral images (hsis) are often with hundreds of measured features (spectral bands) potentially provides more accurate and detailed informat ion for classification and widely used in environmental mapping, geologica l research, and minera l identification in recent years. the cost of co llecting ground-truth of re motely sensed hyperspectral image can be considerably d if f ic u lt a n d e xp e n s iv e . t h e r e fo re , f e techniques have been an important part for hyperspectral image classification. in general feature e xtraction (fe) or dimensionality reduction (dr) plays quite an important role in the field of pattern recognition. linear discriminant analysis (lda) [1] is one of the most well-known linear feature e xtraction methods and has been successfully applied to many fie lds. the purpose of lda is to find a linear t ransformat ion matrix that can be used to project data fro m a h igh-dimensional space into a lo w-dimensional subspace to mitigate the so-called curse of dimensionality [2], [3] or the hughes phenomenon [4], [5]. the hughes phenomenon describes that the ratio of the number of train ing samples and the number of features must be ma intained at or above some minimu m va lue to achieve statistical confidence [5]. otherwise, the classification accuracy will decline with an increase in the dimensionality of data to some e xtent. however, it is not necessary to have sufficient training samp les to keep the ratio in a high-dimensional classification task. the re fo re , by featu re e xt ract ion, the ra t io c an b e re la t iv e ly en la rge d and t he cu rs e o f dimensionality can therefore be improved, this will result in an enhancement of c lassification accuracy. meanwhile, the co mputational time can be reduced as well. * corresponding aut hor, email: jinnminyang@mail.ntcu.edu.t w advances in technology innovation , vol. 2, no. 3, 2016, pp. 68 72 69 copyright © taeti ba s ic a lly , lda h as t h ree inhe rent deficiencies in dealing with classificat ion proble ms. first, lda is only well-suited for normally distributed data [1]. if the distributions are significantly non-normal, the use of lda cannot be e xpected to accurately indicate which features should be e xtracted to preserve co mple x stru ctures needed for classification. second, since the rank of between-class scatter matrix is the number o f c lasses minus one [1], the nu mber of features can be extracted at most rema ins the same. third, the singularity proble m arises when dealing with high-dimensional and small samp le size (sss) data [1, 3, 6-7]. nonparametric linear discriminant analysis such as nonparametric discriminant analysis (nda) [1], nonparametric weighted feature e xtract ion (nwfe) [6] and cosine-based feature extraction (cnfe) [7] provide solutions for circu mventing the previously mentioned problems. the afore mentioned fe methods are spectral-based algorith ms ; in other words, they measure similarity in the spectral space. using only spectral informat ion to classification tasks is insufficient. spatial conte xture info rmation has been proven to be useful to improve the classification of hsi data in recent years [8]-[9]. in the paper, a nonparametric feature e xtraction method, integrating both spectral and spatial information, is proposed. the rest of this paper is organized as fo llows. the proposed method and its e xperiment a re described in section 2. the e xperimental results and discussion are provided in section 3. finally, section 4 gives some conclusions of the paper. 2. method the goal of fe is to find a transformation matrix a which ma xi mi zes th e c lass separab ility in the transformed space, where and denote the within-class and between-class scatter matrices, respectively. that is (1) the ma ximizat ion of (1) is equivalent to solving the generalized eigenvalue decomposition problem where denotes the dimensionality of the transformed space, represent the eigen-pair of , and . thus, the transformation matrix a = [v1 , … , v𝑃 ] can be obtained. the proposed spatial and spectral feature e xtraction method includes two parts, one idea is to incorporate the spatial information into the within -class and scatter matrix design, and the other idea is to incorporate another scatter ma trix to regularize the within -class scatter matrix. 2.1. the nonparametric linear discriminate analysis the within-c lass matrix and between-class scatter matrix o f the nonpara metric linear discriminant feature e xtract ion (nlda) are defined as follows, respectively. (2) (3) where and denote the local mean of tra ining samp le corresponding to the th class and th class, respectively. the local mean of is computed by its -nearest neighbors ( nns) in the sa me c lass or in the different classes as shown in (4). (4) 2.2. the spatial and spectral nonparametric linear discriminate analysis the within-c lass matrix and between-class scatter matrix o f the spatial and spectral n o n p a ra me t ric lin e a r d isc rimin a n t fe atu re extraction (ssnlda) are defined as follows, re spectively.             1 1 i t l s s i i i iw i n i i is x x x xp m m       (5)             1 1 1 inl l t s s i i i i i j jb i j j i s x x x xp m m         (6) where and denote the local mean of tra ining samp le corresponding to the th class and th class, respectively. the local mean of is co mputed by utilizing the advances in technology innovation , vol. 2, no. 3, 2016, pp. 68 72 70 copyright © taeti spectral we ighted local mean and spatial we ighted local mean , as shown in (7). (7) where (8) with (9) and (10) with (11) and denotes the coordinate of training sample , the parameter . another part of is the distance scatter matrix introduce in [8]. based on a window, a training samp le and its pixe l neighbors form a local patch , where the odd number is the width of the neighborhood window. the scatter matrix and (12) whe re s = 𝑧2 − 1 , and . pa ra met er reflects the degree of filtering. regularization is emp loyed to improve the singularity problem in ssnlda. the within-class scatter matrix is replaced by (13) where denotes the diagonal parts of a matrix and . 2.3. dataset the indian pines image, mounted fro m an aircra ft flo wn at 65000-ft a ltitude and operated by the nasa/jet propulsion laboratory, with the size of 145 × 145 pixels has 220 spectral bands measuring appro ximately 20 m across the ground. we have also reduced th e number of bands to 200 by re moving bands covering the region of water absorption: 104-108, 150-163, and 220. there are 16 c lasses in the data set. the total number of samples is 10249, ranging fro m 20 to 2455 in each c lass. in figure 1, the left and right images depict the false color co mposition of three sample bands 50, 27 and 17 and its ground truth of the indian pines dataset, respectively. fig. 1 the left figure depicts indian pines image of band 50, 27 and 17; the right one shows its ground truth. 2.4. experiment design three diffe rent cases, each class with 5 (case i), 10 (case ii), and 20 (case iii) tra ining samp les are investigated to discover the effect on the sizes of training samples in the experiments. the re main ing samp les are e mp loyed as the test samples. the cases i and ii are the so-called il l-p o se d a nd po o rly pos e d c lass ific a t io n problems [7], respectively. they are challenging cases in the fie ld of pattern recognition. in each case, the training and testing datasets are randomly selected. we will repeat each case for 10 times and report the averaged overall accuracy (oa) and standard deviation. two other linear feature e xtraction methods, cnfe and nwfe, a re utilized to compare the classification performance with the proposed ssnlda. the 1-nearest neighbor (1nn) c lassifier is emp loyed. in ssnlda, we adopt a window to form a local patch and the values of ,  and  are set as 0.5. advances in technology innovation , vol. 2, no. 3, 2016, pp. 68 72 71 copyright © taeti table 1 graph representations class # pixels class name 1 46 alfalfa 2 1428 corn-notill 3 830 corn-mintill 4 237 corn 5 483 grass -pasture 6 730 grass -trees 7 28 grass -pasture-mowed 8 478 hay-windrowed 9 20 oats 10 972 soybean-notill 11 2455 soybean-mintill 12 593 soybean-clean 13 205 wheat 14 1265 woods 15 386 buildings-grass-trees -drives 16 93 stone-steel-towers 3. results and discussion table 2 lists the best classification accuracies of the three cases of the indian pines dataset. as we can see from the table, 1nn classifier with ssnlda features can achieve better results than with cnfe and nwfe features. ssnlda provides about 6% improvement as compared with the other two methods. meanwhile, the standard deviations is smaller as well. fig. 2 demonstrates the variations of oas with the reduced dimensions where 5, 10 and 20 training samples are utilized. the proposed ssnlda significantly outperform the other two methods. fig. 3 shows the classification map of the indian pines scene using 1nn classifier for 20 training samples case. the dimensionality of the reduced space is 30. as shown in fig. 3, the 1nn classifier with ssnlda feature can get better results. table 2 classification accuracies (in percent) in indian pines scene case fe oa±std (#features) nwfe 63.03±4.68(28) cnfe 65.10±4.91(15) ssnlda 71.05±3.28(28) nwfe 72.82±2.28(29) cnfe 75.67±1.77(29) ssnlda 81.79±1.81(30) nwfe 80.15±1.76(28) cnfe 81.93±2.17(28) ssnlda 88.08±1.21(30) fig. 2 classification results on the indian pines dataset for the three feature extraction method. (a) nwfe (b) cnfe (c) ssnlda (d) ground-truth fig. 3 classification maps of the indian pines scene using 1nn classifier for 20 training samples case. 4. conclusions in this paper, a spectral and spatial information-based nonparametric feature extraction ssnlda is proposed. from the above results, we find ssnlda can achieve more stable and effective results. in most of cases, 1nn classifier with ssnlda features can obtain better results than other spectral-based fe, cnfe and nlda, particularly when the training sample size is quite small. acknowledgement this study is supported by ministry of science and technology, r.o.c., under the contract number of most 104-2221-e-142005. references [1] k. fukunaga, introduction to statistical pattern recognition, 2nd ed., new yo rk: academic press , 1990. advances in technology innovation , vol. 2, no. 3, 2016, pp. 68 72 72 copyright © taeti [2] r. o. duda, p. e. hart, and d. g. stork, pattern classification, 2nd ed., new yo rk: john wiley & sons , 2001. [3] s. j. raudys and a. k. ja in, “ sma ll sa mple size effects in statistical pattern recognition: recommendations for practitioners ,” ieee transaction on pattern analysis and m achine intelligence, vol. 13 no. 3, pp. 252-264, 1991. [4] d. a. landgrebe, signal theory methods in multispectral re mote sensing, ne w jersey: john wiley and sons, 2003. [5] p. k. varshney and m. k arora, advanced image processing techniques for remotely sensed hyperspectral data, new yo rk: springer, 2004. [6] b. c. kuo and d. a. landgrebe, “nonparametric we ighted feature e xtract ion for classification,” ieee transaction on geoscience and re mote sensing, vol. 42, no. 5, pp. 1096-1105, 2004. [7] j. m. yang, p. t. yu, and b. c. kuo, “a nonparametric feature extraction and its application to nearest neighbor classification for hyperspectral image data,” ieee transactions on geoscience and remote sensing, vol. 48, no. 3, pp. 1279-1293, 2010. [8] y. zhou, j. peng, and c. l. ph ilip chen, “dimension reduction using s patial and spectral regularized local discriminant e mbedding for hyperspectral image classification,” ieee t ransactions on geoscience and re mote sensing, vol. 53, no. 2, pp. 1082-1095, 2015. [9] h. pu, z. chen, b. wang, and g. m . jiang, “a novel spatial–spectral similarity measure for dimensionality reduction and class ification of hyperspectral imagery,” ieee transactions on geoscience and re mote sensing, vol. 52, no. 11, pp. 7008-7022, 2014. [10] hyperspectral re mote sensing scenes, [online]. available: http://www.ehu.eus/ccwintco/index.php?titl e=hyperspectral_re mote_sensing_scenes. [accessed 24 may 2016]  advances in technology innovation, vol. 5, no. 1, 2020, pp. 56-63 efficient rsu selection approaches for load balancing in vehicular ad hoc networks chi-fu huang * , jyun-hao jhang department of computer science and information engineering, national chung-cheng university, chiayi, taiwan received 12 april 2019; received in revised form 23 april 2019; accepted 04 august 2019 doi: https://doi.org/10.46604/aiti.2020.4080 abstract due to advances in wireless communication technologies, wireless transmissions gradually replace traditional wired data transmissions. in recent years, vehicles on the move can also enjoy the convenience of wireless communication technologies by assisting each other in message exchange and form an interconnecting network, namely vehicular ad hoc networks (vanets). in a vanet, each vehicle is capable of communicating with nearby vehicles and accessing information provided by the network. there are two basic communication models in vanets, v2v and v2i. vehicles equipped with wireless transceiver can communicate with other vehicles (v2v) or roadside units (rsus) (v2i). rsus acting as gateways are entry points to the internet for vehicles. naturally, vehicles tend to choose nearby rsus as serving gateways. however, due to uneven density distribution and high mobility nature of vehicles, load imbalance of rsus can happen. in this paper, we study the rsu load-balancing problem and propose two solutions. in the first solution, the whole network is divided into sub-regions based on rsus’ locations. a rsu provides internet access for vehicles in its sub-region and the boundaries between sub-regions change dynamically to adopt to load migration. in the second solution, vehicles choose their serving rsus distributedly by taking their future trajectories and rsus’ loading information into considerations. from simulation results, the proposed methods can improve packet delivery ratio, packet delay, and load balance among rsus. keywords: vanet, rsu, load balance, routing 1. introduction vehicular ad hoc networks (vanets) [1-2] attract a lot of attention in recent years due to lots of potential applications [3-5]. a vanet consists of many highly mobile vehicles. each vehicle is able to communicate with nearby vehicles and share information through the network. vanets can be regards as a particular type of traditional mobile ad-hoc networks (manets). many technologies of vanets inherit from manets, such as multi-hop relays and routing protocols. however, vanets have some different characteristics. the moving paths and speeds are constrained by the pre-existing road map. taking vehicles’ speeds and moving directions into considerations can predict the future positions of vehicles. besides, we usually assume vehicles have no energy constraint. there are two basic communication models in vanets [6], vehicle-to-infrastructure (v2i), and vehicle-to-vehicle (v2v), as shown in fig 1. v2i communication relies on fixed infrastructures, such as roadside units (rsus), to communicate vehicles. rsus connect to each other through a wired network and form a backbone network. rsus also act as gateways, which are entry points to the internet for vehicles. because of huge deployment and maintenance cost, it is nearly impossible to * corresponding author. e-mail address: cfhuang@cs.ccu.edu.tw tel.: +886-5-2720411; fax: +886-5-2720859 advances in technology innovation, vol. 5, no. 1, 2020, pp. 56-63 57 deploy a large amount of rsus such that each vehicle can directly connect to a rsu. to extend the service ranges of rsus, vehicles utilize v2v communication to connect to rsus. in v2v communication, vehicles form a self-organized network without any centralized system. vehicles communicate with each other directly and achieve long-distance communications through multi-hop relays to connect to rsus. with rsus sparsely deployed, a critical issue is how vehicles establish routing paths to delivery their packets to/from rsus [7]. traditional routing protocols of manets are not suitable for vanets due to high mobility of vehicles and consequent control overhead. geographic routing [8] is suitable for vanets due to its low control overhead. the main idea of geographic routing is to choose the vehicle geographically closest to the destination as the next hop to relay packets. naumove et al. [9] proposed the advanced greedy forwarding (agf) algorithm and preferred group broadcasting algorithm (pgb) to improve gpsr performance in vanet. context-aware routing (car) [10] based on agf and pgb could find a route that maximizes the chance of successful delivery. in gytar [11], data packets are forwarded through the selected intersections by a greedy carry-and-forward scheme.. fig. 1 basic communication models in vanets vehicles tend to choose nearby rsus as their entry points to internet. the distance between a vehicle and its servicing gateway affects network performance significantly. the longer distance can result in a higher end-to-end delay and a lower packet delivery ratio. how to deploy rsus becomes a critical issue [12]. intersections are good places to deploy rsus. because when approaching an intersection, vehicles decelerate, which increases the connection time between vehicles and rsus. other criterions to deploy rsus include the area with high vehicular density is preferred, and the distances between any two rsus should be large enough. however, the vehicular density varies from time to time due to the high mobility nature of vehicles. if each vehicle connect to its closet rsu, it is possible that some rsus are overloaded but some are idle. such load imbalance among rsus can prevent the network from fully utilizing rsus’ bandwidths [13-15]. in this paper, we study how to balance the loads of rsus providing internet access for vehicles in an urban environment. there are serval criteria in our design. first, current and future loads of rsus are primary concerns when choosing a serving rsu. second, we should not ask a vehicle to choose a faraway rsu as its serving gateway, even if the rsu is with a very light load. third, the length of routing path between a vehicle and its serving rsu changes as the vehicle moves. therefore, a vehicle should consider its future trajectory when choosing a serving rsu. in this paper, we first propose a rsu-based method, which divides the whole map into several sub-regions. there is only one rsu residences in each sub-region to provide internet access service for all vehicles in its sub-region. the boundaries between sub-regions change dynamically to react to load migrations of rsus. this centralized method is inspired by the cell breathing for wlan [16], which adjusts the beacon transmission range of access points (aps) to balance the loads of aps. to be more scalable, we proposed a distributed vehicle-based method, which allows vehicles choose their serving rsu distributedly. we design a criterion for vehicles to advances in technology innovation, vol. 5, no. 1, 2020, pp. 56-63 58 choose their service rsus based on their future locations, which can prevent vehicles from changing their serving rsus frequently. the main contributions of this paper are listed as follows. 1. we design two methods to balance the loads among rsus for different scenarios. the simulation results shows that both our methods can balance the loads among rsus. 2. with simulation verifications, both of our rsu-based and vehicle-based methods can improve the packet delivery ratio and the end-to-end delay. it is because an overloaded rsu increases the average queuing delay. in the worst case, the rsu may need to drop packets from its queue. 3. the proposed vehicle-based method considers both the driving directions and the future locations of vehicles, which can prevent vehicles from changing their serving rsus frequently. 4. we assist vehicles to choose a suitable rsu with vehicular density, load conditions of rsus, and routing length to improve the overall network performance. the rest of this paper is organized as follows. section 2 presents the system model and define the load-balancing rsu selection problem. in section 3, we describes the proposed methods. section 4 demonstrates the efficiency of the proposed methods through simulations. section 5 discusses the future works and concludes this paper. 2. system model and problem definition in an urban environment, there are n intersections and p road segments. we assume there are total m rsus deployed at a portion of n intersections to provide internet access for vehicles, where n m . to formulate the problem, we translate the city map into a graph g (v, e) as shown in fig. 2, where v is the set of intersections, v’ is the set of intersections with rsu deployed, and e is the set of road segments. for each edge e in e, there is a weight representing its traffic density. except the distance, there are several concerns need to be address when choosing a serving rsu. first, the queuing delay of packets increases as the load of a rsu increases. therefore, the load index ( il ) of rsu i is an important factor to be considered, where li is the ratio of the remaining bandwidth to the total bandwidth of rsu i. second, due to the finite queue length of a rsu, an overloading rsu increases the packet dropping rate. third, a vehicle should take its driving direction and future position into consideration in addition to its current position. otherwise, a vehicle will change its serving rsu frequently. fig. 2 the city map translates into a graph g (v, e) balancing the loads among rsus is important. it can improve not only the packet queuing delay but also the packet delivery ratio for vehicles. we call the problem of how to assign each vehicle to a suitable serving rsu as the load-balancing advances in technology innovation, vol. 5, no. 1, 2020, pp. 56-63 59 rsu selection (lbrs) problem. to evaluate how balance we can achieve, a load-balance index (lbi) of the whole network is defined as {li {li} l 1 m {li} max min bi ,where i . max     (1) the value of lbi is between 0 and 1. a lower lbi implies a more balanced condition. note that, even if a me thod can minimize lbi value but assign vehicles to distancing rsus, it cannot be a good solution. it would increase the end-to-end delay and decrease delivery ratio. therefore, our goal is to assign each vehicle to a serving rsu such that the lbi of the whole network as small as possible without violating the principle of distance criteria. 3. the proposed rsu-based and vehicle-based schemes in this section, two types of schemes, called rsu-based and vehicle-based schemes, are presented to solve the lbrs problem. the rsu-based schemes divides the urban environment into m non-overlapped cells according to the number of rsus. only one rsu residences in a cell to provide internet access service for vehicles in this cell. boundaries between cells define the service ranges of rsus and change dynamically to balance the loads among rsus. in vehicle-based scheme, vehicles choose their serving rsu according to the load index, trajectories, and path lengths to rsus, independently and distributedly. 3.1. rsu-based mechanism in a vanet, due to high mobility of vehicles and the rapidly changing topology, a long delivery path based on multi-hop relays from a vehicle to a rsu may disconnect frequently. to choose a near gateway, we utilize voronoi diagram [16] to partition the network and define the serving ranges of rsus. given a plane and a set a of sits on the plane, the voronoi diagram partitions the plane into numerous cells such that each cell contains exactly one site, denoted as s , in a. the cell contains all points that are closer to s than to any other site in a. in this paper, sites in a are rsus and points are vehicles driving on the roads. we utilize the voronoi diagram to partition the city map into m cells, and each cell contains exactly one rsu. besides, vehicles driving on the roads assist each other in packet forwarding from vehicles to rsus. the distribution of road segments and their lengths are the factors to be concerned, rather than a plane and the euclidian distance. fig. 3 an example of the map-based voronoi diagram resulted from fig. 2 a map-based voronoi diagram considering a city map rather than a plane works as follows. we modify the partition algorithm of the voronoi diagram. instead of the perpendicular bisectors, we find the midpoints of the shortest paths between each pair of rsus. the network is then partitioned by connecting those midpoints. an example is shown in fig. 3. a vehicle in a cell will chooses the rsu in the same cell as its serving gateway, and change to another rsu when driving into another cell. notice that vehicles will not change their serving rsu before their session ends. advances in technology innovation, vol. 5, no. 1, 2020, pp. 56-63 60 next, we further take the load indexes of rsus and vehicular densities of road segments into consideration. the boundaries between rsus adjust dynamically to adapt to loads and densities changes. to distinguish from the above method, we call this method as the weighted map-based voronoi diagram. we make the following modifications. first, we consider the weighted road lengths by the vehicular density of a load segment, instead of the actual lengths. next, when calculating a midpoint, we select the point where the weighted path lengths from the point to two rsus are inversely proportional to their load indexes. rsus calculates the boundaries periodically to maintain the load balance among rsus. the last boundary information is disseminated to vehicles by geocast. once a vehicle receives an updated boundaries information, it selects its serving rsu accordingly. 3.2. vehicle -based mechanism in this type of solution, vehicles utilize their future trajectories and the last load indexes of rsus to choose their serving gateways distributedly. we call this scheme as the trajectory-based scheme since trajectories of vehicles play a major role in the design. compared to the previous methods, there are no clear boundaries defining the serving ranges of rsus. moreover, this solution does not need the density information of road segments, since each vehicle select its rsu individually. fig. 4 a vehicle samples its future trajectory with intersections to calculate a favor value of each rsu in this scheme, vehicles obtain the load index information from rsus by geocast periodically. vehicles are aware of their future trajectories, which can be acquired from the gps navigation system. intersections on a vehicle’s trajectory are selected as representative points, as shown in fig. 4. the path lengths from each representative point i to a nearby rsu j are estimated and denoted as  d i, j . an exponential decay function  h i is used to decay the influences of distancing intersections. then, the vehicle calculates a favor value jf for each rsu j as,     0 1 d l j n ii f h i i, j       (2) where 0 , 1   . a vehicle selects the rsu with the largest favor value as its serving rsu. when a vehicle receives new load indexes of rsus or arrives at a new intersection, the vehicle calculate the favor values of all rsus to decide the best serving rsu. as mention above, vehicle will not change its serving rsu before its current session ends. moreover, we use a predefined threshold to prevent a vehicle from changing its’ serving rsu frequently to avoid the ping-pong effect. 4. simulation results we develope simulations to compare the network performance of the proposed schemes. the simulations adopte a 6 * 6 grid network in a 6 km * 6 km area where the length of each segment is 1 km. there are 6 rsus randomly deployed on different intersections. we generate 300 vehicles and the trajectory of each vehicle is the shortest path between a random advances in technology innovation, vol. 5, no. 1, 2020, pp. 56-63 61 source and destination intersections. communication ranges of vehicles and rsus are the same, which is 350 m. to test heterogeneity and homogeneity of rsus, we set rsus with different bandwidths to the internet, randomly between 5 ~ 20 mbps in case 1 and equally 10 mbps in case 2. fig. 5 packet delivery ratio with different request rates fig. 6 end to end delay with different request rates we first compare the performance of the delivery ratio among voronoi diagram, map-based voronoi diagram, weighted map-based voronoi diagram, and the distributed trajectory-based method. the results are shown in fig. 5. it illustrates the average successful packet delivery ratio among different request rates from source vehicles to a destination rsus. at a low bit rate, the weighted map-based voronoi diagram and the trajectory-based method have a lower delivery ratio due to the longer routing length. however, when the request bit rate increases, the delivery ratio of voronoi diagram and map-based voronoi diagram decrease fast because some rsus are over-loaded and dropping packets from their queues. our proposed methods can relieve the problem of packet dropping and achieve better load balance. as shown in fig 6, we compare the end-to-end delay of packets. in a low traffic condition, both of weighted map-based voronoi diagram and the trajectory-based methods have a longer transmission delay resulted from choosing a farer serving rsu. however, when the request rate increases, both methods can benefit from the shorter queuing delay and reduce the overall end-to-end delay. in the following experiments, we evaluate the load balance effects of the proposed methods in homogeneous and heterogeneous rsu bandwidth settings. as shown in fig. 7, both of weighted map-based voronoi diagram and the trajectory-based methods can relieve the load unbalance problem among rsus. the weighted map-based voronoi diagram method needs to monitor the load information of all rsus and make the decisions of serving range of each rsu in a central server. therefore, it is reasonable that it can achieve a better load-balanced status than trajectory-based method. from fig. 8, we find that some rsus are overloaded when the request bit rate increases; however, our weighted map-based voronoi diagram and the trajectory-based methods can eliminate this problem. (a) heterogeneous bandwidths (case 1) (b) homogeneous bandwidths (case 2) fig. 7 the load balance index with different bandwidth settings of rsus advances in technology innovation, vol. 5, no. 1, 2020, pp. 56-63 62 (a) heterogeneous bandwidths (case 1) (b) homogeneous bandwidths (case 2) fig. 8 the maximum load index among rsus with different bandwidth settings of rsus 5. conclusions and future works in this paper, we discuss the load unbalancing problem among rsus in vanets resulted from the uneven distribution of vehicles and heterogeneity of rsus. then, we proposed two schemes to solve the problem from different aspects. the first one is a centralized rsu-based method, and the second one is a distributed vehicle-based method. both of the proposed methods can balance the loads among rsus and improve the overall packet delivery ratio and latency. in the weighted map -based voronoi diagram, we divides the map into several cells and take the routing length, vehicular density, load indexes of rsus into considerations. this centralized solution can balance the loads among rsus better than the distributed method. however, it needs the vehicular density information and vehicles may change their serving rsu more frequently. therefore, we propose a scalable distributed solution, called trajectory-based method, without the demand of the vehicular density information. moreover, it considers the future locations of vehicles when choosing serving rsu. as a result, it can prevent vehicles from changing their serving rsu frequently. in the future, we tend to use the historical records of vehicular density in a day to analysis the expected delivery ratio and delay to replace the routing length in the current design. in addition, we want to design a reservation mechanism. we will utilize the future trajectory of vehicles to calculate the suitable future serving rsu and reserve bandwidth. this may be able to achieve a better quality of service of vehicles and meanwhile maintain load balancing among rsus. acknowledgement the study is sponsored by minister of science and technology, taiwan under grant no. most 107-2221-e-194-015. references [1] f. cunha, l. villas, a. boukerche, g. maia, a. viana, r. a. f. mini, and a. a. f. loureiro, “data communication in vanets: protocols, applications and challenges,” ad hoc networks, vol. 44, pp. 90-103, july 2016. [2] m. k. priyan and g. u. devi, “a survey on internet of vehicles: applications, technologies, challenges and opportunities,” international journal of advanced intelligence paradigms, vol. 12, no. 1-2, pp. 98-119, 2019. [3] c. englund, l. chen, a. vinel, and s. y. lin, “future applications of vanets,” in vehicular ad hoc networks; c. campolo, a. molinaro, and r. scopigno, eds., springer international publishing: switzerland, pp. 525-544, 2015. [4] h. zhu, k. v. yuen, l. mihaylova, and h. leung, “overview of environment perception for intelligent vehicles,” ieee transactions on intelligent transportation systems, vol. 18, no. 10, pp. 2584-2601, october 2017. [5] c. f. huang and y. h. chen, “message-efficient route planning based on comprehensive real-time traffic map in vanets”, proceedings of engineering and technology innovation, vol. 9, pp. 25-31, 2018. [6] e. uhlemann, “introducing connected vehicles,” ieee vehicular technology magazine, vol. 10, no. 1, pp. 23-31, march 2015. [7] i. wahid, a. a. ikram, m. ahmad, s. ali, and a. ali, “state of the art routing protocols in vanets: a review,” procedia computer science, vol. 130, pp. 689-694, 2018. advances in technology innovation, vol. 5, no. 1, 2020, pp. 56-63 63 [8] s. boussoufa-lahlah, f. semchedine, and l. bouallouche-medjkoune, “geographic routing protocols for vehicular ad hoc networks (vanets): a survey,” vehicular communications, vol. 11, pp. 20-31, january 2018. [9] v. naumov, r. baumann, and t. gross, “an evaluation of inter-vehicle ad hoc networks based on realistic vehicular traces,” proc. of the 7th acm international symposium on mobile ad hoc networking and computing, may 2006, pp. 108-119. [10] v. naumov and t. gross, “connectivity-aware routing (car) in vehicular ad-hoc networks,” ieee infocom 2007-26th ieee international conference on computer communications, 2007, pp. 1919-1927. [11] m. jerbi, s. m. seouci, r. meraihi, and y. ghamri-doudane, “an improved vehicular ad hoc routing protocol for city environments,” ieee international conference on communications, may 2007, pp. 3972-3979. [12] p. prithviraj and g. aniruddha, “maximizing vehicular network connectivity through an effective placement of road side units using voronoi diagrams,” ieee 13th international conference on mobile data management, july 2012, pp. 274-275. [13] u. f. mohd and p. mohammed, “integration of vehicular roadside access and the internet: challenges & a review of strategies,” ieee 2012 international conference on devices, circuits and systems, march 2012, pp. 568-571. [14] g. n. alia, p. h. j. chong, s. k. samantha, and e. chan, “efficient data dissemination in cooperative multi-rsu vehicular ad hoc networks (vanets),” journal of systems and software, vol. 117, pp. 508-527, july 2016. [15] y. dai, d. xu, s. maharjan, and y. zhang, “joint load balancing and offloading in vehicular edge computing and networks,” ieee internet of things journal (early access), vol. 6, no. 3, pp. 4377-4387, june 2019. [16] y. bejerano and s. j. han, “cell breathing techniques for load balancing in wireless lans,” ieee transactions on mobile computing, vol. 8, no. 6, pp. 735-749, june 2009. [17] f. aurenhammer, “voronoi diagrams: a survey of a fundamental geometric data structure,” acm computing survey, vol. 23, no. 3, pp. 345-405, september 1991. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation , vol. 2, no. 3, 2016, pp. 95 98 95 zinc sulfide buffer layer for cigs solar cells prepared by chemical bath deposition rui-wei you, yen-pei fu * department of materials science and engineering, national dong -hwa university, hualien, taiwan received 21 march 2016; received in revised form 28 may 2016; accept ed 02 june 2016 abstract in this study, zns thin films were successfully synthesized by chemica l bath deposition (cbd) with starting materials of nh2-nh2, sc(nh2)2, and znso4‧7h2o. zns thin films were deposited with different time on glass substrates by cbd at 80 o c and ph=9. based on x-ray diffraction (xrd) patterns, it is found that the zns thin films e xh ibit cubic polycrystalline phase. it was found that the optimu m deposition time is 90 min for preparing zns thin film that is suitable as buffer layer for cuin 1-xgaxse2 solar cells. the thin film deposited for 90 min has high transmittance up to 80% in the spectra range fro m 350 n m to 800 n m, and the optica l band gap is about 3.59 ev. ke ywor ds : zns, buffer layer, che mical bath deposition, thin film solar cells , optical property 1. introduction zinc sulfide is a wide-band-gap semiconductor with a range of potential applications in optoelectronic devices . generally, in thin film so la r c e lls base d on cu in s2, cu inse2 , cu(in,ga)se2, cu(in,ga)(sse)2, (cigsse), the buffer layer is mainly the ii-vi type semiconductor, such as cadmiu m sulfide (cds) and zinc su lfide (zns) thin film [1]. the buffer layer with ii-vi type semiconductor is a direct gap semiconductor. the band gap of cadmiu m sulfide is 2.26~2.5 e v, and zinc sulfide is larger than 3.5 e v [2]. the highest efficiency is up to 19.9%, if the cuin1-xgaxse2 thin film solar cell combined with cadmiu m sulfide buffer layer at present [3]. for fear of cadmiu m (cd) toxicity damages our environment, we have to choose a free-cadmiu m process for preparation of buffer layer. we utilize zinc sulfide as buffer layer and the efficiency of cuin1-xgaxse2 thin film solar cell with zinc sulfide buffer layer is up to 18.6% [4]. consequently, there is no toxicity in the use of zinc sulfide and that can lower and lighten the influence on the environment. the buffer layer located between zno windows layer (n-type semiconductor) and cu in1 xga xse2 a bso rp t io n lay e r (p -t yp e semiconductor) is able to eliminate the band discontinuity. the zns buffer layer requires high optical transmission and allows photon to reach absorption layer to e xc ite electron, and then electron-hole pairs (ehp) a re generated. if the buffer layer is too thic k, photoelectron can’t pass through the layer and reach to electrodes. on the contrary, the layer is too thin that couldn’t separate absorption layer and transition conduction electric layer (tco). for better performance, the thic knes s of thin film should be controlled in the range of 30-50 nm. che mical bath deposition (cbd) method has been used for many years to prepare zns large-area and uniform thin film, and it can be prepared under room te mperature [ 5]. utilizing the cbd method produces nano-structure thin film of zinc sulfide with smooth surface and uniform co mposition, and it improves the efficiency of cuin1-xga xse2 thin film solar ce ll. in this study, we attempt to prepare the uniform zinc sulfide thin film by cbd technique using nh3oh, nh2-nh2, sc(nh2)2, and znso4•7h2o as s ta rt in g ma te ria ls , a nd inv est igat e its characterizations such as structural, compositional a n d optical properties. 2. method in this study, zns thin film buffe r layer prepared by chemical bath deposition process . z in c su lfa te (z n so4 •7h2 o) an d th iou re a (sc (nh2)2) were used as the source of zinc ions and sulfide ions, respectively. the reaction s o * corresponding aut hor, email: ypfu@mail.ndhu.edu.tw advances in technology innovation , vol. 2, no. 3, 2016, pp. 95 98 96 copyright © taeti lu t io n was o bt a in ed b y mi xin g 0.1m znso4•7h2o, 0.3m sc(nh2)2 , and 1.5m n2h4, whereby hydrazine (n2h4) was used as a comple x agent. the ph of the reaction solution was adjusted to 9 by ammonia (nh3oh), reaction solution temperature was controlled at 80 o c and the rotational speed of stirrer was controlled in 50 rp m. soda-lime g lass substrates were used as substrates for the deposition of zns films. be fore deposition, the substrates were ultrason ically c leaned with acetone, rinsed with deio nized water and dried in a ir. to investigate the effect of deposition time on properties of zns, the substrates were collected every 30 minutes in which the deposition-time is set in the range of 30 to 180 min. then these samples were c leaned by deionized water and dried with a n2 gas stream. in order to obtain crystalline zns, t hese as-deposited zns specimens required a post-annealing in a tube furnace for l h under ar atmosphere at 300 o c. the average roughness of the zns films was investigated by surface profile measuring system (alpha-step, veeco de ktak 3 st ). the co mposition of zns thin films was analyzed by e nergy dispersive spectrometer (eds, horiba noran instrument). the crystalline phase of the annealed zns films were characterized by x-ray diffractometer (xrd, rigaku d/ max-2500 v) with a wave length of 1.5406å fro m the cu kα radiation with 2θ ranging fro m 20 o to 60 o . op t ic a l p ro p e rt ies o f th in fil ms we re charac terized by an uv-vis spectrometer (jasco v-650 spectrophotometer). the band gaps (eg) of zns films were determined by the relat ionship of the transmittance and thickness of the film. 3. results and discussion fig. 1 shows x-rays diffraction of the zinc sulfide thin film annealed at 300 o c for 1 h under argon atmosphere. zns e xists with two stru ctures, one is cubic with zinc-blende type the other is hexagonal with wurtzite type. zns thin films prepared via the che mica l bath deposited are highly disordered, but it can be transformed into a wurt zite-2h phase by annealing [ 6]. göde et al. reported that an a morphous zns film o btained at bath temperature of 60–70 o c and a wurtzite-2h phase acquired at 80 o c [7]. in this study, zns thin film prepared by cbd is zinc-blend type with cubic structure being in correspondence with jcpds card no. 79-043. the xrd pattern reveals a wide diffraction peak fro m 25 o to 30 o . there are three d iffraction peaks corresponding to (111), (220) and (311); however, the diffract ion peaks in (200) and (311) a re not clear indicating the zinc sulfide thin film with low crystallization. fig. 1 x-rays diffraction pattern of zns thin film at deposition time o f 90 min, then annealed at 300 o c for 1h under ar at mosphere to understand the effect of the deposition time on the composition of films, the energy dispersive spectrometer (eds) was used to analyze the ato mic rat io of zn/s listed in table 1. it is found that the ratio of zn/s is close to 1 for films with various deposition times. table 1 the composition ratio of zn/s at different deposited time composit ion deposit ion t ime (min) 30 60 90 120 150 180 zn (at om %) 50.4 51.9 50.3 51.2 50.8 49.8 s (at om %) 49.6 48.1 49.7 48.8 49.2 50.2 fig. 2 the thic kness of zns films as a function of deposition time, after annealing at 300 o c for 1 h under ar atmosphere advances in technology innovation , vol. 2, no. 3, 2016, pp. 95 98 97 copyright © taeti fig. 2 shows the zns film thickness as a function of deposition time. apparently, the relation between thickness and deposition time is divided into two stages, the first stage is the linear growth for thin film and the thic kness grows fro m 124 n m to 252 n m in the period of 30 ~ 120 min. however, the second stage is the e xponential growth; the thickness significantly increases from 352 n m to 750 n m for deposition time fro m 120 to 180 min. at the second stage, the zns film undergoes homogeneous nucleation. as deposition time increasing above 150 min, ho mogeneous particles begin to deposit on the substrate leading the significant enhancement in g rowth rate of thin film. tab le 2 is zns thin film thic kness and roughness as function of deposition time. the average roughness of film for deposition time of 30 min is significantly high due to the facts that the heterogeneous deposition on substrate is still not uniform at the initia l stage. as depos ition time fro m 30 to 120 min, the roughness is relative ly lower and the thin films become uniform gradually. when deposition time for 180 min, the rate of ho mogeneous deposition on substrate increased, and the average roughness of zns film is increased up to 94 nm. table 2 the thickness and average roughness for zns films with diffe rent deposition time dep osition time (min) 30 60 90 120 150 180 thickness (nm) 124 160 249 252 358 750 average roughness (nm) 87 30 58 34 32 94 fig. 3 shows the optical transmittance in the wavelength range of 300 800 n m for the zns films deposited on the sodium glass substrates as a function of diffe rent deposition time . the zns films reveled h igh optical trans mittance in the range of 70 ~ 80% at visib le wavelength; therefore, the films are suitable as buffer layers in cigs-based solar cells. the h ighest transmittance of 85% is located about wavelength of 425 n m for deposition time of 90 min. as deposition time increased fro m 90 to 150 n m, the absorption edge shifted gradually fro m 400 n m to 500 n m. the sharp absorption feature is due to the uniform zns thin films and the lo w concentration of defects in the films. ho wever, as the deposition time increased, t h ic kn e ss b e ca me t h ic ke r, a n d mo re ho mogeneous particles deposited on the glass substrate leading the optical transmittance lo wer than 80% for films that deposition time is above 90 min. based on optical results, the film deposited for 90 min e xhibited good optical propert ies , and the shortest wavelength of adsorption edge, which could ma ke cigs solar ce ll with a higher short circuit density (vsc) [8]. fig. 3 t ransmission spectra of zns thin films at diffe rent deposition time on g lass substrates . bath conditions: [znso4] =0.1m, [sc(nh4)]=0.3m, [nh2-nh2]=1.5m, ph=9 the optical band gap of zns films could be obtained using the tauc relationship revealed as follows [9]. αhν= a(hν-eg) n (1) where a is a constant, h is planck’s constant, v is the photon frequency, eg is the optical band gap energy, and n is 1/2, respectively. the band gap value was determined fro m the intercept of the straight-line portion o f the (αhν) 2 against the graph on the hν-axis. the band gap values of the zns thin films prepared by cbd with various deposition time a re revea led in table 3. the band gap values are greater than 3.5 e v for the films deposited fro m 30 to 120 min. higher band gap values could match well with cigs solar cell, and the solar cells gain better quantum efficiency [10]. table 3 the band gap for zns thin films with various deposition time dep osition time (min) 30 60 90 120 150 180 band gap (ev) 2.57 3.53 3.59 3.54 3.45 3.30 advances in technology innovation , vol. 2, no. 3, 2016, pp. 95 98 98 copyright © taeti 4. conclusions in this study, the zns thin films with sphalerite structure prepared by chemica l bath deposition using zinc sulfate, thiourea, and co mp le x of hydrazine as starting materia ls. the thic kness of zns thin film varies with deposition time fro m 124 n m of 30 min to 750 n m of 180 min. the zns films reveled h igh optical transmittance in the range of 70 ~ 80% at visible wave length. the band gap values for zns films are in the range of 2.57 ~ 3.61 e v. based on the results, the following bath conditions, [znso4]=0.1m , [sc(nh4)]=0.3m, [nh2-nh2] [ =1.5m , ph=9 and t =80 o c deposited for 90 min, revealed the optimu m zns film properties, which is suitable as buffe r layer for cuin1-xgaxse2 solar ce lls, and the properties of the film are described as follo ws. (1). xrd pattern reveals a cubic zinc b lend structure with the typical co mposition rat io of zn/s = 50.3:49.7, which is very c lose to the stoichiometry of t he zns compound. (2). the high est transmittance of 85% is located about wavelength of 425 n m and the optical band gap is about 3.59 ev. references [1] a. wei, j. liu, m. zhuang, and y. zhao, “preparation and characterizat ion of zns thin films prepared by chemical bath deposition,” materials science in se miconductor processing, vol. 16, pp. 1478-1484, de cember 2013. [2] b. g. street man and s. k. banerjee, “solid state electron devices ,” 6th edition, pearson prentice hall, new jersey, pp. 158-208, 2006. [3] i. repins, m. a. contreras, b. egaas, c. deha rt, j. scharf, c. l. perkins, b. to, and r. noufi, “accelerated publication 19.9% efficient zn o/cds/ cuin ga se2 solar ce ll wit h 81.2% fill fa c to r,” prog ress in photovoltaics: research and applications 16, pp. 235, 2008. [4] r. n. bhattacharya and k. ra manathan, “cu (in,ga )se2 thin film solar ce lls with buffer layer alternative to cds,” sola r en ergy, vol. 77, pp. 679-683, 2004. [5] j. liu, a. wei, and y. zhao, “ effect of diffe rent comp le xing agents on the properties of chemica l-bath-deposited zns thin films,” j. alloys co mpd., vol. 588, pp. 228-234, 2014. [6] i. o. oladeji and l. chow, “synthesis and processing of cds/zns mu lt ilayer films for solar cell application,” thin solid films, vol. 474, pp. 77-83, 2005. [7] f. göde, c. umus, and m. zor, “investigations on the physical properties of the po lycrystalline zns thin films deposited by the chemical bath deposition method,” j. cryst. growth, vol. 299, pp. 136-141, 2007. [8] s. w. shin, s. r. kang, k. v. gurav, j. h. yun, j. h. moon, j. y. lee , and j. h. kim, “a study on the improved growth rate and morphology of che mically deposited zns thin film buffer layer for thin film solar ce lls in acid ic mediu m,” so lar energy, vol. 85, pp. 2903-2911, 2011. [9] s. w. sh in, g. l. agawane, m. g. gang, a. v. moholkar, j. h. moon, j. h. kim, and j. y. lee, “preparat ion and characteristics of chemical bath deposited zns thin films : effects of d ifferent co mp le xing agents,” j. alloys compd., vol. 526, pp. 25-30, 2012. [10] r. n. bhattacharya, k. ra manathan, l. gedvilas, and b. keyes “cu(in,ga)se2 thin-film solar ce lls with zns(o,oh), zn–cd–s(o,oh), and cds buffe r layers,” j. phys. chem. solids, vol. 66, pp. 1862 -1864, 2005.  advances in technology innovation, vol. 5, no. 3, 2020, pp. 156-165 side resistance of drilled shafts socketed into rocks for serviceability and ultimate limit states yit-jin chen * , cheng-chieh hsiao, anjerick topacio department of civil engineering, chung yuan christian university, taoyuan, taiwan received 26 april 2019; received in revised form 08 august 2019; accepted 04 october 2019 doi: https://doi.org/10.46604/aiti.2020.4155 abstract this study evaluates the analysis models of side resistance in rock sections by utilizing a wide variety of load test data. available analytical models including the empirical adhesion factor versus the rock’s uniaxial compressive strength and its root are analyzed and compared statistically to determine the optimum relationships. the interpretation criteria for the l1 and l2 methods are used to analyze the load test results for serviceability and ultimate limit states, respectively. the analysis results show that the relationship model with the empirical adhesion factor versus the root of the rock’s uniaxial compressive strength exhibits better correlation than the one with the rock’s uniaxial compressive strength. moreover, the general coordinate axes regression equation demonstrates better reliability than the semi-logarithmic and full logarithmic axes equations for both limit states. based on these analyses, specific design recommendations for the side resistance of drilled shafts socketed into rocks are developed and provided with the appropriate statistics to verify their reliability. keywords: drilled shaft, socketed into rock, side resistance, statistical analysis 1. introduction limited land for building structures frequently lead to the construction of high buildings to save land occupancy. the structures of such buildings are heavy. to stabilize these structures, a deep foundation is often used to transfer the load of the superstructure to the underlying bearing layer. occasionally, a certain length of the pile foundation is embedded into the rock layer to increase the effects of stabilizing the superstructure. among pile foundations, drilled shafts (also called cast-in-place piles, drilled piers, or bored piles) are frequently used as deep foundation because they produce less noise and vibration during construction and meet the required pile diameter and depth. in addition, drilled shafts can provide sufficient lateral support to resist the force of the superstructure from earthquakes and wind loads. when a pile is subjected to an axial load, the load is transmitted along the length of the pile toward the hard soil or rock layer, as shown in fig. 1. pile capacity includes side and tip resistances for resisting axial load. side resistance plays an important role when a pile is socketed into rocks. in the analysis of side resistance, the analysis concept of the shaft in cohesive soil and rocks is the same. in cohesive soil, the total stress analysis method is frequently used to evaluate unit side resistance (fs). the fs value can be computed on the basis of an empirical coefficient (a), which is the adhesion factor between the soil and the shaft, and the average soil undrained shear * corresponding author. e-mail address: yjc@cycu.edu.tw tel.: +886-3-2654227; fax: +886-3-2654299 advances in technology innovation, vol. 5, no. 3, 2020, pp. 156-165 157 strength (su) over the shaft length. moreover, the α value is related to su, and the initial asu relationship of the drilled shafts was developed by stas and kulhawy [1], as shown in fig. 2. however, the su values in their analysis were obtained from random test types. the su value can be determined from various test types, but the results present an evident difference. therefore, chen and kulhawy [2, 3] later adopted a unique test type for su from the consolidated isotropically undrained triaxial compression (ciuc) test, which is denoted as su(ciuc), the reference plane for a consistent test. the improved aciuc-su(ciuc) relation is illustrated in fig. 3. the data distribution of the improved aciuc-su(ciuc) correlation is superior to that of the previous result shown in fig. 2. these concepts have been expanded to the analysis of side resistance for driven precast concrete piles [4, 5]. fig. 1 axial load resisted by a pile foundation fig. 2 early a-su relation [1] fig. 3 improved aciuc-su(ciuc) relation [2,3] the analysis model of side resistance for drilled shafts socketed into rocks is similar to the conventional total stress analysis in cohesive soil. however, the uniaxial compressive strength (qu) is adopted in rocks instead of the su in cohesive soil. the fs value expressed by using qu or its rootz ( uq ) can be expressed as: su(ciuc)/pa advances in technology innovation, vol. 5, no. 3, 2020, pp. 156-165 158  s uf q (1)  s uf q (2) accordingly, numerous studies have been conducted and have resulted in various relationships in determining the side resistance value of drilled shafts. kulhawy and goodman [6] recommended an  value of 0.15 and found that the value of qu should be less than 5 mpa. they used eq. (1) to determine side resistance. moreover, several researchers have utilized eq. (2) to determine side resistance during analyses. hooley and brooks [7] studied the friction force of piles in overcompacted clay, soft rock, and weathered rock. they recommended a qu value between 0.25 and 3.0 mpa and an α value of 0.15. the american association of state highway and transportation officials [8] suggested that the qu value should be less than 40 mpa and the α value should be between 0.03 and 0.04. horvath et al. [9] suggested that the qu value should be less than 40 mpa, and the α value should be between 0.2 and 0.3. ku et al. [10] conducted related research on the western soft rock of taiwan and recommended an α value of 0.16-0.21 but did not recommend any value for qu. lastly, yang et al. [11] found that the α value of piles socketed into rocks in northern taiwan is between 0.1 and 0.5, and the relation α versus uq is shown in fig. 4. fig. 4 αuq relationship based on northern taiwan's load tests [11] the aforementioned descriptions indicate that the current relations are based on a few load test data, and the range of the values of α is extremely broad and scattered. thus, the current study was conducted to reassess side resistance for the serviceability limit state (sls) and ultimate limit state (uls) of drilled shafts socketed into rocks. moreover, the various influencing factors were analyzed to provide a reliable design recommendation. 2. database for analysis to evaluate the side resistance behavior of drilled shaft foundations socketed into rock sections, this study collected data from taiwan, turkey, the usa, italy, and singapore [12]. a total of 44 load tests embedded into rocks were obtained, including 28 sandstones, 3 mudstones, 3 limes, 3 soft sandstones, 3 hard shales, 2 tuffs, and 3 amphibolite. in addition, 81 sets of pile load data instruments were used from these load tests. therefore, one load test may consist of several instruments for different rock sections. the average a and qu values were adopted for the analysis of each load test. the collected pile information was accompanied by soil/rock layer information, pile foundation information, load–displacement curve, and load distribution curve along the pile length. table 1 provides the details for the geometries, geotechnical parameters, and interpreted results of these load test data. table 2 lists the statistical summary of these data [12]. u aq , mp advances in technology innovation, vol. 5, no. 3, 2020, pp. 156-165 159 table 1 geometries, geotechnical parameters, and interpreted results of load test data shaft no. depth, d diameter, b test depth from gl qu fs(l1) 1 fs(l2) 2 (m) (m) (m) (mpa) (mpa) (mpa) tp1 37.6 0.95 28.0~36.0 12.77 0.268 0.63 tp2 42.6 0.95 34.2~41.0 10.01 0.176 0.53 tp3 20.0 1.2 4.0~19.0 0.60 0.063 0.16 tp4 66.0 1.5 61.5~66.0 4.17 0.160 0.59 tp5 25.5 1.5 21.5~25.5 0.20 0.088 0.24 tp6 48.0 1.5 40.0~48.0 1.21 0.095 0.31 tp7 48.0 1.5 40.0~48.0 0.80 0.080 0.25 tp8 53.0 1.5 49.5~53.0 1.03 0.110 0.52 tp9 30.0 1.2 26.4~30.0 2.26 0.070 0.24 tp10 53.5 2.0 37.2~53.3 2.46 0.060 0.22 tp11 54.5 2.0 38.2~54.5 0.34 0.069 0.17 tp12 24.0 0.8 16.7~24.0 0.70 0.018 0.11 tp13 48.0 1.5 47.7~52.0 0.65 0.090 0.13 tp14 47.7 1.2 41.7~47.7 0.16 0.115 0.14 tp15 59.0 1.2 55.8~59.0 3.00 0.110 0.29 tp16 29.0 1.2 24.9~29.0 0.13 0.060 0.16 tp17 45.0 1.2 35.0~42.5 3.47 0.062 0.20 tp18 38.0 1.0 35.0~37.4 13.00 0.150 0.37 tp19 60.4 1.5 58.4~59.8 13.70 0.300 0.58 tp20 48.0 1.5 44.5~48.0 13.64 0.277 1.15 tp21 46.0 0.41 43.4~45.3 2.54 0.160 0.53 tp22 10.0 1.2 2.5~10.0 1.30 0.053 0.21 tp23 9.1 1.2 3.5~9.1 1.60 0.050 0.22 tp24 8.0 1.2 4.5~8.0 1.00 0.050 0.20 tp25 25.0 1.0 15.5~25.5 4.26 0.135 0.53 tp26 17.0 1.8 3.0~16.4 1.70 0.064 0.24 tp27 36.0 1.0 32.9~35.4 1.03 0.092 0.18 tp28 27.5 1.2 25.5~27.0 1.51 0.075 0.47 tp29 22.8 0.8 12.3~22.2 0.52 0.110 0.35 tp30 20.4 0.8 15.5~19.8 1.86 0.111 0.30 tp31 26.4 1.0 16.8~26.8 0.32 0.082 0.23 tp32 8.7 0.7 5.6~8.8 5.00 0.227 0.50 tp33 11.0 0.7 5.6~11.0 5.00 0.40 tp34 18.5 1.2 11.0~18.5 0.90 0.100 0.15 tp35 39.0 1.2 26.0~37.0 3.00 0.046 0.20 tp36 13.5 1.2 11.0~13.5 6.00 0.148 0.42 tp37 7.3 0.7 2.1~7.3 6.82 0.133 0.46 tp38 13.5 1.35 4 .0~10.0 7.00 0.080 0.34 tp39 11.5 1.5 3.5~9.6 7.77 0.088 0.40 tp40 59.0 1.2 54.6~58 0.26 0.050 0.26 tp41 73.0 1.5 67.0~73.0 0.30 0.053 0.15 tp42 76.0 1.3 71.0~76.0 0.45 0.012 0.22 tp43 40.7 2.2 28.5~39.5 0.37 0.026 0.04 tp44 15.0 1.2 2.0~14.5 0.93 0.053 0.14 1 fs(l1) is the unit friction interpreted from l1 method 2 fs(l2) is the unit friction interpreted from l2 method 3. analysis method two methods were used in this study to obtain the side resistance of the drilled shaft foundation. first, the value was directly obtained from the t-z curve, which defines the shear stress–vertical displacement response of the soil at each particular depth. second, the load–displacement curve and the load distribution curve throughout the pile length were utilized. by considering these test results, the interpretation of kulhawy and hirany (called the l1-l2 interpretation) [13] was adopted to construe unit side resistance under various limit states. advances in technology innovation, vol. 5, no. 3, 2020, pp. 156-165 160 table 2 statistics of load test data for analysis interpretation method n statistical analysis pile geometry qu (mpa)  fs (mpa) pile length d (m) pile diameter b (m) d/b l1 44 range 7.3-76.0 0.41-2.2 6.7-112.2 0.13-13.7 0.011-0.719 0.012-0.50 mean 33.10 1.2 28.13 3.372 0.098 0.104 sd 17.53 0.39 18.35 3.92 0.14 0.078 cov 0.53 0.33 0.65 1.16 1.42 0.75 l2 44 range 7.3-76.0 0.41-2.2 6.7-112.2 0.13-13.7 0.03-1.23 0.04-1.19 mean 33.10 1.2 28.13 3.372 0.277 0.33 sd 17.53 0.39 18.35 3.92 0.31 0.21 cov 0.53 0.33 0.53 1.16 1.12 0.64 this l1-l2 interpretation is a graphical method that utilizes the load–displacement curve. l1 is defined as the endpoint of the initial linear segment. as illustrated in fig. 5, the load and displacement are represented by ql1 and 1l respectively. l2 is the starting point of the final line segment. the load and displacement are represented by 2l q and 2l  respectively. an interpretation example based on one of the pile report data is shown in fig. 6. chen and fang [14] studied the l1 and l2 interpretation by using many load test data and established the interrelationships among l1, l2, and several typical interpretation criteria. their analysis showed that l1 = 0.5l2, on average. this finding indicates a factor of safety of 2 if l1 is used for the design. moreover, l1 should be used in sls and l2 should be used in uls. tang et al. [15] recently conducted statistical analyses to evaluate model factors in a reliability-based design for drilled shafts under axial loading with sls and uls. fig. 5 regions of load–displacement curve fig. 6 l1-l2 interpretation of the t-z curve example [12] 4. analysis results 4.1. correlation of α versus qu eq. (1) was used to evaluate the pile load test data for α, and the results of the l1 and l2 interpretations are presented in figs. 7 and 8, respectively, including the average results of the data obtained from 44 single pile reports. unit side resistance was determined using the load–displacement curve or the t–z curve. the data in figs. 7 and 8 were then plotted into three coordinate axes, namely (a) general coordinate, (b) semi-logarithmic, and (c) full logarithmic axes, to determine the optimum trend for each interpretation. the statistical analysis is provided in table 3 to compare the three equations. the statistical results of the coefficient of determination (r 2 ), standard deviation (sd), and coefficient of variation (cov) are also listed in the figures and tables. fsl2 fsl1 displacement, (mm) s o c k e t f ri c ti o n , fs ( k p a ) final linear region transition region initial linar region displacement l o a d advances in technology innovation, vol. 5, no. 3, 2020, pp. 156-165 161 (a) general coordinate (b) semi-logarithmic coordinate (c) full-logarithmic coordinate fig. 7 a-qu relationships for l1 interpretation (a) general coordinate (b) semi-logarithmic coordinate (c) full-logarithmic coordinate fig. 8 a-qu relationships for l2 interpretation advances in technology innovation, vol. 5, no. 3, 2020, pp. 156-165 162 table 3 statistical results for a-qu relationships interpretation method coordinate form regression equation n r 2 sd cov l1 general coordinate 0.01 0.07 /  uq 44 0.82 0.06 0.59 semi-logarithmic 0.13 0.08 log( )   uq 0.50 0.10 1.00 full logarithmic log( ) 2.60 0.70 log( )    uq 0.75 0.09 0.87 l2 general coordinate 0.07 0.16 /  uq 44 0.81 0.13 0.45 semi-logarithmic 0.36 0.19 log( )   uq 0.60 0.19 0.66 full logarithmic log( ) 1.43 0.68 log( )    uq 0.79 0.15 0.51 the results of the statistical analyses indicated that the three coordinate axes used in the study caused differences in terms of reliability with each interpretation used. as shown by the results, the general coordinate regression equation has the lowest cov for the l1 and l2 interpretations. therefore, the general coordinate regression equation exhibits higher reliability than the other equations. 4.2. correlation of α versus uq (a) general coordinate (b) semi-logarithmic coordinate (c) full-logarithmic coordinate fig. 9  uq relationships for l1 interpretation the analysis presented in this section is commonly used in determining the side resistance of drilled shafts socketed into rock sections. the difference is that the uniaxial compressive strength was analyzed through its root. the results showed that several discrete data points are scattered, resulted in an inconsistent plot. these data were obtained from piles socketed into mudstone (n = 3) because foundations embedded into mudstone tend to have a greater tip capacity than friction. thus, these data were omitted in this section to obtain reasonable results. the new data were then plotted into (a) general coordinate, (b) semi-logarithmic, and (c) full logarithmic axes, as shown in figs. 9 and 10 for the l1 and l2 interpretations respectively. the advances in technology innovation, vol. 5, no. 3, 2020, pp. 156-165 163 data were plotted in the three coordinate axes in order to determine the optimum trend for each interpretation. statistical analysis is provided in table 4 to compare the two interpretations used in the study. (a) general coordinate (b) semi-logarithmic coordinate (c) full-logarithmic coordinate fig. 10  uq relationships for l2 interpretation table 4 statistical results for  uq relationships interpretation method coordinate form regression equation n r 2 sd cov l1 general coordinate 0.03 0.06 /  uq 41 0.56 0.03 0.41 semi-logarithmic 0.10 0.05 log( )   uq 0.40 0.04 0.48 full logarithmic log( ) 2.53 0.48 log( )    uq 0.30 0.04 0.48 l2 general coordinate 0.14 0.12 /  uq 41 0.40 0.09 0.36 semi-logarithmic 0.29 0.12 log( )   uq 0.38 0.10 0.37 full logarithmic log( ) 1.34 0.45 log( )    uq 0.39 0.10 0.37 similarly, the general coordinate regression equation has the lowest cov for the l1 and l2 interpretations. therefore, the general coordinate regression equation exhibits higher reliability than those of the semi-logarithmic and full logarithmic coordinates. 4 .3. comparison of  versus qu and uq from the preceding analysis results, the regression equation developed in the general coordinate axes should be used when analyzing drilled shafts socketed into rocks. moreover, in accordance with the cov values computed in the study, the advances in technology innovation, vol. 5, no. 3, 2020, pp. 156-165 164  uq relationship yielded better correlation than that of α – qu. table 5 provides the comparison of the cov values of the l1 and l2 interpretations of theα – qu and  uq relationships. the  uq relationship resulted in a lower cov than that of the α–qu relationship, providing accurate results for sls and uls. fig. 4 indicates that the current proposed relation between a and uq , and the range of α values is extremely broad and scattered. with the completion of the study, data correlation through analysis was improved, which increased the reliability of its application to sls and uls. table 5 statistical results for αuq relationships interpretation method relationship regression equation n r 2 sd cov l1  uq 0.01 0.07 /  uq 44 0.82 0.06 0.59  uq 0.03 0.06 /  uq 41 0.56 0.03 0.41 l2  uq 0.07 0.16 /  uq 44 0.81 0.13 0.45  uq u0.14 0 q.12 /   41 0.40 0.09 0.36 5. conclusions this study utilized numerous load test data to evaluate the side resistance of drilled shafts socketed into rocks. upon analysis of the test results, the following conclusions were drawn. 1. this study developed α–qu and  uq relationships plotted into three coordinate axes. the three coordinate axes were then compared, and the analysis results showed that the general coordinate axes regression equation exhibited better reliability than the semi-logarithmic and full logarithmic axes equations. 2. the study proved that using the  uq relationship is reliable in the analysis of drilled shafts socketed into rocks for the l1 and l2 interpretations because of its lower cov value than that of the α–qu relationship. the relations recommended in this study also yielded higher reliability than previous ones. 3. to determine sls, the l1 interpretation of the  uq relationship developed in the study using the regression equation, 0.03 0.06 /  uq , is recommended for the developed equation, r 2 = 0.56, sd = 0.03, and cov = 0.41. 4. to determine uls, the l2 interpretation of the  uq relationship developed in the study using the regression equation, u 0.14 0 q.12 /   , is recommended for the developed equation, r 2 = 0.40, sd = 0.09, and cov = 0.36. acknowledgments this research work was supported by the ministry of science and technology (most), taiwan, under grant 107-2221-e-033-014. conflicts of interest the authors declare no conflict of interest. references [1] c. v. stas and f. h. kulhawy, “some observations on undrained side resistance of drilled shafts,” foundation engineering, current principles and practices, asce, pp. 1011-1025, 1984. advances in technology innovation, vol. 5, no. 3, 2020, pp. 156-165 165 [2] y. j. chen and f. h. kulhawy, “case history evaluation of drilled shafts behavior,” electric power research institute, palo alto, report tr-104601, p. 356, 1994. [3] y. j. chen and f. h. kulhawy, “evaluation of undrained side and tip resistances for drilled shafts,” proc. soil and rock america, vol. 2, pp. 1963-1968, 2003. [4] m. c. marcos and y. j. chen, “evaluation of side resistance of driven precast concrete piles,” materials science and engineering, iop, vol. 658, no. 1, 012005, 2019. [5] m. c. marcos, and y. j. chen, “applicability of various load test interpretation criteria in measuring driven precast concrete pile uplift capacity,” international journal of engineering and technology innovation, vol. 8, no. 2, pp. 118-132, 2018. [6] f. h. kulhawy and r. e. goodman, foundation in rock, ground engineers reference book, butterworth and co. ltd., 2005. [7] p. hooley and s. r. l. brooks, “the ultimate shaft frictional resistance mobilized by bored piles in overconsolidated clays and socked into weak and weathered rock,” the engineering geology of weak rock, balkema, pp. 447-455, 1993. [8] american association of state highway and transportation official (aashto), standard specifications for highway bridges, 1992. [9] r. g. horvath, t. c. kenney, and p. kozicki, “methods of improving the performance of drilled piers in weak rock,” canadian geotechnical journal, vol. 20, no. 4, pp. 758-772, 1983. [10] c. s. ku, j. weng, x. liu, and x. lin, “study on side resistance of soft rock foundations,” proc. rock engineering symposium, tamsui, pp. 224-231, 2007. [11] z. y. yang, j. q. shiau, j. ching, y. s. lee, and c. j. chen, “side resistance of pile socketed into rock in taiwan area,” proc. rock engineering symposium, kaohsiung, pp. 75-83, 2010. [12] c. c. hsiao, “evaluation of side resistance for drilled shafts socketed into rocks,” master thesis, department of civil engineering, chung yuan christian university, 2018. [13] a. hirany and f. h. kulhawy, “interpretation of load tests on drilled shafts: axial compression,” foundation engineering: current principles & practices (gsp22), asce, new york, pp.1150-1159, 1989. [14] y. j. chen and y. c. fang, “critical evaluation of compression interpretation criteria for drilled shafts,” journal of geotechnical and geoenvironmental engineering, asce, vol. 135, no. 8, pp. 1056-1069, 2009. [15] c. tang, k. k. phoon, and y. j. chen, “statistical analyses of model factors in reliability-based limit state design of drilled shafts under axial loading,” journal of geotechnical and geoenvironmental engineering, asce, vol. 145, no. 9, pp. 04019042-1-19, 2019. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation, vol. 5, no. 2, 2020, pp. 76-83 machine vision and deep learning based rubber gasket defect detection chao-ching ho * , eugene su, po-chieh li, matthew j. bolger, huan-ning pan graduate institute of manufacturing technology and department of mechanical engineering, national taipei university of technology, taipei, taiwan received 17 may 2019; received in revised form 08 august 2019; accepted 18 october 2019 doi: https://doi.org/10.46604/aiti.2020.4278 abstract this study develops an automated optical inspection system for silicone rubber gaskets using traditional rule-based and deep learning detection techniques. the specific object of interest is a 5 mm × 10 mm × 5 mm mobile device power supply connector gasket that provides protection against foreign body inclusion and water ingression. the proposed system can detect a total of five characteristic defects introduced during the mold-based manufacture process, which range from 10-100 μm. the deep learning detection strategies in this system employ convolutional neural networks (cnn) developed using the tensorflow open-source library. through both high dynamic range image capture and image generation techniques, accuracies of 100% and 97% are achieved for notch and residual glue defect predictions, respectively. keywords: traditional rule-based strategy, deep learning, convolutional neural networks (cnn), image recognition, image processing, deep residual learning 1. introduction mobile device designers and manufacturers have made considerable efforts to increase the "ruggedness" of their products. one such effort has been the industry-wide adoption of the international electromechanical commission (iec) published inclusion protection (ip) that rates electronic devices according to their protection against solid inclusion and water ingression. as mobile device components have become smaller and ip ratings higher, it has become necessary for manufacturers to be able to detect defects that are a few micrometers in size. defect detection at this scale requires automated optical inspection techniques as traditional instrument-aided methods have been rendered obsolete. the demand for silicone rubber-based components in the mobile device industry to provide tight seals to prevent foreign body inclusion and water ingression has been a boost for taiwanese manufacturers. the strength of taiwanese science and engineering has proved fundamental for the development and manufacture of natural and synthetic rubber products that require complex chemical refinement, bridging, and coloring sequences. rubber mold manufacturing techniques can be broadly classified into three categories: sheet molding, injection molding, and extrusion molding, each of which introduces characteristic defects at varying rates and distributions. as mobile device designers have decreased component size and increased ruggedness requirements, manufacturers have realized the necessity of automated optical inspection techniques for detecting defects a few micrometers in size. furthermore, the non-invasive nature of optical inspection techniques provides the speed necessary for mass manufacturing. in 1980, a. cornforth and c. james proposed a rubber gasket inspection system to identify the void and lamination disband errors between the nylon tip and the rubber ring using an ultrasonic device [1]. recently, systems have been developed for the prediction of silicone rubber solidification time through the optical measurement thickness [2]. furthermore, optical * corresponding author. e-mail address: hochao@ntut.edu.tw advances in technology innovation, vol. 5, no. 2, 2020, pp. 76-83 77 inspection-based systems to inspect the bubbles present in solid and liquid silicone rubber in the molding process have also been developed [3]. machine vision measurement provides a non-invasive inspection process widely deployed in a broad range of smart manufacturing systems [4-6], e.g. leather production and manufacturing [7], leather defect-recognition [8], solenoid manufacturing [9], and in-process led chip mounting alignment [10]. all the examples mentioned above feature dim and low reflective surface materials which make the proper illumination for defect detection difficult. in our previous study [11], we found that while machine vision-based systems are capable of defect classification performance, the performance of these systems was affected by variation in lighting conditions, and these systems were incapable of qualitative defect detection. therefore, to ensure the homogeneity of the silicone rubber gaskets with its isolated components, a specified on-line measurement system for this application was developed in this study as shown in fig. 1. this study develops a fully automated optical inspection system for silicone rubber mobile device power supply connector gaskets. the defect detection application employs traditional rule-based and deep learning detection methods. recently, several studies have been conducted on the application of deep learning methods for machine vision-based optical inspection systems [12-16]. (a) the silicone rubber gaskets. (b) inclusion (c) notches (d) inner residual glue fig. 1 examined rubber gaskets (a) inclusion defects (b) skewness fig. 2 two types of side layer defects for the silicone rubber gaskets from side-view 2. rule based image detection methodology the mobile device power supply gasket inspection unit has two separate image capture stations, one is the top view and the other is the side view station as shown in fig. 2. the top view station is used to examine three defects: outer ring inclusion, inner ring notches, and inner ring residual glue, whereas, the side view station checks for two defects: top edge inclusion and metal contact leg skewness. 2.1. top-view station the top view station captures the inspected object images using a basler aca2500-14μm monochrome area scan industrial camera, fitted with a 50 mm cctv lens. at the time of inspection, the camera was set to a ×0.343 zoom for a 16 × 12 mm² sensor size and a precision of 6.4 μm/pixel. advances in technology innovation, vol. 5, no. 2, 2020, pp. 76-83 78 the top view station begins by acquiring multi-exposure images of the silicone rubber seal rings. the image processing application then creates both a contour approximation and a convex hull representation of the outer seal ring. as the eccentricity and straightness of the outer seal ring are rotation invariant, the application uses horizontal and vertical projections to segment the image. the application then calculates the difference in the segmented contour and convex hull representations to predict manufacturing defects as shown in fig. 3. because the top view optical inspection application does not use morphological operations or filtering, the image preserves the geometry of the inspected object at the pixel level. thus, the application can correctly detect inclusion defects as small as 60 μm in size. it is found that traditional rule-based image processing techniques are unable to accurately detect notch and inner residual glue defects because these defects are more qualitative in nature, thereby making them more suitable for deep-learning approaches. fig. 3 top view flowchart using traditional rule based image processing strategy 2.2. side-view station the second inspection station uses a basler ca1600-20μm monochrome area scan industrial camera equipped with an open four-view mirror to capture four side views. while the four-view mirror reduces the overall number of stations needed in the system, it increases the pre-processing time necessary to divide the distinct views as shown in fig. 4. as illustrated in fig. 5(a), the side view defect detection application first partitions the image using a predetermined region of interest (roi). while this roi attempts to consider most variations in the power supply gasket placement, excessively skewed or absent gaskets may lead to unpredictable results. the application uses a binary threshold to the roi. the image processing application segments the image via horizontal projection techniques, which include additional parameters in the segmentation routine to control the magnitude of separation or overlap between the images as presented in fig. 5(b). additionally, a separate metal connection leg absence defect detection application is implemented using contour finding techniques. as shown in fig. 6, the routine centroid calculations are allowed for the labeling of contour centroids to check for potential skew detection. fig. 4 side view flowchart with using traditional rule-based strategy advances in technology innovation, vol. 5, no. 2, 2020, pp. 76-83 79 (a) side view of region of interest. (b) four view segmentation. fig. 5 region of interest and image segmentation (a) the front metal connection leg absence defect (b) the back metal connection leg absence defect fig. 6 leg absence defect detection using contour finding techniques the flow chart of top edge inclusion detection provides an intuitive explanation of the application logic. as the roi is determined from the orientation of the top edge and subsequent affine transformation, the top edge defect detection allows improper placement of the power supply gasket on the fixture. the roi is narrowed to account for the entire top edge and does not prevent errors that may arise when executing the inclusion finding a routine. the top edge material defects are observed by comparing the contour and convex hull representations of the top edge plastic material, as shown in fig. 7. fig. 7 the detected top edge material defects 3. deep learning approach the top view defect detection application is improved by adding images and type labels to provide a labeling framework for potential machine learning applications. therefore, quantitative information for more qualitative surface defects, namely inner ring notches and residual glue defects [17-18] were provided. to detect the surface type defects of the silicone rubber gaskets which have a rough and dim texture, a multi-exposure technique is used to enhance the illumination and highlight the defects. these multi-exposure images are then included in the dataset and trained in a 50-layer resnet network [19-20]. the pixel size range of the defects is limited to avoid the feature vanishing during the convolution operations of the network and to front back advances in technology innovation, vol. 5, no. 2, 2020, pp. 76-83 80 allow for more accurate discernment between the background and the defects. moreover, the deep learning approach can indicate the defect types and defect positions within the predicted image. 3.1. deep learning network structure fig. 8 the employed deep learning framework table 1 resnet-50 structure layer name output size resnet 50-layer inputs 224 × 224 × 3 conv1 112 × 112 × 64 7 × 7, 64, stride 2 pool1 56 × 56 × 64 3 × 3 max pool, stride 2 conv2.x 28 × 28 × 256 1 1, 64 3 3, 64 1 1, 256             × 3, stride 2 conv3.x 14 × 14 × 512 1 1, 128 3 3, 128 1 1, 512             × 4, stride 2 conv4.x 7 × 7 × 1024 1 1, 256 3 3, 256 1 1, 1024             × 6, stride 2 conv5.x 7 × 7 × 2048 1 1, 512 3 3, 512 1 1, 2048             × 3, stride 1 pool5 1 × 2048 reduce mean conv6 1 × 2 1 × 1, 1000, stride 1 table 2 rmpnet-50 structure layer name output size resnet 50-layer inputs 112×112× 3 conv1 56 × 56 × 64 7 × 7, 64, stride 2 pool1 56 × 56 × 64 3 × 3 max pool, stride 2 conv2.x 28 × 28 × 256 1 1, 64 3 3, 64 1 1, 256             × 3, stride 2 conv3.x 14 × 14 × 512 1 1, 128 3 3, 128 1 1, 512             × 4, stride 2 conv4.x 7 × 7 × 1024 1 1, 256 3 3, 256 1 1, 1024             × 6, stride 2 conv5.x 7 × 7 × 2048 1 1, 512 3 3, 512 1 1, 2048             × 3, stride 1 pool5 1 × 2048 reduce mean conv6 1 × 6 1 × 1, 6, stride 1 resnet is chosen as the deep residual network because its residual architecture effectively overcomes the vanishing gradient and explosion issues. as the resnet in tensorflow pre-training models and related high-level libraries are available in the open-source community, many comparative data for resnet are available for industrial use. the network proposed in this paper uses a multi-class classifier constructed with resnet 50-layer architecture. part of the resnet structure is modified by reducing the size of the receptive field, and the size of the feature map increased as shown in tables (1)-(2), respectively. the proposed pmpnet network can be seen in fig. 8. the size of the input picture of our proposed network was 112 × 112 pixels, and there is no max-pooling layer. advances in technology innovation, vol. 5, no. 2, 2020, pp. 76-83 81 fig. 9 the labelled defect types for deep learning 3.2. training and dataset for deep learning network the image of the surface defect and its annotation mask are labeled using the patch-based method. fig. 9 shows the original image and the labeled defect types for deep learning. this work uses the tensorflow open-source software library to develop a deep learning network and the corresponding training classifier. in addition to distinguishing the defects, this proposed method can also identify non-defective background images to increase detection accuracy. the training process is divided into two steps. the first step used the pre-trained model of resnet v1 50-layers for transfer learning. it only trained for five epochs and updated the parameters of the last layer of the convolutional neural network layer. the parameters of the remaining network layers are fixed, the batch size was set to 64, the learning rate is set to 0.001, and the weight decay is set to 0.0004. the second step trained for 50 epochs and updated all network layer parameters, the batch size was set to 64, the learning rate is set to 0.0001, and the weight decay is set to 0.0004. the normalized exponential function is used as the basis of confidence for classifying in the last layer of the network. it is an extension of the sigmoid function and can be termed as the softmax function: 0 j j h j c h j softmax e e    (1) furthermore, mean square error (mse) is used to evaluate the loss function. mse is expressed as: 2 1 1 ˆ( ) n i i i mse n y y    (2) in this study, a common data augmentation method such as mirroring and up and down flipping is used for image preprocessing. 3.3. defect prediction for deep learning network the cutting principle of defect detection is the same as the labeling dataset strategy, although the method and parameters were slightly different. the original large image is cut into several 112 × 112 segment patches in the stride of 64 and the trained defect classifier is used to predict the patches. the fused feature map of the defect category is normalized into a gray-scale map. as deep learning method requires a large amount of data to achieve accurate and efficient results. each silicon rubber gaskets advances in technology innovation, vol. 5, no. 2, 2020, pp. 76-83 82 is rotated by a mechanical turntable to acquire more image data from different lighting angles. the feature map corresponding to different colors according to different types of detected defects is shown in fig. 10. during the prediction stage, the accuracies of the notch and residual glue defects were 97% and 80%, respectively. after rotating the objects under inspection at different angles, a series of images are obtained from different angles for the rubber gasket under evaluation. the collected images are analyzed for deep learning and the classification is chosen with voting. after voting, the prediction accuracies of the notch and residual glue defects were 100% and 97%, respectively. compare to the traditional rule-based image detection approach, the deep learning approach has high classification performance. it can detect non-contour defects (i.e., notches and residual glue) and achieve valid prediction results. fig. 10 the fused feature map after deep learning prediction 4. conclusions in this paper, defect detection for mobile device power supply connector gaskets using rule-based and deep learning image processing techniques is presented. using a rules-based approach, outer ring inclusion and metal contact leg absence defects with 100% accuracy are successfully identified. the inner ring surface notch and residual glue defect detection applications are developed using deep learning approaches. using the pmpnet framework, the overfitting phenomenon is avoided during all the proposed training steps, and the prediction accuracies of 100% and 97% are reached for the notch and residual glue defects, respectively. the high accuracy online inspection system for silicon rubber gaskets is implemented by fusing two different strategies successfully. due to misjudgments caused by the instability of the light source, both approaches are desirable to help determine the stability of the light source processing before the light source correction mechanism. however, these two approaches require manual input to help to evaluate the value of the result. further studies can be investigated to acquire feedback during the training process which will allow computers to make the decisions. acknowledgments the authors would like to thank all of those reviewers who have provided helpful suggestions. we would also like to thank pioneer material precision technology for providing all the experiment samples. conflicts of interest the authors declare no conflict of interest. references [1] a. cornforth and c. james, “ultrasonic system for the inspection of rubber gaskets,” ndt international, vol. 13, no. 1, pp. 15-17, february 1980. advances in technology innovation, vol. 5, no. 2, 2020, pp. 76-83 83 [2] c. c. kuo and y. t. siao, “measuring the solidification time of silicone rubber using optical inspection technology,” optik-international journal for light and electron optics, vol. 125, no. 1, pp. 196-199, january 2014. [3] c. c. kuo and y. r. chen, “rapid optical inspection of bubbles in the silicone rubber,” international journal for light and electron optics, vol. 124, no. 13, pp.1480-1485, july 2013. [4] r. ciobanu, d. rizescu, and c. rizescu ciprian, “automatic sorting machine based on vision inspection,” international journal of modeling and optimization, vol. 7, no. 5, pp. 286-290, october 2017. [5] j. a. marvel, r. bostelman, and j. falco, “multi-robot assembly strategies and metrics,” acm computing surveys, vol. 51, no. 1, pp. 1-32, april 2018. [6] y. x. yang, y. t. lou, m. y. gao, and g. j. ma, “an automatic aperture detection system for led cup based on machine vision,” multimedia tools and applications, vol. 77, no. 18, pp. 23227-23244, september 2018. [7] c. c. ho and t. r. tsai, “machine-vision-based servo control of a robotic sewing system,” advanced science letters, vol. 9, no. 1, pp. 45-49, april 2012. [8] c. c. ho, j. c. li, and t. h. kuo, “multicamera fusion-based leather defects marking system,” advances in mechanical engineering, vol. 5, 347921, january 2013. [9] c. c. ho, y. m. chen, and t. y. chi, “machine vision-based automatic placement system for solenoid housing,” key engineering materials, vol. 649, pp. 9-13, march 2015. [10] c. c. ho, y. m. chen, and p. c. li., “machine vision based in-process led chip mounting system,” measurement and control, vol. 51, pp. 1-11, july 2018. [11] c. c. ho and j. j. liu, “development of auto defect inspection system for cell phone silicone rubber gasket,” proc. tenth international symposium on precision engineering measurements and instrumentation (ispemi 2018), spie press, march 2019, pp. 1105303. [12] y. j. cha, w. choi, and o. büyüköztürk, “autonomous structural visual inspection using region‐ based deep learning for detecting multiple damage types,” computer‐ aided civil and infrastructure engineering, vol. 33, no. 9, pp. 731-747, november 2017. [13] r. ren, t. hung, and k.c. tan, “a generic deep-learning-based approach for automated surface inspection,” ieee transactions on cybernetics, vol. 48, no. 3, pp. 929-940, march 2018 [14] y. j. cha, w. choi, and o. büyüköztürk, “deep learning‐ based crack damage detection using convolutional neural networks,” computer‐ aided civil and infrastructure engineering, vol. 32, no. 5, pp. 361-378, march 2017. [15] g. s. babu, p. l. zhao, and x. l. li, “deep convolutional neural network based regression approach for estimation of remaining useful life,” proc. international conference on database systems for advanced applications (dasfaa), march 2017, pp. 214-228. [16] r. ren, t. hung, k.c. tan, “a generic deep-learning-based approach for automated surface inspection,” ieee transactions on cybernetics, vol. 48, no. 3, pp. 929-940, march 2018. [17] g. psuj, “multi-sensor data integration using deep learning for characterization of defects in steel elements,” sensors, vol. 18, no. 1, january 2018. [18] x. c. yang, h. li , y. t. yu, x. c. luo, t. huang, and x. yang, “automatic pixel‐ level crack detection and measurement using fully convolutional network,” computer‐ aided civil and infrastructure engineering, vol. 33, no. 12, pp. 1090-1109, december 2018. [19] t. wang, y. chen, m. qiao, and h. snoussi, “a fast and robust convolutional neural network-based defect detection model in product quality control,” the international journal of advanced manufacturing technology, vol. 94, no. 9, pp. 3465-3471, february 2018. [20] r. f. wei and y. b. bi, “research on recognition technology of aluminum profile surface defects based on deep learning,” materials, vol. 12, no. 10, may 2019. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation, vol. 5, no. 1, 2020, pp. 33-44 the design of mold clamping mechanisms with two dead-position configurations tzu-hsia chen 1,* , long-chang hsieh 2 1 department of mechanical and computer-aided engineering, feng chia university, taiwan 2 department of power mechanical engineering, national formosa university, taiwan received 28 april 2019; received in revised form 08 june 2019; accepted 11 july 2019 doi: https://doi.org/10.46604/aiti.2020.4164 abstract blow-molding machines are divided into two types: (1) rotary type blow-molding machine, and (2) linear type blow-molding machine. good kinematic characteristics of mold clamping mechanism is the goal pursued by manufacturers of blow-molding machine. traditionally, the mold clamping mechanism has only one dead-position configuration for the mold in the closed position. when the mold is in the open position, the mold clamping mechanism does not have dead-position configuration. this study focuses on the design of mold clamping mechanisms with two dead-position configurations when the mold is in open and closed positions. firstly, with reference to the existing patents, based on the yan's creative design methodology, four design concepts of mold clamping mechanisms are synthesized. then, according to the studies on kinematics of mechanisms, three mold clamping mechanisms with two dead-position configurations are also synthesized. the results of this paper will enhance the research and development (r&d) capability of the mold clamping mechanisms to improve industrial competitiveness. keywords: blow-molding machine, conceptual design, creative design, dead-position configuration, mold clamping mechanism 1. introduction due to the reason that the usage of polyethylene terephthalate (pet) bottles is more and more widely, how to increase the production of pet bottles has become a major issue for the manufacturers. the requirements of the blow-molding machine will be stricter than ever. in order to improve the stability of the production of pet bottles, reducing the vibration and noise during the production process, extending the life of the machine, it is necessary to analyze the kinematics of mold clamping mechanism. good kinematic characteristics of mold clamping mechanism is the goal pursued by manufacturers. blow-molding machines are divided into two types: (1) rotary type blow-molding machine as shown in fig. 1 (a), and (2) linear type blow-molding machine as shown in fig. 1 (b). the motion of linear blow-molding machine is linear and reciprocated. the mold clamping mechanisms for linear type blow-molding machine are simple, stable, and reliable. figs. 2 (a) and 2 (b) show the linear type blow-molding machine and its mold clamping mechanism. traditionally, the mold clamping mechanism of linear type blow-molding machine has only one dead-position configuration for the mold in the closed position. when the mold is in the open position, the mold clamping mechanism does not have dead-position configuration. it results poor kinematic characteristics (larger or infinite acceleration), and causes blow-molding machine unstable. for the mold clamping mechanisms with two dead-position configurations in open and closed positions, it will have better kinematic characteristics (smaller max. acceleration), and causes blow-molding machine more stable. * corresponding author. e-mail address: summer34134@gmail.com tel.: +886-4-2451-7250 ext.3501 advances in technology innovation, vol. 5, no. 1, 2020, pp. 33-44 34 (a) rotary type blow-molding machine (b) linear type blow-molding machine fig. 1 types of blow-molding machines (a) linear blow-molding machine (b) mold clamping mechanism fig. 2 linear blow-molding machine this study focuses on the design of mold clamping mechanisms with two dead-position configurations for linear blow-molding machine. firstly, by referring to the existing patents [1-7] and the studies on the creative design [8-13], four feasible design concepts of mold clamping mechanisms are synthesized. then, according to the studies on kinematics of mechanisms [14-20], three mold clamping mechanisms with two dead-position configurations are also synthesized. when the advances in technology innovation, vol. 5, no. 1, 2020, pp. 33-44 35 mold is in open and closed positions, the mold clamping mechanism will stay in the state of dead-position configuration (meanwhile, the mechanism is converted to desired structure). hence, the mold clamping mechanisms with two dead-position configurations have the advantage of more safety. the results of this research can be used as reference when designing the real mold clamping mechanism and enhance r&d capability of the mold clamping mechanisms to improve industrial competitiveness. 2. existing design (a) clamping mechanism (b) kinematic skeleton fig. 3 the mode clamping mechanism of servo-drive type injection molding machine [1] (a) opening/closing mechanism (b) kinematic skeleton fig. 4 driving device of controlling mode opening and closing for plastic molding machine [2] advances in technology innovation, vol. 5, no. 1, 2020, pp. 33-44 36 lots of researches have been done about the topological characteristics for mold clamping mechanisms. fig. 3 [1] and fig. 4 [2] show planar mechanisms with 6 links and 7 joints. for the mold clamping mechanism shown in figs. 3 and 4, the mold clamping mechanism has a dead-position configuration when the mold is in the closed position. however, when the mold is in the open position, the mold clamping mechanism does not have dead-position configuration. the purpose of this paper is to design the mold clamping mechanism having dead-position configurations when the mold is in the open and closed positions. the mobility of planar mechanism can be obtained by following equation. f=3(n-j-1)+σfi (1) where n is number of links, j is number of joints, and fi is the degrees of freedom of joint i. the mechanisms, shown in figs. 3 and 4, have 6 links, 7 joints (5 revolute pairs and 2 prismatic pair), according to equation of mobility, we get: f=3(n-j-1)+σfi =6(6-7-1)+7=+1 (2) the topological characteristics of mode clamping mechanisms for linear blow-molding machine are concluded as follows: (1) it consists of 6 members and 7 joints. (2) it has ground link (gr, member 1), input slider (sin, member 2), first connecting rod (member 3), rocker (member 4), second connecting rod (member 5), and output slider (sout, member 6). (3) it has 7 joints including 5 revolute joints (jr) and 2 prismatic joints (jp). (4) it is planar mechanism with 1 degree of freedom. according to above reasoning, fig. 5 shows the corresponding generalized chain of the patent [2] shown in fig. 4. fig. 5 generalized chain of the mode clamping mechanism [2] 3. conceptual design in 1998, professor yan proposed a creative design methodology [8], shown in fig. 6, to create all feasible design concepts for mechanical devices. the steps are: step 1: identify the topological characteristics based on the existing designs with required design specifications. step 2: select any existing design and transform it into its corresponding generalized chain. step 3: synthesize all generalized chains that have the same numbers of links and joints as the generalized chain obtained in step 2. step 4: assign types of links and joints to each generalized chain obtained in step 3, to have the atlas of feasible specialized chains based on the algorithm of specialization to meet the design requirements and constraints. step 5: particularize each feasible specialized chain obtained from step 4 to its corresponding schematic format of mechanical device, to have the atlas of mechanical devices. advances in technology innovation, vol. 5, no. 1, 2020, pp. 33-44 37 fig. 6 yan’s creative design methodology [8] 3.1 atlas of kinematic chain according to the process of number synthesis, 2 kinematic chains with 6 links and 7 joints, shown in fig. 7, are synthesized. (a) 6-1 (b) 6-2 fig. 7 atlas of (6, 7) kinematic chains 3.2 design requirements and constrains based on the concluded topological characteristics of existing designs, the design requirements and constrains of the mode clamping mechanisms are concluded as follow: (1) there must be a ground link (gr), input slider (sin,), output slider (sout). (2) the ground link as the frame must be ternary link. (3) the ground link (gr) must be adjacent to input slider (sin) and output slider (opening/closing mold, sout) with prismatic pair. (4) input slider (sin) and output slider (sout) can’t be in the same circuit with 1 degree of freedom. (5) the mode clamping mechanism only includes prismatic pair (jp) and revolute pair (jr). (6) the mode clamping mechanism must have 2 prismatic pairs. 3.3 specialization subject to certain design requirements and constraints, the specializing steps for mode clamping mechanism are concluded as follow: existing designs generalization topological characteristics generalized chains number synthesis atlas of kinematic chains specialization design requirement and constraints atlas of feasible specialized chains particularization atlas of designs advances in technology innovation, vol. 5, no. 1, 2020, pp. 33-44 38 (1) for each generalized (kinematic) chain, identify the ground link (gr) for all possible cases. (2) for each case obtained in step 1, identify input slider (sin, denoted by thick black line ) and output slider (sout, denoted by thick black line ). (3) for each case obtained in step 3, identify the corresponding revolute pair (denoted by ○) and prismatic pair (denoted by ●). based on the design requirements and constraints, 3 feasible specialization chains, shown in figs. 8(a), 8(b), and 8(c), are generated by the process of specialization. (a) 6-1-1 (b) 6-2-1 (c) 6-2-2 fig. 8 atlas of (6, 7) feasible specialized chains 3.4 particularization for each feasible specialized chain, it can be particularized into its corresponding kinematic skeleton by the reverse process of generalization. for the feasible specialized chains shown in figs. 8 (a) ~8 (c), their corresponding kinematic skeletons of mode clamping mechanisms are shown in figs. 9 (a)~9 (c). according to the kinematic skeletons shown in figs. 9 (a) ~9 (c), one possible multiple-joint mechanism is synthesized and shown in fig. 9 (d). (a) mechanisms (i) (b) mechanisms (ii) (c) mechanisms (iii) (d) mechanisms (iv) fig. 9 atlas of feasible mold clamping mechanisms of (6, 7) generalized chain 4. clamping mechanisms with two dead-position configurations 4.1 case i fig. 10 (a) and 10 (b) show the mode clamping mechanism (i) and its corresponding vector coordinate system of the mechanism shown in fig. 9 (a). according to fig. 10 (b), the mechanism has two independent vector loops (vector loops 1 and 2). vector loops 1 and 2 can be expressed as follows: (3) 2 3 4 0ar r r   advances in technology innovation, vol. 5, no. 1, 2020, pp. 33-44 39 (4) according to fig. 10 (b), θ4a＝θ4b+α and θ4c＝θ4a-π+β are the given conditions. if we substitute them into the above equations, then the components x and y can be expressed as: (5) (6) (7) (8) (a) kinematic skeleton (b) vector coordinate system fig. 10 mold clamping mechanism (i) (a) closed position (dead-position configuration, line l1) (b) open position (dead-position configuration, line l2) fig. 11 mold clamping mechanism (i) with two dead-position configurations if r2y＝153.5mm, r3＝176mm, r4a＝190mm, r4b＝r5＝250mm, α=30⁰, then r4c＝127.78mm and β=101.97⁰. according to eqs. (5) ~ (8), we get: (1) if θ4b＝0⁰ (θ4a＝30⁰), then θ3＝-19.41⁰, r2x＝-1.45mm, θ4c＝θ4a-π+β＝-48.03⁰, θ5＝0⁰, and r6＝500mm. fig. 11 (a) shows the corresponding dead-position configuration (line l1) of mode clamping mechanism at closed position. (2) if θ4b＝60⁰ (θ4a＝90⁰), then θ3＝11.97⁰, then r2x＝-171.22mm, θ4c＝θ4a-π+β＝11.97⁰, θ5＝-60⁰, and r6＝250mm. fig. 11(b) shows the corresponding dead-position configuration (line l2) of mode clamping mechanism at open position. according to figs. 11 (a) and 11 (b), the complete mechanism of mode clamping mechanism (i) with two dead-position configurations are shown in figs. 12 (a) and 12 (b). 4 5 6 0 b r r r   2 3 3 4 4 cos cos 0 x a a r r r    2 3 3 4 4 sin sin 0 y a a r r r    4 4 5 5 6 cos cos 0 b b r r r    4 4 5 5 sin sin 0 b b r r   advances in technology innovation, vol. 5, no. 1, 2020, pp. 33-44 40 (a) closed position (dead-position configuration, line l1) (b) open position (dead-position configuration, line l2) fig. 12 complete mechanism of mold clamping mechanism (i) with two dead-position configurations 4.2 case ii fig. 13 (a) and 13 (b) show the mode clamping mechanism and its corresponding vector coordinate system of the mechanism shown in fig. 9 (b). according to fig. 13 (b), the mechanism has two independent vector loops (vector loops 1 and 2). vector loops 1 and 2 can be expressed as follows: (a) kinematic skeleton (b) vctor coordinate system fig. 13 mold clamping mechanism (ii) (9) (10) according to fig. 13 (b), θ3c＝θ3a+π-β are the given conditions. if we substitute them into the above equations, then the components x and y can be expressed as: (11) (12) (13) 2 3 4 0 a r r r   4 3 5 6 0 c r r r r    2 3 3 4 4 cos cos 0 x a a r r r    2 3 3 4 4 sin sin 0 y a a r r r    4 4 3 3 5 5 6 cos cos cos 0 c c r r r r      advances in technology innovation, vol. 5, no. 1, 2020, pp. 33-44 41 (14) if r2y＝162.8mm, r4＝94mm, r3a＝162.8mm, r3c＝156mm, r5＝250mm, and β=90⁰, according to eqs. (11) ~ (14), we get: (1) if θ4＝0⁰, then θ3a＝-90⁰, r2x＝94mm, θ3c＝θ3a+π-β＝0⁰, θ5＝0⁰, and r6＝500mm. fig. 14 (a) shows the corresponding dead-position configuration of mode clamping mechanism at closed position. (2) if θ4＝60⁰, then θ3a＝-30⁰, r2x＝-93.99mm, θ3c＝θ3a+π-β＝60⁰, θ5＝-60⁰, and r6＝250mm. fig. 14 (b) shows the corresponding dead-position configuration of mode clamping mechanism at open position. according to figs. 14 (a) and 14 (b), the complete mechanism of mode clamping mechanism (ii) with two dead-position configurations are shown in figs. 15 (a) and 15 (b). (a) close position (dead-position configuration, line l1) (b) open position (dead-position configuration, line l2) fig. 14 mold clamping mechanism (ii) with two dead-position configurations (a) closed position (dead-position configuration, line l1) (b) open position (dead-position configuration, line l2) fig. 15 complete mechanism of mold clamping mechanism (ii) with two dead-position configurations 4.3 case iii fig. 16 (a) and 16 (b) show the mode clamping mechanism and its corresponding vector coordinate system of the mechanism shown in fig. 9 (c). according to fig. 16 (b), the mechanism has two independent vector loops (vector loops 1 and 2). vector loops 1 and 2 can be expressed as follows: 4 4 3 3 5 5 sin sin sin 0 c c r r r     advances in technology innovation, vol. 5, no. 1, 2020, pp. 33-44 42 (a) kinematic skeleton (b) vector coordinate system fig. 16 mold clamping mechanism (iii) (a) close position (dead-position configuration, line l1) (b) open position (dead-position configuration, line l2) fig. 17 mold clamping mechanism (iii) with two dead-position configurations (15) (16) according to fig. 16(b), θ4b＝θ4a-π+β are the given conditions. if we substitute them into the above equations, then the components x and y can be expressed as: (17) (18) (19) (20) if r2y＝120mm, r3＝120mm, r4a＝250mm, r4b＝120mm, r5＝250mm, and β=90⁰, according to eqs. (17) ~ (20), we get: (1) if θ5＝0⁰, then θ4a＝0⁰, r2x＝380mm, θ4b＝θ4a-π+β＝-90⁰, θ3＝0⁰, and r6＝500mm. fig. 17(a) shows the corresponding dead-position configuration of mode clamping mechanism at closed position. 5 4 6 0 a r r r   2 3 4 6 0 b r r r r    5 5 4 4 6 cos cos 0 a a r r r    5 5 4 4 sin sin 0 a a r r   2 3 3 4 4 6 cos cos 0 x b b r r r r     2 3 3 4 4 sin sin 0 y b b r r r    advances in technology innovation, vol. 5, no. 1, 2020, pp. 33-44 43 (2) if θ5＝-60⁰, then θ4a＝60⁰, r2x＝42.15mm, θ4b＝θ4a-π+β＝-30⁰, θ3＝-30⁰, and r6＝250mm. fig. 17(b) shows the corresponding dead-position configuration of mode clamping mechanism at open position. according to figs. 17(a) and 17(b), the complete mechanism of mode clamping mechanism (iii) with two dead-position configurations are shown in figs. 18(a) and 18(b). according to fig. 18(b), the mold is damaged because of too large stroke of the input slider, so it is not a good design. (a) closed position (dead-position, line l1) (b) open position (dead-position, line l2) fig. 18 complete mechanism of mold clamping mechanism (iii) with two dead-position configurations 5. conclusions traditionally, the mold clamping mechanism does not have dead-position configuration for the mold in the open position. this study focuses on the design of mold clamping mechanisms with two dead-position configurations when the mold is in open and closed positions. firstly, by referring to the existing patents and the studies on the creative design, four feasible design concepts of mold clamping mechanisms are synthesized. then, according to the studies on kinematics of mechanisms, three mold clamping mechanisms with two two dead-position configurations are also synthesized. because of too large stroke of input slider, the design case iii is not good. only two mold clamping mechanisms with two two dead-position configurations (case i and case ii) can be used to design the real mold clamping mechanism for linear blow-molding machine. the results of this research will enhance r&d capability of the mold clamping mechanisms to improve industrial competitiveness. acknowledgement the authors are grateful to chum-power machinery co., ltd. for the support of this research under 107af094. conflicts of interest the authors declare no conflict of interest. references [1] g. x. lai, w. f. su and g. x. lin, the mode clamping mechanism of servo-drive type injection molding machine, taiwan patent, 542097, jul. 11, 2003. [2] f. q. jiang, driving device of controlling mode opening and closing for plastic molding machine, taiwan patent, m257273, feb. 21, 2005. [3] w. t. chang and l. y. wu, ten-link clamping mechanism with double dead-positions, taiwan patent, i382913, jan. 21, 2013. advances in technology innovation, vol. 5, no. 1, 2020, pp. 33-44 44 [4] j. li and g.w. li, all-electric plastic injection machine, taiwan patent, m510231, oct. 11, 2015. [5] w. y. lin, y. c. chang, z. cheng, x. r. chen, and h. y. chen, new fast five-point locking mechanism with two dead-position configurations, taiwan patent, m510233, oct. 11, 2015. [6] y. y chen and j. c. qin, clamping device for mold, taiwan patent, i537121, jun. 11, 2016. [7] w. y. lin, y. c. chang, and y. j. xiao, new five-point clamping mechanism with two dead-position configurations for injection molding machine, taiwan patent, m541943, mar. 21, 2017. [8] h. s. yan, “a methodology for creative mechanism design,” mechanism and machine theory, vol. 27, no. 3, pp. 235-242, may. 1992. [9] h. s. yan, creative design of mechanical devices, springer-verlag, singapore, 1998. [10] w. h. hsieh and s. j. chen, “innovative design of cam-controlled planetary gear trains,” international journal of engineering and technology innovation, vol. 1, no. 1, pp. 01-11, 2011. [11] l. c. hsieh and t. h. chen, “the systematic design of link-type optical fiber polisher with single flat,” journal of advanced science letters, vol. 9, pp. 318-324, 2012. [12] g. s. hwang and c. c. lin, “innovative design of a transmission mechanism for variable speed wind turbines,” jsme journal of advanced mechanical design, systems, and manufacturing, vol.9, no.3, pp. 1-11, 2015. [13] t. h. chen and l.c. hsieh, “the innovative design of the massage mechanisms for massage chair,” advances in technology innovation, vol. 4, no. 2, pp. 116-124, 2019. [14] z. f. yuan, m. j. gilmartin, and s. s. douglas, “optimal mechanism design for path generation and motions with reduced harmonic content,” asme transactions, journal of mechanical design, vol. 126, no. 1, pp. 191-196, 2004. [15] w. h. hsieh, “kinematic synthesis of cam-controlled planetary gear trains,” mechanism and machine theory, vol. 44, no. 5, pp. 873–895, 2009. [16] h. li and y. zhang, “seven-bar mechanical press with hybrid-driven mechanism for deep drawing; part 1: kinematic analysis and optimum design,” journal of mechanical science and technology, vol. 24, no.11, pp.2513-2160, 2010. [17] w. h. hsieh and c. h. tsai, “on a novel press system with six links for precision deep drawing,” mechanism and machine theory, vol. 46, no. 2, pp. 239–252, 2011. [18] l. c. hsieh and t. h. chen, “the systematic design of link-type optical fiber polisher with single flat,” journal of advanced science letters, vol. 9, pp. 318-324, 2012. [19] l. c. hsieh and t. h. chen, “the systematic design of planetary-type grinding devices for optical fiber ferrules and wafers,” transactions of the canadian society for mechanical engineering, vol. 40, no. 4, pp. 619-630, 2016. [20] f. c. chen, y. f. tzeng, and w. r. chen, “on the mechanism design of an innovative elliptical exerciser with quick-return effect,” international journal of engineering and technology innovation, vol. 8, no. 3, pp. 228-239, 2018. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation, vol. 3, no. 1, 2018, pp. 26 35 changes of e-kers rules to make f1 more relevant to road cars albert boretti 1, 2,* 1 department of mechanical and aerospace engineering, benja min m. statler college of engineering and mineral resources, west virginia university, morgantown, usa . 2 military technological college, muscat 111, oman . received 27 oct ober 2016; received in revised form 21 march 2017; accept ed 05 april 2017 abstract today’s f1 hybrid cars are based on very similar power units made up of about the same internal co mbustion engine (ice) and energy recov ery system (ers). because of restrictive design rules permitting too much fuel per race, the internal co mbustion engine is not part icularly fuel e ffic ient. the methodology is based on lap time simu lations and telemetry data for a f1 h car covering one lap of the monaco grand prix. the methodology is based on lap time simu lations and tele metry data for a f1 h car covering one lap of th e monaco grand prix. the p resent limit o f 100 kg of fuel per race is e xcessive. the low power energy recovery system is used strategically rather than fuel savings recovering very little bra king energy. the 4 mj of storable energy is used only when it is s trategically needed. the 2 mj of recoverable energy a llo wed per lap a re a lmost never collected. to return to be technica lly attractive, f1 should permit much more freedom in the definit ion of the ice and the ers. as the goal of the ru les should be lowering the fuel consumption wh ile keeping technical and sporting interest high, the best solution is more freedom to achieve the fastest car within mo re stringent limits of fue l economy. a rea l limit to the total fuel consumption for a race track like monte carlo should be not mo re than 80 kg o f fuel. this would translate in more energy recovery to the ers per lap and better fue l e fficiency of the ice and will ce rtainly help mo re the design of passenger cars . keywor ds : hybrid cars, kinetic energy recovery systems, motorsport, f1, le mans 1. introduction today f1 racing cars, similarly to other popular racing car series, are hybrid cars as the environmental concern must be at least apparently the driving force for every sport activity. the power unit now comprises an internal combustion engine and an energy recovery system. the energy recovery system in theory should recover the waste energy, mostly kinetic, to reduce the fuel energy supply to the internal co mbustion engine. bac kground informat ion on kinetic en ergy recovery systems for rac ing cars can be found in [1], wh ile the specific internal co mbustion engines and kinetic energy recovery systems (kers) fo r 201 4 f1 cars are discussed in [2-4]. the fuel saving goal is however practically eluded by the most part of the f1 tea ms. the kinetic energy recovery system is indeed mostly used strategically to boost the performance of a car, being , otherwise, the internal combustion engine not that powerful as it was in the recent past, by discharging the energy storage in selected lap and not certainly charging and discharging the energy storage every single lap . the lack of freedo m given to engineers to develop a technical solution delivering a target fuel economy is the reason why f1 is not technically challenging in a way that may be benefic ia l to road transport while being attractive to the motor enthusiasts. purpose of the manuscript is to suggest changes of the technical regulations that could improve the energy * corresponding author. e-mail address: a.a.boretti@gmail.com advances in technology innovation, vol. 3, no. 1, 2018, pp. 26 35 27 cop y right © taeti recovery and the fuel conversion effic iency of the internal co mbustion engine for a better fue l economy while permitting top-class motorsport performances. it is shown how a mo re restrictive total fuel usage per race and mo re freedo m to develop the internal combustion engine and the energy recovery system may benefit the interest towards the racing event and the value of the technical development for production cars . 2. lmp1-h vs. f1 hybrid 2015 the latest 2015 le mans race was dominated by the porsches and the audis battling until the very end of the race, wi th four class 1 le mans prototype hybrid (lmp1-h) car manufacturers, the two ge rman plus the japanese toyota and nissan, proposing for the event very d ifferent technical solutions for what concerns the internal co mbustion engine (ice), diesel or gasoline, d ifferent displace ment, turbo or naturally aspirated, and the energy recovery system (ers), e lectric or mechanic or electro-mechanic, o f different powers and different energy storage. the rules set different limitations to the fuel flo w rate per diffe rent ers energy storage limits and fuel selection. these rules give the lm p1 -h engineers the freedo m to develop diffe rent ices and different erss. the developments of alternative solutions are ultimately what are needed to ma ke the race event attractive to the motor enthusiast and be relevant, in long term perspective, to the design of every day passenger cars. the same technical enthusiasm does not certainly apply to todays’ f1 hybrids. ref. [4] provides an assessment of the differe nt kers options available in class 1 le mans proto-type hybrid (lmp1-h). fig. 1 f1 kers (es+mgu-k) and e-boost/whr (es+mgu-h) from [5] todays’ f1 hybrids have power units where simila rly to lmp1-h the ice is one ele ment of the power unit that also includes an ers. table 1 reca lls the present specifications of the ice and the ers. the ers inc ludes two motor generator units (mgu) linked to an energy store (es) recharged by the braking work and eventually the waste heat. the four power unit components, plus the ice and the turbocharger are the six separate e le ments making a power unit. four of each e le ment are available to each driver per season without incurring in a grid penalty. during a race, drivers may use steering whee l controls to switch to different power unit settings, or to change the rate of ers energy harvest. fig. 1 (fro m [5]) presents the f1 kers (es+mgu-k) and e-boost/whr (es+mgu-h). the set-up of the power unit of a f1 hybrid is therefore in princ iple not that far fro m the one of the porsche 919 h winner of the last le mans 2015 race. the d ifference is , however, substantial when the details are considered. f1 is much better fo r what concerns the turbocharger, as the mgu -h fitted to the turbocharger shaft and connected to the es gives wider opportunities than having just a power turbine downstream of the traditional turbocharger turbine to recover a minimal a mount of waste heat at the expenses of increased back flow. however, f1 hybrids are less flexible under all the other aspects. in f1, the turbocharged 1.6-litre v6 engines are pretty much the same fo r every tea m. not only sa me d isplacement and (about) same fuel, but also same number of cylinders, same 90-degree v angle, sa me rev limiter at 15,000 rp m, sa me four poppet valves per cylinder, same direct fuel in jection, same single fixed geometry turbocharged, same bore, same stroke, same crankcase height, almost everything the same to deliver about the same 600 hp or 447 kw of top bra ke powe r, with bra ke mean effective pressure and brake specific fuel consumption curves also expected to be very similar. advances in technology innovation, vol. 3, no. 1, 2018, pp. 26 35 28 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti the 600 hp or 447 kw are the values of peak powe r c laimed by the most part of the teams for the 2014 season. the fuel flow rate is limited to 100 kg/h, that considering a lowe r heating value for gasoline of 44.5 mj/kg translates in a ma ximu m f uel power of 1236 kw or 1658 hp, for a pea k powe r e ffic iency of the ice o f 600/ 1658 = 36.2%. per ru mors, a couple of tea ms outperforming the others this 2015 season could have moved around this limit pro mpting the federation to seek for re medy. if the fuel flow mete r is placed in a certain location of the fuel line, there is always the opportun ity to accumulate fuel downstream of the flow meter, and , therefore, en joy an instantaneous flow rate delivered by the high -pressure fuel injectors more than the instantaneous contemporary reading of the ambient pressure flow meter flow rate . table 1 data of power units of 2015 f1 hybrid cars internal combustion engine disp lacement 1.6 liters rev limit 15,000 rp m pressure charging sin gle turbochar ger, unlimited boost p ressure (but maximum 3.5 bar due to fuel flow limit) fuel flow limit 100 kg/h (but not at the injectors) permitted fuel quantity p er race 100 kg configuration 90° v6 number of cy linders 6 bore 80 mm stroke 53 mm crank height 90 mm number of valves 4 p er cy linder, 24 total exhausts single exhaust outlet, from turbine on car center line fuel direct fuel injection number of power units p ermitted p er driver p er y ear 5 energy recovery sy stems m gu-k rp m m ax 50,000 rp m m gu-k p ower m ax 120 kw energy recovered by m gu-k m ax 2 m j/lap energy released by m gu-k m ax 4 m j/lap m gu-h rp m unlimited energy recovered by m gu-h unlimited every percentage point incre ment o f the ice fuel conversion efficiency everything but difficult to achieve would t ranslate in an increase of the peak power of 12 kw or 17 hp. simila rly, any percentage in crease of the instantaneous flow rate to the injectors would translate in an incre ment of the instantaneous peak power of 4.5 kw or 6 hp. as an additional measure to temporarily increase the peak power, the m gu -h can drive the turbocharger compressor that, otherwise, only depends on the gas expansion through the turbine that is also translating in back pressure for the engine. this can make plausible the large r peak power outputs rumored for 2015. in addition to the ma ximu m fuel flow rate, belo w 10,500 rp m the fuel mass flow must not exceed q = 0.009 n +5.5, with n the engine speed in rpm and q in kg/h. in addition to the fuel flo w limiter placed along the fuel line, the 2014 and 2015 season have seen the introduction of a total fuel per race capped at 100 kg, or 4,450 mj of fuel energy per race, that is certainly a driver for much better fuel economies, but not certainly that strong. this fuel limit properly redefined may be the driver for a better product. fully integrated with the ice is the ers that increases the unit’s overall effic iency by recovering the waste energy fro m the brakes and the exhaust. the recovery of the e xhaust energy is , however, simply the turbocharger turbine that may de liver energy to the energy store that is not delivered to the turbocharger compressor. the ers accounts for an additional 120 kw or 160 hp to de liver about the same powe r output of the past 2.4 liters v8 engines naturally aspirated. the ers co mprises two motor generator units (m gu-k and m gu-h), plus the energy store. the motor generator units convert mechanical and heat energy to electrical energy and vice versa. the mgu -k converts the car kinetic energy generated under braking into electric ity while it acts as a motor under acceleration returning power to the drivetrain . the mgu-h converts the exhaust heat into advances in technology innovation, vol. 3, no. 1, 2018, pp. 26 35 29 cop y right © taeti electric ity but only through the turbocharger, i.e. only recovers the very s mall a mount of energy in the turb ine that would b e otherwise waste-gated when more than the compressor demands. the stored energy can be used to power the mgu -k. the mgu-h controls the speed of the turbo and the power fro m to and fro m the turbocharger shaft. the m gu -h may supply the e xtra energy needed at the compressor when the turbine energy is not enough, as for e xa mp le in low speed oper ating points or during accelerations, in this case ad-dressing the turbolag issues. it may a lso recover the e xtra energy available at the turbine otherwise waste-gated. the m gu-h may increase the power o f the ice at any speed by precise extra boost, opera ting in the best point of the map, with supply or withdraw of e xt ra energy. while the e -boost technology is certainly not new [10-15], the precise boost of the f1 m gu-h linked to the es of the kers may ce rtainly improve the overall powe r and energy management of the vehicle. apart fro m the same design of the m gu and es pure ly e lectric , a re the energy and powe r limits of the es and the m gu -k that makes a huge difference vs. the lmp1 -h ca rs. while lm p1-h cars have ma ximu m re leased energy of 2, 4, 6 or 8 mj/ lap and unlimited re leased power, in f1 a ma ximu m of 4 mj per lap can be transferred fro m the es to the m gu -k and then the drivetrain, but only a ma ximu m of 2 mj per lap can be transferred fro m the m gu-k to the es. more than that, the ma ximu m power of the m gu-k is limited to only 120 kw or 160 hp, wh ile the lmp1 -h a ll have powers of the m gu -k more than 185 kw up to a ma ximu m of 550 kw considered for the nissan gt -r lm nismo. the low power in addition to the limited energy is what makes the fuel saving kinetic energy recovery very difficult. as braking of f1 cars usually occurs with powers large ly e xceeding the propulsive power, up to about 2,000 kw the lo w power m gu-k must be recharged carefu lly, as this recharge t ranslates in a lap time penalty. in f1, the m gu-k is limited to recover 2 mj of energy per lap while the m gu -k may then supply a ma ximu m of 4 mj per lap to the drivetrain but the ma ximu m power in and out is limited to 120 kw (160 hp). th is means that the ers is more strategic rather than energy sav ing, as it can be certainly used to save or gain positions or improve the time of an indiv idual lap, wh ile it is still not convenient and not encouraged to recover and reuse the braking energy at any lap as it would be the case if the fuel energy saving wo uld be the real issue. table 2 summary table of 2015 lmp1-h and f1-h power and energy rules (a) limited to <300 kw in 2016. (b) no driver. (c) with driver lm p1-h (le m ans race track) f1-h no ers ers options released energy m j/lap 0 < 2 < 4 <6 < 8 <4 recovered energy m j/lap 0 <2 released power kw 0 unlimited (a) unlimited (a) unlimited (a) unlimited (a) 120 kw car m ass kg 850 (b) 870 (b) 870 (b) 870 (b) 870 (b) 702 (c) petrol energy m j/lap 150.8 146.3 141.7 137.2 134.9 m ax petrol flow kg/h 95.6 93 90.5 87.9 87.3 100 petrol cap acity carried on-board l 66.9 66.9 66.9 66.9 66.9 fuel technology factor 1.061 1.061 1.061 1.061 1.061 na k technology factor 1 0.983 0.983 0.983 1 diesel energy m j/lap 142.1 140.2 135.9 131.6 127.1 m ax diesel flow kg/h 83.4 83.3 81 78.3 76.2 diesel cap acity carried on-board l 54.8 54.8 54.8 54.8 54.8 table 2 presents a summary of the 2015 lmp1-h and f1-h powe r and energy rules. in case of one lap of the monaco grand prix, 3.337 km long, a f1 of curb weight 702 kg less the driver weight may use 1.28 kg or 57 mj of fuel energy, i.e . 17 mj/km. in case of one lap of the le mans race, the circuit de la sarthe is 13.629 km long; a lmp1-h of curb we ight 870 kg may only use 134.9 mj of fuel energy, i.e. 9.9 mj/km. even if it is not desirable for energy saving to switch on the mgu-k at end of straight for a small-time interval before the driver hits the frict ion brakes, this is what presently makes the largest contribution to the amount of energy available to t he es in f1. the overall lap t ime may also be faster with this strategic recharge because of the faster acceleration up to speed on the advances in technology innovation, vol. 3, no. 1, 2018, pp. 26 35 30 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti next straight may mo re than co mpensates for the lost time . this technique is not , however, in the direct ion of imp roving the fuel economy, as the direct path engine to wheels is much mo re effic ient than the path engine to mgu -k to es to m gu-k to wheels. the m gu-h is in theory uncapped, and an unlimited a mount of energy can be t ransferred between the m gu-h and the es and/or the mgu-k. the m gu-h technology is still far fro m being fu lly developed, but it is expected to help more in terms of ice output by precise boost rather than recovery of waste heat. the use of mgu-h and m gu-k and es may permit fu rther enhanced energy and power management. 3. energy analysis of a f1 2015 lap of monte carlo to understand the present status of energy recovery and fuel economy, tele metry and lap time simulat ions may help. the selected race track is monte carlo. the circuit de monaco is a street circuit of length 3.34 km. the total dis tance is 78 laps or 260.52 km. if we do consider the last 5 years of the monte ca rlo co mpetit ion, 3 with the naturally aspirated 2.4 liters v8 and a very small m gu-k of 60 kw, and the latest 2 with the turbocharged 1.6 liters v6 with the la rger but still sma ll present mgu-k of 120 kw, clea rly the latest f1 are much slo wer than their predecessors no matter the c laim of preserving the ma ximu m power output to preserve performances. in 2011, with sunny, fine and dry conditions, the best qualifying time for pole position was 1:13.556 while the fastest lap during the race was 1:16.234. in 2012, with warm and sunny conditions, about same fine conditions except the threat of showers at the end of the race, the best qualifying time fo r pole position was 1:14.381 while the fastest lap during the race was 1:17.296. in 2013, with sunny and dry conditions, the best qualifying time fo r pole position was 1:13.876 while the fastest lap during the race was 1:16.577. during the first season wit h the new rules, in 2014 with sunny and dry conditions, the best qualifying time for pole position was 1:15.989 while the fastest lap during the race was 1:18.479, roughly 2 seconds slower. finally, this year, 2015, with sunny and dry conditions, the best qualifying time fo r pole position was 1:15.098 wh ile the fastest lap during the race was 1:18.063. therefore, the new ru les have certainly slowed down the cars . some imp rovements have however been achieved in terms of fue l economy, even if the 100 kg of ma ximu m fuel pe r a race is not yet the driving force for the further development of the ice and the ers to drastically reduce the fuel consumption. tele metry and lap time simu lations may be used to compute the like ly performances of f1cars during one lap. the dynamic of rac ing cars and the equations governing the motion of the car a re proposed in [6]. the specific software used in t his paper, [7], is very simple but reliable. not having too much of supporting information as detailed dig itized tele met ry data and vehicle para meters, more co mplicated approaches as for e xa mple [8] only introduces additional difficulties to define the many other additional para meters involved in the simulat ion. the code [7] solves the newton’s equations of motion in the three directions for a point moving along a curved path. the minimu m we ight of the car including the driver but not the fuel was 690 kg in 2014 and it is 705 kg in 2015. the weight of the car is , therefore , ta ken here equal to 720 kg. for what concerns the aerodyn amic d rag, we appro ximate the aerodynamic drag fo rce as ½∙ρ·v 2· cd·a, where ρ is the air density, cd is the drag coeffic ient and a is the frontal car area, and the lift force as ½∙ρ·v2· cl·a where cl is the lift coefficient. we take ρ=1.29 kg/ m3, cd=0.85 a nd cl=2.4 when a=1.5 m2 for the specific very low speed circu it. as the drag force dra mat ically impact on the energy requested by a f1 ca r to cover a lap, the above far from accurate values certainly impact on the accuracy of the energy computation. the simulat ions require definition of fe w additional para mete rs, as the tires radius, the lift (downforce) coeffic ient, the rolling resistance and the longitudinal and lateral friction of tires, the gear ratios of the sequential gearbox, the final d rive ratio, the drive efficiency and a grip ratio, in addition to the specification of the engine power curve and obviously of the race t rack. advances in technology innovation, vol. 3, no. 1, 2018, pp. 26 35 31 cop y right © taeti while some of the latest lap time simu lation codes as [8] also account for lateral and longitudinal we ight transfer, real t ire effects as camber, slip rat io and slip angle, te mperature and pressures, vehicle yaw over-steering or under-steering and, finally banking and grade on the track, the code [7] does not. today’s most sophisticated lap time simu lation tools are fully int egrated with the vehicle manage ment and data acquisition systems. while these tools are very accurate, they also rely on the in -deep knowledge of the detailed vehicle operation that is proprietary data of only the tea ms. without this in -deep knowledge, they are only mo re co mp licate without being more accurate than [7]. fig. 2 lump mass model of a f1 car fig. 2 presents the lump mass model of a f1 car. the three-dimensional computation of the car aerodynamic with moving wheels and ground is proposed in [9]. the ae rodynamic simu lations return drag and lift coeffic ients to be used in the model. the car is modelled as a pa rtic le moving a long a curved path subject to propulsive and braking forces. th is simplified model permits a straightforward evaluation of the energy flow of the car covering one lap of a race track. the ne wton’s equations of motion are solved for the longitudinal, lateral and vertical directions . (a) velocity vs. distance of a f1 car (b) longitudinal acceleration vs. distance of a f1 car fig. 3 velocity and longitudinal acceleration vs. distance of a f1 car covering one qualifying lap at monte carlo fig. 3(a ) presents the velocity vs. distance from te le metry and fro m the simu lation, and fig. 3(b) presents the longitudinal acceleration. the tele metry informat ion was digitized fro m an image, and , therefore, suffers of poor resolution. lap time is 1:15:100. this optimum lap is covered by using the ice and the electric energy the simulat ion produces one velocity value every half a meter o f the race trac k, but the diffe rences in between the two traces are not only due to the different resolution, but also to the model not fully tunable by lack of knowledge of the vehi cle parameters and limited by the simp lified mathemat ics. the lap time is 1:15:100. this optimu m lap is covered by using 13 mj of ice fuel energy delivered with up to a power of 450 kw depending on engine speed plus 4 mj of electric energy delivered with up to a power of 120 kw . advances in technology innovation, vol. 3, no. 1, 2018, pp. 26 35 32 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti (a) propulsive vs. distance of a f1 car (b) bra king powers and propulsive, braking and recoverable energy vs. distance of a f1 car fig. 4 propulsive and bra king powers and propulsive, braking and recoverable energy vs. distance of a f1 car covering one qualifying lap at monte carlo figs. 4(a ) and 4(b) p resent the propulsive and braking powers and propulsive, bra king and recoverable energy vs. distance of a f1 ca r covering one lap at monte carlo as detailed in fig. 3. lap time is 1:15:100. this optimu m lap is covered by using the ice and the electric energy. the propulsive energy is 17.69 mj; the braking energy is 9.20 mj and the theoretically recoverable energy is 1.99 mj. the graphical co mparison of tele metry and model results shows a good accuracy. this comparison is usually enough for this kind of simulat ions and perfectly aligned with the scope of the paper aimed to discuss changes of rules rather than the accuracy of lap time simulations . (a) velocity vs. distance of a f1 car (b) longitudinal acceleration vs. distance of a f1 car fig. 5 velocity and longitudinal acceleration vs. distance of a f1 car covering one standard race lap at monte carlo (a) propulsive vs. distance of a f1 car (b) bra king powers and propulsive, braking and recoverable energy vs. distance of a f1 car fig. 6 propulsive and braking powers and propulsive, bra king and recoverable energy vs. distance of a f1 car covering one standard race lap at monte carlo advances in technology innovation, vol. 3, no. 1, 2018, pp. 26 35 33 cop y right © taeti figs. 5 and 6 p resent same results of fig. 3 and 4 with a diffe rent set up permitting 1.6 s slowe r lap t imes. in fig. 5, lap time is 1:18:700. this lap is covered by using the ice energy only. the grip is reduced 9% vs. the qualify ing conditions reflecting tire usage. in fig. 6, lap t ime is 1:18:700. this lap is covered by using the ice energy only. the grip is reduced 9% vs. the qualifying conditions reflecting t ire usage. the propulsive energy is 16.12 mj; the bra king energy is 8.38 mj and the theoretically recoverable energy is 2.00 mj. 4. discussion traditional limit ing factors fo r the m gu -k bra king energy recovery a re the powe r o f the unit, the total energy storage, the balance in between the front and rear a xle b raking and the balance in between mgu -k and friction braking. certain ly, energy storage at powers much higher than 120 kw also has some downfall for an f1 car. the aero drag is so huge that there is much less energy available at very high vehicle speeds. furthermore, the very high energy numbers only las t for a fraction of second. however, the much higher power and energy storage limits permitted in the lmp1-h series certainly show the way to move. if the energy input fro m the m gu -k to the es may not e xceed 2mj in any one lap and energy re leased fro m the es to the m guk may not e xceed 4mj in any one lap, this means that the continuous use of the kers is eventually limited to just the 2mj recovered and reused per lap, while in the strategic use of the kers, the 4mj could be made available in a lap providing in the previous lap there has been no discharge of the kers. this does not help the fuel e conomy. fro m a global fue l energy perspective, the total fuel available to cover the 78 laps or 260.52 km in monte carlo is 100 kg. this translates roughly in 1.28 kg per every lap of 3.34 km. by assuming a lowe r heating va lue for gasoline of 44.5 mj/kg, this translates in a ma ximu m fue l energy supply of 57.0 mj per lap. the power unit energy requested per lap is less than 16.1 mj. this translates in an average fuel effic iency of only 28.2% requested to the engine without any working kinetic energy recovery. by recovering the 2 mj per lap with the mgu-k, this effic iency could be further reduced to an even smaller 24.7%. these efficiencies are everything but great. while better estimations may certa inly follow the use of the tele metry data and the lap time simu lations tools the teams do have, when considering the 1.5 liters v6 turbo engines of the prev ious turbo era a lmost 30 years ago were a lready operating with effic iencies we ll above 30% in a range of operating points of interest, it does not seem that th e 100 kg per race is a rea lly an up-to date limit set to push forward the boundaries of energy efficiency, as the teams may easily ach ieve th e 28.2% efficiency and avoid recovering the braking energy in norma l laps, recharging the es only strategically and using the mgu-k only when needed to gain/defend a position. the previous analyses are done without any inclusion of the drag reduction system (drs). this overtaking a id permits a driver within one second of a rival car within designated drs act ivation zo nes to alter the angle of the rear wing flap, reducing drag coefficient and thereby achieving a temporary speed advantage. the drs has no relevant impact on the fuel economy. regarding the energy flow through the mgu-h, any transformation of energy type, for e xa mp le mechanical to electrica l to chemical back to electrical and back to mechanical occurs with efficiency far from unity. how powerful is today’s power unit of hybrid f1 when compared with traditional powe rtrains of the past is a question difficult to answer. for ma rket ing purposes, it is co mmon to c laim that today’s power units have the peak power of the ice, plus the peak power of the m gu-k, with the mgu-h possibly further increasing the power of the ice by precise e xtra boost. the ice delivers power to the whee ls as a function of the speed of the crankshaft. the 450 kw of peak power are obtained at high speeds approaching the speed limite r and certain ly not at low speeds. the m gu -k also delivers power to the whee l, but the 120 kw of peak power are now available at any speed of the crankshaft. advances in technology innovation, vol. 3, no. 1, 2018, pp. 26 35 34 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti the best use of the mgu-k is to produce a faster acceleration after a bend and not to increase the top speed on a straight. the mgu-h of untapped power does not deliver any power to the wheels. the mgu-h is only linked to the turbocharger, and may only help the ice to delive r more power by spinning faster the turbine above the balance in between gas expansion in the turbine and air compression in the compressor. if the mgu-k power is supplied at low speed, then the equivalent torque of the engine drastically improves. today’s power unit of hybrid f1 are by fa r less powerful of past traditional powert rains, but certain ly have much better torque. 5. conclusions to return to be technically attractive, f1 should permit much more freedo m in the defin ition of the ice and the ers. as the goal of the rules should be the lowering the fuel consumption while keeping high the technical and sporting interest, the best solution is more freedo m to achieve the fastest car wit hin mo re stringent limits of fuel economy. this would benefit the racing and the everyday car. a real limit should be set to the ma ximu m a mount of fuel to be used for a fixed distance race, and the engineers should be , then, left free to develop the hybrid power unit with at the most a prescribed displace ment of the engine. the p resent limit of 100 kg of fuel per race does not force the teams to recover the 2 mj of energy every lap, and does not force them to use the fuel much more efficiently within the internal combustion engine that what is common practice since decades. a real limit to the total fuel consumption for a race like monte ca rlo should be not more than 80 kg of fue l, that would require the recovery of the 2 mj of energy every lap and an average fuel effic iency of the ice of 30.9% , everything but impossible to reach with today’s technologies, but certainly much better than what is presently delivered by today’s f1 inter nal combustion engines. alternatively, the teams could continue to use only strategically the m gu-k not as a fuel saving measure, but they should, then, improve their internal co mbustion engines to an average fuel e fficiency of 35.2%. these numbers will certainly need the development of novel strategies that may help the design o f passenger cars . references [1] a. boretti, kinetic energy recovery systems for racing cars, pennsylvania: society of automotive engineering (sae) international, 2013. [2] a. boretti, “kers braking for 2014 f1 cars,” society of automotive engineering (sae) technical paper 2012-01-1802, september 17, 2012. [3] a. boretti, “f1 2014: turbocharged and downsized ice and kers boost ,” world journal of modelling and simulation, vol. 9, no. 2, pp. 150-160, 2013. [4] a. boretti and i. aris, “regenerative braking of a 2015 lmp1-h racing car,” society of automotive engineering (sae) technical paper 2015-01-2659, september 27, 2015. [5] m. petrány, “how formula one's amazing new hybrid turbo engine wo rks,” http://jalopnik.com/how-formula-ones -amazi ng-new-hybrid-turbo-engine-works -1506450399, january 22 2014. [6] w. f. milliken and d. l. milliken, race car vehicle dynamics, pennsylvania: soc iety of automotive engineering (sa e) international, 1995. [7] “optimumg-vehic le dynamics solutions,” http://www.optimumg.com, 2017. [8] “lapsim-professional racing simulation software,” http://www.lapsim.nl, 2014. [9] m. s. al muharrami et al., “study of open and closed wheels aerodynamic of racing cars in wind tunnels with movable ground at different car speeds,” 18 th australasian fluid mechanics conference, australia, december 2012. [10] u. hop mann and m. algra in “ diesel engine e lectric turbo co mpound tech nology,” society of automotive engineering (sae) technical paper 2003-01-2294, june 23, 2003. [11] f. millo, f. malla mo, e. pautasso and g. ganio mego, “the potential of electric e xhaust gas turbocharging for hd d iesel engines,” society of automotive engineering (sae) technical paper 2006-01-0437, april 3, 2006. [12] s. ibara ki, y. ya mashita, k. su mida, h. ogita, and y. jinnai, “ develop ment of the “hybrid turbo,” an electrically assisted turbocharger,” mitsubishi heavy industries, ltd. technical review, vol. 43, no. 3, september 2006. http://jalopnik.com/how-formula-ones-amazing-new-hybrid-turbo-engine-works-1506450399 http://jalopnik.com/how-formula-ones-amazing-new-hybrid-turbo-engine-works-1506450399 http://www.optimumg.com/ http://www.lapsim.nl/ advances in technology innovation, vol. 3, no. 1, 2018, pp. 26 35 35 cop y right © taeti [13] a. patterson, r. tett and j. mc gu ire, “ exh aust heat recovery using electro -turbogenerators,” society of automotive engineering (sae) technical paper 2009-01-1604, may 13, 2009. [14] t. katrašnik et al., “an analysis of turbocharged diesel engine dynamic response improvement by electric assisting sys tems,” journal of engineering for gas turbines and power, vol. 127, no. 4, pp. 918-926, july 2005. [15] n. terdich and r. martinez-botas, “experimental efficiency characterization of an electrically assisted turbocharger,” society of automotive engineering (sae) technical paper 2013-24-0122, september 8, 2013. definitions/abbreviations a acceleration lmp1-h le mans prototype 1 hybrid e energy m mass e-kers electric kers mgu motor-generator unit ers energy recovery system mgu-k driveline mgu es energy store mgu-h turbocharger mgu f force p power ice internal combustion engine r force kers kinetic energy recovery systems v velocity  advances in technology innovation, vol. 4, no. 4, 2019, pp. 222-233 development of a cost-effective design of a p-v ventilated greenhouse solar dryer for commercial preservation of tomatoes in a rural setting ebangu orari benedict 1,2,* , dintwa edward 1 , motsamai seraga oboetswe 1 1 department of mechanical engineering, university of botswana, gaborone, botswana 2 department of biosystems engineering gulu university, gulu, uganda received 28 march 2019; received in revised form 16 may 2019; accepted 09 july 2019 abstract commercial preservation of agro-produce by solar drying entails using large-scale dryer units that are generally not affordable by potential users in a rural setting. the objective of this study was to design a photo-voltaic greenhouse solar drying unit with air convection powered by a photovoltaic system as part of a research study to develop a cost-effective commercial solar dryer for the preservation of tomatoes in a rural setting. the design methodology involved. first, the determination of thermophysical properties of local tomatoes and predicting its drying model. second, determination of design parameters; design equations, design conditions and assumptions, and psychrometric analysis. third, application of the design parameters under the climatic conditions of botswana to produce the embodiment of design with the drawings and specifications for a solar hot air dryer unit of 2000 kg batch-load of wet tomatoes. this drying unit is integrated into a solar collector with a cost-friendly thermal energy storage system to store solar energy during the day for use after sunset for better performance. keywords: commercial preservation, thermo-physical properties, embodiment of design, thermal energy storage system 1. introduction the discussion on applicable methods to alleviate the high global postharvest losses, 30-50 %, that negatively impact on the agro-produce value chain is on-going [1]. recent research work has focused on using renewable energy technologies for drying of agro-produce as the preservation method to reduce postharvest losses [2-3]. solar dryers have therefore increasingly become popular especially on account of their relative costs of investment and operation [4-7]. however solar dryers are scarce in botswana despite the endowment with abundant sunshine [8]. this study aims at developing a suitable solar dryer for use in botswana to preserve fruits and vegetables, particularly tomatoes. drying of tomatoes for commercial purpose in a rural set up requires the use of tunnel or greenhouse types of solar dryers. in this paper, the design of a forced convection solar dryer ventilated by fans operated by a photo-voltaic system is presented. various designs of solar dryers for the preservation of fruits and vegetables have been reported in the literature [9]. experiments were carried out on solar drying of pineapple using a solar tunnel dryer, the version of the original hohenheim type solar dryer in bangladesh [10]. this dryer had a transparent polyethylene covered flat plate collector and a drying tunnel. hot air was supplied to the drying tunnel using fans operated by a solar module. however, the tunnel dryer design had limited batch capacity for large scale commercial drying of products. design innovations have been made aimed at increasing the capacity and efficiency of solar dryers. a solar dryer with reverse flat-plate collector absorber to receive concentrated solar irradiation at a higher temperature to increase the efficiency of the dryer was developed [11]. the geometrical construction and manual tracking of the sun rays were the major challenges * corresponding author. e-mail address: benebangu@yahoo.com tel.: +267-76675111 advances in technology innovation, vol. 4, no. 4, 2019, pp. 222-233 223 of this technology. a solar dryer comprising a flat plate collector that used mirror reflectors to enhance the capture of sol ar irradiance at the solar heater unit of a mixed-mode greenhouse dryer was designed and constructed [12]. the dryer was tested by successful drying of bananas and tomatoes. however, to track the sunshine, the mirrors needed to be continuously adjusted during the day and this presents a labor intensive activity. a large scale polycarbonate covered greenhouse / tunnel type solar dryer using forced convection air flow was developed [13]. it was tested for performance and it did demonstrate the potential for drying chilli, coffee and banana. this dryer design had the potential to be developed to dry, at large scale, tomatoes and other horticultural products if auxiliary energy source was incorporated into the dryer system to enable drying without interruption. janjai [14], developed and tested at field level a large-scale solar green-house dryer with a loading capacity of 1000 kg of fruits or vegetables. this greenhouse dryer had a parabolic shape and was covered with polycarbonate sheets. the dryer was backed up by the use of lpg gas energy source and its ventilation fans were powered by photo-voltaic, p-v, module. it was used to dry bananas, coffee, chilli, tomatoes, and other fruits and vegetables. the solar greenhouse dryer was easy to construct and was suitable for commercial-scale applications because of the fast drying rate. however, a cheaper source of auxiliary energy than lpg gas could reduce the operational cost of the dryer. from the reviews of the greenhouse solar dryer technologies, sdts, summarized in table 1, great effort has been put to develop greenhouse solar dryers for fruits and vegetables but few designs have been developed and tested for commercial drying of tomatoes in a rural setting where affordability of the technology is a matter of great concern. table 1 summary of material properties year year highlights 2016 hii, et al. [15] development and recent trends in greenhouse sdts 2016 hawlader, et al. [16] solar tunnel greenhouse type of sdt 2017 sengar and kothari [17] thermal modeling and performance evaluation of arch shape greenhouse for nursery raising 2017 atalay, et al. [18] modeling of the drying process of apple slices: application with a solar dryer and the thermal energy storage system. 2019 badaoui et al. [19] experimental and modelling study of tomato pomace waste drying in a new solar greenhouse: evaluation of new drying models 2019 román‐roldán, et al. [20] computational fluid dynamics analysis of heat transfer in a greenhouse solar dryer “chapel‐ type” coupled to an air solar heating system thermo-physical properties of the material to be dried define the boundary conditions for the drying process. density, specific heat capacity, heat conductivity, thermal diffusivity, moisture content are the properties of tomatoes were assessed from the literature and also determined by laboratory experiments. the experiments were necessary because although the tomato varieties prevalent in botswana are of mexico native origin, solanum lycopersicum, their properties could vary due to climatic and ecological conditions the six common varieties grown in botswana are money maker, expresso, heinz 1379, zeal, zest, and six pack. the farmers in botswana have ranked the zeal variety as the best variety [21]. 2. design methodology the design methodology considers the conceptual design of a phot-voltaic air fan ventilated dome-shaped design of a greenhouse solar dryer for large scale drying of tomatoes. hence the thermo-physical properties are determined in the first step. the design equations and design calculations are discussed in the second step and the embodiment of design is handled in the final step. the flow diagram of the design process is depicted in fig. 1. 2.1. determination of thermophysical properties of tomatoes the experimental setup for determination of moisture content was in respect of an oven drying method used to determine the moisture content of sliced tomatoes. the oven dryer apparatus (carbolite-gero gmbh co., kg 30-3000 °c) was used to dry the tomato samples. it was set to operate at a temperature of 105°c, and the air in the oven was set to ventilate at 1.1 m/s. a calibrated electronic balance type: snowrex lv-6b was used for measurement of weight. advances in technology innovation, vol. 4, no. 4, 2019, pp. 222-233 224 fig. 1 flowchart of the design methodology 2.2. experimental procedure tomato samples of the zeal variety that is most prevalent in botswana were procured from gaborone supermarkets and used as the material for determining the moisture content of the tomato for drying. the tomatoes were first washed with running water. no chemical pre-treatment of the samples was done. a vegetable cutting knife was used to cut the tomato samples to halfslices of about 25 mm diameter. these slices were weighed and placed inside the oven dryer. the oven-method procedure of the hot air oven method according to acac 1990 for determination of moisture content of sliced tomato was followed. the initial weight of each tomato-slice sample with its containment was recorded after determining the weight of the empty container. then the instant weights of each sample were recorded on an hourly basis while drying in the oven. the decreasing masses of the sample slices were monitored to a final stable weight by means of a calibrated digital balance. the end of drying was reckoned by uniform constant weight readings. weight loss was determined at 60-minute intervals during drying. visual inspection of color and texture was relied upon to confirm the dried status of the tomato slices [22]. 2.3. the design equations moisture content was calculated as the ratio of weight loss during the drying period with the weight of the wet or the dry weight of the sample. the value of moisture content usually expressed in percentage, %, or in decimal values, was thus presented either on a wet basis, wb , or dry basis, db . the value of moisture content usually expressed in percentage, %, or in decimal values, was thus presented either on a wet basis, wb or dry basis, db . the moisture content, wet-basis, and moisture content dry-basis were evaluated [23] by o d wb o w w m w   (1) o d db d w w m w   (2) where ow is the weight of the wet material and dw is the weight of the dry material. the relationship between wbm and dbm can be given by two expressions of moisture content of agricultural materials as follows 1 wb db wb m m m   (3) advances in technology innovation, vol. 4, no. 4, 2019, pp. 222-233 225 1 db wb db m m m   (4) the moisture content at time t during drying, tm , is evaluated by [22]. ( 1) 1 t t i o w m m w          (5) where i m is the initial moisture content, db , tw is the instant weight of the sample, ow is the initial weight of the sample. the moisture ratio rm is a function of time that rationally defines the moisture content of the drying process. since fm is relatively negligible in comparison to im and tm . rm is approximately given as the ratio given by [23]. t f t r i f i m m m m m m m     (6) the mass of the moisture to be removed from the food product is calculated by berichte [24] ( ) 100 p i f w f m m m m m    (7) where, pm , is the initial mass of the product to be dried, wm is the mass of water to be removed from the wet material, i m is the initial moisture content, % dry basis, fm is the equilibrium or final moisture content, % dry basis. latent heat of vaporization was calculated by [25]. 3 4.186 10 (597 0.57 )prl tv    (8) where vl is the latent heat of vaporization of water and prt is the drying temperature in °c of the material to be dried. the quantity of heat energy required for drying a given quantity of food materials can be approximated using the basic energy balance equation for the evaporation of water was given norton and sun by [26] 1 2( )wq m l m c t tv a pa   (9) where vl =latent heat of vaporization of water in kj/kg, wm is the mass of water evaporated from the food item, am is the mass of drying air, 1 2( )t t is the difference between initial and final temperatures of the drying air respectively, in kelvin, pa c is the specific heat capacity of drying air at constant pressure kj/kg k. the enthalpy h of air was approximated [27] by 1006.9 (2512131.0 1552.4 )h t w t   (10) where t is the temperature of drying air, °c, and w is the humidity ratio of water vapor, in one kg of dry air. the amount of water in food and agricultural materials affects the quality and perishability of the material. the indicator for perishability of the material is wa the water activity. it is the level of availability of moisture to support degradation like microbial action. the water activity is evaluated as [28] advances in technology innovation, vol. 4, no. 4, 2019, pp. 222-233 226 w o p a p  (11) where p is the partial pressure of water above the material surface and, op is the partial pressure of pure water under the same conditions as the food. the water activity can also be evaluated using the empirical formula [29] by 1 exp[ exp(0.914 0.5639 ln )] 100 rh db e w m     (12) where db m is the moisture content dry basis in kg water vapor/kg pure water and, rh e is the relative humidity of the air surrounding the food at which the material neither gains nor loses its natural moisture and is in equilibrium with the environment. the average drying rate rd for drying the food was evaluated by [26] w r d m d t  (13) where w m is the mass of moisture removed from the food and d t is the drying time in hours. the mass flow rate, am of the air needed for drying, is given by [26] ( ) r a f i d m w w   (14) where rd is the average drying rate, iw is the initial humidity ratio, kg h2o/kg dry air, fw is the final humidity ratio, kg h2o/kg dry air. the volumetric airflow rate av in m3/hr is evaluated by a a a v m   (15) the total useful thermal energy of the drying air, e in joules, required to evaporate the water is given [30] by = ( )a f i de m h h t (16) where the air mass flow rate is am in kg/hr, ih and fh are the final and initial enthalpy of drying air and ambient air respectively in j/kg of dry air and, d t is the drying time in hours the dryer area was calculated from the equation by [31] de a i (17) where the area of the dryer cover is d a in m2, i is the total incident radiation on the greenhouse dryer surface during the drying period, and  is the dryer efficiency which is between 30-50% [32]. the vent area v a was determined by the ratio of the volumetric flow rate of air inside the dryer a v to wind velocity w v by advances in technology innovation, vol. 4, no. 4, 2019, pp. 222-233 227 a v w v a v  (18) 2.4. psychrometric chart thermodynamic parameters were obtained by use of psychrometric calculator developed from psychrometric charts that describe the thermodynamic process of hot air drying [33]. the most common dehydration processes use hot air as the drying medium as the air delivers heat to the product in order to evaporate moisture. the properties of air, most easily understood by psychrometric relationships are critical to understanding the process of evaporation. the dry air can be visualized as a gaseous solution of dry gases in constant proportions and water vapor in varying amount. at least two properties of air must be known in order to use the chart to characterize the drying air. amongst the properties of air that are critical to drying are absolute humidity, enthalpy or heat content, specific volume, and relative humidity. the thermodynamic changes that take place in a solar dryer include heating depicted line a-b of fig. 3 whereby there is a rise in temperature from ta to tb at constant absolute humidity. the relative humidity reduces from ha to hb. this is followed by the heating phase depicted by line b-c whereby the air picks up moisture from the product and the temperature drops from the drying temperature attained by air inside the dryer tb to exit temperature tc. in this process, the relative humidity increases from hb to hc. the humidity ratio will increase from hb to hc. the initial enthalpy, ha, will change to hc at c’. table 2 psychrometric values of drying air item temperature-dry basis in °c relative humidity in % enthalpy in kj/kg humidity ratio in kg/kg specific volume in m 3 /kg ambient temperature 24 55 53.70 0.0116 0.9695 solar collector outlet air 45 21.09 82.36 0.0143 0.913 dryer outlet air 37 50.0 89.28 0.0202 0.890 table 3 design conditions and assumptions s/no items condition and assumption 1 location: botswana, gaborone, university of botswana elevation:1005 m, 24.56 0s, 25.92 0e 2 material for drying tomato 3 drying period using solar energy exclusively, d t , the average bright sunshine hours in a day 30 4 loaded mass, pm kg/batch 2000 5 initial moisture content, mi, %, bw , measured in the laboratory 96 6 final moisture content, mf, %, bw , evaluated using equation 1 at wa =0.55 12 7 average ambient air temperature, tam 0 c 24 8 average ambient relative humidity, rh, % 44 9 maximum allowable temperature, tprmax, 0 c 60 10 maximum allowable product temperature, tpr, 0 c 45 11 specific heat capacity of tomato slices cpp, kj/kg 4.080 12 average incident solar radiation, in mj m 2 /day 21.6 13 wind speed, vw, km/h, from gaborone weather forecast 6 14 collector efficiency.  =25-50 % [32] 30 15 the thickness of half sliced tomato, m 0.03 the changes in the drying air are evaluated by the assistance of the psychrometric chart/calculator. this thermodynamic data is given in table 2. these parameters are used in the design process. the assumption was made that there was no heat loss between the collector and drying chamber; the solar collector outlet temperature was assumed to be equal to the dryer inlet temperature. the conditions and assumptions for the design calculations are given in table 3. advances in technology innovation, vol. 4, no. 4, 2019, pp. 222-233 228 fig. 3 the principle of the psychrometric chart 3. results and discussions 3.1. the tomato thermo-physical properties of tomatoes the thermophysical properties were determined in respect of moisture content, density, specific heat capacity, heat capacity and conductivity for the zeal tomato variety. table 4 gives the results of weight measurements during oven drying. table 4 weight of tomato samples during oven drying expired hours w1, g w2, g w3, g w4 average, g m (t) mrexp 0 63 61 66 63.3 24.01 1 1.05 48 36 50 44.7 16.64 0.69 2.08 32 34 38 34.7 12.69 0.53 3.17 28 30 34 30.7 11.11 0.46 3.95 25 28 30 27.7 9.93 0.41 4.7 22 24 27 24.3 8.61 0.36 5.69 17 21 24 20.7 7.16 0.3 6.75 14 17 20 17.0 5.71 0.24 7.05 9 13 16 12.7 4 0.17 7.95 4 10 12 8.7 2.42 0.1 8.86 3 6 8 5.7 1.24 0.05 9.11 3 7 7 5.7 1.24 0.05 9.72 3 7 7 5.7 1.24 0.05 fig. 4 determination of the initial moisture content of tomato samples fig. 5 moisture content mt, db and mt, wb, wet and dry basis advances in technology innovation, vol. 4, no. 4, 2019, pp. 222-233 229 the results depicted in fig. 4 shows the curves of reducing weight for samples of tomato drying samples. the initial moisture content, wet basis, wb, was determined using equation 2 as 87%, 90%, and 96%, and for tomato samples sample ts1, sample ts2, and sample ts3, respectively. the moisture content for the local botswana tomato in this study was taken as 96% as it was the highest value from the samples that were analyzed. mass of tomato was measured using a weighing digital scale and volume was determined by displacing water from a calibrated flask by the tomato sample. hence the density of the local zeal variety tomato was determined. fig.5 depicts the moisture ratio tm , bw , and tm , db , calculated from the average weight measurements of the oven drying process using eq. (5), the moisture content wet basis was determined using eq. (3). fig. 6 shows the moisture ratio determined from the tomato oven drying experiment using the approximate eq. (6). by use of excel curve fitting solver, the predicted moisture ratio for this drying process was modeled as curve expon mrexp. fig. 6 mrexp determined from experiment and the fitted mrpred the moisture ratio model is given in table 5 for oven drying of tomato slices at 105°c and air velocity of 1.1 m/s was derived by curve fitting model of excel software, is the newton model with regression coefficient r 2 =0.921. table 5 tomato moisture ratio model model formula exp( )mr kt  k r 2 model  exp 0.285mrpred t  0.285 0.921 newton specific heat capacity and thermal conductivity were obtained from the literature [34, 35]. the thermophysical properties of the zeal tomato used in this study are given in table 6. table 6 properties of the round ripe red zeal tomato grown in botswana weight in gram s volume, cm 3 density, kg/m 3 specific heat capacity, kj/kg °c thermal conductivity, w/m k diffusivity in m 2 /s*e-7 initial moisture content, % final moisture content, % 132 130.1 1014 4.080 0.59 1.44 96 12 3.2. design assumptions and calculations the results of calculations done on the basis of equations given in section are summarized in table 7. the parameter of interest is the top area of the solar energy interface. by the determined value of the area, the researcher was able to proceed to complete the design based on priorities for the design concept and other engineering design parameters and procedure and to give specifications of the components and materials to be used. table 7 design assumptions and calculations s/no description equation value 1 mass of water to be evaporated from 2000 kg of fresh tomatoes, and 50 kg for prototype, mw. in kg using equation 4.5 1795 2 latent heat of vaporization in mj using equation 4.6 2.390 3 the energy required for evaporating water from the product, in mj using equation 4.7 4.562 advances in technology innovation, vol. 4, no. 4, 2019, pp. 222-233 230 table 7 design assumptions and calculations (continue) s/no description equation value 4 equilibrium relative humidity, in % erh=aw*100% using equation 4.10 55 5 initial humidity ratio, wi from psychrometric chart/www.psychrometric-calculator.com. 0.0143 6 final humidity ratio, wf from psychrometric chart/www.psychrometric-calculator.com. 0.0202 7 enthalpy, hi, initial enthalpy of ambient air, kj/kg/kgda from psychrometric chart/www.psychrometric-calculator.com. 82.36 8 enthalpy, hf of final moist air kj/kg/kgda from psychrometric chart/www.psychrometric-calculator .com. 89.28 9 average drying rate mdr for td=10 hours per day for 3 days, kg/hr using equation 4.11 59.85 10 airflow rate, ma, kg/hr using equation 4.12 10144.07 11 volumetric air flow rate, va in m 3 /min using equation 4.13 141.0 12 total useful energy, e in mj using equation 4.14 2105.909 13 greenhouse top area ad m2 using equation 4.15 324.98 14 greenhouse with floor area af m2 of dimensions of about 13 m widths by 25m length af=ad 325 15 the linear air flow rate in m/s conversion of va over the flow area 0.007 16 the vent area av m2 when wind velocity vw is 6 km/h using equation 4.16 1.41 17 the diameter of a cylindrical vent d m d=2*sqrt av/π 1.34 3.3. design features and material specifications table 8 specifications of major items of the designed greenhouse dryer s/no item description specification 1 polyethylene ultra violet resistant cover clear 2-3 mm thick film 2 horizontal floor area 325 m 2 . 3 the volume of greenhouse dryer 1035 m 3 4 drying table for tomato slices plastic mesh on an aluminum frame 5 photovoltaic, p-v unit 100-watt, 100 ah solar battery 6 air fan unit 150 m 3 /min 7 inlet/outlet ducts diameter aluminum 1340 mm (a) the isometric drawing (b) top view (c) back view (d) front view fig. 7 drawings of the design of the ventilated greenhouse dryer showing the different views the final drawings of the ventilated greenhouse solar dryer were prepared using solid works software and ansys modeller 17.2. centrifugal blow fan is fitted at the inlet and the exhaust fan of the same rating is fitted at the outlet to exit the drying air. these fans are powered by a photovoltaic system, not shown in the drawing. the dryer base is covered with checked aluminum plate, painted black and is insulated from the ground. inside the dryer is the drying rack where half-sliced tomatoes for drying are laid. the greenhouse is covered by polyethylene ultra-violet film and entry to the inside of the dryer is accessed by opening the door in the front side. the design drawings are given in fig. 7. the major components are listed in table 8. 3.4. cost-effectiveness of design the final design of the greenhouse solar dryer is to be integrated with a solar collector with a thermal energy storage system to enable the drying process to continue after sunset. sensible heat storage, shs, systems are relatively simple and advances in technology innovation, vol. 4, no. 4, 2019, pp. 222-233 231 relatively of lower cost compared to chemical or latent thermal energy storage systems [36]. solid sensible heat storage system using granite rock as the thermal storage material has been identified for use in the design of the thermal energy storage system. in comparison with latent and thermo-chemical optional thermal energy storage systems, the use of sensible rock thermal energy storage system is the most cost-effective as regards technical, cost and environmental criteria. the classification of the shs system using rock material that is low cost, ready availability, negligible maintenance, and operational cost would result in a relatively low-cost solar collector. when the solar collector with the tes system is integrated into the greenhouse unit, the technology becomes cost-effective to fulfill the main objective of commercial drying of tomatoes even after sunset in a rural setting. fig. 8 depicts the schematic drawing of the design of the greenhouse solar dryer with the proposed solar collector. fig. 8 schematic design drawing of the p-v ventilated greenhouse solar hot air tomato dryer 3.5. design materials in selecting the design materials, consideration of their effectiveness and price was made. the ventilated greenhouse dryer’s cover material was chosen based on suitability to interface with the incident solar radiation, strength and price. plastic films and sheets are relatively cheaper than glass. generally, plastics may have long-wave transmittance of up to 0.4 with a maximum temperature of 120°c. for application in low-temperature drying, a maximum of 60°c, the ultra-violet resistant polycarbonate material was recommended but it was scarce in botswana. polyethylene ultra-violet resistant film, 2-4mm thickness, was alternatively available in gaborone and was thus chosen. the insulation material was required for the base of the dryer. the base comprising of an insulated slab made of polyurethane foam, pur/pir that is frequently employed in storage tanks was recommended. the insulation material has an extremely low thermal conductivity of 0.02-0.026w/m k possess excellent mechanical strength, exceptional durability, and moisture resistant. the insulation material was cost-effective and easy to install. the output of the solar collector was the drying air that flowed to the drying chamber via a duct, aluminium and galvanized-tin ducts are the two standard most common materials for fabricating ducts because of their high optical reflectance. aluminum is light-weight, has a low density of about third of other metals. it has a high thermal conductivity of about three times that of steel and can be easily fabricated and installed on-site. however, the galvanized tin duct was relatively cheaper than the aluminium duct and was recommended as the ducting material. airflow inside the solar dryer was taken care of by the use of a solar photo-voltaic, p-v module. the solar panel system was incorporated in the design as part of the dryer technology and was used to power the fanning system for a controlled operation depending on insolation levels. a solar battery 0f 100 ah in the p-v system of one 100 w panel was to be used to store energy in order to save on the high operational cost of using electricity. advances in technology innovation, vol. 4, no. 4, 2019, pp. 222-233 232 4. conclusions the following conclusions were made from the development of the design of the solar dryer unit: (1) the thermo-physical properties of the tomato were determined as weight: 132 g, volume: 130.1 cm3, density: 1014 kg/m3, specific heat capacity: 4.080 kj/kg °c, thermal connectivity: 0.59 w/m k, diffusivity: 1.44e-7 m/s2, initial moisture content: 96% wb , and final moisture content: 12 % wb . (2) the moisture ratio curve for drying tomato slices at a temperature of 105°c and air velocity of 1.5 m/s, under oven-drying, was fitted to the newton model with regression coefficient r2=0.921 using excel software. the drawings of the design of the p-v powered forced convection commercial solar drying unit was produced. locally procured materials were specified in the design consistent with the objective of producing a cost-effective solar dryer. the solar dryer was designed as a mixed-mode type to operate continuously using direct sunshine during the day and stored solar energy at night to dry tomatoes. the design was done under the assumptions and conditions of botswana. in the finalization stage of the design process, all the parameters were applied to produce the embodiment of design with the design drawings and specifications of the ventilated greenhouse solar dryer unit. this unit will be integrated into the solar collector with a solar energy storage system for use after sunset. conflicts of interest the authors declare no conflict of interest. acknowledgment the authors of this paper acknowledge mobility to enhance training of engineering graduates in africa, metega for funding of the post graduate study program at the university of botswana under which this study was made possible. references [1] w. h. organization, “the state of food security and nutrition in the world 2018: building climate resilience for food security and nutrition,” food & agriculture organization of the united nations, 2018. [2] m. güne, b. blywis, f. juraschek, and p. schmidt, “concept and design of the hybrid distributed embedded systems testbed,” freie universität berlin, technology rep., 2008. [3] c. gordon and s. thorne, “determination of the thermal diffusivity of foods from temperature measurements during cooling,” journal of food engineering, vol. 11, no. 2, pp. 133-145, 1990. [4] c. cheapok and p. pornnareay, “introducing solar drying in a developing country: the case of cambodia,” proceedings of world renewable energy congress (wrec), pp. 2198-2201, 2000. [5] a. esper, k. lutz, and w. mühlbauer, “development and testing of plastic film solar air heaters,” solar & wind technology, vol. 6, no. 3, pp. 189-195, 1989. [6] h. c. kaymak, i. ozturk, f. kalkan, m. kara, and s. ercisli, “color and physical properties of two common tomatoes (lycopersicon esculentum mill.) cultivars,” journal of food agriculture and environment, vol. 8, no. 2, pp. 44-46, april 2010. [7] a. fudholi, k. sopian, m. h. ruslan, m. alghoul, and m. sulaiman, “review of solar dryers for agricultural and marine products,” renewable and sustainable energy reviews, vol. 14, no. 1, pp. 1-30, january 2010. [8] w. weiss and j. buchinger, “solar drying,” aee intec publication, a-8200 gleisdorf, feldgasse, vol. 19, 2012. [9] s. lahsasni, m. kouhila, m. mahrouz, l. mohamed, and b. agorram, “characteristic drying curve and mathematical modeling of thin‐layer solar drying of prickly pear cladode (opuntia ficus indica),” journal of food process engineering, vol. 27, no. 2, pp. 103-117, june 2004. [10] b. bala, m. mondol, b. biswas, b. d. chowdury, and s. janjai, “solar drying of pineapple using solar tunnel drier,” renewable energy, vol. 28, no. 2, pp. 183-190, february 2003. [11] r. goyal and g. tiwari, “parametric study of a reverse flat plate absorber cabinet dryer: a new concept,” solar energy, vol. 60, no. 1, pp. 41-48, january 1997. advances in technology innovation, vol. 4, no. 4, 2019, pp. 222-233 233 [12] m. hossain, k. gottschalk, and b. amer, “mathematical modelling for drying of tomato in hybrid dryer,” arabian journal for science and engineering, vol. 35, no. 2, pp. 239, october 2010. [13] m. g. green and d. schwarz, “solar dryer plans,” 2014. [14] s. janjai, “a greenhouse type solar dryer for small-scale dried food industries: development and dissemination,” international journal of energy and environment, vol. 3, no. 3, pp. 383-398, january 2012. [15] c. hii, c. law, and m. cloke, “modeling using a new thin layer drying model and product quality of cocoa,” journal of food engineering, vol. 90, no. 2, pp. 191-198, january 2009. [16] m. hawlader, m. uddin, j. ho, and a. teng, “drying characteristics of tomatoes,” journal of food engineering, vol. 14, no. 4, pp. 259-268, 1991. [17] s. sengar and s. kothari, “thermal modeling and performance evaluation of arch shape greenhouse for nursery raising,” african journal of mathematics and computer science research, vol. 1, pp. 1-9, 2008. [18] h. atalay, m. t. çoban, and o. kıncay, “modeling of the drying process of apple slices: application with a solar dryer and the thermal energy storage system,” energy, vol. 134, pp. 382-391, september 2017. [19] o. badaoui, s. hanini, a. djebli, b. haddad, and a. benhamou, “experimental and modelling study of tomato pomace waste drying in a new solar greenhouse: evaluation of new drying models,” renewable energy, vol. 133, pp. 144-155, april 2019. [20] n. i. román‐roldán, a. lópez‐ortiz, j. f. ituna‐yudonago, o. garcía‐valladares, and i. pilatowsky‐figueroa, “computational fluid dynamics analysis of heat transfer in a greenhouse solar dryer “chapel‐type” coupled to an air solar heating system,” energy science & engineering, 2019. [21] baliyan, s. p. m., shesha rao. (2013). evaluation of tomato varieties for pest and disease adaptation and productivity in botswana. international journal of agricultural and food research. [22] purkayastha, m. d.,nath, a.,deka, b. c. and mahanta, c. l. (2013). thin layer drying of tomato slices. journal of food science and technology, vol. 50, no. 4, pp. 642-653. [23] m. g. s. green, “solar drying technology for food preservation,” gtz-gate, january 2011d. brooker, “mathematical model of the psychrometric chart,” transactions of the asae, vol. 10, pp. 558-560, 1967. [24] b. a. berichte, “experimental studies and mathematical modelling of solar drying system for production of high quality dried tomato,” pp. 1-173, 2007.d. jain and p. b. pathare, “study the drying kinetics of open sun drying of fish,” journal of food engineering, vol. 78, no. 4, pp. 1315-1319, february 2007. [25] s. youcef-ali, h. messaoudi, j. desmons, a. abene, and m. le ray, “determination of the average coefficient of internal moisture transfer during the drying of a thin bed of potato slices,” journal of food engineering, vol. 48, no. 2, pp. 95-101, may 2001. [26] e. akoy, m. ismail, a. ahmed, and w. luecke, “design and construction of a solar dryer for mango slices,” proceedings of international research on food security, natural resource management and rural development-tropentag. university of bonn, germany. [27] v. puri and r. anantheswaran, “the finite-element method in food processing: a review. journal of food engineering,” vol. 19, no. 3, pp. 247-274. [28] c. hii, c. law, and m. cloke, “modeling using a new thin layer drying model and product quality of cocoa. journal of food engineering,” vol. 90, no. 2, pp. 191-198. [29] j. hernandez, g. pavon, and m. garcıa, “analytical solution of mass transfer equation considering shrinkage for modeling food-drying kinetics,” journal of food engineering, vol. 45, no. 1, pp. 1-10, july 2000. [30] d. jain and p. b. pathare, “study the drying kinetics of open sun drying of fish,” journal of food engineering, vol. 78, no. 4, pp. 1315-1319, february 2007. [31] d. wray and h. s. ramaswamy, “novel concepts in microwave drying of foods. drying technology,” vol. 33, no. 7, pp. 769-783. [32] m. sodha, n. bansal, a. kumar, p. bansal, and m. malik, “solar crop drying,” cpr press., boca raton, florida, usa, 1987. [33] d. brooker, “mathematical model of the psychrometric chart,” transactions of the asae, vol. 10, pp. 558-560, 1967. [34] p. greidanus and m. a. verhoeven, “product data of tomatoes,” sprenger instituut, wageningen 0169-3638, 1986. [35] g. giovanelli, b. zanoni, v. lavelli, and r. nani, “water sorption, drying and antioxidant properties of dried tomato products,” journal of food engineering, vol. 52, no. 2, pp. 135-141, april 2002. [36] b. amer, m. hossain, and k. gottschalk, “design and performance evaluation of a new hybrid solar dryer for banana,” energy conversion and management, vol. 51, pp. 813-820, 2010. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation, vol. 5, no. 3, 2020, pp. 182-189 nonlinear dynamic analysis of direct acting tensioner of an offshore floating platform zong-yu chang 1,* , xin duan 1 , zhong-qiang zheng 1 , lin zhao 1 , yu-hu yang 2 , xian-yi zhou 1 , peng zhu 1 , jing-wen he 1 1 college of engineering, ocean university of china, qingdao, china 2 laboratory of mechanism theory and equipment design of ministry of education, tianjin university, tianjin, china received 08 august 2019; received in revised form 25 december 2019; accepted 16 may 2020 doi: https://doi.org/10.46604/aiti.2020.4529 abstract the offshore floating platform is the key equipment in offshore and gas development. the significant heave motions occur with the excitation of wind and waves, which will affect the safety of a riser system. a direct acting tensioner can be applied to reduce the effects on the riser system and be widely used on different kinds of offshore platforms. based on the analysis of the structure and working principle of a direct acting tensioner (dat), the nonlinear dynamic performance of dat riser system was studied. additionally, the dynamic model of the dat riser system is established and the dynamic response was gained by the numerical integration method. the differences of dynamic responses were compared between a linear model and a nonlinear model. the response on different side of the equilibrium position is asymmetric because of the nonlinear stiffness of dat. the results can be helpful for the design of dat. keywords: direct acting tensioner, nonlinear dynamic model, accumulator 1. introduction the explorations of oil and gas in the ocean have been rapidly developed in recent years [1]. various offshore floating platforms are widely used in the deep sea oil and gas exploration [2]. the offshore floating platform undergoes significant heave motion with the excitation of waves, which may lead to a huge pulling force or a riser connecting to the seabed buckled. as a result, it brings about serious accidents. the riser tensioner can reduce the influence of the platform's heave motion on the riser by controlling the relative displacement of the offshore floating platform and the riser, which can ensure that the riser does not buckle due to compression, and does not damage abruptly due to excessive tension [3]. as shown in fig. 1, the riser tensioner can be classified as a wireline riser tensioner system (wrt) and a direct acting tensioner system (dat) [4]. the wireline riser tensioner system is revealed in fig. 1(a). and the direct acting tensioner system is shown in fig. 1(b). the wireline riser tensioner system needs frequent maintenance because the wireline easily worn. the direct acting tensioner system is directly connected to the riser and the platform, which has high reliability and long service life. therefore, direct acting tensioners are welcomed by the industry [5]. scholars have carried out related researches on the performance of riser tensioners. xu [6] used a constant tension model to simulate the riser tensioner, and analyzed the deformation and movement of the riser by applying constant tension to the riser and ignoring the tension change caused by the platform heave motion. yong [7] and liu [8] applied a linear spring to * corresponding author. e-mail address: 854127780@qq.com advances in technology innovation, vol. 5, no. 3, 2020, pp. 182-189 183 simulate the tensioner. it is shown that the tensioner can reduce the influence of the heave motion on the riser to a certain extent. however, the nonlinear stiffness characteristics of the direct-acting tensioner cannot be reflected by the linear spring model. the stiffness of the hydro-pneumatic tensioner will be changed with different stroke. hatleskog [9] analyzed the nonlinear stiffness characteristics of the hydro-pneumatic tensioner and compared with results of the linear spring simulation. the results are indicated that the nonlinear characteristics of the tensioner can improve the compensation performance of the tensioner, thus the nonlinear characteristics of the tensioner cannot be ignored. zhang [10] analyzed the parameters which will affect the compensation performance of the direct-acting riser tensioner. researches demonstrated that the accumulator volume will influence the compensation performance. (a) wireline riser tensioner system (b) direct acting tensioner system fig. 1 wireline riser tensioner system and direct acting tensioner system hyewon lee [11] considered the characteristics of hydro-pneumatic tensioners, and established a mechanical analysis model of the offshore floating platform-direct acting tensioner-riser system. seung-ho ham [12] used a multibody dynamics system to analyze the movement of riser tensioners. studies have shown that accumulator volume and wave amplitude are important parameters affecting the compensation performance of tensioners. the previous analysis method ignored the effect of the tensioner on the riser tension change, applied directly to a constant tension on the top of the riser to simulate the tensioner to solve the dynamic response. in order to solve the above problems, in this study, the dynamic model of the marine riser system has been established. the nonlinear stiffness characteristics caused by the hydraulic and pneumatic system of the direct acting tensioner have been considered in the model. the dynamic response of the system has been solved based on numerical analysis methods. furthermore, the influence of nonlinear stiffness on response has been analyzed. 2. structure and principle of direct acting tensioner fig. 2 the marine riser system inner pipe tensioner riser string lmrp bop stack gps-satellit advances in technology innovation, vol. 5, no. 3, 2020, pp. 182-189 184 the structure of the marine riser system is shown in fig. 2. the upper end of the riser includes the inner pipe and outer pipe connected by a telescopic joint. the inner pipe and the floating platform are connected while the outer pipe has a lower flex joint connected to the lower marine riser package (lmrp) of the blowout preventer (bop) stacked on its lower end. the direct acting tensioner acts on the outer pipe of the riser through a tensioner ring, which is shown in fig. 3. fig. 3 direct acting tensioner system the direct acting tensioner is mainly composed of three parts: support frame, hydraulic cylinder, and tensioner ring. the support frame of the tensioner is installed on the platform. the accumulator provides power to compensate for the riser heave motion through the expansion and contraction of the hydraulic cylinder. 3. dynamic analysis of the marine riser system 3.1. the analysis of tensioner performance fig. 4 the structure of direct acting tensioner system table 1. the parameters of hydraulic cylinder component specifications value unit accumulator volume 3.2 m 3 initial pressure 190.2 bar hydraulic cylinder piston diameter 470 mm piston rod diameter 400 mm piston mass 2510 kg the structure of the hydraulic cylinder is shown in fig. 4 and the parameters of the hydraulic cylinder are shown in table 1. the rodless chamber is connected to the accumulator. the gas change process satisfies the conditions of the adiabatic process; therefore, it is suitable for the equation of state for the ideal gas. v is the volume of the accumulator gas. p is the gas pressure of the accumulator. v0 is the initial volume of the gas, and p0 is the initial pressure of the gas. according to the equation of state for the ideal gas, the four variables satisfy the equations as: support frame hydraulic cylinder tensioner ring hydraulic cylinder accumulator v p gas oil advances in technology innovation, vol. 5, no. 3, 2020, pp. 182-189 185 kk pvvp  00 (1) the pressure p is: 0 0 0 0 0 ( ) = ( ) k k p v v p p p v v a x   (2) where ap is the piston diameter; x is the displacement of the piston and the hydraulic cylinder force fi is: 0 0 0 ( ) k p i p p v f p v a x a p a   (3) 4 1 i i f f    (4) where f is the tensioner force. the relationship between tensioner force and the displacement of the piston is shown in fig. 5. fig. 5 the relationship between force and displacement of hydraulic cylinder it is obvious that the relationship between force and displacement is nonlinear. in addition, the magnitude of stiffness is transforming with the change of the displacement. 1.4 0 1 0 1 1 1 p p a pf k l x a x / v                (5) the stiffness curve is shown in fig. 6. fig. 6 the relationship between stiffness and displacement of riser-direct acting tensioner system 3.2. dynamics modeling of the riser system the model of direct acting riser tensioner with the floating platform and riser system is shown in fig. 2. the tensioner applies tension to the riser through the tensioner ring to suppress the heave motion of the riser. the scheme of the dynamic model is indicated in fig. 7. advances in technology innovation, vol. 5, no. 3, 2020, pp. 182-189 186 fig. 7 dynamics model of the marine riser system generally, the tilt angle for the hydraulic cylinders of the direct-acting tensioner is small. the effect on the tilt angle of the hydraulic cylinders on the lateral direction can be neglected. under the excitation of the wave, the heave motion of platform can be assumed simple harmonic as a . based on eq. (3), the tension of the tensioner f against the riser can be calculated and the motion of system can be written as: 4 4 1 1 ( ) ri i i i mx f mg c x a k x         (6) axx  (7) xkf rr  (8) where m and kr respectively represent the riser mass and stiffness. x and fr are the heave motion and tension of the riser respectively. i and ci represent the number of hydraulic cylinders and the hydraulic cylinder damping coefficient, respectively. the parameters of riser tensioner-riser system are shown in table 2. table 2. the parameters of riser tensioner riser system specifications value unit riser mass 7.49×10 5 kg number of hydraulic cylinders 4 riser stiffness 3430 kn/m hydraulic cylinder damping coefficient 967 n/(m/s) 4. dynamic response analysis of the marine riser system 4.1. dynamic response analysis of the riser (a) the tension of the riser in presence or absence of the tensioner (b) the heave motion of the riser in presence or absence of a tensioner fig. 8 riser response comparison with or without compensation advances in technology innovation, vol. 5, no. 3, 2020, pp. 182-189 187 the results of this part are obtained under the heave motion of the platform where the motion height (h) = 5m, and the period (t) = 20s [13]. the response of the riser is shown in fig. 8. fig. 8(a) shows the tension response of the riser with or without rise tensioner. fig. 8(b) reveals the heave motion of the riser with or without tensioner. according to the figure, it can be seen that with the riser tensioner, the oscillation magnitude of tension and heave motion is reduced obviously about 30% and 50%. 4.2. analysis of nonlinear characteristics of riser tensioner in order to simplify the analysis, the nonlinear model of the riser tensioner can be equivalently linearized. when the displacement range of the piston is from -8m to 8m, the linear fitting curve is obtained as fig. 9. the difference between linear and nonlinear model can be seen in this figure. fig. 9 linear fit of hydraulic cylinder (a) heave motion of riser with 3 meter displacement excitation (b) tension of riser with 3 meter displacement excitation (c) heave motion of riser with 10 meter displacement excitation (d) tension of riser with 10 meter displacement excitation fig. 10 riser response based on the two models, the responses of tension force and heave motion of riser are shown in figs. 10-11 with heave motion amplitude 3 m and 10 m of the platform. fig. 10(a) and 10(b) are the heave motion and tension of the riser with 3 meter displacement excitation. fig. 10(c) and (d) are the heave motion and tension of the riser with 10 meter displacement excitation. x(m) f (n ) advances in technology innovation, vol. 5, no. 3, 2020, pp. 182-189 188 from fig. 10(a) and 10(b), under the small amplitude motion of the platform, the results of the linear model and the nonlinear model agree extremely well. it is suggested that under small amplitude excitation, the linear model can be used instead of the nonlinear model however, when the platform motion reaches large amplitude, there are significant differences between the results of nonlinear model and linear model. it is obvious that the tension curve and motion curve is no longer symmetric in positive and negative directions for the nonlinear model. the responses of the marine riser system with four different amplitudes heave motion of the platform are calculated as figs. 11-12. fig. 11 presents the vertical displacement of the riser and fig. 12 indicates the tension of the riser. as shown in figs. 11-12, the degree of symmetric is increased with the amplitude of the platform motion and the nonlinearity of hydraulic cylinder. fig. 11 the vertical displacement of riser fig. 12 the tension of riser an index of asymmetric is carried out to represent the degree of asymmetric. what is more, the equilibrium position is assumed to be the steady state when the platform is not moving. the index s is defined as the ratio between the positive displacement and the negative displacement of the riser as: p n x s x  (9) where xp is the positive displacement the riser, xn is the negative displacement. the index of asymmetry degree with the different exciting displacement of platform is shown in fig. 13. the comparison of positive displacement and negative displacement is shown in fig. 13. the index of asymmetry degree with the different exciting displacement of platform is shown in fig. 14. fig. 13 comparison of positive displacement and negative displacement fig. 14 asymmetry index with different heave excitation as referred in section 3, the stiffness of the riser tensioner decreases with the increase of the displacement of the riser. the stiffness in a negative direction is larger than the one in a positive direction. therefore, the tensioner has a small displacement and large tension force in the negative direction. it would be beneficial for avoiding the buckling of the rise and keeping fairly tension force on the riser ring. advances in technology innovation, vol. 5, no. 3, 2020, pp. 182-189 189 5. conclusion in this study, the dynamic simulation of the marine riser system is carried out, and the dynamic response of the marine riser system is analysed. dynamic model is established including the floating platform, rise tensioner and riser ring. the simplified model of dat with pneumatic system is developed and the nonlinear stiffness of the riser tensioner is considered. through the numerical simulation, the vertical motion of the marine riser system and the tension of the riser with different wave conditions are analysed. in low-amplitude wave conditions, the system response of the model considering nonlinear stiffness of dat is similar to the result of the linear model; in high -amplitude wave conditions, the significant difference exists between the linear model and the nonlinear model. the asymmetrical response can be seen obviously in the nonlinear model. an index to represent the degree of asymmetry is carried out. the calculation suggests that with the increase in the heave motion amplitude of offshore, the asymmetrical degree will increase. the nonlinear characteristic of direct acting tensioner could be useful to prevent the buckling of the riser. in order to further improve the compensation performance of the riser tensioner, the impact of the riser hysteresis effect on the riser response should be analysed next. at the same time, most of the researches on the nonlinear stiffness characteristics of the riser tensioner are still in the theoretical research simulation stage. a test device should be established in the future to verify the effect of the tensioner nonlinear stiffness on its compensation performance in the experiment. conflicts of interest the authors declare no conflict of interest. references [1] t. wang and y. liu, “dynamic response of platform-riser coupling system with hydro-pneumatic tensioner,” ocean engineering, vol. 166, pp. 172-181, october 2018. [2] b. chen, j. yu, y. yu, l. xu, h. wu, and z. li, “modeling approach of hydro-pneumatic tensioner for top tensioned riser,” journal of offshore mechanics and arctic engineering, vol. 140, no. 5, october 2018. [3] r. g. pestana, f. e. roveri, r. franciss, and, g. b. ellwanger, “marine riser emergency disconnection analysis using scalar elements for tensioner modeling,” applied ocean research, vol. 59, pp. 83-92, september 2016. [4] j. k. woodacre, r. j. bauer, and r. a. irani, “a review of vertical motion heave compensation systems,” ocean engineering, vol. 104, pp. 140-154, august 2015. [5] y. wu, “a new operability and predictability enhanced riser control system for deepwater marine operation: an integrated riser hybrid tensioning system,” ph.d. dissertation, dept. elect. eng., the university of texas at austin, 2015. [6] z. q. xu, w. y. tang, h. x. xue, and z. q. hu, “safety assessment of a top-tensioned tlp production riser in disastrous marine environment,” china ocean engineering, vol. 28, no. 3, pp. 44-49, 2010. [7] y. peng and b. zhao, “analysis of factors influencing the crane heave compensation system compensation effect at deepwater drilling rig,” journal of residuals science & technology, vol. 13, no. 8, 2016. [8] s. liu and l. li, “control performance simulation on heave compensation system of deep-sea mining based on dynamic vibration absorber,” international conference on digital manufacturing & automation, december 2010, pp.441-445. [9] j. t. hatleskog and m. w. dunnigan, “heave compensation simulation for non-contact operations in deep water,” oceans, september 2006, pp. 1-6. [10] y. t. zhang, z. d. liu, h. jiang, g. b. wu, and z. l. zhang, “study on active force of compensation system for floating drilling platform,” oil field equipment, vol. 39, no.4, pp. 1-4, april 2010. [11] h. lee, m. i. roh, s. h. ham, and s. ha, “dynamic simulation of the wireline riser tensioner system for a mobile offshore drilling unit based on multibody system dynamics,” ocean engineering, vol. 106, pp. 485-495, september 2015. [12] s. h. ham, m. i. roh, and j. w. hong, “dynamic effect of a flexible riser in a fully connected semisubmersible drilling rig using the absolute nodal coordinate formulation,” journal of offshore mechanics and arctic engineering, vol.139, no. 5, october 2017. [13] w. shisheng, x. bin, and l. xinzhong, “the motion response analysis of deep water typical tlp in environment conditions of south china sea,” shipbuilding of china, vol. 52, no. 1, pp. 94-101, 2011. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation, vol. 5, no. 2, 2020, pp. 65-75 adaptation of the higher education in engineering to the advanced manufacturing technologies olivier bonnaud 1,2,* , ahmad bsiesy 3,4 1 department of sensors and microelectronics, ietr, university of rennes 1, 35042 rennes, france 2 national coordination for education in microelectronics, gip-cnfm, 38016 grenoble, france 3 university grenoble alpes, cnrs, ltm, f-38000 grenoble, france 4 cime nanotech, university grenoble-alpes, 38016 grenoble, france received 25 april 2019; received in revised form 07 july 2019; accepted 02 november 2019 doi: https://doi.org/10.46604/aiti.2020.4144 abstract the 21st century will be the era of the fourth industrial revolution with the progressive introduction of the digital society, with smart/connected objects, smart factories that are driven by robotics, the internet of things (iot) and artificial intelligence. manufacturing is expected to be performed by smart factories, i.e. the industry 4.0, which is the outcome of steady development of information technology associated with new objects and systems that can automatically supervise and fulfil manufacturing tasks. the industry 4.0 concept relies largely on the ability to design and manufacture smart and connected devices that are based on microelectronics technology. this evolution requires highly-skilled technicians, engineers and phds, all of them well prepared for research, development and manufacturing. their training, which combines knowledge and the associated compulsory know-how, is becoming the main challenge for the academic world. the curricula must therefore include the basic knowledge and associated know-how training in all the specialties of the field. the software and hardware tools used in microelectronics education are becoming highly complex and expensive that the most viable solution for practical training is to share technical facilities and human resources. this strategy has been adopted by the french microelectronics education network, which includes twelve universities and two industrial unions. by sharing human resources and technical facilities, the cnfm network was able to minimize the costs and to train future graduates on up-to-date tools similar to those used in companies. this paper aims to show how the strategy adopted by the french network can help meeting the needs of the future industry 4.0. several examples of innovative activities developed in this strategy will be given. keywords: industry 4.0, microelectronics, engineering education, network structuring, multidisciplinarity 1. introduction the evolution towards digital societies implies a revolution in the industry that relies largely on the tremendous growth of microelectronics, which is at the heart of the internet of things (iot) [1-2]. indeed, the rapid development of microelectronics [3] has made it possible to achieve fast computing, high-speed data processing and communications, and also more sensitive detection and more effective actuation [4]. today, all these functions are integrated in the connected objects that apply to all major societal activities [5], from communications to health, safety, transport, energy [6-7], as well as production in the new industrial scheme [8]. in fact, modern industry takes advantage of intelligent robots [9-10] that are well adapted to the * corresponding author. e-mail address: olivier.bonnaud@univ-rennes1.fr advances in technology innovation, vol. 5, no. 2, 2020, pp. 65-75 66 production environment and have a core based on connected objects. as a result, industry is increasingly dependent on iot, which is necessary to govern the design of new products, their production, as well as to control the flow of raw materials and manufactured objects. all these production and management functions are part of what is known as smart factory industry 4.0 [8]. in this field, engineers, technicians and phd grade holders must be integrated into this design and manufacturing chain. the skills of these actors must combine the knowledge and practical know-how or skills that are essential when manufacturing a real object integrated into its environment and with a well-defined mission profile [11]. electronic hardware platforms (microprocessors, microcontrollers, sensors, actuators, etc.) are at the heart of the smart factory and they are provided by microelectronics industry [12]. higher education programs in engineering must therefore contain the basic knowledge and associated know-how in the following specialties: analog and digital electronics, signal processing, sensors and actuators, on-board electronics, energy harvesting systems, communications, transmission protocols and human-machine communication systems (vision, sound, touch, etc.) [6]. these specialties must meet the future needs of a connected society: low power, high speed, reproducibility, quality, safety, reliability, and low cost. a transformation of higher education is a major challenge to meet the needs of companies and research centers with a permanent behavior oriented towards innovation. this paper addresses these different aspects. 2. microelectronics and iot as mentioned above, microelectronics and its evolution towards nanotechnologies are the driving force behind the evolution towards the digital society and future industry 4.0. over the past sixty years, the integration of elementary devices into integrated circuits, mainly transistors, has grown exponentially, primarily due to the decrease in the lateral dimension of elementary devices. this evolution followed the famous moore's law [3]. nowadays, the minimum dimension entered within the nanometre scale, which is why we are now talking about nanotechnologies. on the one hand, the reduction in size was governed by a better control of the stages of the manufacturing process gradually involving the self-assembly of nano-objects and thin films. on the other hand, the size has also decreased due to the increase in the computing capacity that led to the improvement in the design of increasingly complex circuits. the consequence of this unprecedented evolution is that the lateral dimensions of the devices are currently approaching the inter-atomic distance. thus, the decrease in the size of the elementary transistors and thus the increase in the integration density of the transistors reach a physical limit. however, since the early 2000s, the introduction of the third dimension has provided a new way to further increase integration density. indeed, this approach is based on stacking of elementary devices and circuits on the surface of the previous circuits. this is carried out by involving new deposition techniques of thin films, and the technique of stacking of the circuits themselves based on the thinning of the substrates followed by a sticking process [13-14]. fig. 1 illustrates these two developments. the number of elementary devices in the most advanced electronic systems can count up to several billions [3]. fig. 1 moore’s law [3] and “more than moore” evolution [13]. systems on chip (soc) and systems in package (sip) open the way for connected objects [14] advances in technology innovation, vol. 5, no. 2, 2020, pp. 65-75 67 thanks to this integration scheme, many electronic functions have been developed and by combining several technological processes, the extension to multi-physical systems has become possible. for example, micro-electro-mechanical systems (mems) technology opened the way to create many types of sensors and actuators. the possibility to gather several functions allows the creation of systems on chips (soc) and systems in package (sip) which are the cornerstone of new connected objects. indeed, the connected objects are a combination of several functions that are summarized in fig. 2 [5-6]. they may contain sensors, actuators, signal processing modules, analog/digital and digital/analog converters, energy harvesting cells, and communication modules. depending on the application domain, sensors and actuators are multidisciplinary [15]. with these new systems, the fields of application can cover all societal needs such as communications, health, environment, energy, transport, agriculture, and the new industry 4.0, which is covered in this document and that requires advanced manufacturing technologies. fig. 2 simplified architecture of a connected object with its main functions [6] to summarize, the microelectronics and the associated new connected objects, allow the complexity of the architecture and the systems, increasingly involve multidisciplinary approach, and can be applied to large spectrum of societal needs in the so-called internet of everything or ioe [16-17]. the human resources associated with this field, i.e. the engineers, technicians, and operators working in the manufacturing have to acquire skills corresponding to knowledge and know-how of the field, associated with capabilities of team working. 3. advanced manufacturing technologies and constraints advanced manufacturing technologies should involve many objects, including robots and connected environments. this environment must be secure and reliable in terms of physical objects, intellectual property, data protection and monitoring. for designers and manufacturers, this means a lot of works are dedicated to the safety and reliability of objects and microelectronic circuits, as well as to data transmission and the reduction of possible interferences in the signals. all these points put high challenges on microelectronic devices, circuits and systems. another aspect is less visible but of ever-increasing importance. it is the energy consumption of all these objects within the iot framework. fig. 3 shows the expected evolution of iot energy consumption over the next 20 years. growth is exponential and iot electrical power consumption is expected to reach 15,000 gw in 2040, which is the global energy consumption from all sources (nuclear, oil, gas and renewable energy) by 2018 [18], while the global average power consumption should reach about 125,000 gw [19-20]. it is clear that the planet will not be able to reach this level of production and the reasonable solution is to significantly reduce the consumption of each connected object by a factor of more than 10 and probably close to 100 after 2040. the solutions to this issue require deep changes in current microelectronics field. it is still necessary to reduce the consumption of each elementary device thanks to nanotechnology development, optimize the number of components in the advances in technology innovation, vol. 5, no. 2, 2020, pp. 65-75 68 circuits, create new architectures that sequentially activate the useful zones by putting all the others on standby, i.e. to move from synchronous architectures to asynchronous ones in order to minimize the consumption of the clocks, combine different technologies (ulsi, thin film technologies, organic technologies) and improve the transmission and reception efficiency in signal transmissions, to mention only the most obvious ones. fig. 3 projected annual energy consumption over the next 20 years, equivalent to the world average annual capacity in 2018, all energy sources cumulated (after [20]) creating electronic circuits in heterogeneous technologies and multidisciplinary applications, improving safety, controlling energy consumption are the challenges that future technicians, engineers and phds will have to face. they must therefore be prepared through initial and lifelong training adapted to this evolution. university training in engineering must therefore meet these needs. the french network for education in microelectronics and nanotechnology (cnfm) [21], created in the early 1980s and moving from microelectronics to nanotechnology in the 2000s [22], has adopted a strategy in this direction. 4. french higher education network in microelectronics 4.1. the cnfm french network higher education in microelectronics in france is coordinated by a 35-years old national network, the cnfm, funded by the ministry of higher education with an official structure called “groupement d'intérêt public” (gip) [21-22]. this network was created to share education facilities (platforms) containing expensive micro and nano-fabrication equipment and tools, that require high operating costs. these shared open facilities allow offering practical training to all academic institutions in france that become users of the platforms for training their students in know-how. foreign academic institutions are also users of the platforms in the frame of cooperation agreements. this network is composed of twelve academic members who are respectively in charge of the twelve common centers spread over the french territory as shown in fig. 4, each center being common to several academic institutions located in the same geographic area. the twelve cnfm common centers (red labels) manage platforms i.e. technical facilities for practical hands-on training. among the eighty technical platforms, 7 cleanrooms are mainly dedicated to education (red stars). national services for cad (computer-aided-design) tools provided by the montpellier cnfm common center are dedicated to all the national community and foreign partners of the network. two industrial unions and especially the most representative association acsiel alliance electronics consortium [23] are also the members of the cnfm body. they represent more than 150 industrial sites (orange and green circles). the presence of industrial partners ensures strong links between academic training and industrial needs. indeed, industrial partners provide the cnfm network with the necessary valuable advice on the nature of skills and competencies needed by the work market. hereafter, we will present what we think to be the main characteristics of a good education approach able to ensure i) a high quality of the content, ii) a know-how on updated and industrial tools, iii) an international recognition of the diploma, iv) and sustainability meeting the needs of advanced manufacturing technologies. advances in technology innovation, vol. 5, no. 2, 2020, pp. 65-75 69 fig. 4 map of the french academic and industrial activities in microelectronics [21] the main company sites are identified by orange circles. the twelve common centres of the cnfm (red circles and red stars) manage practical training platforms and are located throughout the metropolitan area. for their practice in cleanroom, students travel from their institution to closest cnfm centres (blue lines). 4.2. the missions and the strategy of the cnfm network as previously mentioned, the very essence of the gip-cnfm network is to share technical platforms, which include technological processes, cad, characterizations, and test, because of their very high equipment and operating costs. in addition, sharing pedagogical approaches between educators, and applying good practices towards students trained in initial training or even in lifelong learning (lll) is also a main educational target. collaboration within the network exists in several forms, via:  the sharing and common use of technological and design platforms,  the collaborative work in the context of national, or international, multi-year projects,  three-day educational workshops and brainstorming seminars involving teachers, researchers and technical collaborators, as well as industry representatives. these joint activities allow the exchange of knowledge and practices in order to produce and disseminate knowledge and know-how to the entire academic community [24] as well as to foster an innovative approach that is mandatory for advanced manufacturing technologies. 4.3. cnfm strategy with industry the main point of the gip-cnfm's strategy is to pool technology, design, characterisation and testing platforms in the broad field of electronics, microelectronics and nanotechnologies. this network provides a permanent link between the academic world and the industrial world that can reflect its needs in the short, medium and long term. the technological advances in technology innovation, vol. 5, no. 2, 2020, pp. 65-75 70 platforms are accessible to all students, whatever their level from undergraduate to post-doctoral students, which allows them to acquire essential know-how in addition to the theoretical knowledge acquired either in traditional form (courses) or in the form of online tools, such as moocs [25-26]. since electronics and microelectronics are at the heart of the majority of innovative connected objects, the training strategy aims to provide skills and know-how in these fields [27] but also in the fields of application such as communication, health, environment, transport, etc. [15]. the existence of multidisciplinary platforms open to initial and lifelong learning must make it possible to meet the needs of industry in terms of the quality and relevance of curricula content. the close link within the cnfm network between academia and industry makes it possible to jointly define learners' needs, build curricula adapted to the objectives and ensure learning that includes both the knowledge and know-how required for a successful 21st century industry [27]. 4.4. collaboration within the network and international extension collaboration within the network exists in several forms, on the one hand through the sharing and common use of technological and design platforms, and on the other hand through collaborative work in the context of national, or international, multi-year projects and through the organisation of steering committees, national educational workshops and reflection seminars, particularly within the framework of the network's orientation council. these joint activities allow for the exchange of knowledge and practices in order to produce and disseminate knowledge and know-how to the entire academic community. this good practice of networking is not so familiar within the french academic world and serves as an example to other fields both in france and also abroad. the use of platforms shared by several academic institutions allows users from different backgrounds and belonging to different institutions to meet and collaborate while acquiring know-how that is essential for all. 5. some answers to the needs of companies involved in manufacturing with the objective of best meeting the needs of manufacturing industry, several parameters and qualities must be taken into account following several brainstorming meetings during 2018 and 2019. their objective is to adapt the training of technicians, engineers and phds, who will have to get along well with companies in industry 4.0. in the qualities required, we find constants, new approaches, and skills. 5.1. adaptation of the content the main aim is to cover the entire design and manufacturing chain of the connected objects that will be at the heart of 21st century industry. the programmes should therefore contain the basic knowledge and associated know-how in the following specialities: analog and digital electronics, signal processing, sensors and actuators, embedded electronics, energy harvesting systems, communications, transmission protocols and human-machine communication systems (vision, sound, touch, etc.). thanks to the annual funding provided by the cnfm to innovative education projects in the frame of the finmina project [28], the platforms are permanently up-dated with new practice in accordance with the evolution of the techniques and the fields of application [29]. these specialities must meet the future requirements of a connected society, namely: low power, low consumption, speed, reliability, low cost. in fact, the 80 platforms cover the entire microelectronics spectrum. 5.2. learning environment the objective of the existing network is to provide a learning environment that allows the trainees to be immersed in a technological environment that prepares them for the industrial environment of the 21st century and more particularly for advances in technology innovation, vol. 5, no. 2, 2020, pp. 65-75 71 industry 4.0. indeed, the current centres include seven cleanrooms that condition users to the requirements of advanced technologies, namely, cleanness concept, adapted clothing, handling with dedicated tools or robots. indeed, the current centres include seven clean rooms that show the users the requirements of advanced technologies: cleanliness concept, adapted clothing, handling with dedicated tools and robots. they allow the use of high chemical purity products to minimize the contamination of integrated electronic devices, and the use of complex equipment that allows the creation of thin layers with thickness control at the atomic scale. the platforms associated with these cleanrooms allow a multidisciplinary orientation by the realization of mems (micro-electrical-mechanical-systems), biomems (biological mems), omems (optical mems), intended for sensors or actuators of connected objects [4]. computer-aided design of electronic systems uses industrial software to acquire concepts that capture high-level abstract models. these electronic systems are usually described by the means of advanced computer languages (such as vhdl) that design the functionalities, and non-functional specifications such as electrical consumption, safety, reliability and performance. due to the complexity of integrated systems, design requires the juxtaposition of many skills and know-how, which implies teamwork and therefore collective problem solving. team project training must meet these requirements and becomes an essential form of the pedagogical approach. within the network, all colleagues apply these principles in the educational programmes of their partner institutions, which automatically implement an experimental approach. 5.3. delivery mechanisms as already mentioned above, the first objective of the network is to provide the corresponding knowledge and know-how. given the organisation of the network, teachers and learners interact face-to-face on concrete issues. indeed, any practical activity is closely supervised because of the complexity, impact and dangerousness of the equipment used. this strong support in practical activities can be compensated by more autonomy in acquiring basic knowledge using online tools such as moocs [26, 30], serious games [31] and flipped classrooms [32]. these tools can also be adapted to prepare practical activities (practical training, supervised projects, etc.) in order to optimize the presence on the platforms and minimize their cost. the introduction of virtual and augmented reality or artificial intelligence can be considered as part of a constructive cooperation between the existing cnfm network and other networks with these competences. 5.4. assessment since practical activities in a high-tech environment are automatically supervised by a tutor or teacher, the evaluation must first be done by the supervisor. however, as these activities are generally carried out in groups, it is quite possible to exploit the group dynamic and its self-assessment, which can be included in experience reports drawn up at the end of their practical training on platforms. 5.5. recognition the official cnfm training network is made up of service units (or cnfm poles), which are not a priori authorised to deliver a diploma. diplomas, certificates or titles are awarded by the academic institutions to which the student users are attached. however, as part of lifelong learning, certificates may be issued specifying the objectives, content and quality of the work carried out on the cnfm platforms. it is expected that the network will become a certification body specifically designed for the training of employees of companies using the network's resources. 5.6. quality the content of the essentially practical activities are established in connection with academic training on the one hand and with companies for their employees on the other. this approach ensures content quality. at the end of practical training, advances in technology innovation, vol. 5, no. 2, 2020, pp. 65-75 72 particularly for continuing training (lll), questionnaires are drawn up to check that the actual training is in line with the objectives defined in advance and that the trainees' perception is correct. a debriefing at the end of the training session should also allow trainers to justify certain approaches that would not have been perceived in the same way by users. 5.7. sustainability of the education network the cnfm network's strategy, which makes it possible to pool resources and maintain good matching with the needs of the socio-economic world, can only be achieved by keeping platform equipment and study topics at the highest level. the field of microelectronics has been evolving very rapidly for decades and training must also keep the pace. thus, this policy requires financial support to update the hardware and software that academic bodies currently have difficulty in assuming despite the pooling organised by the network, which however makes it possible to limit the cost. it is necessary to consider at national and european levels, that the training needs of companies are specifically supported within the framework of industrial sectors with dedicated funding from both public bodies and companies. 6. adaptation of the technical content by the cnfm network 6.1. new technical challenges as a result of the analysis of the context, the technical challenges that the future engineers and doctors will have to face appear at different levels in order to improve the electrical behaviour and the electrical energy consumption of the future connected objects included in robots or cobots, and involved in the advanced manufacturing. their interventions should occur at the level of the elementary devices, of electronic functions and circuits and at the level of systems, as follows:  continuation of the miniaturization of elementary devices, minimizing consumption both in the on-state (ron) and off-states (leakage),  reduction of switching and static losses of power components,  introduction of the third dimension to improve circuit integration in minimizing interconnection losses and improving reliability,  new circuit architectures able to control active and stand-by zones,  new circuit concepts allowing the generalization of asynchronous control of all elementary electronic function, minimizing the synchronous power supply by the clocks of all transistors,  implementation of low temperature and large area technologies,  development of low power sensors and actuators,  optimization of communication devices and protocols to limit the occupation of frequency bands and the data flow. it is clear that all these topics will need capabilities of team working and some multidisciplinary skills for the actors [34]. the choice of the practice of the network thanks to the annual calls organized by the network council is deliberately oriented towards these technical challenges. in the following we give several examples adapted to this strategy. 6.2. example of innovative practice provided by the cnfm centres several examples of realization by students in the cnfm centres during the last years are given. the scope is wide and we highlight only a few images that are representative of the orientation towards advanced manufacturing technologies. fig. 5 shows the designed and fabricated objects and mentions the corresponding cnfm centre of realization. advances in technology innovation, vol. 5, no. 2, 2020, pp. 65-75 73 fig. 5 examples of objects, designed, fabricated and characterized in the microelectronics centres these examples show the complexity of the practice training on the different platforms, the multidisciplinarity of the realizations which are adapted to the local context of the microelectronic centres: computer-aided-design of new architectures for low consumption circuits [33], new sensors and micro-electrical-mechanical systems (mems), new nano-devices involving nanowire-based transistors and two-dimensional graphene-based transistors, radio-frequency waveguides for transceivers, electronics on plastics (plastronics), and lab-on-chip. depending on the complexity of the realization, the students can acquire the knowledge in less than one week in general initial formation, or during several weeks in the frame of projects that are more specialized. the students are using the design tools and the clean-room and characterization facilities of the different centres and of the national services for cad software [34]. 6.3. example of innovative practice provided by the centers since 2011-2012, more than 100 innovative projects have been launched by the cnfm centres. the users, mostly master's and engineering degree students, have experience in practical training and more particularly on innovative subjects. fig. 6 shows the evolution of the network activities over the past seven years. fig. 6 innovative practice on the platforms of the cnfm network over the past six academic years thanks to the national finmina programme [28], more than 6,000 students have acquired experience and know-how on new technological or design tools that meet the new needs of industrial and societal applications. about 1,000 are users of the platforms during their phd, and about 300, mainly employees of companies, are in lifelong learning in a thematic reconversion cycle. advances in technology innovation, vol. 5, no. 2, 2020, pp. 65-75 74 7. conclusions the emergence of advanced manufacturing technologies in industry 4.0 requires an increasing adaptation of the training of engineers and technicians in the field of microelectronics to ensure that they acquire the skills and know-how necessary for the ever-changing iot technologies. the adaptation of training, which must meet quality and efficiency criteria, to the acquisition of skills and know-how is essential to meet economic, industrial and social needs [35]. the french microelectronics and technology training network, with the help of its twelve common platforms with a national mission, enables innovative practical activities to be set up using high-performance tools. this approach, considered original by the international academic community and highly appreciated by french industry, can serve as an example for other foreign countries. acknowledgment the authors want to thank all the members of the french gip-cnfm network for they contribution to many innovative realizations. this work is financially supported by french higher education ministry and by idefi-finmina program (anr-11-idfi-0017). a special thanks to l. chagoya-garzon, secretary of gip-cnfm for her fruitful advice for the proof reading of this paper. conflicts of interest the authors declare no conflict of interest. references [1] h. chaouchi, “the internet of things: connecting objects,” iste ltd, john wiley & sons, may 2010. [2] m. burgess, “what is the internet of things?,” wired magazine, https://www.wired.co.uk/article/internet-of-things-what-is-explained-iot, february 16, 2018. [3] g. e. moore, “cramming more components onto integrated circuits,” electronics magazine, vol. 38, no. 8, pp. 114-117, 1965. [4] v. sharma and r. tiwari, “a review paper on “iot” & its smart applications,” international journal of science, engineering and technology research, vol. 5, no. 2, february 2016. [5] a. schütze, n. helwig, and t. schneider,“sensors 4.0-smart sensors and measurement technology enable industry 4.0,” journal of sensors and sensor systems, vol. 7, no. 1, pp. 359-371, may 2018. [6] o. bonnaud, “new approach for sensors and connecting objects involving microelectronic multidisciplinarity for a wide spectrum of applications,” international journal of plasma environmental science and technology, vol. 10, no. 2, pp. 115-120, december 2016. [7] g. matheron, “microelectronics evolution, keynote, european, microelectronics summit,” paris, france, november 2014. [8] b. p. santos, f. c. santos, and t. m. lima, “industry 4.0: an overview,” proceedings of the world congress on engineering, vol. 2, 2018. [9] m. peshkin and j. e. colgate, “cobots,” industrial robot, vol. 26, no. 5, pp. 335-341, 1999. [10] å . f. berglund, f. palmkvist, p. nyqvist, s. ekered, and m. å kerman, “evaluating cobots for final assembly,” procedia cirp, vol. 44, pp. 175-180, 2016. [11] o. bonnaud and l. fesquet, “practice in microelectronics education as a mandatory complement to the future numeric-based pedagogy: a strategy of the french national network,” international kes conference on smart education and smart e-learning, springer, cham, pp. 1-8, may 2016. [12] m. mckellop, “the future of the semiconductor industry is the internet of things,” the burn-in, https://www.theburnin.com/industry/semiconductors-internet-of-things/, november 28, 2018. [13] r. r. tummala and m. swaminathan, system on package, miniaturization of the entire system, mcgraw-hill education, the1st edition, may 2008. [14] t. simonite, “moore’s law is dead. now what?,” mit technology review, https://www.technologyreview.com/s/601441/moores-law-is-dead-now-what/, may 13, 2016. advances in technology innovation, vol. 5, no. 2, 2020, pp. 65-75 75 [15] o. bonnaud, “the multidisciplinary approach: a common trend for ulsi and thin film technology,” ecs transaction, vol. 67, no. 1, pp. 147-158, 2015. [16] technopedia, “internet of everything (ioe),” https://www.techopedia.com/definition/30121/internet-of-everything-ioe, 2019. [17] idc, “what the iot market has in store for europe in 2019,” https://uk.idc.com/trends/predictions, january 30, 2019 [18] enerdata, “global energy statistical yearbook 2019,” https://yearbook.enerdata.net/total-energy/world-energy-production.html, march 2019. [19] international energy agency, “world energy outlook 2019,” https://www.iea.org, january 2019. [20] enerdata, “world energy production,” https://yearbook.enerdata.net/total-energy/world-energy-production.html, 2019. [21] cnfm, “coordination nationale pour la formation en microélectronique et nanotechnologies,” www.cnfm.fr; gip-cnfm, march 2019. [22] o. bonnaud, p. gentil, a. bsiesy, s. retailleau, e. d. gergam, and j. m. dorkel,“gip-cnfm: a french education network moving from microelectronics to nanotechnologies,” proceeding of ieee global engineering education conference; amman, jordan, pp. 122-127, april 2011. [23] acsiel alliance é lectronique, “components and systems alliance for electronics industry in france,” http://www.acsiel.fr/en-gb/index.aspx, 2019. [24] o. bonnaud, “practice-oriented pedagogical strategy of the french microelectronics and nanotechnologies network,” 14th international conference on education and educational technology, kuala-lumpur, malaysia, pp. 11-19, april 2015. [25] a. fox, “from moocs to spocs,” communications of the acm, vol. 56, no. 12, pp. 38-40, 2013. [26] o. bonnaud and l. fesquet, “mooc and the practice in electrical and information engineering: complementary approaches,” 15th international conference on information technology based higher education and training, istanbul, turkey, september 2016. [27] o. bonnaud, “knowledge and know-how in microelectronics; strategy of innovative practice to balance the new on-line courses approach,” international conference on advanced technology innovation, koh samui, thailand, pp. 10-11, june 2017. [28] finmina, “formations innovantes en microélectronique et nanotechnologies,” http://www.cnfm.fr/?q=fr/pr%c3%a9sentation_finmina, innovative training for microelectronics and nanotechnologies march 2019. [29] o. bonnaud, l. fesquet, p. nouet, and t. m. brahim, “finmina: a french national project to promote innovation in higher education in microelectronics and nanotechnologies,” international conference on information technology based higher education and training, york, uk, proceeding of ithet 2014, september 11-13, 2014. [30] o. bonnaud, “new vision in microelectronics education: smart e-learning and know-how, a complementary approach,” international kes conference on smart education and smart e-learning, pp. 267-275, 2019. [31] f. bruguier, p. benoit, l. dalmasso, b. pradarelli, and l. torres, “amuse: un escape game pour l’enseignement de la sécurité numérique,”, proceeding of 15th jpcnfm 2018, saint-malo, france, vol.15, november 2018. (french) [32] o. bonnaud, y. danto, y.kuang,and y. li, “international flipped class for chinese honors bachelor students in the frame of multidisciplinary fields: reliability and microelectronics,” advanced in technology innovation, vol. 3, no. 3, pp. 126-132, 2018. [33] o. bonnaud and l. fesquet, “innovation for education on internet of things,” proceeding of engineering and technology innovation, vol. 9, pp. 1-8, 2018. [34] o. bonnaud and l. fesquet, “communicating and smart objects: multidisciplinary topics for the innovative education in microelectronics and its applications,” proceeding of 15th international conference information technology based higher education and training, ieee press, pp. 1-5, august 2015. [35] o. bonnaud, “mandatory matching between microelectronics industry and higher education in engineering toward a digital society,” smart education and e-learning 2019, springer, singapore, vol. 24, pp. 255-266, 2020. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation, vol. 4, no. 3, 2019, pp. 140-151 numerical analysis of an industrial-scale steam methane reformer chun-lang yeh * department of marine engineering, national taiwan ocean university, keelung, taiwan received 02 march 2018; received in revised form 24 june 2018; accepted 29 july 2018 abstract a steam reformer of a hydrogen plant is a device that supplies heat to convert the natural gas or liquid petroleum gas into hydrogen via catalysis. it has been often used in the petrochemical industry to produce hydrogen. control of the catalyst tube temperature is a fundamental demand of the reformer design because the tube temperature must be maintained within a range that the tube has minor damage and the catalysts have a high activity to convert the natural gas or liquid petroleum gas into hydrogen. in this research, the effect of the burner on/off manners on the catalyst tube temperature and the hydrogen yield of an industrial-scale side-fired steam methane reformer is investigated. the aim is to seek a feasible burner on/off manner that has acceptable catalyst tube temperature and hydrogen yield, so as to improve the performance and service life of a steam methane reformer. it is found that when one group of burners is turned off, the outer surface temperatures of the tubes are decreased by about 94°c in average, the inner surface temperatures are decreased by about 54°c in average, and the hydrogen yields are decreased by about 4%. when two groups of burners are turned off, the outer surface temperatures of the tubes are decreased by about 175°c in average, the inner surface temperatures are decreased by about 106°c in average, and the hydrogen yields are decreased by about 7.9%. when three groups of burners are turned off, the outer surface temperatures of the tubes are decreased by about 251°c in average, the inner surface temperatures are decreased by about 151°c in average, and the hydrogen yields are decreased by about 11.4%. the catalyst tube temperatures and the hydrogen yields reduce to a greater extent in regions where burners are turned off. when the central groups of burners are turned off, the tube temperatures and the hydrogen yields have greater reductions. on the other hand, when the rear groups of burners are turned off, the tube temperatures and the hydrogen yields have lower reductions. keywords: hydrogen, steam methane reformer, catalyst tube, burner operation 1. introduction a steam reformer of a hydrogen plant is a device that supplies heat to convert the natural gas or liquid petroleum gas into hydrogen via catalysts. hydrogen is an important material for petroleum refineries. it converts crude oil into products with high economic value, e.g., gasoline, jet fuel, and diesel. hydrogen can be produced by a number of ways, e.g., electrolysis, steam methane reforming (smr), partial oxidation reforming, nuclear energy, etc. [1,2] among these ways, smr is the most common commercial method of industrial hydrogen production. smr reaction mainly includes the following three chemical equations: 4 2 2 ch h o co 3h  (1) 2 2 2 co h o co h  (2) 4 2 2 2 ch 2h o co 4h  (3) * corresponding author. e-mail address: clyeh@nfu.edu.tw advances in technology innovation, vol. 4, no. 3, 2019, pp. 140-151 141 the first (smr) and the third reactions are endothermic, the second reaction (water-gas shift (wgs)) is exothermic, and the overall reaction is endothermic. the combustion process in a reformer provides heat to maintain the reforming reaction in a catalyst tube. control of the catalyst tube temperature is a fundamental demand of the reformer design because the tube temperature must be maintained within a range that the tube has minor damage and the catalysts have a high activity to convert the natural gas or liquid petroleum gas into hydrogen. when a steam reformer is operating, the catalyst tubes are subjected to stresses close to the ultimate stress of the tube material. this leads to an acceleration of the creep damage. safety, reliability, and efficiency are the basic requirements of the reformer operation. the catalyst tube should have uniform heat distribution to extend the tube service life. however, the heat distribution in a reformer is practically non-uniform. in addition, maldistribution of the flue gas and fuel gas may result in flame impingement on the catalyst tubes and lead to localized hot spots and high tube wall temperature. these factors may shorten tube life. owing to the rapid development in computer science and technology, as well as the improvements in physical models and numerical methods, computational fluid dynamics (cfd) is widely used in analyzing systems involving heat transfer, fluid flow, and chemical reactions. cfd is also used to simulate systems that cannot be measured easily or simulated experimentally. in recent years, there have been a lot of smr researches using cfd. tran et al. [3] developed a cfd model of an industrial-scale steam methane reformer. the authors pointed out that the reformer cfd model can be considered an adequate representation of the on-line reformer and can be used to determine the risk of operating the online reformer at unexplored and potentially more beneficial operating conditions. di carlo et al. [4] investigated numerically a pilot scale bubbling fluidized bed se-smr (sorption enhanced steam methane reforming) reactor by means of a two-dimensional cfd approach. the numerical results show quantitatively the positive influence of carbon dioxide sorption on the reforming process at different operating conditions, specifically the enhancement of hydrogen yield and reduction of methane residual concentration in the reactor outlet stream. lao et al. [5] developed a cfd model of an industrial-scale reforming tube using ansys fluent with realistic geometry characteristics to simulate the transport and chemical reaction phenomena with the approximate representation of the catalyst packing. the authors analyzed the real-time regulation of the hydrogen production by choosing the outer wall temperature profile of the reforming tube and the hydrogen concentration at the outlet as the manipulated input and controlled output, respectively. mokheimer et al. [6] presented modeling and simulations of the smr process. the model was applied to study the effect of different operating parameters on the steam and methane conversion. the results showed that increasing the conversion thermodynamic limits with the decrease of the pressure results in a need for long reformers so as to achieve the associated fuel reforming thermodynamics limit. it is also shown that not only increasing the steam to methane molar ratio is favorable for higher methane conversion but the way the ratio is changed also matters to a considerable extent. ni [7] developed a 2d heat and mass transfer model to investigate the fundamental transport phenomenon and chemical reaction kinetics in a compact reformer for hydrogen production by smr. parametric simulations were performed to examine the effects of permeability, gas velocity, temperature, and rate of heat supply on the reformer performance. it was found that the reaction rates of smr and wgs are the highest at the inlet but decrease significantly along with the reformer. increasing the operating temperature raises the reaction rates at the inlet but shows very small influence downstream. ebrahimi et al. [8] applied a three-dimensional zone method to an industrial fired heater of the smr reactor. the effect of emissivity, extinction coefficient, heat release pattern and flame angle on the performance of the fired heater are presented. it was found that decreasing the extinction coefficients of combustion gases by 25% caused a 2.6% rise in the temperature of heat sink surfaces. seo et al. [9] investigated numerically a compact smr system integrated with a wgs reactor. heat transfer to the catalyst beds and the catalytic reactions in the smr and wgs catalyst beds were investigated. the effects of the cooling heat flux at the outside wall of the system and steam-to-carbon ratio were also examined. it was found that as the cooling heat flux increases, both the methane conversion and carbon monoxide content are reduced in the smr bed and the carbon monoxide conversion is advances in technology innovation, vol. 4, no. 3, 2019, pp. 140-151 142 improved in the wgs bed. in addition, both methane conversion and carbon dioxide reduction increase with increasing steam-to-carbon ratio. in this paper, the transport and chemical reaction in an industrial-scale steam methane reformer is simulated using cfd. the influence of burner on/off manners on the catalyst tube temperature and the hydrogen yield of an industrial-scale side-fired steam methane reformer is investigated. the aim is to seek a feasible burner on/off manner that has acceptable catalyst tube temperature and hydrogen yield, so as to improve the performance and service life of a steam methane reformer. 2. numerical methods and physical models in this study, the ansys fluent v.17 commercial code [10] is employed to simulate the reacting and fluid flow in a steam methane reformer. the simple algorithm by patankar [11] is used to solve the governing equations. the discretizations of convection terms and diffusion terms are carried out by the second-order upwind scheme and the central difference scheme, respectively. in respect to physical models, by considering the accuracy and stability of the models and by referring to the other cfd researches [3-5] of steam methane reformers, the standard k-ε model [12], discrete ordinate (do) radiation model [13] and finite rate/eddy dissipation (fred) model [14] are adopted for turbulence, radiation and chemical reaction simulations, respectively. the standard wall functions [15] are used to resolve the flow quantities (velocity, temperature, and turbulence quantities) at the near-wall regions. for the steady-state three-dimensional flow field with chemical reaction in this study, the governing equations include the continuity equation, momentum equation, turbulence model equation (k-ε model), energy equation, radiation model equation (discrete ordinates radiation model), and chemical reaction model equation (fred model). among these models, only the fred chemical reaction model is described below while the others and the convergence criterion are not described because they have been introduced in the author’s previous study [16]. consider the general form of the r th chemical reaction as follows: f ,r b ,r k n n i,r i i,r ii 1 i 1 k         (4) where n = number of chemical species in the system 𝜈𝑖,𝑟 ′ = stoichiometric coefficient for reactant i in reaction r 𝜈𝑖,𝑟 ′′ = stoichiometric coefficient for product i in reaction r μi = species i 𝑘𝑓,𝑟 = forward rate constant for reaction r 𝑘𝑏,𝑟 = backward rate constant for reaction r eq. (4) is valid for both reversible and non-reversible reactions. for non-reversible reactions, the backward rate constant, kb,r, is omitted. the species transport equation of a chemical reaction system can be written as   ti i,m i i i t y d y r s sc               (5) where yi, di,m, sct, ri, and si are the mass fraction, diffusion coefficient, turbulent schmidt number, net generation rate, and extra source term of species i, respectively. the net source of chemical species i due to reaction is computed as the sum of the arrhenius reaction sources over the nr reactions that the species participate in �̂� advances in technology innovation, vol. 4, no. 3, 2019, pp. 140-151 143 rn i w,i r 1 i,r ˆr m r    (6) where mw, i is the molecular weight of species i and �̂�𝑖.𝑟 is the arrhenius molar rate of creation/destruction of species i in reaction r. for a non-reversible reaction, the molar rate of creation/destruction of species i in reaction r is given by     j,r j,rn i,r i,r i,r f ,r j 1 j,r r̂ k c               (7) for a reversible reaction,   j,r j,rn ni,r i,r i,r f ,r j 1 j,r b,r j 1 j,rr̂ k c k c                     (8) where 𝐶𝑗,𝑟 = molar concentration of species j in reaction r (kgmol/m 3 ) 𝜂𝑗,𝑟 ′ = rate exponent for reactant species j in reaction r 𝜂𝑗,𝑟 ′′ = rate exponent for product species j in reaction r γ = net effect of third bodies on the reaction rate the forward and backward rate constants for reaction r, kf,r and kb,r, are computed using the arrhenius expression: r er / rt f ,r r k t e     (9) f ,r b,r r k k k  (10) where ar = pre-exponential factor (consistent units) βr = temperature exponent (dimensionless) er = activation energy for the reaction (j/kgmol) r = universal gas constant (j/kgmol-k) kr is the equilibrium constant for the r th reaction and is computed from   n i ,r i ,ri 1 o o atmr r r s h k exp r rt rt                    (11) where patm denotes atmospheric pressure (101, 325 pa). the term within the exponential function represents the change in gibbs free energy, and its components are computed as follows:   o o n r i i,r i,r i 1 s s r r      (12)   o o n r i i,r i,ri 1 h h rt rt      (13) where si o and hi o are the standard-state entropy and standard-state enthalpy (heat of formation). in this study, the kinetic and thermodynamic constants for reactions (1-3) used in chibane and djellouli’s work [17] are adopted. in general, a reformer operates at high temperatures. for a non-premixed reaction, turbulence mixes the reactants and then advects the mixture to the reaction zone for quick reaction. for a premixed reaction, turbulence mixes the lower -temperature reactants and the higher-temperature products and then advects the mixture to the reaction zone for a quick reaction. therefore, advances in technology innovation, vol. 4, no. 3, 2019, pp. 140-151 144 the chemical reaction is generally mixing (diffusion) controlled. however, the flue gas, fuel gas, and air are generally premixed before injecting into the reformer. although the chemical reaction in most regions in a reformer is mixing controlled, in some regions, e.g., the neighborhood of the feed inlet, the chemical reaction is kinetically controlled. in existing chemical reaction models, the eddy dissipation model (edm) [18] can consider simultaneously the diffusion controlled and the kinetically controlled reaction rates. in edm, the net generation rate of species i in the r th chemical reaction is found from the smaller value of the following two reaction rates: r i,r i,r w ,i r r ,r w,r y r m min k m          (14) p p i,r i,r w ,i n j,r w , jj y r m k m        (15) where yp is the mass fraction of any product, p yr is the mass fraction of a particular reactant, r a is an empirical constant equal to 4.0 b is an empirical constant equal to 0.5 in general, the edm works well for a non-premixed reaction. however, for a premixed reaction, the reaction may start immediately when injecting into a reformer. this is unrealistic in practical situations. to overcome this unreasonable phenomenon, ansys fluent provides another model, the finite-rate/eddy-dissipation (fred) model, which combines the finite-rate model and the edm. in this model, the net generation rate of a species is taken as the smaller value of the arrhenius reaction rate and the value determined by edm. the arrhenius reaction rate plays the role of a switch to avoid the unreasonable situation that the reaction starts immediately when injecting into a reformer. once the reaction is activated, the eddy-dissipation rate is generally lower than the arrhenius reaction rate, and the reaction rate is then determined by the edm. 3. results and discussion to validate the numerical methods and physical models used in this study, an industrial-scale steam methane reformer is simulated. the configuration and dimension of the reformer investigated are shown in fig. 1. note that only one half of the reformer is simulated due to its symmetry, as shown in figure 1b,c. the reformer contains 138 reforming tubes and 216 burners on one side (totally 276 tubes and 432 burners). the outer diameter and thickness of a reforming tube are 136mm and 13.4 mm, respectively, while the diameter of a burner is 197 mm. the boundary conditions for the numerical model of the steam methane reformer are described below. these conditions are practical operating conditions that are used by a petrochemical corporation in taiwan. (1) symmetry plane: symmetric boundary condition (2) wall: standard wall function (3) reforming tube inlet: v = 5.4 m/s (in axial direction) t = 912.75 k pgauge = 2.1658 × 10 6 n/m 2 advances in technology innovation, vol. 4, no. 3, 2019, pp. 140-151 145 species mole fraction: ch4 = 0.2029 h2o = 0.6 h2 = 0.12855 co2 = 0.06565 co = 0.00145 n2 = 0.00145 (a) a typical steam methane reformer (b) a numerical model of the steam methane reformer (c) the dimension of the steam methane reformer (d) illustration of the burner positions (e) illustration of the reforming tube positions fig. 1 configuration and dimension of the steam methane reformer investigated (4) reforming tube exit the diffusion flux for all flow variables in the outflow direction are zero. in addition, the mass conservation is obeyed at the exit. advances in technology innovation, vol. 4, no. 3, 2019, pp. 140-151 146 (5) fuel and flue gas inlet (burner inlet): v = 2.404 m/s (in radial direction) t = 673.15 k pgauge = 1.04544 × 10 4 n/m 2 species mole fraction: h2 = 0.0816 ch4 = 0.0474 n2 = 0.49057 o2 = 0.12818 co2 = 0.25225 (6) fuel and flue gas exit: the diffusion flux for all flow variables in the outflow direction are zero. in addition, the mass conservation is obeyed at the exit. the turbulence kinetic energy is 10% of the inlet mean flow kinetic energy and the turbulence dissipation rate is computed using eq. (16). l k c 2/3 4/3   (16) where l = 0.07 l and l is the hydraulic diameter. 3.1. comparison of numerical results with experimental data the simulation results are compared with the experimental data from a petrochemical refinery in taiwan to evaluate the numerical methods and physical models adopted in this study. the temperature of the reformer is detected by an infrared thermographer with radiation emissivity of 0.92, the field of view (fov) 24mm and object distance of 5m. as mentioned above, the real reformer contains 138 reforming tubes and 216 burners on one side (totally 276 tubes and 432 burners). to save simulation time, a simplified model is also calculated and compared. the simplified model contains 6 tubes and 12 burners on one side of the reformer. the arrangement of reforming tubes and burners as well as their dimensions for the simplified model is shown in fig. 2. the flow rates in the reforming tubes and burners for the simplified model are the same as those in the real reformer. therefore, the reforming tubes and burners for the simplified model have larger diameters. fig. 2 the arrangement of reforming tubes and burners as well as their dimensions for the simplified model advances in technology innovation, vol. 4, no. 3, 2019, pp. 140-151 147 the computer used in this study is an asus esc-500-g4 work station of 8 cores with intel core i7-6700 cpu and 64 gb ram. the numbers of cfd cells for the real reformer model and the simplified model are around 4 million and 1.5 million, respectively. the grid mesh is generated by the software gambit and is unstructured. the dimensionless distance from the wall, y*, in the wall function method has been examined after a converged solution is obtained. it was found that the values of y* for the nodes at the wall to their nearest interior nodes vary between 20.0 and 60.0, and lie in the logarithmic layer of the wall function method. this implies that the wall-adjacent cells of the grid mesh in this study are suitable for the use of wall function. the solution of the cfd model for the real reformer is obtained after approximately 30 full days while that for the simplified reformer is approximately 10 full days. fig. 3 compares the average temperatures at the outer surfaces of the reforming tubes. it can be seen that the simulation result agrees well with the experimental data. the deviations from the experimental data using the real reformer model and the simplified model are 2.88% and 3.18%, respectively, which are both acceptable from a viewpoint of engineering applications. the result calculated from the real reformer model agrees better with the experimental data than that from the simplified model, although the latter also gives an acceptable result. fig. 4 shows the simulated hydrogen yield at the reforming tube outlets using a simplified model. the experimental value of the hydrogen yield is 0.698. the simulated value is 0.708. the deviation of the cfd simulation is 1.43%. in the subsequent discussion, the simplified model is used for the parametric study to save simulation time. fig. 3 comparison of the average temperatures at the outer surfaces of the reforming tubes fig. 4 the simulated average mole fraction of hydrogen at the reforming tube outlets 3.2. effect of the burner on/off manner to explore the effect of the burner on/off manner on the thermal field and hydrogen yield, the burners are divided into six groups. the first group ranges from x=0 to x=6.5m, the second group ranges from x=6.5m to x=12.67m, the third group ranges from x=12.67m to x=18.84m, the fourth group ranges from x=18.84m to x=25.01m, the fifth group ranges from x=25.01m to x=31.18m, and the sixth group ranges from x=31.18m to x=37.68m. each group of burners can be controlled on or off. in the following discussion, 12 different manners of the burner on/off are discussed. x (m) t ( c ) 0 5 10 15 20 25 30 35 0 200 400 600 800 1000 experimental data (for tube outer surface) o real reformer model simplified model x (m) 5 10 15 20 25 30 0 0.2 0.4 0.6 0.8 1 y h2 advances in technology innovation, vol. 4, no. 3, 2019, pp. 140-151 148 fig. 5 and 6 compare the average temperatures at the inner and outer surfaces, respectively, of the reforming tubes using different manners of the burner on/off. from the simulation results, it is seen that temperature profiles are obviously influenced by the manners of the burner on/off. when one group of burners is turned off, the outer surface temperatures of the tubes are decreased by about 94°c on average, and the inner surface temperatures are decreased by about 54°c on average. when two groups of burners are turned off, the outer surface temperatures of the tubes are decreased by about 175°c on average, and the inner surface temperatures are decreased by about 106°c on average. when three groups of burners are turned off, the outer surface temperatures of the tubes are decreased by about 251°c on average, and the inner surface temperatures are decreased by about 151°c on average. the tube temperatures reduce in regions where burners are turned off. when the central groups of burners are turned off, the tube temperatures have greater reductions. on the other hand, when the rear groups of burners are turned off, the tube temperatures have lower reductions. the results can also be observed in table 1. (a) group 1 turned off (b) group 2 turned off (c) group 3 turned off (d) group 4 turned off (e) group 5 turned off (f) group 6 turned off (g) groups 1 and 2 turned off (h) groups 3 and4 turned off (i) groups 5 and 6 turned off (j) groups 1, 2 and 3 turned off (k) groups 4, 5 and 6 turned off fig. 5 comparison of the average temperatures at the outer surfaces of the reforming tubes x (m) t ( c ) 5 10 15 20 25 30 35 0 200 400 600 800 1000 o fully opened group 1 closed x (m) t ( c ) 5 10 15 20 25 30 35 0 200 400 600 800 1000 o fully opened group 2 closed x (m) t ( c ) 5 10 15 20 25 30 35 0 200 400 600 800 1000 o fully opened group 3 closed x (m) t ( c ) 5 10 15 20 25 30 35 0 200 400 600 800 1000 o fully opened group 4 closed x (m) t ( c ) 5 10 15 20 25 30 35 0 200 400 600 800 1000 o fully opened group 5 closed x (m) t ( c ) 5 10 15 20 25 30 35 0 200 400 600 800 1000 o fully opened group 6 closed x (m) t ( c ) 5 10 15 20 25 30 35 0 200 400 600 800 1000 o fully opened group 1&2 closed x (m) t ( c ) 5 10 15 20 25 30 35 0 200 400 600 800 1000 o fully opened group 3&4 closed x (m) t ( c ) 5 10 15 20 25 30 35 0 200 400 600 800 1000 o fully opened group 5&6 closed x (m) t ( c ) 5 10 15 20 25 30 35 0 200 400 600 800 1000 o fully opened group 1,2&3 closed x (m) t ( c ) 5 10 15 20 25 30 35 0 200 400 600 800 1000 o fully opened group 4,5&6 closed advances in technology innovation, vol. 4, no. 3, 2019, pp. 140-151 149 for a reformer operating at a high temperature, the heat transfer to the catalyst tubes comes primarily from the radiation of the fired walls and the combustion gas, and secondarily from the convection of the combustion gas. in terms of radiation, the radiation intensity from the middle groups of burners is higher than that from the side groups of burners. this is because the view factors among the middle groups of burners and the reformer tubes are larger than those among the side groups of burners and the reformer tubes. in terms of convection, turning off the upstream groups of burners can alleviate convection to the downstream region and hence can reduce the tube temperature to a higher extent. on the contrary, turning off the downstream groups of burners has little influence on the convection to the upstream region and hence can reduce the tube temperature only to a lower extent. table 1 comparison of the average temperatures at the reforming tube surface operating manner of the burners the average temperature of tube outer surfaces ( o c) the average temperature of tube inner surfaces ( o c) fully opened 866 761 group 1 turned off 770 706 group 2 turned off 769 704 group 3 turned off 767 697 group 4 turned off 768 701 group 5 turned off 776 710 group 6 turned off 780 727 group 1&2 turned off 690 654 group 3&4 turned off 684 649 group 5&6 turned off 699 661 group 1,2&3 turned off 609 605 group 4,5&6 turned off 621 616 the above result can also be observed from fig. 7 which compares the average mole fractions of hydrogen at the reforming tube outlets using different manners of the burner on/off. it can be found from fig. 7 that the simulated hydrogen yield is 0.708 for the case of burners fully opened. the real value of the hydrogen yield is 0.698. the deviation of the cfd simulation is 1.43%. it is also observed that the hydrogen yields reduce in regions where burners are turned off. the more the burners are turned off, the greater the reduction in hydrogen yields will be. table 2 shows the comparison of the average mole fractions of hydrogen at the reforming tube outlets using different manners of the burner on/off. it is observed that when one group of burners is turned off, the hydrogen yields are decreased by about 4%. when two groups of burners are turned off, the hydrogen yields are decreased by about 7.9%. when three groups of burners are turned off, the hydrogen yields are decreased by about 11.4%. when the central groups of burners are turned off, the hydrogen yields have greater reductions. on the other hand, when the rear groups of burners are turned off, the hydrogen yields have lower reductions. fig. 6 comparison of the average mole fractions of hydrogen at the reforming tube outlets x (m) 5 10 15 20 25 30 35 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1yh2 fully opened group 2 closed group 1 closed group 4,5&6 closed group 1,2&3 closed group 5&6 closed group 3&4 closed group 1&2 closed group 6 closed group 5 closed group 4 closed group 3 closed advances in technology innovation, vol. 4, no. 3, 2019, pp. 140-151 150 table 2 comparison of the average mole fractions of hydrogen at the reforming tube outlets operating manner average mole fractions of hydrogen at the reforming tube outlets fully opened 0.708 group 1 turned off 0.681 group 2 turned off 0.678 group 3 turned off 0.676 group 4 turned off 0.675 group 5 turned off 0.681 group 6 turned off 0.686 group 1&2 turned off 0.655 group 3&4 turned off 0.647 group 5&6 turned off 0.653 group 1,2&3 turned off 0.625 group 4,5&6 turned off 0.629 4. conclusions in this research, the effect of the burner on/off manners on an industrial-scale side-fired steam methane reformer is investigated to seek a feasible burner on/off manner that has acceptable catalyst tube temperature and hydrogen yield, so as to improve the performance and service life of a steam methane reformer. it is found that when one group of burners is turned off, the outer surface temperatures of the tubes are decreased by about 94°c in average, the inner surface temperatures are decreased by about 54°c in average, and the hydrogen yields are decreased by about 4%. when two groups of burners are turned off, the outer surface temperatures of the tubes are decreased by about 175°c in average, the inner surface temperatures are decreased by about 106°c in average, and the hydrogen yields are decreased by about 7.9%. when three groups of burners are turned off, the outer surface temperatures of the tubes are decreased by about 251°c in average, the inner surface temperatures are decreased by about 151°c in average, and the hydrogen yields are decreased by about 11.4%. the tube temperatures and the hydrogen yields reduce to a greater extent in regions where burners are turned off. when the central groups of burners are turned off, the tube temperatures and the hydrogen yields have greater reductions. on the other hand, when the rear groups of burners are turned off, the tube temperatures and the hydrogen yields have lower reductions. the result of this paper is helpful in improving the performance and service life of a steam methane reformer. funfing this research and the apc were funded by the ministry of science and technology, taiwan, under the contract most107-2221-e-150-061. acknowledgement the author is grateful to the formosa petrochemical corporation in taiwan for providing valuable data and constructive suggestions to this research during the execution of the industry-university cooperative research project under the contract 104af-86. conflicts of interest the author declares no conflict of interest. abbreviations the following abbreviations are used in this manuscript: cμ turbulence model constant (=0.09) k turbulence kinetic energy (m 2 /s 2 ) advances in technology innovation, vol. 4, no. 3, 2019, pp. 140-151 151 p pressure (n/m 2 ) t temperature (k) v velocity (m/s) xyz cartesian coordinates with origin at the centroid of the burner inlet (m) y mole fraction (%) greek symbols ε turbulence dissipation rate (m 2 /s 3 ) μ viscosity (kg/(m s)) ρ density (kg/m 3 ) τ shear stress (n/m 2 ) references [1] a. s. kimmel, “heat and mass transfer correlations for steam methane reforming in non-adiabatic, process-intensified catalytic reactors,” master thesis, marquette university, milwaukee, wi, usa, 2011. [2] l. castagnola, g. lomonaco, and r. marotta, “nuclear systems for hydrogen production: state of art and perspectives in transport sector,” global journal of energy technology research updates, vol. 1, pp. 4-18, 2014. [3] a. tran, a. aguirre, h. durand, m. crose, and p. d. christofides, “cfd modeling of an industrial-scale steam methane reforming furnace,” chemical engineering science, vol. 171, pp. 576-598, 2017. [4] a. di carlo, i. aloisi, n. jand, s. stendardo, and p. u. foscolo, “sorption enhanced steam methane reforming on catalyst-sorbent bifunctional particles: a cfd fluidized bed reactor model,” chemical engineering science, vol. 173, pp. 428-442, 2017. [5] l. lao, a. aguirre, a. tran, z. wu, h. durand, and p. d. christofides, “cfd modeling and control of a steam methane reforming reactor,” chemical engineering science, vol. 148, pp. 78-92, 2017. [6] e. m. a. mokheimer, m. i. hussain, s. ahmed, m. a. habib, and a. a. al-qutub, “on the modeling of steam methane reforming,” journal of energy resources technology, vol. 137, 012001, 2015. [7] m. ni, “2d heat and mass transfer modeling of methane steam reforming for hydrogen production in a compact reformer,” energy conversion and management, vol. 65, pp. 155-163, 2013. [8] h. ebrahimi, j. s. soltan mohammadzadeh, a. zamaniyan, and f. shayegh, “effect of design parameters on performance of a top fired natural gas reformer,” applied thermal engineering, vol. 28, pp. 2203-2211, 2008. [9] y. s. seo, d. j. seo, y. t. seo, and w. l. yoon, “investigation of the characteristics of a compact steam reformer integrated with a water-gas shift reactor,” journal of power sources vol. 161, pp. 1208-1216, 2006. [10] fluent inc., ansys fluent 17 user’s guide, new york: fluent inc., 2017. [11] s. v. patankar, numerical heat transfer and fluid flows, new york: mcgraw-hill, 1980. [12] b. e. launder and d. b. spalding, lectures in mathematical models of turbulence, london: academic press, 1972. [13] r. siegel and j. r. howell, thermal radiation heat transfer, washington, dc: hemisphere publishing corporation, 1992. [14] y. r. sivathanu and g. m. faeth, “generalized state relationships for scalar properties in non-premixed hydrocarbon/air flames,” combustion and flame, vol. 82, pp. 211-230, 1990. [15] b. e. launder and d. b. spalding, “the numerical computation of turbulent flows,” computer methods in applied mechanics and engineering, vol. 3, pp. 269-289, 1974. [16] c. l. yeh, “numerical analysis of the combustion and fluid flow in a carbon monoxide boiler,” international journal of heat and mass transfer, vol. 59, pp. 172-190, 2013. [17] l. chibane and b. djellouli, “methane steam reforming reaction behaviour in a packed bed membrane reactor,” international journal of chemical engineering and applications, vol. 2, pp. 147-156, 2011. [18] b. f. magnussen and b. h. hjertager, “on mathematical models of turbulent combustion with special emphasis on soot formation and combustion,” proc. symp. (international) on combustion, the combustion institute, 1976. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation, vol. 3, no. 1, 2018, pp. 43 50 cae analysis of secondary shaft systems in great five-axis turning-milling complex cnc machine chih-chiang hong 1,* , cheng-long chang 1 , chun-chen huang 2 , chi-ching yang 3 , chien-yu lin 4 1 department of mechanical engineering, hsiuping university of science and technology , taichung, taiwan, roc. 2 department of industrial engineering and management , hsiuping university of science and technology , taichung, taiwan, roc. 3 department of electrical engineering, hsiuping university of science and technology, taichung, taiwan, roc. 4 l&l machinery industry co., ltd, taichung, taiwan, roc. received 12 february 2017; received in revised form 12 april 2017; accept ed 23 april 2017 abstract the commerc ial co mputer a ided engineering (ca e) software is used to analy ze the linear-static construction, stress and deformation fo r the secondary shaft systems in great five -a xis turning-milling co mple x co mputer numerical control (cnc) mach ine. it is convenient and always only three dimensional (3d) graphic parts needed firstly prepared and further mo re detail used for the co mme rcia l ca e. it is desirable to predict a defo rmed position for the cut tool under externa l pressure loads in the working process of cnc machine. the linear results for static analysis of stresses, displacements in corresponding to the screw shaft locates at top, mediu m and bottom positions of the secondary shaft systems are obtained by using the simulation module of solidworks® . keywor ds : cae, static analysis , linear analysis , solidworks, shaft systems, stress analysis, cnc 1. introduction there are many co mputer a ided engineering (ca e) co mmerc ia l software used to develop and design the computer numerical control (cnc) machine for saving the cost of production. in 2016, afkha mifar et al. [1] used the finite e le ment method (fem) analysis to s imu late the position error of the tooltip in the 3-a xis vertical milling machin ing centers cnc series. in 2015, ma x et al. [2] used catia® and nx™ software to create 3d models for the teaching and studies of fem analysis in the cnc milling machine. in 2014, altintas et a l. [3] simu lated and optimized the cutting process in the virtual machining (vm ) of cnc system. in 2014, soori et al. [4] developed a vm software and created machined parts in the virtual environ ments for 3-a xis cnc. in 2013, chang [5] introduced computer-based technology in the vm to provide a relative ly low setup cost when compared with physical cnc. in 2012, wang et al. [6] used software ansys® (one of the fem codes) to co mpute the static-structural results of the fra me and tool ca rrier fo r the hydraulic swing-type plate shears of cnc equip ments. there are also some other comme rcia l ca e software used in the structural analysis for the engineering system, for e xa mp le: solidw orks® , creo® , inventor® , freeca d (an open -source), abaqus® , hypersize r® and midas® etc.. in 2016, mackrell [7] introduced the mu lti-mechanics module o f software abaqus® for engineer used to work and design in the fie ld of ca e. in 2013, paulo et a l. [8] used software abaqus® to simulate mechanica l behavior for the stiffened a lu minu m panels. in 2010, youn is [9] presented the autodesk software inventor® used for the structural simulat ion in the engineering. to e xecute the fourth industrial revolut ion for the cnc systems, the design and analysis experiences of ca e are novel for the conventional company. in 2016, hong et al. [10] presented the static -structural ca e analysis of great five-a xis turning-milling comple x machine for the cnc system with the solidworks® simu lation module . simp le, c lear and easy steps in the simu lated process are the specific reason of selecting solidworks software for present work. educational version of * corresponding author. e-mail address: cchong@mail.hust.edu.tw advances in technology innovation, vol. 3, no. 1, 2018, pp. 43 50 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti 44 solidw orks 2014 software has been used for conducting the present simulation. in this paper, the linear static stresses and displacements of secondary s haft system of the cnc machines are studied more details and obtained with the solidw orks® simulat ion module. the ma ximu m values of stress and displacement are usually provided to give a basic data for the detail construction of cnc machine. the highlight notes of this paper are also included as follows: (1) it is helpful for engineers to investigate great five-a xis turning-milling co mple x cnc machine data under ca e ana lysis . (2) co mmerc ial ca e solution for the secondary shaft systems under e xternal pressure loads is provided. (3) the linear results are provided by using the simu lation module of solidworks® . for the meaning of "great five-a xis " in this study usually notes that bigger dimensions of work piece e.g. length 5000 mm, dia meter 950 mm than s ma lle r one can be mach ined at working t ime by the five a xes: x-a xis, y-a xis, z -a xis (three translations), a-a xis and b-a xis (two rotations) of wor king platform wh ich moving respectively to the cutting tools . in 2015, yang et a l. [11] presented the general stiffness model of rigidity for tool path planning in five-a xis cnc machin ing to restrain the chatter of the mach ine. in 2015, wagner [12] p resented an optimization co mputer aided design (cad)/co mputer aided manufacture(cam)/ ca e technique for the processing of cutting tool of the comple x 3d s urfaces on 5 a xes cnc machines . the main novelty of their research is used the commerc ia l ca e simu lation module to investigate the linearly static structure analysis in cnc mach ine for t ime saving and obtain the basic data for the construction of cnc machine parts . in general, the using of commerc ia l ca e software have the trustable and acceptable experience in t he data, but the cost of module software is respectively higher when co mpared with personal developed software. usually the c ommerc ial cae software e.g. solidworks® used for the educational version has the great discount in the universities and schools. t he ma in scope of this research is to use the reasonable cost of commerc ial ca e software to save developing time of software and find the useful data of co mputation. there a re so me c o mme rc ial ca e softwares in educational version had by the university e.g. catia® , a nsys® and solidworks® . they are all very good commerc ia l cae softwares to provide the calculation solution. when the commerc ia l ca e software had for the industry is usually very e xpensive and needed some teaching courses to provide for the user. to choose which type of co mmerc ial ca e software is very important based on the user’s like of industrial company when the researcher and teacher of university are in the corporation relationship. the advantages of this paper in co mparison with the other researches performed in this subject are the using solidworks® simu lation module is also used by the employee of industrial co mpany, the upgrading of new cnc machine design performed by the owners of industrial company and the reliable data calculated from the commercial cae software. 2. method the steps of simu lation with the software solidworks® simu lation module in the static structural linear analysis and the general matrix equation of mathe matica l model were used in the computer program to solve for stress and displacement results by hong et al. [10] as follows.     fuk  (1) where [k ] is materia l stiffness matrix, {u} is displace ment vector, {f} is e xternal load vector. it is necessary to prepare the assembling 3d parts of the secondary shaft systems as shown in fig. 1. the dimensions of ma in parts are provided for the secondary shaft system is 1397mm 845mm 1426mm. the main co mponents of secondary shaft system are screw shaft and base. the tool is fixed on the end of scre w shaft to provide dril ling and milling functions. the secondary shaft system can be rotated by the base in rotational motion with respect to y a xis. it is necessary to define the individual materia l of assembl ing 3d parts for the secondary shaft systems. the ma in materia ls of t he secondary shaft system are cast steel. the yie ld stress of cast steel material is 241mpa. to present normal working in the cnc machine, the value of working stress in each materia l of components should under its yield stress value. advances in technology innovation, vol. 3, no. 1, 2018, pp. 43 50 cop y right © taeti 45 three contact boundary conditions (b. c.) of secondary shaft systems are used to computed and analyzed for corresponding to the screw shaft locates at top, mediu m and bottom positions with 100mm apart along y a xis, respectively. the boundary conditions of the secondary shaft s ystem for the base are four sides in clamp b.c. and shown in fig . 2. external pressure loads on left-end of screw shaft and hydraulic pressure loads on the base in secondary shaft system for the screw shaft locates at top position typically is also shown. mesh of grids in the secondary shaft system is shown in fig . 3 and fig. 4. mesh of grids with para meter e le ment length equal to 27.21mm based on curvature mesh is shown in fig . 3, also with proper mesh controlled base on the size of parts, mediu m mesh dens ity and used to generate a proper mesh of grids in the computation and analyses. a typical mesh of g rids in secondary shaft system for the screw shaft locates at top position is sh own in fig. 4. a table is provided to define the characteristics of materials used in the simulation as shown in table 1. fig. 1 assembling 3d parts of the secondary shaft system (a) contact b.c. for the screw shaft locates at top position (b) contact b.c. for the screw shaft locates at medium position (c) contact b.c. for the screw shaft locates at bottom position (d) four sides in clamp b.c. for the base fig. 2 boundary conditions in the secondary shaft system external pressure loads on left-end of screw shaft and hydraulic pressure loads on the base in secondary shaft system for the screw shaft locates at top position typically is also shown. mesh of grids in the secondary shaft system is shown in fig . 3 and fig. 4. mesh of grids with para meter e le ment length equal to 27.21mm based on curvature mesh is s hown in fig. 3, also with proper mesh controlled base on the size of parts, mediu m mesh density and used to generate a proper mesh of grids in the computation and analyses. a typical mesh of g rids in secondary shaft system for the screw shaft locates at to p position is shown in fig. 4. a table is provided to define the characteristics of materials used in the simulation as shown in table 1. advances in technology innovation, vol. 3, no. 1, 2018, pp. 43 50 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti 46 table 1 characteristics of materials component material dimensions yield stress screw shaft carbon steel 1397mm 220mpa ψ303mm base cast steel 1042mm 241mpa 815mm 231mm sliding box 1 1023 carbon steel 600mm 282mpa 540mm 350mm sliding box 2 carbon steel 600mm 220mpa 540mm 350mm fig. 3 mesh of grids parameters in the secondary shaft system fig. 4 typical grids in the secondary shaft system 3. results and discussion firstly, used the solidw orks® simu lation module to obtain the stresses and displacements of static results due to e xternal pressure loads (10mpa perpendicula r to xy plane, para lle l to x a xis and y a xis, respectively) on left -end of the screw shaft and hydraulic p ressure loads (10mpa on the base) of secondary shaft system. the e xte rnal loads and their positions are determined in table 2. static stress and displacement results of secondary shaft system fo r screw shaft locates at top position are shown in fig. 5 and fig. 6, respectively. the ma ximu m value 131m pa of stresses is found in the area of base and the ma ximu m value 0.522mm of displace ments is found in the top are a of secondary shaft system. the ma ximu m va lue (131mpa) of stress due to external pressure loads (10mpa in x, y and z) and hydraulic pressure loads (10mpa ) are less than yield stres s value 241mpa , so the parts of machinery are in safety condition. it suggests that the machinery of secondary shaft system can stand 10mpa e xterna l loads. a linear analysis is considered and clarified that the behavior of structure is also linear in fact, for the ma ximu m value 0.522mm o f displace ments found in the top area of secondary shaft system is very much less than the dimension length 1397mm of the screw shaft. advances in technology innovation, vol. 3, no. 1, 2018, pp. 43 50 cop y right © taeti 47 table 2 external loads and their positions external load position value direction external pressure loads on left-end of the screw shaft 10mpa perpendicular to xy plane external pressure loads on left-end of the screw shaft 10mpa parallel to x axis external pressure loads on left-end of the screw shaft 10mpa parallel to y axis hydraulic pressure loads on top of the base 10mpa parallel to y axis fig. 5 stress for the screw shaft locates at top position fig. 6 displacement for the screw shaft locates at top position fig. 7 stress for the screw shaft locates at top position fig. 8 displace ment for the screw shaft locates at top position secondly, the simplic ity stresses due to the same e xternal p ressure loads (10m pa) place on left -end of screw shaft are studied, when the screw shaft locates at top, mediu m and bottom position s of the secondary shaft system, respectively. static stress and displacement results for the screw shaft locates at top position are shown in fig. 7 and fig. 8, respectively . fig. 9 stress for the screw shaft locates at medium position fig. 10 displacement for the screw shaft locates at medium position advances in technology innovation, vol. 3, no. 1, 2018, pp. 43 50 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti 48 the ma ximu m value 64m pa of stresses is found in the area of base and the ma ximu m va lue 0.1012 mm of displace ments is found in the left-end area of screw shaft and in the left-top-end area of fra me. static stress and displacement results for the screw shaft locates at mediu m position are shown in fig . 9 and fig. 10, respectively. the ma ximu m va lue 61mpa of stresses is found in the area of base and the ma ximu m va lue 0.0864 mm of displace ments is foun d in the left-end area of screw shaft. static stress and displacement results for the screw shaft locates at bottom position are shown in fig . 11 and fig. 12, respectively. the ma ximu m value 67mpa of stresses is found in the area of base and the ma ximu m va lue 0.07839mm of displacements is found in the left-end area of screw shaft. the ma ximu m va lue (0.1012mm) of d isplacement due to static-uniform pressure external loads (10mpa) can be occurred at the left -end area of screw shaft, so the cut tool deflection in the secondary shaft system should be reconsidered for the accuracy operation and safety condition . fig. 11 stress for the screw shaft locates at bottom position fig. 12 displacement for the screw shaft locates at bottom position fig. 13 stress for the screw shaft fig. 14 displacement for the screw shaft fig. 15 stress for the sliding box 1 fig. 16 displacement for the sliding box 1 thirdly, the screw shaft in the detail studies are investigated due to the same e xternal pressure loads (10mpa perpendicular to xy p lane, paralle l to x a xis and y a xis, respectively) on left -end of the screw shaft. static stress and displacement results for the screw shaft are shown in fig . 13 and fig. 14, respectively. the ma ximu m value 65m pa of stresses is found in the corner area of shaft and the ma ximu m va lue 0.058mm o f displace ments is found in the left -end a rea of the screw advances in technology innovation, vol. 3, no. 1, 2018, pp. 43 50 cop y right © taeti 49 shaft. the sliding bo x in the detail studies are investigated due to the s ame e xternal pressure loads (10mpa downward to z a xis) on the one-end-side of slid ing bo x. static stress and displacement results for the sliding bo x 1 material are shown in fig . 15 and fig. 16, respectively. fig. 17 stress for the sliding box 2 fig. 18 displacement for the sliding box 2 the ma ximu m value 309m pa of stresses is found in the sliding -groove area of bo x and the ma ximu m va lue 0.046 mm of displacements is found in the front-central area of the sliding bo x. the ma ximu m value (309m pa) of stress due to external pressure loads (10mpa downwa rd to z a xis) is g reater than yie ld stress value 282m pa, so the sliding bo x of machinery is in un-safety condition. it suggests that the machinery of the slid ing box 1 materia l can't e xceed 10mpa e xte rnal loads. it is more interesting to observe some places/spots where there are some stress concentrations that will occur due to different material properties. static stress and displacement results for the slid ing box 2 materia l are shown in fig. 17 and fig. 18, respectively. the ma ximu m value 2126mpa of stresses is found in the sliding -groove area of bo x and the ma ximu m value 0.05221 mm of displacements is found in the front-central area of the sliding box. the ma ximu m value (2126mpa ) of stress due to extern al pressure loads (10m pa downwa rd to z a xis) is much greater than yield stress value 220mpa (a lmost 10 times), so the sliding box of machinery is in un-safety condition. in the linear analysis, it suggests that the machinery of the slid ing bo x 2 material can't exceed 1m pa e xterna l loads. more informat ive with further analysis related to this study would be the cut position on th e screw shaft in the milling process simulation subject. it would be interesting to investigate the optimize variable pitch fo r the cut under the milling and turning processes with considering the effects of stress and displacement . 4. conclusions in this paper, the static linear stress and displacements of secondary shaft system under e xte rnal pressure loads in cnc mach ines are obtained with the simu lation module of solidworks® . under the action values of externa l pressure loads 10m pa and hydraulic pressure loads 10mpa , when the screw shaft locates at top position, the ma ximu m va lue 131mpa of stresses is found in the area of base and the ma ximu m value 0.522mm o f d isplacements is found in the top area of secondary shaft system. the ma ximu m values of stress and displacement are usually provided to give a basic data for the detail and good construction of secondary shaft system, so the cnc machine can present in normal working condition . acknowledgement the completion of this paper was enabled by a grant 105-iem-1244 from hsiuping university of science and technology, taiwan, roc. references [1] a. afkhamifar, d. antonelli, and p. chiabert, “variational analysis for cnc milling process,” procedia cirp, vol. 43, pp. 118-123, 2016. advances in technology innovation, vol. 3, no. 1, 2018, pp. 43 50 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti 50 [2] a. max, v. lašová, and š. pušman, “enhancement of teaching design of cnc milling machines,” procedia social and behavioral sciences, vol. 176, pp. 571-577, 2015. [3] y. altintas, p. kersting, d. biermann, e. budak, b. denkena, and i. la zoglu, “virtual process systems for part machining operations,” cirp annals manufacturing technology, vol. 63, no. 2, pp. 585-605, 2014. [4] m. soori, b. arezoo, and m. habibi, “virtual machining considering dimensional, geometrical and tool deflection errors in three-axis cnc milling machines,” journal of manufacturing systems, vol. 33, no. 4, pp. 498-507, 2014 [5] k. h. chang, “chapter 2 virtual machining,” in product manufacturing and cost estimating using cad/cae, pp. 39-93, 2013. [6] y. wang, b. cui, k. li, t. zhang, and z. zhang, “structural analysis and experimental research of an cnc hydraulic swing-type plate shears,” aasri procedia, vol. 3, pp. 414-420, 2012. [7] a. mackrell, “multiscale composite analysis in abaqus: theory and motivations,” reinforced plastics, 2016. [8] r. m. f. paulo, f. teixeira-dias, and r. a. f. va lente, “numerical simulation of aluminium stiffened panels subjected to axial compression: sensitivity analyses to initial geometrical imperfections and material properties,” thin-walled structures, vol. 62, pp. 65-74, 2013. [9] w. younis, “chapter 15 dp13 assembly optimization: structural optimization of a lifting mechanism,” up and running with autodesk inventor simulation 2011 (second edition) a step-by-step guide to engineering design solutions, pp. 353-372, available online 26 may 2010. [10] c. c. hong, c. l. chang, and c. y. lin, “static structural analysis of great five-axis turning-milling complex cnc machine,” engineering science and technology, an international journal, vol. 19, no. 4, pp. 1971-1984, 2016. [11] c. yang, z. liqiang, and l. dong, “general stiffness model for five-axis cnc machining,” international journal of research in engineering and science, vol. 3, no. 8, pp. 43-47, 2015. [12] e. wagner, “a new optimization cad/cam/cae technique for the processing of the complex 3d s urfaces on 5 axes cnc machines ,” procedia technology, vol. 19, pp. 34-39, 2015.  advances in technology innovation, vol. 5, no. 2, 2020, pp. 98-111 robust lpv control for attitude stabilization of a quadrotor helicopter under input saturations seif-el-islam hasseni 1,* , latifa abdou 2 1 lmse laboratory, 2 li3cub laboratory, electrical engineering department, university of biskra, algeria received 18 march 2019; received in revised form 29 september 2019; accepted 15 october 2019 doi: https://doi.org/10.46604/aiti.2020.3953 abstract this article investigates the robust stabilization of the rotational subsystem of a quadrotor against external inputs (disturbances, noises, and parametric uncertainties) by the lft-based lpv technique. by establishing the lpv attitude model, the lpv robust controller is designed for the system. the weighting functions are computed by cuckoo search, a meta-heuristic optimization algorithm. besides, the input saturations are also taken into account through the anti-windup compensation technique. simulation results show the robustness of the closed-loop system against disturbances, measurement noises, and the parametric uncertainties. keywords: quadrotor, lpv systems, optimal weighting functions, cuckoo search, input saturations 1. introduction several researchers have focused on quadrotor aircraft because of its advantages such as vertical landing and stationary flight [1]. however, the quadrotor is a nonlinear complex vehicle and very sensible, which motivates the researchers to develop robust controllers. in this article, the development of a robust controller for the quadrotor attitude subsystem is investigated. recently, researchers have been interested in linear-parameter-varying (lpv) theory because of its advantage that allowing them to represent the nonlinear model as the quasi-lpv system. there are several approaches to develop the lpv system from its nonlinear original model. the common ones are the polytopic approach [2] and the linear-fractional-transformation (lft) [3]. in this article, the lft is used because it can exact representative the nonlinear model. on the other hand, one of the efficient robust controllers is the h∞, which is created for the linear-time-invariant systems (lti). in the lpv case, there are no frequency norms. there is a just relative gain in the l2 sense (in the time-domain). in the lti case, the l2 gain and the h∞ norm are equal. hence, the lpv method via the l2 gain concept can be seen as the extension of h∞ synthesis [4]. one of the existed problems on h∞ controller design is weighting functions’ selection. there are many structures such as loop-shaping which is proposed by mcfarlane and glover [5]. loop-shaping is the simplest and the most commonly used method [6]. the other common structure is the mixed sensitivity structure [7]. it is very hard to give a general formula for weighting functions that work in every case [8]. this problem depends on experimental skills [9], but zhou and doyle proposed guidelines to adjust the weighting functions’ parameters for a siso system [10]. in [11], the authors propose to consider each system as a combination of first and second-order sub-systems, which make it easier to find a good weighting function to be used in h∞ control methodology. yet there is no general methodology to adjust the parameters especially when performance specification is difficult to define, plus it gets more complex for mimo systems. * corresponding author. e-mail address: seif.hasseni@univ-biskra.dz advances in technology innovation, vol. 5, no. 2, 2020, pp. 98-111 99 on the other side, the cuckoo search algorithm is a swarm-based optimization algorithm. it is inspired by a kind of bird (cuckoo) behavior in dumping its eggs into random nests. a cuckoo never builds its own nest but lets other species doing it. the algorithm’s essential operations are the global random walk and the local walk. the global walk is similar to practical swarm optimization (pso) and simulated annealing (sa) with lévy flight transition probability. the local walk is a differential operation with such probability, so it is similar to the differential evolution algorithm (de). therefore, each algorithm (sa, pso, and de) could be considered as a special case of the cs algorithm. conversely, cs is a good and efficient combination of sa, pso, and de in one algorithm [12]. many researchers have used a nature-inspired optimization algorithm to determine weighting functions’ parameters for the lpv system [13-14]. this approach is much more required in the mimo systems. in this article, the mixed sensitivity structure is applied, and then the cuckoo search (cs) algorithm is exploited to find optimal weighting functions’ parameters. the second point investigated in this article is the input saturation. it is a real constraint of the actuators. the actuators can’t provide additional effort although the outputs don’t reach the reference signal. such a technique called anti-windup has been used in [15]. an anti-windup compensator has been designed with the same nature and structure of lft-lpv plant to reject the input saturation effect [16]. the article’s outline is as follows; in section 2, the representation of the quadrotor attitude model as the lpv system and the synthesis controller are recalled [17]. section 3 is about a description of the cs algorithm characteristics. besides that, the optimization of the weighting functions problem for the nominal system (without considering the input saturation) is discussed in section 3.2. in section 4, the anti-windup compensator design is investigated. the simulation results will be shown in section 5, and finally, the conclusion in section 6. 2. lft-lpv controller design in this section, the control of the quadrotor’s attitude by the lpv technique is recalled [17]. 2.1. lft-lpv representation of quadrotor attitude as noticed previously, this work focusses on the rotational subsystem of the quadrotor, the framework of three degrees of freedom system is presented on fig. 1. the angles are the three degrees of freedom; roll (φ), the rotation around the x axis; pitch (θ), the rotation around the y axis; yaw (ψ), the rotation around the z axis. fig. 1 the geometric structure of the quadrotor these motions are achieved by four rotors, ( 1 4)i im  . the nonlinear dynamic model of the rotational subsystem is shown on eqs. (1)-(2). it has six states; roll, pitch, yaw; [ ]   and their derivatives; [ ]   . it has three inputs ][u u u  . advances in technology innovation, vol. 5, no. 2, 2020, pp. 98-111 100 4 2 3 1 1 2 3 4 ( ) ( ) 1 ( ) y z r r x x x z x r r y y y x y z z i i j l f f i i i i i j l f f i i i i i t t t t i i                                   (1) where 1 2 3 4r         (2) the considered inputs are expressed as: 4 2 3 1 1 2 3 2 2 4 2 2 2 3 1 2 2 2 2 4 1 2 3 4 ( ) ( ) = ( ) u f f u f f u t t t t b b d                               (3) table 1 os4 parameters parameter description value unit l arm length 0.23 m b thrust coefficient 3.13 10 -5 n.s 2 d drag coefficient 7.5 10 -7 n.m.s 2 ix, iy inertia on x and y axis 7.5 10 -3 kg.m 2 iz inertia on z axis 1.3 10 -2 kg.m 2 jr rotor inertia 6 10 -5 kg.m 2 ωi rotor speed [0, 500] rad.s -1 table 1 shows the model’s parameters from the selected project, os4 [18]. by considering the assumption that the selected system is symmetric ( ),x yi i the yaw model can be linearized as: 1 z u i   (4) there are two subsystems i.e., the roll-pitch (φ, θ) subsystem and the linear one, yaw (ψ) subsystem. fig. 2 show the sheme of the quadrotor attitude. after some mathematical operations (for more details see [17]), the lpv model of the nonlinear subsystem (roll-pitch) with lft representation is presented as: 1 1 1 1 11 d p p p pp p p p a b bx x z c d w y c d d u                            (5) and p pw z  (6) fig. 2 lft-lpv scheme of the quadrotor attitude zρ u wρ y quadrotor attitude θ advances in technology innovation, vol. 5, no. 2, 2020, pp. 98-111 101 where the elements of the matrices in eq. (5) can be obtained by: 0 1 0 0 0 0 0 0 0 0 0 1 0 0 0 0 a              (7) 0 0 0 0 0 0 0 0 0 0 0 0 y z r x x p z x r y y i i j i i b i i j i i                        (8) 1 0 0 0 0 0 0 x y l i l b i                     (9) 0 0 0 1 0 0 0 1 0 1 0 0 0 1 0 0 pc              (10) 1 1 0 0 0 0 0 1 0 c        (11) 4 40 xppd  (12) 1 4 20p xd  (13) 1 2 40p xd  (14) 11 2 20 xd  (15) where the state vector t x       , the input vector t u u u     , the outputs   t y   , pz and pw are the inputs and outputs of the parameters block, 1, 2, 1, 2( ).diag p p p p  table 2 presents the parameter range of roll-pitch subsystem. the yaw subsystem is a simple double integrator system with 2 states   t   , and just one input ( ).u table 2 varying-parameters range parameter description range 1p  1 [ 2, 2] .rad s   2p r 1 [ 500, 500] .rad s   2.2. the controller synthesis in the h∞ controller design, either the lpv subsystem (roll-pitch) or lti subsystem (yaw) needs to interconnect the system with externl inputs (disturbances and noises) and the actuator dynamics. advances in technology innovation, vol. 5, no. 2, 2020, pp. 98-111 102 the rotor’s dynamic in the selected project, [18] can be expressed as: 0.936 ( ) 0.178 1 i di act s s      (16) where i  is the actual rotor speed, di  is the desired rotor speed. from eq. (3), it is possible to extract the inputs’ dynamics. the actuators’ dynamic for each input ( , , )u u u   of pitch, roll, and yaw controls can be expressed as: 2 ( )a du act s u (17) where du is the desired control signal generated by the controller, and au is the actual control signal. fig. 3 lft-lpv representation of closed loop plant-controller as noticed before, both the plant and controller of the roll-pitch subsystem are lpv with lft representation (fig. 3). on the contrary, the yaw subsystem is a simple lti. the augmented plant (p) includes actuators dynamics, external inputs, and weighting functions. the plant’s state space is presented in eqs. (18)-(22), and the controller’s state space is expressed as: 1 2p px ax b w b w b u    (18) 1 2p p pp p p pz c x d w d w d u    (19) 1 1 11 12p pz c x d w d w d u    (20) 2 2 21 22p py c x d w d w d u    (21) p pw z  (22) where x is the states vector; u is the input vector; w is the exogenous inputs; disturbances and noises; y is the measured outputs; z is the controlled outputs; pz and pw are inputs and outputs of the block, 1 1 2 2( , ,..., ).r r k rkdiag p i p i p i  1k k k k kp kpx a x b y b w   (23) 1 11 1k k k k p kpu c x d y d w   (24) 1kp kp k kp kpp kpz c x d y d w   (25) kp kpw z  (26) where kx is the controller states vector, y is the outputs from the plant, u is the controller outputs, kpz , and kpw are inputs and outputs of the block parameter. zkρ zρ wkρ w u y wρ z p k θ θ advances in technology innovation, vol. 5, no. 2, 2020, pp. 98-111 103 after getting the lpv model of attitude in section 2.1, establishing an interconnection between the plant and the different inputs is needed. the inputs are reference signals, disturbances, and noises. the actuator dynamic is also added by the block act 2 as shown in eqs. (16)-(17). furthermore, the weighting functions which represent the desired performance are ordered in mixed sensitivity structure by the errors ( , ,e   ) and the input signals ( , ,u   ). in this article, three types of external inputs are considered, i.e., reference signals ( , ,r   ), disturbances ( , ,d   ), and noises ( , ,n   ) as presented in fig. 4. the controller of each subsystem, the lti controller for the yaw subsystem and the lpv controller for the roll-pitch subsystem, will be designed with the same nature as the plant. fig. 4 closed-loop plant-controller of the quadrotor 3. optimal weighting functions 3.1. the cuckoo search algorithm the cuckoo search (cs) algorithm is a swarm-based intelligent algorithm. it was developed by yang and deb in 2009 [19] and inspired by the cuckoo’s behavior in laying its eggs. a cuckoo doesn’t build its own nest but selects another species’ nest to let them brood young cuckoos. in this article, a simple way of application of cs is used; a nest corresponds to a cuckoo and also to an egg. the standard cs procedure strike a balance between a local and a global random walk. the global walk is presented by lévy flight [19] where a cuckoo searches for a new nest. furthermore, there is a probability pa of cases where the host birds discover that the eggs are not theirs, so they abandon the eggs. the cuckoo seeks for a new solution (nest) far from the actual solution; that’s the local random walk. algorithm 1: improved cuckoo search 1 2 3 4 5 6 7 8 9 10 11 initialize solutions randomly; while max_generation not meet do search by fraction (pc); get a cuckoo randomly by lévy flight; evaluate its fitness fi; get a nest randomly (j) if fi is better than fj replace the worse nest(j) by the trial one; end if fraction (pa) of worse nests are abandoned and build new ones; end while p + + + + + ze zu uφ,θ,ψ rφ,θ,ψ φ,θ,ψ eφ,θ,ψ dφ,θ,ψ nφ,θ,ψ zρ wρ zkρ wkρ we act 2 . quadrotor attitude θ θ wu k advances in technology innovation, vol. 5, no. 2, 2020, pp. 98-111 104 the improved cuckoo search (algorithm 1) was developed by ouaarab et al. [20-21]. its extension to the standard cs is to add a fraction pc of smart cuckoos which seek new solutions in other areas by lévy flight operation, and can be expressed as: 1 ( ) t t i ix x l     (27) where l is the lévy flight function [22],  is the scaling factor. after getting a new solution (trial one), it is compared with a random one (j). if the trial solution is better than j (the worse), it should be replaced by the new solution. in addition to the fraction pc, there are other eggs with probability pa witch are abandoned by the host birds. therefore, the cuckoos search for new nests in the local area is: 1 ( ) t t t i p q at i t i x s x x if r p x x otherwise         (28) where px and qx are random solutions different from ix , and s is a scaling factor. the parameters’ values are chosen on the base of the recommendation of optimal ranges from experts in the field [20], and are shown in table 3. table 3 parameters setting of improved cs parameter value max_generation 20 population size 50 pc 0.5 α 0.1 λ 1.5 pa 0.25 s 0.2 3.2. the optimal weighting functions the good choice of weighting functions significantly affects the system’s nominal performance. in [17], the selection of the weighting functions is implemented by trial-and-error. in this article, the implementation is done by an optimal algorithm (cuckoo search). the weighting functions are espressed as: 0 0 0 0 0 0 e e e e w w w w               (29) 0 0 0 0 0 0 u u u u w w w w               (30) because the pitch and the roll have the same dynamics, their functions are equal. the weighting function of error is considered as a first-order filter. the weighting function of the control is considered as a scalar gain, i.e., 1 2 1 1 e e e s w w k s         (31) u u uw w k   (32) concerning the yaw subsystem, the weighting function of the error is considered as a first-order filter. the weighting function of the control is considered as a simple gain and expressed as: advances in technology innovation, vol. 5, no. 2, 2020, pp. 98-111 105 1 2 1 1 e e s w k s       (33) u uw k  (34) the parameters’ selection of the roll-pitch subsystem and yaw subsystem is achieved separately. in the optimization process, the solutions of roll-pitch problem and yaw problem can be obtained by:  1 2e us k k   (35)  1 2e us k k   (36) it should be noticed that the solutions are presented in real-coded. the optimization process of each problem is solved by the presented cs optimization algorithm in section 2. the fitness functions are the integral square of error (ise) is used for the fitness function, that is, 2 2 0 0 ( ) ( ) t t roll pitchf e d e d        (37) 2 0 ( ) t yawf e d    (38) the fitness (ise) evolutions of each criterion, eqs. (37)-(38), are presented in fig. 5 and fig. 6. the efficiency of the cs algorithm is shown by the presented convergences where the fitness of the roll-pitch subsystem (fig. 5) reaches its optimal (minimum) at 10 th generation, and the fitness of the yaw subsystem (fig. 6) reaches its optimal (minimum) at the 8 th generation. the optimal weighting functions parameters are presented in table 4. fig. 5 fitness evolution of the roll-pitch subsystem (iseφ+iseθ) fig. 6 fitness evolution of the yaw subsystem (iseψ) table 4 optimal parameters of we and wu roll-pitch weφ, weθ ke τ1 τ2 12.66 3.112 65.359 wuφ, wuθ ku=0.902 yaw weψ ke τ1 τ2 2.5 0.3388 8.8968 wuψ ku=0.28 fig. 7 presents the inverse sensitivity functions of both subsystems. the presented weights are the error weighting functions ew  and ,ew  when the weights in the control criteria ( uw  and uw  ) are just scalar gains. the inverse of advances in technology innovation, vol. 5, no. 2, 2020, pp. 98-111 106 sensitivity function of the roll-pitch subsystem ( 1 / ew  ) is a high pass filter, the steady-state error to step input ( 0.078)  , the overshoot magnitude (peak sensitivity, 1.65sm  ), and limit closed-loop bandwidth /0.1937 rad s)( b  . the inverse of the sensitivity function of the yaw subsystem ( 1 / ew  ) is a high pass filter also, the steady-state error to step input ( 0.4)  , the overshoot magnitude (peak sensitivity, 1.55sm  ), and limit closed-loop bandwidth /0.281 rad s)( b  . these results are in line with the recommended guidelines of [10]. fig. 7 desired closed-loop performance let klpv denote the controller of the roll-pitch subsystem, and klti denote the controller of the yaw subsystem, they can be expressed as: 1 1 1 11 1 1 k k kp lpv k k k p kp kp kpp a b b k c d d c d d            (39) 2 2 2 2 k k lti k k a b k c d        (40) the augmented controller k (fig. 4) is obtained by combining the two controllers, and that is: 1 2 0 0 k k k a a a        (41) 1 2 0 0 0 kpk k k bb b b        (42) 1 2 0 0 0 k k k kp c c c c            (43) 111 2 1 0 0 0 0 k pk k k kp kpp dd d d d d            (44) in this stage, the problem of the optimal weighting functions’ parameters is solved, but the problem of the input saturation has not been considered yet. the advantage of this method is that the controller design and the anti-windup compensator synthesis can be carried out independently. the next section is to discuss the anti-windup compensator’s design step. advances in technology innovation, vol. 5, no. 2, 2020, pp. 98-111 107 4. anti-windup compensator the system is a mini-drone which has actuators, rotors, with low energy consumption. if the input saturation is imposed, the controller design considering the input saturation is needed to stabilize the system. the saturation function is presented in eq. (45): ( ) m m m m m m u if u u sat u u if u u u u if u u           (45) in general nonlinear systems, hyperbolic tangent ( ( ) tanh( ))sat u u is used the approximate function of the saturation e.g., the missile control [23] and ship control [24]. many researchers have focused their attention on the anti-windup problem, hence, a modern anti-windup compensator design is developed for the lti systems called coprime factor [25]. 4.1. theoretical background the main advantage of this approach is that the controller and the anti-windup are designed independently. the anti-windup compensator has the same nature and the same representation of the plant and the controller (lft-lpv). fig. 8 general scheme of the coprime factor anti-windup the anti-windup compensator is shown in the block diagram (fig. 8). the windup’s effect is rejected by adding yd, and the plant can be created as multiplier factors. to design the factors, firstly a matrix ( )f  is created according to some lmi conditions. by putting ( ) ( ) ,m f i   it could be possible to generate the other factor ( )n  ; ( ) ( ) ( ).n g m   so, the plant ( )g  could be presented as multiplier factors; 1 ( ) ( ) ( ).g n m     therefore, the design of the anti-windup compensator is based on determination ( )n  and ( )m  of ( ).g  the key is to find ( )f  that minimize the 2l gain. for more details about computing ( ),f  see [16]. previously, the lpv controller for the nominal system was designed without considering the saturation. as mentioned, the compensator’s design is independent of the controller’s design. what’s next is the proof: by assuming ( ) ( ) ( )n g m   in fig. 8 it yields, ( ) m mom u u u u m u     (46) ( ) ( ) ( ) ( )( ( ) ) ( ) nom d m m nom y y y n u g u n u g u m u g u               (47) where nomu is the nominal input generated by the controller, mu is the actuator input. nomy is the nominal output. y is the actual output, du and dy are the outputs of the anti-windup compensator. advances in technology innovation, vol. 5, no. 2, 2020, pp. 98-111 108 the anti-windup state-space is presented as: ( ) ( ) ( ) ( ) ( ) u u u u m d y yu y yu a fb b w b u z c fd d w d u a w y m u f u y c fd d w d u w z                                             (48) 4.2. the compensator design fig. 9 implemented anti-windup compensator in the case of the stabilization of the quadrotor attitude, the rejecting of the input saturation’s effect is done by the form of internal model control (imc) [26]. that means ( ) 0f   ; so, ( )m i  and ( )( ) gn   . the only output of the anti-windup compensator is d y (fig. 9). and where the augmented plant, ( )g  , contains the plant ( )p  which is the lft-lpv representation of the quadrotor attitude and the actuators function (act 2 ). from table 2, the rotor’s saturation is 1 0 500 . i rad s    . by combing it with eq. (3), the saturation values are presented as: (.) (7.825, 7.825, 0.375)sat diag (49) 5. simulation results to show the efficiency of the designed controller, the controller is applied to the nonlinear model itself (eq. (1)). the selected initial angles are 0 0 0 ] ] 3 3 3 [ [ t        rad, where the objective is to stabilize the quadrotor at its origin 0, 0, 0)( . in the next simulation, the nominal performance is taken without considering the disturbances. the application of the controller is done for three modes: the nominal one where no saturation is introduced; the second mode where the input saturation is introduced, and the third mode where the input saturation is introduced by adding the anti-windup compensator. fig. 10 presents the angles’ dynamics of the three modes. fig. 11 and fig. 12 present the inputs and its zoom with the three modes. the benefit of anti-windup compensation is clearly shown. the solid line presents the nominal responses (without saturation). it’s possible to observe a swift response and short settling time with impossible input (reaches 100 n). however, when the saturation is introduced without anti-windup (dotted line), it’s possible to observe a longer settling time and bigger overshoot than the nominal one. by adding the anti-windup compensator (dashed line), the responses are better than the obtained by the saturation without anti-windup compensator, and the responses are almost identical to those in the nominal responses. in fact, this is the objective of the anit-windup compensator. the efficiency of the robustness h∞ controller is shown against the external inputs and the parametric uncertainties with taking the anti-windup compensator. permanent measurement noises of 2 ( 0.034 rad)   are applied on all of the the angles. + g(ρ) ref u + um ũ yd ynom y k(ρ) n(ρ) act 2 p(ρ) advances in technology innovation, vol. 5, no. 2, 2020, pp. 98-111 109 the input disturbances are introduced as impulses with 0.1 d n   in 6, 6.07][ s, 0.1 d n   in 10, 10.07][ s, and 0.02 .d n m   in 14, 14.1][ s. the simulation results are presented in fig. 13. it is observed that the stability is kept despite the presence of disturbances and noises. fig. 10 the angles responses fig. 11 the plant control inputs fig. 12 the plant control inputs (zoom) by adding 50% of the nominal values of the inertia , , , )( x y z ri i i j after 5 seconds as uncertainties, its simulation is shown in fig. 14. compared with the nominal system, it is clearly noticed that there are no big changes in the responses. in addition to disturbances and noises, the system is robust and also against the parametric uncertainties. advances in technology innovation, vol. 5, no. 2, 2020, pp. 98-111 110 fig. 13 the angles responses with presence of input disturbances and measurement noises fig. 14 the angles responses with presence of the uncertainties 6. conclusions in this article, the robust control of a quadrotor attitude has been carried out by using an lft-lpv h∞ controller. two issues are discussed. first, the weighting functions selection where an optimization meta-heuristic algorithm is exploited, cs, to provide the parameters of these functions to get optimal performances. second, the problem of the input saturation which is imposed by the actuators. this problem has been solved by using an anti-windup compensator. the advantage of this approach is that the lpv controller and the anti-windup compensator could be designed separately without any conflict in the system stability’s analysis. the simulation results showed that the system is robust and against the external inputs as well as the uncertainties. conflicts of interest the authors declare no conflict of interest. references [1] l. carrillo, a. lópez, r. lozano, and c. pégard, quad rotorcraft control. london: springer-verlag, 2013. [2] n. vafamand and a. khayatian, “model predictive-based reset gain-scheduling dynamic control law for polytopic lpv systems,” isa transactions, vol. 81, pp. 132-140, 2018. advances in technology innovation, vol. 5, no. 2, 2020, pp. 98-111 111 [3] h. behzad, a. casavola, f. tedesco, and m. a. sadrnia, “use of lpv-lft unknown input observers for the design of fault tolerant sensor reconciliation schemes,” proc. iranian conference on electrical engineering, mashhad, iran, may 2018, pp. 899-905. [4] p. apkarian and p. gahinet, “a convex characterization of gain-scheduled h∞ controllers,” ieee transactions on automatic control, vol. 40, no. 5, pp. 853-864, 1995. [5] d. mcfarlane and k. glover, robust controller design using normalized coprime factor plant descriptions, vol. 138 of lecture notes in control and information sciences, berlin: springer-verlag, 1990. [6] a. d. rosaline and s. ushakumari, “robust loop-shaping controller for load frequency control of an uncertain deregulated power system,” journal of the institution of engineers (india): series b, vol. 100, no. 4, pp. 357-369, auguest 2019. [7] l. yu, c. che, j. guo, and h. wang, “the design of h∞ mixed sensitivity controller for fin/tank roll stabilizer,” proc. 35 th chinese control conference, chengdu, china, july 2016, pp. 2929-2933. [8] s. skogestad and i. postlethwaite, multivariable feedback control, analysis and design, 2 nd ed. chichester: john wiley & sons, 2005. [9] r. w. beaven, m. t. wright, and d. r. seaward, “weighting function selection in the h∞ design process,” control engineering practice, vol. 4, no. 5, pp. 625-633, may 1996. [10] k. zhou and j. c. doyle, essentials of robust control, new jersey: prentice hall, 1998. [11] j. hu, c. bohn, and h. r. hu, “systematic h∞ weighting function selection and its application to the real-time control of a vertical take-off aircraft,” control engineering practice, vol. 8, no. 3, pp. 241-252, march 2000. [12] x. s. yang, nature-inspired optimizations algorithms, x.s. yang (ed.). london: elsevier, pp. 129-139, 2014. [13] a. l. do, o. sename, l. dugard, and b. soualmi, “multi-objective optimization by genetic algorithms in h∞/lpv control of semi-active suspension,” ifac proceedings volumes, vol. 44, no. 1, pp. 7162-7167, january 2011. [14] v. t. vu, o. sename, l. dugard, and p. gaspar, “multi objective h∞ active anti-roll bar control for heavy vehicles,” ifac papersonline, vol. 50, no. 1, pp. 13802-3807, july 2017. [15] j. fu, y. lv, and t. huang, “distributed anti-windup approach for consensus tracking of second-order multi-agent systems with input saturation,” systems & control letters, vol. 130, pp. 1-6, auguest 2019. [16] e. prempain, m. c. turner, and i. postlethwaite, “coprime factor based anti-windup synthesis for parameter-dependent systems,” systems & control letters, vol. 58, no. 12, pp. 810-817, december 2009. [17] s. hasseni and l. abdou, “robust attitude stabilization of a quadrotor by using lft linear-parameter-varying h∞,” proc. 26 th mediterranean conference on control and automation, zadar, croatia, auguest 2018, pp. 241-246. [18] s. bouabdallah, “design and control of quadrotors with application to autonomous flying,” ph.d. thesis, epfl, switzerland, 2007. [19] x. s. yang and s. deb, “cuckoo search via lévy flights,” proc. world congress on nature and biologically inspired computing, coimbatore, india, december 2009, pp. 210-214. [20] a. ouaarab, b. ahiod, and x. s. yang, “discrete cuckoo search algorithm for the travelling salesman problem,” neural computing and applications, vol. 24, no. 7-8, pp. 1659-1669, june 2014. [21] a. ouaarab, b. ahiod, and x. s. yang, “improved and discrete cuckoo search for solving the travelling salesman problem,” cuckoo search and firefly algorithm, theory and applications, x. s. yang (ed.). cham, switzerland: springer, 2014, pp. 63-84. [22] x. s. yang and s. deb, “multiobjective cuckoo search for design optimization,” computers & operations research, vol. 40, no. 6, pp. 1616-1624, june 2013. [23] b. liu, m. hou, and y. li, “field-of-view and impact angle constrained guidance law for missiles with time-varying velocities,” ieee access, vol. 7, pp. 61717-61727, may 2019. [24] m. li, and c. liu, “new second order sliding mode control design for course-keeping control of ship with input saturation,” proc. international conference on advanced mechatronic systems, xiamen, china, march 2018, pp. 462-466. [25] p. f. weston and i. postlethwaite, “linear conditioning for systems containing suturing actuators,” automatica, vol. 36, no. 9, pp. 1347-1354, september 2000. [26] m. c. turner, g. herrmann, and i. postlethwaite, “incorporating robustness requirements into antiwindup design,” ieee transactions on automatic control, vol. 52, no. 10, pp. 1842-1855, 2007. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation, vol. 4, no. 3, 2019, pp. 197-209 performance evaluation for stacked-layer data bus based on isolated unit-size repeater insertion chia-chun tsai * department of computer science and information engineering, nanhua university, chiayi, taiwan received 13 march 2019; received in revised form 11 april 2019; accepted 10 may 2019 abstract the data bus of a stacked-layer chip always supports that data of a program are frequently running on the bus at different timing periods. the average data access time of a data bus to the timing periods dominates the program performance. in this paper, we proposed an evaluated approach to reconstruct a 3d data bus with inserted unit-size repeaters to motivate that the average data access time of the bus on a complete timing period can speed up at least 10%. the approach is trying to insert a number of unit-size repeaters into bus wires along the path of a source-sink pair for isolating extra capacitive loadings at each timing period to reduce their access time. the above process is repeated until no any improvement for each access time. each inserted repeater with just one unit size due to the limited space of a chip area and the minor reconstruction of a data bus in practical. the approach has the advantages of uniform repeater insertion, less extra area occupation, and simplified time-to-space tradeoff. experimental results show that our approach has the rapid capable evaluation for a stacked-layer data bus within one millisecond and the saving in average access time is up to 50.81% with the inserted repeater sizes of 70 on average. keywords: stacked-layer chip, 3d data bus, unit-size repeater, average access time 1. introduction for a stacked-layer chip [1], each layer has own local data bus and a number of tsvs (through silicon vias) is used to vertically connect these local data buses to integrate them to be a 3d global data bus. the 3d global data bus consists of a number of 2d local data buses. data are frequently running on the 2d local data bus or 3d global data bus for executing multiple programs. a data access time is defined as the propagation delay from a source to at least one sink at a timing period. for a program with a number of hundreds or thousands timing periods, its average access time is defined as the total data access times divided by the number of timing periods. the average access time dominates the program performance. p4 p1 p2 p3 p6 p5 c3 c2 c5 c4 p4 p1 p2 p3 p6 p5 c’3 c’2 c’5 c’4 (a) extra loading capacitances c2 to c5 (b) extra capacitive loadings are reduced by inserted repeaters fig. 1 data access on the source-sink pair p1-p6 of a 2d local data bus * corresponding author. e-mail address: chun@nhu.edu.tw tel.: +886-5-2721001#5030 advances in technology innovation, vol. 4, no. 3, 2019, pp. 197-209 198 in nanotechnology, a longer interconnection wire always dominates the propagation delay more than a gate delay because of their incremental wire resistances and capacitances. fig. 1(a) shows a 2d local data bus and there is a bidirectional data access between terminals p1 and p6 at two different timing periods. from the figure, obviously, these extra loading capacitances, c2, c3, c4, and c5 will cause to increase the data access time of the source-sink pair of p1-p6. each extra loading capacitance comes from their branch wire capacitance and terminal capacitive loading along the path of p1-p6 or p6-p1. as shown in fig. 1(b), most of these extra capacitive loadings can be isolated by inserting a bidirectional repeater into each branch wire for the data bus reconstruction. that is, these extra loading capacitances will be dramatically reduced to be c’2, c’3, c’4, and c’5, and c’2 < c2, c’3 < c3, c’4 < c4, and c’5 < c5. the data bus reconstruction will result that the data access time between two source-sink pairs with terminals p1 and p6 can be clearly reduced. the above concept of reconstructing a data bus can be expanded to other source-sink pairs for isolating unnecessary capacitive loadings by inserting repeaters into their branch wires to reduce their access times. for a program ran on a data bus with a number of timing periods, its average access time can thus be reduced and its performance can also be upgraded. however, all the inserted repeaters will also cause extra area occupation. this is the time-to-space tradeoff of a data bus reconstruction, such as the saving in average access time is at least 10% with paying a number of repeater sizes. the repeater insertion was widely applied to a one-way signal path that can effectively reduce their propagation delay, but a few papers discussed the repeater insertion to apply a bidirectional data bus. ismail [2] proposed repeater insertion for the path delay of an rlc-based wire to estimate the delay and their inductive effects of on-chip interconnects. lin [3] presented buffer insertion to construct a clock tree on multimode multivoltage islands. they used adjustable delay buffers (adbs) for controlling the clock delay under a boundary skew. ghoneima [4] introduced the optimal positioning of interleaved repeaters in bidirectional buses. his solution was in focus to reduce noise interference between buses. acton [5] summarized some studies of signal repeater insertion in multi-source multi-sink data bus. daneshtalab et al. [6] proposed an appropriate bus isolation strategy for a 3d stacked-layer chip and had a high-performance inter-layer bus structure (hibs). the hibs can reduce the complexity of bus arbitrators and make the saving in the propagation delay of data communication. thakkar et al. [7] introduced a new architecture called 3d-wiz that is used for reducing the interaction overloading between data bus of drams. the architecture can reduce their access times among any drams. cho et al. [8] presented the analysis of system bus considering the interconnection of tsvs on a stacked-layer soc (system-on a chip). they found the maximum throughput of the system bus of a 3d stacked-layer chip depending on the data bandwidth. mohamed [9] introduced a master-slave data access by adding nocs (network-on-chips) among multiple processors and there was a number of data interchange rules that would limited the access time between processors. khan et al. [10] analyzed the performance for current noc simulation tools in terms of latency, throughput, and energy consumption, but this comparison was just for 2d nocs. tsai [11-12] first conducted repeater insertion and sized the repeaters to minimize the propagation delay for a 3d data bus based on rc delay model, but they do not to consider the capacitive loading effect of un-accessed local data buses. tsai [13] created an effective method associated with embedded isolated switches [14-15] and inserted repeaters for a 3d data bus to reduce their critical access time, but no any considerations about the pre-evaluation in average access time for a data bus reconstruction. most of the above reconstructed data bus methods were based on the repeater insertion and sized them as possible for maximally reducing the data access time. these approaches can decrease the access time effectively, but their data bus would be required to have extra areas for inserting different-size repeaters. this causes the incremental difficulty for reconstructing a data bus at the post refinement step in physical design. the above problem for the optimal solution in the time-to-space tradeoff (data access time minimization vs. repeaters’ locations and sizes) by inserting repeaters into a data bus had been approved to be intractable [16]. how to evaluate the data bus of a stacked-layer chip to run well for reducing the average access time? a few papers conduct to this topic and it is the valuable problem for investigation in advance. advances in technology innovation, vol. 4, no. 3, 2019, pp. 197-209 199 in this work, we proposed an approach to evaluate the bus performance by reconstructing a 3d data bus. with inserting unit-size repeaters into a data bus at each timing period, the average data access time of a bus on a complete timing period can be motivated to speed up at least 10% (here, we call it as basic performance ratio). the approach is trying to insert a number of unit-size repeaters to isolate most of extra capacitive loadings to reduce the access time of each source-sink pair at different timing periods. this process is repeatedly done until no any improvement for each access time. then we can estimate the new average access time of a data bus on a complete timing period. if the saving in average access time with inserted unit-size repeaters is larger than the basic performance ratio, the reconstructed data bus can be accepted for reducing the average access time and applied for most of multiple programs ran on the bus. here, we emphasize the inserted repeater with just a unit size due to the limited space of a chip area and the minor reconstruction of a data bus. the evaluated approach has advantages: uniform repeaters, less extra occupied area, and simplified the time-to-space tradeoff with a basic performance ratio. the demonstrated results show that most of 3d stacked-layer data buses with inserted unit-size repeaters their average access time for any program can be dramatically reduced. 2. problem formulation 2.1. symbols and definitions table 1 shows all the symbols and their definitions that are used to go through the whole article for accordance. table 1 symbols and their definitions symbol definition symbol definition n the total number of terminals of a 3d data bus q the total number of bus wires of a 3d data bus n(n-1) the complete timing period of a 3d data bus pk the kth terminal on a 3d data bus tij the access time from source i to sink j without any inserted unit-size repeaters t’ij the access time from source i to sink j with inserted unit-size repeaters tav the average access time of a 3d data bus without any inserted unit-size repeaters u-tav the average access time of a 3d data bus with inserted unit-size repeaters rw the resistance of a unit-length wire u-size the number of inserted unit-size repeaters cw the capacitance of a unit-length wire rpk the kth unit-size repeater rtsv the resistance of a tsv rb the resistance of a unit-size repeater ctsv the capacitance of a tsv cb the capacitance of a unit-size repeater rfg the resistance of a segment wire (f,g) tb the intrinsic delay of a unit-size repeater cfg the capacitance of a segment wire (f,g) rdi the output driving resistance of a source i r1 the resistance of a bus wire l1 clj the input loading capacitance of a sink j c1 the capacitance of a segment wire l1 c(tg) the lumped capacitance at node g ca the total capacitance of a wiring area a cs the extra capacitance at a node c’a the reduced total capacitance of a wiring area a with inserted unit-size repeaters c’s the reduced extra capacitance at a node with inserted unit-size repeaters cbusj the total capacitance of the jth-layer local bus,e.g., cbus2 m the multiple times of wire capacitance c1 e.g., cs = mc1, m  0 2.2. gartner ś hype cycle phases a 3d stacked-layer data bus as shown in fig. 2 extended from fig. 1, there is a number of n terminals and exists a maximal number of n(n-1) timing periods as well as a number of n(n-1) data access times. the number of n(n-1) timing periods is called the complete timing period of a data bus. generally, an executed program has a number of hundreds or thousands timing periods that data are frequently running on the data bus and these timing periods may cover a complete timing period. if most of data access times for a program at different timing periods can be reduced a little, then its average access time will be decreased, that is, the program performance can thus be promoted. fig. 2(a) shows the bidirectional data access between two terminals p4 located on layer1 and p16 located on layer3 at the different timing periods of a 3d stacked-layer data bus. obviously, their data access times, tp4-p16 and tp16-p4, cover those extra loading capacitances, ca, cb, cc, cd, ce, cf and cbus2. especially, the total capacitance of local bus on layer2, cbus2, will be advances in technology innovation, vol. 4, no. 3, 2019, pp. 197-209 200 a bigger capacitive loading for their data access. as shown in fig. 2(b), if we insert a number of unit-size repeaters to some bus wires, then most of extra loading capacitances can be reduced to be c’a, c’b, c’c, c’d, c’e, c’f and c’bus2, respectively. thus, their data access times, tp4-p16 and tp16-p4, can be effectively reduced. p16 p13 p14 p15 p10 p7 p8 p9 p4 p1 p2 p3 tsv1 layer1 tsv2 layer2 layer3 p17 p18 p11 p12 p6 p5 cbus2 ca ce cf cd cc cb p16 p13 p14 p15 p10 p7 p8 p9 p4 p1 p2 p3 tsv1 layer1 tsv2 layer2 layer3 p17 p18 p11 p12 p6 p5 c’c c’bus2 c’b c’a c’e c’d c’f (a) extra loading capacitances (b) access time is reduced with inserted repeaters fig. 2 data access on the source-sink pair p4-p16 of a 3d stacked layer in fig. 2(a), the access time tij (tji) from source i (j) to sink j (i) along the path of the source-sink pair of p4-p16 based on elmore -rc delay model [17] is represented as below. ( , ) ( . ) +( )( ( )) 2 t fg gij f g path i j di fg c r r c t   (1) where rdi is the output driving resistance of source i, rfg and cfg are resistance and capacitance of a bus wire (f,g), respectively, and c(tg) is the lumped capacitance of branch rooted at node g. it is noted that c(tg) contains those extra capacitive loadings, ca, cb, cc, cd, ce, cf and cbus2. fig. 3 shows the equivalent -rc circuit based on elmore delay model of fig. 2(b) between terminals p4 located on layer1 and p16 located on layer3 with two tsvs and a number of inserted unit-size repeaters for isolating extra loading capacitances. from the figure, extra loading capacitances c11, c12, and c13 on layer1 are isolated from the inserted unit-size repeaters rp11, rp12, and rp13; extra capacitances c21 and c22 on layer2 are isolated from the inserted unit-size repeaters rp21 and rp22; and extra capacitances c31, c32, and c33 on layer3 are isolated from the inserted unit-size repeaters rp31, rp32, and rp33. a unit-size bidirectional repeater has two equivalent sets of input capacitance cb, intrinsic delay tb, and output resistance rb that are inversely connected in parallel. the access time is the scaled-50% propagation delay based on elmore rc delay model. likely a bus wire, a tsv has also the equivalent rc mode [18] with the resistance rtsv and two half capacitances of ctsv/2. the access time t’ij (t’ji) from source i (j) to sink j (i) along the path of a source-sink pair of p4-p16 with isolated unit-size repeaters is represented as below. + ( ) ( , ) ( . ), ( , ) ( )( ( )) 2 t k fg gf g path i j rp path i j di fgij c r r c t      (2) where c(tg) is the lumped capacitance of branch rooted at node g including the capacitances within those isolated unit-size repeaters rp(k). advances in technology innovation, vol. 4, no. 3, 2019, pp. 197-209 201 c13 p16 ctsv2/2 ctsv2/2 rtsv2 ctsv1/2 ctsv1/2 rtsv1 cl p 1 6 p 13 p 1 4 p 1 5 p 10 p 7 p 8 p 9 p 4 p 1 p 2 p 3 ts v1 la yer 1 ts v2 lay er2 lay er3 p 1 7 p 1 8 p 1 1 p 1 2 p 6 p 5 c ’c c’ bus 2 c ’b c ’a c ’e c ’d c ’f cb rb tb rb cb tb cb rb tb rb cb tb cb rb tb rb cb tb cb rb tb rb cb tb cb rb tb rb cb tb p4 cl cb rb tb rb cb tb cb rb tb rb cb tb cb rb tb rb cb tb rd rd c31 rp31 c32 rp32 c33 rp33 c11 rp11 c12 rp12 rp13 c21 rp21 c22 rp22 fig. 3 the equivalent -rc circuit of a source-sink pair p4-p16 shown in fig. 2(b) for a reconstructed data bus with inserted unit-size repeaters, the average access time on a complete timing period is represented as the bus performance. since a unit-size repeater has also including the input capacitance, output resistance, and intrinsic delay, the access times for all the source-sink pairs with inserted repeaters will be affected with each other. thus, we need to estimate a data bus with inserted unit-size repeaters whether its average access time on a complete timing period is decreased or not. that is, we can calculate the saving percentage in the average access time on a complete timing period for the data bus without/with inserted unit-size repeaters. if the saving is over the basic performance ratio, the data bus can be reconstructed with inserted a number of unit-size repeaters that has good performance improvement in average access time. therefore, the problem to evaluate the performance in average access time on a complete timing period for reconstructing the 3d data bus of a stacked-layer chip can be defined as below. given the topology of a stacked-layer data bus that has a number of n terminals and a number of q bus wires on a complete timing period (i.e., the number of n(n-1) timing periods), the objective is to evaluate the possible reconstruction of a data bus by inserting unit-size repeaters into the bus wires such that the saving in average access time with inserted repeaters is at least the basic performance ratio than that of without any inserted repeaters, where the basic performance ratio depending on the user’s definition, such as 10 %. 3. performance evaluation of a stacked-layer data bus 3.1. the estimation of a unit-size repeater insertion to understand the effects in data access time of a source-sink pair, it is required to make the estimation of a data access time before/after inserting a unit-size bidirectional repeater into a bus wire. as shown in fig. 4(a), the access time tij from advances in technology innovation, vol. 4, no. 3, 2019, pp. 197-209 202 source i to sink j along the bus wire l1 based on the elmore delay model can be obtained. if a sink connects the wire segments of a subtree, then the sink has the extra loading capacitance cs and the access time tij will be increased, and tij is represented as below. 1 1 1 ( / 2 ) ( )t ij lj li ljs sdi c c c r c c c cr       (3) where r1and c1 are the resistance and capacitance of a wire l1, respectively, rdi is the output driving resistance of source i, and cli and clj are the input loading capacitances of source i and sink j, respectively. cli rdj rdi clj source/sink i l1 source/sink j cs (a) extra capacitance cs unit-size repeater cli rdj rdi clj source/sink i l1 source/sink j c’s (b) cs is reduced to be c’s cli rdj rdi cb rb tb rb cb tb clj source/sink i l1/2 l1/2 source/sink j bidirectional unit-size repeater c’s (c) inserting repeater into the middle of a bus wire l1 fig. 4 the bus wire l1 is inserted into a bidirectional unit-size repeater to reduce the access time tij, we can insert a unit-size repeater to isolate the subtree wires that can largely decrease the extra loading capacitance cs to be c’s, c’s < cs, as shown in fig. 4(b), that is, eq. (3) is updated to be t’ij and t’ij is denoted as below. 1 1 1 ( / 2 ) ( )t ij lj li ljs sdi c c c r c c c cr        (4) as shown in fig. 4(c), the access time t’ij from source i to sink j can be reduced in advance by inserting a unit-size bidirectional repeater into the middle of a bus wire l1 if it was enough longer, that is, eq. (4) is updated as 1 1 11 1 1 / 2 / 2 / 2( / 4 ) ( / 2 )( / 4 ) ( )t b b li bdiij lj b b ljs s r c c r c c cc c c r c c c c tr             (5) where rb, cb, and tb are the output resistance, input capacitance, and intrinsic delay of a unit-size repeater, respectively. for simplification, we assume that cs is the multiple times of the wire capacitance c1, that is, cs = mc1, m  0. and c’s is sum of the half of capacitance c1 and the input capacitance cb, i.e., c’s = c1/2+cb if m > 0 and c’s = 0 if m = 0. if the source and advances in technology innovation, vol. 4, no. 3, 2019, pp. 197-209 203 sink are also a bidirectional unit-size repeater, then rdi = rb and cli = clj = cb. the access times tij and t’ij from source i to j without/with inserted repeaters are respectively derived as follow. 1 1 1 1 1 1 1 1 1 1 1 ( / 2 ) (2 ) ( / 2 ) (2 / 2 ) (m+1/2)r 2 ( 1) , 0 t b b b b b b bs s b b b b ij c c c r c c c r c c mc r c c c c r c r c m r c m r                  　 (6) if m is progressively large, then the access time tij will be increased, but the access time t’ij always keeps a fixed value that is independent of m. with inserting a unit-size repeater into the wire l1, if its access time t’ij is always less than tij and then the reduced quantity in access time of (tij-t’ij) is obviously meaningful. here, we want to know how the wire length l1 can be inserted a unit-size repeater for effectively reducing the access time. 1 1 11 1 1 / 2 / 2 / 2( / 4 ) ( / 2 )( / 4 ) ( )t b b li bdiij lj b b ljs s r c c r c c cc c c r c c c c tr             (7) case 1: m = 0, 2 1 1 1 / 4 2 ( ) / 4 (2 ) 0w wij ij b b b b b bt t r c r c t r c l r c t        (8) where the unit of rw and rb is ω, the unit of cw and cb is pf, and the unit of tb is ps, and the unit of l1 is µ m. we can derive the wire length l1 (µ m) is 1 2 2 b bb w w r c t l r c   (9) case 2: m > 0, 1 1 2 1 1 / 2 3 ( 1/ 2) ( / 2 (1/ 2 ) ) (3 ) 0 1 1 b b b b b w w w wb b b b b ij ij r c r c m r c t mr c l r c m r c l r c t t t mr c                (10) the wire length l1 (µ m) can be formulated as 2 1 0.5 (0.5 ) (0.5 (0.5 ) ) 4 (3 ) 2 w w w w w wb b b b b b b w w r c m r c r c m r c mr c r c t l mr c         (11) 3.2. the effects of data access time with inserted unit-size repeaters due to the strategy of extra capacitive loading isolation is adopted by inserting unit-size repeaters, the access time of a source-sink pair for the shorter path has larger reduction in extra capacitances than the longer path. for a data bus on a complete timing period, all the bus wires are almost inserted with full unit-size repeaters. the data access time of a source-sink pair for the longer path may increase. fig. 5(a) shows its extended data access of a source-sink pair of p4-p16 in fig. 2(b) that has up to the number of six inserted unit-size repeaters, rp14, rp15, rp16, rp34, rp35,and rp36, along their longer path. repeaters rp14 and rp16 are inserted for isolating extra capacitive loading due to the path of p2-p6, rp15 is inserted for the isolation due to the path of p4-p6, rp34 and rp36 are inserted for the isolation due to the path of p14-p18, and rp35 is inserted for the isolation due to the path of p14-p16. fig. 5(b) shows its equivalent circuit of fig. 5(a) that is the updated bus structure with inserted unit-size repeaters. the access time t’ij from source i to sink j along the path of a source-sink pair of p4-p16 with inserted unit-size repeaters is formulated as below. + ( ) ( ) ( , ), ( . ), ( , ) t ( )( ( )) 2 x xb b bx x k r fg gij f g rp path i j rp path i j di fg c r r c c t t      (12) where rp(k) is the number of isolated unit-size repeaters that are not located on the path of p4-p16 and rp(x) is the number of inserted repeaters that are located on the path of p4-p16. advances in technology innovation, vol. 4, no. 3, 2019, pp. 197-209 204 3.3. the evaluation for reconstructing stacked-layer data bus with inserted unit-size repeaters the evaluated algorithm, evaluate_stacked-layer_databus_reconstruction(), for the bus performance by reconstructing a stacked-layer data bus with inserted unit-size repeaters is introduced in fig. 6 to solve the above problem defined in section 2. the initial step is to read a 3d data bus topology to construct their data structure. then, we calculate the average access time tav of an original 3d data bus without any inserted repeaters on a complete timing period using the function, find_averageaccesstime(), where tav is defined as the total access times divided by the number of n(n-1) timing periods. the for loop in step3 is for each timing of a complete timing period and insert a number of unit-size bidirectional repeaters into the middle of all the branch bus wires along the path of each source-sink pair estimated by eqs. (9) and (11) for isolating the branch capacitive loadings, but at most one repeater is inserted into the middle of a bus wire. the new average access time u-tav of a 3d data bus with inserted unit-size repeaters on a complete timing period is obtained using the same function find_averageaccesstime() in step4. finally, if the saving u-saving in average access time defined as (tav – u-tav) / tav * 100% is larger than the basic performance ratio 10%, then, the 3d data bus can be reconstructed by inserting a number of unit-size bidirectional repeaters in the space depending on the limited chip area. otherwise, give up the reconstruction of a 3d data bus topology. p16 p13 p14 p15 p10 p7 p8 p9 p4 p1 p2 p3 tsv1 layer1 tsv2 layer2 layer3 p17 p18 p11 p12 p6 p5 c’c c’bus2 c’b c’a c’e c’d c’f rp15 rp14 rp35 rp16 rp34 rp36 (a) six inserted unit-size repeaters to the path of p4-p16 c13 p16 ctsv2/2 ctsv2/2 rtsv2 ctsv1/2 ctsv1/2 rtsv1 cl p4 cl rd rd c31 rp31 c32 rp32 c33 rp33 c11 rp11 c12 rp12 rp13 c21 rp21 c22 rp22 cb rb tb rb cb t b cb rb tb rb cb tb cb rb tb rb cb tb cb rb tb rb cb tb cb rb tb rb cb t b cb rb tb rb cb tb cb rb tb rb cb tb cb rb tb rb cb cb rb tb rb cb cb rb tb rb cb cb rb tb rb cb t b cb rb t b rb cb t b cb rb t b rb cb t b cb rb tb rb cb t b rp14 rp34 rp15 rp35 rp36 rp16 (b) the equivalent circuit of the bus structure fig. 5 a source-sink pair p4-p16 has six inserted unit-size repeaters and its equivalent circuit advances in technology innovation, vol. 4, no. 3, 2019, pp. 197-209 205 evaluate_stacked-layer_databus_reconstruction() { /* a 3d data bus topology with the number of n terminals and q bus wires on a complete timing period. */ step1: scan a 3d data bus topology and construct its data structure. step2: compute each source-sink access time of n(n-1) timing periods and get the whole average access time tav by the function of find_averageaccesstime(). step3: for (each timing of a complete timing period) { insert a number of unit-size repeaters to isolate those all the extra capacitive loadings form the source-sink pair; but at most a repeater is inserted the middle of a bus wire. } step4: estimate each source-sink access time of n(n-1) timing periods with inserted unit-size repeaters and calculate the whole average access time u-tav. step5: if (u-saving = (tav – u-tav) / tav * 100% > basic performance ratio 10%) then, the 3d data bus can be reconstructed by inserting a number of unit-size bidirectional repeaters into the space depending on a limited chip area. else, give up the reconstruction of a 3d data bus topology. } fig. 6 the algorithm is used for the evaluation of a data bus performance the time complexity of the proposed evaluated algorithm is o(n 2 ) because the n(n-1) timing periods are executed, where n is the number of terminals. 4. experimental results we have implemented the proposed evaluated algorithm in c language on an i7 cpu@2.7ghz, dual cores with 8gb ram, running ms-windows 10. table 2 shows the parameters of 45nm technology [19] based on elmore rc delay model. terms rw and cw represent the resistance and capacitance of a unit-length wire, respectively. rtsv and ctsv are the resistance and capacitance of a tsv, respectively. rb, cb, and tb denote the output resistance, input capacitance, and intrinsic delay of a unit-size repeater, respectively. table 2 parameters based on 45nm technology a unit-length wire a tsv a unit-size repeater rw cw rtsv ctsv rb cb tb 0.1ω 0.2ff 0.035ω 15.48ff 122ω 24ff 17ps we refer six 3d data bus topologies with 3 stacked layers from [11-12] and reduce them in total length by five times for testing our proposed algorithm. for a data bus, the driving resistances of all the sources and the loading capacitances of all the sinks are assumed to be those parameters of a unit-size repeater. the inserted repeaters into bus wires are also fixed by a unit size due to the limited space of chip area and the minor reconstruction of a data bus. table 3 shows the evaluation in average access time for six 3d data bus topologies that their total lengths are reduced by 5 (marked with r5) on their complete timing periods (marked with -nxn) and 2000 timing periods (marked with -2k), respectively. in the table, #term , #loc, tlength, and #peri are the number of terminals, number of bus wires, total wire length, and number of timing periods, respectively, of a 3d data bus. tav and u-tav are the average access times without/with inserted the number of u-size unit-size repeaters, respectively. u-saving is the saving ratio defined as (tav u-tav) / tav * 100%. since we always try to insert a bidirectional unit-size repeater into each bus wire for conducting the complete timing periods, thus their number of advances in technology innovation, vol. 4, no. 3, 2019, pp. 197-209 206 u-size unit-size repeaters is near double to the bus wires #loc. for all the cases on their complete timing periods (marked with $r5-nxn) and 2000 timing periods (marked with $r5-2k), their corresponded u-tav and u-saving are almost equivalent with each other, for example, the u-savings of test0r5-nxn and test0r5-2k are 37.09% versus 37.27%. these average access times, u-tavs, have better savings, u-savings, in the range of 37.09% to 60.88% and they are always larger than the basic performance ratio 10%. the results show that all the cases are suitable to reconstruct their data bus by inserting a number of unit-size repeaters for reducing the average access time to any programs with a number of hundred or thousand timings ran on the bus. table 3 the evaluation in average access times tav and u-tav for data buses (their total length is reduced by 5, r5) without/with inserted unit-size repeaters on their complete timing periods and 2000 timing periods, respectively example #term #loc tlength complete timing period ($r5-nxn) 2k timing periods ($r5-2k) #peri tav (ns) u-tav (ns) u-size u-saving #peri tav(ns) u-tav(ns) u-size u-saving test0r5-* 18 38 7510μm 306 0.3625 0.2281 76 37.09% 2000 0.3627 0.2275 76 37.27% casefr5-* 15 29 12069μm 210 0.6078 0.2378 58 60.88% 2000 0.6099 0.2386 58 60.88% casegr5-* 10 21 8797μm 90 0.4419 0.2270 42 48.62% 2000 0.4405 0.2240 42 49.16% casehr5-* 9 20 8538μm 72 0.4393 0.2380 40 45.73% 2000 0.4392 0.2398 40 45.40% casejr5-* 21 44 11166μm 420 0.6117 0.2917 88 52.31% 2000 0.6104 0.2903 88 52.43% casekr5-* 30 58 13776μm 870 0.7686 0.3059 116 60.20% 2000 0.7556 0.3075 116 59.30% *: nxn or 2k we extend the evaluation for all the cases that their total lengths are reduced by 10 (marked with r10). table 4 shows their corresponded u-savings of the cases on their complete timing periods (marked with $r10-nxn) and 2000 timing periods (marked with $r10-2k). like the evaluation in table 3, their u-savings are almost equivalent with each other. it is noted that three cases test0r10-nxn (test0r10-2k), casegr10-nxn (casegr10-2k), and casehr10-nxn (casehr10-2k) on their complete timing periods (2000 timing periods) are failed because their corresponded u-savings have -8.58% (-9.79%), 7.24% (7.27%), and 0.33% (-0.1%) under the basic performance ratio 10%. obviously, these three-case data buses are not suitable for inserting a number of unit-size repeaters. table 4 the evaluation in average access times tav and u-tav for data buses (their total length is reduced by 10, r10) without/with inserted unit-size repeaters on their complete timing periods and 2000 timing periods, respectively example #term #loc tlength complete timing period ($r10-nxn) 2k timing periods ($r10-2k) #peri tav(ns) u-tav(ns) u-size u-saving #peri tav(ns) u-tav(ns) u-size u-saving test0r10-* 18 38 3736μm 306 0.1856 0.2015 76 -8.58% 2000 0.1857 0.2038 76 -9.79% casefr10-* 15 29 6020μm 210 0.2703 0.1889 58 30.11% 2000 0.2703 0.1884 58 30.28% casegr10-* 10 21 4388μm 90 0.1952 0.1810 42 7.27% 2000 0.1942 0.1801 42 7.24% casehr10-* 9 20 4259μm 72 0.1908 0.1901 40 0.33% 2000 0.1907 0.1909 40 -0.10% casejr10-* 21 44 5561μm 420 0.2826 0.2479 88 12.27% 2000 0.2830 0.2480 88 12.35% casekr10-* 30 58 6859μm 870 0.3622 0.2601 116 28.18% 2000 0.3521 0.2610 116 25.89% *: nxn or 2k table 5 the evaluation in average access times tav and u-tav for casek (the total length is reduced by 5 or 10, r5 or r10) without/with inserted unit-size repeaters on their complete timing periods and different timing periods, respectively example #term #loc #peri total length reduced by 5 (casekr5-*k) total length reduced by 10 (casekr10-*k) tlength tav(ns) u-tav(ns) u-size u-saving tlength tav(ns) u-tav(ns) u-size u-saving casekr?-nxn 30 58 870 13776μm 0.7686 0.3059 116 60.20% 6859μm 0.3622 0.2601 116 28.18% casekr?-.1k 30 58 100 13776μm 0.6324 0.3008 116 52.44% 6859μm 0.2581 0.2628 116 -1.82% casekr?-.2k 30 58 100 13776μm 0.6446 0.3019 116 53.16% 6859μm 0.2710 0.2588 116 4.53% casekr?-.3k 30 58 300 13776μm 0.6757 0.3103 116 54.07% 6859μm 0.2798 0.2573 116 8.04% casekr?-.5k 30 58 500 13776μm 0.6875 0.3053 116 55.60% 6859μm 0.2964 0.2594 116 12.48% casekr?-.7k 30 58 700 13776μm 0.6960 0.3067 116 55.94% 6859μm 0.3100 0.2610 116 15.82% casekr?-1k 30 58 1000 13776μm 0.7288 0.3086 116 57.66% 6859μm 0.3252 0.2599 116 20.08% casekr?-2k 30 58 2000 13776μm 0.7556 0.3075 116 59.30% 6859μm 0.3521 0.2610 116 25.89% casekr?-3k 30 58 3000 13776μm 0.7619 0.3038 116 60.13% 6859μm 0.3602 0.2613 116 27.45% casekr?-4k 30 58 4000 13776μm 0.7636 0.3022 116 60.42% 6859μm 0.3603 0.2596 116 27.95% casekr?-5k 30 58 5000 13776μm 0.7670 0.3059 116 60.12% 6859μm 0.3614 0.2588 116 28.38% *: by 5 or 10-.3k: 300 timing periods table 5 presents the evaluation in average access time for the data bus topology casek that the total length is reduced by 5 or 10 (marked with r5 or r10) on their complete timing periods (marked with -nxn) and different number of timing periods (marked with -0.1k to -5k). from the table, the u-savings of casekr5-nxn and casekr10-nxn on their complete timing periods advances in technology innovation, vol. 4, no. 3, 2019, pp. 197-209 207 are 60.20% and 28.18%, respectively. for the cases casekr5-.1k to casekr5-5k of a 3d data bus at different number of 100 -5000 timing periods, their average access times have good u-savings in range of 52.44% to 60.42% and they are suitable for inserting a number of unit-size repeaters for the access time reduction. for the cases casekr10-.1k to casekr10-5k of a 3d data bus at different number of 100-5000 timing periods, the u-savings of three cases casekr10-.1k, casekr10-.2k, and casekr10-.3k are -1.82%, 4.53%, and 8.04%, respectively are less than the basic performance ratio 10% and they are not suitable for their data bus reconstruction. fig. 7(a) shows the 3d data bus topology of casekr10-nxn with inserting a number of 116 unit-size repeaters. in the figure, two numbers located on the middle of a bus wire are the sizes of an inserted unit-size bidirectional repeater. fig. 7(b) presents all the access times without/with inserted repeaters to each source-to-sink pair on a complete timing period. the real access time (marked with real_time) of each source-sink pair with inserted repeaters is always less than that the required access time (marked with ireq_time) without any inserted repeaters. the average access times without and with inserted repeaters are 0.3622 ns and 0.2601 ns, respectively. the saving in average access time is up to 28.18%. (a) the 3d data bus topology with inserting the number of 116 unit-size repeaters (b) real access times (real_times) with inserted repeaters are less than the required access times (ireq_times) without inserted repeaters fig. 7 the bus topology and all the required and real access times of case casekr10-nxn advances in technology innovation, vol. 4, no. 3, 2019, pp. 197-209 208 5. conclusions the proposed evaluated approach for the bus performance by reconstructing a stacked-layer data bus based on inserted unit-size repeaters on a complete timing period has been successfully applied for estimating whether the average data access time is reduced more or not. inserting a number of unit-size repeaters for a data bus reconstruction can reduce the impact in the requirements of each repeater area. conducting the complete timing period can cover all the possible data accesses for any programs executed on the bus at different timing periods. evaluating the average access time of a data bus can respond the performance of an executed program in practical. therefore, our evaluated approach is simple but very fast and effective. extending work is to investigate different diverse evaluated approaches such that can suit for the various data bus topologies of emerging stacked-layer chips. acknowledgment this work was partially supported by nhu-107 research project subsidy of nanhua university. conflicts of interest the authors declare no conflict of interest. references [1] ee times, the state of the art in 3d ic technologies, november 27, 2013. [2] y. i. ismail, e. g. friedman, and j. l. neves, “repeater insertion in tree structured inductive interconnect,” ieee transactions on circuits and systems ii: analog and digital signal processing, vol. 48, no. 5, pp.471-481, may 2001. [3] k. y. lin, h. t. lin, t. y. ho, and c. c. tsai, “load-balanced clock tree synthesis with adjustable delay buffer insertion for clock skew reduction in multiple dynamic supply voltage designs,” acm trans. on design automation of electronic systems, vol. 17, no. 3, article 34, 2012. [4] m. ghoneima and y. ismail, “optimum positioning of interleaved repeaters in bidirectional buses,” ieee trans. on cad of integrated circuits and systems, vol. 24, no. 3, pp. 461-669, march 2005. [5] q. ashton acton, issues in electronic circuits, devices, and materials: 2011 edition, scholarly editions, january 2012. [6] m. daneshtalab, m. ebrahimi, and j. plosila, “hibsnovel inter-layer bus structure for stacked architectures,” proc. ieee international conference on 3d system integration (3dic 12), february 2012, pp. 1-7. [7] i. g. thakkar and s. pasricha, “3d-wiz: a novel high bandwidth, optically interfaced 3d dram architecture with reduced random access time,” proc. ieee international conference on computer design (iccad 14), november 2014, pp. 1-7. [8] k. cho, h. s. na, t. w. cho, and y. you, “analysis of system bus on soc platform using tsv interconnection,” proc. ieee asia symposium on quality electronic design (asqed 12), august 2012, pp. 255-259. [9] k. s. mohamed, ip cores design from specifications to production, chap-4 soc buses and peripherals, 1 st ed. switzerland: springer international publishing, 2016. [10] s. khan, s. anjum, u. a. gulzari, and f. s. torres, “comparative analysis of network-on-chip simulation tools,” iet computers & digital techniques, vol. 12, no. 1, pp. 30-38, january 2018. [11] c. c. tsai, “repeater insertion for 3d data bus with tsvs for reducing critical propagation delay,” proc. international conference on computer science and information engineering (csie 15), june 2015, pp. 203-208. [12] c. c. tsai, “an effective algorithm for minimizing the critical access time of a 3d-chip data bus,” international journal of electronics communication and computer engineering, vol. 9, no. 4, pp. 117-123, july 2018. [13] c. c. tsai, “embedded bus switches on 3d data bus for critical access time reduction,” proc. ieee latin american symposium on circuits and systems (lascas 18), february 2018, pp. 1-4. [14] idtqs3245 data sheet, idt co., november 2014. [15] 74vhct126aft data sheet, toshiba co., november 2014. [16] c. c. tsai, d. y. kao, and c. k. cheng, “performance driven bus buffer insertion,” ieee trans. on computer-aided design of integrated circuits and systems, vol. 15, no. 4, pp. 429-437, april 1996. advances in technology innovation, vol. 4, no. 3, 2019, pp. 197-209 209 [17] w. c. elmore, “the transient response of damped linear networks,” journal of applied physics, vol. 19, no. 1, pp. 55-63, january 1948. [18] t. bandyopadhyay, k. j. han, d. chung, r. chatterjee, m. swaminathan, and r. tummala, “rigorous electrical modeling of through silicon vias with mos capacitance effects,” ieee trans. components, packaging, and manufacturing technology, vol. 1, no. 6, pp. 893-903, june 2011. [19] y. cao, w. zhao, e. wang, w. wang, j. velamala, a. balijepali, and s. sinha, "predictive technology model (ptm)," http://ptm.asu.edu, june 1, 2012. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation, vol. 4, no. 3, 2019, pp. 177-196 transmission wheeling pricing in embedded cost using modified amp-mile and mva utility factor methods gaurav jain 1,* , dheeraj kumar palwalia 2 , anuprita mishra 3 1,2 department of electrical engineering, rajasthan technical university, kota, india 3 department of electrical engineering, technocrats institute of technology, bhopal, india received 11 january 2019; received in revised from 27 february 2019; accepted 28 march 2019 abstract transmission wheeling pricing is one of the decisive aspects of present open access electricity market. various methods are available for transmission; however, no method is proved to diverse operating conditions of the power system. these methods are not able to quantify the full recovery of embedded cost. all the variables i.e. remaining charges, used circuit capacity are not counted in the existing methods. this paper explicates two methods, modified amp-mile method, and mva utility factor method, to recover the embedded cost. modified amp-mile method is a customized form of existing amp-mile method. in the mva utility factor method, cost allocation is based on marginal participation (mp). it evaluates the cost, using sensitivity analysis of network power. the proposed methods are tested on an ieee 6-bus system and further verified on hadoti region real 37-bus system. all the results are presented in full recovery model (frm) and partial recovery model (prm). keywords: transmission wheeling pricing, embedded cost recovery, open access, power system economics 1. introduction in reference to the indian electrical network, power plant and electrical utilities are connected to the same transmission network. a nodal point is required to decide transmission pricing by independent power producers and electrical utilities both. the action of one buyer creates an effect on other participants; hence practical cost allocation becomes difficult to investigate [1]. however, transmission cost allocation is a complicated issue in deregulated power system [2]. in past years, different methods for allocation of transmission cost in electric networks are proposed by researchers. capacity usage related to each transaction is calculated for all transmission lines by applying existing methods i.e. average participation method, marginal participation method, distribution factors, equivalent bilateral exchange method, z-bus method, and cooperative game theory. in open access, electricity market allotment of embedded cost is one of the important aspects [3]. each utility has to find a solution with the characteristics of its transmission system and degree of deregulation adopted. various methods are employed at all operational conditions of diverse power systems to obtain such a solution. no technique is capable of evaluating the entire embedded cost. any usage-based cost allocation method must contain three features, i.e. accurate algorithms for transmission usage evaluation, equitable allocation rules and full recovery of embedded cost. based on the above features, cost allocation signifies to identify cost causer for incurring these costs. to determine the causer may be complicated because the non-linear nature of power flow equalities causes difficulty to nature of power flow equations [4]. *corresponding author. e-mail address: jaingauravrtu@gmail.com tel.: +91-9414982965; +91-9602410960 advances in technology innovation, vol. 4, no. 3, 2019, pp. 177-196 178 existing methodologies used for transmission wheeling pricing are classified into rolled-in methods and usage-basedmethods. the rolled-in methods do not provide price signals that are cost reflective and are subdivided into two methods, i.e. postage stamp method and contract path method. in postage stamp method, electric utilities allocate the fixed cost among its users having firm contracts [5]. whereas, in contract path method [6] managed power would be confined to an artificially specified path through the transmission system. on the other hand, usage-based methods require power flow execution of the transmission system and divided into various sub-categories, i.e. mw-mile, modulus, mva-mile, and amp-mile methods. mw-mile methodology [7] is the pricing strategy for the recovery of fixed transmission costs on the basis of actual power flow of the transmission network. in the modulus method, all mediators have to pay for the actual capacity use and additional reserve [8]. mva-miles method is an augmented version of mw-miles; it takes into account the range of the use of network due to their active and reactive power injection/drawn [9]. whereas, amp-mile method is based on the current flow in the system. marginal participation (mp) methods dominate tracing flow methods as there is no electrical principle behind the tracing flow [10]. it is implemented where tracing flow is significantly less. it allocates transmission charges to either generators or demand nodes. the allocation between generators and demands is decided exogenously, therefore it distorts the vocational signal. it assigns power flow sensitivity in each line due to power injection at each bus and the network usage cost. sensitivities or utility factors evaluated as per mp are used to predict the changes in losses, voltages and different branch flow due to change in loads and generations [11]. these sensitivities capture the effects of unbalanced network parameters, load and generator locations. in marginal participation methods, the amp-mile method is enough to attain the aim for the transmission network. even though it has some limitation, i.e. it cannot be implemented on ehv networks and cannot allocate entire embedded cost. hence, additional charges are required to be imposed on agents. in practice, the extent of use (eu) is never 100% of the circuit’s capacity. therefore, the grid looks underutilized and lastly service cost based on the network usage will be smaller than embedded cost. other costs are incurred through supplementary charges [12]. in transmission wheeling, there are many challenges occur to find out proper coat allocation. some challenge is to promote the efficiency of the day-to-day operation of the bulk power market. to resolve the problem of signal locational advantages for investment in generation and demand. to increase the investment in the transmission system for saving customer cost in the energy market. recover the costs of existing transmission assets, which has been investment by the transmission company. in transmission pricing, incremental cost is directly available from economic dispatch. these pricing methods are well suited for rapid on line costing, but limited to presenting economic effects on the wheeling utility’s production cost and total system losses. the main focus is to find ways and means to generate and inject more competition, thereby forcing the conventional monopolistic power market to a competitive market. the transmission of open access has been introduced into the electric power supply industry to alter the traditionally monopolized market. it is desired that transmission prices and payment do not disturb decisions for new generation investment, for generator and for consumer demand. at same time charging must be achieved in a simple and fair form, realistic and adequate for real-time application as well as transparent enough to be politically acceptable. in wheeling,methodology compute a high priority problem due to growth in transmission facilities, cost differentials between utility companies, and dramatic growth in non-utility generation capacity. the modified amp method in the full recovery model and mva utility factor method in full/partial recovery model is analyzed in this paper. cost comparison analysis of the different method is show benefits of modified amp method in transmission pricing. the embedded cost allocations using a different method at different percentage loading are evaluated. a real 37-bus system and ieee 6-bus system network is used in the case study to prove efficiency and applicability of the proposed method. in this paper, the nonlinearity of sensitivity indices of modified amp method and mva utility factor has been established and analyzed. advances in technology innovation, vol. 4, no. 3, 2019, pp. 177-196 179 2. wheeling pricing methodologies the pricing methodology adopted by each utility is depending on the characteristic of the transmission or distribution network. therefore, a particular pricing methodology cannot be applied for all conditions as each methodology has its specific characteristics. deregulated environment reduces the tariff for consumers and improves the efficiency for power suppliers in the long run. transmission pricing has been categorized on the basis of their operating principles i.e. marginal/incremental cost-based pricing [13], embedded cost-based pricing [14], and combination of embedded with incremental cost-based pricing [15]. table 1 literature methodology in embedded wheeling pricing from past to present s.no year author literature points of embedded wheeling pricing 1 1989 d. shrimohammadi • analysis of mw-mile method • cost allocated is proportional to the mw flows • this method doesn’t take care of reactive power • fail to reflect technical operational conditions of network 2 2001 chingtzong su, and ji-horngliaw • analysis of mva-mile method • considering apparent power (mva) for wheeling pricing • more reasonable and valid than the commonly used mw-mile method • unable to consider direction of reactive power flow • not providing true pricing 3 2006 yog r. sood • modified analysis in mw-cost method or mva-cost method • instead of multiplying the changes in the flow in the facility by length, it is multiplied by its cost unlike mw-mile or mva-mile method 4 2006 f. li • analysis of mw + mvar-mile • this method takes care of power factor of network users • it separates the mw and mvar power flows • distinguishes direction of reactive power flow • acknowledges full cost-benefit of network users, especially dg 5 2009 paul m. sotkiewicz and j.mario vignlo • analysis of amp-mile method • cost allocation is based on amp flow caused by individual customer • explicitly account for counterflows and reward dg • drafts the direction of reactive power • but applicable only on distribution network 6 2007-2010 florin and cutsem, xiao and mccalley, k. singh • analysis of congestion management in transmission pricing • probabilistic risk indices to assess real power level security system • congestion management in deregulation environment by utilizing impedance matrix 2.1. marginal/ incremental cost-based pricing methods marginal pricing or marginal wheeling rates are also called an extension of the spot-pricing theory. spot pricing is a method for pricing electricity that maximizes the economic efficiency of the power system [16]. it is difficult to estimate transmission pricing using marginal cost based pricing method as the income would not be sufficient for financing the investment. however, a theory of social benefit, which maximizes real-time price of real and reactive powers, is considered [17]. it flattens the peak power demand and fills the valley of demand. the allocation of transmission payments among different agents depends upon the total energy consumed. an algorithm for optimal pricing includes transmission cost beside generation cost in electricity supply. the report of optimal pricing calculation has to be sent to all the participants. the marginal cost could be minimized with the inclusion of facts devices in an overloaded transmission system. facts devices can change power flows by using system parameters. using spot pricing and marginal cost theories active and reactive power transition costs is calculated while voltage-dependent load models were observed [18]. incremental cost methodologies are defined in two parts as short run and long run cost. short run marginal/incremental cost or spot pricing is economic. it has some bidding to power transmission systems such as entire transmission costs not recovered, the charges acquired exceedingly volatile, rationale for transmission charges and the transmission system is frequently not in most favorable condition, etc. on the other hand, long-run incremental cost methodology depends on forecast advances in technology innovation, vol. 4, no. 3, 2019, pp. 177-196 180 data with uncertainties. it is difficult to obtain convincing prices as many non-deterministic factors are involved [19]. all system costs (existing transmission system, operation, and expansion) are allocated among the system users in proportion to their eu (extent of use) of the transmission resources. the charge for basic transmission service is usually the component of overall transmission service charges. 2.2. embedded cost-based pricing methods embedded cost is defined as the revenue required paying for all existing or any new facilities added to the power system during the contract for transmission service. in general form adequate remuneration of transmission systems and easy to implement. the embedded cost methods allocate the total system cost among the transmission customers, based on extent of use (eu) rule. in embedded methods, all system costs (existing transmission system, operation, and expansion) are allocated among the system users in proportion to their extent of use (eu) of the transmission resources. they can be classified as rolled-in methods and usage-based methods. the main shortcoming of the rolled-in methods is ignorance of actual system operation. as a result, they are likely to send incorrect economic signals to transmission customers. but this problem has overcome by usage-based methods as it evaluates eu in the framework of either load flow or optimal power flow. embedded costs methods are used by the utilities to allocate existing transmission facilities to the transmission wheeling transaction. table 1 represents the literature survey of embedded wheeling pricing. this table easily describes embedded wheeling cost strategy in electricity market from past to present stage. 2.2.1. active power flow based methods the capacity of transmission network used for a transaction is a function of the magnitude of electric power, transmission lines length, and facilities involved in the active power transaction. capacity value provides an equitable means of allocating the cost of transmission facilities among users of the firm transmission service. it takes full account of current generation cost and capacities as well as the transmission of the demand in space and time. the mw-mile method is used to evaluate the transmission pricing leads to the effective recovery of all embedded costs. it includes an analysis of relative reliability contributions of each generator to the unscheduled transmission capacity in a circuit. all transmission users are liable to wages the actual use of capacity and transmission reserves. in practice, it is improper for those who make limited usage of the network. 2.2.2. real power flow based methods j. bialek proposed a tracing flow method for evaluating the flow of electricity through power networks [20]. it allows quantification of active or reactive power flows from a particular source to a specific load, the contribution from each generator, power load flow and losses in a line. kirschen’s method is based on the solution of a series of load flows. it calculates the contribution of the generator to the loads, line flows and transmission pricing. j. bialek proposed another tracing flow methodology is known as unifying tracing-based methodology of transmission pricing for inter-system trades. it is easy, transparent and fast. it can also deal effectively with circular flows. an up-gradation of mw-mile method was introduced in the year 2001. the upgrade technique is called a mva-mile method which reflects the eu of transmission facilities in the system. it enforces to power flow and considers apparent power. it is reasonable and valid in comparison to the commonly used mw-mile method, but big wheeling charge may pick up the total generation cost. monetary path method is also based on tracing flow concept, which proposes an even measurement for transmission usages by active and reactive powers. reactive power allocation method determines real and imaginary currents to handle the system losses and loop flow. the traces from current sources to current sinks are then converted to power contributions. mva method is economic to resolve difficult reactive power pricing and costing issues. all the methods are based on tracing flow concept for usage quantification. advances in technology innovation, vol. 4, no. 3, 2019, pp. 177-196 181 2.3. composite embedded and marginal cost-based pricing methods tthis methodology includes both the existing system cost and marginal costs of transmission transactions to evaluate the collective transmission pricing part by embedded cost and marginal pricing method. the marginal cost based pricing is used to transmission price services. it requires supplement revenue generation as a pricing scheme is not able to care financially to the transmission service providers. this approach discriminates between operating and embedded costs. it develops separate methods in respect of each of these components. the capacity utilized as well as consistency benefits derived by different users for investment recovery payout of charges for investment recovery are considered. it also includes marginal pricing approach to the recovery of operating cost. it is a simple novel method used for topological analysis of power flows based transmission supplement charge allocation in the network. its result is positive in counter-flow contributions from all the users. revenue of transmission company divided into marginal cost and supplementary charges. the marginal cost is evaluated by frm model to estimate the total transmission cost (cost allocation and remaining charges). in supplementary charges, the cost is evaluated by prm model. whereas, locational charges and post stamp charges method is used to estimate the supplementary charges. both methods are used to evaluate remaining charges in supplementary cost. the supplementary charges are allocated in real power as well as reactive power load through mw-mile, mva-mile method. this charge for usage of a separate transmission asset is divided into a locational and non-locational component. wheeling charging strategy and unused capacity of the asset are shown in fig. 1. fig. 1 wheeling charging strategy 2.4. amp-mile method the amp-mile method is an embedded cost allocation method for the medium voltage distribution network. it is based on the extent of use (eu) for circuits is measured in terms of the contribution of each customer to the current flow i.e. the power to current distribution factors (apidflk t & rpidflk t ), at any instant of time. it is the least control of system stability (steady state and transient) in the distribution system. current capacity increases up to the thermal limit. so current flows may attribute network customers, therefore method is acknowledged as “amp-mile” or “i-mile” methodology [1]. thus different extent of use (eu) is found out using current distribution factors and used to accomplish allocation of cost. unambiguously accounts for flow direction to provide better long-term price signals and to alleviate potential constraints [20]. steps to allocate embedded cost, as well as remaining cost are as follows advances in technology innovation, vol. 4, no. 3, 2019, pp. 177-196 182 t t t lk k lk t l apidf pl aeoul ai   (1) t t t lk k lk t l apidf pg aeoug ai   (2) t t t lk k lk t l rpidf ql reoul ai   (3) t t t lk k lk t l rpidf qg reoug ai   (4) where, 𝐴𝐸𝑜𝑈𝐿𝑙𝑘 𝑡 is active eu of circuit 𝑙 at time 𝑡 due to demand at 𝑘𝑡ℎ bus, 𝐴𝐸𝑜𝑈𝐺𝑙𝑘 𝑡 is active eu of circuit 𝑙 at time 𝑡 due to generation at 𝑘𝑡ℎ bus, 𝑅𝐸𝑜𝑈𝐿𝑙𝑘 𝑡 is reactive eu of circuit 𝑙 at time 𝑡 due to demand at 𝑘𝑡ℎ bus, 𝑅𝐸𝑜𝑈𝐺𝑙𝑘 𝑡 is reactive eu of circuit 𝑙 at time 𝑡 due to generation at 𝑘𝑡ℎ bus, 𝐴𝑃𝐼𝐷𝐹𝑙𝑘 𝑡 is active power to current distribution factor of circuit 𝑙 at time 𝑡 due to demand at 𝑘𝑡ℎ bus, 𝑅𝑃𝐼𝐷𝐹𝑙𝑘 𝑡 is active power to current distribution factor of circuit 𝑙 at time 𝑡 due to demand at 𝑘𝑡ℎ bus. various methods are not able to quantify the eu of a transmission network for both active and reactive power flows. the amp-mile method became the base of the research. limitations of amp-mile method have been identified through numerical value. based on these corrections modified amp-mile method is proposed. it recommends applicability on ehv networks by putting more prominence on stability limits. a step ahead introduced a new usage-based cost allocation techniquemva utility factor method with two models: full recovery and partial recovery model. it carries the advantages of modified amp-mile method. it is based on marginal participation; therefore obtained utility factors are prone to the choice of slack bus. therefore distinct slack bus notion has been proposed to allocate the embedded cost of ehv networks. distinct slack bus notion is different from dispersed slack bus concept. the intermediate stages of nonlinear sensitivities are found using modified nr based load flow. the relationship between flow and power injection/withdrawal is nonlinear. the novel sensitivity patterns could assist iso to forecast day-ahead transmission. in full recovery model total eu for all the lines remains unity under all loading conditions. 3. proposed methodology the amp-mile method has some limitations. when the system has fully loaded this method do not recover the full embedded cost. it is relevant only on radial networks since currents are comparative to the thermal capacity of the distribution network (high r/x ratio). it is stated that circuit currents are an approximately linear function of active and reactive power at the bus in a radial network. in amp-mile a reconciliation factor is needed to find eu factors for a given line sum to unity. it confers only two types of sensitivities (eq.(5)) in the form of transmission factors cupf (current utility active factor) and cuqf (current utility reactive factor). the common transmission factor for demand/generation transmutation at the same bus with a difference of sign. l dk l gk i p i p     (5) l dk l gk i q i q     (6) where, 𝐼𝑙 is the absolute value of the current through circuit l, 𝑃𝑑𝑘 is the active power withdrawal due to demand at 𝑘 𝑡ℎ bus, 𝑃𝑔𝑘 is the active power withdrawal due to generation at 𝑘𝑡ℎ bus, 𝑄𝑑𝑘 is the reactive power withdrawal due to demand at 𝑘 𝑡ℎ bus, 𝑄𝑔𝑘 is the reactive power withdrawal due to generation at 𝑘 𝑡ℎ bus. advances in technology innovation, vol. 4, no. 3, 2019, pp. 177-196 183 3.1. modified amp-mile method the modified amp-mile method identifies the nonlinear or linear nature of sensitivities depending on location and topological conditions. its charges allocation has not been stable at variable load levels and different period of time. it indicates that the current sensitivity indices cupftlk and cuqftlk are exhibiting nonlinear nature with respect to active and reactive powers (injection/withdrawal) at a bus of ehv network. modified nr based load flow is used to find current sensitivity indices. in the modified amp-mile method, reconciliation factor is not required to furnish the total eu for a given line equal to unity. it selects new distinct slack bus perception to resolve the load flow values. entire embedded cost of ehv networks is allocated by using non-linear sensitivities and new distinct slack bus notion. if a system has large chances of increase in load and generation on the same bus, comparisons could be made between ∂il ∂pdk⁄ and ∂il ∂pgk⁄ . therefore, current sensitivity indices are expressed as t ldk l dk cupf i p   (7) t lgk l gk cupf i p   (8) t ldk l dk cuqf i q   (9) t lgk l gk cuqf i q   (10) where, 𝐶𝑈𝑃𝐹𝑙𝑑𝑘 𝑡 is current utility active factor of 𝑙𝑡ℎline w.r.t.𝑘𝑡ℎ demand bus at 𝑡𝑡ℎinstant, 𝐶𝑈𝑃𝐹𝑙𝑔𝑘 𝑡 is current utility active factor of 𝑙𝑡ℎline w.r.t.𝑘𝑡ℎ generator bus at 𝑡𝑡ℎ instant, 𝐶𝑈𝑄𝐹𝑙𝑑𝑘 𝑡 is current utility reactive factor of 𝑙𝑡ℎline w.r.t.𝑘𝑡ℎ demand bus at 𝑡𝑡ℎinstant, 𝐶𝑈𝑄𝐹𝑙𝑔𝑘 𝑡 is current utility reactive factor of 𝑙𝑡ℎ line w.r.t.𝑘𝑡ℎ generator bus at 𝑡𝑡ℎinstant. the applicability of the modified amp-mile method is on ehv networks, by putting more prominence on stability limits, instead of thermal capability. in eq. (5), there should be dissimilarity between cupflk t and cuqflk t for load/generation given in eqs. (7-10). the expressions of dissimilar eu’s are where, 𝐴𝐸𝑈𝐷𝑙𝑘 𝑡 is active extent of use by 𝑘𝑡ℎ bus demand for 𝑙𝑡ℎ line at 𝑡𝑡ℎ instant, 𝐴𝐸𝑈𝐺𝑙𝑘 𝑡 is active extent of use by 𝑘𝑡ℎ bus generation for 𝑙𝑡ℎ line at 𝑡𝑡ℎ instant, 𝑅𝐸𝑈𝐷𝑙𝑘 𝑡 is reactive extent of use by 𝑘𝑡ℎ bus demand for 𝑙𝑡ℎ line at 𝑡𝑡ℎ instant, 𝑅𝐸𝑈𝐺𝑙𝑘 𝑡 is reactive extent of use by 𝑘𝑡ℎ bus generation for 𝑙𝑡ℎ line at 𝑡𝑡ℎ instant, 𝑃𝑑𝑘 𝑡 is active demand on 𝑘𝑡ℎ bus at 𝑡𝑡ℎ instant, 𝑃𝑔𝑘 𝑡 is active generation on 𝑘𝑡ℎ bus at 𝑡𝑡ℎinstant, 𝑄𝑑𝑘 𝑡 is reactive demand on 𝑘𝑡ℎ bus at 𝑡𝑡ℎ instant, 𝑄𝑔𝑘 𝑡 is reactive generation on 𝑘𝑡ℎ bus at 𝑡𝑡ℎinstant, 𝐼𝑙 𝑡 is absolute current in 𝑙𝑡ℎ line at 𝑡𝑡ℎinstant. the reconstruction of the algorithm to evaluate cupflk t and cuqflk t inherently attributes the efficacy of slack bus power. it provides true sensitivities, re-establishing and offered a close approximation of circuit currents. the expression of the absolute value of il t will turn to substitute used circuit cost at tth instant (uccl t ) equal to unity to find the allocation of embedded cost. consequently adapted circuit cost at tthinstant (accl t ) would be equal to ccl t and therefore the relation of locational charges change as given below ? ) t t t t t t lk ldk dk l l dk dk l aeud cupf p i i p p i      (11) ? ) t t t t t t lk lgk gk l l gk gk l aeug cupf p i i p p i      (12) ? ) t t t t t t lk ldk dk l l dk dk l reud cuqf q i i q q i      (13) ? ) t t t t t t lk lgk gk l l gk gk l reug cuqf q i i q q i      (14)   1 nbus t t t t t t t t t l ldk dk lgk gk ldk dk lgk gk k i cupf p cupf p cuqf q cuqf q          (15) advances in technology innovation, vol. 4, no. 3, 2019, pp. 177-196 184 where, 𝐴𝐿𝑘 𝑡 is active locational charge due to active demand on 𝑘𝑡ℎ bus at 𝑡𝑡ℎinstant, 𝐴𝐺𝑘 𝑡 is active locational charge due to active generation on 𝑘𝑡ℎ bus at𝑡𝑡ℎinstant, 𝑅𝐿𝑘 𝑡 is reactive locational charge due to reactive demand on 𝑘𝑡ℎ bus at 𝑡𝑡ℎ instant, 𝑅𝐺𝑘 𝑡 is reactive locational charge due to reactive generation on 𝑘𝑡ℎ bus at 𝑡𝑡ℎinstant, 𝐶𝐶𝑙 𝑡 is a level cost for each hour. the expression of remaining circuit charges (𝑅𝐶𝐶 𝑡) are so the total cost of the modified amp-mile method in the full recovery model is the modified amp-mile method increases allocation equivalent to full recovery model in proportion to assorted eu’s. the eu’s of all circuits would be unity under all loading conditions and no need to calculate remaining (supplementary or non-vocational) charges. therefore, keeping ucc equal to unity and evaluate transmission charges plus capacity charges simultaneously. transmission network participants have to pay for used/unused capacity in proportion to their eu. it is justified by the need for system meeting reliability, stability and security criteria for all customer. 3.2. mva utility factor method the mva utility factor method allocates the entire embedded cost of transmission networks. it carries the advantages of previously discussed modified amp-mile method. the non-linear patterns of mva utility factors have been furnished and distinct slack bus notion has been promoted to allocate the embedded cost of power networks. it provides better promises for payments to counterflow creators and gives assurance for prudent implementation. in the proposed technique individual participant’s impact on the system is recognized through mva flow caused by them. therefore, a method is called an mva utility factor method. it is a significant method because no question arises in relation to current limits for the reason that mva flows can be increased under specified constraints of the power network. it wiped out limits of the modified amp-mile method and exploits load flow and derives non-linear sensitivities (mvauf). this illustrates a true understanding of the network and causes a fair allocation. in mva utility factor method, unequal sensitivities at a bus having active/reactive generation and load are represented as [21]. where, 𝑀𝑉𝐴𝑃𝑈𝐹𝑙𝑑𝑘 𝑡 is mva utility active factor of 𝑙𝑡ℎline w.r.t.𝑘𝑡ℎ bus demand at 𝑡𝑡ℎinstant, 𝑀𝑉𝐴𝑃𝑈𝐹𝑙𝑔𝑘 𝑡 is mva utility active factor of 𝑙𝑡ℎ line w.r.t.𝑘𝑡ℎ bus generator at 𝑡𝑡ℎinstant, 𝑀𝑉𝐴𝑄𝑈𝐹𝑙𝑑𝑘 𝑡 is mva utility reactive factor of 𝑙𝑡ℎline w.r.t.𝑘𝑡ℎ bus demand at 𝑡𝑡ℎ instant, 𝑀𝑉𝐴𝑄𝑈𝐹𝑙𝑔𝑘 𝑡 is mva utility reactive factor of 𝑙𝑡ℎline w.r.t.𝑘𝑡ℎbus generator at 𝑡𝑡ℎinstant. 1 linen t t t k lk l l al aeud cc    (16) 1 linen t t t k lk l l ag aeug cc    (17) 1 linen t t t k lk l l rl reud cc    (18) 1 linen t t t k lk l l rg reug cc    (19) 1 [ ] 0 linen t t t l l l rcc cc cc     (20) ? t t t t k k k k total cost al ag rl rg    (21) t t lgk ldk mvapuf mvapuf  (22) t t lgk ldk mvaquf mvaquf  (23) advances in technology innovation, vol. 4, no. 3, 2019, pp. 177-196 185 the transmission network though ehv network does not follow the same for some of the generator buses using load flow. it obtains either equal or unequal sensitivities at generator buses in ehv networks. therefore, eqs. (22-23) does not reflect true operating conditions. many cost allocation methodologies suggested following load flow to avoid unequal sensitivities. it established four different utility factors corresponding to generator buses in ehv networks irrespective of equal or unequal sensitivities. mva flow of a line can be expressed using utility factors of eqs. (24-27) are where, 𝑀𝑉𝐴𝑙 𝑡 is the absolute value of mva through circuit l, 𝑃𝑑𝑘 𝑡 is the active power withdrawal due to demand at 𝑘𝑡ℎ bus at 𝑡𝑡ℎ instant, 𝑃𝑔𝑘 𝑡 is the active power withdrawal due to generation at 𝑘𝑡ℎ bus at 𝑡𝑡ℎ instant, 𝑄𝑑𝑘 𝑡 is the reactive power withdrawal due to demand at 𝑘𝑡ℎ bus at 𝑡𝑡ℎ instant, 𝑄𝑔𝑘 𝑡 is the reactive power withdrawal due to generation at 𝑘𝑡ℎ bus at 𝑡𝑡ℎ instant. formulation of mvauf’s is evaluated using modified nr based load flow. thus attributes the effect of slack bus power. these buses are self-regulating and bounds for each of the generators. it maintains a constant voltage at buses; consequently, no change in line flow occurs due to change in a reactive generation. expression of absolute mva in lthline at tthinstant (mval t ) are the mvauf’s from eqs. (24-27) are employed for the evaluation of eou by each participant, equations given as: it is concluded that total eu due to all participants is unity even though the system is not fully loaded. this methodology is utilized to develop two types of allocation models; partial recovery model (prm) and full recovery model (frm). 3.2.1. partial recovery model (prm) in this modal, a part of the embedded cost is allocated on the basis of electricity usage and remaining charges are imposed on the participants. the two types of charges are; pubc: charge allocated to participants based on the real usage of the network. rc: this portion of allocation reflects a charge to recover the cost of the unused network capacity. it is revealing the security issue of the power system and has to be imposed on all participants. expressions of pubc and rc are: t t t ldk l dk mvapuf mva p   (24) t t t lgk l gk mvapuf mva p   (25) t t t ldk l dk mvaquf mva q   (26) t t t lgk l gk mvaquf mva q   (27) 1 nbus t t t t t t t t t l ldk dk lgk gk ldk dk lgk gk k mva mvapuf p mvapuf p mvaquf q mvaquf q          (28) t t t t lk ldk dk l aeud mvapuf p mva  (29) t t t t lk lgk gk l aeug mvapuf p mva  (30) t t t t lk ldk dk l reud mvaquf q mva  (31) t t t t lk lgk gk l reug mvaquf q mva  both models by a utility depend on its transmission system and extent of deregulation espoused. (32) 1 nline t t t dk lk l l pubcp aeud acc    (33) advances in technology innovation, vol. 4, no. 3, 2019, pp. 177-196 186 where, 𝑃𝑈𝐵𝐶𝑃𝑑𝑘 𝑡 is partial recovery usage-based charges for active demand on 𝑘𝑡ℎ bus at 𝑡𝑡ℎ instant, 𝑃𝑈𝐵𝐶𝑃𝑔𝑘 𝑡 is partial recovery usage-based charges for active generation on 𝑘𝑡ℎ bus at 𝑡𝑡ℎ instant, 𝑃𝑈𝐵𝐶𝑄𝑑𝑘 𝑡 is partial recovery usage-based charges for reactive demand on 𝑘𝑡ℎ bus at 𝑡𝑡ℎ instant, 𝑃𝑈𝐵𝐶𝑄𝑔𝑘 𝑡 is partial recovery usage-based charges for reactive generation on 𝑘𝑡ℎbus at 𝑡𝑡ℎ instant. let 𝐶𝐶𝑙 𝑡 is levelized hourly cost of 𝑙𝑡ℎcircuit and annual circuit cost will be𝐶𝐶𝑙 𝑡 × 8760.then corresponding adapted circuit cost 𝐴𝐶𝐶𝑙 𝑡 at 𝑡𝑡ℎ instant is where𝑈𝐶𝐶𝑙 𝑡 is the used circuit capacity of line 𝑙 for time 𝑡, and defined by where 𝐶𝐴𝑃𝑙 is mva capacity of the line. the remaining charge (rc) express as 3.2.2. full recovery model (frm) accomplishes total recovery of embedded cost by substituting 𝑈𝐶𝐶𝑙 𝑡 unity in eq. (37); consequently, 𝐴𝐶𝐶𝑙 𝑡 would be equal to 𝐶𝐶𝑙 𝑡. expressions of fubc are: where 𝐹𝑈𝐵𝐶𝑃𝑑𝑘 𝑡 is full recovery usage-based charges for active demand on 𝑘𝑡ℎbus at 𝑡𝑡ℎ instant, 𝐹𝑈𝐵𝐶𝑃𝑔𝑘 𝑡 is full recovery usage-based charges for active generation on 𝑘𝑡ℎbus at 𝑡𝑡ℎinstant, 𝐹𝑈𝐵𝐶𝑄𝑑𝑘 𝑡 is full recovery usage-based charges for reactive demand on 𝑘𝑡ℎbus at 𝑡𝑡ℎinstant, 𝐹𝑈𝐵𝐶𝑄𝑔𝑘 𝑡 is full recovery usage-based charges for reactive generation on 𝑘𝑡ℎbus at 𝑡𝑡ℎinstant. in frm, the eq. (38) for remaining charges can be modified accordingly and expressed as: using an analysis of both methodologies a flow chart is shown in fig. 2. data input and newton raphson analysis is the same in both methods. current utility factor logic has been used in mod amp mile method similarly; 1 nline t t t gk lk l l pubcp aeug acc    (34) 1 nline t t t dk lk l l pubcq reud acc    (35) 1 nline t t t gk lk l l pubcq reud acc    (36) t t t l l l acc ucc cc  (37) / t t l l l ucc mva cap (38) 1 nline t t t l l l rc cc acc       (39) 1 nline t t t dk lk l l fubcp aeud cc    (40) 1 nline t t t gk lk l l fubcp aeug cc    (41) 1 nline t t t dk lk l l fubcq reud cc    (42) 1 nline t t t gk lk l l fubcq reug cc    (43) 1 0 nline t t t l l l rc cc cc        (44) advances in technology innovation, vol. 4, no. 3, 2019, pp. 177-196 187 mva sensitivity analysis has been used to calculate the mva utility factor method. used circuit capacity (ucc=1) unity value is used to calculate annual circuit cost which is defined frm wheeling value in both methods. mva utility factor prm model represent adapted circuit cost to evaluate cost allocation (ca) and remaining charges (rc) in the network. start input bus data, line data, unit cost of transmission line calculate cost of each transmission line and perform base case load flow using newton raphson method modified amp-mile method mva utility factor method find four different’s mvapuf value by varying demands / generations at all respective buses formulate the total mva by using demand/ generation sensitivity value calculate extent of use of each transmission line by individual generator and load find four different’s current utility cupf value by varying demands / generations at all respective buses partial recovery model full recovery model calculate annual circuit cost using used circuit capacity(ucc=1) calculate partial recovery cost allocation (ca) charges based on actual usage of the network (pubc) calculate adapted circuit cost (acc) find the remaining charges(rc) calculate total wheeling cost(rc+ca) in prm model stop print results calculate full recovery cost allocation (ca) charges based on actual usage of the network (fubc) calculate total wheeling cost(ca) in frm model formulate the total current (i) by using demand/ generation sensitivity value calculate extent of use of each transmission line by individual generator and load find the locational charges using used circuit capacity (ucc=1) calculate total wheeling cost using frm model fig. 2 flow chart of mod amp-mile and mva utility factor methods 4. case study and results the proposed methodologies have to be employed on some real-time system or ieee standard bus system to prove its efficiency and reliability. in this paper, a real-time 37 bus system and one ieee standard 6 bus system is used as a case study and respective results are discussed. in the case study, modified amp-mile method is applied using full recovery modal, whereas, mva utility factor method is applied using both modals i.e. full and partial recovery modal. single line diagram of real-time 37 bus system is shown in fig.3. the system has 2 generation bus and 48 transmission corridors at 220 kv and 132 kv voltage level. the generation capacity of 1300 mw is assumed as base load condition and total load connected on the system is 911 mw. it is assumed in this analysis that load customer would pay 100% of the transmission cost of services to the transmission utility. the annual revenue requirement of transmission facility is 843.56 crs-inr. the embedded cost to be allocated is assumed proportional to the length of individual transmission lines in rupee/hr. the comprehensive detail and various parameters of the 37 bus system are given in table 2. data of total load connected is collected of each feeder whereas, bus data represent the bus voltage, active/reactive value of generated power and total load connected in different buses. in line data values represent the load connection information for every node , value of the impedance in each line of the system and line charging value in the transmission network. voltage and power factor value is collected for evaluating the bus data. advances in technology innovation, vol. 4, no. 3, 2019, pp. 177-196 188 fig. 3 north indian real test system (power map of 37 bus hadoti region) table 2 cost of generation and load (rupee / hr.) for real 37 bus system bus data of 37-bus transmission system line data of 37-bus transmission system bus no. bus voltage power generated load node connection impedance(p.u.) line charging magnitude angle p q p q from to r x (p.u.) b/2 (pu) (deg) (mw) (mvar) (mw) (mvar) 1 2 0.00024 0.001275 0 1 1 0 1300 0 0 0 1 5 0.0056 0.02975 0 2 1.01 0 0 0 0 0 1 3 0.00896 0.0476 0 3 1.02 0 0 0 0 0 2 6 0.00672 0.0357 0 4 1 0 0 0 0 0 2 15 0.0306397 0.0794612 0 5 1.02 0 0 0 0 0 2 21 0.0132314 0.0343144 0 6 1 0 0 0 0 0 2 22 0.01001 0.02596 0 7 1 0 0 0 18 7.1 2 23 0.0206388 0.0535248 0 8 0.98 0 0 0 28 10.17 2 7 0.0295022 0.0765112 0 9 0.99 0 0 0 10 3.3 2 18 0.03822 0.09912 0 10 0.98 0 0 0 20 7.27 2 36 0.000455 0.00118 0 11 0.98 0 0 0 20 7.24 3 4 0.00456384 0.0242454 0 12 0.99 0 0 0 26 8.52 3 8 0.0352625 0.09145 0 13 1 0 0 0 20 6.57 3 17 0.00864045 0.0224082 0 14 1.03 0 0 0 14 4.61 3 35 0.000455 0.00118 0 15 0.98 0 0 0 40 11.68 4 28 0.01630064 0.0865972 0 16 0.98 0 0 0 20 5.82 4 9 0.0186732 0.0484272 0 17 1 0 0 0 35 12.69 4 12 0.05027295 0.1303782 0 18 1 0 0 0 25 9.87 4 13 0.03069976 0.079617 0 19 0.98 0 0 0 30 7.53 4 17 0.0142506 0.0369576 0 20 0.98 0 0 0 40 10.04 4 19 0.03586401 0.09301 0 21 1 0 0 0 42 16.6 4 34 0.000455 0.00118 0 22 1 0 0 0 80 31.64 5 15 0.00728 0.01888 0 23 0.99 0 0 0 85 33.6 6 31 0.00738848 0.0392513 0 24 0.98 0 0 0 29 7.28 6 21 0.0219583 0.0569468 0 25 1 0 0 0 20 4.98 6 23 0.0219583 0.0569468 0 26 0.99 0 0 0 35 8.76 6 24 0.0281372 0.0729712 0 27 1 0 0 0 30 7.53 6 37 0.000455 0.00118 0 28 1 0 300 0 0 0 8 9 0.175266 0.0454536 0 29 1.02 0 0 0 0 0 8 10 0.0219947 0.0570412 0 30 0.98 0 0 0 35 8.76 8 11 0.053235 0.13806 0 31 1.02 0 0 0 0 0 12 14 0.0344799 0.0894204 0 32 0.98 0 0 0 25 6.25 15 16 0.02548 0.06608 0 33 1 0 0 0 35 8.76 18 20 0.02002 0.05192 0 34 1 0 0 0 42 13.8 19 20 0.02928289 0.0759424 0 35 1 0 0 0 25 9.09 20 24 0.0228046 0.0591416 0 36 1 0 0 0 50 19.75 20 25 0.0154245 0.040002 0 37 1 0 0 0 32 8.02 20 30 0.0250068 0.0648528 0 22 23 0.00455 0.0118 0 advances in technology innovation, vol. 4, no. 3, 2019, pp. 177-196 189 table 2 cost of generation and load (rupee / hr.) for real 37 bus system (continued) 24 31 0.0153517 0.0398132 0 25 29 0.0336245 0.087202 0 26 29 0.0104923 0.0272108 0 27 29 0.0173628 0.0450288 0 28 29 0.00668 0.0354875 0 29 31 0.0079472 0.0422195 0 29 33 0.000455 0.00118 0 30 32 0.042345868 0.1098533 0 30 31 0.00637 0.01652 0 4.1. non-linear nature of sensitivity indices load flow is being used either for transmission network or ehv network to obtain equal sensitivities and unequal sensitivities. an inequality in curve nature is depending upon magnitude. whereas, magnitude depending on the choice of slack bus, location of generator bus and transmission line.during load flow, any bus can be assigned as a slack bus. load flow neglects longitudinal resistance, the conductance of network elements, reactive power flow and considers all voltages equal to unity. in result, the magnitude of injection/withdrawal variation at any generation bus directly influences loss compensation. lines connected to the slack bus would have different flow variations and lines away from the slack bus would have less impact. the equivalent feature is narrated in rudnick’s method [4]. the evaluation of the total impact of injection/withdrawal of electrical power is independent for each bus. amount of allocation depends on currency cost impact. at bus payments are made by the utility to participants, reflecting the proposed technique identifies the negative eu due to reactive demand. (a) non linear cupf t 1g1 for 37 bus system (b) non linear cupf t 21d10 for 37 bus system (c) non linear cupft9d7 for 37 bus system (d) non linear cuqft16d9 for 37 bus system fig. 4 nonlinear cupf curve of generation and load in 37 bus system fig. 4(a-d) shows the nonlinear sensitivity behavior curves for modified amp-mile method. the real or reactive mva sensitivities would not be identical at generator buses. it exhibits nonlinear nature of sensitivities depending upon location and topological conditions. fig. 4(a-d) shows the nonlinear sensitivity curves for mva utility factor method. the estimation and recognition pattern for mvauf’s of non-linear lines with respect to different injections/withdrawals at each bus is obtained with the help of the nr algorithm. moreover, allocated charges would not radically be stable over differing load level due to advances in technology innovation, vol. 4, no. 3, 2019, pp. 177-196 190 non-linear sensitivities. patterns of sensitivities assessed for variation of p and q at all buses for different loading by employing load flow analysis. the sensitivity patterns have many advantages for a deregulated market. (a) provide price signals for the future generation/demand expansion. (b) forecast transmission pricing, if used along with eu values. (c) congestion anticipation and used to manage congestion either by curtailing the demand. (d) assist iso to carry out the day ahead scheduling in the deregulated electricity market. (a) non linear mvapuft16g1 for 37 bus system (b) non linear cupft21d10 for 37 bus system (c) non-linear cupft9d7 for 37 bus system (d) non-linear cuqft16d9 for 37 bus system fig. 5 nonlinear mvapuf curve of generation and load in 37 bus system 4.2. cost allocation andcost curve evaluate the total cost in rupee/hr by using mathematical operation of both the proposed methods. for ieee 6 bus standard system, comparison of the proposed model (partial recovery model and full recovery model) has been brought down. cost allocated (ca), remaining charge (rc), and embedded cost (ec) are to be compared for at all buses by employing different techniques. table 3 shows the comparison of all the costs including rc and % ec of each bus. by observing table 3, execution of the mva utility factor method causes a striking reduction in cost allocated to bus 1 (slack bus), as positive payments by all other methods are turning to negative payments. likewise, cost allocation is negative to bus 5 load and positive to the first generator in amp-mile method. it shows reflecting payments are made by transmission utility to load and recovered from generators, which seems unjustified. this problem gets resolved through proposed mva utility factor method using either partial recovery model or full recovery model. realization of proposed modified amp-mile and mva utility factor method proved to be superior over mw-mile method, mva-mile method, and amp-mile method. its advantages like frm of ehv networks contrasting amp-mile tackles reactive power unlike mw-mile and anticipates the direction of reactive power not like mva-mile. the cost to be allocated in modified amp-mile and mva utility factor method is assumed proportional to the length of transmission lines in rupee/hr. in ieee 6 bus system the total cost evaluated in modified amp-mile and mva utility factor methods in frm condition is 5435 rupee/hr. whereas, in prm condition, mva utility factor allocated cost is 948.137 rupee/hr. it has been observed from table 2, that 100% cost is assigned advances in technology innovation, vol. 4, no. 3, 2019, pp. 177-196 191 as cost allocated or zero amount as remaining charges in modified amp-mile and mva utility factor method (frm only). all the other existing methods along with proposed mva utility factor method (prm) have a significant amount as remaining charges. hence, these methods do not allocate 100% embedded cost. the negative sign represent the return cost given by the transmission company. table 3 comparison of cost allocation in ieee 6-bus system by different methodologies under 100% loading bus no. mw-mile (rupee/hr.) mva-mile (rupee/hr.) ampmile (rupee/hr.) amp-mile (lf) (rupee/hr.) modified amp-mile (rupee/hr.) mva uf method (prm) (rupee/hr.) mva uf method (frm) (rupee/hr.) bus 1 281.5 27.1 239.31 206.0269 1001.3 -119.75 -286.9 bus 2 145.8 227.9 19.65 -99.7790 -485.1 66.96 69.1 bus 3 350.0 487.2 116.36 -152.8764 -1050.8 133.47 814.7 bus 4 254.7 411.7 100.47 253.0793 -103.1 220.202 664.0 bus 5 588.8 592.9 -13.13 440.9396 4519.0 388.798 3024.7 bus 6 553.7 400.3 2.04 271.1701 1553.6 258.456 1149.4 ca 2174.5 2146.6 462.67 918.5604 5435 948.137 5435 rc 3260.5 3288.4 4972.33 4516.4396 0 4486.86 0 %ec 40 39.5 8.51 16.90 100 17.45 100 the locational and remaining charges allocation at different loadings has been estimated using diverse cost allocation techniques. the evaluated cost allocated under the different percentage of base case (bc) loading conditions by different methodologies is shown in table 4. it has been observed that the cost allocation by employing modified amp-mile method is constant for all loading conditions. the percentage embedded cost is 100% are zero remaining charges are allocated in modified amp-mile method irrespective of loading of the system. hence, modified amp-mile method is seen to be the most excellent method for full recovery modal only, under each loading condition. before evaluating the cost of the system, check the degree of congestion at bc loading specifically for overloading, to keep the system secured. table 4 embedded cost (rupee/hr.) allocation in ieee 6-bus system by different methods at different % loading methods loadings 50% 75% 100% 125% 150% mw-mile 1488.6 1860.6 2174.5 3045.2 2980.0 rem. cost (mw) 3946.4 3574.4 3260.5 2389.8 2455 mva-mile 1288.8 1672.6 2146.6 4100.3 3371.6 rem. cost (mva) 4146.2 3762.4 3288.4 1334.7 2063.4 amp-mile (base) 190.153 252.19 462.67 713.08 1107.6 rem. cost(i-mile) 5244.8 5182.8 4972.33 4721.9 4327.4 amp-mile(lf) 882.8 847.6 918.56 1098.9 1264.1 rem. cost (amp-mile(lf)) 4552.2 4587.4 4516.44 4336.1 4170.9 mod. amp-mile 5435.0 5435.0 5435.0 5435.0 5435.0 rem. cost (mod. amp-mile) 0 0 0 0 0 any generator bus can be assigned as a slack bus for estimation of cost and to find cuf’s. consider another generator bus as the new slack bus and then cost allocation associated withthe new slack bus has been evaluated. all the costs, ca, rc, and %ec has been evaluated for the new slack bus by employing different methodologies and compare the results as shown in table 5. the utilities have a different region with significant local load and generation. the injection in a given region may cause an increment in the circuit flows all around the country. resulting tariffs are sometimes nonspontaneous with generators that are close to load centers in a given region receives high tariffs. therefore the currency cost impact of the different slack bus by implementing distinct slack bus notion is used. results for depiction on the ieee 6 bus system are given in table 4 at base case (bc) loading. embedded cost is evaluated using modified amp-mile and mva utility factor method for 37 bus system in a similar manner employed in ieee 6 bus system. frm costs were evaluated using both the above method are shown in table 6 (a) and advances in technology innovation, vol. 4, no. 3, 2019, pp. 177-196 192 table 6(c). whereas, prm costs are shown in table 4 (b) for mva utility factor method only. table 4 illustrates the generation and load cost allocates at different buses for active and reactive power. the allocations by frm at all busses are given in table 6(a) and (c). table 5 cost impact (rupee/ hr) of the new slack bus at 100% base bus no. amp-mile (lf) slack bus -bus1 mod. amp-mile slack bus-bus1 amp-mile(lf) slack bus –bus 2 mod. amp-mile slack bus –bus 2 bus 1 206.0269 1001.3 60.4148 360.1 bus 2 -99.7790 -485.1 234.5645 1042.3 bus 3 -152.8764 -1050.8 -87.6418 -2317.3 bus 4 253.0793 -103.1 144.6832 -272.8 bus 5 440.9396 4519.0 344.1372 4615.5 bus 6 271.1701 1553.6 157.5987 2007.2 ca 918.5604 5435 853.7567 5435 rc 4516.4396 0 3727.4866 0 %ec 16.90 100 15.71 100 it has been observed that the payments are made by transmission utility to participants due to the counter flow of power. in 37 bus system, the total cost evaluated is 13699 rupee /hr. in modified amp-mile and mva utility factor methods (frm condition only.) whereas, in prm condition, mva utility factor allocated cost is12699 rupee/hr. in table 6 real 37 bus system frm model are recover the full allocation cost (ac) in both the technique. this result analysis represents the beneficial effect of the proposed work. the remaining charges are also calculated by the postage stamp method. compare the remaining charges cost in ieee 6 bus and real 37 bus system. the low remaining charges in real 37 bus test system are showing the good stability in pricingwheeling market. that is more beneficial for generators and consumer in the electrical energy market (real-world strategies). (a) modified amp-mile cost curve using frm in 6 bus system (b) mva utility factor cost curve using prm & frm in 6 bus system (c) modified amp-mile cost curve using frm in 37 bus system (d) mva utility factor cost curve using prm & frm in 37 bus system fig. 6 prm and frm cost curves of mva utility factor methods advances in technology innovation, vol. 4, no. 3, 2019, pp. 177-196 193 table 6 cost of generation and load (rupee / hr.) for real 37 bus system (a)modified amp-mile cost (frm) (b) mva utility factor cost(prm) (c) mva utility factor cost (frm) bus active active reactive bus active active reactive bus active active reactive no. load gen load no. load gen load no. load gen load 1 0 -24761 0 1 0 7371 0 1 0 4155 0 2 0 0 0 2 0 0 0 2 0 0 0 3 0 0 0 3 0 0 0 3 0 0 0 4 0 0 0 4 0 0 0 4 0 0 0 5 0 0 0 5 0 0 0 5 0 0 0 6 0 0 0 6 0 0 0 6 0 0 0 7 306.5 0 102.1 7 69.996 0 10.63 7 352.1 0 53.52 8 1832.8 0 339.4 8 311.17 0 48.64 8 674.9 0 103.3 9 495.7 0 39.2 9 48.233 0 5.087 9 -6 0 -8.438 10 1570.8 0 308.8 10 285.67 0 42.96 10 763.2 0 110.2 11 1929.3 0 397.1 11 380.01 0 56.01 11 1153.3 0 164.8 12 1892.2 0 236 12 294.2 0 37.56 12 359.9 0 45.51 13 1459.2 0 181.8 13 160.03 0 18.95 13 358.8 0 38.64 14 1372.9 0 210.7 14 226.81 0 27.36 14 632.1 0 70.26 15 404.9 0 46.7 15 124.06 0 10.79 15 529.7 0 46.38 16 507.9 0 70.7 16 132.13 0 11.11 16 597.2 0 50.32 17 1413.5 0 157.8 17 159.57 0 29.63 17 56.7 0 22.32 18 -272.2 0 -31.5 18 127.24 0 17.02 18 237.7 0 37.08 19 -1100.4 0 -165.4 19 270.3 0 22.51 19 797.9 0 64.66 20 -52.1 0 -51.6 20 311.26 0 22.83 20 617.9 0 54.82 21 -326.2 0 -17.3 21 34.024 0 3.751 21 -56.1 0 -11.68 22 37 0 -11.5 22 132.65 0 34.6 22 59.7 0 28.67 23 209.8 0 -74.8 23 203.7 0 51.1 23 142.4 0 48.96 24 997 0 74.2 24 160.32 0 9.624 24 121.6 0 1.48 25 202.8 0 -14.9 25 122.56 0 9.513 25 238.3 0 19.57 26 1384.6 0 49.9 26 95.874 0 8.288 26 130.1 0 7.14 27 1290.4 0 54.1 27 110.22 0 8.769 27 221.2 0 12.6 28 0 11822 0 28 0 162.1 0 28 0 289.1 0 29 0 0 0 29 0 0 0 29 0 0 0 30 1441.2 0 54.3 30 161.64 0 8.622 30 -119.7 0 -35.27 31 0 0 0 31 0 0 0 31 0 0 0 32 1623.4 0 154.4 32 267.23 0 16.05 32 457.1 0 10.19 33 1252.2 0 45.4 33 47.783 0 5.438 33 -1.6 0 -0.64 34 2253.4 0 253.3 34 189.05 0 21.16 34 -38.8 0 -4.04 35 656.5 0 93.1 35 79.808 0 15.34 35 30.3 0 11.71 36 1.4 0 1.1 36 4.022 0 0.582 36 -4.9 0 -0.71 37 14.6 0 1336 37 71.246 0 1.159 37 14.6 0 -9 total 22799.1 -12939 3839 total 4580.8 7533.1 555.1 total 8321.6 4445 933.3 cost cost cost ca 13699 ca 12669 ca 13699 rc 0 rc 1030 rc 0 %ec 100 %ec 92.48 %ec 100 fig. 6 show cost curve nature of prm and frm for mva utility factor method, whereas, only frm for modified amp-mile method. fig. 6(a) and (b) are drawn by using data of table 3, whereas, fig. 6(c) and (d) are drawn by using data of table 6. the price instability significantly affects the results of both the proposed methods. fundamentally modified amp -mile and mva utility method shows different characteristics with the fluctuations in demand. therefore, transmission price and demand are the essential characteristics for the estimation of cost allocation. variations in the curve are depending on the magnitude. proposed methodologies are fair, accurate and feasible for estimation of cost if the cost allocation is prepared as per the demand. mva utility factor technique is simple in application and provides price signals. the dissimilarity in the cost curve shown in fig. 6(b) is due to prm allocation is 17.45% of the entire embedded cost and remaining 82.55% allocate through supplementary charges, whereas, frm allocated 100% embedded cost for both the methods. similarly, in fig. 6(d), frm allocates 100% embedded cost, but prm allocates 92.48% of the entire embedded cost. in prm energies due to the amount of allocation in both the models depends on circuit capacity. the decrement proportionality is with used circuit advances in technology innovation, vol. 4, no. 3, 2019, pp. 177-196 194 capacity (ucc), which is always less than unity. hence, the quality of decrement may cause a disparity in the sign of allocation of both the model as shown in fig. 6. in results, current utility factor and eu’s are also used in both techniques. implementation of this technique proved to be superior over mw-mile and mva-mile method along with full recovery. table 7 shows the current utility factor and eu’s of reactive power, reflecting the direction of reactive power. through cost allocation by these existing methods appear to be uniform but due to mentioned limitations well through-out to be inexcusable. table 7 current utility factors and eu’s for reactive load at bus 34 of 37 bus system line no. 𝜕𝐼𝑙 𝜕𝑄𝑑34⁄ extent of use of load 34 line no. 𝜕𝐼𝑙 𝜕𝑄𝑑34⁄ extent of use of load 34 line no. 𝜕𝐼𝑙 𝜕𝑄𝑑34⁄ extent of use of load 34 1 0.0006 0.0020 17 -0.0001 -0.0046 33 0.0000 0.0000 2 0.0000 0.0000 18 0.0002 0.0052 34 0.0002 0.0219 3 0.0027 0.0332 19 0.0001 0.0057 35 0.0008 0.0524 4 0.0002 0.0040 20 -0.0009 0.0167 36 0.0003 -0.0481 5 0.0000 0.0000 21 -0.0005 0.2119 37 0.0001 -0.0009 6 0.0001 0.0027 22 0.0045 0.1239 38 0.0003 -0.0385 7 0.0001 0.0013 23 0.0000 0.0000 39 0.0001 0.0025 8 0.0000 0.0000 24 0.0002 0.0871 40 0.0001 -0.0021 9 0.0000 0.0000 25 0.0001 0.0124 41 0.0002 -0.0023 10 0.0002 0.0076 26 -0.0002 0.0049 42 0.0000 0.0000 11 0.0000 0.0000 27 0.0002 0.0163 43 0.0000 0.0000 12 0.0014 0.4823 28 0.0000 0.0000 44 0.0003 -0.0009 13 0.0005 0.0108 29 -0.0001 -0.0090 45 0.0001 -0.0003 14 0.0007 0.0381 30 0.0001 0.0051 46 0.0000 0.0000 15 0.0001 0..0047 31 0.0001 0.0050 47 0.0000 0.0000 16 0.0013 -0.0053 32 0.0001 0.0076 48 0.0002 0.0042 by evaluating the non-linear curve, cost allocation, remaining charges, current utility factor, and eu’s. some important factors are originated. both methods recover 100 embedded costs in a comparison to the existing method. when compare the ieee 6-bus and 37-bus system in mvapuf prm model. the cost allocation value is 17.45% and 92.48 % respectively. therefore 37-bus practical system cost allocation is more suitable as a comparison to the ieee system. so both methodologies are reliable for the practical transmission network. the non-linear nature of curve is used to assessing true portrayal of operating condition load flow has followed to develop fair allocation. the sensitivity patterns can help iso to forecast day-ahead transmission cost as well as plan for day-ahead setting up of open access electricity market. 5. conclusions in open access, electricity market allotment of embedded cost is one of the momentous facets. the two methodologies for allocation of the embedded cost of transmission network i.e. modified amp-mile method and mva utility factor method with two models: full recovery and partial recovery, has been employed in this paper.  the proposed method exploits marginal participation in the load flow framework. the nonlinear sensitivities for the current utility of active and reactive powers have been discussed in this paper. the nonlinear sensitivities in power networks which are used to anticipate congestion and transmission price forecasting.  fair cost allocation in the presence of nonlinear sensitivities is solved by employing proposed modified amp-mile methodology. it is implemented on prices and sensitivities with respect to injections/withdrawals of power in the modern electricity market.  mva utility method in two different modals i.e. frm and prm is employed for estimation of fair cost allocation. non-linear sensitivities and distinct slack bus notion are promoted to allocate the partial or entire embedded cost of the transmission network. advances in technology innovation, vol. 4, no. 3, 2019, pp. 177-196 195  results show that the wheeling charges for modified amp-mile and mva utility factor method have approached more close solutions to mw-mile, mva-mile and amp-mile methods for the 6-bus and 37-bus test system.  comparison of other existing techniques for ieee 6-bus system confirms the effectiveness of the proposed method. the impact of loading on cost allocation by different methods has also been discussed.  both methodologies also justify new distinct slack bus notion and negative payment to evaluate currency cost impact as choice of slackbus affects cost allocation.  proposed methods allocate 100% embedded cost and zero remaining charges. the outcomes of standard ieee 6-bus and real-time 37-bus systems show the validity and efficacy of the proposed techniques. conflicts of interest “the authors declare no conflict of interest.” references [1] p. m. sotkiewicz and j. m. vignolo, “allocation of fixed costs in distribution networks with distributed generation,” ieee transaction on power systems, vol. 21, no. 2, pp. 639-652, may 2006. [2] w. j. lee, c. h. lin and k. d. swift, “wheeling charge under a deregulated environment,” ieee transactions on industry applications, vol. 37, no. 1, pp. 178-183, february 2001. [3] h. hamada and r. yokoyama, “wheeling charge reflecting the transmission conditions based on the embedded cost method,” journal of international council on electrical engineering, vol. 1, no. 1, pp. 74-78, september 2011. [4] c. w. yu and y. k. wong, “analysis of transmission-embedded cost allocation schemes in open electricity markets,” resources, energy and development, vol. 3, no. 1, pp. 1-11, april 2006. [5] j. pan, y. teklu, s. rahman, and k. jun, “review of usage-based transmission cost allocation methods under open access,” ieee transactions on power systems, vol. 15, no. 4, pp. 1218-1224, november 2000. [6] s. m. h. nabavi, s. hajforoosh, s. hajforosh, and n. a. hosseinipoor, “using tracing method for calculation and allocation of reactive power cost,” international journal of computer applications, vol. 13, no. 2, pp. 14-17, january 2011. [7] g. jain, k. singh, and d. k. palwalia, “transmission wheeling cost evaluation using mw-mile methodology,” proc. ieee 3rd international conference in nirma university (nuicone), ahmedabad, pp. 6-8, december 2012. [8] d. shrimohammadi, p. r. gribik, e. t. k. law, j. h. malinowski, and r. e. o’donnell, “evaluation of transmission network capacity use for wheeling transactions,” ieee transactions on power system, vol. 4, pp. 1405-1413, october 1989. [9] f. li, n. p. padhy, j. wang, and b. kuri, “mw+mvar-miles based distribution charging methodology,” ieee power engineering society general meeting, pp. 18-22, june 2006. [10] h. m. merrill and b. w. erickson, “wheeling rate based on marginal cost theory,” ieee transaction on power system, vol. 4, no. 4, pp. 1445-1451, november 1989. [11] d. k. khatod, v. pant, and j. sharma, “a novel approach for sensitivity calculations in the radial distribution system,” ieee transaction on power delivery, vol. 21, no. 4, pp. 2048-2057, october 2006. [12] l. j. billera and d. c. heath, “allocation of shared costs: a set of axioms yielding a unique procedure,” mathematic of operational research, vol. 7, no. 1, pp. 32-39, february 1982. [13] d. shirmohammadi, x. v. filho, b. gorenstin, and m. v. p. pereira, “some fundamental technical concepts about cost based transmission pricing,” ieee transactions on power systems, vol. 11, no. 2, pp. 1002-1008, may 1996. [14] j. w. m. lima, “allocation of transmission fixed charges: an overview,” ieee transactions on power systems, vol. 11, no. 3, pp. 1409-1418, august 1996. [15] r. r. kovacs and a. l. leverett, “a load flow based method for calculating embedded incremental and marginal cost of transmission capacity,” ieee transaction on power system, vol. 9, no. 1, pp. 272-278, february 1994. [16] l. murphy, r. j. kaye, and f. f. wu, “distributed spot pricing in radial distribution systems,” ieee transactions on power systems, vol. 9, no. 1, pp. 311-317, february 1994. [17] j. y. choi, s. h. rim, and j. k. park, “optimal real time pricing of real and reactive powers,” ieee transactions on power systems, vol. 13, no. 4, pp. 1226-1231, november 1998. [18] g. r. yousefi and h. seifi, “wheeling charges with consideration of consumer load modeling,” ieee power systems conference and exposition, vol. 1, pp. 168-173, october 2004. advances in technology innovation, vol. 4, no. 3, 2019, pp. 177-196 196 [19] d. gang, z. y. dong, w. bai, and x. f. wang, “power flow based monetary flow method for electricity transmission and wheeling pricing,” electric power systems research, vol. 74, no. 2, pp. 293-305, january 2005. [20] j. bialek, “allocation of transmission supplementary charge to real and reactive loads,” ieee transaction on power systems, vol. 13, no. 3, pp. 749-754, august 1998. [21] c. g. beckman, “india’s new cost allocation method for inter-state transmission: implications for the future of the indian electricity sector,” the electricity journal, vol. 26, no. 10, pp. 40-50, december 2013. [22] a. bhadoria, v. k. kamboj, m. sharma, and s. k. bath, “a solution of non-convex/convex and dynamic load dispatch problem using moth flame optimizer,” springer inae letters, pp. 1-22, 2018. [23] e. a. surucu, a. k. aydogan, and d. akgul, “bidding structure, market efficiency and persistence in a multi-time tariff setting,” energy economics, vol. 54, pp. 77-87, february 2015. [24] d. streimikiene and i. siksnelyte, “sustainability assessment of electricity market models in selected developed world countries,” renewable and sustainable energy reviews, vol. 57, pp. 72-82, may 2016. [25] a. k. ferdavani, r. a. hooshmand, and h. b. gooi, “analytical solution for demand contracting with forecasting-error analysis on maximum demands and prices,” iet generation, transmission & distribution, vol. 12, no. 12, pp. 3097-3105, may 2018. [26] d. feng, c. liu, z. liu, l. zhang, y. ding, and c. campaigne, “a constraint qualification based detection method for nodal price multiplicity in electricity markets,” ieee transactions on power systems, vol. 32, no. 6, pp. 4968-4969, november 2017. [27] y. yu, t. jin, c. zhong, and c. wang, “a multi-agent based design of bidding mechanism for transmission loss reduction,” electrical power and energy systems, vol. 78, pp. 846-856, june 2016. [28] r. f. blanco, j. m. arroyo, and n. alguacil, “on the solution of revenueand network-constrained day-ahead market clearing under marginal pricing—part i: an exact bilevel programming approach,” ieee transactions on power systems, vol. 32, no. 1, pp. 208-219, january 2017. [29] m. hossain, k. tushar, and c. assi, “optimal energy management and marginal cost electricity pricing in micro grid network,” ieee transactions on industrial informatics, vol. 13, no. 6, pp. 3286-3298, december 2017. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). template encit2010 advances in technology innovation, vol. 5, no. 2, 2020, pp. 112-125 synthesis of formation control systems for multi-agent systems under control gain perturbations kazuki miyakoshi 1,* , shun ito 1 , hidetoshi oya 1 , yoshikatsu hoshi 1 , shunya nagai 2 1 graduate school of integrative science and engineering, tokyo city university, tokyo, japan 2 department of information systems creation, kanagawa university, kanagawa, japan received 25 april 2019; received in revised form 28 july 2019; accepted 26 november 2019 doi: https://doi.org/10.46604/aiti.2020.4136 abstract this paper proposed a linear matrix inequality (lmi)-based design method of non-fragile guaranteed cost controllers for multi-agent systems (mass) with leader-follower structures. in the guaranteed cost control approach, the resultant controller guarantees an upper bound on the given cost function together with asymptotical stability for the closed-loop system. the proposed non-fragile guaranteed cost control system can achieve consensus for mass despite control gain perturbations. the goal is to develop an lmi-based sufficient condition for the existence of the proposed non-fragile guaranteed cost controller. moreover, a design problem of an optimal non-fragile guaranteed cost controller showe that minimizing an upper bound on the given quadratic cost function can be reduced to constrain a convex optimization problem. finally, numerical examples were given to illustrate the effectiveness of the proposed non-fragile controller for mass. keywords: multi-agent systems (mass), consensus, control gain perturbations, guaranteed cost control, lmis 1. introduction the robustness of control systems is an important property, and thus “robust stability analysis” and “robust stabilization problems” have been well studied for a long time [1-2]. especially, quadratic stabilizing controllers and control are typical robust control strategies [3-5]. furthermore, for practical situations, it is desirable to design control systems that achieve not only asymptotical stability but also an adequate level of control performance. one approach to this problem is “the guaranteed cost control”, and was introduced by chang and peng [6]. in the guaranteed cost control approach, a cost function corresponding to control performance introduced, and the resultant controller has the advantage of providing an upper bound on the given cost function. namely, the system performance degradation incurred by the uncertainties is guaranteed to be less than this bound. based on this idea, many significant results have been presented [7-9]. in the work of petersen and mcfarlane [7], the parameter-dependent riccati equation approach adopted, and yu and chu have proposed a guaranteed cost controller design method based on linear matrix inequalities (lmis) [8-9]. on the other hand, it is generally known that feedback systems designed for robustness concerning parameters in the controlled system may require very accurate controllers. however, uncertainties in controllers may appear for imprecision inherent in conversion and roundoff errors in numerical computations. furthermore, it has pointed out that any useful design strategy should generate a controller which also has a sufficient margin for the readjustment of its coefficients [10-11]. in particular, keel and bhattacharyya [10] showed some examples that optimum and robust controllers can produce extremely fragile controllers, concerning vanishingly small perturbations for parameters of the designed controller. namely, any useful design procedure should generate a controller which also has sufficient room for the readjustment of its coefficients because * corresponding author. e-mail address: g1881842@tcu.ac.jp advances in technology innovation, vol. 5, no. 2, 2020, pp. 112-125 113 controller implementation is subject to imprecision inherent in analog-digital and digital-analog conversion. finite word length and finite resolution measuring instruments and roundoff errors in numerical computations. from this viewpoint, there are some efforts to tackle the design problem of robust non-fragile controllers [12-14]. famularo et al. had shown a design method of a robust non-fragile lq controller for linear systems with structured uncertainties in the system matrix [12]. additionally, an lmi-based design method of the decentralized guaranteed cost controller for uncertain large-scale interconnected systems has been presented [13]. in the work of oya et al. [14], for linear continuous-time systems with structured uncertainties included in the system matrix and the input under multiplicative or additive control gain variations, an lmi-based design method of a robust non-fragile stabilizing controller has been proposed. by the way, with rapid technological development, controlled systems become large-scale and complex. one can see that systems are referred to as “large-scale interconnected systems”. for large-scale interconnected systems, “centralized control strategies” might be unrealistic for physical, technical, and societal reasons. also, one can see that “decentralized control strategies” are useful. many researchers have studied decentralized control for large-scale interconnected systems (see [15] and references therein). for large-scale interconnected systems, some design methods for the decentralized robust control had been suggested [16-18]. nagai et al. [17] proposed a decentralized variable to gain robust controller for a class of uncertain large-scale interconnected systems, and guaranteed cost control for large-scale interconnected systems with control gain perturbations has also been studied in [18]. additionally, formation control for multi-agent systems (mass) is known as one of the decentralized control problems and has recently attracted much attention. in general, a multi-agent system is a loosely coupled network of multiple interacting agents, and it is well-known that mass can achieve various tasks efficiently. although there are various topics in the formation control problem for mass, the consensus problem has been well focused. the consensus problem can be applied to various fields such as sensor networks, vehicle formations, mobile robots, unmanned aerial vehicles, and so on. additionally, consensus means that states of all agents are driven to a common state by implementing distributed protocols. therefore, the consensus problems are one of the most important and fundamental in the formation, and lots of existing results for consensus problems have been shown (e.g. [19-22]). the consensus problem for a network of first-order integrators with directed in the formation flow and fixed/switching topology has been studied in [19]. xie and wang have studied convergence analysis of a consensus protocol for a class of networks of dynamic agents with fixed topology. furthermore, zhai et al. proposed the matrix inequality-based stabilization condition and consensus algorithm for mass had been presented [22]. ito et al. [23] proposed an adaptive gain controller design method considering relative distances for mass. although the existing result [23] had dealt with consensus problems for multi-agent systems with a leader-follower structure, control gain perturbations have not been considered. about these results of the consensus problem for mass, uncertainties in controllers were not considered, and the lmi-based design method of non-fragile guaranteed cost controllers for mass with control gain perturbations has little been considered as far as known. from this viewpoint, a consensus problem for mass with leader-follower structures was discussed, and an lmi-based design method of a non-fragile guaranteed cost controller which was presented guarantees a consensus for mass. in this paper, additive control gain perturbations are dealt with and showed that sufficient conditions for the existence of the proposed non-fragile guaranteed cost controller given in terms of linear matrix inequalities (lmis). additionally, a design method of quadratic guaranteed cost control-based consensus strategy which guarantees the upper bound of a given quadratic cost function is considered. moreover, the optimal guaranteed cost control approach which minimizes the upper bound of the quadratic cost function is discussed. the crucial differences between the proposed controller synthesis and the existing results [19-23] are that the proposed controller perturbations and optimal guaranteed cost control–based consensus strategy can be achieved. besides, the proposed non-fragile controller can easily be obtained by solving the constrained convex optimization problem, i.e. the proposed non-fragile guaranteed cost control for a consensus is useful. finally, a numerical example is presented to illustrate the effectiveness of the proposed guaranteed cost controller for formation control systems. advances in technology innovation, vol. 5, no. 2, 2020, pp. 112-125 114 this paper is organized as follows. in section 2, notation and two useful lemmas which were used in this paper are shown. in section 3, the main results presented. namely, an lmi-based design method of a guaranteed cost controller for mass under control gain perturbations. additionally, the optimal guaranteed cost controller which was discussed minimizes the upper bound on a given quadratic cost function. finally, simple illustrative examples were displayed to show the effectiveness of the non-fragile guaranteed cost controller developed in this paper. 2. notations and lemmas the following notations were used in this paper. for a matrix a , the inverse matrix a and its transpose are denoted by 1 a  and t a , respectively. also,  he a means t a a and n i represent n -dimensional identity matrix, and a block diagonal matrix is composed of matrices i a for 1, 2, ,i m are described as  1 2diag , , , ma a a . moreover, real symmetric matrix , 0a a  (resp. 0a  ) means that a is positive (resp. nonnegative) definite matrix. for a vector a , a is euclidian norm, and a represents its included norm for a matrix a . the symbols “ * ” and “ ” mean symmetric blocks in matrix inequalities and equality by definition, respectively. moreover, the kronecker product of matrices m n a   and p qb  is defined as 11 1 1 n m mn a b a b a b a b a b             (1) furthermore, the following two useful lemmas were used in this paper: lemma 1 (schur complement formula [24]): for a given constant real symmetric matrix  , the following items are equivalent: (i) 11 12 22 0 *            , (ii) 11 0  and 1 22 12 11 12 0 t        , (iii) 22 0  and 1 11 12 22 12 0 t        . lemma 2 [14]: for matrices p and h , which have appropriate dimensions and a positive scalar  , the following relation holds: 1t t t t ph h p pp h h     (2) 3. problem formulation the multi-agent system which is considered is composed of 3 agents where an agent indexed by 1 acts as the leader and the other agents indexed by 2, 3, respectively, act as the followers described as the following state equation: ( ) ( ) ( ) i i i d x t ax t bu t dt   (3) where  (1) (2) (3) (4)( ) ( ), ( ), ( ), ( )ti i i i ix t x t x t x t x t is the state of i -th agent and elements, (1) ( ) i x t and (2) ( ) i x t (resp. (3) ( ) i x t and (4) ( ) i x t ) mean the position and velocity on x axis (resp. y axis), 4 4 a   and 4 2 b   are given by advances in technology innovation, vol. 5, no. 2, 2020, pp. 112-125 115 0 1 0 0 0 0 0 0 0 0 1 0 , 0 0 0 1 0 0 0 0 0 0 0 1 a b                          　 (4) one can see that the overall system can be written as ( ) ( ) ( ) t t d x t a x t b u t dt   (5) where  diag , ,ta a a a and  diag , ,tb b b b . moreover,  2 3( ) ( ) ( ) ( ) t t t t l x t x t x t x t and  2 3( ) ( ) ( ) ( ) t t t t l u t u t u t u t in eq. (5) are the vectors of the state and the control input for the overall system, respectively. it is well-known that the information path between agents based on graph theory, and the topology connection which is considered is shown in fig. 1. fig. 1 the topology connection of agents since the topology connection can also be expressed by graph laplacian l defined by the adjacency matrix a and the degree matrix d. in the topology connection, the adjacency matrix a and the degree matrix d are given by 0 0 0 0 0 0 1 0 1 , 0 2 0 1 1 0 0 0 2 a d                       　 (6) thus, the corresponding graph laplacian l can be described as 0 0 0 1 2 1 1 1 2 l d a              (7) now, to consider the control gain perturbations for the i -th agent of eq. (3), there are the following control input:   , , ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) i i k i f i i i j j n u t u t u t k t x t f t x t x t      (8) where i n is the set of neighbors for the i -th agent. 2 4( )k t  is the state feedback gain matrix and 2 4( )f t  is the consensus gain matrix. these gain matrices are defined as ( ) ( )k t k k t (9) ( ) ( )f t f f t (10) note that 2 4 k   and 2 4 f   are the nominal control gain matrices, and the matrices 2 4( )k t    and 2 4 ( )f t    are unknown time-varying parameters which satisfy the relations ( ) kk t   and ( ) ff t   . one can see that additive control gain perturbations were considered in this paper. additionally, 1 k   and 1 f   mean upper bounds of the advances in technology innovation, vol. 5, no. 2, 2020, pp. 112-125 116 control gain perturbations and are known positive scalars. namely, for the overall system of eq. (5) the actual control input implemented is assumed to be ( ) ( ) ( ) ( ) ( ) ( ) ( ) k f t t u t u t u t k t x t f t x t    (11) where  2 3( ) ( ) ( ) ( ) ( , ) t t t t p pl p p u t u t u t u t p k f  . in eq. (11), 6( ) k u t  can be described as  3( ) ( ) ( ) ( ) ( )k tu t i k t x t k t x t  (12) and 6( ) f u t  is the consensus protocol given by   2 3 0 0 0 ( ) ( ) ( ) 2 ( ) ( ) ( ) ( ) ( ) 2 ( ) ( ) ( ) ( ) ( ) ( ) l f t x t u t f t f t f t x t f t f t f t x t l f t x t f t x t                      (13) therefore, the actual control input of eq. (11) can be rewritten as ( ) 0 0 ( ) ( ) ( ) 2 ( ) ( ) ( ) ( ) ( ) ( ) 2 ( ) k t u t f t k t f t f t x t f t f t k t f t                (14) moreover, the closed-loop system found is described as   2 3 ( ) ( ) ( ) ( ) ( ) ( ) 0 0 0 0 ( ) 0 0 ( ) * 0 * 0 ( ) ( ) 2 ( ) ( ) ( ) * * * * ( ) ( ) ( ) 2 ( ) ( ) 0 0 t t t t l k f f d x t a x t b k t x t f t x t dt a b k t x t a b f t k t f t f t x t a b f t f t k t f t x t a bf a bf b bf bf a                                                                 2 3 ( ) ( ) ( ) ( ) ( ) l t t t x t k t f t x t x t                  (15) where 4 4 k a   and 4 4 f a   are given by ka a bk  and 2fa a bk bf   , respectively. also, ( )tk t and ( ) t f t are unknown matrices which can be written as ( ) 0 0 0 0 0 ( ) * ( ) 0 , ( ) ( ) 2 ( ) ( ) * * ( ) ( ) ( ) 2 ( ) t t k t k t k t f t f t f t f t k t f t f t f t                               　 (16) now, the following quadratic cost function introduced:   0 ( ) ( ) ( ) ( ) ( ) ( ) t t t t t t k k k f f f j x t q x t u t r u t u t r u t dt     (17) where 12 12 6 6 , tt k q r     , and 6 6 tf r   are given positive-definite symmetric matrices that are selected by designers. from the above discussion, the control objective in this paper is to design a guaranteed cost controller which minimizes the upper bound on the quadratic cost function of eq. (17). that is to derive gain matrices 2 4 k   and 2 4 f   minimizing the upper bound on the quadratic cost function of eq. (17). advances in technology innovation, vol. 5, no. 2, 2020, pp. 112-125 117 4. main results in this section, a design method of the non-fragile controller showed based on the linear matrix inequality (lmi) framework. firstly, the following theorem is given for the non-fragile controller under additive control to gain perturbations of eq. (11). theorem 1: for the overall system of eq. (15) with the control input of eq. (11) and the quadratic cost function of eq. (17), if there are symmetric positive definite matrix 0s  , matrices k w and f w , and scalars 0, 0, 0     and 0  satisfying the following lmi condition:   122 122 1 1 2 6 1 2 6 12 12 , , , , 0 0 0 0 0 0 0 0 0 0 0 0 5 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 6 0 0 0 0 0 0 0 0 0 0 0 0 0 0 t t t t t t t t k f t t t k f t t t k t f t t k k k f f f t t s w w s s s w w s s s i s i s q w r i w r i s i s i                                                  (18)               11 12 13 22 23 33 , , , , , * , , , , * * , , , , k f f k f k f f k f s w w w s w w s w w w s w w                    (19)            11 12 13 23, , , t t k k f f f f f s w he as bw w w w b w he bw           (20)        22 33, , , , , , , , 2 t k f k f k f s w w s w w he as bw bw bb             (21) then the upper bound on the quadratic cost function of eq. (17) is guaranteed and the closed-loop system of eq. (15) with gain matrices 1 k k w s   and 1 f f w s   are asymptotically stable. moreover, the upper bound on the quadratic cost function of eq. (17) is given by  * (0) (0) (0)tj x x px . proof: using a symmetric positive definite matrix 4 4 p   , the following quadratic function introduced as a lyapunov function candidate:    3, ( ) ( ) t v x t x t i p x t  (22) from eqs. (5), (11), and (15), the time derivative of the quadratic function ( , )v x t which is along the trajectory of the closed-loop system of eq. (15) can be computed as          3 3 3 ( , ) ( ) ( ) ( ) ( ) ( ) ( , ) ( ) ( ) ( ) t t t t t t d d d v x t x t i p x t x t i p x t dt dt dt x t he i p k f b k t f t x t                          (23) where 12 12 ( , )k f    is the matrix described as 0 0 ( , ) k f f a k f bf a bf bf bf a              (24) advances in technology innovation, vol. 5, no. 2, 2020, pp. 112-125 118 therefore, if there exists the state feedback gain matrix 2 4 k   , consensus gain matrix 2 4 f   and the symmetric positive definite matrix 4 4 p   which satisfy the inequality condition     3 ( , ) ( ) ( ) 0t t the i p k f b k t f t        (25) then the quadratic function ( , )v x t satisfies the following inequality: ( , ) 0, ( ) 0 d v x t x t dt    (26) i.e. the quadratic function ( , )v x t becomes a lyapunov function for the closed-loop system eq. (15). by considering the quadratic cost function of eq. (17), there is an inequality condition:   ( , ) ( ) ( ) ( ) ( ) ( ) ( ) 0 t t t t t t t t t t k t t f t k f he p b k t f t q k t r k t f t r f t        (27) instead of the matrix inequality of eq. (25). the matrix 1 s p  introduced, and the change of variables k w ks and f w fs considered. then preand post-multiplying eq. (25) by 4 4 s   and using lemma 2, there is         2 2 5 ( , , ) ( ) ( ) ( ) ( ) ( ) 0 t t t t t t t k f k f t t t t t t t t t t t k t k k t f t f f t s w w b b s s s s s q s w k t s r w k t s w f t s r w f t s                        (28) where 12 12( , , ) k f s w w    is given by 11 12 13 3 3 22 23 33 ( , ) ( ) ( ) ( , , ) ( ) ( , , )( ) * ( , , ) ( ) * * ( , , ) k f f k f k f f k f s w w w s w w i s p k f i s s w w w s w w                         (29)            11 12 13 23, , , t t k k f f f f fs w he as bw w w w b w he bw           (30)      22 33, , , , 2k f k f k fs w w s w w he as bw bw      (31) one can see that 12 12( , , , , ) k f s w w      can be expressed as ( , , , , ) ( , , ) ( ) t t tk f k fs w w s w w b b       (32) therefore, applying lemma 1 to the inequality eq. (28) and simple algebraic gives   122 122 1 1 1 , , , , 0 0 0 0 0 0 0 0 ( , ) 0 5 0 0 0 0 0 0 0 0 0 0 0 0 t t t t t t t t t t tk f k f t k t f t t k k f f s w w s s s w w s i s i k f s q w r w r                                                     (33) where 12 12 ( , )k f      is given by advances in technology innovation, vol. 5, no. 2, 2020, pp. 112-125 119 00 0 00 0 0 0 0 0 00 0 0 0 0 0 ( , ) 00 0 0 0 0 0 0( ) 0 0 ( ) 0 0 ( )0 0 0 0 0 0 t t t t t t t t t t t t t s s s s k f k t k t f t                                                                                                 0 0 0 0 0 ( ) t t t f t                             (34) thus, one can see from lemma 2 that the following inequality can be obtained:   1 1 122 122 1 1 2 6 1 2 6 , , , , ( ) 0 0 0 0 0 0 0 0 0 5 0 0 0 0 0 0 0 0 0 0 0 0 6 t t t t t t t t k f t t t t t k f t k t f t t k k k f f f s w w s s s s s w w s i s i s q w r i w r i                                                 (35) furthermore, by applying lemma 1 to the inequality of eq. (35), the inequality of eq. (35) which was found is equivalent to the lmi of eq. (18). it follows that the result of the theorem is true. therefore, the proof of theorem 1 is completed. (q.e.d) theorem 1 provides a sufficient condition to gain perturbations of the form of eq. (15) for the existence of the guaranteed cost controller which is under control. next, the theorem for the optimal guaranteed cost controller is showed. the lmi of eq. (18) defines a convex solution set, and thus various efficient convex optimization algorithms can be adapted to test whether the lmis are solvable and whether they generate particular solutions. moreover, the parametrized representation exploited to design the guaranteed cost controller with some additional requirements, its solutions parametrize the set of guaranteed cost controllers. in particular, the optimal guaranteed cost controller which minimizes the upper bound on the quadratic cost function of eq. (17) can be obtained by solving a certain optimization problem. the upper bound on the quadratic cost function of eq. (17) depends on the initial value of the overall system. thus, the complementary variable  which was introduced satisfies the following lmi: (0) 0 (0) t x x s       (36) one can easily see that the minimization of  is equivalent to the minimization of the upper bound on the quadratic cost function of eq. (17). consequently, the following theorem developed: theorem 2: consider the overall system of eq. (5) with control gain perturbations, the quadratic cost function of eq. (17) and the control input of eq. (11). if there exists solutions of the following constrained convex optimization problem: 0, , , 0, 0, 0, 0, 0 minimize [ ] subject to eqs.(18) and (34)      k fs w w       (37) then the control input of eq. (11) is an optimal guaranteed cost control which minimizes the upper bound on the quadratic cost function of eq. (17). proof: since the minimization of  is equivalent to the minimization of the upper bound on the quadratic cost function of eq. (17), the result of theorem 2 is obtained by adopting a similar way to the proof of theorem 1 and the existing results (e.g. [13-14, 25]). advances in technology innovation, vol. 5, no. 2, 2020, pp. 112-125 120 remark 1: in this paper, a design problem of non-fragile guaranteed cost controllers which achieve consensus for multiagent systems (mass) with leader-follower structures was studied, and for simplicity mass composed of 3 agents had been considered. additionally, lmi-based design for optimal non-fragile guaranteed cost control had also been discussed. one can easily see that the proposed controller design approach applies to mass without leaders, and the proposed non-fragile controller can be extended to mass composed of more than 3 agents. 5. numerical examples to demonstrate the efficiency of the proposed non-fragile stabilizing controller, a simple example had been run. in this example, the non-fragile controller under the control gain perturbations is considered. also, the simulation results are shown for the proposed robust stabilizing controller designed without thinking of control gain perturbations. the control problem considered here is not necessarily practical. however, the simulation results stated below illustrate the distinct feature of the proposed non-fragile controller, i.e. the proposed non-fragile controller was compared with the conventional consensus controller which was designed by ignoring controller gain perturbations. consider the multi-agent system of eq. (5). in this example, 13.5 10 k     and 1 3.5 10 f     is assumed, respectively. the initial state for agent (0) i x and target value * i x were chosen to be 2 3 4 0 3 3 3 6 (0) , (0) , (0) 3 3 1 2 2 1 l x x x                                        　 　 (38) * * * 2 3 0 1 1 0 0 0 , , 0 1 1 0 0 0 l x x x                                        　 　 (39) as shown in the image in fig. 2. additionally, the weighting parameters q , k r , f r were set as 1.0q  , 12.0 10 k r    , and 12.0 10 f r    . here, unknown parameter ( )k t and ( )f t is given as  1 1 1 1 1 8.0 10 1.0 exp( 0.1 ) cos(10.0 ) 1 1 1 1 k k t t                (40)  1 1 1 1 1 8.0 10 1.0 exp( 0.1 ) cos(10.0 ) 1 1 1 1 k f t t                (41) fig. 2 trajectories for each agent in the state space advances in technology innovation, vol. 5, no. 2, 2020, pp. 112-125 121 by applying theorem 2 and solving the constrained convex optimization problem eq. (36), there are the following values for the lmi solution: 1 1 3 2 1 2 2 1 1 1 2.6795 10 1.0585 10 2.7372 10 2.9836 10 * 4.4751 10 2.3603 10 1.1958 10 * * 3.5287 10 1.5169 10 * * * 3.4045 10 s                                 (42) 1 3 2 4 3 3 2.3088 10 4.5315 2.9135 10 1.2169 10 8.8730 10 8.2894 10 2.3342 10 4.4138 k w                     (43) 9 9 10 9 10 9 10 8 1.1693 10 6.4972 10 4.4497 10 7.0950 10 2.3135 10 5.8497 10 6.6781 10 6.1563 10 f w                           (44) 1 2 6.1981 10 , 1.3859, 5.0597, 6.8027, 3.5265 10             (45) therefore, the feedback gain matrix and the consensus can be calculated as 1 1 1 1 1 1 4.2765 1.1199 10 9.2459 10 4.6627 10 1.7296 10 3.7704 10 7.0206 1.6257 10 k                     (46) 9 8 9 8 8 8 7 7 4.1277 10 1.4406 10 8.2922 10 2.4390 10 2.9470 10 1.9930 10 1.0160 10 2.2850 10 f                         (47) on the other hand, the gain matrices for the conventional consensus controller can be obtained as 1 5 6 5 6 1 1.2613 7.7305 10 1.1617 10 6.6800 10 1.6130 10 4.7165 10 1.2615 7.7315 10 k                      (48) 4 6 6 6 5 6 6 6 2.0593 10 7.9111 10 4.2830 10 9.4814 10 4.7420 10 1.7697 10 2.3861 10 1.5674 10 f                        (49) note that the gain matrices of eq. (47) and (48) in the conventional controller can be designed by solving lmi of eq. (a.1) in appendix and theorem a.1, and it shows an lmi-based design method of the conventional consensus controller. the conventional controller is designed by considering the stabilization of the closed-loop system of eq. (a.2). fig. 3 time histories of the x-position fig. 4 time histories of the control input for x-direction the results of the simulation of this example were depicted in figs. (3)-(10). fig. 3 (resp. fig. 5) shows the time histories of the x-position (resp. y-position), and fig. 4 (resp. fig. 6) presents those of the control input for x-direction(resp. y-position). in these figures, the transient time-response, the manipulated control input, and the actual control input generated advances in technology innovation, vol. 5, no. 2, 2020, pp. 112-125 122 by the proposed consensus controller, i.e. figs. (3)-(6) are the results for the proposed consensus controller which is based on a guaranteed cost control strategy. additionally, the figures (figs. (7)-(10) show time histories of the state and the control input which were generated by the conventional consensus controller. fig. 7 and fig. 8 (resp. fig. 9 and fig. 10) represent the time histories of the x-position and one of the control input for x-direction (resp. the time histories of the y-position and one of the control input for y-direction), respectively. fig. 5 time histories of the y-position fig. 6 time histories of the control input for y-direction fig. 7 transient time-response of the x-position fig. 8 time histories of the control input x-direction fig. 9 transient time-response of the y-position fig. 10 time histories of the control input y-direction advances in technology innovation, vol. 5, no. 2, 2020, pp. 112-125 123 from figs. (2)-(10), the proposed non-fragile controller which was found stabilizes the uncertain linear system eq. (5) under the control gain perturbation eq. (41). furthermore, the proposed non-fragile control strategy can achieve the consensus for multi-agent systems despite control gain perturbations. on the other hand, the conventional controller was found cannot achieve consensus because of the controller gain perturbations are ignored in the controller design stage. therefore, the effectiveness of the proposed non-fragile stabilizing controller is shown. 6. conclusions this study had dealt with a design problem of a non-fragile guaranteed cost controller achieving consensus for mass with leader-follower structures. sufficient conditions had been shown for the existence of the proposed non-fragile guaranteed controller can be reduced to the solvability of lmis. furthermore, the optimal non-fragile guaranteed cost controller, which minimizes the upper bound on a given quadratic cost function had been presented. finally, simple numerical examples were given to illustrate the effectiveness of the proposed formation control system. the simulation result had shown that the closed-loop system was well stabilized despite control gain variations. besides, the consensus for mass can be achieved by the proposed non-fragile controller. additionally, the proposed optimal guaranteed cost controller which had been shown can be designed by solving the constrained convex optimization problem. namely, the proposed nonfragile controller can easily be obtained by using software such as matlab’s lmi control toolbox and scilab’s lmitool. it is worth pointing out that this paper didn’t take into account the time delay that occurred when control was performed via a network. therefore, there is room for expansion to the control system for systems with delays. furthermore, future research subjects include some extensions of the proposed design approach to a broad class of systems such as discrete-time systems and output feedback systems. additionally, the proposed controller synthesis will be extented to formation control for mass consisting of more general agent’s dynamics. conflicts of interest the authors declare no conflict of interest. appendix in this appendix, an lmi-based design method of the conventional consensus controller which is showed was obtained by ignoring control gain perturbations. namely, for i -th agent of eq. (3), the following control input is considered:   , , ( ) ( ) ( ) ( ) ( ) ( ) i i k i f i i i j j n u t u t u t kx t f x t x t      (a.1) thus, the closed-loop system can be described as   2 3 2 3 ( ) ( ) ( ) ( ) 0 0 0 0 0 0 ( ) * 0 * 0 2 ( ) * * * * 2 ( ) 0 0 ( ) ( ) ( )                                                                   t t t t l k l f f d x t a x t b k x t f x t dt a b k x t a b f k f f x t a b f f k f x t a x t bf a bf x t bf bf a x t (a.2) and the following lmi-based design method of the conventional consensus controller can be developed: advances in technology innovation, vol. 5, no. 2, 2020, pp. 112-125 124 theorem a-1: consider the overall system of eq. (5) and the control input of eq. (a.1). if there exists a symmetric positive definite matrix 0s  , matrices k w and f w will satisfy the following lmi condition:             * 2 0 * * 2 t t k f f k f f k f he as bw bw bw he as bw bw he bw he as bw bw                 (a.3) then the closed-loop system of eq. (a.2) with gain matrices 1 k k w s   and 1 f f w s   are asymptotically stable. proof: the result of theorem a.1 can easily be obtained by using a similar way to the proof of theorem 1. references [1] k. zhou, j. c. doyle, and k. glover, robust and optimal control, prentice hall, 1996. [2] k. zhou, essentials of robust control, prentice hall, 1998. [3] b. r. barmish, “stabilization of uncertain systems via linear control,” ieee transaction on automat control, vol. 28, no. 8, pp. 848-850, 1983. [4] i. r. petersen and c. v. hollot, “a riccati equation approach to the stabilization of uncertain linear systems,” automatica, vol. 22, no. 4, pp. 397-411, 1986. [5] p. p. khargonekar, i. r. petersen, and k. zhou, “robust stabilization of uncertain linear systems: quadratic stabilizability and control theory,” ieee transaction on automatic contrtol, vol. 35, no. 3, pp. 356-361, 1990. [6] s. s. l. chang and t. k. c. peng, “adaptive guaranteed cost control of systems with uncertain parameters,” ieee transaction on automatic control, vol. 17, no. 4, pp. 474-483, 1972. [7] i. r. petersen and d. c. mcfarlane, “optimal guaranteed cost control and filtering for uncertain linear systems,” ieee transaction on automatic control, vol. 39, no. 9, pp. 1971-1977, 1994. [8] l. yu and j. chu, “optimal guaranteed cost control of linear systems: lmi approach,” proc. 14th ifac world congress, beijing, china 1999, pp. 541-546. [9] l. yu and j. chu, “an lmi approach to guaranteed cost control of linear uncertain time delay systems,” automatica, vol. 35, no. 6, pp. 1155-1159, 1999. [10] l. h. keel and s. p. bhattacharyya, “robust, fragile, or optimal?,” ieee transaction on automatic control, vol. 42, no. 8, pp. 1098-1105, 1997. [11] p. dorato, “non-fragile controller design: an overview,” proc. american control conference, philadelphia, pennsylvania, usa, pp. 2829-2831, 1998. [12] d. famularo, p. dorato, c. t. abdallah, w. m. haddad, and a. jadbabaie, “robust non-fragile lq controllers : the static state feedback case,” interational journal of control, vol. 73, no. 2, pp. 159-165, 2000. [13] h. mukaidani, y. takato, y. tanaka, and k. mizukami, “the guaranteed cost control for uncertain large-scale interconnected systems,” proc. the 15th ifac world congress, pp. 265-270, 2002. [14] h. oya, k. hagino, and h. mukaidani, “robust non-fragile controllers for uncertain linear continuous-time systems,” proc. the 31st annual conference of ieee industrial society (iecon2005), raleigh, nc, usa, pp. 1-6, 2005. [15] d. d. siljak, decentralized control of complex systems, courier corporation, 1991. [16] z. gong, “decentralized robust control of uncertain interconnected systems with prescribed degree of exponential convergence,” ieee transaction on automatic control, vol. 40, no. 4, pp. 704-707, 1995. [17] s. nagai, h. oya, t. kubo, and t. matsuki, “synthesis of decentralized variable gain robust controllers for a class of large-scale interconnected systems with mismatched uncertainties,” international journal of system science, vol. 48, no. 8, pp. 1616-1623, 2017. [18] h. mukaidani, y. tanaka, and k. mizukami, “guaranteed cost control for large-scale systems under control gain perturbations,” electrical engineering in japan, vol. 146, no. 4, pp. 43-57, 2004. advances in technology innovation, vol. 5, no. 2, 2020, pp. 112-125 125 [19] r. olfati-saber and r. m. murray, “consensus problems in networks of agents with switching topology and timedelays,” ieee transaction on automatic control, vol. 49, no. 9, pp. 1520-1533, 2004. [20] g. xie and l. wang, “consensus control for a class of networks of dynamic agents: fixed topology,” proc. the 44th ieee conference on decision and control, and the european control conference 2005, seville, spain, 2005, pp. 96101. [21] y. zhang and y. p. tian, “consentability and protocol design of multiagent systems with stochastic switching topology,” automatica, vol. 45, no. 5, pp. 1195-1201, 2009. [22] g. zhai, s. okuno, j. imae, and t. kobayashi, “a matrix inequality based design method for consensus problems in multi-agent aystems,” internationl journal of application mathematics and computer science, vol. 19, no. 4, pp. 639646, 2009. [23] s. ito, k. miyakoshi, h. oya, y. hoshi, and s. nagai, “consensus via adaptive gain controllers considering relative distances for multi-agent systems,” advcances in technology innovation, vol. 4, no. 4, pp. 234-246, august 2019. [24] s. boyd, l. el ghaoui, e. feron, and v. balakrishnan, linear matrix inequalities in system and control theory, siam studies in applied mathmatics, 1994. [25] h. oya, k. hagino and h. mukaidani, “guaranteed. cost control for uncertain. linear continuous-time. systems. under conrtol gain perturbations,” (in japanese) trans. of jsme(c), vol. 72, no. 713, pp. 92-101, 2006. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). template encit2010 advances technology innovation, vol. 4, no. 4, 2019, pp. 234-246 consensus via adaptive gain controllers considering relative distances for multi-agent systems shun ito 1,* , kazuki miyakoshi 1 , hidetoshi oya 1 , yoshikatsu hoshi 1 , shunya nagai 2 1 graduate school of integrative science and engineering, tokyo city university, tokyo, japan 2 department of information systems creation, kanagawa university, kanagawa, japan received 25 april 2019; received in revised form 11 june 2019; accepted 28 july 2019 abstract in this paper, for multi-agent systems (mass) with leader-follower structures, we present a linear matrix inequality (lmi)-based design method of an adaptive gain controller considering relative distances between agents. the proposed adaptive gain controller consists of fixed gains and variable ones tuned by time-varying adjustable parameters. the objective of this paper is to derive enough conditions for the existence of the proposed adaptive gain controller which achieves consensus for each agent. the advantages of the proposed adaptive gain controller are as follows; the proposed controller can be obtained by solving lmi, and the proposed control system can achieve consensus and formation control, even if uncertainties are included in the information for relative distances. in this paper, we show that the design problem of the proposed adaptive gain controller can be reduced to the solvability of lmi. finally, simple numerical examples are included to illustrate the effectiveness of the proposed adaptive gain controller for mass. keywords: multi-agent systems (mass), consensus, relative distance, adaptive gain controller, linear matrix inequality (lmi) 1. introduction when we consider designing control systems for dynamical systems, it is necessary to derive a mathematical model for the controlled system, and one can see that optimal control is well known to be a powerful strategy in modern control theory. lq regulator for linear systems is a typical controller, and it ensures asymptotical stability for closed-loop systems with good robustness provided that a mathematical model for a control system describes precisely [1,2]. however, there always exist some gaps between the mathematical model and the controlled system, and the gaps are referred to as “uncertainty”. therefore, controller design methods dealing with uncertainties explicitly have been required, and robust control for uncertain dynamical systems has been extensively studied. one can see that robust control can be classified into “robust stability analysis” and “robust stabilization”, and lots of existing results for robust control strategies have been presented [3-6] and quadratic stabilizing controllers and control are well known robust control strategies[7,8]. note that the conventional robust controller consists of fixed gains which are derived by considering the worst-case variations for uncertainties. in contrast with the conventional robust control with fixed gains, some researchers have presented variable gain robust controllers for uncertain systems [9-11]. such variable gain robust control strategies are more flexible and adaptive comparing with the conventional robust control with fixed gains. on the other hand, the practical systems in modern society have become large-scale and complex due to rapid development of technologies, and such systems are referred to as “large-scale interconnected systems”. since it is difficult to apply centralized control strategies to such large-scale interconnected systems, design problems of decentralized control for *corresponding author. e-mail address: g1881842@tcu.ac.jp advances technology innovation, vol. 4, no. 4, 2019, pp. 234-246 235 large-scale interconnected systems have been well studied (see [12] and references therein). for instance, one can see that large interconnected power distribution systems which have strong interactions, transportation and traffic systems with lots of external signals, water systems which are widely distributed in the environment, energy systems, communication systems and so on are large-scale systems. moreover, formation control has recently attracted much attention, and a multi-agent system, in general, can be described as a network of a few of coupled dynamic units that are called agents. the design problems of formation control for mass are considered as one of the decentralized control problems and it is well -known that mass can achieve various task efficiently. for design problems of formation control for mass, the consensus problem has received a lot of attention, because this problem has drawn substantial attention from various fields such as vehicle formations, unmanned aerial vehicles, mobile robots, sensor networks, and so on. moreover, a consensus which means the states of all agents are driven to a common state by implementing distributed protocols is well accepted as one of the most important and fundamental problem in formation control. thus, a large number of existing results for consensus problem have been presented (e.g. [13-16]). in the work of olfati-saber et al. [13], the consensus problem for a network of first-order integrators with directed information flow and fixed/switching topology has been studied, and convergence analysis of a consensus protocol for a class of networks of dynamic agents with fixed topology have been shown [14]. zhang and tian have studied the mean-square consensus for mass composed of second-order integrators [15], and the matrix inequalitybased stabilization condition and consensus algorithm for mass have been presented [16]. also, a number of the existing results for the leader-follower consensus for mass has been presented([17-19]). “leader-follower” refers to defining a leader (whether real or virtual) and controlling another agent (follower) to follow the leader. in these results for consensu s problem for mass, controllers have fixed gain parameters only, and relative distances between agents cannot be considered explicitly. there are few results of the consensus problem via adaptive gain-based controller considering relative distances between agents for mass. from the above, this paper deals with a consensus problem for mass with leader-follower structures. in this paper, we present a design method of an adaptive gain controller giving considering relative distances between agents. the adaptive gain controller consists of fixed gains and variable ones tuned by time-varying adjustable parameters. in this paper. we show that enough conditions for the existence of the proposed adaptive gain controller can be reduced to linear matrix inequality (lmi). the proposed adaptive gain control strategy has advantages as follows; the proposed controller design approach can handle relative distances between agents explicitly. furthermore, even if the information for relative distances between the other agents and the leader is unknown, but their upper bounds are known, the controller can achieve consensus. furthermore, the proposed consensus control system can be designed by solving lmi. finally, simple numerical examples are included to illustrate the effectiveness of the proposed formation control systems. 2. preliminaries this chapter shows the mathematical notation used in this paper. m n represents an m-by-n real matrix, and ni represents an n-dimensional identity matrix. for matrix a , t a and 1 a  represent transpose and inverse. for a square matrix a , 0( 0)a a  indicates that a is positive definite (positive semidefinite) and 0( 0)a a  is negative definite (negative semidefinite). c is the norm of any matrix c . 1 2( , ,..., )ndiag a a a gives a diagonal block matrix with matrices ia (i = 1, 2, ..., n) on the diagonal. element * in the matrix represents a symmetric element. if a is a m×n matrix and b is a p×q matrix, the kronecker product a b is defined as follows; 11 1 1 n mnm mp nq a b a b a b a b a b              (1) advances technology innovation, vol. 4, no. 4, 2019, pp. 234-246 236 moreover, in this paper, we express the information path between agents based on graph theory [20]. the graph is collection of vertices and edges, and the notation of a graph is ( , ) , where {1, 2,..., }n is the set of n vertices in the graph, and  is the set of edges connecting the vertices. note that there are two types of graph, i.e. undirected graph and directed one. in this paper, we consider the directed graph. furthermore, we introduce an adjacency matrix , a degree matrix and a graph laplacian in order to express the graph algebraically. the adjacency matrix represents the adjacency relation of each vertex of the graph. in the graph ( , ) for a pair ( , )i j  that is, there is an edge from j  to i  the vertex i is said to be adjacent to j . in this case, the adjacent set of vertices i is  | ( , )i j i j   , and the elements ija of ija    is defined as 1 if( , ) and 0 otherwise ij i j i j a      (2) additionally, in the directed graph ( , ) , the in-degree of a vertex represents the number of edges incoming to the vertex and it is denoted as in id . conversely, out-degree means the number of edges outgoing from a vertex. then for the graph ( , ) , the degree matrix n n  and the graph laplacian are defined as 1 2( , ,..., )ndiag a a a (3)  (4) furthermore, the following useful lemma is used in this paper: lemma 1 [21]: for arbitrary vectors  and  , matrices g and h with appropriate dimensions, the following inequality holds: 2 2 t t gh g h    (5) 3. problem formulation fig. 1 multi-agent system in this paper in fig.1, the triangle “ i ” represents i -th agent(i=1,2,3). moreover “ l ” means the leader and the others are followers, and arrows indicate communication paths. then, the adjacency matrix , the degree matrix , and the graph laplacian in fig.1 can be obtained as 0 0 0 2 0 0 2 0 0 1 0 1 , 0 2 0 , 1 2 1 1 1 0 0 0 2 1 1 2                                    (6) now we assume that i -th agent ( , 2, 3i l ) can be described as the following state equation: advances technology innovation, vol. 4, no. 4, 2019, pp. 234-246 237 ( ) ( ) ( ) ( 1, 2,3)i i i d dt x t ax t bu t i   (7) where ( ) n ix t  and ( ) m iu t  are the vectors of the state and the control input, and the state ( ) n ix t  is given by  ( ) ( ) ( ) ( ) ( ) t i xi xi yi yix t x t v t x t v t (8) i.e. the state ( )xix t (resp. ( )yix t ) is the position in x-axis (resp. y-axis), and ( )xiv t (resp. ( )yiv t ) is velocity in x-axis (resp. yaxis) for the i -th agent. in (7), l n a   and l m b   are the system parameters which are defined as 0 1 0 0 0 0 0 0 0 0 1 0 , 0 0 0 1 0 0 0 0 0 0 0 1 a b                          (9) here, in order to consider the relative positions between agents, we introduce the following vectors:  ˆ ˆ t i xi xi yi yid d v d v (10) where xid (resp. yid ) is the desired relative position in x-axis (resp. y-axis) between the i -th agent and the leader agent. similarly, ˆxiv and ˆyiv are the target velocity. note that one can see that 0ld . here, we consider the difference between the actual position of the agent ( ( ))ix t and the desired relative position between the leader and the follower id as a new state of the system. from (7), the state equation of each follower considering the relative positions from leader to follower is expressed as    ( ) ( ) ( ) ( 1, 2.3)i i i i i d x t d a x t d bu t i dt      (11) by introducing the additional state vector ( )ix t described as   ( ) ( ) ˆ( ) ( ) ( ) ( ) ˆ( ) ( ) ( ) ( ) xi xi xi xi xi xi i i yi yi yi yi yi yi i x t d x t v t v v t x t d x t v t v v t x t x t d                                  (12) then one can see from (11) and (12) that the following state equation can be obtained: ( ) ( ) ( )i i i d dt x t ax t bu t  (13) summarizing the state equations of all agents, we get the following total system: ( ) ( ) ( )t t d dt x t a x t b u t  (14) where , , ( )t ta b x t and ( )u t are matrices and vectors given by     3 3 2 3 2 3 0 0 0 0 0 0 , b 0 0 0 0 0 0 ( ) ( ) ( ) ( ) , ( ) ( ) ( ) ( ) t t t t t t t t t t l l a b a i a a i b b a b x t x t x t x t u t u t u t u t                             (15) advances technology innovation, vol. 4, no. 4, 2019, pp. 234-246 238 next, we consider the control input ( )u t . note that consensus problem, “consensus” for agents means that the following relation for i  and j  holds:  lim ( ) ( ) 0i j t x t x t    (16) if 2 4 f   is the consensus gain, it is known that the consensus input for ( )ix t , ( )fiu t is given by [22].  ( ) ( ) ( ) i fi i j j u t f x t x t    (17) in the case of this paper, ( )fiu t is calculated as follows: 2 2 3 3 2 3 ( ) 0 ( ) ( ( ) 2 ( ) ( )) ( ) ( ( ) ( ) 2 ( )) fl f l f l u t u t f x t x t x t u t f x t x t x t          (18) namely,  2 3( ) ( ) ( ) ( ) t t t t f fl f f u t u t u t u t can be represented by the following matrix-vector form: 2 3 0 0 0 ( ) ( ) 2 ( ) ( ) ( ) 2 ( ) l f x t u t f f f x t f x t f f f x t                     (19) additionally, let ( )ku t be the state feedback input for stabilization of the system. by using the feedback gain matrix 2 4 ( )ku t   , the state feedback input ( )ku t can be written as ( ) ( ) ( )ku t i k x t (20) finally, we introduce a compensation input v(t) and consider the following control input: ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) 0 0 0 0 0 2 ( ) 2 ( ) 2 2 k f t u t u t u t v t i k x t f x t v t k f k f f x t f f f d v t f f k f f f f                                       (21) where  2 3( ) ( ) ( ) ( ) t t t t l v t v t v t v t . note that the design method for the compensation input v(t) and the gain matrices 2 4 k   and 2 4 f   is discussed in the next section. from (14) and (21), we have 2 3 0 0 0 0 0 2 2 2 2 0 0 0 0 ( ) 0 0 0 * 0 ( 2 ) ( ) 2 * * ( 2 ) ( ) ( ) ( )( ) ( ) l t tt k d f k f f f f f dt f f k f f f f a bk x t a bf b k f bf x t bf bf a bf bf b k f x t b x t d v tx t a x t                                                                             2 3 2 2 2 3 3 3 2 3 0 0 * 0 ( ) 2 * * 0 0 ( ) ( ) ( ) ( ) (2 ) ( ) ( ) ( 2 ) l k l l f f d b bf d b v t bf bf bf d b a x t bv t bf a bf x t bv t bf d d bf bf a x t bv t bf d d                                                             (22) in (21), ka and fa are the matrices described as ka a bk  , 2ka a bk bf   . advances technology innovation, vol. 4, no. 4, 2019, pp. 234-246 239 from the above, the controller design objective in this study is to derive the consensus gain 2 4 f   , the feedback gain 2 4 k   and compensation input 6 ( )v t  so that the asymptotic stability of the closed-loop system of (22) is guaranteed. 4. main results in this section, the design method of the feedback gain 2 4 k   , the consensus gains 2 4 f   and the compensation input 2 ( ) ( , 2, 3)iv t i l  is shown. we give the following theorem for determining these parameters of the overall system (22). theorem 1. consider the overall system of (14) and the control input of (21). if there exist solutions 0s  , kw and fw of following lmi condition: 11 12 13 22 23 33 11 12 13 22 33 23 * 0 * * , 2 , 2 2 , k k t t t t t k k k t t t t t t t k k f f f f a a bk a a bk bf sa as w b bw w b sa as w b bw bw w b w b bw                                             (23) then the compensation input ( )v t is designed as follows, 2 2 2 3 22 2 3 3 3 2 32 3 0 ( ) ( ) ( ) ( ) 2 ( ) ( )( ) ( ) 2 ( ) ( ) t tl t t t t t t v t f b px t v t v t fd dm b px t b px tv t f b px t fd dm b px t b px t                                 (24) where the matrix p is given by 1 p s   and the feedback gain matrix k and the consensus one f are designed as 1 kk w s   (25) 1 ff w s   (26) moreover, by applying the control input of (27) with the compensation input ( )v t of (24) the gain matrices k (25) and f (26) to the overall system of (14), asymptotic stability of the closed-loop system of (22) is guaranteed. proof: using a positive definite symmetric matrix 4 4t p p    , we introduce the following quadratic function as a candidate for lyapunov function: 3( , ) ( ) ( ) ( ) t v x t x t i p x t  (27) the time derivative of the quadratic function along the trajectory of the closed-loop system of (22) satisfies  3 3( , ) ( ) ( ) ( ) ( ) ( ) ( ) t td d d v x t x t i p x t x t i p x t dt dt dt     (28) advances technology innovation, vol. 4, no. 4, 2019, pp. 234-246 240 as is well known, the stability condition for the closed-loop system is ( , ) 0 d dt v x t  (29) and the time derivative of the quadratic function along the trajectory of the closed-loop system of (22) can be written as 2 2 3 3 3 2 3 3 2 2 3 3 0 0 ( ) ( , ) ( ) ( ) (2 ) ( ) ( ) ( ) ( 2 ) 0 0 ( ) ( ) ( ) ( ) ( ) (2 ) ( ) t k l f f k l t f f a bv t d v x t bf a bf x t bv t bf d d i p x t dt bf bf a bv t bf d d a bv t x t i p bf a bf x t bv t bf d d bf bf a bv t b                                                  2 3 2 2 3 2 2 3 3 2 3 3 ( 2 ) ( ) ( ) * ( ) ( ) ( ) ( ) (2 ) ( ) (2 ) ( ) ( 2 ) ( ) t t t t t k k t t t t f f t f f l l t t t t t f d d a p pa f b p f b p x t pbf a p pa f b p pbf x t pbf a p pa v t v t x t p b v t f d d p b v t f d d v t f d d v t f                                                     2 3 ( ) ( 2 ) t x t d d                (30) where 12 12 tp   is the following symmetric positive definite matrix: 3 0 0 * 0 * * t p p i p p p              (31) here, by introducing the matrix ( , , )p k f and the scalar function ( , , ( ))p f v t which are defined as         2 2 3 2 2 3 3 2 3 3 2 3 2 2 3 2 3 2 3 ( ) ( ) ( , , ( )) ( ) ( ) (2 ) ( ) (2 ) ( ) ( ) ( 2 ) ( ) ( 2 ) 2 ( ) ( ) 2 ( ) 2 ( ) 2 ( ) 2 t l l t t t t t tt l l v t v t p f v t x t p b v t f d d p b v t f d d x t v t f d d v t f d d pb pbv t x t pb pb v t f d d x t pb pb v t f d d                                              3 ( ) t x t   (32) ( , , ) * * * t t t t t k k t t t f f t f f a p pa f b p f b p p k f a p pa f b p pbf a p pa                  (33) one can see that the stability condition for the closed-loop system of (22) is reduced to ( , ) ( ) ( , , ) ( , , ( )) 0 td v x t x t p k f p f v t dt     (34) namely, if the matrix ( , , )p k f is negative definite and ( , , ( ))p f v t < 0 are satisfied, then the quadratic function ( , )v x t becomes a lyapunov function. for leader agent, there is no need the compensation input, i.e. ( ) 0lv t  then we have      2 2 2 3 2 2 2 2 2 3 2 3 3 2 3 3 3 3 2 3 3 3 ( ) ( ) 0 ( ) ( ) (2 ) ( ) ( ) ( ) 2 ( ) ( ) ( ) ( ) 2 ( ) ( ) ( ) ( ) 2 ( ) l l t t t t t t t t t t t t t t t pbv t t pb v t f d d x t v t b px t d f b px t d f b px t t pb v t f d d x t v t b px t d f b px t d f b px t                        (35) where ( )i t ( , 2,3)i l is the i -th term in the right-hand side of (32). since ( ) 0lv t  , we consider the design problem of 2 ( )v t and 3 ( )v t . let 3 ( )t be an auxiliary input for reducing the effect of 3d . in this paper, 3d means the relative position advances technology innovation, vol. 4, no. 4, 2019, pp. 234-246 241 between the leader and the follower 3, and it is unknown to the follower 2. the follower 2 can obtain the information for the upper bound 3dm for the relative position, i.e. 3dm satisfies 3 3d dm . therefore, we consider 2 2 3( ) 2 ( )v t fd t   (36) and one can see that for the third term in the right-hand side of 2 ( )t in (35) the following inequality holds: 3 2 3 2 3 2 ( ) ( ) ( ) t t t t t t d f b px t d f b px t dm f b px t     (37) thus, by selecting 3 ( )t defined as 2 3 3 22 2 ( ) ( ) ( ) ( ) t t t t f b px t t dm b px t b px t    (38) we can obtain 2 2 2 2 3 2 2 2 2 3 22 2 ( ) ( ) 2 ( ) ( ) ( ) 2 ( ) ( ) ( ) 0 t t t t t t t t t t f b px t t d f b px t dm x t pb b px t d f b px t dm f b px t b px t                  (39) similarly, we consider the following compensation input for follower 3: 3 3 2( ) 2 ( )v t fd t   (40) for the third term in the right-hand side of 3 ( )t in (35), we find that the inequality 2 3 2 3 2 3 ( ) ( ) ( ) t t t t t t d f b px t d f b px t dm f b px t     (41) is satisfied, and thus 2 ( )t is designed as 3 2 2 32 3 ( ) ( ) ( ) ( ) t t t t f b px t t dm b px t b px t    (42) then we can obtain the following inequality: 3 ( ) 0t  (43) consequently, if ( )v t is designed as 2 2 2 3 22 2 3 3 3 2 32 3 0 ( ) ( ) ( ) ( ) 2 ( ) ( )( ) ( ) 2 ( ) ( ) t tl t t t t t t v t f b px t v t v t fd dm b px t b px tv t f b px t fd dm b px t b px t                                 (44) then we have advances technology innovation, vol. 4, no. 4, 2019, pp. 234-246 242  2 3( , , ( )) 2 ( ) ( ) 0p f v t t t     (45) once again considering the asymptotic stability condition of (34), the quadratic form term of ( )x t should satisfy ( ) ( , , ) ( ) 0 t x t p k f x t  (46) the inequality of (46) is equivalent to the following condition; ( , , ) 0p k f  (47) in order to design the consensus gain f , and the feedback gain k , we introduce the symmetric positive definite matrix s satisfying 1 s p   and change of variables 2 4 kw ks    and 2 4 fw fs    . moreover, preand post-multiplying (47) by 3( )i s , we get. 11 12 13 22 23 33 11 3 3 0 0 0 0 * 0 * * 0 * * * ** * * 0 * * ( , 2 ) , ( ) ( , , )( ) t t t t t k k t t t f f t f f k k t t t k k a p pa f b p f b ps s s a p pa f b p pbf s s sa p pa a a bk a a bk bf sa as w b bw i s p k f i s                                                               12 13 22 33 232 2 , t t k t t t t t t t k k f f f f w b sa as w b bw bw w b w b bw                  (48) this inequality of (48) is linear matrix inequality (lmi) for s , kw and fw . if the solution of the lmi of (48) exists, the asymptotic stability of the closed-loop system of (22) is guaranteed, and the feedback gain k and the consensus gain f can be obtained as 1 1 , k fk w s f w s     (49) from the above discussion, the proof of theorem1 is accomplished. remark 1: in this paper, we approached the case of network topology such as fig.1 as an example, but the other topological structures can be handled if similar theoretical development is applied. however, it is inevitable that lmi will increase in size and complexity by the number of agents and the topology becomes complicated. remark 2: when the relative position between the leader and the follower is considered explicitly, it is often uncertain or unknown about the relative position between the leader and the other followers. thus, construction of the state equation is generally not easy. as a result, there are not much exiting results which have explicitly dealt with the relative position in the dynamics as far as we know. on the other hand, in this study, it is possible to discuss lmi-based control system design that clearly indicates relative positional relationship by adding an input using the maximum value of relative distance that is known when performing the desired formation. 5. numerical simulation firstly, by solving lmi (48), we have symmetric positive definite matrices 4 4s  , 4 4 p   and matrices 2 4 kw   and 2 4 fw   which are given by advances technology innovation, vol. 4, no. 4, 2019, pp. 234-246 243 1 3 3 1 3 3 3 3 1 3 3 1 1.1574 4.2577 10 2.0326 10 3.4553 10 4.2577 10 1.1574 3.4553 10 2.0326 10 2.0326 10 3.4553 10 1.1574 4.2577 10 3.4553 10 2.0326 10 4.2577 10 1.1574 s                                        (50) 1 1 4 3 1 1 3 4 4 3 1 1 3 4 1 1 9.9925 10 3.6760 10 2.3596 10 2.4249 10 3.6760 10 9.9925 10 2.4249 10 2.3462 10 2.3596 10 2.4249 10 9.9925 10 3.6760 10 2.4249 10 2.3462 10 3.6760 10 9.9925 10 p                                        (51) 3 18 3 1.1651 1.9053 9.7081 10 1.6958 10 9.7077 10 2.1926 1.1651 1.9053k w                 (52) 6 1 6 1 6 1 6 1 6.9246 10 7.9258 10 6.9246 10 5.6843 10 6.9246 10 5.6843 10 6.9246 10 7.9258 10 fw                      (53) then the feedback gain 2 4 k   and the consensus gain 2 4 f   can be calculated as 2 3 1 1.8646 2.3321 1.4595 10 6.8409 10 8.2057 10 2.1978 1.8699 2.3327 k                 (54) 1 1 1 1 1 1 1 1 2.9273 10 7.9213 10 2.1088 10 5.6819 10 2.1088 10 5.6819 10 2.9274 10 7.9213 10 f                      (55) in this example, initial values for the closed-loop system of (22) are selected as follows; 2 3 5 6 1 2 4 2 (0) , (0) , (0) 3 0 2 2 3 4 lx x x                                 (56) furthermore, let ( )r t be the leader’s reference input and ( )lx t be ( ) ( ) ( )l lx t x t r t  . in this example, ( )r t gives the leader to go around a circle of radius 3 with 20[s]. also, give 2d and 3d are so that the followers 2 and 3 leaves the leader by (2, 2) and (-2, 2).         2 3 3cos 0.1 2 2 0.3sin 0.1 0 0 ( ) , , 2 23sin 0.1 0 00.3cos 0.1 t t r t d d t t                                      (57) additionally, 2dm and 3dm are selected as 2 3 6dm dm  . the simulation result of this numerical example is shown in figs. 2 7. in figs.2 5, show the state trajectory of each agent and the shape of the formation every 5[s]. figs.6 and 7 show the time histories of each agent in the x and y-axes, respectively. from figs.2-5, we can see that the leader follows the given trajectory, and the followers 2 and 3 follow the desired relative position. also, from fig.6 and fig.7, looking at the transition of the position coordinates of each agent, it can be seen that the follower moves away from the leader's movement locus by the desired position as time passes. namely, it can be said that the proposed formation control system has been designed, and thus we have shown the effectiveness of the proposed formation control systems. advances technology innovation, vol. 4, no. 4, 2019, pp. 234-246 244 fig. 2 movement of each agent in 0[s] fig. 3 movement of each agent in 10[s] fig. 4 movement of each agent in 20[s] fig. 5 movement of each agent in 40[s] fig.6 time histories of xix (t) fig.7 time histories of yix (t) 6. conclusions in this paper, we present a design method of an adaptive gain controller considering relative distances between agents for mass with leader-follower structure. the proposed adaptive gain controller consists of the state feedback laws with fixed gains and compensation input with adaptive gains which are adjusted by updating rules. we have shown that the sufficient conditions for the existence of the proposed adaptive gain controller are reduced to the solvability of lmi, i.e. the proposed controller can be designed by using software such as matlab’s lmi control toolbox, scilab’s lmitool and so on. in the proposed control strategy, there is no need the information on the target value of the other followers and the information about the upper bound on relative positions is only required. furthermore, the effectiveness of the proposed formation control system has been shown through a simple numerical example. advances technology innovation, vol. 4, no. 4, 2019, pp. 234-246 245 the future research subjects are an extension of the proposed design to such a broad class of systems as discrete-time systems and output feedback systems. moreover, for the proposed adaptive gain controller, improvement of transient performance and guaranteeing disturbance attenuation level are our important future research subjects. additionally, we will study the conservativeness of the proposed controller design and extend the proposed controller synthesis to such a broad class of control systems as a formation for mass consisting of more general agent’s dynamics with uncertainties and consensus via output feedback controllers. conflicts of interest the authors declare no conflict of interest. references [1] m. norton, modern control engineering, pergamon press, 1972. [2] b. d. o. anderson and j. b. moore, optimal control – linear quadratic method –, prentice hall, new jersey, 1990. [3] k. zhou, essentials of robust control, prentice hall, 1998. [4] b. r. barmish, “stabilization of uncertain systems via linear control,” ieee trans. automat. contr., vol. 28, no. 8, pp. 848-850, august 1983. [5] i. r. petersen and c. c. hollot, “a riccati equation approach to the stabilization of uncertain linear systems,” automatica, vol. 22, no. 4, pp. 397-411, july 1986. [6] k. hagino and h. komoriya, “a design method of robust control for linear systems,” ieice trans. fundamentals, vol. j72-a, no. 5, pp. 865-868, may 1989. (in japanese) [7] p.p. khargonekar, i.r. petersen and k. zhou, “robust stabilization of uncertain linear systems: quadratic stabilizability and control theory,” ieee trans. automat. contr., vol. 35, no. 3, pp. 356-361, march 1990. [8] i. r. petersen and d. c. mcfarlane, “optimal guaranteed cost control and filtering for uncertain linear systems,” ieee trans. automat. contr., vol. 39, no. 9, pp. 1971-1977, september 1994. [9] s. yamamoto and k. yamauchi, “a design method of adaptive control systems by a time-varying parameter of robust stabilizing state feedback,” tran. iscie, vol. 12, no. 6, pp. 319-325, may 1999. (in japanese) [10] m. maki and k. hagino, “robust control with adaptation mechanism for improving transient behavior,” int. j. contr., vol. 72, no. 13, pp. 1218-1226, november 1999. [11] h. oya and k. hagino, “robust control with adaptive compensation input for linear uncertain systems,” ieice trans. fundamental, of electronics, communications and computer sciences, vol. e86-a, no. 6, pp. 1517-1524, june 2003. [12] d. d. siljak, decentralized control of complex systems, new york, academic press., 1991. [13] r. olfati-saber and r. m. murray, “consensus problems in networks of agents with switching topology and timedelays,” ieee trans. on automat. contr., vol. 49, no. 9, pp. 1520-1533, september 2004. [14] g. xie and l. wang, “consensus control for a class of networks of dynamic agents: fixed topology,” proc. of the 44th ieee conference on decision and control, and the european control conference 2005, pp.96-101, seville, spain, december 2005. [15] y. zhang and y. p. tian, “consentability and protocol design of multiagent systems with stochastic switching topology,” automatica, vol. 45, no. 5, pp. 1195-1201, may 2009. [16] g. zhai, s. okuno, j. imae and t. kobayashi, “a matrix inequality based design method for consensus problems in multi-agent systems,'' int. j. appl. math. comput. sci., vol. 19, no. 4, pp. 639-646, april 2009. [17] w.ni, and d.cheng, “leader-following consensus of multi-agent systems under fixed and switching topologies,” systems & control letters, vol.59, issues.3-4, pp.209-217, march-april 2010. [18] d.zhang, z.xu, q.g.wang, and y.b.zhao, “leader–follower h_∞ consensus of linear multi-agent systems with aperiodic sampling and switching connected topologies,” isa transactions, vol.68, pp.150-159, may 2017 [19] w.he, c.xu, q.han, f.qian, and z.lang, “finite-time l_2 leader–follower consensus of networked euler–lagrange systems with external disturbances,” ieee transactions on systems, man, and cybernetics: systems, vol.48 issue 11, pp.1920 – 1928, november 2018. [20] t. namerikawa, “control theoretical approach to multi-agent problems,” institute of electronics, information, and communication engineers, ieice technical report nc2008-1, may 16, 2008. (in japanese) advances technology innovation, vol. 4, no. 4, 2019, pp. 234-246 246 [21] s. nagai, h. oya, t. kubo, and t. matsuki, “synthesis of decentralized variable gain robust controllers for a class of large-scale interconnected systems with mismatched uncertainties,” int. j. syst. science, vol. 48, issue 8, pp. 1616-1623, july 2017. [22] c. yoshioka, t. namerikawa,” consensus problem for multi-agent system and its application to formation control,” transaction of the society of instrument and control engineers, vol. 44, no. 8, pp. 663-669, august 2008. (in japanese) copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). microsoft word 4-aiti#5476 31-38.docx advances in technology innovation, vol. 6, no. 1, 2021, pp. 31-38 multicriteria decision analysis on information security policy: a prioritization approach jonathan salar cabrera 1,* , ariel roy luceño reyes 2 , cindy almosura lasco 1 1 institute of computing and engineering, davao oriental state college of science and technology, philippines 2 college of information and computing, university of southeastern philippines, philippines received 06 april 2020; received in revised form 08 august 2020; accepted 11 october 2020 doi: https://doi.org/10.46604/aiti.2021.5476 abstract security is the most serious concern in the digital environment. to provide a sound and firm security policy, a multi-holistic approach must be considered when making strategic decisions. thus, the objective of this study was to evaluate the information security (is) and decision making of davao oriental state university (dorsu) using the analytic hierarchy process (ahp) approach. the four aspects of is, namely, the technology, management, economy, and culture were used with the three is components consisting of confidentiality, integrity, and availability to implement the ahp. the results showed that the technology and management have higher significant values than the economic and cultural aspects. meanwhile, for the is components, the integrity signifies the highest priority followed by confidentiality, lastly, and availability. these results emphasize an imbalance in implementing is policy, which must be addressed to ensure that the data integrity, confidentiality, and availability are balanced, particularly during the information exchange transactions. keywords: analytic hierarchy process, evaluation process, information security, prioritization 1. introduction data is an essential element in the organization that needs to be protected. therefore, attacks usually focus on the internal controls of the organization that enables the hampering of information. such views on the importance of data-enabled modern organizations to take an advanced step in ensuring that data is protected from unauthorized access. this involves integrating technology and the employment of governance and policy enforcement mechanisms to guarantee data protection and compliance. first-world countries like the usa and the united kingdom have been strengthening their laws and regulations on information security (is) to prevent unwanted events such as the recent facebook–cambridge analytica data scandal. it is unfortunate that, in the philippines, the current setting of its is law does not conform yet to the current trend of the is environment. however, in terms of is control and security, organizations can follow international standards. deciding to make the best decision of what security policies to implement is a challenging task, especially when there are several aspects to consider. in the fast-changing information age, the is policy in the organization is also evolving to catch-up with the change. subsequently, all possible options in the is aspects should be considered to develop effective and appropriate policy. literature shows that is developments mainly focused on technical and managerial aspects [1]. however, in the information age, information technology is merely affecting cultural and economic aspects. combining cultural, economic, technology, and management aspects into is-related decisions expand the views from different perspectives. hence, a suitable and appropriate method is highly required to analyze by incorporating those aspects carefully. thus, mcda is highly recommended. * corresponding author. e-mail address: jonathan.cabrera@doscst.edu.ph advances in technology innovation, vol. 6, no. 1, 2021, pp. 31-38 32 in this paper, the ahp approach under mcda is used to evaluate the is decision making of davao oriental state university (dorsu). section 2 describes the related literature of the components and aspects of the information security applied in this study. section 3 discussed the methodology with the mcda-ahp evaluation together with partial results as the steps progress. the results and discussions are discussed in section 4, whereas the conclusion and the recommendations are given in section 5. 2. literature review it is necessary to briefly review the elements of the information system to fully understand the importance of information security. typically, information systems in an organizational setting are blends of software, hardware, and telecommunications networks to collect, produce, and distribute a useful data [2]. on the other hand, is is defined as creating a set of practices that keeps the information and information systems secure from the unauthorized access, usage, leakage, retardation, alteration, or destruction [3-4]. with the significance of the information security, its role is vital due to the digitalization of the business processes of the organization. sharing information using various information technologies provides risk. as a result, security is highly needed. in the fast-changing information era, is plays an important role that the organization should put is as a priority to secure thrust in the digital environment. further, works about information show various matters concerning is policy [5-7]. the confidentiality, integrity, and availability (cia) triad is considered as the core principle of is [8]. however, as stated in the five pillar information assurance data security model [9-10], a continuous debate suggests that aside from confidentiality, integrity, and availability, the is core principle can still be extended to include authenticity and non-repudiation features. cia, or often called the security triad, should always fulfill to achieve the is objectives in the organization. confidentiality, as the first component of the cia triad, denotes thwarting the information leakage to adversaries which is very essential in maintaining the secrecy of personal information held by the system [8]. integrity, on the other hand, is the second component of the cia triad and lies in upholding and assuring the truthfulness and dependability of data. this component denotes that modifications on data must not be made without authorization or proper consent to conduct data changes. aside from the confidentiality, it is necessary that is must provide a message integrity, which refers to ensuring that messages have never been modified, altered or tampered with [8]. lastly, the availability, the third out of three components of the cia triad, is thought as a must since it is significant for any information system to be accessible whenever it is needed. it means that the proper functionality of the computing systems which are to store, process, access, and protect the information must be maintained to ensure the accessibility of security controls and communication channels at all times. guaranteeing availability also encompasses prevention from denial-of-service attacks [8]. subsequently, the cia is always a part of the is aspects, referring to the perspectives of the organization or business. mostly, the organization focuses on the management and technical aspects of is [1]. furthermore, recent studies are giving high emphasis on cultural [11] and economic aspects [12] in the information security. in general, aspects of the information security can be categorized into the technology, management, culture, and economy. technology is the most vital guard to ensure the security of information [13]. since the start of the digital age, apprehensions were mostly directed to safeguarding information, and technology, which includes hardware, information/data, and applications. computers, wired/wireless networks, and internet security were amongst the primary concern [14]. likewise, the management in information security is all about ensuring information handling in the organization. while the economy is another crucial aspect of is, it has been recently recognized that economic concerns play a noteworthy role in warranting the level of security measures within an organization [1]. by disregarding the different economic aspects involved in is which includes investment, incentives, and financial information sharing, it will be difficult to determine the economic benefit of such protections [15]. accordingly, a measurement of the economic aspect of is can be done quantitatively. last of all, the cultural aspect of information security refers to human attributes such as behaviors, attitudes, and values that contribute to the protection of all kinds of information in advances in technology innovation, vol. 6, no. 1, 2021, pp. 31-38 33 a given organization [16]. it is also the least important aspect in almost all organizations. in addition, as stated by ngo et al. [17], information security culture is formed by the conventional conduct and actions of workforces and the organization as a whole, and how things are done. in overall assessment in the information security aspect, the three components (i.e., confidentiality, integrity, and availability) should exist altogether to guarantee that the information is confident in terms of protecting disclosure of information, without any alteration or modification by unauthorized actions as well as it is available when required by authenticated person or systems [13]. 3. methodology information security policies are critical because they must be able to review the risk appetite of an organization's management. the proper evaluation of is policies will not only address the need of an organization to create a mechanism that protects it from internal and external threats, but also help in directing a managerial mindset on implementing security within the organization [18]. hence, with such high regard on the safety, the methodology of this paper is focused mainly on evaluating is policy using the ahp framework under mcda concepts as shown in fig. 1. this approach has three levels: goal, criteria, and indicators. the goal is top-level, which specifies the objective of this paper: the information security policy evaluation; the second level is the criteria, which are four aspects of the information security policy; and last but not least, the indicators which are the three security components. fig. 1 mcda-ahp framework for information security policy assessment mcda is a useful and valuable tool when applying complex decisions. using a structured approach, mcda analyzes and measures a series of alternatives or criteria to discern their relative importance and identify which criterion is the most significant. likewise, the ahp is an mcda approach and a decision support tool introduced by saaty [19-20]. ahp is a hierarchical approach tool used to solve complex decision problems. the ahp hierarchy is a top to bottom approach, from the top is the goal, criteria are in the middle, and at the bottom is the indicators presented in fig. 1 [21]. the weights of each criterion and indicators must be determined using pairwise comparisons. the weights of the criteria and indicators show the importance of decision making. there are six steps to process a complex problem using ahp [19, 22]. as described in the paper of cabrera & lee [21], the first step is when the problem has already been identified, break the problem down into its component factors. second, these component factors are arranged and constructed into a hierarchy, then a pairwise comparison matrix is built on the third step. hence, the decision-makers (dm) can systematically assess the alternatives for each of the chosen criteria or indicators. during the fourth step, the weights of each criterion are calculated based on the values assigned by the dm and on the fifth step, the results are then analyzed to establish the prioritization of the criteria and indicators. after that, the consistency is checked to determine the reliability of the results. advances in technology innovation, vol. 6, no. 1, 2021, pp. 31-38 34 the consistency ratio (cr) is the key component of ahp. the cr should be less than 10%, so that the comparison matrix is considered acceptable, and the judgements of the dm are reliable (see sub-section 3.3 for the computation). the succeeding sub-sections are the processes to come up with the acceptable relative weights in every criterion and indicators. 3.1. pairwise comparison in order to use the ahp, a pairwise comparison matrix must be conducted first. this was done by comparing each criterion of the study based on saaty’s scale [19] as displayed in table 1. the results were in integer values (i.e., 1 to 9). the higher number means that the chosen factor is more important than the other. table 1 the fundamental scale and its description as described by cabrera & lee [21] scale judgement of preference description 1 equally important two factors contribute equally to the objective 3 moderate experience and judgement slightly favor one over the other 5 strong experience and judgement strongly favor one over the other 7 very strong experience and judgement very strongly favor one over the other 9 extremely important the evidence favoring one over other is of the highest possible validity 2, 4, 6, 8 intermediate values when compromise is needed 3.2. normalization the normalization in ahp is a probability assigned to the suitability of each alternative. this step used the normalized matrix. this matrix is used to add the values in each column. in the pairwise comparison matrix, the entry in each column is divided by the sum of the column. then, the result will be input in the corresponding cell in the normalized matrix. if the total value in the column is 1, the results in all cells in the column are normalized values as described in eq. (1). finally, the priority vector (pv) (i.e., the weights of the criterion or indicator) is computed by dividing the sum of the column of the matrix by the number of criteria used (n), as shown in eq. (2), the cij in eq. (1) refers to the value of a criterion or indicator in the pairwise comparison matrix. the xij is the normalized score, while pvij refers to the weights of each criterion or indicator. the pvs give the relative importance weights of the criteria or indicators. 1 / n ij ij ij i x c c = = ∑ (1) 1 / n ij ij j pv x n = =∑ (2) 3.3. consistency analysis the consistency analysis (ca) is the last process in ahp. in order to derive the cr, the ca process has to undergo three steps. the first step is to calculate the consistency measure (cm) which can be obtained through multiplying the pairwise matrix with the pv. the result is then divided into the weighted sum vector with its criterion weights. the second step is calculating the consistency index (ci) as described in eq. (3). the λmax refers to the sum of the cm divided by the n (i.e., number of criteria or indicators). finally, the cr is computed by the ci over the ri as described in eq. (4). max ( ) / ( 1)ci n n= − −λ (3) /cr ci ri= (4) the λmax values are set to 4.03, 3.01, 3.01, 3.04, and 3.07 for the goal, technology, management, economy, and culture, respectively. the values of the random index (ri) developed by saaty [20]and its corresponding number of compared criteria are 0.00, 0.58, 0.90, 1.12, 1.24, 1.32, and 1.41 for the 2, 3, 4, 5, 6, 7, and 8, respectively. advances in technology innovation, vol. 6, no. 1, 2021, pp. 31-38 35 4. results and discussion the advantage of ahp is the ability to quantify the inconsistency in the judgment of the decision-makers. as stated by saaty [19, 22], the cr should be less than ten percent (10%) to guarantee that the decision is reasonably correct. if inconsistency occurs, the survey should be repeated until the cr is less than ten percent (10%). the survey should be repeated in cases where cr is less than ten percent to guarantee the level of consistency at an acceptable level. the researchers fulfilled the pairwise comparison matrix from the survey conducted to the it head of the davao oriental state university. the it head is responsible for managing and supervising the it functions of the university, which includes one main campus and three external campuses, situated on different municipalities in the province of davao oriental. as of this writing, the university has an overall population of approximately 10,000 students with all these data stored in the electronic school's management system (esms) and access to the school's e-learning management system (elms) that are both administered and managed by the information technology services unit. there were five comparison matrices created, representing the it head's opinion in the current is policy implementations according to the ahp framework. in terms of the is policy, the technology criterion is highly prioritized, followed by management, economy, and cultural aspect, respectively as shown in table 2. table 2 matrix concerning the goal criteria t m e c technology (t) 1 2 3 4 management (m) 0.5 1 2 3 economy (e) 0.33 0.5 1 2 culture (c) 0.25 0.33 0.5 1 sum 2.08 3.83 6.5 10 moreover, tables 3-6 show the importance of the three alternatives (i.e., confidentiality, integrity, and availability) for every single criterion. in the technology perspective, integrity is highly prioritized followed by availability and confidentiality. in the case of management, the topmost priority is the integrity. the next is the confidentiality and the last is the availability. finally, economically and culturally speaking, the confidentiality indicator is prioritized followed by the integrity and availability. table 3 pairwise comparison matrix concerning technology criteria c i a confidentiality (c) 1.00 0.33 0.50 integrity (i) 3.00 1.00 2.00 availability (a) 2.00 0.50 1.00 sum 6.00 1.83 3.50 table 4 pairwise comparison matrix concerning the management criteria c i a confidentiality (c) 1.00 0.50 2.00 integrity (i) 2.00 1.00 3.00 availability (a) 0.50 0.33 1.00 sum 3.50 1.83 6.00 table 5 pairwise comparison matrix concerning the economy criteria c i a confidentiality (c) 1.00 3.00 5.00 integrity (i) 0.33 1.00 3.00 availability (a) 0.20 0.33 1.00 sum 1.53 4.33 9.00 table 6 pairwise comparison concerning the culture criteria c i a confidentiality (c) 1.00 3.00 4.00 integrity (i) 0.33 1.00 3.00 availability (a) 0.25 0.33 1.00 sum 1.58 4.33 8.00 on the other hand, tables 7-11 show the normalized matrix derived from tables 2-6, respectively, which are the pairwise comparison matrices. table 7 shows the percentage value of the criteria concerning the goal. it is revealed that technology is the dominant aspect of the overall is policy perspectives, which accounted for 47% followed by 28%, 16%, and 10% for the management, economy, and culture. advances in technology innovation, vol. 6, no. 1, 2021, pp. 31-38 36 furthermore, tables 8-11 represent the pv (i.e., weight) of the three alternatives (i.e., confidentiality, integrity, and availability). table 8 shows that integrity is rank first with a 54% allocation in choosing the technology aspect in is. the availability and confidentiality are only 30% and 16%, respectively. also, the management aspect, integrity is prioritized first with 54% followed by confidentiality with 30% and 16% for availability (see table 9). from the economic point of view, the confidentiality is significantly vital with an accounted value of 63%, while the integrity and availability are 26% and 11%, respectively as shown in table 10. looking at the cultural perspective, as shown in table 11, confidentiality is highly emphasized with a value of 61%. the next is 27% for integrity and 12% for availability. table 7 normalized matrix with the relative weights concerning the goal criteria t m e c total pv cm t 0.48 0.52 0.46 0.40 1.86 0.47 4.05 m 0.24 0.26 0.31 0.30 1.11 0.28 4.04 e 0.16 0.13 0.15 0.20 0.64 0.16 4.02 c 0.12 0.09 0.08 0.10 0.38 0.10 4.02 sum 1.00 1.00 1.00 1.00 1.00 table 8 normalized matrix with the relative weights concerning the technology indicators c i a total pv cm c 0.17 0.18 0.14 0.49 0.16 3.00 i 0.50 0.55 0.57 1.62 0.54 3.01 a 0.33 0.27 0.29 0.89 0.30 3.01 sum 1.00 1.00 1.00 1.00 table 9 normalized matrix with the relative weights concerning the management indicators c i a total pv cm c 0.29 0.27 0.33 0.89 0.30 3.01 i 0.57 0.55 0.50 1.62 0.54 3.01 a 0.14 0.18 0.17 0.49 0.16 3.00 sum 1.00 1.00 1.00 1.00 table 10 normalized matrix with the relative weights concerning the economy indicators c i a total pv cm c 0.65 0.69 0.56 1.90 0.63 3.07 i 0.22 0.23 0.33 0.78 0.26 3.03 a 0.13 0.08 0.11 0.32 0.11 3.01 sum 1.00 1.00 1.00 1.00 table 11 normalized matrix with the relative weights concerning the culture indicators c i a total pv cm c 0.63 0.69 0.50 1.82 0.61 3.13 i 0.21 0.23 0.38 0.82 0.27 3.07 a 0.16 0.08 0.13 0.36 0.12 3.02 sum 1.00 1.00 1.00 1.00 to determine the correctness of the percentage value stipulated in the previous paragraph, cr has to be computed. the cr was found to be 1.1%, 0.8%, 0.8%, 3.3%, and 6.4% for the goal, technology, management, economy, and culture, individually. the cr results specify that in the pairwise comparison and it showed that there is an adequate level of coherency. thus, it can be concluded that the values are coherently acceptable. the last goal of the analysis is to generate global or overall priorities as the final weight of the indicators. the result is shown in table 12 and fig. 2 and 3. based on the results, it can be concluded that integrity is observed as the main priority as compared to confidentiality and availability. integrity accounted for 47%, while confidentiality and availability were 32% and 22%, respectively as shown in fig. 3. advances in technology innovation, vol. 6, no. 1, 2021, pp. 31-38 37 table 12 the overall weight of the indicators goal confidentiality integrity availability overall technology 0.075 0.254 0.141 0.470 management 0.084 0.151 0.045 0.280 economy 0.101 0.042 0.018 0.161 culture 0.061 0.027 0.012 0.100 overall 0.321 0.473 0.216 fig. 2 overall rating on information security aspects fig. 3 overall rating on information security components in the information security aspects, the evaluation result showed that technology is the most important followed by the management while the economic and cultural aspects of is gained the third and fourth spots, correspondingly. as evident to its overall ratings, particularly, dorsu has placed more serious concern on technology with 47% of the ratings, as compared to management with 28%, the economy with 16%, and 10% for culture (see fig. 2). the result reflects that the dorsu implementation for information security policy is imbalanced. various literature shows the importance of cultural [11, 23-25] and economic aspect [12, 16] as the basis on the effective and balances information security policy implementations. the result also shows that information security is a challenging issue in dorsu governance. 5. conclusions and recommendations this study gives a good reason for the application of the ahp approach in the evaluation of is policy. integrating multiple criteria and indicators in the mcda shows a tangible advantage. the mcda-ahp approach showcases the ability to check the inconsistency judgement of the decision-makers in a various criteria scenario. what is more, this approach also displays that decision-makers will be assisted in evaluating the implementation of the is policy. in the aspect of is, technology is found to be the highest priority. likewise, concerning the is component, integrity signifies the highest priority followed by confidentiality and availability. based on the results of this paper, the following recommendations are identified. first, improve the employee’s security awareness by providing training to achieve a comprehensive is culture in the organization. second, economic aspects should be addressed as part of the important factors in is policy. finally, it is essential to note that data integrity, confidentiality, and availability should be balanced, particularly during information exchange transactions bound outside in the organization’s computer networks [26]. other mcda approaches like analytic network process (anp) and fuzzy ahp/anp can also be explored in the future to validate the result in other perspectives. besides, expanding the number of respondents that will include all known decision-makers in the organization that may or may not have an it background in order to give a generalized and holistic view of the result. advances in technology innovation, vol. 6, no. 1, 2021, pp. 31-38 38 conflicts of interest the author declares no conflict of interest. references [1] r. anderson, “why information security is hard: an economic perspective,” proc. annual computer security applications conference, december 2001, pp. 358-365. [2] j. p. laudon and k. c. laudon, management information systems: managing the digital firm, 12th ed. england: pearson education limited, 2012. [3] s. shackleford, a. proia, b. martell, and a. craig, “toward a global cybersecurity standard of care? exploring the implications of the 2014 nist cybersecurity,” texas international law journal, no. 291, 2015. [4] y. p. surwade and h. j. patil, “information security,” e-journal of library and information science, pp. 458-466, january 2019. [5] a. singh, a. vaish, and p. k. keserwani, “information security: components and techniques,” international journal of advanced research in computer science and software engineering, vol. 4, no. 1, pp. 1072-1077, january 2014. [6] a. r. otero, information technology control and audit, 5th ed. new york: auerbach publications, 2018. [7] s. mishra and g. dhillon, “defining internal control objectives for information systems security: a value focused assessment,” proc. european conference on information systems (ecis 2008), january 2008. [8] d. la marca, “aspects of it security,” https://www.mediabuzz.com.sg/research-aug-13/aspects-of-it-security, december 02, 2019. [9] k. brauer, “authentication and security aspects in an international multi-user network,” turku university of applied sciences, thesis, may 17, 2011. [10] h. moghaddasi, s. sajjadi, and m. kamkarhaghighi, “reasons in support of data security and data security management as two independent concepts: a new model,” the open medical informatics journal, vol. 10, pp.4-10, october, 2016. [11] a. alhogail and a. mirza, “information security culture: a definition and a literature review,” world congress on computer applications and information systems, ieee press, october 2014. [12] l. xu, y. li, and j. fu, “cybersecurity investment allocation for a multi-branch firm: modeling and optimization,” mathematics, vol. 7, no. 7, pp. 1-20, july 2019. [13] i. syamsuddin, “strategic information security decision making with analytic hierarchy process,” international research journal of applied basic sciences, vol. 2, no. 11, pp. 426-432, 2011. [14] i. syamsuddin and j. hwang, “the application of ahp to evaluate information security policy decision,” international journal of simulation: systems, science and technology, vol. 10, pp. 46-50, 2014. [15] l. a. gordon and m. p. loeb, “the economics of information security investment,” acm transactions on information and system security, vol. 5, no. 4, pp. 438-457, november 2002. [16] g. dhillon, principles of information systems security: texts and cases, 1st ed., john wiley & sons, 2007. [17] l. ngo, w. zhou, and m. warren, “understanding transition towards information security culture change,” australian information security management conference, june 2014, pp. 67-73. [18] r. dunham, “information security policies: why they are important to your organization,” https://linfordco.com/blog/information-security-policies/, may 5, 2020. [19] t. l. saaty, the analytic hierarchy process: planning, priority setting resource allocation, new york: mc graw-hill, 1980. [20] t. l. saaty, “a scaling method for priorities in hierarchical structures,” journal of mathematical psychology, vol. 15, no. 3, pp. 234-281, june 1977. [21] j. s. cabrera and h. s. lee, “impacts of climate change on flood-prone areas in davao oriental, philippines,” water vol. 10, no. 7, 893, july 2018. [22] t. l. saaty, decision making for leaders: the analytic hierarchy process for decisions in a complex world, 3rd ed. pittsburgh: rws publications, 2012. [23] z. milanović, “information-security culture of youth in serbia,” fbim transactions, vol. 7, no. 1, pp. 110-118, april 2019. [24] t. schlienger and s. teufel, “analyzing information security culture: increased trust by an appropriate information security culture,” international workshop on database and expert systems applications, september 2003, pp. 405-409. [25] l. v. astakhova, “the concept of the information-security culture,” scientific and technical information processing, vol. 41, no. 1, pp. 22-28, april 2014. [26] j. hwang and i. syamsuddin, “information security policy decision making: an analytical hierarchy process approach,” third asia international conference on modeling & simulation, may 2009, pp. 158-163. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation, vol. 4, no. 4, 2019, pp. 247-259 a research survey of electronic commerce innovation: evidence from the literature kai-yu tang 1 , chun-hua hsiao 2,* , mei-chun chen 3 1 department of international business, ming chuan university, taipei, taiwan 2 department of marketing, kainan university, taoyuan, taiwan 3 department of information management, vanung university, taoyuan, taiwan received 11 april 2019; received in revised form 11 june 2019; accepted 02 july 2019 abstract the development of technology has ignited many innovations in business management, especially in the electronic commerce area. the essential example, that is, online stores and online shopping, is a critical evolution and innovation from traditional brick-and-mortar stores to clicks and mortar. following previous research (van oorschot et al.), this present study adopted bibliometric and keyword analysis to review the main characteristics of electronic commerce innovations. focused on the academic sources, the research data used in this study were searched for and collected from the web of science (wos), a renowned academic database which covers the most influential research journals in electronic commerce. based on a combination of several keywords related to “innovation” and “electronic commerce,” the keyword search in the wos was conducted in may 2019. as a result, a total of 334 research articles related to electronic commerce innovations were collected. derived from the bibliometric analysis, some keywords that were seldom used in the earlier decade (2000-2009), but which rapidly grew in use in the recent decade (2010-2018) were found, including m-commerce, platforms, social commerce, online review, and co-creation. in addition, the top 10 influential articles listed in each of the two decades were identified. the results show some of the research trajectories in ec innovations. in the first decade (2000-2009), the top 10 papers focused on traditional it adoption, such as self-service technology, enterprise resource planning systems, and the adoption of general attitude-intention theories such as the technology acceptance model. in the recent decade (2010-2018), researchers have shown more diverse interest in innovative ec applications, such as rfid applications, cloud computing, crowdsourcing, etc. accompanying these ec innovation contexts, in addition to general attitude-intention theories, more theories such as signaling theory, have been adopted. keywords: electronic commerce innovations, research trajectory, systemic literature survey 1. introduction nowadays, one may see many innovations in the development of electronic commerce (ec). for example, bhattacherjee [1] introduced some e-commerce brokerage services, and eastin [2] provided some examples of web technologies applied to consumers’ purchasing behavior, such as online shopping, online banking, and even online investing. indeed, these ec innovations have become part of our daily life. in recent studies, researchers have brought more advanced and innovative ideas into ec business research. while cui et al [3] mentioned some benefits of ec innovations in growing economies in rural china, escobar-rodríguez and bonsón-fernández [4] pointed out other ec applications in the spanish fashion industry. more recently, vakulenko et al. [5] offered the possibility to create an e-customer journey map through innovative management of *corresponding author. e-mail address: maehsiao@gmail.com tel.: +886-3-341-2500; fax: +886-3-3412430 advances in technology innovation, vol. 4, no. 4, 2019, pp. 247-259 248 ec services. these original ideas have acted as a lighthouse for ec business, shedding light on the next-generation ec innovations. however, few studies have focused on the analysis of ec innovations from an academic viewpoint. it is especially necessary because the rigorous nature of academic research has enhanced the reliability of the innovations in real business practices. moreover, researchers can follow some performance indicators of scientific research, such as times cited per year, to identify and trace the high impact ideas, ec innovative business models or practices. three purposes of this study were therefore formed as follows. first is ranking the most productive countries and journal publications in the ec innovations literature; second is identifying the most influential research articles across two decades of observation (the first decade: 2000-2009; the recent decade: 2010-2018); and the last one is highlighting the most critical innovative practices of ec in the two decades of research. 2. data and methods 2.1. the process of data inclusion in this paper, a systematic approach for data inclusion was adopted to achieve the above three research questions. the first process of data inclusion was to identify the keywords related to innovation and electronic commerce from the literature. the selection of keywords referred to previous review studies in the fields of innovation management and electronic commerce. in the field of innovation management, randhawa et al. [6] reviewed “open innovation” in product innovation research. using bibliometric analysis, van oorschot et al. [7] analyzed the “innovation adoption” literature with the issue of technology and social change. in electronic commerce literature, akter and wamba [8] provided a systematic review of “e-commerce” in electronic markets. han et al. [9] conducted a literature survey of “social commerce” in e-commerce research. more recently, gursoy’s [10] presented a critical review to highlight the importance of consumers’ “online review” in ec platforms. as such, we adopted the most popular terminology used in the above reviews, such as “innovative,” “innovation,” “electronic commerce,” “e-commerce,” “e-tailer,” “social commerce,” and “online review” as the keywords for this search. the second step was to select the sourcing database. the web of science (wos) was selected as the primary source as it covers the most influential ec journals in the categories of business, management, and information science, including the international journal of electronic commerce, electronic commerce research and applications, and mis quarterly. as such, the above seven terms were used as keywords for searching for high-quality electronic commerce innovation research in terms of citation counts. (brouthers et al.) next, two researchers carefully removed those redundant and incomplete items selected during the search. a final total of 334 articles published from 2000 to 2018 were obtained as the literature of research interest. the process of data inclusion was completed on 22 may 2019. more information for the selection of highly cited articles and keywords is provided below. 2.2. the selection of high impact articles in this study, the numbers of articles published and their authorship were counted and are shown in table 1. note that only the number of non-repeated authors was reported to show the trend of researchers involved in the field. as shown in table 1, a total of 118 articles with 256 authors in the first decade were found (2000-2009). furthermore, figure 1 shows a highly correlated pattern of publications between the numbers of published papers and involved authors, with a pearson correlation coefficient of 0.98. the scale of publications increased to 216 articles and 558 authors in the recent decade (2010-2018), which approximately doubled the number of articles and authors in the first decade. the growth of authors involved in the ec innovation research in the last five years (2014-2018, as shown in fig. 1) is especially noteworthy, indicating the great attraction of ec innovation research to academics. in this study, we adopted a bibliometric analysis to identify the most influential among the 334 articles for each decade. the main idea of bibliometric analysis is to reveal the impact of scientific advances in technology innovation, vol. 4, no. 4, 2019, pp. 247-259 249 publications. the two often-used bibliometric indicators, total times cited (citations) and times cited per year (average citations), were to identify the most influential articles of ec innovation. all 334 articles in this analysis had received a total of 13,858 citations. the citation information was also collected from the wos document by document, providing evidence of the most critical practices in electronic commerce innovation. the pilot results of the selection of high impact articles are shown in the next section. furthermore, based on the calculation of total citations received by journal publications and the country in which the researchers are affiliated, the indicator of scientific performance (i.e., h-index) was also provided for the analysis of the most high-impact journals and productive countries in the field. table 1 trend of ec innovation research: articles published and authors involved year articles published authors (non-repeated) 2000 5 8 2001 13 31 2002 8 13 2003 11 31 2004 12 20 2005 14 33 2006 14 30 2007 11 22 2008 9 19 2009 21 49 sub-total (2000-2009) 118 256 2010 28 63 2011 15 36 2012 19 55 2013 11 23 2014 13 24 2015 26 71 2016 31 86 2017 38 94 2018 35 96 sub-total (2010-2018) 216 558 total (2000-2018) 334 814 fig. 1 the trend of ec innovation research: articles and authors 2.3. keyword analysis following previous scientometric research [11], we used a keyword analysis to decompose the most valuable content of the corpus from these 334 ec innovation articles. all the wording in the title, keyword, and abstract were gathered. some insignificant stop words were excluded, such as “and,’’ “at,” ‘‘in,” “above,” “under,” and so on. other terms with similar meanings in the context of ec innovation were aligned, for example, “electronic commerce,” “e-commerce,” “innovation,” and “innovations.” next, a process of scraping common terms was conducted to help identify those keywords which are discriminative for representing the main characteristics of ec innovation research. advances in technology innovation, vol. 4, no. 4, 2019, pp. 247-259 250 after the list of keywords was aligned, each keyword was then matched and counted if it appeared in a particular article. note that each article may have multiple keywords. only the keywords which appeared in different articles were counted and accumulated; however, those keywords repeatedly used in a single article were counted once. accordingly, a total of 5,211 keywords were obtained. based on the frequency count, the most critical keywords with significant growth trends and decline trends were selected. a further comparison of selected keywords for the two periods of research (period i: 2000-2009, and period ii: 2010-2018) was conducted to identify the research trends in ec innovation. 3. results and discussion in line with the three purposes of this research, the results of the research survey in ec innovation literature are presented. first is the publication patterns in the ec innovation research, including the most productive countries, influential journals, and representative keywords that are frequently used. second is the most influential research articles in the first decade: 2000-2009. and the last one is the most influential research articles in the recent decade: 2010-2018. further discussions of the research changes over the past two decades are also provided. 3.1. the most productive country and journal publications in the ec innovation literature 3.1.1 . the most productive countries to measure each country’s productivity, only the first authors’ country affiliation was counted because the first author has a major contribution to the research. if the first author has multiple affiliations, the main affiliation was manually checked and coded. as a result, a total of 39 countries were found and listed in ec innovation publications as shown in table 2. we also found that authors from the first 10 countries contributed over 105 articles and 158 articles, accounting for over 89% and 73% of the total publications in the periods of 2000-2009 and 2010-2018, respectively. therefore, these countries were labeled as the top 10 productive countries in ec innovation research, namely the usa, taiwan, china, the uk, canada, australia, south korea, portugal, germany, and spain. to highlight the impact of the top 10 countries, the total citations received are included in parentheses. for example, usa (7309) means that all of the articles published by the authors affiliated with the usa received 7,309 citations in total. fig. 2 visualizes the distribution of ec innovation research with the number of publications, showing the publication trends of the most productive countries in ec innovation across the two phases (2000-2009 and 2010-2018). table 2 the most influential countries in the field of ec innovation # country (citations) number of articles published h-index active years 2000-2009 2010-2018 total trends 1 usa (7903) 59 52 111 (↓) 45 2000-2018 2 taiwan (1032) 15 18 33 (↑) 14 2004-2018 3 china (971) 3 34 37 (↑) 13 2003-2018 4 uk (773) 15 10 25 (↓) 11 2001-2018 5 canada (706) 4 5 9 (↑) 15 2001-2018 6 australia (392) 6 6 12 (-) 10 2000-2016 7 south korea (278) 2 6 8 (↑) 5 2004-2016 8 portugal (277) 0 5 5 (↑) 4 2014-2018 9 germany (212) 0 8 8 (↑) 5 2011-2018 10 spain (198) 1 14 15 (↑) 7 2008-2018 more specifically, the result indicates that over one-third of the publications were from the usa in terms of the total number of articles from 2000 to 2018, (n = 111, 33%); the usa also ranked as the most productive country in period i (n = 59, 50%), but not in period ii (n = 52, 24%). it shows a decreasing trend of ec innovation research in usa publications. however, the overall 7,903 citations of the usa were ranked as the first place, revealing that us researchers have a significant impact both in the quantity and quality of ec innovation research. we also found that the uk ranked second in the period i (n = 15, 13%) in terms of publications, but dropped rapidly in period ii (n = 10, 5%). this trend indicates that the researchers affiliated advances in technology innovation, vol. 4, no. 4, 2019, pp. 247-259 251 with the usa and the uk have dominated the ec innovation research, but an increasing number of researchers from other countries have since joined the field. note that australia’s researchers showed a balanced performance in ec innovation across the two decades. fig. 2 the most productive country in the ec innovation research on the other hand, a growth pattern of ec innovation publications appeared in the remaining countries, led by taiwan. researchers from taiwan have been ranked as the second most productive academics in terms of overall citations received (1,032 times). a total of 15 articles published in period i (2000-2009) and 18 articles in period ii (2010-2018), demonstrating a growth trend of taiwan’s ec innovation research. china ranked third in terms of total citation times. china researchers published three articles in period i, and rapidly developed to the second in period ii (n = 34, 16%). the fifth to tenth countries are canada (#5), south korea (#7), portugal (#8), germany (#9), and spain (#10), while australia (#6) showed a balanced performance in ec innovation. all of them have ec innovation articles in both stages. the following countries (i.e., germany (#9) and portugal (#8)), however, started to publish ec innovation articles in 2011 and 2014, individually. this reveals that some potential research issues from europe and latin america are also worthy of follow-up attention. 3.1.2 the most influential journals fig. 3 the most influential journals in the ec innovation research table 3 shows the most influential journals in the ec innovation research between 2000 and 2018. based on brouthers et al.’s suggestion [12], the rankings of the most influential journals are based on the total number of citations, not on journal ranking or the number of journals published. as such, the results showed that the four journals grew in terms of publication trends ranked by citations received, while another four declined. as a leading journal with a growth trend, mis quarterly (misq) ranked first, with 39 articles receiving 3,729 citation counts, and the h-index was 24. the results suggest the importance of misq in ec innovation research. similarly, the following three journals with growth trends were information & management (7→10), the international journal of electronic commerce (5→12), and technological forecasting and social advances in technology innovation, vol. 4, no. 4, 2019, pp. 247-259 252 change (4→9). the first two journals published at least 17 articles on ec innovation and received about a thousand citations each. technological forecasting and social change was active between 2004 to 2018 with 13 ec innovation articles published among which, focusing on the new applications of ec, wang et al.’s research [13] titled “understanding the determinants of rfid adoption in the manufacturing industry” was the most cited (176 citations). table 3 the most cited journals in the field of ec innovation # journal title (citations) number of articles published h-index active years 2000-2009 2010-2018 total trends 1 mis quarterly (3729) 8 31 39 (↑) 24 2002~2018 2 information & management (1482) 7 10 17 (↑) 12 2002~2016 3 international journal of electronic commerce (972) 5 12 17 (↑) 12 2000~2018 4 information systems research (949) 3 3 6 (-) 5 2001~2018 5 journal of marketing (556) 2 2 (↓) 2 2005~2009 6 management science (485) 2 2 4 (-) 4 2006~2017 7 internet research (456) 11 5 16 (↓) 11 2000~2018 8 organization science (428) 4 1 5 (↓) 5 2001~2017 9 technovation (399) 4 4 (↓) 4 2006~2009 10 technological forecasting and social change (329) 4 9 13 (↑) 6 2004~2018 note that the abbreviations of journal titles. misq is mis quarterly; i&m is information and management; ijec is international journal of electronic commerce; isr is information systems research; jm is journal of marketing; ms is management science; ir is internet research; os is organization science; tfsc is technological forecasting and social change. the remaining journals have mostly shown a downward trend in terms of the number of publications, including journal of marketing, internet research, organization science, and technovation. based on the analysis, only two articles related to ec innovation were published in the journal of marketing; however, the influence was considerable. for example, meuter et al.’s research [14] “choosing among alternative service delivery modes: an investigation of customer trial of self-service technologies” was cited 2005 times, while li et al.’s article [15] “internet auction features as quality signals” was cited 68 times. the two articles pointed out some innovations of alternative service delivery modes and quality signals of the internet auction. these findings can be considered as the research fronts of the online-to-offline model and online review for online shopping. internet research published 16 articles on ec innovation, suggesting that the ec research has become mature and that researchers have turned their interest to new topics in the field. the publication trend of two journals was stagnant, but still active in recent years (information systems research: 2001-2018; management science: 2007-2017) and they received a number of citations from follow-up studies (information systems research: 949 times; management science: 485 times). the changes in publication patterns for the top 10 journals are also presented in figure 3. 3.1.3 the most frequently used keywords further analysis of publication patterns was conducted by using keyword analysis. in this study, a total of 5,210 keywords were obtained, among which 3,785 were found to show an increasing trend (more frequently used in the recent decade than in the earlier one), and the remaining 1,425 showed a decreasing trend. twenty keywords that represent some unique attributes and which are frequently used in the ec innovation research are listed in table 4. the values in each cell are the counts of keywords which appeared in the ec innovation research. the larger the values are, the more frequently the keywords were used in the research articles. in this study, non-repeated counts are presented, showing the mainstreams of ec innovation that most research focused on. two different research trends of keywords were interpreted as follows. advances in technology innovation, vol. 4, no. 4, 2019, pp. 247-259 253 (a) keywords with increasing trends in this study, the counts of most keywords increased. for example, the top two keywords (i.e., innovation and e-commerce) that were used for searching the research data are listed in italics in the parentheses. overall, these two keywords have been most frequently used by researchers: 233 (67%) and 172 (51%) times, respectively. this result provides evidence of content validity for our research data in the ec innovation area. however, the growth trend of these two keywords is different across two research decades. “innovation” was the most adopted keyword in the period i (n=85, 2000-2009) and period ii (n=148, 2010-2018). this indicates a significant growth of “innovation” in the ec innovation research. on the other hand, “e-commerce” appeared as the second most highly used keyword in the first decade (n=82), but the growth in number (n=90) in the recent decade was not as active as that of “innovation.” the following five keywords (#1-5) show the domain-specific characteristics of ec innovation research, namely “model” (173), “data” (121), “implications” (114), “online” (112), and “technology” (105). the first four keywords almost doubled from the first to the present decade (55->118; 35->86; 29->85; 23->89), while the growth of “technology” (40->65) was relatively stable. this result suggests the need for and importance of scholarly research in ec innovation, such as research models, data-driven evidence, online applications, managerial implications for real business operations, and new technologies. the next 10 keywords that rarely appeared in the earlier stage (2000-2009) but which were highly researched in the recent decade (2010-2018) are marked in table 4 (#6-15). they are “m-commerce,” “china,” “platforms” (0->11); “amazon,” “devices,” “image,” “shopping experience” (0->10); “social commerce,” “online review” (0->8), and “co-creation” (0->6). most of the above keywords revealed the emerging applications of ec innovations (e.g., m-commerce, platforms, devices, social commerce, online review, co-creation), while others highlighted the great consumption market, china, and the leading ec company, amazon. it is also interesting to note that “image” and “shopping experience” are specific contexts of consumers’ reviews conducted online. consumers in the digital age are used to shopping online via their smartphones. the images and reviews posted by other consumers have become the first impression of the products they search for. these are popular topics in ec innovation research. table 4 a list of keywords used in the ec innovation articles: comparison of two decades # year counts (all years) 2000-2009 2010-2018 ∆ (innovation) 233 85 148 63 (e-commerce) 172 82 90 8 1 model 173 55 118 63 2 data 121 35 86 51 3 implications 114 29 85 56 4 online 112 23 89 66 5 technology 105 40 65 25 6 m-commerce 11 0 11 11 7 china 11 0 11 11 8 platforms 11 0 11 11 9 amazon 10 0 10 10 10 devices 10 0 10 10 11 image 10 0 10 10 12 shopping experience 10 0 10 10 13 social commerce 9 0 9 9 14 online review 8 0 8 8 15 co-creation 6 0 6 6 16 internet 61 34 27 -7 17 b2b 24 13 11 -2 18 b2c 15 9 6 -3 19 travel agency 5 5 0 -5 20 internet banking 2 2 0 -2 advances in technology innovation, vol. 4, no. 4, 2019, pp. 247-259 254 (b) keywords with decreasing trends note that two main keywords used for the search are listed in italics in the parentheses. the following are the most often used keywords with increasing trends (#1-5); ten keywords that seldom appeared in the earlier stage (2000-2009), but were highly researched in the recent decade (2010-2018) were marked (#6-15); the last five keywords, on the other hand, were found to have a decreasing trend of use in recent ec innovation research (#16-20). table 4 a list of keywords used in the ec innovation articles: comparison of two decades # details of the earlier decade (2000-2009) 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 4 11 7 9 8 9 10 6 7 14 5 13 5 7 9 9 10 7 5 12 1 3 3 3 2 5 9 9 6 5 10 2 0 2 3 3 4 4 5 3 3 8 3 0 1 1 3 3 5 4 3 3 6 4 0 1 1 3 2 4 2 2 0 8 5 2 4 5 4 3 2 7 4 3 6 6 0 0 0 0 0 0 0 0 0 0 7 0 0 0 0 0 0 0 0 0 0 8 0 0 0 0 0 0 0 0 0 0 9 0 0 0 0 0 0 0 0 0 0 10 0 0 0 0 0 0 0 0 0 0 11 0 0 0 0 0 0 0 0 0 0 12 0 0 0 0 0 0 0 0 0 0 13 0 0 0 0 0 0 0 0 0 0 14 0 0 0 0 0 0 0 0 0 0 15 0 0 0 0 0 0 0 0 0 0 16 3 6 0 6 5 3 3 1 1 6 17 2 1 1 2 2 1 2 1 0 1 18 0 1 1 1 0 1 1 2 0 2 19 0 2 0 0 0 0 0 0 1 2 20 0 0 0 1 1 0 0 0 0 0 # details of the recent decade (2010-2018) 2010 2011 2012 2013 2014 2015 2016 2017 2018 21 8 14 7 10 17 22 25 24 18 8 8 4 5 8 13 16 10 1 12 10 7 8 6 15 16 20 24 2 8 7 8 5 6 10 13 13 16 3 14 5 2 2 5 12 11 16 18 4 9 6 4 3 4 12 16 20 15 5 9 7 3 2 4 14 7 9 10 6 0 0 1 0 1 2 2 4 1 7 1 0 1 1 1 1 3 1 2 8 0 1 0 0 0 0 2 5 3 9 1 1 1 0 1 1 0 3 2 10 0 0 1 0 0 2 1 3 3 11 1 0 0 1 0 2 4 2 0 12 1 3 0 0 1 1 0 0 4 13 0 1 0 0 0 1 3 2 2 14 2 0 0 0 1 1 1 2 1 15 1 1 1 0 0 1 1 1 0 16 5 3 4 2 1 3 3 2 4 17 2 0 0 1 1 1 4 1 1 18 1 2 0 1 0 0 2 0 0 19 0 0 0 0 0 0 0 0 0 20 0 0 0 0 0 0 0 0 0 ∆: the difference in keyword counts between the earlier decade (2000-2009) and the recent one (2010-2018). the last five keywords, on the other hand, showed a decreasing trend in recent ec innovation research (#16-20), including “internet” (34->27), “b2b” (24->13), “b2c” (15->9), “travel agency” (5->0), and “internet banking” (2->0). advances in technology innovation, vol. 4, no. 4, 2019, pp. 247-259 255 although the keyword “internet” appeared in 61 articles from 2000 to 2018, it is clear that the figure significantly decreased in recent ec innovation research. we argue that the reason is due to the development of advanced technology. the internet-based environments for commerce have become more general than in the 1990s (the age of the rise of the internet). likewise, ec business models have developed from some essential business to business (b2b) and business to consumer (b2c) models to an integrated model, such as the online-to-offline model (o2o). many industries were also affected and forced to change by the emerging ec innovations, for example, travel agencies versus online travel platforms, and the applications of internet banking versus mobile banking apps that are used on smartphones. 3.2. the most influential research articles in the first decade: 2000-2009 table 5 top 10 research articles of ec innovations in the first decade (2000-2009) # authors title journal main innovations in ec category cites (avg.) 1 [16] understanding and predicting electronic commerce adoption: an extension of the theory of planned behavior mis quarterly (2006) theory of planned behavior; perceived behavioral control; self-efficacy; controllability; technology acceptance model; trust; electronic commerce; consumer behavior information science; management 842 (64.8) 2 [17] post-adoption variations in usage and value of e-business by organizations: cross-country evidence from the retail industry information systems research (2005) technology diffusion; innovation; e-business; it investment; usage; value; back-end integration; firm performance; resource-based view; international perspective information science 530 (37.9) 3 [14] choosing among alternative service delivery modes: an investigation of customer trial of self-service technologies journal of marketing (2005) service delivery modes; self-service technologies; customer trial business & economics 488 (34.9) 4 [18] enticing online consumers: an extended technology acceptance perspective information & management (2002) electronic markets; innovation diffusion; online retailing; technology acceptance model; virtual store information science; management 503 (29.6) 5 [19] the process of innovation assimilation by firms in different countries: a technology diffusion perspective on e-business management science (2006) technology diffusion; innovation assimilation; assimilation process; e-business; competition; firm size; technology integration; international perspective operations research & management science 371 (28.5) 6 [20] the effects of personalization and familiarity on trust and adoption of recommendation agents mis quarterly (2006) trust; electronic commerce; adoption; personalization; familiarity; cognitive trust; emotional trust; recommendation agent; delegation information science; management 338 (26) 7 [21] shaping up for e-commerce: institutional enablers of the organizational assimilation of web technologies mis quarterly (2002) web technology; web implementation; it management; innovation assimilation; structuring actions information science; management 359 (21.1) 8 [22] shopping motivations on internet: a study based on utilitarian and hedonic value technovation (2007) internet shopping; utilitarian motivation; hedonic motivation; shopping motivation; search intention; purchase intention engineering & industrial; management 188 (15.7) 9 [23] informational cascades and software adoption on the internet: an empirical investigation mis quarterly (2009) e-commerce; herding; informational cascades; decision making; network effects; word-of-mouth; software download; online communities; online user review information science; management 143 (14.3) 10 [24] a model of organizational integration, implementation effort, and performance organization science (2005) integration; interdependence; performances; erp implementation; electronic integration management 196 (14) ∆: the difference in keyword counts between the earlier decade (2000-2009) and the recent one (2010-2018). after counting the total and average citations of influential papers in terms of highly cited papers, we identified the top 10 research articles related to e-commerce applications across the two decades. table 5 lists the top 10 highly cited articles along with authors, title, the published journal, main keywords, and citations from 2000-2009. in the following, we will summarize research outlines from the top three articles. as seen in table 5, pavlou and fygenson’s article emerged as the most influential paper in terms of total citations (842) and average citations per year (64.8) [16]. they proposed an integrated model to predict ec adoption based on renowned theories, such as the theory of reasoned action (tra), the theory of planned behavior (tpb), and the technology acceptance model (tam). in addition, other factors including technological characteristics, consumer skills, time and monetary resources, and product characteristics also contributed to the prediction of ec adoption. ranked 2 nd with total citations of 530 and average citations per year of 37.9, zhu and kraemer’s research [17] focused on the post-adoption of the e-retailing industry. based on the innovation diffusion theory (idt) and the resource-based theory, three perspectives of factors including technological, organizational, and environmental factors were employed to accomplish the research purpose. as self-service technologies (sst) have become increasingly popular, meuter et al. [14] adopted variables advances in technology innovation, vol. 4, no. 4, 2019, pp. 247-259 256 (e.g., role clarity, motivation, and ability) obtained from the consumer readiness theory to investigate their influences on consumer trials of sst. as we look inside the articles in the first decade, most of the top 10 highly cited papers are attributed to both the information science and management categories [16-18, 20-21, 23]. in terms of research type, all of them are empirical studies investigating consumers, and only three cover the subject of the organization [15, 19, 24]. in the research contexts, except for the applications of self-service technology (sst) [14] and recommendation agents (ras) [20], most studies applied to ec adoption, including software adoption such as erp [24]. note that four of the top 10 articles are published in the prestigious journal, mis quarterly [16, 20-21, 23]. note that a total of 118 research articles published from 2000 to 2009. the most influential articles (top 10) are listed above. the ranking of the top 10 articles was sorted by the times cited per year (average citations), which is presented in parentheses in the citation column. 3.3. the most influential research articles in the recent decade: 2010-2018. table 6 top 10 research articles of ec innovations in the recent decade (2010-2018) # authors title journal main innovations in ec category cites (avg.) 1 [25] what makes a helpful online review? a study of customer reviews on amazon.com mis quarterly (2010) electronic commerce; product reviews; search and experience goods; consumer behavior; information economics; diagnosticity information science; management 615 (68.3) 2 [26] business models: origin, development and future research perspectives long range planning (2016) business models of innovation, change & evolution, performance & controlling and design. business; development studies; management 124 (41.3) 3 [27] assessing the determinants of cloud computing adoption: an analysis of the manufacturing and services sectors information & management (2014) cloud computing; it adoption; diffusion of innovation (doi); technology-organization-environment (toe) information science; management 177 (35.4) 4 [28] co-creation: toward a taxonomy and an integrated research perspective international journal of electronic commerce (2010) active consumption; co-creation; consumer roles; e-commerce research; taxonomic frameworks business & economics 240 (26.7) 5 [29] trust, satisfaction, and online repurchase intention: the moderating role of perceived effectiveness of e-commerce institutional mechanisms mis quarterly (2014) e-commerce; trust; online repurchase intention; e-commerce; institutional mechanisms; moderation analysis; partial least square modeling information science; management 109 (21.8) 6 [30] what signal are you sending? how website quality influences perceptions of product quality and purchase intentions mis quarterly (2011) signaling theory; signals; cues; website quality; ecommerce; perceived quality; credibility; information asymmetries information science; management 162 (20.3) 7 [31] an integrative model of consumers' intentions to purchase travel online tourism management (2015) innovations diffusion theory; intentions to purchase; online travel shopping; social media; technology acceptance model; theory of reasoned action; theory of planned behaviour hospitality, leisure, sport & tourism; management 81 (20.3) 8 [13] understanding the determinants of rfid adoption in the manufacturing industry technological forecasting and social change (2010) radio frequency identification; technology-organization-environment framework; technology adoption; innovation adoption business; regional & urban planning 176 (19.6) 9 [32] task design, motivation, and participation in crowdsourcing contests international journal of electronic commerce (2011) analyzability; autonomy; co-creation; crowdsourcing; extrinsic motivation; intrinsic motivation; tacitness; task design; variability business & economics 123 (15.4) 10 [33] harnessing the influence of social proof in online shopping: the effect of electronic word of mouth on sales of digital microproducts international journal of electronic commerce (2011) digital microproducts; digital products; electronic word of mouth; ewom; social proofs business & economics 116 (14.5) compared with the articles on ec innovation from 2000 to 2009, the top 10 articles of ec innovation in the recent decade seem to have received fewer citations in terms of both total and average per year citations, with the exception of the top one article. as seen in table 6, mudambi and schuff’s [25] article, ranked as number one, received a total of 615 citations and 68.3 average citations per year. they worked on the critical issue regarding the success of ec business. collected from more than advances in technology innovation, vol. 4, no. 4, 2019, pp. 247-259 257 1,500 reviews from amazon across six products, they found some decisive factors including extremity, review depth, and product type. amblee and bui [33] also targeted amazon’s consumers but worked on another important issue of social influence, i.e., electronic word of mouth (ewom). wirtz et al. [26] presented a position paper (ranked 2 nd ) to articulate the development, evolution, and future research of business models. from the organization perspective, oliveira et al. [27] investigated over 350 firms to assess the determinants of cloud computing adoption based on the theories of diffusion of onnovation theory (doi) and technology-organization-environment (toe). the above are the research outlines of the top three articles. overall, in the recent decade, fives articles were attributed to the business category [13, 26, 28, 32-33], four to the information science management category [25, 27, 29-30], and one to the hospitality & tourism category [31], showing that the ec innovations have been widely applied to various fields. in terms of research type, eight out of 10 articles are empirical research, and only two are qualitative research. while wirtz et al. [26] systematically reviewed the research on business models, zwass [28] conducted a taxonomy of co-creation, another emerging topic. with regard to the research topics, more diverse and innovative themes can be found in the last decade, such as rfid adoption, cloud computing [27], co-creation [28], extrinsic cue signals [30], online travel [31], crowdsourcing contests [32], and recommendation systems [33]. among the top 10 articles, two well-known journals, misq and ijec, each published three articles. note that a total of 216 research articles were published from 2010 to 2018. the most influential articles (top 10) are listed above. the ranking of the top 10 articles was sorted by the times cited per year (average citations), which is presented in parentheses in the citation column. 4. conclusions in line with the main research interest, we have concluded some differences in the development of ec innovation (tables 5 and 6) between the two decades. in summary, the main research streams in the past decade (2010-2018) are more diverse and extensive than the research in the previous decade (2000-2009) in terms of the research categories, research types, and research topics. first, more ec innovation studies were attributed to business and economics fields than to information science and management categories. next, considering the research type, qualitative studies such as research reviews and position papers have emerged in the past decade, indicating that the ec innovation research has matured in recent years. finally, in terms of ec innovation topics, the sst and erp were the foci of research in the early decade, but more innovative applications appeared as hot topics in recent years such as cloud computing, co-creation, and crowdsourcing. we have also identified some theories other than traditional attitude-intention models (e.g., tra, tpb, tam, and idt) that have been used in the contexts of new ec applications over the past decade, including technology-organization-environment (toe), signaling theory, extrinsic and intrinsic motivations, and theory of job design. conflicts of interest the authors declare no conflict of interest. references [1] a. bhattacherjee, “acceptance of e-commerce services: the case of electronic brokerages,” ieee transactions on systems, man, and cybernetics, vol. 30, no. 4, pp. 411-420, 2000. [2] m. s. eastin, “diffusion of e-commerce: an analysis of the adoption of for e-commerce activities,” telematics and informatics, vol. 19, no. 3, pp. 251-267, 2002. [3] m. cui, s. l. pan, s. newell, and l. cui, “strategy, resource orchestration and e-commerce enabled social innovation in rural china,” journal of strategic information systems, vol. 26, no. 1, pp. 3-21, 2017. advances in technology innovation, vol. 4, no. 4, 2019, pp. 247-259 258 [4] t. escobar-rodríguez and r. bonsón-fernández, “analysing online purchase intention in spain: fashion e-commerce,” information systems and e-business management, vol. 15, no. 3, pp. 599-622, 2017. [5] y. vakulenko, p. shams, d. hellström, and k. hjort, “service innovation in e-commerce last mile delivery: mapping the e-customer journey,” journal of business research, vol. 101, pp. 461-468, august 2019. [6] k. randhawa, r. wilden, and j. hohberger, “a bibliometric review of open innovation: setting a research agenda,” journal of product innovation management, vol. 33, no. 6, pp. 750-772, 2016. [7] j. a. van oorschot, e. hofman, and j. i. halman, “a bibliometric review of the innovation adoption literature,” technological forecasting and social change, vol. 134, pp. 1-21, 2018. [8] s. akter and s. f. wamba, “big data analytics in e-commerce: a systematic review and agenda for future research,” electronic markets, vol. 26, no. 2, pp. 173-194, 2016. [9] h. han, h. xu, and h. chen, “social commerce: a systematic review and data synthesis,” electronic commerce research and applications, vol. 30, pp. 38-50, 2018. [10] d. gursoy, “a critical review of determinants of information search behavior and utilization of online reviews in decision making process,” international journal of hospitality management, vol. 76, pp. 53-60, 2019. [11] j. s. liu and l. y. lu, “an integrated approach for main path analysis: development of the hirsch index as an example,” journal of the american society for information science and technology, vol. 63, no. 3, pp. 528-542, 2012. [12] k. d. brouthers, r. mudambi, and d. m. reeb, “the blockbuster hypothesis: influencing the boundaries of knowledge,” scientometrics, vol. 90, no. 3, pp. 959-982, 2011. [13] y. m. wang, y. s. wang, and y. f. yang, “understanding the determinants of rfid adoption in the manufacturing industry,” technological forecasting and social change, vol. 77, no. 5, pp. 803-815, 2010. [14] m. l. meuter, m. j. bitner, a. l. ostrom, and s. w. brown, “choosing among alternative service delivery modes: an investigation of customer trial of self-service technologies,” journal of marketing, vol. 69, no. 2, pp. 61-83, 2005. [15] s. li, k. srinivasan, and b. sun, “internet auction features as quality signals,” journal of marketing, vol. 73, no. 1, pp. 75-92, january 2009. [16] p. a. pavlou and m. fygenson, “understanding and predicting electronic commerce adoption: an extension of the theory of planned behavior,” mis quarterly, vol. 30, no. 1, pp. 115-143, 2006. [17] k. zhu and k. l. kraemer, “post-adoption variations in usage and value of e-business by organizations: cross-country evidence from the retail industry,” information systems research, vol. 16, no. 1, pp. 61-84, 2005. [18] l. d. chen, m. l. gillenson, and d. l. sherrell, “enticing online consumers: an extended technology acceptance perspective,” information & management, vol. 39, no. 8, pp. 705-719, 2002. [19] k. zhu, k. l. kraemer, and s. xu, “the process of innovation assimilation by firms in different countries: a technology diffusion perspective on e-business,” management science, vol. 52, no. 10, pp. 1557-1576, 2006. [20] s. y. x. komiak and i. benbasat, “the effects of personalization and familiarity on trust and adoption of recommendation agents,” mis quarterly, vol. 30, no. 4, pp. 941-960, 2006. [21] d. chatterjee, r. grewal, and v. sambamurthy, “shaping up for e-commerce: institutional enablers of the organizational assimilation of web technologies,” mis quarterly, vol. 26, no. 2, pp. 65-89, 2002. [22] p. l. to, c. c. liao, and t. h. lin, “shopping motivations on internet: a study based on utilitarian and hedonic value,” technovation, vol. 27, no. 12, pp. 774-787, 2007. [23] w. duan, b. gu, and a. b. whinston, “informational cascades and software adoption on the internet: an empirical investigation,” mis quarterly, vol. 33, no. 1, pp. 23-48, 2009. [24] h. barki and a. pinsonneault, “a model of organizational integration, implementation effort, and performance,” organization science, vol. 16, no. 2, pp. 165-179, 2005. [25] s. m. mudambi and d. schuff, “what makes a helpful online review? a study of customer reviews on amazon,” mis quarterly, vol. 34, no. 1, pp. 185-200, 2010. [26] b. w. wirtz, a. pistoia, s. ullrich, and v. gottel, “business models: origin, development and future research perspectives,” long range planning, vol. 49, no. 1, pp. 36-54, 2016. [27] t. oliveira, m. thomas, and m. espadanal, “assessing the determinants of cloud computing adoption: an analysis of the manufacturing and services sectors,” information & management, vol. 51, no. 5, pp. 497-510, 2014. [28] v. zwass, “co-creation: toward a taxonomy and an integrated research perspective,” international journal of electronic commerce, vol. 15, no. 1, pp. 11-48, 2010. [29] y. l. fang, i. qureshi, h. s. sun, p. mccole, e. ramsey, and k. h. lim, “trust, satisfaction, and online repurchase intention: the moderating role of perceived effectiveness of e-commerce institutional mechanisms,” mis quarterly, vol. 38, no. 2, pp. 407-428, 2014. advances in technology innovation, vol. 4, no. 4, 2019, pp. 247-259 259 [30] j. d. wells, j. s. valacich, and t. j. hess, “what signal are you sending? how website quality influences perceptions of product quality and purchase intentions,” mis quarterly, vol. 35, no. 2, pp. 373-396, 2011. [31] s. amaro and p. duarte, “an integrative model of consumers' intentions to purchase travel online,” tourism management, vol. 46, pp. 64-79, 2015. [32] h. c. zheng, d. h. li, and w. h. hou, “task design, motivation, and participation in crowdsourcing contests,” international journal of electronic commerce, vol. 15, no. 4, pp. 57-88, 2011. [33] n. amblee and t. bui, “harnessing the influence of social proof in online shopping: the effect of electronic word of mouth on sales of digital microproducts,” international journal of electronic commerce, vol. 16, no. 2, pp. 91-113, 2011. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation, vol. 8, no. 1, 2023, pp. 38-58 maritime computing transportation, environment, and development: trends of data visualization and computational methodologies thanapong chaichana* college of maritime studies and management, chiang mai university, samut sakhon, thailand school of mechanical, aerospace and automotive engineering, coventry university, coventry, united kingdom received 01 july 2022; received in revised form 21 october 2022; accepted 25 october 2022 doi: https://doi.org/10.46604/aiti.2023.10419 abstract this research aims to characterize the field of maritime computing (mc) transportation, environment, and development. it is the first report to discover how mc domain configurations support management technologies. an aspect of this research is the creation of drivers of ocean-based businesses. systematic search and meta-analysis are employed to classify and define the mc domain. mc developments were first identified in the 1990s, representing maritime development for designing sailboats, submarines, and ship hydrodynamics. the maritime environment is simulated to predict emission reductions, coastal waste particles, renewable energy, and engineer robots to observe the ocean ecosystem. maritime transportation focuses on optimizing ship speed, maneuvering ships, and using liquefied natural gas and submarine pipelines. data trends with machine learning can be obtained by collecting a big data of similar computational results for implementing artificial intelligence strategies. research findings show that modeling is an essential skill set in the 21st century. keywords: maritime computing, management technology, modeling 1. introduction maritime computing (mc) is an interdisciplinary field comprising economics, geography, transportation, logistics, mechanics, computer science, engineering, and technology [1-3]. it is the application of computing techniques to maritime engineering (me). computing techniques are the implemented techniques in computers to solve problems by either step-wise, repeated, or iterative solution methods; also known as in-silico methods. among these techniques, computational fluid dynamics (cfd) and machine learning (ml) are some of the most widely used techniques. cfd is a branch of fluid mechanics that uses numerical analysis and data structures to analyze and solve problems; its concept is to model the problem using geometric approaches to produce visual results [4-7]. ml is a branch of computer science that uses data and algorithms to mimic the way humans learn; its concept is to develop computers to learn from data and then perform learning [8-11]. me is referred to the word ocean engineering in certain academic and professional circles as the field of study dealing with the engineering of boats, ships, submarines, and any other marine vessels, as well as other ocean systems and structures. mc will focus on transportation, environment, and development in the me field. mc transportation is the study of the changes and movement of objects in the sea. mc environment is the study of oceanic issues and environmental simulations. mc development is the study of geometrical simulations and ocean-shaped designs linked to the sea. * corresponding author. e-mail address: thanapong@wavertree.org tel.: +666-5-6728999 advances in technology innovation, vol. 8, no. 1, 2023, pp. 38-58 39 this study is organized as follows. section 2 describes the proposed method, especially presenting research concepts of literature data describing data. section 3 illustrates the results obtained from the algorithmic search. section 4 discusses how the findings will improve future ocean-based businesses. finally, the last section concludes that mc is important in maritime society to accomplish development goals. 2. literature review instead of being an engineer, mc is more like the computationalist and the socialist which uses computers and technology to assist ocean-based businesses at the microscale and macroscale. mc involves a business process management system. currently, several crucial maritime studies employ computation and technology. fig. 1 shows the images of mc and indicates that the relationships between skill sets are mutually linked. the design of numerous ocean-based business processes consists of transportation, logistics, trade, food, environment, data, simulation, and modeling. transportation, logistics, and computing automation were the core of the maritime business proposed by the world maritime university [12]. similarly, bioeconomy involves the conversion of agricultural, organic, and marine resources into materials, energy, fuels, feed, and foods [ 13-14]. (a) basic guidance of mc (b) relationship of mc domains (c) conceptual mc domain fig. 1 the field of maritime computing (mc) mc helps bioeconomy to develop in all aspects. thus, it can provide technological benefits and systems for economic science and bioeconomy [15-16]. economic science, a smaller area of economics, involves rigorous thinking and mathematical applications. unlike the political economy, economic science focuses on the production, distribution, and consumption of goods and services [16]. mc supports economic science through a significant aspect of its modeling skill set as shown in fig. 1(a). it also supports sustainable economic development involved in price fluctuation, banking characteristics, and social development [17-19]. computing is used to translate information, ideas, and practical and mathematical modeling data into digital data. however, visualization techniques are required to determine computational results and process digital data to interpret and verify the outcomes of these results such that they are applicable in real-world conditions [20-21]. mathematical modeling alone is inadequate for explaining the computational consequences [20, 22-24]. a computational research of fluids is noteworthy because fluids are present everywhere and are found in numerous computing applications. therefore, cfd is a preferred concept for ocean-based businesses to classify maritime visualizations and computational methodologies. cfd is an interpreter that provides computational outcomes in terms of data visualizations, including data plots, streamlines, vector changes, contour areas, shading colors, and user-programmable animations [24-26]. advances in technology innovation, vol. 8, no. 1, 2023, pp. 38-58 40 in addition, cfd generates computational results, and these data depend on a study design to solve a specific problem statement. later, visualization methodologies were applied to translate these computational results into graphs, images, and/or animations for understanding the solutions. ml enables computer systems to learn from input data to automatically predict solutions. in this context, the concept of ml and cfds are possible to build big data of similar study designs as input data for ml to predict the computational outcome. the use of computing data involves data collection, which is adequate for forecasting the next state of data-driven applications [27]. massive data collection and automation are usually accomplished through ml. ml can be used to complete routine tasks, simulate virtual contexts, and reduce physical work and manual labor. in the intelligent societies of the 21st century, computing technology has advanced physical labor by inputting human ideas and data into a computer domain. these inputs are then translated into digital data that machines or computers can be used to perform human tasks [28-30]. however, no systematic studies have focused on mc classification. cfd is a key concept in ocean-based businesses that examines maritime visualization and the computational methodologies dedicated to mc transportation, environment, and modeling. the main objective of the present research is to investigate the trends in computer technology and data applications used in mc during the past five decades to help drive and expand ocean studies and support essential skill sets in the 21st century. 3. proposed method mc is generally defined as the analysis of collected material. in this study, a literature review of maritime studies was performed by the preferred reporting items for systematic reviews and meta-analysis (prisma) guidelines [31-32]. a comprehensive electronic search of the literature was conducted using the following databases: sciencedirect, scopus, the iet digital library, ieee-xplore, springer link, web of science, and google scholar. the keywords used to perform the search were mainly included in the form of [cfd+(marine_or_maritime)] [33]. hence, this keyword was preferred. fig. 2 protocol representing research concepts of literature data describing data maritime studies were searched from the 1960s to 2022 in the aforementioned databases and were eligible for inclusion in the present research. articles were included if they were peer-reviewed and published in english. the titles of the articles were analyzed using the and operator in the basic search engine to ensure that the studies met the eligibility criteria before inclusion. in addition, the abstracts of articles were analyzed to determine if the study conditions accurately reflected our study aim and title. the introduction section of each article was scanned for critical thinking. conference abstracts, review articles, advances in technology innovation, vol. 8, no. 1, 2023, pp. 38-58 41 editorial notes, short articles, and other article types were excluded according to the study criteria. finally, full-text original research papers were included to perform the mc classification. fig. 2 shows a pictorial representation of the article included in this study. computer technology advancements began on a massive scale in the past two decades because of the affordable cost of electronics production, the large-scale design of microelectronics, the availability of computer programs and graphic processing unit (gpu) technology, digital literacy, and computing skills [34]. thus, few computational applications in maritime studies would have been available during the 20th century. the three significant key areas found in maritime literature are transportation, environment, and development research. accordingly, keywords, namely “transport,” “environment,” and “modeling,” were used to search for information, along with the aforementioned main keywords (maritime and cfd). these sets of keywords were used to obtain adequate and accurate information to categorize search results into the following phases: the current situation, the prediction of the future situation, and the definition of mc and its basic skillset. 3.1. data classification and critical evaluation strategies the data were collected according to the inclusion criteria described in the previous subsection. the last search was completed on october 02, 2022. each article was manually assessed. an independent assessor (thanapong chaichana) classified and tabulated the data in a microsoft excel spreadsheet for analysis. the data observer (thanapong chaichana) verified the search results of the original mc research, strictly focusing on the algorithmic search for analysis and review. the following characteristics of maritime studies were classified: authorship, year of publication, area of mc research, study design, study purpose, key findings, algorithmic software usage, maritime big-data possibility, artificial intelligence (ai) feasibility, and future direction of research. 3.2. data validation and double-check of results after the search was completed, a double-check procedure was performed to verify the results. first, the presence of both keywords (cfd and marine/maritime) in the titles of maritime articles was confirmed. alternatively, the presence of these keywords was determined in the abstract or introduction section. second, the key areas of mc were classified. the key research areas and keywords used for classification were transport, environment, and modeling. subsequently, all the maritime studies were thoroughly read to ensure that they represented the actual research context in the field of mc. finally, the original research articles were thoroughly evaluated to exclude studies with the same group of authors, the same series of publications, duplicate and similar studies, and articles irrelevant to maritime modeling, environment, and transport research. accordingly, the entire inclusion process, shown in fig. 2, was successfully implemented. 4. results 4.1. search outcome of the maritime literature the electronic search yielded 2,179 articles. after applying the search criteria to screen full-text original research articles, 92 of the 2,179 articles were found to be eligible. these articles were then evaluated to determine whether they focused on three research areas: transportation, environment, and development. a total of 62 articles met the criteria and were reviewed; of these 62 articles, 8 were excluded because they were an extension of conference papers (3 articles), focused on aerospace research (1 article), or had the same name as the first author (4 articles). finally, 54 articles met the research criteria and were included in this meta-analysis. fig. 3 shows the characteristics and key research areas of the 54 articles. there are 14 and 15 articles focused on maritime transportation and the environment, respectively. the remaining 25 articles focused on maritime development. the results advances in technology innovation, vol. 8, no. 1, 2023, pp. 38-58 42 obtained from the algorithmic search offered unique information and more accurate search results when these results are compared to previous studies [31, 35-37]. moreover, a bibliometric search algorithm implemented using scientometric software (e.g., bibexcel [38] or citespace [39]) may reduce the time consumption. however, an accurate search result depends on keywords, boolean conditions, and searching protocol (fig. 2 offered a newly accurate scientometric algorithm). fig. 3 studies focusing on mc transportation, environment, and development 4.2. maritime computing (mc) fig. 4 data trends of mc methodologies, invention of computers, and computational techniques advances in technology innovation, vol. 8, no. 1, 2023, pp. 38-58 43 in the mc research field, interesting findings were observed in the areas of transportation, environment, and development. fig. 4 shows an mc chart with details of its development and data reported over the past five decades. in the current decade, computational results visualized using three-dimensional (3d) and multidimensional computing domains have been reported as normal methodologies for mc. in contrast, in the 1970s and 1980s, mc began with one-dimensional (1d) or two-dimensional (2d) mathematical models, drawings, and domain geometries. technological advancements have led to the visualization and implementation of real-world mc practices. 4.3. mc transportation mc transportation was developed to research the changes and movement of objects in the sea. table 1 presents the categorization of mc transportation. the study design began by simulating the ship speeds. subsequently, a ship maneuvering simulation was performed. liquefied natural gas (lng) is widely used in maritime transportation systems. studies have used lng fuel in transport to model computer-generated lng spills and dispersion processes above the sea surface. studies have also focused on the fuel efficiency and hazard issues of lng. submarine pipelines are used for transporting objects, and they cause damage to submarine landslides and the seabed. commercial software is commonly used for creating algorithms, computations, and simulations as an instant programming tool to perform data visualization, data analysis, and computational methodologies. the computational results of ship simulations could be collected as a maritime databank to form big data, which can be used in the future to plan the ml strategy leading to ai execution. table 1 characteristics of mc transportation study design study purpose key findings algorithmic software usage maritime big-data possibility ai feasibility future direction simulation of 3d flow of maritime lng control valves [40] determine the advanced design of maritime lng control valves with high pressure drop good agreement of computational results with conventional control valves, improvement of flow pattern, reduction of cavitation, and prediction of performance commercial cfd-ace+ n/a n/a n/a simulation of submarine pipelines to examine the interaction of ocean transports with submarine landsides [41] predict submarine pipelines imposed by submarine landslides, a comparison of different shapes (wedge, airfoil, double-ellipse, and arc-angle hexagon) and conventional circular shape revealing the disadvantages of conventional circular pipelines in terms of lift force and drag force when interacting with submarine landsides, a recommendation using other shapes commercial ansys n/a n/a n/a simulation of distribution of lng imposed by air and sea surface temperatures [42] predict the lng dispersion process to spill and smoke clouds to examine sea transport as a major application of maritime fuel good agreement of computational results with experiments; sea surface temperatures impact lng dispersion more than air temperatures do commercial ansys fluent n/a n/a hazard of evaluating lng spill and vapor cloud dispersion advances in technology innovation, vol. 8, no. 1, 2023, pp. 38-58 44 table 1 characteristics of mc transportation (continued) study design study purpose key findings algorithmic software usage maritime big-data possibility ai feasibility future direction simulation of 3d flow of maritime lng control valves [40] determine the advanced design of maritime lng control valves with high pressure drop good agreement of computational results with conventional control valves, improvement of flow pattern, reduction of cavitation, and prediction of performance commercial cfd-ace+ n/a n/a n/a simulation of submarine pipelines to examine the interaction of ocean transports with submarine landsides [41] predict submarine pipelines imposed by submarine landslides, a comparison of different shapes (wedge, airfoil, double-ellipse, and arc-angle hexagon) and conventional circular shape revealing the disadvantages of conventional circular pipelines in terms of lift force and drag force when interacting with submarine landsides, a recommendation using other shapes commercial ansys n/a n/a n/a simulation of distribution of lng imposed by air and sea surface temperatures [42] predict the lng dispersion process to spill and smoke clouds to examine sea transport as a major application of maritime fuel good agreement of computational results with experiments; sea surface temperatures impact lng dispersion more than air temperatures do commercial ansys fluent n/a n/a hazard of evaluating lng spill and vapor cloud dispersion simulation of high-speed catamaran transport to examine fuel economic efficiency [43] predict full-scale resistance values of the ship good agreement of computational results with experiments; the full-scale drag of large catamarans open-source openfoam n/a n/a simulate medium-speed catamarans in shallow water conditions simulation of virtual ship maneuvering using a captive model test [44] determine hydrodynamic coefficients for predicting ship maneuver good agreement of computational results with experiments; primarily designing maritime transport performance commercial star-ccm+ collect computational results of the model ship (dtmb 5512) using the same hydrodynamic model, problems, and initial conditions n/a simulation of 3d hydrodynamics of naca 0012 airfoil (wing) moving above a free surface for super-high-speed ships [45] predict flow field and pressure distribution around the wing on a free surface treating a free surface as a rigid wavy wall, examining the lift/drag ratio involved in a propulsion system n/a collect computational results of the naca 0012 airfoil wing using the same geometry, problems, and initial conditions n/a creation of a new solver for transport equation [46] create a new solver method acceptable results compared to standard solver open-source openfoam n/a n/a n/a advances in technology innovation, vol. 8, no. 1, 2023, pp. 38-58 45 table 1 characteristics of mc transportation (continued) study design study purpose key findings algorithmic software usage maritime big-data possibility ai feasibility future direction prediction of hydrodynamic performance of marine propeller [47] predict the hydrodynamic performance using a neural network a neural network can be trained by cfd data commercial ansys fluent yes, the collection of cfd data yes, ml and cfd data n/a simulation of a boil-off gas generation for lng [48] predict the thermodynamic and hydrodynamic of an lng tank understanding the thermal behavior and characteristics of an lng tank commercial ansys fluent collect the same computing results of an lng tank using the same scale model, problems, and initial conditions n/a simulation of a semi-submersible floating offshore wind turbine [49] analyze the accuracy of the simulation a guideline is provided for the verification and validation of the simulation n/a n/a n/a n/a optimization of a fin-and-tube heat exchanger [50] investigate the optimal tube settings for a fin-and-tube heat exchanger significant improvements in thermal-hydraulic efficiency open-source openfoam n/a n/a n/a prediction of dynamic responses of the ultra-large floating body on maritime airport [51] analyze a case study on an ultra-large floating body under typhoon-wave a reference for the design and construction of maritime airports under typhoon conditions. n/a collect the same computing results of maritime airports using the same scale model, problems, and initial conditions n/a prediction of high-speed craft [52] present geometric deep learning models for engineering design optimizations acceptable results of the deep learning-based surrogate model using cfd data as a ground truth n/a yes, the collection of cfd data yes, ml and cfd data n/a assessment of a marine diesel engine fueled with natural gas [53] calculate the emission and performance of marine diesel engine temperature is a major reason for the increasing use of natural gas and biodiesel content. n/a collect the same computing results of marine diesel engine using the same scale model, problems, and initial conditions n/a fig. 5 shows a bottom view of the flow computation throughout the control valves of maritime lng structures. an example of data visualization in mc transportation revealed velocity streamlines paired with contour plots. among the included studies, those focusing on mc transportation were the smallest in number because an increasing number of ocean studies have recently been conducted. hence, the future direction of research on mc transportation is expected to be economic efficiency and environmental sustainability. fig. 5 mc transportation visualization showing lng control valves [40] advances in technology innovation, vol. 8, no. 1, 2023, pp. 38-58 46 4.4. mc environment the mc environment focuses on environmental simulations and oceanic issues. table 2 lists the collection of mc environments. studies in this area have reported important ocean issues, such as the detection of microparticles of heavy metals in the ocean, the presence of coastal waste particles, underwater robotic observers, renewable energy generated by a tidal stream generator, oxygen dissolved in coastal water, ocean noise reduction, diesel engine emission reduction, and the effect of abalone aquaculture on the ocean. several studies have presented their codes (in-house software) instead of using commercial software for computations, algorithms, and simulations. big data and ai can be used to solve similar computational problems, models, and initial conditions. data obtained from the computing results can be collected if the same geometrical domains are used, such as diesel engines and waste sorting models. table 2 characteristics of the mc environment study design study purpose key findings algorithmic software usage maritime big-data possibility ai feasibility future direction simulation of deployment patterns of tidal stream turbines in the ocean [54] determine single and arrayed tidal stream turbines to generate maritime energy revealing the design of deployment of tidal turbine arrays offering new opportunities to progress renewable energy in-house software n/a n/a n/a simulation of the automatic design process of environmental ocean microfluidic chip [55] detect maritime ecological toxicity of heavy metals in microalgae detect ocean microfluidics-inte grated heavy metals (cu, hg, cd, and zn) using hydraulic analogy commercial cfd-ace+ and orcad n/a the automatic design process of microfluidic dilute network for ocean ecological toxicity assessment routine microalgae bioassays for expanding cell-based screening of environmental health risks simulation of coastal waste particles in a wind-power sorting system [56] determine different airflow rates to classify coastal waste particles into both combustible and noncombustible characteristics identify the arrangement of coastal waste particles to ensure recycling and reduction of coastal waste commercial ansys collect the same computing results of coastal waste particles and airflow rates using the same scale model, problems, and initial conditions n/a simulation of marine diesel engine for catalytic reduction system (crs) [57] predict nox emission reduction using a computing approach and comparison with experimental data revealing the reasonable potential of computational simulation to compute the actual nox reduction rate of crs commercial avl fire collect the same computing results of crs, nozzle model, and nox emission data using the same scale model, problems, and initial conditions n/a simulation of a two-arm marine robotic vehicle to examine hydrodynamic behavior [58] determine the propulsion and manipulation of bioinspired underwater robotic vehicles for observing the marine ecosystem revealing flow development and hydrodynamic force around the bioinspired arms of underwater robotic vehicles in-house software (developed own code) n/a n/a simulate more arm designs and kinematic parameters to optimize the current robotic system simulation of pumped surface water downwelling and dynamic response of oxyflux device (dissolved oxygen in coastal water) [59] determine 1/16 oxyflux model’s dynamic response and pumping performance (estuarine and coastal marine ecosystems) revealing the nonlinear effects of the reduction in the dynamic behavior of oxyflux and the small wave caused by the low-intensity winds of summer (involves predicting anoxia) commercial star-ccm+ n/a n/a improve time-domain computational model and floater shape; extend surge mode (currently have heave mode and pitch mode) and mooring system advances in technology innovation, vol. 8, no. 1, 2023, pp. 38-58 47 table 2 characteristics of the mc environment (continued) study design study purpose key findings algorithmic software usage maritime big-data possibility ai feasibility future direction simulation of submersible abalone cages for marine-deployed aquaculture [60] simulate advanced abalone cages and conventional abalone cages in the exposed ocean environment performing surface and submerged simulations of both advanced and conventional abalone cages; new design of abalone cage commercial flow-3d, fluent, and msc.marc n/a n/a perform economic analysis and risk calculation for newly designed abalone cage simulation of the hydrodynamics of a spherical underwater robot [61] predict the movement of the spherical underwater robot examine hydrodynamic force, velocity, and pressure, basic movement characteristics of robot moving underwater commercial ansys n/a n/a improve the control accuracy of spherical underwater robots simulation of acoustic sounds generated from flows around circular cylinder [62] determine noises occurring while transport moves at increasing speed revealing simulated acoustic sound waves and sound pressure in-house software n/a n/a n/a simulation of floating tidal stream for renewable energy [63] assess floating tidal concepts compared to seabed mounted systems good agreement of computational results with experiments open-source openfoam n/a n/a n/a simulation of a marine environment containing chloride sea salts [64] evaluate a crack growth of chloride-induced stress corrosion cracking in a dry storage system a long-term environmental temperature had an impact on the erosion of long-term storage commercial ansys fluent collect the same computing results of a storage system and marine environment using the same scale model, problems, and initial conditions n/a simulation of underwater gas leakage and dispersion behaviors [65] investigate environmental pollution caused by underwater gas leakage current speed and gas leaking rate mainly affect the underwater gas migration process commercial ansys fluent n/a n/a n/a simulation of environmental loads impact ship maneuverability [66] investigate a real seaway to understand a ship’s maneuvering performance good agreement of computational results with experiments open-source openfoam collect the same computing results of environmental loads and vessel model using the same scale model, problems, and initial conditions prediction of the ship maneuver in waves with different wavelengths and heights simulation of multiphase oil behaviors [67] calculate risk assessment and pollution control of oil spills found that the presence of ice makes the spreading of spilled oil slower n/a n/a n/a n/a simulation of marine cabin's ventilation for reefer containers [68] analyze refrigeration efficiency in a severe marine environment good agreement of computational results with experiments commercial ansys fluent collect the same computing results of reefer containers and the marine environment using the same scale model, problems, and initial conditions n/a fig. 6 shows the tidal stream generators in the natural environment of the ocean. in this case, data visualization in the mc environment was performed using velocity streamlines designed to optimize the arrangement position of tidal stream generators in the ocean to generate renewable energy. future studies can focus on improving the accuracy of the computing results by redesigning the simulations. some included studies focused on mc environments because of the effects of fossil fuels and renewable energy on the global economy and sustainable development goals. advances in technology innovation, vol. 8, no. 1, 2023, pp. 38-58 48 fig. 6 mc environment visualization revealing tidal stream generators in natural flow surroundings [54] 4.5. mc development mc development involves the study of geometrical simulations and ocean-shaped designs linked to the sea. table 3 lists the characteristics of mc development. study designs indicated that maritime development primarily focused on the design and modeling of oceanic objects, such as ships, sailboats, propellers, propeller blades, breakwater slopes, wave gliders, bio-inspired robots, submarines, and deep-water risers. most studies on maritime development have used the knowledge of fluid dynamics and hydrodynamic foundations, as well as the propulsion prediction of objects related to the ocean. open-source and commercial software programs are regularly used to create algorithms, simulations, and computations. the analysis results indicated that maritime big data and ai are likely achievable with similar geometrical domains, initial conditions, and maritime issues. table 3 characteristics of mc development study design study purpose key findings algorithmic software usage maritime big-data possibility ai feasibility future direction simulation of numerous marine hydrodynamic problems [69] develop their own software for naval architecture and ocean engineering visualize marine hydrodynamics open-source openfoam collect similar computational results for ai using the same scale model, problems, and initial conditions efficiency and accuracy of computing solver simulation of sheet cavitation flow around marine propellers [70] predict cavitating flows at low reynolds numbers of the transition-sensitiv e turbulence flow model good agreement of computational results with experiments commercial ansys fluent collect the computational results of the same propeller model using the same flow model, problems, and initial conditions n/a simulation of wave-seabed and breakwater slope [71] predict the impact of wave-seabed on breakwater design increment of the slope of a breakwater altered seabed and liquefaction open-source openfoam, commercial comsol and matlab collect similar wave-seabed and breakwater designs using the same scale model, problems, and initial conditions n/a simulation of optimization of marine contra-rotating propeller [72] determine the best hydrodynamic performance of marine contra-rotating propeller developing a new optimization method for maritime propeller commercial ansys fluent collect the computational results of the same propeller model using the same scale model, problems, and initial conditions efficiency improvement of computing results considering time consumption and technological cost simulation of a wave glider in a head sea [73] predict the dynamic performance of a wave glider in a head sea dependence of propulsion efficiency on the surge of surface boats and passive eccentric rotation of hydrofoil commercial fine/marine and star-ccm+ collect the computational results of the same simulation and wave glider model using the same scale model, problems, and initial conditions improvement of propulsion efficiency, optimum simulation, and design of hydrofoil in different head sea conditions advances in technology innovation, vol. 8, no. 1, 2023, pp. 38-58 49 table 3 characteristics of mc development (continued) study design study purpose key findings algorithmic software usage maritime big-data possibility ai feasibility future direction simulation of the ship with two flapping foils [74] determine the best hydrodynamic performance of a ship with two alike penguin flippers the function of strouhal number, efficiency, and force coefficients of flapping foils commercial ansys fluent collect similar computational results of ship model and function of flapping foils using the same scale model, problems, and initial conditions n/a simulation of deep-water marine riser [75] determine vortex-induced vibration and fatigue damage caused by a very long marine riser good agreement of computational results with experiments n/a collect similar computational results of the same marine riser model using the same scale model, problems, and initial conditions n/a simulation of a four-bladed marine propeller [76] determine the best performance of marine propellers good agreement of computational results with previous results commercial ansys fluent collect similar computational results of the same marine propeller using the same scale model, problems, and initial conditions improvement of velocity and rotational speed simulation of suboff model maneuver [77] determine the flow change around the stern and entire suboff for maneuverability good agreement of yawing moment and yawing force with experimental data commercial ansys collect similar computational results of the suboff model using the same scale model, problems, and initial conditions n/a simulation of bio-inspired marine propulsor by imitating uniform fish fin [78] optimize the movement of maritime fish fin propulsor lunate-shaped fish fin demonstrates the highest efficiency improvement commercial ansys fluent n/a n/a efficiency improvement of computing results by modifying and redesigning the study model simulation of a five-bladed metal marine propeller [79] determine hydrodynamic performance with a change in marine propeller geometry the deformation of the marine propeller causes a negligible change in hydrodynamic performance commercial ansys fluent collect similar computational results of the same marine propeller using the same scale model, problems, and initial conditions analysis of a composite propeller and a comparison with the current result simulation of a composite marine propeller [80] optimize the hydroelasticity performance of the composite marine propeller optimizing the design of the composite marine propeller can reduce the vibratory hub loads commercial ansys collect similar computational results of the same marine propeller using the same scale model, problems, and initial conditions n/a simulation of marine propeller and suboff submarine [81] predicting that the propeller makes submarine underwater noise reduction of submarine underwater noise requires the control of propeller thrust excitation. n/a n/a n/a n/a simulation of fluid displacement under the bow of a ship model [82] predict the breaking wave and flow patterns at the bow of the ship model visualizing the breaking bow wave and wave patterns at the bow of the ship model in-house software (developed own code) n/a n/a examining the interaction of bow wave with shoulder wave for a comprehensive understanding of complex phenomena simulation of novel composite tube (toting wheel) marine current turbines under free flow conditions [83] predict ocean current energy generation from single and arrayed marine current turbines with ten tubes revealing significant power extraction of ocean current energy from arrays of marine current turbines with ten tubes commercial ansys fluent n/a n/a refinement of simulation system, such as changing wheel design (amend blade) to extract more ocean current energy advances in technology innovation, vol. 8, no. 1, 2023, pp. 38-58 50 table 3 characteristics of mc development (continued) study design study purpose key findings algorithmic software usage maritime big-data possibility ai feasibility future direction study of flow around three commercial ships (container ship and two crude-oil ships having bow and stern bulbs) [84] determine flow characteristics around the three commercial ships, a study of ship hydrodynamics measuring flow wave patterns and velocity components depending on the speed of the ship n/a n/a n/a n/a simulation of the unsteady motion of a ship in waves [85] predict the performance of the ship in waves reveal the viability of computing ship motion with two degrees of freedom in-house software n/a n/a the realistic computational approach to simulate six-degrees-of-fre edom motion with nonlinear phenomena simulation of flow around a ship model [86] calculate viscous flow passing through the ship hull with and without a free surface computational approach by comparing computing results with measurement data in-house software n/a n/a n/a simulation for the design of a sailboat [87] compute the design of hull-form development of the sailboat reveal the prediction of both steady and unsteady sailing performances of boats in-house software n/a n/a improve simulation simulation of an x-plane submarine [88] compute the maneuvering coefficients of the submarine good agreement of computational results with experiments snufoam based on open-source openfoam collect similar computational results of x-plane submarine using the same scale model, problems, and initial conditions n/a simulation of the planing hulls [89] analyze complex hydrodynamics problems of the planing hulls cfd automated workflows can be used to study the planing hull hydrodynamics open-source openfoam n/a n/a n/a simulation of ship model maneuvers for autonomous vessels [90] compute the dynamic maneuvers of a container ship with self-propelled free running good agreement of computational results with experiments commercial star-ccm+ collect similar computational results of the same marine propeller using the same scale model, problems, and initial conditions validate and implement a more sophisticated propeller model simulation of ship navigation [91] analyze the energy efficiency of ships a reference for performance monitoring and maintenance prediction of international maritime affairs n/a yes, the collection of cfd big data using the same scale model, problems, and initial conditions collect the accumulated navigation mileage of every ship simulation of blade shape optimization [92] optimize the shape of the blades of the savonius wind turbine the optimized blade performed better than a classical semicircular blade commercial ansys collect similar computational results of the blade model using the same scale model, problems, and initial conditions improvement of simulation and blade shapes simulation of towing tank [93] investigate the effect of muddy seabeds on marine vessels accuracy of the bingham model for marine vessels sailing through fluid mud refresco cfd code n/a n/a n/a advances in technology innovation, vol. 8, no. 1, 2023, pp. 38-58 51 fig. 7 shows a study of ship hydrodynamics to model a ship maneuverer moving in the ocean. this research illustrates a computing example of maritime development, and the developed model explains virtual ship hydrodynamics. future studies can focus on improving the precision and productivity of the computing results, reducing computational costs, improving simulations, and using future technological advancements. most included studies focused on mc development because of the relationship between modeling and design in the field of mc. fig. 7 mc development visualization illustrating the motion of a ship while it turns on the sea surface [69] 5. discussion in the present research, articles that focused on mc transportation, environment, and development were analyzed. systematic reviews and metadata methodologies were used to search and analyze original full-text research articles on maritime literature retrieved from online digital databases. the field of mc focuses mainly on transportation, environment, and development. these mc areas are key inputs for economic benefits, demographic trends, security factors, increasing seaborne and trade demands, skilled workers, automation and technology, and sustainable development goals [12]. this research majorly contributes to the literature in several ways: (1) this study expands research on the mc perspective by providing evidence showing how the key classification of the mc domain promoted ocean-based businesses. (2) it is the first report to discover how mc domain configurations support management technologies. since ocean-based businesses in several countries are a major economic contribution to the world economy through research, business, trade, and innovation, e.g., great britain, the united states, taiwan, and china. (3) the results of the current study provided the history of using computers for ocean-based businesses through the mc domain and assisted in planning future business, research, and innovation. this finding represented an opportunity to use the computational results of the same study designs to build big data and intelligent computation about mc data trends. this study focuses on ocean-based businesses, particularly in the area of maritime studies, to obtain insights into the field of mc. the mc literature contains cfd involved in naval architecture, marine engineering, computer technology, visualization, and computational methodologies. in contrast, other maritime applications (e.g., shipping, freight, airlines, air transport, fisheries, finance, and investment) and other computational studies (those based on blockchain, the internet of things, social networks, and digital marketing) are excluded. consequently, the main aim of mc is to support economic science and bioeconomy. as shown in fig. 4 mc began with the mathematical matrices based on numerical computation in 1990. traditional computing methodologies are typically limited to the invention of computer technology to visualize data and findings. the computational results were mostly presented as 2d contours, line plots, and scatter plots [94-95]. prior to that, computer technology was evolving, focusing on the improvement of microprocessor and semiconductor technologies, and industrialized advances in technology innovation, vol. 8, no. 1, 2023, pp. 38-58 52 [34]. therefore, mc was absent during the 1980s and 1970s (fig. 4). when gpu technology was considerably improved in the 2010s [96], studies on mc began to be increasingly conducted (fig. 4); these findings are presented in fig. 3 (from 2012 to 2019, the number of first authorships increased). an analysis of the original maritime literature indicates that computing and technology are paramount for education, new skill sets, and practical work. maritime development reveals modeling is an essential ground skill set of all other mc areas (fig. 4, development is highlighted as a root). simultaneously, maritime transportation and environment (fig. 4) were in line with the sustainable development goals [97-98]. however, maritime transportation studies have provided benefits not only for increasing trade opportunities, but also improving economics and supply chain management [12, 99-100]. thus, mc transportation is a crucial economic factor. further maritime trends depend on clean energy, renewable energy, environmental recovery, trade, and real-time digital efficiency cost management [101-103]. mc development demonstrates modeling literacy and skills enabled apprentices, learners, and researchers to translate non-time-dependent activities, such as the maritime logistics process of supply chains for scheduling issues, into a time-dependent computer model, depicting a well-structured problem [21, 104-105]. the results shown in tables 1-3 indicate that big data and ai are likely achievable through the data collection of similar computational results for applying the ml approach. the maritime visualization and computational methodologies discovered in this report are used in ocean studies. the maritime visualization outputs of this report in terms of mc transportation, environment, and development are shown in figs. 5, 6, and 7, respectively. moreover, data visualization methodologies are identified, including organizing data into a table, 1d/2d contour plot, 2d cross-sectioned plot, line graph, bar graph, vector plot, streamline plot, streamline contour, streamline surface, vorticity magnitude contour, wave pattern, 3d graphical abstract, 2d geographical map, scatter plot, actual problematic photo compared with computing results, 3d streamlines, 3d particle tracers, 3d cross-sectional views, 3d computer domains, flowchart, and workflow diagrams. mathematical models and techniques for computation used in mc include the navier–stokes equation, fluid-structure interaction, marker-and-cell method, navier–stokes–poisson equation, reynolds-averaged navier–stokes (rans), reynolds stress model, microalgae bioassay, microfluid networks, circuit design, fluid dynamics, hydrodynamics, experimental study, unsteady rans, structural model, wave model, coupled blade element momentum (bem)-cfd model, renormalization group k-epsilon model, 2d schematic diagram, 3d computer design, statistical analysis, error estimation, large-eddy simulation, detached‐eddy simulation, kinematic equations, lattice gas method, lattice boltzmann method (lbm), lattice gas cellular automation, finite difference lbm, arbitrary lagrangian–eulerian method, electronic design automation, and computer-aided design. furthermore, the meshing element approaches comprise both the finite volume approach and the finite element method. in addition, open-source, in-house, and commercial software programs are used in mc to create computations, algorithms, and simulations. however, the open-source software used to develop in-house programs is based on openfoam [106]. the aforementioned mc domain primarily consists of transportation, environment, and development. mc characterizes the computer simulation planning of people’s ocean business processes and tasks. the mc approach generally includes maritime visualization and computational methodologies. mc supports the translation of maritime data and ideal design information into real-world objects, models, and applications [20, 107]. a business process substantially requires a clear goal associated with its producible outputs. hence, these procedures should be simple with tractability and should strongly focus on producing maritime computational results and outputs [21, 108]. mc offers support to economic science and the bioeconomy [15-16, 109-110]. advances in technology innovation, vol. 8, no. 1, 2023, pp. 38-58 53 the analyzed studies published in the years 2020-2022 demonstrated the applications of the proposed concept of ml, cfds, and big data [47, 52, 91]. fig. 4 shows the evolution of computer technology; clearly, the complex problem modeling and computation techniques progressed together with the invention of computers [34]. it seems that the trends of data visualization and computational methodologies will be included soft computing and intelligent computation in the 2020s (fig. 4). furthermore, the global policies on climate change and sustainable development have directly influenced mc research. as a result, the analyzed results in fig. 3 depict the rise of mc transportation and environment significantly in the years 2020-2022. this study has some following limitations: (1) as a comprehensive search was performed in different databases to retrieve full-text original articles, most of the search results did not yield original studies. thus, in some databases, such as google scholar, it is difficult to categorize original research. (2) the sources of digital maritime literature are changing daily owing to the regular publication of new articles, new research submissions, and updated scientific databases. thus, maritime digital databases and literature do not have up-to-date oceanic information. nevertheless, the use of mc began to rise in the 2010s (fig. 4). (3) an automated procedure for document search and analysis is not investigated. hence, a manual analysis is performed to identify the original studies on mc. (4) an automated document search and analysis need to be performed to complement this study. furthermore, no articles from the iet digital library or ieee-xplore databases met the inclusion criteria. additionally, the same articles were indexed in several databases which are the web of science, scopus, sciencedirect, and google scholar. this complicates the final combination of search results. (5) the systematic search is time-consuming, requiring a considerable time to read articles thoroughly to ensure that they met the selection criteria for inclusion in this meta-analysis. 6. conclusion it is the first report to discover how mc domain configurations support management technologies. in this systematic review and meta-analysis, mc studies were identified and reviewed. the areas of mc development, environment, and transportation were successfully categorized. the results show that mc can provide new oceanic knowledge and education to support economic science and bioeconomy. from the analysis, the following observations are made: (1) studies on mc began to increase in the 2010s, while maritime modeling is a key skill set. maritime transportation studies focus on the usage of lng, maneuvering ship’s speed, and submarine pipelines. the maritime environment has been explored for ocean recovery and protection. the trends of maritime data visualization and computation methodologies were observed to be associated with computational coding techniques and the invention of computers. (2) the ml strategy can be applied to the big data collection of similar computing results to implement ai strategies. additionally, mc improvement mainly focuses on promoting mc to the public for an increment of mc research studies. this update will add innovations and technologies; hence it creates an economic impact on ocean-based businesses. (3) future improvements to the mc may involve the use of quantum bits as an alternative to current digital bits for better modeling of specific problems for processor processing. the gap in connecting ml models to cfd results remains; thus, more research is needed to guide the generation of cfd results, and also makes all ml models are interconnected. the simple partial differential equation (pde) of navier–stokes is one of the best methods for hybrid visualization and cfd practices. this hybrid model can be used as a standard guide for individuals to understand how to create specific questions using pde and maritime modeling. therefore, the efficiency of maritime modeling can be improved. advances in technology innovation, vol. 8, no. 1, 2023, pp. 38-58 54 the present report indicates that mc is important in maritime society to accomplish sustainable development goals. based on these primary results, further studies focusing on a case study of each mc classification are required to corroborate the present findings. conflicts of interest the author declares no conflict of interest. references [1] j. b. r. g. souppez, “interdisciplinary pedagogy: a maritime case study,” dialogue: journal of learning and teaching, pp. 37-44, january 2017. [2] m. luo and s. h. shin, “half-century research developments in maritime accidents: future directions,” accident analysis & prevention, vol. 123, pp. 448-460, february 2019. [3] p. hanafizadeh, a. karimi, a. taklifi, and a. hojati, “experimental investigation of two-phase water–oil flow pressure drop in inclined pipes,” experimental thermal and fluid science, vol. 74, pp. 169-180, june 2016. [4] t. chaichana and g. reeve, “computer simulation approach to model virtual geography of seablite source,” 4th international conference on computer communication and the internet (iccci), pp. 174-178, july 2022. [5] t. chaichana, s. wangtueai, s. sookpotharom, d. songtam, and y. chakrabandhu, “computing virtual geography possesses seablite local assets,” international journal on engineering applications (irea), vol. 10, no. 4, pp. 272-278, july 2022. [6] z. sun and t. chaichana, “an investigation of correlation between left coronary bifurcation angle and hemodynamic changes in coronary stenosis by coronary computed tomography angiography-derived computational fluid dynamics,” quantitative imaging in medicine and surgery, vol. 7, no. 5, pp. 537-548, october 2017. [7] z. sun and t. chaichana, “computational fluid dynamic analysis of calcified coronary plaques: correlation between hemodynamic changes and cardiac image analysis based on left coronary bifurcation angle and lumen assessments,” interventional cardiology, vol. 8, no. 6, pp. 713-719, november 2016. [8] t. chaichana, z. sun, m. barrett-baxendale, and a. nagar, “automatic location of blood vessel bifurcations in digital eye fundus images,” proceedings of sixth international conference on soft computing for problem solving, vol. 2, pp. 332-342, 2016. [9] t. chaichana, b. drury, c. s. brennan, and g. reeve, “machine learning predicts future growth of seablite,” unpublished. [10] n. boonnam, t. udomchaipitak, s. puttinaovarat, t. chaichana, v. boonjing, and j. muangprathub, “coral reef bleaching under climate change: prediction modeling and machine learning,” sustainability, vol. 14, no. 10, article no. 6161, may 2022. [11] t. chaichana, s. yoowattana, z. sun, s. tangjitkusolmun, s. sookpotharom, and m. sangworasil, “edge detection of the optic disc in retinal images based on identification of a round shape,” international symposium on communications and information technologies, pp. 670-674, october 2008. [12] world maritime university (wmu), transport 2040: automation, technology, employment the future of work, london: international transport workers’ federation (itf), 2019. https://commons.wmu.se/lib_reports/58 [13] n. befort, “going beyond definitions to understand tensions within the bioeconomy: the contribution of sociotechnical regimes to contested fields,” technological forecasting and social change, vol. 153, article no. 119923, april 2020. [14] s. wydra, “measuring innovation in the bioeconomy – conceptual discussion and empirical experiences,” technology in society, vol. 61, article no. 101242, may 2020. [15] r. kitney, m. adeogun, y. fujishima, á . goñi-moreno, r. johnson, m. maxon, et al., “enabling the advanced bioeconomy through public policy supporting biofoundries and engineering biology,” trends in biotechnology, vol. 37, no. 9, pp. 917-920, september 2019. [16] x. liu, “exploration on relativity of modern economic science: viewpoint based on sustainable development,” china population, resources and environment, vol. 17, no. 1, pp. 11-14, january 2007. [17] h. mohammadi, r. kazemi, h. maghsoudloo, e. mehregan, and a. mashayekhi, “system dynamic approach for analyzing cyclic mechanism in land market and their effect on house market fluctuations,” proceedings of the 28th international conference of the system dynamics society, pp. 1-7, july 2010. [18] m. bolgorian and a. mayeli, “banks’ characteristics, state ownership and vulnerability to sanctions: evidences from iran,” borsa istanbul review, vol. 19, no. 3, pp. 264-272, september 2019. advances in technology innovation, vol. 8, no. 1, 2023, pp. 38-58 55 [19] e. mehregan, “designing a sustainable development model with a dynamic system approach (isfahan province),” neuroquantology, vol. 20, no. 6, pp. 8143-8164, june 2022. [20] t. chaichana, z. sun, a lucey, d. reid, and a. nagar, “the increasing role of use of computer-based methods in disease diagnosis,” digital medicine, vol. 3, no. 4, pp. 150-153, 2017. [21] j. d. sterman, “system dynamics modeling: tools for learning in a complex world,” california management review, vol. 43, no. 4, pp. 8-25, july 2001. [22] k. mcfarlane, t. chaichana, z. sun, j. n. dentith, and p. brown, “femoral neck angle impacts hip disorder and surgical intervention: a patient-specific 3d printed analysis,” proceedings of the 2019 the 3rd international conference on digital technology in education, pp. 155-159, october 2019. [23] t. chaichana, z. sun, and j. jewkes, “impact of plaques in the left coronary artery on wall shear stress and pressure gradient in coronary side branches,” computer methods in biomechanics and biomedical engineering, vol. 17, no. 2, pp. 108-118, 2014. [24] t. chaichana, z. sun, and j. jewkes, “computation of hemodynamics in the left coronary artery with variable angulations,” journal of biomechanics, vol. 44, no. 10, pp. 1869-1878, july 2011. [25] t. chaichana, z. sun, and j. jewkes, “hemodynamic impacts of left coronary stenosis: a patient-specific analysis,” acta of bioengineering and biomechanics, vol. 15, no. 3, pp. 107-112, 2013. [26] t. chaichana, z. sun, and j. jewkes, “computational fluid dynamics analysis of the effect of plaques in the left coronary artery,” computational and mathematical methods in medicine, vol. 2012, article no. 504367, 2012. [27] a. mourad, j. puchinger, and c. chu, “a survey of models and algorithms for optimizing shared mobility,” transportation research part b: methodological, vol. 123, pp. 323-346, may 2019. [28] t. ruiz, l. mars, r. arroyo, and a. serna, “social networks, big data and transport planning,” transportation research procedia, vol. 18, pp. 446-452, 2016. [29] m. gohar, m. muzammal, and a. rahman, “smart tss: defining transportation system behavior using big data analytics in smart cities,” sustainable cities and society, vol. 41, pp. 114-119, august 2018. [30] z. allam and z. a. dhunny, “on big data, artificial intelligence and smart cities,” cities, vol. 89, pp. 80-91, june 2019. [31] z. sun and t. chaichana, “a systematic review of computational fluid dynamics in type b aortic dissection,” international journal of cardiology, vol. 210, pp. 28-31, may 2016. [32] d. moher, l. shamseer, m. clarke, d. ghersi, a. liberati, m. petticrew, et al., “preferred reporting items for systematic review and meta-analysis protocols (prisma-p) 2015 statement,” systematic reviews, vol. 4, article no. 1, 2015. https://doi.org/10.1186/2046-4053-4-1 [33] “cambridge dictionary,” https://dictionary.cambridge.org, may 19, 2022 [34] s. furber, “microprocessors: the engines of the digital age,” proceedings of the royal society a: mathematical, physical and engineering sciences, vol. 473, no. 2199, article no. 20160893, march 2017. [35] m. madhavan, s. wangtueai, m. a. sharafuddin, and t. chaichana, “the precipitative effects of pandemic on open innovation of smes: a scientometrics and systematic review of industry 4.0 and industry 5.0,” journal of open innovation: technology, market, and complexity, vol. 8, no. 3, article no. 152, september 2022. [36] t. chaichana, c. s. brennan, s. osiriphun, p. thongchai, and s. wangtueai, “development of local food growth logistics and economics,” aims agriculture and food, vol. 6, no. 2, pp. 588-602, 2021. [37] y. chakrabandhu, s. chotinun, p. rachtanapun, c. jaisan, s. phongthai, p. klinmalai, et al., “computing survey assessing digital business status: simaon’s pradu hang dam thai native chicken farm,” international conference on inventive computation technologies (icict), pp. 161-165, july 2022. [38] “bibexcel,” https://homepage.univie.ac.at/juan.gorraiz/bibexcel/, may 19, 2019 [39] “citespace,” http://cluster.cis.drexel.edu/~cchen/citespace/, may 19, 2022. [40] y. j. an, b. j. kim, and b. r. shin, “numerical analysis of 3-d flow through lng marine control valves for their advanced design,” journal of mechanical science and technology, vol. 22, no. 10, pp. 1998-2005, october 2008. [41] n. fan, t. k. nian, h. b. jiao, and y. g. jia, “interaction between submarine landslides and suspended pipelines with a streamlined contour,” marine georesources & geotechnology, vol. 36, no. 6, pp. 652-662, 2018. [42] w. c. ikealumba and h. wu, “effect of air and sea surface temperatures on liquefied natural gas dispersion,” energy & fuels, vol. 30, no. 11, pp. 9266-9274, november 2016. [43] m. haase, k. zurcher, g. davidson, j. r. binns, g. thomas, and n. bose, “novel cfd-based full-scale resistance prediction for large medium-speed catamarans,” ocean engineering, vol. 111, pp. 198-208, january 2016. [44] a. hajivand and s. h. mousavizadegan, “virtual simulation of maneuvering captive tests for a surface vessel,” international journal of naval architecture and ocean engineering, vol. 7, no. 5, pp. 848-872, september 2015. [45] s. h. kwag, “lift/drag prediction of 3-dimensional wig moving above free surface,” ksme international journal, vol. 15, no. 3, pp. 384-391, march 2001. advances in technology innovation, vol. 8, no. 1, 2023, pp. 38-58 56 [46] p. peltonen, p. kanninen, e. laurila, and v. vuorinen, “the ghost fluid method for openfoam: a comparative study in marine context,” ocean engineering, vol. 216, article no. 108007, november 2020. [47] m. bakhtiari and h. ghassemi, “cfd data based neural network functions for predicting hydrodynamic performance of a low-pitch marine cycloidal propeller,” applied ocean research, vol. 94, article no. 101981, january 2020. [48] s. wu and y. ju, “numerical study of the boil-off gas (bog) generation characteristics in a type c independent liquefied natural gas (lng) tank under sloshing excitation,” energy, vol. 223, article no. 120001, may 2021. [49] y. wang, h. c. chen, a. koop, and g. vaz, “verification and validation of cfd simulations for semi-submersible floating offshore wind turbine under pitch free-decay motion,” ocean engineering, vol. 242, article no. 109993, december 2021. [50] t. välikangas, m. folkersma, m. d. maso, t. keskitalo, p. peltonen, and v. vuorinen, “parametric cfd study for finding the optimal tube arrangement of a fin-and-tube heat exchanger with plain fins in a marine environment,” applied thermal engineering, vol. 200, article no. 117642, january 2022. [51] t. zhu, s. ke, w. li, j. chen, y. yun, and h. ren, “wrf-cfd/csd analytical method of hydroelastic responses of ultra-large floating body on maritime airport under typhoon-wave-current coupling effect,” ocean engineering, vol. 261, article no. 112022, october 2022. [52] a. abbas, a. rafiee, m. haase, and a. malcolm, “geometrical deep learning for performance prediction of high-speed craft,” ocean engineering, vol. 258, article no. 111716, august 2022. [53] z. zhang, j. lv, w. li, j. long, s. wang, d. tan, et al., “performance and emission evaluation of a marine diesel engine fueled with natural gas ignited by biodiesel-diesel blended fuel,” energy, vol. 256, article no. 124662, october 2022. [54] m. edmunds, r. malki, a. j. williams, i. masters, and t. n. croft, “aspects of tidal stream turbine modelling in the natural environment using a coupled bem–cfd model,” international journal of marine energy, vol. 7, pp. 20-42, september 2014. [55] b. han, g. zheng, j. wei, y. yang, l. lu, q. zhang, et al., “computer-aided design of microfluidic resistive network using circuit partition and cfd-based optimization and application in microalgae assessment for marine ecological toxicity,” bioprocess and biosystems engineering, vol. 42, no. 5, pp. 785-797, may 2019. [56] j. yoon, d. y. kim, and d. kim, “a numerical study on the behavior of coastal waste particles in a wind-power sorting system for renewable fuel production,” waste management, vol. 80, pp. 387-396, october 2018. [57] c. lee, “numerical and experimental investigation of evaporation and mixture uniformity of urea–water solution in selective catalytic reduction system,” transportation research part d: transport and environment, vol. 60, pp. 210-224, may 2018. [58] a. kazakidi, d. p. tsakiris, and j. a. ekaterinaris, “impact of arm morphology on the hydrodynamic behavior of a two-arm robotic marine vehicle,” ifac-papersonline, vol. 50, no. 1, pp. 2304-2309, july 2017. [59] a. antonini, a. lamberti, r. archetti, and a. m. miquel, “cfd investigations of oxyflux device, an innovative wave pump technology for artificial downwelling of surface water,” applied ocean research, vol. 61, pp. 16-31, december 2016. [60] t. kim, j. lee, d. w. fredriksson, j. decew, a. drach, and k. moon, “engineering analysis of a submersible abalone aquaculture cage system for deployment in exposed marine environments,” aquacultural engineering, vol. 63, pp. 72-88, december 2014. [61] c. yue, s. guo, and l. shi, “hydrodynamic analysis of the spherical underwater robot sur-ii,” international journal of advanced robotic systems, vol. 10, no. 5, 2013. https://doi.org/10.5772/56524 [62] h. k. kang, k. d. ro, m. tsutahara, and y. h. lee, “numerical prediction of acoustic sounds occurring by the flow around a circular cylinder,” ksme international journal, vol. 17, no. 8, pp. 1219-1225, august 2003. [63] s. a. brown, e. j. ransley, s. zheng, n. xie, b. howey, and d. m. greaves, “development of a fully nonlinear, coupled numerical model for assessment of floating tidal stream concepts,” ocean engineering, vol. 218, article no. 108253, december 2020. [64] w. y. wang, y. s. tseng, and t. k. yeh, “evaluation of crack growth of chloride-induced stress corrosion cracking in dry storage system under different environmental conditions,” progress in nuclear energy, vol. 130, article no. 103534, december 2020. [65] y. sun, x. cao, f. liang, and j. bian, “investigation on underwater gas leakage and dispersion behaviors based on coupled eulerian-lagrangian cfd model,” process safety and environmental protection, vol. 136, pp. 268-279, april 2020. [66] d. kim, s. song, and t. tezdogan, “free running cfd simulations to investigate ship manoeuvrability in waves,” ocean engineering, vol. 236, article no. 109567, september 2021. [67] m. raznahan, s. s. li, z. wang, m. boufadel, x. geng, and c. an, “numerical simulation of multiphase oil behaviors in ice-covered nearshore water,” journal of contaminant hydrology, vol. 251, article no. 104069, december 2022. advances in technology innovation, vol. 8, no. 1, 2023, pp. 38-58 57 [68] k. ankang, l. fuliang, g. jiandou, d. bin, c. haofeng, and c. a. o. dan, “numerical simulation and experimental study on marine cabin's ventilation for reefer containers,” international journal of refrigeration, vol. 133, pp. 226-234, january 2022. [69] j. h. wang, w. w. zhao, and d. c. wan, “development of naoe-foam-sjtu solver based on openfoam for marine hydrodynamics,” journal of hydrodynamics, vol. 31, no. 1, pp. 1-20, february 2019. [70] m. m. helal, t. m. ahmed, a. a. banawan, and m. a. kotb, “numerical prediction of sheet cavitation on marine propellers using cfd simulation with transition-sensitive turbulence model,” alexandria engineering journal, vol. 57, no. 4, pp. 3805-3815, december 2018. [71] c. liao, d. tong, d. s. jeng, and h. zhao, “numerical study for wave-induced oscillatory pore pressures and liquefaction around impermeable slope breakwater heads,” ocean engineering, vol. 157, pp. 364-375, june 2018. [72] n. m. nouri, s. mohammadi, and m. zarezadeh, “optimization of a marine contra-rotating propellers set,” ocean engineering, vol. 167, pp. 397-404, november 2018. [73] f. yang, w. shi, x. zhou, b. guo, and d. wang, “numerical investigation of a wave glider in head seas,” ocean engineering, vol. 164, pp. 127-138, september 2018. [74] n. p. b. mannam, p. krishnankutty, h. vijayakumaran, and r. c. sunny, “experimental and numerical study of penguin mode flapping foil propulsion system for ships,” journal of bionic engineering, vol. 14, no. 4, pp. 770-780, december 2017. [75] c. kamble and h. c. chen, “cfd prediction of vortex induced vibrations and fatigue assessment for deepwater marine risers,” ocean systems engineering, vol. 6, no. 4, pp. 325-344, december 2016. [76] m. p. kishore, “numerical investigation for cfd simulation of open water characteristics and cavitation inception of marine propeller blade,” journal of maritime research, vol. 13, no. 1, pp. 73-78, january 2016. [77] x. wu, y. wang, c. huang, z. hu, and r. yi, “an effective cfd approach for marine-vehicle maneuvering simulation based on the hybrid reference frames method,” ocean engineering, vol. 109, pp. 83-92, november 2015. [78] m. i. lamas galdo, c. g. rodríguez vidal, and j. d. rodríguez garcía, “optimization of the efficiency of a biomimetic marine propulsor using cfd,” ingeniería e investigación, vol. 34, no. 1, pp. 17-21, april 2014. [79] h. n. das, p. veerabhadra, c. suryanarayana, and s. kapuria, “effect of structural deformation on performance of marine propeller,” journal of maritime research, vol. 10, no. 3, pp. 47-50, december 2013. [80] x. d. he, y. hong, and r. g. wang, “hydroelastic optimisation of a composite marine propeller in a non-uniform wake,” ocean engineering, vol. 39, pp. 14-23, january 2012. [81] y. s. wei, y. s. wang, s. p. chang, and j. fu, “numerical prediction of propeller excited acoustic response of submarine structure based on cfd, fem and bem,” journal of hydrodynamics, ser. b, vol. 24, no. 2, pp. 207-216, april 2012. [82] a. olivieri, f. pistani, and a. d. mascio, “breaking wave at the bow of a fast displacement ship model,” journal of marine science and technology, vol. 8, no. 2, pp. 68-75, october 2003. [83] j. wang, b. gower, and n. müller, “a case study of a novel ducted composite material marine current turbine,” open engineering, vol. 3, no. 1, pp. 80-86, march 2013. [84] w. j. kim, s. h. van, and d. h. kim, “measurement of flows around modern commercial ship models,” experiments in fluids, vol. 31, no. 5, pp. 567-578, november 2001. [85] y. sato, h. miyata, and t. sato, “cfd simulation of 3-dimensional motion of a ship in waves: application to an advancing ship in regular heading waves,” journal of marine science and technology, vol. 4, no. 3, pp. 108-116, december 1999. [86] s. shiotani and y. kodama, “numerical analysis on free surface waves and stern viscous flow of a ship model,” journal of marine science and technology, vol. 3, no. 3, pp. 130-144, september 1998. [87] h. miyata, h. akimoto, and f. hiroshima, “cfd performance prediction simulation for hull-form design of sailing boats,” journal of marine science and technology, vol. 2, no. 4, pp. 257-267, december 1997. [88] y. j. cho, w. seok, k. h. cheon, and s. h. rhee, “maneuvering simulation of an x-plane submarine using computational fluid dynamics,” international journal of naval architecture and ocean engineering, vol. 12, pp. 843-855, 2020. [89] r. ponzini, f. salvadore, e. begovic, and c. bertorello, “automatic cfd analysis of planing hulls by means of a new web-based application: usage, experimental data comparison and opportunities,” ocean engineering, vol. 210, article no. 107387, august 2020. [90] y. jin, l. j. yiew, y. zheng, a. r. magee, j. duffy, and s. chai, “dynamic manoeuvres of kcs with cfd free-running computation and system-based modelling,” ocean engineering, vol. 241, article no. 110043, december 2021. [91] h. j. shaw and c. k. lin, “marine big data analysis of ships for the energy efficiency changes of the hull and maintenance evaluation based on the iso 19030 standard,” ocean engineering, vol. 232, article no. 108953, july 2021. advances in technology innovation, vol. 8, no. 1, 2023, pp. 38-58 58 [92] h. xia, s. zhang, r. jia, h. qiu, and s. xu, “blade shape optimization of savonius wind turbine using radial based function model and marine predator algorithm,” energy reports, vol. 8, pp. 12366-12378, november 2022. [93] s. lovato, a. kirichek, s. l. toxopeus, j. w. settels, and g. h. keetels, “validation of the resistance of a plate moving through mud: cfd modelling and towing tank experiments,” ocean engineering, vol. 258, article no. 111632, august 2022. [94] s. shiotani and y. kodama, “numerical simulation of viscous flows with free-surface around a ship model,” computational mechanics ’95, pp. 838-843, 1995. https://doi.org/10.1007/978-3-642-79654-8_136 [95] t. chaichana, “analysis of computing progress in maritime studies,” proceedings of the 2019 the 3rd international conference on digital technology in education, pp. 150-154, october 2019. [96] c. cook, h. zhao, t. sato, m. hiromoto, and s. x. d. tan, “gpu-based ising computing for solving max-cut combinatorial optimization problems,” integration, vol. 69, pp. 335-344, november 2019. [97] k. l. nash, j. l. blythe, c. cvitanovic, e. a. fulton, b. s. halpern, e. j. milner-gulland, et al., “to achieve a sustainable blue future, progress assessments must include interdependencies between the sustainable development goals,” one earth, vol. 2, no. 2, pp. 161-173, february 2020. [98] p. centobelli, r. cerchione, and e. esposito, “pursuing supply chain sustainable development goals through the adoption of green practices and enabling technologies: a cross-country analysis of lsps,” technological forecasting and social change, vol. 153, article no. 119920, april 2020. [99] s. lan, c. yang, and g. q. huang, “data analysis for metropolitan economic and logistics development,” advanced engineering informatics, vol. 32, pp. 66-76, april 2017. [100] h. h. lean, w. huang, and j. hong, “logistics and economic development: experience from china,” transport policy, vol. 32, pp. 96-104, march 2014. [101] v. w. b. martins, i. s. rampasso, r. anholon, o. l. g. quelhas, and w. l. filho, “knowledge management in the context of sustainability: literature review and opportunities for future research,” journal of cleaner production, vol. 229, pp. 489-500, august 2019. [102] j. webb, “the future of transport: literature review and overview,” economic analysis and policy, vol. 61, pp. 1-6, march 2019. [103] j. l. aleixandre-tudó, l. castelló-cogollos, j. l. aleixandre, and r. aleixandre-benavent, “renewable energies: worldwide trends in research, funding and international collaboration,” renewable energy, vol. 139, pp. 268-278, august 2019. [104] k. gregor, i. danihelka, a. graves, d. j. rezende, and d. wierstra, “draw: a recurrent neural network for image generation,” proceedings of the 32nd international conference on machine learning, vol. 37, pp. 1462-1471, july 2015. [105] k. laxman, “a conceptual framework mapping the application of information search strategies to well and ill-structured problem solving,” computers & education, vol. 55, no. 2, pp. 513-526, september 2010. [106] h. g. weller, g. tabor, h. jasak, and c. fureby, “a tensorial approach to computational continuum mechanics using object-oriented techniques,” computers in physics, vol. 12, no. 6, pp. 620-631, november 1998. [107] j. e. van aken and h. berends, problem solving in organizations: a methodological handbook for business and management students, 3rd ed., cambridge: cambridge university press, 2018. [108] m. e. porter, competitive strategy: techniques for analyzing industries and competitors, new york: free press, 1980. [109] f. d. vivien, m. nieddu, n. befort, r. debref, and m. giampietro, “the hijacking of the bioeconomy,” ecological economics, vol. 159, pp. 189-197, may 2019. [110] t. düppe, “border cases between autonomy and relevance: economic sciences in berlin—a natural experiment,” studies in history and philosophy of science part a, vol. 51, pp. 22-32, june 2015. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation, vol. 5, no. 4, 2020, pp. 216-229 application-based online traffic classification with deep learning models on sdn networks lin-huang chang 1,* , tsung-han lee 1 , hung-chi chu 2 , cheng-wei su 1 1 department of computer science, national taichung university of education, taichung, taiwan 2 department of information and communication engineering, chaoyang university of technology, taiwan received 19 december 2019; received in revised form 10 march 2020; accepted 03 july 2020 doi: https://doi.org/10.46604/aiti.2020.4286 abstract the traffic classification based on the network applications is one important issue for network management. in this paper, we propose an application-based online and offline traffic classification, based on deep learning mechanisms, over software-defined network (sdn) testbed. the designed deep learning model, resigned in the sdn controller, consists of multilayer perceptron (mlp), convolutional neural network (cnn), and stacked auto-encoder (sae), in the sdn testbed. we employ an open network traffic dataset with seven most popular applications as the deep learning training and testing datasets. by using the tcpreplay tool, the dataset traffic samples are re-produced and analyzed in our sdn testbed to emulate the online traffic service. the performance analyses, in terms of accuracy, precision, recall, and f1 indicators, are conducted and compared with three deep learning models. keywords: software defined network, network traffic classification, deep learning, tensorflow 1. introduction artificial intelligence (ai) has been one of the hottest topics currently. machine learning (ml), which is a subset of ai and adjusts itself in response to the data it is exposed to, will enable the machines to improve at the task with experience. among different machine learning schemes, shallow neural networks, such as backpropagation neural network, bayesian network, support vector machine (svm), and c4.5 decision tree, are commonly used to build traffic classifiers for network services in many machine learning-based classification schemes. due to the continuous expansion of network and considerable deployment of the internet of things (iots), great amount of data and information have been collected and consequently promoted the arrival of the big data era. furthermore, with the significant promotion in processing speed of computing hardware such as graphics processing unit (gpu) and tensor processing unit (tpu), the deep learning (dl) using deep neural networks, which provide higher classification accuracy than the shallow neural networks, have been the major focus on ml. the promising ml techniques, such as deep neural networks provides a good opportunity and key to the data-driven machine learning algorithms in the network field. on the other hand, more and more application services have emerged over the internet. the network management issues, including quality of service (qos) setting, network policy, network security as well as intrusion detection, majorly depend on the accuracy of network traffic classification for applications. the emerged applications or services may not be identified by * corresponding author. e-mail address: lchang@mail.ntcu.edu.tw tel.: +886-4-22183812; fax: +886-4-22183580 advances in technology innovation, vol. 5, no. 4, 2020, pp. 216-229 217 the traditional schemes, such as simply checking source/destination ip addresses in the network layer or source/destination port numbers as well as protocols in the transport layer, because they might pass through different network address translation (nat) or virtual private network (vpn), or they may not utilize the standard port numbers or fields. therefore, to provide proper qos for a specific application, it might need to manually verify or identify the network application packets. these processes will significantly reduce the efficiency of network traffic processing and consequently increase the overall delay and processing loading. furthermore, the development of software defined network (sdn) has made network management more effective and efficient [1-2]. unlike traditional networks, sdn separates the network layer into a data plane and a control plane which could be a logical control unit and programmable. the flexibility and programmability of the sdn architecture design make the network routing and management easier and more efficient, especially for the design and handling of network classification which enables the optimal network solutions, such as configuration and resource allocation. the centralized sdn controller is able to collect various real-time network data due to its global network view. the ai techniques can then be applied to the sdn networks by employing network optimization and data analysis based on the real-time and historical sdn network data. therefore in this paper, we propose an application-based online and offline traffic classification with deep learning models on sdn networks. we design the deep learning models resigned in the sdn controller. the sdn controller establishes the match fields of the flow entry and sends them to the open virtual switch (ovs). it also extracts traffic statistics data from the ovs switch. the server ip addresses and transport port numbers of each flow plus the statistics data are designed to be the input features for the deep learning models. the remaining sections of this paper are organized as follows. we describe the protocols, standards, tools, and related research reviews in section 2. section 3 addresses the experimental settings and model design. in section 4, we present the experimental result and performance analysis in this paper followed by the conclusion and future works in section 5. 2. related work we first review some protocols, standards as well as related tools used in this paper. we also survey some literature related to this research. the following subsections will address the brief discussion of each subject. 2.1. deep learning review dl [3], rather than a task-specific algorithm in the ml series, is a broader representation of learning data. the dl aspect includes supervised, semi-supervised, or unsupervised learning. dl can be applied for classification, clustering, and regression. with classification, dl is able to learn the correlation between data and labels, which is known as supervised learning. for some applications, it is not required or unable to know the labels of the data in advance. learning without labels is called unsupervised learning. with clustering, dl does not require labels to detect similarities among groups of data. with regression, the dl is exposed to enough of the right data. it is able to establish correlations between present events and future events. this predictive analytics is different from the classification which might be called a static prediction. given a time series, dl can run regression by reading lots of data from the past and predict some data likely to occur in the future. dl architectures, including convolutional neural network (cnn), deep neural networks (dnns) and recurrent neural network (rnn), multilayer perceptron (mlp), long short-term memory (lstm), and stacked auto-encoder (sae), have been applied to different fields such as computer vision, audio recognition and natural language processing, etc. some of dl models or implementations have achieved comparable or even superior results as compared to human experts. advances in technology innovation, vol. 5, no. 4, 2020, pp. 216-229 218 in this paper, we apply the tensorflow [4] neural network modules as the dl platform for our application-based online traffic classification. the major programming language and development of tensorflow is python. tensorflow supports application programming interfaces (apis) such as keras [5], with the advantage of easy operation, modular design, and flexible scalability, as the high-level development interface. keras, working in conjunction with the back-end tensorflow engine, provides functional apis for building models, training models, evaluating models, as well as predicting results. 2.2. openflow protocol openflow [6] is one of the most dominated sdn communication protocols. by using the software programming, openflow supports the control and communication of the rules signaling and flows forwarding from the centralized controller southbound to the sdn switch. the northbound apis from the control plan to the management plan, on the other hand, is defined to be used for application service and operation management. several consoles, such as ryu [7], floodlight, and opendaylight, have been developed to support openflow. we apply the ryu controller which supports openflow up to versions 1.5 to define the framework of the sdn networks in this research. ryu, being a component-based sdn framework, provides software components with well-defined api, which is easy for developers to create new control applications and network management. it also supports various communication protocols for network devices management, such as openflow and of-config. 2.3. tcpreplay tools tcpreplay [8], an open-source utility, is used to edit and replay captured network packets. it has been used primarily to replay malicious traffic patterns for network intrusion detection. we apply tcpreplay to replay the network traffics of the real-world dataset and inject them into the sdn network environment in this research. 2.4. iscx dataset the iscx [9] dataset, collected by the canadian cyber security institute, is a network traffic dataset used by many universities, companies, and independent researchers around the world. in this paper, we apply the iscx dataset as our training dataset and offline testing dataset. 2.5. survey on ai applied to sdn and related review the flexibility and programmability of the sdn network make the traffic classification, routing optimization, qos/qoe prediction, resource management, and security easier and more efficient. with the advantage of the sdn global network view, the centralized sdn controller is able to collect various real-time and historical network data from the sdn switches at per port and per-flow granularity levels and consequently, we can apply the ai techniques on the sdn networks by employing network optimization and data analysis. we will provide a brief survey of ai or ml techniques applied to the sdn network, including traffic classification, routing optimization, qos/qoe prediction, and security [10]. first, traffic classification techniques in sdn networks include a port-based approach, deep packet inspection (dpi), and ai or ml [11-12]. the port-based approach will not be effective when most applications recently utilize dynamic ports as tcp or udp port numbers. the dpi approach results in high computational cost or difficult pattern update because all traffic flows need to be checked or updated for the exponential growth of applications. currently, more and more encrypted packets of various applications make the dpi approach even impractical. ai or ml techniques are applied to extract knowledge from the traffic flows in sdn, thus ai or ml-based approaches are able to classify encrypted packets correctly with the lower computational cost. advances in technology innovation, vol. 5, no. 4, 2020, pp. 216-229 219 deep learning models have been applied to network traffic classification currently. the research in [13] introduced a sae-based scheme to classify unencrypted data flows. however, their scheme was only applied to unencrypted traffic and could not be deployed to the encrypted data. besides, the dataset used in their research was not open to the public. the research in [14] proposed a scheme to identify encrypted traffic based on sae and cnn models. the research in [15] proposed three deep learning based models, including mlp, sae, and cnn for traffic classification. they developed their models based on all encrypted streaming packets from the open source dataset. however, their research, including training and prediction models, could not be applied to the real network traffic or emulated online flows because they only conducted the offline dataset analysis. secondly, the routing optimization issues in sdn networks typically employ the shortest path first (spf) algorithm and heuristic algorithms [16]. the spf algorithm is a best-effort routing scheme. it is unable to utilize the best network resources due to the simplicity of the algorithm. the high computational cost however would be the shortcoming of heuristic algorithms [17], such as ant colony optimization algorithm. the introduction of ai or ml to the route optimization in sdn can be considered as a decision-making policy which does not need a complex mathematical model once the model has been trained. the reinforcement learning could be an effective mechanism which may provide a near-optimal routing decision quickly. thirdly, based on qos/qoe prediction, the sdn network operators or service providers are able to provide suitable services to the customers according to their expectations which consequently increase customer satisfaction and service. the authors in [18] have proposed a traditional m/m/1 network model and neural network model to train the model with traffic load and overlay routing policy and then to estimate the network delay. the authors in [19] have focused on the qoe prediction for video streaming service in the sdn network by correlating the qos parameters with the qoe values. they applied a supervised learning model to estimate the mean opinion score (mos) value according to the sdn network parameters, such as delay, jitter, and bandwidth, to adjust video parameters, such as bitrate and resolution, in the sdn controller and consequently improve the qoe of users’ experience. lastly, security is always an important issue in sdn network research and operation. intrusion detection system (ids) is an important mechanism and system for network security which in general consists of signature-based ids and anomaly-based ids according to how they identify network attacks [20]. because the signature-based ids has some shortcomings, such as signature update difficulty and high time consumption for all signatures comparison, most researches focus on the anomaly-based ids. the anomaly-based ids is a flow-based traffic identification which in general inspects the packet header information based on flow-granularity information, while the signature-based ids being a payload-based identification which needs to inspect the whole payload of packets. the supervised learning algorithms of ai or ml techniques are often applied in anomaly-based ids by training a predefined model to identify intrusions and normal flows. again, the global view and programmability of sdn networks will facilitate the ai or ml-based ids because of its simplicity and flexibility of data collection and attack reaction, respectively. with proper feature selection and feature extraction, many studies have been conducted for ai or ml-based ids in sdn networks. the authors in [21] employed the data preprocessing, decision-making subsystem, and response subsystem for their predictive data model as a threat-aware ids in sdn. they used a forward feature selection strategy to select appropriate feature sets in the data preprocessing subsystem, applied the decision tree and random forest algorithms to detect malicious flows in the predictive data modeling subsystem, and then employed the reactive routing to install differe nt flow rules and types based on the detection results in the decision making and response subsystems. the authors in [22] used five flow features, including source/destination ip, source/destination port, and packet length, to predict malicious flows for their proposed hmm-based network ids. advances in technology innovation, vol. 5, no. 4, 2020, pp. 216-229 220 in this paper, we will mainly focus on the deep learning models applied to the traffic classification, especially application-based traffics, in the realm of sdn. the sdn switches send the traffic statistics and packet_in messages to the sdn controller for features extraction, which include server ip and port of each flow plus the statistics data. the sdn controller with deep learning models, including mlp, cnn, and sae models, will conduct the application-based traffic classification for each flow by establishing the match fields of the flow. 3. experimental settings and model design 3.1. experimental setting fig. 1 illustrates the network topology for our proposed sdn network testbed. as shown in fig. 1, we first apply the iscx dataset [9] to emulate the offline and online network traffic by injecting them into tcpreplay to reproduce the client/server communications in the sdn testbed. as mentioned earlier, the iscx dataset, including vpn and non-vpn datasets, was collected by the canadian institute for cybersecurity at the university of new brunswick. both vpn and non-vpn datasets consist of six application services, including skype, facebook, hangouts, youtube, and etc. internet router ova ryu controller tcpreplay host1 tcpreplay host2 fig. 1 network topology of the sdn testbed in this research, we deploy tp-link tl-wr1043nd as the open virtual switch (ovs) based on the openflow protocol for the sdn data plan. the ovs is implemented and installed using openwrt [23]. the ryu controller [7] is used in this paper to define the framework of the sdn networks. as defined by openflow standard, whenever a new coming packet does not match any flow entry in the flow table of ovs, the ovs switch will trigger the packet_in mechanism and send it to the ryu controller as routing setting request. for all packets in the same flow, only one packet_in message is sent to the controller for the same flow. in this research, besides the packet_in message for each flow, we also collect the flow statistics data. for the openflow statistics mechanism, the ovs sends the periodical statistics reports to the controller in response to the controller’s statistics request. the ovs statistics reported in 1sec interval, including flow statistics, table statistics, port statistics, queue statistics, group statistics, and meter statistics, could be used as the deep learning features for traffic flow behavior learning. the flow statistics data combined with the packet_in message will be the learning features of the deep learning models designed in the ryu controller for a specific flow. for each sdn flow, keeping the session connected, the ovs statistics report will be sent to the ryu controller continuously in 1sec interval. this will avoid the stochastic learning results due to inconstant or random statistics reports. because of the traffic flow characteristics of sdn networks, the processing data including packet_in message and statistics data in 1sec interval for deep learning in this research is less than the traditional traffic classification which processes data from all packets. advances in technology innovation, vol. 5, no. 4, 2020, pp. 216-229 221 3.2. data pre-processing in this paper, the dataset was collected in a precise and pre-defined amount and the experiments were conducted for three different deep learning models ten times. most input data are the features with a one-hot input scheme except two features, i.e. byte counts and packet counts, which will be discussed in more detail shortly. since deep learning is a mathematical operation of non-linear regression, it is necessary to limit the eigenvalues input to the classification model to a fixed range. therefore, the keys and values are stored in mapping mode using hash handles, and minmaxscaler in scikit-learn [24] to normalize the values and scale them to the range of 0 to 1. the minmaxscaler is shown in eq. (1), where xi represents the selected feature value, xnom represents the normalized value of one specific feature, and min(x) and max(x) are the minimum and maximum values of the feature, respectively. ( ) ( ) ( ) [0, 1] i x nom x x x min x max min     (1) table 1 shows the iscx dataset used in this experiment, which is injected into tcpreplay to emulate the offline and online network for the client/server communications in the sdn testbed. there are six applications, including facebook, gmail, hangouts, netflix, skype, and youtube, for each vpn and non-vpn dataset in the iscx dataset. for the skype application, the dataset was further divided into two classes, skypeaudio and skypevideo. table 1 dataset used in deep learning model labels applications training data amount validation data amount testing data amount facebookchat facebook 9981 1109 11091 gmailchat gmail hangoutchat hangout netflix netflix skypeaudio skype skypevideo youtube youtube table 2 examples of selected features and labels data # top1 top2 top3 top4 top5 top6 … top305 other byte count packet count labels 1 0 0 0 0 0 0 … 0 0 0.08 0.08 skypeaudio 2 1 0 0 0 0 0 … 0 0 5.87 0.16 skypeaudio 3 0 0 0 0 0 0 … 0 0 0.12 0.12 skypeaudio 4 1 0 0 0 0 0 … 0 0 7.57 0.14 skypeaudio 5 1 0 0 0 0 0 … 0 0 8.12 0.15 skypeaudio 6 0 0 0 0 0 0 … 0 0 0.06 0.06 netflix 7 1 0 0 0 0 0 … 0 0 152.17 0.9 skypevideo … … … … … … … … … … … … … 8 0 0 0 0 0 0 … 0 0 0.09 0.09 hangoutchat 9 0 0 0 0 0 0 … 0 0 0.21 0.1 hangoutchat 10 0 0 0 0 0 0 … 0 0 10.89 0.06 netflix 11 0 0 0 0 0 0 … 0 0 0.06 0.06 youtube 12 0 0 0 0 0 0 … 0 0 0.08 0.08 skypeaudio 13 0 0 0 0 0 0 … 0 0 1.53 6 gmailchat 14 0 0 0 1 0 0 … 0 0 1.36 9 gmailchat 15 0 0 0 0 0 0 … 0 0 0.06 1 netflix 16 0 0 0 0 0 0 … 0 0 2.54 7 facebookchat 17 0 1 0 0 0 0 … 0 0 0.5 1 gmailchat … … … … … … … … … … … … … advances in technology innovation, vol. 5, no. 4, 2020, pp. 216-229 222 so, we will have seven labels for all deep learning models. the total amount of training dataset is 11090, with which 90% of them is used for the deep learning training process and 10% of them is used for the deep learning validation process. the total amount of testing dataset is 11091, which is also injected into tcpreplay and then replayed one by one to emulate the online traffic in the sdn network. because we emulate the real traffic by using the iscx open-source dataset, the dataset packets are handled as tcpreplay pcap packets. the packet_in messages due to the flow mismatch in ovs switch and the periodical statistics data responded from ovs switch to the controller are collected on the ryu controller. the server ip addresses and transport port numbers in each flow and the byte counts and packet counts collected from statistics data are selected as the deep learning features in our designed models. the server ip addresses and transport port numbers are handled using the one-hot input scheme. table 2 shows the examples of the selected features and labels in our designed application-based deep learning models. the top 305 most frequent server ip addresses and transport port numbers are selected as the one-hot input for deep learning models. the other column in one-hot input is used for any other ip or port number, not in the top 305 most frequent lists. 3.3. deep learning model design in this paper, we apply three deep learning models, including cnn, mlp, and sae models, for our application-based offline and online traffic classification. 3.3.1. mlp model input layer ae1 hiden ae1 output output layer 32 units ae1 input ... ... ... ... fig. 2 mlp deep learning in our model the first deep learning model we apply in this research is the multilayer perceptron (mlp) model. the designed mlp deep learning model, shown in fig. 2, employs back-propagating supervised learning techniques in artificial neural networks. it consists of an input layer, three hidden layers followed by a dropout layer, and an output layer. the input layer consists of 305 one-hot inputs representing the most frequent server ip addresses and transport port numbers plus one other column one-hot input representing the less frequent server ip addresses and port numbers from the packet_in headers. besides, two inputs are representing the byte counts and packet counts in every one-second periodical statistics data. each of the three hidden layers is composed of 256 neurons. the dropout layer is used to prevent the overfitting issue. the output layer consists of seven neurons, and the softmax function is applied as a classifier for the mlp deep learning model. 3.3.2. sae model in this research, we also apply the stacked autoencoder (sae) as one of the deep learning training models. the proposed sae architecture, containing one single encoding and decoding, is illustrated in fig. 3. the encoders of sae deep learning model, used for dimension reduction or feature extraction, contains 32 neurons in this paper. advances in technology innovation, vol. 5, no. 4, 2020, pp. 216-229 223 input layer ae1 hiden ae1 output output layer 32 units ae1 input ... ... ... ... fig. 3 sae deep learning in our model 3.3.3. cnn model the third deep learning model we apply in our deep learning training is the cnn model which is a typical deep learning model used for classification. the designed cnn model is specified in fig. 4 which consists of three convolution layers, three max-pooling layers as well as one fully connected layer with relu function acting as an activation function, followed by a dropout layer and an output layer. the convolution layer uses 64 filters to process input data. again, the dropout layer is used to handle the overfitting issues. input layer ... max1 poolconv1 layer max2 poolconv2 layer conv3 layer full connect 64 filter 64 filter 64 filter output layer max3 pool d r o p o u t r e lu 2 pool size 2 pool size 2 pool size ... ... ... ... ... ... fig. 4 cnn deep learning in our model 3.3.4. deep learning model training parameters the training dataset is divided in 9:1 ratio into training and verification processes for all three mlp, ase and cnn models in our deep learning training processes. the training parameters, such as epochs, batch size, learning rate, as well as optimizers, of our deep learning models are tabulated in table 3. table 3 deep learning model training parameters models epochs batch rate optimizers mlp 100 512 0.01 adam sae cnn 4. experimental result and analysis 4.1. offline training result in this paper, we first apply three deep learning models, including mlp, sae, and cnn models, for our application-based offline traffic classification. for the model training of offline deep learning processes, the training history for three different deep learning models is presented in fig. 5. the experimental results of the training history include the accuracy and loss advances in technology innovation, vol. 5, no. 4, 2020, pp. 216-229 224 function evolution for both training and validation processes. the accuracy and loss rate within ten epochs of validation process for cnn deep learning model is about 94.95% and 20.26%, respectively. the cnn deep learning model shows fairly steady convergence within ten epochs. however, the loss function of validation process keeps almost unchanged after ten epochs and even goes little bit higher than that of training process. this could be due to the overfitting issue. it might need further optimization of the cnn parameters to reduce the overfitting issue. fig. 5 offline training results for three different deep learning models the accuracy and loss rate within ten epochs of validation process for mlp deep learning model is about 94.95% and 17.63%, respectively. there are some turbulences in training and validation processes after ten epochs. however, it achieves steady convergence after about 30 epochs. the accuracy and loss rate within ten epochs of the validation process for sae deep learning model is about 94.86% and 19.638%, respectively. unlike mlp model, the sae deep learning model shows fairly good convergence within five to ten epochs. this could be due to the fine-tuning training mechanism of the sae pre-training. however, more training epochs, less than 20 epochs, are enough for them to obtain similar error rates. advances in technology innovation, vol. 5, no. 4, 2020, pp. 216-229 225 table 4 offline training result for different models cnn-1d mlp sae training process accuracy 93.35% 93.21% 93.13% training loss rate 16.77% 16.40% 15.97% validation process accuracy 94.86% 94.59% 94.86% validation loss rate 19.64% 15.69% 19.64% the offline training results for three different deep learning models, mlp, sae, and cnn, are summarized in table 4. more than 93% accuracy is achieved for all three different models. the offline training time, directly measured from the system, for cnn, mlp and sae models are 1,518, 111, 25 sec, respectively. the complexity of ml models, especially for the deep learning models, has been significantly increasing over recent years. this is mainly due to the increasing number of deep learning network layers and the volume of dataset which consequently increases the computational cost corresponding to the execution time required training a deep learning network. fortunately, this computational cost is only effective during offline training phase. in this research with online traffic classification in sdn during testing phase, the execution time to identify an application for each flow is within few seconds which is relatively small. the trained models for online testing can reach a very high confidence on classification within several statistics data which was collected in every second. the detailed online testing results will be discussed in the next sub-section. 4.2. online testing result as mentioned earlier, we have injected the iscx dataset into tcpreplay host and then replayed them one by one according to the same processing time and sequence to emulate the online traffic for the client/server communications in the sdn testbed. the server ip addresses and transport port numbers in each flow and the byte counts and packet counts collected from statistics data are stored and selected in ryu controller for further data pre-processing so that the server ip addresses and transport port numbers can be handled using one-hot input scheme. for online testing, we use other dataset files with the same applications and pre-processing as the training model to match the input size of the training model. in order to response to the unknown applications, we add a server to mirror packet in message in controller. the mirror server will label the online applications or services for predicted result comparison. furthermore, in the case of vast amount of packets or flows injected into the ovs switch, such as abnormal intrusion, the mirror server can play a role of ovs buffering through the dropping of the vast injected packets on ovs switch to avoid the bandwidth exhaustion. because of the limitation of the processing speed of the ryu controller and ovs switch hardware, we might encounter statistics packets drop or tcpreplay packets drop when the testing dataset injected into the ovs switch according to the same processing time and sequence. therefore, the incomplete dataset due to the packets drop might cause the accuracy decline of the online prediction for different deep learning models as compared with the offline cases. in fig. 6, we illustrate the predicted result, in terms of confusion matrix, of application-based online traffic classification with mlp deep learning model. the diagonal values in the matrix correspond to the true positives of the predicted result of facebookchat, gmailchat, handoutschat, netflix, skypeaudio, skypevideo, and youtube categories, respectively. from the illustration shown in fig. 6, the number of false positives between skypeaudio and skypevideo categories is relatively high. this is understandable because the flows from these two categories own the same server ip and transport port features. the only possible features which we can identify could be byte counts. however, the dynamic encoding schemes of the audio and video could dim the byte count feature. advances in technology innovation, vol. 5, no. 4, 2020, pp. 216-229 226 the numbers of false positives between netflix and skypeaudio or between youtube and skypeaudio also show relatively higher ratio. although the server ip addresses are different for these applications, the similar multimedia behavior in terms of byte counts and packet counts features for deep learning will lead the miss-identification of such streaming category to the skypeaudio category. as a simple deep back-propagating learning technique, the mlp deep learning model may suffer from such miss-identification easily. therefore, a relatively higher error rates were observed between skypeaudio and skypevideo or between netflix and skypeaudio applications. however, we still achieve about 87% accuracy on average for application-based online traffic classification with mlp deep learning model. especially, we conduct the application services with very similar characteristics, such as multimedia, audio and video streaming, which could be very difficult for network traffic classification. fig. 6 confusion matrix of mlp online test result fig. 7 confusion matrix of sae online test result the predicted result, in terms of confusion matrix, of application-based online traffic classification with sae deep learning model is illustrated in fig. 7. similarly, the number of false positives between skypeaudio and skypevideo categories is relatively high again for the sae deep learning model. the numbers of false positives between netflix and skypeaudio or between youtube and skypeaudio also shows relatively higher ratio. because the sae deep learning model uses encoder and decoder for learning, the selected features need to be further adjusted to distinguish the similar streaming behavior between skypeaudio and skypevideo categories or among different multimedia streaming properties. fig. 8 confusion matrix of cnn online test result besides, the predicted result, in terms of confusion matrix, of application-based online traffic classification with cnn deep learning model is illustrated in fig. 8. similarly, the numbers of false positives between skypeaudio and skypevideo categories are relatively high for the cnn deep learning model. the numbers of false positives between netflix and advances in technology innovation, vol. 5, no. 4, 2020, pp. 216-229 227 skypeaudio or between youtube and skypeaudio also show relatively higher ratio which is believed due to the similar multimedia behavior in terms of byte counts and packet counts features for deep learning. the average accuracy for application-based online traffic classification with cnn deep learning model is about 87.3%. in this paper, we reduce the problem of multiclass classification to multiple binary classification problems by using one vs rest transformation technique [25]. the one vs rest transformation technique will treat a single classifier per class with the predicted labels of that class as positive samples and all other predicted labels as negatives. thus, four indicators, accuracy, precision, recall, and f1, were calculated using the confusion matrix to evaluate the performance of application-based online traffic classification with different deep learning models. the online testing result with four indicators in our sdn testbed is shown in table 5. table 5 online testing result for different models accuracy precision recall f1-score cnn 87.208% 85.142% 84.428% 84.857% mlp 87.167% 84.428% 87.714% 84.714% sae 87.079% 85.142% 84.285% 84.714% based on the experimental results, the average accuracies for different deep learning models are about the same, 87%, which is smaller than our offline training result, above 93%. due to the hardware limitation, we might encounter possible statistics packets drop or tcpreplay packets drop when the testing dataset injected into the ovs switch according to the same processing time and sequence. consequently, the incomplete dataset due to the packets drop might cause the accuracy decline of the online prediction for different deep learning models as compared with the offline cases. on the other hand, the increasing number and variation of application services might increase the difficulty on application-based online traffic classification on sdn networks with deep learning models. although the numbers of false positives between skypeaudio and skypevideo categories are relatively high and the numbers of false positives between netflix and skypeaudio or between youtube and skypeaudio also show relatively higher ratio for all deep learning models, the values of precision, recall and f1-score are all above 84% and quite close to the accuracy value. as we commented, the application services conducted in this paper with very similar characteristics, such as multimedia, audio and video streaming, could be very difficult for network traffic classification. however, with the help of deep learning models, including cnn, mlp and sae, the average online testing accuracies for all models are all above 87% with quite close precision, recall and f1-score values. 5. conclusions deep learning techniques have become one of the most interesting and practical topics being applied on all kinds of fields, such as computer vision, audio recognition and natural language processing. in this paper, we have proposed an application-based online and offline traffic classification, based on deep learning mechanisms, over sdn testbed. the designed deep learning architecture and scheme consists of three deep learning models, mlp, sae, and cnn, in the sdn testbed. we have applied an open network traffic dataset with seven most popular applications as the deep learning training and testing datasets. by using the tcpreplay tool, the dataset traffic samples are re-produced and analyzed in our sdn testbed to emulate the online traffic service. we have conducted performance analysis, in terms of accuracy, precision, recall, and f1 indicators, and compared the results with three deep learning models. the offline training results have achieved more than 93% accuracy on identifying seven popular applications for all three different models. furthermore, we have achieved 87% accuracy for the application-based online testing prediction. advances in technology innovation, vol. 5, no. 4, 2020, pp. 216-229 228 in the future, we will correlate the deep learning parameters, models and accuracy to improve the overall learning performance. the real online packets measurement and analysis in sdn network will be conducted in the next step. the network slicing in sdn network with qos mechanism followed by successful traffic classification could be the further research. acknowledgement this research was supported by research grants from ministry of science and technology, taiwan (most 107-2221-e-142-002-my2) as well as mobile broadband promotion project, ministry of education, taiwan (no. 1070208431g). conflicts of interest the authors declare no conflict of interest. references [1] m. n. a. sheikh, “sdn-based approach to evaluate the best controller: internal controller nox and external controllers pox, onos, ryu,” global journal of computer science and technology, vol. 19, no. 1-e, february 2019. [2] h. c. chu, y. x. liao, l. h. chang, and y. h. lee, “traffic light cycle configuration of single intersection based on modified q-learning”, applied sciences, vol. 9, no. 21, article 4558, october 2019. [3] s. rezaei and x. liu, “deep learning for encrypted traffic classification: an overview,” ieee communications magazine, vol. 57, no. 5, pp. 76-81, may 2019. [4] m. abadi, p. barham, j. chen, z. chen, a. davis, j. dean, et al, “tensorflow: a system for large-scale machine learning,” 12th usenix conference on operating systems design and implementation, august. 2016, pp. 265-283. [5] “keras,” https://keras.io/, june 2019. [6] “open networking foundation,” https://www.opennetworking.org/, june 2019. [7] “ryu sdn framework community,” https://ryu-sdn.org/, june 2019. [8] “tcpreplay,” https://tcpreplay.appneta.com/. [9] g. draper-gil, a. h. lashkari, m. mamun, and a. a. ghorbani, “characterization of encrypted and vpn traffic using time-related features,” proc. 2nd international conference on information systems security and privacy, february 2016, pp. 407-414. [10] j. xie, f. r. yu, t. huang, r. xie, j. liu, c. wang, and y. liu, “a survey of machine learning techniques applied to software defined networking (sdn): research issues and challenges,” ieee communications surveys & tutorials, vol. 21, no. 1, pp. 393-430, 2019. [11] p. amaral, j. dinis, p. pinto, l. bernardo, j. tavares, and h. s. mamede, “machine learning in software defined networks: data collection and traffic classification,” ieee 24th international conference on network protocols, ieee press, november 2016, pp. 1-5. [12] p. wang, s. c. lin, and m. luo, “a framework for qos-aware traffic classification using semi-supervised machine learning in sdns,” proc. ieee international conference on services computing, ieee press, june 2016, pp. 760-765. [13] r. thupae, b. isong, n. gasela, and a. m. abu-mahfouz, “machine learning techniques for traffic identification and classification in sdwsn: a survey,” 44th annual conference of the ieee industrial electronics society in iecon, ieee press, october 2018, pp. 4645-4650. [14] h. k. lim, j. b. kim, j. s. heo, k. kim, y. g. hong, and y. h. han, “packet-based network traffic classification using deep learning,” international conference on artificial intelligence in information and communication, ieee press, february 2019, pp. 046-051. [15] p. wang, f. ye, x. chen, and y. qian, “datanet: deep learning based encrypted network traffic classification in sdn home gateway,” vol. 6, article 18174852, pp. 55380-55391, september 2018. [16] r. hajlaoui, h. guyennet, and t. moulahi, “a survey on heuristic-based routing methods in vehicular ad-hoc network: technical challenges and future trends,” ieee sensors journal, vol. 16, no. 17, pp. 6782-6792, september 2016. [17] a. azzouni, r. boutaba, and g. pujolle, “neuroute: predictive dynamic routing for software-defined networks,” 13th international conference on network and service management, ieee press, january 2017, pp. 1-6. advances in technology innovation, vol. 5, no. 4, 2020, pp. 216-229 229 [18] j. carner, a. mestres, e. alarcn, and a. cabellos, “machine learning based network modeling: an artificial neural network model vs a theoretical inspired model,” ninth international conference on ubiquitous and future networks, ieee press, july 2017, pp. 522-524. [19] t. h. lee, l. h. chang, and w. c. cheng, “design and implementation of sdn-based 6lbr with qos mechanism over heterogeneous wsn and internet,” ksii transactions on internet & information systems, vol. 11, no. 2, pp. 1070-1088, february 2017. [20] n. sultana, n. chilamkurti, w. peng, and r. alhadad, “survey on sdn based network intrusion detection system using machine learning approaches,” peer-to-peer networking and applications, vol. 12, no. 2, pp. 493-501, january 2018. [21] c. song, y. park, k. golani, y. kim, k. bhatt, and k. goswami, “machine-learning based threat-aware system in software defined networks,” 26th international conference on computer communication and networks, ieee press, july 2017, pp. 1-9. [22] t. hurley, j. e. perdomo, and a. perez-pons, “hmm-based intrusion detection system for software defined networking,” 15th ieee international conference on machine learning and applications, ieee press, december 2016, pp. 617-621. [23] “openwrt,” https://openwrt.org/, june 2019. [24] “scikit-learn,” https://scikit-learn.org/stable/index.html, june 2019. [25] m. l. zhang and z. h. zhou, “a review on multi-label learning algorithms,” ieee transactions on knowledge and data engineering, vol. 26, no. 8, pp. 1819-1837, august 2014. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). advances in technology innovation, vol. 6, no. 2, 2021, pp. 128-136 outdoor thermal comfort improvement of campus public space damrongsak rinchumphu 1,2,* , non phichetkunbodee 1,2 , nakarin pomsurin 1,2 , chawanat sundaranaga 2 , sarote tepweerakun 2 , chatchawan chaichana 3 1 department of civil engineering, faculty of engineering, chiang mai university, chiang mai, thailand 2 civil innovation and city engineering laboratory, chiang mai, thailand 3 department of mechanical engineering, faculty of engineering, chiang mai university, chiang mai, thailand received 20 september 2020; received in revised form 31 january 2021; accepted 01 february 2021 doi: https://doi.org/10.46604/aiti.2021.6453 abstract this study focuses on the design of a campus public space, located within the faculty of engineering, chiang mai university, thailand. this area faces extreme temperatures, creating uncomfortable outdoor thermal conditions and hindering activities that are expected to support the learning and social cohesion needs of students. to create the best conditions in this space, three design alternatives such as adding a pond, large trees, or shrubs were considered, and the physiologically equivalent temperature (pet) was used to calculate the outdoor thermal comfort index for each alternative. the alternatives were then compared to the base case. the pet can be calculated using the envi-met simulation software following the appropriate field data collection and calibration process. the results showed that adding large trees in the south-west area is the best design alternative. the pet for this alternative was 3.17 % lower than the base case. in addition, this design workflow is an effective working model for further outdoor public space designs to meet the constraints of effective sustainable development in any tropical campus area. keywords: outdoor public space, thermal comfort, physical equivalent temperature, design alternative 1. introduction the development of a university's outdoor activity area is often in response to the learning and social needs of students, and its building is an important policy of chiang mai university, as it has a focus on creating an environment that is attractive and suitable to be used appropriately [1]. outdoor public space development should be based on indicators that reflect the conditions for the actual use. weather conditions that are not too hot or too cold will have a positive effect on the students who use the outdoor public space for studying as the conditions outside the building are comfortable [2-4]. suitable outdoor thermal conditions for comfort are related to air temperature, relative humidity and wind speed that make the human body feel comfortable [5]. from several research studies, there are a number of outdoor thermal comfort indicators. a commonly used indicator is the physiologically equivalent temperature (pet) [6-8]. the pet is based on modeling the heat balance of the human body. the numerical values are based on the evaporation of moisture related to the heat and air mass, and it is the heat that can maintain the heat balance in humans. pet is also an indicator that can explain the thermal exchange between the body and the environment, and it is used for outdoor comfort analysis. the measured variables are air temperature, relative humidity, wind speed, average heat radiation value and the variables of the human natural state that tries to present them, so they are comparable to human feelings [6, 9]. * corresponding author. e-mail address: damrongsak.r@cmu.ac.th tel.: +66-95-9959519; fax: +66-53-892376 advances in technology innovation, vol. 6, no. 2, 2021, pp. 128-136 129 to develop an outdoor public space that it is more sustainable, this study focuses on outdoor thermal comfort improvement, as represented by the pet around the area within the boundary of the lan witsawa ruamjai, which is the space for outdoor activities on the campus. the result is expected to identify the best alternative design for outdoor public space. the workflow from this study could be used as a framework for improving other campus outdoor public spaces to meet smart campus policy of universities in the future. the paper is structured in five sections: (1) introduction, (2) physiologically equivalent temperature (pet), (3) research methodology, (4) results and discussion and (5) conclusions. 2. physiologically equivalent temperature (pet) there are several indices used to measure the outdoor thermal comfort level [7, 10], for example, predicted mean vote (pmv), effective temperature (et), perceived temperature (pt), physiological equivalent temperature (pet) and universal thermal climate index (utci). according to the research conducted by johansson [11] and klaylee [12], pet is the most suitable indicator for assessing the outdoor thermal comfort. pet can be calculated by a computer program, which represents a realistic state from an airflow simulation program. the most effective simulation software is envi-met v 4.4 [9]. pet is affected by four environmental factors as follows: 2.1. air temperature air temperature is a fundamental factor in climate studies, which always changes over time and varies by season. the comfortable temperature for humans outside a building in greece, according to a study by matzarakis et al. [13], was found to be between 18-23 ºc, with relative humidity and air velocity at optimal conditions. however, srivanit & auttarat [14] conducted a study in chiang mai province, and concluded that the average outdoor comfortable condition was 23.0 – 31.0 o c, which is related to the local conditions of the area in this study. 2.2. relative humidity relative humidity is the ratio of the amount of water vapor in the air at that time compared to the amount of water vapor that the air can accept. the relative humidity in thailand, which is located in the tropical climate zone, during the summer is up to 90%, while in the winter it may drop below 40%, which is approximately 60-70% on average. however, if the relative humidity level is too high, it can make human sweat more result in difficult evaporation, and causes them to feel hot and uncomfortable. while a too low relative humidity will irritate the skin [12]. 2.3. air velocity air velocity is a factor that affects the heat transfer of the body. an uncomfortable situation may be caused by convection when the wind speed increases, and humans will feel colder than the actual temperature. due to the higher cooling rate from the body surface, humans feel 0.4 o c cooler than the actual air temperature when the wind speed increases by 1 km/hr. humans may feel especially cold if the wind speed is too high and the temperature is too low. on the other hand, if the weather is hot and the air velocity is low, the air will not take the heat out of the body, and this causes humans to feel hot and uncomfortable. the wind speed will vary depending on the physical environment and location [9]. it was found that humans feel comfortable with an air velocity of 0.25-0.50 m/s [15]. 2.4. solar radiation solar radiation is the electromagnetic waves emitted from the sun. people often call it sunlight, but the electromagnetic waves are not the only light as there are also other rays [16]. however, the radiation from the sun varies by specific place. advances in technology innovation, vol. 6, no. 2, 2021, pp. 128-136 130 international studies often use a specific city or country to measure the solar radiation when calculating the outdoor thermal comfort. a survey was conducted, and it showed that the average solar radiation value for thailand was 219.9 w/m 2 [17]. furthermore, this study has found 4 factors that are not direct design elements but are consequences of the design and selection of open space. for example, increasing the number of trees or buildings can reduce the temperature of the area. in addition, from the literature review [18-19], there are experiments on modifying the area by changing the surface and green components, such as replacing the land surface with a waterscape. this will increase the water surface area, which affects the temperature and humidity of the surrounding area [19]. increasing the number of trees of various sizes can also decrease the temperature. a tree of any shape and size will affect the shading, wind speed, and solar radiation, resulting in a lower pet value and increasing outdoor thermal comfort in the area, especially in tropical areas such as chiang mai, thailand [20]. these factors can be used when planning and creating outdoor public space design alternatives as well as for determining pet, as per the purpose of this research [21]. 3. research methodology the research methodology includes two parts: the first part is selecting a study area and defining the base case and setting the alternative cases model. the second part is the process of the model simulation and analysis. 3.1. selecting study area and models setting the study area is named lan witsawa ruamjai, an open public space, located in the faculty of engineering, chiang mai university, chiang mai, thailand. it is a space that students can use for relaxation or activities in the university. this research defines the boundary of the study area using aerial images from google earth at the latitude and longitude of 18.7942153, 98.9512263, respectively (fig. 1). fig. 1 aerial view of lan witsawa ruamjai, chiang mai university fig. 2 2d model of study area presented in envi-met version 4.4 advances in technology innovation, vol. 6, no. 2, 2021, pp. 128-136 131 the total area of the space is 3,023.0 m², including 1,114.4 m 2 of hardscape and 1,908.6 m² of softscape, which consists of shrubs and trees. in a radius of 50 m, the area is surrounded by buildings with 16 floors, asphalt concrete road, ditch, green spaces and big trees as shown in fig. 2. according to the design guideline in the trees-nc standard in section 2: site and landscape, the pet value depends on the land surface type, number, and location of trees and shrubs [22]. after determining the existing physical characteristics of the base case, the design alternatives will be developed following the trees – nc standard with three options of adding a pond as a water surface area and adding large trees and shrubs, specifically in the south-west of the study area. the details of the three alternative designs are shown in table 1 and presented as follows: alternative 1: replacing the surface of the space with a 2 m deep pond. the pond would be semi-natural, with a small water treatment system. the pond would increase the relative humidity and also absorbs solar radiation to reduce the pet. alternative 2: plant cork trees, shading, broad and dense foliage over an area of 10 m with perennial trees in the south-west of the study area. these trees will provide shade during daytime, especially in the afternoon when the temperature is at its peak. the large trees will also be able to block radiation from the sun. alternative 3: medium sized shrubs planted around the study area in the south-west. it may add a little shade to the study area and can provide some degree of protection against solar radiation. table 1 details of design alternatives design alternative design type details base case concrete pavement light gray color alternative 1 pond 2 m depth alternative 2 large trees diameter of shading area >10 m two large trees of native plant (cork tree) south – west* alternative 3 shrubs diameter of shading area <5 m medium sized shrubs (banyan tree) south – west* * according to the srivanit & hokao [20] explained that chiang mai, thailand is located at longitude 18.7942153 on, therefore the sun direction will lay on the south side of the area for most of the year, and the sun will raise the air temperature during the afternoon time on the west direction for the whole year, then the most efficient thermal protection concerned must be focused on the south-west direction for this study area. 3.2. model simulation processes the measured meteorological data includes the air temperature, relative humidity, wind speed and solar radiation, as these are the main factors that are directly related to the daily climate. this research was to determine the average daily temperature over the five years 2015-2019 to identify the days with the highest average temperature during the year. the data is from a weather monitoring station of the meteorological department, the northern meteorological center. the monitoring station, located in suthep subdistrict, mueang chiang mai district, chiang mai, is the closest weather monitoring station to the study area, approximately 2.35 kilometers from the site. the highest average five-year (2015-2019) temperature was on april 13, this day will be used in this study. the field data was obtained from a lutron wbgt-2010sd instrument to measure the air temperature, relative humidity, mean radiant temperature, and were calculated using a diameter globe thermometer at 75 mm. the wind velocity was measured using a hot-wire anemometer that was recorded in the testo 435-2 data logger and used to calculate the mean radiant temperature (tmrt) according to eq. (1). advances in technology innovation, vol. 6, no. 2, 2021, pp. 128-136 132 18 0.71 4 4 0.4 1.335 10 [( 273.15) ( )]a mrt g g a v t t t t d       (1) where tg is the globe temperature (°c), va is the wind speed (m/s), ta is the air temperature (°c), d is the globe thermometer diameter (m) and ε is the globe emissivity (ε = 0.95). the air temperature, relative humidity and wind speed were collected from 07:00-18:00. on april 13 th , 2020, by the lutron wbgt-2010sd for the data calibration process. after collecting the weather data, the model calibration process was started by comparing the tmrt between the computer simulation and the field results. the envi-met simulation program was used for calibration, and the results of the field calibration and the five-year average dataset of the meteorological department were compared. by comparing the similarity of the data set with the root mean square error (rmse) and willmott's index of agreement (d), the rmse and d result must be within the acceptable range of 0.66-7.98 and 0.53-0.95, respectively [23]. if they are not in the range, a trial-and-error process must be applied until the rmse and d values are within the acceptable range. after three rounds of calibration, rmse = 1.94 and d = 0.75 were within the acceptable range. the four models in this study, base case and three alternative cases, are presented in fig. 3. the pet simulation results of each model will be presented in the next section. (a) base case (b) alternative 1 (c) alternative 2 (d) alternative 3 fig. 3 models showing approach for outdoor area design by envi-met version 4.4 program 4. results and discussion table 2 average pet during period 07:00-18:00 of base case and different design options. time base case alternative 1 alternative 2 alternative 3 ( o c) (%) ( o c) (%) ( o c) (%) 07:00 23.96 23.95 -0.05 23.95 -0.05 23.95 -0.05 08:00 28.85 28.79 -0.21 28.74 -0.38 28.82 -0.10 09:00 37.42 37.26 -0.43 37.12 -0.82 37.38 -0.10 10:00 43.29 43.06 -0.53 42.86 -0.99 43.23 -0.14 11:00 45.58 45.29 -0.63 45.06 -1.12 45.51 -0.15 12:00 46.46 46.13 -0.71 45.76 -1.51 46.36 -0.21 13:00 48.36 48.26 -0.20 47.80 -1.15 48.32 -0.08 14:00 49.13 49.08 -0.10 48.37 -1.54 49.10 -0.06 15:00 49.45 49.41 -0.08 48.12 -2.68 49.43 -0.04 16:00 49.31 49.27 -0.08 45.82 -7.62 49.30 -0.03 17:00 48.33 48.27 -0.12 40.91 -18.13 48.32 -0.02 18:00 36.91 36.79 -0.32 36.45 -1.25 36.74 -0.47 average 42.25 42.13 -0.29 40.91 -3.17 42.21 -0.12 advances in technology innovation, vol. 6, no. 2, 2021, pp. 128-136 133 the results of the climate data, temperature, and pet from the simulation in the envi-met version 4.4 program on the hottest day in thailand, april 13 during the period 07:00-18:00, measured at 1.40 meters above the ground, are present in table 2. in table 2, the highest pet value predicted on the site was under alternative 3 as the wind was blocked by a long line of trees. the most effective option that gives the lowest pet value is alternative 2, as the trees do not block the wind and give significant shade that blocks the radiation from the sun. the simulation results during the afternoon peak temperature period 13:00-16:00 are shown in figs. 4-8. fig. 4 air temperature at 1.4 m during 13:00-16:00, april 13, 2020 fig. 5 specific humidity at 1.4 m during 13:00-16:00, april 13, 2020 it can be concluded that the hour with the highest pet value for all alternative cases is between 14:00-15:00. according to srivanit & auttarat [14], during this period, the heat accumulated from radiation throughout the day and the rays from the sun causes a high temperature. in addition, the researchers compared the pet values with the base case and the average values obtained in each alternative, and it was found that the difference in the pet during the day compared to the base case advances in technology innovation, vol. 6, no. 2, 2021, pp. 128-136 134 for alternative 2 had the greatest difference from the base case at 1.33 o c and the least difference from the base case of 0.05 degrees as shown in table 2, especially during the period 16:00-17:00. the average value of the pet for all periods of the base case and alternatives 1-3 are shown in table 3. fig. 6 wind speed at 1.4 m during 13:00-16:00, april 13, 2020 fig. 7 mean radian temperature at 1.4 m during 13:00-16:00, april 13, 2020 from table 3, it can be concluded that alternative 2 is the best alternative to improve the outdoor thermal comfort for this area, as adding larger trees to the model has a large effect on the pet value. this is consistent with the studies of gatto et al. [24] and cheung and jim [25], which clearly show the benefits of large trees compared to other cases. table 3 average pet values during day time 6.00-18.00 base case alternative 1 alternative 2 alternative 3 average pet 42.25 42.13 40.91 42.21 different pet from base case ( o c) -0.12 -1.33 -0.04 different pet from base case (%) -0.29 -3.17 -0.09 advances in technology innovation, vol. 6, no. 2, 2021, pp. 128-136 135 fig. 8 pet at 1.4 m during 13:00-16:00, april 13, 2020 5. conclusions this study aims to present an appropriate alternative for improving outdoor public spaces in universities. the study area, named lan witsawa ruamjai, was an open public space, located in the faculty of engineering, chiang mai university, chiang mai, thailand. it is a space where students can relax or perform activities in the university area. to measure the effects of the design alternatives, pet was the index used to choose the best outdoor thermal comfort design. from the results of the three alternative cases, it was found that the option that uses large trees in the south-west locations of the area had the highest pet reduction, compared with the other cases. the results of the study further support the idea of the benefits of large trees for human habitation in urban areas. this is consistent with previous research. in addition, to select the most suitable option by using the simulation process in a quantitative technique. this can give a clear understanding to compare and reduce conflicts when choosing a complicated design alternative. it can be used as an example of a design process for new construction and renovation work in other types of construction project to find a balance for sustainable development that can be effective in the future. acknowledgement this research work was partially supported by the cmu junior research fellowship program. conflicts of interest the authors declare no conflict of interest. references [1] “cmu smart city information,” https://enis.cmu.ac.th/audit_analytic/dashboard.php, may 28, 2019. [2] z. huang, b. cheng, z. gou, and f. zhang, “outdoor thermal comfort and adaptive behaviors in a university campus in china's hot summer-cold winter climate region,” building and environment, vol. 165, pp. 1-11, november 2019. [3] l. zhao, x. zhou, l. li, s. he, and r. chen, “study on outdoor thermal comfort on a campus in a subtropical urban area in summer,” sustainable cities and society, vol. 22, pp. 164-170, february 2016. https://enis.cmu.ac.th/audit_analytic/dashboard.php advances in technology innovation, vol. 6, no. 2, 2021, pp. 128-136 136 [4] f. aram, e. solgi, e. h. garcia, and a. mosavi, “urban heat resilience at the time of global warming: evaluating the impact of the urban parks on outdoor thermal comfort,” environmental sciences europe, vol. 32, pp. 1-15, september 2020. [5] c. m. peccolo, “the effect of thermal environment on learning,” department of health, education and welfare office of education, university of iowa, technical report, 1962. [6] p. höppe, “the physiological equivalent temperature – a universal index for the biometeorological assessment of the thermal environment,” international journal of biometeorology, vol. 43, no.2, pp. 71-75, october 1999. [7] z. fang, x. feng, j. liu, z. lin, c. m. mak, j. niu et al., “investigation into the differences among several outdoor thermal comfort indices against field survey in subtropics,” sustainable cities and society, vol. 44, pp. 676-690, january 2019. [8] f. aram, e. solgi, s. baghaee, e. higueras garcía, a. mosavi, and s. s. band, “how parks provide thermal comfort perception in the metropolitan cores; a case study in madrid mediterranean climatic zone,” climate risk management, vol. 30, pp. 1-14, january 2020. [9] m. srivanit and k. hokao, “evaluating the cooling effects of greening for improving the outdoor thermal environment at an institutional campus in the summer,” building and environment, vol. 66, pp. 158-172, august 2013. [10] k. blazejczyk, y. epstein, g. jendritzky, h. staiger, and b. tinz, “comparison of utci to selected thermal indices,” international journal of biometeorology, vol. 56, no. 3, pp. 515-535, may 2012. [11] e. johansson, “influence of urban geometry on outdoor thermal comfort in a hot dry climate: a study in fez, morocco,” building and environment, vol. 41, no. 10, pp. 1326-1338, october 2006. [12] j. klaylee, “the assessment of physical design for outdoor thermal comfort: case study of thammasat university (rangsit center) (in thai),” bachelor, urban planning, thammasat university, patumthani, http://library.ap.tu.ac.th/searching.php, 2015. [13] a. matzarakis, h. mayer, and m. g. iziomon, “applications of a universal thermal index: physiological equivalent temperature,” international journal of biometeorology, vol. 43, no. 2, pp. 76-84, october 1999. [14] m. srivanit and s. auttarat, “thermal comfort conditions of urban spaces in a hot-humid climate of chiang mai city, thailand,” the 9th international conference on urban climate, july 2015. [15] a. auliciems and s. v. szokolay, thermal comfort, brisbane: plea in association with department of architecture, university of queensland, 1997. [16] j. waewsak, c. chancham, m. mani, and y. gagnon, “estimation of monthly mean daily global solar radiation over bangkok, thailand using artificial neural networks,” energy procedia, vol. 57, pp. 1160-1168, 2014. [17] p. nimnuan and s. janjai, “an approach for estimating average daily global solar radiation from cloud cover in thailand,” procedia engineering, vol. 32, pp. 399-406, 2012. [18] p. vangtook and s. chirarattananon, “an experimental investigation of application of radiant cooling in hot humid climate,” energy and buildings, vol. 38, no. 4, pp. 273-285, april 2006. [19] a. panrare, p. sohsalam, and t. tondee, “constructed wetland for sewage treatment and thermal transfer reduction,” energy procedia, vol. 79, pp. 567-575, november 2015. [20] m. srivanit and k. hokao, “effects of urban development and spatial characteristics on urban thermal environment in chiang mai metropolitan, thailand,” lowland technology international, vol. 14, no. 2, pp. 9-22, december 2012. [21] p. suropan, d. rinchumphu, and m. srivanit, “the study of eco-efficiency from outdoor thermal impacts for hi-end condominium project in central business district of bangkok (in thai),” the national conference on “vernacular creativity wisdom”, faculty of architecture, chiang mai university, january 2017, pp. 57-64. [22] thai green building institute, thai’s rating of energy and environmental sustainability for new construction and major renovation, 1 st ed. bangkok: thai green building institute, 2012. [23] f. salata, i. golasi, r. de lieto vollaro, and a. de lieto vollaro, “urban microclimate and outdoor thermal comfort: a proper procedure to fit envi-met simulation outputs to experimental data,” sustainable cities and society, vol. 26, pp. 318-343, october 2016. [24] e. gatto, r. buccolieri, e. aarrevaara, f. ippolito, r. emmanuel, l. perronace, and j. l. santiago, “impact of urban vegetation on outdoor thermal comfort: comparison between a mediterranean city (lecce, italy) and a northern european city (lahti, finland),” forests, vol. 11, no. 2, pp. 1-22, february 2020. [25] p. k. cheung and c. y. jim, “comparing the cooling effects of a tree and a concrete shelter using pet and utci,” building and environment, vol. 130, pp. 49-61, december 2017. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation, vol. 6, no. 1, 2021, pp. 39-46 the impact of coating ingredients on the aging resistance of topcoat paints by model trees tzu-tsung wong * , shih-hsuan hung institute of information management, national cheng kung university, tainan, taiwan received 20 may 20xx; received in revised form 15 july 2020; accepted 14 september 2020 doi: https://doi.org/10.46604/aiti.2021.5307 abstract topcoat paint is mainly composed of resin and pigment and hence its quality highly depends on the type and proportion of these two ingredients. this study aims at testing the formula of the topcoat paint for finding one that can achieve better quality for anti-aging. various formulas of paint are applied on boards that will be put into ultraviolet accelerated test machines to simulate weathering tests. the gloss and color, before and after the tests, are collected and numerical prediction method m5p is used to grow model trees for discovering the key factors affecting aging. based on the structure and the linear regression models in the trees, a better topcoat paint should be composed of a high proportion of resin and generally a low proportion of pigment. good types of resin and pigment are also identified for keeping color and gloss. keywords: accelerated aging test, linear regression, model tree, numeric prediction, topcoat paint 1. introduction buildings are usually covered by paint to enhance their looking and durability. there are two main categories of paint: undercoating and topcoating. topcoat paint can protect buildings from erosion and the function of undercoat paint is to increase the adhesion of topcoat paint. the life of a building is thus primarily determined by the quality of topcoat paint. the two main materials for composing paint are resin and pigment. the type and proportion of these two ingredients have a great impact on the appearance and aging-resistance properties of topcoat paint. if the relationship between the ingredients and the aging properties can be determined, an enterprise will be able to produce a topcoat paint that can satisfy the requirements of a customer at a relatively low cost. the coating on objects is critical to the protection of the objects and hence this kind of the issue has been studied in several applications [1-4]. accelerated tests were generally adopted to obtain observations for studying the properties of paints [5-7]. resin and pigment are the two main ingredients of paints. many recent studies attempted to explore the individual impact of resin [8-9] or pigment [10-14]. these two ingredients can also be considered together to determine the performance of coatings [15-17]. none of those previous studies applied numeric prediction methods to analyse the impact of resin and pigment on the aging properties of paints. the purpose of this paper is to explore the various formulas of topcoat paint. the data collected for various formulas will be analysed by a numeric prediction method that can produce a tree structure with multiple linear regression models. the tree and the regression models can be helpful in finding the attributes that are critical to the properties of topcoat paint. these results can be employed to compose the topcoat paint for satisfying customer needs. * corresponding author. e-mail address: tzutsung@mail.ncku.edu.tw tel.: +886-6-2757575; fax: +886-6-2362162 advances in technology innovation, vol. 6, no. 1, 2021, pp. 39-46 40 2. data collection the three factors that can affect the life of paint are solar radiation, temperature, and humidity, and solar radiation has the most significant impact. better quality of resin and pigment can enhance the aging-resistant properties of paint. a company uses three types of resin r1, r2, and r3 and three types of pigment p1, p2, and p3 provided by different vendors to produce paint. the proportion of the resin in a formula can be between 30% and 60%, and this value for pigment is generally between 15% and 35%. the four attributes for determining the aging-resistant properties of paint are shown in table 1. the number of possible combinations resulting from the four attributes is 3435 = 180. the paint of each combination will be applied on two boards to collect their data for the aging resistance; accordingly, the number of instances in a data set will be 360. note that the two proportions x2 and x4 are considered as continuous attributes for analysing their impact on the age-resistant properties. table 1 the attributes for determining the aging-resistant properties of paint attribute possible values x1 (resin type) r1, r2, r3 x2 (resin proportion) 30%, 40%, 50%, 60% x3 (pigment type) p1, p2, p3 x4 (pigment proportion) 15%, 20%, 25%, 30%, 35% the lifetime of a system is the duration for the system to operate normally. however, the degradation of the coating in a building occurs gradually as time goes by. the age of paint is thus represented by its colour and gloss. an aluminium board coating by paint will be dried naturally for a week. then its colour and gloss will be measured and recorded as c0 and g0, separately. fig. 1 shows the colour meter to evaluate the colour in a board. for the sake of simplicity, the colour of all formulas is set to be white. a smaller value implies that the white colour is brighter. similarly, the gloss of a board is measured by the gloss meter shown in fig. 2. the gloss values of five fixed positions in a board are measured, and their average will be the gloss level of the board. fig. 1 the colour meter fig. 2 the gloss meter after the initial color c0 and gloss g0 of each board had been obtained, the 360 boards were put into ultraviolet accelerated test machines for t hours. then the color and gloss of each board were measured again as before, and they were denoted as c1 and g1, respectively. this testing process in the ultraviolet accelerated test machines lasted 5t hours, and every board was taken out the machine for measuring color and gloss every t hours. the exposure conditions of the machine were set to be the cycle 2 of x2.1 given in astm g154-16 provided by astm international. let cj and gj be the color and the gloss of a board at jt hours for j = 1, 2, 3, 4, 5. the color difference for jt hours was calculated as dj = cj-c0, and the gloss retention rate for jt hours was computed as ej = gj/g0. the data set for color difference collected at jt hours was denoted as dj in which every instance was represented as <x1, x2, x3, x4, dj>. similarly, the data set for gloss retention rate collected at jt hours was denoted as ej in which every instance was represented as <x1, x2, x3, x4, ej>. let the paint coating in a board be the formula with 40% type r1 resin and 30% type p2 pigment. the color and gloss measured after one-week natural drying are c0 = 0.1 and g0 = 60, individually. after the board is put into an ultraviolet accelerated test machine for 2t hours, its color and gloss becomes c2 = 0.5 and g2 = 51, separately. the instances for this board advances in technology innovation, vol. 6, no. 1, 2021, pp. 39-46 41 in data sets d2 and e2 are <r1, 0.4, p2, 0.3, 0.4> and <r1, 0.4, p2, 0.3, 0.85>, respectively. it is possible that the aging-resistant properties of formula may degrade slowly at the beginning while its degrading becomes quick after a fixed time point or vice versa. the 10 data sets can provide this kind of information to satisfy the needs of topcoat paints. 3. model trees and data analysis linear regression is the most popular tool for numeric prediction. it provides a linear model to analyze the relationships between independent variables and a dependent variable. when the instances in a data set are not collected from the same population, the prediction error resulting from single linear regression model is generally large. more powerful tools for numeric prediction are therefore developed to make more accurate predictions [18-21]. the model tree proposed by quinlan [19] has a tree structure and multiple linear regression models for numeric prediction and interpretation, and its prediction error is generally far smaller than that resulting from single linear regression model. this tool has been employed in several recent applications for numeric prediction [22-26]. this section will briefly describe the way for building a model tree from a data set. the evaluation and the interpretation of a model tree will also be introduced. 3.1. model trees it is similar to decision trees, model trees have a tree structure that contains internal nodes for branching and leaf nodes for prediction. every leaf node in a model tree has a linear regression model to predict the class value of any instance reaching this node. the attribute chosen for branching is determined by its standard deviation reduction. the domain of a continuous attribute is divided into ten equal-width intervals by nine splitting points. the expected standard deviation of the class value for every binary branching on each of the nine splitting points is calculated and the one with the smallest expected standard deviation will be the best splitting point for this attribute. the one that has the smallest expected standard deviation on the class value among all attributes is chosen for branching. after the tree is fully grown, a linear regression model is derived for each node based on the instances reaching that node. the regression models in internal nodes will be used for pruning and the regression models in leaf nodes are employed for predicting the class value. a discrete attribute with m possible values is replaced by m-1 synthetic binary attributes. the expected standard deviation of the class value for every binary attribute can be calculated to determine whether this attribute should be chosen for binary branching in an internal node. example 1 given below shows the way to interpret the learning results for discrete attributes. example 1: suppose that a data set has one continuous attribute x1 and one discrete attribute x2 with four possible values a, b, c, and d for predicting class value y. the order of these values is c, d, a, b based on their average class value summarized from the data set; i.e., values c and b have the largest and the smallest average class values, respectively. then the three synthetic binary attributes for replacing x2 are given in table 2. if the attribute value of x2 appears in a binary attribute, the value of this binary attribute equals 1 and equals zero otherwise. for example, an instance <0.4, d> will be replaced by <0.4, 1, 0, 0>. let the linear regression equation in the leaf node reaching by this instance be y = 1.26x1 – 0.35d-a-b + 0.082b. then the predicted class value of this instance is y = 1.260.4 – 0.351 + 0.0820 = 0.154. table 2 the synthetic binary attributes for replacing the discrete attribute x2 in example 1 attribute value d-a-b a-b b a 1 1 0 b 1 1 1 c 0 0 0 d 1 0 0 advances in technology innovation, vol. 6, no. 1, 2021, pp. 39-46 42 3.2. evaluation and interpretation a small prediction error is the primary concern for numeric prediction. since the color difference and the gloss retention rate have different scales, an evaluation measure should be able to indicate whether a formula of topcoat paint can achieve a small prediction error on both aging-resistance properties. relative absolute error (rae) and root relative squared error (rrse) are therefore adopted for performance evaluation. let yi and yi ’ be the actual and the predicted class value of instance i. then these two measures are calculated as: || || rae ' yy yy ii iii    (1) and 2 2' )( )( rrse yy yy ii iii    (2) where y is the average class value. these two evaluation measures have been normalized, and hence they can be employed to compare the prediction error between the color difference and the gloss retention rate. like decision trees, the branching attributes in a model tree determine the leaf node in which an instance will reach. moreover, the linear regression models in leaf nodes will show critical attributes and their impact level on the dependent variable. there are five model trees for the color difference and similarly for the gloss retention rate. the five model trees for each aging-resistant property will be put together to analyze the critical attributes and their impact levels. those results can be helpful to know which kind of formula will be better for aging resistance. 4. experimental results as described in section 2, the color and gloss of each board were measured for every t hours. the value of t was set to be 250 hours; therefore, the data were collected at 0, 250, 500, 750, 1000, and 1250 hours. the class values in the data sets d1 through d5 and e1 through e5 are collected by this way, and they are divide into two groups for the color difference and the gloss retention ratio. these two groups of data sets will be analyzed by the function m5p with default settings provided by the software weka in the following two subsections for drawing conclusions. only the analysis on the data sets d2 and e2 obtained at 500 hours will be shown in detail because their learning results are simpler for interpretation. then the five data sets in each group will be put together for identifying the critical attributes of topcoat paint in anti-aging. since estimating the lifetime of topcoat paint is not the goal of this study, setting t = 250 hours ensures that enough data can be collected within reasonable time. 4.1. color difference table 3 the characteristics of the five model trees for colour difference data set d1 d2 d3 d4 d5 rae 81.7% 76.2% 75.2% 74.2% 73.9% rrse 86.1% 84.8% 80.8% 80.6% 79.3% branching attributes x1, x2, x3 x1, x2, x3 x1, x2, x3 x1, x2, x3 x1, x2, x3, x4 tree size 8 4 9 8 8 the prediction error and the size of the five model trees for the color difference are summarized in table 3 in which tree size represents the number of leaf nodes. both the relative absolute error and the root relative square error are getting smaller as advances in technology innovation, vol. 6, no. 1, 2021, pp. 39-46 43 the number of testing hours is increasing. the prediction on color difference is more accurate on the topcoat paint exposing longer. attribute x4 appears only in the tree structure for data set d5. this suggests that pigment proportion is the least important among the four attributes. the model tree and its corresponding linear regression models for data set d2 are shown in fig. 3 and table 4, separately. attribute x4 (pigment proportion) is the only one that is not chosen for branching; consequently, it is less important than the other three attributes in determining the color difference of topcoat paints. according to the four linear regression models, the overall coefficients for resin types r1, r2, and r3 are 0, -0.0670 + 0.0950 = 0.0280, and -0.1665 0.0670 + 0.0184 + 0.0950 + 0.0184 + 0.0310 = -0.0697, respectively. these imply that the best and the worst resin types for reducing color difference are r3 and r2, respectively. since the four coefficients for resin proportion are all negative, higher resin proportion is better for reducing color difference. similar to resin types, the overall coefficients for pigment types p1, p2, and p3 are 0, 0.0314, and 0.4623, respectively. the priority order for pigment types is thus p1, p2, and p3. the three coefficients for pigment proportion are all positive; therefore, the pigment proportion should not be high for maintaining the color of topcoat paint. fig. 3 the model tree grown from data set d2 table 4 the linear regression models for the model tree given in fig. 3 attribute lm1 lm2 lm3 lm4 r2-r3 -0.0670 0.0950 r3 -0.1655 0.0184 0.0184 0.0310 x2 -0.0032 -0.0013 -0.0008 -0.0006 p3-p2 0.0122 0.0064 0.0064 0.0064 p2 0.0116 0.2462 -0.0472 0.2203 x4 0.0004 0.0004 0.0136 constant 0.7149 0.6290 0.5873 0.3376 the above analysis is applied to all of the five model trees for the color difference and the results are summarized in table 5. when the testing time is more than 500 hours, the best and the worst resin types are r1 and r2, respectively. higher resin proportion is definitely beneficial for maintaining the color of topcoat paints. the ranks of pigment types are consistent all the time. table 5 also suggests that the pigment proportion should not be high. according to the experimental results given in table 5, the best formula for the color of topcoat paint is the combination of a high proportion of r1 resin and a low proportion of p1 pigment. table 5 the summarization of the linear regression models for colour difference data set d1 d2 d3 d4 d5 rank of resin types r2, r1, r3 r1, r3, r2 r1, r3, r2 r1, r3, r2 r1, r3, r2 resin proportion negative negative negative negative negative rank of pigment types p1, p3, p2 p1, p3, p2 p1, p3, p2 p1, p3, p2 p1, p3, p2 pigment proportion negative positive positive positive positive advances in technology innovation, vol. 6, no. 1, 2021, pp. 39-46 44 4.2. gloss retention ratio table 6 shows the prediction errors and the sizes of the five model trees for the gloss retention rate. the testing time in the machines seems to have almost no impact on the prediction error. all five models trees have branching attributes resin type (x1); as a result, it is the most important attribute for gloss. pigment type (x3) and pigment proportion (x4) become more important as exposing time is getting longer. note that the prediction errors for the gloss retention rate are all smaller than those given in table 3 for the color difference. table 6 the characteristics of the five model trees for gloss retention rate data set e1 e2 e3 e4 e5 rae 65.3% 59.5% 68.3% 65.1% 65.4% rrse 68.9% 61.5% 70.9% 67.5% 67.7% branching attributes x1, x2, x3 x1, x2, x4 x1, x2, x3, x4 x1, x3, x4 x1, x2, x3, x4 tree size 8 6 6 5 8 the model tree and its corresponding linear regression models for data set e2 are shown in fig. 4 and table 7, respectively. the only attribute that is not chosen for branching is x3 (pigment type), and hence it is less important than the other three attributes in determining the gloss of topcoat paints. a small color difference is preferred, while better topcoat paint should have a larger gloss retention rate. the coefficients for resin types r1, r2, and r3 calculated from linear regression model lm1 in table 7 are 1.6714+0.7991 = 2.4705, 1.6714, and 0, respectively. as a result, the best and the worst resin types for maintaining gloss are r1 and r3, respectively. the same result can be obtained from the other five linear regression models. the four coefficients for resin proportion are all positive; accordingly, higher resin proportion is better for the gloss retention ra te. the coefficients for pigment types p1, p2, and p3 calculated from lm1 are 21.9847-17.9402 = 2.0445, 21.9847, and 0, respectively; consequently, the priority order is p2, p1, p3. although the six linear regression models do not have a consistent rank on pigment types, most of them follow this order. half of the six coefficients for pigment proportion are positive. it is thus inconclusive in determining the impact of pigment proportion on the gloss of topcoat paint. fig. 4 the model tree grown from data set e2 table 7 the linear regression models for the model tree given in fig. 4 attribute lm1 lm2 lm3 lm4 lm5 lm6 r2-r1 1.6714 1.6714 1.6714 0.8849 0.8849 0.8849 r1 0.7991 0.7991 0.7991 1.1751 1.1751 1.1751 x2 0.1615 0.3795 0.2530 0.2476 0.4002 0.2023 p2-p1 21.9847 2.5062 -16.0532 11.8200 -5.8016 6.1060 p1 -17.9402 -7.3138 24.6857 -3.3163 5.7759 -4.3692 x4 -0.1888 -1.0879 -0.9935 0.3213 0.4718 0.0083 constant 58.7103 78.9668 80.5453 59.4366 44.5479 73.7545 advances in technology innovation, vol. 6, no. 1, 2021, pp. 39-46 45 the above analysis is applied to all of the five model trees for the gloss retention rate, and the results are summarized in table 8. the best and the worst resin types are r1 and r3, respectively. similar to the analysis performed for the color difference, higher resin proportion is definitely beneficial for maintaining the gloss of topcoat paint. it seems that p2 is the best among the three pigment types for the gloss retention rate. the proportion of pigment in composing topcoat paint should depend on the types of resin and pigment. based on the results obtained from table 8, the best formula for the gloss of topcoat paint is the combination of high proportion r1 resin and generally low proportion p2 pigment. table 8 the summarization of the linear regression models for gloss retention rate data set e1 e2 e3 e4 e5 rank of resin types r1, r2, r3 r1, r2, r3 r1, r2, r3 r1, r2, r3 r1, r2, r3 resin proportion positive positive positive positive positive rank if pigment types p3, p1, p2 p2, p1, p3 p2, p1, p3 p2, p1, p3 p2, p3, p1 pigment proportion negative inconclusive inconclusive inconclusive negative 5. conclusions in this paper, the aging-resistance properties of topcoat paint measured by its color and gloss are investigated by numeric prediction methods. a formula of topcoat paint is mainly composed of resin and pigment. this study analyzed 180 possible formulas of resin type, resin proportion, pigment type, and pigment proportion to determine their impact on the aging-resistance properties. ultraviolet accelerated test machines were the devices for testing the paint covered in aluminum boards to collect their color differences and gloss retention rates for every 250 hours. the data collected at the same time are gathered in a file to grow a model tree for studying the impact of the four attributes for composing topcoat paint. the experimental results on the five data sets for the color difference consistently indicate that the best formula is the combination of a specific type of resin with a high proportion and a specific type of pigment with a low proportion. the experimental results on the five data sets for the gloss retention rate have the same conclusion on the resin type and resin proportion. however, the best pigment types for color and retention are different and the pigment proportion for gloss generally depends on the other three attributes. the information obtained in this study could be helpful in finding better formulas to satisfy the needs of customers and in choosing vendors who provide resin and pigment. the lifetime of topcoat paint is another key factor in determining its quality. the color and gloss of paint obtained from ultraviolet accelerated test machines can be good indicators about its lifetime in natural environments. estimating the lifetime of topcoat paint should thus be an interesting topic for the extension of this study. acknowledgement this research was supported by the ministry of science and technology in taiwan under grant no. 107-2410-h-006-045-my3. conflicts of interest the authors declare no conflict of interest. references [1] a. boubault, c. k. ho, a. hall, t. n. lambert, and a. ambrosini, “durability of solar absorber coatings and their cost-effectiveness,” solar energy materials and solar cells, vol. 166, pp. 176-184, july 2017. [2] k. simunkova, m. panek, and a. zeidler, “comparison of selected properties of shellac varnish for restoration and polyurethane varnish for reconstruction of historical artefacts,” coatings, vol. 8, no. 4, march 2018. [3] j. wu, k. niu, b. su, and y. wang, “effect of combined uv thermal and hydrolytic aging on micro-contact properties of silicone elastomer,” polymer degradation and stability, vol. 151, pp. 126-135, may 2018. advances in technology innovation, vol. 6, no. 1, 2021, pp. 39-46 46 [4] e. kizilkonca and f. b. erim, “development of anti-aging and anticorrosive nanoceria dispersed alkyd coating for decorative and industrial purposes,” coatings, vol. 9, no. 10, september 2019. [5] p. m. carmona-quiroga, r. m. j. jacobs, s. martinez-ramirez, and h. a. viles, “durability of anti-graffiti coatings on stone: natural vs accelerated weathering,” plos one, vol. 12, no. 2, february 2017. [6] p. sanmartin and f. cappitelli, “evaluation of accelerated ageing tests for metallic and non-metallic graffiti paints applied to stone,” coatings, vol. 7, no. 11, october 2017. [7] a. k. soares, r. l. pereira, p. h. gonzalez de cademartori, h. w. dalla costa, and d. a. gatto, “artificial weathering of four coatings applied on woods of two forest species,” nativa: pesquisas agrárias e ambientais, vol. 6, no. 3, pp. 313-320, 2018. [8] r. david, v. s. raja, s. k. singh, and p. gore, “development of anti-corrosive paint with improved toughness using carboxyl terminated modified epoxy resin,” progress in organic coatings, vol. 120, pp. 58-70, july 2018. [9] x. x. yan, “effect of black paste on the property of fluorine resin/aluminum infrared coating,” coatings, vol. 9, no. 10, september 2019. [10] t. ramde, l. g. ecco, and s. rossi, “visual appearance durability as function of natural and accelerated ageing of electrophoretic styrene-acrylic coatings: influence of yellow pigment concentration,” progress in organic coatings, vol. 103, pp. 23-32, february 2017. [11] s. n. roselli, r. romagnoli, and c. deyda, “the anti-corrosion performance of water-borne paints in long term tests,” progress in organic coatings, vol. 109, pp. 172-178, august 2017. [12] m. kohl, a. kalendova, e. cernoskova, m. blaha, j. stejskal, and m. erben, “corrosion protection by organic coatings containing polyaniline salts prepared by oxidative polymerization,” journal of coatings technology and research, vol. 14, no. 6, pp. 1397-1410, november 2017. [13] n. m. ahmed, m. g. mohamed, r. h. tammam, and m. r. mabrouk, “performance of coatings containing treated silica fume in the corrosion protection of reinforced concrete,” pigment & resin technology, vol. 47, no. 4, pp. 350-359, july 2018. [14] q. m. abd el-gawad, n. m. ahmed, m. m. selim, e. hamed, and e. r. souaya, “the anticorrosive performance of cost saving zeolites,” pigment & resin technology, vol. 48, no. 4, pp. 317-328, july 2019. [15] r. s. peres, a. v. zmozinski, f. r. brust, a. j. macedo, e. armelin, c. aleman, and c. a. ferreira, “multifunctional coatings based on silicone matrix and propolis extract,” progress in organic coatings, vol. 123, pp. 223-231, october 2018. [16] w. h. li, d. c. franco, m. s. yang, x. p. zhu, h. p. zhang, y. y. shao, h. zhang, and j. x. zhu, “investigation of the performance of ath powders in organic powder coatings,” coatings, vol. 9, no. 2, february 2019. [17] t. kanbayashi, a. ishikawa, m. matsunaga, m. kobayashi, and y. kataoka, “application of confocal raman microscopy for the analysis of the distribution of wood preservative coatings,” coatings, vol. 9, no. 10, september 2019. [18] a. karalic, “employing linear regression in regression tree leaves,” proc. of the 10th european conference on artificial intelligence, august 1992, pp. 440-441. [19] j. r. quinlan, “learning with continuous classes,” proc. of the 5th australian joint conference on artificial intelligence, november 1992, pp. 343-348. [20] p. chaudhuri, m. c. huang, w. y. loh, and r. yao, “piecewise-polynomial regression trees,” statistica sinica, vol. 4, no. 1, pp. 143-167, january1994. [21] w. y. loh, “regression trees with unbiased variable selection and interaction detection,” statistica sinica, vol. 12, no. 2, pp. 361-286, april 2002. [22] e. ghasemi, h. kalhori, r. bagherpour, and s. yagiz, “model tree approach for predicting uniaxial compressive strength and young's modulus of carbonate rocks,” bulletin of engineering geology and the environment, vol. 77, no. 1, pp. 331-343, february 2018. [23] f. t. matthew, a. i. adepoju, o. ayodele, o. olumide, o. olatayo, e. adebimpe, o. bolaji, and e. funmilola, “development of mobile-interfaced machine learning-based predictive models for improving students' performance in programming courses,” international journal of advanced computer science and applications, vol. 9, no. 5, pp. 105-115, 2018. [24] h. p. su, c. lian, j. c. liu, and h. l. liu, “machine learning models for solvent effects on electric double layer capacitance,” chemical engineering science, vol. 202, pp. 186-193, july 2019. [25] h. s. yi, b. lee, s. park, k. c. kwak, and k. g. an, “prediction of short-term algal bloom using the m5p model-tree and extreme learning machine,” environmental engineering research, vol. 24, no. 3, pp. 404-411, 2018. [26] z. s. khozani, k. khosravi, b. t. binh, b. klove, w. h. m. w. mohtar, and z. m. yaseen, “determination of compound channel apparent shear stress: application of novel data mining models,” vol. 21, no. 5, pp. 798-811, 2019. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). 4__aiti#6517__169-178 advances in technology innovation, vol. 6, no. 3, 2021, pp. 169-178 fuzzy study on the winning rate of football game betting woo-joo lee1, hyo-jin jhan2, seung-hoe choi3,* 1department of mathematics, yonsei university, seoul, korea 2school of aerospace and mechanical engineering, korea aerospace university, goyang, korea 3school of liberal arts and science, korea aerospace university, goyang, korea received 03 october 2020; received in revised form 14 april 2021; accepted 15 april 2021 doi: https://doi.org/10.46604/aiti.2021.6517 abstract this study aims to find variables that affect the winning rate of the football team before a match. qualitative variables such as venue, match importance, performance, and atmosphere of both teams are suggested to predict the outcome. regression analysis is used to select proper variables. in this study, the performance of the football team is based on the opinions of experts, and the team atmosphere can be calculated with the results of the previous five games. elo rating represents the state of the opponent. also, the selected qualitative variables are expressed in fuzzy numbers using fuzzy partitions. a fuzzy regression model for the winning rate of the football team can be estimated by using the least squares method and the least absolute method. it is concluded that the stadium environment, elo rating, team performance, and importance of the match have effects on the winning rate of korean national football (knf) team from the data on 118 matches. keywords: winning rate prediction, elo rating, fuzzy number, fuzzy partition, regression model 1. introduction for the member countries of international federation of association football (fifa), their primary interest is the athletic performance in the world cup qualifiers and the finals [1]. predicting the results of football matches as well as other popular sport matches is also important for fans [2]. thus, the methods for forecasting football results have been studied in various ways [3-10]. researchers can take advantages of the long history of football because the football results have been continuously recorded. since modern football has been first introduced to korea, knf team has played a total of 941 games, with a winning percentage of 54.3% [11]. presently, the primary methods for predicting a football game are sorted into two types: 1) quantitative methods and 2) qualitative methods. the quantitative approach is used to predict the outcome of a match by analyzing the previous datasets of both teams competing in the match [12-16]. on the other hand, the qualitative approach is used to predict the outcome of a match by expressing the characteristics that cannot be expressed numerically. these might include the athletic performance of the teams, the match location, and the physical condition of the players [17]. in addition, there is also a combination of two approaches. however, while the research on quantitative methods of predicting the outcome of sport events is more prominent, the research on qualitative and combined methods seems to be scarce [18-19]. furthermore, the variables that affect the performance of each team before the match include the condition of the players’ participation, the condition of teamwork, and the starting squads. it can be quite exciting for football fans to predict the outcome of matches using variables that can be inferred before the match [3]. * corresponding author. e-mail address: shchoi@kau.ac.kr tel.: +82-2-300-0073; fax: +82-2-300-0492 advances in technology innovation, vol. 6, no. 3, 2021, pp. 169-178 the objective of this study is to propose the quantitative and qualitative variables affecting the winning rate of the football team before the match, and to present a statistical model of the winning rate of the team using the selected variables. accordingly, we express qualitative variables in fuzzy numbers and use the selected quantitative and qualitative variables to infer a fuzzy regression model of the football team’s winning rate. for this purpose, we analyze 118 games of korean national football (knf) team. the overall workflow can be sorted into three steps: (1) collecting raw data: public football in-game data have long been recorded and published online. the quantitative data provided by the korea football association (kfa) on 118 matches from september 2, 2011 to november 14, 2019 are collected. also, the qualitative data are primarily extracted from reviews of football experts. (2) pre-processing and transformation: numerically derived variables such as the winning rate, difference in elo rating, and atmosphere are added, and the details of those are described by eqs. (7)-(12). also, the qualitative variables such as the performance of the team and the importance of a match are transformed into fuzzy numbers. (3) fitting the data into the fuzzy regression model and estimate the coefficients: section 2 presents fuzzy regression model for the winning rate of the football game betting. section 3 gives the numerical results of knf team using the absolute deviation method. section 4 concludes the study. 2. fuzzy regression model for the winning rate of the football game betting this section introduces a fuzzy regression model for the winning rate using triangular fuzzy numbers and fuzzy partitions. regression model (eq. (1)) is presented to analyze the impact of match location (��), elo rating difference (�� ), atmosphere of the national football team (��), performance (��), and importance (��) of the match on the winning rate ( ) of the football team. ( ), , , ,w g e a a gp f p d t a i= (1) to estimate a given regression model in the classical statistical method, the residuals, which are the differences between observations and estimated values, must meet gauss-markov conditions. in other words, the residuals must satisfy the independence, the normality, and the heteroscedasticity [20-21]. if there is a correlation between independent variables, such as atmosphere, athletic performance and match importance, the use of traditional regression will result in multicollinearity problems. in addition, all variables used in the regression model can be expressed as quantitative variables, but sometimes the independent or dependent variables can be expressed as qualitative or categorical. although qualitative data can be quantitatively expressed using probability variables to estimate regression models, this can represent a reduction or omission of the information contained in the qualitative data. thus, it may be more efficient to use qualitative or linguistic data for the regression analysis. the proposed statistical analysis method to account for the causal relationship of such linguistic or qualitative data is the fuzzy regression model. in 1980, professor tanaka introduced the first fuzzy regression model, which applied fuzzy numbers to the regression model introduced by professor zadeh in 1965 [22-23]. since then, fuzzy regression models have been used in many areas [24-26]. data used for fuzzy regression can be expressed in triangular fuzzy numbers. the triangular fuzzy number, denoted by numbers a = (�� , �, �� )� , consists of the mode (�), left endpoint (�� ), and right endpoint (�� ) [26]. the membership functions of the right and left sides of the triangular fuzzy number are shown in fig. 1. 170 advances in technology innovation, vol. 6, no. 3, 2021, pp. 169-178 ( ) ( ) , 0, . 1 a a aa x l l x o w a a ll x − ≤ ≤ −=      (2) and ( ) ( ) 0, , . 1 a a aa x r a x r a rr o x w − ≤ ≤ −=      (3) fig. 1 fuzzy number in this study, we apply a fuzzy partition to represent the winning rate, performance, and importance of the football team in triangular fuzzy numbers. if ��� ∶ �� (0) ⊆ �, � = 1, … , � � 1 is set as a fuzzy partition of the universal set �, it satisfies for each i and any ! ∈ �: ( )0ia φ≠ (4) ( )11 0ni iu a x+= = (5) ( )1 1 1 n ii a x + = =∑ (6) where �� (0) is a 0-level set of a fuzzy set �� [27]. fig. 2 implies that the sum of the membership value at one point is 1. fig. 2 shows a fuzzy segmentation consisting of five points: #$, #%, #&, #', and #(. we use the least absolute deviation method suggested by choi and buckley [24] to estimate the regression coefficients of the fuzzy regression model (eq. (1)). they first transformed the fuzzy number into crisp number to estimate the regression model for the center [24]. fig. 2 fuzzy partition 171 advances in technology innovation, vol. 6, no. 3, 2021, pp. 169-178 3. regression analysis for the winning rate of knf team this section examines the variables that affect the winning rate of knf team in the betting, and estimates the fuzzy regression model for knf team. this study uses the data provided by kfa on 118 matches from september 2, 2011 to november 14, 2019, along with the columns written by the experts who analyzed knf team’s performance [2]. this section explains various quantitative and qualitative variables that have influenced the winning rate of knf team, and it subsequently produces a fuzzy regression model for the winning rate of knf team using the selected variables. 3.1. quantitative data for the model we study 118 sets of knf team’s data to forecast the result of the football matches. the anova results show that knf team’s winning rates not only vary with match types such as friendly matches, world cup qualifiers, east asian cup finals, and world cup matches, but also vary with the region in which the matches take place (� − *#�+, < 0.05) [28]. knf team has its lowest winning rate in the world cup matches and the highest winning rate in the world cup qualifiers. also, the winning rate is higher at home games than at neutral games and away games (table 1). table 1 results according to the match type and region match type number of matches winning rate location number of matches winning rate friendly 60 0.42 home 50 0.56 world cup qualifier 35 0.63 neutral 40 0.43 east asian cup 6 0.48 away 28 0.45 asian cup 11 0.57 world cup 6 0.23 total 118 mean: 0.49 the quantitative data on the football team can be classified into two categories: before the match and after the match. the quantitative data that are available before the match are the team’s fifa ranking, elo rating, and the dividend rate for the football team announced by the betting company. the fifa / coca-cola world ranking, which was released for the first time in august 1993, can be obtained from the website of the international football confederation [29-30]. the elo rating, which was developed by arpad elo in 1997, can be found online [31-32]. the sample correlation (0�1 ) between the fifa and elo rankings for knf team is positive (0�1 = 0.48, � − *#�+, < 0.000). since elo rating is instantly published after every match and fifa ranking is updated every month, it may be more efficient to predict knf team’s winning rate by using elo rating than fifa ranking. kfa releases the results of every match through the knf team website. there are 17 types of data on game scoreboards published by kfa, including goals for (gf), goals against (ga), shots (!�&), shots on target (!�'), fouls (!�(), yellow cards, red cards, offsides, corner kicks (!�4 ), saves, clears, interceptions, aerial ball winning rates, pass success rates (!�$'), ground ball winning rates (!�$(), pass blocks, and ball possession. in order to estimate the regression model proposed in this section, we use the least squares method. this study is conducted based on records provided by kfa from 2011 to 2019. kfa has announced that the online update of 2017-2019 season data is in progress. with the variable selection method commonly used in regression analysis [28], our variables that influence the goal difference (��(�)) of knf team are derived as shown in the following equation (� − *#�+, < 0.000, #�56 % = 0.61): ( ) * 3 6 152.613 0.507 0.115 0.248 0.016 i g e i i i d i d x x x= − + + + + (7) 172 advances in technology innovation, vol. 6, no. 3, 2021, pp. 169-178 where ��8 ∗ is the difference in the elo rating. the above regression equation shows that the variations in knf team’s score difference are statistically significant with shots (!�&), ground ball winning rates (!�$(), and elo ratings (��8 ∗). in addition, the positive results for the estimated coefficients of shots, ground ball winning rates, and elo ratings indicate that as the ranking of the opposing team decreases and the team’s performance increases, the score increases in relation to the knf team’s losses. moreover, the estimated regression equation shows that as the aggressiveness of knf team’s defense increases, the number of losses decreases. the variables used in this study are not standardized because they have different characteristics. as a result, the estimated regression model (eqs. (7) and (8)) show the intercept values. as shown in table 1, knf team’s performance can vary according to match importance such as world cup matches, world cup qualifiers, asian cup finals, and friendly matches. therefore, in this study, we choose to analyze the data on knf team’s world cup qualifier matches, which are held regularly (� − *#�+, < 0.000, #�56% = 0.877). ( ) 5 9 14 159.47 0.644 0.133 0.269 0.095 0.03 i g p i i i i d i g x x x x= − + − − + + (8) the estimated regression equation shows that the location of the match (;<8 ), pass success rates (!�$'), and ground ball winning rates (!�$() are important variables for knf team’s performance. in addition, the estimated regression equation indicates that a high number of fouls (!�() in a match negatively affect the team’s chances of victory because the korean team is one of the top teams in asia. this is the same for the number of corner kicks (!�4 ). 3.2. qualitative data for the model before a football match, the media and fans are interested in predicting the outcome. although football results do not seem easy to predict, prediction methods can be implemented by using variables that can affect the outcome of a match, including the ranking of the two teams, the match location, the atmosphere and teamwork of each team, the physical condition of players, athletic performance, and the importance of the match. the media provides important information about the game through a preview with football experts, and then a betting company sets a dividend on the victory, draw, and defeat of the match according to the customer’s bet. the winning rate of the football team can be predicted before the match based on the opinions of football experts and the information announced by the betting company [33-34]. in this study, we use expert columns and dividend rates published by the betting company, which both provide predictions and analyses of various sports in addition to football to predict the winning rate of knf team. the dividend rate for a match announced by a betting company cannot be unrelated to the outcome of the match [35-36]. the winning rate calculated by the dividends of winning (� ), drawing (�= ), and losing (�> ) for a team in a particular match are equal to: d l w d l w l w d d d p d d d d d d = + + (9) the above expression is given by odds-makers, who estimate the odds in betting or competing [37-38]. the sample correlation (0wg) between the winning rate of knf team and the difference between the scored and lost goal of knf team is: ( )0.616 0.000 wg p valueρ = − < (10) the regression equation for the winning rate of knf team (pw) based on knf team’s goal difference (gd) is as follows (#�56% = 0.374, � − *#�+, < 0.000): 0.436 0.06 w d p g= + (11) 173 advances in technology innovation, vol. 6, no. 3, 2021, pp. 169-178 this shows that the winning rate based on the dividends is related to the outcome of knf team. other important variables that can determine the outcomes of matches are the performances in previous matches, the current teamwork, the current athletic performance, the physical condition of the players competing in the game, and the importance of the game. the current atmosphere of the football team can change depending on the outcome of previous game results. when a team goes on a losing streak, the atmosphere of the team will deteriorate, and negative public opinions can cause significant changes, including the replacement of the coach. therefore, the outcome of the national team’s recent matches is an important factor in predicting the national team’s winning rate. based on the recent five matches, the current atmosphere of the national team is defined as: ( ) ( ) ( ) ( ) ( )1 2 3 4 50.3 0.3 0.2 0.1 0.1i s s s s s a i i i i i t e e e e e− − − − −= + + + + (12) where ,$@a b stands for the elo rating the knf team acquired at t matches before the current match. the importance of the match can affect its outcome. as with other team sports, the importance of the match determines the starting players in the match. while the rookie players are usually assigned in a squad of friendly matches, the best players are assigned for important matches that determine whether the team is qualified for a tournament. in this study, the importance of matches is classified into “very important”, “important”, “normal”, “ordinary”, and “very ordinary”. knf team’s athletic performance is another variable that can affect the outcome of a match. a team’s athletic performance can be evaluated by its ability to score, the ability to defend, and the organizational ability that expresses the organic movement of the defense, offense, and midfield. in addition, the physical strength and condition of the teams participating in the match are also the factors of the team’s performance. this study utilizes the columns of three experts who regularly analyze knf team’s performance and publish their opinions on team organization prior to knf matches. these columns enable us to tally all of the expert-derived positive and negative remarks regarding knf team’s offensive, defensive, and organizational abilities, as well as the strength and physical conditioning of knf team. team performance is expressed as one of the five categories: “very good”, “good”, “normal”, “bad”, or “very bad”. 3.3. fuzzy regression model for the winning rate of knf team in this study, the following fuzzy partitions are used to represent the winning rate of the football team as fuzzy numbers. (1) $: defeat with large margin, denoted by $ = (0, 0, 0.25) (2) %: defeat with small margin, denoted by % = (0, 0.25, 0.5) (3) &: draw, denoted by & = (0.25, 0.5, 0.75) (4) ': win with small margin, denoted by ' = (0.5, 0.75, 1) (5) (: win with large margin, denoted by ( = (0.75, 1, 1)� similar to the winning rate, we apply the following fuzzy partitions for the importance of the match. (1) �� $: very ordinary, denoted by �� $ = (0, 0, 0.25)� (2) �� %: ordinary, denoted by ��% = (0, 0.25, 0.5)� (3) �� &: normal, denoted by �� & = (0.25, 0.5, 0.75)� (4) �� ': important, denoted by �� ' = (0.5, 0.75, 1)� (5) �� (: very important, denoted by �� ( = (0.75, 1, 1)� 174 advances in technology innovation, vol. 6, no. 3, 2021, pp. 169-178 in this study, we also express knf team’s performance along with the winning rate and importance as fuzzy numbers. we present the national team’s performance in a combination of offensive ability, defensive ability, organizational ability, and physical conditioning evaluated by football experts. the overall competition performance of the four categories is equal to 2. thus, the universal set of the knf team’s ability to apply all four categories is equal to � = �0.5, 1.5 . fig. 3 shows the fuzzy division of a football team into five stages, including “very bad”, “bad”, “normal”, “good”, and “very good”. fig. 3 fuzzy partition for the national team’s performance we describe the national team’s performance as fuzzy numbers using the following divisions: (1) �d $: very bad, denoted by �d $ = (0.5, 0.5, 0.75)� (2) �d %: bad, denoted by �d % = (0.5, 0.75, 1)� (3) �d &: normal, denoted by �d & = (0.75, 1, 1.25)� (4) �d ': good, denoted by �d ' = (1, 1.25, 1.5)� (5) �d (: very good, denoted by �d ( = (1.25, 1.5, 1.5)� the football experts’ opinions are combined with the national team’s athletic ability and match importance. the betting company publishes the opinions of three or four experts on its website before national team [2, 34]. table 2 data of knf team number opponent location (��) goal difference (��) elo rating difference (,e) atmosphere (�d) athletic performance (�d) match importance (��) winning rate ( ) 1 lebanon home 6 627 0.6 �d ' �� ' ' 2 kuwait away 0 195 -0.5 �d & �� ' & ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ 26 croatia home -1 -166 -3.3 �d ' �� & & ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ 76 china away -1 271 -0.2 �d & �� ' & ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ ⋮ 117 lebanon away 0 371 -4.3 �d ( �� ' ( 118 brazil neutral -3 -320 -8.3 �d ' �� & $ the following is a summary of the expert opinions on a match against croatia in 2013: “south korea’s recent a-match performance has been bad. however, all of the players in good condition are going to be in the match, and we are expecting much from the europa league players in particular. this friendly match is an important test match ahead of the final qualifying round.” in addition, the expert opinions on the world cup asian qualifying rounds against china in 2017 are as follows: “the 175 advances in technology innovation, vol. 6, no. 3, 2021, pp. 169-178 korean team is currently ranked second in its group for the world cup asia qualifying round. while winning points in away matches in china is important, key players from the europa league are absent due to warnings and injuries. the opposing team, china, will put forth a strong effort because the new coach has yet to win his first victory.” we have compiled the national team’s data based on the expert opinions above. table 2 shows the quantitative data and the fuzzy data of knf team since september 2011. in table 2, the performance, the game importance, and the winning percentage are expressed in the triangle fuzzy numbers. the aim of this study is to infer the regression equation for the winning rate for the national team using the data given in table 2. the most common method is to estimate the fuzzy regression model using the least squares method. however, it may be more efficient to use the least absolute deviation method instead of the least squares method, which is sensitive to outliers. in this study, the three levels of least absolute deviation method proposed by choi and buckley [24] are used to estimate the fuzzy linear regression model for the winning rate of knf team given in table 2. * 1 2 3 4 5j j j j j jw g e a a g p c p c d c t c a c i= + + + + (13) step 1: consider the centroids of the fuzzy number. to find the center of support for the winning rate, the performance and match importance are expressed in the triangular fuzzy number. the triangular fuzzy number used in this study is symmetrical, so each fuzzy number mode becomes a centroid. step 2: estimate the regression coefficient based on the centroid data. the pseudo-estimate is estimated based on the centroid data [(�gh , ��h , �ih , �dh , �� h , ��h ): 5 = 1, … , 118 given in table 2 and the least absolute deviation method, which minimizes the following equations. * 1 2 3 4 5 ˆ ˆ ˆ ˆ ˆ j j j j j jw g e a a g p c p c d c t c a c i= + + + + (14) 118 * 1 2 3 4 5 1 j j j j j jw g e a a g j p c p c d c t c a c i = − − − − −∑ (15) step 3: estimate the spread of the fuzzy number. using the pseudo-estimate k obtained in the second step, the estimate of the winning rate for the knf team l k is as follows. * 1 2 3 4 5 ˆˆ ˆ ˆ ˆ ˆ ˆ ˆ( , 0, ) j j j j j jw g e a a g t p c p c d c t c a c i l r= + + + + + − (16) where �m and �̂ are, respectively, the values minimizing the following subject to � > 0 and � > 0: 118 * 1 2 3 4 5 1 ˆ ˆ ˆ ˆ ˆ w j j j a gj j j p g e a a i j l c p c d c t c l c l l = − − − − − +∑ (17) 118 * 1 2 3 4 5 j 1 ˆ ˆ ˆ ˆ ˆ w j j j a gj j j p g e a a i r c p c d c t c r c r r = − − − − − −∑ (18) the fuzzy regression model for the winning rate of knf team using the data given in table 2 and the three-stage absolute deviation method is: *0.078 0.066 0.039 0.288 0.239, 0.251ˆ ( ) j j j j jw g e c g t p p d a i= + + + + − (19) 176 advances in technology innovation, vol. 6, no. 3, 2021, pp. 169-178 the estimated model indicates that the match location (��), elo rating (,e), athletic performance (��), and match importance (��) affect the national team’s winning rate. the results are in line with the thoughts of both experts and fans. in other words, knf team has a high chance of winning in matches when its performance is good, the importance of match is high, and its elo rating is high. however, the atmosphere of knf team does not seem to have a significant impact on the knf team’s winning rate. we think this result requires further study. 4. conclusions in this study, we suggested the fuzzy regression model for the winning rate of the football team betting, and estimated the regression model of knf team using the data of 118 matches of knf team from 2011 to 2019. before the football match, the qualitative data on the basis of football experts’ reviews were collected to express knf team’s offensive, defensive, and organizational abilities, as well as the match importance, as a fuzzy number. we subsequently examined the impact of the fuzzy numbers defined by using the fuzzy partition and elo rating on the football team’s winning rate. it was found that the winning rate for knf team was influenced by the match location, elo rating, team performance, and match importance. consequently, as all of the coefficients are positive, it turned out that home court advantage increases the possibility of winning. the better the knf team’s performance reviewed in previous games is and the more important the upcoming match is, the higher the winning rate of knf team is expected. however, the atmosphere of knf team was not found to influence the match results. the results of 118 matches of knf team were analyzed with a limited amount of information in this study. however, in the future, it would be useful to extend this research to include more games and extrapolate a model of the winning rate depending on the match type, e.g., world cup matches, regional qualifiers, and friendly matches. acknowledgements the authors thank visual sports and wise toto for offering their data for this study. this work was supported by a national research foundation of korea (nrf) grant funded by the korean government (msit) (no. 2017r1d1a1b03029559). conflicts of interest the authors declare no conflicts of interest. references [1] “fifa,” http://www.fifa.com/index.html, july 05, 2019. [2] h. o. steller, d. sandor, and r. verlander, “issues in sports forecasting,” international journal of forecasting, vol. 26, no. 3, pp. 606-621, july-september 2010. [3] f. wunderlich and d. memmert, “analysis of the predictive qualities of betting odds and fifa world ranking: evidence from the 2006, 2010, and 2014 football world cups,” journal of sports sciences, vol. 34, no. 24, pp. 2176-2184, august 2016. [4] m. carpita, e. ciavolino, and p. pasca, “exploring and modelling team performances of the kaggle european soccer database,” statistical modelling, vol. 19, no. 1, pp. 74-101, february 2019. [5] y. f alfredo and s. m. isa, “football match prediction with tree based model classification,” international journal of intelligent systems and applications, vol. 11, no. 7, pp. 20-28, july 2019. [6] t. g. omomule, a. j. ibinuolapo, and o. o. ajayi, “fuzzy-based model for predicting football match results,” international journal of scientific research in computer science and engineering, vol. 8, no. 1, pp. 70-80, february 2020. [7] e. esme and m. s. kiran, “prediction of football match outcomes based on bookmakers odds by using k-nearest neighbor algorithm,” international journal of machine learning and computing, vol. 8, no. 1, pp. 26-32, february 2018. [8] j. m. pérez-sánchez, e. góme-déniz, and n. dávila-cárdenes, “a comparative study of logistic models using an asymmetric link: modelling the away victories in football,” symmetry, vol. 10, no. 6, june 2018. 177 advances in technology innovation, vol. 6, no. 3, 2021, pp. 169-178 [9] c. p. igiri and e. o. nwachukwu, “an improved prediction system for football a match result,” iosr journal of engineering, vol. 4, no. 12, pp. 12-20, december 2014. [10] h. liu, m. á. gomez, c. lago-peñas, and j. sampaio, “match statistics related to winning in the group stage of 2014 brazil fifa world cup,” journal of sports sciences, vol. 33, no. 12, pp. 1205-1213, march 2015. [11] “korea football association,” http://www.kfamatch.or.kr/svc/man/selectmaininfo.do, july 15, 2019. [12] j. goddard and i. asimakopoulos, “forecasting football results and the efficiency of fixed-odds betting,” journal of forecasting, vol. 23, no. 1, pp. 51-66, january 2004. [13] r. d. baker and i. g. mchale, “forecasting exact scores in national football league games,” international journal of forecasting, vol. 29, no. 1, pp. 122-130, january-march 2013. [14] g. boshnakov, t. kharrat, and i. g. mchale, “a bivariate weibull count model for forecasting association football scores,” international journal of forecasting, vol. 33, no. 2, pp. 458-466, april-june 2017. [15] m. j. dixon and m. e. robinson, “a birth process model for association football matches,” journal of the royal statistical society series d (the statistician), vol. 47, no. 3, pp. 523-538, september 1998. [16] j. h. kim, g. t. ro, j. s. park, and w. h. lee, “the development of soccer game win-lost prediction model using neural network analysis,” korean journal of sport science, vol. 18, no. 4, pp. 54-63, 2007. [17] a. c. constantinou, n. e. fenton, and m. neil, “profiting from an inefficient association football gambling market: prediction, risk and uncertainty using bayesian networks,” knowledge-based systems, vol. 50, pp. 60-86, september 2013. [18] h. rue and o. salvesen, “prediction and retrospective analysis of soccer matches in a league,” journal of the royal statistical society: series d (the statistician), vol. 49, no. 3, pp. 399-418, september 2000. [19] j. sargent and a. bedford, “improving australian football league player performance forecasts using optimized nonlinear smoothing,” international journal of forecasting, vol. 26, no. 3, pp. 489-497, july-september 2010. [20] s. h. choi and j. h. yoon, “general fuzzy regression using least squares method,” international journal of systems science, vol. 41, no. 5, pp. 477-485, march 2010. [21] h. y. jung, j. h. yoon, and s. h. choi, “fuzzy linear regression using rank transform method,” fuzzy sets and systems, vol. 274, pp. 97-108, september 2015. [22] h. tanaka, s. uejima, and k. asai, “linear regression analysis with fuzzy model,” ieee transactions on systems, man, and cybernetics, vol. 12, no. 6, pp. 903-907, 1982. [23] l. a. zadeh, “fuzzy sets,” information and control, vol. 8, no. 3, pp. 338-353, june 1965. [24] s. h. choi and j. j. buckley, “fuzzy regression using least absolute deviation estimators,” soft computing, vol. 12, no. 3, pp. 257-263, may 2007. [25] h. j. jhang, h. kwak, and s. h. choi, “analysis of the outcome for the korean pro-basketball games using regression model,” journal of the korean institute of intelligent systems, vol. 25, no. 5, pp. 489-494, october 2015. [26] l. a. zadeh, “the concept of a linguistic variable and its application to approximate reasoning—i,” information sciences, vol. 8, no. 3, pp. 199-249, 1975. [27] c. mencar, m. lucarelli, c. castiello, and f. a. maria, “design of strong fuzzy partitions from cuts,” 8th conference of the european society for fuzzy logic and technology, september 2013, pp. 424-431. [28] h. k. kim and s. h. choi, statistical analysis with applications, korea: kyung moon sa, 2002. (in korean) [29] j. lasek, z. szlávik, m. gagolewski, and s. bhulai, “how to improve a team’s position in the fifa ranking? a simulation study,” journal of applied statistics, vol. 43, no. 7, pp. 1349-1368, october 2016. [30] k. suzuki and k. ohmori, “effectiveness of fifa/coca-cola world ranking in predicting the results of fifa world cup finals,” football science, vol. 5, pp. 18-25, march 2008. [31] l. m. hvattum and h. arntzen, “using elo ratings for match result prediction in association football,” international journal of forecasting, vol. 26, no. 3, pp. 460-470, july-september 2010. [32] “wfer: the world football elo rating system,” http://www.eloratings.net, june 10, 2019. [33] “sportstoto,” http://www.sportstoto.co.kr/index.jsp, may 12, 2019. [34] “wisetoto,” http://www.wisetoto.com/index.htm, may 12, 2019. [35] “betman,” www.betman.co.kr, june 02, 2019. [36] “william hill,” http://sports.williamhill.com/bet/en-gb, april 04, 2019. [37] “chosunilbo,” http://premium.chosun.com/site/data/html_dir/2014/07/03/2014070300008.html, march 03, 2019. [38] m. j. dixon and p. f. pope, “the value of statistical forecasts in the uk association football betting market,” international journal of forecasting, vol. 20, no. 4, pp. 697-711, october-december 2004. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). 178  advances in technology innovation, vol. 3, no. 4, 2018, pp. 147 156 development of an automated measurement system for checking irradiation non-uniformity of solar simulators chen-huei hsieh * , xui-ze zh department of automation engineering & institute of mechatronoptic systems, chien-kuo technology university, changhua, taiwan, roc received 02 june 2017; received in revised from 13 june 2017; accepted 23 august 2017 abstract generally, for some solar simulator manufactures, the measurement of irradiation non-uniformity is made by placing a pv cell artificially and point by point on each measuring point on the solar module. it records each corresponding output voltage of pv cell. it is time-consuming and easy to produce measurement error. if either an xy table or a multi-axis robot is applied in this case, it is ot only expensive but also difficult to carry and setup. hence, an innovative and automated measurement system, which consists of both a self-developed smart robot and a monitoring and control system, is proposed in this paper to solve aforementioned problems. the monitoring and control system, i.e., constructed by a pc and a self-developed labview-based application program, plays the role of remotely commanding the operation mode of a robot, receiving the irradiance data being transmitted from the robot and calculating the irradiation non-uniformity. once the irradiation non-uniformity is calculated, some operations are made to adjust the irradiation parameters of the solar simulator such that the irradiation non-uniformity can meet the specification, i.e., 2%, 5%, 10% for class a, b or c of solar module, respectively. keywords: solar simulator, pv cell, solar module, irradiation non-uniformity, measurement 1. introduction recently, many works of literature have paid much attention to develop or design the solar simulator [1-6]. polly et al. [1] presented a dual source solar simulator. to test meteorological radiation, sun et al. [2] designed the optical system of the solar simulator. domínguez et al. [3] presented a solar simulator for measuring the performance of large area concentrator photovoltaic (cpv) modules. a new economical large-scale multiple-lamp solar simulator has been designed and constructed by meng et al. [4] to provide a test platform for the simulation of solar radiation at the earth's surface. a low-cost and portable solar simulator by using led (light-emitting diode) lamps has been proposed by kohraku and kurokawa [5]. besides, sabahi et al. [6] designed and constructed an efficient large-scale solar simulator to investigate solar thermal collectors. to measure the characteristic parameters, such as maximum power output [7], operating temperature [8] etc., of solar module more accurate, an efficient multiple-lamp solar simulator has been employed to irradiate the surface of the solar module. before that, the irradiation parameters of the solar simulator need to be checked in advance, such as irradiation non-uniformity, irradiation stability, and so on. the qualification of solar simulator was specified in iec standard 60904-9, and it specified that at least 64 points irradiance on the surface of a solar module is required to be measured. generally, the solar module consists of 6 x 10 or 6 x 12 pv (photovoltaic) cell and each of them is 200mm*200mm in dimension. before checking the irradiation non-uniformity, these 64 points irradiance need to be measured in advance. and then, the irradiation non-uniformity is checked based on these data according to a non-uniformity formula. * corresponding author. e-mail address: chhsieh@ctu.edu.tw tel.: +886-4-7111111 ext. 2416 http://ieeexplore.ieee.org/document/6316346/ advances in technology innovation, vol. 3, no. 4, 2018, pp. 147 156 copyright © taeti 148 the study of [9] pointed out that the most commonly used irradiance non-uniformity measurement methods were “mechanical scanning”, “optical tracing”, and “reference correction”, etc. however, they can not meet the requirement of measuring the non-uniformity of pulse simulator. in [9], a measurement system based on the “shutter method” has been proposed to accomplish an irradiance non-uniformity measurement of pulse simulator. for some solar simulator manufactures, to measure the irradiation non-uniformity, the measurement of irradiance on the irradiated surface of the solar module is made by placing a pv cell artificially and point by point on each measuring point on the surface of the solar module. and then, it records each corresponding output voltage of pv cell. it is time-consuming and easy to produce measurement error. in the way of “mechanical scanning”, if either an xy table or a multi-axis robot is applied to achieve this automated measurement process, it is not only expensive but also difficult to carry and setup. therefore, it is urgent to develop a portal and automated measurement system for checking irradiation non-uniformity of the solar simulator. the purpose of this paper is to propose an innovative and automated measurement system for checking the irradiation non-uniformity of the solar module. this paper is organized as follows. section 2 gives a profile of the proposed system hardware structure. section 3 describes the system software, including graphic user interface (gui), the pc program and the arduino program. in section 4, the experimental results are presented to demonstrate that the proposed scheme is capable of checking the irradiation non-uniformity of the solar module. conclusions are made in section 5. 2. profile of the system structure fig. 1 sketch of robot fig. 2 moving paths for robot the automated measurement system proposed in this paper consists of both a small and smart robot (320mm x 320mm x 75mm) and a monitoring and control pc. in a mechanism, the robot has been well designed for only moving along either the vertical or the horizontal path to avoid turning and to promote position ability. functionally, it is capable of loading a pv cell on its top and moving automatically on a sheet of poster paper which is covered on the surface of the to-be-measured solar module and laid out with several pre-designed tracking paths and position points for the robot. the designed sketch of the robot is shown in fig. 1, and the poster paper (192mm x 96mm) being laid out with tracking paths and position points are advances in technology innovation, vol. 3, no. 4, 2018, pp. 147 156 copyright © taeti 149 shown in fig. 2. in fig. 2, there are 6 horizontal paths and 12 vertical paths. the line-tracking sensor h and the line-tracking sensor v are, respectively, employed for the robot to move along the horizontal path and the vertical path. when the robot moves along a horizontal path, once the line-tracking sensor v senses the vertical path the robot stops for a moment and the irradiance on the surface of the solar module is acquired and stored in the eeprom of arduino mega [10]. the similar working principle is applicable to move along vertical paths for the robot. the structure of the proposed automated measurement system is shown in fig. 3, and serial peripheral interface (spi) signal between the rf module and the arduino mega/uno [10] is further shown in fig. 4. the pv cell detects the irradiance on the irradiated surface of the measuring location of the solar module and its output voltage is sent to an analog input (ai) of arduino mega. the control signals from the pc include operation mode, i.e., automation mode and manual mode, position location, start/restart measurement, start transmission, dc motor control and stepping motor control, can be transmitted through the rf modules located on both the pc side and the robot side. in fig. 1, two sets of dc motors including transmission gears are employed to drive four wheels for driving the robot to move along the horizontal direction, and the other two sets of dc motors including transmission gears are employed to drive four wheels for driving the robot to move along the vertical direction. moreover, two sets of stepping motors including transmission mechanism are employed to control the mechanism which switches either horizontal moving drive wheel or vertical moving drive wheel to contact the floor. fig. 5 reveals this switching effect. on the other side, the monitoring signal from arduino mega like irradiance data being acquired can be transmitted to the pc. since the rf module only has spi, the feedback data transmitted from arduino can not be transferred to pc. an arduino uno is used to bridge the spi of rf module and the usb of pc. spi rf module b arduino mega 2560 monitoring and control pc arduino uno spi usb rf module b push buttom/ switch line-tracking sensor ai di spi di motor driver motor driver dc motor*2 stepping motor*2 pv cell robot spi do do fig. 3 structure of the proposed system hardware rf module (si-4432) adruino mega/uno sdo sdi sck nsel nirq miso mosi sck ss int 0 fig. 4 spi between the rf module and the arduino mega/uno fig. 5 sketch of the robot advances in technology innovation, vol. 3, no. 4, 2018, pp. 147 156 copyright © taeti 150 the main facilities and modules being required for the proposed system are as follows. (1) one pc is used as monitoring and control station. (2) labview 2015 for the design of gui and the communication with the arduino mega. (3) one arduino uno connected with the pc via usb and with the rf module (si4432) via spi. (4) the robot controlled by arduino mega makes it move in horizontal and vertical on the top surface of solar module. (5) one rf module (si4432) connected with the arduino mega via spi. (6) one pv cell connected with arduino mega via ai. (7) two sets of line-tracking sensors connected with arduino mega via digital input (di) are used to sense the horizontal path and vertical path, respectively. (8) two sets of dc motor drivers connected with arduino mega via digital output (do) are used to drive four dc motors for driving the robot to move along horizontal and vertical direction, respectively. (9) two sets of stepping motor drivers connected with arduino mega via do are used to drive two stepping motors for switching either horizontal moving drive wheel or vertical moving drive wheel to contact with floor. 3. description of the system software the system software includes the monitoring and control program in pc and the control and communication program in arduino uno/mega. the monitoring and control program in pc includes gui and the program of commanding robot and receiving measured data from the robot via wireless communication. on the other hand, the robot is controlled by the arduino mega which includes control program and the program to communicate with the arduino uno via the rf module b. the arduino uno functionally acts as the bridge between the pc and the arduino mega. thus, the program in it is used to communicate with the pc via usb and communicate with the arduino mega via the rf module a. 3.1. description of gui the labview-based gui builds up the friendly human operation interface and the communication mechanism between the pc and the robot. from the monitoring and control panel in the pc, shown in fig. 6, the remote connection between the pc and the robot will be achieved immediately once the “start measurement” button is pressed. the gui can determine the operation mode of the robot, i.e., automation model and manual mode. fig. 6 monitoring and control panel in the pc 3.2. illustration of pc programming the programming of pc is responsible for transmitting the control signals to the arduino mega and also receiving data from the arduino mega. besides, the receiving data transferred from the arduino mega are also integrated into gui in pc. the main procedures in pc are briefly described as follows, and its flowchart is shown in fg. 7. advances in technology innovation, vol. 3, no. 4, 2018, pp. 147 156 copyright © taeti 151 start input measuring parameters input specified position transmite specicied position (row, col) move to specified position? receive the measured data from arduino mega end robot operates in automation mode receive all the measured data from arduino mega complete transmission? compute and display nonuniformity complete all the measured points? display the measured data restart measurement ? start measurement ? stop system? restart measurement ? connection ready? auto/manual mode? no yes no yes no yes no yes no no yes no yes yes automanual fig. 7 flowchart of the labview programming (a) set com port for the serial communication and determine the transmission rate, i.e., baud rate (bit per second), as 9600 that need to match the arduino uno transmission rate. (b) set the specification of solar module, i.e., the number of pv cells that a solar module includes. for example, 6 rows x 12 columns means that a solar module has 72 pv cells. (c) check if the wireless communication connection is ready. if yes, turn on “communication ready” indicator and go to the next step. otherwise, repeat step (c). (d) in this step, two operation branches work independently as follows. branch 1: manual mode (i) check if the manual mode is requested. if yes, go to the next step. otherwise, go back to step (d). (ii) set the specified position on the solar module, i.e., the row number and the column number. (iii) check if the “start measurement” button is pressed. if yes, transmit the manual command code and the specified position to the arduino uno and go to next step. otherwise, the same procedure is repeated. (iv) check if the specified position is reached. if yes, turn on the “position ready” indicator and go to the next step. otherwise, the same procedure is repeated. (v) receive the measured data from the arduino uno. (vi) check if the measured data is received. if yes, display the measured data and go to the next step. otherwise, go to step (v). advances in technology innovation, vol. 3, no. 4, 2018, pp. 147 156 copyright © taeti 152 (vii) check if the “restart measurement” button is pressed. if yes, go to the next step (ii). otherwise, go to step (e). branch 2: automation mode (i) check if the automation mode is requested. if yes, go to the next step. otherwise, go back to step (d). (ii) check if the “start measurement” button is pressed. if yes, transmit automation command code to the arduino uno. otherwise, the same procedure is repeated. (iii) check if the measurement of all the measuring points is completed. if yes, turn on “measurement completion” indicator and go to the next step. otherwise, the same procedure is repeated. (iv) check if the “start transmission” button is pressed. if yes, go to the next step. otherwise, the same procedure is repeated. (v) receive all the measured data, i.e., the irradiance on the solar module, from the arduino uno. (vi) check if all the measured data are received. if yes, turn on “transmission completion” indicator go to the next step. otherwise, go to step (v). (vii) display all the measured data on gui, compute irradiation non-uniformity and display it, and save the data measured at this time in a file. the formula to compute irradiation non-uniformity is described as follows.          irradianceirradiance irradianceirradiance minmax minmax uniformity-non (1) (e) check if the “stop” button, i.e., the button to stop system, is pressed. if yes, stop the measurement system. otherwise, go to step (b) 3.3. illustration of the arduino programming there are programs in both the arduino mega and the arduino uno need to be designed and their illustration are, respectively, illustrated as follows: (1) illustration of the arduino uno programming the arduino uno is employed to bridge the communication between the rf module with spi interface and the pc with a usb interface. the programming flowchart is shown in fig. 8. the main procedures are briefly described as follows. (a) set the transmission rate, i.e., baud rate (bit per second), as 9600 that need to match the pc transmission rate. (b) check if wireless communication connection is ready. if yes, turn on “communication ready” indicator and go to the next step. otherwise, the same procedure is repeated. (c) receive command code (sent from pc) and transmit it to the arduino mega. (d) check if the “start transmission” flag is activated. if yes, receive the measured data from the arduino mega and transmit them to pc and go to the next step. otherwise, the same procedure is repeated. (e) check if the “restart measurement” flag is activated. if yes, go to step (b). otherwise, the same procedure is repeated. (2) illustration of the arduino mega programming the arduino mega is employed to control the robot for executing manual or automation operation according to the command code transmitted from pc, to acquire the output voltage of pv cell via ai, and to feedback the acquired data to pc. the programming flowchart is shown in fig. 9. the main procedures are briefly described as follows. (a) check if the wireless communication connection is ready. if yes, turn on “communication ready” indicator and go to the next step. otherwise, the same procedure is repeated. (b) compare if the command code is manual or automated. two operation branches work independently as follows. advances in technology innovation, vol. 3, no. 4, 2018, pp. 147 156 copyright © taeti 153 (c) check if the “start measurement” flag is activated. if yes, go to the next step. otherwise, the same procedure is repeated. branch 1: manual mode (i) control the robot to move to the specified measuring point. (ii) acquire the output voltage of pv cell, i.e., the irradiance on the solar module, via ai and transmit the acquired data to pc. (iii) go to step (a). branch 2: automation mode (i) acquire the output voltage of pv cell i.e., the irradiance on the solar module, via ai. (ii) control the robot to move to the next measuring point that is the required point belongs to the programmed path in automation operation. (iii) store the acquired data in the eeprom of the arduino mega. (iv) check if the measurement of all the measuring points is completed. if yes, transmit all the acquired data stored in eeprom to the pc and go to step (a). otherwise, go to step (i). start connection ready? receive command code transmit command code to arduino mega start transmission? receive measured data from arduino mega and transmit them to pc restart measurement yes no yes no yes no compare command code? position to the specified measuring point store in eeprom acquire pv cell output voltage via ai acquire pv cell output voltage via ai position to next measuring point transmit the acquisition data to arduino uno complete measurement? start transmission? transmit the acquisition data to arduino uno no no yes yes manual auto start set start point row 1, col 1 connection ready? yes no fig. 8 flowchart of the arduino uno programming fig. 9 flowchart of the arduino mega programming advances in technology innovation, vol. 3, no. 4, 2018, pp. 147 156 copyright © taeti 154 4. experimental results and discussion the experiment of this new approach is implemented in a chamber which is equipped mainly with a solar simulator used to irradiate the solar module and a platform used to load a solar module. this chamber is shown in fig. 10. fig. 10 a chamber (1) connect pc to the robot the first step to implement the proposed system is to connect pc to the robot. once the labview-based gui is activated, shown in fig. 6, the connection will be operated immediately whenever the arduino mega is standby. fig. 11 indicates the implementation status after completing this step. fig. 11 setup of the system connection (2) implement automation operation the implementation of the automation operation, shown in fig. 12, shows the measured data, i.e., the irradiance on the solar module, the maximum and minimum value of them and the irradiation non-uniformity. fig. 12 operational results of automation mode advances in technology innovation, vol. 3, no. 4, 2018, pp. 147 156 copyright © taeti 155 (3) implement manual operation the implementation of the manual operation is shown in fig. 13. it shows the number of pv cells, which a solar module has, the specified measuring point (row number and column number) and the measured data from that point are shown in “receive data” field. fig. 13 operational results of manual mode (4) measuring time the experimental results about the measuring time and the moving velocity which the robot operates in automation mode are shown in table 1. generally, if the measurement of irradiation non-uniformity is made by placing a pv cell artificially and point by point on each measuring point on the solar module, each measuring point takes 20 seconds in average to measure. thus, it will totally need 1440 seconds to complete measuring one solar module which is 6 rows x 12 columns in the specification. the measuring time obtained by using the proposed system is significantly improved as compared with that by artificial measurement. table 1 measuring time and moving velocity specification of solar module measuring time (sec) moving velocity (mm/sec) 6 rows x 10 columns 210 56.47 6 rows x 12 columns 240 56.03 5. conclusions in this paper, an innovative and automated measurement system for checking the irradiation non-uniformity of the solar simulator is proposed. the feasibility of this new approach has been demonstrated by implementing the experiment in a chamber with a solar simulator and a test platform. furthermore, the experimental results have shown that the proposed system has merits of meeting the miniaturization and portability demands, low setup cost and higher measuring efficiency. references [1] s. j. polly, z. s. bittner, m. f. bennett, r. p. raffaelle, and s. m. hubbard, “development of a multi-source solar simulator for spatial uniformity and close spectral matching to am0 and am1.5,” ieee international conf. on photovoltaic specialists, ieee press, june 2011, pp. 1739-1743. [2] m. sun, g. zhang, b. yang, x. tao, and g. ding, “optical system design of solar simulator for testing meteorological radiation instrument,” international conf. on optoelectronics and microelectronics, ieee press, october 2012, pp. 600 603. [3] c. domínguez, i. antón, and g. sala, “solar simulator for concentrator photovoltaic systems,” optics express, vol. 16, no. 19, pp. 14894-14901, 2008. [4] q. meng, y. wang, and l. zhang, “irradiance characteristics and optimization design of a large-scale solar simulator,” solar energy, vol. 85, no. 9, pp. 1758-1767, september 2011. [5] s. kohraku and k. kurokawa, “new methods for solar cells measurement by led solar simulator,” proc. of 3rd world conf. on photovoltaic energy conversion, ieee press, may 2003, pp. 4-7. http://ieeexplore.ieee.org/search/searchresult.jsp?searchwithin=%22authors%22:.qt.stephen%20j.%20polly.qt.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchwithin=%22authors%22:.qt.zachary%20s.%20bittner.qt.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchwithin=%22authors%22:.qt.mitch%20f.%20bennett.qt.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchwithin=%22authors%22:.qt.ryne%20p.%20raffaelle.qt.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchwithin=%22authors%22:.qt.seth%20m.%20hubbard.qt.&newsearch=true http://ieeexplore.ieee.org/document/6186290/ http://ieeexplore.ieee.org/document/6186290/ http://ieeexplore.ieee.org/xpl/mostrecentissue.jsp?punumber=6177424 http://ieeexplore.ieee.org/xpl/mostrecentissue.jsp?punumber=6177424 http://ieeexplore.ieee.org/search/searchresult.jsp?searchwithin=%22authors%22:.qt.mingjiao%20sun.qt.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchwithin=%22authors%22:.qt.guoyu%20zhang.qt.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchwithin=%22authors%22:.qt.bin%20yang.qt.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchwithin=%22authors%22:.qt.xue%20tao.qt.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchwithin=%22authors%22:.qt.guipeng%20ding.qt.&newsearch=true http://ieeexplore.ieee.org/document/6316346/ http://ieeexplore.ieee.org/document/6316346/ http://ieeexplore.ieee.org/xpl/mostrecentissue.jsp?punumber=6296429 advances in technology innovation, vol. 3, no. 4, 2018, pp. 147 156 copyright © taeti 156 [6] h. sabahi, a. a. tofigh, i. m. kakhki, and h. bungypoor-fard, “design, construction and performance test of an efficient large-scale solar simulator for investigation of solar thermal collectors,” sustainable energy technologies and assessments, vol. 15, pp. 35-41, june 2016. [7] j. dubard, j. r. filtz, v. cassagne, and p. legrain, “photovoltaic module performance measurements traceability: uncertainties survey,” measurement, vol. 51, pp. 451-456, may 2014. [8] r. eke, a. s. kavasoglu, and n. kavasoglu, “design and implementation of a low-cost multi-channel temperature measurement system for photovoltaic modules,” measurement, vol. 45, no. 6, pp. 1499-1509, june 2012. [9] y. yuan, y. yang, and z. ya, “research of solar simulator irradiance non-uniformity measurement,” ieee international conf. on electronic measurement & instruments, ieee press, october 2011, pp. 307-310. [10] “arduino mega 2560 rev3,” products retrieved from https://store.arduino.cc/usa/arduino-mega-2560-rev3, june 1, 2017. http://ieeexplore.ieee.org/search/searchresult.jsp?searchwithin=%22authors%22:.qt.yafei%20yuan.qt.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchwithin=%22authors%22:.qt.yiqiang%20yang.qt.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchwithin=%22authors%22:.qt.zhang%20ya.qt.&newsearch=true http://ieeexplore.ieee.org/xpl/mostrecentissue.jsp?punumber=6030004 http://ieeexplore.ieee.org/xpl/mostrecentissue.jsp?punumber=6030004 https://store.arduino.cc/usa/arduino-mega-2560-rev3  advancesin technology innovation, vol. 6, no. 3, 2021, pp. 179-190 an experimental study of plastic waste as fine aggregate substitute for environmentally friendly concrete anita intan nura diana 1,* , subaidillah fansuri 1 , akhmad feri fatoni 2 1 department of civil engineering, wiraraja university, sumenep, indonesia 2 department of nursing, wiraraja university, sumenep, indonesia received 29 december 2020; received in revised form 06 april 2021; accepted 07 april 2021 doi: https://doi.org/10.46604/aiti.2021.6930 abstract decomposing plastics, including plastic bottles, is a very difficult process because it takes 50-100 years. every year, the use of plastic bottles is increasing, but only few people are willing to treat plastic bottle waste. in this study, plastic bottle waste is used as a substitute of fine aggregate and shaped in such a way to have a sand-like gradation. the variations of graded plastic bottle waste are 0%, 5%, 10%, and 12%. the test objects for each variation consist of three specimens. data are analyzed by using regression and classical assumption test with spss program. the results of the data analysis show that there is a simultaneous effect on the compressive strength with variations in plastic waste substitution. the compressive strength decreases with the increase in the percentage of plastic added. maximum compressive strength is at the variations of 0% and 5% with19.192 mpa and 16.414 mpa, respectively. keywords: environmentally friendly concrete, plastic waste, compressive strength 1. introduction plastic bottle waste is one of the most common environmental problems in this era. plastic is a material that is difficult to decompose, taking up to 50-100 years. every year, the amount of plastic bottle waste is increasing because people use plastic-based packaging almost every day. nonetheless, people are less aware that behind the use of plastic materials, the recycling process of plastic bottle waste is not optimal. some people recycle plastic bottle waste by converting them into vases, bags, and furniture so that it can be reused. however, the results of the survey conducted in sumenep area regarding the recycling process of plastic bottle waste are still limited. in the field of structure and materials, there have been many studies using plastic bottle waste as a research subject. using plastic bottle waste as construction materials can reduce environmental damages caused by plastic waste pollution. in this study, plastic bottle waste is used as a substitute of fine aggregate and shaped in such a way so that its gradation is similar to the common fine aggregate, i.e. the sand. sand is the main ingredient for making concrete. the sand used in the concrete mixture is black sand, which is more suitable than other types of sand for building construction. the black sand content has a good binder with other concrete mixture materials. behind the excellent use of black sand in building construction, a very serious problem arises. land damage has often occurred due to uncontrolled mining of black sand (illegal mining) around the sand mining area. in the last few years, there have been conflicts that claimed the lives of environmentalists in the mining areas, including lumajang area. every day, black sand mining is carried out in the lumajang area to meet consumer demand without any restrictions on black sand mining which will have a negative impact on the surrounding environment. * corresponding author. e-mail address: anita@wiraraja.ac.id tel.: +62-81332299841 advancesin technology innovation, vol. 6, no. 3, 2021, pp. 179-190 sumenep does not have black sand materials for infrastructure purposes. the black sand in sumenep regency is imported from outside the city. therefore, it is not surprising that the price of black sand in sumenep regency is quite high since the demand is high. as a result, people in sumenep regency have started many private companies and agencies that prioritize in building materials for fine aggregate using black sand. various studies have been carried out on utilizing plastic waste as a substitute of fine aggregate [1-11]. guendouz et al. [1] investigated the use of two types of plastic scrap with the pet and ldpe logos used for the manufacture of fine aggregate in the form of powder in concrete z. diana and depriyanto [2] examined the effect of adding plastic fibers to paving blocks on compressive strength, shock resistance, and water absorption. karimah [3] studied the compressive strength of normal strength concrete by adding 15%, 30%, and 45% ldpe plastic waste. alvine [4] investigated the compressive strength of concrete by adding 20%, 40%, and 60% abs plastic waste. previous research combined plastic waste with other materials as a substitute for concrete mixtures. among these materials are electronic plastic waste and marble dust on hardened properties of high-strength concrete [9] and plastic/rubber waste as environmentally friendly aggregates for concrete mixtures [10]. adela, behanu, and gobena [11] utilized plastic waste as an alternative material for coarse aggregates. several studies have shown that adding a number of plastic waste to the concrete mixture can reduce the quality of the concrete production [9-11]. based on the previous research, further research on the use of plastic waste for concrete mixture is needed. research can be in the form of variations in processing plastic waste and the addition of additives. table 1 literature review ref. plastic type shape replacement level aggregate replacement type application [1] pet + ldpe fibers and powder powder: 10%, 20%, 30%, and 40% fibers: 0.5%, 1%, 1.5%, and 2% fine concrete [2] pet fibers 0%, 0.25%, 0.50%, 0.75%, and 1% coarse paving block [3] ldpe flakes 15%, 30%, and 45% coarse concrete [4] abs flakes and powder 20%, 40%, and 60% coarse and fine concrete [5] e-plastic waste flakes 0%, 5.5%, 11%, and 16.5% coarse concrete [6] ldpe powder particles below 75 μm fine [7] ldpe shredded 5%, 10%, 15%, 20%, and 25% fine concrete [8] ldpe powder 15% and 30% fine concrete [9] e-plastic waste irregular flakes 0%, 10%, 20%, 30%, and 40% coarse and fine concrete [11] plastic bags and plastic bottles irregular 0%, 25%, 35%, and 50% coarse concrete table 1 shows several types and shape of plastic waste that have been used in previous studies. based on the background discussion above, only few researchers in sumenep district utilized plastic waste as substitute of fine aggregate in concrete mixture.this study focuses on the use of graded plastic bottle waste (the plastic type is pet) as a substitute of fine aggregate, and uses the local coarse aggregates originating from duko village, rubaru district, and sumenep regency based on the results of research conducted by fansuri et.al [12]. the use of local materials is intended as a form of sustainability of construction materials by local construction actors in sumenep regency. 2. research methodology 2.1. flow chart of the study fig. 1 shows the flow chart of the approach for this study. the process consists of five steps: problem statement, data collection, analysis data, result and discussion, and conclusions.the five steps are described as follows: 180 advancesin technology innovation, vol. 6, no. 3, 2021, pp. 179-190 step 1: problem statement the most important background is the plastic bottle waste problem that we face in this era. plastic is a material that is difficult to decompose, which takes 50-100 years to decompose. every year the amount of plastic bottle waste is increasing because almost every day people use plastic-based packaging. it takes innovation to turn plastic waste into construction materials (see introductory section for detailed explanations). step 2: data collection data are divided into primary data and secondary data. primary data are collected through experiments in the civil engineering laboratory by testing the aggregate materials for their density, water content, and sieve analysis, by making fresh concrete started with the calculation of mix design, manufacture of test object, slump test, concrete molding, and concrete curing, and by testing the compressive strength of fresh concrete. secondary data are obtained by reviewing similar previous research and by testing the standards used in the study. the standards include the specific gravity test, water absorption of fine and coarse aggregates, fine aggregate and coarse aggregate sieve analysis testing, concrete manufacturing procedures, and concrete compressive strength testing. step 3: data analysis primary data are especially analyzed by using classical assumption tests (linearity test, normality test, and heteroscedasticity test) and linier regression. all the analyses in the study are executed by using spss software. step 4: result and discussion at this stage, all the experimental and analytical results will be displayed in the form of tables, figures, and detailed calculations. then, the obtained experimental results and analysis are compared with the previous literature. step 5: conclusion at this stage, conclusions and recommendations will be obtained from the results of the experiments and analyses that have been carried out. fig. 1 research methodology 2.2. model the model used in this study is fresh concrete with a mixture of plastic bottle waste. plastic bottles are cut into small pieces to resemble grading sand. furthermore, the small pieces of plastic waste used as a substitute for sand for the concrete mixture are divided into several variations. there are three concrete specimens as samples for each variation. the concrete compressive strength test is carried out for 14 days. table 2 presents the test objects for each substitute variation. problem statement data collection primary data secondary data analysis data results and discussion conclusions 181 advancesin technology innovation, vol. 6, no. 3, 2021, pp. 179-190 table 2 plastic gradient variation for each specimen plastic gradient variations specimen for each variation 0% 3 5% 3 10% 3 12% 3 total samples 12 2.3. data analysis the data to be analyzed are obtained from the results of testing in the laboratory. data from the laboratory are presented in tables and graphs. data are analyzed using linear regression with classical assumption tests (linearity test, normality test, and heteroscedasticity test). all the analyses use spss program which are presented in the form of tables, pictures, and descriptions. 3. results and discussion the results of this study consist of several test results for concrete mixtures (fine and coarse aggregate). after concrete compressive strength tests are carried out at the civil engineering laboratory, compressive strength test data are analyzed using spss software. the results of the study are described below. 3.1. fine aggregate experiments there are several experiments carried out in the laboratory to determine the quality of fine aggregate, including water content experiment, specific gravity experiment [13], and sieve analysis experiment [14]. the following is the result of the water experiments in the laboratory. table 3 data testing and calculation of fine aggregate water content experiment 1 2 3 weight of original fine aggregate (w1) 500 gr 500 gr 500 gr weight of oven-dry fine aggregate (w2) 485 gr 487 gr 486 gr water content of fine aggregate (w1 − w2) / w2 × 100% 3.52% 2.67% 2.88% average of water content 3.023% from the test shown in table 3, it can be seen that the average value of fine aggregate water content is 3.023%. the greater the difference between the original weight of the fine aggregate and the weight of the fine aggregate after being heated, the more water the aggregate will contain. from the test shown in table 4, it can be seen that dry specific gravity is 2.56, saturated surface dry (ssd) is 2.62, apparent specific gravity is 2.71, and absorption is 2%. detailed calculations can be seen in table 4. table 4 data testing and calculation of fine aggregate density experiment 1 2 3 average weight of pycno + ssd fine aggregate + water (w1) 1005 1019 1017 weight of ssd fine aggregate (500 gr) 500 500 500 weight of pycno + water (w2) 705 705 705 weight of oven-dry fine aggregate (w3) 490 490 490 dry specific gravity = w3 / (w2 + 500 − w1) 2.45 2.63 2.60 2.56 ssd specific gravity = 500 / (w2 + 500 − w1) 2.50 2.69 2.66 2.62 apparent specific gravity = w3 / (w2 + w3 − w1) 2.58 2.78 2.75 2.71 absorption = [(500 − w3) / 500] × 100% 2% 2% 2% 2% 182 advancesin technology innovation, vol. 6, no. 3, 2021, pp. 179-190 figs. 2-5 illustrate the fine aggregate grading zone graphs. the value to illustrate the graph of the fine aggregate zone is obtained from table 5. from the fine aggregate gradation graphs, it can be seen that the fine aggregate used is at the gradation limit no. 3. furthermore, the amount of sand replaced by plastic bottle waste adjusts the selected gradation. it is done so that the plastic grading is similar to the sand grading used. the materials used in this study are shown in fig. 6. table 5 data testing and calculation of fine aggregate sieve analysis filter size restrained weight (gr) cumulative restrained weight (gr) cumulative restrained (%) cumulative passes (%) mm inch 4.8 3/8 30 30 3 96.95 2.36 8 46 76 7.72 92.28 1.70 12 60 136 13.81 86.19 1.18 16 73 209 21.22 78.78 0.60 30 210 419 42.54 57.46 0.425 40 160 579 58.78 41.22 0.30 50 42 621 63.05 36.95 0.15 100 301 922 93.60 6.40 0.075 200 48 970 98.48 1.52 pan 15 985 100 0 total 985 0 20 40 60 80 100 0.01 0.1 1 10 % c u m m u la ti v e filter size lower limit upper limit % cummulative passes fig. 2 fine aggregate gradation limit no. 1 0 20 40 60 80 100 0.01 0.1 1 10 % c u m m u la ti v e filter size lower limit upper limit % cummulative passes fig. 3 fine aggregate gradation limit no. 2 0 20 40 60 80 100 0.01 0.1 1 10 % c u m m u la ti v e filter size lower limit upper limit % cummulative passes fig. 4 fine aggregate gradation limit no. 3 183 advancesin technology innovation, vol. 6, no. 3, 2021, pp. 179-190 0 20 40 60 80 100 0.01 0.1 1 10 % c u m m u la ti v e filter size lower limit upper limit % cummulative passes fig. 5 fine aggregate gradation limit no. 4 (a) sand (b) plastic bottle waste (c) graded plastic fig. 6 materials used in this study 3.2. coarse aggregate experiment there are several experiments conducted in the laboratory to determine the quality of the local coarse aggregate in duko village, rubaru district [12]. the experiments include sieve analysis experiment [14] and water content experiment, specific gravity experiment [15]. the followings are the results of the experiments. table 6 data testing and calculation of coarse aggregate water content experiment 1 2 3 weight of original coarse aggregate (w1) 500 gr 500 gr 500 gr weight of oven-dry coarse aggregate (w2) 401 gr 401 gr 401 gr water content of coarse aggregate (w1 − w2) / w2 × 100% 24.7% 24.7% 24.7% average of water content 24.7% from the test shown in table 6, it can be seen that the average of coarse aggregate water content value is 24.7%. the greater the difference between the original weight of the fine aggregate and the weight of the fine aggregate after being heated, the more water the aggregate will contain. from the test shown in table 7 below, it can be seen that the dry specific gravity is 2.02, ssd specific gravity is 2.52, apparent specific gravity is 4.03, and absorption is 24.69%. table 7 data testing and calculation of coarse aggregate density experiment 1 2 3 average weight of the test specimen in saturated surface dry condition (w1) 500 500 500 bucket weight in water (w2) 631 631 631 bucket weight + specimen in water (w3) 930 935 932 weight of oven-dry coarse aggregate (w4) 401 401 401 dry specific gravity = w4 / (w2 + w1 − w3) 2.00 2.05 2.02 2.02 ssd specific gravity = w1 / (w2 + w1 − w3) 2.49 2.55 2.51 2.52 apparent specific gravity= w4 / (w2 + w4 − w3) 3.93 4.13 4.01 4.03 absorption = [(w1 − w4) / w4] × 100% 24.69% 24.69% 24.69% 24.69% 184 advancesin technology innovation, vol. 6, no. 3, 2021, pp. 179-190 figs. 7-9 illustrate the coarse aggregate grading zone graphs. the value to illustrate the graph of the coarse aggregate zone is obtained from table 8. based on the experimental results and gradation limit of coarse aggregate sieve analysis, it is obtained that the coarse aggregate gradation is in the coarse aggregate gradation with a maximum size of 40 mm. table 8 data testing and calculation of coarse aggregate sieve analysis filter size restrained weight (gr) cumulative restrained weight (gr) cumulative restrained (%) cumulative passes (%) mm inch 76.20 3 0 0 0 100 50.8 2 0 0 0 100 38.1 1.5 0 0 0 100 25.40 1 318 318 35.14 64.86 19 3/4 235 553 61.10 38.90 13.20 1/2 70 623 68.84 31.16 8.5 3/8 110 733 80.99 19.01 4.75 #4 160 893 98.67 1.33 2.36 #8 12 905 100 0 0.15 100 0 905 100 0 pan 0 905 100 0 total 1000 0 20 40 60 80 100 1 10 100 % c u m m u la ti v e filter size lower limit upper limit % cummulative passes fig. 7 coarse aggregate gradation limit (10 mm) 0 20 40 60 80 100 1 10 100 % c u m m u la ti v e filter size lower limit upper limit % cummulative passes fig. 8 coarse aggregate gradation limit (20 mm) 0 20 40 60 80 100 1 10 100 % c u m m u la ti v e filter size lower limit upper limit % cummulative passes fig. 9 coarse aggregate gradation limit (40 mm) 185 advancesin technology innovation, vol. 6, no. 3, 2021, pp. 179-190 3.3. concrete mix design planning mix design planning is an important stage in concrete planning. based on the mix design planning, the composition of coarse aggregate, fine aggregate, cement, and water can be obtained. in this study, mix design planning uses the indonesian national standard (sni) 03-2834 [16]. data of concrete mixture material are used to create three specimens for each variation, providing the required compressive strength in 23 mpa equivalents to k270. concrete mixture data for each variation is shown in table 9. table 9 data of concrete mixture material concrete mixture material variation 0% (kg) variation 5% (kg) variation 10% (kg) variation 12% (kg) total (kg) cement 3.43 3.43 3.43 3.43 13.72 water 2.035 2.035 2.035 2.035 8.14 coarse aggregate 12.571 12.571 12.571 12.571 50.284 fine aggregate 8.109 7.704 7.298 7.136 30.247 plastic bottle waste 0.405 0.811 0.973 2.189 3.4. concrete slump testing in this study, the stages and calculations of concrete slump testing use the procedure listed in the indonesian national standard (sni) 1972 [17]. fig. 10 shows the tool used in the slum test, namely the abram cone. the test tool must be a mold made of a metal material that is not sticky and does not react with cement paste. the thickness of the metal shall not be less than 1.5 mm, and when it is formed by a spinning process, there shall be no dots in the mold and its thickness is smaller than 1.15 mm. the mold must be cone-shaped with a base diameter of 203 mm, top diameter of 102 mm, and height of 305 mm. the base and top faces of the cone must be open and parallel to each other and perpendicular to the axis of the cone. the mold must be equipped with a treadle section and a handle as shown in fig. 10. fig. 10 mold for slump test (abram cone) table 10 data testing and calculation of slump value plastic waste variation mold of slump test height (cm) height of each specimen (cm) slump value for each specimen (cm) average (cm) 1 2 1 2 0% 30 19 21 11 9 10 5% 30 21 21 9 9 9 10% 30 20 22 10 8 9 12% 30 22 21 8 9 8.5 based on table 10, which shows the slump plan set in 60-180 mm, the results of the laboratory experiments meet the requirements. the lowest average value of the slump test is 85 mm and the highest value is 100 mm. this value is within the planned slump value. 186 advancesin technology innovation, vol. 6, no. 3, 2021, pp. 179-190 3.5. concrete compressive strength this concrete compressive strength test is carried out according to the procedure in the indonesian national standard (sni) 1974 [18]. the detailed results of the concrete compressive strength test can be seen in tables 11-14. table 11 data testing and calculation of concrete compressive strength value in 0% variation sample date weight (kg) p (n) a (mm 2 ) age conversion (14 days) compressive strength (f’c) mpa characteristic concrete (k) kg/cm 2 create test 1 06/05/2020 20/05/2020 8.143 340000 22500 0.88 17.172 206.888 2 06/05/2020 20/05/2020 8.064 395000 22500 0.88 19.949 240.355 3 06/05/2020 20/05/2020 8.083 405000 22500 0.88 20.455 246.440 average of variation: 0% 19.192 231.228 table 12 data testing and calculation of concrete compressive strength value in 5% variation sample date weight (kg) p (n) a (mm 2 ) age conversion (14 days) compressive strength (f’c) mpa characteristic concrete (k) kg/cm 2 create test 1 12/05/2020 26/05/2020 7.926 410000 22500 0.88 20.707 249.483 2 12/05/2020 26/05/2020 7.615 360000 22500 0.88 18.182 219.058 3 12/05/2020 26/05/2020 8.058 205000 22500 0.88 10.354 124.741 average of variation: 5% 16.414 197.761 table 13 data testing and calculation of concrete compressive strength value in 10% variation sample date weight (kg) p (n) a (mm 2 ) age conversion (14 days) compressive strength (f’c) mpa characteristic concrete (k) kg/cm 2 create test 1 18/05/2020 01/06/2020 7.240 240000 22500 0.88 12.121 146.039 2 18/05/2020 01/06/2020 7.200 185000 22500 0.88 9.343 112.571 3 18/05/2020 01/06/2020 7.160 200000 22500 0.88 10.101 121.699 average of variation: 10% 10.522 126.770 table 14 data testing and calculation of concrete compressive strength value in 12% variation sample date weight (kg) p (n) a (mm 2 ) age conversion (14 days) compressive strength (f’c) mpa characteristic concrete (k) kg/cm 2 create test 1 18/05/2020 01/06/2020 7.284 215000 22500 0.88 10.859 130.826 2 18/05/2020 01/06/2020 7.140 210000 22500 0.88 10.606 127.784 3 18/05/2020 01/06/2020 7.247 200000 22500 0.88 10.101 121.699 average of variation: 12% 10.522 126.770 fig. 11 shows the relationship between the variations of plastic waste substitution in the fine aggregate of concrete mixture and the result of compressive strength value. the highest compressive strength is shown by concrete without plastic waste substitution in the concrete mixture (19.19 mpa in 0% variation). there is a decrease in compressive strength that is in line with the increase in the amount of plastic waste added to the concrete mixture. all the results of the compressive strength of the concrete, none of them meets the compressive strength requirements in 23 mpa. there are several factors that cause failure to pass in this experiment. first, the coarse aggregate material used do not meet the overall sni requirements. the aggregate material in this study uses the local aggregate in sumenep area. in a study conducted by fansuri et al. [12] in five local coarse aggregate mining locations, namely batuan village, batu putih village, dasuk village, duko village, and ellak daya village, it can be seen that the aggregate in duko village met the parameters of coarse aggregate as a concrete material, where the water absorption was 1.83% (maximum value is 3%), specific gravity was 187 advancesin technology innovation, vol. 6, no. 3, 2021, pp. 179-190 2.57 (minimum value is 1.8), test value wear and tear using los angeles was 24.8% (maximum value is 40%), and volume weight was 2075.5 kg/m 3 (minimum value is 2200 kg/m 3 ). second, several studies that have been done previously explained that the use of additives in the form of additives (chemical compounds) is needed as the liquid can add strength to concrete with variations of plastic waste. meanwhile, the experiments conducted in this study do not use added materials in the form of additives (chemical compounds). 0 5 10 15 20 25 0% 5% 10% 15% c o m p re s s iv e s tr e n g th v a lu e ( m p a ) variation of plastic waste fig. 11 compressive strength relationship and plastic waste variations 3.6. normality test of concrete normality test in this study uses the spss for windows program. the normality test is used to determine whether the obtained research data has a normal distribution or not. from table 15, it can be seen that the spread of unstandardized residual value is normal because the significant value is 0.200 > 0.05 (normally distributed). therefore, this analysis can be processed to regression analysis because the conditions in the classical assumption test, in this case the residual value, have been stated as normally distributed. table 15 normality test result of unstandardized residual using one-sample kolmogorov-smirnov test n 12 normal parameters a,b mean 0.0000000 std. deviation 31.61054075 most extreme differences absolute 0.194 positive 0.194 negative -0.183 test statistic 0.194 asymp. sig. (2-tailed) 0.200 c,d a. normal test distribution b. calculated from data c. lilliefors significance correction d. lower bound of the true significance 3.7. heteroscedasticity test of concrete this test is used to determine deviation of the classic assumptions of heteroscedasticity because there is an inequality of variance from one residual observation to another, while in the regression model, it is actually expected to be constant. in this test, the researchers use the spss 25 for windows program and the glejtser method. the symptom of heteroscedasticity is shown by the regression coefficient of the independent variable on the absolute value of the residual. in decision making, there is no symptom of heteroscedasticity if the probability value is greater than the alpha value (0.05). from table 16, it can be concluded that heteroscedasticity do not occur in the regression model because the sig. variation of plastic waste to the absolute residual is 0.386 > 0.05. 0% 5% 10% 15% variation of plastic waste 188 advancesin technology innovation, vol. 6, no. 3, 2021, pp. 179-190 table 16 result of heteroscedasticity test of concrete model unstandardized coefficient standardized coefficient t sig. b std. error beta 1 constant 30.647 10.761 2.848 0.017 plastic waste variation -1.189 1.312 -0.275 -0.906 0.386 a. dependent variable: re3 3.8. hypotheses test and regression equations there are two variables in this study, namely independent variable (x) and dependent variable (y). variable x is variation of plastic waste, and variable y is concrete compressive strength. the hypotheses in this study are as follows. (1) ho: there is no simultaneous effect between variable x on variable y. (2) ha: there is a simultaneous effect between variable x on variable y. table 17 anova test result anova a model sum of squares df mean square f sig. 1 regression 23784.941 1 23784.941 21.639 0.001 b residual 10991.489 10 1099.149 total 34776.430 11 a. dependent variable: compressive strength b. predictors: constant, plastic waste variation hypotheses test can be done using the f test and significance test (probability). the test rule is to use the f test under several conditions: if f value ≤ f table, ho is accepted; if f value > f table, ho is rejected. based on the calculation, the f value of the anova table is 21.639 and the f table value is 4.96. thus, it can be concluded that f value > f table. it means ho is rejected, or in other words, ha is accepted (there is simultaneous effect between variable x on variable y). the significance test (probability) is conducted under several conditions: if probability (sig.) > α, ho is accepted; if probability (sig.) < α, ho is rejected. based on table 17, the probability value (sig.) = 0.001 and the significant level value α = 0.05. therefore, it can be concluded that the probability value (sig.) < α. it means ho is rejected, or in other words, ha is accepted (there is a simultaneous effect between variable x on variable y). table 18 coefficient results of compressive strength regression analysis coefficient a model unstandardized coefficients standardized coefficients t sig. 95% confidence interval for b b std. error beta lower bound upper bound 1 constant 235.162 16.853 13.954 0.000 197.611 272.713 plastic waste variation -9.560 2.055 -0.827 -4.652 0.001 -14.139 -4.981 a. dependent variable: compressive strength based on table 18, it can be analyzed that the regression equation model for the simultaneous effect on compressive strength with plastic waste variation is y = 235.162 – 9.560 x, compressive strength without the addition of variation of plastic bottle waste is x = 0, and the value of compressive strength is (y) = 235.162 kg/cm 2 (19.518 mpa). if the addition of variation of plastic bottle waste is x = 1%, it is estimated that the value of compressive strength (y) = 235.162 – 9.560 (1) = 225.602 kg/cm 2 (18.725mpa) is added. every 1% addition of plastic bottle waste will decrease the compressive strength of the concrete by 9.560 kg/cm 2 (0.793 mpa). 189 advancesin technology innovation, vol. 6, no. 3, 2021, pp. 179-190 4. conclusions based on the results of the analysis and discussion in the previous section, the following conclusions are obtained. (1) the linear regression equation for the variation of plastic waste on the compressive strength of concrete is y=235.162– 9.560x, meaning that there is a significant influence between the variation of plastic waste on the compressive strength of concrete. (2) the more plastic waste added into the variation, the lower the concrete compressive strength, where the maximum compressive strength is at the variations of 0% and 5%, namely 19.192 mpa for the variation of 0% and 16.414mpa for the variation of 5%. in general, the concrete which contains more plastic has poorer durability performance. conflicts of interest the authors declare no conflict of interest. references [1] m. guendouz, f. debieb, o. boukendakdji, e. h. kadri, m. bentchikou, and h. soualhi, “use of plastic waste in sand concrete,” journal of materials and environmental science, vol. 7, no. 2, pp. 382-389, october 2016. [2] a. i. n. diana and h. depriyanto, “the effect of using economic plastic fiber (eco plafie) paving block on compressive strength, shock resistance, and water absorption as environmentally friendly products,” national conference on mathematics, science and education, madura islamic university press, october 2018, pp. 19-26. [3] h. karimah, “experimental study of the effect of ldpe cast plastic content on the compressive strength of normal strength concrete,” thesis, department civil engineering, parahyangan catholic university, bandung, jawa barat, 2018. [4] alvine, “experimental study of the effect of abs plastic waste as a partial substitution of concrete aggregates with a design compressive strength of f’c = 35 mpa,” thesis, department civil engineering, parahyangan catholic university, bandung, jawa barat, 2019. [5] s. suleman and s. needhidasan, “utilization of manufactured sand as fine aggregates in electronic plastic waste concrete of m30 mix,” materials today, vol. 33, no. 1, pp. 1192-1197, september 2020. [6] b. v. bahoria, d. k. parbat, and p. b. nagarnaik, “xrd analysis of natural sand, quarry dust, waste plastic (ldpe) to be used as a fine aggregate in concrete,” materials today, vol. 5, no.1, pp. 1432-1438, february 2018. [7] s. s. mathi, v. johnpaul, r. sindhu, p. r. riyas, and n. chidambaram, “mechanical properties of concrete with plastic as partial replacement of fine aggregate,” materials today: proceedings, in press. [8] z. c. steyn, a. j. babafemi, h. fataar, and r. combrinck, “concrete containing waste recycled glass, plastic, and rubber as sand replacement,” construction and building materials, vol. 269, 121242, february 2021. [9] a. evram, t. akçaoğlu, k. ramyar, and b. çubukçuoğlu, “effects of waste electronic plastic and marble dust on hardened properties of high strength concrete,” construction and building materials, vol. 263, no. 1, pp. 1-10, september 2020. [10] x. li, t. c. ling, and k. h. mo, “functions and impacts of plastic/rubber wastes as eco-friendly aggregate in concrete—a review,” construction and building materials, vol. 240, no. 1, pp. 1-13, december 2019. [11] y. adela, m. behanu, and b. gobena, “plastic wastes as a raw material in the concrete mix: an alternative approach to manage plastic wastes in developing countries,” international journal of waste resources, vol. 10, no. 382, pp. 1-7, june 2020. [12] s. fansuri and a. i. n. diana, “characteristics of coarse aggregate and fine aggregate for building materials in sumenep regency,” national conference on mathematics, science, and education, madura islamic university press, october 2018, pp. 9-18. [13] method of test for specific gravity and water absorption of fine aggregate, indonesian national standard 1970, 2008. [14] test method for fine aggregate and coarse aggregate sieve analysis, indonesian national standard astm c136, 2012. [15] method of test for specific gravity and water absorption of coarse aggregate, indonesian national standard 1969, 2008. [16] procedure for making a normal concrete mix plan, indonesian national standard 03-2834, 2000. [17] how to test a concrete slump, indonesian national standard 1972, 2008. [18] how to test concrete compressive strength with cylindrical test objects, indonesian national standard 1974, 2011. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). 190  advances in technology innovation, vol. 6, no. 2, 2021, pp. 106-116 innovative strategy to meet the challenges of the future digital society olivier bonnaud * department of sensors and microelectronics, ietr, university of rennes 1, 35042 rennes, france national coordination for education in microelectronics, gip-cnfm, 38016 grenoble, france received 12 november 2020; received in revised form 30 december 2020; accepted 04 february 2021 doi: https://doi.org/10.46604/aiti.2021.6724 abstract today, we are experiencing a societal revolution with the development of digital technologies, and it brings new challenges. indeed, the number of connected objects, intelligent sensors and iots are increasing exponentially. the same goes for the resulting energy consumption. beyond 2030, without a radical transformation of communication technologies and protocols, the digital world will be at an energy dead end. all these objects are physically realized with microelectronic devices and systems. this analysis of the microelectronics community has led the french government to recognize an electronics sector that is becoming a priority area of industrial policy. the strategic committee of this sector has proposed innovations applied to the entire digital chain including all facets of the microelectronics field and human skills and know-how. the technological and energy issues are thus presented, and the proposed solutions were addressed. they concern both technological and human aspects. this paper ends by giving examples of the implementation of innovative approaches which essentially include the electronic functions involved in connected objects and which are intended to bring the know-how of future actors in the field. keywords: digital society, microelectronics, multidisciplinarity, energy challenge, higher education training 1. introduction it is globally recognized that we are experiencing a societal revolution with the development of digital technologies, connected objects and the internet of things (iot) [1]. while this rapid evolution may seem easy and problem-free, in reality, it raises new and extremely important challenges. indeed, the number of connected objects, smart sensors, artificial intelligence systems (ai) and iots is growing exponentially. all these objects, functions and systems take part in the industrial revolution 4.0. [2], which will involve many robots, collaborative robots [3], connected monitoring, and management. as a consequence, the resulting energy consumption of this digital world is growing exponentially as well. all of these objects and systems are physically realized using microelectronic and nanotechnological devices, circuits and systems, as well as smart sensors and actuators, opening the way to multidisciplinary applications [4]. beyond 2030, without a radical transformation of technologies and communication protocols, the digital world will be in an energy dead-end and most likely also in a digital security dead-end [5]. this analysis by the microelectronics community has led to the french government's recognition of an electronics sector that is becoming a priority axis of industrial policy [6]. a strategic sector committee has been set up, bringing together industry representatives, researchers and academic forces in the field of electronics and microelectronics. it acts as a structure for brainstorming and positions itself as a force for proposals to government ministries. it highlights innovative topics applied to the entire digital chain and which are part of the technical challenges. they will be applied to elementary components and to the architectures of electronic circuits and systems for the detection, processing and storage of information as well as signal * corresponding author. e-mail address: olivier.bonnaud@univ-rennes1.fr advances in technology innovation, vol. 6, no. 2, 2021, pp. 106-116 107 transmission. if these topics seem to be part of a classical approach, in practice the concepts to be implemented are new. indeed, designers and manufacturers have tried in the past to bring new products with new functions to the market as quickly as possible, regardless of energy consumption and environmental impact. their approach has been to implement the existing product using ip (intellectual property) and through improved integration, the size of the circuits and systems has not increased. in addition, in order to produce faster and cheaper, many companies have ceased their activities and relocated the manufacture of electronic devices and systems abroad, particularly, to the far east. while this approach seems very good economically, in the long term it means a loss of skills and foreign dependency that can affect strategic sectors. this evolution generates new challenges that lead to a new strategy. this strategy must respond to an increase in human skills and know-how, both qualitatively and quantitatively, in order to meet these challenges. the main objective of this work is to describe the strategy adopted by the french industrial electronics sector and by the national academic network in charge of higher education in the field of electronics, microelectronics and nanotechnologies, in order to meet the challenges on both levels. the first one deals with the technological aspect with an innovative approach that particularly decreases the power consumption of all iot applications. note that the new connected objects are designed to save energy in many societal applications related to transportation or smart home for example [4]. the second one is the human aspect in the framework of improving knowledge and know-how in order to reduce the shortage of skills in this exponentially growing field. future actors will have to ensure innovation in connected objects and their applications in all sectors of society. thus, this document focuses successively on the strategy set-up by the microelectronics industry sector in france, on the detailed description of the technological and energy challenges that highlight human needs, and finally on the actions carried out within the national training network in microelectronics and nanotechnologies to train future actors in this field. special attention is paid to the current contribution to training and to the innovation strategy that needed to meet future challenges. 2. digital society and microelectronics links in order to fully understand the importance of the strategy and the challenges in the field of microelectronics, this first part deals with the situation of the field in the context of the future digital society. this evolution is global. however, in order to describe concrete actions, the case of france is highlighted. 2.1. dependence of the digital society on the electronics sector the revolution of smart and connected objects coupled with the energy transition is affecting all sectors of industry and the growth of the next few decades will belong to those who can benefit from these technological innovations based on electronics [6]. electronics is the essential industrial base for the production of digital equipment and systems. the electronics industries thus take a central place in the french industrial landscape and are at the heart of the digital transformation [7]. after several years of reflection and consultation, the electronics industry sector was found on march 15, 2019 by the general directorate of enterprises within the ministry in charge of economy, finance and industry [8]. the “electronic sector” is therefore at the heart of the societal evolution towards digital and connected objects (iot, internet of things, internet of every-thing!). in france, it represents 1,100 companies and 230,000 jobs for a turnover of 15 billion euros [6]. in order to meet the challenges, this sector must master the 4 complementary pillars which are: (1) technologies and electronic components including intelligent sensors to create data, (2) the connected objects to process, transmit and develop the associated services, (3) power electronics to support the energy transition and the development of electric mobility, (4) cybersecurity to build the confidence necessary for the development of electronic technologies in industry (cyber-physical systems). advances in technology innovation, vol. 6, no. 2, 2021, pp. 106-116 108 2.2. major role of microelectronics, microand nanotechnologies within this sector, microelectronics occupies a central place since all the objects involved in smart and connected objects are made up of electronic components, circuits and systems. the great majority of which are integrated and carry embedded software that provides the intelligence of the application devices. in addition, many new systems include passive devices and micromechanical elements for sensing or actuations. to resume, the connected objects are primarily made up of multiple microelectronic, microtechnological and communications devices [9] that are judiciously combined in a full system. this evolution has been made possible by the fabulous growth of microelectronic integration over the last sixty years. the number of elementary components in an integrated circuit has increased from a few units to a few billion thanks to the reduction of lateral dimensions in a plane [10]. unfortunately, this reveals a physical limitation related to the distance between atoms and the quantum effects attached to them. but about fifteen years, this limitation has been overcome by playing on the third dimension [11], which makes it possible to have integrated systems of several thousand functions that can each contain billions of elementary components [12]. these results make it possible to have data processing systems in fabulous quantities in ever shorter times and have paved the way for connected objects, iot, ai and the industrial revolution 4.0. all these aspects are taken into consideration in the strategy at the national level. 2.3. main objectives resulting from the analysis of the french electronics sector the french electronics sector has established a strategic committee composed mainly of industrial members but also of academic members. the main goal of this strategic committee is to develop activities capable of meeting the challenges mentioned above. the committee has thus identified four key priorities in its work: (1) maintaining excellence in key digital technologies by increasing r&d&i efforts and developing strategic partnerships; (2) adapting skills by anticipating changing needs (employment and skills) and by developing work-linked training offer; (3) promoting competitive "made in france" electronics manufacturing; (4) positioning the sector as a key player in the digital transformation by supporting smes in other sectors. within the framework of these priorities, the strategic committee set up four working groups in 2019 that produced several documents proposing priority actions and objectives. the first priority directly concerns the purely technological aspects with the associated constraints and challenges. it mainly concerns the policy towards research laboratories and research development departments of companies. the second priority relates to the adaptation of skills and know-how, which presents a challenge in terms of training content at all levels and the number of people trained. everything possible must be done to achieve these objectives that are: (1) raising young people's awareness of the field in order to make it more attractive, (2) initial training from technician to engineer level, (3) advanced training through research to develop innovation, (4) life-long training to enable the company's employees to adapt to innovative technologies. it is clear that we will focus on these two priorities that directly involve universal strategy and actions. to this end, it is useful to detail the different areas of scientific and technical expertise and know-how. in other words, we need to highlight the technical and technological connected objects and challenges advances in technology innovation, vol. 6, no. 2, 2021, pp. 106-116 109 3. connected objects and challenges the digital society and future industry 4.0 are based on internet of things (iot). in terms of physical support, the basic system is the connected object that gathers many electronic functions. in addition, the connected object fills also a mission of control, monitoring processes or robots. in order to well select the skills and the orientation of the training, it is useful to highlight the several associated works, limitations and related challenges. 3.1. usual architecture of a connected object the first mission of a connected object is to retrieve a physical, chemical or biological quantity by means of a sensor that transforms it into an electrical signal, amplifies and treat the signal, transmits it remotely to a processing system in order to have a feedback through the whole chain that can correct the original quantity. it is thus composed of an on-site part, a remote part and a communication system between the two. the connected objects must perform functions such as detecting a physical, chemical or biological analogue signal, transforming it into digital signals via converters, storing and transmitting it via communication protocols, receiving it in databases and data centers, analysis processing centers, and performing the return chain to an actuator that can control the application phenomenon. thus, all these objects have a common architecture as shown in fig. 1, which includes sensors and actuators, electronic modules for signal processing, data transmission and reception modules, but also devices for monitoring, visualization, alarms, memorization, energy recovery and storage, etc. [9]. fig. 1 schematic architecture of a connected object. many electronic functions are involved in the full system 3.2. multiplication of connected objects and sensors as already mentioned, the enormous integration capacity of microelectronics allows multiplying not only the connected objects, but also the sensors and the iot [13]. the number of these functions and objects is growing exponentially, as shown in fig. 2 [14]. it is interesting to note that the development of heterogeneous integration has led to the development of sensor design and fabrication. for example, about thirty sensors are installed in a regular cell phone, while seventy are wired in a standard car. fig. 2 exponential growth rate of the number of iot, connected objects and sensors; a direct effect of the microelectronics integration growth advances in technology innovation, vol. 6, no. 2, 2021, pp. 106-116 110 3.3. energy consumption challenges the consequence of this exponential growth is that the energy consumed by connected objects is also growing exponentially and is becoming globally preponderant [15] as it will be shown below. indeed, one gigabyte of data downloaded from the internet sums up to 5.12 kwh [16]; loading a film online (equivalent to a 5gb dvd) consumes 25 kwh! according to recent literature, the contribution to this consumption is distributed as shown in fig. 3: fig. 3 allocation of power consumption due to connected objects and iot 48% by data centers (servers, routers); 38% by the end user (computer, smartphone, display, battery charging, screen power supply, life boxes, etc.); 14% for communications (cables, optical fibers, repeater-amplifiers, hertzian or satellite transmissions). fig. 4 expected exponential growth of the iot electrical energy consumption without any change in the technologies of iot and connected objects this distribution is approximate since some functions can be assigned to two different sectors. however, the total appears to be correct. as shown in fig. 4, the electricity consumption due to iot is multiplied by 2 each 4 years. in 2018, the electrical energy consumed by the implementation of connected objects reached 2,900 twh, which corresponds to a permanent production of 320 gw out of the 8,800 annual hours and represents 11% of the world's electrical energy consumption [17]. at the beginning of 2020, the consumption was equivalent to about 3 times the energy consumption of world air traffic as shown fig. 4. let us notice that due to the global pandemic linked to “covid-19”, the extrapolation of global air traffic beyond 2020 has been strongly affected since these curves were drawn. assuming current technologies are maintained, this global energy consumption could reach 124,600 twh in 2040, which represents a permanent production of 15,000 gw or 15,000 nuclear reactors. this energy consumption is also equivalent to the global energy production in 2018, regardless of its source and form [18]. it is clear that this situation will become unrealistic and that major changes will have to be made in the design, technology, energy consumption of connected objects, the organization of manufacturing (industry 4.0.) and the transfer and storage of data. in practice, the energy consumption of all physical objects will have to be considerably reduced. the first approach will consist in reducing the consumption of microelectronic devices and circuits, which account for a contribution close to 60%. assuming that the reduction gradually reaches a factor of 10 over the next 20 years, we can expect an evolution of iot-related advances in technology innovation, vol. 6, no. 2, 2021, pp. 106-116 111 energy consumption as shown in fig. 5. global electrical consumption and iot consumption in dash lines are issued of fig. 4. knowing that iot consumption does not depend only on the hardware, instead of undergoing a growth in a ratio of 32, we should limit this growth with a ratio of 6.5. the annual electrical energy consumption of iot is expected to reach nineteen thousand terawatt-hours, which is more appropriate for the planet (full lines) and therefore more realistic. however, global consumption of electrical energy will in any case be twice as high as it is today. fig. 5 expected iot consumption and global electrical energy consumption in 2040. decreasing significantly the consumption of electronics devices and circuits (dash line to solid line), should allow to limit the global electrical energy consumption to 57,000 twh a year by 2040 3.4. induced challenges on microelectronics technologies, integration, consumption, multidisciplinarity in order to achieve this reduction by a factor of 10, many challenges will have to be met. the challenges of the discipline thus appear at different levels. the works of the french microelectronics network and of the working group of the strategic committee of the industrial sector have concluded on the priorities. although already presented in a previous document [18], it is however useful to recall the enumeration of the major points [13]: (1) the further miniaturization of elementary components in order to minimize their consumption both in operation and off-state by reducing leakage currents (tunnel effect); (2) improving the performance of power components and their circuits in order to reduce both switching and static losses. these properties are very important for servers and data centers which have a huge permanent energy consumption; (3) the uses of the third dimension to improve circuit integration in order to minimize interconnection losses and improve reliability. extensive studies will be required to control the power density to be dissipated in these three-dimensional structures; (4) modification of the circuit architecture to separate useful active zones from zones that can be put on standby, a way to significantly decrease the permanent energy consumption of circuits; (5) the generalization of asynchronous control which should minimize the important losses linked to the synchronous power supply by the clocks of all the transistor control electrodes; (6) the implementation of new low-temperature and/or large-area technologies involving new materials (inorganic or organic) for which the electrical consumption can be several orders of magnitude lower; (7) the development of sensors and actuators in specific low-power technologies. these technologies will be able to be integrated thanks to the rapid growing heterogeneous integration, presently. several of them will involve a multidisciplinary approach related to the application domain; (8) optimization of communication devices and protocols to limit frequency band occupation and data flow to keep only the useful according to the mission profile. all these topics are introduced in the strategy of the microelectronics industry sector and the national training network. advances in technology innovation, vol. 6, no. 2, 2021, pp. 106-116 112 3.5. present shortages in human skills revealed by the strategic committee the technical and technological challenges can only be met if the future technicians, engineers, teachers and doctors are capable of dealing with the subjects mentioned and providing innovative solutions that meet the requirements. it is therefore necessary to provide initial and continuing training adapted to this evolution and allowing them to acquire a good basic knowledge of the technique but also the know-how indispensable for design and production [19]. due to the increasing complexity of the discipline, training must be able to cover the basic knowledge that will have to be redefined, but also the acquisition of skills in highly specialized areas. this necessitates the adaptation of study content by introducing new concepts and pedagogical approaches, including more practical learning, but also a strong awareness of multidisciplinarity in relation to the fields of application [20], project-based learning and the ability to work and innovate in a multidisciplinary team. currently, at the global level, companies in the electronics and microelectronics sector report a lack of applications in many skilled jobs. training must fill this gap in the job market. this aim is in line with the missions of the french higher education training network in the field of microelectronics. 4. the national microelectronics network in the electronics sector 4.1. constitution of the cnfm network as already presented in the previous paper (18), this network is entitled cnfm for national coordination for training in microelectronics and nanotechnologies [21], with an official structure as a gip (for public interest group). it is composed of twelve academic members who are respectively in charge of twelve common centers spread over the french territory, each center being common to several academic institutions located in the same geographic area [22]. the twelve cnfm common centers manage platforms and their technical facilities for practical hands-on training. among the eighty technical platforms, 7 cleanrooms are mainly dedicated to education, but also allow research and transfer activities. national services for cad (computer-aided-design) tools provided by the montpellier cnfm common centers are dedicated to all the national community and foreign partners of the network. two industrial unions and especially the most representative association acsiel alliance electronics consortium [23] are also the members of the cnfm body and involved in the national electronic sector policy. they represent more than 150 industrial sites. 4.2. concept and objective of the cnfm network the true essence of the gip-cnfm network is to share technical platforms, which include technological processes, cad, characterizations, and test. this sharing is mandatory due to the very high costs of equipment and functioning expenses. in addition, sharing pedagogical approaches between educators, and applying good practices towards students in initial training or even in life long learning [24] is also a main educational target. fig. 6 the microelectronic network is at the heart of the master plan. innovative practice enables the know-how of future players in the electronics sector to be developed advances in technology innovation, vol. 6, no. 2, 2021, pp. 106-116 113 the objective of the existing network is to provide a learning environment that allows the trainees to be immersed in a technological environment that prepares them for the industrial environment of the 21st century and more particularly for industry 4.0. thanks to an eight years project entitled finmina [25], the policy of permanent innovation in the twelve centers of the network was boosted. it allowed to develop innovative platforms with original practice all along the eight years, from 2012 to 2019 [26]. the strategy was therefore to create new practical training courses in line with the advent of the digital society. fig. 6 summarizes the strategy to address the technological challenges while addressing the human challenges. 4.3. main know-how training activities in the majority of cases, the innovative character of the training activities stems from the research activities carried out by the teacher-researchers of the academic institutions and members of the network. this research is very often conducted in partnership with corporate research and development centers, which makes it possible to direct them towards the expressed by the sector. fig. 7 summarizes the two indicators of the network combining skills and objectives based on an innovative approach. fig. 7 positioning of the cnfm training network in the electronics sector strategy. innovation must improve skills in the priority actions of the electronics sector another point that needs to be stressed is the necessary complementarity of knowledge and know-how across the skill set [27-28]. this complementarity is more essential in the engineering disciplines and more particularly in microelectronics and whatever the technologies [29]. various examples of platform realizations and practical work that were recently set-up by the network are given below. they address all the mentioned aspects of the evolution towards digital, knowing that a good digital technology must solve many problems related to the analogue properties of electronics! they also concern the variety of application fields, which implies the use of physical/electronic, chemical/electronic, biological/electronic transducers, which are mostly miniaturized and integrated. subjects range from classical electronics to microelectronics and nanotechnologies, from very low power to high power, including fast communication electronics and embedded electronics, systems-on-a-chip, lab-on-a-chip, and digital security systems, which gives them a multidisciplinary aspect. subjects range from classical electronics to microelectronics and nanotechnologies, from very low power to high power, including fast communication electronics and embedded electronics, systems-on-a-chip, lab-on-a-chip, and digital security systems, which gives them a multidisciplinary aspect [29]. these techniques can implement multi-physical design tools, integrated silicon technologies, but also technologies based on new inorganic compound and organic materials and large-area thin-film devices for display or energy harvesting [30]. finally, they also involve digital security (in the material and equipment meaning), reliability, and testing whether electrical, physical or physical-chemical from the human to the atomic scale. 4.4. main results on innovative strategy of the network detailed information on these activities is available on the network's website [22] and examples have already been presented in the literature according to their subject matter [28-31]. fig. 8 shows several results of the innovative training strategy, which were provided by the network's inter-university centers. the topics concern technologies, designs, advances in technology innovation, vol. 6, no. 2, 2021, pp. 106-116 114 characterizations, sensors and systems with a multidisciplinary approach as mentioned in the main objectives. on each selected image, the name of the center and the subject are displayed. all innovative practice topics are validated by the representative of the microelectronics industry in the framework of the general assembly of the official structure of the network (gip-cnfm), whose president is also the president of the french union of the microelectronics industry. fig. 8 example of practice realizations by students in several network’s microelectronics inter-university centers. for each picture, the name of the center and the topic are displayed in fact, more than 80 practical and innovative learning platforms have been set up within the network over the last decade. every year, nearly 6,000 users come to acquire know-how on these evolving platforms, with another 11,000 coming to the centers to acquire basic or awareness-raising know-how. this approach has an international interest, especially in china [32-33]. 5. conclusions in an environment of continuous and tremendous technological evolution for more than 60 years, the field of microelectronics allows the development of connected objects and the advent of industry 4.0 applied to many areas (iot) that are experiencing exponential growth. the societal stakes are considerable, and everything must be done to limit the energy consumption of the communicating systems that will invade our daily lives. these challenges will be met thanks to high-quality, innovative training to improve the performance of the associated electronics, which must on the one hand meet these economic and societal requirements, and on the other hand be in line with the priorities of the electronics industry, which has been undergoing restructuring since 2019. the gip-cnfm, a joint national training structure, by providing students and technical staff of industrial companies with skills and know-how in the field, is positioning itself as a major player in symbiosis with the world of industry and research, and is helping to prepare for the future by increasing the pool of trainees both qualitatively and quantitatively. acknowledgment the authors want to thank all the members of the french gip-cnfm network for they contribution to many innovative realizations. this work was financially supported by french higher education ministry and by idefi-finmina program (anr-11-idfi-0017). a special thanks to l. chagoya-garzon, secretary of gip-cnfm for her fruitful advice for the proof reading of this paper. advances in technology innovation, vol. 6, no. 2, 2021, pp. 106-116 115 conflicts of interest the authors declare no conflict of interest. references [1] m. burgess, “what is the internet of things?” wired magazine, https://www.wired.co.uk/article/internet-of-things-what-is-explained-iot, february 16, 2018. [2] a. schütze, n. helwig, and t. schneider, “sensors 4.0-smart sensors and measurement technology enable industry 4.0,” journal of sensors and sensor systems, vol. 7, no. 1, pp. 359-371, may 2018. [3] r. smith, “the future of manufacturing: cobots in the factory,” https://www.tctmagazine.com/tctblogs/guest-blogs/the-future-of-manufacturing-cobots-in-the-factory/, march 04, 2019. [4] o. bonnaud and l. fesquet, “microelectronics at the heart of the digital society: technological and training challenges,” sbmicro, ieeexplore, august 2018, pp. 1-4. [5] e. gelenbe and y. caseau, the impact of information technology on energy consumption and carbon emissions, acm publication, june 2015, pp. 1-15, https://dl.acm.org/doi/pdf/10.1145/2755977. [6] t. tingaud, “la filière industries de l’electronique,” https://www.conseil-national-industrie.gouv.fr/la-filiere-industries-de-l-electronique. [7] o. bonnaud, “mandatory matching between microelectronics industry and higher education in engineering toward a digital society,” smart education and e-learning 2019, springer, vol. 24, pp. 255-266, june 2019. [8] “signature by the minister,” https://www.economie.gouv.fr/signature-contrat-comite-strategique-filiere-industrie-electronique-plan-nano-2022. [9] o. bonnaud and l. fesquet, “innovation for education on internet of things,” proceedings of engineering and technology innovation, vol. 9, pp. 1-8, july 2018. [10] g. e. moore, “cramming more components onto integrated circuits,” electronics magazine, vol. 38, no. 8, pp. 114-117, april 1965. [11] k. cheval, j. coulm, s. gout, y. layouni, p. lombard, d. leonard, et al., “progress in the manufacturing of molded interconnected devices by 3d microcontact printing,” advanced materials research, vol. 1038, pp. 57-60, september 2014. [12] r. r. tummala and m. swaminathan, system on package: miniaturization of the entire system, 1st ed. mcgraw-hill, 2008. [13] o. bonnaud, “new approach for sensors and connecting objects involving microelectronic multidisciplinarity for a wide spectrum of applications,” international journal of plasma environmental science and technology, vol. 10, no. 2, pp. 115-120, december 2016. [14] j. bryzek, “the trillion sensors (tsensors) foundation for the iot,” https://www.iot-inc.com/wp-content/uploads/2015/11/2-janusz.pdf, 2015. [15] enerdata, “global energy statistical yearbook 2019,” https://yearbook.enerdata.net/total-energy/world-energy production.html, march 2019. [16] m. greenfield, “internet des objets (iot): nombre d'appareils connectés dans le monde de 2015 à 2025,” https://fr.statista.com/statistiques/584481/internet-des-objets-nombre-d-appareils-connectes-dans-le-monde--2020/ [17] enerdata, “world energy production,” https://yearbook.enerdata.net/total-energy/world-energy-production.html, 2020. [18] o. bonnaud and a. bsiesy, “adaptation of the higher education in engineering to the advanced manufacturing technologies,” advances in technology innovation, vol. 5, no. 2, pp. 65-75, april 2020. [19] o. bonnaud, “knowledge and know-how in microelectronics; strategy of innovative practice to balance the new on-line courses approach,” international conference on advances in technology engineering, june 2017, pp. 10-11. [20] o. bonnaud and l. fesquet, “communicating and smart objects: multidisciplinary topics for the innovative education in microelectronics and its applications,” 15th international conference information technology based higher education and training, ieee press, august 2015, pp. 1-5. [21] cnfm: “coordination nationale pour la formation en microélectronique and nanotechnologies,” www.cnfm.fr, december 2020. [22] o. bonnaud, p. gentil, a. bsiesy, s. retailleau, e. d. gergam, and j. m. dorkel, “gip-cnfm: a french education network moving from microelectronics to nanotechnologies,” ieee global engineering education conference, april 2011, pp. 122-127. advances in technology innovation, vol. 6, no. 2, 2021, pp. 106-116 116 [23] acsiel alliance é lectronique, “components and systems alliance for electronics industry in france,” http://www.acsiel.fr/en-gb/index.aspx, 2019. [24] b. pradarelli, o. bonnaud, p. nouet, and p. benoit, “cnfm: innovative single national entry-point for lifelong learning in microelectronics,” proceedings of edulearn, july 2019, pp. 1-3. [25] finmina, “formations innovantes en microélectronique et nanotechnologies,” http://www.cnfm.fr/?q=fr/presentation_finmina, march 2020. [26] o. bonnaud, “finmina: a french national project dedicated to educational innovation in microelectronics to meet the challenges of a digital society,” smart innovation, systems and technologies, vol. 188, pp. 31-44, june 2020. [27] o. bonnaud and l. fesquet, “practice in microelectronics education as a mandatory supplement to the future digital-based pedagogy: strategy of the french national network,” 2016 11th european workshop on microelectronics education (ewme), southampton, 2016, pp. 1-6. [28] o. bonnaud, “new vision in microelectronics education: smart e-learning and know-how, a complementary approach,” smart innovation, systems and technologies, vol. 99, pp. 267-275, may 2018. [29] o. bonnaud, “introduction of the multidisciplinary know-how of the french national training network to contribute to the evolution of microelectronic technologies towards connected objects,” 17th international conference on information technology based higher education and training (ithet), august 2018, pp. 1-4. [30] o. bonnaud, “the multidisciplinary approach: a common trend for ulsi and thin film technology,” ecs transaction, vol. 67, no. 1, pp. 147-158, 2015. [31] o. bonnaud and l. fesquet, “innovation in higher education: specificity of the microelectronics field,” 31st symposium on microelectronics technology and devices, ieee press, september 2016, pp. 119-122. [32] o. bonnaud and l. wei, “a way to introduce innovative approach in the field of microelectronics and nanotechnologies in the chinese education system,” proceedings of engineering and technology innovation, vol. 4, pp. 19-21, october 2016. [33] o. bonnaud, y. danto, y. h. kuang, and l. yuan, “international flipped class for chinese honors bachelor students in the frame of multidisciplinary fields: reliability and microelectronics,” advances in technology innovation, vol. 3, no. 3, pp. 126-132, march 2018. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation , vol. 2, no. 3, 2016, pp. 89 94 89 biosignal –based multimodal biometric system kuk won ko and sangjoon lee* school of mechanical and ict convergence engineering, sun moon university, s. korea. received 02 may 2016; received in revised form 25 june 2016; accept ed 26 june 2016 abstract this study concerns personal identification based on electrocardiogra m (ecg) and phot oplethysmogram (ppg) signals. we manufactured a bio-signal measure ment system that can simu ltaneously measure ecg and ppg signals, using which three channels of ecg signal and one channel of ppg signal we re acquired fro m the right-hand index finger of a total of 33 su bsets for 3 minutes. lead-i signal of the three-channel ecg signal and the one-channel ppg signal we re selected for recognition. for each subject, 160 heartbeats were automatically separated from the acquired bio-signals, and a total of 21 features, co mprising 15 ecg features, 4 ecg-related ppg features, and 2 features co ncerning ppg only, were e xtracted fro m each heartbeat. letting the 21 features form a single data point, heartbeat features of each subset were used as the training data for a support vector mach ine (svm) c lassifie r, with the number of data points being adjusted from 10 to 80, and the data points (80 – 150) other than the tra ining data were used as the testing data, in order to investigate the recognition performance indices. as a result, the proposed algorithm showed high recognition performance o f 99.28% accuracy, 0.88% frr, 0.85% far, 99.28% sensitivity, and 99.31% specificity, when there are 80 training data points. moreover, even when there are 10 tra ining data points, the proposed alg orith m showed the performance of 92.77% ac curacy, 7.23% frr, 6.29% far, 92.77% sensitivity, and 93.21% s pecific ity, wh ich can be evaluated as an extre me ly high recognition performance considering that there was a total of 4,950 testing data points. keywor ds : biometric, pattern recognition, personal identification 1. introduction following the rapid recent advances in information technology (it), the demand for b iometric recognition technology is increasing for the purpose of information security and the prevention of privacy invasion. fingerprints, iris, facia l features, dorsal metacarpal veins of the hand, or gait, which are the intrinsic features of human body, are used as the ma in e le ments of e xisting recognition systems. biologica l features for b io metric recognition can be selected the characteristics that have almost no variation with time, or reflect slow variation with time. this study introduces a mu ltimodal bio metric system and algorithm based on electrocardiogra m (ecg) and photoplethysmogram(ppg) signals. it is known that there are around 80,000 – 100,000 contractions in a healthy heart in o rder to circulate the blood in at ria to the vessels of the entire body, within 0.1% cases of wh ich e xh ibit abnorma l ecg waveforms. moreover, ecg s ignals are known to be regular and convenient for pattern classification compared to other bio-signals. ppg signals are bio-signals that are lin ked with ecg signals, and also represent the state of b lood vessels of the subject with heart contraction. furthermore, while ecg and ppg waveforms of an indiv idual subject vary slightly at each time, these signals have the characteristics of unchanging overall pattern. the first study on recognition systems based on bio-signals was introduced by lena biel [1] in 2001, followed by fe w researchers around the world. john m. irvine [ 2] showed the effec tiveness of ecg for personal identification even when the heart is in a stressed state (excite ment or e xe rcise). in addition, t. w. shen [3], steven a. israel [4], konstantinos n. and plataniotis[5] introduced personal identification algorithms based on ecg. the proposed algorithm is based on the detection of feature points fro m ecg and ppg signals obtained simu ltaneously, and a total of 21 features are automatica lly e xt racted, inclu ding 15 features fro m the ecg heartbeat signals * corresponding aut hor, email: mcp94lee@sunmoon.ac.kr advances in technology innovation , vol. 2, no. 3, 2016, pp. 89 94 90 copyright © taeti separated fro m the overall ecg signal, and 6 features from ppg signal, in order to promote the real-t ime operation of the recognition a lg orith m. the proposed algorithm adopts down slope trace waveform (dstw) method [ 6], which shows high peak detection rate, for the detection of p, q, r, s, and t points of the ecg signal. for pattern recognition, one-against-one support vector machine (svm ) a lgorith m was used, which shows excellent recognition results in various fie lds, and is capable of mu lti -class classification. 2. method 2.1. bio-signal measurement system while the previous ecg recognition and disease detection algorithms can be assessed in terms of validity by using e xisting public dat abase files such as mit ecg database for ve rification, there is no public database that includes a simultaneous measurement of ppg s ignals lin ked with ecg signals. there fore, the hardware for simu ltaneous measurement of ecg and ppg signals was manufactured for the acquis ition of b io-signals for the recognition algorith m in this study. fig. 1 ecg and ppg measurement system blockdiagram fig. 1 shows the block diagra m and the photograph of the manufactured ecg and ppg measure ment system hardware . ecg was measured in three channels (la -ra, ll-la, and ra-ll) based on the wille m einthoven triangle method. moreover, low-pass filters and high-pass filte rs with cutoff frequencies of 0.16hz and 160hz, respectively, were imple mented in hardwa re, with the total a mp lification gain of the ecg signal being 1500, as shown in (a-b) and (a-c) of fig. 1. for ppg measurement, the output signal fro m the filter and the origina l signal were d iffe rentially a mp lified by imple menting a high-pass filter with cutoff frequency of 0.1 hz in o rder to prevent the saturation due to the dc offset fro m the light output during a mplification, as shown in the ppg measurement hardware in fig. 1. moreover, a band -pass filter with cutoff frequencies of 0.1hz and 8hz was designed and imple mented. fig. 2 shows an actual measured signal being displayed and stored on pc. fig. 2 measurement software 2.2. peak detection from ecg and ppg signals through dstw fig. 3 dstw peak detection method[6] peaks are detected fro m the ecg signal by generating dstw , as shown in fig. 3, in order to firstly detect the r peaks in ecg, which a re convenient to detect [6]. after the r peak is detected as shown in fig. 4, q, s, p, t , and end of t are sequentially detected in accordance with a given detection rule (eqs. 15). fig. 4 results from the detection of fiducial points from ecg signal (p, q, r, s, t , te) (subject 4) advances in technology innovation , vol. 2, no. 3, 2016, pp. 89 94 91 copyright © taeti fig. 5 results from the detection of ppg peaks (onset, peak) (subject 4) the points to be detected fro m the ppg signal we re selected to be the onset points, which are the initia l points, and ppg peaks. onset peaks are used to detect the fiducial points in accordance with g iven ru les (eq. 6), a fter the ppg peak has been detected as shown in fig. 5.    q k index( [ ]), min arg x n for  r k 30 n r(k)   (1)    s k index( [ ]), min arg x n for    r k n r k 30   (2)     p k index( ), max arg x n for  q k 60 n q(k)   (3)     t k index( ), max arg x n for    s k n s k 100   (4)     k index( ), end min t arg x n for    t k n t k 50   (5)     onset k index( ), min arg p n for    k 100 n k peak peak ppg ppg   (6) where k denotes the indices of each peak fro m one to m (1≤k≤m), and n denotes the indices of the sampled data for x(n ) and p(n). the actual data were samp led at 250 hz, and n=1 imp lies 4 ms 〖 (t〗 _s=1/f_s ) 2.3. feature extraction as shown in fig. 6, the features for b iometric recognition can be e xtracted by detecting the six types of ecg peaks and the two types of ppg peaks. for this a lgorith m, a total of 21 features were established by e xtracting 15 types of ecg features, 3 types of ppg features, and 3 types of features relating to both ecg and ppg. 2.4. support vector machine for classification svm is a new genre of lea rning systems for pattern recognition, devised by corters and vapnik [7]. while svm did not receive much attention initially, it is recently being used in various fie lds including bio informatics, charac ter recognition, handwriting recognition, fac ia l recognition, and object recognition, and is wide ly used for supervised pattern recognition owing to its excellent performance. fig. 6 extraction of 21 features from ecg and ppg peak detection the train ing and classificat ion algorithm for the proposed algorithm was based on one-against-one mu lti-class svm algorith m, which is not affected by the curse of dimensionality. 160 peaks were detected (p, q, r, s, t , and tend) fro m the ecg signals simultaneously acquired fro m each of the 33 test subjects, and, likewise, 160 peaks (onset, and ppgpeak ) we re detected fro m the ppg signals, forming a set of 21 features for each peak. fro m each of the 34 dataset files, the first (10-80) features were used as the training data, and the other (140 – 80) features were used as the testing data. moreover, we e xa mined the diffe rence in recognition rate between the case where only ecg features are considered, and the case where both ecg and ppg features are considered. 3. results and discussion 3.1. testing method for the e xperiment, the testing data were not provided with any informat ion for recognition, and the average and standard deviation of the performance evaluation ele ments were invest igated, such as accuracy, sensitivity, specific ity, false reject ion rate (frr), and fa lse acceptance rate (far) of the result with varying nu mber of training data points. the formulae for frr and far are shown in eqs. 13 and 14, respectively. for bio metric recognition systems, it is common to measure the equal error rate (err) perfo rmance inde x, wh ich is where frr and fa r coincide, and this was achieved by analyzing the frr and fa r, and the resulting err and re ceiver operating characteristic (roc) curve while increasing the threshold value fo r the distance measurement algorith m for the decision of recognition fro m a lower value to a higher value [14]. advances in technology innovation , vol. 2, no. 3, 2016, pp. 89 94 92 copyright © taeti frr = 𝑁𝑤 𝑁𝑡 × 100 (7) far = tr −tn tr × 100 (8) where nw is number of wrong recognition, nt is number of testing data for target subject, tr is total recognition number for taget subject, tn is testing data number. moreover, the sensitivity and specificity of the proposed algorith m we re investigated using the true positives (tp), the fa lse positives (fp), and the false negatives (fn). for a bio metric recognition algorith m, t p, fp, and fn a re d efined as follow:  tp (true positive): nu mber of accurately recognized test subjects.  fp (false positive): number of falsely recognized test subjects . (the number of data points classified as the subject in question – tp)  fn (false negative): the number of subject data points that should have been recognized, however, we re not reco gnized. (the number of testing data points from each subject -tp) sensitivity and specific ity are defined by eqs. 9 and 10 below, respectively. sensitivity = tp tp+ fn × 100 (9) specificity = tp tp+ fp × 100 (10) 3.2 testing result table 1 recognition performance indices with both ecg and ppg signals used p p g and ecg feat ures analysis and mult i class classifier t raining number 10 20 30 40 50 60 70 80 accuracy (%) avr 92 95 96 97 97 98 99 99 st d 10 9.0 5.7 4.8 2.6 2.0 1.6 1.6 frr (%) avr 7.2 4.7 3.1 2.4 2.6 1.2 0.8 0.8 st d 10 9.0 5.7 4.8 2.6 2.0 1.5 1.6 far (%) avr 6.7 4.3 2.8 2.3 2.3 1.1 0.8 0.8 st d 7.8 6.8 4.9 3.6 2.8 2.0 1.3 1.5 sensit ivit y (%) avr 93 95. 96 97 98 98 99 99 st d 10 9.0 5.7 4.8 2.6 2.0 1.6 1.6 specificit y (%) avr 93 95 97 97 98 98 99 99 st d 7.8 6.8 4.9 3.6 2.8 2.0 1.4 1.5 fig. 7 roc curve with varying tra ining data size (15 ecg features used only) fig. 8 roc curve with varying training data size (21 ecg and ppg features used) fig. 7 shows the performance indices of recognition using a mult i-c lass svm classifie r with 15 ecg features detected fro m the actual measure ment data fro m 33 test subjects. 89.92% accuracy, 10.08% frr, 9.79% fa r, 89.92% s e ns it iv it y , a nd 90. 21% s p e c ific it y we re observed when the classifie r is tra ined with 80 heartbeat data obtained from the ecg s ignals advances in technology innovation , vol. 2, no. 3, 2016, pp. 89 94 93 copyright © taeti fro m each subject. this result corresponds to the re c o gn it io n wit h resp e ct to 2,640 t e s t in g heartbeat data from a total of 33 ecg signals. the proposed algorithm was tested based on a total of 21 feature points, by add ing the 6 ppg feature points measured simu ltaneously fro m the same subject to the 15 ecg feature points. recognition perfo rmance indices, which a re shown in table 1, d isplayed the results of 99.28% accuracy, 0.88% frr, 0.85% far, 99.28% sensitivity, and 99.31% specificity. this is the performance when 80 tra ining data points fro m the ecg and ppg signals measured fro m 33 test subjects are used, and 2,970 testing data points were used overall. moreover, as shown in tab le 1 and fig. 8, the accuracy index shows the high performance of 92.7% even when 10 tra ining data points are applied out of the 160 heartbeat data from each subject (4,950 testing heartbeat data). the result of the above test is better than or equal to the p e rfo rma n c e o f t h e e xis t in g c o mmo n re c o g n it io n s ys t e ms , a n d th e re c o g n it io n performance is observed to be e xtre me ly higher compared to the previous test. therefore, it can be deduced that the consideration of the features of two types of signals for the design of b iometric recognition systems leads to a higher recognition performance than when a single type of signal is used. 4. conclusions this paper concerns the imp le mentation of a mu lti-modal bio metric recognition system and algorith m, where in ecg and ppg signals we re used for the recognition. since there is no pre vious database where ecg and ppg signals a re lin ked, a bio metric system that is capable of a c t u a l ec g a n d pp g me a s u re me n t s wa s manufactured, and three channels of ecg signal and one channel of ppg signal were acquired fro m a total of 33 subjects for 3 minutes. lead -i signal of the three channels of ecg signals only and the 1-channel ppg signal were selected for recognition. with the ecg signals acquired fro m e a c h su b je ct , 160 h e a rt be a ts we re separated and 15 ecg features were e xt racted fro m each heartbeat. moreover, 4 ppg features related to ecg and 2 features of ppg only we re selected, so that a total of 21 features we re used for the recognition. the recognition performance indices of the proposed biometric reco gnition system were investigated by adjusting the number of train ing heartbeat data between 10 and 80. the performance indices of the reco gnition based on ecg features only were found to be 89.92% accuracy, 10.08% frr, 9.79% far, 89.92% sensitivity, and 90.21% s p e c ific it y . fu rt h e rmo re , t h e p e rfo r ma n c e indices of the recognition combining ppg were found to be 99.28% accuracy, 0.88% frr, 0.85% far, 99.28% sensitivity, and 99.31% specific ity, when there are 80 heartbeat data for training, which indicate an e xt re me ly high recognition performance. therefore, it can be deduced that the use of the features of a comb ination of mo re than two bio-signals can lead to a higher reco gnition performance than the use of a single bio-signal. moreover, the recognition system based on the combination of the two signals e xhibits the performance of 92.77% accuracy, 7.23% frr, 6.29% far, 92.77% sensitivity , and 93.21% specific ity, even when 10 heartbeat data are used for training, which can be viewed as an extre me ly high recognition performance, considering that there were 4,950 testing heartbeat data. acknowledgement this work was supported by the human re s o u rc e t ra in in g pro g ra m fo r re g io n a l innovation and creativity through the ministry of education and national research foundation of korea (nrf-2014h1c1a1066998). references [1] l. bie l, et al., “ ecg ana lysis: a new approach in human identification,” ieee t ra ns a ct io ns on inst ru me n t at io n an d measurement, vol. 50, no. 3, pp. 808-812, 2001. [2] j. m . irv in e, et a l., “ eig en pu lse : robust human identification fro m ca rdiovascular function,” pattern recognition, vol. 41, no. 11, pp. 3427-3435, 2008. [3] t. w. shen and w.j. to mp kins , “bio metric statistical study of one-lead ecg features and body mass index (bm i) ,” 27th annual international conference engineering in medicine and biology society, ieee press , september, 2005. [4] s. a. israel, et al., “ecg to identify individuals,” pa t t e rn re c o gn it io n , vo l. 3 8, n o . 1, p p . 133-142, 2005. [5] k. n. plataniotis, d. hatzinakos, and j. k. m. lee, “analysis of human electrocard iog ra m ecg for b io met ric re cognit ion ,” proc . o f advances in technology innovation , vol. 2, no. 3, 2016, pp. 89 94 94 copyright © taeti bio metrics symposiums , baltimore, usa, 2006. [6] j. kim, et al., “an event detection algorithm in ecg with 60hz interference and bas eline wandering,” proceedings of the 2nd inte rn at iona l con fe ren ce on inte ra ct ion sciences, information technology (icis ‘09), culture and hu man new york, ny, usa , 2009. [7] c. cortes, and v. vapnik, “support-vector network,” machine learn ing, vol. 20, no. 3, pp. 273-297, 1995.  advances in technology innovation, vol. 3, no. 1, 2018, pp. 09 16 single-crossarm stainless steel stayed columns radek pichal, josef machacek * faculty of civil engineering, czech technical university in prague, prague, czech republic . received 05 june 2017; received in revised form 04 july 2017; accept ed 11 august 2017 abstract stability and strength of imperfect stayed columns are studied in 3d using ansys software. following three tests of stayed columns with a reasonable geometry , the numerical modelling is validated and subsequently compared with available analytical and numerical 2d results. arrangement and values of prestressing of ties and initial deflections of the columns affecting the nonlinear stability problem are discussed in a detail. the effect of nonlinear stress -strain relationship corresponding to common stainless steel material is shown, with respect to loading level corresponding to loss of the column stability. the assembly technique of the stayed columns is taken into account, comparing the method and stability/strengths of columns with fixed or sliding stays in the connection to the central crossarm. finally some recommendations concerning the analysis and use of such stayed columns are given. keywords: stayed columns, stainless steel, prestressing, nonlinear buckling, nonlinear material, sliding stays 1. introduction extremely slender compression columns are required particularly in unique structures both by architects and investors. however, the slenderness is limited by a required strength and possible deflections due to buckling. well-known solution for the problem are frequently used prestressed stay columns, made usually from a central slender column, several lateral crossarms each with two planar arms (arranged in a plane with the column) or crossarms in space with three (arranged in 120º) or f our (arranged in 90º) arms and prestressed stays formed by cables or rods, see fig. 1. (a) grande arche, paris (b) parc del centre del poblenou, barcelona (c) estádio algarve, faro fig. 1 examples of the stayed columns used in famous structures * corresponding author. e-m ail address: m achacek@ fsv.cvut.cz tel.: +420776562811 advances in technology innovation, vol. 3, no. 1, 2018, pp. 09 16 copyright © taeti 10 analytical analysis of the stayed columns with multiple pin -connected crossarms and validated by tests was developed by chu and berge [1] in the sixtieth. following analytical research by other authors was accomplished by smith et al. [2] and ha fez et al. [3]. they analyzed a stayed column with one central crossarm rigidly connected (welded) to the column. the column was ideally straight and concentrically loaded, the stays were fixed to the crossarm in ideally hinge connection and the buckling was supposed to occur in the plane of the crossarm member. they analyzed planar arrangement (with two arms of the crossarm) but the solution covered also space arrangement with 4 arms positioned in 90º (fig. 2). the analysis resulted into critical (buck ling) loading for an arbitrary stays prestressing, considering symmetrical or antisymmetrical mode of buckling. however and more important, they also discovered three zones of the behavior depending on the level of the stays prestressing: zone 1 (up to tmin), where the prestressing in the stays disappears when the applied load is less or equal to the euler load (ncr = ne); zone 2 (up to an optimal prestressing topt), where the stays remain effective until the applied load triggers a buckling; zone 3 (above topt), where all the stays remain active (in tension) even after buckling. higher prestressing than topt increases the column loading and, therefore, decreases the critical column load ncr, see fig. 2. n n stays a , e crossarm a , i , e s s a a a column a , i , ec c c x z y t t t t l a aa a n [kn] t t 3toptmin n cr,max n = n cr,min e opt t ~ 0 .4 t o p t z o n e 1 zone 2 zone 3 a b cr critical load n by hafez et al.cr max n for 2a/l = 0.2 max n for 2a/l = 0.1 (a) geometry of the stayed column (b) critical and maximal loadings fig. 2 geometry, critical and maximal loading acc. to wadee et al. (see later) for initial deflection wo = l/200 vs stays prestress numerous other authors investigated influence of initial imperfections (e.g. wong and temple [4], chan et al. [5]). important findings resulted from the later, concerning recommendations for the maximal loading vs optimal pretension of imperfect columns, stay areas and crossarm lengths with respect to bu ckling modes and, therefore, resulting maximal loading. another study of buckling and postbuckling behavior of “nearly perfect” prestressed stayed columns was presented by saito and wadee [6]. they used both analytical and numerical (abaqus software) analy sis and drew attention to stable post-buckling paths in zone 1 and initial part of zone 2, while unstable post-buckling path in zone 3. they also mentioned a danger resulting from changes in the ambient temperature, necessity of modelling in 3d and in [7] studied significance of interactive buckling in a column with one central crossarm (combination of symmetrical and antisymmetrical modes of buckling). two tests of prestressed stayed columns with a reasonable size of l= 12 m and a = 600 mm were performed by araujo et al. [8]. the specimen were placed in horizontal position, the first one with negligible prestressing while the second with typica l prestressing. significant increase of maximal loading in comparison with the capacity of a column without any sta ys was affirmed. the research was accompanied with numerical analysis and parametric studies on values of initial deflections and diameter of stays (cables or rods), while using ansys software. extension of the studies (with a use of tests by servitova and machacek [9]) the authors published in [10]. large experimental analysis of imperfect prestressed stayed columns with one central crossarm was presented by osofero et al. [11]. totally 18 specimens in vertical position with length of l = 2800 mm and variable arms a = 100 ÷ 420 mm were tested. a knife-edge support of the central column forced the buckling into the prescribed plane by either in symmetric or antisymmetric mode, depending mostly on ratio l/a. the tests provided the actual maximal loads nmax depending on the stays prestress t є (0, 4topt), and pointed out to interactive buckling mode for cases where symmetric and antisymmetric buckling loads approximately coincided . advances in technology innovation, vol. 3, no. 1, 2018, pp. 09 16 copyright © taeti 11 complex numerical analysis of imperfect prestressed stayed columns resulting in the design recommendations was published by wadee et al. [12]. they analyzed planar stayed columns with three levels of initial deflection amplitudes ( l/200, l/400, l/1000) for symmetrical and antisymmetrical buckling modes, using abaqus software. after some adjustments to correlate the numerical values with test data, the results are presented for strengths of the columns in a normalized form of nmax/ncr, symmetrical (for 2a/l < 0.175) and antisymmetrical (for 2a/l > 0.175) modes of buckling and four levels of the stays prestressing: zone 1 (t < tmin), zone 2a (t є(tmin, 0,4topt)), zone 2b (t є(0,4topt, topt)) and zone 3 (t є(topt, 3topt)). the results enabled a determination of approximations for maximal ultimate strength nmax for an arbitrary prestressing up to 3topt (shown for two common values of 2a/l in fig. 2, but looking at the picture be aware of the relation to ncr on vertical axis). this article presents a numerical approach of prestressed stainless steel stayed columns with one central crossarm in 3d validated by own tests. the emphasis is laid on nonlinear behavior of the material, space direction of the column buckling and boundary sliding conditions of stays at the crossarm. 2. experiments, numerical modelling and model validation four tests of stainless steel stayed columns with a reasonable length l = 5000 mm were performed at the lab of the czech technical university in prague [9]. the main column was fitted with hinges at both ends (see fig. 3 (a), (b)). the parameters according to fig. 2 were identical for all tests and as fo llows: the central tube ø 50x2 [mm] (l = 5000 mm, ac = 302 mm 2 , ic = 87009 mm 4 , ec,ini = 184 gpa), the crossarm tubes ø 25x1.5 [mm] (a = 250 mm, aa = 111 mm 2 , ia = 7676 mm 4 , ea,ini = 184 gpa), the stays are macalloy cables 1x19 stainless steel ø 4 mm (ls = 2513 mm, as = 12.6 mm 2 , es,ini = 200 gpa). the stays in all tested columns were sliding at steel saddles of crossarms. material of both tubes was tested in a hydraulic testing machine on a weakened cross -section machined from the full cross sections according to fig. 3(b). the stress -strain relationship of the stainless steel material (1.4301) was derived as an average from three such coupon tensile measurements. it should be noted that initial youn g’s modulus for the stays was accepted in the following nu merical analysis due to rather low stresses at collapse loadings, while multilinear isotropic hardening according fig. 3 (a) for the central column and crossarms . 0.002 0.004 0.006 0.008 200 300 400 500 0 0 100 tensile test [mpa]  = 434.1 mpa ansys modelling 0.2 e = 184.0 gpa in e e1 2 e n e 3 e 4 e 5 (a) tested column in the frame position (b) detail of the hinge at supports and central column tensile specimen (c) stress -strain diagram of the austenitic grade 1.4301 stainless steel material fig. 3 assembly and material testing advances in technology innovation, vol. 3, no. 1, 2018, pp. 09 16 copyright © taeti 12 after assembly of each of the stayed columns a careful measurements o f the space initial deflections was recorded using 3d scanning and electric potentiometers (for more details see [9]). because the initial deflection values of the unprestressed c olumns exceeded in some columns the limits prescribed by en 10219-2 (i.e. l/500), the imposed prestressing was slightly uneven in the four stays to result in the later shown acceptable final initial deflections under each of the selected prestressing. ansys software was used for numerical analysis of both tests and in the following parametrical studies. 3d model was formed using for the central and crossarm tubes beam188 and for cable stays link180 (and no -compression option) elements, all embodying large deflections and material nonlinearity. the steel saddles at tips of the crossa rms were modelled using shell281 elements (fig. 4 (a)) with various frictions coefficients between the saddles and cable stays. after fe meshing the division employed in the analysis was l/250, a/25 and shell elements with area of approx. 23 mm 2 . prestress ing of the stays was achieved by a respective thermal change and external loading of the column by an axial displacement x. (a) fe modelling of saddles (b) column 1: comparison of test and ansys analysis fig. 4 ansys detailed modelling and validatio n with ratio 2a/l = 0.1 and initial deflections roughly in one half wave shape, all the final deflections at the tests and numerical analysis followed the first buckling mode, i.e. one half wave. results of tests and validation of numerical approach is sh own for all 4 tests, using friction coefficient between the saddles and cable stays ν = 0.1 (common for steel-steel friction). column 1: the total applied prestressing in all four stays was 4t = 5.44 kn. the prestressing in each of the 4 stays was slightly different to receive required global imperfection, which in this case had amplitudes at midspan of w0y = 1.9 mm and w0z = 8.3 mm. the test exhibited linear behavior up to approx. 15 kn, followed by a rapid growth of deflection up to maximal ultimate loading of nmax,exp = 17.7 kn and terminated due to enormous deflection, see fig. 4 (b). numerical analysis in 3d covered the initial deflections of one half-sine shape with the above given amplitudes and initial total prestressing 4t, requiring slightly different prestressing in each of the four stays. the comparison of test and numerical analysis is shown in fig. 4 with good agreement. column 2: the total applied prestressing in this case was 4t = 4.54 kn, arranged in the similar procedure as for the column 1, resulting in initial deflections with the mid -span amplitudes w0y = 3.8 mm and w0z = 19.9 mm. up to the external load of 12.5 kn the behavior was nearly linear, followed by enormous increase of deflections and maximal ultimate load of 14.9 kn, when the t est was terminated. here the numerical maximal loading exceeded the test value of approx. 9 %, which was assigned to difficulties with modelling of various prestressing of the 4 stays and keeping the deflections as in the test. column 3: first, column 3a was tested without stays (in the lab reality with attached, but slacked stays). the amplitudes of initial deflections at midspan were w0y = 0.3 mm and w0z = 1.4 mm. the test was terminated due to sudden increase of central deflection (see fig. 5 (b)) under loading nmax,exp  6.5 kn, while common euler’s critical load is ne = 6.3 kn (the difference amounts for 2.8 %). numerical analysis of initially deflected column without stays gives nmax = 6.0 kn < ne. saddle stay crossarm advances in technology innovation, vol. 3, no. 1, 2018, pp. 09 16 copyright © taeti 13 (a) comparison of test and ansys analysis for column 2 (b) comparison of test and ansys analysis for column 3 fig. 5 comparison of test and ansys analysis second, the stays in the column 3b were slightly prestressed with total value of 4t = 3.9 kn and resulting final central amplitudes of initial deflections were w0y = 0.5 mm and w0z = 2.2 mm. this test was terminated due to enormous deflections, giving maximal ultimate load nmax = 16.2 kn. instead of numerical analysis of the slightly prestressed column the results of the column with fully slacked (unprestressed) stays is shown for comparison in fig. 5 (b). the numerical results with nmax = 17.0 kn revealed a small “jump” at point a on the load -deflection curve at the level of critical loading, when a buckling activated initially slacked stays on the concave side of the column. the positive influence of the unprestressed stays on the ultimate loading is, therefore, confirmed in agreement with results of [12]. comparison of results of the three tests with the proposed numerical modelling is satisfying an d justifies use of the model for the following numerical studies . 3. critical and maximal loading in 2d and 3d, material nonlinearity numerical analysis of the prestressed stayed columns needs to respect the 3 zones according to the prestress of stays (see fig. 2). as explained in the introduction, the behavior in the zone 2 involves sudden change of the assembly inner energy due to the central column buckling and instant activating of stays on convex side of the column. therefore, linear buckling analysis can’t be used and geometrically nonlinear one is necessary. however, in such case gnia (geometrically nonlinear analysis with imperfections) need to be used, with negligible initial deflections. in the study were considered values of w0y = w0z = l/500000 = 0.01 mm. direction of maximal deflections in all former described tests was into space (i.e. between the axes y, z). the question arose, whether solution in 3d with the corresponding initial deflections in both axes y, z gives lower critical/maximal loadings. the studied stayed column had the same geometry as in chapter 2, but with fixed stays at the crossarms: the central tube ø 50x2 [mm], the crossarm tubes ø 25x1.5 [mm] the stays as macalloy cables ø 4 mm. stainless steel material with e = 200 mpa was considered for all the central tube, crossarms and stays (here as an initial value due to low stresses). comparison of results for 2d an alysis according to [3] and fem in 3d with four amplitudes of initial deflections (w0 = 0.01 mm, 0.05 mm, 0.10 mm and 25 mm) is shown in table 1. the greatest amplitude corresponds to the design recommendation of eurocode en 1993-1-1 for cold-formed tubes and elastic analysis (l/200 = 25 mm). both symmetric (one half wave initial deflection) and antisymmetric (two half waves with half amplitudes of initial deflection) were analyzed to determine topt and corresponding maximal critical loading or maximal strength nmax of imperfect column in the prestressing range up to 3topt, fig. 6. advances in technology innovation, vol. 3, no. 1, 2018, pp. 09 16 copyright © taeti 14 table 1 influence of initial deflection w0, comparison of 2d and 3d analysis , material nonlinearity w0 [mm] symmetrical initial deflections antisymmetrical initial deflections nmax [kn] topt [kn] nmax,sym [kn] topt [kn] nmax,anti [kn] 0 (2d) 1.41 39.79 1.30 36.79 36.79 0.01 (3d) 1.51 39.73 1.35 36.18 36.18 0.05 (3d) 1.58 39.25 1.43 35.77 35.77 0.10 (3d) 1.61 38.62 1.52 35.43 35.43 25.0 (3d) 22.74 24.84 22.74 0.01 (stainless steel) 1.51 36.54 1.27 31.58 31.58 25.0 (stainless steel) 19.57 19.92 19.57 after analysis of the results , it was concluded (see the first two rows in table 1), that 2d and 3d results, in spite of various directions of buckling (in direction of arms for 2d and into the space in 3d), provide nearly identical critical/strength values. this conclusion may be considered not only for the “ideal” column (i.e. for critical loading), but also for maximal ultimate loading (strength of imperfect stayed columns). the last row of the table 1 presents results for the same stayed column as above but made from stainless steel ma terial as in the tests (see chapter 2). it means that for central column and crossarms instead of constant e = 200 mpa the respective v alues of e1 = 184 mpa, e2 etc. (see fig. 3) were employed. the results for nmax are significantly lower for the stainless steel material and simple reduction from elastic behavior using initial ratio e1/e  0.92 is not sufficient. the gmnia results for stainless steel material are shown in fig. 6 (b). it is obvious that for the given geometry, design initial deflections l/200 and decisive role of symmetric initial deflections (because 2a/l = 0.1 < 0.175), the ratio nmax/ncr is similar to the one given in [12]. w 0 w 0 w 0 5 0 0 0 250250 x y,z x symmetrical mode antisymmetrical mode (a) initial deflection shape (b) comparison of values for “ideal” (w0 = l/500000) and initially deflected (w0 = l/200) stainless steel stayed column fig. 6 gmnia results for “ideal” and imperfect stayed column the changes in deflection shape for such large antisymmetric initial deflections with an increase of prestressing are highlighted by points 1 and 2 in fig. 6. the antisymmetric initial mode (however not decisive in this geometry) is changed after prestressing to an interactive one up to the point 1 and then, from point 2, to antisymmetric one again (see fig. 7) . point 1 point 2 fig. 7 deflection shapes for antisymmetric initial deflections and prestressing corresponding to points 1, 2 advances in technology innovation, vol. 3, no. 1, 2018, pp. 09 16 copyright © taeti 15 4. support of stays at crossarms the stays are formed either by rods or cables. using rods naturally requires fixed fastening to the crossarms, but cables may be fastened either by forked terminal (i.e. fixed) or run continuously over saddles. in the tests described in chapter 2 , the saddles were used, enabling balancing tension in both parts of the respective stay after exceeding frictio n between the saddle and the cable, see fig. 8 (a). such arrangement is advantageous from assembly point of view and saving cable sockets, but obviously reduces in some way strength of the stayed column system. sliding stays fixed stays crossarm leg stay stay crossarm leg saddle forked socket (a) support at crossarm (b) influence of cable slip at saddles fig. 8 alternatives and value of support conditions at crossarm the friction between the saddles and cables may vary between from nearly zero (when using teflon -like lining) and common value for steel-steel contact of ν = 0.10. to analyze the influence of the cable slip at saddles safely, very low friction coefficient ν = 0.01 was used and results are shown in fig. 8 (b). the columns with geometry given in chapter 3 and initial deflection of the central column w0 = l/500000 were analyzed using gmnia. from comparison of results is obvious that the different behavior arises at antisymmetrical mode of buckling only, were reduction of maximal critical load is substantial. extensive studies concerning pres tressed stayed columns with sliding stays and required necessary initial deflections are in progress . 5. conclusions the proposed numerical ansys model was successfully validated by comparison with the four tests of stainless steel stayed columns of reasonable geometry and various prestressing of stays . the detailed studies resulted into the following conclusions: (1) the stayed column even with unprestressed stays (slacked stays) provides significantly higher maximal ultimate loading in comparison with simple column without stays due to activating of stays at the concave side of the column during the buckling. (2) the 2d planar analysis (fe or analytical one) with a buckling in the direction of the arms supplies nearly identical results concerning optimal prestressing, maximal critical and maximal ultimate loading as the 3d space analysis with buckling into space (in between the arms of the crossarm). (3) using nonlinear (stainless steel) material in stayed columns requires gmnia and proper introduction of stainless steel stress -strain relationship. (4) maximal ultimate loading (strengths nmax) of a stayed column must be analyzed with initial deflections of appropriate amplitude and shape. with reasonable amplitudes (e.g. l/200 for cold-formed tubes required by eurocode 3) and common ratios 2a/l, both symmetric and antisymmetric initial deflections may be decisive and corresponding ratios nmax/ncr be greater (for low prestressing) or much lower (for great prestressing) than 1. (5) continuous stays, running over saddles, may be adv antageous for assembly, but when antisymmetric buckling is predominant, a strong reduction in maximal ultimate loading need to be expected . advances in technology innovation, vol. 3, no. 1, 2018, pp. 09 16 copyright © taeti 16 acknowledgement the s upport of the czech grant agency grant gacr no. 17-24769s is gratefully acknowledged. references [1] k. h. chu and s. s. berge, “analysis and design of struts with tension ties ,” journal of the structural division asce, vol. 89, no. st1, pp. 127-163, february 1963. [2] r. j. smith, g. t. mccaffrey, and j. s. ellis, “buckling of a single-crossarm stayed column,” journal of the structural division asce, 11071 st1, pp. 249-268, january 1975. [3] h. h. hafez, m. c. temple, and j. s. ellis, “pretensioning of single-crossarm stayed columns ,” journal of the structural division asce, 14362 st2, pp. 359-375, february 1979. [4] k. c. wong and m. c. temple, “stayed columns with initial imperfections ,” journal of the structural division asce, 108 st7, pp. 1623-1640, 1982. [5] s. l. chan, g. shu, and z. lü, “stability analysis and parametric study of pre-stressed stayed columns ,” engineering structures , vol. 24, no. 1, pp. 115-124, january 2002. [6] d. saito and m. a. wadee, “post-buckling behaviour of prestressed steel stayed columns ,” engineering structures, vol. 30, no. 5, pp. 1224-1239, may 2008. [7] d. saito and m. a. wadee, “numerical s tudies of interactive buckling in prestressed steel stayed columns ,” engineering structures, vol. 31, no. 2, pp. 432-443, february 2009. [8] r. r. araujo, s. a. l. andrade, p. c. g. s. vellasco, j. g. s. silva, and l. r. o. lima, “experimental and numerical as sessment of stayed steel columns ,” journal of constructional steel research, vol. 64, no. 9, pp. 1020-1029, september 2008. [9] k. servitova and j. machacek, “analysis of stainless steel stayed columns ,” proceedings 6 th intern. symp. steel structures, korean society of steel construction, seoul, pp. 874-881, 2011. [10] l. r. o. lima, p. c. g. vellasco, and j. g. s. silva, “numerical modelling of prestressed stayed stainless steel columns ,” tubular structures xiv, taylor and francis, london, pp. 377-382, 2012. [11] a. i. osofero, m. a. wadee, and l. gardner, “experimental study of critical and post -buckling behaviour of prestressed stayed steel columns ,” journal of constructional steel research, vol. 79, pp. 226-241, december 2012. [12] m. a. wadee, l. gardner, and a. i. osofero, “design of prestressed stayed columns ,” journal of constructional steel research, vol. 80, pp. 287-298, january 2013. [13] r. pichal and j. machacek, “3d stability of prestressed stayed columns ,” proceedings 22 nd intern. conf. engineering mechanics, inst. of thermomechanics, academy of sciences of the czech republic, svratka, pp. 462-465, 2016. [14] r. pichal and j. machacek, “buckling and post-buckling of prestressed stainless steel stayed columns ,” procedia engineering, vol. 172, pp. 875-882, 2017. 5___aiti#7835_in press advances in technology innovation, vol. 6, no. 4, 2021, pp. 251-261 optimization of superplastic forming process of aa7075 alloy for the best wall thickness distribution manh tien nguyen 1,* , truong an nguyen 1 , duc hoan tran 1 , van thao le 2 1 faculty of mechanical engineering, le quy don technical university, ha noi, viet nam 2 advanced technology center, le quy don technical university, ha noi, viet nam received 02 june 2021; received in revised form 27 july 2021; accepted 28 july 2021 doi: https://doi.org/10.46604/aiti.2021.7835 abstract this work aims to optimize the process parameters for improving the wall thickness distribution of the sheet superplastic forming process of aa7075 alloy. the considered factors include forming pressure p (mpa), deformation temperature t (°c), and forming time t (minutes), while the responses are the thinning degree of the wall thickness ε (%) and the relative height of the product h*. first, a series of experiments are conducted in conjunction with response surface method (rsm) to render the relationship between inputs and outputs. subsequently, an analysis of variance (anova) is conducted to verify the response significance and parameter effects. finally, a numerical optimization algorithm is used to determine the best forming conditions. the results indicate that the thinning degree of 13.121% is achieved at the forming pressure of 0.7 mpa, the deformation temperature of 500°c, and the forming time of 31 minutes. keywords: superplastic forming, wall thickness distribution, optimization, response surface, aa7075 alloy 1. introduction aa7075 is an aluminum alloy with outstanding mechanical properties such as high strength, light weight, and corrosion resistance. therefore, aa7075 alloy is widely used in the aerospace and automotive industries. this alloy has been used to replace high strength steel on most types of cars and commercial aircraft [1-2]. however, the deformability of this alloy is limited at room temperature. the elongation to fracture for tensile tests of this alloy was achieved below 10% at room temperature (20°c), resulting in its limited applicability in forming processes. to improve its deformability, other forming methods have been used to expand the application of the alloy. superplastic forming (spf) is one of the widely used methods. spf is conducted within a limited range of conditions such as material microstructure, strain temperature, and strain rate. when conditions of spf are met, the material will achieve a degree of deformation many times greater than that of the normal strain condition. in uniaxial tension, the elongations in excess of 200% are usually indicative of superplasticity [3-6]. the forming process for hollow parts from a flat sheet or a pre-profile blank or a tube blank under the action of compressed air pressure is one of the spf methods. the received product may have a shape of forming tools [7-9]. this process can bring some attractive advantages, including large deformation ability, complex products, and a simple or no forming tools required. moreover, the spf technology has a great advantage in the production of single or small series of complex parts from high strength materials. however, this technology also has serious disadvantages that are an uneven distribution of wall thickness after the deformation process greatly affects the quality and working ability of the product [10-11]. * corresponding author. e-mail address: nguyenmanhtiengcalmta@gmail.com tel.: +84 977 817 623 advances in technology innovation, vol. 6, no. 4, 2021, pp. 251-261 the spf by compressed air pressure depends on many parameters such as forming pressure, deformation temperature, forming time, tool size, coefficient of friction, back pressure, etc. [12-15]. the degree of influence of each parameter on the deformation ability is significantly different. the wall thickness distribution of the product is the prioritized criterion in terms of the formed part quality. several studies on improving the distribution of the wall thickness have been carried out. shojaeefard et al. [16] have studied the effect of the die entry radius and the interfacial friction coefficient on the required forming time and thickness distribution of the product. their results have shown that with lower friction coefficients, the sensitivity of the formed part thinning factor to the die entry radius variations is high. it was also concluded that a larger die entry radius along with a lower friction coefficient leads to a shorter forming time and a more uniform final part thickness distribution. jarrar [17] found that for free-bulge forming process of aa5083 at 500°c, single-, two-, and three-step isobaric forming process produce the same part thickness uniformity within shorter forming time compared to the forming process at a constant maximum strain rate. kumaresan et al. [18] performed the finite element simulation using abaqus software to evaluate the deviation of the thickness distribution between the simulation and experiment. the process parameters selected for simulation include pressure, forming time, and initial sheet thickness. also, balasubramanian et al. [19] experimentally and numerically studied the three-stage forming process. discontinuities in the thickness of the formed profile have been observed due to the transition of radius from one stage to the next. results of this study have revealed that an optimum pressure of 0.5 mpa for achieving uniform thickness in the profile exists. alirezaiee et al. [10] proposed a new method to produce a part with a fairly uniform thickness distribution. in their proposed method, a die having a movable segment is used. by moving this segment during the process, the part is formed during two stages and final thickness distribution is controlled. one of the factors affecting the forming process, which has also been mentioned in many studies, is the change in the microstructure of the material during the spf process. since the grain size is a preliminary criterion for obtaining superplasticity in any material, its stability at the deformation temperature plays a significant role. the deformation ability of the material will be significantly reduced as the average grain size increases during the spf process [20-21]. moreover, abnormal grain growth is also a major metallurgical aspect in superplastic deformation [22-23]. however, the aforementioned publications indicated that the process parameter optimizations for sheet spf processes about the formed part quality have not been thoroughly investigated. recently, many technological solutions have been researched to improve the thinning degree of the wall thickness and increase the product quality. the new spf process was developed based on the back pressure principle. besides hardware developments, forming parameter optimization is still an effective approach in terms of costs and time. the aa7075 high-strength aluminum alloy was selected as the workpiece material due to its wide industrial applications. this work aims to optimize the process parameters for improving the wall thickness distribution of the sheet spf process of aa7075 alloy. the considered factors include forming pressure p (mpa), deformation temperature t (°c), and forming time t (minutes), while the responses are the thinning degree of the wall thickness ε (%) and the relative height of the product h*. first, a series of experiments are conducted in conjunction with the response surface method (rsm) to render the relationship between inputs and outputs. subsequently, an analysis of variance (anova) is conducted to verify the response significance and parameter effects. finally, a numerical optimization algorithm is used to determine the best forming conditions. 2. research methodology 2.1. optimization problem the thinning degree of the wall thickness ε (%) is one of the important criteria with regard to the quality of spf products. the goal of the current work is to optimize the process parameters, including forming pressure, deformation temperature, and forming time for minimizing the thinning degree. the schematic diagram of the sheet spf process under the action of 252 advances in technology innovation, vol. 6, no. 4, 2021, pp. 251-261 compressed air pressure is depicted in fig. 1. the die is fastened by bolts, and the pneumatic tube into the die is welded with a sealing cap. the 45 steel is selected as the die material due to the medium forming temperature. the die and the stroke sensor are placed in the furnace during deformations. fig. 1 schematic diagram of the sheet spf process as shown in table 1, the limits of variables are denoted as -1, 0, and +1 for the lower value, middle value, and higher value, and the input variables are denoted as a for forming pressure (mpa), b for deformation temperature (°c), and c for forming time (min). the parameter ranges are determined based on the references [7, 11, 24]. consequently, the optimizing issue is to find the minimum thinning degree based on the optimization process parameters [25]. table 1 levels and their values of the process parameters symbol parameters level -1 level 0 level +1 p a: forming pressure (mpa) 0.7 0.8 0.9 t b: deformation temperature (°c) 500 515 530 t c: forming time (min) 20 30 40 2.2. optimization framework in the box-behnken design (bbd), three levels for each design parameter or factor are required, with all design points lying on the same sphere and with at least three or five runs at the center point. bbd is applied instead of the full-factorial one to decrease the number of experiments and ensure the predicting accuracy. the predictive mathematical model of the thinning degree is then developed with respect to process parameters using rsm [26-28]. subsequently, a fitness analysis is conducted in order to investigate the significance of the proposed model and the factor is chosen. to solve the optimization problem, an advanced technique entitled desirability function (df) is adopted for obtaining optimal values. in addition to the design points, a set of random points are checked to see if there is a more desirable solution. the influence of the parameters on the objective function will help the technology design process. the feasible solutions could be obtained with a minimal cost, from a problem that potentially has a great number of solutions [29-30]. as a result, an integrative approach combining bbd, rsm, and df can be considered a powerful method to find global optimal process parameters with a smaller number of experiments as well as less computational effort. 3. experimental procedures the experimental specimens are cut from a flat sheet of the thickness of 1.2 mm to have a diameter of 50 mm. the material used in this study is an aa7075 alloy. the chemical composition of the aa7075 alloy listed in table 2 is expressed in weight percent (%wt). the aa7075 alloy is prepared to fine stable grain size with an average grain size of about 13 µm. the experimental tests are carried out under spf conditions. the experimental equipment system is shown in fig. 2. 253 advances in technology innovation, vol. 6, no. 4, 2021, pp. 251-261 table 2 chemical composition of the analyzed aa7075 alloy zn mg cu fe cr ti zr mn si al 5.35 2.34 1.32 0.30 0.22 0.04 0.0027 0.024 0.05 balance fig. 2 the experimental equipment system the deformed specimens are exhibited in fig. 3. the position of wall thickness measurement is determined as shown in fig. 4. at that position, the thinning degree is the greatest. the wall thickness values are measured by a jena universal microscope. the overall height of the product is determined and shown in fig. 5. the average response values are observed repeatedly three times. the experimental data of the thinning degree of the wall thickness ε and the relative height of the product h* are exhibited in table 2. the thinning degree of the wall thickness is the ratio between the amount of thinning at a dangerous position and the thickness calculated according to the law of constant volume. the relative height of the product is the ratio between the overall height hi and outside diameter dp of the part. fig. 3 the specimens after spf processes fig. 4 the position of wall thickness measurement of deformed specimens fig. 5 determination of the overall height hi and outside diameter dp of the product 254 advances in technology innovation, vol. 6, no. 4, 2021, pp. 251-261 4. results and discussion 4.1. model fitness investigation in this study, anova is used to determine the adequacy and significance of models. in addition, it is used to evaluate the effect of lack-of-fit on the models and the significance of the individual model coefficient. there are 15 test trials performed based on the design of the experiment method as shown in table 3. anova results of the thinning degree and relative height are shown in table 4 and table 5. the significant coefficients are selected based on the p-values of the parameters considered. the parameters are modeled using a confidence level of 95%. hence, the factors with p-values less than 0.05 are considered significant. table 3 experimental results number of experiments factor 1 factor 2 factor 3 responses a: forming pressure (mpa) b: deformation temperature (°c) c: forming time (min) r1: thinning degree ε (%) r2: relative height of the product h* 1 -1 -1 0 13 0.25 2 +1 -1 0 15 0.32 3 -1 +1 0 16.8 0.30 4 +1 +1 0 16.5 0.37 5 -1 0 -1 16.5 0.34 6 +1 0 -1 17.8 0.27 7 -1 0 +1 17 0.40 8 +1 0 +1 18 0.49 9 0 -1 -1 17.8 0.30 10 0 +1 -1 18.4 0.42 11 0 -1 +1 17 0.33 12 0 +1 +1 20 0.45 13 0 0 0 17.4 0.48 14 0 0 0 17.2 0.48 15 0 0 0 17 0.44 table 4 anova result for thinning degree model source sum of squares degree of freedom mean square f-value p-value significance model 33.48 9 3.72 38.16 0.0004 significant a-a 2.00 1 2.00 20.51 0.0062 b-b 9.90 1 9.90 101.55 0.0002 c-c 0.2813 1 0.2813 2.88 0.1502 ab 1.32 1 1.32 13.56 0.0142 ac 0.0225 1 0.0225 0.2308 0.6512 bc 1.44 1 1.44 14.77 0.0121 a² 7.50 1 7.50 76.90 0.0003 b² 0.7477 1 0.7477 7.67 0.0394 c² 8.87 1 8.87 90.98 0.0002 residual 0.4875 5 0.0975 lack-of-fit 0.4075 3 0.1358 3.40 0.2358 not significant pure error 0.0800 2 0.0400 corrected total 33.97 14 r² 0.9856 adjusted r² 0.9598 predicted r 2 0.8028 adequate precision 27.7014 255 advances in technology innovation, vol. 6, no. 4, 2021, pp. 251-261 as shown in table 4 and table 5, the f-values 38.16 and 25.93 indicate the significance of the quadratic models. in these models, all the single terms (p, t, and t) and quadratic terms (p 2 , t 2 , and t 2 ) are found to be significant model terms. the insignificant terms (pt, pt, and tt) should be eliminated in the design space in order to save computational time. additionally, the r 2 -value of the thinning degree is 0.9856, indicating that 98.56% of the total variations are explained by the model. the r 2 -value of the relative height is 0.9790, indicating that 97.9% of the total variations are explained by the model. moreover, the comparison results from the adjusted r 2 -value and predicted r 2 -value using anova indicate that the quadratic model is more accurate than the linear one. adequate precision is used to measure the signal to noise ratio. the ratio greater than 4 is desirable. in the thinning degree model and the relative height model, the adequate precisions are 27.7014 and 15.1235 respectively, indicating adequate signals. these models can be used to navigate the design space. the lack-of-fit is not significant relative to the pure error. non-significant lack-of-fit is good. table 5 anova result for relative height model source sum of squares degree of freedom mean square f-value p-value significance model 0.0836 9 0.0093 25.93 25.93 significant a-a 0.0036 1 0.0036 10.08 10.08 b-b 0.0136 1 0.0136 37.99 37.99 c-c 0.0162 1 0.0162 45.21 45.21 ab 0.0006 1 0.0006 1.74 1.74 ac 0.0121 1 0.0121 33.77 33.77 bc 0.0009 1 0.0009 2.51 2.51 a² 0.0285 1 0.0285 79.64 79.64 b² 0.0103 1 0.0103 28.85 28.85 c² 0.0009 1 0.0009 2.45 2.45 residual 0.0018 5 0.0004 lack-of-fit 0.0009 3 0.0003 0.7115 0.7115 not significant pure error 0.0009 2 0.0004 corrected total 0.0854 14 r² 0.9790 adjusted r² 0.9413 predicted r 2 0.8039 adequate precision 15.1235 4.2. model for thinning degree and relative height the models of thinning degree and relative height are developed in terms of input parameters using rsm. from the experimental values, the coefficients of the regression equations are calculated. the regression coefficients of insignificant terms are eliminated based on anova results. consequently, the regression response surface models showing the thinning degree of the wall thickness ε and the relative height h* are expressed by eq. (1) and eq. (2) which show the final equation in terms of coded factors. 2 2 2 17.2 0.5 1.1125 0.1875 0.575 0.075 0.6 1.425 0.45 1.55 a b c a b a c b c a b c ε = + × + × + × − × × − × × + × × − × − × + × (1) 2 2 2 0.4633 0.02112 0.0412 0.045 0.0125 0.55 0.015 0.0879 0.0529 0.0154 * a b c a b a c b c a b c h = + × + × + × − × × + × × − × × − × − × − × (2) fig. 6(a) and fig. 6(b) represent the normal probability curves for thinning degree and relative height respectively to check the adequacy of models. as all points are in a straight line, it can be concluded that the models are adequate [29-30]. 256 advances in technology innovation, vol. 6, no. 4, 2021, pp. 251-261 (a) thinning degree (b) relative height fig. 6 residual plots for the developed models 4.3. factor effect analysis 4.3.1. effect of process parameters on thinning degree ε the effects of process parameters on the objective are investigated using the contour plots (2d and 3d) which are considered in full parameter ranges, at each level of the focal position. fig. 7 shows that an increase of the forming time results in a decreased thinning degree, until the optimum value of around 30 min. the thinning degree is then increased with the continuous increase of the forming time. besides, when the strain temperature increases from 500°c to 530°c, the degree of thinning increases. this phenomenon can be explained as follows. when the forming time and deformation temperature increase, the grain grows faster. therefore, it reduces the deformation rate and deformation degree of the specimens. the degree of thickness distribution becomes uneven, the thinning degree at the dangerous position will increase [7, 11, 24]. as shown in fig. 7, when the forming pressure increases, the thinning degree increases. however, when the pressure increases beyond 0.8 mpa, the thinning degree of the wall thickness at the dangerous position decreases. this phenomenon can be explained as follows. when the strain rate is high, the time is not enough for the corresponding diffusion process to create the planes favorable for sliding, so the stress increases and the deformation degree of the specimen decreases [24]. (b) 3d interaction plots of forming pressure and deformation temperature fig. 7 effect of processing parameters on the thinning degree 257 (a) 2d perturbation plot of process parameters advances in technology innovation, vol. 6, no. 4, 2021, pp. 251-261 (c) 3d interaction plots of forming pressure and forming time (d) 3d interaction plots of forming time and deformation temperature fig. 7 effect of processing parameters on the thinning degree (continued) 4.3.2 effect of process parameters on the relative height of product h* in fig. 8, it is shown that the influence of process parameters including forming pressure, deformation temperature, and forming time leads to the relative height of the product. as the forming pressure and deformation temperature increase, the relative height increases. however, when the forming pressure and strain temperature increase to a certain value, the relative height of the product tends to decrease. because of this, the higher strain rate and high temperature will make the grain microstructure of the material grow, reducing the possibility of deformation [24]. the forming time increases (from 20 to 40 min), and the relative height of the product increases accordingly. besides, with a long forming time, the thinning degree of the wall thickness increases. this may result in product destruction. (b) 3d interaction plots of forming pressure and deformation temperature (c) 3d interaction plots of forming pressure and forming time (d) 3d interaction plots of forming time and deformation temperature fig. 8 effect of processing parameters on the relative height of the product 258 (a) 2d perturbation plot of process parameters advances in technology innovation, vol. 6, no. 4, 2021, pp. 251-261 5. optimization results after building the statistical regression equation showing the relationship between process parameters and process response, the equation is used to solve the optimization problem. according to the discussion above, it can be concluded that the process parameters of forming pressure, deformation temperature, and forming time have significant and complex effects on the thinning degree of the wall thickness model. the optimization issue can be solved using rsm. table 6 shows that the thinning degree of the wall thickness ε of 13,121% is observed at p = 0.7 mpa, t = 500°c, and t = 31 min, respectively. table 6 optimization results using rsm constraints name goal lower limit upper limit lower weight upper weight importance a:a in range 0.7 0.9 1 1 3 b:b in range 500 530 1 1 3 c:c in range 20 40 1 1 3 r1 minimize 13 20 1 1 3 solutions number a b c r1 desirability selection 1 0.700 500.000 31.006 13.121 0.983 selected 2 0.700 500.000 31.123 13.121 0.983 3 0.700 500.013 30.671 13.125 0.982 4 0.700 500.000 30.806 13.126 0.982 5 0.700 500.000 31.616 13.127 0.982 6. conclusions in this study, an integrative approach using the physical experiment, rsm model, and the df algorithm was used to optimize the process parameters concerning the thinning degree of the wall thickness. the relationship between process parameters and the thinning degree was constructed through rsm and experimental data. df was applied to determine the optimal design variables and the response. the main conclusions from the research results of the current work can be drawn as follows: (1) the thinning degree is decreased with a higher value of the forming time until it reaches the optimal points. with increased factors, the thinning degree increases. additionally, an increased forming pressure or deformation temperature results in a higher thinning degree value. (2) the mathematical model of the thinning degree developed using bbd and rsm predicts the response values with sufficient accuracy. (3) in the range of studied parameters, the thinning degree of 13,121% is achieved at p = 0.7 mpa, t = 500°c, and t = 31 min. (4) the research results allow determining the relative height of the product h* corresponding to the thinning degree of the wall thickness ε in the range of allowable limit values. conflicts of interest the authors declare no conflict of interest. references [1] n. e. prasad, a. a. gokhale, and r. j. h. wanhill, aluminum-lithium alloys: processing, properties, and applications, oxford: elsevier, 2014. [2] e. a. starke and j. t. staley, “application of modern aluminum alloys to aircraft,” progress in aerospace sciences, vol. 32, no. 2-3, pp. 131-172, 1996. 259 advances in technology innovation, vol. 6, no. 4, 2021, pp. 251-261 [3] r. j. bhatt, “advanced approaches in superplastic forming—a case study,” journal of basic and applied engineering research, vol. 1, pp. 5764, october 2014. [4] i. charit, r. s. mishra, and m. w. mahoney, “multi-sheet structures in 7475 aluminum by friction stir welding in concert with post-weld superplastic forming,” scripta materialia, vol. 47, no. 9, pp. 631-636, november 2002. [5] s. x. mcfadden, r. s. mishra, r. z. valiev, a. p. zhilyaev, and a. k. mukherjee, “low-temperature superplasticity in nanostructured nickel and metal alloys,” nature, vol. 398, no. 6729, pp. 684-686, april 1999. [6] y. wang and r. s. mishra, “finite element simulation of selective superplastic forming of friction stir processed 7075 al alloy,” materials science and engineering: a, vol. 463, no. 1-2, pp. 245-248, august 2007. [7] g. giuliano, superplastic forming of advanced metallic materials: methods and applications, cambridge: woodhead, 2011. [8] c. h. hamilton and n. e. paton, superplasticity and superplastic forming: proceedings of an international conference on superplasticity and superplastic forming, warrendale: the minerals, metals, and materials society,1989. [9] c. k. syn, m. j. o’brien, d. r. lesuer, and o. d. sherby, “an analysis of gas pressure forming of superplastic al 5083 alloy,” international conference on light materials for transportation systems, may 2001, pp. 1-8. [10] m. alirezaiee, r. j. nedoushan, and d. banabic, “improvement of product thickness distribution in gas pressure forming of a hemispherical part,” proceedings of the romanian academy, vol. 17, no. 3, pp. 245-252, july-september 2016. [11] t. g. nieh, j. wadsworh, and o. d. sherby, superplasticity in metals and ceramics, cambridge: cambridge university press, 2005. [12] a. o. caballero, o. a. ruano, e. f. rauch, and f. carreño, “severe friction stir processing of an al-zn-mg-cu alloy: misorientation and its influence on superplasticity,” materials and design, vol. 137, pp. 128-139, october 2017. [13] j. r. groza, j. f. shackelford, m. t. lavernia, and m. t. powers, materials processing handbook, boca raton: crc press, 2007. [14] a. d. jose and j. babu, “experimental studies on thinning characteristics of superplastic hemi-spherical forming,” international journal of emerging technology and advanced engineering, vol. 5, no. 1, pp. 104-109, january 2015. [15] a. smolej, b. skaza, and m. fazarinc, “determination of the strain-rate sensitivity and the activation energy of deformation in the superplastic aluminum alloy al-mg-mn-sc,” materials and geoenvironment, vol. 56, no. 4, pp. 389-399, december 2009. [16] m. h. shojaeefard, a. khalkhali, and e. miandoabchi, “effects of process parameters on superplastic forming of a license plate pocket panel,” advanced design and manufacturing technology, vol. 7, no. 2, pp. 25-33, june 2014. [17] f. jarrar, “designing gas pressure profiles for aa5083 superplastic forming,” procedia engineering, vol. 81, pp. 1084-1089, october 2014. [18] g. kumaresan and a. jothilingam, “experimental and fe simulation validation of sheet thickness optimization in superplastic forming of al alloy,” journal of mechanical science and technology, vol. 30, no. 7, pp. 3295-3300, september 2016. [19] m. balasubramanian, p. ganesh, k. ramanathan, and v. s. s. kumar, “superplastic forming of a three-stage hemispherical 5083 aluminium profile,” journal of mechanical engineering, vol. 61, no. 6, pp. 365-373, arpil 2015. [20] d. harwani, v. badheka, v. patel, and j. andersson, “developing superplasticity in magnesium alloys with the help of friction stir processing and its variants—a review,” journal of materials research and technology, vol. 12, pp. 2055-2075, may-june 2021. [21] v. patel, v. badheka, w. li, and s. akkireddy, “hybrid friction stir processing with active cooling approach to enhance superplastic behavior of aa7075 aluminum alloy,” archives of civil and mechanical engineering, vol. 19, no. 4, pp. 1368-1380, august 2019. [22] v. patel, w. li, a. vairis, and v. badheka, “recent development in friction stir processing as a solid-state grain refinement technique: microstructural evolution and property enhancement,” critical reviews in solid state and materials sciences, vol. 44, no. 5, pp. 378-426, july 2019. [23] v. v. patel, v. badheka, and a. kumar, “friction stir processing as a novel technique to achieve superplasticity in aluminum alloys: process variables, variants, and applications,” metallography, microstructure, and analysis, vol. 5, no. 4, pp. 278-293, august 2016. [24] n. m. tien, n. t. an, t. d. hoan, l. d. giang, and l. t. tan, “experimental study on effects of process parameters on superplastic deformation ability of 7075 aluminium alloy using taguchi method,” international conference on engineering research and applications, december 2019, pp. 328-334. [25] c. f. wu and m. hamada, experiments: planning, analysis, and optimization, hoboken: wiley, 2009. 260 advances in technology innovation, vol. 6, no. 4, 2021, pp. 251-261 [26] s. gopalakannan, t. senthilvelan, and s. ranganathan, “modeling and optimization of edm process parameters on machining of al 7075-b4c mmc using rsm,” procedia engineering, vol. 38, pp. 685-690, 2012. [27] r. kumar and s. chauhan, “study on surface roughness measurement for turning of al 7075/10/sicp and al 7075 hybrid composites by using response surface methodology (rsm) and artificial neural networking (ann),” measurement, vol. 65, pp. 166-180, april 2015. [28] e. salur, a. aslan, m. kuntoglu, a. gunes, and o. s. sahin, “experimental study and analysis of machinability characteristics of metal matrix composites during drilling,” composites part b: engineering, vol. 166, pp. 401-413, june 2019. [29] s. a. sonawane and m. l. kulkarni, “multi-response optimization of wire electrical discharge machining for titanium grade-5 by weighted principal component analysis,” international journal of engineering and technology innovation, vol. 8, no. 2, pp. 133-145, march 2018. [30] c. s. syan and g. ramsoobag, “a differential evolution optimization approach for parameters estimation of truncated and censored failure time data,” advances in technology innovation, vol. 3, no. 4, pp. 185-194, october 2018. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). 261 template encit2010 advances in technology innovation, vol. 5, no. 1, 2020, pp. 10-17 combining cloud computing and artificial intelligence scene recognition in real-time environment image planning walkable area jia-shing sheu * , chen-yin han department of computer science, national taipei university of education, taipei, taiwan received 18 may 2019; received in revised form 11 june 2019; accepted 20 august 2019 doi: https://doi.org/10.46604/aiti.2020.4284 abstract this study developed scene recognition and cloud computing technology for real-time environmental imagebased regional planning using artificial intelligence. tensorflow object detection functions were used for artificial intelligence technology. first, an image from the environment is transmitted to a cloud server for cloud computing, and all objects in the image are marked using a bounding box method. obstacle detection is performed using object detection, and the associated technique algorithm is used to mark walkable areas and relative coordinates. the results of this study provide a machine vision application combined with cloud computing and artificial intelligence scene recognition that can be used to complete walking space activities planned by a cleaning robot or unmanned vehicle through real-time utilization of images from the environment. keywords: tensorflow, computer vision, neural network, scene recognition, cloud computing 1. introduction with the rapid development of technology, high-quality digital cameras have become commonly used for acquiring images. these cameras are used in mobile phones, personal digital assistants, robots, medical systems, and surveillance and home security systems. with machine learning and neural operations, the image in the cloud can be easily identified. this study primarily used cloud resources currently available to upload real-time images from the environment obtained using a digital camera to the cloud and cooperate with cloud machine learning and image recognition for identifying different objects in the real-time image. when objects in a live image can be successfully identified, the environmental conditions observed in the live image can be further distinguished. this study used these resources to map real-time images from the environment to walkable areas by combining images obtained using digital cameras, which can be used with eye-like cameras when combined with cloud machine learning and neural recognition. in the study, section 2 provides a review of the literature and methodology. the structure of the system is discussed in section 3. design and experiments are provided in section 4. finally, conclusions are provided in section 5. 2. literature review and methodology an artificial neural network is a mathematical model based on a neural network system that mimics humans. the neural network aims to establish a neural network model that mimics biological perception in a computer and uses this mathematical statistical model to optimize data learning. in the cognitive system of artificial intelligence, the use of neural network-like algorithms is more advantageous than formal logic inference algorithms. because the neural network has nonfixed, nonconvex, nonlinear, and nonlimiting characteristics, different structures, functions, and learning algorithms can be used [1-3]. neural networks can be classified into various types, such as perceptual neural networks, linear neural * corresponding author. e-mail address: jiashing@tea.ntue.edu.tw tel.: +886-2-27321104 ext. 55425; fax: +886-2-27375457 advances in technology innovation, vol. 5, no. 1, 2020, pp. 10-17 11 networks, inverted-transitive neural networks, and enhanced neural networks. because an inverse-transfer-like neural network can approximate any function with arbitrary precision under appropriate parameter settings, it can effectively predict the problem and provide functions such as environment perception, pattern recognition, and prediction. neural networks have been applied in many fields, including emotion perception, handwriting recognition, image recognition, and language translation [4-7]. a neural network generally has several levels, and each level can provide a sufficient number of neurons. each neuron adds up the input of the previous layer of neurons and enters the conversion of an activation function. each neuron has a distinct connection with the neurons of the subsequent layer, and thus, the output value of the upper layer of neurons is weighted and passed to the neurons of the subsequent layer. the neurons in the same layer are not connected and are distinct [8-9]. fig. 1 neural network-like simulation of how a single neuron operates the neural network is represented by numerous single neurons as shown in fig. 1, and comprises the following components: (1) inputs: receive signals from external or upper neurons. (2) weights: the memory of the neural network. the number of weights corresponds to the number of inputs in the input layer; that is, one weight corresponds to one input, which is multiplied by a conversion function and then added to the other values. (3) transfer function: the transfer function accumulates the product of all input values and weight values and then enters the activation function. (4) threshold: each neuron has a threshold. when the product value is accumulated by the conversion function, the offset value is added. the excitation function is provided to determine whether the signal needs to be output downward. that is, partial migration is used to control the continuation of the value. (5) activation function: an excitation function of the neural network introduces the characteristics of the nonlinear output, and thus, the neural network can perform mapping to learn complex functions (6) outputs/activation: the output values mapped using the excitation function can be used as inputs to the subsequent layer of neurons. advances in technology innovation, vol. 5, no. 1, 2020, pp. 10-17 12 the architecture of the neural network connects the aforementioned single neurons to form a model. some commonly used neural network architectures are inverted neural networks, convolutional neural networks, and recurrent neural networks. 3. system structure the study is primarily based on a neural network-based regional identification system, its focus being on detecting walkable areas. by inputting images to a pre-trained neural network, the objects in the images can be identified. the empty areas between objects indicate walkable areas. fig. 2 presents the architecture of the system obtained from using ief0 [10]. to achieve to mark walkable areas and relative coordinates, this study is divided into three parts: (1) the first part constitutes neural network object identification. when the image enters the neural network, the trained model is used to load the images into the personal computer, identify the object position, and predict the score and object name. (2) the second part involves finding the path. by using the object position and coordinates obtained using the neural network, the distance between the objects is calculated, and the shortest distance between objects is identified to determine the position of the empty area as shown in fig. 3. (3) the third part includes path marking. the two points with the shortest path are connected. the enclosed space inside the line is the walkable area primarily discussed in this paper as shown in fig. 4. fig. 2 the architecture of the whole system fig. 3 the architecture of object detection advances in technology innovation, vol. 5, no. 1, 2020, pp. 10-17 13 fig. 4 the architecture of walkable area marked this experiment used tensorflow as the core machine learning framework and an open source project tensorflow object detection application programming interface [11] as the core of object recognition, where the source code allows the use of trained models that meet their requirements. the pretraining model compiled using the tensorflow detection model zoo was used in this study. the models used for object recognition were faster_rcnn_inception_v2_atrous_coco, ssd_mobilenet_v1_coco, and ssd_mobilenet_v1_fpn_coco. with the bounding box coordinates of the object, the distance between all bounding boxes can be calculated. the algorithm uses the distance between each point to select the shortest path, which is considered a part of the walkable area and is connected into a large block through the plurality of lines. according to an axiom in the euclidean geometry system, “there are two points and only one line,” and each bounding box has four points. thus, for n bounding boxes, the following general formula can be used in eq. (1): 4 2 2 4n n n            (1) in any two bounding boxes, distance d between any two points is measured, which can be recorded as 𝑷𝟏(𝒙𝟏,𝒚𝟏) and 𝑷𝟐(𝒙𝟐, 𝒚𝟐). the distance between all points is calculated using eq. (2). for each length comparison, the two shortest distance between all bounding boxes is selected, and starting and ending point coordinates are recorded. 2 2 2 1 2 1 ( ) ( )d x x y y    (2) the start and end point coordinates are obtained using the aforementioned method, and the line connecting the points can form a closed area, which is the walkable area discussed by the research institute. 4. experimental results table 1 graph representations components specification operating system debian 9 central processing unit amd ryzen 7 2700x random access memory ddr4-3200 16gb graphics processing unit nvidia gtx1080ti a car combined the raspberry pi and cmos camera is used to capture the environmental image as shown in fig. 5. the raspberry pi is a popular embedded system, which is developed by the raspberry pi foundation in the united kingdom. the processor is boardcom bcm2837 soc with a 1.2ghz 64-bit quad-core arm cortex-a53 processor. raspberry pi provides 40-pin general purpose input-output (gpio) connector, that allows user connect peripherals such as lcd1602 advances in technology innovation, vol. 5, no. 1, 2020, pp. 10-17 14 module, sensors, etc. via inter-integrated circuit (i²c), universal asynchronous receiver-transmitter (uart) and serial peripheral interface (spi). raspberry pi also provides the accessories, such as touch panel and camera, the camera which is this experiment used, the camera can get 1024 × 768 images by 5 million pixels, it connects raspberry pi by camera serial interface (csi) with a flexible flat cable. raspberry pi’s power rating is 5v/1.34a (1.5w), we use a mobile power as a power source by microusb, providing maximum 5v/2.4a (12w) output. in this experiment, we use wireless local area network (wlan) and local area network (lan) connect raspberry pi and server. raspbian was chosen for the operating system, it is a debian-based for raspberry pi, officially provided by raspberry pi foundation. raspbian is highly optimized for the raspberry pi line’s low-performance arm cpus, also provide a desktop environment for beginners. in the beginning, raspberry pi boots into the system, then startup streaming service, the protocol is using real time streaming protocol (rtsp). at the server side, the server captures the image from raspberry pi, tensorflow object detection program will detect the object from the image, in this experiment, we use “faster_rcnn_inception_ resnet_v2_atrous_coco” as frozen model to detect object, the reason we chose this model, it can recognize object more than the other two frozen model, but disadvantage of this model is using more time. after detected, the program will generate object coordinate, name and detected score, to validate our theory, we need coordinate data to calculate the walkable area, then, the program will send object coordinate data and source imaged to the walkable area program, name and score data are optional data, but in this experiment, we didn’t use them, those data contain the object name and detected score. the walkable area program was used in python 3 as programming language, in this program, coordinate data from tensorflow object detection is playing an important part in this experiment, first step, we use the coordinate data to mark t he bounding box, second step, using the algorithm to find the walkable area of this image, then, the program will generate the marked image. (a) the raspberry pi (b) cmos camera fig. 5 a car combined the raspberry pi and cmos camera figs. (6-7) present the original input, result detected using tensorflow object detection, and walkable area determined using the program, respectively. the program was developed according to the aforementioned theory. fig. 7 shows the result of the system output, bounding box in green represent the object, the two red lines represent the two shortest distance between two bounding boxes, two lines means we have four coordinates. to show the walkable area, creating an empty layer, same width and height as origin image, using these coordinates to create a polygon on it, filling in blue color in a closed area, after creating the layer, it will overlay on the origin image. the marked area is the walkable area that the system determined. advances in technology innovation, vol. 5, no. 1, 2020, pp. 10-17 15 fig. 6 the environmental image captured by the cmos camera fig. 7 the walkable area determined using the system marked by blue color in fig. 8, this environment is a walking area in the laboratory, in human thinking, we can pass through by the box’s lefthand side, but there was a wheel of the chair in the left, so we pass through between these two objects. the program will detect the objects using fig. 8, then the program used an algorithm to calculate the walkable area between these two objects, shown in fig. 9. you can see that the program marks the walkable area by blue color, so does human thinking. but in fig. 9, some object is not marked, for example, the cabinet in the middle of fig. 9, algorithm skip that object, the blue area is overlapped on it. fig. 8 the environmental image captured by the cmos camera fig. 9 the walkable area determined using the system, marked by blue color fig. 10 the environmental image captured by the cmos camera fig. 11 the walkable area determined using the system, marked by blue color table 2 presents the detail of the detection, we chosen google’s tensorflow object detection api “faster_rcnn_ inception_resnet_v2_atrous_coco” frozen model as object detection model, “ssd_mobilenet_v1_coco,” and “ssd_mobilenet advances in technology innovation, vol. 5, no. 1, 2020, pp. 10-17 16 _v1_ fpn_coco” are not used here, two of them can’t recognize more than two objects, by this reason, the algorithm need s two or more objects input, which is the area that we want to find. table 2 detection details for figs. (7, 9, 11) number of detections detection classes time (sec.) fig. 7 4 toilet, microwave, oven, chair 0.0157 fig. 9 4 refrigerator, chair, bowl 0.0161 fig. 11 2 chair 0.0191 by experiment results, this algorithm needs more object data, accuracy will be increased. choosing a model is a subject in this study, we can train the model base on the existing model, so we should prepare the object that appears indoor frequently, such as sofa, chairs, cabinet, fans, computers, etc. another problem is, vision is limited, the camera cannot capture whole object, only a part of an object, chair foot is an obvious case, sometimes object detection cannot detect what the object is, or it is an object, cause the program cannot find the walkable area. the problem as descript above mentioned, this problem is future work that we need to solve them. 5. conclusions this study proposed the use of recent paths between all objects to determine the walkable area, the experiment showed the result, it proofs the algorithm can find the walkable area by tensorflow object detection. enhanced this algorithm is the most important job in the future, low latency is the goal of this algorithm, the real-time system does not allow high latency. for example, the cleaning robot needs to know which area cannot entry, low latency can speed up the efficiency of the cleaning work. with the popularity of high-bandwidth networks, cloud computing was used to obtain the desired results. with technological developments, the tensor calculation of embedded systems can considerably improve computing efficiency. in the near future, this algorithm can be installed in embedded systems, which will be a significant development for robot vision. conflicts of interest the authors declare no conflict of interest. references [1] s. ren, k. he, r. girshick, and j. sun, “faster r-cnn: towards real-time object detection with region proposal networks,” ieee transactions on pattern analysis and machine intelligence, vol. 39, no. 6, pp. 1137-1149, 2017. [2] t. y. lin, m. maire, s. belongie, l. bourdev, r. girshick, j. hays, p. perona, d. ramanan, c. l. zitnick, and p. dollár, “microsoft coco: common objects in context,” arxiv:1405.0312v3 [cs. cv], february 2015. [3] a. g. howard, m. zhu, b. chen, d. kalenichenko, w. wang, t. weyand, m. andreetto, and h. adam, “mobilenets: efficient convolutional neural networks for mobile vision applications,” arxiv:1704.04861 [cs. cv], 2017. [4] r. girshick, j. donahue, t. darrell, and j. malik, “rich feature hierarchies for accurate object detection and semantic segmentation,” proc. the ieee conference on computer vision and pattern recognition (cvpr), ieee press, june 2014, pp. 580-587. [5] w. liu, d. anguelov, d. erhan, c. szegedy, s. reed, c. y. fu, and a. c. berg, “ssd: single shot multibox detector,” proc. part i: eccv 2016-the 14th european conference, amsterdam, netherlands, october 2016, pp. 21-37. [6] k. he, x. zhang, s. ren, and j. sun, “deep residual learning for image recognition,” proc. 2016 ieee conference on computer vision and pattern recognition (cvpr), ieee press, jun. 2016, pp. 770-778. [7] c. ning, h. zhou, y. song, and j. tang, “inception single shot multibox detector for object detection,” proc. 2017 ieee international conference on multimedia & expo workshops (icmew), ieee press, 2017, pp. 549-554. [8] j. s. sheu and y. l. huang, “implementation of an interactive tv interface via gesture and handwritten numeral recognition,” multimedia tools and applications, vol. 75, no. 16, pp. 9685-9706, 2016. advances in technology innovation, vol. 5, no. 1, 2020, pp. 10-17 17 [9] r. kaluri and p. reddy, “optimized feature extraction for precise sign gesture recognition using self-improved genetic algorithm,” international journal of engineering and technology innovation, vol. 8, no. 1, pp. 25-37, 2018. [10] c. h. chen, c. m. kuo, c. y. chen, and j. h. dai, “retracted: the design and synthesis using hierarchical robotic discrete-event modeling,” journal of vibration and control, vol. 19, no. 11, pp. 1603-1613, 2012. [11] m. abadi, p. barham, j. chen, z. chen, a. davis, j. dean, m. devin, s. ghemawat, g. irving, m. isard, m. kudlur, j. levenberg, r. monga, s. moore, d. g. murray, b. steiner, p. tucker, v. vasudevan, p. warden, m. wicke, y. yu, and x. zheng, “tensorflow: a system for large-scale machine learning,” arxiv:1605.08695 [cs.cv], 2016. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation, vol. 4, no. 4, 2019, pp. 269-275 a biogas desulphurization system with water scrubbing cheng-chang lien * , wei-cheng lin department of biomechatronic engineering, national chiayi university, chiayi, taiwan received 04 april 2019; received in revised form 21 april 2019; accepted 04 june 2019 abstract the biogas is renewable energy with very good potential. it is a gas produced by the bacteria transformation of the organic substance under the anaerobic condition. it is a flammable gas because it contains the hydrogen sulfide (h2s) in its composition. it will cause damage to metal equipment and the corrosion of the pipe. so before using the biogas, the desulphurization process should be carried on. the purpose of this study was to design a water scrubbing system for biogas and carry on the performance test of the desulphurization process in the hog farm. in the field test, the biogas flow rate is changed in the water scrubbing cylinder and under different external circulation water flow rate. the h2s concentration in the biogas is detected before and after water scrubbing, and the desulphurization efficiency of biogas was discussed, furthermore, the ph value of water in the water storage tank was measured. when the h2s adsorbed by water was at steady state, whether the ph value of external circulation water can meet the effluent standard was discussed. from the test results, it showed that the removal efficiency of h2s was 47.7% in 30 minutes of water scrubbing time under 120 ℓ of total water amount, 80 cm of water level in the water scrubbing cylinder, 20 ℓ/min of internal circulation water flow rate and 10 ℓ/min of external circulation water flow rate, and 25 ℓ/min of biogas flow rate. the ph value of water in the water storage tank was 6.22, which met the effluent standard of regulation for discharging directly. this system can reduce the h2s content in the biogas, and purify the quality of the biogas. the operation of this system is convenient and the fabrication cost is low, which is suitable for the small-scale hog farm to purify the biogas. keywords: biogas, water scrubbing, hydrogen sulfide, ph, desulphurization 1. introduction the biogas is considered as renewable energy with very good potential. it is a gas produced by the bacteria transformation of the organic substance under the anaerobic condition. it is a flammable gas, which contains 50-75% of methane (ch4), 25-45% of carbon dioxide (co2), 0-20,000ppm of hydrogen sulfide (h2s), and small amount of gases such as nitrogen (n2), oxygen (o2), hydrogen (h2), and ammonia (nh3) [1]. the production sources of biogas are much extensive, including animal droppings, wastewater, and rubbish landfill, etc. [2]. it can be applied in house fuel, warm-keeping light, generator, electric car, and warm-keeping facilities in the pigsty of hog farm, etc. [3]. the h2s in the biogas is a gas having stink odor and toxicity. it has a very strong corrosive property to some metals. it will cause damage to machinery equipment. therefore, the use of biogas as the fuel will be limited. before using the biogas as the fuel, the h2s must be removed [4]. the h2s in the biogas can be removed through the physical, chemical and biological methods, such as the absorption, adsorption, and bio-reaction. the h2s and co2 in the biogas are the acidic gases. they will be acidified after h2s and co2 are adsorbed in the water [5-7]. the water scrubbing of biogas is a physical purification method. the solubility in water for various gases in the biogas is different. the methane gas is much more difficult to be dissolved in the * corresponding author. e-mail address: lanjc@mail.ncyu.edu.tw tel.: +886-5-2717972; fax: +886-5-2750728 advances in technology innovation, vol. 4, no. 4, 2019, pp. 269-275 270 water. this property is used to remove h2s and co2 by water scrubbing, in order to increase the methane purity in the biogas [8-9]. when selecting the scrubbing column as water scrubbing purification equipment, the desulfurization performance would depend on the scrubber size and design, gas pressure, composition of raw biogas, water flow velocity and water purity. water scrubbing as adsorbent mainly had the low cost and the easy operation, as well as easily increased the contact area and time of gas and liquid to accomplish the continuous purification [10]. the purpose of this study was to design a water scrubbing system for biogas and carry on the performance test of the desulphurization process in the hog farm. the test was carried under different biogas flow rate in the water scrubbing cylinder and external circulation water flow rate, the h2s concentration in the biogas was detected before and after water scrubbing, and the desulphurization efficiency of biogas using water scrubbing was discussed, furthermore the ph value of water in the water storage tank was measured. when the h2s adsorbed by water was at steady state, whether the ph value of external circulation water can meet the effluent standard was discussed. this system can reduce the h2s content in the biogas, and purify the quality of the biogas. 2. materials and method 2.1. design of water scrubbing and desulphurization system fig. 1 illustration of water scrubbing and desulphurization system for biogas the water scrubbing and desulphurization system for biogas is illustrated in fig. 1. the water scrubbing cylinder with 0.003 m thickness made by transparent acrylic is 0.25 m in diameter and 1.20 m in height. it is the main body of water scrubbing and desulphurization system, which is used to observe the internal biogas bubbles and the reaction situation in the water scrubbing cylinder. there are an internal circulation water inlet and an outlet for measurement of h2s in biogas after water scrubbing and desulphurization at the top of the water scrubbing cylinder. there are water outlet and aeration disc at the bottom of water scrubbing cylinder. the aeration disc with 0.23 m in diameter is placed at the bottom of water scrubbing cylinder. there is a biogas inlet under aeration disc. a biogas pump is used to pressurize the biogas. a pressure gauge is used to measure the pressure of biogas flowing into the water scrubbing cylinder. the biogas flow rate is measured by a float type gas flowmeter (f20-100nlpm, lorric ® , taiwan). the biogas enters the water scrubbing cylinder through the bottom of the aeration disc. because there are a lot of holes on the surface of aeration disc, the biogas will enter into the water scrubbing cylinder in a way of many small bubbles, so that the biogas bubbles will contact with water sufficiently for reacting. advances in technology innovation, vol. 4, no. 4, 2019, pp. 269-275 271 the internal circulation water and external circulation water are used for the field test. at the internal water circulation, the water pump is used to pump the circulation water into the water scrubbing cylinder from the top of water scrubbing cylinder, so that the biogas bubbles will contact with water sufficiently for reacting that increase solubility in water, in order to remove the h2s in biogas, and flow back to the water storage tank from the water outlet at the bottom of water scrubbing cylinder. at water external circulation, the external circulation water without adsorbing h2s flows into the water storage tank to upgrade the inner circulation water. the water absorbing h2s is discharged from the water storage tank. the biogas after water scrubbing and desulphurization is discharged from the outlet at the top of the water scrubbing cylinder. the h2s detecting device (120-sm, kitagawa, japan) is used to measure the h2s concentration versus water scrubbing time, in order to calculate the removal h2s efficiency of water scrubbing and desulphurization. at the same time, the ph meter (ts100, suntex, taiwan) is used to measure and record the variation of ph value for the water in the water storage tank during water scrubbing and desulphurization process. 2.2. performance test of water scrubbing and desulphurization system fig. 2 the test flow diagram of water scrubbing and desulphurization system for biogas after the assembly of this water scrubbing and desulphurization system is finished, it is moved to the hog farm to carry on the field test. the biogas used for the test is the swine wastewater treated through the three-step system and produced from the anaerobic fermentation process. before the field test, the total water amount is 120 ℓ in the water storage tank and the acrylic water scrubbing cylinder, the height of water level in the water scrubbing cylinder is 80 cm, and the flow rate of internal circulation water is 20 ℓ/min. the biogas flow rate is changed to 25, 50, and 75 ℓ/min and the external circulation water flow rate is changed to 0 and 10 ℓ/min for carrying on the field test. before the field test, the two-point standard buffer ph 4 and ph 7 were used to calibrate the slope of the electrode for ph meter, and the set time interval was recorded once per minute for ph value. every test time for water scrubbing is 30 minutes. advances in technology innovation, vol. 4, no. 4, 2019, pp. 269-275 272 before starting the field test, the detecting device is used to measure the h2s concentration in biogas without water scrubbing and desulphurization. measure and record the h2s concentration in biogas and the variation of ph value for the water in the water storage tank during water scrubbing and desulphurization process. fig. 2 shows the test flow diagram of water scrubbing and desulphurization system for biogas. the removal efficiency of h2s in the biogas is shown in eq. (1).   ( )% / 100[ ]the removal efficiency a b a  　 (1) where a is the h2s concentration (ppm) before water scrubbing of biogas and b is the h2s concentration (ppm) after water scrubbing of biogas. 3. result and discussion 3.1. variation of removal h2s efficiency fig. 3 the removal of h2s efficiency versus water scrubbing time for different biogas flow rate under 0 ℓ/min of external circulation water flow rate fig. 4 the removal h2s efficiency versus water scrubbing time for 0 and 10 ℓ/min of external circulation water flow rate under 25 ℓ/min of biogas flow rate during the performance test of this system, the total water amount is 120 ℓ, the height of water level in the water scrubbing cylinder is 80 cm, and the flow rate of internal circulation water is 20 ℓ/min. under the external circulation water flow rate is 0 ℓ/min, and the removal of h2s efficiency versus water scrubbing time for the different biogas flow rate is measured and recorded, as shown in fig. 3. from the figure, it is shown that the removal of h2s efficiency is decreased with the increase of water scrubbing time. under 50 and 75 l/min of biogas flow rate, at 12 minutes of water scrubbing time, the removal h2s efficiency is dropped to 4.9% rapidly. after 12 minutes of water scrubbing time, the h2s adsorbed by water has already been closed to the saturated condition, and there is almost no removal h2s efficiency. under 25 l/min of biogas flow rate, at 12 minutes of water scrubbing time, the removal h2s efficiency is dropped to about 24.9%. at 30 minutes of water scrubbing time, the removal of h2s efficiency is dropped to below 7.0%. the water in the water scrubbing cylinder and water storage tank is not discharged by the external circulation water flow. the h2s content in water is increased with the increase of water scrubbing time and reached to the saturated condition finally. under 25 ℓ/min of biogas flow rate, the removal h2s 0 20 40 60 80 100 0 5 10 15 20 25 30 biogas flow rate : 25 ℓ/min biogas flow rate : 50 ℓ/min biogas flow rate : 75 ℓ/min water scrubbing time (min) r e m o v a l h 2 s e ff ic ie n c y ( % ) 0 20 40 60 80 100 0 5 10 15 20 25 30 outside cycle 0 ℓ/min of water flow water scrubbing time (min) r e m o v a l h 2 s e ff ic ie n c y ( % ) advances in technology innovation, vol. 4, no. 4, 2019, pp. 269-275 273 efficiency versus water scrubbing time for 0 and 10 ℓ/min of external circulation water flow rate is shown in fig. 4. when the external circulation water flow rate is 10 ℓ/min, the h2s adsorbed in water will not reach the saturated condition after 12 minutes. the removal of h2s efficiency is kept at a steady value. the external circulation water without adsorbed h2s is added to replace the old water adsorbed h2s can increase the desulphurization effect. fig. 5 illustrates the removal of h2s efficiency versus water scrubbing time for different biogas flow rate under 10 ℓ/min of external circulation water flow rate. 10 minutes before water scrubbing time, the removal of h2s efficiency is decreased with the increase of water scrubbing time under 25, 50, and 75 ℓ/min of biogas flow rate. 10 minutes after water scrubbing time, the variation of removal h2s efficiency is not large. when the water scrubbing time is 30 minutes, the removal h2s efficiency is kept at a constant value respectively under 25, 50, and 75 ℓ/min of biogas flow rate. the removal of h2s efficiency is decreased with the increase of biogas flow rate. from the test results, it is known that when the external circulation water flow rate is increased to 10 ℓ/min, the h2s concentration in water will not reach the saturated condition, and the desulphurization rate will be kept at a certain level steadily. fig. 5 the removal h2s efficiency versus water scrubbing time for different biogas flow rate under 10 ℓ/min of external circulation water flow rate table 1 shows the removal of h2s efficiency versus water scrubbing time for different biogas flow rate under 0 and 10 ℓ/min of external circulation water flow rate and 30 min of water scrubbing time. under 0 ℓ/min of external circulation water flow rate and 30 min of water scrubbing time, the removal h2s efficiency at 25, 50, and 75 ℓ/min of biogas flow rate is 7.0%, 2.1%, and 2.0% respectively, nearly reach the saturated condition. under 10 ℓ/min of external circulation water flow rate and 30 min of water scrubbing time, the removal h2s efficiency at 25, 50, and 75 ℓ/min of biogas flow rate is 47.7%, 33.4%, and 20.7% respectively. it can be seen the removal h2s efficiency in biogas upgrades as the external circulation water flow increases, and steadily remain in a certain one without saturation; this is because the water in the water storage tank is continuously changed so that the removal h2s efficiency can steadily remain after certain water scrubbing time. it is also known that the removal of h2s efficiency is decreased with the increase of biogas flow rate. table 1 the removal h2s efficiency for different biogas flow rate under 0 and 10 ℓ/min of external circulation water flow rate and 30 min of water scrubbing time external circulation water flow (ℓ/min) 10 0 biogas flow (ℓ/min) 25 50 75 25 50 75 trial 1 3 3 3 3 3 3 concentration of h2s (ppm) 5230 7990 9950 11770 12170 12120 the removal efficiency (%) 47.7±0.3 33.4±0.1 20.7±0.9 7.0±0.7 2.1±0.6 2.0±0.1 1 the total water amount is 120 ℓ of, the height of water level in the water scrubbing cylinder is 80 cm and the internal circulation water flow rate is 20 ℓ/min 3.2. variation of ph value in the water storage tank fig. 6 show that the ph value in water storage tank versus water scrubbing time for 0 ℓ/min and 10 ℓ/min external circulation water flow rate under 25 ℓ/min of biogas flow rate. the ph value in the water storage tank decreases with the 0 20 40 60 80 100 0 5 10 15 20 25 30 biogas inflow rate : 25 l/min biogas inflow rate : 50 l/min biogas inflow rate : 75 l/min water scrubbing time (min) r e m o v a l h 2 s e ff ic ie n c y ( % ) advances in technology innovation, vol. 4, no. 4, 2019, pp. 269-275 274 increase of the washing time before 10 minutes of the water scrubbing time. the ph value in the water storage tank tends to be saturated after 10 minutes of the water scrubbing time, it can be seen that the ph value in water storage tank of the external circulating water flow 0 ℓ/min after 10 min of water scrubbing time is lower than the ph value in water storage tank of the external circulating water flow 10 ℓ/min. fig. 7 illustrates the ph value in water storage tank versus water scrubbing time for different biogas flow rate under 10 ℓ/min of external circulation water flow rate. the ph value in water storage tank for the different biogas flow rate is decreased with the increase of water scrubbing time. 10 minutes before water scrubbing time, the ph value in the water storage tank is decreased with the increase of water scrubbing time under 25, 50, and 75 ℓ/min of biogas flow rate. 10 minutes after water scrubbing time, the variation of ph value in water storage tank saturates to a certain value, which is also similar to the variation of removal h2s efficiency versus water scrubbing time. table 2 shows the ph value in the water storage tank versus water scrubbing time under 10 ℓ/min of external circulation water flow rate and 30 min of water scrubbing time. under 30 minutes of water scrubbing time, the ph value in water storage tank for 25, 50, and 75 ℓ/min of biogas flow rate is dropped to 6.22, 6.22, and 6.24 respectively. according to the effluent standard of wastewater specified in government regulation for taiwan, the ph value of the effluent standard is 6.00-9.00. the ph value of field test result does not exceed the effluent standard, and the water can be discharged directly without special wastewater treatment [11]. table 2 the ph value in the water storage tank at different biogas flow rate under 10 ℓ/min of external circulation water flow rate and 30 min of water scrubbing time biogas inflow (l/min) 25 50 75 trial 1 3 3 3 ph in 0 min 7.15 7.19 7.15 ph in 30 min 6.22±0.01 6.22±0.03 6.24±0.01 1 the total water amount is 120 ℓ of, the height of water level in the water scrubbing cylinder is 80 cm and the internal circulation water flow rate is 20 ℓ/min fig. 6 the ph value in water storage tank versus water scrubbing time for 0 ℓ/min and 10 ℓ/min external circulation water flow rate under 25 ℓ/min of biogas flow rate fig. 7 the ph value in water storage tank versus water scrubbing time for different biogas flow rate under 10 ℓ/min of external circulation water flow rate 5.50 5.75 6.00 6.25 6.50 6.75 7.00 7.25 0 5 10 15 20 25 30 outside cycle 0 ℓ/min of water flow water scrubbing time (min) w a te r p h i n w a te r st o ra g e 5.5 5.75 6 6.25 6.5 6.75 7 7.25 0 5 10 15 20 25 30 biogas inflow rate : 25 ℓ/min biogas inflow rate : 50 ℓ/min biogas inflow rate : 75 ℓ/min water scrubbing time (min) w a te r p h i n w a te r st o ra g e advances in technology innovation, vol. 4, no. 4, 2019, pp. 269-275 275 4. result and discussion a water scrubbing and desulphurization system were designed and the performance test of the desulphurization process was carried on in the hog farm. 10 minutes before water scrubbing time, the removal of h2s efficiency is decreased with the increase of water scrubbing time under 25, 50, and 75 ℓ/min of biogas flow rate at 120 ℓ of total water amount, 80 cm of water level, 20 ℓ/min of internal circulation water flow rate and 10 ℓ/min of external circulation water flow rate. 10 minutes after water scrubbing time, the removal h2s efficiency was kept at a constant value under three different biogas flow rates. under 25 ℓ/min of biogas flow rate, the removal h2s efficiency was 47.7% in 30 minutes of water scrubbing time, and the ph value in the water storage tank was 6.22 in 30 minutes of water scrubbing time, which met the effluent standard of regulation and can be discharged directly. as for this water scrubbing and desulphurization system, was often used as a pre-treatment equipment for heating water of biogas combustion, the operation is convenient and the fabrication cost is low, which is suitable for the small-scale hog farm to purify the biogas. conflicts of interest the authors declare no conflict of interest. references [1] m. kaltschmitt and w. streicher, “energie aus biomasse,” regenerative energien in ö sterreich, pp. 339-532, 2009. [2] s. pipatmanomai, s. kaewluan, and t. vitidsant, “economic assessment of biogas-to-electricity generation system with h2s removal by activated carbon in small pig farm,” applied energy, vol. 86, no. 5, pp. 669-674, may 2009. [3] c. c. lien, c. h. ting, and j. h. mei. “piglets comfort with hot water by biogas combustion under controllable ventilation,” advances in technology innovation, vol. 3, no. 2, pp. 86-93, april 2018. [4] j. nägele, h. j. steinbrenner, g. hermanns, v. holstein, n. l. haag, and h. oechsner, “innovative additives for chemical desulphurization in biogas processes: a comparative study on iron compound products,” biochemical engineering journal, vol. 121, pp. 181-187, may 2017. [5] a. l. kohl and r. nielsen, gas purification, 5th ed. houston: gulf publishing company, 1997. [6] g. lastella, c. testa, g. cornacchia, m. notornicola, f. voltasio, and v. k. sharma, “anaerobic digestion of semi-solid organic waste: biogas production and its purification,” energy conversion and management, vol. 43, no. 1, pp. 63-75, january 2002. [7] s. a. marzouk, m. h. al-marzouqi, m. teramoto, n. abdullatif, and z. m. ismail, “simultaneous removal of co2 and h2s from pressurized co2-h2s-ch4 gas mixture using hollow fiber membrane contactors,” separation and purification technology, vol. 86, pp. 88-97, february 2012. [8] s. rasi, j. läntelä, a. veijanen, and j. rintala, “landfill gas upgrading with countercurrent water wash,” waste management, vol. 28, no. 9, pp. 1528-1534, 2008. [9] m. islamiyaha, t. soehartantoa, r. hantoroa, and a. abdurrahman, “water scrubbing for removal of co2 (carbon dioxide) and h2s (hydrogen sulfide) in biogas from manure,” issn 2413-5453, 2: 126-131, 2015. doi: http://dx.doi.org/10.18502/ken.v2i2.367. [10] r. kapoor, p. m. v. subbarao, shah, v. k. vijay, g. shah, s. sahota, and d. singh “factors affecting methane loss from a water scrubbing based biogas upgrading system,” applied energy, vol. 208, pp. 1379-1388, december 2017. [11] water quality protection division, environmental protection administration, executive yuan, taiwan, roc. 2017, “the emission standard for chemical industry stipulated by the environmental protection administration,” article 2, article 5, amended article 2-1, no. 1060101625 order of environmental protection administration, dated on december 25, 2017. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation, vol. 5, no. 2, 2020, pp. 84-97 free vibration frequency of thick fgm circular cylindrical shells with simply homogeneous equation by using tsdt chih-chiang hong * department of mechanical engineering, hsiuping university of science and technology, taichung, taiwan received 04 june 2019; received in revised form 20 july 2019; accepted 17 september 2019 doi: https://doi.org/10.46604/aiti.2020.4380 abstract the objective of this study is to provide the frequency solutions of free vibration in thick fgm circular cylindrical shells by mainly considering both shear correction coefficient and nonlinear coefficient term. this paper investigates the effects of third-order shear deformation theory (tsdt) and the varied shear correction coefficient on the free vibration of thick functionally graded material (fgm), the circular cylindrical shells with simply homogeneous equation under thermal environment. the approach of derivations are given as follows, the varied value of shear correction coefficient is included in the simple homogeneous equation. the nonlinear term of displacement field of tsdt is also included to derive the simply homogeneous equation, some reasonable simplifications in the elements of homogeneous matrix under free vibration of thick fgm circular cylindrical shells are assumed, thus, the natural frequency can be found. three parameters effect on the frequency of thick fgm circular cylindrical shells are computed and investigated, they are nonlinear coefficient c1 term, environment temperature and power law index. there are some main conclusions obtained, generally the natural frequency results are in decreasing value with the mode shape numbers for the thicker circular cylindrical shells. the values of natural frequencies are also affected by the nonlinear coefficient term. keywords: third-order shear deformation theory, tsdt, fgm, shells, vibration, frequency 1. introduction there are some free vibration frequency investigations with shear deformation effect and experimental studies in the functionally graded material (fgm) circular cylindrical shells. in 2019, shahbaztabar et al. [1] presented the free vibration of fgm circular cylindrical shells, including the pasternak foundation and stationary fluid by using the first order shear deformation theory (fsdt). some effects on the result of the natural frequencies are investigated e.g. fluid depth ratio, elastic foundation, volume fraction exponent, geometrical parameters and boundary conditions. in 2019, zippo et al. [2] used the experimental method and the model of finite element method (fem) to study the linear and dynamic behavior of vibrations in the polymeric circular cylindrical shell with fgm equivalent thermal temperature properties. in 2018, baltacıoğlu and civalek [3] used the love’s shell theory and fsdt of the displacements to obtain the numerical results for the circular cylindrical fgm with carbon nanotube reinforced (cntr) panels. for the thick fgm shells, it is necessary to consider the nonlinear terms of displacement theories to obtain more accurate results of analyses, e.g. third-order shear deformation theory (tsdt), higher-order shear deformation theory and triangular function shear deformation theory. in 2018, torabi and ansari [4] presented a formulation of higher-order isoparametric supplement to study the free vibration of fgm shells, considering the structural effects of circular cylindrical, conical, spherical and toroid shells. in 2017, baltacıoğlu and civalek [5] used the extended fem to obtain the frequency of the vibration of cracked fgm shells considering the structural effects of cylindrical * corresponding author. e-mail address: cchong@mail.hust.edu.tw advances in technology innovation, vol. 5, no. 2, 2020, pp. 84-97 85 shell, conical shell and spherical on the vibrations. in 2017, wang and wu [6] presented the free vibration analysis of porous fgm cylindrical shells based on a sinusoidal shear deformation theory (ssdt). in 2014, fazzolari and carrera [7] presented the hierarchical trigonometric ritz formulation (htrf) and used in the free vibration analyses for the doubly curved fgm shells and sandwich shells with fgm core. in 2016, fantuzzi et al. [8] presented free vibration analyses of cylindrical and spherical shells by using the fem and the generalized differential quadrature (gdq) methods. there are some new and improved tsdt used in the investigations of fgms. in 2016, bui et al. [9] presented the numerical results of deflection and frequency by using the tsdt and fem for the static bending behaviors of fgm plates. a similar new tsdt in terms of five un-known variables was used in the eigenvalue equation to calculate the natural frequency. in 2017, do et al. [10] presented the numerical results of deflection and stress by using the tsdt and fem for the static buckling and bending behaviors of fgm plates. the same new tsdt in terms of five un-known variables was used in the bending equation and pre-buckling equation respectively to calculate the numerical solutions without considering the effect of shear correction factors. in 2018, vu et al. [11] presented the numerical results of deflection and frequency by using the tsdt and meshfree method for the static bending, free vibration and buckling behaviors of fgm plates. a similar refined tsdt in terms of four un-known variables was used in the equations to calculate the numerical solutions. fig. 1 two-material thick fgm circular cylindrical shells there are some importance and relevance of studied topics in the dynamics of cylindrical shells composed of fgms. in 2012, zhang et al. [12] presented the nonlinear dynamics of clamped-clamped fgm circular cylindrical shells under an external excitation and uniform temperature change. the similar equations were used and based on the fsdt and von-karman nonlinear strains-displacement relation to obtain the numerical response of displacements. in 2016, dai et al. [13] presented the reviews of coupled mechanics on the fgm cylindrical structures during years 2000-2015. some of the existing mechanical theories and hypotheses were assumed and would be improved in the future for obtaining the more accuracy of the results. in 2008, ansari and darvizeh [14] presented a general analytical approach in arbitrary boundary conditions of fgm circular cylindrical shells. the fsdt of displacements was used to derive the homogeneous linear system and obtained the natural frequency under different boundary conditions. the author has some gdq computational experiences in the composited fgm circular cylindrical shells. in 2017, hong [15] used the approach of fsdt model and the varied shear correction factor to present the numerical gdq results of thermal vibration and flutter of a supersonic air flowed over thick fgm circular cylindrical shells. in 2017, hong [16] used the approach of love’s theory for thin multilayered shells to present the numerical gdq results of displacement and stresses of thin fgm laminated magnetostrictive shells with the value effects of velocity feedback and control gain subjected to thermal vibration. it is interesting to investigate the natural frequency in the tsdt approach of thick fgm circular cylindrical shells under free vibration with simply homogeneous equation and four edges in simply supported boundary conditions. the value effects of three parametric: nonlinear coefficient 1c term, environment temperature and power law index on the natural frequency of thick fgm circular cylindrical shells are investigated. the main contribution and novelty of paper is to provide and investigate the analytic solutions of natural frequencies in the free vibration ,  x,u z ,w r t z r * h l x fgm material 1 fgm material 2 1h 2h advances in technology innovation, vol. 5, no. 2, 2020, pp. 84-97 86 of thick fgm circular cylindrical shells by considering the varied effects of shear correction coefficient and nonlinear terms of tsdt. the motivation of the paper is listed as follows. to obtain the natural frequency values of fgm shells considering the nonlinear coefficient terms of tsdt within the simply homogeneous matrix equation. also by considering the calculated values of shear correction coefficient usually varied with thickness, power law index and environment temperature. 2. formulation for a two-material thick fgm circular cylindrical shells in thermal environment with thickness 1h of fgm material 1 and thickness 2h of fgm material 2. fig. 1 shows a colorful figure that can illustrate the fgm material for the approach of the study. the material properties of power-law function of fgm circular cylindrical shells are considered with a young’s modulus fgme of fgm in the standard variation form of power law index nr , the others are assumed in the simple average form [17]. the properties ip of individual constituent material of fgms are functions of environment temperature t in the following form [18], 1 2 3 0 1 1 2 3( 1 )ip p p t pt p t p t        (1) where 0 1 1 2, , ,p p p p and 3p are the temperature coefficients. the time dependent of nonlinear displacements u , v and w of thick fgm circular cylindrical shells are assumed in the nonlinear coefficient 1c term of tsdt equations [19] as follows, 3 0 1 3 0 1 ( , , ) ( , , ) ( ) ( , , ) ( , , ) ( ) ( , , ) x x w u u x t z x t c z x w v v x t z x t c z r w w x t                          (2) where 0u and 0v are tangential displacements in the in-surface coordinates x and  axes direction, respectively, w is transverse displacement in the out of surface coordinates z axis direction of the middle-plane of circular cylindrical shells, x and  are the shear rotations, r is the middle-surface radius of fgm shell, t is time. coefficient for *2 1 4 / (3 )c h is given as in tsdt approach, in which *h is the total thickness of circular cylindrical shells. the linear time dependent of displacements also can be obtained by letting 1 0c  in the equations. the nonlinear coefficient 1c term of displacement fields of tsdt [19] is used in the thick fgm circular cylindrical shells to investigate the nonlinear value effect on the natural frequency results. for the normal stresses ( x and  ) and the shear stresses ( ,x z   and xz ) in the thick fgm circular cylindrical shells under temperature difference t for the 𝑘 th layer are in the following equations [20-21], 11 12 16 12 22 26 ( ) ( )16 26 66 ( ) 44 45 ( ) ( )45 55 ( ) , , x x x x x xk k k z z xz xzk kk q q q t q q q t tq q q q q q q                                                                                 (3) where x and  are the coefficients of thermal expansion, x is the coefficient of thermal shear, ijq is the stiffness of fgm circular cylindrical shells. x ,  and x are in-plane strains, not negligible z and xz are shear strains. advances in technology innovation, vol. 5, no. 2, 2020, pp. 84-97 87 the dynamic equations of motion with tsdt for a thick fgm circular cylindrical shells are given as follows [22], 2 2 2 0 0 1 1 32 2 2 2 2 2 0 0 1 1 32 2 2 2 2 2 1 2 2 2 2 2 2 2 2 0 1 6 1 32 2 2 2 2 1 + ( ) 1 + ( ) 1 2 1 + ( ) 1 ( ) [ xx x x x x xx x n n u w i j c i x r xt t t n n v w i j c i x r rt t t q q p p p c q x r r xx r w w w i c i c i t t x r                                                                        2 2 0 0 42 2 2 1 0 2 1 42 2 1 0 2 1 42 1 1 ( ) ( )] 1 + ( ) 1 1 + ( ) x xx x xx x u v j x r x rt t m m w q j u k c j x r xt m m w q j v k c j x r rt                                                 (4) where * * * * * * 1 1 2 2 2 2 32 2 , ( , , ) hxx x h x x hxx x h x x hxx x h x x z x xz m m c p q q c p x n n dz n m m zdz m p p z dz p q q                                                                                                                  * * * 2 2 1 ( ) 1 , ( 0,1, 2,..., 6) h h n k k i i k k dz i z dz i           (5) where * n is the total number of layers,  k  is the density of k th ply.   i i 1 i 2j i c i , i 1,4   , 2 2 2 1 4 1 6k i 2c i c i   there are some assumptions for the terms in the strains, e.g. the higher order terms 2( / )w x  , 2[ / ( )]w r   and ( / )[ / ( )]w x w r     can’t be neglected. the von karman type of strain-displacement relations with 0 0/ /v z v r    , 0 0/ /u z u r    and / / / 0xw z z z          are used as eq. (6), by substituting equations (3) and (6) into equation (4), the dynamic equilibrium differential equations with tsdt of thick fgm circular cylindrical shells in terms of partial derivatives of displacements and shear rotations subjected to the expressions terms ( 1 5, ...,f f ) in partial derivatives of thermal loads ( , ,n m p ), mechanical loads ( 1 2, ,p p q ) and inertia terms can be derived and expressed in matrix forms. by assuming that mid-plane strain terms    2 1 / 2 w / x  ,    w / x 1 / r w /     and      2 1 / 2 1 / r w /    are in constant values, the 1 5, ...,f f can be expressed in the derivative terms as eq. (7)-(8), advances in technology innovation, vol. 5, no. 2, 2020, pp. 84-97 88 2 22 30 x x x 1 2 2 22 30 1 2 2 2 zz x u 1 w u w 1 w z c z x 2 x x x x x 2 x 1 v 1 1 w 1 v 1 1 1 w 1 1 w z c z r 2 r r r r r 2 r 0 1 u v w 1 w r x x r                                                                                                          2 2 3 30 x x 0 1 1 20 z 1 20 xz x 1 x 1 u w v 1 w w 1 w z c z z c z r x x x x r x x r v 1 w v 1 w 1 w 3c z z r r r r u w u w 3c z z x r x                                                                                                              　　 w x    (6) 1 1 2 2 2 2 2 3 1 2 2 2 4 1 5 1 1 + 1 + 2 1 ( ) 1 1 + ( ) 1 1 + ( ) xx x x xx x xx x xx x x x n n f p x r n n f p x r p p p f q c r xx r m m p p f c x r x r m m p p f c x r x r                                                                 (7) and * * * * * * 32 11 12 16 2 32 12 22 26 2 32 16 26 66 2 ( , , ) ( ) (1, , ) ( , , ) ( ) (1, , ) ( , , ) ( ) (1, , ) ( , , , , , )s s s s s s s s s s s s h xx xxxx x xh h x xh h x xx x xh i j i j i j i j i j i j n m p q q q t z z dz n m p q q q t z z dz n m p q q q t z z dz a b d e f h                                       * * * * ** * * * * * * * * * * * * 2 3 4 62 2 2 3 4 5 * *2 2 (1, , , , , ) , ( , 1, 2, 6) ( , , , , , ) (1, , , , , ) , ( , 4, 5) s s h s s i jh h i ji j i j i j i j i j i j h q z z z z z dz i j a b d e f h k q z z z z z dz i j        (8) where 1p and 2p are external in-plane distributed forces in x and  direction respectively. q is external pressure load. k is the shear correction coefficient. * h is the total thickness of fgms circular cylindrical shells. the computed and varied values of k are usually functions of total thickness of circular cylindrical shells, fgm power law index and environment temperature [23]. the s s i j q and * * i j q for thick fgm circular cylindrical shells with /z r terms cannot be neglected are used in the following simple forms in 2014 by hong [23], in 2010 by sepiani et al. [24], advances in technology innovation, vol. 5, no. 2, 2020, pp. 84-97 89 2 11 22 2 12 21 44 55 66 16 26 45 / (1 ) ( ) / [(1 / )(1 )] / [2(1 )] / [2(1 / )(1 )] 0 fgm fgm fgm fgm fgm fgm fgm fgm fgm q q e q q e z r q e q q e z r q q q                      (9) where 1 2=( ) / 2fgm   is the poisson’s ratios of the fgm circular cylindrical shells, * * 2 1 1( )[( ) / ] nr fgme e e z h h e    is the young’s modulus of the fgm circular cylindrical shells, nr is the power-law exponent parameter, 1e and 2e are the young’s modulus, 1 and 2 are the poisson’s ratios of the fgm constituent material 1 and 2, respectively. the simpler stiffness forms of s s i j q and * * i j q are used to calculate the stresses, , , , , ,s s s s s s s s s s s s i j i j i j i j i j i j a b d e f h and * * * * * * * * * * * *, , , , , i j i j i j i j i j i j a b d e f h . for example, by using change of variable in integration calculation, the 11a , 11e , 11f , 11h and 44h of thick fgm circular cylindrical shells are given as follows, * 1 2 11 21 2 * 4 2 1 11 21 2 * 5 1 11 2 1 21 2 * 7 11 2 1 21 2 ( ) 1 1 ( ) 2 ( ) ( ) 1 3 3 1 [ ] 4 2( 3) 4( 2) 8( 1) 1 ( ) 2 ( ) 1 2 1 1 1 {( )[ ] } 5 4 3 2( 2) 16( 1) 80 1 ( ) 2 ( ) 1 3 {( )[ 7 1 ( ) 2 n n n n n n n n n n n n r e eh a r h e e e r r r r h e f e e r r r r r h h e e r r                                             1 * 6 2 1 44 1 2 13 2 13 6 4( 5) 4 16( 3) 3 1 ] } 16( 2) 64( 1) 448 ( ) ( ) 1 5 2 1 5 1 [ ] 6 2( 5) 4 3 64( 2) 32( 1) 2(1 ) 2 n n n n n n n n n n n n r r r e r r k h e e h r r r r r r                               (10) 3. vibration frequency the thick fgm circular cylindrical shells with layers in the stacking sequence (0 / 0 )   are used to study the free vibration frequency results with the effects of environment temperature and varied shear correction coefficient calculations, under four sides simply supported boundary condition, no thermal loads ( t 0  ), no in-plane distributed forces ( 1p = 2p =0) and no external pressure load ( q =0) . the free vibration frequency mn with mode shape numbers 𝑚 and 𝑛 for four sides simply supported boundary condition can be derived by simply assuming that 31 1 0i i j   , 0ij ijb e  , 16 26 0a a  , 16 26 0d d  and 45 45 45 0a d f   under the following time sinusoidal displacement and shear rotations forms with amplitudes mna , mnb , mnc , mnd and mne . 0 0 cos( / )sin( / )sin( ) sin( / ) cos( / )sin( ) sin( / )sin( / )sin( ) cos( / )sin( / )sin( ) sin( / ) cos( / )sin( ) mn mn mn mn mn mn x mn mn mn mn u a m x l n r t v b m x l n r t w c m x l n r t d m x l n r t e m x l n r t                       (11) advances in technology innovation, vol. 5, no. 2, 2020, pp. 84-97 90 where m is the number of axial half-waves, n is the number of circumferential waves. by substituting equations (11) into dynamic equilibrium differential equations under free vibration  1 2 5... 0f f f    with the assumed reasonable simplifications of 13 14 15 23 24 25 0fh fh fh fh fh fh      and 3 61 4 0i ij j   in the elements of homogeneous matrix, thus the simply homogeneous equation can be obtained as follows. 11 12 12 22 33 34 35 2 34 44 45 0 2 35 45 55 0 0 0 0 00 0 0 00 0 ,0 0 0 0 0 0 0 mn mnmn mnmn mn mn mn mn mn fh fh afh fh bfh fh fh ck fh fh fh i d ek fh fh fh i                                                          (12) where 2 0 2 2 11 11 66 12 12 66 2 2 22 66 22 2 2 2 4 2 2 2 2 33 55 44 1 11 1 12 1 66 2 4 1 22 1 55 1 55 ( / ) ( / ) ( )( / )( / ) ( / ) ( / ) ( / ) ( / ) ( / ) (2 +4 )( / ) ( / ) ( / ) 3 (2 -3 )( mn mni fh a m l a n r fh a a m l n r fh a m l a n r fh a m l a n r c h m l c h c h m l n r c h n r c d c f                            2 2 1 44 1 44/ ) 3 (2 -3 )( / )m l c d c f n r  (13) and 2 3 2 2 2 34 55 1 11 1 11 1 66 1 66 1 12 1 12 2 1 55 1 55 2 3 2 2 2 35 44 1 22 1 22 1 66 1 66 1 12 1 12 2 1 44 1 44 ( / ) ( )( / ) (2 -3 + )( / )( / ) (6 -9 )( / ) ( / ) ( )( / ) (2 -3 + )( / ) ( / ) (6 -9 )( fh a m l c f c h m l c f c h c f c h m l n r c d c f m l fh a n r c f c h n r c f c h c f c h m l n r c d c f n                  2 3 2 2 2 2 44 11 1 11 1 11 66 1 66 1 66 55 1 55 1 55 2 2 45 12 66 1 12 1 12 1 66 1 66 2 2 2 2 55 66 1 66 1 66 22 1 22 1 22 44 / ) ( -2 + )( / ) ( -2 + )( / ) 6 9 ( + -2 + -2 + )( / )( / ) ( -2 + )( / ) ( -2 + )( / ) 6 r fh d c f c h m l d c f c h n r a c d c f fh d d c f c h c f c h m l n r fh d c f c h m l d c f c h n r a                  2 1 44 1 449c d c f (14) the determinant of the coefficient matrix in equation (12) vanishes for obtaining non-trivial solution of amplitudes can be represented in the simply five degree polynomial equation as follows, thus the mn can be found. 5 4 3 2 (1) (2) (3) (4) (5) (6) 0mn mn mn mn mna a a a a a          (15) where 11 12 11 22 12 12 11 12 11 22 12 12 11 22 11 22 12 12 11 22 11 22 12 12 (1) (2) ( ) (3) [( ) ( ) ] (4) ( ) ( ) (5) [( ) ( ) ] (6) ( ) a sd a fh fh sd sc a fh fh fh fh sd fh fh sc sb a fh fh fh fh sc fh fh sb sa a fh fh fh fh sb fh fh sa a fh fh fh fh sa                        (16) advances in technology innovation, vol. 5, no. 2, 2020, pp. 84-97 91 in which 2 2 0 33 44 2 0 33 55 44 55 33 44 35 35 34 34 2 0 45 45 33 44 55 44 34 35 35 34 45 35 35 44 34 34 55 45 45 33 ( / ) / ( ) / sd k i sc fh sd fh k i sb fh fh fh fh fh fh fh fh fh fh k i fh fh sa fh fh fh fh fh fh fh fh fh fh fh fh fh fh fh fh fh fh                (17) 4. results and discussion table 1 *f for sus304/si3n4 * /l h nr 1 c (1/ mm 2 ) * f present solution, * 1.2h  mm, varied k 1t k 100t k 300t k 600t k 1000t k 5 0.5 0.925925 0 3.178485 8.567362 3.434401 9.391482 3.873990 10.984527 4.111627 12.617598 3.662770 13.538765 1 0.925925 0 3.318237 7.958229 3.570262 8.647485 4.005848 9.985663 4.253132 11.39358 3.862303 12.597763 2 0.925925 0 3.477849 7.132526 3.723252 7.690617 4.151720 8.763615 4.409905 9.831241 4.098108 10.888006 10 0.925925 0 3.756482 6.452526 3.984509 6.885567 4.394140 7.695008 4.670961 8.305713 4.532945 8.563466 8 0.5 0.925925 0 2.238668 6.204801 2.419486 6.788758 2.730914 7.907682 2.897884 9.025515 2.576800 9.594029 1 0.925925 0 2.336488 5.856612 2.514541 6.350593 2.823131 7.296222 2.996852 8.251528 2.716518 8.981841 2 0.925925 0 2.448283 5.856612 2.621661 6.350593 2.925246 6.587167 3.106589 7.313977 2.881696 7.905719 10 0.925925 0 2.643672 5.107512 2.804862 5.428395 3.095301 6.021389 3.289678 6.482052 3.186499 6.638852 10 0.5 0.925925 0 1.926014 5.478031 2.081703 5.986835 2.349735 6.954710 2.830540 7.897100 2.217255 8.325227 1 0.925925 0 2.010152 5.224647 2.163467 5.659943 2.429062 6.485115 2.578504 7.288469 2.337450 7.833433 2 0.925925 0 2.106326 4.906995 2.255628 5.272914 2.663339 5.956491 2.672919 6.573925 2.479556 6.987514 10 0.925925 0 2.274471 4.730258 2.413313 5.024359 2.663339 5.565807 2.830540 5.980343 2.741827 6.078461 the composited thick fgm sus304/si3n4 material is used to implement the numerical computation of vibration under environment temperature t (free stress assumed). the fgm material 1 at inner position of circular cylindrical shells is sus304 (stainless steel), the fgm material 2 at outer position of circular cylindrical shells is si3n4 (silicon nitride) used for the free vibration frequency computations with simply homogeneous equation. for the preliminary fgm circular cylindrical shells study, it did not considered the effect of nonlinear coefficient term on the calculation of varied shear correction coefficient. the varied values of k are usually functions of * h , nr and t in the thick fgm circular cylindrical shells ( 0ijb  ). for l/r=1, 1 2h h , * 1.2h  mm, calculated values of k are increasing with nr (from 0.1 to 10). thus values of k are used for frequency calculations of the free vibration (no thermal loads under no temperature difference ( t =0) including the effects of nonlinear coefficient 1 c term. firstly, for the frequency parameter * 11 2 114 /f r i a values under the effects of 1c = 0.925925/mm 2 and 1 c = 0/mm2 for */l h = 5, 8 and 10 are shown in table 1, where 11 is the fundamental first natural frequency (m = n = 1). for sus304/si3n4 thick circular cylindrical shells under free vibration with * 1.2h  mm, the *f values under 1t k , 100k, 300k, 600k and 1000k with varied k and 1c effects are in the values not greater than 13.538765. advances in technology innovation, vol. 5, no. 2, 2020, pp. 84-97 92 the frequency parameter 2 *11 1 1( / ) /l h e   values under the effects of 1c = 0.925925/mm 2 and 1 c = 0/mm2 for */l h = 5, 8 and 10 are shown in table 2, 1 is the density of fgm material 1, for sus304/si3n4 thick circular cylindrical shells under free vibration with * 1.2h  mm, the ω values under 1t k , 100k, 300k, 600k and 1000k with varied k and 1c effects are in the values not greater than 32.380783. it is easy to judge the machining abnormality by the external sensor signal in this experiment. at this time, the internal condition data could be combined to determine the type of abnormality as chatter. the accuracy of the monitoring is improved. there are a variety of visualization methods in this monitoring system, which could help to quickly locate abnormal positions. in addition, it helped to grasp the information of cnc machine tool conditions at all times, which reduced the difficulty of subsequent fault analysis and diagnosis. table 2  for sus304/si3n4 * /l h nr 1 c (1/ mm 2 )  present solution, * 1.2h  mm , varied k 1t k 100t k 300t k 600t k 1000t k 5 0.5 0.925925 0 5.781036 15.582337 6.101545 16.684873 6.693336 18.978658 7.132169 21.886919 7.123131 26.329359 1 0.925925 0 5.782522 13.868399 6.103089 14.782213 6.694982 16.689058 7.133930 19.110862 7.124928 23.239543 2 0.925925 0 5.783707 11.861481 6.104272 12.608768 6.696166 14.134532 7.135265 15.907033 7.126575 18.934150 10 0.925925 0 5.784393 9.935876 6.104699 10.549432 6.696107 11.726208 7.135397 12.687871 7.128233 13.466383 * /l h nr 1 c (1/ mm 2 )  present solution, * 1.2h  mm , varied k 1t k 100t k 300t k 600t k 1000t k 8 0.5 0.925925 0 6.514712 18.056488 6.877522 19.297414 7.549395 21.860160 8.042832 25.049551 8.017926 29.852609 1 0.925925 0 6.514685 16.329633 6.877464 17.369365 7.549288 19.510704 8.042762 22.144931 8.018013 26.510597 2 0.925925 0 6.514442 16.329633 6.877144 17.369365 7.548845 16.998745 8.042382 18.934526 8.017992 21.996770 10 0.925925 0 6.513346 12.583626 6.875763 13.307022 7.546949 14.681323 8.040542 15.843253 8.017431 16.703767 10 0.5 0.925925 0 7.006078 19.926910 7.396695 21.272380 8.119573 24.032178 8.647910 27.397165 8.623964 32.380783 1 0.925925 0 7.005980 18.209449 7.396562 19.350477 8.119391 21.677169 8.650032 24.450416 8.623955 28.901228 2 0.925925 0 7.005695 16.320789 7.396207 17.289890 8.117177 19.214038 8.649613 21.273336 8.623854 24.302459 10 0.925925 0 7.004657 14.567711 7.394913 15.395722 8.117177 16.963157 8.647910 18.271238 8.623262 19.117235 it is interesting to compare the present vibration values of frequency with some authors' work as shown in the tables (3)-(4). the values of *f vs. * h for sus304/si3n4 under * /l h =10 and 300t k with varied k and 1c effects are shown in table 3. the compared value *f = 8.426538 at * h = 2mm, nr = 0.5 is greater than * f = 8.0 at n= 13 with silicon nitride-nickel under classical shell theory (cst), no external pressure ( 0ek  ) by sepiani et al. in 2010 [24]. the values of  vs. * h for sus304/si3n4 under * /l h =10 and t=700k with varied k and 1c effects are shown in table 4. the compared value  = 2.459972 at 1 c = 0.925925/mm2, * 1.2h  mm, nr = 0.5 is greater than  = 1.71137 with the material variation type a, three layers thickness ratio 1-8-1, the l directional radius of curvature is ∞, * /l h =10, nr = 0.5 for the fgm sandwich shell presented by chen et al. in 2017 [25]. advances in technology innovation, vol. 5, no. 2, 2020, pp. 84-97 93 table 3 comparison of frequency *f for sus304/si3n4 and silicon nitride-nickel 1c (1/mm 2 ) *h (mm) * f present method , * 1/ = 0l h , 300t k , varied k , for 3 4sus304 / si n sepiani et al. 2010, for silicon nitride-nickel, n 13 0.5nr  1nr  2nr  0.925925 1.2 2.349735 2.429062 2.516924 0.333333 2 8.426538 8.711254 9.026930 8.0 0.000033 200 842669.2 871142.9 902712.1 0.000014 300 253980.0 262560.1 272073.1 0.000003 600 18903.19 19542.01 20250.14 0.000001 900 43930.59 45414.97 47060.53 table 4 comparison of frequency  for sus304/si3n4 1c (1/mm2) * h (mm)  present method, * 0/ 1l h  , 700t k , varied k chen et al. 2017, type a, 1-8-1, 0.5nr  0.5nr  1nr  2nr  0.925925 1.2 2.459972 2.550420 2.651571 1.71137 0.333333 2 8.821661 9.146189 9.509387 0.000033 200 882174.8 914629.0 950949.6 0.000014 300 267903.0 277757.0 288785.0 0.000003 600 19897.76 20629.82 21448.99 0.000001 900 46307.33 48010.84 49917.17 table 5 fundamental natural frequency 11 for * 1.2 h mm * /l h nr 1c (1/ mm 2 ) 11 1t k 100t k 300t k 600t k 1000t k 5 0.5 0.925925 0 0.001620 0.004366 0.001730 0.004731 0.001906 0.005406 0.001947 0.005975 0.001614 0.005968 1 0.925925 0 0.001620 0.003886 0.001730 0.004191 0.001907 0.004753 0.001947 0.005217 0.001615 0.005267 2 0.925925 0 0.001620 0.003324 0.001730 0.003575 0.001907 0.004026 0.001948 0.004343 0.001615 0.004291 10 0.925925 0 0.001620 0.002784 0.001731 0.002991 0.001907 0.003340 0.001948 0.003464 0.001615 0.003052 8 0.5 0.925925 0 0.000713 0.001976 0.000761 0.002137 0.000840 0.002432 0.000857 0.002671 0.000709 0.002643 1 0.925925 0 0.000713 0.001787 0.000761 0.001924 0.000840 0.002170 0.000857 0.002361 0.000709 0.002347 2 0.925925 0 0.000713 0.001787 0.000761 0.001924 0.000839 0.001891 0.000857 0.002019 0.000709 0.001947 10 0.925925 0 0.000712 0.001377 0.000761 0.001474 0.000839 0.001633 0.000857 0.001689 0.000709 0.001479 10 0.5 0.925925 0 0.000490 0.001396 0.000524 0.001508 0.000578 0.001711 0.000590 0.001870 0.000488 0.001835 1 0.925925 0 0.000490 0.001275 0.000524 0.001371 0.000578 0.001543 0.000590 0.001668 0.000488 0.001637 2 0.925925 0 0.000490 0.001143 0.000524 0.001225 0.000578 0.001368 0.000590 0.001452 0.000488 0.001377 10 0.925925 0 0.000490 0.001020 0.000524 0.001091 0.000578 0.001208 0.000590 0.001247 0.000488 0.001083 secondly, the natural frequency mn values (unit 1/s) of free vibration ( t 0  ) according to mode shape numbers m and n for the sus304/si3n4 fgm thick circular cylindrical shells are calculated. for the values of fundamental first (m=n=1) natural frequency 11 vs. nr with * 1.2h  mm, varied k and the effects of 1c = 0.925925/mm 2 and 1 c = 0/mm2 for */l h = 5, 8 and 10 are under 1t k , 100k, 300k, 600k and 1000k are shown in table 5. for the values of natural frequency mn vs. m,n=1,2,…,9 with nr = 0.5, 300t k , * 1.2h  mm under varied k and the effects of 1c = 0.925925/mm 2 and 1 c = 0/mm2 for * /l h = 5 and 10 are shown in table 6. advances in technology innovation, vol. 5, no. 2, 2020, pp. 84-97 94 table 6 natural frequency mn vs. m and n under varied k , 1c , 0.5nr  and 300t k 1c (1/ mm 2 ) * /l h 1n 1n 2n 3n 4n 5n 6n 7n 8n 9n 0.925925 5 10 0.001906 0.000578 0.001347 0.000409 0.001098 0.000334 0.000949 0.000289 0.000847 0.000258 0.000772 0.000236 0.000713 0.000218 0.000666 0.000204 0.000627 0.000193 0 5 10 0.005406 0.001711 0.005285 0.001621 0.005244 0.001602 0.005209 0.001595 0.005170 0.001592 0.005126 0.001590 0.005077 0.001588 0.005022 0.001587 0.004963 0.001586 1c (1/ mm 2 ) * /l h 2n 1n 2n 3n 4n 5n 6n 7n 8n 9n 0.925925 5 10 0.001347 0.000409 0.000953 0.000289 0.000778 0.000236 0.000673 0.000204 0.000602 0.000183 0.000549 0.000167 0.000508 0.000154 0.000474 0.000144 0.000447 0.000136 0 5 10 0.002712 0.000861 0.002647 0.000811 0.002633 0.000801 0.002626 0.000797 0.002620 0.000796 0.002613 0.000795 0.002606 0.000794 0.002606 0.000794 0.002590 0.000794 1c (1/ mm 2 ) * /l h 3n 1n 2n 3n 4n 5n 6n 7n 8n 9n 0.925925 5 10 0.001098 0.000334 0.000778 0.000236 0.000635 0.000192 0.000550 0.000167 0.000492 0.000149 0.000449 0.000136 0.000415 0.000126 0.000388 0.000118 0.000366 0.000111 0 5 10 0.001817 0.000578 0.001766 0.000541 0.001757 0.000534 0.001753 0.000532 0.001751 0.000531 0.001748 0.000530 0.001746 0.000530 0.001744 0.000529 0.001741 0.000529 1c (1/ mm 2 ) * /l h 4n 1n 2n 3n 4n 5n 6n 7n 8n 9n 0.925925 5 10 0.000949 0.000289 0.000673 0.000204 0.000550 0.000167 0.000476 0.000144 0.000426 0.000129 0.000389 0.000118 0.000360 0.000109 0.000337 0.000102 0.000317 0.000096 0 5 10 0.001373 0.000438 0.001326 0.000407 0.001318 0.000401 0.001316 0.000399 0.001314 0.000398 0.001313 0.000397 0.001312 0.000397 0.001311 0.000397 0.001309 0.000397 1c (1/ mm 2 ) * /l h 5n 1n 2n 3n 4n 5n 6n 7n 8n 9n 0.925925 5 10 0.000847 0.000258 0.000602 0.000183 0.000492 0.000149 0.000426 0.000129 0.000381 0.000115 0.000348 0.000105 0.000322 0.000098 0.000301 0.000091 0.000284 0.000087 0 5 10 0.001111 0.000355 0.001062 0.000326 0.001055 0.000321 0.001053 0.000319 0.001052 0.000318 0.001051 0.000318 0.001050 0.000318 0.001050 0.000317 0.001049 0.000317 1c (1/ mm 2 ) * /l h 6n  1n 2n 3n 4n 5n 6n 7n 8n 9n 0.925925 5 10 0.000772 0.000215 0.000549 0.000167 0.000449 0.000136 0.000389 0.000118 0.000348 0.000105 0.000317 0.000096 0.000294 0.000089 0.000275 0.000083 0.000259 0.000079 0 5 10 0.000941 0.000300 0.000886 0.000272 0.000880 0.000267 0.000878 0.000266 0.000877 0.000265 0.000876 0.000265 0.000876 0.000265 0.000875 0.000265 0.000875 0.000265 1c (1/ mm 2 ) * /l h 7 n  1n 2n 3n 4n 5n 6n 7n 8n 9n 0.925925 5 10 0.000713 0.000218 0.000508 0.000154 0.000415 0.000126 0.000360 0.000109 0.000322 0.000098 0.000294 0.000089 0.000272 0.000082 0.000254 0.000846 0.000240 0.000073 0 5 10 0.000826 0.000263 0.000761 0.000234 0.000754 0.000229 0.000752 0.000228 0.000751 0.000227 0.000751 0.000227 0.000751 0.000227 0.000750 0.000227 0.000750 0.000227 1c (1/ mm 2 ) * /l h 8n  1n 2n 3n 4n 5n 6n 7n 8n 9n 0.925925 5 10 0.000666 0.000204 0.000474 0.000144 0.000388 0.000118 0.000337 0.000102 0.000301 0.000091 0.000275 0.000083 0.000254 0.000078 0.000238 0.000072 0.000224 0.000069 0 5 10 0.000744 0.000236 0.000668 0.000205 0.000661 0.000201 0.000658 0.000199 0.000658 0.000199 0.000657 0.000199 0.000657 0.000198 0.000657 0.000198 0.000656 0.000198 1c (1/ mm 2 ) * /l h 9n  1n 2n 3n 4n 5n 6n 7n 8n 9n 0.925925 5 10 0.000627 0.000193 0.000447 0.000136 0.000366 0.000111 0.000317 0.000096 0.000284 0.000087 0.000259 0.000079 0.000240 0.000073 0.000224 0.000069 0.000212 0.000064 0 5 10 0.000684 0.000215 0.000596 0.000183 0.000588 0.000179 0.000586 0.000177 0.000585 0.000177 0.000584 0.000176 0.000584 0.000176 0.000584 0.000176 0.000584 0.000176 advances in technology innovation, vol. 5, no. 2, 2020, pp. 84-97 95 fig. 2 1n vs. nr for * =/ 5l h fig. 3 1n vs. nr for * 1/ = 0l h fig. 4 1n vs. t for * =/ 5l h fig. 5 1n vs. nr for * 1/ = 0l h finally, the natural frequency mn values (unit 1/s) vs. nr and t of free vibration ( t =0) according to mode shape numbers m=1 and n (from 1 to 9) for the sus304/si3n4 fgm thick circular cylindrical shells are calculated. figs. (2)-(3) show the values of 1n vs. nr in fgm circular cylindrical shells for thick * /l h = 5, 10 respectively, with the effects of varied k and 1 c = 0.925925/mm2 under 300t k . generally the values of 1n are decreasing with values of n (from 1 to 9) for * /l h = 5, nr = 0.5, 1 and 10. the greatest value of 11 = 0.00191 (unit 1/s) is found for * /l h = 5. the values of 1n are also decreasing with values of n (from 1 to 9) for * /l h = 10, nr = 0.5, 1 and 10. figs. 4-5 show the values of 1n vs. t in fgm circular cylindrical shells for thick * /l h =5, 10 respectively, under the effects of varied k , 1c = 0.925925/mm 2 and nr = 0.5. generally the values of 1n are decreasing with values of n (from 1 to 9) for * /l h = 5, 300t k , 600k and 1000k, the values of 1n are almost in the same for 300t k and 600k, but in greater values than that in the 1000t k . the greatest value of 11 = 0.00191 (unit 1/s) is found for * /l h = 5, 600t k . the values of 1n can stand for higher temperature 1000t k at * /l h = 5. the values of 1n are decreasing with values of n (from 1 to 9) for * /l h = 10, 300t k , 600k and 1000k , the values of 1n are almost in the same for 300t k and 600k, but in greater values than that in the 1000t k . the greatest value of 11 = 0.00059 (unit 1/s) is found for * /l h = 10, 600t k . the values of 1n can stand for higher temperature 1000t k at * /l h = 10. the values of 1n at * /l h = 5 are also found in the greater values than that at * /l h = 10. 5. conclusions the values of natural frequency and frequency parameters are calculated and obtained by using the simply homogeneous equation with the polynomial equation in fifth-order of mn  in the free vibration of thick fgm circular cylindrical shells. the advances in technology innovation, vol. 5, no. 2, 2020, pp. 84-97 96 three items of value effects are considered in nonlinear coefficient term 1 c , shear correction coefficient and environment temperature. some of the important results are found as follows. (a) data investigated in the three kinds of frequency parameters under free vibration with and without the effects of 1 c . (b) generally the values of 1n are decreasing with values of n (from 1 to 9) for * /l h =5 and 10, nr = 0.5, 1 and 10. (c) the values of 1n can stand for higher environment temperature 1000t k at * /l h = 10. (d) the values of 1n vs. environment temperature t at * /l h =5 are found in the greater values than that at * /l h = 10. conflicts of interest the authors declare no conflict of interest. references [1] a. shahbaztabar, a. izadi, m. sadeghian, and m. kazemi, “free vibration analysis of fgm circular cylindrical shells resting on the pasternak foundation and partially in contact with stationary fluid,” applied acoustics, vol. 153, pp. 87-101, 2019. [2] a. zippo, m. barbieri, and f. pellicano, “temperature gradient effect on dynamic properties of a polymeric circular cylindrical shell,” composite structures, vol. 216, pp. 301-314, 2019. [3] a. k. baltacıoğlu and ö. civalek, “vibration analysis of circular cylindrical panels with cnt reinforced and fgm composites,” composite structures, vol. 202, pp. 374-388, 2018. [4] j. torabi and r. ansari, “a higher-order isoparametric superelement for free vibration analysis of functionally graded shells of revolution,” thin-walled structures, vol. 133, pp. 169-179, 2018. [5] a. nasirmanesh and s. mohammadi, “an extended finite element framework for vibration analysis of cracked fgm shells,” composite structures, vol. 180, pp. 298-315, 2017 [6] y. wang and d.wu, “free vibration of functionally graded porous cylindrical shell using a sinusoidal shear deformation theory,” aerospace science and technology, vol. 66, pp. 83-91, 2017. [7] f. a. fazzolari and e. carrera, “refined hierarchical kinematics quasi-3d ritz models for free vibration analysis of doubly curved fgm shells and sandwich shells with fgm core,” journal of sound and vibration, vol. 333, pp. 1485-1508, 2014. [8] n. fantuzzi, s. brischetto, f. tornabene, and e. viola, “2d and 3d shell models for the free vibration investigation of functionally graded cylindrical and spherical panels,” composite structures, vol. 154, pp. 573-590, 2016. [9] t. q. bui, t. v. do, l. h. t. ton, d. h. doan, s. tanaka, d. t. pham et al., “on the high temperature mechanical behaviors analysis of heated functionally graded plates using fem and a new third-order shear deformation plate theory,” composites part b, vol. 92, pp. 218-241, 2016. [10] t. v. do, d. k. nguyen, n. d. duc, d. h. doan, and t. q. bui, “analysis of bi-directional functionally graded plates by fem and a new third-order shear deformation plate theory,” thin-walled structures, vol. 119, pp. 687-699, 2017. [11] t. v. vu , a. khosravifard, m. r. hematiyan, and t. q. bui, “a new refined simple tsdt-based effective meshfree method for analysis of through-thickness fg plates,” applied mathematical modelling, vol. 57, pp. 514-534, 2018. [12] w. zhang, y. x. hao, and j. yang, “nonlinear dynamics of fgm circular cylindrical shell with clamped-clamped edges,” composite structures, vol. 94, pp. 1075-1086, 2012. [13] h. l. dai, y. n. rao, and ting dai, “a review of recent researches on fgm cylindrical structures under coupled physical interactions, 2000-2015,” composite structures, vol. 152, pp. 199-225, 2016. [14] r. ansari and m. darvizeh, “prediction of dynamic behaviour of fgm shells under arbitrary boundary conditions,” composite structures, vol. 85, pp. 284-292, 2008. [15] c. c. hong, “effects of varied shear correction on the thermal vibration of functionally-graded material shells in an unsteady supersonic flow,” aerospace, vol. 4, pp. 1-15, 2017. [16] c. c. hong, “the gdq method of thermal vibration laminated shell with actuating magnetostrictive layers,” international journal of engineering and technology innovation, vol. 7, pp. 188-200, 2017. [17] c. c. hong, “thermal vibration of magnetostrictive functionally graded material shells,” european journal of mechanics a/solids, vol. 40, pp. 114-122, 2013. advances in technology innovation, vol. 5, no. 2, 2020, pp. 84-97 97 [18] c. c. hong, “rapid heating induced vibration of magnetostrictive functionally graded material plates,” transactions of the asme, journal of vibration and acoustics, vol. 134, 021019, pp. 1-11, 2012. [19] s. j. lee, j. n. reddy, and f. rostam-abadi, “transient analysis of laminated composite plates with embedded smart-material layers,” finite elements in analysis and design, vol. 40, pp. 463-483, 2004. [20] s. j. lee and j. n. reddy, “non-linear response of laminated composite plates under thermomechanical loading,” international journal of non-linear mechanics, vol. 40, pp. 971-985, 2005. [21] j. m. whitney, “structural analysis of laminated anisotropic plates,” lancaster: pennsylvania, usa, technomic publishing company, inc., 1987. [22] j. n. reddy, “energy principles and variational methods in applied mechanics,” wiley, new york, 2002. [23] c. c. hong, “thermal vibration of magnetostrictive functionally graded material shells by considering the varied effects of shear correction coefficient,” international journal of mechanical sciences, vol. 85, pp. 20-29, 2014. [24] h. a. sepiani, a. rastgoo, f. ebrahimi, and a. g. arani, “vibration and buckling analysis of two-layered functionally graded cylindrical shell considering the effects of transverse shear and rotary inertia,” materials and design, vol. 31, pp. 1063-1069, 2010. [25] h. chen, a. wang, y. hao, and w. zhang, “free vibration of fgm sandwich doubly-curved shallow shell based on a new shear deformation theory with stretching effects,” composite structures, vol.179, pp. 50-60, 2017. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation , vol. 2, no. 3, 2017, pp. 61 67 61 advanced manufacture of spiral bevel and hypoid gears vilmos simon department of machine and product design , budapest university of technology and economics, hungary. received 20 january 2016; received in revised form 27 may 2016; accept ed 02 june 2016 abstract in this study, an advanced method for the manufacture of spiral bevel and hypoid gears on cnc hypoid generators is proposed. the optmal head-cutter geometry and machine tool settings are determined to introduce the optimal tooth surface modifications into the teeth of spiral bevel and hypoid gears. the aim of these tooth surface mo difications is to simultaneously reduce the tooth contact pressure and the transmission errors, to ma ximize the ehd load carrying capacity of the oil film, and to minimize power losses in the oil film. the proposed advanced method for the manufacture of spiral bevel and hypoid gears is based on machine tool setting variation on the cradle-type generator conducted by optimal polynomial functions and on the use of a cnc hypoid generator. an algorithm is developed for the execution of motions on the cnc hypoid generator using the optimal relations on the cradle-type machine. effectiveness of the method was demonstrated by using spiral bevel and hypoid gear examples. significant improvements in the operating characteristics of the gear pairs are achieved. ke ywor ds : manufacture, spira l bevel and hypoid gears, load distribution, ehd lubrication, cnc generator 1. introduction the new cnc hypoid generators have made it possible to perform nonlinear correct ion motions for the cutting of the face-milled and face-hobbed spiral bevel and hypoid gears. several studies investigated freeform cutting methods using such machines. among them, shih and fong [1] proposed a flank-correct ion methodology derived direct ly fro m the six-a xis cartesian-type cnc hypoid generator. a polynomia l representation of the universal mot ions of machine tool settings on cnc machines was proposed by fan in re f. [2]. chen and wasif [3] presented a new mathe matical model to calculate the cutter system location and orientation and a generic post-processing method to establish the machine kine matic chain and to co mpute the coordinates of the mach ine a xes for the face-milling process on cnc machines. zhang et al. [4] derived the re lative mot ion relat ion among the virtual c radle, generating gear, cutter and workpiece on the cnc hypoid generator. to achieve maximu m life in a gear set, appropriate bearing pattern location with low tooth contact pressure and low loaded transmiss ion error must coexists. the maximum tooth contact pressure and t rans mission e rro r depend substantially on tooth geometry. in order to reduce the tooth contact pressure and the transmission errors, and to decrease the sensitivity of the gear pair to errors in tooth surfaces and to the relative positions of the mating members, carefully chosen tooth surface modifications are usually applied to the teeth of one or both mating gears. as a result of these modifications, a point contact replaces the theoretical line contact of the fully conjugated tooth surfaces. these modifications are introduced into the gear tooth surfaces by applying the appropriate machine tool setting for the man ufacture of the pinion and the gear and/or by using a head-cutter with optimized geometry. the new cnc hypoid generators have made it possible to perform vary ing correction motions during the cutting of face-milled and face-hobbed spiral bevel and hypoid gears. in this paper, a method is presented to determine optimal head -cutter geometry and optima l polyno mial functions for the conduction of mach ine tool setting variat ion in pinion teeth finishing simu ltaneously redu cing ma ximu m tooth contact pressure and transmission errors , ma ximizing ehd load carry ing capacity of the oil film, and min imizing the power losses in the oil film. the developed optimization procedure relies heavily on the loaded tooth contact analysis for the predict ion of ma ximu m tooth contact pressure and trans* corresponding aut hor, email: simon.vilmos@gt 3.bme.hu advances in technology innovation , vol. 2, no. 3, 2017, pp. 61 67 62 copyright © taeti mission errors and on the e lastohydrodynamic lubrication analysis for the calculat ion of ehd load carrying capacity of the oil film, and power losses in the oil film. the load distribution and transmission error ca lculation method e mployed in this study was developed by the author of this paper [5, 6]. the ehd lubrication calculat ions are based on the method presented in refs. [7, 8] the optimization is based on machine tool setting variation on the cradle-type generator conducted by optima l polynomial functions and on optimal head-cutter geometry. in the second step an algorith m is developed for the e xecution of motions on the cnc hypoid generator using the re lat ions on the c rad le -type mach ine. effectiveness of the method was demonstrated by using spiral bevel and hypoid gear e xa mp les. significant reductions in the ma ximu m tooth contact pressure and transmission errors , and improve ments in lubricat ion performances were obtained. 2. manufacture of spiral bevel and hypoid gears on cradle-type generator the concept of an imag inary generating crown gear is used in the gen erating cutting process of the face -hobbed spiral bevel and hypoid pinion and gear teeth (fig. 1). the machine tool settings are: the t ilt angle of the cutter spindle with respect to the cradle rotation axis (κ), the swivel angle of cutter tilt (μ), the radia l mach ine tool setting (e), and the tilt d istance fro m t ilt centre to re ference plane o f head-cutter (hd). to obtain the tooth surface in the generating process, the work gears are rolled with the i maginary generating gear. the coordinate systems   eeee z,y,xk and   iiii z,y,xk are attached to the head-cutter and to the pinion/gear, respectively. the teeth-surfaces of the pinion and of the gear are defined by the following system of eqs. (9)-(10):           , 1 , , 0 3 2 1 4 3 2 1 i e ig h rd t i i i i i c c ec c           r m m m m m m m r    , 0 0 0 i c i c c  v e (1) to co generated gear generating crown gear (c) (t) (w) c0y e 01y c0x head cutter ty t0y tz ' tz t0x tx ' t0z fig. 1 spiral bevel gear hobbing 2.1. variation of machine tool setting para meters the variations of the tilt and swivel angles, tilt d istance, radial mach ine tool setting, and the ratio of roll a re conducted by polynomia l fun ctions of fifth-order:       1 10 1 10 1 10 10 11 12 2 5 15 .... c c c c c c c c c c                        1 10 1 10 1 10 20 21 22 2 5 25 .... c c c c c c c c c c                        1 10 1 10 1 10 30 31 32 2 5 35 .... c c c c c c d h c c c c                       1 10 1 10 1 10 40 41 42 2 5 45 .... c c c c c c e c c c c                        1 10 1 10 1 10 1 50 51 52 2 5 55 .... c c c c c c g i c c c c                  (2) where 1c  is the angle of rotation of the imaginary generating cro wn gear in pin ion tooth surface generation. t h e re fo re , th e ma xi mu m t o o th c on ta ct pressure, ma ximu m transmission error, ehd load carrying capacity, and frict ion factor depend on 33 manufacture parameters: advances in technology innovation, vol. 2, no. 3, 2017, pp. 61 67 63 copyright © taeti   1 2 55 max max 0 1 0 , , , prof prof ji t ij i j mp r r p p r c                   1 2 55 2 max 2 max 0 1 0 , , , prof prof ji t ij i j mp r r r c                       1 2 55 0 1 0 , , , prof prof ji t ij i j mp r r w w r c                   1 2 55 0 1 0 , , , prof prof ji t t t ij i j mp r r f f r c                 (3) 2.2. the optimization of machine tool settings and head-cutter geometry an optimization method is applied to sy stematica lly define optima l head-cutter geometry and machine tool settings to simultaneously minimize ma ximu m tooth contact pressure and angular displacement er ror o f the driven gear, to ma ximize the ehd load carrying capacity of the oil film, and to minimize power losses in the oil film. the proposed optimization procedure relies heavily on the loaded tooth contact analysis for the prediction of ma ximu m tooth contact pressure and transmission errors and on the ehd lubrication analysis to calculate the ehd load carrying capacity of the oil film, and the frict ion factor. the employed methods are developed in refs. [5 8] the goal of the optimization is to minimize tooth contact pressure and transmission errors , to ma ximize the ehd load carrying capacity of the oil film, and to minimize power losses in the oil film while keeping the loaded contact pattern inside the physical tooth boundaries of the pinion and the gear. the applicab le object ive functions can be expressed as       max 2 max max 0 2 max 0 p mp mp mp p f c c p          (4)       0 0 t w f t mpmp mp fw f c c w f     (5) where 0max p and 0max2  are the ma ximu m tooth contact pressure and transmis sion error, 0 w and 0t f are ehd load carry ing capacity of the oil film and the friction factor obtained for the initial va lues of manufacture parameters; p c , c , wc , and fc are non-negative weight coeff icients, expressing their relative importance. the proper constraints are due to the require ments that the contact pattern remains inside the possible contact area defined by load distribution calculation and inside the physical tooth boundaries of the pinion and the gear. it leads to the requirement that the contact load outside the instantly possible contact area should be zero. therefore, the constraint can simply be denoted by   0mpc  (6) whe re c is the tota l o f tooth su rfa ce po ints with instantaneously not e xist ing contact loads. there fo re , it depends on the tooth surface topography th rough the manu factu re p ara me ters mp. the optimization p roble m formu lated according to eqs. (4), (5), and (6) is a nonlinear constrained optimization proble m. functions  mpf and  mpc are not available analytically, they exist numerica lly through the load distribution calculation and ehd lubrication analysis. therefore, the computer simulat ion of load distribution and ehd lubrication must be run, repeatedly, in order to compute the various quantities needed by the optimization algorith m. the load distribution calculation and the ehd lubrication analysis are based on highly nonlinear systems of equations. an approximate and iterative technique is used to perform the load distribution calculation and the ehd lubricat ion analysis . this causes that the calculation of partial derivatives for gradient-based optimization algorith ms to be quite impractical. for this reason, a nonderivative method is selected to solve this particular optimizat ion problem. he re, the pattern search method is used. advances in technology innovation , vol. 2, no. 3, 2017, pp. 61 67 64 copyright © taeti 3. manufacture of face-hobbed hypoid gears on cnc hypoid generator the cnc machine for generation of spira l bevel and hypoid gears is provided with six degrees -of-freedom for three rotational mot ions ( ,  ,  ), and three translational motions (x, y, z, fig. 5). the six a xes of cnc generator are directly driven by the servo motors and able to imple ment prescribed functions of motions. the face-hobbing method requires simu ltaneous six-a xis control (the face-milling method requires only five-a xis control). the following coordinate systems are applied to describe the relations and motions in the cnc generator (fig. 2) : co o rd in a t e s yst e ms   tttt z,y,xk a nd   iiii z,y,xk a re rig id ly co nn ect ed to t he head-cutter and the pinion/gear, respectively. the coordinate transformat ion fro m system t k to system i k performs the following equation: i0 iy , y cy y t0z , x cx i0x t0x co 1o tio z ix t0y iz i0 cz ,z tz ty tx head-cutter workpiece fig. 2 machine-tool setting for pinion tooth-surface finishing on cnc gnerator       0 , , , i i ti cnc t t ti t x y z         r m m m r μ r (7) the location and the orientation of the tool with respect to the pinion/gear are given in coordinate systems that are represented for a conventional, crad le-type generator (fig. 1). an algorith m is developed for the execution of motions on the cnc generator using the relations valid for the cradle-type mach ine. this algorith m is based on the conditions that the relative position of the a xes of the head -cutter and the pinion rotations, 0t z and 0i y , and the a xia l re lative position of the head-cutter and the pinion/gear should be the same whether the pinion/gear is cut on a cradle-type or on a cnc hypoid generator. to ensure the same relative position of the two a xes, 0t z and 0i y , on both the cradle -type and cnc hypoid generating machines, the ele ments of the coordinate transformation matrices should be equal. on the basis of eqs. (1) and (7) the following condition should be satisfied:        0 00 00 0234120 ,,, t tt z tti z tcccii z i zyx em emmmmme r rr    (8) the same re lative position of the head-cutter and the pinion along their a xes in the case of both machines, is satisfied by applying the following condition       0 2 1 4 3 2 0 0 0 t t t o o i i i c c c t o ti t         r m m m m m r m r (9) 4. results and discussion a computer p rogra m was developed to i mple ment the formu lation provided above. by applying this program the optimal mach ine tool settings were calcu lated and functions were developed for the execution of motions on the cnc hypoid generator using the relations on the cradle-type machine. fig. 3 tooth contact pressure distribution in hypoid gear pair when the pinion and gear tooth surfaces are fully conjugate the load distribution calculat ion was performed for 21 instantaneous positions of the pinion and the gear rolling through a mesh cycle. the tooth contact pressure distributions along the potential contact lines for 21 instantaneous positions and for a ll the adjacent tooth pairs engaged for a part icular position of the mat ing advances in technology innovation, vol. 2, no. 3, 2017, pp. 61 67 65 copyright © taeti me mbe rs, fo r the case when no modificat ions are introduced into the pinion teeth of the hypoid gear pair, na me ly straight-lined head-cutter profile and the basic values of machine tool settings are applied, are shown in fig. 3. in this case the pinion and gear tooth surfaces are fu lly conjugate. the obtained ma ximu m tooth contact pressure is 369.5 mpa and the ma ximu m angular displacement error of the driven gear is 8.87 arcsec. the tooth contact pressure distribution for the case when the pinion teeth are manufactured by the head-cutter of optimized geometry and by optimal variation in machine tool settings governed by eq. (2) is shown in fig. 4. it can be observed that the ma ximu m tooth contact pre ssure is reduced to mpa4.332p max  and the maximum transmission error to secarc79.0 max2  . similar reductions in the ma ximu m tooth contact pressure and in the ma ximu m displace ment error of the driven gear a re obtained in the case of a spiral bevel gear pair (figs. 5 and 6). fig. 4 tooth contact pressure distribution in hypoid gear pair when the pinion tooth is manufactured by the h ead -cutter of optimized geo metry and by op timal variation in machine tool settings fig. 5 tooth contact pressure distribution in the spiral bevel gear pair when the pinion and gear tooth surfaces are fully conjugate fig. 6 tooth contact pressure distributions along the potential contact lines when the pinion tooth is manu fa ctu red by opt imi zed head-cutter and machine tool settings fig. 7 pressure distribution in the oil film in spiral bevel gear pair for the basic values of machine tool setting parameters fig. 8 pressure distribution in the oil film in the spiral bevel gear pair for the optima l values of machine tool setting parameters by applying the optima l co mbination of head-cutter geometry and machine tool settings the lubrication performances of the spiral bevel gear pair are imp roved. in figs. 7 and 8 it can be considered that there is a considerable increase in the ehd load carrying capacity and reduction in the power losses in the oil film. advances in technology innovation , vol. 2, no. 3, 2017, pp. 61 67 66 copyright © taeti -50 0 50 100 150 200 250 227 228 229 230 231 232 233 234 235 236 237 238  [deg.]  , [ d e g .] , x ,y ,z [ m m ]]   x y z fig. 9 motion graphs for the cnc hypoid generator for fin ishing the pinion in fun ction of the rotation angle of the head-cutter on the cnc generator -0,04 -0,03 -0,02 -0,01 0 0,01 0,02 0,03 0,04 0,05 0,06 0,07 0,08 -15 -10 -5 0 5 10 15  t [deg.]  x ,  ,  [ d e g .] ,  x ,  y ,  z [ m m ]    x y z fig. 10 diffe rences in motions on the cnc hypoid generator as results of using head-cutte r o f opt imized geo met ry , optima l polynomia l functions for the conduction of variation in machine tool settings and modified roll fo r pin ion tooth flank generation the graph shown in fig. 9, represent the e xecution of motions on the cnc hypoid ge nerator for finishing the pinion teeth governed by eq. (2). the variat ion in mot ion para meters is e xpressed in function of the rotation angle of the head -cutt er on th e cnc ge ne rato r. the differences in the values of mot ion para meters on the cnc hypoid generator, as results of using optima l polynomia l functions for the conduction of variat ion in machine tool settings and mod ified roll for p inion tooth flank generation, are shown in fig. 10. 5. conclusions an advanced method for the manufacture of spiral bevel and hypoid gears on cnc hypoid generator is presented. the optimal head -cutter geometry and mach ine tool settings are det ermined to introduce the optima l tooth modific ations into the teeth of spiral bevel and hypoid gears in order to reduce the tooth contact pre ssure and transmission errors , to ma ximize the ehd load carry ing capacity of the oil film, and to minimize power losses in the oil film. the method is based on machine tool setting variation on the cradle-type generator conducted by polynomia l functions of fifth -order. an algorith m is developed for the execution of mot ions on the cnc hypoid generator using the optima l relations on the cradle-type machine. by applying the head-cutter of optima l geometry and the optima l variation in machine tool settings the following operating parameters are improved: (1) in the case of the hypoid gear pair mo de rate reduction in the ma ximu m tooth contact pressure of 10% and a drastic reduction in the transmission errors of 91% were o btained. (2) for the spiral bevel gear pair significant reductions in the ma ximu m tooth contact pressure of 62% and in the transmis sion errors of 73% were achieved. (3) the ehd load carrying capacity of the oil film is drastically increased for 252% and the power losses in the oil film are reduced for 61% in the case of the spiral bevel gear pair. references [1] y. p. shih and z. h. fong, “flank correction for spiral bevel and hypoid gears on a six-a xis cnc hypoid generator,” asme journal of mechanical design, vol. 130, pp. 062604-1-8, 2008. [2] q. fan, “tooth surface error correction for face-hobbed hypoid gears,” asme journal of mechanical design, vol. 132, pp. 011 004-1-8, 2010. [3] z. c. chen and m . wasif, “a generic and theoretical approach to progra mming and post-processing for hypoid gear machin ing on mu lti-a xis cnc face-milling machines,” inte rn at iona l jou rna l o f advan ced manufacturing and technology, vol. 81 , pp. 135-148, 2015. advances in technology innovation, vol. 2, no. 3, 2017, pp. 61 67 67 copyright © taeti [4] w. zhang, b. cheng, x. guo, m. zhang, and y. xing, “a mot ion control method for face hobbing on cnc hypoid generator,” mechanism and machine theory, vol. 92, pp. 127-143, 2015. [5] v. simon, “load distribution in hypoid gears,” asme journal of mechanical design, vol. 122, no. 4, pp. 529-535, 1998. [6] v. simon, “ load distribution in spiral bevel gears,” asme journa l of mechanica l design, vol. 129, pp. 201-209, 2007. [7] v. simon, “ elastohydrodynamic lubricat ion of hypoid gears ,” proc. of third international po we r t rans miss ion an d ge a rin g conference, asm e journal of mechanica l design, vol. 103, pp. 195-203, 1981. [8] v. simon, “ influence of machine tool setting para meters on ehd lubrication in hypoid gears,” mechanism and machine theory, vol. 44, pp. 923-937, 2009. [9] v. simon, “in fluence of tooth modificat ions on tooth contact in face-hobbed s piral bevel gears,” mechanism and machine theory, vol. 46, pp. 1980-1998, 2011. [10] v. simon, “opt imization of face-hobbed hypoid gears,” mechanism and machine theory, vol. 77, pp. 164-181, 2014. 3__aiti#7309__157-168 advances in technology innovation, vol. 6, no. 3, 2021, pp. 157-168 design philosophy for buildings’ comfort-level performance nanang gunawan wariyatno 1,2 , han ay lie 1 , fu-pei hsiao 3 , buntara sthenly gan 4,* 1 civil engineering department, faculty of engineering, universitas diponegoro, semarang, indonesia 2 civil engineering department, faculty of engineering, universitas jenderal soedirman, purwokerto, indonesia 3 national center for research on earthquake engineering, tainan, taiwan 4 department of architecture, college of engineering, nihon university, koriyama, japan received 15 march 2021; received in revised form 27 april 2021; accepted 28 april 2021 doi: https://doi.org/10.46604/aiti.2021.7309 abstract the data reported by japan meteorological agency (jma) show that the fatal casualties and severe injuries are due to heavy shaking during massive earthquakes. current earthquake-resistant building standards do not include comfort-level performance. hence, a new performance design philosophy is proposed in this research to evaluate the quantitative effect of earthquake-induced shaking in a building. the earthquake-induced response accelerations in a building are analysed, and the response accelerations related with the characteristic property of the building are used to evaluate the number of seismic intensity level (sil). to show the indispensability of the newly proposed comfort-level design philosophy, numerical simulations are conducted to evaluate the comfort level on different floors in a building. the results show that the evaluation of residents’ comfort levels should be considered in the current earthquake-resistant building design codes. keywords: comfort-level, seismic intensity level (sil), quantification, earthquake, design philosophy 1. introduction japan is susceptible to strong earthquakes and volcanic eruptions owing to its location at the junction of four active tectonic plates: the pacific, north american, eurasian, and philippine plates. since the 19th century, 20% of the recorded strong earthquakes worldwide with a magnitude larger than six on the richter scale have occurred in japan [1]. the technological development of earthquake-resistant buildings in japan started in 1891 during the meiji era. in 1919, the first law related to the construction of earthquake-resistant buildings was issued. japan’s design codes are readjusted and renewed whenever a massive earthquake with a high number of casualties occurs. to prevent the collapse of buildings, the latest design codes require that structures should undergo plastic deformations while behaving elastically under small to moderate earthquake conditions. this new design method is effective for conserving human lives in the case of building collapse. however, it does not consider the possibility of deaths due to non-structural failure induced by vibration and shaking. japan’s early warning technology, which was managed by japan meteorological agency (jma), has become very sophisticated based on japan’s experience with mitigating earthquake disasters. this study presents a new and universal design philosophy for quantifying shakings and vibrations during earthquakes based on the jma method. the evaluation process involves the determination of seismic intensity level (sil) with the primary purpose of defining countermeasures or preventive actions for minimizing the number of human casualties. * corresponding author. e-mail address: gan.buntarasthenly@nihon-u.ac.jp tel.: +81-24-9568735 advances in technology innovation, vol. 6, no. 3, 2021, pp. 157-168 there is plenty of existing literature on the earthquake-resistant buildings and the mitigating earthquake disasters. however, to the best of authors’ knowledge, there is no literature on evaluating the comfort-level performance of buildings against the shakings during earthquakes. during the occurrences of earthquakes, sil values are calculated immediately from the earthquake wave information at the ground of every recording station during an earthquake. higher floors in high-rise buildings will experience more vigorous shaking than the ground. therefore, the present work will expand sil’s application from ground shaking to specific floors in a general building. 2. statistics on human casualties, building collapse, and damage for 23 years (since 1996), jma has collected data from 153 locations, including the seismic magnitude, sil, number of human casualties, and building damage or collapse due to significant earthquakes [2-3]. excluding the 2011 tohoku earthquake in which the human casualties and building damage/collapse were primarily due to the tsunami, the comparison between the number of human casualties (death or injured) and the number of buildings that collapsed or were damaged are presented in fig. 1. fig. 1 statistical data on human and building casualties due to earthquakes [2, 4] as shown in fig. 1, the number of injured people is significantly larger than the number of deaths or missing people, and the ratio has increased over the years. however, the number of collapsed buildings’ ratio to the number of damaged structures was far lower. the earthquake-resistant design standards provide excellent earthquake resistance in relation to just the building’s strength. on the contrary, the deaths and injuries are not mainly due to building collapse [5]. strong earthquakes that cause vibrations can result in injuries and fatalities [6-7]. thus, structural damage plays a crucial role in the number of human casualties during earthquakes [8]. the design codes, which only focus on a structure’s strength, do not suffice to protect inhabitants from injury. the behavior of buildings with the reinforced concrete subjected to lateral loads has been tested to evaluate their strengths [9-11]. the casualties attributable to the extensive shaking of a non-collapsed or damaged building should be considered. fig. 2 shows the relationship between the earthquakes’ magnitude on the richter scale and sil [2, 12-13]. as shown in fig. 2, the recorded magnitudes at earthquakes’ epicenters on the richter scale and sil deviate from their linear relationship [2, 14-15]. the non-linear relationship between the magnitudes and sil is caused by the location of the earthquake recording stations where sil is calculated depending on soil conditions, distances, and depths from the earthquake epicenter. hence, the proposed new seismic-performance design code for casualties attributable to non-collapsed structures should be based on sil values, instead of earthquakes’ magnitude. sil values provide a realistic representation of the real shaking because the effects of the building’s distance from the epicenter, the soil conditions, the buildings’ 1 10 100 1000 10000 100000 1000000 1995 1998 2001 2004 2006 2009 2012 2014 2017 f at al it ie s/ in ju rr ie s an d s tr u ct u ra l d am ag e/ fa il u re (l o g s ca le ) year structural damage injuries structural failure dead and/or missing 158 advances in technology innovation, vol. 6, no. 3, 2021, pp. 157-168 characteristics, the earthquake’s magnitude, and human perceptions of the shaking phenomena due to the earthquake are considered. the quantification of human perceptions during shaking provides accurate scientific information of the buildings’ residents. fig. 2 relationship between the magnitude on the richter scale and sil [2, 4] 3. about sil before 1996, in japan, sil was qualitatively estimated based on human perceptions, and the numbers of casualties attributable to earthquakes were considered secondary information. since then, attempts have been made to evaluate sil quantitatively. jma compiled the data from approximately 4000 sil observation stations throughout japan (recently added by fire and disaster management agency (fdma)). sil has been widely utilized in japan to broadcast quick and informative earthquake warnings to the entire country through the television for disaster mitigation when an earthquake hits the country [2, 16-18]. table 1 presents the descriptions of sil values with the corresponding sil (computed m-sil) ranges in 10 categories based on human perceptions and buildings’ damage level. table 1 descriptions of sil (freely translated from japanese) sil m-sil human perception and reaction 0 0–0.4 imperceptible to people. 1 0.5–1.4 felt slightly by a few people inside a quiet room of a building. 2 1.5–2.4 felt by some people inside a quiet room of a building. 3 2.5–3.4 felt by most people in a building. 4 3.5–4.4 most people are startled, hanging objects such as lamps swing significantly, and unstable ornaments may fall. 5 weak 4.5–4.9 many people are frightened and feel the need to hold onto something stable. dishes in cupboards and items on bookshelves may fall. unsecured furniture may move, and unstable furniture may topple over. 5 strong 5.0–5.4 many people find that it is difficult to walk without holding onto something stable. dishes in cupboards and items on bookshelves are more likely to fall. unsecured furniture may topple over. unreinforced concrete-block walls may collapse. 6 weak 5.5–5.9 people find that it is difficult to remain standing. unsecured furniture moves and topples over. doors may become stuck shut. wall tiles and windows may sustain damage and fall. in wooden houses with low earthquake resistance, tiles may fall, and buildings may lean or collapse. 6 strong 6.0–6.4 people find that it is difficult to move without crawling. people may be thrown from their position. most unsecured furniture move and are more likely to topple over. wooden houses with a low earthquake resistance are more likely to lean or collapse (compared with previous cases). large cracks may occur, as well as large landslides and collapses. 7 ≥ 6.5 wooden houses with low earthquake resistance are even more likely to lean or collapse. wooden houses with a high earthquake resistance may lean. the majority of reinforced concrete buildings with low earthquake resistance will collapse. 3.0 4.0 5.0 6.0 7.0 8.0 9.0 3.0 4.0 5.0 6.0 7.0 8.0 9.0 e ar th qu ak e m ag ni tu de ( m ar ch 1 9 9 6 ju n e 2 0 1 8 ) seismic intensity level (march 1996 june 2018) 159 advances in technology innovation, vol. 6, no. 3, 2021, pp. 157-168 additionally, modified mercalli intensity (mmi) is used to classify the shaking effects on humans. although the mmi scale has 12 intensity scales with pictorial illustrations and simple explanations regarding the effects of shaking on humans, the perceptions are essentially qualitative [19-22]. because sil can be evaluated quantitatively, it is an essential evaluation tool in designing buildings to reduce the seismic intensity and the number of human casualties. it should be noted that the sil values are calculated immediately from the earthquake wave information at every recording station during the earthquake. it is most likely that higher floors in high-rise buildings will experience higher shaking sil than the ground, hence more human casualties were due to the unmitigated design of the buildings. 4. calculating sil (m-sil) the procedure for computing m-sil value (the calculated sil value) of the jma intensity scale was presented in [13]. a brief explanation is presented herein for clarity. first, the fast fourier transform (fft) is applied to each of the threecomponent acceleration recorded in the frequency domain of the ground responses [18]. the bandpass filters given in eqs. (1)-(6) are then applied to the frequency domain of the three-component accelerations. after the filters are applied, the frequency domain’s accelerations are transformed back to the time domain. the normalized vectored composition of the three components is used to calculate the amplitude of the acceleration. the intensity is automatically calculated using threecomponent ground acceleration records after the application of a band pass filter [13], as follows: 1 2 3λ λ λ λ= × × (1) 1 1 f λ = (2) 2 4 6 8 10 12 2 1 0.694 0.241 0.0557 0.009664 0.00134 0.000155y y y y y yλ = + + + + + + (3) 38 3 1 f eλ −= − (4) 10 f y = (5) where � represents the dominant frequency, �� represents the filter on-period effect, �� represents the high-cut filter, and �� represents the low-cut filter. by using the filtered time-domain acceleration and its vectored acceleration, m-sil can be determined by using eq. (6). ( )0.3m-sil 2 log 0.94a= + (6) where ��.� is the lowest maximum acceleration of a continuously total duration of 0.3 seconds around the peak acceleration response. according to previous studies [5, 23], ��.� must be determined at the lowest maximum acceleration of a continuously total duration of 0.3 seconds around the absolute values of peak acceleration response. the sil value can be classified by using the computed m-sil value as shown in table 1. fig. 3 illustrates the way to determine the ��.� values. fig. 3(a) shows the band pass filtered absolute acceleration response in the time domain. the dotted line defines the lowest maximum acceleration ��.� of a continuously total duration of 0.3 seconds around the peak acceleration. in the second method, ��.� can also be determined by plotting the cumulative duration against the vectored acceleration, as shown in fig. 3(b). the lowest maximum acceleration ��.� can be determined at the cumulative duration of 0.3 seconds. 160 advances in technology innovation, vol. 6, no. 3, 2021, pp. 157-168 (a) fft spectrum of response acceleration (b) cumulative duration of fft spectrum fig. 3 illustrative response acceleration for calculating ��.� 4.1. implementation of quantitative sil in seismic-performance design the implementation of the quantitative values m-sil derived from eq. (6) should be implemented into the performance design of buildings. fig. 4 shows a flowchart of the proposed seismic-performance design concept. fig. 4 flowchart of the proposed comfort-level performance design using sil 4.2. numerical simulation to demonstrate the effectiveness and usefulness of the proposed method, a numerical simulation is performed on a building. numerical simulations have been proven to be a sophisticated method for evaluating the seismic performance of steel structures [24-25]. fig. 5 shows three prototypes of a two-story steel frame structure. these three prototypes of the same structural members’ steel structures are evaluated and classified according to their strengthening methods. the first structure uses moment-resisting frame (mrf) method, the second structure uses concentrically braced frame (brf) method, and the third structure uses mrf equipped with base isolation (bi) system. the columns are wf 250×250×8×13 sections, and the beams and bracings are wf 250×125×6×9 sections. all structural members are made of grade a36 steel. the span in all the directions is 4.00 m, with a floor-to-floor height of 3.50 m (fig. 5). the bi structure has a typical bi rubber bearing. the strengthening methods of the three structures are designed to fulfill the earthquake-resistant design required by the building codes. vectored acceleration [gal ] c u m m u la ti v e d u ra ti o n [ s] t a a0.3 t = 0.3 sec time history response analysis of building (dynamic analysis during the earthquake wave) input: earthquake wave analysis model (building + soil + foundation) output: building responses （acceleration, velocity, displacement） response acceleration response displacement calculate the value of a0.3 (jma) determine the dominant period, tdom (using fft) plot (tdom, a0.3) onto the sil graph of jma 161 advances in technology innovation, vol. 6, no. 3, 2021, pp. 157-168 (a) perspective view of three prototypes of two-story steel frame structures (b) three prototypes of two-story steel frame structures fig. 5 types of building strengthening for sil evaluation the structures are assumed to be located on soft soil as this type of soil shakes the most during earthquakes. fig. 6 shows a prior-selected response spectrum that is used to generate artificial earthquake waves. the response spectrum is obtained from semarang (smg), a city in central java (7.0051° s, 110.4381° e). in this study, several earthquake waves [26-27] recommended for seismic-performance evaluation are considered for simulation purposes. the response spectrum (fig. 6) is used to generate artificial earthquake waves following the corresponding earthquake waves as tabulated in table 2. in the present numerical simulation, the earthquake waves are used for comparison purposes. to develop and check the consistency of the measured and computed signal processing of the earthquake waves, data consistency assessment function (dcaf) [28-29] is recommended. table 3 shows the maximum acceleration, the lowest maximum acceleration ��.�, and the m-sil calculated using eq. (6) at smg location for different earthquake waves in table 3. it can be observed that the values of m-sil shown in table 3 are determined to be about 5.4 in order to fit with the selected response spectrum of smg. the three structural prototypes are analyzed by using the linear material dynamic time-history analysis with the input earthquake waves taken from the artificial acceleration time-history as shown in table 2. fig. 6 response spectrum for soft soil of smg location 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0 0.5 1 1.5 2 2.5 3 3.5 4 s p ec tr al a cc el er at io n [ g a l ] period t [s] a s 162 advances in technology innovation, vol. 6, no. 3, 2021, pp. 157-168 table 2 artificial acceleration time-history at smg site earthquake year magnitude artificial acceleration time-history big bear 1992 6.46 el centro 1940 6.95 kobe 1995 6.90 loma prieta 1989 6.93 morgan hill 1984 6.19 mammoth lake 1980 6.06 northridge 1994 6.69 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0 10 20 30 40 50 60 a cc el er at io n [ g a l] time [s] -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0 10 20 30 40 50 60 a cc el er at io n [g a l] time [s] -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0 10 20 30 40 50 60 a cc el er at io n [ g a l] time [s] -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0 10 20 30 40 50 60 a cc el er at io n [g a l] time [s] -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0 10 20 30 40 50 60 a cc el er at io n [ g a l] time [s] -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0 10 20 30 40 50 60 a cc el er at io n [ g a l] time [s] -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0 10 20 30 40 50 60 a cc el er at io n [ g a l] time [s] 163 advances in technology innovation, vol. 6, no. 3, 2021, pp. 157-168 table 3 values of m-sil at smg location (ground) earthquake sampling location maximum acceleration [gal] lowest maximum acceleration ��.� [gal] m-sil big bear-smg 280 178 5.4408 el centro-smg 290 173 5.4161 kobe-smg 300 206 5.5677 loma prieta-smg 310 189 5.4929 morgan hill-smg 320 187 5.4837 mammoth lake-smg 310 173 5.4161 northridge-smg 290 166 5.3802 (a) displacement-time response (b) fft spectrum fig. 7 rooftop response (using el centro-smg earthquake wave) table 4 tdom calculation results using fft earthquake sampling location axis floor mrf brf bi tdom tdom tdom [s] [s] [s] big bear x roof 0.356 0.145 2.114 2 nd 0.356 0.145 2.114 y roof 0.475 0.130 2.048 2 nd 0.475 0.130 2.048 el centro x roof 0.370 0.147 2.260 2 nd 0.370 0.147 2.260 y roof 0.468 0.127 2.260 2 nd 0.468 0.127 2.260 kobe x roof 0.377 0.144 2.114 2 nd 0.377 0.144 2.114 y roof 0.471 0.128 2.114 2 nd 0.471 0.128 2.114 loma prieta x roof 0.360 0.148 2.114 2 nd 0.360 0.148 2.114 y roof 0.462 0.133 2.114 2 nd 0.462 0.133 2.114 morgan hill x roof 0.364 0.144 2.114 2 nd 0.364 0.144 2.114 y roof 0.465 0.124 2.114 2 nd 0.465 0.124 2.114 mammoth lake x roof 0.354 0.143 2.260 2 nd 0.354 0.143 2.260 y roof 0.478 0.134 2.185 2 nd 0.478 0.134 2.185 northridge x roof 0.354 0.149 2.114 2 nd 0.354 0.149 2.114 y roof 0.493 0.132 2.114 2 nd 0.493 0.132 2.114 -35 -25 -15 -5 5 15 25 35 0 10 20 30 40 50 60 d is p la c em en t [m m ] time [s] mrf brf bi 0 2000 4000 6000 8000 10000 12000 14000 0.01 0.1 1 10 | y ( fr eq u en cy ) | frequency [hz] mrf brf bi 164 advances in technology innovation, vol. 6, no. 3, 2021, pp. 157-168 table 5 sil evaluation of the rooftop floor earthquake sampling location axis floor mrf brf bi tdom a0.3 jma tdom a0.3 jma tdom a0.3 jma (s) (gal) sil (s) (gal) sil (s) (gal) sil big bear x roof 0.356 702.58 6 strong 0.145 567.61 5 strong 2.114 22.05 4 2 nd 405.20 6 weak 393.82 5 strong 15.98 3 y roof 0.475 749.33 6 strong 0.130 507.91 5 strong 2.048 31.61 4 2 nd 500.03 6 strong 367.23 5 weak 28.26 4 el centro x roof 0.370 757.04 6 strong 0.147 478.93 5 strong 2.260 15.30 3 2 nd 438.34 6 weak 338.93 5 weak 14.68 3 y roof 0.468 702.20 6 strong 0.127 475.62 5 strong 2.260 26.61 4 2 nd 433.15 6 weak 367.74 5 strong 25.64 4 kobe x roof 0.377 522.19 6 weak 0.144 502.98 5 strong 2.114 18.90 4 2 nd 332.22 6 weak 340.01 5 strong 14.87 3 y roof 0.471 700.58 6 strong 0.128 442.08 5 strong 2.114 26.64 4 2 nd 402.12 6 weak 324.12 5 weak 25.98 4 loma prieta x roof 0.360 688.63 6 strong 0.148 512.13 5 strong 2.114 18.13 4 2 nd 422.38 6 weak 382.89 5 strong 16.06 3 y roof 0.462 682.83 6 strong 0.133 468.15 5 strong 2.114 32.05 4 2 nd 404.61 6 weak 344.48 5 weak 30.43 4 morgan hill x roof 0.364 770.13 6 strong 0.144 475.01 5 strong 2.114 18.70 4 2 nd 401.77 6 weak 328.11 5 weak 15.69 3 y roof 0.465 685.59 6 strong 0.124 439.83 5 strong 2.114 30.88 4 2 nd 385.01 6 weak 325.90 5 weak 29.45 4 mammoth lake x roof 0.354 770.13 6 strong 0.143 531.88 5 strong 2.260 22.02 4 2 nd 401.77 6 weak 367.21 5 weak 15.24 3 y roof 0.478 701.33 6 strong 0.134 534.70 5 strong 2.185 26.17 4 2 nd 442.03 6 weak 381.01 5 weak 27.35 4 northridge x roof 0.354 687.90 6 strong 0.149 538.96 5 strong 2.114 19.89 4 2 nd 406.55 6 weak 372.79 5 strong 16.00 3 y roof 0.493 764.98 6 strong 0.132 489.06 5 strong 2.114 28.86 4 2 nd 456.38 6 weak 357.85 5 weak 28.17 4 the displacement response on every floor in the buildings is analyzed using the time-history analysis [30] with a particular earthquake wave input. we determine the dominant period tdom on every floor in the buildings from the history of displacement response. fig. 7 shows the representative displacement responses on the rooftop of three buildings under evaluation using the el centro-smg earthquake wave as the input. a standard fft is performed on the time against displacement response relationship [31-32] to determine the buildings’ time dominant period (fig. 7(a)). fig. 7(b) presents the resulting fft spectrum of the displacement time response of three evaluated buildings using the el centro-smg earthquake wave as the input. table 4 shows the time-dominant period tdom based on the fft calculation, and table 5 presents the sil values for each floor obtained from the lowest maximum acceleration ��.�. in this simulation, the rooftop floor has a higher acceleration ��.� than the second floor for all types of prototypes and all time-history earthquakes in both xand y-axis directions. the acceleration is directly correlated to the sil value. a higher ��.� acceleration resulted in the higher sil value calculated from the acceleration to time-dominant period relationship is plotted in fig. 8 [4]. the results show that the bi prototype has the lowest ��.� acceleration and sil value. the sil value ranges from 3 to 4, corresponding to an ��.� in the range of 15-32 gal. for the mrf prototype, sil ranges from 6 weak to 6 strong, corresponding to 332-770 gal. the brf prototype has an ��.� in the range of 324-539 gal, and sil ranges from 5 strong to 5 weak. a graphical representation of the sil curves as a function of the time-dominant period and maximum acceleration ��.� for the el centro-smg case is shown in fig. 8. an examination of these graphs reveals that the range of the curve corresponds to the ��.� and time-dominant period (tdom) of the prototypes. 165 advances in technology innovation, vol. 6, no. 3, 2021, pp. 157-168 fig. 8 sil as a function of the dominant period and acceleration [2, 33] as shown in fig. 8, the sil values’ categorization between lines is based on a more significant value than the lower sil values. for example, for the bi frame, the x-direction response is designated as sil-3, whereas in reality, the condition closely approaches sil-4. nevertheless, the classification assigns sil to 3. this might lead to improper simulated behavior because the structure closely resembles sil-4. it is suggested that the bandwidth between sil categories should be proposed based on either a mathematical study or an experimental study. the results of the numerical simulation reveal that the strengthening method significantly affects the sil value. bi significantly reduces and stabilizes the sil of every floor, while mrf results in a less sensitive response to sil in all directions. furthermore, these observations suggest that sil is significantly affected by the frame’s configuration and displacement direction, in contrast to the acceleration response. 5. conclusions a method for quantifying sil based on the acceleration and dominant period on every floor of a structure was proposed and demonstrated. the proposed design philosophy for buildings’ comfort-level performance is inherently a new and universal design philosophy that can be applied to different building codes and various earthquake waves. one of the advantages of this sil calculation method is that it can quantify the shaking on each floor of a building during earthquakes, which will aid the quantitative measuring of the residents’ comfort levels. the proposed sil-based performance evaluation is beneficial in the planning, designing, and strengthening stages of building constructions. the silbased method for calculating the level of shaking can also be used to quantitatively evaluate the performance of newly developed earthquake-resistant building reinforcement equipment. additionally, the present proposed method can monitor the shaking in buildings during earthquakes so that suitable measures can be taken quickly. considering the significant potential of the proposed sil method for quantifying the level of shaking, the residents’ comfort level should be accounted for in current earthquake-resistant design building codes. 6. future recommendation experimental works on building prototypes have to be conducted to precisely verify the quantitative shaking intensity level for research directions. for practical applications, the present proposed method could be beneficial for the following: 0.1 1 10 100 1000 10000 0.01 0.1 1 10 100 a cc el er at io n [g a l ] dominant period tdom [s] mrf x y bi x y 0 .3 a 166 advances in technology innovation, vol. 6, no. 3, 2021, pp. 157-168 (1) in the planning and designing stages, to predict the quantitative shaking that will occur in the building on different floors. (2) in the existing stage, to characterize the quantitative shaking subjected to a specific earthquake. (3) in the strengthening stage, to evaluate the quantitative shaking for decision making. (4) in the research and development project stage, to quantify the level of shaking of earthquake instruments such as the base isolator, active/passive control system, and damper equipment. conflicts of interest the authors report no potential competing interests. references [1] “lists, maps, and statistics,” https://earthquake.usgs.gov/earthquakes/browse/m6-world.php, 2018. [2] “major earthquakes that occurred near japan (since 1996),” www.data.jma.go.jp/svd/eqev/data/higai/higai1996new.html, 2018. (in japanese) [3] k. ishibashi, “status of historical seismology in japan,” annals of geophysics, vol. 47, no. 2-3, pp. 339-368, 2004. [4] “seismic intensity and acceleration,” http://www.data.jma.go.jp/svd/eqev/data/kyoshin/kaisetsu/comp.htm, 2018. (in japanese) [5] h. kawasumi, “measures of earthquake danger and expectancy of maximum intensity throughout japan as inferred from the seismic activity in historical times,” bulletin of the earthquake research institute, vol. 29, pp. 469-482, 1951. [6] s. maruyama, y. s. kwon, and k. morimoto, “seismic intensity and mental stress after the great hanshin-awaji earthquake,” environmental health and preventive medicine, vol. 6, no. 3, pp. 165-169, october 2001. [7] m. takegami, y. miyamoto, s. yasuda, m. nakai, k. nishimura, h. ogawa, et al., “comparison of cardiovascular mortality in the great east japan and the great hanshin-awaji earthquakes—a large-scale data analysis of death certificates,” circulation journal, vol. 79, no. 5, pp. 1000-1008, 2015. [8] l. a. dengler and j. w. dewey, “an intensity survey of households affected by the northridge, california, earthquake of 17 january 1994,” bulletin of the seismological society of america, vol. 88, no. 2, pp. 441-462, 1998. [9] m. teguh, “structural behaviour of precast reinforced concrete frames on a non-engineered building subjected to lateral loads,” international journal of engineering and technology innovation, vol. 6, no. 2, pp. 152-164, february 2016. [10] w. t. lin, t. l. weng, and e. t. chang, “deterioration estimation of reinforced concrete building structures using material testing data base,” international journal of engineering and technology innovation, vol. 9, no. 1, pp. 22-37, january 2019. [11] r. hore, s. chakraborty, a. m. shuvon, and m. a. ansary, “effect of acceleration on wrap faced reinforced soil retaining wall on soft clay by performing shaking table test,” proceedings of engineering and technology innovation, vol. 15, pp. 24-34, april 2020. [12] k. r. karim and f. yamazaki, “correlation of jma instrumental seismic intensity with strong motion parameters,” earthquake engineering and structural dynamics, vol. 31, no. 5, pp. 1191-1212, may 2002. [13] k. t. shabestari and f. yamazaki, “attenuation relationship of jma seismic intensity using jma records,” 10th japan earthquake engineering symposium, 1998, pp 529-534. [14] a. tan and a. irfanoglu, “correlation between ground motion based shaking intensity estimates and actual building damage,” 15th world conference on earthquake engineering, september 2012, pp. 28588-28597. [15] n. gunawan, a. han, and b. s. gan, “proposed design philosophy for seismic-resistant buildings,” civil engineering dimension, vol. 21, no. 1, pp. 1-5, march 2019. [16] o. kamigaichi, “jma earthquake early warning,” journal of japan association for earthquake engineering, vol. 4, no. 3, pp. 134-137, 2004. [17] v. sokolov, t. furumura, and f. wenzel, “on the use of jma intensity in earthquake early warning systems,” bulletin of earthquake engineering, vol. 8, no. 4, pp. 767-786, 2010. [18] v. y. sokolov, “seismic intensity and fourier acceleration spectra: revised relationship,” earthquake spectra, vol. 18, no. 1, pp. 161-187, february 2002. 167 advances in technology innovation, vol. 6, no. 3, 2021, pp. 157-168 [19] d. j. dowrick, g. t. hancox, n. d. perrin, and g. d. dellow, “the modified mercalli intensity scale,” bulletin of the new zealand society for earthquake engineering, vol. 41, no. 3, pp. 193-205, september 2008. [20] m. grigoriu, “do seismic intensity measures (ims) measure up?” probabilistic engineering mechanics, vol. 46, pp. 80-93, october 2016. [21] k. t. shabestari and f. yamazaki, “a proposal of instrumental seismic intensity scale compatible with mmi evaluated from three-component acceleration records,” earthquake spectra, vol. 14, no. 4, pp. 711-723, november 2001. [22] h. o. wood and f. neumann, “modified mercalli intensity scale of 1931,” bulletin of the seismological society of america, vol. 21, no. 4, pp. 277-283, december 1931. [23] k. du, b. ding, w. bai, j. sun, and j. bai, “quantifying uncertainties in ground motion-macroseismic intensity conversion equations: a probabilistic relationship for western china,” journal of earthquake engineering, vol. 24, pp. 1-25, may 2020. [24] e. f. deng, l. zong, y. ding, z. zhang, j. f. zhang, f. w. shi, et al., “seismic performance of mid-to-high rise modular steel construction—a critical review,” thin-walled structures, vol. 155, 106924, october 2020. [25] k. kostinakis, i. k. fontara, and a. m. athanatopoulou, “scalar structure-specific ground motion intensity measures for assessing the seismic performance of structures: a review,” journal of earthquake engineering, vol. 22, no. 4, pp. 630-665, 2018. [26] k. a. porter, “an overview of peer’s performance-based earthquake engineering methodology,” 9th international conference on applications of statistics and probability in civil engineering, july 2003, pp. 1-8. [27] y. chen, p. avitabile, and j. dodson, “data consistency assessment function (dcaf),” mechanical systems and signal processing, vol. 141, 106688, july 2020. [28] y. chen, p. avitabile, c. page, and j. dodson, “a polynomial based dynamic expansion and data consistency assessment and modification for cylindrical shell structures,” mechanical systems and signal processing, vol. 154, 107574, 2021. [29] y. chen, d. joffre, and p. avitabile, “underwater dynamic response at limited points expanded to full-field strain response,” journal of vibration and acoustics, vol. 140, no. 5, 051016, october 2018. [30] m. beer, i. a. kougioumtzoglou, e. patelli, and s. k. au, encyclopedia of earthquake engineering, heidelberg: springer berlin heidelberg, 2015. [31] y. chen, p. logan, p. avitabile, and j. dodson, “non-model based expansion from limited points to an augmented set of points using chebyshev polynomials,” experimental techniques, vol. 43, no. 5, pp. 521-543, october 2019. [32] s. günay and k. m. mosalam, “peer performance-based earthquake engineering methodology, revisited,” journal of earthquake engineering, vol. 17, no. 6, pp. 829-858, 2013. [33] “seismic intensity level and acceleration, weather and earthquakes, seismic intensity information 2003,” http://www.jma.go.jp/en/quake/quake_sindo_index.html, 2020. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). 168  advances in technology innovation, vol. 5, no. 3, 2020, pp. 147-155 improved whale optimization algorithm based on inertia weights for solving global optimization problems i-ming chao 1 , shou-cheng hsiung 1,* , jenn-long liu 2 1 department of industrial management, i-shou university, kaohsiung, taiwan 2 department of information management, i-shou university, kaohsiung, taiwan received 08 september 2019; received in revised form 09 december 2019; accepted 20 february 2020 doi: https://doi.org/10.46604/aiti.2020.4167 abstract whale optimization algorithm (woa) is a new kind of swarm-based optimization algorithm that mimics the foraging behavior of humpback whales. woa models the particular hunting behavior with three stages: encircling prey, bubble-net attacking, and search for prey. in this work, we proposed a new linear decreasing inertia weight with a random exploration ability (ldiwr) strategy. it also compared with the other three inertia weight woa (iwwoa) methods: constant inertia weight (ciw), linear decreasing inertia weight (ldiw), and linear increasing inertia weight (liiw) by adding fixed or linear inertia weights to the position vector of the reference whale. the four iwwoas are tested with 23 mathematical and theoretical optimization benchmark functions. experimental results show that most of iwwoas outperform the original woa in terms of solution accuracy and convergence rate when solving global optimization problems. accordingly, the ldiwr strategy produces a better balance between exploration and exploitation capabilities for multimodal functions. keywords: whale optimization algorithm, bubble-net feeding method, inertia weights, exploration and exploitation capabilities 1. introduction optimization plays essential roles in scientific research, management, and industry because numerous real-world problems can mostly model as optimization tasks [1]. in recent years, many meta-heuristic algorithms have been widely applied as powerful tools to solve optimization problems. due to the following reasons: (i) have fewer parameters; (ii) do not require gradient information; (iii) can bypass local optima; (iv) can be utilized to solve the practical problems [2]. there are many popular population-based meta-heuristic optimization algorithms, such as particle swarm optimization (pso) [3], ant colony system (acs) [4], artificial bee colony (abc) [5], cuckoo search (cs) [6], fruit fly optimization algorithm [7], and whale optimization algorithm (woa) [8]. among them, woa, proposed by mirjalili and lewis [8], is competitive. due to the simplicity of woa in implementation and only two main parameters adjusted, the algorithm has shown superior compared to the state-of-art meta-heuristic algorithms from the testing results of different benchmark functions and engineering design problems [8-9]. basically, the algorithm is inspired by the foraging behavior of humpback whales. humpback whales in a group hunt school of krill or small fishes by shrinking and encircling around them to herd them to close to the sea surface and generating bubbles along a helix-shaped or ‘9’-shaped path to perform a bubble-net attack [8-9]. population-based meta-heuristic optimization algorithms have a common feature regardless of their nature — the search process dividing into two phases: exploration and exploitation [8, 10]. the mechanisms of shrinking encircling and spiral * corresponding author. e-mail address: melvinisstrong@gmail.com advances in technology innovation, vol. 5, no. 3, 2020, pp. 147-155 148 updating represent the exploitation phase, and the method of a random search for prey represents the exploration phase. finding a proper balance between exploration and exploitation is the most challenging task in the development of any metaheuristic algorithm due to the stochastic nature of the optimization process. the main problems faced by woa are slow and premature convergence, similar to other meta-heuristic algorithms. therefore, many variants of woa were proposed in the literature. to enhance the convergence speed and exploitation mechanism, mafarja and mirjalili proposed a memetic algorithm by hybrid woa with simulated annealing (sa) by searching the most promising regions located by the woa algorithm to improve the exploitation [9]. because the updated solution is mainly depending on the current best solution got so far. hu et al. [2] proposed the original woa with inertia weights, which is similar to the modified pso algorithm [11], to get an improved woa. in 2018, kaur and arora proposed a chaotic whale optimization algorithm (cwoa) to replace the critical parameter ‘p’ of woa instead of 0.5 probability that whale either chose the encircling or spiral path to update the position during optimization in the original woa [12]. to conclude the literature review; different inertia weight strategies may get various incremental changes in a better solution. in this paper, a new idea was proposed by adding linear decreasing inertia weight to the position vector of the reference whale and remaining the exploration ability of the original woa. to compare with the two best inertia weight methods, obtained from the study of hu et al. [2], the efficiency of the opposite strategy is also observed for linear decreasing inertia weight method. the rest of this paper is organized as follows. section 2 presents the basis of the original woa algorithm. improving woa by four inertia weightings was performed in section 3. in section 4, the experimental results presented and result analyzed. finally, in section 5, conclusions are given. 2. whale optimization algorithm woa is inspired by the individual foraging behavior of humpback whales. fig. 1 [9] shows the special hunting method of humpback whales. woa models the behavior as two phases. the first one is the exploitation phase, including encircling prey and spiral bubble-net attacking method. the second one is the exploration phase when searching randomly for prey. fig. 1 bubble-net feeding behavior of humpback whales [9] 2.1. exploitation phase (encircling prey and spiral bubble-net attacking method) humpback whales hunt prey with two steps: encircling around them and creating bubbles-nets. first, they recognize the locations of the prey and then encircle around them. woa algorithm assumes that the solution of the current best candidate (leader whale) is targeting prey or closing to the optimum. when encircling the prey, the other whales update their positions towards the best whale obtained so far. therefore, encircling prey can be represented by the following equations [8]: ( 1) ( )    j k i p j x t x t a d (1) ( ) ( )   j j j p i d c x t x t (2) advances in technology innovation, vol. 5, no. 3, 2020, pp. 147-155 149 the number of population and variables (dimensions) is assumed as m, and n. 𝑋𝑖 𝑗 is a position matrix, 𝑖 = 1, 2, ⋯ , 𝑀, 𝑗 = 1, 2, ⋯ , 𝑁 represents the position of the i th whale on the j th dimension. 𝑋𝑝 𝑗 represents the position of prey (the leader whale p) on the j th dimension. where t indicates the current iteration, 𝑋𝑝 𝑗 (𝑡) represents the current position on the j th dimension of the leader whale p, and 𝑋𝑖 𝑗 (𝑡) represents the current position of the i th whale on j th dimension; | |denotes the absolute value. xp should be updated every iteration if there is a better solution appearing. a and c are coefficient numbers that are calculated based on random functions as follows: 1 2  a a rand a (3) 2 2 c a rand (4) 2 2 a t maxiter (5) where maxiter is the maximum number of iterations, α is linearly decreasing from 2 to 0 from the start to the end of iterations (in both exploration and exploitation phases). since rand1 and rand2 are random numbers in the intervals [0, 1], then the ranges of a and c are in the intervals [−α,α] and [0, 2], respectively. in the exploitation phase, whales swim around the prey within the shrinking circle as well as move along a spiral-shaped path at the same time to form distinctive bubbles along a 9-shaped path to perform the bubble-net attacking [8]. in other words, there are two types of behavior in the bubble-net attacking that include a shrinking encircling mechanism or spiral updating position. as mentioned before, the value of a is linearly decreased from 2 to 0. since the range of a being in the intervals[−α, α], the absolute value of a decreases for iterations. therefore, the shrinking encircling mechanism can be achieved by using eq. (1) the spiral-shaped path between a whale 𝑋𝑖 𝑗 (𝑡) and the prey 𝑋𝑝 𝑗 (𝑡) can be expressed by eq. (6) [9]. ( 1) cos(2 ) ( )    j bl j i j p x t d e l x t (6) where 𝐷𝑗 ′ = |𝑋𝑝 𝑗 (𝑡) − 𝑋𝑖 𝑗 (𝑡)| represents the distance between the i th whale and the prey (the best solution obtained so far) on the j th dimension, b is a constant for defining the logarithmic spiral shape, and l is a random number in [-1, 1]. to model the two mechanisms in the bubble-net attacking, shrinking encircling, and the spiral-shaped path. mirjalili and lewis assumed that there is a probability of 0.5 to choose between them throughout iterations as in eq. (7) [8, 9]. ( ) 0.5 ( 1) cos(2 ) ( ) 0.5            j p j j i bl j j p x t a d if p x t d e l x t if p (7) 2.2. exploration phase (search for prey) to enhance the exploration mechanism in woa, instead of updating the positions by the location of the best solution found so far. a random whale is selected to guide the search. when |𝐴| ≻ 1, a whale forced to move far away from the best-known whale to execute the global search for prey (exploration). however, when |𝐴| ≺ 1, a whale will perform a local search according to the best solution found so far (exploitation). this mechanism can model as follows [8, 9]: ( ) ( )   j j j rand i d c x t x t (8) ( 1) ( )    j j i rand j x t x t a d (9) where 𝑋𝑟𝑎𝑛𝑑 𝑗 (𝑡) represents the position on j th dimension of a random whale from the population selected by |𝐴| ≻ 1. advances in technology innovation, vol. 5, no. 3, 2020, pp. 147-155 150 3. inertia weight woa as seen in the previous section 2, the original woa updated solution mostly depends on the current best candidate solution. from the literature review, different inertia weight strategies are known to get different changes in solution [2, 11]. inertia weights are introduced ω & ωr ∈ [0, 1] into woa to get the improved whale optimization algorithm (inertia weight whale optimization algorithm, iwwoa) by adding fixed or linear inertia weights to the position of the reference whale. in the exploitation phase (|𝐴| ≺ 1), the updated method is represented by the following equations: ( ) ( )    j j j p i d c x t x t (10) ( ) 0.5 ( 1) cos(2 ) ( ) 0.5                 j p j j i bl j j p x t a d if p x t d e l x t if p (11) in the exploration phase (|𝐴| ≻ 1), the updated mathematical model is as follows: ( ) ( )    j j j rand i d r c x t x t (12) ( 1) ( )     j j i rand j x t r x t a d (13) hu et al. [2] had proved to add inertia weights into woa are competitive with other meta-heuristic algorithms: woa, foa, abc, and pso can enhance the ability of exploitation. hu et al. [2] applied four strategies, the ldiw strategy introduced by shi and eberhart [11] in 1998, sugeno function as an inertia weight (sfiw) introduced by lei et al. [13] in 2006, exponential decreasing inertia weight (ediw) strategy and ciw introduced by lu et al. [14] in 2014. iwwoa1 is the best result of the study [2] by adding smaller fixed inertia weights to the position vector of the reference whale. iwwoa2 uses the better region refer to the paper [11] by adding linear inertia weights to the reference whale to strengthen the local search ability of the original woa. iwwoa3 is the opposite way of iwwoa2. the iwwoa4 represents the new idea by adding linear decreasing inertia weight to the position vector of the reference whale and remaining the exploration ability of the original woa in the meantime. those ways are in the following: (1) iwwoa1: ω = ωr = 0.1 (ciw-the best); (2) iwwoa2: ω = ωr; ω_initial=0.9; ω_final=0.4; ω=ω_initial-(t/max_iter)×( ω_initial-ω_final) (ldiw-revised); (3) iwwoa3: ω = ωr; ω_initial=0.4; ω_final=0.9; ω=ω_initial-(t/max_iter)×( ω_initial-ω_final) (liiw-revised); (4) iwwoa4: ωr=1; ω_initial=0.9; ω_final=0.4; ω=ω_initial-(t/max_iter)×( ω_initial-ω_final) (ldiwr-proposed) where ω and ωr represent the inertia weight in the exploitation and exploration phase, respectively. when ω = ωr = 1, iwwoas become the original woa. ω = ωr = 0.1 have already been known can obtain a better solution in the ciw strategy for woa [2]. inertia weight linearly was learned to decrease from ω_initial=0.9 to ω_final=0.4 can obtain a better solution in the ldiw strategy for pso [11]. the effect of the liiw strategy on woa wants to be realized (from ω_initial=0.4 to ω_final=0.9), and the effect of the ldiwr strategy on woa. 4. results and discussion 4.1. experiments of benchmark functions to test the performance of the original woa [8] and the ldiwr strategy, the former methods proposed by hu et al. [2], the same 23 benchmark functions used in the literature [8] are taken, which consist of 7 (f1-f7) unimodal, 6 (f8-f13) advances in technology innovation, vol. 5, no. 3, 2020, pp. 147-155 151 multimodal and 10 (f14-f23) fixeddimension multimodal functions. those functions are quite different from the test functions of hu et al. [2] in 2017 when consisting of 18 unimodal functions and only 9 multimodal functions. fig. 2 shows the typical 2d plot examples of the test benchmark functions considered in this work. f3 f8 f14 f6 f12 f16 (a) unimodal (b) multimodal (c) fixeddimension multimodal fig. 2 typical 2d plots of benchmark functions for all iwwoas & woa algorithms, the population size and maximum iteration are equal to 30 and 500, respectively. for each benchmark function, iwwoas & woa algorithms were executed independently run 30 times. table 1 and table 2 compares the best, the best mean (mean) and standard deviation (std) of solutions obtained by using the original woa & iwwoas for the benchmark functions (f1-f23). 4.2. evaluation of exploitation capability (functions f1–f7) f1-f7 are unimodal functions because there is only one global optimum solution. these functions usually utilized to evaluate the exploitation ability of meta-heuristic algorithms [8]. as seen from table 1 and table 2, the smaller ciw strategy got 5 in 7 of the best means, and the ldiw and ldiwr strategy got the others. the iwwoas are highly competitive with the original woa algorithm. most of the iwwoas is better than the original woa on the mean and standard deviation of 30 independent runs. furthermore, comparing the ldiw and liiw methods, the performance of those approaches follows this order: ldiw > liiw. because the ldiw strategy (ω = ωr from 0.9 to 0.4) leads the smaller and smaller stride, so it can provide better exploitation ability than liiw (ω = ωr from 0.4 to 0.9), especially in the final stage to find the globally optimal solution of unimodal functions. it is worth mentioning that the iwwoa1 (ciw ω = ωr = 0.1) is the best method on f1, f2, f3, f4, and f7. thus, iwwoas provide excellent exploitation capability, and the smaller ciw strategy is the better choice for unimodal functions. table 1 comparison of the best, mean and std values of the objective functions obtained using the woa and iwwoa1-iwwoa2 (continued) fun original woa iwwoa1(0.1) ciw iwwoa2(0.9->0.4) ldiw f1 1.60450e-73 6.09550e-73 0.00000e+00 0.00000e+00 1.54950e-271 0.00000e+00 f2 3.90430e-49 2.01600e-48 4.50300e-220 0.00000e+00 6.22180e-144 1.63220e-143 f3 4.46769e+04 1.42133e+04 0.00000e+00 0.00000e+00 7.35480e-182 0.00000e+00 f4 5.46388e+01 2.87173e+01 1.43180e-218 0.00000e+00 3.12300e-120 1.61210e-119 advances in technology innovation, vol. 5, no. 3, 2020, pp. 147-155 152 table 1 comparison of the best, mean and std values of the objective functions obtained using the woa and iwwoa1-iwwoa2 fun original woa iwwoa1(0.1) ciw iwwoa2(0.9->0.4) ldiw f5 2.79790e+01 3.73070e-01 2.87286e+01 3.23700e-02 2.79719e+01 2.92100e-01 f6 4.19590e-01 2.67260e-01 6.09030e-01 3.07770e-01 2.81010e-01 1.25790e-01 f7 4.48670e-03 5.20030e-03 6.78090e-05 6.12410e-05 8.85760e-05 7.35660e-05 sum 5 / 0 6 / 0 2 / 5 1 / 6 0 / 1 0 / 2 f8 -9.95404e+03 1.56063e+03 -7.60267e+03 2.17479e+03 -1.10905e+04 1.63330e+03 f9 1.89480e-15 1.02040e-14 0.00000e+00 0.00000e+00 0.00000e+00 0.00000e+00 f10 4.79620e-15 2.31150e-15 8.88180e-16 0.00000e+00 2.42770e-15 1.76050e-15 f11 6.73470e-03 3.62670e-02 0.00000e+00 0.00000e+00 0.00000e+00 0.00000e+00 f12 4.06280e-02 7.38760e-02 3.70340e-02 2.43910e-02 1.24130e-02 5.75200e-03 f13 5.60110e-01 2.38900e-01 4.01050e-01 2.49290e-01 2.35320e-01 8.37050e-02 f14 3.12180e+00 3.50270e+00 6.09460e+00 3.61850e+00 2.60110e+00 3.01320e+00 f15 6.61020e-04 3.82440e-04 8.33010e-04 6.12800e-04 7.54130e-04 5.87520e-04 f16 -1.03160e+00 1.32260e-09 -9.83610e-01 1.86660e-02 -1.02900e+00 2.95320e-03 f17 3.97900e-01 1.16820e-05 5.99420e-01 1.64920e-01 3.98180e-01 4.20450e-04 f18 3.00010e+00 1.57690e-04 1.02615e+01 9.21590e+00 3.01600e+00 2.36150e-02 f19 -3.85460e+00 1.00220e-02 -3.26660e+00 4.38600e-01 -3.83800e+00 2.26320e-02 f20 -3.21410e+00 1.13040e-01 -1.80080e+00 4.58620e-01 -3.24200e+00 9.04730e-02 f21 -8.17300e+00 2.58430e+00 -2.65770e+00 1.11060e+00 -7.31410e+00 2.77040e+00 f22 -7.58410e+00 3.06820e+00 -2.56330e+00 9.32650e-01 -7.09060e+00 2.74490e+00 f23 -6.83310e+00 3.32870e+00 -2.90990e+00 1.65360e+00 -6.44140e+00 2.50680e+00 sum 5 / 5 6 / 4 11 / 3 9 / 3 0 / 3 1 / 3 remarks: 1. data with bold black font indicates the worst among the original woa and the other iwwoas 2. data with bold red font indicates the best among the original woa and the other iwwoas 3. the rows of sum indicate the accumulated number of the worst and the best in unimodal and multimodal functions 4.3. evaluation of exploration capability (functions f8–f23) multimodal functions are more complex than unimodal functions. the former functions include many local optima whose complexity increases exponentially with the number of design variables. consequently, this kind of test problems turns very popular to evaluate the exploration ability of the optimizer [8]. the results reported in table 1 and table 2 for functions f8-f23 show that the ldiwr strategy has an outstanding exploration capability. comparing the performance of the original woa with the inertia weighting methods in this work follows the order: ldiwr > original woa > liiw > ldiw > smaller ciw. the novel ldiwr method obtained 7 best mean solutions on f8-f9, f11-f14, and f23 in the 16 multimodal functions. notably, there is no worst solution obtained 8 best mean solutions on f8-f9, f11-f14, and f22-f23 in the 16 multimodal functions. notably, there is no worst solution obtained by the ldiwr strategy. previously mentioned ldiwr integrated the exploration mechanism of the original woa algorithm (ωr=1), and the better exploitation mechanism of the ldiw strategy on woa, as mentioned before. however, there are 11 worst mean solutions obtained by the ciw approach. it might be decreasing the exploration capability of iwwoa1 by using the smaller ciw strategy (ω = ωr = 0.1). it is worth mentioning that the liiw strategy is better than the ldiw for multimodal functions. because the liiw (ω = ωr from 0.4 to 0.9) leads the larger and larger stride; it could provide better exploration ability than the ldiw (ω = ωr from 0.9 to 0.4), especially in the final stage to find the globally optimal solution of multimodal functions. hence, the ldiwr is more robust than the other iwwoas and the original woa. advances in technology innovation, vol. 5, no. 3, 2020, pp. 147-155 153 table 2 comparison of the best, mean and std values of the objective functions obtained using the iwwoa3 and iwwoa4 fun iwwoa3(0.4->0.9) liiw iwwoa4(0.9->0.4) with random ability-ldiwr mean std mean std f1 1.19800e-190 0.00000e+00 1.86600e-261 0.00000e+00 f2 4.25480e-105 1.38300e-104 1.71000e-141 5.32650e-141 f3 6.60840e-120 2.78580e-119 6.52640e-178 0.00000e+00 f4 1.09580e-75 3.82330e-75 2.99650e-116 1.56740e-115 f5 2.81399e+01 2.80560e-01 2.78947e+01 2.43210e-01 f6 4.88890e-01 1.75960e-01 2.95130e-01 9.86380e-02 f7 8.46060e-05 8.72020e-05 1.49770e-04 1.20400e-04 sum 0 / 0 0 / 1 0 / 1 0 / 3 f8 -1.10724e+04 1.68468e+03 -1.14430e+04 1.51963e+03 f9 0.00000e+00 0.00000e+00 0.00000e+00 0.00000e+00 f10 1.48030e-15 1.32400e-15 2.78300e-15 1.77240e-15 f11 0.00000e+00 0.00000e+00 0.00000e+00 0.00000e+00 f12 2.33280e-02 8.77710e-03 1.20190e-02 4.66670e-03 f13 2.70880e-01 8.10020e-02 2.19660e-01 7.98360e-02 f14 3.48570e+00 3.34880e+00 1.98310e+00 2.10240e+00 f15 3.63690e-04 8.53780e-05 6.74530e-04 5.41630e-04 f16 -1.00900e+00 1.12530e-02 -1.02780e+00 5.95660e-03 f17 3.99250e-01 1.95610e-03 3.98160e-01 3.54870e-04 f18 3.01550e+00 2.42980e-02 3.01610e+00 2.44000e-02 f19 -3.83410e+00 2.84740e-02 -3.83910e+00 2.49720e-02 f20 -3.15340e+00 1.28960e-01 -3.23220e+00 9.90000e-02 f21 -4.99030e+00 6.67050e-02 -7.29060e+00 2.40980e+00 f22 -5.22030e+00 8.95650e-01 -7.66510e+00 2.64640e+00 f23 -5.17970e+00 9.96500e-01 -8.17640e+00 2.53490e+00 sum 0 / 3 0 / 6 0 / 8 0 / 6 remarks: 1. data with bold black font indicates the worst among the original woa and the other iwwoas 2. data with bold red font indicates the best among the original woa and the other iwwoas 3. the rows of sum indicate the accumulated number of the worst and the best in unimodal and multimodal 4.4. analysis of convergence behavior as learned that the quality of the solution and the convergence speed of an algorithm of the global optimal solution enormously depends on the parameter and search strategy of the optimizer [1]. therefore; analyzing the convergence behavior after different inertia weighting was added to the position vector of the reference whale in necessary. convergence curves of the original woa, ciw, ldiw, liiw, and ldiwr were compared in fig. 3 for 23 benchmark functions. iwwoas are competitive enough with the original woa on the mean best fitness in each iteration over 30 runs. the slow convergence and premature convergence problems of woa on unimodal and multimodal functions (f1 to f13) can be seen. as shown in fig. 3, the iwwoas shows three different convergence behaviors when optimizing 23 benchmark functions. firstly, the convergence of the iwwoas tends to be accelerated as iteration increases except the liiw on f1 through f4. secondly, iwwoas trend of convergence within fewer iterations. the adding inertia weight strategy proposed for ciw, ldiw, and liiw restrict the exploration capability to assist them in looking for the promising regions of unimodal and multimodal functions in the initial steps of iteration. they also cause a more rapid converge towards the optimum almost before half of the iterations. this behavior is evident in f5 through f13. thirdly, the original woa and ldiwr are almost faster than the others on the fixed-dimension multimodal functions. their excellent performance is due to their full exploration ability of woa on multimodal functions. advances in technology innovation, vol. 5, no. 3, 2020, pp. 147-155 154 f1 f2 f3 f4 f5 f6 f7 f8 f9 f10 f11 f12 f13 f14 f15 f16 f17 f18 f19 f20 f21 f22 f23 fig. 3 comparison of convergence curves obtained using the original woa and proposed four iwwoas for solving 23 benchmark functions advances in technology innovation, vol. 5, no. 3, 2020, pp. 147-155 155 5. conclusions this work introduced four inertia weighting to the position vector of the reference whale to strengthen the local search ability of the woa. the proposed iwwoas on 23 mathematics benchmark functions were conducted to analyze the exploitation, exploration, and convergence behavior by comparison with the original woa, iwwoa1 (ciw ω = ωr = 0.1), iwwoa2 (ldiw ω = ωr = 0.9 to 0.4), iwwoa3 (liiw ω = ωr = 0.4 to 0.9), and iwwoa4 (ldiwr ω =0.9 to 0.4, ωr =1). iwwoas were found to be competitive enough. according to our analysis, the smaller constant inertia weight strategy (ciw ω = ωr = 0.1) was the better choice for unimodal functions. furthermore, the linear decreasing inertia weight with random exploration ability strategy (ldiwr ω =0.9 to 0.4, ωr =1) preserved the full exploration capability of woa. it possesses the benefit of the inertia weight method, which is also more robust than the other iwwoas and the original woa on multimodal functions. hence, the ldiwr strategy can produce a better balance between exploration and exploitation capabilities for searching solutions and result in an improvement in the convergence speed and optimal solution of the original woa. conflicts of interest the authors declare no conflict of interest. references [1] g. wu, r. mallipeddi, and p. n. suganthan, “ensemble strategies for population-based optimization algorithmsa survey,” swarm and evolutionary computation, vol. 44, pp. 695-711, february 2019. [2] h. hu, y. bai, and t. xu, “improved whale optimization algorithms based on inertia weights and theirs applications,” international journal of circuits, systems and signal processing, vol. 11, pp. 12-26, 2017. [3] j. kenney and r. eberhart “particle swarm optimization,” proc. ieee symp. international conference on neural networks, ieee press, 1995, pp. 1942-1948. [4] m. dorigo and l. m. gambardella, “ant colony system: a cooperative learning approach to the traveling salesman problem,” ieee transactions on evolutionary computation, vol. 1, no. 1, pp. 53-66, april 1997. [5] d. karaboga and b. basturk, “on the performance of artificial bee colony (abc) algorithm,” applied soft computing, vol. 8, no 1, pp. 687-697, february 2008. [6] x. s. yang and s. deb, “engineering optimisation by cuckoo search,” international journal of mathematical modelling and numerical optimisation, vol. 1, no. 4, pp. 330-343, december 2010. [7] w. t. pan, “a new fruit fly optimization algorithm: taking the financial distress model as an example,” knowledge-based systems, vol. 26, pp. 69-74, february 2012. [8] s. mirjalili and a. lewis, “the whale optimization algorithm,” advances in engineering software, vol. 95, pp. 51-67, may 2016. [9] m. m. mafarja and s. mirjalili, “hybrid whale optimization algorithm with simulated annealing for feature selection,” neurocomputing, vol. 260, pp. 302-312, october 2017. [10] m. črepinšek, s. h. liu, and m. mernik, “exploration and exploitation in evolutionary algorithms: a survey,” acm computing surveys (csur), vol. 45, no. 3, pp. 1-33, june 2013. [11] y. shi and r. c. eberhart, “parameter selection in particle swarm optimization,” proc. international conference on evolutionary programming, march 1998, pp. 591-600. [12] g. kaur and s. arora, “chaotic whale optimization algorithm,” journal of computational design and engineering, vol. 5, no. 3, pp. 275-284, january 2018. [13] k. lei, y. qiu, and y. he, “a new adaptive well-chosen inertia weight strategy to automatically harmonize global and local search ability in particle swarm optimization,” proc. ieee symp. 2006 1st international symposium on systems and control in aerospace and astronautics (isscaa), january 2006, pp. 977-980. [14] j. lu, h. hu, and y. bai, “radial basis function neural network based on an improved exponential decreasing inertia weight-particle swarm optimization algorithm for aqi prediction,” abstract and applied analysis, vol. 2014, pp. 1-9, july 2014. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation, vol. 4, no. 4, 2019, pp. 260-268 remote monitoring for the operation status of cnc machine tools based on html5 yu-shun wang, ling-song he * , zhi-qiang gao, jun-feng wang, yang-fan cao school of mechanical science and engineering, huazhong university of science and technology, wuhan, china received 23 april 2019; received in revised form 10 may 2019; accepted 28 july 2019 abstract in order to improve the accuracy of remote monitoring of computer numerical control (cnc) machine tools and reduce the difficulty of monitoring; a remote monitoring method for cnc machine tools based on html5 is proposed in this paper. the core idea of this method is to record external sensor information and internal working condition information in the same time, and then visualize the information in multiple directions. monitoring accuracy is improved through the combined use of internal and external information. in response to the difficult problem of traditional method monitoring; the internal working condition information, external sensor information, 3d model and multimedia information of cnc machine tools are jointly visualized. the 3d model synchronous motion is driven by real-time working condition data. remote low-latency multimedia information transmission is implemented by using cloud live broadcast technology. keywords: cnc machine tool, remote monitoring, html5, 3d model 1. introduction intelligent manufacturing, as the core of industry 4.0, has been the focusing research of global manufacturing field in recent years. as a “mother machine”, the cnc machine tool is the core equipment for intelligent manufacturing. its monitoring of operation status and adaptive control are critical to intelligent manufacturing. as a means of monitoring, the operation status of cnc machine tools and remote monitoring have attracted attention of many researchers. for traditional remote monitoring of cnc machine tools, the monitoring of operation status and fault diagnosis are realized by monitoring the external sensor information (e.g., vibration, sound, current, temperature). however, the sensor signal is the physical quantity measured indirectly. it can only represents the state of the sensor, but not the whole cutting process. in addition, the stability and reliability of the sensor signal can be easily affected by the complex working environment of the cnc machine tool (e.g., electromagnetic, temperature, external noise), which can easily prone to misjudgment [1]. moreover, the data acquisition of the internal operating conditions is ignored, which lead to lack of data (e.g., g code, axis position, feed rate, spindle speed) guidance during condition monitoring and fault diagnosis. therefore, it is not possible to understand the state and the process of processing in a short time, which can easily lead to misjudgment. simultaneously, the degree of data visualization is limited and the interactive operations in data analysis was scanty. in recent years, researchers have investigated in a lot of energy to break through the limits of traditional cnc machine tool monitoring methods. in 2016, avic proposed a four-layer cyber physical system (cps) cnc machining process intelligent platform, which could monitor the machine tool in real time and realize ultra-low delay transmission of processing * corresponding author. e-mail address: helingsong@hust.edu.cn tel.: +86-13554303940 advances in technology innovation, vol. 4, no. 4, 2019, pp. 260-268 261 data and high-fidelity 3d visualization and reproduction [2]. in 2017, eun-young heo et al. proposed a process monitoring system combined with virtual machining that simulates the nc tool path cutting process through virtual machining to predict the static and dynamic characteristics of the operation [3]. the real-time diagnosis was realized by combining the internal working condition data and sensor data of the cnc machine tool with the predicted virtual machining data. in the same year, yi cai et al. merged sensor data and information to build a digital twin virtual machine for network physical manufacturing, which was used for the diagnosis and prediction of cnc machine faults [1]. the above researches all used the working condition information and external sensor information as main data source, which could significantly improve the accuracy of monitoring. however, the cross-platform, visualization, and interaction of the designed system need to be further improved. in view of the advantages of hyper text markup language (html5) can supports the multi-device cross-platform operation, responsive web design, instant update, better interaction, and good support for audio and video[4, 5], a remote monitoring method for cnc machine tools based on html5 is proposed in this paper. this method used the working condition information and external sensor information inside the open cnc system as the main data source, and monitored the machine tool through internal and external information fusion to improve the monitoring accuracy. visualize with html5, make full use of its high-level support for 2d charts, 3d models, audio and video visualization tools, and multi-dimensional visualization of data to reduce the difficulty of monitoring. 2. overall system design the design concept for the overall architecture of the system was the cps model [1-2, 6]. a four-layer architecture based on device awareness layer, network layer, data layer and visualization layer was built, and its system architecture is shown in fig. 1. fig. 1 system architecture the internal working condition data, external sensor data, and multimedia data are acquired through the device sensing layer in this system. the data is collected and parsed, and then the processed data was transmitted to the data layer through the network layer to complete the data service and data storage. the core of the remote monitoring system is the visualization layer. the data processed by the data layer is again transmitted to the visualization layer via the network layer, and the multi-level visual monitoring of the data is completed and the corresponding analysis and processing are performed. it can also provide operators with more real-time, more comprehensive, more decision-making reference value cnc machining process information. this helped the operator to strengthen the monitoring and fault analysis of the real-time operating state of the machine during the entire machining process. advances in technology innovation, vol. 4, no. 4, 2019, pp. 260-268 262 3. system implementation 3.1. data acquisition and transmission the first thing that the remote monitoring system needs is the acquisition and transmission of data. the data collected mainly includes three categories: working condition data acquired from the cnc machine controller, external sensor data acquired from the data acquisition card, and multimedia data acquired from the camera and the microphone, as shown in fig. 2. fig. 2 data acquisition and transmission real-time data acquisition is achieved by transferring data from the network layer to the data layer for data services and data storage. data services are responsible for high-reliability, low-latency transmission of high-capacity, high-real-time data and data pretreatment. the temporary storage and persistent storage of data were mainly completed by data storage. since the transmission control protocol (tcp) has the characteristics of acknowledgment response and retransmission control, both the working condition data and the external sensor data are transmitted based on the tcp to ensure high reliability of data. both temporary storage and persistent storage are used. temporary storage is mainly used for caching and was realized by redis database. since the data in redis is stored in ram, the read/write speed is fast, and the low-latency transmission of the data can be ensured, which could ensure the real-time monitoring. in addition, in order to ensure that the users use the data on demand, the pipeline technology in redis [7] is used to decouple the coupling between the data and the user. different data can be obtained according to different monitoring contents. for persistent storage, the mongodb database is used. the content of the redis database is synchronized to the database by means of master-slave synchronization, which facilitated later playback. the transmission of multimedia data is performed by using web real-time communication (webrtc) technology, and network address translation (nat) penetration is used to ensure low latency of multimedia data transmission. 3.2. 3d model driven by real-time data traditional 3d models are rarely introduced in remote monitoring. even if it has been introduced, it is only used for virtual simulation, and seldom moved synchronously with physical devices. the design of introduce 3d models into the monitoring interface and then use the real-time data to drive 3d model motion had its unique advantages. on the one hand, it added visualization of tool position data and tool trajectory during monitoring, always feedback the current artifact appearance to the monitor, and could view the artifact in all directions by dragging and zooming. on the other hand, by monitoring the processing process, the abnormal position could be quickly located (through the internal working condition data) when signal anomaly. then, the abnormal position could be mapped in real time by 3d model, which could simplify the positioning processing and reduce the difficulty of fault monitoring. in the three-dimensional display of remote monitoring of cnc machine tools, web graphics library (webgl) was responsible for loading and controlling the 3d model, and the browser provided the operating environment [7-8]. in order to advances in technology innovation, vol. 4, no. 4, 2019, pp. 260-268 263 facilitate the control of the 3d model, it has been divided into different components according to the motion mode, and then webgl is used to “virtual assembly” according to its relative position. the mapping relationship between the component and the driving data is established to realize the motion of the assembled cnc machine model through real-time data driving, as shown in fig. 3. fig. 3 mapping of real-time data and 3d model the cnc machine is ready to add the corresponding blank from the blank stock as required. after the tool is calibrated, the data corresponding to g54 in the cnc machine controller is acquired, which was the real workpiece zero point. by establishing a mapping with the zero point of the virtual machine tool workpiece, the mapping relationship between the virtual machine tool and the physical machine tool could be established, thereby realizing the initialization of the data driven model. the mapping equation is as follows: 54 300 v r g x x x    (1) 54 1160 v r g y y y   (2)  54 83.8v r g workpiecez z z h    (3) 0.3 2 / 360 v r s s      (4) where xv, yv, zv, sv is corresponding to the x, y, z axis position and spindle speed in the virtual machine tool respectively; xr, yr, zr, sr is corresponding to the x, y, z axis position and spindle speed in the physical machine tool respectively; xg54, yg54, zg54 is corresponding to the workpiece zero position on the physical machine respectively; hworkpiece is the height of the workpiece in the virtual machine. after the virtual machine model is initialized and the mapping relationship is established, the real-time spindle speed s and the x, y, z axis position could be obtained and used to drive the model motion in real time, as shown in fig. 4. when the initialization of the data-driven model was completed, the machine zero point, tool parameters, and tool initial position could advances in technology innovation, vol. 4, no. 4, 2019, pp. 260-268 264 be obtained. after the table position is partially magnified, the appearance of the workblank could be observed. the real-time data could drive the model to observe the current tool position. after processing, the final rendering of the part could be observed. fig. 4 processing renderings of real-time data driven parts 3.3. low latency video the general architecture of video surveillance in traditional remote monitoring systems are content delivery network (cdn), real time messaging protocol (rtmp) and http live streaming (hls), with a delay of more than 3 seconds, which cannot meet the current requirements for monitoring in real-time. this system drew on the idea of interactive live broadcast, using webrtc technology to control the average delay to about 200ms to realize the real time of video monitoring, as shown in fig. 5. the data transmission is mainly divided into two parts. one was the transmission of signaling, which is realized by exchange the session information, network configuration, media configuration, etc. between the two ends of the peer to peer (p2p) through the cloud platform, and finally the nat penetration is achieved. then a peer-to-peer connection between the device collection end and the kurento server, between the kurento server and the viewer is established [9-11]. another part is the transmission of the video stream. this part directly transmited data through the p2p channel. when the video data is collected by the device, the video is uploaded to the kurento streaming server via the p2p channel, and then the viewer and the kurento terminal are respectively transmitted via the p2p channel. due to the video passes through the p2p channel in both transmissions, there is a small delay and realized low delay of video monitoring. fig. 5 architecture diagram of video monitoring 3.4. remote monitoring prototype system based on html5 based on the above theoretical basis, a remote monitoring prototype system based on html5 is developed, as shown in fig. 6. the monitoring system is mainly used for real-time monitoring and fault diagnosis of cnc machine tools. advances in technology innovation, vol. 4, no. 4, 2019, pp. 260-268 265 the monitoring system visualizes the data in multiple directions, including data display, 2d controls, 3d models, and audio and video. data presentations are intuitive enough to know their specific values, but they did not apply to data that is large in volume or needs to understand trends. the two-dimensional control is realized by html5 canvas technology, which is used to dynamically generate raster-based images [12]. it could be used for a wide range of purposes, such as the dial could directly display the value (more suitable for occasions with slower data changes) and reflectes the trend of change, the chart could display the corresponding waveform directly and switched the different display forms of the data by performing corresponding data processing. the 3d model could be intuitively perceived by the monitor. by driving the 3d model with real data, it is possible to quickly locate the specific machining position and observe the motion, such as linear or circular motion, face milling or shoulder milling. in addition, omnidirectional, multi-angle observations could be performed by rotating the 3d model. through audio and video, you could understand the specific processing environment and reduce the difficulty of monitoring through both visual and auditory aspects. in this design, different visualization methods were adopted for different types of data, which reduced the difficulty of monitoring. fig. 6 interface of remote monitoring prototype system based on html5 4. case study the cutting stability during metal cutting directly affects the machining accuracy and machining efficiency of the machine tool, so real-time monitoring of cutting stability is a key means to evaluate the quality of part machining [13]. the greatest influence on cutting stability was the cutting chatter. cutting chatter refers to a kind of violent self-excited vibration caused by the characteristics of the processing system under the action of no external force during the cutting process[14]. this experiment uses the prototype system to monitor the stability during the process of milling the three-layer pagoda, and analyzes the cause, time and position of the stability problem in time to provide assistance for the rapid diagnosis and analysis. a three-axis vertical milling machine equipped with huazhong cnc system was used in this experiment. the cutting and tool parameters are shown in table 1. during the experiment, the vibration and sound signal waveforms changed significantly at a certain moment, and the phenomenon is shown in fig. 7. by comparing the vibration sound waveforms of fig. 7 (a) and (b), the following characteristics were obtained: from the time domain perspective, the amplitude in the fig. 7 (b) is larger; from the frequency domain perspective, the frequency advances in technology innovation, vol. 4, no. 4, 2019, pp. 260-268 266 domain in the fig. 7 (a) is more discrete. there was no obvious maximum value, and there is a significant peak in the intermediate frequency domain of fig. 7 (b). by listening to the audio data transmitted from the microphone, the sound at time in fig. 7 (b) is sharper, and it is possible to guess that chatter may occur at that moment. at this time, combined with the internal working condition data of g code, axis position data and the visualization of the video and 3d model, it could be found that the running track and the cutting parameters of the workpiece had not changed at this time, and the vibration and sound waveform were proved. the difference is not caused by the change of the running track and the cutting parameters of the workpiece. it could be confirmed that the waveform change at this time is caused by the chatter. table 1 cutter, cutting and workpiece parameters of a three-axis vertical milling machine cutter parameters diameter (mm) 10 helix angle 45° edge number 3 material cemented carbide cutting parameters feed rate (mm) 1/6 axial depth (mm) 7 radial depth (mm) 0.3~0.9 workpiece parameters size (mm) 50*50*50 material aluminum 7075 (a) before chatter (b) chatter fig. 7 cuttig stability experiment vibration frequency domain vibration time domain sound time domain sound frequency domain vibration time domain vibration frequency domain sound time domain sound frequency domain advances in technology innovation, vol. 4, no. 4, 2019, pp. 260-268 267 in addition, when the chatter is successfully monitored, if only the vibration and sound waveforms of the external sensor are passed, it was not possible to quickly locate the position where the chatter occurs and the cause of the chattering. at this time, through the video and data-driven 3d model, the tool trajectory can be visualized in real time, and the chatter position could be quickly located, which played a positive role for subsequent observation and analysis. when the processing is completed, the vibration pattern could be clearly observed by observing the above-mentioned located position under the microscope, as shown in fig. 8. by real-time acquisition and visualization of the internal data of the cnc controller in cnc machine tool, it is possible to obtain internal operating conditions such as spindle rotation speed, feed rate, and override when chattering occurs. this helped to understand the cause of the chatter and provided guidance for suppressing chatter. it is easy to judge the machining abnormality by the external sensor signal in this experiment. at this time, the internal condition data could be combined to determine the type of abnormality as chatter. the accuracy of the monitoring is improved. there are a variety of visualization methods in this monitoring system, which could help to quickly locate abnormal positions. in addition, it helped to grasp the information of cnc machine tool conditions at all times, which reduced the difficulty of subsequent fault analysis and diagnosis. video model partial enlargementmachined part normal surface flutter surface fig. 8 processed workpiece and detailed vibration lines 5. conclusions a method based on html5 for remote monitoring of cnc machine tools is proposed in this paper. this method improved the accuracy of monitoring by combining the internal working condition data of cnc machine tools with external sensor data. a variety of visualization methods, such as data display, 2d controls, 3d models, and audio and video were used to reduce the difficulty of monitoring and analysis. based on this, a prototype system is developed and the corresponding experiments are carried out. the results indicated that the system can significantly improve the accuracy of monitoring and reduce the difficulty of monitoring and analysis. this monitoring system could improve workpiece quality and fault diagnosis accuracy significantly. conflicts of interest the authors declare no conflict of interest. references [1] y. cai, b. starly, p. cohen, and y. s. lee, “sensor data and information fusion to construct digital-twins virtual machine tools for cyber-physical manufacturing,” procedia manufacturing, vol. 10, pp. 1031-1042, 2017. [2] x. h. li and w. y. li, “the research on intelligent monitoring technology of nc machining process,” procedia cirp, vol. 56, pp. 556-560, 2016. advances in technology innovation, vol. 4, no. 4, 2019, pp. 260-268 268 [3] e. y. heo, h. lee, c. s. lee, d. w. kim, and d. y. lee, “process monitoring technology based on virtual machining,” procedia manufacturing, vol. 11, pp. 982-988, 2017. [4] z. lei, w. hu, and h. zhou, “deploying web-based control laboratory using html5,” 2016 13th international conference on remote engineering and virtual instrumentation (rev), madrid, pp. 69-73, 2016. [5] y. sheng, j. huang, f. zhang, y. an, and p. zhong, “a virtual laboratory based on html5,” 2016 11th international conference on computer science & education (iccse), nagoya, pp. 299-302, 2016. [6] j. chen, j. yang, h. zhou, h. xiang, z. zhu, y. li, c. h. lee, and g. xu, “cps modeling of cnc machine tool work processes using an instruction-domain based approach,” engineering, vol. 1, pp. 247-260, 2015. [7] j. y. xia, b. j. xiao, d. li, and k. r. wang, “interactive webgl-based 3d visualizations for east experiment,” fusion engineering and design, vol. 112, pp. 946-951, 2016. [8] l. lianzhong, z. kun, and x. yang, “a cloud-based framework for robot simulation using webgl,” 2015 sixth international conference on intelligent systems design and engineering applications (isdea), guiyang, pp. 5-8, 2015. [9] a. s. rosas and j. l. a. martínez, “videoconference system based on webrtc with access to the pstn,” electronic notes in theoretical computer science, vol. 329, pp. 105-121, 2016. [10] sredojev, d. samardzija, and d. posarac, “webrtc technology overview and signaling solution design and implementation,” in 2015 38th international convention on information and communication technology, electronics and microelectronics (mipro), 25-29 may 2015, piscataway, nj, usa, 2015. [11] k. i. z. apu, n. mahmud, f. hasan, and s. h. sagar, “p2p video conferencing system based on webrtc,” 2017 international conference on electrical, computer and communication engineering (ecce), cox's bazar, pp. 557-561, 2017. [12] s. dumke, h. riemann, t. bluhm, r. daher, m. grahl, m. grün, a. holtz, j. krom, g. kühner, h. laqua, m. lewerentz, a. spring, and a. werner, “next generation web based live data monitoring for w7-x,” fusion engineering and design, vol. 129, pp. 16-23, 2018. [13] z. yongliang and l. zhiyuan, “stability and online monitoring of cutting chatter: a review,” applied mechanics and materials, vol. 121-126, pp. 377-81, 2011. [14] k. j. li, y. x. liu, and z. zhao, “research in multiple factors vibration controlling of cnc milling machine,” 2011 fourth international conference on intelligent computation technology and automation, shenzhen, guangdong, pp. 472-475, 2011. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation , vol. 2, no. 3, 2016, pp. 73 84 73 copyright © taeti optimization of ducted propeller design for the rov (remotely operated vehicle) using cfd aldias bahatmaka 1,* , dong-joon kim 2 , deddy chrismianto 3 1 interdiciplinary program of marine design convergence, pukyong national university, busan, korea. 2 department of naval architecture and marine systems engineering , pukyong national university, busan, korea. 3 department of naval architecture, diponegoro university, semarang, indonesia. received 09 march 2016; received in revised form 19 june 2016; accept ed 21 june 2016 abstract the development of underwater robot technology is growing rapidly. for reaching the best performance, it is important that the inn ovation on rov should be focused on the thruster and propeller. in this research, the ducted propeller thruster is used while three types of shuskhin nozzle are selected. the design is compared in a c cordance with the thruster that has been made as the propulsion device of underwater robots. each type of the thruster model indicates different force and torque. for the analysis, each model is built in co mputer aided design (rh inoceros) program pac kages and comput ational fluid dyna mics (cfd) to find the most optima l model wh ich can produce the h ighest thrust. among the entire model, the kaplan series (ka5-75) with the type c of nozzle has the highest thrust which is 2.53 n or 25.24% of e xt ra thrust. for the optimization o f thrust, genetic a lgorithms (ga) is used. the ga can search for parameters in la rge mult i-d imensional design space. thus, the principle can be applied for determining the initia l propelle r that produces optimu m thrust of rov. the ga has successfully shown able to obtain an optima l set p ara meters for propeller characteristics with the best performance. ke ywor ds : re motely operated vehic le (rov), ducted propeller, cfd, genetic a lgorithms nomenclature ae propeller expanded area a0 propeller disk area ctn regression coefficient of thrust coefficient cqn regression coefficient of torque coefficient d diameter propeller j advance coefficient kt thrust coefficient kq torque coefficient n propeller blade pitch p/d pitch diameter ratio q torque (nm) sn exponent of j t thrust tn exponent of p/d un exponent of ae/a0 vn exponent of z z number of blades ρ density of water 1. introduction re motely operated vehic le (rov) is an instrument formed min i-sized submission vehicle. rov is usually used to explore underwater photography, military operation and underwater pipeline repairing. rov is used for activit ies small cave. rov is designed to have abilities of seabed rescue operations and repairing o f seabed objects from the surface [1]. un d e rwa t e r ro b o t is d es ig n ed a nd ma n u fa c t u re d abso lu t e ly requ ire s ma n y supporting components to improve the operation to perform a variety missions. thruster is one component that has function as locomotor of underwater robot to maneuver horizontally when it moves forward and bac kwa rd and a lso to maneuver vertically to moves up and down [2]. * corresponding aut hor, email: bahat maka.aldias@gmail.com advances in technology innovation , vol. 2, no. 3, 2016, pp. 73 84 74 copyright © taeti the first proposal to use a screw propeller appears to have been made in england by ho oke in 1680, and its first actual use is generally attributed to colonel stevens in a stream-driven boat at new york in 1804. in 1828 a vessel 18 m (60 feet) long was successfully propeller by a screw propelle r designed by ressel. the screw propeller has reigned supreme in the rea lm of ma rine propulsion. it has proved e xtraord inarily adaptable in meeting the incessant quest for propellers to deliver mo re and more thrust under increasingly arduous conditions [3]. the genetic a lgorith ms is one of method that has tool for optimization for many difficult optimization proble m with mult iple object ives. ga methods were shown to be able to replace the traditional computation method and design charts. in this case, for determining the in itia l propeller that optimu m for rov thruster is difficult. for selecting the init ial propelle r, used optim tools in genetic algorithms, the result will be found directly and efficiently. however, the traditional design charts method and previous ga methods were li mited in considering or ma ximu m hydrodynamic aspect alone, such as hydrodynamic efficiency, and the thrust coefficient, etc. hence, the goals of the present research are: to determine the optimu m value o f thrust for rov thruster using 3 para meters are pitch dia meter rat io, e xpanded blade area ratio, and rotational speed of propelle r. applying ga methods to propeller design with consideration of thrust for reach ing the best performance of propeller. 2. method the co mputational flu id dyna mics (cfd) [4], is one branch of fluid mechanics that uses numerical methods and algorithms to solve and analyse problems related to flu id flo w the pu rpose of cfd is to predict accurately on fluid flow, heat transfer, and chemica l reactions in comple x systems, which involve one or a ll the above phenomena. 2.1. rov design in this research, rov project in fig. 1 has been designed and which has dimension as shown as the table 1. table 1 principle dimensions of rov item unit length 601.87 mm beam 409.20 mm height 290.00 mm mass (on air) 13.70 mm weight 5.00 n fig. 1 rov project design 2.2. propeller design the propeller d imension and geometry d esign are listed in the table 2. table 2 principle dimensions of propeller item unit diameter 130.00 mm pitch 78.00 mm amount of blade 5 blades expanded bar 0.75 rake of angle (b5-75) 10.00 degree rake of angle (ka 5-75) 15.00 degree rotational speed 300.00 rpm density of water 1.025 kg/m3 in this case, the propeller dimension was chosen by the product sold in the market. and using this dimension, propeller was designing in cad model. fo r the design used 2 types of design, b-se ries (b5-75) as shown in fig. 2 and kaplan-series (ka5-75) as shown in fig. 3. advances in technology innovation , vol. 2, no. 3, 2016, pp. 73 84 75 copyright © taeti fig. 2 b-series (b5-75) fig. 3 kaplan-series (b5-75) 2.3. numerical method and bboundary conditions in this steps, for the numerical on analysis was using co mputational fluid dyna mics method. several steps for calculat ions such as: geometry, mesh, setup, solution, and result. boundary condition was calcu lating for the open water propeller [5]. the boundary condition can be shown in fig. 4. fig. 4 boundary condition of open water propeller 2.4. validation of thruster in this research to validate the result of the test model, used the software test result that a lre a dy e xis t a nd was con du ct ed b y mulyowidodo, k. et al in bandung technology institute indonesia. va lidation of underwater thruster for shrimp rov-it b is used to determine the e xact boundary condition for using on the boundary condition when analyzing 3 models for rov thruster using cfd-based software. re ference is taken fro m the validated model for the testing thruster. used propeller type of ka5-75 series and based on the theory of wageningen [6], the open water propelle r characteristics conventionally were presented in form of the thrust and torque coeffic ient k t and kq in term of the advance coefficient j, where 2 4t t k n d  (1) 2 5q t k n d  (2) a v j nd  (3) the characteristics of the ship’s propeller for open water test condition are as represented in kt-kq-j chart. type of each propelle r is having the characteristic curves of different p erformance. it can be shown in fig. 5. fig. 5 kt-kq-j chart 2.5. geometry design of nozzle for the calcu lation of nozzle , shushkin nozzle has been chosen which developed (prof.dr.-ing. h. heuser, 1982). the nozzle design is shown in fig. 6. advances in technology innovation , vol. 2, no. 3, 2016, pp. 73 84 76 copyright © taeti fig. 6 pressure process and flow contraction at the nozzle propeller compared the size of kort nozzle tipe a: ld/dp =0,75; di / dp = 1.015; limits: 20mm <(di – dp) < 60 mm; da/di = 1,25; la/ ld = 0,53; lp/ ld = 0,27, lv/ ld = 0,40; lh/ld = 0:33 the size of kort nozzle tipe b: ld/dp = 0,75; di/ dp = 1,015; limits: 20mm < (di –dp) < 60 mm; da/di = 1,25; dk/ di = 1,02; dr/di = 1,035; la/ ld = 0,32; lp/ld = 0,25, lv/ ld = 0,425; lh/ ld = 0,325; lk/lh = 0,925 the size of kort nozzle tipe c : ld/dp = 0,75; di/ dp = 1,015; limits: 20mm < (di – dp) < 60 mm; da/di = 1:20; dk/ di = 1,015; dr/di = 1,030; la=ld = 0,50; lp/ld = 0,50; lv/ ld = 0,40; lh/ ld = 0,35; lk/lh = 0,880 2.6. preliminary propeller design the preliminary propeller design proble m is described in detail in the princ iples of nava l architecture [3]. here is the basic design or initia l design of propelle r with the principa l dimension as: (dependent variables) diameter : 130.00 mm amount of blades (z) : 5 blades rake of angle : 15.00 degree material : mn.bronze (2) density of water : 1.025 kg/m3 (independent variables) p/d : 0.60; 0.65; 0.70 ae/a0 : 0.75; 0.80; 0.85 rotational speed (n) : 5 rps; 10 rps; 15 rps 2.7. performance of computation selecting fro m a propeller series is a simp le method to design a propeller. a mong the series propeller is one of the most often used and studied. and selection of the blade was improved later. the thrust and torque coefficient can be expressed as the table 3. table 3 regression coefficients and exponents of kt-kq n c tn s n tn u n vn n c qn s n tn u n vn 1 0.00880496 0 0 0 0 1 0.00379368 0 0 0 0 2 −0.204554 1 0 0 0 2 0.00886523 2 0 0 0 3 0.166351 0 1 0 0 3 −0.032241 1 1 0 0 4 0.158114 0 2 0 0 4 0.00344778 0 2 0 0 5 −0.147581 2 0 1 0 5 −0.0408811 0 1 1 0 6 −0.481497 1 1 1 0 6 −0.108009 1 1 1 0 7 0.415437 0 2 1 0 7 −0.0885381 2 1 1 0 8 0.0144043 0 0 0 1 8 0.188561 0 2 1 0 9 −0.0530054 2 0 0 1 9 −0.00370871 1 0 0 1 10 0.0143481 0 1 0 1 10 0.00513696 0 1 0 1 11 0.0606826 1 1 0 1 11 0.0209449 1 1 0 1 12 −0.0125894 0 0 1 1 12 0.00474319 2 1 0 1 13 0.0109689 1 0 1 1 13 −0.00723408 2 0 1 1 14 −0.133698 0 3 0 0 14 0.00438388 1 1 1 1 15 0.00638407 0 6 0 0 15 −0.0269403 0 2 1 1 16 −0.00132718 2 6 0 0 16 0.0558082 3 0 1 0 17 0.168496 3 0 1 0 17 0.0161886 0 3 1 0 18 −0.0507214 0 0 2 0 18 0.00318086 1 3 1 0 19 0.0854559 2 0 2 0 19 0.015896 0 0 2 0 20 −0.0504475 3 0 2 0 20 0.0471729 1 0 2 0 21 0.010465 1 6 2 0 21 0.0196283 3 0 2 0 22 −0.00648272 2 6 2 0 22 −0.0502782 0 1 2 0 advances in technology innovation , vol. 2, no. 3, 2016, pp. 73 84 77 copyright © taeti table 3 regression coefficients and exponents of kt-kq (continued) n c tn s n tn u n vn n c qn s n tn u n vn 23 −0.00841728 0 3 0 1 23 −0.030055 3 1 2 0 24 0.0168424 1 3 0 1 24 0.0417122 2 2 2 0 25 −0.00102296 3 3 0 1 25 −0.0397722 0 3 2 0 26 −0.0317791 0 3 1 1 26 −0.00350024 0 6 2 0 27 0.018604 1 0 2 1 27 −0.0106854 3 0 0 1 28 −0.00410798 0 2 2 1 28 0.00110903 3 3 0 1 29 −0.000606848 0 0 0 2 29 −0.000313912 0 6 0 1 30 −0.0049819 1 0 0 2 30 0.0035985 3 0 1 1 31 0.0025983 2 0 0 2 31 −0.00142121 0 6 1 1 32 −0.000560528 3 0 0 2 32 −0.00383637 1 0 2 1 33 −0.00163652 1 2 0 2 33 0.0126803 0 2 2 1 34 −0.000328787 1 6 0 2 34 −0.00318278 2 3 2 1 35 0.000116502 2 6 0 2 35 0.00334268 0 6 2 1 36 0.000690904 0 0 1 2 36 −0.00183491 1 1 0 2 37 0.00421749 0 3 1 2 37 0.000112451 3 2 0 2 38 5.65229e−05 3 6 1 2 38 −2.97228e−05 3 6 0 2 39 −0.00146564 0 3 2 2 39 0.000269551 1 0 1 2 40 0.00083265 2 0 1 2 41 0.00155334 0 2 1 2 42 0.000302683 0 6 1 2 43 −0.0001843 0 0 2 2 44 −0.000425399 0 3 2 2 45 8.69243e−05 3 3 2 2 46 −0.0004659 0 6 2 2 47 5.54194e−05 1 6 2 2 as the functions of the blade number, blade area ratio, p itch ratio and advance coefficient [4]: 39 1 0 untn sn vne t tn n ap k c j z d a             (4) 47 1 0 untn sn vne q qn n ap k c j z d a             (5) where ctn and cqn are the regression coefficients of the thrust and torque coefffic ients, repectively; sn, tn, un, and vn are the exponents of j, p/d, ae/a0, and z, respectively. 2.8. genetic algorithms (ga) the basic theory, genetic algor ithms a re search algorithms based on the mechanics of natural selection and natural genetics. they combine survival of the fittest among string s tructure with structured yet randomized informat ion e xchange to inform a structure algorithms with some of the innovative flair of human search. in every generation, a new set of artific ia l creatures (strings) is created using bits and pieces of the fittest of the old man; an occasional new part is tried for good measure. while randomized, genetic algorithms are no simple random walk. they effic iently e xp loit historica l informat ion to speculate on new search points with e xpected imp roved performance. genetic algorith ms have been developed by john ho lland [7]. based on the theory, the research was conducted with the set of para meters dependent and independent variables. for the first steps were identifying variab les and functions, the prope ller of re mote ly operated vehicle (rov) , there are three variable that we could change randomly as like as p/d (p itch ratio ), ae/a0 (e xpanded blade area rat io), and n (rotational speed). using the three variables, we would get the value of kt and kq. thus, the thrust (t) and torque (q) can be expressed as: function:  2 4tt k n d (6)  2 5qq k n d (7) the fitness function we can select the eqs. (6) and (7), then used the regression table (3) to produce the thrust and torque coefficients. advances in technology innovation , vol. 2, no. 3, 2016, pp. 73 84 78 copyright © taeti fitness function: 39 1 0 untn sn vne t tn n ap k c j z d a             (8) 47 1 0 untn sn vne q qn n ap k c j z d a             (9) the constraint function, used three variables (p/d, ae/a0, n) can be created as: constraint function: 0.60 0.70 p d   (10) 5 15n  (11) 2.9. optimization for the process of optimization shown in fig. 7, used optimization tools that supported by matlab program for the solution of genetic algorith ms method. in this case, ga apply to know how is the optimu m thrust for the in itia l propeller using the parameters and all parame ters setup for ga is shown in table 4. table 4 setting up the ga parameters ga parameters population size = 20 number of generations = 80 selection: stochastic uniform crossover rate = 0.8 crossover function: scattered mutation rate = 0.2 fig. 7 flowchart of genetic algorithms advances in technology innovation , vol. 2, no. 3, 2016, pp. 73 84 79 copyright © taeti 2.10. cfd analysis type a nozzle fig. 8 meshing of type a for b-series (b5-75) fig. 9 meshing of type a for kaplan-series (ka5-75) type b nozzle fig. 10 meshing of type b for b-series (b5-75) fig. 11 meshing of type b for kaplan-se ries (ka5-75) the process of analysis for the nozzle design has been conducted. there were so me different results for the meshing process of all nozzle’s type. for the type a for b-series (b5-75) shown in fig. 8, it has 845,769 nodes and kaplan-se ries (k5-75) shown in fig.9 has 702,840 nodes. the type b for b-series (b5-75) shown in fig.10, it has 1,146,817 nodes and kap lan-series (k5-75) shown in fig.11 has 1,007,934 nodes. the type c for b-se ries (b5-75) shown in fig.12, it has 1,343,789 nodes and kaplan-series (k5-75) shown in fig.13 has 1,206,703 nodes. type c nozzle fig. 12 meshing of type c for b-series (b5-75) fig. 13 meshing of type c for kaplan-se ries (ka5-75) streamline simulation for the simulat ion of strea mline, the input for a ll types were same, and the rotational speed was 300rp m. the results of streamline fro m type a nozzle for b-series (b5-75) shown in fig.14, type a nozzle for kap lan-series (ka 5-75) shown in fig.15, type b nozzle for b-series (b5-75) shown in fig.16, type b nozzle fo r kaplan-series (ka 5-75) shown in fig.17, type c nozzle for b-series (b5-75) shown in fig.18, type a nozzle for kaplan-series (ka5-75) shown in fig. 19. advances in technology innovation , vol. 2, no. 3, 2016, pp. 73 84 80 copyright © taeti type a nozzle fig. 14 strea mline of type a for b-se ries (b5-75) fig. 15 strea mline of type a for kaplan-se ries (ka5-75) type b nozzle fig. 16 strea mline of type b for b-se ries (b5-75) fig. 17 strea mline of type b for kaplan-se ries (ka5-75) type c nozzle fig. 18 strea mline of type c for b-se ries (b5-75) fig. 19 strea mline of type c for kaplan-se ries (ka5-75) fro m a ll the figures, it can be seen that the type c of no zzle was the highest of thrust than type a or type b. pressure contour type a nozzle fig. 20 pressure contour of type a for b-se ries (b5-75) fig. 21 pressure contour of type a for kaplan-series (ka5-75) advances in technology innovation , vol. 2, no. 3, 2016, pp. 73 84 81 copyright © taeti type b nozzle fig. 22 pressure contour of type b for b-se ries (b5-75) fig. 23 pressure contour of type b for kaplan-series (ka5-75) type c nozzle fig. 24 pressure contour of type c for b-se ries (b5-75) fig. 25 pressure contour of type c for kaplan-series (ka5-75) this process was indicating the pressure distribution on the blade area. the results of pressure contour from type a nozzle for b-series (b5-75) shown in fig. 20, type a nozzle for kaplan-series (ka 5-75) shown in fig. 21, type b nozzle for b-series (b5-75) shown in fig. 22, type b nozzle for kaplan-series (ka 5-75) shown in fig. 23, type c nozzle for b-series (b5-75) shown in fig. 24, type a nozzle for kaplan-series (ka5-75) shown in fig. 25. 3. results and discussion fro m the analysis of cfd method, it can be seen on the table 5. for the changes of the force as clearly shown in fig. 26, and the changes of the torque shown in fig. 27. table 5 results of force (t) and torque (q) type of models force (t) (n) torque (q) (nm) b-series (b5-75) propeller without nozzle 1.67 0.016 nozzle a 2.06 0.020 nozzle b 2.12 0.021 nozzle c 2.26 0.023 kaplan (ka5-75) propeller without nozzle 2.02 0.020 nozzle a 2.34 0.022 nozzle b 2.50 0.023 nozzle c 2.53 0.024 fig. 26 chart of force (t) analysis on 300rpm advances in technology innovation , vol. 2, no. 3, 2016, pp. 73 84 82 copyright © taeti fig. 27 chart of torque (q) analysis on 300rpm in this paper, we have described the specification and the design concepts of ducted propeller as the thruster of rov (re motely operated veh icle ). the re we re several designs of the ducted propeller. total o f models were 6 models , b-series (b5-75) and kap lan-se ries (ka5-75) with nozzle’s type a, b, and c. based on the simu lations which we re co nducted, can be concluded that the highest of force (t) was model of kaplan-series (ka5-75) with type c of nozzle. the mode l could produce 2.53 n, or 25.24% of e xt ra thrust. the lowest of torque (q) was model of b-series (b5-75) with type a of nozzle . thus the best model will be used for rov thruster is model of kaplan -se ries (ka5-75) with type c of nozzle. in other hand, the model a lso was conducted for analysis using other software, was used star ccm+ for checking the value of the force (t) and the torque (q). the geo metry shown in fig. 28, the mesh scene as shown in fig. 29, and the results could be seen in fig. 30 for the strea mline simu lation and for the pressure contour shown in fig. 31. geometry fig. 28 geometry of ducted propeller in star ccm+ generating mesh fig. 29 mesh scene in star ccm+ streamline simulation fig. 30 streamline simulation in star ccm+ pressure contour fig. 31 pressure contour in star ccm+ based on the simu lation results both of ansys cfx and star ccm+ is almost same value, the difference of the force (thrust) is around 5%. it means, the analysis of ducted propeller is enough satisfied and accurate. it can be seen on the table 6. table 6 comparison results results ansyscfx star ccm+ force (t) 2.53 2.66 torque (q) 0.024 0.033 advances in technology innovation , vol. 2, no. 3, 2016, pp. 73 84 83 copyright © taeti optimization in the previous chapter has been obtained the model that will be used for the rov (re motely operated vehicle ). the model is the kaplan-series with type c nozzle. further, the optimization analysis can be applied to get the optimal of thrust for this case. with the fitness function and constraints fro m the table 7, the results can be e xported by genetic algorithms as shown in fig. 32. table 7 output of optimum thrust iteration p/d ae/a0 n(rev/s) 51 0.60 0.79 5.00 at the 51 iterat ion using ga, the thrust will be optimu m when the value of p/d (pitch ratio) is 0.60, a e/a0 is 0.79 and the value of rotational speed is 5 rev/s. based on the results from the table 8, the thrust is increasing 27.43% fro m 2.530 n to 3.224 n. and it means the optimization of the thrust for the ducted propeller of rov is reached and satisfied. fig. 32 fitness value through the generations final results table 8 comparison the final results result original new differences force (n) 2.530 3.224 +27.43% torque (nm) 0.024 0.032 +33.33% 4. conclusions in this paper, about a study on development of the propuls ion device for the rov (re motely operated veh icle ). the cfd method has been demonstrated to be more effective as the pro ble m solver fo r determining the optimu m thrust of rov. the nozzle can produce the e xtra thrust for the propeller. cfd model shows these results . the comparison of propeller types such as b-series (b5-75) and kaplan-series (ka5-75) shows that kaplan-series (ka 5-75) has stronger thrust than b-series (b5-75) and the strea mline simulation shows these results also. the highest thrust (t) is fro m the mode l kaplan-series (k5-75) with type c of nozzle . the model can produce 2.53 n or 25.24% of extra thrust. the ga is applied for optimization design of the initial p ropelle r dimension and the results show that it is good for this kind of the proble ms . the optimization method ga is easy to use as optimu m solution in mu lti-dimensional using three variables such as p/d (pitch rat io), a e/a0 (e xpanded blade area ratio), and n (rotational speed) are affecting to the result of the thrust. fro m the results of optimizat ion, the thrust is optimu m when the value of p/d is 0.60, a e/a0 is 0.79 and the value of rotational speed is 5 rev/s or 300 rpm. in this research, for the optimizat ion of the ducted propeller for rov has been proposed and provides the solution to the problem of rov thruster. therefore , cfd method and ga opt imization can be emp loyed due to its ability to solve the problem and its performance mentioned above especially for the best performance of ducted propeller. acknowledgement the authors would like to thanks fo r the support of the bk21 p lus madec hu man re source development group in 2016, pukyong national university, without their help, this paper could not be submitted. references [1] d. christ, roberto, l. wern li sr, robert, the rov manual,user guide for observation, united kingdom: burlington, 2007. [2] s. a. sharkh, m. r. harris, r. m. crowder, p. h. chappell, r. l. stoll, and j. k. sykulski, “design considerations for electric drives for the thrusters of unmanned underwater vehicles ,” the 6th european conference on advances in technology innovation , vol. 2, no. 3, 2016, pp. 73 84 84 copyright © taeti peer electronics and application, vol. 3, pp. 799-804, 1995. [3] v. manen, j. d., and p. v. ossanen, principles of naval architecture, vo lu me ii: resistance, propulsion, and vibration, 2nd ed., new jersey: society of naval architects and marine engineers, 1988. [4] g. kuiper, “new developments and propeller design,” journal of hydrodynamics, ser. b, vol. 22, no. 5, pp. 7-16, 2010. [5] parra and carlos, “nu merica l investigation of hydrodynamic performances of ma rine propeller,” master thesis, galati un iversity, 2013. [6] m. m. bernitsas, d. ray, and p. kinley, “kt, kq, and effic iency curves for the w ag en ing en b-se rie s p rop e lle rs ,” michigan univ., dept. of naval arch itecture and marine eng., usa, 1981. [7] goldberg and e. david, “genetic algorithms in search, optimizat ion, and machine learning,” alabama univ., usa, 1953.  advances in technology innovation, vol. 2, no. 4, 2017, pp. 99 104 99 harvesting atmospheric ions using surface electromagnetic wave technologies louis wai yip liu * , qingteng zhang, yifan chen department of electrical and electronics engineering, southern university of science and technology, xili, nanshan, shenzhen, china. received 19 november 2016; received in revised form 08 march 2017; accept ed 08 march 2017 abstract for the first time, this paper discloses the use of flowing water for capturing atmospheric ions into a dc electricity. the proposed methodology can be employed to neutralize the positively charged pollutants in air, which are believed to be harmful to our health. methodology: atmospheric ions can be collected by a negatively charged antenna which comprises a dielectric layer sandwiched between a top aluminiu m layer and a bottom lead plate. the top aluminiu m layer is used to collect the ambient protons, whilst the bottom lead plate is negatively charged by a negative static electricity extracted from flowing water. the voltage has been measured between the top aluminiu m layer and the bottom lead plate with and without any sunlight. results: without any uv light or other electromagnetic disturbance, the generated voltage has rapidly increased from 200 mv to 480 mv within 5 seconds if the bottom lead plate is co nnected to the negative ion source. without the neg ative ion source, however, the output voltage fell to around 10 mv and any significant voltage rise can be observed even in the presence of an uv light. conclusions: capturing atmo spheric ions is technically feasible. measured results suggest that, when us ed in conjunction with a negative ion source, the proposed device can harvest atmospheric ions without any uv light. ke ywor ds : cosmic rays , surface plasmon polarition, surface plasmonic resonance, kelvin water dropper, kelvin thunder 1. introduction solar energy has been proposed as an alternative to fuel based energy sources which release tons of pollutants to our atmosphere on a daily basis. however, conventional solar cells are usually inactive before 6 am or after 6 p m. there have been many suggestions to overcome this issue. one suggestion is to use molten salts to store the sun's heat du ring the day so that power can be obtained by cooling during the evening time [1, 2]. french national center for scientific research and the university of tokyo has suggested mounting a solar panel in a high-fly ing balloon to tap solar energy above the clouds [2]. unfortunately, neither of these solutions can be carried easily with household facilities . what is little known is an energy source known as cosmic rays. cosmic rays which come from e xploding stars from light years away can be harvested during the evening time. cosmic rays are ionizing particles of cosmic origin traveling at near the speed of light [3-5]. the majority of cosmic particles are the nuclei of atoms, mostly protons. as these cosmic particles penetrate into the ionosphere, they induce nuclear-electromagnetic cascade ionizing air mo lecules all the way down to the surface of the earth. the earth is an entity carrying negative charges, which repels the negative ions on the surface of earth. as a result of the earth's intrinsic negativity, air pollution and the cosmic ray induced ionization, the density of these positive ions in the lower atmosphere becomes 10-20% higher than the negative ions. the surface of the earth, these positive ions ma inly include positively charged dusts, bacteria, pollen, chemicals and fumes. the buildup of these excessive pos itive ions can be removed and turned into energy when they are attracted to an artificial source of negative charges. if these negative charges are freely abundant, then the amount of energy generated will be positively associated with the density of positive ions in the regions where atmospheric ions is to be harvested. the most basic atmospheric ion collector was invented by nikola tesla in 1901 [6-7]. in his atmospheric ion collector, an antenna is used to collect cosmic particles or positive ions. the antenna comprises a sheet of optically smooth metal co mpletely encapsulated in a dielectric. the antenna serves two purposes: 1) it collects the ultraviolet radiations in the form of surface electromagnetic waves; and 2) it attracts atmospheric ions. the captured radiations or a positive ion increases the voltage of the antenna with respect to the service earth connection. the condenser connected between the antenna and the earth has a high capacitance, thereby creating a high electrostatic suction to attract ions at high altitudes. when the voltage is high enough, it charges up a capacitor or a battery. some researchers have replicated the tesla's experiment with limited success [8-13]. however, this kind of atmospheric ion collectors do not work well when placed at low altitudes, where the density of the atmospheric ions is simply too low. in the field of astrophysics, cosmic radiations are usually detected using a scintillator [14]. for example, an array of scintillator surface detectors samples the footprint of the cosmic-ray shower when it reaches the earth's surface. a fluorescence telescope is used to measure the scintillation light generated as the shower passes through the atmosphere. while this technique is highly useful for the purposes of detection of cosmic rays, it may not be readily useful for neutralization of positively charged pollutants at low altitudes. for the first time, this paper proposes a novel collector which uses flowing water as an energy source to capture positively atmospheric ions on the surface of the earth. *corresponding aut hor. email: liaowy@sust c.edu.cn advances in technology innovation, vol. 2, no. 4, 2017, pp. 99 104 100 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti 2. proposed antenna for collecting positively charged air ions to harvest atmospheric ions, it is important to understand which frequency bands are available for direct energy harvesting. these frequency bands will determine what metal is to be used for the antenna (i.e. the part for capturing cosmic radiations). fig. 1 shows a spectrum of radiations from outer-space. it can be seen from fig. 1 that ultraviolet radiations are the only frequency band at which ionizing energy can be conveniently captured at the ground level [15]. these ultraviolet radiations are mainly of solar origin. a small amount of atmospheric ions come from these ultraviolet radiations. the oxygen and ozone molecules in the atmosphere absorb most of the cosmic radiations from outer-space, including the x-rays and gamma-rays from the galaxy. there is no direct access to the primary cosmic particles on the surface of the earth. however, it is incorrect to assume that atmospheric ions cannot be harvested on the surface of the earth. the nuclei, the x-rays and the gamma-rays from the galaxy constantly induce cosmic cascades in the ionosphere, releasing a large amount of energetic ions that are far more penetrating. fig. 1 available spectrum for energy harvesting [15] to maximize the energy which can be harvested, therefore, the antenna should be designed in such a way that it can capture not only ultraviolet radiations from the sun, but also the free ions that are moving randomly in the atmosphere. these two requirements can be easily fulfilled by following the following design strategies: (1) the inco ming uv light must be captured in the form of surface e lectro magnetic waves, which is concentrated on the surface. either alu minu m or silver should be used as the metal for capturing ultravio let lights because these two metals naturally have a surface plasmonic resonance frequency right at the ultraviolet band; and (2) on the other hand, the metal for the antenna, wh ich can be aluminu m or silver, should be biased to certain voltage so that it naturally attracts the nearby atmospheric protons. this can be done by connecting the metal indirectly to a source of negative charges. (3) the induced current must be directed fro m the po sitive end, wh ich is the antenna in this case, towards the negative end, which is the source of negative charges. this can be done with a h igh reverse breakdown voltage. (4) the surface area of the antenna should be as large as possible in ways to ma ximize the chance of ionization. (5) all the metal used in the co llector should not be e xposed and should be dielectrica lly coated in a way to minimize the charge leakage. fig. 2 antenna for capturing atmospheric positive ions and cosmic particles when the bottom lead plate is negatively charged by a negative ion source, the top aluminum layer will be electrostatically induced to a positive potential. if the top aluminum layer is fully covered by the dielectric coating, then the surface of the top dielectric coating will become negative. the atmospheric positive ions will be attracted to the surface of the top dielectric coating and the top aluminum layer will be fu rther charged positively by electrostatic induction. when the potential difference between the top aluminum layer and the bottom lead plate exceeds the threshold voltage of the diode, a current will form. in addition to the atmospheric positive ions, some other cosmic-ray induced particles or ultraviolet radiations from the sun also play a role in ionization. the fast-moving particles in these radiations will be slowed down significantly as they reach the top dielectric layer. the cosmic radiations are ionizing. when the atmospheric ions reach the aluminum layer, ionization is expected to occur. the energy that is created due to this ionization process is mainly electromagnetic in nature. this electromagnetic energy can either be trapped as a result of the intrinsic localized surface plasmon resonance in aluminum or be transmitted as a surface electromagnetic wave, also known as surface plasmon polariton. surface electromagnetic wave differs from the conventional electromagnetic wave in that the former propagates along the interface between two different media whilst the latter propagates through the 3d dimensional space. unlike the conventional electromagnetic wave, which is radiating in all directions, surface electromagnetic waves tend to be transmitted in a more concentrated manner. the efficiency of this surface plasmon polaritan can be optimized by making sure the antenna reaches the surface plasmonic resonance. the surface plasmon resonance occurs when the leaky modes are minimum and the propagation modes are maximum, according to [16]. in the structure as shown in fig. 2, a complete surface plasmon resonance condiadvances in technology innovation, vol. 2, no. 4, 2017, pp. 99 104 101 cop y right © taeti tion will occur mainly at the interface between the top dielectric layer and the aluminum layer (also referred as mi structure) and, to some extent, in the high-k material between the aluminum layer and the bottom lead plate (also referred as mim structure). during surface plasmon resonance, most of the energy will concentrate in the horizontal direction in the forms of propagation modes and there should be very little leakage in the form of radiations in the vertical direction. 2.1. surface plasmon resonance for the mi structure to attain surface plasmon resonance condition in the mi structure (that is, the interface between the top dielectric layer and the aluminum layer), the criterion shown in eq. (1) and (2) must be met, according to [16]: εa aa+ εb ab= 0 (1) where εa and εb are respectively the dielectric constants of the dielectric coating and the aluminu m layer. aa and ab are respectively the thicknesses of the dielectric coating and the aluminum layer. the dielectric constant of aluminu m is obtained according to the drude's model. at uv-c band frequencies, the dielectric constant of aluminum is about -12. dielectric constant of the top dielectric layer is in the neighborhood of 3. the alu minu m thickness b is about 100 microns. according to eq. (1), since a layer of pmma of 30 micron in thickness has been coated on the top of the aluminum layer. the other condition required for surface plasmon resonance in the mi structure is [16]: k a ε a + k b ε b = 0 (2) where k a and kb are propagation constants for top dielectric layer a and the aluminu m layer b. they can be computed using the following mathematical relationships [16]: k a 2 = k x 2 − εa k 0 2 (3) kb 2 = k x 2 − εb k 0 2 (4) ko is the propagation constant of vacuum in the horizontal a xis. k x is the propagation constant in the direct ion of propagation. 2.2. surface plasmon resonance for the mim structure to attain this resonance condition in the mim structure (that is, the high-k dielectric between the alu minu m layer and the bottom lead p late), the following criterion shown in eq. (5) and (6) must be met. the first conditio n is [16]: εb ab+ εc ac+ εd a d= 0 (5) where εb , εc and εd are respectively the dielectric constants of the alu minum layer, the high-k materia l and the bottom lead plate. in addition, ab, ac and ad are respectively the thicknesses of the aluminum layer, the high -k material and the bottom lead p late. dielectric constant of the high -k materia l is in the neighborhood of 10. the thic kness of the bottom lead plate d is about 300 microns. according to eq . (5), the thickness of the high-k materia l should be 72 microns. the other condition required for surface plasmon resonance in the mim structure is [16]: k b ε b + k c ε c + k d ε d = 0 (6) where kb , kc and kd are respectively propagation constants for the aluminum layer, the high-k material and the bottom lead plate. 3. negative ion source to counter balance the incoming positive charges accumulated on the top of the antenna, there must be a corresponding negative charge source. this negative charge source can be generated through frictions between two insulators. in this work, the two insulators are water and plastic. fig. 3 illu strates the whole system for harvesting energy from cosmic radiations and atmospheric ions. the sub-system highlighted by the box on the left is the negative ion source capable of generating a negative voltage with respect to the ground out of flowing water. in the negative ion source as shown in fig. 3, the ends of the plastic tube act as the water inlet and the water outlet. the water inlet and the water outlet are respectively fastened with metal ring a and metal ring b. these metal rings never touch water but, between the metal rings and the flowing water, the dielectric material must have a high dielectric constant. between these two metal rings, pin a, pin b and the ground pin are nailed into the plastic tube, submerging into the water flowing through the plastic tube. pin a is wired to metal ring b, whilst pin b is wired to metal ring a. the ground pin, gnd, is wired to the metal faucet. the ground pin can also be shorted to pin a without negatively affecting the measured results. since water is an insulator, running water through the plastic tube will generate static electricity in the inner surface of the plastic tube. assume that the water running at the region close to metal ring a is positively charged. metal ring a will be electrostatically induced to a negative potential, whilst pin a will carry a positive potential. since pin a is wired to metal ring b, the water at the region close to metal ring b and pin b will be negatively charged. the wire connection between pin b and metal ring a forms a positive feedback which amplifies their negative potential. by the same token, the wire connection between pin a and metal ring b forms another positive feedback, amplifying their positive potential. fig. 3 sche mat ic vie w of the prototype modeling the whole atmospheric ions collecting system advances in technology innovation, vol. 2, no. 4, 2017, pp. 99 104 102 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti as shown in fig. 3, the ions in water tend to have a short-circuit effect on the pathway between pin a and the ground and on the pathway between the ground and pin b. to increase the negative charge for the antenna, it is important to decrease the mobility of ions in the region between pin a and the ground or in the region between pin b and the ground. 4. measurements the prototype as shown in fig. 3 has been constructed and its performance has been measured with and without sunlight in the eleventh floor of a building based in shenzhen. we have used a voltmeter (rigol dm3051) to measure the voltage across the diode as shown in fig. 3 at different times of the day. during the measurement, all the sources of electromagnetic disturbance has been removed, or placed very far away from the proposed system, to rule out any possibility of any non-cosmic energy source. the measurement setup is illustrated in fig. 4. the same measurement was taken repeated every lunch break and evening throughout the last month of autumn. our findings are summarized in the following table: table 1 measured voltage across the diode when the water was running at 150 cm 3 per second s i tu ati on s ge n e ral o bservations fi n al voltage s tarting vol tage s tarti n g vol tage be twe e n th e top al l aye r an d th e ground w i th su n l i gh t volt age rose st eadily for 5 seconds ~680 mv 200 mv 350 mv w i th ou t an y su n l i gh t volt age rose st eadily for 5 seconds ~450 mv 200 mv 350mv fig. 4 measurement setup the voltage between pin a and the ground pin of the proposed negative ion source was very stable and it tends to increases when the water flo w rate increases. when water flow rate increases, the final voltage across the diode also increases with approximately a delay of 2 seconds. this is due to the fact that the atmospheric ions that can be harvested are positively associated with quantity of negative charges from the negative ion source. when the bottom lead plate is disconnected from the negative ion source, however, the output voltage fell to around 10 mv and no significant voltage rise can be observed even in the presence of an uv light. the voltage between the top aluminium layer and the ground connection has been successfully used to charge up a 470 uf capacitor. the experimental setup was found to be highly sens itive to external e lectro magnetic disturbance. it was found that placing a lit co mpact fluorescent next to the proposed antenna can cause huge jump in the measured voltage. 5. discussions in this work, the process of ion harvesting, which is mainly driven by flowing water, does not necessitate consumption of fuels to operate. this water can flow from natural sources, including waterfalls, springs, heavy rainfall, tidal waves or rivers. flowing water carries ions which can be further polarized in the proposed negative ion source. the ions from the proposed negative ion source contribute to a potential difference which can be further increased by connecting the proposed negative ion source in series. the end result is that the output voltage will increases, depending on how many of these proposed negative ion source are connected in series. table 1 shows that the output voltages measured with and without sunlight are clearly very different. sunlight, background radiations or other forms of electromagnetic disturbance can lead to increased voltage across the diode, suggesting that there are indeed background radiations or atmospheric ions being captured. the quantity of the negative charges from the negative ion source determines how much and how fast the energy from atmospheric ions can be captured into a dc electricity. unlike a conventional solar panel, this energy can be continuously harnessed throughout the day. in general, the purer the water is, the more negative charges can be generated from the negative ion source. impure water contains ions which neutralize the generated electrostatic charges. however, our latest experimental results suggest that sea water yields a much higher voltage. increasing the ionization rate is the key to improve the efficiency of an atmospheric ion collector. the expression for rate of ionization at different altitudes, q(h) or different phases of solar cycle can be summarized by the following equation [17-21]: q (h)= ∑ i∫ ei ∞ ∫ a= 0 2 π ∫ θ= 0 π 2 + δθ d(i ) ( e )( de dh ) (i ) sin (θ)dθ da de q (7) where (de/dh) are ionization losses [23-24] of partic les of type i, a is the a zimuth angle and θ is the angle towa rds the vertical. dh takes into account that at a given height h that the particles can penetrate from the space angle (0, hma x=90+dθ ), which is g reater than the upper he misphere angle (0, 90) for flat mode l. di( (e) is the cosmic ray diffe rential spectrum that can be used to calculate cosmic rays during different phases of solar cycle [16-18]. advances in technology innovation, vol. 2, no. 4, 2017, pp. 99 104 103 cop y right © taeti ionizat ion in the atmosphere takes place as a result of cosmic radiat ions or ultraviolet radiations. mounting the atmospheric ion collector at high altitudes in theory can lead to higher ionization rates. mounting a device at high altitudes is not always practically rea listic. ho wever, there are alternatives. at ground levels, air pollution is major source of positive ions that can be easily captured using the proposed device. other ionizat ion sources include the ultraviolet radiations fro m sporadic solar energetic particles, cosmic rays and the ions due to the natural radioactivity of the soil which is bla med for the unwanted radon gas. ultrav iolet radiations tend to be intensive in ozone depleted regions, where the air is usually saturated with positively charged air pollutants. during the process of electric ity generation using the proposed technology, the negative charges fro m the proposed atmospheric ion collector will neutralize the positively charged pollutants in air to some e xtent. whether the energy comes fro m uv light, cosmic rad iations or fro m positively charged pollutants, the total of the energy that can be captured depends very much on the sun and the detectable solar activity. hence, it a lso depends on the diurnal (time of day) effect and a seasonal effect. levels of ionization tend to be low during the winter t ime because the sun is farther away. the activity of the sun is a lso associated with the sunspot cycle. there will be mo re solar rad iations when the sun is active. positive ions are known to be detrimental to our health. the atmospheric positive ions are electrostatically attracted to the body, which naturally carries negative charges, and cause oxidative damages. there have been published reports on the link between lack of negative ions in the body and ailments, including headaches, and/or fatigue [23-24]. in 2013, the international agency for research on cancer (iarc) even classified these particulates as a carcinogen [23]. in its evaluation, the iarc showed an increasing risk of lung cancer with increasing levels of exposure to outdoor air pollution and particulate matter. the process of energy harvesting, which is mainly driven by gravity, involves neutralization of atmospheric positive ions. harvesting atmospheric ions in areas plagued by air pollution is one of the few solutions to nullify the harmful effects of the excessive accumulation of positively charged pollutants in the atmosphere. 6. conclusions in this paper, we have proposed the use of flowing water for capturing atmospheric ions into a dc voltage. measurement has been performed with and without any sunlight. measured results suggest that atmospheric ions can be harvested to an observable extent by the proposed atmospheric ion collectors with or without a sunlight. the harvested energy is positively associated with the purity of water and the quantity of the negative charges from the negative ion source. acknowledgment this work is supported by sustech teaching innovation funds (jg201505), national natural science foundation of china (61401191), guangdong natural science funds for distinguished young scholar (2015a030306032), sust c funds (frgsustc1501a-51, frg-sustc1501a-65), and shenzhen science and technology innovation committee funds (jcyj20150331101823678). references [1] k. sheikh, “new concentrating solar tower is worth its salt with 24/7 power,” scientific american, 2016. [2] d. bie llo, “ how to use s olar energy at night,” https://www.scientificamerican.com/article/how-to-us e-solar-energy-at-night/, 2009. [3] h. v. cane and d. lario, “an introduction to cmes and energetic particles,” space science review, vol. 123, no. 1, pp. 45-56, 2006. [4] l. liu and p. solis, “the speed and lifetime of cosmic ray muons,” mit undergraduate, 2010. [5] n. tesla, “the eternal source of energy of the universe, origin and intensity of cosmic rays,” new york, 1932. [6] p. i. y. ve linov, s. asenovski, k. kudela, j. lastovicka, l. mateev, a. mishev, and p. tonev, “impact of cosmic rays and solar energetic particles on the earth's ionosphere and atmosphere,” journal of space weather and space climate, vol. 3, 2013. [7] n. tesla, method of utilizing radiant energy, u.s. patent, 685,958, nov. 25, 1901. [8] h. plauson, conversion of atmospheric electric energy, u.s. patent, 1540998, jun. 9, 1925. [9] h. plauson, improvement in electric motors, british patent, 157,262, jul. 10, 1922. [10] h. plauson, process and apparatus for converting static atmospheric energy into dynamic electrical energy of any suitable high periodicity, british patent, 157,263, jul. 10, 1922. [11] w. i. pennock, apparatus for collecting atmospheric electricity, u.s. patent, 911,260, feb. 2, 1909. [12] l. g. smith, ionospheric battery, u.s. patent, 3,205,381, sep. 7, 1965. [13] t. yuki, method and apparatus for capturing an electrical potential generated by a moving air mass, u.s. patent, 533,237, jan. 15, 1985. [14] g. f. knoll, radiation detection and measurement, 4th ed. wiley, 2010. [15] http://ecuip.lib.uchicago.edu/multiwavelength -astrono my/images/infrared/history/transmission-page.jpg [16] a. i. smolyakov, “resonant modes and resonant transmission in mult i-layer structures,” progress in electromagnetics research, vol. 107, pp. 293-314, 2010. [17] v. pere z, d. d. ale xander, w. h. bailey, “air ions and mood outcomes: a review and meta-analysis”, bmc psychiatry, vol. 13, no. 1, p. 29, 2013. [18] r. a. baron, g. w. russel, and r. l. arms , “negative ions and behavior: impact on mood, memory, and aggression among type a and type b persons ,” journal of pers onality and social psychology, vol. 48, no. 3, pp. 746-754, 1985. [19] s. wu, f. deng, j. huang, h. wang, m. shima, x. wang, y. qin, et al., “ blood pressure changes and chemical constituents of particulate air pollution: results from the healthy volunteer natural relocation (hvnr) study,” environmental health perspectives, vol. 121, no. 1, pp. 66-72, 2013. [20] m. buchvarova and p. i. y. ve linov, “empirical model advances in technology innovation, vol. 2, no. 4, 2017, pp. 99 104 104 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti of cosmic ray spectrum in energy interval 1 mev-100 ge v during 11-year solar cycle,” advances in space research, vol. 45, no. 8, pp. 1026-1034, 2010. [21] r. sternheimer, “in fundamental principles and methods of particle detection,” methods of experimental physics, vol. v, a. nuclear physics, edited by l.c.l., yuan, and c. s. wu, new york, london, academic press, 1961. [22] m. ackermann, m. ajello, a. allafort, l. baldini, j. ballet, and g. barbiellini, et al., “detection of the characteristic pion-decay signature in supernova remnants”, science magazine, vol. 339, p. 807, 2013. [23] d. loomis, w. huang, and g. chen, “the international agency for research on cancer (iarc) evaluation of the carcinogenicity of outdoor air pollution: focus on china,” chin j cancer, vol. 33, no. 4, pp. 189-196, 2014. [24] k. sa kakibara, “influence of negative ions on drivers ,” r & d review of toyota crdl, vol. 37, no. 1, 2002. template encit2010 advances in technology innovation, vol. 5, no. 1, 2020, pp. 01-09 optimizing engine start systems with application to sailing/coasting and mild hybridization madhusudan raghavan * propulsion systems research lab, general motors r&d, warren, michigan, usa received 25 april 2019; received in revised form 23 june 2019; accepted 20 august 2019 doi: https://doi.org/10.46604/aiti.2020.4148 abstract engine start systems are key to providing a good customer experience for today’s drivers. considerable effort goes into ensuring a smooth and quiet engine start, especially in vehicles equipped with start/stop systems. we present two novel mechatronic starters that are designed to improve start quality by enabling faster and quieter engine starts. in the first proposed concept, the traditional alternator is replaced with a motor/generator unit that is capable of exerting positive torque on the engine as needed, in addition to the conventional power generation function. the motor/generator is selectively connected to the crankshaft via a selectable geared or belted connection to enable different operating modes. this starter executes a 400 ms faster start for a typical engine when compared to a conventional 12v starter. we also present a second starter that uses an integrated two -speed gear train to crank the engine. the cranking gear ratio is changed from the initial high ratio to a lower ratio once the engine starts to spin. this ratio change allows the starter motor to continue to operate in a favorable torquespeed zone and push the engine to a higher pre-ignition rpm than a conventional starter, resulting in a quieter, smoother start. we also present results from incorporating the belted/geared starter concept in vehicles with sailing/coasting mode as well as in mild hybrid propulsion systems. sailing/coasting mode of operation is enabled by the quick engine re-start capability of this starter allowing seamless switching between fuelled and unfuelled engine operation. such an operation could reduce fuel consumption by about 3-6% on the nedc driving cycle, without regenerative braking. one may further hybridize the propulsion system by adding a battery for storing regenerative braking energy. using such an architecture, a 6-8% fuel economy improvement on the wltp certification driving cycle may be achieved, depending on voltage and power levels implemented, as well as energy storage systems included. keywords: start/stop, mild hybrid, silent start, sailing/coasting 1. introduction transportation is the source of approximately 25% of greenhouse gas (ghg) emissions worldwide. with population growth and increased energy use and travel, transportation emissions are growing very fast [1]. these include both motor vehicles and air transport. these ghg challenges are being tackled by the introduction of low carbon fuels, electrification of the propulsion system, and increased renewable electricity generation from wind and solar. some of these initiatives are yet to see widespread adoption due to associated system cost. in this framework, light or mild electrification (under 50v) of the propulsion system appears to have mass-market potential because of the favorable value proposition ratio it offers. smooth, fast engine starts are critical elements of such mild hybrid systems. * corresponding author. e-mail address: madhu.raghavan@gm.com tel.: +1 248 930 5248 advances in technology innovation, vol. 5, no. 1, 2020, pp. 01-09 2 a start/stop system switches off the engine when the vehicle comes to a stop, thereby eliminating the consumption of fuel and the release of emissions during engine idling. however, this creates the need for fast engine restart systems to achieve satisfactory tip-in response when the driver depresses the accelerator to drive away. reference [2] performs an assessment of various start/stop systems and investigates the delay in tip-in response and launch performance when the driver depresses the accelerator when the engine is off. reference [3] describes experiments to study the difference in particulate mass emissions for gdi engines with and without start/stop systems. they found that the additional starts with start/stop systems did not significantly increase the particulate mass emissions. reference [4] presents a coupled magneticthermal model to study the reason for the damage of the starter motor of a start/stop system of a city bus. reference [5] describes the testing of idle-stop systems to see whether real-world fuel savings of such systems are in line with those predicted by the epa fuel economy certification cycles. their results suggest that in many cases the idle stop systems show minimal benefits on the epa cycles but deliver considerable fuel savings during real-world operation. reference [6] describes a novel start/stop system that injects compressed air into appropriate cylinders in the engine to get the engine spinning before it is fueled and sparked. they claim that such a system could potentially replace the conventional starter. reference [7] studies the operation of engine starts and stops in electrified propulsion systems with a focus on the effects of engine temperature on engine cranking torques and start-up emissions. they suggest that it may be advisable to take into consideration engine temperature while crafting engine start/stop strategies. reference [8] describes advancements in lead-acid battery technology, in particular, enhanced flooded batteries and absorbent glass mat batteries, driven primarily by the projected growth in start/stop system market volumes. reference [9] quantifies the co2 potential of start/stop systems by comparing two diesel-powered vehicles in urban driving conditions. one of them has a conventional propulsion system and the other one has a start/stop system. co2 reductions of as much as 20% were obtained with the start/stop system. reference [10] describes a 36v belted alternator starter system with a 7 kw mgu installed on a 1.9l four-cylinder engine. this enhanced start/stop system delivers 12-14% fuel economy improvement on the ftp city cycle and about 1% improvement on the ftp highway cycle. reference [11] describes the creation of a nonlinear control algorithm to execute smooth stops and restarts on a diesel engine, wherein the cranking torques are much higher than for a gasoline engine due to the higher compression ratio. reference [12] describes a 12v belted alternator starter system that can execute engine start/stop functions as well as improve engine responsiveness by means of torque addition to the driveline during transient maneuvers. reference [13] investigates various light electrification architectures ranging from 12v start/stop systems to 48v electrified transmissions, to assess their optimality when applied to a range of vehicle types and motor/generator locations. reference [14] investigates engine start quality, nvh, and cranking speed, and presents experimental data showing that faster cranking is better. in the present work, we describe two new starter concepts that yield fast, smooth, the engine starts suitable for future products. we also show how these starters may be integrated into propulsion architectures to yield fuel economy improvements via sailing/coasting and mild hybridization. 2. mechatronic starter for geared/belted operation in this section, we describe a novel starter system with unique functional operating characteristics. it is created by replacing the traditional alternator with a motor/generator unit (mgu) of approximately the same physical size. this mgu may be used to perform the engine starting functions. since the alternator interacts with the crankshaft via the accessory belt, the mgu can also interact with the crankshaft via this same belt. we include a dual tensioner on this accessory belt as the mgu exerts a negative torque while generating electricity and a positive torque while executing engine starts or driveline torque boosting. the dual tensioner ensures proper belt operation under these different operating conditions. for situations wherein, belt-based engine starting is not advisable (e.g., in extremely cold weather when ice may form on the belt), we need advances in technology innovation, vol. 5, no. 1, 2020, pp. 01-09 3 to retain the services of a geared conventional starter. however, with an innovative mechatronic arrangement, we may potentially eliminate this conventional starter by making the mgu play the role of a geared starter. such an arrangement is shown in fig. 1, with the mechatronic device of interested being indicated by the dashed rectangle. fig. 1 engine with a geared/belted mechatronic starter it is shown in considerably more detail in fig. 2 and in a kinematic diagram format in fig. 3. it is comprised of 2 solenoids which act on yokes, causing a pair of pinion gears to move outwards. as indicated via the red and blue color coding, the solenoids cause their respective pinions to move outwards when actuated, assisted by the lever action of the yokes. when the solenoids are not energized the pinions return to their default positions via return springs. the pinions are mounted on a splined shaft, so they rotate with the shaft but can also translate axially relative to the shaft, as dictated by the solenoids. this translation is coupled with a slight rotation due to the spline angle. the shaft with the splines is mounted to ground (in this case the engine block) via bearings (a revolute joint). the purpose of this mechatronic device is to selectively connect the mgu to the crankshaft flywheel when needed. the pinion gear on the right (in fig. 1) engages with a gear mounted on the mgu. the pinion gear on the left engages with the flywheel which is integral with the crankshaft. thus, this arrangement creates a geared connection between the mgu and crankshaft and the mgu can now execute a geared start similar to the conventional starter. this arrangement thus eliminates the need to carry a conventional starter in addition to the mgu. fig. 2 mechatronic devices to connect the motor/generator to the flywheel to avoid an overconstrained system, an electric clutch is introduced between the mgu and accessory belt sprocket. when the mgu is “geared” to the flywheel, this clutch is opened and the mgu does not exert any torque on the accessory belt. after the geared engine start is executed, and the solenoids are de-energized, the mgu is no longer connected to the flywheel. at this time, the clutch may be engaged to connect the mgu to the belt to allow generation or torque boosting as advances in technology innovation, vol. 5, no. 1, 2020, pp. 01-09 4 appropriate. thus, the mechatronic arrangement of figs. (2)-(3) enables a selectable geared/belted connection of the mgu to the crankshaft. when conditions are favorable (moderate temperatures and humidity), the engine start may be executed via the belt side connection of the mgu to the crankshaft, without having to resort to the geared operation. since the mgu is several times more powerful than the traditional 12v starter motor, the engine starts resulting from this geared/belted arrangement are considerably faster, smoother, and quieter than conventional engine starts. additional details of the operation of this motor/generator-based starter device may be found in reference [14]. fig. 3 kinematic diagram of a mechatronic device 3. two-speed starter system next, we describe a two-speed starter that succeeds in squeezing more cranking effort out of a conventional starter motor. such motors are typically equipped with a single-speed ratio which is optimized to get the engine turning quickly. however, due to the shape of the motor’s torque-speed characteristics, the cranking torque drops rapidly as engine rpm increases. in the proposed invention, we insert a two-speed gearing between the starter motor and the crankshaft flywheel. cranking begins with a high torque ratio to get the engine turning. we then switch to a lower torque ratio, so that the motor can continue to operate in its optimal torque-speed zone and continue to push the engine past its rated pre-ignition rpm. typically ignition occurs when the engine has reached 300-400 rpm. the currently proposed system is able to spin the engine to 700-800 rpm and this results in a smoother start. there are various possible schemes to execute this two -speed gearset. in our study, we have focused on the use of a ravigneaux gearset due to its potential compactness resulting from the shared pinion between adjacent planetary gearsets. alternatively, layshaft gears and clutches could also be used to execute this speed ratio change. fig. 4 shows the steps entailed in this two-speed engine start. the starter motor begins operation along the line labeled gear ratio 1. the speed ratio change is then executed as shown by the segment labeled ratio change. the starter continues to push the engine, its operation state indicated by the line labeled gear ratio 2. note the engine rpm line with the yellow and red stars. with a conventional single-speed starter, engine ignition would occur at the yellow star, at which time the starter motor would have reached its peak rpm. by virtue of the proposed two -speed arrangement, the engine rpm can reach the red star before ignition, and starter motor rpm would still be within its feasible limits (indicated by the dashed red line at the top of the figure). the conventional starter motor generally reaches a maximum output speed of 18000 rpm.when the engine rpm reaches 400 rpm. this translates to a 5:1 ratio for the starter motor gearing, given that the starter pinion to flywheel gear ratio i s approximately 15:135. for our two-speed arrangement, we use the same 5:1 ratio for the first speed and then switch to a 2.5:1 ratio for the second speed. this would allow the engine to reach approximately 800 rpm when the starter motor attains its maximum speed of 18000 rpm. for a generic 2-speed compound planetary ravigneaux gear set, we used the following advances in technology innovation, vol. 5, no. 1, 2020, pp. 01-09 5 gear teeth numbers: 𝑁𝑆𝑢𝑛1 = 50 , 𝑁𝑅𝑖𝑛𝑔1 = 200 , 𝑁𝑃𝑖𝑛𝑖𝑜𝑛1 = 75, 𝑁𝑆𝑢𝑛2 = 74, 𝑁𝑃𝑖𝑛𝑖𝑜𝑛2 = 30 . in our simulations, we executed the ratio change at 280 rpm engine speed, at approximately the mid-point of the start event. for the purposes of the simulation, one may note that the engine inertia is 0.1 kg-m2 and the starter rotor inertia is 0.000241 kg-m2. the starter motor torque begins with a value of approximately 6 n-m at 0 rpm and drops linearly with speed, down to 0 n-m in the neighborhood of 18000 rpm. fig. 4 engine speed (rpm) with two gear starters fig. 5 shows a comparison between the two-speed starter and the conventional single ratio starter. the two-speed starter is able to crank the engine to approximately 800 rpm while the latter saturates at around 400 rpm. these are unfired simulations, hence the wavy behavior of the engine rpm after ignition speed is reached. the red, green, and blue traces for the novel starter result from different assumed actuation times for the speed ratio change (20, 30, and 40 ms respectively), and thus show the sensitivity to actuator capability. additional details of the operation of this two-speed starter device may be found in reference [15]. fig. 5 engine rpm for conventional vs. two-speed starter 4. application to sailing/coasting and mild hybridization of the two proposed starter systems, the first one enables a considerably faster start, but with the added cost of the alternator being replaced by a motor/generator unit. additionally, a bi-directional tensioner must be added to the front-end accessory drive belt, in order to allow driveline torque boosting operation as well as belted engine starts. in contrast, the second proposed starter enables smooth starts by cranking the engine to a higher rpm prior to ignition. in this case, the added cost is that of the two-speed gearset in place of the single ratio gearset of the conventional starter. the first proposed starter concept may be used for sailing/coasting operation as well as in a mild hybrid vehicle. advances in technology innovation, vol. 5, no. 1, 2020, pp. 01-09 6 the terms “sailing” and “coasting” are used interchangeably and refer to the mode of operation when the engine is shut off and disconnected to minimize engine drag losses during decelerations. this is popular in europe and china with the high penetration of manual transmissions. when the engine is decoupled from the driveline during coasting, one possible operating strategy is to keep the engine running at idle for quick re-engagement to the driveline when the driver demands acceleration. this idle operation of the engine during coasting continues to use fuel. the proposed first concept in this pap er gets around this problem, as it enables ultra-fast re-starts and thus potentially allows one to maximize the true “engine off” time periods during coasting, thus maximizing fuel economy. having a quick re-start capability allows one to re-start and reconnect the engine to the driveline with minimum delay and adequate acceleration response. coasting may be thought of as a vehicle transient state between cruising and braking. the various sailing/coasting modes of operation are shown in fig. 6. fig. 6 sailing/coasting a typical driving maneuver is divided into 6 sections or modes. in mode 1, the vehicle is initially stopped (perhaps at a traffic light) with the engine in auto stop mode. in mode 2, the driver releases the brake pedal and depresses the accelerato r pedal. the engine starts and provides torque to accelerate the vehicle as indicated by the linearly increasing speed. once the vehicle reaches the desired cruising speed the driver reduces pressure on the accelerator allowing the vehicle to sail/coast in mode 3. the engine remains on but is disconnected from the driveline. in mode 4, the sailing/coasting operation is continued, but with the engine disconnected and shut off. in mode 5, the driver depresses the brake pedal to slow the vehicle down as needed. in mode 6, the vehicle continues to slow down, but with the motor/generator-based starter ready to make a quick engine re-start in case of a “change of mind” situation, wherein the driver decides to increase speed instead of slowing down (as when a traffic signal turns green). if this change of mind situation does not occur the vehicle comes to a complete stop at the end of mode 6. fig. 7 fuel consumption reduction by using sailing/coasting on a certification driving cycle advances in technology innovation, vol. 5, no. 1, 2020, pp. 01-09 7 fig. 7 shows how the sailing/coasting mode may be used to save fuel on a certification driving cycle. the black line indicates stipulated vehicle speed on the driving cycle. the blue line indicates how the sailing/coasting mode may be used to approximate the sharp decelerations on the stipulated driving cycle with more gradual coasting maneuvers. this allows one to save fuel as indicated by the differences between the red curve (fuel rate for stipulated vehicle speed) and the green curve (fuel rate for the sailing/coasting approximation). note that the green fuel rate drops to zero during the sailing/coasting portions while the red curve remains non-zero on these portions. conservative estimates suggest that this type of sailing/coasting on the nedc driving cycle could save about 3-6% of the fuel consumed. this does not require a large additional battery, as we are not storing any regenerative braking energy for this mode of operation. we had mentioned earlier the ability of the motor/generator-based starter concept in belted mode to execute super-fast restarts and thus maximize fuel savings while ensuring adequate acceleration response. experimental data backing up this claim are shown in fig. 8, where we see engine rpm plots for a fired engine start using the motor/generator (blue) and the conventional starter (red). the engine rpm ramp rate achieved with the motor/generator unit far exceeds that of the conventional starter, resulting in a 400 ms faster spin up to 550 rpm. moving on from sailing/coasting, we can go one step further in mild hybridization by making the additional investment in a battery for storing regenerative braking energy. this leads to additional fuel economy gains. when the motor/generator unit of the starter is coupled to a 120 wh li-ion battery, such a system, with optimal supervisory powertrain control, can yield 6-8% of fuel economy improvement on the wltp driving cycle [12]. fig. 9 shows simulation plots of cumulative battery regeneration energy (i.e., a summation of energy flow into the battery) during the wltp driving cycle, for a 1350 kg. passenger vehicle equipped with this system. the red curve shows the results of a 12v implementation of such a system and the blue curve shows a 48v implementation. the grey curve is the vehicle speed trace during the wltp cycle. overall, approximately 1000 to 1200 kj of regenerative braking energy is captured in the battery during the driving cycle. this energy, when utilized in the propulsion system to offset 12v electrical loads as well as for driveline torque boosting (i.e., exerting electrical torque on the crankshaft via the motor, in place of mechanical torque from the engine), results in the above-mentioned fuel savings. we have been able to confirm this experimentally on instrumented test vehicles. fig. 8 motor/generator vs. a conventional starter fig. 9 cumulative regenerative braking energy into the battery for wltp 5. conclusion we have described two novel starter concepts. the first one can switch between geared and belted operation. this integrated starter enables a very fast and smooth start in belt mode compared to a conventional 12v starter. the second concept uses a two-speed starting device to crank the engine to a higher rpm prior to ignition. it is comprised of an integrated arrangement of gears and clutches that changes the over-all cranking gear ratio during the course of the start event. the two advances in technology innovation, vol. 5, no. 1, 2020, pp. 01-09 8 designs are very comparable and equally applicable to various types of automobiles. the first starter uses two solenoid actuators and sliding surfaces to move the pinions outwards to execute the belt-to-gear mode changes. the second starter uses an additional plane of gears and brakes/clutches to execute the speed ratio change during the start. both require additional packaging space compared to conventional starters. the first starter is capable of faster starts as the alternator is replaced by a motor/generator unit with more power capability than a traditional starter motor. our experiments have shown that this motor/generator-based starter is about 400 ms faster than a conventional starter. we have investigated the use of this fast start capability for sailing/coasting operation wherein the engine is disconnected and shut off during vehicle deceleration. the fast starter allows quick engine re-start and re-connection to the driveline in response to driver power demand. the use of this sailing/coasting mode of operation could save about 3-6% fuel on the nedc driving cycle. beyond sailing/coasting, one may further increase the level of mild hybridization by adding a battery for regenerative braking energy storage. in such an architecture, the motor/generator-based starter, in belted mode, enables hybrid functions such as torque boosting and regenerative braking to achieve a 6-8% improvement in fuel economy on the wltp driving cycle. conflicts of interest the authors declare no conflict of interest. table of notations apu auxiliary power unit ecu engine control unit ecm epa engine control module environment protection agency ev electric vehicle ghg gdi greenhouse gas gasoline direct injection hwfet highway fuel economy driving schedule mgu motor generator unit nedc new european driving cycle nvh noise, vibration, harshness nycc new york city cycle soc state of charge udds urban dynamometer driving schedule wltp world-harmonized light vehicle test procedure references [1] d. greene and s. plotkin, “reducing greenhouse gas emissions from u.s. transportation,” pew center on global climate change, pp. 1-80, may 2003. [2] t. wellmann, k. govindswamy, and d. tomazic, “integration of engine start/stop systems with emphasis on nvh and launch behavior,” in sae technical paper 2013-01-1899. [3] j. storey, m. moses-debusk, s. huff, j. thomas, m. eibl, and f. li, “characterization of gdi pm during vehicle startstop operation,” in sae technical paper 2019-01-0050. [4] z. xu, n. tao, m. du, t. liang, and x. xia, “damage prediction for the starter motor of the idling start-stop system based on the thermal field,” in sae international journal of commercial vehicles. 10(2):2017, 2017-01-9181. [5] j. wishart, m. shirk, t. gray, and n. fengler, “quantifying the effects of idle-stop systems on fuel economy in lightduty passenger vehicles,” in sae technical paper 2012-01-0719. [6] a. inglis, r. timewell, and s. maskerine, “a novel compressed air starting system,” in sae technical paper 2003-012279. [7] n. henein, d. taraza, n. chalhoub, m. lai, and w. bryzik, “exploration of the contribution of the start/stop transients in hev operation and emissions,” in sae technical paper 2000-01-3086. advances in technology innovation, vol. 5, no. 1, 2020, pp. 01-09 9 [8] t. costlow, “powering up the new stop-start systems,” automotive engineering, june 2016. [9] n. fonseca, j. casanova, and m. valdes, “influence of the stop/start system on co2 emissions of a diesel vehicle in urban traffic,” transportation research part d: transport and environment, vol. 16, no. 2, pp. 194-200, march 2011. [10] g. tamai, t. hoang, j. taylor, c. skaggs, and b. downs, “saturn engine stop-start system with an automatic transmission,” sae transactions, vol. 110, pp. 270-280, 2001. [11] m. canova, y. guezennec, and s. yurkovich, “on the control of engine start/stop dynamics in a hybrid electric vehicle,” asme journal of dynamic systems measurement and control, vol. 131, no. 6, pp. 1-12, november 2009. [12] m. raghavan and a. balhoff, “electrical torque addition mechanism for engines with high levels of egr,” in eucomes 2018: proceedings of the 7th european conference on mechanism science, edited by burkhard corves, philippe wenger, mathias hüsing, 2018, pp. 165-172. [13] m. raghavan, “mild electrification across a spectrum of vehicle types,” in fisita technical paper no. f2018-ehv009, 2018. [14] m. raghavan, n. bucknor, and v. donikian, “the kinematics and dynamics of engine start systems,” in iftomm asian mms conference, bangalore, 2018. [15] m. raghavan, “novel mechanisms to improve the start quality of automotive engines,” in iftomm world congress, krakow, 2018. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation, vol. 4, no. 3, 2019, pp. 165-176 the strategy of energy saving for smart shipping heiu-jou shaw 1 , fu-ming tzu 2,* 1 department of system and naval mechatronic engineering, national cheng kung university, tainan, taiwan 2 department of marine engineering, national kaohsiung university of science and technology, kaohsiung, taiwan received 13 march 2019; received in revised form 22 april 2019; accepted 25 may 2019 abstract this paper presents a real record and analysis to reduce carbon dioxide (co2) emission and fuel consumption using the energy efficiency operational index (eeoi) for practical ships. with giant commercial tankers constantly launching into the sea, emission increase with the tonnage is unavoidable. more weight means more pollution. in the experiment, big data of internet of things (iot) were collected to analyze the sailing conditions for over six months through global satellite communications while the ship sailed across the ocean. the results showed that deep draft results in a better performance than shallow draft; that is, the ship carries more cargo but consumes less fuel oil, as opposed to the traditional concept. among the parameters, the draft and cargo mass interacts with the index to identify the fuel consumption at a constant speed; a draft between 11.5 m and 12 m was found to be optimum for this type of container ship. the eeoi is a metric tool to illustrate the variation of fuel consumption and an effective management strategy to reduce the co2 emission quantity. moreover, a constructive model using iot technology and an energy efficiency management strategy presents an accurate big data basis to guide decision making on the vessel. keywords: cargo loading, speed over ground, average draft, fuel consumption, energy efficiency operational index (eeoi) 1. introduction following the booming shipping market, giant container ships now constantly launch into the sea; thus, greenhouse gas pollution is getting more severe due to the increasing carbon dioxide (co2) emission [1, 2]. therefore, an effective method to evaluate the emission [3-5] is urgently needed. the energy efficiency operational index (eeoi) [6, 7] is a metric unit that measures the co2 emission of a ship and also predicts the consumption of the fuel oil in the vessel. presently, energy saving is an extremely necessary task on board since the container ship is always on missions across the oceans so that schedules to improve energy saving are hardly made. furthermore, the international convention for the prevention of pollution from ships, adopted by the international maritime organization (imo), is the standard guiding the energy efficiency of ships [8]. moreover, the measurement of the performance indicator refers to iso19030, which describes the measurement of changes in the hull and propeller parameter for energy efficiency [8]. recently, big data analysis has become a mainstream statistical method for maritime management for global shipping [9]. various shipping companies utilize smart shipping to analyze the navigational attitude, loading, nacelle, and sea weather information, and this improves the energy efficiency to enhance sailing safety. in the globalization era, the shipping industry has become a key factor of the economy in a world where over 90% of global trade is seaborne [10]. moreover, the weight scale of container ships is in megatonnage; thus, in this study, we use current data recorded in a ship to measure the emission * corresponding author. e-mail address: fuming88@nkust.edu.tw tel.: +886-7-8100888#25245; fax: +886-7-5716043 advances in technology innovation, vol. 4, no. 3, 2019, pp. 165-176 166 pollution. if the shipping industry were a country, its co2 emission would be higher than that of germany [2]. the industry accounts for 3% of the world’s co2 emission, ranking as the seventh-largest seventh emitter among countries. in 2013, the imo, aiming at energy saving in ships, enforced regulation for emission control that mandates every shipping company to build a self-project based on the ship energy efficiency management plan (seemp). however, the seemp cannot reflect the real consumption of fuel oil without the eeoi. the eeoi is voluntary behavior to take the measured by own shipping. when a ship sails across the ocean from country to country, especially to an emission control area [11, 12], it follows rather strict regulations; the regulation can help control the pollution and carbon emission quantity. the paper contributes two aspects to reduce co2 emission and save fuel consumption. first, the relationships between power efficiency and speed as well as power efficiency and trim at various drafts are analyzed. second, an eeoi-based strategy for ship energy management is presented. the measured parameters include engine power, trim, draft, fuel consumption, cargo mass, and weather condition. 2. systematic architecture this work is focused on the evaluation of energy saving for ships using the eeoi metric. we selected a practical container ship to install a digital device that functions based on the internet of things (iot) technology. the device collects the sailing information and creates a whole communication system between the headquarter (hq) and ship for energy efficiency by big data analysis (fig. 1). an eeoi-based energy efficiency strategy can then be suggested for improving efficiency and thus saving energy. the installed device consists of a vessel data recorder and an alarm monitoring system (ams). the data are transferred to the hq by amos software (spec tec group holding ltd, italy) using satellite communication and fed back to the ship [13], using the glocalme wifi all-day network service. the devices are installed in the engine room and bridge, and the sub-server is installed inside the electrical room where the big data of a particular parameter are transferred to the main server. this is a novel forward-looking method to manage energy efficiency through big data analysis, whereby the data model is constructed using eeoi, and valuable information is provided to decision makers for smart shipping. fig. 1 communication involving big data using satellite on a practical ship the ship’s big data architecture is divided into three layers. the first layer collects the hardware information of the ship. at this stage, the sensing layer, that is, the installed active sensor, collects information on time from the host (auxiliary machines, and power centers); such information includes navigational information, such as draft, speed, load, and climate condition, and actual operational information. in addition, the fiber optic sensor detects the main shaft speed, and the mass flow meter collects the fuel state. the second layer is the network layer. the information detected by each sensor is collected into the gateway and then enters the router. a wired or wireless network connection is used. when the information packet enters the server, the statistical program automatically begins to connect the satellite communication. the shore-based hq of the traffic control center and the land data center complete the networking function, and the data are scheduled to be sent to the designated advances in technology innovation, vol. 4, no. 3, 2019, pp. 165-176 167 address. the third layer is the application layer. the big data collection is not only used by a single ship; the statistical program structure data are analyzed by the host computer and the server to connect the data to the same type of ship, allowing the fleet to use the data simultaneously. the construction of this system can maximize the application of energy efficiency of the entire ship, which enhances convenient pipeline operations in maritime science for exploration, supporting the practitioners, and reducing marine accidents. the evaluation is based on the eeoi metric unit to examine the fuel oil consumption and co2 emission per distance (nautical mile) for a period of over six months, from feb. 8 to aug. 3, 2018. the evaluated parameters in the experiment include eeoi, cargo mass, fuel consumption, distance, avg. draft, and trim by the fore and stern of a practical ship. 3. data acquisition the data were acquired at a steady rate to measure the hull and propeller parameters [14]; the digital device of big data analysis was used in the experiment to acquire information. the digital data were selected to measure the loading of the container ship every five minutes. the data were collected on board for a voyage of six months: 166 days in the sea and 14 days is in the port; that is, for roughly 92% of the operation time, the ship was in activity. data were still collected when the ship was at the harbor. the steering angle frequently varied due to ship operations, such as entering or departing the harbor, and special conditions. moreover, to fulfill the safety requirements, the speed of the ship was adjusted by the captain, who manipulated the ship to avoid collision during the course. the engine order and rudder order in the interaction often varied for safe navigation. normally, the “ahead” orders of the ship consist of stop, standby, dead slow ahead, slow ahead, half ahead, full ahead. if the ship needs an astern, then the orders include dead slow astern, slow astern, half astern, and full astern. the sensor device installed in the ship is very challenging in the task which the marine equipment must maintain at safety operation, and the installed device cannot affect operation. the acquired parameters, which include fuel consumption, loading change, draft, and trim profile, were analyzed through big data analysis. in addition, due to various weather conditions, the sea wave significantly affected speed. the speed over ground (sog) is the same as the speed displayed on the global position system (gps). furthermore, the ship was manipulated by autopilot while sailing in the wide ocean. however, for safety, the steering gear can be released by hand pilot to avoid an accident. the eeoi is a metric unit to quantify co2 emission and a method to evaluate fuel consumption [6, 7]. according to the imo, the eeoi can be expressed as the product of the co2 factor of transfer coefficient and fuel consumption divided by the product of carrier cargo and sailing distance: average eeoi= ∑ co2factor×foc (tonne) cargo mass (tonne)×sailed distance (nautical mile) n 1 (1) here, n is the total number of voyage segments over the period. a co2 factor of 3.114 (t-co2/t-fuel) was assumed in our experiment, based on the heavy fuel oil at the type of main engine [15]. the foc means fuel consumption during the voyage every five minutes for over six months. hence, the numerator represents the co2 emission of fuel consumption. typically, the large two-stroke cycle of marine engine uses the heavy fuel oil to supply the main power. on the other hand, the denominator represents the cargo mass carried during the voyage. a large denominator, i.e., small eeoi, is based on a certain consumption of fuel oil, and it means higher efficiency of the ship. in this study, the eeoi at the ship varied from 1~3 × 10 −5 , depending on the parameters. thus, we anticipate the measurement method will save fuel oil consumption and reduce the co2 emission to support the marine industry. the main engine (manufactured by man b&w), which is a typical diesel engine of the large two-stroke cycle using heavy fuel oil, was installed in the ship. the technical specifications are presented in table 1. advances in technology innovation, vol. 4, no. 3, 2019, pp. 165-176 168 table 1 the technical data of the ship in the experiment item description unit launch date 2012 main engine type man b&w 12k98mc main engine power 68,640 kw container capacity 8200 teu length over all 333.2 meter depth (molded) 24.2 meter breath (molded) 42.8 meter scantling draft (molded)/summer draft 14.5 meter gross tonnage 90,532 net tonnage 41,396 ton deadweight tonnage/d.w.t 103,235 ton main engine revolution per minute 94 rpm service speed 25 knot 4. result and discussion the experiment was carried out using a practical ship that sailed across the pacific ocean to the atlantic ocean. first, through satellite communication, the digital device installed on the board timely transferred information to the hq, and a big data analysis was performed in the ship. the ship’s route (fig. 2) was from taiwan, hong kong, china to the east coast of the usa throughout panama canal, back and forth. the ship traveled a total distance of 55,000 nautical miles with four voyages during the period. however, temporary disconnection of satellite communication on the vessel was sometimes unavoidable due to unpredictable bad weather or poor transmission at sea. there was a small broken route during the period, as shown in fig. 2, to express bad communication in the vessel. however, the temporary disconnection will not affect the experiment since the data can be saved and transferred to the hq after the communication is restored. fig. 2 electronic chart of 2d route illustrating the trace of the ship a combined statistical distribution of the relative wind speed and direction is presented in fig. 3. the upper fig. 3(a) illustrates the frequency statistics of the relative wind speed at each section; the occurrences from 0 to 60 knots is about 3,000 times at 0 knots, 1,500 times at 10 knots, and 1,000 times at 20 knots, descending sequentially. the lower fig. 3(a) displays the frequency statistics of the wind direction angle, from −180 degrees to +180 degrees. the wind direction angle is 0 degree at the bow, 0 to +180 degrees on the starboard side, and 0 to −180 degrees on the port side. compared with the occurrences of wind speed, the wind direction angle appears denser. except for 180 degrees at 1200 times, the average angle is about 200 times. therefore, the wind direction angle is denser than the relative wind speed. the fig. 3(b) shows the combined distribution of the upper and lower fig. 3(a) diagrams, where blue represents the relative wind direction and red represents the relative wind speed. fig. 4 is a distribution of the relative wind direction and relative wind speed for fuel consumption during the ship navigation. the lateral coordinate presents relative wind speeds, and the longitudinal coordinate displays a relative wind direction. when the ship was heading at the 0-degree course, the positive wind direction came from the bow, whereby the bow advances in technology innovation, vol. 4, no. 3, 2019, pp. 165-176 169 is represented by a 0-degree angle. when the ship hull area was larger, the resistance was greater, which was caused by the positive wind direction. in addition, the color bar on the right of fig. 4 indicates the amount of fuel consumption. the red color indicates large fuel consumption, which sequentially decreases in the downward direction. the blue area presents a small fuel consumption. the wind direction was large, and the relative wind speed was fast; thus, the fuel consumption was relatively high, as indicated by the increase in the red indicator. due to the large area of the hull, the sway was increased during the voyage. once the voyage was increased, the resistance was also large and high fuel consumption was possible. (a) statistics of wind speed and wind direction (b) frequency between wind direction and wind speed fig. 3 distribution of statistics and frequency for wind speed vs. wind direction, respectively fig. 4 comparison of relative wind direction (degree) and speed vs. main engine fuel consumption fig. 5 illustrates the relationships between the speed through water (stw) as well as the sog and the main engine (me) power, main shaft speed, and the relative wind speed. the upper left (a) and upper right (b) figures show the relationships between sog and me power and main shaft speed, respectively. the lower left (c) and right diagrams (d) show the relationships between the stw and the me power and main shaft speed, respectively. the color bar on the right of the figure indicates the relative wind speed. the stronger the red, the weaker the blue. as a result, the ship’s speed and wind speed increase and vice versa. fig. 6 presents cross views of a combined statistical distribution of the me power and stw frequencies. the frequency statistics of the me power at 0~38,000 kw is shown in the upper fig. 6(a). the frequency of the me was up to 46,000 kw, which is used for low-speed operations, and the captain relied on the power; the other different power times were between 1 and 30; this range cannot be clearly displayed due to the limitation of the unit scale in the chart. in the lower fig. 6(a), the occurrences of the stw is shown. the figure shows that the ship’s speed had the second-highest frequency when it was at 0 knots, but the highest frequency was around 18 knots, which also means that the ship’s speed was economical. at the fig. 6(b), the me power is integrated with the stw. as a result, the blue color indicates the stw and the red indicates the me power. on the other hand, the sog depends on the ocean currents and weather at sea and indirectly affects the me output power. advances in technology innovation, vol. 4, no. 3, 2019, pp. 165-176 170 (a) statistics of me power and stw (b) frequency between stw and me power fig. 6 distribution of statistics and frequency for me power vs. stw, respectively fig. 7 shows an exponential function or a polynomial distribution for the relationship between the stw and me power, where the right color bar is red at maximum fuel consumption and blue at minimum fuel consumption. when the stw on the abscissa is higher than 20 knots, the fuel consumption increases in the red or even gray area. both the horizontal and vertical coordinates are exponential functions. among them, the horizontal stripe appearing on the surface indicates that the navigation at this stage was affected by ocean currents. in addition, radio noise sometimes interfered with the data collection on the ship. because the big data hardware was installed in the electrical room, it might have been affected by a heat wave, humidity, and temperature; however, these data can be ignored. fig. 8 illustrates the relationship between the stw and me power. the bar on the right shows the sog. the red indicates the ahead speed of the ship. the higher the speed, the faster the ship. the sog is the actual sailing distance, which is affected by the influence of tidal current, wave, and climate on the water speed. sometimes, the sog was higher than the stw, but the ship’s speed was reduced when it was affected by the resistance of the current. for example, when the sog was in the 17~24 knots, the stw was about 20 knots, which means that the forward or reverse flow or the wind resistance affected the speed of the ship. (a) the relationship between sog and me power (b) the relationship between sog and rpm (c) the relationship between stw and me power (d) the relationship between stw and rpm fig. 5 distribution of speed, power profile and wind speed advances in technology innovation, vol. 4, no. 3, 2019, pp. 165-176 171 fig. 7 a profile illustration involving stw, me power, and fuel consumption fig. 8 a distribution of stw and me power vs. sog fig. 9 illustrates the relationship between the engine speed, me power, and fuel consumption. the right side is a color bar indicating fuel consumption, where red is the highest fuel quantity, which is decremented in turn, and blue is the minimum fuel quantity; the faster the shaft speed, the more the fuel consumption. the figure shows that fuel consumption conditions were normal when the main shaft speed was about 40~60 revolutions per minute (rpm). once the speed exceeded 80 rpm, the fuel consumption was very high, and it falls in the red area or even the gray stage in the figure. from the experiment, it can be concluded that the ship operation was usually under 75% me power and energy was saved. in addition, the maneuvering system was controlled by a governor, which could stabilize the speed of the main shaft. when the ship navigation was affected by the external climate and ocean current, the system sent a feedback signal to adjust the fuel injection to maintain a stable speed of the main shaft. at this moment, the me power will change accordingly. therefore, the profile of me power in fig. 9 is a fluctuation against environmental variation. fig. 10 illustrates the analysis of stw and me power versus the speed of the main shaft, where stw is dependent on the power level of the me. the color pattern on the right represents the main shaft speed. when the stw was high, the me ran at high power, and the fuel consumption increased accordingly. fig. 9 distribution of engine speed and me power vs. foc fig. 10 profile of stw, me power, and shaft speed (rpm) fig. 11 presents the cross views of a combined statistical distribution of the engine speed and me power frequencies. the upper fig. 11(a) depicts the frequency statistics of the engine speed, of which 40 rpm had the highest frequency, about 4,000 times for this ship, due to the needs for special sea detail and along the coastline. moreover, the engine speed can reduce the emission pollution which is used the speed of the same type of ship. second, the speeds were 55, 65, 70, and 75 rpm. furthermore, the lower fig. 11(a) shows the peak frequency, which also conforms to a power of 40 rpm. the fig. 11(b) indicates the frequency statistics of the two parameters, in which the blue color represents the me power and the red color represents the main shaft speed (rpm). both frequencies are integrated into the vertical axis, and the highest peak can be observed as 40 rpm. advances in technology innovation, vol. 4, no. 3, 2019, pp. 165-176 172 (a) statistics of engine speed (rpm) and me power (b) frequency between me power and engine speed (rpm) fig. 11 distribution of statistics and frequency for engine speed vs. me power, respectively fig. 12 presents the relationship between the average draft, trim, and fuel consumption. the right color bar indicates the fuel consumption measured in metric tons. it can be seen that the draft was below 10.5 m. the fuel consumption was still high even if the trim by the stern was between 0 and 0.5 m. the average draft and the load are significant parameters that affect fuel consumption. in addition, improper control of the trim will also increase the ship’s resistance. the trim by the stern indicates that the bow is lifted, and the trim by the head indicates that the bow is sinking. the data show that the bow was always sinking in the containership. from the experiment results, it is recommended that the ship should utilize the ballast water to deliver to the stern section that increases the draft of the aft. fig. 13 illustrates the relationship between the average draft, stw, and me power. the coordinate is the stw and the ordinate is the me power. the right color bar represents the draft; red indicates the draft at deep depth, while blue indicates the draft at shallow depth. the results show that the stw is related to the me power, and the relationship features a curve phenomenon. by statistics, the 12-meter draft was at a speed of over 18 knots. fig. 12 profile of average draft and trim with respect to me fuel consumption fig. 13 profile of stw and me power with respect to average draft fig. 14 presents the cross views of a combined statistical distribution of the average draft and trim frequencies. the frequency statistics of the average draft is depicted in the upper fig. 14(a). the plot shows that the draft depth of 12 m was the most used in the navigation, followed by the adjacency of 11.8 m, and then the depth of about 10 m. the lower fig. 14 (a) graph shows the statistical result of the trim. when the trim is 0 m, the frequency is about 8000 times, and the sequence order is −0.3, −1, and −1.2 m, which can show the tendency at the trim of the ship. as a result, the trim by the head occurred more times than the trim by the stern. the fig. 14(b) shows the distribution of the average draft and trim; blue represents trim and yellow represents the average draft. fig. 15 displays the relationships between the trim and the stw and me power. the result is similar to the previous figure and features an exponential function relationship. the right of the figure shows the height of the trim; where the red color represents the trim by the stern, and the blue represents the trim by the head. as a result, the statistics show that the trim fell to advances in technology innovation, vol. 4, no. 3, 2019, pp. 165-176 173 near 0 m and the frequency of trim by the head was much more than trim by the stern. in addition, fig. 16 shows the distribution of the calculated eeoi statistic. (a) statistics of average draft and trim (b) frequency between average draft and trim fig. 14 distribution of statistics and frequency for average draft vs. trim, respectively fig. 15 profile of stw and me power with respect to trim fig. 16 shows the relationship between the trim, avg. draft, foc, sailing distance, cargo mass, and eeoi. the figure is divided into five operational windows: a, b, c, d, and e. the trim is the difference between the aft draft and the fore draft, and the avg. the draft is an average of the aft draft and fore draft. foc means fuel consumption. the sailing distance is the distance traveled in nautical miles. moreover, sog is the real speed displayed on the gps, which is the vessel speed relative to the earth surface. cargo mass expresses the carrier capability of the ship. the eeoi represents a metric unit of co2 per nautical mile and cargo mass, and it is an indication of co2 emission quantity. as a result, the trim is dependent on the height difference between the fore draft and aft draft, which is one of the major factors that affect the energy-saving strategy. the trim was approximately at 0 m to 0.8 m, an acceptable range considering the various sailing conditions. despite the minor effect of the trim on the eeoi in the experiment, the avg. the draft was a major factor that affected the eeoi. consequently, the operational window of a indicates low draft around 10 m but high emission (eeoi). however, the average drafts around 12 m shown in windows b, c, d, and e correspond to a lower co2 emission than that of window a. thus, the energy saving phenomenon illustrates a quite remarkable and more saving of fuel oil at an average draft of 12 m. on the other hand, cargo mass is a function of draft, and both parameters interact with each other. furthermore, the ballast water is an alternative method to adjust the suitable draft depth. window a indicates light cargo mass, but fuel consumption is worsened and the eeoi increases. windows b, c, d, and e illustrate a heavy cargo mass, but low fuel consumption and low eeoi, which is in the range of 1~3 × 10 −5 . consequently, the cargo mass significantly affected energy efficiency. the eeoi variation for over six months was successfully investigated by statistical analysis of the sailing condition. advances in technology innovation, vol. 4, no. 3, 2019, pp. 165-176 174 fig. 16 eeoi calculation for various cargo masses, distances, focs, avg. drafts, and trims table 2 presents the quantitative intensity of emission index for each window of fig. 16 in the order of the eeoi. window b corresponds to the strategy with the least eeoi. window d also features a low eeoi but has the highest cargo mass than that of the condition if the captain chooses the option. window a corresponds to the worst energy-saving strategy; thus, the captain must avoid this operational condition. table 2 quantitative index for each window according to the eeoi window date duration eeoi × 10 −5 mass (ton) distance (nm) foc (mt) avg. draft (m) trim (m) b 4/3~4/22 1.50 50,803 1.19 0.19 11.86 −0.15 d 5/21~6/4 2.46 54,980 1.43 0.27 11.95 −0.79 e 6/9~7/8 2.17 47,239 1.21 0.22 11.65 −0.29 c 5/3~5/15 2.72 45,337 1.21 0.25 12.01 −0.23 a 2/28~3/18 5.01 27,254 1.38 0.28 10.06 −0.43 table 3 a quantitative result showing the differences and values from max. to min windows eeoi mass distance foc b 10.8% 22.5% 18.6% 15.8% d 17.7% 24.4% 22.2% 22.6% e 15.7% 20.9% 18.9% 18.2% c 19.6% 20.1% 18.9% 20.3% a 36.1% 12.1% 21.4% 23.1% differ 25.3% 12.3% 3.6% 7.3% value 3.51*10 -5 (t-co2/t-nm) 27,726 (ton) 0.23 (nm) 0.09 (ton) table 3 summarizes the analysis results of the percentage difference between the maximum and minimum energy efficiency parameters during different voyages. statistics for interval windows a, b, c, d, and e are presented. the eeoi difference is the amount of co2 emissions, and the eeoi value of each interval is divided by the amount of each interval; for example, for the b interval, the formula is (eeoi value of b)/(eeoi value of a + b + c + d + e). the value for each interval is taken, and then the maximum value is subtracted from the minimum value to obtain the difference in amount. the results show that eeoi could be reduced to 25.3%, which corresponds to a co2 emission reduction of 3.5 × 10 −5 . in the same way, the load capacity was increased by 12.3% or 27,726 tons; the travel distance was increased by 3.6% or 0.23 nautical miles, and fuel consumption was saved by 7.3% or 0.09 tons. therefore, adjusting the load and draft can reduce the co2 emissions and fuel consumption of the ship and provide positive guidance to ship decision makers. advances in technology innovation, vol. 4, no. 3, 2019, pp. 165-176 175 a forward-looking energy efficiency strategy for smart shipping through a big data analysis device is developing in this work. for various conditions of the vessel, the sailing mode can be selected based on an evaluation of energy saving and emission control. the variation of eeoi with the trim, avg. draft, foc, distance, cargo mass indicates a clear relationship. for this type of container ship, a draft between 11.5 m and 12 m is suitably deep to reduce fuel consumption. this energy-saving strategy can significantly save fuel oil and reduce the co2 emission quantity, as opposed to the traditional concept in the shipping industry. furthermore, both cargo mass and ballast technique always interact with the draft depth of the ship. a constructive model using iot technology and an energy efficiency management strategy presents an accurate big data basis to guide decision making on the vessel. the results showed that the eeoi (the co2 emission) was reduced by 25.3%, the ship could carry an additional cargo mass of 12.3%, the travel distance (nautical mile) was increased by 3.6%, and then fuel oil was saved by 7.3%. thus, the maritime industry can improve the energy saving and engine efficiency of ships using the eeoi-based model. acknowledgement this project is financially supported by the program “data analytics for smart shipping and marine energy management” of the ministry of science and technology, taiwan, r.o.c. under grant no. most 107-2218-e-006 -042. moreover, the authors thank enago (www.enago.tw) for providing professional english review services. conflicts of interest the authors declare no conflict of interest. references [1] m. d. a. al-falahi, k. s. nimma, s. d. g. jayasinghe, h. enshaei, and j. m. guerrero, “power management optimization of hybrid power systems in electric ferries,” energy conversion and management, vol. 172, pp. 50-66, september 2018. [2] l. p. perera and b. mo, “emission control based energy efficiency measures in ship operations,” applied ocean research, vol. 60, pp. 29-46, october 2016. [3] m. tichavska, b. tovar, d. gritsenko, l. johansson, and j. p. jalkanen, “air emissions from ships in port: does regulation make a difference?” transport policy, vol. 75, pp. 128-140, march 2019. [4] r. d. prasad and a. raturi, “fuel demand and emissions for maritime sector in fiji: current status and low-carbon strategies,” marine policy, vol. 102, pp. 40-50, april 2019. [5] k. lehtoranta, p. aakko-saksa, t. murtonen, h. vesala, l. ntziachristos, t. ronkko, et al., “particulate mass and nonvolatile particle number emissions from marine engines using low-sulfur fuels, natural gas, or scrubbers,” environmental science & technology, vol. 53, pp. 3315-3322, march 2019. [6] y. h. hou, “hull form uncertainty optimization design for minimum eeoi with influence of different speed perturbation types,” ocean engineering, vol. 140, pp. 66-72, august 2017. [7] t. a. tran, “a research on the energy efficiency operational indicator eeoi calculation tool on m/v nsu justice of vinic transportation company, vietnam,” journal of ocean engineering and science, vol. 2, pp. 55-60, march 2017. [8] shipbuilding — principal ship dimensions — terminology and definitions for computer applications, iso 7462, 1985. [9] h. j. shaw, w. g. teng, and f. m. tzu, “a study on energy efficiency management for smart ships,” 3rd international naval architecture and maritime symposium, proceedings, yıldız technical university, istanbul, turkey, pp. 657-668, 2018. [10] p. kaluza, a. kolzsch, m. t. gastner, and b. blasius, “the complex network of global cargo ship movements,” journal of the royal society interface, vol. 7, pp. 1093-1103, july 2010. [11] m. svindland, “the environmental effects of emission control area regulations on short sea shipping in northern europe: the case of container feeder vessels,” transportation research part d-transport and environment, vol. 61, pp. 423-430, june 2018. [12] l. zhen, m. li, z. hu, w. y. lv, and x. zhao, “the effects of emission control area regulations on cruise shipping,” transportation research part d-transport and environment, vol. 62, pp. 47-63, july 2018. advances in technology innovation, vol. 4, no. 3, 2019, pp. 165-176 176 [13] m. jurjevic, n. koboevic, and m. bosnjakovic, “modelling the stages of turbocharger dynamic reliability by application of exploitation experience,” tehnicki vjesnik-technical gazette, vol. 25, pp. 792-797, june 2018. [14] z. koboevic, d. bebic, and z. kurtela, “new approach to monitoring hull condition of ships as objective for selecting optimal docking period,” ships and offshore structures, vol. 14, pp. 95-103, january 2019. [15] “imo train the trainer (ttt) course on energy efficient ship operation, module 2–ship energy efficiency regulations and related guidelines,” printed and published by the international maritime organization, vol. 2, p. 42, 2016. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). template encit2010 advances in technology innovation, vol. 4, no. 2, 2019, pp. 116 124 the innovative design of the massage mechanisms for massage chair tzu-hsia chen 1* , long-chang hsieh 2 1 department of mechanical engineering, minghsin university of science and technology, taiwan, roc 2 department of power mechanical engineering, national formosa university, yunlin, taiwan, roc received 05 june 2017; received in revised form 09 july 2017; accepted 29 july 2017 abstract the purpose of this paper is to synthesize all the feasible designs of massage mechanisms (including dof=2 and 3) for the massage chair, based on the modified yan’s creative design methodology. first, the topological structure and motion characteristics of existing massage mechanisms are analyzed and its design requirements and constraints are specified mechanisms with beating and kneading functions are mechanisms with 2 degrees of freedom (dof). then, 14 design concepts are generated. one of the design concepts is curried cut, including kinematic design, engineering drawing, and the prototype manufacture. the simulation result showed that new design can produce a wider range of non-uniform output motion than the extiting design. keywords: beating massage, innovative design, kneading massage, massage-mechanism, yan’s creative design methodology 1. introduction due to busy lifestyles in modern days, insufficient time for sports leads to physical illness. recently, because of the rising economic level and health awareness, the concept of “health protection” is well-known. the massage chair is one of the most popular health products. the action of the massage chair is to mimic the massage therapist. with the progress of the society, the demands of the massage chair became more diversified. when people lie on the massage chair, they can enjoy the massage pressure to eliminate muscle fatigue, relieve pressure, and relax. in general, there are two different modes of the massage chairs, like beating massage and kneading massage, and both of them promote the body health. the massage mechanism can be a planar mechanism [1-2] or spatial mechanism [3-5]. this paper focuses on the systematic design of spatial massage mechanism which provides beating massage and kneading massage, including the concept design, the kinematic analysis, the engineering drawing, engineering design and manufacture of a prototype. first, we analyze the topological structure and motion characteristics of the existing massage mechanisms and conclude its design requirements and constraints. all existing massage mechanisms with beating and kneading functions are mechanisms with 2 degrees of freedom (dof). the purpose of this paper is to synthesize all the feasible designs of massage mechanisms (including dof =2 and dof =3). then, based on the design requirements and constraints also modified yan’s creative design methodology [6-11], we can synthesize all feasible design concepts of massage-mechanisms for the massage chair. therefore, one design concept is chosen for the engineering design and prototype manufacture. the beating and kneading massage traces of this new design is simulated by “cosmos” software to verify the new feasibility of design. 2. existing designs before implementing an innovative design, the structure of the massage mechanism must be collected and analyzed from academic papers, catalogues, and technical reports. based on the analysis results, the massage mechanism can be * corresponding author. e-mail address: summer34134@gmail.com tel.: +886-3-559-3142 ext. 3023; fax: +886-3-557-3797 advances in technology innovation, vol. 4, no. 2, 2019, pp. 116 124 2 classified as the planar mechanism or spatial mechanism. fig. 1 shows the patent [1] which is the planar mechanism and provides the kneading massage. fig. 1 improvement structure of massage chair [1] fig. 2 shows the patent [3] which is a spatial mechanism and provides the beating and kneading massages. fig. 2(c) shows its corresponding kinematic skeleton. according to fig. 2, the mechanism has 6 links and 6 joints (5 revolute pairs and 1 spherical pair) and belongs to the spatial mechanism. in the 1st input motion, link 2 provides a beating function and in the 2nd input motion, link 6 provides a kneading function. (a) beating massage (b) kneading massage (c) kinematic skeleton fig. 2 improved double drive system of massage chair [3] the mobility of the spatial mechanism can be obtained by following eq. (1). f 6( ) i n j i f    (1) where n is the number of links, j is number of joints, and fi is the degrees of freedom of joint i. the mechanism shown in fig. 2 has 6 links, 6 joints (5 revolute pairs and 1 spherical pair), according to the equation of mobility, we get: 6( ) 6(6 6 1) 8 2f n j i f i           (2) 3. creative mechanism design methodology the design concept is the initial stage of the engineering design process and also the most difficult part. all existing massage mechanisms with beating and kneading functions are mechanisms with 2 degrees of freedoms (dof). the purpose of this paper is to synthesize all the feasible designs of the massage mechanism (including dof =2 and dof =3). for the mechanisms with 3 dof, there must be 1 redundant degree of freedom. there is no existing mechanism dof =3) which can provide beating massage and kneading massage, therefore, yan’s creative design methodology [6-10] must be modified. fig. 3 shows the modification of yan’s creative design methodology, and the steps are as follows: advances in technology innovation, vol. 4, no. 2, 2019, pp. 116 124 3 fig. 3 modified yan’s creative design methodology [7] (1) identify the existing designs (dof =2) with required design specifications that designers would like to have, and conclude the topological characteristics of these designs. (2) according to the topological characteristics, identify the mobility of the mechanism and its corresponding link and joint types. (3) synthesize the atlas of generalized chains that can be used to design massage mechanism with beating and kneading massages for the massage chair. (4) assign types of members and joints to each generalized chain obtained in step 3, to have the atlas of feasible specialized chains based on the algorithm of specialization to meet needed design requirements and constraints. (5) particularize each feasible specialized chain obtained from step 4 to its corresponding kinematic skeleton, to have the atlas of massage mechanism with beating and kneading massages for the massage chair. 3.1. topological characteristics the first step of the modified creative design methodology is to define the design specifications of mechanical devices that design engineers would like to generate. if there is no special consideration, the degree of freedom of a mechanism is equal to the number of independent inputs for the constraint motion. if the degree of freedom is larger than the number of independent inputs, in general, the mechanism will have unconstraint motion. nevertheless, if the excess degree of freedom is the redundant degree of freedom, it will not affect the moving of other links, the mechanism will still have constraint motion and will still be useful. the purpose of this paper is to invent the massage mechanism with 2 and 3 degrees of freedoms (dof) to provide beating and kneading functions. in this paper, we only concern the mechanism with revolute and spherical pairs. according to eq. (1), if n=6, j=6, jr=5, and js=1, the corresponding generalized chain (6, 6) has 2 degrees of freedom. and, according to eq. (1), if n=5, j=5, jr =3, and js =2, the corresponding generalized chain (5, 5) has 3 degrees of freedom. the results are shown in table 1. advances in technology innovation, vol. 4, no. 2, 2019, pp. 116 124 4 table 1 taxonomy of swarm-based routing protocols degrees of freedom (f) number of links (n) number of joints (j) number of revolute pairs (jr) number of spherical pairs (js) 2 6 6 5 1 3 5 5 3 2 next, the massage mechanisms with beating and kneading functions should have the following topological characteristics: (1) it must be a spatial mechanism (l = 6). (2) it must have 2 input links and 1 output link (massage link). (3) it has a ground link to support or constrain other links. (4) the joints are constrained to be revolute and spherical pairs. (5) for the mechanism with 2 dof, it will have at least 1 spherical pair. (6) for the mechanism with 3 dof, it will have at least 2 spherical pairs. 3.2. atlas of generalized chain (a) (5, 5) generalized chain (b) (6, 6) generalized chain fig. 4 generalized chains for massage mechanism the second step of the modified creative design methodology is to synthesize all possible generalized chains which can be used to synthesize the desired mechanism. there is only 1 generalized chain with 5 links and 5 joints and only 1 generalized chain with 6 links and 6 joints. fig. 4 shows the generalized chains for massage mechanisms. 3.3. design requirements and design constraints design requirements and constraints are determined based on the concluded topological structures. the design requirements and constraints of massage mechanisms for massage chair are: (1) there must be a ground link (gr), first input link for kneading function (i1), second input link for beating function (i2), output link (massage link) (om). (2) the ground link (gr) must be adjacent to 2 input links (i1 and i2) with revolute pairs. (3) the ground link (gr) cannot be incident to spherical pair. (4) the output link (om) can’t be incident to 2 spherical pairs at the same time. (5) for the mechanism with 2 dof, (6, 6) generalized chain must have 5 joints (jr) and 1 spherical pair (js). (6) for the mechanism with 2 dof, (5, 5) generalized chain must have 3 joints (jr) and 2 spherical pairs (jr). (7) for the mechanism with 3 dof, 2 spherical pairs must be adjacent and cause 1 redundant degree of freedom. 3.4. specialization the third step of the modified creative design methodology is to assign specific types of members and joints to each available kinematic chain, subject to certain design requirements to have specialized chains. the specializing steps for massage mechanism are: advances in technology innovation, vol. 4, no. 2, 2019, pp. 116 124 5 (1) for each generalized chain, identify the ground link (gr) for all possible cases. there are 2 possible identifications as shown in figs. 5(a) and 5(b). (2) for each case obtained in step 1, identify the first input link (kneading function) (i1) and the second input link (beating function) (i2). for the specialized chains as shown in figs. 5(a) and 5(b), based on the design requirements, there are 2 possible identifications show in figs. 6(a) and 6(b). (3) for each case obtained in step 2, identify the output link (massage link om). for the specialized chains shown in figs. 5 (a) and 5(b), based on the design requirements, there are 5 possible identifications show in figs. 7(a)-7(e) (4) for each case obtained in step 3, identify the corresponding revolute pairs (denoted by ○) and spherical pairs (denoted by ●). for (5, 5) generalized chain, there are 2 feasible specialized chains shown in figs. 8(a) and 8(b). for (6, 6) generalized chain, there are 12 feasible specialized chains shown in figs. 9(a)-9(e). (a) (b) fig. 5 identify ground link (gr) (a) (b) fig. 6 identify first input link (kneading function) (i1) and second input link (beating function) (i2) (a) (b) (c) (d) (e) fig. 7 identify output link (massage link) (om) (a) (b) fig. 8 atlas of feasible specialized chain of (5, 5) generalized chain advances in technology innovation, vol. 4, no. 2, 2019, pp. 116 124 6 (a) (b) (c) (d) (e) (f) (g) (i) (j) (k) (l) (m) fig. 9 atlas of feasible specialized chain of (6, 6) generalized chain 3.5. particularization (a) (b) fig. 10 atlas of feasible massage mechanisms of (5, 5) generalized chain (a) (b) (c) (d) (e) (f) (g) (h) (i) (j) (k) (l) fig. 11 atlas of feasible designs for massage mechanism of (6, 6) generalized chain advances in technology innovation, vol. 4, no. 2, 2019, pp. 116 124 7 for each feasible specialized chain, it can be particularized into its corresponding kinematic skeleton. particularization is the reverse process of generalization and can be done by applying the generalizing rules in reverse order. figs. 10(a) and 10(b) show 2 feasible massage mechanisms of (5, 5) generalized chain and figs. 11(a)-11(l) show 12 feasible massage mechanisms of (6, 6) generalized chain. the design concept, shown fig. 11(a), is the same as the existing design in fig. 2. therefore, only 11 new designs are synthesized from (6, 6) generalized chain and 2 new designs are synthesized from (5, 5) generalized chain. in this paper, we synthesize 13 new designs of massage mechanisms for the massage chair. 4. engineering design and dynamic simulation after an innovative design, the next step is engineering design. due to the reason of manufacture cost, fig.10 (a) is selected as a design example to carry out, and its solid model is drawn and shown in fig. 12. fig. 13 shows its trace which is simulated by “cosmos”. table 2 shows the comparisons of existing design and new design. according to table 2, we get that new design has larger massage trace than the existing design. if we adjust the initial phase of the kneading input, the range of massage trace will become smaller. table 2 also shows that if the initial phase of the kneading input is increased 7⁰, the corresponding kneading trace is almost close to the existing design. fig. 14 ~16 shows the traces of the new design, respectly. fig. 12 engineering drawing of the design concept shown in fig. 10(a) (a) projection on xy plane (b) projection on xz plane fig. 13 kneading trace of upper roller table 2 the comparisons of existing design and new design kneading massage dispalcement location existing design (initial phase 0) new design (initial phase 0) new design (initial phase 7 of kneading input) x upper roller 2(mm) 2(mm) 2(mm) lower roller 2(mm) 8(mm) 2(mm) y upper roller 7(mm) 11(mm) 7(mm) lower roller 7(mm) 11(mm) 7(mm) z upper roller 21(mm) 41(mm) 22(mm) lower roller 33(mm) 50(mm) 33(mm) beating massage x upper roller 2(mm) 2(mm) lower roller 2(mm) 8(mm) y upper roller 10(mm) 11(mm) lower roller 10(mm) 11(mm) z upper roller 1(mm) 1(mm) lower roller 1(mm) 1(mm) ｘ-ｙ 平面軌跡圖 126 128 130 132 134 136 138 140 179 181 183 185 187 189 191 193 ｘ ｙ ｚ-ｙ平面軌跡圖 108 113 118 123 128 133 138 143 148 153 158 -85 -80 -75 -70 -65 -60 -55 -50 -45 -40 -35 z y advances in technology innovation, vol. 4, no. 2, 2019, pp. 116 124 8 (a) projection on xy plane (b) projection on xz plane fig. 14 kneading trace of lower roller (a) projection on xy plane (b) projection on xz plane fig. 15 beating trace of upper roller (a) projection on xy plane (b) projection on xz plane fig. 16 beating trace of lower roller 5. conclusions in this paper, the new designs of massage mechanisms have been generated by the systematic design methodology. first, the design requirements and design constraints are summarized based on existing designs. then, according to modified yan’s design methodology, 13 new design concepts synthesized. one of the new design concepts is selected as a design example and verified by kinematic simulation. the simulation result showed that the new design can produced a wider range of non-uniform output motion than the existing design. conflicts of interest the authors declare no conflict of interest. x-y 平面軌跡圖 6 8 10 12 14 16 18 20 180 182 184 186 188 190 192 194 x y z-y 平面軌跡圖 -20 -10 0 10 20 30 40 -90 -80 -70 -60 -50 -40 -30 z y x-y 平面軌跡圖 128 130 132 134 136 138 140 142 179 181 183 185 187 189 191 193 x y z-y 平面軌跡圖 128 130 132 134 136 138 140 142 -66 -64 -62 -60 -58 -56 -54 -52 z y x-y 平面軌跡圖 8 10 12 14 16 18 20 22 180 182 184 186 188 190 192 194 x y z-y 平面軌跡圖 8 10 12 14 16 18 20 22 -81 -79 -77 -75 -73 -71 -69 -67 z y advances in technology innovation, vol. 4, no. 2, 2019, pp. 116 124 9 references [1] h. c. chen, improvement of massage chair structure, roc patent, m268028, june 21, 2005. [2] r. c. bai, massage chair with 3d kneading, beating and pressing functions, roc patent, m518561, march 11, 2016. [3] h. j. chang, improved double drive system of massage chair, roc patent, m289645, april 21, 2006. [4] k. b. chen and m. g. fang, massaging apparatus for massage chair, roc patent, i542341, july 21, 2016. [5] r. y. chen, massaging apparatus for massage chair, roc patent, i547273, september 01, 2016. [6] h. s. yan and l. c. hsieh, “conceptual design of gear differentials for automotive vehicles,” journal of mechanical design, vol. 116, no. 2, pp. 565-570, 1994. [7] h. s. yan, “creative design of mechanical devices,” singapore; new york: springer, 1998. [8] h. s. yan, “a methodology for creative mechanism design,” mechanism and machine theory, vol. 27, no. 3, pp. 235242, may. 1992. [9] l. c. hsieh and t. h. chen, “the systematic design of link-type optical fiber polisher with single flat,” journal of advanced science letters, vol. 9, no. 1, pp. 318-324, april 2012. [10] l. c. hsieh, t. h. chen, and s. j. wei, “the innovative design of wheelchair with lifting and standing functions,” proceeding of engineering and technology innovation, vol. 4, pp. 10-12, 2016. [11] w.h. hsieh and s.j. chen, “innovative design of cam-controlled planetary gear trains,” international journal of engineering and technology innovation, vol. 1, no. 1, pp. 01-11, 2011. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation, vol. 4, no. 3, 2019, pp. 125-139 photovoltaic power control using fuzzy logic and fuzzy logic type 2 mppt algorithms and buck converter bachar meryem * , naddami ahmed, fahli ahmed department of electrical engineering, hassan first university, settat, morocco received 13 march 2019; received in revised form 22 april 2019; accepted 25 may 2019 abstract this work presents the analysis, design, simulation and hardware implementation of the classical fuzzy logic (fl) and the proposed fuzzy logic type 2 (flt2) mppt techniques for standalone pv system. fl and flt2 mppt algorithms are simulated via matlab/ simulink and implemented via labview software and compactrio hardware, in different climatic conditions. also, they are compared to the incremental conductance (inc) mppt algorithm, one of the most common used mppt techniques. the studied system consists of pv array, dc/dc converter, mppt controller, batteries and load. the pv array is connected to the dc / dc buck converter that works based on the output pulses of the mppt controller to make the pv system operates at the maximum power point (mpp). thereafter, based on the simulations and the experimental results, a comparison is made to be useful for mppt designers and researchers in this area. keywords: photovoltaic panel, fuzzy logic, fuzzy logic type 2, matlab/simulink 1. introduction solar energy is a renewable, non-polluting and economical source of energy which allows obtaining electricity from the solar irradiation using photovoltaic (pv) cells [1, 2]. despite its advantages, solar energy has a remarkable disadvantage. one of the most relevant problems is that solar energy is intermittent. therefore, the power produced by pv panels is influenced by the climatic conditions (irradiation and temperature) and the load impedance [3, 4]. generally, the intersection of the load and pv panel characteristics is too far from the mpp, thus, it is important to insert a dc/dc converter, between the load and the pv source, for the impedance matching [5]. maximum power point tracking (mppt) algorithms are used for the pursuit of mpp in different climatic conditions [6], by adjusting the duty cycle of the dc/dc converter. generally, the system consists of a pv array, a dc/dc converter, a load and finally batteries. nowadays, several techniques of mppt exist such as perturb and observe (p&o) [7, 8], incremental conductance (inc) [5, 9], constant voltage (cv), open circuit voltage (ocv), and fuzzy logic (fl) [10-12]. mppt algorithms can be distinguished by operating principle, performance, complexity, response time, cost and more. the classic fuzzy logic (fl), introduced in 1965 by lotfi zadeh, allows the representation and processing of imprecise knowledge based on linguistic terms. it relies on human reasoning to convert a linguistic command into an automatic command to control complex systems [13, 14]. fuzzy type 2 (flt2) [15], the proposed mppt technique, is based on fl. it supports uncertainties using three-dimensional membership functions. * corresponding author. e-mail address: meryem.bachar@gmail.com tel.: +212523492455; fax: +212(0)523490354 advances in technology innovation, vol. 4, no. 3, 2019, pp. 125-139 126 in this work, the authors present a comparative study of three mppt algorithms (fuzzy logic, fuzzy logic type 2 and incremental conductance). matlab/simulink environment is used for simulation studies and labview software for experimental tests. this paper is organized as follows: section ii explains the operating principle of a photovoltaic cell and gives the i-v and p-v characteristics of the studied pv array. section iii deals with dc/dc converters and focuses on the dc/dc buck converter, used in this paper. section iv studies the role of mppt algorithms and explain fl and flt2 mppt algorithms. section v presents the simulation results of the photovoltaic system with fl, flt2, and inc algorithms and gives a comparison of them. finally, section vi presents the experimental results of the three studied algorithms by using labview software and compactrio hardware. 2. photovoltaic module in order to study the behaviour of photovoltaic cell, we can model it by an electric circuit, as shown in fig. 1. pv cell usually consists of an iph current generator, which models the conversion of light radiation into electricity, a diode d which represents the pn junction, a parallel resistor rsh and a series resistor rs [16, 17]. fig. 1 electric circuit of pv cell the current generated by the photovoltaic cell is calculated by eq. (1): ph d sh i i i i   (1) eq. (1) can be rewritten by eq. (2) [18]: 0 s sh sh v ir rq s ph v ir i i i e r                     (2) the terms iph, i0, q, a, k, t are the photodiode current, the inverse saturation current, the electron charge, the ideality factor of the pn junction, the boltzmann constant and the temperature of pv cell, respectively. 3. dc/dc buck converter fig. 3 dc/dc buck converter a dc/dc converter is used to convert a dc input voltage to a modified dc output voltage [20]. currently, several dc/dc converters exist such as: buck, boost, buck-boost and full bridge [21, 22]. in this paper, the dc/dc buck converter, advances in technology innovation, vol. 4, no. 3, 2019, pp. 125-139 127 which allows for a lowered voltage, is used. it usually consists of a switch, a diode, a capacitor, and an inductance, as shown in fig. 2 [23]. when the switch s is closed, during the period αt, the diode d is blocked and the voltage across the converter is given by: l e s v v v  (3) when the switch s is open, during the period (1-α) t, the diode is on and the voltage across the converter is given by: l e v v  (4) the voltage passing through the inductance l is given by the following relation: l l di v l dt  (5) the relationship between the input voltage and the output voltage of the buck converter is given in eq. (6): s e v v (6) where α is the duty cycle with 0 <α <1. from eq. (7), we can rewrite the inductance as follows:   1 l l e s s r di l v v v dt f i     (7) the value of the capacity is given by the eq. (8):   r s r r i f c v i esr    (8) the terms fs, ir, vr and esr are switching frequency, ripple current, ripple voltage and effective series resistance, respectively. the switch s is a mosfet controlled by an mppt controller which will be detailed in the following part. 4. maximum power point tracking algorithms (mppt) climate changes during the day influence the power produced by the pv panel. therefore, the operating point does not intersect with the mpp, which causes a loss of power. it is then essential to extract the novel mpp using mppt algorithms. they are used to vary the equivalent resistance of the load to extract the maximum power by automatically varying the duty cycle of the dc/dc converter [10]. nowadays, several methods of mppt exist and can be classified by their tracking techniques to: (1) methods with constant parameters such as constant voltage [24], open circuit voltage [25] and short circuit voltage [26]. (2) methods with trial and error such as the only-current photovoltaic, perturb and observe [28] and dc-link capacitor drop [29]. (3) methods with mathematical calculation such as curve fitting [30], differentiation method and incremental conductance [31, 32]. (4) methods with intelligent prediction such as fuzzy logic control and neural network [33]. in this work, the authors compare two fuzzy logic controls (classic fuzzy logic (fl) and fuzzy logic type 2 (flt2)) with the incremental conductance, in different cases of irradiation. advances in technology innovation, vol. 4, no. 3, 2019, pp. 125-139 128 4.1. classical fuzzy logic (fl) the fuzzy logic algorithm is an intelligent technique. it does not require exact knowledge of the photovoltaic system, which makes its use simple. in addition, it is a robust and powerful technique [34]. a fuzzy controller typically consists of four parts: fuzzification, inference, rule base, and defuzzification, as shown in fig. 4. fig. 3 the fuzzy logic controller diagram in this paper, the inputs of the fl controller are the error and the change of error derror. they are given by the eqs. (9) and (10):           1 1 p k p k error k v k v k      (9) ( ) ( ) ( 1)derror k error k error k   (10) where k, p, and v are the sampling time, the pv panel power and the pv panel voltage, respectively. in the fuzzification stage, the digital inputs are converted into seven linguistic variables that are: positive big (pb), positive medium (pm), positive small (ps), zero (ze), negative small (ns), negative medium (nm), and negative big (nb). fig. 4 and fig. 5 present the linguistic variables of the error and derror. fig. 4 the input of fuzzy logic controller (error) fig. 5 the input of fuzzy logic controller (derror) it is important to understand how the system works to create the rules. in this work, the inference engine based on mamdami method applies 49 rules in the form of if-then as explained in table 1. advances in technology innovation, vol. 4, no. 3, 2019, pp. 125-139 129 table 1 fuzzy logic rule base error derror pb pb pm ps ze ns nm pm ze ze ze nb nb nb ps ze ze ze nm nm nm ze ze ze ze ns ns nm ns ns ns ze ze ze ps nm pm pm ps ns ze ps nb pm pm pm pb ze ze based on these rules, the change in the duty cycle is calculated. the output, as presented in fig. 6, has 7 levels: positive big (pb), positive medium (pm), positive small (ps), zero (ze), negative small (ns), negative medium (nm), and negative big (nb). in the defuzzification stage, the output is converted to a numerical variable to provide an analog signal. thus, the duty cycle of dc/dc buck converter is varied to reach the mpp. fig. 6 the output of fuzzy logic controller (duty cycle) 4.2. fuzzy logic type 2 (flt2) fuzzy logic type 2 (flt2), noted  , models uncertainty better than conventional fuzzy logic. it is characterized by a three-dimensional membership function ( , )u x y  . flt2 can be expressed by eq. (11): ( , ) , xx x u j u x y u x        ,  0,1xj  (11) where x, jx, and ∬ are the primary variable, the primary membership of x and the union of all the cartesian product elements on x, respectively. each three-dimensional membership function has superior and inferior membership functions represented by classical fuzzy logic. the interval between the inferior and superior membership functions is called footprint of uncertainty (fou). it is the new third dimension of the flt2 which gives more precision, compared to the classical fuzzy logic [35]. the block diagram of the flt2, presented in fig. 7, contains five parts: fuzzification, inference engine, rule base, reducer type, and defuzzification. fig. 7 the fuzzy logic type 2 controller diagram advances in technology innovation, vol. 4, no. 3, 2019, pp. 125-139 130 in the fuzzification stage, the two digital inputs (error and derror) are converted into 3 linguistic variables which are: positive (p), zero (z), and negative (n). each variable has two levels: lower u and higher l. the variables then become (pl, pu); (zl, zu), and (nl, nu). the inputs used in this paper are shown in fig. 8 and fig. 9. fig. 8 membership function for error fig. 9 membership function for derror the structure of the rules remains exactly the same in the case of fl algorithm. table 2 summarizes the rules used in the employed flt2 controller. table 2 the rule table of flt2 controller error derror p z n p pb pm z z nm z nm n z n n the inference engine combines fuzzy rules to perform a transformation from fuzzy sets in the input space to fuzzy sets in the output space. there are several methods of inference such as the max-min inference method (mamdami), the max-product inference method (larsen) and the and inference method (sugeno). in this paper, the sugeno method is used. the output of the inference engine block is equal to the weighted average of the output of each fuzzy rule. the flt2 controller differs from the fl by the output processing module which consists, in this case, of two blocks: type reducer and defuzzification. for an flt2 system, each output set of a rule is type 2. consequently, the type reducer is used to get a classic set from the type 2 output sets. the classic set obtained from the type reducer is converted into a well-defined numerical value, in the defuzzification step, to control the dc/dc buck converter. as numerical value ca of the output, it can be obtained by the eq. (12): 1 1 ( ) ( ) ( ) n k kà k a n kà k y u y c x u y      (12) where n, y, and uӑ are, respectively, the number of rules, the output, and the membership function. the simulation results of the two algorithms studied will be presented in the following section. in addition, they will be compared to the inc algorithm. 5. simulation results in this part, the simulation results, under matlab/simulink, are presented. the system, shown in fig. 11, is composed of six tesla solar modules (solar ts250-p150-60), a dc/dc buck converter, mppt controller and four series batteries (48v-165ah). the pv panel characteristics are presented in table 3. advances in technology innovation, vol. 4, no. 3, 2019, pp. 125-139 131 table 3 the specifications of the pv panel parameter value pm 255 w imp 8.32 a vmp 30.6 v isc 8.75 a vsc 37.6 v rsh 529.6538 ω rs 0.29192 ω module efficiency 15.58% fl, flt2, and inc mppt algorithms are tested via the matlab/simulink environment in the case of stable and variable irradiation. fig. 10 the global system in matlab/simulink 5.1. uniform irradiation the simulations of the pv system with the fl, flt2, and inc algorithms are done under uniform irradiation (880w/m²) and fixed temperature (26°c). the voltage and the current of the pv array are used as inputs to calculate the error and the variation of error that represent the input of the fl an flt2 controllers as shown in fig. 11 and fig. 12. they are designed following the steps in the previous section. fig. 11 fl algorithm in matlab/simulink advances in technology innovation, vol. 4, no. 3, 2019, pp. 125-139 132 fig. 12 flt2 algorithm in matlab/simulink fig. 13 the pv voltage with inc, fl, and flt2 mppt algorithms fig. 14 the pv current with inc, fl, and flt2 mppt algorithms fig. 15 the pv power with inc, fl, and flt2 mppt algorithms fig. 13, fig. 14, and fig. 15 present pv voltage, pv current and pv power, respectively. fig. 15 shows that the power obtained by inc oscillates between 1265 and 1272w and the mpp is reached at t=0.25s. with fl, the power is equal to 1300w at time t=0.17s. it reaches the mpp and remains stable. by using an flt2 controller the mpp is reached at time t=0.08s with advances in technology innovation, vol. 4, no. 3, 2019, pp. 125-139 133 remarkable stability. the maximum power is 1335w, which offers better results than those of the fl and inc controllers in terms of speed and output power. fig. 16 shows the output voltage of the dc/dc buck converter. fig. 16 the output voltage of buck converter with inc, fl, and flt2 mppt algorithms 5.2. non-uniform irradiation to test the performance of the three algorithms in different climatic conditions, non-uniform irradiation is applied to the input of the pv array. the irradiation takes 880w/m 2 at the beginning of the simulations. after t=3.2 s, it rapidly decreases to 310w/m 2 and returns to the initial state. fig. 17, fig. 18, and fig. 19 show the voltage, current, and power generated from the pv array with the three algorithms (flt2, fl, inc) in the case of variable irradiation. after the disturbance, the flt2 controller shows a better performance in term of response time, stability and generated power. fig. 17 the pv voltage with inc, fl, and flt2 mppt algorithms fig. 18 the pv current with inc, fl, and flt2 mppt algorithms advances in technology innovation, vol. 4, no. 3, 2019, pp. 125-139 134 fig. 19 the pv voltage with inc, fl, and flt2 mppt algorithms fig. 20 shows the variation of the output voltage of the buck converter in the case of variable irradiation. fig. 20 the output voltage of buck converter with inc, fl, and flt2 mppt algorithms 6. experimental results in order to verify the real working of the studied mppt algorithms, a hardware setup was implemented, as shown in fig. 21. fig. 21 experimental setup with the pv array the setup consists of six tesla solar modules (solar ts250-p150-60), a dc/dc buck converter, compactrio controller, national instrument modules, irradiation sensor (irrb2 thies), temperature sensor (lm35) and load. for voltage measurement, ni 9225 (analog input module, 300vrms) is connected in parallel with the pv array. ni 9247 (analog input module, 50 arms) is connected in series with the pv array to sense the current. all sensed data is given to the national instrument compactrio ni 9025, an embedded real-time controller. it controls the mosfet of the dc/dc buck converter using ni 9474. the studied algorithms are programmed and implemented using labview software. labview is a graphical software from ni for system design, measurement, and control. the choice of this software is based on the ease of implementation of the program in the ni hardware as well as its simple interface of programming and use. advances in technology innovation, vol. 4, no. 3, 2019, pp. 125-139 135 to read the experimental results, the front panel of labview software is used. the observed parameters, collected from the compactrio are irradiation, temperature, pv current, pv voltage, pv power and buck voltage, as presented in fig. 22. fl, flt2, and inc are tested via labview environment in the case of stable and variable irradiation. those experiments were done on may 2, 2019, on a sunny day. the measured parameters are shown in the figures below. fig. 22 labview front panel of the mppt algorithms 6.1. uniform irradiation the experiment of the pv system with the fl, flt2, and inc algorithms are done under uniform irradiation (880w/m²) and fixed temperature (26°c). fig. 23 experimental results of inc mppt algorithm at stable climatic conditions fig. 24 experimental results of fl mppt algorithm at stable climatic conditions fig. 25 experimental results of flt2 mppt algorithms at stable climatic conditions advances in technology innovation, vol. 4, no. 3, 2019, pp. 125-139 136 fig. 23, fig. 24 and, fig. 25 present inc, fl and flt2 experimental results, respectively. fig. 23 shows that the power obtained by inc oscillates between 1217. with fl, the power is equal to 1271w. it reaches the mpp and remains stable. by using an flt2 controller, the mpp reaches 1335w, which offers better results than those of the fl and inc controllers in terms of speed and output power. 6.2. non-uniform irradiation to test the performance of the three algorithms in different climatic conditions, a variation of the incidence angle of irradiation was made. the irradiation takes 880w/m 2 at the beginning of the simulations. after t=3.2 s, it rapidly decreases to 310w/m 2 and returns to the initial state. the variation of the irradiation was obtained by the change of solar incidence angle. fig. 26, fig. 27, and fig. 28 show the experimental results of inc, fl, and flt2 mppt algorithms in the case of variable irradiation. after the disturbance, the flt2 controller shows a better performance in term of response time, stability and generated power. fig. 26 experimental results of inc mppt algorithm at variable irradiation fig. 27 experimental results of fl mppt algorithm at variable irradiation fig. 28 experimental results of flt2 mppt algorithm at variable irradiation 7. discussion from the previous simulations and experimental tests, we can see that the studied mppt controllers make it possible to reach the mpp in the case of fl, flt2, and inc algorithms. in the case of stable and variable irradiation, flt2 controller is faster than fl and inc controllers. in addition, the power obtained with flt2 controller is higher. the number of rules in the fl controller is very high, which stabilizes the system around the mpp but, at the same time, makes the controller more complex and increases the calculation time while the flt2 controller uses a minimized number of rules with lower and upper standard deviations to help avoid uncertainties and reduce computation time. unfortunately, flt2 controller is more difficult to use and understand than fl controller. advances in technology innovation, vol. 4, no. 3, 2019, pp. 125-139 137 the results of this work are presented in table 4. it gives the pv power and the efficiency of the three controllers. the difference between experimental results and simulations is due to several factors such as measurement errors and losses in the conversion chain. table 4 simulation and experimental results of inc, fl and flt2 mppt algorithms simulation results experimental results techniques pv power (w) efficiency (%) pv power (w) efficiency (%) inc 1265-1272 94.22 1217 90.14 fl 1300 96.29 1271 94.14 flt2 1335 98.81 1288 95.40 8. conclusions the objective of this paper is to extract the maximum power of a photovoltaic system using fl, flt2, and inc mppt algorithms. the proposed system was studied under two different conditions; fix and variable irradiation. matlab/simulink environment is used for simulation studies and labview software for experimental tests. the simulation results and experimental tests show that the best mppt technique is flt2. it shows high performance and gives a good track of mpp with a minimized response time, it presents high robustness in case of disturbances. inc shows its limit in response time and stability. in addition, fl algorithm is a robust and powerful technique that does not require exact knowledge of the photovoltaic system. however, flt2 is more difficult to use and understand than fl and inc. the difference between experimental and simulations results is due to several factors such as measurement errors and losses in the conversion chain. as a perspective, the inc, fl, and flt2 mppt algorithms will be tested in the case of a grid-connected system. conflicts of interest the authors declare no conflict of interest. references [1] ilyas, m. ayyub, m. r. khan, a. jain, and m. a. husain, “ realisation of incremental conductance the mppt algorithm for a solar photovoltaic system,” international journal of ambient energy, vol. 39, no. 8, pp. 873-884, 2017. [2] k. p. j. pradeep, c. c. mouli, k. s. p. reddy, and k. n. raju, “design and implementation of maximum power point tracking in photovoltaic systems,” international journal of engineering sciences invention, vol. 4, no. 3, pp. 37-43, march 2015. [3] h. bounechba, a. bouzid, h. snani, and a. lashab, “real-time simulation of mppt algorithms for pv energy system,” electrical power and energy system, vol. 83, pp. 67-78, december 2016. [4] k. rohit, c. anurag, k. govind, s. amritpreet, and y. akhlendra, “modelling/simulation of mppt techniques for photovoltaic systems using matlab,” international journal of advanced research in computer science and software engineering, vol. 7, no. 4, pp. 178-187, april 2017. [5] m. bachar, a. naddami, s. hayani, and a. fahli, “design and dimensioning of desalination mobile unit and optimization of electrical energy with mppt algorithms,” proc. americain institut of physics, pp. 1-18, december 2018. [6] m. r. kumar, s. s. narayana, and g. vulasala, “advanced sliding mode control for solar pv array with fast voltage tracking for mpp algorithm,” international journal of ambient energy, pp. 1-9, 2018. [7] s. yaqin, g. tingkun, d. yan, z. linghan, and t. jingxuan, “photovoltaic array maximum power point tracking based on improved perturbation and observation method,” journal of physics: conference series, vol. 1087, pp. 1-6, 2018. [8] s. y. alsadi, “maximum power point tracking simulation for photovoltaic systems using perturb and observe algorithm,” international journal of web engineering and technology, vol. 2, no. 6, pp. 80-85, 2012. [9] p. srinivas, k. v. lakchmi, and c. ramesh, “simulation of incremental conductance mppt algorithm for pv systems using labview,” international journal of innovative research in electrical, electronics, instrumentation, and control engineering, vol. 4, no. 1, pp. 34-38, 2016. advances in technology innovation, vol. 4, no. 3, 2019, pp. 125-139 138 [10] n. karami, n. moubayed, and r. outbib, “general review and classification of different mppt techniques,” renewable and sustainable energy reviews, vol. 68, pp. 1-18, 2017. [11] m. a. m. ramli, s. twaha, k. ishaque, and y. a. al-turki, “ a review on maximum power point tracking for photovoltaic systems with and without shading conditions,” renewable and sustainable energy reviews, vol. 67, pp. 144-159, 2017. [12] c. larbes, s. m. aït cheikh, t. obeidi, and a. zerguerras, “genetic algorithms optimized fuzzy logic control for the maximum power point tracking in a photovoltaic system,” renewable energy, vol. 34, no. 10, pp. 2093-2100, 2009. [13] u. yilmaz, a. kircay, and s. borekci, “pv system fuzzy logic mppt method and pi control as a charge controller,” renewable and sustainable energy reviews, vol. 81, pp. 994-1001, 2018. [14] v. dabra, k. k. paliwal, p. sharma, and n. kumar, “optimization of the photovoltaic power system: a comparative study,” protection and control of modern power systems, vol. 2, no. 3, pp. 1-11, january 2017. [15] m. biglarbegian, w. w. malek, and j. m. mendel, “on the stability of interval type-2 tsk fuzzy logic control systems,” ieee transactions on systems, vol. 40, no. 3, pp. 798-818, june 2010. [16] j. h. agrawal and m. v. aware, “photovoltaic simulator developed in labview for evaluation of mppt techniques,” proc. international conference on electrical, electronics, and optimization techniques, march 2016, pp. 1142-1147. [17] a. el-leathey, a. nedelcu, s. nicolaie, and r.a. chihaia, “labview design and simulation of a small scale microgrid,” series c: electrical engineering, vol. 78, no. 1, pp. 235-246, march 2016. [18] v. banu, m. istrate, d. machidon, and r. pantelimon, “study regarding modeling photovoltaic arrays using test data in matlab/simulink,” scientific bulletin series c, vol. 77, no. 2, pp. 227-234, 2015. [19] r. sridhar, s. jeevananthan, and p. vishnuram, “particle swarm optimization maximum power point tracking approach based on irradiation and temperature measurements for partially shaded photovoltaic system,” international journal of ambient energy, vol. 38, no. 7, 2017. [20] a. ayaz and l. rajaji, “real-time implementation of the solar inverter with novel mppt control algorithm for residential application,” energy and power engineering, vol. 5, no. 6, pp. 427-435, august 2013. [21] s. kolsi, h. samet, and m. ben amar, “design analysis of dc-dc converters connected to a photovoltaic generator and controlled by mppt for optimal energy transfer throught a clear day,” journal of power and energy engineering, vol. 2, no. 1, pp. 27-34, january 2014. [22] a. attou, a. massoum, and m. saidi, “photovoltaic power control using mppt and boost converter,” balkan journal of electrical & computer engineering, vol. 2, no. 1, pp. 23-27, september 2014. [23] m. bachar, a. naddami, s. hayani, and a. fahli, “optimization of pv panel using p&o and incremental conductance algorithms for desalination mobile unit,” advanced intelligent systems applied to energy, vol. 2, pp. 164-184, january 2019. [24] r. faranda and s. leva, “energy comparison of mppt techniques for pv systems,” wseas transactions on power systems, vol. 3, no. 6, pp. 446-455, 2008. [25] a. montecucco and a. r. knox, “maximum power point tracking converter based on the open-circuit voltage method for thermoelectric generators,” ieee transactions on power electronics, vol. 30, no. 2, pp. 828-839, march 2014. [26] a. dolora, r. faranda, and s. leva, “energy comparison of seven mppt techniques for pv systems,” journal of electromagnetic analysis and applications, vol. 3, pp. 152-162, 2009. [27] s. malathy and r. ramaprabha, “maximum power point tracking based on look up table approach,” advanced materials research, vol. 768, pp. 124-130, september 2013. [28] a. naidu, s. a. kumar, and g. s. reddy, “voltage based p&o algorithm for maximum power point tracking using labview,” innovative systems design and engineering, vol. 7, no. 5, pp. 12-16, 2016. [29] y. shi, r. li, y. xue, and h. li, “high-frequency-link-based grid-tied pv system with small dc-link capacitor and low-frequency ripple-free maximum power point tracking,” ieee transactions on power electronics, vol. 31, no. 1, pp. 328-339, 2016. [30] m. farayola, a. n. hasan, and a. ali, “curve fitting polynomial technique compared to anfis technique for maximum power point tracking,” proc. 8th international renewable energy congress, march 2017. [31] s. e. babaa, m. armstrong, and v. pickert, “overview of maximum power point tracking control methods for pv systems,” journal of power and energy engineering, vol. 2, pp. 59-72, 2014. [32] m. e. başoğlu and b. çakir, “design and implementation of dc-dc converter with inc-cond algorithm,” international journal of electrical energetic electronic and communication engineering, vol. 9, no. 1, pp. 45-48, january 2015. [33] w. m. lin, c. m. hong, and c. h. chen, “neural-network-based mppt control of a stand-alone hybrid power generation system,” ieee transactions on power electronics, vol. 26, no. 12, pp. 3571-3581, july 2011. advances in technology innovation, vol. 4, no. 3, 2019, pp. 125-139 139 [34] m. kermadi and e. m. berkouk, “artificial intelligence-based maximum power point tracking controllers for photovoltaic systems: comparative study,” renewable and sustainable energy reviews, vol. 69, pp. 369-386, 2017. [35] m. s. fadali, s. jafarzadeh, and a. nadeh, “fuzzy tsk approximation using type-2 fuzzy logic systems and its application to modeling a photovoltaic array,” proc. american control conference, pp. 6454-6459, july 2010. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation, vol. 4, no. 3, 2019, pp. 152-164 vibration analysis of two-directional functionally graded sandwich beams using a shear deformable finite element formulation vu nam pham 1, 2 , dinh kien nguyen 2, 3, * , buntara sthenly gan 4 1 thuyloi university, 175 tay son, dong da, hanoi, vietnam 2 graduate university of science and technology, vast, 18 hoang quoc viet, hanoi, vietnam 3 institute of mechanics, vast, 18 hoang quoc viet, hanoi, vietnam 4 department of architecture, college of engineering, nihon university, koriyama, japan received 17 march 2019; received in revised form 10 april 2019; accepted 10 may 2019 abstract free and forced vibration analysis of two-directional functionally graded sandwich (2d-fgsw) beams using a shear deformable finite element formulation is presented. the beams considered in this paper consists of three layers, a homogeneous ceramic core, and two functionally graded skin layers. material properties of the skin layers are supposed to vary in both the thickness and length directions by power gradation laws. based on a refined shear deformation beam theory, in which the transverse displacement is split into bending and shear parts, a novel finite element formulation is derived and employed in the analysis. natural frequencies and dynamic response to a harmonic load of the beams with various boundary conditions are computed, and the influence of the material distribution and the layer thickness ratio on the vibration characteristics of the beams is highlighted. numerical results reveal that the variation of the material properties in the longitudinal direction has a significant influence on the vibration behavior of the beams, and fgsw beams can be designed to achieve desired vibration characteristics by appropriate selection of material grading indexes. keywords: 2d-fgsw beam, refined shear deformation theory, vibration analysis, finite element formulation 1. introduction sandwich structures with high strength-to-weight ratio are widely used to fabricate structural elements in aerospace engineering. in order to improve the mechanical performance of these structures under complex loadings, functionally graded materials (fgms), a new type of composite materials initiated by japanese researchers in 1984 [1], have been incorporated in the sandwich construction in recent years. understanding the mechanical behavior of functionally graded sandwich (fgsw) structures in general, and vibration of fgsw beams, in particular, is crucial for using this new composite material effectively. investigations on the vibration of fgsw beams, the topic discussed in this paper, are briefly discussed below. pradhan and murmu [2] used the modified differential quadrature method to compute natural frequencies of fgsw beams resting on an elastic foundation. the dependence of material properties upon temperature was considered in the work. based on the element free galerkin method, amirani et al. [3] studied free vibration of an fgsw beam with an fgm core. the authors employed two micromechanics models, voigt and mori-tanaka models, to evaluate the effective material properties of the beams, and they showed that the natural frequencies based on mori-tanaka scheme are slightly lower than that using voigt model. adopting a hierarchical displacement field, mashat et al. [4] derived a finite element formulation * corresponding author. e-mail address: ndkien@imech.vast.vn tel.: +84-24-37628006; fax: +84-24-37622039 advances in technology innovation, vol. 4, no. 3, 2019, pp. 152-164 153 for evaluating natural frequencies of laminated and sandwich beams with material properties varying by a power gradation law. free vibration and buckling of fgsw beams were considered by bennai et al. [5] by using a new refined hyperbolic shear deformation beam theory. in [6, 7], vo et al. presented a refined shear deformation theory and then improved it to form the quasi-3d theory by taking thickness stretching effect into account for studying free vibration and buckling of fgsw beams. recently, su et al. [8] employed the modified fourier series to compute fundamental frequencies of fgsw beams resting on an elastic foundation. in many practical circumstances, the unidirectional fgm in the above-cited papers may not be so appropriate to resist multi-directional variations of mechanical and thermal loadings, and development of fgm with material properties varying in two or three spatial directions is necessary. several investigations on vibration analysis of fgm beams with material gradation in both the thickness and length directions have been reported in recent years [9-12]. investigation on the mechanical behavior of 2d-fgsw beams, however, is still very limited. to the authors’ best knowledge, there is only a study by karamanli in [13], where static bending of fgsw beams with material properties varying in both the thickness and length directions by power gradation laws under uniform distributed load was studied by the symmetric smoothed particle hydrodynamics method. in order to explore the behavior of this new type of structure in some further, a finite beam element is formulated in this paper for studying free and forced vibration of two-directional functionally graded sandwich (2d-fgsw) beams. the beams are considered to be formed from three layers, a homogeneous isotropic ceramic core and two fgm skin layers with material properties varying in both the thickness and length directions by power gradation laws. a refined shear deformation beam theory, in which the transverse displacement is split into bending and shear parts, is adopted to derive the element stiffness and mass matrices of the beam element. thus, in addition to the vibration of the 2d-fgsw beams considered herein for the first time, the refined theory which allows to include the shear and rotary inertia effects by dividing the transverse displacement into bending and shear parts is also a new feature of this paper. the theory with the parabolic distribution of shear deformation in the beam thickness does not require a shear correction factor, and with the mentioned advantages, it has been widely adopted in vibration and buckling analysis of fgsw plates [14-17]. using the derived formulation, natural frequencies and dynamic response of the beams with various boundary conditions to a harmonic load are computed, and the effects of material distribution, the layer thickness ratio and the aspect ratio on the vibration behavior of the beams are examined and highlighted. 2. 2d-fgsw beam model a 2d-fgsw beam with a rectangular cross-section (bxh) as depicted in fig. 1 is considered. the beam is assumed to form from three layers, namely a core of pure ceramic and two skin layers made of ceramic-metal fgm. in the figure, the x-axis is chosen on the mid-plane, and the z-axis is perpendicular to the mid-plane and directs upward. denoting z0, z1, z2, and z3 as the vertical coordinates of the bottom layer, layer interfaces, and the top layer, respectively. layer 1: 2d-fgm layer 3: 2d-fgm layer 2: ceramic b 0 z =-h/2 3 z =h/2 y z h x z y 2 z 1 z fig. 1 geometry of 2d-fgsw beam the beam is considered to be made of ceramic and metal whose volume fraction varies in the thickness and length directions by power gradation laws as [13] advances in technology innovation, vol. 4, no. 3, 2019, pp. 152-164 154       1 0 1 0 1 1 2 2 2 3 3 2 1 for , 2 0 for , 1 for , 2 and 1 z x z x n n m n n c m z z x z z z z z l v z z z z z x z z z z z l v v                                 (1) where vc and vm are, respectively, the volume fraction of ceramic and metal; nx and nz are the grading indexes, defining variations of the constituent materials in the xand z-directions, respectively. noting that when nx=0 the beam deduces to the conventional 1d-fgsw beam with the material properties vary in the thickness direction only. the effective property such as young’s modulus and mass density, p(x,z), evaluated by the voigt model is of the forms         1 0 1 0 1 1 2 2 2 3 3 2 ( ) 1 if , 2 , if , ( ) 1 if , 2 z x z x p p m c c c p p m c c z z x p p p z z z z z l p x z p z z z z z x p p p z z z z z l                                      (2) where pm and pc are the properties of the metal and ceramic, respectively. based on the refined third-order shear deformation beam theory [6], the displacements in xand z-directions, u1(x,z,t) and u3(x,z,t), at any point of the beam are respectively given by     3 1 , ,2 3 4 , , ( , ) ( , ) ( , ) 3 , , ( , ) ( , ) b x s x b s z u x z t u x t zw x t w x t h u x z t w x t w x t      (3) where u, wb, ws are the axial displacement, transverse bending and shear displacements of a point on the x-axis, respectively. in eq. (3) and hereafter, a subscript comma is used to denote the derivative with respect to the followed variable, e.g. , / s x w w x   . the strains resulted from eq. (3) are of the forms 3 1, , , ,2 2 1, 3, ,2 4 , 3 4 1 x x x b xx s xx xz z x s x z u u zw w h z u u w h                (4) assuming linearly elastic behavior, the constitutive equations for the beam are of the form 2 , 1 2(1 ) x x xz xz e e           (5) where  is poisson’s ratio, assumed to be unchanged. the strain energy (u) resulted from eqs. (4) and (5) is   2 211 , 12 , , 22 . 23 , ,2 0 2 2 44 , , 66 , 11 22 44 ,2 4 2 4 1 1 8 2 2 2 3 8 16 8 16 3 9 [ ] l x x xz xz x x b xx b xx x s xx v b xx s xx s xx s x u dv a u a u w a w a u w h a w w a w b b b w dx h h h h                      (6) where a11, a12 …, a66, b11, b22, b44 are the beam rigidities, defined as advances in technology innovation, vol. 4, no. 3, 2019, pp. 152-164 155 1 3 2 3 4 6 2 3 4 6 11 12 22 23 44 66 2 2 1 ( , , , , , ) (1, , , , , ) (1, , , , , ) 1 1 i i z ia z e be a a a a a a z z z z z da z z z z z dz           (7) and 1 3 2 4 2 4 11 22 44 1 ( , , ) ( , , ) (1, , ) 2(1 ) 2(1 ) i i z ia z e be b b b z z z da z z dz           (8) in eqs. (6-8), v and a denote the volume and cross-sectional area of the beam, respectively. because the effective young modulus e varies in both the thickness and length directions, the rigidities in eqs. (7) and (8) are functions of x. the kinetic energy (t) of the beam resulted from the displacement field in eq. (3) is as follows 2 2 2 2 234 6644 11 12 , 22 , , , , ,2 2 2 0 8 168 ( 2 ) 2 3 3 9 l b s b s b x b x s x b x s x s x i ii t i u w w w w i uw i w uw w w w dx h h h                (9) in which i11, i12, i22, i34, i44, and i66 are the mass moments, defined as 1 3 2 3 4 6 2 3 4 6 11 12 22 34 44 66 1 ( , , , , , ) ( , ) (1, , , , , z ) ( , ) (1, , , , , z ) i i z ia z i i i i i i x z z z z z da b x z z z z z dz       (10) the above mass moments, as the beam rigidities, also are functions of x. the potential (v) of a harmonic load, f=f0cos(ωt), considered herein has a simple form 0 | cos( )( ) fb s x x v f t w w      (11) in the above equation, the subscript x=xf means that the bending and shear transverse displacements are evaluated at the abscissa of the load f. equations of motion for the beam can be obtained by applying hamilton’s principle to eqs. (6), (9) and (11). however, due to the rigidities and mass moment are functions of longitudinal coordinate x, the coefficients of such equation are dependent on x, and thus a closed-form solution is very difficult to obtain. a finite element formulation is developed below for computing the vibration characteristics of the beam. 3. finite element formulation a two-node beam element with length l is considered in this section. the element contains six degrees of freedom per node, and the vector of nodal displacements is given by { } t u wb ws d d d d (12) where 1 2 1 , 1 2 , 2 1 , 1 2 , 2 { }, { }, { } u wb b b x b b x ws s s x s s x u u w w w w w w w w  d d d (13) are, respectively, the vectors of axial displacements, bending and shear transverse displacements at node 1 and node 2. in eq. (12) and hereafter, a superscript “t” is used to denote the transpose of a vector or a matrix. it should be noted that the order of the nodal displacements is not necessary as in eq. (12), but it is convenient to split the displacements into axial, bending and shear parts. advances in technology innovation, vol. 4, no. 3, 2019, pp. 152-164 156 the displacements inside the element are interpolated from their nodal values according to 0 0 0 0 0 0 uu b wb wb wss ws u w w                         dn n d n d (14) where nu, nwb and nws are the matrices of shape functions for u, wb and ws, respectively. in the present work, the linear function is used for the axial displacement, while the hermite cubic polynomials are employed for wb and ws. in this regard, we can write 1 2 { } u u u l x x n n l l         n (15) and 1 2 3 4 { } w w w w w n n n nn (16) with 2 3 2 3 2 3 2 3 1 2 3 42 3 2 2 3 2 1 3 2 , 2 , 3 2 , w w w w x x x x x x x x n n x n n l l l l l l l l            (17) using the above interpolation scheme, one can write the strain energy for the beam in the form       1 1 2 ne t i i i i u    d k d (18) where ne is the total number of elements, and k is the element stiffness matrix, which can be split into the sub-matrices as (10 x10) aa ab as t ab bb bs t t as bs ss           k k k k k k k k k k (19) in the above, aa k , bb k , ss k , ab k , as k , bs k are the stiffness matrices stemming from axial stretching, transverse bending, transverse shear and couplings of these terms. these sub-matrices can be obtained by twice differentiation of the strain energy u with respect the nodal displacements, for example 2 2 2 2 2 , , b b aa ab bb u u w w u u u          k k k d d d d (20) eq. (20) gives the sub-matrices in the following forms , 11 , , 12 , , 34 ,x2 ( 2 x 2) ( 2 x 4) ( 2 x 4)0 0 0 ,x 22 ,x ,x 44 ,x2 ( 4 x 4) ( 4 x 4)0 0 ,x 66 ,x ,x 11 224 2 ( 4 x 4) 4 , , , 3 4 , , 3 16 8 16 9 l l l t t t aa u x u x ab u x w x as u x w x l l t t bb w x w x bs w x w x t t ss w x w x w a dx a dx a dx h a dx a dx h a dx b b h h h                 k n n k n n k n n k n n k n n k n n n 44 ,x4 0 0 l l w b dx         n (21) similarly, the kinetic energy of the beam can also be written in the form advances in technology innovation, vol. 4, no. 3, 2019, pp. 152-164 157       1 1 2 ne t ii i i t    d m d (22) where / t  d d is the element nodal velocity, and m is the element mass matrix which can be written in sub-matrices as (10 x10) b s b b b b s s b s s s uu uw uw t uw w w w w t t uw w w w w            m m m m m m m m m m (23) the mass sub-matrices in the above equation are obtained by twice differentiation of the kinetic energy with respect to the associated nodal velocities and they have the following forms   11 12 , 11 , 34 ,x2 ( 2 x 2) ( 2 x 4) ( 2 x 4)0 0 0 11 ,x 22 ,x 11 ,x 44 ,x2 ( 4 x 4)( 4 x 4) 0 0 11 ( 4 x 4) 4 , , , 3 4 , , 3 1 b s b b l l l t t t t uu u u uw u w x uw w w u x w l l t t t t m m w w w w bs w w w w t ss w w i dx i dx i i dx h i i dx i i dx h i                           m n n m n n m n n n n m n n n n m n n n n m n n ,x 66 ,4 0 6 9 l t w w x i dx h        n n (24) due to the rigidities and mass moments are functions of x, the integrals in eqs. (21) and (24) are hardly computed explicitly. gauss quadrature is employed herein to compute the stiffness and mass matrices. with the introduced interpolations, the vector of nodal external load given by eq. (11) can be written in the form 0 element under loading cos( ) 0 0 0..... 0 0 0 0...0 0 0 t ex w w f t              f n n (25) the matrix of the shape functions in the above equation is evaluated at x=xf, the abscissa measured from the load to the left end of the element. the equations of motion for the beam in term of finite element analysis can be written in the form [18] ex   md cd kd f (26) where d, d , d are, respectively, the structural vectors of nodal displacements, velocities and accelerations; k, m, and c are structural stiffness, mass and damping matrices, respectively. rayleigh damping, in which the damping matrix c is proportional to a linear combination of mass and stiffness, is employed herein   c k m (27) where α and β are, respectively, the mass and stiffness proportional rayleigh damping coefficient, which are calculated from the critical damping ratio and the natural frequencies as 1 2 1 2 1 2 2 2 ,               (28) in the above,  is the damping ratio, taken by 5% for all numerical computation below. advances in technology innovation, vol. 4, no. 3, 2019, pp. 152-164 158 4. results and discussion using the derived finite element formulation, natural frequencies and dynamic response of the 2d-fgsw beam are computed, and numerical results are reported in this section. to this end, an fgsw beam formed from alumina (ceramic) and aluminum (metal) is considered. the material properties of the constituents are as follows [7]: (1) ec=380 mpa, ρc=3800 kg/m 3 , 0.3 c   for alumina (2) em=70 mpa, ρm=2702 kg/m 3 , 0.3 m   for aluminum for the convenience of discussion, the three numbers as proposed in [6] are used to indicate the layer thickness ratio, for example (2-1-2) means the thickness ratio of the bottom, core, and top layers is 2:1:2. fig. 2 shows the variation of the effective young’s modulus and mass density in the thickness and length directions for (1-1-1) beam made of alumina and aluminum with nx=nz=0.5 according to eq. (2). it can be seen that young’s modulus and mass density continuously vary in both the thickness and length directions of the beam. (a) young’s modulus (b) mass density fig. 2 variation of young’s modulus and mass density for (1-1-1) beam with nx=nz=0.5 4.1. formulation verification before computing vibration characteristics of the beam, the derived formulation is necessary to confirm. since there is no data on the vibration of the 2d-fgsw beam considered herein available in the literature, the comparison is carried out for the static bending of the beam as reported in [13]. table 1 compares the maximum dimensionless deflection of the simply supported beam (ss beam) under uniform distributed load q0 obtained by the present finite element formulation with that of karamanli [13] using the symmetric smoothed particle hydrodynamics method. regardless of the grading indexes, the layer thickness ratio and the aspect ratio, the maximum deflection of the beam obtained in the present paper is in good agreement with that of ref. [13]. the dimensionless deflection in table 1 is defined as follows [13]   3 * 4 0 100 max ( )m e bh w w x q l  (29) where em is young’s modulus of metal. advances in technology innovation, vol. 4, no. 3, 2019, pp. 152-164 159 table 1 comparison of maximum dimensionless deflection ( * w ) of ss beam under static uniform load nx nz source l/h = 5 l/h=20 1-1-1 1-8-1 2-2-1 1-1-1 1-8-1 2-2-1 0.1 0.1 ref. [13] 10.7054 4.7401 10.9470 10.3994 4.4818 9.1047 present 10.8634 4.8064 9.4128 10.4116 4.4848 9.1096 0.5 ref. [13] 7.5039 4.2112 9.5412 7.2199 3.9561 6.5597 present 7.6124 4.2698 6.8473 7.2273 3.9586 6.5680 1 ref. [13] 6.0343 3.9030 6.9428 5.7613 3.6501 5.3608 present 6.1185 3.9570 5.6327 5.7667 3.6525 5.3658 2 ref. [13] 4.8871 3.6275 4.6673 4.6274 3.3772 4.4070 present 4.9572 3.6775 4.7321 4.6313 3.3793 4.4101 0.5 0.1 ref. [13] 8.4793 4.4862 5.7112 8.1706 4.4143 7.3680 present 8.6148 4.5492 7.6764 8.1964 4.2298 7.3839 0.5 ref. [13] 6.5069 4.0580 7.7882 6.2253 4.2331 5.7569 present 6.6011 4.1143 6.0408 6.2338 3.8040 5.7660 1 ref. [13] 5.4735 3.8004 6.1257 5.2055 3.8068 4.9004 present 5.5523 3.8534 5.1692 5.2114 3.5490 4.9064 2 ref. [13] 4.6040 3.5666 4.4251 4.3451 3.3169 4.1669 present 4.6689 3.6155 4.4873 4.3491 3.3190 4.1706 1 0.1 ref. [13] 6.9827 4.2462 6.4602 6.6753 3.5515 6.1562 present 7.0975 4.3050 6.5600 6.7054 3.9922 6.1781 0.5 ref. [13] 5.7178 3.9088 5.3861 5.4388 3.9943 5.1050 present 5.8019 3.9608 5.4616 5.4499 3.6551 5.1153 1 ref. [13] 4.9904 3.6976 4.7624 4.7252 3.6570 4.4948 present 5.0598 3.7487 4.8272 4.7288 3.4477 4.5000 2 ref. [13] 4.3387 3.5031 4.1978 4.0816 3.2549 3.9396 present 4.3982 3.5514 4.2549 4.0843 3.2566 3.9434 4.2. free vibration the fundamental frequency parameters of the ss beam are listed in tables 2 and 3 for various values of the grading indexes and the layer thickness ratio and two value of the aspect ratio, l/h=5 and l/h=20, respectively. the frequency parameters in the tables (and below) are defined as follows 2 i m i m l h e     (30) where ωi is the i th natural frequency of the beam. table 2 fundamental frequency parameter (μ1) of ss beam with l/h = 5 nx nz 1-0-1 2-1-2 1-1-1 1-2-1 1-8-1 2-2-1 0 0.5 3.2500 3.3728 3.5186 3.7904 4.5227 3.7057 1 3.5735 3.7297 3.8755 4.1105 4.6795 4.0189 5 4.4615 4.5756 4.6582 4.7690 4.9899 4.7106 0.5 0.5 3.6300 3.7134 3.8194 4.0263 4.6176 3.9640 1 3.8648 3.9816 4.0942 4.2804 4.7489 4.2086 5 4.5673 4.6620 4.7311 4.8245 5.0125 4.7753 1 0.5 3.8922 3.9549 4.0378 4.2037 4.6942 4.1548 1 4.0761 4.1687 4.2596 4.4123 4.8056 4.3538 5 4.6520 4.7319 4.7906 4.8701 5.0315 4.8282 2 0.5 4.2427 4.2840 4.3402 4.4556 4.8093 4.4223 1 4.3679 4.4318 4.4956 4.6043 4.8918 4.5629 5 4.7785 4.8373 4.8807 4.9398 5.0609 4.9087 5 0.5 4.2427 4.2840 4.3402 4.4556 4.8093 4.4223 1 4.3679 4.4318 4.4956 4.6043 4.8918 4.5629 5 4.7785 4.8373 4.8807 4.9398 5.0609 4.9087 the effects of the grading indexes and the layer thickness ratio on the frequency of the beam are clearly seen from tables 2 and 3. for a given value of the layer thickness ratio, the frequency parameter of the beam increases by the increase of the advances in technology innovation, vol. 4, no. 3, 2019, pp. 152-164 160 material grading indexes, regardless of the aspect ratio. this can be explained by the fact that the volume fraction of the metal, as seen from eq. (1), is lower, and thus the ceramic percentage is larger for the beam associated with higher indexes nx and nz. since young’s modulus of the ceramic is much higher than that of the metal, the rigidities of the beam with higher ceramic content are higher. the mass density of the beam with a higher ceramic content also larger, but for the constituent materials considered in this paper, the increase of the rigidities by increasing nx and nz is much faster than that of the mass moments. this explains the increase of the frequency when increasing the grading indexes. the layer thickness ratio, as seen from table 2 and 3, also influences the increase of the frequency. for example, for the ss beam with l/h=20 and a length index nx=0.5, the frequency parameter μ1 of the (1-0-1) beam increases 21.02% when increasing nz from 0.5 to 5, but the corresponding value is 20.24%, 7.51%, and 8.44% for the (1-1-1), (1-2-1), and (1-8-1) beams, respectively. a careful examination of table 2 and 3 shows that the increase of the frequency by increasing nz depends upon the value of nx also. for example, as seen from table 2, the frequency parameter increases 20.35% when increasing nz from 0.5 to 5 for (1-1-1) beam with nx=0.5, but the corresponding value for the (1-1-1) beam with nx=5 is just 11.44%. in other words, the increase of the fundamental frequency by increasing the thickness index is smaller for the beam associated with a higher length index. table 3 fundamental frequency parameter (μ1) of ss beam with l/h = 20 nx nz 1-0-1 2-1-2 1-1-1 1-2-1 1-8-1 2-2-1 0 0.5 3.3781 3.4953 3.6472 3.9387 4.7473 3.8512 1 3.7147 3.8768 4.0328 4.2889 4.9233 4.1906 5 4.6783 4.8058 4.8987 5.0239 5.2748 4.9578 0.5 0.5 3.7880 3.8629 3.9727 4.1958 4.8533 4.1320 1 4.0307 4.1510 4.2717 4.4760 5.0012 4.3988 5 4.7962 4.9026 4.9806 5.0864 5.3005 5.0306 1 0.5 4.0732 4.1260 4.2112 4.3907 4.9391 4.3413 1 4.2617 4.3563 4.4539 4.6222 5.0651 4.5594 5 4.8911 4.9812 5.0476 5.1380 5.3220 5.0903 2 0.5 4.4581 4.4884 4.5450 4.6700 5.0686 4.6374 1 4.5838 4.6477 4.7160 4.8365 5.1623 4.7923 5 5.0334 5.1000 5.1494 5.2169 5.3554 5.1813 5 0.5 5.0070 5.0168 5.0398 5.0940 5.2756 5.0808 1 5.0587 5.0853 5.1151 5.1690 5.3194 5.1496 5 5.2595 5.2904 5.3134 5.3450 5.4104 5.3284 (a) the first parameter μ1 (b) the second parameter μ2 (c) the third parameter μ3 (d) the fourth parameter μ4 fig. 3 variation of the first four frequency parameters with grading indexes of (1-1-1) ss beam with l/h = 20 advances in technology innovation, vol. 4, no. 3, 2019, pp. 152-164 161 the aspect ratio l/h also slightly influences the change of the frequency parameter and the change in the frequency parameter is not significant for the beam with a lower aspect ratio l/h. for example, the fundamental frequency parameter increases 17.17% for the (2-1-2) beam with nx=1 having l/h=20 when increasing nz from 0.5 to 5, but this increase reduces to 16.42% for the beam with l/h=5. thus, the material gradation, the layer thickness ratio, and the aspect ratio are all important parameters which should be considered in designing the 2d-fgsw beam depends in order to achieve a beam with desired natural frequencies. the effect of the grading indexes on the higher frequency parameters is illustrated in fig. 3-5, where the variation of the first four natural frequency parameters with the grading indexes nx and nx is depicted for the ss, clamped (cc) and cantilever (cf) beams with a layer thickness ratio of (1-1-1) and an aspect ratio l/h=20, respectively. similar to the fundamental frequency, the higher frequencies also increase by increasing the grading indexes nx and nz, regardless of the boundary conditions. the boundary conditions may affect the amplitude of the natural frequencies, but it hardly influences the dependence of the frequency parameters upon the material grading indexes. (a) the first parameter μ1 (b) the second parameter μ2 (c) the third parameter μ3 (d) the fourth parameter μ4 fig. 4 variation of the first four frequency parameters with grading indexes of (1-1-1) cc beam with l/h = 20 (a) the first parameter μ1 (b) the second parameter μ2 (c) the third parameter μ3 (d) the fourth parameter μ4 fig. 5 variation of the first four frequency parameters with grading indexes of (1-1-1) cf beam with l/h = 20 advances in technology innovation, vol. 4, no. 3, 2019, pp. 152-164 162 4.3. dynamic response the dynamic response of the 2d-fgsw beam to a harmonic load p=p0cos(ωt) is investigated in this sub-section. in order to calculate the response of the beam, the average acceleration newmark method is employed to solve eq. (26). the beam with two types of boundary conditions, namely ss and cf beams, with a harmonic load acting at the mid-span and free and, respectively are considered herein. the variation of the dimensionless deflections with the time at the loaded points of the ss and cf beams with an aspect ratio l/h=20 and a layer thickness ratio of (2-1-2) is illustrated in fig. 6 and 7 for an excitation frequency ω=10 rad/s, respectively. the deflections w* in the figures are normalized by the static bending transverse displacements of the ceramic beam, that are 3 48 * ( / 2) for ss beamc b e i w w l pl  (31) 3 3 * ( ) for cf beamc b e i w w l pl  (32) (a) nx=1, nz is variable (b) nz=1, nx is variable fig. 6 variation of mid-span dimensionless deflection with the time of ss (2-1-2) beam under harmonic load with ω=10 rad/s acting at mid-span (a) nx=1, nz is variable (b) nz=1, nx is variable fig. 7 variation of dimensionless deflection at the free end with a time of cf (2-1-2) beam under harmonic load with ω=10 rad/s acting at the free end the effect of the grading indexes on the harmonic response of the beams can be seen from the figures, where the dynamic deflections are clearly decreased by increasing the material grading indexes nx and nz , regardless of the boundary conditions. advances in technology innovation, vol. 4, no. 3, 2019, pp. 152-164 163 this effect can also be explained by the increase of ceramic content for the beam associated with the higher indexes, and this leads to higher rigidities for the beam associated with higher grading indexes, as explained above. as a result, the dynamic deflection of the beam is decreased by increasing the material grading indexes. 5. conclusions in this paper, a finite element formulation was developed for analyzing the free and forced vibration of 2d-fgsw beams. the beams were considered to be formed from three layers, a homogeneous ceramic core and two functionally graded skin layers with material properties varying in both the thickness and length directions by power gradation laws. based on the refined third-order shear deformation theory, in which the transverse displacement is split into bending and shear parts, expressions for stiffness and mass matrices of a two-node beam element were derived and employed in computing natural frequencies and dynamic response of the beams. numerical results obtained by using the formulation was compared with the published data to confirm the accuracy of the proposed formulation. a parametric study has been carried to highlight the effects of the material distribution, the layer thickness ratio and the aspect ratio on the vibration characteristics of the beams. the obtained results reveal that the material distribution and the layer thickness ratio of the beams play an important role on the vibration response of the 2d-fgsw beams. the numerical results of the present paper guide to design 2d-fdsw beams to achieve desired vibration characteristics. though the paper examined only harmonic response of the beam, the finite element formulation developed herein can be employed to compute dynamic response of 2d-fgsw beams subjected to other types of dynamic loads as well. conflicts of interest the authors declare no conflict of interest. acknowledgement this work was supported by vietnam national foundation for science and technology development (nafosted) under grant no. 107.02-2018.23. references [1] m. koizumin, “fgm activities in japan,” composites: part b, vol. 28, pp. 1-4, 1997. [2] s. c. pradhan and t. murmu, “thermo-mechanical vibration of fgm sandwich beam under variable elastic foundations using differential quadrature method,” journal of sound and vibration, vol. 321, pp. 342-362, march 2009. [3] m. c. amirani, s. m. r. khalili, and n. nemati. “free vibration analysis of sandwich beam with fg core using the element free galerkin method,” composite structures, vol. 90, pp. 373-379, october 2009. [4] d. s. mashat, e. carrera, a. m. zenkour, s. a. al khateeb, and m. filippi, “free vibration of fgm layered beams by various theories and finite elements,” composites: part b, vol. 59, pp. 269-278, march 2014. [5] r. bennai, h. a. atmane, and a. tounsi, “a new higher-order shear and normal deformation theory for functionally graded sandwich beams,” steel and composite structures, vol. 19, pp. 521-546, september 2015. [6] t. p. vo, h.-t. thai, t.-k. nguyen, a. maheri, and j. lee, “finite element model for vibration and buckling of functionally graded sandwich beams based on a refined shear deformation theory,” engineering structures, vol. 64, pp. 12-22, april 2014. [7] t. p. vo, h. -t. thai, t . -k. nguyen, f. inam, and j. lee, “a quasi-3d theory for vibration and buckling of functionally graded sandwich beams,” composite structures, vol. 119, pp. 1-12, january 2015. [8] z. su, g. jin, y. wang, and x. ye, “a general fourier formulation for vibration analysis of functionally graded sandwich beams with arbitrary boundary condition and resting on elastic foundations,” acta mechanica, vol. 227, pp. 1493-1514, may 2016. [9] m. şimşek, “bi-directional functionally graded materials (bdfgms) for free and forced vibration of timoshenko beams with various boundary conditions,” composite structures, vol. 133, pp.968-978, december 2015. advances in technology innovation, vol. 4, no. 3, 2019, pp. 152-164 164 [10] z. wang, x. wang, g. xu, s. cheng, and t. zeng, “free vibration of two-directional functionally graded beams,” composite structures, vol. 135, pp.191-198, january 2016. [11] d. k. nguyen, q. h. nguyen, t. t. tran, and v. t. bui, “vibration of bi-dimensional functionally graded timoshenko beams excited by a moving load,” acta mechanica, vol. 228, pp.141-155, january 2017. [12] d. k. nguyen and t. t. tran, “free vibration of tapered bfgm beams using an efficient shear deformable finite element model,” steel and composite structures, vol. 29, pp. 363-377, november 2018. [13] a. karamanli, “bending behaviour of two directional functionally graded sandwich beams by using a quasi-3d shear deformation theory,” composite structures, vol. 174, pp. 70-86, august 2017. [14] m. a. a. meziane, h. h. abdelaziz, a. tounsi, “an efficient and simple refined theory for buckling and free vibration of exponentially graded sandwich plates under various boundary conditions”, journal of sandwich structures and materials, vol. 16, pp. 293-318, march 2014. [15] m. bennoun, m. s. a. houari, and a. tounsi, “a novel five variable refined plate theory for vibration analysis of functionally graded sandwich plates”, mechanics of advanced materials and structures, vol. 23, pp. 423-431, january 2016. [16] h. h. abdelaziz, m. a. a. meziane, a. a. bousahla, a. tounsi, s. r. mahmoud, and a. s. alwabli, “an efficient hyperbolic shear deformation theory for bending, buckling and free vibration of fgm sandwich plates with various boundary conditions,” composite structures, vol. 25, pp. 693-704, december 2017. [17] f. bourada, k. amara, a. a. bousahla, a. tounsi, and s.r. mahmoud, “a novel refined plate theory for stability analysis of hybrid and symmetric s-fgm plates”, structural engineering and mechanics, vol. 68, pp. 661-675, december 2018. [18] m. géradin and r. rixen, mechanical vibrations, theory and application to structural dynamics, 2nd edition. england; chichester, john wiley and sons, 1997. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). https://journals.sagepub.com/doi/abs/10.1177/1099636214526852?journalcode=jsma https://journals.sagepub.com/doi/abs/10.1177/1099636214526852?journalcode=jsma https://journals.sagepub.com/doi/abs/10.1177/1099636214526852?journalcode=jsma callto:16(3),%20293-318 https://www.tandfonline.com/author/bennoun%2c+mohammed https://www.tandfonline.com/author/houari%2c+mohammed+sid+ahmed https://www.tandfonline.com/author/tounsi%2c+abdelouahed https://www.researchgate.net/journal/1225-4568_structural_engineering_mechanics  advances in technology innovation, vol. 5, no. 1, 2020, pp. 45-55 a multichannel mac protocol for iot-enabled cognitive radio ad hoc networks chien-min wu * , yen-chun kao, kai-fu chang department of computer science and information engineering, nanhua university, chiayi, taiwan received 17 march 2019; received in revised form 13 may 2019; accepted 21 august 2019 doi: https://doi.org/10.46604/aiti.2020.3946 abstract cognitive radios have the ability to dynamically sense and access the wireless spectrum, and this ability is a key factor in successfully building internet-of-things (iot)-enabled mobile ad hoc networks. this paper proposes a contention-free token-based multichannel mac protocol for iot-enabled cognitive radio ad hoc networks (crahns). in this, secondary users of crahns detect activity on the wireless spectrum and then access idle channels licensed by primary users. crahns are divided into clusters, and the channel to use for transmission is determined dynamically from the probability of finding idle primary-user channels. the token-based mac window size is adaptive, with adjustment according to actual traffic, which reduces both end-to-end mac contention delay and energy consumption. high throughput and spatial reuse of channels can also be achieved using a dynamic control channel and dynamic schemes for contention windows. we performed extensive simulations to verify that the proposed method can achieve better performance in mobile crahns than other mac schemes can. keywords: internet of things iot, medium access control, cognitive radio ad hoc networks, multichannel, mac, contention-free 1. introduction internet of things (iot) is a global network of devices, each with a unique address and links to other devices. it allows devices (and their users) to communicate with other devices/users at any time, independent of location, network, or service provider. in recent years, the iot has become a topic of intense interest within the field of communication technology. in machine-to-machine (m2m) communication with portable devices, communication methods that depend on maintaining a fixed location will not meet the needs of human users, who will want to move the devices, so communication on the so-called internet of mobile things has become a common application [1]. iot applications such as smart sensors, smart home applications, and monitoring devices must be connected by wireless transmission technology to achieve ubiquitous information access and seamless communication if they are to fulfill the promise of the iot [2]. mobile ad doc networks use a peer-to-peer, infrastructure-free decentralized wireless network and are easy to construct. because of this, there are many practical uses for such networks, including for personal and home use, military use, and facilitation of emergency rescue operations. the nodes of such networks can be m2m iot nodes such as smartphones and smart-sensor nodes. a mobile ad hoc network (manet) is a typical example of an iot-enabled mobile network [3-4]. so-called cognitive radios can make networks more efficient. the main feature of cognitive radio is that it can both sense and access different channels on the wireless spectrum. when a part of the spectrum licensed by primary users (pus) is idle, secondary users (sus) can take advantage of this. sus temporarily uses the licensed spectrum to complete communication * corresponding author. e-mail address: cmwu@nhu.edu.tw tel.: +886-5-6315368; fax: +886-5-6314486 advances in technology innovation, vol. 5, no. 1, 2020, pp. 45-55 46 without interfering with pus and without interfering with other sus, thereby improving the utilization of the wireless spectrum. this concept is part of the next generation of network technology and is known as dynamic spectrum access to form cognitive radio networks. a network of iot devices with cognitive radios will be able to use a dynamic spectrum access scheme to find an available channel in a way that does not interfere with pus’ ability to rely on iot communication within manets. in [5-6], the authors proposed a busy tone-based mac protocol. this protocol uses data-transmission priority to reduce delay. nodes access a channel by checking the broadcast busy tone and then transmitting data when the channel is available. the busy tone prevents the channel from being temporarily grabbed by a general data node. however, under this method, when the amount of data (such as multimedia materials) to be transmitted increases, the functionality of the busy tone will not be sufficient, and the amount of delay for data and multimedia delivery will not be guaranteed. in time-dynamic multiple access (tdma) schemes, as a method to optimize resource utilization, each node is assigned a time slot to avoid collisions caused by contention. in [7], the author proposed a decentralized tdma mechanism for adaptive control of data traffic. in [8], the authors proposed the tdma-based “distributed packet reservation multiple access” (d-prma) scheme, in which higher transmission priority is given to data and multimedia nodes than to general data nodes. however, the number of time slots allocated to the data and multimedia nodes in d-prma is determined from the total number of nodes, meaning that when the number of nodes increases, the system will not scale well. in another approach, a tdma mechanism that combines data and multimedia transmission in a hybrid mac protocol based on the carrier-sense with multiple access with collision avoidance protocol (i.e., csma/ca) has been proposed [9] for ensuring the quality of service (in terms of time delay) for data and multimedia nodes and maximizing channel utilization for general data transmission. in [10-11], the authors suggest that the contention-window size in the mac protocol should be adjusted according to node density and node movement speed to ensure that system performance is not degraded owing to changes in data traffic. clustering is the process of partitioning nodes in a network into many clusters for the purpose of improving system performance. in general, the clustering of nodes that sense the wireless network can improve the scalability and stability of the system. clustering also provides opportunities for nodes in clusters to cooperate on channel sensing and access [12-13]. in an iot-based wireless sensor network using tdma, the nodes are partitioned into multiple clusters. a mac protocol to reduce energy consumption and delays by collecting data within clusters (intra-cluster data) and between clusters (inter-cluster data) has been proposed [14]. ieee 802.11ah is a mac protocol that operates over long distances at low frequencies with low power consumption and allows for a large number of m2m iot nodes [15]. however, the mac protocols in [14-15] cannot satisfy the need for differentiated quality-of-service (qos) within iot manets. an opportunistic spectrum access mac protocol has also been proposed for a cluster-based multichannel wireless network [16]. the authors prove that the nodes using this protocol must repeatedly contend without clustering. multiple occurrences of contention lead to increased delays in data delivery, and increased delays reduce system performance. in [17], the authors proposed a token-based mac protocol (ta-mac) for a mobile network. ta-mac operates in a fixed channel and two-hop environment. however, in real environments, channels are a relatively rare and valuable resource. thus, most systems do not have a fixed channel on which they operate. in [18], the authors proposed a distributed mac protocol (dah-mac) for a manet. however, dah-mac can only be used in a fully interconnected one-hop environment and only in a single-channel environment. in practice, general manets are multi-hop and multichannel environments. toward addressing the deficits of existing protocols, in this paper, a contention-free token-based mac protocol is proposed for an iot-enabled multichannel multi-hop manet based on cognitive radio. in the proposed protocol, the nodes will be divided into some clusters, and the proposed contention-free reservation mechanism is based on a token ring to ensure qos for iot delay. advances in technology innovation, vol. 5, no. 1, 2020, pp. 45-55 47 the remainder of this paper is organized as follows. the system model is introduced in section 2. the detailed token-based mac protocol for iot-enabled manets is introduced in section 3. performance evaluation is discussed in section 4, and the final section presents our conclusions. 2. system model in [17], the authors proposed a token-based adaptive mac protocol (ta-mac) for an iot manet. ta-mac can be used only in an existing single-channel and two-hop environment. however, in the real world, channels are a relatively rare and valuable resource. thus, most systems do not operate on a fixed channel, and the system performance when using a single channel will be much poorer than that when using multiple channels. pu gw su pu su fig. 1 system model for cr-based iot-enabled multichannel manet the use of multiple channels will solve the problems associated with the single-channel restriction if the contention among cognitive radio (cr) devices can be overcome in a cr-based iot-enabled manet. furthermore, though ta-mac can be applied to two-hop environments, there are still many problems to be solved before ta-mac can be applied in multichannel and multi-hop manets. these questions include avoiding interference with channel use by pus, switching between pu channels, and ensuring qos. the system model for a cr-based iot-enabled multi-channel manet is shown in fig. 1. 2.1. clustering in the paper, a cr-based mac protocol will be designed for iot multichannel and multi-hop manets. in these networks, the system nodes will be partitioned into several one-hop clusters, meaning the nodes in each cluster must be completely interconnected within the node. each cluster node will sense the pu channels. the idle pu channel with the highest probability of being successfully used will be chosen as the data channel, ensuring the best stability of the obtained data channel. the cluster formation and cluster head are determined by the degree of importance [18]. 2.2. dynamic data channel the token-based mac protocol presented in this paper is based on the tdma scheme. each su in the token-based mac protocol must receive a token to transmit a message. in traditional tdma, each node can transmit messages in its designated time slots, which are assigned in advance. in the proposed system, the receiver and transmitter information will be included in the token and exchanged with the next transmitter. when a node receives a token, it checks the destination address to see whether the address matches its own. if the destination address is not the node’s address, the node discards the token; otherwise, it accepts the token. choosing the data channel is done via a dynamic scheme, with the choice determined by the probability of successfully picking an idle pu channel. only one transceiver is used for each su node to reduce the hardware cost and improve the practical application. the cluster heads will sense the pu channels, and one of the idle channels will be selected as the data channel. the cluster head then broadcasts the data channel to the cluster members by use of a hello frame. the chosen channel is the dynamic data channel. advances in technology innovation, vol. 5, no. 1, 2020, pp. 45-55 48 3. detailed token-based mac protocol for iot-enabled manet in this paper, time is divided into a number of superframes, each of which has two phases (fig. 2). table 1 shows the symbols used in the data-channel design for our proposed token-based mac protocol. fig. 2 data-channel design of cr-based iot-enabled multichannel manet table 1 symbols used in describing data-channel design for our proposed token-based mac protocol tog t length of sub-periods to gateway fromgm number of mini-slots from gateway fromg t length of sub-periods from gateway intram number of mini-slots inside cluster intra t length of sub-periods inside cluster vt length for each data time slot tog n number of sub-periods to gateway tog tog vm n t length of togt fromg n number of sub-periods from gateway fromg fromg vm n t length of fromgt intran number of sub-periods inside cluster intra intra vm n t length of intrat togm number of mini-slots to gateway chthresholdp success probability threshold 3.1. control channel in fig. 2, the reservation period is divided into three sub-periods: togt , fromgt , and intrat . the sub-periods have togn , fromg n , and intran time slots, respectively. each time slot of togt , fromgt , and intrat contains some mini-time slots, named togm , fromg m , and intram , respectively. each mini-time slot has length vt , indicating that each node can send data within the slot vt period if the token is obtained at this time. therefore, the lengths of togt , fromgt , and intrat are tog tog vm n t , fromg fromg vm n t , and intra intra v m n t , respectively. in the sub-period togt , su nodes send data to the su gateway in the same cluster, and the gateway node acts as the relay node for multi-hop connections. in the sub-period, the su gateway sends data to su nodes in the same cluster. in the sub-period intrat , sus transmits (pairwise) within the same cluster. when the su source node and the su destination node are not in the same cluster, an appropriate token must be acquired before the end of the two sub-periods togt and fromgt , and the data is then sent in each of the selected sub-time slots togm and fromgm . if the su source node and the su destination node are advances in technology innovation, vol. 5, no. 1, 2020, pp. 45-55 49 in the same cluster, then an appropriate token is acquired and transmission is completed in the sub-period intrat . whenever a su node sends data, the token is also sent in a round-robin scheme; the token contains the id of the next su node. the selection of the next su node is determined by the su sender that holds the token. the token is preferentially sent to a su node that has not yet obtained the token during the superframe to ensure that allocation of transmission opportunities is fair. 3.2. detailed token-based mac protocol for iot-enabled manet algorithm 1 shows pseudocode for selection of the dynamic data channel in the token-based mac protocol. algorithm 2 shows pseudocode for our proposed token-based mac protocol in cognitive radio ad hoc networks (crahns). algorithm 1: dynamic data channel selection in token-based mac protocol.  clusters formation completed.  each su node sends an important degree to its one-hop neighbors.  the su node with the highest degree of importance among one-hop neighbors is chosen as headsu and announces itself as the cluster head.  the data channel is each cluster will be selected by the cluster head according to an important degree for each channel. if pu active on equals data channel then. headsu searches a new data channel according to an important degree. else continue end if algorithm 2: token-based mac protocol in crahns.  initially, the frame lengths of togt , fromgt , and intrat are equal  isu judges its own role and waits for a token in the sub-periods togt , fromgt , and intrat .  isu checks the number of token accesses among one-hop neighbors, using a token-ring node table  isu sends the token to the next su, choosing the su with the lowest number of token accesses 3.3. sensing-window phase the time-synchronization function of ieee 802.11 is used for synchronization, and the cluster head transmits a hello frame to all cluster members. in previous studies, it is typical to switch to each potential channel when searching for a channel, but listening to each channel consumes more energy than selective listening. additionally, sensing each pu channel may fail due to sudden activity by a pu or to sensing errors. each su must thus check for each pu in each time slot in the sensing-window phase. the duration of each sensing time slot is about the same as the round-trip time when the su listens for a pu. therefore, we will use the probability of success to decide whether to listen to each channel; this proposed scheme reduces energy consumption. the number of time slots in the sensing-window phase is the number of pu channels. each cluster head can adjust its success probability threshold ( chthresholdp ) so as to achieve the highest efficiency. if a su cannot find a pu spectrum to use during the duration of the beacon interval, then this su must find a pu channel during the next beacon interval. each communication will be attempted a maximum of three times. after three failures, communication is advances in technology innovation, vol. 5, no. 1, 2020, pp. 45-55 50 considered to have failed. when communication has failed, it can remain and contend with other communications in later windows. 3.4. reservation-period phase the reservation period divides the time into a number of time slots and is subdivided into three sub-periods. any node that wants to transmit data must take turns during this period so as to achieve fairness in a probabilistic sense. a su node that obtains the token can transmit data, with the token piggybacked on the data. the token is granted to the next node based on a fairness scheme. if a node is given the token but has no data to transmit, the token is passed to the next node. 3.5. mac protocol dynamic-data-channel table maintenance and frame design to follow the above steps, each su node must maintain a channel-status recording table, a hello recording table, and a token-ring node table. in the iot-based multichannel manet, the hidden problems of inter-cluster and intra-cluster communications can be overcome by the tdma scheme, and qos for data transmission delay can be ensured. in addition, the proposed method also reduces energy consumption by nodes and increases system throughput. table 2 symbols used in control frames for our proposed token-based mac protocol hellosuccess n number of successes chsuccessp probability of successful channel hellofail n number of failures chthresholdp minimum critical probability hellosuccess p probability of success idneighbor number of one-hop neighbors id dvc id for dynamic data channel tokennum number of access tokens received id head id for cluster head role role of neighbor node chsuccess n number of successes for each channel intra intra vm n t length of intrat chfail n number of failures for each channel chthresholdp success probability threshold id node id for the node status join, leave, or dissolve in the cluster id nexthop node id of the next-hop idxsu ids for su neighbor sender su su sender slotreserve number of reserved time slots receiver su su receiver nbrihead number of the neighbor cluster for idhead res power remaining battery energy nbridvc data channel id 3.5.1. hello recording table each su node must maintain a hello recording table that records the number of successes ( hellosuccessn ), the number of failures ( hellofailn ), and the probability of success ( hellosuccessp ). the hello recording table has the following fields (see fig. 3): iddvc , idhead , hellosuccessn , hellofailn , and hellosuccessp . iddvc is the dynamic data channel selected by the cluster head. idhead is the id of the cluster head. iddvc id head hellosuccess n hellofail n hellosuccess p fig. 3 hello recording table 3.5.2. channel-status recording table (csrt) iddvc id head chsuccess n chfailn chsuccess p chthreshold p fig. 4 channel-status recording table the cluster head maintains a csrt, which contains the number of successes ( chsuccessn ) and failures ( chfailn ) for each advances in technology innovation, vol. 5, no. 1, 2020, pp. 45-55 51 channel. the su node selects a channel according to the probability of success. iddvc is the dynamic-data channel selected by cluster head. idhead is the id of the cluster head. chsuccessp indicates the success probability for one channel. chthresholdp indicates the minimum threshold for the success probability of one channel. the csrt has the following fields (see fig. 4): iddvc , idhead , chsuccessn , chfailn , chsuccessp , and chthresholdp . 3.5.3. token-ring node table (trnt) each su node must maintain a trnt to record the status of the tokens in the neighbor nodes in the cluster. the trnt has the following fields (see fig. 5): iddvc , idhead , idneighbor , tokennum , and role . iddvc is the id of the data channel selected by cluster head. id head is the id of the cluster head. idneighbor contains a list of the ids for one-hop neighbor nodes. token num indicates the number of times that the su node received the token. role indicates whether the node is a cluster head, or cluster member or cluster gateway. iddvc id head id neighbor token num role fig. 5 token-ring node table 3.5.4. frame format hello frame: the synchronization frame sent by the cluster head to the su neighbor node. the frame includes two fields (see fig. 6): iddvc and idhead . iddvc is the id of the data channel selected by the cluster head. idhead is the id of the cluster head. iddvc idhead fig. 6 hello format announce frame: a node is to join or leave the cluster or, for the cluster head, to dismiss the cluster. including the following fields: iddvc , idhead , idnode , and status (fig. 7). iddvc is the id of the data channel selected by the cluster head. id head is the id of the cluster head. idnode is the node id. status indicates whether the node wants to join, leave, or dissolve the cluster. iddvc idhead idnode status fig. 7 announce format the token frame is the format of the field is as follows (see fig. 8): idnexthop id dvc id head sender su receiver su 1id su －－ idnsu idnsu idnsu slotreserve respower nbridvc nbrihead －－ nnbrdvc nbrnhead fig. 8 token format there is the node id of the next hop. iddvc is the id of the data channel selected by the cluster head. idhead is the id of the cluster head. sendersu is the sending su . receiversu is the receiving su . 1, ,id idnsu su are the ids of the su neighbors. slot reserve is the number of contiguous time slots that one su can occupy. res power is the battery energy remaining in the su advances in technology innovation, vol. 5, no. 1, 2020, pp. 45-55 52 transmitting node nbri dvc is the id of the data channel selected by the cluster head of the neighboring cluster i . nbrihead is the id of the cluster head of the neighboring cluster i . 4. performance evaluation in our simulation, 10 different topologies are created using 10 different seeds. the results of our simulation show the average values of the 10 different seeds. the simulation is implemented in the c programming language and run on the linux operating system. the traffic is assumed to be uniformly distributed across all nodes with various overall loads. the number of new connections per second is given as an arrival rate. the number of terminated connections per second is given as a departure rate. the multiplicative inverse of the departure rate is also the average lifetime of a connection. “pu on” indicates that a pu is in the active state. “pu off” indicates that a pu is in the idle state. energy detection is easy to implement and commonly used as a spectrum-sensing scheme for sus to sense the active status of pus [19]. the pu active/non-active state is randomly determined in our simulation. the sus is not always-on (i.e., not always transmitting messages), and the number of transmitting sus is not constant. therefore, the number of active pus and sus is not constant, which makes the token-based mac protocol proposed in this paper scalable. because the resources of pus are assigned to and paid for by particular communities (companies, etc.), control by a pu is not impossible in real applications, with the consequence that cognitive radio is a key component of good performance by iot-enabled manets. in all cognitive-radio applications, each node can use two transceivers for transmitting and receiving. in this paper, we need only one transceiver to support both transmission and receiving; we do not need two separate channels. in some scenarios, such as a large assembly, a parade, or a downtown area, there may be hundreds of devices. in the previously described scenario, the requirements for channel bandwidth were high but temporary. in all scenarios, channels are limited resources and controlled by some pus. therefore, a key factor effectively delivering on the promise of the iot is the cognitive radio technique. in our simulation, the number of sus was set to 400 and the number of pus to 8. the transmission ranges for sus and pus were 200 m and 300 m, respectively. the bounding region has dimensions 600 x 600 m 2 . the mean “pu on” duration is 300 s. table 3 shows the parameter values used in the simulation of our proposed token-based mac protocol. table 3 parameter values for our simulation of the proposed token-based mac protocol simulation time 10000 s number of pus 8 number of sus 400 departure rate 0.5 bounding region 600 m*600 m arrival rate 1, …………, 512 transmission range of su 200 m pu sensing error 0%, 10%, 20%, 30% transmission range of pu 300 m mean duration of “pu on” 300 s number of active pus 0, 1, 3 number of seeds 10 the channel spatial reuse,  , is defined as the number of ssu that can simultaneously use one idled licensed channel. we define ɛ as follows: 1 1 1 ( , ) ( ) n m singlei j n multiplei c i j c i         (1) therefore, the throughput per contention window size for this crahn, denoted by  , is defined as follows: [ ] [ ](( 1) data s vacant data ctrl t e ch r e t r n r      (2) advances in technology innovation, vol. 5, no. 1, 2020, pp. 45-55 53 where datar is the data rate for a pu licensed channel, ctrlr is the data rate for a control channel, st is the duration of each time slot, and  e t is the average duration of a beacon interval for the token-based mac. from fig. 9, we know that the token-based mac produced the highest spatial reuse of channels when the arrival rate was 512. we observed that the highest channel spatial reuse in the token-based mac was 4.33 (corresponding to a channel-detection probability = 1.0 and active pus = 1). fig. 9 comparison of channel spatial reuse in token-based mac as a function of arrival rate under different probabilities of channel detection in a multichannel crahn fig. 10 shows the throughput in the token-based mac as a function of the arrival rate in a multichannel crahn. we observed that the highest throughput in the token-based mac was 8,768 bps under an arrival rate of 256 (corresponding to a channel-detection probability = 1.0 and active pus = 1). fig. 10 comparison of throughput in token-based mac as a function of arrival rate under different probabilities of channel detection in a multichannel crahn fig. 11 shows the average mac contention delay per hop in token-based mac as a function of the arrival rate in a multichannel crahn. the average mac contention delay per hop for token-based mac ranged from 12.61 to 14.59 slots (corresponding to a probability of channel detection = 1.0 and active pus = 1) as a function of arrival rate. advances in technology innovation, vol. 5, no. 1, 2020, pp. 45-55 54 fig. 11 comparison of mac contention delay per hop in token-based mac as a function of arrival rate under different probabilities of channel detection in a multichannel crahn 5. conclusions this paper proposed a protocol that used a token-based dynamic-data channel and dynamic-contention window to achieve high channel spatial reuse, high throughput, energy-efficiency, and a low mac contention delay. in this paper, the proposed token-based mac scheme effectively achieved not only lower mac delay but also better per-hop energy efficiency. the simulation results had a best channel spatial reuse for token-based mac of 4.33, a maximum throughput of 8,768 bps under an arrival rate of 256, and a mac contention delay per hop of 12.61 to 14.59 slots, with the probability of channel detection at 1.0 and number of active pus at 1 for all of these values. it is notable that the channel spatial reuse and throughput decreased when the probability of channel detection decreased. conflicts of interest the authors declare no conflict of interest. acknowledgement the authors would like to thank the editor and the reviewers for their valuable comments and suggestions. this work is supported by ministry of science and technology, taiwan, r.o.c. under grant most 107-2221-e-343-001. references [1] a. a. khan, m. h. rehmani, and a. rachedi, “cognitive-radio-based internet of things: applications, architectures, spectrum related functionalities, and future research directions,” ieee transactions on wireless communications, vol. 24, no. 3, pp. 17-25, june 2017. [2] j. gubbi, r. buyya, s. marusic, and m. palaniswami, “internet of things (iot): a vision, architectural elements, and future directions,” future generation computer system, vol. 29, no. 7, pp. 1645-1660, september 2013. [3] h. nishiyama, t. ngo, s. oiyama, and n. kato, “relay by smart device: innovative communications for efficient information sharing among vehicles and pedestrians,” ieee transactions on vehicular technology magazine, vol. 10, no. 4, pp. 54-62, december 2015. [4] j. liu and n. kato, “a markovian analysis for explicit probabilistic stopping-based information propagation in post disaster ad hoc mobile networks,” ieee transactions on wireless communications, vol. 15, no. 1, pp. 81-90, january 2016. [5] p. wang, h. jiang, and w. zhuang, “capacity improvement and analysis for data/data traffic over wlans,” ieee transactions on wireless communications, vol. 6, no. 4, pp. 1530-1541, may 2007. advances in technology innovation, vol. 5, no. 1, 2020, pp. 45-55 55 [6] j. sobrinho and a. krishnakumar, “quality-of-service in ad hoc carrier sense multiple access wireless networks,” ieee journal on selected areas in communications, vol. 17, no. 8, pp. 1353-1368, august 1999. [7] p. wang and w. zhuang, “a collision-free mac scheme for multimedia wireless mesh backbone,” ieee transactions on wireless communications, vol. 8, no. 7, pp. 3577-3589, 2009. [8] s. jiang, j. rao, d. he, x. ling, and c. c. ko, “a simple distributed prma for manets,” ieee transactions on vehicular technology, vol. 51, no. 2, pp. 293-305, march 2002. [9] r. zhang, l. cai, and j. pan, “performance analysis of reservation and contention-based hybrid mac for wireless networks,” in proc. ieee icc’10, pp. 1-5, may 2010. [10] m. wang, q. shen, r. zhang, h. liang, and x. shen, “vehicle-density based adaptive mac for high throughput in drive-thru networks,” ieee internet things journal, vol. 1, no. 6, pp. 533-543, december 2014. [11] w. alasmary and w. zhuang, “mobility impact in ieee 802.11p infrastructureless vehicular networks,” ad hoc network, vol. 10, no. 2, pp. 222-230, march 2012. [12] k. l. a. yau, n. ramli, w. hashim, and h. mohamad, “clustering algorithms for cognitive radio networks: a survey,” journal of network and computer applications, vol. 45, pp. 79-95, october 2014. [13] m. singh and s. k. soni, “spatial correlation-based clustering in wireless sensor network,” international journal of engineering and technology innovation, vol. 7, no. 2, pp. 98-116, february 2017. [14] m. dong, k. ota, a. liu, and m. guo, “joint optimization of lifetime and transport delay under reliability constraint wireless sensor networks,” ieee trans. parallel distrib. syst., vol. 27, no. 1, pp. 225-236, january 2016. [15] m. park, “ieee 802.11ah: sub-1-ghz license-exempt operation for the internet of things,” ieee communication magazine, vol. 53, no. 9, pp. 145-151, september 2015. [16] a. azarfar, j.f. frigon, and b. sanso, “delay analysis of multichannel opportunistic spectrum access mac protocols,” ieee transactions on mobile computing, vol. 15, no. 1, pp. 92-106, january 2016. [17] q. ye and w. zhuang, “token-based adaptive mac for a two-hop internet-of-things enabled manet,” ieee internet of things journal, vol. 4, no. 5, pp. 1739-1753, october 2017. [18] q. ye and w. zhuang, “distributed and adaptive medium access control for internet-of-things-enabled mobile networks,” ieee internet of things journal, vol. 4, no. 2, pp. 446-460, may 2017. [19] x. l. huang, g. wang, f. hu, and s. kumar, “stability-capacity-adaptive routing for high-mobility multihop cognitive radio networks,” ieee transactions vehicular technology, vol. 60, no. 6, pp. 2714-2729, july 2011. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation, vol. 6, no. 1, 2021, pp. 11-20 impact of flax fiber reinforcement on mechanical properties of solid and foamed geopolymer concrete kinga korniejenko * , michał łach, maria hebdowska-krupa, janusz mikuła institute of material engineering, faculty of material engineering and physics, cracow university of technology, cracow, poland received 18 february 2020; received in revised form 02 june 2020; accepted 25 august 2020 doi: https://doi.org/10.46604/aiti.2021.5294 abstract the main objective of this study is to develop the advanced composites for civil engineering applications as material for the building industry, especially for an insulation purpose. the research processes include several steps. firstly, the prototype elements, such as bricks and elevation elements were performed from eco-friendly composite -flax fiber reinforced geopolymer. the elements were designed to take into consideration for environment. geopolymers are environmentally friendly, sustainable, and resource efficient, including energy demand. next, the wall was built from these elements and exposed during the three months in a relevant environment. the main conclusion of the research is that the kind of fibers is important for the mechanical properties of the composite, including the fact that for those different fibers could be more beneficial for different raw materials, giving higher strength properties. the significant influence on the mechanical properties of the composites has the adhesion between fiber and material used as a matrix. the adhesion depends among others on the previous treatment of the fibers. keywords: geopolymer, mechanical properties, foamed material, flax fiber 1. introduction the background for the article is a pressure for the change of the traditional economy that is based on a linear approach to a so-called circular economy in recent years. the main principle of the circular economy is designing out waste and pollution. the rational management of natural resources and the usage of waste materials are getting more and more important. it is also supported by the growing ecological awareness of society, including the consciousness of sustainable development [1-2]. one of the most promising technologies for a circular economy is the geopolymerization with using waste streams. this technology is dedicated to the construction industry. it has a much lower carbon footprint than the technology of manufacturing traditional construction materials [1, 3]. the geopolymer synthesis also enables one to use waste, such as mining tiles or industrial by-products as a raw material for manufacturing new products in low energy and low emission processes [4-5]. one of the limitations for the wide usage of this technology in construction industry is the brittle fracture behavior of the received materials and their relatively low tensile as well as flexural strength [1, 6]. in order to improve tensile and flexural strength, it is possible to reinforce the geopolymer matrix by fibers [2-3]. from a mechanical point of view, reinforcement by fibers is an efficient method to improve material properties, for instance, fracture toughness. the fibers reduce the general effects of cracking, limit the widths of the occurring cracks, especially the propagation of micro-cracks, suppress all brittle * corresponding author. e-mail address: kinga.korniejenko@pk.edu.pl tel.: +48 609974988 https://context.reverso.net/t%c5%82umaczenie/angielski-polski/need+for+a+change advances in technology innovation, vol. 6, no. 1, 2021, pp. 11-20 12 behavior, and enhance ductility [1, 3]. the fibers additions also increase the flexural strength of composites. the fibers improve energy absorption and resistance to deformation [3, 7-9]. in particular, the introduction of short fibers due to their easy fiber dispersion and fiber shape factor; it is an effective way to strengthen geopolymer materials [3]. currently, the reinforcements used in geopolymer composites are usually based on inorganic fibers, such as carbon or glass fibers [3, 10] or synthetic fibers [11]. however, the new research direction usage of eco-friendly materials is clearly visible. from the environmental point of view, the addition of natural fibers is particularly beneficial [1, 12]. this solution is an environmentally friendly alternative to inorganic fibers [1, 13]. nevertheless, natural fibers usually have lower mechanical properties. it does not always allow their desired applications. natural fibers have lower dimensional repeatability than artificial fibers, but these fibers have also many advantages [14-15]. the most important features are: reduction of the environmental impact (closing important life cycles, including co2), low cost of production, low density; the fibers are usually renewable in a short time, non-toxic, and easy to process [1, 16]. the research shows natural fibers could replace the conventional ones in many applications; and gives added value, especially reduction of greenhouse gas emissions and negative environmental impacts [1-2]. contemporary, the flax is one of the most popular fibers used as reinforcement to geopolymer matrix [17]. the research was made on the matrix based on dehydroxylated kaolinite-type clay (new zealand halloysite imerys premium grade) [18], low calcium fly ash (from the eraring power station in nsw) [19-20], fly ash with nano-clay particles [21-22]. the flax fibers were added in an amount: 4 10 wt%. the research results pointed out that mechanical properties are improved by fiber addition, especially flexural strength [17, 23]. the other research shows the possibility of using the geopolymers reinforced by short flax fibers in additive manufacturing technology [24-26]. there were several attempts to investigate foamed geopolymers reinforced by fibers. most of the research was dedicated to reinforcement by artificial fibers such as carbon [27-28] or glass [29-30], including waste products [29-30]. some of the investigations have been done on the composites with natural fibers such as hemp [31-32], abaca [33], cellulose [34-35] and jute [36]. the mineral fibers; for example, basalt have been also tested [28, 37]. the hemp (cannabis sativa) fiber is one of the most widely applied cellulosic fibers as reinforcement in different kinds of composites [31]. the two amounts of short hemp fibers were applied 1.13 and 4.53% [32]. the results show good bonding between matrix and reinforcement, good stability to the thermal variation and improvement of mechanical properties [32]. the research conducted with abaca fibers shows its improving properties [33], including increasing compressive strength and flexural ranged from 19.6 mpa to 36.8 mpa and 2.4 mpa to 6.3 mpa, respectively [33]. the addition of jute fibers also positively influenced on mechanical properties, flame, and water resistance [37]. the investigation with cellulose shows the positive influence of this addition on ph and the possibility of applying this material for eco-friendly pots [34-35]. the flax fibers in foamed geopolymers were only investigated as an addition to long fibers [36]. the results show the positive effect of the addition to flax fibers on impact bending and flexural strength [38]. additionally, the flax fibers were compared with basalt fibers (both were added in amount 2.16 wt%). the results show a different pattern of failure mechanisms dependent on applied reinforcement [38]. the main purpose of the provided research is to develop the advanced composites for civil engineering applications, as material for the building industry. this paper shows the possibilities of using waste streams coming from energy industry (fly ashes and agriculture waste) and short flax fibers (residues for linen production), as raw materials to produce high-added value lightweight and low thermal conductivity geopolymer composites, including foamed products. these kinds of materials could be used as the advanced composites for civil engineering applications, as a construction material as well as insulation one and also for rehabilitating deteriorated concrete [39]. the previous study does not include semi-industrial tests for the foamed geopolymers reinforced flax fibers [40-41]. advances in technology innovation, vol. 6, no. 1, 2021, pp. 11-20 13 2. materials and methods 2.1. materials two kinds of materials have been prepared: solid and foamed. the solid geopolymers based on the fly ash from the chp plant in the city called skawina (lesser poland region, poland) and sand in ratio 1:1. this fly ash has the physical and chemical properties that support the geopolymerization process [42-44]. the second geopolymers according to the metakaolin and sand in ratio 1:1. both kinds of samples were reinforced with 1% by mass of flax fibers. in order to reinforce, the following flax fibers were applied as compositions to solid materials: (1) raw flax fiber shortened (raw straw). (2) long flax fiber after dew retting. (3) cottonized flax fiber. the foamed materials the following flax fibers were applied to: (1) cottonized flax fiber internal layer. (2) shortened flex fiber after water retting external and internal layer. (3) short flax fiber external layer. 2.2. samples preparation samples were prepared using sodium promoter, fly ash or metakaolin, sand, and various flax fibers (each 1%, by mass of the composite). the process of activation has been made by a 10m sodium hydroxide solution combined with the sodium silicate solution (liquid glass at a ratio of 1:2.5). the geopolymers were produced using technical sodium hydroxide flakes and an aqueous sodium silicate solution of r-145 (module 2.5), with a molar density of about 1.45 g/cm 3 . the tap water was used instead of the distilled one. in the case of the foamed plated, the hydrogen peroxide has been used 3.5% for metakaolin and 2% for fly ash based geopolymers. the alkaline solution was prepared by pouring the aqueous solution of sodium silicate over the solid sodium hydroxide. the solution was mixed and allowed to equilibrate until a constant concentration and temperature were achieved. it took nearly 2 hours. the fly ash or metakaolin, sand, alkaline solution, and fibers were mixed for about 15 minutes by using a low-speed mixing machine (to receive the homogeneous paste). on this stage for the foamed materials, a proper amount of hydrogen peroxide was introduced and continuously mixed for approximately 1 min. after the mixing process was complete, the material was poured into a mold and exposed to a temperature of 75 °c for 24 hours. after the heating process, the elements were dismantled and stored under laboratory conditions for a period of 28 days to obtain the full strength. the following kind of samples were prepared (table 1. and table 2.): (1) bricks-solid material  metakaolin with a cottonized flax fibre.  metakaolin with a long flax fibre after dew retting.  metakaolin with a raw flax fibre shortened (raw straw).  fly ash (fa) with a cottonized flax fibre.  fa with a raw flax fibre-shortened (raw straw). the detailed information about composition of prepared samples solid bricks is included in table 1. advances in technology innovation, vol. 6, no. 1, 2021, pp. 11-20 14 table 1 composition of the solid samples bricks no. sample 1 metakaolin with a cottonized flax fiber 2 metakaolin with a long flax fiber after dew retting 3 metakaolin with a raw flax fiber shortened (raw straw) 4 fly ash (fa) with a cottonized flax fiber 5 fa with a raw flax fiber shortened (raw straw) (2) elevation plates-the internal layer (solid material, ca. 1 cm) and external layer (foamed material, ca. 4 cm) a sketch of the elevation plate are presented in fig. 1:  external layer: metakaolin with the shortened flex fibre after water retting, internal layer: metakaolin with the shortened flex fibre after water retting (metakaolin: sand ratio 0.85:0.15, hydrogen peroxide 3.5%).  external layer: fa with the short flax fibre, internal layer: fa with the cottonized flax fibre (fa: sand ratio 0.75:0.25, hydrogen peroxide 2.0%).  external layer: fa with the shortened flex fibre after water retting, internal layer: fa with a shortened flex fibre after water retting (fa: sand ratio 0.85:0.15, hydrogen peroxide 2.0%). fig. 1 sketch of the elevation plate the detailed information about composition of prepared samples-elevation plates is included in table 2. table 2 composition of the elevation plates no. sample external layer sample internal (foamed) layer h2o2 1 metakaolin with a shortened flex fiber after water retting metakaolin with a shortened flex fiber after water retting (metakaolin: sand ratio 0.85:0.15) 3.5% 2 fa with a short flax fiber fa with a cottonized flax fiber (fa: sand ratio 0.75:0.25) 2.0% 3 fa with a shortened flex fiber after water retting fa with a shortened flex fiber after water retting (fa: sand ratio 0.85:0.15) 2.0% fig. 2 the wall made from bricks and elevation plates – first layer fig. 3 the wall made from bricks and elevation plates build on the university campus the both prototype solutions are presented in fig. 2 bricks (small elements) and elevation plates (large elements). in fig. 2, there were two solid bricks-lower based on metakaolin (white color), upper-based on fly ash (gray color). the brick size was: advances in technology innovation, vol. 6, no. 1, 2021, pp. 11-20 15 25 cm × 12 cm × 5 cm. the plate’s size was: 40 cm × 24 cm × 5 cm (fig. 1). the gray elevation plates based on the fly ash reinforced by flax fibers and white elevation plates based on metakaolin reinforced by flax fibers are also shown in fig. 2 (right site). eventually, the two products have been used for creating the prototype wall (fig. 3) for testing in a relevant environment. the wall was placed on the campus of the cracow university on technology (jana pawła ii street, cracow) and left on the open-air for 3 months. next, it was demolished and the chosen parts were investigated according to mechanical properties. 2.3. methods compressive strength tests were carried out according to the methodology described in the standard en 12390-3 (‘testing hardened concrete. compressive strength of test specimens’), because of the lack of separate standards for geopolymer materials. the tests involved cubic samples: 50 × 50 × 50 mm. the samples have been cut from large plates previously prepared (wall elements). tests were performed on a concrete press matest 3000kn with speed of 0.5 [mpa/s]. bending tests were carried out according to the methodology described in the standard en 12390-5 (‘testing hardened concrete. flexural strength of test specimens’), because of the lack of separate standards for geopolymer materials. the tests involved prismatic samples: 50 × 50 × 200 mm (space between supporting points 150 mm). the samples have been cut from large plates previously prepared (wall elements). tests were performed on a universal testing machine-matest 3000kn with speed of 0.05 mpa/s. the calculations were based on the following equation: 3 1 2 cf fl f d d  (1) where3 fcf compressive strength, mpa f maximal load, n l space between supporting points, mm (for conducted tests: 140 mm) d1, d2 sample dimensions, mm 3. experimental results and discussion 3.1. compressive strength the results of the compressive strength tests are shown in table 3 and table 4. table 3 results of compressive strength tests: bricks-solid material no. sample mpa 1 metakaolin with a cottonized flax fiber 66.4 2 metakaolin with a long flax fiber after dew retting 67.6 3 metakaolin with a raw flax fiber shortened (raw straw) 42.9 4 fa with a cottonized flax fiber 58.0 5 fa with a raw flax fiber shortened (raw straw) 90.8 the tests were provided on minimum of 6 samples. the compressive strength test results achieved for bricks (solid material) were between 42.9 to 90.8 mpa. the received values for all types of materials are better than for conventional concrete. the best results were achieved for a fly ash based geopolymer with the raw flax fiber 90.8 mpa. the results were drastically different than for the geopolymer concrete based on the metakaolin with the raw flax fiber-shortened (raw straw) advances in technology innovation, vol. 6, no. 1, 2021, pp. 11-20 16 1% by mass. in the case of the cottonized flax fiber, better results were received for the metakaolin basedmatrix than for fly ash based one. the reverse results were achieved for composites reinforced by raw straw fibers. the better results were for fly ash based geopolymer than metakaolin based material. the investigation was also provided on elevation plates that consist of the foamed material. the plates included elevation part with the solid material and foamed one isolation layer (fig. 4). the grey elements are elevation plates based on the fly ash reinforced by flax fibers in fig. 4. there are visible fibers agglomeration (marked in red arrow). the agglomeration is probably the consequence of not effective mixing process during material manufacturing. it could have an influence on material properties. the white elevation plates based on the metakaolin reinforced by flax fibers are also shown in fig. 4. the layers have been produced with similar materials but based on the metakaolin. the material was also reinforced by flax fibers. the plates did not need additional join between the layers geopolymer binder bounded them. fig. 4 the elevation plates with visible layers and agglomeration of fibers the mechanical properties for foamed samples are presented in table 4. table 4 results of compressive strength tests: elevation plates foamed material no. sample external layer sample internal (foamed) layer h2o2 mpa 1 metakaolin with a shortened flex fiber after water retting metakaolin with a shortened flex fiber after water retting (metakaolin: sand ratio 0.85:0.15) 3.5% 6.8 2 fa with a short flax fiber fa with a cottonized flax fiber (fa: sand ratio 0.75:0.25) 2.0% 6.9 3 fa with a shortened flex fiber after water retting fa with a shortened flex fiber after water retting (fa: sand ratio 0.85:0.15) 2.0% 4.3 the mechanical properties of foamed materials are much lower than solid ones. the same mechanism is presented for all formed materials. the results of compressive strength tests for elevation plates are in range: 4.3 6.9 mpa. the lowest results of compressive strength are for the geopolymer concrete based on fly ash with the shortened flex fiber after water retting 1% by mass. it is almost 38% lower than for other investigated materials based on the metakaolin and fly ash. the pours in foamed layers are open. 3.2. flexural strength the results of the flexural strength tests are shown in table 5 and table 6. the best results were achieved for geopolymer concrete based on the metakaolin with different fibers. this value is definitely better than for traditional cements. as the same as in the case of compressive strength, the foamed materials have much lower properties than solid ones. the flexural strength test results for bricks were between 7.2 and 10.7. the highest value was received for the metakaolin with the cottonized flax fiber 10.7 mpa. the results of flexural strength tests for elevation plates are in the range 0.05 0.86 mpa. however, the results are comparable or even better than other foamed materials accessible on the market, especially foamed organic polymers. the provided research did not show any deterioration process. advances in technology innovation, vol. 6, no. 1, 2021, pp. 11-20 17 table 5 results of flexural strength tests: bricks solid material no. sample mpa 1 metakaolin with a cottonized flax fiber 10.7 2 metakaolin with a long flax fiber after dew retting 10.6 3 metakaolin with a raw flax fiber shortened (raw straw) 7.2 4 fa with a cottonized flax fiber 7.6 5 fa with a raw flax fiber shortened (raw straw) 9.4 table 6 results of flexural strength tests: elevation plates foamed material no. sample external layer sample internal (foamed) layer h2o2 mpa 1 metakaolin with a shortened flex fiber after water retting metakaolin with a shortened flex fiber after water retting (metakaolin: sand ratio 0.85:0.15) 3.5% 0.86 2 fa with a short flax fiber fa with a cottonized flax fiber (fa: sand ratio 0.75:0.25) 2.0% 0.05 3 fa with a shortened flex fiber after water retting fa with a shortened flex fiber after water retting (fa: sand ratio 0.85:0.15) 2.0% 0.65 3.3. the effect of flax fibers addition the article presents the impact of flax fiber reinforcement on the mechanical properties of solid and foamed geopolymer concrete. the prototype elements such as bricks (made from the solid material) and elevation elements (plates including solid and foamed parts) were manufactured and investigated. the two kinds of raw materials have been tested (fly-ash and metakaolin). the research on the mechanical properties compressive and flexural strength were conducted. the compressive strength results for the new materials (solid) were between 42.9 and 90.8 mpa. these values were achieved for the samples with the raw flax fibers shortened (raw straw). the results depended on used raw materials. the other tendency was observed for the samples reinforced by the cottonized flax fibers. in this case, the material based on the metakaolin had a higher compressive strength 66.4 mpa, compared with the samples where fly ash was applied as a raw material 58.0 mpa. the same tendency was noticed in the results of the flexural strength test. for the samples reinforced by the raw flax fibers-shortened (raw straw), better results were achieved for the material based on fly ash 9.4 mpa than for based on the metakaolin 7.2 mpa. the behavior of the material reinforced by cottonized flax fibers for the flexural strength test was similar to the results of the compressive strength test. the composite based on the metakaolin had higher values for the flexural strength 10.7 mpa than the material based on fly ash 7.6 mpa. it shows that the kind of fiber is important for the mechanical properties of the composite, including the fact that for the different raw materials different fibers could be more advantages. the significant influence in this case has the adhesion between the matrix and reinforcement. the adhesion is depended among others on the previous treatment of the fibers. the significant meaning could have also the reaction between fiber and material used as a matrix. the elevation elements (plates including solid and foamed parts) reinforced by the shortened flex fiber after water retting were also compared. the plates based on the metakaolin have better mechanical properties than based on the fly ash. the values for the plates based on the metakaolin were 6.8 mpa for compressive strength and 0.86 mpa for flexural strength. for the composition based on the fly ash, the relevant values were 4.3 mpa and 0.65 mpa, respectively. the different amount of hydrogen peroxide applied in these compositions was not consider, but a larger amount of hydrogen peroxide was added to the material based on the metakaolin; therefore, this material should potentially have lower mechanical properties. 4. conclusions the research shows the possibility of manufacturing the products for construction industry based on the eco-friendly materials. the composites comprise the waste flax fiber and the reinforced geopolymer concrete. the geopolymers includes advances in technology innovation, vol. 6, no. 1, 2021, pp. 11-20 18 the fly ash as well as the metakaolin. the new products were designed to take into consideration for environment, including the energy efficiency, reduction of carbon footprint, waste reduction, and development of a circular economy. the new composites have comparable properties with the results achieved for traditional foamed and solid building materials. the main conclusion of the research is the kind of fibers is important for the mechanical properties of the composite, including the fact that for that different fibers could be more beneficial for different raw materials, giving higher strengt h properties. the significant influence on the mechanical properties of the composites has the adhesion between fiber and material used as a matrix. the adhesion depends among others on the previous treatment of the fibers. the contemporary research gives promising results, but further research is required. the practice applications need further tests to optimize the mechanical properties of the composites as well as investigate the other properties such as water absorption and resistance for a different environments, including long-durability issues. the lca analysis of the chosen products will be also desirable. acknowledgement this work has been financed by the national centre for research and development in poland under the grant: low thermal conductivity geopolymer foams produced with industrial wastes as an innovative material for circular economy (lider/31/0168/l-10/18/ncbr/2019) and the eranet-lac 2nd joint call (http://www.eranet-lac.eu) within the framework of the grant: development of eco-friendly composite materials based on geopolymer matrix and reinforced with waste fibers. conflicts of interest the authors declare no conflict of interest. references [1] k. korniejenko, e. frączek, e. pytlak, and m. adamski, “mechanical properties of geopolymer composites reinforced with natural fibers,” procedia engineering, vol. 151, pp. 388-393, august 2016. [2] k. korniejenko, m. łach, n. doğan-sağlamtimur, g. furtos, and j. mikuła, “fibre reinforced geopolymer composites a review,” 1st international conference on environment, technology and management (icetem), june 2019, pp. 3-13. [3] k. korniejenko, “the influence of short fibres on the properties of composites with geopolymer matrix,” (in polish) ph.d. dissertation, faculty of mechanical engineering, cracow university of technology, cracow, poland, 2018. [4] r. m. novais, j. carvalheiras, m. p. seabra, r. c. pullar, and j. a. labrincha, “innovative application for bauxite residue: red mud-based inorganic polymer spheres as ph regulator,” journal of hazardous materials, vol. 358, pp. 69-81, september 2018. [5] r. m. novais, g. ascensão, n. ferreira, m. p. seabra, and j. a. labrincha, “influence of water and aluminum powder content on the properties of waste-containing geopolymer foams,” ceramics international, vol. 44, no. 6, pp. 6242-6249, april 2018. [6] r. a. sá ribeiro, m. g. sá ribeiro, and w. m. kriven, “a review of particle and fiber-reinforced metakaolin-based geopolymer composites,” journal of ceramics science and technology, vol. 8, no. 3, pp. 307-322, august 2017. [7] g. furtos, l. silaghi-dumitrescu, p. pascuta, c. sarosi, and k. korniejenko, “mechanical properties of wood fiber reinforced geopolymer composites with sand addition,” journal of natural fibers, vol. 1, pp. 1-12, june 2019. [8] a. r. sakulich, “reinforced geopolymer composites for enhanced material greenness and durability,” sustainable cities and society, vol. 1, no. 4, pp. 195-210, december 2011. [9] r. chen, s. ahmari, and l. zhang, “utilization of sweet sorghum fiber to reinforce fly ash-based geopolymer,” journal of material science, vol. 49, no. 6, pp. 2548-2558, december 2013. [10] t. lin, d. jia, p. he, and m. wang, “in situ crack growth observation and fracture behavior of short carbon fiber reinforced geopolymer matrix composites,” materials science and engineering: a, vol. 527, no. 9, pp. 2404-2407, april 2010. advances in technology innovation, vol. 6, no. 1, 2021, pp. 11-20 19 [11] l. yan, b. kasal, and l. huang, “a review of recent research on the use of cellulosic fibres, their fibre fabric reinforced cementitious, geo-polymer and polymer composites in civil engineering,” composites part b: engineering, vol. 92, pp. 94-132, may 2016. [12] a. m. rashad, “a synopsis about the effect of basalt and natural fibers on geopolymer properties,” natural resources conservation and research, vol. 1, pp. 1-9, 2018. [13] g. silva, s. kim, r. aguilar, and j. nakamatsu, “natural fibers as reinforcement additives for geopolymers-a review of potential eco-friendly applications to the construction industry,” sustainable materials and technologies, vol. 23, e00132, april 2020. [14] h. assaedi, f. u. a. shaikh, and i. m. low, “effect of nanoclay on durability and mechanical properties of flax fabric reinforced geopolymer composites,” journal of asian ceramic societies, vol. 5, no. 1, pp. 62-70, march 2017. [15] t. alomayri, f. u. a. shaikh, and i. m. low, “thermal and mechanical properties of cotton fabric-reinforced geopolymer composites,” journal of material science, vol. 48, no. 19, pp. 6746-6752, june 2013. [16] f. pacheco-torgal and s. jalali, “cementitious building materials reinforced with vegetable fibers: a review,” construction and building materials, vol. 25, no. 2, pp. 575-582, february 2011. [17] k. korniejenko, m. łach, m. hebdowska-krupa, and j. mikuła, “the mechanical properties of flax and hemp fibres reinforced geopolymer composites,” iop conference series: materials science and engineering, vol. 379, 012023, may 2018. [18] m. alzeer and k. mackenzie, “synthesis and mechanical properties of novel composites of inorganic polymers (geopolymers) with unidirectional natural flax fibres (phormium tenax),” applied clay science, vol. 75-76, pp. 148-152, may 2013. [19] h. assaedi, t. alomayri, f. u. a. shaikh, and i. m. low, “synthesis and mechanical properties of flax fabric reinforced geopolymer composites,” advances in materials research, vol. 3, no. 3, pp. 151-161, october 2014. [20] h. assaedi, t. alomayri, f. u. a. shaikh, and i. m. low, “characterisation of mechanical and thermal properties in flax fabric reinforced geopolymer composites,” journal of advanced ceramics, vol. 4, no. 4, pp. 272-281, 2015. [21] h. assaedi, f. u. a. shaikh, and i. m. low, “characterizations of flax fabric reinforced nanoclay-geopolymer composites,” composites part b: engineering, vol. 95, pp. 412-422, june 2016. [22] h. assaedi, f. u. a. shaikh, and i. m. low, “effect of nanoclay on durability and mechanical properties of flax fabric reinforced geopolymer composites,” journal of asian ceramic societies, vol. 5, no. 1, pp. 62-70, march 2017. [23] g. silva, s. kim, a. castañeda, r. donayre, j. nakamatsu, r. aguilar, et al., “a comparative study of linen (flax) fibers as reinforcement of fly ash and clay brick powder based geopolymers,” iop conference series: materials science and engineering, vol. 416, 012107, 2018. [24] k. korniejenko, m. łach, s. y. chou, w. t. lin, j. mikuła, d. mierzwiński, et al., “a comparative study of mechanical properties of fly ash-based geopolymer made by casted and 3d printing methods,” iop conference series: materials science and engineering, vol. 660, 012005, 2019. [25] k. korniejenko, m. łach, s. y. chou, w. t. lin, a. cheng, m. hebdowska-krupa, et al., “mechanical properties of short fiber-reinforced geopolymers made by casted and 3d printing methods: a comparative study,” materials, vol. 13, no. 3, 579, january 2020. [26] k. korniejenko and m. łach, “geopolymers reinforced by short and long fibres – innovative materials for additive manufacturing,” current opinion in chemical engineering, vol. 28, pp. 167-172. june 2020. [27] j. h. lee, s. wattanasiriwech, and d. wattanasiriwech, “preparation of carbon fiber reinforced metakaolin based-geopolymer foams,” key engineering materials, vol. 766, pp. 19-27, april 2018. [28] r. krzywoń, j. hulimka, a. jędrzejewska, and m. górski, “application of fibre composite grids as reinforcement of foamed pc and gp concrete,” matec web of conferences, vol. 274, 05002, 2019. [29] r. m. novais, r. c. pullar, and j. a. labrincha, “geopolymer foams: an overview of recent advancements,” progress in materials science, vol. 109, 100621, april 2020. [30] l. senff, r. m. novais, j. carvalheiras, and j. a. labrincha, “eco-friendly approach to enhance the mechanical performance of geopolymer foams: using glass fibre waste coming from wind blade production,” construction and building materials, vol. 239, 117805, april 2020. [31] e. a. correia, s. m. torres, m. e. o. alexandre, k. c. gomes, n. p. barbosa, and s. d. e. barros, “mechanical performance of natural fibers reinforced geopolymer composites,” materials science forum, vol. 758, pp. 139-45, june 2013. [32] b. galzerano, a. formisano, m. durante, f. iucolano, d. caputo, and b. liguori, “hemp reinforcement in lightweight geopolymers,” journal of composite materials, vol. 52, no. 17, pp. 2313-2320, november 2017. advances in technology innovation, vol. 6, no. 1, 2021, pp. 11-20 20 [33] j. p. s. ngo and m. a. b promentilla, “development of abaca fiber-reinforced foamed fly ash geopolymer,” matec web of conferences, vol. 156, 05018, march 2018. [34] m. szechyńska-hebda, j. marczyk, c. ziejewska, n. hordyńska, j. mikuła, and m. hebda, “optimal design of ph-neutral geopolymer foams for their use in ecological plant cultivation systems,” materials, vol. 12, no. 18, 2999, september 2019. [35] m. szechyńska-hebda, j. marczyk, c. ziejewska, n. hordyńska, j. mikuła, and m hebda, “neutral geopolymer foams reinforced with cellulose studied with the ftraman spectroscopy,” iop conference series: materials science and engineering, vol. 706, no. 1, 012017, november 2019. [36] g. silva, j. salirrosas, g. ruiz, s. kim, j. nakamatsu, and r. aguilar, “evaluation of fire, high-temperature and water erosion resistance of fiber-reinforced lightweight pozzolana-based geopolymer mortars,” iop conference series: materials science and engineering, vol. 706, no. 1, 012016, november 2019. [37] v. s. le, p. hájková, v. kovacic, t. bakalova, v. lukáš, c. h. le, et al., “thermal conductivity of reinforced geopolymer foams,” ceramics-silikáty, vol. 63, no. 4, pp. 365-373, 2019. [38] m. frydrych, š. hýsek, l. fridrichová, v. s. le, m. herclík, m. pechociaková, et al., “impact of flax and basalt fibre reinforcement on selected properties of geopolymer composites,” sustainability, vol. 12, no. 1, 118, 2020. [39] r. v. upadhyaya and t. g. suntharavadivel, “optimization of flyash and metakaolin content in mineral based cfrp retrofit for improved sustainability,” international journal of engineering and technology innovation, vol. 9, no. 3, pp. 171-181, july 2019. [40] n. ranjbar and m. zhang, “fiber-reinforced geopolymer composites: a review,” cement and concrete composites, vol. 107, 103498, march 2020. [41] g. bumanis, l. vitola, i. pundiene, m. sinka, and d. bajare, “gypsum, geopolymers, and starch-alternative binders for bio-based building materials: a review and life-cycle assessment,” sustainability, vol. 12, no. 14, 5666, 2020. [42] k. korniejenko, n. p. halyag, and g. mucsi, “fly ash as a raw material for geopolymerisation chemical composition and physical properties,” iop conference series: materials science and engineering, vol. 706, no. 1, 012002, november 2019. [43] k. korniejenko, m. łach, j. marczyk, c. ziejewska, n. p. halyag, and g. mucsi, “fly ash as a raw material for geopolymerisation-mineralogical composition and morphology,” iop conference series: materials science and engineering, vol. 706, no. 1, 012006, november 2019. [44] m. łach, j. mikuła, w. t. lin, p. bazan, b. figiela, k. korniejenko, “development and characterization of thermal insulation geopolymer foams based on fly ash,” proceedings of engineering and technology innovation, vol. 16, pp. 23-29, august 2020. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation, vol. 6, no. 2, 2021, pp. 67-73 influence of cold sensation on plantar tactile sensation for young females tianyi wang 1,* , shima okada 1 , masaaki makikawa 1 , masayuki endo 2 , yuko ohno 3 1 department of robotics, faculty of science and engineering, ritsumeikan university, shiga, japan 2 department of children and women’s health, graduate school of medicine, osaka university, osaka, japan 3 department of mathematical health science, graduate school of medicine, osaka university, osaka, japan received 14 december 2020; received in revised form 31 january 2021; accepted 04 march 2021 doi: https://doi.org/10.46604/aiti.2021.6863 abstract cold sensation (cs) is a cold feeling on people’s hands or feet; this is a well-known health problem for young females. plantar tactile sensation plays an important role in postural control and is affected by skin temperature. however, there is no research focusing on the relation between cs and plantar tactile sensation. in this study, we address the question of whether the cs influences plantar tactile sensation. 32 non cold sensation (non-cs) and 31 cold sensation (cs) young females have participated in this research. a tactile sensation test was conducted at five plantar points (first and fifth toes, first and fifth metatarsal heads, and heel). experimental results showed that although there was no significant difference at the first and fifth toes as well as the first metatarsal head and heel, the sensation threshold at the fifth metatarsal head for cs was lower than the non-cs (21.61 ± 8.10 μm, 27.42 ± 11.02 μm respectively, p < 0.05). it was concluded that plantar tactile sensation for young females with cold sensation was more sensitive compared to healthy subjects. keywords: cold sensation, plantar sensation 1. introduction cold sensation (cs) is frequently observed in females [1]. previous studies reported that about 50% of women over 50 years suffered from cs [2]. cold sensation is a cold feeling at extremities, under the environment where healthy people dose not feel cold [3]. although cs is not perceived as a remarkable symptom [4], it has been proved that cs is not only related to higher frequency of chronic disease, but also associated to health problems during daily life [5]. cold sensation patients frequently complain of cold on their feet. the feet are direct and often the only interface between the human body and the ground [6]. along with the visual and vestibular systems, feedback provided from the somatosensory system at feet plays an important role in postural control [7]. the foot is able to recognize mechanical stimuli through cutaneous mechanoreceptors located in the epidermis and dermis of glabrous skin [8]. there are four types of cutaneous mechanoreceptors: merkel's disk, ruffini's corpuscles, meissner corpuscles, and pacinian corpuscles. except for meissner corpuscles, where the skin temperature will not affect its function, it was reported that cooling the skin reduced the cutaneous sensation and firing response of mechanoreceptors [9]. previous research has revealed the relation between the foot sole temperature and plantar cutaneous sensation only through a passive cooling method such as cold pack [9]. it is still unclear whether cs patients have lower tactile sensation. we consider that knowledge about the relation between cold sensation and plantar tactile sensation will provide useful information for the necessary and appropriate health care for cs patients. * corresponding author. e-mail address: t-wang@fc.ritsumei.ac.jp advances in technology innovation, vol. 6, no. 2, 2021, pp. 67-73 68 as a consequence, the purpose of this paper is to investigate the relation between cold sensation and plantar tactile sensation. this paper consists of five parts. section 2 describes the human subjects and the method. results are shown at section 3. section 4 discusses the experimental results and section 5 concludes this paper. 2. subjects and method 2.1. human subjects we asked sixty-three female students (age: 21.5 ± 1.7 years, height: 157.92 ± 6.07 cm, weight: 51.45 ± 6.16 kg, body mass index: 20.60 ± 2.03 kg/m 2 ) with regular menstrual cycles to participate in the experiment. no subject reported any history of endocrinopathy, cardiovascular disease, gynecological conditions, or connective tissue disease. subjects were required to abstain from alcohol and caffeine for at least one day and from any food at least 2 h before the experiment. before the experiment, all subjects were informed of the purpose and process of this study and provided written informed consent. subjects were free to withdraw from the study at any time. this study was approved by the ethics committee on osaka university hospital (no. 19162, august 2019). the subjects with cold sensation was defined according to our previous study [10]. 2.2. experiment environment the experiment was performed from 2019 july 31 to 2019 october 21, 10:00 to 16:00. the experiment was only conducted during follicular phase; taking consideration about the body temperature fluctuates depending on the phase of the menstrual cycle. local temperature during the experiment was from 30.1 ± 12.3 °c to 22.8 ± 4.7 °c, average 26.9 ± 4.4 °c [11]. the indoor temperature was 24.6 ± 0.6 °c, humidity was 54.5 ± 12.3%. 2.3. plantar cutaneous sensation test tsuneya et al. [12] reported that insufficient toe contact, also known as floating toe, was observed in 76.2% of females and particularly on the fifth toe. as insufficient toe contact could reduce experiment accuracy, before the plantar cutaneous sensation test, we first perform a foot pressure distribution test to check whether subjects' plantar could fully contact the ground. fig. 1 describes the method and example of the foot pressure distribution test. fig. 1(a) shows the overview of foot pressure measurement. i-scan150© nitta pressure sensor sheet (165 × 165 × 0.1 mm) was used to measure the subjects' dominant feet. this sensor sheet contains 1,936 sensor points, and the high resolution could satisfy the need. fig. 1(b) described two types of foot pressure distributions. in this study, if the subject's toe was well contacted, it was defined normal; if any toe was not well contacted, it was defined floating toe and we did not test the plantar tactile sensation at the toe. (a) method for measuring foot pressure distribution (b) left: normal foot pressure distribution; right: fifth toe floating fig. 1 foot pressure distribution test fig. 2 shows an overview of the plantar tactile sensation test. fig. 2(a) shows the system overview of plantar tactile sensation testy system. in this study, © asuka electri, a plantar tactile stimulation platform (400 × 400 × 75 mm) was advances in technology innovation, vol. 6, no. 2, 2021, pp. 67-73 69 utilized. this system mainly comprises three parts: a sensation stimulation platform, a response button, and a computer for motor control and data processing. fig. 2(b) showed the mechanism of plantar tactile stimulation. the contact part is driven using a miniature linear guide with a motor-ball screw system, which can provide variable ranges from 5 to 2000 μm and velocity of shearing movement from 0.1 to 250 mm/s [13]. (a) system overview of plantar tactile sensation test system (b) mechanism of plantar tactile stimulation fig. 2 plantar tactile sensation test system plantar sensation test points are shown in fig. 3(a). in this research, five plantar points were selected as test points. to eliminate serial effects, the stimulation order was set as 1: first toe, 2: heel, 3: fifth toe, 4: first metatarsal head, 5: fifth metatarsal head [14]. white circles denote the test points, and the dotted line with a single arrow denotes the stimuli order. fig. 3(b) describes how to define the sensation threshold. first row denotes the stimulation level; second row shows the subject’s feeling at the feet; third row shows the response to stimulation. twenty stimulation levels were set for the experiment. the subjects were told to press the response button as fast as they can after they felt the movement of the contact part. if they felt the movements and continuously pressed the response button three times, stimulation at the first time when the response button was pressed was defined as the threshold at that plantar point. all tests were conducted on the subjects' dominant feet. (a) foot map with test points and stimuli order (b) method for defining sensation threshold fig. 3 test points and sensation threshold table 1 threshold, stimulation time, and frequency for each stimulation level. level 1 2 3 4 5 6 7 8 9 10 threshold (μm) 5 10 15 20 25 30 35 40 45 50 time (msec) 2 4 6 8 10 12 14 16 18 20 frequency (hz) 500 250 167 125 100 83 71 63 56 50 level 11 12 13 14 15 16 17 18 19 20 threshold (μm) 55 60 65 70 75 80 85 90 95 100 time (msec) 22 24 26 28 30 32 34 36 38 40 frequency (hz) 45 40 38 36 33 31 29 28 26 25 table 1 shows details of the twenty stimulation levels. stimulation range (threshold) was set as 5 μm, 10 μm, 15 μm, …, 90 μm, 95 μm, 100 μm, with the same velocity of 5 mm/s. thus, stimulation time and frequency can be calculated as following: 2 /time threshold velocity  (1) 1/frequency time (2) advances in technology innovation, vol. 6, no. 2, 2021, pp. 67-73 70 2.4. data analysis plantar tactile threshold data were obtained from a special software (“stimulation threshold” ©asuka electri). the student's t-test was used to test the significance of difference between two groups, with the significance level set at 5%. the t-test was performed using matlab r2020a © mathworks. 3. results the subjects’ basic information is summarized in table 2. bar charts in fig. 4 show the results of plantar tactile sensation. left part is the result of sensation threshold, and the right part is the result of response frequency. the horizontal axis shows the amplitude of threshold and frequency, and the vertical axis represents different test points. gray bars represent the non-cs group, and white bars indicate the results of the cs group. notice that the number of test points at the fifth toe (point 3) was different from others. after checking the foot pressure distribution, the floating toe at fifth toe was observed in 9 subjects in the non-cs group and 14 subjects in the cs group. thus, only 23 non-cs subjects and 17 cs subjects had the plantar sensation of fifth toe (point 3). although there was no significant difference of stimulation threshold at first toe, heel, fifth toe or first metatarsal head (p > 0.05), the threshold at the fifth metatarsal head (point 5) for the cs group was significantly lower than the non-cs group (21.61 ± 8.10 μm, 27.42 ± 11.02 μm, respectively, p < 0.05). as regards to stimulation frequency, the metatarsal parts (points 4 and 5) for the cs group showed higher amplitudes than the non-cs group (p < 0.05), 112.94 ± 45.64 hz and 90.87 ± 32.61 hz at point 4, 127.44 ± 35.02 hz and 106.80 ± 44.80 hz at point 5. table 2 age, physical characteristics of subjects. non-cs (n = 32) cs (n = 31) p value age (years) 21.09 ± 1.80 21.97 ± 1.52 < 0.05 height (cm) 158.70 ± 6.28 157.11 ± 5.84 0.30 weight (kg) 52.39 ± 7.04 50.47 ± 5.03 0.22 bmi (kg/m 2 ) 20.75 ± 2.01 20.48 ± 2.08 0.60 value: mean ± standard deviation bmi: body mass index fig. 4 result of threshold and frequency to plantar tactile sensation fig. 5 shows the result of the threshold–frequency plot of the plantar sensation and response ranges for three mechanoreceptors. the blue square represents the meissner corpuscle, the orange square represents the pacinian corpuscle, and the green square represents the ruffini ending. markers in red and blue represent the data of the non-cs and cs groups, respectively. the dashed line is the characteristic curve of threshold–frequency and plotted using the data shown in table 1. standard deviations for the threshold and frequency were plotted in black solid lines. the frequency range of the meissner corpuscle was 3–100 hz, ruffini ending was 15–400 hz, and pacinian corpuscle was 10–500 hz [15]. advances in technology innovation, vol. 6, no. 2, 2021, pp. 67-73 71 fig. 5 threshold–frequency of plantar tactile sensation for the non-cs and cs groups 4. discussion according to the results of the tactile sensation test, the threshold at the fifth toe (point 3) was higher than the other fo ur points both for cs and non-cs groups. ino et al. [14] tested the plantar sensation at the same points in this research and observed that the fifth toe had the highest plantar detection threshold, followed by the first metatarsal head. surprisingly, we observed that young females with cs had a lower tactile threshold at the foot sole, opposite to the findings in the previous study. in [2], we reported that plantar sensation thresholds were higher for the cs group (n = 18) than the non-cs group (n = 11) and that young females with cs were less sensitive at the foot than the healthy subjects. however, the definition of cs in the previous study was completely different in [2]. we used a cooling recovery test to check whether subjects’ plantar temperature could return to a particular level after cooing the feet with 15 °c cold water. although cold water stimulation has been proved as an effective approach to defining cs, we did not crosscheck the result with either the questionnaire or the thermal check methods. thus, we considered that subjects diagnosed with cs in the previous study were different from the cs group in this research. as previously mentioned, there are four types of mechanoreceptors distributed on the plantar glabrous skin. they are physiologically classified according to their adaption characteristics and the size of the receptive fields [16]: merkel's disk, which are slow adapting receptors with small receptive fields (sa ⅰ), generally provide information on indentation and press; ruffini's corpuscles, which are slow adapting receptors with large receptive fields (sa ⅱ), can detect skin stretching; meissner corpuscles, which are fast-adapting receptors with small receptive fields (fa ⅰ), is more responsive to tactile events such as the motion or slippage of an object across the skin; pacinian corpuscles, which are faster adapting receptors with large receptive fields (fa ⅱ), is well known as a mechanoreceptors responding to vibration. according to the result shown in fig. 5, all markers were located at the threshold--frequency response areas of pacinian corpuscles and ruffini endings (orange and green squares). only the markers representing the fifth toes (rhombus in blue and red) and the first metatarsal head of the non-cs (red circle) group were located at the areas of all three mechanoreceptors. according to the movement of the contact part, mechanical stimulation herein involved a multi-sensory approach, including skin motion, stretch, and vibration. thus, it is reasonable to consider that fa ⅰ, fa ⅱ, and sa ⅱ mechanoreceptors participated in the role of sensing the tiny movement of the contact part. in a 2002 study, shimojo et al. outlined that meissner corpuscles receive no effect from skin temperature [17]. despite the receptive fields of pacinian corpuscles being the largest among these four mechanoreceptors, fa ⅱ only takes up 12% of all foot sole cutaneous mechanoreceptors (43 in 364) [18], and the distribution of fa ⅱ along the foot sole is lower than those for advances in technology innovation, vol. 6, no. 2, 2021, pp. 67-73 72 other mechanoreceptors at the same sole area [6]. above all, fast-adapting mechanoreceptors have little influence on plantar tactile sensation for the cs group herein. strzalkowski et al. [19] reported that there were 74 sa ⅱ cutaneous mechanoreceptors among all 364 feet sole cutaneous mechanoreceptors, which was higher than sa ⅰ (63) and fa ⅱ (43). moreover, according to viseux’s work [6], the percentage distribution of sa ⅱ at the metatarsal head was higher than that at the toes and the heel. compared with skin motion and vibration, the metatarsal head seems to be more sensitive to skin stretching. in contrast, it was noticed that lowering the skin temperature enhanced the impulse firing of ruffini endings [18]. thus, it was suspected that the low foot skin temperature for the cs group increased the impulse firing of sa ⅱ mechanoreceptors, thereby improving foot tactile sensation. the mechanical properties of the skin may also partially explain some difference in sensation threshold between the cs and the non-cs groups. cs patients had higher skin hardness at the foot sole [20]; compared with the toes and the heel, the highest skin hardness was observed at the fifth metatarsal head [21]. meanwhile, it was hypothesized that when skin hardness increases, it was easier to have stress concentration and increased sensation at the finger [22].thus, another possible reason to explain the finding in this study is: for young females with cs, lower plantar sensation threshold, namely more sensitivity at the foot, is possibly due to the higher skin hardness at the fifth metatarsal head. the ability to detect tiny movements such as slippage varies if the stimuli direction changes [17]. specifically, the threshold of detecting slippage increases from the vertical to the horizontal direction. in this research, all subjects were asked to put their feet on the test platform along the same direction of contact part movement. thus, discussing the effect of stimulation direction on the sensation threshold was difficult. we will investigate this aspect in a future study. another current technical problem is that controlling the surface temperature of the sensation test platform is difficult. a difference in temperature between the subject's foot sole and contact area may influence the test result. although no subject reported a cold feeling when they put her foot on the test platform during the experiment, the method of avoiding cooling the foot from the experiment equipment should be focused on in a future study. 5. conclusions in this paper, plantar tactile sensation for young females with cold sensation was studied. the experiment results demonstrated that the sensation threshold at the fifth metatarsal head for the cold sensation group was lower than the non-cold sensation group. enhanced impulse firing due to the lower skin temperature and higher skin hardness were considered the reasons for young females with cold sensation being more sensitive. conflicts of interest the authors declare no conflict of interest. references [1] y. m. hur, j. h. chae, k. w. chung, j. j. kim, h. u. jeong, j. w. kim, et al., “feeling of cold hands and feet is a highly heritable phenotype,” twin research and human genetics, vol. 15, no. 2, pp. 166-169, april 2012. [2] t. wang, m. endo, h. uehara, s. suga, m. matsuzaki, g. nakamoto, et al., “plantar tactile sensation for young females with hiesho,” ieee 2 nd global conference on life sciences and technologies, ieee press, march 2020, pp. 329-330. [3] h. mori, h. kuge, s. sakaguchi, t. h. tanaka, and j. miyazaki, “determination of symptoms associated with hiesho among young females using hie rating surveys,” journal of integrative medicine, vol. 16, no. 1, pp. 34-38, january 2018. [4] s. tsuboi, t. mine, y. tomioka, s. shiraishi, f. fukushima, and t. ikaga, “are cold extremities an issue in women’s health? epidemiological evaluation of cold extremities among japanese women,” international journal of women’s health, vol. 11, pp. 31-39, january 2019. advances in technology innovation, vol. 6, no. 2, 2021, pp. 67-73 73 [5] k. h. bae, h. y. go, k. h. park, i. ahn, y. yoon, and s. lee, “the association between cold hypersensitivity in the hands and feet and chronic disease: results of a multicentre study,” bmc complementary alternative medicine, vol. 18, no. 1, pp. 1-8, january 2018. [6] f. j. f. viseux, “the sensory role of the sole of the foot: review and update on clinical perspectives,” neurophysiolgie clinique, vol. 50, no. 1, pp. 55-68, february 2020. [7] a. kavounoudias, r. roll, and j. p. roll, “specific whole-body shifts induced by frequency-modulated vibrations of human plantar soles,” neuroscience letters, vol. 266, no. 3, pp. 181-184, may 1999. [8] s. d. perry, w. e. mcilroy, and b. e. maki, “the role of plantar cutaneous mechanoreceptors in the control of compensatory stepping reactions evoked by unpredictable, multi-directional perturbation,” brain research, vol. 877, no. 2, pp. 401-406, september 2000. [9] c. r. lowrey, n. d. j. strzalkowski, and l. r. bent, “cooling reduces the cutaneous afferent firing response to vibratory stimuli in glabrous skin of the human foot sole,” journal of neurophysiology, vol. 109, no. 3, pp. 839-850, february 2013. [10] t. y. wang, s. okada, m. endo, m. makikawa, and y. ohno, “determination of hiesho among young japanese females using thermographic technique,” advanced biomedical engineering, vol. 10, pp. 11-17, 2021. [11] japan weather association, “osaka weather,” https://tenki.jp/past/2019/07/weather/6/30/47772/, june 05, 2020. [12] m. tsuneya and n. usui, “actual state of toe contact in the upright position among healthy adults,” phys ther japan, vol. 33, no. 1, pp. 30-37, 2006. (in japanese) [13] m. sato, n. takahashi, s. yoshimura, k. yamashita, c. wada, s. ino, et al., “a new device for foot sensory examination employing auto-presentation of shear force stimuli against the skin,” journal of biomechanical science and engineering, vol. 10, no. 2, pp. 1-11, april 2015. [14] s. ino, m. chikai, n. takahashi, t. ohnishi, k. doi, and k. nunowaka, “a pilot study of a plantar sensory evaluation system for early screening of diabetic neuropathy in a weight-bearing position,” 36 th annual international conference of the ieee engineering in medicine and biology society, ieee press, august 2014, pp. 3508-3511. [15] s. choi and k. j. kuchenbecker, “vibrotactile display: perception, technology, and applications,” proceedings of the ieee, vol. 101, no. 9, pp. 2093-2104, september 2013. [16] g. schlee, “quantitative assessment of foot sensitivity: the effects of foot sole skin temperature, blood flow at the foot area and footwear,” ph.d. dissertation, faculty of behavioral and social sciences, chemnitz university of technology, chemnitz, sn, 2010. [17] m. shimojo, “information processing of skin sensation,” journal of the society of instrument and control engineers, vol. 41, no. 10, pp. 723-727, 2002. (in japanese) [18] tokyo medical and dental university, “somatosensory: mechanoceptive sensation,” http://www.tmd.ac.jp/med/phy1/ptext/somat_1.html, june 17, 2020. (in japanese) [19] n. d. j. strzalkowski, r. m. peters, j. t. inglis, and l. r. bent, “cutaneous afferent innervation of the human foot sole: what can we learn from single-unit recordings?” journal of neurophysiology, vol. 120, no. 3, pp. 1233-1246, september 2018. [20] a. ushiroyama, “clinical analysis and treatment of hiesho,” journal of clinical and experimental medicine, vol. 215, no. 11, pp. 925-929, 2005. (in japanese) [21] y. jammes, m. viala, w. dutto, j. p. weber, and r. guieu, “skin hardness and epidermal thickness affect the vibration sensitivity of the foot sole,” clinical research on foot & ankle, vol. 5, no. 3, pp. 1-5, august 2017. [22] h. jeong, m. kaneko, m. higashimori, and k. matsukawa, “improvement of tactile sensitivity under pressing a finger base,” transactions of the society of instrument and control engineers, vol. 43, no. 11, pp. 973-979, november 2007. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). https://tenki.jp/past/2019/07/weather/6/30/47772/ http://www.tmd.ac.jp/med/phy1/ptext/somat_1.html  advances in technology innovation, vol. 5, no. 4, 2020, pp. 248-258 real time scanning-modeling system for architecture design and construction ye zhang * , kun zhang, kaidi chen, zhen xu department of architecture, tianjin university, tianjin, china received 12 march 2020; received in revised form 05 june 2020; accepted 02 july 2020 doi: https://doi.org/10.46604/aiti.2020.5385 abstract the disconnection between architectural form and materiality has become an important issue in recent years. architectural form is mainly decided by the designer, while material data is often treated as an afterthought which doesn’t factor in decision-making directly. this study proposes a new, real-time scanning-modeling system for computational design and autonomous robotic construction. by using cameras to scan the raw materials, this system would get related data and build 3d models in real time. these data would be used by a computer to calculate rational outcomes and help a robot make decisions about its construction paths and methods. the result of an application pavilion shows that data of raw materials, architectural design, and robotic construction can be integrated into a digital chain. the method and gain of the material-oriented design approach are discussed and future research on using different source materials is laid out. keywords: scanning-modeling system, material data, computational design, robotic construction 1. introduction the disconnection between form and material in the architecture field has become a serious problem. with the popularity of 3d modeling software such as sketchup, rhino, and grasshopper, architects are prone to first develop form on a computer screen and only afterwards associate material designations to the predefined geometry model [1]. in improper citing, the digital form is freely rendered with any material bitmap or texture without regard to a material’s rational use or practicality [2], and risking that material information cannot be timely and effectively fed back to the design and construction stages. the separation of material and form or the linear progression from the design intention to materialization has caused some problems in recent decades. firstly, the process of fabricating source material into a variety of sizes and shapes consumes significant time, while a large quantity of material waste is produced. secondly, human-designed rules and solutions are prone to be limited by empirical knowledge [3], especially in projects with irregular surfaces and complex geometries. ignoring source material’s properties makes the design and fabrication process more passive, inefficient, imprecise, and expensive. finally, it hinders expressing building’s regional and cultural characteristics as architectural materials with different personalities in different regions are mass produced through the same processing method. advances in contemporary digital scanning and bespoke robotic technologies [4] enable material information to be simulated, calculated, and well organized. as a result, it is an alternative approach for architecture. in this study, a new, real-time scanning-modeling system for obtaining material information and incorporating the data into a continuous digital chain of computational design and robotic construction are introduced. as fig. 1 shows, traditional designs are determined subjectively by the architects; while a new chain with the scanning-modeling system enables an alternative design approach derived objectively by the data of source material. * corresponding author. e-mail address: yaapp2012@gmail.com tel.: +86-15900241936 advances in technology innovation, vol. 5, no. 4, 2020, pp. 248-258 249 fig. 1 comparison of two modes the scanning-modeling system is regarded as the sensing end of this digital chain. it reads data of raw materials including their sizes and shapes without cumbersome supporting equipment and builds a visual model in real-time. after collecting and visualizing the data, the calculation end of the chain would generate architectural geometry based on material data, site conditions and structural performance. finally, at the action end of the chain, a robot arm would be used to fabricate the design. according to the data obtained during the previous stages, the robot would find an assembly location autonomously without having to be programmed in advance. in sum, this study creates a convenient and rapid material information acquisition system that benefits architectural design and digital construction. 2. literature review and innovation points this study provides a new scanning-modeling system and examines its application in architectural programs. the research contents include autonomous perception, computational generation, and robotic actuation. despite a great deal of research on these three topics over the past several decades, there has been very little research intended to integrate them into the field of architecture. previous studies and new development in measuring and modeling, computational design, and robotic construction are described as follows. 2.1. measuring and modeling table 1 measuring and modeling methods name principle speed accuracy cost stability limitation active method moiré fringes diffraction and interference of light ++ +++ + + only available for objects with regular texture time-of-flight (tof) time taken for the wave to bounce back to the emitter ++ -- low resolution and easy to be disturbed structured lighting encoding and decoding of light + ++ + ++ projection equipment required triangulation solve trigonometric equation + + + data sparseness passive method shape from texture (sft) perspective distortion calculation + + + surface texture information required shape from shading (sfs) ray calculation + light parameters required multi-view stereo (mvs） parallax principle --+ + image error is big silhouette visible shell approximation +++ -++ ++ visible shell only existing 3d reconstruction methods can be divided into two categories: active and passive. the active methods mainly include moire fringe, time-of-flight (tof), and structured lighting and triangulation; while the passive methods mainly include shape from texture (sft), shape from shading (sfs), multi-view stereo (mvs), and silhouette. these methods use mathematical and physical knowledge to rebuild objects. among them, moire fringe used the interference of wave to digital construction construction subjective decision objective decision advances in technology innovation, vol. 5, no. 4, 2020, pp. 248-258 250 deduce the shape; sft, sfs and structured lighting mainly use perspective principle; triangulation, mvs and silhouette use the methods of deformations of triangle positioning. the characteristics of each method are presented in table 1 [5-9]. “+” means the performance is strong while “-” means it is weak. in the field of architecture, the precision of measuring and modeling required is not as strict as other fields. the demands of speed, cost, and stability are higher. therefore, this study is established on the basis of the silhouette method, which develops to improve the speed and efficiency of getting information on material size, and to make it easier to operate as the computing moreover, the information communication mechanism are in a visual interface. 2.2. computational design different from design precedents which rely on intuition and experience to solve design problems, computational design encodes design decisions using a computer language; consequently, the design expression is changed from geometry to logic [10]. the computational design is excellent in dealing with complexity, coming up with many options in a short time, and optimizing the solution. although the computational design is still a relatively young and evolving field [11], some research institutions and researchers have made progress in theoretical development, methodologies, and strategies [12-14]. based on the methodologies and strategies of computational design, this study presents a pavilion which is generated from material information. the major dimensions of computational design in this study include: (1) information transformation: this pavilion makes use of natural branches containing three forks. the length of each fork and the angles between the forks are read by a scanning system and the information is transformed into grasshopper in real time by a plug-in developed by this research. (2) parameter setting: considering the site conditions and the pavilion’s functional uses, the parameters are the location of the anchor points, the camber of the surface, the fractal geometry, and the number of forks at a node. (3) 3d modeling and visualization: the computation of the design and fabrication is conducted in rhino3d which allows designers to test and simulate in a visual environment. (4) processing with data and algorithms: the rules and standards to inform the design direction comprising waste less material, similarity to the target surface, and structural performance. 2.3. robotic construction the use of robots in architectural construction can be traced back to the 1980s [15]. at that time, robotic systems were mainly used in japan, but the efficiency didn’t prove satisfactory and hence received fading research attention. however, with the development of digital tools and manufacturing technologies, robotics in architecture has been a burgeoning research topic since 2010. prior studies have sought to explore and expand the possibilities of robotic fabrication, robotic assembly, and even human-robot interactions on site. however, an important limitation of past research is the material sensing and real-time feedback. this study seeks to fill that gap by emphasizing robotic eye-hand coordination and to break with the typical robot movement pattern which is controlled by humans and not by the robot itself. 3. real time scanning-modelling system 3.1. motivation and framework the primary purpose of the scanning-modeling system is measuring architectural materials on-site automatically and gets their three-dimensional models in real time. the procedure is shown as fig. 2. for each material component, the images from advances in technology innovation, vol. 5, no. 4, 2020, pp. 248-258 251 different perspectives will be scanned by a camera group. an independent program specially written by this study will capture the effective information of the material image and calculate feature points using an open source computer vision library. in order to ensure the efficiency of operational and data transmission, the program is simplified to transmit data of the quantity of controllable feature points with high information concentration. then, a pre-programmed plug-in in rhino/grasshopper will receive data from the scanning program by memory mapping and analyzing outlines from different perspectives. finally, a three-dimensional model of the corresponding material component will be generated within 1 second for subsequent design development. fig. 2 scanning-modelling system 3.2. method this research takes natural branches as an example material to explain how the scanning-modeling system works. the system extracts the contours of the physical branch and reconstructs it digitally. it has two working paths: a software workflow and a hardware workflow. the software workflow is virtual and programmable, while hardware workflow is realistic and transitive. the data circulate and transmit in the software workflow, participating in complex interactions and transformation operations, before finally outputting a concise digital model of the branch. the hardware workflow is very simple, consisting only of material components, sensors (cameras) and a computer. six cameras can complete all of the image acquisition work while the computer can realize image recognition, data processing and 3d model reconstruction. 3.3. hardware workflow the branches (source material), camera group, and computer constitute the hardware workflow of the system. in the laboratory, a 6m-long custom frame to fix the cameras is built, and the background with black curtains to extract information effectively is covered. the conditions and settings for each hardware element are as follows: (1) branches to be tested: this study used natural branches which had minor branches, leaves and twigs pruned away. their lengths range from 75 cm to 245 cm and their radius range from 2cm to 10cm. (2) number and location of cameras: the branch’s radius is set as r, the length of the frame holding the cameras as l and the camera angle as α. an equation can be calculated as: l=2r/sin(α/2) (1) the number of pixels in the corresponding direction of the camera is set as n, the resolution value of the reconstructed model is γ. the calculation is based on the 1st formula: γ=2rcos(α/2)/n=lsinα/2n (2) advances in technology innovation, vol. 5, no. 4, 2020, pp. 248-258 252 the inherent error of this system is not uniform. it is affected by two factors as shown in fig. 3: one is the angle between the object and the camera, and the other is the normal vector of the surface point with respect to the plane of the camera group. to establish this error, this study sets the angle between two adjacent cameras as δ and the inherent error as e. an equation can be computed as: e=r[csc(δ/2)-1] (3) fig. 3 calculations in order for γ> e and considering feasibility, convenience, and accuracy, this research finally takes l = 600cm, α = 0.3π, δ = 0.667π, which means 6 cameras are fixed to a 600 cm-long frame to catch images of the branch. 3.4. software workflow the software workflow is the core of this system. it includes three steps, as shown in fig. 4: (1) generating a description file of calibration cameras. (2) an external program which extracts the image of the branch and calculates related data for visual presentation. (3) a pre-programmed module in rhino receives the data and converts it into a digital branch model in real time. the following sections explain the three steps in more detail. fig. 4 software workflow the first step is to develop a camera calibration file “cameracorrect.py”. the purpose of this step is to determine the mapping relationship between contour pixels and spatial points in the modeling software. the calibration is based on the perspective principle, it does not require a complete camera internal parameter matrix, only a scaling parameter and rotation parameter are taken. because of the usb broadband limitation, an additional module was added in the python script in order to read information from multiple cameras simultaneously. there are five commands “z”, “x”, “f”, “s” and “q” in the script: “z” and “x” to change the camera, “f” to refresh, “s” to display the recorded data, “q” to save the data and exit. after the relevant parameters of the calibration tool are entered, an auxiliary script automatically calculates the final calibration parameters and generates this calibration file. this step only runs once in the whole workflow and is not repeated when a branch is scanned. advances in technology innovation, vol. 5, no. 4, 2020, pp. 248-258 253 the second step is to analyze the branch images and generate both a “points.py” file which contains image information and a “getgeo.gh” file which is used in the modeling software. first, through the gauss filter process and the binary image process, it is able to extract smooth contours of a branch from different perspectives. then, the number of pixels selected in the contour is adjusted by a tolerance parameter and the three-dimensional coordinates of all the selected points are calculated. finally, the three-dimensional array is transferred into grasshopper through shared memory and the spatial position of the points is inversely mapped in rhino according to the former calibration file. the third step in the software workflow is to generate a visual model which runs simultaneously with the second step. a preset program receives array data and reorganizes them into projections from different views in order. by the algorithm of the boolean intersection, all the information work together to generate a digital branch model representative of the scanned physical branch. 3.5. results and discussions the system is characterized by flexibility, reliability, and processing speed. because of the modular organization of the workflow, the perception of material information is expandable. the researchers can increase the number of cameras to improve the accuracy of the system; in addition, expand the type of sensors to capture and analyze other qualities of the source material. flexibility is also reflected in the diversity of operating environments and source materials. although the former branch experiment is carried out in a laboratory, it can be performed in other places including outdoor environments by simply changing the fixing points and corresponding values in the equations. similarly, the system can scan other building materials (source materials). additionally, the system has a high tolerance for faults, built-in redundancy; therefore, it is very reliable. even if some of the sensors are broken, other sensors can still work as usual, so the program can run normally despite the reduced accuracy. last but not the least, this system runs very fast. usually, image processing is extremely cpu intensive, but this system basically eliminates this problem with appropriate abstraction and sufficient optimization. taking the branch project as an example, even in ordinary notebook configuration, a refresh rate of 1 second per time can be achieved. this efficiency not only saves time, but also makes the system more practical to use. 4. application example using the above-mentioned real time scanning-modelling system, a pavilion is built and the application of the system in new architecture model is shown. this pavilion uses abandoned "y" shaped branches as source materials. by the algorithms, the computer can make independent decisions and generate a design based on the quantity and characteristics of the existing materials [16]. then, a robotic arm is used to complete the construction process. 4.1. design of the pavilion 4.1.1. selection and organization of the branches fig. 5 information abstraction advances in technology innovation, vol. 5, no. 4, 2020, pp. 248-258 254 with the scanning-modelling system, the researchers develop digital models of all wooden branches that can be used for the pavilion. next, the researchers distill the information which is useful to design and discard data that does not affect the design such as slight unevenness on the branch surface and the bending amplitude of each fork, etc. the useful parameters are three-dimensional coordinates of the four points p0, p1, p2 and p3 on the branch. the length between every two points and the length and thickness of each fork are shown in fig. 5. because all tree branches are from the same species of the ash tree, their density is considered the same. fig. 6 computational generation process according to the environmental conditions and functional requirements, the approximate dimensions of the pavilion are established. the location of the ground points and the geometry logic are seen as a preliminary step, in addition to stipulating that each vertex must have three forks jointing together, as shown in fig. 6. in the procedure of component selection and organization, the logic is set as follows: (1) all branches are sorted by volume and the three largest load-bearing branches are selected. (2) for any two branches b and c which are connected to the same branch a, if they can intersect, their trajectory of intersection is an ellipse and takes the position with the minimum distance from the central axis of branch a to ensure the overall stability. (3) if branch b and branch c cannot intersect, they are cut shorter to make the intersection possible. according to the information list of all the branches, the computer can quickly calculate the cutting lengths of different options and the branches spatial positions under all possibilities. corresponding pavilion forms are also simulated and all possibilities are sorted by how much material is wasted. (4) the top five options are chosen, then "millipedes" and other analysis tools are used to test and optimize their structural performance. combining the amount of material waste factor, structural properties, and formal aesthetics, the final pavilion design option is selected, as shown in fig. 7. fig. 7 structure performance analysis 4.1.2. detailed design after the design program is generated and optimized, its details are refined. the first is the node. every node in this pavilion is different and because all branches have their own angles, the nodes are customized rather than standardized. traditional rope-wood nodes can adjust the angle and are easy to make; however, they are not strong enough. cnc-customized advances in technology innovation, vol. 5, no. 4, 2020, pp. 248-258 255 tenon-and-mortise nodes are stable, but their fault tolerance ability is weak and the manufacturing takes detailed time. ultimately, a new type of node with universal joints, three-way pipes and jackscrews are designed. the new node is strong, cheap and easy to make, as shown in fig. 8. fig. 8 joints for the base, a cast-in-place concrete foundation and embedded bolt method are tried, but this was eliminated due to their insufficient strength and site environmental limitations. instead, the project uses metal bearing plate and ground pins to establish the base, as shown in fig. 9. fig. 9 base 4.2. robotic construction in the robotic construction stage, the motion of the robot arm and the design of the tool end should be carefully considered. by writing the relevant code in grasshopper and kukaprc, the robotic arm can automatically control its grabbing position and motion movement path according to the generated design. the robotic arm determines the final position of each branch quickly and accurately and then fixes the joint by cooperating with human workers or another robot. if there is any problem during the construction process, the data will be fed back into the digital model for real-time adjustment. in terms of the tool end of the robotic arm, it has to be strong enough to grasp the branch and ensure that it doesn’t twist. an electronically controlled claw is first applied. with the afmotor library in arduino, the opening and closing state, running speed and rotation direction of the gripper can be controlled in real time. however, claws have limitations: its opening and closing range is only 0-90mm and its maximum clamping force is only 16 newton. thus, it is only suitable for grabbing light and small branches. in order to improve the ability to grab, the pneumatic parallel open and close claw are selected and modified. the size between the parallel plates could be adjusted to match the diameter of the branch. the contact surface of the plates is coated with a special rubber material to increase the coefficient of friction; consequently, the branch wouldn’t twist. features of this pneumatic claw are: advances in technology innovation, vol. 5, no. 4, 2020, pp. 248-258 256 (1) wide opening and closing range. (2) greater grabbling ability. (3) suitable for branches with different sizes and weights. (4) it can be deployed in grasshopper, so the robot arm can perform the entire step independently (fig. 10). fig. 10 robotic construction 5. conclusion and future work this research succeeds in developing a real time scanning-modelling system to perceive and analyze material information for guiding the architectural design and robotic construction (figs. 11-12). the material oriented “eye-brain-hand collaboration” chain saves construction materials, greatly reduces the cost of component fabrication, gives unprecedented flexibility to geometry, optimizes structural performance and improves construction work. fig. 11 pavilion in this study, there is still some room for improvement. first of all, the environment perception aspect can be enriched. more types of sensors can be installed and the analysis of algorithms can be optimized; therefore, the system can perceive and utilize other kinds of information such as material strength and toughness. secondly, the computational design part can be further optimized and treated as agent-based simulation and generation. the researchers can develop algorithms that let the agent system provide better outcomes. user interfaces can also be developed. finally, the experimental pavilion relies exclusively on a kuka robotic arm. bespoke robotic arms can be modified and the interoperability of different robotic arms can be developed. advances in technology innovation, vol. 5, no. 4, 2020, pp. 248-258 257 fig. 12 pavilion in future research, the theory and method of this study are expected to be used for other materials in real scale architecture projects. the scripts and scanning-modeling methods can be partially replaced and used for materials with various shapes. for larger scale construction, the locations of sensors could be recalculated using the formula put forward in this paper. the researchers envision that success in digital manufacturing with globally abundant and locally available materials may impact the design approach by considering materials’ personalities. acknowledgement this study was funded by tianjin science foundation (agreement no.: 19jcqnjc07300). conflicts of interest the authors declare no conflict of interest. references [1] k. wu and a. kilian, “design natural wood log structures with stochastic assembly and deep learning,” robotic fabrication in architecture, art and design, springer press, august 2018, pp. 16-30. [2] a. picon, architecture and the virtual: towards a new materiality, wissenschaftliche zeitschrift der bauhaus-universität weimar, 2003. [3] n. leach, “digital cities,” architectural design, vol. 79, no. 4, pp. 6-13, june 2009. [4] f. gramazio, m. kohler, and j. willmann, the robotic touch: how robots change architecture: gramazio&kohler research eth zurich 2005-2013, 1st ed. zurich: park books, 2014. [5] f. chen, g. m. brown, and m. song, “overview of 3-d shape measurement using optical methods,” optical engineering, vol. 39, no. 1, pp. 10-22, january 2000. [6] r. zhu, o. furxhi, d. marks, and d. brady, “millimeter wave surface and reflectivity estimation based on sparse time of flight measurements,” 39th international conference on infrared, millimeter, and terahertz waves (irmmw-thz), ieee press, september 2014, pp. 1-2. [7] j. schlarp, e. csencsics, and g. schitter, “scanning laser triangulation sensor geometry maintaining imaging condition,” ifac-papers online, vol. 52, no. 15, pp. 301-306, december 2019. [8] m. slembrouck, p. veelaert, d. hamme, d. van cauwelaert, and w.philips, “cell-based approach for 3d reconstruction from incomplete silhouettes,” international conference on advanced concepts for intelligent vision systems, springer press, september 2017, pp. 530-541. [9] j. s. franco and e. boyeb, “efficient polyhedral modeling from silhouettes,” ieee transactions on pattern analysis and machine intelligence, april 2009, pp. 414-427. advances in technology innovation, vol. 5, no. 4, 2020, pp. 248-258 258 [10] h. hua, “a bi-directional procedural model for architectural design,” computer graphics forum, vol. 36, no. 8, pp. 219-231, december 2017. [11] g. brugnaro, e. baharlou, l. vasey, and a. menges, “robotic softness-an adaptive robotic fabrication process for woven structures,” proc. 36th annual conference of the association for computer aided design in architecture, october 2016, pp. 154-163. [12] p. eversmann, f. gramazio, and m. kohler, “robotic prefabrication of timber structures: towards automated large-scale spatial assembly,” construction robotics, vol. 1, no. 1-4, pp. 49-60, august 2017. [13] m. maasoumy and a. sangiovanni-vincentelli, “smart connected buildings design automation: foundations and trends,” foundation and trends in electronic design automation, vol. 10, no. 1-2, pp. 1-143, march 2016. [14] r. snooks and g. jahn, “closeness: on the relationship of multi-agent algorithms and robotic fabrication,” robotic fabrication in architecture, art and design, springer press, february 2016, pp. 218-229. [15] j. p. sousa, c. g. palop, e. moreira, a. m. pinto, j. lima, p. costa, et al. “the spiderobot: a cable-robot system for on-site construction in architecture,” robotic fabrication in architecture, art and design, springer press, february 2016, pp. 230-239. [16] r. rust, d. jenny, f. gramazio, and m. kohler, “spatial wire cutting: cooperative robotic cutting of non-ruled surface geometries for bespoke building components,” proc. 21st international conference on computer-aided architectural design research in asia, southeastern university press, march-april 2016, pp. 529-538. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). 2___aiti#7192_in press advances in technology innovation, vol. 6, no. 4, 2021, pp. 213-221 vehicle path planning with multicloud computation services po-tong wang 1 , shao-yu lin 2 , jia-shing sheu 2,* 1 department of electrical engineering, lunghwa university of science and technology, taoyuan, taiwan 2 department of computer science, national taipei university of education, taipei, taiwan received 21 february 2021; received in revised form 09 august 2021; accepted 10 august 2021 doi: https://doi.org/10.46604/aiti.2021.7192 abstract with the development of artificial intelligence, public cloud service platforms have begun to provide common pretrained object recognition models for public use. in this study, a dynamic vehicle path-planning system is developed, which uses several general pretrained cloud models to detect obstacles and calculate the navigation area. the euclidean distance and the inequality based on the detected marker box data are used for vehicle path planning. experimental results show that the proposed method can effectively identify the driving area and plan a safe route. the proposed method integrates the bounding box information provided by multiple cloud object detection services to detect navigable areas and plan routes. the time required for cloud-based obstacle identification is 2 s per frame, and the time required for feasible area detection and action planning is 0.001 s per frame. in the experiments, the robot that uses the proposed navigation method can plan routes successfully. keywords: computer vision, scene recognition, cloud computing, object detection 1. introduction although humans can walk small distances, walking long distances is exhausting and time-consuming. therefore, bicycles, motorcycles, and cars were invented over time, and their continued evolution has made movement convenient and safe. the focus of vehicle development has now shifted to the production of low-pollution electric vehicles, such as electric bicycles, balance bikes, and skateboards. autonomous driving can decrease the burden on human drivers, reduce road congestion, and improve transportation safety [1]. planning a safe pathway is the focus of autonomous driving. the information regarding the surrounding environment is collected to plan the next action. to prevent collisions, the environment and moving objects are monitored in real time for determining appropriate responses. however, the implementation of autonomous driving technologies is difficult. current autonomous vehicles use multiple ultrasonic or optical radars for detecting surrounding objects to create high-quality three-dimensional (3d) models at night and during the day; however, these sensors are expensive. furthermore, inclement weather severely affects the performance of the aforementioned sensors. machine vision technology is used in daily life applications, such as the smart face unlock feature in smartphones, instant text translation, and automatic checkout in stores. the capabilities of hardware equipment, such as high-image-quality cameras, vision processors, and 5g networks, are continually improving. commercially available visual models can accurately perform face, object, and text recognition. this study realizes the recognition of different objects in real-time images by using application programming interfaces (apis), i.e., google cloud vision, amazon rekognition, and azure computer vision. the obtained object information is regarded as the current environmental conditions when determining the walking area. this study uses available resources and machine vision to develop a low-cost and fast method for dynamic path discrimination. * corresponding author. e-mail address: jiashing@tea.ntue.edu.tw tel.: +886-9-38397255; fax: +886-2-27375457 advances in technology innovation, vol. 6, no. 4, 2021, pp. 213-221 the remainder of this study is organized as follows. section 2 describes the relevant literature. section 3 describes the system architecture. section 4 presents a description of the experiments and the experiment results, and section 5 provides the conclusions of this study. 2. literature review and methodology object detection is a technique for classifying objects in an image, and object detection technology has evolved considerably. lecun et al. [2] proposed the lenet image recognition network, which was subsequently modified by krizhevsky et al. [3] into alexnet. the addition of the rectified linear unit and dropout nonlinear activation functions to lenet considerably improved its image recognition rate. thus, machine vision has evolved rapidly. he et al. [4] proposed residual network architecture to solve the problem of overfitting. huang et al. [5] and wang et al. [6] made subtle changes and proposed densenet and cspnet, respectively. similar concepts of cspnet were used to design novel architectures that effectively enhance network identification capabilities [5-6]. howard et al. [7] proposed lightweight network architecture to increase the processor calculation speed and solve the problem of slow network operations in mobile devices and embedded learning. the aforementioned network architectures can be implemented to obtain a backbone network for the rapid learning of image features. many object recognition networks use such backbone networks for image feature extraction. various calculation methods are then incorporated into the network to learn the category and location of images [8]. ren et al. [9] developed a two-stage object recognition, i.e., faster region based convolutional neural networks (faster rcnn), with a high recognition rate by using visual geometry group (vgg) as the backbone network and the region proposal network. the aforementioned architecture was also used in an object recognition network [10]; however, in contrast to faster rcnn, the single-shot detector (ssd) object recognition network performs one-stage identification in real time. numerous sophisticated, fast, and automated object detection networks have been proposed. the initially proposed object detection networks such as “you only look once (yolo)” do not use artificial anchor frames [11]. law et al. [12] proposed cornernet, which is a novel anchorless frame network, for self-learning object detection. duan et al. [13] and tian et al. [14] subsequently improved cornernet and achieved the same results as law et al. [12] without relying on anchor frames. tan et al. [15] developed efficientdet and achieved up to 53.7 average precision (ap) with current one-stage object detectors. achieving simultaneous localization and mapping is critical for developing automated machinery. light detection and ranging technology can be used to sense the surrounding environment and dynamically avoid moving objects, such as crowds and vehicles on the road [16-17]. furthermore, rgbd-based visual sensors were used to establish real-time environmental images and motion paths for augmented reality (ar), virtual reality (vr), and unmanned aerial vehicle (uav) positioning [18]. a vision-based deep learning network can be used to detect the environment for conducting motion detection, walkable area detection, and motion planning [19]. obstacle detection can also performed by object detection, and the technique algorithm will be used to mark walkable areas and relative coordinates [20-21]. avoiding obstacles is a crucial ability for autonomous mobile robots, which must plan suitable movement routes. when moving from the current location to a target location, the most efficient movement route is the shortest path without any restrictions. dijkstra’s algorithm and the a* algorithm are common methods for determining the shortest path. dijkstra’s algorithm is similar to the breadth-first search method; it searches for the shortest distance node outward from the current coordinate point and continues the search until the target point is found. the a* algorithm combines the speed priority and dijkstra methods. the costs from the starting point to the node and from the node to the target point are added and used as the node’s search cost, and the path with the lowest cost is searched for outward from the starting point. using the method, the a* algorithm can plan the optimal path while avoiding obstacles. the obstacle avoidance problem is also called the velocity obstacle (vo) problem. if a robot collides with another robot when maintaining its current speed, the set of all collision events comprises the vo [22]. 214 advances in technology innovation, vol. 6, no. 4, 2021, pp. 213-221 3. system structure this study proposes a real-time path-planning system based on neural networks for unmanned vehicles. the image obtained by an embedded camera is used as the input, and multiple public pretrained neural networks are used to increase the amount of information available on obstacle positions. the drivable area is determined from the blank areas between multiple objects, and safe paths are planned in drivable areas. fig. 1 displays the architecture of the proposed system. three steps are used in this study to plan the path of a self-propelled vehicle: object detection (fig. 2), path planning (fig. 3), and movement control. fig. 1 architecture of the proposed system fig. 2 framework of cloud object detection fig. 3 framework of path planning 3.1. cloud object detection the purpose of object detection is to detect the locations of obstacles and avoid them. an object detector is composed of a backbone, neck, and head. the backbone is a network for obtaining image features, the neck fuses the feature maps from various layers, and the head is used for classification and localization. the model architecture of an object detector is displayed in fig. 4. 215 advances in technology innovation, vol. 6, no. 4, 2021, pp. 213-221 fig. 4 model architecture of an object detector the pretrained networks provided by google cloud platform (gcp), microsoft azure, and amazon web service (aws) are used to detect objects in real-time images and store the data from each platform in the following order: object class name, confidence score, bounding box top boundary value, bounding box bottom boundary value, bounding box left boundary value, and bounding box right boundary value. 3.2. path planning for a vehicle 3.2.1. deletion of the overlapped box first, the coordinates of the object bounding box are determined. next, whether two object bounding boxes overlap is determined. if an overlap is discovered, the bounding box with the higher bottom boundary value is removed. the left and right boundary values of objects a and b are set as (al, ar) and (bl, br), respectively. if the bounding boxes of a and b do not overlap, they must satisfy the following eqs. (1) and (2). l r l r b b a a< < < (1) l r l r a a b b< < < (2) 3.2.2. drivable area detection after removing the overlapping bounding box, the blank area between any two nonoverlapping objects is considered the movement area. the distance between all objects can be calculated from their bounding box coordinates. according to an axiom in the euclidean geometry system, the distance d between any two points p1(x1, y1) and p2(x2, y2) can be calculated using eq. (3). the shortest distance between the bottom endpoints of two bounding boxes, that is, the bottom line of the drivable area, is calculated as follows: 2 2 1 2 1 2 ( ) ( )d x x y y= − + − (3) 3.2.3. set destination the drivable area should be clear and not blocked by intermediate objects. a and b are set as nonadjacent objects, and a is to the left of b. the four endpoint coordinates of the bottom boundaries of the two objects are denoted as al(x1, y1), ar(x2, y2), bl(x1, y1), and br(x2, y2). when n objects exist between a and b, the bottom boundary endpoint coordinates of any object can be denoted as ol(x1, y1) or or(x2, y2). if the area is not blocked, eqs. (4) and (5) are satisfied. when the denominator in eqs. (4) and (5) is 0, the bounding boxes of the two objects overlap. this overlap is eliminated using eq. (1) or (2). 2 1 2 1 2 1 2 1 ax bx ax o x ay by ay o y − − < − − (4) 2 1 2 2 2 1 2 2 ax bx ax o x ay by ay o y − − < − − (5) 216 advances in technology innovation, vol. 6, no. 4, 2021, pp. 213-221 the center point of a robotic car is set as the midpoint of the bottom boundary of the image. the center point of each drivable area can be calculated from the endpoint coordinates of the bottom line of the drivable area. assuming that the two endpoints at the bottom of the drivable area are p1(x1, y1) and p2(x2, y2), the midpoint coordinates of the drivable area, namely pm(xm, ym), can be calculated using eq. (6). eq. (3) is used to calculate the distance between the midpoint of all drivable areas and the center point of the robotic car. the midpoint with the shortest distance to the center point is set as the waypoint, and the straight line from the center point to the waypoint represents the planned route. 1 2 1 2 ( , ) ( , ) 2 2 m m m x x y y p x y − − = (6) 3.3. movement control the coordinates of the center point of the robotic car and the waypoint can be used to calculate the offset angle between the travel direction of the robotic car and the waypoint. first, eq. (3) is used to calculate the distance between the car and waypoint. then, the cosine value of the angle between the planned path and the vertical line is calculated. the angle converted by the cosine value is the offset angle. if the offset angle does not exceed the preset threshold, a forward command is issued to allow the car to move forward. however, if the offset angle exceeds the preset threshold, a turn-left or turn-right command is issued for appropriately controlling the direction of car movement. 4. experimental results table 1 lists the experimental equipment used in this study. the raspberry pi is a single-chip computer developed by the raspberry pi foundation for improving students’ understanding of computing science. the experiment uses docker to construct an image file with the official raspberry pi operating system and the python 3 language for research. table 1 experimental equipment used in this study components specification operating system raspbian central processing unit arm cortex-a72 random access memory 4 gb (lpddr4) camera logitech c310 fig. 5 displays a self-propelled vehicle equipped with a raspberry pi 4 computer and complementary metal oxide semiconductor (cmos) lens assembly for capturing environmental images. the raspberry pi 4 has a 40-pin universal input and output that can be used for external screens and sensors. the raspberry pi 4 is connected to a power source motor for controlling the movement of a robotic car. a logitech c310 fixed-focus lens is used to obtain a 5-million-pixel image with a size of 1280 × 960. this lens is connected to the raspberry pi 4 through a universal serial bus (usb) interface. raspberry pi 4 has a rated power of 5 v/3 a. a mobile power bank is used as a power source through usb-c to provide a maximum output of 5 v/2.1 a to the raspberry pi 4. (a) front view of the vehicle (b) top view of the vehicle fig. 5 the robotic car equipped with raspberry pi 4 and cmos camera 217 advances in technology innovation, vol. 6, no. 4, 2021, pp. 213-221 4.1. drivable area detection the raspberry pi 4 and a streaming service are started. the streaming service is based on real time streaming protocol (rtsp). streaming images are sent to the cloud through cloud platforms, and apis are used for object identification. the object detection services used in this study are provided by google cloud vision, amazon rekognition, and azure computer vision cloud platforms to obtain the position coordinates, category names, and confidence scores for the objects in an image. obstacles are identified by using multiple clouds to increase the amount of information because insufficient information would have negatively affected the suitability of the planned path. moreover, accurate object class names and confidence scores are not required because the aim is to obtain information on surrounding obstacles. fig. 6 displays the results of object marking when using multiple cloud platforms, and the marked boxes of the same color represent the detection results obtained from the same platform. here, the gcp vision is presented by red color and the azure vision is presented by blue color. the program for drivable area detection is written in the python 3 language. first, the overlapping object bounding boxes are removed. then, the drivable area is detected and the waypoint is set. fig. 7 displays the result obtained after subjecting fig. 6 to drivable area detection. the green lines represent the bottom lines of multiple drivable areas. fig. 6 results of the comprehensive mark bounding box of cloud object detection fig. 7 schematic of drivable area detection 4.2. results and analysis fig. 8 depicts the two areas used in the experiment. fig. 8(a) displays a narrow walking passage in a laboratory. this image indicates that the exit of the aisle is on the left; therefore, the robotic car should turn to the left at the end of this aisle to avoid collision with the stacked boxes. fig. 8(b) displays a wide laboratory area with two exits. fig. 9 displays the legend for the different bounding box colors in figs. 10 and 11, which depict the continuous images obtained when the robot car moves according to the system instructions in the narrow and wide laboratory areas, respectively. in these images, different colors are used to represent the obstacles identified using different platforms. the yellow lines in figs. 10 and 11 indicate the real-time planned path for each image. the images in the aforementioned figures prove that visual recognition could be used to obtain a bounding box for path planning. thus, cloud computing facilitates real-time dynamic path planning. (a) narrow area (b) wide area fig. 8 panoramic images of the experimental areas 218 advances in technology innovation, vol. 6, no. 4, 2021, pp. 213-221 fig. 9 legend of the bounding box colors the third perspective the third perspective the third perspective the third perspective the vehicle view the vehicle view the vehicle view the vehicle view (a) at 5 s (b) at 15 s (c) at 25 s (d) at 35 s fig. 10 real-time images of the path planning and movement of the robotic car in the narrow laboratory area the third perspective the third perspective the third perspective the third perspective the vehicle view the vehicle view the vehicle view the vehicle view (a) at 5 s (b) at 15 s (c) at 25 s (d) at 35 s fig. 11 real-time images of the path planning and movement of the robotic car in the wide laboratory area all the obstacles on the ground are marked with white bounding boxes because the correct labeling of the object category is not crucial. moreover, the threshold to limit the object discrimination rate is obtained. multiple travel zones are detected by using a self-developed algorithm, and path planning is completed rapidly by using the obtained obstacle information. detailed experimental data are listed in tables 2 and 3. legend google amazon microsoft obstacles 219 advances in technology innovation, vol. 6, no. 4, 2021, pp. 213-221 table 2 detailed experimental results for cloud object detection figure path planning time (s) cloud platform class name detection confidence score left boundary value right boundary value top boundary value bottom boundary value 10(a) 0.001 gcp packaged goods 0.76 60 259 0 109 aws box 0.95 66 345 139 444 gcp packaged goods 0.69 67 261 64 171 gcp shipping box 0.61 79 340 150 432 azure wall 0.35 656 878 0 391 10(b) 0.02 gcp shipping box 0.76 0 230 85 511 aws box 0.56 0 232 103 500 10(c) 0.001 azure box 0.68 0 238 0 549 gcp shipping box 0.71 2 221 0 489 azure wall 0.56 896 1280 10 503 10(d) 0.001 gcp shoe 0.62 765 1004 285 344 azure cabinet 0.36 1118 1272 0 468 11(a) 0.001 gcp furniture 0.59 3 538 0 386 azure chair 0.38 628 1073 0 249 11(b) 0.001 gcp person 0.73 2 243 2 427 aws high heel 0.55 5 240 133 418 gcp person 0.71 10 254 1 434 azure chair 0.51 829 1253 74 292 azure wall 0.36 862 1216 20 216 11(c) 0.001 azure chair 0.45 2 94 274 481 azure chair 0.44 916 1280 78 267 11(d) 0.001 azure chair 0.38 124 323 0 387 aws airplane 0.82 497 1248 0 288 gcp person 0.69 787 1105 0 264 gcp home appliance 0.54 791 1121 0 266 table 3 average time required for cloud service object identification google cloud vision api 0.356 (s) microsoft azure computer vision api 1.042 (s) amazon web services rekognition api 2.098 (s) 5. conclusions this study developed a method in which the bounding box information returned by multiple cloud object detection services is integrated to detect the drivable area and plan a movement route. the experimental results reveal that robotic cars can perform appropriate path planning by using the developed method. the obtained images indicate that the planned route using the developed method is safe. the goals of this research were to reduce the cost of developing a collision-avoidance system, the cost of computing, and the dependence on an object sensor. cloud vision models are used to reduce the amount of computing in the designed system. the designed system is simple, has low cost, and contains only one lens. routes are planned through simple mathematical operations that do not burden the developed real-time path-planning system. latency is a critical real-time problem that should be considered when using cloud computing local area network signals because it affects system operation. the delay problem can be solved using next-generation high-speed networks that provide high transmission speeds with novel visual recognition models. conflicts of interest the authors declare no conflict of interest. 220 advances in technology innovation, vol. 6, no. 4, 2021, pp. 213-221 references [1] p. h. lin, c. y. lin, c. t. hung, j. j. chen, and j. m. liang, “the autonomous shopping-guide robot in cashier-less convenience stores,” proceedings of engineering and technology innovation, vol. 14, pp. 9-15, january 2020. [2] y. lecun, l. bottou, y. bengio, and p. haffner, “gradient-based learning applied to document recognition,” proceedings of the ieee, vol. 86, no. 11, pp. 2278-2324, november 1998. [3] a. krizhevsky, i. sutskever, and g. e. hinton, “imagenet classification with deep convolutional neural networks,” communications of the acm, vol. 60, no. 6, pp. 84-90, june 2017. [4] k. he, x. zhang, s. ren, and j. sun, “deep residual learning for image recognition,” ieee conference on computer vision and pattern recognition, june 2016, pp. 770-778. [5] g. huang, z. liu, l. v. d. maaten, and k. q. weinberger, “densely connected convolutional networks,” ieee conference on computer vision and pattern recognition, july 2017, pp. 2261-2269. [6] c. y. wang, h. y. m. liao, y. h. wu, p. y. chen, j. w. hsieh, and i. h. yeh, “cspnet: a new backbone that can enhance learning capability of cnn,” ieee/cvf conference on computer vision and pattern recognition workshops, june 2020, pp. 1571-1580. [7] a. g. howard, m. zhu, b. chen, d. kalenichenko, w. wang, t. weyand, et al., “mobilenets: efficient convolutional neural networks for mobile vision applications,” https://arxiv.org/pdf/1704.04861.pdf, april 17, 2017. [8] k. h. shih, c. t. chiu, j. a. lin, and y. y. bu, “real-time object detection with reduced region proposal network via multi-feature concatenation,” ieee transactions on neural networks and learning systems, vol. 31, no. 6, pp. 2164-2173, june 2020. [9] s. ren, k. he, r. girshick, and j. sun, “faster r-cnn: towards real-time object detection with region proposal networks,” ieee transactions on pattern analysis and machine intelligence, vol. 39, no. 6, pp. 1137-1149, june 2017. [10] d. feng, c. haase-schütz, l. rosenbaum, h. hertlein, c. gläser, f. timm, et al., “deep multi-modal object detection and semantic segmentation for autonomous driving: datasets, methods, and challenges,” ieee transactions on intelligent transportation systems, vol. 22, no. 3, pp. 1341-1360, march 2021. [11] w. he, z. huang, z. wei, c. li, and b. guo, “tf-yolo: an improved incremental network for real-time object detection,” applied sciences, vol. 9, no. 16, 3225, august 2019. [12] h. law and j. deng, “cornernet: detecting objects as paired keypoints,” european conference on computer vision, september 2018, pp. 734-750. [13] k. duan, s. bai, l. xie, h. qi, q. huang, and q. tian, “centernet: keypoint triplets for object detection,” ieee/cvf international conference on computer vision, october 2019, pp. 6569-6578. [14] z. tian, c. shen, h. chen, and t. he, “fcos: fully convolutional one-stage object detection,” ieee/cvf international conference on computer vision, october 2019, pp. 9626-9635. [15] m. tan, r. pang, and q. v. le “efficientdet: scalable and efficient object detection,” ieee/cvf conference on computer vision and pattern recognition, june 2020, pp. 10778-10787. [16] j. chen, w. zhan, and m. tomizuka, “autonomous driving motion planning with constrained iterative lqr,” ieee transactions on intelligent vehicles, vol. 4, no. 2, pp. 244-254, june 2019. [17] y. luo, p. cai, a. bera, d. hsu, w. s. lee, and d. manocha, “porca: modeling and planning for autonomous driving among many pedestrians,” ieee robotics and automation letters, vol. 3, no. 4, pp. 3418-3425, october 2018. [18] r. mur-artal, j. m. m. montiel, and j. d. tardós, “orb-slam: a versatile and accurate monocular slam system,” ieee transactions on robotics, vol. 31, no. 5, pp. 1147-1163, may 2015. [19] d. k. dewangan and s. p. sahu, “deep learning-based speed bump detection model for intelligent vehicle system using raspberry pi,” ieee sensors journal, vol. 21, no. 3, pp. 3570-3578, february 2021. [20] c. shao, c. zhang, z. fang, and g. yang, “a deep learning-based semantic filter for ransac-based fundamental matrix calculation and the orb-slam system,” ieee access, vol. 8, pp. 3212-3223, 2019. [21] j. s. sheu and c. y. han, “combining cloud computing and artificial intelligence scene recognition in real-time environment image planning walkable area,” advances in technology innovation, vol. 5, no. 1, pp. 10-17, december 2019. [22] t. xu, s. zhang, z. jiang, z. liu, and h. cheng, “collision avoidance of high-speed obstacles for mobile robots via maximum-speed aware velocity obstacle method,” ieee access, vol. 8, pp.138493-138507, 2020. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). 221  advances in technology innovation, vol. 3, no. 2, 2018, pp. 86 93 piglets comfort with hot water by biogas combustion under controllable ventilation cheng-chang lien 1,* , ching-hua ting 2 , jun-han mei 3 1 department of biomechatronic engineering, national chiayi university, chiayi, taiwan, roc. 2 department of mechanical and energy engineering , national chiayi university, chiayi, taiwan, roc. 3 graduate student, department of bioelectromechanical engineering, national chiayi university , chiayi, taiwan, roc. received 05 june 2017; received in revised form 23 august 2017; accept ed 13 sept ember 2017 abstract the purpose of this study is to develop a hot -water heating system for pig fa rms which use biogas as the energy source while the air quality is regulated using an inverter-controlled fan. the biogas is a by-product from the 3-stage wastewater treat ment process in regula r p ig farms. the b iogas is burned for hot water which is c irculated to wa rm piglet compart ments with regulated, forced ventilation. the hot water is connected to a heat exchanger and hot air is hence blown into the pigsty. to maintain the pigsty at a comfort at mosphere, ventilation is regulated using an inverter-controlled fan. the mechanica l ventilation is to be optimized as a compromise between indoor air quality and ventilation rate. the te mperature uniformity and a ir quality in the p igsty is to be secured for comfortability. expe rimental results show that hot water circu lating at 0.043 m 3 / min and 60°c could keep the pigsty at 28°c for a stocking density of 1.77 pig/ m 2 . forced ventilat ion of 1.7 a ch (air change rate per hour) at 28°c could keep the pigsty comfort in terms of indoor temperature, relative humidity, and carbon -dioxide concentration. keywor ds : biogas, piglets, hot-water heating s ystem, force ventilation 1. introduction biogas is the combustible gas generated through the mic roorganism fe rmentation of organic waste under the anaerobic environment. it is one of the renewable energy to replace the requirement of future energy resource, the generated volume will be affected along with the factors of solid ingredients, fermented temperature, humidity, ph value, microbia l strains and fermentation time of fe rmented organic matter so that the compositions are not e xactly the sa me. the ma in constituent of biogas is methane (ch4) which is around 55-70%, carbon dio xide (co2) appro ximately 30-40%, 0.2-05% of hydrogen sulfide (h2s), and very small a mounts of carbon monoxide, n itrogen and ammonia etc. direct e mission of methane in the atmosphere will cause the global greenhouse effect growing in intensity, and the capability of causing the greenhouse effect per unit of methane is 23 times of carbon d io xide, its harm to the earth environ ment for direct e mission of biogas cannot be ignored indeed [1-4]. the process of 3-stage wastewater treatment facility promoted in the pig farms of ta iwan includes the three stages of solid-liquid separation of e xcreta, anaerobic treat ment and aerob ic treat ment, and large a mount of biogas is generated during the anaerobic proces s, however most of p ig husbandry in ta iwan does not utilize the b iogas but emits to the at mosphere directly. as pointed out from the research, the methane content as generated from p ig sewage is higher than what is generated fro m the waste of crop farm, each hog sewage with the we ight of 60 kg is able to generate appro ximately 0.23 m 3 biogas per * corresponding author. e-mail lanjc@mail.ncyu.edu.tw advances in technology innovation, vol. 3, no. 2, 2018, pp. 86 93 copyright © taeti 87 day. biogas has high heating value and the fuel characteristics as well, it has been considered as the energy resource worth to promote and utilize, after biogas is burned, methan e becomes the carbon dioxide wh ich is able to lo wer the greenhouse effect, if the heat energy generated from the burned biogas is able to utilize, it may save significant energy, therefore, through th e utilization of biogas energy, the requirement of energy saving and carbon reduction can be achieved synchronously [5]. pig let will be weaned about four weeks after birth, the piglet just off from the weaning is very sensitive to the temperature in pigsty, if the daily te mperature in p igsty drop is over 2°c, it will introduce the group diarrhea and respiratory diseases , and significantly affect the growth performance for p iglet. there fore, the te mperature control in pigsty is one of the key factors for enhancing the production performance an d economic benefit of p ig fa rms [6]. the ventilat ion of p igsty can be categorized as the natural ventilation and forced ventilation (mechanica l ventilation). the main purpose of ventilation is to supply the oxygen required for the an ima l breathing, re move the e xcess heat and moisture, p lus reduce the dust and limit the generation of harmful gases, such as ammonia and carbon dio xide etc. as pointed fro m the research, the staffs of livestock farm shall not e xpose to the environmental with the concentration of carbon dioxide (co2) over 1500 pp m and concentration of ammonia (nh3) over 7 ppm, the effective ventilat ion system shall be able to provide the best mic roclimate condition for the working area o f p ig farme rs and anima l living a rea [7-8]. the design of heat preservation fac ility for piglets in netherlands is to use the hot air fro m underground tunnels exchanged heat through the hot -water pipes to heat and preserve the heat for the piglet living area [9]. the purpose of this research is to develop a set of hot-water heating system using biogas as the fuel, and take the advantage of biogas energy generated fro m the 3 -stage wastewater treatment process and apply it to pig farms for the heat preservation of piglets. ut ilize the burning of biogas and heating the water, delivered to the pigsty via pipeline, the hot-water flow blowe rs to blow the hot air through the air heater and exchanges heat in the pigsty to proceed the heating preservation for piglets, the pigsty collocates with the inverter fan to proceed the mechanica l ventilation test with different forced ventilation temperature and regular ventilation percentage, and explore against the temperature uniformity and air quality in the pigsty. 2. material & method 2.1. experimental equipment the biogas generated from the pen manure of pig fa rms through the anaerobic fermentation process treatment, use the pipeline de livered/centralized to store in the red mud rubber bags, since the biogas contains the hydrogen sulfide which is corrosive to the metal equip ment, use the blower to dra w the biogas with the averagely negative pressure of 0.2 mpa through the biogas purification device to re move the hydrogen sulfide in the biogas, and then deliver to the biogas burning furnace a s the source of burning energy. the schematic image of biogas burning hot-water piglet heat preservation system is shown in the fig. 1. the imp le mentation of biogas burning hot-water heat preservation system can be categorized as the biogas hot -water heating device, enclosed pigsty heat preservation system and inverter fan control system. the biogas hot-water heating device includes the biogas -burning furnace, electrical au xiliary heating device and hot -water storage tank. inside the ma in body of biogas -burning furnace is able to store appro ximately 0.7 m 3 water, and 52 pcs of stainless steel pipes are installed as the passage for hot air, when the biogas furnace is burned/heated fro m the bottom, the heated area will be able to increase via t he hot air passed through the hot air passage, so that the heating efficiency is enhanced and the tempe rature of hot-water increases rapidly, wh ile an electrical au xilia ry heating device is added to deal with the situation of insufficient b iogas volume, and the temperature sensor of hot-water is installed to measure the hot-water temperature (tw), so that the heating device is able to operate year-round. advances in technology innovation, vol. 3, no. 2, 2018, pp. 86 93 copyright © taeti 88 when the thermocouple temperature o f a ir (tc) in the enclosed pigsty is lower than the default temperature (tset), the piglet heat preservation system starts the external c ircu lation, de liver the hot -water through the inverter motor and enter into the circulation-type blower in the pigsty to proceed the heat exchange, heat up the air inside the pigsty by blowing the hot air in order to achieve the effect of temperature control and heat preservation in the pigsty. since the pigsty is an enclosed space, the ventilation system is required to ventilate and change air in orde r to avoid the high concentration of noxious gas in the pig sty, this research uses the inverter fan control system to perform the ventilat ion and air change of p igsty, the inverter fan system consists of 2 inverter fans with the ventilation efficiency of 0.64 m 3 /s (slf-730, autofan, korea) collocating with the inverter control system (ecs-3m , va rifan, usa), plus integrates the default temperature (tw) and the defaulted setting of air temperature in the pigsty (tset) together with the low limit air temperature setting function. fig. 1 schematic image of biogas burning hot-water piglet heat preservation system 2.2. experimental methods please locate fig. 2. under the text this research uses temperature/humid ity sensing recorder (u10-003, hobo-onset, usa) and carbon dio xide sensing recorder (attach with te mperature/humidity sensing , telaire 7001, te laire, usa) to measure the air te mperature, hu mid ity and concentration of carbon dioxide in the pigsty. their planned locations are shown in the fig. 2. te mpe rature/humid ity sensing recorders are located left front (lf), left rea r (lr), right front (rf) and right rear (rr) separately, their he ight fro m the raising bedding is 80 c m, the height fro m the ra ising bedding is 25 c m for rear front down (rfd) and right rear down (rrd), the carbon dio xide sensing recorder is erected at the central location of p igsty, the height fro m the raising bedding is 90 c m, it is able to measure the central air te mperature (tm), central re lative humid ity (rhm) and concentration of carbon dioxide (co2) of central location in the pigsty simu ltaneously, both temperature/humidity sensing recorders and carbon dio xide sensing recorde r are configured to record the data once every 30 seconds. fig. 2 schematic image of temperature/humidity s ensing recorders and carbon dioxide sensor location in enclosed pigsty advances in technology innovation, vol. 3, no. 2, 2018, pp. 86 93 copyright © taeti 89 the stocking density of piglets in the pigsty is 1.77 pig/ m 2 , it is assumed that the pigsty is fully enclosed and the air amount flowed into the pigsty (qin) is equal to the air amount flowed out fro m the pigsty (qout), while define the air change rate per hour (ach) in the enclosed pigsty is the times that the outside air a mount of pigsty to replace the equivalent amount of volume space in the pigsty, its calculation formula is shown as eq. (1). ach = qv / v (1) in which, ach: air change rate in p igsty (time/hr), qv: a ir a mount flowed into pigsty from outside of pigsty per hour (m 3 /hr), v: total volu me of enclosed pigsty is 287.28 m 3 , when perform the ventilation and air change at the regular ventilation percentage of 10%, 20% and 30%, a fter conversion, its ventilation and air change rate is 1.7 ach, 3.4 a ch and 5.1 a ch separately. the hot-water temperature (tw) configured at e xperiment is 60°c, hot-water flo w rate is 0.043 m 3 /min and the default temperature in the pigsty (tset) is 28°c, and collocate with the inverter fan control system to perform the ventilation and air change to maintain the a ir quality. as to the setting part of forced ventilation te mperature, when the inside air te mperature of pigsty is higher than the forced ventilation temperature, then start the inverter fan, continue to operate with the bleed air rate of 1.27 m 3 /s until the thermocouple temperature in the pigsty (tc) consistent with the forced ventilation temperature, and the stop. as to the setting part of regula r ventilation percentage, the ventilation and a ir change cycle is once per 10 minutes, if the configured ventilation percentage is 10%, the inverter fan operates 1 minute and then stops for 9 minutes, if the configured ventilation percentage is 30%, the inverter fan operates 3 minutes and then stops for 7 minutes. the calculation method of temperature uniformity (tu) is to take the air te mperature of different measuring location to subtract the air temperature of centra l location (tm) separately, then select the average value of two ma ximu m a ir te mperature difference, this average value is the temperature uniformity, shown as eq. (2). tu = (td1+td2)/2 (2) in which, t d1: largest air te mperature difference between different measuring location and central location (°c), t d2: second largest air temperature difference between different measuring location and central location (°c). this research performs the ventilation and air change of enclosed pigsty by using the biogas burning hot -water heat preservation system collocated with the inverter fan, and performs the e xperiment in the pigsty with stocking density of 1.77 pig/m 2 , measures the air te mperature, hu mid ity and concentration of carbon dio xide at d ifferent location in the pigsty, the forced ventilation te mpe rature configured by the inverter fan is 28°c, and changes the air change rate of ventilat ion to 1.7 a ch, 3.4 ach and 5.1 a ch separately, fro m the calculated temperature uniformity, e xp lore if the air te mperature distribution in the pigsty is uniform or not, and illustrates if the air quality is suitable or not by the concentration of carbon dio xide in the pigsty. 3. results & discussion when the forced ventilation te mperature is 28°c and air change rate of regular ventilat ion is 3.4 ach, the typical diagra m of time variation of the right front (rf), right rear (rr) and right front down (rfd), right rear down (rrd) te mperature in t he pigsty versus the outside temperature of pigsty (t0) for 4 consecutive days is shown as fig. 3, it can be seen fro m fig. 3a , the outside temperature fluctuation of t0 is more obvious in the day and night. the temperature of right front is slightly greater than temperature of right rear in the night which indicates that the pigsty has greater impact by the hot air blo wn fro m the a ir heater. however, on the whole, the te mperature change of rf and rr is not much , it indicates that the overall te mperature advances in technology innovation, vol. 3, no. 2, 2018, pp. 86 93 copyright © taeti 90 change in the pigsty is quite stable. the te mperature change of rfd and rrd can be considered as the temperature change of activity range for p iglets, therefore, it can be seen fro m fig. 3b, the temperature variat ion curves of rfd and rrd nearly overlap which indicates that the temperature change of activity range for the piglets in the overall pigsty is quite consistent. (a) rf, rr and outside temperature (to) (b) rfd, rrd and outside temperature (t o) fig. 3. typical d iagra m of rf, rr, rfd and rrd versus outside temperature of pigsty (t0) for 4 consecutive days under 28°c for forced ventilation temperature and 3.4 ach for regular ventilation under the condition of 28°c for the forced ventilation te mperature and 3.4 a ch for the air change rate of regula r ventilation, the typical diagra m of time variation of the right front (rf), right rear (rr), left front (lf) and left rear (lr) temperature in the pigsty for 4 consecutive days is shown as fig. 4, it can be seen fro m the figure, the te mperature ch ange of right front (rf) and left front (lf), right rear (rr) and left rear (lr) nearly overlap wh ich indicates that the temperature change of left and right sides on the pigsty is quite consistent. fig. 4 typica l d iagra m o f rf, rr, rr and lr versus outside temperature of p igsty (t0) for 4 consecutive days under 28° c for forced ventilation temperature and 3.4 ach for regular ventilation. advances in technology innovation, vol. 3, no. 2, 2018, pp. 86 93 copyright © taeti 91 under the condition of 28°c fo r the forced ventilation te mperature, the statistic table of te mperature and re lative hu midity for the right front (rf), right rear (rr), le ft front (lf), le ft rear (lr), right front down (rfd) and right rear down (rrd) to the diffe rent air change rate of regula r ventilat ion is shown as table 1. it can be seen that the temperature of each location in the pigsty is quite close to the defaulted temperature of 28°c in the p igsty which indicates that the temperature change in the overall pigsty is quite uniform and stable, however, the te mperature of rfd and rrd is slightly lo wer than 28°c, its temperature difference is only 0.29°c ~1.39°c appro ximately. it can also be seen that since the location of rr and lr are more c lose to the inverter fan, rr and lr has higher relat ive hu midity as relat ive to rf and lf, however, the relat ive hu midity of piglet growth area (rfd, rrd) does not change along with the air change rate of regular ventilation i.e . ma intains between 80%-89%. table 1 under 28°c for the forced ventilation te mperature, at the d iffe rent air change rate o f regular ventilation, the statistic table of temperature and re lative hu mid ity for its right front (rf), right rear (rr), le ft front (lf), le ft rear (lr), right front down (rfd) and right rear down (rrd) temperature (°c) air change rate per hour (ach) relative humility (%) air change rate per hour (ach) 1.7 ach 3.4 ach 5.1 ach 1.7 ach 3.4 ach 5.1 ach rf 30.17±0.86 *1 29.57±0.84 30.09±0.74 rf 66.76±9.69 75.71±9.21 65.37±7.37 rr 28.29±0.39 28.71±0.40 28.24±0.28 rr 77.05±6.30 81.67±6.02 77.23±4.57 lf 29.93±0.73 29.44±0.71 *2 lf 69.26±9.05 77.91±8.74 *2 lr 28.32±0.44 28.82±0.43 28.02±0.30 lr 84.96±8.92 86.37±2.77 83.40±2.05 rfd 27.03±0.52 27.71±0.53 27.12±0.43 rfd 87.41±3.74 89.29±4.57 84.91±2.79 rrd 26.61±0.73 27.59±0.62 26.89±0.50 rrd 81.83±4.72 83.79±4.28 80.02±3.42 *1 : mean ± std *2 : lost data due to bit off by the piglets. under the condition of 28°c for the forced ventilation temperature and 3.4 ach for the regular ventilat ion, the typical diagra m of central te mperature (tm), concentration of carbon dioxide in the pigsty and outside temperature of pigsty (to) is shown as fig. 5. it can be seen fro m the figure , the central te mperature (tm) does not have the great fluctuation along with the outside temperature change of pigsty, its te mperature va lue stably maintains around 28°c. the concentration of ca rbon dio xide changes along with the t ime and has the significant fluc tuation, it shall be relat ive to the lifestyle of p iglets. as shown in the figure, the sudden drop of carbon dioxide (a rrow 1) is caused due to the owner entered into the pigsty to observe the night life situation of piglets ; the sudden drop of rear section (arrow 2) is caused due to the owner entered into the pigsty for cleaning. fig. 5 typica l diagra m o f central te mperature (tm), concentration of carbon dioxide and outside temperature of pigsty (t0) for 4 consecutive days under 28°c for forced ventilation temperature and 3.4 ach for regular ventilation at the condition of 28°c for the forced ventilat ion te mperature, the statistic of central te mperature (tm), te mperature error rate, central re lative humid ity (rhm) and concentration of carbon dio xide (co2) in the pigsty under the different a ir change rate of regular ventilation is shown as table 2. it can be seen from the table, under the different setting of air change rate for regular ventilation, central te mperature (tm ) is stable between 28°c-29°c and the standard deviation is very small, it is indicated that the temperature change in the pigsty is quite s table, and the temperature e rror rate is around 1.93% -3.88% , this a lso indicates advances in technology innovation, vol. 3, no. 2, 2018, pp. 86 93 copyright © taeti 92 that the temperature control of this heat preservation system is quite stable. the central re lative humid ity (rhm) is stable between 73.11%-82.21% at diffe rent air change rate of regular ventilat ion, and the concentration of carbon dioxide is ma intained under 1000 ppm. when the air change rate of regular ventilation is 5.1 a ch, the concentration of carbon dio xide in the pigsty is very close to the outdoor concentration of carbo n dioxide, the air quality in the pigsty is good. table 2 statistic of centra l te mperature (tm), te mperature error rate, centra l relat ive humid ity (rhm) and concentration of carbon dioxide (co2) in pigsty at 28°c for forced ventilation te mperature under different a ir change rate of regular ventilation air change rate per hour (ach) 1.7 ach 3.4 ach 5.1 ach tm(°c) 28.49±0.52 29.09±0.46 28.79±0.31 e. r. * (%) 1.93%±1.68% 3.88%±1.65% 2.83%±1.12% rhm(%) 73.11±12.94 77.19±15.99 82.21±8.54 co2(ppm) 896.26±145.08 598.57±294.02 406.80±108.18 * e. r. : error rate (%) =[(tset-tm) /tset]*100% , tset =28°c take the measured temperature of rf, rr, lf, lr, rfd & rrd and the central te mperature (tm) to perform the calculation of te mperature uniformity, the results are shown as table 3, it can be seen fro m the table, at the defaulted temperature 28°c in the pigsty, the temperature uniformity of enclosed pigsty under different air change rate of regular ventilation is between 0.99°c -1.78°c, under the condition of 3.4 ach for the a ir change rate of regular ventilat ion and 28° c for the forced ventilation temperature have better temperature uniformity inside the pigsty. table 3 te mperature un iformity in enclosed pigsty under the diffe rent air change rate of regular ventilation at 28°c for forced ventilation temperature air change rate per hour (ach) 1.7 ach 3.4 ach 5.1 ach temperature uniformity (°c) 1.78 0.99 1.6 4. conclusions the stoking density of piglets in the enclosed pigsty is 1.77 pig/ m 2 , under the setting of 28°c for both defaulted air temperature and forced ventilation te mperature in the pigsty, collocate with different a ir change rate of regular ventilation, i.e. 1.7 a ch, 3.4 ach and 5.1 a ch to perfo rm the e xperiment. the concentration of ca rbon dio xide at diffe rent air change rate of regular ventilation are lower than 900 pp m, it is able to stably ma intain the temperature, re lative humidity and concentration of carbon dioxide in the nursery, and achieve the effective heat preservation effect, and maintain the good air quality in the pigsty. when the air change rate of regular ventilation is configured as 5.1 ach, the concentration of carbon dio xide in the pigsty is quite close to the 400 pp m concentration of carbon dioxide outside the nursery, it is indicated that the air quality in the pigsty is very good. as known fro m the e xpe rimental results, this biogas burn ing hot-water system co llocating with inverter fan to perfo rm the ventilation in enclosed the pigsty, the temperature change in the pigsty is quite un iform, apply to the heat preservation for the enclosed pigsty is feasible and efficient. references [1] p. v. rao, s. b. saroj, r. dey, and s. mutnuri “biogas generation potential by anaerobic digestion for sustainable energy development in india,” renewable and sustainable energy reviews , vol. 14, no. 7, pp. 2086-2094, september 2010. [2] m. van haren and r. fleming, “electricity and heat production using biogas from the anaerobic digestion of livestock manure-literature review,” university of guelph, ridgetown college, 2005. [3] j. b. holm-nielsen, t. a. seadi, and p. oleskowicz-popiel, “the future of anaerobic digestion and biogas utilization,” bioresource technology, vol. 100, no. 22, pp. 5478-5484, november 2009. advances in technology innovation, vol. 3, no. 2, 2018, pp. 86 93 copyright © taeti 93 [4] r. c. saxena, d. k. adhikari, and h. b. goyal, “biomass -based energy fuel through biochemical routes: a review,” renewable and sustainable energy reviews , vol. 13, no. 1, pp. 167-178, january 2009. [5] c. c. su, c. m. hong, m. t. koh, and s. y. sheen, “swine waste treatment in taiwan,” department of livestock management, taiwan livestock research institute, 1994. [6] h. pandorfi and i. j. o. da silva, “eva luation of the behavior of pig lets in d ifferent heating systems using analysis of image and electronic identification,” agricultural engineering international, 2005. [7] k. donham, p. haglind, y. peterson, r. rylander, and l. belin, “environmental and health studies of farm workers in swedish swine confinement buildings ,” british journal of industrial medicine, vol. 46, no. 1, pp. 31-37, january 1989. [8] a. j. heber, j. q. ni, b. l. haymore, r. k. duggirala, and k. m. keener, “air quality and emission measurement methodology at swine fin ishing buildings ,” trans actions of the american society of agricultura l engineers , vol. 44, no. 6, pp. 1765-1778, january 2001. [9] g. p. a. bot and j. h. m. met z, “measurement, eva luation and control of the microc limate in rooms for weaned piglets ,” in: a. v. van wagenberg, j. m. aerts, a. van brecht, e. vranken, t. le roy, d. berc kmans, “climate control based on temperature measurement in the animal-occupied zone of a pig room with ground channel ventilation,” transactions of american society of agricultural engineers, vol. 48, pp. 111-125, january 2005.  advances in technology innovation, vol. 3, no. 1, 2018, pp. 01 08 point-structured human body modeling based on 3d scan data ming-june tsai 1,* , hsueh-yung lung 1 1 dep artment of m echanical engineering, national cheng kung university , tainan, taiwan. received 05 june 2017; received in revised form 13 sept ember 2017; accept ed 20 sept ember 2017 abstract a novel p oint-structured geometrical modellin g for realistic human body is introduced in this p ap er. this technique is based on the feature extraction from the 3d body scan data. anatomic feature such as the neck, the arm p its, the crotch p oints, and other major feature p oints are recogn ized. the body data is then segmented into 6 major p arts. a body model is then constructed by re-samp ling the scann ed data to create a p oint-structured mesh. the body model contains body geodetic landmarks in latitudinal and lon gitudinal curves p assing throu gh those feature p oints. the body model p reserves the p erfect body shap e and all the body dimensions but requires little sp ace. therefore, the body model can be used as a mannequ in in garment industry , or as a manikin in various human factor designs, but the most imp ortant app lication is to use as a virtue character to animate the body motion in mocap (motion cap ture) sy stems. by adding suitable joint freedo ms between the segmented body links, kinematic and dy namic p rop erties of the motion theories can be ap p lied to the body model. as a result, a 3d virtual character that is fully resembled the original scann ed individu al is vividly animatin g the body motions. the gap s between the body segments due to motion can be filled up by skin blending technique usin g the characteristic of the p oint -structured model. the model has the p otential to serve as a standardized dataty p e to archive body information for all custom-made p roducts. keywor ds: 3d body model, body motion animation, feature recognition 1. introduction hu man body modeling is the act that creates a proper shape description for a specific human body. the shape description is a geo metrical representation in the 3d co mputer environ ment and, usually, co mb ined with skin colo ring and te xture mapping to perform more rea listic results. a convenient way to obtain a rea listic representation of human body in the virtual world is using 3d dig itizing technique [1]. recently, a number of 3d scanning systems with the ability to quickly and easily obtain digital human body model are now co mme rcia lly availab le [2-5]. ho wever, the scanned point cloud is commonly scattered and unorganized, and only very little semantic information involved. processing the raw data, therefore , is a cruc ial task in the e xtendable utilit ies of the human body scans. a number of data processing for human body scans, such as the data segmentation for significant parts, the indication of landmarks and feature points, and the dimension measurements, have now been proposed [6-9]. it is obvious that the key point to turn the raw scanned data into useful informat ion should be by means of feature recognition, data extraction, and symbolization. the feature e xt raction plays an important ro le not only in the description of hu man body, but also in the anthropometric analysis of human factor designs. recently, a novel method of body feature e xtract ion fro m a ma rke r -less scanned body was presented [10-11], in which the description of human body features are interpreted into logica l mathemat ical defin itions. tsai and fang [12] patented a feature based data structure for co mputer manikin. accordingly, the body scanned d ata are segmented into 6 six parts: head, torso, two arms and two legs. each part can be encoded into a range image format, and then the featur e points and curves are recognized according to the gradient of the gray scale in the range image format. a point -structured * corresponding author. e-mail address: mjtsai@mail.ncku.edu.tw advances in technology innovation, vol. 3, no. 1, 2018, pp. 01 08 copyright © taeti 2 geometric model is constructed by re-sampling the original 3d scanned data into an interweaving geodetic latitudinal and longitudinal curves, and then achieving an accurate model to describe the human body . there is also an increasing demand for realistic human animation models in various applications such as interactive computer ga mes, virtual reality, and movie a musement industry. although an artificia l mode l is able to serve as useful animation tool to control articu lated mot ions and body surface deformat ions according to different body postures, it is still required a large degree of skill and manual intervention for more rea listic. however, it rese mbles to nobody. currently, a lo t of studies on human motion capture and analysis have been presented [13-16]. so far, there is no direct way to use the personalized body model to animate his own body motion. it is persuasive that using the body model created fro m the scanned human body data to animate his own personalized body motions will yie ld a more rea listic result without laborious motion retargeting intervention because it is a d irect and an e ffic ient way to e xhib it prec ious body motions. by appropriate ly addin g joint freedo ms between the body lin ks to cons truct a kine matic model, the point-structured geometric mode ls obtained by scanned data can naturally and exactly be animated. in this paper, we have also shown that the skin gaps between the body segments due to motion can be filled up by dual quaternion linear blending using the characteristic of the point -structured model. 2. point-structured geometric model the surface informat ion of hu man body can be obtained quickly and easily by using 3d scanning technology. however, the scanned data points without further processing are un-organized and are too huge to be directly used in practice. therefore, a proper way of the data processing with the ability to reduce the a mount of c loud points while e xt racting the significant characteristics of human body will b e e xtre me ly beneficia l for further applicat ions. such a process can be achieved by employing the computational geometry and image processing techniques on the scanned point cloud. according to the geometric characteristics of human body, it is general to divide the body into six topological parts: head, torso, two arms and two legs. the arms can be seg mented fro m the body by the armp it, and the legs can be segmented by the crotch. in the following, we only illustrate the method of point -structured modeling for the torso. 2.1. feature recognition the outside contour of a human body is a very comple x smooth surface, and it is difficult to be described and represented clearly without the help of some body significant features. by the way, the dimensional measure ments of the body used in garment design and anthropometric surveys are a lways dependent on the feature points and feature lines. however, it seems that there is no unanimity on the definit ions of these specific body features in the literatures [17-18]. in order to e xt ract these feature lines and feature points fro m the 3d scanned point cloud automatica lly, the se mantic definit ions of body features are needed to be interpreted into the mathemat ical definit ions. a number of methods based on the image processing techniques, computational geometry and computer graphics have been used to identify these features from the 3d scanned points. and the developed algorithms fo r searching these features automatically have been presented in [10-12]. the body feature lines searching result by [11] is shown in fig. 1(a). a point-structured model of the torso is developed using the concept of geodetic coordinate, which is similar to the longitudinal and latitudinal lines of the earth. it means that the longitudinal cu rves include all the feature curves in the vertical direction of the torso, whereas the latitudinal curves contain all the feature girth lines of the body. therefore, all the significant features can be preserved we ll in such a point -structured representation. the girth lines in horizontal a re important in the description of body curve. consequently, a total of 60 sections are arranged in the representation as shown in fig. 1(b). exc ept advances in technology innovation, vol. 3, no. 1, 2018, pp. 01 08 copyright © taeti 3 for the a lready e xt racted feature g irths wh ich have been assigned as sp ecific orders, the others between each interval are needed to be re-sampled by the method of interpolation fro m the origina l 3d scanned point cloud. the interpolation is conducted with a uniform distribution within each interval. front centerline upper neck line lower neck line shoulder line shoulder girth armpit girth bust girth under bust girth waist girth mid-waist girth hip girth crotch girth side line side line armhole armhole princess lines princess lines shoulder line back centerline 6 0 g ir th s (a) body feature lines searching result by [11] (b) arrangement of structured girth lines fig. 1 body feature lines obtained from the 3d body scanned point cloud and the re -sampled girth lines 2.2. structured points in general, there a re a total of 80 structured points (0~79) used to characterize the circu mference of each latitudinal g irth line. so me specific structured points are needed to correspond to some feature points, respective. it means that they are assigned to locate on the longitudinal feature lines. for e xa mp le, the structured points along the armpit girth a re illustrat ed in fig. 2. therefo re, each structure point has a body geodesic coordinates (bgc) (g, h). where g designates the number of the girth that the point is located; and h is the order of the point on the girth, which also denotes the number of the longitudinal curve. for e xa mp le, the two po ints with numbers (38, 10) and (38, 70) represent the left and right bust points, which are very important land marks of the human body. the bgc are standardized and norma lized in our body model regardless of the gender, age, shape and race of the hu man. similarly, the structured points between two adjacent feature points are a lso generated by interpolation with a uniform distribution. it is clear that the point densities of these intervals are different. the reason is for the curve section with a larger curvature to have more points used for proper representation. fig. 2 structured points of the armpit girth line [12] 2.3. processing result the process to obtain the point-structured model of hu man body is very sophisticated. for the torso, a lot o f significant features with special geometric propert ies are needed to be recognized. based on these features, the concept of longitude and latitude are applied to construct the point-structured torso model. as a result, a feature-based bgc data structure is included in the point-structured model. such a concise representation, as shown in fig. 3, contains all the body features, anthropometric data, and body shape in the torso, it just like a body atlas that can be readily extracted as needed . advances in technology innovation, vol. 3, no. 1, 2018, pp. 01 08 copyright © taeti 4 fig. 3 point-structured torso model and its polygonal mesh in contrast to the torso, to obtain the point -structured models of the t wo arms and two legs are simple r. the a rm can be divided into upper arm and lowe r arm by the elbow girth. due to the cylindrica l shape of arm, its point -structured model can be constructed by using several girth lines distributed along the cylindrica l a xis. like wise, the point -structured model of the leg can be conducted by the same way. a mo re detailed statement of the model construction is shown in [19]. fina lly, a fu ll 3d point-structured body model consisting of only 10,214 structured points is obtained and is shown in fig. 4. the body model is also called the body geometric model (bgm). (a) point cloud (b) point-structured model (c) meshed by triangulation fig. 4 a 3d bgm 3. animation model it is demonstrated in this section how to converse the bgm into a kine matic model that can be used for animation purpose. before the 3d body scanned points can be used for animation, the scanned points should be put in well-organized order, this is our bdm. however, the bgm still cannot be animated because it has no joint freedo m. it means that the segmentation of the body scanned data should be according to the anatomical structure for jo int arrange ment. by the way , a joint model with high degrees of freedom is required for perfo rming much more hu man -like motion. with the kine mat ic analysis based on the joint model, it is not difficult to converse the bgm into an articulated kine mat ic model (bkm ). fortunately, the bgm has been segmented according to the anatomic feature at some joint positions. we just need to add appropriate joint freedoms between the adjacent body links. a realistic human kinematic body can be built for motion animation . 3.1. segmentation a virtual hu man model to perform much more hu man -like mot ion is generally dependent on how many degrees of freedom (dofs) it has. in order to achieve highly realistic human animation, the whole bgm is further divided into 23 parts as advances in technology innovation, vol. 3, no. 1, 2018, pp. 01 08 copyright © taeti 5 shown in fig. 5. these parts will be vie wed as the body links, and a suitable nu mber of joint freedoms are assigned to each of adjacent links. then, the joint-link kine matic model (bkm) is constructed for hu man mot ion animat ion. the jo int -link hierarchica l structure is modeled by five kine matic chains. the first chain consists of hip, waist, chest, neck, and head. the hip is the base fra me in this hie rarchica l structure. cha ins 2 and 3 are co mposed of the scapula, upper arm, lo wer arm, hand and fingers on the left and right, res pectively. the left and right thighs, lower legs, feet and toes are the links of chain 4 and 5, respectively. please note all of these links co me fro m the point -structured bgm which is constructed by the 3d scanned data of a specific person. so fa r, we have finished a compact static bgm with personal shape and appearance. for hu man animat ion, the only require ment is to use the bkm of the specific person and then animate these segmented lin k mode ls by applying the motion data. left toe left foot left lower leg left thigh left fingers left hand left lower arm left upper arm left scapula head neck right toe right foot right lower leg right thigh right fingers right hand right lower arm right upper arm right scapula chest waist hip fig. 5 segmentation for articulated human model 3.2. kinematic model according to the anatomical structure of human body, the type -1 bkm in this study is conducted by 5 kine matic chains with a total of 48 degrees of freedo m, not counting the 6 dof in th e pe lvis. the pelvis is considered as the based lin k of the body, which has 6 dof with respect to the fix coordinate system, and all the other joint freedo ms are co mputed based on this lin k. ho wever, human body has many more jo int freedo ms that performs ve ry co mple x body motion. it is fa milia r that more dofs will lead to a more realistic motion. ho wever, too many dofs also involve a rise in comple xity and computational cost. it is a trade-off that which kind of kine matic mode l is good enough for what kind o f body motion. in our study, it is believed that 48 joint freedoms are suitable for animating most of the body motions . (a) segmented body model (b) bkm with 48 dofs (c) bkm with 31 dof fig. 6 joint arrangement of the articulated models [20] advances in technology innovation, vol. 3, no. 1, 2018, pp. 01 08 copyright © taeti 6 as shown in fig. 6, there are 20 jo ints arranged in the different locations on the bkm. the four jo ints at l_grasp, r_ grasp, l_toe and r_toe each has one dof only. there are 8 jo ints with two dofs, i.e . l_ elbow, l_wrist, l_knee, l_ankle , r_ elbow, r_wrist, r_knee and r_ankle. the three-dof joints are l_ shoulder, l_ leg, r_shoulder and r_leg. all of these joints are regarded as revolute jo ints. subsequently, the four joints (nec k, waist, l_scapula and r_scapula) are modeled as four-dof joints composed of three revolute freedoms and one pris matic freedo m sliding a long the last rotational a xis. consequently, the model possesses totally 48 dofs in the 20 jo ints. other kine matic mode ls with different number of joint freedoms can be created, e.g. the type-2 bkm shown in fig. 6(c ), which has 31 dof to simulate a conventional humanoid robot. it is difficu lt to locate the jo int a xes at e xact position and orientation since the real hu man joints are not simp ly the rev olute or pris matic. if the hu man move ments can be captured by a precision motion trac ker, it is possible to locate the joint a xes accurately. however, using a good bkm with enough joint freedoms, people can replicate the mot ion without proble m. that is why the realistic human animation can be carried out by using the motion capture system. there fore, the motion data recorded fro m a mot ion capture system p lays an important role to produce highly realistic hu man animations. based on the bgm and bkm, tsai and lung [20] e mp loys a self-made mot ion capture system to acquire the space information of a ll lin ks, and then use the method of two-phased optimization to solve the joint angles via inverse kine matics. besides, an intelligent body motion processing system (ibmps) has also been developed by tsai and lung [21 ]. the result shows that highly realistic human animat ions can be achieved using the ibmps. while apply ing the original motion data to bgm and the joint angles to the bkm, the compa rison for four postures is shown in fig. 7, in wh ich all of these body segments are displayed without any blending function for the joint deformations . fig. 7 motion captured model compared with the model after optimization [21] 3.3. sk in blending because the segmented body parts are viewed as the lin ks wh ich are usually treated as rig id bodies, it is in ev idence the relative move ment of two ad jacent lin ks will cause some kinds of splits or penetrations on the joint position, which are in-appropriate for looking in an animation system. in order to have a more authentic appearance, blending function is emp loyed on the structure points to overcome this proble m. fig . 8 illustrated the effect of the blending function. it is obvious that the gaps at the joints (between the adjacent links) have been filled s moothly. the skin b lending is fu lfilled by e mp loying sclerp (quaternion spherica l linear interpolation) by using co nstant volume of the links as the constraint before and after blending. as a result, it yie lds a highly realistic representation with the blending. since the bgm contains the information of skin location, we will know how to move the structure point by studying the skin deformat ion during a body movement. this is another benefit of using the point-structured bgm. advances in technology innovation, vol. 3, no. 1, 2018, pp. 01 08 copyright © taeti 7 fig. 8 the effectiveness of blending function 4. conclusions hu man body modeling is now a very hot research topic. to produce highly rea listic 3d hu man an imation is very useful for many applications. reality is always dependent on both of the realistic appearance and move ments. in this paper, we use the point-structured modeling method to ma intain the real boy shape and reduce the amount of data points considerably. we also construct appropriate bkm to achieve the highly authentic body motion replication. it de monstrated that such a point structured human body model is very useful in the dig ital hu man body modeling and realistic body motion animat ion. the body model can also be used as a mannequin in ga rment industry, or as a manikin in v arious human factor design since it contains all the body dimensions, shape, as well as all body features . acknowledgements this research was supported by the project of national sc ience council, (project number: nsc 99-2221-e-006-018-m y3), and ministry of sc ience and technology of taiwan (pro ject number: most 103-2221-e-006-024) which are greatly appreciative. references [1] m. petrov, a. talapov, t. robertson, a. lebedev, a. zhilyaev, and l. polonskiy, “ optical 3d dig itizers: bring life to the virtual world,” ieee computer graphics and app lication, vol. 18, no. 3, pp. 28-37, may 1998. [2] cyberware, http://www.cyberware.com, 2012. [3] tc2, http://www.tc2.com, 2012. [4] vitus, http://www.vitronic.de, 2012. [5] creaform, http://www.creaform3d.com, 2012. [6] j. h. nurre , j. connor, e. a. le wark, and j. s. collie r, “on segmenting the three -d imensional scan data of a hu man body,” ieee transactions on medical imaging , vol. 19, no. 8, pp. 787-797, august 2000. [7] x. ju, n. werghi, and j. p. siebert, “automatic segmentation of 3d human body scans,” proc. of iasted int. conf. on comp. graphics and imaging, 2000. [8] k. robinette, m. boeh mer, and d. burnsides, “3-d landma rk detection and identification in the ca esa r pro ject,” proc. 3rd int. conf. on 3-d digital imaging and modeling, pp. 292-298, 2001. [9] r. p. pargas, n. j. staples, and j. s. davis, “automatic measurement e xtract ion for apparel fro m a three -dimensional body scan,” optics and lasers in engineering, vol. 28, no. 2, pp. 157-172, september 1997. [10] m. j. tsai, z. p. chen, and y. s. liu, “a study on the automatic feature search fro m 3d body scanners,” proc. of the 20th national conference, society of chinese mechanical engineering, taipei, dec ember 2003. advances in technology innovation, vol. 3, no. 1, 2018, pp. 01 08 copyright © taeti 8 [11] i. f. leong, j. j. fang, and m. j. tsai, “automatic body feature extract ion from a marker-less scanned human body,” computer-aided design, vol. 39, no. 7, pp. 568-582, july 2007. [12] m. j. tsai and j. j. fang, feature based data structure for computer manikin, us patent, 7,218,752 b2, may 15 2007. [13] l. a. wang, w. m. hu, and t. n. tan, “recent development in human motion analysis,” pattern recognition, vol. 36, no. 3, pp. 585-601, march 2003. [14] j. k. aggarwal and q. cai, “hu man motion analysis: a revie w,” co mputer vision and image understanding, vol. 73 , no. 3, pp. 428-440, march 1999. [15] t. b. moeslund and e. granu m, “a survey of co mputer vision -based human motion capture,” co mputer vision and image understanding, vol. 81, no. 3, pp. 231-268, march 2001. [16] r. poppe, “ vision-based human mot ion analysis: an overvie w,” co mputer vision and image understanding, vol. 108, no. 1-2, pp. 4-18, october-november 2007. [17] international organization for standardization, garment construction and anthropometric surveys – body dimensions, reference no. 8559-1989, switzerland: iso, 1989. [18] astm, standard termino logy relat ing to body dimensions for apparel siz ing, astm designation: d 5219-99, usa, 1999. [19] m. j. tsai, h. w. lee, and h. y. lung, “feature-based data structure for digital manikin,” u.s. patent , us 20130069936 a1, march 21, 2013. [20] m. j. tsai and h. y. lung, “two-phase optimized inverse kinematics for motion replication of real human mod els,” journal of the chinese institute of engineers, vol. 37, no. 7, pp. 899-914, april 2014. [21] m. j. tsai, j. h. chao, and t. w. yang, “ construction of a general motion editing system for hu man body and humanoid robots,” proc. asme 2014 int. design engineering technical conf. & computers and information in engineering (idetc/cie 2014), asme press, august, 2014, pp. v01at02a072.  advances in technology innovation, vol. 5, no. 3, 2020, pp. 190-201 innovative design of an elliptical trainer with right timing of the foot trajectory fu-chen chen 1,* , yih-fong tzeng 2 , meng-hui hsu 1 1 department of mechanical engineering, kun shan university, tainan, taiwan 2 department of mechatronics engineering, national kaohsiung university of science and technology, kaohsiung, taiwan received 05 march 2020; received in revised form 06 may 2020; accepted 10 june 2020 doi: https://doi.org/10.46604/aiti.2020.5645 abstract the existing elliptical trainer cannot provide the user with the real jogging exercising mode and does not meet the principles of ergonomics. the purpose of this paper is to propose and study an innovative elliptical trainer that imitates the right timing of the foot trajectory while jogging. first of all, this study proposes and illustrates the structure and function of the innovative elliptical trainer with quick-return effect. then, by using vector-loop method and motion geometry of the mechanism, the proposed innovative mechanism is studied kinematically. a design example is presented for interpreting the design process. at last, the foot trajectory of the innovative elliptical trainer is analyzed and confirmed. the simulation results confirm that the timing of the foot trajectory of the foot support members satisfies the principles of ergonomics, and keeps the user’s legs from injury. keywords: innovative design, mechanism design, kinematics, elliptical trainer 1. introduction jogging is a popular exercise, but it is known that the jogger’s knees may suffer from significant impact especially at the time when the user’s foot hits the ground. the knees could be injured after constantly taking the impacts for a period of time. it has been estimated that between 65-70% of runners will suffer an overuse injury in their lower extremities [1]. about 30 million americans run for recreation or competition. each year between 1/4 and 1/2 of runners lead to serious injury and cause a change in exercise or training [2-4]. therefore, the elliptical trainer is developed to guide the users’ feet to follow a track such that impact forces are minimized and the knees are well protected from being injured. research on the lower extremity movements and forces generated during exercise on the elliptical trainer have demonstrated the elliptical motion which produces lower impact forces than treadmill running during elliptical exercise and walking [5-7]. there are many patents about elliptical trainers, but few researches are studied on the mechanism design of elliptical trainer. shyu et al. [8] presented a novel design of adjustable elliptical trainer and the parameters that affect the elliptical path and inclined angle of the foot trajectory are investigated. knutzen et al. [9] studied the influence of ramp position on joint biomechanics with adjusting the ramp setting during elliptical trainer exercise. although quite a few kinematic structures have been used on the design of elliptical trainers, few of these apparatuses give pedal paths that might boost the lower extremit y kinematics of gaits near the ground. nelson and burnfield [10] proposed a novel design method for elliptical exercisers and created a foot trajectory that more closely mimics the lower extremity kinematics of gait. this design includes the substitution * corresponding author. e-mail address:fcchen@mail.ksu.edu.tw tel.: +886-6-2050496; fax: +886-6-2050509 advances in technology innovation, vol. 5, no. 3, 2020, pp. 190-201 191 of a modified cardan gear system for the typical crank. yin [11] used adams to simulate the dynamics of elliptical exerciser and accurately describe its performance and the characteristic curves of some major components. on conventional elliptical trainer, the joints connecting the flywheel to two supporting members are placed on the flywheel with a 180° phase angle. therefore, when one of the user’s feet is at the front end of a pedal trajectory and about to support the user’s weight, the other is at the rear end of the pedal trajectory as shown in fig. 1(a). in other words, the supporting travel a1 and the striding travel a2 of the pedal trajectory a have almost the same path length. critics of the elliptical trainer have expressed concerns that the foot trajectory seems unergonomic and could cause the knee to harmful loads, resulting in injury [4]. the foot trajectories during treadmill walking or running are tracked by using webcam technology [12] and wearable wireless ultrasonic sensor network [13]. the foot trajectory of real jogging is shown in fig. 1(b). when one of the user’s feet is at the front end of the trajectory and starts supporting the user’s weight, the other one has not yet reached the rear end of the trajectory but still on the way of its backward path. in fact, it could not begin to move forward until it reaches the rear end of the trajectory. as shown in fig. 1(b), the path length of the supporting travel b1 should be shorter than that of the striding travel b2. as a result, the conventional elliptical trainer cannot provide the user with the real jogging exercising mode and does not meet the principles of ergonomics. (a) conventional trajectory (b) real jogging trajectory fig. 1 foot trajectory quick return mechanisms can be seen in every corner of engineering industry in various machines such as shaper, stamping press, power-driven reciprocating saw and so on. several structures of the quick-return mechanisms can be found [14-15], including crank-shaper mechanisms, whitworth mechanism and offset crank-slider mechanism. quick-return mechanisms are usually used in machine tools for the intention of providing the reciprocating cutting tool a slow cutting stroke and a quick return stroke with a constant angular speed of the driving flywheel. the ratio of the time required for the cutting stroke to the time required for the return stroke is called the time ratio (tr) and is greater than unity [15]. the study of quick-return mechanisms has been investigated by quite a few researchers, and many important contributions have been accomplished. dwivedi [16] modified the whitworth quick-return mechanism to construct a high-velocity impacting press. the impacting press machine includes a whitworth quick-return mechanism comprising a crank and a drive arm together with a variable speed d.c. motor, and a flywheel. this study also analyzes the reasons of the unbalanced forces of this high velocity machine and manages to reduce the forces delivered to the foundations. fung et al. [17-18] derived the governing equation of a quick-return mechanism by using the finite element method (fem) with time-dependent length and hamilton's principle. hsieh and tsai [19] proposed a novel design for quick-return mechanism that combines a generalized oldham coupling and a slider-crank mechanism. the proposed quick-return mechanism is more compact and can be balanced easier than a conventional design. chen et al. [20] proposed an elliptical trainer that composes of a conventional elliptical trainer and a draglink mechanism to generate a quick-return effect in order to mimic the timing of the foot trajectory while jogging. but the design procedures are too complicated, it is difficult for the designers to adjust the dimension to meet the timing of the foot trajectory while jogging. advances in technology innovation, vol. 5, no. 3, 2020, pp. 190-201 192 the purpose of this study is to propose an innovative elliptical trainer that takes advantage of an inverted slider-crank mechanism to imitate the timing of the pedal trajectory and study the kinematics of the design. by using the inverted slider-crank mechanism, the design procedures are simple and easy to adjust the dimension to meet the requirement of time ratio in the proposed design. 2. innovative elliptical trainer (a) innovative design (b) detail drawing when right foot at a front end (c) detail drawing when right foot at a rear end fig. 2 innovative elliptical trainer the innovative elliptical trainer proposed in this study is illustrated in fig. 2. the design comprises a frame, two flywheels, two swing handles, two foot support members, two sliders and a timing adjustment wheel. a sliding branch is created on each of the flywheels. each of the foot support members connects to a respective flywheel. when the flywheel rotates about its pivot, the foot support member moves along a closed pedal trajectory c. the pedal trajectory c comprises a supporting travel c1 (from p1 to p3) and a striding travel c2 (from p3 to p1). the timing adjustment wheel rotates about a fixed pivot on the frame. advances in technology innovation, vol. 5, no. 3, 2020, pp. 190-201 193 a right slider sliding along the sliding branch on the right flywheel connects to the right side of the timing adjustment wheel. on the contrary, a left slider sliding along the sliding branch on the left flywheel connects to the left side of the timing adjustment wheel. the right and left sliders are connected to the same timing adjustment wheel only with 180° phase angle. the motion of the elliptical trainer is described as follows. fig. 2 shows that a user stands on the two foot support members, where the right foot support member is pedaled by the user’s right foot and located at a front end p1, and the left one is pedaled by the user’s left foot and located at a point p3. accordingly, when the foot support member pedaled by the user’s right foot is pedaled downward, the right flywheel with the sliding branch is driven to rotate about its fixed pivot. therefore, the timing adjustment wheel is driven to rotate according to the relative movement between the right slider and the sliding branch of the right flywheel. then, the timing adjustment wheel drives the left flywheel with the sliding branch to rotate according to the relative movement between the left slider and the sliding branch of the left flywheel so as to guide the left foot support member pedaled by the user’s left foot to move upward. furthermore, when the foot support member pedaled by the user’s right foot finishes the supporting travel c1 and reaches the point p3, the foot support member pedaled by the user’s left foot is guided to finish the striding travel c2 and reaches the front end p1. the quick-return effect is further explained as follows. when the timing adjustment wheel rotates 180° clockwise (𝑎1 + 𝛽1 = 180° in fig. 2(b) and 2(c)), the respective angle that the right flywheel rotates is less than 180° (𝑎2 + 𝛽2 = 180° in fig. 2(b) and 2(c)). therefore, the rotational speed of the timing adjustment wheel is faster than that of the right flywheel when the foot support member pedaled by the user’s right foot moves within the supporting travel c1. the foot support member pedaled by the user’s right foot is located at the point p3 rather than at the rear end p2 when the foot suppo rt members pedaled by the user’s left foot just reaches the front end p1 of the pedal trajectory c. as the user’s left foot reaches the supporting travel c1, the user’s right foot begins to lift backward and upward. as a result, the user can shift his weight from one leg to the other before both his two legs are extended to their respective extreme positions so that the user is protected from muscle sore and pain. the timing of pedal trajectory c generated by the proposed elliptical trainer is closer to that in the real jogging and meets the principles of ergonomics. 3. motion geometry of the quick-return effect the skeleton drawing of the proposed innovative elliptical trainer assembly is shown in fig. 3. from the vector loop diagram of the innovative elliptical trainer mechanism in fig. 4, the following two vector loop equations can be derived as: 2 4 1r r r 0   (1) 5 6 7 8 9r r r r r 0     (2) decomposing the vectors in eq. (2) into x and y scalar components, the equations become 5 5 6 6 7 7 8r cos r cos r cos r 0      (3) 5 5 6 6 7 7 9r sin r sin r sin r 0      (4) where 𝜃𝑖 is the position angle of vector �̅�𝑖and the angle is defined positive when measured counterclockwise. when the foot support member is at the front end p1 of the pedal trajectory in fig. 2, the position angles of flywheel and foot support link are the same (as shown in figs. 3 and 5(a)), i.e. 𝜃5 = 𝜃6. substituting this into eqs. (3) and (4) yields ( )  5 6 5 7 7 8r r cos r cos r  (5) ( ) = +5 6 5 7 7 9r r sin r sin r  (6) squaring eqs. (5) and (6) and adding them together yields advances in technology innovation, vol. 5, no. 3, 2020, pp. 190-201 194 7 7acos +b sin +c 0   (7) where 7 8a 2r r (8) 7 9b 2r r (9) c ( )     2 2 2 2 7 8 9 5 6r r r r r (10) by solving eq. (7), the position angle of the handle, 𝜃7, is obtained as: ( )        2 2 2 1 7 b a b c 2 tan c a  (11) from eqs. (5) and (6), the angle of the flywheel, 𝜃5, can be expressed as: 1 7 7 9 5 7 7 8 r sin r tan r cos r        (12) fig. 3 skeleton drawing of the innovative elliptical trainer assume that the angle between the flywheel and the sliding branch is 𝑎, as shown in figs. 3-5. when the foot supporting member is at the front end p1 of the pedal trajectory (position 𝑎0𝑎1𝑏0𝑐1𝑑1𝑑0) in fig. 3, the position angle of sliding branch, 𝜃4, can be found by 4 5    (13) by decomposing the vectors in eq. (1) into x and y scalar components and rearranging them yields 2 2 4 4 1r cos r cos r   (14) 2 2 4 4r sin r sin  (15) squaring eqs. (14) and (15) and adding them together yields ( )    2 2 2 4 1 4 4 1 2r 2r cos r r r 0 (16) advances in technology innovation, vol. 5, no. 3, 2020, pp. 190-201 195 fig. 4 vector loop diagram (a) right foot at a front end (b) right foot at a rear end fig. 5 detail drawing of inverted slider-crank mechanism by solving eq. (16), the length of vector, 𝑟4, is obtained as: ( ) ( )     2 2 2 4 1 4 1 4 1 2r r cos r cos r r  (17) from eqs. (14) and (15), the position angle of the timing adjustment wheel, 𝜃2, can be expressed as: 1 4 4 2 4 4 1 r sin tan r cos r       (18) similarly, when the foot support member is at the rear end p2 of the pedal trajectory (position 𝑎0𝑎2𝑏0𝑐2𝑑2𝑑0) in fig. 3, the difference between the position angles of flywheel and foot support link is 180° (as shown in figs. 3 and 5(b)), i.e. 𝜃6 = 𝜃5 − 𝜋. substituting this into eqs. (3) and (4) and yields ( )  5 6 5 7 7 8r r cos r cos r  (19) ( ) = +5 6 5 7 7 9r r sin r sin r  (20) squaring eqs. (19) and (20) and adding them together yields 7 7acos +b sin +c 0   (21) where advances in technology innovation, vol. 5, no. 3, 2020, pp. 190-201 196 7 8a 2r r (22) 7 9b 2r r (23) c ( )     2 2 2 2 7 8 9 5 6r r r r r (24) by solving eq. (21), the position angle of the handle, 𝜃7, is obtained as: ( )        2 2 2 1 7 b a b c 2 tan c a  (25) from eqs. (19) and (20), the angle of the flywheel, 𝜃5, can be expressed as: 1 7 7 9 5 7 7 8 r sin r tan r cos r        (26) as the foot support member is at the rear end p2 of the pedal trajectory in fig. 3, the position angle of the sliding branch, 𝜃4, can be obtained from eq. (13). since the position angle of 𝜃4 is known, the length of vector 𝑟4, 𝑟4, and the position angle of the timing adjustment wheel, 𝜃2, can be obtained from eqs. (17) and (18) as well. assume that ∅12 is the angular displacement of the flywheel as the pedal trajectory moves from the front end p1 to the rear end p2, and ∅21 is from the rear end p2 to the front end p1. the summation of these two angles is 360°, that is 12 21+ 2   (27) time ratio (tr) is the ratio of the time required for the slow stroke to the time required for the quick stroke. if the timing adjustment wheel runs at constant velocity, the time ratio tr can be expressed as: 12 21 tr    (28) assume the time ratio is known, then the rotation angles of the two strokes of the foot trajectory between two extreme positions can be expressed as: 21 2 1 tr     (29) 12 212    (30) as the pedal trajectory moves from the front end p1 to the rear end p2, the flywheel rotates with respect the fixed pivot and the rotation angle is approximately 180°. when the pedal position is at the front end p1, assume that the position angle of the sliding branch (link 4) is perpendicular downwards as shown in fig. 5(a). at this time, 𝜃41 denotes the position angle of the sliding branch and is equal to −90° while the position angle of timing adjustment wheel (link 2) is denoted as 𝜃21. when the pedal position is at the rear end p2, assume that the position angle of the sliding branch is perpendicular upwards as shown in fig. 5(b). at this time, 𝜃42 denotes the position angle of the sliding branch and is equal to 90° while the position angle of timing adjustment wheel is denoted as 𝜃22. from the geometry of the inverted slider-crank mechanism in fig. 5, the position angles of the timing adjustment wheel, 𝜃21 and 𝜃22, are symmetrical. therefore, 𝜃22 = −𝜃21 and the lengths of vector 𝑟4 at these two position are the same. furthermore, ∅21 can be expressed as: 21 22 21 21 22         (31) when the time ratio (tr) is known, the position angle of the timing adjustment wheel can be calculated from eqs. (29) and (31). advances in technology innovation, vol. 5, no. 3, 2020, pp. 190-201 197 furthermore, when the pedal position is at the front end p1 and the position of the sliding branch is downwards, i.e., 𝜃41 = −90°, the position angle of timing adjustment wheel 𝜃2 is equal to 𝜃21. it can be observed that there are three unknown variables (𝑟1, 𝑟2, 𝑟4) in eqs. (14) and (15). assume that the length of the fixed link, 𝑟1, is known, the length of the link 2, 𝑟2, can be obtained as: 1 2 21 r r cos  (32) as the pedal trajectory moves from the front end p1 to the rear end p2, the flywheel rotates with respect to the fixed pivot and the rotation angle of the flywheel is approximately 180°. according to the assumption above, the relationship between the stroke angles (∅12 and ∅21) and the length ratio (𝑟2/𝑟1) of the timing adjustment wheel to the fixed link is shown in fig. 6(a). fig. 6(a) shows the stroke angle ∅12 is larger than the stroke angle ∅21. the smaller the length ratio is, the larger the stroke angle ∅12 is and the smaller the stroke angle ∅21 is. however, when the length ratio of the timing adjustment wheel to the fixed link is increased, the difference between the stroke angles ∅12 and ∅21 is reduced. the relationship between the time ratio and the length ratio is shown in fig. 6(b). from fig. 6(b), the smaller the length ratio of the timing adjustment wheel to the fixed link is used, the larger the time ratio becomes. on the contrary, the larger the length ratio is used, the smaller the time r atio becomes. by using fig. 6(b), it can be used to select proper length ratio to generate desired time ratio. (a) stroke angles (b) time ratio fig. 6 stroke angles and time ratio 4. kinematic analysis when the dimensions of the elliptical trainer mechanism and the position angle of the timing adjusting wheel 𝜃2 are known, the complete kinematic analysis and pedal trajectory can be derived and calculated. from eqs. (14) and (15), the position angle of the sliding branch, 𝜃4, can be expressed as: 1 2 2 4 2 2 1 r sin tan r cos r       (33) by expressing the eq. (2) into x and y scalar component equations and rearranging, the equations become 6 6 7 7 8 5 5r cos r cos r r cos     (34) 6 6 7 7 9 5 5r sin r sin r r sin     (35) where 𝜃5 = 𝜃4 + α. squaring eqs. (34) and (35) and adding them together yields advances in technology innovation, vol. 5, no. 3, 2020, pp. 190-201 198 7 7acos +b sin +c 0   (36) where a ( ) 7 8 5 52r r r cos (37) b ( ) 7 9 5 52r r r sin (38) 2 2 2 2 2 5 6 7 8 9 5 8 5 5 9 5c r r r r r 2r r cos 2r r cos        (39) by solving eq. (36), the position angle of the handle, 𝜃7, is obtained as: ( )        2 2 2 1 7 b a b c 2 tan c a  (40) from eqs. (34) and (35), the angle of foot support link, 𝜃6, can be found by 1 7 7 9 5 5 6 7 7 8 5 5 r sin r r sin tan r cos r r cos            (41) differentiating eqs. (14) and (15) with respect to time yields velocity equations as: ( ) ( ) ( )  2 2 2 4 4 4 4 4r sin r sin cos r     (42) ( ) ( ) ( )  2 2 2 4 4 4 4 4r cos r cos sin r     (43) when the velocity of timing adjusting wheel �̇�4 is known. by using gauss-jordan method to solve eqs. (42) and (43), �̇�4 and �̇�4 can be obtained as: ( )  2 2 4 4 2 4 r cos r     (44) ( )  4 2 2 4 2r r sin    (45) differentiating eqs. (42) and (43) with respect to time yields velocity equations as: ( ) ( ) ( )  6 6 6 7 7 7 5 5 5r sin r sin r sin      (46) ( ) ( ) ( )   6 6 6 7 7 7 5 5 5r cos r cos r cos      (47) where �̇�5 = �̇�4. by using gauss-jordan method to solve eqs. (46) and (47), �̇�6 and �̇�7 can be obtained as: ( ) ( )    5 5 7 4 5 6 7 6 r sin r sin       (48) ( ) ( )    5 5 6 7 5 6 7 6 r sin r sin       (49) therefore, the pedal position (𝑥𝑝, 𝑦𝑝 ) with respect to the fixed pivot of the flywheel can be expressed as: p 5 5 6 p 6x r cos r cos   (50) p 5 5 6 p 6y r sin r sin   (51) differentiating eqs. (50) and (51) with respect to time yields the pedal velocity as: advances in technology innovation, vol. 5, no. 3, 2020, pp. 190-201 199 p 5 5 5 6 p 6 6x r sin r sin      (52) p 5 5 5 6 p 6 6y r cos r cos     (53) 5. design example and discussion to explain the design procedure of the proposed innovative mechanism, a design example is presented here for illustration. referring to a conventional design, assume that the parameters 𝑟5, 𝑟6, 𝑟7, 𝑟8 and 9r are 28.0 cm, 134.0 cm, 93.0 cm, 143.0 cm and 82.0 cm respectively. using eqs. (11) and (12), the angles 𝜃7 and 𝜃5 are calculated as 𝜃7 = −78.37°and 𝜃5 = −3.22° respectively. assume that the position angle of the sliding branch is 𝜃4 = 𝜃41 − 90° when the foot support member is at the front end of the pedal trajectory. the angle between the flywheel and the sliding branch can be calculated from eq. (13) as α = 𝜃5 − 𝜃4 = −86.78°. assume that the time ratio (tr) is 2.0, the angles ∅21 = 120°, ∅12 = 240°and 𝜃22 = −𝜃21 = 60° can be obtained from eqs. (29)-(31). finally, assume that fixed length 𝑟1is 13.0 cm, then the length of link 2 can be computed from eq. (32) as 𝑟2 = 26.0 cm. by using the vector-loop method and the kinematic analysis, motion simulation of pedal trajectory is carried out to validate the feasibility of the proposed innovative mechanism. the pedal trajectories of a conventional and the proposed design are shown in fig. 7. the pedal trajectories are drawn and dotted whenever the timing adjustment wheel is angular displaced with 10° increment. in fig. 7(a), for the conventional elliptical trainer, the distance between any adjacent points near the front and rear ends of the pedal trajectory is shorter and that in the middle of the supporting and striding travel is longer. in fig. 7(b), for the proposed design, the distance between any adjacent points near the bottom of pedal trajectory is shorter and that near the top is longer. if the timing adjustment wheel is operated at a constant speed, the speed of the foot near the front and rear ends is slower and that in the middle of the supporting and striding travels is faster for the conventional elliptical trainer. however, the speed of the foot at the bottom of the pedal trajectory is slower and that at the top is quicker for the proposed design. (a) conventional design (b) proposed design fig. 7 pedal trajectory fig. 8 pedal velocity advances in technology innovation, vol. 5, no. 3, 2020, pp. 190-201 200 assume that the timing adjustment wheel runs at 60.0 rpm, the velocity of the pedal on the foot support member is shown in fig. 8. when the timing adjustment wheel is at about 60° and 250° of the conventional design, i.e. near the front and rear ends of trajectory, the pedal velocity is about 65.0 cm/sec. one the contrary, when the flywheel is at about 150° and 340°, i.e. in the middle of the supporting and striding travels, the pedal velocity is about 175.0 cm/sec. the pedal velocity of the conventional design is not same as the real jogging. however, when the timing adjustment rotates from 60° to 300° in the proposed design, i.e. near the bottom part of pedal trajectory, the pedal velocity changes between 65.0~120.0 cm/sec. when the flywheel is between 300° and 60°, i.e. near the top part of pedal trajectory, the pedal velocity varies between 65.0~350.0 cm/sec. the maximum speed during the striding travel is 3 times than that during the supporting travel. therefore, the average velocity of pedal trajectory on the supporting travel is slower than that on the striding travel. by using the quick-return effect of the inverted slider-crank mechanism, the timing of the pedal trajectory in this innovative design is more similar to the one in real jogging and meets the principles of ergonomics. when using this proposed innovative elliptical trainer design, the user can shift his body weight from one leg to the other before both of his legs extend to their extreme positions. compared to the conventional design, such a motion pattern proposed in this study can protect the user from suffering muscle sore and pain. by using a 3d cad design software, solidworks, the solid model of this innovative design is constructed and shown in fig. 9. for the sake of showing the structure of the innovative design, the flywheel is drawn as a crank in fig. 9. the cad prototype can be used for assembly, simulation and fabrication in order to confirm its feasibility and expedite the commercialization. (a) top view (b) isometric view fig. 9 cad solid model of the innovative elliptical trainer 6. conclusions the purpose of this paper is to propose and study an innovative elliptical trainer that imitates the timing of the foot trajectory during jogging. the procedures for the design of elliptical trainer with required time-ratio are illustrated. the pedal trajectory and the kinematics of the innovative elliptical trainer are analyzed and simulated. the results of the simulation confirm that the velocity of pedal trajectory on the supporting travel in the proposed design is slower tha n that on the striding travel. therefore, the proposed innovative design that mimics the timing of the foot trajectory can satisfy the design requirements and fit the principles of ergonomics for the joggers. based on the results of this investigation, the foot trajectory and the workload of this design can be further modified by adjusting the dimension, the ramp setting and the resistance of the motion. as for the dynamics of the innovative elliptical trainer and its test data under iso standard [21], they will be studied, collected and optimized in the future when the prototype is constructed. conflicts of interest the authors declare no conflict of interest. advances in technology innovation, vol. 5, no. 3, 2020, pp. 190-201 201 references [1] j. h. hoeberigs, “factors related to the incidence of running injuries,” sports medicine, vol.13, no. 6, pp. 408-422, november 1992. [2] g. a. hanks and a. kalenak, running injuries. in: grana wa, kalenak a, editors. clinical sports medicine. philadelphia: w.b. saunders company, pp. 458-465, 1982. [3] a. f. renstrom, “mechanism, diagnosis, and treatment of running injuries,” instructional course lectures, vol. 42, pp. 225-234, 1993. [4] t. f. novacheck, “the biomechanics of running,” gait and posture, vol. 7, no. 1, pp. 77-95, january 1998. [5] t. w. lu, h. l. chien, and h. l. chen, “joint loading in the lower extremities during elliptical exercise,” medicine and science in sports and exercise, vol. 39, no. 9, pp.1651-1658, september 2007. [6] t. w. lu and h. l. chien, “dynamic analysis of the lower extremities during elliptical exercise,” journal of biomechanics, vol. 39, no. 1, p. s193, 2006. [7] j. p. porcari, j. m. zedaker, l. naser, and m. miller, “evaluation of an elliptical exerciser in comparison to treadmill walking and running, stationary cycling, and stepping,” medicine & science in sports & exercise, vol. 30, no .5, p. 168, 1998. [8] j. h. shyu, c. k. chen, c. c. yu, and y. j. luo, “research and development of an adjustable elliptical exerciser,” advanced materials research, vols. 308-310, pp. 2078-2083, august 2011. [9] k. m. knutzen, w. l. mclaughlin, a. j. lawson, b. s. row, and l. t. martin, “influence of ramp position on joint biomechanics during elliptical trainer exercise,” the open sports sciences journal, vol. 3, pp. 165-177, 2010. [10] c. a. nelson and j. m. burnfield, “improved elliptical trainer biomechanics using a modified cardan gear,” asme 2012 international design engineering technical conferences and computers and information in engineering conference, vol. 4: 36th mechanisms and robotics conference, august 2012, pp. 35-42. [11] z. yin, “modal analysis and dynamics simulation of elliptical trainer,” international conference on mechatronics and intelligent robotics, march 2020, pp. 19-25. [12] c. krishnan, e. p. washabaugh, and y. seetharaman, “a low cost real-time motion tracking approach using webcam technology,” journal of biomechanics, vol. 48, no. 3, pp. 544-548, february 2015. [13] y. qi, c. b. soh, e. gunawan, and k. s. low, “ambulatory measurement of three-dimensional foot displacement during treadmill walking using wearable wireless ultrasonic sensor network,” ieee journal of biomedical and health informatics, vol. 19, no. 2, pp. 446-452, march 2015. [14] g. h. martin, kinematics and dynamics of machines, 2nd ed. usa: waveland press, 1982. [15] h. s. yan, mechanisms: theory and applications, singapore: mcgraw-hill, 2016. [16] s. n. dwivedi, “application of a whitworth quick return mechanism for high velocity impacting press,” mechanism and machine theory, vol. 19, no.1, pp. 51-59, 1984. [17] r. f. fung and k. w. chen, “constant speed control of the quick return mechanism driven by a dc motor,” jsme international journal series c mechanical systems, machine elements and manufacturing, vol. 40, no. 3, pp. 454-461, 1997. [18] r. f. fung and f. y. lee, “dynamic analysis of the flexible rod of a quick-return mechanism with time-dependent coefficients by the finite element method,” journal of sound and vibration, vol. 202, no. 2, pp. 187-201, may 1997. [19] w. h. hsieh and c. h. tsai, “a study on a novel quick return mechanism,” transactions of the canadian society for mechanical engineering, vol. 33, no. 3, pp. 487-500, march 2018. [20] f. c. chen, y. f. tzeng, and w. r. chen, “on the mechanism design of an innovative elliptical exerciser with quick-return effect,” international journal of engineering and technology innovation, vol. 8, no. 3, pp.228-239, july 2018. [21] stationary training equipment part 9: elliptical trainers, additional specific safety requirements and test methods, iso 20957-9, pp. 1-14, 2016. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). https://www.scientific.net/author/jenq_huey_shyu_1 https://www.scientific.net/author/ching_kong_chen https://ieeexplore.ieee.org/search/searchresult.jsp?searchwithin=%22first%20name%22:%22yongbin%22&searchwithin=%22last%20name%22:%22qi%22&newsearch=true&sorttype=newest https://ieeexplore.ieee.org/search/searchresult.jsp?searchwithin=%22first%20name%22:%22cheong%20boon%22&searchwithin=%22last%20name%22:%22soh%22&newsearch=true&sorttype=newest https://ieeexplore.ieee.org/search/searchresult.jsp?searchwithin=%22first%20name%22:%22erry%22&searchwithin=%22last%20name%22:%22gunawan%22&newsearch=true&sorttype=newest https://ieeexplore.ieee.org/search/searchresult.jsp?searchwithin=%22first%20name%22:%22kay-soon%22&searchwithin=%22last%20name%22:%22low%22&newsearch=true&sorttype=newest  advances in technology innovation, vol. 4, no. 4, 2019, pp. 211-221 an intelligent optimization system of micro electroforming process for the mesh filter wen-chin chen 1,* , pao-wen lin 2 , shi-bo lin 1 , kuan-ming lin 3 , yun-ru chang 4 1 department of industrial management, chung hua university, hsinchu, taiwan 2 ph.d. program of technology management, chung hua university, hsinchu, taiwan 3 department of foreign languages and literatures, chung hua university, hsinchu, taiwan 4 languages center, chung hua university, hsinchu, taiwan received 12 april 2019; received in revised form 19 may 2019; accepted 17 july 2019 abstract this research integrates the taguchi method, analysis of variables (anova), back-propagation neural networks (bpnn), and hybrid pso-ga to develop an intelligent optimization system of micro electroforming process for the mesh filter. from the outset of discussions with engineers in terms of past related literature survey of the micro electroforming process, the quality characteristics of product and control variables can be well ascertained, then transforming the problem of multiple quality characteristics into a single quality characteristic via the taguchi method and anova. however, the optimal parameter settings (solution) through the taguchi experimental planning is still belong to a discrete optimal solution which is impossible to meet the process stability and quality goals. therefore, this study first identifies the initial weight of bpnn，using hybrid pso-ga with multilayer perceptron (mlp)，in order to improve training efficiency and precision of bpnn. furthermore, the study constructs the signal-to-noise (s/n) ratios (bpnns/n) and quality predictors (bpnnq) based on hybrid pso-ga and bpnn with the experimental data. the optimal parameter settings are obtained through a combination of bpnns/n and bpnnq with modified pso-ga. finally, confirmation experiments are performed to assess the effectiveness of the proposed system. the results show that the proposed system can create the best performance, and the optimal parameters not only enhance the stability in the micro electro forming process but also effectively improve the product quality. keywords: taguchi method, bpnn, pso-ga, micro electroforming, mlp 1. introduction recently the technology of micro electroforming process has been widely used, and the mesh process can be mainly divided into the photolithography process and the micro electroforming process, having been widely used recently. and the process parameter control directly affects product quality and cost. the photolithography process consists of three main components: coating photoresist, exposure, and development. in order to obtain higher resolution, some baking and cooling steps are also adopted in the photolithography process. in the current technology, of photolithography process, entirely seven steps are required in the process: cleaning the substrate, pre-baking, coating the photoresist, soft-baking, exposure, development, and hard-baking. when the photolithography process parameters are not well controlled, defects such as the * corresponding author. e-mail address: wenchin@chu.edu.tw tel.: +886-3-5186585; fax: +886-3-5186575 advances in technology innovation, vol. 4, no. 4, 2019, pp. 211-221 212 excessive image size variation, poor transfer rate, and even transfer failure may occur; therefore, the photoresist must be stripped and the previous process repeated until the inspection is completed. then, the semi-finished mold core formed by photolithography is subjected to a micro-electroforming process. the micro electroforming process consists of five main components: electroforming, photoresist stripping, finished stripping, cleaning and hard baking. as to the current technology of electrochemical micro-electroforming process for making molds, when the process parameters are not well controlled, problems such as the product forming failure and excessive size variation will be directly caused and hence the loss due to the fact that the product will not pass the quality inspection and cannot be reworked. therefore, to improve yield and reduce cost, the parameter setting of the micro electroforming process control factors are even more important. since the micro electroforming process can be applied to a variety of materials, and there are many types and formulations of chemical electroforming fluids, many scholars are devoted to studying the interaction between various chemical electroforming fluids and materials and the related physical phenomena in the process [1-8]. however, through the analysis of the materials, a suitable combination of electroforming liquid and materials, a better process and product quality, and the better process parameters combination all can be obtained. if an inappropriate combination of process parameters is used, it can lead to product defects and excessive process variations. in the past, some scholars adopted the taguchi experimental design method to explore the correlation between process parameters and quality [9-11].however, the taguchi experimental design is a discrete method for solving single quality characteristics, and only the local optimal solution of the pre-selected parameter level can be obtained specific to a single quality characteristic, but not the global continuous optimal solution. therefore, it is necessary to combine the experimental design, smart predictors and applications of related theories for optimization to search for the best combination of process parameters by numerical simulation and prediction[12-14].the above studies only focused on optimizing the process parameters for product quality characteristics, but they did not assess the stability of the process in the micro electroforming process. therefore, this study proposes an intelligent optimization system to find optimal process parameters of multiple quality characteristics in the micro electroforming process. firstly, the taguchi method is used to determine the best combination of parameter settings by calculating the signal-to-noise (s/n) ratio from the experimental data. the highest s/n ratio value is employed to decide the best settings for quality responses. significant factors are determined through analysis of variance (anova). the s/n ratio predictor (bpnnsn) and quality predictor (bpnnq) are constructed by bpnn. in the first stage optimization, bpnns/n is coupled with ga in order to minimize the variations of the process. in the second stage optimization, the optimal parameter settings are obtained via a combination of bpnns/n, bpnnq and hybrid ga-pso. finally, two confirmation experiments are conducted to assess the effectiveness of the proposed intelligent optimization system. this study focuses on not only the optimal process parameters to improve the multiple qualities, but also the stability of the process to enhance the productivity. the research has been motivated by the current development of ai, big data, internet of things (iot) and cloud computing worldwide in general, which especially play their important roles in the future industrial automation systems. 2. research methodologies in this study, an intelligent optimization system is proposed for the micro electroforming process of the mesh filter. the research integrates the taguchi method, anova, bpnn, the improved hybrid pso-ga, statistical process control and other related technologies to obtain the optimization for multi-objective micro-electroforming process. and it has enabled product quality to be maintained within acceptable quality ranges and made the micro electroforming process more stable. firstly, based on the literature reviews and discussions with engineers on the influence of process parameters to their quality characteristics, the control parameters and level are selected for the taguchi orthogonal table experiment. then the electroforming experiment is proceeded in the customized electroforming tank for the research. the micro electroforming workpiece is the mesh filter. in order to ensure that the product can be supplied to the manufacturer's required size advances in technology innovation, vol. 4, no. 4, 2019, pp. 211-221 213 specifications, a discussion with the manufacturer is proceeded, obtaining a product diameter of 555±3μm and a deviation of quantity specification within 5% of the product thickness. therefore, the quality characteristics are set to the diameter roundness and thickness uniformity. after literature reviews and discussions with a number of experts and engineers, the control factors selected for the experiments are the temperature of the electroforming liquid, current density, cathode size, the distance between the anode and the cathode, and oscillating rate. to acquire more accurate data, the study has to consider the lesser number of experimental control factors and standards, so the taguchi l18 (2 1 x 3 4 ) orthogonal table is used for the experiment, in which the temperature of the casting liquid is set at two levels and the remaining control factors are set at three levels. however, the taguchi experimental design is a discrete method for solving single quality characteristics, and only the local optimal solution of the pre-selected parameter level can be obtained specific to a single quality characteristic, but not the global continuous optimal solution. in addition, since the combination of process parameters obtained from the taguchi experimental planning cannot meet both the stability of the micro electroforming process and the best quality of the product, a two-stage optimization must be carried out. the first stage is to obtain the measurement data of the diameter and thickness of the mesh filter by experiments. next, for the taguchi data analysis, the problem of multi-objective quality characteristics needs to be transformed into a single quality characteristic, so that the based-on data can be further analyzed accordingly. the data analysis includes factor response graph analysis and variance analysis, and the relationships between the s/n ratios and the quality characteristics of the experimental control factors can be known. using the s/n ratio factor reaction map, we can find important control factors that have a significant impact on the quality characteristics and classify the control factors to optimize the two steps of the taguchi process. firstly, the step uses the first type factor to modulate the s/n ratios to the maximum value for the purpose of reducing the process variation. secondly, the step adjusts the second type of factor level to approximate the average value of the quality characteristic to the target value. finally, the third type of factor is used to reduce production costs. based on the steps, a set of optimal process parameters of taguchi can be obtained, which can be used as the initial value of the optimization of the second stage. through the factor response analysis and the analysis of the variance, the significant control factors found will be adjusted as the basis of the subsequent solution parameters. in the second stage, the parameter combination obtained by the taguchi experimental planning is used as the basis to establish the s/n ratio predictor (bpnns/n) and quality predictor (bpnnq). however, the initial weight value of the bpnn is often generated in a random manner, and the initial weight value affects the network training speed and prediction accuracy; therefore, this study uses the improved hybrid pso-ga combined with multilayer perceptron (mlp) to obtain and preserve the initial weight required for bpnn. this method not only improves the training speed of bpnn, but also increase the predictive power of it. in this stage, the process parameter combination obtained in the first stage is used as the initial value, and the s/n ratio predictor and the quality predictor are combined with the hybrid pso-ga for global search to find the process parameters that best meet the quality specifications and the most stable quality. the mesh microstructure diameter size target is 555 μm, and the acceptable thickness deviation is within 5%. for the diameter roundness, the measurement method is divided into twenty areas, as shown in fig. 1. the measurement of five points in each area is averaged, thus the measurement can better determine the exact roundness of diameter. formula for roundness is as shown in eq. 1.for the thickness uniformity, the measurement method is divided into 20 points for the inner and outer regions, as shown in fig. 2, taking the percentage of thickness deviation inside and outside. 20 5 1 1 1 d d 20 5 ij i j      (1) dij is the measured value of the mesh microstructure diameter; d is the diameter roundness of the mesh quality characteristics; measuring area has 20 measuring points in five zones; and j is the number of measuring points. the main advances in technology innovation, vol. 4, no. 4, 2019, pp. 211-221 214 purpose of this study is to find out the optimal process parameter combination of the micro electroformed mesh, so that its quality is within the desired range, and the product tends to stabilize and reduces the non-performing rate. the thickness uniformity is shown as eq. 2. t = 1 𝑇𝑀𝑎𝑥 − 𝑇𝑀𝑖𝑛 ∙ 1 4 × 5 × ∑ |∑ 𝑇𝑂𝑖𝑗 − ∑ 𝑇𝐼𝑖𝑗 5 𝑘=1 5 𝑗=1 | 4 𝑖=1 (2) t is the thickness uniformity of the mesh quality characteristics as shown in equation 2. the maximum thickness measurement value is 𝑇𝑀𝑎𝑥 =max (𝑇𝑂𝑖𝑗 , 𝑇𝐼𝑖𝑗 ); the minimum value is 𝑇𝑀𝑖𝑛 =min ( 𝑇𝑂𝑖𝑗 , 𝑇𝐼𝑖𝑗 ); 𝑇𝑂𝑖𝑗 is the thickness measurement of the outer ring; 𝑇𝐼𝑖𝑗 is the thickness measurement of the inner ring; i is the measuring area; and j is the number of measurement points. fig. 1 schematic diagram of mesh diameter measurement fig. 2 schematic diagram of mesh thickness 3. results and discussion 3.1. taguchi experiment in this study, the micro electroforming product is the mesh filter. in order to ensure that the product can be supplied to the manufacturer's required size specifications, a discussion with the manufacturer is proceeded, obtaining a product diameter of 555±3μm and a deviation of quantity specification within 5% of the product thickness. the quality characteristics are diameter roundness and thickness deviation. in addition, the experimental control factors are defined as following five experimental control factors: temperature of the electroforming liquid (tl) (℃), current density (cd) (a/ dm 2 ), cathode size (cs) (dm 2 ), the distance between the anode and the cathode (dac) (cm), and oscillating rate (or) (rate/min). the range of adjustment parameters and the control factor level settings are shown in table 1. the five experimental control factors in this study uses l18(2 1 x 3 4 ) orthogonal table. as shown in table 2, the micro electroforming is performed on a customized experimental machine, and the data of the diameter roundness and the thickness deviation are obtained by measurement, and the s/n ratios are calculated. the diameter roundness adopts the first type formula of the eyesight characteristic, and the thickness deviation adopts the small characteristic formula. 18 groups from no.1 to no.18 are the taguchi experimental data; 5 groups from no. 19 to no. 23 are randomly generated. for the quality characteristic of diameter, the equation of type i formula of the nominal-the-best is used as shown in eq. 3. as for a deviation of quantity specification within 5% of the product thickness, the smaller-the-better is used as shown in eq. 4. the experimental product is shown in fig. 3. ])log[(10 )( log10/ 221 2 smy n my ns n i i      (3) )log(10log10/ 221 2 sy n y ns n i i    (4) where yi is the response value of a specific treatment under i replications, n is the number of replications, y is the average of all yi values, and sis the standard deviation of all yi values. advances in technology innovation, vol. 4, no. 4, 2019, pp. 211-221 215 table 1 showing different crystal growth methods, growth time and approximate sizes of the grown crystal experimental control factors range level 1 level 2 level 3 tl 40-50 40 50 cd 1-5 1.00 3.00 5.00 cs 1-4 1.00 2.25 4.00 dac 9-12 9.0 10.5 12.0 or 20-52 20 36 52 table 2 the results of the diameter and thickness deviation of the taguchi experiment no average diameter (x) (μm) thickness deviation (y) (%) σ(x) σ(y) s/n ratiofor x s/n ratio for y 1 554.310 0.0237 0.1114 0.0018 3.1114 32.4771 2 554.447 0.0303 0.2542 0.0017 4.3085 30.3453 3 555.780 0.0271 0.1100 0.0020 2.0726 31.3317 4 554.330 0.0139 0.2128 0.0016 3.0610 37.0848 5 554.837 0.0315 0.1332 0.0015 13.5251 30.0174 6 555.117 0.0295 0.0907 0.0020 16.6066 30.5767 7 553.903 0.0106 0.1570 0.0022 -0.8895 39.2921 8 555.467 0.0326 0.1986 0.0025 5.8971 29.7218 9 555.347 0.0292 0.1986 0.0019 7.9694 30.6632 10 554.257 0.0171 0.2194 0.0012 2.2136 35.3380 11 554.690 0.0308 0.3568 0.0019 6.5092 30.2133 12 555.700 0.0303 0.0954 0.0024 3.0181 30.3316 13 554.380 0.0081 0.1153 0.0018 4.0044 41.6647 14 554.807 0.0391 0.1069 0.0012 13.1148 28.1516 15 555.067 0.0303 0.1097 0.0024 17.8310 30.3329 16 553.923 0.0072 0.1222 0.0014 -0.6972 42.6330 17 554.577 0.0321 0.1914 0.0023 6.6586 29.8436 18 554.833 0.0367 0.2775 0.0024 9.7959 28.6889 19* 554.767 0.0282 0.1168 0.0016 11.6699 30.9735 20* 555.550 0.0219 0.2117 0.0012 4.5930 33.1608 21* 555.107 0.0408 0.1570 0.0020 14.4356 27.7712 22* 554.740 0.0359 0.1153 0.0024 10.9205 28.8752 23* 553.997 0.0074 0.2084 0.0019 -0.2124 42.3435 fig. 3 the experimental product 3.2. variation of ph value in the water storage tank this study aims to find an optimal combination of process parameters that meet the multi-objective quality. however, taguchi’s experimental design belongs to a single quality response and discrete optimization method, and only can derive local advances in technology innovation, vol. 4, no. 4, 2019, pp. 211-221 216 optimal solution of pre-selected parameter level value for a single quality characteristic. therefore, it is necessary to first transform and integrate the problem of multiple quality objectives of diameter and thickness deviation into a problem of single quality objective, and then to conduct subsequent data analysis based on the experimental data. data analysis includes factor response analysis and anova in order to understand the relationships between experimental control factor pairs of s/n ratio and quality characteristics. s/n ratio response chart can identify important control factors with more significant effects on s/n ratios, while the use of quality characteristics response chart can screen control factors with more significant effects on quality characteristics. the taguchi optimization process parameter combination can be used as the initial value for subsequent optimization, while the significant control factors found by factor response analysis and anova are chosen as the basis for adjusting the process parameters for subsequent optimization. the method is to integrate the individual offsets of the quality characteristics of the diameter and thickness deviation into a total offset to achieve a single target quality. the calculation of the integrated diameter and thickness deviation into a single target quality is as follows [15]: (1) the calculation of the total offset taking the taguchi experimental product as an example, the diameter size of the experimental product is measured as x, and the target value of the quality characteristic is 555μm, so the diameter offset is x-555μm; the thickness deviation measurement is y, and the target value of the quality characteristic is 0%, so the measured value is the thickness deviation offset. thus, the same measurement unit (x-555) and y are assumed to be two vectors, so the vector sum z is its total offset. the calculation is as follows: total offset z = ((x 555) 2 +y 2 ) 1/2 (5) (2) the calculation of the total standard deviation in the taguchi experiment, if the standard deviation of a certain group of diameter is σ(x), and the standard deviation of thickness is σ(y), then the total standard deviation σ(x+y) is as follows 2 2 (x y) (x) (y) 2 cov(x, y)       (6) (3) conduct factors response analysis via anova and main effects plot the s/n ratio response factor table of integrating into a single quality and the main effects plot for s/n ratios of total bias are demonstrated in the study. the factor response chart shows that a set of taguchi optimal process parameter combination can be obtained to meet the multiple quality characteristics. this study will denote the minimum process variation and optimal quality characteristic, and the optimal parameter combination of taguchi experiment and anova. table 3 anova of the total offset quality characteristics df seq ss adj ss adj ms f p contribution significant tl 1 0.00999 0.00999 0.00999 0.34 0.578 0.64% no cd 2 0.39617 0.39617 0.19808 6.67 0.020 25.40% yes cs 2 0.78652 0.78652 0.39326 13.24 0.003 50.42% yes dac 2 0.07188 0.07188 0.03594 1.21 0.347 4.61% no or 2 0.05761 0.05761 0.02880 0.97 0.420 3.69% no error 8 0.23764 0.23764 0.02971 15.24% total 17 1.55980 s = 0.172352, r-sq = 84.76%, r-sq(adj) = 67.62% this integrated diameter and thickness deviation offset becomes the total offset, which is a quality characteristic of single target quality. since the smaller the value is, the better the result will be, the s/n ratio chooses the smaller-the-better formula. for the anova of the total offset of quality characteristics, the significant influence factors are selected according to the advances in technology innovation, vol. 4, no. 4, 2019, pp. 211-221 217 contribution. first, the two control factors of cathode size (50.42%) and current density (25.40%) are spotted, as shown in table 3. the above two control factors have a significant influence on the quality characteristics and can provide the basis for subsequent optimization of the adjustment process parameters. the optimal parameter combination of the taguchi method is shown in table 4. table 4 optimal parameter combination optimal parameter tl cd cs dac or 50 3.00 4.00 9 20 3.3. using mlp combined with hybrid pso-ga to find the initial weight of bpnn as a consequence of using the taguchi experimental analysis, the optimal combination of parameters obtained is the discrete parameter combination established by the original taguchi experimental design, and the quality may not reach the target value. therefore, the research uses the backpropagation neural network (bpnn) to construct the s/n ratio predictor and quality predictor, combining the improved hybrid pso-ga optimized in this study. moreover, in terms of the taguchi experimental analysis, the optimized combination of parameters is used as the initial value of the algorithm search. furthermore, it is hoped to find a set of continuous type of best combination of parameters that can achieve process stability and quality objectives. however, when the traditional bpnn is in sample training, the initial weight is often generated randomly, and it will affect the training speed and accuracy of the neural network. therefore, this study proposes to use the improved pso-ga combined with mlp to solve the initial weight value of bpnn and to find a better set of adaptation as the initial weight value of bpnn. the study uses bpnn to build the s/n ratio predictor and quality predictor, using the improved pso-ga combined with mlp to find the better initial weight for the s/n ratio predictor and quality predictor respectively.the objective function is defined in eq. 7. min 𝑓(w) = 1 36 ∑ ∑(𝑑𝑘 𝑝 − 𝑦𝑘 𝑝 (w))2 2 𝑘=1 18 𝑝=1 (7) where p represents number of pth sample, k represents kth quality characteristic, 𝑑𝑘 𝑝 represents the target value of bpnn indexed by k , 𝑦𝑘 𝑝 (𝑊) represents the output value of mlp indexed by k, w is mlp’s weights. through using mlp and eq. 7 with the modified pso-ga to solve optimization weights, better weight values can often be found. 3.4. establishing s/n ratio predictor and quality predictor the study uses bpnn to establish the s/n ratio predictor and quality predictor. the input values of the s/n ratio predictor and quality predictor are the normalized values of the 18 groups of parameters of the taguchi experiment, and the output value of the predictor is the normalized value of the s/n ratio and the average value of the quality characteristics in the taguchi experiment. in addition, the 19th to 23rd groups in the taguchi experiment are used as the testing data of bpnn. in order to enable bpnn to training convergence, this study adopts the weighting solution of mlp-pso-ga, stated in the previous section, as the initial value of bpnn for training. the s/n ratio predictor training uses 1053 generations, training rmse to be 0.0004940, and testing rmse to be 0.0259. the quality predictor takes 968 generations; the training rmse is 0.00082261, and the test rmse is 0.0253. the error of the predicted value of the two predictors is analyzed within the acceptable range by comparing the error between the predictors and the actual values. 3.5. two-stage process parameter optimization in the first stage, the experiment focuses on maximizing the s/n ratio. the constructed s/n ratio predictors are combined with a ga to identify the process parameter combination with the minimum variance and the most robust process so that the s/n ratio values of diameter roundness (mm) and thickness deviation must be maximized. the taguchi optimal parameter advances in technology innovation, vol. 4, no. 4, 2019, pp. 211-221 218 combination is used as the initial value to carry out the full range global search for the six control factors. the fitness function of ga is presented as follows: max snod, snot s.t. 40≤ x1 ≤ 50, 1≤ x2 ≤ 5, 1≤ x3 ≤ 4, 9≤ x4 ≤ 12, 20≤ x5 ≤ 52, (8) where x(x1, x2, x3, x4, x5) is the process parameter(control factor), snod is the s/n ratio of diameter predicted by de-normalized bpnns/n, and snot is the s/n ratio of thickness deviation predicted by de-normalized bpnns/n. five control factors are temperature of the electroforming liquid (x1), current density (x2), cathode size (x3), the distance between the anode and the cathode (x4), and oscillating rate (x5). this numerical analysis is to conduct the global search for all control factors and obtained the process parameter combination of the first stage multi-objective s/n ratio maximization. the optimal parameter combinations are: x1=49.582, x2=2.152, x3=3.545, x4=9.004, and x5=51.912. min f2 (x) = (yod－555) 2 max snod, snot s.t. yot ≤ 0.01 8≤ x2 ≤ 28, 62≤ x3 ≤ 78 (9) where x(x2,x3)is the process parameter (control factor), yod is the output value of diameter quality predictor after de-normalization, and yot is the output value of thickness deviation quality predictor after de-normalization, and 555 is the target value of diameter quality characteristic, and 0.01 is the target value of thickness deviation quality characteristic as smaller as possible. the main control factors are x2is current density and x3 is cathode size. by conducting a global search for the two significant control factors of the second stage, and combining bpnns/n and bpnnq with modified pso-ga, this study can obtain the process parameter combination meeting the multi-objective quality and minimizing variation. the optimal process parameter combination is shown in table 5. table 5 optimal parameter combination (two-stage process) optimal parameter tl cd cs dac or 49.582 2.002 3.85 9.004 51.912 3.6 confirmation of experiment and discussion due to the accuracy set by the operating machine, the optimized parameter values must be rounded up according to the limits set by the machine. the finally confirmed experimental parameters are shown in table 6. the experimental data will be confirmed according to the above-mentioned quality evaluation methods, and the comprehensive evaluation and comparison tables of the quality of the diameter and thickness will be separately compiled, as shown in table 7 and table 8. the product quality characteristics and ideal functions of this study are based on the manufacturer's requirements for product quality. the diameter roundness specification is 555 ±0.3 μm (target value: 555 μm, tolerance: ±3 μm), and the thickness deviation specification is 5% and expectedly smaller. (target value: 0μm). additionally, for the diameter quality characteristics, the multi-quality optimized cpk value is 1.69, which is much larger than the 0.70 of the taguchi method, and the average diameter value is also the closest to the target value. the standard deviation of 0.058 is also lower than 0.132 of the taguchi method. it is found that the two-stage optimization is better. moreover, for the thickness quality characteristics, after the two-stage optimization, the thickness deviation is reduced from the taguchi method, 0.0281, to 0.0191; the standard deviation is also reduced from 0.0075 of the taguchi method to 0.0036. it can be seen that after the two-stage optimization, not only the diameter is closer to the target value, but also the thickness deviation is reduced, and the process is more stable. the results show that the two-stage optimization is better in the comprehensive evaluation of each quality. advances in technology innovation, vol. 4, no. 4, 2019, pp. 211-221 219 table 6 optimal parameters and machine settings tl cd cs dac or taguchi + anova 50 3.00 4.00 9 20 machine settings 50 3.00 4.00 9 20 two-stage optimization 49.582 2.002 3.850 9.004 51.912 machine settings 50 2.00 4.00 9 52 table 7 a comprehensively evaluation & comparison table of diameter quality cpk average standard deviation taguchi + anova 0.70 554.980 0.132 two-stage optimization 1.69 555.004 0.058 table 8 a comprehensively evaluation & comparison table of thickness quality average standard deviation taguchi + anova 0.0281 0.0075 two-stage optimization 0.0191 0.0036 3.7. process parameters and quality characteristics analysis fig. 4 the effect of current density and cathode size on diameter s/n ratio this section discusses the relationship between process parameters and quality characteristics. the process parameter control factors of this experiment are the temperature, current density, cathode size, distance between cathode and anode, and oscillating rate. according to previous investigation, the current density and cathode size are the most significant factors for the process parameters in this research; therefore, this study uses those factors as variation factors for more in-depth analysis. the cathode size, the cathode-anode distance, and the oscillating rate are fixed according to the optimum parameters of the taguchi experimental analysis, and their values are 50° c, 9 cm and 20 times / min respectively. when the s/n ratio predictor and the quality predictor are used as the variation factors for the predicted current density and the cathode size, the output of the predictor is shown in figs. 4 to 7. as what is shown from figures 4 and 5, for the diameter roundness s/n ratio, the smaller the value of the current density and the larger the value of the cathode size are, the greater the influence on the s/n ratio is. therefore, if it is desired that the diameter roundness s/n ratio can achieve better results, the current density and cathode size parameters are respectively lowered and increased to perform better. for the thickness deviation s/n ratio, the smaller the current density and the larger the cathode size are, the greater the influence on the thickness s/n ratio is. it is clearly observed that the current density interacts with the cathode size. therefore, if a thickness s/n ratio is desired to obtain a better result, the adjustment of the current density should be considered, followed by the cathode size. in addition, as seen from fig. 6 and fig. 7, if the value of the current density is lower and the value of the cathode is higher, the influence on the diameter roundness and thickness deviation is more obvious. thus, if the product process requires a larger size, the parameters with lower current density adjustment and higher cathode size adjustment can obtain better results. the diameter quality characteristic required for this experiment in this research is 555μm; and the thickness quality characteristic is the smaller the better. according to the advances in technology innovation, vol. 4, no. 4, 2019, pp. 211-221 220 trend graph of the diameter quality predictor, the drop point is between the current density of 1~2 a/dm 2 and the cathode size is about 3~4 dm 2 , and according to the trend graph of the thickness quality predictor, the drop point is about between 1~3 a/dm 2 (current density) and 3.5~4 dm 2 (cathode size). therefore, in this study, the optimal parameters of the s/n ratio predictor and the quality predictor are obtained by the hybrid pso-ga. the best parameters are current density 2.00 a/dm 2 and cathode size 4 dm 2 . fig. 5 the effect of current density and cathode size on thickness s/n ratio fig. 6 the effect of current density and cathode size on the diameter fig. 7 the effect of current density and cathode size on thickness 4. conclusions the micro electroforming technology has been widely adopted, and the industry has higher and more requirements for the product precision. how to set appropriate process parameters to meet the quality requirements and improve the production efficiency and process stability often bothers the engineers. therefore, if the optimal process parameters can be found, it will improve product quality and reduce costs. to this end, this research studies the intelligent optimization system of the micro electroforming process parameters for the mesh filter, using the systematic optimization method to effectively find the optimal combination of process parameters. after the actual verification, for the diameter quality characteristics, the multi-quality optimized cpk value is 1.69, which is much larger than the 0.70 of the taguchi method, and the average diameter value is also the closest to the target value. the standard deviation of 0.058 is also lower than 0.132 of the taguchi method. for the thickness quality characteristics, the thickness deviation is reduced from the taguchi method, 0.0281, to 0.0191; the standard advances in technology innovation, vol. 4, no. 4, 2019, pp. 211-221 221 deviation is also reduced from 0.0075 of the taguchi method to 0.0036. therefore, accordingly, the results suggest that t he proposed intelligent optimization system not only makes the diameter closer to the target value but also reduces the thickness deviation and makes the process more stable. conflicts of interest the authors would like to thank material and chemical research laboratories of industrial technology research institute (itri) in taiwan for providing equipment and technical support. references [1] m. kim, j. y. lee, s. c. kwon, d. kim, i. g. kim, and y. choi, “application of small angle neutron scattering to analyze precision nickel mesh for electro-magnetic interference shielding formed by continuous electroforming technique,” physica b: condensed matter, vol. 385-386, no. 2, pp. 914-916, november 2006. [2] s. h. hong, j. h. lee, and h. lee, “fabrication of 50 nm patterned nickel stamp with hot embossing and electroforming process,” microelectronic engineering, vol. 84, no. 5-8, pp. 977-979, may-august 2007. [3] p. m. ming, d. zhu, y. y. hu, and y. b. zeng, “micro-electroforming under periodic vacuum-degassing and temperature-gradient conditions,” vacuum, vol. 83, no. 9, pp. 1191-1199, may 2009. [4] s. w. ryu, s. choo, h. j. choi, c. h. kim, and h. lee, “replication of rose petal surfaces using a nickel electroforming process and uv nanoimprint lithography,” applied surface science, vol. 322, pp. 57-63, december 2014. [5] c. w. park and k. y. park, “an effect of dummy cathode on thickness uniformity in electroforming process, ”results in physics,” vol. 4, pp. 107-112, 2014. [6] w. yu and s. zhou, “the current situation and development trend of electroforming,” advanced materials research, vol. 960-961, pp. 103-108, june 2014. [7] x. chen, l. liu, j. he, f. zuo, and z. guo, “fabrication of a metal micro mold by using pulse micro electroforming,” micromachines, vol. 9, pp. 13, april 2018. [8] m. zhao, l. du, c. du, z. wei, x. ji, z. bai, and x. liu, “quantitative study of mass transfer in megasonic micro electroforming based on mass transfer coefficient: simulation and experimental validation,” electrochimica acta,” vol. 297, no. 20, pp. 328-333, february 2019. [9] m. behagh, a. r. f. tehrani, h. r. s. jazi, and s. z. chavoshi, “investigation and optimization of pulsed electroforming process parameters for thickness distribution of a revolving nickel part,” part c: journal of mechanical engineering science, vol. 227, pp 1-9, 2013. [10] s. d. mohanty, r. c. mohanty, and s. s. mahapatra, “study on performance of edm electrodes produced through rapid tooling route,” journal of advanced manufacturing systems, vol. 16, no. 4, pp. 357-374, december 2017. [11] b. g. park, s. g. kim, and j. s. ko, “a study on the optimization of electroplating conditions for silicon vias using the taguchi experimental design method,” international journal of precision engineering and manufacturing, vol. 20, no. 3 , pp. 437-442, march 2019. [12] x. y. jiang and w. c. chen, “an effective approach for the optimization of cutting parameters,” international journal of computer applications in technology, vol. 4, no. 3, pp. 180-189, 2014. [13] k. h. kim, j. c. park, y. s. suh, and b. h. koo, “interactive robust optimal design of plastic injection products with minimum weldlines,” international journal of advanced manufacturing technology, vol. 88, pp. 1333-1344, 2017. [14] s. kitayama, k. tamada, m. takano, and s. aiba, “numerical optimization of process parameters in plastic injection molding for minimizing weldlines and clamping force using conformal cooling channel,” journal of manufacturing processes, vol. 32, pp 782-790, april 2018. [15] l. v. candioti, m. m. de zan, m. s. camara, and h. c. goicoechea, “experimental design and multiple response optimization using the desirability function in analytical methods development,” talanta, vol. 124, pp. 123-138, june 2014. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation, vol. 5, no. 3, 2020, pp. 135-146 robust adaptive depth control of hybrid underwater glider in vertical plane ngoc-duc nguyen 1 , hyeung-sik choi 2,* , han-sol jin 2 , jiafeng huang 2 , jae-heon lee 2 1 department of electrical and information engineering, seoul national university of science and technology, seoul, korea 2 department of mechanical engineering, korea maritime and ocean university, busan, korea received 25 december 2019; received in revised form 22 march 2020; accepted 09 june 2020 doi: https://doi.org/10.46604/aiti.2020.4142 abstract hybrid underwater glider (hug) is an advanced autonomous underwater vehicle with propellers capable of sustainable operations for many months. under the underwater disturbances and parameter uncertainties, it is difficult that the hug coordinates with the desired depth in a robust manner. in this study, a robust adaptive control algorithm for the hug is proposed. in the descend and ascend periods, the pitch control is designed using backstepping technique and direct adaptive control. when the vehicle approaches the target depth, the surge speed control using adaptive control combined with the pitch control is used to keep the vehicle at the desired depth with a constant cruising speed in the presence of the disturbances. the stability of the proposed controller is verified by using the lyapunov theorem. finally, the computer simulation using the numerical method is conducted to show the effectiveness of the proposed controller for a hybrid underwater glider system. keywords: nonlinear robust adaptive control, depth control, hybrid underwater glider, buoyancy engine 1. introduction underwater glider (ug) is the new type of autonomous underwater vehicle (auv). ug has some advantages over the conventional auv. the first innovation is that it can use low energy to survey the large area of ocean and another feature is that it can glide extremely quietly under the ocean. therefore, there are many researches on this vehicle not only in oceanography but also in military purposes. in [1], a stable gliding condition was derived and an lqr controller was designed for a sawtooth depth of ug using the equilibrium point in the stable condition. a model predictive control was developed in [2] for ug to compensate for the drift due to external disturbances. in [2], the simulation and experiment studies were carried out to find the optimal pitch angle in gliding motion for the maximum speed. in [4], an energy optimal depth controller was developed for a long-range autonomous underwater vehicle with applications to hybrid ug during the level flight. a robust integral sliding mode with a super-twisting algorithm was proposed in [5] for the trajectory tracking problem of autonomous ug with environmental disturbances. in [6], the optimal control of an ug vehicle was proposed using linear quadratic regulator strategy. a new optimal three-dimensional path planning method was developed in [7] for the minimum energy consumption for ugs. in [8], a self-searching optimal control was proposed for pitch-keeping control of ug during descent and ascent under ocean currents and noise disturbances. a novel control algorithm was presented in [9] using reinforcement learning with active disturbance rejection control (adrc) and it was compared with the classical adrc by simulation of ouc-iii glider. in [10], a new hybrid heading tracking control algorithm was proposed using adaptive fuzzy incremental pid and anti-windup compensator to improve the heading control of ugs. a dynamic surface decoupling control algorithm using * corresponding author. e-mail address: hchoi@kmou.ac.kr tel.: +8210-5581-2971 advances in technology innovation, vol. 5, no. 3, 2020, pp. 135-146 136 adrc was proposed in [11] and sea trial results also were presented here with some improvements in overshoot and settling time. in [12], an lqr controller to maintain the sawtooth vertical movement in the gliding motion of ug was presented with the reduced order luenberger observer for estimating the gravity center and buoyancy mass. in [13], to realize the precise navigation for the developed hybrid ug, an attitude reference system was composed of a ring laser gyroscope and a geomagnetic sensor was presented applying extended kalman filter algorithm. in addition, a ray-type shape underwater glider with robust adaptive heading control using propellers was developed and tested in the sea trials [14]. among those gliders, the depth control problem is still a challenging problem. the purpose of this study aims at the depth control of the hug using a combination of the pitch and speed controls, robust to disturbances will be designed. in this research, to contribute to the control design of the torpedo-shaped hybrid underwater glider (hug) with buoyancy engine and propeller, the adaptive robust depth control is proposed for depth keeping control against unknown parameters and environmental disturbances. this controller is designed based on the lyapunov stability theorem and the backstepping technique for underactuated dynamics like hug. the hydrodynamic coefficients of this vehicle were obtained from [15] due to the similar torpedo shape and similar design dimension. the heave motion is controlled through pitch control and speed control with the line-of-sight guidance law. to see the effectiveness of this proposed controller for depth keeping control, the computer simulation using matlab is performed and the stable performance is guaranteed. 2. vertical model of hybrid underwater glider : : surge roll u p k x z n m y : : sway pitch v q : : heave yaw w r body-fixed coordinate system north-east-down coordinate system    0 y o 0 x 0 z e x y z (a) earth-fixed and body-fixed coordinate system (b) developed hug in the field test fig. 1 hybrid underwater glider to find the position of the ug; first, the relationship between the body-fixed coordinate and earth-fixed coordinate is expressed using the angle of the hull and the acceleration as in fig.1. for this, heading and attitude sensors are in the glider [16]. the vertical dynamics of hug can be expressed in the kinematic and kinetic model as eq. (1) [17]. ( ) ( ) ( )       t e jv m c v d v v g j       (1) where 𝜂 = [𝑥 𝑧 𝜃]𝑇 is the position and orientation of the vehicle in inertial frame 𝐸𝑥𝑦𝑧 in fig. 1; 𝜈 = [𝑢 𝑤 𝑞]𝑇 is the translation and angular velocity in the body-fixed frame 𝑂𝑋0𝑌0𝑍0 in fig. 1; 𝐽 is jacobian matrix; 𝑀 = 𝑀𝑅𝐵 + 𝑀𝐴 is the inertia matrix; 𝐶(𝜈) = 𝐶𝑅𝐵 (𝜈) + 𝐶𝐴(𝜈) is the coriolis and centripetal matrix; 𝐷(𝜈) is hydrodynamic damping matrix; 𝑔(𝜂) is the gravitational matrix; 𝜏 is the external forces and moments; 𝑔0(𝜂) is the ballast forces and moments which are generated by buoyancy engine in the hybrid underwater glider (hug); 𝜏𝑒 is the environment disturbances. to obtain the detail dynamic, one can expand eq. (1) into eq. (2), while the dynamic coefficients are obtained from the remus auv model [15] which has the same hull dimension as the developed hug. advances in technology innovation, vol. 5, no. 3, 2020, pp. 135-146 137     22 | | 2 cos sin sin cos | | ( )sin sin q u euu g g w wu u w g gq q x u v z u v q u mz q mx q mwq z wq z x u u w b m z q m x w mx z q mz q muq x                                         | | | | cos| | ( )cos | | ( )sin ( )cos eww yy q u w w g w w q u q q g g q eqb b gi m uq z w w w b q mz u qw w qu m w z wu z qu x uw m q q z w z b x w x b mx                             (2) as this time, 𝑋�̇�, 𝑍�̇�, and 𝑀�̇� are the added mass coefficients; 𝑋𝑢, 𝑍𝑤, and 𝑀𝑞 are linear damping coefficients; 𝑋|𝑢|𝑢, 𝑍|𝑤|𝑤 , and 𝑀|𝑞|𝑞 are nonlinear damping coefficients; 𝑊 and 𝐵 are the weight and buoyancy force respectively in the neutral buoyancy condition; 𝑥𝑔 and 𝑧𝑔 are coordinates of gravity center in the body-fixed frame; 𝑥𝑏 and 𝑧𝑏 are coordinates of buoyancy center in the body-fixed frame; 𝜏𝑤 is the control force from buoyancy engine; 𝑚 and 𝐼𝑦𝑦 are the vehicle mass and the y-axis inertia; 𝜏𝑞 is the control moment from mass-shifter; 𝜏𝑤 is the buoyancy force generated by the pump inside hug body; 𝜏𝑢 is the thruster force; 𝜏𝑒𝑢 , 𝜏𝑒𝑤 , and 𝜏𝑒𝑞 are disturbances from ocean currents and wave and [𝜏𝑒𝑢 𝜏𝑒𝑤 𝜏𝑒𝑞 ] 𝑇 = 𝐽𝑇 𝜏𝑒 . m g 2 c r  bu sealed at atmosphere pressure, 1atmsealed at atmosphere pressure, 1atm o-ringo-ring seawaterseawater motor and gearmotor and gear ball-screwball-screw fig. 2 buoyancy engine diagram in this hug system, the buoyancy engine will let the water in or out by moving piston along the cylinder as revealed in fig. 2. during this process, the volume of this vehicle will decrease or increase according to the position of the piston. if the weight is equal to the buoyancy force in the neutral condition, this glider descends when its volume is reduced and ascends toward the water surface when its volume is increased. in order to specify the force that this buoyancy engine can produce, the travel distance of the piston and the radius of cylinder should be defined. the buoyancy force is equal to the weight of seawater going in or out and it is indicated in eq. (3). as this point, in fig. 2, 𝑢𝑏 is the position of piston; 𝑅𝑐 is the radius of the cylinder; 𝜌 is the density of seawater; 𝑔 is the gravitational acceleration. during the operation of the buoyancy engine, the center of the buoyancy is shifted along the 𝑂𝑥0 axis by eq. (4). 2 w b c u r g   (3) 2 2 ( ) 2 b c b p b c b nb u r u x x r u v      (4) where, 𝑋𝑝 is the position of the piston in the neutral position along the 𝑂𝑥0 axis in the body-fixed frame; 𝑉𝑛𝑏 is the volume of the vehicle in the neutral condition of the buoyancy engine. the movable mass in fig. 3 can move along the ox0 axis; therefore, the center of gravity in this axis 𝑥𝑔 can be defined as: stat stat m m g m x m u x m   (5) advances in technology innovation, vol. 5, no. 3, 2020, pp. 135-146 138 where, 𝑚𝑠𝑡𝑎𝑡 = 𝑚 − 𝑚𝑚 is the static mass; 𝑚𝑚 is the weight of the movable mass; 𝑢𝑚 is the position of the moving mass; 𝑥𝑠𝑡𝑎𝑡 is the position of the static mass and it is assumed to be very small because the origin of body-fixed frame is located near to the center of gravity. furthermore, 𝑥𝑔 can be approximated as:. m g m m x u m  (6) the moment produced from the mass-shifter can be computed by the product of the net buoyancy force and the location of the center of gravity as found in eq. (7). in addition, the moment of inertia is also changed because of the change in mass distribution following eq. (8). 𝐼𝑛𝑦 is the moment of inertia in the neutral condition of moving mass. m q g m m wx w u m    (7) 2 ( )yy ny m m mi i m u sgn u (8) m g moving massmoving mass mu motor and gearmotor and gear ball-screwball-screw fig. 3 mass shifter diagram 3. depth control 3.1. pitch control the pitch angle in the proposed design of the hug is changed by controlling the moving mass position and the pitch angle dynamics can be described as eq. (9), which are from eq. (2).  33 2 system 1: system 2: , , q eq q m q f           (9) where 33   qym i m (10) 2 ( ) ( ) ( )sin ( )cos            g g w w q u g b g bq qf mz u qw mx w qu m w z wu z qu x uw m q q z w z b x w x b  (11) in this system, the buoyancy force is only controlled by on and off mode for descending and ascending. therefore, the moment produced by moving mass is used and designed using the adaptive backstepping technique for the vertical dynamics. two differential equations in eq. (9) are considered as two subsystems for the backstepping control. for one equation, the virtual control law is set for 𝑞𝑑 so that 𝜃 → 𝜃𝑑 as 𝑡 → ∞, and then the control moment 𝜏𝑞 is designed for controlling 𝑞 → 𝑞𝑑 as 𝑡 → ∞. the errors of system 1 and 2 are defined in: 1 2 d d e e q q         (12) error dynamics of the pitch dynamics are defined as below: 1 2d d d de q e q          (13) advances in technology innovation, vol. 5, no. 3, 2020, pp. 135-146 139 the error dynamics in eq. (13) can be stabilized by designing 𝑞𝑑 = �̇�𝑑 − 𝑘1𝑒1 with the positive control gain, 𝑘1. then, the error dynamics of system 1 can be obtained as: 1 1 1 2e k e e  (14) derivative of the virtual control is: 2 1 1 1 2d dq k e k e   (15) error dynamics of the subsystem 2 is obtained as: 33 33 33 2 332 d q eqd m qm e m q m q f       (16) using the sliding mode control technique for finding the control law for pitch control, one can choose the lyapunov function for this system as below: 2 2 1 2 1 33 2 2 2 2 1 1 1 2 2 2 v e m e a p a   (17) where 𝑌2𝑎2 = −𝑓2 + 𝑚33�̇�𝑑; 𝑃2 = 𝐼9×9 is the identity matrix; �̃�2 = �̂�2 − 𝑎2; 𝑌2 = [�̇� �̇� 𝑞𝑢 𝑤𝑢 𝑞𝑤 |𝑞|𝑞 𝑠𝑖𝑛𝜃 𝑐𝑜𝑠𝜃 �̇�𝑑 ] is the regressive vector for updating the parameter vector 𝑎2 defined as: 32 3[ ( )( )( ) ( ) ( ) ]            t g g w g q u w g g b g bq q a mz mx m mx z x z mz m z w z b x w x b m (18) the derivative of the above lyapunov function is expanded in eq. (19). to stabilize the dynamics in this equation, the control input is designed as eq. (20) with the saturation function in the switching control. 2 1 2 1 1 1 2 2 2 2 2 2 ˆ( ) tq eqv k e e e e y a a p a         (19) 2 2 2 2 2 1 2 2 ˆ ˆ ( )q eq e y a k e e k sat        (20) substituting the control law (20) into the first derivative of the lyapunov function (19) will result to: 2 2 12 2 1 1 2 2 2 2 2 2 2 2 2 2 2 2 ˆ ˆ( ) ( )eq eq e v k e k e k e sat e e y a a p a           (21) by choosing the adaption law as �̇̂� = −𝑃2𝑌2 𝑇 𝑒2, the derivative of 𝑉2can be derived as: 2 2 2 2 1 1 2 2 2 2 2 2 ˆ( ) ( )eq eq e v k e k e k e sat e         (22) it is observed that if 𝑘2∆ > 𝐹𝑒𝑞 and |𝜏𝑒𝑞 − �̂�𝑒𝑞 | ≤ 𝐹𝑒𝑞 , then �̇�2 ≤ −𝑘1𝑒1 2 − 𝑘2𝑒2 2 ≤ 0. here, 𝐹𝑒𝑞 is defined relating with the environmental condition. following the direct lyapunov method for v2, the control law (20) is proved to be globally asymptotically stable (gas). 3.2. speed control the vehicle speed will be controlled by thruster attached at the tail of hug. the speed dynamics will be investigated to carry out the speed controller. the system 3 is defined as eq. (23) which is the 4th equation in eq. (2). advances in technology innovation, vol. 5, no. 3, 2020, pp. 135-146 140  311system 3: , , sinw u eum u f           (23) where 11   um m x (24) 2 3 ( ) ( ) ( )sin       g g q w u uf mz q mx z q m z wq x u u w b  (25) the speed error is defined as: 3 de u u  (26) and the speed error dynamics can be formulated as: 11 3 11 3 11sinw ud dm e m u u f m u        (27) the lyapunov candidate can be designed as: 2 1 3 11 3 3 3 3 1 1 2 2 tv m e a p a  (28) where 113 [ ( )( ) ( ) ]        t g g q w wu u a mz mx z m z x b w m (29) 3 3 3 ˆ a a a (30) furthermore, its derivative is formulated as eq. (31) using the dynamics in eq. (27). 1 3 3 3 3 3 3 3( )u euv e y a a p a       (31) where 𝑌3 = [�̇� 𝑞 2 𝑤𝑞 |𝑢|𝑢 𝑠𝑖𝑛𝜃 �̇�𝑑 ]. the speed control law can be designed as: 3 3 3 3 3 3 3 ˆ ˆ ( )u eua e y k e k sat       (32) applying this control law into (31) yields: 2 13 3 3 3 3 3 3 3 3 3 3 3 3 ˆ ˆ( ) ( ) teu eu e v k e k sat e e e y a a p a          (33) the last two term in eq. (25) can be eliminated by choosing �̇̂�3 = −𝑃3𝑌3 𝑇 𝑒3, then �̇�3 can be reformulated as: 2 3 3 3 3 3 3 3 3 ˆ( ) ( )eu eu e v k e k sat e e        (34) if the control gain 𝑘3δ is chosen as: 3 | |ˆeu euk     (35) then, the derivative of the lyapunov function 𝑉3 is negative definite as: advances in technology innovation, vol. 5, no. 3, 2020, pp. 135-146 141 2 3 3 3 3 3 3 2 3 3 ˆ (n.d) eu euv k e k e e k e       (36) this control law is proved to be gas using lyapunov function 𝑉3 as shown in eq. (28). 3.3. los depth-keeping guidance for navigation of the hug, a line of sight (l.o.s) method based on the calculated heading angle was used [18]. the line-of-sight (los) guidance in fig. 4 is applied for path following control for the vertical plane. it is assumed that the position information of the hug is available from the inertial navigation system. then, a two-dimensional los guidance is constructed to apply the desired pitch angle 𝜃𝑑 to the controller. the way-point in the los guidance is given by the operator and it contains 2 components, 𝑥𝑘 and 𝑧𝑘 . the path variables by the previous waypoint and the current waypoint are defined by 𝑥𝑘 , 𝑧𝑘 , 𝑥𝑘−1 and 𝑧𝑘−1. by solving eqs. (37)-(38) for the virtual los point (𝑥𝑙𝑜𝑠, 𝑧𝑙𝑜𝑠 ), the desired pitch angle can be computed by eq. (39). 𝐿𝑝𝑝 is the length of the vehicle; 𝑛 is the positive gain that combines with 𝐿𝑝𝑝 to define the los radius. 2 2 2( ) ( ) ( )pplos losz z x x nl    (37) 1 1 1 1 los k k k los k k k z z z z x x x x          (38) arctan          los d los z z x x  (39) robust adaptive control huglos disturbance , d d x z d   , ,x z  e , ,u w q parameter adaptation d u fig. 4 algorithm of depth control for the energy efficient operation, hug will glide down by only using the buoyancy engine and mass shifter. when it reaches the desired depth, the depth control is applied to keep it at the desired depth. the adaptation law is used to estimate the nonlinear unknown term in the pitch and surge dynamics. the estimation algorithm for the unknown parameters is used in the control action while stabilizing the dynamics. furthermore, the sliding mode control is used to deal with the environmental disturbances from currents and waves. in the following section, the simulation of the diving, depth-keeping and redurfacing are presented. 4. simulation and discussion in order to validate the good performance of the proposed depth control, the simulation of heaving motion control of the hug is presented using the matlab with the vehicle parameters which is presented in table 1. the descending and ascending motion is actuated by only buoyancy engine and mass shifter. when the hug goes down close to the desired depth, the buoyancy engine is controlled to stay in the neutral buoyant condition or zero net buoyancy. the thruster and mass shifter are used to drive the vehicle along the desired depth with the designed speed. in this simulation, the desired depth is 200 m advances in technology innovation, vol. 5, no. 3, 2020, pp. 135-146 142 ( 1 2 200 z z m ), the length of trajectory distance is 300 m, the desired speed is 2 knots ( 1 /du m s ). also, environmental distrubances are chosen as the sine function with the magnitude of 2n, -1n and 2nm (  | | 2 1 2 t e   ) for 𝐸𝑥 , 𝐸𝑧 and 𝐸𝑦 axes respectively. table 1 vehicle parameters parameters value parameters value parameters value m 70kg 𝑋�̇� −9.3 × 10 −1kg 𝑍|𝑤|𝑤 −1.31 × 10 2kg/m 𝐼𝑦𝑦 1.29kgm 2 𝑍�̇� −3.55 × 10 1kg 𝑀�̇� −4.88kgm 2/rad 𝑚𝑚 1.5kg 𝑍�̇� −1.93kgm/rad 𝑀�̇� −1.93kgm 𝑊 − 𝐵 ±0.7kg 𝑋|𝑢|𝑢 −3.9kg/m 𝑀|𝑞|𝑞 −1.88 × 10 2kgm2/rad2 the hug dives with −30° pitch angle to the depth of 200 m, then it will use the thruster to keep at the constant depth, and after moving 300 m at the desired depth it will ascends with a positive pitch angle of 30° ( 30d ). fig. 5 is indicated that hug glides down with the small oscillation due to the environment effects. this phenomenon can be observed in the pitch tracking control in fig. 6. at the desired depth, the output of los guidance is seen as the wave form because of the sine function of the disturbances. it shows that the control law drives the hug to track the path of los guidance very well in the presence of disturbances. fig. 5 trajectory of hug with environmental disturbances (a) longitudinal coordinate (b) depth (c) heading performance in depth control fig. 6 the position and orientation of hug advances in technology innovation, vol. 5, no. 3, 2020, pp. 135-146 143 fig. 7 is described velocities of the vehicle in the body-fixed coordinates. the speed control results are presented for the surge velocity 𝑢 with the desired speed is 1 m/s. it is noted that the speed control is only used in the cruise period. the robust adaptive speed control shows good tracking performance along the desired trajectory under the bounded disturbances. additionally, the control force and torque for the depth control are simulated in fig. 8. the control force of the buoyancy engine is limited in the range from -2 n to 2 n. the range of the pitching torque generated by mass shifter is in the range from -20 nm to 20 nm. (a) speed control performance (b) sway velocity (c) pitch velocity fig. 7 velocities of the hug in the body-fixed coordinates (a) net buoynacy force (b) moment induced by moving mass (c) thruster force during the depth control fig. 8 force and moment from the robust adaptive controller as stated above, the stability proof is revealed that the errors are bounded in fig. 9. 𝑒1 and 𝑒2 are from pitching control system, and they show good convergence performance, even though the desired angle changes due to the los guidance and external disturbance effects. especially, during the transition period from gliding to cruising or vice versa, 𝑒1 and 𝑒2 can quickly converge near to zero in fig. 9. for speed control, the convergence only appears in the cruising mode while keeping the advances in technology innovation, vol. 5, no. 3, 2020, pp. 135-146 144 (a) pitch angle error (𝑒1 is pitch angle error) (b) virtual control error (𝑒2 is the virtual control error) (c) speed control error in the desired depth (𝑒3 is the speed control error) fig. 9 the tracking errors from 3 sub-controller in crusing period hug to the desired depth of 200 m. in the desired depth, the speed of the vehicle is controlled at a constant value of 1 m/s or 2 knots. at the same time, the error converges quickly close to zero in fig. 9. fig. 10 the parameter adaptation of vector �̂�2 fig. 11 the parameter adaptation of vector �̂�3 two main adaptation processes are indicated in fig. 10 for vector 𝑎2 and 𝑎3 respectively. there are 9 parameters in vector 𝑎2 (18) and 6 parameters in vector 𝑎3 (29). parameters 7 and 8 of vector 𝑎2 in fig. 10 are sensitive to the external disturbances with the wave curve updating process. these parameters are defiend as –(zgwzbb) and (xgwxbb) respectively in advances in technology innovation, vol. 5, no. 3, 2020, pp. 135-146 145 (18). these terms are from the restoring moment due to the change of the gravity center and buoyancy center. in addition, in vector 𝑎3, there is one parameter that proportionally adapts to the change of disturbances, which is parameter 4 defined as 𝑋|𝑢|𝑢 in the damping force in the system 3. this phenomenon is shown in fig. 11, parameter 4 shows the waving curve in the speed control. 5. conclusions in this study, the depth control for nonlinear vertical dynamics of the hybrid underwater glider was proposed and its performance was validated through simulations. the robust adaptive control law has presented the stable performance against the parameter change and external disturbance. it is guaranteed that the error is close to zero as the time go to infinite. according to the direct lyapunov method, the stability of this control law has been established and proved. the depth control algorithm uses the los guidance law for keeping the hug to the desired depth. the computer simulation using matlab was performed to verify the good and stable performance of depth keeping control using the proposed control algorithm. in the future research, this algorithm will be implemented in the developed hug at the laboratory and, the water tank and sea trials of the hug employing the proposed depth control will be carried out. acknowledgement this study is part of a project supported by the civil-military dual-use technology “data collection system with underwater glider” (grant no. 19-sn-mu-01) and a government project entitled “development of hybrid underwater drone for accurate inspection of the underwater pipeline and cable”. conflicts of interest the authors declare no conflict of interest references [1] m. g. joo and z. qu, “an autonomous underwater vehicle as an underwater glider and its depth control,” international journal of control, automation and systems, vol. 13, no. 5, pp. 1212-1220, july 2015. [2] i. abraham and j. yi, “model predictive control of buoyancy propelled autonomous underwater glider,” proc. of the american control conference, july 2015, pp. 1181-1186. [3] s. k. jeong, h. s. choi, j. h. bae, s. s. you, h. s. kang, s. j. lee, et al., “design and control of high speed unmanned underwater glider,” international journal of precision engineering and manufacturing-green technology, vol. 3, no. 3, pp. 273-279, july 2016. [4] b. claus and r. bachmayer, “energy optimal depth control for long range underwater vehicles with applications to a hybrid underwater glider,” autonomous robots, vol. 40, no. 7, pp. 1307-1320, february 2016. [5] m. mat-noh, m. r. arshad, and r. mohd-mokhtar, “nonlinear control of autonomous underwater glider based on super-twisting sliding mode control (stsmc),” 7th ieee international conference on system engineering and technology, october 2017, pp. 71-76. [6] r. d. s. tchilian, e. rafikova, s. a. gafurov, and m. rafikov, “optimal control of an underwater glider vehicle,” procedia engineering, vol. 176, pp. 732-740, 2017. [7] y. liu, j. ma, n. ma, and g. zhang, “path planning for underwater glider under control constraint,” advances in mechanical engineering, vol. 9, no. 8, pp. 1-9, august 2017. [8] z. huang, h. zheng, s. wang, y. liu, j. ma, and y. liu, “a self-searching optimal adrc for the pitch angle control of an underwater thermal glider in the vertical plane motion,” ocean engineering, vol. 159, pp. 98-111, july 2018. [9] z. q. su, m. zhou, f. f. han, y. w. zhu, d. l. song, and t. t. guo, “attitude control of underwater glider combined reinforcement learning with active disturbance rejection control,” journal of marine science and technology, vol. 24, no. 3, pp. 686-704, 2019. advances in technology innovation, vol. 5, no. 3, 2020, pp. 135-146 146 [10] h. sang, y. zhou, x. sun, and s. yang, “heading tracking control with an adaptive hybrid control for under actuated underwater glider,” isa transactions, vol. 80, no. 10, pp. 554-563, july 2018. [11] d. song, t. guo, w. sun, q. jiang, and h. yang, “using an active disturbance rejection decoupling control algorithm to improve operational performance for underwater glider applications,” journal of coastal research, vol. 34, no. 3, p. 724, may 2018. [12] m. g. joo, “a controller comprising tail wing control of a hybrid autonomous underwater vehicle for use as an underwater glider,” international journal of naval architecture and ocean engineering, vol. 11, no. 2, pp. 865-874, july 2019. [13] s. k. jeong, h. s. choi, j. il kang, j. y. oh, s. k. kim, and t. q. m. nhat, “design and control of navigation system for hybrid underwater glider,” journal of intelligent and fuzzy systems, vol. 36, no. 2, pp. 1057-1072, march 2019. [14] n. d. nguyen, h. s. choi, and s. w. lee, “robust adaptive heading control for a ray-type hybrid underwater glider with propellers,” journal of marine science and engineering, vol. 7, no. 10, october 2019. [15] t. t. j. prestero, “verification of a six-degree of freedom simulation model,” phd thesis, dept. mechanical engineering, university of california at davis, 2001. [16] d. h. ji, h. s. cho, s. k. jeong, j. y. oh, s. k. kim, and s. s. you, “a study on heading and attitude estimation of underwater track vehicle,” advances in technology innovation, vol. 4, no. 2, pp. 84-93, april 2019. [17] handbook of marine craft hydrodynamics and motion control, john wiley & sons, 2011. [18] d. jung, s. hong, j. lee, h. cho, h. choi, and m. vu, “a study on unmanned surface vehicle combined with remotely operated vehicle system,” proceedings of engineering and technology innovation, 2018, vol. 9, no. 7, pp. 17-24, july 2018. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation, vol. 3, no. 2, 2018, pp. 59 69 improving activated sludge wastewater treatment process efficiency using predictive control ioana nascu 1 , ioan nascu2,* 1 artie mcferrin department of chemical engineering, texas a&m, college station tx, usa. 2 department of automation, technical university of cluj-napoca, romania. received 05, june 2017; received in revised form 07, june 2017; accept ed 09, july 2017 abstract this paper investigates the performance of a new predict ive control approach used to improve the energy efficiency and effluent quality of a conventional wastewater treatment plant (wwtp). a modified variant of the well-known generalized predictive control (gpc) method has been applied to control the dissolved oxygen concentration in the aerobic bioreactor of a wwtp. the quadratic cost function was modified to a positional imple mentation that considers control signal weighting and not its increments, in ord er to minimize the control energy. the activated sludge process (asp) optimizat ion using the proposed variant of the gpc a lgorith m provides an improved aeration system effic iency to reduce energy costs. the control strategy is investigated and evaluated by performing simulations and analyzing the results. both the set point tracking and the regulatory performances have been tested. moreover, the effects of some tuning parameters are also investigated. the results show that this control strategy can be efficiently used for dissolved oxygen control in wwtp. keywords: predict ive control, process optimization, process model, wastewater treat ment plant, activated sludge treatment 1. introduction wastewater treatment plants are key infrastructures for ensuring a proper protection of our environment. biologica l treatment is an important and integral part of any wwtp. the activated sludge process is the most commonly used technology to treat sewage and industrial wastewaters due to its flexibility, high reliability and cost-effectiveness, as well as its capacity of producing high quality effluent. an overv iew of the activated sludge wastewater treat ment process mathemat ical modeling is presented in [1] and application of asp models can be found in [2]. the asps a re d ifficult to be controlled because of their co mple x and nonlinear behavior. however, the optima l control of the biologica l reactors plays an important role in t he operation of a wwtp and the efficiency of most wwtp is an important issue still to be improved . a good control of wwtp processes could lead to better water quality and t o an effic ient use of energy [3-4]. this research area is a key part of keeping the environ ment clean and nowadays has received great emphasis due to the strict regulations fo r the discharged water. many o f the wwtp are operated in a less -than-optimal manner with respect to both treatment and energy efficiency, causing high costs and inefficient operation in order to meet the regulations. * corresponding author. e-mail address: ioan.nascu@aut.utcluj.ro tel.: +40-264-40122 advances in technology innovation, vol. 3, no. 2, 2018, pp. 59 69 copyright © taeti 60 the dissolved oxygen concentration has a ma jor impact in the activated sludge process. to control the dissolved oxygen concentration, the a mount of air provided by the blowe rs in the ae ration tank is modified. the activated sludge process is the most energy consuming process es in the whole wwtp, nearly half of the energy consumed in a wwtp being used in the aeration tank. an over dosage of aeration is unwanted because it brings increased costs and just little or even no gain in the quality of water. fo r values of dissolved oxygen concentration in t he aeration tank above 2 mg/l, the increase of the aeration flow begins to have a lower effect in the quality of effluent and at values of 4-6 mg/ l doesn’t have any effect. in the absence of adequate control systems, to reduce the effect o f disturba nces on the flow or load of the effluent, it is sometimes preferred to operate at high concentrations of dissolved oxygen in the aerat ion tank. using the blowe rs in manual operation at constant flow rates during the periods of reduced wastewater intake will produce a loss of energy. thus, optimizing the aerat ion process defines an important objective to reduce energy consumption and imp rove energy effic iency. severa l strategies have been proposed for controlling the dissolved oxygen concentration. so me researchers have focused on the importance of we ll -tuned models and simu lation p latforms in the process of designing the controllers for the dissolved oxygen concentration [5 -6]. other researchers have focused on designing multip le model controllers instead of nonlinear comp le x ones [7]. nevertheless, because of the high nonlinearit ies of the process, robust controllers are required to maintain an optimu m setpoint, regardles s of the changes in the operating point. su ch control strategies have also been proposed, including adaptive [8-9], predict ive [10-11], fuzzy [12] or fractional order piμdλ control [13]. optimizing and ma intaining the dissolved oxygen set point define important objectives for researchers in wwtp c ontrol. optimization of the dissolved oxygen set point is not the purpose of this paper. in practice, an appropriate dissolved oxygen set point is determined either manually by e xperienced operators or automatica lly through optimization algorith ms. in this paper, we assume the appropriate set point is prescribed by the optimizing part of a mult ilayer hie rarchica l control structure and the proposed control system will be responsible for forcing the plant to follow this set -point. a modified variant of the well-known generalized predict ive control method [14] has been applied to control the d issolved oxygen concentration in the aeration tank of an activated sludge process . the asp process optimization using the proposed variant of the gpc a lgorith m provides an improved aeration system effic iency to reduce energy costs. the gpc a lgorith m is well known and consists of applying a control sequence that minimizes a quadratic cost function defined over a prediction horizon. the consideration of weighting o f control incre ments in the cost function in gpc allows an incre mental imp le mentation which ensures offset -free re ference tracking and disturbance rejection but does not provide minimizat ion of the control energy. to a lleviate the above mentioned limitat ion, the aim of this paper is to propose a different variant of the quadratic cost function. in order to minimize the control energy, the quadratic cost function was modified to a positional imp le mentation that considers control signal weighting and n ot its incre ments. to evaluate the performance of the proposed design for asp process optimizat ion, simulat ion results are presented and discussed in detail. the asp process was first modeled and the models were calibrated and validated based on a combination of laboratory tests and plant operating measured data. the proposed control strategy is investigated and evaluated by performing simulations and analyzing the results. both the set poin t tracking and the regulatory performances have been tested. moreover, the effects of some tuning parameters are also investigated. the results show that this control strategy ca n be efficiently used for dissolved oxygen control in wwtp. 2. process description and modeling the activated sludge wastewater treatment processes are very co mple x, with large, uncontrollable input disturbances, significant nonlinearities and characterized by uncertainties regarding their para meters. the most widely used model s to describe these processes is the activated sludge model nr.1 (asm 1) proposed by the international water association (iwa) [15]. hav ing thirteen state variables and eight dynamic processes, this model is highly co mple x, but it provides deep insight in advances in technology innovation, vol. 3, no. 2, 2018, pp. 59 69 copyright © taeti 61 the behavior of the process. since the model a lso contains a large number of b iokinetic and stoichiometries parameters, for control purposes it is necessary to simplify it into a simpler model, especially if a hierarchical control structure is used. the modelling process and the model ca libration are made based on data obtained from a conventional activated sludge system operating under aerobic conditions and whose main purpose is to ensure the removal o f colloida l and dissolved carbonaceous organic matter. the res idual water that needs to be treated is coming fro m a factory that processes and is painted cotton, a milk factory and from domestic households. fig. 1. wwtp's biological treatment process schematic configuration. the wastewater first enters the aerated bioreactor where the treatment based on asp takes place. the clea r water and the sludge are separated due to gravity in the secondary settler. in orde r to keep b iological sustainability, the act ive sludge is recirculated and the bioreactor is aerated using an aeration network where air is being blown with fine bubbles. in our previous work [16-17] we have developed a reduced model to the asp based wastewater treatment process where the simplest possible case was taken into consideration. only the re moval of organic matter is considered, while bio logical phosphorus and nitrogen re moval is neglected . the following co mponents are treated in the model: one organic matter component, one microorganism co mponent and dissolved o xygen. for the model used in this paper, t wo processes are considered to take place in the aeration tank: the reduction of organic substance in heterotrophic aerobic bacteria and the reduction of ammonia n itrogen with autotrophic aerobic bacteria. the carbonaceous conversion is integrated in a consistent manner with the transformations of nitrogen. the following components are treat ed in the model: one organic matter component, two nitrogen components, two microorganism components and dissolved oxygen. the developed model is based on the following assumptions: the content of the aeration tank is considered perfect stirred; there are no d irections in the secondary settler; the bio mass concentration in the e ffluent is negligible; the o xygen concentration and substrate are neglected in the recycled sludge; the active sludge is the only recycled component into the aeration tank. in this case, there are 6 equations . that can be written for the aeration tank considered as a completely mixed reactor. eqs. (1-6) correspond to the mass balance eqs. for:heterotrophic (1) and autotrophic (2) b io mass, biodegradable substrate (3), ammonia nitrogen (4), nitrite and nitrate nitrogen (5) and dissolved oxygen(6) concentration. ( ) ( ) ( ) ( )(1 ) ( ) [ ( ) ( ) ] ( ) b,h r b,h b,h h ha h b,h dx t = r d t x t d t + r x t + m t + m t b x t dt     (1) ( ) ( ) ( ) ( )(1 ) ( ) [ ( ) ] ( ) b,a r b,a b,a a a b,a dx t = r d t x t d t + r x t + m t b x t dt    (2) ( ) ( ) ( ) ( ) ( )[( ) ( ) ( )]s h ha b,h s sin h ds t m t + m t = x t d t 1+ r s t + s t dt y   (3) bioreactor, s,x,do settler, xr recycled sludge, xr, rd waste sludge, x, βd aeration,w influent, sin, d, doin effluent se, (1-β)d advances in technology innovation, vol. 3, no. 2, 2018, pp. 59 69 copyright © taeti 62 , , ( ) 1 1 = ( ) ( ) ( ) ( ) ( )[(1 ) ( ) ( )] 2.86 no ha b h a b a no noin h a ds t yh m t x t m t x t d t r s t s t dt y y       (4) ( ) 1 [ ( ) ( )] ( ) ( ) ( ) ( ) ( )[(1 ) ( ) ( )]nh xb h ha bh xb a ba nh nhin a ds t = -i m t + m t x t i + m t x t d t + r s t + s t dt y (5) 1( ) ( ) ( ) (1 ) ( ) ( ) ( )[(1 ) ( ) ( )] [ ( )]h in maxh b,h a b,a h a yddo t 4.57 = m t x t m t x t d t + r do t + do t ++aw do do t dt y y       (6) the mass balance equations for the recycled biomass are: )())(()()1)(( )( ,, , txrtdtxrtd dt tdx hrbshbs hrb   (7) , , , ( ) ( )(1 ) ( ) ( )( ) ( ) rb a s b a s rb a dx t d t r x t d t r x t dt     (8) under steady state conditions, from mass balance equations in the settling tank the resulting concentrations in the effluent ssef, snhef and snoef are: ( ) (1 ) ( ) sef s s t = + r s t (9) ( ) (1 ) ( ) nhef nh s t = + r s t (10) (t)s r)+(1(t)s nonoef  (11) the equations for the heterotrophic growth of the biomass in aerobic (µh) and anoxic (µha) conditions are: h hmax , ( ) ( ) ( ) ( ) ( ) s s s o h s t do t t k s t k do t      (12) h a hmax , ( ) ( ) ( ) ( ) ( ) ( ) s oh no g s s o h no no s t k s t t k s t k do t k s t          (13) while the equation for the growth of autotrophic mass µa is: )( )( )( )( (t) , hmaxa tdok tdo tsk ts aonhnh nh      (14) where xb,h and xb,a represent the active heterotrophic and autotrophic biomass concentration, ss, snh, sno, and do concentration of b iodegradable organic matter, a mmonia nitrogen, n itrite -n itrate, and dissolved oxygen in the aerated bioreactor, ssin, snhin, and snoin concentrations in the influent, xrb,h and xrb,a recycled heterotrophic and autotrophic bio mass concentrations, d and ds dilution rates (ratio of influent flow to volu me of the aerated bioreactor and settler), w aeration rate, α o xygen transfer rate, r the ratio of recycled sludge flo w to influent flow, β the ratio of waste flow to influent flow, ya autotrophic biomass yield factor, yh heterotrophic bio mass yield factor, ixb conversion coefficient for the nitrogen mass, koh and koa oxygen saturation coefficients at half for heterotrophic/autotrophic biomass, knh and kno ammonia and nitrate saturation coeffic ients at half for autotrophic bio mass, ks organic substrate saturation coefficient, ηg correct ion coeffic ient for µh in ano xic conditions. this model has 8 state variables: xb,h, xb,a, xrb,h, xrb,a, ss, snh, sno, and do. the kinetic and stoichiometric para meters va lues obtained after the model calibrat ion are : bh=0.034; ba=0.002; yh=0.54; ya=0.13; ixb=0.068; α =0.016; domax=10; ηg =0.8; µhmax=0.127; µamax=0.02; koh=0.2, koa=0.4; ks=130; kno=0.9; knh=1; β =0.015. advances in technology innovation, vol. 3, no. 2, 2018, pp. 59 69 copyright © taeti 63 3. control strategy the aeration system, the b lower and p iping model and th e proportional–integral–derivative (pid) control of the air flow is presented in [18-19]. following this work, this paper investigates the performance of a mod ified variant of the we ll-known generalized pred ictive control (gpc) method to control the dissolved oxygen concentration (do) in the aerated bioreactor of an activated sludge process , considered as process output variable (fig. 2). we assume that the appropriate set point for the dissolved o xygen concentration is given and the predict ive control system is used to maintain this set point. the aeration a ir flow (w) is considered as manipulated variable. fig. 2 predictive control schemes the generalized predictive controller is one of the most relevant design methods of model-based pred ictive control (mbpc). the standard gpc synthesis is based on a linear p rocess model, carima controlled auto-regressive integrated moving-average, a quadratic cost function and a control law, both using an incremental structure (the actual control signal incre ment δuis co mputed) [14]. th is incre mental imple mentation ensures offset -free behavior in closed loop control systems. the activated sludge process is the most energy consuming process es in the whole wwtp, nearly half of the energy consumed in a wwtp being used for the aeration. an important step in developing the proposed control strategy is the repara metrizat ion of the cost function in the pred ictive a lgorith m to contain a measure of energy consumed by ae ration process. this could exploit the fluctuation of operating conditions by realizing significant energy savings. since the aeration air fl ow w is the manipulated variable resulted fro m the con troller output u, to minimize the aeration flow and not its variations the repara metrized cost function of the predictive algorithm has to contain the controller output u, instead of the control output incre ment δu. this will lead to a positional imp le mentation based on a positional form for the process model and controller cost function. consider the following modified mbpc cost function: 2 1 2 2 1 2 1 ( , , ) [ ( ) ( )] [ [ ( 1)] unn u r j n j j n n n e y t j y t j u t j                   (15) where: y r is the future refe rence sequence, n1 is the minimu m costing horizon, n2 is the ma ximu m costing horizon, nu is the control horizon,and ρ is a control-weighting coefficient. the use of the carma process model instead of the carima model : 1 1 1 ( ) ( ) ( ) ( ) ( ) ( )a q y t b q u t k c q e t       (16) will lead to a positional form for the controller and therefore the controller will not have an integrator. for simp lic ity, in this development c (q -1 )=1 is chosen. to derive a j step ahead predictor of the process output y(t+j), on considers the polynomial identity:` advances in technology innovation, vol. 3, no. 2, 2018, pp. 59 69 copyright © taeti 64 1 1 1 1 ( ) ( ) ( ) j j j e q a q q f q       (17) where ej(q -1 ) and fj(q -1 ) are polynomia ls uniquely defined, given a(q -1 ) and the prediction interval j, o f degree j and respectively n (n the process order). based on eq. (16) and eq. (17) we obtain: 1 1 1 1 ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) j j j y t j e q b q u t j k f q y t e q e t j            (18) the optimal predictor, given measured data up to time t (including t) is written as: 1 1 1 ( | ) ( ) ( ) ( ) ( ) ( ) ( ) j j j y t j t g q u t j k f q y t e q e t j           (19) where )()()( 111   qbqeqg jj (20) for simplic ity, in the derivation below, n1 is set to 1, n2 to n, nu to n and k to 1. for j = 1,...,n, the optima l predictor eq. (18) can be written:: 1 0 1 0 1 ( ) ( ) [ ( ) ] ( ) ( ) ( ) ( ) 1 1 y t 1 g u t g q g u t f y t e q e t 1           (21) 1 1 1 1 0 2 1 0 2 ( ) ( ) ( ) [ ( ) ] ( 1) ( ) ( ) ( ) ( ) 1 2 y t 2 g u t g u t 1 g q q g g u t f q y t e q e t 2                  (22) 1 ( 1) 1 0 1 ( ) ( ) ( ) [ ( ) ... ] ( 1) ( ) ( ) ( ) ( ) n n -1 0 n n 1 n n y t n g u t ... g u t n 1 g q q g g u t n f q y t e q e t n                        (23) on observe that the predictor y (t+j), consists of three terms: one including the past known control actions and the filtered measured process outputs, the second depending on future control actions which must be determined and the third, depending on the future noise signals. let f( t+j) be the component of y (t+j), which includes all the known terms at a timely moment: 1 1 0 1 1 ( ) ( )f(t ) [g (q ) g ]u t f y t      (24) 1 1 1 2 1 0 2 ( 2) [ ( ) ] ( 1) ( ) ( )f t g q q g g u t f q y t          (25) 1 ( 1) 1 0 ( ) [ ( ) ... ] ( 1) ( ) ( ) n 1 n n n f t n g q q g g u t n f q y t              (26) then eq. (19) can be rewritten in the vectorial form: efugy  (27) where y, u, f and e are vectors of the form: [ ( 1), ..., ( )] , x 1 t y y t y t n n    (28) [ ( ), ..., ( 1)] , x 1 t u u t u t n n   (29) [ ( 1), ..., ( )] , x 1 t f f t f t n n    (30) 1 1 1 [ ( ) ( 1), ..., ( ) ( )] , x 1 t n e e q e t e q e t n n      (31) and the matrix g is then lower triangular of dimension n x n : advances in technology innovation, vol. 3, no. 2, 2018, pp. 59 69 copyright © taeti 65                   03n2n1n 012 01 0 g...ggg ............... 0...ggg 0...0gg 0...00g g (32) for n u < n the matrix g is then of dimension n x n u:                   nun g...ggg ............... 0...ggg 0...0gg 0...00g g 3n2n1n 012 01 0 (33) and y, u, f and e are vectors of the form: [ ( 1), ..., ( )] , x1 t y y t y t n n   (34) [ ( ), ..., ( 1)] , x1 t u u u u t u t n n   (35) [ ( 1), ..., ( )] , x1 t f f t f t n n   (36) 1 1 1 [ ( ) ( 1), ..., ( ) ( )] , x1 t n e e q e t e q e t n n      (37) the cost function becomes: ( ) [( ( ) ] [( ( ) ]) ) t tt t u r r r r j 1, n, n = e y y + u = e gu + f +e gu + f +e + uy y y yu u  (38) assuming that e[e t u]=0, e[e]=0, and e[e t e] is not affected by u, the first derivative of the previous equation gives : [ ( ) ] [( ) ( )] t t t r r j = 2e gu + f + e + iu = 2e g + i u + f y yg g g u     (39) for the first derivatal ive equate zero the control vector u is obtained: 1 ( ) ( ) t t r u g g i g y f     (40) only the first element of u vector, u(t), must be determined and this value represents the current controller output: )()( fytu r t  (41) where α t = [α 1... α n] is the first row of the (g t g+ρi) -1 g t matrix. note that if the equation for the calculation of the controller output (u) has the same form as that for the incremental gpc a lgorith m (δu), the diophantine equation form and calculat ion of polynomials ej and gj is different. 4. simulation results the assessment of the developed control system is done through numerica l simulation in matlab/sim ulink environment. the nonlinear model of the activated sludge wastewater treatment process given by eqs. (1)-(14) was used to simu late the process dynamics. in the gpc a lgorith m the pred iction of the process output is based on a linear process model. to obtain the linear state space model and the transfer function fro m w to do the model was linearized a roun d an operating point. an eighth order transfer function was obtained. to reduce its order to two, from the linear state space model a balanc ed advances in technology innovation, vol. 3, no. 2, 2018, pp. 59 69 copyright © taeti 66 state-space realizat ion was first computed and then the smallest 6 diagonal entries of the balanced grammians were e liminated using modred. simila r results were obtained using the input and output data obtained during simulat ions of the nonlinear model dynamics for sma ll variations around the considered operating point and a recursive least square algorithm to estimate a second order discrete transfer function. the steady state values of the input variables are: do=0.064[h -1 ]; doin0=0.5[mg/l]; sin0=765[mg/l]; w 0=100[m 3 /h] and r0=0.8. the steady state value for the considered process output is do0 =1.36 mg/ l. the aeration air flo w (w) is considered as the manipulated input, the other inputs being considered as disturbances . the air flow values were limited between w min = 50m 3 /h and w max =210m 3 /h. the controller design para meters are : n=6, nu=1, sa mpling period ts= 0.01 h. diffe rent aspects, such as setpoint changes and effects of load disturbances , have been analyzed. in fig. 3 the setpoint tracking for a step fro m do0 =1.36 mg/l to do1 = 2mg/l in do setpoint at the time mo ment t=1h is presented. all the process inputs excepting the manipulated input have been considered as constants and equal to their steady state values. using the original gpc control a lgorith m that has an incremental form there is no steady s tate error between the process output (continuous line) and the setpoint. the developed gpc control algorith m has a positional form and the steady state error is increasing with the value o f the control -we ighting coeffic ient, ρ( dotted line). however, it can be observed that with the increasing value of this coefficient, the aeration air flow w is decreasing and also the total a mount of air consume d to reach the new setpoint. fig. 3 setpoint tracking performances for process output (do) and contro l output (w). incre mental gpc continuous line, positional gpc for different values of the control-weighting coefficient (ρ) dotted line for the ne xt simu lation scenarios a constant setpoint is considered and the regulatory performance during a simu lation test when the disturbances presented in fig. 4 (a step disturbance with an a mp litude equal to 10% of the steady state value sin0, fro m t=10h to t=20h) and fig. 5 (rando m d isturbances for a large simu lation time) are applied on the most significant input for the process output: influent organic matter concentration ssin. fig. 4 regulatory performances. step ssin input disturbances fig. 5 regulatory performances. random ssin input disturbances 1 1.2 1.4 1.6 1.8 2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2 t [ h ] d o [ m g / l ] ro=0.002 ro=0.003 ro=0.004 incremental gpc positional gpc, ro=0.001 1 1.2 1.4 1.6 1.8 2 100 120 140 160 180 200 t [ h ] w [ m 3 / h ] incremental gpc positional gpc, ro=0.001 ro=0.002 ro=0.004 ro=0.003 advances in technology innovation, vol. 3, no. 2, 2018, pp. 59 69 copyright © taeti 67 fig. 6 shows the regulatory performance during simulat ion test when the disturbances presented in fig. 4 a re applied. using the incre mental gpc control algorith m, there is no steady state error between the process output (continuous line) and the setpoint. using the positional gpc control algorith m the steady state error in the process output is increasing with the value of the control-weighting coefficient, ρ (only the case ρ=0.002 is shown by dotted line in the figure). with the increasing value of this coefficient, the aeration air flo w w is decreasing and also the total amount of air consumed to reject the disturbance. as can be seen in the figure representing the control output w, the surplus of air needed to reject the disturbance using a positional gpc control a lgorith m (dotted line) represents 72% of the surplus of air needed to reject the disturbance using incremental gpc (continuous line). of course it needs to consider the disadvantage of the steady state error and of the response time. fig. 6 regulatory performances fo r step disturbances presented in fig . 4, process ,output (do) and control output (w). incremental gpc continuous line, positional gpc with ρ=0.002 dotted line, no control dashed line. fig. 7 shows the regulatory performance during simulat ion test when the disturbances presented in fig. 5 a re applied. three cases are presented: (i) control using positional gpc (dotted line ), (ii) control using incre mental gpc (continuous line) and (iii) no control (dashed line). the do setpoint is fixed at 2 mg/l and is kept constant during the simu lation. the disadvantage of using the positional gpc is the steady state error and the advantage is a low power consumption. therefore, choosing the control-weighting coefficient value will be based on compromise between performance and power consumption. in the case of fig. 7, the average amount of air necessary for aeration is 2647 cubic meters daily if incre mental gpc control is used. positional gpc with a value ρ=0.002 leads to an average amount of air necessary for aeration of 2348 cubic meters daily. considering the percentage, this means 88% of the average a mount of air necessary for aeration using incremental gpc, i.e. a 12% reduction in a ir flo w. the blo wers operate under a predictable set of laws concerning speed, power and pressure. in accordance with affinity la ws, flow is proportional to motor speed; and power is proportional to the cube of motor speed. thi s means that already min ima l reductions in blower air flow can provide savings in en ergy consumption. reducing the blower air flow by 12% decreases the power requirement by 32%. fig. 7 regulatory performances for load disturbances presented in fig . 5, process output (do) and control output (w). incremental gpc continuous line, positional gpc with ρ=0.002 dotted line, no control dashed line. 10 15 20 25 30 100 110 120 t [ h ] w [ m 3 / h ] no control incremental gpc positional gpc, ro=0.002 10 15 20 25 30 0.7 0.8 0.9 1 1.1 1.2 1.3 1.4 t [ h ] d o [ m g / l ] positional gpc, ro=0.001 no control incremental gpc 0 200 400 600 800 1000 1200 1 2 3 4 5 d o [ m g / l ] t [ h ] incremental gpc no control positional gpc, ro=0.002 0 200 400 600 800 1000 1200 60 80 100 120 140 w [ m 3 / h ] t [ h ] no control incremental gpc positional gpc, ro=0.002 advances in technology innovation, vol. 3, no. 2, 2018, pp. 59 69 copyright © taeti 68 5. conclusions the wastewater treatment plants are considered comple x processes due to the strong nonlinearit ies, large variable time constants and continuous perturbations present in the influent. this study evaluates the performance of a positional gpc control algorith m for the dissolved oxygen concentration in the activated sludge process of a wwtp. both the setpoint tracking and the regulatory performances have been tested and compared with those obtained using the incremental gpc. the design parameters for both controllers are the same and the simulations provide information on the compromise between control performances (steady state error and response time ) and savings in energy consumption. the aerat ion flo w and, by default, the b lower’s speed is allowed to be lowered when the operatin g conditions of the wwtp permit. the power consumed by blowers is proportional to the cube of air flo w. this means that already min ima l reductions in blo wer a ir flow can provide savings in energy consumption. since the presented control system is responsible for forc ing the plant to follow the setpoint prescribed by the optimizing part of a mult ilayer hiera rchica l control structure, it re ma ins to be analyzed in what degree the overall performances of the hierarchical control system will be affected by the steady state error of this control loop. acknowledgement the support of the ro manian national authority for scientific research, uefiscdi, under grant caseau 274/ 2014 and pn-iii-p2-2.1-ci-2017-0202 is gratefully acknowledged. references [1] u. jeppsson, “modelling aspects of wastewater treatment processes,” ph.d. thesis, dept. of industrial electrica l eng. and automation, lund university, sweden, 1996. [2] d. brjdanovic, s. c. f. meijer, c. m. lopez-va zquez, c. m. hooijmans, m. c. m. van loosdrecht, “applications of activated sludge models,” london:iwa publishing, 2015. [3] r. katebi, m. a. johnson, and j. w ilkie, control and instrumentation for wastewater treat ment plants, springer, london, 2012. [4] m. a. brdys , m. grochowski, t. gminski, k. konarcza k, and m. dre wa, “ hierarch ical predict ive control o f integrated wastewater treatment systems,” control engineering practice, vol. 16, no. 6, pp. 751-767, june 2008. [5] n. s. iordache, c. petrescu, g. necula, and busuioc, “municipal wastewater treat ment improve ment using computer simu lating, advances in waste management,” proc. the 4th wseas international conf. waste management, water pollution, air pollution, indoor climate (wwai’10), may 2010, pp. 95-100. [6] i. naşcu, g. vlad, s. folea , and t. buzdugan, “deve lopment and application of a pid auto -tuning method to a wastewater treatment process,” 2008 ieee international conf. automation, qua lity and testing, robotics , ieee press, august 2008, pp. 229-234. [7] j. kocijan, n. hvala, and s. strmčn ik, “multi-model control of wastewater treat ment reactor,” system and control: theory and applications , pp. 49-54, 2000. [8] g. vlad, r. crişa, b. mureşan, i. naşcu, and c. dǎrab, “ develop ment and application o f a predict ive adaptive controlle r to a wastewater treatment process,” 2010 ieee international conf. automation, ieee press, july 2010, pp. 219-224. [9] s. m irghasemi, c. j. b. macnab, a. chu, “ dissolved oxygen control of activated sludge biorectors using neural-adaptive control,” proc. ieee symp. computational intelligence in control and automation, dec. 2014. pp. 1-6. [10] b. ho lenda, e. do mo kos, a. redey, and j. faza kas, “dissolved oxygen control of the activated sludge wastewater treatment process using model p redictive control,” co mputers and chemica l engineering, vol. 32, no. 6, pp. 1270-1278, june 2008. [11] i. naşcu and i. naşcu, “modelling and optimizat ion of an activated sludge wastewater treat ment process ,” european sy mp. computer aided process engineering , ieee press, december 2016, pp. 1159-1164. https://www.google.ro/search?newwindow=1&sa=n&hl=ro&biw=1476&bih=770&tbm=bks&tbm=bks&q=inauthor:%22reza+katebi%22&ved=0ahukewijnzr89rxjahwh_q4khcbbcreq9agiajaj https://www.google.ro/search?newwindow=1&sa=n&hl=ro&biw=1476&bih=770&tbm=bks&tbm=bks&q=inauthor:%22jacqueline+wilkie%22&ved=0ahukewijnzr89rxjahwh_q4khcbbcreq9agibdaj http://www.sciencedirect.com/science/article/pii/b9780444634283501983 advances in technology innovation, vol. 3, no. 2, 2018, pp. 59 69 copyright © taeti 69 [12] c. belch ior, r. a raújo, and j. landeck, “ dissolved oxygen control of the activated sludge wastewater treat ment process using stable adaptive fuzzy control,” computers and chemical engineering, vol. 37, pp. 152-162, february 2012. [13] g. harja, i. nascu, c. muresan, and i. nascu, “improve ments in dissolved oxygen control of an activated sludge wastewater treatment process,”circuits systems and signal processing, vol. 35, no. 6, pp. 2259-2281, june 2016. [14] d. w. clarke, c. mohtadi, and p. tuffs, “ generalized predictive control part 1. the basic algorith m,” automatica, vo l. 23, no. 2, pp. 137-148, march 1987. [15] w. gu jer, m. henze, t . m ino, and m . van loosdrecht, activated sludge models asm1, asm2, asm 2d and asm 3, iwa publishing, 2000. [16] s. cristescu, i. naşcu, and i. naşcu, “sensitivity analysis of an activated sludge model for a qastewater treat ment plant,” proc. international conf. system theory, control and computing, pp. 595-600, october 2015. [17] i. muntean, r. both, r. crisan, and i. nascu, “ rga analysis and decentralized control for a wastewater treatment plant,” proc. the ieee international conf. industrial technology, march 2015, pp. 453-458. [18] g. ha rja , c. mureşan, g. vlad and i. naşcu, “ fract ional order pi control strategy on an activated slud ge wastewater treatment process,” 17th international conf. system theory, control and co mpu ting (icst cc), ieee press, october 2015. [19] g. harja, g. vlad, and i. naşcu. “dissolved oxygen control strategy for an activated sludge wastewater treatment process ,” recent advances in electrical engineering series. proc. the 19th international conf. systems (cscc'15), ju ly 2015 pp. 453-458. http://ieeexplore.ieee.org/search/searchresult.jsp?searchwithin=%22authors%22:.qt.grigore%20vlad.qt.&newsearch=true  advances in technology innovation, vol. 5, no. 1, 2020, pp. 18-32 frame-synchronous blind audio watermarking for tamper proofing and self-recovery hwai-tsu hu * , ying-hsiang lu department of electronic engineering, national i-lan university, yilan, taiwan received 25april 2019; received in revised form 28 may 2019; accepted 22 august 2019 doi: https://doi.org/10.46604/aiti.2020.4138 abstract this paper presents a lifting wavelet transform (lwt)-based blind audio watermarking scheme designed for tampering detection and self-recovery. following 3-level lwt decomposition of a host audio, the coefficients in selected subbands are first partitioned into frames for watermarking. to suit different purposes of the watermarking applications, binary information is packed into two groups: frame-related data are embedded in the approximation subband using rational dither modulation; the source-channel coded bit sequence of the host audio is hidden inside the 2 nd and 3 rd -detail subbands using 2 n -ary adaptive quantization index modulation. the frame-related data consists of a synchronization code used for frame alignment and a composite message gathered from four adjacent frames for content authentication. to endow the proposed watermarking scheme with a self-recovering capability, we resort to hashing comparison to identify tampered frames and adopt a reed–solomon code to correct symbol errors. the experiment results indicate that the proposed watermarking scheme can accurately locate and recover the tampered regions of the audio signal. the incorporation of the frame synchronization mechanism enables the proposed scheme to resist against cropping and replacement attacks, all of which were unsolvable by previous watermarking schemes. furthermore, as revealed by the perceptual evaluation of audio quality measures, the quality degradation caused by watermark embedding is merely minor. with all the aforementioned merits, the proposed scheme can find various applications for ownership protection and content authentication. keywords: blind audio watermarking, lifting wavelet transform, 2 n -ary adaptive quantization modulation, rational dither modulation, tamper proofing, self-recovery 1. introduction in the age of cloud sharing and mobile access, digital resources (such as speech, image, audio and video files) on the internet keep increasing dramatically in recent years. ironically, owing to the availability of convenient computer software, tampering multimedia data is also rampant nowadays. protection against intellectual property infringement thus becomes an important issue. digital watermarking is considered a promising countermeasure to cope with this issue [1-2]. digital watermarks can be embedded in noise-tolerant multimedia signals to fulfill the goals of content authentication, copyright protection, covert communication, etc. based on the information required for extraction, watermarking schemes can be divided into non-blind and blind categories. non-blind schemes require the original image and/or watermark for extraction, whereas blind schemes require neither. depending on the application scenario, audio watermarks can also be classified as robust or fragile. robust watermarking is meant to be resilient to modification attempts, whereas fragile watermarking makes the embedded information sensitive to any modifications [3]. among the audio watermarking schemes developed for content authentication, the early purpose of the embedded watermark was focused on the detection and localization of tampered area [4-6]. * corresponding author. e-mail address: hthu@niu.edu.tw tel.: +886-3-9317343; fax: +886-3-9369507 advances in technology innovation, vol. 5, no. 1, 2020, pp. 18-32 19 a potential application of fragile watermarking in the field of audio processing is the self-recovery technique, which embeds a watermark into the audio itself to combat the tampering situations. the embedded watermark is often a compressed version of the original content generated via data compression and coding schemes. the amount of the watermark that survives the tampering can help the receiver to not only locate the tampering areas but also to recover the lost content with a certain quality. a few self-recovery schemes for image signals have been proposed so far, such as [7-10]. speech signal self-recovery was also attempted in [11-13]; nonetheless, studies of the self-recovery schemes for audio signals were relatively limited. the method proposed in [14] divides the audio into 4 segments and embeds the feature parameters of every segment into the less significant bits (lsbs) of another randomly selected segment. for this method, self-recovery is feasible only if the lsbs are completely retrievable. by contrast, the method in [15] embeds the control bits for self-recovery in the integer discrete cosine transform (intdct) domain and then employs a compressive sensing technique to retrieve the tampered intdct coefficients. although this method is capable of recovering the audio signal tampered by content replacement attacks, it can only restore the attacked signals up to 0.6 % with acceptable quality. furthermore, the size of the replaced segment must remain identical to enable tampering detection and signal recovery. in fact, it is more common to encounter a situation where the replaced segment holds a different size. the method in [16] attempts to solve such a size discrepancy problem using a synchronization strategy. however, the adopted synchronization strategy appears oversimplified. when the length of the received audio signal is shorter than that of the original, it simply adds a set of zeros at the end of the audio instead of aligning the signal back to the correct position. one common drawback of the foregoing audio watermarking schemes for self-recovery is that they all lack the countermeasures to cope with cropping and/or time-shifting attacks. a minor time mismatch can disrupt the watermark extraction for subsequent self-recovery. motivated by the work done in [14-16], we propose an efficient blind audio watermarking scheme that is capable of achieving tamper proofing and self-recovery in the presence of arbitrary content replacement attacks. the remainder of this paper is organized as follows. section 2 presents two watermarking schemes designed for attaining frame-synchronous blind audio watermarking in the lifting wavelet domain. section 3 outlines the procedures used in watermark embedding and extraction. the framework for self-recovery is discussed in section 4. section 5 evaluates the proposed scheme in terms of imperceptibility, temper proofing, self-recovery, and processing time. in order to illustrate the advantages of the proposed scheme more clearly, section 5 also provides a comparative evaluation between the proposed self-recovery scheme and the one in [16]. finally, conclusions are given in section 6. 2. lwt-based watermarking schemes among the transforms used to perform audio watermarking, dwt appears to be the most popular due to its perfect reconstruction and good multi-resolution characteristics. in particular, many dwt-based schemes take advantage of quantization index modulation (qim) [17] to achieve effective watermark embedding. to reduce computational and memory overhead, we adopted a lifting scheme to implement the dwt in this study. a lifting wavelet transform (lwt) comprises three steps: split, prediction, and update for signal decomposition, and another three steps: update, prediction, and merge are needed for signal reconstruction. the lwt saves computational time and enables frequency localization to overcome the weakness of the traditional wavelet. it is regarded as the second-generation wavelet transform [18]. fig. 1 presents the procedural flow for watermark generation and embedding. as illustrated in the right branch of fig. 1, we first apply a 3-level lwt to decompose a host audio signal into one approximation subband and three detail subbands, each corresponding to a specific frequency range. in particular, the daubechies-8 basis [19] is used as a wavelet function in the process of lwt. note that audio watermarking is preferably implemented in low-frequency subbands with relatively high intensity, as these subbands are more tolerable to signal alteration with less impairment in perceptual quality. theoretically, for audio sampled at 44.1 khz, the 3 rd -level approximation subband spans a frequency range from 0 to 2756 (= 3 22050 / 2 ) hz, advances in technology innovation, vol. 5, no. 1, 2020, pp. 18-32 20 which is suitable for robust watermarking. hence, in our design the approximation subband is reserved for embedding the crucial information including synchronization code, frame index, and hash data derived from the channel-coded bit stream. the 2 nd and 3 rd -level detail subbands are used to hide fragile watermarks that is responsible for data authentication and signal recovery. fig. 1 watermark generation and embedding 2.1. rational dither modulation following the application of lwt to the audio signal, we employ the rational dither modulation [20, 21] to carry out binary embedding in the approximation subband. let (3) ( ) a c n denote the th n coefficient in the -level approximation subband. by referring to the qim [17], the embedding of a binary bit  ( ) 0,1aw n  into (3) ( ) a c n can be formulated as   (3) ( )(3) (3) ( ) ( ) ( ) 1 ˆ ( ) sgn ( ) ( ) 2 2 2 a na a a n a k c n w n c n c n w n                  (1) where  sgn  ,  ,    represent the sign, absolute, and floor functions, respectively. ( )n stands for the step size for quantizing coefficients. the subscript ‘a’ in the symbol ( )aw n implies the targeted approximation subband wherein the watermark bit is embedded and extracted. the use of the magnitude rather than amplitude in eq. (1) aims at excluding the trouble of sign flipping. in accordance with the formulation in eq. (1), the watermarking error, which is defined as the difference between (3) ˆ ( ) a c n and (3) ( ) a c n , can be assumed to have a uniform distribution over ( ) ( ) / 2, / 2 n n     with a variance of 2 ( ) / 12 n  . one of the key features of the rdm lies in the acquisition of ( )n , which is recursively derivable from previous coefficients through      1010 log 1/121/ 2 2 (3) 20 ( ) 1 1 ˆ ( ) 10 repf bl n a i c n i l             (2) 3 rd advances in technology innovation, vol. 5, no. 1, 2020, pp. 18-32 21 where l stands for the length of the involved coefficients. similar to the manner in [20, 21], the embedding strength is adaptively controlled at the maximum tolerable level of the human auditory system [22, 23]. the term   2010 repf b signifies a multiplicative factor for adjusting the embedding strength. ( ) rep f b is the auditory masking threshold in unit of decibel for the bark scale repb . ( ) 0.275 15.025 rep rep f b b      (3) with  representing a clearance gap for imperceptibility. repb can be obtained from a representative frequency repf using the following empirical formula [24]:    1 1 213tan 0.00076 3.5 tan ( / 7500)rep rep repb f f     (4) here we chose the center of the -level approximation subband as the representative frequency, i.e. 3 1 0.5 2 2 s rep f f    (5) where s f is the sampling frequency. overall, eq. (2) jointly takes into account of the psychoacoustic features (i.e., ( ) rep f b ), the quantization error distribution (i.e.,  1010log 1 / 12 ), and the root-mean-square of previously processed coefficients (i.e., (3) ˆ{ ( ) 1, , } a c n i i l  ). the watermark extraction in rdm requires the derivation of quantization step ( )n  based on the same formula presented in eq. (2). subsequent to the acquisition of the 3 rd -level approximation coefficient (3) ( ) a c n , the bit ( )aw n residing in (3) ( ) a c n can be determined by (3) ( ) ( ) ( ) mod 2 0.5 , 2 a a n c n w n             (6) where  mod ,x y denotes the modulo operation, which returns the remainder after the division of x by y . the tilde symbol atop a participating variable implicates the effect due to possible attacks. 2.2 n 2 -ary adaptive quantization index modulation analogous to rdm, the n 2 -ary adaptive quantization index modulation (aqim) modifies the coefficient magnitude according to a n 2 -ary number  ( ) 0,1, , 2 1ndw n   .   ( ) ( ) ( ) ( ) ( ) 1 ˆ ( ) sgn ( ) max 0, ( ) 2 2 2 j dj j d k d d k dn n k c n w n c n c n w n                      (7) where max  denotes the maximum value drawn from a set of data. ( ) ( ) j d c n is the th n coefficient in the th j -level detail subband. the floor function within the above equation may, however, render a negative value that is illegitimate to the definition of a magnitude. in case a negative outcome occurs, we simply replace the negative value with zero. in contrast to the case in rdm, the quantization step k  is computed from the energy of all the coefficients in a frame indexed by an integer k . given that the watermarking errors maintain a power ratio  in decibels, the relationship betwee k  and  can be mathematically expressed as follows: 3 rd advances in technology innovation, vol. 5, no. 1, 2020, pp. 18-32 22       1 2 ( ) 0 10 1 2 ( ) ( ) 0 1 2 ( ) 0 2 1 ( ) 10 1 ˆe ( ) ( ) 1 ( ) / 12 f c f l j d if l j j d d ic l j d if k c i l c i c i l c i l                    (8) where f l is the frame length. cl denotes the number of the coefficients involved in the quantization. basically, c fl l . by referring to eqs. (3) and (4), the value of  can also be estimated as  10( ) 10 log 1/ 12repf b    (9) in [20, 25-27], it was demonstrated that the quantization step size can be adaptively retrieved from a watermarked audio as long as the energy level remains unchanged throughout the watermarking process. the modification on ( ) ( ) j d c n in eq. (7) inevitably cause variations in energy, which makes the retrieved k  different from the one used in watermark embedding. thus, the recovered watermark bits may become inaccurate. this conflict can be settled by first minimizing the overall energy variation of the first c l coefficients and then tuning the other coefficients in the range between 1 c l  and f l . more specifically, we first sort the coefficient magnitudes, termed ( ) ( ) ( ) j i d i l c l  , in descending order: 0 1 1 1 ˆ ˆ ˆ ˆ ˆ( ) ( ) ( ) ( ) ( ) ci i l l l l l l             (10) where i l , which is drawn from  0,1, , 1cl  , signifies the index associated the th i largest magnitude. when applying eq. (7) to the th i l coefficient, the optimal solution 1( )il is ( ) 1 ˆ ˆ( ) ( ) ( ) j i i d i l l c l   (11) and the suboptimal 2 ( ) i l becomes    ( ) 2 ˆ ˆ ˆˆ( ) , if ( ) ( ) & ( ) ( ) ˆ ( ) , otherwise. j i k i d i i k i i k l l c l l l l               (12) in general, coefficients with large magnitudes contribute more variations in energy. to minimize the overall energy variation in a frame, we select between 1 ( ) ' i l s and 2 ( ) ' i l s for the coefficient magnitudes in the top o l ranks.         11 2 2 2 2 0, , 1 0 {1,2} ˆˆ arg min ( ) ( ) ( ) ( ) o i i o o i ll i n i i i i n i l i i l n n l l l l                 (13) subject to the constraint that the accumulated energy must be less than the overall energy, i.e.       1 11 1 1 2 2 2 2 2 ( ) ( ) ( ) 0 0 0 ˆ( ) ( ) ( ) ( ) ( ) f fc c c i o c l ll l l j j j n i i d i d d i i l i i l i l l c l c i c i                   (14) the search for  ˆin in this study is done by a brutal force approach. thus, the required computation is exponentially proportional to o l . this study chooses o l as 8. substituting ˆ ( )in i for the magnitudes in  ˆ ( ) 0 1i ol i l    , i.e. advances in technology innovation, vol. 5, no. 1, 2020, pp. 18-32 23 ˆ ˆ ˆ( ) ( ) ( ) ii i n i l l l      , yields the least energy variation achievable by the o l coefficients. once the magnitude for the th i l coefficient is determined, the corresponding detail coefficients can be modified as follows:    ( ) ( ) ˆ ( ) ˆ ( ) sgn ( ) ; 0,1, , 1; 0,1, , 1 ( ) ij j d i d i c i c i l c l c l i l l l l           (15) where  represents an infinitesimal number added to the denominator to avoid dividing by zero. the violation of the constraint (14) implies that the energy collected from the first c l coefficients exceeds the total amount. it will be impossible to compensate for the excessive portion by regulating the energy over the remaining coefficients, i.e.,  ( ) 1( ) , , , 1jd c c fc i i l l l  . in case the inequality (14) cannot maintain after adjusting the first ol coefficients, we then proceed with the next o l coefficients in the top ranks, i.e.,  ( )ˆ ( ) , , 2 1jd i o oc l i l l  , and rerun the adjustment process. previously altered detail coefficients in the sorted sequence shall remain intact. the adjustment process continues until the constraint shown in eq. (14) is satisfied. finally, to ensure a perfect match with the original energy level, we use the remaining ( )f cl l coefficients to absorb the energy discrepancy       1/ 2 1 1 2 2 ( ) ( ) ( ) ( ) 0 0 1 2 ( ) ˆ( ) ( ) ˆ ( ) ( ) ; , , 1. ( ) f c f c l l j j d d j j i i d d c fl j d i l c i c i c k c k k l l c i                           (16) after completing the watermark embedding, we reconstruct the audio by taking the inverse lwt with respect to all subband coefficients. to retrieve the embedded watermark bits from the watermarked audio, we follow the same steps used in the embedding process. the quantization step size k  can be obtained using eqs. (8). the th i 2 -ary n number, termed ( ) d w i , is determined based on the qim rule: ( ) ( ) ( ) mod 2 0.5 , 2 j d n n d k c i w i             (17) 3. self-recovery framework one of the main features of the proposed watermarking scheme is the self-recovery capability. in order to achieve tamper proofing and self-recovery concurrently, we incorporate the source-channel coding and hashing techniques into the proposed watermarking scheme. the basic idea is to use the frame-partitioned source-channel coded data as the watermark in the embedding phase, and examine the watermark for tamper detection and self-recovery in the extraction phase. in this study, we adopt a mpeg-1 audio layer iii codec (termed mp3 for short) to perform a lossy data-compression of the host audio. in consideration of the limited watermarking capacity, the audio signal is encoded at a very low bitrate of 16 kilobits per second (kbps). this is actually achieved by down-sampling the audio by a factor of 4 and then applying the mp3 codec to convert the audio to a bit stream of 64 kbps. the bit stream is further divided into frames of size 2448 (=21538), which can be regarded as 2 message words, each containing 153 bytes. in this study, reed-solomon (rs) codes on the galois fields 8 (2 )gf [28] are employed to recover the information destroyed by tampering attempts. for each message word, we use a  255, 153 rs code to form an augmented word of length 255. this arrangement enables the rs code to correct 51   255 153 / 2  errors in a row of 255 symbols. in other words, the tolerable tampering rate of the applied rs code is 20% (i.e., 51/255). as a result, the number of bits that were supposed to advances in technology innovation, vol. 5, no. 1, 2020, pp. 18-32 24 embed in each frame is expanded from 2448 to 4080 (=22558). given that the sampling rate of the audio is 44100 hz, this amount of binary bits shall be embedded into a frame with its length no less than 441002448/16000. hence we choose the frame length as 6656 samples and embed 4080 bits into the 3 rd and 2 nd -level detail subbands after performing 3-level lwt decomposition. as shown in fig. 2, the 3 rd and 2 nd -level detail subbands respectively consist of 832 and 1664 coefficients in each frame. we tactically embedded 816 octal numbers (3 bits per coefficient) into the 3 rd -level detail subband and 1632 binary bits (1 bit per coefficient) into 2 nd -level detail subband using the 2 -ary n aqim discussed in section 2, thus rendering a total of 4160 bits to accommodate the need of channel-coded data for audio recovery. the extra 80 bits (= 4160 4080 ) are reserved for the need of file headers. also note that we have applied a distinct 2 -ary n aqim to each detail subband. the use of 8-ary aqim in the 3 rd -level detail subband stems from the consideration that this subband usually contains relatively higher intensity than that found in the 2 nd -level detail subband. according to the formula for 2 -ary n aqim given in eq. (8), a high energy level also leads to a large quantization step that is supposedly more capable of resisting against malicious attacks. 4. procedures for watermark embedding and extraction 4.1. watermark embedding the procedure for watermarks embedding is detailed in the following. step. 1: apply the 64 kbps mp3 codec to a down-sampled audio signal and convert the output file into a bit stream. compose the bit sequence as an array of message words with a size of 153 bytes (or equivalently counted as 153 8-bit symbols). step. 2: append the parity symbols to each message word after applying a  255, 153 rs encoder. step. 3: the symbols are scrambled among words via the use of . 1key . this operation allows the rs code to detect and correct symbol errors in the corrupted word. step. 4: divide the message array into groups, each holding 4 consecutive words. for each group,  record the frame index as a 16-bit integer and encode this integer using a  31, 16 bch encoder [29].  record the total number of frames as a 16-bit integer and encode this integer using a  31, 16 bch encoder.  use . 2key to randomly permute the symbol sequence in each message word.  apply the md5 hash algorithm [30] to the first 153 symbols in each word and draw 16 hash bits from each hashed output to form a composite hash representation of 64 bits.  pack the frame-related information as a bit sequence of length 128, as shown in fig. 2. step. 5: perform 3-level lwt on the host audio step. 6: partition the coefficients in each subband into frames. for a frame length of 6656 audio samples, there are 832 and 1664 coefficients contained in the 3rd and 2nd-level subbands, respectively. step. 7: distribute the channel-coded symbols obtained in steps. 3 and 4 to the audio frames. for each frame,  embed the synchronization code and frame-related information alternately into the 3rd-level approximation subband.  embed 2448 bits into the 3rd-level detail subband using 8-ary aqim.  embed 1632 bits into the 2nd-level detail subband using binary aqim. step. 8: take inverse lwt to obtain the watermarked audio. advances in technology innovation, vol. 5, no. 1, 2020, pp. 18-32 25 fig. 2 arrangement of watermark bits 4.2. watermark extraction fig. 3 outlines the procedure for watermark extraction, tampering detection, and self-recovery. the required steps are outlined as follows: step. 1: conduct 3-level lwt. step. 2: extract the embedded bits from the approximation subband using rdm discussed in section 2. step. 3: apply a matched filter to the extracted bit sequence. the synchronization code in reverse order serves as the filter coefficients. given that  ( ) 0,1n  denotes the synchronization code of length syncl , feeding the extracted ( )aw n into the matched filter results in, fig. 3 block diagram of watermark extraction for tampering detection and audio recovery    1 0 ( ) 2 ( 1 ) 1 2 ( ) 1 syncl sync a i m n l i w n i         (18) ideally, a salient peak of value sync l occurs whenever ( ) a w n perfectly matches with the synchronous code. 255 bytes 255 bytes 255 bytes 255 bytes : : : 255 bytes 255 bytes 255 bytes 255 bytes advances in technology innovation, vol. 5, no. 1, 2020, pp. 18-32 26 step. 4: for each frame,  retrieve the frame-related data and the hash bits from two consecutive frames; acquire the number of total frames and frame index using a  31, 16 bch decoder.  extract 2448 bits from the 3rd-level detail subband using 8-ary aqim.  extract 1632 bits from the 2nd-level detail subband using binary aqim.  rearrange these (2448+1632) bits as two words, each comprising 255 8-bit symbol (1 byte per symbol).  place these two words in the corresponding index entry. step. 5: use . 2key to restore the symbol sequence in each message word. step. 6: use . 1key to restore the original permutation of the symbol array. step. 7: pass the message word to the rs decoder to obtain the source-coded audio symbols. step. 8: generate the hash bits from each word using the md5 hash algorithm and compare these bits with those recorded in the approximation subband. if the hash bits are identical, the symbol sequence is assigned to the location indicated by the frame index. otherwise, the frame is labeled as tampered at the receiver. step. 9: use a mp3 decoder to decompress the audio signal from the extracted watermark bits and up-sample the output by a factor of 4. step. 10: if the audio frame has been tampered, then we substitute the up-sampled audio signal for the tampered audio content. since the rs decoding process is capable of removing 51 (=  255 153 / 2 ) errors in a row of 255 symbols, tampering is recoverable as long as the tampering rate is below 0.2 (=51/255); otherwise, the rs decoder and recovery process fail. 5. performance evaluation the test materials in the following experiments comprised twenty-four 30-second music clips collected from a variety of compact discs, including vocal arrangements and ensembles of musical instruments. the music clips can be classified into four categories: classical (3), pop (7), rock (7), soundtracks (7). all audio signals were sampled at 44.1 khz with 16-bit resolution. the parameters used in the proposed watermarking scheme were set as follows: 2  , 0 8l  , 128 sync l  ; rep f  1378.1, 4134.4, and 8268.8 hz for the 3 rd -level approximation subband, 3 rd -level detail subband, and 2 nd -level detail subband, respectively; the back-tracing length l used in the rdm was set to 416. 816cl  and 832fl  were chosen for the 8-ary aqim in the 3 rd -level detail subband, while 1632 c l  and 1664 f l  were for the binary aqim in the 2 nd -level detail subband. 5.1 imperceptibility test the quality of the watermarked audio signal was evaluated using the snr defined in eq. (19) along with the perceptual evaluation of audio quality (peaq) metric [31].   2 10 2 ( ) 10 log ˆ( ) ( ) n n s n snr s n s n               (19) advances in technology innovation, vol. 5, no. 1, 2020, pp. 18-32 27 where ( )s n and ˆ( )s n denote the original and watermarked audio signals, respectively. the peaq simulates the subjective evaluation of human subjects. it renders an objective difference grade (odg) between -4 and 0, signifying a perceptual impression from “very annoying” to “imperceptible”. in this study, the peaq metric for the imperceptibility test was an implementation released by the tsp lab at mcgill university [32]. table 1 summarizes the experiment results with respect to the test materials. for each audio signal of 30 seconds long, there are over 198 frames of size 6656 can be embedded and at least 99 of them contain the synchronization code. embedding the synchronization codes and frame-related data into the 3 rd -level approximation subband rendered an average snr of 28.34 db, which led to an average odg score around -0.30. the subsequent embedding of the channel coded symbols into the 3 rd and 2 nd -level detail subbands brought the snr to 27.36 db and caused the odg to slightly drop to -0.44. such a result suggests that the proposed watermarking schemes merely have minor influence on perceptual quality. moreover, for all audio files in the test, the embedded synchronization codes were perfectly detected using the matched filter. fig. 4 shows one such example, wherein the peaks with a height of 128 repeatedly appear for every 1664 approximation coefficients. table 1 quality measures of the watermarked audio after applying rdm and 𝟐𝑵-ary aqim to the audio signals quality measure rdm (in 3 rd -level app. subband) n 2 -ary aqim (in 3 rd & 2 nd -level detail subbands) rdm+ n 2 -ary aqim snr [db] mean 28.34 35.59 27.36 standard deviation 0.25 3.92 0.46 odg mean -0.30 -0.22 -0.44 standard deviation 0.38 0.26 0.43 (a) audio siganl (b) output of the matched fliter fig. 4 matched filtering with respect to the watermark bits obtained by the rdm 5.2 tamper detection and recovery a representative audio signal was employed to demonstrate the competence of the proposed scheme for tampering detection and localization. we conducted three types of attacks (namely, deletion, substitution, and insertion) on the audio signal with the self-recovering watermark embedded. the deletion attack cropped the leading 25000 samples of the watermarked audio signal. the substitution attack replaced the watermarked audio signal with zero over the range between 325001 and 375000. for the insertion attack, we appended 50000 samples of random noise at the end of the watermarked audio signal. both the substitution and insertion attacks represent possible attempts on counterfeiting the audio signal. advances in technology innovation, vol. 5, no. 1, 2020, pp. 18-32 28 (a) audio with fragile watermarks embeded (b) tampered audio (c) output of the matched fliter fig. 5 illustration of three types of tampering attack (a) audio with fragile watermarks embeded (b) tampered audio (c) recoverd audio fig. 6 illustration of tampering detection and audio signal recovery fig.5 (a) and (b) respectively present the original and tampered watermarked audio signals. the watermark bits hidden in the 3 rd -level approximation subband are extracted, bipolarized (i.e.,    0,1 1,1  ), and finally fed into a matched filter. fig. advances in technology innovation, vol. 5, no. 1, 2020, pp. 18-32 29 5 (c) depicts the output of the matched filter. a sharp peak with its magnitude greater than a predefined threshold (e.g., 80) can serve as an indicator to demarcate the frame boundary. the tampered signal is then processed using the watermark extraction and self-recovery procedures discussed in sections 2-4. more specifically, subsequent to frame synchronization, the hash bits are used to verify the veracity of individual frame content. as shown in fig. 6(b), a nonzero level (delineated as a bold solid red line) signifies intact frames and a zero level specifies the occurrence of tampering. all the data extracted from the 2 nd and 3 rd -level detail subbands are then employed to reconstruct an mp3 decompressed version of the audio signal. eventually, the lost contents in the tampered frames are replaced by the reconstructed ones, which are drawn in red in fig. 6(c). this typical example demonstrates that our scheme not only accurately locates the tampered audio frames but also possesses a self-recovering capability. 5.3 processing time the proposed self-recovery scheme comprises five basic modules to carry out watermark embedding. the first module involves source-channel encoding and hashing technique jointly used to constitute the bit sequence for audio recovery. as the watermarking is accomplished in three low-to-middle frequency subbands, we need a 3-level lwt and another inverse lwt to decompose and recompose the audio signal. these two transformations are extra computational burdens for the watermarking performed in the lwt domain. the operations situated in between the lwt and ilwt contain two sorts of watermark embedding, namely, the synchronization code sequence in the 3 rd level approximation subband and the channel-coded bit stream in both the 2 nd and 3 rd detail subbands. as for the process of watermark extraction, we only need a 3-level lwt to decompose the watermarked audio. the detection of the synchronization code enables the alignment of frame boundary, which facilitates the watermark retrieval in the 2 nd and 3 rd detail subbands. possible errors in the watermark bits shall be amended with the assistance of the rs channel coder. eventually, the veracity and integrity of the received audio can be authenticated using hashing comparison. table 2 processing time required for each program module in watermark embedding and extraction processes program modules in watermark embedding processing time [sec] mean standard deviation conduct (1) data encoding & hashing (2) bit arrangement 6.377 0.193 perform lwt 0.947 0.009 embed sync_code using rdm 0.780 0.019 embed channel-coded data using aqim 2.852 0.075 perform ilwt 0.954 0.008 overall 11.910 0.258 program modules in watermark extraction processing time [sec] mean standard deviation perform lwt 0.957 0.013 align frames via the detection of sync_code 0.027 0.007 extract watermark (coded data) 0.029 0.004 perform channel-decoding and hashing comparison 1.267 0.037 restore the tampered signal if necessary overall 2.281 + 0.048 + we implemented the proposed watermarking algorithm in a matlab environment operating with a 4 ghz intel(r) core(tm) i7-4790k cpu and 32 gb ram. table 2 lists the average computation time for the twenty-four 30-second audio signals in the test set. in general, it takes 11.91 seconds to complete the watermark embedding for an audio file of 30 second long. among the five modules in the whole process, the data encoding and hashing consume about 53.54% of the computational time. the actual embedding in lwt subbands requires 5.533 seconds in total. compared to the lengthy computation required in the embedding process, the time spent on watermark extraction is greatly reduced while extracting watermark bits from the 2 nd and 3 rd detail subbands using aqim. advances in technology innovation, vol. 5, no. 1, 2020, pp. 18-32 30 5.3 comparative evaluation in order to illustrate the advantages of our proposed scheme more clearly, we make a comparison between ours and the scheme proposed by gomez-ricardez and garcia-hernandez in [16]. the scheme in [16] is chosen for comparison based on the following two similarities. first, just like the manner we have done in this study, it employs a channel coder to protect the watermark. second, this scheme is also claimed to be robust against the content replacement attack if the affecting portion is less than 20% of the whole audio. table 3 summarizes the comparison. the method in [16] is indeed capable of restoring the substituted segment when the size of substitution remains unchanged. restoring the audio segment destroyed by the insertion attack is also possible if the tampered area is accurately located and the whole audio is properly trimmed and aligned. however, dealing with the deletion attack is problematic. for example, deleting a small section of the audio in the middle, then shifting the remaining part ahead, and finally padding zeros at the end can easily cripple the watermark extraction for the scheme in [16]. the cause is ascribable to the fact that the deletion misplaces a large portion of the watermarked audio and thus devastates the channel code information. for the same sake, the scheme in [16] cannot survive the cropping or time-shifting attacks, which are known to disrupt the frame synchronization for correct watermark extraction. by contrast, with the incorporation of the self-synchronization feature discussed in section 2, the proposed scheme can withstand all the aforementioned attacks (i.e., insertion, deletion, substitution, cropping, and time-shifting). table 3 comparison results attack types resistance the proposed scheme in [16] cropping / time shifting yes no deletion yes no substitution yes partially feasible insertion yes partially feasible lsb erasure yes no another advantage of the proposed scheme is that it is quite capable of resisting against minor attacks such as lsb erasure. table 4 presents the extraction results when 1 and 2 lsbs are deliberately obliterated. the results indicate that even in the case of 2 lsbs erasure the proposed scheme can perfectly extract 19 out of 24 embedded watermarks from the 2 nd detail subband. moreover, because the maximum ber (i.e., 1.172%) is less than 20% 1/8, the reed-solomon code capable of correcting 20% erroneous 8-bit symbols is sufficient to recover the original watermark bits. table 4 comparison results # of erasure bits erase 2 lsbs erase 1 lsb embedding location 2 nd detail subband 3 rd detail subband 2 nd detail subband 3 rd detail subband # of watermarks without errors 19 21 22 23 # of watermarks with errors 5 3 2 1 largest ber among the watermarks with errors 1.172% 0.081% 0.223% 0.001% 6. conclusion in this paper, we have proposed a novel watermarking scheme to not only authenticate the veracity and integrity of the received audio but enable the recovery of tampered contents via the exploitation of source-channel coding. after the application of 3-level lwt to the audio signal, the proposed scheme performed two types of watermarking processes in a frame-synchronous manner. a compressed version of the original signal protected with the rs code was embedded into the 3 rd and 2 nd -level detail subbands using 2 -ary n aqim, while the frame-related data and hash bits were embedded into the 3 rd -level approximation subband using rdm. the experiment results indicated that the watermark embedding resulted in an average snr of 27.36 db and an average odg score around -0.44 for a test set of twenty-four audio clips, suggesting that the advances in technology innovation, vol. 5, no. 1, 2020, pp. 18-32 31 watermarked audio is nearly perceptually indistinguishable from the original one. in the phase of watermarking extraction, the rdm proved to be effective in tracking synchronization codes, thus facilitating the frame alignment and watermark extraction. the 2 -ary n aqim also demonstrated its competence in performing multi-bit data hiding in the lwt domain. more importantly, the ability of tracing frame boundaries empowered the proposed scheme to combat with the cropping and replacement attacks that no previous self-recovery watermarking schemes could easily handle. as there is plenty of room for hiding extra information in the 3 rd -level approximation subband, our future work will be focused on adding other robust watermarks to reinforce copyright protection. conflicts of interest the authors declare no conflict of interest. acknowledgment this research work was supported by the ministry of science and technology (most), taiwan, under grant 107-2221-e-197-021. references [1] n. cvejic and t. seppänen, digital audio watermarking techniques and technologies: applications and benchmarks. hershey: information science reference, igi global, 2008. [2] x. he, watermarking in audio: key techniques and technologies, youngstown, n.y.: cambria press, 2008. [3] m. steinebach and j. dittmann, “watermarking-based digital audio data authentication,” eurasip journal on advances in signal processing, vol. 2003, no. 10, pp. 1001-1015, 2003. [4] m. q. fan, p. p. liu, h. x. wang, and h. j. li, “a semi-fragile watermarking scheme for authenticating audio signal based on dual-tree complex wavelet transform and discrete cosine transform,” international journal of computer mathematics, vol. 90, no. 12, pp. 2588-2602, 2013. [5] ghobadi, a. boroujerdizadeh, a. h. yaribakht, and r. karimi, “blind audio watermarking for tamper detection based on lsb,” proc. 2013 15th international conference on advanced communications technology (icact), ieee press, january 2013, pp. 1077-1082. [6] n. n. hurrah, s. a. parah, n. a. loan, j. a. sheikh, m. elhoseny, and k. muhammad, “dual watermarking framework for privacy protection and content authentication of multimedia,” future generation computer systems, vol. 94, pp. 654-673, 2019. [7] h. he, f. chen, h. tai, t. kalker, and j. zhang, “performance analysis of a block-neighborhood-based self-recovery fragile watermarking scheme,” ieee transactions on information forensics and security, vol. 7, no.1, pp. 185-196, 2011. [8] q. han, l. han, e. wang, and j. yang, “dual watermarking for image tamper detection and self-recovery,” 2013 9th international conference on intelligent information hiding and multimedia signal processing, october 2013, pp. 33-36. [9] x. zhang, z. qian, y. ren, and g. feng, “watermarking with flexible self-recovery quality based on compressive sensing and compositive reconstruction,” ieee transactions on information forensics and security, vol. 6, no. 4, pp. 1223-1232, 2011. [10] w. l. tai and z. j. liao, “image self-recovery with watermark self-embedding,” signal processing: image communication, vol. 65, pp. 11-25, july 2018. [11] s. sarreshtedari, m. a. akhaee, and a. abbasfar, “a watermarking method for digital speech self-recovery,” ieee/acm transactions on audio, speech, and language processing (taslp), vol. 23, no.11, pp. 1917-1925, 2015. [12] w. lu, z. chen, l. li, x. cao, j. wei, n. xiong, et al., “watermarking based on compressive sensing for digital speech detection and recovery (†),” sensors, vol. 18, no. 7, pp. 2390, 2018. [13] s. li, z. song, w. lu, d. sun, and j. wei, “parameterization of lsb in self-recovery speech watermarking framework in big data mining,” security and communication networks, 2017. [14] f. chen, h. he, and h. wang, “a fragile watermarking scheme for audio detection and recovery,” 2008 congress on image and signal processing, vol. 5, pp. 135-138, 2008. advances in technology innovation, vol. 5, no. 1, 2020, pp. 18-32 32 [15] menendez-ortiz, c. feregrino-uribe, j. j. garcia-hernandez, and z. j. guzman-zavaleta, “self-recovery scheme for audio restoration after a content replacement attack,” multimedia tools and applications, vol. 76, no. 12, pp. 14197-14224, june 2017. [16] j. j. gomez-ricardez and j. j. garcia-hernandez, “an audio self-recovery scheme that is robust to discordant size content replacement attack,” 2018 ieee 61st international midwest symposium on circuits and systems (mwscas), 2018, pp. 825-828. [17] chen and g. w. wornell, “quantization index modulation: a class of provably good methods for digital watermarking and information embedding,” ieee trans. information theory, vol. 47, no. 4, pp. 1423-1443, 2001. [18] w. sweldens, “the lifting scheme: a custom-design construction of biorthogonal wavelets,” applied and computational harmonic analysis, vol. 3, no. 2, pp. 186-200, 1996. [19] daubechies, ten lectures on wavelets. philadelphia, 1992. [20] h. t. hu and l. y. hsu, “a dwt-based rational dither modulation scheme for effective blind audio watermarking,” circuits, systems, and signal processing, vol. 35, no. 2, pp. 553-572, 2016. [21] h. t. hu and l. y. hsu, “supplementary schemes to enhance the performance of dwt-rdm-based blind audio watermarking,” circuits, systems, and signal processing, vol. 36, no. 5, pp. 1890-1911, 2017. [22] x. he and m. s. scordilis, “an enhanced psychoacoustic model based on the discrete wavelet packet transform,” journal of the franklin institute, vol. 343, no. 7, pp. 738-755, 2006. [23] t. painter and a. spanias, “perceptual coding of digital audio,” proc. ieee, vol. 88, no. 4, pp. 451-515, 2000. [24] h. traunmüller, “analytical expressions for the tonotopic sensory scale,” the journal of the acoustical society of america, vol. 88, no. 1, pp. 97-100, 1990. [25] h. t. hu, l. y. hsu, and h. h. chou, “variable-dimensional vector modulation for perceptual-based dwt blind audio watermarking with adjustable payload capacity,” digital signal processing, vol. 31, pp. 115-123, 2014. [26] h. t. hu and l. y. hsu, “robust, transparent and high-capacity audio watermarking in dct domain,” signal processing, vol. 109, pp. 226-235, 2015. [27] h. hu and t. lee, “high-performance self-synchronous blind audio watermarking in a unified fft framework,” ieee access, vol. 7, pp. 19063-19076, 2019. [28] s. lin and d. j. costello, error control coding, second edition: prentice-hall, inc., 2004. [29] g. forney, jr., “on decoding bch codes,” ieee trans. information theory, vol. 11, no. 4, pp. 549-557, 1965. [30] b. den boer and a. bosselaers, “collisions for the compression function of md5,” workshop on the theory and application of cyptographic, berlin, heidelberg, pp. 293-304, 1994. [31] itu-r recommendation bs.1387, “method for objective measurements of perceived audio quality,” december 1998. [32] p. kabal, “an examination and interpretation of itu-r bs.1387: perceptual evaluation of audio quality,” tsp lab technical report, dept. electrical & computer engineering, mcgill university, 2002. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). 2__aiti#6666__146-156 advances in technology innovation, vol. 6, no. 3, 2021, pp. 146-156 dockless shared bicycle flow control by using kernel density estimation based clustering shang-yuan chen * , tzu-tien chen school of architecture, feng chia university, taichung, taiwan received 31 october 2020; received in revised form 29 january 2021; accepted 03 february 2021 doi: https://doi.org/10.46604/aiti.2021.6666 abstract since dockless sharing bicycles have become an indispensable means of everyday life for urban residents, how to effectively control the supply and demand balance of bikes has become an important issue. this study aims to apply kernel density estimation based (kde-based) clustering analysis and a threshold-based reverse flow incentive mechanism to encourage the users of bicycles to adjust the supply and demand actively. and it takes shanghai jing’an temple and its surroundings as the research area. its practical steps include: (1) compilation and processing of the needed data, (2) application of kde-based clustering, partitioning, and grading, and (3) incentives calculation based on dockless shared bicycle flow control system. the study finds that the generalization function of kde-based clustering can be used to estimate the density value at any point in the study area to support the calculation of the incentive mechanism for bicycle reverse flow. keywords: classification, clustering, unsupervised neural networks learning, r language, choropleth map 1. introduction the promotion of shared bicycles seeks to resolve the problem of short-distance urban transportation, and shared bicycles are an important means of contemporary green transportation [1]. the docking shared bicycles found throughout taiwan are chiefly aimed at tourists. however, unlike dockless shared bicycles, docking shared bicycles do not allow users to obtain and return the bikes at any location, thus docking shared bicycles cannot easily become important means of everyday transportation for urban residents. in addition, although dockless v-bikes made a short-lived appearance in taiwan during 2017, little data concerning v-bikes is available. as a consequence, this study relies on observations and analysis of dockless shared bicycles in shanghai, china to gain a better understanding of dockless shared bicycles and methods of improving their effectiveness, as well as specific methods of making them more acceptable to city residents and a more important means of transportation. after the shared bicycles are introduced on a large scale in a city, the supply and demand problems of bicycles involving different areas often emerge. for instance, at certain important transportation nodes, the use of bicycles frequently increases during peak times, and may cause bicycle shortages. therefore, it is necessary to dispatch bicycles in order to resolve the supply and demand problems. however, the current dispatching aprroach is often too inefficient to be of much use. furthermore, it also imposes logistics costs on bicycle companies because manpower and trucks are required to move bicycles from oversupplied areas to shortage areas. besides, since bicycles cannot be used while they are being transported, the capacity of urban transportation is reduced. as a consequence, how to dispatch shared bicycles in real-time, or in advance, is an especially important issue. * corresponding author. e-mail address: shangyuanc@gmail.com tel.: +886-4-24517250 advances in technology innovation, vol. 6, no. 3, 2021, pp. 146-156 in this study, we apply cluster analysis with unsupervised learning to shared bicycle flow control and dispatching, and employ artificial intelligence to extract the real and full-scale transportation rules from open data. this study’s analysis of the times and locations at which the oversupply or shortage of shared bicycles may occur in shanghai will help bicycle companies to perform dispatching ahead of time. furthermore, clustering results can be used in conjunction with incentive mechanisms employing threshold values to encourage bicycle users to actively perform bicycle dispersal and dispatching, which will boost the carrying capacity of urban transportation systems, and reduce the risk and cost of ineffective dispatching. 2. literature review the scope of this study includes shared bicycles and cluster analysis; the following is a review of the literature: 2.1. shared bicycles shared bicycle systems include docking and dockless models. the world’s first intelligent public bicycle system was the vélo’v docking public bicycle system, which was established in lyon, france, in 2005 [2]. nevertheless, docking shared bicycles have never caught on in china. since the residents of china consider bicycles to be everyday means of transportation, many dockless shared bikes began appearing in china during the second half of 2016. they are not dependent on the fixed and managed parking areas, and can be left in any vacant space along roads. consumers only need to use wechat pay, alipay, or some other apps to pay a specified deposit before using the bikes. the following explains the clustering principles of the docking and dockless shared bicycles, and the differences between the two. (1) docking shared bicycles and their clustering principles: it is much easier and more clear-cut to calculate the number of docking bikes that are taken or returned than dockless bikes. to predict the number of bikes at each group of docks in a city as a whole and after clustering, it is first necessary to perform the grouping of the docking points. in the case of docking shared bikes, because the locations at which bicycles can be obtained or returned are limited to places where docks are present, it is only necessary to perform clustering of the fixed docks, instead of the bikes themselves. if within a specific geographical scope, we divide n docks into m groups (n>m), calculate the probability of bikes moving among the m groups, and analyze their historical data (including time, climate, notable events, and other influencing factors), we can then predict future movement trends [3]. (2) dockless shared bikes and their clustering principles: unlike the case of shared bicycles with fixed docks, the clustering of dockless shared bicycles changes with time, and it is challenging to calculate check-out and check-in probabilities. one feasible approach that can be employed to cluster dockless shared bicycles is to perform clustering of the coordinates of the bicycles’ check-out and check-in locations. however, it is still hard to perform direct clustering due to the fact that check-out points and check-in points have an irregular and macro distribution. moreover, the clustering of dockless shared bicycles must also reflect the precondition that users within each cluster can easily find out usable shared bicycles within a comfortable walking distance. besides, the most suitable walking distance is affected by climatic conditions, vertical distance, and a city’s state of economic development [4]. in the case of shanghai community residents, taking a walking distance of 787m as a dividing point, residents would choose to walk when the distance is shorter than 787m, but would very seldom choose to walk when the distance is longer than 787m [5]. this study, therefore, adopts the round figure of 787m as the scope of the most suitable walking distance; 787m is established as the baseline scope of clusters in the remainder of this paper. since 787m is roughly equivalent to the distance that can be walked in 10 minutes, it is a representative figure of daily walking distance. as shown in fig. 1, in a zone, the distance for users to find a bicycle should not exceed the “optimum walking distance”. the minimum diagonal of the equilateral rectangle is 787m, which corresponds approximately to the cell grid with a side length of 550m, reflecting the unit distance of the longitude and latitude grid in the shanghai area. 147 advances in technology innovation, vol. 6, no. 3, 2021, pp. 146-156 fig. 1 the hypothesis for the cell grid 2.2. cluster analysis as described above, the methods and algorithms used to cluster shared bikes affect the shared bicycle flow prediction and dispatching ability. in machine learning, the use of clustering methods and the derivation of clustering algorithms are referred to as “cluster analysis”. cluster analysis is defined as the division of different objects into homogeneous subgroups [6]. data clustering is ordinarily classified as unsupervised learning. while clustering and classification are easily confused, particularly in data mining, the difference between clustering and classification lies in the fact that clustering is a process of grouping to form samples, where the examples are adjacent in space. in contrast, classification involves attaching labels to samples, where the labels are derived from supervised external standards. in other words, clustering constitutes an unsupervised grouping process, where groups are formed after data has been input naturally [7], but the classification is supervised and involves other external information. unsupervised neural networks can uncover essential characteristics, or that may be overlooked, from data on their own, without any preconceived expectation of the output values, and can, therefore, perform clustering of data [8-9]. different clustering methods of unsupervised learning use mathematical formulas as their algorithms. although cluster analysis has been used to develop various types of algorithms [10-16], those algorithms lack a uniform classification. in terms of classification, this study synthesizes the views of gan et al. (2007) [11] and saxena et al. (2017) [14]. gan created the parsing tree based on the characteristics of clustering algorithms. from the top, this tree lists hard clustering and fuzzy clustering. hard clustering includes partitional and hierarchical approaches. hierarchical approaches may consist of either divisive or agglomerative methods [11]. also, saxena et al. suggested the following three additional categories for applying clustering techniques: distance-based methods, model-based methods, and density-based methods (fig. 2) [14]. according to the nature of the dockless shared bicycle problem, they are often distributed in arbitrary shape in a city. therefore, this study proposes to employ density-based clustering algorithms. most partitioning methods are based on the distance between objects. such methods can find only spherical-shaped clusters and encounter difficulty in discovering clusters of arbitrary shapes. however, the density-based method can be used to filter out the noise or outliers and identify the clusters of arbitrary shapes. there are two types of density-based approaches. one is density function, which is a total mathematical function, e.g. kernel density estimation based (kde-based) clustering, and the other is density-based connectivity, e.g. density-based spatial clustering of applications with noise (dbscan) [17]. 148 advances in technology innovation, vol. 6, no. 3, 2021, pp. 146-156 fig. 2 the clustering method parsing tree comparing the kde-based clustering with dbscan methods, the former offers the advantage of the setting of generalization function, which is convenient to partition and grading and is capable of being the basis for incentive mechanism. however, the latter can only be used to mark those clusters that meet a pre-set density requirement by the control radius ε and minpts quantity. kde-based clustering has long been used for detecting traffic accident hot spots and has proven to be useful in traffic accident analysis [18]. in particular, the ordering capabilities of kde-based clustering help compare the spatial structure to determine or assign the importance of mitigation measures [19-20]. as a consequence, the following research will discuss the introduction of kde-based clustering applied to the dockless shared bicycle flow control system. 3. theory and method as mentioned above, this study proposes a “dockless shared bicycle flow control system” that is an incentive mechanism to encourage the reverse flow of bikes in which kde-based clustering is introduced to partition and grading as the basis of threshold control. the activity of the dynamic bicycle control system is not evenly distributed but varied in intensity and scope following people’s everyday activities. because the system’s most important function is to maintain a balance between supply and demand in clusters with different densities, the incentive mechanism is used to reverse supply and demand imbalances during earlier periods. this solution brings up the question of how to perform partitioning and grading of bicycle check-out points and check-in points. when check-outs and check-ins span different clusters, how should incentives be provided to reverse bicycle movement trends? the following explains kde-based clustering partition and grading and dockless shared bicycle flow control system. 3.1. kde-based clustering partition and grading kde-based clustering analysis over a density surface is based on the technologies of spatial smoothing and spatial interpolation [21-22]. the kernel function analyzes and converts the point distribution of the plane in the area, and infers the spatial density distribution in the area. the spatial density is the property that describes the spatial distribution in an intensive or sparse pattern in quantity. therefore, the point data is mapped from the inseparable 2-dimensional space to the separable 3-dimensional space. since a density value is calculated as an attribute of each grid cell in the sub-region, it is, therefore, possible to represent the spatial distribution employing homogenous and easily comparable areas with choropleth map visualization. this estimated density is used to cover the top of the research domain with a grid cell, and the estimation of density is performed based on each grid cell’s center point (fig. 3). based on defined kernel function and bandwidth, kde provides an estimate of the intensity in each position of the grid cell by moving three-dimensional functions. the defined kernel function weights events within its sphere of influence according to their distance from the point at which the intensity is estimated (fig. 4) [23]. in other words, the nature of the optimization and generalization function is a significant characteristic of this density estimation method, which can perform fitting and generalization of large bodies of complex data following the approximation function and bandwidth, and can possess estimation ability. 149 advances in technology innovation, vol. 6, no. 3, 2021, pp. 146-156 commonly used density functions include uniform, epanechnikov curve, quartic curve, and gaussian curve, etc. (fig. 5). among them, quartic and gaussian are more suitable radial basis function kernels but the quadratic kernel is less computationally intensive than the gaussian kernel and can be used as an alternative when using the gaussian becomes too time-consuming and tedious [20]. after optimization, the system can estimate the density value of any position in the study area. since a density value is calculated as an attribute of each quadrat sub-region, it is, therefore, possible to represent the spatial distribution employing homogenous and easily comparable areas with choropleth map visualization [20]. the general form of a kernel density estimator is expressed as: 21 ( , )1 ( ) n i i d s c f s k h h = =      ∑ (1) where f(s) is the density measured at location s, h is the search radius (bandwidth) of the kde [only events within h are used to estimate f(s)], ci is the observed event point, k[ ] is the weight of event ci at distance d(s, ci) to location s. the kde usually models the so-called kernel function, k, as a function of the ratio between d(s, ci) and h so that the “distance decay effect” can be taken into account in density estimation. the evaluation of kde requires two parameters: the bandwidth h and the kernel function k, which determine the weighting of the points. in this study, we use the quartic kernel function, which is one of the most commonly used features: [ ]2 2 ( , )( , ) 3 1 4 d s cd s c k h h = −          (2) fig. 3 performing density estimation based on the center point of each grid cell [24] fig. 4 weighting (ωi) based on a specific kernel function and bandwidth [25] 150 advances in technology innovation, vol. 6, no. 3, 2021, pp. 146-156 fig. 5 kernel functions in a standard coordinate system [26] 3.2. dockless shared bicycle flow control system it is crucial that shared bicycle companies possess real-time or ahead of time dispatching and management ability. this study proposes a dockless shared bicycle flow control system model and an incentive mechanism for the reverse flow of bikes based on threshold values. the clustering results can be used in conjunction with incentive mechanisms employing threshold values to encourage bicycle users to perform bicycle dispersal and dispatching actively. it boosts the carrying capacity of urban transportation systems, and reduces the risk and cost of ineffective dispatching. as shown in fig. 6, this system considers the number of check-outs within a unit time and a unit area to be the bicycle demand of that area, and the number of check-ins within a unit time and a unit area to be the bicycle supply of that area. it is necessary to use incentives or compulsory rules to control the rate of check-outs and check-ins to ensure that each region has a proper amount of bicycle stock. when the stock is excessively high—higher than the upper threshold—the system triggers the check-out incentive mechanism. when the stock is too low—below the lower limit—the system activates the check-in incentive mechanism. these incentive mechanisms controlling reverse flow based on threshold values can be expected to ensure that the stock can be in a better state than when not using the incentive mechanisms. apart from this, when the system detects that the bicycle density reaches the level of oversupply or shortage, the shared bicycle company have to dispatch trucks to perform transport and force the system back within the normal range [27]. fig. 6 dockless shared bicycle control system model 151 advances in technology innovation, vol. 6, no. 3, 2021, pp. 146-156 for every partition, the most important aspect is maintaining a balance between supply and demand within each cluster in this dynamic system, and incentives must seek to reverse imbalances in supply and demand during the previous period. in this study, we consequently design a check-out incentive formula (referring to eq. (3)) and a check-in incentive formula (referring to eq. (4)), in which x is the check-out rate, y is the check-in rate, �� is the time, and n is an integer. if ����� ≤ ���� at the time ��, then the check-out incentive amount is expressed as: 1 1 2 i n nt t y x − − − (3) where � is a whole number and is the check-out sequence at a time ��. by eq. (3), at the time ��, the incentive for check-out of the first bicycle is maximum, and the impetus for the check-out of each subsequent bike decreases progressively with the increasing root of 1/2 with the increase in the number of bikes checked out: 1/2 1 , 1/2 2 , 1/2 3 , and so on. the incentive for the final bicycle checked-out approaches 0 without limit, but never reaches 0. if ����� ≤ ����at the time � , then the check-in incentive amount is expressed as: 1 1 2 j m mt t x y − − − (4) where j is a whole number and is the check-in sequence at a time � . by eq. (4), at time � , the incentive for check-in of the first bicycle is maximum, and the impetus for the check-in of each subsequent bike decreases progressively with the increasing root of 1/2: 1/2 1 , 1/2 2 , 1/2 3 , and so on. the incentive for the final bicycle checked-in approaches 0 without limit, but never reaches 0. as described above, if a bicycle is checked out from area p at time �� and checked in area q at � , the incentive amount is the sum of the check-in incentive in area p during at time tout and the check-out incentive in area q during at time ���. the total incentive amount can be represented as: 1 1 1 1 2 2 i j n n m mt t t t y x x y sum − − − − − − = + (5) 4. realization and verification this section realizes and tests a dockless shared bicycle flow control system following the preceding theory and method. we employ 101,843 data items issued in august 2016 for the shanghai open data innovative application competition to perform modeling and verification [28]. these data items include order numbers, bicycle serial numbers, user serial numbers, check-out times, check-out longitudes, check-out latitudes, check-out times, check-in longitudes, check-in latitudes, and routes. we use this data to find the optimal clustering methods for shared bikes in shanghai by examining relatively small areas (the area around shanghai’s jing’an temple) and relatively large areas (the part of shanghai within the fourth ring roads—121.75°e-121.125°e, 30.95°n-31.45°n) and also establish the models corresponding to the state of clustering. this control system applies the kde-based clustering analysis to propose an incentive mechanism for the reverse flow of bikes based on partition and threshold to encouraging bicycle users to evacuate and dispatch actively. we employ the following three steps in this process: (1) compilation and processing of the needed data, (2) application of kde-based clustering, partitioning, and grading, and (3) incentives calculation based on dockless shared bicycle flow control system. in the development of the application, the rstudio software is used as an integrated development environment for writing r language programs [29]. 152 advances in technology innovation, vol. 6, no. 3, 2021, pp. 146-156 4.1. stage 1: compilation and processing of data as of august 2016, 223,000 mobike bikes are in use in shanghai [28]. since each mobike bike in china is ridden by 5.4 persons per day [30], we can roughly estimate that there are approximately 36,000,000 use records each month for mobike bikes in shanghai during august 2016. based on the shanghai area’s latitude and longitude, as well as a maximum suitable walking distance of 787m, which corresponds approximately to the cell grid with a side length of 550m, we can estimate that there is an average of approximately 2,857 bikes in each 550m×550m grid square at that latitude and longitude (discarded, damaged, and under-repair bikes were not subtracted). assuming that 50% of bikes in the urban area of shanghai can still be found, we can estimate that there are approximately 1,428 bikes in each grid square. the following modeling, therefore, employs a common stock of 1,428 bikes in each 550m×550m grid square as the reference bicycle stock value for shanghai during august 2016. 550m×550m squares are designated as grid cells for kde-based clustering calculations. 4.2. stage 2: application of kde-based clustering, partitioning, and grading as described above, the activity of the dynamic bicycle control system is not evenly distributed but varied in intensity and scope following people’s everyday activities. because the system’s most important function is to maintain a balance between supply and demand in clusters with different densities, the incentive mechanism is used to reverse supply and demand imbalances during earlier periods. the following is an explanation of how this is done: following the analysis in section 3.2, the kernel density spatial clustering method is used to perform partitioning and grading of check-out and check-in density in the study area around shanghai’s jing’an temple on 8/1/2016. in this study, we set 787m as bandwidth and choose the quartic kernel function to execute kde-based clustering. (1) check-out density partitioning and grading: as shown in table 1, according to the results of calculations, the level 1 area around jing’an temple has a check-out density of approximately 3,600, while the adjacent peripheral area has a check-in density of roughly 900. this number shows that the density difference between the central region and the adjacent outer space is around 3,600�900=2,700. following the depth of the purple color in fig. 7, the system recommends that there be 12 clusters; when 2,700 points are divided among levels 1 to 12, we obtain the clustering thresholds for each level shown in table 1. table 1 check-out clustering thresholds for each level areas with different check-out clustering levels level 1 level 2 level 3 level 4 level 5 level 6 level 7 level 8 level 9 level 10 level 11 level 12 upper threshold ≥3600 3375 3150 2925 2700 2475 2250 2025 1800 1575 1350 1125 lower threshold 3375 3150 2925 2700 2475 2250 2025 1800 1575 1350 1125 ≤900 fig. 7 partitioning of check-out density in the area around shanghai’s jing’an temple on 8/1/2016 fig. 8 partitioning of check-in density in the area around shanghai’s jing’an temple on 8/1/2016 153 advances in technology innovation, vol. 6, no. 3, 2021, pp. 146-156 (2) check-in density partitioning and grading: as shown in table 2, according to the results of calculations, the level 1 area around jing’an temple has a check-in density of approximately 3,100. in contrast, the adjacent peripheral area has a check-in density of roughly 800. this number shows that the density difference between the central region and the adjacent outer space is around 3,100�800=2,300. following the depth of the orange color in fig. 8, the system recommends that there be 11 clusters; when 2,300 points are divided among levels 1 to 11, we obtain the clustering thresholds for each level shown in table 2. table 2 check-in clustering thresholds for each level areas with different check-in clustering levels level 1 level 2 level 3 level 4 level 5 level 6 level 7 level 8 level 9 level 10 level 11 upper threshold ≥3100 2891 2681 2472 2263 2055 1845 1636 1427 1218 1009 lower threshold 2891 2681 2472 2263 2055 1845 1636 1427 1218 1009 ≤800 4.3. stage 3: incentives calculation based on dockless shared bicycle flow control system as shown in figs. 7-8, during the period from 12:00 to 18:00 on 8/1/2016 (i.e., the period following 06:00 to 12:00 on 8/1/2016), the first bicycle is checked-out from the check-out point and returned at the check-in point half an hour later, which is during the same period. the calculations of the check-outs and check-ins are presented as follows. (1) as for the check-outs, the estimated check-out rate x at the check-out point is 1600, and this area is in a level 7 check-in cluster (threshold: 1,845-1,636); since 1600 < 1636 (implying that the check-out rate x is slower than the minimum estimated check-in rate), consequently when x��� ≤ x���� ≤ x���, and x���� < y���. the check-out incentive system should be activated. according to eq. (3), y���� is entered as y���, 1 1636 1600 18 2 − = (6) (2) as for the check-ins, the estimated check-in density y of the check-in point is 2,800; and this area is in a level 2 check-out cluster (threshold: 3,375~3,150); since 2750 < 3150 (implying that the check-in rate y is slower than the minimum estimated check-out rate), when y��� ≤ y�&�� ≤ y���, x��� ≥ y�&��. the check-in incentive system should be activated. according to eq. (4), when x�&�� is entered as x���, 1 3150 2750 200 2 − = (7) (3) the total incentive is the sum of the two amounts. according to eq. (5), a total incentive amount of 218 units is obtained. 18 200 218sum = + = (8) the preceding is a discussion of partitioning and grading and the use of an incentive mechanism relying on threshold values to reverse the flow of bikes. the data for the period from 12:00 to 18:00 on 8/1/2016 obtained from on-site observations is very consistent with the results of our model. because 8/1/2016, monday, is the first working day of the week, it is easy to find shared bikes in the outer marginal area of jing’an temple. but in contrast, the jing’an temple center is located in a dense commercial and office area, so it is often difficult to find bikes that can be borrowed. 154 advances in technology innovation, vol. 6, no. 3, 2021, pp. 146-156 5. conclusions with kde-based clustering, this study can use the choropleth map visualization to represent spatial distribution to achieve partitioning and grading. it can support the calculation of the threshold-based incentive mechanism for bicycle reverse flow. the analysis of the implementation results is as follows: (1) if an area is in a peak usage period, the flow rate is fast (i.e., the pick-up rate is fast, and the return speed is also fast). alternatively vice versa, if the relative time is in the off-peak period. when the absolute value of the difference between the check-out rate and the check-in rate is in a range, the number of bikes in the area is considered to be in a state in which regular operation can be maintained. (2) however, if a domain is in a fast check-out speed for a long time, and the check-in speed is very slow or almost stationary, it means that the number of bikes in the area has been slower than the shortage threshold. conversely, if the check-in rate is rapid for a prolonged period within a specific area, but the check-out rate is almost 0, this indicates that the number of bikes in that area quickly rises above the oversupply threshold. when states of oversupply or shortage occur, trucks should be dispatched to move and redistribute the bikes in order to ensure that the system returns to a healthy state. kde-based clustering has its constraints to establish dispersal and dispatching strategies. as analyzed above, although the incentive mechanism can alleviate the situation of imbalance between supply and demand, the oversupply or shortage of bicycles still cannot be completely ruled out. the dbscan clustering method, is better than kde-based clustering obviously in this situation. therefore, how to use the dbscan cluster analysis to find out the oversupply or shortage of distribution of bikes in the study area to establish dispersal and dispatching strategies will be the core issue of the next research. acknowledgments this work was supported by project: cluster analysis of unsupervised learning applied to dockless shared bicycle traffic prediction and scheduling, most 108-2221-e-035-005. conflicts of interest the authors declare no conflict of interest. references [1] s. k. chang, h. w. chang, and y. w. chen, green transportation, slow, friendly, and sustainable: people-oriented transportation environment makes the city smoother and life better, taipei: neonaturalism, 2013. (in chinese) [2] w. zhu, j. y. he, and d. wang, “methods and empirical research on the distribution of public bicycle systems in france: case study on paris and lyon,” urban planning international, vol. 30, no. s1, pp. 64-70, 2015. [3] y. li, y. zheng, h. zhang, and l. chen, “traffic prediction in a bike-sharing system,” proceedings of the 23rd sigspatial international conference on advances in geographic information systems, no. 33, november 2015, pp. 1-10. [4] j. speck, walkable city: how downtown can save america, one step at a time, new york: north point, november 2012. [5] n. wang and y. c. du, “resident walking distance threshold of community,” transport research, vol. 1, no. 2, pp. 20-24, june 2015. (in chinese) [6] m. negnevitsky, artificial intelligence: a guide to intelligent systems, london: pearson education limited, 2011. [7] j. c. príncipe, n. r. euliano, and w. c. lefebvre, neural and adaptive systems: fundamentals through simulations, new york: wiley, 2000. [8] m. c. su and x. d. zhang, machine learning: neural networks, fuzzy systems, and gene algorithms, 2nd ed. taipei: chuan hwa, 2004. (in chinese) [9] a. d. kulkarni, computer vision and fuzzy-neural system, new jersey: prentice hall ptr, 2001. 155 advances in technology innovation, vol. 6, no. 3, 2021, pp. 146-156 [10] c. c. aggarwal and c. k. reddy, data clustering: algorithms and applications, london: chapman and hall/crc, 2013. [11] g. gan, c. ma, and j. wu, data clustering: theory, algorithms, and applications, usa: society for industrial and applied mathematics, 2007. [12] j. han, m. kamber, and j. pei, data mining: concepts and techniques, 3rd ed. massachusetts: morgan kaufmann, 2012. [13] s. k. popat and m. emmanuel, “review and comparative study of clustering techniques,” international journal of computer science and information technologies, vol. 5, no. 1, pp. 805-812, 2014. [14] a. saxena, m. prasad, a. gupta, n. bharill, o. p. patel, a. tiwari, et al., “a review of clustering techniques and developments,” neurocomputing, vol. 267, pp. 664-681, december 2017. [15] p. n. tan, m. steinbach, and v. kumar, introduction to data mining, 1st ed. london: pearson, pp. 491-493, 2005. [16] s. theodoridis and k. koutroumbas, pattern recognition, 4th ed. usa: elsevier, december 2009. [17] h. shah, k. napanda, and l. d’mello, “density-based clustering algorithms,” international journal of computer sciences and engineering, vol. 3, pp. 54-57, 2015. [18] z. xie and j. yan, “detecting traffic accident clusters with network kernel density estimation and local spatial statistics: an integrated approach,” journal of transport geography, vol. 31, pp. 64-71, july 2013. [19] t. k. anderson, “kernel density estimation and k-means clustering to profile road accident hotspots,” accident analysis & prevention, vol. 41, no. 3, pp. 359-364, may 2009. [20] w. yu and t. ai, “the visualization and analysis of urban facility pois using network kernel density estimation constrained by multi-factors,” boletim de ciências geodésicas, vol. 20, no. 4, pp. 902-926, october 2014. [21] m. c. jones, “variable kernel density estimates, and variable kernel density estimates,” australian journal of statistics, vol. 32, no. 3, pp. 361-371, 1990. [22] t. c. bailey and a. c. gatrell, interactive spatial data analysis, harlow: longman, 1995. [23] a. c. gatrell, t. c. bailey, p. j. diggle, and b. s. rowlingson, “spatial point pattern analysis and its application in geographical epidemiology,” transactions of the institute of british geographers, vol. 21, no.1, pp. 256-274, 1996. [24] t. hart and p. zandbergen, “kernel density estimation and hotspot mapping: examining the influence of interpolation method, grid cell size, and bandwidth on crime forecasting,” policing: an international journal of police strategies & management, vol. 37, no. 2, pp. 305-323, may 2014. [25] g. o. ricardo, “kernel density estimation,” http://csyue.nccu.edu.tw/ch/kernel%20estimation(ref).pdf, august 2019. [26] “kernel (statistics),” https://en.wikipedia.org/wiki/kernel_(statistics), august 2019. [27] s. y. chen and t. t. chen, “application of cluster analysis with unsupervised learning to dockless shared bicycle flow control and dispatching,” computer-aided design and application, vol. 17, no.5, pp. 1067-1083, 2020. [28] “soda open data,” http://shanghai.sodachallenges.com/data.html, august 2019. (in chinese) [29] j. p. lander, r for everyone: advanced analytics and graphics, boston: addison-wesley professional, december 2013. [30] beijing planning design research institute, “2017 sharing bike, and urban development white paper”, http://www.199it.com/archives/581592.html, 2017. (in chinese) copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). 156 advances in technology innovation, vol. 3, no. 4, 2018, pp. 185 194 a differential evolution optimization approach for parameters estimation of truncated and censored failure time data chanan s. syan, geeta ramsoobag * department of mechanical and manufacturing engineering university of the west indies, st. augustine, trinidad, west indies. received 05 june 2017; received in revised form 07 september 2017; accepted 14 september 2017 abstract most practically collected datasets are plagued with issues of incompleteness and inaccuracies which will cause erroneous reliability modeling and poor maintenance decisions. this work outlines an investigation into the use of heuristic techniques to estimate the parameters of a stochastic life distribution by using failure data with truncation and censoring. a maximum likelihood estimation (mle) approach was utilized in which the loglikelihood function is modified to account for the truncation and censoring factors. a differential evolution (de) algorithm developed in matlab r2013a minimizes the negative log likelihood (nll) function and obtains optimum parameters for the 2 parameters (2-p) weibull distribution. results obtained from a series of designed experimental tests generalized the relationship between the increasing levels of truncation and censoring individually on the β and η parameters. the impact of the modified nll technique was examined under cases of left truncated and right censored (ltrc) data through evaluations of the mse metric which were compared to estimations made under the normal nll equation. truncation and censoring percentages were increased from 0% to 50% for testing the modified nll approach. it is clear from the low mse values (error) that this approach is successful at estimating the parameters closer to the true values. this approach was applied to failure data of a gas engine power generator utilized in offshore gas production. the results were compared with those obtained from traditional weibull analysis in the reliasoft weibull/alta package. keywords: heuristics, truncation, censoring, maximum likelihood estimation, reliability analysis 1. introduction maintenance optimization algorithms have now become an indispensable tool that asset intensive organizations utilize to reduce costs and increase profit margins. a significant aspect of this process entails the collection and analysis of lifetime data. weibull analysis and other data fitting methods compose the basis of interpretations for equipment degradation over time. however, a drawback of these analyses is the dependency on accurately collected and complete datasets [1]. in many cases, the quality of failure time data has been a major cause for concern. truncation, censoring or even missing events are a few of the problems with most practically collected data [2]. in many analyses, these factors are unaccounted and risk impacting failure predictions. this further impacts the accuracy of maintenance planning in the optimization stages. the popularity of such data in engineering and other fields such as medical sciences has prompted several areas of research to address the above mentioned issues [3-4]. maximum likelihood estimation (mle) has been ubiquitous in such cases. its adaptability towards treating with truncated and censored data is noted in a number of past publications [5-6]. * corresponding author. e-mail address: chanan.syan@sta.uwi.edu tel.: +1 868 662 2002 ext. 82074 advances in technology innovation, vol. 3, no. 4, 2018, pp. 185 194 copyright © taeti 186 traditionally adopted gradient based methods have been utilized to maximize the likelihood function established. in this regard, the predominance of issues such as trapping at local maxima or minima is noted, leading to failure of convergence in many cases [7]. heuristics have been proposed for a number of years to estimate function parameters with much success [8]. a novel aspect of the work reported in this paper investigates the applicability and accuracy of extending heuristic global optimization techniques towards the likelihood function. this would alleviate the issues that have been mentioned while providing simplicity in dealing with the complex function which traditional techniques are not able to address. section 2 of this paper details the findings of past literature; section 3 defines the methodology which was undertaken in the current study; section 4 outlines the results illustrating the truncation and censoring impact and application to a real case study of an offshore gas power generator. section 5 draws conclusions to date from the results and presents future research work. 2. literature research 2.1. left truncation and right censoring although there are various categories of incomplete datasets, the current work targets three specific classifications namely left truncated, right censored, and left truncated and right censored (ltrc) data. for defining these three (3) classifications, consider failure time data collected from 100 identical power transformers between the period 1980 and 2008. the classification of failure cases can be completely defined by four (4) categories as illustrated in fig. 1. failure case 1 represents the class of right censored data which remains in operation at the end of the observation period of 2008. case 2 is a left truncated observation occurring between installation year and prior to the end of the observation period. case 3 illustrates a combined class of left truncation and right censoring. case 4 represents a transformer which was installed and failed prior to observation and as a result no data is available. case 4 is termed as missing data. fig. 1 classification of truncated and censored data [9] the issue of ltrc data has been explored widely in the fields of medicine and is becoming increasingly popular in engineering applications. perhaps the most popular technique was applied by balkrishnan and mitra [5] and liu [10] which involved parameter estimation through a modified likelihood function for ltrc data. an expectation maximization (em) and newton raphson (nr) algorithm was used for comparison purposes to execute the optimization of the estimated function. mitra [11] used this approach to estimate the parameters of a weibull, gamma and log-normal distribution for simulated ltrc failure times with characteristics of that as defined by hong [2] for power transformers. furthermore, emura and shiu [9] adopted and transformed the maximization approach through the use of a one-step nr algorithm for simplicity. although significant developments have been made drawbacks are apparent in terms of complexity, nonconvergence and local maxima or minima trapping when dealing with gradient decent based optimization techniques [7]. advances in technology innovation, vol. 3, no. 4, 2018, pp. 185 194 copyright © taeti 187 2.2. heuristics optimization techniques differential evolution heuristic techniques are a branch of evolutionary algorithms which perform global optimization searches based on stochastic evolution rather than gradient based decent [12]. for this reason, such techniques remain unaffected by the drawbacks of traditional search techniques. genetic algorithms (ga), particle swarm optimization (pso) and simulated annealing (sa) are amongst some of the more popularly applied algorithms for parameter estimation purposes. the differential evolution (de) algorithm was developed by storn and price in the nineties as a recent advancement in this area [13-14]. de offers the advantage of determining global minima/maxima points despite the selection of initial values, fast convergence and fewer limitations of control parameters [15]. das et al. [16] presented an exhaustive survey on des and their adaptations over the years. furthermore, wang and huang [17] applied de to estimate stochastic parameters illustrating the change in population probability density function (pdf) upon crossover, mutation and selection operations. lobato et al. [18] compared de to sa for the estimation of radiative properties through an inverse problem approach. 2.3. maximum likelihood estimation the mle method as defined by scholz [19] is as follows: consider a random vector of observations x = (x1, x2,…,xn) defined by the joint probability distribution function over the n dimensional euclidean space r n . the likelihood function is defined as given in eq. (1). 𝐿𝑥 (𝜃) = ∏ 𝑓(𝑥𝑖 ; 𝜃1, 𝜃2, … , 𝜃𝑟 ) 𝑤ℎ𝑒𝑟𝑒 𝜃1, 𝜃2, 𝜃3, … , 𝜃𝑟 𝑎𝑟𝑒 𝑡ℎ𝑒 𝑝𝑎𝑟𝑎𝑚𝑒𝑡𝑒𝑟𝑠 𝑜𝑓 𝑡ℎ𝑒 𝑑𝑖𝑠𝑡𝑟𝑖𝑏𝑢𝑡𝑖𝑜𝑛 𝑛 𝑖=1 (1) the pdf for 2-p weibull with θ= (β, η) is given in eq. (2) below. 𝑓(𝑥; 𝛽, 𝜂) = 𝛽 𝜂 ( 𝑥𝑖 𝜂 ) 𝛽−1 𝑒 −( 𝑥𝑖 𝜂 ) 𝛽 (2) substitution of eq. (2) into eq. (1) and taking logs we obtain eq. (3) as shown below. 𝐿𝐿(𝑥𝑖 : 𝛽, 𝜂) = 𝑛 (𝑙𝑜𝑔 𝛽 𝜂 ) + ∑ {(𝛽 − 1) 𝑙𝑜𝑔 ( 𝑥𝑖 𝜂 ) − ( 𝑥𝑖 𝜂 ) 𝛽 } 𝑛𝑖=1 (3) the log-likelihood function is extended to include left truncated and right censored data with truncation indicator, ν, and censoring indicator, 𝛿, as shown in eq. (4) [11]. both indicators are binary in nature [1,0] with 𝜈𝑖 𝑜𝑟 𝛿𝑖 = 0 if the sample element is truncated or censored and 𝜈𝑖 𝑜𝑟 𝛿𝑖 = 1 if it is not. mitra [11] tested the approach using multiple unit systems where in some cases the dataset belonging to a single unit may or may not be truncated or censored. for a truncated unit, the truncation time is given by 𝜏𝑖. 𝐿(𝛽, 𝜂) = ∑ 𝛿 {𝑙𝑜𝑔𝛽 − 𝑙𝑜𝑔𝜂 + (𝛽 − 1)𝑙𝑜𝑔 𝑥𝑖 𝜂 }𝑛𝑖=1 − ( 𝑥𝑖 𝜂 ) 𝛽 + ∑ (1 − 𝜈) ( 𝜏 𝜂 ) 𝛽 𝑛 𝑖=1 (4) although de has been applied extensively towards parameter estimation its extension in the context of dealing with incomplete datasets is yet to be investigated. thus, the main focus of the current work is to analyze the application of the de heuristic global optimization technique for parameter determination purposes as applied to truncated and censored datasets. the modified mle technique will be implemented within the de algorithm to account for the incomplete data. 2.4. mean squared error performance metric past authors have often implemented the mean squared error (mse) metric for determining the levels of success attained through their approaches [2, 9, 11]. mse estimates the error between the predicted and true values as represented in the undermentioned eq. 5. for n estimates, if xîis the vector of estimates and xiis the vector of true values then the mse is calculated as: 𝑀𝑆𝐸 = 1 𝑛 ∑ (𝑋�̂� − 𝑋𝑖 ) 2𝑛 𝑖=1 (5) advances in technology innovation, vol. 3, no. 4, 2018, pp. 185 194 copyright © taeti 188 3. methodology matlab r2013a software was utilized for coding the de algorithm in the current study. the algorithm would refer to the normal and modified nll equations with the aim being to minimize the function and follows the standard flow diagram proposed by storn and price [13]. to effectively test the capability of estimating the parameters with minimum error, data were simulated using reliasoft’s weibull/alta monte carlo simulation (mcs) package. this approach also allowed for comparative testing with both complete and incomplete datasets. complete datasets were firstly simulated with the choice of beta values selected to cover cases of reliability growth (β<1) and reliability degradation (β>1). for many industrial systems a combination of static and rotational equipment typically coexists. thus to properly represent the failure nature for both types, beta values of 0.6, 2.5 and 4.5 were randomly selected. table 1 shows the full composition of the data simulation program. additionally, four sample sizes of 20, 100, 150 and 200 were selected in the current study. in comparison to previous work, this range of sample sizes is smaller. however, this was the choice in the current work since much of the practical data collected from the industry conform to sample sizes within the range of 20-200. thus, to allow for conformity in the practical application section similar sample sizes were selected. since the mcs allowed for different levels of censoring to be included, censored samples corresponding to 5%, 20% and 50% for each weibull parameter vector set and sample size were also simulated. all data generations were then exported, chronologically ordered and then artificially left truncated by omitting the initial number of failures corresponding to the truncation percentage as a function of sample size. a total of 384 tests were performed, and the mse values were calculated for each dataset tested. fig. 3 and fig. 5 illustrate a comparison of the pdf plots for the true and estimated weibull parameters. table 1 dynamics of the simulated datasets used for testing the de algorithm parameter varied # of variations details references weibull distribution parameter vector 3 beta, β=0.6 , eta, η=500 beta, β=2.5, eta, η=2500 beta, β=4.5, eta, η=4000 [20] sample size 4 20, 100,150 and 200 n/a percentage censoring 4 0%,5%, 20% and 50% [5], [7], [11] percentage truncation 4 0%,5%, 20% and 50% [5], [7], [11] total # of datasets 192 total # of tests 384 (normal and modified nll) 4. results 4.1. truncation effect (a) effect of truncation percentages on beta and eta estimates with normal nll (b) effect of truncation percentages on beta and eta estimates with modified nll fig. 2 effect of truncation percentages of beta and eta values 2000 2200 2400 2600 2800 0 1 2 3 0 5 20 50 e ta ( η /h r s) b e ta ( β ) beta20 beta100 beta150 beta200 eta20 eta100 eta150 eta200 advances in technology innovation, vol. 3, no. 4, 2018, pp. 185 194 copyright © taeti 189 (a) effect of increased truncation percentages on the pdf curve with normal nll l (b) effect of increased truncation percentages on the pdf curve with modified nll fig. 3 effect of increased truncation percentages on the pdf curve the individual effect of truncation and censoring percentages was investigated initially. fig. 2 shows the impact when using (a) the normal nll and (b) the modified nll functions for varying percentages of truncation. the true beta and eta values for the data were 2.5 and 2500 hours respectively and the results are displayed for all sample sizes. the estimated values of β and η have surpassed the true values when utilizing the normal nll function. significant improvement can be noted with the use of the modified nll equation in fig. 2(b). the values for beta and eta are estimated much closer to the true values even at increased levels of truncation. the accuracy of the estimates is, however, dependent on the sample size. at larger sample sizes of 150 and 200 the accuracy of estimation was higher as expected. fig. 3 further illustrates the overall effect of left truncation on the weibull pdf, where it can be noted that the truncation percentage increases, the distribution is shifted to the right indicating increased characteristic life. therefore, as the truncation percentage increases, the errors in characteristic life would also increase. larger truncation percentages cause inaccurate magnification in β values resulting in the higher failure rate than the true value, again introducing errors in reliability of the asset. 4.2. the censoring effect figs. 4(a) and 4(b) illustrate the impact of right censoring on both beta and eta parameters for true beta =2.5 and true eta=2500 for all sample sizes. an increase in the beta values and reduction in the eta values were noted with increasing percentages of censoring. increased values of beta indicate a higher failure rate for the asset than the true value while decreasing eta implies a reduction in characteristic life. the censoring impact is noted in the pdf curve as illustrated in fig. 5(a) as it shifts it to the left along the time axis. this reduction in eta decreases the characteristic life and exaggerates the state of unreliability. utilization of the de algorithm has proven effective in nullifying the effects of censoring as demonstrated in fig. 5(b). there is a minimal impact on the pdf curves even at a high percentage of censoring. an added benefit noted with the use of the de algorithm was its ability to enforce repeatability in the results. under different internal parameters, the de algorithm shows no variation in the results, indicating that it is robust. (a) effect of censoring percentages on beta and eta values with normal nll (b) effect of censoring percentages on beta and eta values with modified nll fig. 4 effect of censoring percentages on beta and eta values 0 1000 2000 3000 0 2 4 6 0 5 20 50 e ta ( η /h rs ) b e ta ( β ) beta20 beta100 beta150 beta200 eta20 eta100 eta150 eta200 0 2000 4000 0 1 2 3 0 5 20 50 e ta ( η /h rs ) b e ta ( β ) beta20 beta100 beta150 beta200 eta20 eta100 eta150 eta200 advances in technology innovation, vol. 3, no. 4, 2018, pp. 185 194 copyright © taeti 190 (a) effect of increased censoring percentages on the pdf curve with normal nll (b) effect of increased censoring percentages on the pdf curve with normal nll fig. 5 the effect of increased censoring percentages on the pdf curve. 4.3. left truncation and right censoring (a) mse values for sample size 20 with normal nll equation (b) mse values for sample size 100 with normal nll equation (c) mse values for sample size 150 with normal nll equation (d) mse values for sample size 200 with normal nll equation fig. 6 mse values for three simulated weibull parameter vector sets under varying combinations of truncation and censoring the majority of testing investigated different percentage combinations of ltrc data. comparison of the estimates illustrated the benefits when utilizing the modified nll equation and de algorithm. table 2 summarizes the results of mse for different distribution parameters at increasing combinations of truncation and censoring. although the percentages of truncation were increased from 5% to 50%, the increase in the mse values remained low, indicating that the estimated 0 200 400 600 800 1000 0 ,0 5 ,5 2 0 ,2 0 5 0 ,5 0 0 ,5 0 ,2 0 0 ,5 0 5 ,0 2 0 ,0 5 0 ,0 5 ,2 0 5 ,5 0 2 0 ,5 5 0 ,5 2 0 ,5 0 5 0 ,2 0 m s e percentage truncation/percentage censoring, %t/%c true beta =4.5 and true eta=4000 hrs true beta=2.5 and true eta = 2500 hrs true beta=0.6 and true eta =500 hrs 0 5000 10000 15000 20000 0 ,0 5 ,5 2 0 ,2 0 5 0 ,5 0 0 ,5 0 ,2 0 0 ,5 0 5 ,0 2 0 ,0 5 0 ,0 5 ,2 0 5 ,5 0 2 0 ,5 5 0 ,5 2 0 ,5 0 5 0 ,2 0 m s e percentage truncation/percentage censoring, %t/%c true beta =4.5 and true eta =4000 hrs true beta = 2.5 and true eta = 2500 hrs true beta = 0.6 and true eta = 500 hrs 0 5000 10000 15000 20000 25000 30000 0 ,0 5 ,5 2 0 ,2 0 5 0 ,5 0 0 ,5 0 ,2 0 0 ,5 0 5 ,0 2 0 ,0 5 0 ,0 5 ,2 0 5 ,5 0 2 0 ,5 5 0 ,5 2 0 ,5 0 5 0 ,2 0 m s e percentage truncation/percentage censoring, %t/%c true beta = 4.5 and true eta= 4000 hrs true beta = 2.5 and true eta = 2500 hrs true beta = 0.6 and true eta = 500 hrs 0 5000 10000 15000 20000 25000 30000 35000 0 ,0 5 ,5 2 0 ,2 0 5 0 ,5 0 0 ,5 0 ,2 0 0 ,5 0 5 ,0 2 0 ,0 5 0 ,0 5 ,2 0 5 ,5 0 2 0 ,5 5 0 ,5 2 0 ,5 0 5 0 ,2 0 m s e percentage truncation/percentage censoring, %t/%c true beta = 4.5 and true eta = 4000 hrs true beta = 2.5 and true eta = 2500 hrs true beta = 0.6 and true eta= 500 hrs advances in technology innovation, vol. 3, no. 4, 2018, pp. 185 194 copyright © taeti 191 values are much closer to the true values. a range of the mse from 0.89 to 16,133 was obtained in the case of true β=0.6 and true η=500 under the normal nll while a range from 0.13 to 5.97 was noted under the modified nll and de algorithm for truncation and censoring percentages increasing from 0% to 50%. similar ranges were obtained with the other weibull parameter vector sets as well. the reduction in mse levels were noted for all sample sizes tested. table 2 shows the results for sample size of 100. in the current work, a smaller range of sample sizes was tested compared to past publications. this was done to make the practical and simulated datasets of similar sizes for comparison purposes. table 2 mse values for normal and modified nll equations at sample size 100 for combinations of percentage truncation and censoring (%t/%c) mean squared error (mse) true β=0.6 and true η=500 true β=2.5 and true η=2500 true β=4.5 and true η=4000 percentage truncation/percentage censoring 0/0 5/5 20/20 50/50 0/0 5/5 20/20 50/50 0/0 5/5 20/20 50/50 normal nll 0.89 54.0 1220 16133 0.115 53.7 1221 12599 0.17 53.8 1220.6 5313.7 modified nll 0.13 5.09 5.51 5.97 0.116 4.78 5.39 6.00 0.17 4.78 5.39 6.01 the reduction and stabilization of mse values are noted in fig. 7 (a)-(d). the larger differences which were noted across the weibull distributions at higher percentages of censoring were also minimized. in addition, contrary to fig. 6 which shows a steady increase in the mse for increasing sample sizes, these values have remained consistently lower throughout all sample sizes when the modified nll and de algorithm are implemented. this indicates a much more precise and accurate estimation to the true values. (a) mse values for sample size 20 with modified nll equation (b) mse values for sample size 100 with modified nll equation (c) mse values for sample size 150 with modified nll equation (d) mse values for sample size 200 with modified nll equation fig. 7 mse values for three simulated weibull parameter vector sets under varying combinations of truncation and censoring 0 2 4 6 8 10 12 14 0 ,0 5 ,5 2 0 ,2 0 5 0 ,5 0 0 ,5 0 ,2 0 0 ,5 0 5 ,0 2 0 ,0 5 0 ,0 5 ,2 0 5 ,5 0 2 0 ,5 5 0 ,5 2 0 ,5 0 5 0 ,2 0 m s e percentage truncation/percentage censoring, %t/%c true beta=4.5 and true eta =4000 hrs true beta= 2.5 and true eta=2500 hrs true beta=0.6 and true eta=500 hrs 0 5 10 15 20 0 ,0 5 ,5 2 0 ,2 0 5 0 ,5 0 0 ,5 0 ,2 0 0 ,5 0 5 ,0 2 0 ,0 5 0 ,0 5 ,2 0 5 ,5 0 2 0 ,5 5 0 ,5 2 0 ,5 0 5 0 ,2 0 m s e percentage truncation/percentage censoring, %t/%c true beta=4.5 and true eta=4000 hrs true beta=2.5 and true eta=2500 hrs true beta=0.6 and true eta=500hrs 0 5 10 15 20 25 0 ,0 5 ,5 2 0 ,2 0 5 0 ,5 0 0 ,5 0 ,2 0 0 ,5 0 5 ,0 2 0 ,0 5 0 ,0 5 ,2 0 5 ,5 0 2 0 ,5 5 0 ,5 2 0 ,5 0 5 0 ,2 0 m s e percentage truncation/percentage censoring, %t/%c true beta=4.5 and true eta=4000 hrs true beta=2.5 and true eta=2500hrs true beta=0.6 and true eta=500hrs 0 2 4 6 8 10 12 14 0 ,0 5 ,5 2 0 ,2 0 5 0 ,5 0 0 ,5 0 ,2 0 0 ,5 0 5 ,0 2 0 ,0 5 0 ,0 5 ,2 0 5 ,5 0 2 0 ,5 5 0 ,5 2 0 ,5 0 5 0 ,2 0 m s e percentage truncation/percentage censoring, %t/%c true beta=4.5 and true eta=4000hrs true beta=2.5 and true eta=2500hrs true beta=0.6 and true eta=500hrs advances in technology innovation, vol. 3, no. 4, 2018, pp. 185 194 copyright © taeti 192 4.4. practical application the data used for the practical validation was the failure times relating to an offshore platform gas engine generator. the power generator is a critical asset on the self-sustainable platform since it provides a source of power to many other pieces of equipment. since the failure time data from the organization’s computer maintenance management system (cmms) is incomplete, modified approaches are required to account for these deficiencies. traditional weibull analysis has been performed on the generator by using reliasoft’s weibull/ alta software package. subsequently, the dataset with n=40 was, then, entered into the developed algorithm and tested to determine the estimated parameters of beta and eta which would account for the ltrc effect. fig. 8 illustrates the obtained results and compares it to the traditional weibull analysis. since there is no information on the failures prior to the observable period, the truncation percentage was calculated by using time. the number of failures obtained during an eight-year observable period from 2007 to 2014 was used to estimate the numbers of failures missing over a five-year period. however, it should be noted that this truncation percentage is calculated on the assumption of a constant failure rate. to validate this assumption, a graphical cumulative failures versus time trend test was performed on the data for which the correlation coefficient (r 2 ) was found to be 97%. this allows for reasonable justification of the assumption and calculation of the truncation percentage. the final values noted were 40% left truncation and 5% right censoring. the estimated values of beta and eta were found as 3.32 and 91,058 hours respectively under the de algorithm and modified nll as opposed to 4.13 and 95,747 hours under the traditional approach based on the normal nll equation not considering the truncation and censoring impact. since the percentage of truncation is 40% whilst the censoring impact is only 5% the increases in both beta and eta values are higher than the values obtained under the developed de with modified nll equation. this is similar to the demonstrated cases of truncation impact in the experimental study in section 4.1, where increases in both beta and eta values were noted with higher levels of truncation. fig. 8 pdf comparison plot of traditional weibull testing and the modified approach for a power generator 5. conclusions much of the maintenance databases are plagued with truncated, censored, or missing data, thus the accurate prediction is hampered leading to the poor maintenance decisions. reductions in the mses show that the parameter estimations are much closer to the true values when using the differential evolution (de) algorithm approach for comprehensive range of combinations of ltrc. results from experimentation with the individual effect of increasing percentages of truncation highlighted a general tendency to progressively overestimate the values of both β and η. in the case of censored data, while the β value is increased, the η value decreased with larger percentages of censoring. under the practical application, a difference of 0.81 was noted in the β parameter and 4,689 hours for η. both values were greater under the normal nll (traditional technique) as was illustrated in cases with larger values of truncation (40%) present in the data. both values advances in technology innovation, vol. 3, no. 4, 2018, pp. 185 194 copyright © taeti 193 remained greater than 1 indicating the asset was in a state of reliability degradation. the algorithm has provided a simple and efficient way of correcting such issues. the use of the current approach however requires knowledge of the number of data points missing as illustrated in the practical application which may not always be available. the future work will target further improvement and refinement of the algorithm so that these data points can be estimated more accurately. acronyms mle maximum likelihood estimation de differential evolution nll negative log-likelihood ll log-likelihood mse mean squared error 2-p 2 parameters ltrc left truncated and right censored em expectation maximisation nr newton raphson mcs monte carlo simulation pdf probability density function ga genetic algorithms pso particle swarm optimisation sa simulated annealing cmms computer maintenance management system notations β beta parameter η eta parameter θ parameter vector for 2-p weibull distribution (β, η) n sample size of data points xi data point i from sample x vector of data points ν truncation indicator τ censoring indicator lx(θ) likelihood equation r 2 correlation coefficient references [1] a. h. c. tsang, w. k. yeung, a. k. s. jardine, and b. p. k. leung, “data management for cbm optimization,” journal of quality in maintenance engineering, vol. 12, no. 1, pp. 37-51, 2006. [2] y. hong, “reliability prediction based on complicated data and dynamic data,” ph.d. dissertation, iowa state univ., iowa, 2009. [3] a. carrión, h. solano, m. l. gamiz, and a. debón, “evaluation of the reliability of a water supply network from rightcensored and left-truncated break data,” water resources management, vol. 24, no. 12, pp. 2917-2935, september 2010. [4] m. lopez, m. elisa, r. cao, and i. v. keilegom, “smoothed empirical likelihood confidence intervals for the relative distribution with left‐truncated and right‐censored data,” canadian journal of statistics, vol. 38, no. 3, pp. 453-473, august 2010. [5] n. balakrishnan and d. mitra, “likelihood inference based on left truncated and right censored data from a gamma distribution,” ieee transactions on reliability, ieee press, july 2013, pp. 679-688. [6] b. bhattacharya, d. l. shrestha, and d. p. solomatine, “neural networks in reconstructing missing wave data in sedimentation modelling,” in proc. 30 th congress of international association of hydraulic engineering and research (iahr), ieee press, august 2003, pp. 209-216. [7] n. balakrishnan and d. mitra, “some further issues concerning likelihood inference for left truncated and right censored lognormal data,” communications in statistics-simulation and computation, vol. 43, no. 2, pp. 400-416, may 2013. [8] g. d. maringer, portfolio management & heuristic optimization, 2nd ed. new york: springer us, 2006. [9] t. emura and s. k. shiu, “estimation and model selection for left-truncated and right-censored lifetime data with application to electric power transformers analysis,” communications in statistics-simulation and computation, ieee press, march 2016, pp. 3171-3189. [10] j. liu, “analyzing left-truncated right-censored data with uncertain onset time with parametric models,” ph.d. dissertation, the university of texas school of public health, 2012. advances in technology innovation, vol. 3, no. 4, 2018, pp. 185 194 copyright © taeti 194 [11] d. mitra, “likelihood inference for left truncated and right censored lifetime data,” ph.d. dissertation, mc master univ., 2013. [12] x. l. dong, s. q. liu, t. tao, s. p. li, and k. l. xin, “a comparative study of differential evolution and genetic algorithms for optimizing the design of water distribution systems,” journal of zhejiang university science a, vol. 13, no. 9, pp. 674-686, september 2012. [13] r. storn and k. price, “differential evolution-a simple and efficient heuristic for global optimization over continuous spaces,” journal of global optimization 11, vol. 11, no. 4, pp. 341-359, december 1997. [14] j. tvrdik, “competitive differential evolution and genetic algorithm in ga-ds toolbox,” technical computing prague, humusoft press, 2006, pp. 99-106. [15] d. karaboga and s. ö kdem, “a simple and global optimization algorithm for engineering problems: differential evolution algorithm,” turkish journal of electrical engineering & computer sciences, january 2004, pp. 53-60. [16] d. swagatam, s. s. mullick, and p. n. suganthan, “recent advances in differential evolution-an updated survey,” swarm and evolutionary computation, vol. 27, pp. 1-30, april 2016. [17] w. ling and f. huang, “parameter analysis based on stochastic model for differential evolution algorithm,” applied mathematics and computation, vol. 217, no. 7, pp. 3263-3273, december 2010. [18] f. s. lobato, v. steffen jr, and s. n. j. antônio, “a comparative study of the application of differential evolution and simulated annealing in radiative transfer problems,” journal of the brazilian society of mechanical sciences and engineering, vol. 32, pp. 518-526, december 2010. [19] f. w. scholz, “maximum likelihood estimation,” encyclopedia of statistical sciences, john wiley and sons, inc., 1985. [20] r. b. abernethy, the new weibull handbook, 2nd ed. texas: barringer & associates, 2006.  advances in technology innovation, vol. 2, no. 4, 2017, pp. 113 118 113 securing ad hoc wireless sensor networks under byzantine attacks by implementing non-cryptographic methods shabir-sofi 1,* , roohie-naaz 2 1 department of information technology, national institute of technology srinagar, india. 2 department of computer sciences & engineering, national institute of technology srinagar, india . received 10 december 2016; received in revised form 16 february 2016; accept ed 08 march 2017 abstract ad hoc wire less sensor network (wsn) is a collection of nodes that do not need to rely on predefined infra structure to keep the network connected. the level of security and performance are always somehow re lated to each other, therefore due to limited resources in wsn, cryptographic methods for securing the network against attacks is not feasible. by zantine attacks disrupt the communicat ion between nodes in the network without regard to its own resource consumption. this paper discusses the performance of cluster based wsn co mparing lea ch with advanced node based clusters under byza ntine attacks. this paper also proposes an algorithm for detection and isolation of the compromised nodes to mit igate the attacks by non-cryptographic means. the throughput increases after using the algorithm for isolation of the ma licious nodes, 33% in case of gray hole attack and 62% in case of black hole attack. ke ywor ds : byzantine attacks, cluster based wire less sensor network, advanced node, gray hole, black hole, non-cryptographic 1. introduction wireless sensor network (wsn) is a type of ad hoc networks having la rge nu mber of the nodes. the nodes of the wsn may be static or mobile as in case of other ad hoc networks. the wire less sensor networks pose unique challenges as the sensor nodes are limited in their energy, computation and communicat ion capabilities. a lso, the sensor nodes are deployed in inaccessible areas t o monitor physical environ ment. the sensor nodes may be thousands in nu mber to co llect ively mon itor an a rea. as a result, the e xisting security mechanisms are inadequate [1]. since a ll the nodes in an area usually detect common phenomenon, this leads to high data redundancy. to save energy and prolong network lifetime , an effic ient way is to aggregate the raw data before they are transmitted to the base station as the sensor nodes are resource limited and energy co nstrained. data aggregation is an essential paradigm to eliminate data redundancy and reduce energy consumption [2-3]. the leve l of security and performance are somewhat related to each other. a wsn applicat ion usually requires diffe rent functionalities, sensing, storing data, and data communicat ion. sensing usually require a large number of nodes to ensure coverage and few resources on each node. in contrast, data transmission and data storage require more system resources . data aggregation is an essential parad ig m to e liminate data redundancy and reduce energy consumption. the data ag-gregation is used in wsn to reduce the communication overhead and prolong the network lifetime. however, an adversary may co mpro mise some nodes and use them to forge fa lse values as the aggregation result. for securing data aggregation, we need to detect the malic ious nodes which add to overhead due to encryption, decryption and sharing of keys . tiered network design with functional partit ion prolongs network lifet ime instead of homogeneous network. clustering in wsn, where groups of sensor nodes select their c luster head depending on the energy level [4 , 14] or in some applications the cluster can be fixed at the time of deployment [5]. whether the cluster head is pre-decided or selected by the individual nodes of the group the network will be ad hoc in either case. for many applicat ions, the sensed readings are sens itive and thus demand for data security, confidentiality, integrity and freshness. however, the tight resource co nstraints of wire less sensors restrict the adoption of trad itional co mputation intensive algorith ms. a co mpro mised storage agent may reveal its saved readings, drop important readings, compose forged data readings and reply old data readings. without carefully designed security enhancements, the above attacks can leave the network useless in a hostile environment. there is no secure boundary in ad hoc networ ks, ma king the network susceptible to attacks, since ad hoc networks suffer fro m all-weather attacks which may come fro m any node in the network. the re are other lin k attacks also which can jeopardize the ad hoc network [6]. these include eave sdropping, active interfe ring and leakage of secret info rmat ion, data tampering, message reply, message conta mination and denial of service attacks . the attacks where aim is to gain control over wsn nodes by some unrighteous means and then using these compro mised nodes to execute further malic ious actions. the threats of such attacks are usually fro m inside the network and these threats are more dangerous than the threats from outside the network. these attacks are diff icult to detect as they come fro m co mpro mised nodes , which behave well before they are comp ro mised. a good e xa mple of this type of threats comes fro m the potential byzantine fa ilures encountered in the routing protocol for the ad hoc networks. in a by zantine fa ilu re, a set of nodes are comp ro mised in such a way that the incorrect and ma licious behaviour cannot be directly detected because of the cooperation among these compro mised nodes when *corresponding aut hor. email: address:shabir@nitsri.net t el.: +91-9419009971 advances in technology innovation, vol. 2, no. 4, 2017, pp. 113 118 114 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti they perform ma licious behaviours. the compro mised nodes may seemingly behave we ll; however they may actually make us e of the fla ws and inconsistencies in the routing protocol to undetectable destroy the routing fabric of the network, generate and advertise new routing information that contain non-existent link, provide fake lin k state information, o r even flood other no des with routing traffic. it is co mmon in ad hoc networks that benign failures such as path breakages, transmission impairments and packet dropping, happen frequently. hence ma lic ious failures will be mo re difficult to detect especially when adversaries change their attack pattern and their attack target in different periods of time. 1.1. attack s in ad hoc network s there are nu merous types of attacks in ad hoc network, which may be classified into two types, externa l attacks and internal attacks. external attack, in which the attacker aims to cause congestion propagate fake routing info rmat ion or d isturb nodes from providing services. in internal attack, in wh ich the adversary wants to gain access to the network act ivities, either by some impe rsonation or by directly co mpro mising a current node and using it as basis to conduct its ma licious behaviors [7]. in an internal attack adversary can capture some nodes in the network and ma ke them look like benign nodes, these nodes join the network as the norma l nodes and begin to conduct the ma licious behaviors like propagating fake routing info rmat ion and begin inappropriate priority to access some confidential informat ion [22]. the internal attacks are sometimes more severe threat to the security than externa l attacks as they are difficult to detect at an early stage. 1.2. routing attack s routing attacks are classified into two categories: a ttacks on routing protocols and attacks on packet fo rwarding. the ma in influences brought by the attacks on routing include network partit ion, route loop, resource deprivation and route hijack. because of the mobility and constantly changing topology of the mobile ad hoc ne tworks, it is very difficult to validate all the route messages as a result, impersonating another node to spoof route message, advertising false route metric to misrepresent topology, flooding route discovery, modifying route reply message, generating bogus route error to disrupt a working route, suppressing route error to mislead others may occur. in packet forwarding/delivery selfishness and den ial-of-serv ice are the two main strategies applied for the attack. 1.3. byzantine attack s when a network device suffers a byzantine fault it is assumed to be controlled by an adversary who uses the device to disrupt the network [16]. the goal of the by zantine node is to disrupt the communicat ion of other nodes in the network, without regard to its own resource co nsumption. these cause byzantine failu res which include the omission failures and commission failures. as for instance in omission failures if a node fail to rece ive a request or fail to send a response and in commission failures if a node process a request incorrectly or sending an incorrect or inconsistent response to a request. in ad hoc networks, the by zantine attacks are as: blac k hole attack, gray hole attack, flood rushing attack and wormhole attack. wireless sensor networks are favorite targets of byzantine attacks because of their limited dynamic t opology etc. [21]. 1.4. black hole attack it is a basic byzantine attack [9] where adversary stops forwa rding data packets, but still partic ipates in the routing protocol correctly. as a result, whenever the adversarial node is selected as part of a path by the routing protocol, it prevents commun ication on that path. most routing protocols are disrupted by black hole attacks because they render the normal methods of route maintenance useless . 1.5. gray hole attack it is a special case of black hole attack where an attacker could create a grey hole, in which it is selectively drops some packets but not others, for exa mp le forwarding some packets but not data packets[10]. 1.6. wormhole attack if more than one node is compromised, it is reasonable to assume that these nodes interact in order to ga in an additional advantage. this a llo ws the adversary to perform a more effective attack. one such attack is by zantine wormhole where two adversaries tunnel packets between each other in order to create a shortcut (or wormhole) in the network. the adversaries can send a rou te request and discover a route across the ad hoc network, then tunnel packets through the non-adversarial nodes to execute the attack. the adversaries can use the low cost appearance of the wormhole links in order to increase the probability of being elected as part of the route and then attempt to disrupt the network by dropping all of the data packets. the wormhole attack is strong attack which can be performed even if only two nodes are compromised . 1.7. flood rushing attack a flood rushing attack [12] e xplo its the flood duplicate suppression technique used by many routing protocols. this attack takes place during the propagation of legit imate flood and can be seen as a “race” between the legitimate flood and the adversarial variant of it. if an a dversary successfully reaches some of its neighbors with its own version of the flood packet before they receive a version through a leg itimate route, then those nodes will ignore the legitimate version and will propagate the a dversarial version. th is may result in the continual ab ility to establish an adversarial-free route, even when authentication techniques are used. when a node wants to send a packet, it will send route request packet and if it receives a route reply first fro m a norma l behaving node, then ev erything will work fine. however, if it gets reply fro m an attacker node, all the packets will not reach the destination or there may be selective dropping. in both the cases the delivery ratio will decrease. therefore, identification of such nodes is the first step in preventing their partic ipation in the data transfer. also, a route reply fro m an attacke r node can advances in technology innovation, vol. 2, no. 4, 2017, pp. 113 118 115 cop y right © taeti reach the source node earlier than a norma l node if it is near to the source node. since each node in a homogen eous wsn, acts as router, the data transmission fro m source to the gateway occurs via diffe rent sensor nodes, while in case of heterogeneous network the indiv idual nodes may or may not participate in the routing process . the homogenous wsn can be treated as a special case of ad hoc networks where the number o f nodes is very large as compared to the ad hoc network. the detection and isolation of an attacker node is difficult. also, the packet delivery ratio will be lesser. in heterogeneous wsn the nodes are grouped in clusters and each no de in the cluster transmits its data via the c luster head (ch) [4, 14]. since the nodes in cluster are fewer as compared to the nodes in a homogeneous wsn the chances of detection and isolation of the attacker node are more. karlof et a l. [13] proposed selective forwa rding attack for the first time in wire less sensor networks and suggested that multipath forwa rding to counter the attack. but, the algorith m fails to suggest a method to isolate the attacking node. marti et a l. [11] p roposed a technique called watchdog, in which a node continuously monitors the neighboring nodes to which the packet is sent and to check whether the packet is fo rwa rded or not. but the algorith m fails to detect the attacker in the presence of selective forwarding attack. 2. comparison leach and advanced node in fig. 1, lea ch vs. advanced node based network the probability of sustaining the black hole or gray hole attack is more in the advanced node based network as compared to lea ch based network. also the life cyc le of the nodes in advanced node based network is more than the leach based network. during the cluster head s election process and after becoming cluster head the node consume a lmost n+1 times the energy consumed by an individual sensor node. since data aggregation as well as the routing of the other informat ion fro m and to the nodes is carried through the cluster head, in addition to its own sensing and data transmission which leads to quicker e nergy depletion. fig. 1 advanced node based protocol vs. leach protocol. (energy in mah and time in days with 1 hour operation for each sensor node) 3. results and discussion in this work, we develop a non-cryptographic type of defense by checking the forwarding of the upstream nodes by overhearing their transmission. we consider ad hoc on demand vector routing protocol to imple ment these a ttacks . in the black ho le attack, a node will part icipate in routing but will drop a ll the packets it receive [11]. the ma licious node will a lways advertise in the network that it has a fresher route to the destination by setting the s equence number to a large value and will reply to the broadcast route request packet before other nodes send a reply. thus, the attacker node will attract all the traffic in its transmission range towards itself and then drop the packets. this type of situation will decrease the packet delivery rat io, but at the same time the energy of the black node will decrease rapidly resulting in self-immo lation of the node. however, during the time o f the data transmission the other nodes which send the packet to the black node will result in decrease of their energy due to repeated transmissions for the same packet. th is will decreas e their energy and result in reduced life cycle of the node. in gray hole attack, the attacker node drop selective packets a ccording to some criteria or randomly [4]. this type of attack is difficult to detect, especially in wireless scenario where pac kets are dropped because of the congestion, channel capacity etc. this a lgorith m is based on the probability of attack which depends on the ratio of number of packets to the nu mber of packets transmitted. if the probability of attack is greater than the probab ility of blac k hole attack and it is true twice then the attack is black hole attack and if the probability of attack is greater than the probability of gray hole attack and it is true twice then the advances in technology innovation, vol. 2, no. 4, 2017, pp. 113 118 116 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti attack is gray hole attack. after the detection of the attack all the nodes are sent a broadcast not to include the node in any future routing for transmission of packets. the co mplete algorithm for different scenario is given as under: scenario: (as shown in fig. 2) case 1: homogeneous ad hoc or wireless sensor network wireless sensor network is a large network of sensors which have the ability to communicate with each other. these sensor nodes are transmitting the data from one sensor to another for further transmission to the sink node. in ad hoc networks ad hoc on demand vector is a source initiated advanced on demand routing protocol. each sensor node has a routing table that stores the information of the next hop node to route the destination. when a source node wants to route a packet to sink node, it uses the specified route if a fresh route to the sink is not available otherwise it will update its table for shortest route by the route discovery using route request message to its neighboring nodes. in gray-hole attack the malicious node selectivity or randomly forwards packets passing through it. sink node after receiving packet from the source node, unicast (route reply) message en -route neighboring node from which it receives the packet. in black-hole attack, the malicious node pretends as if it has the shortest path to the sink node and drops all the packets. case 2: heterogeneous wireless sensor network with leach based cluster head the wireless sensor network is partitioned into clusters and each cluster consists of a group of sensor nodes which may or may not transmit data to the destination via the neighboring nodes. mostly, the nodes communicate directly with the cluster head. the cluster head is chosen which is having the maximum energy level amongst the cluster nodes. as in leach the process of selecting or electing a cluster head is repeated after a certain interval of time. the number of nodes in a cluster is less as compared to the case 1 [14]. but the gray-hole and black-hole attack is possible if the nodes communicate with the cluster head via intermediate or neighboring nodes. also, the attacks are possible if the node which acts as cluster head is compromised. the severity of the attack may be manifold as all the data packets from each and every node will be dropped. case 3: heterogeneous wire less sensor networks with advanced node as cluster head the wireless sensor network is partitioned into clusters as in case of case 2 but the cluster head is predefined and the advanced node which acts as a cluster head is presumed to have higher energy, processing power and range [5] as compared to the normal sensor nodes. the possibility of gray-hole and black-hole attacks is less as compared to the case 1 or case 2. as it will be difficult to compromise the cluster head which is responsible for the transmission of the data from the nodes to the gateway. all the nodes will d irectly communicate with the cluster head, but as a special case the nodes may also communicate with the cluster head via the intermediate or neighboring nodes within that cluster. in the earlier case, the probability of compromising a node is lesser. also, it is possible to use the cryptographic algorithms like key exchange mechanisms between the nodes and the cluster head during data transmission. 4. algorithm for detection and isolation of byzantine nodes by non-cryptographic methods in either case of by zantine attacks, gray -hole o r black-ho le attack, the detection of the type of attack is first step. after we know the type of attack our ne xt pr iority is to identify the comp ro mised nodes in the network. the algorith m detects these nodes by non -cryptographic methods, checking the forwarding o f the nodes by ove rhearing their transmission and isolation of these nodes so that cannot take part in routing. the scenario is shown in fig. 2. fig. 2 gray-hole and blac k-hole attack detection and isolation scenario 4.1. assumptions before the imp le mentation of the algorithm we have taken certain assumption. since there are many other fa ctors which could cause the change in the throughput which we have ta ken as solely by the by zantine attacks. like con-gestion due to buffer overflow is not insignificant, as we need to restrict the upstream node fro m delivering packets when the downstream node does not have sufficient space. (b) in p ractical cases black hole attack may not drop all the packets; it has its dependence on other factors as well. (c ) as the signal power decreases the range is also decreased, but in case of wsn, the nodes are at a very short distances for a decrease in energy is no t affected too much e xtend as compared to long distance communication. (d) in mu lti-hop communicat ion each node maintains the table of the routing informat ion during the transmission of packets, but here each node will be having additionally the attack table, this may add some overhead to the packets . a. no packet is dropped due to buffer overflow b. black-hole attack drops all the packets it receives c. range is not getting affected by decrease in the en ergy level of a node d. each node will maintain an attack table notation and parameters nid node identifier nt total no. of packets transmitted by a node nd total no. of packets dropped by a node nl total packet loss nl= ntnd pa probability of packets successfully received pb probability of presence of black-hole pg probability of presence of gray-hole pa= nl/ nt nr reporter node na attacker node ch cluster head cid cluster id advances in technology innovation, vol. 2, no. 4, 2017, pp. 113 118 117 cop y right © taeti 4.1.1. algorithm for homogeneous network for a particular interval: 1. calculate value ofnl nl = nl nd; 2. define values of pb and pg //threshold values as per the scenario 3. calculate pa pa = nl/ nt; 4. if (pa >= 2pb){ then broadcast packets to all ns and rs with nid of both reporter node and attacker node. type –of-attack = b;} else if (pa >= 2pg){ broadcast packets to all ns and rs with nid of both reporter node and attacker node. type-of-atack = g;} else if (pa>pg and pa >pb){ broadcast packets to all ns and rs with nid of both reporter node and attacker node. type-of-attack = g;} else{ print(“no attacker node found”); and broadcast nid of sender node to all ns and rs”. type-of-attack = nil;} 4.1.2. algorithm dedicated cluster head assumption is that cluster heads are pre assigned with identified cluster nodes . for a particula r cluster and for a particular interval: 1. assign a node cid with maximu m power 2. calculate value of nl for that particular cluster nl = nt – nl; 3. define value of pb and pg for a cluster 4. calculate pa pa = nl/nt; 5. if (pa >= 2pb){ broadcast packets to all ns, rs and ch with nid of both nr and na type-of-attack = b;} else if (pa >= 2pg){ broadcast packets to all other ns, rs and ch with nid of both nr and na type-of-attack = g;} elseif (pa >pg and pa >pb){ broadcast to all ns, rs and chs of cluster with n id of both nr and na type-of-attack = g;} else{ print (“no attack found”) 4.1.3. algorithm heterogeneous network assumption is that network is div ided into clusters <=100. for a part icula r c luster and for a part icular interval: 1. choose a node randomly as ch and assign cid 2. calculate value of nl for that particular cluster nl = nt – nd; 3. define value of pb&pg for a cluster. 4. calculate pa pa = nl/nt; 5. if (pa >= 2pb){ if (attacker nid = cid of ch){ broadcast packets to all other chs with cid of attacker ch type-of-attack = b;} else { broadcast packets to all ns, rs and ch with nid of both nr& na. type-of-attack = b;} elseif( pa>= 2pg){ if (attacker nid = cid) { broadcast packets to all chs with cid of attacker ch type-of-attack = g;} else{ broadcast packets to all ns and rs and to ch of cluster with nid of both nr and na type-of-attack = g;} } else if (pa>pg and pa>pb){ if(attacker nid = cid){ broadcast packets to all other chs with cid of attacker ch type-of-attack = g;} else{ broadcast packets to all ns, rs& ch with nid of both nr and na type-of-attack = g;} } else{ print(“no attack found”); and broadcast nid of sender node to all ns, rs and chs type-of-attack = nil; } 5. results based on the algorithm for dete ction and isolation fro m the simu lation results as is evident from the fig. 4 throughput vs. time. in itia lly, we simu late the network with no attack; the throughput is 90-95%. then, as we introduce the black hole attack in the network, throughput decreases to 3%-5%. now, as the network uses the isolation algorith m, throughput increases to 67%. simila rly in advances in technology innovation, vol. 2, no. 4, 2017, pp. 113 118 118 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti case of gray hole attack, initia lly we simu late the network without attack and the throughput is 90 -95% . then, we introduce the gray hole attack and the throughput decreases to 35%-50%. after using the isolation algorithm the throughput increases to 88%. fig. 3 time(s) vs. throughput (%) fig. 4 time vs. throughput (%) 6. conclusion as shown in fig. 3 and fig. 4, there is re ma rkab le improve ment in the throughput after using the proposed al-gorithm for isolation of the ma lic ious nodes. the alg orith m will be mo re suited to the applications were we require to have energy e fficient design. since the a lg orith m is a non-cryptographic one and purely depend on the probability of packets successfully received, there fore probability of presence of blac k hole nodes and probability of presence of gray ho le nodes may vary in some cases. but the a lgorith m will be useful for the sensor networks where we can't use the cryptograph ic algorith ms to tackle the security proble m due to the fact that increased processing and communication time will increase the energy consumption. if we part itioned the network into clusters then the gray hole or the b lack hole attack will re main confined to its own cluster only without affecting the other clusters in the network till the cluster head itself is not compro mised. but if we use the advanced node in the network as a cluster head then the probability of cluster head to be compromised will be lesser due to the fact that the node is pred efined cluster head and we can also use the cryptographic mechanis ms like the key e xchange etc. for secure transmission with the processing to be done centrally at the cluster head (advanced node), as it is having higher processing, commun ication and energy as compared to the member nodes of the cluster. references [1] a. perrig, j. stankovic, and d. wager, “security in wireless sensor networks ,” communication of the acm, vol. 47, no. 6, pp. 53-57, june 2004. [2] d. estrin, r. govindan, and j. heidemann, s. kumar, “next century challenges: scalable coordination in sensor networks ,” proc. acm international conf. mobile computing and networking, acm press , 1999, pp. 263-270. [3] y. yu, b. krishnamachari, v. k. prasanna, “energy-latency tradeoffs for data gathering in wireless sensor networks ,” proc. ieee co mputer and communication societies, ieee press , 2004. [4] s. e. khediri, n. nasri, a. wei, and a. kachouri, “a new approach for clustering in wireless sensors networks based on leach,” procedia computer science, vol. 32, pp. 1180-1185, 2014. [5] s. a. sofi and r. naa z, “ energy efficient routing protocol for structured deployment of wireless sensor networks ,” international conf. next generation networks, iet press, september 2010, p. 10. [6] s. sofi, e. malik, r. baba, h. baba, and r. mir, “analysis of byzantine attacks in ad hoc networks and their mitigation,” international conf. computing and information technology, february 2012, pp.794-799. [7] w. li and a. joshi, “security issues in mobile ad hoc net works a survey,” https://pdfs.semanticscholar.org/b221 /99c61df5445836c5f1bbd0ea6f02dabefd6b.pdf, 2007. [8] y. c. hu, a. perrig, and d. b. johnson, “rushing attacks in wireless ad hoc network routing protocols ,” proc. 2nd acm workshop on wireless security , acm press , 2003, pp. 30-40. [9] c. karlof and d. wagner, “secure routing in wireless sensor networks: attacks and countermeasures ,” ad hoc networks , vol. 1, no. 2, pp. 293-315, 2003. [10] s. marti, t. j. giuli, k. lai, and m. baker, “mitigating routing misbehavior in mobile ad-hoc networks,” proc. 6th annual international conf. mobile computing and networking, pp. 255-265, august 2000. [11] h. li, k. li, w. qu, and i. stojmenovic, “secure and energy efficient data aggregation with malicious aggregator identification in wireless sensor networks ,” future generation computer systems, vol. 37, pp. 108-116, 2014. [12] y. zhao, y. zhang, z. qin, and t. znati, “a co-commitment based secure data collection scheme for tiered wireless sensor networks ,” journal of systems architecture , vol. 57, no. 6, pp. 655-662, 2011. [13] manju, s. chand, and b. kumar, “improved-coverage preserving clustering protocol in wireless sensor networks ,” international journal of engineering and technology of innovation, vol. 6, no. 1, pp. 16-29, 2016. [14] r. duche and n. sawade, “energy efficient fault tolerant sensor node failure detection in wsns,” international journal of engineering and technology of innovation, vol. 6, no. 3, pp. 190-210, 2016. [15] m. abdelhakim, l. e. lightfoot, j. ren, and t. li, “distributed detection in mobile access wireless sensor networks under byzantine attacks,” ieee transactions on parallel and distributed systems , vol. 25, no. 4, pp. 950-959, april 2014. [16] m. young and r. boutaba, “overcoming adversaries in sensor networks: a survey of theoritical models and algorithm approaches for tolerating malicious interferences ,” ieee communications survey and tutorials, vol. 13, no. 4, 2011.  advances in technology innovation, vol. 5, no. 3, 2020, pp. 166-181 evaluation of pan-sharpening techniques using lagrange optimization mutum bidyarani devi 1,* , rajagopalan devanathan 2 1 department of electronics and communication engineering, 2 department of electrical and electronics engineering, hindustan institute of technology and science, chennai, india received 19 may 2019; received in revised form 07 december 2019; accepted 04 march 2020 doi: https://doi.org/10.46604/aiti.2020.4288 abstract earth’s observation satellites, such as ikonos, provide simultaneously multispectral and panchromatic images. a multispectral image comes with a lower spatial and higher spectral resolution in contrast to a panchromatic image which usually has a high spatial and a low spectral resolution. pan-sharpening represents a fusion of these two complementary images to provide an output image that has both spatial and spectral high resolutions. the objective of this paper is to propose a new method of pan-sharpening based on pixel-level image manipulation and to compare it with several state-of-art pansharpening methods using different evaluation criteria. the paper presents an image fusion method based on pixel-level optimization using the lagrange multiplier. two cases are discussed: (a) the maximization of spectral consistency and (b) the minimization of the variance difference between the original data and the computed data. the paper compares the results of the proposed method with several state-of-the-art pan-sharpening methods. the performance of the pan-sharpening methods is evaluated qualitatively and quantitatively using evaluation criteria, such as the chi-square test, rmse, snr, sd, ergas, and rase. overall, the proposed method is shown to outperform all the existing methods. keywords: image fusion, pan-sharpening, optimization, spectral consistency 1. introduction with the fast-growing number of earth’s observation satellites, satellite data providing information about the surface of the earth is increasing. this information has been applied in different fields for various end-users’ applications, such as urban planning, agriculture, forestry, and mining. depending on sensors onboard the satellite, different types of images of the earth’s surface received. some sensors provide single-channel data while other sensors provide both single and multichannel data. single-channel data or monochrome images, such as the panchromatic (pan) image, usually comes with a high spatial resolution associated with a low spectral resolution. on the other hand, the multispectral (xs) image comes with a low spatial resolution and a high spectral resolution. for example, the commercial-launched satellite sensor, such as ikonos, provides 1 m pan and 4 m xs (r, g, b, and near-infrared bands) images. the analysis and the usage of the remotely sensed data are user-dependent. sometimes the processing of high-quality images is needed for certain applications such as classification and target detection. since satellite data are available in different resolutions spatially and spectrally as well as at different scales and times, data fusion has been applied successfully to obtain both high resolution spatial and spectral image. data fusion can be defined as the process of combining two or more incoming signals which complement each other and produce an output signal which has more information content than the incoming inputs. more formally, data fusion is a formal framework that expresses means and tools for the alliance of data of the same scene originating from different sources. it aims * corresponding author. e-mail address: bidyarani.mutum@gmail.com advances in technology innovation, vol. 5, no. 3, 2020, pp. 166-181 167 at obtaining information of greater quality; the exact definition of greater quality will depend on the application” [1]. image fusion plays an important role in the remote sensing field where satellite images come with complementary resolutions that are being made available in the public domain. image fusion, in particular, is defined as the “combination of two or more different images (of the same scene) to form a new image by using a certain algorithm” [2]. one important application of image fusion has been increasing the resolution of an xs imagery by using higher-resolution pan data. the output consists of an xs image whose resolution is both spatially and spectrally high. such an image fusion process usually termed as pan-sharpening. pan-sharpening involves the integration of a panchromatic band and a multispectral band of different resolutions. generally, a panchromatic band comes with a high spatial resolution and a low spectral resolution, while a multispectral band comes with a lower spatial resolution and a high spectral resolution. sometimes, the integration of these images is required when a very high-quality image is needed. a panchromatic image provides details of the scene observed while lacking in color properties. nevertheless, a multispectral image can be useful in detecting objects with a similar spectral signature but belonging to different classes. in the pan-sharpening process, the properties of these complementary images are combined to get an output image that has a high resolution spatially and spectrally. however, one primary challenge in pan-sharpening is to preserve the spectral properties of the multispectral band in the output image. the assessment of the fused image can be performed using various evaluation parameters. the most commonly used fusion methods include intensity-hue-saturation (his) transform, brovey transform, principal component analysis (pca), smoothing filter-based intensity modulation (sfim), high pass filter (hpf), and multiplicative transform. in the his [3-6] pan-sharpening method, the rgb color space is transformed into ihs color space. the transformation can be performed by using three bands only at a time. the step consists firstly of resampling the xs images to the same spatial resolution of pan. secondly, the transformation of resampled rgb to ihs color space is carried out. thirdly, the intensity component is a histogram matched with panchromatic band data. fourthly, the intensity component is replaced by the histogram matched with panchromatic band data. finally, the reverse transformation is applied to get the new r, g, and b fused images. the brovey method [7], on the other hand, is a combination of pan and xs images. the method involves multiplication and division operations. the brovey method is limited to only three bands of the xs channels. each xs band is divided by the sum of all the three bands and multiplied by pan. smoothing filter-based intensity modulation (sfim) is a smoothing algorithm [8] where a low pass filter is applied to a high-resolution pan channel. the low resolution multispectral is multiplied with the high-resolution pan band, which is divided by the low pass filtered pan band. this is done for every band of the multispectral channel. the high pass filter (hpf) [8] method involves high pass filtering of pan band with a window size of a 3×3 filter. in this case, the multispectral band is multiplied with the high pass filtered pan band divided by 2. principal component analysis (pca) [9, 10] is another commonly used method applied to image fusion. first, the principal components (pc) are computed for each multispectral band, then the first pc is replaced by the pan band, and an inverse transformation is applied to get the fused output. lastly, multiplication transform [8] is another popular algorithm used for image fusion. this method is very simple to implement. each of the low-resolution multispectral bands is multiplied with the high-resolution pan band with a corresponding weight. a square root is taken on the final output to avoid excessive brightness values. earlier work of the authors [11] proposed a solution to the problem of pan-sharpening, which aims in maintaining the spectral consistency of the xs channels in the fused image. the primary objective of the current work is to compare the proposed pan-sharpening technique [11] to the existing fusion methods of his, brovey, and other standard methods (namely, sfim, hpf, pca, and multiplication methods) based on several evaluation criteria such as chi-square test, rmse, snr, sd, ergas, and rase. the main contribution of the paper stated as (a) restatement of the authors’ earlier proposal for pan-sharpening based on linear regression and lagrange optimization, (b) application of the proposed algorithm of pan-sharpening to ikonos data, (c) validation of the results of the proposed pan-sharpening method, (d) complexity analysis of the proposed pan-sharpening algorithm, and, (e) finally, a comparison of the performance of the proposed pan-sharpening advances in technology innovation, vol. 5, no. 3, 2020, pp. 166-181 168 method to the other fusion methods viz. his, brovey, sfim, hpf, pca, and multiplication methods based on several evaluation indices, namely, chi-square test, rmse, snr, sd, ergas, and rase. the organization of the rest of the paper is as follows: section 2 gives an overview of the existing work related to the pan-sharpening comprising of state-of-the-art methods including model-based methods. section 3 deals with a brief description of the proposed fusion method formulated as an optimization problem with an objective function and the lagrange multiplier-based solution. the proposed method is applied to ikonos data as described in section 4. simulation results of the proposed method and comparison with other fusion methods based on several evaluation criteria are discussed in section 5. finally, section 6 presents the main discussions of the experimental results followed by the conclusion. 2. related work the state-of-art methods include the intensity-hue-saturation transformation (ihs) [3-6], the brovey method [7], and principal component analysis (pca) [12]. many researchers have reported that though these conventional methods produce a spatially high-resolution fused image, images are spectrally distorted. hong [13] proposed a method that involves the integration of “ihs and wavelet” to preserve the spatial details. zhang and hong [14] also proposed a method of integrating ihs and wavelet to solve the spectral distortion problem. a method combining pca and nonsubsampled contourlet transform was reported by [15] to overcome the drawback of spectral distortion. andreja [16] proposed a method integrating ihs and brovey methods with a multiplicative (multi) method for maintaining the spatial and spectral details. yang [17] proposed an ihs based on pan-sharpening technique using ripplet transform and compressed sensing for improved spectral properties. the pan-sharpening method based on bayesian theory was proposed by [18]. a comparison between different popular image fusion techniques was reported in the works of [19-20]. performance evaluation of fusion algorithms can also be seen from [20-22]. an assumption that the downsampled fused image should be similar to the original xs image was proposed by [23]. meng [24] proposed a pan-sharpening technique which uses an edge-preserving guided filter based on a three-layer decomposition. to strengthen pan-sharpening methods, [25] proposed a method involving the prior modification of the panchromatic image. this method preserves simultaneously spatial and spectral quality. there also exist a few variational models reported by researchers. ballester [26] proposed the first variational model called p+xs. fang [27] also proposed a model to fuse pan and xs based on certain assumptions. moller [28] proposed a model called variational wavelet for pan-sharpening (vwp). deng [29] also proposed a variational model based on kernel hilbert space and heaviside function. super resolution-based pan-sharpening can be found in the works of [30-32]. the fusion of satellite images has been done on data obtained from various sensors. it can be done on the data coming from the same sensor or different sensors onboard. examples of single-source sensors include ikonos, quickbird, landsat, spot, etc. ikonos data fusion has been reported in the works of [6, 23, 33-34]. fusion using spot data has been reported in the works of [3, 35]. several works have been reported integrating ikonos with landsat data as well as spot data with landsat data [4-5]. a review of different pan-sharpening techniques can be seen in the works of [2, 36-39]. 3. proposed fusion method 3.1. linear regression the first step in our fusion method was to build a linear regression model based on the assumption that a strong correlation exists between the panchromatic and the multispectral bands. the regression model can be defined as follows ar bg cb p   (1) advances in technology innovation, vol. 5, no. 3, 2020, pp. 166-181 169 where r, g, and b represent the deviations from the respective sample mean of red, green, and blue spectral band intensity data. a, b, and c are the regression coefficients. eq. (1) can also be written as xu p (2) where x = [r g b] is an n × 3 matrix containing columns of n samples of red, green, and blue color data. besides, u= [a b c] t represents the vector of regression coefficients determined. the superscript t stands for matrix transpose. the regression coefficients can be calculated as   1 t t u x x x p   (3) it is assumed that ( x t x ) -1 exists. 3.2. lagrange optimization using (1) as a constraint, an objective function for minimization is formulated to achieve spectral consistency and variance matching of the xs bands as follows.       2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 1 1 1 1 1 1 1 1 1 1 1 1                                                                                       r jk r jk jk r g jk j j j j g jk jk g b jk j j j b jk jk b jk jk j j j r r r g n n n n g g b n n n b b ar n n n i bg m                               k jk jk j cb p (4) where rjk, gjk, and bjk are deviations from the respective sample mean for the red, green, and blue band, respectively at pixel location (j, k) of the n× n image. besides, r  , g  , and b  are the parameters for red (r), green (g), and blue (b) band, respectively forming a convex combination of spectral consistency and variance matching. 2 r  , 2 g  and 2 b  are the variances of the original red (r), green (g), and blue (b) band data, respectively, and jk  is the lagrange multiplier to enforce the constraint (1). the solution to the above-mentioned minimization problem had been reported in an earlier work of the authors [11] and is briefly described in the following section. two independent cases can be derived from eq. (4). case 1: with 1, , , j j r g b   , for red, green, and blue respectively. case 2: when 0, , , j j r g b   , for red, green, and blue respectively. the first case deals with minimizing the spectral inconsistency between the actual multispectral data and the computed data. the second case compares the variances between the actual multispectral data and the computed data assuming a gaussian distribution of the intensity data. proposition 1: 1, , , j j r g b   for red, green, and blue, respectively. eq. (4) can be written as   2 2 2 2 2 2 2 2 2 1 1 1                                jk jk jk jk jk jk jk jk j j j j r g b ar bg cb n n m p n in (5) without loss of generality, the three bands, namely red, green, and blue can be considered independently. differentiating eq. (5) concerning rjk, the solution to eq. (4) is given as advances in technology innovation, vol. 5, no. 3, 2020, pp. 166-181 170 2 2 2 ,    jk jk fo a r the red p r a b c band (6) the pan-sharpened output for the red band is obtained as (1) ,   new jk jk mean r r r (7) where (1) , new jk r is the pan-sharpened high-resolution output at the location (j,k), and rmean is the sample mean of the red band. superscript (1) stands for case 1. similarly, if we differentiate (4) concerning gjk and bjk respectively, the solution to eq. (4) is given as 2 2 2 ,    jk jk for the green bp g a b c band (8) the pan-sharpened output for the green band is obtained as (1) ,   new jk jk mean g g g (9) where (1) , new jk g is the pan-sharpened high-resolution output at the location (j, k), gmean is the sample mean of the green band, and 2 2 2 ,    jk jk for the blue cp b a b c band (10) the pan-sharpened output for the blue band is obtained as (1) ,   new jk jk mean b b b (11) where (1) , new jk b is the pan-sharpened high-resolution output at the location (j,k), and bmean is the sample mean of the blue band. proposition 2: 0, , , j j r g b   for red, green, and blue band, respectively. eq. (4) can be written as   2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 1 1 1 1 1 1                                                                        jk jk r jk jk g j j j j jk jk b jk jk jk jk jk j j j r r g g n n n n b b ar bg cb p n in n m (12) without loss of generality, the three bands, namely red, green, and blue can be considered independently. differentiating eq. (9) concerning rjk, gjk, and bjk respectively, the solution to eq. (4) is given as 2 2 2 ,    jk jk r r g b ap r e e a b c e for the red e band (13) 2 2 2 ,    jk jk g g r b bp g e e a b c e e for the green band (14) and 2 2 2 ,    jk jk b b r g for the b cp lub e e a b c e e e band (15) advances in technology innovation, vol. 5, no. 3, 2020, pp. 166-181 171 respectively, the pan-sharpened output for the red, green, and blue band is respectively obtained as ( 2) , ( 2) , ( 2) ,       new jk jk mean new jk jk mean new jk jk mean r r r g g g b b b (16) where (2) (2) (2) , , , , , new jk new jk new jk r g b is the pan-sharpened high-resolution output and rmean, gmean, bmean are the respective sample means of the red, green, and blue band respectively. superscript (2) stands for case 2. remark 1: er, eg, and eb are the errors associated with the multispectral red, green, and blue band which represents the variance difference between the actual and the computed data. eqs. (13)-(15) are circuitous involving error ratios that are dependent on the solution of rjk, gjk, and bjk, respectively. for example, the ratios /g re e and /g be e in eq. (14) for gjk depend on rjk, gjk, and bjk. similarly, for the error ratios involved in eqs. (13) and (15). the circuitous relation is resolved [11] by showing that the error ratios, in the optimal case, can be considered as the ratio of the variances of the respective spectral band. for example, the error ratios in eq. (14) is in the form 2 2 / /  g r g r e e and 2 2 / /  g b g b e e . similar results apply for the error ratios in eqs. (13) and (15). the solution of case 1 and case 2 as given in eqs. (6), (8) and (10) as well as eqs. (13)-(15), respectively represent high-resolution deviation (from the respective mean) in the multispectral band for each channel. remark 2: the fusion process can be performed simultaneously for all the three bands considered. (1) (1) (1) , , , , , new jk new jk new jk r g b (when added to respective mean) represent the projected high-resolution multispectral band based on case 1 while (2) (2) (2) , , , , , new jk new jk new jk r g b (when added to respective mean) represent the high-resolution multispectral band based on case 2. remark 3: since the eqs. (6)-(11) and (13)-(16) are in closed form, the computation of (1) (1) (1) , , , , , new jk new jk new jk r g b and (2) (2) (2) , , , , , new jk new jk new jk r g b for each location (j, k) is done in a single operation. hence, the complexity of pan-sharpening computation is o(n) where n is the number of data points. 3.3. proposed fusion method fig. 1 flowchart of the proposed fusion method advances in technology innovation, vol. 5, no. 3, 2020, pp. 166-181 172 fig. 1 shows the methodology of the proposed fusion method indicating the sequence of steps involved as 1, 2, etc as shown. first, a panchromatic band and multispectral band images were taken. second, down-sampling the panchromatic data to the lower resolution of multispectral data since the resolution differs in the ratio of 4:1. for example, down-sampling of 8×8 pixels pan data gives 2×2 pixels points. thirdly, the deviation from the respective sample means was calculated for both the pan and xs data. fourthly, the regression coefficients were calculated using eq. (3), and the fifth step consists of applying the proposed two cases of fusion methods to the dataset with the respective means of the band added to the deviations (eq s. (6)-(16)). finally, the projected high-resolution multispectral data was down-sampled for comparison with the original low-resolution multispectral data. 4. evaluation criteria the proposed method is applied to ikonos data, the results of which are presented in the following section. the results of the proposed method of pan-sharpening are also compared with that of other state-of-the-art methods on the same data in that section. to compare the various results, using a common set of evaluation criteria which are described in this section. 4.1. chi-square test the chi-square test [40] is computed as   2 2 i i i i o e e     (17) where i e and i o are the expected and the observed data points, respectively, and i=1, 2…n. a small p-value indicates a good fit. the lower the p-value is, the more significant the result is. a low p-value indicates the probability of getting a better fit than the one evaluated is low. 4.2. root mean square error (rmse) root mean square error (rmse) [41] is defined as 2 1 ˆ( )     n i i i x x rmse n (18) where i x is the original value, while ˆ i x is the computed value. a lower rmse value indicates good spatial and spectral properties. 4.3. signal to noise ratio (snr) the signal to noise ratio (snr) [42] can be calculated as follows 2 1 2 1 ˆ( ) ˆ( )       n i i n i i i x snr x x (19) where i x and ˆ i x are the original and computed data, respectively. a higher snr value indicates a good result. 4.4. spectral discrepancy (sd) spectral discrepancy (sd) [41] is usually done to check the spectral quality of the fusion result and is computed as follows advances in technology innovation, vol. 5, no. 3, 2020, pp. 166-181 173 1 1 ˆ      n i i i sd x x n (20) a lower sd value indicates a good spectral quality of the fusion result. 4.5. erreur relative globale adimensionelle de synthese (ergas) “erreur relative globale adimensionelle de synthese (ergas)”, an error-index proposed by [43] which specifies the global picture quality of the fusion output and is given as 2 1 1 ( ) 100 ( ) q q h rmse q ergas l q q         (21) where (h/l) is the ratio of the pixel sizes between the pan and xs; ( )q and q represents the mean of the qth channel and the index of the band respectively. as proposed by [43], a lower value of ergas indicates a good quality or better fusion output, or the results are considered to be of low quality. 4.6. relative average spectral error (rase) rase index is represented as a percentage to predict the spectral quality of the fusion output [44]. 2 1 100 1 ( ) k i i rase rmse k m k    (22) where m is the average pixel value in the spectral band considered, k indicates the number of bands, and rmse is the root mean square error of the k th channel. like the ergas, the result can be interpreted similarly. a lower value of the rase index indicates a good spectral quality of the fusion output. 5. experimental results and analysis ikonos satellite image was used for the fusion process. ikonos satellite image has been used by many researchers for various applications. in this work, an illustration of the proposed methods and other fusion methods were performed on a real ikonos image consisting of a 32×32 pixels panchromatic band and corresponding 8 ×8 pixels multispectral band (shown in figs. 2 and 3, respectively). the dataset consists of smaller areas selected from a larger ikonos satellite image that is freely available for use. the ikonos satellite image provides two types of images (a) a panchromatic image (1 m spatial resolution) and (b) a multispectral image (4 m spatial resolution) comprising of 4 bands namely, red, green, blue, and nir. in this study, only the first three bands were considered. fig. 2 32×32 panchromatic band fig. 4 shows the fusion result based on the proposed method and fig. 5(a) through (f) shows the fusion results based o n his, brovey, sfim, hpf, pca, and multiplicative methods, respectively. in this work, the fusion result was obtained only for advances in technology innovation, vol. 5, no. 3, 2020, pp. 166-181 174 the three bands, namely, red, green, and blue of the multispectral channels. all the figures are presented in grayscale since each band can be represented as a single color. therefore, the grayscale image will be equally effective. fig. 3 8 ×8 multispectral band (a) case 1 (b) case 2 fig. 4 fusion result based on proposed fusion method advances in technology innovation, vol. 5, no. 3, 2020, pp. 166-181 175 for the assessment of the spectral and spatial quality of the fusion result, the projected high-resolution xs images are down-sampled so that they are in the same dimension as the original xs. spectral consistency means that the down-sampled data from the high-resolution xs should be close to the original xs data [45]. (a) ihs method (b) brovey method (c) sfim (d) high pass filter fig. 5 fusion results advances in technology innovation, vol. 5, no. 3, 2020, pp. 166-181 176 (e) pca (f) multiplicative methods respectively fig. 5 fusion results (continued) table 1 performance evaluation of the proposed fusion method (case 1 and case 2) evaluation criteria multispectral band case 1 case 2 chi square test (p value) red 0.01 0.0002 green 1.4×10 -6 1.4×10 -6 blue 3.9×10 -4 0.48 rmse red 4.75 4.43 green 4.21 4.21 blue 4.45 5.86 snr red 7.04 7.53 green 11.39 11.4 blue 10.13 7.70 spectral discrepancy red 3.72 3.46 green 2.54 2.53 blue 2.22 3.35 ergas 2.82 2.98 rase 10.63 11.61 variance red (21.17) 31.83 25.882 green (17.81) 0.002 0.002 blue (10) 3.5 12.5 in this study, the following evaluation factors were used for the qualitative and quantitative analysis of the discussed methods; chi-square test, rmse, snr, sd, ergas, and relative average spectral error (rase) for assessing the fusion results. table 1 shows the fusion result based on case 1 and case 2 of the proposed fusion method. in table 1, columns correspond to the evaluation criteria discussed in section 4, the multispectral band considered and the results of case 1 and case 2 of the proposed method. the rows of table 1 correspond to the evaluation criteria, viz. chi-square test, rmse, snr, sd, ergas, relative average spectral error (rase) as well as the red, green, and blue band data for each criterion used. table advances in technology innovation, vol. 5, no. 3, 2020, pp. 166-181 177 2 shows the performance of the existing fusion methods, viz. his, brovey, sfim, hps, pca, and multiplicative methods under the evaluation criteria mentioned in the previous section. the columns of table 2 correspond to the existing fusion methods listed above, and the rows of table 2 correspond to the different evaluation criteria, viz. chi-square test, rmse, snr, sd, ergas, and relative average spectral error (rase) as explained in section 4. the comparisons made from table 1 and table 2 for the dataset used in this work can be summarized as follows. table 2 performance evaluation of his, brovey, sfim, hpf, pca and multiplication methods evaluation criteria multispectral band ihs brovey sfim hpf pca multiplication chi square test (p value) red 0.92 1 0.56 1 1 0 green 0.09 1 0.99 1 1 0.1 blue 0.15 1 0.97 1 1 7x10 -4 rmse red 5.38 23.5 7.19 16.02 26.27 3.15 green 5.39 33.38 10.12 23.09 32.88 5.65 blue 5.46 31.71 9.56 21.89 43.88 4.51 snr red 5.72 0.43 5.31 1.12 2.28 11.40 green 8.28 0.44 5.36 1.07 2.45 7.58 blue 7.74 0.44 4.85 1.08 2.04 9.24 spectral discrepancy red 4.62 23.24 4.33 15.85 26.05 2.42 green 4.62 33.19 6.08 22.99 32.84 5.19 blue 4.63 31.57 5.77 21.83 43.87 4.02 rase 14.04 233.21 6 31 15 4 variance red (21.17) 58.4 3.05 37.19 17.52 52.49 35.67 green (17.81) 52.31 4.13 53.07 24.68 17.55 42.63 blue (10) 38.04 3.23 50.45 23.51 10.05 41.58 table 1 shows the performance evaluation of the proposed method (two cases) based on the evaluation parameters used. table 2 gives the performance evaluation of the state-of-art methods, namely his, brovey, sfim, hpf, pca, and multiplication methods using the evaluation indices discussed. the evaluation indices discussed in section 4 have been applied to our proposed method and each of the existing method mentioned in the previous section. the evaluation of each methods is done using the same dataset so that a comparative analysis can be made from the results shown in table 1 and table 2. a numerical analysis between the proposed method and the state-of-art methods based on the evaluation criteria mentioned above is as follows. 5.1. chi-square test from table 1 and table 2, it can be seen that case 1 and case 2 of the proposed method give the smallest p-value for all the xs bands. except for the blue band, the multiplication method gives the least p-value of 0.0007. the p-value indicates the probability of getting a better fit. smaller the p-value better the goodness of fit. therefore, there is a small chance that other methods will provide a better fit than the proposed method. on the other hand, brovey, hpf, and pca methods correspond to a maximum p-value of 1 for all the xs bands. therefore, the fit is poorer in those cases compared to the proposed cases. 5.2. root mean square error (rmse) the rmse value indicates the error between the computed and the observed data. it shows how far the computed data deviate from the observed data. larger the value of rmse, greater the error between the dataset points. the rmse for case 1 was found to be 4.75, 4.21, and 4.45, respectively for the red, green, and blue band while case 2 gives corresponding values of 4.43, 4.21, and 5.86. the rmse of the proposed method was found to be similar compared to that of the multiplicative method which gives a value of 3.15, 5.65, and 4.51, respectively for the red, green, and blue band. but the proposed two cases give a lower value compared to the remaining methods. the brovey and pca methods result in the largest rmse value amongst all the methods indicating that the difference between the original xs and the computed xs is very large. for example, the brovey method gives an rmse value of 23.5, 33.38, and 31.71, respectively for the red, green and blue band. advances in technology innovation, vol. 5, no. 3, 2020, pp. 166-181 178 5.3. signal to noise ratio (snr) the snr value gives the amount of information contained in the fused output. the snr is the highest for the proposed two cases with a value of 7.04, 11.39, and 10.13 for case 1; 7.53, 11.4 and 7.70 for case 2 for the three bands respectively. the multiplication method also results in similar values of 11.40, 7.58, and 9.24, respectively for the three bands. brovey, hpf, and pca result in the least snr value. for example, the hpf method results with a lower snr value of 2.28, 2.45, and 2.04 for the red, green, and blue band, respectively. 5.4. spectral discrepancy (sd) a low value of sd indicates a good spectral quality. sd is the lowest for the proposed method followed by the multiplication method, while brovey and pca methods give the highest discrepancies. two proposed cases result in the values of 3.72, 2.54, and 2.22 for case 1; 3.46, 2.53, and 3.35 for case 2 for the red, green, and blue band, respectively. on the other hand, the existing methods give a higher value than the proposed methods. for example, a pca based method gives a value of 26.05, 32.84, and 43.87, respectively for the red, green and blue band. as mentioned above, sd indicates the level of spectral quality of the results. higher the value, lower the spectral information. 5.5. erreur relative globale adimensionelle de synthese (ergas) the ergas values of the proposed two cases compare quite well with a value of 2.82 for case 1, and 2.98 for case 2 with the multiplication method value of 3. the remaining methods give a higher ergas value, for instance, hpf and pca give a value of 23 and 22, respectively. a low ergas value indicates a good fusion output. brovey, hpf, and pca methods result in a maximum ergas value and hence indicate a poor output. 5.6. relative average spectral error (rase) the sfim and multiplication methods give a lower rase value of 6 and 4, respectively against the proposed two cases (case 1 and case 2) with a value of 10.63 and 11.61. the brovey method results in the maximum rase value of 233.31. similar to ergas, a low rase value indicates a good result of the output. in terms of variance matching, for both the two cases of the proposed fusion method, the variance of green bands is almost constant with a very small value of 0.002 against the actual variance of 17.81. however, compared to the other methods, case 2 gives the best approximation for red and blue bands of 25.88 and 12.5 against the original variance values of 21.17 and 10, respectively. ihs gives a larger value than the original multispectral variance. and, brovey method results in the lower variance value compared to the actual multispectral variance. though pca seems to result in a closer approximation to the actual variance for green and blue bands, the result is not significant since other evaluation parameters of pca such as the p-value, rmse, sd, ergas, and rase values do not relate well with its variance data. case 2 of the proposed method gives a better approximation to the actual variance for red and blue bands. there are certain specific cases where the existing fusion methods seem to indicate a better result than that of the proposed method. for example, the p-value for the blue band obtained by the ihs method and multiplication method is smaller than the p-value of the proposed method of case 2. the critical value (cv) obtained through his and multiplication methods are small, so the p-value results are small. the rmse value provided by the multiplication method for the red band is also smaller than that provided by the proposed method. the multiplication transform method also gives a higher snr value for the red band. because the rmse value of the multiplication method is smaller and according to snr definition, smaller rmse results in large snr. lastly, the sfim and multiplication transform methods give a lower rase value compared to the proposed method. however, in summary, it is clear that overall the proposed cases of 1 and 2 outperform all the existing methods. advances in technology innovation, vol. 5, no. 3, 2020, pp. 166-181 179 6. conclusion in this paper, a new pixel value-based image fusion method of a high-resolution panchromatic band and a low-resolution multispectral band were proposed. a linear regression relationship between the panchromatic and multispectral band was formulated. a lagrange multiplier based objective function which seeks to maximize spectral consistency and tracks variance of the given data independently was formed. the results of the proposed pan-sharpening method were compared with a few existing image fusion methods such as his, brovey, sfim, hpf, pca, and multiplication methods on a common set of ikonos satellite data. the comparison was made based on seven evaluation criteria. considering overall performance, the proposed method came out favorably when compared with all other existing methods because the majority of the evaluation indices gave a better result for the proposed method. however, there were a few cases of isolated improved data for the existing methods on certain criteria compared to the proposed method as pointed out under the discussion section. nevertheless, improved results of the existing methods seemed to be sporadic and not supported all criteria, so the performance of the existing methods put into question. hence, it can be concluded that the proposed method outperforms the existing methods based on common data and independent criteria. as future work, it is necessary to study how far the improved performance of the proposed method is robust in the sense that if the input data is changed how will it affect the comparative performance. conflicts of interest the authors declare no conflict of interest. references [1] l. wald, “data fusion: a conceptual approach for an efficient exploitation of remote sensing images,” 2nd international conference, france, pp. 17-24, 1998. [2] c. pohl and j. l. v. genderen, “review article multisensor image fusion in remote sensing: concepts, methods and applications,” international journal of remote sensing, vol. 19, no. 5, pp. 823-854, 1998. [3] t. m. lillesan. w. j. carper, and r. w. kiefer “the use of intensity-hue-saturation transformations for merging spot panchromatic and multispectral image data,” photogrammetric engineering and remote sensing, vol. 56, no. 4, pp. 459-467, april 1990. [4] j. p. chavez and p. s. chavez “comparison of the spectral information content of landsat thematic mapper and spot for three different sites in the phoenix, arizona region,” photogrammetric engineering & remote sensing, vol. 54, no. 12, pp. 1699-1708, january 1988. [5] k. edwards and p. a. davis, “the use of intensity-hue-saturation transformation for producing color shaded-relief images,” photogrammetric engineering & remote sensing, vol. 60, pp. 1369-1374, november 1994. [6] t. m. tu, p. s. huang, c. h. huang, and c. p. chang, “a fast intensity-hue-saturation fusion technique with spectral adjustment for ikonos imagery,” ieee geoscience and remote sensing letters, vol. 1, no. 4, pp. 309-312, october 2004. [7] a. r. gillespie, a. b. kahle, and r. e. walker, “color enhancement of highly correlated images. ii. channel ratio and “chromaticity” transformation techniques,” remote sensing of environment, vol. 22, no. 3, pp. 343-365, august 1987. [8] d. b. yuan, x. hong, s. yu, l. li, and y. zhao, “analysis of four remote image fusion algorithms for landsat7 etm+ pan and multi-spectral imagery,” international journal of online engineering, vol. 10, no. 3, p. 49, 2014. [9] j. qu, y. li, and w. dong, “guided filter and principal component analysis hybrid method for hyperspectral pansharpening,” journal of applied remote sensing, vol. 12, no. 1 pp. 1-18, 2018. [10] m. ghadjati, a. moussaoui, and a. boukharouba, “a novel iterative pca-based pansharpening method,” remote sensing letters, vol. 10, no. 3, pp. 264-273, 2019. [11] m. b. devi and r. devanathan, “pansharpening using data-centric optimization approach,” international journal of remote sensing, vol. 40, no. 20, pp. 7784-7804, 2019. [12] j. p. chavez, s. c. sides, and j. a. anderson, “comparison of three different methods to merge multiresolution and multispectral data: landsat tm and spot panchromatic,” photogrammetric engineering and remote sensing, vol. 57, no. 3, pp. 265-303, march 1991. advances in technology innovation, vol. 5, no. 3, 2020, pp. 166-181 180 [13] g. hong, y. zhang, and b. mercer, “a wavelet and ihs integration method to fuse high resolution sar with moderate resolution multispectral images,” photogrammetric engineering & remote sensing, vol. 75, no. 10, pp. 1213-1223, october 2009. [14] y. zhang and g. hong, “an ihs and wavelet integrated approach to improve pan-sharpening visual quality of natural colour ikonos and quickbird images,” information fusion, vol. 6, no. 3, pp. 225-234, september 2005. [15] s. ourabia and y. smara, “a new pansharpening approach based on nonsubsampled contourlet transform using enhanced pca applied to spot and alsat-2a satellite images,” journal of the indian society of remote sensing, vol. 44, no. 5, pp. 665-674, october 2016. [16] a. svab lenarcic and k. oštir, “high-resolution image fusion: methods to preserve spectral and spatial resolution,” photogrammetric engineering and remote sensing, vol. 72, no. 5, pp. 565-572, may 2006. [17] c. yang, q. zhan, h. liu, and r. ma, “an ihs-based pansharpening method for spectral fidelity improvement using ripplet transform and compressed sensing,” sensors, vol. 18, no. 11, pp. 1-20, november 2018. [18] t. wang, f. fang, f. li, and g. zhang, “high-quality bayesian pansharpening,” ieee trans image process, vol. 28, no. 1, pp. 227-239, january 2019. [19] k. nikolakopoulos, “comparison of nine fusion techniques for very high resolution data,” photogrammetric engineering & remote sensing, vol. 74, pp. 647-659, 2008. [20] h. li, l. jing, and y. tang, “assessment of pansharpening methods applied to worldview-2 imagery fusion,” sensors, vol. 17, no. 1, pp. 1-30, january 2017. [21] e. i. ulzurrun, c. g. martin, j. m. ruiz, a. g. pedrero, and d. r. esparragon, “fusion of high resolution multispectral imagery in vulnerable coastal and land ecosystems,” sensors, vol. 17, no.2, pp. 1-23, february 2017. [22] h. li, l. jing, y. tang, and h. ding, “an improved pansharpening method for misaligned panchromatic and multispectral data,” sensors, vol. 18, no. 2, pp. 1-16, february 2018. [23] c. chen, y. li, w. liu, and j. huang, “image fusion with local spectral consistency and dynamic gradient sparsity,” 2014 ieee conference on computer vision and pattern recognition, pp. 2760-2765, 2014. [24] x. meng, j. li, h. shen, l. zhang, and h. zhang, “pansharpening with a guided filter based on three-layer decomposition,” sensors, vol. 16, no. 7, pp. 1-15, july 2016. [25] a. sekrecka and m. kedzierski, “integration of satellite data with high resolution ratio: improvement of spectral quality with preserving spatial details,”, sensors, vol. 18, no. 12, pp. 1-22, december 2018. [26] c. ballester, v. caselles, j. verdera, and b. rougé, “a variational model for p+xs image fusion,” international journal of computer vision, vol. 69, pp. 43-58, 2006. [27] f. fang, f. li, c. shen, and g. zhang, “a variational approach for pansharpening,” ieee transactions on image processing, vol. 22, no. 7, pp. 2822-2834, 2013. [28] m. möller, t. wittman, a. l. bertozzi, and m. burger, “a variational approach for sharpening high dimensional images,” siam journal on imaging sciences, vol. 5, no. 1, pp. 150-178, 2012. [29] l. j. deng, g. vivone, w. guo, m. dalla mura, and j. chanussot, “a variational pansharpening approach based on reproducible kernel hilbert space and heaviside function,” ieee trans image process, vol. 27, no. 9, pp. 4330-4344, september 2018. [30] c. kwan, j. h. choi, s. h. chan, j. zhou, and b. budavari, “a super-resolution and fusion approach to enhancing hyperspectral images,” remote sensing, vol. 10, no. 9, pp. 1-28, september 2018. [31] y. qu, h. qi, and c. kwan, “unsupervised and unregistered hyperspectral image super-resolution with mutual dirichlet-net,” computer vision and pattern recognition, 2019. [32] r. borsoi, t. imbiriba, and j. carlos moreira bermudez, “super-resolution for hyperspectral and multispectral image fusion accounting for seasonal spectral variability,” ieee transactions on image processing, vol. 29, pp. 116-127, july 2019. [33] h. aanæs, j. r. sveinsson, a. a. nielsen, t. bovith, and j. a. benediktsson, “model-based satellite image fusion,” ieee transactions on geoscience and remote sensing,” vol. 46, no. 5, pp. 1336-1346, 2008. [34] h. aanaes, a. a. nielsen, t. bovith, j. r. sveinsson, and j. a. benediktsson, “spectrally consistent satellite image fusion with improved image priors,” proc. of the 7th nordic signal processing symposium-norsig 2006, ieee press, janury 2006, pp. 14-17. [35] g. cliche, f. bonn, and p. teillet, “integration of the spot panchromatic channel into its multispectral mode for image sharpness enhancement,” photogrammetric engineering & remote sensing, vol. 51, pp. 311-316, 1985. [36] x. meng, h. shen, h. li, l. zhang, and r. fu, “review of the pansharpening methods for remote sensing images based on the idea of meta-analysis: practical discussion and challenges,” information fusion, vol. 46, pp. 102-113, 2018. advances in technology innovation, vol. 5, no. 3, 2020, pp. 166-181 181 [37] p. ghamisi, b. rasti, n. yokoya, q. wang, b. hofle, l. bruzzone, and p. m. atkinson, “multisource and multitemporal data fusion in remote sensing: a comprehensive review of the state of the art,” ieee geoscience and remote sensing magazine, vol. 7, no. 1, pp. 6-39, march 2019. [38] g. vivone, l. alparone, j. chanussot, m. dalla mura, a. garzelli, g. a. licciardi, and l. wald, “a critical comparison among pansharpening algorithms,” ieee transactions on geoscience and remote sensing, vol. 53, no. 5, pp. 2565-2586, december 2014. [39] s. kahraman and a. erturk, “review and performance comparison of pansharpening algorithms for rasat images,” journal of electrical & electronics engineering, vol. 18, no. 1, pp. 109-120, 2018. [40] r. l. plackett, “international statistical review”, vol. 51, pp. 59-72, 1983. [41] m. fallah and a. azizi, “quality assessment of image fusion techniques for multisensor high resolution satellite images (case study: irs-p5 and irs-p6 satellite images),” in: wagner w., székely, b. (eds.): isprs tc vii symposium – 100 years isprs, vienna, austria, vol. 38, january 2010. [42] yuhendra, i. alimuddin, j. t. s. sumantyo, and h. kuze, “assessment of pan-sharpening methods applied to image fusion of remotely sensed multi-band data,” international journal of applied earth observation and geoinformation, vol. 18, pp. 165-175, 2012. [43] t. ranchin, l. wald, and m. mangolini, “the arsis method: a general solution for improving spatial resolution of images by the means of sensor fusion,” fusion of earth data, cannes, france, pp. 53-58, february 1996. [44] c. myungjin, “a new intensity-hue-saturation fusion approach to image fusion with a tradeoff parameter,” ieee transactions on geoscience and remote sensing, vol. 44, no. 6, pp. 1672-1682, 2006. [45] a. vesteinsson, j. r. sveinsson, j. a. benediktsson, and h. aanaes, “spectral consistent satellite image fusion: using a high resolution panchromatic and low resolution multi-spectral images,” proc. 2005 ieee international geoscience and remote sensing symposium, vol. 4, pp. 2834-2837, july 2005. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  advances in technology innovation, vol. 6, no. 1, 2021, pp. 47-66 an overview of own tracking wireless sensors with gsm-gps features hamed yassin-kassab, marius-corneliu rosu * applied electronics and information engineering, politehnica university of bucharest, romania received 27 september 2019; received in revised form 28 december 2019; accepted 20 january 2020 doi: https://doi.org/10.46604/aiti.2021.4793 abstract wireless sensors (ws) mobility and pause time have a major impact directly influencing the energy consumption. lifetime of a ws network (wsn) depends directly on the energy consumption, thus, the hardware and software components must be optimized for energy management. this study aims to combine a compact hardware architecture with a smart energy management efficiency in order to increase ratio lifetime/energy consumption, to improve the operating time on a portable tracking system with gps/gsm/gprs features and own power. in this paper we present the evolution of own ws tracking architecture with gps/gsm/gprs features, basic criterion being the lifetime combined with low power consumption. concern was focused on hardware and software areas: large number of physical components led to reconsideration of hardware architecture, while for software, we focused on algorithms able to reduce the number of bits in transmitted data packets, which help to reduce energy consumption. the results and conclusions show that the goal was achieved. keywords: ws, gps/gsm/gprs features, portable devices, operating lifetime/energy consumption 1. introduction a typical embedded system involves the following main features: sense, process and store, transmit, and power management as shown in fig. 1, while the first order radio energy consumption model is shown in fig. 2 [1-2]. since computers and software have made their presence in every component and as part of communication and automation, the structure described above has changed, and concern has materialized in both hardware and software direction. as electronic components manufacturing technology has evolved, specialized companies appeared in achieving their specific features dedicated to the purpose of use; thus, at least hardware designers can choose components with low noise and minimal power consumption. in the software part, the remote data transmission algorithm has focused on minimizing the number of usable bits for data packets, the next packet following will be transmitted or not, depending on the decision taken locally in the ws, so that the active transmission time would be shorter, which reduces the energy effort in the sensor. ideally, we want a network to have a long life and to operate unattended, but there are factors that limit the capacity of the power source. the issue of energy consumption has been divided into two different directions: first, a large number of efficient energy communication protocols have emerged, including mac protocols, routing and self-organizing, which take into account the particularities of wireless sensor networks (wsn). secondly, local dynamic power management (dpm) strategies are developed to detect and minimize the impact of inefficient activities on an individual node. 1.1. energy model description as already mentioned in the introduction, the model used for radio energy dissipation is shown in fig. 2 and, according with literature, it is a first order radio model [3-4]; the definitions used parameters are shown in table 1 [3]. * corresponding author. e-mail address: marius-corneliu.rosu@sdettib.pub.ro tel.: +4-0768-957-856 advances in technology innovation, vol. 6, no. 1, 2021, pp. 47-66 48 fig. 1 typical embedded system fig. 2 energy dissipation model in radio transmission table 1 graph representations parameter definition unit eelec energy dissipated during transmission / reception 50 nj/bit εfs free transmitter amplifier model 10pj/bit/m 2 εmp multi-path model of the transmitter 0.0013pj/bit/m 4 k data length 2,000 bits d0 threshold for distance √𝜀𝑓𝑠/𝜀𝑚𝑝 m when k-bits message is transmitted over a distance a d, it will consume an amount of etx energy according with eq. (1) [5]: 0 4 0 , ( ) , s fs f 2 elec tx elec k e + k ε d if d < d e k,d = k e + k ε d if d d (1) and the erx energy needed to receive the transmitted message is given by relation eq. (2) [3] ( )rx elece k = k e (2) to forward this message, the amount of efx energy needed is given by expression eq. (3) [3], where d0 it is defined in table 1 and expresses the threshold for changing the amplification patterns. 0 4 0 , ( , ) ( , ) ( ) , fs fs 2 elec tx rxfx elec 2k e + k ε d if d < d e k d = e k d + e k = 2k e + k ε d if d d (3) 2. background and related work 2.1. background in this section we will briefly review the theories and technologies relevant to the subject of this paper. as autonomous devices, available energy and other resources are limited in size and cost, being directly dependent on the range of available energy (e.g. battery energy density or external energy capture devices); computational resources, storage and communication, all of these vary from one system to another. the design challenges for such ws devices and are illustrated in table 2. table 2 design challenge issues design issues determined factors dynamics of the network mobility of ws, target and sink for ws node location deterministic or ad-hoc inter-node communication direct or multi-hop communication data transfer models continue, event-triggered, on demand, hybrid data aggregation internal network (partial total) or network external tracking systems are a way to monitor, detect, and then notify users of certain activities. such systems use different types of technologies that are covered by the internet of things (iot). to determine the most accurate location of an object, the global positioning system (gps) represents the main technology for such purpose. advances in technology innovation, vol. 6, no. 1, 2021, pp. 47-66 49 a gps tracking system involves three layers [6]: (1) space layer, composed of at least 24 geostationary satellites orbiting twice a day; (2) control layer, consisting of 5 stations across the globe, monitors and manages satellites by updating orbital data and clock correction. (3) user layer, consisting of all kinds of gps receivers; they have the ability to receive data from satellites. once the receiver sets its position with the gps satellites, the information is usually delivered in one of the following formats: 1. latitude and longitude and 2. universal transverse mercator (utm); gps receiver provides location and time information about any object anywhere, anytime, as long as the communication path is not obstructed by any obstacle. when ws is in contact with at least three satellites, two-dimensional position (latitude and longitude) is determined by triangulation method [7]; we must remember that the satellite emits a signal with a certain frequency. if four or more satellites are detected by our ws, then it is able to calculate the position in three dimensions: latitude, longitude and altitude; after that it is possible to calculate vehicle’s average speed and moving direction. gps technology has the following drawbacks [6-7]: (1) they consume large amounts of energy; (2) it is not efficient for indoor locations; (3) it needs a free line of sight for at least three satellites; (4) detects position approximately; (5) yield is weaker in crowded areas. however, simultaneously it is also necessary the access by mobile communication network gsm/gprs who made possible online tracking of the system; we make it clear that gps tracking uses the global satellite network, while gsm/gprs is the mobile communications terrestrial network. a clarification in this respect can be found by the reader in table 3 [8]. similar systems have been designed and built by researchers, tailored to the specific requirements and purpose pursued (e.g. in the health field). there are five major classes, generally accepted techniques for energy efficiency in ws, summarized in tables 4-5 and shows how each technique addresses energy-consuming processes. use of energy must be tightly controlled; energy efficiency means adapting the power cycle to the node's operating mode. table 3 comparation gps–gsm/gprs terms of comparison gps gprs semantics global positioning system general packet radio service scope it provides positioning services offers voice&data services for mobile telephony applications navigation, surveillance, mapping, gis, and others access on e-mail, multimedia messages, video calls, etc. operating communicates with a collection of satellites orbiting the earth. communicates with a terrestrial tower. minimum number of stations 3 1 use gps is used anywhere: sky, earth, water, etc. gprs limited range and available only on land. costs expensive economic advances in technology innovation, vol. 6, no. 1, 2021, pp. 47-66 50 table 4 design challenge issues techniques description data reduction reducing the amount of produced, processed and transmitted data. examples: data compression and aggregation. reducing the protocols size (rps) goal: increase protocol efficiency by reducing overall costs. transmission periods, adapted to network stability or distance to the transmission source. examples: stratification approach or avalanche optimization of transmitted data. efficiency routing routing protocols must be developed so that the network life will grow by reducing energy transmission and avoid nodes with low residual energy. examples: opportunistic protocols taking advantage of node mobility; protocols using geographic coordinates of nodes; protocols that hierarchize nodes to simplify routing; protocols that send data only to nodes interested in avoiding unnecessary transmissions. cycle of service express the fraction of times when the nodes are active; active / inactive periods must be tailored to the application-specific requirements, which can be further subdivided. they are related to the medium access protocol. examples: 1. techniques centered on selecting active nodes in wsn; 2. techniques that perform the on / off mode of active node radios when communication involves the node. topology control adjusts transmission power, maintaining network connectivity; a low topology is created based on local information. table 5 techniques influence on wsn processes stage techniques sensing processing communication collision over-listening package control passive listening interference data reduction: major major secondary secondary secondary rps major secondary secondary major secondary efficiency routing major secondary major secondary major cycle of service major major major major secondary major major topology control major major major major 2.2. related work ws development has facilitated the emergence of increasingly diverse applications. however, the issue of energy and how to maximize the life of the network in a very limited budget of energy, are still the design priorities. as we have shown in table 4, five main classes are considered energy-consuming. next, there are mentioned the techniques that address the energy efficiency challenge in the literature, divided into the five classes corresponding to table 4; a brief reference will also be mentioned. a. regarding data reduction – literature proposals are divided into 3 main categories depending on the data processing stage: acquisition, processing, and communication. b. regarding protocols size – schemes to reduce the size of protocols and implicitly their costs, lead ultimately on saving the limited energy resources. c. regarding efficiency routing – raises problems with the design of routing protocols for a network; they seek load balancing, minimizing energy consumption for end-to-end transmission of a packet, avoiding low-energy nodes. d. regarding cycle of service – are techniques for planning activity alternating the active and inactive periods to turn off the radio subsystem whenever possible, and providing an operational network for application. e. regarding topology control algorithms – refers to building and maintaining a low topology that preserves energy while maintaining network connectivity and coverage as shown in [9]. there is an optimum distance transmission that minimizes energy dissipation while keeping a connected topology, as shown in [10]; methods presented in [11] are listed. in table 6 we made a brief review of data reduction solutions for earlier mentioned classes, as we have found in papers. advances in technology innovation, vol. 6, no. 1, 2021, pp. 47-66 51 table 6 state of the art in energy consumption reduction main classes processing steps techniques achievements a data reduction data acquisition 1. based on sampling [12-15] 2. based on prediction in [16] centralized prediction; in [17] prediction is done by grouping sensors; in [18] describes the communication protocol with neighbors. data processing 1. data compression several techniques: command encoding; pipe-line or dis31bution compression, in [19]. 2. data aggregation categories: structure cluster cu protocols leach in [20-21], pegasis in [22]; structure tree with the algorithm appointed dctc in [23-24]; in [25] low cost tree; in [26] and in [27], based smt and mst. without structure, in [28] for dynamic events, or tod in [29], a scalable technique. data transmission they are related to data acquisition and processing techniques. main classes reduction method techniques achievements b protocols size through regular adaptive transmission 1. based on adaptation to network changes in [30], it is suggested to adapt the message period to network stability. in [31] trickle algorithm presents a more complex approach. 2. based on distance adaptation to the information source. in [32] basic idea is the concept fish eye and the period of transmission increases with distance from the source. by stratified optimization 1. top-down approach all techniques listed for sensors with limited resources were given in [33]. 2. the bottom-up approach 3. apps-centered approach 4. mac-centered approach 5. integrated approach by optimizing floods 1. based on multiple relocations it has been found to be more difficult to find for most graphs a set of minimally connected nodes [34]. 2. based on dominant domains (dd) in [35-37] a heuristic distribution is used, where a dominant set is originally built and then redundant nodes are removed. other authors, [38-39], use a leading node to assign a rank to each node. 3. based on neighborhood negotiations in [40], any data is described by a descriptor called metadata, which is unique and shorter than actual data. main classes reduction method techniques achievements c efficiency routing by data-driven protocols technology spin [40] directed diffusion technique (did) [22] technology cadr, cougar. [22] by hierarchical protocols technology leach [22] technology pegasis [22] technology teen [22] technology apteen [22] advances in technology innovation, vol. 6, no. 1, 2021, pp. 47-66 52 table 6 state of the art in energy consumption reduction (continued) main classes reduction method techniques achievements c efficiency routing by geographic protocols technology gear [22] technology gaf [22] technology speed [20] by opportunistic protocols based on spatial and natural diversity. [41] based on mobility. variants based on router node mobility protocols are proposed in [42-45]. protocols based on mobile relays are in [46-48]. main classes reduction method techniques achievements d cycle of service high level granularity linear programming technique. [49-50] gaf technique [22] span technique [51-54] low level granularity based on “time division multiple access” (tdma) proposed in: [55]–trama; [56]-flama; [57]-flexitp; [58]-asap. based on “sensor media access control” (smac) variants: [59]-sync messages; [60]-smac; [61]-dmac; [62]-bmac. hybrid between tdma and “carrier sense multiple access” (csma) variants: [63]zmac; [64] protocol tdma/ca based on colors nodes provided by serena main classes reduction method techniques achievements e topology control algorithms directed lmst (dlmst) dlmst [11] directed rng (drng) drng [11] residual energy aware dynamic (read) read [11] on the other hand, researchers have developed systems with different design hardware, performing the same task. we focused on real-time detection for location and speed, and in addition, the vehicle temperature and to require the imposition of new features and constraints on the measured parameters. industry has developed around an ever-changing technology with a wide range of applications aimed at the environment, health systems, military and robotics, as we have shown in table 6. typically, such applications use a large number of nodes in a self-organizing multi-hop wsn, running on batteries and equipped with one or more sensors, as well as low power processors and transceivers. in this section we analyzed the main factors who contributes on reduction of wsn's autonomous operation time, the factors that influence the architecture of the networks, the energy problems that may occur and the state-of-the-art achievements in the field; we have also developed a classification of energy efficiency techniques by describing the algorithms that lead to the reduction of energy consumption which were synthesized in table 6. in addition, for a complete analysis of real-time tracking devices, we recommend to the interested readers to consult chapter3 from reference [65]. instead of conclusions: it is obvious that the domain is still open to researchers, the subject of consumption and limited resources over time being more present than ever. 3. hardware development a node consumes energy and most of it, with values between 60% and 80% [66-67], during communication. at the same time improving energy storage has an annual average growth rate of 6%, as stated in [68]. due to many constraints, the advances in technology innovation, vol. 6, no. 1, 2021, pp. 47-66 53 complexity of nodes development is limited. depending on applications, nodes can be designed with enhanced local decision capacity or not. as we pointed out in the introduction, the paper presents the evolution of our own ws, with gps gsm/gprs features; the design and improvement criterion were the stand-alone operating time by reducing energy consumption. our focus is on both hardware and software sides, and figs. 3-5 show the evolution on the hardware enhancement component. as can be seen, the large number of physical components in fig. 3 led to a reconsideration in the design concept (see figs. 4 and 5). (a) gps side view (b) gsm/gprs side view fig. 3 first hardware solution the demand for choosing the electronic components in the following modules was reported in the energy consumption balance of the active components (see appendix 2 with datasheet reference). overall the entire hardware system, besides module from this work, includes the following units: (1) the mobile units specialized in “machine to machine’’ tracking; (2) the alarm station; (3) hardware systems to optimize energy consumption and battery charge that ensures autonomy of ws; (4) hardware analysis and filtering systems for analogue and/or digital signals. the module is based on the dual gps-gsm/gprs connection, and performs remote positioning and target monitoring via sms or internet; it has statistical features for mileage, detection and alarm when power is interrupted, as well as other customizable features. external interfaces make it possible to connect a radio frequency identifier (rfid), to a serial port, a fuel sensor, an led display, etc. the modules in this paper have been modified for improved performance particularly energy consumption and thus increase autonomy in operation. however, they need to be explained as follows: fig. 3 shows the first prototype, made on double front wiring, one dedicated to gps communications (fig. 3(a)), the other side to gsm communications (fig. 3(b)). as a result of increasing number of simultaneous surveillance and different manufacturers for gps and gsm modems (gps-u-blox, gsm-huawei, cpu-philips), rewriting of the programs in modules according to the adaptations needed for the transmission and reception was not fully compatible and satisfactorily, optimization and technical compatibility not being 100% possible at that time. in addition to this variant we didn’t take into account the optimization of the data transmission, the main requirement being the stability of the transmission and its non-stop maintenance. in an analysis of the total energy consumption, we found that it exceeded 250ma / h, which, subject to the requirement of autonomy, was limited to a maximum of 24 hours. fig. 4 it is the intermediate step of improving the first module, in the sense of redesigning the double-sided wiring, the use of the same manufacturer for gps and gsm modems and an advanced risc machine (arm) microprocessor (μp). advances in technology innovation, vol. 6, no. 1, 2021, pp. 47-66 54 fig. 4 the second hardware solution fig. 5 the third hardware solution however, there were several drawbacks related to energy consumption (80ma max gsm modem / h, max 30ma gps / h and μp max 15ma / h relative to a supply voltage of 3.3v), but also in local limitation at μp level of data processing before the transmission, and with this version we have succeeded, within 30 days, to deliver without loss of quality and accuracy, only 5mb of data, thanks to software development supported by μp. full description can be found in [5]. since energy efficiency remains the main theme, we continued to improve the module at hardware point of view. fig. 5 represents the new hardware concept that improves the life of its own power supply (if it is used to continuously track container shipping). also, through a physical interface, it can be linked to a hardware and software management interface for data and alarms transmission and reception, thus becoming a coordinating node in a network topology. the design concept is also a double face circuit, the two components responsible for the gps-gsm/gprs communication are from u-blox with a consumption of 5ma/h in stand-by; entire assembly consumes 75ma/h in transmission-reception mode, which satisfies one of the objectives stated in paper. equipped with a μp from microchip (pic32mx775) with a core 80mhz/32-bit mips 105dmips m4k ® [69], we were able to write a program to detect all events that required a "wake-up" decision from sleep mode, to perform all instructions for data transmission or reception, consumption in sleep mode being determined insignificant (5ma). another aspect of hardware design, as we have already noted, is the quality of the active components used; so, using the processor at a minimum computing power and working frequency, the results are reflected in improving the overall energy balance because, as we said, energy consumption increase with increasing work frequency and computing power. 4. the software application table 7 loop code distribution [5] original code modified code int a[n], b[n], c[n], i; for( i = 5; i < 3500; i += 50 ) c[i] = val; for( i = 5; i < 2000; i += 100 ) b[i] = val; for( i = 5; i < 1000; i += 100 ) a[i] = val; for( i = 0; i < n; i++ ) { for( j = 0; j < n; j++ ) { a[j] = a[j] + c[i]; b[j] = b[j] + c[i]; } } int c[n], a[n], b[n], i; for( i = 5; i < 3500; i += 50 ) c[i] = val; for( i = 5; i < 2000; i += 100 ) b[i] = val; for( i = 5; i < 1000; i += 100 ) a[i] = val; for( i = 0; i < n; i++ ) { for( j = 0; j < n; j++ ) { a[j] = a[j] + c[i]; } } for( i = 0; i < n; i++ ) { for( j = 0; j < n; j++ ) { b[j] = b[j] + c[i]; } } advances in technology innovation, vol. 6, no. 1, 2021, pp. 47-66 55 allows mobile units to be viewed by using maps; the resident program in mobile units gps/gsm/gprs or gsm/gprs is written in the programming language c/c++/java, while for the communications server we only used the java programming language. both programs, from mobile units and communication server, perform specific functions. simultaneously, for the remote data transfer algorithm we focused on minimizing the number of usable bits in data packets; next data packet is transmitted or not, depending on the decision taken locally in ws, the decision being in fact only on the data difference calculated by the sensor between the transmitted and the current packet. such transmission duty cycle is shortened, leading to a decrease in the sensor energy effort. a very useful method to decrease energy consumption is also optimizing the code. as we shown in [5], there are many optimization techniques; of all of them, we expose to the reader a modified programing code solution so that it will require a minimum cpu usage (see table 7) [5]. the costumed software solutions call a few tricks and even though at first glance it gives the impression that it is laborious; in its decipherment it can be seen that by intelligent use of structured programming blocks that accelerates work instructions specific to the programming language (interrupts, semaphores, timers, reducing execution time, loops technique by vector addition and distribution loops, reducing the instantaneous power, cache memory usage, storage 0 instead of 1), simultaneously with tailored software solutions that minimize transmission time (special protocols, data compression, eliminate transmission errors, choosing optimal level of security), the use of cpu is directed to the minimum calculations required. 5. gps tracking system real-time tracking with one of the modules described in section 3, are made according to the diagram in fig. 6: internet access to the system is via web/ftp servers that are loaded with a series of html pages containing “server-side” scripts and “client-side” scripts, which are actually files such as xml, css, swf, etc. fig. 6 real-time gps tracking system note: alarms generated by mobile units or fixed gprs communicators are transmitted by dual gprs / gsm; so we have the guarantee that the alarms will be dispatched regardless of the provider's gsm / gprs network; in this way the success rate can be up to 100%. we remind that gps tracking uses the global satellite network, while gsm / gprs mobile terrestrial network. these servers are designed and configured so that they can perform certain functions: (1) to communicate bidirectionally with java servers through: commands, messages, events, gps positions, data base (db) information etc., with the purpose of mediating customer-car links. advances in technology innovation, vol. 6, no. 1, 2021, pp. 47-66 56 (2) communicate with db to get mobile unit reports, db personalization by customers, viewing routes, positions, on-line journeys, points of interest, areas of interest, events, etc. (3) communicate with other users (customers, operators, administrators, etc.) through programs that are localized to the administrator or clients, for mobile unit management, back-ups, log management, etc. (4) to configure, display and maintain the maps functionality. (5) to execute the download over the air (dota) operation through ftp protocols. (6) ensure the possibility of on-line uploading, directly by clients, of back-up files. 5.1. the software system enables mobile units to be viewed using maps, which have the following features: (1) vectorized format, which makes their size to be very small (under 100kb each). (2) they are always on the client's computer, so they "move very fast" to any mouse action (zoom, left / right, up / down). (3) objects displayed on the map are disjoint, so size map remains the same no matter how many items are displayed on this. (4) both the server and client computer information, can be displayed on a map, at the same time. (5) at any change of displayed information, the map remains the same, the refresh sensation being insensible. (6) regardless of users’ number accessing the same map, it will "move" as fast as it would on the client's computer. (7) the map can display simultaneously in real time, approx. 1000 (one thousand) cars. (8) maps can be displayed on any pc, laptop, pda or smart phone. (9) any change in the field will be done very quickly on the appropriate maps, because the software required for editing and georeferencing is proprietary. (10) it can compile a map in its own format for any city in the world in a few days. (11) each map has an associated chart network (corresponding to the access paths) through which may represent routes or find optimal paths between points; the software needed to create this network as proprietary code. (12) georeferencing of maps is also done using proprietary software code. (13) access to information is the rule type user / password / rights. (14) optional, it can be implemented and viewed on google maps; on request, any public maps, for which the client provides user ids (yahoo, open street, garmin, google etc.), can be used. as practical implementation at national level, we scaled romania map at 1: 25,000, being able to realize various views (e.g. view by time and geographic area). 5.2. fleet/machine to machine monitoring typically, the tracking unit program algorithm passes through three steps: measuring, recording and transmitting data over the gsm network; maximum power consumption is recorded in the last round when data is transmitted via gsm network. to reduce energy consumption, the workflow implemented on the tracking unit goes through five stages. to the three steps mentioned we introduced two more sequences, memorize and compare, designed to reduce data transmission stage. advances in technology innovation, vol. 6, no. 1, 2021, pp. 47-66 57 how it works? in the step memorizes, μp stores the output of the sensors as variables in its memory; then in compare step, the new values of the acquired data are compared with those in the previous step. ex.: for geographical azimuth (latitude and longitude) the new values measured are compared with those already previously stored; if they are identical, then compare other physical values (temperature): if the difference between two consecutive measurements does not exceed a default threshold, then the data is not transmitted, otherwise the data is transmitted. monitoring is done in real time, data transmission from the mobile unit to the servers being made by gps / gsm / gprs continuously 24h / 7 days. pseudocode algorithm that guide in 5 steps shows as follows: measures temperature, position and speed; if position is identical to the previous position; if temperature is identical to the previous temperature; if the difference between the previous temperature and the current temperature is within the threshold; do not transmit data; else transmit data; else transmit data; for more detailed analyzes, besides the website, there is a special monitoring software with capabilities in car behavior analysis. alarms generated by mobile units or fixed gprs communicators are transmitted dual gsm / gprs.; success rate of alarms as they reach dispatch center, regardless of mobile communication network gsm / gprs, is approaching by 100%. the application runs on any computer with internet connection and windows. a large number of monitors can be installed and this does not affect system load, or limit real-time access to monitored vehicles or fixed targets; a remote tracking application we also described in [70] the main interface of the screen control, is divided into two divisions: self-monitoring track division and the alarms monitoring division from either monitored vehicles or fixed targets viewing mobile units is done using maps that provide information about mobile unit positions in several variants: routes, current positions, all in real time, for groups or subgroups of cars, or individual cars in any of the above-mentioned variants. note: all programs that use maps are proprietary code and maps are loaded in vector format from google maps! 5.3. system maintenance for system maintenance, a series of dedicated programs are used, that perform database management functions, traffic analysis, information stored in the system, georeferencing, associated network drawing maps, server monitoring and backup. these are able to perform both functions of management activities, as well as analysis and providing reports; figs. 7 and 8 are such examples for database management functions. these mean the following: fig. 7 is a report generated for a car for 24 hours. the upper part is a complete report, while the two-column table is the synthesis of the day. fig. 8 is a 24 hours summary for a client who owns a fleet of 27 cars; green means stationary, blue stops, and movement ocher. in addition, fig. 8 has a “blind zone” column, which means that the vehicle could not be located by gps. the other data are easy to understand. advances in technology innovation, vol. 6, no. 1, 2021, pp. 47-66 58 fig. 7 real-time gps detailed report for one car fig. 8 real-time gps detailed cars report 6. experiments and results we have already mentioned in the previous section, that in tracking algorithm we introduced two more intermediate sequences aimed at reducing energy consumption for the device. the determinations were made on two levels: hardware and software. in this section we have summarized the main results. 6.1. the hardware development the experiment was designed to measure the energy consumption of the tracking device. for this purpose, we used the monsoon power monitor (mpm) hardware to determine energy consumption [71], the second and third hardware solution (figs. 4 and 5), and the appropriate mpm graphic user interface (gui) available at [72], installed on a laptop with an intel advances in technology innovation, vol. 6, no. 1, 2021, pp. 47-66 59 centrino 2 duo processor, 8gb of ram, and windows 10 pro os.; the description of gui software for these measurements is provided also in [71]. the graph gui scale was set as follows: (1) time (bottom) was shown every 100ms, with 10 units gap. (2) on left side power, with a scale gap of 500mw. (3) on right side current, with a scale gap of 500ma. and the measurements results are shows in fig. 9. the battery size was 550mah, which is for a 9v battery, on display is show the number of samples, the total consumed energy, average power and the average tension for the entire duration of the measurement. fig. 9 power measurements results 6.2. the software development in this section we have summarized the main results for the software applications. the following figures highlight comments made in sections 4 and 5 of this paper; fig. 10 is a custom gui for logging into the gps tracking system, written as we said in java. fig. 10 gui logging interface in section 5 we said that real-time tracking is done by gps and communication of positions through gsm/gprs; here too we mentioned that through the proprietary software we were able to build our own tracking maps, with different levels of resolution and accuracy. in figs. 11-13 maps are carried on 3 levels of accuracy, from general to detail on the area of interest (fig. 13 is the city of bucharest). in fig. 14 we gave a detail plan; is from faculty of electronics (we noted). advances in technology innovation, vol. 6, no. 1, 2021, pp. 47-66 60 fig. 11 level 1 accuracy of own tracking maps fig. 12 level 2 accuracy of own tracking maps fig. 13 level 3 accuracy of own tracking maps fig. 14 detail of level 3 accuracy of same map fig. 15 geofencing results fig. 16 real-time position repot fig. 17 real-time travel routes advances in technology innovation, vol. 6, no. 1, 2021, pp. 47-66 61 regarding real time mobile units viewing and tracking, we have implemented our own program that allows us to restrict the area of traffic by drawing the displaced area of movement (general called geofencing), being physically visible on the map. fig. 15 illustrates such a restriction points that is actually the fence limiting movement; if the border is passed, the system sends sms alert to administrator, so permanent surveillance on pc monitor is not necessary. the particularity towards other similar achievements is the use of our own maps rather than google's. the second is the possibility to choose how to delimit the allowed area: in fig. 15 we showed only the punctuation of boundaries, which are linked to each other, and also through physical marking an area of travel which can be adjusted by simply dragging whit mouse. our programs are also written to generate a series of reports (of which two have already been mentioned in figs. 7 and 8). another thought-based and implemented feature is the tracking of real-time positioning of the device being tracked; by the colors allocated for various situations, the administrator, after a certain time of use, can realize the deviations from the imposed rules. fig. 16 illustrates such a report, and in fig. 17 we have illustrated another type of query that shows the movement of a fleet on a certain day and pre-set hourly. 7. conclusions this paper presents detailed aspects related to progress of our own wn tracking device, with gsm / gps / gprs features. thus the concern of improvement and optimization of energy consumption had aimed on both hardware and software component. developed hardware modules are those of section 3, figs. 3-5, specifying that they work in real time, and the side of software applications is done in section 4. we show that hardware solution from fig. 5 represents the new concept that improves the life of its own power supply and provides the ability to bind to a hardware and software management dispatch system, for data and alarm transmission and reception, thus becoming a coordinating node in a network topology. as far as software design is concerned, we have looked for the management and handling programs to be as easy and simple to use as possible, that's why the programs needed for the mobile units were written in combined language c / c ++ / java, and for servers only java programming language. we adopted the differential transmission strategy for significant data packets, the difference measurements are calculated in mobile units and only data that is perceived by the software as alerts are transmitted; the user is responsible for choosing the communication interface and defining all related metrics. we introduced two other intermediate software stages in the data transmission, described by the pseudocode on subsection 5.2 thus managing to improve the balance of energy consumption; system maintenance is done by dedicated programs that perform tasks and database management. section 6 describes the results obtained for both hardware and software solutions. in subsection 6.1 we made a brief description of the experiment, results and methods used in the hardware area, while subsection 6.2 is allocated to software results, in fact much more spectacular in our opinion. the programs for which we presented the results of subsection 6.2 are proprietary code, runs on any computer with an internet connection and windows system and not limit real-time access to monitored vehicles or fixed targets. particularities with respect to other similar achievements in the field: 1. the use of own maps adapted after google; 2. the possibility to choose the delimitation mode for the allowed area. however, we consider the following achievements for further development: (1) an architecture for a new device that includes a wifi module, bluetooth and internal battery; (2) wireless charging of electric cars; advances in technology innovation, vol. 6, no. 1, 2021, pp. 47-66 62 (3) capturing energy from surrounding electromagnetic sources that deliberately transfers energy from: radio frequency, solar and wind energy, thermal energy; (4) application in the existing intelligent buildings without additional wiring, a wireless monitoring system based on passive infrared sensors (pir); (5) using thermal energy dissipated by wireless devices for generating electricity. (6) we can develop our application for any other type of real-time tracking. the paper present our efforts to improve the use of ws in wn applications. these have been materialized in a number of practical applications for locating various fixed and mobile targets like: (1) mobile targets with the hardware modules described in section 3 and shown in figs. 3-5; (2) monitoring applications of the fixed oil wells with the camera and passive infrared sensor (pir); (3) monitoring of maritime containers. therefore, we can say that the purpose, for which we have designed all hardware and software solutions presented in this paper, has been reached. conflicts of interest the authors declare no conflict of interest references [1] j. wang, j. cho, s. lee, k. c. chen, and y. k. lee, “hop-based energy aware routing algorithm for wireless sensor networks,” ieice transactions on communication 2010, vol. e93.b, no. 2, pp. 305-316, 2010. [2] w. heinzelman, “application-specific protocol architectures for wireless networks,” ph.d. thesis, massachusetts institute of technology: cambridge, ma, usa, pp. 84-86, 2000. [3] j. wang, j. u. kim, s. lei, y. niu, and l. sungyoung, “a distance-based energy aware routing algorithm for wireless sensor networks,” sensors, vol. 10, pp. 9493-9511, 2010. [4] o. younis and s. fahmy, “heed: a hybrid, energy-efficient, distributed clustering approach for ad hoc sensor networks,” ieee transactions on mobile computing, vol. 3, no. 4, pp. 366-379, october-december 2004, doi:10.1109/tmc.2004.41. [5] c. h. yassin and m. c. rosu, “energy-saving strategies in monitoring for wireless sensor networks,” the 21th ieee international symposium for design and technology in electronic packaging (siitme 2015), brasov, romania october 2015. [6] s. murugananham and p. r. mukesh, “real time web-based vehicle tracking using gps,” world academy of science, engineering and technology, vol. 61, pp. 91-99, 2010. [7] n. chadil, a. russameesawang, and p. keeratiwintakorn, “real-time tracking management system using gps, gprs and google earth,” international conference on electrical engineering/electronics, computer, telecommunications and information technology, krabi, 2008, pp. 393-396. [8] s. m. kale and a. kumar, “location tracking and data compression for accurate energy efficient localization using gsm system,” international journal of innovative research in science, engineering and technology, vol. 3, no. 3, pp. 10163-10167, march 2014. [9] m. a. labrador and p. m. wightman, “topology control in wireless sensor networks; with a companion simulation tool for teaching and research,” springer science and business media b. v., 2009. [10] f. ingelrest, d. simplot-ryl, and i. stojmenovic, “optimal transmission radius for energy efficient broadcasting protocols in ad hoc networks,” ieee transactions on parallel and distributed systems, vol. 17, no. 6, pp. 536-547, june 2006. [11] r. zhang and m. a. labrador, “energy-aware topology control in heterogeneous wireless multihop networks,” proceedings 2nd international symposium on wireless pervasive computing, puerto rico, 2007. [12] g. anastasi, m. conti, m. di francesco, and a. passarella, “energy conservation in wireless sensor networks: a survey,” ad hoc networks, vol. 7, no. 3, pp. 537-568, 2009. [13] r. willett, a. martin, and r. nowak, “backcasting: adaptive sampling for sensor networks,” third international symposium on information processing in sensor networks, ipsn, berkeley, ca, usa, 2004, pp. 124-133. advances in technology innovation, vol. 6, no. 1, 2021, pp. 47-66 63 [14] a. d. marbini and l. e. sacks, “adaptive sampling mechanisms in sensor networks,” proceedings lcs, 2003. [15] a. jain and e. y. chang, “adaptive sampling for sensor networks,” dmsn '04 proceedings of the 1st international workshop on data management for sensor networks: in conjunction with vldb 2004, pp. 10-16, 2004. [16] s. goel and t. imielinski, “prediction-based monitoring in sensor networks: taking lessons from mpeg,” acm sigcomm computer communication review, vol. 31, no. 5, pp. 82-98, 2001. [17] b. gedik, l. liu, and p. s. yu, “asap: an adaptive sampling approach to data collection in sensor networks,” ieee transactions on parallel and distributed systems, vol. 18, no. 12, pp. 1766-1783. [18] s. goel, a. passarella, and t. imielinski, “using buddies to live longer in a boring world,” fourth annual ieee international conference on pervasive computing and communications workshops, (percomw'06), pisa, 2006. [19] n. kimura and s. latifi, “a survey on data compression in wireless sensor networks,” international conference on information technology, coding and computing (itcc'05), 2005, vol. 2, pp. 8-13. [20] w. heinzelman, a. chandrakasan, and h. balakrishnan, “energy-efficient communication protocol for wireless microsensor networks,” proceedings of the 33rd annual hawaii international conference system sciences, maui, hi, usa, january 2000, vol. 2, pp. 1-10. [21] w. b. heinzelman, a. p. chandrakasan, and h. balakrishnan, “an application-specific protocol architecture for wireless microsensor networks,” ieee transactions on wireless communications, vol. 1, no. 4, pp. 660-670, october 2002. [22] k. akkaya and m. younis, “a survey on routing protocols for wireless sensor networks,” ad hoc networks, vol. 3, no. 3, pp. 325-349, may 2005. [23] w. zhang and g. cao, “dctc: dynamic convoy tree-based collaboration for target tracking in sensor networks,” ieee transactions on wireless communications, vol. 3, no. 5, pp. 1689-1701, september 2004. [24] w. zhang and g. cao, “optimizing tree reconfiguration for mobile target tracking in sensor networks,” ieee infocom, hong kong, 2004, vol. 4, pp. 2434-2445. [25] a. goel and e. deborah, “simultaneous optimization for concave costs: single sink aggregation or single source buy-at-bulk,” algorithmica, vol. 43, pp. 5-15, 2003. [26] r. cristescu, b. beferull-lozano, and m. vetterli, “on network correlated data gathering,” ieee infocom 200, hong kong, 2004, vol. 4, pp. 2571-2582. [27] y. zhu, k. sundaresan, and r. sivakumar, “practical limits on achievable energy improvements and useable delay tolerance in correlation aware data gathering in wireless sensor networks,” second conference ieee secon, santa clara, ca, usa, 2005, pp. 328-339,. [28] k. w. fan, s. liu, and p. sinha, “on the potential of structure-free data aggregation in sensor networks,” proceedings 25th conference infocom, barcelona, april 2006, pp. 1-12. [29] k. w. fan, s. liu, and p. sinha, “scalable data aggregation for dynamic events in sensor networks,” proceedings of the 4th international conference on embedded networked sensor systems (sensys'06), boulder, co, usa, 2006, pp. 181-194. [30] s. mahfoudh and p. minet, “an energy efficient routing based on olsr in wireless ad hoc and sensor networks,” 22nd international conference on advanced information networking and applications -workshops, okinawa, 2008, pp. 1253-1259. [31] http://tools.ietf.org/html/rfc6206 [32] g. pei, m. gerla, and t. w. chen, “fisheye state routing: a routing scheme for ad hoc wireless networks,” ieee international conference on communications icc global convergence through communications, new orleans, la, usa, 2000, vol. 1, pp. 70-74. [33] m. van der schaar, “cross-layer wireless multimedia transmission: challenges, principles, and new paradigms,” ieee wireless communications, vol. 12, no. 4, pp. 50-58, 2005. [34] m. garey and d. johnson, computers and intractability: a guide to the theory of np completeness. w. h. freeman & co. new york, ny, usa, 1990. [35] j. wu and h. li, “on calculating connected dominating set for efficient routing in ad hoc wireless networks,” proceedings the 3rd international workshop on discrete algorithms and methods for mobile computing and communications dialm, 1999, pp. 7-14. [36] f. dai and j. wu, “an extended localized algorithm for connected dominating set formation in ad hoc wireless networks,” ieee transactions on parallel and distributed systems, vol. 15, no. 10, pp. 908-920, 2004. [37] f. ingelrest, d. simplot-ryl, and i. stojmenovic, “smaller connected dominating sets in ad hoc and sensor networks based on coverage by two-hop neighbors,” 2nd international conference on communication system software and middleware, bangalore, india, 2007, pp. 1-8. [38] k. alzoubi, p. j. wan, and o. fieder, “distributed heuristics for connected dominating sets in wireless ad hoc networks,” journal of communications and networks, vol. 4, no. 1, pp. 22-29, 2002. [39] b. han, h. h. fu, l. li, and w. jia, “efficient construction of connected dominating set in wireless ad hoc networks,” ieee international conference mass 2004, (ieee cat. no.04ex975), pp. 570-572, fort lauderdale, fl, usa, 2004. [40] e. o. blass, j. horneber, and m. zitterbart, “analyzing data prediction in wireless sensor networks,” ieee vehicular technology conference, 2008, pp. 86-87. advances in technology innovation, vol. 6, no. 1, 2021, pp. 47-66 64 [41] k. zeng, w. lou, j. yang, and d. r. i. brown, “on geographic collaborative forwarding in wireless ad hoc and sensor networks,” international conference on wireless algorithms, systems and applications (wasa2007), 2007, pp. 11-18. [42] l. jun and j. p. hubaux, “joint mobility and routing for lifetime elongation in wireless sensor networks,” proceedings of 24th annual joint conference of the ieee computer and communications societies, vol. 3, pp. 1735-1746, miami, fl, 2005. [43] m. bhardwaj and a. p. chandrakasan, “bounding the lifetime of sensor networks via optimal role assignments,” proceedings of 21st annual joint conference of ieee computer and communications societies, new york, usa, 2002, vol. 3, pp. 1587-1596. [44] m. cagalj, j. p. hubaux, and c. enz, “minimum-energy broadcast in all-wireless networks: np-completeness and distribution issues,” proceedings of 8th annual international conference on mobile computing and networking, atlanta, usa, 2002, pp. 172-182. [45] i. papadimitriou and l. georgiadis, “energy-aware routing to maximize lifetime in wireless sensor networks with mobile sink,” journal of communications software and systems, vol. 2, no. 2, pp. 141-151, june 2006. [46] l. pelusi, a. passarella, and m. conti, “opportunistic networking: data forwarding in disconnected mobile ad hoc networks,” ieee communications magazine, vol. 44, no. 11, pp. 134-141, november 2006. [47] r. c. shah, s. roy, s. jain, and w. brunette, “data mules: modeling a three-tier architecture for sparse sensor networks,” proceedings ieee international workshop on sensor network protocols and applications, usa, 2003, pp. 30-41. [48] c. h. ou and k. f. ssu, “routing with mobile relays in opportunistic sensor networks,” 18th ieee annual international symposium on personal, indoor and mobile radio communications, athens, 2007, pp. 1-5. [49] s. meguerdichian and m. potkonjak, “low power 0/1 coverage and scheduling techniques in sensor networks,” ucla technical reports 030001, january 2003. [50] k. chakrabarty, et al., “grid coverage for surveillance and target location in distributed sensor networks,” ieee transactions on computers, vol. 51, no. 12, pp. 1448-1453, december 2002. [51] b. chen, k. jamieson, h. balakrishnan, and r. morris, “span: an energy efficient co-ordination algorithm for topology maintenance in ad hoc wireless networks,” acm wireless networks, vol. 8, no. 5, pp. 481-494, september 2002. [52] x. wang, g. xing, y. zhang, c. lu, r. pless, and c. gill, “integrated coverage and connectivity configuration in wireless sensor networks,” proceedings of sensys, pp. 28-39, la, usa, 2003. [53] m. cardei and d. du, “improving wirelessn sensor network lifetime through power aware organization,” acm journal of wirless networks, vol. 11, no. 3, may 2005. [54] m. cardei, m. thai, y. li, and w. wu, “energy-efficient target coverage in wireless sensor networks,” proceedings of 24th annual joint conference of the ieee infocom, miami, fl, usa, 2005, pp. 1976-1984. [55] v. rajendran, k. obraczka, and j. j. garicia-luna-aceves, “energy-efficient, collision-free medium access control for wireless sensor networks,” proceedings acm sensys, la, ca, november 2003, pp. 181-192. [56] v. rajendran, “energy-efficient, application aware medium access for sensor networks,” proceedings 2nd ieee conference on mobile ad-hoc and sensor systems (mass), 2005, pp. 8-630. [57] w. l. lee, a. datta, and r. cardell-oliver, “flexitp: a flexible-schedule-based tdma protocol for fault-tolerant and energy-efficient wireless sensor networks,” ieee transactions on parallel and distributed systems, vol. 19, no. 6, pp. 851-864, june 2008. [58] s. gobriel, d. mousse, and r. cleric, “tdma-asap: sensor network tdma scheduling with adaptive slot stealing and parallelism,” 29th ieee international conference on distributed computing systems, montreal, qc, 2009, pp. 458-465. [59] w. ye, j. heidemann, and d. estrin, “medium access control with coordinated adaptive sleeping for wireless sensor networks,” ieee/acm transactions on networking, vol. 12, no. 3, pp. 493-506, june 2004. [60] t. van dam and k. langendoen, “an adaptive energy efficient mac protocol for wireless sensor networks,” proceedings of 1st international conference on embedded networked sensor systems, november 2003, pp. 171-180. [61] g. lu, b. krishnamachari, and c. s. raghavendra, “an adaptive energy efficient and low-latency mac for data gathering in wireless sensor networks,” proceedings 18th international parallel and distributed processing symposium, new mexico, usa, april 2004, vol. 224, pp. 26-30. [62] j. polastre, j. hill, and d. culler, “versatile low power media access for sensor networks,” proceedings 2nd acm conference on embedded networked sensor systems (sensys), maryland, november 2004. [63] i. rhee, a. warrier, m. aia, and j. min, “z-mac: a hybrid mac for wireless sensor networks,” ieee/acm transactions on networking, vol. 16, no. 3, pp. 511-524, june 2008. [64] s. mahfoudh, g. chalhoub, p. minet, m. misson, and i. amdouni, “node coloring and color conflict detection in wireless sensor networks,” future internet, vol. 2, no. 4, pp. 469-504, 2010. [65] a. salman, “real time vehicle tracking system and energy reduction,” university of waterloo library, ontario, canada, 2017. advances in technology innovation, vol. 6, no. 1, 2021, pp. 47-66 65 [66] i. f. akyildiz, w. su, y. sankarasubramaniam, and e. cayirci, “wireless sensor networks: a survey,” computer network, vol. 38, pp. 393-422, march 2002. [67] c. m. sadler and m. martonosi, “data compression algorithms for energy-constrained devices in delay tolerant networks,” proceedings of the 4th international conference on embedded networked sensor systems, sensys, boulder, usa, 2006, pp. 265-278. [68] i. buchmann, “batteries in a portable world: a handbook on rechargeable batteries for non-engineers,” 3rd ed. richmond: cadex electronics, 2011. [69] https://www.microchip.com/wwwproducts/en/pic32mx775f256h. [70] m. c. rosu, a. iliesiu, y. c. hamed, and s. g. obreja, “a gi proposal to display ecg digital signals wirelessly real-time transmitted onto a remote pc,” advances in technology innovation, vol. 3, no. 3, pp. 118-125, march 2018. [71] https://msoon.github.io/powermonitor/powertool/doc/power%20monitor%20manual.pdf [72] https://en.freedownloadmanager.org/windows-pc/monsoon-powertool-free.html appendix 1 – abbreviations from table 6 apteen adaptive periodic teen asap accelerated sap bmac berkeley mac cadr constrained anisotropic diffusion routing cougar data-centric cadr protocol csma carrier sense multiple access dctc direct connect text client protocol did directed diffusion technique dlmst directed lmst dmac directional mac drng directed rng flama flow-aware medium access flexitp flexible based tdma protocol gaf geographic adaptive fidelity gear geographical and energy aware routing leach low-energy adaptive clustering hierarchy protocol lmst local minimum spanning tree mac medium access control mst multiple shared tree pegasis power-efficient gathering in sensor information systems read residual energy aware dynamic rng random number generator serena scheduling router nodes activity smac sensor media access control smt steiner minimum tree span switch port analyzer speed routing protocol with qos spin sensor protocols for information via negotiation sync data synchronization tdma time division multiple access tdma/ca time division multiple access/ control access teen threshold-sensitive energy efficient sensor network protocol tod tree on directed acyclic graph trama traffic-adaptive medium access protocol zmac zebra mac appendix 2 – datasheets links 1. huawei gtm900 gsm/gprs wireless module; available: https://www.digchip.com/datasheets/download_datasheet.php?id=4375635&part-number=gsm900 2. ublox neo-6m gps modul; available: https://www.waveshare.com/w/upload/2/2c/neo-6-datasheet.pdf. 3. nxp semiconductors lpc2361/62 µ c; available: https://www.nxp.com/docs/en/data-sheet/lpc2361_62.pdf. advances in technology innovation, vol. 6, no. 1, 2021, pp. 47-66 66 4. ublox leon-g1 series gsm/gprs module; available: https://www.u-blox.com/sites/default/files/products/documents/ leon-g1_datasheet_%28ubx-13004887%29.pdf. 5. u-blox max-6q-0-000 gps modul; available: https://www.iot-lab.info/wp-content/uploads/2013/10/gps_max6.pdf. 6. sim900 gsm/gprs modul; available: https://datasheet.octopart.com/sim900-simcom-datasheet-17594122.pdf. 7. stm32f103rb microcontroller (µ c); available: https://www.st.com/content/st_com/en/products/microcontrollers-micro processors/stm32-32-bit-arm-cortex-mcus/stm32-mainstream-mcus/stm32f1series/stm32f103/stm32f103rb.html. 8. microchip-pic32 µ c; available: https://ro.mouser.com/datasheet/2/268/60001156j-1315355.pdf. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). microsoft word 2-aiti#6494 74-89.docx advances in technology innovation, vol. 6, no. 2, 2021, pp. 74-89 development of a small intelligent weather station for agricultural applications yi-hua chung 1 , jun-fu huang 2 , yuan-chen hu 2 , chen-kang huang 2,* 1 department of chemistry, national taiwan university, taipei, taiwan 2 department of biomechatronics engineering, national taiwan university, taipei, taiwan received 29 september 2020; received in revised form 08 december 2020; accepted 01 february 2021 doi: https://doi.org/10.46604/aiti.2021.6494 abstract it is known that climate change causes a decrease in the profit gained from agricultural production. this work designs and establishes weather boxes equipped with functions of rainfall prediction, frosting forecast, and lightning detection. with the wireless connection and the build-in decision mode, weather boxes can deliver early-warning by sending texting messages to the users and actuating the corresponding action to response the extreme climate. to implement rainfall and frosting prognostication, two different datasets are analyzed by the technology of data mining. one of the datasets is acquired from the central weather bureau, and the other is from the proposed weather box monitoring the agricultural environment. from the experimental results, the prediction model constructed from the data which is collected by the proposed weather box exhibits a higher accuracy in rainfall forecasting than those based on the central weather bureau. keywords: weather box, rainfall, frosting, lightning, early warning 1. introduction climate change and agriculture gain have a reciprocal relationship. climate change may cause a decrease in the profit gained from agricultural production. therefore, accurate weather forecasting becomes an important but challenging task to researchers for a long time. in addition, numerous people are aware of climate-changing. for example, orchardists pay attention to the amount of rainfall, which affects the fruit quality. the frosting, which will impact the flavor and aromatics of the tea, is concerned by tea growers. lightning is responsible for plantation fires and the death of the crops [1-3]. recently, the continuous advances in sensors, sensing systems, and the internet of things (iot) [4-5] technology significantly impact several fields. for instance, in the agriculture fields, to accomplish the task of the remote environment monitoring for the farmland, developers can utilize single-board microcontrollers or computers to connect with various sensors to collect and accumulate a large amount of local environmental data at first. the data can be analyzed to get a more undiscovered relationship between the data via the technology of data mining [6-12]. after all, those relations can be used to achieve the goal of future predicting. currently, some companies and organizations, such as airports and agriculture agencies, collect and accumulate the weather data by their own weather stations to fulfill some of their particular needs. the data, which is for weather monitoring or predictions, become more precise when a greater amount of data are collected and analyzed. * corresponding author. e-mail address: ckhuang94530@ntu.edu.tw tel.: +886-2-3366-5351 advances in technology innovation, vol. 6, no. 2, 2021, pp. 74-89 75 official weather stations collecting professional data are precious. however, these weather stations are mostly found in the research institutions located in urban area far from the remote agricultural locations, which may cause the grabbed data to have some potentially hidden problems. fortunately, the concepts, such as free software and free hardware can significantly reduce the cost of a station. to construct a low-cost weather station, the developers can modify the existing version to reduce or increase the functions and costs based on the needs. sensors are elements that translate a usually non-electrical value to an electrical value, which can be measured, amplified, or even modified. to manage and control these sensors and share the collected data, some new low-cost single board computers may be used, for example, arduino and raspberry pi. it is commonly known that the central weather bureau (cwb) has already set up numerous weather stations nowadays. nevertheless, most of the weather stations are built in the megalopolis instead of some distant regions. furthermore, most of the commercially available weather stations are too expensive to be afforded. low cost weather stations are required to set up in remote agricultural area. the functions of the weather boxes constructed in this work are shown in fig. 1. this paper is purposed to construct a weather box with the ability to perform rainfall, frosting prediction, and lightning detection. for the first aim, rainfall prognostication, a weather box is constructed with the wemosd1 choosing to be the microcontroller. three sensors, bme280, fc-37, and gy-49 are used to sense the humidity, temperature, pressure, rain, and solar radiation as the local weather parameters to accumulate the amounts of the local weather data in three experimental fields, including taipei city, taoyuan city, and yilan city. after stockpiling the local weather data, data sets from (1) weather box or (2) the historical dataset released by the cwb, which consists of several atmospheric attributes, are analyzed by the techniques of data mining to extract the hidden relationships among weather parameters. above all, as the weather box releases a heavy rain alerting by the prediction of the weather forecasting model, it can also send out the alarming message to the users by the communication application. fig. 1 different weather boxes with various functions constructed in this work second, the frosting module is added into the system with both of the frosting point exploring and recording functions. when in the exploring mode, the testing surface is lowered by the thermoelectric cooler controlled by raspberry pi. during the surface temperature drop, photos are taken and analyzed thereafter. when the frost is first observed, the surface temperature is considered as the frosting point for the corresponding environment temperature, pressure, and relative humidity. when in the recording mode, a webcam is set to take photos at the testing surface regularly. the series of photos are analyzed to find out the frosting occurrence and its corresponding surface temperature. lastly, regarding the lightning detection, the weather box attained the greatest performance with precision = 66.7% and recall = 22.2% on may 27, 2020, in taipei, and precision = 71.1% and recall = 45.8% on july 16, 2020, in yilan. obviously, the presentation of the detecting results in yilan is better than those in taipei as the section of result and discussion shown. the main causation for the poor performance of the recall is because of the small bandwidth for the detecting frequency of the lightning events. unfortunately, the data released by the cwb is constructed by the cg flash and ic flash events. however, the as3935 sensor can only detect some of the cg flash events, and all of the ic flash events can not be sensed by the as3935 due to the narrow bandwidth of it. the principal reason for the subpar precision is on account of the detection method we chose, radio detection. though we put effort into finding the experimental fields without noises, the system is still easy to be affected by the disturbers in the environment. advances in technology innovation, vol. 6, no. 2, 2021, pp. 74-89 76 2. literature review low-cost weather stations were developed in previous literature [13]. to establish the weather station, r. c. brito’s team firstly proposed an architecture structured in layers, which considered the scalability of the system and allowed the connection of new sensors to support the needs of monitoring new data. each feature in this structure was equipped with specific hardware, and the sensors would send the measured data back to the arduino through different means, such as cable, analogically, or digitally. the arduino used i 2 c communication protocol to exchange information with the raspberry pi. the raspberry pi was connected to the internet through rj-45 cable, which sent the data to a web server, the database. it also shared the data in rest format, which could be read by web pages, desktop applications, and mobile devices. the connection to the internet was done through a wifi or cellular (gsm, 3g, 4g) network. proprietary resources weren't used in the creation of this weather station. all the elements, such as hardware, software, and communication protocols were free to use and low-cost to construct. in the literature and this work, the classification in matlab toolbox, including decision tree [14], support vector machine [15-17], k nearest neighbor [18], and bagged tree [19-20], was used to find the best classification to construct the rainfall prediction model of rainfall. the last case is lightning detection. discharge activity is a weather phenomenon, including two types. lightning of cloud to cloud is called ic flash, and lightning cloud to ground is called cg flash. besides, the cg lightning has a more harmful destructive power to the targets on the ground, including human beings, forest creatures, and so on. commonly, there are three ways to detect the lightning event, inclusive of radio detection, radar detection, and optical lightning detection [21]. in [22], an intelligent lightning detection system was constructed. the system was built on a lightning signal collection module, central control module, display module, memory module, time module, and host computer module, which is established to cope with the problems of the high cost and the lacking the weather stations for detecting lightning. the 32-bit arm stm32f103 was used to act as the core, microcontroller, of the system. as the lightning occurrs, the lightning signal acquisition module will output a signal pulse which implies the fast sampling intensity of the lightning in real-time, and the sampled data is transited to the arm stm32f103. after receiving the data, the microcontroller will count the number of lightning and the strength through tft color screen display, automatically record the time of lightning occurs, and store these data in the sd card. meanwhile, via rs485 bus, the arm stm32f103 will transmit processed data to the pc. monitoring software in the pc which allows the users to get monitoring instantaneously about the number of lightning occurring and the lightning intensity. moreover, the data can be stored on a computer hard drive, with retrospective queries functions. after owning the ability to detect whenever lightning occurs, another task that needs to be solved is lightning positioning development. as the previous paragraph mentioned, the method of radar to detect lightning is mainly used for thunderstorm activity detection, and it is impossible to detect single lightning and its intensity parameters. therefore, in order to judge the real placement that the lightning happens, the tome-of-arrival (toa) method and the magnetic direction method (mdf) are used in combination. referring to the main purpose of distant regions’ remote environmental monitoring, the way of radio detection is chosen to be used to sense the lightning events, and the events of cg flash are mainly focused on. to complete lightning detection, the as3935 [23] sensor is used to support our decision model to make the arbitration and the raspberry pi acts as the core. two experimental fields, including national taiwan university in taipei city and the national yilan university in yilan city, are set. when the microcontroller is connected to the internet, the collected local lightning events data will be uploaded to the intelligent biosensing platform, an online website. otherwise, the data is accumulated in the micro sd card slotted on the microcontroller. after piling up the data, the data collected by the weather box is compared with the lightning events data released by the cwb, which is open access, as the benchmark to ensure the precision and the accuracy of the data collected by the weather box. advances in technology innovation, vol. 6, no. 2, 2021, pp. 74-89 77 the results of rainfall prognostication show that the best prediction model that covered the data collected by the weather box is knn (k=3). as for the data released by the cwb, bagged tree is the best. besides, the matthews correlation coefficient (mcc) [24] is used as the index to measure the qualities of the classifications used in the experiments. as the experimental result showed in the results section, the best mcc number of the models to predict rainfall before raining for 2-hr of the data collected by the weather box is 0.925 (knn), which is better than 0.596 (bagged tree), the outstanding mcc of the data released by the cwb. compared to the cwb data, the reasons why the data collected by the weather box, lead to more precise and instantaneous prediction models are on account of the higher location congestion of the weather stations and the more frequent interval of the detecting time. for the frosting prediction, the alarming would take the corresponding actions when the temperature < 5°c, less cloud cover, and the relative humidity greater than 60% were occurred simultaneously. 3. material and methods 3.1. rainfall forecast (1) weather box construction fig. 2 the rainfall weather box designing flow chart fig. 3 the rainfall weather box experimental set-up diagram to reach the goal of rainfall forecast, a weather box is constructed to collect the data, which has a coverage of a small area, in order to provide an instantaneous and short-range weather prediction that reflects the real local weather accurately. considering the problem of accessing internet [20] in the remote region and the cost of the weather box, wemosd1 is chosen advances in technology innovation, vol. 6, no. 2, 2021, pp. 74-89 78 to be the core of the weather box for its’ low-cost and self-contained wi-fi networking capabilities. the weather box of rainfall prediction was equipped with three different detecting sensors, including bme280, fc-37 and gy-49. they are used to sense the humidity, temperature, pressure, rain and radiation and used four li ion 186500 batteries in parallel as its’ power supply. the weather box designing flow chart is shown in fig. 2 and fig. 3. besides, the voltage given by the 3.7v batteries does not meet with the specification of the wemmosd1. we cope with the dilemma by assembling a step-up transformer, which is constructed with a boost convertor, lm-2577, and a protection circuit module, xd-58a, which can increase the voltage to the scope of application. furthermore, ssd1306-oled is selected as the display module to be the monitor for our climate box. there are five layers for the module to set out distinct information. the first layer shows the time and date instantly. the second and third layers illustrate two kinds of datasets that are both grabbed from the openweathermap api, a website provided by the cwb open information. primarily, the second layer, which presents the instantaneous rainfall prediction, indicates some messages, including the character to judge the weather today, t symboling the temperature, h representing the humidity, p designating as the pressure, and w stood for the wind speed. then, the third layer denotes weather forecasting with three days in succession, which helps us to predict the climate during the upcoming three days. the forth layer, which demonstrates the dataset supervised by our sensors straight away. additionally, the data of the instantaneous rainfall prediction provided by the cwb open information is used to compare with the dataset supervised by our sensors. after gripping the data, there are two different ways for data storage. ultimately, the fifth layer acting as the recording layer aims to determine which way is used to store the data and checked whether the sd card works correctly. cooperating with the internet accessibility, if the internet is connected, the data sensed will be sent to the cloud, thingspeak, or the data will only be stored in the sd card module. notably, when the reading of the fc-37, the rain sensor exceeds the setting threshold. a text message will automatically be sent to the users for the heavy rain advisory by using ifttt, a freeware web-based service that creates chains of simple conditional statements [25], or a communication application. (2) data gathering from weather box as completing to manufacture the weather box, the experimental fields are set in the national taiwan university in taipei, taoyuan city, and pear orchard located at yilan city. since the bme280 sensor needs to be situated in the outside of the weather box to get a precise record. therefore, all of the weather boxes for experimental fields are located at the places that sheltered from the storm to satisfy the needing and prolong the service life of the sensor at the same time, as shown in fig. 4. nevertheless the rain sensor, the fc-37, demands to be exposed under the rain to catch the information of the instant raindrop so that the fc-37 sensor are extended out of the shelter. fig. 4 the rainfall prediction weather box the first experimental field is national taiwan university. it is worth mentioning that the rain detector is leaned in a slanted position to let the water roll down through the edge to avoid having puddles on the sensor, as shown in fig. 5. in addition, the second experimental field is set up in taoyuan city to be a comparison to the first field to find out whether the advances in technology innovation, vol. 6, no. 2, 2021, pp. 74-89 79 experimental location will impact our rainfall weather box or not shown in fig. 6. the last field is located in the pear orchard in yilan city. apart from the other two fields, this field grabs another two specific weather parameters, inclusive of soil humidity and soil electrical conductivity. the method of supervised learning is used and corresponding class labels are given as either sunny or rainy to construct a new rainfall prediction model, which is distinguishing from the previous weather boxes. fig. 5 the rain detector was leaned in a slanted position (taipei) fig. 6 the rain detector was leaned in a slanted position (taoyuan) (3) data pre-processing after accessing two datasets (the daily meteorological report provided by the cwband the data accumulated by our rainfall weather box) the data must have been rectified by data cleaning for the following three different cases. firstly, for the data collected by our weather box every ten minutes, the anomalous data caused by the problems of instrument and data transfer should be disposed of. the reason why those data can be abandoned is that it will not have a significant change of weather status in ten minutes (the time interval that we collected one piece of data). for the data provided by the cwb, there are two parameters that needs to be processed. that is, if the symbol of “v” is presented in wind direction, the current value of the wind direction will be replaced by taking the average of the previous data and the next one. likewise, if the symbol of precipitation is “t”, which implies the data is less than 0.1mm, the precipitation value will be altered by replacing it with 0.1mm. in order to design a rainfall-forecast model which could give out the alarm two hours earlier, the method of supervised learning is used. a rainfall label is constructed for the labeling schema, which is defined as follows: 1, if rain or two hours before rain 0, otherwise label  =   (1) the methods used for feature scaling are min-max normalization and standardization (z-score normalization) in this paper. although there are 17 weather parameters, some of them are lost. thus, only eight of them are chosen to be the training features, including temperature, dew point, relative humidity, atmospheric pressure for weather station and sea level, wind advances in technology innovation, vol. 6, no. 2, 2021, pp. 74-89 80 speed, wind direction, and solar radiation. after that, the 10-fold cross validation measurement is used to test and improve those constructed models. all of the machine learning model is developed in matlab. 3.2. frosting prediction the raspberry pi controls two sensors, including the bme280 and ds18b20. the bme280 detects temperature, humidity, and pressure in the air; the ds18b20 measures the temperature of the surface. in the hoarfrost point finding mode with the temperature controller and the thermoelectric cooler module the hoarfrost point is measured in any environment by gradually decreasing the surface temperature. in normal mode, the thermoelectric cooler module only have the function of heat dissipation by running fans. by the heat pipe in the thermoelectric cooler module, the surface temperature is close to the atmospheric temperature to simulate the leaf in the outdoor environment for a long time. the picture of the frosting prediction weather box is shown in the fig. 7. the experimental field is set in the tea plantation in fushoushan farm in taichung. fig. 7 the frosting prediction weather box 3.3. lightning detection for lightning detection, a lightning detecting weather box is constructed with raspberry pi acting as the microcontroller. the raspberry pi is connected to two sensors, as3935 and bmp180. the weather box is illustrated in fig. 8. following ensuring the connection of the hardware, the first case is tuning the loop antenna. the loop antenna is designed to have a resonance frequency at 500khz, which is the same as the lightning frequency of most lightning events of cg flashes, by adding or removing the tuning capacitors from 0pf to 120 pf in the step of 8pf to optimize the performance of the signal validation and distance estimation. the embedded algorithm will check the incoming signal pattern to reject the potential man-made disturbers [26]. fig. 8 the lightning detection system in a thermometer shelter in yilan advances in technology innovation, vol. 6, no. 2, 2021, pp. 74-89 81 in this paper, the sensor interacts with the raspberry pi via spi protocol, and python2 is chosen as the developing language with package “spidev” (a python module for interfacing with spi devices from user space via the spidev linux kernel driver). there are four focused programmable settings in this paper. the first one is analog-front-end (afe), which is a setting that can amplify and demodulate the ac-signal picked up by the antenna. the second one is mask disturber (mask_dist), which can shield the man-made disturber from the as3935. the third one is spike rejection (srej), which is used to increase the robustness against false alarms from such disturbers with its range between 0 and 15. the last one is noise floor level threshold (nf_lev). the output signal of afe is also used to measure the noise floor level, which is continuously compared to the noise threshold. whenever the noise floor level exceeds the noise threshold, as3935 issues an interrupt, int_nh, to inform the raspberry pi that it cannot operate properly due to the high input noise received by the antenna in this work. after the pre-work of the system is finished, whenever an event happened, the int pin will go high, which implies that an event had happened. if the event is judged as a lightning event, the estimated distance of the lightning will be recorded in the register and stored in the micro sd card, as the algorithm flow chart shown in fig. 9. the collected data will be sent to the intelligent biosensing platform within 6 hours, and the lightning detection system status can be monitored remotely with the weather boxes working automatically. fig. 9 the algorithm works flow chart of lightning detection system 4. results and discussion 4.1. rainfall forecast (1) taipei, data collected by weather box the instantaneous rainfall prediction model uses four parameters as inputs, including temperature, humidity, pressure, and solar radiation at that time to determine whether it is raining or not. as table 1 shows, all of the models used in this advances in technology innovation, vol. 6, no. 2, 2021, pp. 74-89 82 research reaches an accuracy of over 90%, and the model using knn is the best. for constructing a model to predict raining before 2 hours, the extra models are built in two cases, inclusive of “before raining for 1-hr prediction model” and “before raining for 2-hr prediction model”. as the results portrayed in table 2 and table 3, both of the outstanding models for the two cases are the models utilizing knn classifiers. table 1 the result of instantaneous rainfall prediction model (taipei) classifier accuracy precision recall f1 measure mcc no rain rain no rain rain no rain rain decision tree 94.3% 94.9% 93.9% 97.2% 89.0% 95.9% 91.3% 0.867 bagged tree 96.9% 97.0% 97.7% 99.0% 93.8% 97.9% 95.7% 0.928 knn(k=3) 97.4% 98.3% 96.4% 98.7% 96.0% 98.5% 96.1% 0.941 svm 90.8% 92.1% 88.3% 95.1% 83.1% 93.4% 85.6% 0.786 table 2 the result of before raining for 1-hr prediction model (taipei) classifier accuracy precision recall f1 measure mcc no rain rain no rain rain no rain rain decision tree 94.0% 93.8% 93.7% 96.9% 87.7% 95.4% 90.9% 0.863 bagged tree 96.7% 96.7% 97.5% 98.4% 93.5% 97.4% 95.4% 0.925 knn(k=3) 97.1% 98.1% 96.1% 98.5% 95.7% 98.3% 95.9% 0.934 svm 90.4% 91.0% 88.0% 94.5% 82.3% 92.7% 84.9% 0.782 table 3 the result of before raining for 2-hr prediction model (taipei) classifier accuracy precision recall f1 measure mcc no rain rain no rain rain no rain rain decision tree 91.7% 91.5% 92.2% 96.3% 84.1% 93.8% 87.8% 0.821 bagged tree 96.4% 96.1% 97.3% 98.2% 93.4% 97.1% 94.9% 0.922 knn(k=3) 96.6% 97.8% 96.0% 98.3% 94.5% 98.0% 95.2% 0.925 svm 88.6% 89.9% 88.1% 94.1% 80.5% 91.9% 83.8% 0.750 (2) taoyuan, the data collected by weather box similar to the method used to predict rainfall in taipei, there are also three models constructed in this part, inclusive of the “instantaneous rainfall prediction model”, the “before raining for 1-hr prediction model”, and the “before raining for 2-hr prediction model”. as the results showed in table 4, table 5, and table 6, the models using knn as a classifier have an exceptional outcome. it is worth mentioning that the models using bagged tree as their classifier in taoyuan are better than those in taipei. table 4 the result of instantaneous rainfall prediction model (taoyuan) classifier accuracy precision recall f1 measure mcc no rain rain no rain rain no rain rain decision tree 94.3% 96.2% 86.1% 97.0% 82.5% 96.4% 83.9% 0.804 bagged tree 96.8% 97.2% 94.3% 99.0% 88.1% 97.9% 90.9% 0.891 knn(k=3) 97.1% 97.4% 94.4% 99.1% 88.9% 98.1% 91.6% 0.899 svm 89.9% 92.9% 73.5% 94.2% 70.0% 93.4% 71.7% 0.651 table 5 the result of before raining for 1-hr prediction model (taoyuan) classifier accuracy precision recall f1 measure mcc no rain rain no rain rain no rain rain decision tree 93.2% 95.1% 86.0% 97.0% 80.1% 95.9% 82.8% 0.783 bagged tree 96.2% 97.0% 92.1% 99.0% 87.3% 97.9% 89.6% 0.877 knn(k=3) 96.4% 97.2% 93.1% 98.1% 88.2% 97.4% 90.5% 0.884 svm 88.8% 91.5% 72.4% 94.1% 68.1% 92.9% 71.0% 0.639 advances in technology innovation, vol. 6, no. 2, 2021, pp. 74-89 83 table 6 the result of before raining for 2-hr prediction model (taoyuan) classifier accuracy precision recall f1 measure mcc no rain rain no rain rain no rain rain decision tree 93.2% 92.9% 85.0% 96.1% 76.1% 94.4% 80.2% 0.753 bagged tree 96.0% 96.8% 92.1% 98.2% 88.1% 97.4% 90.1% 0.881 knn(k=3) 96.0% 96.9% 92.0% 98.2% 88.0% 97.5% 90.0% 0.882 svm 87.7% 91.0% 74.6% 93.8% 65.6% 92.4% 70.0% 0.629 (3) yilan, the data collected by weather box apart from the models used in the previous fields, models in this part are constructed differently for comparison to the previous ones. the models constructed in this part use six parameters as inputs, including temperature, humidity, soil humidity, soil electrical conductivity, wind speed, and solar radiation. as the results illustrated in table 7, the performance in yilan is not as good as those in other experimental fields. besides, the model using svm as a classifier have a recall of zero to predict rainfall. table 7 the result of instantaneous rainfall prediction model (yilan) classifier accuracy precision recall f1 measure mcc no rain rain no rain rain no rain rain decision tree 87.9% 93.1% 71.1% 91.8% 73.2% 92.4% 72.1% 0.646 bagged tree 91.3% 90.2% 86.3% 96.7% 72.9% 93.3% 79.0% 0.736 knn(k=3) 87.3% 90.8% 72.0% 92.1% 69.3% 91.4% 70.6% 0.626 svm 77.9% 78.0% 22.0% 100% 0% 87.6% 0% 0 (4) taipei, the data provided by central weather bureau as shown in tables 8 to 10, it can be found that models trained by the data collected by the weather boxes in taipei have lower accuracies. the models using bagged tree as their classifier perform best in this part. moreover, the accuracy and recall of predicting rainfall decrease for longer predictions. table 8 the result of instantaneous rainfall prediction model (cwb_taipei) classifier accuracy precision recall f1 measure mcc no rain rain no rain rain no rain rain decision tree 86.7% 90.8% 69.2% 94.1% 55.3% 92.4% 61.4% 0.537 bagged tree 87.8% 90.2% 74.4% 95.7% 56.3% 92.8% 64.0% 0.574 knn(k=5) 84.8% 90.5% 61.1% 90.3% 60.4% 90.3% 60.0% 0.508 svm 85.7% 88.2% 69.3% 95.4% 46.6% 91.6% 55.7% 0.481 table 9 the result of before raining for 1-hr prediction model (cwb_ taipei) classifier accuracy precision recall f1 measure mcc no rain rain no rain rain no rain rain decision tree 84.6% 90.7% 69.8% 94.1% 54.1% 91.9% 60.9% 0.523 bagged tree 86.3% 90.0% 74.3% 95.4% 56.4% 92.4% 64.1% 0.575 knn(k=5) 83.3% 90.2% 61.4% 91.3% 58.2% 90.7% 60.0% 0.501 svm 83.8% 88.3% 69.5% 95.2% 46.3% 91.6% 55.5% 0.487 table 10 the result of before raining for 2-hr prediction model (cwb_ taipei) classifier accuracy precision recall f1 measure mcc no rain rain no rain rain no rain rain decision tree 83.3% 86.1% 72.3% 93.5% 54.1% 89.3% 61.8% 0.523 bagged tree 85.7% 88.1% 78.3% 94.0% 60.4% 90.9% 68.1% 0.596 knn(k=5) 82.4% 86.7% 63.4% 88.7% 60.2% 87.4% 61.7% 0.495 svm 82.3% 84.8% 72.1% 94.6% 49.1% 89.2% 58.4% 0.487 advances in technology innovation, vol. 6, no. 2, 2021, pp. 74-89 84 three types of models use four classifiers to find out the best classifier to fulfill the task of rainfall forecast. as the results shown in table 1 to table 3, for the three types of the models, it is obvious to recognize that the all of the models using the knn as a classifier have the best performance, and those who used svm have the worst. as for the models utilize the data provided by the cwb in taipei used eight training features, inclusive of temperature, dew point, relative humidity, atmospheric pressure (weather station and sea level), wind speed, wind direction, and solar radiation. the training results of these models indicate that the models using bagged tree as their classifier have the top achievement, and those using svm as their classifier look wretched as the results shown in table 8 to table 10. by the comparison of the models between the two different datasets, the prediction results suggest that the models which use the data collected by the weather box rather than the cwb lead to the increase of the predicting accuracies. additionally, it is clear to find that the accuracies decrease continuously from the “instantaneous rainfall prediction model” to the “before raining for 1-hr prediction model” until the “before raining for 2-hr prediction model”. it is thought to because of the wrong prediction of the rainfall. as the high density of the orange dots that denote raining between the humidity of 70% and80% in the humiditypressure graph in fig. 10, this situation implies that the prediction models might have had the wrong prediction in this zone caused by the high density. this might also be the reason why the rainfall prediction results at the yilan experimental field are worse than those at the taipei. because the weather in yilan is more prone to rain, and the detected humidity maintains at a high level which might cause the misjudgments. fig. 10 the humiditypressure graph provided by the cwb in order to have a higher accuracy of the rainfall prediction before raining for two hours earlier, the construction of the data is modified to build more accurate prediction models. more data are being labeled as “raining” in the new and “before raining for 2-hr prediction model” in the following extra experiments by adjusting the ratio of the raining data to no-raining data to find out the best constitution of the training datasets for the four classifiers as presented in table 11, though this action might have led to the ability to forecast no-rain decrease. for the following experimental results, these models of cwb and of our weather box are tested during other raining days. the experimental results are demonstrated in table 12 and table 13. it is suggested that because the shorter data collecting period(ten minutes) which is more frequent than an hour, the period time of cwb grabbing data, causes a higher accuracy as the reason in the previous part. additionally, the models built by the data collected by the weather box are overfitted, which leads to the poor performance as presented in table 12 and comparing to the results in table 11. for the results in the experimental field in taoyuan, the best recall of the models built to forecast the rainfall in taoyuan is 88%, which is lower than the best recall among all models in taipei, 94.5%. this is caused by the different locations where the weather boxes are placed. the weather boxes in the taoyuan experimental field are placed indoors. even though we try to extend the weather boxes to the outdoors, they are still placed directly indoors. nonetheless, the weather boxes in taipei advances in technology innovation, vol. 6, no. 2, 2021, pp. 74-89 85 experimental fields are placed in the outdoors, which receive the weather parameters directly. another possible reason is the terrain, since the topography of taoyuan city is tableland, dissimilar to basin, the terrain of taipei. for the models in yilan, though they use more weather parameters than those in taipei, they perform poorly. for example, as the table 7 showed, the model using svm as classifier has no ability to predict rainfall. the reason is because of the lack of the label of the state of weather. to fix up this problem, the corresponding method is taken at the time using the relative humidity to substitute the label of the state of weather. the fixing approach is applicable because it is obvious that as soon as the fc-37 sensor of the weather box detectes the raindrop, the reading of the humidity will increase instantly as response. the results reveal that this solution is not a proper way to take. therefore, there are some recommended improving ways to implement these models. for example, adding some extra indicators such as rain gauge and raindrop sensor for the experimental modules might help predicting the rainfall. otherwise, the collected data should be analyzed through the differential method to find the trend of the rain point. table 11 the result of before raining for 2-hr best-ratio prediction model (weather box_ taipei) classifier accuracy precision recall f1 measure mcc no rain rain no rain rain no rain rain decision tree 93.7% 94.2% 92.2% 95.8% 90.1% 94.9% 91.1% 0.862 bagged tree 93.5% 94.1% 92.3% 96.1% 89.8% 95.0% 91.0% 0.858 knn(k=3) 93.9% 92.4% 96.0% 98.5% 86.9% 95.3% 91.2% 0.868 svm 92.0% 91.1% 93.8% 96.5% 83.5% 93.7% 88.3% 0.825 table 12 result of before raining for 2-hr best-ratio prediction model (1) (cwb_ taipei) classifier ratio accuracy precision recall f1 measure mcc no rain rain no rain rain no rain rain decision tree 75:25 83.7% 88.3% 64.5% 91.1% 57.7% 89.6% 60.9% 0.506 bagged tree 50:50 79.7% 92.7% 52.7% 80.3% 77.8% 86.0% 62.8% 0.513 knn(k=5) 50:50 79.0% 91.7% 51.8% 80.4% 74.2% 85.7% 61.0% 0.487 svm 50:50 81.1% 92.0% 55.3% 83.0% 74.4% 87.3% 63.4% 0.520 table 13 the result of before raining for 2-hr best-ratio prediction model (2) (cwb_ taipei) classifier ratio accuracy precision recall f1 measure mcc no rain rain no rain rain no rain rain decision tree 75:25 72.6% 91.0% 47.1% 70.4% 79.2% 79.4% 59.1% 0.434 bagged tree 50:50 77.4% 92.2% 53.2% 76.4% 80.6% 83.6% 64.1% 0.508 knn(k=5) 50:50 77.1% 87.1% 53.5% 81.5% 63.9% 84.2% 58.2% 0.429 svm 50:50 69.8% 92.7% 44.5% 64.8% 84.7% 76.2% 58.3% 0.430 4.2. frosting prediction via some tentative, there are three situations found to be the conditions of frosting, involving “as the temperature is lower than 5°c”, “the cloud cover is less”, and “the relative humidity is higher than 60%”. under these situations, the weather box will send out a warning message to alarm the users to take some protection procedures, such as spraying water when the phenomenon of frosting is occurring or applying irrigation before the crops are frosted at nights soon. 4.3. lightning detection after constructing the lightning-detecting weather box, we start to collect the lightning data and optimize the related settings based on the feedback of the collected data during the plum rain season in taiwan. besides, we use the lightning data received by the central weather bureau as the standard to evaluate the recall and precision of the data we receive. moreover, five minutes is set as a time-interval. that is, if the central weather bureau and the weather box we constructed both receive lightning events in the same time-interval, we consider that the data detected from both resources are the same flash event. some parameters are defined as the followings, the units of them are the number of time-intervals. advances in technology innovation, vol. 6, no. 2, 2021, pp. 74-89 86 : true positive if both resources received data (2) : false positive if both weather box received data (3) : false negative if only cwb received data (4) + true positive recell true positive false negative = (5) + true positive precision true positive false positive = (6) the reason why we use time-interval to calculate the recall is because the as3935 sensor will only record the most recent data among all the detected lightning events during one time-interval. the estimated distance does not represent the distance to a single lightning event but the estimated distance to the leading edge of the storm. for instance, if there exist three lightning events which estimated distances are 1-km, 3-km, and 5-km at the same time-interval, the as3935 will only record the lightning event of the 1-km one during this time-interval. as the results shown in table 14, both of the recall and the precision outcome are not good enough, especially the performance of the recall. besides, compared to the results in the field of taipei in table 14, the results, no matter the recall or the precision, in yilan in table 15 present better. the reasons of the poor performance of the lightning detection system are discussed in the following. table 14 the part of lightning events result in the field of taipei data received during 2020/05/26 parameter true positive false positive false negative recall precision counts 2 3 6 25.0% 40.0% data received during 2020/05/27 parameter true positive false positive false negative recall precision counts 2 1 7 22.2% 66.7% data received during 2020/05/28 parameter true positive false positive false negative recall precision counts 1 1 9 10.0% 50% table 15 part of lightning events result in the field of yilan data received during 2020/07/01 parameter true positive false positive false negative recall precision counts 35 17 78 31.0% 67.3% data received during 2020/07/10 parameter true positive false positive false negative recall precision counts 17 8 42 28.8% 68.0% data received during 2020/07/16 parameter true positive false positive false negative recall precision counts 27 11 32 45.8% 71.1% at the beginning of the experiment, it is found out that the constructed lightning detection system is vulnerable to be interference with external disturbance. the data collected in march fills with a distance of 1 km. there are some possible reasons to discuss. the first one is because of the noise generating by the raspberry pi itself. another is that the ntu's internet base station formed an interference source that is 1-km away from the place where we collect the data. the fixing-up solution we adopt is extending the wires to make the microcontroller become farther away from the as3935. this demand is due to the detection way of as3935, radio detection, which is known to be sensitive to any other nearby interference [21]. advances in technology innovation, vol. 6, no. 2, 2021, pp. 74-89 87 so during april, we put the effort into finding the proper places with fewer disturbers to collect the lightning data, and the experimental field is finally set on the top floor of the department of bio-mechatronics engineering of ntu for less man-made interference. the power supply used at that time is solar power, and the weather box is placed under the solar panels to shield from the rainfall. however, after several times failing to receive data, we ultimately find out that the solar panel will block the reception of electromagnetic waves. the lightning detection system is finally placed in a thermometer shelter to avoid the rain from destroying the system. besides, the system is connected to the power wire of raspberry pi to the socket as the power supply. additionally, the pieces of lightning events collected by the weather box are far less than those provided by the cwb during one storm in the experimental field of taipei. it is supposed that the number of weather stations of the weather box in taipei is far less than the number of weather stations of cwb in taipei. furthermore, it is easy to reveal that the recall and the precision of the system perform worse than expected. the first reason causing the poor recall is due to the narrow bandwidth of the as3935 to collect the lightning events. it is known that the frequency bandwidth of cg flash is 3-3000khz, and the frequency bandwidth of ic flash is 30,000-300,000khz. however, according to the datasheet of as3935, the as3935 only receive lightning events with frequency in the range of (500 ± 33) khz. there are large amounts of the lightning events that are not able to be received by the as3935, which have caused the number of false negatives soaredand poor performance of recall. for the precision, though the precision is better than the recall in the taipei field, it still does not meet with the expected result. the main reason to be discussed is the disturbance. raspberry pi 4 and raspberry 3b+ are used as the microcontroller at the beginning. however, the experimental data points out that the noises generated by raspberry pi 4 will misjudge as the lightning events, which give rise to the number of the true positive and decreased the precision of the detection system. moreover, as mentioned in the first paragraph of this section, though the chosen experimental field in taipei is as possible as interference-free, the unexpected man-made disturbance can not be avoided completely, which increases the true positive and the decreased the precision. additionally, it is worth mentioning that even if the datasheet of as3935 claims that as3935 is subjected to lightning events with a radius of 40 kilometers, according to the actual use, the range of lightning events that can be detected as completely as possible is 20 kilometers. besides, compared to the results of taipei experimental field, the results in yilan perform noticeably. two main differences are responsible for it. firstly, the disturbance with a radius of 40 kilometers of the yilan field are lesser than the taipei field. secondly, the lightning detection system in yilan had an automatic scheduling function, and the system will retune the antenna every minute to ensure the accuracy of every moment. additionally, this system can also be connected remotely to observe and monitor the working status of the entire system at any time. into the bargain, from the lightning observation document of the cwb, an experiment is done to count the lightning events collected by different lightning-detecting systems during a storm in [26]. the plotting-results reveal the number of lightning events collected by different systems. for instance, the lightning events collected by the cwb system are 76, which is larger than the number of events detected by the jma system, 0, at 15:00. however, at 15:22, the lightning events collected by the jma system are 1137, which is far larger than the number of events detected by the cwb system, which is 76. it is shown from the results that the number of the collected lightning events has a significant relationship with the bandwidth that the system can collect lightning and the mainly-focused detecting frequency, which is 500khz for as3935. also, using the number of lightning events provided by the cwb as the benchmark may not have been a suitable method to examine the performance of the system. 5. conclusion a small intelligent weather station was developed, designed and established. it detected and collected rainfall, frosting, and lightning. the microprocessor could send out data via internet or store data locally. results showed that prediction model advances in technology innovation, vol. 6, no. 2, 2021, pp. 74-89 88 based on the locally collected data was more accurate than using data from weather bureau. frosting could be observed and recorded by an interval-shooting webcam. the lightning prediction was limited due to the frequency restriction of currently available sensors. for future work, we hope to increase the amount of lightning data by trying to use multiple weather boxes to collect data in the same area at the same time. besides, we also aim to find other lightning sensors that detect different bandwidths of the lightning events to complete our lightning detection function. conflicts of interest the authors declare no conflict of interest. references [1] m. a. cooper and r. l. holle, current global estimates of lightning fatalities and injuries, reducing lightning injuries worldwide, springer, 2019. [2] r. l. holle, “annual rates of lightning fatalities by country,” 20th international lightning detection conference, april 2008. [3] r. s. cerveny, p. bessemoulin, c. c. burt, m. a. cooper, z. cunjie, a. dewan, et al. “wmo assessment of weather and climate mortality extremes: lightning, tropical cyclones, tornadoes, and hail,” weather, climate, and society, vol. 9, pp. 487-497, 2017. [4] s. pattar, r. buyya, k. r. venugopal, s. iyengar, and l. patnaik, “searching for the iot resources: fundamentals, requirements, comprehensive review, and future directions,” ieee communications surveys & tutorials, vol. 20, pp. 2101-2132, april 2018. [5] p. sethi and s. r. sarangi, “internet of things: architectures, protocols, and applications,” journal of electrical and computer engineering, vol. 2017, january 2017. [6] s. aftab, m. ahmad, n. hameed, m. s. bashir, i. ali, and z. nawaz, “rainfall prediction using data mining techniques: a systematic literature review,” international journal of advanced computer science and applications, vol. 9, no. 5, pp. 143-150, 2018. [7] d. j. hand and n. m. adams, “data mining,” wiley statsref: statistics reference online, pp. 1-7, 2014. [8] j. han, j. pei, and m. kamber, data mining: concepts and techniques, elsevier, 2011. [9] a. iqbal and s. aftab, “a feed-forward and pattern recognition ann model for network intrusion detection,” international journal of computer network & information security, vol. 11, no. 4, pp. 19-25, april 2019. [10] z. chao, f. pu, y. yin, b. han, and x. chen, “research on real-time local rainfall prediction based on mems sensors,” journal of sensors, vol. 2018, 2018. [11] s. zainudin, d. s. jasim, and a. a. bakar, “comparative analysis of data mining techniques for malaysian rainfall prediction,” international journal on advanced science, engineering and information technology, vol. 6, pp. 1148-1153, 2016. [12] c. m. bishop, pattern recognition and machine learning, springer, 2006. [13] r. c. brito, f. favarim, g. calin, and e. todt, “development of a low cost weather station using free hardware and software,” latin american robotics symposium (lars) and 2017 brazilian symposium on robotics (sbr), november 2017, pp. 1-6. [14] a. geetha and g. nasira, “data mining for meteorological applications: decision trees for modeling rainfall prediction,” ieee international conference on computational intelligence and computing research, 2014, pp. 1-4. [15] p. s. yu, t. c. yang, s. y. chen, c. m. kuo, and h. w. tseng, “comparison of random forests and support vector machine for real-time radar-derived rainfall forecasting,” journal of hydrology, vol. 552, pp. 92-104, september 2017. [16] c. cortes and v. vapnik, “support-vector networks,” machine learning, vol. 20, no. 3, pp. 273-297, september 1995. [17] l. c. liang and l. t. chen, “improved svm classifier incorporating adaptive condensed instances based on hybrid continuous-discrete particle swarm optimization,” advances in technology innovation, vol. 1, no. 2, pp. 53-57, september 2016. [18] t. cover and p. hart, “nearest neighbor pattern classification,” ieee transactions on information theory, vol. 13, no. 1, pp. 21-27, january 1967. [19] l. breiman, “bagging predictors,” machine learning, vol. 24, no. 2 pp. 123-140, 1996. advances in technology innovation, vol. 6, no. 2, 2021, pp. 74-89 89 [20] s. rana and r. garg, “slow learner prediction using multi-variate naïve bayes classification algorithm,” international journal of engineering and technology innovation, vol. 7, no. 1, pp. 11, january 2017. [21] j. lai, y. liu, j. du, and q. li, “lightning detection technology and application,” international conference on meteorology observations (icmo), december 2019, pp. 1-5. [22] z. yang and s. jiang, “design of lightning detection system based on arm,” international conference on lightning protection (iclp), october 2014, pp. 346-350. [23] as3935: franklin lightning sensor ic, https://www.mouser.tw/new/ams/ams-as3935/ [24] s. boughorbel, f. jarray, and m. el-anbari, “optimal classifier for imbalanced data using matthews correlation coefficient metric,” plos one, vol. 12, no. 6, pp. e0177678, june 2017. [25] welcome to collective.ifttt! https://collectiveifttt.readthedocs.io/en/latest/ [26] y. shibai, l. f. tsai, and y. q. lee, “analysis of the characteristics of each lightning detection system in taiwan,” 2018, http://photino.cwb.gov.tw/conf/history/108/a3/a3_10_l_%e7%99%bd%e6%84%8f%e8%a9%a9_%e5%90%84% e9%96%83%e9%9b%bb%e5%81%b5%e6%b8%ac.pdf (chinese) copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). 4___aiti#7988_in press_20211213 advances in technology innovation, vol. 7, no. 1, 2022, pp. 41-55 determination of natural frequency and critical velocity of inclined pipe conveying fluid under thermal effect by using integral transform technique jabbar hussein mohmmed * , mauwafak ali tawfik, qasim abbas atiyah department of mechanical engineering, university of technology, baghdad, iraq received 28 june 2021; received in revised form 02 september 2021; accepted 03 september 2021 doi: https://doi.org/10.46604/aiti.2021.7988 abstract this study proposes an analytical solution of natural frequencies for an inclined fixed supported euler-bernoulli pipe containing the flowing fluid subjected to thermal loads. the integral transform technique is employed to obtain the spatial displacement-time domain response of the pipe-fluid system. then, a closed-form analytical expression is presented. the effects of various geometric and system parameters on the vibration characteristics of pipe-fluid system with different flow velocities are discussed. the results illustrate that the proposed analytical solution agrees with the solutions achieved in previous works. the proposed model predicts that the pipe loses the stability by divergence with the increasing flow velocity. it is evident that the influences of inclination angle and temperature variation are dramatically increased at a higher aspect ratio. additionally, it is demonstrated that the temperature variation becomes a more harmful effect than the internal fluid velocity on the stability of the pipe at elevated temperature. keywords: pipe conveying fluid, natural frequency, critical flow velocity, finite fourier sine transform, laplace transform 1. introduction fluid-conveying pipes are very important parts in many engineering structures and have many potential applications in industry. practical applications include chemical plants, nuclear reactors, oil and gas industries, heat exchangers, fuel lines of transportations, monitoring and controlling tubes, marine risers, and so on. other applications in engineering devices involve jet pumps, some kinds of valves and parts of hydraulic machinery, and submarine systems, etc. hemodynamics and the pulmonary and urinary systems also may be including some applications of the pipes conveying fluid [1-2]. the function of these pipes is very important and can be considered the blood circulation of human body. the fluid flowing inside these pipes influences the dynamic behaviors of the piping systems, which play a significant role in the stability and the performance of the systems. the influences of the fluid and structure interaction usually lead to the vibration of the pipe and even the rupture of the pipe. thus, a convenient knowledge of the dynamic behavior of these types of systems is essential and has motivated many scholars over the last six decades. extensive investigations have been executed on the vibration analysis of the fluid-conveying pipes subjected to various end and loading conditions. the first correct motion equation of the structure-fluid coupling vibration for a pipe containing flowing fluid was proposed by housner [3]. thereafter, the vibration characteristics of pipe-fluid system have been increasingly studied by several other scholars. stein and tobriner [4] discovered that the internal pressure of flowing fluid has inverse effect on the natural frequency of the system. païdoussis [5] further investigated the dynamic behavior of conservative system, predicting that the system may be undergoing the flexural oscillatory instability by divergence at a certain high flow speed, and that at higher flow speed the system can be subjected to coupled-mode flutter. plaut and huseyin [6], hatfield et al. [7], and lesmez et al. [8] further analyzed the plain and u-bend piping systems. * corresponding author. e-mail address:10493@uotechnology.edu.iq tel.: +9647805886894 advances in technology innovation, vol. 7, no. 1, 2022, pp. 41-55 nowadays, with the existence of larger pipeline systems, the demand for inclined pipes have been grown due to their special conditions, and it is therefore worthy being concerned about. moreover, these pipes can be possessed various aspect ratios of length to external diameter. a better understanding of the influence of the aspect ratio and inclination angle on the dynamic behavior of a pipe is very essential to design thin components in different industries. additionally, when the pipes are designed and analyzed, it is very important to take account of the working conditions which can affect the dynamic behavior and integrity of such components. one of the important examples of such operating conditions is the temperature variation, which is undergone during the conveyance of fluid in pipes. the pipes in such cases suffer the compressive stresses resulting from the thermal effect in addition to those generated by the fluid flowing through the pipe. such pipes can experience buckling and show the complicated dynamical behavior related to their divergence even at a very low internal flow velocity of fluid [9]. consequently, temperature is one of the main concerns among other factors and operating conditions in designing slender structures like the pipeline systems containing flowing fluid. recently, qian et al. [9] studied the effects of both linear and non-linear stress-temperature cases on the stability of pipe containing flowing fluid. zhao et al. [10] investigated the stability of the simply supported pipes induced by pulsating fluid velocity and thermal loading. kukla [11] studied the temperature distribution and lateral vibration of beam induced by external heating source. blandino and thornton [12] performed a study about the vibration characteristic of an internally heated beam. alfosail et al. [13] numerically evaluated the dynamic behavior of inclined risers. gan et al. [14] investigated the stability of an inclined cantilevered pipe containing flowing fluid. yang and wang [15] analyzed the dynamic behavior of an inclined beam induced by moving loads. from the literature review, it is observed that no previous works have been reported on the response of pipes due to a combination effect of internal fluid, inclination angle, aspect ratio, and thermal loads. furthermore, it is noted that most of the above-mentioned analyses have been based on numerical or approximate approaches in tackling linear and non-linear dynamics issues of these slender structures. this approximation is valid only for small values of the applied loads and self-weight. it has been found that there are quite a few articles that use analytical approaches to analyze the dynamic problems of the pipes conveying fluid. an analytical solution is very important for the improvements and verifications of effective applied numerical simulation tools. to address the lack of research in this aspect, the analytic solution based on integral transform technique (itt) approach (via the combination of finite fourier sine and laplace transforms) is adopted in this work to analyze the dynamic behavior of an inclined and doubly fixed supported pipe subjected to the thermal loads. firstly, the frequency-domain response of the pipe conveying fluid is obtained. then, the corresponding inverse transforms on the displacement frequency-domain responses of the pipe are conducted to obtain the spatial displacement time-domain responses. moreover, the effects of changing the aspect ratio, inclination angle, and temperature variation on the natural frequencies, as well as the critical velocities of the pipes conveying fluid under different fluid speeds are evaluated. this work is structured as follows. section 1 introduces the problem under study. section 2 presents the methods used to solve motion equations. section 3 provides the model description and solution procedure of governing equations. in section 4, results are analyzed and discussed. finally, the paper ends in section 5 with the summary and conclusions of the current work. 2. methods 2.1. laplace transform laplace transform is a linear transform that is widely applied in the dynamic problems of structure. it provides easy and powerful means to transform partial differential equations into ordinary differential equations, or transform ordinary differential equations into integral form equations, which make them much easier to be solved. laplace transform directly obtains the solution of differential equations with certain boundary values without determining the general solution first 42 advances in technology innovation, vol. 7, no. 1, 2022, pp. 41-55 [16-17]. many successful applications of laplace transform were done to study the dynamic behavior of the components with and without external forces [18-19]. assume that �(�) is a real function of time t, which is defined in the real domain [0, +∞), the laplace transform and its inversion are defined by: ( ) ( ) ( )0 [ ] st s t t l f f e dtψ ∞ −= = ∫ (1) 1 ( ) ( ) ( ) 1 [ ] 2 i st t s si f l e ds i ε ε ψ ψ π + ∞− − ∞ = = ∫ (2) where l is laplace transform, l −1 is the inverse laplace transform, s represents a complex number in the laplace domain, and si (i = 1, 2, 3, …) are the singularities of �(�). 2.2. finite fourier sine transform the fourier transform method was utilized into the field of structural dynamics by many researchers [20-21]. fourier transform is helpful in solving the differential and partial differential equations of non-periodic functions. it can be applied to solve some boundary-value and initial-value problems for partial differential equations. fourier transform is usually used in order to reduce the order of differential equations. it can be employed to convert the infinite degree of freedom systems into a superimposed double or single degree of freedom system. the finite sine fourier transform for a function of �(�), which is a real function of coordinate x in space (0 ≤ x ≤ l), can be represented as follows [22]. 1 ( ) ( ) ( )0 [ ] sins x x nf f f n xdx fπ= =∫ (3) ( ) ( ) 1 ( ) 1 [ ] 2 sin n xs n n f f f n x fπ ∞− = = =∑ (4) where �� is fourier transform, �� is the inverse fourier transform, and n is an integer. 3. model description 3.1. governing equation the system considered in the present investigation comprises a linear elastic, clamped ends, and the inclined pipes made of polypropylene-random (pp-r) material with total length (l), inner and outer diameter (di and do), thickness (t), cross-sectional area (ap), mass per unit length (mp), young’s modulus (e), and moment of inertia (i). the fluid passing through the pipe has a mass per unit length mf with an axial internal fluid flowing velocity υ. a represents the cross-sectional area of the flow and is assumed to be a constant. the fluid has no compression properties and has no viscosity. unlike the discrete systems (simple or torsional), the continuous systems, like pipe or beam, possess an infinite number of degrees of freedom that make them more difficult to be treated. hence, many assumptions were made to simplify the working process with these systems [23-24]. figs. 1 and 2 present the geometry and cartesian coordinate system, assuming that the origin of the (x, y, z) framework is placed in the space at the left end of the pipe. the properties of pipe and fluid and the numeric parameters considered in this study are presented in table 1. the current study focuses on finding the natural frequencies and critical velocities of axial flow fluid-carrying pipes under different conditions. since the non-planar motion (three-dimensional motion) does not affect the values of the natural frequencies or critical velocities, it is neglected and the planar motion (two-dimensional motion) of the pipes is adopted. fig. 3 illustrates the fluid and pipe elements, and shows all the forces acting on the system. 43 advances in technology innovation, vol. 7, no. 1, 2022, pp. 41-55 table 1 numeric values of used parameters specification value material pp-r fluid water outer diameter do (m) 0.025 inner diameter di (m) 0.018 thickness t (m) 0.0035 length l (m) 0.5-1.25 aspect ratio (length to outer diameter) l/do (20-50) do modulus of elasticity e (gpa) 0.8 at 25°c 0.38 at 50°c 0.23 at 70°c density of pipe ρp (kg/m 3 ) 909 density of fluid ρf (kg/m 3 ) 1000 coefficient of expansion α (1/k) 0.3 × 10 -4 fig. 1 three-dimension geometry of the clamped-end pipeline fig. 2 schematic of the clamped-end inclined pipe (a) fluid element (b) pipe element fig. 3 free-body diagram of the fluid and pipe element to simplify the derivation process, some assumptions are adopted: considering the vertical motion (y-axis) of the pipe; ignoring internal damping (viscoelastic effect), dissipation, external tension, and internal pressurization; assuming a small deformation; treating the pipe as an euler-bernoulli beam model. then, using the newtonian method by balancing forces on 44 advances in technology innovation, vol. 7, no. 1, 2022, pp. 41-55 the fluid and pipe elements in the transverse and longitudinal directions, and after some substitutions processes and mathematical calculations, the following motion equation of the heated and inclined pipe can be derived (more details about the assumptions and derivation process can be found in the work of mohmmed et al. [25]): ( ) ( ) ( ) ( ) ( ) 4 2 2 2 2 4 2 2 2 2 2 2 sin sin cos 2 0 f p f f p f p f f p y y y y y ei m m l x g m v n m m g m m g x x x x x y y m v m m x t t θ θ θ ∂ ∂ ∂ ∂ ∂ − + − + + + + + + ∂ ∂ ∂ ∂ ∂ ∂ ∂ + + + = ∂ ∂ ∂ (5) where ei is the flexural rigidity of the pipe; g is gravitational constant; θ is the inclination angle of the pipe with horizontal axis; qs represents internal shear stress; p is internal fluid pressure; s is the internal perimeter (internal section circumference) of the pipe; f is the transverse force between pipe wall and fluid (per unit length); t represents longitudinal tension. m is bending moment; q represents the transverse shear force in the pipe; y(x, t) is the lateral displacement of the pipe; x and t are the axial coordinate and time, respectively. n represents the thermal load in the pipe due to temperature variation and can be calculated as: n = αeap∆t, where α is thermal expansion coefficient (1/°c) and l is pipe length (m). ∆t = (tpt)x − ti is the temperature change (°c). ti is the initial temperature of the pipe (°c). (tpt)x is the instantaneous temperature of the pipe after the time (°c). the terms of eq. (5) can be described as follows: the first component represents the acceleration resulting from the restoring force, and the second and third components represent the accelerations due to the gravity force and the centrifugal force respectively. after that, the forth component associates with the acceleration due to the thermal force caused by the temperatures variation, then the fifth and sixth components illustrate the static force because of the gravity effect. the seventh component stands for the acceleration due to the coriolis forces, following the acceleration because of the system’s inertia force. rearranging eq. (5) gives: ( )( ) ( ) ( ) ( ) 4 2 2 2 4 2 2 2 sin sin cos 2 0 p p pf f f f f pf y y y y ei m m l x g n m v m m g m m g m v x x tx x y m m t θ θ θ    ∂ ∂ ∂ ∂− + − − − + + + + + ∂ ∂ ∂∂ ∂ ∂+ + = ∂ (6) eq. (6) can be rewritten in non-dimensional form as follows: ( ) 4 2 2 2 2 1 2 4 2 2 1 sin sin cos 2 0g n u g g u t t φ φ φ φ φ ζ θ θ θ β ζ ζ ζ ζ ∂ ∂ ∂ ∂ ∂ − − − − + + + + = ∂ ∂ ∂ ∂ ∂ ∂    (7) where y l φ = (8) x l ζ = (9) f f p m m m β = + (10) ( ) 3m m gl g ei + = (11) 1 2 f m u vl ei =       (12) 45 advances in technology innovation, vol. 7, no. 1, 2022, pp. 41-55 1 2 2 1 f p ei t l m m = +       (13) 2 nl n ei = (14) ( ) 2 1/ 21 sin sin 2 cosg n u g u gφ ζ θ φ θφ β φ φ θ′′′′ ′′ ′ ′− − − − + + + = −   ɺ ɺɺ (15) 3.2. solution procedure in order to discrete eq. (15) and transform it to an ordinary differential equation, the term η(ζ, τ) is used to decompose the equation into space and time as follows: ( ) ( ) ( ), t tφ ζ ζ= γ λ (16) where λ(�̅) the generalized coordinate of the system and γ(ζ) is a trial/comparison function satisfying both the geometrical and natural boundary conditions. the pipe end boundary conditions considered in the current study is clamped-clamped support, which can be symbolized as c-c. the boundary conditions at the edges can be expressed as: clamped (ζ = 0) clamped (ζ = 1), c-c at ζ = 0, γ = 0, and �γ/�ζ = 0; at ζ = 1, γ = 0, and �γ/�ζ = 0. substituting eq. (16) into eq. (15) gives: ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )2 1 21 sin sin 2 cos t g n u t g t u t t g ζ ζ θ ζ θ ζ β ζ ζ θ  ′′′′ ′′ ′ ′γ λ − − − − γ λ + γ λ + γ λ + γ λ   = − ɺ ɺɺ (17) to solve eq. (17), itts are used in this work. first, based on eq. (1), by assuming zero initial conditions case, laplace transform is introduced to suppress the time dependency. ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( )2 1 2 21 sin sin 2 cos s g n u s g us s s s g s ζ ζ θ ζ θ ζ β ζ ζ θ ′′′′ ′′ ′ ′γ λ − − − − γ λ + γ + γ λ + γ λ = −     (18) next, for reducing the difficulty of solving eq. (18), a finite fourier sine transform is introduced to convert the infinite degree of freedom pipe-fluid system into a superimposed double degrees of freedom system. thus, by using eq. (3), finite fourier sine transform with regard to spatial (space) coordinate γ is conducted on eq. (18), and the following equation can be obtained: ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) 1 1 1 2 0 0 0 1 1 1 1 2 2 0 0 0 1 0 sin sin sin sin sin sin sin 2 sin sin cos 1 sin s n d g n u s n d g s n d g s n d us s n d s s n d g n d s ζ πζ ζ θ ζ πζ ζ θ ζ ζ πζ ζ θ ζ πζ ζ β ζ πζ ζ ζ πζ ζ θ πζ ζ ′′′′ ′′ ′′λ γ − − − λ γ + λ ⋅ γ ′ ′+ λ γ + λ γ + λ γ = − ∫ ∫ ∫ ∫ ∫ ∫ ∫ (19) γ(ζ) is beam eigen function and varies according to boundary condition for the clamped supported case: ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) cosh l cos l cosh cos [ sinh sin ] sinh l sin l r r r r r r r r λ λ ζ λ ζ λ ζ λ ζ λ ζ λ λ − γ = − − − −       (20) 46 advances in technology innovation, vol. 7, no. 1, 2022, pp. 41-55 ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) sinh l sin l cosh cos [ sinh sin ] cosh l cos l r r r r r r r r λ λ ζ λ ζ λ ζ λ ζ λ ζ λ λ + γ = − − − −       (21) where ��is the root of the equation cosh(��l)cos(��l). the using of terms included in eq. (20) or (21) for fixed support may contain some difficulty and complexity in determining the integral transformation for each term in eq. (19). thus, for simplification, a polynomial function can be applied for this type of support system: ( ) 2 3 4 0 1 2 3 4 a a a a aζ ζ ζ ζ ζγ = + + + + (22) applying the boundary conditions of c-c gives: ( ) ( )4 3 2 42 aζ ζ ζ ζγ = − + (23) using orthogonal functions gives: ( )4 5 4 3 2 1 3 70 70 315 540 420 126 a a a a a a− + − + =       (24) by substituting � = 1 in eq. (21), �� = 25.20 can be obtained for the first mode. by substituting eq. (23) in eq. (19), performing the integral transformation, and reorganizing and combining similar terms, the following equation is yielded: ( ) ( ) ( ){ }{ } ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) 1 5 5 12 1 2 2 3 3 1 3 3 1 11 3 3 3 3 1 1 3 3 3 3 24 [1 1 ] 12 24 2 [1 1 ] [1 1 ] 1 [1 1 ] 2 24 242 [1 1 ] [1 1 ][1 1 ] sin 24 2 1 12 [1 1 ] [1 2 1 ] [1 1 ] n n n n n nn n n n n n s s u n u n n n n n nn g n n n π β π π π π π ππ θ π π π + + + + ++ + + + − − + + − + + − − + + − + − − + − ++ − − + − + − + − + + + −                         ( ) ( ) 4 1 cos .1 s g f a s θ λ = −                              (25) where ( ) ( ) ( ) 1 1 0 [1 1 ] 1 sin 1 n n d f n πζ ζ π + + − = =∫ (26) eq. (25) can be simplified by assuming: ( )( )1 21 3 3 12 2 1 1 n f z u n β π = + −     (27) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) 1 11 3 3 3 3 1 2 2 1 1 3 3 3 3 2 24 242 [1 1 ] [1 1 ][1 1 ] 24 [1 1 ] sin 24 2 1 12 [1 1 ] [1 2 1 ] [1 1 ] n nn n f n n n n n n nn z n u g n n n n π π ππ θ π π π π + ++ + + + + − − + − ++ − − = + − − + + − + − + − + + + −                         (28) 47 advances in technology innovation, vol. 7, no. 1, 2022, pp. 41-55 ( ) ( ){ }{ } ( ) ( )1 12 1 25 5 3 3 4 24 1 1 [1 1 ] [1 1 ] cos .1 n n f f g s z s z s f n n a s θ π π + + ∴ + − − + − + + λ = − (29) additionally, the following equations can be assumed: ( ) ( ) 1 1 1 1 5 5 3 3 24 1 [1 1 ] [1 1 ] f f n n z n n ϕ π π + + = + − − + − (30) ( ) ( ) 22 1 1 5 5 3 3 24 1 [1 1 ] [1 1 ] f f n n z n n ϕ π π + + = + − − + − (31) ( ) ( ) ( ) ( ) 4 1 1 5 5 3 3 2 2 1 1 cos .1 24 1 [1 1 ] [1 1 ] n n f f g f a s n n s s s θ π π ϕ ϕ + + − + − − + − ∴ λ = + + (32) by applying the fourier-laplace inversion, the spatial displacement time-domain response (dynamic response !) for the inclined fixed supported pipe conveying fluid can be given: ( ) ( ) ( ) 4 4 4 2 2 cos .f , 24 2 f g t t a n n θ φ ζ ζ π π − = γ −     (33) ( ) ( ) ( ) 2 1 1 2 1 2 1 2 2 1 1 1 f f f f f f f f f f f f t t t e e α α α α α α α α α α − − = + − −       (34) 2 1 12 1 2 4 f f f f i ϕ ϕ α ϕ= + − (35) 2 1 12 2 2 4 f f f f i ϕ ϕ α ϕ= − − (36) by isolating characteristic equations, the dimensionless fundamental natural frequency (ωn) and its complimentary value (ωm) are: 2 2 2 1 1 12 2 2 2 4 4 f f f n f n f ϕ ϕ ϕ ω ϕ ω ϕ= − ⇒ = −           ∓ (37) 2 m f ω ϕ= − (38) the fundamental natural frequency ωn is adopted in this work. the aforementioned equations are solved for each value of variables by using matlab 2019b software program to determine the fundamental natural frequency of the system. the simplified flow chart is illustrated in fig. 4. 48 advances in technology innovation, vol. 7, no. 1, 2022, pp. 41-55 fig. 4 the flowchart for matlab code 4. results and discussion 4.1. validation of analytical model to validate the analytical calculation method proposed in this work, the fundamental natural frequency is calculated at stationary fluid case (u = 0) and at fluid velocities u = 0.5, u = 1, and u = 1.5, respectively, and compared with those in the work of ni et al. [26] and liang et al. [27]. table 2 presents the geometry and material properties used in this comparison, and lists the obtained results. it is clear from table 3 that the relative errors between the results in current study and those in refs. [26-27] are less than 2.46%. these little differences may be attributed to the using of different models in the two references. in refs. [26-27], the pipe is modeled according to timoshenko theory; in the current work the euler-bernoulli beam is adopted. it can be concluded from this comparison that the applied integral transform method (via the combination of the finite fourier sine transform and laplace transform) gives quite reliable results provided that the fluid flow velocity is moderate. table 2 parameters of pipe and fluid young’s modulus pipe size pipe length pipe density fluid density 210 gpa do = 324 mm 32 m 8200 kg/m 3 908.2 kg/m 3 di = 292 mm table 3 comparison of dimensionless natural frequency of the fixed supported pipe with different fluid velocities frequency fluid velocity current value ref. [26] relative error % ref. [27] relative error % ωn ωn ωn u = 0 22.3087 22.2597 0.22 22.42 0.49 u = 0.5 22.2534 22.0910 0.73 22.27 0.07 u = 1 22.0864 21.9224 0.74 22.11 0.10 u = 1.5 21.8053 21.4165 1.81 21.28 2.46 4.2. natural frequency in the current study, the natural frequency and stability of an elastic, inclined, and doubly fixed supported pipe conveying a newtonian and incompressible fluid at different temperature are investigated. the natural frequencies of the pipe are analytically calculated based on eq. (37). the values of the system parameters used to obtain the results are listed in table 1. to connect the subcritical, critical, and post-critical vibratory behaviors, the continuity profile of the real and imaginary components of pipe-fluid system’s natural frequencies (instead of separated profile as widely observed in the literature) is adopted in this study by plotting the natural frequencies in absolute values against the internal flow velocity. start input constants g, ρf, ρp, di, do, ti, and α define variables v, l/do, θ, (tpt)x, and t calculate the parameters a, ap, mf, mp, and i define the parameters e, n, a4, and n calculate the dimensionless parameters β, g, "#, and �̅ set the interval of ζ find the terms $% , $%&, !% , and !% & find the natural frequency '( plot the results print output end 49 advances in technology innovation, vol. 7, no. 1, 2022, pp. 41-55 the natural frequency of the pipe-fluid system is demonstrated in figs. 5-9. two zones are identified by observing the profile of fundamental natural frequency with respect to the influence of flow speed in figs. 5 and 8; where the two zones illustrate the divergence case when the flow is lower and higher than the critical points of flow. in general, the results show that the fundamental natural frequency of the system is affected by factors such as fluid velocity, and tends to reduce as the velocity of fluid flowing within the pipe increases. with the increasing fluid velocity, the fundamental natural frequency of the pipe vanishes, and then the divergence instability happens. the corresponding fluid velocity, at which the fundamental natural frequency vanishes, is named critical velocity. this is because the increase of the velocity of fluid flowing inside the pipe leads to the weakness of the pipe stiffness. almost similar conclusions in the subcritical fluid velocity range (1st zone) were observed in the work of ni et al. [26] and in the supercritical regime (2nd zone) in the work of adelaja [28] and olunloyo et al. [29], while for the whole spectrum results the same trend was found in the work of plaut [30]. fig. 5 presents the effect of different aspect ratio (length to diameter ratio) on the natural frequency. it is clear that the natural frequencies of the fluid-conveying pipe decrease with the increase of length to diameter ratio. similarly, the critical fluid velocity decreases with the increase of the aspect ratio. this suggests that the pipe-fluid system will hastily reach divergence instability situation with higher aspect ratio. the reason is that the higher length to diameter ratio leads to the increase of the pipe weight resulting in the weakness of its stuffiness. the same trend was observed in the work of liang [27] and olunloyo et al. [31] with the increase of the pipe length. meanwhile, by observing fig. 5, it is evident that the level of reduction in the magnitude of natural frequencies and critical velocity for the same aspect ratio reduces with the increase of the support angle. at the same time, by observing fig. 5, it is clear that the level of reduction in the value of the natural frequency and critical velocity decreases for the same aspect ratio with the increase of the inclination angle. (a) ) = 0° (b) ) = 15° (c) ) = 30° fig. 5 the natural frequency value versus fluid velocity with different aspect ratios for the case of t = 25°c and n = 1 50 advances in technology innovation, vol. 7, no. 1, 2022, pp. 41-55 the effect of the pipe’s inclination angle and aspect ratio on the fundamental natural frequency of the structure is studied further in figs. 6 and 7 respectively. in general, the results indicate the increase of natural frequencies with the increase of the inclination angle. in addition, it can be observed that the enhancement in the natural frequency values is larger at higher aspect ratio. for example, the natural frequency value increases approximately by 0.18% (from 23.104 for θ = 0° to 23.146 for θ = 60°) at aspect ratio = 20 while its value increases by 5.2% (from 22.223 for θ = 0° to 23.380 for θ = 60°) at aspect ratio = 50. additionally, the results in fig. 7 show an interesting phenomenon. when the aspect ratio equals 50 and the inclination angle is in the range of 40-90, the influence of the inclination angle on the fundamental natural frequencies becomes higher than the influence of the aspect ratio on the fundamental natural frequencies, i.e., the increase of inclination angle compensates the loss in natural frequency value due to the increase of aspect ratio. also, it can be observed from fig. 7 that the natural frequency of the pipe has an increasing trend up to the inclination angle of θ = 90° and a reverse trend afterward, i.e., beyond the angle 90° the behavior is reflected. this can be explained as follows: for an inclined pipe, the gravitational force generated by the weight of the pipe and the fluid is divided into two components, the lateral component and the axial component relative to the pipe axes. when the angle of inclination of the pipe is * 90,, the axial component exerts a tensile force on the pipe, which increases with the increase of the inclination angle of the pipe and the aspect ratio. this in turn reduces the impact of the lateral component, responsible for the occurrence of the lateral displacement of the pipe, and thus increases natural frequencies of the inclined pipe. on the contrary, as the pipe’s inclination angle ) 90,, the axial component gradually decreases and the lateral component becomes larger with the increase of the inclination angle which causes the natural frequency become lower and eventually same as the horizontal case. (a) l/d = 20 (b) l/d = 30 (c) l/d = 50 fig. 6 the natural frequency value versus internal fluid velocity with different inclination angles and aspect ratios for the case of t = 25°c and n = 1 51 advances in technology innovation, vol. 7, no. 1, 2022, pp. 41-55 fig. 7 the natural frequency value versus inclination angle with different aspect ratios for the case of t = 25°c, u = 1, and n = 1 fig. 8 describes the effect of the temperature variation on the fundamental natural frequency and the critical velocity of the pipe conveying fluid for different aspect ratios. the results reveal that the temperature variation has a great influence on the fundamental natural frequencies and the critical velocity. here, as temperature increases, a significant reduction in fundamental natural frequency and critical velocity is observed. thus, the temperature variations have a softening effect on the pipe stiffness, which is consistent with the work of ashley et al. [32] where the natural frequency was found to be inversely proportional to the temperature and with the work of orolu et al. [33] where the critical velocity reduces with the increase of temperature. also, it is noted that the temperature effect on the fundamental natural frequency and the critical velocity of the pipe conveying fluid is considerably growing at higher aspect ratio. for example, the natural frequency and critical velocity reduce from “23.23 and 7.46” to “22.89 and 7.35” respectively at l/d = 20, while the lowering is from “23.2 and 7.38” to “4.55 and 0.09” respectively at l/d = 50. it can be inferred from these results that the pipe may be losing their stability even with zero fluid velocity with the increasing temperature. this suggests that the temperature effect becomes more essential parameter, for engineers and designers, than the velocity of fluid flowing inside the pipe at higher aspect ratio as designing the pipelines systems. the temperature effect on natural frequencies is further examined by plotting in fig. 9 the variation of the fundamental natural frequencies with the pipe’s inclination angle for different temperatures. the curves in fig. 9 manifest that when the variation in the pipe temperature is slight, the change of the inclination angle has very little effect on the fundamental natural frequency. however, as the pipe temperature increases, the influence of inclination angle becomes more noticeable. (a) l/d = 20 fig. 8 the natural frequency value versus fluid velocity with different temperatures and aspect ratios for the case of θ = 0° and n = 1 52 advances in technology innovation, vol. 7, no. 1, 2022, pp. 41-55 (b) l/d = 30 (c) l/d = 50 fig. 8 the natural frequency value versus fluid velocity with different temperatures and aspect ratios for the case of θ = 0° and n = 1 (continued) fig. 9 the natural frequency value versus pipe’s inclination angle with different temperatures for the case of u = 1, n = 1, and l/d = 50 5. summary and conclusions the itt for natural frequency and dynamic response computation of pp-r pipe containing incompressible flowing fluid is proposed in this article. the euler-bernoulli beam model is applied to obtain the transverse vibration of the pipe containing flowing fluid. the closed-form analytical expression used to calculate both natural frequency and overall dynamic of the fixed supported pipe containing flowing fluid is derived based on the combination of finite fourier sine and laplace transforms and their inverse transform. the effect of various parameters (internal flow velocity, aspect ratio, inclination angle, and temperature variation) on the vibration characteristic of the pipe conveying fluid is presented. the comparison of the numerical results of natural frequency at different internal fluid velocity with the results in published references is conducted and shows a good agreement. the results reveal that the fundamental natural frequency of the pipe containing flowing fluid reduces as the internal flow velocity increases. once the fundamental natural frequency reduces to zero, the divergence instability happens. the findings of the current study can be summarized as follows: (1) the proposed analytical method has a clear concept, suitable for hand computation, and gives a theoretical basis for more engineering applications of the inclined fixed supported pipe conveying fluid under thermal loads. (2) the aspect ratio, temperature variation, and inclination angle strongly affect both the natural frequency and critical velocity of the system. 53 advances in technology innovation, vol. 7, no. 1, 2022, pp. 41-55 (3) there is a strong coupling between the aspect ratio of pipe length to its outside diameter with temperature variation and inclination angle. (4) the temperature variation is a major concern rather than the internal fluid velocity in the design of the pipe containing flowing fluid at higher aspect ratio. (5) the inclination angle has larger impact on vibration characteristics at higher aspect ratio, which should be paid attention in engineering. (6) the natural frequency and critical velocity of the pipe-fluid system decrease while its dynamic deflection increases with the increase of temperature. the divergence can occur even when the fluid velocity equals zero with the increasing temperature of the pipe. (7) the natural frequency and critical velocity of the pipe conveying fluid decrease while its dynamic deflection increases with the increase of aspect ratio. conflicts of interest the authors declare no conflict of interest. references [1] m. p. païdoussis, fluid-structure interactions: slender structures and axial flow, 2nd ed., london: academic press, 2014. [2] v. olunloyo, c. osheku, and p. olayiwola, “concerning the effect of a viscoelastic foundation on the dynamic stability of a pipeline system conveying an incompressible fluid,” journal of applied and computational mechanics, vol. 2, no. 2, pp. 96-117, 2016. [3] g. w. housner, “bending vibration of a pipe line containing flowing fluid,” journal of applied mechanics, vol. 19, no. 2, pp. 205-208, june 1952. [4] r. a. stein and m. w. tobriner, “vibration of pipes containing flowing fluids,” journal of applied mechanics, vol. 37, no. 4, pp. 906-916, december 1970. [5] m. p. païdoussis, “flutter of conservative systems of pipes conveying incompressible fluid,” journal of mechanical engineering science, vol. 17, no. 1, pp. 19-25, february 1975. [6] r. h. plaut and k. huseyin, “instability of fluid-conveying pipes under axial load,” journal of applied mechanics, vol. 42, no. 4, pp. 889-890, december 1975. [7] f. j. hatfield, d. c. wiggert, and r. s. otwell, “fluid structure interaction in piping by component synthesis,” journal of applied mechanics, vol. 104, no. 3, pp. 318-325, september 1982. [8] m. w. lesmez, d. c. wiggert, and f. j. hatfield, “modal analysis of vibrations in liquid-filled piping systems,” journal of fluids engineering, vol. 112, no. 3, pp. 311-318, september 1990. [9] q. qian, l. wang, and q. ni, “instability of simply supported pipes conveying fluid under thermal loads,” mechanics research communications, vol. 36, no. 3, pp. 413-417, april 2009. [10] d. zhao, j. liu, and c. q. wu, “stability and local bifurcation of parameter-excited vibration of pipes conveying pulsating fluid under thermal loading,” applied mathematics and mechanics, vol. 36, no. 8, pp. 1017-1032, august 2015. [11] j. k. kukla, “application of the green functions to the problem of the thermally induced vibration of a beam,” journal of sound and vibration, vol. 262, no. 4, pp. 865-876, may 2003. [12] j. r. blandino and e. a. thornton, “thermally induced vibration of an internally heated beam,” journal of vibration and acoustics, vol. 123, no. 1, pp. 67-75, january 2001. [13] f. k. alfosail, a. h. nayfeh, and m. i. younis, “natural frequencies and mode shapes of statically deformed inclined risers,” international journal of non-linear mechanics, vol. 94, pp. 12-19, september 2017. [14] c. gan, s. jing, s. yang, and h. lei, “effects of supported angle on stability and dynamical bifurcations of cantilevered pipe conveying fluid,” applied mathematics and mechanics, vol. 36, no. 6, pp. 729-746, may 2015. [15] d. s. yang and c. m. wang, “dynamic response and stability of an inclined euler beam under a moving vertical concentrated load,” engineering structures, vol. 186, pp. 243-254, may 2019. 54 advances in technology innovation, vol. 7, no. 1, 2022, pp. 41-55 [16] b. ram, engineering mathematics, india: pearson education, 2009. [17] x. liang, z. wu, l. wang, g. liu, z. wang, and w. zhang, “semianalytical three-dimensional solutions for the transient response of functionally graded material rectangular plates,” journal of engineering mechanics, vol. 141, no. 9, pp. 1-17, september 2015. [18] y. d. li and y. r. yang, “forced vibration of pipe conveying fluid by the green function method,” archive of applied mechanics, vol. 84, no. 12, pp. 1811-1823, july 2014. [19] h. b. wen, y. r. yang, p. li, y. d. li, and y. huang, “a new method based on laplace transform and its application to stability of pipe conveying fluid,” shock and vibration, vol. 2017, 1472601, 2017. [20] z. lai, l. jiang, and w. zhou, “an analytical study on dynamic response of multiple simply supported beam system subjected to moving loads,” shock and vibration, vol. 2018, 2149251, 2018. [21] x. tan, h. ding, and l. q. chen, “nonlinear frequencies and forced responses of pipes conveying fluid via a coupled timoshenko model,” journal of sound and vibration, vol. 455, pp. 241-255, september 2019. [22] l. jiang, y. zhang, y. feng, w. zhou, and z. tan, “dynamic response analysis of a simply supported double-beam system under successive moving loads,” applied sciences, vol. 9, no. 10, 2162, may 2019. [23] w. t. thomson, theory of vibration with applications, london: unwin hyman, 1988. [24] m. ali, i. alshalal, and j. h. mohmmed, “effect of the torsional vibration depending on the number of cylinders in reciprocating engines,” international journal of dynamics and control, vol. 9, pp. 901-909, january 2021. [25] j. h. mohmmed, m. a. tawfik, and q. a. atiyah, “natural frequency and critical velocities of heated inclined pinned pp-r pipe conveying fluid,” journal of achievements in materials and manufacturing engineering, vol. 107, no. 1, pp. 15-27, 2021. [26] q. ni, z. l. zhang, and l. wang, “application of the differential transformation method to vibration analysis of pipes conveying fluid,” applied mathematics and computation, vol. 217, no. 16, pp. 7028-7038, april 2011. [27] x. liang, x. zha, x. jiang, l. wang, j. leng, and z. cao, “semi-analytical solution for dynamic behavior of a fluid-conveying pipe with different boundary conditions,” ocean engineering, vol. 163, pp. 183-190, september 2018. [28] a. o. adelaja, “natural frequencies of pressurized hot fluid conveying pipes,” fuoye journal of engineering and technology, vol. 3, no. 2, pp. 131-135, september 2018. [29] v. o. olunloyo, c. a. osheku, and p. s. olayiwola, “a note on an analytic solution for an incompressible fluid-conveying pipeline system,” advances in acoustics and vibration, vol. 2017, 8141523, 2017. [30] r. h. plaut, “postbuckling and vibration of end-supported elastic pipes conveying fluid and columns under follower loads,” journal of sound and vibration, vol. 289, no. 1-2, pp. 264-277, january 2006. [31] v. o. olunloyo, c. a. osheku, and s. i. kuye, “vibration and stability behaviour of sandwiched viscoelastic pipes conveying a non-newtonian fluid,” 29th international conference on ocean, offshore, and arctic engineering, pp. 99-111, june 2010. [32] h. ashley and g. haviland, “bending vibrations of a pipe line containing flowing fluid,” journal of applied mechanics, vol. 17, no. 3, pp. 229-232, september 1950. [33] k. o. orolu, t. a. fashanu, and a. a. oyediran, “cusp bifurcation of slightly curved tensioned pipe conveying hot pressurized fluid,” journal of vibration and control, vol. 25, no. 5, pp. 1109-1121, march 2019. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). 55 3___aiti#8492_in press_20211213 advances in technology innovation, vol. 7, no. 1, 2022, pp. 30-40 clustering analysis with embedding vectors: an application to real estate market delineation changro lee * department of real estate, kangwon national university, chuncheon, south korea received 16 september 2021; received in revised form 15 november 2021; accepted 16 november 2021 doi: https://doi.org/10.46604/aiti.2021.8492 abstract although clustering analysis is a popular tool in unsupervised learning, it is inefficient for the datasets dominated by categorical variables, e.g., real estate datasets. to apply clustering analysis to real estate datasets, this study proposes an entity embedding approach that transforms categorical variables into vector representations. three variants of a clustering algorithm, i.e., the clustering based on the traditional euclidean distance, the gower distance, and the embedding vectors, are applied to the land sales records to delineate the real estate market in gwacheon-si, gyeonggi province, south korea. then, the relevance of the resultant submarkets is evaluated using the root mean squared errors (rmse) obtained from a hedonic pricing model. the results show that the rmse in the embedding vector-based algorithm decreases substantially from 0.076-0.077 to 0.069. this study shows that the clustering algorithm empowered by embedding vectors outperforms the conventional algorithms, thereby enhancing the relevance of the delineated submarkets. keywords: clustering, categorical data, high-cardinality, entity embedding, market delineation 1. introduction machine learning is rapidly expanding to various applications; particularly, it has been used with great success in several applications such as computer vision, natural language processing, speech recognition, and time-series forecasting [1-4]. there are two main types of tasks within the field of machine learning: supervised and unsupervised learning. in a supervised learning framework, the algorithm learns on a labeled dataset. however, an unsupervised learning framework provides unlabeled data that the algorithm attempts to learn by extracting meaningful features without any guidance from the labels. clustering is a de facto standard tool employed in unsupervised learning and is extensively used as a technique for discovering hidden patterns in data, such as consumer groups based on demographic profiles, or real estate submarkets based on property characteristics [5]. although clustering analysis is a popular tool in unsupervised learning, it experiences difficulty when categorical variables are dominant in the dataset. clustering is the process of grouping similar data points. the similarity between data points is calculated by a distance measure, which is essentially based on the geometry and distance in the euclidean space. this concept of physical distance is well-suited to continuous data, but it is not directly applicable to categorical data. for instance, categorical data such as the color of products with each element being black, blue, and red cannot be clustered based on the distance between the three colors. several methods, such as one-hot encoding approaches and specialized distance metrics for categorical data, have been proposed in the literature to solve this problem; however, these methods have not performed satisfactorily. this study attempts to overcome this limitation by using an entity embedding approach. the entity embedding maps categorical data into metric spaces, such as the euclidean space, and thus can alleviate the discrete properties inherent in categorical data. * corresponding author. e-mail address: spatialstat@naver.com tel.: +82-33-250-6833 advances in technology innovation, vol. 7, no. 1, 2022, pp. 30-40 this study offers a way to enhance the clustering performance via an entity embedding approach, which has been actively used in the field of machine learning, particularly in natural language processing tasks. first, a study area is chosen to delineate the real estate market. because the real estate market is always localized, it is routine to segment the market before performing the main tasks, such as property price estimation for tax assessment [6]. second, to construct appropriate submarkets within the study area, the sales records of lots sold from 2016 to 2018 are analyzed and grouped by using several variants of a k-means clustering algorithm, respectively. an entity embedding approach is applied to handle high-cardinality variables when performing clustering analysis. finally, the predictive accuracy of different sets of submarkets created earlier is compared in the context of property valuation. the contribution of this study can be summarized as follows: (1) by adopting the framework proposed in this study, clustering algorithms can be utilized in a more efficient manner when applied to the tasks in which high-cardinality categorical data are dominant. (2) this study shows that the entity embedding approach can be used effectively in processing structured data, i.e., traditional tabular format data with rows and columns, beyond the field of unstructured data, such as images and free-form texts. (3) to the best of the authors’ knowledge, real estate markets have not been delineated with the aid of entity embedding. this study attempts to construct real estate submarkets by exploiting the entity embedding approach. the remainder of this study is organized as follows. section 2 presents the background information on clustering analysis and real estate market delineation. section 3 describes the dataset, embedding vectors, and silhouette score used to determine the optimal number of clusters. the results and interpretations are provided in section 4. a summary of this study and conclusions are presented in section 5. 2. literature review 2.1. clustering analysis and entity embedding for a clustering algorithm to group observations together, the notion of (dis)similarity among observations must be defined first. a popular dissimilarity metric, i.e., a distance metric for clustering, is the euclidean distance. however, the euclidean distance is only applicable to continuous variables and not categorical variables. when categorical variables must be used for clustering, the simplest technique to calculate the euclidean distance is to convert each element in a categorical variable to a separate binary dummy variable, and this is often called the one-hot encoding approach. although this approach is applicable to a clustering algorithm, it becomes cumbersome and inefficient for processing data when categorical variables are dominant in the dataset. alternatively, a special distance metric that can handle both continuous and categorical data types may be utilized, and a well-known such metric is the gower distance [7]. to calculate the gower distance, an appropriate distance metric for each variable type is selected and scaled to fall between 0 and 1; the euclidean distance is chosen for continuous variables, and the dice coefficient is chosen for categorical variables [8]. the dice coefficient is a well-known similarity measure for categorical data [9]. subsequently, the average value of all variables is calculated to create the final gower distance. the dice coefficient is a measure that replaces the euclidean distance metric with a matching similarity measure, which is easily applicable to categorical data. many clustering algorithms proposed for grouping categorical variables are implemented based on the modified concept of (dis)similarity, such as the dice coefficient; k-modes is one such well-known algorithm [10]. k-modes is a clustering algorithm that is specialized in classifying categorical data. in addition, useful modifications to the k-modes have been proposed recently to achieve better performance [11-12]. 31 advances in technology innovation, vol. 7, no. 1, 2022, pp. 30-40 however, the above approaches become very inefficient when a categorical variable is highly cardinal; that is, the number of unique values in a categorical variable is very large. to solve this problem, this study utilizes an entity embedding approach for high-cardinality categorical variables. an entity embedding technique is used for mapping categorical values to a multidimensional space with fewer dimensions than the original number of levels, where the values with similar function outputs are close to each other [13-14]. an entity embedding is usually created via neural network training in the form of embedding vectors. the entity embedding technique is widely used in the field of natural language processing because words can be viewed as the agglomeration of high-cardinality categorical variables [15-18]. according to the work of guo et al. [13], the traditional one-hot encoding of a categorical variable can be expressed as: ii x a x → δ (1) where ���� denotes the kronecker delta, and the possible values for � are the same as ��. the element is nonzero only when � � ��. based on the work of guo et al. [13], the entity embedding of �� can be expressed as: i ii a a x a x x w w→ ∑ =δ (2) where � represents the weight connecting the one-hot encoding layer to the embedding layer in a neural network. thus, the mapped embedding is the weight of the embedding layer and can be learned in the same way as the parameters of other neural network layers. in short, entity embedding is an approach of converting a categorical variable into a vector representation. the entity can be a word in natural language processing or a level in a categorical variable. synonyms in natural language or similar levels in a categorical variable come to have similar vector values after neural network training. fig. 1 illustrates the vector representation of entity embedding for the categorical variable “pet breeds”. a few studies utilizing entity embedding for clustering have been reported in the literature [19-20]. however, they attempted to apply the entity embedding technique to unstructured data such as imagery data, more specifically, the modified national institute of standards and technology (mnist) database. this image dataset is already pre-processed public data. this study differs from them in that the entity embedding is applied to structured data, that is, tabular format data with rows and columns, and the real estate dataset used in the study is collected from scratch and cleaned thoroughly. by adopting the entity embedding approach, especially for the data with high-cardinality categorical variables, several problems can be mitigated. first, an unrealistic amount of computational resource consumption can be avoided owing to the traditional one-hot encoding of high-cardinality variables. second, different values of categorical variables can be treated in a meaningful manner, instead of processing them completely independently from each other. third, the feature engineering step is avoided because the embeddings, by nature, can intrinsically group similar values together, thereby removing the need for domain experts to learn the relationships between values in the same categorical variable. finally, the learned embedding vectors can be visualized using a dimensionality reduction technique, providing additional insights to business practitioners. fig. 1 vector representation of entity embedding in the case of “pet breeds” 32 advances in technology innovation, vol. 7, no. 1, 2022, pp. 30-40 in this study, the embedding vectors are created by neural network training for high-cardinality variables and added to the existing dataset that comprises continuous variables. subsequently, to overcome the inefficiency observed in the one-hot encoding and gower distance approaches, the euclidean distance is calculated for this combined dataset. 2.2. application to real estate market delineation because buyers and sellers require different types of properties for different reasons, the real estate market is almost always divided into submarkets within those property types and sales motivations. because real estate is best characterized by its fixed location, its market delineation usually involves defining geographical boundaries [21-22]. it is routine that licensed appraisers and automated valuation models first identify the relevant submarket boundaries and then analyze the supply and demand of the proposed properties within a specific submarket. a submarket is generally associated with a group of similar properties in terms of price level and geographical location. various methods have been proposed to delineate the real estate market; examples of these methods range from well-established approaches, such as automatic zoning procedure [23-24] and spatial ‘k’luster analysis by tree edge removal algorithm [25-26], to more recently suggested methods, such as adaptive density-based spatial clustering [27-28]. most of these methods are based on a range of data-driven algorithms. the most popular algorithm for market delineation is a clustering algorithm. conversely, a clustering procedure is performed using relevant variables such as price and geographical location, following which the clustering results are evaluated considering both internal and external validity measures [29-30]. representative indices for internal validity measures include inertia and silhouette score; however, the criteria for external validity measures can vary depending on the dataset and research purpose [31-34]. this study adopts a k-means clustering algorithm to group similar observations (sales records of land in this study) into several distinct clusters, which serve as real estate submarkets for the purpose of price estimation. several studies utilized price estimation results to test market segmentation, and they proved that market delineations improved the accuracy of price estimation [35-37]. 3. embedding vectors and silhouette score 3.1. dataset this study estimates the land prices in gwacheon-si, gyeonggi province, south korea. with a population of over 13 million, gyeonggi province is the most populous province in south korea [38]. the gwacheon-si government is one of the 45 local governments in gyeonggi province, and gwacheon-si comprises both urban and rural areas. this mixed landscape is the main reason behind choosing gwacheon-si for the analysis because the heterogeneity strongly indicates the need for market segmentation. the dataset is periodically provided in a comma-separated value file format by the ministry of land, infrastructure and transport (molit) website, and have been released by the government since 2006, when the act on report on real estate transactions was enforced. these land sales records are used in a range of government administrations, such as monitoring the real estate market and tax assessment for traded properties. the records are also used in private sectors, such as for the valuation of collaterals for loan approval. the dataset comprises the lots sold in gwacheon-si from 2016 to 2018, and includes the attributes of 1,111 samples. these attributes include sales price, zone improvement plan (zip) code, lot shape, and bearing. the following variables are used for the clustering analysis: sales price, longitude and latitude (hereafter denoted as x-coordinate and y-coordinate, respectively), zip code, and zone. by employing these five variables, the lots close to each other in terms of sales price, 33 advances in technology innovation, vol. 7, no. 1, 2022, pp. 30-40 geographical location, and zoning are expected to group together to form a submarket. although more variables, such as lot shape (regular or irregular) and bearing, are available as inputs, they reflect the physical characteristics of individual lots, rather than neighborhood/submarket characteristics; thus, these other variables are not employed for market delineation. table 1 presents the descriptive statistics of sales samples. table 1 descriptive statistics of the 1,111 lots sold from 2016 to 2018 variable min. mean median max. sales price (krw/m 2 ) 6,938 1,539,433 379,753 22,783,807 zip code (25 levels) #13820: 245 (22.1%) #13840: 164 (14.8%) #13814: 101 (9.1%) #13813: 81 (7.3%) #13801: 75 (6.8%) #13824: 73 (6.6%) zone (7 levels) green belt: 951 (85.6%) residence 3: 103 (9.3%) residence 5: 40 (3.6%) natural & green area: 8 (0.7%) residence 2: 6 (0.5%) residence 4: 2 (0.2%) residence 1: 1 (0.1%) *note: only primary levels are presented in the zip code for readability. the median sales price of lots in gwacheon-si is 379,753 krw/m 2 , and most area is zoned for the green belt (85.6%). as shown in the table, zip code and zone are categorical variables. particularly, the zip code is highly cardinal with 25 levels. all continuous variables (sales price, x-coordinate, and y-coordinate) are scaled to have a mean of zero and a standard deviation of one before being fed into the clustering algorithm. a portion of the dataset (222 samples, 20%) is reserved to evaluate the performance of clustering algorithms. the land sales records are collected by local governments and released to the public on a monthly basis. the records can be utilized to detect overheating spots in real estate submarkets and diagnose the sustainability of the overall market. in the future, local governments need to collect and agglomerate the land sales data in a more frequent cycle, for example, on a weekly or daily basis, to adapt to a rapidly changing real estate market. in addition to optimizing the data accumulation process, data-driven algorithms based on machine learning must be developed and deployed in administration tasks ranging from disclosing the real estate market in a transparent manner to detecting fraudulent land transactions. 3.2. creating embedding vectors in the case of a high-cardinality categorical variable, such as the zip code, a one-hot encoding technique cannot be efficient in handling a large number of elements in a variable; thus, the high-cardinality categorical variable should be represented in the form of embedding vectors. the embedding vectors utilized in this study are obtained from the training results of a neural network. the aforementioned network is a fully connected layer network with the following architecture: four input variables are employed, and then two embedding layers corresponding to the categorical variables are additionally specified and added to the network. an appropriate number of dimensions has to be determined for each embedding layer, and the prediction performance for various dimension sizes is reviewed through the usual cross-validation process. the number of dimensions assigned to each categorical variable through this cross-validation process is 10 and 3 for the zip code and zone, respectively. the number of dimensions in fig. 2 is selected based on the results of the heuristic grid search. the grid search uses five-fold cross-validation to evaluate the possible combinations of values (the number of dimensions for the zip code and zone in this study), and the mean squared error (mse) between the observed land prices and the predicted prices is used as a criterion for 34 advances in technology innovation, vol. 7, no. 1, 2022, pp. 30-40 the evaluation. as a rule of thumb, half the number of original levels is often used as a reference number for dimensions [39]. the zip code comprises 25 levels, and the zone consists of 7 levels, as shown in fig. 2. thus, according to this rule of thumb, good candidate numbers of dimensions for the zip code and zone are 12-13 and 3-4, respectively. hence, the number of dimensions for the zip code is chosen as 10, and that for the zone is chosen as three, considering the results of the heuristic grid search and the rule of thumb together. finally, to include more parameters to capture minor data nuances, three hidden layers are added to the end of the architecture. the output layer with one neuron corresponds to the land price estimated by the neural network. the final architecture of the neural network used to obtain the embedding vectors is presented in fig. 2. the resultant embedding vectors assume the following form: a 25 × 10 matrix for the zip code and a 7 × 3 matrix for the zone. table 2 presents the embedding vectors of the zone. interpreting the learned embedding vectors always involves subjective judgments, and this study does not attempt to interpret their meanings because the primary goal is to reuse them in a subsequent clustering algorithm and achieve a performance better than those of the baseline algorithms. fig. 2 architecture of the neural network used to obtain embedding vectors table 2 embedding vectors (7 × 3 matrix) of the zone learned from neural network training zone vector 1 vector 2 vector 3 residence 1 0.21375 0.20492 0.24875 residence 2 -0.13061 -0.18887 -0.21148 residence 3 0.01262 0.01311 0.01660 residence 4 0.04791 0.04156 -0.00550 residence 5 -0.02817 -0.02647 -0.02822 natural & green area 0.00644 -0.03270 0.01594 green belt 0.01215 -0.02023 -0.00067 3.3. silhouette score when fitting a clustering algorithm such as k-means to a dataset, it is always subject to the judgment of a researcher to determine the optimal number of clusters, making the results vulnerable to criticism of subjectivity. two popular methods are used to overcome this criticism: inertia and silhouette score analysis [40-41]. the former is defined as the mean squared distance between each observation and its closest centroid. the lower the inertia is, the better the algorithm is. however, this approach suffers from a limitation: as the number of clusters increases, the inertia always becomes lower. the silhouette score is a better measure to determine the number of clusters. it measures the extent of closeness between each observation in a cluster and the observations in the neighboring clusters, providing a way to assess the number of clusters. it varies between -1 and 1. a score close to 1 indicates that the observation is far from the neighboring clusters. a score close to 0 denotes that the observation is very close to the decision boundary between two neighboring clusters, and a negative score indicates that those observations might have been assigned to a wrong cluster. this study uses the silhouette score for the analysis. 35 advances in technology innovation, vol. 7, no. 1, 2022, pp. 30-40 a k-means clustering algorithm is used to delineate the real estate market, and three distance metrics are utilized. first, the euclidean distance is applied for the five input variables, and categorical variables are converted to binary variables using the one-hot encoding approach. second, the gower distance, which is capable of dealing with both continuous and categorical variables, is used. third, the euclidean distance is applied again, although this time the categorical variables are converted to the continuous ones using entity embedding vectors. the continuous input variables (xand ycoordinates) are standardized before calculating the distance metrics. the k-means algorithm is implemented primarily following the clustering algorithm described in the work of kaufman et al. [41]. fig. 3 shows the respective silhouette score for k-means clustering based on the following three distance metrics: the euclidean distance, the gower distance, and the euclidean distance empowered by embedding vectors (hereafter denoted as euclidean distance 1, gower distance, and euclidean distance 2, respectively). for the categorical variables, the one-hot encoding approach is used in euclidean distance 1, and the entity embedding approach is employed in euclidean distance 2. after reviewing the visual depictions in the figure, three, four, and three clusters are chosen for each algorithm. as shown in panels (a) and (b) of fig. 3, there are points with higher scores than those of the chosen points; however, the model sparsity is also considered when choosing the optimal number of clusters. (a) euclidean distance 1 (b) gower distance (c) euclidean distance 2 fig. 3 silhouette score for each clustering algorithm 4. results 4.1. results and evaluation fig. 4 shows the market delineation results from the three k-means algorithms based on the euclidean distance 1, gower distance, and euclidean distance 2, respectively. each k-means algorithm produces three, four, and three submarkets, respectively, as shown in the panels (a), (b), and (c) in the figure. the three submarkets delineated by the k-means algorithm based on euclidean distances 1 and 2 appear to correspond approximately to northern, southern, and eastern parts of the study area. the four submarkets identified by the k-means algorithm utilizing gower distance correspond approximately to northern, middle, southern, and eastern parts of the study area. (a) euclidean distance 1 (b) gower distance (c) euclidean distance 2 fig. 4 market delineation results 36 advances in technology innovation, vol. 7, no. 1, 2022, pp. 30-40 the clustering results generally need to be checked using external validity measures, which can vary depending on the dataset and application areas, as explained earlier. this study uses the predictive accuracy of price estimation for the external validity measure. conversely, the delineation results are evaluated on the basis of the accuracy of the predicted price for each submarket. this approach is often used to compare the resultant submarkets delineated by clustering algorithms [42]. the price is estimated using the well-established hedonic pricing model, as follows [43-45]: i i i i i i i price zip zone year shape bearing area= + + + + + (3) where pricei denotes the sales price per square meter of lot i, zipi and zonei denote the areas to which lot i belongs, and yeari denotes the year in which lot i is sold. shapei of lot i has two levels: regular and irregular shape. bearingi of lot i comprises four levels: east, west, south, and north. finally, areai represents the size of lot i measured in square meters. then, the root-mean-square error (rmse) criterion is used for comparing the predictive accuracy, as follows [46]: ( )2 1 1 ˆ n i rmse y y n = = −∑ (4) where � denotes the price predicted by the pricing model, and y denotes the observed price. table 3 presents the predictive accuracy (rmse) for each submarket. prices in which rmse is measured are standardized to have a mean of zero and a standard deviation of one. although the marginal difference in average rmse exists between euclidean distance 1 and gower distance, the average rmse for the submarkets created by euclidean distance 2 is significantly reduced compared with the former two results. the average rmse in euclidean distance 2 is reduced by 10% approximately, from 0.076-0.077 (euclidean distance 1 and gower distance) to 0.069. this decrease could be attributed to the capability of embedding vectors used in euclidean distance 2 to extract the intrinsic relationships between levels in a categorical variable. conversely, by identifying the meaningful patterns inherent in categorical data and representing the patterns in the form of numerical vectors, the entity embedding approach can enhance the relevance of delineated submarkets. table 3 predictive accuracy for each submarket constructed by euclidean distance 1, gower distance, and euclidean distance 2, respectively rmse euclidean distance 1 gower distance euclidean distance 2 submarket 1 0.075 0.074 0.071 submarket 2 0.104 0.088 0.090 submarket 3 0.049 0.097 0.047 submarket 4 0.048 average 0.076 0.077 0.069 4.2. interpreting the resultant submarkets the average rmse for the submarkets constructed by euclidean distance 2 is the lowest. the submarkets are illustrated in fig. 5, which shows the subway line and arterial road. the clustering algorithm agglomerated the individual sales lots in gwacheon-si into three submarkets: northern, southern, and eastern submarkets. the northern submarket, which has old single-family houses, is a typical residential area with a well-established urban infrastructure. the southern submarket can be characterized by the landscapes mixed with public facilities and apartments; the public facilities include gwacheon city hall and the old central government complex. the eastern submarket mainly comprises low hills, small mountains, and sparsely located houses. in fig. 5, the area indicated by the dotted circle is classified as belonging to the southern submarket by euclidean distance 1, but is reclassified as belonging to the northern submarket when euclidean distance 2 is applied. it is the area located close to the subway line and appears to be difficult to delineate in a confident manner. consultations from the local experts, such as real 37 advances in technology innovation, vol. 7, no. 1, 2022, pp. 30-40 estate brokers and property appraisers, also confirm the ambiguity of the submarket membership of this area. some experts classify the dotted area as belonging to the northern submarket in gwacheon-si, while others consider that the area should belong to the southern submarket. although the topic of the submarket membership of the area is controversial even among domain experts, the data-driven clustering by euclidean distance 2 labels the area as belonging to the northern submarket in gwacheon-si, and the validity is proved by the lowest rmse obtained from a hedonic pricing model. fig. 5 submarkets constructed by euclidean distance 2 5. conclusions a clustering algorithm becomes inefficient or unstable when categorical variables are dominant in the input dataset, and the situation worsens in the case of high-cardinality categorical variables. this study attempted to enhance the performance of a clustering algorithm by employing an entity embedding approach to high-cardinality variables. gwacheon-si was chosen for the analysis, and a clustering algorithm was applied to the sales records of lots to delineate the real estate market. embedding vectors for the zip code and zone were learned from neural network training and subsequently applied to the clustering algorithm. the results showed that the submarket delineation created by the euclidean distance-based clustering algorithm equipped with embedding vectors outperformed the ones achieved by baseline models, such as the ordinary euclidean distance-based algorithm and the gower distance-based algorithm. this study offered an efficient alternative to handle these categorical variables when applying clustering algorithms: the clustering analysis results can be improved by using embedding vectors learned from neural network training, which has been utilized universally in machine learning applications that handle unstructured data. this study might promote the rapid adoption of machine-learning tools in the field of structured data. the dataset used in the study comprised 1,111 lots, but the data size may be insufficient to draw a generalizable conclusion. different results would possibly be observed if the proposed approach was applied to different datasets or different local areas. empirical experiments at a more extensive scale need to be attempted in future studies. conflicts of interest the author declares no conflict of interest. references [1] v. goyal, g. singh, o. tiwari, s. punia, and m. kumar, “intelligent skin cancer detection mobile application using convolution neural network,” journal of advanced research in dynamical and control systems, vol. 11, no. 7, pp. 253-259, 2019. 38 advances in technology innovation, vol. 7, no. 1, 2022, pp. 30-40 [2] a. aggarwal, m. alshehri, m. kumar, p. sharma, o. alfarraj, and v. deep, “principal component analysis, hidden markov model, and artificial neural network inspired techniques to recognize faces,” concurrency and computation: practice and experience, vol. 33, no. 9, e6157, may 2021. [3] m. alshehri, m. kumar, a. bhardwaj, s. mishra, and j. gyani, “deep learning based approach to classify saline particles in sea water,” water, vol. 13, no. 9, 1251, 2021. [4] a. aggarwal, a. rani, p. sharma, m. kumar, a. shankar, and m. alazab, “prediction of landsliding using univariate forecasting models,” internet technology letters, in press. [5] i. h. witten and e. frank, data mining: practical machine learning tools and techniques, cambridge: morgan kaufmann, 2016. [6] a. c. goodman and t. g. thibodeau, “housing market segmentation and hedonic prediction accuracy,” journal of housing economics, vol. 12, no. 3, pp. 181-201, september 2003. [7] j. c. gower, “a general coefficient of similarity and some of its properties,” biometrics, vol. 27, no. 4, pp. 857-871, december 1971. [8] l. r. dice, “measures of the amount of ecologic association between species,” ecology, vol. 26, no. 3, pp. 297-302, july 1945. [9] p. legendre and l. legendre, numerical ecology, burlington: elsevier science, 2012. [10] z. huang, “extensions to the k-means algorithm for clustering large data sets with categorical values,” data mining and knowledge discovery, vol. 2, no. 3, pp. 283-304, 1998. [11] s. s. khan and a. ahmad, “cluster center initialization algorithm for k-modes clustering,” expert systems with applications, vol. 40, no. 18, pp. 7444-7456, december 2013. [12] n. sharma and n. gaud, “k-modes clustering algorithm for categorical data,” international journal of computer applications, vol. 127, no. 17, pp. 1-6, october 2015. [13] c. guo and f. berkhahn, “entity embeddings of categorical variables,” https://arxiv.org/pdf/1604.06737.pdf, april 22, 2016. [14] v. efthymiou, o. hassanzadeh, m. rodriguez-muro, and v. christophides, “matching web tables with knowledge base entities: from entity lookups to entity embeddings,” international semantic web conference, pp. 260-277, october 2017. [15] j. pennington, r. socher, and c. d. manning, “glove: global vectors for word representation,” conference on empirical methods in natural language processing, pp. 1532-1543, october 2014. [16] o. abdelwahab and a. elmaghraby, “uofl at semeval-2016 task 4: multi domain word2vec for twitter sentiment classification,” 10th international workshop on semantic evaluation, pp. 164-170, june 2016. [17] z. chen, y. huang, y. liang, y. wang, x. fu, and k. fu, “rglove: an improved approach of global vectors for distributional entity relation representation,” algorithms, vol. 10, no. 2, 42, 2017. [18] m. aydoğan and a. karci, “turkish text classification with machine learning and transfer learning,” international artificial intelligence and data processing symp., pp. 1-6, september 2019. [19] j. xie, r. girshick, and a. farhadi, “unsupervised deep embedding for clustering analysis,” international conference on machine learning, pp. 478-487, june 2016. [20] x. guo, l. gao, x. liu, and j. yin, “improved deep embedded clustering with local structure preservation,” 26th international joint conference on artificial intelligence, pp. 1753-1759, august 2017. [21] c. wu and r. sharma, “housing submarket classification: the role of spatial contiguity,” applied geography, vol. 32, no. 2, pp. 746-756, march 2012. [22] b. keskin and c. watkins, “defining spatial housing submarkets: exploring the case for expert delineated boundaries,” urban studies, vol. 54, no. 6, pp. 1446-1462, 2017. [23] s. openshaw, “a geographical solution to scale and aggregation problems in region-building, partitioning and spatial modelling,” transactions of the institute of british geographers, vol. 2, no. 4, pp. 459-472, 1977. [24] d. p. claessens, s. boonstra, and h. hofmeyer, “spatial zoning for better structural topology design and performance,” advanced engineering informatics, vol. 46, 101162, october 2020. [25] r. m. assunção, m. c. neves, g. câmara, and c. da costa freitas, “efficient regionalization techniques for socio‐ economic geographical units using minimum spanning trees,” international journal of geographical information science, vol. 20, no. 7, pp. 797-811, 2006. [26] w. lin and y. li, “parallel regional segmentation method of high-resolution remote sensing image based on minimum spanning tree,” remote sensing, vol. 12, no. 5, 783, 2020. 39 advances in technology innovation, vol. 7, no. 1, 2022, pp. 30-40 [27] z. cai, j. wang, and k. he, “adaptive density-based spatial clustering for massive data analysis,” ieee access, vol. 8, pp. 23346-23358, 2020. [28] n. jabeur, a. u. h. yasar, e. shakshuki, and h. haddad, “toward a bio-inspired adaptive spatial clustering approach for iot applications,” future generation computer systems, vol. 107, pp. 736-744, june 2020. [29] w. m. rand, “objective criteria for the evaluation of clustering methods,” journal of the american statistical association, vol. 66, no. 336, pp. 846-850, december 1971. [30] p. j. rousseeuw, “silhouettes: a graphical aid to the interpretation and validation of cluster analysis,” journal of computational and applied mathematics, vol. 20, pp. 53-65, november 1987. [31] s. eldridge, d. ashby, c. bennett, m. wakelin, and g. feder, “internal and external validity of cluster randomised trials: systematic review of recent trials,” british medical journal, vol. 336, 876, april 2008. [32] m. rezaei and p. fränti, “set matching measures for external cluster validity,” ieee transactions on knowledge and data engineering, vol. 28, no. 8, pp. 2173-2186, august 2016. [33] x. li, w. liang, x. zhang, s. qing, and p. c. chang, “a cluster validity evaluation method for dynamically determining the near-optimal number of clusters,” soft computing, vol. 24, no. 12, pp. 9227-9241, 2020. [34] s. s. kumar, s. t. ahmed, p. vigneshwaran, h. sandeep, and h. m. singh, “two phase cluster validation approach towards measuring cluster quality in unstructured and structured numerical datasets,” journal of ambient intelligence and humanized computing, vol. 12, no. 7, pp. 7581-7594, 2021. [35] c. a. lipscomb and m. c. farmer, “household diversity and market segmentation within a single neighborhood,” the annals of regional science, vol. 39, no. 4, pp. 791-810, december 2005. [36] y. tu, h. sun, and s. m. yu, “spatial autocorrelations and urban housing market segmentation,” the journal of real estate finance and economics, vol. 34, no. 3, pp. 385-406, 2007. [37] z. liu, j. cao, r. xie, j. yang, and q. wang, “modeling submarket effect for real estate hedonic valuation: a probabilistic approach,” ieee transactions on knowledge and data engineering, vol. 33, no. 7, pp. 2943-2955, july 2021. [38] kostat, “statistics korea: population and households,” http://kostat.go.kr/portal/eng/pressreleases/8/1/index.board, 2020. [39] a. koul, s. ganju, and m. kasam, practical deep learning for cloud, mobile, and edge: real-world ai and computer-vision projects using python, keras, and tensorflow, sebastopol: o’reilly media, 2019. [40] a. struyf, m. hubert, and p. rousseeuw, “clustering in an object-oriented environment,” journal of statistical software, vol. 1, no. 4, pp. 1-30, february 1997. [41] l. kaufman and p. j. rousseeuw, finding groups in data: an introduction to cluster analysis, hoboken: john wiley & sons, 2009. [42] s. c. bourassa, f. hamelink, m. hoesli, and b. d. macgregor, “defining housing submarkets,” journal of housing economics, vol. 8, no. 2, pp. 160-183, june 1999. [43] s. rosen, “hedonic prices and implicit markets: product differentiation in pure competition,” journal of political economy, vol. 82, no. 1, pp. 34-55, january-february, 1974. [44] s. catma, “the price of coastal erosion and flood risk: a hedonic pricing approach,” oceans, vol. 2, no. 1, pp. 149-161, march 2021. [45] p. m. campos, j. s. thompson, and j. p. molina, “effect of irrigation water availability on the value of agricultural land in guanacaste, costa rica: a hedonic pricing approach,” e-agronegocios, vol. 7, no. 1, pp. 38-55, 2020. [46] d. wackerly, w. mendenhall, and r. l. scheaffer, mathematical statistics with applications, belmont: cengage learning, 2014. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). 40  advances in technology innovation, vol. 3, no. 4, 2018, pp. 166 178 dynamic biometric signature an effective alternative for electronic authentication vladimír smejkal 1,* , jindřich kodl 2 1 moravian university college olomouc, olomouc, czech republic. 2 authorized expert of cryptology and information systems security, prague, czech republic. received 07 june 2017; received in revised form 13 june 2017; accepted 23 december 2017 abstract the use of dynamic biometric methods for the authentication of people provides significantly greater security than the use of the static ones. the variance of individual dynamic properties of a person, which protects biometric methods against attacks, can be the weak point of these methods at the same time. this paper summarizes the results of a long-term research, which shows that a dbs demonstrates practically absolute resistance to forging and that the stability of signatures provided by test subjects in various situations is high. factors such as alcohol and stress have no influence on signature stability, either. the results of the experiments showed that the handwritten signature obtained through long practice and the consolidation of the dynamic stereotype, is so automated and stored so deep in the human brain, that its involuntary performance also allows other processes to take place in the cerebral cortex. the dynamic stereotype is composed of psychological, anatomical and motor characteristics of each person. it was also proven to be true that the use of different devices did not have a major impact on the stability of signatures, which is of importance in the case of a blanket deployment. the carried out experiments conclusively showed that the aspects that could have an impact on the stability of a signature did not manifest themselves in such a way that we could not trust these methods even used on commercially available devices. in the conclusion of the paper, the possible directions of research are suggested. keywords: biometric authentication methods, dynamic biometric signature, electronic signature, authentication 1. introduction one negative aspect of the current trend of connecting ever more devices in cyberspace (primarily thanks to the internet of things) is the increased risk of them being attacked and exploited. all devices and all information systems must, therefore, be constructed from the very start to be secure against possible risks. the key assumptions for constructing secure information systems are ensuring the quality identification and authentication of people, assets, and events in the system – on the one hand to prevent cyber attacks and other criminal activities in cyberspace, and on the other to record all incidents and their circumstances. this is an important tool for investigation, and also a source of feedback for improving security measures. the opinion that multifactor authentication is essential to ensure an adequate level of security for information systems is already practically universal [1]. it is also necessary to understand that in open systems we do not ensure proper data protection only through integrity and trust, but a sophisticated authentication process is a very important point in terms of the protection of * corresponding author. e-mail address: smejkal@znalci.cz. advances in technology innovation, vol. 3, no. 4, 2018, pp. 166 178 copyright © taeti 167 data, respectively the assets of the subject in question. it is only possible to launch the authorisation process which will provide the authenticated person with access to ict services after successful and reliable authorization. the used authentication methods should meet the requirements for variability both from the perspective of the technologies and used systems and from the perspective of the users themselves. the proposed solutions must also respect the legislation in force in different countries and permit the execution of legal acts according to the intentions of both sides. the authentication methods must provide a high level of protection against their breaking or exploitation, while at the same time remaining user-friendly. they must be secure, easy to use, unobtrusive and also have reasonable implementation costs. the addition of an extra factor for authentication also brings an increase in the technological, organization, and primarily financial demands of such a solution. hence, this field is still being developed, and the search for the “holy grail of authentication” continues. in particular, at a time when there is more and more emphasis on protection of personal data with concurrent pressure to both simplify and make the process of identification and authentication more pleasant for the user. for these reasons, it is important to focus on the issue of multifactor authentication, in particular, those where biometric methods play an important role. 2. biometric authentication methods biometry is a set of scientific findings, which focuse on the investigation and subsequent practical use of measurable characteristics of living organisms with the aim of unequivocally identifying them ( first identification) or verifying them ( then authentication) [2]. biometric authentication methods appear to be a reasonable compromise between demands on users and/or tools for authentication while not reducing the level of security. there are many biometric methods, but they can basically be split into three groups: (1) static (2) static with testing for the presence of a person (3) dynamic we must make a fundamental distinction between static and dynamic methods, where static methods are basically a continuation of the authentication principle that the user “has something”, even if this means something that is a part of their physiology. this means that there remains a risk of the falsification of biometric information (faking fingerprints, iris image, etc.) or their use through coercion. hence, for example, at the current time, the authentication methods used for mobile telephones can serve at most only to reassure users or for simple “superficial” protection. dynamic authentication methods have been arousing interest. we can assume a higher level of protection from abuse, as we are moving from the variant that the user “has something” to the variant in which the user “knows something” and, what is more, they “do not know what they know”. these are processes where the primary impulse arises in the central nervous system in the human brain with a predefined intensity and duration. the nervous system then activates the relevant muscles in a defined order, so that the user can perform a certain activity a signature, a certain gait, a gesture, and so on. in this case, therefore, biometry is based on the characteristics of a person’s behavior. the use of static methods connected with a test for the presence of a (living) person is something of an intermediate step. this is usually performed by detecting body temperature or, at a more sophisticated level, by detecting the circulation of blood in vessels or by measuring oxyhaemoglobin concentration. advances in technology innovation, vol. 3, no. 4, 2018, pp. 166 178 copyright © taeti 168 for each method, there are several characteristics that will determine its usefulness. these are, in particular: accuracy and reliability (expressed through parameters such as frr, far, fer, fir, fmr, fnmr [3]), security (protection against forgery and coercion, as well as against attacks on biometric images and samples), user acceptability, costs of acquisition and operation, etc. sometimes, these qualities can be directly in opposition a simpler, and more accurate method might be easier to attack (e.g. the faking of fingerprints). depending on the number of features used for recognition biometric systems, we can use a unimodal method (if only one of the biometric characteristics is used), or a multimodal one (if more than one characteristic is used). 3. dynamic biometric signature a dynamic biometric signature appears increasingly appropriate to authenticate people and also documents containing important information. it contains information about how the signature was created, and thus reflects characteristics of the signers, their habits and behavior, as well as the fact that they made the signature consciously. these characteristics represent a biometric footprint that is unique for each person and cannot be reproduced by a forger (unlike the actual image of the signature itself, which only makes up one of the parameters of the biometric footprint). these include the basic features of a handwritten signature: • the duration of the signature process, including the periods between strokes • points and curves in different parts of the signature • the pressure exerted by the pen on the pad during different parts of the signature • the overall size of the signature • the form and shape of the signature • the length and angle of lines, arcs and curves, the number of loops • the speed of individual stokes, acceleration and deceleration one of the important attributes of a dbs is that it contains not only the element that the writer is alive, but also the fact that the signature was created by the writer consciously. therefore, there is no need to develop additional mechanisms to test whether the subject is present, alive, or not unlike with static biometric methods (checking the print of a finger, palm, iris etc.). it is also legally beneficial, in that we can rely on the (theoretically rebuttable) assumption that people knew what they were signing [4]. a special tablet (pad) is currently used for scanning a dbs. when acquiring signature data, biometric data [mostly x(t), y(t), p(t), t)] are also acquired. this biometric information is also used to calculate additional parameters, defined using the iso/iec 19794-11 standard [5]. this standard contains a description of the mandatory parameters and formats of the biometric data. the system then scans the parameters of the signature, adds them to the information from the signed document, for example, the user name, current time and date, and about the device used. these data are encrypted and create the so-called biometric data that are sent for further processing. in this way, we acquire authentication data for the signer with subsequent securing against forgery. at the same time, the process simulates the standard process of a traditional signature. a dbs may be used as a handwritten signature, but it is also possible to use it in the form of a one-off password or graphical pattern that the user writes on the sensor or shows with a gesture for example using a mobile phone. a dbs would appear to be an effective alternative to the only technology used for electronic signatures to date a signature based on asymmetric cryptography. it can be used with an advantage in cases when advances in technology innovation, vol. 3, no. 4, 2018, pp. 166 178 copyright © taeti 169 the implementation of certificates (partly due to their limited validity period) and the secure “concealment” of private keys significantly interfere with the normal activities of signers. from the perspective of users, it is also the most pleasant method, as the act of signing is something we do today and every day without needing any special knowledge, skills or the ownership of any secret. handwritten signatures are also one of the most socially accepted biometric features. 4. risks involved with the use of a dynamic biometric signature recently, the static authentication tools provided by the makers of mass-used devices (mobile phones, laptops, etc.), such as fingerprint, palm or face scanners have been accepted in a sickly optimistic way. at the same time, the successful cases of attacking these devices are known. e.g. [18-21]. (new authentication tool face id on iphone x has also been subjected to attacks since it was released. they have been unsuccessful so far [22], but according to the authors, it is only a matter of time (and money). perhaps, the attacks failed because of the fact that face id is not a mere static method. greater safety of the new authentication method is based on the fact that face id revolutionizes authentication on iphone x, using a state-of-the-art truedepth camera system made up of a dot projector, infrared camera, and flood illuminator, and is powered by a11 bionic to accurately map and recognize a face. these advanced depth-sensing technologies work together to securely unlock iphone, enable apple pay, gain access to secure apps, and many more new features. face id projects more than 30,000 invisible ir dots. the ir image and dot pattern are pushed through neural networks to create a mathematical model of your face and send the data to the secure enclave to confirm a match while adapting to physical changes in appearance over time. all saved facial information is protected by the secure enclave to keep data extremely secure, while all of the processing is done on device and not in the cloud to protect user privacy. face id only unlocks iphone x when customers look at it, and it is designed to prevent spoofing by photos or masks [23]. an infrared camera, then, captures the distortion of that grid as the user rotates his or her head to map the face's 3-d shape a trick similar to the kind now used to capture actors' faces to morph them into animated and digitally enhanced characters. [24]. despite the high level of quality of face id that greatly eliminated the possibility of forgery (falsifying the sample), the weaknesses of static biometric methods remain the same. table 1 comparison of dynamic and static biometric authentication methods property static methods dynamic methods forgery (imitation) yes no stealing (reuse) yes no compulsion yes complicated possibility of revocation (substitution by a different sample) no partially impact of the internal state of an organism no no impact of the environment no no ageing very little has not been studied change as a result of illness or injury yes yes inability to use (for objective reasons) improbable in the case of a serious disorder, e.g. inability to write (agraphia or dysgraphia). critics of this method of identification, respectively authentication of people point to the lower credibility of a dbs arising from insufficient uniqueness of generated signature samples in particular in connection with the possible change in the motor skills of people as they age or due to stress, the “rationalization” (simplification) of the signature due to the increasing skill of the writer after numerous repetitions, etc. the experiments described below show that this is not true. table 1 contains the comparison of selected properties of dynamic and static biometric authentication methods in terms of safety and protection against abuse. advances in technology innovation, vol. 3, no. 4, 2018, pp. 166 178 copyright © taeti 170 the second main counterargument is the need for high security for the database of signature samples, respectively biometric samples when signing. this is another separate issue [6]. 4.1. dbs credibility regarding the first objection, it is important to state that the dbs rests on the repeated performance of the same movement. it means a handwritten signature, which is considered to be a highly-qualified movement, during which some general laws apply, and where the latest research shows that some basic characteristics of a signature, respectively its characteristic parameters, can be determined and described mathematically:  the 2/3 pl (power law) states the known effect of the curvature of trajectory on the speed of movement, and the momentary angular velocity is proportionate to the momentary curvature with a 2/3 power relationship. a study focusing on the issue of handwriting, and its compliance with the 2/3 pl has confirmed the relationship between the trajectory of the curvature of the writing and the speed of movement [7, 17].  the principle of isochrony applies to the temporal characteristics of human movement and describes the invariance of the time to perform a movement in relation to its amplitude, meaning that the duration of a movement remains almost the same irrespective of its scope. this fundamental also applies for smaller parts of the assessed movement, which means that this part of the movement can be geometrically changed (for example enlarged), while the time to complete this sub-movement remains the same.  another easy to observe a feature of human movement is its smoothness. the minimum jerk model enables the assessment of trajectory creation, where the sections of a movement are examined in relation to the maximized smoothness. (jerk is a vector-based physical quantity characterizing movement, and describes the rate of change of acceleration.) the natural laws indicated above, which are discussed in studies for activities connected with writing, can also be naturally applied to the handwritten signature. studies have also shown that for samples of a signature or even of only its part, the time pattern of determined sections remains the same, irrespective of changes in geometry. hence, changes in movement caused by emotions or stress will not impact the overall timing of the movement (in our case the signature). thus, as soon as identification points are determined for a handwritten signature, the time relationship between them will always remain the same [8-11]. 4.2. dbs forgery when trying to forge a signature, the forger must contend with the speed profile of the original. even if the image is the same, the velocity profile will differ with a level of probability approaching one. it is important to note that a certain variability is actually permitted for the coordinates (x, y), while the time relationship between these through points must remain unchanged. small changes in the geometric shape of the signature (amplitude, scale) are not significant compared to the actual timing of the movement. an attempt at forging a signature must always be visible at the time level, and this especially if the forger only has access to the visual form of the signature [6]. as a part of the research performed by the authors in 2014, the characteristics of a dbs during an attempt at forging the signature of another person whose signature sample was available as an image were examined, inter alia. 102 university students of different ages took part in the test. the test subjects were supplied with a sample of the signature of a bank client, who had made the signature during a standard transaction. the forgers were given both the first and second names of the signer. in this phase of the test, the test subjects had the chance to practice the signature before imitating it and thus create as advances in technology innovation, vol. 3, no. 4, 2018, pp. 166 178 copyright © taeti 171 convincing copies as possible. the signature training was done on paper, and also on a signing pad (in this phase the result of the signing was not scanned). after the training, the test subjects were tasked with imitating the signature sample five times with the maximum possible accuracy. fig. 1 shows that the attempt at imitating the client’s signature according to the sample reveals diametrical differences in terms of the time taken to create the sample and the forgery, while at first glance they appear very similar: fig. 1 time: the original signature above, the forgery below it is clear at first glance how long it took to create the forged signature; without mentioning the detailed differences in the individual time sections. when analyzing the scope of the performed signature (according to the number of image points), we see the different progress of the individual characteristics – for example, speed in fig. 2: fig. 2 writing speed: the original signature above, the forgery below a handwriting expert was also involved in the experiments and was provided with the forgery attempts produced by the test subjects. the analysis focused on assessing the compliance between the signatures created by the test subjects and the bank client’s signature. the results confirmed the conclusions arising from the data acquired by the validation server. the results of the examined set of signatures showed that out of the 190 forgeries, not even a short-term rehearsal of the signature to be forged was successful [12]. advances in technology innovation, vol. 3, no. 4, 2018, pp. 166 178 copyright © taeti 172 the experiments showed that the biometric data acquired during the creation of the signature provide such a set of information that enables an unequivocal opinion during subsequent verification in the case of a dispute over the authenticity of a signature. it was shown that the manufacturer’s software can be used to process these data. the configuration of the data that influences the amounts of the calculated indicators cannot – with the exception of the acceptance level – be changed in the programs delivered by the manufacturer. the configuration of the compliance level that results in agreement/rejection/non-rejection of a validated dbs will have to be addressed in the validation server, where it is possible to set the acceptable level according to the determined policy of the dbs operator. one interesting result of the experiments is that it will only be necessary to use the results and conclusions of an analysis of biometric data performed by a handwriting expert in extreme cases of the verification of the authenticity of a signature, while it is usually enough to deploy the validation server. in such a case, the biometric data are an ideal source of information for the handwriting expert compared to a situation in which they can only use two short texts, namely signatures on paper, for a comparison [12]. 4.3. dbs stability for different people other research assessed the stability of a dbs across a heterogenous set of tested people. three hypotheses were formulated: i. when using a standard (routine) signature, there is a high degree of similarity among individual signatures. the signatures are stable and display a small degree of variability. ii. this also applies to special circumstances, such as drunkenness, stress or other influences. iii. for some people, even the falsified (intentionally changed by them) signatures display a high degree of stability. a total of 10 signatures were recorded on the form from each person within the framework of the testing of the group of heterogeneous people. people signed in the usual standard manner during the first six signatures (hereafter simply referred to as the “authentic signatures”). they were, then, called upon during the next four signatures to try to change the signatures in such a way so that they could object that they were not the authors of the signatures during any eventual verification of authenticity (hereafter simply referred to as the “inauthentic signatures” or the “falsified signatures”). this involves a fairly typical variant where the first 6 signatures are completely routine, while the signatures starting with the seventh are falsifications of the person’s own signature (the person was asked to sign the same name, but different) with certain variability. after testing the group of 24 people which included:  people aged 12 to 92.  people with a primary, secondary and tertiary education (university students and graduates, including postgraduate education).  people working both manually and mentally.  the following evaluation was performed. a total of 240 results were acquired when comparing the authentic signatures, 138 when comparing the inauthentic signatures and 444 when comparing the authentic and inauthentic signatures with each other. the first finding involved the fact that in most people the first signature undertaken on the sensor showed a high degree of variance of compliance in relation to the other signatures, which is a consequence of the necessity of getting used to the method of signing on the sensor. for this reason, the first signature of each person was always omitted from the evaluation which resulted in the evaluated set of data being more homogeneous (with a smaller variance and standard deviation). advances in technology innovation, vol. 3, no. 4, 2018, pp. 166 178 copyright © taeti 173 it was found that the degree of compliance (stability) for authentic signatures is substantially higher than for inauthentic signatures, and this with a level of significance of 0.01. 4.4. the influence of the consumption of alcohol on a dbs the testing of influences on the biometric signature’s dynamic was carried out on a group of 9 university educated people with an average age of 35.2 ± 14.2 (for 22 to 60). the measurement (signatures and filling in the tests) was carried out in a restaurant during its regular operations, i.e. in an environment with significant background noise. all 9 people carried out 2 x 10 signatures (before and after drinking alcohol). people who did not reach a breath alcohol concentration of at least 1‰ were eliminated from further processing. during the course of the experiment, which lasted 4 hours, 6 people reached an average breath alcohol level of 1.4 ± 0.2 ‰ (from 1.0 to 1.7‰). the measurements were taken using an alcoscan al9000l (korea) digital alcohol tester. people took the brickemkamp-zillmer variant d2 attention test at the beginning and the end of the measuring. the d2 selective attention test is a widely used test in psychological diagnostics. it enables the evaluation of the attention, performance (speed) and error rate under time pressure. it is appreciated for its stable results and the high mutual correlation of the individual evaluated parameters. we used the brickenkamp-zillmer d2 test, published in association with the hogrefe gottingen publishing house [13]. the speed of performance after the consumption of alcohol did not change in a demonstrable manner. on the other hand, the number of errors after the consumption of alcohol increased by more than twice (2.4x). the error rate with consistent results was statistically evaluated using a paired t-test and welch’s test comparing the characteristics before (b as before) and after (a as after) the consumption of alcohol and a significant difference was found in the error rate (ch%) and in the attention variability (vs) at a level of significance of 0.05. the degree of compliance of the signatures after the consumption of alcohol at a level of significance of 1% does not differ from the degree of compliance of the signatures before the consumption of alcohol, not even in the case of people who displayed a significant increase in errors during the attention test. it has been demonstrated that: i. when using the standard (routine) signature, there is a high degree of similarity between the individual signatures for a particular person. the signatures of individual persons are stable and they display only a small degree of variability. the signature is a stable tool for the identification and authentication of a person. ii. (a) in some people, even their falsified (intentionally changed by them) signatures display a high degree of stability. (b) in some people, these falsified signatures are not sufficiently distinguishable from the authentic signatures. iii. the influence of alcohol on the realization of routine (authentic) signatures has not been proven. even in cases where people displayed a high degree of stability in their inauthentic signatures, the dynamic parameters of the data which were received during the creation of the dbs provide sufficient room for their analysis by a trained person or a handwriting expert, who has a significant amount of information available of the sort which is simply unable to be acquired when comparing classic signatures on paper [14]. it is common for the person providing a signature to be exposed to stress, and one reason for this is the importance of the situation in which they are appending the signature. after all, stress and very often negative stress are the most common emotions in human life. for this reason, we were interested in whether and in what way stress influences the quality and constancy of a dbs. advances in technology innovation, vol. 3, no. 4, 2018, pp. 166 178 copyright © taeti 174 in our experiments, we used the extreme situations in which test subjects in survival courses (”x-tream” course) at the university of defence of the czech republic found themselves [15]. test subjects undertook a series of demanding tasks and gradually reached a level of stress and physical and mental exhaustion, hunger, sleep deprivation and therefore stress. we originally formulated two hypotheses: i. stress does not influence the stability of a signature. ii. there is a significant correlation between performances achieved in a d2 attention test and the stability of a dbs. the tests incorporated in our experiment included a brickenkamp-zillmer d2 attention test and provided ten signatures on a signotec tablet that reads the biometric characteristics of a signature. equipment and procedures were the same as before. [14] each test subject undertook the tests before the load commenced (at the beginning of the survival course), in the middle of the course and at the end – indicated as s(tart), m(iddle), f(inish). a total of 26 people took part. the results of our experiments showed that even though the physical and mental load on the test subjects rose, the stability of their dynamic biometric signatures was high or actually improved during the experiment. the variability of the signatures of most individual people did not differ significantly at individual stages. it was also again confirmed that the use of a 1 st signature as “practice”, not included in the results, reduces the variability of signatures among all the test subjects. hypothesis i., that stress has no influence on the stability of a signature, was confirmed, at a significance level of 0.01. hypothesis ii., that there is a significant correlation between performances achieved on a d2 attention test and the stability of a dbs, was not confirmed. it can be assumed that a signature, as a stereotype embedded in the central nervous system over the long-term, can be influenced by outside factors to a lesser extent than specific activity performed at a stage of stress. we consider the stability of a dbs to be confirmed to such an extent by the range of experiments which we conducted [4, 6, 12, 14] to be able to use this method to identify and authenticate people or the documents which they have signed with a high degree of reliability and verifiability. 4.5. stability of a dynamic biometric signature created on various devices within further experiments, we focused on examining whether the use of various devices for dbs scanning affects the stability of a dbs of the signer. in our experiments, we used all the available pads produced by signotec, which differ from each other in terms of their design, the size of the signature field, resolution, sampling rate, and even the scanning method used – a regular pen or a special pen using the ert (electromagnetic resonance technology). the purpose of the experiments was to show the possible change of the stability of a dbs of a signer depending on the scanning device. as the sample represented people of both sexes aged 20 to 65, the size of the heterogeneous sample used was statistically representative enough. the following hypotheses were formulated: i. the participants will cope with the difficulties connected with the changing circumstances of the signing depending on the technical design of the pad in a different way: h0 the stability of signatures of a particular person on each device does not significantly change (mean and variance of the degree of compliance of signatures for each device belonging to the same basic set). h1 there is a statistically significant difference in the means and variances of the degree of compliance of signatures of a particular person on individual devices. advances in technology innovation, vol. 3, no. 4, 2018, pp. 166 178 copyright © taeti 175 ii. the stability of signatures achieved on individual devices will statistically significantly differ: h0 the mean degree and variance of compliance of signatures for each device do not significantly change (mean and variance of the degree of compliance of signatures for each device belonging to the same basic set). h1 there is a statistically significant difference in the means and variances of the degree of compliance of signatures on individual devices. 4.5.1. testing of hypothesis i the result characterizing the technology as a whole, i.e. without differentiation of types of devices and signers (i.e. for all people on all devices) is as follows see table 2: table 2 summary results on the degree of compliance of signatures x [%] m2 σ [%] 79.330 173.290 13.164 the selective mean of the degree of compliance of signatures came under an accepted level of compliance of biometric signatures 60% only in case of two people. in order to test the homogeneity of variances of the degree of compliance of signatures of each participant on all devices, the bartlett's test was used [25]. the values in the b test ranged from 20.341 to 609.934, i.e. the p-value was between 0.000 and 0.005. for all participants, the null hypothesis was, therefore, rejected at the significance level 0.01 and thus at the significance level 0.05 (as the p-value was <0.01 for all participants). using the cochran-cox test [26], a pair of devices, where the hypothesis on the compliance of means of the degree of compliance was rejected at the significance level 0.01, was found for each participant. the simple sorting test (analysis of variance, anova) that would keep the probability of error of the first kind at the level 0.05 or 0.01 could not be used with regard to the results of the bartlett's test. we can conclude that the participants coped badly with various designs of devices. this is due to the fact that selective variances of the degree of compliance are significantly different for all participants on individual devices, and there are also differences in means of compliances. 4.5.2. testing of hypothesis ii the following values of selective means and unbiased estimates for variances of the degree of compliance of signatures were detected on the stated devices see table 3 below: table 3 selective means and unbiased estimates for variances of the degree of compliance the signotec device and scanning method x [%] s 2 alpha ert 80.342 113.019 delta ert 76.749 238.268 gamma ert 78.971 232.027 omeganew td 76.022 228.052 omegaold td 83.002 125.844 sigmalite wd 77.097 148.574 sigmanew td 85.233 139.194 sigmaold td 77.195 120.338 compliance of variances was verified by the bartlett's test (b = 13.597, k-1 = 7, α = 0.01 and 0.05, p-value = 0.059), so the hypothesis on the compliance of all variances was accepted at the significance level 0.01 and at the significance level 0.05. advances in technology innovation, vol. 3, no. 4, 2018, pp. 166 178 copyright © taeti 176 the simple sorting test (anova) [26] gave the following results f = 2.565, k = 7, n-k = 306, p-value = 0.014, α = 0.01 and 0.05, f0.01= 2.700 and f0.05= 2.039, where f1-α (k-1, n-k) is (1-α) quantile of the fisher-snedecor distribution for the significance level α, so compliance of all means was accepted at the significance level 0.01 and rejected at the significance level 0.05. the results of the scheffe's test of multiple comparisons [27] would enable to determine between which two above-mentioned means the statistically significant differences exist. the scheffe's test accepts the equality of all 28 pairs of means of the degree of compliance of signatures at the significance level 0.01 and 0.05 (28 is the number of possible options for all pairs of devices). when using different devices, there were found no differences between the mean values of the degree of compliance (x) and values of the variance of the degree of compliance (σ 2 ) that is at the significance level 0.05 for variances and at the significance level 0.01 for means. it can, therefore, be noted that, despite the technological differences between individual devices, the stability of signatures (indicated by variance) does not change when changing the device. also, the degree of compliance of signatures on all devices does not statistically significantly differ at the significance level 0.01. the experiment brought the following results: hypothesis i. – the participants will cope with the difficulties connected with the changing circumstances of the signing depending on the technical design of the pad in a different way: the null hypothesis h0 claiming that the stability of the signatures of a particular person on each device does not significantly change that is at the significance level 0.01 and thus at the significance level 0.05, was disproved. a pair of devices, where the hypothesis on the compliance of means of the degree of compliance was rejected at the significance level 0.01, was found for each participant. therefore, the hypothesis h1 claiming that there is a statistically significant difference in the means and variances of the degree of compliance of signatures of a particular person on individual devices was confirmed. hypothesis ii. – the stability of signatures achieved on individual devices will statistically significantly differ: the null hypothesis h0 claiming that the mean degree and variance of compliance of signatures for each device do not significantly change, because there were found no differences between the values of means and variances of the degree of compliance when using different devices, that is at the significance level 0.05 for variances and at the significance level 0.01 for means, was confirmed. a detailed description of the experiment can be found in the report from [28]. 5. conclusions dbs has been an important alternative to classic electronic signatures based on cryptographic methods. it can be used with an advantage in cases when the implementation of certificates (partly due to their limited validity period) and the secure “concealment” of private keys significantly interferes with the normal activities of signers. from the perspective of users, it is also the most pleasant method, as the act of signing is something we do today and every day without needing any special knowledge, skills or the ownership of any secret. handwritten signatures are also one of the most socially accepted biometric features[16]. through the implemented tests, the authors of this paper have shown that the level of compliance, stability, and imperviousness to influence during the creation of signatures performed by the test subjects in different situations is high. a dbs shows practically absolute resistance to imitation, and the stability of signatures made by the test subjects in different advances in technology innovation, vol. 3, no. 4, 2018, pp. 166 178 copyright © taeti 177 situations is high. factors such as alcohol and stress have no influence on signature stability either. the results of the experiments have shown that a handwritten signature acquired through long practice and the compaction of the dynamic stereotype, which is composed of the physiological, psychological, anatomical and motor characteristics of each person, is automatic to such an extent and stored deep in the human brain, that its involuntary performance enables other processes to take place concurrently in the cortex. a dbs is not a replacement for a cryptographic electronic signature, but an important alternative that can be used in cases when the use of certificates, the secure storage and “policing” of private keys, etc. would significantly impact routine and stable processes, and potentially form a barrier discouraging normal users (contracting parties). its advantage over a cryptographic electronic signature, when a computer de facto does the signing, is the existence of this “handwritten” quality, which is only ensured through a legal declaration when a private key is used. in further research, the authors want to focus on other possible impacts occurring as a result of the external factors that can influence the process of signing, e.g. a change in the body position. they will also take into account the additional authentication options and abilities to imitate a dbs, this time not as a static image of the signature, but in a process of monitoring and imitating its dynamics. references [1] v. smejkal and j. kodl, “development trends of electronic authentication,” proc. of the 42nd annual conf. 2008 ieee international carnahan conf. on security technology, ieee press, october 2008, pp. 1-6. [2] r. rak, v. matyáš, and z. říha, “biometry and identity in forensic and commercial applications,” grada publishing, 2008. [3] m. gamassi, m. lazzaroni, m. misino, v. piuri, d. sana, and f. scotti, “accuracy and performance of biometric systems,” proc. of the 21st ieee instrumentation and measurement technology conf. (ieee cat. no. 04ch37510), ieee press, november 2004, pp. 510-515. [4] v. smejkal and j. kodl, “strong authentication using dynamic biometric signature,” proc. of 45th annual 2011 ieee international carnahan conf. on security technology (iccst), ieee press, october 2011, pp. 340-344. [5] iso/iec 19794-11:information technology. biometric data interchange formats, part 11: signature/sign processed dynamic data, ieee standard, 2013. [6] v. smejkal, j. kodl, and j. jr. kodl, “implementing trustworthy dynamic biometric signature according to the electronic signature regulations,” proc. of 47th annual 2013 ieee international carnahan conf. on security technology (iccst), ieee press, october 2014, pp. 165-170. [7] f. lacquaniti, c. terzuolo, and p. viviani, “the law relating the kinematic and figural aspects of drawing movements,” vol. 54, no. 1-3, pp. 115-130, october 1983. [8] f. lacquaniti, “central representations of human limb movement as revealed by studies of drawing and handwriting,” vol. 12, no. 8, pp. 287-291, 1989. [9] s. kandel, j. p. orliaguet, and p. viviani, “perceptual anticipation in handwriting: the role of implicit motor competence,” perception & psychophysics, vol. 62, no. 4, pp. 706-716, january 2000. [10] a. j. thomassen and h. l. teulings, “time, size and shape in handwriting: exploring spatio-temporal relationships at different levels,” time, mind, and behavior, springer berlin heidelberg, pp. 253-263, 1985. [11] y. wada and m. kawato, “a theory for cursive handwriting based on the minimization principle,” biological cybernetics, vol. 73, no. 1, pp. 3-13, june 1995. [12] v. smejkal and j. kodl, “assessment of the authenticity of dynamic biometric signature,” the results of experiments. proc. of 48th annual 2014 ieee international carnahan conf. on security technology (iccst), ieee press, october 2014, pp. 45-49. [13] r. brickenkamp and e. zillmer, d2-test of attention, seattle: hogrefe & huber, 1998. [14] v. smejkal, j. kodl, l. sieger, d. novák, and j. schneider, “the dynamic biometric signature. is the biometric data in the created signature constant?” in proc. of 49th annual 2015 ieee international carnahan conf. on security technology (iccst), ieee press, september 2015, pp. 385-390. advances in technology innovation, vol. 3, no. 4, 2018, pp. 166 178 copyright © taeti 178 [15] e. ambrozová, j. koleňák, d. ullrich, v. okorný, and d. cibulka, “x-tream index and multi-parameter personality dimension for managers cognition and decision-making in the modern corporate environment,” global journal for research analysis, vol. 4, no. 3, pp. 1-7, 2015. [16] r. tolosana, r. vera-rodriguez, j. ortega-garcia, and j. fierrez, “increasing the robustness of biometric templates for dynamic signature biometric systems,” in proc. of 49th annual ieee international carnahan conf. on security technology (iccst), ieee press, september 2015, pp. 229-234. [17] j. kodl jr., “mechanisms of human arm motion planning in the presence of multiple solutions,” imperial college, 2010. [18] a. rattani and a. ross, “automatic adaptation of fingerprint liveness detector to new spoof materials,” 2014 ieee international joint conf. on biometrics (ijcb), ieee press, december 2014, pp. 1-8. [19] a. al-ajlan, “survey on fingerprint liveness detection,” 2013 international workshop on biometrics and forensics (iwbf), lisbon, ieee press, april 2013, pp. 1-5. [20] l. ghiani, d. yambay, v. mura, s. tocco, g. l. marcialis, f. roli, and s. schuckers, “livdet 2013 fingerprint liveness detection competition 2013,” 2013 international conf. on biometrics (icb), ieee press, september 2013, pp. 1-6. [21] m. drahanský, o. kanich, and e. březinová, “challenges for fingerprint recognition-spoofing, skin diseases and environmental effects,” in handbook of biometrics for forensic science, advances in computer vision and pattern recognition, 2017. [22] a. greenberg, “we tried really hard to beat face id-and failed (so far),” https://www.wired.com/story/tried-to-beat-face-i d-and-failed-so-far/. [23] the future is here: iphone x. apple inc. press release, september 2017. [24] a. greenberg, “how secure is the iphone x's faceid? here's what we know,” https://www.wired.com/story/iphone-x-facei d-security/. [25] g. w. snedecor and w. g. cochran, “statistical methods. eighth edition,” iowa state university press, 1989. [26] w. g. cochran and g. m. cox, experimental & designs, 2nd ed. new york: john wiley and sons, 1957. [27] h. scheffé, the analysis & variance, new york: john wiley and sons, 1999. [28] v. smejkal, j. kodl, l. sieger, f. hortai, and p. tesař, “stability of a dynamic biometric signature created on various devices,” ieee press, december 2017. 3___aiti#9227___258-269 advances in technology innovation, vol. 7, no. 4, 2022, pp. 258-269 effects of data standardization on hyperparameter optimization with the grid search algorithm based on deep learning: a case study of electric load forecasting tran thanh ngoc, le van dai * , lam binh minh faculty of electrical engineering technology, industrial university of ho chi minh city, ho chi minh city, vietnam received 04 january 2022; received in revised form 24 march 2022; accepted 25 march 2022 doi: https://doi.org/10.46604/aiti.2022.9227 abstract this study investigates data standardization methods based on the grid search (gs) algorithm for energy load forecasting, including zero-mean, min-max, max, decimal, sigmoid, softmax, median, and robust, to determine the hyperparameters of deep learning (dl) models. the considered dl models are the convolutional neural network (cnn) and long short-term memory network (lstmn). the procedure is made over (i) setting the configuration for cnn and lstmn, (ii) establishing the hyperparameter values of cnn and lstmn models based on epoch, batch, optimizer, dropout, filters, and kernel, (iii) using eight data standardization methods to standardize the input data, and (iv) using the gs algorithm to search the optimal hyperparameters based on the mean absolute error (mae) and mean absolute percent error (mape) indexes. the effectiveness of the proposed method is verified on the power load data of the australian state of queensland and vietnamese ho chi minh city. the simulation results show that the proposed data standardization methods are appropriate, except for the zero-mean and min-max methods. keywords: deep learning, grid search, data standardization method, hyperparameter, electric load forecasting 1. introduction according to the united states energy information administration (us-eia), worldwide energy demand is expected to rise by 50%, with rising countries in asia leading the way. this rising demand would place considerable strain on the current energy infrastructure and jeopardize global environmental health by increasing greenhouse gas emissions from conventional power sources [1]. in the united states and europe, an estimated 40% of electricity consumption and 38% of co2 emissions come from the construction industry [2]. currently, the construction industry tends to use sustainable energy sources to replace limited energy sources. as a result, the use of renewable energy sources has been increasing, the design of buildings must be improved, and the building energy demand needs to be forecasted. therefore, it is necessary to apply the energy load forecasting method because it has both economic and infrastructure advantages. this method can predict future electricity consumption and help power companies make economically viable plans and decisions [3]. short-term load forecasting plays an important role in the power industry, including power system planning, power generation planning, and the power supply-demand balance [4-5]. if load forecasting is accurate, significant cost reductions in control operations and decision-making, such as dispatch, unit commitment, fuel allocation, power system security assessment, and off-line analysis, would be realized. on the contrary, if there is an error in the forecast of electricity demand, there will be an increase in operating costs. * corresponding author. e-mail address: levandai@iuh.edu.vn advances in technology innovation, vol. 7, no. 4, 2022, pp. 258-269 in the past decade, many methodologies and techniques have been proposed to solve the problem of short-term power load forecasting. they can be classified into two groups of methods. the first group relates to using statistical methods, such as multiple regression, exponential smoothing, and autoregressive integrated moving average (arima) [6-7]. the second group employs artificial intelligence techniques, such as support vector machines (svm) and artificial neural networks (anns) [8-9]. recent development based on anns and deep learning (dl) networks is one of the methods applied to solve the problem of load forecasting. the dl architecture includes different models, such as long short-term memory networks (lstmn), convolutional neural networks (cnn), deep belief networks, and deep boltzmann machine networks. among them, lstmn and cnn are popular in the problem of power load forecasting [10-11]. the main feature of the dl model is that the accuracy of the load-forecasted results highly depends on its hyperparameters. therefore, determining these hyperparameters for dl models is important [12-13]. recently, some algorithms, such as grid search (gs), random search (rs), and genetic algorithm (ga), have been applied to determine the hyperparameters of the dl model, among which the gs algorithm was widely applied [14-15]. in addition, the characteristics of the input data are also important factors affecting the accuracy of the dl model. moolayil [16] introduced some data standardization methods to solve this problem. however, raschka et al. [17] and yang et al. [18] did not show any interest in employing data normalization on the gs algorithm. as a result, this may be a leading disadvantage for these studies. to overcome this disadvantage, this study proposes an input data standardization method on the gs algorithm to determine the hyperparameters of the dl model, including cnn and lstmn, for energy load forecasting. for this proposed method, the model data are split into training and testing sets. for the training step, the gs algorithm is performed to determine the hyperparameters of the dl model corresponding to each data normalization method. for the testing step, the predicted errors of these optimal models are compared, and thereby the proposed methodology can evaluate the impact of the data normalization methods on the gs algorithm to the dl model. the error value in the dl model is usually determined based on the error evaluation indexes of the actual value and predicted value of the model, such as the mean absolute error (mae) and the mean absolute percent error (mape). the forecasting results of the dl model are significantly affected by the scale and size of the data. therefore, it is necessary to standardize the data during training and forecasting for the dl model. in this study, the methods, such as zero-mean, min-max, max, decimal, sigmoid, softmax, median, and robust, are proposed to standardize the input data of the dl model, and the hyperparameter values of cnn and lstmn models are established based on epoch, batch, optimizer, dropout, filters, and kernel. the novelty and contributions of this study include the following aspects: (i) introduce a data standardization method on the gs algorithm to determine the optimal hyperparameters of the dl model, including cnn and lstmn for the energy load forecasting; (ii) consider the error evaluation indexes of mae and mape of actual and predicted values for determining the optimal hyperparameters of the dl model through the epoch, batch, optimizer, dropout, filters, and kernel; (iii) conclude that the zero-mean and min-max are two of the data standardization methods and not the best methods on the gs algorithm for determining the optimal hyperparameters of the dl model. this study consists of five sections. section 1 presents the urgency, settlement, and unresolved issues of the load forecasting problem. section 2 describes the principle, hyperparameters, gs algorithm, and data normalization method for the dl model. the experimental procedures and settings are presented in section 3. section 4 presents the results and discussion. finally, the conclusions and future research aspects are presented in section 5. 2. methodology 2.1. deep learning structures artificial intelligence is the ability of a machine to imitate intelligent human behavior. machine learning (ml) is part of artificial intelligence that allows a system to learn and automatically improve from experience. dl is an application of ml that 259 advances in technology innovation, vol. 7, no. 4, 2022, pp. 258-269 uses complex algorithms and deep neural nets to train a model. dl models are built using several algorithms, such as cnns, lstmns, recurrent neural networks (rnns), generative adversarial networks (gans), radial basis function networks (rbfns), and so on [19]. in this study, two widespread dl networks, lstmn and cnn, are used to resolve the problem. the procedure is performed as follows. 2.1.1. lstmn network the difference between the rnn and feed-forward neural network (ffnn) is that the rnn is a model that can create a correlation between the previous information and the current state. a simple rnn structure is shown in fig. 1, in which the output signal is determined based on a linear transformation and nonlinear activation. the output signal can be calculated under the tangent function as follows [20]: 1 tanh( ( , ) ) t t t h w h x b−= + (1) where ht-1 denotes the (t-1)th output signal, xt denotes the t th input signal, and b denotes the bias. the lstmn is a modified rnn model developed by song et al. [21]. the difference between the lstmn and the rnn is that the lstmn can process long-term dependencies. fig. 2 describes the lstmn structure. each block has two parallel lines going in and out, representing the cell state and hidden state information. the general structure of lstmn has four layers of neural networks composed of three inputs (i.e., ct-1, ht-1, and xt) and two outputs (i.e., ct and ht). therefore, this lstmn structure can be described by the following equations [21]: the authors identify the information from the previous cell state ct-1 that should be removed by the following forget gate ft. 1 ( ))( , t f t t f f w h x b−= × +σ (2) the authors identify the input signal xt that should be stored in the cell state ct in the input gate, in which the input information and the candidacy cell state ��� should be updated by: 1 ( ( , ) ) t i t t i i w h x b−= × +σ (3) 1 tanh( ( , ) ) t c t t c c w h x b−= × +ɶ (4) the previous cell state ct is updated by combining ct-1 and ���: 1t t t t t c f c i c−= × + × ɶ (5) the outcome ht in the output gate is confirmed based on the output information ot and ct: 1 ( ))( , t o t t o o w h x b−= × +σ (6) tanh( ) t t t h o c= × (7) in which w is the input weight and f, i, and o represent the forget, input, and output gates, respectively. a a ht-1 ht ht+1 xt-1 xt xt+1 tanh fig. 1 simple rnn architecture 260 advances in technology innovation, vol. 7, no. 4, 2022, pp. 258-269 ht ct-1 ht-1 xt σ tanh ct ht ft it ot t cɶ σ σ tanh fig. 2 lstmn architecture 2.1.2. cnn network cnn is a feedforward neural network with a structure similar to that of human neurons. fig. 3 depicts the cnn structure developed based on the convolutional cnn structure introduced in the work of bon et al. [22], which includes convolution, pooling, and fully linked layers. the input data are convolved using many filters for the convolution layer, and a feature map is formed when a bias term is added. then, a nonlinear function is applied. the pooling layer’s primary goal is to lower the resolution of the feature maps to aggregate the input data. there are several sorts of pooling procedures, the most prevalent of which is the max-pooling strategy. finally, fully connected layers process the convolutional layers’ outputs [21]. ... ... input convolution layer pooling layer fully connected layer output fig. 3 the architecture of the cnn model 2.2. hyperparameters in general, the accuracy of a dl model depends on its hyperparameters. therefore, determining the hyperparameters for dl plays an extremely important role. a hyperparameter is a configuration that is external to the model, whose value cannot be estimated from data and all values are set before the started network training. in this study, the hyperparameter values of lstm and cnn are established, as listed in table 1. epoch refers to the number of times to expose the model to the whole training dataset; batch refers to the number of samples within an epoch after which the weights are updated; dropout refers to the process of randomly omitting a fraction of the hidden neurons. for each training case, each hidden neuron is randomly omitted from the network with a fixed probability p, where p can be chosen in the range [0,1]. optimizer refers to the optimization algorithm which plays an important role in improving the accuracy of the dl network. the optimizer is a mathematical algorithm that uses derivatives, partial derivatives, and the chain rule in calculus to understand how much change the network will see in the loss function by making a small change in the weight of the neurons. filter is one of the most important cnn hyperparameters, which is the number of filters that will be learned by the convolutional layer. in a cnn, a convolution filter iterates over all of the input components, executing convolution operations to extract input characteristics; 32, 64, 128, and so on are the most frequent number of filters. the convolution window width and height are determined by kernel size. the kernel size might be an odd integer, such as (3, 3), (5, 5), (7, 7), and so on. there have been many proposed methods to determine the dl hyperparameters in recent years, such as the gs, rs, gradient-based optimization, bayesian optimization (bo), ga, and particle swarm optimization (pso). of these methods, gs is widely employed due to its simplicity and efficiency. therefore, gs is the choice to determine the dl hyperparameters in this study. 261 advances in technology innovation, vol. 7, no. 4, 2022, pp. 258-269 table 1 graph representations lstmn model cnn model epoch epoch batch batch optimizer optimizer dropout filter kernel 2.3. grid search method the gs is a comprehensive search process through the predefined subclass of the value’s combinatory of the model’s hyperparameters. the operation principle of gs is illustrated in fig. 4. it is composed of two hyperparameters, x and y [23-24]. the x is established by three values {x1, x2, x3}, and the y is established by three values {y1, y2, y3}. as a result, their combination is nine value pairs. the gs will perform a search for the optimal model based on these values, and the optimal hyperparameter corresponds to the dl model with the smallest error. x y 3 y 2 y 1 y 2 x 1 x 3x fig. 4 the operation principle of the gs method the error value in the dl model is usually determined based on the error evaluation indexes of the actual and predicted values of the model, such as mean square error (mse), mae, and mape. these evaluation indexes can be described as follows [25]: 2 1 1 ˆmse n i i i y y n = = −∑ (8) 1 1 ˆmae n i i i y y n = = −∑ (9) 1 ˆ1 mape 100 n i i i i y y n y= − = ×∑ (10) where �� is the actual value i th , and ˆ i y is the predicted value i th . 2.4. data normalization many studies have shown that the forecasting results of the dl model are significantly affected by the scale and size of the data [26]. thus, it is necessary to standardize the data during training and forecasting for the dl model. in this study, the methods zero-mean, min-max, max, decimal, sigmoid, softmax, median, and robust are proposed to standardize the input data of the dl model. the mathematical models of these methods can be described as follows [16]: mean std zero-mean normalization : x x x x ′ − = (11) min max min min-max normalization: x x x x x − = − ′ (12) 262 advances in technology innovation, vol. 7, no. 4, 2022, pp. 258-269 max max normalization: x x x =′ (13) decimal normalization: 10 j x x′ = (14) min1 sigmoid normalization: , 1 a std x x x a xe − ′ − = = + ∀ (15) min 1 softmax normalization: , 1 a a std x xe x a xe − − −− = =′ + ∀ (16) med median normalization: x x x =′ (17) med 75 25 robust normalization: , x x x iqr x x iqr − = = −′ ∀ (18) where � and �� are the original and standardized date value, respectively; xmean, xstd, xmin, xmax, and xmed are the mean, standard deviation, min, max, and median values of x, respectively; x25 and x75 are the 25 th quantile and the 75 th quantile values of x, respectively; j is the smallest integer that satisfies the condition of max|��| ≤ 1. 2.5. grid search method based on data normalization based on the theoretical base of the dl structure, hyperparameter, and the date standardization method, as presented in sections 2.1, 2.2, and 2.4, the proposed gs algorithm applied to standardize the data for the dl model is shown in fig. 5. the procedure is done in the six steps below. the procedure is applied to each data standardization method, as introduced in section 2.4, to determine the error value. this error value is then compared to evaluate the effect of these methods on the gs algorithm for the dl network. step 1: the original data is processed and the input-target pairs of ( ��� ��,��� �� ) and ( �����,����� ) are determined corresponding to the training and test processes, respectively. step 2: the training and test data are standardized by using the methods described in section 2.4 to determine (��� �� � ,��� �� � ) and (����� � ,����� � ). the next step is the training process. step 3: the gs algorithm is applied to determine the optimal hyperparameters of the dl model from the variable value combination of each hyperparameter cfg = {cfgi}, i= 1: n, in which n is the total number of combinations. the vector cfgi depends on the dl model. for the lstmn model, cfgi = {ei, bi, oi, di} is used, whereas cfgi ={ei, bi, oi, fi, ki} is used for the cnn model, in which e, b, o, d, f, and k notate the hyperparameters of epoch, batch, optimizer, dropout, filters, and kernel, respectively. for this step, to overcome overfitting during training, the cross-validation (cv) technique is applied, and the dl model is run repeatedly at the same time. then, the average value of the model is used to increase its reliability. the next steps are the test processes: step 4: the dl is used with the obtained hyperparameters in step 3 to predict the �������� � . step 5: �������� is calculated by using the value �������� � . step 6: the error value of the dl is determined based on the difference between �������� and ����� by using eqs. (8)-(10). 263 advances in technology innovation, vol. 7, no. 4, 2022, pp. 258-269 process data scaling data data train x train y testx testy mse ' train x ' train y using cnn and lstmn models (testing process) un-scaling data ' test x ' test y' predict y evaluate by using eqs. (8)-(10) predict y test y mae mape using cnn and lstmn models (training process) cfgop t cfg fig. 5 the gs methodology based on data normalization 3. simulation data setup 3.1. data the half-hourly load demand data of queensland state, australia, and the hourly load demand data of ho chi minh city, vietnam, are used to verify the effectiveness of the proposed method. the selected data are divided into two different cases, corresponding to the lstmn and the cnn. these datasets have different periods, and their statistical properties are shown in table 2. fig. 6 shows the ytrain value waveform in case 1 according to the data normalization methods. 6000 5000 (1)-none 1.0 -1.0 0.0 (2)-zero-mea n 0.75 0.25 0.50 (3)-min-max 0.9 0.7 0.8 (4)-max 0.6 0.5 (5)-decimal 0.75 0.25 0.50 (6)-sigm oid 0.5 -0.5 0.0 (7)-softmax 1.0 0.8 (8)-median 0.0 -1.0 (9)-robust m w p u p u p u p u p u p u p u p u 0 100 200 300 time (half hour) (2)-zero-mea n (7)-softmax time (half hour) 3000 2000 (1)-none m w -1.0 0.25 0.50 (3)-min-max p u 0.6 0.8 (4)-max p u 0.3 0.2 (5)-decimal p u 0.75 0.25 0.50 (6)-sigm oid p u 1.00 0.75 (8)-median p u 0.0 -0.5 (9)-robust p u 0 50 100 150 0.5 1.0 0.0 p u 0.5 -0.5 0.0 p u (a) queensland state (b) ho chi minh city fig. 6 the ytrain test value waveform in case 1 according to the data normalization methods 264 advances in technology innovation, vol. 7, no. 4, 2022, pp. 258-269 table 2 data characteristics description case 1: lstm model case 2: cnn model queensland state ho chi minh city queensland state ho chi minh city xtrain xtest xtrain xtest xtrain xtest xtrain xtest time (day) 05/10/14 05/23/14 05/24/14 05/30/14 25/11/18 22/12/18 12/23/18 12/29/18 03/29/14 05/23/14 05/24/14 05/30/14 10/28/14 12/22/14 12/23/18 12/29/18 size (672,48) (336, 48) (672, 24) (168, 24) (2688, 48) (336, 48) (1344, 24) (168, 24) min (mw) 4,304.46 4,404.48 1,347.70 1,873.90 4,279.21 4,404.48 1,347.70 1,873.90 mean (mw) 5,535.20 5,591.45 2,917.94 2,844.65 5,589.60 5,591.45 2,951.42 2,844.65 max (mw) 6,917.66 6,824.76 3,945.90 3,695.20 6,984.78 6,824.76 3,945.9 3,695.20 std (mw) 6,38.78 6,54.69 6,02.94 553.73 679.70 654.69 589.33 553.73 3.2. simulation value setup the values of the optimal hyperparameters for the dl model are listed in table 3. for the lstm model, the total number of the hyperparameter combinations, represented by cfgi = {ei, bi, oi, di}, is 81. the set value for the cv cycle is 2 (i.e., the training dataset is divided into two subsets corresponding to two times of training and testing). for the cnn model, the total number of the hyperparameter combinations, written as cfgi = {ei, bi, oi, fi, ki}, is 243. the set value for the cv cycle is equal to 3 (i.e., the training dataset is divided into three subsets corresponding to three times of training and testing). the set value for the number of repetitions is two times for both lstm and cnn (i.e., each model is trained twice). the error measurement of the gs algorithm used in the training process is mae. table 3 the values of the optimal hyperparameters for the dl model hyperparameter lstm model cnn model epoch (e) 100, 300, 500 300, 500, 700 batch (b) 10, 30, 50 30, 50, 70 optimizer (o) adadelta, adam, adamax adagrad, adam, sgd dropout rate (d) 0.1, 0.3, 0.5 filter (f) 48, 80, 112 kernel (k) 3, 5, 7 number of combinations (cfg) 81 243 4. experimental results and analyses tables 4 and 5 show the experimental results produced while using lstm and cnn based on the gs algorithm during training for the data normalization scenarios, respectively. these tables illustrate that the dl model’s ideal hyperparameters have distinct values for each data normalization approach and for various queensland and ho chi minh city datasets. in addition, it shows the same values of the optimal hyperparameter set in some cases. for example, in the instance of queensland state data where the lstmn is applied, the max and median approaches yield the same values of the ideal hyperparameters as the normal method (original data). table 4 the obtained results of optimal hyperparameters when using lstmn method queensland state ho chi minh city epoch batch dropout optimizer epoch batch dropout optimizer normal 500 10 0.1 adam 500 10 0.3 adam zero-mean 500 10 0.1 adamax 500 30 0.1 adam min-max 500 10 0.1 adamax 500 10 0.1 adam max 500 10 0.1 adam 300 10 0.1 adam decimal 500 10 0.3 adam 500 10 0.3 adam sigmoid 500 10 0.1 adamax 500 10 0.1 adam softmax 500 10 0.1 adamax 500 10 0.1 adam median 500 10 0.1 adam 300 10 0.1 adam robust 500 50 0.1 adam 500 10 0.1 adam 265 advances in technology innovation, vol. 7, no. 4, 2022, pp. 258-269 table 5 the obtained results of optimal hyperparameters when using cnn method queensland state ho chi minh city epoch batch optimizer filter kernel epoch batch optimizer filter kernel normal 700 50 adam 112 7 700 30 adam 80 5 zero-mean 700 70 adam 112 3 500 50 adam 112 7 min-max 500 50 adam 112 3 700 50 adam 112 7 max 700 30 adam 112 3 700 30 adam 112 5 decimal 700 50 adam 112 7 700 70 adam 112 7 sigmoid 700 50 adam 80 7 700 30 adam 80 7 softmax 300 70 adam 80 7 700 50 adam 80 7 median 700 50 adam 80 5 700 30 adam 80 7 robust 500 70 adam 112 7 700 50 adam 80 5 table 6 the mae when using the lstmn method mae (mw) mape (%) training test training queensland ho chi minh city queensland ho chi minh city queensland ho chi minh city normal 546.17 534.95 567.59 504.53 10.68 20.07 standard 33.31 26.68 39.94 48.58 0.73 1.73 min-max 40.21 35.37 39.14 46.81 0.70 1.76 max 44.04 50.82 44.18 50.09 0.81 1.85 decimal 45.63 47.32 43.74 53.42 0.80 2.00 sigmoid 40.19 56.87 40.96 68.12 0.73 2.43 softmax 34.43 30.88 36.64 43.95 0.66 1.61 median 41.98 65.02 42.44 60.23 0.77 2.24 robust 34.16 28.62 37.24 39.31 0.67 1.42 table 7 the mae when using cnn method mae (mw) mape (%) training test training queensland ho chi minh city queensland ho chi minh city queensland ho chi minh city normal 52.66 38.95 52.94 46.97 0.94 1.71 standard 32.98 25.60 37.03 38.72 0.67 1.41 min-max 35.98 26.02 37.18 38.65 0.66 1.42 max 44.57 34.58 41.64 38.62 0.74 1.42 decimal 40.76 37.25 39.38 43.80 0.71 1.61 sigmoid 35.79 30.40 38.01 42.89 0.68 1.56 softmax 33.17 24.48 35.51 36.19 0.64 1.37 median 40.78 34.01 39.48 39.25 0.70 1.45 robust 26.21 23.63 33.56 36.03 0.60 1.32 table 6 presents the mae and the mape error of the training and test stages of the lstmn model. fig. 7 shows the boxplot chart of these maes and mapes corresponding to table 6. the obtained results show the effectiveness of the data normalization method for the gs algorithm in the lstm model. specifically, the mae is significantly reduced when a data normalization method is applied. for queensland data, the mae of the test process is 567.59 mw and the mape is 10.68% without using any data normalization technique, while they decrease to 44.18 mw and 0.81% at max, respectively, when the proposed data normalization methods are applied. similarly, table 7 and fig. 8 show the mae, the mape, and their boxplots of the training and test stages for the cnn model. again, the observed results show that applying data normalization methods significantly reduces both the mae and mpae. in other words, the performance of the gs algorithm is greatly improved with data normalization. moreover, the effectiveness of applying data normalization techniques can be divided into three groups. the first group of applying the softmax and robust methods yields small maes. the second group, which presents medium maes, includes the zero-mean and the min-max. the third group that provides medium maes consists of the max, decimal, sigmoid, and median methods. 266 advances in technology innovation, vol. 7, no. 4, 2022, pp. 258-269 500 400 300 200 100 0 1 2 3 4 5 6 7 8 9 queensland ho chi minh normalization m a e ( m w ) (a) mae of the training process 500 400 300 200 100 0 1 2 3 4 5 6 7 8 9 queensland ho chi minh normalization m a e ( m w ) 17.5 12.5 10.0 7.5 2.5 0 1 2 3 4 5 6 7 8 9 queensland ho chi minh normalization m a p e ( % ) 5.0 15.0 20.0 (b) mae of the test process (c) mape of the test process fig. 7 the boxplot of the maes and mapes when using lstmn 50 40 30 20 10 0 1 2 3 4 5 6 7 8 9 queensland ho chi minh normalization m a e ( m w ) (a) mae of the training process 50 40 30 20 10 0 1 2 3 4 5 6 7 8 9 queensland ho chi minh normalization m a e ( m w ) 1.6 1.2 1.0 0.6 0.2 0 1 2 3 4 5 6 7 8 9 queensland ho chi minh normalization m a p e ( % ) 0.4 0.8 1.4 (b) mae of the testing process (c) mape of the test process fig. 8 the boxplot of the maes and mapes when using cnn 267 advances in technology innovation, vol. 7, no. 4, 2022, pp. 258-269 5. conclusions this study presents an approach to examine the effect of data normalization methods on the gs algorithm for determining the optimal hyperparameters of the dl model, including the lstmn and cnn, for energy load forecasting. the power load data of the australian state of queensland and the vietnamese city of ho chi minh were used to verify the reliability of the proposed method. the error evaluation indexes of mae and mape of the actual and predicted values are established based on the epoch, batch, optimizer, dropout, filters, and kernel to determine the optimal hyperparameters of the dl model. the effectiveness of applying data normalization techniques can be divided into three groups. the first group of the applications of the softmax and robust methods yielded small maes. the second group, which presented medium maes, included the zero-mean and the min-max. the third group that provided medium maes consisted of the max, decimal, sigmoid, and median methods. the results showed that both mae and mape were much smaller when applying data normalization. in addition, out of the eight proposed data normalization methods, zero-mean or min-max was not the best method for the gs algorithm for determining the optimal hyperparameters of the dl model. conflicts of interest the authors declare no conflicts of interest. references [1] a. m. omer, “energy use and environmental impacts: a general review,” journal of renewable and sustainable energy, vol. 1, no. 5, article no. 053100, 2009. [2] k. amasyali, et al., “a review of data-driven building energy consumption prediction studies,” renewable and sustainable energy reviews, vol. 81, pp. 1192-1205, january 2018. [3] e. naml, et al., “artificial intelligence-based prediction models for energy performance of residential buildings,” recycling and reuse approach for better sustainability, pp. 141-149, 2019. [4] y. jiang, et al., “stochastic receding horizon control of active distribution networks with distributed renewables,” ieee transactions on power systems, vol. 34, no. 2, pp. 1325-1341, march 2019. [5] j. c. lópez, et al., “parsimonious short-term load forecasting for optimal operation planning of electrical distribution systems,” ieee transactions on power systems, vol. 34, no. 2, pp. 1427-1437, march 2019. [6] s. khan, et al., “forecasting day, week and month ahead electricity load consumption of a building using empirical mode decomposition and extreme learning machine,” 15th international wireless communications and mobile computing conference, pp. 1600-1605, june 2019. [7] t. t. ngoc, et al., “grid search of exponential smoothing method: a case study of ho chi minh city load demand,” indonesian journal of electrical engineering and computer science, vol. 19, no. 3, pp. 1121-1130, september 2020. [8] t. t. ngoc, et al., “grid search of multilayer perceptron based on the walk-forward validation methodology,” international journal of electrical and computer engineering, vol. 11, no. 2, pp. 1742-1751, april 2021. [9] k. krishnakumari, et al., “hyperparameter tuning in convolutional neural networks for domain adaptation in sentiment classification (htcnn-dasc),” soft computing, vol. 24, no. 5, pp. 3511-3527, march 2020. [10] y. yu, et al., “short-term load forecasting using deep belief network with empirical mode decomposition and local predictor,” ieee power and energy society general meeting, pp. 1-5, august 2018. [11] x. wang, et al., “lstm-based short-term load forecasting for building electricity consumption,” ieee 28th international symposium on industrial electronics, pp. 1418-1423, june 2019. [12] m. zahid, et al., “electricity price and load forecasting using enhanced convolutional neural network and enhanced support vector regression in smart grids,” electronics, vol. 8, no. 2, article no. 122, 2019. [13] n. m. aszemi, et al., “hyperparameter optimization in convolutional neural network using genetic algorithms,” international journal of advanced computer science and applications, vol. 10, no. 6, pp. 269-278, march 2019. [14] j. brownlee, deep learning for time series forecasting: predict the future with mlps, cnns, and lstms in python, new york: machine learning mastery, 2018. [15] s. mukhopadhyay, deep learning and neural networks, advanced data analytics using python, berkeley: apress, 2018. 268 advances in technology innovation, vol. 7, no. 4, 2022, pp. 258-269 [16] j. moolayil, learn keras for deep neural networks: a fast-track approach to modern deep learning with python, new york: springer, 2019. [17] s. raschka, et al., python machine learning: machine learning and deep learning with python, scikit-learn, and tensorflow 2, hoboken: wiley, 2019. [18] l. yang, et al., “on hyperparameter optimization of machine learning algorithms: theory and practice,” neurocomputing, vol. 415, pp. 295-316, july 2020. [19] s. motepe, et al., “power distribution networks load forecasting using deep belief networks: the south african case,” ieee jordan international joint conference on electrical engineering and information technology, pp. 507-512, april 2019. [20] x. zhang, et al., “deep neural network hyperparameter optimization with orthogonal array tuning,” international conference on neural information processing, pp. 287-295, december 2019. [21] x. song, et al., “time-series well performance prediction based on long short-term memory (lstm) neural network model,” journal of petroleum science and engineering, vol. 186, article no. 106682, march 2020. [22] n. n. bon, et al., “fault identification, classification, and location on transmission lines using combined machine learning methods,” international journal of engineering and technology innovation, vol. 12, no. 2, pp. 91-109, february 2022. [23] u. michelucci, applied deep learning: a case-based approach to understanding deep neural networks, new york: apress, 2018. [24] s. v. subramanian, et al., “deep-learning based time series forecasting of go-around incidents in the national airspace system,” aiaa modeling and simulation technologies conference, article no. 0424, january 2018. [25] t. t. ngoc, et al., “support vector regression based on grid search method of hyperparameters for load forecasting,” acta polytechnica hungarica, vol. 18, no. 2, pp. 143-158, january 2021. [26] a. s. girsang, et al., “stock price prediction using lstm and search economics optimization,” iaeng international journal of computer science, vol. 47, no. 4, pp. 758-764, november 2020. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). 269 3___aiti#7330_in press advances in technology innovation, vol. 6, no. 4, 2021, pp. 222-234 experimental and theoretical analysis of cracking moment of concrete beams reinforced with hybrid fiber reinforced polymer and steel rebars hiep dang vu 1 , duy nguyen phan 2,* 1 civil engineering faculty, hanoi architectural university, vietnam 2 faculty of civil engineering, mientrung university of civil engineering, vietnam received 18 march 2021; received in revised form 11 may 2021; accepted 12 may 2021 doi: https://doi.org/10.46604/aiti.2021.7330 abstract this study aims at experimentally and theoretically investigating the cracking moment (mcrc) of hybrid fiber reinforced polymer (frp)/steel reinforced concrete (rc) beams. six hybrid glass frp (gfrp)/steel and three gfrp rc beams with various gfrp and steel reinforcement ratios are tested in four-point bending scheme. experimental results indicate that both gfrp and steel rebars affect mcrc, but the effect of steel reinforcement is more significant. when the steel reinforcement ratio increases to 1.17%, mcrc goes up to 15.9%, while the same value for gfrp is only 9.7%. an analytical method is proposed based on the plain section assumption and nonlinear behavior of materials for estimating mcrc. the proposed model shows a good agreement with the experimental data conducted in this study and collected from the literature. the results of the parametric study give evidence of the positive effects of hybrid reinforcement ratios and elastic modulus of frp on mcrc of hybrid rc beams. keywords: concrete beam, fiber reinforced polymer (frp), hybrid reinforcement, cracking moment, flexural behavior 1. introduction due to some special characteristics, such as superior corrosion resistance, low strength/weight ratio, non-conductive, non-magnetic, fiber reinforced polymer (frp) rebars, are widely used as an alternative to steel rebar for concrete structures [1-2]. it is well known that there are four common types of frp rebars used to reinforce concrete components: aramid frp (afrp), basalt frp (bfrp), carbon frp (cfrp), and glass frp (gfrp). only the elastic modulus of cfrp is equivalent or higher than that of steel bar. the elastic modulus of other frp types is much lower than that of steel bar, which causes large deflection and crack width of frp reinforced concrete (rc) bending elements [3-6]. to overcome this drawback of frp rc beams, many researchers proposed combining traditional steel bars to frp bars in the tension zone of the concrete beam. as a result, hybrid frp/steel rc beams are formed [7-9]. in practice, the hybrid frp/steel rc beams were also met in a type of rc beams strengthened with frp sheets. although the flexural behavior of frp/steel rc beams has been extensively investigated, the data on cracking behavior is still limited. this study presents experimentally and theoretically the investigation on the cracking behavior of hybrid frp/steel rc beams, and introduces an analytical method for predicting the cracking moment of frp/steel rc beams. first, three groups of beams, which contain six hybrid gfrp/steel and three gfrp rc beams, are cast and tested in the four-point pending scheme to clarify the cracking behavior. second, an analytical method based on the plane section assumption and strain compatibility is * corresponding author. e-mail address: nguyenphanduy@muce.edu.vn tel.: +84-917688903 advances in technology innovation, vol. 6, no. 4, 2021, pp. 222-234 proposed for estimating the cracking moment. this method considers the nonlinear behavior of concrete and the contribution of reinforcements. finally, a parametric study on the effect of longitudinal reinforcement ratios and elastic modulus of frp on the cracking moment is done by using the proposed analytical method. 2. literature review previous studies of frp/steel rc beams mainly focused on the flexural behavior, evaluating the cracking moment, failure modes and load-bearing capacity, short-term and long-term deflections, crack width, crack spacing, stiffness, and ductility. the results of the previous studies revealed the positive effect of steel reinforcement in improving the flexural behavior of frp/steel rc beams [4-16]. for the design purposes of hybrid frp/steel rc beams, many researchers proposed analytical methods to estimate load-carrying capacity [10, 12-13, 17-19] and limits of reinforcement ratios [20], to determine failure modes [10-16], and to estimate crack width, crack spacing, and midspan deflection [21-22]. some researchers tried to use the existing design codes with some modifications for the calculation of hybrid frp/steel rc beams [4, 18, 23-24]. regarding the cracking moment of frp/steel rc beams, very few researchers focused on the cracking behavior and proposed the formulas to calculate the cracking moment of frp/steel rc beams. by using the existing design codes with the modulus of rupture and transformed uncracked sections, mohamed [25] compared the theoretical and experimental cracking moments of frp/steel rc beams, and reported that the theoretical cracking loads were significantly smaller than the experimental values. kartal and kalkan [26] developed two cracking moment estimates of frp/steel rc beams, one for the gross moment of inertia and the other for the uncracked transformed moment of inertia. in each method, three different tensile strengths of concrete, i.e., the experimental value calculated from the prismatic beam tests and the second and third values obtained from empirical flexural tensile strength of eurocode 2 and aci 318m, were applied. the results showed that the uncracked transformed moment of inertia method with a modulus of rupture expression according to the aci 318 m gave the best agreement between theoretical and experimental results. besides, the authors reported that ignoring the contribution of the longitudinal reinforcements in the calculation may lead to underestimation of the cracking moment. maleki and kheyroddin [18] used the recommendations of aci440.1r-15 and csa-s806-12 to estimate the first cracking moment of gfrp/steel rc beams. the results showed that these design codes underestimated the cracking moment of tested beams. valivonis and skuturna [27] experimentally investigated the cracking moment of rc beams strengthened by cfrp laminates. they reported that cfrp laminates significantly increased the critical tension strains of the concrete and cracking moment (from 56% to 106%), and that cfrp laminates also influence the expansion of cracks and restrict the development of the crack. these authors proposed an analytical equation for estimating the cracking moment by using curvilinear diagrams to describe the compressed concrete and the concrete in tension. however, the comparison results showed a large deviation between theoretical and experimental values. gao et al. [28] carried out an experimental study on the flexural behavior of one-way slab strengthened by frp sheets, and pointed out the influence of the amount of strengthening cfrp and gfrp sheets on cracking moment of rc one-way slab. in particular, the cfrp sheet with a high elastic modulus has a stronger effect on the cracking moment in comparison with a gfrp sheet. due to the complication of a method for estimating the cracking moment based on the plane cross-section assumption and equilibrium condition, these authors proposed an equivalent-conversion method for determining the cracking moment of rc slabs strengthened with frp, in which the formula of elastic materials is used with a plasticity coefficient. the proposed method revealed good agreement between experimental and theoretical values. a lot of experimental research on the flexural behavior of hybrid frp/steel rc beams has been reported along with many proposals for prediction models of cracking moments. however, the majority of these proposals are based on the existing design codes (aci and eurocode), which neglected the contribution of reinforcements. this fact led to the underestimation of the cracking moment of frp/steel rc beams. 223 advances in technology innovation, vol. 6, no. 4, 2021, pp. 223-234 3. experimental investigation six hybrid gfrp/steel rc and three gfrp rc beams with dimensions of 150 × 250 × 2700 mm (width × height × long) and different gfrp and steel reinforcement ratios are cast and loaded in a four-point bending scheme. all nine testing beams are divided into three groups to evaluate the influence of gfrp/steel reinforcement on the cracking load. in each group, the gfrp reinforcement area (af) is fixed, and the steel reinforcement area (as) varies (fig. 1 and table 1). specifically, the group of beams #1 (beams b1, b2, and b3) is reinforced with 2g10 (two gfrp bars with a diameter of 10 mm, af = 1.225 cm 2 ), the group of beams #2 (beams b4, b5, and b6) is reinforced with 2g14 (af = 2.65 cm 2 ), and the group of beams #3 (beams b7, b8, and b9) is reinforced with 3g14 (af = 3.97 cm 2 ). in addition, during the analysis of the test results, the groups of beams with fixed longitudinal steel reinforcement and varied gfrp reinforcement are also created. the group of beams with fixed 2s10 (two steel bars with the diameter of 10 mm) contains beams b2, b5 and b8, and the group of beams with fixed 2s14 (two steel bars with the diameter of 14 mm) contains beams b3, b6, and b9. the loading span is 2400 mm, of which the length of the pure bending zone is 400 mm (fig. 2(a)). as shown in fig. 1, the gfrp and steel rebars are arranged in two layers, wherein the gfrp rebars are placed in the undermost layer with a concrete cover thickness (cf) of 25 mm, while the steel rebars are arranged in the inner layer (second layer) with a concrete cover thickness (cs) of 50 mm. all the actual dimensions of the testing beams are re-measured after casting, as shown in table 1. the shear reinforcements from plain round steel bars with the diameter of 6 mm and the spacing of 100 mm are put in the shear spans to prevent shear failure. in the middle span, the spacing of stirrups is 200 mm to form a reinforcement cage and reduce the influence of stirrups on flexural behavior. two 6 mm diameter steel rebars are used as compressive reinforcements. the beam specimens are designed with references to aci 440.1r-15 [29] and recommendations by previous researchers [12, 20]. accordingly, the gfrp reinforcement ratio (��) varies from 0.35% to 1.18%, and the steel reinforcement ratio (��) ranges from 0.52% to 1.13%. details of geometries and reinforcements of specimens are presented in table 1. (a) gfrp rc beams (b) hybrid gfrp/steel rc beams (c) reinforcement cage fig. 1 beam’s reinforcement (unit: mm) table 1 details of beam specimens group of beams beam id b, mm h, mm ��, mm ��, mm h0f, mm h0s, mm af, cm 2 as, cm 2 ��, % ��, % rm, mpa #1 b1.2g10 150 253 21 232 1.23 0.35 37.2 b2.2g10-2s10 152 254 30 54 224 200 1.23 1.57 0.36 0.52 41.6 b3.2g10-2s14 151 252 27 72 225 180 1.23 3.08 0.36 1.13 39.5 #2 b4.2g14 148 252 31 221 2.65 0.81 42.6 b5.2g14-2s10 151 251 31 52 220 199 2.65 1.57 0.80 0.52 41.0 b6.2g14-2s14 152 254 33 62 221 192 2.65 3.08 0.79 1.06 40.5 #3 b7.3g14 148 255 26 229 3.97 1.17 45.5 b8.3g14-2s10 153 254 33 59 221 195 3.97 1.57 1.18 0.53 42.8 b9.3g14-2s14 155 255 23 58 232 197 3.97 3.08 1.11 1.01 42.9 *note: b, h, ��, ��, h0f, and h0s are the geometric dimensions of cross section as shown in fig. 1. �� = as /(bh0s) and ��= af /(bh0f) are the steel and gfrp reinforcement ratios, respectively. the beam id is formed from 3 parts: the first part (e.g., b1) is the order number of the beams, the second part (2g10, 2g14, and 3g14) refers to the number and the diameter of gfrp bars, and the last part (2s10 and 2s14) indicates the number and the diameter of steel bars. b refers to beam, g refers to gfrp, and s refers to steel. 200 ø6 af h 0 f 2 5 0 a f c f 2ø6 ( )asca sc c sc a c f c s h 0 f h 0 s ø6 a s a f f a s 2ø6 ( )asc c sc 224 advances in technology innovation, vol. 6, no. 4, 2021, pp. 222-234 (a) schematic diagram (b) test setup fig. 2 flexural test the concrete with the desired compressive strength of 40 mpa and the water-to-cement ratio of 0.55 is produced from ordinary portland cement. the actual average compressive strength (rm) is evaluated from compressive tests on 150 × 150 × 150 mm cubic specimens after 28 days of curing, as shown in table 2. the ribbed gfrp rebars with nominal diameters of 10 mm and 14 mm are used in this study and are manufactured by vietnam frp trading and production joint stock company. the gfrp bars are produced from continuous high-strength e-glass fiber and vinyl ester resin. the hot-rolled plain steel bar with a diameter of 6 mm is used for stirrups and compressive reinforcement, and the hot-rolled ribbed steel bars with diameters of 10 mm and 14 mm are used as longitudinal rebars. the mechanical properties of gfrp, round plain steel, and ribbed rebars according to tensile tests are shown in table 2. table 2 mechanical properties of reinforcements rebars yield strength, mpa tensile strength, mpa young’s modulus, gpa gfrp rf = 997 ef = 44.3 the 6 mm plain round steel σsy = 309 rs = 358 es = 200 the ribbed steel (diameter ≥ 10 mm) σsy = 412 rs = 577 es = 200 the beams are loaded after a 28-day curing period in a four-point bending scheme until failure (fig. 2). three 100 mm strain gauges (s1, s2, and s5) are placed at the top, side, and bottom surfaces at the midspan of the test beams to record the compressive and tensile strains in concrete. two 5 mm strain gauges (s3 and s4) are attached on the surface of tensile steel and gfrp rebars at midspan before casting and are protected by silicon to measure the strains in rebars during the test. a linear variable differential transformer (lvdt) is fixed at midspan to measure the deflection, and two digital indicators i1 and i2 are placed at supports to eliminate the displacements of holders. during the test, the loads on the beam are applied in a step-by-step procedure, and the values of load from loadcell, the data from lvdt, and strain gauges are automatically collected by the datalogger sts-wifi system. 4. experimental results and discussion the first cracking moment of tested beams is obtained from the load versus strain curve of the concrete and reinforcements. this value can also be determined from the load versus midspan deflection curve. fig. 3 and fig. 4 illustrate the load-strain and load-midspan deflection curves for a typical tested beam b2.2g10-2s10. it is worth to note that until the yield of steel (after the concrete cracks), the trends of the load versus strain curves of the concrete and reinforcements and the load versus midspan deflection curves of the remained tested beams are similar to those of the beam b2.2g10-2s10. as shown in fig. 3, there are leaps on the load-strain curves as the first crack appears. the leap is also noticed on the load-midspan curve at the moment when the first crack appears (fig. 4). the experimental cracking moments of tested beams (mcrc,e) are presented in table 3. the load-carrying capacities (mu,e) and failure modes of tested beams are also recorded and presented in table 3 to evaluate the ratio between cracking moment and load-carrying capacity. the relationship between the cracking moments and distribute beam loadcell hydraulic jack 150 1000 400 1000 150 2700 ø6@200 ø6@100 gfrp bars steel bars s1 s3 s4 s5 2 5 0 s2 i2i1 lvdt 225 advances in technology innovation, vol. 6, no. 4, 2021, pp. 223-234 the steel reinforcement ratios of the groups of beams with fixed gfrp reinforcement ratios are illustrated in fig. 5. the relationship between the cracking moments and the gfrp reinforcement ratios of the groups of beams with fixed steel reinforcement ratios are also presented in fig. 6. fig. 3 load versus strain curve of the concrete and reinforcements (beam b2.2g10-2s10) fig. 4 load versus midspan deflection curve (beam b2.2g10-2s10) table 3 the cracking moment of tested hybrid gfrp/steel rc beams beam id experimental theoretical εs,e ×10 4 εf,e ×10 4 εbt,e ×10 4 σs,e, kn σf,e, kn σs,e /σf,e σs,e /σsy, % σf,e /ffu, % mcrc,e, knm mu,e, knm mcrc,e / mu,e failure mode εs,t ×10 4 εf,t ×10 4 εbt,t ×10 4 mcrc,t, knm mcrc,t / mcrc,e b1.2g10 1.65 1.72 7.3 0.73 5.35 24.1 0.22 rg 1.23 1.67 5.73 1.07 b2.2g10-2s10 0.86 1.29 1.73 17.2 5.7 3.02 4.17 0.57 6.20 34.4 0.18 sy-rg 0.82 1.12 1.70 6.60 1.06 b3.2g10-2s14 0.83 1.12 1.40 16.6 5.0 3.32 4.03 0.5 6.20 38.1 0.16 sy-cc 0.57 1.15 1.74 6.30 1.02 b4.2g14 1.31 1.47 5.8 0.58 6.20 41.4 0.15 cc 1.11 1.65 6.30 1.02 b5.2g14-2s10 1.00 1.14 1.76 20 5.1 3.92 4.85 0.51 6.40 48.4 0.13 sy-cc 0.83 1.10 1.72 6.47 1.01 b6.2g14-2s14 0.75 0.99 1.62 15 4.4 3.41 3.64 0.44 6.50 56.5 0.12 sy-cc 0.70 1.08 1.76 6.76 1.04 b7.3g14 1.30 1.64 5.8 0.58 6.65 52.4 0.13 cc 1.18 1.65 6.85 1.03 b8.3g14-2s10 0.73 1.03 1.51 14.6 4.6 3.17 3.54 0.46 6.40 50.0 0.13 sy-cc 0.75 1.08 1.72 6.94 1.08 b9.3g14-2s14 0.93 1.39 1.86 18.6 6.2 3.00 4.51 0.62 6.80 56.7 0.12 sy-cc 0.76 1.21 1.77 7.40 1.09 *note: rg refers to the rupture of gfrp, sy refers to steel yielding, and cc refers to concrete crushing. fig. 5 cracking moment versus steel reinforcement ratio curves of the groups of beams with fixed gfrp reinforcement ratios fig. 6 cracking moment versus gfrp reinforcement ratio curves of the groups of beams with fixed steel reinforcement ratios -3 0 3 6 9 12 15 s1 s2 s3 s 5 0 20 40 60 80 strain ε×10 3 l o a d , k n first crack s4 5.0 5.5 6.0 6.5 7.0 0 0.4 0.8 1.2 c ra c k in g m o m e n t, k n m steel reinforcement ratio µ s , % group of beams #1 (b1, b2, b3) group of beams #2 (b4, b5, b6) group of beams #3 (b7, b8, b9) 6.00 6.25 6.50 6.75 7.00 0.3 0.6 0.9 1.2 c ra c k in g o m e n t m , k n m gfrp reinforcement ratio µ f , % grou s 2 5 8p of beams with fixed 2 10 (b , b , b ) beams with fixed 2 1 (b , b , b )group of s 4 3 6 9 226 advances in technology innovation, vol. 6, no. 4, 2021, pp. 222-234 it is well known that the cracking moment depends on beam dimensions, properties of materials, and reinforcement ratios [28, 30]. it can be seen in figs. 5-6 that the reinforcements significantly affect the cracking moment of hybrid gfrp/steel rc beams. as gfrp or steel reinforcement ratios increase, the cracking moment of tested beams linearly increases. as can be seen in fig. 5, in the groups of hybrid beams with fixed gfrp reinforcement ratio, the effect of steel reinforcement on cracking moment decreases with the increase of the gfrp reinforcement ratio. in the group of beams #1 (�� = 0.36%), when the steel reinforcement ratio increases from 0 to 1.13%, the cracking moment of hybrid gfrp/steel beams increases to 15.9%, while in the group of beams #2 (�� = 0.8%) and #3 (�� = 1.17%) the corresponding values are 4.8% and 2.3% respectively. when the steel reinforcement ratios are fixed, the cracking moment of hybrid gfrp/steel rc beams also increases with the increase of the gfrp reinforcement ratio. however, the effect of gfrp reinforcement on the cracking moment of the hybrid beam is less than that of steel rebars. in the groups of beams with fixed steel reinforcement (groups 2s10 and 2s14 in fig. 6), the cracking moments of hybrid beams increase to 8.8% and 9.7% respectively. for the tested beams, the experimental cracking moment to load-carrying capacity ratio (mcrc,e /mu,e) varies from 0.12 to 0.22 (table 3), and this ratio decrease with the increase of the gfrp or steel reinforcement ratio. (a) beams b1.2g10 (b) beams b2.2g10-2s10 (c) beams b3.2g10-2s14 (d) beams b4.2g14 (e) beams b5.2g14-2s10 (f) beams b6.2g14-2s14 (g) beams b7.3g14 (h) beams b8.3g14-2s10 (i) beams b9.3g14-2s14 fig. 7 distribution of strains on the cross-section at the fist cracking moment fig. 7 presents the distribution of strain on the cross-section of tested beams at the moment when the first crack appears. as can be seen, the plane cross-section assumption is satisfied for hybrid frp/steel rc beams. to clarify the contribution of each type of longitudinal reinforcements to the cracking moment, the maximum experimental tensile strains and stresses of gfrp (��,� and ��,�) and steel rebars (��,� and ��,�) before the appearance of the first crack are reported in table 3. the 0-0.5-1.0-1.5 0.5 1.0 1.5 2.0 0 50 100 150 200 250 h e ig h t, m m strain × 104 0-0.5-1.0-1.5 0.5 1.0 1.5 2.0 0 50 100 150 200 250 h e ig h t, m m strain × 10 4 0-0.5-1.0-1.5 0.5 1.0 1.5 2.0 0 50 100 150 200 250 h e ig h t, m m strain × 10 4 0-0.5-1.0-1.5 0.5 1.0 1.5 2.0 0 50 100 150 200 250 h e ig h t, m m strain × 10 4 0-0.5-1.0-1.5 0.5 1.0 1.5 2.0 0 50 100 150 200 250 h e ig h t, m m strain × 10 4 0-0.5-1.0-1.5 0.5 1.0 1.5 2.0 0 50 100 150 200 250 h e ig h t, m m strain × 10 4 0-0.5-1.0-1.5 0.5 1.0 1.5 2.0 0 50 100 150 200 250 h e ig h t, m m strain × 10 4 0-0.5-1.0-1.5 0.5 1.0 1.5 2.0 0 50 100 150 200 250 h e ig h t, m m strain × 10 4 0-0.5-1.0-1.5 0.5 1.0 1.5 2.0 0 50 100 150 200 250 h e ig h t, m m strain × 10 4 227 advances in technology innovation, vol. 6, no. 4, 2021, pp. 223-234 maximum tensile strain in concrete (� ,�) at this moment is also presented in table 3. due to the elastic behavior of reinforcements in this stage, the values of the tensile stresses in gfrp and steel rebars are determined by multiplying the measured tensile strains by the corresponding young’s moduli (44.3 gpa and 200 gpa, table 2). as can be seen in table 3, due to the high young’s modulus of steel rebars, at the first cracking moment, the stress in steel bars is about 3.02 to 3.92 times higher than that in gfrp rebars, and is about 3.00% to 4.85% of the yield strength. that fact has proven that both gfrp and steel reinforcement affects the cracking moment of hybrid gfrp/steel rc beams, but the contribution of gfrp reinforcement in this stage is negligible as compared to steel, and the stress in gfrp is about 0.44% to 0.73% of ultimate tensile strength. these results are consistent with the findings reported by most other researchers [27-28, 31] and are contrary to the results presented in [32]. in addition, experimental research results show that, as the first crack appears, the maximum tensile strain in the outermost concrete fiber (� ,�) reaches the value of 1.47×10 -4 to 1.86×10 -4 , which meets the recommended ultimate tensile strain of concrete � � = 1.5×10 -4 in sp 63.13330:2018 [33]. in addition, the recorded maximum compressive strains in the outermost concrete fiber (� ,�) before the appearance of the first crack vary from 0.91×10 -4 to 1.67×10 -4 , which are much less than the ultimate compressive strain of concrete. 5. an analytical method for calculating the cracking moment of hybrid frp/steel rc beam as reported in the introduction and as indicated from the experimental results, the hybrid gfrp/steel reinforcement significantly affects the cracking moment. therefore, using the recommendations of the existing design codes which neglect the contribution of reinforcements leads to the underestimation of the cracking moment. this section introduces an analytical method for calculating the cracking moment of frp/steel rc beams based on the plane cross-section assumption and equations of equilibrium. the bilinear stress-strain relationship of concrete introduced in sp 63.13330:2018 [33] is used in the calculation (fig. 8). at the first cracking moment, the maximum strain in the outermost tensile concrete fiber reaches the ultimate value � � = 1.5×10 -5 according to the stress-strain relationship in fig. 8. before the concrete cracks, assuming that the concrete in compression behaves elastically, the stress distributes in triangular form. in the tension zone, the stress distributes in trapezoid form with the maximum stress equal to the tensile strength rbt according to the stress-strain relationship in fig. 8. the distributions of strain and stress on the cross-section are presented in fig. 9. fig. 8 bilinear stress-strain relationship of concrete [33] (a) cross-section (b) distribution of strains (c) distribution of stress fig. 9 stress and strain distributions for calculating the cracking moment σ r σ 8×10 -5 15×10 -5 0.0015 arctge ε b2 ε b ε b1,red εbt1,redεbt2 εbt rb b bt bt b,red h hh b a a f s 0 f 0 s a a s f asc a sc ε b εs εf ε = 15×10 bt2 εsc -5 ε = 8 × 1 0 b t1 ,r e d -5 h hh 0 f 0 s a a s f a sc σb σsc σs σf rbt x h -x ( ) h -x 7 1 5 ( ) h -x 8 1 5 228 advances in technology innovation, vol. 6, no. 4, 2021, pp. 222-234 the maximum strains in the outermost compressive concrete fiber � , in the compressive steel rebars �� , in the tensile steel rebars ��, and in the tensile gfrp rebars �� are determined according to � � by using the plane cross-section assumption, i.e., according to the strain distribution shown in fig. 9(b). the stress in concrete and reinforcements can be found by using the obtained strains in materials. before the first crack appears, the stress in the outermost compressive fiber of concrete � is determined by eq. (1): 2 1, 10 bt b b b b bred b red x r rx e h x h x ε σ ε ε = = = − − (1) where rb is the prismatic strength of concrete; x is the compression zone height; � is the maximum strain in the outermost compressive concrete fiber, which can be determined according to the eq. (2) based on the plain cross section in fig. 9(b); ebred is the reduced modulus of concrete, which is determined by eq. (3) according to sp 63.13330 : 2018 [33]. ( )2 /b bt h xxε ε −= (2) 1, / bred b b red e r ε= (3) the stress in compressive reinforcement rebars is as follows: ( )2bt sc s sc sc s x a e e h x ε σ ε − = = − (4) where es is the modulus of elasticity of steel; �� is the strain in compressive steel rebars and is calculated according to fig. 9(b) by the following equation. ( ) ( )2 /sc bt scx a h xε ε= − − (5) the stress in tensile steel rebars is as follows: ( )2bt s s s s s h x a e e h x ε σ ε − − = = − (6) where �� is the strain in tensile steel reinforcement obtained from fig. 9(b) as follows: ( ) ( ) 2 / s bt s h x a h xε ε= − − − (7) the stress in tensile frp rebars is as follows: ( )2bt f f f f f h x a e e h x ε σ ε − − = = − (8) where �� is the strain in gfrp rebars and determined according to fig. 9(b) as follows: ( ) ( )2 /f bt fh x a h xε ε= − − − (9) the resultant forces of the compression zone of concrete (nb), of the compressive reinforcement (nsc), of the tensile steel (ns) and gfrp (nf) rebars, and in the tension zone of concrete (nbt) are calculated using the following equations: 229 advances in technology innovation, vol. 6, no. 4, 2021, pp. 223-234 2 2 20 b b b bx brx n h x σ = = − (10) ( )2bt s sc sc sc sc sc e x a a n a h x ε σ − = = − (11) ( )2bt s s s s s s e h x a a n a h x ε σ − − = = − (12) ( )2bt f f f f f f h x a e a n a h x ε σ − − = = − (13) ( ) ( ) 1 2 7 4 15 15 bt bt bt bt bt h x r b h x r b n n n − − = + = + (14) the equation of horizontal force equilibrium is as follows: ( ) ( ) ( ) ( )2 22 2 11 20 15 bt f f fbt s sc sc bt s s s btb h x a e ae x a a h x a e a h x r bbrx h x h x h x h x εε ε − −− − − − + = + + − − − − (15) by expanding eq. (15) and setting the constants a (16), b (17), and c (18), eq. (15) is re-written as eq. (19): 11 20 15 b bt r b r b a = − (16) ( )2 22 15 bt bt s s s sc f f r bh b e a e a e aε= + + + (17) ( ) 2 2 11 15 bt bt s s s s sc sc f f f s s f f r bh c e a a e a a e a a e a h e a hε= − + − − − (18) 2 0ax bx c+ + = (19) by solving eq. (19) and choosing the compatible root meeting the condition 0 < x < h, the equations for calculating the compression zone height is expressed as follows: 2 4 2 b ac b x a − − = (20) by using the compression zone height x (eq. (20)) and defining the resultant forces in materials (nb, nsc, ns, nf, and nbt) according to eq. (10) to eq. (14), the first cracking moment mcrc can be determined by taking the moment about the axis passing through the neutral axis (eq. (21)). the comparison results between the experimental cracking moments and the theoretical values (mcrc,t) of the tested beams obtained by eq. (21) in table 3 show good agreement. the deviation between the experimental and theoretical cracking moment is less than 9%. in addition, the theoretical strains in the outermost compressive concrete fiber, in the steel tensile reinforcement, and in the tensile gfrp reinforcement obtained by eq. (2), eq. (7), and eq. (9) respectively are also in good agreement with the experimental values (table 3). ( ) ( ) ( ) ( ) ( )1 223 162 3 30 45 bt btb crc sc sc s s f f n h x n h xxn m n x a n h x a n h x a − − = + − + − − + − − + + (21) 230 advances in technology innovation, vol. 6, no. 4, 2021, pp. 222-234 to verify the applicability of the proposed theory for determining the cracking moment of concrete beams reinforced with a combination of steel bars and different types of frp bars, the experimental data of 24 tested hybrid frp/steel rc beams in the literature are collected and compared as shown in table 4. the comparison results prove the accuracy in estimating the cracking moment of concrete beams reinforced with hybrid frp/steel rebars. the average value is 1.006, the standard deviation is 0.076, and the mean value is 0.995. table 4 the comparison between experimental and theoretical cracking moments of concrete beams reinforced with a combination of different types of frp and steel bars ref. beam id b, mm h, mm af, mm as, mm af, mm 2 as, mm 2 asc, mm 2 asc, mm es, gpa type of frp ef, gpa rm, mpa mcrc,e, knm mcrc,t, knm mcrc,t / mcrc,e [25] b10/8s 100 200 22 64 100.6 157 0 0 200 gfrp 39 30 2.5 2.39 0.96 b10/6 100 200 21 64 56.6 157 0 0 200 gfrp 41 30 2.5 2.37 0.95 b12/8 100 200 22 65 100.6 226 0 0 200 gfrp 39 30 2.7 2.42 0.90 b12/6s 100 200 21 65 56.6 226 0 0 200 gfrp 41 30 2.3 2.4 1.04 [23] gfrp-40s 150 250 42 98 253.4 142.6 142.6 210 200 gfrp 48 32.7 5.4 5.4 1.00 cfrp-40s 150 250 40 95 142.6 142.6 142.6 210 200 cfrp 103 32.7 5.7 5.4 0.95 gfrp-60s 150 250 42 98 253.4 142.6 142.6 210 200 gfrp 48 49.3 7.43 7.11 0.96 cfrp-60s 150 250 40 95 142.6 142.6 142.6 210 200 cfrp 103 49.3 6.3 7.11 1.13 [26] b2s3 200 300 30 30 118.3 339.3 1.57 270 200 bfrp 43 25 13.2 11.85 0.90 b3s2 200 300 30 30 177.5 226.2 1.57 270 200 bfrp 43 25 11.6 11.5 0.99 b4s1 200 300 30 30 236.7 113.1 1.57 270 200 bfrp 43 25 10.4 11.1 1.07 g2s3 200 300 30 30 259.8 339.3 1.57 270 200 gfrp 35 25 11.5 11.9 1.03 g3s2 200 300 30 30 389.7 226.2 1.57 270 200 gfrp 35 25 10.1 11.6 1.15 g4s1 200 300 30 30 519.6 113.1 1.57 270 200 gfrp 35 25 10.5 11.2 1.07 g1s5 200.8 301.9 30 30 117.5 565.5 0 0 200 gfrp 46 24.4 13.3 12.7 0.95 g2s4 199.8 301.1 30 30 234.9 452.4 0 0 200 gfrp 46 24.4 12.7 12.3 0.97 g3s3 200.6 304.4 30 30 352.4 339.3 0 0 200 gfrp 46 24.4 10.9 12.2 1.12 g4s2 198.6 304.6 30 30 469.9 226.2 0 0 200 gfrp 46 24.4 11 11.8 1.07 b1s2 199.8 308 30 30 59.2 226.2 0 0 200 bfrp 43 24.4 11.8 11.7 0.99 b1s4 200 300 30 30 59.2 452.4 1.57 270 200 bfrp 43 25 14 12.2 0.87 g1s4 200 300 30 30 129.9 452.4 1.57 270 200 gfrp 35 25 13.2 12.3 0.93 b2s1 199.2 301.7 30 30 118.3 113.1 0 0 200 bfrp 43 24.4 10.8 10.9 1.01 g1s2 198.6 304.9 30 30 117.5 226.2 0 0 200 gfrp 46 24.4 11.4 11.5 1.01 g2s1 202 301.6 30 30 234.9 113.1 0 0 200 gfrp 46 24.4 9.9 11.1 1.12 average standard deviation mean 1.006 0.076 0.995 6. parametric study as mentioned above, besides the section geometry and mechanical properties of concrete, the longitudinal reinforcements also significantly affect the cracking moment of hybrid frp/steel rc beams. in this section, a parametric study of the effects of the longitudinal frp and steel reinforcement ratios and young’s modulus of frp rebars on the cracking moment of hybrid frp/steel rc beams is carried out using the proposed formula (eq. (21)). first, the influence of hybrid frp/steel reinforcement ratios on the cracking moment are performed with the following input data: b×h = 300×600 mm; rb = 30 mpa; rbt = 2.5 mpa; eb = 32 gpa; es = 200 gpa; σsy = 400 mpa; ef = 50 gpa; rf = 1000 mpa. the distances from the centroid of the steel rebars and frp rebars to the outermost tensile concrete fiber are chosen equally: af = as = 40 mm. by considering the recommendations in [20], the steel and frp reinforcement ratios are selected to vary from 0% to a maximum 4% (i.e, the area of each type of reinforcements varies from 0 cm 2 to 67.2 cm 2 ). by conducting nonlinear regression analysis, the relationship between cracking moment calculated by eq. (21) and hybrid steel/frp reinforcement ratios is expressed by eq. (22). it should be noticed that, the adjusted coefficient of multiple determination (r 2 ) of regression model in eq. (21) is higher than 0.9999. 231 advances in technology innovation, vol. 6, no. 4, 2021, pp. 223-234 2 2 , 3.597 15.077 5.49 0.1567 crc crc b f s f s m m µ µ µ µ= + + + − (22) where mcrc,b is the cracking moment of plain concrete beam; µf and µs are in percent (%). the response surface which expresses the relationship among cracking moment, frp, and steel reinforcement ratios according to eq. (22) is shown in fig. 10. the results of the parametric study on the current section and properties of beam show that the cracking moment increases to 93.7% when the steel reinforcement ratios increase from 0% to 4%. meanwhile, the corresponding value for the case of the increasing frp reinforcement is only 24.3%. this outcome affirms the above conclusion of the lower contribution of frp reinforcement on the cracking moment of hybrid frp/steel rc beam in comparison with the steel reinforcement. in addition, as reported in previous studies, the ratio af /as significantly affects the flexural behavior of frp/steel rc beams. as found from the parametric study, this ratio also significantly affects the cracking moment of frp/steel rc beams. namely, with a constant total area of hybrid frp/steel reinforcement, the cracking moment decreases with an increase in the af /as ratio. the effect of young’s modulus of frp on the cracking moment is also investigated on the above-mentioned cross section and properties of materials. the cross section is reinforced with three groups of steel and frp reinforcement: the first group — as = 4 cm 2 and af = 4 cm 2 (minimum hybrid frp/steel reinforcement ratio [20]), the second group — as = 25 cm 2 and af = 25 cm 2 (compatible hybrid frp/steel reinforcement ratio [20]), and the third group — as = 50 cm 2 and af = 50 cm 2 (maximum hybrid frp/steel reinforcement ratio [20]). the modulus of elasticity of frp reinforcement varies from 50 gpa to 200 gpa. it can be seen in fig. 11 that the cracking moment of hybrid frp/steel rc beams is linearly proportional to young’s modulus of frp reinforcement. besides, the effect of young’s modulus of frp reinforcement on cracking moment increases with the increase of the frp reinforcement ratio. fig. 10 the response surface among cracking moment, frp, and steel reinforcement ratios fig. 11 the relationship between young’s modulus of frp reinforcement and cracking moment 7. conclusions in this study, the cracking moment of hybrid frp/steel rc beams was experimentally and theoretically investigated. based on the study results, the following conclusions may be drawn: µ , % µ , % 4.0 3.0 2.0 1.0 0.0 1.0 2.0 3.0 60.0 80.0 100.0 120.0 140.0 f m ,k n m cr c 0.0 s 60 85 110 135 160 50 100 150 200 e , gpa m ,k n m c rc f a a = = 4 cms f 2 a = a = 25 cm s f 2 a = a = 50 cm s f 2 232 advances in technology innovation, vol. 6, no. 4, 2021, pp. 222-234 (1) the cracking moment of hybrid frp/steel rc beams is linearly proportional with the steel and frp reinforcement ratios. (2) in the scope of the experimental study, the cracking moment to load-carrying capacity ratio of hybrid gfrp/steel rc beams varies from 12% to 22%, and this ratio reduces with increasing total hybrid reinforcement ratio. (3) both steel and frp rebars affect the cracking moment of hybrid frp/steel rc beams. however, due to higher modulus of elasticity of steel in comparison with the frp reinforcement, the contribution of steel rebars on the cracking moment of hybrid frp/steel rc beams are more significant in comparison with frp reinforcement. (4) the proposed analytical method, which is based on the nonlinear stress-strain relationship of concrete and considered the contribution of hybrid reinforcements, accurately estimates the cracking moment of hybrid frp/steel rc beams. (5) the modulus of elasticity of frp reinforcement also affects the cracking moment of frp/steel rc beams, the cracking moment increases with the increase of young’s modulus of frp reinforcement. it should be stressed that the above conclusions are based on the study results of concrete beams made from normal strength concrete reinforced with optimal hybrid frp/steel reinforcement ratio. the applicability of these conclusions to other types of concrete and very high hybrid reinforcement ratios is unknown. further studies are necessary. conflicts of interest the authors declare no conflict of interest. references [1] s. a. jabbar and s. b. h. farid, “replacement of steel rebars by gfrp rebars in the concrete structures,” karbala international journal of modern science, vol. 4, no. 2, pp. 216-227, june 2018. [2] c. jeetendra, k. suresh, and a. hussain, “application of frp in concrete structures,” international journal of engineering associates, vol. 4, no. 8, pp. 50-51, august 2015. [3] c. barris, l. torres, j. comas, and c. miàs, “cracking and deflections in gfrp rc beams: an experimental study,” composites part b: engineering, vol. 55, pp. 580-590, december 2013. [4] w. qu, x. zhang, and h. huang, “flexural behavior of concrete beams reinforced with hybrid (gfrp and steel) bars,” journal of composites for construction, vol. 13, no. 5, pp. 350-359, octorber 2009. [5] d. lau and h. pam, “experimental study of hybrid frp reinforced concrete beams,” engineering structures, vol. 32, no. 12, pp. 3857-3865, december 2010. [6] j. zhang, w. ge, h. dai, and y. tu, “study on the flexural capacity of concrete beam hybrid reinforced with frp bars and steel bars,” 5th international conference on frp composites in civil engineering, september 2010, pp. 304-307. [7] s. h. alsayed and a. m. alhozaimy, “ductility of concrete beams reinforced with frp bars and steel fibers,” journal of composite materials, vol. 33, no. 19, pp. 1792-1806, october 1999. [8] m. a. aiello and l. ombres, “structural performances of concrete beams with hybrid (fiber-reinforced polymer-steel) reinforcements,” journal of composites for construction, vol. 6, no. 2, pp. 133-140, january 2002. [9] k. h. tan, “behavior of hybrid frp-steel reinforced concrete beams,” 3rd international symposium on non-metallic (frp) reinforcement for concrete structures, october 1997, pp. 487-493. [10] p. d. nguyen, v. h. dang, and n. a. vu, “performance of concrete beams reinforced with various ratios of hybrid gfrp/steel bars,” civil engineering journal, vol. 6, no. 9, pp. 1652-1669, september 2020. [11] i. kara, a. ashour, and m. köroğlu, “flexural behavior of hybrid frp/steel reinforced concrete beams,” composite structures, vol. 129, pp. 111-121, october 2015. [12] l. pang, w. qu, p. zhu, and j. xu, “design propositions for hybrid frp-steel reinforced concrete beams,” journal of composites for construction, vol. 20, no. 4, 04015086, august 2016. [13] x. gu, y. dai, and j. jiang, “flexural behavior investigation of steel-gfrp hybrid-reinforced concrete beams based on experimental and numerical methods,” engineering structures, vol. 206, 110117, march 2020. [14] b. jia, s. liu, x. liu, and r. wang, “flexural capacity calculation of hybrid bar reinforced concrete beams,” materials research innovations, vol. 18, no. supp 2, pp. s2-836-s2-840, may 2014. 233 advances in technology innovation, vol. 6, no. 4, 2021, pp. 223-234 [15] j. zhang, w. ge, h. dai, and y. tu, “study on the flexural capacity of concrete beam hybrid reinforced with frp bars and steel bars,” 5th international conference on frp composites in civil engineering, september 2010, pp. 304-307. [16] y. yang, z. y. sun, g. wu, d. f. cao, and z. q. zhang, “flexural capacity and design of hybrid frp-steel-reinforced concrete beams,” advances in structural engineering, vol. 23, no. 7, pp. 1290-1304, may 2020. [17] n. duy phan and d. viet quoc, “limiting reinforcement ratios for hybrid gfrp/steel reinforced concrete beams,” international journal of engineering and technology innovation, vol. 11, no. 1, pp. 1-11, january 2021. [18] f. maleki and a. kheyroddin, “investigation of cracking moment in rc beams with gfrp bars,” vol. 2, pp. 37-47, january 2017. [19] x. ruan, c. lu, k. xu, g. xuan, and m. ni, “flexural behavior and serviceability of concrete beams hybrid-reinforced with gfrp bars and steel bars,” composite structures, vol. 235, 111772, march 2020. [20] g. xingyu, d. yiqing, and j. jiwang, “flexural behavior investigation of steel-gfrp hybrid-reinforced concrete beams based on experimental and numerical methods,” engineering structures, vol. 206, pp. 110-117, march 2020. [21] w. ge, j. zhang, d. cao, and y. tu, “flexural behaviors of hybrid concrete beams reinforced with bfrp bars and steel bars,” construction and building materials, vol. 87, pp. 28-37, july 2015. [22] h. alkhudery, o. albuthbahak, and h. alkatib, “investigation of effective hybrid frp and steel reinforcement ratio for concrete members,” journal of engineering and applied sciences, vol. 13, no. 13, pp. 5150-5161, september 2018. [23] s. kim and s. kim, “flexural behavior of concrete beams with steel bar and frp reinforcement,” journal of asian architecture and building engineering, vol. 18, no. 2, pp. 89-97, may 2019. [24] w. ge, y. wang, a. ashour, w. lu, and d. cao, “flexural performance of concrete beams reinforced with steel-frp composite bars,” archives of civil and mechanical engineering, vol. 20, pp. 1-17, june 2020. [25] a. s. mohamed, “flexural behavior and design of steel-gfrp reinforced concrete beams,” aci materials journal, vol. 110, no. 6, pp. 677-686, november 2013. [26] s. kartal and i. kalkan, “first cracking moments of hybrid fiber reinforced polymer-steel reinforced concrete beams,” international journal of civil and environmental engineering, vol. 13, no. 9, pp. 527-533, 2019. [27] j. valivonis and t. skuturna, “cracking and strength of reinforced concrete structures in flexure strengthened with carbon fibre laminates,” journal of civil engineering and management, vol. 13, pp. 317-323, december 2007. [28] d. gao, p. you, and z. changhui, “crack resistant performance and crack width of one-way slabs strengthened with frp,” applied mechanics and materials, vols. 71-78, pp. 3810-3815, july 2011. [29] aci committee 440, “guide for the design and construction of structural concrete reinforced with fiber-reinforced polymer (frp) bars,”american concrete institute, aci440.1r-15, march 2015. [30] j. c. mccormac and r. h. brown, design of reinforced concrete, 10th ed. united states: wiley, 2015. [31] w. j. ge, a. f. ashour, j. yu, p. gao, d. f. cao, c. cai, et al., “flexural behavior of ecc-concrete hybrid composite beams reinforced with frp and steel bars,” journal of composites for construction, vol. 23, no. 1, 04018069, february 2019. [32] s. f. liu, m. y. hou, x. liu, and r. h. wang, “analysis of cracking moment and immediate deflection in frp rebars reinforced concrete beams,” advanced materials research, vols. 671-674, pp. 618-621, march 2013. [33] concrete and reinforced concrete structures, russian technical standard sp 63.13330, 2019. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). 234 1___aiti#9252___155-168 advances in technology innovation, vol. 7, no. 3, 2022, pp. 155-168 observer-based quadratic guaranteed cost control for linear uncertain systems with control gain variation satoshi hayakawa * , yoshikatsu hoshi, hidetoshi oya graduate school of integrative science and engineering, tokyo city university, tokyo, japan received 11 january 2022; received in revised form 23 april 2022; accepted 24 april 2022 doi: https://doi.org/10.46604/aiti.2022.9252 abstract this study proposes a method for designing observer-based quadratic guaranteed cost controllers for linear uncertain systems with control gain variations. in the proposed approach, an observer is designed, and then a feedback controller that ensures the upper bound on the given quadratic cost function is derived. this study shows that sufficient conditions for the existence of the observer-based quadratic guaranteed cost controller are given in terms of linear matrix inequalities. a sub-optimal quadratic guaranteed cost control strategy is also discussed. finally, the effectiveness of the proposed controller is illustrated by a numerical example. the result shows that the proposed controller is more effective than conventional methods even if system uncertainties and control gain variations exist. keywords: polytopic uncertainty, quadratic guaranteed cost control, observer-based controller, control gain variation, linear matrix inequality (lmi) 1. introduction in the design of control systems for dynamical systems, it is necessary to derive a mathematical model for the controlled system. if the mathematical model represents the control system precisely, then the desired control system can be designed by various control design strategies. however, it is unavoidable that there are some gaps between an original controlled system and its mathematical model, and these gaps are known as “uncertainty.” therefore, robust controller design methods that can explicitly deal with uncertainties have been well studied. a large number of robust control strategies have been proposed [1-3]. most conventional robust controllers have been designed by solving linear matrix inequalities (lmis) and have fixed gains that are designed by considering the worst-case variation for uncertainties. in fact, it is desirable to design robust control systems with not only robust stability but also satisfactory control performance. to achieve this, chang and peng [4] proposed guaranteed cost control. in this design method, there is an upper bound on a given performance index. the degradation of the system performance caused by uncertainties is guaranteed to be below this bound. many studies have adopted this concept [5-7]. in the work of moheimani and peterson [6], the riccati equation approach [5] was extended to uncertain time-delay systems, and a guaranteed cost controller design method that solves a certain parameter-dependent riccati equation was proposed. yu and chu [7] proposed a design method for guaranteed cost controllers for linear uncertain time-delay systems that uses the lmi approach. studies on robust control generally assume that the full state of the controlled systems can be measured. however, in practice, the full state of systems cannot be obtained due to practical constraints. to overcome this problem, some observer-based quadratic stabilizing controllers have been presented [8-11]. for example, oya and hagino [10] proposed an observer-based guaranteed cost controller for polytopic uncertain systems. the polytopic representation allows the structure of uncertainties to be directly represented. * corresponding author. e-mail address: itustellar00@gmail.com advances in technology innovation, vol. 7, no. 3, 2022, pp. 155-168 on the other hand, keel and bhattacharyya [12] pointed out that it is necessary for a controller to tolerate some uncertainty when the control input is implemented. controller implementation involves the uncertainties inherent in analog-to-digital and digital-to-analog conversion and roundoff errors in numerical computations. thus, a nonzero margin of tolerance is required for the controller design. many design methods that consider control gain variation have been proposed [13-16]. yang et al. [13] proposed a design method for �� control for linear systems with addition control gain variation. famularo et al. [14] considered not only control gain variations but also the uncertainty of the system matrix. oya et al. [15] proposed a design method for a robust controller for linear uncertain systems with control gain perturbation. however, the problem of observer-based quadratic guaranteed cost control for linear uncertain systems with control gain variation has not been discussed. this study proposes a method for designing an observer-based guaranteed cost controller for linear uncertain systems with control gain variation. in this study, the design approach is separated into two steps [10, 17]. in the first step, an observer is designed; in the second step, a feedback controller that guarantees the upper bound on the given quadratic cost function is derived. sufficient conditions for the existence of the proposed controller are given in terms of lmis. the proposed control system can thus be designed using software such as matlab’s lmi control toolbox and scilab’s lmitool. 2. preliminaries this section presents two lemmas that are used in this study. lemma 1 shows the relation between matrices and a positive constant. lemma 2 is the schur complement formula. lemma 1 [16]: for matrices p and h that have appropriate dimensions, the following formula is obtained. 1t t t t ph h p pp h h+ ≤ +γ γ (1) where � is a positive constant. lemma 2 (schur complement formula [16]): for a given constant real symmetric matrix ξ, the following items are equivalent. 11 12 12 22 (i) 0 t ξ ξ  ξ = >  ξ ξ  (2) 1 11 22 12 11 12(ii) 0 and 0 t −ξ > ξ − ξ ξ ξ > (3) 1 22 11 12 22 12(iii) 0 and 0 t−ξ > ξ − ξ ξ ξ > (4) 3. problem formulation this study considers the linear uncertain system described by the following: ( ) ( ) ( ) ( )x t a x t b u t= +ɺ θ (5) ( ) ( )y t cx t= (6) where � ∈ ℝ�× and � ∈ ℝ�×� denote the known nominal matrices, and (�) ∈ ℝ� , �(�) ∈ ℝ , and �(�) ∈ ℝ� are the vectors of the state, control input, and output, respectively. the full state cannot be measured. in eq. (5), �(�) is supposed to have appropriate dimensions and the following structure: 156 advances in technology innovation, vol. 7, no. 3, 2022, pp. 155-168 1 ( ) n k k k a a a = = + ∑θ θ (7) in eq. (7), the matrix � ∈ ℝ�×� represents the known nominal value for system parameters, and the matrix ak, k = 1, 2, ……, n, denotes the structure of the uncertainties. the parameter � � (��,……,��)� represents unknown parameters that belong to the following parameter set: 1 | 1, 0 for 1, , n n k k k k nθ θ θ =   ∆ ∈ = ≥ =     ∑≜ ℝ … (8) furthermore, for ∀� ∈ ∆, it is assumed that the pair (�(�),�) and (�,�(�)) are controllable and observable respectively. now the following full-state observer is introduced: ˆ ˆ ˆ( ) ( ) ( ) ( )( ( ) ( ))x t ax t bu t g t y t cx t= + + −ɺ (9) where �(�) ∈ ℝ�×� is the observer gain matrix which is described as: ( ) ( )g t g g t= + ∆ (10) ( ) g g t∆ ≤ ε (11) in eqs. (10) and (11), �(�) ∈ ℝ�×� shows the uncertainty for the observer gain matrix, and �� is a known constant that represents the upper bound for ∆�(�). in other words, ∆�(�) fluctuates in the range given in eq. (11). the actual control input �(�) is defined as: ˆ( ) ( ) ( )u t k t x t−≜ (12) where (�) ∈ ℝ ×� is the control gain matrix which satisfies the following relation: ( ) ( )k t k k t= + ∆ (13) ( ) k k t∆ ≤ ε (14) in eq. (14), (�) ∈ ℝ ×� represents the uncertainty of the controller gain matrix, and the known constant �! is the upper bound for ∆ (�). in this study, the control input with ∆ (�) is considered so as to design the observer-based quadratic guaranteed cost controller under control gain variation. namely, the manipulated control input for the uncertain linear system in eqs. (5) and (6) is �(�) ≜ # $(�). fig. 1 shows the configuration of the proposed control system. fig. 1 configuration of the proposed control system 157 advances in technology innovation, vol. 7, no. 3, 2022, pp. 155-168 for the linear uncertain system given in eqs. (5) and (6), the following quadratic cost function is defined: { } 0 ( ) ( ) ( ) ( )t tj x t q x t u t r u t d t ∞ = +∫ (15) where the weighting matrices % ∈ ℝ�×� and & ∈ ℝ × are positive definite and can be selected by designers. by introducing an estimation error vector '(�) ≜ (�) # $(�), from eqs. (5) and (9), the following estimation error system is obtained: ˆ( ) ( ( ) ( ) ) ( ) ( ) ( )ee t a g t c e t a x t= − +ɺ θ θ (16) where �((�) is the matrix given by �((�) ≜ �(�) # �. moreover, an augmented vector ((�) ≜ ( $(�) '(�))� is introduced. then, from eqs. (5), (6), (9), (13), and (16), the following augmented system is derived: ( ) ( ) ( ) e e x t x t= ωɺ θ (17) ( ) ( ) ( ) ( ) ( ) ( )e a b k t g t c a a g t c −  ω =  −  θ θ θ (18) moreover, by using the estimated error vector '(�), the control input in eq. (12), and the augmented vector ((�), the quadratic cost function in eq. (15) can be rewritten as follows: 0 ( ) ( ) ( ) e t x e ej x t t x t d t ∞ = γ∫ (19) ( ) ( ) ( ) t q k t r k t q t q q  + γ =      (20) note that because the weighting matrices % ∈ ℝ�×� and & ∈ ℝ × in eq. (18) are positive definite, the matrix γ(�) in eq. (20) is semi-positive definite. applying lemma 2 to eq. (20), the semi-positive definiteness of the matrix γ(�) can be obtained as follows: 1 ( ) ( ) ( ) ( ) t t q k t rk t qq q k t rk t −+ − = (21) the definition of an observer-based quadratic guaranteed cost control is as follows. definition: the control input in eq. (12) is an observer-based quadratic guaranteed cost control for the linear uncertain system in eqs. (5) and (6) and the quadratic cost function in eq. (19) provided that the closed-loop system in eq. (17) is asymptotically stable for ∀� ∈ ∆ and there exists a positive constant +∗( ((0)) that satisfies ./0 ≤ +∗( ((0)). from the above discussion, the objective of this study is to design the observer gain matrix and the control gain matrix that guarantee the upper bound on the quadratic cost function in eq. (19). 4. main results this section shows an lmi-based design method for the observer gain matrix � ∈ ℝ�×� and the control gain matrix ∈ ℝ ×� that ensures the upper bound of the quadratic cost function. it is difficult to design both gain matrices simultaneously because of the uncertainty parameters. thus, the observer gain matrix � is first designed, and then the control gain matrix is determined. 158 advances in technology innovation, vol. 7, no. 3, 2022, pp. 155-168 4.1. design of observer gain matrix this study considers the design of the observer gain matrix � that stabilizes the following system obtained by ignoring the estimate $(�) in eq. (16): ( ) ( ( ) ( ) ) ( )e t a g t c e t= −ɺ θ (22) now, let 2�('̅) ≜ '̅�(�)4('̅(�) be a lyapunov function candidate. 4( ∈ ℝ�×� is a symmetric positive definite matrix. from lemma 1, lemma 2, and a previously reported result [1], a sufficient condition for the asymptotical stability of the system in eq. (22) is obtained as follows: ve x ( ) ( ) 0 , t t t t e e e e g e g e n y a a y h c c h c c y y i θ θ γ ε θ ε γ  + − − + < ∀ ∈ ∆  −  (23) from eq. (23), the observer gain matrix � can be easily designed as follows: 1 e eg y h −= (24) 4.2. design of control gain matrix in this section, the control gain matrix that minimizes the upper bound on the quadratic cost function in eq. (19) is designed. the following quadratic function is introduced as a lyapunov function candidate: ( ) ( ) ( ) t k e e ex x t x tλ≜v (25) where λ ∈ ℝ6�×6� is a symmetric positive definite matrix. the time derivative of the quadratic function 2!( () along the trajectory of the augmented system in eq. (17) can be computed as: ( ) ( )( ( ) ( )) ( ) t t k e e ex x t x t= ω λ + λωɺ θ θv (26) because the matrix γ(�) in eq. (20) is semi positive definite, the following inequality is considered: v e x( ) ( ) ( ) 0 , t tω λ + λ ω + γ < ∀ ∈ ∆θ θ θ (27) if a symmetric positive definite matrix λ and a control gain matrix that satisfy the matrix inequality in eq. (27) exist, then the following relation holds: ( ) ( ) ( ) ( ) 0 t k e e ex x t t x t< − γ <ɺv (28) namely, the augmented system in eq. (17) is quadratically stable and ((∞) � 0 holds. from '(�) ≜ (�) # $(�), the asymptotical stability of the linear uncertain system in eqs. (5) and (6) is ensured. furthermore, by integrating both sides of the inequality in eq. (27) from 0 to ∞, the following equation can be obtained: * 0 ( ) ( ) (0) (0) ( (0)) e t t e e e e exj x t x t dt x x x ∞ = λ < λ =∫ j (29) therefore, if matrices λ and that satisfy the lmi in eq. (27) exist, the asymptotical stability of the system in eqs. (5) and (6) is ensured and the upper bound on the quadratic cost function in eq. (19) is given by eq. (29). 159 advances in technology innovation, vol. 7, no. 3, 2022, pp. 155-168 now, by introducing the auxiliary parameter 8 ∈ ℝ�, the following matrix is considered (remark 1): ( ) ( ) ( ) t nq i k t rk t qt q q  + + γ =      δ δ (30) from eqs. (20) and (30), the relation γ9(�) # γ(�) ≥ 0 holds. therefore, the inequality in eq. (27) also holds provided that the following condition is satisfied: v e x( ) ( ) ( ) 0 , t tω λ + λ ω + γ < ∀ ∈ ∆δθ θ θ (31) here, ; ≜ diag(@,@() ≜ λa�(@,@( > 0 ∈ ℝ�×�) and c ≜ @ are defined. then, preand post-multiplying eq. (31) by ;, the condition in eq. (31) can be written as: v e x( ) ( ) ( ) 0 , t tω + ω + γ < ∀ ∈ ∆δθ θ θs s s s (32) the inequality in eq. (32) is organized as follows: 11 12 vex 12 22 ( ) ( , ) 0 0 0, 0 0( , ) ( , ) t n t e e t t s sq i q s sq qt t ψ ψ + + < ∀ ∈ ∆ ψ ψ                     θ δ θ θ θ (33) 1 1 ( ) ( ) ( ) ( ) ( ) t t t t t t t a s s a b b s k t b b k t s s k t r k t sψ = + − − − ∆ − ∆ +w w (34) 12 ( , ) ( ) ( ) t e e et sa gcs g t csψ = + + ∆θ θ (35) 22 ( , ) ( ) ( ) ( ) ( ) t t t t t e e e e e et a s s a gcs s c g s c g t g t csψ = + − − − ∆ − ∆θ θ θ (36) using lemma 1 and lemma 2, the inequality of eq. (33) can be described as follows: 11 12 * 12 22 vex 1 2 ( ) 0 0 ( ) ( ) 0 0, 0 0 0 t t t e e k m s s s s r i −  ψ ψ        ψ ψ + γ < ∀ ∈ ∆           − +   δ θ θ θ θ ξε w w (37) 2 2 1 1 1k g n t t t t a s s a b b b b i s s s sψ = + − − + + + + ε α η ε α ξ w w (38) 1 2 ( ) ( ) t e e s a g c sψ = +θ θ (39) 2 2 2 1 ( ) ( ) ( ) 1 g t t t t e e e e n e e t e e a s s a g c s s c g i s c c s s c c s = + − − + + + ψ θ θ θ β ε β η (40) note that γ9∗ in eq. (37) is the matrix expressed as: * n q i q q q + γ       ≜δ δ (41) 160 advances in technology innovation, vol. 7, no. 3, 2022, pp. 155-168 one can easily see that the matrix γ9∗ is positive definite because its positive definiteness is equivalent to % + δf� # %%a�% � δf� > 0. because the matrix γ9∗ is a positive definite, lemma 2 can be applied to the inequality condition in eq. (37). as a result, the following condition is obtained: ( ) 11 12 2212 1 2 vex 1 * 00( ) 0 ( ) 00 0 0( ) 0 00 0 000 0 0 0 0000 0( ) 0 0 0 0,000 0 0 00 00 0 0 0 000 00 0 0 0 0 0 0 0 0 0 0 0 0 − − − γ ψ ψ ψψ − + − −ψ = < − − ∀ ∈ ∆                              δ θ ε θθ ξε αε ξθ β η θ t k tt t eee k m nk n e l e l e ss s ss cs c r i is is cs i cs i s s w w (42) 2 1 1 t t t t g n a s s a b b b b iψ = + − − + +α η εw w (43) 1 2 ( ) ( ) t e e s a g c sψ = +θ θ (44) 2 2 2 ( ) ( ) ( ) t t t e e e e g n a s s a g c s s c g iψ = + − − +θ θ θ β ε (45) the condition in eq. (42) is an lmi for ; > 0,c,g > 0,h > 0,i > 0 and j > 0. if the solution ; > 0,c,g > 0,h > 0,i > 0 and j > 0 of the lmi in eq. (42) exists, an observer-based quadratic guaranteed cost control law is obtained as follows: ˆ( ) ( )u t k x t−≜ (46) 1 k s −= w (47) from the above, the following theorem for designing the observer-based quadratic guaranteed cost controller is obtained: theorem 1: by solving the lmi in eq. (23), the observer gain matrix � is derived as � � 4(a��( in advance. if there exists the solution ; > 0,c,g > 0,h > 0,i > 0 and j > 0 for ∃8 > 0 satisfying the lmi, vex( ) 0,ψ < ∀ ∈ ∆θ θ (48) then the control gain matrix can be computed as � c@a� and the control law �(�) � # $(�) becomes an observer-based quadratic guaranteed cost control. 4.3. sub-optimal guaranteed cost control because the lmi in eq. (42) has a convex solution ; > 0,c,g > 0,h > 0,i > 0 and j > 0, it can be optimized by using software such as matlab’s robust control toolbox. in this section, the following optimization problem is considered: , , , , , * ( (0)) subject to eq. (42) and 0, 0, 0, 0, 0minimize ex > > > > >α β η ξ α β η ξ s w j s (49) 161 advances in technology innovation, vol. 7, no. 3, 2022, pp. 155-168 if the optimization problem in eq. (49) is solved, then a sub-optimal observer-based quadratic guaranteed cost control can be obtained. however, the upper bound +∗( ((0)) in eq. (49) depends on the initial augmented vector ((0). note that the error '(0) cannot be utilized because the initial state (0) cannot be completely observed. thus, to avoid this dependence, it is assumed that the initial vector ((0) is a random vector that satisfies em ((0) (�(0)n � f6� and em ((0)n � 0. then, the upper bound on the quadratic cost function in eq. (29) is given as op+∗q ((0)rs � tu{λ}. therefore, the minimization problem of tu{λ} minimized subject to the lmi constraint in eq. (42) can be derived. moreover, by introducing a complementary variable ∑ ∈ ℝ6�×6�, which is a symmetric positive definite matrix, the following relation is considered: 2 2 0 0 n n i i σ  σ ≥ λ > ⇔ ≥   s (50) then, the minimization problem of tu{λ} can be transformed into that of tu{∑} because the condition in eq. (50) is an lmi in ∑ and ;. consequently, the optimization problem in eq. (49) can be reduced to the following constrained convex optimization problem: , , , , , , { } subject to eqs. (42) and (50) and 0, 0, 0, 0, 0minimize tr σ σ > > > > > α β η ξ α β η ξ s w s (51) finally, the following theorem can be obtained: theorem 2: if there exists the solution ; > 0,c,g > 0,h > 0,i > 0 and j > 0 that satisfies the constrained convex optimization problem in eq. (51), there exists a sub-optimal observer-based quadratic guaranteed cost control. note that by using the solution of the lmi in eq. (23), the observer gain matrix � is derived as � � 4(a��( in advance. if the solution to the constrained convex optimization problem is obtained, the control gain matrix can be computed as � c@a�. therefore, the control law �(�) � # $(�) is a sub-optimal observer-based guaranteed cost control. remark 1: this study introduced the auxiliary parameter 8 in eq. (30). if 8 is zero, the positive definiteness of the matrix γ9∗ is reduced to the relation % # %%a�% � 0. then, lemma 2 cannot be applied to the inequality in eq. (37). if the parameter 8 is a positive scalar, then the lmi in eq. (42) can be obtained. however, if the parameter 8 is set to a larger value, the result will be more conservative. therefore, 8 is set to be small as possible. 5. simulation this section demonstrates the effectiveness of the proposed method. as an example, the following aircraft model is considered [18]: 0.091 0.097 0 1 0 0 0 0 0 0 1 0 0 0 0 0 5.43 0 0.686 3.62 2.87 0.638 0 0 ( ) ( ) ( ) 0.56 0 0 0.122 0.127 0.459 0 0 0 0 0 0 10 0 10 0 0 0 0 0 0 10 0 10 x t x t u t − −               − − = +    + −       −       −    ɺ ω (52) ( )( ) 0 1 0 0 0 0 ( )y t x t= (53) where the parameter y represents the uncertainties and is assumed to vary in the range of m#1.01.0n. let case 1 and case 2 be y � #1.0 and y � 1.0, respectively; these two cases are the worst cases of the uncertainties. it is assumed that the initial state and the initial estimate are (0) � (1.0 0 0 0 0 0)� and $(0) � (0 0 0 0 0 0)�, respectively. the state variables are 162 advances in technology innovation, vol. 7, no. 3, 2022, pp. 155-168 shown in table 1. this simulation sets the weighting matrix of the quadratic cost function, the parameter 8, and the variations of and � as follows: 5 6 21 .0 , 4 .0 , 1 .0 1 0q i r i −= = = ×δ (54) 1 1 1 1 1 1 ( ) 0 .1(1 .0 | c o s (1 0 ) |) 1 1 1 1 1 1 t k t e t −       ∆ = − π (55) 1 1 1 1 1 1( ) 0 . 1(1 .0 | c o s (1 0 ) |) ( )t tg t e t−∆ = − π (56) in addition, �! and �� are set as �! � 0.35 and �� � 0.25, respectively. by solving the lmi condition in eq. (23), the observer gain matrix � is derived as: ( )10.9786 9.4 665 59.8127 11.64 23 14.4855 3.4610 tg = − − − (57) then, by applying theorem 2 and solving the constrained convex optimization problem, the control gain matrix is obtained as: 5 1 .3 2 7 3 1 .5 3 2 6 3 .7 1 0 2 3 4 .5 1 8 3 1 .7 9 2 0 1 .3 7 2 5 1 6 2 .0 6 5 6 7 .9 5 5 9 0 .2 5 4 0 1 2 4 .4 2 2 7 1 .3 7 2 3 5 .1 4 4 4 k −  =  − −  (58) consequently, the upper bound on the quadratic cost function op+∗q ((0)rs is obtained as op+∗q ((0)rs � 6.4975 × 10c. this simulation compares the results of the proposed method, the conventional linear quadratic regulator (lqr), and the work of oya et al. [10]. oya et al. [10] proposed a method for designing observer-based quadratic guaranteed cost control for uncertain systems. the control gain matrix in eq. (59) is the result obtained by using the lqr: 0 .5 2 7 2 0 .2 3 1 5 0 .2 6 7 3 0 .4 3 3 6 0 .1 5 7 0 0 .0 2 2 3 0 .3 2 9 5 0 .0 4 9 5 0 .0 8 5 7 0 .6 4 7 8 0 .0 2 2 3 0 .1 2 7 0 k −  =  −  (59) �d in eq. (60) and d in eq. (61) are the observer gain matrix and the control gain matrix obtained from the design method in the work of oya et al. [10]: ( )4 .7 6 6 3 4 .7 3 2 8 3 3 .2 6 2 6 6 .0 9 6 8 6 .2 5 4 4 1 .2 6 4 4 trh = − − − (60) 9 .6 0 2 0 2 7 .1 4 0 7 1 5 .0 4 6 8 6 .3 8 1 4 3 .2 9 6 7 0 .8 4 0 0 4 .0 7 8 7 6 .3 0 7 6 3 .9 8 3 2 5 .2 4 9 9 0 .8 4 0 0 0 .6 5 1 0 rk −  =  −  (61) figs. 2-5 and figs. 6-9 show the results of the lqr and the method of oya et al. [10]. from the results, the lqr and the method of oya et al. [10] did not achieve asymptotical stability. figs. 6 and 7 show that the state diverged in case 1 by the method of oya et al. [10]. control gain variation was not considered in the work of oya et al. [10], and thus the system could not be stabilized. table 1 state variables of example aircraft �(�) dimensionless slide-slip velocity (dsv) 6(�) roll c(�) roll rate e(�) yaw rate f(�) aileron angle g(�) rudder angle 163 advances in technology innovation, vol. 7, no. 3, 2022, pp. 155-168 fig. 2 time histories of �(�) # c(�) by lqr (case 1) fig. 3 time histories of e(�) # g(�) by lqr (case 1) fig. 4 time histories of �(�) # c(�) by lqr (case 2) fig. 5 time histories of e(�) # g(�) by lqr (case 2) fig. 6 time histories of �(�) # c(�) by the method of oya et al. [10] (case 1) fig. 7 time histories of e(�) # g(�) by the method of oya et al. [10] (case 1) fig. 8 time histories of �(�) # c(�) by the method of oya et al. [10] (case 2) fig. 9 time histories of e(�) # g(�) by the method of oya et al. [10] (case 2) 164 advances in technology innovation, vol. 7, no. 3, 2022, pp. 155-168 on the other hand, figs. 10-14 show the results for the proposed controller design method. as shown, even in the presence of system uncertainties and control gain variation, the proposed method achieved asymptotic stability. this demonstrates the effectiveness of the proposed quadratic guaranteed cost controller. fig. 10 time histories of �(�) # c(�) by the proposed method (case 1) fig. 11 time histories of e(�) # g(�) by the proposed method (case 1) fig. 12 time histories of �(�) # c(�) by the proposed method (case 2) fig. 13 time histories of e(�) # g(�) by the proposed method (case 2) fig. 14 time histories of the input by the proposed method 6. conclusions this study proposed a method for designing an observer-based quadratic guaranteed cost controller for linear uncertain systems with control gain variation. in the proposed approach, the observer gain matrix was first designed, and then the control gain matrix was determined. the design parameter 8 was introduced. the design of an observer-based quadratic guaranteed cost controller was reduced to an lmi condition. moreover, a robust sub-optimal guaranteed cost controller was investigated. the results of this study are a natural extension of those in the work of oya et al. [10]. although the uncertainty in the input 165 advances in technology innovation, vol. 7, no. 3, 2022, pp. 155-168 matrix has been considered in the work of oya et al. [10], the proposed design method can be easily applied to such a problem. by introducing additional actuator dynamics and constituting an augmented system, the uncertainties in the input matrix are embedded in the system matrix of the augmented system. in future work, the proposed adaptive robust controller synthesis will be extended to a broad class of systems, including uncertain linear systems with time delays and decentralized control for large-scale interconnected systems. in the proposed design method, if the parameter 8 is set to a larger value, the result will be more conservative. therefore, reducing conservatism should also be considered. conflicts of interest the authors declare no conflicts of interest. references [1] h. oya, et al., “observer-based robust control giving consideration to transient behavior for linear systems with structured uncertainties,” international journal of control, vol. 75, no. 15, pp. 1231-1240, october 2002. [2] h. oya, et al., “adaptive robust control scheme for linear systems with structure uncertainties,” ieice transactions on fundamentals of electronics, communications, and computer sciences, vol. e87-a, no. 8, pp. 2168-2173, august 2004. [3] h. oya, et al., “robust control giving consideration to time response for a linear systems with uncertainties,” transactions of the institute of systems, control, and information engineers, vol. 15, no. 8, pp. 404-412, august 2002. [4] s. s. chang, et al., “adaptive guaranteed cost control of systems with uncertain parameters,” ieee transactions on automatic control, vol. 17, no. 4, pp. 474-483, august 1972. [5] s. o. r. moheimani, et al., “optimal quadratic guaranteed cost control of a class of uncertain time-delay systems,” iee proceedings—control theory and applications, vol. 144, no. 2, pp. 183-188, march 1997. [6] i. r. petersen, et al., “optimal guaranteed cost control and filtering for uncertain linear systems,” iee transactions on automatic control, vol. 39, no. 9, pp.1971-1977, september 1994. [7] l. yu, et al., “an lmi approach to guaranteed cost control of linear uncertain time delay systems,” automatica, vol. 35, no. 6, pp. 1155-1159, june 1999. [8] s. h. park, et al., “h∞ control with performance bound for a class of uncertain linear systems,” automatica, vol. 30, no. 12, pp. 2009-2012, april 1994. [9] r. petersen, “a riccati equation approach to the design of stabilizing controllers and observers for a class of uncertain linear systems,” ieee transactions on automatic control, vol. 30, no. 9, pp. 904-907, september 1985. [10] h. oya, et al., “observer-based guaranteed cost control for polytopic uncertain systems,” the japan society of mechanical engineers, vol. 71, no. 710, pp. 89-98, october 2005. [11] s. nagai, et al., “a point memory observer with adjustable parameters for a class of uncertain linear systems with state delay,” proceedings of engineering and technology innovation, vol. 11, pp. 38-45, january 2019. [12] l. h. keel, et al., “robust, fragile, or optimal?” ieee transactions on automatic control, vol. 42, no. 8, pp. 1098-1105, august 1997. [13] g. h. yang, et al., “h∞ control for linear systems with additive controller gain variations,” international journal of control, vol. 73, no. 16, pp. 1500-1506, february 2000. [14] d. famularo, et al., “robust non-fragile lq controllers: the static state feedback case,” proceedings of the 1998 american control conference, vol. 2, pp. 1109-1113, june 1998. [15] h. oya, et al., “guaranteed cost control for uncertain linear continuous-time systems under control gain perturbations,” the japan society of mechanical engineers, vol. 72, no. 713, pp. 72-101, january 2006. [16] k. miyakoshi, et al., “synthesis of formation control systems for multi-agent systems under control gain perturbations,” advances in technology innovation, vol. 5, no. 2, pp. 112-125, april 2020. [17] d. rosinová, et al., “output feedback stabilization of linear uncertain discrete systems with guaranteed cost,” 15th triennial world congress, vol. 35, no. 1, pp. 211-215, july 2002. [18] m. noton, modern control engineering, new york: pergamon press, 1972. [19] s. nagai, et al., “synthesis of decentralized variable gain robust controllers with guaranteed l2 gain performance for a class of uncertain large-scale interconnected systems,” journal of control science and engineering, vol. 2015, article no. 342867, 2015. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). 166 advances in technology innovation, vol. 7, no. 3, 2022, pp. 155-168 appendix in this appendix, the extension to ℒ6 gain performance is discussed. the following uncertain system is considered: ( ) ( ) ( ) ( ) ( )xx t a x t b u t d t= + +ɺ θ ω (a1) ( ) ( ) ( )zz t c x t d t= +ɺ ω (a2) where (�) ∈ ℝ� , �(�) ∈ ℝ , i(�) ∈ ℝ� , and y(�) ∈ ℝj are the vectors of the state, input, output, and disturbance, respectively. the following full-state observer is introduced: ˆ ˆ ˆ( ) ( ) ( ) ( )( ( ) ( ))x t ax t bu t g t z t cx t= + + −ɺ (a3) let the input �(�), the estimation error '(�) and the augmented vector ((�) be �(�) ≜ # (�) $(�), '(�) ≜ (�) # $(�), and ((�) ≜ ( $(�) '(�))�, respectively. the augmented system and the estimation error system can be obtained as follows: ˆ( ) ( ( ) ( ) ) ( ) ( ) ( ) ( ( ) ) ( ) e x z e t a g t c e t a x t d g t d t= − + + −ɺ θ θ ω (a4) ( ) ( ) ( ) ( )e ex t x t t= ω +ɺ θ ωd (a5) ( ) ( ) ( ) ( ) ( ) ( ) e a b k t g t c a a g t c − ω = −       θ θ θ (a6) ( ) ( ) ( ) z x z g t d t d g t d = −       d (a7) please refer to the work of nagai et al. [19] for a definition and a lemma about ℒ6 gain performance. the observer gain matrix � is designed as eq. (24) in section 4.1. to design control gain matrix , the following lyapunov function and hamiltonian are defined: ( ) ( ) ( ) t k e e e x x t x tλ≜v (a8) * 2 ( , ) ( ) z ( ) ( ) ( ) ( ) ( ) t t e k e h x t x t z t t t+ −ɺ≜ γ ω ωv (a9) here, (�∗)6 ≜ � and (�) ≜ (f� f�)k/$(l)((l)m � n ((�) are introduced. furthermore, the following equation can be obtained: ( ) ( ) ( ) ( ) ( , ) ( ) ( ) ( )( ) t t t t t zt t e e e t t t z z z p c c t c d x t h x t x t t tt d c d d i                      λ ω + ω λ + λ + = λ + − θ θ ω ωγ t t d t d t (a10) to satisfy �( (, �) < 0, the following inequality is considered: v ex ( ) ( ) ( ) 0 , ( ) t t t t t z t t t z z z p c c t c d t d c d d i λ ω + ω λ + λ + < ∀ ∈ ∆ λ + −         θ θ θ γ t t d t d t (a11) ; ≜ diag(@,@() ≜ λa�(@,@( > 0 ∈ ℝ�×�) and c ≜ @ are defined, then preand post-multiplying both sides of eq. (a11) by diagq;,fjr and using lemmas 1 and 2, the following theorem is obtained: 167 advances in technology innovation, vol. 7, no. 3, 2022, pp. 155-168 theorem a: the system in eqs. (a1) and (a2) is asymptotically stable if there exists a solution ; > 0,c,g > 0,h > 0,i > 0,p > 0 and j > 0 satisfying the following lmi: 11 12 22 12 ( ) 0 0 0 0 ( ) 0 0( ) 0 0 0 00 0 0 00 0 0 ( ) 0 00 0 0 00 0 000 000 0 00 00000 00 0 0000 0 00 0000 t tz z k tt t tt x z e z e ee t t t t tt t t z z x z z e z z pz z e l n k l e l l z l z gd sc d sc s d gd s c d s c s cs c d dd d g d cs d d id g d cs cs ics i s i cs i i d i d  +ψ ψ ψ − +ψ − + −+ − ψ = − − − − −  θ ε θθ γ θ α ε β η ξ λ vex 0,                  <                  ∀ ∈ ∆θ (a12) 2 2 1 1 t t t t g n g n a s s a b b b b i iψ = + − − + + +α η ε ξ εw w (a13) 12 ( ) ( ) t e esa gcsψ = +θ θ (a14) 2 2 22 ( ) ( ) ( ) g g t t t e e e e n na s s a gcs s c g i iψ = + − − + +θ θ θ βε λε (a15) by solving the lmi in eq. (23) in advance, the observer gain matrix � is derived as � � 4(a��(. if the solution of the lmi condition in eq. (a12) is obtained, then the control gain matrix can be computed as � c@a�. therefore, the control law �(�) � # $(�) is the observer-based control. 168 2___aiti#8933_in press advances in technology innovation, vol. 7, no. 2, 2022, pp. 92-104 behavior of built-up cold-formed steel stub columns infilled with washed bottom ash concrete mohd syahrul hisyam mohd sani 1,* , fadhluhartini muftah 1 , nor maslina mohsan 1 , bishir kado 2 1 school of civil engineering, college of engineering, universiti teknologi mara (uitm) cawangan pahang, bandar jengka, malaysia 2 department of civil engineering, faculty of engineering, bayero university, kano, nigeria received 18 november 2021; received in revised form 25 january 2022; accepted 26 january 2022 doi: https://doi.org/10.46604/aiti.2022.8933 abstract the main objective of the study is to determine the behavior of built-up cold-formed steel (cfs) stub columns infilled with washed bottom ash (wba) concrete and also their failure mode. five proportions of wba as sand replacement in concrete and five specimens of built-up cfs stub columns infilled with wba are produced in this study. there are four parts of the testing conducted: material properties of cfs, material properties of wba concrete, mechanical properties of the connection, and mechanical behavior of built-up cfs stub columns. the result shows that the specimen with 25% wba is reported to have the highest value of compressive strength in the material properties of wba concrete and the mechanical behavior of built-up cfs stub column. the percentage difference of the ultimate load of the built-up cfs column filled with normal concrete and filled with wba concrete is noted to have a range of 3% to 25%. keywords: built-up section, cold-formed steel, stub column, washed bottom ash concrete, buckling 1. introduction cold-formed steel (cfs), sometimes identified as a thin-walled structure, is generally used as construction material either structural or non-structural element. cfs is produced by using steel sheets through a rolling process in ambient temperature, and is different from the hot-rolled steel produced by using high technology and huge machine in high temperature with an energy-consuming process. cfs is with many advantages such as being lightweight, ease of transportation, quick installation and erection, corrosion resistance, high strength to weight ratio, etc., and is normally utilized as a wall frame or panel, roof truss structure, and storage rack. cfs is formed in a variety of shapes, thicknesses, and cross-section areas, with the channel, hat, angle, and zee section normally found in the construction market. this section is becoming popular due to its ability for minimizing the total production or material cost. this popular section is classified as an open section which is common in the unsymmetrical section and exposed to structural integrity issues and failure. the development of cfs is constantly changing in line with technological changes such as the production of complex shapes and cross-sections, enhancing the quality of steel material, producing corrosion resistance methods, and improvising methods of forming [1]. in overcoming the issue of structural integrity, this section is established by using a combination of two or more sections to produce the symmetrical section and sometimes it becomes a closed section. selvaraj and madhavan [2] mentioned about the cfs section with an unsymmetrical or single symmetrical or open section that failed due to the instability effect. also, they mentioned that cfs is designed to transform to the built-up section which is classified as a close section or * corresponding author. e-mail address: msyahrul210@uitm.edu.my tel.: +60123267055 advances in technology innovation, vol. 7, no. 2, 2022, pp. 92-104 double symmetrical section. the new section which is recognized as the built-up section has existed in a variety of shapes, cross-section areas, and dimensions. the basic of the built-up section by using channel section is a back-to-back configuration or known as the i-section and face-to-face configuration or also as box-up or a hollow section. fig. 1 illustrates the example of a built-up face-to-face and back-to-back cfs section by using a variety of sections. muftah et al. [3] reported the result of the behavior of built-up face-to-face cfs section with difference channel section which is fastened by using bolt and nut and known as outstand and extended stiffener under bending load. by producing the built-up section, the production cost of the section can be reduced and the cost of using an expert and high technology machine can be reduced. the build-up section is formed by using a variety of fasteners or connectors, for instance, bolt and nut, full welding, spot welding, self-tapping screw, self-drilling screw, rivet, and innovation fastener. li et al. [4] have stated that the compression capacity in the axial condition of the built-up column section is allowed twice and more than twice of the basic section in producing the closed or symmetrical section. nie et al. [5] mentioned that there are a lot of researchers who studied screw spacing and thickness effect on the strength of built-up back-to-back cfs columns that are broadly utilized in engineering activity due to quick assembly and installation. meza et al. [6] stated that the performance and fundamental knowledge of built-up cfs columns are still limited. the main objective of the study is to determine the mechanical behavior of the built-up cfs columns under compression load and observe the failure mode of the columns. furthermore, the study is also to determine the suitable proportion which would provide the optimum mechanical behavior of built-up cfs columns with different proportions of the washed bottom ash (wba) as sand replacement in concrete for solving the structural integrity issues and failures. (a) face-to-face and back-to-back [8] (b) back-to-back [9] fig. 1 the example of the built-up cfs section 2. built-up cold-formed steel column the built-up cfs column is normally provided of the mechanical behavior, especially the ultimate load value with twice or more than twice of the individual section under axial compression [4]. roy and lim [7] have reported the built-up face-to-face 93 advances in technology innovation, vol. 7, no. 2, 2022, pp. 92-104 channel column which is classified as a hollow section with twice the strength of the individual section, and have promoted that the stability of the section is suitable to use in frames. in american iron and steel institute (aisi s100-16) and australia and new zealand (as/nz) specification, the built-up cfs section is designed by modifying the slenderness ratio (kl/r)m: 22 m o i kl kl a r r r = +                  (1) where (kl/r)o is the overall slenderness ratio, a is the length of the intermediate fastener, and ri is the radius of gyration (minimum). cfs with a thin and slender section as compression or flexural member tends to have the buckling failure, such as local, distortional, global and lateral buckling, web crippling, and torsion when subjected to load. the local, distortional, global, and lateral buckling of the cfs section is illustrated in fig. 2. several factors involved in the buckling failure of the cfs section are the cross-section, shape, imperfection, slenderness ratio, and height. normally, the column is divided by referring to the slenderness ratio into three categories: short, intermediate, and slender columns; the short column fails due to yielding, while the slender column fails due to buckling [10]. rokilan and mahendran [11] stated that the local buckling affected the cfs section to fail because the cfs section has a larger width-to-thickness ratio and is not similar to the hot-rolled steel section. selvaraj and madhavan [2] described that the individual cfs section, which is classified as an open or slender section, has failed in several ways such as local, distortional, and global buckling when the structural section is exposed to the instability conditions. nie et al. [5] reported that local buckling and local-flexural buckling happened for the cfs closed section, and the built-up closed section with two channel sections could avoid the distortional buckling when subjected to compressive load. li et al. [4] reported that the study on the mechanical behavior of the built-up cfs, especially the effect of the distortional buckling, is still limited, and no research has been discussed due to the complexity of the cross-section. (a) local (b) distortional (c) global (d) lateral fig. 2 the buckling failure of the cfs section 3. built-up cold-formed steel column with concrete the built-up cfs incorporated with concrete and mortar as the column structure is increased due to an increase in the demand for the tall building construction. the composite structure is produced to be excellent in strength, buckling resistance, seismic resistance, and fire resistance, and is also capable to reduce the production cost and material cost. ibanez et al. [12] reported that the combination of steel and concrete is broadly used in huge infrastructure and tall buildings because of its economic aspects, good structural behavior and bearing capacities, and better ductility. the built-up cfs column with concrete can delay the deformation and crack of the concrete when subjected to axial compression, and acts as a permanent formwork that replaces the timber formwork. zhu et al. [13] reported that the normal and self-consolidating concrete with a design strength of 30 mpa is utilized and filled into cfs tubes with 200 mm × 200 mm, square hollow section. mohd sani et al. [14] analyzed the resistance of the built-up face-to-face cfs column, which is fastened by bolt and nut and filled with normal concrete for the height of the column of 900 mm. qu et al. [15] studied the axial 94 advances in technology innovation, vol. 7, no. 2, 2022, pp. 92-104 compression behavior of the rectangular cfs tubes filled with concrete under two different loading methods. in design standard eurocode 4, the plastic resistance to compression (npl,rd) of the composite column based on the ultimate limit states is determined by the equation: , 0.85a y ck s sk pl rd c ma c s a f f a f n a= + +      γ γ γ (2) where aa is the cross-section areas of structural steel; ac is the cross-section areas of concrete; as is the cross-section of reinforcement; fy is the yield stress of steel; fck is the strength of concrete; fsk is the yield stress of steel reinforcement; ɣma, ɣc, and ɣs are the partial safety factors at the ultimate limit states. in general, the stub column or known as the short column is tested to obtain the structural behavior information as comprehensive and inclusive parametric studies to further assess the failure mode for considering varieties of the steel slenderness value and grades. besides, when the length of the beam or slab is extended for any reason, the stub column is utilized as a structure and specifically acts as a concentrated point load which is designed and built above the beam. the stub column is sometimes planned as an interior aspect and utilized to improve the stability of the building. the testing of the stub column is conducted to evaluate the effect of the width-thickness ratio (b/t) on buckling behavior and bearing capacity. bottom ash (ba) is the waste product from coal-fired electric power plants which physically is lightweight and has a granular, porous, and coarsen surface. ba is collected from the furnace or boiler kiln at the bottom side, and another waste product, which is normally recognized as fly ash, is collected from the precipitator process. normally, the total percentage of waste product of ba is more than the fly ash, and sometimes the ba is dumped at the nearest area for storing or recycling. if the ba is not properly stored, there will be environmental effects (such as air pollution, water pollution, and groundwater contamination) and human health effects (such as respiratory diseases and cancer risk) occurred imminently. thus, ba is proposed for the 3r (reduce, reuse, and recycle) process which is utilized as structural fill material, concrete ingredient material, and road base. ba with similar size of the sand is becoming popular to replace sand in concrete and lastly promoting the lightweight concrete. there is no information from previous studies on the utilization of ba filled in built-up cfs section as column or beam. nowadays, the normal concrete is shifted from traditional material to waste material which is parallel to the sustainable development program. the traditional material in normal concrete is replaced or substituted with waste material to solve the environmental problem which occurred from the beginning of the quarry activity. mainly, the co2 emission from aggregate occurred from the excavation, blasting, and transportation activities using electricity. fayaz et al. [16] have noted that sand is scarce due to the indian government imposing harsh restrictions on the sand quarrying at the river, which causes construction activities to be affected. the sand quarrying activities at the river normally produced environmental damage and river erosion [16]. besides, sand mining activities have also created a lot of problems such as riverbeds becoming deeper, riverbank collapsed, vegetation losses on a riverbank and, aquatic life and agriculture sector interrupted [17]. they have also separated the aggregate into three categories, i.e., recycled coarse aggregate, normal coarse aggregate, and fine aggregate that are being used in concrete manufacturing which contributed to co2 emission of 39%, 42%, and 19%, respectively. from the observation and analysis, the information of the built-up cfs created with spot weld is still limited rather than other fasteners to form a symmetrical and closed section. besides, the built-up cfs stub column infilled with normal concrete is considered a common of the study, but nowadays they are combined with special concrete for increasing the strength and improving the stability of the structure. the combination of built-up cfs with special concrete is recognized as new research activity and there is no code of practice that fully describes it. the arrangement and complete experimental setup for the built-up cfs stub column subjected to axial compression are not explained well in previous studies, especially the support condition and the imperfection analysis. 95 advances in technology innovation, vol. 7, no. 2, 2022, pp. 92-104 previous studies have not enlightened the utilization of special concrete in the built-up cfs stub column as a structural component to reduce the overall weight and production cost of the structure, such as the optimum percentage of waste material for replacing the traditional material. roy et al. [18] have reported that the information and design guidelines of the cfs stub column in the australia-new zealand code of practice (as/nzs 4600) are classified as conservative when compared with a slender column. roy et al. [19] stated that there are limited information and study on the determination of the strength due to axial compression for built-up face-to-face cfs and the effect of the spacing of the fastener. ferhoune and zeghiche [20] reported that very few studies by experiments have been conducted on the built-up cfs stub column which is with welding fastener and filled with normal concrete or special concrete. 4. specimen preparation and experimental setup the cfs channel section with double intermediate web stiffeners and with a dimension of web element of 75 mm, flange elements of 34 mm, lipped element of 8 mm, the thickness of 1 mm, and steel grade of 550 mpa is selected as shown in fig. 3. the section properties of the cfs channel section are tabulated in table 1. cfs channel section is clean and clear before starting by checking the material properties of the section using a coupon tensile test specimen. cfs is cut on the web and flange elements which are situated vertically as similar as the column condition by referring to bs en 10002-1:2001 [21]. then, two cfs channel sections are located face to face to produce the built-up cfs section as same as the square hollow section by using spot weld on three locations as shown in fig. 4. the spot weld with the width of 5 mm and with three numbers is located at the top, middle, and bottom of the built-up section in two parts (left and right) by referring to the study of roy et al. [19]. the height of the specimens is constant at 250 mm. the ba collected from the furnace or boiler kiln is prepared for the cleaning and washing process to form the wba with appropriate sizes. for material properties of the wba concrete, the concrete with grade 20 is designed by using material density and cast in 5 times accordingly to the proportion of sand replacement, 0%, 25%, 50%, 75%, and 100%. the total specimen of the concrete for material properties is 30 cubes, and the mix of all specimens is without using a superplasticizer. the concrete is cured in the water-curing tank for 7 days and 28 days of compressive strength. the built-up cfs section, as shown in fig. 4(a), is filled with normal concrete as a control specimen and wba concrete with 25%, 50%, 75%, and 100% to form a column. the total specimen of the built-up cfs column with wba concrete is 12 specimens and 3 specimens with normal concrete. the built-up cfs with normal and wba concrete is cured for 28 days before testing. the imperfection and residual stress of the cfs section are ignored in the study. in the experimental activity, there are four parts which include material properties of cfs, material properties of wba concrete, mechanical properties of the connection, and mechanical behavior of built-up cfs stub column. the universal testing machine (utm) with a capacity of 30 kn is used for determining the material properties of the cfs test. the mechanical properties of connection are divided into two parts: shear connection and pull-out connection test. from the material properties of cfs, the ultimate strength, yield strength, elastic modulus, and deformation at ultimate load are observed. furthermore, there are four specimens for the shear connection test and three specimens for the pull-out connection test proposed for checking the mechanical properties of the connection. the utm with a capacity of 100 kn is utilized. the ultimate load of all connection test specimens is determined and the failure mode is observed. next, the material properties of the normal and wba concrete, especially the ultimate load and compressive strength, are determined by using an auto compression machine with a capacity of 3000 kn. the failure mode of the concrete under compressive strength is observed. lastly, for the mechanical behavior of the built-up cfs stub column, the ultimate load of the column is evaluated and the failure mode of the column for all proportions of wba concrete is observed. the built-up cfs column without concrete is also determined for the comparison study. the experimental setup of the mechanical behavior of the built-up cfs stub column has followed the study of mohd sani and muftah [22] as shown in fig. 5. 96 advances in technology innovation, vol. 7, no. 2, 2022, pp. 92-104 fig. 3 the cross-section and dimension of the cfs channel section table 1 the section properties and dimensions of the cfs channel section parameter value and unit web (d) 75 mm lipped (l) 8 mm area (a) 148 mm 2 second moment of area (ixx) 0.135 × 10 6 mm 4 section modulus (zxx) 3.605 × 10 3 mm 3 radius of gyration (rx) 30.22 mm flange (f) 34 mm thickness (t) 1 mm yield strength (fy) 550 mpa second moment of area (iyy) 0.025 × 10 6 mm 4 section modulus (zyy) 2.240 × 10 3 mm 3 radius of gyration (ry) 12.94 mm (a) top (b) front fig. 5 experimental setup for testing the mechanical behavior of the built-up cfs stub column fig. 4 the view of the built-up cfs column 5. results and discussion the results and discussion for the four parts of the experiment are analyzed here for achieving the objective of the study. the results and discussion are started with the material properties of cfs, continued with the material properties of wba concrete and mechanical properties of the connection test, ending with the mechanical behavior of the built-up cfs stub column. 5.1. material properties of cold-formed steel (cfs) the result of the material properties of cfs is tabulated in table 2. from the table, the highest value of the ultimate load is the flange element, and the lowest value of the ultimate load is recorded at the web element. the percentage difference of 5.89% for ultimate load, 5.88% for ultimate strength, 6.43% for yield stress, 2.44% for elastic modulus, and 4.06% for deformation at ultimate load are recorded between web and flange elements. the ultimate load and strength of the flange element are more than the web element because of the process of bending the flat element into a new element. the ultimate and yield strength is vividly increased from web to flange element due to strain hardening and cold-rolling process in the ambient temperature. 97 advances in technology innovation, vol. 7, no. 2, 2022, pp. 92-104 besides, the increment is also due to cold-forming, which produces new strength, is added with existing strength, and decreases the ductility. therefore, the ultimate load and strength of the element are dependent on the bending process. if the element is without a bending process, the ultimate load and strength would show the lowest value. the cfs which was bought from the malaysia construction market has met the quality and is suitable for further work. dinis et al. [23] reported that the elastic modulus of the web element is higher than the flange element when testing by using a coupon tensile specimen. fig. 6 illustrates the example of the coupon specimen after failing, and fig. 7 shows the stress-strain graph of the cfs. table 2 the result of the material properties of cfs element ultimate load (kn) ultimate strength (mpa) yield strength (mpa) elastic modulus (gpa) deformation at ultimate load (mm) web 6.71 536.6 524 205 4.43 flange 7.13 570.1 560 200 4.25 fig. 6 the coupon specimen after testing 5.2. material properties of washed bottom ash (wba) concrete the material properties of wba concrete, especially the ultimate load and compressive strength, are tabulated in table 3. from the observation, the color of the specimen without wba is brighter if compared with the specimen with wba which shows dark grey. fig. 8 and fig. 9 illustrate the relationship between the compressive strength and the age of curing and the relationship between the compressive strength and the proportion of wba, respectively. the highest and lowest value of compressive strength is 25% and 100% respectively of the wba concrete specimen. the ultimate load and compressive strength of the wba concrete are increased by increasing 25% wba and decreased by increasing more than 25% wba. this is because the wba concrete with a high percentage of wba has the fresh mix very dry and the hardened concrete too brittle. the compressive strength of all specimens is increased from 7 days to 28 days as calculated approximately 24.26% of 0%, 26.15% of 25%, 25% of 50%, 21.05% of 75%, and 24% of 100%. the percentage achieved from the calculation is noted as significant and acceptable. the result of ultimate load and compressive strength with 25% wba which is more than the control mix is similar to the study by kim et al. [24]. kim et al. [24] reported that the specimens with 25%, 50%, and 75% fine ba aggregate have shown more value of compressive strength compared with the control mix. the percentage difference of the 0.00e+00 1.00e+02 2.00e+02 3.00e+02 4.00e+02 5.00e+02 6.00e+02 0 0.02 0.04 0.06 0.08 0.1 0.12 s tr e ss ( m p a ) strain (mm/mm) web flange fig. 7 the stress-strain graph of the material properties of cfs 98 advances in technology innovation, vol. 7, no. 2, 2022, pp. 92-104 compressive strength between all specimens with control is 7.34% of 25%, 18.81% of 50%, 71.78% of 75%, and 87.62% of 100% wba concrete specimen. the failure mode of the 0% (control mix), 25%, 50%, and 75% are shown in fig. 10. 100% wba concrete specimen is classified as brittle concrete as shown in fig. 11 with the overall side, and the corner of the cube is broken. table 3 the ultimate load and compressive strength of the wba concrete specimen specimen 7 days 28 days ultimate load (kn) compressive strength (mpa) ultimate load (kn) compressive strength (mpa) 0% (control) 153.5 15.3 202.2 20.2 25% 160.7 16.1 218.7 21.8 50% 123.0 12.3 164.0 16.4 75% 44.9 4.5 57.56 5.7 100% 19.5 1.9 24.66 2.5 fig. 10 the failure mode of specimen fig. 11 the failure mode of 100% wba specimen after compressive strength test 0 2 4 6 8 10 12 14 16 18 20 22 7 days 28 days c o m p re ss iv e s tr e n g th (m p a ) age of curing 0% (control) 25% 50% 75% 100% 0 5 10 15 20 25 0% (control) 25% 50% 75% 100% c o m p re ss iv e s tr e n g th (m p a ) specimen 7 days 28 days 0% (control mix) 25% 50% 75% fig. 8 the relationship of the compressive strength and the age of curing fig. 9 the relationship of the compressive strength and different specimens 99 advances in technology innovation, vol. 7, no. 2, 2022, pp. 92-104 5.3. mechanical properties of the connection test the shear connection test is conducted and the result is shown in table 4. the percentage difference between the full weld and spot weld is determined and noted to have 77.49% for 1sw, 73.60% for 2sw, and 27.67% for 3sw. 68.87% and 63.50% are reported when the 3sw specimen is compared with 1sw and 2sw, respectively. 3sw is classified as an appropriate connection method for the built-up cfs section that gives the optimum value and shows the highest value of ultimate load with less energy consumption and less cost compared with the full weld method. the failure mode of all specimens is observed to have the break at the location of the weld. fig. 12 illustrates the ultimate load between the methods of connection for shear testing. the graph illustrates that the ultimate load is increased by increasing the number of the spot weld. the result of the pull-out connection test is tabulated in table 5. three specimens are tested and two-surface full weld (2sfw) specimens obtain the highest ultimate load value of 37.03 kn. the percentage differences between the three-number spot weld (3sw) specimen with 1sfw and 2sfw are 27.10% and 55.04%, respectively. the deformation at ultimate load between 3sw and 1sfw is illustrated similarly and the value is not too far between them, around 9.37%. the deformation at ultimate load for 2sfw is 10.64 mm and demonstrates that the connection with full weld at two surfaces is the most practical method to the joint between two sections. with the less deformation at ultimate load among the specimen, 2sfw is considered an appropriate connection method but the specimen is categorized as more costly and has high energy consumption when using full weld and compared with other connection methods. 3sw specimen is classified as a suitable connection method that provides a significant ultimate load for joining between two specimens without high cost and energy consumption. all specimens fail due to breaking at the weld. table 4 the shear connection test result specimen symbol ultimate load (kn) failure mode full weld fw 13.37 break at the weld one-number of spot weld 1sw 3.01 break at the weld two-number of spot weld 2sw 3.53 break at the weld three-number of spot weld 3sw 9.67 break at the weld table 5 the pull-out connection test result specimen symbol ultimate load (kn) deformation at ultimate load (mm) three-number spot weld 3sw 16.65 25.14 one-surface full weld 1sfw 22.84 27.74 two-surface full weld 2sfw 37.03 10.64 5.4. mechanical behavior of built-up cold-formed steel (cfs) stub column the mechanical behavior of the built-up cfs stub column is tabulated in table 6 and fig. 13. the highest value of the ultimate load and compressive strength is 25% wba specimen, and the lowest value of the ultimate load and compressive 0 2 4 6 8 10 12 14 1sw 2sw 3sw fw u lt im a te l o a d ( k n ) specimen fig. 12 the graph of the ultimate load according to the specimen or method of connection 100 advances in technology innovation, vol. 7, no. 2, 2022, pp. 92-104 strength is 100% wba specimen. the ultimate load of the built-up cfs stub column without concrete is reported to have 73.5 kn. when the built-up cfs column is filled with normal concrete, the ultimate load is increased by around 55.86%. although when the built-up cfs column is filled with wba concrete with a different proportion, the ultimate load is improved approximately 66.53% of 25%, 54.148% of 50%, 48.24% of 75%, and 46.55% of 100% wba specimen. the percentage difference of the ultimate load between the control specimens with wba concrete is noted as having 24.18% for 25% wba specimen, 3.72% for 50% wba specimen, 14.72% for 75% wba specimen, and 17.42% for 100% wba specimen. the failure mode of the specimen is also tabulated in the table and illustrated in fig. 14. all specimens are observed to have local buckling and distortional buckling. nie et al. [5] mentioned about the local buckling existing for the short column at web and flange elements when subjected to axial compression. the web and flange elements are deformed and moved out from the original as shown in the red circle in fig. 15. the specimens do not fail at the connection area between one cfs channel with another cfs channel. the cfs section shows significant failure in buckling rather than concrete either normal or wba concrete which does not illustrate the crack or failure as illustrated in fig. 16. from the experimental activity, the built-up cfs with wba concrete of 25% is shown as the specimen with optimum value, and the built-up cfs with wba concrete of 50% is represented quite similar with control specimen. besides, the 75% wba and 100% wba specimens are observed to fail on the surface of the column due to the brittleness of the wba when fully replaced with sand. the pattern of the failure mode of all specimens is observed having the same condition and proven by previous studies, nie et al. [5] and nie et al. [25]. table 6 the result of the mechanical behavior of the built-up cfs stub column specimen ultimate load (kn) compressive strength (mpa) failure mode 0% (control) 166.5 31.71 local and distortional buckling 25% 219.6 41.83 local and distortional buckling 50% 160.3 30.53 local and distortional buckling 75% 142.0 27.05 local and distortional buckling 100% 137.5 26.19 local and distortional buckling fig. 13 the ultimate load of the built-up cfs column with different proportions of wba concrete fig. 14 the failure mode of the built-up cfs specimen from the front view 0 50 100 150 200 250 0% (control) 25% 50% 75% 100% u lt im a te l o a d ( k n ) specimen without concrete 0% wba 25% wba 50% wba 75% wba 100% wba 101 advances in technology innovation, vol. 7, no. 2, 2022, pp. 92-104 fig. 15 the failure mode of the built-up cfs specimen at the connection area fig. 16 the failure mode of the built-up cfs specimen from the top view 6. conclusions and recommendations from the result and observation of the complete experimental activity, several conclusions and recommendations are drawn as shown here. (1) the compressive strength of the wba concrete with 25% of wba is increased approximately 7.34% when compared with the control specimen. the result of compressive strength is increased between 21% and 26% when the concrete is fully submerged in water in early strength age with mature strength. the wba concrete with more than 25% of wba is shown to decrease and recorded to have 19% to 88% compared with the control mix. from the observation of the failure mode, the wba concrete is noted to have brittle conditions when achieving more than 50% or until 100% of wba. (2) the built-up cfs stub column with 25% wba concrete is shown to have the highest value of ultimate load and compressive strength with 219.6 kn and 41.83 mpa respectively and recorded to have 24.19% compared with the control specimen. the ultimate load and compressive strength of the built-up cfs column are increased when the wba concrete is less than 25% and decreased when the wba concrete is more than 25%. (3) all specimens failed the local and distortional buckling on the steel surface, but there are no cracks or failure on the concrete surface detected. thus, the utilization of 25% wba as sand replacement can reduce the production cost and environmental problems produced from the site of the sand quarry. for further study, the casting and mixing process of the fresh concrete must be added with the superplasticizer for controlling the strength and workability of the specimen. the built-up cfs intermediate and slender column should be designed and established to determine the relationship between the slenderness ratio and strength of the column. lastly, the built-up cfs column with a variety of the fastener should be designed, produced, and discussed, and the imperfection aspect should be added in numerical analysis to evaluate the mechanical behavior. acknowledgements the authors kindly thank the malaysia ministry of higher education under the fundamental research grant scheme (frgs/1/2018/tk01/uitm/02/2) for the financial support and universiti teknologi mara (uitm) cawangan pahang for the facility support, especially laboratory machines and equipment. sincerest gratitude is extended to the college of engineering, uitm cawangan pahang members and staff for providing technical assistance and advice. without concrete 0% wba 25% wba 50% wba 75% wba 100% wba without concrete 0% wba 25% wba 50% wba 75% wba 100% wba 102 advances in technology innovation, vol. 7, no. 2, 2022, pp. 92-104 conflicts of interest the authors declare no conflict of interest. references [1] h. d. craveiro, j. p. c. rodrigues, and l. laím, “buckling resistance of axially loaded cold-formed steel columns,” thin-walled structures, vol. 106, pp. 358-375, september 2016. [2] s. selvaraj and m. madhavan, “design of cold-formed steel built-up columns subjected to local-global interactive buckling using direct strength method,” thin-walled structures, vol. 159, article no. 107305, february 2021. [3] f. muftah, m. s. h. m. sani, and m. m. m. kamal, “flexural strength behaviour of bolted built-up cold-formed steel beam with outstand and extended stiffener,” international journal of steel structures, vol. 19, no. 3, pp. 719-732, september 2018. [4] y. li, t. zhou, l. zhang, j. ding, and x. zhang, “distortional buckling behaviour of cold-formed steel built-up closed section columns,” thin-walled structures, vol. 166, article no. 108069, september 2021. [5] s. f. nie, t. h. zhou, y. zhang, and b. liu, “compressive behavior of built-up closed box section columns consisting of two cold-formed steel channels,” thin-walled structures, vol. 151, article no. 106762, june 2020. [6] f. j. meza, j. becque, and i. hajirasouliha, “experimental study of cold-formed steel built-up columns,” thin-walled structures, vol. 149, article no. 106291, april 2020. [7] k. roy and j. b. p. lim, “numerical investigation into the buckling behaviour of face-to-face built-up cold-formed stainless steel channel sections under axial compression,” structures, vol. 20, pp. 42-73, august 2019. [8] q. y. li and b. young, “experimental and numerical investigation on cold-formed steel built-up section pin-ended columns,” thin-walled structures, vol. 170, article no. 108444, january 2022. [9] b. chen, k. roy, a. uzzaman, g. raftery, and j. b. p. lim, “axial strength of back-to-back cold-formed steel channels with edge-stiffened holes, un-stiffened holes and plain webs,” journal of constructional steel research, vol. 174, article no. 106313, august 2020. [10] m. s. h. m. sani, f. muftah, and c. s. tan, “investigation on mechanical performance of slender cold-formed steel channel column,” key engineering materials, vol. 706, pp. 107-111, august 2016. [11] m. rokilan and m. mahendran, “design of cold-formed steel columns subject to local buckling at elevated temperatures,” journal of constructional steel research, vol. 179, article no. 106539, april 2021. [12] c. ibanez, d. hernández-figueirido, and a. piquer, “effect of steel tube thickness on the behaviour of cfst columns: experimental tests and design assessment,” engineering structures, vol. 230, article no. 111687, january 2021. [13] a. zhu, x. zhang, h. zhu, j. zhu, and y. lu, “experimental study of concrete filled cold-formed steel tubular stub columns,” journal of constructional steel research, vol. 134, pp. 17-27, july 2017. [14] m. s. h. m. sani, f. muftah, m. f. muda, and c. s. tan, “resistance of built-up cold-formed steel channel columns filled with concrete,” jurnal teknologi, vol. 78, no. 5-2, pp. 99-104, may 2016. [15] x. qu, z. h. chen, and g. j. sun, “axial behaviour of rectangular concrete-filled cold-formed steel tubular columns with different loading methods,” steel and composite structures, vol. 18, no. 1, pp. 71-90, january 2015. [16] d. fayaz, a. malik, m. zakir, and t. a. malik, “effects on the properties of concrete by partial replacement of cement with carbon black and natural fine aggregate with robo sand,” international journal of technical innovation in modern engineering and science, vol. 4, no. 6, pp. 1-8, june 2018. [17] a. c. sankh, p. m. biradar, s. j. naghathan, and m. b. ishwargol, “recent trends in replacement of natural sand with different alternatives,” proceedings of the international conference on advances in engineering and technology, pp. 59-66, may 2014. [18] k. roy, t. c. h. ting, h. h. lau, r. masood, r. alyousef, h. alabduljabbar, et al., “cold-formed steel lipped channel section columns undergoing local-overall buckling interaction,” international journal of steel structures, vol. 21, no. 2, pp. 408-429, january 2021. [19] k. roy, c. mohammadjani, and j. b. p. lim, “experimental and numerical investigation into the behaviour of face-to-face built-up cold-formed steel channel sections under compression,” thin-walled structures, vol. 134, pp. 291-309, october 2018 [20] n. ferhoune and j. zeghiche, “experimental behaviour of concrete-filled rectangular thin welded steel stubs (compression load case),” comptes rendus mécanique, vol. 340, no. 3, pp. 156-164, 2012. [21] metallic materials—tensile testing. method of test at ambient temperature, bs en 10002-1:2001, 2001. 103 advances in technology innovation, vol. 7, no. 2, 2022, pp. 92-104 [22] m. s. h. m. sani and f. muftah, “experimental investigation of built-up cold-formed steel stub column with and without concrete-filled,” aip conference proceedings, vol. 2447, article no. 020001, august 2021. [23] p. b. dinis, b. young, and d. camotim, “local-distortional interaction in cold-formed steel rack-section columns,” thin-walled structures, vol. 81, pp. 185-194, august 2014. [24] y. h. kim, h. y. kim, k. h. yang, and j. s. ha, “effect of concrete unit weight on the mechanical properties of bottom ash aggregate concrete,” construction and building materials, vol. 273, article no. 121998, march 2021. [25] s. nie, t. zhou, m. r. eatherton, j. li, and y. zhang, “compressive behaviour of built-up double-box columns consisting of four cold-formed steel channels,” engineering structures, vol. 222, article no. 111133, november 2020. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). 104  advances in technology innovation, vol. 7, no. 4, 2022, pp. 279-294 learning representations for face recognition: a review from holistic to deep learning fabian barreto 1* , jignesh sarvaiya 2 , suprava patnaik 3 1 department of electronics and telecommunication, xavier institute of engineering, mumbai, india 2 department of electronics, sardar vallabhbhai national institute of technology, surat, india 3 school of electronics, kalinga institute of industrial technology, bhubaneswar, india received 18 august 2021; received in revised form 07 january 2022; accepted 08 january 2022 doi: https://doi.org/10.46604/aiti.2022.8308 abstract for decades, researchers have investigated how to recognize facial images. this study reviews the development of different face recognition (fr) methods, namely, holistic learning, handcrafted local feature learning, shallow learning, and deep learning (dl). with the development of methods, the accuracy of recognizing faces in the labeled faces in the wild (lfw) database has been increased. the accuracy of holistic learning is 60%, that of handcrafted local feature learning increases to 70%, and that of shallow learning is 86%. finally, dl achieves human-level performance (97% accuracy). this enhanced accuracy is caused by large datasets and graphics processing units (gpus) with massively parallel processing capabilities. furthermore, fr challenges and current research studies are discussed to understand future research directions. the results of this study show that presently the database of labeled faces in the wild has reached 99.85% accuracy. keywords: learning representations, deep learning, autoencoders, variational autoencoders 1. introduction in the modern world, automatic face recognition (afr) is embedded into smart e-commerce applets for better personalization and marketing of commodities, such as hair styling and digital makeup. consumer-based photography has become a new trend in selecting a range of products that suit consumers’ needs, with social media platforms providing facial recognition services to attract diverse users. conventional facial recognition (fr) requirements are limited to basic security and access control applications, and are implemented in more advanced ways. examples include accessing historical data and using cloud-based database identification and closed-circuit television (cctv) video-supported tracking, leading to better enforcement of the law. facial identification has become essential for forensics, surveillance, border control, lie detection, and access id verification. fr, in its various dimensions, is currently a research area in computer vision, and is the process of detecting and locating faces from a background, normalizing face images, and performing face verification (fv) or face identification (fi). there are two separate tasks for face matching while conducting fr, namely: fv and fi. in fv, one determines whether a given test image is from the same person being verified, while the fi aims to recognize the facial images of persons already enrolled in the database [1]. to verify genuineness, the output of fr is either “yes” or “no,” which may be a result of the class number corresponding to the input image. in the fv, the input image is assumed to be a sample from a known possible class of inputs. * corresponding author. e-mail address: frfabiansj@xavier.ac.in tel.: +919833916407 advances in technology innovation, vol. 7, no. 4, 2022, pp. 279-294 regarding face detection (fd), in 2001, viola and jones [2] used haar-like features to detect human faces. a 24 × 24 pixel can have over 160,000 haar-like features. the framework used the concept of integral images to perform intensive computation and the adaptive boost (adaboost) algorithm to select the best features from different subsets. wang et al. [3] categorized fd and recognition development into four broad representation learning types: holistic, handcrafted, shallow, and deep learning. traditionally, fr techniques have been divided into two major categories: geometric and photometric techniques. here, geometric techniques find distinct features and spatial positioning to form a template that is used to compare and eliminate variances in face images. photometric approaches are distilled out and use hidden statistical properties that account for the entire input of facial images. popular photometric approaches include principal component analysis (pca) using the eigenface algorithm and linear discrimination analysis (lda) using the fisherface algorithm. the holistic approach uses low-dimensional representations in the form of a manifold or a linear subspace. however, this approach is limited by variations such as face appearances that introduce different statistical distributions, which are difficult to manage. the early twentieth century saw a transition to handcrafted local feature-based methods. inherent face changes are managed through local descriptors, such as local binary patterns (lbps), gabor filters, and histograms of oriented gradients (hogs). these local features help remove redundant and meaningless information from raw representation; thus, they provide greater robustness than existing methods and are greatly invariant to transformation. the limitations of these approaches are that they suffer from a lack of compactness, distinctiveness over a large sample space, and acceptance for real-time applications, as well as being slow and susceptible to poor generalization. shallow representation learning, with a oneor two-layer representation, improved the distinctness of the codebook. noticeable shallow approaches included the learning-based (le) approach, discriminant face descriptor (dfd), feature vector, and pcanet. however, these approaches were not robust to the complex non-linear nature of the face. deep learning (dl) is a revolutionary approach that has changed the facial recognition landscape. in 2012, alexnet achieved state-of-the-art (sota) recognition accuracy and propelled research toward dl for computer vision. researchers have used a convolutional neural network (cnn) that exhibited strong invariance to face pose, lighting, expression, and other variations to achieve high accuracy. thus, this research addresses recognition accuracy and investigates the complexity of learning a large number of features, dependency on datasets, protocols addressing application scenarios, and model interpretability. this research also addresses variations encountered owing to cross-posed, aging, and other adversarial conditions. since the 1990s, remarkable advances have been made in fd and recognition. this study aims to review the development of learning representations for fr in the past three decades and has resulted in an accuracy increase of 39.85% for labeled faces in the wild (lfw) database from the earlier methods used three decades ago. the remainder of this study is organized as follows. section 2 describes the initial holistic representation of the learning stage for fr, and section 3 describes the transition to a handcrafted stage. section 4 presents the shallow learning phase, and section 5 deals with the dl phase and some challenges and current research studies. finally, section 6 provides the conclusions of this study. 2. review of holistic learning the earliest holistic stage begins by using eigenfaces, motivated by sirovich and kirby [4], to efficiently represent face images using pca. they then transition to fisherface algorithms and lda and later to independent component analysis (ica), leading to sparse representation-based classification (src), a particular case of collaborative representation-based classification (crc). later, researchers used distance metric learning with improved class separability, meaning that the holistic stage can assume certain distributions (linear, manifold, and sparse) from which it arrives at a low-dimensional representation. however, these assumptions do not hold firm ground on the variations in facial features. 280 advances in technology innovation, vol. 7, no. 4, 2022, pp. 279-294 2.1. principal component analysis (pca) ballantyne et al. [5] mentioned the pioneering work of woody bledsoe and his afr team. they manually classified face images with landmarks (e.g., eye centers and mouth) and saved the metrics in a database. goldstein et al. [6] enhanced the accuracy by using 21 specific subjective markers on the face. the work of turk and pentland [7] in 1991 gave a new direction to using eigenfaces (pca) to develop the first afr system. varying the illumination and pose conditions is a challenging task for this method. it is essential to understand that a particular eigenfeature may not be related to recognition, but to the direction of illumination. hence, an increase in eigenfeatures does not necessarily lead to better accuracy. pca can only set apart the linear dependencies in the pixel pair of a facial image. pca is a method for expressing data vectors in their principal components (pcs), where the largest variances in the data indicated the direction of the pcs (fig. 1). pcs capture the most significant data information and correspond to the eigenvectors given by the largest eigenvalues of the autocorrelation matrix of the data vectors. pca computes the most representational basis for looking at the dataset and generally works as follows. first, it calculates the covariance matrix of the given data points and calculates the eigenvectors and corresponding eigenvalues sorted in decreasing order. then, the first k eigenvectors are chosen from the n eigenvectors (k < n), yielding the novel k dimensions. thus, the original n higher dimensions were transformed into k fewer dimensions. fig. 1 original space (x1, x2) and pca reduced space (pc1, pc2) 2.2. linear discrimination analysis (lda) lda constructs a subspace that differentiates between different face images, while fisher discriminant analysis classifies face images into groups based on their facial features. zhao et al. [8] used lda for fr because it encodes discriminatory information. they used pca to project the face image to a subspace and used an lda to obtain a linear classifier in the subspace. the pure lda approach, however, does lead to an overfitting problem and does not perform well for samples from different classes and samples with diverse backgrounds. 2.3. independent component analysis (ica) ica describes a subspace method that transforms data from high to low dimensions. it finds a linear transformation that leads to the minimization of the statistical dependence between its components. however, unlike pca, it provides an improved probabilistic model, a greater response to high-order statistics, and better reconstruction in noisy environments [9]. a set of statistically independent basis images for a set of face images is found by separating the independent components of the facial images (fig. 2). here, let s be a set of statistically independent source images, which is unknown, with x as the source of the face images and a as an unknown combination matrix. wi is a matrix of learned filters which in turn produces outputs u that are statistically independent. ica outputs in rows that are wix = u. 281 advances in technology innovation, vol. 7, no. 4, 2022, pp. 279-294 fig. 2 image synthesis model 2.4. hidden markov model (hmm) in a hidden markov model (hmm), patterns are characterized as parametric random processes. these parameters can be estimated precisely and logically. samaria et al. [10] used the hmm model to represent the statistics of facial images. they converted a two-dimensional face image to a one-dimensional sequence. as shown in fig. 3, the face is split into regions (e.g., the forehead, eyes, nose, mouth, and chin). after determining the hidden states (five in the given figure), the hmm is trained to learn the state transitional probability. after training on the output probability, the class was determined. although hmm has a better detection rate, it also has a higher false-alarm rate. fig. 3 five-state hmm 2.5. bayesian model schneiderman et al. [11] derived a probabilistic model for fr using local regions, such as the eyes, nose, and mouth. their statistical model captured the more unique patterns of the human face, such as the intensity patterns around the eye, to represent the local features more uniquely. they also modelled the joint probability of local features and positions, as human faces are easily recognized because of their proper spatial arrangement. they used the bayesian decision rule, also known as maximum a posteriori (map), and calculated a larger probability for a given input image x, namely, p(face | x) or p(not face | x), indicating whether a face was selected. yang et al. [12] presented two advantages of using a naive bayes classifier; that is, it provided a better estimation of the subregion conditional density functions and provided an map to understand the joint statistics of a local feature and its position. 2.6. locality preserving projection (lpp) he et al. [13] proposed an appearance-based laplacian method for facial recognition by using locality preserving projections (lpps) to map facial images into a subspace. eigenfaces (pca) preserve the global surface of the face image, whereas the fisherface algorithm (lda) preserves discriminating information. the advantage of lpp over pca and lda is 282 advances in technology innovation, vol. 7, no. 4, 2022, pp. 279-294 that it preserves local features and detects the essential face manifold surface, where the nearest-neighbor graph models this surface. the face images in the lower-dimensional subspaces are called laplacian faces. facial recognition was performed in three steps. laplacian faces were calculated from the given training face image samples, and the test image is then projected onto the laplacian face subspace. finally, the nearest-neighbor classifier identifies a new face. as this method considers the face manifold, it considers varying illumination conditions. 2.7. sparse representation-based classification (src) and collaborative representation-based classification (crc) src and crc belong to sparse representation-based classifiers. the test input image was a linear connection between the recorded images. the test image can be recognized as the combination coefficients for the target faces, which are larger than the others. in src/crc, the test face images are coded over others with sparsity constraints, such as l1 minimization. src/crc uses the reconstruction error to determine the face image. in the work of wright et al. [14], the discriminative property of an src model for classification was used, while in the work of zhang et al. [15] and zhang et al. [16], it was shown that the good performance of src is primarily due to the collaborative representation of the test face image with training samples across different classes. 2.8. distance metric learning in distance metric learning, one learns a distance metric for the input space of face images from a given set of similar/dissimilar points in the training face images. yang et al. [17] categorized the algorithms for distance metric learning into supervised and unsupervised methods. supervised training face images are placed into pairwise constraints: pairs of same-class data points in the equivalence constraints and those that belong to different classes in equivalence constraints. supervised learning can be global or local, where global satisfies pairwise constraints simultaneously and local only meets local pairwise constraints. supervised learning includes supervised global learning, local adaptive supervised learning, neighborhood component analysis, and relevant component analysis (rca), while unsupervised learning includes linear-like pca and multidimensional scaling. they also include nonlinear embedding methods such as isometric mapping, linear embedding, and laplacian eigenmaps. jin et al. [18] presented a regularized distance metric learning algorithm that is robust for high-dimensional data. here, the generalization error of regularized distance metric learning is independent of dimensionality. the algorithm was tested with the baselines of the euclidean distance metric, mahalanobis distance metric, large margin nearest neighbor classifier, information-theoretic metric learning, and rca and was comparable to sota approaches for distance learning. 3. review of handcrafted local feature learning to enhance the holistic method, researchers started using handcrafted local features. they used gabor wavelets, elastic bunch graph matching (ebgm), local binary patterns (lbp), and high dimensional local binary patterns (hd-lbp). these methods did achieve robust performance. however, as the features increased, there was a problem of distinctiveness, and the large size created the problem of non-compactness. 3.1. gabor wavelet (filter) gabor introduced the gabor wavelet (or gabor filter) in 1946 as a band-pass filter and has an impulse response given by a gaussian function, multiplied by a harmonic function. its resolution is optimal in both the domains of space and frequency. daugman [19] generalized the 1-d gabor filters to two-dimensional gabor filters. liu et al. [20] described a facial recognition gabor feature classifier where gabor wavelets first transform the face images to obtain the augmented gabor fv and then pass through an enhanced fisher discrimination model. their results showed that the classifier can discriminate gabor features with 283 advances in technology innovation, vol. 7, no. 4, 2022, pp. 279-294 low dimensionality and increased discrimination. barbu [21] proposed a 2-d gabor filter for human fr. he used 2-d gabor filter banks, which help extract different orientation and scale features from the input face image, resulting in 3-d face feature vectors. one disadvantage is that gabor features have high dimensionality and result in redundancy [22]. a hybrid method uses gabor filters and another technique such as pca to reduce redundancy. principal gabor filters that help reduce redundancy are described in the work of štruc et al. [23]. here, they used orthonormal linear combinations and derived a gabor face representation. however, the tradeoff is that the filters are not optimally localized in the space and frequency domains. 3.2. local binary pattern (lbp) the human face can be viewed as consisting of micro-patterns and hence can use an lbp as a face descriptor [24-25]. lbp was first proposed for texture description [26], where it was observed that certain lbp are key properties of texture and sometimes represent over 90% of all 3 × 3 patterns present in the textures. after thresholding, a histogram that functions as a texture descriptor can be created (fig. 4). these patterns have uniform circular structures with few spatial transitions and were used as templates. the lbp operator is only a 3 × 3 neighborhood; therefore, it is difficult to capture the features that are dominant for large-scale structures, with later models using neighborhoods of different sizes to correct this issue. lbp efficiently summarizes the local structures of facial images, where each pixel was compared with its neighboring pixels. an example is shown in fig. 5. here, each pixel is compared with its eight neighbors by subtracting the center pixel value. the encoding process is done in the following steps. encode a 0 for negative; otherwise, encode a 1. concatenate all binary values in a clockwise direction. begin from the top-left neighbor and move clockwise. convert the binary to a decimal value, the label (lbp codes) for the given pixel. lbp is a non-parametric method that converts the face image into an array of integer labels. huang et al. [27] surveyed lbp and its variants that offer better performance and improved the robustness of the original lbp. isnanto et al. [28] used lbp and haar cascade classifier on low-resolution images for multi-object fr. fig. 4 lbp histogram fig. 5 lbp operator 3.3. elastic bunch graph matching (ebgm) bolme [29] described the ebgm fr algorithm. it recognizes new facial images by localizing landmark features and then finds the similarity measure. facial landmark points were selected manually from a set of model face images with variations. gabor jets are the names given to the gabor wavelets extracted from the landmark point and the jets from the model form a face bunch graph. each node contains a stack of n jets (n = model image). here, the edge is the distance 284 advances in technology innovation, vol. 7, no. 4, 2022, pp. 279-294 between landmark points (fig. 6). the limitation of the ebgm is that one needs to rely on the model’s manual ground truth for landmark selection at the initial recognition stage. lahasan et al. [30] proposed a method to overcome this shortcoming by posing the ebgm as an optimization problem by using harmony search (hs) to find the optimal facial landmarks using the manual method. fig. 6 ebgm process 3.4. scale-invariant feature transform (sift) scale-invariant feature transform sift was proposed by lowe [32-33]. it creates descriptors that are scale-, rotation-, and translation-invariant and possesses high dimensionality. fr tasks use sift features [33-34] to reliably match images. this process includes extracting sift keypoints from the face image. how can one find the test image? by finding the matching features. the euclidean distance was used as the measure; however, a challenge is the reliable extraction of consistent sift descriptors. as shown in fig. 7, the sift algorithm has four stages: keypoint detection, keypoint localization, orientation assignment, and keypoint descriptor generation. keypoint detection uses the difference of the gaussian (dog) function to detect feature points, and each keypoint is assigned one or more orientations during the orientation assignment stage. in the last stage, each keypoint is assigned to a vector descriptor. given that the algorithm is computationally intensive, the actions are performed only at positions that go through the first test. fig. 8 shows the sift features of a 64 × 64 image, its noisy version, and matching features. fig. 7 stages of the sift algorithm (a) sift keypoints of the original image (64 × 64) (b) with added noise (c) sift keypoint matching fig. 8 implementation of sift 285 advances in technology innovation, vol. 7, no. 4, 2022, pp. 279-294 3.5. histogram of oriented gradient (hog) dalal et al. [35] developed grids of hog descriptors, which have the advantage of capturing the gradient (edge) structure, a characteristic of the local shape. the grids count the occurrence of edge orientations in the local neighborhood of the face image. facial images were split into small and linked regions (cells), and a histogram of the edge orientations was computed for each cell. the histograms were normalized to account for the illumination and combined to form the hog descriptor. the hog is invariant to 2d rotation and scaling. using locally normalized hog features with an overlapping dense grid yielded better results. déniz et al. [36] proposed a method for building a robust hog descriptor by using a regular grid, combining hog descriptors at different scales, and applying a reduction in linear dimensions. 4. review of shallow learning the shallow learning-based (le) local descriptor phase uses local filters to learn distinctiveness and a codebook to achieve compactness. as this was a shallow representation, a oneor two-layer representation, it is not robust to the complex nonlinearity of face images. the method also improves one characteristic, such as pose, light, or expression, but does not address unconstrained changes in the face image in general. 4.1. learning-based (le) cao et al. [37] proposed a new le descriptor that was compact, discriminative, and easy to extract. they list the disadvantages of existing handcrafted methods, as it is challenging to obtain an optimal encoding and unevenly distributed. their process consisted of extracting face landmarks that aligned nine different parts of the face separately, which were fed into the dog filter to remove lowand high-frequency illumination variations. each pixel has a low-level fv encoded by an le encoder. pca-reduced histograms were concatenated and then normalized to obtain the le descriptor, and the similarity of the le descriptors of the face pair was measured using the l2 distance norm. the nine component similarity scores were then fed into a pose-adaptive classifier, which resulted in fv. 4.2. discriminant face descriptor (dfd) lei et al. [38] described a technique for acquiring a dfd. discriminant local features learn by minimizing the feature differences between the same face images and maximizing those between different face images. the discriminative capability is performed in three steps: learning discriminant image filters, determining the optimal neighborhood sampling, and constructing the dominant patterns. they also used coupled dfds to view heterogeneous facial data. 4.3. feature vector sánchez et al. [39] described the feature vector method for image classification based on the principle of gaussian mixture distribution. they proposed using the fisher kernel framework and described their blocks by deviation from a gaussian mixture distribution with diagonal covariance. visual vocabulary is a gradient vector for the model parameters. their method encoded the (probabilistic) count of occurrences and higher-order statistics. the authors listed the advantages of their method as having better results than efficient linear classifiers and compression with a very low loss of accuracy. 4.4. pcanet chan et al. [40] proposed a baseline model for image classification called pcanet, a precursor to dl models. pcanet consists of cascaded pca to learn from multistage filter banks, binary hashing, and blockwise histograms and has two variations: randnet and ldanet. in randnet, they replaced pca filters with random filters of the same size at each layer, whereas in ldanet, the supervision of a classification problem was improved by using supervised training. lda is used to 286 advances in technology innovation, vol. 7, no. 4, 2022, pp. 279-294 learn the filters. pcanet eliminated image variability and provided effective accuracy with well-preprocessed images in the datasets. however, pcanet may not sufficiently account for the variability of challenging face images. however, the pcanet is a valuable baseline for studying dl architectures. 5. review of deep learning the fr landscape saw a fundamental shift with the introduction of alexnet, which uses dl. deepface [41], deepid [42-43], facenet [44], arcface [45], and adaptiveface [46] have paved the way for an evolution of network architectures, algorithms, and datasets to answer the multi-faceted fr problem. the accuracy results for the lfw database [47] explain the fr development stages. for the holistic stage, the accuracy was 60%, while for handcrafted, it increased to 70%, shallow to 86%, and finally, for dl, especially for deepface, it approached human-level performance of 97% for the unconstrained fr. in the early days of the afr, the focus was more on developing fd algorithms and less on developing face image datasets. there has been organic growth in the datasets over the past two decades because it has come from the research community in terms of the need for a large number of face images with varying conditions and diversity. another development has been the challenge to go beyond recognizing faces from laboratory-controlled to unconstrained face images. afr research has progressed enormously, with some simple datasets achieving 99% accuracy, which has resulted in the development of more complex datasets that can facilitate new directions for fr research. the number of face images in the datasets and their variations has increased over the years. the past decade with fr research moving toward dl approaches has resulted in the growth of large training datasets required to implement dl algorithms effectively. taskiran et al. [48] classified face image datasets as image-based or video-based. they may also be 3d or hyperspectral/infrared datasets. some of the datasets were private, whereas others were public. these datasets are essential for benchmarking new afr algorithms. a database’s choice depends on the given problem that one intends to solve or a property that one wants to test and also depends on the size of the training set required to test the algorithm. some databases, such as facebook, google, celebfaces+, and vggface, were used for training, and others, such as lfw, ytf, and ijb-c, were used for testing. 5.1. artificial intelligence (ai), machine learning (ml), and deep learning (dl) john mccarthy, the father of artificial intelligence (ai), coined the term ai in his 1955 proposal for the dartmouth conference, usa, in 1956. on a broader scale, ai explores theories and applications to broaden human intelligence and envisions the creation of a future where intelligent machines have human-like perception and cognition. researchers have made significant progress in understanding and improving learning algorithms; however, the challenge of ai remains [49]. as shown in fig. 9, dl is a subfield of machine learning (ml), and ml is a subset of the broader field of ai. some examples of ml problems include classification, clustering, and prediction. traditional ml techniques are constrained to process data in a basic form and domain experts are required to carefully perform feature extraction [50]. dl is a subset of the ml and learns multiple representations and abstraction levels to understand the data. the raw input was transformed to a higher and more abstract level (fig. 10). these transformations can help learn complex and intricate functions. fig. 9 relationship of ai, ml, and dl 287 advances in technology innovation, vol. 7, no. 4, 2022, pp. 279-294 (a) ml (b) dl fig. 10 ml and dl approaches 5.2. artificial neural network (ann) the unique human brain, especially how neurons interact, has inspired scientists. artificial neural networks (anns) are hardware and software implementations of neural structures in the human brain. the history of neural computing originated with the work of mcculloch and pitts in 1943. the warren mcculloch and walter pitts model (mcp model, known as the linear threshold gate model) is a binary classifier [51], where the weights were manually adjusted by a human. in the 1950s, rosenblatt published the perceptron algorithm, which automatically learns weights without human involvement [52]. this was an enhanced version of the mcp model. the perceptron model adds extra information representing the bias and variable weight values. the 1969 publication by minsky and papert [53] weakened neural network research for nearly a decade (1969-1986). they believed that using perceptrons in practical applications was futile without an adequate basic theory. in 1979, fukushima developed a neural network with multiple pooling and convolutional layers called neocognitron, which used a hierarchical and multilayered design that learned how to recognize visual patterns [54]. rumelhart revived neural network research in 1986 using a backpropagation (bp) algorithm. the neural network iteratively learns weights that are then used to predict class labels. given sufficient hidden units and sufficient training data multilayers, feedforward networks can closely approximate any function. in 1989, yann lecun demonstrated bp at the bell labs. he combined cnns with bp to read handwritten digits. in 1997, long short-term memory for recurrent neural networks (rnns) was developed by hochreiter and schmidhuber, with a gating mechanism to regulate the information to be kept or discarded at each time step. 5.3. the deep learning phase in 2009, fei-fei li launched the challenging benchmark dataset, imagenet [55]. between 2011 and 2012, krizhevsky created alexnet, a cnn. as shown in fig. 11, alexnet has five convolutional layers, followed by max-pooling layers and three fully connected layers. instead of using tanh and sigmoid activation functions, he used rectified linear units (relus), which increased the speed and dropout. alexnet showed that a greater depth resulted in high performance and, despite being computationally expensive, is feasible because of graphics processing units (gpus). in 2014, deepface used neural networks to identify faces from the lfw dataset with 97.35% accuracy, an improvement of 27% over previous efforts [41]. in 2015, the facenet model, using googlenet-24, achieved 99.63% accuracy for the google dataset [44]. in 2018, ring loss model using resnet-64 achieved 99.5% accuracy for the ms-celeb dataset [56] and arcface model using resnet-100 achieved 99.83% accuracy for the ms-celeb dataset [45]. in the work of yan et al. [57], the use of vargfacenet resulted in an accuracy of 99.85% for the lfw database. fig. 11 alexnet architecture 288 advances in technology innovation, vol. 7, no. 4, 2022, pp. 279-294 fig. 12 autoencoder model fig. 13 variational autoencoder the evolution of dl is described in detail by schmidhuber [58]. he explains the hierarchical representation learning for different supervised/reinforcement learning and the various advancements in both feedforward (acyclic) neural networks (fnns) and recurrent (cyclic) neural networks (rnns). he also described the evolution of restricted boltzmann machines (rbms), as well as the constituents of multilayer learning architectures, such as the deep belief networks (dbns). advances in dl meant working with high dimensional data, which could be reduced to codes of lower dimensionality. in 2006, hinton and salakhutdinov [59] trained an “autoencoder” network. autoencoders [60] are used for dimensionality reduction, denoising, and outlier detection and are made up of three sections, as shown in fig. 12. the encoder encodes the data to the hidden layer (code) which results in an output h = f(x). the decoder then outputs r = g(h). the training minimizes a mean squared error loss function. deep autoencoders use numerous internal intermediate representations, and these deep layers help learn more intricate and complex data patterns. convolutional autoencoder (caes) [60] helps integrate the convolutional advantage of a cnn. the encoder is thus made up of convolutional layers and the decoder of deconvolutional layers. thus, caes extract features and gives a feature map containing the image’s significant points. one limitation of an autoencoder is that it has a deterministic latent-space representation. although the autoencoder learns the input data, it may lack relevant information, which may be due to random encoding in the latent space or empty space. to overcome this, kingma et al. [61] suggested a variational autoencoder (vae), as shown in fig. 13, which uses a probability distribution for latent space code representation. an inference model q(z | x) for vae is described in [62]. here,  denotes the variational parameters, optimized for q(z | x)  p(x | z). here, q(z | x) approximates the posterior p(z | x) of the generative model and is optimized using the evidence lower bound (elbo) [63]. in 2014, goodfellow et al. [64] introduced generative adversarial networks (gans) as well as an adversarial network framework. a generative model is matched against a competitor, which they call a discriminative model, and the latter learns to determine whether the query face image is from the model distribution or given data distribution [64]. both thrive on competition to improve their methods till one cannot be distinguished from the other. 5.4. some current research in dl for fr developing different deep fr methods and their deployment in real-world applications requires a systematic performance evaluation. iandola et al. [65] provided an evaluation framework for different datasets and sota methods. they used the following criteria: data augmentation, network architecture, loss function, training strategy, and model compression. the varied sizes of the datasets, such as casia-webface, vgg-face, ms-celeb-1m, and megaface for training and lfw and ytf for testing the models, make comparisons difficult. here, both the datasets and architectures vary. a critical part of the evaluation is the loss function, which imposes stricter requirements for fr, as it has to discriminate and separate the features from the embedding space. the training strategy also plays an important role in terms of the learning rate and batch size. with 289 advances in technology innovation, vol. 7, no. 4, 2022, pp. 279-294 the modern trend of using fr in mobile and embedded devices, they also evaluated squeezenet [66] and mobilenet [67], which use compressed models and give better performance. they concluded that the deep resnet series has advantages over other architectures, and the batch and feature normalization optimizes performance. deployment of fr models, especially unconstrained faces on embedded or mobile devices, needs to meet the challenge of recognizing low-resolution faces at a low computational cost. this problem is addressed in the work of ge et al. [68] by using the selective knowledge distillation approach and calling it the teacher-student model. they used a two-stream cnn, one with high resolution (hr), which collected the essential facial features used to tune the other lr network using regression and classification. li et al. [69] also take on the challenging task of working with low-resolution unconstrained face images. they explore good-performing models using the scface [70] and uccsface [71] datasets. to visually learn the network, they pre-train it with dcgan [72]. new trends for unconstrained, very low-resolution fr were explored in [73]. they present a classification of very low-resolution fr approaches, characterizing them as heterogeneous or homogeneous based on their belongingness to different or same domains, respectively. the heterogeneous approach can be classified into projection (coupled mapping) and synthesis (super-resolution (sr)) methods. in a homogeneous approach, they discussed lightweight ccns. they listed the challenges for very low-resolution fr as the availability of datasets for real-world applications, the dearth of discriminative features, discrepancies in the domain, and the efficiency of existing solutions. one of the challenges in fr is the development of a pipeline that can simultaneously perform fd, alignment, and recognition. other parameters, such as pose and gender, may also be required in some instances. a cnn pipeline for the different processes is described by ranjan et al. [74]. they use a deep pyramid single-shot face detector (dpssd) and a new loss function called crystal loss. they evaluated their end-to-end system on the iarpa janus benchmarks ijb-a [75], ijb-b [76], ijb-c [77], and iarpa janus challenge set 5 (cs5) datasets to obtain sota performance. they also mentioned that some of the challenges facing current fr systems are dataset bias and domain adaptation. in mid-march 2020, the world health organization (who) declared the coronavirus disease 2019 (covid-19) be a pandemic [78]. dl has been extensively used in the analysis of the covid-19 pandemic, as elaborated in the work of heidariet al. [79], for disease prediction, disease monitoring, drug testing, and vaccine development. who issued guidelines for wearing a mask to prevent the transmission of the disease. abboah-offei et al. [80] provided a detailed analysis of facemasks to control the transmission of respiratory viral infections, and the french government tested ai-based cctv software to detect whether travelers wore masks or not [81]. the fr research community is engaged in developing systems to monitor the facemasks worn by people. fig. 14 depicts a block diagram of face mask detection using ml or dl. mbunge et al. [82] and nowrin et al. [83] provide a comprehensive review of mland dl-based facemask detection techniques. most of the facemask detection algorithms are cnn-based. a few are hybrid as they use dl and ml approaches like support vector machine (svm) and decision tree (dt). cnn-based models include mobilenetv2 [84], resnet [85], and vgg-16 cnn [86]. mobilenet and resnet perform better than vgg-16 cnn. mobilenetv2 exhibits better performance because it is a lightweight classifier. srcnet [87] uses an sr network and a classification network to perform three-class classification with an accuracy of 98.7%. facemasknet [88], a three-class classifier, has an accuracy of 98.6%. retinafacemask [89], which uses both resnet and mobilenet, incorporates transfer learning to achieve sota results. some challenges for face mask detection are elaborated in the work of nowrin et al. [83]. these include the availability of benchmarked datasets, variation in mask designs, processing speed for real-time applications, and variations in image resolution and masked face reconstruction. fig. 14 face mask detection block diagram 290 advances in technology innovation, vol. 7, no. 4, 2022, pp. 279-294 6. conclusions this study reviewed the vast literature on the development of different approaches for afr. over time, a transition from shallow to modern sota methodologies for dl has been observed. early fr methods used limited images and a laboratory-controlled environment. however, with the advent of dl models, the lfw database achieved 99.85% accuracy. this was possible because of gpus’ massively parallel processing capabilities and large training and testing datasets. the challenges faced by dl models were also examined. as networks deepen, the complexity of the deep convolutional neural network (dcnn) model increases. a deep autoencoder or vae that preserves some interclass discrimination information and intraclass similarity can feed a dcnn with a lower complexity to reduce the overall dcnn complexity. the performance decreases when the images have low resolution, variations in illumination, and blurry quality. hence, dl methods must be made more robust under adverse conditions. the advent of new mobile communication technologies presents the challenge of integrating personalized fr applications that can be accessed by mobile users over different clouds and networks. conflicts of interest the authors declare no conflicts of interest. references [1] g. guo, et al., “a survey on deep learning based face recognition,” computer vision and image understanding, vol. 189, article no. 102805, december 2019. [2] p. viola, et al., “rapid object detection using a boosted cascade of simple features,” ieee computer society conference on computer vision and pattern recognition, pp. 1-9, december 2001. [3] m. wang, et al., “deep face recognition: a survey,” https://arxiv.org/pdf/1804.06655v2.pdf, april 2018. [4] l. sirovich, et al., “low-dimensional procedure for the characterization of human faces,” journal of the optical society of america a, vol. 4, pp. 519-524, march 1987. [5] m. ballantyne, et al., “woody bledsoe: his life and legacy,” ai magazine, vol. 17, no. 1, pp. 7-20, 1996. [6] a. j. goldstein, et al., “man-machine interaction in human-face identification,” bell system technical journal, vol. 51, no. 2, pp. 399-427, 1972. [7] m. a. turk, et al., “face recognition using eigenfaces,” ieee computer society conference on computer vision and pattern recognition, pp. 586-587, january 1991. [8] w. zhao, et al., “subspace linear discriminant analysis for face recognition,” http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.7.6280&rep=rep1&type=pdf, april 1999. [9] m. s. bartlett, et al., “face recognition by independent component analysis,” ieee transactions on neural networks, vol. 13, no. 6, pp. 1450-1464, december 2002. [10] f. samaria, et al., “hmm-based architecture for face identification,” image and vision computing, vol. 12, no. 8, pp. 537-543, october 1994. [11] h. schneiderman, et al., “probabilistic modeling of local appearance and spatial relationships for object recognition,” ieee computer society conference on computer vision and pattern recognition, pp. 45-51, june 1998. [12] m. h. yang, et al., “detecting faces in images: a survey,” ieee transactions on pattern analysis and machine intelligence, vol. 24, no. 1, pp. 34-58, august 2002. [13] x. he, et al., “face recognition using laplacianfaces,” ieee transactions on pattern analysis and machine intelligence, vol. 27, no. 3, pp. 328-340, january 2005. [14] j. wright, et al., “robust face recognition via sparse representation,” ieee transactions on pattern analysis and machine intelligence, vol. 31, no. 2, pp. 210-227, april 2008. [15] l. zhang, et al., “sparse representation or collaborative representation: which helps face recognition?” international conference on computer vision, pp. 471-478, november 2011. [16] l. zhang, et al., “collaborative representation based classification for face recognition,” https://arxiv.org/vc/arxiv/papers/1204/1204.2358v1.pdf, april 2012. [17] l. yang, et al., “distance metric learning: a comprehensive survey,” https://www.cs.cmu.edu/~liuy/frame_survey_v2.pdf, may 2006. 291 advances in technology innovation, vol. 7, no. 4, 2022, pp. 279-294 [18] r. jin, et al., “regularized distance metric learning: theory and algorithm,” advances in neural information processing systems, vol. 22, pp. 862-870, december 2009. [19] j. g. daugman, “two-dimensional spectral analysis of cortical receptive field profiles,” vision research, vol. 20, no. 10, pp. 847-856, january 1980. [20] c. liu, et al., “a gabor feature classifier for face recognition,” 8th ieee international conference on computer vision, pp. 270-275, july 2001. [21] t. barbu, “gabor filter-based face recognition technique,” proceedings of the romanian academy, vol. 11, no. 3, pp. 277-283, march 2010. [22] m. yang, et al., “gabor feature based sparse representation for face recognition with gabor occlusion dictionary,” european conference on computer vision, pp. 448-461, september 2010. [23] v. štruc, et al., “principal gabor filters for face recognition,” 3rd international conference on biometrics: theory, applications, and systems, pp. 1-6, september 2009. [24] t. ahonen, et al., “face recognition with local binary patterns,” european conference on computer vision, pp. 469-481, may 2004. [25] t. ahonen, et al., “face description with local binary patterns: application to face recognition,” ieee transactions on pattern analysis and machine intelligence, vol. 28, no. 12, pp. 2037-2041, october 2006. [26] t. ojala, et al., “a comparative study of texture measures with classification based on featured distributions,” pattern recognition, vol. 29, no. 1, pp. 51-59, january 1996. [27] d. huang, et al., “local binary patterns and its application to facial image analysis: a survey,” ieee transactions on systems, man, and cybernetics, part c (applications and reviews), vol. 41, no. 6, pp. 765-781, march 2011. [28] r. r. isnanto, et al., “multi-object face recognition using local binary pattern histogram and haar cascade classifier on low-resolution images,” international journal of engineering and technology innovation, vol. 11, no. 1, pp. 45-58, january 2021. [29] d. s. bolme, “elastic bunch graph matching,” master thesis, department of computer science, colorado state university, co, 2003. [30] b. m. lahasan, et al., “recognizing faces prone to occlusions and common variations using optimal face subgraphs,” applied mathematics and computation, vol. 283, pp. 316-332, june 2016. [31] d. g. lowe, “object recognition from local scale-invariant features,” 7th ieee international conference on computer vision, pp. 1150-1157, september 1999. [32] d. g. lowe, “distinctive image features from scale-invariant keypoints,” international journal of computer vision, vol. 60, no. 2, pp. 91-110, november 2004. [33] j. luo, et al., “person-specific sift features for face recognition,” ieee international conference on acoustics, speech, and signal processing, pp. 593-596, april 2007. [34] c. geng, et al., “face recognition using sift features,” 16th ieee international conference on image processing, pp. 3313-3316, november 2009. [35] n. dalal, et al., “histograms of oriented gradients for human detection,” ieee computer society conference on computer vision and pattern recognition, pp. 886-893, june 2005. [36] o. déniz, et al., “face recognition using histograms of oriented gradients,” pattern recognition letters, vol. 32, no. 12, pp. 1598-1603, september 2011. [37] z. cao, et al., “face recognition with learning-based descriptor,” ieee computer society conference on computer vision and pattern recognition, pp. 2707-2714, june 2010. [38] z. lei, et al., “learning discriminant face descriptor,” ieee transactions on pattern analysis and machine intelligence, vol. 36, no. 2, pp. 289-302, june 2013. [39] j. sánchez, et al., “image classification with the fisher vector: theory and practice,” international journal of computer vision, vol. 105, no. 3, pp. 222-245, december 2013. [40] t. h. chan, et al., “pcanet: a simple deep learning baseline for image classification?” ieee transactions on image processing, vol. 24, no. 12, pp. 5017-5032, september 2015. [41] y. taigman, et al., “deepface: closing the gap to human-level performance in face verification,” proceedings of the ieee conference on computer vision and pattern recognition, pp. 1701-1708, june 2014. [42] y. sun, et al., “deep learning face representation by joint identification-verification,” advances in neural information processing systems, pp. 1988-1996, december 2014. [43] y. sun, et al., “deepid3: face recognition with very deep neural networks,” https://arxiv.org/pdf/1502.00873.pdf, february 2015. 292 advances in technology innovation, vol. 7, no. 4, 2022, pp. 279-294 [44] f. schroff, et al., “facenet: a unified embedding for face recognition and clustering,” proceedings of the ieee conference on computer vision and pattern recognition, pp. 815-823, june 2015. [45] j. deng, et al., “arcface: additive angular margin loss for deep face recognition,” proceedings of the ieee/cvf conference on computer vision and pattern recognition, pp. 4690-4699, june 2019. [46] h. liu, et al., “adaptiveface: adaptive margin and sampling for face recognition,” proceedings of the ieee/cvf conference on computer vision and pattern recognition, pp. 11947-11956, june 2019. [47] g. b. huang, et al., “labeled faces in the wild: a database for studying face recognition in unconstrained environments,” workshop on faces in real-life images: detection, alignment, and recognition, pp. 1-11, october 2008. [48] m. taskiran, et al., “face recognition: past, present and future (a review),” digital signal processing, vol. 106, article no. 102809, november 2020. [49] y. bengio, learning deep architectures for ai, hanover: now publishers, 2009. [50] y. bengio, et al., “representation learning: a review and new perspectives,” ieee transactions on pattern analysis and machine intelligence, vol. 35, no. 8, pp. 1798-1828, march 2013. [51] s. hayman, “the mcculloch-pitts model,” international joint conference on neural networks, pp. 4438-4439, july 1999. [52] f. rosenblatt, “the perceptron: a probabilistic model for information storage and organization in the brain,” psychological review, vol. 65, no. 6, pp. 386-408, november 1958. [53] m. minsky, et al., perceptrons, cambridge: mit press, 1969. [54] k. fukushima, “neocognitron: a self-organizing neural network model for a mechanism of pattern recognition,” proceedings of the u.s.-japan joint seminar, pp. 267-285, february 1982. [55] a. krizhevsky, et al., “imagenet classification with deep convolutional neural networks,” advances in neural information processing systems, vol. 25, pp. 1097-1105, december 2012. [56] y. zheng, et al., “ring loss: convex feature normalization for face recognition,” proceedings of the ieee conference on computer vision and pattern recognition, pp. 5089-5097, june 2018. [57] m. yan, et al., “vargfacenet: an efficient variable group convolutional neural network for lightweight face recognition,” proceedings of the ieee/cvf international conference on computer vision workshops, pp. 2647-2654, october 2019. [58] j. schmidhuber, “deep learning in neural networks: an overview,” neural networks, vol. 61, pp. 85-117, january 2015. [59] g. e. hinton, et al., “reducing the dimensionality of data with neural networks,” science, vol. 313, no. 5786, pp. 504-507, july 2006. [60] i. goodfellow, et al., deep learning, cambridge: mit press, 2016. [61] d. p. kingma, et al., “auto-encoding variational bayes,” https://arxiv.org/pdf/1312.6114v4.pdf, december 2013. [62] d. p. kingma, et al., “an introduction to variational autoencoders,” https://arxiv.org/pdf/1906.02691v1.pdf, june 2019. [63] c. doersch, “tutorial on variational autoencoders,” https://arxiv.org/pdf/1606.05908v1.pdf, june 2016. [64] i. goodfellow, et al., “generative adversarial nets,” proceedings of the 27th international conference on neural information processing systems, pp. 2672-2680, december 2014. [65] m. you, et al., “systematic evaluation of deep face recognition methods,” neurocomputing, vol. 388, pp. 144-156, may 2020. [66] f. n. iandola, et al., “squeezenet: alexnet-level accuracy with 50x fewer parameters and <0.5 mb model size,” https://arxiv.org/pdf/1602.07360.pdf, november 2016. [67] a. g. howard, et al., “mobilenets: efficient convolutional neural networks for mobile vision,” https://arxiv.org/pdf/1704.04861.pdf, april 2017. [68] s. ge, et al., “low-resolution face recognition in the wild via selective knowledge distillation,” ieee transactions on image processing, vol. 28, no. 4, pp. 2051-2062, november 2018. [69] p. li, et al., “on low-resolution face recognition in the wild: comparisons and new techniques,” ieee transactions on information forensics and security, vol. 14, no. 8, pp. 2000-2012, august 2019. [70] m. grgic, et al., “scface-surveillance cameras face database,” multimedia tools applications, vol. 51, no. 3, pp. 863-879, february 2011. [71] a. sapkota, et al., “large scale unconstrained open set face database,” ieee 6th international conference on biometrics: theory, applications, and systems, pp. 1-8, september 2013. [72] a. radford, et al., “unsupervised representation learning with deep convolutional generative adversarial networks,” https://arxiv.org/pdf/1511.06434v1.pdf, november 2015. [73] l. s. luevano, et al., “a study on the performance of unconstrained very low resolution face recognition: analyzing current trends and new research directions,” ieee access, vol. 9, pp. 75470-75493, may 2021. 293 advances in technology innovation, vol. 7, no. 4, 2022, pp. 279-294 [74] r. ranjan, et al., “a fast and accurate system for face detection, identification, and verification,” ieee transactions on biometrics, behavior, and identity science, vol. 1, no. 2, pp. 82-96, april 2019. [75] b. f. klare, et al., “pushing the frontiers of unconstrained face detection and recognition: iarpa janus benchmark a,” proceedings of the ieee conference on computer vision and pattern recognition, pp. 1931-1939, june 2015. [76] c. whitelam, et al., “iarpa janus benchmark-b face dataset,” proceedings of the ieee conference on computer vision and pattern recognition workshops, pp. 90-98, july 2017. [77] b. maze, et al., “iarpa janus benchmark-c: face dataset and protocol,” international conference on biometrics, pp. 158-165, february 2018. [78] “who director-general’s opening remarks at the media briefing on covid-19–11 march 2020,” https://www.who.int/director-general/speeches/detail/who-director-general-s-opening-remarks-at-the-media-briefing-oncovid-19---11-march-2020, march 11, 2020. [79] a. heidari, et al., “the covid-19 epidemic analysis and diagnosis using deep learning: a systematic literature review and future directions,” computers in biology and medicine, vol. 141, article no. 105141, february 2022. [80] m. abboah-offei, et al., “a rapid review of the use of face mask in preventing the spread of covid-19,” international journal of nursing studies advances, vol. 3, article no. 100013, november 2021. [81] h. fouquet, “paris tests face-mask recognition software on metro riders,” https://www.bloombergquint.com/politics/paris-tests-face-mask-recognition-software-on-metro-riders, may 07, 2020. [82] e. mbunge, et al., “application of deep learning and machine learning models to detect covid-19 face masks–a review,” sustainable operations and computers, vol. 2, pp. 235-245, january 2021. [83] a. nowrin, et al., “comprehensive review on facemask detection techniques in the context of covid-19,” ieee access, vol. 9, pp. 106839-106864, july 2021. [84] p. khandelwal, et al., “using computer vision to enhance safety of workforce in manufacturing in a post covid world,” https://arxiv.org/ftp/arxiv/papers/2005/2005.05287.pdf, may 2020. [85] m. loey, et al., “fighting against covid-19: a novel deep learning model based on yolo-v2 with resnet-50 for medical face mask detection,” sustainable cities and society, vol. 65, article no. 102600, february 2021. [86] s. v. militante, et al., “real-time facemask recognition with alarm system using deep learning,” 11th ieee control and system graduate research colloquium, pp. 106-110, august 2020. [87] b. qin, et al., “identifying facemask-wearing condition using image super-resolution with classification network to prevent covid-19,” sensors, vol. 20, no. 18, article no. 5236, september 2020. [88] m. inamdar, et al., “real-time face mask identification using facemasknet deep learning network,” https://ssrn.com/abstract=3663305, july 2020. [89] m. jiang, et al., “retinamask: a face mask detector,” https://arxiv.org/pdf/2005.03950v1.pdf, may 2020. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). 294  advances in technology innovation, vol. 7, no. 1, 2022, pp. 19-29 implementation of 20 nm graphene channel field effect transistors using silvaco tcad tool to improve short channel effects over conventional mosfets vinod pralhad tayade 1, 3, * , swapnil laxman lahudkar 2 1 department of electronics and telecommunication engineering, aissms institute of information technology, pune, india 2 department of electronics and telecommunication engineering, jspm’s imperial college of engineering and research, pune, india 3 department of electronics and telecommunication engineering, government polytechnic, nashik, india received 17 july 2021; received in revised form 01 september 2021; accepted 02 september 2021 doi: https://doi.org/10.46604/aiti.2021.8098 abstract in recent years, demands for high speed and low power circuits have been raised. as conventional metal oxide semiconductor field effect transistors (mosfets) are unable to satisfy the demands due to short channel effects, the purpose of the study is to design an alternative of mosfets. graphene fets are one of the alternatives of mosfets due to the excellent properties of graphene material. in this work, a user-defined graphene material is defined, and a graphene channel fet is implemented using the silvaco technology computer-aided design (tcad) tool at 100 nm and scaled to 20 nm channel length. a silicon channel mosfet is also implemented to compare the performance. the results show the improvement in subthreshold slope (ss) = 114 mv/dec, ion/ioff ratio = 14379, and drain induced barrier lowering (dibl) = 123 mv/v. it is concluded that graphene fets are suitable candidates for low power applications. keywords: graphene, mosfet, silvaco tcad, graphene fet, 2d low power design, 2d-material 1. introduction conventional metal oxide semiconductor field effect transistors (mosfets) have a limitation when scaled down to nanometer channel lengths. their performance is degraded and short channel effects emerge, which degrade the overall performance of devices. the circuit designed using the scaled device consume more static and dynamic power. hence, the era of conventional mosfets has come to an end and the device design community is in search of an alternative of conventional mosfets. it is important to search for a convincing material, which can be used at small channel lengths. graphene is a promising material due to its higher mobility, better electrical conductivity, and better thermal conductivity. the implementation of this atomically thin material needs to be defined in the silvaco technology computer-aided design (tcad) tool for any further use of graphene material applications. the purpose of this study is to investigate and use the significant properties of the promising graphene material as a channel to replace conventional mosfets. the mobility of graphene material is high, and hence it can be used as a channel material for small geometry devices. the work on graphene channel fets is still in its initial stage, as it needs to further improve and optimize the ion/ioff ratio, subthreshold slope (ss), and drain induced barrier lowering (dibl) of these short channel parameters. in the present scenario, graphene fets are the demanding devices for low power, radio frequency (rf), * corresponding author. e-mail address: taydevinod@gmail.com tel.: +91-9766334676 advances in technology innovation, vol. 7, no. 1, 2022, pp. 19-29 biosensor circuits, and high-speed analog very-large-scale integration (vlsi) designs. one major concern for designing graphene fets is the availability of software tools because graphene is not available in conventional tcad tools such as the silvaco tcad and synopsys sentaurus tcad tools. hence, one purpose of the study is to add graphene as a user-defined material in the silvaco tcad tool, which can be used for any applications. the study is organized as follows. section 2 presents the detailed literature review. section 3 describes the design methodology used for implementation. section 4 provides the design of 100 nm fet using silicon and graphene material channels, defining the graphene material in the silvaco tcad tool. section 5 discusses the scaling of the device to 20 nm fet using silicon and graphene material channels. section 6 focuses on the results and the comparison with other published results. finally, the study is concluded. 2. literature review in 2015, international technology road map for semiconductors (itrs) discussed various emerging transistor structures in nano-scales [1]. graphene fets are one of the suitable candidates for future high-density and high-speed circuits [2]. graphene fets are designed by using top gate, bottom gate, and side gate approach to improve device parameters. various substrate materials like silicon substrates, sic substrates, and hexagonal boron nitride (hbn) substrates are reported as per the study. a bi-layer graphene sheet or graphene nanoribbon (gnr) is used to introduce the bandgap in a graphene fet due to which it becomes a suitable candidate for digital applications [3]. initially, novoselov et al. [4] extracted graphene from carbon and proposed that graphene could be the best possible metal for fet applications. in addition to the scalability to true nanometer sizes, graphene also offers linear current-voltage (i-v) characteristics, ballistic transport, and huge sustainable currents (9108 a/cm 2 ). graphene transistors show a rather modest on-off resistance ratio, which is a natural drawback of a material having zero bandgaps [4] schwierz [5] focused on graphene transistors’ status, prospects, and problems. the author reported the classification and detailed analysis of various graphene transistors developed in recent years. the challenge of graphene fets is the opening of the bandgap of the defined size and the reliable approach compatible with standard semiconductor processing steps. for logic operations, a bandgap of 0.4 ev or more will be required [5]. marmolejo-tejada et al. [6] presented a study of gnr based fets. the authors concluded that gnr-fets can be used for switching applications, and can offer a high ion/ioff ratio and the ss near its ideal value [6]. chen et al. [7] proposed a simulation program with integrated circuit emphasis (spice) compatible model of mos type gnr-fets with doped reservoirs, currents, and charge models which closely match with numerical tcad simulations. they observed that gnr-fets are promising compared to silicon complementary metal oxide semiconductors (cmos) since these devices have either lower power or lower delay. gnr-fets are still promising candidates for low-power applications [7]. chen et al. [8] proposed a bi-layer graphene-based electrostatically doped tunnel field effect transistor (bed-tfet), and studied the operation principle of the bed-tfet and its performance sensitivity to the device design parameters. agarwal et al. [9] proposed a bi-layer graphene tunneling field effect transistor (blg-tfet) suitable for digital cmos logic circuits. a bandgap opening is induced in blg using both top-bottom asymmetric chemical doping and vertical electric field. the proposed blg-tfet shows better characteristics for ultralow-power applications, specifically in low to medium-speed applications [9]. lv et al. [10] proposed segmented edge saturation (ses) as a novel method to design high-performance tfets using smooth gnr. both high on-to-off current (ion/ioff) ratio and large ion are obtained [10]. rassekh and fathipour [11] reported a junctionless transistor (jlt) at 10 nm gate length, which is suitable for low power applications. the authors reported few parameters, i.e., ion/ioff ratio = 4.2 × 10 4 , threshold voltage (vth) = 327 mv, dibl = 218 mv/v, ss = 109.9 mv/dec., and the comparison is carried out with silicon-on-insulator (soi) fin-shaped field effect 20 advances in technology innovation, vol. 7, no. 1, 2022, pp. 19-29 transistors (finfets) [11]. boukortt et al. [12] reported a soi n-channel finfet using the silvaco tool at 8 nm gate length, and demonstrated the effects of gate work function on various parameters. the authors reported few parameters such as ss = 63.13 mv/dec, ion/ioff ratio = 10 6 , and dibl = 85.30 mv/v [12]. pravin et al. [13] demonstrated the effectiveness of using a high-k dielectric material. the authors used hfo2 material as dielectrics instead of sio2. these authors observed that dibl is reduced by 61.5%, delay is reduced by 4%, and ion/ioff ratio is equal to 10 9. [13]. ning et al. [14] demonstrated a flexible fet using the chemical vapor deposition (cvd) technique. the proposed fet exhibits ion/ioff ratio = 400 on a bending surface [14]. he et al. [15] studied the temperature effect on graphene fets for rf applications. the authors found that graphene fets could be used up to 200°c temperature using sic substrates [15]. tamersit and djeffal [16] reported gnr-fets using graded gate engineering. the implemented device shows considerable improvement in ss, voltage gain, and cutoff frequency as compared to normal gnr-fets [16]. in another implementation, tamersit [17] observed that in gnr-fets, short channel parameters can be improved by using the junctionless and multigate technology. the author reported the improvement in ss, dibl, and threshold voltage roll-off [17]. radsar et al. [18] reported the performance improvement in gnr-fets by changing the gate dielectrics with high dielectric coefficient material lanthanum aluminate. the authors observed the improvement in ion/ioff ratio, ss, and dibl as compared to other dielectric materials [18]. fahad et al. [19] solved analytical models from gnr-tfets using schrodinger equations. the designed form channel length is 20 nm. the authors found close agreement of ion/ioff ratio and ss with a numerical quantum simulation method [19]. the graphene and sio2 oxide interface can also degrade the performance of the device due to the mismatch of structure and tunneling of hot electrons into the oxide [20]. it leads to the degradation of drain current (id) and vth. hence, to improve these parameters, it is proposed that the use of hfo2 having high-k could be a suitable material as an oxide [21]. for the fabrication of graphene layers on silicon substrates, a rapid cvd system is used. using this system, a graphene layer having the thickness in 2 to 3 nm can be formed. this represents that the layer of graphene material on the silicon substrate offers the expected mobility for graphene channel fets. the rapid cvd fabricated the graphene material to be used as channel material, which is equivalent to 1 nm atomic thick layer. with reference to this, in the proposed research work, the graphene material thickness is considered to be 1.5 to 2 nm [22]. 3. design methodology the id of conventional mosfets depends on various parameters. as per eq. (1), it is observed that id is directly proportional to the mobility of electrons (µ n) and applied drain-source voltage (vds) for n-channel mosfets. id is inversely proportional to the channel length (l). when scaling of mosfets is carried out at that time, the channel length is reduced in nanometer size and vds is also reduced. due to the reduction in vds, the overall power consumption of the device reduces, and id also degrades. hence, to improve id, the mobility of electrons can be increased, but the mobility of silicon material is limited. to increase the mobility, a new promising material graphene is used in this design, which has a mobility of 30000 cm 2 /v.s. due to the increase in mobility, the performance of the device can be improved. the short channel parameter ss is directly proportional to depletion capacitance (cd) and inversely proportional to oxide capacitance (cox) as per eq. (2). to improve ss to its ideal value of 60 mv/dec., cox can be reduced by changing the dielectric material used under the gate terminal. in this design, hfo2, the dielectric material is used to improve ss. the improvement in ss also improves the ion/ioff ratio of the device, which is an essential factor for the digital logic application of the device and low power consumption. dibl is another short channel effect that depends on vds and vth as per eq. (3). dibl can be controlled by improving vth, which again depends on channel materials and oxide materials. in this research work, the effect of graphene material as a channel is studied, and the simulation results are obtained using the silvaco tcad tool. the four designs are discussed in further sections. 21 advances in technology innovation, vol. 7, no. 1, 2022, pp. 19-29 2. ( ) . .[2( ) ] 2 ox d gs ds ds n c w i lin v vt v v l     (1) 60(1 ) d ox c ss c   (2) dd low th th high low dd dd v v dibl v v    (3) 4. design of device having 100 nm channel length 4.1. design of silicon mosfet with 100 nm channel length the silicon channel mosfet is designed by using the dimensions as per the standard examples from the silvaco tcad tool. fig. 1 shows the designed structure of 100 nm channel length device. the total length of the device in x-direction is 500 nm and the height of the device in y-direction is 65 nm, thus a proper aspect ratio is maintained. the source terminal length in x-direction ranges from 0 nm to 200 nm, and the contact of source ranges from 0 nm to 100 nm, as shown in fig. 1. the channel region ranges from 200 nm to 300 nm with a height of 15 nm. the drain region ranges from 300 nm to 500 nm, and the drain contact ranges from 400 nm to 500 nm in x-direction. a sio2 dielectric material is deposited with 2.5 nm thickness. it ranges from 175 nm to 325 nm over the channel region. a polysilicon contact of 2.5 nm thickness is used to apply gate voltage. the bulk starts from 20 nm to 65 nm in y-direction. the default body voltage is zero. fig. 1 silicon channel mosfet with 100 nm channel length 4.2. simulation results of 100 nm mosfet the designed 100 nm mosfet is simulated, and various parameters are extracted. initially, the drain current to gate-source voltage (id-vgs) characteristic is plotted. vth is extracted from this plot, the observed value is vth = 0.303v. for dibl extraction, the device is simulated for two different drain-drain voltage (vdd), i.e., vdd(min) = 0.1 v and vdd(max) = 1 v. the obtained dibl value is 0.0279 mv/v. the ss value, which decides the speed of the device, is observed to be 79.72 mv/dec (near its ideal value of 60 mv/dec). ion/ioff ratio is obtained by finding the values of ion at vdd = 1 v and ioff at vdd = 0 v. the ratio observed is 1.73e 10 , which is higher enough to switch off the device and for low leakage current. fig. 2 shows the id-vgs characteristics. fig. 3 shows the id-vds characteristics to plot and find the saturation slope. the three curves for three different gate voltages are plotted: vgs1 = 0.3 v, vgs2 = 0.6 v, and vgs3 = 1 v. the observed saturation slope value is 2.78e-05. from the characteristics, it is observed that the device offers a very low id for vgs1 = 0.3 v and vgs2 = 0.6 v, and offers sufficient id for vgs3 = 1 v. 22 advances in technology innovation, vol. 7, no. 1, 2022, pp. 19-29 4.3. design of graphene fet with 100 nm channel length the graphene material is not directly available in the silvaco tcad tool for simulation purposes. hence, the design of the graphene channel fet is implemented using the silvaco tcad tool by adding graphene as a user-defined material. graphene is an extract of carbon having very high carrier mobility and a 2-d structure. to implement graphene fet using silvaco tcad, one major issue is that graphene needs to be directly available in the tool. hence, the user needs to define a user-defined material by changing the properties of the existing material. in the silvaco tool, various materials are available which can be used as an alternative of graphene. one attempt has been reported by mobarakeh et al. [23]. a 3c-sic material is used as a graphene channel. another design using the silvaco tcad user-defined material is demonstrated by kuang et al. [24]. in silvaco tcad, three materials have the properties which are close to that of graphene material, as shown in table 1. in this work, insb is used as a base material because it has low energy bandgap and higher electron and hole mobility, which is the closest to the properties of graphene material. table 1 different material parameters which are close to graphene material [25] material eg (ev) mun (cm 2 /v.s) mup (cm 2 /v.s) nc (per cc) nv (per cc) ni (per cc) vsatn (cm/s) vsatp (cm/s) 3c-sic 2.2 1000 50 6.59e+18 1.68e+18 1.1 2.00e+7 1.00e+6 insb 0.17 78000 750 4.16e+16 6.35e+18 1.92e+16 1.00e+6 1.00e+6 inas 0.35 33000 460 9.33e+16 8.12e+18 9.99e+14 1.00e+6 1.00e+6 note: eg is energy bandgap; mun is the mobility of electronics; mup is the mobility of holes; nc is the effective density of state (conduction band); cc is cubic per centimeter; nv is the effective density of state in valence band; ni is intrinsic carrier concentration; vsatn is the saturation velocity of electrons; vsatp is the saturation velocity of holes. as per the syntax of user-defined materials, the following statement is included in the code. this statement includes the properties of the user-defined graphene material: “material material = graphene eg300 = 0.7 affinity = 4.07 mun = 30000 mup = 30000 nc300 = 4.16e16 nv300 = 6.35e18 index.file = graphene.nkuser.group = semiconductor user.default = insb”. in this statement, a .nk file for graphene material is formed by preparing a table of 499 entries of wavelength and its corresponding refractive index. as demonstrated by weber [26], the energy bandgap for this simulation is considered 0.7 ev. from these different experiments carried out by han et al. [27] and chen et al. [28], it is observed that in gnr, if the ribbon width is reduced below 20 nm, then a sufficient energy bandgap can be achieved, and hence the nanoribbon device can be used as the switching device. 1.00e-14 1.00e-12 1.00e-10 1.00e-08 1.00e-06 1.00e-04 1.00e-02 1.00e+00 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 d ra in c u rr e n t (a /µ m ) gate voltage (v) 1.00e-24 1.00e-21 1.00e-18 1.00e-15 1.00e-12 1.00e-09 1.00e-06 1.00e-03 1.00e+00 0 0 .0 1 0 .0 3 0 .0 5 0 .1 0 .2 0 .3 0 .4 0 .5 0 .6 0 .7 0 .8 0 .9 1 d ra in c u rr e n t (a /µ m ) drain voltage (v) si_100 nm vgs1 = 0.3 v si_100 nm vgs2 = 0.6 v si_100 nm vgs3 = 1 v fig. 2 id-vgs characteristics of si-channel mosfet (100 nm) fig. 3 id-vds with three values of vgs and y-axis with log scale 23 advances in technology innovation, vol. 7, no. 1, 2022, pp. 19-29 4.4. defining graphene structure fig. 4 shows the original structure of 100 nm silicon mosfet, which is modified to form the graphene fet. all dimensions are as per the silicon 100 nm design; only the channel material from 200 nm to 300 nm in x-direction and the one from 5 nm to 20 nm in y-direction are replaced with the user-defined graphene material. fig. 5 shows the id-vgs curve by varying the vgs values from 0 v to 1 v with a step of 0.1 v. it has a close agreement with conventional mosfets. fig. 6 shows the family of id-vds curve for three different gate voltages, i.e., vgs1 = 0.3 v, vgs2 = 0.6 v, and vgs3 = 1 v. this curve shows that the graphene fet enters into the saturation region and works like a normal silicon mosfet. fig. 4 structure of 100 nm graphene channel fet fig. 5 id-vgs curve for 100 nm graphene channel fet 5. design of device having 20 nm channel length 5.1. design of graphene fet with 20 nm channel length the graphene channel mosfet is designed. as shown in fig. 7, the original 100 nm device is scaled down to 20 nm device by keeping the aspect ratio. the total length of the device in x-direction is 100 nm and the height of the device in y-direction is 13 nm, thus a proper aspect ratio is maintained. the source terminal length in x-direction ranges from 0 to 40 nm and the contact of the source ranges from 0 to 20 nm. the channel region is formed between 40 nm to 50 nm in x-direction with a height of 20 nm in y-direction. the drain region starts from 60 nm and ends at 100 nm, and the drain contact starts from 80 nm to 100 nm in x-direction. hfo2 dielectric material is deposited with 2 nm thickness; it ranges from 37.5 nm to 62.5 nm over the channel region. the contact of 2 nm thickness is used to apply gate voltage. the bulk starts from 4 nm to 13nm in y-direction. the default body voltage is zero. fig. 8 shows id-vgs characteristics of the graphene fet for vds = 0.8 v, and fig. 9 shows id-vds characteristics for three different gate voltages, i.e., vgs1 = 0.3 v, vgs2 = 0.6 v, and vgs3 = 0.8 v. the device offers more id for vgs = 0.8 v. 1.00e-13 1.00e-12 1.00e-11 1.00e-10 1.00e-09 1.00e-08 1.00e-07 1.00e-06 1.00e-05 1.00e-04 1.00e-03 1.00e-02 1.00e-01 1.00e+00 0.00 0.10 0.20 0.30 0.40 0.50 0.60 0.70 0.80 0.90 1.00 d ra in c u rr e n t (a /µ m ) gate voltage (v) 1.00e-15 1.00e-14 1.00e-13 1.00e-12 1.00e-11 1.00e-10 1.00e-09 1.00e-08 1.00e-07 1.00e-06 d ra in c u rr e n t (a /µ m ) drain voltage (v) drain current (vgs1 = 0.3 v) drain current (vgs2 = 0.6 v) drain current (vgs3 = 1 v) fig. 6 id-vds characteristics for three gate voltages 24 advances in technology innovation, vol. 7, no. 1, 2022, pp. 19-29 fig. 7 structure of 20 nm graphene channel fet fig. 8 id-vgs curve of 20 nm graphene channel fet 5.2. design of 20 nm silicon channel mosfet the implementation of a 20 nm silicon channel mosfet is designed and simulated to compare to the results with the 20 nm graphene channel fet. the structure is shown in fig. 10. the channel is replaced with silicon material and hfo2 is used as the dielectric. in the id-vgs characteristics for vds = 0.8v, id increases linearly with an increase in vgs after vth. in the id-vds curve for vgs1 = 0.3 v, vgs2 = 0.6 v, and vgs3 = 0.8 v, the small geometry silicon device enters into the saturation region after a pinch-off point. the characteristics of the 20 nm silicon channel mosfet are embedded with the 20 nm graphene fet design in the result section. fig. 10 structure of 20 nm silicon channel fet 1.00e-07 1.00e-06 1.00e-05 1.00e-04 1.00e-03 1.00e-02 1.00e-01 1.00e+00 0.00 0.08 0.16 0.24 0.32 0.40 0.48 0.56 0.64 0.72 0.80 d ra in c u rr e n t (a /µ m ) gate voltage (v) -1.00e-03 0.00e+00 1.00e-03 2.00e-03 3.00e-03 4.00e-03 5.00e-03 6.00e-03 7.00e-03 0 0 .0 1 0 .0 1 0 .0 2 0 .0 4 0 .0 8 0 .1 6 0 .2 4 0 .3 2 0 .4 0 .4 8 0 .5 6 0 .6 4 0 .7 2 0 .8 d ra in c u rr e n t (a /µ m ) drain voltage (v) drain current (vgs1 = 0.3 v) drain current (vgs2 = 0.6 v) drain current (vgs3 = 1 v) fig. 9 id-vds curve for three different values of vgs 25 advances in technology innovation, vol. 7, no. 1, 2022, pp. 19-29 6. results and discussion 6.1. discussion of 100 nm channel length design the simulation of both the silicon channel and graphene channel fets is carried out. for the supply voltage of 1 v, it is observed that the graphene fet shows ss = 68.45 mv/dec. and the silicon fet shows ss = 79.6 mv/dec. the ss of graphene fet is close to the ideal value of 60 mv/dec. hence, it is a suitable candidate for low-power applications. another short channel parameter dibl is also reduced to a value of 202 mv/v in the graphene fet as compared to 279 mv/v in the si-mosfet, which is better for stable operation of the device. the ion/ioff ratio of graphene fet is observed equal to55962, which is still less than that of silicon mosfet, so it needs to be improved further. the id(max) of silicon mosfet is more than that of graphene fet. table 2 shows a comparison of all parameters. in the graphene fet design, by changing gate dielectric material from sio2 to hfo2, the improvement in parameters is observed. fig. 11 shows overlay characteristics of both fets. from the characteristics, it is observed that the si-mosfet offers more current for different gate voltages; hence, further scaling of these two devices is carried out and a 20 nm fet is designed to observe and improve short channel effects. from table 2, it is observed that graphene material can be used as a channel material to replace silicon because the behavior of the graphene fet is found to be identical to conventional mosfets. the id of silicon fet is more than that of graphene fet due to the large channel length and sufficient bandgap of silicon material. the ss in graphene fet for hfo2 dielectric material is observed to be 68.45 mv/decade. it is due to an increase in cox, which depends on dielectric material and oxide thickness (tox). as the dielectric constant of hfo2 is high and tox is only 2.5 nm which is low, the cox value increases. as per eq. (2), ss depends on cox. the ss decreases due to an increase in the cox value. this decrease in ss is close to the ideal value of ss. the ion/ioff ratio is still more in the silicon fet due to the high ion caused by the sufficient bandgap available in silicon material. the dibl is reduced in the graphene fet due to more control of gate voltage on channel carriers because of better conductivity of graphene material and high-k gate dielectric material. fig. 11 overlay characteristics of 100 nm si-mosfet and graphene fet table 2 comparison of silicon and graphene channel mosfets at 100 nm gate length device/parameter si-mosfet graphene fet (with sio2) graphene fet (with hfo2) channel length (nm) 100 100 100 vdd(max) (v) 1 1 1 electron mobility (cm 2 /v.s) 1000 30000 30000 hole mobility (cm 2 /v.s) 500 30000 30000 bandgap (ev) 1.08 0.7 0.7 channel material silicon graphene graphene gate dielectric material sio2 sio2 hfo2 vth (v) 0.303 0.322 0.357 subthreshold swing (mv/dec) 79.6 93.0 68.45 id(max) (a/µm) 0.000122 2.18e-08 1.69e-6 ion/ioff 1.73e10 984.21 55962.56 dibl (mv/v) 279 202 202 1.00e-24 1.00e-21 1.00e-18 1.00e-15 1.00e-12 1.00e-09 1.00e-06 1.00e-03 1.00e+00 d ra in c u rr e n t (a /µ m ) drain voltage (v) si_100 nm vgs1 = 0.3 v si_100 nm vgs2 = 0.6 v si_100 nm vgs3 = 1 v gr_100 nm vgs1 = 0.3 v gr_100 nm vgs2 = 0.6 v gr_100 nm vgs3 = 1 v 26 advances in technology innovation, vol. 7, no. 1, 2022, pp. 19-29 6.2. discussion of 20 nm channel length design the implemented graphene channel fet and silicon channel fet at 20 nm channel length are compared as shown in table 3. it is observed that the graphene fet offers vth = 0.040 v, and silicon mosfet offers vth = 0.021 v. the observed ss is 114 mv/dec in the graphene fet and 115 mv/dec in the silicon mosfet. these values need further improvement. the id value is more in the graphene fet for vds = 0.8 v, as compared to the silicon mosfet. the observed ion/ioff ratio is 14379, which shows the improvement in the graphene fet, so it is suitable for low power applications. the observed dibl in the graphene fet is 123 mv/v, which is less than that of the silicon channel mosfet. the current work is also compared with the already published work, and it can be seen that the graphene fet shows the improvement in id(max) due to the high mobility of graphene material and dibl parameters. fig. 12 shows the comparison curve of id-vgs for vds = 0.8 v. it is observed that the graphene fet offers a high id value for the same voltage of vds as compared to the silicon mosfet. fig. 13 shows the combined curve of id-vds for three different gate voltages. the comparison shows that the graphene fet offers more id than the silicon mosfet, which is a benefit for the high-speed operation of the device. as per table 3, the id(max) is the highest as compared to others due to the excellent electrical conductivity of graphene material and higher mobility. however, due to this, there is an increase in ss, which is still lower than that of the silicon fet. the id of the silicon fet decreases due to short channel effects. the ion/ioff ratio of the graphene fet increases due to the low leakage in graphene material in “off” conditions. the high value of ion also contributes to the increase of this ratio. the dibl parameter of the graphene fet is also lower than that of the si-mosfet due to the better control of gate voltage on the channel in the short device. the id of the graphene fet is the highest among all published results shown in table 3 due to the higher mobility of graphene material. ss, dibl, and ion/ioff ratio needs to be further improved for low power applications. table 3 comparison of the parameters in graphene and silicon mosfets device si-mosfet (this work) graphene fet (this work) soi-jlt [11] finfet [12] gnr-tfet [19] channel length (nm) 20 20 10 8 20 vdd(max) (v) 0.8 0.8 0.8 0.9 0.1 electron mobility (cm 2 /v.s) 1000 30000 1000 1000 not mentioned hole mobility (cm 2 /v.s) 500 30000 500 500 not mentioned bandgap (ev) 1.08 0.7 1.08 1.08 0.289 base material silicon user-defined graphene silicon silicon graphene gate dielectric material hfo2 hfo2 hfo2 sio2 sio2 vth (v) 0.0218 0.040 0.327 subthreshold swing (mv/dec.) 115 114 109.9 63.13 27.4 id(max) (a/µm) 0.0027 0.00638 330 × 10 -6 0.00001 4.4 × 10-6 ion/ioff 6401 14379 420000 10 6 116 dibl (mv/v) 129 123 218 85 fig. 12 comparison curve of id-vgs for vds = 0.8 v 0.00e+00 2.00e-04 4.00e-04 6.00e-04 8.00e-04 1.00e-03 1.20e-03 1.40e-03 0 0 .0 8 0 .1 6 0 .2 4 0 .3 2 0 .4 0 .4 4 0 .4 8 0 .5 6 0 .6 4 0 .7 2 0 .8 d ra in c u rr e n t (a /µ m ) gate voltage (v) drain current (gr_20 nm) drain current (si_20 nm) 27 advances in technology innovation, vol. 7, no. 1, 2022, pp. 19-29 fig. 13 combined curve of id-vds for three different gate voltages 7. conclusions the silvaco tcad tool does not have an inbuilt graphene material; hence, a user-defined graphene material is added in this study. this material is formed by using the parameters of the existing materials, which are close to that of graphene, and the simulation is carried out. the first graphene channel fet with 100 nm channel length is implemented using hfo2 material as the dielectric under gate terminal, and compared with another implementation of 100 nm silicon channel fet. the results are in good agreement with each other. the high dielectric material hfo2 offers less leakage current; hence, it improves the ion/ioff ratio of the device. the graphene fet has improved ss and dibl parameters. to study the short channel effects, further scaling of the graphene fet is carried out to the 20 nm channel length. this small channel device also shows the improvement in dibl, ss, and id(max) over the 20 nm silicon fet and other published results. the saturation curves of the both are plotted and compared, and it is observed that the graphene fet provides more id as compared to the silicon mosfet. as a high ion/ioff ratio is observed, it is concluded that the graphene channel fet can be a perfect replacement for a conventional silicon mosfet at a small channel length for low power and high-speed applications. further improvement of the implemented fet using the user-defined graphene material could be achieved by using a double gate structure. conflicts of interest the authors declare no conflict of interest. references [1] the international technology roadmap for semiconductors, “international technology roadmap for semiconductors 2.0, 2015 edition, beyond c-mos,” https://www.semiconductors.org/wp-content/uploads/2018/06/6_2015-itrs-2.0-beyond-cmos.pdf, 2015. [2] v. tayade and s. lanudkar, “a review of emerging devices beyond mosfet for high performance computing,” international conference on emerging smart computing and informatics, pp. 34-38, march 2020. [3] i. meric, c. r. dean, n. petrone, l. wang, j. hone, p. kim, et al., “graphene field-effect transistors based on boron-nitride dielectrics,” proceedings of the ieee, vol. 101, no. 7, pp. 1609-1619, july 2013. [4] k. s. novoselov, a. k. geim, s. v. morozov, d. e. jiang, y. zhang, s. v. dubonos, et al., “electric field effect in atomically thin carbon films,” science, vol. 306, no. 5696, pp. 666-669, october 2004. [5] f. schwierz, “graphene transistors: status, prospects, and problems,” proceedings of the ieee, vol. 101, no. 7, pp. 1567-1584, july 2013. [6] j. m. marmolejo-tejada and j. velasco-medina, “review on graphene nanoribbon devices for logic applications,” microelectronics journal, vol. 48, pp. 18-38, february 2016. [7] y. y. chen, a. sangai, a. rogachev, m. gholipour, g. iannaccone, g. fiori, et al., “a spice-compatible model of mos-type graphene nano-ribbon field-effect transistors enabling gateand circuit-level delay and power analysis under process variation,” ieee transactionsons on nanotechnology, vol. 14, no. 6, pp. 1068-1082, november 2015. -1.00e-03 0.00e+00 1.00e-03 2.00e-03 3.00e-03 4.00e-03 5.00e-03 6.00e-03 7.00e-03 0 0 .0 1 0 .0 2 0 .0 4 0 .0 8 0 .1 6 0 .2 4 0 .3 2 0 .4 0 .4 8 0 .5 6 0 .6 4 0 .7 2 0 .8 d ra in c u rr e n t (a /µ m ) drain voltage (v) gr_20 nm vgs1 = 0.3 v gr_20 nm vgs2 = 0.6 v gr_20 nm vgs3 = 0.8 v si_20 nm vgs1 = 0.3 v si_20 nm vgs2 = 0.6 v si_20 nm vgs3 = 0.8 v 28 advances in technology innovation, vol. 7, no. 1, 2022, pp. 19-29 [8] f. w. chen, h. ilatikhameneh, g. klimeck, z. chen, and r. rahman, “configurable electrostatically doped high-performance bilayer graphene tunnel fet,” ieee journal of electron devices society, vol. 4, no. 3, pp. 124-128, may 2016. [9] t. k. agarwal, a. nourbakhsh, p. raghavan, i. radu, s. de gendt, m. heyns, et al., “bilayer graphene tunneling fet for sub-0.2 v digital cmos logic applications,” ieee electron device letters, vol. 35, no. 12, pp. 1308-1310, december 2014. [10] y. lv, w. qin, q. huang, s. chang, h. wang, and j. he, “graphene nanoribbon tunnel field-effect transistor via segmented edge saturation,” ieee transactions on electron devices, vol. 64, no. 6, pp. 2694-2701, june 2017. [11] a. rassekh and m. fathipour, “a single-gate soi nanosheet junctionless transistor at 10-nm gate length: design guidelines and comparison with the conventional soi finfet,” journal of computational electronics, vol. 19, no. 2, pp. 631-639, june 2020. [12] n. e. i. boukortt, b. hadri, a. caddemi, g. crupi, and s. patanè, “3-d simulation of nanoscale soi n-finfet at a gate length of 8 nm using atlas silvaco,” transactions on electrical and electronic materials, vol. 16, no. 3, pp. 156-161, 2015. [13] j. c. pravin, d. nirmal, p. prajoon, and j. ajayan, “implementation of nanoscale circuits using dual metal gate engineered nanowire mosfet with high-k dielectrics for low power applications,” physica e: low-dimensional systems and nanostructures, vol. 83, pp. 95-100, september 2016. [14] j. ning, y. wang, x. feng, b. wang, j. dong, d. wang, et al., “flexible field-effect transistors with a high on/off current ratio based on large-area single-crystal graphene,” carbon, vol. 163, pp. 417-424, august 2020. [15] z. he, c. yu, q. liu, x. song, x. gao, j. guo, et al., “high temperature rf performances of epitaxial bilayer graphene field-effect transistors on sic substrate,” carbon, vol. 164, pp. 435-441, august 2020. [16] k. tamersit and f. djeffal, “boosting the performance of a nanoscale graphene nanoribbon field-effect transistor using graded gate engineering,” journal of computational electronics, vol. 17, no. 3, pp. 1276-1284, september 2018. [17] k. tamersit, “a computational study of short-channel effects in double-gate junctionless graphene nanoribbon field-effect transistors,” journal of computational electronics, vol. 18, no. 4, pp. 1214-1221, december 2019. [18] t. radsar, h. khalesi, and v. ghods, “improving the performance of graphene nanoribbon field-effect transistors by using lanthanum aluminate as the gate dielectric,” journal of computational electronics, vol. 19, no. 4, pp. 1507-1515, december 2020. [19] m. s. fahad, a. srivastava, a. k. sharma, and c. mayberry, “analytical current transport modeling of graphene nanoribbon tunnel field-effect transistors for digital circuit design,” ieee transactions on nanotechnology, vol. 15, no. 1, pp. 39-50, january 2016. [20] f. djeffal, t. bentrcia, m. a. abdi, and t. bendib, “drain current model for undoped gate stack double gate (gsdg) mosfets including the hot-carrier degradation effects,” microelectronics reliability, vol. 51, no. 3, pp. 550-555, march 2011. [21] m. a. abdi, f. djeffal, z. dibi, and d. arar, “a two-dimensional analytical subthreshold behavior analysis including hot-carrier effect for nanoscale gate stack gate all around (gasgaa) mosfets,” journal of computational electronics, vol. 10, no. 1-2, pp. 179-185, june 2011. [22] x. ma, w. gu, j. shen, and y. tang, “investigation of electronic properties of graphene/si field-effect transistor,” nanoscale research letters, vol. 7, no. 1, 677, december 2012. [23] m. s. mobarakeh, n. moezi, m. vali, and d. dideban, “a novel graphene tunneling field effect transistor (gtfet) using bandgap engineering,” superlattices and microstructures, vol. 100, pp. 1221-1229, december 2016. [24] y. kuang, y. liu, y. ma, j. xu, x. yang, x. hong, et al., “modeling and design of graphene gaas junction solar cell,” advances in condensed matter of physics, vol. 2015, 326384, 2015. [25] silvaco international, “atlas user’s manual, device simulation software, volume i,” http://statistics.roma2.infn.it/~messi/sic/sic_21-01-04/atlas98-v1_users.pdf, november 1998. [26] o. weber, “fdsoi vs finfet: differentiating device features for ultra low power & iot applications,” ieee international conference on ic design and technology, pp. 1-3, may 2017. [27] m. y. han, b. ö zyilmaz, y. zhang, and p. kim, “energy band-gap engineering of graphene nanoribbons,” physical review letter, vol. 98, no. 20, 206805, may 2007. [28] z. chen, y. m. lin, m. j. rooks, and p. avouris, “graphene nano-ribbon electronics,” physica e: low-dimensional systems and nanostructures, vol. 40, no. 2, pp. 228-232, december 2007. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). 29  advances in technology innovation, vol. 7, no. 3, 2022, pp. 181-194 maintenance initiation prediction incorporating vibrations and system availability lasithan lasyam gopikuttan 1,* , shouri puthan veettil 2 , rajesh vazhayil govindan 2 1 apj abdul kalam technological university, kerala, india 2 department of mechanical engineering, model engineering college, kerala, india received 04 october 2021; received in revised form 25 december 2021; accepted 26 december 2021 doi: https://doi.org/10.46604/aiti.2022.8618 abstract as per iso-10816, electric motors up to 15 kw are classified as class i machines, and the major reason for their failure is that the vibrations in them are above the alert limit. this study presents a new model for predicting the condition-based maintenance (cbm) initiation points through vibration measurement in a system of class i machines. the proposed model follows the accelerated life testing (alt) procedure. alt includes the formation of an artificial wear environment in bearings to analyze the resultant system vibrations on system availability. the artificial wear environment created is close to the real industrial situation. the results show that the prediction of the cbm initiation points is based on the established relation between the system availability and vibrations. furthermore, a relation between the available time for maintenance initiation and different vibration velocities is demonstrated. keywords: availability, condition-based maintenance, alert limit, alarm limit, acceleration factor 1. introduction reliability and availability management plays a prime role in the success of a company. for the economic performance of an industrial plant, maintaining high availability, reliability, and maintainability for the plant machines and their subsystems is crucial [1-3]. the machines in an industrial process plant can fail because of a wide range of reasons. the major reason behind machine failure is the unchecked magnitude of vibrations above the alert limit. the availability of a system or machine is affected badly when the machine failure occurs. the direct impact of the vibration levels on the system availability and the usability of that influence in condition-based maintenance (cbm) programs are not reported in the literature so far [4-5]. over the years, maintenance has advanced to adopt the aspect of reliability. reliability is a design attribute that shows the expected acceptable performance of an item. reliability-centered maintenance (rcm) is used to decide the maintenance requirements of any physical asset to ensure its satisfactory operation. cbm is a subclass of rcm [3]. typically, the objective of a cbm program is to devise a maintenance policy that optimizes the system performance according to the criteria such as cost, availability, and reliability [3]. early failure detection is the responsibility of most cbm programs [4]. for the cbm application, a potential failure-functional failure (p-f) curve can be plotted between the failure resistance or health condition of a machinery system and the time period of its failure [3]. it can be inferred from this curve that as time increases, failure resistance decreases to complete the system failure. this curve can be used to explain various stages of system failure, i.e., failure initiation (i), potential failure (p), and functional failure (f). along this curve, a point “p” can be identified, where there is a potential to fail, * corresponding author. e-mail address: lasithanlg@gmail.com tel.: +91-9744000988 advances in technology innovation, vol. 7, no. 3, 2022, pp. 181-194 and a point “f” beyond “p” can be detected, where the system will not perform as expected. after the point “p,” the health condition decreases rapidly. there is no hard data to define p-f intervals [3]. therefore, the best strategy is to employ methods that could effectively ascertain the machine condition before the potential failure occurs, and this should permit scheduling the repair activity before the p-f interval. shin et al. [5] examined the cbm strategy from various perspectives and addressed the data, procedure, and techniques for implementing the cbm approach. li et al. [4] proposed a cbm model for assuring average system availability and plant safety. electrical motors are integral parts of the majority of the machines installed in an industrial plant [3, 6]. the failure of any one of these motors degrades the machine performance, which, in turn, influences the overall plant availability. manjare et al. [6] reviewed machine learning (ml)-based fault detection techniques and predictive maintenance (pdm)/cbm strategies for electrical motors used in industrial plants. furthermore, the vibration data acquired from accelerometer sensors are extensively used for data analysis. kumar et al. [7] presented a comprehensive review of various faults in electric motors, failure causes, and advanced condition monitoring and diagnostic techniques. however, the availability aspects associated with the failure of electric motors and maintenance initiation points are not considered in these studies. in this experimental study, the deterioration in a system of electric motors with five hp-rated power class i machines is simulated by deliberately creating the vibrations above the alert limit. an analysis is performed for the values of the alert and alarm limits of vibration velocity (root mean square (rms)). the alert limit corresponds to the failure initiation, and the alarm limit corresponds to the potential failure for the system [3]. the maintenance initiation points for cbm are defined only in accordance with the shrinkage pattern of system availability with the vibration velocity for a particular system load. the maintenance initiation points of cbm for different system loads are predicted according to this relation. the proposed approach can be applied to accurately predict the value of vibration velocity (rms) for the maintenance initiation points of class i machines. the frequent failure of certain class i machines in an industrial process plant is mainly due to bearing wear, which gives rise to vibrations in the machines. as the wear increases, the system vibration level increases. in this study, the system failure is simulated by artificially wearing the bearings. the resultant vibration velocity is measured as a time series. the time recorded during the experiment is changed by suitable transformation to the corresponding useful life under normal operating conditions of the machines. this study establishes a relation between the vibration velocity and the projected useful life period of the bearings in the system for different possible system loads. from the value of a maintenance initiation point of vibration velocity for a possible system load, the corresponding life of the bearings in the system can be calculated. by considering possible system loads, this study establishes a general expression between the available maintenance initiation time for cbm and the vibration velocity for the system under consideration. in the expression, the vibration velocity considered is less than the defined value of maintenance initiation vibration velocity. the remainder of this study is organized as follows. section 2 provides a detailed review of the extant literature. section 3 explains the experimental setup of a class i machine and the loading arrangement for the same. the experimental procedure of the accelerated life testing (alt) and vibration measurement is also explained in this section. section 4 details the experimental data and failure analysis; also, the system availability is modeled in terms of vibration velocity (between alert and alarm limits), and the resulting curves are plotted. the equations for the available maintenance initiation time and the value of maintenance initiation points of the system are established in this section. finally, section 5 concludes this study by summarizing the findings. 2. literature review and background study the term “reliability” can be defined as the probability that under the stated operating conditions, a system or equipment will perform its intended function satisfactorily for a specified interval [2]. the general expression for reliability with time period t and failure rate λ [2] is given by: 182 advances in technology innovation, vol. 7, no. 3, 2022, pp. 181-194 ( ) t r t e    (1) where λ = 1/mtbf. mtbf refers to the mean time between failures. from eq. (1), it can be inferred that as system failure resistance decreases, its reliability decreases. the term “availability” is used to indicate the probability of a system or equipment being in operating condition at any time t, given that it is in operating condition at t = 0 [2]. to make a system in operating condition at time period t, it must be functional. in other words, it must not fail, and if there is a failure at t, it must be repaired. thus, availability involves the characteristics of both reliability and maintainability. the latter is defined as the probability of repairing a failed system or component in a specified period [2]. availability features allow the system to stay operational even when faults occur, whereas reliability means it is likely to perform perfectly, and maintainability implies that even if something does go wrong, it can be rectified effortlessly [2-3]. the general expression for the system availability a is given as eq. (2) [2]: 1 1 mtbf a mtbf mttr mttr      (2) where mttr refers to the mean time to repair. vibration monitoring might be considered the grandfather of cbm/pdm and provides the foundation for the cbm programs of most facilities [3, 8-9]. bianchini et al. [10] conducted an experimental study for cbm through vibration monitoring on submersible well pumps following vibration severity standards such as iso 10816-7 (2009). however, in their study, the availability of pumps corresponding to the vibration level was not set. sulaiman et al. [11] investigated the effect of high vibration (above acceptable limit) in gas turbines installed in al ghubra power & desalination company and suggested using online condition monitoring methods to determine the running condition of gas turbines in advance to plan maintenance activities to avoid sudden failure of the turbine. in industrial situations, the process plants often operate with certain machines, which can be classified as class i, ii, iii, and iv on the basis of vibration severity standards iso 10816 and whose failure is mainly due to severe vibrations. the standard provides a reference for evaluating vibration severity in machines operating over the range of 600 to 12,000 rpm [8-9]. these machines are often subjected to more than the acceptable limits of vibration, which invariably leads to an increase in their failure rate, especially in the case of components with rotating parts. rotating machinery is extensively used in today’s industries, and some of these are extremely critical for the successful operation of the plant. the machine collapse may result in costly downtime of the plant. the faults in rotating machinery, such as machine being out of balance or alignment, gear fault, resonance, bent shafts, bearing failures, mechanical looseness, can be identified by measuring and analyzing the vibration generated by the machine [8-9]. the rolling element bearings in rotating machinery allow a relative movement and bear shaft load [12]. the life of a bearing depends on its use, and its expected life is solely based on experience [13]. the main cause of failure of industrial bearing is rolling contact fatigue (rcf). the rcf wear mechanism involves false brinelling, characterized by plastically formed indentations, which are generally caused by vibration due to overload. the cause of the wear is that lubricant is squeezed out between the contact area of rolling elements and raceways, resulting in direct metal-to-metal contact. vibration causes wear of the surfaces in contact, and fine abrasive particles are rapidly produced, which results in a characteristic groove with the oxide acting as an abrasive [14]. failure analysis of the bearing can be investigated by making artificial defects on various elements of the bearing and analyzing it with a vibration signature tool for monitoring its condition [15]. usually, life testing under normal operating conditions of mechanical parts with high reliability is expensive in terms of both capital and time. hence, it is desirable to accelerate the testing procedure for gathering the failure data. the chief objective of accelerated testing procedures is to reduce the time required for life testing by strategies such as intensified stress levels. 183 advances in technology innovation, vol. 7, no. 3, 2022, pp. 181-194 physical acceleration or true acceleration means that operating a unit at high stress produces the same failures that would occur at typical use stresses, except that they happen much quicker. then, by extrapolating the results suitably to the “normal use” conditions, a reasonably accurate estimate of the life of the component under the “normal use” conditions can be obtained [2, 16]. however, the issue of prediction accuracy associated with extrapolating data outside the range of testing has not yet been addressed comprehensively [16]. acceleration factors show how the time -to-fail at a particular operating stress level (for one failure mode or mechanism) can be used to pred ict the equivalent time-to-fail at a different operating stress level [17-18]. thus, an efficient diagnosis system is needed to predict the condition and consistent lead time of the machine. vibration analysis is a method used for monitoring the condition of the machine. effective vibration signal extracting techniques, therefore, have a critical part in diagnosing a rotating machine. in detecting early fault generation, frequency domain features in the vibration signals are generally more effective as comp ared to the time domain features [19]. tools such as neural networks, hybrid systems, and fuzzy logic are being employed for increasing the effectiveness of fault diagnosis [20]. it is normally accepted that the vibration velocity recorded over the range of 10 hz (600 rpm) to 1000 hz (60,000 rpm) provides the finest indication of a vibration’s severity on rotating machines [8-9, 11]. as most of the rotating machines operate between 600 rpm and 60,000 rpm, the vibration velocity is the best candidate for vibration measurement and analysis [8, 11], whereas above 60,000 rpm vibration acceleration is only the fine indicator [8, 11]. the rms value is directly related to the energy content of the vibration and thus its destructive capability [8-9]. the indicators used for the initiation of cbm involve vibration signatures, temperature changes, and process parameters of machining such as spindle speed, depth of cut, and feed rate. by the analysis of these indicators, the failure of the machine can be predicted, and the corrective measures must be performed [21]. the process parameter used in the proposed model is vibration velocity (rms), and its influence on system availability was analyzed. the remaining useful life (rul) is the life period left on any system of machines at a particular time of operation [3]. from a specific time of operation of the system, the rul is the time period up to its functional failure. by taking rul into account, a plant engineer can schedule system maintenance, which can avoid unexpect ed system downtime. because of this, the prediction of rul has the highest priority in cbm programs. the method used to calculate rul depends on the kind of data available [3]. coppe et al. [22] proposed the use of a simple crack growth model is proposed t o predict the system rul influenced by the fatigue failure mechanism. kang et al. [23] implemented an ml -based approach for automating the prediction of rul of equipment in continuous production lines. han et al. [24] developed a method for rul prediction for manufacturing systems using a mission reliability-oriented approach based on the functional dependence of components. in this study, the alert limit of vibration, alarm limit of vibration, and the system availability are considered for defining the maintenance initiation points of cbm for class i machines of specific power rating. a relation is established between the system availability and the process parameter vibration velocity (rms). it is observed that the system availability starts to decline at a point much before the alarm limit of vibration (potential failure), and the maintenance is initiated at this point for the proposed model. from this point onward, the reliability begins to drop rapidly. the time available for maintenance initiation can be predicted from a given level of vibration, which is less than the defined value of maintenance initiation vibration level. the functional failure occurs beyond the potential failure and is not considered in this study [3]. it is evident that even though the relation between vibration velocity and system availability can be deduced, no study has predicted the time available to initiate the cbm from a given level of vibration of the machine [4, 25]. in this study, an attempt is made to ascertain the time available for initiating the cbm by having a measure of the vibration level. 184 advances in technology innovation, vol. 7, no. 3, 2022, pp. 181-194 3. experimental setup and design of alt according to the vibration severity standards table for iso 10816 [8-9], class i industrial machines are individual parts of engines and machines, integrally connected with the complete machine in its normal operating condition (electrical motors of up to 20 hp are typical examples of machines in this category). according to iso 10816, the speed of rotation of class i machines lies in the range of 600 to 12,000 rpm. on the one hand, the upper limit of vibration for a good operating condition, defined as the good operating limit, is 0.71 mm/sec, and for a satisfactory condition, defined as an acceptable or alert limit (failure initiation), is 1.8 mm/sec (table 1). on the other hand, the corresponding value for the unsatisfactory condition of vibration level, defined as the alarm limit (potential failure), is 4.5 mm/sec. above the alarm limit, potential failure leads to a functional failure, and some functions of the asset stop working [3]. therefore, the failure probability of machine components subjected to vibration can be defined such that the failure probability is zero for vibration levels below the alert limit and 100% as it reaches the alarm limit. three-phase squirrel-cage induction motors are widely used as industrial drives because they are self-starting, reliable, and economical [8]. around 70% of the failures that occur in electric motors are mechanical in nature [26]. the bearing failure is the leading cause (~51%) of most mechanical failures [27]. the causes of bearing failure include excessive loads, overheating, true and false brinelling, spalling, contamination, lubricant failure, loose and tight fits, corrosion, and misalignment [9]. the contamination is caused by foreign substances getting into bearing lubricants or cleaning solutions. examples of such foreign substances include dirt, abrasive grit, dust, and steel chips from contaminated work areas. solid particle contamination is a serious problem in all industrial sectors, which causes wear in contact surfaces [28]. the wear happens in the inner and outer races of the bearing, and when the ball passes over these races, the phenomenon called “ringing” occurs, which is similar to the vibrations in a car moving on an irregular road. as wear increases, vibration increases. the vibration standards iso 10816 can be used as a reference for evaluating the severity of vibrations. the proposed model creates the wear in the raceways of the bearing of a class i machine intentionally, and the resultant vibrations above the alert limit can be measured and recorded using vibration measurement systems. figs. 1(a)-(b) indicate the line diagram and the photograph of the practical setup used for experimentation. it consists of a delta-connected three-phase induction motor, a long shaft with necessary detachable couplings at the ends, bearings and brackets, and a mechanical loading mechanism with cooling accessories. the rated voltage, current, power, rpm, and frequency of the motor used are 415 v, 7.1 a, 5 hp, 1440 rpm, and 50 hz, respectively. similarly, the shaft material used is en32 grade steel with a diameter of 18 mm and a total length of 75 mm. the shaft is connected to the motor using a flexible element jaw coupling, and a spider rubber bush is placed on the inside of this coupling. the shaft is simply supported at the two bearings, which are placed inside two bearing brackets. the distance between the bearing ends is 65 cm. a brake drum is mounted on the shaft at a distance of 56 cm from the bearing bracket 1. the bearings used are deep groove ball bearing skf 6202 z. the inside diameter of the bearing is 15 mm, and accordingly, the diameter of the shaft when it passes through the bearing portion is reduced to suit this value. as per the data sheet of skf 6202 z bearing, the fatigue load limit (maximum radial load) is 0.16 kn. for analysis purposes, the bearings fitted inside the bearing brackets 1 and 2 are referred to as bearing 1 and bearing 2, respectively. during experiments, the failure of bearing 2 is created by developing the wear in it by adding c10 coarse (10-micron grain size) silicon carbide paste between its inner and outer races. table 1 iso 10816 vibration severity levels for class i machines operating condition of class i machines vibration velocity in mm/sec (rms value) good 0.28 to 0.71 satisfactory (acceptable) 1.12 to 1.80 unsatisfactory (monitored closely) 2.80 to 4.50 unacceptable 7.10 to 45.90 185 advances in technology innovation, vol. 7, no. 3, 2022, pp. 181-194 (a) line diagram of the experimental setup (b) practical setup fig. 1 experimental setup for vibration measurement in class i machines the shaft design is based on the combined torque and bending moment (static analysis). on the basis of the design, the appropriate shaft diameter is calculated as 18 mm. a modal analysis of the combined system involving the shaft and the brake drum is performed using ansys software. the analysis revealed “n” modes of vibration and corresponding natural frequencies. the natural frequencies of the system obtained for the first and second modes are 166 hz (9,960 rpm) and 332 hz (19,920 rpm), respectively. the natural frequencies are observed to have an increasing trend for further modes of vibration. in this study, a variable frequency drive (vfd) is used between the motor and input three-phase supply to fix the motor speed at 1,465 rpm. also, the ratio of the motor’s applied voltage and applied frequency is set as 8.3 v/hz. as the natural frequency values obtained for all the modes of vibration for the designed system are noted to be much higher than 1,465 rpm, the resonance condition is avoided to make the design safe. the loading arrangement used is illustrated in fig. 2 and consists of a brake drum, brake shoe, and loading wheel. the material of the brake shoe is compressed asbestos, and that of the brake drum is en8 grade steel. the brake drum is connected to the shaft at a distance of 56 cm from the bearing bracket 1. the brake drum has a diameter of 150 mm and a thickness of 40 mm. the loading of the motor is achieved by turning the loading wheel, which makes the brake shoe inside the brake shoe bracket to press against or pull away from the rotating hollow brake drum. the system load applied is noted from the load indicator dial. the hinge, brake shoe bracket, and bearing bracket are made of mild steel. the electric motor, bearing brackets, brake shoe bracket, load indicator, hinge, and cooling water arrangement of the brake drum are mounted on a frame made using a light gauge rectangular hollow cross section galvanized iron tube (is1239). rubber bush dampers are used for vibration isolation from the floor. during loading, because of the friction between brake drum and brake shoe, heat is generated inside the hollow brake drum. it is removed by constantly circulating cooling water through the hollow brake drum. the water flow is controlled by a ball valve. the temperature of water at the inlet and exit are measured using k-type thermocouples, and the steady readings obtained are 27°c and 63°c, respectively. fig. 2 line diagram for loading arrangements 186 advances in technology innovation, vol. 7, no. 3, 2022, pp. 181-194 the alt conducted in this experiment follows the procedure detailed in regattieri's work [29]. the experimental setup is designed such that the failure of bearing 2 is the reason for the failure of the system. experimental trials are conducted by continuous measurement of the rms vibration values of the system from the moment the silicon carbide paste is added to the final failure state of the system. during each trial, under a constant radial load, the failure of the system is created by developing the wear in bearing 2. the failure so created is assessed in terms of the induced vibrations of the order of the magnitude above alarm limit vibrations (glut vibrations) in the system by comparing with the vibration severity standards (iso 10816). the trials are repeated by changing the radial load on bearing 2 from 14 kg to 18 kg in steps of 1 kg. eight sets of experiments are conducted under each trial to ensure repeatability. these radial loads are selected such that the fatigue limit of the bearing is never exceeded. the loads on bearing 2 are calculated considering equilibrium conditions of the shaft and assuming a simply supported configuration at bearings 1 and 2. during the experiments, after the motor starts, the system steadies itself in a few seconds and begins to operate in a stable condition. the vibration level is measured to ensure that there is no vibration interference, such as the vibration of the mounting of the load indicator in the system. thereafter, 10 g silicon carbide paste is added in bearing 2 between its inner and outer races from the outside to develop the wear in it. the paste is added thrice at the intervals of 5 minutes for every set of the experiment. this results in the failure of the system as observed from the recorded rms vibration values (table 1) when the alarm limit is reached. the system used in the measurement, preprocessing, storage, and postprocessing of the vibration signals consists of an accelerometer sensor, compactdaq, a personal computer (pc), and the labview software (version 2017). compactdaq, a data acquisition platform built by national instruments, includes a broad set of compatible hardware and software; it is a usb-powered plug-and-play type and requires no data card. compactdaq integrates hardware for data i/o with labview software to enable engineers to collect, process, and analyze the sensor vibration data. the accelerometer sensor used is a uniaxial integrated electronic piezoelectric (iepe; made by pcb piezotronics). the setup, used to sense the vibrations in a radial direction perpendicular to the axis of the rotating shaft, is mounted magnetically above the bearing bracket 2, where the maximum level of the vibration is obtained. the sensor has the sensitivity of 100 mv/g, measurement range ±50g, resonant frequency 25 khz, and frequency range 0.5 hz to 10 khz. compactdaq is used to acquire and process the analog signal coming from the uniaxial accelerometer sensor. the signal conditioner, which is inbuilt in the data acquisition (daq) module, removes noise from the signal and supplies a constant current excitation of 2.1 ma to the accelerometer sensor. compactdaq connects to the accelerometer sensor via a wired i/o daq module. a shielded twisted pair (stp) cable is used to carry the analog signal from the accelerometer sensor to the daq module connected to the daq chassis, which, in turn, connects a pc where labview software is installed (fig. 3). the function of the daq chassis includes the synchronization and transfer of digital signals from the daq module into the computer. the synchronization involves determining or enforcing and ordering events on signals. the digital signals from the daq chassis can be stored as binary values in the computer. a military standard connector is used to connect the accelerometer sensor to the stp cable. the signals from the daq chassis are transmitted through a shielded usb cable to the computer. in this experiment, the labview software band-pass filter passes the frequencies of vibration data in the range of 200 hz to 25 khz. this software is used to convert the measured value of instantaneous vibration accelerations into instantaneous vibration velocities by integration, and from these instantaneous values, rms values of vibration velocities are calculated. the rms value of 51.2k samples of vibration velocity is calculated at each second. fig. 3 block diagram for vibration measurement setup in the proposed system 187 advances in technology innovation, vol. 7, no. 3, 2022, pp. 181-194 the daq module used in this experiment is ni 9234 with ni 9171. ni 9171 is the usb interface for powering ni 9234. the daq module ni 9234 has the maximum sampling rate of 51.2 kilo samples/sec(fs) with an inbuilt hardware-based anti-aliasing filter (low pass) with a cut-off frequency of 0.45 × fs (= 23.04 kilo samples/sec). the specifications of the daq module are enumerated as follows: (1) four analog input channels (2) ±5v input range (from accelerometer sensor to daq module) (3) simultaneous sampling (4) ac/dc coupling (5) operating temperature: -40°c to 70°c (6) maximum allowed vibration and shock are 5 g and 50 g, respectively. (7) analog-to-digital converter (adc) amplitude resolution: 24-bit 4. failure analysis, discussion, and results the following assumption is made during failure analysis: (1) failure and repair rates for each subsystem are constant and statistically independent [1]. (2) failure of machines with moving parts is caused by vibrations alone and occurs when the vibration velocity is above the acceptable or alert limit of vibration. while loading, a lot of heat is created inside the hollow brake drum because of the friction between brake drum and brake shoe. this heat is removed using a cooling water arrangement. the cooling water from the tank enters inside the hollow brake drum through a pvc pipe, and its flow is controlled by a ball valve. the centripetal force produced by the rotation of the brake drum pushes the hot water inside the drum to exit through a copper tube. the temperature of hot water coming from the copper tube is measured as 63°c. (3) the parameter for measuring mechanical vibration is taken as a velocity of vibration. the velocity is the best representation of true energy generated by a machine when the relative or bearing cap data are used [9]. (4) availability under consideration is inherent availability or steady-state availability. the steady-state availability represents the long-term availability after the system settles. at the early stages of operation or before the system settles, it may be wobbly because of the issues in (i) employee training, (ii) determining a genuine/good spare part keeping policy, (iii) determining the number of maintenance personnel, and (iv) optimizing the effectiveness/efficiency of maintenance. (5) the life of the bearing is assumed to be 12,000 hours and is taken as the reference life in the analysis [13]. this reference life corresponds to the maximum radial load acting on the bearings used in the experiment. (6) the transformations used are linear, i.e., the time-to-fail at high stress is multiplied by a constant (the acceleration factor) to obtain the equivalent time-to-fail at use stress [17-18]. (7) the bearing failure (potential) probability is 0% below the alert limit of vibration, and it is 100% at and above the alarm limit. according to vibration severity standards iso 10816 shown in table 1, the satisfactory/acceptable vibration level in class i machines is in the range 1.12 to 1.80 mm/sec (rms). the unsatisfactory level (monitor closely) of vibration is in the range 2.80 to 4.5 mm/sec (rms). the vibration level 1.80 mm/sec (rms) can be defined as alert limit or failure initiation stage as per 188 advances in technology innovation, vol. 7, no. 3, 2022, pp. 181-194 p-f curve for class i machines [3]. the vibration level 4.5 mm/sec (rms) can be defined as alarm limit or potential failure stage as per p-f curve. therefore, the system failure probability is zero below the alert limit, and 100% potential failure happens at and above the alarm limit. in this study, bearing 2 is subjected to alt by causing the wear in it by adding some silicon carbide paste. the bearing fails because of the excessive wear so created. as the wear in the bearing increases, the vibration in the system also increases. during the experiment, the rms value of the vibration velocity for a particular system load is recorded after the silicon carbide paste is added. the measurement of rms vibration velocity continues until the alarm limit of vibration is reached. the availability and reliability of the system are calculated by extrapolating the time corresponding to this recorded rms value of the vibration velocity to normal-use conditions. the rated life of a ball bearing in working hours can be expressed as [13-14]: 3 6 10 10 60 h c l p n          (3) where l10h = rated bearing life in hr; c = dynamic load capacity (n; constant for identical bearings); p = load applied on the system (n); n = speed of rotation of the motor (rpm). in the industrial application of bearings, the speed of rotation is relatively constant, and the desired life is expressed in terms of hours of service. in this study, the bearing speed is kept constant at 1,465 rpm. the expected life for bearings in industrial applications is in the range of 12,000 to 20,000 hr [13]. if two groups of identical ball bearings are tested under loads p1 and p2 for respective lives l1 and l2, then 3 1 2 2 1 l p l p        (4) the data sheet for the skf 6202 z bearing (bearing 2) mentions that the maximum radial load that can be applied on it is 160 n. in the experimental setup, the shaft is simply supported at the two bearing ends. when the radial load applied on the system is 18 kg (176.58 n), by using equilibrium equations for the shaft configuration, the load acting on bearing 2 is found to be 15.51 kg (152 n). when l1 = 12,000 hr and p1 = 160 n, the life corresponding to the load of p2 = 15.51 kg (152 n) on bearing 2 is calculated as l2 = 13996.21 hr. similarly, the life of the ball bearing for different system radial loads (m kg) can be calculated and is encapsulated in table 2. table 2 life of bearing calculated for different system loads radial load applied on the system (m kg) radial load on bearing 2, whose failure due to excess vibration is considered (kg) life of bearing 2 based on the reference life of 12,000 hr (hr) 14 12.06 29,771.38 15 12.92 24,212.20 16 13.78 19,959.70 17 14.65 16,609.25 18 15.51 13,996.21 4.1. experimental data and failure analysis eight trials of the experiment are performed for a particular system load to ensure repeatability. each trial involved the time series vibration level measurement up to the alar m limit. thus, a total of 40 trials are conducted for the system loads varying from 14 to 18 kg. by performing the error analysis for the data collected, inappropriate trial measurement is removed. 189 advances in technology innovation, vol. 7, no. 3, 2022, pp. 181-194 furthermore, a total of 40 bearings fail during the experiment. the silicon carbide paste is added between the inner and outer races of the bearing to create the wear. the common bearing defect is the indentations in the raceways, and when the ball passes over the raceways, mechanical vibration is induced. table 3 summarizes the measured failure data of vibration velocity (between alert to alarm limits), the time between failures, and the projected useful life period for bearing 2 under the 14 kg system load obtained for each of the six different trials. the estimated projected time for the bearing for its useful life period is calculated following assumption 6 for the analysis. for an applied system load of m kg, the useful life period of the bearing in the system l hr (during its normal operating conditions) can be calculated from the observed failure time of t s (during experimentation) as given by eq. (5) [17-18]: 3600 m af t l   (5) where af m (the acceleration factor for system load m kg) is the proportionality constant. for a system load of 14 kg, the value of the acceleration factor used in the analysis is af 14 = 34,786.423. the experiment is repeated for the system loads of 15 kg, 16 kg, 17 kg, and 18 kg in a similar way, and the values are calculated. using curve-fitting techniques, the relation between the vibration velocity and the projected time for the system loads from 14 kg to 18 kg is found to be parabolic in nature with goodness of fit, r 2 ≥ 0.95. the relationship can be written as: 2 p p v at b t c    (6) where v represents the vibration velocity recorded in mm/hr between alert to alarm limits for the system load (m kg); tp represents the projected time calculated in hrs using the acceleration factor (af m ) corresponding to the vibration velocity (v). the system load (m kg), coefficients (a, b, and c) in eq. (6), goodness of fit (r 2 value) for eq. (6), and acceleration factor (af m ) obtained from this study are summarized in table 4. using curve-fitting techniques, the relation between the vibration velocity v (mm/hr) and the average value of projected time for system loads from 14 kg to 18 kg, tpa (hr), is found to be parabolic in nature with the r 2 value 0.9986. the relationship can be expressed as: 2 0.00003 0.0583 2169.7, 6480 16200 pa pa v t t for mm hr v mm hr       (7) for any given vibration velocity, the corresponding projected time can be obtained by solving eq. (7). the values of vibration velocity in mm/hr and the corresponding values of mtbf and failure rate for the 14 kg system load are encapsulated in table 5. the mtbf (12,998.2 hr) corresponding to vibration velocity 6,480 mm/hr (1.8 mm/sec) is obtained by deducting the value of average useful life period of the bearing, 16,773.18 hr (corresponding to 6,480 mm/hr) from the useful average life 29,771.38 (corresponding to 16,200 mm/hr). similarly, the mtbf (11,492.4 hr) corresponding to vibration velocity 7,200 mm/hr is obtained by deducting the value of average useful life period of the bearing, 18,278.98 hr (corresponding to 7,200 mm/hr) from the useful average life 29,771.38 (corresponding to 16,200 mm/hr). likewise, other values from 10,255.55 hr to 0.00 hr are obtained. in accordance with table 5, the variation of failure rate (per hr) of the experimental setup of class i machine with vibration velocity (mm/hr) for the system load of 14 kg, can be drawn as shown in fig. 4. the system availability can be calculated using the eq. (2). the repair corresponding to bearing failure involves (a) changing the bearing and (b) (occasionally) changing the shaft. the mttr is calculated as 47 min (0.783 hr). 190 advances in technology innovation, vol. 7, no. 3, 2022, pp. 181-194 table 3 the measured failure data and the projected useful life period for bearing 2 under the 14 kg system load vibration velocity (mm/hr) time (s) projected time for bearing for its useful life period (hr) trial 1 trial 2 trial 3 trial 4 trial 5 trial 6 average value 6,480 1,902 1,734 1,700 1,692 1,694 1,693 1,735.83 16,773.18 7,200 1,988 1,914 1,870 1,858 1,860 1,860 1,891.67 18,278.98 7,920 2,074 2,026 2,008 2,002 2,004 2,004 2,019.67 19,515.83 8,640 2,200 2,176 2,164 2,162 2,162 2,162 2,171.00 20,978.15 9,360 2,268 2,270 2,270 2,270 2,269 2,270 2,269.50 21,929.94 10,080 2,374 2,374 2,374 2,374 2,374 2,373 2,373.83 22,938.10 10,800 2,438 2,448 2,454 2,454 2454 2,454 2,450.33 23,677.31 11,520 2,496 2,520 2,526 2,528 2,528 2,528 2,521.00 24,360.16 12,240 2,570 2,614 2,622 2,624 2,623 2,622 2,612.50 25,244.31 12,960 2,714 2,733 2,738 2,740 2,739 2,738 2,733.67 26,415.13 13,680 2,806 2,838 2,846 2,848 2,848 2,847 2,838.83 27,431.35 14,400 2,896 2,934 2,944 2,946 2,946 2,945 2,935.17 28,362.21 14,760 2,936 2,969 2,980 2,982 2,982 2,982 2,971.83 28,716.51 15,120 2,968 3,004 3,014 3,016 3,015 3,016 3,005.50 29,041.83 15,480 2,998 3,034 3,042 3,044 3,043 3,044 3,034.17 29,318.83 15,840 3,028 3,060 3,068 3,069 3,068 3,069 3,060.33 29,571.68 16,200 3,060 3,086 3,091 3,075 3,090 3,084 3,081.00 29,771.38 table 4 the system load, coefficients, goodness of fit, and acceleration factor obtained from this study system load (m kg) coefficients goodness of fit (r 2 value) acceleration factor (af m ) a b c 14 +1.7491 × 10 −5 −0.0803 +2828.30 0.997 34,786.423 15 +2.2053 × 10 −5 −0.0313 +3975.10 0.999 32,723.243 16 −1.4072 × 10 −5 +1.4825 −7885.70 0.987 32,694.021 17 +2.2995 × 10 −5 +0.5762 −294.48 0.989 33,196.922 18 +2.4260 × 10 −4 −3.7696 +21356.00 0.997 37,301.122 table 5 the vibration velocity, mtbf, and failure rate of the system for the 14 kg system load vibration velocity (mm/hr) mtbf (hr) failure rate (per hr) 6,480 12,998.20 0.000076934 7,200 11,492.40 0.000087014 7,920 10,255.55 0.000097508 8,640 8,793.23 0.000113724 9,360 7,841.44 0.000127528 10,080 6,833.28 0.000146343 10,800 6,094.07 0.000164094 11,520 5,411.22 0.000184801 12,240 4,527.07 0.000220894 12,960 3,356.25 0.000297952 13,680 2,340.03 0.000427345 14,400 1,409.17 0.000709637 14,760 1,054.87 0.000947988 15,120 729.55 0.001370711 15,480 452.55 0.002209724 15,840 199.70 0.005007524 16,200 0.00 ∞ 191 advances in technology innovation, vol. 7, no. 3, 2022, pp. 181-194 fig. 4 variation of failure rate (per hr) with vibration velocity (mm/hr) for the 14 kg system load fig. 5 variation of system availability with vibration velocity for different system loads fig. 5 shows the variation of system availability with vibration velocity (mm/hr) for different system loads from 14 kg to 18 kg. it is found that there is a slight decrease in availability between alert and alarm limit of vibration, and at the latter, there is a sudden decrease in availability to a value of zero for all system loads. it can be noted that at the maintenance initiation point, the system availability begins to decline. this is because, from this point onward, the reliability starts to decrease rapidly. the value of the maintenance initiation point for possible system loads is obtained as 12,176 mm/hr (3.4 mm/s). the value 3.4 mm/s is considered to be the maintenance initiation point for a system of class i machines. by considering all the possible system loads, the relation between the time available for maintenance initiation ta (hr) and the vibration velocity v (mm/hr) is given by: 2 0.00008 2.5944 20196 , 6480 12176 a t v v for mm h m hrr mv     (8) 4.2. error analysis keeping the system configuration and applied load unaltered, eight trials of the experiment are conducted, and the rms vibration velocity values and the corresponding times are recorded. using grubbs’ test at a significance level (α) = 0.02, the trials which involve outliers are detected and removed for a particular system load. the trials that are removed are two trials for the system loads of 14 kg, 15 kg, and 17 kg, and three trials for the system loads of 16 kg and 18 kg. the remaining trials of time (corresponding to the vibration velocity between alert and alarm limits) for the system loads of 14 kg, 15 kg, 16 kg, 17 kg, and 18 kg are subjected to error analysis to find the margin of error at 95% confidence interval. the maximum value of percentage margin of error (at 95% confidence interval) in the average time is calculated as 3.5%, which corresponds to the average time of 1735.83 s (corresponding to the vibration velocity 1.8 mm/s) for the system load of 14 kg. the minimum value of the percentage margin of error in the average time is 0.013%, which corresponds to the average time of 2373.83 sec (corresponding to the vibration velocity 2.8 mm/sec) for the system load of 14 kg. 0 0.001 0.002 0.003 0.004 0.005 0.006 6480 8480 10480 12480 14480 16480 f a il u re r a te ( p e r h r) vibration velocity (mm/hr) 192 advances in technology innovation, vol. 7, no. 3, 2022, pp. 181-194 fig. 6 vibration velocity vs the average time corresponding to system loads (14 kg to18 kg) with error bars fig. 6 depicts the variation of vibration velocity between the alert and alarm limits against the average time of different trials (recorded in alt of bearing) for system loads from 14 kg to 18 kg. error bars (corresponding to the margin of error) in the average time are also shown in this figure. 5. conclusions in this experimental study, the failure of the machine was assessed by considering vibrations above the alert limit. accelerated life tests were conducted by creating a severe environment of high stress in the bearing of a class i machine system for experimentation. by a suitable external mechanism, the wear rate in the bearing was increased rapidly until failure occurred. the resulting vibration data were captured as a time series for analysis. the vibration was recorded as vibration velocity (rms). using a linear transformation, the time recorded was converted to the corresponding value in its normal use. a suitable loading mechanism was designed to apply possible magnitudes of system load. the major contributions of the proposed experimental study are summarized below. (1) it is demonstrated that the cbm initiation point of vibration velocity can be accurately predicted by the analysis of the pattern of system availability with vibrations to ensure the long life of the process plant machines and to enhance the overall performance of the plant. (2) the proposed model established a relationship between the vibration velocity and the projected useful life period of the bearing for the system of the electric motor with rated power 5 hp (class i machine) applied with different possible system loads. for a given value of maintenance initiation point of vibration velocity and a possible system load, the corresponding life of the bearing in the system can be computed using the proposed relation to conduct the systematic maintenance of the machines effectively. (3) in the proposed model, a general equation is developed between the time available for maintenance initiation and vibration velocity for the given system by considering different possible system loads. (4) the experimental study revealed that in cbm, the vibration parameter can be used as a reliable characteristic to predict the maintenance initiation point. conflicts of interest the authors declare no conflict of interest. references [1] h. d. goel, j. grievink, p. m. herder, and m. p. weijnen, “integrating reliability optimization into chemical process synthesis,” reliability engineering and system safety, vol. 78, no. 3, pp. 247-258, december 2002. [2] s. s. rao, reliability-based design, 1st ed., new york: mcgraw-hill inc., 1992. [3] r. gulati, maintenance and reliability best practices, 2nd ed., new york: industrial press inc., 2012. 193 advances in technology innovation, vol. 7, no. 3, 2022, pp. 181-194 [4] y. li, s. peng, y. li, and w. jiang, “a review of condition-based maintenance: its prognostic and operational aspects,” frontiers of engineering management, vol. 7, no. 5, pp. 323-334, july 2020. [5] j. h. shin and h. b. jun, “on condition based maintenance policy,” journal of computational design and engineering, vol. 2, no. 2, pp. 119-127, january 2015. [6] a. a. manjare and b. g. patil, “a review: condition based techniques and predictive maintenance for motor,” international conference on artificial intelligence and smart systems, pp. 807-813, march 2021. [7] s. kumar, d. mukherjee, p. k. guchhait, r. banerjee, a. k. srivastava, d. n. vishwakarma, et al., “a comprehensive review of condition based prognostic maintenance (cbpm) for induction motor,” ieee access, vol. 7, pp. 90690-90704, july 2019. [8] a. r. mohanty, machinery condition monitoring: principles and practices, 1st ed., usa: crc press, 2014. [9] r. k. mobley, root cause failure analysis, 1st ed., burlington: elsevier, 1999. [10] a. bianchini, j. rossi, and l. antipodi, “a procedure for condition-based maintenance and diagnostics of submersible well pumps through vibration monitoring,” international journal of system assurance engineering and management, vol. 9, no. 3, pp. 999-1013, february 2018. [11] s. k. s. a. adawi and g. r. rameshkumar, “vibration diagnosis approach for industrial gas turbine and failure analysis,” british journal of applied science and technology, vol. 14, no. 2, pp. 1-9, january 2016. [12] s. pattabhiraman, g. levesque, n. h. kim, and n. k. arakere, “uncertainty analysis for rolling contact fatigue failure probability of silicon nitride ball bearings,” international journal of solids and structures, vol. 47, no. 18-19, pp. 2543-2553, september 2010. [13] v. b. bhandari, design of machine elements, 4th ed., new delhi: mcgraw hill education india private limited, 2017. [14] r. k. upadhyay, l. a. kumaraswamy, and m. s. azam, “rolling element bearing failure analysis: a case study,” case studies in engineering failure analysis, vol. 1, no. 1, pp. 15-17, january 2013. [15] s. kulkarni and s. b. wadkar, “experimental investigation for distributed defects in ball bearing using vibration signature analysis,” procedia engineering, vol. 144, pp. 781-789, december 2016. [16] c. zhang, i. chuckpaiwong, s. y. liang, and b. b. seth, “mechanical component lifetime estimation based on accelerated life testing with singularity extrapolation,” mechanical systems and signal processing, vol. 16, no. 4, pp. 705-718, july 2002. [17] j. b. bernstein, reliability prediction from burn-in data fit to reliability models, 1st ed., london: elsevier, 2014. [18] nist/sematech, “engineering statistics handbook,” https://www.itl.nist.gov/div898/handbook/ april 2012. [19] m. vishwakarma, r. purohit, v. harshlata, and p. rajput, “vibration analysis and condition monitoring for rotating machines: a review,” materials today: proceedings, vol. 4, no. 2, pp. 2659-2664, december 2017. [20] a. boum, n. y. j. maurice, l. n. nneme, and l. m. mbumda, “fault diagnosis of an induction motor based on fuzzy logic, artificial neural network and hybrid system,” international journal of control science and engineering, vol. 8, no. 2, pp. 42-51, august 2018. [21] b. k. p. kumar, y. basavaraj, m. j. sandeep, and n. k. kumar, “optimization of process parameters for turning machine using taguchi super ranking method: a case study in valve industry,” materials today: proceedings, vol. 47, no. 10, pp. 2505-2508, may 2021. [22] a. coppe, m. j. pais, r. t. haftka, and n. h. kim, “using a simple crack growth model in predicting remaining useful life,” journal of aircraft, vol. 49, no. 6, pp. 1965-1973, november 2012. [23] z. kang, c. catal, and b. tekinerdogan, “remaining useful life (rul) prediction of equipment in production lines using artificial neural networks,” sensors, vol. 21, no. 3, article no. 932, january 2021. [24] x. han, z. wang, m. xie, y. he, y. li, and w. wang, “remaining useful life prediction and predictive maintenance strategies for multi-state manufacturing systems considering functional dependence,” reliability engineering and system safety, vol. 210, no. 11, article no. 107560, june 2021. [25] e. quatrini, f. costantino, g. d. gravio, and r. patriarca, “condition-based maintenance—an extensive literature review,” machines, vol. 8, no. 2, article no. 31, june 2020. [26] a. j. bazurto, e. c. quispe, and r. c. mendoza, “causes and failures classification of industrial electric motor,” ieee andescon, pp. 1-4, october 2016. [27] r. karakolev and l. dimitrov, “analysis of electrical motor mechanical failures due to bearings,” iop conference series: materials science and engineering, vol. 393, no. 1, article no. 012064, august 2018. [28] g. k. nikas, “a state-of-the-art review on the effects of particulate contamination and related topics in machine-element contacts,” proceedings of the institution of mechanical engineers, part j: journal of engineering tribology, vol. 224, no. 5, pp. 453-479, march 2010. [29] a. regattieri, f. piana, m. gamberi, f. g. galizia, and a. casto, “reliability assessment of a packaging automatic machine by accelerated life testing approach,” procedia manufacturing, vol. 11, pp. 2178-2186, january 2017. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). 194 4___aiti#7853_in press advances in technology innovation, vol. 6, no. 4, 2021, pp. 235-250 a recursive least-squares approach with memorizing factor for deriving dynamic equivalents of power systems ali karami * faculty of engineering, university of guilan, rasht, iran received 06 june 2021; received in revised form 01 august 2021; accepted 02 august 2021 doi: https://doi.org/10.46604/aiti.2021.7853 abstract in this research, a two-stage identification-based approach is proposed to obtain a two-machine equivalent (tme) system of an interconnected power system for transient stability studies. to estimate the parameters of the equivalent system, a three-phase fault is applied near and/or at the bus of a local machine in the original multimachine system. the electrical parameters of the equivalent system are calculated in the first stage by equating the active and reactive powers of the local machine in both the original and the predefined equivalent systems. the mechanical parameters are estimated in the second stage by using a recursive least-squares estimation (rlse) technique with a factor called “memorizing factor”. the approach is demonstrated on new england 10-machine 39-bus system, and its accuracy and efficiency are verified by computer simulation in matlab software. the results obtained from the tme system agree well with those obtained from the original multimachine system. keywords: network reduction, transient stability, critical clearing time, system identification 1. introduction in modern power systems, for economic and security reasons, individual power companies are connected together with tie lines to form an interconnected system. sophisticated computer programs are usually used for analysis and control of large power systems. however, it is not necessary to model the whole power network when the aim is to investigate a specific phenomenon in a small part of the system. in addition, sometimes it is very difficult to model the entire system in detail due to the lack of complete system data and/or inaccuracies in the system data [1]. to reduce the computational requirement, it is highly desirable to model only a small part of the system called “internal system” with enough accuracy, and represent the rest part of the system known as “external system”, by an equivalent model. power system equivalencing (or reduction) is generally divided into two main groups, namely static equivalencing and dynamic equivalencing. the static equivalents are employed in investigating the steady-state problems of power systems such as power flow and system planning studies. the dynamic equivalents are used in dynamical studies of power systems, including large and/or small disturbance stability problems as well as dynamic security assessment. this study is concerned with dynamic equivalencing of power systems for transient stability studies [2-3]. to reduce the computational burden of transient stability analysis, a dynamic equivalent is sometimes used for multimachine power systems [4]. however, a single-machine-infinite-bus (smib) system is usually employed for this purpose. an infinite bus is used to approximately model a large area in a power system [1]. an equivalent synchronous machine rather than an infinite bus was considered in the work of li et al. [5] to include the external system effect on the internal system. this early work was the main motivation for the present study. however, the method proposed in this study differs significantly from the method of this early work. * corresponding author. e-mail address: karami_s@guilan.ac.ir tel.: +98-13-33690274; fax: +98-13-33690271 advances in technology innovation, vol. 6, no. 4, 2021, pp. 235-250 the main objective of this study is to derive a two-machine equivalent (tme) system of a multimachine power system, by looking into the system from the bus of a local machine. the desired parameters of the defined equivalent system can be separated into two groups. the first group consists of the electrical parameters and the second group consists of the mechanical parameters of the tme system. to estimate the parameters of the equivalent system, a systematic two-stage identification-based approach is proposed. the electrical parameters of the equivalent system are calculated in the first stage of the approach by using an innovative and simple method. the mechanical parameters of the equivalent system are estimated in the second stage of the approach by employing a recursive least-squares estimation (rlse) technique with a factor called here “memorizing factor”. one advantage of the proposed methodology is that it is simple to understand. in addition, it is relatively easy to implement in a computer program. the effectiveness of the proposed methodology is demonstrated on the new england 10-machine 39-bus test system. the simulation results obtained are thoroughly described and discussed. the rest of the study is organized as follows. section 2 presents the literature review. section 3 represents a brief description of the rlse algorithm. the multimachine power system model and problem statement are described in section 4. section 5 presents the first stage of the proposed approach. the second stage of the proposed approach is described in section 6. the simulation results are presented in section 7. section 8 discusses the performance of the approach, and finally section 9 is devoted to the concluding remarks. 2. literature review research in power system reduction dates back to several decades ago; however, only a few references are mentioned here due to space limitation. the most well-known methods of static equivalencing are ward equivalent [6] and radial equivalent independent (rei) equivalent [7]. the ward equivalent is based on gaussian elimination technique, and the rei equivalent is indeed an extension of the ward equivalent. these methods separate a power system into three parts: internal, boundary, and external systems. the size of power system model is decreased by replacing the external system with small equivalents, whereas the internal system remains unchanged [1]. an extended ward equivalent method has been used for power system contingency analysis and static security assessment in the work of srivani et al. [8]. fu et al. [9] have proposed a hybrid method based on artificial neural network (ann) and ward-type equivalency for fast and online voltage security assessment of power systems. the proposed approach possesses good properties of ward equivalent method, and can update the parameters of the equivalent model for representing real-time topology change of the external system. lee et al. [10] have presented a static network equivalent model for korean power systems. the proposed equivalent model preserves the overall transmission network characteristics focusing on power flows among the areas in korean power systems. the already existing methods of dynamic equivalencing are modal methods, coherency methods, and measurement or simulation-based methods [11]. modal methods are based on the linearized state-space model of power systems. in the modal methods, the system equations are first linearized. then, the eigenvalues, eigenvectors, and participation factors of the linearized equations are employed to eliminate the less relevant part of the system [12]. paternina et al. [13] have proposed a strategy for developing dynamic equivalents in large-scale power system by mode preservation. using the state-space representation of power systems, other reduction techniques have also been investigated in the literature. rergis et al. [14] have proposed a general approach for constructing reduced-order models of power systems, which is based on a truncated balanced realization (tbr) linear reduction procedure. chaniotis et al. [15] have described the use of krylov subspace methods in the model reduction of power systems. zhu et al. [16] have proposed a method for large-scale power system model reduction based on the extended krylov subspace technique. ritschel et al. [17] have reported a balanced truncation model reduction approach for reducing two commonly used nonlinear and dynamical models of power grids. 236 advances in technology innovation, vol. 6, no. 4, 2021, pp. 235-250 in coherency methods, coherent groups of generators are first identified by using a suitable technique. then, the generators in coherent groups are represented by an equivalent generator [18]. unlike the modal methods, the coherency methods retain the physical models of the generators in an equivalent form [1, 11]. ayon et al. [19] have presented a methodology to identify coherent areas based on sliding power spectral density algorithm. miah [20] has proposed a coherency-based dynamic equivalent of a power system external area for transient stability assessment. based on the use of coherency aggregation techniques and nonlinear black-box optimization, matar [21] has developed a dynamic model reduction methodology. chittora et al. [22] have proposed a coherency based dynamic reduction technique for reducing the size of the power system network, and have tested it on a network having high voltage direct current (hvdc) transmission line. in measurement or simulation-based methods, the external system response is either measured or simulated, and system identification or curve-fitting techniques are used to determine the model parameters [11]. in the system identification based methods, a structure is first defined for the equivalent system. the measured data are then utilized to estimate the parameters of the predefined equivalent system. these methods require no knowledge of the parameters and topology (structure) of the external system. stankovic et al. [23] have proposed an ann-based strategy for the identification of the reduced-order dynamic equivalents, which uses only measurements at points where the internal and external systems are interfaced. chakrabortty et al. [24] have described an algorithm for constructing simplified interarea models of large power networks, using dynamic measurements available from phasor measurement units (pmus). ramirez et al. [25] have proposed a robust dynamic reduction approach to reduce the computational burden and time of transient stability studies. rueda et al. [26] have introduced an identification approach of the dynamic equivalent parameters, from measured operational dynamic responses associated to different disturbances. in order to increase both model accuracy and simulation speed, zhang et al. [27] have proposed a measurement-based dynamic equivalent. tong et al. [28] have proposed an artificial neuro-fuzzy inference system (anfis)-based dynamic reduction approach, in which the wide-area measurements obtained by pmus at the boundaries between the reduced system and the study system have been used to represent the external area. jiang et al. [29] have proposed a measurement-based reduction approach using system identification techniques, in which instead of a single event, multiple grid events with different patterns have been considered for estimating the parameters of the equivalent loads. almeida et al. [30] have reported the development of three software tools for the computation of dynamic equivalent models of power systems. the available techniques for power system dynamic reduction suffer from two major drawbacks: (1) they are usually time-consuming; (2) they are difficult to implement in a conventional transient stability program. for instance, as noted in the work of ayon et al. [19], the eigenvalues and eigen-matrix of the linearized system equations are usually difficult and also time-consuming to compute in the modal based methods. 3. rlse algorithm with memorizing factor assume that a dependent variable y can be evaluated by a linear function of k independent variables denoted by a row vector x = (x1, x2, …, xk) [31], then eq. (1) can be obtained: 1 1 2 2 ( x ) ... x φ k k y f x x xφ φ φ= = + + + = (1) where φ = ���,��,… ,� � shows an unknown column vector of parameters to be estimated, and t denotes transpose. assume further that s samples of measurements y(t) and x1(t), x2(t), …, xk(t) are available at time t = 1, 2, …, s. in the least-squares (ls) technique, the aim is to fit the s samples by the linear function of eq. (1). by filling the samples into eq. (1), a set of linear equations is obtained and can be written in matrix form as: y x φ= (2) 237 advances in technology innovation, vol. 6, no. 4, 2021, pp. 235-250 where y denotes an s × 1 vector and x shows an s k× matrix. note that if s > k, that is for the case in which the number of samples is greater than the number of unknown parameters (i.e., overdetermined system of equations), there exists no exact solution for the k unknown parameters φ. notice that sometimes a time-varying process is encountered and it is required that the estimated parameters are updated when new observations (samples) are available. to improve the computational efficiency, rlse technique can be employed in such cases. in rlse algorithm, the previously estimated parameters are utilized to obtain new estimations for the unknown parameters. to reduce the effect of past data points on the current estimate, a forgetting factor (� ) is used in the ls criterion. the forgetting factor � is a number less than 1 (e.g., � = 0.99) and causes the recent measured data to affect the estimation more, i.e., past samples are discounted. the rlse updating procedure with a forgetting factor � can be summarized as follows [31]: p ( ) x ( 1) g ( 1) [ x ( 1) p ( ) x ( 1)] t t f t t t t t tλ ++ = + + + (3) φ( 1) φ( ) g( 1)[ ( 1) x( 1)φ( )]t t t y t t t+ = + + + − + (4) p ( 1) [p ( ) g ( 1) x ( 1)p ( )] / f t t t t t λ+ = − + + (5) where g(t+1) denotes an k × 1 vector and p(t) is an k × k matrix. the matrix p is usually initialized to a very large identity matrix since no information is available in the beginning of rlse algorithm. the initial values of unknown parameters, that is φ(0), can be obtained by ls method or can be simply assumed as a vector of zeroes. in some cases, however, a time-varying process may be encountered, in which the older samples convey more information about the “true” parameter vector φ than the newly available data points. in such cases, more weights are to be given to the older samples so that the parameter values are weighted more by the older samples. to fix this problem, a factor called “memorizing factor” (��) is introduced in this study. the idea of a memorizing factor is very simple. it can be shown that if a number of more than 1 is set to the forgetting factor � used in eqs. (3)-(5), then rather than the older data points, less weight is given to the newly available samples. in other words, if the forgetting factor � is set to a value just more than 1, then compared with the older samples, the recent samples are discounted. this forgetting factor of more than 1 is called here the memorizing factor ��. notice that the greater the memorizing factor is, the less the influence of the recent data points on the estimation algorithm is. it is to be noted that the equations for rlse method with a memorizing factor �� are almost same as eqs. (3)-(5). the only difference is that the forgetting factor � is replaced by the memorizing factor �� in those equations. as will be seen later on in this study in section 7.1., the parameters of the tme system are estimated with much more accuracy if a memorizing factor (e.g., �� = 1.002) is used in the estimation algorithm. 4. power system model and problem statement consider that there is a power system consisting of n generators. in the classical model, the ith generator is represented with internal electromotive forces (emf) ��� = ��∠�� behind direct-axis transient reactance ��� � as shown in fig. 1. in fig. 1, it is assumed that a local machine i is connected to the rest of the power system. the aim is to derive a tme system for the entire power system, by looking into the system from the bus of this local machine. the loads are assumed to be represented as constant shunt admittances in the classical model of power system. the reduced (internal) admittance matrix yred (or yint) of the system is then obtained [32]. 238 advances in technology innovation, vol. 6, no. 4, 2021, pp. 235-250 i i e δ∠ dixj ′ip iq fig. 1 the connection of a local machine to a multimachine system letting ��� ≅ �� − �� and denoting the elements of matrix ���� as ��� = ��� + � ��, the active power and reactive power of the ith machine can be expressed as follows [32]: 2 1 ( cos sin ) n j i i i ii i j ij ij i j ij ijp e g e e g e e bδ δ = ≠ = + +∑ (6) 2 1 ( sin cos ) n j i i i ii i j ij ij i j ij ijq e b e e g e e bδ δ = ≠ = − + −∑ (7) the dynamics of the synchronous machine in classical model are governed by the following differential equations [32]: ; with 1, 2, ..., i i s d i n dt δ ω ω= − = (8) 1 [ ( )] ; w ith 1, 2, ...,i mi i i i s i d p p d i n dt m ω ω ω= − − − = (9) where !�� is the input mechanical power (assumed constant), "� is the rotor inertia constant, and #� is the natural damping coefficient of the ith machine. also, %& is the electrical synchronous speed. to simulate the behaviour of the actual system in the event of a large disturbance, it is necessary to numerically solve eqs. (8) and (9) in both faulted period and post fault period. having obtained the time variation of the rotor angles, the time variation of the ith machine’s active power (!�) and reactive power ('�) can be easily obtained by using eqs. (6) and (7), respectively. as can be seen in eqs. (6) and (7), both the angle of ith machine itself and the angle of remaining system machines are used in calculating !� and '�. it is also required to use the elements of the reduced matrix ����. it is important to mention here that the information of the whole system is implicitly included in the expressions for !� and '�. in other words, !� and '� are indeed global system variables, in which the effects of both the machine itself and the rest of the network are included. suppose that a large disturbance occurs near and/or at the bus of local machine i. furthermore, assume that it is only required to approximately evaluate the transient stability of this machine with respect to the rest of the system. a dynamic equivalent model of a multimachine system, viewed from the bus of this local machine, is derived for this purpose. here, machine i represents the internal system (of interest), and the rest of the system represents the external system (to be reduced). instead of an infinite bus, the aim is to replace the external system by an equivalent machine e in series with a lossless transmission line of reactance ��, as depicted in fig. 2. fig. 2 shows the defined tme system. the equivalent machine e shown in fig. 2 is represented as a classical machine with an internal emf ��� = ��∠��, a rotor inertia constant "�, and a natural damping coefficient #�. direct-axis transient reactance of the equivalent machine e is included in the reactance of transmission line �� for the sake of simplicity. 239 advances in technology innovation, vol. 6, no. 4, 2021, pp. 235-250 , im id ip iq dixj ′ ejx , em ed i i e δ∠ e e e δ∠ fig. 2 the tme system of a power system to determine the parameters of the tme system, the original multimachine system is initially excited to obtain the time variations of several system variables. a three-phase fault of short duration (e.g., 50 ms) is considered at the terminal of the local machine for this purpose. the obtained time solution for the system variables is then saved. notice that the two-stage approach is performed only in post fault period. for practical application of the proposed approach, the required time variations of the system variables can be obtained from real system events. the desired parameters of the tme system can be separated into two groups. the first group consists of three electrical parameters (i.e., ��, ��, and ��); whereas the second group includes "� and #�, representing the mechanical parameters of the equivalent machine e. it is to be noted that all parameters of the local machine i are assumed to be known. 5. the first stage of the proposed approach to estimate the parameters of the tme system, a three-phase short circuit is applied near the terminal of machine i in the original multimachine system. in other words, a time-domain simulation is carried out in the original test system. from the simulation, the active and reactive powers of machine i (i.e., !� and '�), are obtained at each time-step in post fault period. the time variation of local machine angle (i.e., ��), is also obtained from the time-domain simulation. the electrical parameters of the tme system are then estimated as follows. the active and reactive powers of machine i in fig. 2 can be calculated by using the following equations [1]: sin( )i ei i e edi e e p x x δ δ= − ′ + (10) [ cos( )]ii i e i e edi e q e e x x δ δ= − − ′ + (11) the electrical parameters ��, ��, and �� can be estimated by equating the active and reactive powers of machine i, obtained from the original system, with the corresponding powers calculated from eqs. (10) and (11). in other words, those electrical parameters are calculated in such a way that if they are substituted into eqs. (10) and (11), the active and reactive powers of machine i from the tme system become equal to their corresponding powers from the original system. there are only two known variables (i.e., !� and '�) here, but three unknown parameters (i.e., ��, ��, and ��) need to be estimated. therefore, many solutions can be found for the unknown parameters. to obtain suitable solutions, the rotor angle �� of machine e is assumed to be equal to the inertia weighted average (iwa) angle of all machines in the external system. the angle ��() is defined as: 1 1 n iwa j j j it m m δ δ = ≠ = ∑ (12) where "* = ∑ "� , �-�.� , and n is the number of generators. replacing the assumed value for �� from eq. (12) into eqs. (10) and (11), one can find �� and �� by the following equations (after a little algebra): 240 advances in technology innovation, vol. 6, no. 4, 2021, pp. 235-250 [ cos( ) sin( )] i i e i i e i i e e p e p qδ δ δ δ = − + − (13) sin( )i ee i e di i e e x x p δ δ ′= − − (14) it should be emphasized that, at each time-step of integration, it is required to re-calculate the parameters ��, ��, and �� by repeatedly solving eqs. (12)-(14). then, the time solutions of ��, ��, and �� are saved. the values obtained for these parameters at the end of the first stage are considered their final estimated values. 6. the second stage of the proposed approach to estimate the values of "� and #�, it is assumed that the time variation of �� obtained from the first stage of the approach fits the following second-order swing equation: 2 2 e e e e me e d d m d p p dtdt δ δ + = − (15) where !�� represents the input mechanical power and !� denotes the electrical output power of machine e. note that the defined tme system shown in fig. 2 is purely reactive; therefore, the electrical output power of machine i is completely absorbed by the equivalent machine e. the equivalent machine e acts here as a synchronous motor, and this leads to the following equations: e ip p= − (16) me mip p= − (17) substituting eqs. (16) and (17) into eq. (15), it yields: 2 2 e e e e i mi d d m d p p dtdt δ δ + = − (18) to estimate the values of "� and #�, the following approach is used. since time variation of �� is known, time variations of the first and second time-derivatives of �� (i.e., � / � = 0��/02 and � 3 � = 0 ���/02 �) can be calculated by using suitable numerical differentiation techniques. the first and second time derivatives of eδ for the lth step in post fault period are denoted by �/��4 and � 3 ��4 , respectively. substituting � / ��4 and � 3 ��4 into eq. (18) gives: miieeee plpldlm −=+ )()()( ... δδ (19) where !��4 shows the electrical output power of machine i, for the lth step in post fault period. notice that eq. (19) represents a discrete and algebraic equation, which is also linear with respect to "� and #�. also, eq. (19) needs to be satisfied at each time-step in post fault. therefore, at the end of the post fault, there will be a set of algebraic equations from which only two unknown parameters are to be estimated. a set of overdetermined linear algebraic equations needs to be solved here. in this study, the rlse algorithm with a memorizing factor is employed to estimate the values of "� and #� (referring to section 3). based on what mentioned so far, the main steps of the proposed method are summarized in the flowchart illustrated in fig. 3. in fig. 3, dt denotes the time-step used in the numerical integration, and t represents the total simulation time. 241 advances in technology innovation, vol. 6, no. 4, 2021, pp. 235-250 start perform time-domain simulation to solve swing equations f irst sta g e o f th e a p p ro a c h is t > t ? no yes no yes end s e c o n d sta g e o f th e a p p ro a c h define a fault scenario in the power system set t = t0 , where t0 is a few time after the fault clearing time calculate by using eq. (12), and then setiwaδ iwae δδ = e x c ite th e sy ste m b y a p p ly in g a fa u lt obtain and by solving eqs. (13) and (14)ee ex set t = t + dt save final estimated values of , , and ee eδ ex obtain and for the post-fault period using numerical differentiation techniques eδɺ eδɺɺ set t = t0 , where t0 is a few time after the fault clearing time obtain and by employing an rlse algorithmem ed set t = t + dt is t > t ? save final estimated values of and em ed read power system data and perform power flow analysis fig. 3 the proposed two-stage equivalencing approach 7. simulation results the proposed two-stage approach is applied to the new england 10-machine 39-bus (or the ieee 39-bus) test system. a single-line diagram of this system is shown in fig. 4. the bus and line data of this system can be found in the work of pai et al. [32]. the new england test system is called “the test system” hereafter, for the sake of simplicity. the time-domain simulations are carried out by using a computer program written by the author in matlab software. this program solves system differential equations by utilizing the fourth-order runge-kutta integration technique. however, power flow solutions for the test system are obtained by using the matpower package [33]. a time-step of 0.001 s is used in all calculations. the classical model is used and a uniform damping (i.e., 567806 = #�/"�; with i = 1, 2, …, n) of 1.0 is assumed for all machines of the test system. fig. 4 single-line diagram of the new england 39-bus test system 242 advances in technology innovation, vol. 6, no. 4, 2021, pp. 235-250 7.1. main results of simulation to excite the test system, a three-phase self-clearing type fault of 0.05 s duration is considered at bus 35. the simulation is carried out for 10 s in order to capture full dynamic behaviour of the test system. the location of the fault is very close to generator 2, as shown in fig. 4. for this fault, the internal system consists of generator 2 and the rest of the system represents the external system. generator 2 is the local machine here (i.e., i = 2), and the aim is to obtain a tme system from the bus of this local machine. it is to be noted that the proposed approach is performed only for post fault period. the time variation of the local machine angle (rotor angle of generator 2), obtained from the time-domain simulation in the original test system, is plotted in fig. 5 (blue curve). the time variation of the equivalent machine angle ��, obtained from the first-stage of the approach, is also illustrated in fig. 5 (red curve). the variation of the local machine active power is plotted in fig. 6. the variations of �� and �� are also illustrated in figs. 7 and 8, respectively. figs. 5, 7, and 8 clearly show that the values of ��, ��, and �� progressively settle at certain values as the time increases. the final estimates of several system variables including the rotor angle of both the local machine and the equivalent machine, along with some other system variables such as the inertia constant and damping coefficient of the local machine, are listed in table 1. the values of "� and #� are estimated in the second stage of the approach. before presenting the simulation results for "� and #�, it is worthwhile to investigate the effect of the inertia constant and damping coefficient of a machine on its dynamic behaviour (or on its rotor angle oscillations). for this purpose, consider the swing equations of a synchronous machine, e.g. eq. (18). as can be seen in eq. (18), only for the case in which the machine angle varies with time, its dynamic behaviour will be affected by its inertia constant and damping coefficient. this is due to the fact that when the machine angle remains constant, the first and second time derivatives of its angle will be equal to zero. for such a case, the machine inertia constant and damping coefficient are indeed omitted in the machine swing equations. consequently, the machine inertia constant and damping coefficient will have no effect on the rotor angle oscillations. instead, if the machine angle quickly varies with time, the machine dynamic behaviour is greatly affected by its inertia constant and damping coefficient. table 1 final estimated values for parameters of the tme system �� (rad) �� (p.u.) �� (p.u.) �� (rad) �� (p.u.) !�� (p.u.) "� (p.u.) #� (p.u.) "� (p.u.) #� (p.u.) 0.7597 0.9709 0.0051 1.1694 1.2255 6.3339 0.1607 0.1607 2.2152 0.9447 from the above discussion, it is concluded that in the beginning of post fault period, rotor angle of a synchronous machine conveys much information regarding its inertia constant and damping coefficient. therefore, instead of a forgetting factor, a memorizing factor can be used in the rlse algorithm of the second stage. as mentioned in section 3, a forgetting factor is a number less than 1; whereas a memorizing factor is a number more than 1. if a memorizing factor is used in the rlse algorithm, then compared with the recent measurements, more weights are given to the older data points. fig. 5 variations of the local and equivalent machines angles fig. 6 active power variations of the local machine after an 0.05 s fault on the original system 243 advances in technology innovation, vol. 6, no. 4, 2021, pp. 235-250 fig. 7 voltage magnitude variations of the equivalent machine fig. 8 variations of the transmission line reactance the time variations of "� and #� obtained by utilizing a fixed memorizing factor of 1.002 are illustrated (green curves) in figs. 9 and 10, respectively. it can be easily seen in figs. 9 and 10 that, except for the first few seconds, the values of "� and #� remain almost constant in post fault period. the time plots in figs. 9 and 10 are consistent with the time variations of the local machine and equivalent machine angles shown in fig. 5. as can be seen in fig. 5, those angles progressively settle at certain values in post fault period. however, for the first few seconds in post fault period, considerable time variations in both rotor angles can be observed. the final estimated values of "� and #� are also given in table 1. to show the advantage of memorizing factor, a fixed forgetting factor of 0.99 is also used in the second stage of the proposed approach, and then the time variations of "� and #� are obtained. the time variations obtained for "� and #� are plotted (red curves) in figs. 11 and 12, respectively. as can be seen, now there exist big fluctuations in the values of "� and #�. as a matter of fact, it is very difficult or even impossible to estimate the values of "� and #�, in this case. this proves the superiority of the memorizing factor to the forgetting factor. the time variations of "� and #� obtained by considering three different values for the memorizing factor are also obtained and plotted in figs. 9 and 10, respectively. the values chosen for the memorizing factor are 1.004, 1.01, and 1.05. in addition, the time variations of "� and #� obtained by employing ls technique are illustrated in figs. 9 and 10, respectively. for the case of ls technique, the memorizing factor is assumed to be one (i.e., ��= 1.00). as can be seen in figs. 9 and 10, depending on the value of the memorizing factor, both "� and #� vary differently over time. furthermore, for each value of the memorizing factor, both "� and #� progressively settle at some different values in the post-fault period. the final estimated values of "� and #�, for various values of the memorizing factor, are listed in table 2. fig. 9 inertia constant variations of the equivalent machine by using memorizing factor 244 advances in technology innovation, vol. 6, no. 4, 2021, pp. 235-250 fig. 10 damping coefficient variations of the equivalent machine by using memorizing factor fig. 11 inertia constant variations of the equivalent machine by using forgetting factor fig. 12 damping coefficient variations of the equivalent machine by using forgetting factor table 2 comparison of parameters of the tme system �� "� (p.u) #� (p.u.) cct (sec) 1.00 1.6071 1.0257 0.251 1.002 2.2152 0.9447 0.254 1.004 2.4208 0.9575 0.255 1.01 1.6417 2.0033 0.253 1.05 1.7766 2.2474 0.254 in an rlse based estimation method, a system of overdetermined equations needs to be solved (referring to section 3). for a system of overdetermined equations, there exists no solution that satisfies all the equations exactly. therefore, rlse algorithm is used to estimate the unknown parameters in an optimal sense. in ls algorithm, to obtain solutions for the unknown 245 advances in technology innovation, vol. 6, no. 4, 2021, pp. 235-250 parameters, each equation is weighted equally. however, in rlse algorithm, based on the value chosen for the memorizing factor (or the forgetting factor), each equation is weighted differently. therefore, for each value chosen for ��, the parameters "� and #� vary differently over time. the system of overdetermined equations obtained from eq. (19) is solved in the second stage of the proposed approach. as can be seen in eq. (19), when the first and second time derivatives of the local machine angle become zero, and at the same time, the right hand side of eq. (19) becomes zero, then no new equation is added to the set of aforementioned overdetermined system of equations. as shown in fig. 5, after a few seconds in the post-fault period, rotor angle of the equivalent machine remains unchanged. therefore, after a few seconds, the first and second time derivatives of the equivalent machine angle become zero. in addition, as shown in fig. 6, the active power of the local machine remains unchanged and reaches to its input mechanical power after a few seconds. consequently, the right hand side of eq. (19) becomes zero. in fact, the right hand side of eq. (19) represents the dependent variable y used in eq. (4). moreover, the first and second time derivatives of the equivalent machine angle are indeed the independent variables denoted by the vector x in eq. (4). referring to eq. (4), it can be concluded that when both y and x are equal to zero, the unknown vector of parameters φ is no longer updated by rlse algorithm. the vector φ consists of the mechanical parameters "� and #� of the equivalent machine e. as mentioned in section 3, the greater the memorizing factor is, the less the influence of the recent data points (or equations) on the estimation algorithm is. however, if the memorizing factor is a value very greater than 1, then almost all equations in a system of overdetermined equations are greatly discounted. this means that even those past equations that are very important for estimating the unknowns are neglected. for such a case, the parameters cannot be estimated by rlse algorithm. the curves shown in figs. 9 and 10 confirm this fact. as seen, for a memorizing factor of greater than 1.05, rlse algorithm may fail to correctly estimate the parameters "� and #�. the memorizing factor leading to the best (or optimal) estimation for the unknown parameters will depend on the system of overdetermined equations being considered. as stated before, the values of "� and #� cannot be estimated by using rlse algorithm with a forgetting factor. for the case of forgetting factor, past samples (equations) are discounted. the less the forgetting factor is, the less the influence of past equations on the estimation algorithm is. the variations of "� and #� obtained by using a fixed forgetting factor of 0.986 are also illustrated (blue curves) in figs. 10 and 11, respectively. as shown, compared with the case of a fixed forgetting factor of 0.99, there exist higher fluctuations in the values of "� and #� in this case. this is due to the fact that the less the forgetting factor, the less the effect of past equations in estimating the parameters "� and #�. it should be noted that only the values of "� and #� estimated in the second stage of the approach are dependent on the value of the memorizing factor. the parameters ��, ��, and �� are estimated in the first stage of the approach and their values do not depend on the value chosen for the memorizing factor. therefore, for all cases considered in this sub-section, the values of ��, ��, and �� are the same as those given in table 1. 7.2. results for the derived tme system in this sub-section, the derived tme system is used to replicate the behaviour of the original unreduced system by considering a three-phase fault on both systems. a three-phase self-clearing type fault of 0.1 s duration is considered at bus 2 in the original test system. the same fault is also considered at bus 2 in the derived tme system. as it is well known, relative rotor angles rather than absolute rotor angles need to be used in evaluating the stability of a power system. to obtain the relative rotor angles, it is required to take a reference rotor angle. the relative rotor angle in the tme system is easily obtained as: ��� − �� , where �� and �� are the angles of local machine 2 and equivalent machine e, respectively. for the original test system, the inertia weighted average angle (��()) of machines in the external system is used as an approximate reference rotor angle for local machine 2. the angle ��() is defined earlier in eq. (12). this relative rotor angle is denoted as: ��� − ��() . 246 advances in technology innovation, vol. 6, no. 4, 2021, pp. 235-250 the time variation of the relative angle ��� − ��() is shown in fig. 13 (solid blue curve). the time variation of the relative angle ��� − �� is also illustrated in fig. 13 (dotted red color). as can be seen in fig. 13, the relative rotor angle ��� − �� of the tme system closely approximates the relative rotor angle ��� − ��() of the original system. the time variations of the active and reactive powers of local machine 2 on both systems are shown in figs. 14 and 15, respectively. it can again be observed in figs. 14 and 15 that the active and reactive powers of the local machine in the tme system closely approximate the corresponding powers in the original system. to further investigate the effectiveness of the proposed method, a three-phase self-clearing type fault is considered at bus 2 on both the tme system and the original test system. the critical clearing time (cct) corresponding to this fault on both systems is then obtained. note that cct is used as an index for transient stability analysis of power systems [34]. the cct is found to be 0.254 s for the tme system, whereas the cct of 0.228 s is obtained for the original system. the values of cct on both systems are obtained via a trial-and-error approach and performing several time-domain simulations. it is obvious that the cct obtained from the tme system is near to its value in the original system, and this confirms the correctness of the proposed approach. fig. 13 relative rotor angles of local machine after an 0.1 s fault on the original and tme systems fig. 14 active power output of local machine after an 0.1 s fault on the original and tme systems fig. 15 reactive power output of local machine after an 0.1 s fault on the original and tme systems the ccts for other values of "� and #� mentioned in table 2 are also obtained in the tme system. the obtained ccts are listed in table 2. as seen in this table, the ccts obtained by using different values for the memorizing factor are almost equal to each other. this can be justified as follows. on the one hand, it is to be noted that the differences between the values of "� and #� are rather small, for various values of the memorizing factor. on the other hand, it can be noticed that the assumed fault is a severe three-phase short circuit at the terminal of the local machine. for this fault, the rest of the system, which is modelled here by an equivalent machine, has a little effect on the dynamic performance of the local machine. therefore, the values of "� and #� will only have a little impact on the cct value as well. 247 advances in technology innovation, vol. 6, no. 4, 2021, pp. 235-250 8. discussion the simulation results presented in section 7 shows that there exist some differences between the behaviour of the original system and that of the derived tme system. these differences may be associated with the model defined for the tme system. as shown in fig. 2, the equivalent machine e is assumed to be connected to the local machine i through a lossless transmission line. as a modification, a resistance can be included in series with the transmission line in the defined tme system. it is expected that by this modification, the tme system can approximate the behaviour of the original system with more accuracy. in this case, however, all the necessary equations mentioned in sections 5 and 6 are to be changed appropriately. in this study, the parameters of the equivalent system are obtained by using simulation data from a single fault scenario. in addition, just one system operating condition is considered. in order to improve the performance of the reduced (equivalent) model, multiple faults at different locations with different system operating conditions can be considered for estimating the parameters of the equivalent model. furthermore, the approach proposed in this study can be used in combination with a heuristic optimization-based method such as particle swarm optimization (pso) algorithm [35], to enhance the estimation accuracy of the tme system parameters. the time variations of the required system variables are obtained by considering a three-phase fault on the original test system in this study. it is important to point out that the required time variations of system variables can be acquired from an actual system disturbance event or an actual field measurement on a multimachine power system. the first three steps of the proposed approach, shown by the dotted red color in fig. 3, can be avoided if the measured (recorded) data are utilized in estimating the parameters of the tme system. furthermore, if some of the required system data are unavailable, they can be estimated by employing a suitable state estimation technique. the use of local measurements available from pmus is also an alternative way for estimating the required data [24]. nevertheless, these issues have not been addressed in the present study. the tme system is derived for the classical model of power system; however, it can be used for more accurate transient stability analysis of the local machine as well. to this end, the detailed generator model with excitation system and other control systems can be applied to the local machine in the derived tme system. the tme system can then be employed for more accurate design of power system stabilizer (pss) [1] or other controllers for the local machine. as described in section 7.1., the idea of memorizing factor is inspired by analyzing the behaviour of synchronous generators, following a three-phase fault on a multimachine power system. however, the memorizing factor based rlse algorithm can, in general, be used for solving any set of over-determined algebraic equations in which the older equations are more important than the recent ones. as a matter of fact, the dynamic equivalencing approach proposed in this study can be considered one of the applications of the memorizing factor concept. as stated in section 7.1., for a given system of overdetermined equations, there exists an optimal value for the memorizing factor that leads to the best estimation of the unknown parameters. future research can be focused on finding such an optimal value for the memorizing factor. 9. conclusions in this study, a systematic two-stage identification-based approach was presented to tackle the challenging problem of multimachine power systems dynamic equivalencing. in particular, a tme system was derived for a multimachine power system, by looking into the system from the bus of a local system machine. first, the tme system was defined, and then the desired parameters of the predefined equivalent system were estimated by using a two-stage approach. the electrical parameters of the equivalent system were obtained by utilizing a simple technique in the first stage of the approach. the mechanical parameters of the equivalent system were obtained in the second stage of the approach by employing an innovative rlse technique with a “memorizing factor”. the effectiveness of the proposed methodology was validated by using the new england 39-bus test system. 248 advances in technology innovation, vol. 6, no. 4, 2021, pp. 235-250 many simulation results were presented to show the correctness of the proposed approach. the advantages of using a memorizing factor rather than a forgetting factor were justified and also verified by many simulation results. instead of the traditional smib system, the derived tme system can be employed for transient stability analysis as well as other dynamical studies in multimachine power systems. it can also be used for more accurate design of pss or other controllers for the local machine. the memorizing factor based rlse algorithm introduced in this study is completely new, and can be applied in any system identification-based problem in which the older samples are more important than the newly available data points. the concept of varying memorizing factor can be regarded as a future work. conflicts of interest the author declares no conflict of interest. references [1] j. machowski, z. lubosny, j. bialek, and j. r. bumby, power system dynamics: stability and control, 3rd ed. hoboken: john wiley & sons, 2020. [2] a. karami and k. m. galougahi, “improvement in power system transient stability by using statcom and neural networks,” electrical engineering, vol. 101, no. 1, pp. 19-33, april 2019. [3] e. c. machado and j. e. pessanha, “foundations on the hamiltonian energy balance method for power system transient stability analysis: theory and simulation,” journal of control, automation, and electrical systems, vol. 31, no. 1, pp. 226-232, february 2020. [4] y. f. gao, j. q. wang, t. n. xiao, and d. z. jiang, “fast emergency control strategy calculation based on dynamic equivalence and integral sensitivity,” frontiers of information technology and electronic engineering, vol. 20, no. 8, pp. 1119-1132, august 2019. [5] x. y. li and o. p. malik, “estimation of equivalent models for emergency state control of interconnected power-systems based on multistage recursive least-squares identification,” iee proceedings c (generation, transmission, and distribution), vol. 140, no. 4, pp. 319-325, july 1993. [6] j. b. ward, “equivalent circuits for power-flow studies,” transactions of the american institute of electrical engineers, vol. 68, no. 1, pp. 373-382, july 1949. [7] p. dimo, nodal analysis of power systems, lancashire: abacus press, 1975. [8] j. srivani and k. s. swarup, “power system static security assessment and evaluation using external system equivalents,” international journal of electrical power and energy systems, vol. 30, no. 2, pp. 83-92, february 2008. [9] y. fu and t. s. chung, “a hybrid artificial neural network (ann) and ward equivalent approach for on-line power system voltage security assessment,” electric power system research, vol. 53, no. 3, pp. 165-171, march 2000. [10] b. g. lee, j. lee, and s. kim, “development of a static equivalent model for korean power systems using power transfer distribution factor-based k-means++ algorithm,” energies, vol. 13, no. 24, 6663, december 2020. [11] u. d. annakkage, n. k. c. nair, y. liang, a. m. gole, v. dinavahi, b. gustavsen, et al., “dynamic system equivalents: a survey of available techniques,” ieee transactions on power delivery, vol. 27, no. 1, pp. 411-420, january 2012. [12] j. h. chow, power system coherency and model reduction, springer science and business media, 2013. [13] m. r. a. paternina, j. m. ramirez-arredondo, j. d. lara-jiménez, and a. zamora-mendez, “dynamic equivalents by modal decomposition of tie-line active power flows,” ieee transactions on power systems, vol. 32, no. 2, pp. 1304-1314, march 2017. [14] c. m. rergis, r. j. betancourt, and a. r. messina, “order reduction of power systems by modal truncated balanced realization,” electric power components and systems, vol. 45, no. 2, pp. 147-158, 2017. [15] d. chaniotis and m. a. pai, “model reduction in power systems using krylov subspace methods,” ieee transactions on power systems, vol. 20, no. 2, pp. 888-894, may 2005. [16] z. zhu, g. geng, and q. jiang, “power system dynamic model reduction based on extended krylov subspace method,” ieee transactions on power systems, vol. 31, no. 6, pp. 4483-4494, november 2016. [17] t. k. ritschel, f. weiß, m. baumann, and s. grundel, “nonlinear model reduction of dynamical power grid models using quadratization and balanced truncation,” automatisierungstechnik, vol. 68, no. 12, pp. 1022-1034, november 2020. 249 advances in technology innovation, vol. 6, no. 4, 2021, pp. 235-250 [18] r. podmore, “identification of coherent generators for dynamic equivalents,” ieee transactions on power apparatus and systems, vol. 97, no. 4, pp. 1344-1354, july 1978. [19] j. j. ayon, e. barocio, i. r. cabrera, and r. betancourt, “identification of coherent areas using a power spectral density algorithm,” electrical engineering, vol. 100, no. 2, pp. 1009-1019, june 2018. [20] a. m. miah, “study of a coherency-based simple dynamic equivalent for transient stability assessment,” iet generation, transmission, and distribution, vol. 5, no. 4, pp. 405-416, april 2011. [21] m. matar, n. fernandopulle, and a. maria, “dynamic model reduction of large power systems based on coherency aggregation techniques and black-box optimization,” international conference on power systems transients, june 2013, pp. 1-7. [22] s. chittora and s. n. singh, “coherency based dynamic equivalencing of electric power system,” 18th national power systems conference, december 2014, pp. 1-6. [23] a. m. stankovic, a. t. saric, and m. milosevic, “identification of nonparametric dynamic power system equivalents with artificial neural networks,” ieee transactions on power systems, vol. 18, no. 4, pp. 1478-1486, november 2003. [24] a. chakrabortty and a. salazar, “building a dynamic electro-mechanical model for the pacific ac intertie using distributed synchrophasor measurements,” european transactions on electrical power, vol. 21, no. 4, pp. 1657-1672, may 2011. [25] j. m. ramirez, b. v. hernández, and r. e. correa, “dynamic equivalence by an optimal strategy,” electric power system research, vol. 84, no. 1, pp. 58-64, march 2012. [26] j. l. rueda, j. cepeda, i. erlich, d. echeverría, and g. argüello, “heuristic optimization based approach for identification of power system dynamic equivalents,” international journal of electrical power and energy systems, vol. 64, pp. 185-193, january 2015. [27] x. zhang, y. xue, s. you, y. liu, z. yuan, j. chai, et al., “measurement-based power system dynamic model reductions,” north american power symposium, september 2017, pp. 1-6. [28] n. tong, z. jiang, s. you, l. zhu, x. deng, y. xue, et al., “dynamic equivalence of large-scale power systems based on boundary measurements,” american control conference, july 2020, pp. 3164-3169. [29] z. jiang, n. tong, y. liu, y. xue, and a. g. tarditi, “enhanced dynamic equivalent identification method of large-scale power systems using multiple events,” electric power systems research, vol. 189, 106569, december 2020. [30] a. b. almeida, r. reginatto, and r. j. g. c. d. silva, “a software tool for the determination of dynamic equivalents of power systems,” irep symposium bulk power system dynamics and control, august 2010, pp. 1-10. [31] k. j. keesman, system identification: an introduction, london: springer, 2011. [32] m. a. pai, energy function analysis for power system stability, boston: kluwer academic publishers, 1989. [33] r. d. zimmerman, c. e. murillo-sánchez, and r. j. thomas, “matpower: steady-state operations, planning, and analysis tools for power systems research and education,” ieee transactions on power systems, vol. 26, no. 1, pp. 12-19, february 2011. [34] a. karami and s. z. esmaili, “transient stability assessment of power systems described with detailed models using neural networks,” international journal of electrical power and energy systems, vol. 45, no. 1, pp. 279-292, february 2013. [35] a. slowik, swarm intelligence algorithms: a tutorial, boca raton: crc press, 2021. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). 250  advances in technology innovation, vol. 7, no. 4, 2022, pp. 242-257 circularly polarized antennas using characteristic mode analysis: a review vamshi kollipara, samineni peddakrishna * school of electronics engineering, vit-ap university, amaravati, india received 19 october 2021; received in revised form 20 december 2021; accepted 21 december 2021 doi: https://doi.org/10.46604/aiti.2022.8739 abstract characteristic mode analysis (cma) can be used in antenna designs to solve radiation problems. this review focuses on the existing development methodologies for circularly polarized (cp) antennas and their axial ratio bandwidth (arbw) improvement using cma. to discuss the physical insights related to cp radiation, this study systematically examines different antenna design structures used in previous research. it investigates the impact of modal parameters such as the eigenvalue, modal significance (ms), characteristic angle (ca), surface current, far-field radiation behavior on cp radiation, and arbw for various antenna designs. in addition, it discusses the comparative analysis of various antenna design approaches in terms of antenna performance parameters such as the operating frequency band, arbw, and gain. the results show that cma provides more valuable information for the selection of feed position in antenna designs than the conventional full-wave simulation approach. keywords: characteristic mode analysis (cma), eigenvalues, modal significance, characteristic angle 1. introduction today’s advanced wireless communication technology development is being driven by different integrated challenges. the antenna design methodology is one of the interesting challenges with desired radiation behavior [1-3]. an independent antenna design approach contributes to the optimization of a communication system [4-5]. sometimes, in such a communication system, unnecessary power losses may occur if the polarization of the receiving and transmitting antennas is not matched. by using circularly polarized (cp) antennas, power losses due to polarization mismatch can be avoided. cp antennas are essential in various applications due to their ability to combat multi-path interference and mitigate linear polarization (lp) problems such as faraday rotation [6]. cp radiation can be achieved by combining two orthogonal lp radiations with equal magnitude and quadrature-phase excitations. additionally, various techniques have been used, such as adjusting patch shapes with single or multiple feed networks through parameter sweeping [7] and using automated optimization methods [8]. most of the antenna designs typically follow a cut-and-try approach, based on the engineering experience by full-wave simulation without understanding the natural resonance characteristics of structures. this procedure becomes somewhat trivial and lacks physical insights. because of this, the final simulation characteristics such as the radiation efficiency, input impedance, and radiation patterns of antennas are dependent on the resonance properties themselves from the exciting external feed. hence, the antenna resultant current distribution does not reflect the natural resonance characteristics with the improper feed. thus, to understand and analyze the resultant current distribution that does reflect the natural resonant behavior, the theory of characteristic mode analysis (cma) is one of the fundamental approaches of recent days in the antenna domain. cma analyzes the resonant behavior of an antenna structure using a source-free method and it decomposes the full-wave current into an individual mode [9-10]. this will help understand the physical insights and optimal feed arrangements, and excite the desired modes by suppressing undesired modes [11]. the optimization can thus be done on the designs to the desired radiation performance. * corresponding author. e-mail address: krishna.samineni@gmail.com advances in technology innovation, vol. 7, no. 4, 2022, pp. 242-257 cma was originally addressed by garbacz [12] and formulated with turpin [13]. they demonstrated a specific set of characteristic modes (cms) for an arbitrary perfect electric conductor (pec) obstacle that is independent of any specific source. there is only the assumption that cms have been given instead of an explicit definition. therefore, harington and mautz [14] refined the cm theory for pec objects by relating the surface current to the tangential electric field and defined that an infinite number of independent current modes can naturally exist in an arbitrary structure. later, it has been extended to different structures [15-16]. however, the appropriateness of their formulations has not been exhibited properly in solving practical problems. in recent years, numerous investigations into these earlier formulations have been addressed [17-19]. the most important applications of these findings include mobile handset antenna designs [20], radio-frequency identification (rfid) tag antennas [21], universal serial bus (usb) dongle antennas [22], aircrafts [23], unmanned aerial vehicles (uav) [24-25], ships [26], and land vehicles [27-29]. recently, the study has been extended to a dual-port fractal ultrawideband (uwb) multiple-input multiple-output (mimo) antenna for portable handheld wireless devices [30] and a 4-port mimo antenna for 5.8 ghz wlan applications [31]. very recently, some 5g antennas have also been proposed using mimo technology [32-36]. the cm theory has become a standard approach in these developments. in this article, the authors explore the cm-based methodological advancement for a variety of critical cp antenna designs. this article intends to assemble and organize the advanced research achievement made in the area of cm studies, especially for the cp achievement with wide axial ratio bandwidth (arbw). it gives preliminary knowledge about the cm parameters and helps readers get a deeper understanding of various practical antenna developments using cm-based methodologies. the organization of the article is in the following manner. section 2 explains the basic cm parameters that are required for understanding the natural resonance behavior of an antenna. section 3 describes the comprehensive methodologies in the literature for the cp achievement and arbw enhancement using cma. section 4 concludes the discussion followed by a list of references. 2. cm parameters this section discusses the formulation of cms and their parameters that are useful in antenna designs. the initial phase of any antenna design using cma involves the modal analysis. this will help understand the modal behavior of geometry. mathematically, it can be thought of as eigenvalue analysis [37]. typically, cma begins by extracting the resonance information from cm parameters. these cm parameters are obtained by solving an eigenvalue equation that is derived from the method of moment impedance matrix as shown in eq. (1) [13]. [ ] [ ] [ ]z r j x  (1) from this, an eigenvalue equation is formulated as eq. (2): [ ][ ] [ ][ ]n n nx i r i  (2) here, x and r are the imaginary and real parts of the generalized impedance matrix [z], respectively. vector in is the eigen or characteristic current, where n represents the index of each mode. λn is the real eigenvalue. this represents the possible modes that are naturally supported by the structure. by observing the real eigenvalue λn, the physical insight corresponding to the natural resonance information is obtained. 2.1. eigenvalue (λn) an eigenvalue (λn) provides useful information about the natural resonant frequency of an intended antenna design, especially as a function of frequency. the eigenvalue is useful to identify the mode information because its magnitude is proportional to the total stored field energy within a radiating antenna. based on the stored electric and magnetic energy, the associated mode information can be classified as resonant mode, inductive mode, and capacitive mode. 243 advances in technology innovation, vol. 7, no. 4, 2022, pp. 242-257 in particular, if it shows λn = 0, i.e., the stored magnetic energy is the same as the stored electric energy as represented by eq. (3), the associated modes are said to be resonant. * * . . n n m n v v h h dv e e dv    (3) if it shows λn > 0, i.e., the stored magnetic energy is more than the stored electric energy as represented by eq. (4), the associated modes are said to be inductive. * * . . n n m n v v h h dv e e dv    (4) if it shows λn < 0, i.e., the stored magnetic energy is less than the stored electric energy as represented by eq. (5), the associated modes are said to be capacitive [6]. * * . . n n m n v v h h dv e e dv    (5) to understand the behavior of eigenvalues for various modes supported by the structure, a typical rectangular patch with 25 × 160 mm is considered to represent the eigenvalues for various modes [38]. as can be seen from fig. 1, the eigenvalues of mode 1 and mode 4 are approaching zero, and these modes are identified as resonance frequencies across the frequency band. in addition, it shows the sign of the eigenvalue of mode 3 indicating the ability to store electrical energy (λn < 0) and mode 2 indicating the ability to store magnetic energy (λn > 0). however, if higher frequencies are considered with higher-order modes, more than one mode turns into the resonance conditions. hence, for good polarization purity, it is not easy to excite a single mode. also, extracting the modal resonance properties for higher frequencies becomes quite difficult because they are tightly grouped and cannot be differentiated from each other. therefore, to visualize those resonating characteristics, additional parameters such as modal significance (ms) and characteristic angle (ca) are defined along with eigenvalues. fig. 1 variation of eigenvalues for different modes [38] 2.2. modal significance (ms) ms is another way of extracting the resonance characteristics by using eigenvalues, and it measures the potential contribution of each mode. ms is the intrinsic property of each mode and is defined in eq. (6): 1 | 1 | n ms j    (6) 244 advances in technology innovation, vol. 7, no. 4, 2022, pp. 242-257 from eq. (6), it can be observed that the eigenvalue λn tends to be zero, and the ms becomes one and the mode starts resonating. the ms range transforms to [[0, 1]] from the much higher eigenvalue range [−∞, +∞]. therefore, in many cases, to investigate the resonant behavior, it is more convenient to use ms than eigenvalues, especially over wide frequency band applications. additionally, ms is also used to determine how many significant modes will be considered in the antenna design from the contribution of eigenvalues. to distinguish the significant modes and non-significant modes, the half-power ms is defined. from this, if ms ≥ 0.707, the associated modes are significant. if ms < 0.707, the associated modes are non-significant. further, the half-power ms also reinforces the definition of each cm bandwidth, especially in the case when the feed structure is not available in the initial stage of the design. the half-power bandwidth of each cm is defined in eq. (7). h l res f f bw f   (7) where fres, fh, and fl are the resonant frequency, upper half frequency, and lower half frequency, respectively. they are determined from eq. (6) in the following manner. if λn = 0 and ms = 1, then the associated mode is a resonant mode (fres). if λn = ±1 and ms = 0.707, then the associated mode is the lower and upper half power frequency band mode (fl and fh), as shown in eq. (8): 1 1 ( ) ( ) 0.707 1 2 l h n ms f ms f j       (8) as a further extension, fig. 2 shows the associated ms behavior of each mode supported by the same structure across the frequency for four different modes [38]. as observed from the figure, ms approaches unity for mode 1 and mode 4 and is considered to contribute to the radiation. mode 2 and mode 3 are considered non-significant modes. fig. 2 variation of ms for different modes [38] 2.3. characteristic angle (ca) ca is another way to show the mode behavior near resonance. it is the phase lag between the real characteristic current on the surface and the tangential electric field. this can be extracted from the eigenvalue using eq. (9). 1 180 tan ( ) n n     (9) from eq. (9), if λn = 0, the phase lag between the electric field and real current on the surface is 180° out of phase, and the mode is said to be the most effective radiating mode of the radiating element. on the opposite, if λn = ±∞, the attained phase lag is 90° or 270°, and the mode is said to be non-radiating or cavity resonance mode [9]. in this case, the modal current generates a null field in the exterior region. additionally, if the phase angle varies between [90°, 180°] and [180°, 270°], the modes are said to be inductive modes and capacitive modes, respectively. 245 advances in technology innovation, vol. 7, no. 4, 2022, pp. 242-257 to further understand the associated modal behavior of each mode supported by a structure, the variation of ca across the frequency for four different modes is shown in fig. 3 [38]. from the figure, it can be easily understood that the variation of ca for mode 1 and mode 2 attains 180° phase lag at a certain frequency and these are considered resonant modes. mode 3 and mode 4 are varying between [90°, 180°] and [180°, 270°] over a given frequency range, so they are said to be inductive modes and capacitive modes, respectively. fig. 3 variation of ca for different modes [38] 2.4. eigen current or field in addition to the characteristic parameters, the real eigen currents are important to illustrate various possible currents or fields that are naturally supported by the structure. thorough investigations into modal currents and modal fields can yield useful information on how to feed the structure for certain desirable radiation patterns. to excite the desired mode, a source needs to be positioned at a location with high characteristic currents. to understand how to place the source to excite all of the desired cms from the characteristic currents, the modal currents of the first three normalized modes are illustrated in fig. 4 [39]. as from the current distribution, the maxima and minima of the individual cms j1, j2, and j3 are primarily observed at the edges of the major and minor axis and can be characterized by a sinusoidal behavior. as regards this current behavior, two feeding methods have been proposed: the capacitive coupling element (cce) and inductive coupling element (ice) methods. for efficient mode excitation, cce and ice have to be placed at the location with the minimum and maximum characteristic current, respectively. (a) surface current and field (b) sinusoidal behavior of current fig. 4 modal current and field pattern variation [39] however, practical radiating antennas typically consider the effect of source excitation. from the source excitation point of view, modal decompositions can tell how well each of the designs is excited in the desired modes. this can be done by calculating the quantities called modal excitation coefficient (mec), modal weighting coefficient (mwc), and modal input power from eqs. (10)-(14). the total current on the structure is assumed to be linear superposition of the orthogonal set of mode currents [14, 37]. this can be extracted from eq. (10). . n ntot nj jc    (10) where cn is the mwc and jn is the characteristic current of the mode n. 246 advances in technology innovation, vol. 7, no. 4, 2022, pp. 242-257 mwc is an important guiding parameter for an antenna design to determine the appropriate feed point and feed structure telling which mode carries the maximum current. 1 | | . | . | | 1 | n n i n c j e ds j     (11) | | . | | n n msc v (12) where vn is the mec which describes an easy way to excite the mode from its excitation. | | . n n i v j e ds  (13) from eq. (12), it can be understood that a large ms and mec are necessary to excite the desired operating modes. another important parameter is the modal input power, which is described as the power in each mode and used to find which mode is exited for radiation behavior problems. the modal input power pin is determined by the mwc cn, as shown in eq. (14). 21 | | 2 in n n p c  (14) the above quantities explain how a design can succeed in the excitation of the desired modes. finally, this section summarizes the standard values for the measure of a resonant mode from the above parameters, i.e., eigenvalue (λn) = 0, ms = 1, and ca = 180°. they are essentially expressing the same thing differently, and which parameter is used depends on the personal design requirement. if the design requirement is only on the primary resonant mode or to identify a few lower-order resonant modes, the parameter extraction of the eigenvalue (λn) is sufficient. to extract the behavior of higher-order modes or over a wide frequency band of operation, ms is an important parameter. when multiple modes need to be excited for certain radiation performance (i.e., circular polarization), the ca and current or field distribution are of great importance in the antenna design. 3. cp antennas using cma cma and its applications in antenna designs for enhancing various radiation parameters are described by various research groups in the literature. however, this section has limited the review to a parameter called circular polarization. the underlying mechanism of this parameter is readily available to be read through many electronic databases individually. this review process is highlighted, with all those approaches together, by classifying various groups based on the feed networks and the type of geometry. this will help further understand the insight into the design of the cp antenna and its arbw enhancement using cma. antenna polarization is usually defined as the orientation of an electric field as a function of time, at a fixed position in space. the polarization type of an antenna can be identified from the axial ratio (ar) [40-41]. if ar is zero or infinity, it is lp. for unity ar, it is circular polarization. if ar is between zero to one, it is elliptical polarization. however, for practical applications, the most acceptable ar is 1.414 = 3 db for circular polarization. to generate circular polarization in antennas, a well-known fact is to excite two orthogonal modes with equal amplitude and quadrature (90°) phase difference. various techniques have been described in the literature using singleor dual-feeding techniques [42-43]. in a single-feeding method, a small perturbation is required for the orthogonal mode and a quadrature phase shift at the feed point. the perturbation segment is in the form of a slit, a slot, a truncated segment, or an added stub [44-46]. however, the single-feed antennas are structurally simple but suffer from narrow arbw. to overcome this, a 247 advances in technology innovation, vol. 7, no. 4, 2022, pp. 242-257 dual-feed network is one of the solutions. however, dual-feed cp antennas require an additional power divider to split the input signal with equal amplitude and an additional quadrature wavelength transmission line to generate 90° phase shifts. this enhances the design complexity. to compensate for the two issues, another type of cp antennas called slot and stub antennas has been proposed [47-49]. however, most of these antennas are usually accomplished by tuning the patch shapes and feed positions through full-wave simulation. those designs have not been verified for the similarity between the natural resonance characteristics and the physical insight of the feed location. to address the above two characteristics and further optimization of the structure, cma has been identified by many research groups. using cma and its parameter extraction, one can easily identify the circular polarization of the design from the following three conditions: (1) the first condition to satisfy circular polarization is to identify adjacent orthogonal modes and this can be easily verified from the modal current distribution analysis for various modes. (2) the next condition is that both orthogonal fields are with uniform magnitude and it can be verified from the ms between two orthogonal modes. if the ms is the same for two orthogonal modes at a certain frequency means, it satisfies the required condition. (3) finally, the last condition for circular polarization is a quadrature-phase difference between the modes and it can be verified from ca. if the ca difference between the orthogonal modes is 90° means, it satisfies the required condition. with the use of the above three conditions, various cp antennas have been designed in the literature. these antennas mainly differ in the way by which the lp modes are excited properly with the help of cma. to review those proposed structures, they are divided into two parts. the first part of the cp antennas is considered without metasurface (mts) and the second part is with mts. 3.1. cp antennas without mts to demonstrate how cma is useful to achieve circular polarization, a simple rectangular asymmetrical u-shaped slot antenna as shown in fig. 5 is considered [50]. cma has been carried out without considering the feed structure. fig. 6 shows the first two modes of ms and ca across the frequency band from 2.0-2.7 ghz. it is found that the two modes have the same large ms and a 90° phase difference at 2.3 ghz, which satisfies the above three conditions. as a result, for cp operations, these two modes operate together at the center frequency. next, to find the optimal feed position, the characteristic currents as shown in figs. 7(a) and (b) have been characterized in the horizontal mode and vertical mode with respect to the modes j1 and j2. then, to excite two modes properly, the vertical mode is subtracted from the horizontal mode, i.e., performed j1 − j2 between the two modes. as a result, the minimum current has been obtained at the inner edge of the u-slot’s long arm, as shown in fig. 7(c). this location specifies that the two orthogonal modes represent a similar current amplitude. the two orthogonal modes that have comparable current density location points (indicated with the black dot in fig. 5) will generate far-field cp radiation upon excitation. to generate the circular polarization, an equal current magnitude and 90° ca phase difference has been generated from an equal crossed dipole to an unequal crossed dipole [51]. it has been observed that a square contour parallel dipole has shown a 90° phase difference for degenerating modes such as mode 2 and mode 3 by analyzing individual orthogonal current modes [52]. the designs discussed above have a low profile and can be fabricated easily [53-59]. however, some of those structures exhibit a narrow arbw because the orthogonality and phase difference is achieved at a single frequency. to enhance the arbw with a single-feed mechanism from cma, tran et al. [53.] coupled a higher band c-shaped monopole cp antenna with a lower band square patch with a c-shaped slot aperture, as shown in fig. 8. to understand the arbw enhancement, the ms, ca, current distribution, and far-field radiation of a c-shaped monopole and c-shaped slotted patch are presented in fig. 9 and fig. 10, respectively. as observed from fig. 9, the first two modes are exactly the same, with a 90° phase difference at around 5.5 ghz radiating in a broadside direction. from fig. 10, a similar phenomenon is observed for the first two 248 advances in technology innovation, vol. 7, no. 4, 2022, pp. 242-257 fundamental modes at around 2.1 ghz with broadside radiation. in addition, it is noticed that the same ms is quite close to 90° phase difference between the modes j2 and j4 as well as the modes j2 and j5 at 3.2 and 3.5 ghz, respectively. however, the current directions of the modes j2 and j4 are observed on the opposite side instead of having quadrature-phase differences. hence, it has been considered that the c-shaped slotted patch is proposed to radiate cp waves at 2.1 and 3.5 ghz. fig. 5 asymmetric u-slot antenna for circular polarization [50] (a) ms (b) ca fig. 6 cm parameters of the asymmetric u-slot antenna [50] (a) mode j1 (b) mode j2 (c) j1 − j2 fig. 7 current distribution of the u-slot antenna at 2.3 ghz [50] (a) top layer (b) bottom layer fig. 8 geometry of the cp c-shaped monopole antenna [53] (a) ms fig. 9 modal characteristics of the c-shaped monopole patch [53] 249 advances in technology innovation, vol. 7, no. 4, 2022, pp. 242-257 (b) ca (c) current distribution and far-field radiation fig. 9 modal characteristics of the c-shaped monopole patch [53] (continued) (a) ms (b) ca (c) far-field radiation fig. 10 modal characteristics of the c-shaped slotted patch [53] (a) s11 parameters (b) ar fig. 11 parameters of the c-shaped monopole and c-shaped slotted patch antenna [53] once the desired modes have been selected for wide cp radiation from the above two shapes, the feed point from the c-shaped monopole patch is subsequently identified from the current distribution. it is identified as the feed point at the corner by subtracting the current mode j2 from the mode j1 of the c-shaped patch. then, the patch is excited with a monopole arrangement and the arbw is obtained ranging from 4.6 to 6.1 ghz. next, to enhance the arbw, the slotted c-shaped patch aperture is excited through a coupling mechanism with the directly fed c-shaped patch monopole. the final geometry of the cp antenna top and bottom layer design is shown in fig. 10. here, the cp radiation at 2.1 ghz and 3.5 ghz is predicted based on the desired 250 advances in technology innovation, vol. 7, no. 4, 2022, pp. 242-257 modes (j1, j2, and j5) of the current distribution of the c-shaped slotted patch. therefore, the position of the c-shaped patch monopole plays an extensive role in the performance of the c-slotted patch. to obtain the optimum coupling position, the design is fine-tuned using numerical simulation. the simulated s11 and arbw results of an individual and optimized patch antenna are shown in fig. 11. the results show that a very wide arbw of 96.1% can be achieved in this configuration. using similar kinds of techniques, a cpw-fed rectangular slot antenna has been designed for wider arbw [54-57]. the cp conditions are attained using an asymmetric slot [54] and asymmetric stubs [55] on the ground plane by altering the current distribution. additionally, the arbw has been improved by employing two symmetric inverted l-stubs with vertical strips parallel to an antenna and a pair of asymmetric inverted l-strips with spiral slots, respectively. in another design, a circular ring cpw-fed antenna has been proposed with two orthogonal clock angle microstrip lines on opposite sides of the substrate [56]. to attain a quadrature-phase difference ca, the angle between the microstrip lines is optimized to achieve circular polarization. recently, a cp loop antenna has been designed with a broad arbw and impedance bandwidth with a single-feeding technique [57]. here, a 90° phase difference is attained by loading the lumped inductors. a pair of degenerated mode resonance points is split by properly positioning the feeds and loading the inductors with the aid of cma. using this method, an 8.3% arbw and a 47% impedance bandwidth are achieved at 2.4 ghz wlan band. in summary, this subsection has reviewed the cma of optimal antenna designs without mts for cp radiation and improvement in ar. the optimal designs discussed above [55-57] demonstrate that cma provides important information for cp performance and arbw enhancement from cm parameters and current or field distribution. on the other hand, the designs described above illustrate how to optimize the feed position to provide cp radiation and improvement in arbw from the modal current distribution. 3.2. cp antennas with mts this subsection investigates how cma can be used to tailor the mts for the purpose of designing cp antennas. various mtss are exploited in the cp radiation and arbw improvement [58]. to demonstrate the usefulness of cma with respect to mts, the geometry of the mts antenna is shown in fig. 12. it comprises two dielectric layers and three metallic layers. the three metallic layers (i.e., the mts, cross-shaped slot, and feed structure) are placed on the top, middle, and bottom of two substrates. to analyze the cp behavior, cma is first performed on the top mts, and ms is demonstrated for the first four modes across the frequency band, as shown in fig. 13(a). then, the current distribution and far-field patterns of these modes at the respective resonance frequencies, such as 6 ghz (for the modes j1 and j2) and 6.5 ghz (for the modes j3 and j4), are observed as shown in figs. 13(b) and (c). as observed from the modal currents, the modes j1 and j2 have an identical current distribution except for the 90° phase difference. however, they are a pair of degenerate modes and cannot provide 90° ca. the modes j3 and j4 are self-symmetrical modes with out-of-phase current distribution and appear null in the broadside direction. hence, only mode 1 and mode 2 have been considered for cp radiation with simultaneous excitation with a 90° phase difference from the feed structure. (a) top view of mts (b) cross slot (c) back view of the microstrip line (d) side view of the antenna fig. 12 configuration of the mts-based antenna [58] 251 advances in technology innovation, vol. 7, no. 4, 2022, pp. 242-257 (a) ms (b) current distribution (c) far-field patterns fig. 13 mts cm parameters [58] to excite only the first two modes with a 90° phase difference and suppress higher-order modes, an adaptive feed network has been employed with a combination of a cross-slot and a microstrip meander-line (figs. 13(b) and (c)). as a result, cp radiation is formed. to investigate the cp performance, the s parameter and ar performance are compared in fig. 14. further, various mts antennas have been proposed in the literature based on the orthogonality principle of surface current distribution and identical ms between adjacent or lower-order modes and higher-order modes [59-62]. a perforated h-shaped mts [59] and a corner-truncated patch mts with capacitive loading strips [60] have been proposed by exciting adjacent higher-order modes and lower-order modes, respectively. recently, another non-uniform chebyshev distribution mts has been exploited as a superstrate by exciting two alternative modes such as mode 1 and mode 3 with an equal electric field and relative phase difference close to 90° [61]. apart from the arbw enhancement, the cp mts antenna has been proposed for radar cross-section (rcs) reduction [62]. here, by using cma, a linearly polarized slot antenna is converted to a cp antenna with the help of polarization-dependent mts and is additionally used for rcs reduction. in another design, a dual cp mts antenna has been proposed [63]. (a) s11 (b) ar fig. 14 mts antenna parameters [58] 252 advances in technology innovation, vol. 7, no. 4, 2022, pp. 242-257 in this design, the desired modes are chosen and excited by a modified slot antenna with two orthogonal feeds. the modified cross slots, called two diagonal linear slot arms and a circular ring slot, introduce another resonance for wide arbw. apart from the arbw enhancement, the mts is also used for polarization conversion [64]. here, various etching mechanisms, like diagonal slot, cross slot, and corner truncation, have been used on a linearly polarized rectangular mts. to achieve circular polarization, this has been done with the exciting slot antenna. modal analysis is used to verify cp conditions and optimize them. recently, an artificial magnetic conductor (amc)-based reflector cp antenna using a cpw-fed structure has been proposed for 5g sub-6 ghz communications with unidirectional radiation [65]. the periodic metallic square patches amc improve arbw from 27.27% to 51.67% and gain increases from 3.3 dbic to 8.7 dbic than the conventional pec. from the above discussion, it can be summarized that orthogonality is verified before the selection of a specific feed design. if any of the cp conditions are not satisfied for a simultaneous 90° phase difference for excitation, it has been optimized from the feed network and further analyzed. in some designs, the quadrature phase shift has not been attained from the cma and has been compensated by properly designing the feed network. moreover, the largest ms of different modes at the same frequency and different frequencies together with the 90° phase differences can generate narrowband and wideband arbw, respectively. studies have also indicated how to optimize cp antenna designs for two linear modes using mts. after identifying the cp waves in a particular frequency band with the cma, antenna designers concentrate their effort on the feed structure. moreover, in the literature, most of the cma has been carried out for isolated patches by neglecting the ground plane and dielectric material. the accuracy of the resonant frequencies will be affected by such kind of simplification. this is because the characteristic fields and currents are dependent on the size and shape of the patch as well as the dielectric substrate. hence, while considering cma, the dielectric substrate and ground plane cannot be ignored when seeking resonant frequencies of microstrip antennas. moreover, fig. 15 represents a potential mts utilized for circular polarization and arbw enhancement using cma. additionally, the design approaches discussed above are consolidated in fig. 16 based on the utilized feed network. moreover, the analysis of these parameters is performed before the selection of source excitation. thus, there is a degree of freedom in the feed network selection and feed location in the final antenna design. therefore, the techniques for the selection of feed networks and their final optimization with the help of cma are reviewed in the above section from the mentioned literature [50-65]. fig. 17 shows the antenna designs for various feeding methods. additionally, the important observations of all these antenna designs are summarized with their achieved operating band and arbw as shown in table 1. fig. 15 typical mts geometry utilized for circular polarization using cma fig. 16 cp antennas designed using cma with different feeding techniques 253 advances in technology innovation, vol. 7, no. 4, 2022, pp. 242-257 fig 17 typical cp geometry using cma with various feeding methods table 1 summary on various antenna design approaches for circular polarization ref. antenna design approach ar band (ghz) arbw (% and ghz) gain (dbi) observation [50] (a) an asymmetric u-slot antenna (b) a redundant e-slot antenna (a) 2.28-2.37 (b) 2.2-2.55 (a) 3.8 and 0.09 (b) 3.7 and 0.35 (a) 5.86 (b) 8.87 (a) an offset probe feed provides better ar performance than a center feed. (b) an insignificant mode is suppressed to get a redundant e-shaped slot antenna. [51] two crossed-dipole antennas 0.282-0.295.5 4.65 and 13.5 the length of one dipole is increased slightly to get a 90° phase shift. [54] a small semicircular slot rectangular antenna 2.25-4.4 64.6 and 2.15 the inverted-l stubs are used for desired phase difference and diagonal corner truncation for orthogonal modes. [55] an i-shaped radiating patch 6.6-11.8 56.5 and 5.2 5.5 the simultaneous excitation of even and odd modes from a rectangular stub, inverted-l stub, and spiral slot provides the wideband radiation behavior. [56] a clock-shaped antenna from a metallic ring antenna 2.38-5.8 84.6 and 3.42 3.9 a 90° phase difference is introduced by the two opposite microstrip lines in the x and y directions. [59] a crossed shape aperture with h-shaped unit cell mts 5.2-6 14.3 and 0.8 9.4 the additional required phase for circular polarization is compensated via a cross-shaped strip with an aperture. [60] a corner-truncated patch along with a pair of inserted capacitive loading strips 3.3-3.6 8.5 and 0.3 6.57 from cma, the pair of inserted diagonal capacitive loading strips of a corner-truncated patch is optimized for attaining quadrature phase difference and arbw improvement. [61] a non-uniform mts superstrate layer excited by a stripline through a rectangular slot 1.99-2.37 17.4 and 0.265 7.1-8 the phase difference between the orthogonal modes is observed only 60° and exploited as an inductive exciter for an additional 30° phase shift. [62] a rectangular patch as a polarization-dependent mts 5.83-6.32 9.05 and 0.49 6.4 the degenerated modes are achieved using polarization-dependent mts with 74° phase difference and excited with a linearly polarized slot antenna to achieve cp radiation. [63] an mts excited with a hybrid feed system consisting of a cross slot and a microstrip line 2.15-2.95 31.3 and 0.8 7.01 due to a cross-slot on the ground plane, 90° phase difference is presented via microstrip meander line excitation. [64] a square patch with a diagonal slot, crossed slot, and corner-truncated mts 2.32-2.46 2.55-2.58 2.52-2.54 5.8 and 0.14 1.1 and 0.03 0.7 and 0.02 5 3.5 3.5 due to asymmetry on the square patch, a phase difference is created for circular polarization. 4. conclusions this article concentrated on modal parameters that are required to analyze the natural mode resonance and radiating behavior, followed by a novel approach for designing an antenna using cma to improve arbw. the cm parameters together with characteristic currents explicitly gave useful information for analysis of the antenna before excitation. in particular, this review 254 advances in technology innovation, vol. 7, no. 4, 2022, pp. 242-257 presented how the arbw is enhanced in various developed structures from the theoretical formation analysis of the cm theory. the information provided by these developed structures helps understand valuable insights for the selection of feed position to maximize the antenna performance. moreover, the authors’ abundant contributions to the study of the cm theory were summarized based on the feed structures, utilizing mts with concluding remarks. fully exploiting and making use of cma in antenna designs could significantly enhance 5g and mm-wave applications from the dual-polarized cp antenna perspective. from a future perspective, if cp antennas can fulfill the needs of long-distance communication as done by high gain linearly polarized antennas, the cm-based approach becomes attractive in antenna designs. numerical models, physical interpretation insights, and cma would be available for a more detailed analysis of cp antennas with and without mts. this approach is also helpful to demonstrate that simultaneous mimo operations increase channel capacity and throughput. for efficient mimo performance, the radiation diversity is also explored for its low correlation coefficient. additionally, further research is going on to improve other important parameters, such as gain, efficiency, compactness, and polarization purity. finally, these challenges require the cma method to become more efficient for designing cp antennas. conflicts of interest the authors declare no conflict of interest. references [1] r. s. uqaili, j. a. uqaili, s. zahra, f. b. soomro, and a. akbar, “a study on dual-band microstrip rectangular patch antenna for wi-fi,” proceedings of engineering and technology innovation, vol. 16, pp. 1-12, august 2020. [2] s. patil, v. kapse, s. sharma, and a. k. pandey, “low profile wideband dual-ring slot antenna for biomedical applications,” proceedings of engineering and technology innovation, vol. 19, pp. 38-44, august 2021. [3] v. kollipara, s. peddakrishna, and j. kumar, “planar ebg loaded uwb monopole antenna with triple notch characteristics,” international journal of engineering and technology innovation, vol. 11, no. 4, pp. 294-304, september 2021. [4] a. kumar and a. p. s. pharwaha, “design and optimization of micro-machined sierpinski carpet fractal antenna using ant lion optimization,” international journal of engineering and technology innovation, vol. 10, no. 4, pp. 306-318, september 2020. [5] k. a. khan and s. m. nokerov, “optimization of multi-band characteristics in fan-stub shaped patch antenna for lte (cbrs) and wlan bands,” proceedings of engineering and technology innovation, vol. 18, pp. 25-35, april 2021. [6] s. s. gao, q. luo, and f. zhu, circularly polarized antennas, uk: ieee press, 2013. [7] r. zenter, z. sipus, and j. bartolic, “optimization synthesis of broadband circularly polarized microstrip antennas by hybrid genetic algorithm,” microwave and optical technology letters, vol. 31, no. 3, pp. 197-201, november 2001. [8] r. l. haupt, “antenna design with a mixed integer genetic algorithm,” ieee transactions on antennas and propagation, vol. 55, no. 3, pp. 577-582, march 2007. [9] y. chen and c. f. wang, characteristic modes: theory and applications in antenna engineering, hoboken: wiley, 2015. [10] b. yang and j. j. adams, “computing and visualizing the input parameters of arbitrary planar antennas via eigenfunctions,” ieee transactions on antennas and propagation, vol. 64, no. 7, pp. 2707-2718, july 2016. [11] b. b. q. elias, p. j. soh, a. a. al-hadi, p. akkaraekthalin, and g. a. vandenbosh, “a review of antenna analysis using characteristic modes,” ieee access, vol. 9, pp. 98833-98865, july 2021. [12] r. j. garbacz, “modal expansions for resonance scattering and phenomena,” proceedings of the ieee, vol. 53, no. 8, pp. 856-864, august 1965. [13] r. garbacz and r. turpin, “a generalized expansion for radiated and scattered fields,” ieee transactions on antennas and propagation, vol. 19, no. 3, pp. 348-358, may 1971. [14] r. harrington and j. mautz, “theory of characteristic modes for conducting bodies,” ieee transactions on antennas and propagation, vol. 19, no. 5, pp. 622-628, september 1971. [15] e. safin and d. manteuffel, “reconstruction of the characteristic modes on an antenna based on the radiated far field,” ieee transactions on antennas and propagation, vol. 61, no. 6, pp. 2964-2971, june 2013. [16] t. bernabeu-jiménez, a. valero-nogueira, f. vico-bondia, e. antonino-daviu, and m. cabedo-fabres, “a 60-ghz ltcc rectangular dielectric resonator antenna design with characteristic modes theory,” ieee antennas and propagation society international symposium, pp. 1928-1929, july 2014. 255 https://ieeexplore.ieee.org/author/37085500002 https://ieeexplore.ieee.org/author/38183890100 https://ieeexplore.ieee.org/author/38278107100 https://ieeexplore.ieee.org/author/38270911200 https://ieeexplore.ieee.org/author/38271010200 advances in technology innovation, vol. 7, no. 4, 2022, pp. 242-257 [17] r. t. maximidis, c. l. zekios, t. n. kaifas, e. e. vafiadis, and g. a. kyriacou, “characteristic mode analysis of composite metal-dielectric structure based on surface integral equation/moment method,” 8th european conference on antennas and propagation, pp. 2822-2826, april 2014. [18] h. alroughani, j. ethier, and d. a. mcnamara, “observations on computational outcomes for the characteristic modes of dielectric objects,” ieee antennas and propagation society international symposium, pp. 844-845, july 2014. [19] y. chen and c. f. wang, “surface integral equation based characteristic mode formulation for dielectric resonators,” ieee antennas and propagation society international symposium, pp. 846-847, july 2014. [20] z. miers, h. li, and b. k. lau, “design of bezel antennas for multiband mimo terminals using characteristic modes,” 8th european conference on antennas and propagation, pp. 2556-2560, april 2014. [21] r. rezaiesarlak and m. manteghi, “on the application of characteristic modes for the design of chipless rfid tags,” ieee antennas and propagation society international symposium, pp. 1304-1305, july 2014. [22] e. a. elghannai and r. g. rojas, “design of usb dongle antenna for wlan applications using theory of characteristic modes,” electronics letters, vol. 50, no. 4, pp. 249-251, february 2014. [23] j. chalas, k. sertel, and j. l. volakis, “nvis synthesis for electrically small aircraft using characteristic modes,” ieee antennas and propagation society international symposium, pp. 1431-1432, july 2014. [24] a. krewski, “2-port mimo antenna system for high frequency data link with uavs using characteristic modes,” ieee antennas and propagation society international symposium, pp. 364-365, july 2013. [25] y. chen and c. f. wang, “electrically small uav antenna design using characteristic modes,” ieee transactions on antennas and propagation, vol. 62, no. 2, pp. 535-545, february 2014. [26] y. chen and c. f. wang, “shipboard nvis radiation system design using the theory of characteristic modes,” ieee antennas and propagation society international symposium, pp. 852-853, july 2014. [27] b. a. austin and k. p. murray, “the application of characteristic-mode techniques to vehicle-mounted nvis antennas,” ieee antennas and propagation magazine, vol. 40, no. 1, pp. 7-21, february 1998. [28] k. p. murray and b. a. austin, “synthesis of vehicular antenna nvis radiation patterns using the method of characteristic modes,” iee proceedings—microwaves, antennas, and propagation, vol. 141, no. 3, pp. 151-154, june 1994. [29] m. ignatenko and d. filipovic, “application of characteristic mode analysis to hf low profile vehicular antennas,” ieee antennas and propagation society international symposium, pp. 850-851, july 2014. [30] a. mohanty and b. r. behera. “characteristics mode analysis: a review of its concepts, recent trends, state-of-the-art developments and its interpretation with a fractal uwb mimo antenna,” progress in electromagnetics research b, vol. 92, pp. 19-45, march 2021. [31] a. mohanty and b. r. behera, “cma assisted 4-port compact mimo antenna with dual-polarization characteristics,” aeü —international journal of electronics and communications, vol. 137, article no. 153794, may 2021. [32] a. h. jabire, h. x. zheng, a. abdu, and z. song, “characteristic mode analysis and design of wide band mimo antenna consisting of metamaterial unit cell,” electronics, vol. 8, no. 1, article no. 68, january 2019. [33] t. li and z. n. chen, “design of dual band metasurface antenna array using characteristic mode analysis (cma) for 5g millimeter-wave applications,” ieee-aps topical conference on antennas and propagation in wireless communications, pp. 721-724, september 2018. [34] b. feng, x. he, j. cheng, and c. sim, “dual-wideband dual-polarized metasurface antenna array for the 5g millimeter wave communications based on characteristic mode theory,” ieee access, vol. 8, pp. 21589-21601, january 2020. [35] k. gao, x. ding, l. gu, y. zhao, and z. nie, “a broadband dual circularly polarized shared‐aperture antenna array using characteristic mode analysis for 5g applications,” international journal of rf and microwave computer-aided engineering, vol. 31, no. 3, article no. e22539, march 2021. [36] a. abdelaziz, h. a. mohamed, and e. k. hamad, “applying characteristic mode analysis to systematically design of 5g logarithmic spiral mimo patch antenna,” ieee access, vol. 9, pp. 156566-156580, november 2021. [37] a. kishk and l. shafai, “different formulations for numerical solution of single or multibodies of revolution with mixed boundary conditions,” ieee transactions on antennas and propagation, vol. 34, no. 5, pp. 666-673, may 1986. [38] a. sharif, j. ouyang, f. yang, r. long, and m. k. ishfaq, “tunable platform tolerant antenna design for rfid and iot applications using characteristic mode analysis,” wireless communications and mobile computing, vol. 2018, article no. 9546854, 2018. [39] r. martens, e. safin, and d. manteuffel, “inductive and capacitive excitation of the characteristic modes of small terminals,” loughborough antennas and propagation conference, pp. 1-4, november 2011. [40] c. a. balanis, antenna theory: analysis and design, 3rd ed., new jersey: john wiley & sons, 2005. [41] y. dong, h. toyao, and t. itoh, “compact circularly-polarized patch antenna loaded with metamaterial structures,” ieee transactions on antennas and propagation, vol. 59, no. 11, pp. 4329-4334, november 2011. [42] j. s. row and c. y. ai, “compact design of single-feed circularly polarized microstrip antenna,” electronics letters, vol. 40, no. 18, pp. 1093-1094, september 2004. 256 advances in technology innovation, vol. 7, no. 4, 2022, pp. 242-257 [43] q. zhu, s. yang, and z. chen, “modified corner-fed dual polarised stacked patch antenna for micro-base station applications,” electronics letters, vol. 51, no. 8, pp. 604-606, april 2015. [44] j. y. sze and c. c. chang, “circularly polarized square slot antenna with a pair of inverted-l grounded strips,” ieee antennas and wireless propagation letters, vol. 7, pp. 149-151, march 2008. [45] y. sung, “axial ratio-tuned circularly polarized square patch antenna with long stubs,” international journal of antennas and propagation, vol. 2018, article no. 7068560, 2018. [46] c. f. zhou and s. w. cheung, “a wideband cp crossed slot antenna using 1-λ resonant mode with single feeding,” ieee transactions on antennas and propagation, vol. 65, no. 8, pp. 4268-4273, august 2017. [47] c. y. d. sim, h. d. chen, and l. zuo, “cpw-fed square ring slot antenna with circular polarization radiation for wimax/wlan applications,” microwave and optical technology letters, vol. 57, no. 4, pp. 886-891, february 2015. [48] s. dentri, c. phongcharoenpanich, and k. kaemarungsi, “single-fed broadband circularly polarized unidirectional antenna using folded plate with parasitic patch for universal uhf rfid readers,” international journal of rf microwave computer-aided engineering, vol. 26, no. 7, pp. 575-587, september 2016. [49] s. b. vignesh, nasimuddin, and a. alphones, “stubs-integrated-microstrip antenna design for wide coverage of circularly polarised radiation,” iet microwaves, antennas, and propagation, vol. 11, no. 4, pp. 444-449, september 2016. [50] y. chen and c. f. wang, “characteristic-mode-based improvement of circularly polarized u-slot and e-shaped patch antennas,” ieee antennas and wireless propagation letters, vol. 11, pp. 1474-1477, november 2012. [51] j. p. ciafardini, e. a. daviu, m. c. fabrés, n. m. m. hicho, j. a. bava, and m. f. bataller, “analysis of crossed dipole to obtain circular polarization applying characteristic modes techniques,” ieee biennial congress of argentina, pp. 1-5, june 2016. [52] j. f. lin and q. x. chu, “a modal approach to evaluating the axial ratio of circularly polarized antennas,” ieee international conference on computational electromagnetics, pp. 202-204, february 2016. [53] h. h. tran, n. nguyen-trong, and a. m. abbosh, “simple design procedure of a broadband circularly polarized slot monopole antenna assisted by characteristic mode analysis,” ieee access, vol. 6, pp. 78386-78393, 2018. [54] k. saraswat and a. r. harish, “analysis of wideband circularly polarized ring slot antenna using characteristics mode for bandwidth enhancement,” international journal of rf and microwave computer-aided engineering, vol. 28, no. 2, article no. e21186, september 2017. [55] m. m. s. jaiverdhan, r. p. yadav, and r. dhara, “characteristic mode analysis and design of broadband circularly polarized cpw-fed compact printed square slot antenna,” progress in electromagnetics research m, vol. 94, pp. 105-118, july 2020. [56] m. han and w. dou, “compact clock-shaped broadband circularly polarized antenna based on characteristic mode analysis,” ieee access, vol. 7, pp. 159952-159959, november 2019. [57] y. chen, x. li, z. qi, h. zhu, and s. zhao, “a single-feed circularly polarized loop antenna using characteristic mode analysis,” international journal of rf and microwave computer‐aided engineering, vol. 31, no. 7, article no. e22648, july 2021. [58] x. gao, g. tian, z. shou, and s. li, “a low-profile broadband circularly polarized patch antenna based on characteristic mode analysis,” ieee antennas and wireless propagation letters, vol. 20, no. 2, pp. 214-218, december 2020. [59] c. zhao and c. f. wang, “characteristic mode design of wide band circularly polarized patch antenna consisting of h-shaped unit cells,” ieee access, vol. 6, pp. 25292-25299, 2018. [60] y. juan, w. yang, and w. che, “miniaturized low-profile circularly polarized metasurface antenna using capacitive loading,” ieee transactions on antennas and propagation, vol. 67, no. 5, pp. 3527-3532, may 2019. [61] f. a. dicandia and s. genovesi, “characteristic modes analysis of non-uniform metasurface superstrate for nanosatellite antenna design,” ieee access, vol. 8, pp. 176050-176061, 2020. [62] p. k. rajanna, k. rudramuni, and k. kandasamy, “characteristic mode-based compact circularly polarized metasurface antenna for in-band rcs reduction,” international journal of microwave and wireless technologies, vol. 12, no. 2, pp. 131-137, march 2020. [63] s. liu, d. yang, and j. pan, “a low-profile broadband dual-circularly-polarized metasurface antenna,” ieee antennas and wireless propagation letters, vol. 18, no. 7, pp. 1395-1399, july 2019. [64] p. v. kumar and b. ghosh, “characteristic mode analysis of linear to circular polarization conversion metasurface,” electromagnetics, vol. 40, no. 8, pp. 605-612, march 2020. [65] h. askari, n. hussain, d. choi, m. a. sufian, a. abbas, and n. kim, “an amc-based circularly polarized antenna for 5g sub-6 ghz communications,” computers, materials, and continua, vol. 69, no. 3, pp. 2997-3013, august 2021. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). 257 advances in technology innovation, vol. 3, no. 2, 2018, pp. 70 77 energy-effective predictive temperature control for soy mash fermentation based on compartmental pharmacokinetic modelling sophia ferng 1 , ching-hua ting 2,* , chien-ping wu 2 , yung-tsong lu 3 , cheng-kuang hsu 1 , robin yih-yuan chiou 1 1 department of food science, national chiayi university, chiayi, taiwan, roc. 2 department of mechanical and energy engineering, national chiayi university, chiayi, taiwan, roc. 3 department of biomechatronic engineering, national chiayi university, chiayi, taiwan, roc. received 07 june 2017; received in revised form 05 sept ember 2017; accept ed 02 oct ober 2017 abstract co mpart ment modelling has been successfully used in pharmacokinetics to describe the kinetics of drug distribution in body tissues. in this study, the technique is adopted to describe the dynamics of temperature response and energy exchange in a soy mash fermentation system. the object ive is to provide a prec ise temperature-controlled at mosphere for effect ive fermentation with the pre mise of energy saving. in analogy to pharmacokinetics, water and mash tanks are treated as compartments, energy flow as drug delivery, and the temperature as the drug concentration in a specific co mpart ment. the model allows us to estimate the time of injecting a certa in a mount of energy to a specific tank (co mpart ment) in a cost-effective way. thus, model-based temperature control and energy management can be possible. keywor ds: soy mash fermentation, temperature, predictive control, energy -effective, compartment model 1. introduction soy sauce was invented by the chinese about 3500 years ago , and the modern producing technology was developed by the japanese about 500 years ago. the production consists of solid -state fermentation, mash fermentation, and flavouring. the quality and cost of a soy sauce product are ma inly determined by the mash fermentation stage. in tradit ion, soy mash is placed in a pottery tank wh ich is e xposed to sunshine for 4~12 months. the duration of e xposure under the sun is season-dependent as the enzyme and the microbes in the mash are sensible to temperature variations [1-4]. despite the cost raised by a longer fermentation time and more manpower, tradit ional soy sauces still deserve of popularity as their special flavours are superior to cheap, chemical ones . because the mash fermentation is manipulated outdoors, the processes of fe rmentation are easily affected by the climate, and the mash is vulnerable to a lien contamination [5]. this may lead to a produce with unstable quality and a lien contamination. to overcome this proble m, we moved the fermentation fro m outdoors to indoors and controlled the fe rmentation c limate [5], as shown in fig. 1. in the system, the pottery containing soy mash is bathed in a water tank and the temperature of the batching water is regulated to meet the require ments of various fermentation stages [5-6]. it has been demonstrated that a controlled fermentation climate can c reate a we ll environ ment fo r the enzy mes and microbes partic ipating in mash fermentation [7-8]. thus, the produce can arrive at a better quality in a shorter time in comparison with the traditional approach [9]. * corresponding author. e-mail address: cting@mail.ncyu.edu.tw tel.: +886-5-2717642; fax: +886-5-2717561 advances in technology innovation, vol. 3, no. 2, 2018, pp. 70 77 copyright © taeti 71 fig. 1 setup of the energy-saving soy sauce fermentation system [5]. as fig. 1 illustrates , the system consists of a heat pump which supplies hot water circulation fo r regulat ing the fermenter with desired temperature settings . the system is controlled with a programmab le logic controller (plc) and supervised with an industrial personal co mputer (ipc). the current principle of operation is to let the heat pu mp work at noontime for the benefit of a better operational efficiency and the hot water produce is stored in a storage tank for later use , usually in night time . the plc man ipulates the solenoid valves and the circulation pump once the tempe rature of the mash is below a certain leve l. these two solenoid valves assure correct water flow directions. hot water will c irculate for 3 min for adequate heat injection into the soy mash. the duration was determined through e xperimental studies via tria l-and-error. fig. 2 de monstrates the temperature profiles of the soy mash, the at mosphere, and the circu lation wate r. clea rly, the mash can ferment at an environ ment with a mo re stable te mperature regardless of a vary ing at mospheric climate. though, the temperature variation in the fermenter has been imp roved to be with in 1°c, much improved regardless of vary ing climate, it still fluctuates as a result of simple on/off control using solenoid valves. (a) temperature responses of the traditional procedure (b) temperature responses of on/off control fig. 2 the temperature profiles of the soy mash, the atmosphere, and the water in the bathing tank the price of electric ity is diffe rent between peak and off-peak times . the heat pu mp has a ma ximu m efficiency at noontime . ho wever, the demand of hot water circulation is usually in night time, several hours after the hot water preparation. the stored hot water will inevitably lose some heat to the space before being consumed. there should be an optimu m time of operat ing the heat pump in term of e lectric ity cost [10-12]. it would be possible to operate the heat pump at a most cost-effective way if we can determine when hot water circulation is demanded. the objective of this study is to develop a temperature and energy prediction model using the compartment modelling technique which has been successfully used in pharmacology for drug ad ministration [ 13]. as the system described in fig. 1, we can partition the system into several unique co mpart ments and describe mathe matica lly the energy e xchanges among the advances in technology innovation, vol. 3, no. 2, 2018, pp. 70 77 copyright © taeti 72 compart ments . in analogy to pharmacokinetics, the water storage tanks and the fermentation tanks are treated as compart ments, energy e xchange among tanks and the at mosphere as drug delivery, the te mperature as drug concentration, electrica l energy injected into the heat pump can be treated as a dose input. we can estimate energy flows, temperatures responses , and hence, electric ity consumption. based on this, we can determine the best time of heat pu mp operation and control the time and duration of hot water circulat ion using simple and cheap solenoid valves . hence, the profit can be promoted as a result of better and efficient fermentation without too much of energy consumption and therefore , secured environmental affect. accordingly, a compromise can be arrived at among economy, ecology, and energy (3e). 2. materials and methods 2.1. compartment modelling fig. 3 illustrates the proposed compartment mode l. the heat pu mp is identified as an energy supply to the hot water circulat ion system. the soy mash pot (co mpart ment 2) is bathed in hot water ( co mpart ment 1). there is heat e xchange in between. co mpa rt ment 1 will inevitably loss heat to the atmosphere. co mpart ment 3 stores hot water p roduce fro m the heat pump. heat energy is supplied fro m co mpart ment 3 to co mpart ment 1 via water circulation. coo led water is to be circulated back to co mpart ment 3 fro m co mpart ment 1. co mpart ment 1 is equipped with a well-designed stirrer that mixes incoming hot water and the e xisting cool water efficiently. in other word, energy is in jected into co mpart ment 1 as a dose bolus rather than via heat transfer. thus, there is no 𝑘31 and 𝑘13 corre lation in between. to simplify the proble m, the storage tank (co mpart ment 3) is treated as an infinite tank comparing to compart ment 1. since the hot water storage tank is well insulated, no-heat-loss is assumed, ie. 𝑘30 = 0 . eq. (1) describes the dynamics of energy transfer based on the compart mental model; where 𝑇1 is the temperature o f co mpart ment 1, 𝑇2 is the te mperature of co mpart ment 2, and r(t) is the injected energy dose:    1 10 12 1 21 2 2 12 1 21 2 dt r t k k t k t dx dt k t k t dx                (1) 𝑘12：the transfer constant of heat in circulation water transfer to soy mash (𝑠 −1 ) 𝑘21 ：the transfer constant of heat in soy mash transfer to circulation water (𝑠 −1 ) 𝑘10：the transfer constant of heat in circulation water dissipate into the air (𝑠 −1 ) 𝑘30 ：the transfer constant of heat in storage water dissipate into the air (𝑠 −1 ) fig. 3 the proposed compartment model 2.2. determination of system parameters fig. 4 shows the experimental setup for determin ing the k coefficients. tanks were filled with water of different temperatures and hence, energy flowed fro m high te mperature to low te mperature sites. te mperature responses were recorded every 10 min till thermodynamic equilibriu m. acquired data were fitted to first-order equations using the matlab curve fitting toolbox. advances in technology innovation, vol. 3, no. 2, 2018, pp. 70 77 copyright © taeti 73 (a) for 𝑘12 (b) for 𝑘21 (c) for 𝑘10 fig. 4 experimental setups for determining the k coefficients other para meters to be determined are the volume of the circulat ion water, the volu me of the soy mash and the hot water circulation time. 2.3. energy consumption cost function the heat pump e xtracts energy from the air to heat up water. the ambient te mperature is the key factor that influences the effic iency of heat pump operation [11]. the higher the a mb ient temperature is, the higher the heat pump effic iency. hence to operate the heat pu mp at noontime will have the ma ximu m operational e fficiency. ho wever, the noontime is categorized as a peak t ime in e lectric ity pricing, ie. the p rice of electric ity is h igher than other times. the mash fermentation tank will have a lower te mperature at n ight time and therefore, the hot water prepared at noontime is not used for several hours . in other word, the hot water produced by the heat pump at noontime will loss a big amount of energy to the atmosphere. thus, it may be more profitable by operating the heat pump away fro m noontime as a co mpro mise among electric ity price, heat pu mp effic iency, and energy loss. accordingly, an energy consumption cost function is to be developed to account for the efficiency of the heat pump, the electricity pricing policy, and the heat loss of the hot water storage tank. 3. results and discussion 3.1. system parameters te mperature responses were acquired with a sampling period of 10 min for 6 h wh ich is long enough for the system to reach thermodynamic equilibriu m. the acquired data are fitted to a first-order equation using the matlab curve fitting toolbox. fig. 5 shows the temperature responses. (a) for 𝑘12 (b) for 𝑘21 (c) for 𝑘10 fig. 5 temperature responses for determining the compartment -model coefficients fig. 5(a) is the temperature profile for 𝑘12 and can be described as: 𝑇 = 5.727𝑒 −𝑡 7.143 ⁄ + 39.37 °c (2) with a t ime resolution of 10 min (600 s ). the t ime constant is 7.143 × 600 = 4285.8 s and its reciprocal gives 𝑘12 = 2.33 × 10 −4 𝑠 −1. advances in technology innovation, vol. 3, no. 2, 2018, pp. 70 77 copyright © taeti 74 fig. 5(b) is the temperature profile for 𝑘21 and can be described as: 𝑇 = 9.86𝑒−𝑡 5 .821 ⁄ + 36.52 °c (3) the time constant is 5.821 × 600 = 3492.6 s and its reciprocal gives 𝑘12 = 2.86 × 10 −4 𝑠 −1 . fig. 5(c ) is the temperature profile for k 10 . the bathing tank is cladded with a good insulation materia l. hence as the profile shows, this is a very slow heat transfer s ystem. it took 11 days to reach thermodynamic equilibriu m and resulted in 1716 record ings . these data are down-sampled with a factor of 50. the profile can be described with first -order dynamics, as: 𝑇 = 18.85𝑒 −𝑡 8.523 ⁄ + 28.81 °c (4) the new sampling period is 60 × 10 × 50 = 3 × 104 s , the time constant is 8.523 × 30000 = 255690 s , and the coefficient is 𝑘10 = 3.91 × 10 −6 𝑠 −1. 3.1.1. other constants co mpart ment 1 has a space of 220 litres for circulat ion water and co mpart ment 2 acco mmodates 160 litres of soy mash. 3.1.2. thermodynamics of the circulation water the purpose of water c irculat ion control is to regulate the temperature of the soy mash in co mpart ment 2 through controlling the heat in jected into co mpart ment 1. fig. 6 illustrates the conceptual thermodynamics of the two co mpart ments. co mpart ment 1 has a water d istributor designed to mix the inco ming water with the e xisting one in an effective way. hence the two med ia a re assumed to mix instantaneously. the first la w of thermodyna mics describes the heat balance of the thermodynamic system, as: 𝑄𝑖𝑛 + 𝑄𝑜𝑟𝑖𝑔𝑖𝑛 − 𝑄𝑜𝑢𝑡 = 𝑄𝑓𝑖𝑛𝑎𝑙 (5) fig. 6 heat balance in the batching tank substituting 𝑄 = 𝑚 × ℎ𝑓 to eq. (4) and discretizing give, �̇� ℎ × ∆𝑡 × ℎ𝑓 (𝑇ℎ ) + 𝑚𝑏 × ℎ𝑓 (𝑇𝑏 ,𝑘−1 ) − �̇� ℎ × ∆𝑡 × ℎ𝑓 (𝑇𝑏 ,𝑘−1 ) = 𝑚𝑏 × ℎ𝑓 (𝑇𝑏 ,𝑘 ) (6) with �̇� ℎ the circulating water flow rate, ℎ𝑓 the enthalpy of water at a specific te mperature, 𝑚𝑏 the mass of the bathing water, 𝑇ℎ the temperature of the incoming hot water, 𝑇𝑏 the temperature of the bathing water, ∆𝑡 the heating duration, and k the time stamp. the enthalpy can be approximated with the specific heat, ie. ℎ𝑓 ≅ 𝐶𝑝 × 𝑇 , hence eq. (6) can be rewritten as : �̇� ℎ × ∆𝑡 × 𝐶𝑝 × 𝑇ℎ + 𝑚𝑏 × 𝐶𝑝 × 𝑇𝑏 ,𝑘−1 − �̇�ℎ × ∆𝑡 × 𝐶𝑝 × 𝑇𝑏 ,𝑘−1 = 𝑚𝑏 × 𝐶𝑝 × 𝑇𝑏 ,𝑘 (7) rearranging the above gives the temperature response: 𝑇𝑏 ,𝑘 − 𝑇𝑏 ,𝑘−1 = ∆𝑇𝑏 = �̇� ℎ ×∆𝑡 𝑚𝑏 (𝑇ℎ − 𝑇𝑏 ,𝑘−1 ) (8) the follo wing values are used for simu lation validation: �̇�ℎ = 10kg/min, ∆𝑡 = 3 min, 𝑚𝑏 = 220 kg, 𝑇ℎ = 47℃. the water storage tank (co mpart ment 3) has a capacity of 1500 lit res and only 30 litres are de manded for the c irculat ion. advances in technology innovation, vol. 3, no. 2, 2018, pp. 70 77 copyright © taeti 75 hence it is reasonable to assume a constant 𝑇ℎ . the initia l te mperature of the bathing water 𝑇𝑏,k −1 is assumed as 36.8°c. substituting the above values into eq. ( 8) results in ∆𝑇𝑏 = 1.39°c , ie. a 3-min heat in jection heats up the bathing water by 1.39°c . this information will be used in compartment modelling. 3.1.3. validation the compart ment model of fig. 3 was validated using the matlab simbiology toolbo x using the para meters identified in section 3. fig. 7 shows simu lation and actual responses . there is a strong degree of coherence in between. this promising result indicates that the proposed compartment mode l can e ffective ly describes the energy fl ows of the fermentation system. fig. 7 simulation and actual responses it is clea r that the t iming of operating the heat pu mp is the key issue of saving energy cost for the fermentation system. the timing should account for the operational effic iency, electricity pricing, and heat losses from tanks. thus, the above three factors are transcribed as an energy cost function : minimis ing the function for the least energy bill. in this study, the heat pump was set to produce hot water at 47℃. the energy for 1 litre of hot water produce is: 𝑄 = 𝑚 × 𝐶𝑝 × ∆𝑇 = 1 × 4.2 × (47 − 𝑇𝑡 ) = 197.4 − 4.2 𝑇𝑡 (9) with 𝑇𝑡 the temperature of the water in the storage tank, which can be estimated using the compartment model . the performance of a heat pump is described with the so-called coeffic ient of pe rformance (cop). it is a ratio of the heat generated, q, by the heat pump to the energy consumed, w, by the heat pump [14], as: 𝐶𝑂𝑃 = 𝑄 𝑊 (10) the work done by the heat-pump compressor is: 𝑊 = 𝑄 𝐶𝑂𝑃 = 197 .4−4.2 𝑇𝑡 𝐶𝑂𝑃 𝑘𝐽 (11) the cop coefficient is sensitive to the amb ient temperature 𝑇𝑎 . table 1 lists the cops of the heat pump (chp -80y, sun tech, taiwan). the hotter the ambient is, the larger the cop. table 1 cops of the heat pump at different ambient temperatures ambient (°m) -20 -15 -10 -5 0 2 7 10 16 20 25 30 35 43 cop 2.2 2.4 2.6 2.5 2.6 2.8 3.7 4.0 4.2 4.3 4.4 4.5 4.4 4.3 advances in technology innovation, vol. 3, no. 2, 2018, pp. 70 77 copyright © taeti 76 the comp ressor of the heat pump is assumed to have an effic iency η=0.9. the work de manded by the heat pump to produce 1 litre of hot water is: 𝑊𝑖 = 𝑊 𝜂 = 197.4 − 4.2 𝑇𝑡 𝐶𝑂𝑃 ÷ 0.9 = 219.33 − 4.67 𝑇𝑡 𝐶𝑂𝑃 kj (12) or in kwh, as: 𝑊𝑖 = 219.33 − 4.67 𝑇𝑡 𝐶𝑂𝑃 ÷ (3.6 × 103 ) = 0.060925 − 0.001297 𝑇𝑡 𝐶𝑂𝑃 𝐾𝑊𝐻 (13) the cost c, in ntd, that the heat pump take to produce 1 litre of hot water is: 𝐶 = 𝐸 𝐶𝑂𝑃 (0.060925 − 0.001297 𝑇𝑡 ) (14) where e is the price of unit kwh in nt d/kw h, a value varies at peak and off-peak times. table 2 shows the pricing policy of taiwan power co mpany (tpc). the above cost function is a function of the time of heat pump operation (transcribed as cop and e) and heat losses from the storage tanks (transcribed as 𝑇𝑡 ). table 2 pricing policy of taiwan power company 4. conclusions we have built a co mpart ment model for describing te mperature response, heat exchange, and energy demand of a soy mash fermentation system. there is a strong coherence between simulat ion and actual results . simu lation results show that the model can accurate ly predict the trend of te mperature response. thus, our ne xt work is to imp le ment model -based predictive temperature control based on the developed compart ment model. this will g ive a more precise te mperature control. the model a lso describes the heat exchange, or energy flow, between tanks (compart ments). this feature allows us to estimate the optimu m t ime of running the heat pump. hence we can arrive at a p ro mising te mperature control with an economic electricity bill. acknowledgements this work was sponsored by the ministry of sc ience and technology of the roc (taiwan) govern ment under the grant most 105-2221-e-415-030, and by the research center for energy technology and strategy of national chengkung university, taiwan, roc. references [1] n. w. su, m. l. wang, k. f. kwok, and m. h. lee, “effects of temperature and sodium chloride concentration on the activities of proteases and amylases in soy sauce koji,” journal of agricultural and food chemistry, vol. 53, no. 5, pp. 1521-1525, march 2005. [2] y. zhu and j. tramper, “koji: where east meets west in fermentation,” biotechnology advances, vol. 31, no. 8, pp. 1448-1457, december 2013. summertime non-summertime peak time 07:30~22:30 3.24 3.18 00:00~07:30 22:30~24:00 00:00~07:30 22:30~24:00 sun and off-peak day mon~fri sat 2.06 1.39 1.33 all day 1.39 1.33 off-peak time off-peak time half-peak time 07:30~22:30 1.39 1.33 2.14 off-peak time advances in technology innovation, vol. 3, no. 2, 2018, pp. 70 77 copyright © taeti 77 [3] j. s. lee, s. j. rho, y. w. kim, k. w. lee, and h. g. lee, “evaluation of biological activities of the short -term fermented soybean extract,” food science and biotechnology, vol. 22, no. 4, pp. 973-978, august 2013. [4] c. cui, m. zhao, d. li, h. zhao, and w. sun, “biochemical changes of traditional chinese -type soy sauce produced in four seasons during processing,” cyta journal of food, vol. 12, no. 2, pp. 166-175, october 2014. [5] s. h. ferng, c. p. wu, y. t. lu, w. h. huang, t. c. lin, c. k. hsu, j. s. ju, and c. h. ting, “automatic control of soy mash fermentation for performance promotion using green energy with energy conservation,” the 8th international symp. machinery and mechatronics for agriculture and biosystems engineering (ismab), niigata, japan, may 2016. [6] t. r. chen, and r. y. y. chiou, “microbial and compositional changes of inyu (black soybean sauce) broth during fermentation under various temperature conditions ,” journal of the chinese agricultural che mica l soc iety, vol. 34, no. 2, pp. 157-164, 1995. [7] x. gao, c. cui, h. zhao, m. zhao, l. yang, and j. ren, “changes in volatile aroma compounds of traditional chinese-type soy sauce during moromi fermentation and heat treatment,” food science and biotechnology, vol. 19, no. 4, pp. 889-898, 2010. [8] n. w. su, m. l. wang, k. f. kwok, and m. h. lee, “effects of temperature and sodium chloride concentration on the activities of proteases and amylases in soy sauce koji,” journal of agricultural and food chemistry, vol. 53, pp. 1521-1525, august 2005. [9] n. x. hoang, s. ferng, c. h. ting, w. h. huang, r. y. y. chiou, and c. k. hsu, “optimizing the initial moromi fermentation conditions to improve the quality of soy sauce,” lwt food science and technology, vol. 74, pp. 242-250, december 2016. [10] t. olsson, “evaluating machine learning for predicting next -day hot water production of a heat pump ,” 2013 fourth international conf. power engineering, energy and electrical drives (powereng), may 2013. [11] j. cummings and c. withers, “energy savings and peak demand reduction of a seer 21 heat pump vs. a seer 13 heat pump with attic and indoor duct systems ,” us department of energy, march 2014. [12] m. kassai, “a developed method for energy saving prediction of heat -and energy recovery units ,” energy procedia, vol. 85, pp. 311-319, january 2016. [13] w. b. runciman and r. n. upton, anaesthetic pharmacology review: pharmacokinetics , castle house publications, 1994. [14] c. m. faye, d. p. jerald, d. s. jeffrey, heating, ventilating, and air conditioning: analysis and design, 6th ed. john wiley & sons , 2004.  advances in technology innovation, vol. 3, no. 1, 2018, pp. 17 25 prediction primary available blend biodiesel of waste oil from aurantiochytrium sp. for general diesel engines shu-yao tsai 1 , hsiang-yu lin 2 , guan-yi lu 1 , chun-ping lin 1,* 1 department of health and nutrition biotechnology, asia university, taichung, taiwa n, roc. 2 department of neonatology, children’s hospital, china medical university hospital, taichung, taiwan, roc. received 05 june 2017; received in revised form 27 july 2017; accept ed 11 august 2017 abstract chemical and enzyme transesterification were compared by discussing preliminary transesterification of waste oil of aurantiochytrium sp., which was then used in transesterification for the primary available blend biodiesel for a general diesel engine in this study. we made progress on the winterized characteristics of the waste oil’s biodiesel of aurantiochytrium sp. and its biodiesel, which included the reactivity parameters and properties. this approach led to the development of a novel idea for the evaluation of kinetic parameters of winterization, along with obtaining the suitable operation and storage conditions of biodiesel. therefore, the waste oil of aurantiochytrium sp. could be developed for biodiesel production and successfully made into a suitable blend diesel. overall, we acquired the best condition of mixtures and the highly mixed rate of petrodiesel: biodiesel = 80 : 20 (activation energy of winterization 21.32 kj/mol; onset temperature of winterization -4.15 °c; heat of combustion 43.15 mj/kg; kinematic viscosity 3.51 mm 2 /s; flash point 67.5 °c), which was an appropriate blend biodiesel from the waste oil’s biodiesel of aurantiochytrium sp. keywords: waste oil, aurantiochytrium sp., biodiesel, winterization, blend biodiesel 1. introduction this study focused on aurantiochytrium sp., which is a type of microalgae, the oil of which is refined for docosahexaenoic acid (dha) for incorporating into health food or food additives [1]. more than 20 mass% of the solid waste of aurantiochytrium sp. is discarded as rubbish under current operating procedures in food factories. in this study, we implemented the preliminary transesterification of solid waste of aurantiochytrium sp., to determine the advantages and disadvantages of chemical and enzyme transesterification. chemical [1-7] and enzyme [8-15] transesterification were compared to discuss the preliminary transesterification of waste oil of aurantiochytrium sp., which could then be beneficially used in the development of transesterification for an available mixture of biodiesel. the main stages o f chemical transesterification are as follows [1-7]: (a) saponification fixes the fatty acids, which are used for enhancing the yield of the transesterification reaction and removing impurities, glycerol, and free fatty acid; (b) reduction reaction, which reverts the fatty acids for convenient converting to biodiesel; and (c) acid-catalysed reaction, which makes the fatty acids and methanol for fully transesterifying into the fatty acid methyl ester. the main stages of enzyme transesterification are as follows: (a) triglyceride hydrolysis by hydrolytic enzymes is * corresponding author. e-m ail address: cp.lin@ asia.edu.tw tel.: +886-4-23323456; fax: +886-4-23321206 advances in technology innovation, vol. 3, no. 1, 2018, pp. 17 25 copyright © taeti 18 used for enhancing the yield of the transesterification reaction and removing impurities, glycerol, and free fatty acid [ 14]; and (b) immobilized enzyme reacts with the fatty acid , which reverts the fatty acids for converting to biodiesel [15]. we also developed the winterized characteristics of the waste oil’s biodiesel of aurantiochytrium sp. and its biodiesel, which included the reactivity parameters and properties [16-18]. differential scanning calorimetry (dsc) was used to obtain the winterization temperature and the other detailed enthalpy of the exothermicity of the biodiesel (b100) and the various propor tion of biodiesel for mixing with petrodiesel (b0). the parameters and reactivity propertie s could be applied to the designs of the operation, application, and transportation safety conditions [1, 16-17]. it is an important project for developing high quality available blend biodiesel from the waste oil of aurantiochytrium sp. an available blend of biodiesel has the appropriate heat of combustion, kinematic viscosity [16], and storage safety [18]. it is an important performance indicator of biodiesel for mixing with b0, which is concerned with effective fuel for diesel engines. here, the heat of combustion, kinematic viscosity, and storage safety of blend biodiesel were tested by dsc [16-17], flash point tester [18], viscometer [19-20], and oxygen bomb calorimeter [21-23], and these were also compared to the various proportions of b100 for mixing with b0. then, the best condition of mixtures was obtained as an available blend diesel of waste oil’s biodiesel of aurantiochytrium sp. 2. experimental and method 2.1. samples the waste oil of aurantiochytrium sp., which was supplied directly from vedan enterprise corp. in taiwan, was stored frozen at –20 °c. the original fatty acid profile of waste oil was conducted by gas chromatography (gc) analysis [ 24]. 2.2. transesterification first, the fixed fatty acid composition was saponified with the three eq uivalents of sodium hydroxide under ca. 90 °c for three hours. then, the impurities, the free fatty acids, glycerol, and microalgae cell wall were removed to form the waste oil soap. the oil soap was included in the further transesterification. we then obt ained a high purity fatty acid for the next step of the reduction reaction. second, the reduction reaction was mixed with sulphuric acid to form fatty acids under 85 °c for three ho urs. finally, the acid-catalysed reactions were mixed with dil. sulphuric acid (less than 0.02 mass% ratio of fatty acid) and methanol (fatty acids: methanol = 1:5, an amount of methanol more than five -times that of the fatty acids is better) under reflux overnight [1]. for preparation of the 0.1 m of disodium hydrogen phosphate-sodium dihydrogen phosphate buffer solution, the buffer solution was maintained at ph 7. candida rugose is a type of hydrolytic enzyme of lipase, which showed the preferred results of the hydrolysis reaction under a ph 7 buffer solution [8-15]. a 0.1 mass % of c. rugose was mixed with solid waste oil of aurantiochytrium sp. (triglycerides), and then the mixture was conducted for 260 rpm and stirring for 12 hours under isothermal 35 °c. from the above results, we removed the solid product and the oil layer of the upper layer (fatty acids), and then mixed this with c. antarctica (lipase acrylic resin of immobilized enzyme) (less than 2 mass% ratio of fatty acid) [15], methanol (fatty acids : methanol = 1:1), and isooctane (fatty acids: isooctane = 2:1) using a reflux apparatus under the isothermal conditions of 35 °c for 260 rpm and stirred for 16 hours . 2.3. gas chromatography (gc) analysis gas chromatography (gc) analysis fatty acid methyl esters (fames) were analysed by gas chromatograph agilent 6890n (agilent technologies, usa) equipped with a flame ionization detector (gc-fid) and a restek rt-2340 nb cap. column (105 m × advances in technology innovation, vol. 3, no. 1, 2018, pp. 17 25 copyright © taeti 19 0.25 mm × 0.20 µm). the oven temperature was initially set at 170 °c and then programmed to 250 °c using helium as the carrie r gas, at a flow rate of 1.1 ml/min and held for 30 min. the injector and detector temperatures were 250 °c. the split ratio was 1:80. fatty acids were identified by comparing the retention time of fame peaks with supelco 37 fame mixture standards (sigma) [1, 16, 24]. 2.4. differential scanning calorimetry (dsc) tests the temperature–programmed screening experiments were performed with dsc ta q20-rcs90 (ta instruments, usa). for the dsc analysis on the samples sealed in 20 μl aluminium pans, the lid was pressed onto the crucible using the pressure of a heavy mechanical force, and the seal tightened the crucible; the test cell was then sealed manually by a special tool equipped with ta’s dsc [1, 16]. approximately 1.7 mg to 2.5 mg of the sample was used for obtaining the experimental data. the b0 and b100 of the dynamic tests of the scanning rate selected for the programmed temperature ramp we re 2, 4, 6, and 8 °c/min and b2-b50. the scanning rate selected for comparing the programmed temperature ramp was 4 °c/min for the range of temperatures cooling down from 30 °c to -40 °c for each for each winterization phase behaviour experiment. in all of studies with the dsc thermal analysis, high purity nitrogen was the purge gas, and the flow rate was 50 ml/min. 2.5. diesel, biodiesel, and blend biodiesel winterization k inetic evaluation here, the proto-kinetic equation was applied to evaluate the win terization of the phase transfer for the exothermic reaction as follows [17]: 1 2 0 (1 ) ae n nrt ir k e      (1) where ea is the activation energy, k 0 is the pre-exponential factor, r is the ideal gas constant, α is the degree of conversion of a reaction or stage, and n1 and n2 are the reaction order of the phase transfer of the winterized exothermic reaction [17]. 2.6. biodiesel and blend biodiesel k inematic viscosity measurement the kinematic viscosity measurement is important for diesel engine fuel, and it is also an important indicator for biodiesel. to charge the sample into the viscometer, we placed the viscometer into the holder for fixing and inserted it into the consta nt temperature bath. the viscometer holder fit a cannon-fenske routine viscometer 75 u534 (cannon instrument company, usa) [16, 19]. we aligned the viscometer vertically in the bath by means of a small plumb bob in the tube. approximately 10 minutes was allowed for the sample to come to the bath temperature of 40 °c. we repeated this process three times for each sample to measure the efflux time and check the run. 2.7. biodiesel and blend biodiesel heat of combustion measurement the heat of combustion analysis of the samples involved a pa rr 1341 oxygen bomb calorimeter instrument (parr instrument company, usa) [16, 21-22]. we did a heat of combustion analysis for all of the samples as follows. to prepare, the water temperature was approximately 1.5 °c below room temperature, for which it was not necessary to use exactly 2 kg, but the amount selected must be duplicated within ± 0.5 g for each experiment. we opened the filling connection control valve slowly and watched the gauge as the bomb pressure rose to the desired filling pressure (usually 30 bar, but never more than 40 bar). then, we closed the control valve. instead of weighing the bucket, it was filled from an automatic pipette or from any other volume tric device if the repeatability of the filling system was within +/ – 0.5 ml and the water temperature was held within a 1 ºc range. we advances in technology innovation, vol. 3, no. 1, 2018, pp. 17 25 copyright © taeti 20 let the stirrer run for 5 minutes to reach equilibrium before starting a measured run [16, 21-22]. approximately 0.5 g of all of the samples was used for acquiring the experimental data. at the end of this period, we recorded the time on the timer of the parr 6775 digital thermometer and read the temperature. 2.8. flash point measurement the flash point was determined by hfp 360 pensky -marten flash point tester (walter herzog gmbh, germany), which met the requirements of the astm d93b standard [23, 25], and it was used to estimate the flash points of the b0, b100, and the various proportions of b100 for mixing with b0. the astm d93 test method was applied to measure the flash points for each sample [23, 25]. the tester amalgamated control devices programmed the instrument to heat the sample at a specific designated rate within a temperature range close to the expected flash point. the flash point was automatically tested by using an ignit er at specified temperature test intervals. 3. results and discussion 3.1. the results of transesterification from the sequences of the saponification, reduction reaction, and acid -catalysed reactions, which were conducted for the full process of transesterification, and excluding the impurities, the free fatty acids, and glycerol, we successfully obtained the biodiesel of waste oil of aurantiochytrium sp. [1, 16]. fig. 1(a) shows the fatty acid profile of waste oil from gc analysis, which verified the fatty acid composition of the waste oil and confirmed that the amount of palmitic acid and dha was more than 75 mass% of total fatty acids [1]. in addition, from fig. 1(b), the c14-c24 of the fatty acids were more than 92 mass% of the total fatty acid methyl ester profile by chemical transesterification, but via the enzyme transesterification, they were only 60 mass%. (a) fatty acid composition of waste oil (b) fatty acid methyl ester composition of waste oil fig. 1 fatty acid and fatty acid methyl ester composition of wa ste oil of aurantiochytrium sp fig. 2 shows that the dsc tests programmed temperature ramps were 2, 4, 6, and 8 °c/min for the range of temperatures cooling from 30 to -40 °c for each b0 and b100 experiment, respectively. moreover, figs . 2 (a) and 2 (b) s how the onset temperature of ca. -6 °c and 20 °c for the winterization of b0 and b100, respectively. b100 was the onset temperature (ca. 20 °c) of winterization slightly below room temperature, which would clog the oil lines and affect the diesel engine pe rformance at the low ambient temperature. advances in technology innovation, vol. 3, no. 1, 2018, pp. 17 25 copyright © taeti 21 (a) b0 for the range of temperatures cooling down from 30 to -40°c with scanning rates of 2, 4, 6, and 8 °c/min (b) b100 for the range of temperatures cooling down from 30 to -40°c with scanning rates of 2, 4, 6, and 8 °c/min fig. 2 dsc thermal curves of heat flow versus temperature 3.2. kinetic parameter of winterization table 1 comparisons of the b0 and b100 kinetic parameters for the evaluation with scanning rates of 2, 4, 6, and 8 °c/min kinetic parameter sample condition a 2 4 6 8 ln(k0)/ln(1/s) b b0 7.6886 7.0606 6.8166 8.0858 ea c 26.7237 23.8218 22.5479 24.1341 n1 d 1.1179 1.1126 1.0977 1.1352 n2 e 1.0108 1.2552 1.1630 2.7679 δη f 5.1302 7.6498 7.0392 11.0410 ln(k 0)/ln(1/s) b100 4.1213 4.5116 4.4545 4.7674 ea 14.8082 14.2822 14.2331 14.6001 n1 1.1501 1.1493 1.1126 1.1045 n2 2.0334 2.5039 2.0072 1.9112 δη 169.0000 179.6522 162.9779 168.3870 the b0 and b100’s kinetic parameters of winterization were evaluated as listed in table 1. from table 1, we compared the ea and of winterization, b0 and b100 ca. 24 kj/mol and 14 kj/mol, respectively. we found that the simulation method of applying the proto-kinetic equation could be appropriately used in the exothermic reaction for winterization of waste oil’s biodiesel of aurantiochytrium sp. moreover, from table 1, the results of the parameter evaluation for proto -kinetic simulation demonstrated that the model provided much more consistent results for b0 and b100. therefore, while analysing the b0 and b100 kinetic parameters of winterization, we obtained a better model by proto -kinetic simulation in this study. similar to the various proportions of b100 for mixing with b0 of b2-b50 blend biodiesel, table 2 and fig. 3 show that the dsc onset temperature, peak maximum temperature and enthalpy measured for the b2-b50 clearly discriminated the differences. fig . 3 shows that there was an onset temperature from ca. -6 °c to 6 °c for the winterization of b2-b50 by dsc analyses. an important characteristic of a good blend biodiesel is the ability to avoid the winterization phenomenon under the low ambient temperature. therefore, while at the ambient low temperature, a reliable blend biodiesel cannot be solidified in the pipeline and tank, wh ich could cause engine failure [1, 16]. in contrast to tables 1 and 2, we observed the results of the proto -kinetic simulation for b0–b100, in which the kinetic parameters provided clear and specific results. the ea values of b0-b100 were in the range of 23.82–14.28 kj/mol. moreover, table 2 shows that the result was explicit: b0-b100 of ea, along with the higher proportion of b100, became smaller. from the above results of the kinetic parameter evaluation of winterization, we obtained the highly proportio nal mixture of b100, which was easy for winterizing the blend biodiesel. advances in technology innovation, vol. 3, no. 1, 2018, pp. 17 25 copyright © taeti 22 table 2 results of dsc test and proto-kinetic equation simulation from b2 to b50 kinetic parameters for the evaluation at a scanning rate of 4 °c/min sample b2 b5 b8 b10 b20 b30 b50 mass a 2.0 2.0 2.0 2.0 2.0 1.9 2.0 exoto b -6.22 -4.85 -5.71 -5.34 -4.15 -1.03 6.09 exotp c -7.31 -14.54 -14.50 -12.66 -7.67 -1.78 5.70 exoδh d 10.40 13.50 19.00 22.70 34.20 46.90 81.10 ln(k 0)/ln(1/s) e 7.9899 6.9724 6.5301 6.3565 6.6144 6.3312 5.2386 ea f 26.8404 23.6787 23.0120 22.4650 21.3152 21.3428 18.7669 n1 g 1.0718 1.1073 1.0886 1.0935 1.1686 1.1335 1.1055 n2 h 0.6631 1.8153 0.9095 1.0190 1.6253 1.4865 2.2704 δη i 11.2995 18.4967 19.3829 23.1463 34.9629 47.6906 82.7576 fig. 3 dsc thermal curves of heat flow vers us temperature for the b2-b100 for the range of temperatures cooling down from 30 to -40 °c with a scanning rate of 4 °c/min 3.3. the results of k inematic viscosity measurement from fig. 4, we obtained the b0-b100 of the kinematic viscosity average valu es range from 3.14 mm 2 /s to 4.35 mm 2 /s. generally, the biodiesel would be a fuel for a diesel engine. the kinematic viscosity range followed the astm d6751 and en 14214 specification standards [23, 24], from 1.9 mm 2 /s to 6 mm 2 /s and from 3.5 mm 2 /s to 5.0 mm 2 /s, respectively. it was too thin for b2-b10 as a blend biodiesel for en 14214 specification standards, and it could not provide the appropriate lubrication of diesel engines. we also observed that b20-b100 (3.51-4.35 mm 2 /s), whose values of kinematic viscosity were included in a suitable range, was also concerned with the effective fuel for diesel engine, such as the winterization, the heat of combustion, the f lash point, and the storage conditions of blend biodiesel (see fig . 4). then, the best conditions of the mixtures were obtained as an available blend biodiesel in this study. 3.4. heat of combustion measurement we obtained the best condition of mixtures and the maximum proportion of biodiesel as a good blend biodiesel of waste oil’s biodiesel. in this study, for the heat of combustion of blend biodiesel, the basic value was b0 (43.98 kj/g), and this value was used as a comparison benchmark for other diesels. from fig. 4, we obtained the heat of combustion and showed that b100, b50, and b30 were les s than the others, which also showed that they were unsuitable for a blend biodiesel. from the comparisons of all samples heat of combustion, the mixed ratio of blend biodiesel smaller than b20 was acquired. this was the fuel that was suit able for use in compression ignition diesel engines in this study. meanwhile, from fig . 4, we observed the b2-b20 (43.30-43.15 mj/kg), whose values of heat of combustion were included in the suitable range, and this also showed that the highly mixed ratio of b lend biodiesel was b20 (43.15 mj/kg) in this study. advances in technology innovation, vol. 3, no. 1, 2018, pp. 17 25 copyright © taeti 23 biodiesel (mass%) 0 20 40 60 80 100 f la sh p o in t (o c ) 40 60 80 100 120 140 160 180 200 220 k in e m a ti c v is c o si ty ( m m 2 /s ) 2.8 3.0 3.2 3.4 3.6 3.8 4.0 4.2 4.4 4.6 h e a t o f c o m b u st io n ( m j/ k g ) 38 39 40 41 42 43 44 flash point flash point regr. r 2 = 0.948, y=1.26x+48.74 kinematic viscosity kinematic viscosity regr. r 2 = 0.968, y=0.01x+3.20 heat of combustion heat of combustion regr. r 2 = 0.975, y=-0.05x+43.75 fig. 4 heat of combustion, kinematic viscosity, and flash points of various proportions of b0 for mixing with b100 3.5. flash point measurement here, the results of the flash point tests of b0, b100, and the various proportions of b100 for mixing with b0 also followed the u.s. department of transportation (dot 49 cfr 173.120) regulations [18]. from fig. 4, we obtained the b0-b100 of the flash point temperature range from 45 °c to 184 °c, individually. from fig. 4, we also observed the b2-b100 (63-184 °c) of flash points were all above 60.5 °c. this temperature of flash points exceeded the safety condition that was related to a suitable fuel fo r diesel engine. this study also showed that adding the waste oil’s biodiesel of aurantiochytrium sp. could increase the storage and transport safety of diesel. the above transesterified results of fatty acid via gc analyses and dsc tests repeatedly corrobor ated the accuracy of the transesterification and the specific characteristic, proving that waste solid byproducts of aurantiochytrium sp. oil formed biodiesel. then, by conducting the dsc, bomb calorimeter, viscometer, and flash point tester, we compared the winterization, kinematic viscos ity, and heat of combustion of the various proportions of b100 for mixing with b0, which could establish the best conditions of the mixture b20 as an available blend biodiesel. 4. conclusions the waste oil of aurantiochytrium sp. can be developed for biodies el production and successfully made into a suitable blend biodiesel. we obtained the best conditions of mixtures and the highly mixed rate of b20 (ea of winterization 21.32 kj/mol at dsc scanning rate 4 °c/min; onset temperature of winterization -4.15 °c at dsc scanning rate 4 °c/min; heat of combustion 43.15 mj/kg; kinematic viscosity 3.51 mm 2 /s; flash point 67.5 °c) as an appropriate blend biodiesel from the waste oil’s biodiesel of aurantiochytrium sp. thus, from a green energy perspective, this project addressed the problem of “will not be occupied” arable land, ease of management, and ease of extensive cultivation for the production of biomass. we will enhance the transesterification technology for the waste oil of aurantiochytrium sp. to find improved energy efficiency, a less polluting method, and appropriate chemical and physical stabilities, which will be used in the mass production of green energy in futur e work. advances in technology innovation, vol. 3, no. 1, 2018, pp. 17 25 copyright © taeti 24 acknowledgements the authors are indebted to the donors of national science council (nsc), taiwan, r.o.c. under the contract no s.: most 105-2221-e-468-001& most 102-2221-e-468-001-my2 and the department of medical research, china medical university hospital, china medical university, taiwan, r.o.c. under the contract no.: dmr-100-165 for financial support. in addition, we are grateful to vedan enterprise corp. in taiwan for providing the microalgae oil and its waste oil. references [1] s. y. tsai, h. y. lin, g. y. lu, and c. p. lin, “solid byproducts of aurantiochytrium sp. oil made into the biodiesel oil made into the biodiesel,” journal of thermal analysis and calorimetry, vol. 120, no. 1, pp. 563-572, april 2015. [2] j. wang, l. cao, and s. han, “effect of polymeric cold flow improvers on flow properties of biodiesel from waste cooking oil,” fuel, vol. 117, pp. 876-881, january 2014. [3] a. hayyan, m. a. hashim, m. e. mirghani, m. hayyan, and i. m. alnashef, “esterification of sludge palm oil using trifluoromethane sulfonic acid for preparation of biodiesel fuel,” korean journal of chemical engineering, vol. 30, no. 6, pp. 1229-1234, june 2013. [4] s. k. hoekman, a. broch, c. robbins, e. ceniceros, and m. natarajan, “review of biodiesel composition, properties, and specifications,” renewable and sustainable energy reviews, vol. 16, no. 1, pp. 143-169, january 2012. [5] x. meng, j. yang, x. xu, l. zhang, q. nie, and m. xian, “biodiesel production from oleaginous microorganisms,” renewable energy, vol. 34, no. 1, pp. 1-5, january 2009. [6] j. v. gerpen, “biodiesel processing and production ,” fuel processing technology, vol. 86, no. 10, pp. 1097-1107, june 2005. [7] p. sivakumar, k. anbarasu, and s. renganathan, “bio-diesel production by alkali catalyzed transesterification of dairy waste scum,” fuel, vol. 90, no. 1, pp. 147-151, january 2011. [8] m. xiao, s. mathew, and j. p. obbard, “biodiesel fuel production via transesterification of oils using lipase biocatalyst ,” global change biology bioenergy, vol. 1, no. 2, pp. 115-125, april 2009. [9] n. sangaletti, m. cea, m. a. b. regitano-d’arce, t. m. f de souza vieirac, and r. navia, “enzymatic transesterification of soybean ethanolic miscella for biodiesel production ,” journal of chemical technology and biotechnology, vol. 88, no. 11, pp. 2098-2106, november 2013. [10] f. hasan, a. a. shah, and a. hameed, “industrial applications of microbial lipases ,” enzyme and microbial technology , vol. 39, no. 2, pp. 235-251, june 2006. [11] m. aarthya, p. saravananb, m. k. gowthamana, c. rosea, n. r. kaminia, “enzymatic transesterification for production of biodiesel using yeast lipases: an overview,” chemical engineering research and design, vol. 92, no. 8, pp. 1591-1601, august 2014. [12] r. abdulla, and p. ravindra “immobilized burkholderia cepacia lipase for biodiesel production from crude jatrophacurcas l. oil,” biomass and bioenergy, vol. 56, pp. 8-13, september 2013. [13] a. guldhe, b. singh, t. mutanda, k. permaul, and f. bux, “advances in synthesis of biodiesel via enzyme catalysis: novel and sustainable approaches ,” renewable and sustainable energy reviews , vol. 41, pp. 1447-1464, january 2015. [14] t. c. kuo, j. f. shaw, and g. c. lee, “conversion of crude jatropha curcas seed oil into biodiesel using liquid recombinant candida rugosa lipase isozymes ,” bioresource technology, vol. 192, pp. 54-59, september 2015. [15] x. fang, y. zhan, j. yang, and d. yu, “a concentration-dependent effect of methanol on candida antarctica lipase b in aqueous phase,” journal of molecular catalysis b: enzymatic, vol. 104, pp. 1-7, june 2014. [16] s. y. tsai, w. s. tseng, c. t. wu, and c. p. lin, “dwarf-castor oil made into a suitable biodiesel,” procedia engineering, vol. 84, pp. 940-947, 2014. [17] c. p. lin and j. m. tseng, “green technology for improving process manufacturing design and storage management of organic peroxide,” chemical engineering journal, vol. 180, pp. 284-292, january 2012. [18] g. b. wang, c. c. chen, h. j. liaw, y. j. tsai, “prediction of flash points of organosilicon compounds by struct ure group contribution approach,” industrial & engineering chemistry research , vol. 50, no. 22, pp 12790-12796, september 2011. [19] cannon-fenske opaque viscometers, cannon instrument company, 2014, www.cannoninstrument.com. [20] astm d6751, standard specification for biodiesel fuel blend sto ck (b100) for middle distillate fuels, astm international, west conshohocken, pa, 2015, www.astm.org . [21] parr bulletin 1341, plain jacket calorimeter, parr instrument co., moline, il, 1977. advances in technology innovation, vol. 3, no. 1, 2018, pp. 17 25 copyright © taeti 25 [22] a. m. halpern, g. c. mcbane, experimental physical chemistry. 3rd ed., new york: freeman, w. h., pp. 5.1-5.15, 2006. [23] a. s. gundogar and m. v. kok, “thermal characterization, combustion and kinetics of different origin crude oils ,” fuel, vol. 123, pp. 59-65, may 2014. [24] f. gasparini, j. r. de o. lima, y. a. ghani, r. r. hatanaka, r. sequinel, d. l. flumignan, and j. e. de oliveira, “en 14103 adjustments for biodiesel analysis from different raw materials, including animal tallow containing c17,” world renewable energy congress 2011, bioenergy technology, may, 2011, pp. 101-108. [25] g. knothe, “analyzing biodiesel: standards and other methods ,” journal of the american oil chemists ' society, vol. 83, no. 10, pp 823-833, october 2006.  advances in technology innovation, vol. 3, no. 4, 2018, pp. 195 205 a comparison of automotive waste heat recovery systems madhusudan raghavan 1,* , yong sheng he 2 1 general motors r&d center, pontiac, mi 48340, usa 2 general motors global propulsion systems, milford, mi 48380, usa received 05 june 2017; received in revised form 09 december 2017; accepted 05 january 2018 abstract the purpose of this research study is to assess schemes for recovering energy that is lost via an automobile’s exhaust. three systems for waste heat recovery, viz., (1) electric turbo-compounding, (2) organic rankine cycles, and (3) thermoelectrics are studied. this involves detailed computer simulations of the waste heat recovery systems, integrated within a vehicle propulsion system operating over certification driving cycles. the simulations suggest that fuel economy improvements of 3-5% could be achieved by converting waste heat into useful work. keywords: waste heat recovery systems, turbo-compounding, organic rankine cycle, thermoelectrics 1. introduction fig. 1 transportation sector energy use (mmbtu, million btus) per capita the international energy outlook [1] offers the following forecasts regarding population growth and energy consumption in the future. global population grows 28% from 6.9 billion in 2010 to 8.8 billion in 2040. population remains concentrated in asia and africa. their shares of the global population are 70% in 2010 and 72% in 2040. most other regions continue to grow through 2040. projected gdp rises by an average of 3.6% per year globally from 2010 to 2040, ranging from a high of 5.7% for china to a low of 0.7% for the middle east. total world marketed energy will grow by 56% from 2010 to 2040, from 524 quads to 820 quads (quad = quadrillion btu) with transportation sector projected usage shown in fig. 1. against this backdrop of increasing population, energy consumption, urbanization, congestion and pollution, we are looking at ways to improve vehicle fuel economy by minimizing automotive engine fuel consumption. in raghavan [2], we explored * corresponding author. e-mail address: madhu.raghavan@gm.com tel.: +1 248 930 5248 advances in technology innovation, vol. 3, no. 4, 2018, pp. 195 205 copyright © taeti 196 mechanical hybrid solutions to improving propulsion system efficiency. in the present work we have explored three thermal schemes for automotive waste heat recovery (a) electric turbo-compounding, (b) organic rankine cycles, and (c) thermoelectrics, in the context of vehicle simulations on certification drive cycles. the difference between this work and prior studies is that in each case we have sized the respective system for vehicle implementation and we evaluate the efficacy of each system over the various certification drive cycles with numerous transients, as opposed to running each system at its best steady-state operating point. our modeling tools are sufficiently detailed that we are able to base our architecture decisions on the simulation outcomes. we present the results of these studies and offer a comparative assessment of these 3 systems. 2. electric turbo-compounding electric-assisted turbocharging systems were developed previously for diesel engines as described in hopmann et al., [3], millo et al., [4], and balis et al., [5]. this kind of system consists of an electric machine integrated into the turbocharger shaft. the electric machine can work as a motor to improve transient response or work as a generator to recover energy. a turbo-compounding system with an extra downstream turbo-generator to recover waste energy was designed for diesel engines, see vuk [6]. different types of electric turbo-compounding systems (etc) were compared for turbocharged gasoline energy exhaust heat recovery as discussed by wei et al., [7]. the system that we investigated is shown in fig. 2. in order to increase the flexibility of the exhaust energy recovery and minimize the energy loss, the turbo-generator may be driven by a variable nozzle turbine (vnt). fig. 2 etc system with vnt at engine low load or high speed operating conditions, the throttle would be partly or fully closed to transfer the exhaust energy to the turbo-generator. at engine low speed and high load operating conditions, the vnt nozzle would be partly or fully closed to ensure enough exhaust energy flows through the turbocharger turbine. the performance of the etc system was studied on a 1.8-liter in-line four-cylinder turbocharged and intercooled four-stroke gasoline engine (compression ratio of 9:1, rated power of 125 kw at 5500 rpm and maximum torque of 235 nm at 2000~4000 rpm). the turbine of the turbo-generator was designed to meet the engine full-load power requirement, and the turbine geometry was optimized for better engine part-load efficiency, as discussed in zhuge et al., [8]. the optimized turbo-generator turbine parameters are as follows: inlet tip radius 23.5 mm, inlet hub radius 23.5 mm, inlet blade height 4.0 mm, inlet blade angle 0°, exit tip radius 19 mm, exit hub radius 7 mm, exit blade angle 70°, number of blades 12, volute radius 40.5 mm, volute area 245 mm 2 . the gasoline engine was modelled using the engine 1d simulation software tool, gt-power, including gas exchange, fuel combustion, heat transfer, and waste energy recovery by both the turbocharger for intake compression and the turbo-generator for electric power generation. the compressors and turbines were modeled with turbo through-flow models based on compressor two-zone models and the turbine mean-line flow models, instead of using compressor and turbine performance maps. this model-based method allowed engine system optimization by adjusting the design parameters of the compressor and advances in technology innovation, vol. 3, no. 4, 2018, pp. 195 205 copyright © taeti 197 turbine. this was critical to achieve the optimal performance of the etc system. a vehicle model of a typical passenger car was used to evaluate the etc system performance under us06 and ftp75 drive cycles in gt-drive. parameters of the vehicle were as follows: vehicle mass 3150 lb, vehicle frontal area 1.997 m 2 , vehicle drag coefficient 0.348, final drive gear ratio 3.55, final drive efficiency 0.97, tire rolling radius 289.7 mm, tire rolling resistance factor 0.0122. table 1 shows our key simulation results. the mechanical power output from the engine and the electric power output from the turbo-generator were both considered in evaluating the overall fuel efficiency of the engine with the etc system. fig. 3 shows electric power generated by the turbo-generator during the us06 driving cycle. the key takeaway from this exercise was that a turbo-compounding system sized for a mid-sized vehicle offered a 1.6% fuel economy improvement for the ftp75 city cycle, no improvement on the highway cycle and a 4.48% improvement on the us06 cycle. these fuel economy improvements were computed by placing the turbo-generator output on the 12v electrical bus so that this recovered energy could either be stored in a 12v (li-ion) battery or directly offset the electrical accessory loads by reducing the electrical generator load. table 1 drive cycle results for turbo-generator certification cycles us06 ftp75 average engine power output (kw) 15.1 4.76 average electric power output (kw) 1.05 0.297 improvement of fuel efficiency (%) 4.48 1.61 fig. 3 electric power generation [watts] on us06 3. organic rankine cycles a simplified schematic of a rankine cycle power generator is shown in fig. 4. for a vehicle application, the available heat energy from the hot exhaust gas of an internal combustion engine ( inq  ) could be utilized to vaporize the working fluid to drive a turbine for power generation. the organic rankine cycle (orc) uses an organic compound as the working fluid. this class of fluid, typically a refrigerant, requires much less input energy to boil and is thus potentially better suited to applications where the quality of the waste heat is variable, as in an automotive application. it also freezes at a much lower temperature, which is another advantage. kadota et al., [9], used engine coolant and exhaust heat along with a water-based rankine system incoporating an axial piston expander with a motor/generator. they claimed a system efficiency of 13%. ringler et al., [10], also used coolant and exhaust heat with a vane expander. the working fluid was water and ethanol and they claimed a 16% system efficiency. teng et al., [11], used the egr heat and exhaust heat as the heat source. their working fluid was ethanol and they used a turbine for expansion. they claimed an estimated efficiency of 14%. briggs et al., [12], used exhaust heat as the heat source on a r245a based system that operated in the supercritical region with an efficiency of 13.7%. advances in technology innovation, vol. 3, no. 4, 2018, pp. 195 205 copyright © taeti 198 fig. 4 rankine cycle schematic under steady-state operating conditions, a rankine cycle waste heat recovery (whr) system can achieve relatively high efficiency levels. given a fixed temperature differential and sufficient system warm-up time, the efficiency levels quoted are the highest that can be achieved. however, it is unlikely such levels are possible under transient vehicle operating conditions and typical usage profiles for light-duty vehicles. to obtain a more realistic assessment of in-vehicle operating efficiency, a dynamic model of the organic rankine cycle system was investigated at gm r&d using the amesim software environment. the model is shown schematically in fig. 5. the rankine system has 6 main components: an evaporator (or boiler), a turbine expander with a bypass, a condenser, a receiver and pump. the receiver is also commonly referred to as a surge tank. fig. 5 amesim organic rankine cycle model the amesim model is a full two-phase flow transient model that allows analysis of component sizing, refrigerant charge determination, and component behavior including heat exchanger (hex) effectiveness as a function of temperature, and heat exchanger wall and fluid temperatures. an example of the model output is shown in fig. 6, which represents system operation on the us06 cycle. the expander and condenser power are shown as negative values while the evaporator and pump power are positive. the expander peak power reaches 2 kw based on a heat input into the evaporator of approximately 15 kw. the heat advances in technology innovation, vol. 3, no. 4, 2018, pp. 195 205 copyright © taeti 199 input to the evaporator is extracted from the engine exhaust gas and transferred to the rankine cycle working fluid via a gas-liquid heat exchanger. we decided not to use the heat from the engine coolant to keep the mechanical circuitry simple. the intermittent nature and variability of the on-cycle exhaust gas heat is evident from the evaporator power input. fig. 6 modeled dynamic orc whr operation power fig. 7 location of orc system on vehicle fig. 8 exhaust gas heat loss schematic using actual vehicle experimental data, an exhaust gas model was implemented and tuned to provide a reasonable estimate for the exhaust gas temperature. the vehicle data was obtained from a buick lacrosse equipped with belt alternator starter (bas) mild hybrid system. fig. 7 shows the locations in the vehicle of the catalyst-out (red dot) and resonator-in (blue dot) temperatures used to tune the exhaust model. the location of the whr system evaporator is assumed to be downstream of the catalytic converter and resonator, in the approximate location of the blue dot. we anticipate the evaporator heat exchanger to be about the size of a shoe-box. advances in technology innovation, vol. 3, no. 4, 2018, pp. 195 205 copyright © taeti 200 the exhaust temperature model is based on the simple heat loss model shown in fig. 8. the exhaust gas loses heat to the exhaust pipe, which in turn loses heat to the environment. inputs are the exhaust flow rate and the fraction of the exhaust power reaching the whr system. taking a control volume around a section of pipe, the heat transfer is approximated by the equation: ( ) p gas exhaust pipe pipe gas pipe mc t q h a t t    (1) where m is the exhaust gas mass flow, cp is the exhaust gas specific heat, exhaust q is the exhaust gas heat power and hpipe is the convection coefficient for heat transfer between the gas and the pipe wall. there are two temperature states, the exhaust gas temperature tgas and the pipe wall temperature tpipe. the exhaust mass flow is estimated from the air fuel ratio and fuel rate reported by the engine subsystem from the vehicle model. using estimates of the heat power into the exhaust system derived from the test data, different exhaust temperature model tunings were developed for the city, highway and us06 cycles to match the test data trends of the catalyst-out temperature as much as possible. our choice of powertrains was influenced by the question of how best to utilize the power generated by a whr system. the power is generated mechanically but delivering it directly to a drive shaft introduces some complications which require redesign of the driveline components. converting the whr system’s mechanical output to electricity is more easily accomplished in principle (e.g. via a generator), and this energy may be readily used on the electrical system of a mild hybrid vehicle. the vehicle model that we used, had an 8kw evaporator, an 85% efficient isentropic expander, r600a as the working fluid and an 89% electric generator to convert expander shaft power into electricity on a 4000-lb test weight class (twc) suv vehicle. the whr system added 35 kg to the mass of the vehicle and this was accounted for in our modeling, as was the thermal inertia of the system. additionally the condenser was treated as integrated with the radiator in the vehicle, so no additional cooling capacity was added to support whr system operation. table 2 drive cycle results for orc whr certification cycles ftp city ftp highway us06 average evaporator temperature ( o c ) 157.8 178.0 175.5 average exhaust gas temperature ( o c ) 394.0 453.9 542.2 fuel economy improvement (%) 2.7 5.1 4.6 the resulting whr system efficiencies and fuel economy improvement over the baseline mild hybrid model are shown in table 2. since the operation of the whr system actually hinders engine warm up during cold start, a by-pass valve was used to avoid the whr system during start up. the integrated whr/vehicle model provides a more realistic assessment of the on-cycle fuel economy benefits of an orc whr system. in the mild-hybrid vehicle, the engine is on as long as the vehicle is in motion, providing longer engine-on time. the engine is also running at lower cycle average efficiencies, leading to more waste heat and higher exhaust temperatures. with higher exhaust temperatures, the average evaporator temperature increases, leading to higher carnot and actual system efficiency. the study also showed that the large estimated steady state efficiencies are difficult to achieve if the vehicle operation is highly transient. higher efficiencies may be obtainable in off-cycle conditions like long-distance steady-state high-speed cruising. 4. thermoelectrics a thermoelectric converter (tec) is shown schematically in fig. 9. the tec is constructed of p-type and n-type semiconductor thermoelectric (te) materials connected electrically in series and thermally in parallel via hot and cold junctions. the operating principle is the seebeck effect by which a temperature differential between the hot and cold junction causes current to flow across an electrical load connected, generally for convenience, to the cold junctions. the current flow advances in technology innovation, vol. 3, no. 4, 2018, pp. 195 205 copyright © taeti 201 results from the voltage potential difference across the te material induced by the temperature differential. the teg is a collection of tecs plus the heat exchangers and other power electronics required to make a working whr system. there have been numerous prior studies on waste heat recovery systems based on thermoelectric generators (tegs) for potential automotive applications. see for example, hussain et al., [13], lagrandeur et al., [14], and stobart et al., [15]. at gm r&d, we explored the use of teg electrical power derived from waste heat to try and achieve a 5% fuel economy gain on the us06 drive cycle. the amount of electrical power required suggested that a vehicle with some degree of powertrain electrification would be required to use the power generated to displace fuel. the 5% fuel economy improvement target required that the cycle average power output from the teg be 1.6 kw. as a potential target application we selected a mild hybridized full-size truck (fst), as the large engine offers good opportunities for waste heat harvesting. fig. 9 thermoelectric converter schematic in this study, a simulink vehicle model was used to simulate the behaviour of an integrated teg whr system to evaluate fuel economy improvement, in the presence of transient effects, such as unsteady heat input, system warm-up, and thermal inertia. for the hybrid vehicle, the power bus distributes power between electrical subsystems such as the high voltage energy storage system and other components on the bus, such as a teg. the teg model is connected to the vehicle power bus similar to other components such as the power steering, the cooling fan, the vacuum pump, the air-conditioner, and the dc/dc converter. the vehicle model provides instantaneous exhaust power, coolant temperature, and fuel flow rate to the teg model subsystems. the fuel flow rate is used to estimate the exhaust mass flow using the approximation that the mass flow rate is proportional to the air/fuel ratio (afr). the exhaust mass flow and the exhaust power are inputs to the exhaust gas temperature estimation model. the teg physics are modelled in the lumped-parameter teg subsystem model, which receives exhaust gas temperature at the teg inlet, the exhaust mass flow, and the coolant temperature as inputs. it outputs electrical power, which is optionally scaled by an efficiency factor (value of 0.9) to reflect losses in the electrical power transfer to the bus. the details of the above model are as follows. given an average teg power output requirement of 1.6 kw for the mild hybrid truck, and a 10% conversion efficiency teg , the average teg heat power input required is 16.5kw, from the following equation. teg outteg integ p p   (2) the teg system efficiency, teg , is a function of the tec module efficiency, the heat exchanger effectiveness and any power conditioning (pc) efficiencies: hextecpcteg   (3) advances in technology innovation, vol. 3, no. 4, 2018, pp. 195 205 copyright © taeti 202 the tec efficiency is calculated from the following equation, where ht is the hot junction temperature, ct is the cold junction temperature and the mean temperature t = 0.5*( ht + ct ). 1 1 ( )( ) 1 h c tec c h h t tzt t t zt t       (4) in the above, zt is the dimensionless thermoelectric figure of merit based on the intrinsic properties of the te materials. assuming a drive cycle average tc of 100 deg c and th of 400 deg c, a representative zt is approximately 0.9 for the te material being developed for gm r&d. thus, tec = 0.1743. assuming average power conditioning efficiency ( pc ) of 95% (dc/dc converters) and heat exchanger effectiveness of 0.6 (i.e., hex = 60%), the overall teg efficiency, teg , is 9.9% which has been rounded to 10%. thus, for the fst, the average teg input power required is 16.5 kw, assuming a 10% teg level conversion efficiency. the average engine exhaust heat power is approximately 27 kw, assuming 30% of the fuel energy is dissipated as exhaust heat. the teg model is designed to capture spatial temperature distribution effects in the teg. a tec array is represented by a single thermal capacitance node representing the hot sink for the array, again using the electrical analog for the heat transfer model, shown in fig. 10. heat transfer from the exhaust gas heat exchanger to the hot sink is represented by a thermal resistance r_ex_hs. the heat transfer into the tec and out to the coolant is represented by r_hs_cool + rte. since each array is thermally connected to adjacent arrays, the heat transfer between adjacent arrays, j, and k, is modeled by thermal resistance r_hs_jtok. the temperature at the hot sink, t_hs_j, is determined by the state equation (eq. (5)). the resistance data is given in table 3. ( _ _ _ _ ) ( _ _ _ _ ) ( _ _ _ _ ) ( _ _ _ _ ) _ _ _ _ _ _ _ _ dq t ex j t hs j t hs i t hs j t hs j t hs k t hs j t cool j dt r ex hs r hs jtok r hs jtok r hs cool         (5a) dt jhsdt hsmc dt dq p __ *__ (5b) where cp_m_hs is the specific heat of the tec material. fig. 10 resistor diagram model for tec table 3 tec heat transfer model electric analog parameters thermal resistance parameter value hsexr __ 0.04123 k/w ter 0.25 k/w coolhsr __ 0.0041 k/w jtokhsr __ 0.047 k/w advances in technology innovation, vol. 3, no. 4, 2018, pp. 195 205 copyright © taeti 203 hsmc p __ 1021 j/k fig. 11 teg2 schematic of physical arrangement the teg is made up of 7 arrays as shown in fig. 11. the power output section incorporates experimental data from the skutterudite material made on the kg scale for the gm r&d project. curve fits for these data produced the following equations for the seebeck coefficient and resistivity: n-type material 77210 5601.33690.53437.3)(   etetetn (6a) 78212 2490.76386.19721.6)(re   etetetln p-type material 57210 2320.31472.64724.4)(   etetetp (6b) 68211 4659.22302.23142.1)(re   etetetl p the tecs’ open circuit voltage is defined by eq. (7),   ht ct n ht ct poc dttndttnv )()(  (7) where n is the number of te leg pairs in series, np, is the temperature dependence of the seebeck coefficient, and tc and th are the cold and hot junction temperatures, respectively, such that th tc = junct . the total electrical resistance of the te module is calculated from eq. (8),             2 )(re)(re 2 )(re)(re mod cnhncphp tltltltl a nl r (8) where l is the te module leg length and a is the leg’s cross sectional area. for the simulation, n = 32, l = 0.004m and a = 0.000016m 2 . in order to calculate voc, we integrate  tp and  tn from eqs. (6a) and (6b): ht ct ht ct oc tet e t e tet e t e v                       72 7 3 10 52 7 3 10 5601.3 2 3690.5 3 3437.3 2320.3 2 1472.6 3 4724.4 (9 ) the total power per tec array is then computed by combining the powers of the various individual modules. in the vehicle model the power generated by the teg appears as surplus power which can be used as needed. the model contains an exhaust bypass control to prevent the peak hot junction temperature from exceeding a preset limit by diverting flow from going into the teg. the exhaust backpressure model also causes exhaust to bypass the teg if the predicted back-pressure exceeds 5 kpa. we simulated several drive cycles using the above model. results from the us06 cycle are advances in technology innovation, vol. 3, no. 4, 2018, pp. 195 205 copyright © taeti 204 shown in fig. 12. key points to note are as follows. the modelled exhaust gas temperature at the teg inlet is the blue curve, the exhaust gas heat power into the teg is the red curve, and the corresponding teg output power is in green. the orange curve shows the teg output power assuming a constant exhaust gas temperature of 800 o c at the teg inlet. this is an upper bound on the teg output power. the teg hot side and cold side temperatures are shown respectively by the purple (te_hot) and teal (te_cold) curves. from the green (actual) and orange (max output) lines, we observe that the maximum teg output capability is reached between 270 sec and 420 sec on the high-speed portion of the us06 drive cycle. the actual fe gains over the us06 cycle are shown in table 4. the fe improvements were close to zero for the ftp city and highway cycles. the us06 result showed a fuel economy improvement of approximately 2% but fell considerably short of the 5% goal. fig. 12 mild hybrid fst with teg whr system table 4 us06 drive cycle results for teg engine output (kj) fe gain (%) no teg (basesline) 15756 0 teg 15441 2.0 5. conclusions in this study, three different waste heat recovery systems (electric turbo-compounding, organic rankine cycle, and thermoelectrics) were explored and their potential to improve vehicle fuel economy were evaluated. (1) the electric turbo-compounding system with a turbo-generator driven by a vnt, when applied to a mid-size 3150-lb vehicle, showed fuel economy gains of 1.6% on the ftp city cycle, 0% on the ftp highway cycle and 4.5% on the us06 cycle, respectively. (2) the organic rankine cycle system, with r600a as the working fluid, improved the fuel economy of a mild hybrid 4000-lb suv vehicle by 2.7% on the ftp city cycle, 5.1% on the ftp highway cycle, and 4.6% on the us06 cycle. (3) the thermoelectric generator system, when applied to a mild hybrid 6500-lb full-size truck, provided little or no fuel economy improvement on the ftp city and highway cycles, and a 2% fuel economy improvement on the us06 cycle. while we do not claim that these trends will be the same for all applications (heavy duty trucks may show a different trend), we are reasonably confident (based on the fidelity of our sub-system models) that the above numbers are representative and accurate for light duty vehicle applications. integrated whr/vehicle system models are essential to providing realistic assessment of the on-cycle fuel economy benefits of the various whr systems. specific introductions of these systems in the light duty vehicle market will be determined by the cost-benefit ratios of the respective technologies. advances in technology innovation, vol. 3, no. 4, 2018, pp. 195 205 copyright © taeti 205 references [1] “u. s. energy informationadministration(eia),” http://www.eia.gov/forecasts/archive/ieo13/pdf/0484(2013).pdf, 2013. [2] m. raghavan, “propulsion architectures using mechanical energy storage,” proc. the institution of mechanical engineers, part k: journal of multi-body dynamics, vol. 230, no. 3, pp. 242-250, 2016. [3] u. hopmann and m. c. algrain, “diesel engine electric e-turbo compound technology,” sae technical paper, 2003. [4] f. millo, f. mallamo, e. pautasso, and g. ganio mego, “the potential of electric exhaust gas turbocharging for hd diesel engines,” politecnico di torino, sae technical paper, 2006. [5] c. balis, c. middlemass, and s. m. shahed, “design & development of e-turbo for suv and light truck applications,” conf. deer 2003: garrett engine boosting systems, june 23, 2003. [6] c. t. vuk, “electric turbo-compounding a technology whose time has come,” proc. deer, diesel engine emissions reduction conference, 2003, august 21-25. [7] w. wei, w. zhuge, y. zhang, and y. he, “comparative study on electric turbo-compounding systems for gasoline engine exhaust energy recovery,” university of amsterdam, asme paper gt2010-23204, 2010. [8] w. zhuge, l. huang, w. wei, y. zhang, and y. he, “optimization of an electric turbo-compounding system for gasoline engine exhaust energy recovery,” the university of hong kong, department of mechanical engineering, sae technical paper, 2011. [9] m. kadota and k. yamamoto, “advanced transient simulation on hybrid vehicle using rankine cycle system,” sae international journal of engines, vol. 1, no. 1, pp. 240-247, 2008 [10] j. ringler, m. seifert, v. guyotot, and w. hubner, “rankine cycle for waste heat recovery of ic engines,” sae international journal of engines , vol. 2, no. 1, pp. 67-76, 2009 [11] h. teng, j. klaver, t. park, g. l. hunter, and b. van der velder, “a rankine cycle system for recovering waste heat from hd diesel engines whr system development,” sae technical paper, 2011 [12] t. e. briggs, r. wagner, k. d. edwards, s. curran, and e. nafziger, “a waste heat recovery system for light duty diesel engines,” oak ridge national laboratory, sae technical paper, 2010. [13] q. hussain, d. brigham, and j. maranville, “thermoelectric exhaust heat recovery for hybrid vehicles,” sae international journal of engines, vol. 2, no. 1, pp. 1132-1142, 2009. [14] j. lagrandeur, d. crane, and a. eder, “vehicle fuel economy improvement through thermoelectric waste heat recovery,” 2005 diesel engine-efficiency and emissions research (deer) conference, chicago, il, august 2008. [15] r. stobart, g. dong, j. li, and m. a. wijewardane, “performance analysis of tegs applied in the egr path of a heavy duty engine for a transient drive cycle,” gt suite conference, october 2011. https://www.eia.gov/ http://www.eia.gov/forecasts/archive/ieo13/pdf/0484(2013).pdf https://saemobilus.sae.org/search/?op=doauthsearch&auth=millo%2c%20f.&subscribedonly=true https://saemobilus.sae.org/search/?op=doauthsearch&auth=nafziger%2c%20eric&subscribedonly=true  advances in technology innovation, vol. 5, no. 2, 2020, pp. 126-134 computer science and information management, soochow university, taipei, taiwan received 17 march 2019; received in revised form 29 august 2019; accepted 02 october 2019 doi: https://doi.org/10.46604/aiti.2020.819 abstract computer aided geometric design employs mathematical and computational methods for describing geometric objects, such as curves, areas in two dimensions (2d) and surfaces, and solids in 3d. an area can be represented using its boundary curves, and a solid can be represented using its boundary surfaces with intersection curves among these boundary surfaces. in addition, other methods, such as the medial-axis transform, can also be used to represent an area. although most researchers have presented algorithms that find the medial-axis transform from an area, a algorithm using the contrasting approach is proposed; i.e., it finds an area using a medial-axis transform. the medial-axis transform is constructed using discrete points on a curve and referred to as the skeleton of the area. subsequently, using the aforementioned discrete points, medial-axis circles are generated and referred to as the muscles of the area. finally, these medial-axis circles are blended and referred to as the blended boundary curves skin of the area; consequently, the boundary of the area generated is smooth. keywords: medial-axis transform, hermite curve, bezier curve, computer-aided geometric design 1. introduction in the domain of computer-aided geometric design, an area can be represented using various methods, such as via its boundary curves, via its medial-axis transform [1], or via set operations, such as union, intersection, and difference, of numerous primitive geometric objects. to describe a curve or area in two dimensions (2d), researchers attempted to use different curve representations in different applications. for example, cinque, levialdi, and malizia [2] used a cubic bezier curve to describe the shape via its boundary curve with orientation. yang, lu, and lee [3] used a bezier curve to describe the shape of chinese calligraphy characters [4]. chang and yan [5] derived an algorithm to simulate hand-drawn images using a cubic bezier curve. cao and kot [6] derived an algorithm to embed data using electronic inks without the loss of data. furthermore, algorithms were proposed to derive the interior/exterior offset curves and medial axis of an area using the boundary curves [7]. in 3d, surface representations are used to simulate natural objects, such as flowers [8]. in addition, the circle packing problems, i.e., the arrangement of circles with same or different radii inside a specified area, were also investigated by researchers [9-11]. in this study, methods are proposed to grow an area. an area is initialized using two approaches, one with the use of medial-axis circles, and the other via points on a curve. subsequently, the algorithms are proposed to grow the areas using the initialized area. given a cubic bezier curve with n discrete points on it, n circles are attempted to find centered at these n points with different radii, such that the neighboring circles are tangent to each other. this is the initial area that is constructed from a curve that has points on it. next, the area along different directions is grown; the first direction is along the two end points of the curve, as depicted in fig. 1(a); the second direction is along the two sides of the curve, as depicted in fig. 1(b). a single circle is also considered to be an initial area, in this case; the area then grows along different directions, as depicted in fig. 1(c). * corresponding author. e-mail address: chiang@gm.scu.edu.tw tel.: +886-2-23111531 ext. 3801; fax: +886-2-23756878 area-construction algorithms using tangent circles ching-shoei chiang * , hung-chieh li advances in technology innovation, vol. 5, no. 2, 2020, pp. 126-134 127 the initial area generated by the aforementioned curve is referred to as the skeleton of initial area, this curve is referred to as the skeleton of the initial area and these circles as the muscle of the initial area. the union of all the muscles is referred to as the central muscle of the area. upon extending the area along different directions, the circular arc connecting two tangent circles needs to be blended such that the boundary of the extended area is smooth. the entire process, from skeleton to muscles, and, subsequently, from muscles to skins, leads to the generation of the desired area. the generation of the skeleton and skin is simple. therefore, the extension of the muscles is mainly emphasized. in the first section, the problem is illustrated. in the second section, an algorithm is proposed to generate the central muscles, and in the third section, the extending of the muscles is introduced from one side of the skeleton, or from every direction of a circle muscle. the conclusions are presented in the last section. (a) along two end points of the central muscle (b) along one side of the central muscle (c) along multiple directions of a circle fig. 1 muscle growth 2. generating central muscles consider the following problem. given n distinct ordered points pi, where i = 0, …, n, a set of circles has to be found, ci, whose circles are centered at these points pi, and each circle is tangent to its neighboring circles. in this problem, there are n+1 variables (ri, where i = 0, …, n) with only n constraint equations (li = dist(pi, pi+1), i = 0, …, n–1). therefore, one more constraint equation needs solving the aforementioned problem. the following algorithm can be obtained by inputting the value r0: algorithm 1: 1. input the distinct ordered points, i.e., p0, p1, …, pn. 2. input r0 3. for i in 0, 1, …, n–1: 4. calculate ri+1 = li – ri, where 0 < r0 < l0 although the neighboring circles are tangents to each other, two neighbors of the same circle cannot be ensured that they would not intersect each other. therefore, to avoid the aforementioned intersection, constraints are placed on the radius of each circle using the following theorems: theorem 1: given pi, where i = 0, ..., n, if ci-1 did not intersect ci+1, where i = 1,.., n–1, then ri > (li-1+li-dist(pi+1,pi-1))/2. (see fig. 2) proof: the result is derived directly from the following equation: dist (pi+1, pi-1)> ri-1+ri+1, where ri-1=li-1–ri and ri+1=li–ri. advances in technology innovation, vol. 5, no. 2, 2020, pp. 126-134 128 theorem 1 is used to calculate the lower bound of the radius of circle centered at pi, and these radii are denoted by mri, where i = 1,…, n–1. notably, li-1+li-dist (pi+1,pi-1)>0 by triangle inequality. consider the case wherein two circles ci = c (pi, mri) and ci+1 = c(pi+1, mri+1) , where i = 1, …, n–2 intersect each other; therefore, it is impossible to find a set of circles that are tangents to each other, i.e., the circles do not intersect each other. therefore, the following example is provided: fig. 2 intersection of two neighboring circles (a) invalid input (b) minimal radius (c) cj-1 is tangent to cj+1 (d) better result fig. 3 circles with different radii example 1: given four points, namely, p0 = (0, 0), p1 = (2, 5), p2 = (4, –5), and p3 = (8, 2). the minimal radius can be calculated, called mri, for p1, p2, and p3 as mr1 = 4.59, mr2 = 5.78, and mr1 + mr2 > l1 = 10.198, respectively (see fig. 3(a)). in this case, irrespective of the values are assigned to r1 and r2, there is always the case wherein c0 intersects c2 or c1 intersects c3. example 2: using the input p0 = (0, 0), p1 = (2, 1), p2 = (4, –1), p3 = (6, 1), and p4 = (8, 0), mr1, mr2, and mr3 are calculated as 0.47, 0.82, and 0.47, respectively, and the circles centred at pi with radius mri are depicted in fig. 3(b). assuming j to be the index of the maximum value of mri, rj = mrj can be set. consequently, other radii are calculated starting from rj, and rj+1, rj+2, …, rn and rj-1, rj-1, …, r0 are also calculated using an approach similar to that used in algorithm 1. accordingly, it yields the following algorithm: algorithm 2: 1. input the distinct ordered points, i.e., p0, p1, …, pn. check whether the input is appropriate for the algorithm. if the input is inappropriate, a new set of points is requested for as the input data. 2. calculate mri, where i = 1, ..., n–1. compute index j for max{mri, where i = 1, …, n–1}. set rj = mrj. 3. from rj, repeat rk+1 = lk – rk for k = j, …, n–1. 4. from rj, repeat rk-1 = lk-1 – rk for k = j, …, 1. in this case, the circles are found with pj as their centers and mrj their radii, where mrj + mrj+1 < lj. a set of radius rk is found, where k=1,2, …, j-1, j+1, …n, such that the circles will not intersect locally (ck-1 and ck are tangent to each other, and cj and ck+1 are also tangent to each other; in addition, ck-1 and ck+1 may be tangent to each other but cannot intersect otherwise). for the case rj = mrj, circle cj-1 will be tangent to cj+1, as depicted in fig. 3(c). advances in technology innovation, vol. 5, no. 2, 2020, pp. 126-134 129 example 3: using the same input as that used in example 2, r2 = mr2 = 0.82 was fixed and r3 and r4 was calculated to be 2 and 0.24, respectively. in addition, r1 and r0 are equal to 2 and 0.24, respectively, as depicted in fig. 3(c). the radius of cj is increased to avoid c j-1 and cj+1 being tangents to one another, as depicted in fig. 3(c). a “better” chain of circles must be produced, such that all circles ci-1 and ci+1 on the chain are separated locally. the “better” chain of circles may be the case wherein the radii of circles are within a small range. theoretically, to minimize the value of max {ri, i=0, n} – min {ri, i=0, n}. using the algorithm, all mri are calculated, where i = 1, …, n–1, and mci, where mci are the circles centered at pi with radius mri. if mci-1 did not intersect mci+1, where i = 1, …, n–1, a set of ri is found, such that ci is tangent to ci+1 with the property that ci-1 did not intersect ci+1. moreover, rj has both a lower bound, i.e., mrj, and an upper bound, which could be min{dist(pj, pj+1)-rj+1, dist(pj, pj-1)-rj-1}; notably, upon increasing rj, circles mcj-1 and mcj+1 will not intersect each other. the radius was discretized from its lower bound to upper bound and tried each case and, subsequently, produced the “better” chain of circles satisfying the case wherein max{ri, i=0, n} – min{ri, i=0, n} was minimum. therefore, there is the following algorithm can be found: algorithm 3: 1. input the distinct ordered points, i.e., p0, p1, …, pn. check whether the input is appropriate for the algorithm. if the input is not appropriate, ask for a new set of points as the input. 2. calculate mri, where i = 1,.., n–1. compute index j for max{mri, where i=1,.., n–1}. set upper = min{li–mri+1, li-1–mri-1} 3. set m such that it divides r from its lower bound to upper bound into m+1 values. 4. for rj from mrj to its upper bound with step (upper – mrj)/m 4.1 from rj, repeat rk+1 = lk – rk for k = j, …, n–1. 4.2 repeat rk-1 = lk-1 – rk for k = j, …, 1. 4.3 calculate max {rj} – min {rj} for j = 1, …, n–1. memorize k when max {rk} – min {rk} is minimum. example 4: using the same input as that used in example 2 to have the lower and upper bounds for r2 (0.82, 2.36). this range linearly was discretized into 11 different radii and observed that when r2 =1.44, a “better” chain of circles was produced. the result is depicted in fig. 3(d). a sequence of points, p = [p0, p1, …, pn] is used to produce the associated radii for point pi using the above algorithm. p is called the skeleton/bone of the area. the produced circles are referred to as the muscles of the area. if the path that connects these ordered points pi has a sharp angle, it is highly probable that the circles generated using algorithm 3 intersect other circles locally. to avoid the aforementioned situation, skeleton must be constructed using points on bezier or hermite curve. consider the case wherein points lie on a bezier curve b(t), where t is equal to 0, 0.25, 0.5, 0.75, and 1.0. the points are used on the curve as the input for algorithm 3. example 5: consider the bezier curve whose control polygon is denoted by [p0, p1, p2, p3], where p0 = [0, 0], p1 = [2, 6], p2 = [7, –1], and p3 = [9, 3]. the control polygon and bezier curve are depicted using gray line and red curve, respectively, in fig. 4. using algorithm 3, the muscles are calculated to be c(0, 0, 1.969), c(1.969, 2.438, 1.164), c(4.5, 2.25, 1.374), c(7.03, 1.688, 1.219), and c(9.0, 3.0, 1.147), as depicted using green circles in fig. 4. the case is selected wherein the difference between the maximum radius and minimum radius is minimum. notably, the polygon that connects points b(0), b(0.25), b(0.5), b(0.75), and b(1.0) did not have sharp angles, and the values of mri, where i =1, 2, and 3, are considerably small, i.e., 0.32, 0.007, and 0.199, respectively. mc1 and mc3 are depicted as left and right red circles in fig. 4. advances in technology innovation, vol. 5, no. 2, 2020, pp. 126-134 130 fig. 4 skeleton on bezier curve 3. generating external muscles after generating the central muscles the generation of external muscles is launched. for this, continuous circles that are tangents to one side of the central muscle must be produce, as depicted in fig. 1(b). consider 3 circles c0, c1, and c2, as depicted in fig. 5. a sequence of circles wants finding, such that the following conditions are satisfied: the first circle c0 is tangent to c0 and c1; the last circle cn is tangent to c1 and c2; circle ci between c0 and cn is tangent to c1 and its neighbors ci-1 and ci+1, as depicted in fig. 5(a). fig. 5(b) and fig. 5(c) depicts the cases for n = 0 and n = 1. algorithms are developed for this case in this section. two subcases are considered, the first case wherein circles ci, where i = 0, …, n, have the same radius, and the second case wherein the radii of these circles have the arithmetic progression property. the first case is started from considering the case wherein a circle is tangent to 3 circles, as depicted in fig. 5(a); this case comprises a well-known problem called apollonius’s problem, which has 8 solutions. on the basis of the orientation of circles (a circle oriented clockwise has positive radius, and that oriented counter-clockwise has negative radius), the problem can be simplified by solving a quadratic equation. after solving the equation, two circles are founded, one contains all the three circles, and the other contains none of the three circles. the wanted circle, in this case, is the one that contains none of the three circles. (a) n+1 circles (b) one circle (c) two circles fig. 5 construction of circles along one side for finding a circle that is tangent to the three circles, the problem can be solved algebraically. assume that the three circles have (xi, yi) as their centers and ri as their radii, where i = 1, 2, and 3. the circle centered at (x, y) with radius r is tangent to these three circles. therefore, there are three equations as (x – xi) 2 + (y – yi) 2 = (r – ri) 2 , where i = 0, 1, and 2, with three variables x, y, and r. because each of the aforementioned three equations are of degree two, they should have eight solutions in total. these three equations are referred to as eq0, eq1, and eq2. notably, by performing eq0 – eq1, the terms with degree two are eliminated, and, accordingly, the equation obtained is a linear equation. the system of equation can be simplified {eqn0, eqn1, eqn2} into {eqn0, eqn0 – eqn1, eqn0 – eqn2}, where the new system of equation is of degrees two, one, and one, respectively; therefore, the number of solutions becomes two. in the algorithm, the binary approach is used to find the radius of the tangent circle. this algorithm is introduced because it can be easily extended to the case with more circles, as depicted in fig. 5(b) and 5(c). before the algorithm, there are the following theorems. theorem 2: three circles are tangent to each other, and their radii are x, y, and z. furthermore, the angle  of the triangle that is formed by connecting their centers has the opposite side of length y + z, as depicted in fig. 6(a). then: advances in technology innovation, vol. 5, no. 2, 2020, pp. 126-134 131 2 2 cos( ) x xy xz yz a x xy xz yz        (1) proof: the result is derived directly from the cosine theorem, that is, (y + z) 2 = (x + y) 2 + (x + z) 2 – 2(x + y)(x + z)cos(). using the equation in theorem 2, there are four variables, namely, x, y, z, and , with one equation. given any three variables, the fourth one can be calculated. assume that there are two circles with radii x and y, and angle . accordingly, radius z can be calculated. besides, if x and y are known, and assumed radius z,  can be calculated as well. (a) one circle tangent to two circles (b) one circle tangent to three circles fig. 6 relation between radii and angles now, consider 3 circles ci, where i = 0, 1, and 2. assume that c0 is tangent to c1 and that c1 is tangent to c2. in addition, assume that the circle that is tangent to c0, c1, and c2 has radius r. from this assumption and considering fig. 6(b), angles 1 and 2 can be calculated using the equation in theorem 2, and, subsequently, their sum is compared with the angle between vectors p0 – p1 and p2 – p1. if the sum is less than the angle between the vectors, radius r enlarged, and if the sum is greater than the angle between the vectors, radius r reduced, until the solution is found within a tolerance. this aforementioned idea can be implemented using the following algorithm: algorithm 4: 1. input three circles c0, c1, and c2. #assume that c0 is tangent to c1 and that c1 is tangent to c2. 2. calculate the angle between vectors p0-p1 and p2-p1, and refer to it as . 3. set r, rmin, and rmax as 1, 0, and 999, respectively. 4. set eps = 0.001 5. calculate angles 1 and 2 using the assumed radius r. 6. while |1 + 2 – | > eps: if 1 + 2 > : rmax = r r = (rmax + rmin)/2 else: rmin = r r = (rmax + rmin)/2 7. display the result. advances in technology innovation, vol. 5, no. 2, 2020, pp. 126-134 132 example 5: consider three circles c0 (–4, 3, 3), c1 (0, 0, 2), and c2 (4, 0, 2). notably, c0 is tangent to c1 and c1 is tangent to c2. circle c0 (2, 6.60, 4.90) is found, as depicted in fig. 7(a). now consider the case wherein n+1 circles want adding. in this case, the first circle c0 is tangent to c0 and c1; the last circle cn is tangent to c1 and c2; circle ci between c0 and cn is tangent to both c1 and its neighbors ci-1 and ci+1, as depicted in fig. 5(c). a similar idea as that for algorithm 4 can be applied. in algorithm 4, the sum of two angles, 1 and 2, are compared with angle , and now the sum of n + 2 angles is compared with angle . it is referred to as the extended algorithm 4. example 6: using the same input as that used in example 5, 2 circles that are tangent to each other want finding, where the first circle is tangent to c0 and c1 and the second circle is tangent to c1 and c2. using the extended algorithm 4, these two circles are found centered at (0.047, 3.045) and (2, 2, 97) each of radius r = 1.046, as depicted in fig. 7(b). example 7: given three circles c0 (–4, –3, 3), c1 (0, 0, 2), and c2 (4, 0, 2). using the extended algorithm 4, 5 circles are found between c0 and c2, such that these 5 circles are tangent to c1. the circles are found centered at (–2.697, 0.513), (– 2.031, 1.846), (–0.767, 2.636), (0.723, 2.648), and (2.0, 1.88) each of radius 0.745, as depicted in fig. 7(c). (a) one circle (b) two circles (c) five circles fig. 7 tangent circles with similar radii (a) 2 circles with dr = 0.5 (b) 5 circles with dr = 0.2 (c) 5 circles with dr = 0.25 fig. 8 tangent circles with different radii let people consider the case wherein the radii of the circles along one side of initial area are in an arithmetic progression. assume that dr is the common difference. accordingly, the first circle has radius r; the second circle has radius r – dr; the third circle has radius r – 2*dr, …, and so on. using the binary approach as that employed in algorithm 4, to find 2 or 5 tangent circles, examples 8 and 9 are referred to as depicted in fig. 8(a) and (b), respectively. example 8: using the same input data as those used in example 5 and setting dr = 0.5, two circles are found: c0 (0.31, 3.31, 1.32) and c1 (2.0, 1.99, 0.82), as depicted in fig. 8(a). example 9: using the same input data as those used in example 7 and setting dr = 0.2, 5 circles are found: c0 (–3.03, 1.09, 1.22), c1 (–1.43, 2.66, 1.02), c2 (0.41, 2.79, 0.82), c3 (1.64, 2.04, 0.62), and c4 (2.14, 1.13, 0.42), as depicted in fig. 8(b). upon increasing dr, the difference between the radii of c0 and c4 is also increased. the difference can be seen in radii by comparing fig. 8(c), where dr = 0.25, with fig. 8(b), where dr = 0.2. now, consider the case wherein the external muscle/circle grows along every direction of a muscle/circle, as depicted in figs. 1(c) and fig. 9. in this case, a list of angles is needed; the list indicates the direction along which new circles want advances in technology innovation, vol. 5, no. 2, 2020, pp. 126-134 133 extending. the extension might have no solution. consider a unit circle centered at the origin, with the list of angles, alist, where the first element in alist is 0. other cases can be converted into this situation by translating, rotating, and scaling the circles in the problem. with this initial condition, it yields the following algorithm: fig. 9 growth direction of circles algorithm 5: (eps is a considerably small number) 1. input c(0,0,1) and a list of angles, alist. the length of alist denotes len(alist) 2. assume the initial radius of the first circle along the x+ axis, and refer to it as r0. 3. calculate r1 using the equation in theorem 2 with the angle data provided by alist. continue this process until all ri, where i = 1, 2, …, len(alist) – 1 have been calculated. 4. calculate the sum of the angles between the angle generated by connecting the center of three circles as c0cc1, c1cc2, …, cncc0, and refer to it as rsum. 5. while abs(rsum – 2)>eps: if rsum > 2: reduce the assumed radius r0 else: increase the assumed radius r0. example 10: assuming the input as alist=[0, 90, 180, 270], the four circles obtained are c(4, 0, 3), c(0,3,2), c(–4, 0,3), and c(0, –3, 2), as depicted in fig. 10(a). notably, the radii of these circles are not the same, although their directions of extension are east, north, west, and south, respectively. there are more than one solution, and the algorithm finds an appropriate one. (a) four directions (b) five directions fig. 10 grow external muscle example 11: assuming the input as alist=[0, 72, 144, 216, 288], the five circles obtained are c(2.5, 0, 1.5), c(0.73, 2.42, 1.36), c(–2.02, 1.47, 1.5), c(–1.91, –1.39, 1.36), and c(0.77, –2.38, 1.5). to generate skin to wrap an area, such as the area in figs. (7)-(10), the interior and exterior circle muscles must be identified. for two exterior muscles tangents to one another, the skin to connect these two circles can be easily produced. a advances in technology innovation, vol. 5, no. 2, 2020, pp. 126-134 134 circular arc is an appropriate choice to blend these two circles, provided the radii of the circles contain the circular arc. using the chosen radius, both the center of the circle and the start/end angle of the circular arc can be easily obtained. 4. conclusion and future research tangency among circles has numerous important applications, such as in packing problems. although most researchers concentrate on finding circles inside an area, a contrasting approach is employed by constructing the area inside out, from a 1d curve (skeleton) to 2d area (union of circles/muscles), in order to ensure a smooth boundary area (g 1 -continuity boundary curve). the algorithm used to extend the area is based on the binary approach, and it can be easily used for growing areas. future research will emphasize on the design of area-growth directions. for example, the area generated from piecewise cubic bezier curve, so that the whole area is extended to produce an elongated area, certain data structures, such as medial-axis transforms, can be easily generated. the other example is to automatically grow an area to approach a bounded one. acknowledgment this work was supported in part by the national science council in taiwan under grant most 104-2221e-031-005. conflicts of interest the authors declare no conflict of interest. references [1] c. s. chiang, “the medial axis transform of the region defined by circles and hermit curve,” international conference on computer science and engineering (iccse), july 22-24 2015. [2] l. cinque, s. levialdi, and a. malizia, “shape description using cubic polynomial bezier curves,” pattern recognition letters, vol. 19, pp. 821-828, 1998. [3] h. m. yang, j. j. lu, and h. j. lee, “a bezier curve-based approach to shape description for chinese calligraphy characters,” proceedings of the sixth international conference on document analysis and recognition, pp. 276-280, 2001. [4] c. s. chiang and l. y. hsu, “describing the edge contour of chinese calligraphy with circle and cubic hermite curve,” 2013 computer graphics workshop, july 2013. [5] h. h. chang and h. yan, “vectorization of hand-drawn image using piecewise cubic bezier curves fitting,” pattern recognition, vol. 31, no. 11, pp. 1747-1755, 1998. [6] h. cao and a. c. kot, “lossless data embedding in electronic inks,” ieee transactions on information forensics and security, vol. 5, no. 2, pp. 314-323, 2010. [7] l. cao, z. jia, and j. liu, “computation of medial axis and offset curves of curved boundaries in planar domains based on the cesaro’s approach,” computer aided geometric design, vol. 26, no. 4, pp. 444-454, 2009. [8] p. qin and c. chen, “simulation model of flower using the integration of l-systems with bezier surfaces,” international journal of computer science and network security, vol. 6, no. 2, pp. 65-68, 2006. [9] g. l. orick, k. stephenson, and c. collins, “a linearized circle packing algorithm,” computational geometry 64, pp. 13-29, 2017. [10] o. angel, t. hutchcroft, a. nachmias, and g. ray, “unimodular hyperbolic triangulations: circle packing and random walk,” inventiones mathematicae 206.1, pp. 229-268, 2016. [11] k. e. stange, “the sensual apollonian circle packing,” expositiones mathematicae 34.4, pp. 364-395, 2016. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). 4___aiti#9080___270-278 advances in technology innovation, vol. 7, no. 4, 2022, pp. 270-278 wave transmission and energy dissipation in a box culvert-type slotted breakwater nastain 1,2,* , suripin 3 , nur yuwono 4 , ignatius sriyana 3,† 1 doctoral program of civil engineering department, diponegoro university, indonesia 2 civil engineering department, jenderal soedirman university, indonesia 3 civil engineering department, diponegoro university, indonesia 4 civil and environmental engineering department, gadjah mada university, indonesia received 14 december 2021; received in revised form 12 april 2022; accepted 13 april 2022 doi: https://doi.org/10.46604/aiti.2022.9080 abstract this research is conducted to examine the transmission wave and energy dissipation of a box culvert-type slotted breakwater, which is designed as a breakwater structure with a watertight wall at the top and a box culvert type hole at the bottom. the process involves physical modeling of this structure in the laboratory. the hole and wave parameters are varied to determine the breakwater performance. the results show that the transmission coefficient (kt) value is reduced as the relative hole height (hl/d) value is decreasing and the relative hole length (b/l) and wave steepness (h/l) values are increasing. the energy dissipation coefficient (kd) value increases with an increment in hl/d, h/l, and b/l but starts to decrease after reaching the maximum, which is the optimum h × b/l 2 value. this optimum value is found to be 0.0034(hl/d) 2.618 depending on the (hl/d) value, while the maximum kd value is recorded to be 0.70. keywords: breakwater, box culvert, wave transmission, energy dissipation 1. introduction coastal protection using rubble mound breakwaters requires a large amount of construction material, and the dimensions required for deeper waters are usually higher, leading to an increase in the quantity of materials needed [1-4]. the use of natural stone in large quantities for construction, however, can damage mining sites and quarries and deplete limited natural resources. it is also not advisable to use rubble mound breakwaters because they require a good subgrade bearing capacity [1, 5-6]. one of the efforts directed towards solving this problem is using a wall-style perforated breakwater which can save on the quantity of construction material required and is also considered environmentally friendly due to its ability to ensure effective water circulation [4, 7-9]. however, the design of the model, in the form of a wall, shows it has a very small and negligible relative wall thickness (b/l) [10], making it less effective in reducing the incident wave height when the porosity of the hole (ε) and wave length (l) is large. theoretically, the energy dissipation coefficient (kd) of the thin wall perforated breakwater model is 0 when the porosity of the hole is 0, or 1 while the maximum is only 0.3 [11]. several studies have been conducted on the box culvert-type slotted breakwater with a focus on different types, since it was first proposed by jarlan [12], after investigating the hollow barrier model in front of the upright wall structure. for example, mackay and johanning [13] studied the hollow vertical wall model using analytical and numerical models. vijay et al. * corresponding author. e-mail address: nastain@unsoed.ac.id † corresponding author. e-mail address: sriyana808@gmail.com advances in technology innovation, vol. 7, no. 4, 2022, pp. 270-278 [14] examined a multi-layer perforated wall model using physical and numerical models. furthermore, george and cho [11] analyzed a single wall model with a continuous horizontal perforation in the middle, while nikoo et al. [15] focused on a two-layer perforated wall model. ahmed and schlenkhoff [16] also studied a double wall model with holes in the middle, ahmed [9] examined a single wall model with vertical holes in the middle using a numerical model, and rageh and koraim [10] analyzed a single wall model with horizontal holes at the bottom. another study by rageh and koraim [1] focused on the vertical wall model with holes at the bottom, and ariyarathne [17] examined a massive structural model with holes at the top, while suh et al. [2] focused on the kaison model with vertical holes in some of the walls. however, no attention has been paid to the cross-sectional shape of the hole used, despite the important effect of the hole height (hl) and length (b) on the performance of a perforated breakwater in reducing wave height (hi) and incident wave energy (ei). furthermore, the use of a box culvert-type with a relatively large hole length (b) is expected to increase the effectiveness of the breakwater in reducing wave height and energy with a large hole porosity (ε). it is, therefore, important to note that the box culvert-type of the slotted breakwater has two parts: the upper (in the form of a waterproof wall structure) and the lower (in the form of a box culvert model). this research aims to examine the transmission and dissipation of wave energy through a box culvert-type slotted breakwater. the process involves determining the important parameters of hole and wave structure, such as relative hole height (hl/d), relative hole length (b/l), and wave steepness (h/l), using non-dimensional analysis. moreover, the parameters are varied to determine the breakwater performance using the transmission coefficient (kt) and energy dissipation coefficient (kd) values as indicators. 2. experimental program the research is conducted by physically modeling a wave flume, which is equipped with a wave generator, a wave damper, and a wave probe. the experiments are conducted to determine the transmission coefficient (kt) and energy dissipation coefficient (kd) of waves in the box culvert-type perforated breakwater model using different wave and hole structure parameters, obtained through non-dimensional analysis [18]. 2.1. materials the schematic diagram of the wave flume, model location, and wave probe tool is presented in fig. 1. the wave flume used has a length (p) of 15 m, a width (l) of 0.3 m, and a height (t) of 0.45 m, with one flap-type wave generator installed at one end and a wave absorber at the other end, to reduce reflections. moreover, three wave probes (wp) installed at the front and back of the model, with 0.01 cm accuracy, are used to measure the water level elevation. each of these probes is calibrated before being used for measurements. furthermore, the breakwater model is placed in the middle of the wave flume, or approximately 0.5 flume lengths from the wave generator or the wave absorber. fig. 1 wave flume, model location, and wave probe tool 2.2. physical model the breakwater model is produced using an acrylic material with a thickness (b) of 0.01 m and divided into two parts. the upper part is in the form of a watertight wall structure with a constant thickness (b) of 0.01 m. sinking height (hp) is varied at 271 advances in technology innovation, vol. 7, no. 4, 2022, pp. 270-278 0.100, 0.050, 0.025, and 0.000 m from the still water level, while the bottom is in the form of a box culvert model hole with the height (hl), varying at 0.100, 0.150, 0.175, and 0.200 m from the bottom of the flume and the length (b) at 0.2, 0.4, and 0.6 m. this means that 12 models are used for the experiment. the sketches and pictures of the box culvert-type slotted breakwater model are presented in fig. 2 and fig. 3. (a) side view (b) front view fig. 2 sketch of the box culvert-type slotted breakwater model fig. 3 model of the box culvert-type slotted breakwater 2.3. wave height measurement wave heights, including the incident (hi), reflection (hmax and hmin), and transmission (ht), are measured using a wave probe connected to the wave tide meter (wtm), and a computer is used to record the waves from the tide meter. two wave probes are used to measure the incident wave (hi) and reflection (hmax and hmin) wave heights. wave probe 1 (wp-1) is placed at a location 1.25l in front of the breakwater model to measure the minimum reflection wave height (hmin), and wave probe 2 (wp-2) is placed at a location as far as l in front of the breakwater model, to measure the maximum reflection wave height (hmax). this is according to the two-point method of dean and dalrymple [19], mani [20], murali and mani [21], koraim [22], koraim [5] and koraim [23]. to measure the transmission (ht) wave height using one wave probe, wp-3 is located at a distance of 2.0 m behind the breakwater model (wave damper side) [5, 22-23]. the wave length (l) based on the wave period (t) is calculated by using the dispersion equation, according to the linear wave theory. the position of the wave crest or the highest water level (the quasi-antinodes), as far as l from the breakwater model, is visually observed and marked as the wp-2 position. the position of the lowest water level (the quasi-nodes), as far as 1.25l in front of the breakwater model, is determined based on the l value which is calculated as the position of the wp-1 location. the measurement of wave height is repeated three times by shifting the positions of wp-1 and wp-2 slightly to the right and to the left from the initial position, so that the values of hmin and hmax are obtained. these placements are indicated in fig. 4. fig. 4 placement of the wave probe location on the model 272 advances in technology innovation, vol. 7, no. 4, 2022, pp. 270-278 the experiment is conducted at a constant water depth (d) of 0.2 m with the wave period t varied at 1.61, 1.34, 1.15, and 1.02 seconds or (h/l) at 0.0161, 0.0238, 0.0329, and 0.0452. moreover, four variations of relative hole height (hl/d) are used: 0.500, 0.750, 0.875, and 1.000, while the twelve variations of relative hole length (b/l) employed are: 0.093, 0.115, 0.138, 0.160, 0.186, 0.230, 0.277, 0.279, 0.319, 0.345, 0.415, and 0.479. analysis is conducted to determine the relationship between the transmission coefficient (kt) and energy dissipation coefficient (kd) of waves using non-dimensional parameters of the hole and wave structures, such as the relative hole height (hl/d), relative hole length (b/l), and wave steepness (h/l), as shown in eq. (1). , ( , , ) t d l k k f h d b l h l= (1) the transmission coefficient (kt) is calculated using the transmission wave height data (ht), as indicated in eq. (2), while the energy dissipation coefficient (kd) is determined using eq. (4) [16, 24]. moreover, the incident wave height (hi) and reflection wave height (hr) are calculated based on the maximum (hmax) and minimum reflection wave height (hmax) using partial standing wave theory in eq. (5) and eq. (6) [19]. t t i k h h= (2) r r i k h h= (3) 2 2 (1 ) d t r k k k= − − (4) max min 2 i h h h + = (5) max min 2 r h h h − = (6) 3. results and discussion 3.1. effect of the relative hole height (hl/d) and wave steepness (h/l) on the transmission coefficient (kt) and energy dissipation coefficient (kd) the experiment is conducted using a constant hole length of the breakwater model (b) of 0.6 m, while (hl/d) is varied at 0.500, 0.750, 0.875, and 1.000as well as h/l at 0.0161, 0.0238, 0.0329, and 0.0452. the effect of the relative hole height (hl/d) and wave steepness (h/l) on the transmission coefficient (kt) is presented in fig. 5, while the effect on the energy dissipation coefficient (kd) is depicted in fig. 6. fig. 5 shows that a higher h/l and lower hl/d produce a lower kt, as indicated by b = 0.6 m, hl/d = 0.5 – 1.0, and h/l = 0.0161 – 0.0452 which produces a kt = 0.41 – 0.11 or causes a 73.17% reduction. the smallest value (0.11) is found at hl/d = 0.5 and h/l = 0.0452, while the largest (0.41) is at hl/d = 1.0 and h/l = 0.0161. this is associated with the ability of a greater h/l value to cause a steeper wave while a smaller hl/d value can cause a smaller porosity of the structure. this condition makes it difficult for the wave to pass through the breakwater, thereby making the transmission wave height smaller. the multivariate nonlinear regression analysis of the experimental data shows the relationship between the wave transmission coefficient (kt) with the hl/d and h/l functions, as shown in eq. (7), where r 2 = 0.967. 0.016 1.087 6.594( ) 0.321( ) 6.226 t l k h l h d= − + + (7) 273 advances in technology innovation, vol. 7, no. 4, 2022, pp. 270-278 fig. 5 kt function of hl/d and h/l fig. 6 kd function of hl/d and h/l fig. 6 shows that an increment in hl/d and the reduction in h/l lead to an increase in the kd as indicated by b = 0.6 m, hl/d = 0.5 – 1.0, and h/l = 0.0161 – 0.0452, which produces a kd = 0.41 – 0.69 or causes a 68.29% increase. the smallest value (0.41) is recorded at hl/d = 0.5 and h/l = 0.0452, while the highest (0.69) is at hl/d = 1.0 and h/l = 0.0161. this reveals that a higher hl/d or porosity of the structure and smaller h/l or wave steepness can significantly reduce the wave energy. 3.2. effect of the relative hole length (b/l) and relative hole height (hl/d) on the transmission coefficient (kt) and energy dissipation coefficient (kd) the experiment is conducted using a constant steepness wave value h/l of 0.0452 while the length of the breakwater model hole (b) is varied at 0.2 m, 0.4 m, and 0.6 m and the hl/d at 0.500, 0.750, 0.875, and 1.000. the effect of the relative hole length (b/l) and relative hole height (hl/d) on the transmission coefficient (kt) is presented in fig. 7, while the effect on energy dissipation coefficient (kd) is specified in fig. 8. fig. 7 shows that a higher b/l and lower hl/d cause a reduction in kt, as indicated by the use of h/l = 0.0452, b/l = 0.160 – 0.479, and hl/d = 0.5 – 1.0, which produces a kt = 0.43 – 0.11 value or causes a 51.16% decrease. the smallest value (0.11) is found at b/l = 0.479 and hl/d = 0.5, while the highest (0.43) is recorded at b/l = 0.160 and hl/d = 1.0. this is associated with the ability of a greater b/l value to produce a greater frictional effect of fluid turbulence with the hole length b, as well as the smaller hl/d, which causes a smaller porosity of the structure. this condition makes it difficult for the wave to pass through the breakwater, thereby making the transmission wave height smaller. the multivariate nonlinear regression analysis of the experimental data shows the relationship between the wave transmission coefficient (kt) with the hl/d and h/l functions as presented in eq. (8), where r 2 = 0.977. 0.128 1.050 0.932( ) 0.245( ) 0.849 t l k b l h d= − + + (8) 274 advances in technology innovation, vol. 7, no. 4, 2022, pp. 270-278 fig. 7 kt function of b/l and hl/d fig. 8 kd function of b/l and hl/d fig. 8 shows that a lower b/l and higher hl/d cause an increase in kd, which is indicated by 0.41 – 0.69 on the 68.29% increment recorded with h/l = 0.0452, b/l = 0.160 – 0.479, and hl/d = 0.5 – 1.0. meanwhile, the smallest value (0.41) is recorded with b/l = 0.479 and hl/d = 0.5, while the highest (0.69) is found with b/l = 0.160 and hl/d = 1.0. this indicates that higher hl/d or porosity of the structure and smaller b/l can significantly reduce the wave energy. 3.3. effect of the wave steepness (h/l), relative hole length (b/l), and relative hole height (hl/d) on the value of transmission coefficient (kt) and energy dissipation coefficient (kd) the experiment is conducted with the length of the breakwater model hole (b) varying from 0.2 m to 0.4 m and 0.6 m, while the hl/d is 0.500, 0.750, 0.875, and 1.000 and h/l is 0.0161, 0.0238, 0.0329, and 0.0452. the effect of the wave steepness (h/l), relative hole length (b/l), and relative hole height (hl/d) on the transmission coefficient (kt) is presented in fig. 9, while the effect on the energy dissipation coefficient value (kd) is illustrated in fig. 10. fig. 9 shows a higher h/l and b/l, which is represented as h × b/l 2 , and a lower hl/d is able to reduce kt as indicated by the use of h/l = 0.0161 – 0.0452 and b/l = 0.093 – 0.479, as well as hl/d = 0.5 – 1.0, to produce kt = 0.58 – 0.11 or 81.03% reduction. meanwhile, the smallest value (0.11) is found at h/l = 0.0452, b/l = 0.479, and hl/d = 0.5, while the highest value (0.58) is recorded at h/l = 0.0161, b/l = 0.093, and hl/d = 1.0. this is associated with the ability of a greater h/l and b/l to cause a steeper wave and hole length, leading to a greater frictional effect of fluid turbulence with hole length b, while a smaller hl/d is observed to cause a smaller structural porosity. this makes it difficult for the wave to pass through the breakwater, making the transmission wave height smaller. moreover, the multivariate nonlinear regression analysis of the experimental data shows the relationship between the wave transmission coefficient (kt) and h/l, b/l, and hl/d functions, as shown in eq. (9), where r 2 = 0.979. 0.839 2 0.323 0.075( ) [( )]t lk h d h b l −= + × (9) 275 advances in technology innovation, vol. 7, no. 4, 2022, pp. 270-278 fig. 9 kt function of hl/d, b/l, and h/l fig. 10 kd function of hl/d, b/l, and h/l fig. 10 shows the increment in hl/d, h/l, and b/l, which is represented by h × b/l 2 and able to increase the kd, but the value decreases when the maximum value is obtained and the h × b/l 2 is at its optimum. this means more h × b/l 2 lead to a reduction in the kd. this optimum h × b/l 2 is recorded to be 0.0034(hl/d) 2.618 depending on the hl/d value. meanwhile, the maximum �� value is 0.7. this is because the higher hl/d can produce greater structural porosity so that the energy reduction is greater. furthermore, the greater the value of h/l and b/l, the effect of frictional fluid turbulence with the length of the hole will also be larger, so the energy reduction will be even greater. however, after reaching the optimum value of h × b/l 2 , the frictional effect of fluid turbulence with the hole length of b will be more of the wave resistance. as a result, the waves will be more likely to be reflected as reflection waves, and the energy reduction process will be reduced. the optimum value of h × b/l 2 is influenced by the value of hl/d used. the greater the value of hl/d, the higher the optimum value of h × b/l 2 . figs. 11-12 present the graphic nomogram of the relationship between kt and kd box culvert-type slotted breakwater with functions h/l, b/l, and hl/d for regular waves based on eq. (9) and eq. (4). this nomogram can, therefore, be used to determine the values of kt and kd while planning box culvert-type slotted breakwater for beach protection in the field. fig. 11 nomogram of kt function of hl/d, b/l, and h/l 276 advances in technology innovation, vol. 7, no. 4, 2022, pp. 270-278 fig. 12 nomogram of kd function of hl/d, b/l, and h/l 4. conclusions in this study, the transmission and dissipation of a box culvert-type slotted breakwater was designed as a breakwater structure (with a watertight wall at the top and a box culvert type hole at the bottom) and investigated. the results show that the wave transmission coefficient (kt) was reduced as the hl/d decreased and b/l and h/l increased, as indicated by h/l = 0.0161 – 0.0452, b/l = 0.093 – 0.479, and hl/d = 0.5 – 1.0 which produced a kt = 0.58 – 0.11 or caused an 81.03% reduction. moreover, the equation for the relationship among kt, h/l, b/l, and hl/d is kt = 0.075(hl/d) 0.839 × [(h × b/l2)]-0.23. the energy dissipation coefficient (kd), however, increased as the hl/d, h/l, and b/l increased but later started reducing after reaching the maximum value at the optimum h × b/l 2 . this means that more h × b/l 2 led to a reduction in the kd. this optimum h × b/l 2 was recorded to be 0.0034(hl/d) 2.618 depending on the hl/d value. meanwhile, the maximum �� value was 0.7. conflicts of interest the authors declare no conflicts of interest. acknowledgments the authors would like to thank the staff and management of the hydraulic and hydrology laboratory, research center of engineering science, gadjah mada university, for providing the facilities to conduct this experiment. references [1] o. s. rageh, et al., “the use of vertical walls with horizontal slots as breakwaters,” 13th international water technology conference, pp. 1689-1710, march 2009. [2] k. d. suh, et al., “wave reflection from partially perforated-wall caisson breakwater,” ocean engineering, vol. 33, no. 2, pp. 264-280, february 2006. [3] k. d. suh, et al., “wave reflection and transmission by curtainwall-pile breakwaters using circular piles,” ocean engineering, vol. 34, no. 14-15, pp. 2100-2106, october 2007. [4] k. d. suh, et al., “an empirical formula for friction coefficient of a perforated wall with vertical slits,” coastal engineering, vol. 58, no. 1, pp. 85-93, january 2011. [5] a. s. koraim, “hydraulic characteristics of pile-supported l-shaped bars used as a screen breakwater,” ocean engineering, vol. 83, pp. 36-51, june 2014. [6] l. t. somervell, et al., “estimation of friction coefficient for double walled permeable vertical breakwater,” ocean engineering, vol. 156, pp. 25-37, may 2018. [7] z. huang, et al., “hydraulic performance and wave loadings of perforated/slotted coastal structures: a review,” ocean engineering, vol. 38, no. 10, pp. 1031-1053, june 2011. 277 advances in technology innovation, vol. 7, no. 4, 2022, pp. 270-278 [8] n. nurisman, “efektivitas submerged breakwater berlubang terhadap gelombang datang,” jurnal maritim indonesia, vol. 7, no. 2, pp. 165-175, december 2019. (in indonesian) [9] h. ahmed, “wave interaction with vertical slotted walls as a permeable breakwater,” ph.d. dissertation, department of civil engineering, bergischen universität wuppertal, germany, wuppertal, 2011. [10] o. s. rageh, et al., “hydraulic performance of vertical walls with horizontal slots used as breakwater,” coastal engineering, vol. 57, no. 8, pp. 745-756, august 2010. [11] a. george, et al., “hydrodynamic performance of a vertical slotted breakwater,” international journal of naval architecture and ocean engineering, vol. 12, pp. 468-478, december 2020. [12] g. jarlan, “a perforated vertical wall breakwater,” the dock and harbour authority, vol. 41, no. 486, pp. 394-398, 1961. [13] e. mackay, et al., “comparison of analytical and numerical solutions for wave interaction with a vertical porous barrier,” ocean engineering, vol. 199, article no. 107032, march 2020. [14] k. g. vijay, et al., “wave interaction with multiple slotted barriers inside harbour: physical and numerical modelling,” ocean engineering, vol. 193, article no. 106623, december 2019. [15] m. r. nikoo, et al., “multi-objective optimum design of double-layer perforated-wall breakwaters: application of nsga-ii and bargaining models,” applied ocean research, vol. 47, pp. 47-52, august 2014. [16] h. ahmed, et al., “numerical investigation of wave interaction with double vertical slotted walls,” international journal of environmental, ecological, geological, and mining engineering, vol. 8, no. 8, pp. 536-543, april 2014. [17] h. a. k. s. ariyarathne, “efficiency of perforated breakwater and associated energy dissipation,” master thesis, department of civil engineering, texas a&m university, texas, 2017. [18] n. yuwono, perencanaan model skala hidraulis, yogyakarta: pt kanisius, 2021. (in indonesian) [19] r. g. dean, et al., water wave mechanics for engineers and scientists, new jersey: world scientific publishing, 1991. [20] j. s. mani, “design of y-frame floating breakwater,” journal of waterway, port, coastal, and ocean engineering, vol. 117, no. 2, pp. 105-119, march 1991. [21] k. murani, et al., “performance of cage floating breakwater,” journal of waterway, port, coastal, and ocean engineering, vol. 123, no. 4, pp.172-179, july 1997. [22] a. s. koraim, “the hydrodynamic characteristics of slotted breakwaters under regular waves,” journal of marine science and technology, vol. 16, pp. 331-342, may 2011. [23] a. s. koraim, “mathematical study for analyzing caisson breakwater supported by two rows of piles,” ocean engineering, vol. 104, pp. 89-106, august 2015. [24] b. m. webb, et al., “wave transmission through artificial reef breakwaters,” coastal structures and solutions to coastal disasters joint conference, pp. 432-441, september 2015. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). 278  advances in technology innovation, vol. 8, no. 3, 2023, pp. 210-218 english language proofreader: chih-wen teng effect of pin diode integration on patch antennas for frequency reconfigurable antenna applications boyapati bharathidevi, jayendra kumar* school of electronics engineering, vit-ap university, amaravathi, andhra pradesh, india received 10 january 2022; received in revised form 05 august 2022; accepted 04 may 2023 doi: https://doi.org/10.46604/aiti.2023.9235 abstract pin diodes are commonly used to design reconfigurable antennas owing to their sufficient isolation, lower cost, and ease of fabrication. this study aims to explore the effect of biasing conditions of a pin diode radio frequency (rf) switch on a frequency-reconfigurable antenna. this approach investigates the contribution of the forward diode current and the reversed biased voltage on the shift in the operating band, the impedance matching, and the radiation efficiency of a reconfigurable antenna. the benefits and drawbacks of different approaches to modeling pin diode rf switches are demonstrated on ansys electromagnetic switch. the result shows a significant match between simulated and measured operating bands, impedance matching, and radiation efficiency. the proposed rf switch model can be used as a practical simulation model for implementing various reconfigurable microwave components. keywords: biasing circuit, microstrip antenna, pin diode switch, reconfigurable antenna, rf switch 1. introduction microstrip antennas are one of the most popular printed antennas which play a significant role in today’s wireless communication systems. extensive developments including bandwidth enhancement, surface wave reduction, gain enhancement, and radiation efficiency improvement are aimed at fulfilling the requirements of existing and future communication systems, which have made the microstrip antenna an important element in the field of radio frequency (rf) and microwave engineering. in recent years, research interests have been focused on increasing the features of microstrip antennas as multi-band [1-3], wideband [4], multi-polarization [5], switched parasitic multiple-input-multiple-output antennas [6], and frequency, polarization, and pattern diversity. reconfigurability can be achieved through two methods, (i) modifying the length of the current path or (ii) placing the slotted filters into the antenna structure using rf switches [7]. naturally, changing the state of the rf switches modifies the impedance levels which in turn modulates the frequency response of the antenna. the application of numerous reconfigurable antennas has been reported in the literature, and novelty is observed in terms of the type of reconfigurability (frequency, pattern, or polarization), design simplification, or improvement in the antenna parameters. the rf switch is a key component of reconfigurable antenna modeling and biasing conditions have a significant effect on its performance, durability, and robustness. an electronic rf switch draws a significant amount of current; thus, it is essential to estimate the requirement of the power supply for reconfigurable operations. previous research has proposed the different models use, including an ideal open and short circuit and a diode equivalent circuit without direct current (dc) blocking capacitors of the rf pin diode switch for reconfigurable antennas. the dc blocking capacitors and rf blocking inductors are necessary components to protect the rf signal and diode biasing source, * corresponding author. e-mail address: kumar.jayendra@gmail.com advances in technology innovation, vol. 8, no. 3, 2023, pp. 210-218 211 respectively. investigation into the isolation and insertion loss of these switches after integrating them into a microwave system is required to ensure the desired switching operations. however, the switching operation of a pin diode rf is greatly dependent on the biasing elements and conditions. moreover, due to the rapid development of computer-added, it is convenient and common practice to use a software package to design and analyze a new system. although the accuracy of the simulation results is completely dependent on the accuracy of the component models, defined in the software package. nevertheless, the reconfigurable antenna is a wellestablished field, neither modeling nor biasing conditions of the rf switches for reconfigurable antenna application are discussed in the literature. this paper aims to demonstrate the (i) effect of the different rf switch models and (ii) biasing conditions on the radiation performance of a reconfigurable antenna. the use of different rf switches in optical [8-9], mems [10], fets [11], pin diodes [12], and micro-fluids [13] reconfigurable antennas have been proposed. optically controlled reconfigurable antennas are extremely delicate, thus, making them difficult to implement for practical applications [8-9]. mems rf switches require higher operating voltage and are relatively higher in cost [10]. on the contrary, fet-based switches require low operating power and are cheaper, but suffer from higher loss and poor linearity. pin diode-based switches offer a low-cost and low-loss solution [12]. hence, most of the reconfigurable antennas, frequency [14-18], polarization [19-21], and pattern [22-23] proposed in the literature are developed using a pin diode rf switch. vian and popovicet.al. [8], achieved frequency, pattern, and polarization reconfigurability by incorporating eight pin diodes in a single antenna structure. the compound reconfigurability of the antenna is remarkable but the integration of numerous pin diodes makes the overall structure too complicated for fabrication and installation. however, considerable results have been obtained so far due to ease of integration and moderate isolation. thus, in this work, the pin diode rf switch was chosen to identify the effect of the rf switch modeling and biasing the conditions on the accuracy, durability, and performance of reconfigurable antennas. numerous studies of reconfigurable antennas using different switching elements have been proposed for decades. the novelty of most of these antennas lies in their structure or switching mechanism. however, a detailed investigation of the antenna characteristics concerning the biasing conditions is unavailable in the previous studies. the investigation is imperative to establish the trade-off between antenna performance and dc power consumption. the present study emphasizes the antenna characteristics against the supplied forward current and reverses bias voltage. the effects of bias conditions on the performance of a pin diode-integrated reconfigurable antenna are reported using a simple slotted rectangular patch antenna. a trade-off between the antenna radiation performance and the diode current was established. a detailed analysis of a frequency reconfigurable antenna against the rf switch model and the biasing condition of the switch has been presented. a pin diode rf switch integrated rectangular slot has been incorporated into the patch of a conventional rectangular patch antenna to obtain dual-band frequency reconfigurability. in the simulation, the performance of the antenna has been analyzed using an ideal switch model and an actual switch model. thereafter the forward diode current and the reverse diode capacitance were varied to observe the effect of the biasing conditions on the performance of the antenna. the prototype of the frequency reconfigurable antenna was developed, and its scattering parameter and radiation efficiency were experimentally analyzed for different biasing conditions. during simulation, it was found that the other antenna parameters, such as radiation pattern, half-power beam width, etc., were not significantly affected by the biasing conditions, thus, experimental results have not been presented. the ideal model of the switch has been shown to exhibit an unacceptable frequency shift after fabrication, making it unacceptable for simulation. this study proposed the switch model has a good agreement amongst simulation and measured performance of the antenna. advances in technology innovation, vol. 8, no. 3, 2023, pp. 210-218 212 2. configurations and modeling of rf switch series and shunt are commonly used configurations of pin diode rf switch [24] as shown in fig. 1. the isolation and insertion loss of a series configuration switch depends on the off-state diode capacitance (ct) and the on-state diode resistance (rs), respectively. on the other hand, the isolation and insertion loss of a shunt configuration depends on the onstate diode resistance (rs) and the off-state diode capacitance (ct), respectively. however, in a shunt configuration switch, the switching element is not directly connected to the rf transmission line; thus, providing lower insertion loss than the series configuration switch. moreover, the series configuration of the pin diode rf switch (shown in fig. 1.) is easy to integrate with reconfigurable antenna applications as has been widely reported in the previous literature. in fig. 1, capacitors c1 and c2 are the dc blocking capacitors, r1 is a current limiting resistor, l1 is used to protect the dc power supply from rf signals, and ls is the diode inductance. using ansys electromagnetics suite, a high-frequency-structural-simulator (hfss), an equivalent circuit was designed by assigning the lumped r-l-c boundary to the multiple 2d or 3d interconnected structures. a single structure can be modeled as a single component or a parallel combination of r-l, r-c, l-c, or r-l-c circuits. however, it is necessary to define the direction of the current flow while assigning the lumped r-l-c boundary. (a) on-state (b) off-state fig. 1 series configuration of an rf switch in this work, a 2d sheet was chosen to model a pin diode rf switch in the on-state and the off-state, as shown in fig. 1. excluding biasing elements, four interconnected structures are required to model a pin diode rf switch in the on-state or the off-state, as shown in fig. 2. for the on-state, the parallel combination of r and ct is replaced by rs as shown in fig. 2(b). moreover, the dimensions of sheets are chosen based on the dimensions of the practical surface-mount capacitors and diodes. the values of ls, ct, and rs can be found in the technical datasheet of the respective diode [25]. (a) on-state (b) off-state fig. 2 model of pin diode rf switch in the ansys electromagnetics suite, hfss 3. performance analysis of the frequency reconfigurable antenna in this section, a conventional rectangular patch antenna was designed, and a rectangular slot integrated with two pin diode rf switches introduced frequency reconfigurability. the antenna was designed on an fr4 substrate of thickness 1.6 mm and fed through a 50 ω co-axial sub-miniature version-a (sma) connector. the fr4 substrates are commonly used for printed circuit board (pcb) development and are much cheaper compared to other low-loss substrates such as rogers rt duroid. advances in technology innovation, vol. 8, no. 3, 2023, pp. 210-218 213 also, a low-frequency patch antenna exhibits acceptable performance for the fr4 substrate of loss tangent 0.02. the configuration of the antenna and magnified view of the switch model is shown in fig. 3. the proposed antenna configuration and the magnified view of the biasing arrangement are shown in fig. 3. (a) test antenna layout (b) switch model fig. 3 configuration of the frequency reconfigurable antenna (l1 = 1 mh, r1 = 100 ω, c1 = c2 = 1 uf, rs, and ls are taken from datasheet) 3.1. simulation analysis this section presents the performance of the frequency reconfigurable antenna against different switch models and biasing conditions. in the ideal condition, for the off-state, all the circuit elements and biasing line have been removed and for the on-state, two copper sheets were used to establish the connection. the switch model discussed in section 2 was used as a practical switch. the biasing conditions were controlled by changing the on-state resistance, and the off-state capacitance, as presented in the technical datasheet of the pin diode nxp bap65-02 [25]. the proposed antenna is intended to operate in 400 mhz wireless medical telemetry service (wmts) and 700 mhz global system for mobile (gsm) bands. thus, the pin diode nxp bap65-02 is chosen, which is known to support the intended frequency range. if another diode is used, the series inductance, on-state resistance, and off-state capacitance will change accordingly. the pin diode nxp bap65-02 has sufficiently good isolation in the off-state and low insertion loss in the onstate, as presented in its technical datasheet [25]. the antenna response was obtained for the ideal switch and the practical switch model, as shown in fig. 4. in an ideal case, the switch positions are open and short in the off-state and the on-state, respectively. in practice, for the on-state, a diode resistance of 0.35 ω is set, corresponding to 100 ma forward diode current. similarly, in the off-state, the capacitance was set to 0.375 pf, corresponding to 20 v reverse voltage. the variation in nxp bap65-02 on-state diode forward resistance and off-state junction capacitance against the forward current and the reverse voltage, respectively listed in table 1, is reproduced from the technical datasheet [25]. (a) scattering parameter (s11) of the antenna for different diode models (b) radiation efficiency of the antenna for different diode models fig. 4 performance of the antenna for the ideal condition of the switch and pin diode rf switch model advances in technology innovation, vol. 8, no. 3, 2023, pp. 210-218 214 table 1 nxp bap65-02 characteristics [25] forward characteristics reverse characteristics forward current (ma) forward resistance rs (ω) reverse voltage (vr) junction capacitance cd (pf) 1 1 0 0.65 5 0.65 1 0.55 10 0.56 3 0.5 100 0.35 20 0.375 the scattering parameters and the radiation efficiency of the antenna are shown in fig. 4. it is evident in fig. 4(a) that there is a considerable difference between impedance matching as well as operating bands for both conditions. moreover, in the off-state, the operating bands are completely different, confirming the incapacity of the ideal diode model. to develop an electronically reconfigurable antenna, practical diode(s) are integrated into the antenna structure. when these active elements are added to the antenna structure, the distributed reactance sufficiently changes, leading to the shift in series or parallel resonance center frequency. thus, the shift in the operating band should be accounted for to develop a practical electronically reconfigurable antenna. in fig. 4(b), it can be observed that the ideal model of the diode exhibits much better radiation performance than a practical model, which confirms the significant losses due to the integration of pin diode rf switches. although, it should be noted that the antenna has poor radiation efficiency in the off-state than the on-state. the performance of the reconfigurable antenna was further analyzed to consider the different biasing conditions. in the on-state, the forward diode current was increased stepwise to observe its effect on the shift in the operating band, the impedance matching, and the radiation efficiency. in the off-state, the reversed biased voltage is varied to analyze the antenna performances. the forward current and reverse bias voltage change did not shift the operating bands, as shown in fig . 5(a). however, a remarkable change in the impedance matching can be observed in the off-state. a higher forward diode current yields better radiation efficiency, whereas the reverse bias voltage did not affect the radiation performance, as shown in fig. 5(b) . the simulated radiation patterns against different biasing conditions are shown in fig. 6. it has been observed that there is a negligible effect of the biasing conditions on the radiation pattern in both the off and on conditions. however, in the on state, the back radiation pattern of the antenna is slightly affected, as shown in fig. 6(b). it has been noticed in fig. 5(b) that higher forward current yields more radiation from the antenna structure, which holds for the front as well as back radiations . therefore, as the forward current increases, the back radiation of the antenna also increases due to improvement in the radiation efficiency. (a) scattering parameter (s11) of the antenna in on-state and off-state against different forward current and reverse voltages, respectively (b) radiation efficiency of the antenna in on-state and off-state against different forward current and reverse voltages, respectively fig. 5 simulated performance against different biasing conditions advances in technology innovation, vol. 8, no. 3, 2023, pp. 210-218 215 (a) off-state e-plane radiation pattern at 400 mhz (b) on-state radiation pattern at 700 mhz fig. 6 e-plane radiation pattern of the antenna in on-state and off-state against different forward current and reverse voltages, respectively 3.2. experimental analysis a physical prototype of the antenna was developed and experimentally analyzed to validate the simulated observations. the prototype of the antenna with the s11 parameter measurement setup is shown in fig. 7. the pcb of this design can be developed using a traditional chemical etching process or a pcb milling machine. a pcb milling machine is a non-chemical computer-aided machine that creates a high-quality pcb by milling the metals from a pcb laminate. the computer guides the milling bits as per the layout provided. to realize a prototype with an ideal diode model, the switch positions were open and short for the off-state and the on-state, respectively. (a) scattering parameter measurement setup (b) top view of the prototype (c) top view of the prototype fig. 7 s11 measurement setup and prototype of the antenna for a practical realization, two surface-mount capacitors were soldered at the anode and cathode of a pin diode nxp bap65-02. the rf switches were biased using a dc power supply. the s11 parameters were measured using a two-port network analyzer. the radiation efficiency (η = g / d) is measured using the gain (g) / directivity(d) method, as suggested by huang [26] and kumar et al. [27]. the gain of the antenna was measured using the two-antenna method and divided by the simulated directivity to estimate the radiation efficiency. the discussion on radiation efficiency presented is derived from the measured gain. however, the measured gain curve is not presented in this paper. the measured s11 parameters and the radiation efficiency of the antenna against forward diode current are shown in fig. 8. since the reverse bias voltage does not significantly affect the antenna performance, the corresponding experimental analysis was not conducted. an unacceptable shift in the operating band for the off-state and a slight shift in the on-state can be observed in fig. 8(a) for the ideal and the practical diode models, as observed during simulation analysis. advances in technology innovation, vol. 8, no. 3, 2023, pp. 210-218 216 (a) measured scattering parameter (s11) against the forward current (b) measured radiation efficiency against the forward current fig. 8 measured performance against different biasing current however, for a forward current of 10 ma, the antenna has extremely poor impedance matching. in the on-state, the measured radiation efficiency for the ideal diode model is nearly 11% better than the practical model with a diode forward current of 100 ma. as observed in the simulation, the measured radiation efficiency of the antenna is extremely poor in the on-state. the simulated and measured radiation patterns of the fabricated prototype have been presented in fig. 9. the antenna radiation pattern is measured using a radiation pattern measurement setup that consists of an rf source, a spectrum analyzer, a turntable with an automated data logger and 360° rotatable antenna stand to hold the antenna under test (aut), and another 360° rotatable antenna stand to hold the source antenna. this setup is placed in an electromagnetically shielded chamber. in the on-state, the antenna has perfect broadside radiation, and the co and cross-polarization isolation is more than 18 db. in the off-state, the antenna has a slightly tilted main lobe, and the isolation between co and cross-polarization is more than 16 db. a good agreement has been demonstrated between simulation and measured results which validates the proposed work. (a) simulated and measured e-plane radiation at 690 mhz in on-state (b) simulated and measured e-plane radiation at 420 mhz in off-state fig. 9 simulated and measured e-plane radiation pattern of the fabricated prototype 4. conclusions an investigation on reconfigurable antenna characteristics against different diode models and biasing condition of the diode has been carried out. during the simulation, the reactance of a diode should be considered to develop a practical antenna to avoid a mismatch between simulated and measured results. increasing the diode forward current can improve the impedance matching and the radiation efficiency of an antenna. the reverse bias voltage has an insignificant effect on the radiation performance. however, the off-state radiation efficiency is poor when the diode is not active. the result shows that the advances in technology innovation, vol. 8, no. 3, 2023, pp. 210-218 217 proposed rf switch model has good impedance matching, radiation efficiency, and co and cross-polarization isolation. the proposed method also be performed for other active switching elements such as a varactor diode, field-effect transistor, and bipolar junction transistors. conflicts of interest the authors declare no conflict of interest. references [1] j. kumar, b. basu, f. a. talukdar, and a. nandi, “multimode-inspired low cross-polarization multiband antenna fabricated using graphene-based conductive ink,” ieee antennas and wireless propagation letters, vol. 17, no. 10, pp. 1861-1865, october 2018. [2] r. s. uqaili, j. a. uqaili, s. zahra, f. b. soomro, and a. akbar, “a study on dual-band microstrip rectangular patch antenna for wi-fi,” proceedings of engineering and technology innovation, vol. 16, pp. 1-12, august 2020. [3] a. kumar and a. p. s. pharwaha, “design and optimization of micro-machined sierpinski carpet fractal antenna using ant lion optimization,” international journal of engineering and technology innovation, vol. 10, no. 4, pp. 306318, september 2020. [4] s. patil, v. kapse, s. sharma, and a. k. pandey, “low profile wideband dual-ring slot antenna for biomedical applications,” proceedings of engineering and technology innovation, vol. 19, pp. 38-44, august 2021. [5] w. yang, m. xun, w. che, w. feng, y. zhang, and q. xue, “novel compact high-gain differential-fed dualpolarized filtering patch antenna,” ieee transactions on antennas and propagation, vol. 67, no. 12, pp. 7261-7271, december 2019. [6] r. rajmohan, k. s. vishvaksenan, and r. kalidoss, “compact four port slot based mimo antenna with isolation enhancement,” international journal of electronics, vol. 108, no. 7, pp. 1075-1088, april 2021. [7] j. kumar, b. basu, and f. a. talukdar, “modeling of a pin diode rf switch for reconfigurable antenna application,” scientia iranica: international journal of science and technology, vol. 26, no. 3, pp. 1714-1723, may–june 2019. [8] j. vian and z. popovic, “a transmit/receive active antenna with fast low-power optical switching,” ieee transactions on microwave theory and techniques, vol. 48, no. 12, pp. 2686-2691, december 2000. [9] c. j. panagamuwa, a. chauraya, and j. c. vardaxoglou, “frequency and beam reconfigurable antenna using photoconducting switches,” ieee transactions on antennas and propagation, vol. 54, no. 2, pp. 449-454, february 2006. [10] y. xu, y. tian, b. zhang, j. duan, and l. yan, “a novel rf mems switch on frequency reconfigurable antenna application,” microsystem technologies, vol. 24, pp. 3833-3841, march 2018. [11] j. kumar, b. basu, f. a. talukdar, and a. nandi, “stable‐multiband frequency reconfigurable antenna with improved radiation efficiency and increased number of multiband operations,” iet microwaves, antennas & propagation, vol. 13, no. 5, pp. 642-648, april 2019. [12] j. kumar, f. a. talukdar, and b. basu, “frequency reconfigurable e‐shaped patch antenna for medical applications,” microwave and optical technology letters, vol. 58, no. 9, pp. 2214-2217, september 2016. [13] x. yang, j. lin, g. chen, and f. kong, “frequency reconfigurable antenna for wireless communications using gaas fet switch,” ieee antennas and wireless propagation letters, vol. 14, pp. 807-810, december 2015. [14] a. boufrioua, “frequency reconfigurable antenna designs using pin diode for wireless communication applications,” wireless personal communications, vol. 110, no. 4, pp. 1879-1885, february 2020. [15] t. khan and m. rahman, “design of low-profile frequency reconfigurable antenna for multiband applications,” international journal of electronics letters, vol. 10, no. 1, pp. 30-47, september 2022. [16] a. bhattacharjee and s. dwari, “a monopole antenna with reconfigurable circular polarization and pattern tilting ability in two switchable wide frequency bands,” ieee antennas and wireless propagation letters, vol. 20, no. 9, pp. 1661-1665, september 2021. [17] j. hu, x. yang, l. ge, z. guo, z. c. hao, and h. wong, “a reconfigurable 1 × 4 circularly polarized patch array antenna with frequency, radiation pattern, and polarization agility,” ieee transactions on antennas and propagation, vol. 69, no. 8, pp. 5124-5129, august 2021. [18] j. kumar, b. basu, f. a talukdar, and a. nandi, “graphene-based multimode inspired frequency reconfigurable user terminal antenna for satellite communication,” iet communication, vol. 12, no. 1, pp. 67-74, january 2018. advances in technology innovation, vol. 8, no. 3, 2023, pp. 210-218 218 [19] s. w. lee and y. sung, “compact frequency reconfigurable antenna for lte/wwan mobile handset applications,” ieee transactions on antennas and propagation, vol. 63, no. 10, pp. 4572-4577, october 2015. [20] b. liu, j. h. qiu, s. c. lan, and g. q. li, “a wideband-to-narrowband rectangular dielectric resonator antenna integrated with tunable bandpass filter,” ieee access, vol. 7, pp. 61251-61258, march 2019. [21] a. panahi, x. l. bao, k. yang, o. o’conchubhair, and m. j. ammann, “a simple polarization reconfigurable printed monopole antenna,” ieee transactions on antennas and propagation, vol. 63, no. 11, pp. 5129-5134, november 2015. [22] s. xiao, c. zheng, m. li, j. xiong, and b. z wang, “varactor-loaded pattern reconfigurable array for wide-angle scanning with low gain fluctuation,” ieee transactions on antennas and propagation, vol. 63, no. 5, pp. 2364-2369, may 2015. [23] j. ren, x. yang, j. yin, and y. yin, “a novel antenna with reconfigurable patterns using h-shaped structures,” ieee antennas and wireless propagation letters, vol. 14, pp. 915-918, november 2015. [24] the pin diode circuit designers’ handbook. watertown, ma: microsemi corporation, 1998. [25] “bap65-02 silicon pin diode” https://www.nxp.com/docs/en/data-sheet/bap65-02.pdf, january 28, 2019. [26] y. huang, handbook of antenna technologies, 1st ed, singapore: springer, 2016. [27] j. kumar, b. basu, f. a. talukdar, and a. nandi, “graphene-based wideband antenna for aeronautical radionavigation applications,” journal of. electromagnetic waves and applications, vol. 31, no. 18, pp. 2046-2054, august 2017. copyright© by the authors. licensee taeti, taiwan. this article is an open-access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). https://www.nxp.com/docs/en/data-sheet/bap65-02.pdf 5___aiti#8714__56-65 advances in technology innovation, vol. 7, no. 1, 2022, pp. 56-65 effects of the vibration amplitude in vibratory stress relief on the fatigue life of structures cuong bui manh 1,* , duong nguyen van 2 , si do van 1 , manh phan van 1 , van thao le 3 1 department of engineering mechanics, le quy don technical university, hanoi, vietnam 2 department of materials science and engineering, le quy don technical university, hanoi, vietnam 3 advanced technology center, le quy don technical university, hanoi, vietnam received 16 october 2021; received in revised form 08 december 2021; accepted 09 december 2021 doi: https://doi.org/10.46604/aiti.2021.8714 abstract this research aims to investigate the effects of vibration amplitude in vibratory stress relief (vsr) on the fatigue strength of structures with residual stress. experiments are carried out on specimens with residual stress generated by local heating. flat specimens made of a36 steel are prepared to be suitable for setting up fatigue bending tests on a vibrating table. several groups of samples are subjected to vsr at resonant frequencies with different acceleration amplitudes. the results show that vsr has an important influence on the residual stress and fatigue limit of steel specimens. the maximum residual stress in the samples is reduced about 73% when the amplitude of vibration acceleration is 57 m/s 2 . the vsr method can also improve the fatigue limit by up to 14% for steel samples with residual stress. keywords: vibratory stress relief (vsr), residual stress, fatigue life, fatigue limit, vibration amplitude 1. introduction residual stress is the internal stress existing between parts of workpieces in the absence of external loads. this stress is generated when the workpieces experience inhomogeneous deformation, e.g., heating, cooling, and plastic deformation [1-2]. many technological processes, such as welding, quenching, and metal forming, can produce residual stress. residual stress often leads to the loss in the geometry accuracy of workpieces after machining, and causes the degradation of corrosion resistance in various environments [3]. residual stress also has effects on the mechanical properties of materials and therefore on the load-bearing capacity and life service of components. especially, the tensile residual stress which usually appears in welding has negative effects on the fatigue limit and fatigue life of workpieces [4-5]. relieving residual stress is one of the major concerns in mechanical engineering processes. there are many relieving methods implemented in manufacturing: annealing, local heating, monotonic overloading, shot pinning, vibratory stress relief (vsr), etc. among the mechanical methods for relieving residual stress, vsr is considered an effective, flexible, inexpensive, and eco-friendly solution [6-8]. in vsr, cyclic external loads are applied to the workpieces with residual stress for a certain length of time, and this process is believed to cause microplastic deformation in micro regions of the workpieces and lead to the relaxation of residual stress [8-9]. this technology has been successfully implemented in the manufacturing of many important workpieces, such as welded shafts, marine shafts, large rails, large surface plates from stainless steel, and thin parts from aluminum alloy [10-13]. there have been lots of publications investigating the vsr effects on the residual stress state and the mechanical properties of specimens [7-8, 12, 14-16]. however, the research examining the vsr effects on the fatigue characteristics of workpieces is still limited. fang et al. [17] demonstrated that the fatigue life of welding steel samples could be improved up to * corresponding author. e-mail address: buimanhcuongkck@lqdtu.edu.vn advances in technology innovation, vol. 7, no. 1, 2022, pp. 56-65 25% by using vsr. munsi et al. [18] also carried out a study on vsr for welded steel bars at non-resonant frequency with two load levels for a short interval of time, only 5 s. they found that applying vsr at high loads and a small number of cycles could reduce the residual stress and increase the fatigue life of structures up to 17%. meanwhile, the thermal method could reduce the residual stress and reduce the fatigue life of structures up to 43%. djuric et al. [19] studied the vsr effects on the properties of martensitic steel, and found an increase of fatigue damage due to the applied vsr treatment. in the work of gao et al. [20], the fatigue characteristics of ti-6al-4v titanium alloys were determined after vsr at different vibration amplitudes. the fatigue limit of samples was found to be slightly reduced with the increasing vsr amplitude (several percent to 10%) while the residual stress could also be reduced up to 60%. song and zhang [21] applied vsr for 7075-t651 aluminum alloy, and found that the fatigue life of specimens could be enhanced to 6.3% with the help of vsr. gao et al. [22] also conducted vsr for 7075-t651 aluminum alloy, and concluded that low amplitude vsr could reduce the residual stress and improve the fatigue life of 7075-t651 aluminum alloy. in this study, vsr is conducted on specimens of astm a36 steel, which is a popular material for weld structures. the residual stress is created by local heating, which is similar to the welding process and different from the pre-strain method in recent literature [19, 21-22]. fatigue bending tests are conducted on a vibrating table with flat specimens, and the fatigue limit is determined by an improved staircase method. herein, this study focuses on the effects of vsr amplitude on the fatigue properties of a36 steel with residual stress, demonstrating the vsr ability of reducing the residual stress and improving the fatigue limit. 2. materials and experimental methods 2.1. test specimens in this work, fifty-five identical flat test specimens, which are designed according to astm e466 for bending fatigue tests, are extracted from a large carbon steel panel (fig. 1). the specimens’ thickness is 6 mm. the end part of each specimen is with a threaded hole m5 (fig. 1(a)) to attach additional weight, which creates an addition flat bending load and increases the stress at the a-a cross-section during the vibration test. tables 1 and 2 present the mechanical properties (e.g., density, modulus, and yield strength (ys)) and the chemical compositions of the carbon steel specimens. (a) geometry and dimensions of specimens (b) fabricated specimens fig. 1 design and preparation of specimens table 1 mechanical properties of low carbon steel density (kg/m 3 ) modulus of elasticity (gpa) poisson’s ratio yield strength (mpa) ultimate tensile strength (mpa) 7850 200 0.3 296 440 57 advances in technology innovation, vol. 7, no. 1, 2022, pp. 56-65 table 2 chemical compositions of specimens element c si mn p s cr mo ni al wt. % 0.1452 0.0087 0.4483 0.0139 0.0028 0.0054 0.0030 0.0091 0.0723 element co cu ti v nb w pb fe wt. % 0.0080 0.0118 < 0.0010 < 0.0020 0.0041 0.0470 < 0.0100 99.183 2.2. generation of the residual and dynamic stress in specimens first, for each steel specimen, high residual stress is induced by spot heating in the transition zone (fig. 2(a)). the displacement velocity of the heat source and the heating time are controlled so that the temperature at the center of the heating zone is around 1000°c. the heat source is moved along line a-a (fig. 1(a)) at a speed of 2 mm/s. after the local heating, each sample is rapidly cooled to room temperature by quickly putting it into a water tank. to monitor the dynamic strain and stress, an hbm strain gage (type 1-lm11-3/350ge) is bonded onto the middle of line a-a. the strain gauge is attached to the surface of specimens in the region of interest (roi) (fig. 2(b)). the dynamic stress is generated and controlled by varying the exciting acceleration amplitude and maintaining the constant frequency of a shaker. therefore, it is necessary to survey the relationship between the exciting acceleration amplitude of the samples and the dynamic stress generated at the cross-section (a-a) before conducting vsr and fatigue tests. an lds electrodynamic shaker (model v830-335t) with an lds laser usb vibration controller (model las200, s/n 10870124) is used for generating the vibration of the samples. an lms noise and vibration test system (lms scadas mobile system) is used for signal acquisition, monitoring the displacement at the end of the samples and the dynamic strain in roi, as shown in fig. 3(a). (a) location of local heating to create residual stress (b) strain gauge attachment on a specimen fig. 2 generating residual stress and attaching a strain gage to monitor the dynamic strain and stress (a) the lms noise and vibration test system and the experimental setup (b) natural frequency of a sample with residual stress fig. 3 experiment for monitoring the displacement and the dynamic strain in roi 58 advances in technology innovation, vol. 7, no. 1, 2022, pp. 56-65 the data acquired by the lms scadas mobile system enables studying and establishing the relationship of the exciting frequency and acceleration amplitude with the dynamic strain and stress. the sine vibration experiment is set at 30 m/s 2 with a frequency range from 260 to 300 hz. a sweep rate is set at 0.25 oct/min to exactly determine the resonance frequency of test specimens. the specimens are heated locally to generate the residual stress. the transmissibility function is used for the resonance search, and is computed based on the ratio of the dynamic strain and exciting acceleration, as shown in fig. 3(b). fig. 3(b) shows that the resonant frequency of the test specimens with residual stress is about 283.59 hz. after the controller finds the resonant frequencies in the sweep range, the sine dwell tests are performed. the controller runs a single sine tone at the resonant frequency instead of sweeping through the frequency range, with the acceleration amplitudes of 60 m/s 2 , 90 m/s 2 , 120 m/s 2 , 150 m/s 2 , and 180 m/s 2 , respectively. subsequently, the dynamic stress is calculated from the dynamic strain by hook’s law, as shown in eq. (1). σ = e × ε (1) where σ is the stress (n/m2), ε is the strain (m/m), and e is the modulus of elasticity (n/m2). 2.3. vsr tests vsr tests are conducted on the lds electrodynamic shaker in a scheme like the one in dynamic stress tests. 56 specimens are randomly selected and divided into 7 groups denoted by a, b, c, d, e, f, and g. each group consists of 8 test specimens. three test specimens in group a are randomly selected to measure the residual stress before vsr, and the remaining five samples of group a are used for determining the fatigue limit. the vsr processes are conducted for the specimens in 6 groups (b, c, d, e, f, and g) using the lds electrodynamic shaker with the acceleration amplitudes of 29 m/s 2 , 43 m/s 2 , 57 m/s 2 , 71 m/s 2 , 99 m/s 2 , and 156 m/s 2 , respectively. each sample in these 6 groups is stimulated by the shaker within 10 minutes. the exciting frequency is set at 283.59 hz according to the resonant frequency of the test specimens. after the vsr processes, three samples in each group (b, c, d, e, f, and g) are randomly selected to measure and check the residual stress, and the remaining five test specimens are used for determining the fatigue limit. the diagram of using different test specimen groups is shown in fig. 4. 56 test samples group a: without conducting vsr, 3 of 8 test samples in the group are used for measuring the residual stress. group b: vsr is conducted with the exciting acceleration amplitude of 29 m/s 2 . group c: vsr is conducted with the exciting acceleration amplitude of 43 m/s 2 . group d: vsr is conducted with the exciting acceleration amplitude of 57 m/s 2 . group e: vsr is conducted with the exciting acceleration amplitude of 71 m/s 2 . group f: vsr is conducted with the exciting acceleration amplitude of 99 m/s 2 . group g: vsr is conducted with the exciting acceleration amplitude of 156 m/s 2 . each group contains 8 test samples. after vsr, 3 of 8 test samples in each group are used for measuring the residual stress. the other 5 samples are used for determining the fatigue limit. fig. 4 diagram of different specimen groups 59 advances in technology innovation, vol. 7, no. 1, 2022, pp. 56-65 2.4. measuring residual stress three test specimens of group a are used for measuring the residual stress; the measured value is considered the residual stress before vsr. for the other 6 groups (b, c, d, e, f, and g), after vsr is conducted, three samples are randomly chosen from each group for the measurement of residual stress. the residual stress on the samples before and after vsr is then compared to each other to assess the ability of reducing the residual stress in different loading modes. the measurement of residual stress is performed according to astm e837-13a [23], using a rs200 system (vishay group, usa) and an ea-06-062re-120 strain gauge rosette. fig. 5 shows the residual stress measuring system marked rs200. when measuring the residual stress, the electrical resistance-strain gauge rosettes are attached to the surface of specimens in the same position as in the measurement of dynamic strains (fig. 3(b)). the residual stress on the surface layer is determined by astm e837-13a, as described in the work of gao et al. [16]. fig. 5 specimens and the residual stress measuring system marked rs200 2.5. fatigue test the fatigue tests with a flat bending load are also conducted on the lds electrodynamic shaker, as shown in fig. 6. in each group, 5 test samples are used to determine the fatigue limit. the fatigue test parameters are as follows: the frequency of fatigue vibration test at the resonance frequency of samples is 283.59 hz, the stress ratio is -1, the ambient temperature is 27°c, and the humidity is 60%. the fatigue limit can be determined by using the staircase method [24-25] (fig. 7). the test is performed for the first specimen at a stress level around the fatigue limit. if the result of the above test is “non-fracture”, the second specimen is tested at the stress level with an increment of d. on the other hand, if the result is “fracture”, the second specimen is tested at the stress level with a decrement of d. the fatigue test is repeated until a given number of specimens is used, then the fatigue limit is calculated. this fatigue test method has been widely accepted to examine the fatigue limit of metals. there were also some modified staircase procedures to reduce the number of test specimens and hence lower the cost and running time of this test [26-27]. fig. 6 a specimen and strain gauge in the fatigue test on the lds electrodynamic shaker 60 advances in technology innovation, vol. 7, no. 1, 2022, pp. 56-65 in this study, a modified staircase method proposed by international council for combustion engines is implemented for bending fatigue tests with a limited number of specimens [28]. herein, the first specimen is subjected to a stress level that is most likely well below the average fatigue strength. when this specimen survives 5 × 106 cycles, this same specimen is subjected to a stress level one increment above the previous. this is continued with the same specimen until failure. then, the number of cycles is recorded, and the next specimen is subjected to a stress that is at least 2 increments below the level where the previous specimen fails. the first stress amplitude level is determined, depending on the tensile strength of the material and according to the following empirical expression [29]: σ1 = 0.68 × (0.55 − 0.0001 × σu) × σu (2) where σu is the material tensile strength. when σu = 440 mpa, the first stress amplitude level σ1 is about 152 mpa. the next stress level applied on a specimen is determined by the results obtained in the preceding test. if the preceding specimen, i, fails at the level σi, the next stress level will be σi + 1 = σi – 2d. if the preceding specimen is not broken, σi + 1 = σi + d, where d is a predefined difference in applied stress levels. in this study, d is determined as d = 0.068 × σ1 ≈ 10.4 mpa. the fatigue testing results of five samples in group c is illustrated in fig. 8. there are 5 specimens that undergo the bending tests. the first specimen is broken in the third test at a stress level of σ2. therefore, the second specimen is firstly tested at a stress level σ0 = σ3 − 2d, and then it is tested at elevated stress levels until it is broken at the fourth test with a stress level of σ3. subsequently, the third specimen is tested, and the tests are repeated until the fifth specimen is broken. based on the test results as shown in fig. 8, the fatigue limit is calculated by eq. (2) [24]: 0 1 2 r a d f = + ±     σ σ (3) where σ0 is the lowest stress level (i = 0) in which the less frequent event run-out appears; f and a are the parameters calculated by eqs. (4) and (5). f i= ∑ (4) i a f= ∑ (5) where i is the number of stress levels, and fi is the number of run-outs obtained at the stress level i. determining the load amplitude fatigue test samples are broken or not at the load circle 5 × 106 decreasing the stress amplitude by 2d = 20.8 mpa determining the fatigue limit of sample groups yes no if all 5 samples are tested increasing the stress amplitude by d = 10.4 mpa start fig. 7 diagram illustrating the process of determining the fatigue limit 61 advances in technology innovation, vol. 7, no. 1, 2022, pp. 56-65 fig. 8 log sheet of a modified staircase test 3. results and discussion 3.1. dynamic stress the correlation between the exciting acceleration amplitude and the strain amplitude on the surface of test specimens in roi is shown in fig. 9. with each case of the exciting acceleration amplitude, there is a corresponding graph that shows the change of strain over time. it can be seen that, in the sweep sine mode, the largest strain response is achieved when the exciting frequency of the lds shaker coincides with the natural frequency of the samples. fig. 10 shows the dynamic stress at different exciting acceleration amplitudes. this dynamic stress-vibration acceleration correlation can be approximately estimated by eq. (6): σ = 2.1049 × a + 47.953 (6) where a is the amplitude of vibration acceleration (m/s 2 ); σ is the principal stress on the surface of test specimens in roi (mpa). it is revealed that the dynamic stress linearly increases with the amplitude of vibration acceleration. fig. 9 correlation between strain and amplitude of vibration acceleration fig. 10 dependence of stress on exciting acceleration 3.2. residual stress and fatigue limit the results of measuring residual stress according to astm e837-13a include a variety of stress components. to compare and match with the results of dynamic stress measurement, the residual stress components will be chosen as the first principal stress. the residual stress of the samples after vsr is shown in fig. 11. it is noted that the samples without vsr are considered the ones with the vibration amplitude of zero. it is clear that the residual stress will be reduced when the amplitude of exciting acceleration increases. according to the graph, the residual stress in the samples without using vsr is 216 mpa. in all the cases using vsr, the residual stress is reduced by more than 36%. in particular, the maximum residual stress reduction can be attained by up to 73% when the 62 advances in technology innovation, vol. 7, no. 1, 2022, pp. 56-65 vibration amplitude reaches 57 m/s 2 . this is corresponding to the total stress exceeding 30% of the ys of the material. in contrast, when the amplitude of vibration acceleration exceeds 57 m/s 2 , the residual stress will increase as the vibration amplitude increases. this may be explained by the intensive deformation of the samples. the process of determining the fatigue limit of sample groups is conducted according to the modified staircase method and is shown in fig. 7. the experimental data is processed, and the final results of the fatigue limit are given in fig. 12. the experimental results of the fatigue limit demonstrate that the fatigue limit of samples is generally increased with the vsr treatment. the fatigue limit of test specimens increases as the vibration acceleration amplitude increases from 0 to 57 m/s 2 , and it reaches a peak of 14% when the vibratory exciting acceleration amplitude is 57 m/s 2 . after that, if the vibratory exciting acceleration amplitude continues increasing, the fatigue limit will gradually decrease. the result is in good agreement with the studies on tensile fatigue characteristics after vsr [19, 21]. it is obvious from the tests that, if the amplitude of the vibration acceleration is very high, the fatigue limit of specimens will be decreased to the value even lower than the value of the samples without vsr. in general, if the vibration amplitude in vsr is too high or the vsr time is too long, micro voids will be formed and specimens’ fatigue limit will be lowered. fig. 11 residual stress in post-vibrating samples with different vibration acceleration amplitudes fig. 12 fatigue limit of sample groups according to vibration acceleration amplitudes 4. conclusions in this study, the vsr effects on the residual stress and fatigue limit of samples are investigated. the key findings of this research are described as follows: (1) the dynamic stress generated through vsr has a linear relationship with the amplitude of vibratory exciting acceleration. at the amplitude of vibratory exciting acceleration lower than 57 m/s 2 , the residual stress is reduced with the increase in the amplitude of vibratory exciting acceleration. the maximum principal residual stress is reduced about 73% when the amplitude of vibratory exciting acceleration is equal to 57 m/s 2 . as the amplitude of vibratory exciting acceleration exceeds 57 m/s 2 , the residual stress will increase. however, it is generally still lower than those in the cases without using vsr. (2) vsr also has an effect on the fatigue limit of the samples made from low-carbon steel with residual stress. vsr can improve the fatigue limit by up to 14% for the samples under residual stress. however, if the vibrating samples are with very high amplitude of acceleration, their fatigue limit will be reduced to the values that are lower than those without vsr. (3) when the vibrating test samples are with moderate acceleration amplitude, vsr is able to both reduce the residual stress and increase the fatigue limit. if the amplitude of vibration acceleration is very high, this effect will be possibly reversed. the residual stress will be improved less, and the fatigue limit can be lower than the cases without vsr. 63 advances in technology innovation, vol. 7, no. 1, 2022, pp. 56-65 symbols and abbreviated terms symbol unit definition vsr vibratory stress relief ys mpa yield strength σ mpa stress ε m/m strain e mpa modulus of elasticity σu mpa material tensile strength σ1 mpa the first stress amplitude level σi mpa the i th stress level σ0 mpa the lowest stress level (i = 0) σr mpa fatigue limit d mpa predefined difference in applied stress levels a m/s 2 amplitude of vibration acceleration conflicts of interest the authors declare no conflict of interest. references [1] j. s. robinson, m. s. hossain, and c. e. truman, “residual stresses in the aluminum alloy 2014a subject to pag quenching and vibratory stress relief,” the journal of strain analysis for engineering design, in press. [2] q. wu, d. p. li, and y. d. zhang, “detecting milling deformation in 7075 aluminum alloy aeronautical monolithic components using the quasi-symmetric machining method,” metals, vol. 6, no. 4, 80, april 2016. [3] m. j. vardanjani, m. ghayour, and r. m. homami, “analysis of the vibrational stress relief for reducing the residual stresses caused by machining,” experimental techniques, vol. 40, no. 2, pp. 705-713, april 2016. [4] h. sasahara, “the effect on fatigue life of residual stress and surface hardness resulting from different cutting conditions of 0.45% c steel,” international journal of machine tools and manufacture, vol. 45, no. 2, pp. 131-136, february 2005. [5] m. benedetti, v. fontanari, p. scardi, c. a. ricardo, and m. bandini, “reverse bending fatigue of shot peened 7075-t651 aluminum alloy: the role of residual stress relaxation,” international journal of fatigue, vol. 31, no. 8-9, pp. 1225-1236, 2009. [6] m. s. patil, r. r. wayakole, and k. d. sarode, “vibratory residual stress relief in manufacturing—a review”, international journal of engineering sciences and research technology, vol. 6, no. 5, pp. 609-613, may 2017. [7] h. song, h. gao, q. wu, and y. zhang, “effects of segmented thermal-vibration stress relief process on residual stresses, mechanical properties and microstructures of large 2219 al alloy rings,” journal of alloys and compounds, vol. 886, 161269, december 2021. [8] c. a. walker, a. j. waddell, and d. j. johnston, “vibratory stress relief—an investigation of underlying process,” proc. of the institution of mechanical engineers, part e: journal of process mechanical engineering, vol. 209, no. 1, pp. 51-58, february 1995. [9] g. cai, y. huang, and y. huang, “operating principal of vibratory stress relief device using coupled lateral-torsional resonance,” international journal of vibroengineering, vol. 19, no. 6, pp. 4083-4097, 2017. [10] m. c. sun, y. h. sun, and r. k. wang, “the vibratory stress relief of a marine shafting of 35# bar steel,” materials letters, vol. 58, no. 3-4, pp. 299-303, january 2004. [11] d. rao, j. ge, and l. chen, “vibratory stress relief in manufacturing the rails of a maglev system,” journal of manufacturing science and engineering, vol. 126, no. 2, pp. 388-391, may 2004. [12] d. rao, d. wang, l. chen, and c. ni, “the effectiveness evaluation of 314l stainless steel vibratory stress relief by dynamic stress,” international journal of fatigue, vol. 29, no. 1, pp. 192-196, january 2007. [13] h. gong, y. sun, y. liu, y. wu, y. he, x. sun, et al., “effect of vibration stress relief on the shape stability of aluminum alloy 7075 thin-walled parts,” metals, vol. 9, no. 1, 27, january 2019. 64 advances in technology innovation, vol. 7, no. 1, 2022, pp. 56-65 [14] c. walker, “a theoretical review of the operation of vibratory stress relief with particular reference to the stabilization of large-scale fabrications,” proc. of the institution of mechanical engineers, part l: journal of materials: design and applications, vol. 225, no. 3, pp. 195-204, july 2011. [15] a. m. fayrushin, r. r. chernyatyeva, and d. n. yakovleva, “the effects of vibration treatment in the process of welding on the structure of metal of seam weld,” iop conference series: earth and environmental science, vol. 459, 062110, 2020. [16] h. gao, s. wu, q. wu, b. li, z. gao, y. zhang, et al., “experimental and simulation investigation on thermal-vibratory stress relief process for 7075 aluminum alloy,” materials and design, vol. 195, 108954, october 2020. [17] d. x. fang, f. h. sun, z. k. gong, a. x. jia, and y. a. qu, “improving fatigue life of welded components by vibratory stress relief technique,” international journal experiment mechanical, vol. 6, pp. 89-95, 1991. [18] a. munsi, a. j. waddell, and c. walker, “the influence of vibratory treatment on the fatigue life of welds: a comparison with thermal stress relief,” strain, vol. 37, no. 4, pp. 141-149, 2001. [19] d. djuric, r. vallant, k. kerschbaumer, and n. enzinger, “vibration stress relief treatment of welded high-strength martensitic steel,” welding in the world, vol. 55, no. 1, pp. 86-93, 2011. [20] h. j. gao, y. d. zhang, q. wu, and j. song, “experimental investigation on the fatigue life of ti-6al-4v treated by vibratory stress relief,” metals, vol. 7, no. 5, 158, may 2017. [21] j. song and y. zhang, “effect of vibratory stress relief on fatigue life of aluminum alloy 7075-t651,” advances in mechanical engineering, vol. 8, no. 6, pp. 1-9, 2016. [22] h. gao, y. zhang, q. wua, j. song, and k. wen, “fatigue life of 7075-t651 aluminum alloy treated with vibratory stress relief,” international journal of fatigue, vol. 108, pp. 62-67, march 2018. [23] standard test method for determining residual stresses by the hole-drilling strain-gage method, astm e837-13a, 2013. [24] m. kuroda, a. sakida, n. oguma, m. nakagawa, t. matsumura, and t. sakai, “historical review on origin and application to metal fatigue of probit and staircase method and their future prospects,” journal of the society of materials science, vol. 70, no. 3, pp. 221-228, march 2021. [25] c. müller, m. wächter, r. masendorf, and a. esderts, “accuracy of fatigue limits estimated by the staircase method using different evaluation techniques,” international journal of fatigue, vol. 100, pp. 296-307, july 2017. [26] h. xue, r. r. li, l. x. wu, w. j. peng, and j. q. huang, “the fatigue performance evaluation of pressure vessel steel using modified staircase method,” advanced materials research, vol. 989, pp. 879-882, 2014. [27] i. m. w. ekaputra, r. t. dewa, g. d. haryadi, and s. j. kim, “fatigue strength analysis of s34mnv steel by accelerated staircase test,” open engineering, vol. 10, no. 1, pp. 394-400, 2020. [28] international council on combustion engines working group 4 (cimac wg4), “iacs ur m53, appendix iv, guidance for evaluation of fatigue tests,” report, october 16, 2009. [29] strength calculation and testing: methods of fatigue strength behavior calculation, gost 25.504-82, 1982. (in russian) copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). 65 microsoft word 3-v8n2(2023)-aiti#8364(111-120).docx advances in technology innovation, vol. 8, no. 2, 2023, pp. 111-120 short-term rainfall prediction using supervised machine learning nusrat jahan prottasha1,*, anik tahabilder2, md kowsher3, md shanon mia1, khadiza tul kobra1 1department of computer science, daffodil international university, dhaka, bangladesh 2department of computer science, wayne state university, detroit, michigan, usa 3department of computer science, stevens institute of technology, hoboken, new jersey, usa received 30 august 2021; received in revised form 16 may 2022; accepted 02 june 2022 doi: https://doi.org/10.46604/aiti.2023.8364 abstract floods and rain significantly impact the economy of many agricultural countries in the world. early prediction of rain and floods can dramatically help prevent natural disaster damage. this paper presents a machine learning and data-driven method that can accurately predict short-term rainfall. various machine learning classification algorithms have been implemented on an australian weather dataset to train and develop an accurate and reliable model. to choose the best suitable prediction model, diverse machine learning algorithms have been applied for classification as well. eventually, the performance of the models has been compared based on standard performance measurement metrics. the finding shows that the hist gradient boosting classifier has given the highest accuracy of 91%, with a good f1 value and receiver operating characteristic, the area under the curve score. keywords: rain prediction, machine learning, supervised classification, agriculture resource, crops yield 1. introduction agriculture plays a vital role in the development of many developing countries [1]. iot-based smart agriculture model is being implemented worldwide to increase crop yields. the use of intelligent tools in farming can increase the production of crops and also minimize the damage due to disasters. the economy of south asian countries, including bangladesh, india, china, and pakistan, depends more on agriculture. but there are always some natural disasters, including rain and floods, that create huge demolition of crops and property. therefore, a good rain prediction model is necessary to forecast the rain to reduce the risk to life and also to maintain the agriculture farms in a better way. in addition, a rain prediction model helps farmers take early flood measurements and properly manage water resources. observing the significance of rain prediction, researchers have developed a lot of devices to predict rainfall, but none of them is worth noting in terms of short-term rain prediction. hence, it has not been adopted eventually by the end-level user to forecast the rain situation. however, machine learning techniques can make a more accurate prediction because of their underlying technology. researchers have implemented neural networks (nn) in rainfall prediction and showed that the nn-based model usually exceeds the performance of the numerical weather prediction model. this study aims to develop a short-term rain prediction model that can effectively and accurately predict rainfall. in this proposed work, several relevant machine learning models have been used to predict rainfall, and finally, a performance comparison has been made to determine the best suitable model. * corresponding author. e-mail address: jahannusratprotta@gmail.com advances in technology innovation, vol. 8, no. 2, 2023, pp. 111-120 112 in this project, the twenty-nine most optimistic classifiers have been used from eleven different categories. all these models have been trained and tested with a relevant rainfall dataset to implement this prediction model. the data was collected from a popular and recognized public repository and split into training, validation, and testing data. since the raw data came from natural weather resources, it has been preprocessed before going to the training phase. a few preprocessing techniques have been implemented to prepare the raw data, such as missing value check, feature selection, features scaling, dimension reduction, etc. after analyzing all the models and comparing them, it was found that the hist gradient boosting classifier (hgbc) has shown the highest accuracy of 91%. a lot of other models have shown the second-highest accuracy of 90%. the contribution of this paper can be summarized below: (1) a pipeline for estimating rain prediction has been developed. (2) diverse types of classifiers have been used to ensure the best model that suits diverse types of data. (3) a comparison among all trained models has been described to measure the comparative performance. the rest of the sections of this paper is organized as follows. section 2 explains the related work of various classification techniques for rainfall prediction. section 3 describes the major technology components used, including the dataset, preprocessing, and algorithm. section 4 describes the methodology that has been used to solve the proposed problem. later on, section 5 contains the experiments and the results. this article is wrapped up in section 6 by discussing the conclusion and future works. 2. related work a country’s agriculture largely depends on rain, and there is a lot of research on forecasting rain. all the earlier methods of rain forecasting are mainly statistical and numerical analysis based [1]. also, some methods predict the rain by analyzing radar images. denœux and rizand [2] have proposed a model that performs deep learning-based analysis on radar images to predict rainfall. however, with the recent advancement of machine learning, many new machine learning-based models have been proposed for rainfall prediction. researchers like shah et al. [3] have developed a simple polynomial regression-based model to predict the rain to benefit agricultural products. asha et al. [4] have proposed a hybrid machine learning classification model for predicting rainfall, and it has shown better performance than the ordinary ml-based model. sakthivel and thailambal [5] have also demonstrated such a hybrid approach for rain prediction, predicting continuous long period rainfall. naidu et al. [6] presented the changes in rainfall patterns in numerous agro-climatic zones using machine learning approaches. besides, dinh et al. [7] have used a support vector machine (svm)-based method to measure the rain forecast and the soil erosion due to the rain. on the other hand, abdel-kader et al. [8] showed a vigorous hybrid technique by particle swarm optimization (pso) and multi-layer perceptron (mlp) for the prediction of rainfall. also, samsiahsani et al. [9] evaluated many machine learning classifiers based on malaysian data for rainfall prediction. similar models have been developed to predict the flood forecast due to heavy rainfall and ocean waves. luk et al. [10] mentioned data scarcity as a limitation in modeling such a predictive model. abbot and marohasy [11] have shown the application of nn in rainfall prediction based on a dataset from queensland, australia. among the short-term prediction models, a model by shah et al. [12], is a good invention for predicting very short-term rainfall. on the other hand, some researchers focused on heavy rainfall only. research by sangiorgioet al. [13] has made improvements in determining heavy rainfall based on water vapor measurement using a nn-based model. han et al. [14] have mentioned the limitations of such a predictive model for forecasting rainfall and flood by determining the major uncertainties. advances in technology innovation, vol. 8, no. 2, 2023, pp. 111-120 113 unlike those works, in this project, several rain-forecasting models for upcoming rain prediction have been developed using twenty-nine different machine learning classifiers. additionally, their performances have also been compared, and hence the best machine learning model for rain prediction has been determined. 3. dataset and algorithm description the structural dataset and algorithms are a machine learning model’s two most essential parts. this section will provide a brief description of the dataset and algorithm. the features of the dataset will be discussed in more details form. then the data preprocessing steps will also be explained in the subsequent section. finally, all the algorithms that have been used to make the models will be described in brief. 3.1. dataset to implement the proposed model, the rain prediction database from kaggle has been used. the dataset [15] comprises the precipitation estimation from the years 1901 to 2015 for each state of australia. each observation contains 19 qualities (person-months, annual, and combinations of 3 continuous months) for subdivisions. the rain prediction unit mentioned in the dataset is measured in millimeters (mm). table 1 summarizes the dataset and its features. the feature name and the description of the feature are shown in the left column and the right-side column, respectively. the dataset is robust and has an adequate number of observations and features, which ensures the quality of the data and guarantees a model with good accuracy if modeled properly. table 1 the description of the dataset feature details description of the feature location this is the name of the location maxtemp max temperature recorded in degrees centigrade mintemp min temperature recorded in degrees centigrade windgustspeed the speed of wind gust in km per hour windgustdir the direction of the strongest wind gust windspeed9am wind speed averaged over 10 minutes before 9 a.m. winddir9am the direction of the wind gust at 9 a.m. windspeed3pm averaged wind speed over 10 min before 3 p.m. winddir3pm the direction of the wind gust at 3 p.m. humidity9am relative humidity at 9 a.m. measured in percentage humidity3pm relative humidity at 3 p.m. measured in percentage temp9am the temperature at 9 a.m. measured in degrees celsius temp3pm the temperature at 3 p.m. measured in degrees celsius pressure9am atmospheric pressure at 9 a.m. measured in hpa pressure3pm atmospheric pressure at 3 p.m. measured in hpa rainfall the rainfall recorded for the day (mm) raintoday if precipitation in the 24 h to 9 a.m. exceeds 1 mm it will be 1, otherwise 0 (mm) rain tomorrow if it will rain or not on the next day, it is output as binary target variable 3.2. pre-processing data preprocessing is a crucial step that helps improve the quality of data to advance the extraction of important bits of knowledge from the data. it refers to preparing the raw data in a format that will be neat, clean, and understandable to the machine learning model. advances in technology innovation, vol. 8, no. 2, 2023, pp. 111-120 114 the flowchart of data preprocessing steps is illustrated in fig. 1. in the subsequent section. after the data is collected, it goes through a cleaning process, and then the missing values are handled logically. then the data encoding is performed, and important features get selected. eventually feature scaling process is used to bring all the features on a similar scale, and 10-fold cross-validation (cv) is implemented. each stage of data preprocessing is described elaborately below. data cleaning is the method of preparing data for analysis by removing or adjusting incorrect, fragmented, unimportant, duplicated, or improperly organized data. it includes further tuning of data by fixing spelling, and syntax errors, and adjusting mistakes, such as dealing with empty data, invalid values, and recognizing duplicate data points. there were some missing, some duplicates, some invalid, and some incomplete values in this dataset. the following actions have been taken to correct those data. (1) a part of the data points is repeated in row and column segments. hence, all the duplicate data was removed, and only a single instance was kept. (2) a few rows and columns were almost empty. that corresponding row or column has been removed from the dataset. (3) some instances were shown to have some invalid values. that instance was also deleted to make the dataset solid and readable. fig. 1 the entire data preprocessing procedure missing values are a common occurrence in the dataset. therefore, they need to be handled to prepare the dataset properly. if the data cannot pass the statistical test, then it is removed from the dataset. besides, most predictive algorithms can't handle any missing value. hence, this issue must be solved before the data is fed into the model. in most machine learning work, people utilize various techniques such as mean, median, and mode methods to handle missing values. but the most appropriate method for managing missing data is to remove the full row for categorical features and replace the missing data with the nearest neighbors for numerical data. the k-nearest neighbors (knn)-based imputation method has been implemented in this work for a more accurate missing value imputation. the not-a-number (nan) data has been replaced by getting the nearest value by considering the three neighbors. categorical data could be a subjective feature whose values are taken by label encoding or one-hot representation. it implies that categorical data must be encoded into numbers before it is fed to the model. in the dataset, there are six categorical factors “location,” “windgustdir,” “winddir9am,” “winddir3pm,” “raintoday,” and “rain tomorrow.” one hot encoding is one of the most popular methods for encoding categorical variables, which has been used in this project. it is one of the widespread approaches, and it works well unless the categorical variables count is too high. it makes a new binary column for each category, demonstrating the presence of each possible value from the categorical data. feature selection is the strategy to figure out highly related input variables when creating a predictive model. reducing the number of input variables is desirable to reduce the computational cost in modeling and increase the model’s performance. the dataset contains 21 components, and the p-value has been tested to check the probability of the null hypothesis. features advances in technology innovation, vol. 8, no. 2, 2023, pp. 111-120 115 with a p-value less than 0.05 have been discarded. after checking multicollinearity, a vital separation is maintained from those components which appear redundant and don’t back the p-value assumption. besides, to handle the numerical feature, the pearson correlation coefficient has been used, which is characterized within condition-1, and for categorical features, anova-f has been used, which is described as: ( ) ( ) ( ) ( ) 1 2 2 1 1 n i i i n n i i i i x x y y r x x y y = = = − − = − −    (1) after performing the feature selection, eighteen features were kept, including location, mintemp, maxtemp, rainfall, windgustdir, windgustspeed, winddir9am, windspeed9am, windspeed3pm, humidity9am, humidity3pm, pressure9am, pressure3pm, temp9am, temp3pm, raintoday, and rain tomorrow. the numerical data are mostly skewed or nonstandard deviation in data analysis due to outliers, multi distributions, exceptionally exponential distributions, and more. this issue was solved by changing the numeric value into a categorical feature. to implement this method, a discretization process that changes over the numerical value into different distribution work has been applied as shown below: ( ) ( ) ( ) ( ) 2 1 2 1 / 1 / n k g i n k g i n x x k f x x n k = = − − = − −   (2) feature scaling is one of the significant procedures required to standardize the independent features of the working dataset. there are different strategies for feature scaling such as min-max scaling, variance scaling, standardization, mean normalization, unit vectors, etc. in this work, min-max scaling has been used as a feature scaling method, and the exchange range has been set between 0 and 1. mathematical formulla of the min-max scaling has been described as: ( ) ( ) ( ) min max min x x x x x − ′ = − (3) cv is a resampling method to train, test, and validate the model using different data in each iteration. there are lots of cv methods to perform this operation. this paper uses a 10-fold cv technique where the whole dataset is separated into ten folds. each section is used either for training, validation, or test set. after data preprocessing, there were 142193 rows as samples and 17 columns as features. 3.3. algorithms a few of the most suitable machine learning and deep learning-based models have been implemented to build the best model. some of them are neighbor relationship-based, some are naive bayes (nb) theory-based, some are based on svm, and so on. table 2 shows a summary of all the algorithms that have been used in this research. for the knn, five neighbors have been considered to find the similarity. for mlp, the learning rate is 0.001 with two hidden layers of 56 units. for radius learning, the radius value was set to 0.1. the gini index has been used to split decisions between the decision tree and the random forest tree. table 2 the summary of the algorithm used methodology algorithms methodology algorithms neighbors classifier k-nearest neighbors discriminant analysis classifier linear discriminant analysis radius neighbors classifier quadratic discriminant analysis advances in technology innovation, vol. 8, no. 2, 2023, pp. 111-120 116 table 2 the summary of the algorithm used (continued) methodology algorithms methodology algorithms neighbors classifier nearest centroid support vector machine classifiers linear svc ensemble classifiers adaboost classifier linear svc bagging classifier nu svc gradient boosting classifier linear model stochastic gradient descent classifier hist gradient boosting classifier ridge classifier random forest classifier ridge classifier cv naive bayes classifiers bernoulli nb passive aggressive classifier multinomial nb logistic regression cv categorical nb logistic regression complement nb perceptron gaussian nb impact learning [16] semi-supervised classifiers label propagation gaussian process classifiers radial basis function label spreading neural network mlp classifier 4. methodology to complete the whole workflow, a total of four steps have been executed: data collection and preprocessing, training model using supervised learning methods, testing, and performance analysis. a popular and acceptable dataset from the kaggle platform [11] has been adopted for this project. this dataset was split into three parts: the training part, the validation part, and the testing part. after gathering all raw data, the dataset goes through data preprocessing steps, which have been used to make the dataset outliers-free and more solid. these data preprocessing steps also help in increasing the performance of the models [17]. as a result, diverse preprocessing methods such as cleaning data, missing value checks, handling the categorical data, feature selection, and feature scaling have been applied. the machine learning models are made by training with the data preprocessed earlier. the training and testing methods of all those models are different. from all the training methods, 29 classifiers have been used so that the performance can be compared and the best suitable model can be selected. most of the models listed in the table showed good performance, but some didn’t fit very well. the complete methodology of this proposed model is shown in fig. 2. fig. 2 the overview of the methodology of the proposed work 5. experiments and results the model was built and then trained with the preprocessed dataset. in this section, the performances of all the algorithms have been compared. besides, various experimental parameter has been tuned for performance analysis and evaluation. in addition, the experimental setup to accomplish the entire task has also been described. for this model, 11 statistical performance metrics have been considered for performance analysis and comparison. advances in technology innovation, vol. 8, no. 2, 2023, pp. 111-120 117 the whole work has been completed in google colab, and python has been provided as a simulation environment by google. a machine learning framework named sci-kit learns and deep learning framework keras have also been used to implement the classification algorithm. in addition, the matplotlib library for data visualization, graphical representation, and data analysis has been used in this project. table 3 the summary of the algorithms that have been used in this research model accuracy f1 score rs ps fbs hl js mc auc bac cks neighbors classifier knc 0.868 0.752 0.672 0.749 0.809 0.202 0.604 0.468 0.749 0.689 0.411 nc 0.813 0.719 0.663 0.661 0.749 0.257 0.561 0.378 0.74 0.68 0.338 rnc 0.803 0.708 0.654 0.648 0.736 0.267 0.55 0.356 0.731 0.671 0.316 ensemble classifiers adc 0.893 0.802 0.72 0.791 0.856 0.177 0.656 0.559 0.797 0.737 0.507 bc 0.894 0.802 0.72 0.793 0.857 0.176 0.657 0.561 0.797 0.737 0.508 gbc 0.904 0.818 0.734 0.812 0.875 0.166 0.675 0.594 0.811 0.751 0.54 hgbc 0.91 0.833 0.751 0.817 0.885 0.16 0.692 0.619 0.828 0.768 0.569 rfc 0.907 0.823 0.737 0.819 0.881 0.163 0.68 0.604 0.814 0.754 0.549 naive bayes classifiers mnb 0.811 0.718 0.662 0.659 0.747 0.259 0.56 0.376 0.739 0.679 0.335 conb 0.733 0.686 0.679 0.63 0.712 0.337 0.512 0.36 0.756 0.696 0.297 cnb 0.731 0.686 0.683 0.632 0.713 0.339 0.512 0.365 0.76 0.7 0.299 gnb 0.695 0.66 0.669 0.618 0.692 0.375 0.482 0.336 0.746 0.686 0.263 cc 0.899 0.813 0.731 0.799 0.866 0.171 0.668 0.58 0.808 0.748 0.529 semi-supervised classifiers lp 0.884 0.787 0.712 0.758 0.832 0.186 0.64 0.521 0.789 0.729 0.476 ls 0.902 0.816 0.733 0.808 0.872 0.168 0.673 0.589 0.81 0.75 0.536 discriminant analysis classifier lda 0.901 0.817 0.736 0.802 0.869 0.169 0.673 0.588 0.813 0.753 0.537 qda 0.708 0.656 0.641 0.603 0.684 0.362 0.483 0.294 0.718 0.658 0.237 svm classifiers lsvc 0.902 0.812 0.727 0.811 0.872 0.168 0.669 0.586 0.804 0.744 0.529 nusvc 0.899 0.809 0.726 0.802 0.865 0.171 0.665 0.576 0.803 0.743 0.522 sgdc 0.894 0.795 0.709 0.8 0.857 0.176 0.65 0.555 0.786 0.726 0.496 rdc 0.9 0.802 0.714 0.813 0.867 0.17 0.658 0.572 0.791 0.731 0.511 rdcv 0.9 0.802 0.714 0.813 0.867 0.17 0.659 0.572 0.791 0.731 0.511 pac 0.85 0.623 0.568 0.767 0.709 0.22 0.506 0.322 0.645 0.585 0.202 lrcv 0.9 0.815 0.733 0.801 0.868 0.17 0.671 0.584 0.81 0.75 0.533 lr 0.901 0.815 0.733 0.802 0.869 0.169 0.672 0.585 0.81 0.75 0.534 pr 0.804 0.716 0.666 0.653 0.742 0.266 0.556 0.373 0.743 0.683 0.332 il 0.902 0.813 0.727 0.811 0.872 0.168 0.669 0.586 0.804 0.744 0.53 gaussian process classifiers gpc 0.903 0.818 0.736 0.807 0.873 0.167 0.675 0.592 0.813 0.753 0.54 neural network classifier mlpc 0.906 0.833 0.757 0.805 0.879 0.164 0.691 0.613 0.834 0.774 0.568 advances in technology innovation, vol. 8, no. 2, 2023, pp. 111-120 118 here, the twenty-nine most suitable machine learning models have been used to predict the rainfall possibility. a total of 11 statistical measurements have been considered and listed in table 3. the table shows that the hgbc has predicted the best accuracy of 0.91, and the f1 score is 0.833. the random forest tree classifier has obtained the best accuracy of 0.907 with an f1 score is 0.823 from the branch of ensemble classifiers. the mlpc has acquired good accuracy, which is 0.906, along with an f1 score of 0.833 from the section of the nn classifier. moreover, from the section on the neighbor’s classifier, it can be noticed that the knc has shown the best accuracy of 0.868, and the f1 score is 0.752. also, from the section on nb algorithms, cc has shown the best accuracy of 0.899 and its f1 score is 0.813 among all nb classifiers. after that, ls also has placed the best accuracy of 0.902, and its f1 score is 0.816 from the branch of semi-supervised classifiers. besides, it can also be seen from the discriminant analysis section that the linear discriminant analysis has proved the best position of accuracy, 0.901, with an f1 score of 0.817. fig. 3 comparison of accuracy among the models next, lsvc also has the best accuracy of 0.902, and its f1 score is 0.812 from the branch of svm classifiers. finally, gpc has figured out the best accuracy result is 0.903 with an f1 score of 0.818 from the section of gaussian process classifiers. all-inclusive, by analyzing all the sections of algorithms for the rain prediction model, it came out that the hgbc is the winner with an accuracy of 0.91, and the f1 score is 0.833. fig. 3 shows the accuracy comparison among all the models. a roc curve is a graph showing the performance of a classification model at all classification thresholds. most of the algorithms show a high chance that the classifier will be able to distinguish the positive class values from the negative class values. 6. conclusion and future work in this work, a machine learning model has been presented that can determine whether it will rain or not on the very next day. real data from australia has been adopted from the kaggle platform and implemented in the model. this data-driven model is more accurate than any other statistical or numerical-based model. the primary purpose here is to find the best classifiers for predicting rainfall. for this reason, various machine learning classifiers have been implemented. eventually, the most significant performance metrics have been compared, including accuracy, f1 scores, roc, auc, and hgbc have shown the best accuracy with a good score. all the models used here have been trained based on the dataset of a particular zone. therefore, rain prediction accuracy may vary depending on the dataset characteristics. training this model with a dataset that has a sample collected from diverse places can make this model more general and suitable for all locations in this world. this proposed model will be able to forecast the rain for the short-term, specifically for the next day. a new model can be built that will predict the rain for the long term in the future. combining this two may build a complete solution for rain prediction. moreover, this machine learning-based model may not be easily useable for general people. therefore, mobile and advances in technology innovation, vol. 8, no. 2, 2023, pp. 111-120 119 computer apps can be built so those general people can easily use them. a deep learning and nn model approach can be used to improve the result. undoubtedly, there is a plan to evaluate the other country’s data for forecasting the rain using this model. thus, this model is expected to be a universal and easy rain prediction tool for ordinary people. nomenclature hgbc hist gradient boosting classifier bc bagging classifier roc receiver operating characteristic gbc gradient boosting classifier auc area under the curve rfc random forest classifier nn neural networks mnb multinomial naive bayes svm support vector machine conb complement naive bayes pso particle swarm optimization cnb categorical naive bayes classifier mlp multi-layer perceptron gnb gaussian naive bayes mm measured in millimeters cc calibration classifier cv cross-validation lp label propagation knn k-nearest neighbors ls label spreading nan not-a-number lda linear discriminant analysis nb naive bayes qda quadratic discriminant analysis rs recall score lsvc linear support vector classifier ps precision score nusvc nu support vector classification fbs f-beta score sgdc stochastic gradient descent classifier hl hamming loss rdc ridge classifier js jaccard score rdcv ridge classifier cv mc matthew’s correlation pac passive aggressive classifier bac balanced accuracy lrcv logistic regression cv cks cohen’s kappa lr logistic regression knc k neighbors’ classifier pr perceptron classifier nc nearest centroid il impact learning rnc radius neighbor’s classifier gpc gaussian process classifier adc adaboost classifier mlpc multi-layer perceptron classifier conflicts of interest the authors declare no conflict of interest. references [1] k. t. sohn, j. h. lee, and s. h. lee, “statistical prediction of heavy rain in south korea,” advances in atmospheric sciences, vol. 22, no. 5, pp. 703-710, 2005. [2] t. denœux and p. rizand, “analysis of radar images for rainfall forecasting using neural networks,” neural computing and applications, vol. 3, no. 1, pp. 50-61, march 1995. [3] b. k. shah, s. thapa, r. s. diyali, s. hk, and s. maharjan, “rain prediction using polynomial regression for the field of agriculture prediction for karnatakka,” international journal of advances in engineering and management, vol. 2, no. 3, pp. 62-66, march 2020. [4] p. asha, a. jesudoss, s. prince mary, k. v. sai sandeep, and k. harsha vardha, “an efficient hybrid machine learning classifier for rainfall prediction,” journal of physics: conference series, vol. 1770, no. 1, article no. 012012, march 2021. [5] s. sakthivel and g. thailambal, “effective procedure to predict rainfall conditions using hybrid machine learning strategies,” turkish journal of computer and mathematics education, vol. 12, no. 6, pp. 209-216, april 2021. advances in technology innovation, vol. 8, no. 2, 2023, pp. 111-120 120 [6] d. naidu, b. majhi, and s. k. chandniha, “development of rainfall prediction models using machine learning approaches for different agro-climatic zones,” handbook of research on automated feature engineering and advanced applications in data science, igi global, 2021. [7] t. v. dinh, h. nguyen, x. l. tran, and n. d. hoang, “predicting rainfall-induced soil erosion based on a hybridization of adaptive differential evolution and support vector machine classification,” mathematical problems in engineering, vol. 2021, article no. 6647829, 2021. [8] h. abdel-kader, m. abd-el salam, and m. mohamed, “hybrid machine learning model for rainfall forecasting,” journal of intelligent systems and internet of things, vol. 1, no. 1, pp. 5-12, 2021. [9] n. samsiahsani, i. shlash, m. hassan, a. hadi, and m. aliff, “enhancing malaysia rainfall prediction using classification techniques,” journal of applied environmental and biological sciences, vol. 7, no. 2s, pp. 20-29, april 2017. [10] k. c. luk, j. e. ball, and a. sharma, “an application of artificial neural networks for rainfall forecasting,” mathematical and computer modeling, vol. 33, no. 6-7, pp. 683-693, march 2001. [11] j. abbot and j. marohasy, “application of artificial neural networks to rainfall forecasting in queensland, australia,” advances in atmospheric sciences, vol. 29, no. 4, pp. 717-730, june 2012. [12] c. shah, c. hendahewa, and r. gonzalez-ibanez, “rain or shine? forecasting search process performance in exploratory search tasks,” journal of the association for information science and technology, vol. 67, no. 7, pp. 1607-1623, july 2016. [13] m. sangiorgio, s. barindelli, r. biondi, e. solazzo, e. realini, g. venuti, et al., “improved extreme rainfall events forecasting using neural networks and water vapor measures,” 6th international conference on time series and forecasting, pp. 820-826, september 2019. [14] d. han, t. kwong, and s. li, “uncertainties in real-time flood forecasting with neural networks,” hydrological processes: an international journal, vol. 21, no. 2, pp. 223-228, january 2007. [15] j. young, “rain in australia,” https://www.kaggle.com/jsphyg/weather-dataset-rattle-package, october 30, 2007. [16] m. kowsher, a. tahabilder, and s. a. murad, “impact-learning: a robust machine learning algorithm,” proceedings of the 8th international conference on computer and communications management, pp. 9-13, july 2020. [17] c. v. z. zelaya, “towards explaining the effects of data preprocessing on machine learning,” ieee 35th international conference on data engineering, pp. 2086-2090, april 2019. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). 6___aiti#8504___143-154 advances in technology innovation, vol. 7, no. 2, 2022, pp. 143-154 low access latency and high throughput for full-duplex cognitive radio using mobility caching placement strategy madhunala srilatha 1,* , anantha bharathi 2 1 department of electronics and communication engineering, vardhaman college of engineering, hyderabad, india 2 department of electronics and communication engineering, university college of engineering, osmania university, hyderabad, india received 18 september 2021; received in revised form 11 november 2021; accepted 12 november 2021 doi: https://doi.org/10.46604/2022.8504 abstract in a remote framework, the densely allotted radio range is not effectively utilized by an authorized primary user, which limits a secondary user to avoid disturbance to a primary user while utilizing a channel. to solve the shortage problem of accessible range, this study proposes a mobility caching placement strategy (mcps) based sliding window scheme to optimize the latency and throughput of cognitive radio system. the performance parameters of the proposed system are analyzed using energy detection metrics by considering full-duplex communication. by the matlab analysis and the evaluation with and without self-interference suppression, the results demonstrate that the entrance inactivity of the mcps based sliding window is diminished by a factor of 1.9 as contrasted with the existing full-duplex systems. with the improved latency and throughput, the proposed method can avoid the ruinous impacts of drawn-out self-residual suppression more effectively compared to the existing approaches. keywords: full-duplex cognitive radio, latency, mobility caching placement strategy, self-residual suppression 1. introduction one of the biggest difficulties in remote framework today is the shortage of accessible range, which can be resolved by using one of the best multiple access schemes, i.e., non-orthogonal multiple access (noma) [1]. nonetheless, ongoing investigations have demonstrated that while the radio range is thickly allotted, it is frequently not intensely involved or utilized by an authorized primary user (pu) [2]. a recurrence coordinated cognitive radio (cr) system is proposed to exploit this circumstance while permitting an unlicensed secondary user (su) to entrepreneurially reuse authorized recurrence groups without creating interference to the authorized user. one of the major necessities of such a framework is that su ought not to produce destructive obstruction to pu [3-4]. therefore, su’s handset must be equipped for detecting the radio channel to decide whether pu is available or not by checking various performance parameters [5]. for range re-utilization, it requires low-inertness medium access for high-need users or high-need transmission. in this framework, a low latency-sensitive continuous transmission requires to be on hold for a split second as pu powerfully gets to the channel for getting its pressing message over. such situations are of specific significance for the ongoing administration, for example, virtual/increased reality and independent vehicles. therefore, it is required to get a latency of less than 1 msec to meet the forthcoming 5g systems [6]. javed et al. [7] have given various methods to sense the pu spectrum with its pros and cons. framework models that intermittently stop the su transmission to detect the channel have been broadly proposed to recognize the beginning of the pu transmission and given various performance parameters’ comparison. these methodologies present visually impaired interims, where the su framework is transmitting and becomes incapable of recognizing the beginning of the pu * corresponding author. e-mail address: ma_srilatha2@yahoo.co.in tel.: +91-9290885384 advances in technology innovation, vol. 7, no. 2, 2022, pp. 143-154 transmission. diminishing the interim between progressive detecting spaces will decrease the proficiency and throughput of the su framework; however, this will enhance its range detecting abilities. to solve this problem, khalid et al. [8] and singh et al. [9] described the use of a full-duplex scheme that senses the spectrum and transmits the data simultaneously, and alves et al. [10] provides protection to the user’s spectrum. irio et al. [11] and gaspard et al. [12] developed a method to expand the throughput of remote correspondence framework, where the remote terminal is permitted to transmit and sense at the same time in the same recurrence band. the throughput can also be increased by increasing the detection probability [13]. tiwari et al. [14] proposed a cooperative medium access control (mac) protocol called mlcp, which improves the throughput even more by enabling simultaneous transmission of pu and su. kumar et al. [15] proposed a clustering algorithm along with an energy efficient power allocation scheme due to which huge improvement in throughput is achieved. however, sustaining the improvement under imperfect self-interference cancellation is not addressed. mokhtarzadeh et al. [16] proposed a method to increase throughput for a full-duplex cr system. previous research has analyzed the exchange of detection which requires further improvement to detect proficiency. the best way to enhance the identification of pu is by utilizing full-duplex cr under self-residual suppression (srs), which senses the channel for every �� test. this makes high access inertness as pu may begin the transmission whenever possible and can identify the nearness of pu more rapidly. ayesha et al. [17] and ayesha et al. [18] have described the alternative approaches of simultaneously sensing and transmission, using srs for improving the trade-off between sensing the spectrum and su throughput. however, the faultless srs cannot be achieved below the noise floor. in the existing work, the performance analysis of spectrum sensing is done by using various schemes and protocols, e.g., multi-objective modified grey wolf optimization algorithm [19], noma network under power splitting based simultaneous wireless information and power transfer (swipt) [20], noma based mobile edge computing with imperfect channel state information (csi) [21], decode-and-forward (df) and amplify-and-forward (af) techniques in noma based cognitive radio network (crn) [22], signal segmentation [23], neural network based multilayer perception model [24], and the channel allocation scheme based on greedy algorithm [25]. cache technology is used to even improve the performance of parameters of spectrum sensing which includes various strategies: cache capacity allocation, cache replacement, cache utilization, and cache placement [26]. in the work of zheng et al. [27] and xia et al. [28], the performance improvement of parameters is analyzed including energy consumption optimization, system transmission performance, and latency. in this study, a mobility caching placement strategy (mcps) based sliding window scheme is proposed. this method significantly reduces access latency for the full-duplex system under srs to protect pu. additionally, it improves the su throughput which maximizes the data transmission. this study emphasizes the issue of compromising the performance parameters of spectrum sensing (such as throughput and access latency) in the full-duplex cr under srs. with different schemes, e.g., half-duplex, full-duplex using slotted window, and full-duplex using sliding window, the performance of parameters is evaluated with and without srs. to optimize the sensing parameters, a mcps based sliding window is proposed. compared to the existing methods, the proposed method will exchange more information bits by taking decisions based on each sample, resulting in a quick detection of pu, in turn lowering the latency, and improving the su throughput. the study is organized as follows. the system model and frame structure of the existing methods and the proposed method is described in section 2. in section 3, the mcps based sliding window scheme is discussed. the performance analysis of latency and throughput is analyzed in section 4. analytical and simulation results are presented in section 5. finally, section 6 concludes the study. 2. system model while su is using the spectrum of pu, no interference in the pu network should occur. to ensure this, su needs to accurately identify the white spaces for the transmission and should vacate and opt for the subsequent band when pu appears. the 144 advances in technology innovation, vol. 7, no. 2, 2022, pp. 143-154 first scenario is considered hypothesis h1. when pu is using the spectrum, the signal received at su is expressed in eq. (1). the second scenario is considered hypothesis h0. while pu is not using the spectrum, the received signal can be expressed in eq. (2). ( ) ( ) ( )s n x n u n= + (1) ( ) ( )s n u n= (2) in the above equation, ���� is the signal which needs to be identified, ���� is the gaussian noise which is additive white in nature, and n is the sample index. the parameters of performance using detection algorithm will be analyzed based on two probabilities: the detection probability (pd) with hypothesis h1 for identifying the presence of licensed user activity and the false alarm probability (pf) with hypothesis h0 for detecting the licensed user when the licensed user is actually not present. pu is effectively protected when the probability of detection is high. in another scenario, su can use the frequency band recurrently when the false alarm probability is low and is available. to guarantee better performance of the system, the detection probability shall be as much as possible, and the false alarm probability shall be as minimum as possible. cr system schemes are half-duplex and full-duplex, which are addressed in this section. the presence of detection of pu is done by using a metric m, which is compared with a given threshold � for both schemes. pu is present if �, otherwise pu is absent and su can use the channel. energy detection is the frequently used metric for sensing the spectrum [29]. ( ) 2 1 1 s n ns m s n n = = ∑ (3) where ���� is the su received signal. if the pu information is known prior, other detection methods can be used. using decision metric, the detection probability pd and the false alarm probability pf can be evaluated with the help of a hypothesis test [7]. ( )0/rfp p m h= > γ (4) ( )1/rdp p m h= < γ (5) the cr system frame structure is shown in fig. 1. initially, the cr system has to sense the spectrum and determine the status of pu. if the frequency band of pu is detected as idle, then the su transmitter will make use of the band to transmit information to the su receiver for the frame duration. if pu starts transmission, then su has to cease its transmission, to avoid interference to pu. again, in the next frame, su will access the frequency band and the process repeats. fig. 1 frame structure of cr system under srs 145 advances in technology innovation, vol. 7, no. 2, 2022, pp. 143-154 2.1. half-duplex cr system in this scheme, spectrum sensing has to be scheduled sequentially with transmission, as shown in fig. 2. the frame contains two parts: � �� sample window for transmission and �� sample window for sensing the frequency band. �� samples are used to know the status of pu using eq. (1) and thus decisions will be made for every � sample. it is clear that under the half-duplex scheme, su has to cease the transmission each time the spectrum sensing is established, which leads to the higher access latency of pu and the reduction of su throughput by a factor of ���� � . fig. 2 frame structure of half-duplex cr system 2.2. simultaneous detection and transmission with slotted window scheme the frame structure of simultaneous detection and transmission with a slotted window using srs is shown in fig. 3 [16]. it contains a single frame, in which both the sensing and transmission will be performed simultaneously to lead to the improvement in detection probability and the decisions made for every ��. in this scheme, as there is no separate sensing slot, the su throughput will not reduce, and su can detect the status of pu immediately which leads to a reduction of latency. fig. 3 frame structure of slotted window full-duplex cr system 2.3. sliding window scheme for simultaneous detection and transmission the frame structure of simultaneous detection and transmission with a sliding window using srs is shown in fig. 4, where the decisions will be made for every sample compared to the slotted window (for every ��). this scheme can be actualized effortlessly in computerized equipment employing a first-in first-out (fifo) cushion. it is imperative to take a note that progressive choice measurements are not free as just a single new sample is included (and one is expelled). fig. 4 frame structure of sliding window full-duplex cr system 2.4. quadrature phase shift keying the major advantages of digital modulation techniques are the improved capacity to transmit huge data with high immunity and the easy detection of its distinct transmission state at the receiver in a noisy medium. while converting analogue signals to digital signals, a trade-off is always made due to the loss of some information in the quantization process. proper selection of the digital modulation technique is crucial especially when the bandwidth and time slots are limited in 146 advances in technology innovation, vol. 7, no. 2, 2022, pp. 143-154 uplink-downlink transmission. the efficacy of a modulation technique is restrained in terms of its bandwidth and power efficiency. the capability of a modulation technique in accommodating data within a limited bandwidth is regarded as bandwidth efficiency whereas the ability to preserve the bit error probability of the digital message at low power levels is treated as the power efficiency of a modulation technique. with regards to the digital modulation schemes, for the same number of bits transmitted and received, binary phase shift keying (bpsk) modulation has less bit error rate and bandwidth for different values of signal-to-noise ratio (snr), as compared to binary amplitude shift keying (bask) and binary frequency shift keying (bfsk) modulation. to detect the message at the output, minimum snr and low bit error rate with higher performance is required. with low power consumption, bpsk modulation technique provides the best performance even for less snr. bandwidth can be saved even more in m-ary phase shift keying (m-psk) modulation schemes by maintaining symbol rate constantly and increasing the number of bits per symbol. for example, in quadrature phase shift keying (qpsk), one of the 4 possible phase shifts per symbol can be transmitted rather than transmitting only 2 possible phase shifts so that 2 bits per symbol (i.e., for the same bandwidth with twice the information) can be transmitted [30]. also, with the same energy efficiency, one can achieve double spectral efficiency. higher compound m-psk modulation schemes can also be possible (8-psk, 32psk, 64psk, etc.) to transmit more bits per symbol and in turn achieve higher bit rate. however, this employs higher vulnerability to noise as the symbols together get the closer and increased bit errors. in order to make a trade-off between bit rate and probability of error, qpsk is opted in the proposed system. 3. mcps based sliding window scheme caching techniques are attracted expressively since they can decrease backhaul traffic and also improve performance metrics. caching placement strategies can be of uncoded or coded type. both are used for optimizing the performance of parameters. the mcps scheme allows decisions to be taken on a multiple sample transfer at a time and increases the concurrent transmission and sensing using intra interference suppression, which checks the presence of pu continuously in a spectrum band and gives a perfect position. the operation of mcps is shown in fig. 5 and the flowchart is shown in fig. 6. the number of input samples and coding rate is calculated by considering snr, number of symbols present at the transmitter, and threshold value using eqs. (6) and (7). ( ) ( )1 ..... 2 x bs bs n r snr n snr n= × × + + (6) / ( 10 )tx snr n i r h thr n= × + × (7) where n represents the number of samples, h is the analysis of h equation, snr represents the signal-to-noise ratio, rx is the receiver, thr represents the threshold value, ri represents the coding rate, and ntx represents the number of symbols at the transmitter. a signal is passed through the qpsk modulator, then the modulated data is analyzed, processed using eqs. (8)-(10), and then transmitted through the channel. ( ) ( ) ( ) ( )1 i bs bs iipq r w l n k w l n r k = × − × × × − ×φ φ (8) ( ) ( ) bs tx w l ipq w l n r = × × (9) 147 advances in technology innovation, vol. 7, no. 2, 2022, pp. 143-154 ( ) ( )ipq w s w l ifft= × (10) where ���� represents the phase of the signal, ���� is the line width, ipq represents the quadrature input, rtx represents the random information at transmitter, and ��s� is the sample width. parameter optimization is done by using eqs. (11)-(17). ( ) ( ) ( ) ( ) 2 s i tx txt bs bs w l qp np c r n n j w l r w l n= × × × − × × × × (11) ( ) ( ) ( ) ( )itt r w s qp ifft qp r w l = × × (12) ( ) ( ) tx t tsqp r m qp conj qp w s = × ×  ∑ (13) ( ) ( )( )v p txt r t rp q max qp conj qp r = × −   (14) ( )( ) ( ) ( ) tx tsrn t r r m qp conj qp w s = × −∑ (15) sqp srn m m gain = (16) ( )1010 logmse gain= × (17) fig. 5 operation of mcps fig. 6 flowchart of mcps 148 advances in technology innovation, vol. 7, no. 2, 2022, pp. 143-154 4. performance analysis 4.1. analysis of latency and throughput in mcps based sliding window the impact of detecting time on the throughput is examined in this section. in each edge of the frame structure span tf, su starts sensing the channel for a period of sensing time (��), and if the channel is free (which indicates pu is not using the channel), then su utilizes the channel (spectrum) for information transmission. on the other hand, if pu wants to use the channel, then su has to evacuate the channel and use another channel that is free for transmission without creating interference to pu. if the length of the corresponding edge of pu is tprf, then the throughput of su that could be transmitted when pu is not utilizing the spectrum is tprf + tf. the throughput (th) of su is given by: ( )1sf cf f t t th p p t −= − (18) at the point, when pu has an exponential on-off movement display, with the mean spans of “on” period signified by β, then pc is given by: exp prf fc t t p       + = − β (19) expecting that the casing of span tprf, tf, and false caution likelihood has been settled, then the standardized throughput of su is: ( )1 expsf prf ff f t t t t th p t       − + = − − β (20) for the case of h0 (i.e., pu is available but wrongly detected), the probability density function (pdf) of the su received signal (s) is p0(x), and the probability of false alarms pf is: ( ) ( )0 0/fp p s h p x dx ∞ = > = ∫ ε ε (21) the detection threshold ɛ is calculated as: ( )2 1 1u f sq p f− ε = σ τ +  (22) on the other hand, when pu is using the channel (for the case of h1), the pdf of the su received signal (s) is p1(x), and the detection probability pd is: ( ) ( )1 1/dp s h p x dx ∞ = > = ∫ ε ε (23) the latency (l) of the system is calculated as: ( )         ∩∩ ++         ∩ +         += − c i cc i iccc ddd d pn dd d pn d d pndpnl 12121 3 3 1 2 211 ... ..... (24) where ni is the number of samples when pu begins with i th decision point, di denotes the event when pu is detected at i th decision once transmission starts, and ��� represents the complementary event when pu is not detected during ith decision. 149 advances in technology innovation, vol. 7, no. 2, 2022, pp. 143-154 4.2. algorithm the algorithm of mcps based on latency and throughput optimization is described in table 1, where the input parameters need to be specified first and then the eqs. (8)-(10) are analyzed. furthermore, the optimization of latency and throughput are calculated based on the analytical values. table 1 algorithm of mcps input: ��, ���, ���, ��, ��, � !, �!# output: return the estimated data in the qpsk domain [var] begin for i = 1 to $!�% &' &( j = 1 to $!�% &' &( )*ℎ,� , �. = ���0!, 1!, ��, ��� for k = 1…………. $!�% &' using 8-10 in var �� parameter optimization learn (latency variance) δ3� update (var name) 43� using � → �� + 1� or �� 1� throughput = ∑ �9:�;:���3;:��<=> �? = @(ab% c c d� ��efghi end 5. analytical and simulation results this section explains the analytical and simulation results of the proposed mcps technique under perfect srs and imperfect srs, as compared with the existing methods. simulation results show the improvement in performance parameters by using the proposed technique. simulation parameters of the mcps scheme are shown in table 2. table 2 simulation parameters parameter value channel length 128 number of bits 10 5 number of sub-carriers 128 number of transmitters 2 number of receivers 2 fast fourier transform size 64 sensing window size 16 5.1. perfect self-interference suppression fig. 7 depicts the substantial improvement of latency and throughput with the mcps based sliding full-duplex method compared to the existing methods under perfect srs. since half-duplex allows either sensing or transmission at a time and cannot perform both tasks at a time, the latency of this method is almost half of a su frame length and throughput is also very less. unlike half-duplex, the slotted full-duplex method allows both sensing and transmission at a time, and the latency of this method has been improved for a duration of ��/2 compared to half-duplex since, for each �� duration, there is a provision of taking decisions. as the decisions are taken for every sample, the sliding full-duplex method can detect the usage of the channel by pu quickly. furthermore, compared to the sliding full-duplex, the latency of mcps is improved by a factor of 0.98; compared to the slotted full-duplex, the mcps latency is improved by a factor of 3.7; compared to half-duplex, the mcps latency is improved by a factor of 20. fig. 8 illustrates the different quantiles of optimization parameter trade-off for the proposed mcps technique and the existing sliding and slotted techniques. the performance of parameters depends on self-residual, noise, and signal realizations. the results show the improvement of the proposed technique compared to the existing techniques. 150 advances in technology innovation, vol. 7, no. 2, 2022, pp. 143-154 fig. 7 latency versus throughput comparison for all the 4 methods under perfect srs fig. 8 quantiles of optimization parameter trade-off under perfect srs 5.2. imperfect self-interference suppression fig. 9 depicts latency improvement for mcps with an increase in srs, measured with respect to the noise floor which is calculated by considering throughput to 0.9. when compared with the existing methods, the slotted full-duplex method has higher latency for all values of self-interference, but the proposed method can withstand the improvement of self-interference by a 3.2 factor. fig. 10 depicts the throughput improvement for mcps with increasing srs. the mcps method has improved the throughput under the pu protection. table 3 shows the comparison of the parameters of the sliding full-duplex and mcps methods. fig. 9 latency versus srs fig. 10 throughput with increasing srs for the latency of 16 samples table 3 comparison of parameters parameters sliding window full-duplex mcps no. of transmitters (tx) 2 2 no. of receivers (rx) 2 2 channel length 64 up to 128 number of bits 10 4 10 5 channel size 52 52 max latency 0.90 0.79 throughput 0.78 0.89 error rate 0.44 0.42 snr 0.97 0.99 151 advances in technology innovation, vol. 7, no. 2, 2022, pp. 143-154 6. conclusions and future scope this study investigated cr from the point of view of ensuring the essential and high priority users under perfect srs and imperfect srs by diminishing the latency of the framework. by inferring systematic recipes for latency and throughput of the existing methods, the proposed method is beneficial in terms of its capacity to make choices at each sample. additionally, the results of simulation show that the proposed sensing framework can achieve a huge improvement in latency and throughput under srs, which is diminished by a factor of 1.9 contrasted with the current full-duplex system. the performance parameters can be further extended by using vehicular ad-hoc networks, which provide cooperation in cr and share the channel status while switching between the channels when pu exists. furthermore, the parameters can be extended by using a testbed that tests the performance and practicability of cr framework. list of acronyms noma non-orthogonal multiple access snr signal-to-noise ratio pu primary user bask binary amplitude shift keying su secondary user bfsk binary frequency shift keying cr cognitive radio m-psk m-ary phase shift keying mac medium access control rx receiver srs self-residual suppression ntx number of symbols at transmitter swipt simultaneous wireless information and power transfer qpsk quadrature phase shift keying csi channel state information ���� the phase of the signal df decode-and-forward ���� line width af amplify-and-forward ipq quadrature input crn cognitive radio network rtx random information at transmitter mcps mobility caching placement strategy ��s� sample width pd probability of detection mse mean square error pf probability of false alarm tf frame structure span fifo first-in first-out ɛ detection threshold bpsk binary phase shift keying th throughput of su �� sensing time β mean span of “on” period �� transmission time tprf length of the corresponding edge of pu ���� detected signal pdf probability density function ���� additive white gaussian noise l latency m decision metric ni number of samples at i th decision point � threshold di event when pu detected at ith decision ���� received signal ��� complimentary event when pu is not detected during ith decision n number of input samples conflicts of interest the authors declare no conflict of interest. references [1] a. kumar and k. kumar, “multiple access schemes for cognitive radio networks: a survey,” physical communication, vol. 38, 100953, february 2020. [2] federal communications commission, “spectrum policy task force report”, federal communications commission, technical report fcc 02-155, november 15, 2002. 152 advances in technology innovation, vol. 7, no. 2, 2022, pp. 143-154 [3] j. tiwari, a. prakash, and r. tripathi, “a novel cooperative mac protocol for safety applications in cognitive radio enabled vehicular ad-hoc networks,” vehicular communications, vol. 29, 100336, june 2021. [4] o. salameh, h. bruneel, and s. wittevrongel, “performance evaluation of cognitive radio networks with imperfect spectrum sensing and bursty primary user traffic,” mathematical problems in engineering, vol. 2020, 4102046, 2020. [5] a. el rharras, m. saber, a. chehri, r. saadane, n. hakem, and g. jeon, “optimization of spectrum utilization parameters in cognitive radio using genetic algorithm,” procedia computer science, vol. 176, pp. 2466-2475, october 2020. [6] b. kodavati and m. ramarakula, “latency based re-enforcement learning over cognitive software defined 5g networks,” 7th international conference on advanced computing and communication systems, pp. 1848-1853, march 2021. [7] j. n. javed, m. khalil, and a. shabbir, “a survey on cognitive radio spectrum sensing: classifications and performance comparison,” international conference on innovative computing, pp. 1-8, november 2019. [8] w. khalid and h. yu, “sensing and utilization of spectrum with cooperation interference for full-duplex cognitive radio networks,” 11th international conference on ubiquitous and future networks, pp. 598-600, july 2019. [9] v. singh, a. gadre, and s. kumar, “full-duplex radios: are we there yet?” proc. of 19th acm workshop on hot topics in networks, pp. 117-124, november 2020. [10] h. alves, t. riihonen, and h. a. suraweera, full-duplex communications for future wireless networks, singapore: springer, 2020. [11] l. irio, a. t. abusabah, and r. oliveira, “residual self-interference estimation in in-band full-duplex wireless systems,” international wireless communications and mobile computing, pp. 1130-1134, june 2020. [12] g. gaspard and d. s. kim, “optimal sensing and interference suppression in 5g cognitive radio networks,” journal of communications, vol. 15, no. 4, pp. 303-308, april 2020. [13] c. m. wu, y. c. kao, and k. f. chang, “a multichannel mac protocol for iot-enabled cognitive radio ad hoc networks,” advances in technology innovation, vol. 5, no. 1, pp. 45-55, december 2019. [14] j. tiwari, a. prakash, and r. tripathi, “a multichannel link-layer cooperation protocol (mlcp) for cognitive radio ad hoc network,” international conference on vlsi, communication, and signal processing, pp. 191-200, october 2019. [15] a. kumar and k. kumar, “energy-efficient resource optimization using game theory in hybrid noma assisted cognitive radio networks,” physical communication, vol. 47, 101382, august 2021. [16] h. mokhtarzadeh, a. taherpour, a. taherpour, and s. gazor, “throughput maximization in energy limited full-duplex cognitive radio networks,” ieee transactions on communications, vol. 67, no. 8, pp. 5287-5296, august 2019. [17] a. ayesha and s. m. chaudhry, “self-interference cancellation for full-duplex radio transceivers using extended kalman filter,” national academy science letters, vol. 43, no. 7, pp. 631-634, february 2020. [18] a. ayesha, m. rahman, a. haider, and s. m. chaudhry, “on self-interference cancellation and non-idealities suppression in full-duplex radio transceivers,” mathematics, vol. 9, no. 12, 1434, june 2021. [19] g. eappen and t. shankar, “multi-objective modified grey wolf optimization algorithm for efficient spectrum sensing in the cognitive radio network,” arabian journal for science and engineering, vol. 46, no. 4, pp. 3115-3145, november 2020. [20] v. aswathi and a. v. babu, “full/half duplex cooperative relaying noma network under power splitting based swipt: performance analysis and optimization,” physical communications, vol. 46, 101335, june 2021. [21] h. jiang, y. wang, x. yue, and x. li, “performance analysis of noma-based mobile edge computing with imperfect csi,” eurasip journal on wireless communications and networking, vol. 2020, 138, july 2020. [22] a. kumar and k. kumar, “relay sharing with df and af techniques in noma assisted cognitive radio networks,” physical communication, vol. 42, 101143, october 2020. [23] k. kirubahini, j. j. triphena, p. g. s. velmurugan, and s. j. thiruvengadam, “optimal spectrum sensing in cognitive radio systems using signal segmentation algorithm,” international conference on wireless communications, signal processing, and networking, pp. 118-121, august 2020. [24] s. dasmahapatra, s. patnaik, s. n. sharan, and m. gupta, “performance analysis of prediction-based spectrum sensing for cognitive radio networks,” 2nd international conference on intelligent communication and computational techniques, pp. 271-274, september 2019. [25] r. li and p. zhu, “spectrum allocation strategies based on qos in cognitive vehicle networks,” ieee access, vol. 8, pp. 99922-99933, may 2020. 153 advances in technology innovation, vol. 7, no. 2, 2022, pp. 143-154 [26] l. li, g. zhao, and r. s. blum, “a survey of caching techniques in cellular networks: research issues and challenges in content placement and delivery strategies,” ieee communications surveys and tutorials, vol. 20, no. 3, pp. 1710-1732, march 2018. [27] x. zheng, g. wang, and q. zhao, “a cache placement strategy with energy consumption optimization in information-centric networking,” future internet, vol. 11, no. 3, 64, march 2019. [28] j. xia, f. zhou, x. lai, h. zhang, h. chen, q. yang, et al., “cache aided decode and forward relaying networks: from the spatial view,” wireless communications and mobile computing, vol. 2018, 5963584, april 2018. [29] h. shehata and t. khattab, “energy detection spectrum sensing in full-duplex cognitive radio: the practical case of rician rsi,” ieee transactions on communications, vol. 67, no. 9, pp. 6544-6555, september 2019. [30] n. fatima, s. a. siddiqui, and a. ahmad, “comparative performance analysis of digital modulation schemes with digital audio transmission through awgn channel,” international conference on power electronics, control, and automation, pp. 1-3, november 2019. copyright© by the authors. licensee taeti, taiwan. this article is an open-access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). 154 5___aiti#8909___131-142 advances in technology innovation, vol. 7, no. 2, 2022, pp. 131-142 finite element and neural network based predictive model to determine natural frequency of laminated composite plates with eccentric cutouts under free vibration mohamed rida seba * , said kebdani applied mechanics laboratory, faculty of mechanical engineering, university of science and technology oran mohamed-boudiaf, oran, algeria received 12 november 2021; received in revised form 29 december 2021; accepted 30 december 2021 doi: https://doi.org/10.46604/aiti.2022.8909 abstract this research proposes a predictive model to identify changes in the mechanical and geometrical properties of composite plates with eccentric cutouts based on natural frequency. finite elements (fe) and neural networks are used to develop the model based on machine learning. first, the numerical analysis of free vibration is performed by the fe model on the laminated composite plates with a stacking sequence [0/90]2s under a clamped-free (cfff) boundary condition. the outputs of the fe model (520 configurations) are then utilized to train the artificial neural network (ann) model through the levenberg-marquardt method, and the developed ann model is then used to evaluate the influence of various parameters on the natural frequency. the results show that the changes in the mechanical and geometrical properties of composite plates have impacts on the natural frequency. furthermore, the findings of the ann model are substantially identical to those of the numerical model, with a small margin of error. keywords: artificial neural network, free vibration, finite element model, cutout, natural frequency 1. introduction laminated composite plates have been widely used in structural engineering due to their reduced weight, extended durability, and fatigue resistance. because of these qualities, they are gaining attention in other engineering fields. cutouts are generally used for ventilation, i.e., the passages for cables and fluids. they can have different shapes, e.g., square, circular, oval, or triangular shapes. however, their existence can have a major impact on the vibratory [1-3], static [4-6], and buckling [7-9] behavior of structures. concerning the vibration behavior of structures, pham et al. [10] considered plates that are completely or partially in contact with fluid and analyzed them with isogeometric analysis (iga) for free vibration. pham et al. [11-12] used the finite element (fe) method to investigate the hygro-thermo-mechanical vibration of double-curved and functionally graded porous (fgp) sandwich plates as well as nanoplates made of functionally graded materials (fgm). pham et al. [13] employed the es-mitc3 element to analyze the free vibration of fgp annular-nanoplates with non-uniform thickness. in addition, nguyen et al. [14] used the es-mitc3 element to investigate the free vibration of fgp plates positioned on partially supported elastic foundations (psef). rai [15] also used fe to examine the nonlinear behavior of reinforced concrete (rc) deep beams. pham et al. [16] conducted a monte carlo simulation using fe analysis to evaluate the natural frequency of rc beams. the free vibration of fgp nanoplates lying on a two-parameter elastic media foundation was explored by pham et al. [17]. * corresponding author. e-mail address: mohamedrida.seba@univ-usto.dz advances in technology innovation, vol. 7, no. 2, 2022, pp. 131-142 recently, vibration response has emerged as an essential technique in structural health monitoring (shm) [18]. a change in a structure’s natural frequency is one of the indicators of a change in its mechanical or geometrical properties or of the presence of damage. several researchers have investigated natural frequency to detect composite plate defects, such as delamination [19-20] and cracks [21-22]. nowadays, the influence of cutouts on composite laminates is studied mainly using numerical analysis, with the fe analysis more specifically. sivakumar et al. [23] presented a ritz fe model to analyze the free vibration of laminates with cutouts. ovesy and fazilati [24] proposed two variants of the finite strip method (fsm) for analyzing the free vibration of composite plates with cutouts. venkatachari et al. [25] used the extended fe technique to investigate the impact of environmental factors on the free vibration of structures. boay [1] developed an fe method for calculating the free vibration of symmetric laminated composite plates with a central hole. artificial neural network (ann) is an artificial intelligence technology widely used for prediction in a variety of engineering fields. several studies in different fields are presented here. truong et al. [26] integrated ann with differential evolution (de) to optimize the material distribution of bidirectional functionally graded (bfg) beams in free vibration. yildirim [27] investigated the free vibration of axially functionally graded (afg) and transversely functionally graded (tfg) beams using the ann model. furthermore, for fgm beams with varied gradation orientations and layer counts, the natural frequency was estimated using the fe approach. to analyze functionally graded annular plates under various boundary conditions, jodaei et al. [28] applied both the differential quadrature and ann techniques. tran et al. [29] developed an ann model to forecast the fundamental frequency of fgm plates by fe in a thermal environment using the es-mitc3 element. due to the intricacy of laminated composites with cutouts, artificial intelligence was introduced to detect changes in natural frequency (i.e., mechanical and geometrical changes in the structures). reddy [30] proposed a method for predicting the natural frequency of laminated composite plates using ann under clamped boundary conditions. altabey [31] predicted changes in the natural frequency of plates supported elastically. timchenko and osetrov [32] proposed convolutional neural networks (cnn) for predicting the natural frequency of composite plates. ann was also used to evaluate the environmental effect on the vibrational response of a skew composite laminated sandwich plate [33]. based on the findings of previous studies, this work aims to expand the use of the fe model to analyze the free vibration of nonlinear layered plates with eccentric square cutouts under clamped-free (cfff) boundary conditions. the study of several parameters, such as cutout size ratio (d/a), number of cutouts, modular ratio (e1/e2), length-to-width ratio (a/b), and thickness ratio (h/a), is carried out using the fe software. the results of the fe model are used to develop an ann model to predict natural frequency. the strategy is to use 240 data points to train, test, and validate the developed model. the study is limited to the first two modes of vibration. the influence of cutout size, number, and modular ratio on the natural frequency is then studied in a more general scope. 2. the fe model 2.1. free vibration analysis the free vibration analysis of structures requires the solution of the following equation, called the eigenvalue problem. the frequency of plates can be obtained easily by the solution of the standard characteristic equation: 2 [ ]{ } [ ]{ } {0}k m∏ − ∏ =ω (1) where �∏� denotes the transverse displacement vector, k is the stiffness matrix, and m is the mass matrix. the formula can be directly used to calculate the natural frequency of the plates under free vibration. 132 advances in technology innovation, vol. 7, no. 2, 2022, pp. 131-142 2.2. material the model developed for this study, as shown in fig. 1, is inspired by the elastic composite stacking sequence [0/90]2s considered by sinha et al. [34]. the composite laminate square model is modified by introducing a group of eccentric cutouts. the sizes and number of cutouts, in addition to the dimensions and the modular ratios of composite plates, are all considered parameters. the properties of the material used in this study are listed in table 1. the position of the cutouts according to the x-axis is considered ex/a = 0.125 for all the scenarios. table 1 mechanical properties of the material used in this study material glass fiber-reinforced polymer (gfrp) [34] e1 (n/m 2 ) 16.07 × 10 9 e2 (n/m 2 ) 16.07 × 10 9 g1 (n/m 2 ) 2.81 × 10 9 v12 0.25 density � (kg/m 3 ) 1664 fig. 1 laminated composite plate with dimensions 2.3. fe simulations the simulations are undertaken using abaqus. to determine the natural frequency, the fe method analysis is performed in a cfff configuration on several specimens of laminated plates with cutouts. the size, number, length, thickness, and modular ratios of specimens are all different from one another. 2.4. convergence as a starting point, the simulation convergence with respect to different mesh sizes is verified by calculating the natural frequency for two mode shapes under boundary conditions (cfff), as shown in table 2. the result in terms of mesh density and mode shapes with respect to the chosen mesh size of 2 is represented in fig. 2. table 2 natural frequency in two mode shapes versus mesh size size natural frequency (hz) for laminated composite plates with cutouts [0/90]2s first mode shape second mode shape 4 69.605 26.187 3.5 62.817 26.168 3 56.777 26.156 2.5 52.696 26.145 2 50.318 26.140 1.5 49.119 26.137 (a) mesh density (b) first mode shape of the plate (c) second mode shape of the plate fig. 2 models of fe and mode shapes for laminated composite plates 133 advances in technology innovation, vol. 7, no. 2, 2022, pp. 131-142 2.5. validation of the fe model the numerical model is validated by comparing the natural frequency in hz for cfff plates that have the length of 0.235 m corresponding to the first mode predicted by this study to those provided by sinha et al. [34] with the same parameters. as shown in table 3, the parameters used in this study, especially the size and position of the cutouts, are same as those in the work of sinha et al. [34]. table 3 comparison of the parameters used in this study and in the work of sinha et al. [34] 3. development of the ann model 3.1. artificial intelligence artificial intelligence is considered one of the most important and fastest developing fields in scientific research. this is due to the difficulty of finding solutions to many problems using conventional methods. anns are preferred among all the artificial intelligence types due to many characteristics: good data security throughout the whole network, the ability to function with less information, excellent fault tolerance, and the ability to train a machine using distributed memory and parallel processing capability. 3.2. parameters and numerical data the numerical model built in the previous section is used to acquire a dataset of 260 natural frequency values in each vibration mode. the feed-forward backpropagation network is implemented using matlab. the following entry variables are fed to the input layer: the size ratio (d/a), number of cutouts, modular ratio (e1/e2), length-to-width ratio (a/b), and thickness ratio (h/a) of the plates. the neurons are in the hidden layer, whereas the number of neurons is in the output layer. the number of neurons in the last layer is two, which represents the number of vibration modes. the input and hidden layers are served by tan-sigmoid transfer functions, while the output layer is served by a linear transfer function. table 4 lists the parameters used to construct the dataset. the neural network is designed, developed, and deployed using the neural network toolbox in matlab. table 4 simulation parameters size ratio (d/a) 0.1/0.2/0.3 number of cutouts 1/2/4 modular ratio (e1/e2) 1/2/3/4 length-to-width ratio (a/b) 1/2/1.5 thickness ratio (h/a) 0.012/0.018/0.024 in this study, a multilayer feed-forward neural network (ffnn) is used to determine the natural frequency. it consists of a single input layer, one or more hidden layers, and a single output layer. a neural network with a single hidden layer can handle the most complicated functions. the basic neural network model is denoted by: ( )ij i j i j w x b= +∑ρ ψ (2) fig. 3 depicts a neuron’s schematic structure. parameter [34] this study error position of cutout size of cutout experimental fe model ex/a = ey/b = 0.25 d/a = 0.1 28 26.2 26.13 1.87 d/a = 0.2 32 26.8 26.76 5.24 ex/a = 0.25 and ey/b = 0 d/a = 0.1 24 26.2 26.14 2.14 d/a = 0.2 20 26.9 26.8 6.8 134 advances in technology innovation, vol. 7, no. 2, 2022, pp. 131-142 fig. 3 structure of a neuron �� denotes the set of inputs for every neuron. the collection of outputs for each neuron is indicated by ��, while the bias set for each neuron is given by ��. the weight coefficient �� is multiplied by every input and then summed with a bias b of neurons to generate the net input n, which may be written as: 1 k j j jn w x b = = +∑ (3) the mathematical notation ji corresponds to the input i in neuron j. the net input n is then sent via an active function , which gives the neuron output �. ( )ρ = f n (4) the hyperbolic tangent sigmoid activation function is used in this investigation. the following formula may be used to express it: ψ − − − = + n n n n e e e e (5) as a result, fnnn with a single input layer and one hidden layer in fig. 4 implements the equation below: 2 2 2 1 1 1 2 1 1 1 )( n k i j ij j iia w w x b b = = = +        +  ∑ ∑ψ ψ (6) where � signifies the entire network output. the activation functions of the hidden layer and output layer, accordingly, are represented by � and . k and n denote the number of inputs and neurons in the hidden layer, respectively. � is the neuron’s bias within the output layer. �� is the weight that connects the ith hidden layer resource to the output layer neuron. fig. 4 depicts the conceptual architecture of the model. fig. 4 architecture of the proposed ann predictive model 135 advances in technology innovation, vol. 7, no. 2, 2022, pp. 131-142 3.3. levenberg-marquardt algorithm training neural networks are basically nonlinear least-squares problems that can be solved using a class of nonlinear least-squares algorithms. among them, the levenberg-marquardt approach is regarded as the most efficient algorithm for training anns. it is based on newton’s approach, which was developed for minimizing sums of squares error functions, such as the form below: 1 2 1 ( ) ( ) ( ) ( ) 2 t i n jf x u u x u xx = = =∑ (7) where �� is the error in the n th pattern, u is the vector with elements ��, and � = ( �, , �,……, �) � contains all of the network’s weights. the sum of squared errors function is denoted as: 1 ( ) ( ) ( ) 2 t f w u w u w= (8) newton’s approach is used to maximize a performance index �( ): 1 1n n n nw w a g − + = − (9) 2 ( ) n n w w a f w = ≡∇ (10) ( ) n n w w g f w = ≡∇ (11) where � �( ) is the hessian matrix and ��( ) is the gradient obtained as follows. ( ) 2 ( ) ( ) t f w j w u w=∇ (12) 2 ( ) 2 ( ) ( ) 2 ( ) t f w j w j w s w= +∇ (13) j(w) represents the jacobian matrix: 11 11 11 1 2 12 12 12 1 2 1 1 1 1 2 1 1 1 1 2 2 2 1 1 2 2 ( ) n m m m p p p n p p p pm p n m n n u u u w w w u u u w w w u u u w w w j w u u u w w w u u u w w w u u w = ⋮ ⋮ ⋱ … … ⋱ … … … … … … … ⋱ … … … ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ ∂ 2 n pmu w w                                                … ∂ ∂ (14) where p stands for the number of training patterns and m defines the number of output patterns. 136 advances in technology innovation, vol. 7, no. 2, 2022, pp. 131-142 2 1 ( ) ( ) ( ) n i i is w v w v w = = ∑ ∇ (15) the hessian matrix can be estimated as follows if s(x) is considered to be small. 2 ( ) 2 ( ) ( ) t f w j w j w=∇ (16) substituting eqs. (11) and (15) into eq. (8), the gauss-newton method can be obtained as: 1 )∆ ( ( ()( ) ) t t k k k k kw j w j w j w u w −  = −   (17) the gauss-newton method has a problem in that the matrix may not be invertible. this may be avoided by making the following changes to approximate the hessian matrix: g h i= + µ (18) as a result, the levenberg-marquardt algorithm is expressed as: 1 ∆ ( ( )() ) )( t t k k k k k kw j w j w i j w u w −  = − +   µ (19) where i is the identity matrix and the amount � is referred to as the learning parameter in neural computing. the learning parameter is reduced as the iterative procedure nears its end. the levenberg-marquardt algorithm is used in this study, as in the work of hagan et al. [35] and lv et al. [36]. 3.4. validation of the ann model the development of an ann model begins by feeding all the captured data as “given” inputs and “desired” outputs. the data is then divided into three sets: training, validation, and test sets, with the proportions 70%, 15%, and 15%, respectively. the regression coefficient (r) and the mean squared error (mse) are used to validate the generated model’s appropriateness. for all the data, r and mse are 0.99995 and 0.143, respectively. as the optimum network, a hidden layer with eleven neurons is chosen to minimize mse. table 5 shows the findings. the results predicted by the ann model are extremely close to the numerical ones. this proves that ann can successfully forecast the natural frequency of laminated composite plates with a cutout. to examine the performance of the proposed model even more deeply, a regression analysis of the predicted as well as the numerical results is illustrated in fig. 5. the regression coefficients (r) are calculated to determine the correlation between the ann predicted values and those obtained from the fe model, as shown in fig. 6. they are divided into three sets: the training, validation, and test data sets. the value of the coefficient r is between zero and one. the degree of correlation increases as r tends toward one. the ann structure (5-11-2) is the best one in table 5. the convergence tests of ann results are done with mse and the regression correlation coefficient (r): 2 1 1 ˆ( ) n i i mse y y n = = −∑ (20) 2 2 1 2 1 1 ˆ1 ( ) ( ) n n i i i i i r y y y y = =    = − − −       ∑ ∑ (21) where y is the actual value, �� is the predicted value of y, and �� is the mean value of y. 137 advances in technology innovation, vol. 7, no. 2, 2022, pp. 131-142 fig. 5 performance of ann training (a)training results of a neural network regression (b) validation results of a neural network regression (c) test results of a neural network regression (d) neural network regression in all data sets fig. 6 correlation between the values predicted by the ann model with structure (5-11-2) and by the fe model table 5 mse and regression coefficient according to the number of hidden layer neurons ann structure the performance of training training: r validation: r test: r mean squared error (mse) 5-1-2 40.2782 0.98244 0.98178 0.98124 49.157 5-2-2 11.8044 0.99125 0.99529 0.98987 24.133 5-3-2 11.8873 0.99421 0.99661 0.99225 16.094 5-4-2 12.9589 0.99613 0.99452 0.99575 11..775 5-5-2 6.0636 0.99802 0.99788 0.99750 6.028 5-6-2 8.8178 0.99376 0.99785 0.99353 15.627 5-7-2 9.9689 0.99729 0.99736 0.99609 8.449 5-8-2 0.8028 0.99974 0.99968 0.9990 7.940 5-9-2 1.6912 0.99951 0.99953 0.99905 1.588 5-10-2 0.5544 0.99985 0.99980 0.99988 0.433 5-11-2 0.1218 0.99995 0.99995 0.99995 0.12 5-12-2 0.5212 0.99989 0.99980 0.99967 0.424 5-13-2 0.3645 0.99996 0.99987 0.99979 0.294 5-14-2 0.3186 0.99994 0.99990 0.99982 0.244 5-15-2 0.5560 0.99993 0.99982 0.99981 0.298 138 advances in technology innovation, vol. 7, no. 2, 2022, pp. 131-142 3.5. comparison between the numerical and predicted natural frequency for all data sets figs. 7(a)-(b) provide a comparison between the numerical and the ann-predicted natural frequency of the first mode shapes. in terms of absolute error (ae), the highest values are 2.94% and 1.58% for each mode shape, while the lowest values are 0.03% and 0%. overall, the error is close to zero, which proves the validity and the accuracy of the ann model. it is noteworthy that ae is determined using the following relation to the positive value. ˆae y y= − (22) where y is the actual value, and �� is the predicted value of y. the proposed ann model is then used to investigate the effect of several variables on the natural frequency as a function of both geometrical and mechanical parameters. only one parameter is changed at a time, while all others are maintained constant. the sensitivity of the composite plate characteristics with respect to the cutout parameters is investigated in the following sections. (a) the values of natural frequency in the first mode shape (b) the values of natural frequency in the second mode shape fig. 7 comparison of the natural frequency values between the numerical model and the ann model with structure (5-11-2) in two mode shapes 4. analysis of parameters’ effect 4.1. effect of the modular ratio (e1/e2) the influence of the modular ratio on the natural frequency of the cfff laminate composite plate [0/90]2s is studied with the following characteristics: length-to-width ratio = 1, thickness ratio = 0.012, and the square cutout with a size ratio d/a = 0.1. the change in natural frequency of the two vibration modes is predicted for the following combinations: e1/e2 = 1, 2, 3, and 4. table 6 shows that, for the two mode shapes, the value of the frequency decreases as the modular ratio increases. it is worth noticing that the natural frequency corresponding to the modular ratio of 2 and the ones corresponding to 3 and 4 are very close. the lowest natural frequencies are 20.36 hz and 46.37 hz for each mode, and the highest ones are 26.140 hz and 50.316 hz. table 6 natural frequencies of the laminated square plate with different parameters (the modular ratio, cutout size ratio, and number of cutouts) modular ratio (e1/e2) cutout size ratio (d/a) number of cutouts 1 2 3 4 0.1 0.2 0.3 1 2 4 natural frequency of the first mode vibration 26.14 22.41 21.06 20.36 26.14 26.80 27.98 36.06 37.92 39.46 natural frequency of the second mode vibration 50.31 47.78 46.66 46.37 50.31 49.49 48.16 73.16 73.21 73.66 139 advances in technology innovation, vol. 7, no. 2, 2022, pp. 131-142 4.2. effect of the cutout size table 6 also shows the variation in the first two natural frequencies with respect to three different cutout size ratios: d/a = 0.1, 0.2, and 0.3 in the same laminated plate having length-to-width ratio a/b = 1 and thickness ratio h/b = 0.012. it is obvious that the size of the cutout affects the natural frequencies of the composite plate. according to the results, the natural frequency of the specimen and the size of the cutout are proportional. for the first two modes, when the size of the cutout d/a changes from 0.1 to 0.2, 0.1 to 0.3, and 0.2 to 0.3, the natural frequency increases by 2.54%, 1.64%, and 2.23%, then by 7.03%, 4.26%, and 5.53%. 4.3. effect of the cutout number in this part, the effect of the number of cutouts on natural frequency is investigated. the laminated square plate is composed of eight elastic layers [0/90]2s with cfff support in the borders. the natural frequency of the first two mode shapes of the composite plate is determined using a frequency response study. the square laminate utilized has a length-to-width ratio of a/b = 1 and a thickness ratio of h/b = 0.01, whereas the cutout size ratio equals d/a = 0.1 in all three configurations. the results are displayed in table 6. for both the vibration modes, the number of cutouts has an inverse effect on the natural frequency of the laminated composite plates. according to the results, as the number of cutouts increases, the specimen’s natural frequency decreases by a modest amount. the change does not exceed 0.7% for the second vibration mode. 5. conclusions in this study, a neural-network-based approach is proposed to assess changes in the geometrical and mechanical properties of composite plates through the prediction of natural frequency. the main idea is to simulate the composite model numerically with the fe method and use its output to construct and train a successful ann predictive model. the model uses natural frequency as an indicator. after validation, the ann model is used to identify the changes in structures by the prediction of natural frequency. the effects of the main geometrical and mechanical characteristics (e.g., the cutout size ratio (d/a), the number of cutouts, and the modular ratio (e1/e2)) on the natural frequency were investigated. the findings of this study can be summarized as follows: (1) the constructed ann model agrees with the fe model, indicated by a mean squared error near zero. the greatest ae between the numerical model and the ann forecasting model was found to be 4.92% and 1.75% for the first and the second vibration modes, respectively. (2) the natural frequency of the laminate plates is inversely proportional to the e1/e2 ratio. (3) the natural frequency is inversely proportional to the number of cutouts. (4) the natural frequency increases with the increase in the size of the cutouts. conflicts of interest the authors declare no conflict of interest. references [1] c. g. boay, “free vibration of laminated composite plates with a central circular hole,” composite structures, vol. 35, no. 4, pp. 357-368, august 1996. 140 advances in technology innovation, vol. 7, no. 2, 2022, pp. 131-142 [2] y. kim and j. park, “a theory for the free vibration of a laminated composite rectangular plate with holes in aerospace applications,” composite structures. vol. 251, 112571, november 2020. [3] b. aidi, m. shaat, a. abdelkefi, and s. w. case, “free vibration analysis of cantilever open-hole composite plates,” meccanica, vol. 52, no. 11-12, pp. 2819-2836, september 2017. [4] a. baltaci, m. sarikanat, and h. yildiz, “static stability of laminated composite circular plates with holes using shear deformation theory,” finite elements in analysis and design, vol. 43, no. 11-12, pp. 839-846, august 2007. [5] t. k. paul and k. m. rao, “finite element stress analysis of laminated composite plates containing two circular holes under transverse loading,” computers and structures, vol. 54, no. 4, pp. 671-677, february 1995. [6] w. zhen and c. wanji, “stress analysis of laminated composite plates with a circular hole according to a single-layer higher-order model,” composite structures, vol. 90, no. 2, pp. 122-129, september 2009. [7] e. gungor and o. talman, “the effects of circular cutout on buckling analysis of laminated composite plates using fem,” proc. of the american society for composites: 34th technical conference, pp. 173-179, september 2019. [8] m. a. komur, f. sen, a. ataş, and n. arslan, “buckling analysis of laminated composite plates with an elliptical/circular cutout using fem,” advances in engineering software, vol. 41, no. 2, pp. 161-164, february 2010. [9] a. l. narayana, k. rao, and r. v. kumar, “buckling analysis of rectangular composite plates with rectangular cutout subjected to linearly varying in-plane loading using fem,” sadhana, vol. 39, no. 3, pp. 583-596, june 2014. [10] q. h. pham, p. c. nguyen, v. k. tran, and t. nguyen-thoi, “isogeometric analysis for free vibration of bidirectional functionally graded plates in the fluid medium,” defence technology, in press. [11] q. h. pham, t. t. tran, v. k. tran, p. c. nguyen, t. nguyen-thoi, and a. m. zenkour, “bending and hygro-thermo-mechanical vibration analysis of a functionally graded porous sandwich nanoshell resting on elastic foundation,” mechanics of advanced materials and structures, in press. [12] q. h. pham, v. k. tran, t. t. tran, t. nguyen-thoi, p. c. nguyen, and v. dong, “a nonlocal quasi-3d theory for thermal free vibration analysis of functionally graded material nanoplates resting on elastic foundation,” case studies in thermal engineering, vol. 26, 101170, august 2021. [13] q. h. pham, t. t. tran, v. k.tran, p. c. nguyen, and t. nguyen-tho, “free vibration of functionally graded porous non-uniform thickness annular-nanoplates resting on elastic foundation using es-mitc3 element,” alexandria engineering journal, in press. [14] p. c. nguyen, q. h. pham, t. t. tran, and t. nguyen-thoi, “effects of partially supported elastic foundation on free vibration of fgp plates using es-mitc3 elements,” ain shams engineering journal, in press. [15] p. rai, “non-linear finite element analysis of rc deep beam using cdp model,” advances in technology innovation, vol. 6, no. 1, pp. 1-10, january 2021. [16] p. c. nguyen, t. n. van, and h. t. duy, “stochastic free vibration analysis of beam on elastic foundation with the random field of young’s modulus using finite element method and monte carlo simulation,” 6th international conference on geotechnics, civil engineering, and structures, pp. 499-506, october 2021. [17] q. h. pham, p. c. nguyen, v. k. tran, and t. nguyen-thoi, “finite element analysis for functionally graded porous nano-plates resting on elastic foundation,” steel and composite structures, vol. 41, no. 2, pp. 149-166, october 2021. [18] d. m. amafabia, d. montalvão, o. david-west, and g. haritos, “a review of structural health monitoring techniques as applied to composite structures,” structural durability and health monitoring, vol. 11, no. 2, pp. 91-147, february 2017. [19] d. chakraborty, “artificial neural network based delamination prediction in laminated composites,” materials and design, vol. 26, no. 1, pp. 1-7, february 2005. [20] a. khan, d. k. ko, s. c. lim, and h. s. kim, “structural vibration-based classification and prediction of delamination in smart composite laminates using deep learning neural network,” composites part b: engineering, vol. 161, pp. 586-594, march 2019. [21] s. khatir, d. boutchicha, c. le thanh, h. tran-ngoc, t. n. nguyen, and m. abdel-wahab, “improved ann technique combined with jaya algorithm for crack identification in plates using xiga and experimental analysis,” theoretical and applied fracture mechanics, vol. 107, 102554, june 2020. [22] p. gayathri, k. umesh, and r. ganguli, “effect of matrix cracking and material uncertainty on composite plates,” reliability engineering and system safety, vol. 95, no. 7, pp. 716-728, july 2010. [23] k. sivakumar, n. g. r. iyengar, and k. deb, “free vibration of laminated composite plates with cutout,” journal of sound and vibration, vol. 221, no. 3, pp. 443-470, april 1999. [24] h. r. ovesy and j. fazilati, “buckling and free vibration finite strip analysis of composite plates with cutout based on two different modeling approaches,” composite structures, vol. 94, no. 3, pp. 1250-1258, february 2012. [25] a. venkatachari, s. natarajan, m. haboussi, and m. ganapathi, “environmental effects on the free vibration of curvilinear fibre composite laminates with cutouts,” composites part b: engineering, vol. 88, pp. 131-138, march 2016. 141 advances in technology innovation, vol. 7, no. 2, 2022, pp. 131-142 [26] t. t. truong, s. lee, and j. lee, “an artificial neural network-differential evolution approach for optimization of bidirectional functionally graded beams,” composite structures, vol. 223, 111517, february 2020. [27] s. yildirim, “free vibration of axially or transversely graded beams using finite-element and artificial intelligence,” alexandria engineering journal, in press. [28] a. jodaei, m. jalal, and m. h. yas, “free vibration analysis of functionally graded annular plates by state-space based differential quadrature method and comparative modeling by ann,” composites part b: engineering, vol. 43, no. 2, pp. 340-353, march 2012. [29] t. t. tran, p. c. nguyen, and q. h. pham, “vibration analysis of fgm plates in thermal environment resting on elastic foundation using es-mitc3 element and prediction of ann,” case studies in thermal engineering, vol. 24, 100852, april 2021. [30] m. r. s. reddy, b. s. reddy, v. n. reddy, and s. sreenivasulu, “prediction of natural frequency of laminated composite plates using artificial neural networks,” engineering, vol. 4, no. 6, pp. 329-337, 2012. [31] w. a. altabey, “prediction of natural frequency of basalt fiber reinforced polymer (frp) laminated variable thickness plates with intermediate elastic support using artificial neural networks (anns) method,” journal of vibroengineering, vol. 19, no. 5, pp. 3668-3678, august 2017. [32] g. timchenko and a. osetrov, “prediction of natural frequency of composite plates with non-canonical shape using convolutional neural networks,” ieee khpi week on advanced technology, pp.129-132, november 2020. [33] v. kallannavar, s. kattimani, m. e. m. soudagar, m. a. mujtaba, s. alshahrani, and m. imran, “neural network-based prediction model to investigate the influence of temperature and moisture on vibration characteristics of skew laminated composite sandwich plates,” materials, vol. 14, no. 12, 3170, june 2021. [34] l. sinha, d. das, a. n. nayak, and s. k. sahu, “experimental and numerical study on free vibration characteristics of laminated composite plate with/without cut-out,” composite structures, vol. 256, 113051, january 2021. [35] m. t. hagan and m. b. menhaj, “training feedforward networks with the marquardt algorithm,” ieee transactions on neural networks, vol. 5, no. 6, pp. 989-993, november 1994. [36] c. lv, y. xing, j. zhang, x. na, y. li, t. liu, et al., “levenberg-marquardt backpropagation training of multilayer neural networks for state estimation of a safety critical cyber-physical system,” ieee transactions on industrial informatics, vol. 14, no. 8, pp. 3436-3446, august 2018. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/). 142 microsoft word 4-v8n2(2023)-aiti#9854(121-135).docx advances in technology innovation, vol. 8, no. 2, 2023, pp. 121-135 building information modeling in the architecture and construction industry jerome jordan faz famadico* department of civil engineering, adamson university, manila, philippines received 14 april 2022; received in revised form 19 october 2022; accepted 22 october 2022 doi: https://doi.org/10.46604/aiti.2023.9854 abstract this study aims to investigate the benefits, risks, barriers, and approaches of building information modeling (bim) implementation in the architecture, engineering and construction (aec) industries. descriptive research methods such as surveys and key informant interviews are used to gather data. respondents in the survey come from different aec companies and are selected with a purposive sampling method. descriptive statistics and one-way analysis of variance (anova) are utilized to analyze the data. the narrative analysis method is also performed to validate the research findings through a desk review of secondary data. the result shows that the major benefit of bim is earlier and more accurate design visualization, while the main risk is accountability and control of data entry into the model. moreover, the major barrier to bim implementation is the high acquisition cost, and the most recommended approach is to increase the availability of bim technology. keywords: building information modeling, digitized construction, bim technology, aec industry, bim framework 1. introduction the construction industry is considered one of the most complex and fragmented industries. its management processes are known to be complicated since uncertainties and risks are mostly inevitable in any construction project [1]. according to leeds’ research [2], construction companies confront many challenges in the productivity, profitability, labor, performance, and sustainability of their projects. most of these companies experience a profit reduction and low productivity due to several factors, such as delays in project completion and changes in the design and construction schedules. in addition, inconsistency in generating information also increases the difficulty of project planning, resulting in misinterpretation of plans and misunderstanding among project stakeholders. increasing productivity and efficiency has been the primary goal in the architecture, engineering, and construction (aec) sector. likewise, contractors and designers find ways to eliminate errors, omissions, and changes in project plans as well as designs while managing a vast amount of building information throughout the project life cycle. information management is one of the best approaches to achieving these goals [3]. however, due to a lack of information technology (it) adoption and utilization, information management needs to be improved in the aec sector in developing countries like the philippines. although it is currently being applied in the construction project life cycle, its usage is partial and isolated. it does not support collaboration and coordination between different systems and disciplines. however, over the past three decades, the construction sector has experienced considerable advancements in using information technology to improve design and construction operations [4]. the most promising development during this period is building information modeling (bim) [5]. * corresponding author. e-mail address: jerome.jordan.famadico@adamson.edu.ph advances in technology innovation, vol. 8, no. 2, 2023, pp. 121-135 122 bim is a state-of-the-art technology steadily making its way into various aec industry operations. it comprises the process of creating and application of a computer-generated model to replicate and simulate a facility’s planning, design, building, and operation [6]. bim models are now changing how buildings and infrastructures work. bim has recently attracted increased attention, particularly in the aec sector. many projects have successfully implemented bim because of the perceived significant benefits, such as decreased construction cost and duration, increased design quality, improved field productivity, and reduced conflicts and changes, to name a few [7-8]. in more advanced countries like the usa, japan, australia, and singapore, bim is already mandated for all construction projects, but bim awareness of the construction industry in the philippines still needs to improve. several factors contribute to this slow adoption of bim in the country. aside from financial constraints, a lack of awareness and knowledge of bim makes the philippines lag behind its neighboring countries in digitizing construction. in the philippines and other developing countries, the use of bim in the construction sector is still in its early stages. this lack of widespread acceptance of bim is related to the risks and challenges hindering its effectiveness [8]. most of the construction projects in the country are still entrenched in traditional processes, and fully adopting bim is difficult due to a shortage of skilled manpower and resources to build bim capabilities. more importantly, there is no government mandate to support the implementation of bim in all construction projects. based on the foregoing, this study analyzes the benefits, risks, barriers, and challenges associated with bim and provides recommendations for future bim adoption. the purpose of this study is to contribute to the limited but increasing empirical pieces of literature and studies on bim. moreover, it aims to discuss the practical application of bim in the philippine construction industry, especially the industry stakeholders’ current levels of bim deployment. it also identified the benefits of bim and the problems that come with it. it would be helpful to address the prevailing issues and problems related to bim in the construction industry after providing recommendations and formulating guidelines. the proposed framework of guidelines for adopting and implementing bim is hoped to significantly improve the design, construction, and operation of different infrastructures in the country. 2. literature review 2.1. main concepts of bim for the past decade, the aec sector has been at the forefront of the digital revolution. at this level of development, bim was introduced. bim is a digital tool for communication, collaboration, scheduling, and visualization among project participants across the whole project life cycle [5, 9]. according to azhar et al. [6], a computer-generated model was proposed and used in the bim process to simulate facility planning, design, building, and operation. bim aids the construction sector by improving user productivity and bringing notable advantages throughout the life cycle of a structure, particularly in facilities management, construction, and design. ismail et al. [10] stated that bim’s accurate geometrical representation of building components in a digital form is its most important advantage. in the case study conducted by yan and demian [7], the shortening of project duration was the main advantage of bim deployment while reducing human resources and project cost, sustainability, quality improvement, and creativity came next. 2.2. bim adoption in different countries in recent years, the adoption of bim has significantly increased in several countries worldwide, particularly in many developed nations such as singapore, the usa, and japan. singapore, for instance, is one of the leading countries considering bim as a standard practice. the country regularly conducts bim conferences, workshops, and seminars on the utilization of bim in construction projects [11]. advances in technology innovation, vol. 8, no. 2, 2023, pp. 121-135 123 on the other hand, lorek [12] discussed a brief history of bim and indicated that although bim is not yet mandated in the usa, its adoption may significantly grow in the years to come. in japan, ishizawa et al. [13] analyzed the bim protocols in the country, and they found that the japanese government has introduced the use of bim for almost ten years already. the japanese government issued guidelines for bim adoption to motivate and encourage japanese practitioners to continue bim utilization. the review article by kaneta et al. [14] revealed the problems of bim implementation in japan and the opposition by general contractors and architectural firms. the reason is that clients are often unaware of the value and incentive that bim can bring. most developed countries actively employ bim, while the implementations in developing countries are rare and less advanced than in most developed countries. several pieces of research show how construction firms struggle with limitations such as the socio-economic and technological environment in developing countries. for instance, in india, the construction industry has only a few bim users with limited bim knowledge [10]. although it was acknowledged that bim could address many problems in the construction sector, its adoption is still considered premature. since no specific initiatives coming from the indian government, some private organizations took the initiative to make bim mandatory for a few of their projects [10]. meanwhile, in malaysia, the construction industry development board (cidb) has undertaken several initiatives to improve bim implementation, such as setting up a reference project using bim, assigning committees to monitor bim activities, and conducting bim seminars and workshops [15]. however, despite these initiatives, a low bim adoption in malaysia was still reported [10]. similar experiences were also observed in indonesia. according to ismail et al. [10], indonesians were highly aware of bim execution in the construction industry. nevertheless, the technology used in the country was low due to a lack of bim implementation standards and regulations, which is why only large projects have utilized bim in their design [10]. in south africa, there is a lack of strong support for the implementation of bim technology from governmental agencies. nonetheless, there are also no uniform standards or guidelines for bim implementation. according to a survey conducted by kekana et al. [15], the construction industry in south africa is technologically aware of bim, but its application is still low due to unwillingness to change the traditional methods of practice. furthermore, they pointed out that most of the bim users in organizations were managers or designers only, and they usually utilized bim for particular purposes such as cost estimation, construction simulation, and management [15]. according to korff’s research [16], bim implementation in south africa has slowed down on account of an underfunded public sector, disinterest in the private sector towards bim uptake, and a cheap labor force. in addition, pakistan also experienced a very slow adoption of bim technology compared to other neighboring countries. therefore, some studies were conducted regarding the knowledge and barriers to bim implementation among pakistani construction players [17]. the majority of pakistani respondents claimed that they need more knowledge about bim and the barriers to its implementation include the belief that current practice is still serving them well and the limitation of its adoption in the local market. 2.3. current status of bim adoption in the philippines in the philippines, 1/3 of the construction industry investors started enacting the bim process, especially in developing 3d models with bim software. however, the remaining 2/3 is still unaware of the bim and all its advantages [18]. furthermore, gonzalez [19] stated that only about 20% of the surveying industries utilized bim software for quantity takeoff, and the rest were content with using the traditional way. most of the users of bim in the aec industry are involved in projects globally that need to be submitted in bim formats, and they can compete internationally. work and skill demands in the country at a local level are mostly related to autocad, and neither universities nor colleges have offered bim-based schooling. consequently, several construction companies and owners have not yet engaged in bim advances in technology innovation, vol. 8, no. 2, 2023, pp. 121-135 124 and realized its potential. the cost of hardware and software is also a massive deterrent to updating the latest versions of bim. as a result, local market support for bim is very low [20]. regarding education, some factors restrained the bim courses from being properly included in the engineering curricula, i.e. the high cost of bim software and hardware, a lack of certified mentors and trainers, and a low level of bim awareness [18]. 2.4. synthesis even if most developing countries emphasized a low bim adoption in certain regions, the advantages of this technology have been acknowledged as well. but more crucially, the difficulties or hurdles associated with its implementation must receive serious consideration if the bim capabilities can be properly leveraged to raise its usage. in the past decade, the adoption of bim has improved significantly in the aec industry. however, the implementation of bim in different countries could have been improved due to numerous barriers. alreshidi et al. [21] found that the major barriers hindering companies from implementing bim were resistance to change by senior employees, lack of training, and disintegration of the project team. based on the foregoing review of different literature and studies about bim, there seems a need to further investigate the benefits, risks, and barriers of bim, particularly in developing countries such as the philippines, where bim implementation is still at a very low level. 3. methodology this section describes the research design and methods utilized in the study. additionally, it also covers the tools and techniques employed during the data collection process. it specifically outlines the research locale, sample and sampling techniques, data gathering methods and processes, statistical treatment of data, and limitations of the study. 3.1. research design this study utilized descriptive and survey research methods, which systematically and accurately describe the facts and characteristics of a given population to provide an accurate account of the characteristics of a particular individual, situation, and group. moreover, survey research gathers data from a population to determine the status concerning variables or subjects under investigation, which is to determine the benefits, risks, barriers, and approaches of bim. 3.2. conceptual framework fig. 1 conceptual framework of the study as presented conceptual framework in fig. 1, the inputs of this research include the prevailing benefits, risks, and barriers of bim and the existing approaches in adopting and implementing bim in construction projects. statistical treatment of data is contained in the inputs of the study, which served as the foundation for a thorough examination of research findings, conclusions, and recommendations. the process and methodology section of this study comprise evidence-based empirical advances in technology innovation, vol. 8, no. 2, 2023, pp. 121-135 125 research using a survey questionnaire, desk review or analysis of project documents as secondary data, key informant interviews, and analysis of findings. finally, the output of this study is a management model that would serve as guidelines to the contractors, consultants, owners, and construction industry professionals for adopting and implementing bim processes in construction projects. 3.3. research locale this research was conducted in the cities included in the national capital region (ncr) of the philippines. ncr is the ideal location for this study owing to its diversity and technological advancement. it is also a residence for the largest construction projects, owners, developers, consultancy, and engineering design companies. 3.4. sample and sampling techniques a purposive sampling method was applied in this study, and the data were obtained from respondents from different companies representing contractors, consultants, and their clients in selected cities in ncr, particularly manila, quezon, makati, pasig, taguig, and muntinlupa. purposive sampling was used in this study to select a particular population or expert group that can answer the research questions best. specifically, among the types of purposive sampling methods, the expert sampling method was used since the study requires inputs and opinions from industry experts who are knowledgeable in bim and currently working and involved in bim implementation in their respective companies. a total of 105 respondents, who could answer and return the questionnaires to the researcher, served as the study participants. fig. 2 shows that 31.43% of all the respondents are clients, 35.24% are consultants, and 33.33% are contractors. the sample population is quite equally distributed among the three groups of respondents and represents a mix of private and government respondents. fig. 2 distribution of respondents 3.5. data gathering methods a questionnaire was developed for this study to collect data; answer the research questions; and assess the perception of contractors, consultants, and clients on bim. this questionnaire is made up of five parts. part i contained the demographic profile of the respondents, such as profession, years of using bim, position in the company, and nature of their companies. parts ii, iii, iv, and v include questions regarding the benefits, risks, barriers, and approaches of bim in philippine construction projects. 3.6. data gathering process before administering the survey questionnaires to the target sample respondents, a pilot test was conducted on selected respondents who were experts in bim. the objectives of the pilot test are to determine the validity and reliability of the questions and confirm whether the questionnaires are easy to accomplish and specific to the research questions. in addition to the survey questionnaire, key informant interviews were conducted with a select group of respondents to get first-hand knowledge of the project's background and history to learn more about how their projects adopt and implement bim. client 31.43% consultants 35.24% contractor 33.33% advances in technology innovation, vol. 8, no. 2, 2023, pp. 121-135 126 the important information gathered from the interviews and observations was used as evidence-based information. it is great to help assess and interpret the findings on the current state of bim implementation in various building projects in the philippines. furthermore, project documents were collected, particularly those relevant to the project's commercial, contractual, and design components. these were used to validate and corroborate the study findings during the desk review of secondary data. the narrative analysis method was applied to analyze these secondary data, and the researcher used the aforementioned collected documents to interpret and explain or validate the research findings. 3.7. statistical treatment of data a series of questionnaires were distributed to the intended respondents to achieve the research objectives. the data gathered from the administered questionnaires were tallied, classified, categorized, and evaluated by the study's objectives. percentage and weighted mean were utilized as descriptive statistics in this study. the percentage can be found by: 100% f p n = × (1) where p is the percentage, f is the frequency of responses and n is the total number of cases. the weighted mean can be found by: x f x n =  (2) where �̅ is the computed value of the weighted mean, f is the frequency, x is the unit weight, and n is the total number of respondents. the one-way analysis of variance (anova) was also performed to find out if there were any significant differences in perceptions regarding the benefits, risks, and barriers of bim among all the respondents. anova can be computed as follows: mst f mse = (3) where f is the anova coefficient, mst is the mean sum of squares of treatment, and mse is the mean sum of squares of error. mst can be computed as: 1 sst mst p = − (4) ( ) 2 sst n x x= − (5) where sst is the sum of squares of treatment, p is the total number of population, and n is the total number of samples. lastly, mse can be found by: sse mse n p = − (6) ( )1sse n s= − (7) where sse is the sum of squares of error, s is the standard deviation of the samples, and n is the total number of observations. 3.8. limitations of the study the unit of analysis is based solely on the perception and opinions of the respondents from different aec companies. the first limitation is related to the data used for construction projects which are limited to only five selected cities and advances in technology innovation, vol. 8, no. 2, 2023, pp. 121-135 127 municipalities in the capital region of the philippines. it could only cover part of the area due to difficulty in the retrieval of questionnaires and the unavailability of the respondents. thus, the statistical strength of the total sample is relatively moderate owing to the small sample size used in the analyses. nevertheless, the perceptions and opinions gathered and generated from the three groups of respondents represented the entire study population. 4. results and discussion this section presents the results of the conducted research survey, analysis of gathered data, and interpretation and discussion of research findings. it also includes a proposed research-validated set of guidelines for adopting bim in various construction projects. 4.1. benefits of building information modeling (bim) table 1 benefits of bim benefits of bim mean std. deviation rank interpretation provides more accurate data visualizations at the earlier stages of the design. 4.52 0.71 1 very often design errors, omissions, conflicts, and clashes are detected before construction. 4.51 0.64 2 very often it provides preliminary insight into design problems and presents opportunities for continuous improvement of the design. 4.44 0.71 3 very often proposals are better understood through accurate visualization of plans. 4.42 0.65 4 very often bim can integrate multiple design disciplines into one plan. 4.39 0.75 5 very often average 4.46 0.69 very often based on the results in table 1, earlier and more accurate design visualization is one of the most significant benefits of bim in the philippine aec industry [20, 22]. bim allows the compilation of every discipline of a project into one complete design, including 3d models and detailed floor plans. through this approach, project stakeholders can see or imagine the project in a real-world scenario, a feature that the traditional paper or 2d design plans fail to deliver. moreover, accurate visualization of the plans also leads to a better understanding of project proposals, according to the respondents. the next benefit of bim is the detection of errors, omissions, conflicts, and clashes. according to key-informantinterviews (kiis) and the research experience from the respondents, the bim toolset helps in automatically detecting clashes among building elements, such as electrical conduit or ductwork that run into a beam. by modeling various electrical and mechanical utilities and structural members, clashes are discovered early in the project and thus reducing costly on-site clashes. bim has an earlier insight into design problems as well. since bim can detect design errors and omissions early, engineers and architects can easily modify and improve the design according to the client’s specifications. interestingly, the respondents' top benefits of bim are confirmed in several pieces of research conducted in different countries [7, 20, 22-23]. according to these studies, the main advantages of bim were precise design visualization, cost and time savings, and coordination and collaboration among many disciplines. on account of the early detection of errors, conflicts, and collisions, bim not only can improve visualization and creativity but also reduce cost and time. coordination and collaboration of different design disciplines were also the perceived benefits of bim in their respective countries. although past studies were conducted at different times and settings, the same benefits were observed in bim implementation regardless of the location. advances in technology innovation, vol. 8, no. 2, 2023, pp. 121-135 128 4.2. risks of bim based on table 2, it can be deduced that respondents rated the control of data entry and the responsibility for any errors and inaccuracies as the most critical risks of bim. the risk of controlling data entry into the model and changing a bim model has become a recent issue in large and complex construction projects [24]. designers, consultants, and contractors may change the design according to their preferences without informing or sharing ideas with the entire project team. this would create confusion among the parties involved and cause conflicts and flawed interactions between different trade disciplines. another key concern about bim is the accountability for any inaccuracies and errors in the design. responsibilities are blurred since each party has participated and contributed in the planning stages, design, revisions, and input to the bim model. uncertainty about what aspect of the model each party had contributed to the project would probably lead to confusion over the liability for any mistakes and conflicts in the design. table 2 risks of bim benefits of bim mean std. deviation rank interpretation control in entering the data into the model and accountability for any errors and inaccuracies in it. 3.49 0.98 1 sometimes financial risks due to a large amount of investment in purchasing bim software and training staff. 3.38 0.99 2 sometimes software licensing issues may arise. 3.23 1.08 3 sometimes experienced senior staff who are used to traditional processes may resist changes. 3.08 0.90 4 sometimes unrealistic project scheduling may cause delays in project completion. 3.07 0.93 5 sometimes average 3.25 0.98 sometimes in addition, another most commonly identified risk of bim is related to financial risk which is caused by a large amount of investment and staff training. according to respondents during the kiis, there is no assurance that they can recover the costs spent in purchasing the bim software packages and skills training of their staff, which can be very costly. in the philippines, for instance, the current price for a single licensed revit bim software is around 6,000 to 7,000 us dollars, while the training cost is around 100 to 200 us dollars per person. according to ham et al. [25], the percentage of bim investment to the whole project cost ranges from 0.00% to 0.91%. this can impose much financial burden on a company, especially start-ups and small-scale construction and design firms. resistance to change and delay due to unrealistic project scheduling are also identified respectively as the top four and five most important risks of bim. it is inherent to most people to resist any changes to the status quo and traditional practices. khalil et al. [23] indicated that it is challenging to change people’s behaviors once they have grown accustomed to certain ways and patterns. that is why introducing new technology such as bim may face opposition and rejection from senior staff and thus fail its implementation. respondents are also concerned about the reliability of the scheduling capabilities of bim. this issue was also raised in the study of wang and chien [26], wherein they reported that aec professionals need to adapt to the new bim-based technique management process to eliminate the risk of delay caused by unrealistic scheduling. the degree of occurrence is quite different since the identified five major risks of bim are sometimes happening according to the data. it is observed that since bim is just in its early stages of implementation in the country, the identified risks still need to be fully realized by project stakeholders. the respondents are more aware of the immediate benefits of implementing bim than the risks attributed to it. advances in technology innovation, vol. 8, no. 2, 2023, pp. 121-135 129 4.3. barriers to bim as shown in table 3, interesting findings can be observed from the processed data that almost all respondents agreed that the top three most important barriers are related to the costs needed to implement bim in various construction projects fully. these costs are identified as costs derived from the acquisition of bim software licenses and skills training of staff. with limited funding and budget, especially for government projects and small and medium-sized enterprises (smes), these identified costs are regarded as barriers that hinder companies from implementing or adopting bim in their projects. table 3 barriers to bim benefits of bim mean std. deviation rank interpretation expensive bim software licenses. 4.25 0.90 1 very often high cost of personnel training and bim implementation. 3.86 0.96 2 often limited project funding and budget. 3.68 1.00 3 often lack of qualified and skilled professionals to handle or operate bim tools and software. 3.64 1.00 4 often expensive human-based services costs. 3.63 0.86 5 often average 3.81 0.94 often according to the respondents, it is too risky to invest much money into something that still needs to be established or proven effective in improving construction processes. however, based on the research experience, the cost is not the primary cause of the slow adoption of bim; it is the need for more support and the willingness of project stakeholders to utilize this new technology. this is supported by kouide et al. [27], who also found that the interest and willingness of project managers and engineers play an important role in bim implementation apart from costs. therefore, khalil et al. [23] suggested that to ensure the success of bim implementation, the government or private clients must impose it in the contract through mandatory legislation and supervision. the next barriers to bim are related to manpower, especially the need for more trained professionals and expensive human-based services costs such as salary for skilled bim modelers. the lack of trained or skilled professionals is one of the major factors which hinder the full implementation of bim [23]. only a few professionals know or have undergone proper training and education regarding bim. this situation may be because only a few students graduating from engineering or architecture programs have the skills and knowledge about bim, and most colleges and universities in the country have not yet incorporated bim in their curricula. according to sabongi [28], the possible reason why bim is not yet included in the curriculum is that there is no room for new courses in the existing curriculum since there is a limit to the number of courses per degree program. there is already an increasing demand for bim modelers but with limited trained professionals; thus, the cost of hiring such skilled manpower resources also increases. consequently, most respondents agreed that the high cost of hiring skilled professionals hinders companies from adopting bim. 4.4. approaches to implementing bim based on table 4, respondents suggested that the aec industry must increase the availability of bim technologies and software to implement bim. interestingly, even though respondents find the acquisition of bim software quite expensive, they still chose this approach to start adopting bim in their projects. in other words, if they are given sufficient funds and budget, they would invest in bim tools and software to establish bim in their respective companies. besides, another best approach towards successful bim implementation is the establishment of bim project execution guidelines and suggesting ways to move from traditional practices into bim. this approach will greatly help companies with little or no idea how they would start implementing bim. advances in technology innovation, vol. 8, no. 2, 2023, pp. 121-135 130 table 4 approaches to implementing bim approaches to implementing bim mean std. deviation rank interpretation bim technology should be made more widely available. 4.10 0.82 1 often develop bim project execution guidelines to facilitate bim implementation 4.01 0.89 2 often increase bim technology research in higher learning institutions. 3.99 1.02 3 often establish feasible ways how to move from traditional practice into bim. 3.98 0.81 4 often organize bim workshops to raise awareness among stakeholders. 3.96 1.08 5 often average 4.01 0.93 often another approach identified by the respondents is to increase research for bim technology in colleges and universities. azhar [22] suggested that bim must be included in the curricula of engineering and architecture programs to increase bim awareness among students, and the education and training of bim software should be implemented. additionally, research in the field of bim must also be conducted to continuously improve bim processes and come up with innovations in bim technology. finally, the respondents agree that to increase awareness regarding bim, workshops, seminars, and training must be conducted. this approach addresses the problem regarding the need for more awareness and knowledge about bim. 4.5. differences in perception about the benefits, risks, and barriers of bim between clients, consultants, and contractors a situation is hypothesized that there is no significant difference between the client, consultant, and contractor’s responses about the benefits, risks, and barriers of bim in various construction projects in the philippines. to test this hypothesis, the following tables show the results of the one-way anova on the differences in perceptions among the respondents. to see if there is a significant difference in the respondents’ responses, a 0.05 level of confidence is employed. the results in table 5 revealed significant differences in the perceptions of clients, consultants, and contractors on one of the five identified benefits of bim, which has something to do with the high-level customization of building systems (p = 0.030). the significant differences in perceptions indicated that respondents disagree on how often this identified benefit happens or is experienced in construction projects. table 5 differences in perception between client, consultant, and contractor regarding the benefits of bim benefits of bim sum of squares df mean square f sig. project coordination and collaboration between groups 0.27 2.00 0.13 0.38 0.686 within groups 36.17 102.00 0.35 total 36.44 104.00 constructability assessment and risk aversion between groups 0.50 2.00 0.25 1.10 0.337 within groups 23.10 102.00 0.23 total 23.59 104.00 high-level customization of building systems between groups 2.06 2.00 1.03 3.62 *0.030 within groups 29.03 102.00 0.28 total 31.09 104.00 optimization of project schedule and cost estimates between groups 3.32 2.00 1.66 2.97 0.056 within groups 57.04 102.00 0.56 total 60.36 104.00 improved project documentation between groups 1.51 2.00 0.76 1.48 0.233 within groups 52.18 102.00 0.51 total 53.69 104.00 *significant at 0.05 level of confidence advances in technology innovation, vol. 8, no. 2, 2023, pp. 121-135 131 from the foregoing exposition, it is safe to assume that the disagreement between clients, consultants, and contractors on high-level customization as one of the major benefits of bim is due to the differences in the scope of work and contract conditions that the two groups of respondents are engaging with. consultants are generally more engaged in revising plans and designs, while clients suggest changes in the plans. consultants may seem to regard bim as beneficial when it comes to customization, but clients think otherwise, as revealed by their low mean scores. although bim enables consultants to change and customize complex designs more easily, this would mean additional costs for clients. the same findings were observed in the study presented by akdag and maqsood [29]. they found that clients typically do not support the use of bim for their projects due to a lack of knowledge and hesitancy to invest in a novel and innovative technology on the market. as to the risks of bim, the data revealed significant differences in perceptions about the risks related to data integrity and security (p = 0.016), as shown in table 6. consultants or designers, being the ones who mostly use bim software to design and model structures, are more aware of the risks of possible loss and mishandling of bim data and inaccuracies of the data entered into the model. on the other hand, clients seldom use bim software and pass on to consultants or designers the task of modeling the design they want. as a result, clients mostly do not experience these risks and are unaware of them [29]. table 6 differences in perception between client, consultant, and contractor regarding the risks of bim benefits of bim sum of squares df mean square f sig. legal risks between groups 1.31 2.00 0.66 1.09 0.341 within groups 61.60 102.00 0.60 total 62.91 104.00 data integrity and security between groups 4.17 2.00 2.09 4.31 *0.016 within groups 49.39 102.00 0.48 total 53.56 104.00 project-related risks between groups 1.26 2.00 0.63 1.43 0.245 within groups 44.98 102.00 0.44 total 46.24 104.00 financial risks between groups 1.79 2.00 0.90 1.32 0.272 within groups 69.20 102.00 0.68 total 70.99 104.00 *significant at 0.05 level of confidence table 7 differences in perception between client, consultant, and contractor regarding the barriers of bim benefits of bim sum of squares df mean square f sig. social-organizational barriers between groups 1.01 2.00 0.51 1.13 0.327 within groups 45.85 102.00 0.45 total 46.86 104.00 financial-related barriers between groups 3.15 2.00 1.58 2.45 0.091 within groups 65.56 102.00 0.64 total 68.71 104.00 technical-related barriers between groups 0.13 2.00 0.07 0.15 0.863 within groups 45.35 102.00 0.44 total 45.48 104.00 contractual-related barriers between groups 1.20 2.00 0.60 0.89 0.415 within groups 69.19 102.00 0.68 total 70.40 104.00 legal-related barriers between groups 0.20 2.00 0.10 0.16 0.854 within groups 63.53 102.00 0.62 total 63.72 104.00 *significant at 0.05 level of confidence advances in technology innovation, vol. 8, no. 2, 2023, pp. 121-135 132 concerning the barriers, the data in table 7 reveal no significant differences in the perception of all the identified barriers of bim. these findings suggest that clients, consultants, and contractors agreed on the identified barriers of bim in the philippine aec industry. 4.6. proposed framework of guidelines for implementing bim fig. 3 bim implementation guidelines advances in technology innovation, vol. 8, no. 2, 2023, pp. 121-135 133 to effectively adopt and implement bim in building construction projects in ncr, the following step-by-step procedures on using bim in designing, constructing, and operating building and infrastructure projects are therefore recommended. fig. 3 shows a research-validated guideline that provides companies with a comprehensive approach to adopting and implementing bim in their projects. 5. conclusions in this research, the small but growing pieces of literature and studies regarding bim were analyzed. the descriptive research method was applied to investigate the benefits, risks, barriers, and approaches of bim implementation. research findings were then used to formulate recommendations and design guidelines for implementing bim in construction projects. in light of the previous findings, the following conclusions were drawn: (1) the benefits of bim were visualizations of a design earlier and more accurately and the detection of design errors, omissions, and clashes. other benefits were providing preliminary insights into design problems, a better understanding of proposals, and integrating multiple disciplines into one plan. (2) the risks were the responsibility and control of data entry of the model, financial risks, and software licensing issues. other risks such as resistance to change by senior staff and delays in project completion were also experienced. (3) the barriers to bim implementation were the expensive acquisition of bim software licenses, high cost of personnel training and bim implementation, limited project funding and budget, lack of trained professionals to handle bim tools and software, and expensive human-based services costs. (4) the most common practices or approaches in addressing the challenges of bim implementation were: increasing the availability of bim technology, establishing bim execution guidelines, innovating research for bim technology, organizing the transition process from traditional practice to bim, and conducting workshops on bim. (5) the respondents have significantly different perspectives on the benefits of bim, particularly on bim’s high level of customization of building systems. however, fewer differences in all of the top five risks of bim were observed, except for data integrity and security. (6) a research-validated framework or guidelines was designed to help the aec sector implement bim in their construction projects. this framework will give clients, consultants, and contractors a thorough plan for how to start using bim in their projects. 6. recommendations one of the objectives of this study is to recommend how to adopt and implement bim in the aec industry, particularly in construction projects. therefore, considering the results of the study as reflected in the findings and conclusions, the following recommendations are offered: (1) to immediately start implementing bim in the philippine aec industry, bim technology should be required in the construction contracts for consultants and contractors to adopt this new technology. it is also necessary that an information manager should be assigned to each project to manage the processes and procedures for information exchange, initiate and implement project and asset information plans, assist in the preparation of project outputs, and implement the bim information exchange protocol. (2) it is also recommended that aec companies conduct bim awareness seminars and workshops to educate their respective employees and spread awareness and appreciation of the benefits of bim. benchmarking other companies already using bim is also an effective way to motivate senior employees to embrace change and consider adopting and using bim technology. advances in technology innovation, vol. 8, no. 2, 2023, pp. 121-135 134 (3) it is highly recommended that companies look for ways to increase their funds and budget to acquire and implement this new technology. aec companies should treat these bim-related expenses as investments toward realizing the benefits, which would eventually outweigh the costs. (4) aec companies should start conducting training and workshops for their employees. they can also hire trainers from institutions offering courses for bim and/or send some of their employees abroad for exposure and learn about benchmarking bim implementation. moreover, institutions of higher learning should start bim courses and increase research on utilizing bim in various projects. conflicts of interest the author declares no conflict of interest. references [1] h. lindblad and s. vass, “bim implementation and organisational change: a case study of a large swedish public client,” procedia economics and finance, vol. 21, pp. 178-184, 2015. [2] r. leeds, “top 4 challenges facing the construction industry,” https://www.designingbuildings.co.uk/wiki/top_4_challenges_facing_the_construction_industry, june 01, 2019. [3] a. monteiro and j. p. martins, “sigabim: a framework for bim application,” proceedings of the xxxviii iahs world congress, pp. 01-08, april 2012. [4] g. s. evuri and n. amiri-arshad, “a study on risks and benefits of building information modeling (bim) in a construction organization,” b. s. thesis, department of computer science and engineering, university of gothenburg, goteborg, sweden, 2015. [5] c. eastman, p. teicholz, r. sacks, and k. liston, bim handbook: a guide to building information modeling for owners, managers, designers, engineers, and contractors, 2nd ed., new jersey: john wiley & sons, inc., 2011. [6] s. azhar, m. hein, and b. sketo, “building information modeling (bim): benefits, risks and challenges,” http://ascpro0.ascweb.org/archives/cd/2008/paper/cpgt182002008.pdf, april 02, 2008. [7] h. yan and p. demian, “benefits and barriers of building information modelling,” https://hdl.handle.net/2134/23773, january 01, 2008. [8] a. ghaffarianhoseini, j. tookey, a. ghaffarianhoseini, n. naismith, s. azhar, o. efimova, et al., “building information modelling (bim) uptake: clear benefits, understanding its implementation, risks and challenges,” renewable and sustainable energy reviews, vol. 75, pp. 1046-1053, august 2017. [9] w. kymmell, “building information modeling: planning and managing construction projects with 4d cad and simulations,” 1st ed., new york: mcgraw-hill, 2008. [10] n. a. a. ismail, m. chiozzi, and r. drogemuller, “an overview of bim uptake in asian developing countries,” aip conference proceedings, vol. 1903, no. 1, article no. 080008, november 2017. https://doi.org/10.1063/1.5011596 [11] w. zheng, “a comprehensive analysis of building information modeling/model (bim) policies in other countries and its adoption strategies in ontario,” master thesis, engineering and public policy, walter g. booth school of engineering practice, mcmaster university, hamilton, ontario, august 2013. [12] s. lorek, “global bim standards: is your country next? constructible,” https://constructible.trimble.com/constructionindustry/global-bim-standards-is-your-country-next, april 05, 2018. [13] t. ishizawa, y. xiao, and y. ikeda, “analyzing bim protocols and users surveys in japan to understand the current japanese bim environment, through the comparison with different countries,” proceedings of the 23rd international conference of the association for computer-aided architectural design research in asia, vol. 2, pp. 31-40, may 2018. [14] t. kaneta, s. furusaka, and n. deng, “overview and problems of bim implementation in japan,” frontiers of engineering management, vol. 4, no. 2, pp. 146-155, july 2017. [15] g. kekana, c. aigbavboa, and w. thwala, “understanding building information modelling in the south africa construction industry,” proceedings of the organization, technology and management in construction (otmc), september 2015. https://core.ac.uk/download/pdf/54199032.pdf [16] s. korff, “bim in south africa: the good, the bad and the promising,” https://www.buildinganddecor.co.za/bim-insouth-africa-the-good-the-bad-and-the-promising, july 09, 2019. advances in technology innovation, vol. 8, no. 2, 2023, pp. 121-135 135 [17] a. abbas, z. u. din, and r. farooqui, “achieving greater project success & profitability through pre-construction planning: a case-based study,” procedia engineering, vol. 145, pp. 804-811, 2016. [18] v. n. r. villasenor, f. c. villasenor, and d. r. gonzales, “educating green building stakeholders about the benefits of bim the philippines experience,” joint apec-asean workshop how building information modeling standard can improve building performance, june 2013. [19] d. r. gonzalez, analysis of the impact of building information modeling (bim) in construction production from the filipino's perspective, manila: mapua institute of technology, 2013. [20] d. migilinskas, v. popov, v. juocevicius, and l. ustinovichius, “the benefits, obstacles and problems of practical bim implementation,” procedia engineering, vol. 57, pp. 767-774, 2013. [21] e. alreshidi, m. mourshed, and y. rezgui, “factors for effective bim governance,” journal of building engineering, vol. 10, pp. 89-101, march 2017. [22] s. azhar, “building information modeling (bim): trends, benefits, risks, and challenges for the aec industry,” leadership and management in engineering, vol. 11, no. 3, pp. 241-252, july 2011. [23] i. g. khalil, a. mohamed, and z. smail, “building information modelling in morocco: quo vadis?” third international sustainability and resilience conference: climate change, pp. 479-483, november 2021. [24] i. a. m. ya’acob, f. a. m. rahim, and n. zainon, “risk in implementing building information modelling (bim) in malaysia construction industry: a review,” e3s web of conferences, vol. 65, article no. 03002, 2018. [25] n. ham, s. moon, j. h. kim, and j. j. kim, “economic analysis of design errors in bim-based high-rise construction projects: case study of haeundae l project,” journal of construction engineering and management, vol. 144, no. 6, june 2018. https://doi.org/10.1061/(asce)co.1943-7862.0001498 [26] c. c. wang and o. chien, “the use of bim in project planning and scheduling in the australian construction industry,” international conference on construction and real estate management, pp. 126-133, september 2014. [27] t. kouide, g. paterson, and c. thompson, “bim as a viable collaborative working tool: a case study,” proceedings of the 12th international conference on computer aided architectural design research in asia, pp. 57-68, april 2007. [28] f. j. sabongi, “the integration of bim in the undergraduate curriculum: an analysis of undergraduate courses,” proceedings of the 45th asc annual international conference, april 2009. http://ascpro0.ascweb.org/archives/cd/2009/paper/ceue90002009.pdf [29] s. g. akdag and u. maqsood, “a roadmap for bim adoption and implementation in developing countries: the pakistan case,” archnet-ijar: internationa journal of architectural research, vol. 14, no. 1, pp. 112-132, january 2020. copyright© by the authors. licensee taeti, taiwan. this article is an open access article distributed under the terms and conditions of the creative commons attribution (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).