proceedings of engineering and technology innovation , vol. 2, 2016, pp. 04 05 4 a low-cost design of earthquake detector with rescue message deliver using mobile device chuen-ching wang * , gi-hong wei department of information technology, kao yuan university, kaohsiung, taiwan. received 30 january 2016; received in revised form 15 february 2016; accept ed 18 march 2016 abstract the earthquake detector is a complex and expensive device. it is necessary to design a simpler and cheaper earthquake detector for common use. our research employs the ability of android-based mobile device to design a minima l cost detector for sensing the earthquake. therefore, as the earthquake occurring, the mobile device can detect the earthquake and issue a warning alarm to the phone owner. at the same time, a rescue messages including disaster location, earthquake level, and related information to the disaster center. experimental results show that the proposed method can achieve the usage of disaster rescue. keywor ds : earthquake detection, earthquake early warn ing system, ce ll broadcast service 1. introduction because earthquake occurrence is difficult to predict, the earthquake hazard plays a major role in the natural disaster. for seismic hazard mitigation, a practical earthquake forecast method appears to be far from realization, because of the extreme complexity involved in the earthquake processes. there are two types of seismic waves:[1-2]body waves that propagate through the earth’s interior and surface waves that propagate along the earth’s free surface or along other discontinuities in the earth’s interior. body waves traveling faster than the surface waves are classified to two types: compressional (longitudinal) and shear (transverse). compressional waves travel about 1.7 times faster than transverse waves and are often called p waves or primary waves. transverse waves are frequently called s waves or secondary waves. p waves are always the first among seismic waves that reach the recording station. rock particles affected by a propagating p wave oscillate backward and fo rward in the same direction as the wave propagates. in the case of s waves, particles vibrate in planes perpendicular to the direction of travel [3]. in this investigation, an earthquake detector is designed using the accelerator which e xists in the mob ile device to immediate ly ca lculate a c celeration, and therefore the earthquake amp litude can be detected. 2. earthquake detection design due to the rapid development of mobile communicat ions, mob ile devices currently can provide anyone, anywhere, and any when to access the internet through the communication network. for the reason of mobility, we imple ment the proposed method in the mobile device to reduce the cost of earthquake detector. we e xpress the main process shown as follows. first, when earthquake is occurring, the earthquake signal will be detected by g-sensor, which can sense the acceleration and orientation. ne xt , the acceleration is calcu lated by the co mputation module and transferred as earthquake magnitude. the earthquake magnitude can be used for determin ing whether an alarm/ message is issued, or not. that is, if the earthquake magnitude is greater than the specific threshold, then the location information (lat itude and longitude) is obtained by gps wh ich built in mobile device and thus the location information can be passed to the alarm module to issue alarm and rescue messages to notify the specific person or disaster center. in order to obtain the s ma llest possible errors, the phone accelerometer is compared with a standard accelerometer located at the national center for research on ea rthquake engineering (ncree) [4]. fig. 1 shows our tes t phone (marked with red bo x) is set on the accelerometer calibration system. the process is shown as follows : * corresponding aut hor. email: t 90261@cc.kyu.edu.tw proceedings of engineering and technology innovation , vol. 2, 2016, pp. 04 05 5 copyright © taeti (1) set the phone on the fixed standard accelerometer (2) adjust the device parameters (3) repeat step 1 and 2 until the phone output a reasonable range. fig. 1 comparison of the accelerator in mobile device and standard accelerometer at ncree 3. conclusions we have integrated the gravity accelerome ter, gps module and co mmunication modules and other hardware system to imp le ment an earthquake detection app in the mobile device. de monstrations show that the app can detect the earthquake intensity, latitude and longitude informat ion, time of occurrence, phone number and other messages and forward these informat ion to the specified relat ives or disaster center using mobile commun ication. the proposed scheme with advantages of easy installing and no extra hardware cost. acknowledgment the research is supported by the nsc 104-2221-e-244-018-, national science council, taiwan. reference [1] “central weather bureau,” http://www.cwb.gov.tw/ [2] s. l. kra mer, geotechnical ea rthquake engineering, nj: prentice-hall, engelwood cliffs, 2003. [3] a. magro-ca mpe ro, r. l. fle ischer, and r. s. likes, “ changes in subsurface radon concentration associated with earthquakes,” j. geophys. res., vol. 85, no. b6, pp. 3053-3057, 1980. [4] “national center for research on eart hquake engineering,” http://www.ncree.org/  proceedings of engineering and technology innovation , vol. 2, 2016, pp. 26 27 26 oppa-ac: optimal path planning based on ant colony algorithm for temporary isolated node in wsn aripriharta 1, 2 , hendrick 1, 3,* , thi thuy lieu nguyen 1 , gwo-jia jong 1 1 department electrical engineering, national kaohsiung university of applied sciences, kaohsiung, taiwan. 2 state university of malang, indonesia. 3 state polytechnic of padang, indonesia. received 01 april 2016; received in revised form 28 april 2016; accept ed 10 may 2016 abstract in real wireless sensor network application (wsn), several nodes may suffer fro m link fa ilure problem. link failure is a proble m that may exist in the presence of obstacles which blocked the wireless connection between nodes. while the link between nodes is blocked, the node will be temporary isolated from the cluster; thus, their data will not reach the destination node. the temporary node isolation problem becomes more challenging if the data should be arrived on time. the traditional clustering algorithm (leach) is not considered the temporary isolated node, which cause longer waiting time for data delivery in certain rounds. in order to solve this problem, we proposed the ant colony-based optimal path planning algorithm (oppa -ac). the opp-a c was improved the leach algorith m by providing alternative path for temporary isolated nodes and guarantee their data arrived in the destination. based on the experimental result, the opp-a c surpassed the traditional algorithm in term of waiting time. ke ywor ds : tempora ry isolated node, ant colony, opp-ac, wsn 1. introduction in wireless sensor network application (wsn), the nodes may suffer from link failure problem due to the environmental effects, such as: radiation, temperature rising, fog, rain and obstacle. link failure is a problem that may exist in the presence of obstacles which blocked the wireless connection between nodes. therefore, those nodes cannot send data through the original forwarding path for a period of time (or temporal) because their co mmunication range is decreased [2-5] or blocked. sink ch1 temporarily isolated node normal node ch2 ch3 fig. 1 temporary isolated node since lea ch was not considered this problem, those nodes are temporarily isolated fro m their ch. the te mporary node isolation problem beco mes more challenging if the data should be arrived on time. moreover, the temporarily isolated nodes (fig. 1) cause packet drops, increase the average delay, and increase the energy consumption; in turn, the perfo rmance of network become degrades [3]. recently, there has been a few re lated works in te mporarily node isolation p roble m [2-5]. however, those existing works focused on connection reestablishment. there fore, we pre sent the ant colony-based optima l path planning algorith m (oppa -ac) to select the forwarding path for tempora rily isolated nodes . our objec tive is to reduce the waiting time and congestion in wsn. 2. method the structural synthesis of ccpgts will be performed based on the creative design met hodology process [7-8]. the p roposed algorith m *corresponding aut hor. email: hendrickpnp77@gmail.co m proceedings of engineering and technology innovation , vol. 2, 2016, pp. 26 27 27 copyright © taeti consists of two steps: (a) clustering process, (2) data transmission. in the first step, we adopt lea ch fo r the ch selection [1]. the second step involves the proposed path selection mechanis m in order to reduce the end to end delay and packet drops in wsn. in order to determine the forwa rding path for temporarily isolated nodes, we used link quality as the pheromone of ants by favoring received signal strength indication (rssi). the rssi value is collected each time the tempora rily isolated node receive adv message fro m nearby chs. rssi va lues is normalized for fu rther calculation by opp-ac, and it given by: 𝑅𝑆𝑆𝐼𝑛𝑜𝑟𝑚 = 𝑅𝑆𝑆𝐼𝑖 𝑅𝑆𝑆𝐼𝑚𝑎𝑥 (1) the objective function is the lin k quality (lq) wh ich depends on rssi va lue. therefore , the isolated node will selects the path with better lq. moreover, ant colony a lgorith m select a chs node with probability pij k = 𝜏𝑖𝑗 𝛼 𝜂𝑖𝑗 𝛽/ ∑ 𝜏𝑖𝑠 𝛼𝜂𝑖𝑠 𝛽 𝑠𝜖𝑙𝑖𝑠𝑡𝑘 for j∈ 𝑙𝑖𝑠𝑡𝑘 , otherwise pij k = 0. moreover, 𝜂𝑖𝑗 = 1/𝑑𝑖𝑗 and the pheromone (𝜏𝑖𝑗 = 1 𝑅𝑆𝑆𝐼𝑛𝑜𝑟𝑚 ), =0.7 and 𝛽 = 0.3. 3. results and discussion we validated the proposed algorith m through simu lations and comparison with the lea ch in term o f end to end de lay. we simu lated 35 nodes in 50x50m 2 square area, the data length ia 30 byte, elec = 50nj/bit, eda=0.5nj/bit, amp=50nj/bit/m 4 , and fs= 10pj/bit/m 2 . based on the e xperimental result (fig. 2), the opp-ac surpassed the lea ch a lgorith m in term of end to end delay. 4. conclusions in this paper, three new designs of the ccpgt have been generated in a systematic methodology. the feasibility of the new designs is verified by conducting kine mat ic simu lation. 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 alternatives for driving a variable speed input mechanism. e n d t o e n d d e la y nodes leach opp-ac fig. 2 end to end delay acknowledgement this work was supported by dikti funded by ministry of research and tech nology and higher education, indonesia under contract number: 124.67/e4.4/2014. references [1] m. mehdi afsar and mohammad-h. tayarani-n, “clustering in sensor networks: a literature survey,” journal of network and co mputer applications, vol. 46, pp. 198-226, 2014. [2] a. roy, p. ka r, and s. m isra, “ detection of dumb nodes in a stationary wire less sensor network,” annual ieee india conference (indicon ‘14), 2014, pp. 1-6. [3] p. ka r, a. roy, s. m isra, and m. s. obaidat, “energyeffic ient connectivity reestablish ment in wsn in the presence of du mb nodes,” ieee international conference on co mmunicat ion workshop (iccw ‘15), 2015, pp. 1485-1490. [4] a. roy, a. mondal, and s. misra, “ connectivity re-establishment in the presence of dumb nodes in sensor-cloud infrastructure: a game theoretic approach,” ieee 6th international conference on cloud co mputing technology and science (cloudco m ‘14), 2014, pp. 847-852. [5] s. misra, p. kar, a. roy, and m. s. obaidat, “existence of dumb nodes in stationary wireless sensor networks,” journal of system and software , vol. 91, pp. 135-146, 2014.  proceedings of engineering and technology innovation , vol. 3, 2016, pp. 13 15 13 behavioural assessment of blood-brain barrier opening induced by various ultrasound parameters feng-yi yang * , sheng-fang huang department of biomedical imaging and radiological sciences , national yang-ming university, taipei, taiwan. received 23 march 2016; received in revised form 15 april 2016; accept ed 02 may 2016 abstract focused ultrasound (fus) -induced opening of the blood-brain barrier (bbb) in the presence of microbubbles is prospective for a targeted drug delivery to the bra in lesion. ho wever, the safety of this technology has been addressed recently. the purpose of this study was to assess the behavioural changes follo wing the fus induced-bbb disruption (bbbd). the behavioural assessment of rats were tested using the open field and hole-board fro m day 1 through day 32 a fter undergoing bbbd induced by fus with either a mild or heavy parameter. we found that heavy bbbd induced behavioral changes, including significantly increased locomotor activity and a longer latency to nose poke for baits . additionally, mild bbbd resulted in significant decrease of central activity. therefore, the behavioural changes after fus induced bbbd should be considered before clinica l application. ke ywor ds : behavioural evaluation, focused ultrasound, blood-brain barrier, memory 1. introduction focused ultrasound (fus) with mic robu bbles has offered the potential to produce bbb disruption (bbbd) noninvasively in specific regions of the brain [1, 2]. however, the greatest limitat ion on the use of fus-induced bbbd in clin ical practice consists of safety concerns relating to cavitation in the bra in. mechanica l effects may be responsible for fus-induced bbbd, but inert ial cavitation could usually cause hemorrhaging or apoptosis in the brain tissue from the neighboring vessel [3]. a lthough no significant negative effects resulting fro m histological e xa mination, further evaluation of brain functions follo wing fus-induced bbbd were still needed. such research will increase our knowledge of the behaviora l a lterations that occur following bbbd and allow for a better assessment of the safety of this technique in terms of its effects on bra in functions. therefore , the purpose of this study was to exa mine the impact of bbbd in terms of behavioral altera tions following fus e xposure in the presence of microbubbles. 2. method all procedures involving anima ls were conducted in accordance with the guidelines for the care and use of laboratory animals. th e study protocol was approved by the anima l ca re and use committee of national yang ming university. male sprague-dawley (sd) rats weighing fro m 300 to 350 g we re used in this study. 2.1. ultrasound setup and behavioural assessment the ultrasound system and equipment setup were the same as used in our previous study [4]. the bbb disruption (bbbd) can be quantified based on the extravasation of evans blue (eb), which binds to albumin. activ ity in the open fie ld was tested with the automated fle x-fie ld/open field photobeam activity sy stem on post-sonication days 1, 9, 18, and 32. spatial lea rning ability was studied by means of a hole-board apparatus in which food rewa rds were used as positive motivation. 2.2. number synthesis three rats fro m each group we re prepared for histological observation. the rats were perfused with saline and 10% neutral buffered * corresponding aut hor. email: fyyang@ym.edu.t w proceedings of engineering and technology innovation , vol. 3, 2016, pp. 13 15 14 copyright © taeti forma lin on days 1 and 9 a fter the fus sonication. the brain slices were stained by tunel staining (deadend co lorimetric t unel system, g7130, pro mega, mad ison, wi, usa) in order to detect dna frag mentation and apoptotic bodies within the cells . all values are shown as means ± sem. the behavioral assessment data were analy zed using the mann-whitney u test. other data were analyzed using the unpaired student’s t test. statistical significance was defined as a p value  0.05. 3. results and discussion fig. 1 effects of rats with fus -induced bbbd on the open field activ ity. (a) the nu mber of center entries was significantly increased in the bbbd-heavy group on days 1 and 9 follo wing sonication. (b) but the number of center entries was significantly decreased in the bbbd-mild group on day 18 following sonication. * denotes significant differences compared to the sham group at the same time points. (*, p<0.05; ***, p<0.005, n = 8) bbbd-heavy e xposed rats entered the center of the open field significantly more frequently on days 1 and 9 post-sonication compared with sham group (fig. 1(a)). bbbd-mild e xposure significantly reduced the number of center entries on day 18 post-sonication compared with sham group (fig. 1(b)). the bbbd-heavy group e xhibited a significantly longer latency and latency to first baited hole compared with the sham group (fig. 2(a)). no s ignificant diffe rences in latency and latency to first baited hole were found between the bbbd-mild group and sham group (fig. 2(b)). fig. 2 ho le-board testing was performed on day 12 fo llo wing sonication. the latency and latency to the first baited hole were recorded in (a) the bbbd-heavy group and (b) the bbbd-mild group. * denotes a significant diffe rence relative to sham group. (*, p<0.05; **, p<0.01, n = 8) to our knowledge, the current study is the first to exp lore the e ffect of bbbd induced by fus on behavioral alterations in an anima l model. the findings fro m this study indicate that bbbd-heavy rats produced hyperactivity rela tive to sham animals under open field test. proceedings of engineering and technology innovation , vol. 3, 2016, pp. 13 15 15 copyright © taeti moreover, the bbbd-heavy rats revea led spatial me mo ry impa irment in the hole-board test compared to sham group. in contrast, bbbd-mild rats only e xhibited an xiety-re lated behaviors in the open field test. 4. conclusions in this paper, fus-induced bbbd repre sents a major advance in the targeted drug delivery of the bra in. the current study suggests the possibility that manipulation of fus-induced bbbd might produce a variety of behavioral changes after enhanced drug delivery. further investigations of behavioral alterations in anima ls fo llowing fus -induced bbbd are needed in order to avoid abnorma l bra in function ing in humans following future clinica l uses of fus-induced bbbd. acknowledgement this study was supported by grants from the ministry of science and technology of taiwan (no. most 104-2314-b-010-003-m y3). references [1] f. y. yang, t . t. wong, m. c. teng, r. s. liu, m. lu, and h. f. liang, et a l., “focused ultrasound and interleukin-4 receptor targeted liposomal do xorubic in for enhanced targeted drug delivery and antitu mor effect in glioblastoma multiforme ,” journal of controlled release, vol. 160, no. 3, pp. 652-658, june 2012. [2] f. y. yang, w. y. chang, j. j. li, h. e. wang, j. c. chen, and c. w. chang, “pharmacokinetic analysis and uptake of 18f-fbpa-fr sfter ultrasound-induced blood-brain barrie r disruption for potential enhancement of boron delivery for neutron capture therapy,” journal of nuclear medicine, vol. 55, pp. 616-621, april 2014. [3] f. yang, w. fu, w. chen, w. yeh, and w. lin, “quantitative evaluation of the use of microbubbles with transcranial focused u ltrasound on blood¡vbra in-barrie r disruption,” ultrasonics sonochemistry, vol. 15, no. 4, pp. 636-643, april 2008. [4] f. y. yang, w. y. chang, j. c. chen, l. c. lee, and y. s. hung, “quantitative assessment of ce rebral glucose metabolic rates after blood-brain barrie r disruption induced by focused ultrasound using fdg-m icropet ,” neurolmage, vol. 90, pp. 93-98, april 2014.  proceedings of engineering and technology innovation, vol. 5, 2017, pp. 37 40 applications of thermal images for monitoring surficial temperature changes of naked slope chien-yuan chen * , zhe-hao liu department of civil and water resources engineering, national chiayi university, chiayi, taiwan, roc. received 02 july 2017; received in revised form 06 august 2017; accept ed 10 august 2017 abstract global climate change causes increases in the torrential rainfall brought by typhoons and the monsoon in taiwan. torrential rain in turn causes landslides, debris flows, and the formation of e arth dams. most dams were formed in remote mountainous areas and are difficult to reach for safety evaluation at the beginning of their formation. a long distance and non-destructive testing methodology is necessary for evaluating the safety of landslide d ams. this study used an infrared imager for monitoring naked slopes. the thermography can detect surficial radiation temperature changes in the slope to locate potential unstable areas for further monitoring. this study proposes radiation temperature change (t) per unit of time (δt) as an index (t /δt) for nondestructive monitoring. the index was used for monitoring and analysis of artificial earth dams constructed a t huishun farm in nantou county. the results of the analysis show that the failure zone of the artificial dam exhibited the greatest change in the index and the potential failure mode could be predicted once the dam breached. the proposed model could be used for potential unstable slope monitoring. keywords: thermography, landslide, monitoring 1. introduction typhoon morakot hit taiwan in 2009 and brought torrential rainfall in southern taiwan. the typhoon caused 17 landslide dams [1]. the failure modes of landslide dams include piping, overtopping, and complex modes [2]. stream flow blocks by landslides could cause ponding water upstream and flood disasters downstream once they are breached . natural dam stability evaluation is urgent for dam breach disaster prevention and mitigation. thermal images have been used to examine a wide variety of problems [3-4]. it has been commonly used for monitoring activities of landslides in recent years [5-7]. it has the advantages of non-contact and long distance as camera-like characteristics, and is suitable for both long -term monitoring and short-term inspection for sites with surficial radiation temperature changes . 2. field model test the artificial earth dam field test was conducted at huishun farm in nantou county in central taiwan. the test procedure included construction of an artificial earth dam in the downstream region of landou creek. the dam size was 25 m in length and * corresponding author. e-m ail address: chieny uc@ m ail.ncy u.edu.tw tel.: +886-5-2717686; fax: +886-5-2717693 proceedings of engineering and technology innovation, vol. 5, 2017, pp. 37 40 copyright © taeti 38 2 m in height and it blocked the entire creek [8]. water was stopped upstream first and released during tests. the test was made on 10 april 2014, and modeled the failure processes and modes of natural earth dams. the thermal imager was set up on a bridge for monitoring of the upstream face of the dam and at the side bank for the downstream face (fig. 1). thermal images were tak en every few minutes with the increasing water table during the test until the dam breached . (a) field photography corresponding thermography fig. 1 the artificial earth dam and corresponding thermal image (taken 10 april 2014) 3. methodology the study proposed a new index for monitoring the surficial cha nges of soil slope by water infiltration using thermal images. the index (t /δt) is a measurement of the surficial radiation temperature changes (t) in the time period between the times th e two images were taken (δt). the changes of surficial radiation temperature over time are used for monitoring during the field test for potential failure mode evaluation. the surface of the earth dam downstream was divided into grids for monitoring the temperature changes. there are 27 grids monitored as marked in the image from left to right: a1~a3, b1~b8, c1~c8, and d1~d8 (fig. 1). the temperature is taken as an average within the grid. fifteen successive images were taken for the analysis. a positive value of the index shows the tempe rate is increasing. a negative index shows reduction in the temperate due to seepage and piping with the increasingly blocked water table. 4. results and analysis fig. 2 monitoring of surficial radiation temperature changes on 14:15:56 fig. 3 monitoring of surficial radiation temperature changes on 14:28:39 monitoring began at 14:15:56 and ended at 14:34:26. the surficial temperature variations with time at the monitoring grids for the first two images are shown in fig. 2. all the grids showed a positive index with increasing surficial temperature during the proceedings of engineering and technology innovation, vol. 5, 2017, pp. 37 40 copyright © taeti 39 sunshine, at the beginning of the test. soil erosion and piping occurred at gird no. 26 (d7 in fig. 1) 12 minutes later with a negative value for the index (fig. 3). with increasing time, another piping occurred in grid no. 8 (b5) (fig. 4). it is speculated that the type of failure mode is seepage caused piping if the blocked water table is not rising at the current stage. the dam breached in a complex mode with piping first followed by overtopping failure (fig. 5). the thermal images showed a clear lower index in the grids with potential seepage-induced piping failure, though the dam failed by overtopping with increasing river flow. piping failure occurs in low permeability dam material with high soil shear strength under high discharge flow. it should be note that the piping water in the upper grids could flow down to the lower grids without piping, causing a lower index in all these grids. the res ults show that the index (t /δt) is suitable for monitoring surficial radiation temperat ure changes of naked slopes in a given time period and may be applied for monitoring potential landslide slopes . (a) field photography (b) corresponding thermography (c) the index (t /δt) changes fig. 4 monitoring of surficial radiation temperatu re changes on 14:33:55 (a) field photography (b) corresponding thermography fig. 5 monitoring of surficial radiation temperature changes on 14:34:26 (continued) proceedings of engineering and technology innovation, vol. 5, 2017, pp. 37 40 copyright © taeti 40 (c) the index (t /δt) changes fig. 5 monitoring of surficial radiation temperature changes on 14:34:26 5. conclusions a long distance and non-destructive testing methodology using an infrared imager is adopted for monitoring the safety of artificial dams. the thermography can detect surficial radiation temperature changes in the slope t o locate potential unstable areas for further monitoring. the radiation temperature change per unit of time is proposed as an index (t /δt) for nondestructive monitoring. the index was used for monitoring and analysis of artificial earth dams constructed at huishun farm in nantou county. the results of the analysis show that the piping failure zone of the artificial dam exhibited the greatest change in the grid and the potential failure mode could be predicted. the index reflects the surficial radiation temperature changes and can be used for monitoring the stability of naked slopes . acknowledgement the authors would like to thank the ministry of science and technology, taiwan roc, for financially supporting this research under contract no. most 105-2625-m-415-001. references [1] c. y. chen and j. m. chang, “landslide dam formation susceptibility analys is based on geomorphic features ,” landslide, vol. 13, no. 5, pp. 1019-1033, october 2016. [2] o. korup, “recent research on landslide dams a literature review with s pecial attention to new zealand,” progress in physical geography, vol. 26, no. 2, pp. 206-235, june 2002. [3] i. baroň, d. bečkovský, and l. míčai, “application of infrared thermography for mapping open fractures in deep -seated rockslides and unstable cliffs,” landslides, vol. 11, no. 1, pp. 15-27, february 2014. [4] t. nolesini, w. frodella, s. bianchini, and n. casagli, “detecting slope and urban potential unstable areas by means of multi-platform remote sensing techniques: the volterra (italy) case study,” remote sens ing, vol. 8, no. 9, p. 746, september 2016. [5] j. h. wu, h. m. lin, d. h. lee, and s. c. fang, “integrity assessment of rock mass behind the shot creted slope using thermography,” engineering geology, vol. 80, no. 1-2, pp. 164-173, august 2005. [6] s. martino and p. mazzanti, “integrating geomechanical surveys and remote sensing for sea cliff slope stability analys is: the mt. pucci case study (italy),” natural hazards and earth system sciences, vol. 14, pp. 831-848, april 2014. [7] s. mineo, g. pappalardo, f. rapisarda, a. cubito, and g. di maria, “integrated geostructural, seismic and infrared thermography surveys for the study of an unstable rock slope in the peloritani chain (ne sicily),” engineering geology, vol. 195, pp. 225-235, september 2015. [8] c. y. chen, s. c. chen, and k. h. chen, “earth dam monitoring by using infrared thermography detection,” proceedings of engineering technology and innovation, vol. 4, pp. 40-42, 2016.  proceedings of engineering and technology innovation , vol. 3, 2016, pp. 22 24 22 recognition of concurrent control chart patterns in autocorrelated processes using support vector machine chau-chen torng * , chon-how hwang department of industrial engineering and management, national yunlin university, yunlin, taiwan. received 16 february 2016; received in revised form 18 march 2016; accept ed 09 april 2016 abstract control chart pattern recognition (ccpr) is an important issue in statistical process control because unnatural control chart patterns (ccps) e xhibited on control charts can be associated with specific causes that adversely affect the manufacturing processes. in recent years, many mach ine learning techniques have been su ccessfully applied to ccpr. however, such existing research for ccpr has mostly been d eveloped for identification of basic ccps (sh ift patterns, trend patterns, cyclic pattern and systematic pattern). little attention has been given to the identification of concurrent ccps (two or more basic ccps occurring simultan eously) which are co mmonly encountered in practical manufacturing processes. in addition, these e xisting researches also assume the process data are independently and identically distributed which may not be appropriate for certain manufacturing processes. this study proposes a support vector machine (svm) approach to identify concurrent ccps for a mu ltivariate process with autocorrelated observations which can be characterized by a first order autoregressive (ar(1)) mode l. the nume rical results indicate that the proposed model can effect ively identify two concurrent identical ccps but for those cases involving one trend pattern and one shift pattern, their recognition accuracy deteriorates to around 20% to 50% depending on the autocorrelation coeffic ients used in the data model. ke ywor ds : control chart pattern, support vector machine, autocorrelation process 1. introduction the effectiveness of the use of control charts depends on the ability to recognize patterns. the common types of ccp e xhib ited on statistical process control charts have been formalized in early literature. in most literature, the fo llo wing seven typical types of basic ccp, i.e ., up ward and downwa rd sh ift pattern (usp and dsp); upward and down ward t rend pattern (utpand dtp); cyclic pattern (cp); systemat ic pattern (sp); natural pattern (np), are usually addressed. identificat ion of unnatural patterns canfacilitate early detection of an out -of-control process and the diagnostic search process by narrowing down the set of possiblecauses that must be investigated. for instance, shift patterns may indicate changes in materia l, mach ine or operator, while trend patternsmay indicate tool wear. cyclic patterns may ind icate voltage fluctuation in power supply [1-3]. there is a crucia l need for an auto matic and effective analysis and interpretation methodology for control chart pattern recognition (ccpr), which enables indication of the real state of the manufacturing process. ccpr studies consist of the description, the identification, the e xp lic it classification, and the e xtraction of patterns in data. in recent years, many mach ine learning techniques have been successfully applied to ccpr. ho wever, such e xisting research for ccpr has mostly been developed for identification of basic ccps (sh ift patterns, trend patterns, cyclic pattern and systematic pattern). little attention has been given to the identification of concurrent ccps (two or more basic ccps occurring simu ltaneously) wh ich are commonly encountered in practical manufacturing processes. in addition, these e xisting researches also assume the process data are independently and identically distributed which may not be appropriate for certain manufactu ring processes. for those processes involving continuous manufacturing operations (including the manufacture of food, chemicals, and paper * corresponding aut hor. email: t orngcc@yuntech.edu.tw proceedings of engineering and technology innovation , vol. 3, 2016, pp. 22 24 23 copyright © taeti and wood products), the consecutive observ ations of product characteristics are often highly correlated. the purpose of this paper is to propose a support vector machine (svm ) approach to identify concurrent ccps for a mu ltivariate process with autocorrelated observations. 2. method this study uses simu lations to generate the required sets of ccp e xa mp les for tra ining, testing, and performance evaluation of the proposed model. the mathemat ical e xp ressions for ccp generation are e xp ressed in a general form that consists of process mean, common cause variation and special d isturbance from assign able causes. the equations and parameters used to generate the data points are shown in table 1. table 1 ccp equations and parameters pat t erns equat ions paramet ers nat ural pat t ern 𝑥𝑖 = 𝜇 + 𝑟𝑖𝜎 =0, =1 t rend pat t ern 𝑥𝑖 = 𝜇 + 𝑟𝑖𝜎 + 𝑡 × 𝑖  26.0,10.0t shift pat t ern 𝑥𝑖 = 𝜇 + 𝑟𝑖𝜎 + 𝑢 × 𝑠   10 0.3,0.1 oru s   cyclic pat t ern 𝑥𝑖 = 𝜇 + 𝑟𝑖𝜎 + 𝑎 × sin 2𝜋𝑡 𝛺  0.3,0.1a syst emat ic pat t ern 𝑥𝑖 = 𝜇 + 𝑟𝑖𝜎 + 𝑑(−1) 𝑖  0.3,0.1d the follo wing equations are used to simulate the autocorrelated data. 𝑋1𝑡 = 𝜇 1 + 𝜙11 𝑋1 ,𝑡−1 +𝜙12 𝑋2 ,𝑡−1 + 𝜀1𝑡 (1) 𝑋2𝑡 = 𝜇 2 + 𝜙21 𝑋1 ,𝑡−1 +𝜙22 𝑋2 ,𝑡−1 + 𝜀2𝑡 (2) where 𝜙𝑖𝑗 are autocorrelated coefficients , and 𝜙11 = {1.0, 0.7 0.4, 0.1}, 𝜙22 = {1.0, 0.7, 0.4, 0.1}. for tra ining and testing the concurrent patterns, a svm based on radial basis kernel fun ction is chosen inthis study. related para meters c and γ for this ke rnel were varied in the fixed ranges [2-5, 25]. using the simu lated training and testing exa mp les, the optima l (c, γ) was determined as [1, 0.03]. the observation window sizes of 24 and 32 are selected according to the suggestions by guh and tannock [4] and hachicha and ghorbel [1]. 3. results and discussion the concurrent ccps we re a co mbination of any t wo of the basic unnatural ccps. thus, 13 types of concurrent ccp, na me ly usp mixed with utp (usp+utp), usp mixed with dtp (usp+dt p), usp mixed with cp (usp+cp), usp mixed with sp (usp+sp), dsp mixed with ut p (dsp+utp), dsp mixed with dtp (dsp+dt p), dsp mixed with cp (dsp+cp), dsp mixed with sp (dsp+sp), utp mixed with cp (utp+cp), ut p mixed with sp (ut p+sp), dtp mixed with cp (dt p+cp), dt p mixed with sp (dtp+sp), cp mixed with sp (cp+sp) are addressed in this study. table 2 accuracy of ccpr simu lation results in table 2 show that the rate of accuracy for patterns of utp, dtp, utp+usp, utp+dsp, utp+sp, dtp+usp, dtp+dsp, dt p+sp is worse than the other patterns. it is noticed that if the trend pattern was mixed with other patterns, their recognition performance would decrease. 4. conclusions in this paper, a support vector mach ine approach is proposed to identify concurrent ccps for a mu ltivariate process with autocorrelated observations which can be characterized by proceedings of engineering and technology innovation , vol. 3, 2016, pp. 22 24 24 copyright © taeti afirst order autoregressive (ar (1)) model. the numerical results indicate that the proposed model can effectively identify two concurrent identical ccps but for those cases involving one trend pattern and one shift pattern, their reco gnition accuracy deterio rates to around 20% to 50% depending on the autocorrelation coefficients used in the data model. acknowledgement the support of the minister of education (taiwan), under grant most 104-2221-e224 026 is gratefully acknowledged. references [1] w. hachicha and a. ghorbel, “a survey of control-chart pattern-recognition literature (1991–2010) based on a new conceptual classification scheme,” co mputers & industrial engineering, vol. 63, no. 1, pp.204-222, 2012. [2] l.j. xie, n. gu, d.l. li, and z.q. cao, “concurrent control chart patterns recognition with singular spectrum analysis and support vector machine,” co mputers & industrial engineering, vol. 64, no. 1, pp. 280-289, 2013. [3] w. a. yang, w. zhou, w. liao, and y. guo, “identification and quantification of co ncurrent control chart patterns using extreme -point symmetric mode deco mpos ition and e xtre me learn ing mach ines ,” neurocomputing, vol. 147, pp. 260-270, 2015. [4] r. s. guhand j. tannock, “recognition of control chart concurrent patterns using a neural network approach,”international journal of production research, vol. 37, no.8, pp.1743-1765, 1999.  proceedings of engineering and technology innovation , vol. 2, 2016, pp. 20 22 20 identifying islamic values in the embodiment of makassar traditional houses in barombong coastal mursyid mustafa * , ananto yudono, ria wikantari, afifah harisah department of architectural engineering, hasanuddin university, makassar, indonesia. received 11 march 2016; received in revised form 18 april 2016; accept ed 03 may 2016 abstract indonesia is a country with ma jority muslim population. however, in the arch itectural design, people was not significantly using the isla mic values and tend to be affected by western mo dern architecture on their house embodiment. therefore, this research e xplores a house arch itecture based on islamic values, by taking the case of traditional houses on the co ast. the coastal areas chosen because it still has trad itional houses condensed with local cultural community. using observational techniques, interviews with selected informants and questionnaires, and supported by literature rev iew, this research was analyzed by qualitative method. this paper aims to identify isla mic values in the embodiment of makassar traditional houses in the baro mbong coastal area, and enrich the concept of islamic-va lued on house architecture. the research findings are the e mbodiment of makassar traditional houses in baro mbong coastal area wh ich are generally shaped the house terrace called as bale-ba le and paladang, which are dee med to have isla mic values, like : (1) in isla m, the way to visit a house is coming fro m the front, instead of behind. evidently, this way e xisted in traditional house, (2) ro le of the bale-bale as an e xtension of the main roo m which can be used to interact with neighbors and the community (hablunminannas) and paladang can serve as a transition zone to screen guests before we lco ming the m inside the house (privacy) and guests to greet before the residents of the house is ready to receive, (3) in the tradition of building houses, already recognized the co ncept of space-oriented four wind directions (sulapa’ appaka), which significantly according to the concept of islam which require orientation to the qibla oriented home. the study concluded that with the local wisdom possessed by the makassar traditional house on barombong coastal area discovered the islamic values in its realization. ke ywor ds : ba ro mbong coastal, isla mic va lues , makassar traditional house 1. introduction indonesia is a country with ma jority muslim population. however, in the arch itectural design, people were not significantly using the isla mic values and tend to be affected by western mo dern architecture on their house embodiment. therefore this research e xplores a house arch itecture based on islamic values, by taking the case of traditional houses on the coast. the coastal areas chosen because it still has trad itional houses condensed with local cultural community which is in accordance with statement of rapoport (1977, 2004), gustafson (2001), in wan isma il [1] where the built environment is influenced by culture, belief and past experiences of its inhabitants . makassar ethnicity of south sula wesi in tradition of building their house alway s used a stage-house form, which is in macrocosmos view div ided by three parts, lower part (ra wa balla), the body of the house (kale ba lla ), and upper part. fig. 1 example of makassar traditional house *corresponding aut hor. email: mursyidmust afa58@gmail.com proceedings of engineering and technology innovation , vol. 2, 2016, pp. 20 22 21 copyright © taeti a common makassar traditional house consists of bale-bale -and lego-lego, as shown in the fig. 1. idawa rni [2] e xp la in ba le-bale as a place for socialization, play, rest and work, it’s also cheap and easy to ma ke because it does not require specific space for placing. rad ja et al. [3]also e xa mine the types and characteristics of bale bale and identify ba le bale as a co mmon space for socialization. 2. method this study uses observational method with reference to the literature. observation phase of this research was done by using interviews with people from various social strata in the study area. in addition it a lso conducted observations on aspects of social, cu ltural, and physical morphology of tradit ional houses in the study location. as for the results of the study will be analyzed using qualitative methods. 3. results and discussion after conducting interviews with selected informants and distributing questionnaires and observation of social, cultura l, and physical morphology of the tradit ional house in baro mbong coastal obtained the following results: (1) in islam, the way to visit a house is coming from the front, instead of behind. evidently, this way existed in traditional house. mustafa [4] stated that architectural interpretations of the qur'an regarding honoring or welcoming neighbors can be expla ined that the courtesy visit is required to reach the house from the front and saluted the residents of the house a ma ximu m of three times and if it does not get permission from the owner of the house, guests must go home. this way is also found in traditional malay house as islam-practicing country according to othman [5] whereas a spacious main verandah or serambi located in front of the house as a vital design element for receiving and entertaining guests . (2) ro le o f the ba le -ba le as an e xtens ion of the ma in roo m wh ich can b e used to inte ract with ne ighbo rs and the co mmun ity (hab lun minannas ) and p a ladang can serve as a trans it ion zone to sc reen guests befo re we lco ming the m inside the house (p rivacy ) and guests to greet befo re the residents of the house is re ady to rec e ive. ha kim [6] have selected qur’an ic verses and saying of the p rophet rega rding bu ild ing and isla mic c it ies p lann ing p rinc ip le. a say ings of the prophet re lated to p rivacy found in trad it iona l house is: ‘o ye who be lieve, enter not houses other than you r o wn, unt il ye have asked pe rmiss ion and saluted those in the m: th at is best fo r you, in ord er that ye may h eed.’ th is saying of prophet para lle l to the use o f ba le -ba le and pa ladang as transit ion zon e to scre en guest before we lco ming the m ins ide the house, given the o wner o f the house so me priv acy. (3) in the tradition of building houses , already recognized the concept of space-oriented four wind directions, which is assumed to be associated with building placement orientation, or direction of the shoreline (sulapa’ appaka ph ilosophy) which is significantly according to the concept of islam which require orientation to the qibla-oriented home. akil [7] in his research explain sulapa’ appaka philosophy are visible in the coastal cities whose topography is relatively flat. 4. conclusions in this paper, with the local wisdom possessed by the makassar traditional house on barombong coastal area discovered the islamic values in its realization. acknowledgement this research is grate fully ac knowledged by depart ment of arch itecture enginee rin g, faculty of enginee ring, hasanuddin unive rsity , makassar, south sulawesi, indonesia. references [1] w. h. w. is mail, “ cultural determinants in the design of bugis house,” procedia social and behavioral sciences , vol. 50, pp. 771-780, july 2012. [2] i. asmal, “ba le-bale as a space for social interaction,” procedia-social and behavioral sciences , vol. 179, pp. 176 -182, april 2015. [3] a. m. radja, t. suzuki, y. yoshizumi, m. kousaka, k. ya mauchi, s. matsubara, and t. oku, “study on bale bale in lae lae isle, makassar, indonesia,” journal architecture proceedings of engineering and technology innovation , vol. 2, 2016, pp. 20 22 22 copyright © taeti planning, aij, vol. 77, no. 675, pp. 969-978, may 2012. [4] m. mustafa, r. wikantari, a. harisah, and a. muftiradja, “kajian tekstual nilai-nila i keisla man untuk arsitektur rumah tinggal,” iplbi, universitas sam ratulangi, manado, 2015. (in indonesian) [5] z. othman, r. aird, and l. buys, “privacy, modesty, hospitality, and the design of muslim ho mes: a literature review,” frontiers of architectural research, vol. 4, issue 1, pp. 12-23, march 2015. [6] b. s. ha kim, arabic islamic cities: build ing and planning principles, routledge ed, london, 2013. [7] a. akil, a. yudono, i. latief, and a. m. rad ja , “app licat io n of bug inese makassarese culture in makassar city, indonesia,” international review for spatial planning and sustainable development, vol. 2, no. 4, pp. 68-80, 2014.  proceedings of engineering and technology innovation , vol. 2, 2016, pp. 08 10 8 study on the wear resistance of sinter alloying steel by deep cryogenic treatment chao-hsu cheng * department of mechanical and automatic engineering, chung chou university of science and technology, changhua, taiwan. received 19 february 2016; received in revised form 16 march 2016; accept ed 20 april 2016 abstract sintered alloy steel powder with four d ifferent hardenability of fn0200, fn0208, fd0408, and flnc4808 are applied to form the sintered alloy steel. the sintered alloy steel, being cooled down to room te mperature, is further soaked in -196℃ liquid nitrogen for 4hrs deep cryogenic treatment. with deep cryogenic treatment, the retained austenite in the sintered a lloy steel would change the phase and become martensite, and the apparent hardness of sintered alloy steel would be enhanced. further analy zing the effect of deep cryogenic treatment on the wear resistance of sintered alloy steel, the wear resistance would be improved when the apparent hardness of sintered alloy steel is enhanced. however, the deep cryogenic treatment would change the sintered alloy steel microstructures because of distinct hardenability. such research findings show the significant effect of deep cryogenic treatment on the sintered alloy steel with high hardenability, where the best hardness and wear resistance could be acquired from flnc4808 sintered alloy steel through the deep cryogenic treatment. keywor ds : sintered alloy steel, deep cryogenic treatment, hardenability, wear resistance 1. introduction research on sub-zero treat ment actually has been done in the early 20 th century to change the structure and function of steel by lower than free zing point. in 1937, a. p. gu lyaev [1] first researched sub-zero treat ment fo r h igh-speed steel and proposed the theory of sub-zero down to -80℃. so far, it is known that deep cryogenic treatment (-196℃) could enhance the ratio of ma rtensite and reduce austenite residue to improve the wear resistance of steel. deep cry ogenic treat ment is a lso broadly used for ma chinery industry internationally to enhance the wear resistance of ferrous products, such as die steel [2], powder h igh-speed steel [3], and cast iron [4], and the performance o f non -ferrous products, like magnalium [5]. nevertheless, deep cryogenic treatment has not been applied to the research on sintered alloy steel. this study therefore intends to discuss the effects of deep cryogenic treat ment on the me chanical properties and wear performance of sintered alloy steel. 2. method sintered alloy steel with four d ifferent mu ltiplying factors of fn0200 (fe1.2cu2.5ni 0.3c), fn0208 (fe 1.2cu2.5ni0.9c), fd0408 (fe 1.5cu4.0ni0.5mo0.9c), and flnc4808 (fe 2.0cu2.7ni1.25mo0.2mn 0.9c) is utilized in this study. the alloy contents are shown in table 1. the as -sintered steel are further proceeded deep cryogenic treatment fro m roo m te mpera ture down to cryogenic temperature -196℃ with 0.3℃/ min for 4hr and then slowly increased to room te mperature with 0.3℃/ min. finally, sintering sintered alloy steel and cryogenically treated sintered steel are tempered with 180℃ for 1hr to re move residual stress caused by phase change. furthermore, the mechanical p roperty is e xperimented. the above test pieces are measured the s urface vic kers mic ro hardness (hv 1) with 1kg load. the prepared φ5 cylinder wears test pieces and placed on the pin-on-disc wea r test platform, *corresponding aut hor. email: chaohsu@dragon.ccut.edu.t w proceedings of engineering and technology innovation , vol. 2, 2016, pp. 08 10 9 copyright © taeti fig. 1. the wear test platform rotates an skd11 disc with the speed 60 rp m, and the wear test piece is placed 47.5 mm a way fro m the disc center shaft, while a 2 kg load is loaded on top of the wear test piece. each test piece is proceeded the wear test of the wear distance 15,000m, and the measured wear we ight loss is calculated the wear coeffic ient of the test piece with the following equation. k＝       m mm nf m 3  (1) where k: wear coefficient; m: wear amount, g; ρ: density, g/mm 3 ; f: positive load on wear test piece, n; t: wear distance, m. table 1 sintered alloy steel contents fig. 1 pin-on-disc wear test schematic 3. results and discussion 3.1. the hardness of sintered alloy steel the effects of deep cryogenic treatment on the hardness of sintered alloy steel with different hardenability are further analyzed. the as -sintered steel after deep c ryogenic treat ment, where the apparent hardness could be improved by deep cryogenic treatment, regard less of the hardenability. sintered a lloy steel with d istinct hardenability would enhance the apparent hardness with increasing hardenability, revealing that sintered alloy steel would be hardened with deep cryogenic treatment. especially, most of the austenite structure in high ca rbon high hardenability alloy steel (fd0408) is changed into ma rtens ite and the austenite residue is dropped down to the lowest that the apparent hardness is increased from hv1 516 to hv1 733. the improvement of wear resistance could enhance sintered alloy steel parts to a higher level. 3.2. the wear resistance of sintered alloy steel the sintered alloy steel wear test pieces with diffe rent hardenability are placed on the pin-on-disc wear test platform for the 15,000m wear test to measure the wear we ight. the sintering fn0200 a lloy steel with the lowest hardenability is worn the weight 0.0547g with 15,000m, wh ile the sintering flnc4808 a lloy steel with the highest hardenability is mere ly worn 0.0014g with 15,000m; a fter deep cryogenic treat ment, flnc4808 further drops down to 0.0009g, fig. 2. the result e xpla ins that sintered alloy steel could be imp roved the wear resistance with deep cryogenic treatment. fig. 2 wear we ight of sintered alloy steel wea r test substituting the density (ρ), wea r distance (t), test piece wear we ight (m), and wear load (f) of the sintered alloy steel test pieces for eq. (1) to analyze the wea r coefficient, the results are shown in fig. 3. fig. 3 co mparisons of wear coefficient of sintering sintered alloy steel after cryogenic treatment proceedings of engineering and technology innovation , vol. 2, 2016, pp. 08 10 10 copyright © taeti fro m the curve in fig. 3, the sintered a lloy steel with deep cryogenic treatment could effectively reduce the surface wear of sintered alloy steel parts. high carbon high hardenability sintered alloy steel (flnc4808) with deep cryogenic treatment could further enhance the wear coefficient fro m k=0.72(× 10 -6 mm 3 /nm) to k=0.46(× 10 -6 mm 3 /nm), and the metallurgical structure is mostly changed into martensite. among sintered alloy steel with distinct hardenability, the wear coefficient (k=0.46× 10 -6 mm 3 /nm) of flnc4808 sintered alloy steel with deep cryogenic treatment outperforms the wear coeffic ient (k=2.25× 10 -6 mm 3 /nm) of fn0208 sintered alloy steel with deep cryogenic treatment, and even better than fd0408 sintered alloy steel (k=0.93× 10 -6 mm 3 /nm). the result presents that a mong sintered alloy steel with diffe rent hardenability, flnc4808 sintered alloy steel with deep cryogenic treatment could acquire the best wear resistance and effectively reduce the surface wear of sintered alloy steel parts. 4. conclusions (1) as-sintered alloy steel with higher hardenability would enhance the apparent hardness as austenite changes the phase into martensit. (2) austenite of sintered alloy steel with deep cryogenic treatment would change into martensite. such a change enhances the structure apparent hardness. (3) regarding the hardness performance and wear resistance of sintered alloy steel with distinct hardenability, high hardenability alloy steel (flnc4808) with the hardenability above 40 presents the best hardness performance and wear resistance combination. (4) flnc4808 sintered alloy steel with deep cryogenic treatment presents the best wear resistance, which is suitable for machine parts requiring high wear resistance. acknowledgement this research (project no.: nsc 101-262 2-e-235-001-c c3) was cosponsored by the national science council of r.o.c. and jabon international technology co. ltd., which is gratefully acknowledged . references [1] a. p. gu lyaev, “imp roved methods of heat treating high speed steels to improve the cutting properties ,” metallurgy, vol. 12, pp. 65-70, 1937. [2] b. podgornik, et. al., “imp roving tribological p roperties of tool steels thro ugh combination of deep-cryogenic t reat ment and plasma nitriding,” wear, vol. 288, pp. 88-93, 2012. [3] v. leskovšek and b. podgornik, “ vacuu m heat treatment, deep cryogenic treat ment and simultaneous pulse plasma nitrid ing and tempering of p/m s390mc steel,” materials sc ience and engineering: a, vol. 531, pp. 119-129, 2012. [4] t. slatter, r. lewis , and a. h. jones , “the influence of cryogenic processing on wear on the impact wear resistance of low carbon steel and lame lla r graphite cast iron ,” wear, vol. 271, pp. 1481-1489, 2011. [5] k. m . asl, et. al., “ effect of deep cryogenic treatment on microstructure, creep and wear behaviors of az91 magnesium alloy ,” materials sc ience and engineering: a, vo l. 523, pp. 27-31, 2009.  proceedings of engineering and technology innovation , vol. 4, 2016, pp. 01 03 1 spatial analysis of subway ridership: rainfall and ridership kyoung-seon park 1 , sung-hi an 1 , hyunmyung kim 1 , chang-hyeon joh 2 , wonho suh 3,* 1 department of transportation engineering, myongji university, yongin, korea. 2 department of geography, kyung hee university, suwon, korea. 3 department of transportation and logistics engineering, hanyang university, ansan, korea . received 30 january 2016; received in revised form 16 february 2016; accept ed 03 march 2016 abstract in-vehicle congestion of the urban ra ilway system is the most important indicator to reflect the operation state of the urban railway. to provide the good service quality of urban railway, the crowdedness of the urban railway should be managed appropriately. the weather is one of the critical factors for the crowdedness. that is because even though the crowdedness of the urban railway is the same, passengers feel more uncomfortable in ra iny weather condition. indeed if specific sections and stations suddenly are concentrated excessive demand, it will lead far mo re serious problem. there fore, this study analysis the relationship between the nu mber of urban railway passenger and rainfa ll intensity in seoul met ropolitan subway system and then conducts the spatial analysis to deduct passenger demand patterns. this study is expected to be useful base study in order to manage the co ngestion at the urban railway station effectively by considering the different rainfall intensity. ke ywor ds : subway ridership, spatial analysis, ridership analysis 1. introduction the crowdedness of the urban railway is the most important indicator to re flect the operation state of the urban railway. to provide the good service quality of urban railway, the crowdedness of the urban ra ilway should be managed appropriately. to evaluate the crowdedness in the urban railway, quantitative factors and qualitative factors are needed. the weather is one of the critica l factors to influence the crowdedness. this is because even though the crowdedness of the urban railway is the same, passengers feel more uncomfortable in ra iny weather condition. indeed if specific sections and stations suddenly are concentrated excessive demand, it will lead far more serious problem. however, few literatures had been studied to analyze the relationship between the weather and traffic de mand. this is because collecting the traffic de mand and weather data was difficult. however, these data have been opened to the public; it is possible as get detailed information such as passenger data. park and lee analyzed the passenger’s transfer pattern during rainfall in busan [1]. this study showed that the ratio of the passenger’s mode choice is different with the different amount of rain. this research also showed that the ratio of mode choice is more influenced by the rainfall on weekend. lee et a l. conducted the relationship between the number of public transportation passenger and the weather, especially for ra in with the smart ca rd data [2]. this study revealed that both the number of bus and urban railway passengers were reduced in the rain. with these results, they pointed out that the public transportation passengers at seodaemun-gu, dongdaemun-gu, and jung-gu were easily influenced by the rain intensity. yi et a l., ana lyzed the rain intensity and the bus travel time to verify the quality of bus services. the study showed that the quality of the bus services was greatly influenced by rain start in the morning peak hours [3]. however, these researches focused on the specific regions so that it could not show a detail analysis. therefore, this study constructs the database set about the urban railway passenger with the * corresponding aut hor. email: iamwonho@gmail.com proceedings of engineering and technology innovation , vol. 4, 2016, pp. 01 03 2 copyright © taeti rainfa ll intensity in seoul and conducts the sp atial analysis to deduct passenger demand patterns. this paper collects ra infall data which a re collected fro m july to september and urban railway passenger data. with these data, this study analyzed the relationship between the number of urban ra ilway passenger and rainfall intensity. 2. method seoul is one of the biggest cities in the world and eight subway lines are connected so people can reach every single place in seoul. so many people use subway for co mmuting and other travel purposes. according to the kang et al., up to 36 % of people responded that they choose subway as a travel mode [4]. this paper analys es the rainfall intensity to show its impact on de mand of the seoul urban railway with the urban railway passenger data. this study use railway passenger data which are collected fro m ju ly to september 2012 and 2013. daily ra infa ll data which are used are collected fro m the automatic weather stations (aws). to analyse the relationship between the rainfall intensity and the subway passenger ridership, this study selects in that same period rainfa ll data and then conducts a spatial analysis. urban railway pas senger dataset from seoul metro (line 1-4) and seoul metropolitan rapid transit corporation (line 5-8) are obtained and total number of board passengers in july to september 2012, and 2013 are e xtracted. the total number of stations on line 1 to 4 in the 2012-2013 is 119 stations. however, total number of stations on line 5 to 8 is different. in 2012, there are 148 stations on line 5 to 8. however, 9 stations which are gulpocheon station, kkachiul station, bucheon cityhall station, bucheon stadium station, samsan gymnasium station, sangdong station, sinjung-dong station, and chun-ui station data from october 2013 added. 3. results and discussion 3.1. urban railway demand analysis this study set the number of passenger data in level 1 as a standard. based on this standard, this study compares each level’ passenger number and standard and then calculates redu ction ratio. as a result, this study finds out that the number of passengers is decreased during the rainfa ll. in leve l 2, total nu mber of passengers is increased by 0.53% than level 1. at level 3, total number of passengers is decreased by 1.84% and level 4, total number of passengers is decreased by 3.17%. de mand was reduced by 2.65% at level 5. at the level 6, it was reduced by 5.39% . detailed results are given in fig. 1. fig. 1 re lationship between urban railway passengers by rainfall level this study also conducted the passengers’ ridership on each station. the results show that the ridership on yeouinaru station and ttukseom resort station is highly influenced by amount of the rain. this is because people visit these stations for le isure activities. on the other hand, the ridership on samsung station, gasan digital co mplex station, and gangnam station does not have a significant difference by the rain. these stations are highly involved with work trip . contrary to this pattern, hangnyeoul station subway passenger rate is 47% increase fro m relationship. regard less of the rainfall, it can imply that this is because of the hangnyeoul station event. 3.2. spatial analysis on subway network by rainfall level a gis software can be used to store, analyze and allows spatial data layers. first, this study analyzes the high number of the board and alight passenger station in seoul. fig. 2 depicts the high total number of board passenger station analysis data from top 1 to top 10; gang-nam station: 37,181,419persons, jam-sil station: 27,974,601 persons, seoul station: 26,244,942 persons, sillim station: 26,201,372 persons, etc. fig. 2 a lso shows the high total number of alight passenger station: gang -nam station: 38,498,150 persons, jam-sil station: 25,588,647 persons, sillim station: 25,491,673 persons, hongik university station: 25,019,830 persons . proceedings of engineering and technology innovation , vol. 4, 2016, pp. 01 03 3 copyright © taeti fig. 2 board and alight passenger station analysis in seoul subway system 4. conclusions the crowdedness of the urban railway is the most important indicator to re flect the operation state of the urban railway. the weather is one of the critica l factors to influence the crowdedness. that is because even though the crowdedness of the urban railway is the same, passengers feel more uncomfortable in ra iny weather condition. indeed if specific sections and stations suddenly are concentrated excessive demand, it will lead far more serious problem. this study finds out that each station’s passenger is decreased by the rainfall level, e xcept for level 2. the nu mber of passengers in level 6 is decreased about 5.39%. the stations which are highly re lated to leisure activity are sen sitive to the rainfall. also, we ana lyzed the sensitivity area of the rainfa ll level. the station was analyzed by using a high sensitivity for the station leisure activities such as yeouinaru station and ttukseom resort station. on the other hand, sa msung station, gasan dig ital co mple x station, gangnam st ation was analyzed and the sensitivity is low. with the gis progra m, this study conducts the spatial ana lysis to show the relationship between the rainfall intensity and the subway passenger ridership in gis map. this study focuses on the entire date of data and passenger data. to conduct more accurate analysis, the study should consider peak-hour and non-peak hour data. acknowledgement this work is supported by a grant nrf-2014r1a1a2054793 and transportation & logistics research program 15ctap-c097344 of korean government. references [1] k. park and s. lee , “a study on the effect of adverse weather conditions on public transportation mode choice,” journal of civil engineering, ko rean society of civil engineers , vol. 32, no. 1, pp. 23-31, 2012. [2] k. lee, j. eu m, s. you, j. min, and k. yang, “the impact of ra in on public t ransit ridership in seoul,” the korean society for railway, 2014. [3] c. yi, j. ko, y. kang, and t. lee, “the impact of ra infall on public transport service in seoul focusing on the changes in punctuality and speed of bus service”, journal of korea planners association, vol. 46, no. 7, pp. 73-87, 2011. [4] y. kang, c. yi, and s. lee , “a study on the factors affecting crowd ing degree in the subway train considering the seoul metropolitan subway network and the land use of its catchment area,” journal of korea planners association, pp. 203-218, 2014.  proceedings of engineering and technology innovation , vol. 3, 2016, pp. 25 27 25 a wireless sensor network-speech recognition scheme using deployments of multiple kinect microphone array-sensors ing-jr ding * and shih-kai lin department of electrical engineering, national formosa university, yunlin, taiwan. received 22 february 2016; received in revised form 28 march 2016; accept ed 13 april 2016 abstract speech recognition has successfully been utilized in lots of applications recently. with the development of the kinect sensor device from microsoft, speech recognition could be further promoted to be used in an ubiquitous environment where a wireless sensor network using kinect sensors is deployed. this study develops a wire less sensor network (wsn)-speech recognition scheme using deployments of mu ltiple kinect microphone-array sensors. presented speech recognition by kinect-wsn could effectively capture the acoustic data made from the talking speaker and then perform the corresponding voice command control on certain target. in this study, different strategies to deploy multiple kinect microphone-array sensors for constructing an ubiquitous kinect-wsn speech recognition environment are investigated. several different acoustic sensing data fusion methods are also explored for achieving superior performance on kinect-wsn speech recognition. the presented method in this paper is evaluated the efficiency and effectiveness in an 5m×5m laboratory env ironment in which any of four test speakers is to ma ke the voice command anywhere. developed kinect microphone array sensor-deployed wsn speech recognition in this work is finely utilized in various different applications in control. keywor ds : speech recognition, kinect microphone arraysensor, wire less sensor network 1. introduction speech recognition has been a matured technique for human machine-interaction (hci) in the recent years. with the development of internet of things (iots) technology nowadays, the smart home scenario that most of equipment and devices in a fa mily are connected and commun icated with each other via wireless networks will be a practical integrated application. conventional speech recognition viewed as the category of voice control interactions is that the voice command data provided by the specific user is acquired by the mic rophone in a very short distance from the user [1, 2]. such voice-control speech recognition application can be widely seen in speech recognition on the smart phone platform and speech recognition on the central mu ltimedia control panel platform in car. in the new technology of iots nowadays, different to conventional voice command-control speech recognition, in order to control all things connected to the internet in a home, office or other indoor environments, a new strategy for speech recogn ition developments, wireless sensor network (wsn)-speech recognition will be attracted much attention and a new and challengeable technique issue. this study explores the utilization of the kinect sensor [3, 4] for sensing and then acquiring the acoustic voice command of the user in an office environment. as many persons know, in addition to gesture recognition by kinect [5-7], the kinect device can also be used to perform speech recognition due to the embedded microphone array design composed of four microphones [8]. speech recognition in this work will be performed in an acoustic sensing area that is properly deployed by multiple kinect microphone-arrays. in the presented scheme of wsn-speech recognition by multip le kinect microphone array-sensors in this paper, technical issues such as the (1) deployment method of the kinect mic rophone array, (2) the establishment of client server-based wireless sensor network by kinect sensors, (3) investigations of acoustic sensing data fusion methods, and (4) the possible application with practice in a real life using presented wsn-speech recognition by kinect microphone array sensors will be considered, which will be detailed in the following section . * corresponding aut hor. email: ingjr@nfu.edu.t w proceedings of engineering and technology innovation , vol. 3, 2016, pp. 25 27 26 copyright © taeti 2. speech recognition via wireless sensor network by using kinect microphone array-sensors the frame work of wsn-speech recognition by kinect microphone array-sensors explored in this study mainly contains kinect sensor deployments by two kinect microphone-arrays, client-server wsn establishments using tcp/ip protocol, acoustic sensing data fusion using a simple and computationally fast strategy. the developed framework is further performed in an application of voice sensing and remote control to the multimedia player component on a smart phone, which is depicted in fig. 1. fig. 1 wsn-speech recognition by deployed kinect microphone array-sensors and its control application fig. 2 sensed data fro m a kinect microphone array-sensor for data fusion calculations (four-channel voice data contained simultaneously in the unique kinect sensor) as depicted in fig. 1, a voice command fro m a speaking user is sensed by two kinect microphone array-sensors that are properly deployed in an office space. each sensed data from each of these two kinect sensors is sent to the server via the tcp/tp protocol, and the server end performs data fusion calculations for determining the recognition result of the sensed voice command. after the data fusion estimate, the recognized voice command is then sent to the smart phone device via the bluetooth protocol to carry out a series of functional control on the multimedia player application program. in this work, two kinect microphone array-sensors are designed to be appropriately localized inside to acquire almost most of all possible voice data. fig. 2 shows sensed data in a form of four channels. all these data come fro m the unique kinect microphone array-sensor. two kinect microphone array-sensors deployed in this work are composed of 2 sets of the four-channel sensed voice data. these data are then considered the content of the voice command using data fusion calculations. the data fusion strategy employed in this study is a voice energy-based method. as shown in fig. 2, each voice data fro m the kinect sensor have the different values of energy. the value of the voice energy is dependent on the amplitude value of the data in certain time duration. the microphone in the kinect microphone array has the large value of voice energy in case the voice data source (i.e. the speaking user) is located extreme ly near the microphone. co nversely, when the microphone in the kinect microphone array is far away from th e voice data source, the estimated voice energy to this microphone will be significantly small. based on the above design thought-line, the primary principle of the data fusion method in this study is that the microphone with the sensed data of large-sized values will have more effects on the recognition decision of voice commands. the simplest method based on such the designed fusion principle is that the data fusion result is the recogn ition outcome of the microphone receiver where the sensed data has the largest values of energies. in the part of control applications, the blu etooth (bt) protocol is employed in this work to handle transmissions of the fused recognition command. the bt connection tunnel is firstly established in an initialization process to form a peer-to-peer connection pair between the server (the command provider) and the end -device of the smart phone (the command receiver). for speeding up command transmissions via bt, a co mmand table containing a series of labels, each label with a text form representing a corresponding voice command, is properly devised. the multimedia player application platform in the smart phone will be finely operated by “remotely sensed voice commands made by the speaking user” under the regulation of the presented method. proceedings of engineering and technology innovation, vol. 3, 2016, pp. 25 27 27 copyright © taeti 3. conclusions in this paper, a wireless sensor network-speech recognition approach is presented by deploying the kinect microphone array-sensors. co mpared to conventional speech recognition, the presented frame work considering sensor deployments, sensing data fusion, wireless communication scheme establishments, and possible e xtension application with p ractice provides an acoustic sensing way for command control in the application of internet of things. in addition, the presented approach with the use of kinect sensors will a lso avoid the property of ‘surveillance’ and therefore can be much mo re acceptable by the users. acknowledgement this research is partially supported by the ministry of sc ience and technology (most) in taiwan under grant m ost 104-2815-c-150-01 7-e. references [1] 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, no. 1, pp. 38-47, 2014. [2] i. j. ding and y. m. hsu, “an hmm -like dynamic t ime wa rping scheme for automat ic s peech recognition,” mathematica l proble ms in engineering, vo l. 2014, artic le id 898729, 8 pages , 2014. [3] i. tashev, “kinect development kit : a toolkit for gestureand speech based human-machine interaction,” ieee signal processing magazine, vol. 30, no. 5, pp. 129–131, 2013. [4] z. zhang, “m icrosoft kinect sensor and its effect,” ieee mult imedia , vol. 19, no. 2, pp. 4-10, 2012. [5] i. j. ding and c. w. chang, “an eigenspace-based method with a user adap tation scheme for hu man gesture recognition by using kinect 3d data,” applied mathemat ical modelling, vol. 39, no. 19, pp. 5769-5777, 2015. [6] i. j. ding and c. w. chang, “feature design scheme for kinect-based dtw human gesture recognition,” multimedia tools and applications , pp. 1-16, july, 2015 [7] k. qian, j. niu and h. yang, “developing a gesture based re mote human-robot interaction system using kinect,” international journal of smart ho me, vol. 7, no. 4, pp. 203-208, 2013. [8] k. ku matani, t. ara kawa , k. ya ma moto, j. mcdonough, b. ra j, r. singh and i. tashev, “microphone array processing for distant speech recognition: towards rea l-world deployment,” proc. asia-pacific signal & information processing association annual summit and conference (apsipa asc), 2012.  proceedings of engineering and technology innovation , vol. 3, 2016, pp. 01 03 1 hybrid ofdm/ook modulations in ocdma scheme for free space optics chih-ta yen * , chia-yu liu department of electrical engineering, national formosa university, yunlin, taiwan . received 01 february 2016; received in revised form 15 march 2016; accept ed 02 april 2016 abstract the study proposes hybrid analog/digital transmission format scheme wh ich integrated optical code-division mult iple -access (ocdma) and polarization mu lt iple xing technique in free space optics (fso) trans mission. orthogonal frequency division mu ltiple xing (ofdm ) transmits as the analog format and (on -off keying) ook transmits as digita l format in the study, respectively. in the proposed hybrid ocdma system, it has high-speed transmission, signal security and low cost…etc. the multip le access interference (mai) can be e fficiently e liminated by using the balanced detection scheme at the receive end. keywor ds : optical code-division multiple-access (ocdma), free space optics (fso), orthogonal frequency division multiplexing (ofdm), multiple access interference (mai) 1. introduction in recent years, peoples become closer and closer relationship between the networks. be cause of hybrid coaxia l cable transmission network change into a fiber optical trans mission network, making people have a better internet e xperience. in the current study, we use the method of different states of polarizations to transmit analog/digital signals and th en into optical code-division mult iple -access (ocdma) networks. due to polarization modulator without using electronic circuit or impressed voltage to control, and has the same intensity of light signals in the outputs, which can mit igate fiber nonlinear e ffects and ma king the overa ll structure greatly improve the reliability of the tran smission side. at present, wired transmission interface is using fiber or coa xial cable to transmit signal. but in some place because of terrain effect (e.g., cliffs desert, both ends of the bridge). wireless transmission system is one of the solutions to solve topographical constraints , but it is always disturbed by channel interfe rence and shelter, severely reducing its effectiveness . we take advantage of the free space optics (fso) approach as the way for last mile trans missions [1]. in this paper, we present optical spectral amp litude coding ocdma (sac-ocdma) combin ing with polarization mu lt iple xing to transmit analog signal of orthogonal frequency division mu ltiple xing (ofdm) and dig ital signal of on-off keying (ook) signals at the sa me time . in the past year, sac-ocdma system is attracted more attention because of the multip le access interference (mai) elimination and pre serve the orthogonality among the users in the system [2]. ofdm technique has been widely adopted in broadband wired and wireless co mmunicat ion systems , because it is robustness to the inter-symbol interfe rence (isi) and multi-path fading by a dispersive channel [3]. 2. system structure description fig. 1 and 2 illustrate the structure of the proposed base stations (transmitters) and control station (receivers). this ocdma mult iple xing structure can transmit analog signal and digita l signal at the same time and still preserve the ability of mai e limination. we transmit ofdm signal as analog signal and ook signal as digita l signal. in this hybrid system, each base station needs a spectral encoder and a decoder in the receiver. the dfb laser array source is co nnected a 1 ( 1)n  splitter which p roviding light source to each base station, shown in fig. 1. the dbf laser source performs as optical carri* corresponding aut hor. email: chihtayen@gmail.com proceedings of engineering and technology innovation , vol. 3, 2016, pp. 01 03 2 copyright © taeti ers pass through a polarizer a polarization co ntrolle r (pc), and a polarization bea m splitter (pbs). the pbs allocates the orthogonal horizontal and vertical states of polarizat ions (sops) upper and lower arms, respectively. between base station and control station, fso is used to be this structure channel. the control station received a composite signal comprising the transmitted chips of all the active base station. as shown fig. 2, the received signal in the control station is divided in to (n−1) -th parts by using the 1×(n−1) splitter, and passed into each receiver. the signal was produced vertical and horizontal po larization states by using pbs. two sops of vertica l and horizontal output from the pbs which are split into an upper branch and the lower branch. the two diffe rent signals of vertical and horizontal passed balance detector, wh ich co mputed the correlation d iffe rence between the two spectra and eliminate the interfe rence of base station with un-matching codeword ind ividually. the upper branch uses ofdm de modulator to revert analog origina l signal and lo wer branch uses digital ook de modulator to revert dig ital ook signal. encoder #j ofdm transmitter #l pc pbs#1 vertical sop horizontal sop mzm mzm ( ) l c i base station #l 1 x k c o m b in e rdfb laser array source 1xk splitter pbc base station #1 base station #(n-1) digital signal transmitter #l fso channel fig. 1 the transmitter of proposing system 1x2 splitter 1x2 splitter ( ) l c i ( )lc i ( ) l c i ( )lc i + + ofdm demodulator #l desired data #l receiver #l decoder #j horizontal sop 1 x (n -1 ) s p litte r pbs#2 vertical sop receiver #1 receiver #(n-1) control station digital signal demodulator #l desired data #l pd #l, v1 pd #l, v2 pd #l, h1 pd #l, h2 fig. 2 the receiver of proposing system before analysis of the proposed system [4], the ofdm signal of the k-th base station is )(, to rfk : )2cos()( ,,, tfato rfkrfkrfk  (1) where ,k rfa is the a mplitude of ofdm signal, and ,k rff is the carrier frequency. and the analog ofdm signal is modulated by a mult i-wavelength dfb laser as optical carriers and then pass through awg encoder is:     cn i krfksk irectictopto 1 , )()()(1)(  (2) where ( )kc i denote the i-th element of the k -th whc codeword,  is the modulation inde x, sp is the transmitted optical power, and ( )rect i function is given by:     0 0 ( ) 2 2 2 2 2 c c c c rect i u v v n i n u v v n i n                        (3) at the receiver end, the instantaneous power spectral density (psd) of the vertical and horizontal sop’s received optical signals can be expressed by the following: , 1 1 ( , ) 1 ( ) 2 ( ) ( ) cnk sr v s o rf k i k p r v t p q t c i rect i             (4)       k k nc i kk sr h irecticb p vr 1 1 )()( 2 )(  (5) where psr is the received effective power at the input port of each receiver,  is the light source bandwidth, k is the number of base station. the relat ionship between bit-error-rate (ber) can employ as:   1 / 8 2 ber erfc snr (6) 3. performance analysis and discussion fso performance by several para meters for reference including geo metric loss, link marg in, the received power and bit error rate. in this paper, it considers the received power and bit error rate [5]. fig. 3(a) and 3(b ) illustrates the bit error rate (ber) re lative to active users in d ifproceedings of engineering and technology innovation , vol. 3, 2016, pp. 01 03 3 copyright © taeti ferent bit rate at the analog/digital signals, re spectively. it had more active user when it had low bit rate. (a) analog signal (b) digital signal fig. 3 illustrates the bit error rate (ber) rela tive to active users in different bit rate 4. conclusions in this study, we present hybrid an alog/digital spectral ocdma system. th is system transmits two different kinds of format signal wh ich contain analog and dig ital signals by using polarizat ion multip le xing technique. the proposed hybrid system can transmit analog and digital signals at the same time such as digital t v and mob ile phone system simu ltaneously. fso is the main t ransmission channel in the study, which has the advantages of low cost, easy construction and terrain without restrictions … etc. the simu lation results show that both ofdm signal as analog signal and ook modulation as digital signal in the proposed hybrid ocdma system perfo rm good performance in fso transmission. acknowledgement this study was supported in part by the ministry of sc ience and technology most 104-2221-e-150-044. references [1] p. t jondronegoro, “free-space optics for fixed wire less broadband,” pp. 2-20, oct. 2004. [2] z. wei, h. m. h. shalaby, and h. ghafouri-shira z, “modified quadratic congruence codes for fiber bragg-grat ing-based spectral-a mp litude-coding optical cdma systems,” j. lightwave technol., vol. 19, pp. 1274-1281, sep. 2001. [3] m. se lvi and k. murugesan, “performance of ofdm based fso commun ication systems using m-ary psk modulation,” international journal of co mputer applications , vol. 49, pp. 41-45, july 2012. [4] j. f. huang and c. t. yen, “phase noise suppression in multilevel optical code-division multiple-access network coding system with embedded orthogonal polarizations ,” optical engineering, vol. 45, pp. 065005, june 2006. [5] v. sharma and g. kaur, “modelling of ofdm-odsb-fso trans mission system under different weather conditions ,” advanced computing and co mmunication technologies (acct ), pp. 154-157, april 2013.  proceedings of engineering and technology innovation , vol. 2, 2016, pp. 14 16 14 synthesis and annealing study of siliconand fluorine-containing low bandgap conjugated polymers for solar cell applications cheng-hsien wu, cian-hao huang, pei-ni song, tzong-liu wang * department of chemical and materials engineering, national university of kaohsiung, kaohsiung, taiwan. received 07 march 2016; received in revised form 12 april 2016; accept ed 09 may 2016 abstract two donor-acceptor alternating conjugated copolymers were synthesized as the donor materials of the active layer in polymer solar cells via stille coupling reaction. both copolymers consisted of dithienosilole as the donor unit and nonfluorinated 2,1,3-benzooxadiazole or fluorinated 2,1,3-benzooxadiazole as the acceptor unit. the nonfluorinated polymer and fluorinated polymer were designated as p1 and p2, respectively. the structures of copolymers were confirmed by ft-ir, 1 h nmr, and 13 c nmr. optoelectronic properties of the polymer were investigated and observed by uv-vis spectrum, photoluminescence spectrum, and cyclic voltammetry. both polymers exhibited a panchromatic absorption ranging from 300 nm to 1100 nm and displayed low band gaps of 1.46 ev and 1.42 e v, respectively. both of the synthesized materials were used as the donor material in the bulk heterojunction (bhj) solar cells and then power conversion efficiency (pce) measurements were conducted in different weight ratios of the polymer:pc61bm blends . compared to the nonfluorinated one, the fluorinated polymer exhibited a higher pce of 2.67% at room temperature under the illumin ation of am 1.5 (100 mw/cm 2 ). study of the effect of annealing on the performance of the p1/pc61bm devices using a 1:2 blend ratio of p1:pc61bm exhibited the highest short-circuit current density of 5.88 ma/cm 2 and a power conversion efficiency of 2.76% at annealing temperature of 125 o c. the bulk heterojunction polymer solar cell employing p2 and pc61bm at a blend ratio of 1:2 exhibited the highest short-circuit current density of 6.44 ma/cm 2 and a power conversion efficiency of 3.54% at annealing temperature of 150 o c. keywor ds : donor-acceptor, solar cell, dithienosilole, 2, 1, 3-benzooxadiazole, low band gap, power conversion efficiency 1. introduction except the “weak donor–strong acceptor” concept for the design of d-a alternating copolymers with low band gaps and high vocs for solar cell applications [1], adding elec tron-withdrawing groups to the polyme r backbone is also considered as an effic ient approach. recently, fluorine has attracted much attention as an electron-withdrawing group in the synth esis of conjugated polyme rs for h igh -efficiency solar cells [2-3]. according to a theoretical study by brédas et a l., the introduction of fluorine atom as a substituent of the conjugated polymer would lo wer both of the hom o and lum o energy levels of the synthesized polymer [4]. on the othe r hand , s ilo le-cont ain ing a romat ic mo iet ies hav e attracted mu ch attent ion as ne w bu ild ing units o f con jugated po ly me rs because of the ir un ique optoe lect ron ic p ro p ert ies. as rega rds the ir ca rbon ana logues, silo le-conta in ing po ly me rs have quit e lo w-ly ing lum o leve ls, wh ich are att ributed to the fact that the -orb ita l of the si-c bond can e ffect ive ly interacts with the -orb ita l o f the butad iene frag ment [ 5]. by trans forming pcpdt bt to the s ilo le-conta in ing po ly me r, poly{[4,4'-b is(2-ethy lhe xy l)d ith ieno (3,2-b; 2', 3'-d )silo le ]-2,6-d iy l-a lt-(2,1,3-ben zoth id ia zo l e)-4,7-d iy l} h as been re cent ly rea lized and a pce of as high as 5.6% has been ach ieved [6]. there fo re , the d ithieno[3,2-b :2',3'-d]-s ilole (dt s) un it has frequent ly used as th e s i*corresponding aut hor. email: t lwang@nuk.edu.t w proceedings of engineering and technology innovation , vol. 2, 2016, pp. 14 16 15 copyright © taeti lo le -conta in ing donor un it in the deve lop ment of high-efficiency pscs . here in, we have investigated the effect of silole-containing and fluorination on the photovoltaic performances of the d -a conjugated copolymers. t wo d-a copoly mers co mprising dts unit and nonfluorinated and fluorinated 2,1,3-benzoo xad iazo le (bo) unit has been synthesized to study the effect of silicon and fluorine substitution. the effect of annealing on the device performance of the poly mer solar ce lls fabricated fro m the nonfluorinated and fluorinated polymers has been investigated in detail. 2. method 2.1. synthesis of polymers in this study, two d-a type copolymers were synthesized via stille coupling reaction. for convenience, they are designated as p1 and p2. the synthetic routes are shown in fig. 1. fig. 1 polymer synthesis 2.2. annealing study the dev ic e st ruct ure o f pscs fo r cu rrent density -vo ltage (j -v) me asu re ments is it o/ pedot : pss/po ly me r: pc61bm/ lif/a l. the films of act ive layers we re ann ea led d irect ly on top o f a hot p late in the glove bo x, and the te mpe ratu re was mon ito red by using a thermo couple touch ing the top o f the su bstrates. a fte r re mova l fro m the hotp late , the substrates were immed iat e ly put onto a meta l plate at the roo m te mp eratu re . ult rav io let -v isib le (uv-vis ) spectroscopic ana lysis was conduct ed on a perkin– elme r la mbda 35 uv-vis spectrophoto mete r. the j-v cu rves we re measured using a ke ith ley 2400 sourc e mete r, unde r illu min at ion fro m a sola r simu lato r. the intensity o f so la r s imu lato r was set with a prima ry re fe rence c e ll and a spectra l correct ion facto r to g ive the pe rformanc e u n der the am 1.5 (100 mw/c m2) g loba l re fe rence spectrum (iec 60904-9). 3. results and discussion fig. 2 j-v characteristics of devices for p1 a fter annealing at different te mperatures for 30 min fig. 3 j-v characteristics of devices for p2 a fter annealing at different te mperatures for 30 min figs. 2 3 show the j-v characteristics under white light illu mination (100 mw/c m 2 ) for photovoltaic devices subjected to thermal annealing. for the p1 poly mer, the device annealed at 125 °c shows the best efficiency of 2.76% , while the p2 poly mer shows the optimu m efficiency of 3.54% at 150 °c. thus, it is c lear that proceedings of engineering and technology innovation , vol. 2, 2016, pp. 14 16 16 copyright © taeti annealing treatment affects the device perfo rmance significantly and that there e xists opt imu m range of the annealing te mperature. it is noteworthy that the voc values of p2 a re higher than those of the p1 polymer because the introduction of f into the conjugated backbone has lowered the hom o and lumo energy levels of the copolymer. due to the values of voc a re re markable for the p2 poly me r, the power conversion efficiencies of the solar ce ll for p 2 a re much improved. 4. conclusions two d–a type copolymers based on dts as the donor unit and nonfluorinated and fluorinated bo as the acceptor unit have been synthesized and e mp loyed as the donor materia ls in the active layer of bhj -type polyme r solar cells. as a result of annealing treatment at an optimu m condition (125 °c/30 min ), the pv cell performance of p1 was dra matica lly enhanced and the power conversion efficiency of device reached to 2.76%. through the modification by fluorination on the bo units, the photovoltaic pe rformance of p 2 at 150 °c was much imp roved due to the increased jsc and enhanced voc and the power conversion efficiency of the device reached 3.54% under wh ite light illu mination (100 mw/cm 2 ). acknowledgement we gratefully acknowledge the support of the ministry of science and technology in taiwan through grant nsc 99-2221-e-390-001-m y3. references [1] h. zhou, l. yang, s. stoneking, and w. you, “a weak donor-strong acceptor strategy to design ideal poly mers for organic solar cells ,” acs appl. mater. & interfaces vol. 2, pp. 1377-1383, 2010. [2] h. y. chen, j. h. hou, s. q. zhang, y. y. liang, g. w. yang, y. yang, l. p. yu, y. wu, and g. li, “poly me r solar ce lls with enhanced open-circuit voltage and efficiency,” nat. photonics vol. 3, pp. 649-653, 2009. [3] h. x. zhou, l. q. yang, a. c. stuart, s. c. price, s. b. liu, and w. you, “development of fluorinated benzothiadiazo le as a structural unit for a poly mer solar ce ll of 7 % efficiency,” angew. chem., int. ed. vol. 50, pp. 2995-2998, 2011. [4] j. l. brédas and a. j. heeger, “influence of donor and acceptor substituents on the electronic characteristics of poly (para phenylene vinylene) and poly (paraphenylene),” che m. phys. lett., vol. 217, pp. 507-512, 1994. [5] s. ya maguchi, y. ita mi, and k. ta mao, “ group 14 metalloles with thienyl groups on 2,5-positions:  effects of group 14 ele ments on their π-e lectronic structures ,” organometa llics , vol. 17, pp. 4910-4916, 1998. [6] j. hou, h. y. chen, s. zhang, g. li, and y. yang, “synthesis, characterizat ion, and photovoltaic properties of a low band gap polymer based on silole-containing polythiophenes and 2,1,3-benzothiadia zole,” j. am. che m. soc., vol. 130, pp. 16144-16145, 2008.  proceedings of engineering and technology innovation , vol. 3, 2016, pp. 07 09 7 design of slim led coupler for collimated light source chun-hao sun, tun-chien teng * department of mechatronic engineering, national taiwan normal university, taipei, taiwan. received 25 february 2016; received in revised form 28 march 2016; accept ed 16 april 2016 abstract this study proposed a slim coupler to collimate the light e mitted fro m an led, which can be used as a collimated light source. the coupler is substantially a 2-d co mpound parabolic collector (cpc); its bottom surface has longitud inally e xtending v-groove microstructures thereon. the parabolic side wa lls re flect the propagating light to converge only in the transverse direction, but the v-groove mic rostructures reflect the light and make its propagating direction gradually rotates around a central a xis of the cpc. therefo re, the angular d istribution of the light finally converges after several times of reflection on both side walls and v-groove microstructures. moreover, the illu minance distribution on the outlet of the v-groove cpc becomes more uniform than a cpc without v-groove mic rostructures. the effects of v-groove microstructures on both angular and illu minance distributions of the light emerging fro m the outlet of the cpc is analyzed, and the feasibility of providing a uniform co llimated light source is verified by conducting simula tion. keywor ds : collimation, coupler, led, compound parabolic collector, illumination 1. introduction the liquid crystal display (lcd) needs an e xtra p lanar light source such as a backlight. as energy-saving is paid more attention an u ltra-collimated planar light source (ucpls) becomes an attractive solution to ra ise energy efficiency of the lcd. the ucpls concentrates the light toward the observer to avoid waste of energy. moreover, it is essential for achieving some advanced functions to greatly raise energy efficiency. severa l typica l designs of ucpls were proposed. the first is to fabricate a light guide plate (lgp) with special microstructures thereon that directly reflect the light propagating within the lgp into the norma l direct ion [1]. the second is to form grat ing dots on the lgp surface to diffract the light propagating within the lgp into the normal d irect ion. [2]. the third is to utilize an lgp co mposed of a stack of mu ltip le layers of diffe rent refractive inde x to ma ke the inner light emerge with narrow vertica l angular distribution [3]. the fourth is to pre-collimate the light emitted fro m an led by an ext ra optical co mponent before it enters an lgp and then to reflect the collimated light into the norma l d irection through microstructures on the lgp [4]. a mong those designs, using an e xtra pre-collimating co mponent is a simple r way but needs thicker or wider volu me to accommodate the component, which is adverse to a slim backlight. thus, we proposed a ucpls design, in which a slim led coupler was used to pre-collimate the light e mitted fro m an led and the overall thickness was 5 mm [5]. in this paper, we further analy ze the effects of para meters of the coupler on both spatial and angular d istrib ution of the output light to attain the optima l condition. 2. principle the led coupler used in the backlight must account for both light mixing and light-collimation. the coupler has a profile of the cpc to convert divergent light entering a small area (inlet) into collimated light emerging from a large area (outlet). the relationship between the angular distribution range (i.e., the “half-angle”) of the light entering the inlet and that of the outlet of cpc can be expressed as follows : 1 2 2 2 1 1 sin sin t n t n    (1) 2 11 2 22 2 1 )sin( )sin(   n n a a  (2) * corresponding aut hor. email: walt er.teng@ntnu.edu.tw proceedings of engineering and technology innovation , vol. 3, 2016, pp. 07 09 8 copyright © taeti eq. (1) is applicable for the 2-d case, where t1 and t2 is the thickness of the inlet and outlet of the cpc, respectively; eq. (2) is applicab le for the 3-d case, where a1 and a2 is the area of the inlet and outlet of the cpc, respectively; and n1 and n2 respectively represent the refractive inde x of the mediu m surrounding the inlet and outlet. the two equations indicate that the degree of collimation of the light e merging fro m the outlet depends on the ratio of the outlet area to the inlet area. as the ratio increases, both width and thickness of the cpc outlet become larger, thus adverse to backlight application. moreover, output a uniform collimated light bea m is another focus if we intend to transversely connect mu ltip le of the cpcs side by side to form sufficiently wide and uniform co llimated light source for the lgp. for a cpc that can be accommodated in a slim backlight whose space is relatively wide but very thin, the horizontal angular distribution of the output light of the cpc would be re latively narrow, but the vertical angular distribution would be re latively wide. to solve this issue, an approach combining a cpc with v-groove microstructures was proposed. the coupler was designed as a thin and long cpc with a rectangular cross -section, and an array of v-groove microstructures longitudinally e xtended along the z-a xis was form on its bottom surface as shown in fig. 1. the light propagating along the z-a xis within a thin but wide cpc e xhib its a narrow horizontal angular distribution (projected on the x-z plane) and a re latively wider vertica l angular distribution (projected on the y -z p lane). when a light beam with a narrow horizontal angle but comparatively wider vertical angle is incident to the slope facet of the v-groove microstructures, it is reflected by total internal re flection (tir), and then continues propagating conversely with a narro w vertica l but co mparatively wider horizontal angle. such a light beam with a wider horizontal angle is more likely to hit the curved surface on the left or right side of the cpc. consequently, it is reflected by tir, and then continues propagating with a narrower horizontal angle; in other words, the beam b ecomes more collimated. after many similar cycles, the original light propagating within the cpc converges both vertically and horizontally. in the process, the length of the cpc and the angle of the apex of the v-groove microstructures have effects on both spatial and angular distributions of the output light. fig. 1 cpc with v-groove microstructures 3. results and discussion in this study, we established an optical model whose related para meters are detailed as follows. a cpc was made of poly methyl methacrylate (pmma), whose dimensions of its inlet and outlet were 3.6 1.2 (mm) and 13.82 2.4 (mm), respectively; its length was 50– 70 mm. a series of simu lations we re performed for different conditions such as: ape x angles of v-grooves and types of the profile of the cpc. ape x angles included 90 o , 110 o , 130 o , and 150 o ; types of the profile had three: width and thickness simultaneously increasing (case a); width increasing first and then thickness (case b); thickness increasing first and then width (case c). table 1 uniformity of illu minance vs. ape x angles case a case b case c 0 o 0.80/0.87 0.88/0.90 0.83/0.85 90 o 0.69/0.76 0.79/0.90 0.69/0.85 110 o 0.92/0.94 0.97/0.98 0.95/0.95 130 o 0.84/0.99 0.98/0.99 0.80/0.97 150 o 0.89/0.98 0.98/0.99 0.78/0.93 table 1 lists uniformity of illu minance in the three cases under the conditions of different apex angles. uniformity is defined as the ratio of minimal illu minance to ma xima l. angle of zero means the bottom of the cpc is bare flat; the uniformity data behind the slash are those of the cpc that is e xtended by an extra length of 20 mm with a constant cross -section. the results indicate that the apex angle of 90 o is adverse to uniformity for the three cases; apex angles of 110 o , 130 o , and 150 o imp rove uniformity for case a and b; only the apex angle of 110 o improves uniformity for case c. case b performs the best uniformity; case c performs the worst. however, a ll the uniformity is improved for the three cases when the cpc is extended. proceedings of engineering and technology innovation , vol. 3, 2016, pp. 07 09 9 copyright © taeti angular distributions of output light of the cpc for case a, b, and c are shown in fig. 2. the left insets are those in the vertical direct ion; the right insets are those in the horizontal direc tion. the effects of ape x angles of v-groove microstructures on angular distribution are demonstrated. the v-groove microstructure narrows the angular distribution in the vertical, and up to 47% at most as co mpared with the cpc without microstructures. however, it also slightly broadens the angular distribution in the horizontal. a mong the three cases, case b has narrower and smooth vertical angular d istrib ution, but its horizontal angular distribution is wider. considering uniformity of illu minance, case b is an optimal solution. fig. 2 angular distributions of output light of the cpc: (a) case a; (b) case b; case c 4. conclusions in this paper, effects of v-groove mic rostructures on both angular and illu minance distributions have been demonstrated for the three types of cpcs. in the slim volu me , the vertica l angular distribution and illu minance uniformity of the output light is narrowed and improved, respectively. the cpc of the profile with width increasing first and then thickness is an optima l solution for a slim coupler p roviding a uniform collimated light source because of its smooth angular distribution and e xcellent illu minance uniformity. references [1] j. h. lee , h. s. lee, b. k. lee , w. s. choi, h. y. cho i, and j. b. yoon, “simp le liquid crystal display backlight unit co mprising only a single-sheet mic ropatterned polydimethylsilo xane (pdm s) light-guide plate,” opt. lett., vol. 32, no. 18, pp. 2665-2667, 2007 [2] s. r. park, o. j. kwon, d. shin, and s. h. song, “ grating mic ro-dot patterned light guide plates for led backlights,” opt. express , vol. 15, no. 6, pp. 2888-2899, 2007. [3] j. w. pan, c. w. fan, “high lu minance hybrid light guide plate for backlight mo dule applicat ion,” opt. expres s, vol. 19, no. 21, pp. 20079-20087, 2011. [4] w. h. yang, h. h. lin, c. j. hsu, and y. n. pao, “ led coupler lens array for one dimensional co llimat ing backlight,” idw., vol. 11, pp. 1399-1400, 2011. [5] t. c. teng and l. w. tseng, “slim p lanar apparatus for converting led light into collimated polarized light uniformly e mit ted fro m its top surface,” opt. express , vol. 22, no. s6, pp. a1477-a1490, 2014.  proceedings of engineering and technology innovation , vol. 3, 2016, pp. 31 33 31 experimental study of the crack control of concrete by self-healing of synthetic fiber reinforced cementitious composites synthetic fiber heesup choi 1,* , masumi inoue 1 , takechi ai 1 , tomoya nishiwaki 2 , myungkwan lim 3 , hyeonggil choi 4 1 department of civil engineering, kitami institute of technology, hokkaido, japan . 2 department of architecture and building science, tohoku university, sendai, japan . 3 graduated school of engineering, hankyong national university, kyonggido, korea . 4 faculty of environmental technology, muroran institute of technology, hokkaido, japan . received 25 january 2016; received in revised form 10 february 2016; accept ed 06 march 2016 abstract in this study, it is possible to disperse effec tively cracked using synthetic fiber, an e xa mination of the most suitable self-healing conditions was performed on the above crack width 0.1mm. as a result, e ffective c rack dispersion using polyvinyl alcohol (pva ) fibers with pola r oh groups, as well as improved self-healing for cracks that are la rger than 0.1 mm in width, posing concerns of co2 gas and cl penetration, were observed. also, co3 2 reacts with ca 2+ in the concrete crack, resulting in the precipitation of a carbonate compound, caco3. based on this, it is deemed possible for the recovery of effec tive watertightness to be made fro m c racks that are la rger than 0.1 mm in width. in addition, it was determined that, as for the most suitable self-healing conditions in the inside and surface of the cracks, calciu m hydro xide (ca(oh)2) solution with co2 mic ro-bubble was more effective in pro moting the self-healing capability than water with co2 micro-bubble. keywor ds : micro crack, pva, co2 micro-bubble, s elf-healing,ca(oh)2, caco3 1. introduction although concrete is one of the most widely used construction materials, it is characterized by substantially low tensile strength in comparison to its compression strength, and occurrence of cracks is unavoidable. in addition, cracks progress due to environmen tal conditions including da mage by freezing, neutralization and salt damage, etc. moreover, detrimental da mages can occur in concrete structures due to the permeat ion of deteriorat ing ele ments such as cl and co2, etc. meanwh ile, under an environment in wh ich moisture is being supplied, if the width of the crac k is sma ll, a phenomenon of self-healing in which the portion of the crack is filled in due to the rehydration of the ce ment particles and precip itation of caco3 has been confirmed. as for the self-healing mechanis m of concrete, the reaction between ca 2+ in concrete and co3 2 dissolved in water produces caco3, a carbonate compound that does not easily dissolve in water. this phenomenon leads to the filling and c losing of crac ks [1]. the calc ite crystal reactions are shown in equations [1-3] h2o + co2⇔h2co3 ⇔h + + hco3-⇔2h + + co3 2 (1) ca 2+ + co3 2 ⇔ caco3 {phwater＞ 8} (2) ca 2+ + hco3 ⇔ caco3 + h + {7.5 ＜phwater＜8} (3) therefore, this study effectively disperses the cracks in the cracks in the ce mentitious composite materia ls by using synthetic fiber, and, for the cracks with a width of more than 0.1mm, e xecuted the review of the optima l * corresponding aut hor. email: hs-choi@mail.kitami-it.ac.jp proceedings of engineering and technology innovation , vol. 3, 2016, pp. 31 33 32 copyright © taeti self-healing conditions along with the revie w of a diverse range of self-healing performance. 2. experimental overview the mixture proportions of the mo rtar a re summarized in table 1. portland ce ment (c, density: 3.16 g/c m 3 , mean d ia meter: 10 μ m), quartz sand as the fine aggregate (s, surface-dry density: 2.61 g/cm 3 , mean dia meter: 180 μ m), and a high-performance water reducing agent as an admixture (sp, density: 1.05 g/cm 3 , ma in constituent: polycarboxylate-based superplasticizer) we re used. as for the synthetic fibers, pva (fiber dia meter: 40 μ m, fiber length: 12 mm, density: 1.3 g/c m 3 ) and polyethylene (pe) (fiber dia mete r: 12 μ m, fiber length: 12 mm, density: 0.97 g/c m 3 ) and polypropylene (pp) (fiber dia mete r: 65 μ m, fiber length: 12 mm, density: 0.91 g/cm 3 ) were used. table 1mixture proportions of the mortar type s/c (wt.%) w/c (wt.%) sp/c (wt.%) fiber (vol.%) pva 0.4 0.3 0.4 1.2 pe 0.45 pp 0.3 table 2 experimental factors and conditions experimental factors conditions fiber pva, pe, pp self healing water + micro-bubble ph 6.0 ca(oh)2 + micro-bubble ph 8.5 temperature 20°c crack (target of crack width: 0.3mm) tensile load period of self-healing 7 days as the experimental procedure, in st ep a, an analysis of the permeability coeffic ient i mmed iately following the introduction of the cracks by the tensile loading test (prior to self-healing), and an analysis of the types of the hydrates prior to self-healing by using the raman spectroscopy were executed. in st ep b, a comparison and evaluation were made by using the method applied in st ep a in order to eva luate the changes in the permeability of each of the specimens due to self-healing, changes in the structure within the cracks and a quantitative evaluation the self-healing precip itated substances. the experimental factors and conditions are summarized in table 2. 3. results and discussion 3.1. permeability coefficient as the results of the experiment, w+m b in st ep b, when compared with that in st ep a, displayed the trend of an increase in the re sistance to permeability by about 40 fo lds for pva, 3.5 fo lds for pe and 1.5 folds for pp(fig 1). meanwhile, ca +mb in st ep b, when co mpared with that in st ep a, displayed the trend of an increase in the resistance to permeability by about 460 folds for pva , 60 fo lds for pe and 6.0 folds for pp (fig 2). in addition, in the comparison of the permeability coeffic ient follo wing self-healing, ca+m b, in co mparison to w+mb, displayed the trend of improve ment in the re sistance to permeability by appro ximately 15 folds for pva, 17 fo lds for pe and 4.0 fo lds for pp (fig 3). fro m the afore mentioned results, it can be discerned that the resistance to permea bility is improved in the order of pva>pe>pp, regardless of the conditions of self-healing. in particular, pva with oh-radica l d isplays a more e ffective self-healing performance [2], and, it was confirmed that the conditions of ca+m b are more advantageous than the conditions of w+mb for the pro motion of self-hea ling performance. therefore , for the micro crac ks with a width of mo re than 0.1mm for wh ich substantial permeat ion of deteriorating ele ments fro m the e xternal into the internal aspects of the concrete is anticipated, it is deemed that the generation and precipitation of the self -healing substances were pro moted due to the mixing of the pva fiber with the oh radica l a long with the introduction of the conditions of self -healing of the saturated ca(oh)2 solution (ca 2+ ) that contains co2 micro bubbles (co3 2). 3.2. review of the crack section due to self-healing fig 4 shows the e xperimental results of ra man spectroscopy. here, the locations of the occurrence of the peak of a wave generated by a laser at the crack section to which the white colored precipitated substance of pva specimen is attached and that at the sectio ns without cracks were co mpared. as the results of the e xperiment, with the peak of the wave of ca co3powder as the subject of comparison, there was almost no occurrence of the peak in the proceedings of engineering and technology innovation , vol. 3, 2016, pp. 31 33 33 copyright © taeti wave of the sections without cracks that coincides with that of the caco3. ho wever, the peak of the wave in the c rack section accurately coincides with the peak of the wave of ca co3 powder. accordingly, it is concluded that the ma jority of the white colored precipitated substance was generated following self -healing due to caco3. fig. 1 permeab ility coeffic ient of water + micro bubble (w + mb). fig. 2 permeability coeffic ient of ca (oh)2 + micro-bubble (ca + mb). fig. 3 co mparison of permeability coefficient ratio fig. 4 precipitated substances of self-healing 4. conclusions in this paper, it was confirmed that the effective restoration of water t ightness and the ma jority of the self-hea ling products was due to ca co3 by using synthetic fiber with polarity, along with the effect of inducing a multip le number of hairline cracks. in addition, it was confirmed that the self-healing conditions of the saturated ca(oh)2 solution, wh ich supplied the co2 micro-bubble, displayed the most effective self-healing performance in the surface and internal sections of the cracks. acknowledgement this work was supported by the young researcher program of lixil 2015. references [1] c. edvardsen, “water permeability and autogenous healing of c racks in concrete,” aci materials journal, vol. 96, no. 4, pp. 448-454, 1999. [2] d. ho mma , h. m ihashi, and t. nishiwa ki, “self-healing capability of fiber re inforced cementit ious composites ,” journal of advanced concrete technology, vol. 7, no. 2, pp. 217-228, 2009. 0.0000001 0.000001 0.00001 0.0001 0.001 0.01 0.1 1 pva pe pp c o e ff ic ie n t o f w a te r p e rm e a b il it y (m /s e c ) step. a step. b 0.0000001 0.000001 0.00001 0.0001 0.001 0.01 0.1 1 pva pe pp c o e ff ic ie n t o f w a te r p e rm e a b il it y (m /s e c ) step. a step. b 0.0001 0.001 0.01 0.1 1 pva pe pp r a ti o o f c o e ff ic ie n t o f w a te r p e rm e a b il it y w+mb ca+mb 1 = standard 0 5 10 15 20 25 0 500 1000 1500 2000 in te n s it y ( c o u n ts ) raman shift/cm-1 0 5 10 15 20 25 0 500 1000 1500 2000 in te n si ty ( c o u n ts ) raman shift/cm-1 0 5 10 15 20 25 0 500 1000 1500 2000 in te n si ty ( c o u n ts ) raman shift/cm-1 non crack caco3 crack caco3 p owder  proceedings of engineering and technology innovation, vol. 7, 2017, pp. 20 24 preliminary study on a system for visualization of big data in smes yasuo uchida 1,* , miao xinyun 1 , seigo matsuno 1 , yasushi iha 2 , makoto sakamoto 3 1 department of business administration, national institute of technology, ube college, ube, japan 2 department of media information engineering, national institute of technology, okinawa college, nago, japan 3 department of computer science and systems engineering, university of miyazaki, miyazaki, japan received 19 july 2017; received in revised form 30 july 2017; accept ed 03 august 2017 abstract the 2012 white paper on information and co mmun ications in japan issued by the ministry of internal affa irs and co mmunications of japan advocates use of big data under its “special the me: ict -induced and accelerated disaster recovery and japan’s re-birth.” however, the analysis in the japan users association of informat ion systems’ white paper on its 2014 it trend survey for co mpanies reports that less than 10% of co mpanies utilize big data, and it would appear that progress in its use is centered on large firms. under such conditions, use of big data is becoming a challenge for the purpose of ensuring the survival and success of sm es as we ll. as a result, r&d and technological support for smes are becoming pressing issues. however, at present there has been almost no academic research concerning policies and future direct ions for use of big data at smes. accordingly, th is study conducted the modelizat ion of the procedure fo r v isualization of big data fo r sm es. specifically, we o rganized the procedure as a tutorial, fro m obtaining data of japanese hot -spring areas using web scraping, to visualizing them using the visualization software cytoscape keywor ds: big data, visualization, smes, cytoscape 1. introduction this study is intended to research and develop a system for visualizat ion of big data suited to smes, as a tactica l informat ion tool to support sm es’ strategies for success under conditions of increasing ly intense global co mpetition. that is, it aims to probe a fra me work that is easy to adopt and superior in terms of operability for the collection, storage, analysis, a nd use of big data. at the same time, it also a ims to eluc idate e mpirica lly the ideal fo rm of a strategic information infrastructure for smes and challenges in its operation and administration. in this study, we carried out a preparatory consideration of visualization of big data by smes. specifically, we organized the procedure as a tutorial, fro m obtaining data of japanese hot-spring areas using web scraping, to visualizing the m using the visualization software cytoscape. 2. trends in use of big data at smes in japan at present, there are very few e xa mp les of successful use of big data by japanese smes. in addition, how big data is used at smes depends on individual planning by each company. accordingly, this paper will begin by summarizing measures taken and research trends related to the use of big data at japanese smes. it also will e xa m ine a nu mber of e xa mp les of early adopters. * corresponding author. e-mail address: uchida@ube-k.ac.jp tel.: +81-836-35-7567; fax: +81-836-35-7567 proceedings of engineering and technology innovation, vol. 7, 2017, pp. 20 24 copyright © taeti 21 for e xa mple , the report “ enriched living and economy fro m connected it : the va lue and re liab ility of big data”[1] fro m the research group on it infrastructure for living and the economy of the informat ion -technology promotion agency, japan (ipa) (it infrastructure for living and the economy of the informat ion -technology promotion agency, japan) both e xpla ins in simp le terms what big data means for managers of co mpanies aiming to provide new services using big dat a and identifies results such as expansion of business opportunities through summa rizing e xa mp les of early adopters of big data, advantages and issues in service realizatio n, and efforts to resolve these. in addition, the 2014 white paper on small and mediu m enterprises in japan [2] fro m the sma ll and mediu m enterprise agency mentions use of data on corporate transactions (big data) as a “key” to revitalization of regional economies. looking at the activit ies of sm es in the fie ld, in nove mber 2014 the osaka cha mber of co mmerce and industry published the results of a survey intended to ascertain matters such as expectations, needs, and issues involved in use of bi g data by second-tier co mpanies and smes [3]. while the results of this survey show that approximate ly 81% of co mpanies are interested in “informat ion (data)” as “useful for manage ment purposes,” respondents also identify the following as the top three “issues in use” of data:  “difficulty of understanding the cost-effectiveness of use of information (data)” (64.9%)  “lack of human resources to analyze information (data)” (56.9%)  “lack of understanding of methods of using information (data)” (34.0%) accordingly, we decided to proceed with research and development focusing on these three points. first of all, we identified as a necessary condition the ability to use personal computers having specificat ions like those used in ord inary administrative-leve l operations instead of high -priced co mputers, to keep costs down as much as possible. we also decided t o use, in principle, software such as open -source software that can be used free of charge as tools needed for analysis and visualizat ion. another prerequisite we identified was that the data analysis must be of a degree capable of being conducted b y emp loyees who have the skill levels needed to analyze data using spreadsheet software (such as microsoft exce l), since it is difficult for sm es to secure staff that have sp ecialized data analysis skills. furthermo re, we decided to provide hints on use of data by describing specific examples of methods of their use. 3. visualization of big data 3.1. steps from data collection through visualization the data subject to visualizat ion can be broken down into two main categories. the first consists of data in the possession of the co mpany itself. in this case, the co mpany has ascertained the content of the data sufficiently and it is easy for it t o process the data on its own. the other category consists of data that is present on the internet. in this case, it is difficult to understand the structure of the data and they are not easy to obtain. however, somet imes sm es will want to obtain and utilize these data . accordingly, this study will consider the steps used when obtaining and processing data present on the in ternet. since the main objective of this study is to illustrate a data processing model, we limited the purposes of visualization itself to the foll owing content:  subject data to be collected: data on hot-springs resorts in japan, published on the internet  purpose of visualization: to visualize the locations and water qualities of hot -springs resorts  steps in visualization: obtaining data through web scraping [4], conducting a number o f preprocessing steps, and then using cytoscape [5] to import the data as network information and visualize it in the form of graphs. proceedings of engineering and technology innovation, vol. 7, 2017, pp. 20 24 copyright © taeti 22 3.2. data acquisition and processin g when obtaining data through web scraping, the permission of the data provider must be obtained in advance. there is a need to consider how to avoid burdenin g the servers and network when actually obtaining the data. besides , the end-user license agree ment must be co mplied with fo r the data obtained. a lthough we used the python language [6] as a software environment for obtain ing and processing data, we arrang ed the model as one consisting of steps that could be used even by non-specialists, with consideration for ease of use. (1) analysis of web pages there was a need to analy ze the data structure of web pages and identify the data obtained. th is can be done using the view source” feature of a web browser (fig. 1). fig. 1 example of displaying a web page’s source (2) we used a python program to obtain the desired data from within web pages through web scraping. in this study, we obtained only data on the names and water quality of hot-springs resorts . (3) we used the python program to look up the latitude and longitude of the hot -springs resorts in google maps [7]. (4) we processed the above data using exce l and other tools and saved it as network data. an e xa mp le of the fo rmat of the data is provided below. in th is case, we used latitude as the y-a xis va lue on the graph a fter inverting positive and negative signs, since display coordinates and axial directions on the monitor are reversed . sample network data format: prefecture name, hot-springs resort name, water quality, x coordinate (longitude), y coordinate (latitude) 3.3. visualization using cytoscape cytoscape is a tool for visualizat ion of networks (through a graph structure). for this reason, the subject of processing needs to have a network structure. accordingly, we decided to analy ze the locations of hot -springs resorts and their local prefecture capitols, as an exa mp le of a network. fig. 2 shows an exa mple of v isualization of information on hot -springs resorts in ya maguchi prefecture resulting fro m loading network data to cytoscape and color-coding the informat ion by water quality. proceedings of engineering and technology innovation, vol. 7, 2017, pp. 20 24 copyright © taeti 23 in the center of the graph is the ya maguchi prefecture capitol. fro m this graph, the reader can identify the mutual positioni ng from the latitudes and longitudes on the map and the water quality from the color coding of the hot -springs resort names. fig. 2 visualization of hot-springs resorts in yamaguchi prefecture 4. discussion in this study, ultimately we visualized structured data. first of all, the origina l data source of web page source text (htm l) is semistructured data [8]. we followed the method of web scraping to obtain data fro m the content of web pages. ne xt, when the data obtained as in this study are composed of multip le files, there is a need for steps such as data abstraction and combination. this p rocess requires use of data-processing tools and programming languages. in addition, occasionally it is impossible to apply general-purpose tools to conversion of semistructured to structured data, and in such cases one must rely on programming languages. also, in the case of a locale such as japan that employs mult ibyte characters, sometimes text code conversion is required [9]. 5. conclusion this study employed a preparatory consideration of a system for visualizat ion of big data at smes, eluc idating a number of require ments. that is, it showed that in processes such as data collection and data processing there are many cases in which it is difficult to process the data using general-purpose tools alone. topics for future study include development of independent tools to supplement general-purpose tools as well as development of general-purpose models for the steps involved in visualization and preparation of tutorials suitable for use by smes. acknowledgement this work was supported by jsps kakenhi grant number 15k03639 . references [1] it infrastructure for living and the economy of the information -technology promotion agency, japan, “enriched living and economy from connected it: the value and reliability of big data,” http://www.ipa.go.jp/files/000001884.pdf. [2] small and medium enterprise agency, “2014 white paper on small and medium enterprises in japan,” http://www.chusho.meti.go.jp/pamflet/hakusyo/h26/pdf/h26_pdf_moku ji.ht ml. [3] the osaka chamber of commerce and industry, “results of survey on use of big data,” press release, 2014. [4] l. richardson, “beautiful soup,” available via https://www.crummy.com/software/beautifulso up/. cited 26 january 2016 https://www.crummy.com/software/beautifulsoup/ proceedings of engineering and technology innovation, vol. 7, 2017, pp. 20 24 copyright © taeti 24 [5] cytoscape consortium, “cytoscape,” http://www.cytoscape.org/. [6] python software foundation, “python,” https://www.python.org/. [7] python software foundation, “pygeocoder,” https://pypi.python.org/pypi/pygeocoder. [8] d. quass, a. rajaraman, y. sagiv, j. ullman , and j. widom, “querying semi structured heterogeneous information,” journal of systems integration, vol. 7, no. 3, pp. 381-407, 1997. [9] “the unicode consortium,” http://unicode.org/.  proceedings of engineering and technology innovation, vol. 7, 2017, pp. 31 36 using text mining to extract issues for school: an empirical study of the social platform-dcard hsin-yi wang, shu-fen chiou * , jung-wen lo department of information management, national taichung university of science & technology, taichung, taiwan, roc. received 18 july 2017; received in revised form 15 august 2017; accept ed 30 august 2017 abstract nowadays, social network within sentiment analysis has become the main trend in text mining domain. there are many platforms have been analyzed, such as facebook, twitter, instagram, and so on. in our manuscript, we attempt to extract the information about the sentiment polarity of messages (positive, neutral or negative) in a social platform “dcard”. the users of dcard are taiwanese college students, and anonymou s post is being used this in social platform, therefore, the user can express their opinion more freedom. we use dcard to the sentiment polarity of messages in extract the information about the school; moreover, the school could get the feedback from this finding to improve their policy. in this paper, we used python to scrap the web page, and the sentiment lexicon would be built. keywords: text mining, big data, social platform, sentiment 1. introduction nowadays, internet is used for communication widely. people used the internet to be browsing the web and collecting data (79.9%), using community websites (24.1%), playing online games (19.4%), listening to music or watching movies (17.8%), (15%) shopping online (10.3%) [1]. for community websites, people prefer communicating via the internet services over talking face-to-face or writing letters. they are more often writing blogs or posting messages on social networks and the personality will be presented by habitual vocabularies they used. in this research, we t ry to s analyzed chinese vocabularies on a social platform named dcard [2]. with the accumulation of large amounts of educational data, the use of advanced statistical techniques, such as data mining, exploring the potentially useful information or realizing some knowledge from bunch of data [3]. another important material for structured data is unstructured data composed of free text ; for example, scientific research papers, patented technical documents, qualitative interview data, and open questionnaires the content of the analysis, etc. with the analysis or research value. analysis of these unstructured data exploration methods, relying on the free text of the advanced processing and statistical operation s, that is, text mining technology[4]. dcard anonymous posting mechanism has successfully attracted tens of thousands of domestic and foreign college students to become platform members, this research is using dcard community behavior of the chinese corpus , the part of the analysis from the dcard text behavior can be analyzed everyone comment on the community on the web. * corresponding author. e-m ail address: fen057@ nutc.edu.tw tel.: +886-4-22196391 proceedings of engineering and technology innovation, vol. 7, 2017, pp. 31 36 copyright © taeti 32 in this paper, we use the word surveying technology to analyze the administrative quality of the campus in order to avoid the loss of the students, but also to attract the resources and improve the quality of education. the results data can be used as reference for the follow-up research reference and campus administrators. 2. related works in this section, we first introduce the technique which we use to extract chinese vocabularies. and we describe the social platform named dcard 2.1. python python is a widely used high-level programming language for general-purpose programming, created by guido van rossum and first released in 1991. an interpreted language, python has a design philosoph y which emphasizes code readability (notably using whitespace indentation to delimit code blocks rather than curly braces or keywords), and a syntax which allows programmers to express concepts in fewer lines of code than possible in languages such as c++ or java. the language provides constructs intended to enable writing clear programs on both a small and large scale [5]. 2.2. dcard dcard is the first exclusive college student dating site platform in taiwan. the official website is show in fig. 1 [6]. the dcard platform attracts many college students to join to expand their circle of friends, but to join must provide school e -mail, with the real name and photos through the administrator can be certified . dcard using anonymous system to register, many s tudents published articles to take anonymous system to express their own emotions, and these acts through the text presented in the article. fig. 1 dcard website 2.3. text mining in a large number of data, there are digital structure of the structured data and text, sound, image of the unstructured data. there are already hundreds of ways to deal with structured data. [7] text mining from biological literature is emerging as one of the main issues in bioinformatics research, and nlp methods are regarded as being useful to raise the potential of text mining from this literature. while the techniques are separated relatively into domain-portable, reference materials, except for corpora. [8]. proceedings of engineering and technology innovation, vol. 7, 2017, pp. 31 36 copyright © taeti 33 text mining is a new field, the text of the survey can be collec ted from the text of meaningful messages, for the file can also be analysed out of his specific purpose of the target, the text is undiscovered, invisible, and difficult in the algorithm. h owever, in modern culture, the document is the most common tool for the formal exchange of messages. the field of textual exploration usually involves a document whose function is to convey a factual message or opinion. [9] the steps of knowledge exploration : [10]  data collection  data cleansing  data conversion  application of exploration technology  the results are presented and interpreted the methods of knowledge exploration:  association analysis  classification  clustering  summarization  prediction  sequence analysis 3. our proposed method the work presented in this article covers, on the one hand, the extraction of information about user’s positive/neutral/negative sentiments from the text title they write. we investigate the detection of sentiment changes with respect to the “usual” sentiment of each us er [11]. in fig. 2, we present the methods ’ flowchart with each purpose below: fig. 2 method table 1 marking positive 其實我們的學校真的挺不錯的(actually our school is really good) negative 畢業證書之學校爛行政(graduation certificate of the school rotten administration) neutral 問學校語言學習軟體(ask for school language learning software) proceedings of engineering and technology innovation, vol. 7, 2017, pp. 31 36 copyright © taeti 34 table 2 dcard to word web text table 3 preprocessing 2 before the process after the process 【沒朋友 徵室友，我又來了】 (【no friends levy roommate, i have come】) 沒朋友徵室友我又來了 (no friends levy roommate, i have come) table 4 keywords main classification keyword course 課程 課名 老師 administration 足球隊(football team) 行政(administrative) 校園(camp us) step 1: we need data collection and marking. this research using the article title to classification to positive , neutral and negative includes school course and administration tow part. we give an example in table 1. step 2: we use python to extract dcard website to the text content. we extract dcard website by using python api url. we extract dcard website convert to text content, combine and export to document. table 2 are our examples. step 3: sorting out the data we export, and remove the extra number and english letter (the web content which is not created by the dcard user), and convert the full shape character to half shape character. step 4: remove the punctuation in the data (e.g. #, semicolon, comma and space) like table 3. step 5: create keywords. classification the pre-process data we created for the school course and administration. for an example in table 4, we set the course and its name for the school course partition, and the “soccer team” for administration due to the students usually use 4. data analysis and results 4.1. data we collect the dcard information from 2014 to 2017 in table 5. the results of the original data have number of 24939. the data have been 4941 after pre-process finishing. we the separate these into "administration", "courses" and "other" three classifications : table 5 data main classification data administration 1098 courses 729 other 3114 table 6 keywords dictionary main classification keywords administration 行政 (administrative) 足球隊 (football team) 停車場 (parking lot) 學校 (school) 校園 (camp us) 系學會 (dep artment of science) 智慧大師 (m aster of wisdom) courses 課 (class) 課程名稱 (class name) 老師名稱 (teacher) rs (rs) 多益 (toeic) 科系 (dep artment) proceedings of engineering and technology innovation, vol. 7, 2017, pp. 31 36 copyright © taeti 35 table 7 polarity administration courses positive 251 211 negative 321 215 neutral 526 303 4.2. keyword dictionary in the classification keywords shown in table 6, we found that students had certain classification of keywords . for instance, students often use the "足球隊" to describe the attitude of the campus administration. the following keywords which are usually being used: 4.3. data polarity we will separate relatively into positive, negative and neutral, the following is a review of a variet y of data classification number: 5. discussion in fig. 3, studies have investigated the title dcard reviews, we found that 22% of students expressed the article ideas for the administration, 15% of people have questions or ideas for courses in dcard, and the others are expressed other deman ds. fig. 3 data fig. 4 administrative in fig. 4, we also found that students in the comments can really achieve the administration and course, there are 1098 administrative from administration. 251 is the positive title, 321 negative title and 526 neutral title. 52 percent of the students published their perspective, point of view and opinions on the dcard, 29% of whom expressed negative title, which showed that students were less likely to be administratively in school (including school systems). fig. 5 course there are 729 titles in the course, with 211 positive titles, 215 negative titles and 303 neutral titles, which shows that st udents have 58% of the ideas and suggestions in the course or teacher, and the 42% of the people have other problem in course. we can see the results in fig. 5. proceedings of engineering and technology innovation, vol. 7, 2017, pp. 31 36 copyright © taeti 36 6. conclusions this research explores students' perceptions of campus quality, extracts web pages, use keywords and the terminology of classified sentences to achieve the desired results. use the positive and negative sentences to explore the ideas, perspectives, point of views of students on the administration and the curriculum the negative title, it can be extracted separately review and achieve to improvement the purpose campus administration and curriculum. furthermore, future research can be published on dcard and use the text to explore the students for the campus administration ideas and recommendations make research more in-depth can improve the improvement. there are two limitations to this paper, which can be discussed in subsequent studies : 1. research methods can only see the school administration is missing, but cannot see what kind of administrative issues are 2. title cannot tell what kind of problem, is it can only know the advantages and disadvantages references [1] s. w. wu, “applying text mining technique to analyze the software quality characteristics of mobile games, ” in department of industrial management, national pingtung university of science and technology , pp. 1-49, 2013. [2] x. z. chang and y. m. li, “using text mining to predict personality based on social behavior,” 2012. [3] r. s. baker and k. yacef, “the state of educational data mining in 2009: a review and future visions ,” journal of educational data mining, vol. 1, no. 1, pp. 3-17, 2009. [4] y. h. tseng and y. i. lin, “the application of content mining techniques to the analysis of educational evaluation research trends ,” journal of research in education sciences, vol. 56, no. 1, pp. 129-166, 2011. [5] g. van rossum, python, https://zh.wikipedia.org/wiki/python#.e5.8f.82.e8.80.83.e6.96.87.e7.8c.ae, 1991. [6] c. y. jian, “dcard,” https://www.dcard.tw/, 2011. [7] c. m. young, “applies by text mining to the support systems for coding icd-9-cm —a study of admission note and discharge summary,” taipei medical university information management, pp. 1-68, 2004. [8] j. d. kim, t. ohta, y. tateisi, and j. tsujii, “genia corpus —a semantically annotated corpus for bio -textmining,” bioinformatics, vol. 19, pp. i180-i182, 2003. [9] m. hall, e. frank, g. holmes, b. pfahringer, p. reutemann, and i. h. witten, “the weka data mining software: an update,” acm sigkdd explorations newsletter, vol. 11, no. 1, pp. 10-18, 2009. [10] y. h. tseng, “research and development on automatic information organization and subject analysis in recent decades ,” journal of educational media and library sciences , vol. 51, pp. 3-26, 2014. [11] a. ortigosa, j. m. martin, and r. m. carro, “sentiment analysis in facebook and its application to e-learning,” computers in human behavior, vol. 31, pp. 527-541, 2014. microsoft word 6-#934-typesetting finish pdf列印.doc proceedings of engineering and technology innovation, vol. 5, 2017, pp. 31 36 causal relationships among factors affecting the buildability in executing construction projects in vietnam hai pham1,*, soo-yong kim2, truong-van luu3, phu-son lam4 1interdisciplinary program of construction engineering and management, pukyong national university, busan, korea. 2department of civil engineering, pukyong national university, busan, korea. 3department of civil engineering, international university, ho chi minh city, vietnam. 4department of civil engineering, ho chi minh city university of technology, vietnam. received 07 july 2017; received in revised form 10 august 2017; accepted 13 august 2017 abstract this paper aims to identify causal relationships among factors affecting the buildability in executing construction projects in vietnam. first, through a comprehensive literature review, the study identified a full set of attributes that affect the buildability of construction projects. subsequently, through discussions and interviews with experts, the study determined main attributes affecting the buildability in the context of the construction industry in vietnam. after that, a survey questionnaire was developed which was based on identified attributes. the data after collected, was analysed by structural equation model (sem) and the expected result from the sem model is a structural model. this structural model includes four factors, namely: design applies advanced machinery, materials and construction methods; design uses available resources at local for saving costs; design applies safe construction methods; and design uses prefabricated and typical components. the key finding of the study is to provide insights into causal relationships among factors affecting the buildability in executing construction projects in vietnam. keywords: buildability, structural equation model, construction projects, vietnam 1. introduction the construction industry is one of the most developed industries in vietnam in recent years. there are many projects to be completed corresponding to the expectation of investors as well as meeting aesthetic and quality requirements. however, there are also many projects completed with differences to initial designs. particularly, many projects after completed design cannot be constructed or have delays in construction schedule, cost overruns due to revisions of drawings to suit practical requirements. currently, there are many methods of project implementation in vietnam but design-bid-build method is still the most common method. for this traditional method, design and construction companies usually work separately. thus a completed design which cannot match the construction is unavoidable. therefore, to finish projects with the high efficiency in terms of aesthetics, ease for construction as designed, limitation of many revisions, reduction of incurred costs and schedule; design companies need to consider more carefully about the * corresponding author. e-mail address: haiphamcpm@gmail.com tel.: +825-1-6297718 proceedings of engineering and technology innovation, vol. 5, 2017, pp. 31 36 copyright © taeti 32 buildability for each design. to obtain this purpose, this study needs to finish the following two objectives: identify attributes that affect the buildability for the process of project implementation in the context of the construction industry in vietnam; and discover causal relationships among constructs affecting the buildability for projects deployment. 2. research methodology to achieve the study objectives above, firstly a comprehensive literature review is conducted to determine attributes that affect the buildability for projects deployment in the context of the construction industry in vietnam. then, a survey questionnaire is developed to collect data for the analysis of the study. the questionnaire consists of two parts: the first part presents the attributes affecting the buildability, the respondents are asked to evaluate the items based on a five-point likert scale according to convention: 1 (strongly disagree ) to 5 (strongly agree). the second section is designed to collect personal information of the respondents such as years of work experience as well as positions in projects. the next step, a structural equation modelling (sem) is deployed to discover causal relationships that affect the buildability for implementing projects. the sem model in this study includes two models: the measurement model and structural model. the purpose of the measurement model is to evaluate the extent to which observed variables load up their underlying constructs. the purpose of the structural model is to discover causal relationships among constructs affecting the buildability. the fit of both the measurement model and structural model is assessed by several goodness of fit (gof) indices. if the two models achieve the minimum requirements of these indices, they are considered to be fit into the data of the study and accepted. finally, discussion of the result of the sem model is also shown. the next section summarizes the factors influencing the buildability to be identified through a literature review and the implementation of the sem model. 3. structural equation model (sem) 3.1. confirmatory factor analysis (cfa) in testing measurement scale, the cfa method in structural equation model has many advantages rather than traditional methods. the reason is that cfa allows the testing of theory structure of measurement scale such as relationship between a research concept and other concepts without bias due to measurement error. moreover, cfa can also check convergence value as well as discriminant value of measurement scale without taking many research steps as traditional methods [10-13]. the model fit is assessed by many goodness of fit (gof) indices as follows:  chi-square (cmin).  chi-square/df (cmin/df).  comparative fit index (cfi).  tucker & lewis index (tli).  root mean square error approximation (rmsea). the model is considered as suitable to data when p-value (chi-square test) higher than 0.05. however, chi-square is dependent on sample size. if a model gets the values of gfi, tli, and cfi higher than 0.9 [10-13]; cmin/df lower than 2; rmsea lower than 0.08, then the model is considered as suitable to data or compatibility to data. proceedings of engineering and technology innovation, vol. 5, 2017, pp. 31 36 copyright © taeti 33 seven main factors covering attributes for the sem model are denoted as in table 1. these attributes are found out based on previous studies of the authors, such as: [1-9] table 1 factors affecting the buildability attributes factors affecting the buildability denoted design applies advanced materials design applies advanced machinery, materials and construction methods b1 design applies advanced machinery and construction methods design considers many options for projects foundation with complex geological conditions design considers supply conditions to imported construction equipment and materials design considers relocation of temporary prop structures to create more space for construction design uses available resources at local for saving costs design uses available resources at local for saving costs b2 design considers minimization of materials waste design considers convenient traffic conditions for workers in sites design ensures foundation stability for projects being executed and surrounding buildings design applies safe construction methods b3 design considers safe construction methods when working underground ground and shape of floors are not too complex design considers sufficient workspace for workers and machinery during the construction design minimizes complex components minimization of complex components and maximization of standardized components for fast construction b4 design maximizes standardized components for fast construction design minimizes complex components design considers safe construction conditions b5 design maximizes standardized components for fast construction design considers the operation possibility of machinery and equipment on sites work coordination and construction process consideration b6 design considers safe conditions for projects when heavy equipment dropped design of construction process considers creating easy conditions for components installation design uses prefabricated and typical components b7 designers, contractors and project management boards cooperate closely and regularly the fig. 1 presents attributes as well as factors affecting the buildability for projects deployment. fig. 1 the original cfa model proceedings of engineering and technology innovation, vol. 5, 2017, pp. 31 36 copyright © taeti 34 fig. 2 the final cfa model (standardized) the final cfa model (fig. 2) is revised several times by adding errors to observed variables according to modification indices (mi). after that, variables which have standardized weights lower than 0.5 are eliminated. data of the study is evaluated based on goodness of fit (gof) indices: chi-square/df = 1.53 < 2; cfi = 0.95 > 0.9; gfi = 0.93 > 0.9; tli = 0.91 > 0.9; rmsea = 0.064 < 0.08. therefore, the final cfa model fits into the data of the study and accepted. the standardized weights are higher than 0.5 and the unstandardized weights are also statistically significant (p-value = 0.00), so the factors achieve the convergent validity. the correlation coefficients among the factors in the model are lower than 0.9 thereby the scale achieves discriminant validity. 3.2. the structural model from the results obtained in the previous analysis of cfa, combining with the consideration of relationships of the main factors, a structural model (fig. 3) is proposed to describe the mutual influence of the main factors as follows: fig. 3 the structural model among the main factors the result of the structural model among the main factors: chi-square/df = 1.64 < 2; cfi = 0.94 > 0.9; gfi = 0.93 > 0.9; tli = 0.91 > 0.9; rmsea = 0.07 < 0.08. therefore, the structural model among the main factors fits into the data of the study and accepted. from the structural model above, it can be concluded that design applies advanced machinery, materials and construction methods has the positive influence on design uses prefabricated and typical components. design consultants generally use proceedings of engineering and technology innovation, vol. 5, 2017, pp. 31 36 copyright © taeti 35 traditional structural systems (such as reinforced concrete systems poured at sites) without much attention to advanced machinery, materials and construction methods. prefabricated components often have advantages in terms of quality as produced in factories and prestressed components will meet requirements on the aesthetic. the increase of using advanced machinery and construction methods also helps to shorten time for construction of prefabricated components. design enhancing the application of new materials will also help reduce costs for projects such as using light brick to make partition will help to reduce loads thereby reducing the volume of foundation and frame systems. the increase of using lightweight and good quality materials as well as advanced machinery and construction techniques will improve safe construction methods as designing. design using locally available materials to reduce transportation and supply costs will negatively impact on design which considers use of prefabricated components. projects which currently use reinforced concrete structures poured in sites under traditional methods are very popular, thereby local workers are mostly familiar with construction process. in addition, contractors are also easy to find the supply of locally available materials. this contributes to limitation of using prefabricated components and advanced construction methods. 4. bootstrap test to assess the reliability of estimates, in quantitative research methods using sampling method, one sample is usually divided into two subsamples. the first subsample is used to estimate model parameters, and the second is for reassessment. another method is to repeat the study by other samples. two methods above are generally impractical because structured methods often require large samples, so this takes more time and cost [14]. in such cases, bootstrap is a suitable method to replace [15]. bootstrap is a repeated sampling method with replacement. the collected sample consists of 132 observations. the sample for bootstrap test consists of 500 observations. the test result is given in the following table: table 2 error of standardized regression weights parameter se se-se mean bias se-bias cr b2  b1 0.111 0.004 0.663 -0.009 0.005 -1.80 b3  b1 0.132 0.004 0.475 0.008 0.006 1.33 b7  b1 0.472 0.015 1.074 0.016 0.021 0.76 b7  b2 0.462 0.015 -0.597 -0.032 0.021 -1.52 the result in table 2 shows that there is no statistically significant difference in parameter values between the collected sample and the sample for bootstrap test. all cr values are smaller than z = +/1.96 of the normal distribution (no statistical significance at the confidence level 95%). thus, the estimated parameters of the sem model are reliable. 5. conclusions in this study, a theoretical model is developed and tested rigorously using data collected through survey questionnaires covering the factors affecting the buildability for projects deployment in the context of the construction industry in vietnam. the sem method is used to empirically validate this theoretical model to discover causal relationships between constructs. the result of the sem model indicates that design applies advanced machinery, materials and construction methods has the positive influence on design uses prefabricated and typical components. design enhancing the application of new proceedings of engineering and technology innovation, vol. 5, 2017, pp. 31 36 copyright © taeti 36 materials will also help reduce costs for projects such as using light brick to make partition will help to reduce loads thereby reducing the volume of foundation and frame systems. the increase of using lightweight and good quality materials as well as advanced machinery and construction techniques will improve safe construction methods as designing. design using locally available materials to reduce transportation and supply costs will negatively impact on design which considers use of prefabricated components. however, there are also some limitations to this study: the sem model is validated by the data collected from the respondents working in the construction industry in vietnam. for this reason, a recommendation made is that there should be further studies to be carried out to discover causal relationships between constructs affecting the buildability in other regions. consequently, similarities and differences will be summarized and compared to this study. references [1] p. t. i. lam, f. w. h. wong, and a. p. c. chan, “contributions of designers to improving buildability and constructability,” journal of design studies, vol. 27, no. 4, pp. 457-479, july 2006. [2] r. m. zin, m. z. a. majid, and c. fadhil, “neural network model for design constructability assessment,” jurnal teknologi, vol. 40, pp. 27-40, june 2004. [3] code of practice on buildable design, 2013 ed. singapore, building and construction authority, august 2013. [4] a. m. jarkas, “an investigation into the influence of buildability factors on productivity of in situ reinforced concrete construction,” ph.d. thesis, university of dundee, 2005. [5] e. saghatforoush, b. trigunarsyah, and e. g. too, “assessment of operability and maintainability success factors in provision of extended constructability principles,” 9th international congress on civil engineering, may 2012. [6] a. m. jarkas, “the effects of buildability factors on formwork labor productivity of grade beams,” journal of construction engineering, vol. 25, no. 2, pp. 240-265, august 2010. [7] a. m. jarkas, “buildability factors that influence micro-level formwork labour productivity of beams in building floors,” journal of construction in developing countries, vol. 16, no. 1, pp. 1-18, 2011. [8] a. m. jarkas, “the effects of buildability factors on rebar fixing labour productivity of beamless slabs,”australasian journal of construction economics and building, vol. 10, no. 1-2, pp. 16-35, january 2010. [9] a. griffith, “a critical investigation of factors influencing buildability and constructability,” ph.d. thesis, heriot-watt university, uk, 1984. [10] j. f. h. jr, w. c. black, b. j. babin, and r. e. anderson, multivariate data analysis, 7th ed. person prentice hall, nj, 2010. [11] j. b. ullman, “structural equation modeling: reviewing the basics and moving forward,” journal of personality assessment, vol. 87, no. 1, pp. 35-50, august 2006 [12] r. e. schumacker and r. g. lomax, a beginner’s guide to structural equation modeling, 3rd ed., new york, ny, taylor and francis group, 2010. [13] k. schermelleh-engell and h. moosbrugger, “evaluating the fit of structural equation models: tests of significance anddescriptive goodness-of-fit measures,” methods of psychological researchonline, vol. 8, no. 2, pp. 23-74, may 2003. [14] j. c. anderson and d. w. gerbing, “structural equation modeling in practice: a review and recommended two-step approach,” psychological bulletin, vol. 103, no. 3, pp. 411-423, 1988. [15] r. e. schumacker and r. g. lomax, a beginner’s guide to structural equation modeling, 2nd ed., new jersey: lawrence erlbaum associates, 2004. 4___peti#7790__38-44 proceedings of engineering and technology innovation, vol. 19, 2021, pp. 38-44 low profile wideband dual-ring slot antenna for biomedical applications shilpee patil 1,* ,vinod kapse 1 , shruti sharma 1 , anil kumar pandey 2 1 department of electronics and communication engineering, noida institute of engineering and technology, greater noida, utter pradesh, india 2 department of electronics and communication engineering, gl bajaj institute of technology and management, greater noida, utter pradesh, india received 26 may 2021; received in revised form 22 june 2021; accepted 23 june 2021 doi: https://doi.org/10.46604/peti.2021.7790 abstract in this study, a low-profile, co-planar waveguide (cpw) fed, wideband, and dual-ring slot antenna design for biomedical applications is proposed. the proposed antenna has a total area of 10 mm × 10 mm and a height of 0.4 mm, and is designed by using a thin and biocompatible fr4 epoxy (εr = 4.4) substrate to accomplish human body isolation and great flexibility obtained by implantation. this wideband antenna covers a large bandwidth of industrial scientific and medical (ism) frequency band, including 902.8 mhz to 928 mhz, 1.395 ghz to 1.4 ghz, 1.427 ghz to 1.432 ghz, 2.4 ghz to 2.485 ghz, and above. the simulation results of return loss, voltage standing wave ratio (vswr), impedance matching, gain, and radiation pattern of the proposed antenna are obtained through high frequency structure simulator (hfss) 14 software. keywords: biomedical application, dual-ring slot, ism band, implantable antenna 1. introduction the application of antenna is as much as a vast antenna topic itself. some of the applications include automobiles, flush-mounted applications, satellites, cars, aircrafts for biomedical purposes, etc. for biomedical purposes, the antenna is not only used as a device to take and pass body information, but is also used for monitoring, tracking, and for many other purposes. however, for the implantation of an antenna, the size of the antenna is one of the main issues introduced for antenna design. as for introducing an antenna inside the human body, the overall dimensions of the antenna must be tiny and flexible, showing a large bandwidth and a low specific absorption rate (sar) value. many of the research articles depict various issues in the field of antenna design technology to implement a small antenna for biomedical application. in previous years, several multiband slot antennas have been proposed [1-5]. das and mitra [6] present a small, wideband, flexible, and implantable slot antenna design that provides high gain. to improve the antenna gain, their proposed metamaterial (mtm) array has a very high value of epsilon on the superstrate material of an antenna. the gain of that antenna has been enhanced by about 3db using mtm. bashir et al. [7] present a miniaturized wideband implantable antenna for biomedical application. this reported antenna contains the radiating element with rectangular and circular slots and a ground plane with a rectangular slot on it. the maximum gain achieved by this antenna is -12 dbi at 2.45 ghz. the study shows how the parameters such as bandwidth and gain of antenna are varied by positioning the designed antenna in different human body phantoms, e.g., skin, muscles, stomach, small intestine, large intestine, etc. xu et al. [8] present an antenna with an open-end slot on the ground plane, which is dual band in nature and has resonance at 2.45 ghz in the industrial scientific and medical * corresponding author. e-mail address: shilpeepatil21@gmail.com tel.: +919811143811 proceedings of engineering and technology innovation, vol. 19, 2021, pp. 38-44 (ism) band and at 402 mhz in the medical implant communication system (mics) band. the result shows a measured bandwidth of 52.6% at the lower band. for the upper band, the observed resonant frequency shifts to 2.47 ghz and a measured bandwidth of 4.4% is obtained for s11 < -10 db. nesasudha and fairy [9] present a low-profile antenna design for biomedical applications in the frequency range of 2.65 to 3 ghz. this antenna is evident with three distinct layers of substrate materials, i.e., fr4 substrate, bakelite substrate, and roger substrate (ro3010). among the three substrates, the fr4 possesses a return loss of -32.6 db which is the best result for the radiation of the antenna. guo and liu [10] present a single-fed miniaturized circularly polarized microstrip patch antenna. the influence of different body phantoms is presented to calculate the sensitivity of the designed antenna. the simulated value of impedance bandwidth of the reported antenna covering from 2.36 to 2.55 ghz is 7.74%. xiao and li [11] propose an antenna design for an implantable medical device. a miniature semi-circular implantable planar inverted-f antenna (pifa) is designed and proposed in this work. there are three arc-shaped slots that are embedded in a semi-circular patch due to which the proposed antenna achieves reduction in size. chauhan et al. [12] present radio frequency (rf)-linked implantable medical devices. the antenna presented in [13] is implemented to work in the ism band (2.4-2.4835 ghz). the antenna is inserted within a silicon layer and fed by a whole strip line. the result shows the thin dimension of the antenna, which allows the antenna to flex along the curvature of the body without high performance loss, having a higher bandwidth and a good return loss. yang and xiao [14] present a wideband single-fed implantable antenna for the application of ism biomedical at 2.4 ghz. the simulated result of the s11 bandwidth is below -10 db ranging from 2.24 to 2.59 ghz (35%). for the low mode, the maximum of the realized gain -20.8 dbi at boresight is obtained. yang et al. [15] present a design to develop an implantable circularly polarized microstrip patch antenna embedded in a lossy material by addressing its total quality factor. liu et al. [16] design and verify a single-fed miniaturized wide-beamwidth circularly polarized implantable antenna operating in the ism band, for subcutaneous real-time glucose monitoring applications. xia et al. [17] and gani et al. [18] present two antenna systems that operate on multi bands, including the mics band and the ism band. in this study, two triangular rings with equal spacing are designed with the co-planar waveguide (cpw) feeding technique. at the operating frequency of 2.45 ghz, s11 is less than -10 db that shows its proper nature for implantation. the proposed design is a low-profile antenna, which shows a wideband from 0.78 ghz to 5 ghz including 902.8 mhz to 928 mhz, 1.395 ghz to 1.4 ghz, 1.427 ghz to 1.432 ghz, and 2.4 ghz to 2.485 ghz. it covers the ism band and demonstrates a wide range of applications. 2. design configuration the proposed antenna design consists of two concentric triangular ring slots of the equal width d with cpw fed. the design is fabricated on 10 mm × 10 mm fr4 epoxy substrate with a dielectric constant 4.4 and a height of 0.4 mm. therefore, the overall dimension of the antenna is 10 × 10 × 0.4 mm 3 . although a different type of feeding technique is available, in the proposed antenna, the cpw fed with 50 ω matching is used by using appropriate ground plane spacing (s) and trace width (w). fig. 1(a) shows the top view of the designed antenna, where d is the spacing between the triangular rings, s is ground plane spacing, and w is tracing width. their values are presented in table 1. for the design and simulation of the proposed antenna, high frequency structure simulator (hfss) software [19] is used in this study. hfss is a three-dimensional electromagnetic software for designing and simulating high frequency antennas, and is commonly used for antenna design and for the design of complex rf electronic circuits. table 1 design parameter of implantable slot antenna parameter s l d h w a b value (mm) 0.5 10 0.75 0.4 1.05 60 40 39 proceedings of engineering and technology innovation, vol. 19, 2021, pp. 38-44 (a) top view of antenna (b) antenna design in hfss software (c) antenna design in hfss software (continued) fig. 1 antenna configuration 3. antenna performance in biomedical applications, the designed implantable antenna should be situated in a body-like model. for this purpose, skin, fat, muscle, etc., have been used. the proposed antenna is simulated inside the muscle box having a relative permittivity of 52.79 and bulk conductivity of 1.705 as shown in fig. 2. to achieve the radiation characteristics of the proposed antenna, the total height of the muscle box is taken 10 mm whereas the radiation box is taken 20 mm. the value of a and b is given in table 1. fig. 2 antenna placement in muscle box and radiation box 3.1. return loss and input impedance the proposed dual concentric triangular slot antenna is simulated inside the muscle model. the simulated result for return loss characteristics is illustrated in fig. 3. fig. 3 s11 parameter of antenna 40 proceedings of engineering and technology innovation, vol. 19, 2021, pp. 38-44 the s11 characteristics of the proposed antenna represent the amount of power reflected from the antenna, so the s11 parameter is known as the reflection coefficient or return loss of the antenna. therefore, according to power radiation, if s11 is 0 db, it means 100% reflection. in other words, if no power is into the antenna, all power gets reflected. on the other hand, if s11 is less than -10 db, it means 90% of available power is already delivered to the antenna and 10% gets reflected back which is the best result for an antenna. fig. 3 shows the values from 0.78 ghz to 5 ghz under -10 db, representing that at least 90% input power is delivered to the device and the reflected power is less than 10%, which is sufficient for many applications. the simulated s11 results show that the proposed antenna covers many ism band frequencies which include 902.8 mhz to 928 mhz, 1.395 ghz to 1.4 ghz, 1.427 ghz to 1.432 ghz, 2.4 ghz to 2.485 ghz, and above. it is useful for many wireless applications. due to the radiating triangular slot patch at the top of the antenna, the ground is assumed like an effective magnetic current. the image current with respect to the ground assumed as a perfect electric conductor has the same magnitude and phase. fig. 4 illustrates the input impedance characteristics for the proposed antenna at the desired band. fig. 4 input impedance characteristics of antenna 3.2. voltage standing wave ratio the voltage standing wave ratio (vswr) describes the amount of mismatch between the proposed antenna and its feed line connected to it. the value of vswr is in the range of 1 to infinity. vswr is 1 when return loss is infinity. fig. 5 illustrates the value of vswr between 2.5 db to 5 db of the proposed antenna. fig. 5 voltage standing wave ratio 3.3. radiation pattern the three-dimensional radiation pattern describes the strength of radiation emitted in a particular direction as well as the gain of an antenna in terms of electric and magnetic fields. the distance of radiations from the origin represents the strength of radiation emitted in that direction. fig. 6 shows that the power transmitted along the z axis is maximum. 41 proceedings of engineering and technology innovation, vol. 19, 2021, pp. 38-44 fig. 6 3d radiation pattern of the proposed antenna e-plane and h-plane are basically the principal planes for the propagation of waves or radiation fields, where the e-plane shows the plane containing the electric field and h-plane shows the plane containing a magnetic field. co-polarization means that the polarization of the transmitting and receiving antenna is the same, and cross polarization means that the polarization of the transmitting and receiving antenna is different. the proposed antenna has been simulated at 0.915 ghz and 2.45 ghz. fig. 7(a) represents the x-z plane graph and fig. 7(b) represents the y-z plane graph at 0.915 ghz. similarly, fig. 7(c) represents the x-z plane graph and fig. 7(d) represents the y-z plane graph at 2.45 ghz. (a) the x-z plane at 0.915 ghz (b) the y-z plane at 0.915 ghz (c) the x-z plane at 2.45ghz (d) the y-z plane at 2.45ghz fig. 7 radiation pattern of the antenna 3.4. the simulated surface current distribution to accomplish an effective operation of the proposed antenna, it is required to cover most of the impedance bandwidth by axial ratio bandwidth of the antenna. to introduce the current distribution perturbation on the surface of the proposed antenna, dual triangular shaped slots are cut into the ground with cpw feed line. fig. 8 illustrates the simulated surface current distribution at 2.45 ghz for time phase 0˚ and 90˚. as a result, an improved current distribution is obtained within the operating band. 42 proceedings of engineering and technology innovation, vol. 19, 2021, pp. 38-44 (a) phase angle 0˚ (b) phase angle 90˚ fig. 8 the simulated surface current distribution at 2.45 ghz 3.5. antenna gain the simulated value of gain for the proposed antenna with respect to its desired frequency is plotted in fig. 9. the frequency is in the range of 0.78 ghz to 5ghz. the maximum simulated gain is observed -17.0411 db at 0.92 ghz and -14.572 db at 4.02 ghz. the noticeable radiation characteristics with acceptable gains for desired bands and a wideband including ism band are achieved with a small size compared to other implantable antennas. this proposed antenna is compared with different designs of the implantable antennas in table 2. fig. 9 gain vs. frequency table 2 performance comparison with different implantable antennas ref. frequency (ghz) bandwidth (mhz) gain (dbi) dielectric material antenna shape volume (mm 3 ) [7] 2.45 483 -12 rogers ultralam circular 9.8 [15] 2.45 73 -20.47 rogers rt/duroid 6010 zig zag 21 [16] 2.45 300 -17 rogers 3210 circular 91.75 [17] 2.45 246 -21.2 rogers rt/duroid 6010 spiral 31.5 [18] 2.45 105 -22.3 rogers rt/duroid 6010 zig zag 52.5 the proposed work 2.45 4220 -14.6 fr4 epoxy triangular 40 4. conclusions the proposed work presents a cpw fed slot antenna and its design procedure, which is suitable for biomedical applications at a very low cost. this antenna is characterized by its small size, wideband, and single feeding property. the proposed antenna has appreciable gain and good radiation pattern characteristics over a large impedance bandwidth which 43 proceedings of engineering and technology innovation, vol. 19, 2021, pp. 38-44 covers a wide range from 0.8 ghz to 5 ghz and above. this antenna is for biomedical purpose as it covers the ism band range, including 902.8 mhz to 928 mhz, 1.395 ghz to 1.4 ghz, 1.427 ghz to 1.432 ghz, and 2.4 ghz to 2.485 ghz. the maximum gain of the antenna is -17.0411db at 0.92 ghz and -14.572db at 4.02 ghz. conflicts of interest the authors declare no conflict of interest. references [1] j. pourahmadazar, c. h. ghobadi, j. nourinia, n. felegari, and h. shirzad, “broadband cpw-fed circularly polarized square slot antenna with inverted l-strip for uwb applications,” ieee antennas wireless propagation letters, vol. 10, pp. 369-372, april 2011. [2] b. k. kanaujia, s. k. gupta, j. kishor, and d. gangwar, printed antennas: theory and design, boca raton: crc press, 2020. [3] s. patil, a. k. singh, b. k. kanaujia, and r. l. yadava, “a compact, dual wide-band circularly polarized, modified square ring slot antenna for c and ku band applications,” international journal of microwave and wireless technologies, vol. 11, no. 2, pp. 182-189, october 2019. [4] s. patil, a. k. pandey, and v. k. pandey, “a compact, wideband, dual polarized cpw-fed asymmetric slot antenna for wireless systems,” journal of microwaves, optoelectronics, and electromagnetic applications, vol. 19, no. 3, pp. 343-355, september 2020. [5] s. patil, a. k. singh, b. k. kanaujia, and r. l. yadava, “a low-profile triple-band circularly polarized wide slot antenna for wireless systems,” international journal of microwave and wireless technologies, vol. 11, no. 1, pp. 67-75, august 2019. [6] s. das and d. mitra, “a compact wideband flexible implantable slot antenna design with enhanced gain,” ieee transactions on antennas and propagation, vol. 66, no. 8, pp. 4309-4314, may 2018. [7] z. bashir, m. zahid, n. abbas, m. yousaf, s. shoaib, m. a. asghar, et al., “a miniaturized wide band implantable antenna for biomedical application,” uk/china emerging technologies, october 2019, pp. 1-4. [8] l. j. xu, y. x. guo, and w. wu, “dual-band implantable antenna with open-end slots on ground,” ieee antennas and wireless propagation letters, vol. 11, pp. 1564-1567, 2012. [9] m. nesasudha and j. j. fairy, “low profile antenna design for biomedical applications,” international conference on signal processing and communication, july 2017, pp. 139-142. [10] y. guo and c. liu, “miniaturized implantable circularly polarized patch antenna,” ieee international workshop on electromagnetics, august 2014, pp. 271-272. [11] s. q. xiao and r. q. li, “antennas design for implantable medical devices,” ieee international conference on computational electromagnetics, february 2015, pp. 61-63. [12] a. chauhan, g. chauhan, and g. kaur, “implantable antennas in biomedical applications,” international conference on computational intelligence and communication networks, december 2015, pp. 1-5. [13] t. a. aleef and a. biswas, “design and measurement of a flexible implantable stripline-fed slot antenna for biomedical applications,” international conference on electrical engineering and information communication technology, september 2016, pp. 1-5. [14] z. j. yang and s. xiao, “a wideband implantable antenna for 2.4 ghz ism band biomedical application,” international workshop on antenna technology, march 2018, pp. 1-3. [15] z. j. yang, l. zhu, and s. xiao, “an implantable circularly polarized patch antenna design for pacemaker monitoring based on quality factor analysis,” ieee transactions on antennas and propagation, vol. 66, no. 10, pp. 5180-5192, august 2018. [16] x. y. liu, z. t. wu, y. fan, and e. m. tentzeris, “a miniaturized csrr loaded wide-beamwidth circularly polarized implantable antenna for subcutaneous real-time glucose monitoring,” ieee antennas and wireless propagation letters, vol. 16, pp. 577-580, july 2016. [17] z. xia and h. li, “miniaturized dual-band implantable antenna,” ieee international conference on microwave and millimeter wave technology, may 2018, pp. 1-3. [18] i. gani and h. yoo, “multi-band antenna system for skin implant,” ieee microwave and wireless components letters, vol. 26, no. 4, pp. 294-296, march 2016. [19] “ansys hfss,” https://www.ansys.com/en-in/products/electronics/ansys-hfss, 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/). 44 microsoft word 3-peti#5291 23-29.docx proceedings of engineering and technology innovation, vol. 16, 2020, pp. 23 29 development and characterization of thermal insulation geopolymer foams based on fly ash michał łach 1,* , janusz mikuła 1 , wei-ting lin 2 , patrycja bazan 1 , beata figiela 1 , kinga korniejenko 1 1 institute of materials engineering, faculty of materials engineering and physics, cracow university of technology, cracow, poland 2 department of civil engineering, national ilan university, yilan, taiwan received 18 february 2020; received in revised form 30 april 2020; accepted 04 june 2020 doi: https://doi.org/10.46604/peti.2020.5291 abstract the main purpose of the article is to present the differences in the parameters of geopolymer foams obtained in the same way, which is associated with difficulties in controlling the foaming process. difficulties in controlling the foaming process of geopolymers are the direct reason for the lack of implementation of such materials nowadays. the article shows the results for experimental research, especially research into insulation, physical, and mechanical properties for the designated foamed materials. microspheres (75%), sand (5%) and fly ash (20%) were used to produce foamed geopolymers. hydrogen peroxide was the foaming agent. heat conduction coefficients of 0.08-0.07 w/mk were obtained. the material density was obtained at the level of 363-375 [kg/m 3 ] and the compressive strength was 520-683 [kpa]. the results showed that geopolymers can be a good alternative to conventional insulation materials, but the foaming technology should be developed so that it is stable and allows for reproducible material parameters. keywords: geopolymer, thermal insulation, mechanical properties, inorganic foams 1. introduction for many years, scientists have been developing insulation materials that have the best durability, non-flammability, and a relatively low price. one of such materials is foamed geopolymer. however, in order to make a foamed geopolymer be a real alternative to replace the currently used insulation materials, it must be met the relevant strength standards and passed the non-flammability test. different values of thermal conductivity and strength of foamed geopolymers are given in scientific paper. nevertheless, they usually differ from the thermal conductivity of popular materials such as polystyrene [1-2]. the best values in terms of thermal conductivity that scientific studies have shown so far are those in studies by liefke [3]. this is a material called trolit, which has a thermal conductivity of 0.037 w/mk. the selected parameters of this material are presented in table 1. nonetheless, since the material was first described [3-4], such a good parameter for geopolymer foams has not been noted again. scientists have tested a lot of compositions by using different materials and are trying to create the optimal material. a good quality of geopolymer insulating material which has density of 313 kg/m 3 , a compressive strength of 0.63 mpa, and a thermal conductivity 0.07 w/mk was obtained, for example, from a mixture of metakaolin and biomass ashes [5]. usually various types of additives are introduced such as diatomite [6]. the replacement of the diatomite in the formulation affects the morphological structure of the foams and then influences their mechanical properties due to a different type of * corresponding author. e-mail address: michal.lach@pk.edu.pl proceedings of engineering and technology innovation, vol. 16, 2020, pp. 23 29 24 chemical bond between sodium silicate and the solid phase. in particular, the mechanism of consolidation of hybrid foams based on the diatomite has changed from geopolymerisation to silicate polycondensation. therefore, the strength parameters improved with the increase of diatomite content [6]. metakaolin-based geopolymers foamed with h2o2, which the amounts from 0.25 wt.% up to 1.25 wt.%, were characterised by a compressive strength from 6 mpa (for 0.25% h2o2) to 0.36 mpa (for 1.25% h2o2). thermal conductivity was from 0.298 w/mk (for 0.25 wt.% h2o2) up to 0.172 w/mk (for 1.25 wt.%h2o2) [7]. to obtain the optimal parameters of the geopolymer foams, the metakaolin and fly ash are often used together. samson and cyr [8] proved that the most optimal composition for their type of raw materials was 62.5% metakaolin with 12.5% fly ash and 25% activator. the composition was expanded using h2o2 in amounts of 1, 1.5, and 2 wt.% and densities from 225 to 506 kg/m 3 and thermal conductivities from 0.07 to 0.12 w/mk were obtained. these studies have also shown that the use of surfactants has a huge impact on the porous structure [8]. it is also possible to use rice husks as a foaming agent [9]. additives to geopolymers are also very often introduced to improve strength properties. for instance, strength parameters of geopolymers can be improved by adding the lime and the gypsum [10]. fly ash and metakaolin beyond their use for not only use in the production of geopolymers, but also used in other technologies, for example, they can be used as a filling material in mineral composites [11]. table 1 physical characteristics of trolit™ foams [1, 2] bulk density [kg/m 3 ] 200 800 max. temperature of application [ o c] 1000 max. thermostability [ o c] 1200 thermal conductivity [w/mk] ≥0.037 (depending on density) pore parameter [mm] 0.5-3.0 fire protection [din 4102] inflammable, classification a1 compressive strength [n/mm 2 ] 0.5-2.0 tensile strength [n/mm 2 ] ≈250 this study presents the results obtained from foamed geopolymer materials based on fly ash. the results of compressive and bending strength tests depending on the density of the obtained materials are to be presented. in this study, the authors want to draw the problems when foaming geopolymers occur and show the fact that despite of attempting to control foaming conditions, the process will be different each time. the material described in this research differs significantly from that described in the literature because it contains up to 75% of microspheres. only 20% by the weight of fly ash was subjected to the foaming process. this material which is the combination of the open porosity obtained from foaming geopolymer and closed porosity obtained by microspheres is produced. 2. materials and methods fly ash from the skawina chp production plant was used as a precursor. the oxide composition of the precursor, determined by the xrf (x-ray fluorescence), is shown in table 2. table 2 oxide content of the fly ash from the skawina chp plant oxide sio2 al2o3 fe2o3 cao mgo na2o k2o so3 tio2 p2o5 bao content [%] 55.89 23.49 5.92 2.72 2.61 0.59 3.55 0.16 1.09 0.82 0.20 the geopolymers were produced by 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 mixture did not use distilled water. the alkaline solution was prepared, for instance, the solid sodium hydroxide was poured over an aqueous solution of sodium silicate and water. the solution was thoroughly mixed and allowed to equilibrate until a constant concentration and temperature were achieved. the geopolymer foams were prepared in such a way that solid constituents, such as fly ash, microspheres, and sand were mixed until reaching a homogeneous state, and afterwards an alkaline solution -10m sodium hydroxide solution with the addition of sodium silicate in the ratio 1:2.5 was introduced. the material was mixed for 10 minutes. an appropriate amount of proceedings of engineering and technology innovation, vol. 16, 2020, pp. 23 29 25 hydrogen peroxide was then introduced and continuously mixed for approx. 1 minute. after the mixing process was complete, the material was poured into a mould and exposed at a temperature of 75°c for a period of 24 hours. subsequently, it was the heating process; the element was dismantled and stored under laboratory conditions for a period of 28 days in order to obtain full strength. examples of the foamed geopolymer materials are presented in figs. 1 and 2. fig. 1 an example of foamed geopolymer panels fig. 2 an example of foamed geopolymer panels the thermal conductivity coefficient was calculated in accordance with pn-en 12667:2002. the measurements were carried out under the steady-state conditions of heat transfer at an average sample temperature amounting to 10°c, for the samples which dimensions were 300×300×100 mm. the applied temperature difference was 20°c for the thickness of the sample (t1=0°c, t2=20°c, tmean=10°c). the tests were conducted in a closed test chamber (having thermally-insulated walls) with the one-sample plate apparatus hfm 436/3/0, netzsch, provided with a heat-flow density sensor. the results were compared with the certified reference values irmm-440 no 30. the bulk density of the product was calculated as a mass ratio per unit volume of the sample, without reference to the certified specimen. the flexural strength test was conducted in accordance with en 12089 on samples of dimensions 300×50×150 mm. compressive strength tests were carried out according to en 826 on samples of dimensions 50×50×50 mm. 3. results the tests were implemented for one and the same composition, whose composition and heat transfer coefficients are given in table 3. however, three different variants of geopolymer panels were obtained due to the different courses of the foaming process. five different samples and measurements were made for each variant. for the obtained results, the standard deviation was calculated and the values of which were quite huge and exceeded several percent because of the heterogeneous porosity of the material. macroscopic and visual observations showed that the distribution of pores in the samples is uneven. the authors of the study often face the problem of uneven distribution of pores and their different sizes. the process of foaming geopolymers is difficult to control and attention. moreover, it should be paid to stabilise the produced foams. these problems cause a large dispersion in the obtained test results, despite the same process and composition, different results are obtained. table 3 composition of foamed geopolymers composition amount h2o2 thermal conductivity [w/mk] geofoam1 75% microspheres +5% sand +20% flyash 3% wt. 0.08094 geofoam2 75% microspheres +5% sand +20% flyash 3% wt. 0.07430 geofoam3 75% microspheres +5% sand +20% flyash 3% wt. 0.07492 table 4 test results of mechanical properties of obtained geopolymer foams density [kg/m 3 ] flexural strength [kpa] compressive strength [kpa] geofoam1 375.8±5.9 447.61±85.6 683.3±28.87 geofoam2 363.8±7.3 362.91±72.2 520.0±34.64 geofoam3 364.0±18.0 369.38±74.8 526.7±80.8 table 4 and fig. 3 show the results of testing the strength properties of the foamed geopolymer panels. the consequences of tests on the flexural strength and compressive strength are presented, depending on the density of the geopolymer material. the highest value of flexural strength is 447 kpa and the compressive strength is 683 kpa. proceedings of engineering and technology innovation, vol. 16, 2020, pp. 23 29 26 fig. 3 graphical representation of the density and strength of the materials it is the results of the sem / eds scanning microscope below showing the morphology for one of the samples to illustrate the heterogeneity of the material (figs. 4 and 5). point chemical composition analyses and visual observations show that not all the fly ash particles have been expanded. their composition indicates a smaller amount of na and an increased amount of ca (point 4). the observed structure is heterogeneous. fig. 4 microstructure of foamed geopolymer (a) point 1 (b) point 2 fig. 5 chemical point analysis 0 100 200 300 400 500 600 700 800 geofoam1 geofoam2 geofoam3 [k g /m 3 ]; [k p a ] flexural strength[kpa] compressive strength [kpa] density [kg/m3] proceedings of engineering and technology innovation, vol. 16, 2020, pp. 23 29 27 (c) point 3 (d) point 4 (e) point 5 fig. 5 chemical point analysis (continued) there are also variables at content points; for example, the fe and si/al ratios are variable. obviously, all these parameters affect the quality of the geopolymer foams obtained. in the drawing, points 3-5 show spherical particles, probably microspheres used to make geopolymers. they are characterized by a much higher content of al than the foamed structure revealed in points 1 and 2. thanks to the introduction of hollow spherical particles in the form of microspheres, it was possible to achieve both open and closed pore structure. the researchers would like to pay special attention to the heterogeneity of the obtained materials. the heterogeneity comes from the difficulties in controlling the foaming process and problems with the stabilization of the produced foams. 4. summary figs. 1 and 2 show the geopolymer materials characterized by uneven pore size distribution. it’s more distinctly visible in fig. 2 due to the low stability of the pores produced and their tendency to join. foamed structures disintegrate very quickly and proceedings of engineering and technology innovation, vol. 16, 2020, pp. 23 29 28 create larger pores coming to the surface. this is one of the reasons for the great difficulty in foaming geopolymer materials. scientists are attempting to incorporate various types of stabilizers into geopolymer foams. the uneven distribution of pores is also found in the photos from the scanning electron microscope in fig. 4. the geopolymer foams have similar density values in the range of 363-375 kg/m 3 . compressive and bending strengths vary for individual samples but to a small extent. similarly, the thermal conductivity coefficient differs for individual tests. despite the fact that the materials were made in the same way and their composition is identical, different parameters of the finished products were obtained. it is proved that the process of foaming geopolymers is very difficult to control. the lack of the stability of the obtained foams can lead to different parameters such as the heat transfer coefficient. even small differences may have higher thermal conductivity coefficients in this case. it should also be emphasized that in the presented research only 20% of fly ash was used to compose the geopolymer mixture and only these 20% of the components were subjected to the foaming process. if more fly ash and fewer microspheres were used, which additionally stabilize the produced foams, the differences in the results obtained would be greater. difficulties in controlling the foaming process, foam stability, and uneven pore size distribution are one of the main reasons for the lack of commercial implementations of this type of material. all research achievements in this area, new experiences are the most beneficial and one should strive to overcome these problems. the parameters of currently obtained foamed geopolymers are different, unfortunately, from the developed materials based on resins such as irmm-440. these materials are characterized by a very low density of 66-74 kg/m 3 and a heat transfer coefficient of 0.029 w/mk. when designing the geopolymers, the researchers should seek to obtain such the attractive parameters. the obtained thermal conductivity values of foamed geopolymers are better than when using the microwave method of foam production. when using microwaves to produce foamed geopolymers, it was possible to obtain a minimum thermal conductivity 0.2387w· m-1· k-1 and higher density values [12]. 5. conclusions foam geopolymer materials based on fly ash are interesting alternatives to currently popular materials such as polystyrene, mineral wool, and glass. they are a non-flammable material characterized by relatively good insulation parameters. the tests showed that the geopolymer foams, at a density of 375 kg/m3, have a compressive strength of about 0.68 mpa and the bending strength of 0.45 mpa. as the density decreases, the strength properties decrease. it is expected that this material will soon be implemented on an industrial scale as an alternative for the currently used insulation materials in construction. preparation for implementation, however, requires further research related to the stability and repeatability of the obtained geopolymer foams. it should also be striven to make it possible to use various waste materials for the production of foamed inorganic polymers. acknowledgements this work was supported by the polish national centre for research and development under the lider grant no: lider/31/0168/l-10/18/ncbr/2019. conflicts of interest the authors declare no conflict of interest. proceedings of engineering and technology innovation, vol. 16, 2020, pp. 23 29 29 references [1] m. łach, k. korniejenko, and j. mikuła, “thermal insulation and thermally resistant materials made of geopolymer foams,” procedia engineering, vol. 151, pp. 410-416, december 2016. [2] m. łach, d. mierzwiński, k. korniejenko, and j. mikuła, “geopolymer foam as a passive fire protection,” matec web of conferences, vol. 247, pp. 1-6, december 2018. [3] e. liefke, “industrial applications of foamed inorganic polymers,” geopolymere’99 proceedings, : institut géopolymère, december 1999, pp. 189-200. [4] j. davidovits, geopolymer chemistry and application, 4th ed. saint-quentin: institut géopolymère, 2015. [5] a. n. murri, v. medri, e. papa, l. laghi, c. mingazzini, and e. landi, “porous geopolymer insulating core from a metakaolin/biomass ash composite,” environments, vol. 4, no. 4, pp. 1-13, december 2017. [6] b. liguori, i. capasso, v. romeo, m. d auria, m. lavorgna, d. caputo, s. iannace, et al., “hybrid geopolymeric foams with diatomite addition: effect on chemico-physical properties,” journal of cellular plastics, vol. 53, no. 5, pp. 525-536, february 2017. [7] n. a. jaya, y. m. liew, m. m. a. b. abdullah, and c. y. heah, “porous metakaolin geopolymers with tailored thermal conductivity,” iop conference series: materials science and engineering, vol. 551, no.1, pp. 1-5, august 2019. [8] g. samson and m. cyr, “porous structure optimisation of flash-calcined metakaolin/fly ash geopolymer foam concrete,” european journal of environmental and civil engineering, vol. 22, no. 12, pp. 1482-1498, april 2017. [9] h. i. riyap, c. n. bewa, c. banenzoué, h. k. tchakouté, c. h. rüscher, e. kamseu, et al., “microstructure and mechanical, physical and structural properties of sustainable lightweight metakaolin-based geopolymer cements and mortars employing rice husk,” journal of asian ceramic societies, vol. 7, no. 2, pp. 199-212, april 2019. [10] v. sahu, a. srivastava, and v. gayathri, “effect of lime and gypsum on engineering properties of badarpur fly ash,” international journal of engineering and technology innovation, vol. 6, no 4, pp. 294-304, september 2016. [11] r. v. upadhyaya and t. g. suntharavadivel, “optimization of fly ash 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, may 2019. [12] m. nadeem, e. ulhaq, f. ahmed, m. asif rafiq, g. hameed awan, and m. zain-ul-abdein, “effect of microwave curing on the construction properties of natural soil based geopolymer foam,” construction and building materials, vol. 230, pp. 117074, 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/).  proceedings of engineering and technology innovat ion, vol. 6, 2017, pp. 17 22 the development of a new paper pallet machine chi-wei chen * department of marketing and distribution management, tajen university, pingtung, taiwan, roc. received 13 july 2017; received in revised form 15 july 2017, accept ed 19 july 2017 abstract in modern industry world, logistics plays an import role of enhancing the effic iency of shipping goods by use of pallets. the shipping pallets used in modern industry sites are generally made o f meta l, plastic, wood, or paper. among them, the wood pallet is used most often by many firms bec ause of low manufacturing cost. however, a wood pallet is not the best shipping selection by use in the area of air cargo transportation because of its weight and fro m the perspective of environ mental preservation. thus, for a ir cargo transportation, paper pallet is a better shipping selection because of its light weight advantage and also a good shipping tool for upgrading the competitiveness of taiwan’s electronic industry by reducing its air cargo shipping cost. in current paper pallet manufacturing mach ine requires many workers to work with and also generates many wasted papers, which is not a good way fro m the perspective of reducing cost and environmental preservation. a new manufacturing approach was thus proposed and a new machine was also designed to manufacture the newly designed paper pallet. the new paper pallet design was successfully imp le mented and the new prototype machine was also successfully developed. at least 20% of paper was saved comparing to the same size of paper pa llet made by the cu rrent best existing paper pallet machine. and the workers used in new manufacturing process of making paper pallet were a lso reduced at least 10 workers comparing to the process of current existing paper pallet manufacturing machine. keywor ds: paper pallet design, paper pallet machine, air cargo transportation , supply chain management, logistics management 1. introduction during the product manufacturing process and shipping process of finished goods, logistic s area was not being paid enough attention by many industries. thus, in 1962, peter drucke r ca lled the area of logistic s is “the economy’s dark continent” [1-2]. the ma jor reason is that many manufacturing firms are always seeking or ma king new machines, or develop new manufacturing processes to increase their productivities. in recent 20 years, there are more manufacturing firms realize to increase the efficiency of supply chain operation is also an import mean to enhance overall productivities. thus, this paper is intended to propose a new paper pallet design and a new paper pallet machine to enhance the efficiency of logistics operation . traditional ma rine shipping cargo is in the size of 20 or 40-foot long(x8-foot width x 8-foot height) container and usually use pallets with standard specification of 80x120c m or 100x120c m(euro spec.), 110x110c m(jis spec.), 110x110co m or 100x120c m(cns -ta iwan spec.) . in 2007, singapore, shanghai, hong kong ,shen zhen, yingkou(liaonian, china), kaoshiung are in the top 10 lis t of ma rine cargo import and e xport amount (20-foot equivalent unit, tue) , which was shown in table 1 and the international airport air cargo shipping amount was shown in table 2[3]. *corresponding author. e-mail address: cthomas@tajen.edu.tw tel: +886-8-7624002 ext.3213 proceedings of engineering and technology innovation, vol. 6, 2017, pp. 17 22 copyright © taeti 18 table 1 2007 world harbor ranking by import/export cargo cargo import/export amount (tues) ranking harbor name country tues 1 singapore singapore 27,935,500 2 shanghai china 26,152,400 3 hong kong china 23,998,440 4 shenzhen china 21,103,800 5 yingkou (liaonian, china) 13,713,000 6 busan south korea 13,254,703 7 rotterdam netherlands 10,790,604 8 dubai uae 10,653,026 9 kaoshiung taiwan 10,256,829 10 hamburg germany 9,917,180 in the case of marine shipping cargo, each tue needs at least about 10 standard pallets to support unit loads of products. therefore, huge a mount of pallets are needed in this globalizat ion modern world. also huge amount of pallets are made of wood and some special goods requested plastic pallets. and some heavy products need metal pa llets to carry goods. wood pallets are most often used by manufacturing firms because of its low cost. however, if goods are manufactured in ta iwan or china and are needed to be shipped to developed countries, such as usa, uk, or most european countries, wood pallets are needed to add up smoking treat ment and discard cost(usually burning cost), which is not a cost effect method to handle overse a shipping. the most difficult tas k for ta iwan or ch ina manufacturing firms to reduce the shipping cost is that an effic ient mach ine is not existed in modern wo rld wide. paper pa llets are always only one -third of weight a mount of the same size wood pallets. in the area of air ca rgo shipping, paper pallets have large advantages of reducing shipping cost. and about 75% of wood pallets could be replaced by paper pallets. regarding a ir cargo transportation, as in table 2, ta iwan and ch ina occupied top list of international a ir cargo ran king and indeed both taiwan and ch ina have many e lectronic products are needed to be shipped to countries worldwide. thus paper pallets are in big he lp of reducing shipping cost by us ing its light weight advantage and also in the perspective of environmental preservations. table 2 2008 international airport ranking by import/export cargo cargo import/export amount (tons) ranking international airport country tons 1 memphis international airport usa 3,695,438 2 hong kong international airport china 3,660,901 3 shanghai pudong international airport china 2,602,916 4 incheon international airport south korea 2,423,717 10 singapore changi airport singapore 1,883,894 15 taiwan taoyuan international airport taiwan 1,493,120 the proposed newly designed paper pallet mach ine is going to generate a new environ mental protection business with a manufacturing process without waste water and waste air produced. its manufacturing material could be fro m recycled papers and its scraps could be recycled and shipped to paper manufacturing factories to be used as raw materials. 2. existing paper pallet machine and paper pallet designs as mentioned above, both taiwan and china have la rge a mount of electronic products need to be shipped by a ir cargo to many fore ign countries worldwide. thus, the demand of paper pallets is numerous. co mparing to the end price between wood and paper pallets, wood pallets will not have much more advantages than paper pallets. ho wever, in the case of air cargo s hipping, paper pallets will be a better selection than wood pallets. in 2009, eu ropean union was intended to force other countries sellers to use paper pallets to ship goods to enter european un ion countries. this dec ision was hold back, because there is no efficient paper pallet machine available in recent industry markets. proceedings of engineering and technology innovation, vol. 6, 2017, pp. 17 22 copyright © taeti 19 in taiwan, there are about 30 sma ll co mpanies in the business as paper pallets suppliers. the manufacturing method used is all in a very old fashion way, pure ly made by hu man hands and simple tools, and without any mechanica l design s and calculations. the situation in china is that some manufacturing firms are using a paper pallet machine designed by netherland , which required many (at lease 20) worke rs to finish a complete paper pallet. the paper pallet machine developed by netherland was not designed in a highly automat ion way because during its manufacturing process that it still needs many human workers to perform tasks for work -in-p rocess. the manufacturing process es, including stage 1 and stage 2 shown in fig. 1 and fig. 2 respectively, are as follows: stage 1. manufacturing the core material of a bee-hives type paper pallet: fig. 1 the flow chart of making core materials of a bee-hives type paper pallet stage 2. forming a paper pallet fig. 2 the flow chart of making a complete bee-hives type paper pallet this type of paper pa llet is strong enough to support a unit load o f e lectronic products whose average we ight is usually less than 500kgs. usually for the bee-hives type paper pallets, static carrying strength could go up to more than 1000kgs, while dynamic carrying we ight usually less than 1000kgs because of the strength limits of the be -hives type core materia l. fro m cost management point of view, the bee-hives type paper pallet consumes too much human powers. 3. a proposed new paper pallet design and a new designed paper pallet m achine 3.1. a new paper pallet design the proposed paper pallet design has about 20 years of innovation patent protection. the main design of paper pallet ma in board differs fro m that of bee-hives type paper pallet ma in board is the core material inside the main board. the core material of the proposed paper pallet has wave type structure and during the manufacturing process, the core materia l does not need human powers to perform any cutting procedure and thus will not generate any scraps. the core mate ria l of e xisting bee-h ives type paper pallets need human powers (at least 2 persons) to perform stretching and cutting procedure before putting core materia l inside industry papers for succeeding wrapping procedure. the major drawback of the previous process is the time needed for bee-hives type core materia l after pasting glue, heating and drying glues used for adhere outside industry papers and core materia l. the density of glue and the temperature of heating box are key factors affecting the manufacturing efficiency. proceedings of engineering and technology innovation, vol. 6, 2017, pp. 17 22 copyright © taeti 20 the proposed wave type structure also faced the challenge of density of glue used and the heating temperature adjustment, which we re overcome eventually. the majo r challenges of wave type structure core materia l could be separated into two parts. first part is during shaping the wave type structure procedure. it will need very strong forces to transform a flat carton paper to wave type structure, while the wave height is the key factor of affecting final strength of paper pallets. second part is to place a wave type structure core material inside a pre -mold ing carton flat paper and to perform a wrapping procedure. as mention above, in this stage, the glue density and heating temperature setting are very important factors to affec t the strength of final paper pallets. this paper ma inly concentrates efforts on the main board design. the manufacturing of legs used for the proposed paper pallet ma in board is a minor proble m co mparing to the consideration for designing the paper palle t machine. the legs could be manufactured by special designed small molds. the samples of the proposed new paper pallet design are shown in fig 3 as follows: fig. 3 the samples of proposed paper pallets 3.2. the proposed new paper pallet machine in order to produce the paper pallets mentioned above, new manufacturing procedures are developed and are shown in fig. 4 as follows . a prototype machine was also developed to perform the following procedures : fig. 4 the proposed new paper pallet machine working procedures as shown in fig. 4, whole manufacturing procedures without any cutting process, therefore, it will not generate any scraps. the ma jor variab le cost for the ma in boards of paper pallet is fro m the weight of carton papers and indu stry papers used, and also fro m the wo rking hours of human powe rs used. just simply co mparing the manufacturing procedures between bee-hives type paper pallet and the proposed paper pallet, the latter shows a more efficient manufacturing procedure. 3.2.1. wave type structure shaping in this step, a design engineer needs to calculate the strength of carton flat paper to be used [4] and the feeding speed of ma in machine to perform wave type structure shaping [5-8]. if the selection of paper strength is not correct, it will increase the risk of tearing apart the carton flat papers during the shaping wave type structure process , which is an important procedure that proceedings of engineering and technology innovation, vol. 6, 2017, pp. 17 22 copyright © taeti 21 any engineer needs to be very cautious to avoid making mistakes. if a carton flat paper was tearing apart in the paper pallet mach ine, the whole machine has to be shut down and the process of removing the scraps inside the paper pallet machine is needed. the situations mentioned above must be tested during test runs. as the green light starts for making the main boards of proposed paper pallet, the machine must be sure that it will perform smoothly. the feed ing speed of ma king each ma in board could be adjusted up to 60 ma in boards per minute, however, the co ming cha llenge is the pasting and heatin g speed must be adjusted to match up the feeding speed of main machine. 3.2.2. gluing, pasting and heating procedures the key factors of adhering carton flat paper with wave type structure are not the a mount of glue used, nor the brand na me of glue used, but the density of glue used and the time needed to mix the surface fibers tangled together for both carton flat papers and wave type structure core materia ls. when an engineer find the right density of glue and the time needed to put both carton flat papers and wave type structure together, he has solved one of the most fetal problems of making paper pallets. for the p roposed paper pallet mach ine, before starting the machine, a stack of carton flat papers will put in front of mach ine in order to feed into the machine and performing wave type structure shaping procedure. in the same time, the bottom position of the paper pallet mach ine which is designed and placed under surface ground 2 meters depth with ma jor functions of feeding in carton flat papers, g luing, pasting on the bottom of the wave type structure. as soon as accomplishing the procedures mentioned above, the following step will be performing heating procedure to adhere both parts. the following step is wrapping another pre-mold ing carton flat paper on the work-in-process item which is a wave type structure was attached on a carton flat paper. the purpose of this step is to avoid the work-in-process item decreasing its peak heights. as its peak height decreased, the strength of the paper pallet decreased. the holding force could be applied by e xtending the machine structure of a strip of conveyors and apply ing jig and fixtures to withhold the work-in-process on the conveyors while continue its glue drying process, and which can be added up to 10, maybe 100 meters or more , if needed, depending on the space of factories one has or what budget one has. 3.3. purchasing cost, weight, and other factors among wood pallets, existing paper pallets in taiwan, and the proposed paper pallet table 3 comparisons of wood pallets and paper pallets (according to a pallet with 100cmx100c m, dynamic carrying weight 800kgs) proposed paper pallets wood pallets existing paper pallets (in taiwan) weight/each 7kgs 18~40kgs 8~12kgs manufacturing process shaping, pasting cutting, sawing, nailing gluing, pasting, cutting market price(nt$) 250 280~300 260~450 waste treatment recycled burning or burying recycled contaminant none wood scraps paper scraps automation feasible half automation hand made as shown in table 3, market price and light weight advantages of proposed paper pallets are better than that of wood pallets. co mpare to the producing speed between existing paper pallets manufacturers in taiwan and the proposed paper pallet mach ine, the latter shows much better automation advantage. the proposed paper pallet machine a lso could be applied in china, since many taiwan’s e lectronic manufacturers have invested in china and their products are needed to be shipped to countries worldwide. proceedings of engineering and technology innovation, vol. 6, 2017, pp. 17 22 copyright © taeti 22 3.4. other advantages of using paper pallets final step of discarding wood pallets is burning which will generate 100kgs of carbon dio xide per 100c mx100c m wood pallet. it was indicated in 2006[9] that one ton of carbon right trading price in european union ma rket is 20~25euro. thus, in the case of shanghai port (china), the ma rine shipping a mount is 26,152,400 t ues (in 2007). if 75% of marine shipping amount could be transferred to paper pellets shipping, the shipping amount will be 19,614,300 tues. each tue ne eds about 10 pallets to accomp lish unit load tasks, therefore, yearly shanghai port will consume 196,143,000 paper pallets. each paper pallet will reduce 100kgs of carbon dioxide generation which yearly could reduce 196,143,000 x 100 kgs = 19,614,300 tons of carbon dio xide. the previous a mount could be transferred to money wh ich is 19,614,300 x 20 euro = 392,286,000 euro per year. another fact is using recycled papers to ma ke pa llet s could reduce cutting trees. thus, using the previous data fro m shanghai port, yearly it could use 196,143,000 paper pa llets. each paper pallet we ights 7.5 kgs, yearly shanghai could generate 196,143,000 x 7.5kg / 1000kg = 1,471,070 tons of recycled papers. one ton of recycled paper will save cutting 17 trees, wh ich means that shanghai port yearly could save 1,471,070 x 17trees = 25,008,190 trees. 4. conclusions and future studies in this paper, a ne w design of paper pallet was proposed and a new paper pallet mach ine was also introduced. the ne w proposed machine could resolve the issues related to the paper pallet machine used in ch ina that was designed by netherland. the labor cost is increasing dramat ically late ly in ch ina, especially in southern-east coast area. low labor wage is no longer a huge advantage of investing manufacturing firms in ch ina. electronic manufacturing firms in ta iwan or china need to pay more attention to enhance the efficiency of global logistic management issues. also, the advantages of using paper pallets to acquire e xtra economic returns by engaging carbon right trading market and save cutting trees to protect our environment were a lso introduced. econo mic development a lso could ma int ain the sustainable environment preservations. the issue regarding shearing force of ma king wave type structure for the proposed paper pallet needs to be re-e xa mined carefully. expe riments could also be adopted for testing the strength ranges of carton flat papers used without tearing apart during the shaping procedure. an alternative is to develop a different method of feeding carton papers, not only increasing feeding speed but also reducing the risks of occurring carton paper jams. references [1] j. d. wisner, g. k. leong, and k. tan, principles of supply chain management-a balanced approach, south western, a division of thomson learning, inc., 2005. [2] peter f. drucker, “the economy’s dark continent,” fortune, p.14, april 1962. [3] douglas c. long and e. s. su, global logistics management, 2nd ed. taipei:hwatai publishing co., pp.248-249, june 2010.(in chinese) [4] a. c. ugural, mechanics of materials , taipei;new york:mcgra w-hill, 1991. [5] m. f. spotts, t. e. shoup, design of machine elements, 7th ed., new jersey: simom & schuster, 2002. [6] j. l. moum, z. lin, s. s. shei, etc., applied force-kinematics, taipei:gou-li publisher, january 1995.(in chinese) [7] mikell p. groover, automation, production system, and computer integrated manufacturing, new jersey:prentice-hall, inc., 1987. [8] mikell p. groover, fundamentals of modern manufacturing-materials, processes, and systems, 2nd ed., new jersey: prentice-hall, inc., 2002. [9] liberty times, taipei, 10 june, 2006.(in chinese) 3___peti#9025___20-26 proceedings of engineering and technology innovation, vol. 21, 2022, pp. 20-26 evaluation of the shapley additive explanation technique for ensemble learning methods tsehay admassu assegie * department of computer science, injibara university, injibara, ethiopia received 05 december 2021; received in revised form 02 january 2022; accepted 03 january 2022 doi: https://doi.org/10.46604/peti.2022.9025 abstract this study aims to explore the effectiveness of the shapley additive explanation (shap) technique in developing a transparent, interpretable, and explainable ensemble method for heart disease diagnosis using random forest algorithms. firstly, the features with high impact on the heart disease prediction are selected by shap using 1025 heart disease datasets, obtained from a publicly available kaggle data repository. after that, the features which have the greatest influence on the heart disease prediction are used to develop an interpretable ensemble learning model to automate the heart disease diagnosis by employing the shap technique. finally, the performance of the developed model is evaluated. the shap values are used to obtain better performance of heart disease diagnosis. the experimental result shows that 100% prediction accuracy is achieved with the developed model. in addition, the experiment shows that age, chest pain, and maximum heart rate have positive impact on the prediction outcome. keywords: model explanation, random forest, ensemble method, interpretability 1. introduction feature visualization and model explanation techniques play an important role to develop interpretable heart disease diagnostic models and improve the diagnostic accuracy of ensemble learning methods. to implement automated ensemble methods for heart disease diagnosis, model explanation techniques are required [1]. recent research focuses on analyzing the prediction outcome of ensemble learning methods and the interpretation of automated models for creating trust, reliability, and adoption of the ensemble learning methods. the explanation of ensemble learning methods is significant in getting insights into the prediction process of such methods and improving the accuracy of heart disease diagnosis [2]. in addition to improving the prediction accuracy, improving the interpretability of the predicted outcome is important for the adaptability and reliability of such automated methods because the explainability of the automated models can increase the trust and reliability of the prediction outcome. random forest is one of the most widely implemented methods for heart disease prediction due to high precision [3]. designing and implementing interpretable ensemble learning methods for heart disease diagnosis has become one of the recent research topics in machine learning [4]. the reason is that, in recent years, the size of heart disease datasets is growing and the complexity of heart disease diagnosis increases the need for ensemble learning methods to obtain better prediction outcome. however, the application of ensemble learning methods to heart disease diagnosis requires additional research effort to develop the automated models that are automatically from the dataset features. moreover, the prediction results of heart disease diagnosis provided by ensemble methods are more promising than those provided by linear methods such as support * corresponding author. e-mail address: tsehayadmassu2006@gmail.com tel.: +251-9-21114923 proceedings of engineering and technology innovation, vol. 21, 2022, pp. 20-26 vector machines [5]. thus, to develop reliable and adaptable ensemble learning methods for heart disease diagnosis, model interpretation is highly required. the reasoning for the prediction outcome made by the ensemble learning methods is important for domain experts to make decisions on whether the outcome is accepted. this study aims to improve the accuracy of an existing ensemble learning method for heart disease prediction by employing the feature visualization technique and the shapley additive explanation (shap) values with the real-world heart disease datasets obtained from kaggle data repository. experiments are conducted to analyze the features with significant influence on the prediction outcome by the shap values. overall, this study aims at achieving the following specific objectives. 1) the study provides a comprehensive and thorough analysis of the literature on the shap model explainability technique. 2) the study explores heart disease features that have significant influence on the prediction outcome of an ensemble method using random forest. 3) the study designs and implements a trusted and reasonable ensemble-based heart disease diagnosis method. 4) the study evaluates the performance of the developed ensemble method for heart disease diagnosis and analyzes the effectiveness of the shap technique in explaining the ensemble method. 2. literature review recent research has shown significant improvement in the performance of heart disease diagnosis using ensemble learning methods such as random forest on complex training samples [6]. the evaluation and interpretation of ensemble learning models for heart disease diagnosis have become an active field of research [7]. model interpretation is gaining much attention and research effort due to the fact that data scientists have noticed the importance of explaining why machine learning models have predicted an instance as a positive or negative class in the medical domain. interpretability is aimed to increase the trustworthiness and reliability of the predictive outcome made by machine learning models on risky decisions, particularly in the medical domain such as the prediction of diabetes disease [8]. interpretability has become a major concern in developing robust and trusted machine learning models for heart disease diagnosis. numerous previous works have investigated the predictive models of heart disease using machine-learning ensemble methods and deep learning algorithms. literature also shows that high accuracy is attained with the ensemble and deep learning methods [9]. however, the models are not widely adopted due to the lack of interpretability and transparency. model transparency, auditability, and explainability of ensemble learner models are critical to provide a maximally accurate heart disease prediction [10]. additionally, domain experts, regulators, or policy makers demand medical diagnostic models to be transparent and interpretable. thus, in medical diagnosis, the very simple predictive models such as logistic regression or decision trees are still widely employed for medical diagnosis. however, the superior predictive power of modern machine learning algorithms in solving complex problems with the growing size of datasets is not leveraged with simple models. thus, much research effort is needed to develop understandable medical diagnostic models and frameworks for making the “black box” machine learning models transparent, auditable, and explainable. the interpretability of models is gaining much research attention due to three reasons. first, predictive models are widely employed in different areas. second, simple models are not effective to provide good performance on complex problems. lastly, there is growing failure of adoption of complex models due to the interpretability and transparency problems. interpretable and explainable machine learning is regarded as a process, encompassing three high-level stages [11]. first, model interpretability is critical to understand ensemble learning models and datasets. second, model interpretability is important to improve the performance of models using explainable artificial intelligence (ai) methods. third, model interpretability and explanation are applied to refine and optimize machine-learning models. for instance, this study presents an application where interpretability is important. for heart disease prediction, the goal is to identify the patients to be classified into heart disease positive or heart disease negative classes. specifically, the problem is framed as a binary classification problem where the goal is to predict whether a patient is suffering from heart disease or not by considering the 21 proceedings of engineering and technology innovation, vol. 21, 2022, pp. 20-26 related factors: heart rate, maximum heart rate achieved (thalach), presence or absence of chest pain (cp), fasting blood sugar level (fbs), cholesterol level (chol), presence or absence of the exercise-induced angina, total resting blood pressure (trestbps), sex, age, slope, resting electrocardiogram (restecg), old peak, and other patient conditions. the contribution of this study is the use of the shap values to explain and interpret the prediction outcome of random forest with the shap technique. the literature shows that only few works are published regarding the explainability of ensemble learning methods on heart disease diagnosis [12]. however, the existing models do not explain the prediction outcome. to the best of the author’s knowledge, there are no prior published works on the use of interpretable random forest-based heart disease prediction models. this study explores the application of shap in the machine-learning context for medical diagnosis using heart disease datasets as a case study. overall, the study is designed to answer the following research questions: (1) how to increase the reliability of ensemble learning methods such as a random forest model for heart disease diagnosis? (2) how does the random forest model make predictions with heart disease datasets? (3) how to improve the prediction outcome of an ensemble learning method for heart disease diagnosis? (4) what is the influence of heart disease features on the prediction outcome of the random forest model? overall, the design and implementation process of the proposed automated heart disease diagnosis system is demonstrated in fig. 1. the purpose of this study is to develop a more interpretable and accurate automated heart disease diagnosis model using the random forest model and the shap technique. the reliability of the proposed model will be illustrated with the model prototype, and verified with real-world heart disease datasets. fig. 1 the process of model explanation with shap 3. design methodology and implementation the complexity of ensemble learning methods and the growing size of real-world heart disease datasets limit the ability to understand what a model has learned or why a given prediction is made, acting as a barrier to the adoption of machine learning [13]. additionally, from a legal or regulatory perspective, it is required to explain the prediction outcome of a machine-learning model. machine learning explainability (mlx) is a process of explaining and interpreting an ensemble learning model. to conduct the study, the author collects 1025 real-world heart disease datasets from university of california irvine (uci), obtained from a publicly available kaggle data repository, to the scientific community for experimentation. in this study, shap is employed to generate different explainers: objects used to compute the shap values for each sample in the datasets. these values represent the impact of certain features on the prediction outcome. positive values contribute to increasing the final probability, and negative values contribute to decreasing the probability. the experiment is performed on the jupyter notebook in python 3.7. the datasets are in comma-separated value (csv) file format. to be used for analysis, the data is imported to the jupyter notebook python environment using panda’s data analysis library in the form of the data frame. the libraries used for data transformation include pandas, matplotlib, scientific learning kit, and shap. to develop a model for heart disease diagnosis, the author employs random forest algorithms. the datasets are divided into a training set and a testing set. the training set includes 70% of the datasets or 717 observations, and the testing set includes 30% of the datasets or 308 observations. heart disease dataset random forest model shap random forest model random forest model 22 proceedings of engineering and technology innovation, vol. 21, 2022, pp. 20-26 4. existing design fig. 2 shows an existing ensemble learning-based automated heart disease diagnosis system. as can be seen, a model is developed using a feature selection method to improve the prediction outcome. while the application of feature selection improves the prediction accuracy of the system, the reliability and interpretability of the existing model limit the adaptability of such a model in the real-world scenario. thus, to address the existing gap (i.e., the lack of reasoning) for the prediction outcome of the ensemble learning model on heart disease diagnosis, the study proposes a new model with the help of the shap technique to increase the trust of domain experts on using automated models in decision-making. overall, this study aims to introduce the shap technique for model explanation as demonstrated in fig. 1. fig. 2 the existing design 5. experimental results and discussion this section presents the application of shap values for explaining and interpreting the prediction outcome of a random forest model on heart disease diagnosis. the experiment is conducted with shap for heart disease prediction with random forest algorithms. to evaluate the proposed random forest model, the author employs shap for improving the model performance and interpretability. the experiment is designed to explain the prediction output of the random forest model (i.e., the prediction of heart disease positive or heart disease negative). the author employs the heart disease prediction outcome as a case study to evaluate the interpretability of a random forest model using shap. the empirical analysis is conducted to evaluate shap-scores using heart disease datasets. all experiments are implemented in python 3.7 on jupyter notebook environment. the experiment is performed on intel(r) core(tm) i7-8565u cpu, 1.80 ghz, 1992 mhz, 4 core(s), 8 logical processor(s) using microsoft windows 10 pro 64-bit version. for data management and optimization, panda’s library is used. the shap values of each feature in heart disease datasets are demonstrated in fig. 3. as observed from fig. 3, chest pain (cp), the number of major vessels colored by fluoroscopy (ca), and thalassemia (thal) are the factors that contribute to the model’s prediction. thus, the number of major vessels colored by fluoroscopy (ca), chest pain (cp), and thalassemia (thal) are the features that have the greatest influence on the prediction outcome of the ensemble model. as demonstrated from the experimental result in fig. 3, a patient with chest pain is most likely predicted as a heart disease patient by the developed random forest model. the number of major vessels colored by fluoroscopy (ca) has the second most impact on the model output. as observed from the plot given in fig. 3, the features with lower shap values include resting electrocardiogram (restecg), fasting blood sugar level (fbs), and slope. feature selection with chi-square and correlation feature scaling using min-max scaler cleveland heart disease dataset data pre-processing classification using ensemble-learning algorithms performance analysis result and analysis 23 proceedings of engineering and technology innovation, vol. 21, 2022, pp. 20-26 the features with lower shap values have less impact on the model output as compared to the features with higher shap values. the features having high shap values impact the model output on the positive class (i.e., heart disease patients). in contrast, the features with lower shap values have positive impact on the predictive outcome of heart disease negative (i.e., patients without heart disease). the feature “resting electrocardiogram (restecg)” shows that the lowest shap value (approximately 0.4 shap value as illustrated in fig. 3) has positive impact on the predictive outcome of the negative class (class o). another important plot for analyzing the impact of features on the model output is the shap waterfall plot, demonstrated in fig. 4. fig. 4 demonstrates the impact of heart disease features on the random forest model output on heart disease prediction. as observed from fig. 4, the number of major vessels colored through fluoroscopy (ca) has the highest impact for the class of heart disease positive, but is the second most impactful feature for the positive prediction outcome using the shap summary plot demonstrated in fig. 3. ranked below the number of major vessels colored by fluoroscopy (ca), chest pain (cp), and thalassemia (thal) are the features which influence the model’s prediction towards the class of heart disease positive. in contrast, slope, age, resting electrocardiogram (restecg), exercise-induced angina, and total resting blood pressure (trestbps) influence the model’s prediction towards the class of heart disease negative. features such as maximum heart rate (thalach), fasting blood sugar level (fbs), and cholesterol level (chol) have lower effects on the model’s prediction as compared to other heart disease features as demonstrated in fig. 4. fig. 5 demonstrates the interaction plot between maximum heart rate (thalach) and cholesterol level (chol). as illustrated in fig. 5, maximum heart rate (thalach) and cholesterol level (chol) have a linear relationship with each other, showing that they are highly correlated. the strong correlation between them indicates their dependency on each other. fig. 3 shap values of each feature fig. 4 shap waterfall plot fig. 5 shap interaction plot of maximum heart rate (thalach) vs cholesterol level (chol) fig. 6 roc curve of the developed model 24 proceedings of engineering and technology innovation, vol. 21, 2022, pp. 20-26 fig. 7 interaction plot of total resting blood pressure (trestbps) vs age fig. 8 interaction plot of maximum heart rate (thalach) vs age fig. 6 demonstrates the receiver operating characteristic (roc) curve of the random forest model for heart disease diagnosis. as can be seen, the area under the roc curve is 1.00, which shows that the model is perfect for the detection of heart disease. moreover, the area score or the area under curve (auc) value of 1.00 evidently proves that the model performs well and does not have the prediction outcome of false positive and false negative classes on the test set conducted in the experiment. fig. 7 demonstrates the shap interaction plot between total resting blood pressure (trestbps) and age. as demonstrated in fig. 7, the shap interaction values for total resting blood pressure (trestbps) are higher for the patients above the age of 50 years. thus, the plot evidently proves that there is high positive interaction between age and total resting blood pressure (trestbps). fig. 8 illustrates the shap interaction value for maximum heart rate (thalach) and age. as observed from fig. 8, the shap interaction value for maximum heart rate (thalach) is higher for the patients above the age of 45 years. thus, the plot appears to prove that there is high positive interaction between age and maximum heart rate (thalach). 6. conclusions this study presents an explainable and automated heart disease diagnosis model using the shap model explanation technique and random forest-based ensemble learning method. with the shap technique, an interpretable random forest model with 100% prediction accuracy is successfully developed. the shap technique is effective for providing the intuition and the reasoning behind the predictive outcome of ensemble learning methods such as random forest for heart disease diagnosis. for future work, it is recommended to use various model explanation techniques and machine learning algorithms, such as extreme boosting and deep learning, to identify whether a patient is suffering from heart disease or not. the results of future work will facilitate the trusted and reliable automated intelligent heart disease diagnosis models that would become alternatives to the healthcare experts. conflicts of interest the author has no conflicts of interest. references [1] s. liu, x. wang, m. liu, and j. zhu, “towards better analysis of machine learning models: a visual analytics perspective,” visual informatics, vol. 1, no. 1, pp. 48-56, march 2017. [2] k. aas, m. jullum, and a. løland, “explaining individual predictions when features are dependent: more accurate approximations to shapley values,” artificial intelligence, vol. 298, article no. 103502, september 2021. [3] a. chatzimparmpas, r. m. martins, i. jusufi, and a. kerren, “a survey of surveys on the use of visualization for interpreting machine-learning models,” information visualization, vol. 19, no. 3, pp. 207-233, july 2020. 25 proceedings of engineering and technology innovation, vol. 21, 2022, pp. 20-26 [4] j. zhou, a. h. gandomi, f. chen, and a. holzinger, “evaluating the quality of machine learning explanations: a survey on methods and metrics,” electronics, vol. 10, no. 5, article no. 593, march 2021. [5] p. linardatos, v. papastefanopoulos, and s. kotsiantis, “explainable ai: a review of machine learning interpretability methods,” entropy, vol. 23, no. 1, article no. 18, january 2021. [6] h. s. yan, m. c. tsai, and m. h. hsu, “an experimental study of the effects of cam speeds on cam-follower systems,” mechanism and machine theory, vol. 31, no. 4, pp. 397-412, may 1996. [7] d. farrugia, c. zerafa, t. cini, b. kuasney, and k. livori, “a real-time prescriptive solution for explainable cyber-fraud detection within the igaming industry,” sn computer science, vol. 2, no. 3, article no. 215, may 2021. [8] k. futagami, y. fukazawa, n. kapoor, and t. kito, “pairwise acquisition prediction with shap value interpretation,” the journal of finance and data science, vol. 7, pp. 2244, november 2021. [9] m. chaibi, e. m. benghoulam, l. tarik, m. berrada, and a. e. hmaidi, “an interpretable machine learning model for daily global solar radiation prediction,” energies, vol. 14, no. 21, article no. 7367, november 2021. [10] p csókaa, f. illés, and t. solymosi, “on the shapley value of liability games,” european journal of operational research, in press. [11] c. m. viana, m. santos, d. freire, p. abrantes, and j. rocha, “evaluation of the factors explaining the use of agricultural land: a machine learning and model-agnostic approach,” ecological indicators, vol. 131, article no. 108200, november 2021. [12] s. n. payrovnaziri, z. chen, p. rengifo-moreno, t. miller, j. bian, j. h. chen, et al., “explainable artificial intelligence models using real-world electronic health record data: a systematic scoping review,” journal of the american medical informatics association, vol. 27, no. 7, pp. 1173-1185, 2020. [13] k. dissanayake and m. g. m. johar, “comparative study on heart disease prediction using feature selection techniques on classification algorithms,” applied computational intelligence and soft computing, vol. 2021, article no. 5581806, 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/). 26  proceedings of engineering and technology innovation, vol. 7, 2017, pp. 01 07 a designed enclosure unit for loop-mediated isothermal amplification arayan wongniyom 1 , jyh-jian chen 2,* 1 department of tropical agriculture and international cooperation, national pingtung university of science and technology, pingtung, taiwan, roc. 2 department of biomechatronics engineering, national pingtung university of science and technology, pingtung, taiwan, roc. received 17 july 2017; received in revised form 05 august 2017; accepted 25 august 2017 abstract the gene amplification technique is used in medical science and biotechnology popularly. the loop-mediated isothermal amplification (lamp) is a nucleic acid amplification method that can be utilized in low-resource settings and does not require expensive or complex instruments. the annealing process of lamp is performed at a constant temperature (around 65 °c) for an hour. from previous works, phase change materials (pcms) have been applied for the latent heat energy storages due to their isothermal operation during phase transitions. the utilization of the melting temperature of the paraffin wax to maintain the suitable temperature for lamp process is introduced in our study. to enhance the heat transfer inside the energy storage unit, several pin fins are constructed on the inner surface of the unit which is 50 mm × 50 mm × 50 mm in inner size. the influence of various numbers of pin fins on the temperature distribution is examined by the numerical simulation. the heat transfer inside the paraffin enclosure with fins is better than that without fin. it can be found that the temperature uniformity in the pcm inside the enclosure with fins shows much better. though the numbers of the fin are increased, some transfer energy from the heat source remains at the enclosure wall and cannot penetrate into the pcm. the duration for all pcm melted from solid phase to liquid phase is over 10 minutes. in the future, the geometric parameters on the duration for phase transition of pcm inside the lamp chamber will be investigated. keywords: phase change materials, lamp, paraffin wax 1. introduction the detection of nuclei-acid sequences is popularly used in medical research, biotechnology and food science. the loop-mediated isothermal amplification (lamp) is a biotechnique that can amplify deoxyribonucleic acid (dna) within 60 minutes under isothermal conditions (50-65 °c). this technique, a highly sensitive technique that can be used in low-resource settings and does not require expensive or complex instruments [1], is suitable for small-scale hospitals, primary care facilities and clinical laboratories in developing countries [2-3]. in previous studies, the lamp technique was able to yield the amplification of 109-bp of dna [4], and applied to detect the virus as follows: mycobacterium [1], tuberculosis [5], m. avium and m. intracellulare in sputum [3], porcine kobuvirus [6], etc. * the lamp process is to anneal dna and primers with an appropriate constant temperature. to build a reaction device for lamp process, one of the most important issues is to design a reaction chamber that can be kept at a constant temperature for a period of time. when heat energy is supplied to or extracted from a phase change material (pcm), there is only a little * corresponding author. e-mail address: chaucer@mail.npust.edu.tw tel.: +886-8-7703202 ext. 7029, 7839(l) mailto:chaucer@mail.npust.edu.tw proceedings of engineering and technology innovation, vol. 7, 2017, pp. 01 07 copyright © taeti 2 change of temperature inside the pcm. therefore, pcm can be used to stabilize the temperature in an application for lamp process. in general, pcm refers to solid-liquid phase changes only [7]. paraffin wax is the most commonly used commercial organic pcm because of its high heat of fusion, varied phase change temperatures, negligible supercooling, lower vapor pressure in a melt and chemically inert and stable behavior [8]. however, due to low thermal conductivity of paraffin wax, it requires large surface area for heat exchange [9]. kandasamy et al. [10] studied about using the paraffin wax as the pcm-based heat sink. the results from simulations and experiments have shown that pcm-based heat sink can be used in intermittent-use devices. nayak et al. [11] improved the cooling of the electronics devices by using pcm with fins made of thermal conductivity enhancer material. pakrouh et al. [12] presented a numerical investigation to optimize the geometry of pcm-based fin heat sinks. the result showed the fin height affects the base temperature. hosseinizadeh et al. [13] carried out a comparison between heat sink with and without pcm. their results revealed that fins thickness larger than 4 mm did not improve the performance of the heat sink. kamkari et al. [14] demonstrated the experimental results of pcm melting in an enclosure with and without horizontal partial fins. results indicated that when the surface-averaged nusselt number was reduced, increasing the number of fins decreased the melting time and increased the total heat transfer rate. lamp has been utilized for the detection of dna and/or rna from a wide range of bacterial and viral pathogens [15-17]. curtis et al. [15] presented a non-instrumented nucleic acid amplification (nina) heating device. the heat from the exothermic chemical reaction was enough to anneal dna sample for 60 minutes. however, the changing of the ambient temperature had an effect on the temperature stability during the amplification process and it was difficult to control the heat from the chemical reaction. in our lamp device, a cartridge heater is utilized to generate heat energy for melting pcm and then the melting pcm is used for stabilizing the sample temperature during the dna amplification. this process does not need any thermal controller and the integration of the lamp device was easy. 2. numerical methodology a schematic diagram of three-dimensional the enclosure physical model consist of lid, fin array and cavity, the dna sample put on the top surface of the lid is shown in fig. 1(a). considering the symmetry of the geometry, only one quarter of the physical model is included for computation. the pin fin array and the cavity are made of aluminum which contains the pcm. a uniform heat power is applied at the bottom surface of the cavity. initially, the solid pcm and enclosure are in thermal equilibrium with the ambient temperature, 20 °c. the solid and liquid phases of pcm are homogeneous and isotropic. thermal resistance across the wall of the container is neglected and lateral sides of the rectangular enclosure are well insulated. the melting point of pcm is 65 °c. the effect of natural convection within the melt pcm is neglect because the space into enclosure is moderately narrow and fin array barrier velocity flow when phase transition to liquid. the pcm undergoes the transition from the solid state to the liquid state. there are governing equations as follows. heat conduction in liquid region: 𝜕𝑇𝑙 𝜕𝑡 = 𝛼𝑙 𝜕2𝑇𝑙 𝜕𝑥2 + 𝛼𝑙 𝜕2𝑇𝑙 𝜕𝑦2 + 𝛼𝑙 𝜕2𝑇𝑙 𝜕𝑧2 , 𝑡 > 0 (1) heat conduction in solid region: 𝜕𝑇𝑠 𝜕𝑡 = 𝛼𝑠 𝜕2𝑇𝑠 𝜕𝑥2 + 𝛼𝑠 𝜕2𝑇𝑠 𝜕𝑦2 + 𝛼𝑠 𝜕2𝑇𝑠 𝜕𝑧2 , 𝑡 > 0 (2) interface temperature: 𝑇(𝑥, 𝑦, 𝑧, 𝑡) = 𝑇𝑚 (3) proceedings of engineering and technology innovation, vol. 7, 2017, pp. 01 07 copyright © taeti 3 stefan condition: 𝑘𝑠 𝜕𝑇𝑠 𝜕𝑥 − 𝑘𝑙 𝜕𝑇𝑙 𝜕𝑥 = 𝐻𝑓 𝜌 𝑑𝑋 𝑑𝑡 , 𝑡 > 0 (4) where 𝛼 is thermal diffusivity. k is the thermal conductivity. 𝛾 is the melt fraction during the phase change process and is defined by the following relations: 𝛾 = 0 , if 𝑇 < 𝑇𝑠; 𝛾 = 0 , if 𝑇 > 𝑇𝑙 ; 𝛾 = 𝑇−𝑇𝑠 𝑇𝑙−𝑇𝑠 , if 𝑇𝑠 < 𝑇 < 𝑇𝑙 . tm is melting temperature, the subscript l and s is liquid and solid phase respectively. a grid system is composed of 32768 to 61952 hexahedral cells, which depends on the numbers of fin. the numbers of fin are varied from 4 to 49 pin fins, h is fin height, which is varied from 20 mm to 47 mm and the fin maintains 2 mm thickness of whole models shown in fig. 1(b). a constant heat power of 14 w is applied at the bottom surface of the enclosure. a finite volume approach is adopted by using computational fluid dynamics (cfd) software esi-cfd (v2006, cfd research corporation, huntsville, al, usa) to model the pcm melting process. a structured grid system with good orthogonality is recommended for accurate prediction and reliability. in this study the geometric model for the three dimensional rectangular structure is considered. in the computation of the enthalpy of the simulation model, the spatial discretizations are performed using a second-order upwind scheme. the conditions of convergence can be divided into two kinds. the first one is the maximum number of the solver iterations. the second one determines the convergence criteria to be used. in this study, we perform 100 sweep times and use 5×10-5 as the convergence criterion. grid sensitivity is evaluated by using different element sizes to ensure that the simulated results are independent of element size. the element size is repeated until the similarity in results is within the acceptable solution tolerance (0.1%). this study use the pcm, which has the melt point near the amplification process of lamp technique is the rubitherm® rt64hc paraffin wax has a solid density of 880 kg/m 3 , a liquid density of 780 kg/m 3 , a specific heat of 2000 j/kg-k, a latent heat of 2.3×10 5 j/kg and a thermal conductivity of 0.2 w/m-k. aluminum has a density of 2700 kg/m 3 , a specific heat of 880 j/kg-k and a thermal conductivity of 190 w/m-k. (a) the quarter part of the enclosure and pcm (b) bottom view of the heat sinks with 4, 9, 20, 28, 36, and 49 pin fins fig. 1 the numerical models of 2 mm thickness fin 3. results and discussion in this section, we present the thermal characteristics obtained from the three dimensional simulations and the results of the parametric studies are shown. the discussion concerns the calculated temperatures and melt fractions of the pcm as functions of time. in our study, heat conduction is the main heat transfer mechanism inside the enclosure. the heater generates heat flux is attached to the bottom of the aluminum enclosure with a high thermal conductivity. with an increase in time, the proceedings of engineering and technology innovation, vol. 7, 2017, pp. 01 07 copyright © taeti 4 heat energy from pin fin array and cavity through to the solid pcm, melting of the pcm and then heating of the molten pcm to temperature above its melting point. the material of the fins is identical to the enclosure. therefore the temperature of the enclosure and the fins increase more than that of pcm. the temperature of the lid might exceed the temperature suitable for amplification process in lamp technique. the purpose of the fins is enhancing thermal conductivity of pcm that contain inside an enclosure, the heat flux flow with heat conduction from the heat source at the bottom cavity aluminum to pcm. the large disparity thermal conductivity between aluminum and pcm that make effect to the intensive heat flux flow in material. the fins array is proficiently constrained the temperature for annealing the sample on the top surface enclosure shown in fig. 2(a). in order to assess the influence of fin, melt fraction with time variation is evaluated for the pcm with fins and without fin (a) temperature contour of 36 pin fins and fin height of 47 mm at top view at 110 s (b) compared melting fraction varied (c) compared melting fraction varied fin height. fig. 2 the melt contour inside enclosure of pcm and melt fraction effected of number of pin fin and fin height after 1800s, there are three cases of pcm no complete melting to liquid as follows: without fin, 20pins fin and 47mm fin height, and 49 pins fin and 20mm fin height fig. 2(b) and 2(c) show the temporal variations of melt fractions during melting processes for finned enclosure and enclosure without fun. a rapid increase in temperature of pcm is observed with fins than that of pcm without fins. this is because the longitudinal finned material is able to transfer heat from the heat source quickly. the higher heat transfer rate was induced by the presence of fins near the heat source causes faster melting of pcm. thus, the total melting time is reduced compared to the system without fins. the fins accelerate the heat transfer rate from the surface of heat source to and through the pcm. the melt fraction of pcm with 49 pin fins and fin height of 47 mm is higher than that of other cases. regarding to the enclosure with 36 pin fins, the melt fraction is lower than that with 49 pin fins, but is almost the same. therefore, the enclosure with 36 pin fins and fin height of 47 mm is capably performed as well. proceedings of engineering and technology innovation, vol. 7, 2017, pp. 01 07 copyright © taeti 5 (a) compare without fin with 49 pin fins, 47 mm fin height (b) at base cavity (c)pcm fig. 3 average temperature profiles the gradients of the temperature contours are large at the bottom of the enclosure during the observed times because the applied heat source is at the bottom of the enclosure. the pcm has a low thermal conductivity. therefore, the large heat flow along aluminum wall causes high temperature, the pcm temperature is improving almost lid temperature when applied fin array the 49 pins fin and 40mm fin height shown in fig. 3(a). although concentrated temperature at the base cavity in 49 pins and 40 mm fin height is most (neglect without fin case) fig. 3(b), the temperature profiles of pcm almost 49 pins and 47 mm fin height case after 900s. the solid pcm was heated by the adjacent liquid pcm. hence fin array enhanced and increased the melting rate. the extended surfaces of the fin array are inserted into pcm and lead to more heat transfer than the enclosure without fin, as shown in fig. 3(c). the fin array and cavity structures cause the heat energy penetrated into the pcm. the largely different thermal conductivity between aluminum and pcm has the effect to intensive gradient temperature occur on surface heat transfer of aluminum, demonstrated in fig. 4(a,b), and the space without fin remain solid state indicate that the heat transfer between liquid to solid not well. the melting rate is enhanced by pin fin array. the number of pin fin and fin height directly influence the surface heat transfer. when the surface heat transfer is increased, the heat energy for phase transition is reduced in fig. 4(c). it is because the large heat flux through surface fin to pcm change to complete the melting, the liquid pcm cannot absorb energy. the influence of various fin heights and numbers of pin fin on melt fraction when each fin surface heat transfer is almost equal is shown in fig. 4(c). the surface heat transfer are 22840 mm 2 with 49 pin fins and 20 mm fin height, 22528 mm 2 proceedings of engineering and technology innovation, vol. 7, 2017, pp. 01 07 copyright © taeti 6 with 28 pin fins and 47 mm fin height and 22520 mm 2 with 20 pin fins and 47 mm fin height, respectively. the 49 pin fins and 40 mm fin height has surface heat transfer 30680 mm 2 less surface heat transfer than 49 pin fins and 47 mm fin height (33424 mm 2 ) and use latent heat energy for phase transition. the difference of the melt fractions among these cases are within 10% after the beginning of the duration of 300 s. therefore, the latent heat energy storage depends on the surface of the heat transfer. the saved energy for each model can be found by integrating the area under the relative latent heat energy and time. the slope of each graph indicates the energy to be used within 1800s. the group of 49 pin fins saves more energy than others with large space between adjacent fins. (a)top view cross-section at half of fin length (b) front view cross-section at middle of first row of fin (c) latent heat energy and surface heat transfer fig. 4 temperature contours of 36pin 47mm fin height model at 300 (a) the area under each the graph (b) the slope value each graph of latent heat energy fig. 5 the latent energy in each case used to melt pcm in 1800s proceedings of engineering and technology innovation, vol. 7, 2017, pp. 01 07 copyright © taeti 7 4. conclusion in this paper, the numerical study of the designed pcm enclosure for lamp was presented. the transient simulation of heat conduct problem of an enclosure with pcm is shown with several pin fins inside the enclosure. the main parameters about enhancing the heat transfer were the number of pin fin and the fin height. the simulation results show an enhancement of melting rate and a decreasing temperature gradient near the heating wall of enclosure with fin array extended from lid. the thorough pin fin position distribution and the number of pin fin can be used to save the energy to melt pcm. the simulation study shows the pin fin array technique can improve the potential of pcm for applying in the lamp machine. acknowledgment the authors would like to thank the ministry of science and technology of the republic of china for financially supporting this research under contract no. most 105-2313-b-020-007-. references [1] e. aryan, m. makvandi, a. farajzadeh, k. huygen, p. bifani, and s. l. mousavi et al, “a novel and more sensitive loop-mediated isothermal amplification assay targeting is6110 for detection of mycobacterium tuberculosis complex,” microbiological research, vol. 165, no. 3, pp. 211-220, 2010. [2] y. mori and t. notomi, “loop-mediated isothermal amplification (lamp): a rapid, accurate, and cost-effective diagnostic method for infectious diseases,” journal of infection and chemotherapy, vol. 15, no. 2, pp. 62-69, 2009 [3] t. iwamoto, t. sonobe, and k. hayashi, “loop-mediated isothermal amplification for direct detection of mycobacterium tuberculosis complex, m. avium, and m. intracellulare in sputum samples,” journal of clinical microbiology, vol. 41, no. 6, pp. 2616-2622, 2003. [4] t. notomi, h. okayama, h. masubuchi, t. yonekawa, k. watanabe, and n. amino et al, “loop-mediated isothermal amplification of dna,” nucleic acids research, vol. 28, no. 12, pp. e63-e63, 2000. [5] c. c. boehme, p. nabeta, g. henostroza, r. raqib, z. rahim, and m. gerhardt et al., “operational feasibility of using loop-mediated isothermal amplification for diagnosis of pulmonary tuberculosis in microscopy centers of developing countries,” journal of clinical microbiology, vol. 45 no. 6, pp. 1936-1940, 2007. [6] c. li, j. chen, h. shi, x. zhang, d. shi, and x. han et al., “rapid detection of porcine kobuvirus in feces by reverse transcription loop-mediated isothermal amplification,” virology journal, vol. 11, no. 1, pp. 73, 2014. [7] h. mehling and l. f. cabeza, heat and cold storage with pcm, berlin, springer, 2008. [8] m. k. rathod and j. banerjee, “thermal stability of phase change materials used in latent heat energy storage systems: a review,” renewable and sustainable energy reviews, vol. 18, pp. 246-258, 2013. [9] m. akgün, o. aydın, and k. kaygusuz, “experimental study on melting/solidification characteristics of a paraffin as pcm,” energy conversion and management, vol. 48, no. 2, pp. 669-678, 2007. [10] r. kandasamy, x. q. wang and a. s. mujumdar, “transient cooling of electronics using phase change material (pcm)-based heat sinks,” applied thermal engineering, vol. 28, no. 8, pp. 1047-1057, 2008. [11] k. c. nayak, s. k. saha, k. srinivasan, and p. dutta, “a numerical model for heat sinks with phase change materials and thermal conductivity enhancers,” international journal of heat and mass transfer, vol. 49, no. 11, pp. 1833-1844, 2006. [12] r. pakrouh, m. j. hosseini, a. a. ranjbar, and r. bahrampoury, “a numerical method for pcm-based pin fin heat sinks optimization,” energy conversion and management, vol. 103, pp. 542-552, 2015. [13] s. f. hosseinizadeh, f. l. tan, and s. m. moosania, “experimental and numerical studies on performance of pcm-based heat sink with different configurations of internal fins,” applied thermal engineering, vol. 31, no. 17, pp. 3827-3838, 2011. [14] b. kamkari and h. shokouhmand, “experimental investigation of phase change material melting in rectangular enclosures with horizontal partial fins,” international journal of heat and mass transfer, vol. 78, pp. 839-851, 2014. [15] k. a. curtis, d. l. rudolph, i. nejad, j. singleton, a. beddoe, and b. weigl et al., “isothermal amplification using a chemical heating device for point-of-care detection of hiv-1,” plos one, vol. 7, no. 2, p. e31432, 2012. [16] j. singleton, j. l. osborn, l. lillis, k. hawkins, d. guelig, and w. price et al., “electricity-free amplification and detection for molecular point-of-care diagnosis of hiv-1,” plos one, vol. 9, no. 11, p. e113693, 2014. [17] j. singleton, c. zentner, j. buser, p. yager, p. labarre, and b. h. weigl, “instrument-free exothermic heating with phase change temperature control for paper microfluidic devices,” spie proceedings, vol. 8615, p. 86150r, 2013.  proceedings of engineering and technology innovation, vol. 12, 2019, pp. 26-34 job mix formulation for bituminous concrete grade ii using c# programming v. venkat ramayya 1,* , v. vinayaka ram 2 , s. krishnaiah 3 1 department of civil engineering, mahatma gandhi institute of technology, hyderabad, india 2 department of civil engineering, bits, hyderabad, india 3 department of civil engineering, jntua, anantapuramu, india received 16 january 2018; received in revised form 04 march 2018; accepted 17 april 2018 abstract bituminous concrete mix design is a complex process and different methods are proposed by a number of researchers. marshall’s method of mix design is one such process, which became very popular among the practitioners in india and elsewhere. superpave method of mix design which takes into account climatic and traffic factors for aggregate and binder selection. it has gained its prominence and is observed to be superior to any other well-known mix design processes. aggregate gradation adopted for the mix design plays a crucial role in the performance of the mix. many researchers have developed mechanisms to work out job mix formulations (jmf) so as to reach the appropriate aggregate mix in the field. the analytical phase of jmf deals with the method of blending of aggregates from the identified stockpiles in order to fulfill the requirements of gradation as specified by agencies for a given layer of flexible pavements. methods proposed for aggregate blending includes the graphical methods, trial and error method and optimization techniques. the aim of these techniques is two-fold; the first is to optimize the cost involved in blending from the known unit cost of individual stockpiles; and the other is to fulfill the gradation recommendations of the local agencies and parameters suggested in other methods. in this research, an attempt has been made to develop c# (c-sharp) application in .net framework using the sequential search technique to work out the best blend of aggregate particles by ensuring the requirements mentioned in morth’s (ministry of road transport & highways, india) recommended practice. bailey’s method and the maximum density line have been recommended in 0.45 power chart and incorporating optimization criteria. grade ii bituminous concrete has been chosen for the research. it was clearly observed that the fine aggregate fraction in the stockpiles influences the final job mix formulation. the devleloped tool can accommodate up to 5 stockpiles. it was also observed that the bailey’s requirements are more stringent when compared with morth requirements as far as jmf is considered. the developed tool is capable of generating the blends with recycled asphalt pavement (rap) aggregate blends. keywords: bituminous concrete, maximum density line, bailey’s parameters, aggregate blending, c# programming, job mix formula. 1. introduction the flexible pavement is a layered system where the load transfer takes place through the grain to grain contact [1]. in each of these layers, an aggregate blend with different sizes will be used along with the bitumen binder, which forms the binder matrix and holds the aggregates in place. the bituminous concrete mix design is a complex process. marshall’s method of mix *corresponding author. e-mail address: ramayya@mgit.ac.in tel.: +91-99669 20375 proceedings of engineering and technology innovation, vol. 12, 2019, pp. 26-34 27 design is the most popular in india and elsewhere [2].the marshall specimen prepared using the optimum binder content (obc) shall satisfy the parameters like stability, flow, volume filled with bitumen (vfb), volume in mineral aggregates (vma) and volume of voids (vv) [9]. agencies from different countries have standardized the requirements of these parameters and it is mandatory to verify these parameters before adopting a given mix in the field. a number of internal and external factors are responsible for distress in flexible pavements. aggregate gradation is one of the internal factors. the gradation adopted is going to influence the performance of dense graded, open graded and stone matrix asphalt mixes. research has shown that aggregate gradation will influence the performance of porous asphalt mixes. these pavements are now in use for effective stormwater management.the performance of porous asphalt mixes is studied through permeability and air voids [3]. the aspects of strength and durability of these mixes are related to the gradation [3 ]. the aggregate gradation adopted is known to have its influence on mobility, ability to be compacted, stiffness, durability and permeability of the asphalt mixes [4]. mixes with ensured continuity of gradation were observed to have better rutting resistance [6]. apart from the gradation chosen for the mix, the nominal maximum particle size (nmps) is also observed to have an influence on pavement performance. with the increase of nmps, the rutting resistance is observed to improve while mixes with smaller nmps, they found to exhibit better raveling resistance [7]. therefore it can be stated that the gradation of the mix should be given utmost importance and the method of grade selection should be rational to ensure satisfactory pavement performance. several methods were proposed by researchers for assessment of gradation requirements of layers in flexible pavements [8]. since it is not practically possible to sieve the aggregates in the field, the target gradation is attained by blending aggregate stockpiles from one or more quarry sites in a specific proportion called job mix formula. the methods proposed for blending includes graphical method, triangular chart method & asphalt institute graphical method and optimization methods [8]. graphical methods cannot be used when number of stockpiles is more than 3. as of today, there are no tools available for integrating different philosophies of gradation selection and hence there is a potential for research in this area. 2. objective of the research in view of the research gap identified, the objective of the present research is to develop an integrated software tool using c# programming for job mix formulation (jmf) of grade 2 of bituminous concrete mix wearing course. morth specifications, bailey’s parameters, 0.45 power gradation chart and cost per unit of the stock pile have been simultaneously considered to workout different combinations of aggregate blends. 3. problem formulation the blends are to be filtered as per three philosophies mentioned previously. the objective function is . 1 min c p cp m i j j j    (1) pp jmfupper i (2) pplowerjmfi  (3) 1 1 p m j j   (4) the jmf for a specified blend is worked out as proceedings of engineering and technology innovation, vol. 12, 2019, pp. 26-34 28 1 1 jmf p pp i to n m i j ij j     (5) ci unit material cost of the ith blend m number of designated stock piles in the mix pj proportion of aggregates of the jthstockpile cpj cost per unit of jth stockpile jmfi job mix formula for ith particle n number of designated particles in the mix ppij percent passing of the ith particle for jth stock pile obtained from sieve analysis pplower recommended lower percent passing ppupper recommended higher percent passing all particles in the mix should satisfy the condition specified in the eqs. (2) and (3) above for a successful trial blend. apart from the requirements of gradation, bailey’s method addresses the issues related to the ability of the mix to get compacted, segregation and continuity of gradation. the method is not very popular, due to the lack of extensive field verification of the pavement stretches designed using this method [13]. in this method, guidelines have been laid down through three parameters viz. coarse aggregate ratio (ca ratio), coarser portions of the fine aggregate ratio (fac) and finer portions of fine aggregate ratio (faf). these ratios will be calculated based on nmps of the mix, which is one sieve greater than the sieve on which at least 10% of the particles are retained as per trb circular e-c044 [13]. the three ratios specified are linked to half, primary, secondary and tertiary control sieves so as to distinguish the coarser and finer portions of the mix. 4. inputs for determination of jmf in order to find the proportion of aggregate blends for a specified type of layer and hence jmf for a stated nmps, it is required to collect samples from the stockpiles of quarry sites and perform the sieve analysis. the percent passing of each of the particle sizes is to be obtained in sieve analysis and will become input for determination of the jmf. the maximum density line is plotted using eq. (6). where size i size of ith particle in the mix sizemsa maximum size aggregate 5. need for a computational tool the spreadsheets are now being used to find a blend which satisfies the grading requirements of the specified layer fulfilling eqs. (2) and (3) for all the particles in the mix. this is a trial and error method with several proportions satisfying the grading requirements. in order to generate all those proportions which fulfilled the requirements of gradation and to address the issues of compaction, segregation and cost per unit of the blend, excel spreadsheet has its own limitations. to accomplish this, an interactive software interface is developed by using the c# (c-sharp) language. c# is a very useful tool which supports the use of forms in windows environment with graphic user interface (gui) as front end, the user can input data and it is possible to manipulate the data with suitable computations, manage and export data *100 0.45 pp size i i size msa          (6) proceedings of engineering and technology innovation, vol. 12, 2019, pp. 26-34 29 files in the back end using object oriented programming (oop) concept. thus c# programming has been chosen to develop the interactive gui software for the assessment of jmf in the current research. the software requires microsoft net 4.5 framework or higher version for execution. 6. need for a computational tool in order to select a suitable blend, the following criteria have been adopted and embedded in the software. (i) the proportions generated by the software will be examined for the morth requirement. (ii) bailey parameters will be evaluated for the percent passing of each blend obtained. the parameters will be worked out as per the following formulae. where ca ratio coarse aggregate ratio pp half sieve percent passing half sieve pcs primary control sieve = 0.22 x nmps scs secondary control sieve = 0.22 x pcs tcs tertiary control sieve = 0.22 x scs pp pcs percent passing primary control sieve pp scs percent passing secondary control sieve pp tcs percent passing tertiary control sieve the factor 0.22 had arrived after considering the analysis of 2-d and 3-d packing of different shaped particles. 2-d analysis has shown the particle diameter ratio equal to 0.155 (all round) and 0.289 (all flat) with an average of 0.22 for angular and subangular particles [12]. these parameters will be computed from percent passing of each blend generated in the process. (iii) for each of these proportions generated by the software, the level of gradation closer to the maximum density line is evaluated by the statistical parameter root mean square error (rmse) [5]. where ppmax density percent passing corresponding to the i st rticle for the maximum density line in the mix pp jmf percent passing corresponding to the i st rticle in the trial blend i index for the particle in the mix n total number of particle sizes in the mix the blend of aggregates which fulfill the requirements of morth and bailey with the least of rmse and with the least cost per unit of the blend will be recommended for final implementation. in the present research work, the nominal maximum particle size was chosen as 12.5 mm. the ca ratio between 0.50 and 0.65 and other two ratios in the range 0.35 to 0.50 have been chosen, as recommended in bailey’s specifications. 100 pp pp pp ca ratio pcshalf sieve half sieve                (7) pp fa ratio pp scs c pcs  (8) pp fa ratio pp tcs f scs  (9) ( ) 2max. 1 pp pp rmse n n ijmfdensity i     (10) proceedings of engineering and technology innovation, vol. 12, 2019, pp. 26-34 30 (1) ca ratio: the three ratios discussed above are chosen based on nmps. it was observed that the mixes with lower values than recommended ca ratio, and are subjected to segregation and are generally gap-graded mixes, while the mixes with higher values of ca ratio will have issues related to continuity of gradation and difficulty in compaction. (2) fac ratio: while the mixes with lower values of fac ratios will have non-uniform gradation and will have problems of compaction, the higher values will indicate the presence of excessive amounts of fine aggregate leading to tender mixes. the mixes with higher values of fac ratio will show a hump in the 4.75 mm region and below when plotted on a 0.45 power gradation chart. (3) faf ratio: this ratio explains how fine aggregates pack together and influence the voids in the mix. typically, dense graded mixes will have faf less than 0.50. also, vma increases with the decrease in this ratio. 7. need for a computational tool the flow chart for the process incorporating the above said criteria is presented in fig. 1. fig. 1 the process flow chart 7.1. sample source code the sample source code is presented below for the control ‘find jmf’ button. private void button3_click_1(object sender, eventargs e) { //find jmf button string str1, str2, str3, str4, str5; for (j = 0; j < 10; j++) { proceedings of engineering and technology innovation, vol. 12, 2019, pp. 26-34 31 str1 = datagridview2.rows[j].cells[1].value.tostring(); ppsize1[j] = double.parse(str1); } for (j = 0; j < 10; j++) { str2 = datagridview2.rows[j].cells[2].value.tostring(); ppsize2[j] = double.parse(str2); } for (j = 0; j < 10; j++) { str3 = datagridview2.rows[j].cells[3].value.tostring(); ppsize3[j] = double.parse(str3); if (noofcoarsepiles == 2) ppsize3[j] = 0; } for (j = 0; j < 10; j++) { str3 = datagridview2.rows[j].cells[4].value.tostring(); ppsize4[j] = double.parse(str3); } for (j = 0; j < 10; j++) { str5 = datagridview2.rows[j].cells[5].value.tostring(); mf[j] = double.parse(str5); } //location of maximum density line for (i = 0; i < 10; i++) { strtemp = datagridview2.rows[i].cells[0].value.tostring(); size[i] = double.parse(strtemp); ssr[i] = math.pow(size[i], 0.45); if (i > 0) maxden[i] = 100 * math.pow(size[i] / size[0], 0.45); else maxden[0] = 100; } int looper = 0, morth = 0; // three stockpiles in coarse aggregates if (noofcoarsepiles == 3) {//outermost loop for (s1 = 1; s1 <= 95; s1++) {//1 tester = 0; for (s2 = 1; s2 <= 95; s2++) {//2 if (98 s1 s2 > 0) {//3 s34 = 98 s1 s2; proceedings of engineering and technology innovation, vol. 12, 2019, pp. 26-34 32 {//4 for (s3 = 1; s3 <= 95; s3++) {//5 bandmchecker = 0; s4 = s34 s3; if (s4 > 0) {//6 if ((s1 + s2 + s3 + s4 + s5) == 100) {//7 tester = 0; newrmssum = 0; jmfsum = 0; bandmchecker = 0; looper = looper + 1; for (i = 0; i < 10; i++) { jmf[i] = 0.01 * (s1 * ppsize1[i] + s2 * ppsize2[i] + s3 * ppsize3[i] + s4 * ppsize4[i] + 2 * mf[i]); jmfsum = jmfsum + jmf[i]; if (jmf[i]<=upper[i]&&jmf[i]>= lower[i]) tester = tester + 1; newrmssum = newrmssum + (jmf[i] maxden[i]) * (jmf[i] maxden[i]); } 8. results of software for local quarry stockpiles sample sieve analysis was performed for all the four aggregate stockpiles, procured from a local quarry. the results of sieve analysis are considered for generating the alternative aggregate blends through the software developed during the present study. the user can edit the data of percent passing and generate the blends. fig. 2 results of the software showing generated blends the results of c# program are presented in fig. 2 for two aggregate stockpiles. in the initial trial, the gradation of dust could generate blends fulfilling only morth gradation. subsequently, change of gradation of dust could give blends to fulfill both morth and bailey criteria; therefore it can be observed that gradation of dust will influence jmf. out of 4656 combinations of possible blends generated by the software, 36 blends had fulfilled morth requirements (as displayed in yellow colour) and 26 blends fulfilled bailey’s requirements. the minimum and maximum rmse values were observed to be proceedings of engineering and technology innovation, vol. 12, 2019, pp. 26-34 33 7.52 and 9.53 respectively. the dependency of bailey’s parameters on aggregate proportions is presented in fig. 3. no significant observations could be made with regard to the variation of other stockpiles with bailey parameters and hence not discussed. the successful jmf of the 3 blends with minimum dust and maximum ca # 1 & ca # 2, maximum dust with minimum ca #1 and maximum dust with minimum ca # 2 fulfilled both morth & bailey parameters with rmse values, 8.68, 10.54 and 10.45 respectively have been chosen from the results and are plotted and presented in fig.4. also, superpave control points were also marked on the plot for verification. even though the restricted zone has lost its significance [4] in superpave gradation, out of research interest, it was also plotted in fig. 3. it can be clearly observed that all the 3 blends with designated % feed have fulfilled superpave control points, while the blend with minimum dust, maximum ca #1 and ca #2 is observed to be the closest to the maximum density line, with least rmse of 8.68 and also with the lowest cost per unit and hence recommended for final implementation. fig. 3 variation of ca # 1 with ca ratio for dust 6 fig. 4 aggregate gradation for selected blends with % feed along with superpave control points fulfilling morth & bailey criteria 9. conclusions this research has presented the development of software for the assessment of jmf for bc grade ii. excel spreadsheets are been widely used to find the desired blend limiting with few successful blends. the software is developed by the c# programming for arriving at the jmf had to fulfill morth & bailey requirements. it is capable of generating blends fulfilling morth and bailey parameters; and will be able to assess the nearness to maximum theoretical density line along with the cost per unit of the blend. a sample source code was also presented. the results of a case study were also discussed wherein the gradation of dust obtained from a local quarry and was observed to fulfill morth requirements with 36 combinations of aggregate blends and 26 combinations to fulfill bailey requirements. the gradation of dust plays a vital role in designing the jmf. from the study, it can also be concluded that the bailey requirements are more stringent than morth and the superpave gradation. the tool developed with c# programming in the .net framework will facilitate the designers to arrive at most optimal blend and will serve in the initial phase of the mix design process fulfilling recommendations of local agencies, bailey parameters and superpave. conflicts of interest the authors declare no conflict of interest. proceedings of engineering and technology innovation, vol. 12, 2019, pp. 26-34 34 references [1] y. h. haung, pavement analysis and design, 2nd ed, pearson education, usa. [2] active standard, “standard test method for marshall stability and flow of asphalt mixtures,” west conshohocken, pa, 2015. [3] t. n. mansour, bradley j. putman, “influence of aggregate gradation on the performance properties of porous asphalt mixtures,” journal of materials civil engineering, vol. 25, no.2, pp. 281-288, 2013. [4] cooley jr, l a. zhang, jianping, kandhal, p. s. hand, a. j. epps martin a, “significance of restricted zone in superpave aggregate gradation specification,” transportation research circular no. e-c043, transportation research board, national research council, washington, d.c., 2001. [5] r. j. hyndman, a. b. koehler, “another look at measures of forecast accuracy,”. international journal of forecasting, vol. 22, no. 4, pp. 679-688, 2006. [6] b. e. ruth et al, “aggregate gradation characterization factors and their relationships to fracture energy and failure strain of asphalt mixtures,” journal of association of asphalt paving technologies. vol. 71, pp. 310-344, 2002. [7] h. liu, p. hao and j. xu, “effects of nominal maximum aggregate size on the performance of stone matrix asphalt,” applied sciences, vol. 7, no. 126, 2017. [8] mario t. tabiucanon, p. adulbhan, stephen and s. y. chen, “a probabilistic programming model for blending aggregates,” applied mathematical modelling, vol. 4, pp. 257-260, 1979. [9] the asphalt institute, “mix design methods for asphalt concrete and other hot-mix types,” ms-2 (4th edn), maryland, 1974. [10] w. r. vavrik, “bailey method for gradation selection in hot-mix asphalt design transportation research circular,” trb circular number e-c044, 2002. [11] specifications for base and surface course, section 500, materials for structures, morth, government of india, 2015. [12] w. r. vavrik, w. j. pine, g. huber, s. h. carpenter and r. bailey, “the bailey method of gradation evaluation: the influence of aggregate gradation and packing characteristics on voids in the mineral aggregate,” journal of the association of asphalt paving technologists, vol. 70, pp. 132-175, 2001. [13] w. r. vavrik, w. j. pine and s. h. carpenter, “aggregate blending for asphalt mix design: the bailey method,” presented at the 81st annual meeting of the transportation research board, washington, d.c., vol. 1789, pp. 146-153, 2002. [14] w. r. vavrik, “quality control of hot-mix asphalt: controlling aggregate interlock through the bailey method,” in proceedings of the 1st international conference on quality control and quality assurance of construction materials, dubai municipality, united arab emirates, 2001. 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.sciencedirect.com/science/article/pii/0307904x80901924#! https://www.sciencedirect.com/science/article/pii/0307904x80901924#! 4___peti#8870___27-33 proceedings of engineering and technology innovation, vol. 21, 2022, pp. 27-33 development of non-contact real-time monitoring system for animal body temperature kuo-hsiung tseng 1,* , meng-yun chung 1 , wei-hong jhou 1 , wei-jer chang 2 , chen-han xie 2 1 department of electrical engineering, national taipei university of technology, taiwan 2 morepets company limited by shares, taiwan received 09 november 2021; received in revised form 13 february 2022; accepted 14 february 2022 doi: https://doi.org/10.46604/peti.2022.8870 abstract body temperature is an important indicator of health monitoring. however, since animals are covered with fur, it is difficult to obtain their accurate body temperature with the traditional infrared measurement technology. to deal with this problem, this research proposes a non-contact real-time monitoring system using an infrared method combined with object detection. the system is developed based on general infrared thermal imaging technology and an infrared thermal imaging module with an image tracking algorithm. yolo is used to detect animals, and a thermal imaging camera is used to measure the body surface temperature of animals. the result shows that the proposed system can accurately measure the body temperature of animals without being influenced by animals’ fur. in the future, it can be applied to monitor the body temperature of sick animals in veterinary hospitals. keywords: animal body temperature sensing, internet of things, image detection, infrared thermography 1. introduction infrared thermometers were developed in the mid-1980s [1-2]. with pyroelectric sensors, infrared thermometers can effectively detect the heat radiation emitted from a target surface and measure the temperature of the surface in a non-contact way [3-4]. infrared thermometers have been extensively used for body temperature measurement, but their limitation is that they can only conduct close-distance measurement with a small detection range. infrared thermography is a technique combined with image processing to generate visual images with different colors indicating different temperatures on different parts of the target surface [5-7]. this technique is characterized by remote observation, timeliness, and heat distribution observation. notably, it has been used in animal medical diagnosis for decades, e.g., the diagnosis for mammals. mammals generate heat energy by metabolism to maintain constant body temperature [8]. as their skin surface temperature can reflect the condition of tissue metabolism and blood circulation, the lesion of surface inflammation or circulation abnormality can be found when the body temperature is observed to be abnormal. thus, body temperature is an important index for diagnosing diseases. many studies attempted to use infrared thermometers to measure the body temperature of animals, including cats, pigs, and cattle [9]. however, due to animals’ fur and physiological structures, the probability of error is quite high when using infrared thermometers, making it difficult to replace conventional thermometers with infrared thermometers. furthermore, when using conventional thermometers in veterinary clinics, veterinarians have to insert the thermometers into animals’ anuses to measure the core body temperature, making the animals suffer from severe stress and discomfort and even resulting in injuries of the animals or the veterinarians. moreover, it is impractical to use conventional thermometers for repeated measurement, especially for hospitalized animals that require constant monitoring to check whether there is fever or hypothermia. hence, there is an urgency to develop a non-contact continuous animal temperature monitoring system to meet the existing clinical need [10]. * corresponding author. e-mail address: f10473@mail.ntut.edu.tw tel.: +886-2-27712171#2173; fax: +886-2-27317187 advances in technology innovation, vol. 21, 2022, pp. 27-33 in addition to the study need and clinical need, there is also a commercial need that motivates the researchers to develop a non-contact monitoring system for measuring the body temperature of animals. livestock’s commercial value is highly dependent on the health conditions of livestock. however, there is still no effective approach to monitor the health conditions of livestock because of the limitations in fields, manpower, and equipment. infrared thermography is a possible solution that can monitor animals without direct contact and simultaneously measure the temperature distribution of multiple targets in the field of view [11]. infrared thermography is free from the effect of ambient light source conditions and thus is highly suitable for animal temperature monitoring, aiding the early detection of infectious diseases. this research uses the infrared method combined with object detection to develop a non-contact real-time monitoring system to accurately measure the body temperature of animals, without being influenced by animals’ fur. in the rest of this article, the research method and measurement equipment used are explained in section 2, the results and discussion are presented in section 3, and the concluding remarks are provided in section 4. 2. materials and research method 2.1. experimental construction the experimental process of this study is shown in fig. 1. first, a field (e.g., veterinary hospitals, ranches, or farms) is selected as the site where the experiment is conducted. then, the type of animal subjects is selected, and the non-contact and contact methods are used to measure the body temperature of animal subjects. after that, the obtained temperature data are compared and sent to a veterinarian and the owners of animal subjects. in this study, a veterinary hospital is chosen as the experimental field, and an adult pet dog is selected as the animal subject. this study performs target detection for image recognition in the early stage [12]. a visible-light/infrared twin-lens module is used for detection, and is combined with an image tracking algorithm for animal temperature monitoring with target tracking and positioning functions. two groups are set up in this study: a control group (the traditional body temperature measurement conducted by a veterinarian) and an experimental group (the non-contact infrared measurement using the proposed system). an adult pet dog is used for both groups. in the early setup, the compensation for fur interference cannot be regulated accurately, and the module’s precision is adjusted by comparing the data of contact measurement. it must be noted that, unlike human beings, animals cannot tell their conditions. thus, in this study, the dog subject’ body temperature is initially measured to examine and determine fever or other symptoms. subsequently, the measured body temperature is transferred via the internet of things (iot) to the computer of the verification domain as the criterion of diagnosis and treatment for the veterinarian. in addition, as the owner of the dog subject is concerned about the status of the dog, instant information is also sent via iot to the owner’s mobile phone for first-hand updates on the dog’s condition at the hospital [13]. a temperature and humidity sensor is used to detect the dog subject’s internal temperature and humidity, determining whether the subject’s condition is influenced by external factors. fig. 1 experimental process field selection pet selection non-contact temperature measurement contact temperature measurement data comparison send message veterinarian owner pet´s state 28 proceedings of engineering and technology innovation, vol. 21, 2022, pp. 27-33 an flir c5 thermal imaging camera is used in this study to measure the body temperature of the dog, as shown in fig. 2. the specification of the camera is shown in table 1. the camera has a size of 135 × 85 × 25 mm and is only 0.19 kg, which is suitable for conducting the measurement. it can store more than 5000 images by its internal memory and onboard flir ignite cloud connectivity with wi-fi. the infrared sensor provides 160 × 120 (19,200 pixels) images, and can be used under the ambient temperature of -10 to 50°c to measure the subjects having the temperature of -20 to 400°c. the flir camera is a long wave infrared (lwir) camera and receives infrared radiation with a wavelength of 8-14 µ m. under the atmospheric window, water vapor and carbon dioxide have weak absorption of the radiation. this facilitates the infrared body temperature measurement. with the ambient temperature of 15 to 35°c, the accuracy of measuring the subjects with 0 to 100°c is ±3°c, and the accuracy of measuring the subjects with 100 to 400°c is ±3%. parameters such as ambient temperature, ambient humidity, and the emissivity of the subjects can be adjusted. the lightness and portability of the camera make it easy to set up experiments. before each experiment, the environmental parameters (temperature and humidity) of the camera are adjusted according to the indoor temperature and humidity count value. the emissivity of the animal fur can also be set in the range of 0.98-1. fig. 2 flir c5 infrared thermal imaging camera table 1 specification of an flir c5 thermal imaging camera model c5 camera size (l × w × h) 138 × 84 × 24 mm infrared ray (ir) sensor 160 × 120 (19,200 pixels) operating temperature range -10 to 50°c (14 to 122°f) object temperature range -20 to 400°c (-4 to 752°f) spectral range 8 to 14 µ m accuracy at ambient temp. 15 to 35°c (59 to 95°f): 0 to 100°c (32 to 212°f): ±3°c (±5.5°f) 100 to 400°c (212 to 752°f): ±3% company flir country estonia 2.2. infrared radiation in spectra, the wavelength exceeding 0.75 µm is called infrared ray (ir). from the wavelength 0.75 µm to the longest wave band, ir can extend to 1000 µm. the wavelength range covered in this section can be divided into three according to the application domains: 0.75-2.5 µm (near-ir), 2.5-25 µm (mid-ir), and above 25 µm (far-ir). 2.2.1. planck’s law planck’s law, also known as planck’s black body radiation law, is a mathematical relationship proposed by a german physicist max planck in 1990. it refers to the relationship between the radiance and frequency of electromagnetic radiation emitted from a black body at an arbitrary temperature t [14]. 2.2.2. wien’s displacement law wien’s law, also known as wien’s displacement law, was proposed by wilhelm wien. it is a law describing the inverse relationship between the peak wavelength of spectral radiosity of a black body’s electromagnetic radiation and its temperature in physics [15]. the relationship is expressed as eq. (1). wien’s displacement law also describes that the hotter an object is, the shorter the radiation wavelength or the higher the radiation frequency is. 29 advances in technology innovation, vol. 21, 2022, pp. 27-33 max b λ t = (1) where λmax is the peak wavelength of radiation (unit: m), and t is the absolute temperature of the black body (unit: k). b is the proportional constant, called wien’s displacement constant; the value is 2.897 7729 × 10 –3 m· k. 2.2.3. stefan-boltzmann law the stefan-boltzmann law, also known as stefan’s law, means that the total energy radiated from the surface of a black body per unit area in unit time (radiosity or energy flux density of object) j* is proportional to the fourth power of thermodynamic temperature t (absolute temperature) of the black body [16]. the relationship is expressed in eq. (2). * 4 j σt= ε (2) where the radiosity j* has the dimension of power density given in eq. (3). according to the international system of units, the standard unit of the radiosity j* is j/(s × m 2 ), i.e., w/m 2 . 2 energy time distance× (3) the standard unit of absolute temperature t is k. ϵ is the radiation coefficient of the black body; if the black body is an absolute black body, then ϵ = 1. the scale factor σ is called stefan-boltzmann’s constant or stefan’s constant. given that it can be worked out of the known basic physical constants of the nature, it is not considered a basic physical constant. the value of the constant is expressed as eq. (4). 5 4 8 1 2 4 2 3 2π k σ 5.670400 10 js m k 15c h − − − −= = × (4) where k is the value of boltzmann’s constant 1.38064852 × 10 -23 j/k, h is the planck’s constant 6.62607015 × 10 −34 (j· s), and c is the light velocity in a vacuum with the value of 3.00 ×10 8 m/s. 2.3. object detection the traditional thermal imaging equipment mostly performs data acquisition for a single object. thus, the information analysis and processing are local and have low volume computation. to the extent of the information of the whole space, the obtained thermal image information will expand greatly, and the demand for computing resources will grow sharply, becoming the system load [17]. without an intelligent zed improvement method, the system will fail. therefore, this project uses the visible-light/infrared twin-lens module and visible light features for recognizing image objects. meanwhile, target locking is established automatically to reduce the range of image information analysis to improve the excessive operation [18]. this study uses yolo object detection technology, where the image input to the prediction result of yolo only depends on one convolutional neural network (cnn). multiple bounding boxes are predicted simultaneously by using cnn, and the probability of an object is calculated for each box. the object is recognized in the local region compared to the sliding window and region proposals [19]. yolo aims at the full image in the training and operational phases, paving the way for a better detection effect on the background. the background error detection rate of yolo is only half of that of a fast region-based convolutional neural network (fast r-cnn). further, yolov4 has better recognition speed and accuracy and less hardware requirement than yolov3 [20]. this study employs the jetson nano developer edition for development. this edition is characterized by powerful computation capability, lightness, easy disposition, and applicability to mobile equipment. 30 proceedings of engineering and technology innovation, vol. 21, 2022, pp. 27-33 3. results and discussion this study employs one adult dog as the animal subject for non-contact and contact temperature measurements (the contact one is the control group and the non-contact one is the experimental group). for the control group, the body temperature of the adult dog is measured using the rectal thermometer. as can be seen in table 2, the dog’s body temperature got from the rectal thermometer is higher than the normal human body temperature which is in the range of 36-37°c. for the experimental group, the flir c5 thermal imaging camera is used. the eye and abdomen temperatures of the dog are measured at a fixed time in the morning from january 26 2021 to february 02 2021. as shown in figs. 3-5, the square frame is the measuring area of a thermal imager, and the white point is the center point of the measuring area. the temperatures at the red and blue points are the maximum and minimum temperatures in the measuring area, respectively. in fig. 3, it is observed that the density of fur distribution can influence the temperatures measured. the red point is the area with sparse fur and the blue point is the area with fur. the maximum temperature measured in the abdomen area occurs in the region covered with sparse fur. meanwhile, in fig. 4, the region having the maximum temperature measured on the face is the eye. the eye temperature is different from the abdomen temperature. it can be seen that the difference in fur density affects the value of temperature, as shown in fig. 5. this is the reason why this study applies the infrared temperature measurement to animals. in addition to the material emissivity, the fur density also causes measurement difficulties and errors. as shown in fig. 6, it is observed that the eye temperature is approximate to the actual body temperature, and the side temperature is a little lower than the actual body temperature. table 2 result of the contact temperature measurement date temperature (°c) 2021/1/26 39.4 2021/1/27 39.0 2021/1/28 38.6 2021/1/29 38.7 2021/1/30 38.7 2021/1/31 38.6 2021/2/1 38.7 2021/2/2 39.0 fig. 3 flank temperature fig. 4 eye temperature fig. 5 effect of fur density 24 26 28 30 32 34 36 38 2021/1/26 2021/1/27 2021/1/28 2021/1/29 2021/1/30 2021/1/31 2021/2/1 2021/2/2 t e m p e ra tu re (˚ c ) date eye temperature(°c) abdomen temperature(°c) fig. 6 result of the non-contact temperature measurement 31 advances in technology innovation, vol. 21, 2022, pp. 27-33 4. conclusions in asian countries (e.g., taiwan), the number of dogs and cats could reach 3 million in 2020, exceeding the number of children below the age of 15. with the increasing number of dogs and cats, the number of animal hospitals has also been increased to more than 1,700. to ensure that the aging pets receive adequate medical care, the pet owners pay close attention to their pets’ medical service quality. since the intelligentized monitoring systems can save the labor cost for animal hospistals and thus reduce the medical payments for pet owners, admitting such intelligentized monitoring systems will be the future trend. in clinical scenarios, the body temperature of hopitalized animals need to be continuously monitored by veterinarians. however, when veterinarians employ traditional thermometers to measure the temperature through animals’ anuses, the animals are severely discomforted and the veterinarians may be injured if the animals defend themselves. to deal with this problem, a non-contact continuous animal temperature monitoring system is desired. this study combines non-contact infrared measurement with object detection, and uses an optical lens to conduct the measurement. specific regions can be extracted by the yolo algorithm to reduce excessive operation information. the temperatures of the subjects with specific positions can be captured by infrared thermal imaging technology and instantly conveyed to veterinarians. with the proposed system, the risk and inconvenience induced by contact measurement can be avoided. acknowledgements the authors would like to thank the ministry of science and technology (most 109-2622-e-027-033-) for the financial support of this research. also, the authors would like to thank mr. jue-hao chen (department of electrical engineering, national taipei university of technology, taiwan) and yuan lin animal hospital for their kind assistance. conflicts of interest the authors declare that there are no conflicts of interest regarding the publication of this study. statement of ethical approval all procedures performed in studies involving animals were in accordance with the ethical standards of the institution or practice at which the studies were conducted. references [1] j. l. hatfield, “measuring plant stress with an infrared thermometer,” hortscience, vol. 25, no. 12, pp. 1535-1538, december 1990. [2] e. j. sadler, c. r. camp, d. e. evans, and j. a. millen, “corn canopy temperatures measured with a moving infrared thermometer array,” transactions of the american society of agricultural engineers, vol. 45, no. 3, pp. 581-591, 2002. [3] t. matsukawa, m. ozaki, t. nishiyama, m. imamura, and t. kumazawa, “comparison of infrared thermometer with thermocouple for monitoring skin temperature,” critical care medicine, vol. 28, no. 2, pp. 532-536, february 2000. [4] s. abbasbandy, “homotopy analysis method for heat radiation equations,” international communications in heat and mass transfer, vol. 34, no. 3, pp. 380-387, march 2007. [5] c. a. balaras and a. a. argiriou, “infrared thermography for building diagnostics,” energy and buildings, vol. 34, no. 2, pp. 171-183, february 2002. [6] b. b. lahiri, s. bagavathiappan, t. jayakumar, and j. philip, “medical applications of infrared thermography: a review,” infrared physics and technology, vol. 55, no. 4, pp. 221-235, july 2012. [7] s. bagavathiappan, b. b. lahiri, t. saravanan, j. philip, and t. jayakumar, “infrared thermography for condition monitoring—a review,” infrared physics and technology, vol. 60, pp. 35-55, september 2013. [8] l. s. peck, m. s. clark, s. a. morley, a. massey, and h. rossetti, “animal temperature limits and ecological relevance: effects of size, activity, and rates of change,” functional ecology, vol. 23, no. 2, pp. 248-256, march 2009. 32 proceedings of engineering and technology innovation, vol. 21, 2022, pp. 27-33 [9] d. m. newsom, g. l. bolgos, l. colby, and j. a. nemzek, “comparison of body surface temperature measurement and conventional methods for measuring temperature in the mouse,” journal of the american association for laboratory animal science, vol. 43, no. 5, pp. 13-18, september 2004. [10] g. j. tattersall and v. cadena, “insights into animal temperature adaptations revealed through thermal imaging,” the imaging science journal, vol. 58, no. 5, pp. 261-268, july 2010. [11] p. a. warn, m. w. brampton, a. sharp, g. morrissey, n. steel, d. w. denning, et al., “infrared body temperature measurement of mice as an early predictor of death in experimental fungal infections,” laboratory animals, vol. 37, no. 2, pp. 126-131, april 2003. [12] n. a. othman and i. aydin, “a new deep learning application based on movidius ncs for embedded object detection and recognition,” 2nd international symposium on multidisciplinary studies and innovative technologies, pp. 1-5, october 2018. [13] t. h. chung, w. s. jung, e. h. nam, j. h. kim, s. h. park, and c. y. hwang, “comparison of rectal and infrared thermometry for obtaining body temperature of gnotobiotic piglets in conventional portable germ free facility,” asian-australasian journal of animal sciences, vol. 23, no. 10, pp. 1364-1368, october 2010. [14] j. c. cuevas and f. j. garcía-vidal, “radiative heat transfer,” acs photonics, vol. 5, no. 10, pp. 3896-3915, september 2018. [15] j. k. baird and t. r. king, “a wien displacement law for impact radiation,” international journal of impact engineering, vol. 23, no. 1, pp. 39-49, december 1999. [16] w. shi, j. cao, q. zhang, y. li, and l. xu, “edge computing: vision and challenges,” ieee internet of things journal, vol. 3, no. 5, pp. 637-646, october 2016. [17] g. montambaux, “generalized stefan-boltzmann law,” foundations of physics, vol. 48, no. 4, pp. 395-410, march 2018. [18] q. zhu, h. zheng, y. wang, y. cao, and s. guo, “study on the evaluation method of sound phase cloud maps based on an improved yolov4 algorithm,” sensors, vol. 20, no. 15, article no. 4314, august 2020. [19] j. m. zurada, introduction to artificial neural systems, saint paul: west publishing, 1992. [20] a. bochkovskiy, c. y. wang, and h. y. m. liao, “yolov4: optimal speed and accuracy of object detection,” https://arxiv.org/pdf/2004.10934v1.pdf, april 23, 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/). 33 6___peti#8695___50-59 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 50-59 effect of gear materials on the surface contact strength of spur gears eniyew eskezia * , misganaw abebaw department of mechanical engineering, institute of technology, university of gondar, gondar, ethiopia received 14 october 2021; received in revised form 21 january 2022; accepted 22 january 2022 doi: https://doi.org/10.46604/peti.2022.8695 abstract in gear applications, the initiation of cracks at or near the contact surfaces of gear mates occurs mostly due to the surface contact mode of gear failure. this study aims to investigate the influence of gear materials on the contact strength of spur gears. in this study, four different gear materials are selected and the contact stress on spur gear mates is analyzed. hertz’s contact stress equation and ansys 16 are used for the theoretical analysis and finite element (fe) method, respectively. the results of hertz’s equation are compared with the results of ansys 16. the results show that the contact stress on spur gear mates varies when different gear materials are used in both methods. this indicates that the surface contact strength of spur gears is greatly influenced by the type of gear materials. keywords: spur gear, contact stress, analytical method, hertz’s equation, finite element method 1. introduction gears are the basic machine elements in a mechanical power transmission system and most industrial rotating machines. humans have used them for different applications for several years. however, some problems still exist. for instance, gear tooth surface damage, which is caused by wear and surface fatigue, can result in the change of material properties in different working conditions [1-3]. scholars have investigated the solution to the problems by using various methods. among them, theoretical and numerical methods have been used frequently due to low research expenses. contact stress is the main cause of gear failure. it initiates the crack propagation around the contact surfaces of gear mates, and stress concentration may develop based on it [2, 4-5]. to increase gears’ load-carrying capacity and component reliability, it is important to improve the surface strength of gears. surface strength is the factor that determines the gear life rather than bending strength due to the large enough teeth profile of gears. different techniques were adopted by researchers to increase the working ability, efficiency, and working life of gears. furthermore, the surface life of gears can be improved by subjecting the gear materials to different surface treatment processes. gear parameters also influence the gear performance greatly [3-6]. mechanical properties of a material are the properties that encompass a response to an applied load. they reveal the elastic and plastic behavior of the material upon applied forces. these properties are used to classify and identify metals. strength, ductility, hardness, toughness, impact strength, etc. are some mechanical properties of metals. strength is the ability of metals to withstand deformation due to external loads, whereas ductility is the tendency of metals to be drawn into wire before fracturing under tensile loads [4-5]. the main purpose of this study is to investigate the influence of gear materials on the contact strength of spur gears by using the theoretical method (hertz’s contact stress equation) and finite element (fe) method (ansys workbench). both gear designers and manufacturers could take advantages of the findings in this study. * corresponding author. e-mail address: enyew.eskezia@uog.edu.et tel.: +251-9-12802371; fax: +251-058-1141240 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 50-59 2. materials, methods, and conditions the contact stress on spur gears is the main concern in the research field. the analytical, experimental, and fe methods are the basic tools to compute and predict the contact stress on gears. any method out of the three can support the other one for validating the results [3]. 2.1. analytical methods and conditions gear failure is categorized into two general groups: bending and contact failure. bending failure occurs in the root of gear teeth, whereas contact failure occurs on the gear surface when the gear strength is inadequate. the bending stress can be calculated by the lewis formula, whereas the hertz’s equation is used for the contact stress. the design procedures developed by american gear manufacturing association (agma) and international organization for standard (iso) were also used by the researchers to design spur gears for different applications [5-8]. in machine elements, which work under pressure loading and have rolling-sliding motion, the value of contact stress depends on the size of contact areas. the contact stress can be calculated based on elastic theory. the damage of gear teeth due to the contact stress commonly occurs either on the pitch line (only rolling movement) or away from the pitch line (both sliding and rolling movements) [9]. 2.1.1. contact problem modeling using parallel cylinders the contact problem modeling is created by using parallel cylinders in order to apply the hertz’s equation to spur gear mates. the mating gears are considered equivalent to the contact cylinders with the same radius of curvature at contact point [9]. fig. 1 below illustrates the contact modeling. as shown in fig. 2 below, the shape of the contact area of the two cylinders is rectangular (2b × l). however, the stress is distributed along the width (2b) with an elliptical shape. the maximum pressure on this contact area occurs along the midway of the width (b). the value of this maximum pressure is calculated by eq. (1). (a) cylindrical contact in compression [10] (b) contacted spur gear teeth [11] fig. 1 contact problem modeling using parallel cylinders fig. 2 ellipsoidal-prism pressure distribution [9] 51 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 50-59 from the above figure, the maximum pressure, pmax, can be formulated as: 2 max p f bl= π (1) where f is the total contact force. half of the contact width of the pressure distribution can be obtained by: 4 e e fr b le = π (2) where ee is the equivalent modulus of gear material and re is the equivalent radius of curvature. 1 2 1 1 1 e e e e = + (3) 1 2 1 1 1 e r r r = + (4) at pitch point, r1 = rp and r2 = rg. this is given by: sin 2 p p r d= α (5) sin 2 g g r d= α (6) where d refers to pitch diameter, p refers to pinion, g refers to gear, and α refers to pressure angle. the equivalent radius will be: sin 1 1 1 1 2 1 1 ( ) e p g p g r r r d d = = + + α (7) the tangential and normal force components (ft and fn) and the torque (tp) can be related to the pinion and gear. cos cos pt n p tf f f r = = = α α (8) substituting eqs. (2), (7), and (8) into eq. (1), the maximum contact pressure or hertz stress can be obtained. cos p e max e p t e p lr r = απ (9) in the gear contact kinematics, there are three steps as identified in fig. 3. in the beginning, the contact occurs through a combination of rolling and sliding movements between the teeth. then, the sliding movement has the frictional effect, and there is simple rolling contact in the pitch diameter region. after this point, both sliding and rolling movements will occur again [12]. in this study, the authors have considered some assumptions. the materials in contact are considered homogeneous. the load which is normal to the contact tangent plane causes the contact stress. in other words, there are no tangential forces between the surfaces as the contact area is considered very small. thus, the frictional effect is totally neglected. 52 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 50-59 fig. 3 mechanics of gear tooth contact [12] 2.1.2. material requirements the crucial aspect to consider in gear engineering and design is the selection of gear materials. strength, durability, and cost (both material and manufacturing cost) are the most important factors when selecting the gear materials. the purpose of material selection is to find the right combination of physical properties that satisfy the requirements of certain applications at the lowest cost. depending on the shape, nature, and final application of gears, designers and manufacturers choose one from an unlimited number of gear materials to create the finished product. the most commonly used materials include many types of steel, brass, bronze, cast iron, ductile iron, aluminum alloys, powder metals, and plastics [13-14]. table 1 below displays the mechanical properties of four engineering gear materials selected for this study. table 1 properties of gear materials selected for this study selected gear materials modulus of elasticity (gpa) poisson’s ratio density (kg/m 3 ) case-hardened wrought steel 206 0.3 7800 astm class 35 cast iron 114 0.29 7150 gray cast iron (gg-30) 91 0.25 7050 beryllium-aluminum alloy (albemet 162) 179 0.165 2070 2.1.3. geometrical design of spur gear mesh fig. 4 below shows the standard meshing spur gears. the standard meshing spur gears are the reference circles of two gears that contact and roll each other. based on some set values, all the geometrical specifications of the gears are calculated [15]. in this study, the gear and pinion are considered to have identical geometrical parameters which are presented in table 2. the input parameters of the mating gear pair (gear train) are specified in table 3. fig. 4 geometry of standard meshing spur gears [15] table 2 specification of the meshing spur gears no. description of gear geometry symbol formula value 1 module m set value 2 mm 2 number of teeth z 30 3 pressure angle α 20° 4 face width b 12 mm 5 pitch circle diameter d � � � � � 60 mm 6 base circle diameter db �� � � � �� 56.38 mm (a) point of first contact (b) pitch point contact (c) point of last contact a direction of roll direction of frictional forces 53 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 50-59 table 2 specification of the meshing spur gears (continued) no. description of gear geometry symbol formula value 7 outside circle diameter do �� � � � (� + 2) 64 mm 8 addendum ha ℎ� � � 2 mm 9 center distance a � � 1 2 (� + �) � � 60 mm 10 pitch co-efficient re �� � 1 2 �� ( 1 �� + 1 �� ) 5.13 table 3 input parameters for the gear train 2.1.4. model calculation of contact stress analysis by hertz’s equation now, let’s take grey cast iron (gg-30) material for both gears. by using eq. (10), the material co-efficient (equivalent modulus) can be obtained as shown in eq. (11). 2 2 1 2 1 1 1 ee v v e e = − − + (10) 2 2 4 4 1 48513.3 mpa 1 0.25 1 0.25 9.1 10 9.1 10 ee = =− − + × × (11) the hertz contact stress can be calculated using eq. (9). 955000 48513.3 2258.03 mpa 20 30 5.13 cos 20 h max p × = = = × × × × σ π (12) 0.56 2258.03 1264.41 mpa vonmiss = × =σ (13) by using the same approach, the maximum hertzian contact stresses of the other gear materials selected for this study are calculated. the maximum contact stresses calculated based on the theoretical analysis (hertz’s equation) are presented in table 4. table 4 maximum theoretical contact stress for the selected gear materials by hertz’s equation selected gear materials maximum contact stress by hertz’s equation (mpa) case-hardened wrought steel 1194.56 astm class 35 cast iron 1177.75 grey cast iron (gg-30) 1264.41 beryllium-aluminum alloy (albemet 162) 1325.82 2.2. finite element method the fe method is a numerical technique used for solving the physical problems encountered in engineering and mathematical modeling. finite element analysis (fea) is the simulation of any given physical phenomenon by applying the fe method. fea is used to mathematically model the physical phenomenon and solve the complex structural, thermal transfer, fluid flow, and no. input parameters value 1 nominal input power (p) 10 kw 2 pinion speed (n) 100 rpm 3 nominal torque on the pinion shaft �� � 9550 � � � 955 nm 54 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 50-59 multi-physics problems. engineers use fea to reduce the number of experiments and/or physical prototypes and optimize the machine elements in their design phase for faster and better product developments while saving expenses. fea uses mesh generation techniques to divide a complex physical problem into finite small elements coupled with the fe algorithm [16-19]. many contact problems of rotational machine components can be solved by using the fe method. ansys workbench, one of the frequently used fe software, provides a computerized approach for predicting the response of the object to the applied loads [6, 19]. in the present study, ansys 16.0 workbench package is used to determine the maximum allowable contact stress on spur gears. the steps in ansys 16 workbench solution procedures are explained in the following sub-sections. the fe results will be compared and discussed with the theoretical results in section 3. 2.2.1. constructing the geometric model of the matting spur gears the geometric model of the gear teeth profile is constructed by using the solidworks modeling software. the geometrical data of the gears are taken from table 2. the generated model is saved in the ansys supported geometry file formats such as iges and acis. fig. 5 shows a three-dimensional (3d) geometrical model of a single spur gear done by the solidworks software. fig. 5 three-dimensional model of a spur gear by using solidworks 2.2.2. importing the model to ansys workbench the generated gear model is imported to the ansys workbench interface for the analysis purpose. all the dimensions and geometries of the gear mesh are calculated and presented in table 2 above. the imported model of assembled gears is presented in fig. 6 below. fig. 6 assembled spur gear set 2.2.3. specifying the material properties engineering materials may be specified after the geometry is imported. the specific properties of gear materials under analysis, such as young’s modulus, poisson’s ratio, and density, are used for fea. in the ansys workbench interface, the engineering data can be selected from the analysis tab and the corresponding values for each material can be inserted. 2.2.4. conducting the mesh refinement meshing is the process of dividing the entire model into small cells so that at every cell the equations are solved. it gives an accurate solution and improves the quality of the solution. mesh refinement is conducted in order to generate fine mesh elements. fig. 7 shows the mesh model of the gear mates. 55 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 50-59 (a) refined mesh of gear assembly (b) close-up view of tooth flanks fig. 7 meshed assembly in ansys workbench simulation 2.2.5. setting the boundary condition boundary condition refers to the external load on the border of the model structure. this is normally done by considering that the tooth contact is at the pitch point as illustrated in fig. 3(b). in the lower gear, the fixed support boundary condition is introduced. besides, the other gear is subjected to frictionless support and torque/moment in a clockwise direction as illustrated in fig. 8 below. fig. 8 setting the boundary condition 3. results and discussion 3.1. results the von-mises stress at the contact region for the contact model of mating spur gears is displayed on the solution processor of ansys 16 workbench. the images below (figs. 9-12) show the values for four different materials considered in this study. (a) equivalent stress for case-hardened wrought steel (b) close-up view of tooth fig. 9 von-mises stress for case-hardened wrought steel 56 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 50-59 (a) equivalent stress for grey cast iron gg-30 (b) close-up view of tooth fig. 10 von-mises stress for grey cast iron gg-30 (a) equivalent stress for astm class 35 cast iron (b) close-up view of tooth fig. 11 von-mises stress for astm class 35 cast iron (a) equivalent stress for albemet 162 (b) close-up view at tooth fig. 12 von-mises stress for beryllium-aluminum alloy (albemet 162) 3.2. discussion the main aim of this study is to analyze the change of the contact stress on spur gears with different gear materials. the analysis is done by varying the gear materials under constant loading conditions and geometric parameters by using the hertz theory and fe methods (ansys 16). the results of the analysis by these two methods are compared and presented in table 5. the selection of appropriate gear materials for different applications is the key task in gear design and/or manufacturing processes. the use of these different gear materials provides a range of contact stress which is useful in material selection [9]. as illustrated in table 4, the results of contact stress in spur gear mates using the theoretical (hertz’s equation) and fe method (ansys 16) agree with each other with a maximum error of 2.9%. this consequently indicates that the fe simulation (ansys workbench) is compatible and almost identical with the theoretical analysis (hertz’s equation) for the materials under analysis. this fact is demonstrated in fig. 13. 57 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 50-59 table 5 comparison of maximum contact stress for the selected gear materials selected gear materials maximum contact stress by ansys 16 (mpa) maximum contact stress by hertz’s equation (mpa) error (%) case-hardened wrought steel 1195.2 1194.56 -0.05% astm class 35 cast iron 1202.6 1177.75 -2.1% grey cast iron (gg-30) 1231.3 1264.41 2.6% beryllium-aluminum alloy (albemet 162) 1286.8 1325.82 2.9% fig. 13 shows that the contact stress from each material is different, though the analysis is done by keeping all conditions constant except the type of materials. this indicates that the type of gear materials has a great influence on gear failure due to the contact stress. 4. conclusions the surface strength of the gear tooth is a crucial parameter to prevent the failure of mating gears. in this study, the theoretical approach is used to predict the surface contact stress on the spur gear mates, and is done on different gear materials to study their influence on the surface contact strength of spur gears. moreover, the study shows that the effective way of predicting the surface contact stress is to use the 3d geometrical contact model of spur gears under the fe method (ansys 16). the accuracy of the fe results for the selected materials is verified by comparing the fe results with the analytical results using hertz’s standard formula. the meshed in-volute spur gear is properly modeled by rotating cylindrical contacts for the theoretical investigation of surface contact stresses under the applied torque/moment. the 3d spur gear model is done in the modeling software (solidworks) and imported to ansys 16 workbench for fea for the selected gear materials. the correlation between the theoretical formulation and fe results is compared with tolerable accuracy. the comparison shows that the results are well-matched with a maximum of 2.9% accuracy error. both the results of fe and theoretical methods show that the value of contact stress varies as the gear materials are changed. this indicates that the type of gear materials has a direct influence on the surface contact strength of spur gear mates. thus, gear designers and/or gear manufacturers have to consider gear materials as a factor for the surface contact failure of gears. conflicts of interest the authors declare no conflicts of interest. 1100 1150 1200 1250 1300 1350 case-hardened wrought steels astm class 35 cast iron grey cast iron (gg-30) beryllium-aluminum alloy (albemet 162) c o n ta c t s tr e s s i n m p a hertizian contact stress contact stress from ansys 16 fig. 13 comparison of the contact stresses obtained by theoretical analysis and ansys simulation 58 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 50-59 references [1] n. d. narayankar and k. s. mangrulkar, “contact stress and bending stress analysis of spur gear by analytical method,” international journal on theoretical and applied research in mechanical engineering, vol. 6, no. 1-2, pp. 1-3, 2017. [2] p. s. rao and c. vamsi, “contact stress and shear stress analysis of spur gear using ansys and theoretical,” international journal of modern studies in mechanical engineering, vol. 2, no. 2, pp. 9-14, 2016. [3] a. d. kalani, j. a. vadher, and r. k. jani, “fatigue failure of spur gears–a review,” international journal of modern trends in engineering and research, vol. 4, no. 11, pp. 182-186, november 2017. [4] p. singh, g. verma, and d. l. singh, “design and analysis of stress induced in spur gear tool profile using catia and ansys,” international journal of engineering and techniques, vol. 3, no. 5, pp. 118-124, september-october 2017. [5] x. deng, “analysis and prediction of gear fatigue life,” iop conference series: earth and environmental science, pp. 1-5, april 2019. [6] p. s. rao, n. sriraj, and m. farookh, “contact stress analysis of spur gear for different materials using ansys and hertz equation,” international journal of modern studies in mechanical engineering, vol. 1, no. 1, pp. 45-52, june 2015. [7] p. d. devan and v. r. muruganantham, “contact and bending stress analysis of gears–a review,” international journal for research in applied science and engineering technology, vol. 5, no. 12, pp. 2561-2565, december 2017. [8] k. sivakumar, t. m. saravana, s. balaguru, r. sabarish, and r. hariharan, “contact stress analysis of a spur gear tooth pair for two different materials,” international journal of applied engineering research, vol. 10, no. 68, pp. 344-349, january 2015. [9] r. l. norton, cam design and manufacturing handbook, new york: industrial press inc., 2002. [10] b. gupta, a. choubey, g. v. varde, “contact stress analysis of spur gear,” international journal of engineering research and technology, vol. 1, no. 4, june 2012. [11] a. r. hassen, “contact stress analysis of spur gear teeth pair, ” world academy of science, engineering, and technology, vol. 58, no. 1, pp. 597-602, 2009. [12] d. walton and a. j. goodwin, “the wear of unlubricated metallic spur gears,” wear, vol. 222, no. 2, pp. 103-113, november 1998. [13] “selection of gear material,” https://gearmotions.com/selection-of-gear-material/, december 05, 2017. [14] j. r. davis, gear materials, properties, and manufacture, ohio: asm international, 2005. [15] khk stock gears, “calculation of gear dimensions,” https://khkgears.net/new/gear_knowledge/gear_technical_reference/calculation_gear_dimensions.html, december 20, 2017. [16] m. j. khan, a. mangla, and s. h. din, “contact stress analysis of stainless steel spur gears using finite element analysis and comparison with theoretical results using hertz theory,” international journal of engineering research and applications, vol. 5, no. 4, pp. 10-18, april 2015. [17] j. liu, c. wang, and w. wu, “research on meshing stiffness and vibration response of pitting fault gears with different degrees,” international journal of rotating machinery, vol. 2020, article no. 4176430, 2020. [18] r. dharshini and m. vasundara, “analysis of load factors and modes of failure on spur gear,” international research journal of engineering and technology, vol. 5, no. 6, pp. 1618-1625, june 2018. [19] m. bozca, “investigation of the effective parameters of scuffing failure in gears,” ukrainian journal of mechanical engineering and materials science, vol. 6, no. 1, pp. 1-7, july 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/). 59  proceedings of engineering and technology innovation, vol. 22, 2022, pp. 01-09 manufacture and design an apparatus for measuring the thermal resistance of building materials ammar saliby * , béla kovács department of mathematics analysis, miskolc university, miskolc, hungary received 17 february 2022; received in revised form 22 april 2022; accepted 23 april 2022 doi: https://doi.org/10.46604/peti.2022.9500 abstract a good thermal design of buildings plays a key role in reducing the thermal loads of heating and air conditioning systems and thus reducing energy consumption. this study develops a steady-state apparatus for measuring the thermal resistance of building materials by using an absolute technique. the apparatus is designed to accommodate the full and actual size of the tested samples of building materials used in the syrian market to ease testing. therefore, suitable samples do not have to be built for the apparatus; the thermal resistance is measured for many samples, and the results are compared with the thermal insulation code for buildings in the syrian arab republic. the results show that the difference between the resistance values obtained from the experiments and the syrian thermal insulation code is due to the materials used in the local markets and their non-compliance with the code. these materials need testing and quality control during the manufacturing process. keywords: thermal resistance, thermal conductivity coefficient, steady-state, absolute technique 1. introduction in this era, wasted energy is the biggest problem in industrial, commercial, and residential fields. all these fields have a common factor, i.e., buildings. a significant amount of energy is used to make the buildings comfortable to live or work in. however, a considerable amount of energy is wasted on the building envelope. this problem is unavoidable, but the amount of energy wasted could be minimized. additionally, building materials could be improved to make better designs for energy-sustainable buildings. furthermore, updating standards and regulations can help keep up with the latest metrology technology. generally, there are two topics that receive much attention in material engineering, i.e., structure and thermal. this study will focus on the thermal side. there are multiple thermal parameters used for energy calculations in heat transfer applications. these values are available in books and references. however, with the emergence of new materials, these sources need new-found methods with these parameters [1]. the heat transfer discipline depends on explaining thermal energy transfer and its rate under specific conditions. the key elements in determining heat transfer are thermal properties, temperature difference, and dimensions. these factors form the definition of thermal resistance [2]. thermal resistance (r) in this study is limited to conduction only. it is defined as the resistance of energy flow in a structural element through its thickness [2]. therefore, increasing this resistance means the ability of the structural elements to resist the energy flow and reduce the heat transfer. these materials are considered poor conduction materials, called thermal insulators. the thermal resistance is calculated by dividing the thickness of the material (δ) by its thermal conductivity (λ), and its unit is m 2 .°c/w, as shown in eq. (1). * corresponding author. e-mail address: a.saliby@hotmail.com proceedings of engineering and technology innovation, vol. 22, 2022, pp. 01-09 2 r    (1) the thermal resistance increases when the thermal conductivity decreases or the heat sink area becomes larger while the spreading thermal resistance value stays the same. also, the effective thermal conductivity rises when the heat source area is smaller, and the heat sink area is larger [3-5]. additionally, the thermal resistance increases when the thickness of the material increases. the cost of building materials increases exponentially with wall thicknesses. therefore, finding the optimal thickness can minimize the energy waste and the overall cost of the building. many methods have been developed and used to measure the thermal resistance of different materials [6]. however, most of these methods have accuracy limits based on the temperature and thermal conductivity of the samples. proceeding from the importance of thermal resistance in heat transfer calculations, it is necessary to determine and measure the resistance value experimentally. this study identifies the methods of measuring the thermal resistance, develops an apparatus to measure the thermal resistance, and tests some samples of building materials by using the developed apparatus. the thermal resistance of these samples is also calculated theoretically based on the thermal insulation code for buildings in the syrian arab republic. finally, a comparison is made between the thermal resistance values obtained based on the code and the apparatus. 2. methodology and techniques 2.1. steady-state method in the steady-state method, temperatures are measured at a heat source and a heat sink. the temperature difference is calculated at a known distance and a constant heat flow through a sample. fig. 1 shows a schematic diagram of steady-state methods. (1) absolute technique: this method is used for the sample in the form of a cuboid or cylinder. the sample is placed between a heat source and a heat sink, and the temperature difference is measured over a known distance to the sample by thermocouples, which are the most widely used devices because of their cheapness and measurement accuracy of less than 1% [7]. (2) comparative technique: in this method, the tested sample is placed with another sample made of a material with a known thermal conductivity coefficient. the thermal conductivity is derived from the measured temperature gradients in the tested sample and the thermal conductivity of the reference materials [8]. (3) radial heat flow technique: this method allows experiments to be carried out at various temperatures, especially when the temperatures are more than 700°c, where the heat loss by radiation cannot be neglected. normally, this method uses the sample with a circular or duct shape. heat is provided to the sample by using the heaters at the axis, and the heat flows out diagonally. thermocouples are used to measure the temperature with an accuracy of ±0.1°c [9]. (4) parallel thermal conductance technique: this method is suitable for the sample with small diameters, such as carbon fibers, with a diameter of 10-100 mm. with this method, it is difficult to measure the temperature by thermocouples and determine the heat flow. a parallel heat conduction method is proposed for the needlelike sample that cannot support heaters and thermocouples [10]. before measuring the thermal resistance of the sample, the sample is measured concerning the sample support first to determine the heat loss rates related to the sample support. then, the test sample is connected to the sample support, and the thermal resistance is measured again. proceedings of engineering and technology innovation, vol. 22, 2022, pp. 01-09 3 (a) absolute technique (b) comparative technique (c) radial heat flow technique (d) parallel thermal conductance technique fig. 1 steady-state methods 2.2. transient-state method the transient-state method determines the energy dissipation as a function of time. several transient methods have been developed to eliminate the negatives of the steady-state such as the heat loss and long waiting time to reach a constant temperature difference. fig. 2 presents a schematic diagram of transient-state methods. (1) pulsed power technique: this method is similar to the steady-state method, but the heating current is a square pulse wave. one of the advantages of this method is that it allows quick measurements, especially when there are many samples [11]. (2) hot wire technique: this method measures the temperature rise at a known distance from the linear heat source. the method assumes a one-dimensional heat flow inside the sample, which means that the heat source is linear with an infinite length and infinitesimal diameter. when there is an electric current of constant intensity passing through the hot wire, the heat conduction coefficient of the sample can be derived. the available time is tested by using the resulting temperature change at a known distance from the hot wire during a period. the hot wire method is used to measure the materials with a low coefficient of thermal conductivity (such as plastics and fibers) [12]. (3) transient plane source (tps) technique: this method uses a thin metal strip or a metal disc as a flat heat source and a temperature sensor. the metal disc is electrically isolated and placed between two test samples; all other surfaces are thermally insulated. a small constant electric current is applied to the metal disc during the experiment to heat it. the thermal properties of the test samples can be determined by observing the temperature increase for a short period, and the measurement accuracy of the temperature sensor is ±0.01°c [13]. (4) laser flash technique for thermal diffusivity: thermal contact resistance is a significant source of error when measuring the temperature. the thermal diffusion method is used by laser flash to sense the temperature without contact to achieve high accuracy. this method uses rapid optical heating, and the temperature is measured using thermocouples, an infrared detector, or an optical pyrometer. the sample is prepared by coating or spraying a layer of graphite or any high emissivity material on both sides, to make an absorbent from the front side and an emitter on the backside to sense the temperature [14]. proceedings of engineering and technology innovation, vol. 22, 2022, pp. 01-09 4 (a) pulsed power technique (b) hot wire technique (c) transient plane source technique (d) laser flash technique for thermal diffusivity fig. 2 transient-state methods there are two general types of apparatus for determining the thermal resistance of materials. the first type (the hot plate apparatus), which is the most common, is used to determine the thermal resistance from one material surface to the other surface. the apparatus consists of a heating section lined with double-sided copper plates. the copper plates are divided into two parts. these two parts are completely separated by an air gap that prevents any metallic contact between the inner and outer parts of the heating plate, as the inner part is the test part. the outer part is a protective and insulating ring to prevent heat loss during the experiment [15-17]. the second type (hotbox apparatus) is used to determine the total thermal resistance of the air on one side to the air on the other side of the material. the hot plate apparatus is generally accepted to determine the heat resistance. the main goal of any device is to obtain accurate results. this result can be achieved by taking accurate temperature measurements and achieving a correct heat flow direction through the material. special precautions must be taken to prevent heat loss from the test values and ensure that the heat flow lines are perpendicular to the test surfaces. 3. equipment and procedures used in apparatus design 3.1. purpose of apparatus the design of the apparatus, which is based on the absolute technique in the steady-state method, aims to measure the thermal resistance by applying a constant heat flow during the experiment and measuring the temperature difference between the heat source and the heat sink. 3.2. apparatus design and manufacture the thermal resistance measurement apparatus is designed based on the absolute technique in the steady state, where a constant heat source heats the sample, and the temperature difference is measured over a distance known to the sample by temperature sensors. a thermal insulator surrounds the sample and the heating part to ensure the one-dimensional heat transfer and reduce the heat loss by convection and radiation. fig. 3 shows the design of the apparatus in solidworks, where the dimensions are set based on the building materials. proceedings of engineering and technology innovation, vol. 22, 2022, pp. 01-09 5 fig. 3 apparatus design in solidworks 3.3. basic parts the thermal resistance measurement apparatus for building materials mainly consists of a metal structure made of iron bars and a number of thermal insulation types depending on the temperatures to prevent the heat loss. table 1 shows the basic parts of the apparatus. the insulation materials, such as polystyrene, mineral wool, and glass, can be replaced with foam geopolymer materials based on fly ash. they are a non-flammable material characterized by relatively good insulation parameters [18]. fig. 4 illustrates the horizontal projection of the apparatus with its constituent parts. table 1 basic parts of the proposed apparatus no. part name dimensions length (cm) width (cm) height (cm) 1 metal frame 60 60 40 2 aluminum plate 40 1 20 3 mineral wool 40 5 40 4 polystyrene board 40 10 40 (a) dimensions of the apparatus (b) the apparatus after assembly fig. 4 horizontal projection of the apparatus after assembly 3.4. other parts the apparatus contains other systems to complete its work. these systems are the heat source, control system, and sensors. an electrical heater is used as the heat source to provide heat to the system. the control system is used to maintain the temperature at around 200°c, display the temperature from the sensors, and log the data. the instruments used in these systems are described below. proceedings of engineering and technology innovation, vol. 22, 2022, pp. 01-09 6 fig. 5 thermocouples (type k) fig. 6 arduino uno r3 board fig. 7 max6675 assembled printed circuit board (pcb) fig. 8 lcd (type 2004a) fig. 9 power display fig. 10 electric heater on the aluminum plate fig. 11 power supply (1) thermocouples: six k-type thermocouples, which have a length of 100 mm and a diameter of 5 mm, are shown in fig. 5. (2) data logger: the code is designed and written to be compatible with the heat resistance apparatus based on the arduino uno r3 microcontroller. it is an open-source electronic board to easily develop many control projects using the open-source programming language arduino c, as shown in fig. 6 [19]. (3) analog-digital converter (adc) and cold junction compensation: as shown in fig. 7, this study selects an max6675 integrated circuit compatible with k-type thermocouples and a 12-bit adc with a heat-sensing diode [20]. (4) temperature display and data recording: as shown in fig. 8, a liquid-crystal display (lcd) screen (4 × 20) is installed to read the temperature values measured by thermocouples. (5) display power provided: a power screen, as shown in fig. 9, is installed to read the power values provided by the electric heater. (6) electric heater: an electric heater works as a heat source, wraps the heater, and is fixed by brackets on the aluminum plate. the electric heater with a length of 1 m and a diameter of 5 mm is installed as shown in fig. 10. (7) electrical power supply: as shown in fig. 11, an electrical supply source (12 v/5 a) is installed to supply the equipment with constant voltage. 3.5. apparatus calibration apparatus calibration is done to determine the amount of heat flowing through the sample and to determine the error that results from the heat loss by convection, radiation, contact resistance, and resistance of aluminum plates. the calibration is done using a material with a known thermal conductivity coefficient, such as gypsum with dimensions of 40 × 20 cm, a thickness of 1.6 cm, and a thermal conductivity coefficient of 0.16 w/m.°c. the heat is presented on the front side of the sample at a temperature of tai = 200°c by an electric heater. after reaching the steady state, the mean temperature at the front face of the sample is tai = 200°c, and the mean temperature at the opposite side is tao = 85°c. fig. 12 shows the mean temperatures tai and tao. through this calibration, the value of the heat provided is known without losses by convection, radiation, contact resistance, and resistance of aluminum plates. the error value from the losses is included in the calculation code for arduino uno. proceedings of engineering and technology innovation, vol. 22, 2022, pp. 01-09 7 4. tests and experiments experiments are conducted on the materials used in the syrian local markets, where two types of the most famous samples are selected to verify their thermal resistance and compare the results with the syrian thermal insulation code. (1) the sample is a hollow cement block: the sample has a thickness of 10 cm and is placed within the apparatus. the experiment is conducted by adding heat to the front side of the sample using the electric heater. after the system reaches the steady state, the mean temperature at the front side (the heat source) tai is 200°c, and the mean temperature at the opposite side tao is 85°c, the temperature difference δt is 115°c. the heat of 125 w is introduced to the system, resulting in the thermal conductivity coefficient λ of 1.28 w/m.°c. fig. 13 illustrates the tested sample’s dimensions, and fig. 14 shows the mean temperature tai measured on the front side of the sample (heat source) over time and the mean temperature tao on the backside of the sample (heat sink). (2) the sample is a wall section: the wall section contains a hollow cement block with a thickness of 10 cm and two layers of cement (plaster) with a thickness of 2 cm. the experiment is conducted by adding heat to the front side of the sample using the electric heater. after the system reaches the steady state, the mean temperature at the front side (the heat source) tai is 200°c, the mean temperature at the opposite side tao is 65°c, and the temperature difference δt is 135°c. the heat of 125 w is introduced to the system, resulting in the thermal conductivity coefficient λ of 1.62 w/m.°c. fig. 15 illustrates the tested sample’s dimensions, and fig. 16 shows the mean temperature tai measured on the front side of the sample (heat source) over time and the mean temperature tao on the backside of the sample (heat sink). fig. 13 dimensions of the hollow cement block 0 50 100 150 200 250 0 0.5 1 1.5 2 t e m p e ra tu re [ °c ] time [hrs.] tai tao 0 50 100 150 200 250 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 t e m p e ra tu re [ °c ] time [hrs.] tai tao fig. 14 mean temperatures (tai and tao) of the hollow cement block fig. 12 mean temperatures (tai and tao) of the gypsum board proceedings of engineering and technology innovation, vol. 22, 2022, pp. 01-09 8 fig. 15 dimensions of the wall section fig. 16 mean temperatures (tai and tao) of the wall section as shown in table 2, the physical and thermal characterization of the tested materials includes: length (a), width (b), weight (w), density (ρ), thermal conductivity coefficient (λ), thermal resistance (r), thermal resistance according to the code [21] (r*), absolute error, and relative error. table 2 physical and thermal characterization of materials materials characteristics hollow cement block wall section a (cm) 40 40 b (cm) 20 20 δ (cm) 10 14 w (kg) 12.1 16.4 ρ (kg/m 3 ) 1513 λ (w/m.°c) 1.282 1.624 r (m 2 .°c/w) 0.078 0.086 r * (m 2 .°c/w) 0.100 0.139 absolute error 0.022 0.053 relative error 22 % 38 % 5. conclusions this study designs an apparatus for measuring the thermal resistance of the building materials used in the syrian arab republic. the results indicate that the thermal resistance values obtained by the experiment differ from the values obtained by the thermal insulation code for buildings in the syrian arab republic. this difference is due to the materials used in the local markets and their non-compliance with the code. these materials need testing and quality control during the manufacturing process. knowing the heat resistance value is necessary to calculate the lost or gained thermal energy through the structural element and to judge the efficiency of thermal insulation with various structural elements. the lower the value of the thermal conductivity coefficient, the greater the thermal insulation capacity and the higher the percentage of savings in thermal energy used for heating and air conditioning purposes. to measure the thermal resistance, it is essential to develop an apparatus which can work in different positions to achieve the horizontal and vertical heat transfer, determine the amount of heat, and reduce the measurement errors and heat loss. 0 50 100 150 200 250 0.0 1.0 2.0 3.0 4.0 t e m p e ra tu re [ °c ] time [hrs.] tai tao proceedings of engineering and technology innovation, vol. 22, 2022, pp. 01-09 9 nomenclature r thermal resistance (m 2 .°c/w) tai mean temperature (heat source) (°c) tao mean temperature (heat sink) (°c) w weight (kg) δ thickness (cm) λ thermal conductivity coefficient (w/m.°c) ρ density (kg/m 3 ) conflicts of interest the authors declare no conflicts of interest. references [1] y. a. çengel, heat transfer: a practical approach, 2nd ed., new york: macgraw-hill, 2003. [2] f. p. incropera, et al., fundamentals of heat and mass transfer, 6th ed., new york: john wiley & sons, 2007. [3] m. kim, et al., “numerical case study and modeling for spreading thermal resistance and effective thermal conductivity for flat heat pipe,” case studies in thermal engineering, vol. 31, article no. 101803, march 2022. [4] m. kim, et al., “numerical investigation and modeling of thermal resistance and effective thermal conductivity for two-phase thermosyphon,” case studies in thermal engineering, vol. 27, article no. 101358, october 2021. [5] j. h. moon, et al., “boiling-driven, wickless, and orientation-independent thermal ground plane,” international journal of heat and mass transfer, vol. 167, article no. 120817, march 2021. [6] d. zhao, et al., “measurement techniques for thermal conductivity and interfacial thermal conductance of bulk and thin film materials,” journal of electronic packaging, vol. 138, no. 4, article no. 040802, december 2016. [7] standard test method for steady-state heat flux measurements and thermal transmission properties by means of the guarded-hot-plate apparatus, astm c177, 2019. [8] standard test method for thermal conductivity of solids by means of the guarded-comparative-longitudinal heat flow technique, astm e1225, 2004. [9] standard test method for steady-state heat transfer properties of pipe insulation, astm c335/c335m, 2017. [10] b. m. zawilski, et al., “description of the parallel thermal conductance technique for the measurement of the thermal conductivity of small diameter samples,” review of scientific instruments, vol. 72, no. 3, pp. 1770-1774, march 2001. [11] o. maldonado, “pulse method for simultaneous measurement of electric thermopower and heat conductivity at low temperatures,” cryogenics, vol. 32, no. 10, pp. 908-912, april 1992. [12] standard test method for thermal conductivity of refractories by hot wire (platinum resistance thermometer technique), astm c1113/c1113m, 2009. [13] standard test method for measurement of thermal effusivity of fabrics using a modified transient plane source (mtps) instrument, astm d7984, 2016. [14] standard test method for thermal diffusivity by the flash method, astm e1461, 2013. [15] m. m. terzić, et al., “development of a single-sided guarded hot plate apparatus for thermal conductivity measurements,” thermal science, vol. 20, no. 1, pp. 321-329, 2016. [16] w. hemminger, et al., “a guarded hot-plate apparatus for thermal conductivity measurements over the temperature range -75 to 200 [°c],” international journal of thermophysics, vol. 6, no. 5, pp. 489-498, 1985. [17] u. hammerschmidt, “guarded hot-plate (ghp) method: uncertainty assessment,” international journal of thermophysics, vol. 23, no. 6, pp. 1551-1570, 2002. [18] 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. [19] “arduino uno r3,” http://www.arduino.cc, june 2022. [20] “cold-junction-compensated k-thermocoupleto-digital converter (0°c to +1024°c),” http://www.maximintegrated.com, june 2021. [21] thermal insulation code for buildings in the syrian arab republic, syrian engineers association, 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/). http://www.arduino.cc/ http://www.maximintegrated.com/  proceedings of engineering and technology innovation, vol. 13, 2019, pp. 10-19 impact of dfig in wind energy conversion system for grid disturbances sasmita behera * , sudeep kumar behera, bibhuti bhusan pati department of electrical & electronics engineering, veer surendra sai university of technology, burla, odisha, india received 09 april 2018; received in revised form 11 december 2018; accepted 25 january 2019 abstract in this work, a grid-connected doublyfed induction generator (dfig) is studied for the transient and steady response. the vector control technique controls the pulse width modulation (pwm) of both the back-to-back converters interfacing rotor to the grid. reactive power supply and dc bus voltage are managed by the grid-side inverter. active power and rotor angular speed are adjusted by the machine side inverter facilitating power generation for varying wind. the effect of voltage and frequency deviation from the grid on the control is observed. the controllers are found to work satisfactorily except for large frequency variation. the current harmonics are also within the allowed limit. the proposed controllers are expected to satisfy the revised grid code for wind energy. keywords: vector control, pwm, voltage sag, voltage swells 1. introduction increasing share of wind power in grid connection draws attention to the study of its effect on power system stability. in the wind energy conversion system (wecs), different generators are in use. the majority of variable speed wecs have doubly fed induction generator (dfig) [1]. this is due to the efficient energy conversion, low cost, less mechanical stress etc. [2]. the increased proportion of wind power has forced the grid codes to be modified from its isolation by crowbar [1] to ride through the control during the grid disturbance [3]. active and reactive power control for changing wind speed is accomplished by a stator flux oriented vector control with its real axis aligned to the stator flux vector rotating at the synchronous speed in [4]. the performance of such a system for various balanced and unbalanced fault conditions is explored in [5 -6]. there are many possible ways to control the outputs of the generator with different orientation of the real axis of the reference frame [7]. minor but frequent disturbances in the grid such as voltage dip, sag and swell in the grid lead to poor of power quality, reliability,and equipment failure [8]. such disturbances associated with voltage can be supplemented by reactive power balance. different external reactive power support devices have also been in research, but with the additional cost [9-11]. in [12], upholding of dfig by internal control for low voltage in stator voltage oriented reference frame is studied. in another study, the swell case of voltage is observed in [13]. xiao et al. [14] have tentatively defined a maximum limit beyond which the crowbar protection should be activated. selectively harmonics have been eliminated by proper control of dfig in [15]. to minimize rotor current that activates crowbar during low voltage, additional feed-forward control has been proposed in [16]. instead of voltage oriented vector control, a direct current control [17] has been proposed for faster control. mohseni et al. [18-20] have proposed switching to hysteresis control during transient rather than utilizing proportional-integral (pi) current control for steady-state in order to satisfy grid code for the ride-through requirement. but, harmonics with hysteresis control is beyond the standard. pi with resonance control has been proposed in [21-22], to reduce torque pulsation and unbalanced voltage [23]. in addition, to reduce the transient time, torque oscillation for voltage swells virtual impedance control strategy has been proposed in [24]. *corresponding author. e-mail address:sasmitabehera2000m@gmail.com proceedings of engineering and technology innovation, vol. 13, 2019, pp. 10-19 11 the model predictive control (mpc) for switching converter instead of pwm has been reviewed in [25]. though it was faster than pwm, the thd and steady-state performance were inferior. in a comparative study, tremblay et al. [26] have shown robustness to noise and better control of the dfig by pi control in vector control in comparison to direct power control (dpc) and direct torque control (dtc) approach. in recent literature, a demagnetizing current control based on vector control has been proposed for faster control for low voltage [27]. the studies are also limited to sag, swell or harmonic reduction. having reviewed the literature for improvement of control to meet the latest grid codes for wecs, the vector control with pi controller is still unsurpassed though many other control techniques, which have been developed [28-32]. herein, the vector control technique controls the pwm of both the back to back converters interfacing rotor to the grid to maintain reliable power. the sections 2-5 embody the equations governing the different parts of the model. response to grid disturbances is presented in section 6 concluding the impact of dfig in this condition. 2. model of wt-dfig system the power of the wind drives the wind turbine (wt) coupled to dfig. with the gearbox in between, the rotational speed is stepped up suitably and the dfig generates electricity from the mechanical input. fig. 1 shows a dfig vector control strategy. for the accurate design of the dynamic model in the d-q frame, the transient reactance of the generator is considered. the equivalent circuits of dfig and induction machine are analogous [7, 33]. fig. 1 overall structure of the dfig model definingthe different electrical circuit parameters in the d-q frame as in eqs. (1-5), it yields   qr rr sss d l l e )( '  (1)   dr rr ms q l l e )( '  (2) lx ssss  (3)                  l l lx rr m ssss 2 '  (4) r l t r rr '0 (5) proceedings of engineering and technology innovation, vol. 13, 2019, pp. 10-19 12 the dfig is modelled as in eqs. (6) to (10) :     ixe t v l l esixx t rv dt dix qssq s dr rr m ddsss s sds s dss '' ' 0 '' ' 0 ' 1 1 1         (6)     ixe t v l l esixx t rv dt dix dssq s qr rr m qqsss s sqs s qss        '' ' 0 '' ' 0 ' 1 1 1  (7)   ixxe t v l l e dt de qssdqr rr m sqs d s '' ' 0 ' ' 1  (8)   ixxe t v l l e dt de dssqdr rr m sds q s '' ' 0 ' ' 1  (9) dr rr mdr dsi il l  (10) i il lqr rr qr m qs  (11) where , l , r , x , v , e , i are flux linkage, inductance, reactance, voltage, emf, and current respectively. the first subscript d, q stands for the directors and the quadrature component respectively, and the second subscript r and s represent as rotor and stator respectively. the repetition of subscript s and r for l stand for the corresponding self-inductance, whereas single m stands for mutual inductance. t is the synchronous speed, s is the slip, 0 ' t is the electrical time constant of the rotor,the superscript ′ represents a transient condition. equating the lhs of eqs. (3) and (4), and neglecting rotor resistance dsv .then substituting for dr and qr , it reduces to ( ) ( ) dr s r rr qr s r m qs v l i l i        (12) ( ) ( ) qr s r rr dr s r m ds v l i l i       (13) 3. the drive train to model drive train, the two-mass model is simple and it amply represents the dynamics of wt suitable for transient state analysis [34]. the two masses are the generator-gearbox and the hub with blades. mathematically, it is expressed by tt dt d h sw s t   2 (14)   )1( s dt d strt t  (15) tt dt ds h seg 2 (16) dt d dkt t stss    (17) where th and gh represent the inertia of the hub and the generator, respectively, t is the turbine angular speed, r is the generator rotor angular speed, t is the shaft torsion angle, sd and sk are the is the stiffness and damping coefficients of shaft respectively, st is the shaft torque, wt is the effective torque input to the wt, et is the electromagnetic torque. wt and et are given by proceedings of engineering and technology innovation, vol. 13, 2019, pp. 10-19 13 v cr t w t pa w 3 25.0    (18) / e s s t p  (19)    0068.0)54.0 116 (5176.0 21   ec i i p (20)  3 1 035.0 08.0 11     i (21) t w r v    (22) where ap is the air density, r is the radius swept by wt blade, wv is the velocity of wind,  is the pitch angle,  is the blade tip speed ratio, sp is the stator active power, pc is the power coefficient. by controlling t , maximum pc and maximum power from wind can be achieved. 4. converter modeling from fig. 2(b), the active power equation is: dcr gp p p  (23) where r p is the rotor power that charges the dc bus capacitor with dc p , and delivers g p to the grid. these are given by r dr dr qr qrv i v ip   (24) g dg dg qg qgv i v ip   (25) dc dc dc dc dc c dt dv v i vp    (26) where v stands for the voltage and i the current with the second subscript r and g represent for rotor and grid respectively. dc v , dc i and c are the voltage, current,and capacitance of the capacitor in dc link. now, rewriting eq. (18), the power balance in terms of voltages and currents gives: ( )dc vdc dg dg qg qg dr dr qr qr dv c v i v i v i v i dt     (27) 5. controller modeling fixing d-axis of the d q reference frame with dsv , qsv nullifies and dsv equals the grid voltage. gp and reactive power g q can be regulated separately by dg i and qg i respectively in a decoupled manner by vector control resolving three phases to d q components [35]. the stator active power and voltage can also be adjusted separately by qr v and dr v as described below. 5.1.machine-side control the controller on the rotor-side retains the terminal voltage to set point and controls the dfig output active power varying with the wt torque [27]. the control of active power and voltage are achieved by qr v and drv , respectively. the control structure is given in fig. 2(a), while the control is expressed as: proceedings of engineering and technology innovation, vol. 13, 2019, pp. 10-19 14  2 1 1 21 2p p i i m rrqr qr r s ds r s qrpv x i x s i s ik k k k l l        (28)  2 3 3 23 4p p i i rrdr dr r s drvv x i x s ik k k k l      (29) where the gain of the pi controller is xy k . the first subscript x with notation p and i represent proportional and integral gains respectively. the second subscript 1, 2, 3 are for the power controller, machine-side current controller, and grid voltage controller, respectively. dg ef i r and dg ef i r are the references for the respectivecomponents of the current on the rotor, s ref v  is the reference for stator terminal, ref p is the reference of active powerfound from r p b t ef tb p    (30) where tb is the wt angular speed base, bp is the maximum output active power at tb . 5.2.grid-side control the grid-side converter controller (fig. 2(b)), controls dcv by dgi and the reactive power by dgi . the change in intermediary states 1 x , 2x and 3x , shown in fig. 2(b) are expressed as: 1 dc ref dc dc dx dt v v v    (31) 2 1pdg idgdg ref dg dc dg dx dt k ki i v x i      (32) 3 qg ref qg dx dt i i   (33) 2 2pg igdg dt dx k kv x  (34) 3 3igqg pg dt dx k kv x  (35) (a) machine-side converter control (b) grid-side converter control fig. 2 the control of converters [12] where pdg k and idg k are the proportional and integral gains of the dc bus voltage controller, respectively. its output dg ef i r is the set point for dg i ; pg k and ig k are the proportional and integral gains of the grid-side converter current regulator, respectively, dc ef v r is the dc bus voltage set point and dq efi r is the reference for dqi [7]. dc link voltage drop due to increased active power demand due to or wind speed variation in sub-synchronous speed operation is controlled by a change in dg v (cf. eq. (22)). reactive power is controlled by controlling dgv . pwm gating pulses are generated in a similar manner from above two grid voltages as shown in fig. 2(a). proceedings of engineering and technology innovation, vol. 13, 2019, pp. 10-19 15 6. result discussion the wecs considered is integrated to a 120 kv, 3 phase 2500 mva grid with transformers in between. the model is built in matlab/simulink with system specifications: w v :15 m/s,dfig: 9kw, 575 v, 6 pole, 1440 rpm, 60 hz; dc bus voltage 1150 v with a capacitor of 60 mf; frequency of the pwm carriers 1.62 khz and 2.7 khz for machine-side and grid-side converters respectively.to reduce the injection of harmonics from dfig via the grid side converter to the grid, the switching frequency is kept high. 6.1. balanced grid the stator currents and voltages of dfig under steady-state condition (the wind speed constant at 15m/s) are depicted in fig. 3(a). the active power (blue) produced by dfig is 0.9 pu which is excluding electromechanical losses from the mechanical input, which is similar to the findings in [4]. the reactive power (green) reference is 0 kvar. the dc bus voltage remains stable at 1150 v. the rotor angular speed is 1.2 pu, the rated value. (a) no disturbance (b) with voltage sag fig. 3 simulation results of performance parameter variation with times 6.2. voltage sag on the grid grid voltage sag occurs due to generation outage, line outage and switching of a heavy fluctuating load. when voltage sag of 0.15 pu [16, 21] is uniformly applied in all phases at 0.05 s for 9 cycles, the effects are depicted in fig. 3(b). normall y the deviation is limited to ±10% as per ieee 1159. abnormal violations of grid voltage, which sag beyond ±10%, are taken to verify the effectiveness of the control system. on abrupt deviation of the grid voltage from its rated, the stator current rises and the active power p fluctuates and the active power control competently brings it to its rating within the next 4 cycles. for voltage sag reactive power is supported by dfig with an increased injection of vqg and adjusting vdr . it supplies reactive power to the grid at that time and settles to 0 kvar by appropriate control action. for a variation of up to 25%, such rise of current due to sag has a negligible effect on dfig and the controller works well. the dc bus voltage oscillates at the switching instants and comes down to its set value of 1150 v by the grid side converter. the rotor speed is constant (1.2 pu) as v w is fixed. proceedings of engineering and technology innovation, vol. 13, 2019, pp. 10-19 16 6.3. voltage swells on the grid under symmetrical rise of voltage by 0.2 pu at 0.05 s continuing for 9 cycles, the variations are depicted in fig. 4(a).the grid voltage may swell under outage of a large load. when the grid voltage swells, the stator current falls to maintain powerconstant. the reactive power q falls suddenly due to a decrease in power demand and then comes back to 0 kvar within 4 cycles which is better than that in [8] and [13]. it sustains as per australian grid code [20]. finally, the system settles to its set value. 6.4. fall of grid frequency when the grid frequency drops from 60 hz rated to 58 hz at 0.05 s, the proposed control strategy does not act upon it as seen from fig. 4(b). such condition of frequency fall may arise due to loss of large generation. though the case is quite unusual, as the iec standard limits grid frequency deviation to ±0.5 hz, it is shown to check the prominent effect in the simulation. it is indicative of the modification of control required to counter to such large change in frequency. however, for the deviation within standard limits the controller performs to stabilize. since the machine rated frequency is 60 hz, it is not responding to 58 hz, and the machine collapses with a gradual fall of rotor angular speed and dc bus voltage. compared to the grid, the capacity of the generator is small, hence this dfig can’t support frequency stabilization of the grid. (a) voltage swell (b) with voltage sag fig. 4 simulation results of performance parameters variation with times 6.5. current harmonics fig. 5 thd of the stator current the harmonics in the stator current of the dfig are also analyzed and given in fig. 5. the dominant odd harmonics have suppressed considerably. the thd is 1.89 %, which is satisfied as per the standards. proceedings of engineering and technology innovation, vol. 13, 2019, pp. 10-19 17 7. conclusions a dfig-wt system with a back-to-back vsc is studied with its pwm controlled by pi controllers without a crowbar. the impact of dfig is investigated under steady state, for a large deviation of ±20% in voltage and 3% in the frequency of the grid. the stator voltage, current, active and reactive power exchange between the grid and the dfig are observed. in conclusion, the dfig supports grid within stability limits for the cases being discussed except for variation in frequency with the implementation of the vector control strategy. in the grid-connected dfig, the harmonic suppression measure has already been presented. since the harmonicfilter at the input to the controller, it is expected that the control will work in the presence of harmonics. the stator current has 1.89% thd, well within the limit. further study can be focused on the improvement of control performance in the larger rating of dfig and unbalanced faults. acknowledgement the facilities created by the project aicte/8023/rid/rps/19/11-12 dt.15.5.12 of the department entitled “transient stability analysis and control of power system with excitation control”, are gratefully acknowledged. conflicts of interest the authors declare no conflict of interest. references [1] j. lópez, e. gubía, e. olea, j. ruiz, and l. marroyo, “ride through of wind turbines with doubly fed induction generator under symmetrical voltage dips,”ieee transactions on industrial electronics,vol. 56, no. 10, pp. 4246-4254, 2009. [2] r. mittal, k.s. sandhu, and d.k. jain, “an overview of some important issues related to wind energy conversion system (wecs),” international journal of environmental science and development, vol. 1, no. 4, pp. 351-363, 2010. [3] a.k. pathak, m.p. sharma, and m. bundele, “a critical review of voltage and reactive power management of wind farms,”renewable and sustainable energy reviews,vol. 51, pp. 460-471, 2015. [4] b. c. babu and k.b. mohanty, “doubly-fed induction generator for variable speed wind energy conversion systems modeling &simulation,” international journal of computer and electrical engineering, vol. 2, no. 1, pp. 1793-1806, february, 2010. [5] k. s. rao and m. v. kumar, “analysis of doubly fed induction generator under varies fault conditions,” international journal of engineering research and applications, vol. 3, no. 6, pp. 2102-2106, november-december 2013. [6] x. yan, g. venkataramanan, p.s. flannery, y. wang, and b. zhang, “evaluation of the effect of voltage sags due to grid balanced and unbalanced faults on dfig wind turbines,”european power electronics and drives journal,vol. 20, no. 4, pp. 51-61, 2010. [7] f. wu, x.p. zhang, k. godfrey, and p. ju, “small signal stability analysis and optimal control of a wind turbine with doubly fed induction generator,”iet generation, transmission and distribution,vol. 1, no. 5, pp. 751-760, 2007. [8] n. edomah, “effects of voltage sags, swell and other disturbances on electrical equipment and their economic implications,” in iet electricity distribution-part 1, 20th international conference and exhibition oncired, 2009, pp. 1-4. [9] m.e. hossain, “low voltage ride-through capability improvement methods for dfig based wind farm,”journal of electrical systems and information technology, vol. 5, no. 3, 2018. [10] s. agalar and y.a. kaplan, “power quality improvement using sts and dvr in wind energy system,”renewable energy,vol. 118, pp. 1031-1040, 2018. [11] s. priyavarthini, c. nagamani, g.s. ilango, and m.a. a. rani, “an improved control for simultaneous sag/swell mitigation and reactive power support in a grid-connected wind farm with dvr,”international journal of electrical power and energy systems,vol. 101, pp. 38-49, 2018. [12] r. d. shukla and r. k. tripathi, “dynamic performance of dfig based wecs under different voltage sag,”international journal of chemtech research,vol. 5, no. 2, pp. 980-992, 2013. proceedings of engineering and technology innovation, vol. 13, 2019, pp. 10-19 18 [13] s. choudhury, k.b. mohanty, and b.k. debta, “investigation on performance of doubly-fed induction generator driven by wind turbine under grid voltage fluctuation,” inieee 10thinternational conference onenvironment and electrical engineering, 2011, pp. 1-4. [14] s. xiao, h. geng, h. zhou, and g. yang, “analysis of the control limit for rotor-side converter of doubly fed induction generator-based wind energy conversion system under various voltage dips,”iet renewable power generation,vol. 7 no. 1, pp. 71-81, 2013. [15] m. kesraoui, a. chaib, a. meziane, and a. boulezaz, “using a dfig based wind turbine for grid current harmonics filtering,”energy conversion and management,vol. 78, pp. 968-975, 2014. [16] j. liang, w. qiao, and r.g. harley, “feed-forward transient current control for low-voltage ride-through enhancement of dfig wind turbines,”ieee transactions on energy conversion,vol. 25, no. 3, pp. 836-843, 2010. [17] s. li, t.a. haskew, k.a. williams, and r.p. swatloski, “control of dfig wind turbine with direct-current vector control configuration,”ieee transactions on sustainable energy,vol. 3, no. 1, pp. 1-11, 2012. [18] m. mohseni, m.a. s. masoum, and s.m. islam, “low and high voltage ride-through of dfig wind turbines using hybrid current controlled converters,”electric power systems research,vol. 81, no. 7, pp. 1456-1465, 2011. [19] m. mohseni, s.m. islam, and m. a. s. masoum, “enhanced hysteresis-based current regulators in vector control of dfig wind turbines,”ieee transactions on power electronics,vol. 26, no. 1, pp. 223-234, 2011. [20] m. mohseni and s.m. islam, “transient control of dfig-based wind power plants in compliance with the australian grid code,”ieee transactions on power electronics,vol. 27, no. 6, pp. 2813-2824, 2012. [21] h. xu, w. q. zhang, h. nian, and j. li, “improved vector control of dfig based wind turbine during grid dips and swells,” inieeeinternational conference onelectrical machines and systems (icems), 2010, pp. 511-515. [22] j. hu, h. nian, h. xu, and y. he, “dynamic modeling and improved control of dfig under distorted grid voltage conditions,”ieee transactions on energy conversion,vol. 26, no. 1, pp. 163-175, 2011. [23] h. nian, y. song, p. zhou, and y. he, “improved direct power control of a wind turbine driven doubly fed induction generator during transient grid voltage unbalance,”ieee transactions on energy conversion,vol. 26, no.3, pp. 976-986, 2011. [24] z. xie, x. zhang, x. zhang, s. yang, and l. wang, “improved ride-through control of dfig during grid voltage swell,”ieee transactions on industrial electronics,vol. 62, no. 6, pp. 3584-3594, 2015. [25] j. rodriguez, m.p. kazmierkowski, j.r. espinoza, p. zanchetta, h. a. rub, h.a. young, and c.a. rojas, “state of the art of finite control set model predictive control in power electronics,”ieee transactions on industrial informatics,vol. 9, no. 2, pp. 1003-1016, 2013. [26] e. tremblay, s. atayde, and a. chandra, “comparative study of control strategies for the doubly fed induction generator in wind energy conversion systems: a dsp-based implementation approach,”ieee transactions on sustainable energy,vol. 2, no. 3, pp. 288-299, 2011. [27] m.k. döşoğlu, “a new approach for low voltage ride through capability in dfig based wind farm,” international journal of electrical power and energy systems,vol. 83, pp. 251-258, 2016. [28] manaullah, a.k. sharma, h. ahuja, and a. singh, “performance comparison of dfig and scig based wind energy conversion systems,” in innovative applications ofcomputational intelligence on power, energy and controls with their impact on humanity, pp. 285-290, november 2014. [29] s.k.tiwari, b. singh, and p.k. goel, “design and control of microgrid fed by renewable energy generating sources,”ieee transactions on industry applications,vol. 54, no. 3, pp. 2041-2050, 2018. [30] d.s. l. simonetti, a.e. a. amorim, and f.d. c. oliveira, “doubly fed induction generator in wind energy conversion systems,” inadvances in renewable energies and power technologies, pp. 461-490, 2018. [31] y. deng, z. xing, and q. zhang, “analysis of electromagnetic transient characteristics of doubly-fed induction generator under grid voltage swell,”cpss transactions on power electronics and applications,vol. 3, no. 2, pp. 111-118, 2018. [32] s. e. d. silveira, s. m. silva, and b. j. c. filho, “fault ride-through enhancement in dfig with control of stator flux using minimized series voltage compensator,”iet renewable power generation,pp.1234-1240, april 2018. [33] a. panda and m. tripathy, “optimal power flow solution of wind integrated power system using modified bacteria foraging algorithm,”international journal of electrical power and energy systems,vol. 54, pp. 306-314, 2014. [34] e.b. muhando, t. senjyu, a. uehara, and t. funabashi, “gain-scheduled h ∞ control for wecs via lmi techniques and parametrically dependent feedback part ii: controller design and implementation,”ieee transactions on industrial electronics,vol. 58, no. 1, pp. 57-65, 2011. proceedings of engineering and technology innovation, vol. 13, 2019, pp. 10-19 19 [35] m. yamamoto and o. motoyoshi, “active and reactive power control for doubly-fed wound rotor induction generator,” ieee transaction on power electronics, vol. 6, no. 4, pp. 624-629, 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/).  proceedings of engineering and technology innovation, vol. 15, 2020, pp. 42 51 a learning-based em clustering for circular data with unknown number of clusters shou-jen chang-chien, wajid ali, miin-shen yang * department of applied mathematics, chung yuan christian university, taoyuan, taiwan received 04 february 2020; received in revised form 04 march 2020; accepted 21 april 2020 doi: https://doi.org/10.46604/peti.2020.5241 abstract clustering is a method for analyzing grouped data. circular data were well used in various applications, such as wind directions, departure directions of migrating birds or animals, etc. the expectation & maximization (em) algorithm on mixtures of von mises distributions is popularly used for clustering circular data. in general, the em algorithm is sensitive to initials and not robust to outliers in which it is also necessary to give a number of clusters a priori. in this paper, we consider a learning-based schema for em, and then propose a learning-based em algorithm on mixtures of von mises distributions for clustering grouped circular data. the proposed clustering method is without any initial and robust to outliers with automatically finding the number of clusters. some numerical and real data sets are used to compare the proposed algorithm with existing methods. experimental results and comparisons actually demonstrate these good aspects of effectiveness and superiority of the proposed learning-based em algorithm. keywords: clustering, circular data, mixtures of von mises distributions, em algorithm, learning schema 1. introduction since von mises [1] proposed a distribution on circular data, watson and williams [2] considered statistical inferences for von mises distributions. afterward, circular data were widely applied in biology, geology, medicine, oceanography, and meteorology [3-5]. clustering is a useful tool for data mining. according to the statistical point of view, clustering methods can be generally divided into two categories. one is a model-based approach. another is a nonparametric approach. in the model-based approach, the expectation and maximization (em) algorithm [6-8] is the most used method. for a nonparametric approach, an objective function of dissimilarity measures is generally considered in which partitional clustering is the most popular, such as k-means [9-11], fuzzy c-means (fcm) [12-13], and possibilistic c-means (pcm) [14-15]. generally, the em algorithm was popularly used in the analysis of circular data based on mixtures of von mises distributions [16-19]. however, these em clustering algorithms are sensitive to initials and not robust to outliers with a given number of clusters a priori. in this paper, we construct a learning-based schema for em, and then propose a learning-based em algorithm on mixtures of von mises distributions for clustering circular data that is free of initials and robust to outliers. the proposed clustering method can also automatically find an optimal number of clusters. we apply the proposed algorithm to real circular data and several comparisons are given to demonstrate its effectiveness and superiority. 2. learning-based em clustering for circular data in yang et al. [20], they proposed a robust em clustering for gaussian mixture models. the algorithm can automatically find an optimal cluster number of data sets and solve the drawbacks of em that is sensitive to initial values. in this paper, we * corresponding author. e-mail address: msyang@math.cycu.edu.tw proceedings of engineering and technology innovation, vol. 15, 2020, pp. 42 51 43 will modify the approach proposed by yang et al. [20] such that it can handle circular data. we mention that yang et al. [21] used the idea of the robust em clustering on the mixtures of von mises-fisher distributions. we know that von mises-fisher distributions are well used to handle data higher than three dimensions, i. e. data on the unit hypersphere. however, the mixtures of von mises distributions are well for handling circular data. in this paper, we especially focus on circular data based on the mixtures of von mises distributions with angles measured by sine and cosine that is different from yang et al. [21]. suppose that the circular data set 𝑋 = {𝜃1, 𝜃2, ⋯ , 𝜃𝑛} is a random sample of size n from a mixture model 1 ( ; , ) ( ; )       c i i i i f a a f (1) where ai denotes the proportion of the ith class with 1 1   c ii a (2) and 𝑓𝑖 is the pdf of ith class with parameter ∅𝑖 . let class variable 𝑧 = {𝑧1, 𝑧2, ⋯ , 𝜃𝑐 } with 𝑧𝑖 = {𝑧𝑖1, 𝑧𝑖2, ⋯ , 𝑧𝑖𝑛 } 𝑇 where 𝑧𝑖𝑗 = 𝑧𝑖 (𝜃𝑗 ) = 1 if 𝜃𝑗 arises from the ith class and 𝑧𝑖𝑗 = 0 if 𝑥𝑗 arises from other class, for 𝑖 = {1, 2, ⋯ , 𝑐} and 𝑗 = {1, 2, ⋯ , 𝑛}. thus, the joint probability density of the complete data { 𝜃1, 𝜃2, ⋯ , 𝜃𝑛 , 𝑧1, 𝑧2, ⋯ , 𝑧𝑐 } shall be 1 2 1 2 1 1 ( , , , , , , ; , ) [ ( ; )]           ijn c n c j i i i j i z f z z z a a f (3) the log-likelihood for the complete data is given by 1 2 1 2 1 1 ( , ; , , , , , , ) (ln ( ; ))           n c em n c j i ij i i j i l a z z z z a f (4) the em algorithm consists of two steps with “expectation”, shorten by e, and “maximum”, shorten by m. in the e step, the expectation 1 ( ) ( ) / ( )        c ij j i i j i s s s j s e z a f a f (5) is used to substitute for missing data 𝑧𝑖𝑗 . in the m step, estimates of parameters are used by maximizing 1 1 ( ) ( )(ln ( ; ))        c n em i j ij j i i j i e l e z a f (6) with the restriction of eq. (2). the von mises distribution is the most common used distribution on circular data. if we consider that the density 𝑓𝑖 = (𝜃; ∅𝑖 ) is the von mises distribution 𝑉𝑀 = (𝑣𝑖 ; 𝑘𝑖 ) with mean direction 𝑣𝑖 and the concentration 𝑘𝑖 with -1 0 ( ; ) ( ; , )=(2 ( )) exp ( cos( ))      i i i i i i i i f f v k i k k v , 0 2 , 0     　 i k (7) 2 -1 0 0 ( ) (2 ) exp ( cos( )     i i ii k k v d (8) is the modified bessel function of order zero, then update equations for 𝑎𝑖 , 𝑣𝑖 , and 𝑘𝑖 can be obtained as follows: 1 /    n j i ia z j n (9) 11 1 sin( ) tan ( ) cos( )        n j ij n j ij j i j z v z (10) proceedings of engineering and technology innovation, vol. 15, 2020, pp. 42 51 44 11 1 cos( ) ( ) )        n j ij n j ij j i i z v k a z (11) where 𝐴−1(𝑥) a function that can be computed from batschelet’s table (see fisher [3]). in order to solve drawbacks for em, yang et al. [20] give an entropy term of the proportion to the em log-likelihood function. according to their idea, we propose the learning-based em objective function for mixtures of von mises distributions as follows: 1 n 1 11 ( , , , , , , ) (ln ( ; , )) ln          c c c ij i i j i i i j ii j z a v k z a f v k n a (12) thus, in the e step, the expectation �̂�𝑖𝑗 = 𝐸(𝑧𝑖𝑗 |𝜃𝑗 ) is used to substitute for missing data 𝑧𝑖𝑗 with 1 ˆ ( ) ( ; , ) / ( ; , )       c ij ij j i i j i i s s j s s s z e z a f v k a f v k (13) in the m-step, under the constraint eq. (2), we need to maximize 1 n 1 11 ( ( , , , ; , , )) (ln ( ; , )) ln          c c c ij i i j i i i j ii e j z a v k z a f v k n a (14) taking partial derivatives with respect to 𝑎𝑖 , and then the following update equation is obtained ( ) ( ) ( ) ( ) ( ) 1 (ln ln       c new em old old old old i i i i s s s a a a a a a (15) where 1 ˆ /   nem i ijj a z n (16) eq. (15) provides us a method to seek the number of clusters. because the number of clusters is not larger than n, ( ) 1/ new i a n (17) or ( ) 0 new i a (18) is not reasonable. in this situation, we must discard the ith class. thus, the new number of clusters can be obtained with ( ) ( ) ( ) ( ) ( ) | { | 1/ , 1, 2, , }|    new old new new old i i c c a a n i c (19) in order to satisfy the constraints ( ) 1 1 new c kk a   (20) and ( ) 1 ˆ 1 new c k jk z   (21) we adjust 𝑎𝑘′ and �̂�𝑘′𝑗 by proceedings of engineering and technology innovation, vol. 15, 2020, pp. 42 51 45 ( ) 1 = /    new c k k s s a a a (22) ( 1) 1 ˆ ˆ ˆ= /      t c k j k j sj s z z z (23) the purpose of the parameter β is to control competition of proportions. following yang et al. [20], we set β as follows: ( ) ( ) ( ) ( )1 ( ) ( ) ( ) exp (| ) 1 min{ , }      new c new old emi i ci old old c n a a a c a e (24) where ( ) ( ) ( ) max old old c i i a a (25) ( ) max em em c i i a a (26) ( ) ( ) ( ) 1 ln   old c old old i ii e a a (27) when the cluster number c is stable, let 𝛽 = 0. we also use the m-step to estimate 𝑣𝑖 and 𝑘𝑖 . we take partial derivatives with respect to 𝑣𝑖 and 𝑘𝑖 , respectively. then, update equations are obtained as follows: 1 1 1 tan ( sin( ) / cos( ))        n n i ij j ij j j j v z z (28) 1 1 1 ( cos( ) / )       n n i ij j j ij j j k a z v z (29) to solve the initialization problem, we let the initial number of clusters be n. it means that (0) (0) (0) 1 2 1 2 ( , , , ) ( , , , )   n n v v v (30) (0) (0) (0) 1 2 ( , , , ) (1/ , 1/ , , 1/ ) n a a a n n n (31) we then consider the maximum likelihood estimate of the concentration of von mises distribution as initial value 𝑘 𝑖 (0) with (0) 1 ( ), 1, ,    i k a r i c (32) where 2 2 1/ 2 1 1 1 1 [( cos )] ( sin )       n n j j r j j n n (33) is the sample mean resultant length. thus, the proposed learning-based em for circular data can be summarized as follows: learning-based em algorithm for circular data step 1: fix 𝜀 > 0. give initials 𝛽(0) = 1, 𝑐 (0) = 𝑛, 𝑎𝑖 (0) = 1 𝑛 , and assign (𝑣1 (0) , 𝑣2 (0) , ⋯, 𝑣𝑛 (0) )= (𝜃1, 𝜃2, ⋯, 𝜃𝑛). step 2: compute 𝑘𝑖 (0) using eq. (32) and set 𝑡 = 1. step 3: compute �̂�𝑖𝑗 (0) with (0) i a , (0) i  , and (0) i  using eq. (13). proceedings of engineering and technology innovation, vol. 15, 2020, pp. 42 51 46 step 4: compute 𝑣𝑖 (𝑡) with �̂�𝑖1 (𝑡−1) , �̂�𝑖2 (𝑡−1) , ⋯, �̂�𝑖𝑛 (𝑡−1) using eq. (28). step 5: update to 𝑎𝑖 (𝑡) with �̂�𝑖1 (𝑡−1) , �̂�𝑖2 (𝑡−1) , ⋯, �̂�𝑖𝑛 (𝑡−1) and 𝑎𝑖 (𝑡−1) using eq. (15). step 6: compute 𝛽(𝑡) with 𝑎𝑖 (𝑡) and 𝑎𝑖 (𝑡−1) using eq. (24). step 7: update 𝑐 (𝑡−1) to 𝑐 (𝑡) by discard those clusters with 𝑎𝑖 (𝑡) ≤ 1/𝑛 and adjust 𝑎𝑖 (𝑡) and �̂�𝑖𝑗 (𝑡−1) by eqs. (22) and (23). if 𝑡 ≥ 60 and 𝑐(𝑡−60) − 𝑐 (𝑡) = 0, then let 𝛽(𝑡) = 0. step 8: compute �̂�𝑖 (𝑡) with �̂�𝑖1 (𝑡−1) , �̂�𝑖2 (𝑡−1) , ⋯, �̂�𝑖𝑛 (𝑡−1) using eq. (29). step 9: compute �̂�𝑖𝑗 (𝑡) with 𝑎𝑖 (𝑡) , 𝑣𝑖 (𝑡) , and �̂�𝑖 (𝑡) using eq. (13). step 10: compute 𝑣𝑖 (𝑡+1) with �̂�𝑖1 (𝑡) , �̂�𝑖2 (𝑡) , ⋯, �̂�𝑖𝑛 (𝑡) using eq. (28). step 11: compute 𝑑(𝑣 (𝑡+1), 𝑣 (𝑡)) in a convenient norm 𝑑. if 𝑑(𝑣 (𝑡+1), 𝑣 (𝑡)) < 𝜀, stop else 𝑡 = 𝑡 + 1 and return to step 5. 3. examples and comparisons in this section, we make comparisons between the proposed learning-based em and em algorithms on numerical and real data sets. example 1: in fig. 1(a), there is a 2-cluster data set which is generated from a mixture of two von mises distributions 0.4vm(60°, 6.5)+0.6vm(180°, 7.5) with 200 data points. if the angle is an observation, then x-coordinate represents cos and y-coordinate represents sin. we first implement em with c = 2 for this data set. the clustering results of em are shown in fig. 1(b). the data set is well separated, and so we obtain the same clustering results of em with most random initial values. the estimates parameters are as: �̂�1 = 0.4274, �̂�2 = 0.5726, �̂�1 = 61.0254 °, �̂�2 = 178.2695 °, �̂�1 = 6.54, �̂�2 = 7.426. we can see that em has good clustering results and estimates close to the given parameters. we use the learning-based em without a priori cluster number. figs. 1(c)-1(f) are the states of the cluster centers (denoted by “*” symbol) at iteration = 0, 5, 10, 20, respectively. the cluster centers also imply the mean directions. we can see that the cluster number decreases from 200 to 2. therefore, the learning-based em can find an optimal cluster number with c * =2. the clustering results of the learning-based em are the same with em (with a priori cluster number c=2), as shown in fig. 1(b). the parameter estimates of the learning-based em are as: �̂�1 = 0.4257, �̂�2 = 0.5743, �̂�1 = 60.8349 °, �̂�2 = 178.1072 °, �̂�1 = 6.54, �̂�2 = 7.426. the results are closed to those of em. to compare the performance of the proposed learning-based em with em, we use the criterion of mean squared error (mse) to evaluate the accuracy of the two algorithms. we generate 100 data sets from 0.4vm (60°, 6.5)+0.6vm(180°, 7.5) and computer mses from em and the learning-based em. the mse is defined as sum of se of each data set for mse 100   where the squared error (se) for v is defined as 2 2 1 2 ˆ ˆ( 60) ( 180) se 2      from the 100 data sets, we obtain that the mse value from em is 2.3561, and the mse value from the learning-based em is 2.2193. the proposed learning-based em performs better than em. proceedings of engineering and technology innovation, vol. 15, 2020, pp. 42 51 47 (a) 2-clusters data set (b) clustering results of em and the learning-ased em (c) the states of cluster centers from the learning-based em at iteration = 0 (d) the states of cluster centers from the learning-based em at iteration = 5 (e) the states of cluster centers from the learning-based em at iteration = 10 (f) the states of cluster centers from the learning-based em at iteration = 20 fig. 1 2-cluster circular data set with their clustering results furthermore, we consider the influence of noisy points on the proposed learning-based em and em algorithms. we add 50 randomly noisy points which are generated from the interval [0°, 240°] into 2-cluster data set and call noisy data set, as shown in fig. 2(a). the noisy points are denoted by the "+" symbol. we first implement em with c = 2 for the noisy data set. fig. 2(b) shows the clustering results of em. compare fig. 2(b) with fig. 1(b), we find that one of the original 200 data points is classified into another cluster. the estimates parameters are as follows: �̂�1 = 0.4223 , �̂�2 = 0.5477 , �̂�1 = 61.9240 ° , �̂�2 = 178.4338 °, �̂�1 = 4.8590, �̂�2 = 6.5400. we also implement the learning-based em with unknown number c of clusters for the noisy data set. the final states of the cluster centers are shown in fig. 2(c). the learning-based em find the optimal cluster number c * =2 after 51 iterations. the clustering results of the learning-based em are shown in fig. 2(d). compare fig. 2(d) with fig. 1(b), we find that two clusters in the original 200 data points are the same. the estimates parameters are as -1.5 -1 -0.5 0 0.5 1 1.5 -1.5 -1 -0.5 0 0.5 1 1.5 x y -1.5 -1 -0.5 0 0.5 1 1.5 -1.5 -1 -0.5 0 0.5 1 1.5 x y -1.5 -1 -0.5 0 0.5 1 1.5 -1.5 -1 -0.5 0 0.5 1 1.5 x y iteration = 0 -1.5 -1 -0.5 0 0.5 1 1.5 -1.5 -1 -0.5 0 0.5 1 1.5 x y iteration = 5 -1.5 -1 -0.5 0 0.5 1 1.5 -1.5 -1 -0.5 0 0.5 1 1.5 x y iteration = 10 -1.5 -1 -0.5 0 0.5 1 1.5 -1.5 -1 -0.5 0 0.5 1 1.5 x y iteration = 20 (convergence) proceedings of engineering and technology innovation, vol. 15, 2020, pp. 42 51 48 follows: �̂�1 = 0.4490, �̂�2 = 0.5510, �̂�1 = 61.1672 ° , �̂�2 = 178.1286 ° , �̂�1 = 4.8590, �̂�2 = 5.8520. to compare the influences of noisy points on the two algorithms, we compute the increasing rate of mse. we generate 50 randomly noisy points from the interval [0°, 240°] for the data set. then we compute mses for em and the learning-based em from 100 noisy data sets. the increasing rate is defined as (the noisy data set) (the original data set) (the original data set) mse mse increasing rate 100% mse    (34) table 1 shows these mses with its increasing rate. from table 1, we find the increasing rate of the learning-based em is smaller than em, which means that the noisy points have a smaller influence on the learning-based em. (a) noisy data set (b) clustering results of em (c) the final states of cluster centers of the learning-based em (d) clustering results of the learning-based em fig. 2 2-cluster data set with 50 noisy points and their clustering results table 1 mses for em and the learning-based em with its increasing rates algorithms em learning-based em mse(the original data set) 2.3561 2.2193 mse (the noisy data set) 2.8719 2.4621 increasing rate 21.89% 10.94% example 2: in this example, we apply the proposed learning-based em and em algorithms to a real data set, sudden infant death syndrome (sids) data of mooney et al. [17]. the sids data set consists of the sids cases in the uk by a month of death for the years 1983-1998, exhibiting a seasonal pattern with a winter peak. mooney et al. [17] modified all of 12 months to 31 days, i.e. the sids data were month-corrected to 31 days by multiplying february by 1.097 and the 30 day months by 1.033. then they found that the sids data for the year 1998 could be fitted by a mixture of two von misses distributions. table 2 shows the corrected number of sids cases for the year 1998. we transform table 2 to circular data. we map the 12 months from 0 to 360 in degree. every month has equal length (31 days), and so [0, 360] should be divided into 12 equal length intervals. january corresponds to the interval [0°, 30°], february corresponds to the interval [30°, 60°] and so on. based on -1.5 -1 -0.5 0 0.5 1 1.5 -1.5 -1 -0.5 0 0.5 1 1.5 x y -1.5 -1 -0.5 0 0.5 1 1.5 -1.5 -1 -0.5 0 0.5 1 1.5 x y -1.5 -1 -0.5 0 0.5 1 1.5 -1.5 -1 -0.5 0 0.5 1 1.5 x y iteration = 51 (convergence) -1.5 -1 -0.5 0 0.5 1 1.5 -1.5 -1 -0.5 0 0.5 1 1.5 x y proceedings of engineering and technology innovation, vol. 15, 2020, pp. 42 51 49 frequency in each month, we generate random numbers from the uniform distribution on the corresponding interval. 402 observations and its rose diagram (circular histogram) are shown in fig. 3(a) and fig. 3(b), respectively. clearly, we can se e two modes on the intervals [150°, 180°] and [330°, 360°] from fig. 3(b). we implement em with cluster number c=2 for the data set in fig. 3(a). the clustering results of em are affected by different initials. figs. 3(c)-3(e) are results with different initials. if we implement em for this data set with 100 random initials, then we obtain 23 of 100 results as shown in fig. 3(c), 72 of 100 results as shown in fig. 3(d), and 5 of 100 results as shown in fig. 3(e). we also implement the learning-based em without giving c for the data set in fig. 3(a). fig. 3(f) shows the cluster number decreases from 402 to 2. therefore, the learning-based em finds an optimal cluster number c * =2. the final states of cluster centers are shown in fig. 3(g). we can see that the locations of two mean directions are consistent with two modes in fig. 3(b). the clustering results of the learning-based em are shown in fig. 3(h). obviously, the results are different from em. note that the learning-based em sets all data points as initials. thus, an initial condition is not necessary for the learning-based em. if we perform the learning-based em for 100 times, then 100 clustering results are the same. the estimated parameters of the learning-based em, em (with fig. 3(d)) and mooney et al. [17] are shown in table 3. we find the estimates of the learning-based em are closed to mooney et al. [17]. according to mean direction estimates, we know that there are two peaks in sids cases for the year 1998. one peak is in early june and the other is in early december. table 2 corrected number of sids case for the year 1998 jan feb mar apr may jun jul aug sep oct nov dec total 40 31 25 26 29 33 25 20 27 40 43 63 402 table 3 estimates of parameters for sids data algorithms em learning-based em mooney et al. [17] �̂�1 153.32° 151.84° 154.48° �̂�2 337.66° 340.33° 340.52° (a) the sids data set (b) rose diagram to sids data for the year 1998 (c) clustering results of em with different initials (d) clustering results of em with different initials fig. 3 clustering results of learning-based em and em algorithms for sids data set -1.5 -1 -0.5 0 0.5 1 1.5 -1.5 -1 -0.5 0 0.5 1 1.5 x y 20 40 60 80 30 210 60 240 90 270 120 300 150 330 180 0 -1.5 -1 -0.5 0 0.5 1 1.5 -1.5 -1 -0.5 0 0.5 1 1.5 x y -1.5 -1 -0.5 0 0.5 1 1.5 -1.5 -1 -0.5 0 0.5 1 1.5 x y proceedings of engineering and technology innovation, vol. 15, 2020, pp. 42 51 50 (e) clustering results of em with different initials (f) cluster number obtained by the learning-based em under different iterations (g) final states of cluster centers of the learning-based em (h) clustering results of learning-based em fig. 3 clustering results of learning-based em and em algorithms for sids data set (continued) 4. conclusions in general, the em algorithm for clustering circular data quite depends on initializations with a given number of clusters a priori. to solve these drawbacks of em, a learning-based em algorithm is proposed for circular data in this paper. the proposed learning-based em can automatically find an optimal number of clusters for different circular data sets without any initialization and also robust to outliers. some numerical data from mixtures of von mises distributions and real circular data are used for comparisons of the proposed learning-based em with em algorithms. furthermore, we also consider noisy points and outliers in data sets. these comparisons and experimental results actually demonstrate the effectiveness and usefulness of the proposed learning-based em clustering algorithm for circular data. although the proposed learning-based em works well for circular data, it is only fitted for 2-dimensional angular data sets. in our future work, we will further construct a learning-based em algorithm for spherical data sets, i.e. 3-dimensional angular data, and then apply it to extrasolar planet taxonomy. conflicts of interest the authors declare no conflict of interest references [1] r. von mises, “uber die ‘ganzzahligkeit’ der atomgewicht und verwandte fragen,” physikal z., vol. 19, pp. 490-500, 1918. [2] g. s. watson and e. j. williams, “on the construction of significance tests on the circle and the sphere,” biometrika, vol. 43, no. 3/4, pp. 344-352, december 1956. [3] n. i. fisher, statistical analysis of circular data, cambridge:cambridge university, in press, october 1995. -1.5 -1 -0.5 0 0.5 1 1.5 -1.5 -1 -0.5 0 0.5 1 1.5 x y -1.5 -1 -0.5 0 0.5 1 1.5 -1.5 -1 -0.5 0 0.5 1 1.5 x y iteration = 39 (convergence) -1.5 -1 -0.5 0 0.5 1 1.5 -1.5 -1 -0.5 0 0.5 1 1.5 x y 0 5 10 15 20 25 30 35 40 2 50 100 150 200 250 300 350 400 450 iteration c lu st e r n u m b e r proceedings of engineering and technology innovation, vol. 15, 2020, pp. 42 51 51 [4] n. masseran, a. m. razali, k. ibrahim, and m. t. latif, “fitting a mixture of von mises distributions in order to model data on wind direction in peninsular malaysia,” energy conversion and management, vol. 72, pp. 94-102, april 2013. [5] l. p. rivest and s. kato, “a random‐effects model for clustered circular data,” canadian journal of statistics, vol. 47, no. 4, pp. 712-728, august 2019. [6] a. p. dempster, n. m. laird, and d. b. rubin, “maximum likelihood from incomplete data via the em algorithm,” journal of the royal statistical society, series b, vol. 39, no. 1, pp. 1-22, september 1977. [7] g. j. mclachlan and k. e. basford, mixture models: inference and applications to clustering, new york: marcel dekker, 1988. [8] j. yu, c. chaomurilige, and m. s. yang, “on convergence and parameter selection of the em and da-em algorithms for gaussian mixtures,” pattern recognition, vol. 77, pp. 188-203, december 2017. [9] j. macqueen, “some methods for classification and analysis of multivariate observations,” proc. 5th berkeley symposium on mathematical statistics and probability, university of california, in press, vol. 1, no. 14, pp. 281-297, june 1967. [10] d. pollard, “quantization and the method of k-means,” ieee transaction on information theory, vol. 28, no. 2, pp. 199-205, march 1982. [11] m. s. yang and k. p. sinaga, “a feature-reduction multi-view k-means clustering algorithm,” ieee access, vol. 7, pp. 114472-114486, august 2019. [12] j. c. bezdek, pattern recognition with fuzzy objective function algorithms, new york: plenum, in press, july 1981. [13] m. s. yang and y. nataliani, “a feature-reduction fuzzy clustering algorithm with feature-weighted entropy,” ieee transactions on fuzzy systems, vol. 26, no. 2, pp 817-835, april 2018. [14] r. krishnapuram and j. m. keller, “a possibilistic approach to clustering,” ieee transaction on fuzzy systems, vol. 1, no. 2, pp. 98-110, may 1993. [15] m. s. yang, s. j. chang-chien, and y. nataliani, “a fully-unsupervised possibilistic c-means clustering method,” ieee access, vol. 6, pp. 78308-78320, december 2018. [16] r. bartels, “estimation in a bidirectional mixture of von mises distributions,” biometrics, vol. 40, no. 3, pp. 777-784, september 1984. [17] j. a. mooney, p. j. helms, and i. t. jolliffe, “fitting mixtures of von mises distributions: a case study involving sudden infant death syndrome,” computational statistics and data analysis, vol. 41, no. 3/4, pp. 505-513, october 2002. [18] n. sanusi, a. zaharim, s. mat, and k. sopian, “a weibull and finite mixture of the von mises distribution for wind analysis in mersing, malaysia,” international journal of green energy, vol. 14, no. 12, pp. 1057-1062, september 2017. [19] y. ban, x. alameda-pineda, c. evers, and r. horaud, “tracking multiple audio sources with the von mises distribution and variational em,” ieee signal processing letters, vol. 26, no. 6, pp. 798-802, march 2019. [20] m. s. yang, c. y. lai, and c. y. lin, “a robust em clustering algorithm for gaussian mixture models,” pattern recognition, vol. 45, no. 11, pp. 3950-3961, may 2012. [21] m. s. yang, s. j. chang-chien, and w. l. hung, “learning-based em clustering for data on the unit hypersphere with application to exoplanet data,” applied soft computing, vol. 60, pp. 101-114, june 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/). 1___peti#8420__01-11 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 01-11 a method based on only currents for determining fault direction in radial distribution networks integrated with distributed generations ngo minh khoa * , tran xuan khoa faculty of engineering and technology, quy nhon university, binh dinh, vietnam received 05 september 2021; received in revised form 12 december 2021; accepted 13 december 2021 doi: https://doi.org/10.46604/peti.2021.8420 abstract nowadays, more distributed generations (dgs) are connected to a radial distribution network, so conventional overcurrent relays cannot operate correctly when a fault occurs in the network. this study proposes a method to determine the fault direction in a three-phase distribution network integrated with dgs. the obtained pre-fault and fault currents are utilized to extract their phasors by the fast fourier transform, and the phase angle difference between the positive-sequence components of the pre-fault and fault currents is used. moreover, the method only uses the local current measurement to calculate and identify the phase angle change of the fault current without using the voltage measurement. matlab/simulink software is used to simulate the three-phase distribution network integrated with dgs. the faults with different resistances are assumed to occur at backward and forward fault locations. the simulation results show that the proposed method correctly determines the fault direction. keywords: directional protection relay, distributed generation, fault direction, positive-sequence current, radial distribution network 1. introduction integrating distributed generations (dgs) into a radial distribution network has many significant benefits, such as reducing the total power loss, improving the voltage quality, etc. however, there are some existing problems on a radial distribution network integrated with dgs, i.e., the power flow, control, operation, reliability, security, and protection problem of the network [1-2]. an adaptive quantum-inspired evolutionary algorithm was then proposed to improve the power flow and voltage profile in the distribution network integrated with dgs [3-6]. in addition, the protection problem of the distribution network integrated with dgs is studied in many works [7-10]. conventional overcurrent protection relays are usually used to protect a radial distribution network. many researchers only use the current measurement at the relays via current transformers (cts) to operate when the current measurement exceeds the pickup values set in the relays. because the power flow of the distribution network integrated with dgs is changed depending on the penetration level of the network sources, the conventional overcurrent protection relays will not correctly determine the fault direction when a fault occurs in the network [7]. in horak’s work [8], a scheme of directional overcurrent relays was designed based on the phase relationship of voltages and currents at the relay location to determine the fault direction. in the work of jang et al. [9], an adaptive approach for relay protection was applied to a distribution network integrated with a wind farm. the method proposed in eissa’s work [10] * corresponding author. e-mail address: ngominhkhoa@qnu.edu.vn tel.: +84 988 371 737 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 01-11 2 utilized a novel current polarized directional element technique to determine the fault direction on a transmission line. to completely overcome the protection problem, the directional overcurrent protection relays are applied to ensure the selection factor of the relays in the situation. for the directional overcurrent protection relays, the direction of the fault current flowing through the relay location is determined by using the phase angle of voltage and current. the voltage polarity is used as a reference value for determining the fault direction [11]. therefore, voltage transformers are utilized to transfer the high voltage on the primary side to the low voltage on the secondary side and to feed the protection relays. similarly, cts are simultaneously utilized to feed the small-scale currents to the relays. in general, when a short circuit occurs in the network, the fault current phasors at the relays are usually located in two distinct areas: the forward location and the backward location, as shown in fig. 1. this study will exploit the information from these two areas to develop an innovative method for determining the fault direction in a radial distribution network integrated with dgs. i iforward u ibackward fig. 1 two fault identification areas for directional overcurrent protection relays motivated by the above-mentioned works, an innovative method is developed in this study for determining the fault direction in a radial distribution network integrated with dgs. in summary, the main contributions of this study cover three aspects: (i) the proposed method only uses the current measurement at the relay location to determine the phase angle difference between the positive components of the pre-fault and fault currents; (ii) the proposed method is embedded in a relay to detect and protect all types of faults that occur in a radial distribution network integrated with dgs, including phase-to-ground (lg) faults, phase-to-phase (ll) faults, two-phase-to-ground (llg) faults, and three-phase-to-ground (lllg) faults; and (iii) the proposed method is more cost-effective for investing in voltage transformers, as compared to the conventional methods. the rest of this study is organized as follows. section 2 presents the literature review. section 3 describes the background methodology of the proposed method for determining the fault direction in a radial distribution network integrated with dgs. the simulation results and discussion are contained in section 4, and finally section 5 concludes this study. 2. literature review several published works are related to the methods applied to directional overcurrent protection relays in power systems [12-18]. voima et al. [12] presented an adaptive protection scheme to protect the medium voltage networks integrated with dgs, particularly in island operation modes. ukil et al. [13] proposed a novel approach to detect the possibility of fault direction using only the currents at the relays. yousfi et al. [14] developed a method based on the adaline neural network and the instantaneous power theory for extracting the online symmetrical components and phase angle from the fault current. eissa [15] proposed a new technique for directional overcurrent protection based on the post-fault current signals and directional reference current signals. the voltage measurement at the relays did not require determining the fault direction in the technique. proceedings of engineering and technology innovation, vol. 20, 2022, pp. 01-11 3 furthermore, samet et al. [16] developed a high-speed algorithm for determining the fault current, which only used the current measurement at the relays. in the algorithm, the sign of the summation of multiplied faults by the samples of pre-fault current and the direction of power flow in a normal power system was a criterion to determine the fault direction. the effectiveness of the method was shown by the speed for determining the fault direction in less than one-eighth of the cycle of power frequency. samet et al. [16] also proposed a novel directional overcurrent protection scheme for the distribution networks integrated with dgs. this protection scheme calculated the fault direction using a micro-genetic algorithm through numerical relays which were located on the network to protect and detect any changes in the configuration as well as recalculate the setting of directional overcurrent protection relays. on the other hand, to find the optimum relay setting for the minimum time to interrupt the power supply, nascimento et al. [17] and khond et al. [18] developed a new technique using the linear programming problem approach to optimize the relay setting in distribution networks. the power flow in the distribution networks with the penetration of dgs is changed depending on the level of penetration. the adaptive and flexible algorithms for directional overcurrent protection relays were mentioned in many publications [19-25]. brahma et al. [19] developed an adaptive protection scheme applied to directional overcurrent relays to protect the distribution networks with dgs. balyith et al. [20] proposed another novel protection scheme without the need of communication assistance to determine the relay setting to minimize the relays’ overall operating time. zhan et al. [21] proposed a genetic algorithm for the location and sizing optimization of dgs in distribution networks by investigating their relay protection coordination. another adaptive overcurrent coordination scheme based on the evolution algorithm was developed in the work of shih et al. [22] to enhance the relay sensitivity and overcome the drawbacks of dgs. jia et al. [23] presented an improved scheme based on high-frequency impedance to manage the adaptability problem when determining the fault direction in the network with inverter-interfaced renewable energy generations. to overcome the challenges of overcurrent protection in the distribution networks integrated with dgs, the local measurement information was used to detect the operating status and the faulted section in the network [24]. in the work of jones et al. [25], directional overcurrent relays were used to solve difficult problems for distribution feeder protection with high penetration of dgs. concerning the protection issue, ieee standard 1547-2003 [26] presents the criteria and requirements for the interconnection of dgs with power systems, and ieee standard p1547.4 [27] provides alternative approaches and good practices for the design, operation, and integration of dg island systems with power networks. therefore, these two ieee standards are considered in this work. 3. proposed method for determining fault direction 3.1. single-phase network the fault direction in a radial distribution network integrated with dgs can be determined by the proposed method, which uses the phase angle difference between the pre-fault and fault currents at the protection relay location. the method is developed based on the phase angle change of the fault current compared with the pre-fault current, as shown in fig. 2. the dg at the busbar a is connected to the grid source via two segments: line 1 (from the busbar a to the busbar b) and line 2 (from the busbar b to the busbar c). it is assumed that dg is generating the power to supply a load located at the busbar b and transferring it to the grid source via line 2. therefore, the power flow direction in the distribution network is normally from the busbar a to the busbar c. at the end b of line 2 (from the busbar b to the busbar c), a protection relay is established to get the current from ct. the proposed method is applied to the relay to determine the fault direction when a fault occurs in the lines. two fault locations, including the backward fault location (f1) on line 1 and the forward fault location (f2) on line 2, are investigated in the network as shown in fig. 2. proceedings of engineering and technology innovation, vol. 20, 2022, pp. 01-11 4 r relay zdg udg f1 f2 zgrid ugrid z11 z12 z21 z22 a b c ct i dg source grid source line 1 z1=z11+z12 line 2 z2=z21+z22 qdg pdg load cb2cb1 fig. 2 power network integrated with dgs when a fault occurs at f1 and f2, the pre-fault current and the fault current are calculated as follows. the pre-fault current at the relay location is: a c u u i z − = (1) where ua and uc are the voltages at the busbar a and the busbar c, respectively; z = z1 + z2 is the impedance from the busbar a to the busbar c. when an lllg fault occurs at f1, the fault current flow from the grid side to f1 is given as follows: c f1 f1 u i z = (2) where zf1 = z2 + z12 is the impedance from the busbar c to f1. similarly, when an lllg fault occurs at f2, the fault current flow from dg to f2 is given as follows: a f2 f2 u i z = (3) where zf2 = z1 + z21 is the impedance from the busbar a to f2. when a fault occurs at f1 and f2, the fault current flow through ct is determined as follows: 1 f1 i i i= − (4) substituting eqs. (1) and (2) into eq. (4), the following equations are established: a c c 1 f1 u u u i z z − = − (5) 2 f2 i i i= + (6) substituting eqs. (1) and (3) into eq. (6), the following equation is established: a c a 2 f2 u u u i z z − = + (7) it is assumed that the voltage at the busbar a (ua) and the voltage at the busbar c (uc) have the same voltage magnitude and phase angle. f1 and f2 are assumed at the busbar a and the busbar c, respectively. the fault impedances zf1 and zf2 are equal to the impedance of the network z. from eqs. (5) and (7), it is obvious that the currents i1 and i2 are opposite phase angles. in general, the relationship between these currents is depicted in fig. 3. proceedings of engineering and technology innovation, vol. 20, 2022, pp. 01-11 5 i1 ua uc i2if2if1 i ∆ϕ1 ∆ϕ2 note: ∆ϕ1 = ϕ1 – ϕ > 0 ∆ϕ2 = ϕ2 – ϕ < 0 -if1 fig. 3 vector diagram representing the relationship among the currents i1, i2, and i as can be seen in fig. 3, the phase angle difference between the pre-fault current and the fault current is positive (∆φ1 > 0) when a fault occurs at f1. reversely, the phase angle difference between the pre-fault current and the fault current is negative (∆φ2 < 0) when a fault occurs at f2 (∆φ2 < 0). 1 1 0ϕ ϕ ϕ∆ = − > (8) 2 2 0ϕ ϕ ϕ∆ = − < (9) 3.2. three-phase network for a three-phase power network, four types of faults (i.e., lg, ll, llg, and lllg faults) can appear in the network [28-30]. therefore, to perform fault analysis in the network as shown in fig. 2, a positive equivalent rule is used to calculate the fault current flow at f1 and f2. the positive-sequence equivalent diagrams for the fault locations at f1 and f2 are illustrated in figs. 4(a) and (b), respectively, where the additional impedance z∆ depends on the fault types shown in table 1. the proposed method for determining the fault direction in the distribution network integrated with dgs can be described as follows: step 1: faults are detected by comparing the fault current with the pickup current at the relay. step 2: the pre-fault and fault currents at the protection relay location are acquired. step 3: the phasors of the pre-fault and fault currents are estimated by using the fast fourier transform. step 4: the positive-sequence current components are computed from the phasors. step 5: the phase angle difference is calculated. step 6: the fault direction is determined according to the positive equivalent rule. zdg udg f1 zgr id ugr id z11 z12 z21 z22 a b ci pos z∆ ct zdg udg f2 zgr id ugr id z11 z12 z21 z22 a b c z∆ ct i pos (a) for the fault at f1 (b) for the fault at f2 fig. 4 positive-sequence equivalent circuit to calculate fault types proceedings of engineering and technology innovation, vol. 20, 2022, pp. 01-11 6 table 1 additional impedance of four fault types fault location fault type additional impedance z∆ f1 lg ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) zero zero zero zero zero neg neg neg neg neg dg 11 12 2 grid dg 11 12 2 grid zero zero zero zero zero neg neg neg neg neg dg 11 12 2 grid dg 11 12 2 grid z z z z z z z z z z z z z z z z z z z z z ∆ + + + + + + = + + + + + + + + + ll ( ) ( ) ( ) ( ) neg neg neg neg neg dg 11 12 2 grid neg neg neg neg neg dg 11 12 2 grid z z z z z z z z z z z ∆ + + + = + + + + llg ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) zero zero zero zero zero neg neg neg neg neg dg 11 12 2 grid dg 11 12 2 grid zero zero zero zero zero neg neg neg neg neg dg 11 12 2 grid dg 11 12 2 grid zero zero zero zero zero dg 11 12 2 grid zero dg 1 z z z z z z z z z z z z z z z z z z z z z z z z z z z z ∆ + + + + + + + + + + + + + + = + + + +( ) ( ) ( ) ( ) ( ) ( ) neg neg neg neg neg dg 11 12 2 grid zero zero zero zero neg neg neg neg neg 1 12 2 grid dg 11 12 2 grid z z z z z z z z z z z z z + + + + + + + + + + + lllg 0z ∆ = f2 lg ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) zero zero zero zero zero neg neg neg neg neg dg 1 21 22 grid dg 1 21 22 grid zero zero zero zero zero neg neg neg neg neg dg 1 21 22 grid dg 1 21 22 grid z z z z z z z z z z z z z z z z z z z z z ∆ + + + + + + = + + + + + + + + + ll ( ) ( ) ( ) ( ) neg neg neg neg neg dg 1 21 22 grid neg neg neg neg neg dg 1 21 22 grid z z z z z z z z z z z ∆ + + + = + + + + llg ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) zero zero zero zero zero neg neg neg neg neg dg 1 21 22 grid dg 1 21 22 grid zero zero zero zero zero neg neg neg neg neg dg 1 21 22 grid dg 1 21 22 grid zero zero zero zero zero dg 1 21 22 grid zero dg 1 z z z z z z z z z z z z z z z z z z z z z z z z z z z z ∆ + + + + + + + + + + + + + + = + + + +( ) ( ) ( ) ( ) ( ) ( ) neg neg neg neg neg dg 1 21 22 grid zero zero zero zero neg neg neg neg neg 21 22 grid dg 1 21 22 grid z z z z z z z z z z z z z + + + + + + + + + + + lllg 0z ∆ = 4. simulation results to evaluate the effectiveness of the proposed method, the three-phase power network presented in fig. 2 is simulated using the matlab/simulink software. the parameters of the power network elements are given in table 2. the nominal frequency of the network is 50 hz, and the nominal voltage is 22 kv. four types of faults, including lg, ll, llg, and lllg faults, are emulated at f1 and f2, respectively. it is assumed that each fault is started at the time of 0.1 seconds, and the total simulation time is set at 0.3 seconds. besides, the fault resistance at the fault locations is established at different values to analyze the performance of the proposed method. the simulation results of the fault current at the relay location are performed in the time domain. table 2 parameters of the three-phase power network element parameters dg source the frequency f = 50 hz the voltage u = 1.05 × 22 kv the short-circuit power sn = 300 mva the positive-sequence impedance z1 = 0.1613 + j0.0051 ω the zero-sequence impedance z0 = 0.4840 + j0.0154 ω grid source the frequency f = 50 hz the voltage u = 1.05 × 22 kv the short-circuit power sn = 500 mva the positive-sequence impedance z1 = 0.0968 + j0.0031 ω the zero-sequence impedance z0 = 0.2904 + j0.0092 ω line ab the line length l = 25 km the positive-sequence impedance z1 = 0.1153 + 0.3299 ω/km the zero-sequence impedance z0 = 0.413 + 1.0430 ω/km line bc the line length l = 25 km the positive-sequence impedance z1 = 0.1153 + 0.3299 ω/km the zero-sequence impedance z0 = 0.413 + 1.0430 ω/km load the nominal voltage u = 22 kv the power rating p = 10 mw the power factor pf = 1.0 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 01-11 7 the fault current in a circuit is changed from the original values in a steady-state operating mode to the faulty ones. thus, to carry out the fault simulations in this section, the steady-state voltages and currents of the network are calculated by the power flow of the network. it is assumed that dg is generating the power to the grid source via line 1 and line 2. the steady-state results of the distribution network are shown in table 3. it is obvious that the voltages at the three busbars are 1.042 pu, 0.9436 pu, and 1.043 pu, respectively. the current measurement at the overcurrent relay at the busbar b is the magnitude of 0.3599 pu and the phase angle of 42.39°. after establishing the initial steady-state values as shown in table 3, fault simulations are carried out to verify the proposed method above. the simulation results for the two fault locations (f1 on line 1 and f2 on line 2) are simulated and analyzed as follows. table 3 steady-state results of the distribution network bus name voltage current ua (pu) ub (pu) uc (pu) ia (pu) ib (pu) ic (pu) busbar a 1.042∠-2.028° 1.042∠-122° 1.042∠118° 1.132∠-7.822° 1.132∠-127.8° 1.132∠-112.2° busbar b 0.9436∠-25.1° 0.9436∠-145.1° 0.9436∠-94.9° 0.3599∠42.39° 0.3599∠-77.61° 0.3599∠162.4° busbar c 1.043∠-29.84° 1.043∠-149.8° 1.043∠90.16° 0.3563∠42.19° 0.3563∠-77.81° 0.3563∠162.2° the simulation results of the phase angle difference for f1 on line 1 are illustrated in fig. 5. figs. 5(a), (b), (c), and (d) show the results of lg faults, ll faults, llg faults, and lllg faults, respectively. for each case, five fault resistances (0, 5, 10, 15, and 20 ω) are set at the fault locations for evaluating the phase angle difference between the positive-sequence of the pre-fault current and the fault current at the relay location. it is obvious that the phase angle difference is dramatically increased from the value of 0 degrees at the time t = 0.1 seconds. after that, it reaches a positive value. this information is used to determine the backward fault direction. for f2 on line 2, the phase angle difference between the positive-sequence components of the pre-fault current and the fault current is shown in fig. 6. in this case study, five fault resistances (0, 5, 10, 15, and 20 ω) are also set at the fault locations. all the case studies are simulated in matlab/simulink and the results are shown in figs. 6(a), (b), (c), and (d) for the lg faults, ll faults, llg faults, and lllg faults, respectively. as can be seen in fig. 6, the phase angle difference is dramatically decreased at the time t = 0.1 seconds. it then reaches a negative value. therefore, the fault direction of f2 is the forward fault direction. in addition, the fault resistance at the two fault locations is varied from 0 to 20 ω with a step of 1 ω in each case study to evaluate the effectiveness of the proposed method. for each fault resistance, the apparent impedance from the grid source and the dg source to the fault location is also changed in both the magnitude and phase angle; these simulation results are shown in fig. 7. as can be seen in fig. 7, the x-axis and y-axis show the real and imaginary components of the fault current at the relay location, respectively. these simulation results show clearly that the red and blue nodes in fig. 7 represent the forward and backward faults, respectively. (a) lg faults (b) ll faults fig. 5 phase angle difference when faults occur at f1 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 01-11 8 (c) llg faults (d) lllg faults fig. 5 phase angle difference when faults occur at f1 (continued) (a) lg faults (b) ll faults (c) llg faults (d) lllg faults fig. 6 phase angle difference when faults occur at f2 fig. 7 two distinct areas for the backward faults and the forward faults 5. discussion in this study, a method is developed based only on the currents at the relay location for determining the fault direction in a radial distribution network integrated with dgs. a typical radial distribution network is modeled and simulated in the matlab/simulink software. the power flow of steady state in the network is the necessary initial condition to determine the proceedings of engineering and technology innovation, vol. 20, 2022, pp. 01-11 9 fault direction when faults occur in the network. two fault locations, including the backward and forward locations, are considered to confirm the capability of the proposed method. for each case study, the positive-sequence components of the pre-fault and fault currents are extracted using the fast fourier transform, and then the phase angle difference is calculated to determine the fault direction. this work also establishes four types of faults, including lg faults, ll faults, llg faults, and lllg faults, as well as the different fault resistances ranging from 0 to 20 ω. 6. conclusions this study proposes an innovative method for determining the fault direction in a radial distribution network integrated with dgs. the fast fourier transform is applied to extract the phasors of the pre-fault and fault currents at the relay location. the proposed method determines the fault direction based on the phase angle difference between the positive-sequence components of the pre-fault and fault currents at the relay location. the analysis results confirm that the phase angle difference is positive for the faults in the backward direction and negative for the faults in the forward direction. the effectiveness of this method is verified by performing the simulation of a three-phase radial distribution network integrated with dgs. four types of faults with different fault resistances and locations are simulated to evaluate the method. the simulation results confirm that the protection relay applied by this method determines the fault direction correctly. furthermore, because the proposed method only requires the local current measurement without the voltage measurement, it can be easily implemented in conventional non-directional overcurrent relays. the directional overcurrent relays applied by the proposed method can be utilized for future smart grids, displacing the traditional directional overcurrent relays that utilize the reference voltage phasors for estimating the fault direction. conflicts of interest the authors declare no conflict of interest. nomenclature ����� voltage of the grid source ��� �� zero-sequence impedance of the segment from f1 to b ����� impedance of the grid source ��� impedance of the segment from b to f2 � ���� � � negative-sequence impedance of the grid source � �� � � negative-sequence impedance of the segment from b to f2 ����� �� zero-sequence impedance of the grid source ��� �� zero-sequence impedance of the segment from b to f2 ��� voltage of the dg source ��� impedance of the segment from f2 to c �� active power of the dg source ��� � � negative-sequence impedance of the segment from f2 to c ��� reactive power of the dg source ��� �� zero-sequence impedance of the segment from f2 to c ��� impedance of the dg source �△ additional impedance � �� � � negative-sequence impedance of the dg source �� phase-a voltage ��� �� zero-sequence impedance of the dg source �� phase-b voltage ��� impedance of the segment from a to f1 �� phase-c voltage ��� � � negative-sequence impedance of the segment from a to f1 �� phase-a current ��� �� zero-sequence impedance of the segment from a to f1 �� phase-b current ��� impedance of the segment from f1 to b �� phase-c current � �� � � negative-sequence impedance of the segment from f1 to b references [1] w. a. elmore, protective relaying: theory and applications, 2nd ed., new york: marcel dekker, 2003. [2] k. kauhaniemi and l. kumpulainen, “impact of distributed generation on the protection of distribution networks,” 8th iee international conference on developments in power system protection, pp. 315-318, april 2004. proceedings of engineering and technology innovation, vol. 20, 2022, pp. 01-11 10 [3] g. manikanta, a. mani, h. p. singh, and d. k. chaturvedi, “simultaneous application of distributed generator and network reconfiguration for power loss reduction using an adaptive quantum inspired evolutionary algorithm,” international journal of energy technology and policy, vol. 17, no. 2, pp. 140-179, april 2021. [4] g. manikanta, a. mani, h. p. singh, and d. k. chaturvedi, “effect of voltage dependent load model on placement and sizing of distributed generator in large scale distribution system,” majlesi journal of electrical engineering, vol. 14, no. 4, pp. 97-121, december 2020. [5] g. manikanta, a. mani, h. p. singh, and d. k. chaturvedi, “adaptive quantum-inspired evolutionary algorithm for optimizing power losses by dynamic load allocation on distributed generators,” serbian journal of electrical engineering, vol. 16, no. 3, pp. 325-357, 2019. [6] g. manikanta, a. mani, h. p. singh, and d. k. chaturvedi, “placing distributed generators in distribution system using adaptive quantum inspired evolutionary algorithm,” 2nd international conference on research in computational intelligence and communication networks, pp. 157-162, september 2016. [7] p. b. eriksen, t. ackermann, h. abildgaard, p. smith, w. winter, and j. r. garcia, “system operation with high wind generation,” ieee power and energy magazine, vol. 3, no. 6, pp. 65-74, october 2005. [8] j. horak, “directional overcurrent relaying (67) concepts,” 59th annual conference for protective relay engineers, pp. 1-8, april 2006. [9] s. i. jang, j. h. choi, j. w. kim, and d. m. choi, “an adaptive relaying for the protection of a wind farm interconnected with distribution networks,” ieee pes transmission and distribution conference and exposition, pp. 296-302, september 2003. [10] m. m. eissa, “evaluation of a new current directional protection technique using field data,” ieee transactions on power delivery, vol. 20, no. 2, pp. 566-572, april 2005. [11] r. v. r. de carvalho, f. h. t. vieira, s. g. de araújo, and c. r. lima, “a protection coordination scheme for smart grid based distribution systems using wavelet based fault location and communication support,” ieee pes conference on innovative smart grid technologies, pp. 1-8, april 2013. [12] s. voima, h. laaksonen, and k. kauhaniemi, “adaptive protection scheme for smart grids,” 12th iet international conference on developments in power system protection, pp. 1-6, april 2014. [13] a. ukil, b. deck, and v. h. shah, “smart distribution protection using current-only directional overcurrent relay,” ieee pes innovative smart grid technologies conference europe, pp. 1-7, october 2010. [14] f. l. yousfi, d. o. abdeslam, t. bouthiba, n. k. nguyen, and j. merckle, “adaline for online symmetrical components and phase-angles identification in transmission lines,” ieee transactions on power delivery, vol. 27, no. 3, pp. 1134-1143, june 2012. [15] m. m. eissa, “current directional protection technique based on polarizing current,” international journal of electrical power and energy systems, vol. 44, no. 1, pp. 488-494, january 2013. [16] h. samet, t. ghanbari, m. a. jarrahi, and h. j. ashtiani, “efficient current-based directional relay algorithm,” ieee systems journal, vol. 13, no. 2, pp. 1262-1272, september 2018. [17] j. p. nascimento, n. s. d. brito, and b. a. souza, “an adaptive overcurrent protection system applied to distribution systems,” computers and electrical engineering, vol. 81, 106545, january 2020. [18] s. v. khond and g. a. dhomane, “optimum coordination of directional overcurrent relays for combined overhead/cable distribution system with linear programming technique,” protection and control of modern power systems, vol. 4, no. 1, pp. 1-7, april 2019. [19] s. m. brahma and a. a. girgis, “development of adaptive protection scheme for distribution systems with high penetration of distributed generation,” ieee transactions on power delivery, vol. 19, no. 1, pp. 56-63, january 2004. [20] a. a. balyith, h. m. sharaf, m. shaaban, e. f. el-saadany, and h. h. zeineldin, “non-communication based time-current-voltage dual setting directional overcurrent protection for radial distribution systems with dg,” ieee access, vol. 8, pp. 190572-190581, october 2020. [21] h. zhan, c. wang, y. wang, x. yang, x. zhang, c. wu, et al., “relay protection coordination integrated optimal placement and sizing of distributed generation sources in distribution networks,” ieee transactions on smart grid, vol. 7, no. 1, pp. 55-65, january 2016. [22] m. y. shih, a. conde, z. leonowicz, and l. martirano, “an adaptive overcurrent coordination scheme to improve relay sensitivity and overcome drawbacks due to distributed generation in smart grids,” ieee transactions on industry applications, vol. 53, no. 6, pp. 5217-5228, december 2017. [23] k. jia, z. yang, y. fang, t. bi, and m. sumner, “influence of inverter-interfaced renewable energy generators on directional relay and an improved scheme,” ieee transactions on power electronics, vol. 34, no. 12, pp. 11843-11855, december 2019. proceedings of engineering and technology innovation, vol. 20, 2022, pp. 01-11 11 [24] p. mahat, z. chen, b. bak-jensen, and c. l. bak, “a simple adaptive overcurrent protection of distribution systems with distributed generation,” ieee transactions on smart grid, vol. 2, no. 3, pp. 428-437, september 2011. [25] d. jones and j. j. kumm, “future distribution feeder protection using directional overcurrent elements,” ieee transactions on industry applications, vol. 50, no. 2, pp. 1385-1390, april 2014. [26] interconnecting distributed resources with electric power systems, ieee standard 1547-2003, 2011. [27] draft guide for design, operation, and integration of distributed resource island systems with electric power systems ieee standard p1547.4, 2011. [28] ieee recommended practice for conducting short-circuit studies and analysis of industrial and commercial power systems, ieee standard 3002.3-2018, 2019. [29] n. m. khoa, n. a. toan, and d. d. tung, “experimental study on fault ride-through capability of vsc-based hvdc transmission system,” archives of electrical engineering, vol. 70, no. 1, pp. 173-188, march 2021. [30] n. m. khoa and d. d. tung, “locating fault on transmission line with static var compensator based on phasor measurement unit,” energies, vol. 11, no. 9, 2380, september 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/). 5___peti#8556_in press proceedings of engineering and technology innovation, vol. 21, 2022, pp. 34-43 modified digital correlation technique for accurate phase measurement in multi-frequency bio-impedance analysis sruthi s * , rasika dhavse, jignesh sarvaiya department of electronics engineering, sardar vallabhbhai patel national institute of technology, gujarat, india received 25 september 2021; received in revised form 05 december 2021; accepted 06 december 2021 doi: https://doi.org/10.46604/peti.2021.8556 abstract in bio-impedance analysis (bia), high-frequency low-amplitude alternating current (ac) signals can incur time delays due to the capacitive nature of human cell membranes, and the characteristics of human tissues can be assessed from these delays in terms of phase changes. to accurately measure the phase changes, this work proposes a modified digital correlation-based phase measurement method. the accuracy of the general correlation technique is improved through digital direct synthesis (dds) and digital correlation of unipolar square input signals. the proposed method is established through memory management and frequency adjustment. the result shows that, compared to the existing methods, the proposed method needs fewer hardware components, has better accuracy of 0.2° and higher frequency compatibility from 5 khz to 1 mhz, and requires lower cost (140 usd). the method can be applied for the bia of all types of tissues (recently used in covid detection and care) and for the applications where efficient phase measurement is required. keywords: phase measurement, bio-impedance analysis, direct digital synthesis, unipolar square signal, digital correlation 1. introduction time delays can occur when signals pass through different media. these delays are depicted as the phase differences of the propagated signals with respect to the original signals generated from sources. the measurement of phase differences can efficiently help assess the characteristics of the media. thus, the delays between test signals and reference signals need to be determined precisely [1]. interferometry, phase doppler anemometry (pda) and ultrasound imaging, signal demodulation of bio-signals, surface acoustic wave (saw) sensing, coriolis mass flowmetry (cmf), space systems, etc. are some of the applications which demand a precise measurement of phase differences. a phase measurement system should be analyzed based on its accuracy, sensitivity, simplicity, portability, response time, operating frequency, and noise immunity. in the case where synchronous detection/frequency synchronization is to be carried out to test the reliability of phase measurement techniques, the test signals and reference signals are generated from the same source. in bio-impedance analysis (bia), high-frequency low-intensity sinusoidal alternating current (ac) signals are applied to the tissue under test (tut) through electrodes by a drive module, and the voltage response of tut is measured by a sense module through electrodes, as shown in fig. 1. the signal used for excitation suffers phase alteration while traveling through tut. this is caused due to the capacitive nature of cell membranes of the cells constituting tissues. the signal used for excitation is an ultra-weak signal which also gets affected by additive noise sources like motion artifacts, other bio-signals, etc. [2]. it is essential to employ a precise ac constant source to excite tissues [3]. for a very weak signal, the detection system should be highly sensitive and accurate. three dispersion mechanisms (α, β, and γ) are used to characterize the anomalous electric properties of * corresponding author. e-mail address:sruthisindu@gmail.com tel.: +917069778270 proceedings of engineering and technology innovation, vol. 21, 2022, pp. 34-43 bio-materials when subjected to the external ac signals of low frequency to very high frequency. due to the capacitive nature of cell membranes, the tissue characteristics including intracellular, extracellular, and membranous properties are obtained only in the β dispersion range (5 khz to 1 mhz) [2, 4]. recently, the bia during covid care is a hot topic of research [5-6]. the purpose of the study is to devise a digital correlation-based phase measurement method for bia. the method is intended to be simple, portable, cost effective, accurate, and high frequency compatible. for the first time, the phase measurement for bio-impedance signals by using digital correlation method is demonstrated. tut sense module drive module drive electrodes sense electrodes bio-impedance device fig. 1 simple block diagram depicting bia 2. related work conventional phase measurement methods in bia, such as lissajous pattern analysis [7], virtual vector voltmeter [7], xor-filter [8], sine fitting [9], and zero crossing detectors [10], are easy to implement but are outdated. they are not feasible in portable devices and have frequency limitations. yang et al. [11] used an off-the-shelf chip ad8302 for bio-impedance phase measurement since it directly gives the magnitude and phase of the signal under study. this technique is expensive and is sensitive to direct current (dc) components, so it is not effective. olivarez et al. [12] utilized an lcr meter to measure bio-impedance; the meter is expensive and bulky, and is not application specific. analog phase measurement systems are characterized by bulky outdated designs, long-term voltage drifts, non-linearity, and cross talks at high frequencies. digital electronics and arm-based computational methods could minimize the effect of these characteristics, leading to an ultraprecision measurement. an accurate phase measurement is traditionally realized by using digital signal processing (dsp). the typical digital-type phase measurement schemes include digital zero crossing detector, fast fourier transform (fft), quadrature detection, phase-locked loop (pll), and correlation. hu et al. [13] used a zero-crossing algorithm for the sake of simplicity and accuracy, but it is done in an off-line manner due to hardware constraints. also, the digital zero crossing detection is influenced by harmonic distortion and high computation time. kamat et al. [14] used a discrete fourier transform (dft) based off-the-shelf ad5933, and controlled the system using field programmable gate array (fpga) which is computationally efficient. the dft based measurement is affected by additive noises as it involves inverse tan function. also, the frequency-domain method suffers from spectral leakage. the quadrature detection adds an additional phase shift of 90° followed by the use of sophisticated multipliers leading to complexity [1]. pll is also employed to lock into the phase of low amplitude signals, but pll is available as an off-the-shelf (analog and digital) and built-in module (in fpga, software, and neuronal) and requires high quality external circuits [15]. bertotti et al. [16] suggested a peak detection-based phase measurement, but it requires two reference sources increasing hardware complexity. the state-of-art devices such as sine-phase [17] and eliko [18] have focused on the accuracy compromising the cost. therefore, it is necessary to devise a phase measurement method which is simple, compatible with portable devices, cost effective, accurate, and high frequency compatible. these features lead to effective phase measurement. knapp et al. [19] introduced a generalized cross correlator (gcc) based method to estimate phase differences by locating the cross-correlation peak of filtered versions of two received signals. the optimum performance is attained only when the signals and noises are gaussialy distributed [19]. on the basis of accuracy, the correlation-based phase measurement tends to yield better results if modified in an application specific way. drawing inspiration from their work [19], liu et al. [20], luo et 35 proceedings of engineering and technology innovation, vol. 21, 2022, pp. 34-43 al. [21], shen et al. [22], liang et al. [23], etc. utilized the correlation method for time delay/phase difference measurement. till now, the correlation method has been applied to the estimation of time delay in the field of pda, interferometry, and remote sensing. however, the digital correlation-based phase measurement is not employed in many viable applications due to certain limitations, such as frequency drift errors and phase errors. this study proposes a digital correlation-based phase measurement method, which is simple, portable, cost effective, accurate, and high frequency compatible. for the first time, the phase measurement for bio-impedance signals by using digital correlation method is demonstrated. the experimental functionality and reliability of this method is supported by simulation and analytical treatment. the exhaustive error analysis is carried out, and the remedial measures are suggested and implemented. finally, the proposed setup is compared with the existing ones. 3. hardware realization for phase measurement using digital correlation in a bio-impedance system, a drive module offers a high-frequency low-amplitude excitation current to tut through drive electrodes, and a sense module measures the impedance of tut from the voltage response through sense electrodes. the tissue impedance is to be obtained as magnitude and phase. the methodology employed here for bio-impedance phase measurement is a part of a bio-impedance measurement device whose development is under progress. the excitation signal can incur phase changes while passing through a biological tissue. the phase difference between the excitation signal and the response signal is measured by the sense module to give bio-impedance phase. the accurate measurement of phase has always been critical. for testing the reliability of digital correlation, synchronous measurement approach is used. it requires the introduction of the phase difference between the reference signal (like excitation signal) and test signal (dummy bio-impedance signal). these signals can then be processed by digital correlation method, and the phase difference can be obtained. 3.1. experimental setup for phase measurement the controllers used in dsp boards do not accept negative signals. hence, a dc offset is often added before processing them. a typical method of introducing a delay in signals is to propagate the signal through resistor capacitor (rc) networks. rc networks are unfavorable in the present application as they remove dc components and also entail the use of variable passive components with changing frequencies. another option is to use a two-channel programmable function generator, which can produce two sine waves with a dc offset. the phases from 1° to 360° and the frequencies from 5 khz to 1 mhz can be set using it. however, such generator is bulky and cannot be a part of portable devices. as such, direct digital synthesis (dds) chips are employed to produce the reference signal and the test signal in this work. the experimental setup for performing the phase measurement is as shown in fig. 2. dds 1 (ad9837) vref vdd= 5v dgnd vout agnd fsync sclk sdata dds 2 (ad9837) vdd= 5v dgnd vout agnd fsync sclk sdata vdd=5 v lt1017 4 2 3 8 1 c1 lt1017 4 2 3 8 1 10µ f 10µ f tm4c123gxl gnd scl sda vdd= 5 v pf_2 pf_3 d1 d2 controllercomparators sine generators x(p) y(p) vtest cs1 cs2 sdata sclk 1 6 * 2 l c d c2 vdd=5 v display phase =0 phase =0 phase =phase = fig. 2 schematic diagram of hardware realisation for phase measurement 36 proceedings of engineering and technology innovation, vol. 21, 2022, pp. 34-43 the reference (vref) and test (vtest) sinusoidal signals of 0.624 mvpp and the frequencies from 5 khz to 1 mhz are generated using dds chips. the phase measurement using digital correlation method is performed using arm cortex m4 based texas board tm4c123gxl [24]. the board can accept signals through analog and digital pins. the analog pins of the board receive the signals and convert them to digital signals before processing them. however, sampling the signals with a 10-bit analog-to-digital converter (adc) can affect the phase information and can add a/d quantization noise. in order to avoid this, the conversion of sine waves to unipolar square waves is done and then given to the digital pins of the board. dds (ad9837 from analog devices, inc.) are employed to produce the reference signal and the test signal. dds ensures faster switching between output frequencies. additionally, it has fine frequency resolution of 0.06 hz, wide spectrum of operation till 8 mhz, and signal-to-noise ratio (snr) of -63 db. the calculation of frequency word (fw) is shown in eq. (1), and the required frequency is set by sending 16-bit control word through serial peripheral interface (spi) bus. 1 . . 2 . . ndes freq fw ref freq = × (1) where n1 is 28 for an in-built frequency register, the desired frequency (des. freq.) is the frequency of the input sinusoidal signals, and the reference frequency (ref. freq.) is 16 mhz. the phase register content is set after calculating its content as shown in eq. (2) and sending its content as 16-bit control word. 22 2 n phase register = × φ π (2) where n2 is 12, phase register is 12 bits, and � is true phase. for example, the data line value of c200 is the equivalent 16-bit control word for setting the phase of dummy bio-impedance signal as 45°. comparators are used to convert sine waves to square waves without altering the phase information. it is essential to select a comparator integrated circuit (ic) which can sense low input voltages with favorable bandwidth and match input impedance with respect to previous stages. lt1067 from analog devices, inc. is experimentally chosen from a set of comparators, as it can detect a minimum of 100 mv input sine wave and convert it into unipolar square wave. in real time applications, weak signals are magnified by subsequent amplifier stages, and hence the sensitivity of 100 mv is ample. it also functions on single supply, and this feature is essential to avoid negative signals entering the subsequent digital board. comparator circuits generate unipolar square waves. noise interferences are kept away by using constant current source of 215 kω output impedance and 218 µ v root mean square noise for excitation [3]. symmetrical instrumentation amplifiers with common mode rejection ratio (cmrr) above 80 db for rejection of electrode noise and motion artifacts are also used. the simulative noise analysis of comparator circuit is carried out in multisim, and a considerable low value of output noise of 41 pv/hz power is obtained. this work only focuses on the method of phase measurement, and other real time measurement challenges are considered during the development of bio-impedance devices. the texas board with 64 pin and 80 mhz crystal has two spi buses, one of which is used for the communication and control of dds chips. the reference square signal (c1) and test square signal (c2) obtained at the outputs of comparators are given to the digital pins of the board (pf2 and pf3) and then processed digitally to deduce the phase information. this simple method uses only two comparators and a digital board. the comparator works on 5 v dc single supply, and the board supplies this power. the board is universal serial bus (usb) powered. 37 proceedings of engineering and technology innovation, vol. 21, 2022, pp. 34-43 3.2. phase measurement using digital correlation the phase difference of � is induced between two unipolar square signals using dds. these signals are processed by tm4c123gxl board after acquiring their samples through digital pins. the floating point unit (fpu) of the board performs digital correlation by utilizing the samples of input signals. even though the input signal is continuous, the large sample size conserves the phase information within the board. both auto correlation and cross correlation of waves are performed to obtain the correlation coefficient as presented in eq. (3). 0 2 2 0 0 [ ( ) x][ ( ) y] [ ( ) x] [ ( ) y ] n p n n p p x p y p x p y p = = = − − = − − ∑ ∑ ∑ γ (3) where γ is the correlation coefficient, n is the sample number of input unipolar square signals, p represents the sample number, x(p) is the reference unipolar square signal, x� is the mean of x(p), y(p) is the test unipolar square signal, and y� is the mean of y(p). c1 and c2 signals have an amplitude of 5 v with an offset of 2.5 v. the digital pin senses and stores 1 for 5 v and 0 for 0 v. x� and y� are 0.5 considering the stored inputs. the maximum value of γ occurs when � is 0°, and the minimum value of γ occurs when � is 180°. these values are 1 and -1 respectively, and γ is 0 at 90°. this establishes a relationship between phase difference and γ, as shown in eq. (4). ˆ 90 90 ° °= − ×φ γ (4) where �� is the measured phase in degrees, and γ is the correlation coefficient. once the correlation is performed and the correlation coefficient is obtained, the phase difference is displayed on a liquid crystal display (lcd) monitor in degrees. this value is compared with the known value set by the dds chip. the observation is recorded for a set of phase values from 0° to 360° in a range of frequencies from 5 khz to 1 mhz. it is necessary to analyze the errors occurred during the experimentation and resolve them to apply the method in real time applications to improve the accuracy. the digitalread() command accepts the samples of unipolar signals. 4. modelling for phase measurement using digital correlation once the relationship between phase difference and correlation coefficient is established, it is necessary to further analyse the functionality and challenges of the digital correlation method before its real time utilisation. 4.1. modelling by simulation unipolar square waves are generated in matlab environment using square function. the square waves are defined by amplitude (0.5 v), offset (0.5 v), phase, number of samples (200-5000), and frequency (5 khz to 1 mhz). the phase adjustment in the degrees from 0° to 360° (�) is converted to radians and is introduced to one of the signals treated as the test signal. the correlation coefficient is calculated as per eq. (3), and the measured phase (��) is obtained using eq. (4). the true phase and measured phase for 2000 samples at 5 khz are recorded, and the error is tabulated in table 1. a maximum error of 0.2° (the best possible accuracy) is observed at 5 khz for true phase from 0 to 180. for the true phase greater than 180°, the value of γ repeats and hence the �� value is also repeated. this is not an issue in phase measurement as the phase differences in all major applications are confined within 180°. also, it is observed that as frequency increases with the fixed sample size, the error increases (addressed in section 5.1.). 38 proceedings of engineering and technology innovation, vol. 21, 2022, pp. 34-43 table 1 simulative phase measurement using digital correlation true phase, �(°) correlation coefficient, (γ) measured phase, �� (°) error, �� � (°) 0.000 1.000 0 0 1.000 0.992 0.895 -0.105 10.000 0.948 9.895 -0.105 45.000 0.501 44.978 -0.022 50.000 0.449 49.835 -0.165 60.000 0.337 59.910 -0.090 90.000 0.000 89.955 -0.045 120.000 -0.107 119.820 -0.180 150.000 -0.331 149.865 -0.135 180.000 -1.000 180.000 0.000 270.000 0.000 90.000 -180.000 360.000 1.000 0.000 -360.000 4.2. experimental modelling the phase measurement using digital correlation method is performed by the fpu of arm cortex m4 based texas board tm4c123gxl, and the experimental setup is illustrated in section 3.1. the digital pins are configured as input pins, and the unipolar signals are applied to these pins. 2000 samples of these signals are stored as arrays on which the correlation operation (repeated multiplication and cumulative addition of array elements/input samples) is performed to extract γ (eq. (3)). the true phase set by dds is measured using γ (eq. (4)) as the measured phase. the result obtained at 5 khz is tabulated in table 2. during the experimental modelling of the proposed method, it is seen that a maximum permissible error of 0.870°is observed at 5 khz. the error is more in experimental modelling than in simulation due to analog elements like comparators. the measurement process is repeated for different frequencies. it is observed that as the frequency increases with the fixed sample size, the error also increases just like in the simulation of digital correlation. the reasons for this error increase are discussed in section 5.1. table 2 experimental phase measurement using digital correlation true phase, �(°) correlation coefficient, (γ) measured phase, �� (°) error, �� � (°) 0.000 1.000 0 0 1.000 0.992 1.800 0.800 10.000 0.948 10.680 0.680 45.000 0.501 45.860 0.860 50.000 0.449 50.640 0.640 60.000 0.337 60.750 0.750 90.000 0.000 90.760 0.760 120.000 -0.107 121.020 0.020 150.000 -0.331 150.87 0.870 180.000 -1.000 180.000 0.000 270.000 0.000 90.660 -180.000 360.000 1.000 0.000 -360.000 5. results and discussion the characteristics of biological tissue are assessed using bio-impedance. bio-impedance can be represented as magnitude and phase varying with respect to time. sinusoidal carrier signals with bio-impedance information are processed to extract magnitude and phase. the proposed method uses digital correlation to measure the bio-impedance phase. sinusoidal signals are conditioned into unipolar square signals before phase measurement. the correlation method is used to compare the 39 proceedings of engineering and technology innovation, vol. 21, 2022, pp. 34-43 unipolar square signals and yield a correlation coefficient. the correlation coefficient represents the phase information. the phase measurement using correlation method is made viable through a series of simulative and experimental analyses. initially, the simulation and experimentation of phase measurement is carried out at a certain frequency. then, the frequency is varied. other parameters like amplitude, offset, sample size, etc. are kept the same for all measurements. it is seen that the occurring error increases with respect to frequency for a fixed sample size. 5.1. error analysis for the simulation of phase measurement using digital correlation the frequency and sample size of signals are varied in the simulation of phase measurement using digital correlation. the error is plotted versus true phase for various frequencies and sample sizes as shown in fig. 3. at 5 khz, as the sample size increases from 200 to 5000, the error (2.5° to 0.1°) decreases. this is because the sample size denotes the sampling frequency. according to nyquist criterion, the sampling frequency should be at least twice the input frequency. in the simulation, a single cycle is divided into samples. for an input frequency of 5 khz, the number of samples should be at least 2000. any value below that (e.g., 200) does not extract full phase information due to under-sampling whereas a large sample size (e.g., 5000) yields more information from input signals. as the frequency increases to 10 khz, the sample size of 5000 is not enough and need to be increased to fulfill nyquist criteria. a small error occurs even if the sample size is optimum with respect to frequency. this error is due to the harmonics of fundamental frequency of square wave and is negligible at higher frequencies. (a) at 5 khz for 200 samples (b) at 5 khz for 2000 samples (c) at 5 khz for 5000 samples (d) at 10 khz for 2000 samples fig. 3 simulation of phase measurement using digital correlation 5.2. experimental error analysis for phase measurement using digital correlation as explained in section 4.2., the phase measurement of signals with 5 khz is carried out smoothly with a maximum error of 0.870° for 2000 samples. the input signals at the digital pins are sampled by digitalread() and stored as global variables in random access memory (ram). the digital read time for 2000 samples is found out through programming as 9.8 ms, and the 40 proceedings of engineering and technology innovation, vol. 21, 2022, pp. 34-43 sampling frequency is hence 200 khz. the input frequency is changed by programming dds and is set to a higher frequency of 100 khz. it is seen that if the sample size is kept the same, there is a drastic increase in phase error just like in simulation. the reason for the phase error is explained in section 5.1. the error is nullified by increasing the sample size, but arm cortex m4 board has a 32 kb ram which restricts the sample size to 3800 (2000 is a safe value to avoid instability issues). in order to process the signals with higher frequencies, more memory in ram is needed. the limitation of ram affecting sample size is confirmed by trying to use an atmega 328p arduino board instead of texas board. the sample size of input signals is limited in atmega328p to 300 due to 2 kb ram. external rams (like 23lc1024, 47lc16, etc.) can be interfaced to the board through i2c or spi to accommodate large sized global variable arrays to solve this shortcoming. another alternative is to use boards with microcontrollers having ram with compatible large memories (like arduino due, tms320f28335, etc.). the bia experimental setup using modified phase measurement technique for a biological tissue (egg) is as shown in fig. 4. tetrapolar bio-impedance (bi) measurement is done by inserting needle electrodes up to 1 cm into the tissue of a hard-boiled country egg. bi is measured using lcr meter, and bi phase measurement is conducted at 5 khz, 50 khz, and 100 khz. lcr meter has frequency limitation up to 200 khz and uses voltage excitation with bipolar measurement [2].the results are tabulated in table 3. fig. 4 bio-impedance measurement of biological tissue (egg) table 3 comparison of bi measurement with lcr meter and proposed method frequency (khz) bi using lcr meter bi using digital correlation phase (°) phase (°) 5 -89.65 -88.12 50 -64.56 -65.04 100 -30.19 -30.33 5.3. comparison between the proposed method and the existing methods the existing methods of bio-impedance measurement are compared based on signal processing type, accuracy, frequency compatibility, cost, and applications. the proposed phase measurement method is a part of the bio-impedance device which is being developed. the technique used here is the correlation method [20-23]. to analyze where the proposed method stands, a comparison between the proposed method and other traditional and contemporary techniques is made, as tabulated in table 4. a careful study of these methods reveals that the traditional techniques are bulky, slow, and limited by operating frequency. other advanced techniques overcome the frequency limitation up to a certain extent and offer a better accuracy. however, they are complex. the commercially available phase detectors are expensive, sensitive to dc components, frequency limited, and offer moderate accuracy. the proposed digital correlation method is a part of bio-impedance device which is superior to all above techniques in terms of accuracy, frequency compatibility, portability (stand-alone module), speed, and cost effectiveness. the developed bio-impedance device with the present method of phase measurement costs around 140 usd. a/d quantization errors are nil in this method because there is no adcs. 41 proceedings of engineering and technology innovation, vol. 21, 2022, pp. 34-43 table 4 comparative analysis of phase measurement methods ref. method (°) type of signal processing accuracy (°) max. operating frequency (khz) approx. cost (usd) applications [1] quadrature detection analog 0.1 100 200 impedance measurement [7] lissajious loop pattern analog 1 100 280 impedance measurement [7] virtual vector voltmeter analog 1 100 3485 bio-impedance measurement for body mass [8] xor-filter analog 1 10 150 bio-impedance measurement for urolithiasis [9] sine wave fitting simulative 0.1 1 278 impedance measurement [10] auto-balance bridge analog 1 10000 200 bio-impedance of cell [11] phase detector (ad8302) analog 0.3 100 2787 bio-impedance measurement [12] lcr meter analog 0.1 20 2000 bio-impedance measurement for diabetes neuropathy [14] fpga method of samples digital 0.003 100 500 impedance cardiography [15] phase-locked loop digital 0.1 10000 250 impedance measurement [16] peak detection-based phase measurement analog 0.3 100 200 impedance measurement [17] sine phase software based 1 16600 6272 impedance measurement [18] eliko software based 0.1 349 6690 impedance measurement [20] correlation digital 0.15 1000 range finder [23] sine-tan method analog 0.01 1000 5000 interferometry [25] fft-inverse discrete fourier transform digital 0.1 10000 impedance measurement this study modified digital correlation digital 0.002 1000 140 bio-impedance analysis 6. conclusions on the basis of the studies conducted, it can be concluded that the digital correlation-based phase measurement method is a reliable means to characterize the media through which the signals pass. this characterization is done by measuring the delay/phase changes incurred by the signals. in the hardware realization, a reference wave from sources is compared with the wave at the receiving end and looked upon for phase differences by the experimental setup. the phase measurement using correlation method is not widely used in many viable applications including bio-impedance phase measurement due to the demerits such as frequency drift and phase errors. these issues are addressed in this work in a systematic manner. the best possible accuracy of 0.2° is obtained by realizing the proposed method for measuring the induced phase differences between two sine waves in a frequency range of 5 khz to 1 mhz. this modified phase measurement technique is a part of sense module of bio-impedance device under development. this modified method can be incorporated in portable devices for any application which needs accurate phase measurement, as it uses fewer numbers of hardware components and can be operated on usb power. conflicts of interest the authors declare no conflict of interest. references [1] m. sedlacek and m. krumpholc, “digital measurement of phase difference—a comparative study of dsp algorithms,” metrology and measurement systems, vol. 12, no. 4, pp. 427-448, january 2005. [2] s. grimnes and o. martinsen, bioimpedance and bioelectricity basics, london: academic press, 2008. 42 proceedings of engineering and technology innovation, vol. 21, 2022, pp. 34-43 [3] s. sruthi, r. dhavse, and j. n. sarvaiya, “precise, portable and wide band alternating current source for extremely low current values,” analog integrated circuits and signal processing, vol. 102, no. 2, pp. 273-282, february 2020. [4] f. simini and p. bertemes-filho, bioimpedance in biomedical applications and research, cham: springer international publishing, 2018. [5] h. p. f. x. moonen, f. j. l. zanten, l. driessen, v. de smet, r. slingerland-boot, m. mensink, et al., “association of bioelectric impedance analysis body composition and disease severity in covid-19 hospital ward and icu patients: the biac-19 study,” clinical nutrition, vol. 40, no. 4, pp. 2328-2336, october 2021. [6] m. n. r. jafari, z. esmaeilbeig, and m. nemati, “overcoming the bottleneck in covid-19 detection: a machine-learning approach to improve accuracy of electrochemical impedance spectroscopy (eis) detection sensitivity,” open access journal of biomedical science, vol. 3, no. 1, pp. 676-679, october 2020. [7] j. g. webster and h. eren, measurement, instrumentation, and sensors handbook: electromagnetic, optical, radiation, chemical, and biomedical measurement, boca raton: crc press, 2017. [8] a. silverio and j. d. tatel, “development of a variable frequency impedance measuring system,” 14th international symp. on integrated circuits, pp. 564-567, december 2014. [9] p. m. ramos, f. d. silva, and a. c. serra, “improving sine-fitting algorithms for amplitude and phase measurements,” 17th international measurement confederation world congress, pp. 614-619, june 2003. [10] n. li, h. xu, w. wang, z. zhou, g. qiao, and d. d. li, “a high-speed bioelectrical impedance spectroscopy system based on the digital auto-balancing bridge method,” measurement science and technology, vol. 24, no. 6, 065701, may 2013. [11] y. yang, j. wang, g. yu, f. niu, and p. he, “design and preliminary evaluation of a portable device for the measurement of bioimpedance spectroscopy,” physiological measurement, vol. 27, no. 12, 1293, 2006. [12] j. p. olivarez, f. a. olivares, a. p. medina, j. d. carmona, a. r. agundis, a. e. calderon, et al., “bioimpedance phase angle analysis of foot skin in diabetic patients: an experimental case study,” innovation and research in biomedical engineering, vol. 36, no. 4, pp. 233-239, 2015. [13] g. hu, j. xu, g. w. auner, j. smolinski, and h. ying, “digital phase detection approach and its application for aln dual-mode differential surface acoustic wave sensing,” sensors and actuators b: chemical, vol. 132, no. 1, pp. 272-279, january 2008. [14] d. k. kamat, s. nimkar, and p. m. patil, “bio-impedance computation based on hardware implementation using fpga,” international journal of electrical engineering, vol. 6, no. 1, pp. 12-19, january-june 2014. [15] “what is direct digital synthesiser dds,” https://www.electronics-notes.com/articles/radio/frequency-synthesizer/dds-direct-digital-synthesis-synthesizer-what-isbasics.php, 2012. [16] f. l. bertotti, m. s. hara, and p. j. abatti, “a simple method to measure phase difference between sinusoidal signals,” review of scientific instruments, vol. 81, no. 11, 115106, 2010. [17] “sinephase impedance analyzer,” https://sinephase.com/impedance-analyzer-model-16777k-lcr-meter/, august 2016. [18] “impedance spectroscopy,” https://eliko.ee/quadra-impedance-spectroscopy/, december 2017. [19] c. knapp and g. carter, “the generalized correlation method for estimation of time delay,” ieee transactions on acoustics, speech, and signal processing, vol. 24, no. 4, pp. 320-327, august 1976. [20] y. z. liu and b. zhao, “phase-shift correlation method for accurate phase difference estimation in range finder,” applied optics, vol. 54, no. 11, pp. 3470-3477, april 2015. [21] j. luo and e. e. konofagou, “a fast normalized cross-correlation calculation method for motion estimation,” ieee transactions on ultrasonics, ferroelectrics, and frequency control, vol. 57, no. 6, pp. 1347-1357, june 2010. [22] y. l. shen and y. q. tu, “correlation theory-based signal processing method for cmf signals,” measurement science and technology, vol. 27, no. 6, 065006, may 2016. [23] y. r. liang, h. z. duan, h. c. yeh, and j. luo, “fundamental limits on the digital phase measurement method based on cross-correlation analysis,” review of scientific instruments, vol. 83, no. 9, 095110, september 2012. [24] texas instruments, “tiva™ tm4c123gh6pm microcontroller data sheet,” https://www.ti.com/lit/ds/symlink/tm4c123gh6pm.pdf, 2014. [25] w. wang, y. tang, x. zhang, and k. wang, “design of a recursive single-bin dft algorithm for sparse spectrum analysis,” cluster computing, vol. 20, no. 2, pp. 1483-1492, april 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/). 43  proceedings of engineering and technology innovation, vol. 1, 2015, pp. 19 26 19 cop y right © taeti credit card risk assessment using artificial neural networks hsi-chieh lee 1, 2, chen-chi yu 1 , tzu-miao lin 1 1 department of information management, yuan ze university, chungli, taoyuan, taiwan r.o.c. 2 department of computer science and information engineering, national quemoy university, kinmen, taiwan r.o.c. received 16 april 2015; received in revised form 08 may 2015; accept ed 18 june 2015 abstract in recent years, the supply and demand of the plastic currency market has been rapidly increasing, especially, for the growth rate in the credit card market. it has increased about 16 times in the last 10 years. many banks devoted to make a large investment in credit card marketing for the sake of getting the maximum profits in the worldwide market. however, most banks are trying to reduce the requirements for credit card application in order to increase the motivation of the customers for applying their credit cards. as a result, many banks somehow ignore the risk management of credit card approval which leads to the increases of bad debt in the credit card market. when this scenario happens year by year, those banks will not get profits from the credit card market but a great loss. in this study, a total of 113,048 entries were used which included fundamental customer data, credit card data, and customer history data from joint credit information center (jcic) of taiwan. we used the characteristics of artificial neural networks and grey theory to find out the potential factors of the bad credit and finally used the correlation method to find out the higher (important) relative variables (parameters) of bad credit. 80,000 entries were randomly selected as training data and the remaining 33,084 entries were used as testing data. the experimental results shown that the accuracy of forecasting rate for the proposed early warning system was an overall of 92.7%. these results suggested that the once the collections of the new customer data were available, the proposed approach could be used as an early warning system which can be used to decrease the risk of credit card approval. ke ywords : credit card risk assessment, artific ia l neural networks, self-organizing map (som), grey theory, electronic commerce 1. introduction daily items for human beings have been developed completely and in wide array to satisfy the diverse needs of individuals, given the ceaseless advancement of technologies. the money market, an indispensable part of human livelihood, has also transformed from the “barter system” of early times, to transactions with currency, and contemporary transactions with plastic money. in taiwan, the use of plastic money for transactions is common and everywhere. this implies that plastic money has been gradually and surely replacing cash in transactions, and has emerged as the primary form of currency. however, the excessive inflation of credit limit granted in favor of the consumers and the improper approval of credit limits by banks resulted in high debts beyond the means of the consumers for retirements and high non-performing loans for the banks. this is the problem of the so-called “credit card slaves” in taiwan, which hit the society hard wave after wave. in light of the said issued, this paper proposes to apply the neural network with an attempt to map out an effective early warning system for detecting the characteristics of potential credit risks of card holders to avoid the loss of banks due to improper review and approval of credit cards. 1.1. background and motivation according to joint credit information center [1], there were 3,676,000 credit cards circulated in 1995. this number increased to 45,494,000 in 2005. similarly, credit card loans rose from nt$272,387 million in 1995 to nt$1,420,984 in 2005. the data of loans from credit card charges and cash advances provided a clue to the reasons why banks scrambled for a share of the credit card market. the statistics of loans from credit card charges increased by 7.5 times in the decade of 1995 to 2005. the cash advance business also mushroomed by almost 15 times in the same period (fig. 1, 2). fig. 1 volume of credit cards in circulation over the years *corresponding aut hor, email: imhlee@gmail.com proceedings of engineering and technology innovation, vol. 1, 2015, pp. 19 26 20 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti fig. 2 growth trend of loans from credit card charges and cash advances yet, the number of new cards issued, the amount of credit card loans, and cash advances significantly reduced in 2006. this phenomenon was the result of one card holder possessing many credit cards because banks tended to lower the entrance barrier for granting the applications for credit cards and cash cards in order for a larger quantity of issuance. after the “credit card slaves” storm, banks have learned their lesson. they changed from focusing on the quantity of card issuance to screening for clients with good credit standing and therefore attempted to construct an efficient early warning system under objective conditions to reduce non-performing loans. this remains the motive of this study. 1.2. the objective of study this study adopts the method of data mining whereby the potential credit risks of credit card clients are explored based on the characteristics of neural network clustering. in practice, the personal information of clients, the borrowing records of the clients at joint credit information center (jcic) and the patterns of spending by credit card holders and the payment records are included for analysis for finding out the factors related to bad credit of card holders and delinquency in payment. these factors will be used as the variables in the analysis for constructing an early warning system for credit evaluation. 1.3. literature review literature on risk management in the financial sector will be introduced in this section. theories of the neural network model also will be discussed here. 1.3.1. risk management risk literally means the possibility of suffering loss. risk can be managed through identification, analysis, and responding measures. risk can be analyzed from the perspectives of general causes, the probability of occurrence, the effect of risk, and the damage or loss that is caused (converted into monetary form). rosenbloom (1972) defined risk as “the uncertainty of the occurrence of loss”0, webster’ s dictionary defined risk as the opportunity of damage or loss or being in danger. these definitions focused on the possibility of a negative result. further. williams & henis (1981) pointed out that: risk management is the management method through identification, assessment and control of risk at minimal cost in order to minimize the loss that is caused 0. focusing on risk management of banks, lin (1997) proposed a few principles [4]: (1) properly evaluate the influences of different risks on the profitability of banks and these influences shall be subject to good financial management; (2) properly evaluate the capacity of the bank management in early warning of the banking sector and assessment of the strength and weakness of the bank; (3) understand the capacity of the bank in assuming risks on the basis of the competitive advantage of the bank in the industry and take the changes in the macro economic environment under control in pursuing its goal with relevant adjustment of risk level; (4) viable risk management and monitoring systems will be necessary: (5) good management and the training of good professionals. risk management is important to enterprises of any industry. better risk control will help to lower the intensity of loss. 1.3.2. neural network model a neural network is one form of artificial intelligence and is a data processing system imitating the biological neurotic system. with large number of linked artificial receptive neurons, this system imitates the learning capacity of a biological neurotic system and simulates learning and memory through a parallel computation framework. it is applied to categorization, reflection, identification and optimization functions. the neural network system is a simulation of the biological neurotic system in operation through ceaseless learning and correction of data for correct output. the neural network system features large volume of parallel processing, non-linear output, and multiple layering in forecasting. the processing element (or pe) is the fundamental component unit of the neural network system and the processing unit can be subcategorized as input and output values. the sum of the output values from each layer is the input value of the upper layer. generally, the function of the product of the input value and weighed links is expressed in the following equation: ( ) ij i j i a f net f w x           a=the output signal of neural network pe xi=external input of data net=cluster function θj=threshold value of neural network pe proceedings of engineering and technology innovation, vol. 1, 2015, pp. 19 26 21 wij=the linked weights of all neura l network pe in between data paths of the pe. the symbol denotes the strength of the effect of the ith pe of the former layer on the jth pe of the latter layer. f is the transfer function of neural network pe. it is the conversion of the sum of another pe input value and the linked weights to output value through the pe conversion function. the conversion function usually sorts out the output value like the sigmoid function and is mathematically expressed as: 1 ( ) 1 net f net exp    the output of s function fa ll between 0 to 1, and the s function input/output curve is usually skewed. therefore, the net signal is very small and approaches zero. at this point, the output value approaches infinity. if net is a positive value and approaches infinity, the output va lue approaches 1. besides, another commonly used conversion function is hyperbolic tangent function and its output is also in an s shape, and is mathematically expressed as follows: ( ) net net net net exp exp f net exp exp      the hyperbolic tangent function is symmetrical with its origina l point and is a function w ith output function between -1 and 1. this function works faster than other neural functions in computation for results [5]. 1.3.3. neural network model this section gives an introduction on the application software, techniques, and relational analys is. the clustering techniques and neural network algorithm used clementine of spss proposed by kohonen. grey relational analysis is adopted for the sampling process of card users in poor credit standing. 1.3.4. clustering technique clustering is the method through which data with the same or similar characteristics are grouped. clustering is a form of non-monitoring learning skill and in the process of clustering the distance between each data and the center of the cluster will be mapped out. clusters of the shortest distance to the center will be formed into the same group. the main feature of clustering is that no attribute has been assigned to the data and clustering is d irectly conducted. after the clustering process, the clusters will be based on the results of clustering to trace the characteristics of the cluster elements and the criteria for clustering. taking this study as an example, the symptoms of c lients w ith potential risk are unclear before the process. therefore, clustering is used for digging out the characteristics of different clusters. with the different intens ifies of the clusters, different weights are assigned to construct a system for forecasting clients with bad credit standing. the self-organizing map neural network, or som, was proposed by kohonen (1988) [6]. in brief, this system is a network framework with 2 layers, the input layer and output layer (or competitive layer). the characteristic of som is to search the characteristics of the targets and reflect the vector of any dimension to a one or two dimensiona l map (fig. 3.2). the output layer neurons will base on the characteristics of the input and display in the output space in topological structure. this map reflects the characteristics of the input vector. therefore, this is called som. the strength of the neural network proposed by kohonen is that it is a parallel neural network structure and provides the characteristics of analysis and the relation of the organizations. therefore, it has high confident level as a technique of categorization. assuming there are k neurons in this network. if wkt (n) x (n) = max kt wi (n) x (n) i=1, 2,…a this means that the kth neuron is the ultimate winner. adjustment of the chain values            1 ? j j k j w n x n w n if j k w n w n        where, x＝(x1,x2,...,xp )t means the input vector of the network; wj=[wj1,wj 2,...,wjp]t, represents the chain value vector of the jth neuron. p represents the dimension, and η is the parameter of learning rate, n represents the dispersion time. usually, each chain value of the neuron will be normalized with growing rate at 1. therefore, the rule for sorting out the winner could be corrected as: fig. 3 dimension kohonen model proceedings of engineering and technology innovation, vol. 1, 2015, pp. 19 26 22 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti if ∥x－wk∥<∥x－wj∥, then neuron k will be the winner until the formation of the characteristic map (fig. 3). 1.3.5. clustering technique teng (1982) proposed the grey system theory for studying the cognition of human be ings on different research and things from simple to sophistication. it is just like the waves in the sea coming one after another. in other words, many things are neither absolute ly black (blank) nor white (fully transparent) to human be ings. instead, it is in a state of grey (half understood and half unknown), which implies that information is not complete or cannot be described in full detail [7]. grey relation ana lys is refers to certain degree of association between 2 elements of the systems or events. if the elements become congruent due to changes, the association between the two is high, or vice versa. degree of association is the variance of the geometric shapes of the curves. assuming a reference value x0, and n comparative values, then x1, x2,…, xn, the relationa l coefficient ξ(xi) of each comparative value matrix and the reference value can be expressed by the following equation:                   0 0 0 0 | | | | | | | | i i i k i k i i i i k min min x k x k max max x k x k k x k x k max max x k x k           ζ(zeta) is the differentiation coefficient and 0 < ζ < 1 while , implies that it is the smallest absolute value between two values (comparative and reference values). after the comparison of a ll values is completed, the smallest value of the small values is sorted and marked as δ min. similarly, represent the greatest absolute value between two values (comparative and reference values). after the comparison of all values is completed, the greatest value of the great values is sorted and marked as δ max. relationa l coeffic ient ξ(xi) can be simplified and expressed as maxoi(k) maxmin )(       k i furthermore, the relationa l coeffic ient may vary at different points of the reference value and the comparative value along the curve, and there is more than one value of the degree of association. therefore, if a ll the points a long the curve are linked and make it a relational coeffic ient value, this will be the mean of such value and is expressed by ri. the equation of degree of association ri will be:  k n r i n k i  1 1   finally, after adding up a ll relationa l coeffic ient values and then dividing by the total number of values, we get the degree of association. due to the demands on credit card risk assessment and prediction capability of artific ia l neural networks, there are many literatures regarding the deployment of nove l techniques on the problem of credit card risk assessment in recent years [8-10]. 2. experiment 2.1. study sample processed data 113,048 grouping of data testing data 33,048 training data 80,000 use neural network clustering to find out the characteristics of factors most influential on bad credit status like characters of the clusters for constructing the credit early warning system) are the variables meet the requirements screen out the variables credit early warning system n y is the forecast ideal adjust the weights of the parameters end y n fig. 4 flow figure of constructing the early warning system samples of this study inc luded the master file of card holders, credit card payment record information, and the information of jcic, which made up a total of 113,048 entries. thirteen variables are sorted and the samples are divided into training group and testing group. 80,000 samples were selected at random for the training group. the characteristics of bad credit standing of this group are traced through neural network analys is. weights were assigned to these characteristics for constructing an objective client credit evaluation system. the remaining 33,084 samples were subject to testing under the testing group (fig. 4). the result from forecasting is compared with the actual bad credit status clients of proceedings of engineering and technology innovation, vol. 1, 2015, pp. 19 26 23 the banks exemplified by failure to renew credit cards or confirmed by the processing banking officers. if the variance between the forecast and the actual result is wide , it means that more influentia l factors exist and additiona l weight shall be assigned in the scoring. the same princ iple is applied for repeated adjustment of weights. if the forecast approximates the ratio of clients in poor credit standing, it means the characteristics of the variables used in this study are sufficient to forecast clients of potentia lly poor credit status. 2.2. the benchmark for banks to make judgment on poor credit status different ba nks have different criteria for judging poor credit status of c lients. ba nks who provided sa mple for this study offered both interna l and externa l information for judging poor credit status of c lie nts. they are : i. interna l information: this is the information ava ilable from the database of the ba nks and inc ludes : 1) loan information of the ba nk – overdue cash card accounts of more tha n 15 da ys of de linquency for the last payment. 2) loan information of the bank – overdue loan accounts of more than 30 da ys of de linquenc y for the last payment. 3) marked by the bank as gd (1~9) une mployed/ins uffic ient inc omes. 4) marke d by the bank as gf (1~9) dec lare d le ga lly inc ompetent. 5) marke d by the bank as gl (1~9) to be ass iste d by guara ntor for settleme nt.6) marked by the bank as gm (1~9) bearing informa tion on ass ignment of right to debts for repayment. 7) marked by the bank as gn (1~9) retirement of loa ns by a re lative. 8) tota l loans from the bank exceed approved limits. ii. externa l information: this inc ludes information of the c lie nts from other banks , jcic transaction record, loa n record and payment record. externa l information us ua lly c onta ins the follow ings : 1) loan information –unusua l credit approva l. 2) loan information – record of dec lined account. 3) there were more than 3 “insuffic ient pa yme nt for the minimum” or “overdue account for payment in full” of cre dit cards over the las t s ix months. 4) delinque ncy of cash card payment for more tha n 31 days (inc lus ive) over the las t s ix months. 5) delinque ncy of cons umer loa ns for tha n 31 da ys (inc lus ive) over the las t s ix months. 6) supple mentary/s pec ia l me ntion information indicated as une mployed/ins uffic ient inc omes. 7) supple mentary/s pec ia l mention informa tion from jcic indicated dec lared le ga lly incompetent. 8) supple mentary/s pec ia l me ntion information from jcic indicated ass ignme nt of right to debts for retirement of loans. 9) supple mentary/s pec ia l me ntion information from jcic indicate d retirement of loans by a relative. 10) tota l loa ns from other banks exceed approved limits. further to the aforementioned interna l and external information, other special requirements are availed to the banks for reference: 1) no renewal of credit cards for student card holder who holds more than 3 cards. 2) supplementary proof of financia l capacity is required if the tota l unsecured loans exceeds 22 times of monthly salary. 3) if any of the said conditions applies to particular clients of the bank in credit judgment, such client shall be classified as “no card renewal” client. 2.3. analysis of the clustering results the clustering characteristics of the kohonen mode l are adopted and cross computed with a ll input variables. the result showed that each cluster shows their resemblance. under this method, there are 22 clusters grouped. data for the 22 clusters are permuted in sequential order. clusters ranked at the higher echelon of the hierarchy are the variables denoting clients in better credit standing, and vice versa. the table of variables ranking showed that the top 6 poor credit accounts, clusters 6, 9, 14, 18, are congruent with the 6 clusters confirmed by the banks as in poor credit status through ranking and manual confirmation (therefore renewal of cards is declined). therefore, this study further explores the factors and characteristics of the 4 clusters for poor credit standing. they will be listed as the reference in subsequent forecasting of clients of poor credit standing. the relations between the variables and the status of bad credit are expressed in terms of grey relation (fig. 5). fig. 5 forecasting on clusters of bad credit status we can see the comparison of relations between the clusters and the factor values directly from fig. 5. in this study, cluster 18 is used as a control for comparison. grey relation analys is is introduced as follows: i. standardization: the characteristics are sorted through analys is under the kohonen mode l. the data of the 4 clusters are listed (the primary data indicated the ranking among the clusters. the greater value of the data, the poorer the credit status). ii. find proceedings of engineering and technology innovation, vol. 1, 2015, pp. 19 26 24 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti out the maximum variance of the clusters and the minimum variance, for finding out the maximum value and the minimum value , we must first find out the “mapping of variance” table for the maximum value of variance as =1.59, which means δ max=1.59. the minimum variance is ＝0 and is expressed by δmin=0. iii. calculation of relationa l coefficient: it is ξi (k). assuming ζ=0.5, then the relational coefficient of x1 and x0, ξi (k), can be expressed as follows: (1) (2) 57.0 59.15.057.0 59.15.00 max)3( maxmin )3( 01 1           (3) 5.0 59.15.040.0 59.15.00 max)4( maxmin )4( 01 1           (4) the same principle applies to the computation of all other variables.  k n r i n k i  1 1   (5) iv. calculate for the value of: the degree of association between comparative value x1 and corresponding reference value x0.   5723.0 4 5.057.064.01 4 1 1 4 1 1     kr k  the same principle applies to the computation of all other variables. v. the above results are ranked in sequentia l orders and the following result is yielded. r4>r5 >r6> r7> r8> r13> r2> r14> r9 >r11> r16> r12> r1> r15> r1> r17> r3 2.4. construct the early warning system mapping of variables for judgment (fig. 6) can be given the benchmark for mark deduction of each variable. for example, the factor of “cash advance”, if the data field is “y”, it means 8% of the scores for this factor shall be deducted. the sample princ iple applies to the computation of all other variables. the user interface of the early warning system (fig.7) for the forecasting of potentia l risks of credit card holders. enter the id card number of the card holder and press “activate for scoring”, the system will retrieve the data of this holder from the database and process each count of data for determining if they meet the default standards. the total scores for this id will be shown in the fie ld of “total score”. and recommendation will be shown in the fie ld of “advice result”, including the recommendation for “renewal of card” or “no renewal”, and so forth. outstanding amount 7% pass_due_amt stauts of payment 8% pcd payment record in the last 6 months(1~6)48% pay_code_12 payment range 7% pmt cash advance 8% cash approved limit/payement range 8% limit / pmt type of loan 7% account_code is it greater than 0 is it>=4 is it>=2 is it greater than 800,000 y is it smaller than 0.00015 is it a, b, or c full score=100 if the conditions set forth in each field are met, deduct marks from the field. total scores oustanding amount/payment range 7% pass_due_cdoe / pmt is it greater than 0 fig. 6 mapping of variables for judgment fig. 7 the user interface of the early warning system 1 59.15.00 59.15.00 max)1( maxmin )1( 01 1           64.0 59.15.074.0 59.15.00 max)2( maxmin )2( 01 1           proceedings of engineering and technology innovation, vol. 1, 2015, pp. 19 26 25 3. conclusions fig. 8 distribution of “renewal” and “no renewal” accounts judged by the banks currently, each bank has its own method of evaluation in credit risk management. for transparency of information, each bank can access to the database of jcic for the information of its clients on loans w ith other banks, credit information and related payment record. banks can be more accurate in forecasting credit risk if they could have more complete information. in this study, the common fie lds and data of jcic are adopted, and the clients of the banks are subject to analys is with the data provided. neural network system and grey relation analysis are applied to find out the variables possibly related to poor credit status of clients. this section gives a description on the empirical findings of this study. fig. 8 is the population distribution of the clients for deciding the renewals of cards in the testing group. the distribution shows that 93% of all samples constituted c lients being granted the renewal of cards. only 4% of the clients are determined as “no renewal” of cards due to poor credit standing. there are 3% of the samples judged by the banks as “no renewal” after supplementary information on financ ial pos ition has been provided, and shown in the credit record as between “renewal” and “no renewal” of cards (fig. 8). table 1 shown that t1 stands for forecasted “renewal of card” in this study while t2 stands for “renewal of card” determined by banks. f1 stands for forecasted “no renewal of card” in this study while f2 stands for “no renewal of card” determined by banks. (t1,t2)( ※ noyr 2) the forecasting rate is 91.21%, and (f1,f2) the forecasting rate is 1.46%. the total of the two is 92.67%, which means that the forecasted results under the early warning system are congruent with the actual judgment of the banks. in table 2, 49% of the forecasted clusters (t1, f2) were randomly selected by banking professiona l for verifying the credit status of the card holders. the findings indicated that of the 160 samples being selected (note 3), 33.33% are indeed in good credit standing but were erroneously judged by the banks as “no renewal of card”. this would be significant loss of the banks for losing such good potential clients. further, 1.82% of the forecasted clusters (f1, t2) in table 3 are subject to analysis. the findings ind icated that of the 60 randomly selected samples (note 4), 33.39% are indeed clients in potentia l poor credit status but erroneously judged by the banks as “renewal of card” clients. in other words, the proportion of bad debts from 1/3 of these clients is much higher than other clients. the banks shall pay attention and take precautions. this study has led us to know that banks have shortcomings in making judgment on renewal or no renewal of cards. this study also helps to explore a group of margina l c lients falling in between “renewal” and “no renewal” and associated strongly w ith bad credit status, which is critical for decis ion of renewal or no renewal. this could be served as reference for the banks in the evaluation of credit card clients. table 1 comparing the forecasted rate of the early warning system banks this study t2 f2 t1 91.21% 5.49% f1 1.84% 1.46% table 2 t1f2 distribution of sample clustering t1f2( 160 samples) renewal % no renewal % judged by banking officers 53 33.33 106 66.67 table 3 f1t2 distribution of sample clustering f1t2(60 samples) renewal % no renewal % judged by banking officers 20 33.90 39 66.10 ※note 1: “no renewal for student cards” refers to the same student who holds more than 3 credit cards or applies for credit cards with more than 3 banks at the same time. if these conditions are met, the banks shall treat this group of clients as potentially in poor credit status and determined as “no renewal of card”. ※ note 2: (t1, t2) in this study stands for: (t1) forecasted as “renewal of card” and (t2) the judgment of the banks is also “renewal of cards”, and so forth. ※note 3: at the proportion of 5.49% of all samples, 10% or 160 samples are selected at random. ※note 4: at the proportion of 1.82% of all samples, 10% or 60 samples are selected at random. proceedings of engineering and technology innovation, vol. 1, 2015, pp. 19 26 26 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti re fe re nces [1] joint credit information center, taiwan, http://www.jc ic.org.tw/index. htm [2] j. s. rosenbloom, “a case study in t. khabaza and c. shearer (1995), data mining with clementine,” integral solutions limited, 1972. [3] c. a. willia ms and r. m. henis, risk management and insurance, 4th ed., new york: mcgraw-hill, 1981. [4] c. r. lin, finance innovation and risk management, 1st ed., 1997. [5] r. a. neibet, “h ow to choose a data mining sui te,” http://www. dmre view.c om/artic le_s ub.cfm? articleid=1000465, 2004. [6] t. kohonen, self-organizing maps, 3rd ed. springer, berlin, heidelberg, 2001. [7] c. l. teng, h. kuo, grey theory and its application, 1996. [8] j. n. crook, d. b. edelman, l. c. thomas, “recent developments in consumer credit risk assessment,” european journal of operational research, vol. 183, no. 3, 16, pp. 1447-1465, december 2007 [9] l. yu, s. wang, k. k. lai,“2007 credit risk assessment with a multistage neural network ensemble learning approach,” expert systems with a pplications , vol. 34, no. 2, pp. 1434-1444, february 2008 [10] b. twala, “multiple classifier application to credit card risk assessment,” expert systems with a pplications , vol. 37, no. 4, pp. 3326-3336, april 2010. 3___peti#7217__24-34 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 24-34 supervised learning based classification of cardiovascular diseases arif hussain 1 , hassaan malik 1, 2, * , muhammad umar chaudhry 1, 3, 4 1 department of computer science, national college of business administration and economics, multan, pakistan 2 department of computer science, university of management and technology, lahore, pakistan 3 department of electrical and computer engineering, sungkyunkwan university, suwon, korea 4 aihawks, multan, pakistan received 26 february 2021; received in revised form 21 august 2021; accepted 22 august 2021 doi: https://doi.org/10.46604/peti.2021.7217 abstract detecting cardiovascular disease (cvd) in the early stage is a difficult and crucial process. the objective of this study is to test the capability of machine learning (ml) methods for accurately diagnosing the cvd outcomes. for this study, the efficiency and effectiveness of four well renowned ml classifiers, i.e., support vector machine (svm), logistics regression (lr), naive bayes (nb), and decision tree (j48), are measured in terms of precision, sensitivity, specificity, accuracy, matthews correlation coefficient (mcc), correctly and incorrectly classified instances, and model building time. these ml classifiers are applied on publically available cvd dataset. in accordance with the measured result, j48 performs better than its competitor classifiers, providing significant assistance to the cardiologists. keywords: cardiovascular disease, machine learning, artificial intelligence 1. introduction cardiovascular disease (cvd), which is considered to be one of the most fatal diseases, includes a cluster of diseases such as aortic aneurysm, angina, rheumatic heart disease, coronary heart disease, congenital heart disease, stroke, peripheral arterial disease, etc. [1]. cvd affects heart muscles (myocardial infarction) and vasculatures (hypertension). the only two methods to reduce the occurrence of fatal events due to cvd are reducing cholesterol intake and increasing proper exercise. the detection of cvd in the early stage is a difficult and crucial step. cvd are complex and heterogeneous, as it is caused by multiple factors including environmental and behavioral factors. artificial intelligence (ai) has an ability to exploit the big data largely used in patient care. the big data refers to the large volume and variety of the data that are complex in nature. conventional data processing methods and techniques are insufficient to effectively examine a large amount of data [2]. the big data generally displays high capacity, reliability, and various types of features. as a result, scalability and accuracy is improved by using the big data techniques as compared to the traditional methods such as regression and correlation-based models. the big data is based on four vs [3]: volume, velocity, variety, and veracity. the techniques of ai and machine learning (ml) in the big data have achieved tremendous success that helps commerce and fabrications, in order to highlight the hidden patterns and to predict future prospects. researchers use ai methods to track down the disease related contexts, e.g., controlling tuberculosis [4] and predicting extended-spectrum β-lactamases (esbl) [5] and heart diseases [6]. * corresponding author. e-mail address: f2019288004@umt.edu.pk tel.: +92-333-3873355 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 24-34 ml methods are not only used for disease classification, but also widely used in industries. ml based framework are effectively used for evaluating the data collected from smart meters to analyze the original and fake data [7]. several types of ml classifiers are also used for the identification of chatter vibration [8]. in addition, ml techniques are also utilized to distinguish cyber-attacks to perform the paradigm and authentication [9]. ml algorithm is not a panacea for prediction related tasks. even a perfect ml model depends upon the quality and magnitude of the dataset from which it is trained. there are variant kinds of algorithms for classification and prediction of cvd outcomes. this study provides a comparative analysis among the performance matrices of ml classifiers, i.e., support vector machine (svm), logistics regression (lr), naive bayes (nb), and decision tree (j48), using the cvd dataset in diagnosing and making significant decisions to assist the clinical and health experts. the objective of this study is to evaluate and achieve the best accuracy with the lowest error rate in diagnosing the disease. consequently, the efficiency and effectiveness of those approaches in terms of precision, sensitivity, specificity, accuracy, correctly and incorrectly classified instances, and model building time are evaluated. in summary, the main contribution of this study is discussed below: (1) the four ml classifiers applied for cvd diagnosis and the detailed state-of-the-art ml classifiers are discussed. (2) the comparative analysis of ml classifiers is observed in terms of the performance evaluation matrices for cvd classification. (3) j48 model achieves the highest classification accuracy. 2. related work ai is a comprehension of mathematical designing that has the capability to improve health services by using new health care strategies and procedures. in ml, classification is considered to be the most important and essential part. a lot of research has been conducted by using ml to classify cvd from health care related datasets, proving that ml has good performance. haq et al. [6] measured the efficiency of seven algorithms for ml, i.e., k-nearest neighbors (knn), lr, decision tree (dt), nb, artificial neural network (ann), svm, random forest (rf), and measured the efficiency of three algorithms for selecting features, i.e., relief feature selection, minimal-redundancy-maximal-relevance (mrmr), and least absolute shrinkage and selection operator (lasso) to analyze the best classifier for heart diseases. lr is proven to be the most accurate classifier with the accuracy of 84% before applying feature selection. svm achieves the accuracy of 80% after applying feature selection algorithms. li et al. [10] applied different ml approaches to classify heart diseases. they proposed a diagnostic system that was accurate and efficient in producing the significant results. the heart disease dataset was pre-processed by removing the null values. they applied feature selection techniques to find the appropriate outputs. the result of experiment clarifies that the proposed feature selection algorithm (fcm/m) is suitable to diagnose the heart disease with svm. as compared to previous methods, the proposed system (srm-fcm/m) achieves better precision. latha and jeeva [11] presented a method of ensemble classification on cleveland heart dataset to increase the accuracy of weak algorithms by combining multiple classifiers [12]. the study found that ensemble techniques such as bagging and boosting effectively improved the predicted accuracy of weak classifiers in identifying the risk of heart disease. a maximum increase of 7% accuracy was achieved with the help of ensemble classification. raza et al. [13] evaluated different ml techniques and combined them with different feature selection methods to find the best technique for classification [14]. they applied four heart disease datasets and different methods like principal component analysis, chi-squared testing, relief, and symmetrical uncertainty to create distinctive feature sets [15]. by doing this, they achieve the significant accuracy of 85% on all datasets. alaa et al. [16] performed a comparison analysis of the ml techniques 25 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 24-34 based on autoprognosis framework, traditional techniques, and non-traditional techniques on the improvement of cvd and risk predictions. the research found that the ml based autoprognosis framework has the highest accuracy rate for cvd risk prediction as compared to traditional and non-traditional techniques. soni [17] compared different methods for heart disease prediction. the author combined ml methods with data mining techniques to be used in clinical decision support system. nb, sequential minimal optimization (smo), rf, decision table algorithms were used for the detection of heart diseases. few parameters including electrocardiogram (ecg) results were also employed. moreover, the author focused on multiparameter classification using weka tool. different parameters are analyzed, including ecg and heart rate variability (hrv). by using these data, nb and dt achieved good diagnostics accuracy of both 85% in predicting heart diseases. fitriyani et al. [18] proposed a method for heart disease diagnosis at the initial stage. they designed a module named clinical decision support systems (cdss) for heart disease prediction. moreover, they also conducted a comparative analysis with other medical modules and achieved the classification accuracy of 95.90% and 98.40% for statlog and cleveland datasets, respectively. 3. materials and methods 3.1. dataset for this study, cvd dataset [19] is used to develop a cvd diagnosis system based on ml. the sample size of the dataset is 70 thousand patients containing 11 features and one target value. the input features have three types: objective (realistic information), examination (result of medical checkup), and subjective (patient’s information). there are two classes on the target output mark, which reflect whether a cardio patient or a regular subject is the individual entity being examined. therefore, the extracted dataset consists of a matrix of 70k/12 attributes. the complete information and definition of 70k instances with 12 dataset characteristics is given in table 1. table 1 feature information of cvd dataset id parameters short code description 1 age ag age in days 2 height hgt height in cm 3 weight wgt weight in kg 4 gender gen male = 2; female = 1 5 systolic blood pressure sb_hi upper value (mmhg) 6 diastolic blood pressure db_lo lower value (mmhg) 7 cholesterol chl 1: normal; 2: above normal; 3: well above normal 8 glucose glu 1: normal; 2: above normal; 3: well above normal 9 smoking smk 1 = yes; 0 = no 10 alcohol alc 1 = yes; 0 = no 11 physical activity act 1 = yes; 0 = no 12 presence or absence of cvd cardio 1 = yes; 0 = no 3.2. the proposed methodology the proposed framework is designed (fig. 1) to classify cardio patients. for this study, the performance of variant kinds of ml predictive models for cvd diagnosis on full features of datasets is tested. the well renowned ml classifiers, i.e., svm, lr, nb, and dt (j48), are applied in the system. the model’s validation and performance evaluation matrices are computed. the methodology of the proposed system is divided into four phases. in the first phase, the pre-processing of cvd dataset is done by removing the duplicates and null values. secondly, the dataset is divided with the ratio of 85% for training and 15% for testing process. predictive ml classifiers are applied in the third phase. in the fourth and the final phase, the classifier’s performance is calculated in terms of sensitivity, specificity, matthews correlation coefficient (mcc), and f1-score. 26 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 24-34 fig. 1 framework for cvd prediction 3.3. dataset pre-processing dataset pre-processing is a crucial step for achieving the significant outputs of ml classifiers, which should be trained and tested in a powerful manner. pre-processing techniques such as standard scalar and removal of missing values are enforced to the dataset for adequate use in the classifiers. all these data pre-processing techniques are used in this study. 3.4. machine learning classifiers classification is the process of predicting the class of given data points. ml classification algorithms are used to classify the data into their respective domains. in this study, four well-known predictive ml classifier algorithms with their hypothetical background are presented. 3.4.1. logistic regression lr algorithm is used to observe the discrete set of classes [20]. in lr, binary and multiclass classification methods are performed. in binary classification, the prediction is completed by the predictive variable y when y ∈ �0,1�, where 0 is known as negative class and 1 is positive class. in multiclass classification, the prediction is applied to the predictive value y when y ∈ �0,1,2,3�. in order to classify binary classes [0, 1], the hypothesis �� ��� is devised, and the threshold output value is �� at 0.5. if the hypothesis �� � 0.5, it will predict y = 1; if the value is less than 0.5, it will predict y = 0. the hypothesis tends to limit the cost function between 0 and 1, e.g., 0 � �� � 1. the sigmoid function of lr can be written as: 1 1 ( ) xh x fθ −+= ℓ (1) lr is applied on the dataset to measure the probability of the occurrence of binary outcomes such as healthy (0) and cardiac (1) patient. therefore, eq. (1) of the sigmoid function is used to convert the outcomes into the categorical values. similarly, the cost function of lr can be written as: 1 1 1 ( ) cost ( , ) i i i j h x y m θ θ = =   ∑ (2) eq. (2) shows that the cost function takes two parameters in the input. ������ is hypothesis function, ������ is the output, and m represents the number of samples. the cost function is used to measure how badly the lr model is performing for predicting the relationship between y and x. before fitting the parameters to training the data with eq. (2), a constraint is added to prevent the model from overfitting. however, gradient descent (gd) as the optimization algorithm is applied by minimizing j(θ) as a function of θ to find optimal parameters. it takes partial derivative of j with respect to θ, and updates θ via each iteration with a selected learning rate of 0.5 until gd is converged. 27 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 24-34 3.4.2. support vector machine svm is considered to be the most powerful ml technique. svm is a supervised learning algorithm which is mostly used for classification problems as support vector classification (svc) [21]. this approach seeks to classify hyperplanes that are capable of unravelling datasets into elevated-dimensional spaces of features. the margin is called the distinction between datasets, and the svm maximizes the margin [22]. when dealing with binary classification problems, the instances are alienated with the hyperplane �� × � +� 0, where w is known as dimensional coefficient vector, b is the offset value, and x is the dataset value. some of the popular kernel choices are linear ��� × ���, polynomial ���� × �� +�� �, � > 0, ! � 2�, and radial basis function exp%−� ∥ �� −�( ∥ )* [23]. the purpose of using svm is to identify a hyperplane in binary dimensional space that distinctly classifies the data points. therefore, to separate the binary classes of data points, hyperplanes use the maximum distance between data points of both classes. the linear separable hyperplane is used for decision boundaries that assist in classifying the data points. the data points falling on either side of the hyperplane can be attributed to different classes. the size of the hyperplanes is dependent on the numbers of input features. when there is a misclassification done by the model on the prediction of the class of the data point, the loss is included along with the regularization parameter to perform gradient update. however, the required features are extracted from the dataset and split into the training and testing data. the learning rate of the model is set to 0.0001 and the regularization parameter λ is set to 1/epochs. therefore, the regularizing value reduces the number of epochs increased. 3.4.3. naive bayes nb is an algorithm for supervised learning based on the conditional probability theorem to evaluate the class function vector [24]. nb algorithm assumes that all the variables in the dataset are not correlated to each other. it can be written as: ( | ) ( ) ( | ) ( ) p x c p c p c x p x × = (3) where p(c|x) is the posterior probability of class c; p(c) is the probability of class c; p(x|c) is the likelihood probability of the predictor given class c; p(x) is the prior probability of predictor. when working with the dataset, the values of the dataset associated with each class are distributed according to a normal distribution. the developing process of nb model is that the data is defined by a normal distribution with no co-variance between dimensions. this nb model is fit by finding the mean and standard deviation of the points within each label. the cvd database of various patients is based on 11 columns including gender, age, height, and weight (table 1). the target is whether the patient is normal or having cardiac problems. if the targeted column p(c) has value 1, then it means that the person has heart problems, and if the value is 0, then it means that the person is healthy. 3.4.4. decision tree classifier the dt (j48) supervised algorithm is based on tree shape structures where every node is a decision node [25]. decisions are taken on the basis of an algorithmic method that identifies the ways to divide a dataset based on diverse conditions [26]. in dt, internal and external nodes are linked together, which make the decision done easily and efficiently. 4. results and discussion this section contains the outcomes produced by the ml classifiers. the outcomes of ml classifiers are evaluated in terms of different performance evaluation matrices. the cross-validation (cv) approach is applied, and the performance matrices of each ml classifiers are also calculated. the details are given in the following subsections. 28 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 24-34 4.1. k-cross-validation cv is probably considered to be the most important techniques used to validate the stability of the ml model on new data [27]. in k-fold cv, the dataset is separated into k equal size in which k-1 groups are used for training and the rest is for testing purposes. the performance of the classifiers is based on the k results. in this study, k =15 cv is applied, which means that 85% data is applied for training process and the rest is for testing the classifiers. 4.2. performance evaluation matrices various performance evaluation matrices are used in this study. the binary confusion matrix 2 × 2 is used because two categories are included in the proposed output variables. in addition, it provides two types of correct and incorrect prediction. a confusion matrix is shown in fig. 2. fig. 2 confusion matrix classification accuracy shows the performance, and the classification error shows the incorrect classification of the model. they can be written as: fp fn classification error tp tn fp fn + = + + + (4) tp tn accuracy tp tn fp fn + = + + + (5) sensitivity is also known as “true positive (tp) rate”. it confirms that if a diagnostic test is positive, then the subject is a cardio patient. it can be written as: 100% tp sensitivity tp fn = × + (6) specificity is the “true negative (tn) rate” classifier for detecting negative cases. when the diagnostic test is negative, the person is healthy. it can be mathematically written as: 100% tn specificity tn fp = × + (7) the mathematical expression of precision is written as: 100% tp precision tp fp = × + (8) mcc represents the prediction ability of a classifier with the values which vary between [−1, +1]. if mcc classifier has the value +1, this indicates that the predictions of the classifier are ideal; if mcc classifier has the value -1, it indicates that the classifier produces incorrect predictions; the value near 0 implies that random predictions are created by the classifier. mcc’s mathematical expression is written as: 29 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 24-34 ( ) ( ) ( ) ( ) tp tn fp fn mcc tp fp tp fn tn fp tn fn × − × = + + + + (9) receiver operating characteristic curve (roc) analysis compares the tp rate and false positive (fp) rate in the classification results while the area under the curve (auc) characterizes the roc of classifier. the larger the value of auc is, the more effective the performance of the classifier is. 4.3. results and discussion this section contains the outcomes generated by the classification models in terms of variant kinds of performance evaluation matrices. four different ml algorithms are tested for accuracy, i.e., nb, lr, j48, and svm on full features of cvd dataset. all computations are performed in weka software on an intel® core™ i7-6700 cpu @ 3.40 ghz personal computer with installed memory (ram) of 16 gb. the 15-fold cv test is applied, and the data is analyzed visually to figure out the distribution of values in terms of effectiveness and efficiency. the effectiveness of all four classifiers in terms of model building time, accuracy, correctly and incorrectly classified instances is calculated (table 2). the results of the 15-fold performance assessment of four classifiers nb, lr, j48, and svm on cvd dataset are shown in table 2. the processing time of each classifier with nb is 0.16 seconds in fig. 3, which is very fast computationally compared to the lr, j48, and svm. furthermore, the processing time of svm is 148.93 seconds which indicates that svm is a lazy learner. from comparative graph (fig. 4), it is easily noted that j48 has the highest value of correctly classified instances and the lower value of incorrectly classified instances while nb has the lowest value of correctly instances and the higher value of incorrectly classified instances. in table 3, nb shows poor performance that has 58.87% classification accuracy, 27.77% specificity, 58.9% sensitivity, and 22.6% mcc. the nb regression specificity value is 27.77%, indicating that the likelihood of diagnostic tests is negative, and the individual does not have cvd. in addition, 58.9% sensitivity shows the likelihood that the diagnostic test is positive for the person. for lr, the experiment is performed by using 15-fold cv. however, the performance of lr is good as shown in table 3. table 2 evaluation of classifiers in term of time, correctly instances, and incorrectly instances evaluation criteria classifiers nb lr j48 svm time taken to build a model 0.1 1.28 3.33 148.93 correctly classified instances 41209 50502 51159 45405 incorrectly classified instances 28791 19498 18841 24595 0.16 1.28 3.33 148.93 0 20 40 60 80 100 120 140 160 nb lr j48 svm time in seconds fig. 3 classifier processing time in seconds 30 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 24-34 fig. 4 comparative graph of correctly and incorrectly classified instances table 3 classifier 15-fold cv classification performance evaluation evaluation criteria classifiers nb lr j48 svm accuracy (%) 58.87 72.1457 73.0843 64.8643 specificity (%) 27.77 67.79 70.19 59.87 sensitivity (%) 58.9 72.1 73.1 64.9 mcc 0.226 0.445 0.462 0.299 precision (%) 64.4 73.3 73.2 65 the lr classifier achieves 72.1457% accuracy, 67.79% specificity, 72.1% sensitivity, and 44.5% mcc. similarly, svm has the 64.86% classification accuracy, 59.87% specificity, 64.9% sensitivity, and 29.9% mcc. the performance of j48 is excellent which achieves the highest 73.0843% classification accuracy, the best 70.19% specificity, 73.1% sensitivity and 46.2% mcc value. the output of j48 outperforms the other three classifiers in terms of precision, sensitivity, specificity, and mcc, as shown in fig. 5. j48 has 73.0843% predictive accuracy, 73.1%sensitivity, and 70.19% specificity. lb, which has 72.1547% classification precision, is the second significant classifier. the worst performance is observed in terms of precision, sensitivity, and specificity for nb classifiers. in fig. 6, the accuracy 73.0843% obtained by j48 is greater than the accuracy obtained by nb, lr, and svm (which varies between 58.87% and 72.1457%). it is observed that j48 and lr have high mcc values, while nb and svm have the lowest mcc values with 15-fold cv on the full feature’s dataset, as depicted in fig. 7. therefore, it means that j48 and lr are the highest prediction ability classifiers as compared to nb and svm. simulation errors are found to test the better performance of classifiers. to do so, the measured efficacy of all four classifiers is in provisos of kappa statistic (ks), mean absolute error (mae), root mean squared error (rmse), relative absolute error (rae), and root relative squared error (rrse). ks, mae, and rmse show numeric values while rae and rrse are in percentage. the detailed results are shown in table 4. the graphical representation for simulation of errors is presented in fig. 8. 0 10000 20000 30000 40000 50000 60000 nb lr j48 svm correctly classified instances incorrectly classsified instances 0 10 20 30 40 50 60 70 80 nb lr j48 svm accuracy (%) specificity (%) sensitivity (%) fig. 5 performance in terms of precision, specificity, and sensitivity of classifiers 31 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 24-34 table 4 summary of simulation errors evaluation criteria classifiers nb lr j48 svm ks 0.17 0.44 0.4617 0.29 mae 0.41 0.39 0.3611 0.35 rmse 0.53 0.43 0.43 0.59 rae (%) 83.85 79.09 72.21 70.27 rrse (%) 106.10 87.60 86.91 118.55 from table 4, the outcomes represent the chance of having a best classification (0.4617) with the least warning error rate (0.3611) formed by j48. furthermore, the second-best classification (0.4429) with error rate (0.3955) is produced by lr. nb and svm have the highest value of error rate, which explain the large number of incorrectly classified instances for each algorithm. the value of mae is observed to evaluate the quality of the classifiers. from table 4, it is clearly identified that j48 and svm have minimum mae values as compared to other classifiers nb and lr (fig. 9). 58.87 72.1457 73.0843 64.8643 0 10 20 30 40 50 60 70 80 nb lr j48 svm accuracy 0.226 0.445 0.462 0.299 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 nb lr j48 svm mcc 0 0.2 0.4 0.6 0.8 1 1.2 ks mae rmse rae rrsse nb lr j48 svm fig. 6 classifier accuracy test with 15-fold cv on complete dataset features fig. 7 mcc of all four classifiers with 15-fold cv fig. 8 compared simulation error with respect to ks, mae, rmse, rae, and rrse 32 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 24-34 j48 and svm have the nearest predictive values to the target values as compared to lr and nb. the confusion matrix is used for evaluating the performance of a classification model. the confusion matrix of all four classifiers is presented in fig. 10 below. (a) j48 (b) lr (c) svm (d) nb fig. 10 confusion matrix 5. conclusions and future work in this modern era, the amount of data collected from health surveillance has rapidly increased with the advancement of the medical technology. it is foreseeable that ai platform will enable the analysis of multiple infectious diseases including cvd, which is considered to be one of the main causes of death worldwide. in this study, variable generalization performance of supervised learning algorithms for cvd diagnosis was evaluated on cvd dataset by using four well-known supervised ml classifiers. in accordance with the obtained output, the higher probability of achieving an improved result can be inferred in terms of the accuracy of the cvd classification when j48 is applied. in the future work, the ensemble of pre-trained models would be applied for classifying cvd. conflicts of interest the authors declare no conflict of interest. statement of ethical approval for this type of study, statement of human rights is not required. statement of informed consent for this type of study, informed consent is not required. references [1] j. mackay, g. a. mensah, s. mendis, and k. greenlund, the atlas of heart disease and stroke, geneva: world health organization, 2004. [2] z. s. wong, j. zhou, and q. zhang, “artificial intelligence for infectious disease big data analytics,” infection, disease, and health, vol. 24, no. 1, pp. 44-48, february 2019. 0.3 0.32 0.34 0.36 0.38 0.4 0.42 0.44 nb lr j48 svm mean absolute error fig. 9 mae for nb, lr, j48, and svm 33 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 24-34 [3] s. schneeweiss, “learning from big health care data,” the new england journal of medicine, vol. 370, no. 23, pp. 2161-2163, june 2014. [4] a. blumenthal, “artificial intelligence to fight the spread of infectious diseases,” https://phys.org/news/2018-02-artificial-intelligence-infectious-diseases.html, february 20, 2018. [5] k. e. goodman, j. lessler, s. e. cosgrove, a. d. harris, e. lautenbach, j. h. han, et al., “a clinical decision tree to predict whether a bacteremic patient is infected with an extended-spectrum β-lactamase-producing organism,” clinical infectious diseases, vol. 63, no. 7, pp. 896-903, october 2016. [6] a. u. haq, j. p. li, m. h. memon, s. nazir, and r. sun, “a hybrid intelligent system framework for the prediction of heart disease using machine learning algorithms,” mobile information systems, vol. 2018, 3860146, december 2018. [7] m. elsisi, k. mahmoud, m. lehtonen, and m. m. darwish, “reliable industry 4.0 based on machine learning and iot for analyzing, monitoring, and securing smart meters,” sensors, vol. 21, no. 2, 487, january 2021. [8] m. q. tran, m. k. liu, and m. elsisi, “effective multi-sensor data fusion for chatter detection in milling process,” the international society of automation transactions, in press. [9] m. elsisi, m. q. tran, k. mahmoud, d. e. a. mansour, m. lehtonen, and m. m. darwish, “towards secured online monitoring for digitalized gis against cyber-attacks based on iot and machine learning,” institute of electrical and electronics engineers access, vol. 9, pp. 78415-78427, june 2021. [10] j. p. li, a. u. haq, s. u. din, j. khan, a. khan, and a. saboor, “heart disease identification method using machine learning classification in e-healthcare,” institute of electrical and electronics engineers access, vol. 8, pp. 107562-107582, june 2020. [11] c. b. c. latha and s. c. jeeva, “improving the accuracy of prediction of heart disease risk based on ensemble classification techniques,” informatics in medicine unlocked, vol. 16, 100203, july 2019. [12] k. w. johnson, j. torres soto, b. s. glicksberg, k. shameer, r. miotto, m. ali, et al., “artificial intelligence in cardiology,” journal of the american college of cardiology, vol. 71, no. 23, pp. 2668-2679, june 2018. [13] a. raza, a. mehmood, s. ullah, m. ahmad, g. s. choi, and b. w. on, “heartbeat sound signal classification using deep learning,” sensors, vol. 19, no. 21, 4819, january 2019. [14] r. spencer, f. thabtah, n. abdelhamid, and m. thompson, “exploring feature selection and classification methods for predicting heart disease,” digital health, vol. 6, pp. 1-10, march 2020. [15] k. shameer, k. w. johnson, b. s. glicksberg, j. t. dudley, and p. p. sengupta, “machine learning in cardiovascular medicine: are we there yet?” heart, vol. 104, no. 14, pp. 1156-1164, july 2018. [16] a. m. alaa, t. bolton, e. di angelantonio, j. h. rudd, and m. van der schaar, “cardiovascular disease risk prediction using automated machine learning: a prospective study of 423,604 uk biobank participants,” public library of science one, vol. 14, no. 5, e0213653, may 2019. [17] v. d. soni, “detection of heart disease using machine learning techniques,” international journal of scientific and technology research, vol. 9, no. 8, pp. 285-288, august 2020. [18] n. l. fitriyani, m. syafrudin, g. alfian, and j. rhee, “hdpm: an effective heart disease prediction model for a clinical decision support system,” institute of electrical and electronics engineers access, vol. 8, pp. 133034-133050, july 2020. [19] h. malik, m. s. farooq, a. khelifi, a. abid, j. n. qureshi, and m. hussain, “a comparison of transfer learning performance versus health experts in disease diagnosis from medical imaging,” institute of electrical and electronics engineers access, vol. 8, pp. 139367-139386, june 2020. [20] b. alić, l. gurbeta, and a. badnjević, “machine learning techniques for classification of diabetes and cardiovascular diseases,” 6th mediterranean conference on embedded computing, pp. 1-4, june 2017. [21] x. zhai and c. tin, “automated ecg classification using dual heartbeat coupling based on convolutional neural network,” institute of electrical and electronics engineers access, vol. 6, pp. 27465-27472, may 2018. [22] s. r. saufi, z. a. b. ahmad, m. s. leong, and m. h. lim, “challenges and opportunities of deep learning models for machinery fault detection and diagnosis: a review,” institute of electrical and electronics engineers access, vol. 7, pp. 122644-122662, august 2019. [23] m. mohammed, m. b. khan, and e. b. m. bashier, machine learning: algorithms and applications, boca raton: chemical rubber company press, 2017. [24] r. y. goh and l. s. lee, “credit scoring: a review on support vector machines and metaheuristic approaches,” advances in operations research, vol. 2019, 1974794, march 2019. [25] n. friedman, d. geiger, and m. goldszmidt, “bayesian network classifiers,” machine learning, vol. 29, pp. 131-163, november 1997. [26] m. mohri, a. rostamizadeh, and a. talwalkar, foundations of machine learning, cambridge: massachusetts institute of technology press, 2018. [27] r. r. isnanto, a. f. rochim, d. eridani, and g. d. cahyono, “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. 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/). 34 4___peti#7653__35-56 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 35-56 artifact removal methods in eeg recordings: a review mariyadasu mathe 1, 3, * , padmaja mididoddi 2 , battula tirumala krishna 1 1 department of electronics and communication engineering, jawaharlal nehru technological university kakinada, andhra pradesh, india 2 department of electronics and communication engineering, velagapudi ramakrishna siddhartha engineering college, andhra pradesh, india 3 gandhi institute of technology and management, andhra pradesh, india received 10 may 2021; received in revised form 21 july 2021; accepted 22 july 2021 doi: https://doi.org/10.46604/peti.2021.7653 abstract to obtain the correct analysis of electroencephalogram (eeg) signals, non-physiological and physiological artifacts should be removed from eeg signals. this study aims to give an overview on the existing methodology for removing physiological artifacts, e.g., ocular, cardiac, and muscle artifacts. the datasets, simulation platforms, and performance measures of artifact removal methods in previous related research are summarized. the advantages and disadvantages of each technique are discussed, including regression method, filtering method, blind source separation (bss), wavelet transform (wt), empirical mode decomposition (emd), singular spectrum analysis (ssa), and independent vector analysis (iva). also, the applications of hybrid approaches are presented, including discrete wavelet transform adaptive filtering method (dwt-afm), dwt-bss, emd-bss, singular spectrum analysis adaptive noise canceler (ssa-anc), ssa-bss, and emd-iva. finally, a comparative analysis for these existing methods is provided based on their performance and merits. the result shows that hybrid methods can remove the artifacts more effectively than individual methods. keywords: eeg signal, canonical correlation analysis (cca), ssa-bss, emd-iva 1. introduction electroencephalogram (eeg) is a voltage test recording the electrical activity of the neurons in the brain. eeg is obtained in various frequencies including delta (1-4 hz), theta (4-8 hz), alpha (8-13hz), beta (13-30 hz), and gamma (greater than 30 hz) [1]. eeg brain rhythms with different frequency ranges are shown in table 1 [2]. eeg is suitable for brain signal acquisition because of its non-intrusiveness, portability, high temporal resolution, and cost-effectiveness. brain signal acquisition and processing are mainly used for clinical research purposes to diagnose multiple brain disorders, e.g., sleep disorders, depression, schizophrenia, and epileptic disorder [3]. however, eeg processing and analysis are affected by a couple of conditions, such as technical and physiological conditions. during the signal acquisition process, the most prevalent disorders (artifacts) are ocular, muscle, and cardiac disorders, and electrooculogram (eog) is cumbersome [4]. table 1 basic brain rhythms and their frequency range [2] serial number brain rhythm frequency range state of the brain 1 delta 1-4 hz sleeping/unconscious 2 theta 4-8 hz imagination 3 alpha 8-13 hz calm consciousness 4 beta 13-30 hz focused consciousness 5 gamma >30 hz peak performance * corresponding author. e-mail address: mmathe@gitam.edu tel.: 9866420248 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 35-56 these artifacts, power line interference, impedance fluctuation, and wire movement collide and intrude the energy with eeg signals. due to these artifacts, it is difficult to obtain the correct diagnosis and analysis of eeg. therefore, artifact removal methods are required to reduce the artifacts and preserve the originality of the actual eeg signals. the placement of electrodes in the 10-20 system is shown in fig. 1. fig. 1 depicture of 10-20 electrode placement system [5] researchers developed automated approaches and techniques wherein the eeg signals are recorded using singleand multi-channels. different methods are proposed for the mitigation of artifacts from singleand multi-channels. the accuracy of the methods using single-channels is less than the one using multi-channels, but the complexity over time is less when using single-channels. the basic process includes recording the eeg data with precaution and removing the epochs in eeg data, which can cause the removal of some valuable data [6]. generally, artifact removal techniques are categorized into two types: regression-based method and blind source separation (bss) based method [7]. so far, no specific method can remove all the artifacts. researchers are trying to suppress the artifacts by improving the existing techniques and combing two and more methods to get the artifact-minimized eeg signals. the main objective of this study is to give an overview on the existing methodology to remove the artifacts in eeg recordings and to choose the better method based on artifact removal types. the remaining paper is organized as follows. the artifacts affecting the eeg recordings are discussed in section 2. literature review is clearly explained in section 3. the primary artifact removal techniques’ advantages and disadvantages proposed by various researchers are discussed in section 4. the comparative analysis for multiple artifact removal methods is presented in section 5. 2. about the artifacts human brain is a combination of billions of neurons, and the activity of this brain region is measured by recording the electrical signals generated by the neurons. eeg is such a signal recording acquired by placing the electrodes on the scalp. the recorded eeg signals include the signals from the brain neurons and the neurons of other body parts. the signals generated by other than brain neurons are called artifact signals. these artifacts are due to ocular artifacts (oas), muscle artifacts, cardiac artifacts, electrode placement, impedance mismatching, etc. [8]. non-physiological artifacts can be removed by simple methods such as high-pass filtering and low-pass filtering because these artifacts’ frequency does not overlap with the desired eeg signals [9]. however, the removal of physiological artifacts requires different methods because the frequency spectrum of physiological artifacts overlaps with the desired eeg signals [10]. oas are generated from eye movements (ems) and eye blinks, and are propagated to the scalp region and overlap with the eeg signals. the oa frequency range is petite and is recorded by eog by placing the electrodes (one above the eye and the other on one side of the eye) as shown in fig. 2. eeg signals are deteriorated by eog signal components and vice versa, so bi-directional interference is introduced [11]. the amplitude of eog is higher than the amplitude of eeg. muscle artifacts are originated from different muscle groups such as contraction, stretch, talking, sniffing, swallowing, etc. the frequency range of 36 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 35-56 muscle artifacts is vast and is overlapped entirely with the actual eeg signals. the amplitude of muscle artifacts depends on the activity of the muscle and is measured by electromyography (emg). because of the broad spectrum, muscle artifacts cannot be obtained by single-channels. the autocorrelation of muscle artifacts is very small and independent of the eeg signals both temporally and spatially [8]. cardiac artifacts are originated from the heart and are measured by electrocardiography (ecg). cardiac artifacts follow some regular patterns, so removing the cardiac artifacts is simple by using reference signals. fig. 2 eog and emg elements of a sleep montage setup [5] for the better analysis of human brain and the diagnosis of patients, artifacts must be removed. generally, the recorded eeg signals are between the ranges of 0.01 hz to 100 hz. different types of the brain rhythms and the artifacts generated by the brain and body are shown in fig. 3 and fig. 4. (a) delta wave (b) theta wave (c) alpha wave (d) beta wave (e) gamma wave fig. 3 graphical representation of different brain rhythms fig. 4 different physiological artifacts [8] 37 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 35-56 3. literature survey 3.1. related works in 2013, zou et al. [12] have addressed an automatic algorithm for general artifact identification. it was composed of electrode-scalp impedance information and an event-associated feature-oriented clustering algorithm. the former identified the non-biological artifacts, and the latter identified the artifacts with physiological origins. it could efficiently remove, separate, and detect the non-biological and physiological artifacts. it efficiently improved the signal quality. the classification accuracy was also improved. mahajan and morshed [13] have proposed a uniquely robust, unsupervised, and computationally speed statistical algorithm that employed modified multiscale sample entropy (mmse) and kurtosis to recognize the independent eye blink artifactual components automatically. these components were denoised by the biorthogonal wavelet decomposition. the persistent neural activity was preserved in the independent components (ics), and the artifactual activity was removed. their proposed method attained average specificity, sensitivity, and execution time, which are better than the ones in other previously existing methods. in 2014, acharjee et al. [14] have described the conflict of gradient artifact removal from the eeg signals recorded in a concurrent manner with functional magnetic resonance imaging (mri) acquisition. the artifacts were estimated by the quasi periodicity and its similarity over distinct channels with the help of independent vector analysis (iva). their proposed method utilized spatio-temporal information to evaluate the artifacts for a specific channel with the support of spatial dependences. hence, it offered robustness in terms of uncontrollable variations such as fluctuations and head movement during acquisition. in 2015, khatun et al. [15] have evaluated the unsupervised wavelet transform (wt) approach for removing oas in the case of a single-channel eeg system. a group of 7 raw eeg dataset was investigated. two familiar wt techniques, i.e., stationary wavelet transform (swt) and discrete wavelet transform (dwt), were used. four wt basis functions were utilized for removing oas, i.e., bior4.4, sym3, coif3, and haar, having statistical threshold (st) and universal threshold (ut). their proposed method was an efficient tool for real-time applications. in 2016, tibdewal et al. [16] have performed detection of em artifacts and classification of non-artifactual/artifactual eeg time series together with multiple em artifactual zones. artificial neural network (ann) was employed for the identification. the researchers also proposed a computationally fast, robust, simple, and novel time-amplitude algorithm. their proposed method provided the ann model accuracy for the classification of the artifacts, and detected and marked the multiple em artifactual zones automatically and appropriately. liu et al. [17] have introduced a practical and efficient filtering algorithm based on multiscale entropy and multivariate empirical mode decomposition (memd) for computing the depth of anesthesia (doa). the mean entropy found the artifact-free intrinsic mode functions (imfs). the impact of distinct levels was investigated by means of the simulated data. the complexity was also computed for monitoring the doa. the ann was used for the range mapping for correlating the measurements with bispectral index (bis). the bis and entropy values contained a more robust correlation, and it also efficiently filtered the artifacts. in 2017, barua et al. [18] have proposed an algorithm which is known as automated artifact handling in eeg (arte). it was employed as a pre-processing step in the driver monitoring application, and functioned based on the wavelets, hierarchical clustering, and independent component analysis (ica). it was analyzed on a driver sleepiness study, and was evaluated with the help of a clinical neurophysiologist using quantitative measures. the outcomes demonstrated that it minimized the effect of artifacts in the recorded eeg signals. it was a data-driven one, and it did not depend on manually defined thresholds or extra reference signals. sai et al. [19] have introduced a new technique for artifactual component identification with the help of a pre-trained support vector machine (svm). the researchers also proposed an extendable and robust system that enabled removing the artifacts without arbitrary thresholding. the target artifacts were removed in a successful manner. the training 38 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 35-56 used a group of features such as amplitude range, shannon’s entropy, variance, and kurtosis. the multiple forms of artifacts in the multi-channel eeg were also accommodated. li et al. [20] have introduced a unified framework based on canonical correlation analysis (cca) to remove artifacts. the raw signals constructed a pair of matrices based on the sources via the autocorrelation maximization. these sources were eradicated by placing them as zeros, and the remaining sources reconstructed the artifact-free eeg. the actual and simulated recorded data assessed this framework. it lessened the associated motion artifacts and contained benefits for eliminating the gradient artifacts with respect to the computational complexity and combination of characteristics. roy and shukla [21] have compared the effectiveness of enhanced empirical mode decomposition (eemd) with distinct interpolations based on an artifact removal technique. the input single-channel eeg signals were transformed to multi-channel signals, and the motion artifact randomness was eradicated by dwt filtering and cca. their proposed method acted as a supplement for the traditional algorithms and provided enhancements in difference signal to noise ratio (dsnr) and several performance parameters. in 2019, chang et al. [22] have evaluated artifact subspace reconstruction (asr) during the simulated driving experiments. the ic classifier and ica separated the artifacts for assessing the efficiency of asr. the temporal activity power was minimized. it enhanced the quality of a subsequent ica decomposition. the brain activities were retained, and non-brain signals were removed. their proposed method was an automatic and powerful artifact removal technique for online real-time eeg applications or offline data analysis like brain-computer interfaces and clinical monitoring. cheng et al. [23] have addressed an efficient technique for removing different artifacts for the single-channel eeg. it was a combination of second-order blind identification and singular spectrum analysis (ssa) method. the possible cases contaminated by ecg, eog, and emg artifacts were analyzed. their proposed method removed different artifacts and acted as a promising tool for biomedical signal processing applications. saini et al. [24] have introduced a robust framework for detecting and eliminating oas based on turning point count and variational mode decomposition (vmd). it exploited the efficiency in two stages, i.e., vmd-i and vmd-ii. it was composed of four components: eeg signal decomposition into two modes using vmd-i, rejection of low-frequency baseline components, processed eeg signal decomposition into three modes using vmd-ii, rejection of mode containing oas based on turning point count-based threshold criteria. the oas were eliminated with minimal loss in the entire local rhythms and reconstructed eeg signals. liu et al. [25] have labeled a robust and effective muscle artifact removal technique with the help of fast multivariate empirical mode decomposition (fmemd) and cca in the case of few-channel eeg. the input eeg recordings were decomposed into various multivariate imfs. the cca processed the imfs for measuring the underlying sources. in the final step, the sources with less autocorrelations were eradicated. the average correlation coefficient (cc) and the accuracy were largely consistent. their proposed method worked much robust in the low sampling rate based on the benchmark and real data. dora and biswal [26] have suggested an ecg artifact correction algorithm for the automatic diagnosis and analysis of single-channel eeg signals. it employed a modified and enhanced signal decomposition version for attaining band limited intrinsic mode functions (blimfs). it was helpful when the signal was composed of correlated properties. it suppressed the qrs complexes and estimated the ecg reference. it offered less distortion and was computationally more intensive than other methods. it overcomes the shortcomings such as the requirement of a reference ecg channel and the requirement of r-r interval or amplitude thresholding for qrs complex identification. in 2020, dora et al. [27] have introduced an adaptive ssa algorithm for muscle artifact removal. the mobility threshold was decided in an adaptive manner by neural network regressor (nnr). their proposed method was trained using several contamination levels. it discriminated among several contamination levels and performed superior to the traditional single-channel algorithms. yedurkar and metkar [28] have proposed a technique for removing physiological artifacts and 39 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 35-56 positioning the epileptic region. a hybrid method was recommended based on multi-resolutional analysis and adaptive filtering (mraf) . in the initial step, the eeg signals were decomposed by dwt that localized the epileptic region. the multi-resolutional soft thresholding removed the abrupt changes. moreover, the adaptive filtering (af) eradicated the low-frequency components that described the physiological artifacts. it retained most of the seizure signals that were available in the validated datasets. islam et al. [29] have introduced a technique for removing motion-related artifacts in the field of epilepsy. initially, it involved the eeg signal recording with the help of a wearable eeg headset. it was colored using few motion artifacts produced in a lab-controlled experiment. it was followed by the spectral and temporal spectral signal characterization and artifact removal by ica. the actual clinical eeg data tested it, and the outcomes described an average improvement in accuracy for detecting the seizures and prediction. bajaj et al. [30] have addressed an algorithm based on wavelet packet decomposition (wpd) that permitted the presumed artifact removal by optimizing intuitive parameters. it was composed of 2 tuning parameters and 3 operating modes. it was compared with comparative wavelet-oriented technique and ica-oriented techniques. the performance could be enhanced by proper optimization of the parameters for an individual predictive method. it was accomplished better for artifactual noise removal and neural information retention computed using cc and mutual information (mi) than the comparative wavelet-oriented technique. in 2021, noorbasha and sudha [31] have labeled the conflict of eog artifact removal and separation of different cerebral activities in a single-channel contaminated eeg. a new method based on ica and ssa together with swt was proposed. the single-channel contaminated eeg signals were transformed into multivariate information. then, ica separated the source signals as distinct ics. the swt accomplished the thresholding for preserving the eeg signals and dividing the actual artifacts. their proposed method provided better artifact separation characteristics than the traditional approaches. 3.2. utilized datasets several kinds of datasets are used in the considered previous research, including the manual dataset, mendeley, mit-bih polysomnographic database, eeg during mental arithmetic task database, synthetic dataset, simulated dataset, temple university hospital (tuh) eeg database, gold-standard dataset, epileptic eeg signal dataset from the local hospital, freiburg eeg dataset, semi-simulated dataset, real dataset, open-source dataset, etc. table 2 lists different kinds of datasets used in the considered works in a detailed manner. most of the considered works utilize the manual dataset, followed by a real dataset, simulated dataset, mit-bih polysomnographic database, etc. the manual dataset is used in 50% of the works, real dataset was used in 20% of the works, simulated dataset was used in 15% of the works, mit-bih polysomnographic database was used in 10% of the works, and the remaining datasets was used only in 5% of the works. table 2 different datasets used in the considered works dataset previous work manual dataset [12-15, 18-19, 22-23, 25, 31] mendeley [24] mit-bih polysmnographic database [24, 26] eeg during mental arithmetic task database [24] synthetic dataset [31] simulated dataset [17, 20, 27] tuh eeg database [27] gold standard dataset [28] epileptic eeg signal dataset from local hospital [28] freiburg eeg dataset [29] semi-simulated dataset [26] real dataset [16-17, 20, 30] open-source dataset [21] 40 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 35-56 3.3. simulation platforms the platforms used in the considered works are given in fig. 5. here, matlab and labview softwares are used for the implementation. 92% of the works use the matlab software, and 8% use the labview software. 3.4. performance measures the performance measures used in the analyzed research are depicted in table 3, including accuracy, sensitivity, specificity, mean square error (mse), root mean square error (rmse), relative root mean square error (rrmse), etc. among the considered works, time measure is used in 25% of the works, cc is used in 45% of the works, mi is used in 25% of the works, accuracy is used in 20% of the works, sensitivity is used in 25% of the works, specificity is used in 15% of the works, mse is used in 10% of the works, mean absolute error (mae) is used in 20% of the works, rrmse is used in 25% of the works, rmse is used in 15% of the works, signal to noise ratio (snr) is used in 30% of the works, power spectral density (psd) is used in 10% of the works, signal to artifact ratio (sar) is used in 10% of the works, and the remaining measures such as euclidean distance, normalized mean square error (nmse), spectral coherence, agreement rate, signal quality index (sqi), relative error (re), normalized root mean square error (nrmse), variance, power, maximum absolute error (max), normalized maximum absolute error (nmax), normalized root mean square difference (nrd), percentage root mean square difference (prd), autocorrelation, peak signal to noise ratio (psnr), precision, false alarm rate (far), coherence, mean entropy value (mev), dsnr, and spectral distortion improvement are used only in 5% of the contributions. table 3 performance measures used in the previous research ref. time cc mi accuracy sensitivity specificity mse mae rrmse rmse snr psd sar miscellaneous [12] √ euclidean distance [13] √ √ √ √ √ √ spectral coherence and agreement rate [14] √ [15] √ √ √ nmse [16] √ [17] √ √ mev [18] √ √ √ sqi, re, and nrmse [19] √ √ √ [20] √ √ √ [21] √ dsnr and spectral distortion improvement [22] variance and power [23] √ √ √ √ [24] √ √ √ √ √ √ max, nmax, nrd, and prd [25] √ √ √ √ autocorrelation [26] √ √ √ √ [27] √ √ √ psnr [28] √ √ √ precision [29] √ √ far [30] √ √ coherence [31] √ √ √ 92% 8% contribution matlab labview fig. 5 simulation platforms used in the literature 41 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 35-56 4. survey on the existing artifact removal techniques 4.1. artifact avoidance and rejection the first basic technique to remove artifacts is to instruct the patient to shun unnecessary actions such as ems and stay calm until the completion of the process. however, since it is not possible to keep quiet, artifacts are recorded in the eeg signals [3]. the second way is to remove the epochs in eeg data by observation. this method is also not available because the data is vast, and the expertise must be analyzed appropriately. therefore, a separate automatic procedure is needed to remove the artifacts as much as possible [32]. 4.2. regression methods the regression method is a straightforward and commonly used method for removing artifacts in eeg signals. in this method, a reference signal is required to remove artifacts. eeg signals are generally most affected by eog and emg signals. the regression method depends on the subtraction of artifact signals from the eeg signals. it assumes that oas and eeg signals are linearly uncorrelated [33]. identifying the reference artifact signals is very difficult because eeg and eog are both neural activities [2]. while subtracting eog signals, some useful information related to the original eeg signals was also removed; loss of data is there. the simple mathematical analysis is shown here in eq. (1). ( ) ( ) ( ) ; with 1, 2, ,oseeg l eeg l eog l l o= + = ⋯γ (1) here, the recorded eeg signals are denoted as eegs the fraction of the eog signals is denoted as �, and the original eeg signals are denoted as eego respectively. 4.3. filtering methods 4.3.1. adaptive filtering the performance of wt depends upon the mother wavelet effectively. the wt efficiency is improved if the artifact component’s frequency does not overlap with the usage frequency of the eeg signals. unfortunately, ecg, emg, and eog artifact frequencies are coinciding with the desired eeg signals. in ica, the improvement performance of artifact removal depends on the reference signal and a large amount of data. however, af overcomes the problem of extensive data and frequency band overlap [34]. fig. 6 presents a block diagram to implement af. fig. 6 basic block diagram of adaptive filter [34] more than one channel can be selected as a reference channel, and af is employed mainly to minimize power line interference artifacts. however, this has two particular limitations: (i) the noise content in the original signals must be corrected with some reference noise signals; (ii) the original signal helpful information may be uncorrected with the reference noise. 42 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 35-56 4.3.2. kalman filtering kalman filter (kf) is one of the appropriate estimators for the analysis of linear systems and signals. it is a recursive filtering algorithm to remove the artifacts in eeg signals. the inputs used for this algorithm are the state of the system, recorded signal, and past observations. in this algorithm, a reference signal is not required; kf estimates the state of the system and the reference signal from the contaminated part of the eeg signals. the recorded eeg signals are denoted as eegr, true eeg signals are denoted as eegt, and oas are denoted as eog. kf uses a state-space model to represent the complete analysis. the state-space model of the kf is as follows [35]: ( ) ( 1) ( 1) ( 1)x n ax n bu n cw n= − + − + − (2) ( ) ( ) ( )z n hx n v n= + (3) where z(n) is the measured signal; x(n) is the system’s state; a, b, c, and h are the matrices obtained from different regions of the acquired eeg signals; w(n) and v(n) are the process noise and measurement noise respectively; u(n) is the control input for the system. the measured signal is the combination of eegt and artifactual signals like eog [36]. ( ) ( ) ( ) t z n eeg eog hx n v n= + = + (4) after estimating eeg and eog components, kf is applied to remove the artifacts. in this algorithm, two steps are there. (1) estimate the state variables. (2) compare the previous estimation with the measured signal to obtain an improved version of the state variables. the disadvantage of kf is that it assumes the errors are gaussian, and the system is a linear model. the measurement of covariance matrices for process noise and measured noise is tough to calculate because the practical recorded eeg signals are not a linear combination of the underlying source signals. the extension of kf is extension kalman filter (ekf) and unscented kalman filter (ukf), which work on nonlinear systems. 4.4. blind source separation (bss) bss is a separation approach of underlying sources from the mixed signals without knowing prior information about the origin of signals and mixing parameters. the main plan of bss is to recover the sources from mixed signals [37]. let us consider x as an observed eeg signal from the scalp electrodes, which is a linear mixture of underlying sources (s) with unknown mixing matrix a, eq. (5) is obtained. x as= (5) bss algorithm is reverted to recover the underlying sources. y ws= (6) where y is the estimation of sources, and w is an un-mixing matrix. these algorithms are used to remove artifact components and to reconstruct artifact-free eeg signals. in the following section, some of the bss algorithms are discussed. 4.4.1. principle component analysis (pca) principle component analysis (pca) is one of the simple data reductions and bss techniques to remove the artifacts using the orthogonality principle [8]. berg and scherg proposed pca to minimize oas. pca is a multiple data analysis procedure that involves the transformation of many correlated variables into a smaller number of uncorrelated variables. pca is defined 43 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 35-56 as “linear projection that transforms multivariate data into a set of components called principle components (pcs)” [38]. pcs are calculated using a simple algorithm called singular value decomposition (svd). in svd, eigenvalues are computed from the covariance matrix (c) [36]. most significant eigenvalues are related to the highest variance, and the lowest eigenvalues are related to the lower variance. assume the recorded eeg signals (eegr) combine the original eeg signals (eego) and the artifacts from eog and emg, eq. (7) is obtained. r oeeg eeg eog emgα β= + × + × (7) where α and β are real constants. the drawback of pca is the orthogonality principle only because it cannot separate the artifacts if they have the same amplitude as original eeg data. the orthogonality principle is not always held due to the nonlinearity between the electrical signals generated by brain neurons from eye blinks and ems compared with original eeg signals [39]. the advancement in pca is done by robust pca and kernel pca [40]. the performance of pca is better compared to the regression method and filtering methods. pca can remove oas, but other artifacts cannot be removed. therefore, researchers preferred alternative methods such as ica, cca, empirical mode decomposition (emd), iva, ssa, etc. 4.4.2. independent component analysis (ica) ica is a powerful bss method to remove the artifacts in eeg signals. it converts the multi-dimensional data into different ics. in this method, the observed eeg signals’ primary assumption is a linear mixture of underlying independent sources [39]. pca is unable to differentiate the artifacts if the neural activity and artifacts have the same amplitude. assume that the recorded eeg signals x(k) use n channels and that the number of underlying sources s(k) are n [2], eq. (8) is obtained. ( ) ( )x k as k= (8) where x(k) = [x1(k), x2(k), …, xn(k)] t and s(k) = [s1(k), s2(k) , …, sn(k)] t . t and k are transpose operator and time index operator, respectively. a ϵ r nxn is mixing the unknown matrix. the goal of ica is to recover s from x without knowing a and s. for this, a separate un-mixing matrix w should be found such that eq. (9) is obtained. ( ) ( )y k w x k= (9) where y is the expected value of s. y is precisely equal to s if and only if w = a -1 . this model is by assuming no noise model. for experimental conditions, some noises are also included. a simple model of how the electrodes are placed to record the eeg signals from the brain is shown in fig. 7. fig. 7 a simple model of recording eeg signals by electrodes [2] in the preprocessing of eeg signals using ica, two main steps are there; one is centering, and the second is whitening [37]. in the centering step, the eeg data is centered by subtracting the mean from the data. in the whiting step, the eeg data is de-correlated. ica uses higher-order statistics (hos) to identify independent artifactual components. unlike pca, 44 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 35-56 orthogonality is not required; only statistical independence is needed. the main drawback of ica is that the number of channels used to collect the data should be huge compared to underlying artifactual sources. the second drawback is that ica removes the artifacts whose topologies are stereotyped like eog, but muscle artifacts are not stereotyped because the muscle artifacts are from different sources. 4.4.3. canonical correlation analysis (cca) hotelling in 1936 has proposed cca. the entire concept of this approach is to find the vectors for two dataset variables so that the correlation between the projections of the variables on this basis vector is mutually maximized. this analysis relies on the coordinate system where the variables are described; however, if there is a strong linear relationship between the sets of multi-dimensional variables, they may depend on the coordinate system. this correlation relation may not be visible, but it may seek a pair of linear transformations. each set of variables are transformed, and the corresponding coordinates are correlated to the maximum [41]. cca is also treated as one of the powerful tools used in bss to remove artifacts in eeg signals. this analysis mainly uses second-order statistics (sos) with a less computational cost than ica (which uses hos) [42]. cca removes muscle artifacts easily because muscle artifacts do not have stereotyped topography [43]. muscle artifacts are distributing a total range of frequency spectrum. the autocorrelation of muscle artifacts is very low, while the eeg signal autocorrelation is high. cca uses this property to remove the emg artifacts which are mainly affected in the frequency range of 20hz-50hz. consider that there are a multi-channel eeg signal x(t) and its delayed version y(t) such that y(t) = x(t-1). the basis vectors are x = w t x and y = w t y. in this scenario, cca maximizes the correlation between x and y by removing the mean of x and y [3]. , , [ ][ ] ( , ) max max max ( )( )[ ] [ ] [ ] [ ] t t tt x y x xy t tx y x yt t t t t t x xx y yy yx x x y w w w w e w w w c we xx x y w c w w c we xx e yy e w w e w w ρ = = = (10) where cxx and cyy are auto-covariance matrices of x and y, respectively. cxy is the cross-covariance matrix of x and y. by calculating the derivative of eq. (10) with respect to wx and wy and equating to zero, the following equations are derived. ^ ^ 1 1 2 x x xx xy yy yx c c c c w wρ− − = (11) ^ ^ 1 1 2 y y yy yx xx xy c c c c w wρ− − = (12) where ρ is the canonical cc. ρ is maximized such that artifact components can be removed. if ρ is minimized, then these components approach to artifactual features. 4.5. wavelet transform (wt) the observed eeg signals are not entire because physiological and non-physiological signals contaminate the observed signals. for the better analysis of eeg signals, artifact signals are removed. in pca and ica, the drawback is maintaining orthogonality and large data requirements. these drawbacks are overcome by wt. wt decomposes the time-domain eeg signals by convolving a scaled and translated wavelet function [44]. , 1 ( ) ( ) a b n b n aa ψ ψ − = (13) where a and b are real numbers. n = 1, 2, 3, …., n. dwt removes oas because the wavelets have resembled oas [45]. here, some of the basic wavelets are shown in fig. 8. 45 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 35-56 (a) haar (b) shannon or sinc (c) daubechies 4 (d) daubechies 20 (e) gaussian or spline (f) biorthogonal (g) mexican hat (h) coiflet fig. 8 graphical representation of different basic wavelets dwt is a non-redundant and outstanding transform technique that aims to remove the artifacts from the time domain eeg signals. in this analysis, the signal is passed through a series of low-pass and high-pass filters to attain approximate and detailed coefficients [44]. the process is iteratively repeated until the desired frequency is reached. when this is achieved, the output is down-sampled by a factor of 2 and up-sampled to get the reconstructed signal. the basic block diagram of dwt is shown in fig. 9. fig. 9 dwt graphical decomposition representation [44] (hpf = high-pass filter; lpf = low-pass filter) dwt’s advantage is that it can provide good temporal resolution for high frequency components and better frequency resolution for lower frequency components [19]. the major drawback of dwt is its time invariance. however, eeg data is not time-invariant and is randomly varying. this problem can be overcome by swt, but the drawback is redundant information and slowness. in matlab, it is implemented by swt function [46]. 4.6. empirical mode decomposition (emd) the practical recorded eeg signals are random stochastic signals, but not static signals. in the standard methods of artifact removal, the eeg signals are projected onto the basis vectors. in wt, the basis vectors are predefined. the experimentally recorded eeg data is not following the same pattern for different patients and, at other conditions, follows the random nature. using predefined basis vectors to remove artifacts is not perfect because some proper eeg signals are also removed. a new method is proposed in 1998 to process nonlinear stochastic random data to overcome this problem [34]. emd decomposes the eeg signals into some oscillatory functions called imfs [47]. the derived imfs must follow two criteria. (1) in the whole dataset, the number of extremes and the number of zero-crossings must be equal or differ by at most one. (2) at any point, the mean value of the envelope is given by the local maxima, and the minima must be zero. 46 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 35-56 it is a data-driven method; unlike wt, the decomposition is not done by using predefined bass functions. here are the steps for performing emd [48]: (1) for the given eeg signal x(t), identify all the local maxima and minima of the given signal. (2) interpolate between the maxima and minima to calculate the upper envelope u(t) and lower envelope l(t), respectively. (3) calculate the average m1 from the lower and upper envelopes and subtract it from the signal x(t) to get the first imf. 1 ( ) ( ) 2 u t l t m + = (14) 1 1 ( )h x t m= − (15) (4) check if h1 satisfies the two criteria for imf. if not, repeat the steps (1)-(3). after i th iteration, the conditions are satisfied. 1( 1) 1 1i i i h m h− − = (16) then, c1 = h1i becomes the first imf. the stopping criteria are explained by huang et al. [49-50], which are given by cauchy convergence test. the first imf is expected to have a high frequency. 1 1 ( )x t c r− = (17) where r1 is the first remainder, which contains remaining frequencies. now, r1 is the input signal for the next sifting process. the second imf is c2. 1 2 2 r c r− = (18) 1n n n r c r − − = (19) this procedure is stopped when the residue is not satisfied with the imf criteria and is a monotonic function. the given eeg signal is represented as: 1 ( ) [ ( ) ( )] i i i i x t c t r t = = +∑ (20) emd decomposes the signals based on amplitude and frequency information. the emg artifacts cannot be separated from the eeg signals due to the overlapping at higher frequencies, but these kinds of emg artifacts can be removed using some standard forms of recent artifact removal techniques. 4.7. singular spectrum analysis (ssa) ssa is a powerful decomposition method for the removal of artifacts in single-channel and multi-channel eeg signals. the bss method is mainly applied to multi-channel data. in ssa, the procedure is primarily divided into two steps, and each step is further divided into two phases [51]. (1) decomposition: (i) time delay embedding; (ii) svd. (2) reconstruction: (i) grouping; (ii) diagonal averaging. 47 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 35-56 consider that there are n sampled signals x(t) = x1, x2, …, xn, then in the first embedding step, the x(t) is mapped as x whose size is k × p [52]. 1 2 2 3 1 1 2 1 , , , p p n l l n x x x x x x x x x x x x x + +         = =           ⋯ ⋯ ⋯ ⋮ ⋮ ⋱ ⋮ ⋯ (21) where l is the embedding size. the general value l = n/2. xi is a lagged vector. the second step is svd. by using svd, the trajectory matrix x is defined as: i i i i x vuλ= (22) where λi is eigenvalue; ui and vi are left and right singular eigenvectors. now the trajectory matrix x is written as: 1 l i i x x = = ∑ (23) as the indices in are divided into r groups. the matrix x after grouping is: 1 j r i j x x = = ∑ (24) in the last step of diagonal averaging, the sub-matrix ��� is translated to new series by the averaging of anti-diagonal elements. the final trajectory matrix is represented as: 1 ( ) ( ) l rc j j x t x t = = ∑ (25) the ssa method can remove muscle artifacts better compared to bss methods. by combing the ssa with other bss methods, artifacts can be removed further. 4.8. independent vector analysis (iva) the eeg data is contaminated by noises/artifacts while recording from the scalp. there are many methods for removing eog artifacts, e.g., ica, cca, emd, ssa, etc. the removal of muscle artifacts is difficult; only a few methods are there because muscle artifacts have high amplitude and variable topographical distribution, and overlap with the entire eeg spectrum [53]. ica separates the muscle artifacts effectively using hos by linearly un-mixing the eeg data into ics if and only if the artifacts are stereotyped. however, muscle artifacts are not stereotyped because these are generated from a group of muscles. cca extracts the sources maximally autocorrelated and mutually uncorrelated using sos. comparing with eeg signals, muscle artifacts have a low correlation. cca uses this property to remove the artifacts. cca and ica are applicable mainly for multi-channel data. kim et al. [54] have proposed an iva approach similar to ica, and have addressed the permutation issue when dealing with acoustic source separation. in recent times, this approach has been applied to eeg signals and could remove muscle artifacts successfully. iva operation which relies on “source extraction from different datasets and maximally dependent sources” within the dataset is maximally independent [53]. in this analysis, the source independence within the dataset and dependence between the datasets exist in parallel. in iva, the source component vector (scv) was defined from different datasets [54]. 48 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 35-56 the p th scv is �� = ��� , �� �,…,�� � (p = 1, 2, …, p), which is a random vector independent of all other scvs. �� � (m =1, 2, …, m) represents the p th source component in the m th dataset. the main intention of iva approach is to identify the independent scv from the multi-dimensional dataset, which is achieved by minimizing the mi among the scv estimates �̂�. 1 2 ˆ ˆ ˆ[ ; ; ...... ] iva p i i s s s≜ (26) where ���̂�� is the mi within the p th scv. initially, iva assumes that the data follow laplacian distribution (iva-l), which uses hos without considering linear dependence using sos. anderson et al. [53] proposed iva-g with the gaussian distribution model, which uses sos and does not consider linear dependence. recently, chen et al. [55] in 2017 implemented a novel iva algorithm with multivariate power exponential distribution (mpe), which uses both sos and hos to get both advantages like linear independence and correlation. the drawback of iva is computational complexity and simulation time. it is mainly applied in laboratory applications, not in online applications. 4.9. hybrid methods artifact removal in eeg signals is vital for analyzing brain conditions and diagnosing the patient if any abnormalities are there. researchers are trying some new techniques by combing more than one existing previously discussed method for better removal of artifacts. by combining one or more methods, the advantages of each technique can be used to develop a new modality for completely removing the artifacts. a hybrid method utilizes the combination of several filters to eliminate the artifacts. it can eliminate the significant artifacts from the eeg signals without adjusting the interpretation of the neural state in the real and simulated eeg data. it improves the previous versions of total variation denoising and emd variants. 4.9.1. discrete wavelet transform adaptive filtering method (dwt-afm) for removing the artifacts in eeg signals using ica, a reference signal is required. this problem can be resolved by using dwt. the reference eog signal can be generated from the given data by selecting low-frequency wavelet coefficients to remove oas [56]. unlike ica, a large amount of information is also not required for better removal of artifacts. however, the disadvantage of dwt is that it cannot remove the artifacts effectively if the spectrum of artifacts overlap with the original eeg signals [44]. the afm is also called adaptive noise canceler (anc). anc can solve this problem. the drawback of anc is that it requires a reference signal for the removal of artifacts. by combing dwt and anc, both spectral overlap and reference signal problems can be solved [33]. for oa removal, dwt is used for generating reference eog signal, and this reference signal is used as an input reference for anc. with this hybrid method, the removal of artifacts is better compared to using dwt and anc independently. 4.9.2. discrete wavelet transform blind source separation (dwt-bss) dwt-bss is one of the hybrid methods to remove the artifacts more effectively. the bss technique is generally applicable to multi-channel data. the disadvantages of both methods are discussed in previous sections. to overcome these disadvantages by combing these methods, the advantages in both methods can be combined. based on single-channel and multi-channel, one of these methods can be used as the first method to apply. first, dwt is applied to decompose the time series signal into wavelet coefficients for the single-channel technique, and these coefficients are applied to bss (ica or cca) [46]. for multi-channel data, dwt or bss can be used first, and the resultant signal is applied to the second method for the removal of artifacts. if ica and cca are applied first, the time domain signal is decomposed into ics and ccs, and by using dwt, these coefficients are further decomposed to better remove the artifacts from the eeg data [19]. 49 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 35-56 4.9.3. empirical mode decomposition blind source separation (emd-bss) bss techniques are used for decomposing multi-channel data into separate ics and ccs using ica and cca, respectively. these techniques are mainly helpful for laboratory experiments, where multi-channel data is available. however, for ambulatory or single-channel data analysis, different methods should be used, e.g., dwt, emd, ssa, etc. the performance can be increased in terms of artifact removal using a hybrid model combing multiple methods. dwt decomposes the data using predefined mother wavelets [57]. however, for nonlinear and non-stationary stochastic data like eeg signals, dwt is not sufficient. emd is a data-driven technique that effectively decomposes randomly varying data like single-channel eeg signals into imfs [10]. then, by using any bss like ica or cca, these imfs can be further decomposed to remove artifacts. for multi-channel data, emd can also apply channel-wise to decompose into imfs [43]. 4.9.4. singular spectrum analysis adaptive noise canceler (ssa-anc) ssa-anc is one of the hybrid methods for the removal of artifacts from single-channel and multi-channel eeg signals. it can be applied to portable devices, where multi-channel data collection is not possible because ica and cca are multi-channel data methods. in the anc method for eog artifact removal, a reference eog artifactual signal is required. in a hybrid method like ssa-anc, the reference signal for removing eog artifacts for anc is given by ssa, like dwt, based on eigenvalues. in dwt, last few low-frequency components are used for generating the eog reference signal [56]. in ssa, the reference signal is also generated from the lower eigenvalues, which are developed while applying svd for the trajectory matrix [58]. the performance is improved in terms of artifact removal compared with ssa and anc because the input reference eog signal is given for anc, which is generated from the output of ssa. 4.9.5. singular spectrum analysis blind source separation (ssa-bss) ica and cca are powerful bss methods that use sos and hos to remove oas and muscle artifacts from the recorded eeg signals [57]. bss methods are multi-channel techniques, and each one has its own advantages and disadvantages. bss methods are combined with ssa as a hybrid method to remove artifacts from the recorded eeg signals. ssa decomposes the single-channel or multi-channel eeg data, and the decomposed data is further processed by bss methods to remove artifacts. 4.9.6. empirical mode decomposition independent vector analysis (emd-iva) iva uses sos and hos for the removal of artifacts [55]. by combing with emd, iva removes the artifacts effectively compared with iva independently. emd method decomposes the single-channel signals into multiple imfs, and these imfs are applied to iva to remove artifacts effectively. to overcome the disadvantages in emd, noise-assisted emd is proposed, which is eemd. eemd is applied to single-channel, but memd is applied to multi-channel data also. by combining memd and iva, zou et al. [12] proposed a method memd-iva for removing the artifacts from multi-channel signals. 5. comparative analysis of the existing methods a tabulated survey of literature and methods are presented in table 4. no independent method can remove all the artifacts. the categorization of procedures is done based on the reference channel requirement. the automatic process can be applied online, whether single-channel or multi-channel, but prior information is required to improve the performance to remove artifacts. some methods are used for physiological artifact removal, and some are for non-physiological artifact removal. regression methods are elementary for the removal of artifacts but require reference channels and multi-channel data. filtering methods remove the artifacts by first estimating the artifactual components and subtracting them from the original eeg contaminated signals, but part of the eeg signals is also removed while subtracting. in kf, the first artifactual region is 50 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 35-56 identified, and by estimating noise, it will be subtracted from only the contaminated area [35]. bss methods are very effective and helpful for the removal of artifacts. most researchers use ica to remove artifacts but are restricted to stereotyped artifacts only [61]. muscle artifacts are not stereotyped and thus cannot be removed by ica effectively. cca can overcome this problem, but it increases computational cost and simulation time, making cca not being applied in online applications. ica and cca are multi-channel data processing methods that cannot be used for ambulatory analysis where few channels or single-channel data is available. table 4 comparative analysis of artifactual methods discussed in the survey method reference electrode required prior knowledge automatic online single-/ multi-channel ref. regression yes no yes no multi [33] af yes yes yes yes both [34] kf no yes yes yes both [35] pca no no no no multi [8] ica no no yes yes multi [2] cca no no yes yes multi [3] dwt no no no no both [44] emd no no no yes both [48] ssa no no yes no both [52] iva no no no no both [54] dwt-afm no no yes yes both [56] dwt-bss no no no no both [46] emd-bss no no no no both [10, 59] ssa-anc no no yes yes both [57] ssa-bss no no no no both [57] emd-iva no no no no both [55, 60] emd and ssa methods can be applied for both single-channel and multi-channel, but the disadvantage is mode mixing and aliasing in emd. ssa method removes only one type of artifacts. an extension of ica from 1-d data is iva technology, which can be used to remove both ocular and muscle artifacts as it takes advantage of both cca and ica methods. automatic methods can remove artifacts without manual intervention. regression methods can be used as intuitive if the reference channel is there [8]. bss methods are used as automatic if a subsequent procedure like svm is applied to verify artifactual components. individually each method has its advantages and disadvantages. researchers are combing more than one method like a hybrid method to get better results to use the benefits of both approaches. in hybrids methods, multi-channel strategies can be applied for single-channels by applying single-channel techniques first. the output is applied as the input for the multi-channel approach. eemd-ica method removes the eog artifacts, and eemd-cca removes emg artifacts from a single-channel, even though ica and cca are multi-channel methods [62]. eemd decomposes the single-channel data into multiple imfs. these imfs can be applied to ica or cca for further processing to remove artifacts. compared to wt-ica, eemd-ica removes artifacts effectively because eemd is a data-driven method. fmemd-cca is used to remove muscle artifacts from a few channel methods with less computational time [25]. wt-ica combined with svm was proposed to remove oas automatically. the recursive least square (rls) notch filter method is proposed to remove ecg artifacts because regression and filtering methods cannot delete the ecg artifacts [63]. the memd-iva method is proposed to remove the muscle artifacts from the acquisition channels, in which the inter-channel interference is also considered one of the noises [60]. some recommendations are suggested to remove muscle artifacts from a few channel data [64]. all the above methods remove the artifacts effectively by hybrid application. apart from these methods, researchers will propose one optimal method to remove all the artifacts in the future. 51 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 35-56 6. conclusions eeg signals are generated from the brain and recorded from the scalp using electrodes. different physiological and non-physiological artifacts contaminate these time-domain eeg signals. in this study, many methods are summarized from various author’s publications, but no specific method can eradicate the artifacts with high accuracy and efficiency. the advantages and disadvantages of each method are discussed in this study. some methods are used primarily to remove artifacts such as eog, emg, and ecg, but no specific method can remove all the artifacts. regression methods and filtering methods require artifactual reference signals for the removal of artifacts. bss methods remove the artifacts effectively, but they require multi-channel extensive data and are computationally complex. dwt, emd, and ssa methods are used for single-channel and multi-channel, but the efficiency of removing artifacts is not up to the level. by combining more than one method, i.e., the hybrid method, the efficiency is improved compared to individual methods. still, cost and complexity is increased, and this may not be suitable for online applications. in this aspect, researchers are finding an optimal solution for minimizing the artifacts in the eeg signals. in the future, with the existing artifact removal techniques, one specific method will be implemented by combing present trends, e.g., machine learning, artificial intelligence (ai), and genetic algorithm (ga), with increased accuracy and efficiency for the removal of artifacts. nomenclature eeg electroencephalogram max maximum absolute error bss blind source separation nrd normalized root mean square difference oa ocular artifacts prd percentage root mean square difference emg electromyography nmse normalized mean square error ecg electrocardiography psnr peak signal to noise ratio mmse modified multiscale sample entropy far false alarm rate iva independent vector analysis mev mean entropy value wt wavelet transform dsnr difference signal to noise ratio swt stationary wavelet transform af adaptive filtering dwt discrete wavelet transform eog electrooculogram st statistical threshold ekf extension kalman filter ut universal threshold ukf unscented kalman filter em eye movement pca principle component analysis ann artificial neural network pc principle component doa depth of anaesthesia svd singular value decomposition bis bispectral index cca canonical correlation analysis arte automated artifact handling in eeg anc adaptive noise canceler svm support vector machine cc correlation coefficient asr artifact subspace reconstruction mi mutual information eemd enhanced empirical mode decomposition sos second-order statistics sqi signal quality index hos higher-order statistics imf intrinsic mode function ai artificial intelligence memd multivariate empirical mode decomposition ga genetic algorithm fmemd fast multivariate empirical mode decomposition snr signal to noise ratio blimf band limited intrinsic mode function mse mean square error ssa singular spectrum analysis mae mean absolute error nnr neural network regressor re relative error mraf multi-resolutional analysis and adaptive filtering rmse root mean square error wpd wavelet packet decomposition rrmse relative root mean square error ica independent component analysis nrmse normalized root mean square error ic independent component psd power spectral density tuh temple university hospital sar signal to artifact ratio 52 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 35-56 conflicts of interest the authors declare no conflict of interest. statement of ethical approval (a) statement of human rights for this type of study, statement of human rights is not required. (b) statement on the welfare of animals for this type of study, statement on the welfare of animals is not required. statement of informed consent for this type of study, informed consent is not required. references [1] h. hu, s. guo, r. liu, and p. wang, “an adaptive singular spectrum analysis method for extracting brain rhythms of electroencephalography,” peerj, vol. 5, e3474, june 2017. [2] a. k. abdullah, c. z. zhang, a. a. a. abdullah, and s. lian, “automatic extraction system for common artifacts in eeg signals based on evolutionary stone’s bss algorithm,” mathematical problems in engineering, vol. 2014, 324750, august 2014. [3] m. m. n. mannan, m. a. kamran, and m. y. jeong, “identification and removal of physiological artifacts from electroencephalogram signals: a review,” ieee access, vol. 6, pp. 30630-30652, 2018. [4] r. zafar, a. qayyum, and w. mumtaz, “automatic eye blink artifact removal for eeg based on a sparse coding technique for assessing major mental disorders,” journal of integrative neuroscience, vol. 18, no. 3, pp. 217-229, 2019. [5] a. morley, l. hill, and a. g. kaditis, “10-20 system eeg placement,” https://www.ers-education.org/lrmedia/2016/pdf/298830.pdf, march 15, 2016. [6] s. k. goh, h. a. abbass, k. c. tan, a. al-mamun, c. wang, and c. guan, “automatic eeg artifact removal techniques by detecting influential independent components,” ieee transactions on emerging topics in computational intelligence, vol. 1, no. 4, pp. 270-279, august 2017. [7] s. r. sreeja, r. r. sahay, d. samanta, and p. mitra, “removal of eye blink artifacts from eeg signals using sparsity,” ieee journal of biomedical and health informatics, vol. 22, no. 5, pp. 1362-1372, september 2018. [8] x. jiang, g. b. bian, and z. tian, “removal of artifacts from eeg signals: a review,” sensors, vol. 19, no. 5, 987, march 2019. [9] s. s. menon and k. krishnamurthy, “a study of brain neuronal and functional complexities estimated using multiscale entropy in healthy young adults,” entropy, vol. 21, no. 10, 995, october 2019. [10] k. t. sweeney, s. f. mcloone, and t. e. ward, “the use of ensemble empirical mode decomposition with canonical correlation analysis as a novel artifact removal technique,” ieee transactions on biomedical engineering, vol. 60, no. 1, pp. 97-105, january 2013. [11] m. chavez, f. grosselin, a. bussalb, f. d. v. fallani, and x. navarro-sune, “surrogate-based artifact removal from single-channel eeg,” ieee transactions on neural systems and rehabilitation engineering, vol. 26, no. 3, pp. 540-550, march 2018. [12] y. zou, v. nathan, and r. jafari, “automatic identification of artifact-related independent components for artifact removal in eeg recordings,” ieee journal of biomedical and health informatics, vol. 20. no. 1, pp. 73-81, january 2016. [13] r. mahajan and b. i. morshed, “unsupervised eye blink artifact denoising of eeg data with modified multiscale sample entropy, kurtosis, and wavelet-ica,” ieee journal of biomedical and health informatics, vol. 19, no. 1, pp. 158-165, january 2015. 53 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 35-56 [14] p. p. acharjee, r. phlypo, l. wu, v. d. calhoun, and t. adalı, “independent vector analysis for gradient artifact removal in concurrent eeg-fmri data,” ieee transactions on biomedical engineering, vol. 62, no. 7, pp. 1750-1758, july 2015. [15] s. khatun, r. mahajan, and b. i. morshed, “comparative study of wavelet-based unsupervised ocular artifact removal techniques for single-channel eeg data,” ieee journal of translational engineering in health and medicine, vol. 4, pp. 1-8, march 2016. [16] m. n. tibdewal, r. r. fate, m. mahadevappa, a. k. ray, and m. malokar, “classification of artifactual eeg signal and detection of multiple eye movement artifact zones using novel time-amplitude algorithm,” signal, image, and video processing, vol. 11, no. 2, pp. 333-340, february 2017. [17] q. liu, y. f. chen, s. z. fan, m. f. abbod, and j. s. shieh, “eeg artifacts reduction by multivariate empirical mode decomposition and multiscale entropy for monitoring depth of anaesthesia during surgery,” medical and biological engineering and computing, vol. 55, no. 8, pp. 1435-1450, august 2017. [18] s. barua, m. u. ahmed, c. ahlstrom, s. begum, and p. funk, “automated eeg artifact handling with application in driver monitoring,” ieee journal of biomedical and health informatics, vol. 22, no. 5, pp. 1350-1361, september 2018. [19] c. y. sai, n. mokhtar, h. arof, p. cumming, and m. iwahashi, “automated classification and removal of eeg artifacts with svm and wavelet-ica,” ieee journal of biomedical and health informatics, vol. 22, no. 3, pp. 664-670, may 2018. [20] j. li, y. chen, f. taya, j. lim, k. wong, y. sun, et al., “a unified canonical correlation analysis-based framework for removing gradient artifact in concurrent eeg/fmri recording and motion artifact in walking recording from eeg signal,” medical and biological engineering and computing, vol. 55, no. 9, pp. 1669-1681, september 2017. [21] v. roy and s. shukla, “effective eeg motion artifacts elimination based on comparative interpolation analysis,” wireless personal communications, vol. 97, no. 4, pp. 6441-6451, december 2017. [22] c. y. chang, s. h. hsu, l. pion-tonachini, and t. p. jung, “evaluation of artifact subspace reconstruction for automatic artifact components removal in multi-channel eeg recordings,” ieee transactions on biomedical engineering, vol. 67, no. 4, pp. 1114-1121, april 2020. [23] j. cheng, l. li, c. li, y. liu, a. liu, r. qian, et al., “remove diverse artifacts simultaneously from a single-channel eeg based on ssa and ica: a semi-simulated study,” ieee access, vol. 7, pp. 60276-60289, 2019. [24] m. saini and u. satija, “an effective and robust framework for ocular artifact removal from single-channel eeg signal based on variational mode decomposition,” ieee sensors journal, vol. 20, no. 1, pp. 369-376, january 2020. [25] y. liu, y. zhou, x. lang, y. liu, q. zheng, y. zhang, et al., “an efficient and robust muscle artifact removal method for few-channel eeg,” ieee access, vol. 7, pp. 176036-176050, 2019. [26] c. dora and p. k. biswal, “correlation-based ecg artifact correction from single channel eeg using modified variational mode decomposition,” computer methods and programs in biomedicine, vol. 183, 105092, january 2020. [27] c. dora, r. n. patro, s. k. rout, p. k. biswal, and b. biswal, “adaptive ssa based muscle artifact removal from single channel eeg using neural network regressor,” innovation and research in biomedical engineering, in press. [28] d. p. yedurkar and s. p. metkar, “multiresolution approach for artifacts removal and localization of seizure onset zone in epileptic eeg signal,” biomedical signal processing and control, vol. 57, 101794, march 2020. [29] m. s. islam, a. m. el-hajj, h. alawieh, z. dawy, n. abbas, and j. el-imad, “eeg mobility artifact removal for ambulatory epileptic seizure prediction applications,” biomedical signal processing and control, vol. 55, 101638, january 2020. [30] n. bajaj, j. r. carrión, f. bellotti, r. berta, and a. de gloria, “automatic and tunable algorithm for eeg artifact removal using wavelet decomposition with applications in predictive modeling during auditory tasks,” biomedical signal processing and control, vol. 55, 101624, january 2020. [31] s. k. noorbasha and g. f. sudha, “removal of eog artifacts and separation of different cerebral activity components from single channel eeg—an efficient approach combining ssa-ica with wavelet thresholding for bci applications,” biomedical signal processing and control, vol. 63, 102168, january 2021. [32] m. a. klados, c. papadelis, c. d. lithari, and p. d. bamidis, “the removal of ocular artifacts from eeg signals: a comparison of performances for different methods,” 4th european conference of the international federation for medical and biological engineering, pp. 1259-1263, november 2008. [33] y. chen, q. zhao, b. hu, j. li, h. jiang, w. lin, et al., “a method of removing ocular artifacts from eeg using discrete wavelet transform and kalman filtering,” ieee international conference on bioinformatics and biomedicine, pp. 15-18, december 2016. [34] a. zhang and w. li, “adaptive noise cancellation for removing cardiac and respiratory artifacts from eeg recordings,” 5th world congress on intelligent control and automation, pp. 15-19, june 2004. 54 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 35-56 [35] h. shahabi, s. moghimi, and h. zamiri-jafarian, “eeg eye blink artifact removal by eog modeling and kalman filter,” 5th international conference on biomedical engineering and informatics, pp. 356-367, october 2012. [36] a. mur, r. dormido, and n. duro, “an unsupervised method for artefact removal in eeg signals,” sensors, vol. 19, no. 10, 2302, may 2019. [37] b. mijović, m. de vos, i. gligorijević, j. taelman, and s. van huffel, “source separation from single-channel recordings by combining empirical-mode decomposition and independent component analysis,” ieee transactions on biomedical engineering, vol. 57, no. 9, pp. 2188-2196, september 2010. [38] v. roy and s. shukla, “a survey on artifacts detection techniques for electro-encephalography (eeg) signals,” international journal of multimedia and ubiquitous engineering, vol. 10, no. 3, pp. 425-442, 2015. [39] g. wang, c. teng, k. li, z. zhang, and x. yan, “the removal of eog artifacts from eeg signals using independent component analysis and multivariate empirical mode decomposition,” ieee journal of biomedical and health informatics, vol. 20, no. 5, pp. 1301-1308, september 2016. [40] x. chen, a. liu, h. peng, and r. k. ward, “a preliminary study of muscular artifact cancellation in single-channel eeg,” sensors, vol. 14, no. 10, pp. 18370-18389, october 2014. [41] d. r. hardoon, s. szedmak, and j. shawe-taylor, “canonical correlation analysis: an overview with application to learning methods,” neural computation, vol. 16, no. 12, pp. 2639-2664, december 2004. [42] w. de clercq, a. vergult, b. vanrumste, w. van paesschen, and s. van huffel, “canonical correlation analysis applied to remove muscle artifacts from the electroencephalogram,” ieee transactions on biomedical engineering, vol. 53, no. 12, pp. 2583-2587, 2006. [43] x. chen, x. xu, a. liu, m. j. mckeown, and z. j. wang, “the use of multivariate emd and cca for denoising muscle artifacts from few-channel eeg recordings,” ieee transactions on instrumentation and measurement, vol. 67, no. 2, pp. 359-370, february 2018. [44] s. khatun, r. mahajan, and b. i. morshed, “comparative study of wavelet-based unsupervised ocular artifact removal techniques for single channel eeg data,” ieee journal of translational engineering in health and medicine, vol. 4, 2000108, 2016. [45] p. gajbhiye, r. k. tripathy, a. bhattacharyya, and r. b. pachori, “novel approaches for the removal of motion artifact from eeg recordings,” ieee sensors journal, vol. 19, no. 22, pp. 10600-10608, november 2019. [46] b. s. raghavendra and d. n. dutt, “wavelet enhanced cca for minimization of ocular and muscle artifacts in eeg,” world academy of science, engineering, and technology, vol. 5, no. 9, pp. 419-424, january 2011. [47] e. karatoprak and s. seker, “an improved empirical mode decomposition method using variable window median filter for early fault detection in electric motors,” mathematical problems in engineering, vol. 2019, 8015295, 2019. [48] b. yang, t. zhang, y. zhang, w. liu, j. wang, and k. duan, “removal of electrooculogram artifacts from electroencephalogram using canonical correlation analysis with ensemble empirical mode decomposition,” cognitive computation, vol. 9, no. 5, pp. 626-633, october 2017. [49] n. e. huang and z. wu, “a review on hilbert-huang transform: method and its applications to geophysical studies,” reviews of geophysics, vol. 46, no. 2, rg2006, june 2008. [50] n. e. huang, m. l. c. wu, s. r. long, s. s. shen, w. qu, p. gloersen, et al., “a confidence limit for the empirical mode decomposition and hilbert spectral analysis,” proceedings of the royal society a: mathematical, physical, and engineering sciences, vol. 459, no. 2037. pp. 2317-2345, september 2003. [51] s. xu, h. hu, l. ji, and p. wang, “embedding dimension selection for adaptive singular spectrum analysis of eeg signal,” sensors, vol. 18, no. 3, 697, march 2018. [52] q. liu, a. liu, x. zhang, x. chen, r. qian, and x. chen, “removal of emg artifacts from multichannel eeg signals using combined singular spectrum analysis and canonical correlation analysis,” journal of healthcare engineering, vol. 2019, 4159676, 2019. [53] m. anderson, t. adali, and x. l. li, “joint blind source separation with multivariate gaussian model: algorithms and performance analysis,” ieee transaction on signal process, vol. 60, no. 4, pp. 1672-1683, april 2012. [54] t. kim, t. eltoft, and t. w. lee, “independent vector analysis: an extension of ica to multivariate components,” international conference on independent component analysis and signal separation, march 2006, pp. 165-172. [55] x. chen, a. liu, q. chen, y. liu, l. zuo, and m. j. mckeown, “simultaneous ocular and muscle artifact removal from eeg data by exploiting diverse statistics,” computers in biology and medicine, vol. 88, pp. 1-10, september 2017. [56] h. peng, b. hu, q. shi, m. ratcliffe, q. zhao, y. qi, et al., “removal of ocular artifacts in eeg—an improved approach combining dwt and anc for portable applications,” ieee journal of biomedical and health informatics, vol. 17, no. 3, pp. 600-607, may 2013. 55 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 35-56 [57] l. zou, x. chen, g. dang, y. guo, and z. j. wang, “removing muscle artifacts from eeg data via underdetermined joint blind source separation: a simulation study,” ieee transactions on circuits and systems ii: express briefs, vol. 67, no. 1, pp. 187-191, january 2020. [58] a. k. maddirala and r. a. shaik, “removal of eog artifacts from single channel eeg signals using combined singular spectrum analysis and adaptive noise canceler,” ieee sensors journal, vol. 16, no. 23, pp. 8279-8287, december 2019. [59] x. chen, a. liu, j. chiang, z. j. wang, m. j. mckeown, and r. k. ward, “removing muscle artifacts from eeg data: multi-channel or single-channel techniques,” ieee sensors journal, vol. 16, no. 7, pp. 1986-1997, april 2016. [60] x. xu, x. chen, and y. zhang, “removal of muscle artefacts from few-channel eeg recordings based on multivariate empirical mode decomposition and independent vector analysis,” electronics letters, vol. 54, no. 14, pp. 866-868, july 2018. [61] x. chen, q. liu, w. tao, l. li, s. lee, a. liu, et al., “remae: user-friendly toolbox for removing muscle artifacts from eeg,” ieee transactions on instrumentation and measurement, vol. 69, no. 5, pp. 2105-2119, may 2020. [62] x. chen, q. chen, y. zhang, and z. j. wang, “a novel eemd-cca approach to removing muscle artifacts for pervasive eeg,” ieee sensors journal, vol. 19, no. 19. pp. 8420-8431, october 2019. [63] c. dai, j. wang, j. xie, w. li, y. gong, and y. li, “removal of ecg artifacts from eeg using an effective recursive least square notch filter,” ieee access, vol. 7, pp. 158872-158880, october 2019. [64] x. chen, x. xu, a. liu, s. lee, x. chen, x. zhang, et al., “removal of muscle artifacts from the eeg: a review and recommendations,” ieee sensors journal, vol. 19, no. 14, pp. 5353-5368, july 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/). 56  proceedings of engineering and technology innovation , vol. 2, 2016, pp. 06 07 6 the sensor collaboration structure of industrial 4.0 hao-ting lin * department of mechanical and computer-aided engineering, feng chia university, taichung, taiwan. received 02 february 2016; received in revised form 15 march 2016; accept ed 05 april 2016 abstract as informat ion and technology grow up globally, the business should look for the opt ima l productivity effic iency and add smart manage ment to ma intain the co mpany’s inte rests and value. in order to achieve the smart productivity, vast businesses pay lots of attention to invest in industrial 4.0 and change their business model. in this research, the sensor collaboration structure will be proposed and established to connect the information base and thing base. human can ma ke decisions and communicate between the physical domain and the cyber domain. in the physical doma in, the sensors can gather the data from the machine, the production line, the factory and the product. by sensor collaboration, the signals and info rmat ion can transfer quickly and simu ltaneously in the cyber domain. through the networking between information base and thing base, dec ision make rs can easily predict, judge and handle the tasks. as a result, in formation base, thing base and decision makers can beco me a triangle system to achieve the vertical integration and level contact. ke ywor ds : industrial 4.0, sensor collaboration, physical domain, cyber domain 1. introduction nowadays, the pursuing goal for the manufacturing industry is to establish the smart fac tories. the first industrial revolution is to exe cute machines instead of human to conduct large-scale p roduction. the second industrial revolution is to add the concept of assembly line for improving production effic iency. in the third industrial revolution, with ele ments of electronics and informat ion technology, the production costs are declined. with the development of globalization, the factory must be supplemented with intelligent manage ment to facilitate more rapid changes in the industry pulsation. co mbining with the advantages of three previous industrial revolution, the industry 4.0 is coming to search for the highest effic iency of production resources and to achieve the purpose of the intelligent production. in this paper, the sensor collaboration structure will be proposed for the smart factory. the concept of industrial 4.0 has been popular in eu rope, especially in ge rman manufacturer industries. smart mach inery can contin uously share inventories, problems info rmation and demands through the internet, so the production process and lead time can be fle xib ly adjusted. cyber-physical production systems (cppss) can not only connect machines with each other, but also create the whole values of the stages in the product life cycle. sensors and control ele ments make the machines, factories, transportation, internet and humans together [1]. fig. 1 shows the history of the industry revolution. fig. 1 history of the industry revolution 2. the architecture of industrial 4.0 the architecture of industrial 4.0 which has the vertical integration of the s mart production system, the horizontal integration of the ne tworks, through engineering interaction, and exponential technologies. 2.1. the vertical integration of the smart production system cppss ma ke the factories vary their inventories and demands fle xibly. smart factories *corresponding aut hor. email: haotlin@fcu.edu.t w proceedings of engineering and technology innovation , vol. 2, 2016, pp. 06 07 7 copyright © taeti establish customized and personal production through data integration and smart sensing techniques. cppss a lso pro mote the automation manage ment, ma intenance manage ment and resource connection. 2.2. the horizontal integration of the network s the horizontal integration which is an o ptimized real-time networks can create the higher value and mo re fle xib le for proble m solving. the internet based production system can provide the integrated platform for logistics, inventories, production, marketing and sales b etween in-house and outsourcing. 2.3. through engineering interaction industrial 4.0 has a cross -disciplinary through engineering to connect overall supply chain from products to customers. data and information can be obtained in each stage of the product life cycle to establish the model fro m the prototype to the end product. 2.4. exponential technologies artificia l intelligence (ai), advanced robot and sensing technology can make the production personalized and fle xible. ai can not only an alyze big data, design the more fle xible path for the automated guided vehicle (a gv) and save costs of the supply chain management (scm), but also enhance the cooperation between h uman and machinery. in addition, the 3d p rinting provide the new solution for the production mode to reduce inventories and lead time for the firm. 3. conclusions in this paper, the proposed architecture of industrial 4.0 has the vertical integration, the horizontal integration, through engineering and e xponential technologies. the sensor collaboration for industrial 4.0 can collect and interact with hu man and machinery. managers can co rrectly ma ke a decision by the cyber-physical system. acknowledgement this research was sponsored by ministry of science and technology, taiwan under the grant most 104-2218-e-035 -017. references [1] x. yue, h. cai, h. yan, c. zou, and k. zhou, “cloud-assisted industrial cyber-physical systems: an insight,” microprocessors and microsystems , vol. 39, no. 8, pp. 1262-1270, 2015.  proceedings of engineering and technology innovation, vol. 2, 2016, pp. 01 03 1 optical design of direct projection bicycle light hsuan-jung tsai * , tsang-yen hsieh, jyh-liang wang department of electronic engineering, ming chi university of technology, new taipei city, taiwan. received 01 february 2016; received in revised form 15 march 2016; accepted 03 april 2016 abstract optical lighting systems usually demand specific illuminance distributions to comply with regulatory requirements for various applications. this study demonstrates an optical design of bicycle light integrated with a special fresnel lens and a reflector. to enhance the light intensity, the proposed bicycle light adopts direct projection rather than reflection. the fresnel lens is used to concentrate light and achieve the specific distribution of illuminance. the reflector provides a parabola reflective surface and reduces the divergence of light to greatly cast out parallel light, which improves the illuminance. for optimum case, the simulation is based on the total led power of 180 lm, and the optical power is analyzed as 85 lm with a distance of 10 m away from the bicycle light. the illuminance distribution of bicycle light and cut-off line can meet the requirements of german stvzo regulations. keywords: fresnel lens, bicycle light, reflector, direct projection, stvzo 1. introduction bicycle light provide people with safe riding at night. it requests enough brightness on the road and should avoid the light radiation to the people’s eye. therefore, the optical design with cut-off line is important and indeed necessary. german road traffic licensing regulations (germany stvzo regulations) clearly define the use of bicycle safe and bicycle lighting requirements as following [1] (1) at least 20 lux in hv, the maximum illuminance emax shall be found at this point. if emax is found lateral to hv, its value shall not exceed 1.2 times the value of hv. (2) within the range delimited by an angle of 4° to both sides starting from hv up to and including point l1 respectively r1 as well as 1.5° below hv up to and including point 2: at least half of the maximum illuminance emax. (3) up to 5° below hv and between point 2 and point 3 including the latter: at least 2.5 lux and from there 4° to the left and to the right up to and including the points l5 respectively r5: at least 2.0 lux. (4) in the horizontal plane 3.4° above hv and beyond (zone 1): not more than 2.0 lux. optical design of bicycle light can usually classified as reflective mode and direct mode. reflective mode requires a reflective apparatus. direct mode uses the convex lens and forms optical imaging or non-imaging depending on the application. in this study, a direct projection bicycle light integrated with a reflector is considered and investigated due to the better light efficiency. fig. 1 the illustration of germany stvzo regulations [1] 2. method the requirements and constraints of optical design are based on the germany stvzo regulations. the regulations require power consumption of 6.2w at 12v. the luminous flux of light source embedded in headlights should exceed a value of 42 lm at test voltage. the required illuminance must correspond to the guideline shown in fig. 1. *corresponding author. email: pot.martian@gmail.com proceedings of engineering and technology innovation, vol. 2, 2016, pp. 01 03 2 copyright © taeti fig. 2 the fresnel lens designed with inward grooves facing [3] a typical imaging fresnel lens with inward grooves facing has an analytical solution indicated by tver’yanovich [2]. the structure of fresnel lens will be performed and based on the creative design methodology process. in accordance with fig. 2, three equations can be set up to describe the lens. the prism angle α is the goal of a simulation written as [2-3], where n is a material medium, ω is divergence angle of light, α is light emission angle, β is the sum of α and ω, r is lens radius, and f is lens focal length. 𝑛 𝑠𝑖𝑛 𝛼 = 𝑠𝑖𝑛 𝛽 (1) 𝑡𝑎𝑛 𝜔 = 𝑅 𝑓 (2) 𝛽 = 𝛼 + 𝜔 (3) 3. simulation and result the optical characteristics of bicycle light are designed to fit german stvzo regulations. 3d modeling and components of the proposed bicycle light, which is equipped a commercial led lamp (cree xlamp xp-g2) as the light source. a specific fresnel lens is adopted in the optical design of bicycle light. the fresnel lens combines two focal points together to concentrate light and presents a rectangle distribution of illuminance, shown in fig. 3. fig. 3 (a) 3d cad of fresnel lens, and (b) the simulated illuminance distribution of the led lamp with the specific fresnel lens the simulated illuminance distribution of the led lamp with the specific fresnel lens only provide central illuminance with rectangle shape, and does not extend the illuminance distribution to reach the requirements of l1 and r1. thus, the optical design of bicycle light combined with fresnel lens cannot fit german stvzo regulations. moreover, a reflector is proposed to improve the optical design. the reflector provides a parabola reflective surface and reduces the divergence of light to greatly cast out parallel light, revealed in fig. 4. it evidently indicates that the simulated illuminance distribution can be coherent with the real illuminance of projection. fig. 4 (a) simulated illuminance distribution of the led lamp with a reflector, (b) the real illuminance of projection with a distance of 1 meter, and (c) the reflector entity fig. 5 schematic mechanisms of direct projection bicycle light fig. 5 gives the final design of direct projection bicycle light. the design combines the specific fresnel lens and reflector. the combination demands some modulations of the physical dimensions to achieve the optimum simulation. by the trial of optical simulation to fine tune the design of bicycle light, an optimum illuminance proceedings of engineering and technology innovation, vol. 2, 2016, pp. 01 03 3 copyright © taeti distribution can be made. fig. 6 demonstrates the simulation result of the optimum design. the simulated illuminance distribution is similar as an extended rectangle and in compliance with the requirements of germany stvzo regulations. fig. 6 simulated illuminance distribution of the proposed direct projection bicycle light 4. conclusions a direct projection bicycle light is proposed in this research. the optical design of bicycle light is integrated with a special fresnel lens and a reflector. the fresnel lens combines two focal points to concentrate light and achieve the specific distribution of illuminance. the reflector provides a parabola reflective surface and reduces the divergence of light. the simulated illuminance distribution of bicycle light and cut-off line can meet the requirements of german stvzo regulations. references [1] road traffic licensing regulations, germany: government printing office, 2012. [2] e. v. tver’yanovich, “profiles of solar engineering fresnel lenses,” gelioteknika, vol. 19, no. 6, pp. 31-34, 1984. translated into english in applied solar energy, vol. 19, no. 6, pp. 36-39, 1984. [3] r. leutz and n. suzuki, nonimaging fresnel lenses, paris: springer, may 2001. cut-off line liline microsoft word 7-#990-typesetting finish 列印pdf.doc proceedings of engineering and technology innovation, vol. 7, 2017, pp. 37 40 traffic safety evaluation based on vision and signal timing data won-ho suh*, jin-woo park, jung-ryul kim department of transportation and logistics engineering, hanyang university, ansan, korea. received 18 july 2017; received in revised form 11 august 2017; accepted 30 august 2017 abstract with the advancement of recent image processing technologies numerous imaging applications have been developed to collect traffic information. in this study, image processing technology is applied to roadway safety to detect and measure the conflict between pedestrians and vehicular traffic and investigate the correlations between the conflict and signal timing data. by using an image processing technique, pedestrian and vehicular movements are detected and the distance between these two are calculated to detect conflicts between them. in this study, machine learning techniques are utilized to facilitate more accurate and efficient detection. it is assumed that the intersection is accident-prone if conflicts between pedestrian and vehicular traffic appear more frequently than other intersections and correlations exist depending on the status of signal timing, for example at the end of green signal for a particular movement. this approach is expected to identify accident-prone intersections without actually experiencing crashes; therefore, this would potentially reduce social costs associated with traffic accidents. keywords: image processing, vehicle detecting, traffic safety, pedestrian safety 1. introduction pedestrians are vulnerable road users. when they are involved in collisions with vehicles, there are considerably higher chances of being severely or fatally injured, while a large number of collisions between vehicles are property damage only. surrogate safety measures has been developed as a complementary method to improve pedestrian safety and offer more in depth analysis than relying on historic accident data alone. there is a growing effort to use recent image processing technologies for a conflict analysis between vehicles and pedestrians. in this study, vehicle tracking and pedestrian tracking technologies currently developed and used are reviewed. 2. vehicle tracking model-based tracking utilizes a priori knowledge of typical objects. for a vehicle tracking, vehicles in an image captured from a video are tracked. these methods recognize vehicles. however, this method tends to fail when there is an occlusion for the target object. also, this tracking method is limited in recognizing all the vehicle types on the roadway. region-based tracking method is to identify connected regions of the image associated with each target vehicle. regions are usually acquired from background subtraction. also, this method utilizes kalman filters to process available information from the image including motion, size, color, shape, texture, and centroid. it is known that this tracking method is * corresponding author. e-mail address: wonhosuh@hanyang.ac.kr tel.: +82-31-400-5154; fax: +82-31-436-8147 proceedings of engineering and technology innovation, vol. 7, 2017, pp. 37 40 copyright © taeti 38 computationally efficient and the detection accuracy is generally good in free-flowing traffic conditions. however, if traffic is congested – when background images do not change over a short period of time and vehicles partially occlude one another, this method tends to have problems, since separating individual vehicles are becoming challenging. sometimes, such vehicles are identified as one vehicle, resulting in low detection rates. contour-based tracking method uses the boundary curves of the moving object. for example, this method tracks pedestrians by identifying the contour of a pedestrian’s head. this method is known to have more efficient description of objects than region-based methods. however, this method still shares the weakness of region-based tracking when traffic condition is congested. feature-based tracking does not track object as a whole in image processing. instead, this method tracks unique features. there features can be points or lines of the object which are distinct from other objects. when the target object is partially occluded, some of the distinctive features of the moving object remain visible and detectable, resulting in successful tracking. the advantage of this method over other tracking methods is that the same algorithm can be applied for tracking in different lighting or traffic conditions. for example, this method utilizes window corners or bumper edges during the day when the entire object is visible. during the night, this method uses tail lights, since other features are not detectable in night time lighting conditions. currently, this method is advanced and computational efficient allowing this method to be used in real time processing. an example of vehicle tracking is provided in fig. 1. fig. 1 vehicle tracking [1] 3. pedestrian tracking to identify conflicts between vehicles and pedestrians, all movements of vehicular and pedestrian traffic need to be tracked from video frames. especially, this is challenging since pedestrians sometimes move in group and change directions, while vehicular movements are somewhat predictable. to overcome the issues and increase the accuracy of tracking, there are three tracking methods which are currently popular; tracking by detection, tracking using flow, and tracking with probability. tracking by detection method detects objects using background modeling and subtraction with the current image or deformable templates. tracking using flow is similar to feature based tracking. this method is popular in traffic monitoring and pedestrian counting. tracking with probability uses a bayesian tracking framework. this method also utilizes other filers and methods, such as particle filters, markov chain, and monte carlo methods. an example of pedestrian tracking is provided in fig. 2. also, fig. 3 presents an example of vehicle and pedestrian tracking in an intersection. proceedings of engineering and technology innovation, vol. 7, 2017, pp. 37 40 copyright © taeti 39 fig. 2 pedestrian tracking [1] fig. 3 vehicle and pedestrian tracking [2] 4. calculating conflicts to calculate conflicts between vehicular and pedestrian traffic, potential accident-prone conditions need to be identified. these conditions are illustrated in fig. 4 and fig. 5. based on the calculated time gap between the vehicle and pedestrian, conflict data can be obtained. fig. 4 left turning vehicle conflict example [3] fig. 5 thru moving vehicle conflict example [3] 5. preliminary results and future works the research team is in a process of developing vehicle and pedestrian tracking method. the team was successfully tracked vehicles in a video collected from an intersection in seoul, korea (fig. 6). the team is planning to continue to develop the tracking algorithm and calculate the conflicts between vehicular and pedestrian traffic. the team is utilizing machine learning techniques to reduce the detection error and increase the efficiency of the program. the output of this research is proceedings of engineering and technology innovation, vol. 7, 2017, pp. 37 40 copyright © taeti 40 expected to improve traffic analysis [4] by detecting pedestrian and vehicular movements and facilitate more accurate and efficient detection. it is assumed that the intersection is accident-prone if conflicts between pedestrian and vehicular traffic appear more frequently than other intersections and correlations exist depending on the status of signal timing, for example at the end of green signal for a particular movement. this approach is expected to identify accident-prone intersections without actually experiencing crashes; therefore, this would potentially reduce social costs associated with traffic accidents. fig. 6 vehicle tracking program snapshot acknowledgment this work was sponsored by nrf-2017r1d1a1a09000606 and transportation & logistics research program id-97344 of the korean government. references [1] k. ismail, t. sayed, and n. saunier, “automated analysis of pedestrian-vehicle: conflicts context for before-and-after studies,” transportation research record: journal of the transportation research board, vol. 2198, pp. 52-64, 2010. [2] a. laureshyn, “superpixel based road user tracker,” workshop on the comparison of surrogate measures of safety extracted from video data, transportation research board 93rd annual meeting, january 2014. [3] t. madsen, “automatic detection of conflicts at signalized intersections,” workshop on the comparison of surrogate measures of safety extracted from video data, transportation research board 93rd annual meeting, january 2014. [4] w. suh, “mitigating initialization bias in transportation modeling applications,” proceedings of engineering and technology innovation, vol. 3 pp. 34-36, 2016.  proceedings of engineering and technology innovation , vol. 2, 2016, pp. 11 13 11 fire resistant study for long span beams made of fr 490 steels in-kyu kwon * department of fire protection engineering, kangwon national university, samcheok, korea. received 22 february 2016; received in revised form 02 april 2016; accept ed 10 may 2016 abstract building could be severely influenced by fire and it can be demo lished so that the ma jor structural ele ments such as columns and beams required to be coated by fire protection materials according to each condition. especially, a steel fra med building regarded as a weaken structures when the build ing is engulfed with a severe fire condition. therefore, the fire resistant steel (fr 490) has been developed to improve the stru ctural properties of an ordinary structural steels at high temperature. generally, the fire resistance of structural beam built with a structural steel can be measured by horizontal furnace using a standard testing method by each nation’s standard. ho wever, the real length of the bea m showed the tendency longer than that of the horizontal furnace. therefore, it is considered more important to know the rea l fire resistant performance of structural bea ms. in this paper, to evaluate the fire resistant performance of long span beams made of the fr 490, the fire engineering method is done us ing not only materials properties at high temperatures but a heat transfer, a stress theory. the results showed that as the length of beams are longer, the ma ximu m load bearing capacities at high temperatures are lower and the deflections are increased. ke ywor ds : fr 490, long span beam, fire re sistant performance, structural stability at high temperature 1. introduction when steel fra med buildings are covered with a severe fire, the structural stabilit ies can be damaged and this can cause a progressive co llapse. therefore, a ll most of countries require the prima ry structural elements such as columns, beams, trusses the fire protection according to building situation. in ko rea, the fire protection depends on the buildings height and use. the way to satisfy the fire protection is divided into two parts. one is to obey the building regulation and standard, the other is to design with an e ngineering technique. the last one is known to an advanced engineering des ign or a performance based fire design. the na me can be used differently by nation or by person who uses it. the basic concept of the advanced engineering d esign consists of a fire scenario and a structural analysis using mechanical properties and heat transfer theory. until now, the advanced fire engineering method is known to practice to most of countries but not my country. at now, my country is trying to accept the perfo rmance-based engineering method in building industry and this paper is one of proces s to do it. in general, the structural steel is known as a weaken materia l when it e xposed to a fire. be cause the atomic structure is inclined to recede away as the temperature of it is getting high. therefore, to protect the normal strength of the structural steel at high temperature situation, a proved thickness of passive fire protective ma terials should be applied. a fr 490 had developed to lessen the tendency to reduce mechanical properties such as yield strength and an elastic modulus at high temperature rather than those of sm 490. the fr 490 has the same mechanica l property as the sm 490 has. the purpose of this study is to evaluate the structural stability of long span beams made of the fr 490 at high temperature using mechanica l propert ies and heat transfer and heat stress analysis. 2. method the structural stability of long span beam can be measured by vertical furnace in terms of deflection with loading. but the vert ical furnace is fixed with a ma ximu m d imension, 4 meters in length and 3 meters in width. but the size of the *corresponding aut hor. email: kwonik@kangwon.ac.kr proceedings of engineering and technology innovation , vol. 2, 2016, pp. 11 13 12 copyright © taeti vertical furnace is dependent on each fire facility. therefore, when the longer bea m is required the fire resistance, the furnace is no longer fun ctioned. fro m middle 1900’s, a ca lculation method started fro m european nations based on an engineering development and as a reasonable alternative to furnace test. until now, korea didn’t utilize the fire engineering method and has used a prescriptive method fro m early 1970’s. therefore , even though the fr 490 showed a better performance than that of sm 490, the steel construction fie ld not allow the fr 490 as a structural steel. to utilize the better structural steel than sm 490 into the construction market, an analytica l evaluation is conducted using mechanical pro perties with long span beams. applied mechan ical properties are shown in table 1. table 1 mechanical properties of fr 490 properties temperature (℃) regression equation properties t≤500 cold value (315 mpa) 500<t≤900 -0.72t + 674.49 elastic modulus t≤200 cold value (210 gpa) 200<t≤900 -211.73t + 254,343 long span beams consist of 4 types. the basic length of the beam is 4100 mm. this length is standard size in vertical furnace in korea. the ma ximu m span is reaching to 5000 mm. the ana lytical para meters are shown in table 2. table 2 analysis parameters section h-400x200x8x13 section area(84.12 cm 2 ) support conditions simple beam span of beam (mm) 4100, 4400, 4700, 5000 fire curve standard fire curve, ks f 2257-1,6 3. results and discussion based on an one dimension heat transfer theory the surface temperatures of the beam are calculated like fig. 1. as the beam is not protected any fire protective materia l, the surface temperature is going up rap idly and a fter 20 minutes the temperature is simila r to standard fire curve. fig. 1 steel temperature history versus times as the surface temperature is going up the ma ximu m load of each beam is going down slowly and it is steep about 300℃. the decreasing patterns of long span beams are shown in fig. 2. at the table , it is shown as the lengths of beams are longer the reduction ratios are higher. fig. 2 ma ximu m loadings versus surface temperatures fig. 3 deflections versus elapsed times deflection is very important factor to rec ognize the structural stability in beams. as the time is passing, each beam’s deflection is shown in fig. 3. as the time is passing in other words, the surface temperature of each bea m is going up, the deflection is larger and the longer the beams are, the larger the deflections are. 0 100 200 300 400 500 600 700 800 900 1000 0 10 20 30 40 50 60 t e m p e ra tu re (℃ ) time(minutes) standard fire curv e steel temperature(fr 490) 0 50 100 150 200 250 300 350 0 100 200 300 400 500 600 700 800 900 m a x im u m l o a d ( k n ) temperature(℃) simple beam(4100mm) simple beam(4400mm) simple beam(4700mm) simple beam(5000mm) 0 5 10 15 20 25 30 0 10 20 30 40 50 60 d e fl e c ti o n ( m m ) time (min) simple beam(4100mm) simple beam(4400mm) simple beam(4700mm) simple beam(5000mm) proceedings of engineering and technology innovation , vol. 2, 2016, pp. 11 13 13 copyright © taeti 4. conclusions to know the structural behavior of beams made of fr 490 at high temperature, an eng ineering method was applied and the followings are obtained. (1) as surface temperature of bea m is going up, the load-bearing capacity is going down. (2) as the lengths of beams are longer, the maximu m load capacities are decreased. (3) deflection is increased highly when the surface temperature of bea m is going up and the length of beam is longer. acknowledgement this research was supported by a grant (code 15a udp -b100356-01) fro m an urban-architectura l research progra m funded by ministry o f land, infrastructure and transport of korean government. references [1] y. sa ku moto, t. ya maguchi, m. ohash, and m. sa ito, “high-te mperature propert ies of fire resistant steel for buildings,” journal of structural engineering, vol. 119, pp. 392-407, 1992. [2] i. k. kwon, “experimental study on ma king databases for fire resistant steel at high temperature,” journal of korean institute of fire science and eng ineering, vol. 27, no. 5, pp. 1-7, 2013. [3] swedish institute of stee l construction, “fire engineering design of steel structures,” stockholm, sweden, 1976. [4] h. y. chung, c. h. lee, w. j. su, and r. z . lin, “application of fire -resistant steel to beam-to colu mn mo ment connections at elevated temperatures,” journal of constructional steel research, vol. 66, pp. 289-303, 2010.  proceedings of engineering and technology innovation , vol. 3, 2016, pp. 04 06 4 security enhancement on reconfiguring coded wavelength with tunable wavelength filter array triggered chaotic sequences yao-tang chang 1,* , jen-fa huang 2 , yen-chung huang 2 and yan-tai liou 1 1 department of information technology, kao yuan university, kaohsiung, taiwan . 2 department of electrical engineering, national cheng kung university, tainan, taiwan . received 02 february 2016; received in revised form 17 march 2016; accept ed 02 april 2016 abstract in current study, the reconfigurab le opt ical code-div ision mu lt iple -access (ocdma ) scheme is imp le mented that the chaos sequence is c reated non-linea r time-va riant sequence as secret key and then t rigger tunable wavelength filte r array as rando m wavelength allocation . in the encryption, the distribution o f light ca rrie r is designed and imp le mented by using tunable wave length filte r array triggered chaot ic sequence. in add ition, the a rrayed wavegu ide grating (aw g) router is re written with ma ximal length code (m-sequence) to act as encoder. in the decryption, sy mmet ric sche me and ba lanced photo-detector is presented and reconfigu rable mechanis m is fo llowed the enc ryption synchronously either public o r p rivate channel. hence, the mu ltip le ac cess interfe rence (mai) is cancelled co mplete ly wh ile chaotic sequence is varied synchronously in trans mitte r and re ceiver. co mpared to p revious reconfigurab le scheme by t riggered reg ister and switches afte r awg router, the simulat ion results show that the secret key nu mbe r of proposed cryptography is significantly inc reased to avoid eavesdropping attack in physical layer. keywor ds : reconfigurable coded wavelength, tunable wavelength filter array, chaotic sequence, optical code-division multiple-access (ocdma) 1. introduction the optica l code-div ision mu ltip le -access (ocdma) techn ique has attracted con sidera ble attention for the applicat ion in local-a rea networks because it p rovides a burst and asynchronous mult iple-access environ ment in both the time and the spectral domains [1-6]. the recon figu rable sche me was p resented with triggering register and switches config ured beh ind aw g route r in prev ious wo rk [7]. howeve r, the t rigge ring varies of reg ister was limited by m-sequence code pattern result ing in few reconfigurab le state. the propos ed cryp tography is significantly increased while the tunable wavelength filte r a rray is configu red in front of coded awg router. the proposed encryption/decryption is presented with tunable wavelength filte r array triggered chaotic sequence in section 2. sect ion 3 eva luates the secret key nu mber o f proposed cryptography to avo id eavesdropper’s attacking in physical laye r. finally, we p rovide so me concluding remarks and future works . 2. the proposed cryptography configured with random wavelength distribution as shown in fig. 1, the reconfigurable ocdma-based encryption is designed that the chaos sequence is created non-linear time-variant sequence as secret key and then trigger tunable wavelength filter array as rando m wavelength allocation of broadband light source. the fbg is generally used to play wavelength filter role. in current study, the tunable wav elength filter array is triggered with chaotic s equence. broadband light source is designed to depend on many varies of tunable wavelength filter array. * corresponding aut hor. email: t 10066@cc.kyu.edu.tw proceedings of engineering and technology innovation , vol. 3, 2016, pp. 04 06 5 copyright © taeti in the dec rypt ion end sho wn in fig . 2, symmet ric sche me and ba lanc ed ph o to-detecto r is presented and recon figurable mechan is m is fo llo wed the enc rypt ion synchronously e ithe r pub lic or private channe l . hence , the mu lt ip le access inte rfe rence (ma i) is c ance lled co mp lete ly wh ile the tunable wave length filte r a rray is trigged by the sa me rando m p ara mete r and in it ia l va lu e of ch aotic sequence chaot ic sequence synchronously in receiver end. fig. 1 the proposed encryption with tunable wavelength filter array fig. 2 the proposed decryption with tunable wavelength filter array 3. the secret key evaluation of proposed cryptography by s elect ing rando m pa ra mete r and in it ial value o f chaot ic sequence as the secret key to trigger tunable wavelength filter array , many of rando m wave length allocat ions is obtained resulting fro m tunable wave length filte r d istribut ion and then the proposed cryptography is imp le mented and chara cterized with the re configurab le and fle xib le wave length hopping to avoid the tracking easily by eavesdropping. here , the d istributed pattern l of broadband light source can be possibly assigned and ob tained by probab ility funct ion o f statistics in eq . (1). n / m n n m n 2m m l m m m m n! = m!                     (1) where m donates the code length of m-sequence code and provide the usage of m authorized users. n denotes the ma ximu m chipped wave length nu mbe r of broadband light source. fig. 3 the number of secret key for diffe rent code length co mpa red to p revious recon figurable scheme by triggered register and s witches be hind aw g router, the simu lat ion results show that the secret key nu mbe r of proposed cryptography is sign ificant ly inc reased shown in fig . 3. for 63 authorized users’ considerat ion, the inc reasing pe rfo rmance of p roposed scheme is more 10e+36 times than previous scheme [7]. fig . 3 the eva luation and co mpa rison of the total nu mber of secret key with the proposed wave length filter array and p revious sche me [7]. proceedings of engineering and technology innovation , vol. 3, 2016, pp. 04 06 6 copyright © taeti 4. conclusions since the ocdma scheme is characterized with h igher bandwidth and confidentiality, this have much attractive for many researches to enhance the security by increasing the secret key (i.e., code family) of authorized users. the reconfigurable cryptography applied the chaos sequence to create non-linear t ime -variant sequence as secret key and trigger tunable wavelength filter a rray as random wavelength allocation. in addit ion, the aw g router is re written with ma xima l length code (m-sequence) to act as encoder. hence, the more rando m wavelength allocations is obtained resulting fro m tunable wavelength filter distribution and imple ment the reconfigurable and fle xib le wavelength hopping to avoid the tracking easily by eavesdropping. co mpared to previous reconfigurable scheme by triggered register and switches behind awg router, the simu lation results show that the secret key number of proposed cryptography is significantly increased to avoid eavesdropper’s attacking in physical layer. for 63 authorized users’ consideration, the increasing performance of proposed scheme is more 10e+36 times than previous scheme [7]. hence, the degree of confidentiality for proposed cryptography will be evaluated and verified secure enhancement of the ocdma scheme in future works. references [1] j. a. sa lehi, “ code division mult iple -access techniques in optical fiber networks —part i: fundamental principles,” ieee trans. commun., vol. 37, no. 8, pp. 824-833, august 1989. [2] y. t. chang, j. f. huang, c. c. wang, c. t. yen, h. c. cheng, and l. w. chou, “adaptive modified time-spreading and wavelength-group-hopping embedded m-sequence code for improved confidentiality over synchronous networks ,” optical engineering, vol. 50, no. 5, pp. 055001, may 2011. [3] m. kavehrad and d. zaccarin, “optica l code-division-mu ltip le xed systems based on spectral encoding of noncoherent sources,” j. lightw. technol., vol. 13, no. 3, pp. 534-545, march 1995. [4] h. ta kahashi, k. oda, h. toda, and y. inoue, “transmission characteristics of arrayed waveguide n × n wavelength mu lt iple xer,” j. lightw. technol., vol. 13, no. 3, pp. 447-455, march 1995. [5] z. wang, j. chang, and p. r. prucnal, “theoretical analysis and experimental investigation on the confidentiality of 2-d incoherent optical cdma system,” ieee j. lightwave technol., vo l. 28, no. 12, pp. 1761-1769, june 2010. [6] y. t. chang and y. c. lin, “ dynamic re configurable encryption and decryption with chaos/m-sequence mapping algorith m for secure h.264/a vc v ideo streaming over ocdma passive optical network,” multimedia tools and applicat ion, vol. 74, no. 15, pp. 1931-1948, july 2015. [7] y. t . chang, c. c. sue, and j. f. huang, “robust design for reconfigurab le coder/decoders to protect against eavesdropping in spectral a mp litude coding optical cdma networks,” j. lightw. technol, vol. 25, no. 8, pp. 1931-1948, augus t 2007.  proceedings of engineering and technology innovation , vol. 2, 2016, pp. 23 25 23 experimental study of progressive compression method of resin delivery in liquid composite molding chih-yuan chang * , wei-ru chen department of mechanical and automation engineering , kao yuan university, kaohsiung, taiwan. received 18 march 2016; received in revised form 13 april 2016; accept ed 16 may 2016 abstract a new technique of resin delivery, which we refer to as the progressive compression method (pcm), has been invented to reduce filling time associated with the vacuum assisted resin transfer mold ing (va rtm) process. in the method, the bag is divided into several segments. during infusion, a ll seg mented bags are pulled away fro m the prefo rm by the vacuum. hence, resin is easily infused into the loose preform. once enough volume of resin is infused, the vacuum within the segmented bags is released in a step-wise manner. the at mospheric pressure of the heated air is progressively applied on the segmented bag that is inflated to compact the wetted preform and drive the resin through the re main ing dry preform. the resin flo w is enhanced since the dry preform re ma ins loose during the filling process. the research aims to investigate the effect of seven process parame ters on the pcm co mp lete filling process by applying taguchi’s method. all chosen factors are designed with two levels. expe rimental results show that the predicted optimu m settings are higher vacuu m pressure, more co mpression segment, later init iation of the ne xt co mpression segment, higher air te mperature, later introduction of the heated air, lower init ial height of the cavity and less excess infused resin for reducing the filling time. co mpared with typical va rtm without flow enhancement, pcm at the optimu m settings reduces the filling time by 72.85%. ke ywor ds : filling process , taguchi’s method, vacuum assisted resin transfer molding 1. introduction vacuum assisted resin transfer molding (va rtm) process is a popular method for the manufacture of co mposite components. in a typical va rtm, the dry preform is placed on the rigid mold and sealed with an elastic bag to create an a irtight mo ld cav ity. the in let line is opened and the resin is driven by atmospheric pressure through the infusion line a fter the vacuum is applied within the mo ld cavity. on ce the preform is comp letely saturated, the removal of the unnecessary resin fro m the wetted preform is performed to achieve the uniform thickness of the part. fig. 1 schematic diagram for pcm x x x resin vacuum atmosphere (a) (b) (c) (d) vacuum vacuum atmosphere atmosphere vacuum *corresponding aut hor. email: yuan@cc.kyu.edu.t w proceedings of engineering and technology innovation , vol. 2, 2016, pp. 23 25 24 copyright © taeti a key disadvantage of vartm is a long infusion time because the driving pressure is lo w and the flow resistance offered by the co mpacted preform is high. recent methods of flow enhancement have been devised. in the present study, the progressive compression method (pcm), sharing the common characteristic with the articulated resin transfer molding [1] and the vacuum assisted compression resin transfer mo lding [2], is proposed to improve infusion time as shown in fig. 1. in the process , the s equential in flat ion motion of the segmented bag is designed to squeeze out the resin within the wetted preform and drive the resin through the re main ing dry and loose preform. the resin flo w is thus enhanced. in the present study, the effect of seven process parameters, including vacuu m pressure in the mold cavity (factor a), nu mber of the compression segment (factor b), co mpression timing of the ne xt segment (factor c), te mpera ture of the heated air (factor d), in itiating segment of the heated air (factor e), initia l height of the cavity (factor f) and volu me of the infused resin (factor g), on the pcm comp lete filling process is investigated by applying taguchi’s method. the typica l va rtm filling process is also performed for comparison purpose. 2. method 2.1. materials in the e xperiments, four layers of bi-directional fiber mat (t gfw-600) we re stacked together to form a preform. the used resin was commerc ial unsaturated polyester resin system (eterna l che mica l, 2597pt -6) including 0.5% o f the catalyst. before resin infusion, resin and hardener (m ekpo) we re mixed well in a we ight ratio 100: 1. the dimensions of the rectangular cavity were 280 mm × 150 mm × 10 mm. the cavity thickness could be adjusted by placing an extra plate in the cavity. 2.2. taguchi’s method the chosen factors are designed with two levels as shown in table 1. thus, the l8(2 7 ) standard matrix can be applicable . eight experiments are performed and the factors corresponding to each column are varied a mong the levels within the matrix. table 1 chosen factors fact or level 1 level 2 a. vacuum pressure (kp a ) 100 80 b. number of compression segment 2 3 c. compression t iming of t he next segment (sec) 5 10 d. t emperature of the heated air (℃) 20 40 e. init iat ing segment of t he heated air 1 st segment 3 rd segment f. init ial cavit y height (mm) 7 10 g. volume of infused resin (ml) 100 130 for factor b, the position of the compression segment is shown as fig. 2. fig. 2 position of the compression segment the minimizat ion of filling process is the goal of optimization. a signal-to-noise (s/n) ratio depicting the ‘small the better’ characteristic can be calculated using the formula.          n i i y n ns 1 2 10 1 log10/ (1) where yi is the measured property and n corresponds to the number of sa mples in each test tria l. the measured property is compiled in decibe l (db) in the matrix e xperiment. the s/n ratio is also denoted by η. 3. results and discussion performing the technique of the analysis of variance (anova) investigates the relative effect of different factors on the filling t ime as shown in table 2. the variance ratio, denoted by r, is the ratio of the mean square (ms) due to a factor and the error mean square. that is to say, the larger the value of r, the more important that factor is in influencing the process resp onse η. therefore, the process variables are ranked in order of importance of vacuum pressure, volume of infused resin, initia l cavity height , tempera ture of the heated air, co mpression timing of the next segment, nu mber o f co mpression segment and initiat ing segment of the heated air. re ferring to the sum of squares (ss) colu mn , notice that factor a (vacuum p ressure) ma kes the largest contribution to the total sum of squares, name ly, (40.93/57.67) × 100 = 70.97%. factors air outlet air outlet air inlet proceedings of engineering and technology innovation , vol. 2, 2016, pp. 23 25 25 copyright © taeti g (volu me of infused resin), f (in itia l cavity height) and d (temperature of the heated air) ma ke a 20.12% , 6.09% and 2.26% contribution to the total, respectively. factor e (in itiating segment of the heated air) makes the least co ntribution to the total, on ly 0.09% . this is because the temperature of the heated air at the outlet is found to be roughly 24℃. analyses of the experimental η show that the predicted optimu m settings are a1b2c2d2e2f1g1. table 2 anova factor average  by factor level df ss ms r 1 2 a. -53.8 -58.3 1 40.93 40.93 669.8 b. -56.2 -56.0 1 0.07* 0.07* 1.1 c. -56.2 -55.9 1 0.20 0.20 3.2 d. -56.5 -55.7 1 1.30 1.30 21.4 e. -56.1 -56.0 1 0.05* 0.05* 0.9 f. -55.4 -56.7 1 3.51 3.51 57.5 g. -54.9 -57.3 1 11.60 11.60 189.9 error 0 0 t otal 6 57.67 (error) (2) (0.12) (0.06) * indicates the sum of squares added to estimate the error sum of squares indicated by parentheses. 4. conclusions a new vartm process, called progressive compression method, is successfully developed to reduce the filling time . in the present study, the effects of seven process parameters on the filling t ime are investigated by taguchi’s method. expe rimental results show that the vacuum pressure, the volume of infused resin and the temperature of the heated air are significant variab les for reducing the filling time , while the in itiat ing timing of the heated air appears to be trivial. references [1] j. h. choi and c. k. h. dharan, “mold fill time and void reduction in resin transfer molding achieved by articulated tooling,” journal of composite materials , vol. 36, no. 19, pp. 2267-2285, october 2002. [2] c. y. chang, “experimental analysis of mold filling in vacuum assisted compression resin transfer molding,” journal of reinforced plastics and composites, vol. 31, no. 23, pp. 1630-1637, december 2012.  proceedings of engineering and technology innovation , vol. 3, 2016, pp. 19 21 19 flexible macroblock ordering scramble encryption techniques for h.264/avc videos yih-chuan lin 1,* , wei-siang wang 1 , yao-tang chang 2 1 department of computer science and information engineering, national formosa university, yunlin, taiwan. 2 department of information technology , kao yuan university, kaohsiung, taiwan. received 01 february 2016; received in revised form 12 march 2016; accept ed 02 april 2016 abstract in this paper, a ne w v ideo encryption method through scramb ling the co mpressed videos is presented, which is targeted for h.264/a vc video encryption in order to provide the greatest content protection in the compressed domain. the proposed algorithm uses a macroblock switching mechanism to scra mble the video content for the compressed video sequences . only the instantaneous decoded reference (idr) pictures are scrambled to take the advantage of drift e rror propagation fro m a ll the inter-prediction fra mes. the proposed encryption technique is designed to perform after the encoding process on the compressed bitstream such that bitstream is modified by the parser directly. fina lly, the scra mbled video is still format-co mpliant to a general h.264/a vc d ecoder and can only be recovered by the authorized user that owns private key. based on experimental results, the scene in the scrambled video can be effective ly protected by the proposed scheme with low co mputational co mplexity and negligible bitrate overhead. ke ywor ds : vio lation of privacy, h.264/a vc, video encryption, protection effect 1. introduction with the rapid advance of mu ltimed ia communicat ion and network technology, digita l video applications are easy to be found everywhere. the comp ressed bitstream is transmitted over the internet, decoded on a c lient and showed. digital v ideo surveillance and applications can be found ubiquitously in our daily lives. for manage ment purposes, the captured videos are usually encoded and transmitted over the internet to a third-party video service provider. vio lation of personal privacy and possible leakage o f v ideo scene are addressed in the study. in the case of video surveillance system, in order to archive a huge surveillance data that generated fro m each ip ca me ra, the captured video usually trans mitted to third-party servers . these servers are so-called delegates that provide high capacity for storage; reliable bandwidth resources for end users to access; and simp le manage ment interface fo r video owners. ho wever, there may have some untrustworthy system administrators that take a pee k at the uploaded video content in third-party service provider. selective encryption is a particu lar technique to encrypt the video content; only sens itive part of the bitstream is encrypted. for examp le, the stream c ipher is just applied to ce rtain parts that are sensitive for the video scene; s elected bits are used xor bit wise operation with cipherte xt that generated by stream c ipher. shahid et al. [1] proposed a format-co mp liant selective encryption method. for ca vlc, signs of trailing ones and magnitude’s level suffix in the specified range are encrypted. for cabac, the magnitude’s eg0 sub-suffix in the specified range and signs of non-zero quantized transform residual are encrypted. it is no change in final bitrate. wang et al. [2] proposed a tunable encryption scheme; besides the above-mentioned, signs of motion vector and intra prediction mode are also encrypted. obviously, the corresponding computational cost is relatively lowe r than full encryption. 2. method in this study, we imple ment two video encryption algorith ms that prevent the sensitive * corresponding author. email: lyc@nfu.edu.t w proceedings of engineering and technology innovation , vol. 3, 2016, pp. 19 21 20 copyright © taeti content fro m un-authorized access in the untrustworthy cloud. the encrypted video only recovered from the trusted person who owns a privacy key. the proposed s ystem arch itecture is shown in fig. 1. there e xist two video scrambling mechanisms that can be used to satisfy diffe rent situations in the video encryption process . fig. 1 proposed system scenario the proposed video scrambling methods are separately described as follows : 2.1. scrambling with other videos in a video surveillance environ ment, the surveillance ca me ra is wide ly d istributed over the building. the first scheme is to scramb le mu lti-way v ideo contents simu ltaneously ; every sequence contains part of pictures that belong to the others . the coded idr pictures in each s equence are denoted by v = {v1, v2, ..., vi}, where i is the number of input sequences. the ith coded idr p icture can be represented by vi = {vi1, vi2, ..., vij}, where j is the number of slice groups. after the encryption process, the scrambled idr picture is turned into }ˆ,...,ˆ,ˆ{ˆ kjk2k1i vvvv  , where 1 ≤ k ≤ i. (a) generate the h.264/avc bitstream: each came ra contains an h.264/a vc encoder, the fmo configurations are listed below:  the fmo map type is dispersed mode.  4 slice groups in a coded picture. (b) detect the idr picture: the detector parses the input bitstream and looks for the nalu type to indicate whether the idr p icture has happened or not. if the idr picture is detected, the sentinel value will be signaled. (c) redirect the nalu path: the dispatcher is controlled by a bitstream detector. if the sentinel value is true, the selector signal will be set to the specified value and the input nalu will be redirected to the corresponding output position. the above-mentioned selector signal is determined by a chaotic sequence that proposed in our previous study [3]. (d) data embedding: be fore transmitting the bitstream to the cloud storage, it is necessary to embed the privacy key informat ion in the bitstream. the re lated data hiding process can be found in our previous work [4]. 2.2. scrambling by itself in some situations, there is only one monitor in the building. the second method is to scra mble single video content by itself; the video content can be recovered independently. the ma in idea is that all the mac roblocks in the identical picture will be switched to a new position while the final bitstrea m is still format-co mp liant to a genera l h.264/a vc decoder. the subset of macroblocks in a picture can be listed below:  mbrow = {mb1, mb2, ..., mbr-1}, where r is the number of macroblocks in a row.  mbcolumn = {mbr, mb2r, ..., mb( c-1)×r}, where c is the number of macroblocks in a column.  mbrest = {mbk | r+1 ≤ k ≤ r× c 1, and k ≠ nr, n ∈n + }. let set x is the original macroblock position and set y is the scramb le mac roblock position. given a one-to-one and onto function f that assigns the set x to the set y. every ele ment of y is the image of unique element of x. due to the dependence on each adjacent mac roblock, a huge drift error may be caused by switching macroblock position. however, there is a problem in switching macrobloc k position directly. because coeff_token is encoded by the look-up table, switching macroblock position directly may cause a violation of the standard format. we need to re-encode the coeff_tok en to ensure that it can be found in the look-up table. 3. results and discussion the proposed video encryption scheme has been imp le mented in a video stream platform that is based on the joint model. proceedings of engineering and technology innovation , vol. 3, 2016, pp. 19 21 21 copyright © taeti 3.1. multi-way videos scrambling (a) encrypted hall monitor (b) encrypted container fig. 2 the encrypted video, frame #149, p picture fig. 2 illustrates a visual co mparison of original videos and scramb led videos in the 8-way ca me ra environment. as can be seen, the proposed scheme p rovides a high scra mb ling effect that protects the video content from unauthorized access. 3.2. single video scrambling (a) encrypted hall monitor (b) encrypted container fig. 3 the encrypted video, frame #149, p picture fig. 3 de monstrates the scramb ling e ffect which is quite d ifferent fro m the first method and no video content is leak out. this method provides a better scrambling effect. 4. conclusions in this paper, we propose two encryption schemes that provide quite high scra mb le e ffect for the video content in different scenario. it is a real-t ime application since only simp ly parsing and re-encoding in the co mpressed domain a re needed. although our approaches are applied after the co mpression, we can s till p reserve format-co mpliance. future works a im to seamlessly combine two methods to achieve more scrambling effects. references [1] z. shahid, m. chau mont, and w. puech, “fast protection of h.264/a vc by selective encryption of cavlc and caba c for i and p fra mes,” ieee t rans. circuits syst. video technol., vol. 21, no. 5, pp. 565-576, march 2011. [2] y. wang, m. o'neill, and f. kurugollu, “a tunable encryption scheme and analysis of fast selective encryption for ca vlc and caba c in h.264/a vc,” ieee t rans. circuits syst. video technol., vol. 23, no. 9, pp. 1476-1490, february 2013. [3] y. t. chang and y. c. lin, “the scra mb ling cryptography imp le mented with chaotic sequence trigger optical switch algorith m in wdm passive optical network,” proc. ieee int. ca rnahan conference on security technology, september 2015. [4] w. s. wang and y. c. lin, “a tunable data hiding scheme for ca bac in h.264/a vc video streams,” proc. ieee int. sy mposium on ne xt-generat ion electronics, may 2015.  proceedings of engineering and technology innovation, vol. 6, 2017, pp. 28 31 angular velocity analysis of an epicyclic gear train using fundamental circuits and a block diagram yi-chang wu 1,* , mi-ching tsai 2 1 department of mechanical engineering, national yunlin university of science & technology, yunlin, taiwan, roc. 2 departnent of mechanical engineering, national cheng kung university of science and technology, roc. received 18 july 2017; received in revised form 09 august 2017; accept ed 20 august 2017 abstract the prediction of angular velocity is a prerequisite to investigate the power flow and mechanical efficiency of an epicyclic gear train. this paper presents a useful tool, which combines the fundamental circuit and a block diagram, to analyze the angular velocity of an epicyclic gear train. it is a visualized and straightforward method without manipulating tedious kinematic equations, which makes t he angular velocity analysis of epicyclic gear trains more simplified. a compound epicyclic gear train used in a synchronous differential device is taken as an example to demonstrate the analysis process of the proposed approach. the velocity ratio of the epicyclic gear train is also derived. keywords: angular velocity, epicyclic gear train, fundamental circuit, block diagram 1. introduction an epicyclic gear train is a compact gear mechanism with the characteristics of a wide speed -ratio range, high efficiency, and light weight, which is commonly used in automotive differentials, automatic transmissions, and in -hub bicycle multi-speed transmissions [1]. for an ordinary gear train, it is relatively easy to observe the direction of motion of each gear element and the power flow through the gear train. it is more difficult, however, to visualize the kinematic behavior of an epicyclic gear tr ain directly by observation. there are several approaches used to deal with the angular velocity analysis of epicyclic gear trains, such as: the vector loop method [2], the relative velocity method [3], the tabular method [4], the algebraic method [5], and the fundamental circuit method [6, 7], which have been proposed over the last several decades. for these methods, the fundamental circuit method seems to be a straightforward one. a fundamental circuit is made up of two meshing gears assoc iated with a gear pair and a carrier to maintain a cons tant centre distance between these two gears. for a planetary gear mechanism with k gear pairs, k fundamental circuit equations can be listed. however, the velocity ratio of a planetary gear mechanism is obtained by tediously manipulating these fundamental circuit equations, which frequently causes human error. to overcome this drawback, the fundamental circuit method associated with a block diagram, which is based on the control community, is used to analyze t he angular velocity of a compound epicyclic gear train and calculate the velocity ratio of this gear mechanism. 2. a compound epicyclic gear train fig. 1 shows an epicyclic gear train coupled with an ordinary gear train used in a synchronous differential device. the epicyclic gear train consists of two sun gears (links 1 and 2), a compound planet gear (link 4-4*), and a carrier (link 3). a reverted * corresponding author. e-m ail address: wuy c@ y untech.edu.tw tel.: +886-5-5342601 ext. 4125; fax: +886-5-5312062 proceedings of engineering and technology innovation, vol. 6, 2017, pp. 28 31 copyright © taeti 29 ordinary gear train, which consists of two external gearsets (links 5 and 6, links 7 and 8), is further applied to constrain the motion of the two sun gears (link 1 and 2) of the epicyclic gear train. links 1 and 8 form an external compound gear, and links 2 and 5 also form a compound gear. therefore, the whole compound epicyclic gear train is a 1-dof (degree of freedom) gear mechanism, i.e., a single-input and single-output mechanism. the input link of this gear mechanism is the carrier (link 3), and the output link could be link 2 or link 1. 6 7 85 2 3 1 4 input 4 * fig. 1 an epicyclic gear train coupled with an ordinary gear train 3. angular velocity analysis 3.1. fundamental circuits for the epicyclic gear train depicted in fig. 1, there are two fundamental circuits. these can be identified as fundamental circuit i: (1, 4*)3 and fundamental circuit ii: (2, 4)3, respectively. the related fundamental circ uit equations can be listed as follows : 1 4*1 4 4*1 3 ( 1) 0        (1) 2 42 4 42 3 ( 1) 0        (2) where i is the angular velocity of link i, gear ratios are 4*1 4* 1/z z   and 42 4 2/z z   , and jz is the number of teeth on gear j. the negative sign represents the opposite rotational direction of an external gearset. due to the geometric constraint of the reverted ordinary gear train, the relationship between angular velocities 1  and 2  for links 1 and 2 is: 1 2 k  (3) where k = (-z7/z8)*(-z5/z6). 3.2. block diagram the kinematic characteristics of an epicyclic gear train can be carried out by using a block diagram [8]. eqs (1) and (2) are further expressed as: proceedings of engineering and technology innovation, vol. 6, 2017, pp. 28 31 copyright © taeti 3 0 30 3 1 4*1 4 3 ( )       (4) 2 3 42 4 3 ( )       (5) based on the linear relationships shown in equations (4) and (5), the related block diagram of the gear mechanism shown in fig. 1 is plotted as shown in fig. 2. r4*1 r42 + － － ω4 ω3 + + k +ω2 ω1 #2 ω3 #1 fig. 2 block diagram of the gear mechanism shown in fig. 1 3.3. velocity ratio the velocity ratio of a gear mechanism is defined as the ratio of the output shaft speed to the input shaft speed, i.e., the transfer function. by investigating the block diagram shown in fig. 2, the angular velocity ratios between 2  , 3  and 4  can be easily obtained as: 2 42 4*1 4 4*1 42 1 k k            (6) 3 42 4*1 4 4*1 42 1 k k k            (7) when link 2 is the output link and link 3 is the input link, the velocity ratio of this compound epicyclic gear train is 2 42 4*1 3 42 4*1 k          (8) here, a set of gear-teeth, as listed in table 1, of the presented compound epicyclic gear train shown in fig. 1 is taken as an example to determine the velocity ratio. the related gear ratios are 4*1 16 / 16 1     and 42 12 / 20 0.6     . in addition, the parameter ( 28 / 26) *( 24 / 30) 0.86k      . by applying equation (8), the velocity ratio of the gear mechanism is 2 3 / [ 0.6 ( 1)] / [( 0.86)( 0.6) ( 1)] 0.26           . since the value of the velocity ratio is less than 1, this gear mechanism can be used as a gear reducer when link 2 is the output link and link 3 is the input link. due to the enginering reality, only the coaxial lnks rotate about the stationary shaft a-a, i.e., links 1, 2 and 3, can be assigned as the input or output links. there are a total of 3 2 6p  different arrangements shown in table 2. the value of the velocity ratio for each case can be determined based on the block diagram shown in fig. 2. as depicted in table 2, the rotations of the output and input links for cases i, ii, iii and v are in opposite directions. case iv can be used as a gear multiplier, while case vi is a gear reducer. proceedings of engineering and technology innovation, vol. 6, 2017, pp. 28 31 copyright © taeti 31 table 1 the number of gear-teeth for each gear element of the compound epicyclic gear train z1 z2 z4 z4* z5 z6 z7 z8 16 20 12 16 24 30 28 26 table 2 six different arrangements of the input and output links for the compound epicyclic gear train case i ii iii iv v vi input link 1 1 2 2 3 3 output link 2 3 1 3 1 2 velocity ratio -1.16 -4.47 -0.86 3.85 -0.22 0.26 4. conclusion in this paper, the fundamental circuit method associated with a block diagram is applied to analyze the velocity ratio of a compound epicyclic gear train. the kinematic equations of an epicyclic gear train can be quickly listed by using fundamental circuits. a block diagram is used to represent the linear relationship of the angular velocity of each link. the velocity rat io can be easily obtained by investigating the blo ck diagram without tediously manipulating these kinematic equations. this provides a novel and useful tool to guide mechanical engineers in predicting the speed ratio of an epicyclic gear train. acknowledgement the authors are grateful to the ministry of science & technology (taiwan, r.o.c.) for supporting this research under grants no. most 104-2628-e-224-001-my3. references [1] y. c. wu, “structural analysis of existing multi-speed transmission hubs of bicycles,” proc. engineering and technology innovation, vol. 4, pp. 43-45, 2016. [2] d. gibson and s. kramer, “symbolic notation and kinematic equations of motion of the twenty -two basic spur, planetary gear trains ,” asme journal of mechanisms, transmissions and automation in design , vol. 106, no. 3, pp. 333-340, 1984. [3] z. levai, “structure and analysis of planetary gear trains ,” journal of mechanisms , vol. 3, no. 3, pp. 131-148, 1968. [4] r. l. norton, kinematic and dynamics of machinery, 1st ed. singapore, mcgraw-hill, 2009. [5] w. h. hsieh, “an experimental study on cam-controlled planetary gear trains,” mechanism and machine theory, vol. 42, no. 5, pp. 513-525, 2007. [6] f. freudenstein and a. t. yang, “kinematics and statics of coupled epicyclic spur gear trains ,” journal of mechanism and machine theory, vol. 7, no. 2, pp. 263-375, 1972. [7] y. c. wu and c. w. chang, “development of 3-speed rear hub bicycle transmissions with gear-shifting mechanisms ,” transactions of the canadian society for mechanical engineering, vol. 39, no. 3, pp. 407-418, 2015. [8] m. c. tsai, c. c. huang, and b. j. lin, “kinematic analysis of planetary gear systems using block diagrams ,” journal of mechanical design, vol. 132, no. 6, pp. 065001-1-065001-10, 2010  proceedings of engineering and technology innovation , vol. 2, 2016, pp. 28 30 28 color stability enhancement of white organic light-emitting diodes using a charge control layer shui-hsiang su * , yu-cheng lin, meiso yokoyama department of electronic engineering, i-shou university, kaohsiung, taiwan. received 07 april 2016; received in revised form 01 may 2016; accept ed 26 may 2016 abstract blue phosphorescent iridium complexes iridium (iii) bis [(4,6-di-fluorophenyl)-pyridinato-n,c2'] (firpic) and yellow phosphorescent iridium complexes iridium(iii) bis (4-phenylthieno [3,2-c] pyridinato-n,c2′) acetylacetonate (po-01) was doped into the small molecular phosphorescent host n, n_-dicarbazolyl-3, 5-benzene (mcp) to fabricate white phosphorescent organic light-emitting diodes (white pholeds). device current efficiency is enhanced by inserting a charge control layer (ccl) into the emitting layer. the peaks of pholed e lectrolu minescent (el) spectrum locate at 472 n m (firpic) and 560 nm (po-01). a h igh current efficiency white pholed has been fabricated by the use of firp ic and po -01 as the double e mitting layer (em l), in wh ich the mcp is used as the ccl inserted. the doping concentration of po-01 is optimized and the carrie r transport mechanis m of ccl is discussed. the optimized current efficiency is 30.06 cd/a. the cie coordinates locate at (0.33, 0.41) and vary with in (± 0.01, ±0.01) under driving voltage of 5-15v. keywor ds : oled, charge control layer (ccl) 1. introduction organic light-e mitting diodes (oleds) have attracted much interest over the past few years because of their properties of self-e mission, fast response time, h igh lu minance, low cost, and ease of fabrication, a mong other characteristics . 1,2 white organic light-e mitting diodes (oleds) are one of the most pro mising technologies which will like ly replace the e xisting liquid c rystal display (lcd). in particula r, white oleds (woleds) have drawn particular attention because of their use in full-color displays combined with red-green-b lue (rgb) co lor filters, liquid crystal display (lcd) backlights, and next-generation light sources. 3-9 white light emission can be obtained by mixing light of two comple mentary colors (such as red/bluish green, blue/orange, or green/magenta) or the three prima ry colors (red, green, and blue) fro m sma ll mo lecules and/or polyme rs. to obtain high luminance of a typical layered oled, the light-e mitting layer is generally doped with various fluorescent or phosphorescent dyes. in this paper, iridium (iii) bis [(4, 6-di-fluorophenyl)-pyridinato-n, c2’] (firp ic) and iridium (iii) bis (4-phenylthieno [3,2-c] pyridinato-n,c2′)acetylacetonate (po-01) are used as blue and yellow dopants. doping concentrations are optimized and the carrier transporting mechanism of ccl is discussed. 2. experiment the structures of woled devices used in this study were ito/ m-mtdata/α-npb/ mcp/ mcp:firpic/ mcp:po-01/t pbi/ lif/al. a thin charge control layer (ccl) layer of mcp is sandwiched between mcp:firp ic and mcp:po-01 em l to study the optoelectronic performance of woled devices. devices we re fabricated on glass substrates that had been precoated with a 150-n m-thick layer of indium-t in-o xide (it o) with a sheet resistance of 10 /square. cleaned and uv-ozone-treated it o substrates were loaded into an evaporation system at a base pressure of under 10 -6 torr. the woleds were formed such that the intersections between the ito anode and the cathode stripes each had an area of 0.24 cm 2 . the current density-voltage (j-v), lu minance-current density-luminous efficiency (l-j -ηl) and power effic iency-current dens ity-externa l quantum efficiency (ηp-j-ηext) characteristics were measured using a pr650 spectroscan spectrometer and a keithley 2400 p ro*corresponding aut hor. email: shsu@isu.edu.t w proceedings of engineering and technology innovation , vol. 2, 2016, pp. 28 30 29 copyright © taeti grammable voltage-current source. the el spectra and the co mmission internationale d’ec la irage (cie) coordinates of these devices were a lso obtained using a pr650 spectroscan spectrometer. 3. results and discussion an oled device without a ccl has the structure of it o/ m-mt data/α-npb/ mcp/ mcp:firpic/ mcp:po-01/tpbi/ lif/al. the optimized current effic iency (ce) is 20.1 cd/a and lu minance (l) is 16200 cd/m 2 . a thin ccl layer of mcp sandwiched between mcp:firpic and mcp:po-01 em l increase the ce to 24.1 cd/a. the thickness of each emitt ing layer is modified to observe the luminous characteristics. the optimized current efficiency of 30.06 cd/a and lu minance of 17000 cd/m 2 have been achieved. its cie coordinates locate at (0.33, 0.41) and vary within (±0.01, ±0.01) under driving voltage of 5-15v. fig. 1 shows the triplet e xc iton transfer mechanism in white pholed. fig. 1 the triplet states of t ebl, ccl and em ls . the trip let e xciton transfer mechanism in white pholed 4. conclusions in this paper, the fabrication of high co lor stability woleds with mcp as a ccl layer is demonstrated. the cie coordinates locate at (0.33, 0.41) and vary within (± 0.01, ±0.01) u nder driving voltage of 5-15v. the optimized current efficiency reaches 30.06 cd/a. expe rimental results reveal that we have successfully fabricated high efficiency and color stability woled by optimizing the doping concentration of po-01 and the thickness of ccl. acknowledgement the authors would like to thank the ministry of science and technology of the republic of china, fo r financially supporting this research under contract no. most 104-2221-e-214 -041 references [1] c. w. joo, j . w. sh in, j. moon, j. w. huh, d. h. cho, j. lee, s. k. park, n. s. cho, j. h. han, h. y. chu, j. i. lee , “highly efficient white transparent organic light e mitting diodes with nano-structured substrate,” org. electron., vol. 29, pp. 72, 2016. [2] r. joyce, k. singh, s. va rghese, j. akhtar, “effective c leaning process and its influence on surface roughness in anodic bonding for semiconductor device packaging ,” mater. sc i. se micond. process., vol. 31, pp. 84, 2015. [3] b. liu , j. zou , z. zhou , l. wang, m. xu, h. tao, d. gao, l. lan, h. ning, j. peng, “efficient single-e mitting layer hybrid white organic light-e mitting diodes with low efficiency roll-off, stable color and extreme ly high lu minance,” j. ind. and eng. chem., vol. 30, pp. 85, 2015. [4] c. w. tang and s. a. vanslyke, “ organic electrolu minescent diodes ,” appl. phys. lett., vol. 51,p p. 913, 1987. [5] g. t . chen, s. h. su, c. c. hou, and m. yo koya ma, “ effects of thermal annealing on performance of organic light-emitt ing diodes ,” j. electrochem. soc., vo l. 153, pp. j159, 2007. [6] j. f. li, s. f. chen, s. h. su, k. a. hwang, and m. yo koya ma, “full-wavelength white organic light-emitt ing diodes with blue fluorescence and phosphorescent iridiu m co mple xes ,“ j. electrochem. soc., vol. 153, pp. h195, 2006. [7] x. du, y. huang, s. tao, x. yang, x ding, x. zhang, “highly efficient wh ite fluorescence/phosphorescence hybrid organic light emitting devices based on an effic ient hole-transporting blue emitter,” dyes pigments, vol. 115, pp. 149, 2015. [8] y. s. tsai, a. ch ittawanij, l. a. hong, c. y. ou, f. s. juang, c. c. wang, s.h. la i, “adjusting dopant concentrations in solution process to optimize the white pho sphorescent organic light-e mitt ing diodes ,” microelectron. eng., vol. 138, pp. 31, 2015. http://www.engineeringvillage2.org/controller/servlet/controller?cid=quicksearchcitationformat&searchword1=%7bchen%2c+guan-ting%7d&section1=au&database=1&yearselect=yearrange&sort=yr http://www.engineeringvillage2.org/controller/servlet/controller?cid=quicksearchcitationformat&searchword1=%7bsu%2c+shui-hsiang%7d&section1=au&database=1&yearselect=yearrange&sort=yr http://www.engineeringvillage2.org/controller/servlet/controller?cid=quicksearchcitationformat&searchword1=%7bhou%2c+cheng-chieh%7d&section1=au&database=1&yearselect=yearrange&sort=yr http://www.engineeringvillage2.org/controller/servlet/controller?cid=quicksearchcitationformat&searchword1=%7byokoyama%2c+meiso%7d&section1=au&database=1&yearselect=yearrange&sort=yr http://www.ecsdl.org/vsearch/servlet/verityservlet?key=jesoan&possible1=li%2c+jian-feng&possible1zone=author&maxdisp=25&smode=strresults&aqs=true http://www.ecsdl.org/vsearch/servlet/verityservlet?key=jesoan&possible1=su%2c+shui-hsiang&possible1zone=author&maxdisp=25&smode=strresults&aqs=true http://www.ecsdl.org/vsearch/servlet/verityservlet?key=jesoan&possible1=yokoyama%2c+meiso&possible1zone=author&maxdisp=25&smode=strresults&aqs=true proceedings of engineering and technology innovation , vol. 2, 2016, pp. 28 30 30 copyright © taeti [9] b. s. kim, j. y. lee, “interlayer free hybrid white organic light-e mitting diodes with red/blue phosphorescent e mitters and a green therma lly activated delayed fluorescent emitter,” org. electron., vol. 21, pp. 100, 2015.  proceedings of engineering and technology innovation , vol. 3, 2016, pp. 10 12 10 optimization of coupler inlet for planar solar concentrator tien-hsiang yu, tun-chien teng * department of mechatronic engineering, national taiwan normal university, taipei, taiwan. received 25 february 2016; received in revised form 25 march 2016; accept ed 13 april 2016 abstract in this study, we imp le mented optimization for the coupler inlets of the planar solar co ncentrator, wh ich was proposed in our prior work. the planar solar concentrator had a waveguide slab to carry an array of light-collecting ele ments thereon. the light entering the elements is focused on the coupler in lets and guided into the slab with propagating therein through total internal re flection (t ir). thus, all the light fro m the elements is guided to the ends of the slab and highly concentrated. because the coupler in lets couple light into the slab and also decouple the light propagating in the slab in subsequent interactions, they must be optimized to balance between coupling effic iency and decoupling loss. the optimized arrangements of coupler in lets included three types here: uniform-thic kness platform, stepped-thickness platform, and stepped-width platform. in the simulat ion, the three types have the same area o f the inlet to keep the concentrator with the same toleran ce of the incident angle of light. we analyzed both the optical efficiency and range of the vertical angular distribution of the light reaching the outlet end of the slab; then made a co mparison for the three inlet arrange ments. the simulation results demonstrate the stepped-thickness and stepped-width platforms provide higher efficiency and concentrated irradiance. ke ywor ds : planar concentrator, coupler, solar energy, illumination design 1. introduction traditionally, the planar concentrator is equivalent to a device that divides a single large concentrated element into an array of optics (lens) coupled to a co mmon slab waveguide. the light entering each lens is focused on localized coupling mic rostructures embedded on the bottom of the waveguide and reflected at predicted angles, thereby propagating in the waveguide through total internal reflection (t ir), so the light fro m a ll the lenses is guided to the ends of the waveguide [1]. the advantages of the planar concentrator include a compact volume and a light we ight. because the coupling mic rostru ctures on the waveguide surface couple light into the waveguide and also decouple the light guided in the waveguide in subsequent interactions, the decoupling loss increases with the waveguide length. to manage the decoupling loss, various designs of the waveguide and co upler geo metries we re p roposed to reduce or eliminate interactions with the coupling mic rostructure [1-4]. however, these designs require an accurate align ment of the lens array and the coupling mic rostructure, which is adverse to assemble a large product. thus, we proposed a planar solar concentrator in our prior work, which had a waveguide slab to carry an array of t ir light-collecting ele ments thereon without requiring a lignment [5]. the light entering the e le ments is focused on the coupler inlets and guided into the slab with propagating therein through tir. thus, all the light fro m the ele ments is guided to the ends of the slab and highly concentrated. because the coupler in lets couple light into the slab and a lso decouple the light propagating in the slab in subsequent interactions , we further optimized them to ba lance between coupling e ffic iency and decoupling loss in this study. 2. design concept a coupler inlet is the junction of each t ir collector and the slab, through which the light focused by the tir collector enters the slab below. because the field angle of the sun itself is 0.52 o , the sunlight is focused onto a spot, rather than on a point; therefore, the coupler in let must have an area enough for the focused light to pass through. moreover, the t ir co llector focuses the light within a larger spot area if t racking tolerance is considered. thus, a coupler inlet with a * corresponding aut hor. email: walt er.teng@ntnu.edu.tw proceedings of engineering and technology innovation , vol. 3, 2016, pp. 10 12 11 copyright © taeti larger a rea contributes to an increased coupling efficiency. however, a coupler inlet with a larger a rea a lso raises the probability of decoupling the light propagating in the slab and thus reduces the propagation effic iency. because the total optical e ffic iency depends on the product of the coupling effic iency and the propagation efficiency, securing a balance between th em is critical. fig. 1 a tir collector on a slab with a slope-facet inlet in our prior work, the tolerance of trac king was set to ±0.5° for the rotation angle of the x-a xis and the z-a xis (i.e.  and , respectively; referring to fig. 1). the optima l a rea of the inlet is a trape zoid with two base widths of 0.9 mm and 0.3 mm, and a height of 1.05 mm. in this study, we optimized the arrangements of the inlets to raise the optical effic iency. the arrangements for optimization included three types: uniform-thic kness platform (case a), stepped-thickness platform (case b), and stepped-width platform (case c). in case a, the slab was flat; all the inlets were lined on the same horizontal level. in case b, the slab had a stepped thickness, and thus the location of the former inlet was higher than the later. in case c, the slab had a stepped width, and the inlets were on the same horizontal level, but the slab was tiled at a s mall angle with the line of inlets. the purpose of the arrangements of case b and c is to prevent the coupled light fro m lea king out of other inlets ahead. the slab of case b was the thickest; that of case c was widest; those of case a and b had the same width; those of case a and c had the s a me thic kness. therefore, case a had the smallest area of the outlet end of the slab and thus the largest geometric concentration ratio; case b had the same area as case c. however, case b and case c had the higher propagation efficiency. moreover, the inlet end of the slab in case b and c had a slope facet to reduce the range of the vertical angular distribution of the light reaching the outlet end of the slab, wh ich help the light be further concentrated by attaching an extra co mpound parabolic co llector (cpc) to the outlet end. therefore, we analyzed both the optical effic iency and range of the vertical angular distribution for the three cases under diffe rent para meters of the slab; then made a co mparison to find the optimal conditions . 3. results and discussion in this study, we established an optical model that had a t ir collector on a narro w slab with a slope-facet inlet. then, we varied the angle of the slope facet (slope angle ) and the thickness of the slab for conducting simu lations to find the optima l condit ions. moreover, we made a co mparison for the three arrangements of coupler inlets under the optima l condition. in the optical model, the t ir co llector had a p itch of 10 mm, a height of 15 mm, and a width of 15 mm; the slab had a width of 1.0 mm, and a length of 60 mm; the t ir collector and slab were made of poly methyl methacrylate (pmma). the used ray-tracing software was lighttools 8.3. fig. 2 slope angles vs. ranges of vertical angular distributions and efficiency fig. 2 shows the effects of slope angles on both the optical efficiency and range of vertical angular distributions. the light is incident at an angle (=0.4, =0) with an angular range of 0.26° (half-angle ). the range of vertica l angular distributions substantially decreases as the slope angle decreases, but increases as the thickness of the slab decreases. moreover, slope angles have litt le effects on the effic iency, but the efficiency dramat ically drops if the slope angle is below 4°. furthermore, the effic iency is not affected by the proceedings of engineering and technology innovation , vol. 3, 2016, pp. 10 12 12 copyright © taeti thickness of the slab. fig. 3 shows the effect of slope angles on the optical effic iency for the light at inc ident angles of (0, 0), (0.4, 0), (0.4, 0.4), and (0.8, 0). except the incident angle (0, 0), the other incident angles have the simila r trend, dra matic drop in the effic iency when the slope angle below 4°. fig. 3 slope angels vs. optical efficiency finally, we assembled five t ir collectors according to the three types of arrange ments for conducting simulat ion. case a had a slab with a thickness and width of 1.25 mm and 1 mm, respectively; case b had a slab with a thic kness and width of 2.5 mm and 1 mm, respectively; case c had a slab with a thic kness and width of 0.5 mm and 5 mm, respectively. case b and c had an inlet with a slope facet of 4°. the irra diance of the outlet end of case a is 1.9 times that of case b or c. however, the range of the vertical angular distribution in case a was 96°; those in case b and c were 22°, which means case b and c can be further concentrated 3.8 times case a by attaching an extra cpc the outlet end of the slab. overall, case b and c can provide the concentrated irradiance 2 times case a at most. 4. conclusions in this paper, the arrangements of coupler inlets are ana lyzed and optimized. as compared with the uniform thickness platform, the stepped-thickness and steppedwidth platforms with a slope-facet inlet end o f 4° provide higher efficiency and double concentrated irradiance. references [1] j. h. karp, e. j. t re mb lay, and j. e. ford, “planar micro-optic solar concentrator,” opt. exp ress , vol. 18, no. 2, pp. 1122-1133, 2010. [2] y. liu, r. huang, and c. k. madsen, “design of a lens -to-channel waveguide system as a solar concentrator structure,” opt. express , vol. 22, no. s2, pp. a198-a204, 2014. [3] o. selimog lu, and r. turan, “ exp loration of the horizontally staggered light guides for high concentration cpv applications,” opt. express , vol. 20, no. 17, pp. 19137-19147, 2012. [4] b. l. unger, g. r. sch midt, and d. t . moore, “dimp led planar light guide solar concentrators,” osa international optica l design confe rence and opt ical fabrication and testing, paper itue5p, 2010. [5] h. y. wu and s. c. chu, “ ray-lea kage-free saw tooth-shaped planar light guide solar concentrators,” opt. exp ress , vol. 21, no. 17, pp. 20073-20089, 2013. [6] t. c. teng and w. c. lai, “planar solar concentrator featuring align ment-free total-interna l-re flect ion collectors and an innovative compound tracker,” opt. express , vol. 22, no. 17, pp. a1818-a1834, 2015.  proceedings of engineering and technology innovation, vol. 3, 2016, pp. 28 30 28 using unsupervised machine learning to detect peer-to-peer botnet flows andrea e. medina paredes 1 , yuan-yuan su 2 , wei wu 3 , hung-min sun 4,* 1 institute of information systems and applications, national tsing hua university, hsinchu, taiwan . 2 department of computer science, national tsing hua university, hsinchu, taiwan. 3 fujian provincial key laboratory of network security and cryptology, school of mathematics and computer science, fujian normal university, fuzhou, fujian, china. 4 department of computer science, national tsing hua university, hsinchu, taiwan. received 22 february 2016; received in revised form 19 march 2016; accept ed 06 april 2016 abstract the wa r aga inst botnet infection is fought every day by users that want to feel safe against any threat of compro mise hosts. in this paper we are going to focus on the behavior of peer 2 pee r (p2p) botnets, which a long with hybrid botnets is a growing trend a mong attackers. the ma in approach will consist of a behavior comparison among features e xtracted fro m network flows, focusing only in the flows fro m p2p applications including p2p botnets. ke ywor ds : clusters, network flows, p2p botnets, unsupervised learning 1. introduction malic ious software such as botnets has been around for quite a t ime already and it keeps improving, evolving and growing, as for the detection systems, they try to keep track of these new e me rging botnets trends but some fa il to provide a definite and accurate solution to this problem. for e xa mp le, this past may “8chan”, a website composed of user-created boards, reported a series of ddos attacks coming fro m the “hola!” network, a popular virtual private network use for vie wing b locked v ideos and t v shows from other countries, which counts with a pool of 6 million ip addresses [1]. activities such as adding signatures to d atabases, protecting servers against hackers, the use anti-virus software to protect computers fro m getting infected, track c&c server activ ities and so many other actions are taken in co nsideration, but still cybercriminals find a way to go around the security measures. the use of supervised learning models is one of many approaches that can be use to deal with botnets, classifiers like support vector mach ines (svm ) have shown great accuracy separating botnet network flo ws fro m normal flows [ 2], other methods like decision tree a lgorith ms have been put to the test as well, measuring how accurate the decision tree classifies the data [3]. the drawback of the previous mention methods is that most of them need labelled data in order to function and only yields better results when the botnet signature is already known. in hopes to contribute to these efforts, in this paper we propose the use of unsupervised m achine learning algorithms for the fight against botnet detection. a comparison a mong three clustering algorith ms using network flo ws e xtracted fro m a set of features, will be carried out thorough out this paper. the rest of this paper, is organized with the following: the method use, the e xperiment design with the respective o bservations and finally the analysis of the results. 2. method the structural synthesis of ccpgts will be performed based on the creative design met hodology process [7-8]. the design require ments and design constraints are summarized based on the characteristics of the mechanism. 2.1. approach using unsupervised learning traffic can be classified by selecting its at* corresponding aut hor. email: hmsun@cs.nthu.edu.tw proceedings of engineering and technology innovation , vol. 3, 2016, pp. 28 30 29 copyright © taeti tributes which distinguishes their behavior, we want the unsupervised learning algorith m to find the patterns hidden among the p2p flows. to facilitate the algorith m detection a previous process to select the most relevant features will be carry out and then these input will be feed to the clustering algorithms in order to co mpare their overall performance creating clusters based on the characteristics of those features. then the resulting cluster will be cross validated in order to ensure the legitimacy of the outputs. fig . 1 shows a flowchart with the overall process: fig. 1 overa ll process of the approach using unsupervised learning 2.2. clusters evaluation measurements in this paper we choose simple k-means, farthest first [4] and db scan [5] to wo rk with clustering algorith ms to co mpare wh ich is the best fit to classify p2p botnets traces using network flows e xt racted fro m the pac kets send within a network [6].the data is d ivided later into subsets for evaluating the mach ine learning algorith ms, the sets contain p2p tra ffic fro m both kind’s malic ious flows and non malicious flows. pe rformance statistics are calculated for all the trials. the c lass is ignored during all the evaluations for the clusters. the validation measures are used to evaluate the credibility of the clusters, in this case due to the high imba lance between classes, to keep the rea l scenario of a network, we can’t only re ly on the accuracy measurement of the clusters. the classification oriented measures of cluster validity are described below: • the nu mber of correctly classified instances as malicious is referred to as the true positive (tp). • the number of instances classified as malicious but should be normal and therefore rejected, is refe rred to as the false positive (fp). • the number of instances classified as normal but are actually malicious, is referred to as the false negative (fn). • the number of normal instances from a class correctly rejected is referred to as true neg ative (tn). 3. results and discussion we need to utilize clustering algorithms which can handle such diffe rences in the data along with the first 10 top ranked attributes from the feature e xtraction. as mention before we used weka [6] to run this three clusters. 3.1. dataset assemble for testing the dataset used in this paper was obtained fro m a previous research group that made the datasets publicly available , their paper is about a p2p traffic categorization system called “peer rush” [8]. the labeled data of all four p2p applications (emule , utorrent, vu ze and fros twire) along with zeus and waledac were used for testing purposes. the dataset was divided into 3 co mbinations of subsets, containing both kinds of flows that are labeled e ither non-malicious or malicious: • dataset 1 (zeus traces ): a total of 17,940 flows are contained in the dataset, 95% non-malicious traces and 5% malicious traces. • dataset 2 (waledac traces ): a total of 12,310 flows are contained in the dataset, 93% non-malicious traces and 7% malicious traces. • dataset 3 (zeus and waledac traces ): a total of 12,334 flows are contained in the dataset, 92% non-malicious traces and 8% malicious traces. 3.2. unsupervised learning comparison the accuracy of all the algorith ms is shown in fig. 2. db scan performed significantly we ll for a ll the situations assigned, each change of dataset dimin ishes slightly the accuracy, but in genera l it ma intains the highest percentage. simp le proceedings of engineering and technology innovation, vol. 3, 2016, pp. 28 30 30 copyright © taeti k-means improved in the last test but still had some imba lance in the number of correctly classified ma lic ious instances that were retrieved. fig. 2 accuracy of unsupervised algorithms for all datasets in table 1 we can co mpare the real perfo rmance values of each algorithm. db scan proves to be a worthy contender against p2p botnets flows and may be used to improve the precision of detection systems along with other security tools. we believe that each algorith m performs depending on the quality of the data and the previous preprocessing of it. table 1 average measurement values for each algorithm 4. conclusions the use of unsupervised learning was proposing for classifying p2p traffic flows in comparison to the previous methods using supervised learning. the results lead us to believe the data is suitable for a density based cluster, since db scan a lgorith m performed we ll on every situation, obtaining high precision class ifying p2p botnet flows and retriev ing most of these malic ious flows from the normal p2p traffic. acknowledgement this research was supported in part by the ministry of science and technology , taiwan, under the grants most 104-3115-e-007-004 and most 103-2221-e-007-073-m y3. references [1] r. price, “ business insider,” http://www.bu sinessinsider.co m/hola -used-for-botnet-onchrome, may 2015. [2] p. barthaku r, m. dahal, and m. k. ghose, “a fra mework for p2p botnet detection using svm,” in international conference on cyber-enabled distributed co mputing and knowledge discover, sanya, 2012. [3] w. h. liao and c. c. chang, “peer to peer botnet detection using data min ing scheme,” international conference on internet technology and applications, wuhan, 2010. [4] s. hochbaum, “a best possible heuristic for the k-center problem,” in mathematics of operations research, 1985. [5] p. n. tan, m. steinbach, and v. ku mar, “cluster analysis: basic concepts and algorith ms,” introduction to data mining, pearson, pp. 487-559, 2005. [6] m. ha ll, e. frank, g. ho lmes, b. pfahringer, p. reutemann, and i. h. witten, “the weka data mining software: an update,” sigkdd explorations, 2009. [7] b. rahbarin ia, r. perd isci, a. lan zi, and k. li, “peerrush: min ing for unwanted p2p traffic,” detection of intrusions and malware, and vulnerability assessment, springer, vol. 7967, pp. 62-82, 2013. https://www.most.gov.tw/?menu_id=c53d1d66-f9ed-11e4-9a55-b8ac6f2d65c0&l=en  proceedings of engineering and technology innovation , vol. 3, 2016, pp. 16 18 16 a fast anticollision algorithm with early adjustment of frame length for the epcglobal uhf class-1 generation-2 rfid standard wen-tzu chen * department of transportation and communication management, national cheng kung university, tainan, taiwan. received 27 january 2016; received in revised form 16 february 2016; accept ed 11 march 2016 abstract this paper deve lops a fast algorith m to identify a large nu mbe r of tags in a short period of t ime durat ion in rfid systems. the ma in goal of the proposed algorith m is to meet the require ment in supply chain manage ment. in this paper, we de rive mathe mat ica l mode l to analy ze the tag identify process. based on the analysis, we can dec ide the optima l fra me length and when the current read round is early ended. we find that the optimal fra me length should be set to 1.72 times of the nu mbe r o f tags when the rat io bet ween co llision -slot duration and e mpty-slot durat ion is 4. through the use of early ad justment o f fra me length , the rfid interrogator e xa mines the fitness of fra me le mgth only at three t ime slots in each read round. the primary advantage of ou r algorith m is the ab ility to ach ieve a good comp ro mise between th roughput performance and computation complexity. keywor ds : rfid, tag identification, logistic management, read efficiency, anti-collision algorithm 1. introduction fo r mana g ing a la rge a mount o f ob jects and ident ify ing the ir ids, th e rad io frequen cy ide nt ificat ion (rfid) te chno logy ope rat ing on u lt ra -h igh frequen cy (uhf) b and is in cre as ing ly be ing used espec ia lly in log ist ic mana ge ment . to meet th e requ ire me nt o f rea d ing a dense popu lat ion o f ob jects, th e epcg loba l dev e loped an a ir inte rfa ce , na me d uhf class -1 gene rat ion -2 o r iso 18000-6 type c [1]. the a lgorith ms for identify ing mu lt iple tags in rfid systems have been wide ly studied in the lite rature [2-6]. the epcgloba l uhf class -1 generat ion-2 specifies an algo rith m based on dynamic fra me slotted aloha (dfsa ) to solve the rfid antico llis ion pro b le m. the author in [ 4] indicated that the read performance of dfsa depends ma in ly on the accuracy o f backlog estimate and fra me length setting. fo r the dfsa a lgorith m, a ma ximu m throughput of 36.8% can be achieved only if the fra me length is set to the nu mbe r of tags . one o f the most interesting p rocedures is how to dynamica lly ad jus t fra me length to ach ieve ma ximu m throughput when a b ig d iffe rence exists between frame length and tag quantity . 2. proposed algorithm this paper proposes a simple and efficient anticollision algorith m for the epcg lobal cass -1 generation-2 protocol and describes clear ru les for early adjustment mechanism of fra me length. this mechanism allows the interrogator to early end the current read round at any time slot and enter into a new round with a new fra me length. hence, the interrogator will e xa mine whether the current fra me length requires adjustment or not and how to adjust it if necessary. to achieve a good compro mise between computation comple xity and throughput performance, the proposed algorithm e xa mines the fitness of fra me length at only the three quart iles in each read round, i.e. at the l/4 th , l/2 th , and (3/4)l th slots. the procedures of the proposed method is shown in fig. 1. a key performce of anticollision a lgorith m is normalized throughput, *corresponding aut hor. email: wt chen@mail.ncku.edu.tw proceedings of engineering and technology innovation , vol. 3, 2016, pp. 16 18 17 copyright © taeti ( )/ s s c e u s t s t c t e t       (1) s, c, and e are the numbers of successful, collision, and e mpty slots. the t ime intervals of the successful, collision, and e mpty slots are ts, tc, and te, respectively. query(q) queryrep yes q=qopt update q with qopt no count (s, c, e) at the three quartiles slot estimate n=(s+2.39c)l/i qopt=round(log2(a·n)) fig. 1 proposed anti-collision procedures at the three time slots, the interrogator counts the number of singly occupied slots si, the number of co llision slots ci, and the number of empty slots ei. note that i is equal to l/4, l/2, or (3/ 4) l here . the tag quantity to be read at the beginning of the round can be estimated by [6]. ilcsn ii /)39.2(ˆ  (2) given tag quantity n, we atte mpt to determine an optima l fra me length lopt to ma ximize the throughput in (1). ta king the derivative of u with respect to l and setting the derivative to zero lead to an optimal frame length. nal  opt (3) in eq. (3), a is a coeffic ient. table 1 lists the coefficients for different values of tc/te. table 1 coefficients between lopt and n tc/te a (lopt=a×n) 1 1 2 1.30 3 1.53 4 1.72 5 1.89 6 2.05 7 2.19 8 2.32 9 2.44 after co mputing the optima l fra me length, the interrogator should check whether the cu rrent fra me length is appropriate or not. the fra me length must be a power of 2 due to the uhf class -1 generation-2 standard. hence, the optimal q value can be obtained by: qopt=round(log2lopt) (4) 3. simulation results the read performance of the proposed algorith m and the refe rence schemes was exa mined by carrying out computer simu lations based on the monte carlo technique. fig. 2 co mparison of throughput performance for anticollision algorithms fig. 3 time-saving efficiency the throughput performance of the proposed algorith m is presented in fig. 2. as mentioned before, our method uses early adjustment mechanis m for fra me length and performs at most three e xa minations in each read round for the fitness of fra me length. it can be found that our method has better throughput performance as compared with the other schemes. particularly, for a large numbe r of tags, i.e. 1000, the proposed algorithm can achieve up to 10% throughput imp rovement as compared with the proceedings of engineering and technology innovation , vol. 3, 2016, pp. 16 18 18 copyright © taeti typical dfsa. another advantage of the proposed method is the stability. when the nu mber of tags is increased, the proposed method can keep the throughput stable. fig. 3 presents the time -saving effic iency for different algo rith ms. it can be observed fro m fig . 3 that our a lgorith m can obta in 5 -10 % time -saving e ffic iency as the tag quantity is greater than 200. 4. conclusions this paper proposes an efficient rfid anticollision algorithm to improve the read performance for the epcglobal uhf class -1 generation-2 standard. the results show that the proposed algorithm can ach ieve up to 400 tags/s read speed. for a large number o f tags, i.e. 1000, the proposed algorith m can achieve up to 10% throughput imp rovement as compared with the typical dfsa. we believe that this advantage comes fro m the early adjustment mechanis m for fra me length. another advantage of the proposed method is the stability. when the nu mber of tags is increased, the proposed method can keep the throughput stable. acknowledgement the support of taiwan’s ministry of science and techno logy, unde r grant m ost 104-222 1-e-006-200 is gratefully acknowledged. references [1] epcg lobal, epc radio-frequency identity protocols class -1 generation-2 uhf rfid protocol for communicat ions at 860 mhz-960 mhz, version 1.2.0, oct. 2008. [2] h. vogt, “ efficient object identification with passive rfid tags,” international conference on pe rvasive co mputing , london, uk, pp. 98-113, 2002. [3] d. lee, o. bang, s. im, and h. lee, “ efficient dual bias q-algorithm and optimu m weights for epc c lass 1 generation 2 protocol,” 14 th european wire less conference, pp. 1-5, june 2008. [4] w. t. chen, “an accurate tag estimate method for imp roving the performance of an rfid anticollision a lgorithm based on dynamic fra me length aloha,” ieee trans. autom. sci. eng, vol. 6, no. 1, pp. 9-15, jan. 2009. [5] b. knerr, m. holze r, c. angerer, and m. rupp, “slot-wise ma ximu m likelihood estimation of the tag population size in fsa protocols,” ieee trans. co mmun ications, vol. 58, no. 2, pp. 578-585, feb. 2010. [6] w. t. chen, “a feasible and easy-to imple ment anti-collision algorith m for the epcg lobal uhf class -1 generation-2 rfid protocol,” ieee t rans. autom. sci. eng. , vol. 11, no. 2, pp. 485-491, apr. 2014.  proceedings of engineering and technology innovation , vol. 3, 2016, pp. 34 36 34 mitigating initialization bias in transportation modeling applications wonho suh * department of transportation and logistics engineering, hanyang university, ansan, korea . received 30 january 2016; received in revised form 14 february 2016; accept ed 08 march 2016 abstract traffic simulat ion model is a useful tool to evaluate real world transportation solutions in a risk free environ ment. traffic simu lation model requires some form of initia lization before their outputs can be considered meaningful. models are typically init ialized in a particula r, often “empty” state and therefore must be “warmed-up” for an unknown a mount of simulation time before reaching a “quasi-steady-state” representative of the systems’ performance. the portion of the output series influenced by the arbitrary init ia lization is refe rred to as the in itia l transient and is a widely recognized proble m in other areas, but less emphasized in the tran sportation application. after reviewing methods of accounting for the initia l transient bias, this paper selects and evaluates three techniques; two popular methods from the genera l simu lation fie ld, welch’s and mser method, and one from the current state of the practice in the transpo rtation application, vo lu me ba lancing. vissim models were created to compare the selected methods. after presenting the results of each method, advantages and criticis ms of each are discussed as well as issues that arose during the imple mentation. it is hoped that this paper informs the current practice in transportation application as to how to account for the in itia l transient in order to continue facilitating meaningful and reliable results. ke ywor ds : traffic simu lation, in itia lizat ion bias, simulation analysis 1. introduction traffic simu lation mode ling has become an increasingly vital instrument for the transport ation analysis. traffic simu lation modeling provides the flexibility to manipulate conditions that could influence network operations in a risk-free environ ment, allow ing for co mple x network ana lysis, testing assumptions and possible outcomes to determine their potential for imple mentation [1]. there a re a nu merous means to employ traffic simulat ion modeling to aid in the analysis and decision ma king process. thus, it is e xtre me ly impo rtant that the simula tion output is both meaningful and reliable. one area often overlooked in transportation is guidelines to govern the init ialization of tra ffic simulation models [2]. the simulat ion start-up problem is of significant interest and has been widely studied in simu lation re lated fie lds [3]. when a model is initia lized in a condition uncharacteristic of steady-state of the real-world condition it is attempting to represent bias may be introduced in the simulation’s output. the bias can in turn lead to inaccurate results and possibly faulty conclusions [3]. there are two categories of methods for mitigating the initia lization bias problem. the most common approach is truncation, or discarding the initia l data influenced by the starting conditions. the second approach is intelligent initia lization, o r starting the model in a state with a high probability o f be ing equilibriu m/steady state. however, it is not always convenient or even practical to start the simulat ion in steady state [2]. more importantly, determin ing equilibriu m a priori in a tra ffic simu lation model can be difficult and arbitrary. for e xa mp le, determining a priori how many vehic les to queue at each signal, where to place all the vehicles on a link, and what init ial speed may be nearly impossible in many instances. the need to eliminate initia lization bias, also known as the start-up problem, is a widely-recognized challenge with simu lation analy* corresponding author. email: iamwonho@gmail.com proceedings of engineering and technology innovation , vol. 3, 2016, pp. 34 36 35 copyright © taeti sis [3]. th is occurs because non -terminating simu lations do not have predefined run lengths or in itia l conditions. the simulat ion processes must be initia lized arb itrarily, wh ich creates bias in steady-state parameter estimates. although methods of re moving init ialization bias e xist, there is currently no largely accepted method that performs suitably in all applicat ions. add itionally, there is an overall negligence of the initial transient problem in practice [2]. the purpose of this paper is to exa mine and evaluate the effectiveness of several techniques used for eliminating initia lization bias fro m transportation applications. 2. method methods used to mitigate init ialization b ias attempt to inform the treatment of data affected by the initial transient for discrete, stochastic simu lation models. and as a result these methods seek to provide more accurate, meaningful and re liab le results for simu lation output. the methods can be grouped into the following categories as described by robinson [4]: graphical, heuristic, statistical, in itia lizat ion bias testing, and hybrid methods. 2.1. graphical methods the most common methods to identify the initia l transient are graphical procedures [4]. graphica l procedures consist of a visual inspection of the output time series to determine the e xtent of the initia l transient. a major advantage of the graphica l method is its simp licity and the reliance on few assumptions. a disadvantage is that these methods are typically highly subjective and the truncation points could vary based on the analyst’s judgment. fish man’s and welch’s method are two are exa mp les of graphical methods. 2.2. heuristic methods heuristic methods provide definitive rules or formulas to determine the length of the warm-up period [4]. the advantages of these methods are lack of user specific subjectivity, eas e of imple mentation, and only a few assumptions are generally needed. marginal standard erro r ru le (mser), conway’s rule , crossing of the means rule , and replicated batch means are categorized as heuristic methods. 2.3. statistical methods statistical methods rely on statistical principles to determine the wa rm-up period [4]. disadvantages tend to include the comple xity of these procedures, constraining assumptions, and increased computing time . randomization test, welch’s regression-based method, n-ska rt, and automated simulat ion analysis procedure (asap) fall into this category. 2.4. initialization bias testing the goal of in itia lizat ion bias testing is to determine if bias is present in the data due to the initia l transient. the ma jority of these metho ds build upon the work of schruben [4]. the estimates of the mean and variance are used to compute a test statistic which is co mpared to an appropriate f distribution [ 5]. hypothesis testing is performed with the null hypothesis that no initia lization bias e xists. these procedures can also be used in union with previously described methods to determine if in itia lizat ion bias has been successfully removed. 2.5. hybrid methods hybrid methods are typically a co mbination of two methods, usually in itia lizat ion bias tes ting method and either a graphica l or heuristic method. these methods are typically co mple x and can require large amounts of data [4]. 3. results and discussion hoad et al. performed a comprehensive review on the e xisting methods of estimating the length of the warm-up period and found 42 methods for detecting the extent of the wa rm-up period [6]. these methods were evaluated and graded based on the following crite ria : accuracy and robustness of method, simp licity o f the method, ease of potential automat ion, generality, number o f para meters required, and co mputation time [7]. the list was then narrowed down to six methods for further evaluation. graphica l methods were e xc luded due to their need for human intervention and their subsequent inability to be automated. of the six methods, mser substantially outperformed the rest while the other methods either severely underestimated the truncation point or required an e xtre me ly large number of computational resources. proceedings of engineering and technology innovation , vol. 3, 2016, pp. 34 36 36 copyright © taeti as the research team looks to info rm transportation analysts as to how to mitigate init ia lization bias, the team investigated detailed analyses of three available truncation methods. the methods selected were we lch’s method, due its simp lic ity and use in practice, mser, for its effectiveness in identifying the truncation point and use in industry, and the volume method that is currently being use in a couple of traffic applications . it was found that welch’s method and mser method provide co mparable results for the truncation points for when the simulation model has reached steady-state. the results from imple menting these procedures indicate that 1) welch’s method would be easy to imple ment in practice and 2) mser method, which selects the truncation point by selecting the point that minimizes the width of the confidence interval about the truncated sample mean, provides co nsistent results with possibility to be fully aut omated. for the m ser method the truncation point can be determined based on each replicate run, while welch’s approach gives a single truncation point that is determined fro m and can be applied to all replications. the mser method has the potential to be a robust and useful tool since it can be included in large automated replication process without human interpretation. 4. conclusions the goal of this research is to explore d ifferent methods of mit igating in itia lizat ion bias for transportation applications. the in itia liza tion bias proble m has often been neglected in practice and when unaccounted for it can yie ld inaccurate, unreliable and less meaningfu l re sults. throughout the process of imp le menting initia lizat ion bias minimization procedures, several issues arose. overall, it is also important to have the analyst involved in the decision so that the decision can be made acco rding to the objective of the study and the application model. acknowledgement this work is supported by a grant nrf-201 4r1a 1a2054793 and transportation & logistics research progra m 15ctap-c097344 of ko rean government. references [1] d. ni, “a fra me work for new generation transportation simulation,” proc. ieee, winter simu lation conference, monterey, ca, 2006. [2] s. taylor, “analy zing methods of mit igating initia lization bias in transportation simulation models ,” master thesis, depart ment of civil and environ mental engineering, georgia institute of technology, atlanta, georgia, usa, 2010. [3] a. m. law and w. d. ke lton, “simu lation modeling and analysis ,” 4th ed. ne w york: mcgraw-hill se ries in industrial eng ineering and management science, 2007. [4] s. robinson, “a statistical process control approach to selecting a warm-up period for discrete-event s imu lation,” european journal of operational research, vol. 176, no. 1, pp. 332-346, jan. 2007. [5] j. white, k. preston, m. j. cobb, and s. c. spratt, “a comparison of five s teady-state truncation heuristics for simulation,” proc. winter simu lation conference, orlando, fl, 2000. [6] k. hoad, s. robinson, and r. davies , “automating warm-up length estimation,” proc. winter simulat ion confe rence, miami, fl, 2008. [7] n. m. steiger, “steady-state simulation analysis using asap3,” proc. w inter simulation conference, washington, dc, 2004.  proceedings of engineering and technology innovation , vol. 2, 2016, pp. xx xx 17 combustion synthesis of mullite/metal boride composites chun-liang yeh * , chi-chian chou, che-han kang department of aerospace and systems engineering, feng chia university, taichung, taiwan. received 14 march 2016; received in revised form 27 april 2016; accept ed 12 may 2016 abstract formation of co mposite materia ls co mposed of mullite (3al2o32sio2) and transition meta l borides (nbb2 and tab2) was studied by self-propagating high-temperature synthesis (shs). starting mate ria ls inc luded not only meta l o xides (nb2o5 and ta2o5) and boron oxide (b2o3) as the sources of meta llic e le ments and boron, but also al and si powders as the reducing agents. the evolution of mu llite fro m in situ formed sio2 and al2o3 and synthesis of nbb2 and ta b2 were investigated. the effect of e xcess si addit ion was studied on the co mbustion temperature, fla me -front propagation velocity, and phase composition of the fina l product. for formation of the nb b2/mu llite composites, the combustion velocity about 2.5 mm/s and reaction temperature a round 1500 o c decreased slightly as the si content increased . however, a considerable decrease in co mbustion front velocity fro m 2.74 to 1.43 mm/s and in reaction temperature fro m 1600 to 1250 o c was observed for the production of the tab2/mullite composites . the xrd patterns of the fina l products confirmed the role of e xcess si in the improve ment of silicothermic reduction of b2o3 and subsequent evolution of nbb2, ta b2, and mu llite. the eds analysis indicated an atomic proportion close to that of 3al2o32sio2 for the mullite grains synthesized in this study. keywor ds : combustion synthesis , mullite, transition metal borides , xrd, sem 1. introduction mullite co mposites have attracted great attention due to the broadness of their industrial applications. mullite with stoichiometries ranging fro m re latively silica-rich 3al2o32sio2 (3:2 mu llite) to alu mina-rich 2al2o3sio2 (2:1 mu llite) is the only stable crystalline phase in the alu minosilicate system [1]. fo r further imp roved fracture toughness, fle xu ra l strength, and therma l shock resistance, mullite-based composites containing zro2 and al2o3 we re e xtensively studied [2-3]. late ly, other cera mic addit ives such as tic and tib2 have gained increasing attention [4-5]. a variety of processing techniques with d iffe rent starting materials have been employed to prepare mu llite in e ither monolithic or co mposite forms, and they include the sol-gel method, reaction sintering, spark plasma sintering, solution combustion synthesis, self-propagating high-temperature synthesis (shs), and therma l e xp losion. the shs method takes advantage of highly exothermic reactions, and has the merits of low energy require ment, short processing time, simplic ity, high produ ctivity, and structural and functional diversity of final products [6-7]. in addit ion to the y ield of al2o3, high e xothermicity of the thermite reac tion is a great benefit fo r the shs process. therefore, the thermite-based shs technique has been utilized to prepare mullite co mbined with ceramic or intermetallic phases. this study aims to fabricate nb b2/mu llite and tab2/ mu llite co mposites by combustion synthesis in the shs mode. the shs process involves coreduction of metal o xides (nb2o5 and ta2o5) and b2o3 by al and si. due to the evaporation loss of si at elevated temperatures, the effect of e xcess si addit ion was studied on the combustion temperature, fla me -front velocity and phase composition. 2. method the starting materials of this study include nb2o5 (stre m che micals, < 45 m, 99.9%), ta2o5 (alfa aesar, < 45 m, 99.85% ), b2o3 (stre m che micals, 99.9% ), al (showa che mica l co., < 45 m, 99.9% ), and si (st re m che micals, < 45 m, 99% ). the reactant mixtures we re formulated as 13/ 11nb2o5 (or ta2o5) + 26/ 11b2o3 + 6al + (2x )si with x = 1.0–2.0 for studying the effect of e xcess si (x > 1.0) addition. *corresponding aut hor. email: clyeh@fcu.edu.t w proceedings of engineering and technology innovation , vol. 2, 2016, pp. 17 19 18 copyright © taeti sa mple co mpacts were prepared in a cylindrical shape with 7 mm in dia meter, 12 mm in length, and 50% in re lative density by compressing the well-mixed reactant powders. co mbustion experiments were performed under high-purity (99.99% ) a rgon of 0.15 mpa. the combustion propagation rate and reaction temperature were prec isely measured. deta ils of the e xperimental setup and technique were reported elsewhere [8]. 3. results and discussion a typical co mbustion sequence illustrated in fig. 1 is associated with formation of tab2/mu llite fro m a powder compact with x = 1.0. it is evident that upon ignition the combustion front was readily formed and propagated progressively along the sample, confirming sufficiency of the reaction exothermicity. fig. 1 a sequence of self-sustaining combustion wave for the format ion of the nb b2/mu llite composite, the combustion velocity of about 2.5 mm/s and react ion te mperature of a round 1500 ℃ decreased slightly as the si content e xceeded the stoichiometric a mount. this implies a re duction in the reaction e xothermic ity, because the endothermic mu llit ization process is enhanced. however, a considerable decrease in combustion front velocity fro m 2.74 to 1.43 mm/s and in reaction te mperature fro m 1600 to 1250 ℃ was observed for the production of the tab2/ mullite co mposites . this was attributed to the yield of tasi2 a lo wer e xothermic phase in comparison to tab2. in contrast, no silic ide phases existed in the nbb2/mullite composite. besides mullite, niobiu m borides of three phases, nbb, nb3b4, and nbb2, we re identified in the final product. the presence of nbb and nb3b4 as the minor phases means a defic iency of boron. this was probably caused by lack of sufficient si to fully carry out meta llothermic reduction of b2o3. for the reactant co mpact with e xcess si of x = 2.0, both nbb and nb 3b4 are considerably reduced. the xrd pattern of the product synthesized fro m an e xactly stoichio metric sample is shown in fig. 2(a). first of all, the evolution of mu llite fro m in situ formed sio2 and al2o3 was achieved and tab2 was identified as the do minant boride phase, confirming coreduction of ta2o5 and b2o3 by both al and si. however, the presence of free al2o3 means that the evaporation loss of si brings about a shortage of sio2. the lack of si correspondingly caused incomplete reduction of ta2o5 and b2o3. for the si-e xcess samples of x = 1.5 and 2.0, fig. 2(b) and (c) show almost no free al2o3 and a slight decrease in tab for the resulting products . this implies that the degree of silicothermic reduction of b2o3 was enhanced. however, e xcess si considerably increased the amount of tasi2, indicative of lack of boron for the formation of tab2. fig. 2 xrd patterns of tab2/mullite synthesized fro m samp les with si of (a ) x = 1.0, (b ) x = 1.5, and (c) x = 2.0 fig. 3 sem micrograph and eds spectrum of nbb2/ mullite synthesized fro m a samp le of x = 1.5 proceedings of engineering and technology innovation , vol. 2, 2016, pp. 17 19 19 copyright © taeti the typical microstructure of fracture surface of the nbb2/mu llite co mposite synthesized fro m a sa mple of with x = 1.5 is d isplayed in fig. 3. for those plate-like gra ins formed into a structure with rectangular opens, the eds analysis advises their atomic proportion of o:al:si = 63.09:28.05:8.86, wh ich matches closely with that of 3:2 mu llite. the other constituent nbb2 e xists as fine particles distributed over thin mullite plates. 4. conclusions preparation of nbb2/mullite and tab2/mullite was conducted by combustion synthesis in the shs mode involv ing metallothermic reduction of nb2o5, ta2o5, and b2o3. co mbustion of the sample adopting two reducing agents (al and si) was more e xothermic and speedy than that with pre-added sio2 and using al as the only reductant. it was found that the si content about 50 mo l% in e xcess of the stoichiometric quantity was required by the si-containing sa mples to facilitate the silicothermic reduction of b2o3 to the utmost extent. as a result, the format ion of nbb2 or ta b2 and mullite was improved. the mu llite gra ins synthesized in this study showed an atomic proportion close to that of 3al2o32sio2. in conclusion, the advantages of utilizing two reductants include more e xothe rmic co mbustion, higher self-sustainability, shorter reaction time, and better phase transformation. acknowledgement this research was sponsored by the ministry o f sc ience and technology of ta iwan under the grant of most 104-2221-e-035-057. references [1] h. schneider, j. schreuer, and b. hild mann, “structure and properties of mu llite -a re view,” journal of the euro. ce ra mic society, vol. 28, no. 2, pp. 329-344, 2008. [2] i. zake -t iluga, v. svin ka, r. sv inka, and l. grase, “therma l shock resistance of porous al2o3-mu llite cera mics,” ce ra mics international, vol. 41, no. 9, part a, pp. 11504-11509, november 2015. [3] n. m. rendtorff, s. gó me z, m. r. gauna, m. s. conconi, g. suare z, and e. f. aglietti, “dense mu llite-zirconia -zirconiu m t itanate cera mic composites obtained by reaction sintering,” cera mics international, vol. 42, no. 1, part b, pp. 1563-1572, january 2016. [4] d. ghahre mani, t. ebad zadeh, and a. maghsodipour, “densificat ion, mic ro structure and mechanical properties of mu llite-t ic co mposites prepared by spark plasma sintering,” cera mics international, vol. 41, no. 2, part a, pp. 1957-1962, march 2015. [5] z. i. za ki, “co mbustion synthesis of mullite-t itaniu m boride co mposite,” ce ra mics international, vol. 35, no. 2, pp. 673-678, march 2009. [6] a. g. merzhanov, “co mbustion and e xp losion processes in physical che mistry and technology of inorganic materia ls,” russian chem. rev., vol. 72, pp. 289-310, 2003. [7] k. morsi, “the dive rsity of co mbustion synthesis processing: a rev iew,” journal of materials sc ience, vol. 47, no. 1, pp. 68-92, january 2012. [8] c. l. yeh and j. z. lin, “co mbustion synthesis of cr-al and cr-si intermeta llics with al2o3 additions fro m cr2o3-al and cr2o3-al-si react ion systems,” interme tallics , vol. 33, pp. 126-133, february 2013.  proceedings of engineering and technology innovation, vol. 7, 2017, pp. 41 44 robust algorithms for regression analysis based on fuzzy objective functions tai-ning yang * , chih-jen lee, jenn-dong sun, chun-jung chen department of computer science and information engineering, chinese culture university, taipei, taiwan, roc. received 11 july 2017; received in revised form 17 august 2017; accept ed 26 august 2017 abstract in this paper, we address the issues related to the design of fuzzy robust linear regression algorithms. the design of robust linear regression analysis has been studied in the literature of statistics for over two decades. more recently various robust regression models have been proposed for processing noisy data. we proposed a new objective function by using fuzzy complement and derive improved algorithms that can produce good regression analysis from the spoiled data set. data set from the u.s. department of transportation is used to evaluate the performance of the regression algorithms . keywords : robust regression, fuzzy complement, linear regression analysis 1. introduction linear regression analysis is the study of linear relationships between variables. as a basic and 79 popular statistical technique, it has been widely used in various fields. since linear regression analysis algorithms have to process data from the real world, it should have the ability to cope with the outlier defined as the observation point that is distant from other observ ations. robustness theory is concerned about solving problems subject to model perturbation or outlier. according to huber [1], a robust algorithm not only performs well under the assumed model, but also produces a satisfactory result under the deviation of the assumed model. more recently, many researchers proposed various robust algorithms for regression analysis. kopsinis et al. [2] proposed a mechanism for iteratively detecting and excluding corrupted data. papageorgiou et al. [3] splited the noise into two compon ents: the inlier bounded noise and the outliers . they constructed a robust method in the framework of greedy algorithms. huang et al. [4] developed an effective convex approach that used recent advances on rank minimization and applied the method in computer vision applications. cheng et al. [5] introduced a robust adaptive loss function to measure the representation loss. nurunnabi et al. [6] used global polynomial functions and designed robust algorithms for extracting the ground points in laser scanning 3 -d point cloud data. unlike previous approaches, our proposed robust regression analysis is based on fuzzy objective functions. 2. traditional linear regression analysis regression analysis is a statistical process for estimating the dependent variable from one or more independent variables . the target dependent variable is formulated as a function of the independent variables called the regression function. when the function is linear, the process is called linear regression analysis. * corresponding author. e-m ail address:tny ang@ faculty .pccu.edu.tw proceedings of engineering and technology innovation, vol. 7, 2017, pp. 41 44 copyright © taeti 42 , i i i d x y  denotes the i-th data pair. 1 2( , ,.., ) k i i i i x x x x is the vector of k independent variables and iy is the target dependent variable. there are n data pairs. 1 2 0 1 2 ( ) .. k i i i k i f x w w x w x w x     is the linear estimation function that is the linear combination of input components. the weight 0 1( , ,.., )kw w w w is the coefficients vector for estimation.   2 ( ) ( ) i i i e x y f x  is the loss function. the objective function of traditional linear regression analysis for minimization is 1 n i i e   . the following is the online algorithm of linear regression derived by g radient descent approach. step 1. initially set the iteration count t , iteration bound t, learning coefficient 0 (0,1]  and the weight w. step 2. while t is less than t, do steps 3-7. step 3.compute 0 (1 / ) t t t   and set 1i  step 4. while i is less than n, do steps 5-6. step 5. update the weight: 0 0 ( ( )) new old t i i w w y f x   (1) 1 1 1 ( ( )) new old t i i i w w y f x x   (2) ( ( )) new old k k k t i i i w w y f x x   (3) step6. add 1 to i. step 7. add 1 to t. the above algorithm is known to fail when outliers exist. 3. robust linear regression analysis based on fuzzy objective functions for tackling the noise, we add a noise cluster in which the data has a constant influence . assume that there is an outlier cluster outside the data cluster. i u is the membership of i x in the data cluster, while the standard fuzzy complement of i u , 1( ) i u , is the membership of i x in the noise cluster. the fuzziness variable, m, determines the influence of small i u compared to large i u . following the fuzzy theory, we propose a robust linear regression objective function: 1 1 ( ) ( ) (1 ) , n n m m i i i i i rlg u e x u       (4) subject to i u in [0,1] and m in [1,  ). compute the gradient of rlg with respect to i u and get 1/ ( 1) 1 , when ( ) is minimized. 1 ( ) i mi ru l e x g     (5) proceedings of engineering and technology innovation, vol. 7, 2017, pp. 41 44 copyright © taeti 43 substituting this membership back and after simplification, we get 1 1/ ( 1 )1 1 ( ) ( ) ( ) 1 ( ) m n i mi i rlg e x e x       (6) following the multidimensional chain rule, the gradient of rlg with respect to w is: 1/( 1) ( ) ( )1 ( )( ) ( ) ( )( ) 1 ) ) ( (mi i mii e x e xrlg rlg e xw e x w w             (7) let ( ) i x denote 1/ ( 1) 1 ( ) ( ) 1 ( ) m mi e x    . m is called the fuzziness variable in the literature of fuzzy clustering. the following is the proposed algorithm. step 1. initially set the iteration count t, iteration bound t, learning coefficient 0 (0,1]  , soft threshold  and the weight w. step 2. while t is less than t, do steps 3-7. step 3.compute 0 (1 / ) t t t   , set 1i  . step 4. while i is less than n, do steps 5-6. step 5. update the weight: 0 0 ( ( ))( ) new old t i ii xw w y f x   (8) 1 1 1 ( ( ( ))) new old t ii i i xw w y f x x   (9) ( )( ( )) new old k t i i ii k k xw w y f x x   (10) step 6. add 1 to i. step 7. add 1 to t. 4. simulations fig. 1 results of traditional linear regression fig. 2 results of the proposed linear regression proceedings of engineering and technology innovation, vol. 7, 2017, pp. 41 44 copyright © taeti 44 to show the experimental differences between the traditional and the proposed, we use the data set from the u.s. department of transportation. the data pair is the population and fatal motor vehicle crashes per state in 2015. there are 51 pairs corresponding to 50 states and the district of columbia. for convenience, we scale down the data to be (population/107, crashes/103). in the following experiments, we set iteration bound t =1000, learning coefficient 0 0.3  and the fuzziness variable 3m  . the noisy data set is generated by adding 5000 crashes to the first 3 data. fig. 1 shows the traditional linear regression is greatly affected by the outliers while fig. 2 shows the proposed one is slightly affected by the outliers. as suggested by huber [1], the constant influence is set as the mean of the ( )ie x from the result of the traditional linear regression. the initial weight w in the robust approach is also set by the result of the traditional linear regression. 5. conclusions with consideration of outliers, we propose a fuzzy objective function for robust linear regression. the derived algorithm adapts the estimated component according to the c urrent membership of the input data. thus, the influence of outliers is alleviated. the results of a simple simulation comparing the traditional linear regression and the robust linear regression correspond to our expectations. acknowledgement the support of the ministry of science and technology, (taiwan), under grant most 105-2221-e-034-016 is gratefully acknowledged. references [1] p. j. huber and e. m. ronchetti, robust statistics, 2nd, new york: wiley, 2009. [2] y. kopsinis , s. chouvardas , and s. theodoridis , “iterative randomized robust linear regression,” international conference on acoustics, speech and signal processing, pp. 5436-5540, april 2015. [3] g. papageorgiou, p. bouboulis, and s. theodoridis , “robust linear regression analysis —a greedy approach,” ieee trans actions on signal processing, vol. 63, no. 15, pp. 3872-3887, may 2015. [4] d. huang, r. cabral, and f. de la torre, “robust regression,” ieee trans actions on pattern analysis and machine intelligence, vol. 38, no. 2, pp. 363-375, february 2016. [5] g. l. cheng, f. y. zhu, s. m. xiang, y. wang, and c. h. pan, “semisupervised hyperspectral image classification via discriminant analysis and robust regression,” ieee journal of selected topics in applied earth observations and remote sensing, vol. 9, no. 2, pp. 595-608, september 2016. [6] a. nurunnabi, g. west, and d. belton, “robust locally weighted regression techniques for ground surface points filtering in mobile laser scanning three dimensional point cloud data,” ieee trans action on geoscience and remote sensing, vol. 54, no. 4, pp. 2181-2193, november 2015.  proceedings of engineering and technology innovation, vol. 7, 2017, pp. 14 19 study of alloy springs with magnetorheological dampers for vibration isolator device dyi-cheng chen * , li-ren chen department of industrial education and technology, national changhua university of education, changhua 500, taiwan. received 16 july 2017; received in revised form 19 august 2017; accept ed 30 august 2017 abstract vibrat ion is a factor that must be controlled during the manufacturing process; variation in workpiece dime nsions often results in inaccuracies due to v ibration. this study adopted a spring and electro magnetic -repulsion and magnetorheological da mper that can absorb the energy of e xte rnal vibrat ions and deduced the influential vibrat ion factors. ansys was e mp loyed to determine the energy that could be absorbed by the vibration isolation device under mach ine vibrat ion, and the taguchi method of quality engineering was used to design the structure of the device (metal spring, wire dia meter, and materia l). the usabil ity of the product was e xa mined for applicat ion in computer nume rica l control and tradit ional machines. the considered parameters of the magnetorheological fluid were density, coefficient of elasticity, and poisson’s ratio. the results indicated that sprin g wire dia meter e xerted the strongest effect on the device’s performance and that the electrical current provided to the damper could be buffered . keywor ds : vibration isolation device, alloy spring s, magnetorheological dampers 1. introduction serving as a semiactive control element, a magnetorheological (mr) damper is a smart damping device that is controlled by a magnetic field [1]. an mr damper with semiactive control integrated with the controllability and flexibility of active dampin g can adjust its damping force in response to load force and structure and is characterized by passive stability and energy conservation. the dynamics of an mr damper under different input voltages [2] enable the input voltage to pass through a current driver, which generates an electric current. the electric current then passes through a coil to generate a magnetic field; adjusting the input voltage thus changes the magnetic field strength. kim et al. [3] proposed an innovative friction pendulum system and mr damper that utilize the isolator and auxiliary damper of a smart base isolation system. a fuzzy logic controller is generally used to regulate an mr damper because it has inherent robustness and its capability in processing nonlinear and uncertainty data. beijen et al. [4] demonstrated that a control scheme with two sensors was superior that with sensor fusion because cyclic shaping and greater stability are possible. van der sandea et al. [5] proposed that by changing the weighting filter, different controllers could be d esigned to emphasize comfort or controllability. 2. research methods 2.1. simulation analysis of the organization solidworks was used to design the vibration isolation device and mechanism shown in fig . 5. the overall design concept of the mechanism is displayed in figs . 1 and 2. during an earthquake, the springs absorb vibration. because earthquake movement can occur in various directions—for example, horizontal or vertical—this mechanism was designed to enable absorption regardless of the direction of movement. * corresponding author. email address: dcchen@cc.ncue.edu.tw proceedings of engineering and technology innovation, vol. 7, 2017, pp. 14 19 copyright © taeti 15 fig. 1 vibration isolation device 2.2. ansys ansys is a fin ite ele ment analysis software package; its processing operations can be divided into a preprocessor, postprocessor, and solver. the postprocessor module d isplays calculated results using graphical methods such as a color contour, gradient, vector, part icle trac k, three-dimensional section, transparent, and semit ransparent displays. ca lculated results can be displayed or output in the forms of charts or curves . 2.3. taguchi quality engineering the offline quality control of taguchi e mploys three -stage quality engineering design to achieve a robust product. in the first stage, which is system design, engineering personnel determine the ideal product design concepts according to their professional knowledge, e xperience, and mutual discussion. in the second stage, parameter design, re levant process parameters are selected and an orthogonal arrange ment is used to conduct an experiment, the purpose of the orthogonal arrangement is to reduce the sensitivity of the product toward interference by noise and thus reduce variation in product quality. the third st age, tolerance design, focuses on process operating conditions and analyzes causes of p roduct variation to determine the product’s tolerance range. the loss function can be used to calculate product cost after the tolerance design stage. loss functions are used in quality engineering to obtain a balance between cost and quality. this researc h project e mployed an orthogonal array (oa) to plan the e xpe riment. the response table and factor effect diagra m we re used to determine the effect of each factor on a ta rget value and then to select the most favorable parameter combination . 3. results of research and analysis 3.1. the setting of magnetorheological fluid prior to simu lation analysis, the para meters of the mr fluid (mrf) had to be set, including its density, coeffic ient of elasticity, and po isson’s ratio. the poisson’s ratio of a fluid is 0.5. in consideration of the changes to the mrf under di fferent magnetic currents, the poisson’s ratio of the mrf was set at that of rubber (i.e., 0.45). the density of magnetorheological fluids is known fro m the description of lo rd's magnetorheological da mper products. the elastic modu lus of the para mete rs set for the magnetorheological fluid is required to co mple x modulus g * , the modu lus of the modulus of the material show the relationship between the stress -strain ratio, the modulus of the complex modulus g * g * = g' + ig" (1) g ' is young's coeffic ient of storage modulus, and the viscoelastic materia l by the vibrat ion of the storage energy elastic part. g "for the loss modulus is the viscoelastic materia l by the vibration force when the part of the energy consumption. sun et al. [6] obtained e xperimental data to obtain a set of non -linear re lationship between the modulus and loss modulus of the magnetorheological material and the magnetic field : damper spring mrf proceedings of engineering and technology innovation, vol. 7, 2017, pp. 14 19 copyright © taeti 16 g' =3.11x10 -7 b 2 + 3.56 x10 -4 b + 5.78x10 -1 (2) g" =3.47x10 -9 b 2 + 3.85 x10 -6 b + 6.31 x10 -3 (3) b is the magnetic induction intensity, unit mt. diffe rent magnetic fie lds result in diffe rent magnetic induction intensities, as illustrated in fig. 15. the magnetic indu ction intensity was substituted into (1) to obtain the complex modulus g*, which is detailed in table 1. table 1 mrf-140cg magnetorheological damper parameters current m agnetic induction (mt) g'(m pa) g"(m pa) g * (m pa) 0 0 0.578 0.006 0.578+0.006i 0.5 350 0.74 0.008 0.74+0.008i 1 550 0.867 0.009 0.867+0.009i 3.2. the setting of magnetorheological fluid table 2 d isplays the settings of various materia l para meters. the force e xe rted above the shock absorber mechanis m during each simu lation was 5000 n; pressure was applied downward, and the total deformat ion and von mises stress were determined. table 2 material parameters table material al jis g 3522 si-cr alloy sus316 m rf-140cg density (g/cm 3 ) 2.7 7.85 7.86 8.02 3.64 young's modulus (m pa) 69000 205882 207000 186274 1.03 poisson's ratio 0.34 0.3 0.3 0.3 0.45 3.3. application of taguchi method to be optimized this study select four spring factors and one mrf factor for the vibration isolation device (table 3). tab le 4 lists the l27(3 5 ) oa co mprising the total deformat ion and von mises stress obtained through 27 simulat ions with various parameter values. table 3 earthquake device mechanism parameter factor a b c d e factor wire diameter (mm) sp ring center diameter (mm) sp ace(mm) sp ring material current(a) level 1 6 50 20 jis g 3522 0 level 2 8 60 25 si-cr alloy 0.5 level 3 10 70 30 sus316 1 table 4 l 27(35) proceedings of engineering and technology innovation, vol. 7, 2017, pp. 14 19 copyright © taeti 17 as shown in tables 5 and 6, the most favorable parameters obtained for four factors at three levels were a spring wire dia meter of 6 mm (a 1), spring center dia meter of 60 mm (b2), spring coil spacing of 25 mm (c2), spring material of chro me silicon (d2), and damper current of 0 a (e). regarding the degree of impo rtance to the vibration isolation mechanism, the parameters were in a descending order of spring wire dia meter > spring ma teria l > spring coil spacing > spring center diameter > damper current. fig. 3 plots the s/n (smaller) factor response of the total deformation. table 5 total deformation s / n ratio factor reaction level a b c d e 1 20.34 16.52 16.41 16.70 17.60 2 16.41 19.57 19.68 19.89 17.60 3 16.04 16.71 16.71 16.21 17.60 delta 4.30 3.05 3.27 3.68 0.00 rank 1 4 3 2 5 table 6 total deformation of the best parameters a1 b2 c2 d2 e wire diameter (mm) sp ring center diameter (mm) sp ace(mm) sp ring material current(a) 6 60 25 si-cr alloy 0 the most favorable obtained para meters were used to construct a model, and a nsys analysis was perfo rmed to obtain the following results: greatest total deformation of 0.169 mm (fig. 4 (a)) and greatest von mises stress of 97.12 m pa ( fig. 4 (b)). as shown in tables 7 and 8, the most favorable para meters obtained for the four factors at the three levels we re a spring wire dia meter of 10 mm (a3), spring center diameter of 50 mm (b1), spring coil spacing of 20 mm (c1), spring material of sus316 (d3), and damper current of 0 a (e). the para meters had the follo wing order o f importance in their effect on the vibration isolation mechanis m: spring center dia meter > spring wire d ia meter > spring coil spacing > spring materia l > da mper current. fig. 5 shows the s/n (smaller) factor response diagram of von mises stress. table 7 von mises s / n ratio factor reaction level a b c d e 1 -39.78 -39.65 -39.71 -41.87 -40.45 2 -41.90 -42.00 -39.76 -39.75 -40.45 3 -39.66 -39.69 -41.87 -39.72 -40.45 delta 2.24 2.35 2.16 2.15 0.00 rank 2 1 3 4 5 table 8 von mises of the best parameters a3 b1 c1 d3 e wire diameter (mm) sp ring center diameter (mm) sp ace(mm) sp ring material current(a) 10 50 20 sus316 0 the most favorable obtained para meters we re used to construct a model, and ansys analysis was perfo rmed to dete rmine the greatest total deformation of 0.166 m (fig. 6 (a)) and greatest von mises stress of 97.38 gpa (fig. 6(b)) . 4. conclusions in this study, the investigated mrf was treated simila r to a solid to configure the material para meters. during the simulation process, these parameters we re simp lified by overlooking viscosity. optimal values of total deformation and von mises stress were obtained through simulat ion analysis followed by taguchi analys is. quality characteristics were selected based on the smalle r the better situation to minimize product variation. the optimal para meter results obtained through simu lation analysis revealed that spring wire dia meter had the most crucia l e ffect on the performance of the vibrat ion isolation mechanism; increasing the spring wire dia meter would increase the seismic performance of the mechanism. proper selection of the spring center dia meter and spring coil spacing also enhances seismic perfo rmance. the results of this study can serve as a useful reference for relevant industry. proceedings of engineering and technology innovation, vol. 7, 2017, pp. 14 19 copyright © taeti 18 fig. 2 mrf-140cg magnetorheological damper magnetic field and magnetic induction change map fig. 3 total deformation s / n ratio calculation (a) total deformation (b) von mises stress fig. 4 total deformation model and von mises stress fig. 5 von mises stress s / n ratio calculation proceedings of engineering and technology innovation, vol. 7, 2017, pp. 14 19 copyright © taeti 19 (a) total deformation (b) von mises stress fig. 6 total deformation model and von mises stress references [1] s. narasimhan, s. nagarajaiah, and e. a. johnson, “smart base-isolated benchmark building part iv: phase ii sample controllers for nonlinear isolation systems,” structural control and health monitoring. vol. 15, no. 5, pp. 657-672, july 2008. [2] s. j. dyke, b. f. spencer jr., m. k. sain, and j. d. carlsonl, “modeling and control of magnetorheological dampers for seismic response reduction,” smart materials and structures, vol. 5, no. 5, pp. 565-575, 1996. [3] h. s. kim and p. n. roschke, “design of fu zzy logic controlle r fo r sma rt base isolation system using genetic algorith m,” engineering structures , vol. 28, no. 1, pp. 84-96, january 2006 [4] m. a. beijen, d. tjepkema, and j. van dijka, “two-sensor control in active vibration isolation using hard mounts ,” control engineering practice, vol. 26, pp. 82-90, may 2014. [5] t. p. j. van der sandea, b. l. j. gysenb, i. j. m. besselinka, j. j. h. paulides , e. a. lo monova , and h. nijmeijer, “robust control of an e lectro magnetic active suspension system: simulat ions and measureme nts,” mechatronics, vol. 23, no. 2, pp. 204-212, may 2013. [6] q. sun, j. x. zhou, and l. zhang, “an adaptive bea m model and dynamic characteristics of magnetorheological materia ls ,” journal of sound and vibration, vol. 261, no. 3, pp. 465-481, march 2003.  proceedings of engineering and technology innovation, vol. 5, 2017, pp. 25 30 application of serious game model on simulation training for decision makings of project management hung-ming chen * , kai-wei wei department of civil & construction engineering, national taiwan university of science & technology, taipei, taiwan, roc. received 23 july 2017; received in revised form 07 august 2017; accept ed 10 august 2017 abstract this study develops a novel serious game foster decision making skills in project management based on role playing. it provides features such as accident event handling, resource allocation, selections of subcontractors, procurement and stock management, adjustment of project scheduling, and performance evaluations to be exercised by user online. in doing so, users can strengthen their project skills by managing a virtual project. importantly, the proposed game provides users with virtual experiences to implement management and control tasks of a construction project, which is often difficult to offer in conventional training. the proposed system is an online game deve loped by using advanced web development tools with integrations of mysql for project data management and ms project for project schedule update on the server side. users are presented with a virtual project in dynamic scene image depicting a construction s ite, which changes as the game progresses. there are also images present various construction events. during game play, users can function as a project manager whom is responsible for making various decisions involving accident handling and resource alloca tion as the virtual construction project progresses. results of the decisions made are obtained by real-time simulation based on project data, then are visually and interactively replied to the users to understand the subsequent effects of their actions. in this training mode, decision-based simulation can provide realistic and reliable consequences of their decisions. the virtual reality feature of this computer game offers an economic and viable alternative to actual project management experiences. keywords: serious game, decision simulation, virtual training, project management 1. introduction an increasing number of computer game modes that integrate a diverse array of knowledge skills have been adopted to facilitate professional staff training. in terms of training professional staff in construction engineering, related projects are characterized by customization, large scale nature, long construction time, high cost, unforeseeable circumstances, and poten tial risk involving construction safety. in contrast with other industries, a construction project must draw from the accumulated experiences of its professional staff. however, accumulating relevant experiences is rather difficult owing to the unique characteristics of construction projects, such as the ir large scale nature and high cost. therefore, providing virtual work experiences through computer games that simulate a construction project setting is a viable alternative. virtual operations can mitigate the risks incurred by direct physical exercise and are not subject to external weather factors. moreover, users can be * corresponding author. e-m ail address: hungm ing@ m ail.ntust.edu.tw tel.: +886-2-27376585; fax: +886-2-27376606 proceedings of engineering and technology innovation, vol. 5, 2017, pp. 25 30 copyright © taeti 26 oriented on unfamiliar situations repetitively. in contrast with training in a conventional setting, computer games can save a substantial amount in project costs and waiting time by switching game scenes frequently and simulating different conditions, ultimately allowing trainees to accumulate relevant knowledge skills and experiences cost effectively. computer game applications in training can be divided into operational and role -playing modes. trainees use the operational mode to simulate various workplace operations, e.g., driving an excavator or operating large -scale machinery, with a particular emphasis on fostering the adaptability of trainees to certain tasks [1-6]. alternatively, trainees use the role-playing mode to manage a virtual company or shop from a managerial perspective and to devise responsive strategies based on information that the computer provides, ultimately allowing trainees to accumulate managerial experience [7-11]. conversely, conventional modes for training construction engineering managers include classroom instruction, orientation on business practices, and site visits; of classroom instruction is the most prevalent. as is generally assumed, students acquire p ractical experience in construction management courses, when they actually lack praxis -based opportunities for exercise. although some training curriculums provide students with opportunities to practice in an enterprise setting, student learning of constr uction work is insufficient owing to the short training period. conventional learning is limited in its ability to only assign stude nts a low-level task without actually participating in management tasks. this explains why students often struggle upon ente ring the industrial workplace, eventually learning only after a trial and error approach towards correct industrial practices and expe rience acquired over time. the training mode of a computer game allows the trainee to act as managerial staff involved in engineering projects. in addition to enhancing the simulation, security, and playfulness of training, integrating a computer game with construction engineering supplements training in addition to normal class instruction. at present, this computer game mod e can offer a simulative scene to trainees, allowing them to feel by themselves and to visualize site situation using pictures and words, which will no longer be subject to individual imagination. during simulation, the game can incorporate the proposed so lution derived from classroom case studies for analogous calculation and comparison with original estimates. as for security, the computer game can eliminate potential risks in an actual workplace training site and simulate accidents through a video clip, offering trainees with the emotions of an on -site incident without fear of injury. therefore, this work presents a novel computer simulation game to train construction project managers. the computer game contributes to the development of assistant training software based on decision-making management principles that combine simulation with engineering managerial knowledge. in this game, users serve as a construction site project manager responsibl e for a series of decisions, as presented in 3d and game formats. related tasks include arranging labor, material quantity and equipment for each project, controlling the cash flow and construction period, selecting a contractor, ensuring safety in construction site management, performing field inspection and qualit y control, as well as forecasting weather-related factors to achieve construction project goals. the virtual construction management practices taught in this computer game are an economical, safe, and interesting means of filling the gap between teacher-student interaction and simulation of actual construction management training in the field. by making decisions in a gaming environment, trainees can understand life -like situations found in actual cases through repeated feedback and practice according to performance evaluation. as is expected, this virtual construction game provides trainees with the unique opportunity to manage a virtual construction site. by applying concepts taught in virtual construction, trainees c an learn from experiences involving ins pection and verification, thus clarifying case studies presented in the classroom and strengthening their perceptiveness. additionally, via this virtual medium, trainees can understand the implications of damage caused by improper management practices, wit hout concern over actually losing money and compromising security. trainees can repeat industrial practices through simulation based on their training results. proceedings of engineering and technology innovation, vol. 5, 2017, pp. 25 30 copyright © taeti 27 2. content and scope of virtual construction the proposed game allows users to thoroughly understa nd decision making concepts in construction engineering. although the main training objects are the on -site managerial staff, game contents encompass tasks of the project manager, construction site owner, and contractor, they are not limited to specific ro les. moreover, contents are delivered by assuming that users can comprehend situations based on clear instruction from the 3d animations, although it is not expected to simulate th e actual situation completely and accurately. as the proposed game simulates all work tasks involved in making construction project management -related decisions and schedules, this work selects the contents for a course called project planning and control as the primary source for simulati on. initially, virtual construction scripts are sorted out from course-related contents involving construction management and administrative operations. simulated construction contents are divided into four parts, as listed in the following. based on the contents of each section, several parts are selected as the simulation script. those contents are described briefly as follows: (1) construction contracts and management: specifications based on construction contracts, bid package format, constriction claims and disputes resulting from contracts, as well as provisions of changing orders following commencement of the project. the contract content format is added to the game script in order to describe the task of virtual construction in the proposed game; (2) subcontract: construction bidding and the selection methods for a construction site owner to choose which company to cooperate with, including quality control methods to manage the subcontract after the project is initiated. the proposed game attempts to incorporate methods of selecting contractors and their effects into operating costs, quality, and time, as well as other effects into the game script; (3) construction procedures: confirmation of the contract before starting the construction, daily management after the project is initiated, and accepting prov isions after construction is completed. the proposed game attempts to incorporate management tasks conducted by the site manager into the game script in order to provide trainees with management training-related experience; and (4) cost estimation and control: quantity takeoffs in the training of construction operation, measures to control costs and scheduling, and measures to allocate resources efficiently. the proposed game attempts to incorporate factors aimed at balancing resource allocation and r controlling costs and scheduling into the game script to provide trainees with resource allocation-related skills. the simulated contents originate from course instruction. additionally, the game elements are embedded in the virtual construction of the script with reference to current operating modes and elements of a simulated business game in the current market. finally, the estimated game script is prepared as the design basis of the simulated training contents for the game af ter discussing details with experienced staff. fig. 1 shows the main contents of the game script of the proposed virtual construction. tasks in this work include drawing up a construction contract initially, selecting collaborating companies, allocating resour ces efficiently after the project is initiated, controlling costs and schedules, handling unforeseen emergencies such as those involving construction safety, and evaluating the performance of construction quality after completion. cumulatively, the abo ve tasks form the basis for establishing a game-based training mode for decision making in construction management. 3. analysis of system requirements once the game script for virtual training and presentation of the game contents and interactive modes are established, this work further analyzes operating conditions based on the above considerations to prepare the details of system development. the training mode has the following system functions. proceedings of engineering and technology innovation, vol. 5, 2017, pp. 25 30 copyright © taeti 28 fig. 1 contents of the proposed training mode 3.1. decision mak ing one of the key responsibilities of a construction manager is to make decisions in construction management. whether making the appropriate management decision or not is one of the determinant factors for the success or failure of project management, and it is one of the key competencies that needs to be acquired first. based on the script in this study, the management decision making given to trainees in the game includes two parts: decision making for emergencies and decision making for selecting the contractor. emergencies are generally divided into two parts: those caused by climate and those caused by human factors. the disaster triggered randomly by the climate system of the game causes delays or a shut -down, and thus users must make a decision immediately. with regard to accidents caused by human factors, many accidents questioning construction safety caused by the incompetence of the cooperating company or by the improper habits of the construction workers are triggered in order for users to prepare solutions and select contractors carefully. 3.2. resource allocation and management during construction, the manager must arrange the resources including labor, money, and machinery, for reacting to various uncertain factors. these are very important as they can affect the schedule of the whole construction. therefore, the virtual construction management presented in this study provides trainees with the function of adjusting resources used in construction. the number and type of resources initially planned for each work item are the reference basis. after the scheduling is completed, the actual resource number is divided by the reference number. the ratio is divided into several blocks for differ ent days, and the stepwise method is applied to reduce the days. the number of resources after adjustment is multiplied by unit price and days after changing. the obtained number will be the cost of the operation. 3.3. schedule and progress control due to uncertainties in the simulation, the system allows users to adjust the operating resources according to the encountered situation. the original schedule and progress control are then changed with the occurrence of events and related arrangement. moreover, based on the schedule and progress control, it is necessary to update the schedule and th e cost. therefore, the function of the dynamic treatment of schedule and progress control is required. proceedings of engineering and technology innovation, vol. 5, 2017, pp. 25 30 copyright © taeti 29 to attain the requirement of schedule and progress control, which is achieved by integrating ms project, each game data in the main program is saved in a database, and then it is converted to the specified format set by the ms project. all the kinds of operations in virtual construction can be arranged using this software. after finishing the operation arrangement, the updated scheduling information is stored back to the database and the main program for updating game status through reading the database. 3.4. evaluation of management performance in current project management practice, the evaluation of management performance as a management tool has been introduced. according to the resources to invest and the schedule, the performance index is calculated theoretically to help the manager control and determine in real time the effectiveness of the management strategy in use. using the evaluation index ca n analyze the reasons why the results are not effective. this is to discuss further the occurrence point of the reason and to c onsider and acquire the improved method and make adjustments . the game should provide the function of evaluating management perfo rmance in a virtual construction. based on performance evaluation function offered by the system, users can check, during the game training period or after, their management strategy for resource allocation, cost and time control, and decision making. the proper evaluation theories, including the score index and s curve, are selected to calculate real-time indexes and control curves based on all kinds of status data in the process of game such as loss amount caused by construction safety, types of accidents , and expenses. 4. system implementation the training system for the proposed game is developed based on the above functions. five functional bars allow users to control and summon all functions. the functions of each function menu are listed as follows: (1) main function toolbar: users can access the contract contents and select the collaborating company from a list of many contractors, adjust the number of resources in each construction, shorten or extend the duration, rearrange the operating time to control construction operations, and make verifications and adjustments by accessing the performance scores; (2) time status bar: this menu indicates the date in the game world; the operation is run based on it. users can accelerate or decelerate the game speed, as well as stop the game; (3) project status bar: game climate can affect operation efficiency and the construction period. an icon denotes the current weather situation. conditions of the worker, owner, finances, and duration are denoted by three lights, i.e. red, y ellow, and green lights, to remind users to control them in real time; (4) sub function toolbar for daily management : users can give management instructions, such as orient the trainees, using various management functions offered in this functional toolbar; (5) display area of virtual scene: the animations of construction operations and animations of triggered events are displayed here. users can switch for a closer view of scenes on -site by clicking the building under construction . 5. effectiveness evaluation fifty students enrolled in a course on project planning and control at national taiwan university of science and technology tested the effectiveness of the proposed game. user survey statistics reveal an agreement level of 4.32 out of 5 t hat this game benefits learning and that course contents are clearly understood. this game -based training is regarded to perform better in simulation, interest, interactivity, learning efficiency, and immersion than other training modes do. the leading f ive features selected are management decision making, 3d display of virtual construction site, tutorial hints, project scheduling, as well as resource allocation and management, respectively. proceedings of engineering and technology innovation, vol. 5, 2017, pp. 25 30 copyright © taeti 30 6. conclusions by applying role-playing functions and simulation games, the proposed game orients students on decision making in project management. for instance, in a virtual work site, users can act as site directors to conduct a series of daily duties . based on the understanding and analysis of managerial experience and construction processes accumulated from managing virtual projects, a virtual training system design is planned to fulfill the function demand. also, users do not need to be concerned with poor decisions that can incur property damage or budgetary loss. additionally, the proposed virtual game can make modifications according to the feedback from the performance evaluation index after training. this feature helps users to determine and compensate for management limitations and other details to prevent time and financial loss result ing from a lack of experience or wrong judgment in their future work. the proposed game provides a virtual presence that allows users to observe the incident on site, as well as interact with each other in the decision -making process, which is in contrast with conventional classroom teaching methods. moreover, the proposed game also ensures a better performance in terms of calculating costs and safety in practical training, which represents an advantage over practical training in enterprises, construction s ite visits, and other training modes. references [1] i. javier and d. abraham, “application of virtual reality technology for the improvement of safety in the steel erection process,” international conf. computing in civil engineering, june 2005, pp. 1-11. [2] h. dib, m. f. shiratuddin, and d. fletcher, “learning construction surveying in a virtual environment: a proposed approach,” proc. of 12th international conference in civil and building engineering & 2008 international conference on information technology in construction, october 2008. [3] c. m. li, i. c. yeh, s. f. chen, t. y. chiang, and l. c. lien, “virtual reality learning system for digital terrain model surveying practice,” asce journal of professional issues in engineering education and practice, vol. 134, no. 4, pp. 335-345, october 2008. [4] c. c. lu, s. c. kang, s. h. hsieh, and r. s. shiu, “improvement of a computer-based surveyor-training tool using a user-centered approach,” advanced engineering informatics, vol. 23, no. 1, pp. 81-92, january 2009. [5] h. dib and n. adamo-villani, “a virtual steel sculpture for structural engineering education: development and initial findings ,” journal of educational technology systems , vol. 44, no. 4, pp. 430-449, june 2016. [6] n. adamo-villani and h. dib, “a study of the effects of teaching avatars on students' learning of surveying mathematics ,” international journal of information and communication technology education , vol. 12, no. 2, pp. 1-13, april 2016. [7] i. d. tommelein, d. r. riley, and g. a. howell, “parade game: impact of work flow variability on trade performance,” asce journal of construction engineering and management, vol. 125, no. 5, pp. 304-310, september 1999. [8] k. nassar, “simulation gaming in construction: er, the equipment replacement game,” journal of construction education, vol. 7, no. 1, pp. 16-30, september 2002. [9] w. c. wang, r. j. dzeng, and y. h. lu, “integration of simulation-based cost model and multi-criteria evaluation model for bid price decisions ,” journal of computer-aided civil and infrastructure engineering, vol. 22, no. 3, pp. 223-235, february 2007. [10] a. sherif and h. mekkawi, “a computer-aided-learning tool for teaching construction engineering decision making: an excavation s tage educational game,” proc. of 12th international conference in civil and building engineering & 2008 international conference on information technology in construction, october 2008. [11] h. dib, n. adamo-villani, and j. yu, “computer animation for learning building construction management: a comparative study of first person versus third person view,” proc. of the 1st international conference on e-learning, e-education, and online training, pp. 76-84, september 2014.  proceedings of engineering and technology innovation, vol. 9, 2018, pp. 01 08 innovation for education on internet of things olivier bonnaud 1,2,* , laurent fesquet 3,4 1 microsensor and microelectronic department, university de rennes 1, rennes france 2 gip-cnfm, national coordination for education in microelectronics and nanotechnologies, 38016 grenoble, france 3 tima, grenoble-inp, 38016 grenoble, france 4 cime-nanotech, grenoble alpes université, 38016 grenoble, france received 06 march 2018; received in revised form 18 may 2018; accepted 17 july 2018 abstract the internet of things (iot) and related objects are becoming more prevalent around the world with exponential growth for the next fifteen years. this evolution implies innovation in many fields of technology, whose core is in microelectronics. indeed, iot deals with all societal applications such as health, the environment, transport, energy and communications. thus, connected objects involve many technological components: sensors and actuators, signal processing circuits, data transmission circuits and systems, energy recovery systems, which directly depend on the performance of microelectronics. to create new connected objects, innovation is the main driver. innovation results from the combination of a multidisciplinary approach, links between disciplines and the necessary know-how of engineers and technicians. this paper deals with the orientation of pedagogy towards these objectives through the development of dedicated and innovative platforms in microelectronics. these platforms are developed by the french national microelectronics education network (cnfm). after presenting the context of iot and the evolution of microelectronics technologies, this article highlights the main components of connected objects applied to many societal applications. each component of the objects requires specific microelectronic devices or circuits. innovation appears in the nature of platforms, the multidisciplinary approach of training, the permanent links between disciplines, and the adaptation to new educational tools, mainly online. the results of the training on innovative platforms are presented and discussed. keywords: innovation, internet of things, microelectronics, nanotechnologies, education 1. introduction the internet of things (iot) and related connected objects are becoming more prevalent around the world. about two hundred billion of these objects are expected to be in service in ten years, globally. this evolution implies innovation in many areas of technology, for which the core is in microelectronics. indeed, the iot concerns all societal applications such as health, environment, transport, energy, and communications [1-2]. connected objects involve many technological components: sensors and actuators, signal processing circuits, data transmission circuits and systems, and energy harvesting systems [3]. all these specialties are directly dependent of the microelectronics capabilities. to create new connected objects, innovation is the main engine [4-5]. this innovating behavior results of the combination of the multidisciplinary approach, the links between disciplines and the necessary know-how of engineers and technicians [5-6]. this paper deals with the orientation of pedagogy towards these objectives through the development of dedicated and innovative platforms in the broader field of * corresponding author. e-mail address: olivier.bonnaud@univ-rennes1.fr tel.: +33-6-08078736 proceedings of engineering and technology innovation, vol. 9, 2018, pp. 01 08 copyright © taeti 2 microelectronics. these platforms are developed by the french national network for education in microelectronics (cnfm) [7-8]. after presenting the context of iot and the evolution of microelectronics technologies, this article highlights the main components of connected objects applied to many societal applications. each element of the objects requires specific microelectronics devices or circuits. the innovation appears in the nature of the platforms, the multidisciplinary approach of the training, the permanent links between disciplines, and the adaptation to the new pedagogical tools, mainly online [6]. for the future technicians, engineers and phd, the challenge is high and thus included in the strategy of the cnfm. the results of the training on innovative platforms are presented and discussed. 2. the context of iot: connected objects and data centers the internet of things (iot) and related connected objects are becoming more prevalent around the world. internet of things is supposed to include all the information stored in the data centers and all the connected objects that are transmitting information through the internet for remote control, security, but also production in the industry 4.0 [9]. industry 4.0 represents the fourth industrial revolution in manufacturing and industry. industry 4.0 is the current industrial transformation with automation, data exchanges, cloud, cyber-physical systems, robots, big data, ai, iot and (semi-)autonomous industrial techniques to realize smart industry and manufacturing goals in the intersection of people, new technologies and innovation. this evolution will imply a huge development of connected objects and exchange of data through the internet (or equivalent network). taking into account this perspective, about two hundred billion of these connected objects are expected to be in service in ten years, globally [10], as shown in fig. 1. this estimation seems much lower than the future real situation if we take into account the coming of 5g protocol for which witching and sensing will correspond to 100 billion of permanent units and tracking and tragging should correspond to 1 thousand billion of temporary units per year in short future [11-12]. fig. 1 expected evolution of connected objects. the growth is exponential, similarly to moore’s law. in 2025, about one hundred billions of things should be connected [10]. this evolution implies innovation in many areas of technology, for which the core is the microelectronics. indeed, the iot concerns all societal applications such as health, environment, transport, energy, and communications. connected objects will be involved in these application domains. their internal architecture involves many technological components: sensors and actuators, signal processing circuits, data transmission circuits and systems, and energy harvesting systems. fig. 2 shows a simplified architecture of a connected object [10]. the analysis of the prospect of such a development leads to a huge problem of energy consumption which also exponentially increases with the introduction of new connected objects on the markets. the challenge for innovation will be to drastically reduce the consumption of basic electronic devices, to modify circuit architectures to reduce the instantaneous consumption of billions of transistors, to extend standby operation in order to minimize the power consumption of electronic proceedings of engineering and technology innovation, vol. 9, 2018, pp. 01 08 copyright © taeti 3 systems. in addition, storing and processing data in large data centers require very high power consumption. indeed, several predictions make aware of this problem of power. us researchers predict that electricity consumption will triple in the next five years, with an additional one billion people in developing countries [13]. the communications industry currently uses about 20% of the world's electricity, which hampers climate goals. the demand for energy-hungry data centers, which store digital data from billions of smartphones, tablets and devices connected to the internet, continues to grow. in less than 10 years, more than 50% of the world's electricity will be used for communications if the current technology does not change. fig. 2 simplified architecture of a connected object. it is mainly composed of sensors and/or actuators, signal processing, emission and transmission module, visualization, alarms and controls, and energy harvesting when it is possible [10]. 3. evolution of the microelectronics technologies fig. 3 evolution of the integration. the moore’s law is expected to reach limits. however, thanks to new concepts, the integration is always going-up and becoming “more than moore” with a similar exponential growth. (after m. swamithan et al. [17]). it is clear that the incredible growth of iot and connected objects is a result of the constant evolution of the associated hardware through the integration of circuits and systems. for more than sixty years, integration has grown exponentially and follows the heuristic law proposed by g. moore [14]. this increase in integration is mainly the result of the decrease in the size of the elementary semiconductor devices, mainly the transistors. the lateral size has gone from a few tens of microns to several nanometers in the most recent productions. because the size of the elementary devices becomes close to the size of the atoms, the lateral decay in a plane (planar technology) has reached its limits if we keep the same architecture of the devices. that is why since the early 2000s, the growth of integration has resulted from new concepts based on i) the introduction of the third dimension in elementary devices, such as finfet (fin field effect transistor [15]), ii) the stacking of many integrated circuits [16-17] iii) the development of heterogeneous technologies combining bulk silicon integration with thin film devices. the stacking leads to system-on-chip (soc) and system-in-package (sip) technologies that can gather in the same system several proceedings of engineering and technology innovation, vol. 9, 2018, pp. 01 08 copyright © taeti 4 functions and several dies. thus, the evolution of integration has found new growth with an exponential evolution similar to moore's law and called “more than moore”, as shown fig. 3. the system density reached today one million per square centimeter, each system containing thousands and thousands elementary devices. 4. new tools for education and their limitations in such a context of digitalization and connection of the society, innovation also focuses on education. new teaching tools have been created for over twenty years. at first, texts and images were loaded on private servers. in the second period, the educational tools were accessible on the internet with new possibilities, for example by introducing animations and applets allowing several simulations [18]. thanks to the huge development of data centers and internet communications, today these types of tools have improved considerably and their number has increased considerably. one can find many approaches to massive open online courses, mooc [19], and more specifically several courses dedicated to microelectronics [20]. these online courses can contain many modules and interactive components as shown in fig 4. this figure shows the classic content of these tools that are very practical for training students. these last can go further in the theoretical approach. fig. 4 content of a massive open online course. due to the improvement of the storage capacities, animations, video and simulations can be included in the applets of the moocs. (after o. bonnaud et al. [3]). note that a first limitation to the use of these tools is maintenance. after a few years, the software that supports these tools change and it is more and more difficult to ensure the updates of the online course. in addition, the creation of such tools is a very long process and the updating of the content can also be very difficult if the designers who have contributed to their setting up are no longer present in the structure. a second important limitation to these tools is the lack of experimental training that can provide the students with know-how. if it is possible to create specific classes devoted to the assimilation of knowledge, for example through problem-solving or flipped-classes [21], understanding and the acquisition of skills require experiential training and associate knowledge. however, in the context of engineering, such a policy is almost impossible to apply by a single academic institution. 5. know-how in the frame of mutualized platforms as mentioned above, the orientation of pedagogy towards these objectives requires the development of dedicated and innovative microelectronics platforms. but in microelectronics, the practice requires sophisticated equipment such as workstations capable of receiving complex and heavy software for cad, characterization benches capable of analyzing proceedings of engineering and technology innovation, vol. 9, 2018, pp. 01 08 copyright © taeti 5 components and devices at the nanoscale, cleanrooms providing equipment for manufacturing, components and circuits in conditions similar to foundries but with lower requirements. all of this equipment is very expensive, representing hundreds of millions of us dollars in total, and the only realistic solution is to share the investments and operations between several institutions. thus, twelve common centres were gradually created in the 1980s in france spread throughout the territory and mainly located in areas where activities carried out by industries and academic institutions were already developed. these platforms were created and financially supported, at least partially, by the national microelectronics education network (cnfm) with the aim of minimizing duplication. in fact, apart from the common basic know-how, the centres are specialized in terms of technologies (silicon integrated circuits, nanotechnologies, thin film technologies, organic technology, etc.), circuit nature (analogue, digital, hf, high power, displays, etc.), and fields of application (health, biology, automotive, etc.). fig. 5 shows the distribution on the territory of the twelve centres of the french national network, already detailed in a previous article [7]. in addition to the twelve academic partners, two industrial organizations are partners in this network [22-23]. these two industrial structures represent the majority of companies that are working in france in the field of electronics, microelectronics, and electrical engineering. their participation in the network is important to define the strategy to be adopted in order to best meet their needs. today, the twelve centres manage more than eighty platforms, including seven cleanrooms. the montpellier centre is in charge of the purchasing and the distribution of all microelectronics cad software licenses including cadence, mentor graphics, synopsis, silvaco, altera, xilinks, and many others. this centre is also in charge of the national platform for testing mixed integrated circuits. indeed, the centre of montpellier manages a national equipment allowing a test remote from all the other centres. 89 academic institutions and 60 laboratories are users of the software and this platform. but all these tools are in permanent evolution which justifies a permanent update of the software and the equipment. fig. 5 cnfm: french national network. the twelve centres are spread all over the french territory. (after o. bonnaud et al. [3]). the cad tools are managed by the “national service” insured by the cnfm centre of montpellier. the consequence is that the french network has an incentive policy towards the partners, leading them to create new platforms, new practices and new software every year. this policy was strengthened in 2012 thanks to a national program dedicated to innovative education launched by the office of the commissioner for large investments for the future [24]. the gip-cnfm responded to the call and successfully completed the finmina project for innovative training in microelectronics and nanotechnology [25]. as part of this project, some 60 platforms have been created or updated which have been included in the menu of microelectronics and nanotechnology centres. proceedings of engineering and technology innovation, vol. 9, 2018, pp. 01 08 copyright © taeti 6 6. innovative training on platforms the themes of the innovative platforms cover a broad spectrum. we focus on a selection of recent realizations that preferentially relate to connected objects and are open to initial training. all these subjects are intended to give students some know-how. fig. 6 shows several examples of end objects that have been designed, manufactured, and characterized by students on the platforms of different network centers. the top three images yield results on the design of a fully programmable gate array (fpga) circuit, a digital circuit, and a digital-to-analog converter. the four middle images show the design and electrical characterizations after the fabrication of an rf transmitter circuit, a sensor on a flexible substrate and a connected building energy controller. the three lower images show a three-dimensional sensor associated with its electronics in "plastronic" technology, and a connected drone. the most complex objects are designed and produced by groups of students in the frame of projects or internships. they are all included in the "innovative training", that is to say, they have been introduced gradually since 2011. fig. 6 realization by students of different circuits and systems that can be involved in the connected objects. they can involve different technologies, from silicon integrated circuits to flexible and thin film technologies. fig. 7 left number of users on french platforms. this number has increased in recent years and reaches 16,000 in 2017. right number of users of innovative practices, which has increased rapidly since the starting of finmina project. many other examples could be presented but this is not the purpose of this paper. information on the platforms and their use by students can be obtained on the network's website [7]. for a multidisciplinary purpose adapted to connected objects, practical works on sensors and actuators are also proposed to students [26]. they all give experience and bring their know-how to the students. in order to show the effectiveness of the network, we establish each year a survey on practical training. we have plotted the evolution of the number of users and of the number of hours in the network and the part of the innovative training in fig. 7 and 8. fig. 7 shows the growth of the number of students mainly in initial training. the innovative practice has proceedings of engineering and technology innovation, vol. 9, 2018, pp. 01 08 copyright © taeti 7 led to a significant increase in the number of students since the start of the finmina program. it is the result of the opening of the subjects proposed to other disciplines, more particularly to health, biology and mechanics. fig. 8 shows the number of hours per year. the total number of hours of initial training is almost constant while the innovative part has significantly increased (right). indeed, the volume of practical training per student has followed a very bad evolution for several years, the universities considering that the cost is high, they try to reduce the practical training in the courses. the average time spent on the platforms has therefore decreased. however, more students are users which maintains the total activity. fig. 8 left – evolution of the number of hours spent on the french platforms. right – the total number of initial training is almost constant while the innovative part has significantly increased since the beginning of the finmina project. 7. conclusion and discussion the know-how remains very important in science and engineering and more particularly in microelectronics which is at the heart of the internet of things, but the cost is higher and higher. the current strategy is to share technology platforms dedicated to practical training. this strategy was applied by the french network cnfm which created twelve inter-university centers. these centers host approximately 14,000 undergraduate students each year on more than 80 technology platforms. to maintain the level of know-how that students can acquire and meet the needs of society [27], platforms are continually updated as part of an innovative strategy. innovation appears in the nature of the platforms, the multidisciplinary approach of training and the adaptation to new pedagogical tools. this approach has proved its interest through the growing number of users. in the field of microelectronics, the danger comes from the permanent growth of investments and the new tendency of several universities and institutions to reduce the volume of practical training even if the online courses can’t make the same contribution to the acquisition of skills. practice is a necessary complement to theoretical knowledge. the national network policy needs to be strengthened to maintain the quality of training and skills of graduate students who can contribute to the development of the internet of things. acknowledgment the author wants 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. references [1] o. bonnaud and l. fesquet, “communicating and smart objects: multidisciplinary topics for the innovative education in microelectronics and its applications,” proc. of 15th international conf. information technology based higher education and training, ieee press, august 2015, pp. 1-5 [2] g. matheron, keynote, microelectronics evolution, european, microelectronics summit, paris, november 2014. proceedings of engineering and technology innovation, vol. 9, 2018, pp. 01 08 copyright © taeti 8 [3] o. bonnaud and l. fesquet, “mooc and the practice in electrical and information engineering: complementary approaches,” proc. of 16th international conference on information technology based higher education and training, ieee press, december 2016, pp. 1-4. [4] o. bonnaud and l. fesquet, “innovation in higher education: specificity of the microelectronics field,” proc. of 31st symposium on microelectronics technology and devices (sbmicro 2016), august 2016, pp. 119-122. [5] 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, 2016. [6] o. bonnaud and l. fesquet, towards multidisciplinarity for microelectronics education: a strategy of the french national network, proc. of 2015 ieee international conf. microelectronics systems education, ieee press, may 2015, pp. 1-4. [7] “cnfm: national coordination for education in microelectronics and nanotechnologies,” www.cnfm.fr. [8] o. bonnaud, p. gentil, a. bsiesy, s. retailleau, e. dufour-gergam, and j. m. dorkel, “gip-cnfm: a french education network moving from microelectronics to nanotechnologies,” proc. of ieee global engineering education conference, ieee press, april 2011, pp. 122-127. [9] “industry 4.0: the fourth industrial revolution guide to industrie 4.0,” https://www.i-scoop.eu/industry-4-0/, february 2018. [10] o. bonnaud, “new challenges for higher education in the microelectronics field,” proc. of 10th annual international conference of education, research and innovation 2017, november 2017, pp. 7666-7675. [11] g. fettweis, “on platform design for 5g signal processing implementation,” keynote of latin america symposium on circuits and systems, february 2018. [12] g. fettweis, “5g technology breaking grounds from thingbook to the tactile internet,” http://www.5gsummit.org/ docs/slides/gerhard-fettweis-5gsummit-siliconvalley-11162015.pdf, february 2018. [13] “tsunami of dat could consume one fifth of global electricity by 2025,” https://www.theguardian.com/environment/2017/dec/11/tsunami-of-data-could-consume-fifth-global-electricity-by-202 5, february 2018. [14] g. e. moore, “cramming more components onto integrated circuits,” electronics magazine, vol. 38, no. 8, pp. 114-117, april 1965. [15] “finfet,” https://www.computerhope.com/jargon/f/finfet.htm, april 2017. [16] t. simonite, “moore’s law is dead. now what? ,” https://www.technologyreview.com/s/601441/moores-law-is-dead-now-what/, may 13, 2016. [17] m. swaminathan and j. m. pettit, 3rd system integration workshop, 2011. [18] o. bonnaud et al., “microelectronics online course,” https://microelectronique.univ-rennes1.fr/index21.html, 2013. [19] a. fox, “from moocs to spocs,” communications of the acm, vol. 56, no. 12, pp. 38-40, december 2013. [20] l. stuchlikova, a. kosa, p. benko, and d. donoval, “massive open online courses in microelectronics education,” proc. 10th european workshop on microelectronics education, ieee press, may 2014, pp. 31-36. [21] 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, 2018. [22] “acsiel components and systems alliance for electronics industry,” http://www.acsiel.fr/en-gb/index.aspx. [23] “fieec. federation of electronics, electrics and communication industries,” http://www.fieec.fr. [24] “cgi: commissariat général aux investissements,” http://investissementsdavenir.bpifrance.fr/commissariat_general_a_l_ investissement. [25] finmina: formations innovantes en microélectronique et nanotechnologies (innovation education in microelectronics and nanotechnologies). idefi progam: excellence initiative for innovation in education. see website of cnfm, idefi project: anr-11-idfi-0017, 2011. [26] o. bonnaud, “new approach for sensors and connecting objects involving microelectronic multidisciplinarity for a wide spectrum of applications,” international journal of plasma environmental science & technology, vol. 10, no. 2, pp. 115-120, 2016 [27] o. bonnaud and l. fesquet, “innovative practice in the french microelectronics education targeting the industrial needs,” proc. of ieee microelectronics system conference, ieee press, may 2017, pp. 15-18.  proceedings of engineering and technology innovation, vol. 18, 2021, pp. 36-42 security enhancement in image watermarking using combined medium sub-band wavelet approach for copyright protection sridhar balakrishnan * department of electronics and communication engineering, mlr institute of technology, hyderabad, india received 23 february 2020; received in revised form 24 february 2021; accepted 25 february 2021 doi: https://doi.org/10.46604/peti.2021.5316 abstract watermarking in multimedia content has secured attractive approaches in research society since nowadays the transmission of digital content in wireless medium has been enormous. along with the distribution of digital content, it is important to claim ownership. a non-blind detection technique is proposed in this paper. first, an original image undergoes the wavelet decomposition such as ll1, lh1, hl1, and hh2. afterward, the medium level sub-band coefficients of the image are subject to the comfortable shares for merging. next, one of the shares from the lh1 band is merging with another one of the shares in hl1 band. in this stage, the copyright mark is fetched into the merged sub-band coefficients. finally, an inverse wavelet transform is applied to receive the watermarked image. to prove the authentication, the original image and the watermarked image undergo the same operation and acquire copyright information. experimental results achieve that the proposed approach can withstand various image processing attacks. keywords: copyright protection, image sharing, sub-band fusion, wavelet transform, watermarking 1. introduction digital data possess the characteristics of making multiple copies without any difference. in the fast-growing wireless communications, a huge number of digital information is transmitted every second. the third party can easily acquire and claim ownership of the information. consequently, the protection of licensed innovation is a critical thought for the present world [1-2]. watermarking in digital contents is used to protect the multimedia information [3-6]. the watermarking technology has developed rapidly in the present era, and has given rise to the exponential growth of research findings in the field of copyright marking over the years. in a digital image watermarking system, the data to be embedded is called the digital copyright image, and the image where the watermark image to be embedded is called the host image. digital image watermarking can be classified based on the domain in which the data fetching processes can be carried out. the two major domains in watermarking technology are spatial and transform domains. least significant bits (lsb) belongs to the spatial domain [7]; discrete fourier transform (dft) [8], discrete cosine transform (dct) [9], and discrete wavelet transform (dwt) belong to the transform domain [10]. watermarking in the spatial domain is represented by modifying spatial characteristics directly, such as pixel values. on the other hand, watermarking in the transform domain converts images to the frequency signal first and then modifies the coefficient of certain frequencies. the singular value decomposition (svd) is a well-known technique for factorizing a rectangular matrix, real or complex, which has been widely employed in the image processing application such as image compression, face recognition, watermarking and texture classification. robust copyright marking approach should withstand the identity data under different * corresponding author. e-mail address: sridharbece@gmail.com tel.: +91-9894173468 proceedings of engineering and technology innovation, vol. 18, 2021, pp. 36-42 image processing attacks, including filtering, compression, and cropping. robustness, capacity, and imperceptibility [11] are the three characteristics of watermarking system, but all watermarking approaches only satisfy two characteristics. the rest of the manuscript is formed as follows: related work is highlighted in section ii. the proposed approach and the experimental results are discussed in sections iii and iv. the conclusion is placed in section v. 2. related works the extensive literature gathered and related to image watermarking techniques is critically inspected and exhibited in this section. further, the summary of the literature review is also furnished at the end of the section. lin et al. [12] developed a marking system under dwt. this algorithm performs the dwt on the original image and fetches the watermark into the hl and lh sub-bands. the experimental results demonstrate that the developed technique yields better results than the previous methodologies in terms of the quality of the watermarked image, the recovered watermark with or without attack, and time efficiency. the experiment results reported in the paper also suggest that this method can achieve ownership protection. a novel technique was invented for robust wavelet-based watermarking [13]. this approach addresses the way to fetch signature data in wavelet transform coefficients at the selected group, varying the watermark strength according to the range of sub-band information so that the group can be enabled wherever the respective coefficients reside. at first, the original image can be decomposed into four levels by dwt. next, the edges of the sub-band can be identified by the sobel edge detector. the pipeline of the copyright is carried out through the sub-band coefficients that lie on edges, and in that zone the level of the disturbance is less noticeable, with a sub-band level-dependent strength. in addition to that, there are only a few coefficients ready to embed the watermark around the corners, utilizing a distinct strengthening factor captured by a morphological dilation operation. mohammed and sidqi [14] clarified a multi-band wavelet-based image watermarking approach. right now, a versatile strategy is created depending on the selectivity parameter of band coefficients for hiding the watermark. the benefits of this methodology are using a large number of wavelet groups during the inserting procedure. surekha & sumathi [15] examined the copyright protection by using an image-based discrete wavelet transform through genetic algorithm. the scaling variable of the copyright is the most vital parameter that enhances the visual straightforwardness and strength against attacks. abdallah et al. [16] proposed a blind watermarking technique under wavelet transform, by which an identity mark is fetched into the coarsest scale wavelet coefficients. the image undergoes the three-level decomposition and the copyright mark is installed into the coarse scale. sood and goyal [17] proposed the genetic algorithm (ga)-bacterial foraging algorithm (bfo) and dwt for watermarking. in this method, the five-level dwt is utilized in the spatial transform. yahya et al. [18] developed an algorithm based on discrete wavelet transform to embed a watermark in the selected coefficient blocks only. for the extraction of copyright, a probabilistic neural network is employed. the experimental result demonstrates that the proposed approach provides copyrighted images with a high degree of imperceptibility and robustness to common attacks. from the above reviews, it is clear that we cannot install much data into the host image under frequency domain because the quality of the original image may be degraded significantly. spatial domain provides successful recovery of watermark at receiver end, but the robustness is very limited. hence the identification of marking a region of copyright data plays a vital role, and our proposed approach gives the direction to implant a watermark into the merged sub-bands coefficients under wavelet. the remaining section gives the proposed approach, and follows with experimental results and the conclusion. 3. proposed approach the watermarking technique under wavelet is used positively for concealing the copyright information to achieve imperceptibility in digital content. excellent spatial localization and multi-resolution are remarkable characteristics of 37 proceedings of engineering and technology innovation, vol. 18, 2021, pp. 36-42 wavelets. in our proposed approach, daubechies wavelets are enforced for the decomposition of the original image because the wavelets are orthonormal and compactly supported scaling and wavelet functions. initially original gray image [𝑌]𝑚𝑥𝑛 is subject to the wavelet transformation such as ll1, lh1, hl1 and hh1. fig. 1 shows the procedure of proposed approach. fig. 1 work flow of proposed approach 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 dwt ll ll ll ll lh lh lh lh ll ll ll ll lh lh lh lh ll ll ll ll lh lh lh lh ll ll ll ll lh lh lh lh hl hl hl hl hh hh hh hh hl hl hl hl hh hh hh hh hl hl hl hl hh hh hh hh hl hl hl hl hh hh hh hh y                           (1) rxc 1,1 1,2 1,n 2,1 2,2 2,n m,nm,1 m,2 w w w . . w w w w . . . . . . w w . w                  (2) 1 1,1 1 1,2 1 1,n 1 1 1,n 1 2 ,1 1 2 ,2 1 2 ,n 1 1 2 ,n 1 m 1,1 1 m 1,2 1 m 1,n 1 1 m 1,1 1 m ,1 1 m ,2 1 m ,n 1 1 m ,n lh w hl w . . . . lh w hl w hl w lh w . . . . hl w lh w . . . . . . . . . . . . . lh w hl w lh w hl w hl w lh w hl w lh w x                                                                    (3) original image ∑ dwt idwt watermarked image combined sub-band (lh1 1 +hl1 1 ) relocate the sub-bands into original position share the medium level coefficients (lh1 1 , lh1 2 , hl1 1 , hl1 2 ) α*watermark image 38 proceedings of engineering and technology innovation, vol. 18, 2021, pp. 36-42 w w 1 1 1 1 1 1 1 1 w w 1 1 1 1 1 1 1 1 w w 1 1 1 1 1 1 1 1 w w 1 1 1 1 1 1 1 1 w w 1 1 1 1 1 1 1 1 w w 1 1 1 1 1 1 1 1 w w 1 1 1 1 1 1 1 1 w w 1 1 1 1 ll ll ll ll lh lh lh lh ll ll ll ll lh lh lh lh ll ll ll ll lh lh lh lh ll ll ll ll lh lh lh lh hl hl hl hl hh hh hh hh hl hl hl hl hh hh hh hh hl hl hl hl hh hh hh hh hl hl hl hl hh r  1 1 1 1 hh hh hh                           (4) w 1 1 w 1 1 widwtll lh hl hh y       (5) [ ] w r rw mxn w r r mxn dwt ll lh y hl hh       (6) ( ) ( ) 1 1 1 1 r r 1 1lh hl lh hlw      (7) secret image sharing has received considerable attention in recent years. secure assurance of images is taken as a primary concern in military or commercial based applications. in our proposed framework, the original image is grouped into alternative pixels shares. next, the sub-band coefficients are merged, and finally, normal marked structures can be obtained by combining the allotments into a single image. what follows is the proposed algorithm: algorithm 1: image sharing begin: read the rows do read the columns/2 do if mod (i,2) ≠ 0 first share (odd, even) else first share (even, odd) endif end end second share=original image-first share end 4. experimental results this combined sub-band based marking approach is executed using matlab. for installing the identity information, we utilize a cameraman image of size 256×256 and the copyright image of size 32×32. the original image, the combined sub-bands image, the watermark image, and the watermarked image are displayed in fig. 2. (a) original image (b) dwt of an image fig. 2 experimental results of proposed approach 39 proceedings of engineering and technology innovation, vol. 18, 2021, pp. 36-42 (e) watermarked image (f) extracted watermark fig. 2 experimental results of proposed approach (continued) 4.1 invisibility test the validation of the method is concentrated on the invisibility of the embedding data it generates. to estimate the imperceptibility, mean square error (mse) and peak signal to noise ratio (psnr) are the basic parameters. the difference between the original image o (i, j) and the watermarked image w (i, j) is mse. the average mse of this proposed scheme is 0.0012. psnr is used to justify the degree of excellence of the watermarked image. the psnr of our proposed approach is 61.7259 db. the formulas of mse and psnr are expressed as ( ( ) ( )) 2 i , j 1 mse o i, j w i, j xy   (8) 2 10 255 psnr 10log mse         (9) the correlation co-efficient is another parameter to find out the robustness of the techniques against the image processing attacks. 2 2 ( ) ( ) ( ) ( ) xy xy x y xy xy x y x y o o e e r o o e e                    (10) the correlation coefficient value may be one or zero, depending on whether the watermarked image and the original image are identical or not. eq.10 shows the formula to generate the correlation coefficient between two images. 4.2 attacks in image watermarking the most common attacks of image watermarking are noise attacks, geometrical attacks, and filtering attacks. in our method, different noises (e.g. salt & pepper, speckle, gaussian, and poisson with default noise density) are introduced in a watermarked image with default noise density and the robustness of the system is measured. also, some additional evaluation of median filtering and low pass filtering attack on the watermarked image and measure the performance of the algorithm. table 1 demonstrates the psnr values between dwt and svd. table 2 shows the performance of an algorithm with different attacks. (c) combined share of medium level sub-band image (d) watermark image 40 proceedings of engineering and technology innovation, vol. 18, 2021, pp. 36-42 table 1 psnr values of the proposed approach with different bands table 2 performance of the proposed approach attacks watermarked image watermark image psnr (db) psnr (db) no attacks 61.7259 43.9923 salt & pepper 56.3432 36.1278 speckle 53.4509 36.2309 gaussian 45.7534 35.3267 poisson 37.0921 34.7877 median filtering 38.7204 37.9241 low pass filtering 39.4012 39.3809 fig. 3 performance sketches of dwt and svd the above fig. 3 represents the performance curve between the dwt and svd. in this noted that dwt based watermarking systems hold better psnr value than the svd. also, this proposed approach has secured a higher psnr value than 37.676 db (the value obtained by tanwar and khurana [19]). 5. conclusions digital image watermarking innovation has been drawing the attention of researchers and experts as another technique for ensuring copyright for digital images. it is acknowledged by installing a make-able data which is imperceptible to the human visual systems, and it is called a watermark. henceforth watermarking in images is the process in which a discrete data embedded within an image, imposing unperceived modifications on the image. in this paper, the sub-band sharing based methodology is exhibited. here, the copyright information is covered into the combined wavelet coefficients; henceforth the level of robustness is increased. in this approach, psnr and correlation coefficient is 61.7259 db and 0.9985 respectively. the defined methodology acquires less complexity and high efficiency. 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1 45 50 55 60 65 70 scaling factor p s n r ( d b ) dwt svd scaling factor dwt svd psnr(db) psnr(db) 0.01 74.8455 68.8240 0.02 68.8249 62.8034 0.03 65.3031 59.2815 0.04 62.8043 56.7828 0.05 60.8661 54.8446 0.06 59.2825 53.2609 0.07 57.9435 51.9220 0.08 56.7837 50.7622 0.09 55.7607 49.7391 0.1 54.8455 48.8240 41 proceedings of engineering and technology innovation, vol. 18, 2021, pp. 36-42 conflicts of interest author declares no conflict of interest. references [1] q. li and i. j. cox, “using perceptual models to improve fidelity and provide resistance to valumetric scaling for quantization index modulation watermarking,” ieee transaction on information forensics and security, vol. 2, no. 2, pp. 127-139, june 2007. [2] a. piva, t. bianchi, and a. de rosa, “secure client-side st-dm watermark embedding,” ieee transactions on information forensics and security, vol. 5, no. 1, pp. 13-26, march 2010. [3] a. piva, f. bartolini, and m. barni, “managing copyright in open networks,” ieee internet computing, vol. 6, no. 3, pp. 18-26, may-june 2002. [4] b. sridhar, “a wavelet based watermarking in video using layer fusion technique,” pattern recognition and image analysis, vol. 28, no. 3, pp. 537-545, july 2018. [5] v. m. potdar, s. han, and e. chang, “a survey of digital image watermarking techniques,” 3rd ieee international conference on industrial informatics (indin), august 2005, pp. 709-713. [6] y. wang, j. f. doherty, and r. e. van dyck, “a wavelet based watermarking algorithm for ownership verification of digital image,” ieee transactions on image processing, vol. 11, no. 2, pp. 77-88, february 2002. [7] s. dey, a. abraham, and s. sanyal, “an lsb data hiding technique using natural number decomposition,” third international conference on intelligent information hiding and multimedia signal processing, november 2007, pp. 473-476. [8] g. ren, c. zhang, and x. yang, “blind mesh watermarking based on the featured points in the frequency domain,” second international workshop on knowledge discovery and data mining, january 2009, pp. 701-704. [9] n. hubballi and d. p. kanyakumari, “novel dct based watermarking scheme for digital image,” international journal of recent trends in engineering, vol. 1, no. 1, pp. 430-433, 2009. [10] b. sridhar, “cross-layered embedding of watermark on image for high authentication,” pattern recognition and image analysis, vol. 29, no. 1, pp. 194-199, january 2019. [11] b. sridhar, “a wavelet based watermarking approach in concatenated square block image for high security,” journal of automation, mobile robotics and intelligent systems, vol. 12, no. 3, pp. 68-72, december 2018. [12] h. j. lin, c. w. lu, and c. m. chiang, “dwt-based watermarking technique associated with embedding rule,” 10th wseas international conference on signal processing, computational geometry and artificial vision, august 2010, pp. 23-28. [13] j. n. ellinas, “a robust wavelet-based watermarking algorithm using edge detection,” world academy of science, engineering and technology, vol. 1, no. 10, january 2007. [14] a. a. mohammed and h. m. sidqi, “robust image watermarking scheme based on wavelet technique,” international journal of computer science and security (ijcss), vol. 5, no. 4, pp. 394-404, september-october 2011. [15] p. surekha and s. sumathi, “implementation of genetic algorithm for a dwt based image watermarking scheme,” ictact journal on soft computing: special issue on fuzzy in industrial and process automation, vol. 2, no. 1, pp. 244-252, july 2011. [16] h. a. abdallah, m. m. hadhoud, a. a. shaalan, and f. e. abd el-samie, “blind wavelet-based image watermarking,” international journal of signal processing, image processing and pattern recognition, vol. 4, no. 1, pp. 15-28, 2011. [17] s. sood and a. goyal, “watermarking relational databases using genetic algorithm and bacterial foraging algorithm,” international journal of information and computation technology, vol. 4, no. 17, pp. 1877-1884, 2014. [18] a. n. yahya, h. a. jalab, a. wahid, and r. m. noor, “robust watermarking algorithm for digital images using discrete wavelet and probabilistic neural network,” journal of king saud university—computer and information sciences, vol. 27, no. 4, pp. 393-401, october 2015. [19] p. tanwar and m. khurana, “improved psnr and nc in digital image watermarking using rdwt and svd,” international journal of advanced research in computer science and software engineering research, vol. 6, no. 5, pp. 955-959, may 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/). 42  proceedings of engineering and technology innovation, vol. 8, 2018, pp. 15 -22 a direct lyapunov-backstepping approach for stabilizing gantry systems with flexible cable t. l. nguyen 1,2,* , m. d. duong 1 , and t. h. do 1 1 school of electrical engineering, hanoi university of science and technology , vietnam. 2 institute for control engineering and automation, hanoi university of science and technology , vietnam. received 11 january 2018; received in revised form 10 february 2018; accept ed 06 march 2018 abstract trolley positioning and payload swinging control problem of a fle xible cable gantry crane system are addressed in this paper. the system’s equations of motion that couple the crane’s cable and actuators dynamics are derived via e xtended ha milton’s principle. the control signal is designed based on the lyapunov direct meth od to derive control force and backstepping technique is e mployed to determine input signal for the actuator. the stability of the c losed loop system is proven analytically. nu merical simulat ions are included to demonstrate the effectiveness and robustness of the closed-loop system. keywor ds: flexible systems, overhead crane, field oriented control, lyapunov direct method 1. introduction nowadays, gantry crane systems are widely used in industrial and logistic applicat ions because of their fle xibility in load handling. however, swinging payload phenomenon causes slowing down goods handling operations and can be a potential threat to human and surrounding devices. certain types of payload s can ignite multi-modes or double-lin k pendulum effects [1-4]. in addition, characterized as a class of under-actuated systems, precisely controlling trolley position and suppressing payload vibration simultaneously pose many challenges for control engineers. in order to overco me the a fore mentioned control proble m, various approaches are considered. a conventional robust linear control la w is proposed to control the overhead crane [5]. since crane system is a nonlinear coupling system, instead of linear control, many researchers focus on nonlinear control approaches. a decoupling control law is proposed to asymptotically stabilize tro lley position and swing angle of the payload [6]. however, the designed control only guarantees bounded swing angle. an improve ment [7] is made with varying crane rope length that is meaningful in practice is considered. a switching control action is derived based on feedback linearization technique. position control and vibration suppression o f gantry crane with coupling e ffect between trolley and payload mot ions are ta ken int o account [8]. however, the obtained results are relatively limited in practice because of system’s parameters and actuator’s dynamics variat ions. in order to dea l with system uncertainty, an adaptive mechanis m is integrated in proposed control la w suggested [9]. well-known with its robustness against system uncertainties and disturbances, sliding mode control is applied in gantry control [10]. ho wever, it is need to cooperate with a pre-shape input to gain better performances [11]. severa l adaptive schemes [12-14] fo r gantry control also presented. intelligent control schemes are also considered to control the crane system such as fuzzy control [15] or neural network [16]. instead of feedback controls, some other researches consider feed -forwa rd control approaches where control actions from operators are modified befo re sending to the gantry actuators as shown [17-18]. the advantage of pre-shape input technique * corresponding author. e-mail address: lam.nguyentung@hust.edu.vn proceedings of engineering and technology innovation, vol. 8, 2018, pp. 15 -22 copyright © taeti 16 over feedback control is that measure ment of system states is not required but a fu ll knowledge of the system must be availab le. to rectify this drawback, pre-shape input method can be hybridized with a robust control as indicated [19-20]. the limitation of afore mentioned studies is an assumption of pendulum mot io ns for the payload. the assumption results in a system of ordinary differentia l equations govern system motions. however, practica l applications have shown that it is not the case, and gantry cable actually considered as a fle xible system whose motions are modeled as a system of pa rtial diffe rential equations. boundary controllers for stabilizing the fle xib le rope crane system based on lyapunov’s direct approach are developed [21-23]. the fle xib le rope where coupled longitudinal-t ransverse, transverse-transverse motions and 3d model are investigated [24-27]. dyna mics of the rope without model truncation is investigated, however the dynamics of the actuator are totally ignored. the ignorance of the actuator dynamics might to system instability . this paper directly designs a gantry control in consideration with fle xib le rope but in other direction. we con struct a distributed model of the overhead crane in which the mass and the fle xible of payload suspending cable are fully taken into account. the analytical mechanics inc luding ha milton ’s principle is used to construct the crane model. based on the obtaine d model, the controller is then designed systematically with the help of backstepping control. the rest of the paper is organized as fo llo ws. the mathe matica l mode l of fle xible overhead crane system is presented in section 2. in section 3, control design is developed based on the lyapunov direct method and backstepping technique. the numerical simu lation is performed in section 4 to show the efficiency of the proposed control design. conclusions and further studies are presented in section 5. 2. mathematical model an overhead crane system is illustrated in fig. 1, where fig. 1 a gantry crane system ( , )y x t and dy are the transverse motions of the crane’s cable and the target position of the payload, respectively. 0p is the cable tension, is the mass per unit length of the cable , and l is the length of the cable. m and m are tro lley’s and payload’s masses, respectively. the force (t)f is generally generated by an induction motor. the kinetic energy of the system is given as 2 2 2 0 0 ( ) ( ) ( ) 2 2 2 l y y lm y m t dx t t t (1) proceedings of engineering and technology innovation, vol. 8, 2018, pp. 15 -22 copyright © taeti 17 in eq. (1) and fro m now onward the argu ment ( , )x t is o mitted for neat representation. in addit ion , y(0) and y(l) are used to denote y(0,t) and y(l,t), respectively. the potential energy can be expressed as 2 0 0 ( ) 2 lp y p dx x (2) where 0 p is the tension of the cable. the work done by the external control force is given as w ( ) ( )f t y l (3) re mark 1. bending stiffness of the cable is considerably s mall so that potential energy due to bending stiffness can be ignored. the cable is assumed to be inextensible, and the cable deforms in oxy plane. the extended hamilton’s principle is expressed as follows 2 1 ( w) t t t p dx (4) substituting eqs . (1), (2) and (3) into eq. (9) results in 2 1 2 2 2 2 0 02 2 2 0 (0) (0) (0) (0) ( ) (0) 0 0 t l t t ly y y y y p dx p m y dt m y f t y dt xt x t t (5) using integration by parts, the equations of motions and boundary conditions of the crane system can be given as 0 0 tt xx y p y (6) 0 x (0) (0) ( ) tt my p y f t (7) tt 0 x ( ) ( ) 0my l p y l (8) the force acting on the trolley (t)f can be calculated in term of the motor torque as (t) m b i f m r (9) where b r is the radius of the dru m, i is the transmission ratio of the gearbo x and is the e ffic iency of the t ransmission system. mathematical model of the asynchronous motor can be written as follows : 1 1 1 1 1 sd sd s sq rd rd sd r r s i i i u t t l (10) 1 1 1 1 1 sd s sd sd rd rd sq s r r s i i i u t t t l (11) 1 1 rd sd rd s rd r r i t t (12) 1 1 rq sq s rd rq r r i t t (13) 3 2 m m p rd sq m k z i (14) where sdi and sqi are d irect and quadrature co mponents of stator current. 𝑇𝑟 and 𝑇𝑠 are rotor and stator time constants, pz is number of pole pairs, and is total magnetic leakage factor. 𝐾𝑚 = 𝐿𝑚 𝐿𝑟 , where 𝐿𝑚 and 𝐿𝑟 are mutual and rotor inductance. proceedings of engineering and technology innovation, vol. 8, 2018, pp. 15 -22 copyright © taeti 18 rd rd m l and , rq rq m l where rd and rq are dq components of the rotor flu x. the coupled electrical -mechanica l sys tem is rewritten as follow: 0 0 tt xx y p y (15) 0 x (0) p y (0) tt sq my i (16) 1 2 3 4sq sd sq sq i i i u (17) 0 x (l) p y (l) 0 tt my (18) where 1 2 3 1 1 1 , , s rd s r t t (19) and 4 1 3 , 2 m p rd s b i k z l r (20) re mark 2. eqs. (15)-(18) is derived under a condition that the rotor flu x o rientation is obtained, i.e., 0. rq moreover, it is assumed that sd i and rd are kept constants by current and flu x controllers and their values are available for feedback. in addition, the current controller has the ability of decoupling 𝑖 𝑠𝑑 and 𝑖 𝑠𝑞 . 3. control design the control objective is to simu ltaneously stabilize the t rolley and the payload at the desired position. an investigation of the system g iven in eqs. (15)-(18) shows that the system is of strict-feedback form. hence, in this paper, backstepping technique will be e mployed to design the control input.𝑢𝑠𝑞 the choice of back-stepping as a design tools make it ready if system para meters adaptation is needed. the control design process comprises of two steps. in order to satisfy the control objective, at first we take sq i as a control and define. sq z i (21) where is a virtual control. consider the following lyapunov candidate function 2 2 2 20 d 0 0 w (l) (0) [ (0) y ] 2 2 2 2 2 l l t x t t p m m y dx y dx y y y (22) where d is a strictly positive constant. it is straightforward to show that v can be lower and upper bounded as below 2 2 2 2 2 1 0 0 w [ (l) y (0) [ (0) ] l l t x t t d y dx y dx y y y (23) and 2 2 2 2 2 2 0 0 w [ (l) y (0) [ (0) ] l l t x t t d y dx y dx y y y (24) where 2 2 2 2 2 1 0 0 1 min( , , (l), (0), [ (0) ] ) 2 l l t x t t d y dx y dx y y y y (25) proceedings of engineering and technology innovation, vol. 8, 2018, pp. 15 -22 copyright © taeti 19 and 2 2 2 2 2 2 0 0 1 min( , , (l), (0), [ (0) ] ) 2 l l t x t t d y dx y dx y y y y (26) taking time derivative with respect to eq.(26),then 0 t x 0 0 1 1 0 w = p (l) (l) (0)[ (0)] ( (0) ) (0) (0){ [ (0) ]} l t x t x d t t d y y p y y y z p y y y y y z y y (27) eq. (27) suggests that virtual control 𝑎1 can be chosen as follows (0) [ (0) ] t d ky y y (28) where k is a strictly positive constant. in the second step, the actual control input 𝑢𝑠𝑞 is designed to regulate 𝑧1 at the origin. to archive this target, we consider a lyapunov candidate function as follows 21 w 2 v z (29) taking time derivative with respect to eq. (29), it yields 2 1 1 1 2 3 4 (0) [ (0) ( 1) (0)] 1 t sd sq sq tt t v k y z i i u k y y (30) the actual control input 𝑢𝑠𝑞 can be derived as 34 1 2 1 1 (0) ( 1) (0) sq sd sq tt t u i i k y y (31) where 2 k is positive constant. remark 3: the control input sd u can be derived based on backstepping control method as follows 2 1 1 2 2 1 2 22 ˆ ˆ1 1 1 1 1ˆ ˆ( )( ) rd rd sd sd s sq rd sd rd r r r r d d u i i c i c t t c z z d z s t r t dt tdt (32) where 2 2 2 1 1 ( ) ( ) r t (33) 1 c and 2 c are strictly positive constant, 1 z and 2 z are errors between desired and virtual control when designing the flu x controller. the control design is co mpleted and it is straightforwa rd to show that with the selected control input sq u render the first time derivative of the lyapunov candidate function v as 1 2 (0) 0 t v k y k z (34) eq. (34) shows that (t)v is upper bounded by (0)v . this consequently implies that ( )ty l , (0)ty and (0)t dy y are bounded. further investigation of eq. (34) can prove exponential convergence to dy of (0)y and ( ).y l 4. numerical simulations in order to verify the effectiveness of the proposed feedback control. simu lations are carried out using an induction motor of 7.5kw (other motor power ranges can be applied with no loss of generality). the closed loop system is simu lat ed in proceedings of engineering and technology innovation, vol. 8, 2018, pp. 15 -22 copyright © taeti 20 matlab/simulink environment. simulation parameters are given in table 1. table 1 induction motor parameters parameter value nominal power pn = 2,5kw pairs of pole pp = 2 nominal current un = 340v nominal speed n = 1400 rpm stator resistance rs = 2.521ω stator inductance ls= 0.1825 h rotor resistance rs = 0.976 rotor inductance lr = 0.1858 h inertial moment j = 0,117 kg.m2 the simulat ion is carried out with the mass of the trolley is of 100kg, payload mass and cable length are of 400kg and 5m, respectively. simu lation scenario is to regulate trolley and payload position to a desired position of 5m fro m the init ial condition. it is assumed that initially the positions of the trolley and payload are coinciding. fig. 2 system response of the trolley and payload without control fig. 3 system response of the trolley and payload with control fig. 2 clearly represents large fluctuation of the trolley and the payload. this phenomenon is undesirable in practice. fig. 4 velocity response of the trolley with control fig. 5 control input it can be seen from the simulation results the effect of control action to the system. without control, the payload variation with the ma ximu m va lue of appro ximate 1m around the desired position. when the control is activated, the swing angle of the payload is reduced considerably. finally, the payload mass and cable length are assigned to 600kg and 7m, respectively. due to heavier load and longer cable, higher payload fluctuation is expected in comparison with the previous results. nu merical simu lation indicates that the effectiveness of the proposed control design. trolley position is regulated at the proceedings of engineering and technology innovation, vol. 8, 2018, pp. 15 -22 copyright © taeti 21 desired value after 20s, and the payload also tracks the target after a few oscillat ions. in addition, the control input is of the applicable practice range. fig. 6 system response of the trolley and payload without control fig. 7 system response of the trolley and payload with control fig. 8 velocity response of the trolley with control fig. 9 control input 5. conclusion a design of a position and vibration suppression control of a gantry crane system is demonstrated in this paper. based on energy approach, a system of partia l differentia l and ordinary differentia l equations that govern the system’s mot ions includ ing cable and actuator dynamics are derived. the system dynamics represent fle xib ility of the gantry cable. the lyapunov direct method and backstepping technique are employed to design the controller. stability and the effectiveness of the closed -loop system are verified ana lytically and illustrated numerically. the direct e xtension of the paper is to consider the motion of the system in three-dimensional space. references [1] r. blevins, formulas for natural frequency and mode shape, new york: van nostrand reinhold co., 1979. [2] w. singhose, d. kim, and m. kenison, “input shaping control of double-pendulum bridge crane oscillat ions,” journal of dynamic systems, measurement and control, vol. 130, no. 3, pp. 1-7, may 2008. [3] r. manning, j. cle ment, d. kim, and w. singhose, “dynamics and control of bridge cranes transporting distributed-mass payloads,” journal of dynamic systems, measurement and control, vol. 132, no. 1, pp. 014-505, january 2010. [4] d. kim and w. singhose, “performance studies of human operators driving double -pendulum bridge cranes,” control engineering practice, vol. 18, no. 6, pp. 567-576, 2010. [5] y. sakawa and h. sano, “nonlinear model and linear robust control of overhead traveling cranes,” nonlinear analysis, theory, methods & applications, vol. 30, no. 4, pp. 2197-2207, 1997. [6] y. fang, w. e. dixon, d. dawson, and e. zergeroglu, “nonlinear coupling control laws for an under-actuated overhead crane system,” ieee/asme trans actions on mechatronics, vol. 8, no. 3, pp. 418-423, september 2003. proceedings of engineering and technology innovation, vol. 8, 2018, pp. 15 -22 copyright © taeti 22 [7] h. park, d. chwa, and k. s. hong, “a feedback linearization control of container cranes: varying rope length,” international journal of control automation and systems , vol. 5, no. 4, pp. 379-387, august 2007. [8] h. h. lee, “an anti-swing trajectory control of overhead cranes with high -speed hoisting,” proc. amer. control conf., denver, june 2003, pp. 1440-1445. [9] h. h. lee, “a new approach for the anti-swing control of overhead cranes with high -speed load hoisting,” international journal of control, vol. 76, no. 15, pp. 1493-1499, october 2003. [10] h. h. lee, y. liang, and d. segura, “a sliding-mode anti-swing trajectory control for overhead cranes with high -speed load hoisting,” journal of dynamic systems measurement and control, vol. 128, no. 4, pp. 842-845, december 2006. [11] h. h. lee , “motion planning for three-dimensional overhead cranes with high -speed load hoisting,” international journal of control, vol. 78, no. 12, pp. 875-886, august 2005. [12] j. h. yang and k. s. yang, “adaptive coupling control for overhead crane systems,” mechatronics, vol. 17, no. 2-3, pp. 143-152, march-april 2007. [13] j. h. yang and k. s. yang, “adaptive control for 3-d overhead crane systems,” proc. a mer. control conf., june 2006, pp. 1832-1837. [14] a. t. le and s. g. lee, “3d cooperative control of tower cranes us ing robust adaptive techniques ,” journal of the franklin institute, vol. 354, no. 18, pp. 8333-8357, 2017. [15] a. benhidjeb and g. l. gissinger, “fuzzy control of an overhead crane performance comparison with classic control,” control engineering practice, vol. 3, no. 12, pp. 1687-1696, december 1995. [16] j. h. suh, j. w. lee, y. j. lee, and k. s. lee, “anti-sway position control of an automated transfer cran e based on neural network predictive pid controller,” journal of mechanical science and technology, vol. 19, no. 2, pp. 505-519, december 2005. [17] t. a. le, g. h. kim, m. y. kim, and s. g. lee, “partial feedback linearization control of overhead crane with varying cable lengths ,” international journal of precision engineering and manufacturing, vol. 13, no. 4, pp. 501-507, 2012. [18] n. c. singer and w. p. seering, “preshaping command inputs to reduce system vibration,” journal of dynamic systems, measurement and control, vol. 112, no. 1, pp. 76-82, march 1990. [19] w. singhose, w. seering, and n. singer, “ residual v ibration reduction using vector diagrams to generate shaped inputs,” asme j. of mechanical design, vol. 116, no. 2, pp. 654-659, june 1994. [20] t. singh and s. r. vada li, “ robust time-delay control,” journal o f dyna mic systems, measurement and control, vol. 115, no. 2a, pp. 303-306, 1993. [21] s. joshi,. “position control of a flexible cable gantry crane: theory and experiment,” proc. the american control conference, june 1995, pp. 2820-2824. [22] b. d’andréa-novel and j. m. coron, “exponential stabilization of an overhead crane with flexible cab le via a back-stepping approach,” automatica, vol. 36, no. 4, pp. 587-593, 2000. [23] g. lodewijks, “anti-sway control of container cranes as a fle xible cable system,” proc. ieee international conf. control applications, ieee press, 2004, pp. 1564-1569. [24] s. s. ge, s. zhang, and w. he, “vibration control of a coupled nonlinear string system in transverse and longitudinal directions ,” proc. the ieee conf. decision and control, ieee press, 2011, pp. 3742-3747. [25] x. he, w. he, j. shi, and c. sun, “boundary vibration control of variable length crane systems in two -d imensional space with output constraints ,” proc. ieee/asme transactions on mechatronics, ieee press, 2017, vol. 22, no. 5, pp. 1952-1962. [26] t. l. nguyen and m. d. duong, “nonlinear control of flexible two-dimensional overhead cranes, adaptive robust control systems ,” intech, doi: 10.5772/intechopen.71657, 2018. [27] j. k. liu, h. qin, and w. he, “modelling and vibration control for a flexible string sy stem in three-dimensional space,” iet control theory & applications , vol. 9, no. 16, pp. 2387-2394, 2015. https://www.researchgate.net/journal/0022-0434_journal_of_dynamic_systems_measurement_and_control  proceedings of engineering and technology innovation, vol. 15, 2020, pp. 08 15 analysis of association between caesarean delivery and gestational diabetes mellitus using machine learning nisana siddegowda prema 1,* , mullur puttabuddi pushpalatha 2 1 department of information science and engineering, vidyavardhaka college of engineering, mysuru, india 2 department of computer science and engineering, sri jayachamarajendra college of engineering, mysuru, india received 16 september 2019; received in revised form 23 december 2019; accepted 09 january 2020 doi: https://doi.org/10.46604/peti.2020.4740 abstract the study aims to analyze the association between gestational diabetes mellitus (gdm) and other risk factors of cesarean delivery using machine learning (ml). the dataset used for the analysis is from the pregnancy risk assessment survey (prams), considered in two scenarios, i.e., all the data is taken, and all the data of the women who developed gdm. further, the data is developed in two groups data-i and data-ii by considering multiparous and primiparous women details, respectively. the correlation analysis and major classification algorithms are applied to the data. it is founded that the top risk factors for the first time cesarean delivery are the age, height, weight, race of the women, presence of hypertension and gestational diabetes mellitus. the major risk factor for repeated cesarean delivery is the previous cesarean delivery. the presence of gdm is also one of the risk factors for cesarean delivery. keywords: c-section, cesarean delivery, gdm, machine learning 1. introduction a cesarean (c-section) delivery is a surgical procedure in which a fetus delivered through an incision in the mother's abdomen and uterus. over the past 30 years, there is an increase in c-section delivery. the cdc reported that the highest ever national cesarean birth rate was 29.1% in november, 2005 [1]. the factors associated with deciding the type of deliveries are social-demographic, medical, obstetric history, antenatal history, etc. it is very important to know the presence of any of the above-mentioned risk factors at earliest which will help the doctors and the patients to make the right decision at the right time. 1.1. complications associated with c-section delivery the potential complications to mothers are infection and bleeding which can lead to anemia. further complications are pregnancies, pain, medicine, anesthesia negative reactions, etc. the rate of maternal mortality is higher in c-section deliveries when compared to vaginal deliveries. complications for the baby are low birth weight and respiratory problems. there is a possibility of injury for the baby during incision. 1.2. machine learning in healthcare machine learning (ml) is a technique of training a machine to recognize patterns using data and an algorithm. the prediction accuracy of the machine increases with the data and the complexity of the rules fed into the machine. healthcare is one of the leading applications of ml which has massive and enormous data. ml assists healthcare professionals in analyzing the data and helps them in the right decision making [2]. * corresponding author. e-mail address: premans@vvce.ac.in proceedings of engineering and technology innovation, vol. 15, 2020, pp. 08 15 9 the purpose of this work is to apply ml and statistical techniques for the analysis of the association between cesarean delivery with gdm and other risk factors, as well as to apply the predictions model for the prediction of c-section delivery for the first time delivery and repeated deliveries. 2. literature survey sunantha sodsee [3] proposed a cephalopelvic disproportion (cpd) based on the nearest neighbors (nn) algorithm for the prediction of cesarean delivery. in the proposed nn model, two determined threshold distances were used to identify the nearest and farthest neighbors. the highest performance was achieved by the proposed model [3]. obstetric and pregnancy factors were used in predicting the type of delivery using data mining models by sonia pereira et al. [4], and applied different data mining algorithms in different scenarios; they have achieved the highest accuracy of 84%. farhad soleimanjan et al. [5] used decision tree c4.5 for the prediction of cesarean delivery, the attributes considered are age, blood pressure, and heart problems. the obtained accuracy by the classifier is 86%. decision tree and artificial neural network were used to classify the births into normal and caesarian by ayesha sana et al. [6], the classification accuracy obtained was 80% and 82%, respectively. the authors also used association rules for the identification of caesarian birth patterns. in the study, the authors showed that high blood pressure, lack of education, and pulse rate are associated with caesarian birth. alisha kamat et al. [7] proposed a prediction model using naive bayes and id3 classifiers to determine the type of delivery based on ultrasonography, urine and blood reports of pregnant women. ml techniques like a decision tree and naïve bayes were applied for the prediction of pregnancy-related risk factors [8] and to predict normal or abnormal stages of pregnancy [9]. a classification model was proposed, and it allows an estimation of the interval for the value of the apgar score depending on mother and newborn data [10]. abbas et al. [11] proposed a decision support system using ml techniques to assist the physicians in adopting the correct decisions. they had used neural networks, knn, naïve bayes, and svm classifiers for the analysis of risk factors of c-section delivery. md rafiul hassan et al. [12] proposed a feature selection algorithm coupled with automated classification using ml techniques to analyze and predict ivf pregnancy in greater accuracy. they had used five attributes to asses the prediction ability of ivf pregnancy and five different ml models, namely mlp, svm, c4.5, cart, and random forest. prema pushpalatha [13] used svm and logistic regression for the diagnosis of preterm birth in the pregnant woman having either diabetes mellitus or gdm; the highest accuracy obtained is 86%. prema pushpalatha [13] also conducted a review on applications of data mining techniques in preterm birth prediction. furthermore, the most commonly used data mining technique which was founded is classification in that the usual techniques are support vector machine (svm) and logistic regression. the most commonly considered risk factors are socio-demographic, behavioral (lifestyle), and pregnancy history. further, the application of deep neural networks for the prediction of diabetes was proposed, for five-fold cross-validation, and achieved good results [15]. a prediction of the model using support vector machine and the nearest neighbor was proposed for the prediction of breast cancer, they have obtained the highest accuracy of 99.68% for support vector machine in the training phase [16]. 3. materials and methods 3.1. dataset the data used for this work is the pregnancy risk assessment survey (prams) collected from the centers for control and prevention (cdc). the prams data will be helpful to identify the group of women and infants with high-risk health issues. the data set contains about 41000 instances with more than 350 attributes related to maternal and child healthcare. the data set used for this work is reduced to 13550 instances after removing missing values instances and the outliers. outliers are detected proceedings of engineering and technology innovation, vol. 15, 2020, pp. 08 15 10 using the interquartile range. the prams dataset has about 350 attributes, in that few are redundant and many are irrelevant to the present study, hence only the required attributes are considered. about 20 attributes (as risk factors) are considered, which are related to the following factors:  maternal age  height and weight of the mother  obesity  diabetes mellitus(dm & gdm)  hypertension  lifestyle  obstetric history  educational level  maternal race  urban or rural most of the values are all categorical, few attributes are of type ordinal. the analysis of the data is done by considering complete data and the details of the women who developed gdm, respectively. further, the data is developed in data-i and data-ii groups by considering multiparous and primiparous women details, respectively. in the data set containing the details of women having gdm about 9% and 6% cases are c-section deliveries in data-i and data-ii respectively as shown in fig. 1. fig. 1 c-section delivery distribution in gdm and non-gdm data 3.2. association between the features feature association refers to the statistical relationship(s) between the feature variables. nominal variables are measured at the nominal level and have no inherent ranking. there are many statistical measures available to measure the dependency between the variables, for our data the following measures are used: (1) the correlation coefficient: the strength of the relationship between two variables is measured using the correlation coefficient. the correlation may be negative or positive is decided by the correlation coefficient values, -1 and 1, respectively; the value 0 indicates there is no relationship. (2) cramer's v: by using cramer’s v to calculate the strength of association, if the chi-square test is significant, the v value is between 1 and 0 which tells strong and little association between the variables respectively. (3) feature selection: the following features selection techniques are applied to the data to figure out the most important features. proceedings of engineering and technology innovation, vol. 15, 2020, pp. 08 15 11  correlation-based feature-subset selection: this technique is applied for the selection of the best subset of features. in this method subset of attributes is evaluated by considering the individual predictive ability of each feature along with the degree of redundancy between them. subsets of features that are highly correlated with the class while having low inter-correlation are preferred.  cfssubseteval: evaluates the worth of a subset of attributes by considering the individual predictive ability of each feature along with the degree of redundancy between them. subsets of features that are highly correlated with the class while having low intercorrelation are preferred.  gainratioattributeeval: evaluates the worth of an attribute by measuring the gain ratio concerning the class. gaines(class, attribute)=(h(class)-h(class|attribute))/h(attribute).  infogainattributeeval: evaluates the worth of an attribute by measuring the information gain concerning the class. infogain(class,attribute)=h(class)-h(class|attribute).  onerattributeeval: evaluates the worth of an attribute by using the oner classifier.  symmetricaluncertattributeeval: evaluates the worth of an attribute by measuring the symmetrical uncertainty concerning the class.  symmu(class, attribute)=2*(h(class)-h(class|attribute))/h(class)+h(attribute) 3.3. classification it is a supervised technique of grouping objects into different labeled classes. the classifiers used for this work are random forest, naive bayes, logistic regression and the nearest neighbor classifier(knn). (1) naivebayes: this is a classifier based on bayes theorem with independence assumptions between the features, which uses the maximum likelihood method. (2) random forest: a random forest is an ensemble approach that can also be thought of as a form of the nearest neighbor predictor. each decision tree is constructed by using a random subset of the training data. ensembles are a divide-and-conquer approach used to improve performance. (3) logistic regression: logistics is the regression analysis when the dependent variable is binary. it is used to describe data and to explain the relationship between one dependent binary variable and one or more nominal, ordinal, interval or ratio-level independent variables. (4) knn: it is a non-parametric, lazy learning algorithm, which can be used for both classification and regression. an object is classified by a majority vote of its neighbors, with the object being assigned to the class most common among its k nearest neighbors. if k = 1, then the object is simply assigned to the class of that single nearest neighbor. 4. result and discussion 4.1. the correlation coefficient the correlations between all the individual features are calculated for data-i and data ii as shown in fig. 2 and fig. 3, respectively. it can be observed from the matrix that if a mother is diabetic then there is a strong chance of getting, gdm in the current pregnancy. features like hypertension, age, weight, and gdm are associated with c-section delivery. for repeated c-section, a previous c-section feature has a strong correlation. proceedings of engineering and technology innovation, vol. 15, 2020, pp. 08 15 12 fig. 2 correlation matrix of data-i fig. 3 correlation matrix of data-ii cramer’s v test is done on the data-i and data-ii separately, and the values are tabulated in table 1. from the v values, it can be observed that there is a strong association between previous c-section delivery and repeated c-section delivery in data-i. there is a moderate association between c-section delivery and the age, weight, height of the women in both the data sets. the maternal chronic condition like diabetes and hypertension are also associated with c-section delivery. table 1 top features after cramer’s v test data set used feature cramer’ v data-i (repeated c-section )  previous c-section  gdm  hypertension  mothers weight  mothers height 0.9038 0.0469 0.0602 0.1695 0.0900 data-ii (first c-section)  mothers age  gdm  mothers weight  mothers height  hypertension 0.0581 0.0449 0.1454 0.0720 0.0799 proceedings of engineering and technology innovation, vol. 15, 2020, pp. 08 15 13 4.2. feature selection the top-ranked attributes by the above-mentioned features selection techniques are shown in table. 2, it can be observed in data-ii that, the most ranked feature is hypertension, body mass index (bmi), the height of the women, age as well as weight. and, in data-i, the attributes previous c-section is the top-ranked attribute by all the techniques. along with previous c-section, the other top-ranked attributes for data-i is height, weight, bmi, and age of the women. table 2 the ranked features using various features selection techniques sl..no method selected features: for data-ii selected features: for data-i 1 cfssubseteval: bestfit  hypertension  mothers height  previous live birth  previous c-section  mothers bmi 2 correlationattributeeval  hypertension  mother weight  gdm  previous c-section  mothers weight  mothers age 3 gainratioattributeeval  hypertension  gdm  diabetes  previous c-section  mothers weight  hypertension 4 infogainattributeeval  hypertension  mothers bmi  mothers age  previous c-section  mothers bmi  mothers age 5 onerattributeeval  mothers race  maternal education  mothers height  previous c-section  mothers race  smoking 6 symmetricaluncertattributeeval  hypertension  mothers bmi  mothers age  previous c-section  mothers bmi  mothers weight 7 relieffattributeeval  mothers race  mothers bmi  maternal education  previous c-section  mothers bmi  maternal education 4.3. classification table 3 accuracy obtained for the classifiers data set used classifier accuracy in % for all features for top -7 features for gdm data data-i (repeated c-section) knn(k=15) 96 96 95 naive bayes 96.5 96.5 96.9 logistic regression 96.5 96.5 96.8 random-forest 95 96.5 95 data-ii (first c-section) knn(k=15) 88.9 88.8 84.8 naive bayes 88.6 88.9 84.6 logistic regression 88.9 88.9 84.8 random-forest 88.9 88.9 84.9 the above-mentioned classifiers were applied for data-i and data-ii in both the scenarios by considering all the attributes, and the accuracy obtained by the algorithms are shown in table 3. for the prediction, different classifier are used firstly; random-forest adopted about ten trees and maximum instances for splitting is taken as five. for the k-nearest neighbor classifier, the value of k chosen is 15, and the measure of the distance used is euclidean. the different k values have experimented with knn. if the value k is 15, the highest accuracy is achieved. the classifiers were evaluated using 10 fold cross-validation which the original sample is randomly partitioned into 10 equal-sized subsamples. of the 10 subsamples, a single subsample is retained proceedings of engineering and technology innovation, vol. 15, 2020, pp. 08 15 14 as the validation data for testing the model, and the remaining subsamples were used as training data. the cross-validation process is then repeated 10 times, with each of the 10 subsamples used exactly once as the validation data. the accuracy of the classifiers is more for the data-ii compared to data-i in both the scenarios. logistic regression and naïve bayes classifiers are giving the highest accuracy for both the data sets. there is a decrease in the accuracy of the classifiers in the first c-section in the dataset containing gdm women’s details when compared with the accuracy of the data. the same classifiers are applied for the top-selected features by using the above-mentioned feature selection methods but there is not much improvement in the accuracy of the classifiers. the roc analysis curves of both data are shown in fig. 4 and fig. 5. fig. 4 roc for data-i fig. 5 roc for data-ii 5. conclusions a machine learning-based decision support system has been presented, and the top risk factors for cesarean delivery have been analyzed in both the first times and repeated pregnancy cases. the major risk factors for the first time cesarean delivery are the age, height, weight, race of the mother, and the presence of hypertension. for repeated cesarean, the delivery of previous cesarean delivery is the major risk factor. there is an association between gdm and cesarean delivery in the data set considered for the study. the classification model has achieved the highest accuracy of 96% for multiparous and 89% for primiparous mothers in predicting c-section delivery. in the future, the analysis will be done with paternal risk factors, including ethnicity, financial status, and stress-related factors. the data set is imbalanced, and the ratio of negative classes is more compared to the positive instances. therefore, there is a lot of scopes to improve the classification accuracy handling imbalanced dataset problems. conflicts of interest the authors declare no conflict of interest. references [1] “reasons for a cesarean birth,” https://americanpregnancy.org/labor-and-birth/reasons-for-a-cesarean/, october 13, 2019 [2] p. chowriappa, s. dua, and y. todorov, introduction to machine learning in healthcare informatics, machine learning in healthcare informatics, springer, 2014, pp. 1-23. [3] s. sodsee, “predicting caesarean section by applying nearest neighbor analysis,” procedia computer science, vol. 31, pp. 5-14, december 2014. [4] s. pereira, f. portela, m. f. santos, j. machado, and a. abelha, “predicting type of delivery by identification of obstetric risk factors through data mining,” procedia computer science, vol. 64, pp. 601-609, december 2015. proceedings of engineering and technology innovation, vol. 15, 2020, pp. 08 15 15 [5] f. soleimanian, p. mohammadi, and p. hakimi, “application of decision tree algorithm for data mining in healthcare operations: a case study,” international journal of computer applications, vol. 52, no. 6, pp. 21-26, august 2012. [6] a. sana, s. razzaq, and j. ferzund, “automated diagnosis and cause analysis of cesarean section using machine learning techniques,” international journal of machine learning and computing, vol. 2, no. 5, p. 677-680, october 2012. [7] a. kamat, v. oswal, and m. datar, “implementation of classification algorithms to predict mode of delivery,” international journal of computer science and information technologies, vol. 6, no.5, pp. 4531-4534, 2015. [8] b. lakshmi, t. indumathi, and n. ravi, “a comparative study of classification algorithms for predicting gestational risks in pregnant women,” international conference on computers, communications, and systems, november 2015, pp. 42-46. [9] r. sawant and n. gaikwad, “hybrid prediction method for pregnancy data set,” 2015 1st international conference on next generation computing technologies, september 2015, pp. 918-920. [10] r. robu and ş. holban, “the analysis and classification of birth data,” acta polytechnica hungarica, vol. 12, no. 4, pp. 77-96, july 2015. [11] s. a. abbas, r. riaz, s. z. h. kazmi, s. s. rizvi, and s. j. kwon, “cause analysis of caesarian sections and application of machine learning methods for classification of birth data,” ieee access, november 2018, pp. 67555-67561. [12] m. r. hassan, s. al-insaif, m. i. hossain, and j. kamruzzaman, “a machine learning approach for prediction of pregnancy outcome following ivf treatment,” neural computing and applications, september 2018, pp. 1-15. [13] n. s. prema and m. p. pushpalatha, “machine learning approach for preterm birth prediction based on maternal chronic conditions,” emerging research in electronics, computer science and technology, springer, 2019. [14] n. s. prema and m. p. pushpalatha, “prediction of preterm birth using data mining-a survey,” iioab journal, vol. 10, no. 2, pp. 13-17, january 2019. [15] s. i. ayon and m. islam, “diabetes prediction: a deep learning approach,” international journal of information engineering and electronic business, vol. 2, pp. 21-27, march 2019. [16] m. m. islam, h. iqbal, m. r. haque, and m. k. hasan, “prediction of breast cancer using support vector machine and k-nearest neighbors,” 2017 ieee region 10 humanitarian technology conference (r10-htc), december 2017, pp. 226-229. 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/).  proceedings of engineering and technology innovation, vol. 21, 2022, pp. 01-09 application of ai face recognition technology in swipe card attendance systems for hospitals te-kwei wang 1,* , yu-hsun lin 2 , kai-ping li 1 1 department of electrical engineering, ming chi university of technology, new taipei city, taiwan 2 department of business and management, ming chi university of technology, new taipei city, taiwan received 26 november 2021; received in revised form 26 february 2022; accepted 27 february 2022 doi: https://doi.org/10.46604/peti.2022.8984 abstract traditional swipe card attendance systems for hospitals cannot effectively protect employees’ personal information and ensure that the employees are swiping their own cards. to solve the problem, the present study proposes a novel hospital swipe card attendance system using an artificial intelligence (ai) face modeling system with an open-source face database. the proposed system employs a multi-task cascaded convolutional network (mtcnn) algorithm and facenet to improve the performance of face recognition. the system can compare the face of the one who swipes a card with the faces of cardholders in the database, thereby preventing the one from clocking in on behalf of others. the results show that the application of ai technology in the hospital swipe card attendance system can realize the promise of protecting employees’ personal information and verifying employees’ identities. keywords: face recognition, ai face recognition technology, swipe card attendance system, personal information 1. introduction attendance systems have been widely used in the world to record when employees start and stop their work. from basic attendance systems using paper time cards to advanced attendance systems using chips or swipe cards, today’s systems can successfully keep track of the attendance of employees. however, most of the systems cannot ensure that the person currently clocking in is the cardholder. since the identity of the person clocking in still needs to be checked, the cost of manpower is inevitable. the new-generation fingerprint-based biometric attendance systems can identify who is clocking in by recognizing human biological characteristics. the systems can verify users’ identity when users place their fingers or palms on the readers of systems. they are convenient and can prevent one from clocking in on behalf of others. however, the use of fingerprint-based biometric attendance systems may lead to the risk of getting infectious diseases such as covid-19. to avoid the risk, face recognition-based biometric attendance systems are the alternatives. the earliest face recognition technology was realized by manually marking facial features (e.g., eyes and mouth) and using a computer for calculation [1-2]. today’s face recognition technology is realized based on artificial intelligence (ai) by collecting a huge number of face photos. face recognition technology has been widely used and proposed in recent years. the face is important personal information of an individual. if the individual is in a country with personal property protection laws and has not authorized others to record his/her face, then the data of his/her face cannot be collected and aggregated for big data. * corresponding author. e-mail address: yslin@mail.mcut.edu.tw tel.: +886-2-29089899; fax: +886-2-29084533 proceedings of engineering and technology innovation, vol. 21, 2022, pp. 01-09 2 in face recognition, face detection and alignment are two fundamental tasks [3]. mo et al. [3] disclosed that scholars tend to overlook the correlation between face detection and face alignment, and design the hardwired accelerators which produce extra communication and region overhead and cannot achieve energy-efficient acceleration. to remedy this shortcoming, mo et al. [3] proposed an accelerator based on a multi-task cascaded convolutional network (mtcnn) algorithm to improve acceleration performance by considering face detection and multi-face alignment. mtcnn was proposed by zhang et al. [4] in 2016. compared with other similar methods, their proposed face recognition method based on mtcnn has the following several advantages: (1) decreasing multiply-accumulate (mac) operations and memory access (22.8% and 24.8%); (2) improving the intersection of joint computations and speeding the hardware inference sorting process (16.0%); (3) decreasing memory capacity (38.3%) and achieving similar resolution, same throughput, and high pipeline utilization with only about half multipliers; (4) improving the hardware utilization of the fully connected layer (16.7%). recently, gu et al. [5] have applied mtcnn to improve face recognition performance in a classroom. when using face detection algorithms based on cnn to detect faces in the classroom, gu et al. [5] have noted that the face frame is not easily identified due to several problems including occlusions, large pose changes, and small-scale faces. they have thus validated their proposed face recognition method using the improved mtcnn algorithm, which can achieve better precision and efficiency than certain leading-edge approaches. to have good precision and efficiency, the hospital swipe card attendance system proposed in this study is therefore also integrated with the mtcnn algorithm. this study aims to develop a face recognition-based hospital swipe card attendance system that can record the attendance and verify the identity of employees at the same time. with the proposed system, when an employee swipes a card, the employees’ card number is used as the search condition, and the system automatically retrieves related data (i.e., the facial feature, euclidean distance, and the threshold value) from the human resource (hr) database. the system will then identify the person who has swiped the card. thus, integrating the function of human identification into the hospital swipe card attendance system can prevent the cases of swiping cards on behalf of others and achieve the purpose of privacy protection. the rest of this article contains the following sections: literature review of face recognition and mtcnn, proposed system architecture, system implementation, simulation process and result of the proposed system, and conclusions. 2. literature review 2.1. face recognition in the 1960s, face recognition was proposed by woody bledsoe, helen chan wolf, and charles bisson and was realized by manually marking facial features (e.g., eyes and mouth) and using a computer for calculation [1-2]. due to the technical limitations at that time, no major breakthroughs have been made. in the 1970s, goldstein, harmon, and lesk increased the number of the marked features to 21 items, including hair color, lip thickness, etc. although the accuracy was therefore improved, face recognition was still realized by manually marking facial features, which was very time-consuming and labor-intensive. in 1988, sirovich and kirby introduced linear algebra to improve face recognition performance. in 1991, turk and pentland realized how to detect faces in pictures to perform the work of sirovich and kirby. their work is one of the earliest automatic face recognition. later in the 1990s, the defense advanced research projects agency (darpa) and the national institute of standards and technology (nist) launched into the facial recognition technology (feret) program, supplying 2413 face data of 856 people to encourage the use of commercial facial recognition. about ten years later, nist began to conduct face recognition vendor tests (frvt). based on feret, nist provided the united states government with an evaluation of available commercial face recognition systems and technologies to determine if there were better face recognition technologies. proceedings of engineering and technology innovation, vol. 21, 2022, pp. 01-09 3 in 2006, the face recognition grand challenge (frgc) started to promote and develop the existing face recognition system in the united states. later, in 2010, facebook began to introduce the face recognition function. google proposed facenet in 2015 and obtained 99.63% accuracy using the database “labeled faces in the wild (lfw)”. in 2016, zhang et al. [4] proposed mtcnn for face detection and image pre-processing. 2.2. multi-task cascaded convolutional networks (mtcnn) mtcnn was proposed by zhang et al. [4] in 2016. it can be applied to face detection. with mtcnn, face recognition can achieve a good accuracy level and fast execution speed [5]. mtcnn uses the cascaded convolutional network structure to complete face detection and face alignment at the same time, thereby outputting the location of eyes, nose, and mouth in the face box. 3. proposed system architecture before the proposed face recognition system is developed, mtcnn is used to realize face detection. the flowchart of face recognition based on mtcnn is shown in fig. 1. first, the image pyramid is used to adjust the size of photos so that the photos can be used as model inputs. the second step is to input the photos to p-net. p-net is a fully convolutional network (fcn), and its output is a 32-dimensional feature vector, finding out whether the input possibly contains faces. if so, then the next thing to do is to identify the face candidate frames that match the face area with the original face photo through the operation of non-maximum suppression (nms). the third step is the operation of r-net. r-net is a convolutional neural network (cnn), which uses bilinear interpolation to expand the output of p-net to 24 × 24 and check whether the output of p-net contains faces. if there is a face, r-net finds the candidate frame and uses nms to filter it, using the same way as the operation of p-net. the fourth step is the operation of o-net, which is a cnn. o-net uses bilinear interpolation to expand the output of r-net to 48 × 48 as the input for the final face detection and the key point extraction. facenet is used to realize the proposed ai face recognition system [6]. it is a face recognition system proposed by google in 2015. since facenet uses an algorithm with the feature of easy calculation and available application, it has been widely applied to face verification and recognition. the architecture of the facenet model is shown in fig. 2. facenet uses a cnn to map the face photos to the euclidean space [7]. it employs triplet loss to compare two faces’ similarities in terms of the distance in the euclidean space. that is, the more similar the two faces in the photos, the smaller the distance. on the contrary, the more dissimilar the two faces in the photos, the greater the distance. as shown in fig. 3, the principle of triplet loss is that the operation of feature training is firstly conducted to select the wrong face that is most similar to the original sample (the smallest difference) and the correct face that is most dissimilar to the original sample (the largest difference) at the same time. next, iterative training is performed until the error is minimized, thereby improving the definition of facial features in the model to achieve the best recognition results. start input a photo adjust size is the photo size suitable? p-net r-net o-net end yes no fig. 2 the framework of the facenet model fig. 1 the operation flowchart of mtcnn fig. 3 the operation process of triplet loss proceedings of engineering and technology innovation, vol. 21, 2022, pp. 01-09 4 the operation process of triplet loss is to map the face photos to the euclidean space. the purpose of the operation is to make the dissimilar reference faces 𝑋𝑖 𝑎 (anchor) and 𝑋𝑖 𝑛 (negative) far away and make the similar reference faces 𝑋𝑖 𝑎 (anchor) and 𝑋𝑖 𝑝 (positive) close to each other in the euclidean space. the mathematical expression of the relationship can be expressed as: 2 2 2 2 ( ) ( ) ( ) ( ) , ( , , ) a p a n a p n i i i i i i i f x f x f x f x x x x      (1) the loss function can be obtained by: 2 2 2 2 ( ) ( ) ( ) ( ) n a p a n i i i i i f x f x f x f x            (2) the facenet architecture is different from other face recognition architectures that require classifiers. facenet directly performs spatial mapping, and the mapped dimension is only 128 dimensions. by using the triplet loss, facenet does not require a classifier. the accuracy of facenet is tested based on the academic data set, achieving an accuracy of 99.63%±0.09 and 95.12%±0.39 by using the lfw database and youtube faces database, respectively. the classification test is carried out using the face photo data set of facenet, and the results are shown in fig. 4. fig. 4 results of the classification test using the face photo data set of facenet start end clock-in machine initialization read a card number when a card is swiped check whether the card number exists in the database? error message: the card number has not been found in the database. please try again. take a face photo of the person who swiped the card does the photo match the cardholder photo in the database? download the feature(vector) of the card holder photo from the database calculate the feature vector using facenet calculate the euclidean distance rename the photo title as the card number upload the photo to the database upload the time of swiping the card to the database end is the distance <0.7? upload the time of swiping the card to the database end upload the time of swiping the card to the database notify the hr department of abnormal swiping incidents yes yes no no no yes fig. 5 the proposed swipe card attendance system proceedings of engineering and technology innovation, vol. 21, 2022, pp. 01-09 5 fig. 5 shows the de-capitalization process of the proposed swipe card attendance system [8]. the system is operated based on mtcnn, facenet, and face recognition. the system can only receive cardholders’ card numbers and personal information from the radio frequency identification (rfid) cards in the existing attendance system. it is possible to prevent the occurrence of clock-in incidents. when one swipes a card on the clock-in machine of the proposed system, the machine reads the information of clock-in, and the cardholder’s name and card number recorded in the rfid card are used as the basis to search the database. that is, the proposed system reads the rfid card to obtain the card number and use it as a search condition to query the corresponding cardholder’s photo in the database. because the information of card numbers is available only in searching the database, personal information can be protected. the system can compare the face of the one who swipes the card with the faces of cardholders in the database, thereby preventing the one from clocking in on behalf of others. 4. system implementation 4.1. clock-in machine after the proposed swipe card attendance system is started, the clock-in machine is operated. the operation process of the machine is shown in fig. 6 [9]. the rfid reader is in the detection state. when an rfid card approaches the rfid reader, the rfid reader starts reading the card’s number. start card inspection is a card captured? conflict handling obtain the card number end yes no fig. 6 the operation flowchart of rfid swiping machine 4.2. face recognition mtcnn is a popular cnn-based detector for face recognition [5, 10-12]. as shown in fig. 7, the proposed system uses mtcnn for face detection and face cropping. it resets the size of face photos to 160 × 160 and removes redundant impurities in the photos after whitening. then, the photos are input to facenet to get their feature vectors. the euclidean distance between the feature vectors of the photos is calculated and face recognition is performed according to the set threshold. start use mtcnn for face detection and cropping perform photo scaling and whitening use facenet to calculate feature vectors calculate the euclidean distance identify the person according to the threshold end fig. 7 the operation flowchart of face recognition proceedings of engineering and technology innovation, vol. 21, 2022, pp. 01-09 6 5. simulation process and result of the proposed system 5.1. the participants involved in the simulation testing of the proposed system there are two participants involved in the simulation testing of this study. one participant is labeled as naf09. another is labeled as aa015. 5 photos for the participant naf09 are collected in the dataset. 7 photos for the participant aa015 are collected in the dataset. a total of 12 photos are in the data set. 5.2. the hardware used in the simulation testing of the proposed system table 1 shows the specification of system hardware used in this study. the central processing unit (cpu) for the proposed system is intel®core™ i5-10600k @4.1ghz. the graphics processing unit (gpu) is nvidia geforce gtx 1650. the system has a capacity of 16 gb memory. the software for developing the system is python version 3.8.8 with cuda (version 11.2) and tensorflow (version 2.6.0). table 1 specification of system hardware central processing unit (cpu) intel® core™ i5-10600k @4.1ghz graphics processing unit (gpu) nvidia geforce gtx 1650 computer memory 16gb-3000mhz python version 3.8.8 cuda version 11.2 tensorflow version 2.6.0 5.3. the simulation method fig. 8(c) shows the simulation process of the proposed system. step 1 is to load all photos of a participant. step 2 is to choose one photo as the reference photo and use the rest as the comparison photos. step 3 is to calculate the euclidean distance between the reference photo and each comparison photo in sequence and use the euclidean distance 0.7 as the identification criterion. if the euclidean distance is less than or equal to 0.7, then the photo is identified as “the photo of the participant”. that is, the identification result is “correct”. otherwise, the photo is identified as “not the photo of the participant”. that is, the identification result is “wrong”. step 4 is to count the number of correct identification results and divide it by the number of comparison photos to obtain the success rate of the reference photo identification. step 5 is to choose another photo as the reference photo and repeat steps 3 and 4 until all the photos have been tested as reference photos. step 6 is to load all photos of the other participant and repeat steps 2 to 5. (a) photos of the participant naf09 (b) photos of the participant aa015 fig. 8 the simulation of the proposed swipe card attendance system proceedings of engineering and technology innovation, vol. 21, 2022, pp. 01-09 7 start load all photos of a participant choose one photo as the reference photo and use the rest as the comparison photos calculate the euclidean distance euclidean distance �0.7? correct identification result calculate the success rate of identification is the reference photo testing finished? calculate the accuracy rate is the test of the participant finished? is the test of all the participants finished? wrong identification result end no yes yes yes yes no no no (c) the flowchart of the simulation process fig. 8 the simulation of the proposed swipe card attendance system (continued) (a) success rate of face recognition using naf09’s photos (minimum: 80%, maximum: 100%) (b) success rate of face recognition using aa105’s photos (minimum: 57%, maximum: 100%) fig. 9 success rate of face recognition using photos of each participant as shown in fig. 9, the success rate of face recognition using aa015’s photos is not satisfactory because the minimum is 57%. compared with the success rate using aa015’s photos, the success rate using naf09’s photos is better because the minimum is 80% and the maximum is 100%. overall, such results may imply that more samples are necessary to enhance the recognition accuracy. 5.4. the simulation result of the proposed system each participant has three or more photos with different angles for facial identification in the database. during the simulation, each participant uses the other participant’s rfid card to test the function of the proposed system. if the actual result shows “true”, it means that the person is swiping the card on behalf of the other. if the result shows “false”, the person who swiped the card is the cardholder. the simulation result is shown in table 2. proceedings of engineering and technology innovation, vol. 21, 2022, pp. 01-09 8 table 2 the simulation result of the proposed system the face photo of the person who swipes a card the face photo retrieved from the database euclidean distance actual result naf09_1.jpg aa015_2.jpg 1.1176 true naf09_1.jpg aa015_3.jpg 1.1005 true naf09_1.jpg aa015_4.jpg 1.2087 true aa015_1.jpg naf09_2.jpg 1.1015 true aa015_1.jpg naf09_3.jpg 1.0909 true aa015_1.jpg naf09_4.jpg 1.1197 true 6. conclusions the present study develops a swipe card attendance system with a face recognition function for hospitals. the test results verify that the system can identify the event where a card is swiped by a non-cardholder. when this case happens, the system will send an email to notify the hr department, which effectively reduces the labor cost of the hr department for attendance checking. the current decision-making method only uses euclidean distance 0.7 as the criterion, which is slightly rough. in the future, false acceptance rate (far) [13], false rejection rate (frr) [14], and equal error rate (eer) [15] will be added to the experiment to strengthen the decision-making process of face recognition. also, infrared scanning [16] will be added to increase the features. moreover, with only two participants, the present study may produce the issue of false positives or false negatives in the identification process, thus decreasing the accuracy of face recognition. further research should recruit more participants to join the research to improve the rigorousness of the research method. this study just focuses on one face in one frame. for those cases with several faces in one frame and with low-resolution photos, further research can employ multi-object face recognition using local binary pattern histogram to improve the success rate of face recognition [17]. it should be noted that, face recognition not only can be used in hospital swipe card attendance systems but also can be employed in monitoring the breath function of patients [18]. future research can integrate face recognition with thermal images to identify the patients and assess their health conditions. conflicts of interest the authors declare no conflicts of interest. references [1] i. goodfellow, y. bengio, and a. courville, deep learning, massachusetts: mit press, 2016. [2] s. z. li and a. k. jain, handbook of face recognition, new york: springer science and business media, 2005. [3] h. mo, l. liu, w. zhu, q. li, h. liu, s. yin, et al., “a multi-task hardwired accelerator for face detection and alignment,” ieee transactions on circuits and systems for video technology, vol. 30, no. 11, pp. 4284-4298, november 2020. [4] k. zhang, z. zhang, z. li, and y. qiao, “joint face detection and alignment using multitask cascaded convolutional networks,” ieee signal processing letters, vol. 23, no. 10, pp. 1499-1503, august 2016. [5] m. gu, x. liu, and j. feng, “classroom face detection algorithm based on improved mtcnn,” signal, image, and video processing, in press. [6] f. schroff, d. kalenichenko, and j. philbin, “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. [7] j. hearty, advanced machine learning with python, birmingham: packt publishing, 2016. [8] h. yang and x. han, “face recognition attendance system based on real-time video processing,” ieee access, vol. 8, pp. 159143-159150, july 2020. proceedings of engineering and technology innovation, vol. 21, 2022, pp. 01-09 9 [9] q. miao, f. xiao, h. huang, l. sun, and r. wang, “smart attendance system based on frequency distribution algorithm with passive rfid tags,” tsinghua science and technology, vol. 25, no. 2, pp. 217-226, april 2020. [10] n. zhou, r. liang, and w. shi, “a lightweight convolutional neural network for real-time facial expression detection,” ieee access, vol. 9, pp. 5573-5584, january 2021. [11] b. yu and d. tao, “anchor cascade for efficient face detection,” ieee transactions on image processing, vol. 28, no. 5, pp. 2490-2501, may 2019. [12] x. li, z. yang, and h. wu, “face detection based on receptive field enhanced multi-task cascaded convolutional neural networks,” ieee access, vol. 8, pp. 174922-174930, october 2020. [13] s. saito, y. tomioka, and h. kitazawa, “a theoretical framework for estimating false acceptance rate of prnu-based camera identification,” ieee transactions on information forensics and security, vol. 12, no. 9, pp. 2026-2035, september 2017. [14] n. merhav, “false-accept/false-reject trade-offs for ensembles of biometric authentication systems,” ieee transactions on information theory, vol. 65, no. 8, pp. 4997-5006, august 2019. [15] d. h. kaye, “the error of equal error rates,” law, probability, and risk, vol. 1, no. 1, pp. 3-8, july 2002. [16] r. he, j. cao, l. song, z. sun, and t. tan, “adversarial cross-spectral face completion for nir-vis face recognition,” ieee transactions on pattern analysis and machine intelligence, vol. 42, no. 5, pp. 1025-1037, may 2020. [17] r. r. isnanto, a. f. rochim, d. eridani, and g. d. cahyono, “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. [18] h. hendrick, a. aripriharta, s. k. chen, m. h. tung, t. c. chiang, and g. j. jong, “nostril in rgb imaginary by using ni vision labview,” proceedings of engineering and technology innovation, vol. 4, pp. 37-39, october 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/).  proceedings of engineering and technology innovation, vol. 20, 2022, pp. 12-23 closed loop v/f speed control of multilevel inverter fed induction motors using pid and fuzzy logic controllers aravinda shilpa konathala * , mamillapalli rose mary † , bade mounica, gudla meghana, jyothula renuka devi department of electrical engineering, andhra university college of engineering for women, andhra university, andhra pradesh, india received 08 august 2021; received in revised form 21 september 2021; accepted 22 september 2021 doi: https://doi.org/10.46604/peti.2021.8232 abstract the volts per hertz (v/f) speed control method makes induction motors available to be used in variable speed applications. however, the great harmonic distortion in supply voltage is a major issue of this method. therefore, this study focuses on decreasing the harmonic distortion in the supply to induction motors. for that, the use of thirty-one-level multilevel inverters (mlis) is investigated. another drawback of the v/f speed control method is the use of proportional-integral-derivative (pid) controllers as speed controllers, which are difficult to be tuned precisely. to overcome this, the use of fuzzy logic controllers is investigated. the whole model is developed and simulated in matlab/simulink by varying two variables, i.e., reference speed and load torque, one at a time using each controller. the observations for the changes in speed, torque, stator current, and supply voltage to the induction motors are made. the system’s performance using fuzzy logic controllers is found superior. keywords: multilevel inverter, induction motor, constant v/f speed control, pid controller, fuzzy logic controller 1. introduction induction motors have been widely used since the day of their invention. they are employed in almost all industrial, robotic, and domestic applications because of their advantages, e.g., robustness, inexpensiveness, simpleness and lightness in construction, maintenance-free operation, etc. although they possess some disadvantages like time-varying and non-linear dynamics, the advantages surpass the disadvantages. hence, they are used in various applications including fans, blowers, cranes, traction, electric vehicles, conveyors, etc. induction motors are remarkably known as constant speed motors because the drop in their speed from no-load to full-load is small. this limited their demand for constant speed applications. the advancements of the power electronics field paved a path for the speed control of induction motors. many speed control methods were developed to control the speed of induction motors to use their advantages even in variable speed applications. the variable voltage and variable frequency method or the volts per hertz (v/f) speed control method is found efficient among all the speed control methods. using this method, the speed of the motors can be controlled smoothly over a wide range [1-3]. the motors are fed from a voltage source inverter (vsi) to obtain the variable voltage and variable frequency supply. the vsi produces a twoor three-level voltage output that contains a greater number of harmonics. due to this, the motors develop a pulsating torque which is undesirable and needs to be overcome. to remove the harmonics from the inverter output, the inverter output is passed through a filter, whose * corresponding author. e-mail address: aravindashilpacf@andhrauniversity.edu.in † corresponding author. e-mail address: rosemarymamillapalli@gmail.com proceedings of engineering and technology innovation, vol. 20, 2022, pp. 12-23 size and cost increase with voltage requirement, increasing the cost of the whole system. chitra et al. [4] proposed that by employing a five-level cascaded h-bridge (chb) multilevel inverter (mli) in the place of a conventional vsi, the harmonic distortion and hence the torque pulsation can be reduced. the size of the filter, if employed, can also be reduced. however, their proposed method has some drawbacks, such as the increased number of switching devices and the increased complexity in the control circuit of the inverter. the use of five-level chb mli reduces the harmonic distortion in the inverter output as compared to the use of a vsi. however, it is high enough to cause heating of the motor winding. due to the non-linear nature of the induction motor control dynamics, an efficient control algorithm is required to control the speed of induction motors. the conventional controllers used for the induction motor speed control are proportional-integral (pi) controllers. however, because of the use of pi controllers, the overshoots of greater amplitude are observed in the speed response of induction motors, which requires a larger settling time to attain a steady state. hence, proportional-integral-derivative (pid) controllers can be used to decrease the overshoots slightly and hence the settling time [5]. this study focuses on employing an mli of a greater number of levels(thirty-one levels) to feed an induction motor for its variable frequency control so that an input supply of less harmonic distortion is obtained and the use of a filter can be avoided. also, the use of a pi or pid controller as a speed controller causes overshoots, and the exact mathematical model of the system is needed for tuning. hence, this study also focuses on investigating the use of a fuzzy logic controller, which has better performance than a pid controller as a speed controller. 2. variable voltage and variable frequency control if f is supply frequency and p is the number of stator poles, the synchronous speed (𝜔𝑠) of an induction motor is given by: 4 s f p    rad/s (1) since the induction motor is an asynchronous motor, it cannot run at synchronous speed, i.e., the speed at which the stator field rotates. the difference between the synchronous speed and the rotor speed is called the slip speed. hence, the rotor speed (𝜔𝑟) of the induction motor is given by:  1 sr s   rad/s (2) or   4 1r f s p     rad/s (3) where s is the slip. from the above equation, it is clear that the speed of the induction motor can be changed either by varying the number of poles, the slip, or the supply frequency. the speed controlled by varying the supply frequency is proved to be efficient. the equation for the stator induced electromotive force is given by: 4.44e k tf  v (4) where k and t are constants. if only the supply frequency is varied without changing the supply voltage, then the air-gap flux of the induction motor varies. however, the induction motor is designed to operate at the knee point of the magnetizing characteristic. therefore, if the flux increases beyond the rated value, then a larger amount of currents will be drawn by the motor to produce that flux. if the flux decreases below the rated value, the motor cannot develop the required torque. hence, it is required to be ensured that the flux remains at the rated value. for that, the magnitude of the supply voltage has to be changed 13 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 12-23 along with the supply frequency. due to this reason, this method of speed control is called the variable voltage and variable frequency control or constant v/f control. however, the voltage is kept at some higher value for low frequencies to overcome the stator resistance drop [6]. to produce this voltage of variable magnitude and frequency, the induction motor is fed from a vsi. the vsi produces voltage with greater harmonic distortion due to which the pulsating torque is produced and the motor winding gets heated up. thus, mli is a suitable alternative producing an approximate sinusoidal output waveform with less harmonic distortion. 3. multilevel inverter (mli) mli is a power electronic device that can provide the desired alternating current (ac) voltage levels at the output using multiple low direct current (dc) voltages as the inputs. mli generates higher voltage waveforms from lower voltage components. the major purpose of mli is to operate in high power applications using low-rating devices. therefore, mli is a preferred choice in industries for medium and high power applications. multilevel power conversion was introduced more than twenty years ago. over the past years, the demand for mlis is increasing gradually due to their advantages such as the output voltage with decreased harmonic content, the lower electromagnetic interference, and the decreased dv/dt stress on the switches. due to the reduction in the harmonic content, the voltage wave obtained from mlis is almost sinusoidal; hence, a higher wave quality is obtained when compared to vsis. mlis are generally operated at lower switching frequencies when compared to vsis; hence, the stresses on switches are also reduced. for low and medium voltage applications, the importance of mlis is discovered in nearly every subject of electrical engineering together with renewable power systems, high voltage direct current (hvdc) applications, industrial drive applications, uninterruptible energy supplies, etc. mlis are widely used in drives and different allied regions in industries. due to these advantages, numerous configurations of mlis are being developed which may fall under one of the three basic topologies, i.e., flying-capacitor (fc), neutral-point-clamped (npc), and chb topologies, or a few hybrids of those [7-9]. cascaded mlis are further classified into two types, i.e., symmetric and asymmetric types, based on the magnitude of dc sources. the input dc sources have equal magnitude for the symmetric type while the dc sources have unequal magnitude for the asymmetric type. however, the asymmetric chb mli has more advantages because of its ability to produce a greater number of levels with a less number of components [10-11]. in this study, a thirty-one-level asymmetrical mli is used to feed the motor. the topology contains 17 switches of which s1, s2, s3, and s4 form the first h-bridge, s5, s6, s7, and s8 form the second h-bridge, and s9, s10, s11, and s12 form the third h-bridge. this mli is controlled by multicarrier sinusoidal pulse-width modulation [12]. the topology is shown in fig. 1. fig. 1 thirty-one-level asymmetrical mli 14 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 12-23 the above topology gives an output voltage with total harmonic distortion (thd) as low as 4.00% [12]. it is very much less when compared to the thd presented in the output voltage of a twoor three-level vsi. since the thd is less, the drawbacks which occur by using the vsi in the v/f speed control can be overcome efficiently. 4. speed controller to control the speed of the induction motor which has a highly non-linear dynamic behavior, a strenuous control algorithm is required. the most commonly used controllers for this purpose include pi controllers, proportional-derivative (pd) controllers, and pid controllers, or a combination of these. pid controllers are proved to be superior because they provide reliable and robust control if tuned properly. to set the parameters of these controllers, the plant model has to be linearized by complex mathematical calculations. despite this effort, the values obtained are only approximate, and the variations that occur in the model due to the practical conditions add to the complexity. further, decent performance is obtained only at one operating speed. the drawbacks mentioned above show a requirement for a more robust, reliable, cost-effective, and less complex speed controller. a fuzzy logic controller possessing all these qualities seems to be a better alternative for the conventional controller [13-17]. from the last three decades, the demand for fuzzy control has increased gradually because of its characteristics such as better non-linearity handling capability, knowledge-based algorithm, being independent from exact plant modeling, and having a control algorithm that replicates the human thought process. along with these attributes, a fuzzy logic controller provides good response in stability, precision, and rapidity. hence, this controller is simple to design and also to use as it requires almost no knowledge of complex mathematical calculations but a qualitative knowledge of the system. 5. the proposed speed control model fig. 2 shows the block diagram of the proposed closed-loop v/f speed control model of an induction motor. the main components of the system include a speed controller, a slip regulator, a flux controller, and an mli supplying power to the induction motor. fig. 2 block diagram of the proposed model as shown in fig. 2, the actual speed of the rotor is compared with the reference speed in a comparator and a speed error is generated, which is the slip speed. this slip speed is passed through a speed controller which produces the output with reduced noise and good steady-state accuracy. when a pid controller is used, it performs proportional, integral, and derivative action on the speed error, and produces the controlled output. fig. 3 shows the block diagram of the fuzzy logic controller, which simulates the behaviour of the pid controller. when a fuzzy logic controller is used as a speed controller, it requires two inputs namely “error” and “derivative of error”, which are normalized to match the fuzzifier limits. the inference system then processes these two fuzzy inputs using the fuzzy control rules and the database. these are given by the programmer based on the chosen membership function and fuzzy rule table 15 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 12-23 which is shown in table 1, to give the output fuzzy variable. the obtained fuzzy outputs are defuzzified by the defuzzifier to produce a crisp value, i.e., change-of-control. this value is integrated using an integrator and limited to a safe value using a saturation limiter. this value is then multiplied with suitable integral gain and then added to the change-of-control signal which is multiplied by a suitable derivative gain. this makes the fuzzy logic controller imitate a pid controller. the controlled output is then obtained from the fuzzy logic controller. fig. 3 block diagram of fuzzy logic controller table 1 fuzzy rule table for output e ∆e nl nm ns ze ps pm pl nl nl nl nlm nm nms ns ze nm nl nlm nm nms ns ze ps ns nlm nm nms ns ze ps pms ze nm nms ns ze ps pms pm ps nms ns ze ps pms pm plm pm ns ze ps pms pm plm pl pl ze ps pms pm plm pl pl this controlled output, whether obtained from the pid controller or fuzzy logic controller, is fed to a slip regulator, which controls the slip, thereby limiting the operation of the motor in the stable region of torque-speed characteristics, i.e., from the synchronous speed to the speed corresponding to the maximum torque. also, the stator currents and the torque produced are limited within the safe value by controlling the slip. this controlled and regulated slip speed is added to the rotor speed, which gives the synchronous speed. the obtained synchronous speed is then converted into the respective frequency and fed to the flux control block, where the modulating signal of the required magnitude and frequency will be generated. this modulating signal is used to control the switching of the mli which feeds the induction motor. thus, the mli shown in fig. 1 produces a voltage of required magnitude and frequency. the motor runs at the reference speed specified, at the rated flux, and below the rated speed. 6. simulation results and discussion the proposed speed control model is simulated using matlab/simulink tool, as shown in fig. 4. a three-phase squirrel-cage induction motor of the rated voltage 400 v and speed 1500 rpm is chosen. from the motor parameters, the slip-speed value corresponding to the maximum speed is 55.8 rps. hence, this value is chosen as a limiting value in the slip regulator. to feed the induction motor, the dc sources in the three-phase mli are selected to be v1 = 22.6 v, v2 = 43.2 v, v3 = 86.4 v, and v4 = 172.8 v so that a peak voltage of v1 + v2 + v3 + v4 = 325 v will be generated and its root mean square (rms) value will be 230 v which is the required per phase voltage. fig. 5 shows the simulation model of the mli. 16 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 12-23 as shown in fig. 6, the flux control block limits the frequency to 50 hz, which is the rated frequency, and produces the modulating signal of the required frequency. the voltage is boosted to a slightly higher value at low frequencies to overcome the stator resistance drop. fig. 7 shows the simulation model of the fuzzy logic controller. a pid-type fuzzy logic controller is developed using the mamdani inference engine. the inference system consists of seven membership functions corresponding to the two inputs “error” and “change-of-error”, and eleven membership functions corresponding to the output “change-of-control”, as shown in fig. 8. fig. 4 simulation model of the proposed speed control model fig. 5 simulation model of the proposed thirty-one level mli fig. 6 simulation model of flux control block fig. 7 simulation model of fuzzy logic controller 17 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 12-23 (a) input membership functions (b) output membership functions fig. 8 input and output membership functions of fuzzy logic controller to test the viability of the proposed model, the model is simulated under two conditions using both pid and fuzzy logic controllers: (1) variable speed and constant load torque (2) constant speed and variable load torque 6.1. case 1: variable speed and constant load torque a constant load torque of 20 n-m is kept on the motor, and the reference speed is changed stepwise from 1000 rpm to 1400 rpm at t = 2 sec and then to 700 rpm at t = 4 sec. the changes in speed response (fig. 9), torque response (fig. 10), stator currents (figs. 11-12), and mli output (figs. 13-14) are noted as follows. fig. 9 variations in speed response fig. 10 variations in torque response 18 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 12-23 fig. 11 variations in stator currents using pid controller fig. 12 variations in stator currents using fuzzy logic controller fig. 13 variations in mli output using pid controller fig. 14 variations in mli output using fuzzy logic controller from the results obtained in case 1, the following are the observations made. the induction motor tracks the reference speed precisely using both pid and fuzzy logic controllers. though the overshoots are observed in the performance of both controllers, the settling-time is observed to be lower while using the fuzzy logic controller than the pid controller. the fuzzy 19 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 12-23 logic controller portraits a better response when the reference speed is changed. it tends to approach the new reference faster. the magnitude and the frequency of the output voltage of mli have also changed with the reference speed. table 2 provides the comparative analysis between the pid and fuzzy logic controller’s performance for case 1. table 2 comparative analysis for variable speed and constant load torque parameter pid controller fuzzy logic controller rise time 0.1 sec 0.1 sec settling time 0.5 sec 0.2 sec overshoot more less speed of response slow quick range of speed control 600 rpm to 1500 rpm 300 rpm to 1500 rpm 6.2. case 2: constant speed and variable load torque a constant reference speed of 1200 rpm is given to the motor, and the load torque on the motor is changed stepwise from 5 n-m to 30 n-m at t = 2 sec. the changes in speed response (fig. 15), torque response (fig. 16), stator currents (figs. 17-18), and mli output (figs. 19-20) are noted. fig. 15 variations in speed response fig. 16 variations in torque response fig. 17 variations in stator currents using pid controller 20 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 12-23 fig. 18 variations in stator currents using fuzzy logic controller fig. 19 variations in mli output using pid controller fig. 20 variations in mli output using fuzzy logic controller from the results obtained in case 2, the following are the observations made. the induction motor tracks the reference speed precisely using both controllers even when the torque is changed. as the load torque on the motor is changed, there is a disturbance in the motor speed. it is pronounced more when a pid controller is used as compared to the case when the fuzzy logic controller is used. the speed response due to the fuzzy logic controller shows almost negligible change when the load torque on the induction motor is changed. however, there is a small steady-state error for high load torque in fuzzy controller response which is almost negligible. in the torque response also, the oscillations are observed to be lower at starting and also at high load torque when the fuzzy logic controller is used when compared to the pid controller. the magnitude and the frequency of the output voltage of mli have also changed with the load torque. table 3 presents the comparative analysis between the pid and fuzzy logic controller’s performance for case 2. table 3 comparative analysis for constant speed and variable load torque parameter pid controller fuzzy logic controller rise time 0.1 sec 0.1 sec settling time 3 sec 0.2 sec overshoot more less speed of response slow quick range of torque control 0 n-m to 35 n-m 0 n-m to 35 n-m 21 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 12-23 7. conclusions in this study, a v/f closed-loop control scheme for mli fed induction motors developed in matlab/simulink is shown. a thirty-one-level mli is employed. the model is simulated using two controllers (conventional pid and fuzzy-pid controllers) in two cases (reference speed change and load torque change). from the results, it is evident that the magnitude of the torque ripples is reduced to a great extent as compared to the case when a three-level inverter is employed [3]. since the induction motor is fed from the thirty-one-level mli, which produces voltages of low harmonic distortion as compared to a three-level vsi, the harmonic distortion in the stator currents is also reduced. the power loss is reduced and hence the efficiency is increased as compared to the case when a two-level or a three-level inverter is employed. both the conventional pid and fuzzy-pid controllers provide a good response. the rise time is observed to be same for both controllers. however, the overshoots and settling-time are less when a fuzzy controller is used. the rotor speed tends to approach the new reference speed faster when the fuzzy controller is used than the pid controller. torque ripples are also observed to be slightly lower in the case of the fuzzy controller. hence, this v/f control model of mli fed induction motors can be employed in low, medium, and high power applications with the optimum usage of input power. from the results, the fuzzy logic controller shows better performance as compared to the pid controller. hence, the pid controller can be replaced by the fuzzy logic controller. conflicts of interest the authors declare no conflicts of interest. references [1] a. agrawal, r. s. lodhi, and p. nema, “a review of speed control methods of induction motors,” iosr journal of electrical and electronics engineering, vol. 13, no. 3, pp. 9-14, may-june 2018. [2] a. desai, s. deshmukh, u. chaugule, m. bonde, and j. satheesh, “speed control of 3-phase induction motor using v/f method,” academia, vol. 7, no. 4, pp. 6224-6226, 2017. [3] d. jee and n. patel, “v/f control of induction motor drive,” thesis, department of electrical engineering, national institute of technology, rourkela, odisha, 2013. [4] a. chitra, w. r. sultana, j. vanishree, s. sreejith, s. jose, and a. j. pulickan, “performance comparison of multilevel inverter topologies for closed loop v/f controlled induction motor drive,” energy procedia, vol. 117, pp. 958-965, june 2017. [5] r. bharti, m. kumar, and b. m. prasad, “v/f control of three phase induction motor,” international conference on vision towards emerging trends in communication and networking, pp. 1-4, march 2019. [6] c. u. ogbaku and m. u. agu, “effects of load and speed variations in a modified closed loop v/f induction motor drive,” nigerian journal of technology, vol. 31, no. 3, pp. 365-369, 2012. [7] i. colak, e. kabalci, and r. bayindir, “review of multilevel voltage source inverter topologies and control schemes,” energy conversion and management, vol. 52, no. 2, pp. 1114-1128, february 2011. [8] a. el-hosainy, h. a. hamed, h. z. azazi, and e. e. el-kholy, “a review of multilevel inverter topologies, control techniques, and applications,” 19th international middle-east power systems conference, pp. 1265-1275, december 2017. [9] i. e. tashiwa, g. d. dung, and b. s. adole, “review of multilevel inverters and their control techniques,” european journal of engineering research and science, vol. 5, no. 6, pp. 659-664, june 2020. [10] a. dubey and a. k. bansal, “cascaded h-bridge multilevel inverter,” international journal of control theory and applications, vol. 9, no. 7, pp. 3029-3036, 2016. [11] r. agrawal and s. jain, “comparison of reduced part count multilevel inverters (rpc-mlis) for integration to the grid,” international journal of electrical power and energy systems, vol. 84, pp. 214-224, january 2017. 22 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 12-23 [12] k. a. shilpa, m. revathi, and k. r. sudha, “design and implementation of thirty one level multilevel inverter,” international conference on smart and intelligent systems, pp. 547-555, february 2021. [13] v. varun, b. govindarajan, and s. nayak, “speed control of induction motor using fuzzy logic approach,” thesis, department of electrical engineering, national institute of technology, rourkela, odisha, 2011. [14] t. d. dongale, s. r. jadhav, s. v kulkarni, t. g. kulkarni, r. r. mudholkar, and m. d. uplane, “performance comparison of pid and fuzzy control techniques in three phase induction motor control,” international journal on recent trends in engineering and technology, vol. 7, no. 2, pp. 5-8, 2012. [15] s. pati, a. panda, and s. mohanty, “a comparative performance study of scalar controlled induction motor using pid controller and fuzzy pid controller,” international conference on circuits, power, and computing technologies, pp. 904-909, march 2014. [16] s. pati, m. patnaik, and a. panda, “comparative performance analysis of fuzzy pi, pd and pid controllers used in a scalar controlled induction motor drive,” international conference on circuits, power, and computing technologies, pp. 910-915, march 2014. [17] m. vahedpour, a. r. noei, and h. a. kholerdi, “comparison between performance of conventional, fuzzy and fractional order pid controllers in practical speed control of induction motor,” 2nd international conference on knowledge-based engineering and innovation, pp. 912-916, november 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/). 23 2___peti#9210_in press proceedings of engineering and technology innovation, vol. 21, 2022, pp. 10-19 a solar energy harvester for a wireless sensor system toward environmental monitoring saeed mohsen * department of electronics and communications engineering, al-madina higher institute for engineering and technology, giza, egypt received 01 january 2022; received in revised form 13 february 2022; accepted 14 february 2022 doi: https://doi.org/10.46604/peti.2022.9210 abstract harmful environments can cause severe health problems to individuals. thus, this study proposes a solar-powered wireless sensor system to monitor the physical parameters of an ambient environment in real-time. this system is developed based on two sensors and a nodemcu board that includes a microcontroller with a wi-fi chip. this system is built to measure the ambient temperature, relative humidity, atmospheric pressure, and ultraviolet (uv) index. the power supply of the system is a solar energy harvester, which consists of a solar cell, a dc-dc converter, and a rechargeable battery. this harvester is practically tested outdoors under direct sunlight. the proposed system experimentally consumes an average power of 40 mw over one hour, and the lifetime of this system is 123 hours in the active-sleep mode. the results demonstrate that the system can sustainably operate for monitoring the environmental data. keywords: wireless sensor system, environmental monitoring, solar energy harvester, rechargeable battery, internet of things (iot) 1. introduction in the last decade, wireless sensor systems had a robust effect in environmental monitoring applications [1-2]. these systems are extensively utilized to measure environmental parameters in real-time [3-6]. batteries are often utilized to power the sensor systems, but they have a bounded lifetime [7-8]. energy harvesting (eh) methods are utilized to extend the lifetime of the batteries [9-12]. these methods utilize various energy sources, e.g., solar [13], wind [14], thermal [15], mechanical [16], and radio frequency [17] sources. solar energy is characterized by the higher power density compared to other energy sources [18]. however, the illumination rates of solar energy can change instantaneously. therefore, energy-storage elements and dc-dc converters are needed to store and regulate the power extracted via eh methods. two energy-storage elements, i.e., a super-capacitor and a rechargeable battery, are utilized as energy storage. super-capacitors have lower energy densities than rechargeable batteries [19-20]. thus, in this work, a battery is utilized. moreover, a harvester utilizing solar energy is implemented to overcome the generic battery problem [21-26]. this proposed harvester grants the sensor system a more long-term lifetime. in previous researches, different sensor systems were presented for environmental monitoring applications [27-30]. senivasan et al. [27] implemented a harvester based on solar energy to power a wireless sensor network (wsn) mote, which included two nimh batteries, a voltage regulator, and a temperature sensor. its power consumption was 94.29 mw. moreover, it had a lifespan of 38 hours, and the solar cell was large in the area. chieng et al. [28] developed a sensor system that had a rechargeable battery with a capacitance of 1500 mah. this system had a zigbee module and a temperature sensor, consumed * corresponding author. e-mail address: g17082131@eng.asu.edu.eg proceedings of engineering and technology innovation, vol. 21, 2022, pp. 10-19 a high power of 91.41 mw, and worked for 40 hours. joris et al. [29] introduced a sensor system that had a sigfox module and a sht21 sensor. the supply was a 90 mah battery. the system consumed the power of 107.7 mw, and its lifetime was 20 hours. al rasyid et al. [30] presented a sensor network that had a zigbee module, a co sensor, and a co2 sensor. the network consumed the power of 180 mw, and its lifetime was 50 hours. this study proposes a self-sustainable wireless sensor system for the perpetual monitoring of environmental parameters. to monitor the parameters continuously, the lifetime of the proposed wireless sensor system should be prolonged. a low-power software algorithm is developed to increase the system lifetime until 123 hours, so the proposed system could work without replacing the battery. additionally, the system is powered by a perpetual solar energy harvester to reach a higher lifetime. the harvester is developed based on a rechargeable battery that has high energy density. the system is implemented with the testing outdoors and consumes 40 mw. the sensor system measures four physical parameters: ambient temperature, relative humidity, atmospheric pressure, and ultraviolet (uv) index. the data of the system are gathered by two sensors, monitored on a serial monitor application, and sent to a cloud service via wi-fi. the primary contributions of the study are as follows. (1) the study develops a self-powered wireless sensor system to continuously monitor the ambient temperature, atmospheric pressure, relative humidity, and uv index. (2) the study designs a solar energy harvester to power the system with a solar cell and a rechargeable battery, saving the power consumption of the system with sleep commands and low-power hardware components. the study is organized as follows. section 2 reviews the works related to the architectures of sensor systems. section 3 introduces the proposed architecture of the solar energy harvester and wireless sensor system. the hardware implementation of the proposed wireless sensor system with the harvester is described in section 4. the results are demonstrated in section 5 and discussed in section 6. the conclusion of the study is provided in section 7. 2. related works many wireless sensor systems utilized solar energy harvesters. for instance, capella et al. [31] utilized a 5.4 cm × 4.2 cm photovoltaic panel, with a super-capacitor of 50 f and a battery together. the power system was sufficient to prolong the lifetime of the sensor system to one year. in the work of alippi et al. [32], a maximum power point tracking (mppt) circuit was proposed for a wsn. this circuit was low-cost and had small-size dimensions. the wsn monitored the luminosity of a marine system at various depths. it also measured the temperature of the system. a power management unit was operated with rechargeable batteries, which were realized to increase the lifetime. in the work of jang et al. [33], a wsn was proposed to monitor environmental data with different structures. it used a solar cell with 21 cm × 11.4 cm as well as the batteries which had a capacitance of 10,000 mah for 8 nodes. the voltage of batteries was 4.15 v, and the voltage decreased when the solar cell did not charge the system. in the work of ingelrest et al. [34] and barrenetxea et al. [35], a wsn was presented to monitor harmful environments. a solar energy harvester was integrated per station to supply the sensor system, and a long-term system was deployed. the sensor system could last for 180 days. lazarescu et al. [36] utilized a solar energy harvester in a gateway system because this gate consumed more power compared to the sensor system. the gateway was placed inside a birdhouse. the system was tested under direct sunlight, and the lithium-ion batteries could not be replaced due to the fixing place of the birdhouse. 3. proposed architecture fig. 1 demonstrates the proposed structure of the solar energy harvester as well as the wireless sensor system. the harvester includes a solar cell to detect the sunlight irradiance and a dc-dc converter to step up the solar cell output. a rechargeable battery is used as energy storage, and a nodemcu board is used for processing the physical signals of two sensors (one for 11 proceedings of engineering and technology innovation, vol. 21, 2022, pp. 10-19 measuring the humidity, temperature, and pressure, and the other for sensing the uv index). further, wi-fi is used to transmit the environmental data to a cloud service. eventually, the environmental data of sensors are monitored on a serial monitor application. fig. 2 illustrates the schematic diagram of the harvester with the sensor system. in this figure, the utilized solar cell is mpt 3.6-75 and is connected to the input of a ltc3105 dc-dc converter to charge a 18650 lithium-ion rechargeable battery. the used sensors are bme280 and gy1145, while the utilized nodemcu board is gizwits wifi witty esp-12f. fig. 1 the proposed structure of the solar energy harvester as well as the wireless sensor system fig. 2 the schematic diagram of the harvester with the sensor system 4. implementation of the wireless sensor system 4.1. hardware implementation the setup of the wireless sensor system with the energy harvester is demonstrated in fig. 3. it is implemented using the following hardware components: solar cell, dc-dc converter, rechargeable battery, bme280 sensor, gy1145 sensor, and nodemcu board. table 1 shows the characteristics of the used components in the wireless sensor system. the total cost of the proposed wireless sensor system is usd 120. fig. 3 the setup of the wireless sensor system with energy harvester 12 proceedings of engineering and technology innovation, vol. 21, 2022, pp. 10-19 table 1 characteristics of the components used in the system component voltage current bme280 sensor 3.3 v 3.6 µ a gy1145 sensor 3.3 v 1.4 µ a nodemcu board 3.3 v 2.5 ma wi-fi chip 3.3 v 70 ma the system includes a low-power bme280 sensor, chosen from bosch ® for sensing an environment’s ambient temperature, relative humidity, and atmospheric pressure. this sensor draws a current of 3.6 µ a and can sense the ambient temperature from -40 to 85°c, the relative humidity from 0 to 100%, and the atmospheric pressure from 300 to 1100 hpa. the gy1145 sensor measures the uv index and draws a current of 1.4 µ a. the bme280 and gy1145 sensors are chosen because they are convenient for all environmental applications whether indoors or outdoors. the brain of the system is a nodemcu board from espressif systems ® , which has small dimensions of 29.9 mm × 31.5 mm. moreover, the nodemcu board is utilized for processing the environment’s data. the board consumes a low current of 2.5 ma, and 3.3 v is its operating voltage. the nodemcu board is an open-source development board and has a built-in wi-fi chip. the wi-fi chip is used to transmit the data to a cloud service. this wi-fi chip has a range of 400 m and its current consumption is 70 ma. the nodemcu board has a small size, but it consumes a high current due to its wi-fi chip. the nodemcu board is chosen because it can be easily programmed with arduino c language. in addition, this board is intended for internet of things (iot)-based applications. therefore, the proposed wireless sensor system can be used in iot applications. the harvester contains a solar cell manufactured by sundance solar ® . the dimensions of the solar cell are 7.2 cm × 6.0 cm. the cell is a thin-film and amorphous mpt 3.6-75 cell, and it gives the power of 180 mw at 1000 w/m 2 irradiance. the cell can convert light into electric power. the utilized solar cell has a maximum current and voltage of 50 ma and 3.6 v, respectively. based on the calculations made, the mpt 3.6-75 cell is chosen. the solar cell is connected in series to power a dc-dc converter, which is utilized to fix the solar cell output voltage. the converter is an ltc3105 converter. in other words, this converter charges one battery as a backup energy source. therefore, this battery will have an overall capacitance of 3800 mah and a voltage of 4.2 v, with an area of 11.7 cm 2 . a mcp1700 low dropout voltage regulator is used to fix the battery voltage to 3.3 v because the nodemcu board works at this value. the bme280 and gy1145 sensors, as well as the wi-fi chip, are fixed on the breadboard. this board has a total area of 3.0 cm × 7.0 cm. 4.2. software implementation the flowchart of the wireless sensor system is described in fig. 4. it comprises sequential steps. the bme280 and gy1145 sensors are configured first, and then i 2 c communication protocols are initialized for the mentioned sensors. then, the initialization is carried out to the wi-fi, and the nodemcu board is identified on the ubidots platform. the nodemcu board of the system works in two modes (active and sleep), so the sensor system works in the active mode. then, the sensor system verifies the connection to the wi-fi network. if the status of the connection is equal to one, the nodemcu board continuously reads the environmental data of the utilized sensors. after the data is sent to the ubidots platform for 10 seconds through the wi-fi, the sensors are turned off and the wi-fi goes into sleep. then, the nodemcu board is transferred into the sleep mode for a duration of 50 seconds. a serial monitor application is used for monitoring the data received from the system. in the active mode, the bme280 and gy1145 sensors are configured and i 2 c protocols are initialized to these sensors. the nodemcu board is also configured. 13 proceedings of engineering and technology innovation, vol. 21, 2022, pp. 10-19 fig. 4. the flowchart of the wireless sensor system 4.3. wireless communication the wi-fi chip type “esp8266mod” is utilized. the performance of the wi-fi chip depends on many factors, e.g., data rate, energy consumption, and security. the wi-fi has the power consumption of 70 ma at the active mode, and its data rate is 72.2 mbps. the wi-fi is secure in transmission for certain data from one device to others. it provides multi security modes in data encryption and authentication. the operating frequency of the wi-fi is 2.4 ghz. it is a wireless protocol convenient for wireless systems. in the proposed system, the utilized wi-fi is version 5.8, and it can save energy. the wi-fi is experimentally tested, and the measured current consumption is 0.2 ma at the sleep mode, showing that the wi-fi protocol is efficient. moreover, the utilized wi-fi is suitable for eh systems and convenient with wireless systems including sensors. the wi-fi chip is compatible with ascii commands, which are utilized in configuring communications. the utilized protocol in the wi-fi depends on the ieee 802.11 b/g/n. additionally, the data of utilized sensors connect to the nodemcu board of the wireless sensor system through i 2 c protocols. 5. results from the simulation of the solar energy harvester, fig. 5 demonstrates the i-v characteristic curve of the solar cell under direct sunlight irradiance of 1000 w/m 2 . the sunlight represents the source of illumination to validate the electrical characteristics of the utilized solar cell. it is clear that the maximum current of the solar cell is 50 ma at a voltage of 3.6 v. the output current and voltage values of the solar energy harvester are tested using two multimeters (one for the current and the other for the voltage). table 2 demonstrates the measured conversion efficiency of the solar energy harvester at various input power values. these values are measured at various irradiance levels of the sunlight using a solar power meter. these irradiance levels are 200, 400, 600, 800, and 1000 w/m 2 , and the area of the solar harvester is 0.00432 m 2 . thus, the input power values to the solar harvester are 4319, 3446, 2588, 1730, and 859 mw with conversion efficiencies of 4.63, 3.47, 4.25, 2.89, and 2.91%, respectively. the corresponding output power values of the solar harvester are measured using two multimeters. it is clear that the maximum conversion efficiency (ƞ) is 4.63%. fig. 6 illustrates the p-v characteristic curve of the solar cell under direct sunlight irradiance of 1000 w/m 2 . this figure shows that the solar cell’s maximum power (mpp) is 180 mw at 3.6 v. practically, the voltage and current of the solar cell are measured via various resistors and 2 multi-meters (one multi-meter for measuring the current and the other for the voltage). the irradiance rate is measured on a sunny day, via a solar power meter. 14 proceedings of engineering and technology innovation, vol. 21, 2022, pp. 10-19 fig. 5 i-v characteristic curve of the solar cell under 1000 w/m 2 table 2 conversion efficiency of the solar energy harvester output power (mw) 200 150 110 50 25 input power (mw) 4319 3446 2588 1730 859 conversion efficiency (ƞ %) 4.63 3.47 4.25 2.89 2.91 fig. 6 p-v characteristic curve of the solar cell under 1000 w/m 2 fig. 7 demonstrates the discharging curve of the 3800 mah rechargeable battery for the system in the active-sleep mode. in this mode, the battery takes roughly 123 hours for discharging from 4.2 to 3.3 v, so the sensor system is powered without the harvester for 123 hours. the voltage of the battery stops at 3.3 v during the discharging operation because the nodemcu board of the system works at this value. the system does not run if the voltage of the battery is less than 3.3 v. the voltage measurements for the battery are taken via a multi-meter. fig. 8 demonstrates the charging and discharging voltages from the solar cell and the battery over 539 hours. the charging experiment is executed outdoors with 1000 w/m 2 and is started at 12:00 pm on 11 june 2021 (sunny day). the discharging experiment is terminated after 539 hours. it is clear that the sensor system operates during different periods for discharging and charging the battery. the longest period for the wireless sensor system is tested to be 246 hours without being charged by the solar energy harvester. eventually, the results demonstrate that the wireless sensor system is supplied using the solar energy harvester so that it can sustainably operate. fig. 7 the discharging curve of 3800 mah battery fig. 8 the charging and discharging voltages for the battery during 539 hours 15 proceedings of engineering and technology innovation, vol. 21, 2022, pp. 10-19 fig. 9 a screenshot of the serial monitor application showing the monitored environmental data fig. 10 relative humidity measured by bme280 fig. 9 reveals a screenshot of the displayed environmental data on a serial monitor application, which is conducted for monitoring the status of an ambient environment during the active period (600 seconds/hour) of the system. the monitored environmental data are instantaneously visualized on the application for every active period of the system. the environmental data (i.e., relative humidity, ambient temperature, atmospheric pressure, and uv index) can change continuously. at first, the uv index of the environment is shown to be 0.00, the relative humidity is 31.79%, the ambient temperature is monitored at 36.56°c, and the atmospheric pressure is 1008.28 hpa. then, the uv index, relative humidity, ambient temperature, and atmospheric pressure are observed to be 0.00, 31.71%, 36.60°c, and 1008.26 hpa, respectively. from the serial monitor application, one can obtain the data of an ambient environment over time. fig. 10 demonstrates the relative humidity of an environment during the active time of bme280. the relative humidity is measured under various physical conditions of an environment. during the measurements, the wireless sensor system samples the bme280 sensor at a sampling rate of 100 hz. it can be seen that the relative humidity continuously changes from 0 to 10 seconds and the maximum humidity percentage is 64. bme280 can sense the relative humidity from 0 to 100%. the humidity monitoring is achieved by the serial monitor application of arduino, and then the results are saved in an excel file. 6. discussion table 3 illustrates the comparison of recent wireless sensor systems. in the proposed work, the developed system has 2 sensors (bme280 and gy1145), whereas the systems in [27-29] have only one sensor. the number of physical parameters of the proposed work is 4, while the system parameters in [27-29] are fewer. the wi-fi technology utilized in this study has a higher range than the technologies in [27-28]. additionally, the lifetime of the proposed system is the longest compared to [26-28], and the power consumption of the proposed system is less than [27-29]. the proposed system consumes 40 mw. this value is achieved via the used microcontroller, sensors, and the developed algorithm. the microcontroller is operated in two modes (sleep and active), while the sensors are switched off sometimes to save power. therefore, the proposed wireless sensor system has lower power consumption than [27-30], and the lifespan of the proposed work is longer than [27-30]. it can be seen that there is a trade-off among the type of energy storage, the amount of environmental data, the value of power consumption, and the type of the wireless technology chosen when developing the wireless sensor system in this study. 16 proceedings of engineering and technology innovation, vol. 21, 2022, pp. 10-19 table 3 comparison of recent wireless sensor systems reference [27] [28] [29] [30] the proposed work sensors temperature temperature sht21 tgs2442 for co and tgs4161 for co2 bme280 and gy1145 numbers of physical parameters one one two two four (relative humidity, ambient temperature, atmospheric pressure, and uv index) wireless technology micaz 2.4 ghz 75 m zigbee 2.4 ghz 100 m sigfox 2.4 ghz 10 km zigbee 2.4 ghz 100 m wi-fi 2.4 ghz 400 m lifetime of system (hrs) 38 40 20 50 perpetual, and 123 without a harvester power consumption (mw) 94.29 91.41 107.7 180 40 7. conclusions this study introduces a perpetual solar energy harvester to supply a wireless sensor system for environmental monitoring. the solar energy harvester is designed to increase the lifetime of the wireless sensor system from 2 to 123 hours. the power consumption of the system is 40 mw. the sensors’ data of the system are instantaneously monitored on the serial monitor application. the system is tested experimentally to measure the physical parameters (i.e., ambient temperature, relative humidity, atmospheric pressure, and uv index) of an ambient environment in a continuous mode. finally, the designed solar energy harvester enables the sustainable operation of the proposed monitoring system in sunny environments. in the future, the author will consider a battery that has a capacity higher than 3800 mah. further, the power performance of the solar cell will be tested at different bending angles. the testing of the cell is important to verify the sustainability of the solar energy harvester. additionally, the product of the wireless sensor system could be manufactured using a printed circuit board. one could apply the wireless sensor system for pollution monitoring. also, one could use the sensors under humid environments, e.g., precipitation. conflicts of interest the author declares no conflict of interest. references [1] f. wu, c. rüdiger, and m. r. yuce, “real-time performance of a self-powered environmental iot sensor network system,” sensors, vol. 17, no. 2, article no. 282, february 2017. [2] t. wu, f. wu, j. m. redoute, and m. r. yuce, “an autonomous wireless body area network implementation towards iot connected healthcare applications,” ieee access, vol. 5, pp. 11413-11422, june 2017. [3] m. lapinski, c. b. medeiros, d. m. scarborough, e. berkson, t. j. gill, t. kepple, et al., “a wide-range, wireless wearable inertial motion sensing system for capturing fast athletic biomechanics in overhead pitching,” sensors, vol. 19, no. 17, article no. 3637, august 2019. [4] m. sadrawi, c. h. lin, y. t. lin, y. hsieh, c. c. kuo, j. c. chien, et al., “arrhythmia evaluation in wearable ecg devices,” sensors, vol. 17, no. 11, article no. 2445, october 2017. [5] l. xie, p. chen, s. chen, k. yu, and h. sun, “low-cost and highly sensitive wearable sensor based on napkin for health monitoring,” sensors, vol. 19, no. 15, article no. 3427, august 2019. [6] p. f. silva, e. e. c. santana, m. s. s. pinto, r. c. s. freire, m. a. oliveira, g. fontaglland, et al., “flexible wearable pre-fractal antenna for personal high-temperature monitoring,” wireless personal communications, vol. 114, no. 3, pp. 1983-1998, may 2020. 17 proceedings of engineering and technology innovation, vol. 21, 2022, pp. 10-19 [7] a. decker, “solar energy harvesting for autonomous field devices,” iet wireless sensor systems, vol. 4, no. 1, pp. 1-8, march 2014. [8] a. dionisi, d. marioli, e. sardini, and m. serpelloni, “autonomous wearable system for vital signs measurement with energy-harvesting module,” ieee transactions on instrumentation and measurement, vol. 65, no. 6, pp. 1423-1434, february 2016. [9] t. wu, m. s. arefin, d. shmilovitz, j. m. redoute, and m. r. yuce, “a flexible and wearable energy harvester with an efficient and fast-converging analog mppt,” ieee biomedical circuits and systems conference, pp. 336-339, october 2016. [10] g. mestre, a. ruano, h. duarte, s. silva, h. khosravani, s. pesteh, et al., “an intelligent weather station,” sensors, vol. 15, no. 12, article no. 31005, december 2015. [11] r. hamid and m. r. yuce, “a wearable energy harvester unit using piezoelectric-electromagnetic hybrid technique,” sensors and actuators a: physical, vol. 257, pp. 198-207, april 2017. [12] m. v. ramesh, “design, development, and deployment of a wireless sensor network for detection of landslides,” ad hoc networks, vol. 13, pp. 2-18, february 2014. [13] s. mohsen, a. zekry, k. youssef, and m. abouelatta, “a self-powered wearable wireless sensor system powered by a hybrid energy harvester for healthcare applications,” wireless personal communications, vol. 116, no. 4, pp. 3143-3164, september 2020. [14] a. ce-ce and s. xiao-xia, “wireless sensor network in wind and solar hybrid street lamp application,” 27th chinese control and decision conference, pp. 3335-3339, may 2015. [15] p. c. dias, f. j. o. morais, m. b. d. m. frança, e. c. ferreira, a. cabot, and j. a. s. dias, “autonomous multisensor system powered by a solar thermoelectric energy harvester with ultralow-power management circuit,” ieee transactions on instrumentation and measurement, vol. 64, no. 11, pp. 2918-2925, november 2015. [16] m. wijesundara, c. tapparello, a. gamage, y. gokulan, l. gittelson, t. howard, et al., “design of a kinetic energy harvester for elephant mounted wireless sensor nodes of jumbonet,” ieee global communications conference, pp. 1-7, december 2016. [17] r. wallace, “monopole pcb antenna with single or dual band option,” https://www.ti.com/lit/an/swra227e/swra227e.pdf?ts=1647261207955&ref_url=https%253a%252f%252fwww.google. com%252f, march 2013. [18] f. wu, c. rudiger, and m. r yuce, “design and field test of an autonomous iot wsn platform for environmental monitoring,” 27th international telecommunication networks and applications conference, pp. 206-211, november 2017. [19] m. f. othman and k. shazali, “wireless sensor network applications: a study in environment monitoring system,” procedia engineering, vol. 41, pp. 1204-1210, december 2012. [20] a. mathieu, j. kern, and g. ferré, “internet of wine: a low-cost solution for stock management improvement,” 25th ieee international conference on electronics, circuits, and systems, pp. 189-192, december 2018. [21] s. mcgarry and c. knight, “development and successful application of a tree movement energy harvesting device to power a wireless sensor node,” sensors, vol. 12, no. 9, pp. 12110-12125, september 2012. [22] s. climent, a. sánchez, s. blanc, j. v. capella, and r. ors, “wireless sensor network with energy harvesting: modeling and simulation based on a practical architecture using real radiation levels,” concurrency and computation: practice and experience, vol. 28, no. 6, pp. 1812-1830, october 2016. [23] g. manes, g. collodi, r. fusco, l. gelpi, and a. manes, “a wireless sensor network for precise volatile organic compound monitoring,” international journal of distributed sensor networks, vol. 8, no. 4, article no. 820716, april 2012. [24] s. mohsen, a. zekry, k. youssef, and m. abouelatta, “an autonomous wearable sensor node for long-term healthcare monitoring powered by a photovoltaic energy harvesting system,” international journal of electronics and telecommunications, vol. 66, no. 2, pp. 267-272, june 2020. [25] x. silvani, f. morandini, e. innocenti, and s. peres, “evaluation of a wireless sensor network with low cost and low energy consumption for fire detection and monitoring,” fire technology, vol. 51, no. 4, pp. 971-993, october 2014. [26] y. j. wong, r. nakayama, y. shimizu, a. kamiya, s. shen, i. z. m. rashid, et al., “toward industrial revolution 4.0: development, validation, and application of 3d-printed iot-based water quality monitoring system,” journal of cleaner production, vol. 324, article no. 129230, november 2021. [27] s. senivasan, m. drieberg, b. s. m. singh, p. sebastian, and l. h. hiung, “an mppt micro solar energy harvester for wireless sensor networks,” ieee 13th international colloquium on signal processing and its applications, pp. 159-163, march 2017. 18 proceedings of engineering and technology innovation, vol. 21, 2022, pp. 10-19 [28] l. j. chien, m. drieberg, p. sebastian, and l. h. hiung, “a simple solar energy harvester for wireless sensor networks,” 6th international conference on intelligent and advanced systems, pp. 1-6, august 2016. [29] l. joris, f. dupont, p. laurent, p. bellier, s. stoukatch, and j. m. redouté, “an autonomous sigfox wireless sensor node for environmental monitoring,” ieee sensors letters, vol. 3, no. 7, pp. 1-4, july 2019. [30] m. u. h. al rasyid, i. u. nadhori, a. sudarsono, and y. t. alnovinda, “pollution monitoring system using gas sensor based on wireless sensor network,” international journal of engineering and technology innovation, vol. 6, no. 1, pp. 79-91, january 2016. [31] j. v. capella, a. bonastre, r. ors, and m. peris, “in line river monitoring of nitrate concentration by means of a wireless sensor network with energy harvesting,” sensors and actuators b: chemical, vol. 177, pp. 419-427, february 2013. [32] c. alippi, r. camplani, c. galperti, and m. roveri, “a robust, adaptive, solar-powered wsn framework for aquatic environmental monitoring,” ieee sensors journal, vol. 11, no. 1, pp. 45-55, january 2011. [33] s. jang, h. jo, s. cho, k. mechitov, j. a. rice, s. h. sim, et al., “structural health monitoring of a cable-stayed bridge using smart sensor technology: deployment and evaluation,” smart structures and systems, vol. 6, no. 5-6, pp. 439-459, july 2010. [34] f. ingelrest, g. barrenetxea, g. schaefer, m. vetterli, o. couach, and m. parlange, “sensorscope: application-specific sensor network for environmental monitoring,” acm transactions on sensor networks, vol. 6, no. 2, article no. 17, january 2010. [35] g. barrenetxea, f. ingelrest, g. schaefer, and m. vetterli, “wireless sensor networks for environmental monitoring: the sensorscope experience,” ieee international zurich seminar on communications, pp. 98-101, march 2008. [36] m. t. lazarescu, “design and field test of a wsn platform prototype for long-term environmental monitoring,” sensors, vol. 15, no. 4, pp. 9481-9518, april 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/). 19 6___peti#7522__53-61 proceedings of engineering and technology innovation, vol. 19, 2021, pp. 53-61 tractor-implement tillage depth control using adaptive neuro-fuzzy inference system (anfis) aristide timene 1,* , ndjiya ngasop 2 , haman djalo 1 1 department of physics, faculty of sciences, university of ngaoundere, cameroon 2 department of electrical engineering, energy and automation, national school of agro-industrial sciences, university of ngaoundere, cameroon received 20 april 2021; received in revised form 10 may 2021; accepted 11 may 2021 doi: https://doi.org/10.46604/peti.2021.7522 abstract this study presents a design of an adaptive neuro-fuzzy controller for tractors’ tillage operations. since the classical controllers allows plowing depth errors due to the variations of lands structure, the use of the combined neural networks and fuzzy logic methods decreases these errors. the proposed controller is based on adaptive neuro-fuzzy inference system (anfis), which permits the generation of fuzzy rules to cancel the nonlinearity and disturbances on the implement. the design and simulations of the system, which consist of a hitch-implement mechanism, an electro-hydraulic actuator, and a neuro-fuzzy controller, are conducted in solidworks and matlab software. the performance of the proposed controller is analyzed and is contrasted with a proportional integral derivative (pid) controller. the obtained results show that the neuro-fuzzy controller adapts perfectly to the dynamics of the system with rejection of disturbances. keywords: tractor-implement, plowing depth, neuro-fuzzy controller 1. introduction on farmlands, a tractor’s electro-hydraulic hitch system lowers, holds, or lifts implements into the ground depending on the desired working depth [1]. to command the hydraulic hitch system, modern tractors are supplied with position control and draft force control system. however, variations in soil conditions (e.g., texture and organic matter) causes plowing depth errors. this situation directly decreases root penetration and plant growth. for a better approximation of land structures, it is recommended to mix the position and draft control [2-3]. in [2], a combined position-draft control was used to adjust the tillage depth of the implement. in their study, the participation ratio of the position target and the draft force target are mixed before being processed for regulation. then, a fuzzy controller sends an appropriate signal to the electro-hydraulic system to cancel the plowing depth error. this strategy makes it possible to limit the changes in plowing depth due to the changes in soil resistance. shafaei, loghavi, and kamgar [3] also prove the capacity of fuzzy controller to decrease plowing depth errors, driving wheel slip, and fuel consumption in tillage operations. furthermore, the fuzzy logic controller is also used in [4] to compensate nonlinearities and reduce pressure fluctuation in tractor’s hydraulic hitch. if fuzzy logic is often used in the control of the electro-hydraulic system of the tractor to overcome nonlinearities, it nevertheless has limits, in particular on the accuracy of the information expressed in natural language after mixing position and draft. to overcome this drawback, the current trend is to use hybrid architectures to take advantage of fuzzy logic and neural networks. the use of adaptive neuro-fuzzy inference system (anfis) offers the possibility of modeling a priori knowledge and linguistic decision rules obtained by experts in the field [5-6]. * corresponding author. e-mail address: aristidetimene@yahoo.com proceedings of engineering and technology innovation, vol. 19, 2021, pp. 53-61 indeed, much research has proven the usefulness of the anfis method, developed by jang [7] for the control of systems. the neuro-fuzzy controller was used in [8] to ensure that one leg of a quadruped robot follows the desired trajectory. in [9], the neuro-fuzzy controller was used to stabilize the climatic variables inside greenhouse system at required levels for crops development. an adaptive neuro-fuzzy controller was designed in [10], for the speed control of permanent magnet synchronous motor. the proposed controllers were satisfactory in terms of easy settling time, zero peaks overflow, and high robustness. in this work, a controller combining two intelligent techniques (fuzzy logic and neural networks) is designed to decrease tractors’ plowing depth errors. this study comprises six sections. first, the hitch-implement model is presented in section 2. then, section 3 describes the electro-hydraulic actuator and section 4 presents the proposed controller. to show the effectiveness of the enhanced neuro-fuzzy control strategy, the simulations examples are provided and commented in section 5. finally, a concluding summary ends the study. 2. modelling of hitch-implement mechanism a tractor’s three-point hitch mechanism is modeled in solidworks software. the kinematics of the up and down movement of the system is described in [11]. the common mechanism of the system is shown in fig. 1. the tractor model is category 2, which has a maximum lifting capacity of 3546 kg with a power of 90 hp [12]. the dimensions of the model are listed in table 1. for a more realistic simulation, the matlab/simscape multibody model of the tractor’s hitch system is imported from solidworks software with the physical properties to calculate the movement. the matlab model of the hitch-implement mechanism is given in fig. 2. fig. 1 computer-aided design (cad) model of the tractor’s three-point hitch mechanism table 1 parameters of the three-point hitch linkages part name measure lift arm length 295 mm lower link length 946 mm lift rod length 765 mm vertical length from lift arm pivot point to upper link pivot point 130 mm upper link length 650 mm mast height 610 mm vertical length from upper link pivot point to lower link pivot point 460 mm external lift cylinder length 380-560 mm stroke of the cylinder 180 mm 54 proceeding of engineering and technology innovation, vol. 19, 2021, pp. 53-61 fig. 2 hitch-implement mechanism in matlab in this work, two implement have been used: chisel plow and moldboard plow. the parameters listed in table 2 are provided from the data in [13]. the force applied to the implement on the ground is the disturbance of the system. it is expressed by [13]: 2 ( ) s e d f t a b v c v w h= × + × + × × × (1) where f is the implement draft (n); ts characterizes a dimensionless soil texture adjustment parameter for a category of soil s; a, b, and c are machine specific parameters; v is the tractor field speed (km/h); we is the machine width (m) or number of rows; hd is the tillage depth (cm). table 2 parameters of implements and field operations parameters chisel plow moldboard plow plowing depth (cm) 30 10 weight (kg) 380 200 a 107 116 b 6.3 0 c 0 2.3 soil type clay fine working width (m) 2.1 1.04 tractor speed (km/h) 7 5 draft force (n) 8095 2429 draft range (%) 50 40 3. modelling of hydraulic actuator the actuator of the tractor is its electro-hydraulic system (fig. 3), which allows the movement of hitch-implement mechanism. it consists of three main parts, namely the hydraulic pump, proportional valve, and the cylinder. the pump is operated by the tractor engine running at 1500 rpm. it provides a flow rate of 40 l/min. a pressure relief valve assembled on the main distribution block limits the pressure to 194 bars. the two-module servo valve, type bosch ehr5 [14], directs the flow of fluid from the pump to the cylinder. ports b and f are the actuator outputs to the hitch-implement mechanism. 55 proceedings of engineering and technology innovation, vol. 19, 2021, pp. 53-61 according to the manufacturer [14], the lifting valve control signal must be within an operating range of 1 to 3.35 a and the lowering valve control signal must be within an operating range of -1 to -3.35 a. to lift the implement, a positive signal is sent to the solenoid of the valve. likewise, a negative signal is sent to the solenoid of the valve to lower the implement. the cylinder has a total stroke of 180 mm. its base is fixed on the main frame of the tractor, and its rod is on the lift arm. the position of the lift arm measured by a sensor is an input to the controller. the overall diagram of the tractor’s tillage depth control system is shown in fig. 4. fig. 3 electro-hydraulic actuator in matlab/simscape fluids fig. 4 block diagram of the tillage depth control system 4. design of controller 4.1. mixed position-draft control strategy the combination of the position control and draft control allows dual goals, including tillage depth and tillage resistance. the target of position control is the depth parameter. the depth deviation is written as: d m e h h= − (2) where e is the error in depth, hd is the desired depth, and hm is the measured depth. the draft control is used to modify tillage depth according to the soil resistance, using eq. (1): 2 ( ) d s e f h t a b v c v w = × + × + × × (3) 56 proceeding of engineering and technology innovation, vol. 19, 2021, pp. 53-61 the depth deviation can be expressed by using eq. (2): 2 ( ) d m s e f f e t a b v c v w − = × + × + × × (4) where e is the error in depth, fd is the desired draft force, and fm is the measured draft force. here, the participation ratio of the position target and the draft is adjusted according to the mix coefficient (m) proposed in [2]. the tillage depth deviation in the mixed control can be written as follows: ( ) 2 (1 ) ( [ ] ) d m d m s e f f e m h h m t a b v c v w − = + − − × + × + × × (5) when m = 0, the control mode is position; when m = 1, the mode is draft control; when 0 < m <1, the control mode is mixed. the schematic diagram of the proposed controller is illustrated in fig. 5. the user from the tractor panel enters the parameters of the desired depth hd, coefficient m, tractor speed v, and type of implement. in the hitch-implement mechanism, the signals are measured by a force sensor and a position sensor. the data acquisition and calculation module collects the above signals and calculates the depth deviation e. then, the neuro-fuzzy controller calculates the command value of the electro-hydraulic proportional valve which adjusts the plowing depth of the implement. fig. 5 schematic diagram of the neuro-fuzzy controller with mix draft-position strategy 4.2. the proposed adaptive neuro-fuzzy controller design in matlab software, the anfis editor graphical user interface is available in neuro-fuzzy designer app. firstly, in simulink environment, a proportional integral derivative (pid) controller is designed and applied to the system in order to obtain the input and output data sets for training anfis. oksanen, eriksson, and mikkola [15] proposed the practical method for tuning a pid hitch controller. two modes of control are considered: upwards and downwards with different parameters. these parameters are summarized in table 3. table 3 parameters of the pid controller parameter upward downward kp 1.22 0.612 ki 0 0 kd 0 0.174 the data set consists of the error in depth (e), the change in error (ce), and the controller output. using the given input/output data set, the toolbox constructs a fuzzy inference system (fis) whose membership function parameters are adjusted using a hybrid least-squares and back-propagation gradient descent algorithm method. this allows the fuzzy systems 57 proceedings of engineering and technology innovation, vol. 19, 2021, pp. 53-61 to learn from the data they are modeling. after several training processes with 20 epochs, a suitable neuro-fuzzy controller is obtained for tillage operations. a computing flowchart of anfis is provided in fig. 6. the data set is a three-dimensional vector. 70% of the data are used for training, 15% are used for testing, and 15% are used for verification. the controller has seven membership functions for the two inputs with 49 rules, and the gaussian combination membership functions have been chosen for fuzzification. after training, the error tolerance is close to zero as shown in fig. 7, and the membership functions using grid method are shown in fig. 8. note that the input data have been normalized in the range of -1 to 1. fig. 6 computing flowchart of the anfis model fig. 7 training error of the proposed anfis (a) error (e) (b) change in error (ce) fig. 8 membership functions after training 5. results and discussion the graphs in figs. 9-11 show the positions of the chisel plow under pid and neuro-fuzzy control, for different value of mix coefficient (m = 0, m = 0.25, and m = 0.5). at the instants t = 0s and t = 55s, the implement starts from the upper position. the mechanism takes some time to attend the desired upper position (20 cm) because lifting speed of the hitch mechanism is limited by the hydraulic power system. then, a lowering signal is applied at t = 5s, and the implement penetrates progressively 58 proceeding of engineering and technology innovation, vol. 19, 2021, pp. 53-61 under the soil until the desired depth (30 cm). the more the value of the coefficient m increases, the implement becomes more sensitive to soil resistance. this is beneficial for preventing tractor slippage and reducing fuel consumption. however, for large values of the mixing coefficient (m > 0.5), the implement moves away from the desired depth. (a) pid (b) neuro-fuzzy fig. 9 position of the chisel plow for m = 0 (a) pid (b) neuro-fuzzy fig. 10 position of the chisel plow for m = 0.25 (a) pid (b) neuro-fuzzy fig. 11 position of the chisel plow for m = 0.5 the graphs in fig. 12 show the positions of the moldboard under pid and neuro-fuzzy control, for different values of the mix coefficient (m = 0, m = 0.25, and m = 0.5). between t = 10s and t = 60s, the implement follows the desired depth (10 cm). note that for a small plowing depth, increasing the m coefficient does not have a significant effect on the desired depth. this is due to the fact that, at shallow plowing depths, soil resistances are reduced. (a) pid (b) neuro-fuzzy fig. 12 position of the moldboard plow 59 proceedings of engineering and technology innovation, vol. 19, 2021, pp. 53-61 in this work, the well-known integral time absolute error (itae) objective function is used to evaluate the proposed controller. the itae criterion is defined by: ( ) 0 itae t e t dt ∞ = ∫ (6) table 4 performance of the different controllers implement controller m value depth mean min max standard deviation itae chisel plow neuro-fuzzy m = 0 29.88 29.07 30.85 0.25 0.32 m = 0.25 28.68 27.01 30.23 0.54 2.02 m = 0.5 28.29 26.43 30.28 0.64 2.45 pid m = 0 29.47 28.92 31.61 0.45 0.76 m = 0.25 27.96 26.9 30.70 0.77 2.27 m = 0.5 27.67 26.22 30.96 0.78 2.71 moldboard plow neuro-fuzzy m = 0 9.94 9.66 10.18 0.09 0.11 m = 0.25 9.84 9.51 10.23 0.11 0.41 m = 0.5 9.69 9.23 10.08 0.17 0.65 pid m = 0 9.93 9.28 10.52 0.20 0.23 m = 0.25 9.83 9.15 10.52 0.21 0.63 m = 0.5 9.68 8.95 10.34 0.25 0.86 table 4 shows the comparative analysis between the proposed neuro-fuzzy controller and the pid controller. from table 4, the following remarks and interpretations are deduced. for different values of m, in the case of neuro-fuzzy controller, the depth mean value is near to the desired depth compared to the pid controller. for the chisel plow, the value of itae is 0.32, for m = 0, in the case of the neuro-fuzzy controller; which is almost half of the pid controller. in addition, the standard deviations on the desired depth for different m values are shown in table 4. it appears that, the standard deviation values for the proposed controller are lower than those obtained by the pid controller. these parameters shows that the system offers a great robustness toward the load perturbations and a better stability using the neuro-fuzzy controller rather than the pid controller. 6. conclusions this research presented an adaptive neuro-fuzzy controller for tractor’s tillage operations. the design of this controller was carried out under matlab / neuro-fuzzy, after several attempts to arrive at an optimal architecture. firstly, a cad model of the three-point hitch system was designed under solidworks software. after that, the designed model was exported under matlab/simscape multibody. then, an electro-hydraulic valve was simulated under matlab /simscape fluids to actuate the hitch-implement mechanism. finally, the proposed controller was used in the system for the control of two implements hitched to the tractor. the performance index on itae indicates good responses in the case of neuro-fuzzy tillage depth controller which can replace other controllers. in the future, the proposed controller can be implemented in an agricultural tractor. conflicts of interest the authors declare no conflict of interest. references [1] b. bhondave, t. ganesan, n. varma, r. renu, and n. sabarinath, “design and development of electro hydraulics hitch control for agricultural tractor,” sae international journal of commercial vehicles, vol. 10, no. 1, pp. 405-410, 2017. [2] j. han, c. xia, g. shang, and x. gao, “in-field experiment of electro-hydraulic tillage depth draft-position mixed control on tractor,” iop conference series: materials science and engineering, vol. 274, no. 1, 012028, 2017. [3] s. m. shafaei, m. loghavi, and s. kamgar, “a practical effort to equip tractor-implement with fuzzy depth and draft control system,” engineering in agriculture, environment, and food, vol. 12, no. 2, pp. 191-203, july 2019. 60 proceeding of engineering and technology innovation, vol. 19, 2021, pp. 53-61 [4] c. liu, j. zhao, j. gu, y. du, z. li, z. zhu, et al., “pressure control algorithm based on adaptive fuzzy pid with compensation correction for the tractor electronic hydraulic hitch,” applied sciences, vol. 10, no. 9, 3179, may 2020. [5] b. haznedar and a. kalinli, “training anfis structure using simulated annealing algorithm for dynamic systems identification,” neurocomputing, vol. 302, pp. 66-74, august 2018. [6] s. subha and s. nagalakshmi, “design of anfis controller for intelligent energy management in smart grid applications,” journal of ambient intelligence and humanized computing, vol. 11, no. 6, pp. 1-11, june 2020. [7] j. s. jang, “anfis: adaptive-network-based fuzzy inference system,” ieee transactions on systems, man, and cybernetics, vol. 23, no. 3, pp. 665-685, may-june 1993. [8] v. bakırcıoğlu, m. a. şen, and m. kalyoncu, “adaptive neural-network based fuzzy logic (anfis) based trajectory controller design for one leg of a quadruped robot,” 5th international conference on mechanics and control engineering, december 2016, pp. 82-85. [9] h. oubehar, a. selmani, a. ed-dahhak, a. lachhab, m. e. h. archidi, and b. bouchikhi, “anfis-based climate controller for computerized greenhouse system,” advances in science technology and engineering systems journal, vol. 5, no. 1, pp. 8-12, january 2020. [10] r. shanthi, s. kalyani, and p. m. devie, “design and performance analysis of adaptive neuro-fuzzy controller for speed control of permanent magnet synchronous motor drive,” soft computing, vol. 25, no. 2, pp. 1519-1533, january 2021. [11] h. bentaher, e. hamza, g. kantchev, a. maalej, and w. arnold, “three point hitch mechanism instrumentation for tillage power optimization,” biosystems engineering, vol. 100, no. 1, pp. 24-30, may 2008. [12] agricultural wheeled tractors-rear mounted three-point linkage—categories 1n, 1, 2n, 2, 3n, 3, 4n, and 4, gost iso 730-2019, 2020. [13] agricultural machinery management data, asae d497.7, 2011. [14] bosch rexroth ag, “hitch control valves ehr5-oc, ehr5-ls, ehr23-em2 re,” https://www.naptechniek.nl/images/catalogus/pdf/ehr23_re66125_2013-07.pdf, august 19, 2020 [15] l. eriksson, t. oksanen, and k. mikkola, “pid controller tuning rules for integrating processes with varying time-delays,” journal of the franklin institute, vol. 346, no. 5, pp. 470-487, june 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/). 61  proceedings of engineering and technology innovation , vol. 4, 2016, pp. 43 45 43 structural analysis of existing multi-speed transmission hubs of bicycles yi-chang wu * department of mechanical engineering, national yunlin university of science & technology, yunlin, taiwan. received 01 february 2016; received in revised form 10 march 2016; accept ed 02 april 2016 abstract this paper presents the analysis of topological structure of multi-speed transmission hubs for bicycles . three existing transmission hubs with 8-speed, 11-speed and 12-speed are introduced. the topological structure of each gear mechanism, which is the main body of a transmission hub, is analyzed. the degrees -of-freedom (dof) of the transmission hub is calculated by the kutzbach’s equation. the related clutching sequence table for each transmission hub is also listed . keywor ds : structural analysis, multi-speed, transmission hub, bicycle 1. a sturmey-archer 8-speed transmission hub (a) cross -section abc abc 1 234 91011 678 5 第一組第二組第三組 inputoutput (b) sketch diagram (c) clutching sequence table fig. 1 sturmey-archer 8-speed transmission hub figs. 1(a) and 1(b) show the cross -section and sketch diagram of the sturmey-archer 8-speed transmission hub [1], respectively. it is a gear mechanism consists of 11 links, 6 gear pairs and 10 turning pairs. according to the kutzbach’s criterion, the degrees -of-freedom (dof) of this gear mechanism is: dof = 3(n-1)-2j1-1j2 = 3(11-1)-2*10-1*6 = 4. because it is a 4-dof mechanism, four constraints are required for precise controlling this mechanism. except for an input constraint, three additional constraints should be provided, which can be achieved by controlling the engagement of mechanical clutches within the gear mechanism. as can be seen in the clutching sequence table shown in fig. 1(c), three mechanical c lutches are activated at each gear stage. 2. a shimano 11-speed transmission hub figs. 2(a) and 2(b) show the cross -section and sketch diagram of the shimano 11-speed transmission hub [2], respectively. it consists of three sets of compound planetary gear trains, which is a 12-link gear mechanism with 7 gear pairs and 11 turning pairs. according to the kutzbach’s criterion, the dof of this gear mechanism is: dof = 3(n-1)-2j1-1j2 = 3(12-1)-2*11-1*7 = 4. for this * corresponding aut hor. email: wuyc@yuntech.edu.tw proceedings of engineering and technology innovation , vol. 4, 2016, pp. 43 45 44 copyright © taeti transmission hub, 8 mechanical clutches are installed for controlling the power-flow path at each gear stage. fig. 2(c) depicts the clutching sequence table of this transmission hub. (a) cross -section abc ab c d 12345 6 7 8 9 10 11 12 第一組第二組第三組 inputoutput e (b) sketch diagram (c) clutching sequence table fig. 2 shimano 11-speed transmission hub 3. a sachs 12-speed transmission hub figs . 3(a ) and 3(b) show the cross -section and sketch diagram of the sachs 12-speed transmission hub [3], respectively. it consists of two sets of compound planetary gear trains as transmission units and one basic planetary gear train as a d ifferentia l unit. th is is a 14-link gear mechanis m with 9 gear pairs and 13 turning pairs. according to the kutzbach’s criterion, the dof of this gear mechanis m is: dof = 3(n-1)-2j1-1j2 = 3(14-1)-2* 13-1*9 = 4. for this transmission hub, 8 mechanical clutches are installed for controlling the power-flo w paths of the two sets of transmission units. fig. 3(c ) shows the clutching sequence table of this transmission hub. (a) cross -section 23456 9 7 8 14 1 12 13 11 output a bc d abcde input 10 \ (b) sketch diagram (c) clutching sequence table fig. 3 sachs 12-speed transmission hub 4. conclusions based on the above analysis, we can conclude that the ma in body of a mult i-speed transmission hub used for b icycles is usually a comple x planetary gear train for provid ing with several forwa rd gears. mechanical clutches, proceedings of engineering and technology innovation , vol. 4, 2016, pp. 43 45 45 copyright © taeti especially for one-way c lutches, are emp loyed on coaxia l links, i.e., the sun gears, carriers and ring gears, for correctly guiding the powe r-flo w path at each gear stage. the number of activated clutches at each gear stage is related to the dof of the planetary gear mechanism. acknowledgment the author is grateful to the ministry of science & technology (taiwan, r.o.c) for supporting this research under grants no. most 104-2628-e-224-001-m y3. references [1] ricke ls & sram, a mu lti-speed hub gear, world patent, 01/92094, 2001. [2] shimano inc., bicycle hub transmission, u. s. patent, 2009/0023542, 2009. [3] m. gerhard, mult i-speed hub for bicycles, u. s. patent, 5527230, 1996. (microsoft word peti#5273 20230802 \(way, \256\332\276\332\257\310\245\273 1\).docx) proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx english language proofreader: yen-chun hsieh integration of membrane bioreactor and reverse osmosis for textile wastewater treatment and reclamation: a pilot-scale study po-hsun lin1,*, pin-hui lee2, emily k. kin3 1graduate institute of environmental engineering, national central university, taoyuan, taiwan 2eclean-tech enterprise co., ltd, taoyuan, taiwan 3department of environmental science and engineering, northwestern university, evanston, il, usa received 13 april 2023; received in revised form 27 june 2023; accepted 28 june 2023 doi: https://doi.org/10.46604/peti.2023.5273 abstract membrane bioreactor (mbr) technology, a combination of traditional activated sludge and membrane filtration, has been widely used for industrial wastewater treatment and reclamation. this paper highlights a pilot-scale mbr system treating textile wastewater from a textile factory in taiwan. over 7 months of continuous operation, the average mbr influent chemical oxygen demand (cod) is 332 mg/l, and the average effluent cod is 38 mg/l, which results in approximately 88% cod removal. a reverse osmosis (ro) module is installed after 2 months of mbr operation and uses the mbr permeate as its influent. the ro produces pure water with average cod, conductivity, and color of 7 mg/l, 16 μs/cm, and 7 pt-co, respectively. the ro permeate is suitable for reuse in manufacturing processes, and the ro membrane shows stable performance with tmp, which is less than or equal to 0.5 kg/cm2 during the test. the study demonstrates the great feasibility of mbr combined with ro for treating and reclaiming textile wastewater. keywords: membrane bioreactor, textile wastewater, reverse osmosis, transmembrane pressure 1. introduction the textile processing industry is considered a water-intensive sector as it uses large quantities of water during production processes including scouring, bleaching, dyeing, printing, and finishing. it is reported that more than 100,000 commercial textile dyes are available in the market and approximately 700,000-1,000,000 tons of dyes are produced while 280,000 tons are discharged annually from the textile industry to the global environment [1]. the composition of textile wastewater is highly complex and depends on various factors, including the type of fabric, dyeing method, and chemicals used in the process [2]. textile wastewater is often rich in color and of extreme ph with various toxic and recalcitrant organic compounds [2]. thus, several countries have introduced more stringent discharge limits for textile wastewater. stricter regulations are compelling treatment plants to upgrade the existing wastewater treatment systems or adopt new treatment technologies. furthermore, the reclamation of purified effluent will become increasingly relevant due to water scarcity, raising water costs, and the preservation of natural water resources [3]. a membrane bioreactor (mbr) is a hybrid process combining activated sludge units and membrane filtration. the degradation of pollutants occurs inside the bioreactor, while the separation of the treated wastewater from the activated sludge is completed in a membrane module. the small pore size of the membrane can retain a high concentration of microorganisms in the bioreactor, and produce very high-quality recyclable treated water [4-5]. * corresponding author. e-mail address: phlin@ncu.edu.tw proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 2 the advantages of mbr include high solid-liquid separation efficiency, small footprint, high volume loading, highefficiency organic removal, low sludge production, and higher removal of nutrients, organic and persistent organic pollutants (pop) over conventional activated sludge processes [6-8]. however, membrane fouling, defined as the deposit of materials on the membrane surface with the possible existence and growth of microorganisms on it, is the main challenge of the mbr process. it can lead to a decline in permeate flux or an increase in transmembrane pressure (tmp) over time, which results in higher operational costs for membrane cleaning and eventually reduces the lifespan of the membranes [9-10]. over the last two decades, mbrs have attracted great attention due to their ability to produce high-quality effluent. it is currently considered one of the most promising technologies to treat domestic and industrial wastewater and is suitable for direct recycling or further additional purification steps [10-11]. it was reported that 22.4% of the compound annual growth rate (cagr) was expected for the mbr market [12]. a recent business communication company (bcc) report revealed that the market size of mbr was 3.0 billion usd in 2019, and is expected to reach 4.2 billion usd by 2024, with a cagr of 7% [13]. several studies have demonstrated the effectiveness of mbr in treating textile wastewater [11, 14-17]. furthermore, the combination of mbr with other membrane technologies, such as nanofiltration (nf) and reverse osmosis (ro), is applied for the treatment and reclamation of textile wastewater. in the research conducted by li et al. [18], high water recovery could be achieved through the recirculation of nanofiltration concentrate to the mbr in the mbr-nf hybrid process for textile wastewater treatment. in another study, a laboratory-scale mbr-ro unit was employed to treat textile wastewater. the results indicate that the mbr system can achieve 99%, 90.3%, and 82.5% reduction of total suspended solids (tss), chemical oxygen demand (cod), and color, respectively. it is highly effective for pre-ro treatment [18]. however, more practical experience through field trials in textile factories is still needed. in this study, a pilot-scale submerged hollow fiber mbr with a capacity of up to 400 l/d is applied to treat the textile wastewater from a textile factory for 7 months. in addition, an ro module using the mbr permeate as its influent was installed to produce pure water. this is one of the few successful pilot mbr-ro tests conducted in taiwan for textile wastewater treatment and reclamation. this study examines water quality parameters such as cod, color, oil, and conductivity, as well as operating parameters including tmp, permeate flux, and recovery rate. it provides a new concept for recycling wastewater in textile mills. 2. materials and methods this section introduces the materials and methods used to investigate the efficiency of the integration of mbr and ro in treating and reclaiming textile wastewater. the wastewater treatment processes implemented by the textile factory, including physical-chemical and biological processes, are described. the wastewater source and characteristics, the pilot-scale mbr and ro plant, as well as operational conditions, are also explained. the section provides a comprehensive overview of the experimental setup and procedures used to evaluate the performance of mbr and ro in treating textile wastewater. 2.1. description of the wastewater treatment processes of the textile factory fig. 1 illustrates the schematic diagram of the wastewater treatment processes implemented by the textile factory. wastewater produced from various manufacturing processes is collected, mixed in the equalization tank, and passed through a cooling tower to reduce the water temperature. the treatment processes involve physical-chemical processes such as rapid mix, slow mix, and dissolved air flotation (daf) system. aluminum sulfate (alum) was added as a coagulant in the rapid mix tank, while polymer was added as a flocculant in the slow mix tank to bridge small flocs to larger flocs. if necessary, sodium hydroxide (naoh) is used to maintain neutral ph. the daf system can generate micron-sized bubbles that attach to the suspended particles, causing them to float to the surface where they can be skimmed off. the daf can effectively remove proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 3 suspended solids, as well as minor biological oxygen demand (bod) and cod. subsequently, several activated sludge systems are employed as biological treatments to remove major bod and cod. finally, the effluent from the clarifier undergoes rapid mixing, slow mixing, and daf processes again before being discharged. fig. 1 the schematic diagram of the wastewater treatment processes of the textile factory 2.2. wastewater source and characteristics the daily wastewater quantity ranges approximately between 2500 and 3200 m3/day (cmd). the cod and ph of the raw wastewater range from 600 to 800 mg/l and 8.58.7, respectively. in this study, the treated wastewater from the first daf system was used as the influent of the pilot-scale mbr plant. the cod of the mbr influent ranged from 200 to 500 mg/l; the color ranged from 300 to 400 pt-co. during the test, the mbr influent was switched to the mixed liquor of the activated sludge system to increase biomass concentrations and evaluate membrane fouling under high biomass conditions. the cod of this stream ranged from 100 to 200 mg/l. 2.3. pilot-scale mbr and ro plant fig. 2 the pilot-scale mbr module the picture of the pilot-scale mbr module is shown in fig. 2. the entire module has dimensions of 1 m wide, 2 m long, and 1.8 m high, with a capacity of up to 400 l/d. the system consisted of an activated sludge tank with polyvinylidene fluoride (pvdf) hollow fiber membranes (suez, zeeweed® 500d), with a membrane length of 60 cm and a pore size of 0.04 μm. proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 4 the system also included a product water tank, a programmable logic controller (plc) system, a water pump, and a blower. the effluent from the first daf system flowed into the activated sludge tank with hollow fiber membranes, and the resulting filtrate was collected in the product water tank. the picture of the pilot-scale ro module is shown in fig. 3. the system consisted of an influent tank, which stored the mbr product water, a 90 cm ro membrane, a water pump, a plc system, and a 5 μm filter installed before the ro membrane. the ro membrane utilized in the system was a spiral-wound element with a polyamide thin-film composite supplied by suez. the purpose of the filter was to protect the ro membrane from large particles that could potentially clog the membrane. fig. 3 the picture of the pilot-scale ro module 2.4. operational conditions at the beginning of the test period, the mbr was inoculated with 10 liters of activated sludge obtained from the activated sludge system in the textile factory. the schedule of the important working items is listed in table 1. the initial mbr permeate flux was 15 lmh and increased to 20 lmh on the 36th day. the initial backwash flux was 34 lmh and was conducted for 60 sec with a 9 min interval of producing water. as the system achieved stability, the producing water interval was increased, while the backwash time was decreased, as shown in table 1. during the first 20 days of ro operation, the permeate flow rate ranged from 2.3 to 2.8 ml/min with a recovery rate of 13 to 17%. subsequently, the ro brine with a flow rate ranging from 33 to 35 ml/min was recycled to increase the recovery rate from 44 to 48%. table 1 the schedule of the important working items during the test period time (day) working items 1 1. inoculated with 10 l of activated sludge obtained from the activated sludge system in the textile factory. 2. started running the pilot mbr module. 16 inoculated with 10 l of activated sludge obtained from another activated sludge system. 25 increased filtration time (9 to 12 min) and decreased backwash time (60 to 30 sec). 36 increased permeate flux from 15 to 20 lmh. 40 1. increased mbr filtration time from 12 to 15 min. 2. started running the pilot ro module. 56 recycled the ro brine to increase the recovery rate. 83 switched the mbr influent from the daf to the mixed liquor of the activated sludge system. 97 switched mbr influent back to the daf proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 5 3. results and discussion in this section, the experimental tests of the pilot-scale mbr using different sources of wastewater were conducted for 7 months. the temporal variations in cod and mlss were observed during the operation. membrane fouling in the mbr system was monitored and evaluated. in addition, the operation and performance of the ro module were also extensively monitored and analyzed. these results provide valuable insights into the potential applications of mbr and ro in textile wastewater treatment and reclamation. 3.1. pilot-scale mbr the pilot-scale mbr has been operated for approximately 7 months (198 days). fig. 4fig. 4 illustrates the temporal variations of cod in the mbr influent and effluent. the cod of the daf effluent ranged from 169 to 523 mg/l (average: 332±84 mg/l) and the mbr effluent ranged from 5 to 88 mg/l (average: 38±24 mg/l), resulting in an average removal efficiency of 88%. the fluctuations in the influent cod levels can be attributed to the variation of the manufacturing processes, in which different chemicals and additives are applied. in addition, the variability in production quantities and batch sizes in the textile factory can also contribute to cod fluctuations. on the other hand, the cod of the activated sludge mixed liquor ranged from 107 to 200 mg/l (average: 165±39 mg/l), and the mbr effluent ranged from 50 to 143 mg/l (average: 94±35 mg/l). the average removal efficiency was only 43%, significantly lower than the removal efficiency using the daf effluent as the mbr source wastewater. fig. 4 mbr influent and effluent cod as a function of time. (as: activated sludge) the mixed liquor-suspended solids (mlss) of the mbr as a function of time is presented in fig. 5. during the first 80 days of operation, the mlss in the mbr system gradually increased due to the growth of the biomass in the system. during the 83rd to 97th day, there was an mlss spike when the activated sludge mixed liquor was used as the mbr influent. theoretically, such high mlss should result in better cod removal efficiency. however, the cod removal decreased from 83% to 43%. two phenomena were observed: (1) the pressure during membrane backwash was significantly increased from 10 kpa to 50 kpa (data not shown); (2) the oil and grease were observed in the mbr system. therefore, two samples were collected from the activated sludge system on two different dates, and the oil concentrations were measured. the results were 310 and 503 mg/l, respectively (fig. 5), which was significantly higher than the normal oil concentration that mbr can treat. abass et al. [19] found that increasing oil and grease concentration in the mbr influent from 83 mg/l to 260 mg/l accelerated membrane fouling, despite the repeated physical cleaning operation performed [20]. it was found that the oil could be from the kitchen or manufacturing processes and accumulate in the activated sludge. the high oil concentration can cover proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 6 the microorganism biomass in the mbr and significantly affect the ability to degrade the organics in the wastewater [20]. the oil could also cause membrane fouling for long-term operation, reducing the filtration efficiency, requiring more frequent chemical cleaning, and eventually increasing operating costs. therefore, the mbr influent was switched back to daf effluent to prevent oil contamination. meanwhile, part of the mlss in the mbr was discharged to remove oily biomass, which made the mlss low during the 111th to 147th day. fig. 5 mlss and oil concentration in mbr over time membrane fouling is a major drawback of the mbr process, leading to a decline in permeate flux or an increase in the tmp over time. this subsequently results in higher operating costs, reduced treatment capacity, and ultimately shortens the life span of membranes [21-22]. to evaluate whether the wastewater from the textile factory severely fouled the hollow fiber membrane, the tmp was monitored throughout the test period. fig. 6 shows the mbr tmp of the produced water as a function of operation time. fig. 6 mbr tmp of producing water as a function of operation time the tmp started from -1.3 kpa, gradually decreased to -5.3 kpa on the 75th day, and remained stable until the end of the test. the final tmp was still significantly lower than the tmp requiring chemical cleaning (usually -50 kpa, but it varies by different manufacturers). the membrane permeability (the ratio of flux to tmp) during the test period ranged from 3.9 to 11.2 proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 7 lmh/kpa, similar to the permeability reported in other mbr systems for industrial wastewater reuse (i.e., 3 to 1000 lmh/kpa) [23]. these results indicate that the hollow fiber membrane used in this study performed extremely well and was suitable for treating the wastewater produced by this factory. however, the mbr effluent did not meet the reused standards directly (cod � 30 mg/l, and conductivity � 1800 μs/cm) [24-25]. therefore, from a water reuse perspective, the implementation of other advanced treatment technologies should be considered to achieve reuse standards. 3.2. pilot-scale ro the pilot-scale ro module has been operated for more than 4 months (135 days). during the test period, cod, color, and conductivity of the ro influent, effluent, and brine were monitored as shown in fig. 7. it is noted that ro brine was not recycled for the first 7 days of operation, which resulted in a recovery rate of 13~17%. afterward, ro brine was recycled once till the end of the test, and the average recovery rate was 50%. the average cod, color, and conductivity of the ro permeate was 7 mg/l, 7 pt-co, and 16 μs/cm, respectively, and the removal efficiency of cod, color, and conductivity were 92%, 98%, and 99%, respectively. the ro permeate is qualified for reuse in the manufacturing processes within this textile factory, enabling a significant reduction in the quantity of freshwater used in the production process. (a) cod of the ro influent, effluent, and brine (b) color of the ro influent, effluent, and brine (c) conductivity of the ro influent, effluent, and brine fig. 7 cod, color, and conductivity of the ro influent, effluent, and brine as a function of time similar to mbr systems, fouling is a primary issue in ro applications. generally, ro membrane fouling is classified into inorganic fouling, organic fouling, biofouling, and colloidal fouling [26]. among them, biofouling is a significant concern for ro membranes because microorganisms are widespread in most waters. bacteria can secrete extracellular polymeric proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 8 substances (eps) that facilitate their attachment to membrane surfaces. this can lead to the formation of biofilms. biofilms provide an environment for bacterial colonization and continuous release of eps and organic compounds such as polysaccharides and proteins. these substances accumulate within the membrane [26]. ismail et al. [27] indicate that biofouling accounts for over 45% of all membrane fouling. in this study, the tmp of the ro membrane remained below 0.5 kg/cm2 during the entire test (ranged from 0.1 to 0.4 kg/cm2, data not shown), which was still below the critical pressure required for chemical cleaning (i.e.,1 kg/cm2). the results obtained not only indicated that fouling or scaling of the ro membrane was not an issue when applying the mbr effluent as its inlet water but also demonstrated the feasibility of combining mbr and ro for the treatment and reclamation of textile wastewater. furthermore, the average cod, color, and conductivity of the ro brine were 85 mg/l, 340 pt-co, and 1730 μs/cm, respectively. these values were below the wastewater discharge standards [28], indicating that the ro brine is likely suitable for direct discharge without additional treatment. 3.3. summary and suggestions the study demonstrated that mbr resembles a highly effective system for treating daf effluent. combining with the ro module can produce high-quality permeate suitable for textile processes in the factory, indicating the feasibility of the mbr-ro system for wastewater treatment and reclamation in full-scale treatment processes. the water characteristics of each treatment stage and corresponding pictures of water samples are shown in fig. 8. based on the test results, it is recommended to replace the current sludge in the activated sludge system with new sludge when constructing a full-scale treatment process. since the current sludge contains high-concentration oil due to long-term accumulation from the wastewater (i.e., kitchen wastewater or manufacturing processes). wastewater streams containing oil need to be pretreated to decrease the impact on the sludge and membranes. high oil concentration could cover the biomass and significantly affect its ability to degrade the organics in the mbr, resulting in low cod removal. the oil accumulating in the system can foul the membrane and cause severe membrane clogging. furthermore, part of the oil can penetrate the membrane and reach the ro module, and eventually foul the ro membrane. fig. 8 water characteristics of each treatment stage and corresponding pictures of water samples 4. conclusions the pilot mbr system was continuously operated for 7 months, and the pilot ro module was operated for more than 4 months with a recovery rate of 50%. the experimental results lead to the following conclusions: (1) the average cod of mbr effluent is 38±24 mg/l when applying the daf effluent as its inlet water, which results in an average removal efficiency of 88%. proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 9 (2) the tmp of the mbr ranged from -3 to -5 kpa with a permeate flux of approximately 20 lmh during the test. the results demonstrate that the hollow fiber membrane used performed extremely well, hence it can apply to the wastewater produced from the factory. (3) the average cod, color, and conductivity of the ro permeate are 7 mg/l, 7 pt-co, and 16 μs/cm, respectively. the removal efficiency of cod, color, and conductivity are 92%, 98%, and 99%, respectively. (4) the result indicates that the ro permeate is suitable for reuse in the manufacturing processes. additionally, the tmp of the ro membrane was ≤ 0.5 kg/cm2 during the entire test. (5) the results of this study demonstrate the great feasibility of using mbr combined with ro to treat and reclaim textile wastewater. acknowledgment the authors gratefully acknowledge eclean-tech enterprise co., ltd for providing the pilot equipment and financial support. the authors would also like to express their appreciation to ms. chiou-mei chen for her helpful discussion. conflicts of interest the authors declare no conflict of interest. references [1] h. ali, “biodegradation of synthetic dyes—a review,” water, air, & soil pollution, vol. 213, no. 1-4, pp. 251-273, november 2010. [2] d. a. yaseen and m. scholz, “textile dye wastewater characteristics and constituents of synthetic effluents: a critical review,” international journal of environmental science and technology, vol. 16, no. 2, pp. 1193-1226, february 2019. [3] m. k. shahid, a. kashif, p. pathak, y. choi, and p. r. rout, “water reclamation, recycle, and reuse,” clean energy and resource recovery: wastewater treatment plants as biorefineries, vol. 2, pp. 39-50, 2022. [4] n. c. cinperi, e. ozturk, n. o. yigit, and m. kitis, “treatment of woolen textile wastewater using membrane bioreactor, nanofiltration and reverse osmosis for reuse in production processes,” journal of cleaner production, vol. 223, pp. 837-848, june 2019. [5] l. m. l. k. b. lindamulla, v. jegatheesan, k. b. s. n. jinadasa, k. g. n. nanayakkara, and m. z. othman, “integrated mathematical model to simulate the performance of a membrane bioreactor,” chemosphere, vol. 284, article no. 131319, december 2021. [6] c. v. faria, b. c. ricci, a. f. r. silva, m. c. s. amaral, and f. v. fonseca, “removal of micropollutants in domestic wastewater by expanded granular sludge bed membrane bioreactor,” process safety and environmental protection, vol. 136, pp. 223-233, april 2020. [7] s. kitanou, m. tahri, b. bachiri, m. mahi, m. hafsi, m. taky, et al., “comparative study of membrane bioreactor (mbr) and activated sludge processes in the treatment of moroccan domestic wastewater,” water science and technology, vol. 78, no. 5, pp. 1129-1136, october 2018. [8] a. yurtsever, e. basaran, d. ucar, and e. sahinkaya, “self-forming dynamic membrane bioreactor for textile industry wastewater treatment,” science of the total environment, vol. 751, article no. 141572, january 2021. [9] a. sengar and a. vijayanandan, “effects of pharmaceuticals on membrane bioreactor: review on membrane fouling mechanisms and fouling control strategies,” science of the total environment, vol. 808, article no. 152132, february 2022. [10] m. wu, y. chen, h. lin, l. zhao, l. shen, r. li, et al., “membrane fouling caused by biological foams in a submerged membrane bioreactor: mechanism insights,” water research, vol. 181, article no. 115932, august 2020. [11] m. brik, p. schoeberl, b. chamam, r. braun, and w. fuchs, “advanced treatment of textile wastewater towards reuse using a membrane bioreactor,” process biochemistry, vol. 41, no. 8, pp. 1751-1757, august 2006. [12] p. krzeminski, l. leverette, s. malamis, and e. katsou, “membrane bioreactors – a review on recent developments in energy reduction, fouling control, novel configurations, lca and market prospects,” journal of membrane science, vol. 527, pp. 207-227, april 2017. proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 10 [13] s. al-asheh, m. bagheri, and a. aidan, “membrane bioreactor for wastewater treatment: a review,” case studies in chemical and environmental engineering, vol. 4, article no. 100109, december 2021. [14] c. lubello and r. gori, “membrane bio-reactor for advanced textile wastewater treatment and reuse,” water science and technology, vol. 50, no. 2, pp. 113-119, july 2004. [15] p. schoeberl, m. brik, m. bertoni, r. braun, and w. fuchs, “optimization of operational parameters for a submerged membrane bioreactor treating dyehouse wastewater,” separation and purification technology, vol. 44, no. 1, pp. 6168, july 2005. [16] n. o. yigit, n. uzal, h. koseoglu, i. harman, h. yukseler, u. yetis, et al., “treatment of a denim producing textile industry wastewater using pilot-scale membrane bioreactor,” desalination, vol. 240, no. 1-3, pp. 143-150, may 2009. [17] x. zheng and j. liu, “dyeing and printing wastewater treatment using a membrane bioreactor with a gravity drain,” desalination, vol. 190, no. 1-3, pp. 277-286, april 2006. [18] k. li, q. liu, f. fang, x. wu, j. xin, s. sun, et al., “influence of nanofiltration concentrate recirculation on performance and economic feasibility of a pilot-scale membrane bioreactor-nanofiltration hybrid process for textile wastewater treatment with high water recovery,” journal of cleaner production, vol. 261, article no. 121067, july 2020. [19] o. k. abass, f. fang, m. zhuo, and k. zhang, “integrated interrogation of causes of membrane fouling in a pilotscale anoxic-oxic membrane bioreactor treating oil refinery wastewater, ” science of the total environment, vol. 642, pp. 77-89, november 2018. [20] s. h. zahraa, h. a. gzar, and h. h. shaimaa, “evaluation of laboratory submerged membrane bioreactor and reverse osmosis system for treating textile wastewater and reuse,” journal of physics: conference series, vol. 1973, article no. 012185, august 2021. [21] s. k. tseng, “the effect of grease on the activated sludge process in wastewater treatment,” journal of the chinese institute of engineers, vol. 2, no. 2, pp. 143-149, may 1979. [22] i. s. chang, p. l. clech, b. jefferson, and s. judd, “membrane fouling in membrane bioreactors for wastewater treatment,” journal of environmental engineering, vol. 128, no. 11, pp. 1018-1029, november 2002. [23] v. jegatheesan, b. k. pramanik, j. chen, d. navaratna, c. y. chang, and l. shu, “treatment of textile wastewater with membrane bioreactor: a critical review,” bioresource technology, vol. 204, pp. 202-212, march 2016. [24] j. hoinkis, s. a. deowan, v. panten, a. figoli, r. r. huang, and e. drioli, “membrane bioreactor (mbr) technology – a promising approach for industrial water reuse,” procedia engineering, vol. 33, pp. 234-241, 2012. [25] a. rozzi, f. malpei, l. bonomo, and r. bianchi, “textile wastewater reuse in northern italy (como),” water science and technology, vol. 39, no. 5, pp. 121-128, 1999. [26] p. schoeberl, m. brik, r. braun, and w. fuchs, “treatment and recycling of textile wastewater—case study and development of a recycling concept,” desalination, vol. 171, no. 2, pp. 173-183, january 2005. [27] f. ismail, k. c. khulbe, and t. matsuura, reverse osmosis, 1st ed, amsterdam: elsevier, 2018. [28] m. tawalbeh, l. qalyoubi, a. al-othman, m. qasim, and m. shirazi, “insights on the development of enhanced antifouling reverse osmosis membranes: industrial applications and challenges,” desalination, vol. 553, article no. 116460, may 2023. [29] the water pollution control act, effluent standards 14, 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/). 3___peti#8589___20-29 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 20-29 porous maize stalk cellulose fiber-reinforced geopolymer composites for heat insulation on the bottom side of a local electric stove addisu workiye * , eyassu woldesenbet department of mechanical engineering, addis ababa institute of technology, addis ababa university, ethiopia received 30 september 2021; received in revised form 24 january 2022; accepted 25 january 2022 doi: https://doi.org/10.46604/peti.2022.8589 abstract the objective of this work is to develop porous maize stalk cellulose fiber-reinforced geopolymer composites for heat insulation on the bottom side of an electric stove using the solid impregnation method. heat loss measurement is conducted using an infrared thermometer. moreover, the temperature effect on the composites is investigated. the maize stalk cellulose fibers are very essential to anticipate the cracking phenomenon generated by high temperatures. the degradation of the fibers causes the formation of small cavities in the matrix, and thus leads to high temperatures. the experimental result shows that it takes 22 minutes to boil water using the proposed electric stove, whereas it takes 29 minutes using the existing local electric stove. by using the proposed electric stove to boil water, 113,793,148.104 kwh of energy per year at the national level can be saved. keywords: geopolymer, kaolin, insulation, composites, organic oil 1. introduction environmentally friendly heat insulation materials are critical (especially for the structure of buildings and the components of stoves, furnaces, etc.) to sustain development in this era of construction and manufacturing industries. geopolymer is a non-flammable material and has potential for the application of thermal insulation [1]. geopolymer, which is a novel family of cementitious materials, is environmentally friendly due to less emission of co2 and less energy consumption during its formation. it is synthesized from different types of aluminosilicate materials. most of the existing local electric stoves have been manufactured by small facilities without any regard for energy efficiency standards and safety. these stoves have many disadvantages, such as poor insulation, lack of temperature regulation, bulkiness, and overall poor design that encourages wastage of heat. it is therefore vital to develop heat insulation materials for the stoves. energy can be saved by applying locally available materials, such as clay, wood ash, and maize stalk cellulose fibers, for heat insulation on the bottom of local electric stoves. the heat loss from the bottom side of a locally made electric stove is high due to the absence of heat insulation. the purpose of this study is to develop eco-friendly porous maize stalk cellulose fiber-reinforced geopolymer composites from locally available materials for heat insulation on the bottom of local electric stoves to save energy. the study is organized as follows. section 2 presents the detailed literature review. section 3 describes the materials and methodology used to conduct the research work. section 4 presents the results and discussion of the finding. finally, the study is concluded. * corresponding author. e-mail address: natanium27@gmail.com proceedings of engineering and technology innovation, vol. 22, 2022, pp. 20-29 2. literature review kaolinite is among the most common minerals. it is a white and fine particle, which is a plate-like hydrated aluminum silicate crystalline mineral formed over millions of years by the hydrothermal decomposition of granite rocks [2]. normally, suitable kaolin contains 70-73% of sio2, 18-20% of al2o3, 0.4-1% of fe2o3, and 0-0.8% of tio2, without mno [3]. the main component of kandites is kaolinite, whose chemical formula is a12si2o5(oh)4 or a12o3⋅2sio2⋅2h2o(as2h2). the crystalline structure of kaolinite is classified as a two-layered sheet silicate. one layer contains aluminate groups (alo4(oh)2 octahedrons), and the other contains silicates (sio4 tetrahedrons) [4-5]. metakaolin is prepared from kaolin clay by calcination under well-controlled conditions and transformed into a highly reactive amorphous aluminosilicate [6]. metakaolin is formed from very pure kaolin clay by calcination at low temperatures (700-800°c). the transformation of kaolin to metakaolin is an endothermic process, where a large amount of energy is required to remove the chemically bonded hydroxyl ions [7]. 2 2 5 4 2 2 7 2al si o (oh) ----al si o +2h o [8] (1) the main requirement for developing a stable geopolymer is that the raw materials must be extremely amorphous, have adequate reactive glassy content with low water requirement, and can free aluminum. the alkaline solution is among the valuable things in producing a geopolymer. the alkaline activators such as sodium hydroxide, potassium hydroxide, sodium silicate, and potassium silicate are used to activate aluminosilicate materials. geopolymer is synthesized by mixing the aluminosilicate materials with strong alkaline solutions [9-10]. it has advantages such as high mechanical strength, good thermal and chemical stability, etc. [11-13]. recent research suggested developing highly porous and open-cell geopolymers by emulsion templating with vegetable oils for applications of acoustic and thermal insulators [14-16]. moreover, h2o2, aluminum powder, and silica fume are often chosen as blowing agents to develop alkali-activated foams. the decomposition of h2o2 generates oxygen gas inside the cement matrix, resulting in the formation of macrospores in different sizes and shapes. however, the decomposition of aluminum powder and silica fume generates hydrogen gas. the potential problems arise when the porous structure produced by foaming agents is stable only for a short time and then falls for a while (before the bonding process—geopolymerization) [17]. moreover, the repeatability of the results obtained on an industrial scale is still regarded as a challenge of using geopolymer materials as insulators in buildings and other applications [18]. table 1 previous studies using vegetable oils for the formation of highly porous geopolymers precursor alkali ion oil type and the amount added h2o2 wt% added method ref. fly ash naoh corn (4-12 wt%) direct incorporation [19] metakaolin koh olive (20-45 wt%) 0-20 wt% [20] metakaolin koh sunflower, canola, and olive (0-10 wt%) about 6 wt% pre-emulsification [21] metakaolin naoh soy bean (20 wt%) 6 wt% direct incorporation [22] metakaolin naoh canola (2-6 wt%) 0.5-1.5 wt% [23] metakaolin and fly ash koh sunflower (25 wt%) 6 wt% [24] table 1 shows the materials, amount incorporated, and methods used in previous studies for the formation of highly porous geopolymers. vegetable oils were used for emulsion templating with saponification reactions. the three main processing methods of adding the organic liquid into the alkali-activated and ordinary portland cement materials are direct incorporation into the reactive slurry, pre-emulsification ahead of the addition of the solid precursor, and solid impregnation ahead of the addition of the reactive mixture. these process routes are clearly described in fig. 1. usually, water is used to mix portland cement powder, whereas alkaline silicate aqueous solutions are used to mix alkali-activated material powder [25]. the organic liquid can be added at different steps of composite manufacturing, using distinct approaches. 21 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 20-29 fig. 1 three processing routes of adding the organic liquid into alkali-activated materials or portland cement [25] in ethiopia, the locally manufactured electric stoves are predominantly open resistor types and are used in households mainly for cooking or reheating “wot”, making tea, heating water for washing, etc. cafes, restaurants, and hotels use these stoves rarely in the absence of kerosene, wood fire, and cylinder-based gas stoves [26]. the resistance of the heating element used in the stoves is measured during the assessment. it varies from 31.1 ῳ to 39.9 ῳ, which corresponds to the power diameter (p = v 2 / r), at 220 v of 1.6 kw to 1.2 kw respectively. the average power demand of the stoves is estimated to be 1.4 kw. the locally manufactured electric stoves use either singleor double-heating elements. the most common one, the 22 cm-diameter stove, is produced using single resistors with a diameter of 0.6 or 0.7 mm and a length of about 90 cm. most of the existing stoves in ethiopia (fig. 2) have been manufactured by small facilities without any regard for energy efficiency standards and safety. the stoves have many disadvantages, such as poor insulation, lack of temperature regulation, bulkiness, and overall poor design that encourages wastage of heat. the heat loss intensities are caused by conduction, convection, and radiation on the bottom and the top of the stoves. there is significant heat loss on the bottom of the existing locally made stoves through radiation. the heat intensity measured on the bottom of the 22 cm-diameter stove reaches 395°c at the end of a cold-start test, for a duration of 27 minutes. the intensity further reaches 398°c at the end of a hot-start test. the temperature on the top of the stove on a vertical test plane 1.5 cm from the part body reaches 217°c at the end of the hot start test phase. the input energy (qin) is the electric energy and is equal to the sum of the output energy (thermal, (qu)) and the losses. in u iinput energy (q ) = utilized energy (q ) + energy loss (q ) (2) the useful output may be electric power, mechanical work, or heat. the utilized energy is the properly consumed energy of the electric stove for cooking and it is determined as: u i f v f i f pq = (m m ) c + (t t ) m c− × − × (3) where qu is the heat utilized (kj), mi is the initial weight of water (g), mf is the final weight of water (g), cv is the water vaporized heat (2.260 kj/g°c), tf is the final water temperature (°c), and ti is the initial water temperature (°c). energy conversion efficiency = useful output energy (utilized energy) / input energy (4) 22 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 20-29 the potential problems of a locally manufactured electric stove are illustrated in fig. 3. to solve the heat loss problems, bottom heat insulation enclosure and clay plate support have been designed. heat insulation using locally available materials are applied to minimize the heat loss due to radiation and conduction from the clay plate. the insulators could be fiberglass, lightweight pumice, or wood ash. heat intensities on the bottom of the prototype stove have been reduced to 190°c. ethiopian standard es 3406:2007 related en 60335-2-6:2015 was used for measuring the performance of electric ranges, hobs, ovens, and grills for the households. the projected peak power demand and energy consumption of local stoves, including losses, in the 2019 ethiopian fiscal year (efy) are estimated to be 887 mw and 3,794 gwh respectively [27]. however, further research work is still needed to reduce the heat loss. maize stalk cellulose fiber-reinforced calcined kaolin clay-based geopolymer composite has been developed in previous work [28]. what makes this research work different is that the porous network is developed by using the solid impregnation method with organic oils for the insulation on the bottom of a local electric stove. 3. materials and methods this study emphasizes ethiopian kaolinitic clay mineral and its ability to form a cement geopolymer after thermal and alkaline activation. kaolinitic clay is abundantly available in ethiopia and is used in the production of brick and ceramic industries. the raw material for this study is collected from chagal around d/libanose district in the north shewa zone, as shown in fig. 4. in this study, the chemical composition analysis of chagal clay soil is done by atomic absorption spectrometer at the geological survey of ethiopia. apart from atomic absorption spectrometer (aas), the libo2 fusion, hf attack, gravimetric, and colormetric methods are carried out for the samples of chagal clay soil to perform the complete silicate analysis at the geological survey of ethiopia. calcination is conducted at kombolcha polytechnic college using a thermocomputer tc 60 type fl350 developed by kilns & furnaces ltd. the maximum temperature of calcination is 1300°c. the kaolin clay calcined at 850°c is used as the precursor material. sodium silicate with a factor of 2.17 (sio2/na2o) and naoh m12 concentration are used as an alkaline activator in this research work [28]. besides, the maize stalk cellulose fibers extracted from maize stalks using the retting process are used as reinforcement. the fiber surface morphology is modified using sodium hydroxide to reduce its hydrophilic behavior as shown in fig. 5, where the tensile strength and young’s modulus of the single cellulose fiber are determined using the astm d 3822 standard. 1184 mpa and 16 gpa are achieved for tensile strength and young’s modulus respectively [28]. the sodium hydroxide treated short maize stalk cellulose fibers with a length of 2 mm are used as reinforcement in this research work. fig. 4 clay raw materials collected from chagal in ethiopia fig. 5 sodium hydroxide treated maize stalk cellulose fibers fig. 2 an existing local electric stove [27] high power demand low energy efficiency of a locally manufactured electric stove absence of standards type, size, and installation of heating elements size of the stove and power rating lack of research and inovation heat loss bottom, lateral, and top heat losses lack of insulation lack of heat intensity controll mechanisms production of clay plate fig. 3 the problem tree of a locally manufactured electric stove 23 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 20-29 3.1. solid impregnation the solid impregnation method is used to develop porous maize stalk cellulose fiber-reinforced geopolymer composites. this processing method is applied to introduce organic liquid into cementing materials. usually, the alkaline-activated material powder is activated with an alkaline silicate aqueous solution. the solid impregnation method consists of the process of impregnating solid powder with organic liquid. the cementing precursor itself is impregnated with the organic oil and then mixed with the activating solution. the organic liquid is adsorbed onto calcined kaolinite clay; the particle size of charcoal ash is 75 µm in the early stage. then, dried maize stalk cellulose fibers (5 wt%) are added and mixed in a high shear mixer for 2 minutes at a high speed of 800 rpm, and sodium water glass is slowly added to the precursor and mixed until complete integration occurs. after that, the slurry is poured into a high-strength plastic mold and sealed using a plastic cover, being remolded after two days and cured after 28 days. the developed porous maize stalk cellulose fiber-reinforced geopolymer composites are shown in fig. 6. table 2 summarizes the materials, amount incorporated, and the method used for the development of porous maize stalk cellulose fiber-reinforced geopolymer composites with vegetable oils. table 2 development of porous geopolymer composites using vegetable oils aluminosilicate source oil type fiber type alkali ion method used calcined kaolin (50 wt%) coal ash (50 wt%) sunflower (2-6 wt%) maize stalk cellulose fiber (5 wt%) naoh solid impregnation 3.2. thermal behavior of maize stalk cellulose fiber-reinforced geopolymer composites thermogravimetric analysis is performed with sdt q600 v20.9 build 20 (thermobalance universal v4.5a) developed by ta instruments, inc. the device consists of an electronic balance which is placed inside an oven. also, the device is coupled with a control microprocessor. 35.5980 mg of samples in the form of fine powder are slid and placed inside the analysis equipment for testing, and then calcined to 750°c, increasing 10°c per minute. moreover, the temperature effect on the geopolymer composites is investigated. the composite samples with the size of 15 × 10 × 600 mm (w × h × l) are heated in a ventilated furnace to assess the effect of temperature on the surface and microstructure. the heating process is carried out at a rate of 6°c/min for a duration of one hour until the target temperatures of 200°c, 400°c, 600°c, and 800°c are recorded. 3.3. proposed electric stove the heat loss on the bottom of the electric stove in a form of radiation constitutes the major portion of the heat loss. no heat insulator is used between the clay plate and the body of the 22 cm-diameter stove. the existing local stove needs modification due to the following limitations: lack of standards on the product, lack of heat intensity control mechanisms and the switch to control the heat level, the gap between the cookware base and the heating elements, frequent burning of a heating element due to spillage, poor quality of the heating elements, wear of the clay plate and the framework, sub-optimal/poor and inefficient design and workmanship, etc. [29]. in this study, maize stalk cellulose fiber-reinforced calcined kaolin-based geopolymer composites with a thickness of 25 mm are used as a thermal insulator on the bottom of the clay plate to reduce the amount of heat loss as shown in fig. 6. the grooves made on the clay plate are made with four concentric circles having a “u” shaped section. the depth and width of grooves are 8 to 11 mm and 5 to 6 mm respectively. the wound resistor with a diameter of 4-4.5 mm rests on the bottom of the grooves. this layout lets much of the heat from the resistor be transmitted to the clay plate through conduction. two ceramic connectors are used to connect the “woraj” (connecting leads) coming from the resistor to a switch. the connecter is attached on the bottom of the thermal insulator plate. the prototype uses a bimetal strip with a knob having five levels to achieve heat intensity control. heat intensity control enables users to utilize the heat retained in the stove and allows slow cooking. the proposed local electric stove heat insulation setup is shown in fig. 7. 24 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 20-29 fig. 6 heat insulation composite fig. 7 heat insulation setup 3.4. heat level measurement using an infrared thermometer the heat level is measured on the top, bottom, and side of the stove using an infrared thermometer for one hour with 10-minute time intervals as shown in fig. 8. besides, the time of boiling water is also recorded. the investigation is conducted by boiling a liter of water and measuring the required amount of time and energy to boil up the water. fig. 8 heat level measurement on the proposed electric stove 4. results and discussion one of the most significant findings of this work is that the calcined kaolin clay collected from chagal in ethiopia can be used as a potential precursor material to develop the geopolymer as a matrix for porous maize stalk cellulose fiber-reinforced composites. chagal clay is calcined in the laboratory from the room temperature to 500°c at a rate of 7.9°c/min and from 500°c to 950°c at a rate of 7.9°c/min for two hours to obtain the precursor material. table 3 shows the chemical or the oxide compositions of the sampled soils, which are characterized by libo2 fusion, hf attack, gravimetric, colormetric, and aas methods. the result confirms the availability of enough sio2 and al2o3 to be used as a precursor for geopolymer formation. since the calcination temperature rate and cooling rate are slow, the kaolinite clay has enough time to eliminate the crystalline bound water of hydration from the whole material in the furnace. the clay is in its most reactive state when the calcination temperature leads to a loss of hydroxyls and results in a collapsed and disarranged structure. the result of differential scanning calorimetry-thermogarvimetric analysis (dsc-tga) is shown in fig. 9. the first peak under 126.40°c is related to the dewatering process. this process is related to the adsorbed free water or evaporable water on the surface and the porosity of the samples. at the second peak under 284.53°c, thermal degradation occurs due to the decomposition of maize stalk cellulose fibers which generate a total of 15.61% weight loss. fig. 10 presents the pictures of maize stalk cellulose fiber-reinforced metakaolin-based geopolymer composite specimens exposed to 600°c and 800°c. it can be seen that metakaolin-based geopolymers are reinforced by the 2 wt% chopped maize stalk cellulose fibers which do not develop serious cracks on the surface of geopolymer composites. this indicates that maize stalk cellulose fibers are very essential to anticipate the cracking phenomenon generated by high temperatures. the degradation of the fibers causes the formation of small cavities in the matrix, and thus leads to high temperatures. fume is observed during the degradation process as the temperature increases to 800°c. 25 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 20-29 table 3 chemical composition of chagal clay soil samples (%) soil type sio2 feo3 al2o3 mgo cao na2o k2o tio2 p2o5 mno h2o loi chagal 1 52.82 8.84 15.80 11.14 5.80 1.70 1.22 0.75 0.21 0.2 0.15 2.84 chagal 2 51.32 8.86 16.10 11.08 5.90 1.82 1.14 0.76 0.22 0.2 0.18 2.88 fig. 9 dsc-tga of maize stalk cellulose fiber-reinforced geopolymer composites (a) exposed to 600°c (b) exposed to 800°c fig. 10 maize stalk cellulose fiber-reinforced calcined kaolin-based geopolymer composites exposed to varying temperatures the structure of the composites becomes more porous, and the expanded water vapor gets out without significant damage to the microstructure. such degradation of the maize stalk cellulose fibers may be useful to the behavior of geopolymer composites under thermal exposure. at high temperatures, when water is not removed fast enough from the composites, internal vaporization may generate high pressure inside the matrix. the porosity and small channels built by the degradation of the maize stalk cellulose fibers may lower the internal vapor pressure and thus minimize the probability of cracking. the existing local electric stove has several limitations in terms of heat loss, size, and lack of heat intensity control mechanisms according to the finding of ethiopia energy authority [29]. the proposed electric stove uses porous maize stalk cellulose fiber-reinforced geopolymer composites as a thermal insulator which decreases the heat loss on the bottom. it is environmentally friendly and developed from locally available raw materials. the literature supports the potential of using organic oils to develop porous geopolymers from different precursors [19-24]. in this study, the porous network is formed by using the solid impregnation method with sunflower oil. the heat level on the top, bottom, and side is measured using an infrared thermometer for one hour with 10-minute time intervals. the result is shown in fig. 11. the measured heat level is 440°c, 63°c, and 62°c respectively. the result shows that the temperature level on the top increases and the temperature level on the bottom decreases when compared to the existing local stove. moreover, the proposed electric stove has heat intensity control mechanisms and a switch to control the heat level. the size is also reduced to 100 × 260 × 260mm (h × l × w). the innovative contribution of this study is that the construction relevant material is applied to a non-construction work (i.e., the insulation for electric stoves). 26 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 20-29 fig. 11 temperature level versus time table 4 comparison of the insulation material, the heat level, and the time taken for boiling water stove type insulation material heat level (top) heat level (bottom) time ref. existing local electric stove without insulation 215°c 398°c 36 min [27] existing local electric stove wood ash or lightweight pumice 217°c 190°c 29 min [27] proposed local electric stove porous geopolymer composites 440°c 63°c 22 min this study table 5 comparison of the time and energy required for boiling water item existing stove proposed stove time taken 29 min (0.48 hr) 22 min (0.366 hr) power consumption 0.676 kwh 0.512 kwh no. of stove (single) 635,212 635,212 saved power per day 103920.68 kwh table 4 shows the comparison between the existing local electric stoves and the proposed local electric stove regarding the insulation material, the heat level, and the time taken for boiling water. the heat level on the bottom side of the proposed electric stove is reduced to 63°c, and the heat level on the top surface is increased to 440°c. this confirms that the porous maize stalk cellulose fiber-reinforced geopolymer composite is a better insulation material than wood ash or lightweight pumice. table 5 shows the comparison between the existing local electric stove and the proposed local electric stove regarding the time and energy required for boiling one liter of water. with the proposed electric stove, the amount of energy saved per year at the national level for boiling water is determined as follows. total energy saved per year = 103,920.68 kwh 365 days 3 times/day = 113,793,148.104 kwh× × (5) total amount of money = 103,920.68 kwh 365 days 0.5 birr = 18,972,878.512 birr 3× × × = 56,918,635.53 birr per year (6) 5. conclusions porous maize stalk cellulose fiber-reinforced geopolymer composites are successfully developed by using the solid impregnation method with vegetable oils, and are used as a thermal insulator on the bottom of the clay plate to reduce heat loss. the measured average heat loss on the bottom side within 45 minutes is 63°c. this value is the best when compared to the heat loss (190°c) under the cases using wood ash or lightweight pumice as thermal insulators. this confirms that the heat loss on the bottom side is below the thermal degradation temperature (284.5°c) of the maize stalk cellulose fibers inside the geopolymer matrix, as shown in the dsc-tga results. therefore, no fume or crack is observed on the surface of the geopolymer composites which are used as insulators. 27 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 20-29 the proposed electric stove has the following advantages: light weight, small size, and controllable heat intensity level. it takes 22 minutes to boil water using the proposed electric stove, whereas it takes 29 minutes to boil water using the existing local electric stove. this is due to the top surface temperature (440°c) on average of 45minutes. at the national level, 113,793,148.104 kwh of energy per year will be saved using the proposed electric stove to boil water. conflicts of interest the authors declare no potential conflicts of interest concerning the research, authorship, and/or publication of this article. references [1] j. fen, et al., “development of porous fly ash-based geopolymer with low thermal conductivity,” materials and design, vol. 65, pp. 529-533, january 2015. [2] h. h. murray, “traditional and new applications for kaolin, smectite, and palygorskite: a general overview,” applied clay science, vol. 17, no. 5-6, pp. 207-221, november 2000. [3] h. h. murray, “overview–clay mineral applications,” applied clay science, vol. 5, no. 5-6, pp. 379-395, march 1991. [4] w. e. worrall, “clays: their nature, origin and general properties”, 1st ed., london: maclaren and sons, 1968. [5] v. srivastava, “clays: types and applications,” bulletin of the catalysis society of india, vol. 11, no. 5, pp. 56-77, 2012. [6] g. brindley, et al., “the kaolinite-mullite reaction series: ii, metakaolin,” journal of the american ceramic society, vol. 42, no. 7, pp. 314-318, july 1959. [7] v. m. malhotra, et al., “pozzolanic and cementitious materials,” 1st ed., amsterdam: gordon and breach, 1996. [8] b. b. sabir, et al., “metakaolin and calcined clays as pozzolans for concrete: a review,” cement and concrete composites, vol. 23, no. 6, pp. 441-454, december 2001. [9] b. singh, et al., “geopolymer concrete: a review of some recent developments,” construction and building materials, vol. 85, pp. 78-90, june 2015. [10] s. h. sanni, et al., “performance of alkaline solutions on grades of geopolymer concrete,” international journal of research in engineering and technology, vol. 2, no. 11, pp. 366-371, november 2013. [11] w. d. rickard, et al., “performance of solid and cellular structured fly ash geopolymers exposed to a simulated fire,” cement and concrete composites, vol. 48, pp. 75-82, april 2014. [12] t. bakharev, “resistance of geopolymer materials to acid attack,” cement and concrete research, vol. 35, no. 4, pp. 658-670, april 2005. [13] p. palmero, et al., “geopolymer technology for application-oriented dense and lighted materials, elaboration and characterization,” ceramics international, vol. 41, no. 10, pp. 12967-12979, december 2015. [14] c. bai, “high-porosity geopolymer foams with tailored porosity for thermal insulation and waste water treatment,” journal of materials research, vol. 32, no. 17, pp. 3251-3259, september 2017. [15] z. zhang, et al., “geopolymer foam concrete: an emerging material for sustainable construction,” construction and building materials, vol. 56, pp. 113-127, february 2014. [16] c. bai, g. franchin, h. elsayed, a. conte, and p. colombo, “high strength metakaolin-based geopolymer foams with variable macroporous structures,” journal of the european ceramic society, vol. 36, no. 16, pp. 4243-4249, december 2016. [17] m. łach, et al., “determination of the influence of hydraulic additives on the foaming process and stability of the produced geopolymer foams,” materials, vol. 14, no. 17, article no. 5090, september 2021. [18] m. łach, “geopolymer foams–will they ever become a viable alternative to popular insulation materials?–a critical opinion,” materials, vol. 14, no. 13, article no. 3568, july 2021. [19] s. k. kaliappan, et al., “effect of pore-forming agents on geopolymer porosity and mechanical properties,” 3rd international conference on applied science technology, pp. 1-7, september 2018. [20] c. bai, et al., “low-temperature fabrication of sic/geopolymer cellular composites,” composites part b: engineering, vol. 137, pp. 23-30, march 2018. [21] p. hua, et al., “adsorption of acid green and procion red on a magnetic geopolymer based adsorbent: experiments, characterization, and theoretical treatment,” chemical engineering journal, vol. 383, article no. 123113, march 2020. [22] m. s. cilla, et al., “geopolymer foams obtained by the saponification /peroxide/gelcasting combined route using different soap foam precursors,” journal of the american ceramic society, vol. 100, no. 8, pp. 3440-3450, august 2017. 28 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 20-29 [23] j. fiset, et al., “macroporous geopolymer designed for facile polymers post-infusion,” ceramic and concrete composite, vol. 110, article no. 103591, july 2020. [24] m. s. cilla, et al., “effect of process parameters on the physical properties of porous geopolymers obtained by gelcasting,” ceramics international, vol. 40, no. 8, pp. 13585-13590, september 2014. [25] c. shi, et al., “recent progress in low-carbon binders,” cement and concrete research, vol. 122, pp. 227-250, august 2019. [26] m. h. kebede, et al., “home appliance efficiency improvements for energy conservation in debre berhan city: ethiopia,” american journal of energy engineering, vol. 6, no. 2, pp. 10-14, july 2018. [27] “project document on locally manufactured electric stoves: energy efficiency standards and labeling,” http://196.188.108.86/images/resources/directorate/energyefficiency&conservation/1-6%20final%20project%20docu ment%20locally%20manufactured%20electric%20stoves%20april%2019%202017.pdf, 2017. [28] a. workiye, et al., “development of maize stalk cellulose fiber reinforced calcined kaolin clay geopolymer composite,” proceedings of engineering and technology innovation, vol. 16, pp. 30-38, august 2020. [29] “design of prototype of locally manufactured electric stoves,” http://196.188.108.86/media/attachments/eea%20publicaions/energy%20related%20publications/21%20locally %20manufactured%20electric%20stove%20design%20of%20prototype%20and%20list%20of%20required%20material s.pdf, 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/). 29 5___peti#9976___40-49 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 40-49 increasing the power output of a pv solar system by using a cooling-reflector assembly naseer kareem kasim 1 , hazim hamoud hussain 2 , alaa najem abed 2,* 1 ministry of electricity, training and energy research office, baghdad, iraq 2 department of atmospheric science, college of science, mustansiriyah university, baghdad, iraq received 25 april 2022; received in revised form 07 june 2022; accepted 08 june 2022 doi: https://doi.org/10.46604/peti.2022.9976 abstract there are various methods that can be employed to increase the lifespan and power output of photovoltaic (pv) systems. this study aims to increase the power output of a grid-connected pv system by using a water-cooling unit and solar reflectors. the pv modules of the current pv system are divided into two clusters. the first cluster, which is considered an improved cluster, has a solar reflector-cooling unit added to it, while the second cluster is used as a reference. the results show that the maximum efficiency and performance ratio values of the improved and reference pv modules at 10:30 am are 14.7% & 13.7% and 97.5% & 91.2%, respectively. the maximum electrical power values of the improved and reference pv modules at 12:00 pm are 2.55 w and 1.69 w, respectively. the maximum gain value for electrical power is 43%. keywords: performance, efficiency, cooling, solar reflectors, grid-tied 1. introduction renewable energy resources can aid in solving global energy insecurity [1]. electrical energy requirements are gradually increasing, while fossil fuel resources used in traditional systems for electrical power generation are limited and are predicted to become less and more expensive in the future [2]. traditional fuels harm the environment by causing acid rain, global warming, ozone layer depletion, air pollution, and ground and surface water pollutants. global warming is a major issue that harms the environment and threatens mankind as a result of carbon dioxide emissions into the atmosphere. the global amount of co2 emissions in 2014 was about 32,381 million tons, whereas the use of solar pv systems decreases the amount of co2 emissions by more than 100 gigatons (international energy agency 2018) [3]. the concentrated solar photovoltaic (csp) material decreases approximately 50% to 60% of the total cost of the photovoltaic (pv) module [4]. solar energy applications are classified into two kinds: solar pv energy and thermal energy. presently, engineers and scientists are working hard to make clean electricity affordable on a large scale, and have been trying for a long time to develop low-cost, high-efficiency, and easy-to-manufacture solar modules with high productivity [5]. there is more work on the use of solar reflectors in both pv solar and solar thermal energy as well as the use of cooling units with solar reflectors to increase energy output and decrease the temperature of pv modules, but in different ways. more recent work has focused on improving the winter period production of pv modules using solar reflectors [6]. tabaei et al. [7] found that using aluminum foil reflectors boosts the power output of polycrystalline silicon by 14%. when utilizing aluminum foil reflectors with water film, the maximum power output is 50%. a new way to drive the variable speed input mechanism is an open topic to be investigated. zubeer et al. [8] designed a water-cooled concentrated pv system to improve the pv solar * corresponding author. e-mail address: wiseman_1988@yahoo.com tel.: +964 7706395463 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 40-49 module efficiency. the empirical results showed that the ultimate module temperature of the pv panel, concentrated pv system, and water-cooled concentrated pv system is 57.5, 64.1, and 36.5°c, respectively. in addition, the power output of the water-cooled concentrated pv system and concentrated pv system is improved by 24.4% and 10.65%, respectively. in addition, electrical efficiency is increased from 14.2% to 17% by using reflectors and water-cooling units with the pv panel (water-cooled concentrated pv). pavlov et al. [9] found that the use of mounted solar reflectors resulted in 35% increases in daily power during clear days at specific times of the year. bahaidarah et al. [10] designed a solar system that is cooled by flowing water on the backside of the pv module under the climatic conditions of al-dhahran (ksa). the results showed that the pv module’s operating temperature dropped by about 20% and its electrical efficiency increased by 9%. khaled et al. [11] designed a hybrid pv-thermal solar collector system in duhok city to enhance the pv module efficiency. the cooling method leads to an increase in the electrical efficiency by about 3% as compared with the pv solar collector system. bahaidarah et al. [12] evaluated the electrical and thermal performance of a pv system’s v-trough. the results revealed that the power gain of the pv module when using the cooling technique is about 22.8 % and is 31.5 % when using the cooling units with the v-trough system. according to some studies, the best location in high latitudes is a vertical pv module with horizontal solar reflectors (planar concentrator) [13]. ahmad et al. [14] developed multi-level fin heat sinks (mlfhs) made of an extruded alumnum material with a novel geometry attached to the back side of the pv module. at solar irradiance and ambient temperature of 941 w/m 2 and 36.17°c, respectively, a significant drop in the pv module temperature of 8.45°c was observed. under outdoor operating circumstances, the heat sink increased overall power production by up to 9.56%. solanki and sangani [15] conducted an experimental investigation in mumbai, india, in 2007. they discovered that the v-trough system (2-sun) boosts power output by 44% for passively cooled pv modules [15]. in 2009, planar concentrators of various materials were mounted to the upper and bottom edges of pv solar modules to examine their effectiveness. experiments are carried out using stainless, aluminum, chrome, and steel solar reflectors to determine the most efficient reflector material capable of producing the greatest power with the least amount of heat. it has been shown that chromium reflectors produce 27.65% more energy than aluminum reflectors and 34.05% more energy than stainless steel reflectors [16]. this research aims to improve the power efficiency of a grid-connected pv system to achieve good economic feasibility. to achieve the goal, the thin film pv module technology and the behavior of the grid-connected pv system are studied, and a pv solar system is designed and equipped with facilities for reducing the excess heat generated by the pv modules. in this study, the cooling method is used to reduce the excess heat and help enhance the efficiency of the pv solar modules. in addition, since the lifetime of the pv modules is inversely related to temperature, pv module deterioration is minimized by cooling. 2. pv solar system description the present pv system is located in baghdad/al-taji town at latitude 33.3°n and longitude 44.4°e, as shown in fig. 1. as illustrated in fig. 2, the present pv solar system consists of 30 modules designed in 5 strings with 6 series-connected modules. the data sheet for the pv modules and the information on the pv system are shown in table 1. fig. 1 present pv solar system fig. 2 block circuit of the present pv solar system 41 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 40-49 table 1 pv system information and pv module data sheet system information value module data sheet value inverter model sma sb-5000t-21 module model ts-165c2 cigs modules number 30 max power (p max) 165 w inverter size 5.30 kwp open circuit voltage (voc) 88.7 v inverter efficiency 97% short circuit current (isc) 2.66 a system size 5 kwp max power voltage (vmpp) 68.5 v pv modules tilt angle 30° max power current (impp) 2.41 a temperature coefficient of p max -0.30% / °c max reverse current (ir) 6.5 a array area 32 m 2 operating temperature -40°c to 85°c. 3. solar reflectors and pv solar modules the present pv system in this work is divided into two clusters. the first cluster comprises 12 pv modules and is classified as an improved cluster, while the second cluster comprises 18 pv modules and is classified as a reference cluster. the improved cluster (pv modules) is then compared to the reference cluster (pv modules) to determine the increment percentage (gain) in pv solar system electrical parameters. the improved cluster receives more solar radiation (from the solar reflectors and the sun) than the reference cluster. as shown in fig. 3, the solar reflectors are attached in front of the improved cluster, whereas the second cluster (18 pv modules) remains without solar reflectors. in the current pv system, the inverter has two inputs: the reference cluster input contains 18 pv modules (2970 wp), and the improved cluster input contains 12 pv modules (1980 wp). the data is obtained via a speedwire that connects an inverter to the computer. as shown in fig. 4, an inverter shows data in two clusters (a and b). cluster (a) represents the improved cluster data, whereas cluster (b) represents the reference cluster data. in the current system, an inverter is made up of two inverters that use maximum point tracker (mppt) technology [17]. fig. 3 the improved and reference pv solar module clusters fig. 4 screenshot of the inverter data shown on pc 4. preparation for the study the present experiment has been carried out in baghdad/al-mansour company (latitude 33.3°n and longitude 44.4°e). efficiency, electrical power, and performance ratio (pr) have all been improved, as have solar radiation, pv module temperature, and ambient temperature. temperatures of reference and improved pv modules (clusters) are recorded. 42 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 40-49 all electrical parameters of the reference and improved clusters are recorded. the data for each cluster is gathered from an inverter, which shows the pv system data independently in (a and b) format, as previously described. the current research starts from 7:30 am to 6:00 pm on 14 april 2020, when the weather is clear. the cooling process starts from 10:00 am to 12:00 pm on the day of work. the tank water temperature, which is 29°c at 10:00 am, rises to 33°c at 12:00 pm. the temperature of the pv modules causes this rise in tank water temperature, as the water absorbs heat from these pv modules and then returns to the water tank. the cooling process is stopped for two hours to study the temperature gradient of the pv modules to determine how long the cooling process can last. after two hours, the cooling process begins again for one hour. the cooling unit is shown in figs. 5 and 6. the cooling unit, like with the solar reflectors, is added to the first cluster (improved cluster). when the water tank tap is opened, water flows toward the pv modules inside the perforated pipe that extends along with the improved cluster (pv modules). water then drips onto these pv modules, being collected in the gutter and eventually collected in underground water storage. the water pump then pumps water back into the tank, and this cycle continues constantly for the specific time. water loss during the cooling process is minimal in this experiment. although the water tank is enclosed in a cardboard box, it is heated by ambient heat and direct solar radiation. most of the heat is eliminated by water flowing along the gutter, then to storage, and finally to the water tank, as shown in figs. 5 and 6. figs. 5 and 6 show the schematic setup and the real setup of the cooling-solar reflector assembly, respectively. the solenoid valve device displayed in fig. 5 is utilized to control water flow when the electricity is cut off. this valve closes the water tap when the electricity is cut off, which is important because when the electricity is cut off, the submersible water pump stops and no longer lifts water to the tank. the electricity has not been cut off in the current research. fig. 5 schematic setup of the cooling-solar reflector assembly fig. 6 real setup of the cooling-solar reflector assembly 43 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 40-49 the cooling process begins when the tank water tap is opened and water flows through the perforated pipe (upper header). water falls through the perforations and absorbs the heat from the pv modules. when water falls on the pv modules, it flows down and accumulates in a long gutter that extends along the improved pv modules, and then it flows toward the underground water storage and returns to the water tank via the submersible water pump. when water falls on the pv modules and flows down, it travels 10 meters inside the gutter from the last pv module to the underground water tank. this traveled distance makes the water emit heat, and some water heat is also emitted when it accumulates in the underground water storage, which is buried at a depth of 35 cm below the earth’s surface, as illustrated in figs. 5 and 6. 5. system performance evaluation the reference and improved clusters are evaluated using the previously described performance parameters (pr, efficiency, and power). power is obtained directly from an inverter via speedwire, while the efficiency, pr, and solar irradiance are estimated using a set of equations. a digital thermometer is used to measure the temperature of the pv modules and the ambient air. 5.1. efficiency efficiency is classified into: array, system, and inverter efficiencies. the array efficiency (ηpv) is based on the dc power output, while the system efficiency (ηsys) is based on the ac energy output [18-19]. the array efficiency is the ratio of the dc energy output to the in-plane solar irradiation multiplied by the area of pv modules (pv array) [20]. the array efficiency is calculated as follows: dc pv mt e η h a = (1) the system efficiency is calculated as follows: ac sys mt e η h a = (2) where ht is the in-plane solar irradiation, am is the solar pv array area, and edc is the dc energy. the inverter efficiency is calculated as follows [21]: ac inv dc e e η = (3) the inverter efficiency ranges from 96% to 97% because it is indoor [22]. 5.2. performance ratio (pr) pr is a dimensionless quantity that indicates the overall effect of losses (due to inverter inefficiency and cell temperature, losses in wiring and protection diodes, poor module at low irradiance, partial shading, module mismatch, etc.) on the rated power capacity [23]. it is defined as the ratio of ac energy (eac) delivered to the grid to the energy production of an ideal loss-less pv system with 25°c cell temperature and the same solar irradiation. pr allows comparison of pv systems regardless of azimuth angle, tilt angle, solar radiation resources and their nominal (rated) power capacity, etc. [24-25]. it is given as follows: f r y pr y = (4) where yf is the final yield estimated in eq. (5), and yr is the reference yield estimated in eq. (6). 44 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 40-49 the final yield (yf) is the yearly, monthly, or even daily output ac energy of the pv system divided by the rated power of the connected pv system at standard test conditions (1 kw/m 2 and 25°c cell temperature) [26-27]. the intention to compare a similar solar pv power plant in a specific geographic region leads to the formulation of this representative. yf is given as follows: ac pf pv.rated kwh e y ( p kw )= (5) where eac is the ac energy output in kwh, and ppv.rated is the system rated power. the reference yield (yr) is defined as the ratio of total in-plane solar irradiation (kwh/m2) to the reference irradiance (1 kw/m 2 ). this parameter represents an equal number of hours at the reference irradiance and is given by [28-29]: p t r r h y (kwh kw ) h = (6) where hr and ht are the reference irradiance and the in-plane solar irradiation (s.ir), respectively. when the eqs. (5) and (6) are substituted into eq. (4), eq. (7) is attained [30]. rac tpv ,rated e h pr p h = (7) pr can also be given in eq. (8), which is used to estimate pr in this study. lactua ref η pr η = (8) the actual power (pac) and rated power (prated) are estimated by eqs. (9) and (10) as follows: mrac actual p h η a= (9) mrrated ref p h η a= (10) the actual efficiency (ηactua) is estimated by eq. (11) as follows: actual mref ref η η 1 β t )]( t[= − − (11) where ηref is the rated efficiency (15.2%), am is the area of reference and improved pv modules (13.04 m 2 each), β is the temperature coefficient that equals -0.3%/°c, tref is the reference modules’ temperature (25°c), and tm is the actual pv modules’ temperature. from eq. (10), eq. (12) is attained as follows. ac r m actual p h a η = (12) where pac, am, and ηactual are the actual power, improved cluster area, and actual efficiency, respectively. eqs. (12) and (8) are used to estimate the solar irradiance (hr) and actual efficiency (ηactual) respectively. power increment percentage (pincp) is given by eq. (13) as follows: im ref ref pincp (p p ) / p= − (13) where pim and pref are the power of the improved and reference clusters respectively. 45 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 40-49 since the reference cluster contains 18 pv modules and the improved cluster contains 12 pv modules, before all the calculations, the power of the reference cluster is divided by 18 and multiplied by 12 to accomplish an equalization in power (number of pv modules) between the improved and reference clusters, so that each cluster contains 12 pv modules. the cooling-reflector assembly means the water-cooling unit and solar reflectors. 6. results and discussion fig. 7 shows the performance increment percentage (pr incp), as well as the pr of the reference and improved pv modules (clusters). the maximum values of pr incp (pr gain), reference pr, and improved pr at 12:00 pm are 6.3%, 91.2%, and 97.5%, respectively, where the temperatures of reference pv modules (tref), improved pv modules (timp), and ambient air are 32°c, 55°c, and 30°c, respectively. when the cooling process is stopped, the pr of the improved pv modules decreases gradually until it equals the pr of the reference pv modules after 25-33 seconds. this is because when the cooling process stops, the temperature of the improved pv modules increases due to the solar reflectors. this means that solar reflectors reduce the pr by raising the temperature of the pv solar modules while increasing the power output of the pv modules. pr is an important parameter because it reveals all of the impacts of losses on the grid-connected pv system. the pr of a pv system indicates how close it approaches ideal performance during real operation and allows comparison of pv systems independent of location, tilt angle, orientation, and their nominal rated power capacity. fig. 7 shows that the pr increases and decreases when the temperature of the pv modules decreases (during the cooling process) and increases (when the cooling process stops). fig. 8 shows the efficiency increment percentage (η incp), as well as the efficiency of the reference and improved pv modules (clusters). the maximum value of efficiency incp (the gain) at 10:30 pm is 1%, where the reference and improved pv modules efficiencies are 13.7% and 14.7%, respectively. the temperatures of the reference pv modules (tref), improved pv modules (timp), and ambient temperature are 55°c, 32°c, and 30°c, respectively. the maximum efficiency values synchronize with the lowest ambient temperature, namely at the beginning of sunrise and sunset as displayed in fig. 8, where the efficiencies of the reference and improved pv modules are 14.7% and 14.8%, respectively. the efficiency increases with cooling and decreases when using solar reflectors without cooling because they increase the temperatures of the pv modules, while the efficiency increases when using cooling with solar reflectors, as shown in fig. 8. fig. 9 illustrates the power output of the improved and reference clusters. in this figure, the pincp bars and the power curve of the improved pv modules have crests and bottoms. the cooling process causes the crests, whereas stopping the cooling process causes the bottoms. the maximum power values of the reference and improved pv modules at 12:00 pm are 1.69 kw and 2.55 kw, respectively, as shown in fig. 9. the maximum pincp (power gain) at 12:00 pm is 51.3%. the daily average of pincp is 32%. this gain is large and has a significant influence on the economic feasibility of installing a pv solar system. each technique (cooling and solar reflectors) has an impact on a different aspect. the cooling technique increases the voltage by increasing the energy band gap of the semiconductor, reducing the potential barrier resistance. while the solar reflector technique increases the current by increasing electrons, the two techniques together enhance the electrical power of the pv solar system. fig. 10 illustrates the solar irradiance increase percentage (s.ir incp), as well as the reference and improved pv modules’ solar irradiance (s.ir). the maximum values of s.ir incp and the reference and improved pv modules (s.ir) at 12:00 pm are 43%, 938.7 w/m 2 , and 1338.7w/m 2 , respectively. the minimum values at 6:00 pm are 3%, 88.6w/m 2 , and 88.7w/m 2 , respectively. the incp value (43%) corresponds to 403.6 w/m 2 . the daily average of s.ir incp is 28%, which corresponds to 272w/m 2 . during the time period from 4:00 pm to 6:00 pm and from 7:30 am to 8:00 am, the solar irradiance values of the reference and improved pv modules converge, because the solar radiation reflected from the solar reflectors to the pv modules is deflected away from them. in contrast, during midday, when there is no deflection of solar radiation, it is directly reflected on the pv modules. as a result, the improved cluster gets more solar radiation than the reference cluster, resulting in a divergence in s.ir values between the two clusters. 46 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 40-49 0 10 20 30 40 50 60 70 86.0% 88.0% 90.0% 92.0% 94.0% 96.0% 98.0% 0 7 :3 0 a m 0 8 :0 0 a m 0 8 :3 0 a m 0 9 :0 0 a m 0 9 :3 0 a m 1 0 :0 0 a m 1 0 :0 7 a m 1 0 :1 5 a m 1 0 :3 0 a m 1 0 :4 5 a m 1 1 :0 0 a m 1 1 :1 5 a m 1 1 :3 0 a m 1 1 :4 5 a m 1 2 :0 0 p m 1 2 :1 5 p m 1 2 :3 0 p m 1 2 :4 5 p m 1 2 :5 0 p m 0 1 :0 0 p m 0 1 :1 5 p m 0 1 :2 0 p m 0 1 :3 0 p m 0 1 :4 5 p m 0 2 :0 0 p m 0 2 :3 0 p m 0 2 :4 5 p m 0 3 :0 0 p m 0 3 :1 0 p m 0 3 :2 0 p m 0 3 :3 0 p m 0 3 :4 0 p m 0 3 :5 0 p m 0 4 :0 0 p m 0 4 :1 5 p m 0 4 :3 0 p m 0 4 :4 5 p m 0 5 :0 0 p m 0 5 :3 0 p m 0 6 :0 0 p m p v m o d u le s t e m p e ra tu re p e rf o rm a n ce r a ti o day time improved cluster pr % reference cluster pr % timp in celsius tref in celsius 0 10 20 30 40 50 60 70 13.0% 13.2% 13.4% 13.6% 13.8% 14.0% 14.2% 14.4% 14.6% 14.8% 15.0% 0 7 :3 0 a m 0 8 :0 0 a m 0 8 :3 0 a m 0 9 :0 0 a m 0 9 :3 0 a m 1 0 :0 0 a m 1 0 :0 7 a m 1 0 :1 5 a m 1 0 :3 0 a m 1 0 :4 5 a m 1 1 :0 0 a m 1 1 :1 5 a m 1 1 :3 0 a m 1 1 :4 5 a m 1 2 :0 0 p m 1 2 :1 5 p m 1 2 :3 0 p m 1 2 :4 5 p m 1 2 :5 0 p m 0 1 :0 0 p m 0 1 :1 5 p m 0 1 :2 0 p m 0 1 :3 0 p m 0 1 :4 5 p m 0 2 :0 0 p m 0 2 :3 0 p m 0 2 :4 5 p m 0 3 :0 0 p m 0 3 :1 0 p m 0 3 :2 0 p m 0 3 :3 0 p m 0 3 :4 0 p m 0 3 :5 0 p m 0 4 :0 0 p m 0 4 :1 5 p m 0 4 :3 0 p m 0 4 :4 5 p m 0 5 :0 0 p m 0 5 :3 0 p m 0 6 :0 0 p m p v m o d u le s t e m p e ra tu re e ff ic ie n cy day time improved cluster η % reference cluster η % timp in celsius tref in celsius 0% 10% 20% 30% 40% 50% 60% 0 0.5 1 1.5 2 2.5 3 0 7 :3 0 a m 0 8 :0 0 a m 0 8 :3 0 a m 0 9 :0 0 a m 0 9 :3 0 a m 1 0 :0 0 a m 1 0 :0 7 a m 1 0 :1 5 a m 1 0 :3 0 a m 1 0 :4 5 a m 1 1 :0 0 a m 1 1 :1 5 a m 1 1 :3 0 a m 1 1 :4 5 a m 1 2 :0 0 p m 1 2 :1 5 p m 1 2 :3 0 p m 1 2 :4 5 p m 1 2 :5 0 p m 0 1 :0 0 p m 0 1 :1 5 p m 0 1 :2 0 p m 0 1 :3 0 p m 0 1 :4 5 p m 0 2 :0 0 p m 0 2 :3 0 p m 0 2 :4 5 p m 0 3 :0 0 p m 0 3 :1 0 p m 0 3 :2 0 p m 0 3 :3 0 p m 0 3 :4 0 p m 0 3 :5 0 p m 0 4 :0 0 p m 0 4 :1 5 p m 0 4 :3 0 p m 0 4 :4 5 p m 0 5 :0 0 p m 0 5 :3 0 p m 0 6 :0 0 p m p in c p im p ro v e d a n d r e fe re n ce c lu st e rs p o w e r day time pincp (%) improved cluster power (kwp) reference cluster power (kwp) 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 0 200 400 600 800 1000 1200 1400 0 7 :3 0 a m 0 8 :0 0 a m 0 8 :3 0 a m 0 9 :0 0 a m 0 9 :3 0 a m 1 0 :0 0 a m 1 0 :0 7 a m 1 0 :1 5 a m 1 0 :3 0 a m 1 0 :4 5 a m 1 1 :0 0 a m 1 1 :1 5 a m 1 1 :3 0 a m 1 1 :4 5 a m 1 2 :0 0 p m 1 2 :1 5 p m 1 2 :3 0 p m 1 2 :4 5 p m 1 2 :5 0 p m 0 1 :0 0 p m 0 1 :1 5 p m 0 1 :2 0 p m 0 1 :3 0 p m 0 1 :4 5 p m 0 2 :0 0 p m 0 2 :3 0 p m 0 2 :4 5 p m 0 3 :0 0 p m 0 3 :1 0 p m 0 3 :2 0 p m 0 3 :3 0 p m 0 3 :4 0 p m 0 3 :5 0 p m 0 4 :0 0 p m 0 4 :1 5 p m 0 4 :3 0 p m 0 4 :4 5 p m 0 5 :0 0 p m 0 5 :3 0 p m 0 6 :0 0 p m s .i r in c p s o la r ir ra d ia n ce ( s .i r) w /m 2 day time s.ir incp improved cluster s.ir reference cluster s.ir fig. 7 pr incp vs pr of the improved and reference clusters fig. 8 η incp vs the efficiency of improved and reference clusters (pv modules) fig. 9 pincp vs the power of improved and reference clusters fig. 10 s.ir incp vs the s.ir of reference and improved clusters 47 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 40-49 fig. 11 displays the temperature of the reference pv modules (tref), the temperature of the improved pv modules (timp), and ambient temperature. the maximum values of tref and timp are recorded at 12:00 pm at 55°c and 32°c, respectively, at an ambient temperature of 30°c. the minimum values are recorded at 6:00 pm at 32°c and 31°c respectively, at the ambient temperature of 26°c. the temperature of improved pv modules drops by 23°c due to cooling. this indicates that, despite the use of solar reflectors, which increases heat, the cooling technique is very effective in removing heat from the pv modules. the cooling unit increases voltage by decreasing resistance. the maximum ambient temperature occurs at 3:00 pm, whereas the maximum pv module temperature occurs at 12:00 pm. the timp curve has a top and a bottom. when the cooling process stops, the top appears, and when the cooling process begins, the bottom appears. the temperature of the pv modules drops rapidly at the beginning of the cooling process and slowly rises when the cooling process stops until it stabilizes at the final temperature, as shown in fig. 11. 7. conclusions because of the use of cooling-solar reflectors, significant gains in performance parameters (efficiency, pr, and power) and solar radiation are made. the power gained (pincp) is estimated at 32% as a daily average. pr and efficiency are enhanced by 6% and 1%, respectively. as a daily average, the achieved solar irradiance (s.ir incp) is estimated to be 31%. the maximum power output gain value is 51%. the cooling technique has a considerable impact on reducing heat generated by pv modules. the water used for cooling has a considerable impact on the cleaning of pv modules. the cooling unit reduces the temperature of the pv modules by 23°c, permitting solar reflectors to be added to the upper and bottom edges of the pv solar modules to increase power output even more. when utilizing solar reflectors without cooling, pr and efficiency decrease because they raise the temperature of the pv modules. based on the results of this study, it is recommended that researchers improve the performance of pv solar modules by cooling the back side of the pv modules and adding solar reflectors to the upper and bottom edges of the pv modules to increase power output even more. conflicts of interest the authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence this study. references [1] r. karmouch, et al., “solar cells performance reduction under the effect of dust in jazan region,” journal of fundamentals of renewable energy and applications, vol. 7, no. 2, pp. 1-4, march 2017. [2] a. h abedin, et al., “a critical review of thermochemical energy storage systems,” the open renewable energy journal, vol. 4, no. 1, pp. 42-46, august 2011. 0 10 20 30 40 50 60 70 0 7 :3 0 a m 0 8 :0 0 a m 0 8 :3 0 a m 0 9 :0 0 a m 0 9 :3 0 a m 1 0 :0 0 a m 1 0 :0 7 a m 1 0 :1 5 a m 1 0 :3 0 a m 1 0 :4 5 a m 1 1 :0 0 a m 1 1 :1 5 a m 1 1 :3 0 a m 1 1 :4 5 a m 1 2 :0 0 p m 1 2 :1 5 p m 1 2 :3 0 p m 1 2 :4 5 p m 1 2 :5 0 p m 0 1 :0 0 p m 0 1 :1 5 p m 0 1 :2 0 p m 0 1 :3 0 p m 0 1 :4 5 p m 0 2 :0 0 p m 0 2 :3 0 p m 0 2 :4 5 p m 0 3 :0 0 p m 0 3 :1 0 p m 0 3 :2 0 p m 0 3 :3 0 p m 0 3 :4 0 p m 0 3 :5 0 p m 0 4 :0 0 p m 0 4 :1 5 p m 0 4 :3 0 p m 0 4 :4 5 p m 0 5 :0 0 p m 0 5 :3 0 p m 0 6 :0 0 p m t e m p e ra tu re i n c e ls iu s day time timp in celsius tref in celsius ambient temperature in celsius fig. 11 ambient temperature, tref, and timp 48 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 40-49 [3] m. z. jacobson, et al., “a path to sustainable energy by 2030,” scientific american, vol. 301, no. 5, pp. 58-65, november 2009. [4] a. n. al-shamani, et al., “experimental studies of rectangular tube absorber photovoltaic thermal collector with various types of nanofluids under the tropical climate conditions,” energy conversion and management, vol. 124, pp. 528-542, september 2016. [5] h. k. elminir, et al., “effect of dust on the transparent cover of solar collectors,” energy conversion and management, vol. 47, no. 18-19, pp. 3192-3203, november 2006. [6] s. c. seitel, “collector performance enhancement with flat reflectors,” solar energy, vol. 17, no. 5, pp. 291-295, november 1975. [7] h. tabaei, et al., “improving the effectiveness of a photovoltaic water pumping system by using booster reflector and cooling array surface by a film of water,” iranian journal of science and technology transactions of mechanical engineering, vol. 39, no. m1, pp. 51-60, april 2015. [8] s. a. zubeer, et al., “experimental and numerical study of low concentration and water-cooling effect on pv module performance,” case studies in thermal engineering, vol. 34, article no. 102007, june 2022. [9] m. pavlov, et al., “experimental and numerical study of the influence of string mismatch on the yield of pv modules augmented by static planar reflectors,” ieee journal of photovoltaics, vol. 5, no. 6, pp. 1686-1691, september 2015. [10] h. m. bahaidarah, et al., “experimental evaluation of the performance of a photovoltaic panel with water cooling,” ieee 39th photovoltaic specialists conference, pp. 2987-2991, june 2013. [11] s. khaled, et al., “numerical and experimental investigation for hybrid photovoltaic/thermal collector system in duhok city,” journal of environmental engineering and landscape management, vol. 28, no. 4, pp. 202-212, december 2020. [12] h. m. bahaidarah, et al., “a combined optical, thermal and electrical performance study of a v-trough pv system—experimental and analytical investigations,” energies, vol. 8, no. 4, pp. 2803-2827, april 2015. [13] j. a. duffie, et al., solar engineering of thermal processes, photovoltaics, and wind, hoboken: john wiley & sons, 2020. [14] e. z. ahmad, et al., “outdoor performance evaluation of a novel photovoltaic heat sinks to enhance power conversion efficiency and temperature uniformity,” case studies in thermal engineering, vol. 31, article no. 101811, march 2022. [15] c. s. sangani, et al., “experimental evaluation of v-trough (2 suns) pv concentrator system using commercial pv modules,” solar energy materials and solar cells, vol. 91, no. 6, pp. 453-459, march 2007. [16] a. n. abed., “performance improvement of cigs pv solar grid tied system using planer concentrators, case study: baghdad,” ph.d. dissertation, department of atmospheric science, college of science, mustansiriyah university, 2020. [17] u. yilmaz, et al., “improved mppt method to increase accuracy and speed in photovoltaic systems under variable atmospheric conditions,” international journal of electrical power and energy systems, vol. 113, pp. 634-651, december 2019. [18] l. c. de lima, et al., “performance analysis of a grid connected photovoltaic system in northeastern brazil,” energy for sustainable development, vol. 37, pp. 79-85, april 2017. [19] a. n. abed, et al., “efficiency and performance improvement via using optical reflectors of on-grid cigs pv solar system,” karbala international journal of modern science, vol. 6, no. 1, article no. 5, march 2020. [20] n. m. kumar, et al., “performance analysis of 100 kwp grid connected si-poly photovoltaic system using pvsyst simulation tool,” energy procedia, vol. 117, pp. 180-189, june 2017. [21] k. attari, et al., “performance analysis and investigation of a grid-connected photovoltaic installation in morocco,” energy reports, vol. 2, pp. 261-266, november 2016. [22] n. k. kasim, et al., “studying the performance of second-generation pv solar technology under baghdad climate,” research journal in advanced sciences, vol. 1, no. 2, pp. 37-48, october 2020. [23] a. m. khalid, et al., “performance ratio–crucial parameter for grid connected pv plants,” renewable and sustainable energy reviews, vol. 65, pp. 1139-1158, november 2016. [24] t. ozden, et al., “long term outdoor performances of three different on-grid pv arrays in central anatolia–an extended analysis,” renewable energy, vol. 101, pp. 182-195, february 2017. [25] n. m. obaid, et al., “performance analyses of 15 kw grid-tied photo voltaic solar system type under baghdad city climate,” journal of engineering, vol. 26, no. 4, pp. 21-32, march 2020. [26] v. sharma, et al., “performance analysis of a 190 kwp grid interactive solar photovoltaic power plant in india,” energy, vol. 55, pp. 476-485, june 2013. [27] n. m. obaid, et al., “performance assessment of first grid-tied pv solar system under baghdad city climate condition,” iraqi journal of science and technology, vol. 10, no. 1, pp. 63-71, october 2019. [28] m. s. adaramola, et al., “preliminary assessment of a small-scale rooftop pv-grid tied in norwegian climatic conditions,” energy conversion and management, vol. 90, pp. 458-465, january 2015. [29] h. rezk, et al., “energy performance analysis of on-grid solar photovoltaic system–a practical case study,” international journal of renewable energy research, vol. 9, no. 3, pp. 1292-1301, september 2019. [30] n. k. kasim, et al., “performance analysis of grid-connected cigs pv solar system and comparison with pvsyst simulation program,” international journal of smart grid, vol. 3, pp. 172-179, december 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 (http://creativecommons.org/licenses/by/4.0/). 49  proceedings of engineering and technology innovation, vol. 1, 2015, pp. 01 04 1 cop y right © taeti the study of crew physical and psychological evaluation model and sports intervention method guo-qing li * , bin zhang department of physical education, harbin engineering university, harbin, 150001, china. received 13 june 2015; received in revised form 27 july 2015; accept ed 01 august 2015 abstract it is very important to research the physical and psycho logy of crew for ship safe. the artificial neural network theory is applied to the crew health evaluation, according to the crew health evaluation index. the corresponding bp neural network for health state evaluation established. training samples of neural network obtained using the expert experience and st atistical data. with the aid of computer and matlab software, the evaluation results can be conveniently achieved. bp neural network model that was trained can be operated "off-line", evaluate the crew physical and psychological health status accurately and effectively. and the guidance and recommendations of sports items intervention measures were given by crew health expert system. all these research will help the crew to improve the level of physical and psychology health. ke ywor ds : physical and psychological evaluation, sports intervention, bp neural network 1. introduction fig. 1 block diagram range of study ocean shipping has the characteristics of drifting time longer at sea, away from land and shore. so the crew's phys iological status, psychological status and level of technology have put forward higher requirements. it has been shown that human misplay is the main cause of ship accident factors from the large number of maritime case analysis at internal and abroad. therefore, it is of great significance to analyze and research the physical and psychological status. which can take the corresponding methods and preventive measures to reduce the dynamic human misplay factors , as we known these factors would cause ship mechanical accident. the research of human factors and ergonomics is much, but it is not much to achieve the practical application. especially the establishment of crew physical and psychological health evaluation model is rarely. this paper will made crew physical and psychological health evaluation model for ocean shipping at sea eventually, to track and evaluate crew health status, take intervention measures, reduce human accidents. system frame as shown in fig. 1. the research field of this paper is in the lengthways range of dashed. 2. theory of bp neural network bp neural network trained the network using back propagation method, and then network weights can be got. the main idea is modify the weight or threshold to make the error function decline along the negative gradient direction. signal transmission and error back-propagation are two process procedures of bp neural network learning. in the forward prop agation, the signal is inputted from input layer, and propagated to output layer after processed by hidden layer. each neuron only could affect the neurons state of the next layer. the error propagates back if desired output has not been got. output error which is the basis to modify the neurons weight back propagate to input layer through hidden layer, repeat until output error permitted[1-3]. essentially, bp neural network is a nonlinear optimization problem issues for a set of input and output samples. it can be seen as a mapping from n-dimensional to m-dimensional [4]. bp neural network structure is shown in fig. 2. input layer output layer hidden layer fig. 2 structure of neural network corresponding aut hor. e-mail address: liguoqing@hrbeu.edu.cn proceedings of engineering and technology innovation , vol. 1, 2015, pp. 01 04 2 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti 3. crew physical and psychological health evaluation model definition: the research is limited in the dynamic factors of the ocean crew physical and psychological health evaluation index, variable factors, adjustable factors, controllable factors) 3.1. index of physical and psychological health the key research question of this paper is to abstract characteristic variable that can reflect the state of physical and psychological from the index of crew physical and psych ological health. however it is gratified for researchers that a series of index can be adjusted for model and as the application of the model, the extraction index of evaluation system will be closer to the crew physical and psychological state. embodied in the bp neural network model structure is to use the model as simple as possible to evaluate accurately the crew physical and mental health status. 3.2. the modeling of crew physical and psychological health the theory has been proven three layer forward feedback neural network can approximate arbitrary continuous function with arbitrary accuracy. the bp neural network of this paper is three layer structures [5-6]. the input signal is move the spread to the hidden layer nodes, through the transfer function then output signal of hidden layer nodes is transmitted to the output layer. transfer function from the input layer to the hidden layer and hidden layer to the output layer used the tansig function and purelin function. the theory has been proven three layer forward feedback neural network can approximate arbitrary continuous function with arbitrary accuracy. the bp neural network of this paper is three layer structures. the input signal is move the spread to the hidden layer nodes, through the transfer function then output signal of hidden layer nodes is transmitted to the output layer. transfer function from the input layer to the hidden layer and hidden layer to the output layer used the tansig function and purelin function. the ocean crew physical and psychological health at sea evaluation model chose the inde x of fat igue, sleep, appetite, food egestion, loneliness, anxiety, depression, panic, e motionality tension, thinking ability, self -control ability as the input of bp neural network, and crew health level, health, sub-health, unhealthy, serious unhealthy as output. the nu mber of h idden nodes depends on many factors such as the quantity of training sample and the rule comple xity of samp le contains. according to experience hidden nodes is 8. so me of the variables results of the physical and psychological health were chosen as bp neural network tra ining set. the tra ining samp les are shown in table 1. table 1 the training sample set of bp neural network m1 m2 m3 m4 m5 m6 m7 m8 m9 m10 m11 d1 0.20 0.05 0.10 0.10 0.10 0.00 0.05 0.00 0.00 0.00 0.00 d1 0.20 0.05 0.05 0.05 0.05 0.00 0.05 0.00 0.00 0.05 0.00 d1 0.00 0.10 0.00 0.00 0.00 0.10 0.00 0.05 0.00 0.10 0.00 d2 0.10 0.20 0.00 0.00 0.50 0.20 0.20 0.00 0.00 0.00 0.00 d2 0.00 0.05 0.35 0.10 0.30 0.50 0.20 0.20 0.10 0.10 0.10 d2 0.00 0.10 0.00 0.00 0.00 0.20 0.10 0.20 0.15 0.00 0.20 d3 0.20 0.75 0.10 0.10 0.10 0.30 0.00 0.00 0.30 0.20 0.20 d3 0.60 0.05 0.05 0.05 0.25 0.15 0.15 0.60 0.80 0.25 0.40 d3 0.25 0.10 0.00 0.00 0.60 0.10 0.05 0.05 0.30 0.10 0.20 d4 0.00 0.60 0.50 0.50 0.60 0.10 0.10 0.10 0.10 0.30 0.40 d4 0.90 0.30 0.40 0.30 0.10 0.70 0.15 0.25 0.60 0.20 0.20 d4 0.60 0.60 0.05 0.05 0.00 0.15 0.15 0.75 0.10 0.35 0.40 d-health level, d1hea lth, d2sub-health, d3unhealthy, d4seriously unhealthy, mindex of health status, m1fatigue, m2sleep, m3appetite, m4food egestion, m5lonely, m6an xiety, m7depression, m8panic, m 9-e motional tension, m10thinking ability, m11self-control ability since centralized degree of the sample data is high, the data is processed by norma lization. every feature is in the 0-1 interval. the interval value indicated the degree, for e xa mple : if the sleep state is good, then the value is recorded 0.00, insomnia recorded 1.00. the state of sleep quality drop a little , but can be a good rest, recovery is perfect, recorded 0.60. sleep quality drops badly, the re covery is bad, recorded 0.40. mat la b software is used to train sample set. 3.2. the modeling of crew physical and psychological health after network training, the test will be carried on. the purpose of network test is to see whether the network has good generalization ability. select 4 set of data as the network testing input, the test results are shown in fig. 4. from fig. 3-4 we can see, the generalization ability of the network is good, you can use the following bp neural network for ocean crew to evaluate health state. fig. 3 output of no. 1 test sample fig. 4 output of no. 2 test sample proceedings of engineering and technology innovation , vol. 1, 2015, pp. 01 04 3 cop y right © taeti table 2 the test sample of network no. m1 m2 m3 m4 m5 m6 m7 m8 m9 m10 m11 1 0.15 0.00 0.00 0.00 0.60 0.35 0.40 0.00 0.10 0.10 0.10 2 0.80 0.70 0.10 0.10 0.30 0.30 0.50 0.10 0.25 0.20 0.20 4. the sports items intervention expert system the crew should strengthen the physical e xe rcise, learn to rela x themselves. according to the ship chara cteristic, it is beneficia l to health to adjust measures to local conditions to carry out some physical e xerc ise and a proper a mount of entertain ment activ ities. the health will be improved .working state is energetic. the ability that copes with stress will be enhanced. the crew rela xed to control themselves when they meet pressure and feet nervous. all these main ly need the crew own conscious physical training in different ways. the crew sports consciousness is weakness, some of them do not grasp the scientific physical e xe rc ise knowledge and methods, have no phys ical e xerc ise habits. and the work burden is heavier during navigation, phy sical and mental fat igue, sports inert is larger, the necessary physical guidance and supervisory mechanism is not pe rfect. this research will eventually build physical and psychological health state expert system of ocean crew sports intervention to help them providing refe rences and suggestions that how to choose sports items at sea. 4.1. sports project selection principle of crew health state intervention according to the physical and psychological health state as mentioned above, this paper put forward the e xercise treat ment, chose several easy to carry out on board. these sports have better effect on co mmon physical and psychological illness. brief introduction is as followed [4-6]. (1) tai chi body jo ints, muscles and bones will each pull, press and be diastolic, when p laying ta i chi. massage effect will be in visceral due to abdominal breathing, and it is beneficia l for intestinal peristalsis with increasing of dia phragmatic upper and lowe r a mp litude. it is benefic ial for them to pract ice ta i ch i at sunrise or sunset on the sp acious deck in the vast sea. tai chi can train the cre w to be happy and pleased with them, and be in a calm mood. (2) ear action practice the helix covers the ears when doing exe rcise, rub the earhole with the root of palm (i.e. massage earhole that be covered by helix), rub to the ear feel itch, when the blood circulate to the ear. it is help to relief tinnitus. (3) sit-ups this sports can e xerc ise the leg strength also has very good effect on the waist, of course you can use the squat method. (4) barbell practice carry barbell with shoulder, helps to imp rove rena l function, also can achieve the effect of allev iating tinnitus. (5) walking on the deck walking e xerc ise method is very simple, because the tonnage of ocean shipping is very la rge. the b road deck area is equivalent to two football fields. the crew can ma ke full use of the convenient conditions, when the rising in every morn ing the crew can enjoy the beauty on the sea and walk or jog on the open deck. this sport not only may increase the cardiopulmonary function, but also can cu ltivate their origina l nature, condition psychology. this sports mode has good effect on rehabilitation and preve ntion of bronchitis for crew. (6) climbing gangway ladder a certain number of gangways are in ship limited area. the crew can enhance their ability of respiratory system by climb ing the gangway ladder coordinating with the move ment of breath train ing. first suction with the nose, then narrow lips to breathe out, each climb2 steps breathe again, this sport can be realized on every ship. (7) comfortable slow motion in cabin the crew work on board fo r long -term, often c limb cabin, so is very tired. in e course of time they will be facing lu mbago and skelalgia, which cannot relieve by energetic exe rcise, but massage, and do moderate exe rcise. for deck narrow, poor conditions of the ship, the crew can take co mfortable and slow mot ion in cabin to fo r rehabilitation. the concrete forms can reference interface of expert system. please see figs . 6-7. 4.2. the interface of crew health evaluation model and sports items intervention expert system fig. 6 the interface of crew health evaluation model fig. 7 the interface of sports items intervention expert system proceedings of engineering and technology innovation , vol. 1, 2015, pp. 01 04 4 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti 5. conclusions the crew health evaluation model based on bp network and the sports items intervention expert system established in the paper. the system can track and guide n the crew physical and psychological health status at sea in real time. the research work has important practical significance and great practical value, which may enhance the crew physical and psycholog ical health, improve work efficiency, promote the harmony and stability of the ship small environment, maintain the normal life order and ensure safety at sea. china ocean ships have no crew physical and psycholog ical evaluation system in using at present. although training data is limited to the present study training model, this study software is carried in ship will have a large number of training data, and classification ability of the network will be constantly improved. references [1] l. c. jiao, neural network calculation, xi’an xidian univer sity press, 1996. (in chinese) [2] c. r. yuan, artificial neural networks and its application, bei jing tsinghua university press, 1999. (in chinese) [3] c. d. xing, j. m. yang, q. s. ren, theory of artificial intelligent and its application, shang hai dong hua university press, 2005. (in chinese) [4] m. a. gonzalez, “an integrated logistics support system for training crew medical officers in advanced cardiac life s upport,” ph.d. dissertation, dept. industrial engineering, cullen college of engineering, houston univ., houston, tx, 1995. [5] g. q. li, “discuss effect of physical exercise on the positive role of sub health,” journal of yichun university, vol. 32, no. 12, 2010. [6] m. q. chen, f. y. qiao, “research on s ports effect of occupation disease prevention and treatment for crew,” science and technology innovation herald, no. 35, 2011. (in chinese)  proceedings of engineering and technology innovation, vol. 1, 2015, pp. 27 31 27 cop y right © taeti an analysis of managing sustainable competitiveness for semiconductor manufacturers hung-chi hsiao 1,* , taketoshi hibiya 1 , masaru nakano 1 , yih-lin cheng 2 , hung-ching wen 3 1 graduate school of system design and management, keio university, yokohama, japan . 2 department of mechanical engineering, national taiwan university of science and technology, taipei, taiwan. 3 department of management science, national chiao tung university, hsin chu, taiwan. received 16 april 2015; received in revised form 05 may 2015; accept ed 18 june 2015 abstract innovative capability is considered inevitable for firms to sustain their competitiveness. in the recent rapidly changing global competition environment, the traditional integrated device manufacturer (idm) model in semiconductor industry is facing the limitation of sustaining its profitability and co mpetitiveness. idm’s focusing both chip design and manufacturing for various application segments disperse its resources of innovating sustainable competiveness. this study develops an analysis framework with incorporating data envelopment analysis (dea) approach to measure the efficiency through proper input and output variables setting. this framework aims at providing guidelines for developing firm’s business and technology strategies. we conducted a dea analysis by co llecting financial data from twenty-six leading semiconductor manufacturing companies, including twenty idms and six foundries. the results reveal that the foundry companies have higher competitive efficiency than those of idms. the empirical analysis suggests that adopting the asset-light business model may provide idms a better resource allocation and help the increase of relative efficiency scores. ke ywor ds : data envelopment analysis, integrated device manufacturer, foundry, efficiency 1. introduction the semiconductor industry has grown to be worth more than 300 billion u.s. dollars annually [1]. the industry is driven by continuous strong demand from downstream segments, such as computers, communications equipment, co nsumer electronics and industrial instruments. as global co mpetition intensifies and commoditization increases, semico nductor companies rightfully conclude that distinctive capabilities by innovating continuously are needed to attain and sustain high performance. producing the right product in the right quantities at a competitive cost is the keystone of innovation. the worldwide semiconductor industry is undergoing several forms of business model change. pure integrated d evice manufacturer (idm), asset-light idm, and pure integrated circuit (ic) design (fabless) are three distinct types of semiconductor business models. the pure idm model is that a semiconductor company which designs, manufactures, and sells ic products. the asset-light idm model maintains an internal manufacturing facility and outsources some process development and product manufacturing to silicon contract foundry companies (foundries), which operates a semico nductor fabrication facility (fab) for the purpose of fabricating the designs of other companies. fabless companies design their own chips but have no production facility so they outsource manufacturing to idms or foundries. a symbiotic relationship has been developed between the foundries and the fabless companies as they pushed each other to higher levels of co mpetency [2]. in the past two decades, idms’ overall share of the semiconductor market dropped sharply from 98% in 1992 to 75% in 2011, while the global fabless companies’ share increased 23% from 1992-2011 [3]. the fabless-foundry (de-integration) model has been pretty successful. this de-integration scheme helps both sides on reducing the uncertainties of pricy technology developments. the rapid decline in idm market shares suggests that the strong competitiveness from fabless companies and foundries makes it hard for idms to maintain core competencies in both ic design and ic manufacturing. it is crucial for semiconductor idms to understand how the business models and technology innovations they adopted affect their operation performance. the semicondu ctor industry is an interesting case which now shows rapid but partially disintegrated phenomena in the industry evolution process. the data envelopment analysis (dea) introduced by charnes et al. [4], is a nonparametric, linear programming-based methodology for identifying the relative efficiency of a homogenous set of decision making units (dmus) in the presence of multiple inputs and outputs. the efficiency score of efficient dmus is one, while inefficient dmus score b etween zero and one [5]. dea builds an efficient frontier comprising of all the efficient units, thus allowing a comparison of the best performers and providing the firms with guidelines on where and how to improve their capabilities [6]. this study applies dea to evaluate the performance of 26 leading semiconductor companies which include 20 idms and six foundry companies that possess the ic chips manufacturing competency. this analysis identifies the efficient and inefficient semiconductor companies (i.e. the dmus) by defining the proper output and input factors, and provides business managers a tool for assisting decision making of business model and technology alternatives for their organizations. this paper will detail how managers can use the results to determine areas that need specific attention, and more importantly sp ecifically what those areas need to focus on in order to become as efficient as their peer groups. the rest of the paper is structured as follows. section 2 provides the analysis framework for characterizing how a company’s performance is affected by business model and *corresponding aut hor. email: alex_hsiao@z2.keio.jp proceedings of engineering and technology innovation, vol. 1, 2015, pp. 27 31 28 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti technology innovations and addressing the dea model used to measure performance. section 3 presents the results and discussions. conclusions are given in the last section. fig. 1 ic chip’s die cost trend for full-node and half-node technologies 2. research and theoretical framework 2.1. analytical framework during the past decade, many idms watch closely at unburdening their operations and shift fro m capita l inte nsive manufacturing to a partial disintegrated asset -light idm business model. the selection of an appropriate business model needs to be made based on the conditions and constraints specific to the idms, such as forceful competition fro m the rising fab less companies and foun dries, escalating r&d costs for developing process technology, and massive investment in manufacturing capacity of state-of-the-art technologies. further, technology innovation affects semiconductor companies more than other major considerations. for instance, the availability of re -usable intellectual property (ip) bloc ks affects chip size (i.e. chip cost), the power consumption reduction of power trim technology, the innovative water immersion lithography technique using ultraviolet light at 193 n m wavelength to prolong the equipment life cycle, and the adoption of “half-node” process technology nodes. for e xa mple , the p rocess node at 150n m is ca lled a “half-node”—it offers a half step to the next technology node, between 130 and 180 n m. so me foundry customers like to co mp lete their design in one process node and then use a linear, optical shrink to actually fabricate their d evices in the half-node. this a llo ws them to take advantage of a unit cost reduction because of a smaller d ie and also achieve some performance improve ment (fig. 1). m igrating chip design using the smalle r shrink factor is less of a challenge, especially if the co mpany plans ahead. success in imple menting such shrinks requires pro -active considerations that allow e xecution at the half-node with no design rule v iolations and no yie ld or re liab ility issues 2. business model and technology innovations are contin uously being adopted by semiconductor manufacturing companies for sustaining the competitiveness, but it is difficult to do direct evaluation and comparison on the effectiveness and contribution of specific innovations due to the disparate units involved, esp ecially when involving business matters. this paper develops an analytic fra me work for semiconductor manufacturers to assist decision making in incorporating business model innovation as an essential part with technology innovation into strategic planning (fig. 2). firstly the business and technology strategies of firms could be identified based on the industry features and market trends, e.g. idm and foundry. ne xt, firms could further identify the innovations adopted (or to be adopted) corresponding to the selected strategies. then, the pre-defined re lative performance indicators of firms could be evaluated using the dea technique. finally, by analyzing the dea scores and the relat ionships between performance indicators and the innovations, firms could select salient business strategies for achieving sustainable competitiveness . fig. 2 the analytic framework 2.2. theoretical framework 2.2.1 efficiency measures by dea methods the dea approach was introduced by charnes et al. charnes , cooper and rhodes (ccr) model [4] is based on the assumption of constant returns to scale (crs) when calculating the technical efficiency (te). subsequently, banker et al. [7] proposed the bcc model that assumes variable returns to scale (vrs) and formed a more restricted feasible region than the ccr model. bcc model provides pure technical efficiency (pte) scores greater than or equal to those obtained assuming crs. a ratio of te to pte provides a measu rement of scale efficiency (se) [8]. assume that the objective of each dmu is to minimize its inputs, keeping the output level constant in the variable returns to scale (vrs). the pure technical efficiency (pte) of the target dmuo (o=1,…, n) can be computed as a solution to the following linear programming problem [9] pteo = min o (1) s.t. ,,...,1, 1 mixx ioo n j jij    ,,...,1, 1 sryy ro n j jrj    .0;1 1   j n j j  where n is the number of dmu; m and s are the number of inputs and outputs, respectively. let xij and yrj be the amount of the ith input consumed and the amount of the rth output produced by jth dmu, respectively. the pte of the target dmuo is defined as p t e equal to o. by varying the inde x “o” over all dmus, we a rrive at the pte in each dmu. if pt e is equal to one, then the dmuo is technically efficient. if pt e is sma ller than 1, then the dm uo is technically ineffic ient. if ∑ 𝜆𝑗 = 1 𝑛 𝑗=1 is dropped from eq. (1), then the technology is said to exhibit constant returns to scale (crs). the technical efficiency (te) of the target dmuo is defined as te equal to o under the input-oriented crs model [4]. business strategy technology strategy business model innovation technology innovation company performance company relative efficiency dea method technology innovation • re-usable ip • power trim • immersion lithography • smart shrink (half-node) business model innovation • design platform alliance i.e. design reference flow, design for manufacturing (dfm) • service innovation i.e. multi-layer-mask (mlm) multi-project wafer (mpw) • technology development alliance i.e. common platform alliance mass production start + 1 y ear + 2 y ear volume / time die cost technology a technology b half node technology migration : • require brand new design with risks • longer dev elopment with bigger inv estment • larger die shrink merit. smart shrink (half node) : • less risk • minimize the non recurring engineering (nre) dev elopment • quick return on inv estment proceedings of engineering and technology innovation, vol. 1, 2015, pp. 27 31 29 cop y right © taeti se equal to unity indicates that this dmu is operating at the most productive scale, in which the te is equal to the pte. otherwise, the dmu could be at decreasing returns to scale if a proportional increase of all input levels produces a less than propositional increase in output levels, and vice versa for increasing return to scales. refer to reference [11] for more detail about the bcc model. there is a wealth of literature on both basic and applied research in dea. it has been widely applied to assess the relative efficiency of organizations, such as the banking industry [10], seaports [11], the high-tech industry [6], and non-profit organizations like a government [12]. it is always a difficult subject for the inefficient dmus to realize the factors causing the inefficiency, although it is obvious that either reducing inputs or increasing outputs will improve their performances [13]. this study employs the dea technique to measure technical efficiency (te), pure technical efficiency (pte), scale efficiency (se), and returns of scale (rts) for each semico nductor company. envelope calculations were carried out for each of the two years 2008 and 2009 using the “learning version” of dea-solver software [14]. 2.2.2 selection of variables (performance indicators) the selection of input and output variables for dea applications in the semiconductor-related industries can be traced to the literature [6, 15-20]. chen and chen [21] noted that th e output items should represent the result of the main operations objectives of organizations, and the input items should be the operation factors contributing to the output. four input variables are selected for this analysis, which are accessible from public data sources. they are considered the key indicators of managing a firm’s competitiveness including both business and technology innovative activities.  total assets includes total current assets such as inventories and liquid assets of cash, accounts receivable, marketable securities, and total fixed assets such as buildings and structures, machinery and equipment, and investment in securities and common stocks. in general, both idms and foundries own high level of total assets because of the nature of semiconductor industry which requires expensive fabrication facilities.  cost of goods sold (cogs), represents the costs of raw material, labours and related fringe benefits, overhead, d epreciation of manufacturing plant and production facilities, rent for land and buildings, and royalties for patent, designs, and other industrial rights. a cog is considered as a tangible value of a company and the difference between the companies’ efficiencies.  research and development (r&d) expenses, repres ents any expenses associated with the r&d of a company’s goods or services. the aim of r&d expenses is immediate invest and to bring the company success in the future. r&d expenses have the characteristic of technology innovation and operation efficiency perspective.  selling, general, and administrative (sg&a) expenses, includes advertising, sales commissions, distribution-related expenses, sales promotion and other selling expenses, o fficers ’ and directors ’ remuneration and bonuses, payroll, and fringe benefits. the main process perspective is that it helps the company to manage the internal and external part to develop competitive advantages. the output variable is net sales, representing sales revenues of finished goods and merchandise, and operating income from services rendered. net sales, considered a tangible value of a firm, are the gross increase in owner’s equity resulting from business activities. firms are usually seeking to increase their net sales for economies of scale and cost advantage. market leaders with a growing market share represent the firms’ reputation and clout in the industry 2.3. sample and data we select a sample of 26 leading semiconductor comp anies (20 idms and six foundry companies) with data over the two years 2008 and 2009. the semiconductor market experienced downward cycles in the duration this period. this is an interesting period to study because many semiconductor manufacturers operating in the period had to face industry -wide problems such as enormous r&d expenses for advanced technology development, increase of capital expenditures, new product transitions, design patent protection, unit price erosion, soft demand, production over-capacity, global supply chain problems, and other logistical issues [22]. the list of 26 semiconductor companies contains a broad representation of geographic regions. each of these companies is treated as a dmu. the memory ics market had become commoditized due to standardization and excess entry. in addition, intel corporation uses in-house manufacturing facilities to fabricate most of the microprocessor chips. thus, memory companies and intel haven’t been taken into consideration for the sample dmus. criteria to select the dmus are: (1) company’s semiconductor sales ranked within top 50 idm and top 20 foundry market segment in 2009 and (2) individual company’s net sales from the semiconductor segment surpassed 50 % of total net sales. the data of input and output variables are taken from their financial statements which can be accessed through company’s investor relations website. 3. empirical results and discussion 3.1. dea scores the results of the dea pure technical efficiency (pte), scale efficiency (se), and the nature of returns to scale (rts) of each company are shown in table 1. for the purpose of comparison, sample companies are divided into two main groups, idm and foundry groups, according to their business models. referring to the pte scores, 26 dmus could produce the same level of measured output with 7.79-9.67 % less inputs in 2008-2009 on average, holding the current input ratios co nstant. approximately 62-65 % of 26 dmus need to reduce their input if they are to become efficient. the rest of the companies are regarded as efficient. in addition, foundry companies have higher scores on average than idms in 2008-2009. the pte mean scores of idms range from 0.8957-0.9110 and foundry firms range from 0.9287-0.9592 in 2008-2009 (fig. 3). from the “mean” summary of pte scores, the foundries require 4.08-7.13 % surplus inputs on average in order to achieve the efficient output in 2008-2009 while idms requires 8.90-10.43 %, respectively. we consider that idm model presents relatively insufficient performance because resources are dispersed by taking care of both chip design and manufacturing. in addition, idms and foundry companies received higher se scores than pte scores in 2008-2009. the result reveals that the overall technical inefficiencies of all the proceedings of engineering and technology innovation, vol. 1, 2015, pp. 27 31 30 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti idm and foundry dmus are caused by inefficient operation rather than the scale inefficiencies. this also suggests that managers should focus firstly on removing the technical inefficiency of a company, and then the company can be subject to improving their scale efficiencies. the dominant effect of scale indicates that most idms have been operating at a non-optimal scale of operations. table 1 efficiency scores of the 26 semiconductor companies fig. 3 pte comparison between selected idms and foundry companies table 1 reports approximately 75 % of idm companies and 17-33 % of foundry companies in 2008-2009 (pte<1) need to reduce their inputs if they are to become efficient. the rest of companies are regarded as efficient (pte=1). the pte efficient dmu ratio of all foundry companies increased from 67 % in 2008 to 83 % in 2009. this might be due to the fact that the foundry focuses on ic manufacturing with diversified customer base. it is thus easier for the foundry to be run with a higher efficiency even during downturn. in addition, the gap of mean pte scores between idms and foundries widened in 2009 compared to 2008 (fig. 3) primarily due to the fierce competition caused by semiconductor downturn from 2008 to 2009. according to the se scores, three out of 20 idms in 2008, four out of 20 idms in 2009, four out of six foundries in 2008, and three out of six foundries in 2009 yield scale efficiency. the se scores of idms were worse performer than foundry cluster during 2008-2009. for the lower level scale efficiency (se) scores of sample dmus, it could be treated as support for future mergers and acquisitions between firms. to make manufacturing better, faster, and more economical, idms could consider unburdening their operations from cap ital intensive manufacturing to an asset-light idm business model, if those companies are gradually losing their economies of scale in the semiconductor battleground. 3.2. returns to scale to determine the current operating region for a scale inefficient company, following the result of zhu and shen [23], one can easily estimate the returns to scale (rts) by the te and pte scores. if the target company’s te is equal to pte, then constant return to scale (crs) prevails; otherwise, the target company indicates decreasing returns to scale (drs) if a proportional increase of all input levels produces a less than proportional increase in output levels and vice versa for increasing returns to scale (irs). table 1 indicates that 26 dmus operated with crs/drs/irs mixed patterns. companies that operated with drs are too large in scale for their production results. the company scale could be decreased to attempt scale efficiency, if drs prevail. companies that operated with irs are too small in dimension for their production results. moreover, companies, which have been operating at irs, could achieve significant cost savings and efficiency gains by increasing its scale of operations. further mergers and acquisitions among firms to increase the scale of operations could be considered in order to achieve optimal size, significant cost savings, and hence efficiency gains. the results also imply that, in order to reduce the surplus inputs for improving the operational efficiency, the relatively insufficient dmus with idm business model could leverage the resources of the efficient dmus in the foundry companies by adopting the asset-light, or in extremes, the fabless business model. 3.3. dea sensitivity analysis table 2 dea sensitivity analysis results in a sensitivity analysis we obtained a different efficient frontier comprising of all the efficient units according to the ccr model under the exclusion of only one of the input variables. the result reflects which key factor influences the dmus the most. table 2 reveals that total assets is the significant sensitive indicator for idm co mpanies from 2008 to 2009, followed by cogs, r&d e xpenses, and sg&a expenses. foundry group held sg&a e xpenses is the 2008 2009 2008 2009 2008 2009 1 analog devices, inc. (u.s.a.) 0.9626 0.9675 crs crs 0.9540 0.8485 2 advanced micro devices, inc. (u.s.a.) 0.7801 0.8576 crs crs 0.9762 0.9027 3 atmel corporation (u.s.a.) 0.9414 0.9459 drs crs 0.9979 0.9991 4 cypress semiconductor corporation (u.s.a.) 0.8702 0.9203 crs irs 0.9467 0.9364 5 fairchild semiconductor international, inc. (u.s.a.) 0.9381 0.9032 drs irs 0.9980 0.9965 6 freescale semiconductor, inc. (u.s.a.) 0.8621 0.8430 irs crs 0.9991 0.9886 7 infineon technologies ag (germany) 0.7595 0.9224 irs crs 0.9992 0.9730 8 linear technology corporation (u.s.a.) 1.0000 1.0000 crs crs 1.0000 1.0000 9 maxim integrated products, inc. (u.s.a.) 0.9180 0.8934 irs irs 0.9972 0.9993 10 microchip technology, inc. (u.s.a.) 1.0000 0.9638 irs irs 0.9781 0.9689 11 national semiconductor corporation (u.s.a.) 0.9740 0.8898 crs irs 0.9953 0.9969 12 nec electronics corporation (japan) 1.0000 1.0000 drs crs 0.9814 1.0000 13 nxp semiconductors (netherlands) 0.6212 0.6557 crs crs 0.9230 0.9925 14 on semiconductor corporation (u.s.a.) 0.9392 0.9443 drs crs 0.9974 0.9961 15 renesas electronics corporation (japan) 0.9373 1.0000 drs drs 0.9062 0.9159 16 rohm co., ltd. (japan) 0.7150 0.7915 irs crs 0.9986 0.9991 17 skyworks solutions, inc. (u.s.a.) 0.8603 0.8937 irs irs 0.9882 0.9800 18 stmicroelectronics n.v. (switzerland) 0.8347 0.8285 crs crs 0.9987 0.9893 19 texas instruments incorporated (u.s.a.) 1.0000 1.0000 crs crs 1.0000 1.0000 20 vishay intertechnology, inc. (u.s.a.) 1.0000 1.0000 crs crs 1.0000 1.0000 21 advanced semiconductor manufacturing co. (china) 1.0000 1.0000 crs irs 1.0000 0.9629 22 china resources microelectronics (china) 1.0000 1.0000 crs crs 1.0000 1.0000 23 taiwan semiconductor manufacturing co. (taiwan) 1.0000 1.0000 crs crs 1.0000 1.0000 24 united microelectronics corporation (taiwan) 0.8579 1.0000 crs drs 0.9961 0.9966 25 vanguard international semiconductor co. (taiwan) 1.0000 1.0000 crs crs 1.0000 1.0000 26 towerjazz semiconductor (israel) 0.7145 0.7553 irs irs 0.9500 0.9852 mean of 20 idms (from dmu 1 to dmu 20) 0.8957 0.9110 0.9818 0.9741 mean of 6 foundry companies (from dmu 21 to dmu 26) 0.9287 0.9592 0.9910 0.9908 mean of 26 dmus 0.9033 0.9221 0.9839 0.9780 dmu semiconductor companies pte (bcc) rts se 0.6 0.7 0.8 0.9 1 0 1 2 3 4 5 6 7 8 9 1 0 1 1 1 2 1 3 1 4 1 5 1 6 1 7 1 8 1 9 2 0 2 1 2 2 2 3 2 4 2 5 2 6 output/input (pte scores) 20 idm dmus 6 foundry dmus ■ idms (2008)  idms (2009) △ foundries (2008)  foundries (2009) 2009 2008 caused by business and technology innovation mean of 20 idms in 2008 and 2009 mean of six foundries in 2008 and 2009 2008 2009 2008 2009 original te 0.8801 0.8870 0.9222 0.9506 inputs excluding total assets te 0.7237 0.7433 0.8901 0.8867 change % -18% -16% -3% -7% sensitivity rank 1 1 2 2 inputs excluding cogs te 0.7590 0.7505 0.9049 0.9325 change % -14% -15% -2% -2% sensitivity rank 2 2 4 3 inputs excluding r&d expenses te 0.8381 0.8433 0.8933 0.9473 change % -5% -5% -3% 0% sensitivity rank 3 3 3 4 inputs excluding sg&a expenses te 0.8481 0.8693 0.8260 0.8328 change % -4% -2% -10% -12% sensitivity rank 4 4 1 1 20 idm companies six foundry companies proceedings of engineering and technology innovation, vol. 1, 2015, pp. 27 31 31 cop y right © taeti highest sensitive indicator from 2008 -2009, followed by total assets. foundry companies can get a competitive a dvantage by skilfully managing the sg&a e xpenses. both idm and foundry groups are capital intensive manufacturers. they fabricate products with a massive depreciation of the equipment. idm firms can significantly enhance market performance with minimu m efforts to leverage foundry’s capabilities of r&d and capacity. in today’s rapidly evolving world, companies need to constantly adjust their business models to changes in their environment . 4. conclusions this research provided an analysis framework of int egrating the resource constrains of business model and technology innovations from the firm’s financial perfo rmances evaluated by dea model. we properly incorporated four input variables; total asset, cogs, r&d e xpense, sg&a, and one output; net sales for dea analysis. we examined the managerial performance efficiency of 26 leading semiconductor companies over the time period of 2008-2009. the results of the dea analyses show consistency with industry observation. the findings can be briefly summarized as fo llows. first, the foundry companies with less resource loss are more efficient on average than idms, mainly because foundry companies focus on their core competence in chip manufacturing. second, the idm and foundry companies operated with crs/drs/irs mixed patterns. comp anies operated at drs (a proportional increase of all input levels produces a less than proportional increase in output levels) could consider reducing their scale dimension to improve competitive efficiency. ic device manufacturers with prevailing irs could increase their scale of operation for sustainable competitiveness. third, foundry group held sg&a expenses are the highest sensitive indicator from 2008-2009, followed by total assets. the idm group’s most significant sensitive indicator is total assets , followed by cogs. the asset-light business model can provide idms a better resource allocation and increase relative efficiency scores. acknowledgement the authors wish to thank the ke io university gcoe program for “sy mb iotic, safe and secure system design” and for financial support. references [1] wsts, http://www.ests.org/download/bbhist-26.xls, 2012. [2] r. kumar, fabless semiconductor implementation, mcgraw hill, new york, 2008. [3] gartner, http://my.gartner.com/, 2012. [4] a. charnes, w. w. cooper, and e. rhodes, “measuring the efficiency of decision making units,” european journal of operational research, vol. 2, no. 6, pp. 429-444, 1978. [5] l. m. seiford and j. zhu, manage,“profitability and marketability of the top 55 u.s. commercial banks ,” management science, vol. 45, no. 9, pp. 1270-1288, september, 1999. [6] f. h. liu and p. h. wang, “dea malmquist productivity measure: taiwanese semiconductor companies,” international journal of production economics , vol. 112, no. 1, pp. 367-379, 2008. [7] r. d. banker, a. charnes, and w. w. cooper, “some models for estimating technical and scale inefficiencies in data envelopment analysis ,” management science, vol. 30, no. 9, pp. 1078-1092, 1984. [8] r. färe, s. grosskopf, and c. a. k. lovell, production frontiers, cambridge university press, cambridge, 1994. [9] r. c. leachman, s. ding and c. f. chien, “economic efficiency analysis of wafer fabrication,” ieee transactions on automation science and engineering, vol. 4, no. 4, pp. 501-512, 2007. [10] m. asmild, j. c. paradi, d. n. reese, and f. tam, “measuring overall efficiency and effectiveness using dea” european journal of operational research, vol. 178, pp. 305-321, 2007. [11] k. cullinane, t. wang, d. song, and p. ji, “the technical efficiency of container ports: comparing data envelopment analysis and stochastic frontier analysis ,” transportation research part a: policy and practice, vol. 40, no. 4, pp. 354-374, may 2006. [12] p. hughes and m. edwards, “leviathan vs . lilliputian: a data envelopment analysis of government efficiency ,” journal of regional science, vol. 40, pp. 649-669, 2000. [13] c. kao and s. n. hwang, “efficiency decomposition in two-stage data envelopment analysis: an application to non-life insurance companies in taiwan,” european journal of operational research, vol. 185, pp. 418-429, 2008. [14] w. w. cooper, l. m. seiford, and k. tone, introduction to data envelopment analysis and its uses: with dea-solver software and references, springer, new york, 2006. [15] s. thore, f. phillips, t. w. ruefli, and p. yue, “dea and the management of the product cycle: the us computer industry,” computer and operations research, vol. 23, no. 4, pp. 341-356, 1996. [16] g. kozmetsky and p. yue, “comparative performance of global semiconductor companies,” international journal of management science, vol. 26, no. 2, pp. 153-175, 1998. [17] y. chen and a. i. ali, “dea malmquist productivity measure: new insights with an application to computer industry,” european journal of operational research, vol. 159, no. 1, pp. 239-249 (2004) [18] j. k. sengupta, “nonparametric efficiency analysis under uncertainty using data envelopment analysis ,” international journal of production economics, vol. 95, no. 1, pp. 39-49, 2005. [19] d. d. wu and c. b. ho, “productivity and efficiency analysis of taiwan’s integrated circuit industry,” international journal of business performance management, vol. 56, pp. 715-730, 2007. [20] y. s. chen and b. y. chen, “using data envelopment analysis (dea) to evaluate the operational performance of the wafer fabrication industry in taiwan,” journal of manufacturing technology management, vol. 20, pp. 475-488, 2009. [21] t. y. chen and l. h. chen, “dea performance evaluation based on bsc indicators incorporated: the case of semiconductor industry,” international journal of productivity and performance, vol. 56, no. 4, pp. 335-357, 2007. [22] y. c. tang and f. m. liou, “does firm performance reveal its own causes? the role of bayesian inference,” strategic management journal, vol. 31, no. 1, pp. 39-57, 2010. [23] j. zhu and z. h. shen, “a discussion of testing dmus’ returns to scale,” european journal of operational re-search, vol. 81, pp. 590-596, 1995. 5___peti#7712__45-52 proceedings of engineering and technology innovation, vol. 19, 2021, pp. 45-52 the effect of band-gap on tio2 thin film considering various parameters majedul haque mithun, abu sayed, imteaz rahaman * department of electrical and electronic engineering, rajshahi university of engineering and technology, rajshahi, bangladesh received 19 may 2021; received in revised form 27 june 2021; accepted 28 june 2021 doi: https://doi.org/10.46604/peti.2021.7712 abstract the aim of this work is to measure the effect of band-gap on tio2 thin films by changing tetrabutylorthotitanate (tbot), diethanolamine (dea), and temperature. the sol-gel method is experimentally introduced to find out the better band-gap of tio2 thin films by varying the concentration of tbot (4 ml to 10 ml), dea (2 ml to 5 ml), and temperature (350°c to 650°c). with the help of an ultraviolet-visible spectrophotometer for the wavelength of 300-900 nm, these thin films are characterized concerning optical properties (transmittance spectra, absorbance spectra, direct band-gap, and indirect band-gap). the direct and indirect band-gaps are found 3.38 ev and 3.25 ev respectively, which are close to or within the standard band-gap range of tio2 (3.2 ev to 3.35 ev) and are found at 8 ml tbot, 3 ml dea, and a temperature of 550°c. keywords: automatic sol-gel, optical properties, band-gap, tio2 1. introduction titanium dioxide (tio2) is widely used in photocatalytic applications as it provides chemical stability, non-toxicity, and low cost [1-2]. nowadays, tio2 is also considered an energy conversion material. it could be electrode materials for lithium-ion batteries [3] embedded in the membrane of polymer electrolyte fuel cells [4]. tio2 is one of the popular candidates, as its band-gap is quite wide, and therefore only the ultraviolet region of the light could be absorbed [5]. the sample preparation conditions, crystal phase, surface area, size distribution, and porosity [6-7] are mainly responsible for the photocatalytic properties of tio2. additionally, the photocatalytic movement of anatase tio2 is higher than rutile tio2 because of the fact that the band-gap energy of anatase (3.23 ev) is higher than the rutile (3.00 ev) [8]. however, some methods, namely chemical routes such as sol-gel dip-coating [9] and sol-gel spin-coating [10], consume less energy and do not require expensive equipment. many research groups have demonstrated that the nano-tio2 films prepared from the sol-gel method can make good photoanodes of dye-sensitized solar cells [11] and photo electrocatalytic hydrogen production devices [12]. the structural, electronic, and optical properties of iron (fe)-doped tio2 thin films by sol-gel technique are investigated, which shows that the increase of illumination intensity causes the increase of photocurrents [13]. the machine learning algorithm is an important criterion to learn types of atoms considering structural geometrical data of anatase tio2 nanoparticles [14]. a magnetron sputtering technique is used for coating sulfur electrodes into the tio2 thin film with the variation of deposition times [15]. the gaussian process regression model is developed for the predictions of anatase tio2 photocatalysts’ energy band gaps depending on the lattice parameters and surface area [16]. the tio2 thin-film preparation using sol-gel spin coating process is briefly investigated along with optical and material characteristics for future research scope [17]. a cd-si co-doped tio2 * corresponding author. e-mail address: imteaz.eee.ruet@gmail.com tel.: +8801632988128 proceedings of engineering and technology innovation, vol. 19, 2021, pp. 45-52 hierarchical coating on the surface of a glass slide is fabricated, which exhibits superhydrophobic properties. besides, the optical band-gap and surface chemical ligands of the prepared thin films are studied [18]. any change of charge distribution of the crystal unit cells leads to the variation of the single-oscillator parameters [19]. this study serves many purposes to optimize tio2 thin films. first of all, it fabricates tio2 thin films by sol-gel method. characterization of optical properties of tio2 thin films by ultraviolet-visible spectrophotometer is another focus of this research work. measuring the effect of band-gap on tio2 thin films by changing various parameters, i.e., tetrabutylorthotitanate (tbot), diethanolamine (dea), and temperature, is a major work to fulfill the research goal. finding out the better band-gap of tio2 thin films with the concentration of tbot, dea, and temperature is the ultimate focus of this study. 2. experimental details in this section, the sol-gel transition, sol-gel process, preparation of tio2 thin films, substrate cleaning, and preparation of precursor solution are briefly described. 2.1. sol-gel process in this compound method, the “sol” (or arrangement) is gradually developed towards the arrangement of a gel-like diphasic framework containing both the fluid stage and strong stage whose morphologies go from discrete particles to consistent polymer organizations. the arrangement of polymer network includes interfacing the metal habitats with oxo (m-om) or hydroxo (m-oh-m) spans. m − or + h2o → m − oh + roh (1) m − oh + ho − m → m − o − m + h2o (2) m − or + ho − m → m − o − m + roh (3) 2.2. preparation of tio2 thin film tio2 thin film can be prepared by using four steps: substrate cleaning, precursor solution preparation, substrate dipping and withdrawing, and heat treatment. the whole process is shown in fig. 1. cleaning the substrate preparation of the precursor solution substrate dipping and withdrawing heat treatment step-1 step-2 step-3 step-4 fig. 1 different steps of sol-gel process 2.3. preparation of precursor solution at first, a specific amount of tbot is dissolved into a specific amount of absolute ethanol to yield a specific concentrated solution, and then it is magnetically stirred for 2 hours. after stirring for 2 h at room temperature, a mixed solution of water and ethanol in the ratio of 1:10 is added dropwise. the resultant alkoxide solution is stirred at room temperature for hydrolysis reaction for 2 h. hence, the precursor solution is made through this process. then, the glass substrate is dipped in and pulled 46 proceedings of engineering and technology innovation, vol. 19, 2021, pp. 45-52 back from the solution to make thin layer of tio2 film on glass substrate. the dipping and withdrawal speed is 2 mm/sec. the substrates cover with the gel films dried at 600°c before calcining at 550°c for 1 h. the magnetic stirrer employed in this work is shown in fig. 2. fig. 2 magnetic stirrer used for stirring solution 3. result analysis the result is analyzed into three parameters. tio2 film is deposited by tbot variation, dea variation, and temperature variation, which is briefly shown in this section. a low-cost dye-sensitized solar cell is fabricated later successfully with the sol-gel derived tio2 thin film/photoelectrode. to do this chlorophyllin-sodium copper salt dye, carbon counter electrode and ki-based electrolyte are also used. this result shows that the photoelectron conversion efficiency of the tio2 thin film/photoelectrode, deposited with 1.0 g of polyethylene glycol (peg), is the highest among the samples of this investigation. 8 ml tbot and 3 ml dea are used to prepare the precursor solution [20]. that is why 8 ml tbot and 3 ml dea are used in this research. 3.1. deposition of tio2 film by tbot variation in this section, firstly, the transmittance and absorbance spectra of thin films for various concentrations of tbot keeping dea steady at 3 ml is prepared. fig. 3(a) shows the optical conveyance spectra of the prepared films. all the films have a sharp cut-off at around 380 nm wavelength and reach to the top at around 400 nm. the film which has tbot concentration of 8 ml is close to the standard curve. corresponding absorbance spectra is shown in fig. 3(b). according to tauc law reliance of assimilation, the co-productive α on photon vitality (hν) can be expressed in eq. (4). ( ) m h a h egα ν ν= − (4) where m is equivalent to 1/2 and 2 for immediate and aberrant changes separately, α is ingestion coefficient, a is edge width boundary, and hν is photon energy. the equation of α is expressed in eq. (5). 0 4 kπ α λ = (5) the transmitted light can be absorbed by the film material. in the fundamental absorption region, the transmission t is given in eq. (6) [21]. 0 4 exp( ) k t t a π λ − = (6) 47 proceedings of engineering and technology innovation, vol. 19, 2021, pp. 45-52 where k0 is the extinction coefficient. from eq. (6), t is film thickness and λ is the wavelength of incident light. if k0 << n, then the principle variation of t occurs in the exponential term, and the pre-exponential term a which accounts for reflecting effect is close to unity. in this regard, the equation of t is shown in eq. 7 [1]. exp( t)t α= − (7) then, from eq. (7), the absorption coefficient [22] can be expressed as: 100 2.303 log( ) t α = (8) hence, by knowing the value of transmittance t, the value of absorption coefficient α can be determined. again, eq. (9) describes the photon energy. 1240 (ev) hc e hν λ λ = = = (9) fig. 3(c) shows the curves of (αhυ) 2 versus photon energy hν for direct band-gap transitions of tio2 thin films for different concentrations of tbot keeping dea constant at 3 ml. it can be shown from fig. 3 that for 8 ml tbot the band-gap (3.28 ev) is within the range of the standard band-gap (3.20 3.35 ev). direct band-gap transitions of tio2 thin films for different concentration of tbot keeping dea constant at 3 ml is shown in fig. 3(c). it can be shown from the fig. 3 that for 8 ml and 10 ml tbot, it is close to the standard band gap range of 3.20 3.35 ev of tio2. for 4 ml tbot, it is within the standard range. however, for 6 ml tbot, it is very low from the standard mark. indirect band-gap transitions of tio2 thin films for different concentration of tbot keeping dea constant at 3 ml is shown in fig. 3(d). 300 400 500 600 700 800 900 0 50 100 wavelength (nm) t ra n sm it ta n ce ( % t ) 4 ml tbot 6 ml tbot 8 ml tbot 10 ml tbot 300 400 500 600 700 800 900 wavelength (nm) 0 20 40 60 80 100 120 140 160 4 ml tbot 6 ml tbot 8 ml tbot 10 ml tbot a b so rb a n c e (a) transmittance spectra (b) absorbance spectra 2.0 2.2 2.4 2.6 2.8 3.0 3.2 3.4 3.6 3.8 4.0 3.10 ev 3.38 ev 3.42 ev 3.55 ev 10 ml tbot 6 ml tbot 8 ml tbot 4 ml tbot photon energy (ev) ( α h υ )2 0.0 0.3 3.0 2.7 2.4 2.1 1.8 1.5 1.2 0.9 0.6 2.2 2.4 2.6 2.8 3.0 3.2 3.4 3.6 3.82.0 0.0 0.3 0.6 0.9 1.2 1.5 1.8 2.1 2.4 2.7 3.0 4.0 photon energy (ev) 3.00 ev 3.30 ev 3.11 ev 3.25 ev (α h υ )1 / 2 10 ml tbot 4 ml tbot 6 ml tbot 8 ml tbot (c) direct band-gap transition (d) indirect band-gap transition fig. 3 various parameters of tio2 thin films for different concentration of tbot keeping dea constant at 3 ml 48 proceedings of engineering and technology innovation, vol. 19, 2021, pp. 45-52 the variation of tbot keeping dea constant at 3 ml is shown in table 1. in this table, direct and indirect band-gaps are given. the band-gaps are measured with 4 ml, 6 ml, 8 ml, and 10 ml tbot. here, 8 ml tbot is highlighted for its better performance. table 1 tbot variation keeping dea constant at 3 ml variation direct band-gap (ev) indirect band-gap (ev) 4 ml tbot 3.58 3.28 6 ml tbot 3.44 2.70 8 ml tbot 3.28 3.18 10 ml tbot 3.49 3.11 from figs. 3(c)-(d) and table 1, it is clear that both direct and indirect band gap for 8 ml tbot is close to or within the standard range. therefore, for further optimization, the concentration of tbot is fixed at 8 ml in this study. 3.2. deposition of tio2 film by dea variation from the previous section, this work has the optimized value of tbot at 8 ml. now, keeping tbot at 8 ml, this section varies dea. tio2 films are prepared by dea variation by sol-gel process so that the surface of prepared films is grown uniformly. it can be clearly understood that the sol-gel derived tio2 films are visually transparent. first, the transmittance and absorbance spectra of thin films for different concentrations of dea keeping tbot constant at 8 ml is shown. fig. 4(a) shows the optical transmittance spectra of the prepared films. all the films have a very sharp cut-off at around 380 nm wavelength and reach to the peak at around 400 nm. the films are highly transparent in the visible range and have low transmittance at the ultraviolet region. it is noted that the average transparency of the sol-gel derived tio2 films is 81% in the visible range. the film which has dea concentration of 3 ml is closer to the standard curve. corresponding absorbance spectra are shown in fig. 4(b). direct band-gap transitions of tio2 thin films for different concentration of tbot keeping dea constant at 3 ml is shown in fig. 4(c). indirect band-gap transitions of tio2 thin films for different concentration of dea keeping tbot constant at 8 ml is shown in fig. 4(d). the variation of dea keeping tbot constant at 8 ml is shown in table 2. 300 400 500 600 700 800 900 wavelength (nm) 0 20 40 60 80 100 120 140 t ra n s m it ta n c e ( % t ) 2 ml dea 3 ml dea 4 ml dea 5 ml dea 300 400 500 600 700 800 900 wavelength (nm) -20 0 20 40 60 80 100 120 140 a b so rb a n c e 2 ml dea 3 ml dea 4 ml dea 5 ml dea (a) transmittance spectra (b) absorbance spectra 2 2.5 3 3.5 photon energy (ev) 0 0.5 1 1.5 2 2.5 3 3.5 4 ( α h υ )2 2 ml dea -3.21 ev 3 ml dea -3.30 ev 5 ml dea -3.35 ev 4 ml dea -3.30 ev (c) direct band-gap transition 2.2 2.4 2.6 2.8 3.0 3.2 3.4 3.6 3.82.0 0.0 0.3 0.6 0.9 1.2 1.5 1.8 2.1 2.4 2.7 3.0 4.0 photon energy (ev) (α h υ )1 / 2 ( e v 1 /2 m -1 /2 ) 2 ml dea -2.89 ev 3 ml dea -3.25 ev 4 ml dea -2.95 ev 5 ml dea -2.80 ev (d) indirect band-gap transition fig. 4 various parameters of tio2 thin films for different concentration of dea keeping tbot constant at 8 ml 49 proceedings of engineering and technology innovation, vol. 19, 2021, pp. 45-52 table 2 dea variation keeping tbot constant at 8 ml variation direct band-gap (ev) indirect band-gap (ev) 2 ml dea 3.21 2.89 3 ml dea 3.30 3.25 4 ml dea 3.30 2.95 5 ml dea 3.35 2.80 from figs. 4(c)-(d) and table 2, it is clear that both direct and indirect band gap for 3ml dea is within the standard range. therefore, for further optimization, the concentration of dea is fixed at 3 ml in this study. 3.3. deposition of tio2 film by temperature variation the transmittance and absorbance spectra of thin films for temperature variation keeping tbot and dea constant at 8 ml and 3 ml respectively are shown. fig. 5(a) shows the optical transmittance spectra of the prepared films. all the films have a very sharp cut-off at around 380 nm wavelength and reach to the peak at around 400 nm. the films are highly transparent in the visible range and have low transmittance at the ultraviolet region. it is noted that the average transparency of the sol-gel derived tio2 films is 80% in the visible range. the film which has a temperature of 550°c is close to the standard curve. corresponding absorbance spectra is shown in fig. 5(b). direct band-gap transitions of tio2 thin films for temperature variation keeping tbot and dea constant at 8 ml and 3 ml respectively are shown in fig. 5(c), presenting the curves of [h] 2 versus photon energy hν. it can be shown from fig. 5 that for 550°c it is within the standard band-gap range of 3.20 3.35 ev of tio2. for 650°c, it is close to the standard range. however, for 350°c it is very low from the standard mark. indirect band-gap transitions of tio2 thin films for temperature variation keeping tbot and dea constant at 8 ml and 3 ml respectively are shown in fig. 5(d). 300 400 500 600 700 800 900 wavelength (nm) 10 20 30 40 50 60 70 80 90 100 t ra n s m it ta n c e ( % t ) 350 degree (c) 550 degree (c) 450 degree (c) 650 degree (c) 300 400 500 600 700 800 900 wavelentgh (nm) -20 0 20 40 60 80 100 120 140 a b s o r b a n c e 350 degree (c) 550 degree (c) 450 degree (c) 650 degree (c) (a) transmittance spectra (b) absorbance spectra 2 2.5 3 3.5 4 photon energy (ev) -5 0 5 10 15 20 25 30 35 ( α h υ )2 350 degree (c) -3.68 ev 450 degree (c) -3.60 ev 550 degree (c) -3.38 ev 650 degree (c) -3.63 ev 2 2.5 3 3.5 4 photon energy (ev) 0 0.5 1 1.5 2 2.5 350 degree (c) -2.99 ev 450 degree (c) -2.88 ev 550 degree (c) -3.25 ev 650 degree (c) -3.19 ev (α h υ )1 / 2 (c) direct band-gap transition (d) indirect band-gap transition fig. 5 various parameters of tio2 thin films for temperature variation keeping tbot and dea constant at 8 ml and 3 ml respectively 50 proceedings of engineering and technology innovation, vol. 19, 2021, pp. 45-52 temperature variation is shown in table 3 keeping dea and tbot constant at 3 ml and 8 ml respectively. from figs. 5(c)-(d) and table 3, it is clear that both direct and indirect band gap for 550°c is close to or within the standard range. therefore, the optimum temperature is 550°c for a better band gap by keeping tbot and dea at 8 ml and 3 ml respectively. the direct and indirect band-gaps are found 3.38 ev and 3.25 ev respectively, which are close to or within the standard band-gap range of tio2 (3.2 ev 3.35 ev) and are found at 8 ml tbot, 3 ml dea, and a temperature of 550°c. finally, the whole result is shown in table 4. table 3 temperature variation keeping dea and tbot constant at 3 ml and 8 ml respectively variation direct band-gap (ev) indirect band-gap (ev) 350°c 3.68 2.99 450°c 3.60 2.88 550°c 3.38 3.25 650°c 3.63 3.19 table 4 summary of the whole experiment tbot variation keeping dea constant at 3 ml variation direct band-gap (ev) indirect band-gap (ev) 4 ml tbot 3.58 3.28 6 ml tbot 3.44 2.70 8 ml tbot 3.28 3.18 10 ml tbot 3.49 3.11 dea variation keeping tbot constant at 8 ml variation direct band-gap (ev) indirect band-gap (ev) 2 ml dea 3.21 2.89 3 ml dea 3.30 3.25 4 ml dea 3.30 2.95 5 ml dea 3.35 2.80 temperature variation keeping dea and tbot constant at 3 ml and 8 ml respectively variation direct band-gap (ev) indirect band-gap (ev) 350°c 3.68 2.99 450°c 3.60 2.88 550°c 3.38 3.25 650°c 3.63 3.19 4. conclusions this study mainly focuses on the measurement of the effect of band-gap on tio2 thin films by changing tbot, dea, and temperature. at 550°c, the band-gap is better than the rest of the combination (350°c, 450°c, and 650°c). therefore, the direct and indirect band-gap are found 3.38 ev and 3.25 ev respectively, which are close to or within the standard band-gap range of tio2 at 8 ml tbot, 3 ml dea, and a temperature of 550°c. conflicts of interest the authors declare no conflict of interest. references [1] a. velasco-hernández, r. a. esparza-muñoz, f. j. de moure-flores, j. santos-cruz, and s. a. mayén-hernández, “synthesis and characterization of graphene oxide-tio2 thin films by sol-gel for photocatalytic applications,” materials science in semiconductor processing, vol. 114, 105082, august 2020. [2] s. salaeh, m. kovacic, d. kosir, h. kusic, u. l. stangar, d. d. dionysiou, et al., “reuse of tio2-based catalyst for solar driven water treatment; thermal and chemical reactivation,” journal of photochemistry and photobiology a: chemistry, vol. 333, pp. 117-129, january 2017. 51 proceedings of engineering and technology innovation, vol. 19, 2021, pp. 45-52 [3] t. hasegawa, a. b. beleke, and m. mizuhata, “membrane modification by liquid phase deposition using small amount of tio2 for high temperature operation of polymer electrolyte fuel cells,” journal of power sources, vol. 233, pp. 148-156, july 2013. [4] d. dambournet, i. belharouak, and k. amine, “tailored preparation methods of tio2 anatase, rutile, brookite: mechanism of formation and electrochemical properties,” chemistry of materials, vol. 22, no. 3, pp. 1173-1179, february 2010. [5] y. liao, w. que, q. jia, y. he, j. zhang, and p. zhong, “controllable synthesis of brookite/anatase/rutile tio2 nanocomposites and single-crystalline rutile nanorods array,” journal of materials chemistry, vol. 22, no. 16, pp. 7937-7944, april 2012. [6] t. noguchi, a. fujishima, p. sawunyama, and k. hashimoto, “photocatalytic degradation of gaseous formaldehyde using tio2 film,” environmental science and technology, vol. 32, no. 23, pp. 3831-3833, december 1998. [7] d. h. kim, h. s. hong, s. j. kim, j. s. song, and k. s. lee, “photocatalytic behaviors and structural characterization of nanocrystalline fe-doped tio2 synthesized by mechanical alloying,” journal of alloys and compounds, vol. 375, no. 1-2, pp. 259-264, july 2004. [8] w. t. chiou, w. y. wu, and j. m. ting, “growth of single crystal zno nanowires using sputter deposition,” diamond and related materials, vol. 12, no. 10-11, pp. 1841-1844, october-november 2003. [9] w. g. lee, s. i. woo, j. c. kim, s. h. choi, and k. h. oh, “preparation and properties of amorphous tio2 thin films by plasma enhanced chemical vapor deposition,” thin solid films, vol. 237, no. 1-2, pp. 105-111, january 1994. [10] m. sreemany and s. sen, “influence of calcination ambient and film thickness on the optical and structural properties of sol-gel tio2 thin films,” materials research bulletin, vol. 42, no. 1, pp. 177-189, january 2007. [11] p. r. mishra, p. k. shukla, a. k. singh, and o. n. srivastava, “investigation and optimization of nanostructured tio2 photoelectrode in regard to hydrogen production through photoelectrochemical process,” international journal of hydrogen energy, vol. 28, no. 10, pp. 1089-1094, october 2003. [12] j. jiu, s. isoda, m. adachi, and f. wang, “preparation of tio2 nanocrystalline with 3-5 nm and application for dye-sensitized solar cell,” journal of photochemistry and photobiology a: chemistry, vol. 189, no. 2-3, pp. 314-321, june 2007. [13] f. kara, m. kurban, and b. coşkun, “evaluation of electronic transport and optical response of two-dimensional fe-doped tio2 thin films for photodetector applications,” optik, vol. 210, 164605, may 2020. [14] h. kurban, m. kurban, p. sharma, and m. m. dalkilic, “predicting atom types of anatase tio2 nanoparticles with machine learning,” key engineering materials, vol. 880, pp. 89-94, 2021. [15] t. zerrin, m. kurban, m. m. dickson, m. ozkan, and c. s. ozkan, “suppression of the shuttle effect in li-s batteries via magnetron sputtered tio2 thin film at the electrode-electrolyte interface,” acs applied energy materials, vol. 3, no. 2, pp. 1515-1529, february 2020. [16] y. zhang and x. xu, “machine learning band gaps of doped-tio2 photocatalysts from structural and morphological parameters,” acs omega, vol. 5, no. 25, pp. 15344-15352, june 2020. [17] a. deyasi, p. debnath, a. k. datta, and s. bhattacharyya, photonics, plasmonics, and information optics: research and technological advances, boca raton: crc press, 2021. [18] s. n. k. abad, n. n. ilkhechi, m. adel, and m. mozammel, “hierarchical architecture of a superhydrophobic cd-si co-doped tio2 thin film,” applied surface science, vol. 533, 147495, december 2020. [19] a. amini, m. s. zakerhamidi, and s. khorram, “changes of the tio2 thin film electro-optical properties in the ar and n2 glow discharge plasma due to effects of the electric field in plasma’s sheath,” physica b: condensed matter, vol. 612, 412820, july 2021. [20] m. f. hossain, s. biswas, t. takahashi, y. kubota, and a. fujishima, “effect of structure and surface morphology of sol-gel derived tio2 photoelectrode on the performance of dye-sensitized solar cells,” journal of vacuum science and technology a: vacuum, surfaces, and films, vol. 26, no. 4, pp. 1007-1011, july 2008. [21] g. liu, c. han, m. pelaez, d. zhu, s. liao, v. likodimos, et al., “enhanced visible light photocatalytic activity of cn-codoped tio2 films for the degradation of microcystin-lr,” journal of molecular catalysis a: chemical, vol. 372, pp. 58-65, june 2013. [22] a. chen, w. f. chen, t. majidi, b. pudadera, a. atanacio, m. manohar, et al., “mo-doped, cr-doped, and mo-cr codoped tio2 thin-film photocatalysts by comparative sol-gel spin coating and ion implantation,” international journal of hydrogen energy, vol. 46, no. 24, pp. 12961-12980, april 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/). 52  proceedings of engineering and technology innovation , vol. 4, 2016, pp. 22 24 22 the realization of healthcare combined with bluetooth and nfc technology jia-shing sheu 1,* , ho-nien shou 2 , fa-min luo 1 , guan-ci zeng 1 1 department of computer science, national taipei university of education, taipei, taiwan. 2 department of aviation & communication electronics, air force institute of technology, kaohsiung, taiwan. received 29 january 2016; received in revised form 23 february 2016; accept ed 06 march 2016 abstract the quality of life and function are enhanced with the superior technologies integrated and applied to meet the de mands of daily life , achieving lower social costs and ensuring the well-be ing of the social public. furthermore, to achieve the goals of enhancing the le isure industry and extending parenting education, so as to facilitate users’ physical and mental health, broaden their horizons toward quality life, enhance their cultural standing, and enable them to appropriately stretch outwards in a safe and secure setting, in this paper, the single -chip microcontroller with integrated control of radio frequency identification was combined with the global gps system, the usb pedometer, and the bluetooth wireless sensory technology and applied in tourist ro lls and the basic physical and mental health care system in travel. through the technological integration, the wire less sensor bluetooth device featuring node transfer function and near field co mmunicat ion (nfc) application sends informat ion of the tourist reaching the designated position and corresponding activity range data to the handheld platform of the main controlle r (leader) inquiring about the personal health in formation o f the user (tourist), thereby enhancing the security of group activities and even allowing the leader to grasp and offer prompt assistance to the tourist should any accident take place. ke ywor ds : global positioning system, bluetooth, near field co mmun ication, usb pedometer 1. introduction with the technological advancement in life and the progress in medica l technology, life e xtension is producing an impact on the future of taiwan’s lifestyle, while the operational stru cture is still relat ively undergoing change. taiwan’s society is currently facing serious proble ms associated with “aging” and “declining birthrates.” aging will lead to heavy social costs, resulting in a considerable burden on children and the society and in country’s resource-related burdens. hence, if innovative travel services can be integrated to benefit me mbers’ healthand security-related care, it will be good news to society. this being said, if wire less technical applications can be integrated to achieve this end, it will be considered as a practical innovation in the applications. although currently there is no e xperimental market to validate this point, through the realization of the idea, the elderly most difficult to care for will have access to secure and healthy activities . the same will a lso apply to the general public. technology can generate the second identification card with ease (i.e . the provision of personal e mergency protection related in formation), thus deriving industrial applications and generating new travel service models [1-4]. in view of the above considerations and the demand of the application side, the system platform device in this paper integrated the gps, bluetooth, usb pedo meter, and nfc functions into a four-in-one wireless travel passport device using an 8-byte single chip. as for the wireless detector(bluetooth), the protocol in line with the ieee802.15.4 standard was adopted, while the dual cassette body was adopted as the design to set up the radio frequency identification sy stem with 13.56mhz (iso/iec 15693 specifica tion) nfc reader, through which related sto red data can be read by way of the non -contact electronic tag [5-7]. the pedometer records the number of steps taken by the me mber and co nverts the calories consumed to provide the actual * corresponding aut hor. email: jiashing@tea.ntue.edu.t w proceedings of engineering and technology innovation , vol. 4, 2016, pp. 22 24 23 copyright © taeti fitness data, while the absolute position refers to the latitude and longitude data obtained from the gps use, which is then transmitted to the ma in controller’s (leader’s) platform, thereby achieving the relative re lationship between the designated position reached and the distance. at the same time, an e mergency button was designed to notify the ma in controlle r (leader) through interruption in case of an eme rgency [8-10]. 2. the system architecture fig. 1 system prototype machine (left ) and system function block diagram (right) the design ideas of the system are as shown in fig. 1. co mple mentary to the satellite pos itioning data, and co mbined with g roup management, personal information identification, and wire less sensor network transfer as the main functions, the relative positions and basic physical and mental state of the me mbers during travel within safe ranges are all monitored by the ma in controlle r to achieve the following functions: (1) after filling out the questionnaire on personal health information, the tour group me mbe r will log in to the management center and enter the informat ion in the rfid ca rd. the rfid card features the function of me mbe r identification and also serves as a reference for e me rgency rescue and medica l aid. (2) during the tour period, if there are acco mpanying nurses, me mbers e xperiencing discomfo rt can be treated immed iately. meanwhile, for me mbe rs requiring special heath care, the me mbers’ physiological health information can be collected and sent back to the manage ment center daily in the spare time during trave l to enable their re latives and fa mily me mbers to inquire an d understand the health condition of the members. 3. results in this paper, the man-mach ine control interface of the pda-side host adopts the receiving software developed by microsoft visual basic 2005, as shown in fig. 2. the interface program shows all the information received by the controlled-side and calculates the pos ition-related info rmation. meanwhile, it can instantly notify the main controller regarding the status between members. fig. 2 host of human interface control software 4. conclusions with the more rapid development of the various informat ion technologies, if the technologies can be readily integrated into the life functions where appropriate and be applied in cultural arts, the efficiency can be ma ximized, the quality of life enhanced, and ma ximu m re sults achieved at the lowest social costs, which would be great news for all. in this paper, a whole ne w technology application was imple mented: the use of the microprocessor as the man-machine technology, the integration of the various system control modules, and the s enfc module bluetooth gps micro-controller uart selector key emergency power supply pedometer i2c i/f uart i/f_a uart i/f_b gpio gpio usb proceedings of engineering and technology innovation , vol. 4, 2016, pp. 22 24 24 copyright © taeti quence of actions, as well as the functions of the informat ion processing and reading. this device was placed in the actual testing environment. when applied in tourist roll ca ll, its superior function was indeed displayed. it helped the leader save a great a mount of time, and showed good results in short-distance monitoring and warning fo r me mbe rs needing care. when a situation arose, the most appropriate and instant rescue and care was given within the shortest time possible. acknowledgement the support of ministry of science and technology, under grant most 104-2221e-344-002 is gratefully acknowledged. references [1] m. kaur, m. sandhu, n. mohan and p. s. sandhu, “rfid technology principles, advantages, limitations & its applications ,” international journal of co mputer and electrical engineering, vol. 3, no. 1, pp. 1793-8163, feb., 2011. [2] s. shepard, radio frequency identification, 1st ed. mcgraw-hill professional, 2004. [3] k. fin kenze ller, rfid handbook: ra dio-frequency identificat ion fundamentals and applications , 3rd ed. w iley john & sons inc., june 2010. [4] k. albrecht, rfid tag—you’re it, sc ientific american inc., sep. 2008. [5] r. x. xiao, w. r. su, and y, j. xu, “the nerve center of the operation o f rfid technology-rfid middlewa re,” foreseeing innovative new digiservices, sep. 2004. [6] j. a xe lson, usb co mp lete–the developer's guide, gotop information inc., jan. 2010. [7] g. o. oh, d. y. kim, s. i. kim, and s. y. rhew, “a quality evaluation technique of rfid middlewa re in ubiquitous computing,” proceedings of the international conference on hybrid information technology, vol. 2, pp. 730-735, nov. 2006. [8] y. c. liang, s. sun, x. peng, and f. ch in, “tutorial 2: e me rging wire less s tandards for wran, wifi, wimedia and zig bee,” the 10th ieee singapore international conference on co mmunicat ion systems, pp. 27-29, 2006. [9] perfecting just in t ime production, rfi d j o-urnal, http://www.rfid journal.co m/art icle/ articleprint/530/-1/4, november, 2003 [10] philips, mf1 ic s50 s pecification, philips electronics n.v., jan. 15th, 2007. 5___peti#8455__57-67 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 57-67 dynamic power management model for a wireless sensor node rakhee kallimani, sridhar iyer* department of electronics and communication engineering, sg balekundri institute of technology, karnataka, india received 10 september 2021; received in revised form 30 october 2021; accepted 31 october 2021 doi: https://doi.org/10.46604/peti.2021.8455 abstract dynamic power management (dpm) is an efficient technique to design low-power and energy-efficient nodes for wireless sensor networks. this article demonstrates the stochastic behaviour of an input event arrival which is modelled with first-in first-out (fifo) queue and a single server. an event-driven sensor node is developed based on semi-markov model. the article investigates the factors affecting the performance of the individual sensor node with detailed analysis considering power consumption and lifetime to be the performance metrics under study. the results demonstrate the impact of the change in event arrival and the probability of change detection on the performance of the node. it is observed that (i) the number of generated events increases with the change in the average value of the distribution which affects the service time in turn resulting in a variation of the server utilization, and that (ii) the increase in the detection probability increases the power consumption decreasing the lifetime of the node. keywords: dynamic power management, wireless sensor node, semi-markov model 1. introduction with recent progress in wireless communication and network technology, the conventional methods f information gathering, processing, and communication are no longer applicable to the present and the future user demands. furthermore, recent advance in embedded technology has motivated many researchers to analyze wireless sensor networks (wsns), the networks composed of individual nodes positioned at distinct locations. in wsns, each sensor node is composed of (i) a dedicated sensing unit to sense the variation in the environment, (ii) a processing unit to process the information, and (iii) a communication unit to communicate with the neighboring node or a sinking node. each unit in the sensor node is supplied with power for the node’s effective functioning, and the node can be portable or non-portable, based on the source of power and application. in specific, portable and battery-operated devices have attracted researchers’ attention recently. furthermore, wsns have been extensively used in various applications such as military, health care, biomedical, agricultural, etc. researchers and developers have shown keen interest in the field of embedded systems, especially in wsns. wsns embed sensor nodes; each individual node collects the information from the environment and then communicates to either a next node or gateway for further analysis [1]. these nodes are deployed at remote places where human intervention is rare, and hence it is expected for the nodes to be self-monitored, self-controlled, inexpensive, and power-aware [2]. the primary objective of the networks is to monitor the surrounding environment with the nodes powered by a battery. thus, performance and power are the main design metrics that have attracted the attention of researchers. the significance of power management is gaining the attention that (i) the nodes are defined to perform tasks with efficient performance, (ii) the replacement/recharging of batteries is a challenge, and (iii) the networks may fail prematurely due to the failure of an individual power-hungry node. in other words, there is a need for managing the power of each node individually and improve the network performance. in the state-of-art networks, the performance is dependent on the battery-powered nodes as these nodes are resource-constrained. * corresponding author. e-mail address: sridhariyer1983@gmail.com proceedings of engineering and technology innovation, vol. 20, 2022, pp. 57-67 in the embedded systems, there are numerous methods employed to manage the power at different levels of design. the current investigation is related to the power management at the system level such that the power consumption is reduced dynamically and there is no cost of hardware replacement. one such efficient management technique is dynamic power management (dpm). the technique tracks the system activity and the power manageable components, then dynamically switches the system from active state to passive state (low-power state), or switches off the components if required. this contributes to significantly improving the power and performance of the nodes. academicians and researchers have therefore conducted various stochastic techniques to design such nodes. to improve the performance of the entire networks, this study proposes a dpm based sensor node model. the rest of the article is structured as follows. section 2 reviews the related work. the methodology of the proposed dpm node is detailed in section 3. section 4 presents the assumed simulation parameters in modelling the node, and provides the discussion on the obtained results. lastly, the study is concluded in section 5. 2. related work in a wsn embedding several sensor nodes, each node has subunits for sensing, processing, and communication. these units are powered by a unit responsible for sourcing power. the power unit sources the power to each unit [3]. much analysis is carried out in managing the power of these energy-hungry nodes at the communication unit; however, the processing unit and power unit are neglected. popovici et al. [4] explored the power management methods at the system level. the work focused on ultra-low-power wake-up radio receivers. the work also presented the design of a wake-up radio receiver and demonstrated the low-power consumption at low cost. pughat et al. [5] conducted a detailed survey to discuss the different power management techniques at the operating system level along with their limitations. they also provided the scope for researchers to contribute the power saving techniques; their work motivated the present study towards stochastic control of power-saving. healy et al. [6] reviewed the available operating systems for sensor nodes. they listed four operating systems: tinyos, mantis os, sos, and t2; the supporting hardware was also presented. in their work, the need of managing the power at the operating system level was addressed. the component-based modelling is more flexible and easier for developers to experiment with power consumption for enhancing the battery lifetime. their work pointed out the critical issue of operating system in power management, and this motivated the present work. john et al. [7] presented a model for power consumption considering sensor nodes to be in different low-power modes, and simulated the wsn using opnet simulator. the major objective of their work was to determine the average power consumption and lifetime of the nodes. the researchers analyzed the energy consumption using opnet simulator of ad hoc nodes (ieee 802.11). chung et al. [8] emphasized the power management techniques considering the service request and controlling the policy in the design of modelling at the system-level power management. the time-out based, predictive, and stochastic techniques of dpm policy was discussed with the merits and the limitations. the dpm in the known stationary and unknown stationary environment was studied considering the nature of workloads. the authors introduced the adaptive management policies for non-stationary loads based on the sliding window concept. the power management for non-stationary workloads at multiple devices remains the limitation of the work. lee et al. [9] introduced a novel concept in dpm; the analysis of their proposed hybrid model was done by using rescaled range. the time-out method was used in the hybrid model to perform the comparison between the predicted value and threshold value of time, and a decision was taken based on the rescaled range analysis. to model the time series prediction problem, artificial neural network (ann) was used with multilayer perceptron (mlp) and back propagation (bp). their interest in work motivated the attention towards the analysis of workload characteristics of portable devices. the research on identifying the cluster head in wsn by the smart-in-a-smart-intelligent approach using evolutionary game theory was proposed by arivalai et al. [10]. 58 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 57-67 markov model is used to analyze the power and performance of wsn in the stochastic environment. many researchers have contributed to the improvement of wsn by performing the analysis for power consumption and lifetime. pughat et al. [11] developed a wsn model related to stochastic modelling, and compared its power efficiency with the conventional node. yamawaki et al. [12] used an adaptive suspending scheduling method to improve the battery lifetime, and mentioned different techniques to manage the power for wsn applications. pughat et al. [13] analyzed the trade-off between the power and performance of a wsn node. kallimani [14] surveyed the power management techniques at the system level, and emphasized the need for designing a power-managed embedded system. pughat et al. [15] worked on the stochastic behaviour of event arrivals to sensor nodes considering modelling the sensor nodes to sense the forest fire detection. the novelty of the work is the analysis of power consumption performed by implementing the control logic of switching the inactive components to low-power modes by fuzzy logic. their work motivated the present work in performing the analysis of power consumption. the authors also mentioned that their proposed approach can be applied to all energy constraint nodes. kallimani et al. [16] developed a stochastic sensor node using python, and demonstrated the performance analysis of the discrete event-based environment. also, they emphasized that the performance can be further improved using an analyzer unit. ever et al. [17] investigated the performance and availability requirement, and pointed out the impact of queue capacity and system scalability of the clustered internet of things (iot) systems. concerning the industrial iot networks, a novel power management scheme was proposed in the work of kiran et al. [18]. the authors formulated a mathematical model, and analyzed the network’s performance. their proposed model achieved 74.82% energy-saving using the 3d markov chain and m/g/1/k queue. ieee 802.15.4 mac was analyzed, and the effectiveness of the developed model was compared with simulation and real-time testbed deployed at the authors’ campus. the authors concluded that there is 97% accuracy with simulation and 94% accuracy with experimental results compared to the proposed model. lukman et al. [19] focused on power consumption, mobility, and end-to-end communication. the authors developed a sensing platform to sense environmental parameters such as humidity and temperature. they demonstrated the results on power consumption and end-to-end communication (wsn-iot). the work is at the mac layer of iot based stack for schedule-driven sensor nodes, using zigbee communication modules to establish end-to-end communication. the researchers emphasized the need of enhancing the lifetime in terms of the battery of individual nodes. jawad et al. [20] performed an extensive analysis study on power consumption on wireless agricultural system (was) using a sleep/wake scheme. the authors proposed an algorithm called sword to minimize the power consumption and also the data communication of the sensor nodes. the results of the proposed algorithm proved to be 86.54% more reliable than the sleep/wake scheme. the work majorly focuses on reducing the power consumption by transmitting only the data which is more than the threshold value; otherwise the communication unit will be in the sleep mode. cañete et al. [21] contributed detailed performance analysis and queuing systems for monitoring application. the analysis’s outcome shows the significance of reliability at the data link and the network layer. the reason for the congestion of the network is demonstrated to be the delayed occupation of the queue. phung-duc et al. [22] carried out a study on delay performance considering a multi-server queue system using markov chain. zhang et al. [23] used markov decision process to analyze the decision policy on energy consumption of the nodes by switching the nodes’ status. the proposed model is applicable for all the events that arrive at the sensor input. however, this is also the model’s limitation as the event arrived could be an undesired event and the event serviced without the analysis could lead to an increase in the power consumption. considering that there are queued events in the memory, the extensive work is conducted by de cuypere et al. [24]. their proposed model considers two queue systems representing the battery and the sensor data buffer, and the results of experimentation reveal that the data arrival has a limited impact on the performance measures. 59 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 57-67 nguyen et al. [25] identified the best routing protocol with minimum delay, and compared their analysis with simulation results. the delays of candidate routes from sensor nodes to sink nodes were evaluated by the team using queuing theory. sun et al. [26] investigated the m/g/1 queuing theory, and proposed a three-way handshaking in-band full-duplex medium access control (mac) mechanism. the obtained results show that the full duplex access point outperforms the half-duplex access point. furthermore, li et al. [27] investigated and proposed a displacement sensor for structural health monitoring, and detected the early corrosion based on fiber bragg sensor. the wsn in the industrial environment is an important application, and cao et al. [28] formulated an optimization problem considering the deployment problem. the authors continued the work of cao et al. [29] for wireless data center network, and formulated an optimization problem considering the coverage to be objective. there is also the contribution from researchers in reducing the power consumption of wsn for varied applications using different technologies; however, there is limited work concerning the contribution on lifetime analysis and extending the lifetime by dynamically switching the processor power into the low-power modes. the existing literature details the limitation of developing the model using a stochastic approach. the existing work in literature also uses either discrete or continuous markov model. however, the current scenario demands the consideration of a semi-markov model for wsn applications to control the power of a node. the aforementioned research motivates the need of managing the power at the processing unit simultaneously with the power unit. furthermore, if the power is optimally managed at the node level, it would enhance the lifetime of the network to promising levels. thus, in this article, to improve the performance in terms of the lifetime by optimally managing the power at the node level, the dpm technique is implemented [30]. to manage the power at the node level, the best possible approach is to extend the lifetime of a battery by employing energy harvesting capabilities, or to dynamically manage the power of the processing unit by using any of the dpm algorithms. the effect of wsn parameters affecting the lifetime of wireless sensor nodes is presented in the work of younus [31]. as a novelty in this study, the analysis of a node is considered in terms of performance parameters such as lifetime and power consumption by varying the probability of change detection and event arrival considering first-in first-out (fifo) queue discipline and single server using the dpm technique based on semi-markov model. 3. methodology this article demonstrates a simulation model of a wsn node considering the stochastic environment. the simulation of the wsn node is developed based on queuing theory, considering the event arrival to be stochastic in nature. the proposed semi-markov model is at the node level with an event generation subsystem, a power manager subsystem, and a single server demonstrating the use of dpm. the input is the event arrival generated by the event generation subsystem. the purpose of the power manager subsystem is to monitor the arrival of events and compute the power consumption and lifetime of the node. in the current article, a markovian model for an event-based application is proposed to meet the modelling requirement. the semi-markov model is defined by a set of states, and the transition time varies for each state. the considered node is consisting of states, namely s0, s1, s2, and s3, as pre-processing, analysing, main processing, and communication states, respectively. fig. 1 explains the transition of states. the transition from s0 to s1 is with the transition probability of 1, and the transition probability of s1 to s2 is denoted as �, i.e., the probability of change detection. the transition from s1 to s0 is (1-�). the transition probability of s2 to s3 is 1. here, s0 is the pre-processing state, which filters the event arrived. s1 is an analyser state, wherein the filtered event is analysed for the change in the value from the previous value. if there is a change detected, then it is termed as a desired event and moves to the next state for processing, otherwise it transits to the previous state. s2 depicts the main processor, which wakes up only upon the desired event, otherwise the main processor would be in its low-power state. s3 is the communication state, which communicates only when the desired events are processed by the main processor. 60 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 57-67 fig. 1 states of a sensor node two hypothesis statements are formulated. the first statement states that the power consumption can be managed dynamically by waking up the main processor only for the desired event passed by the analyser state. for all the undesired events, the main processor would be in its low-power state. the second hypothesis statement proposes to improve the lifetime of the node as multiple computations of the processor can be minimized for the same event. the assumptions for developing the simulation model are as follows: (1) poisson distribution is used with a mean of “5” for generation of event arrival “�”. (2) exponential distribution is used with a mean of “1” for the inter-arrival time distribution. (3) the probability of change detection “�” is varied in the range of 0 to 1. (4) the system has four power states and simulation parameters, as shown in table 1 and 2, respectively [8]. (5) the switching cost of transition from s1 to s2 and s0 is neglected. (6) the event-based random number generator is used to generate entities with poisson distribution. (7) the time-based entity generator is used with exponential distribution. (8) generated events are stored in the fifo queue. the model is developed using the following equations: 1 2 3 ( )= + + + + a v g a n p r tx p r e t p s a t p s t p s c c (1) ( )= + + a v g a n p r tx t t a t t (2) ( ) ( ) ( ) λ λ 0 avg avg p s + e p ow er c onsum ed p = 1 + t (3) ( ) ( ) λ λ 0 avg avg 1 + t b at_e ngbattery e nergy l ife tim e = = pow er c onsum ed ps + e (4) where the battery energy (���_�� ) is assumed to be of aa battery and is used to calculate the lifetime of the node with average energy consumption (� ��). the power (p) is consumed by the node and is calculated using ��, ��, ��, and �� power state modes. the time spent in analyser state is denoted as � �. ��� is the time spent in processing state, ��� is the time spent in transmission state, and � �� is the average time consumption. �� is the energy cost of processing and �� is the energy cost of communication unit [16]. the model is developed using matlab, simevents library package [32]. the basic abstract model of the system is shown in fig. 2, and the simulation of the proposed model is shown in fig. 3. the abstract model consists of a service provider, a service requestor, and a power manager unit. the service requestor is analogous to the generating events and is passed to the 61 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 57-67 memory, holding these events to be serviced by the service provider such as a processor. the main processor has power manageable components, wherein the power manager block observes the states of each block and accordingly issues a command to the processor and manages the power dynamically by switching the power manageable components into the power states. the dpm model as shown in fig. 3 is developed by integrating the event generation subsystem to a single server and a power manager subsystem. the events to be served by the server are departed from the fifo queue in the event generation subsystem. the power manager subsystem has an energy subsystem. the time subsystem is developed based on eqs. (1)-(2) and a unit to detect the change in the arrival. the utilization of the server is obtained from the output port of the single server. the power consumption is computed by the power manager subsystem built on the governing eq. (3). the subsystem also evaluates the lifetime of the node based on eq. (4). table 1 power state of the system [16] states s0 s1 s2 s3 �� on off off off �� on on off off �� on on on off �� on on on on table 2 simulation parameters [16] parameter value �� 2.3 mw �� 2.3 mw �� 237.5 mw �� 307.9 mw �� 2.206 mj ��� 2 s ��� 0.175 s � � 0.05 s ���_�� 1863 j �� 6.9 µj fig. 2 abstract model fig. 3 developed dpm model 62 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 57-67 the event generation subsystem shown in fig. 4 generates the events by poisson distribution with a mean of 5 using the random generator block. the inter-arrival time is provided with an exponential distribution of the mean of 1 using the time-based entity generator. the set attribute block sets the characteristics of the event. the schedule time-out block schedules the time-out attribute. the generated event is then passed through a fifo queue; fifo is set with 1000 events as capacity and is further passed to the single server for the service. the output of this subsystem is the number of arrivals. the power manager block is depicted in fig. 5, which has basic functionality to observe the workload of the service requestor (arrival) or/and service provider (server) and issues the command to dynamically switch the states. the power manager block is built using the time subsystem and energy subsystem. this subsystem also computes the performance parameters such as � (the probability of change detection) and lifetime in days. the energy subsystem is shown in fig. 5, and the time subsystem is built using the governing equations of the model. the functionality of the block is to compute the average energy and time using the simulation parameters fed as constant and α as varying in the range of 0 to 1. fig. 4 event generation model fig. 5 proposed model of power manager 4. simulation results and discussion the obtained simulation results for a single server queuing model are shown in fig. 6. the scope of simulink is used to obtain these graphs. it can be observed from the figure that the number of events increases with an increase in time, and the probability of change detection increases with the change in the events increasing over time thus simulating the real-time scenario. 63 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 57-67 furthermore, the power consumption increases with the variation in time. the major power consumption occurs in the communication state and the processing state. as the processor needs to compute the events before communicating to the transmission unit, it needs to be in the active state. this leads to more power consumption as the processor is active for all the events that occur, irrespective of the need to be computed or not. the lifetime of the node is seen to decrease with the variation of the probability of change detection, thus reflecting the deviation in the sensed value with the event arrived and suggesting that it needs to be computed before communicating to the next node. the first hypothesis statement stated is that the power consumption is reduced if the event is passed through the analyser state. in other words, this state will analyse if the event is desired or not, and if the event is analysed as an undesired event, then the main processor do not need to compute. otherwise, the event is desired and needs to be processed and communicated. thus, the consumed power of the node is minimum if the event is passed through the analyser state. furthermore, the consumption of power is observed to increase as the probability of change detection increases, indicating that the occurred event is the desired event which needs to be processed by the main processor and hence switching the processor to an active state. if the event is undesired or the same event occurs, then the main processor can be switched to a low-power sleep mode to save power dynamically. fig. 6 shows the event arrival from the event generation subsystem, the probability of change detection termed as α, the computed power consumption and lifetime of the node from the power manager subsystem, the utilization of the server, the energy consumption from the energy subsystem, the number of events departed from the queue, and the waiting time with the variation. the obtained results demonstrate the impact of the probability of change detection on the power consumption and the lifetime of the node. the probability of change detection is directly proportional to power consumption, wherein it is inversely proportional to the lifetime of the node. it can also be observed that the generated events are increasing. also, the utilization of server is observed to be initially at 100 %. as the number of events increases and the waiting time of the events increases, the utilization of the server is affected. (a) event arrival vs time (b) probability of change detection vs time (c) power consumption vs time (d) lifetime vs time (e) server utilization vs time (f) energy consumption vs time fig. 6 impact of change detection and event arrival on lifetime and power consumption 64 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 57-67 (g) number of events vs time (h) average time fig. 6 impact of change detection and event arrival on lifetime and power consumption (continued) table 3 analysis with the variation of mean arrival rate arrival rate (mean) change detection (α) power (mw) lifetime (days) no. of event arrival (�) utilization of server 1 0.984 236.98 78.61 64 64% 2 0.9868 237.26 78.51 76 75% 3 0.9878 237.37 78.48 82 81% 4 0.9884 237.44 78.46 86 85% 5 0.9888 237.40 78.47 84 83.34% 6 0.9888 237.5 78.46 87 86.35% 7 0.9888 237.4 78.46 86 86.35% 8 0.9888 237.5 78.45 88 85.35% 9 0.9888 237.5 78.44 90 89.36% 10 0.9888 237.5 78.44 90 89.36% table 4 analysis of the performance parameters affected by the change in event arrival and the probability of change detection event arrival (�) change detection (α) power consumption (mw) lifetime (days) 1 0 2.3 8100 2 0.5 162.9390534 114.3372298 3 0.666666667 193.6065091 96.22610359 4 0.75 206.6079223 90.17079205 5 0.8 213.7946332 87.13969908 6 0.833333333 218.3545791 85.31994191 7 0.857142857 221.505375 84.10631119 8 0.875 223.8128 83.23920705 9 0.888888889 225.5755013 82.58875583 10 0.9 226.9660071 82.08277634 11 0.909090909 228.0909442 81.67794677 14 0.928571429 230.4637264 80.83701625 20 0.95 233.0158707 79.95163566 30 0.966666667 234.9602226 79.2900168 35 0.971428571 235.509369 79.10513319 37 0.972972973 235.6868688 79.04555777 39 0.974358974 235.8459136 78.99225269 44 0.977272727 236.1795006 78.88068164 49 0.979591837 236.444267 78.79235235 62 0.983870968 236.9310868 78.63045854 66 0.984848485 237.0419839 78.59367229 79 0.987341772 237.3243196 78.50017239 80 0.9875 237.3422118 78.4942546 81 0.987654321 237.3596593 78.48848475 82 0.987804878 237.3766785 78.48285735 83 0.987951807 237.393285 78.4773672 84 0.988095238 237.4094936 78.47200934 85 0.988235294 237.4253184 78.46677904 65 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 57-67 table 3 assists in studying the effect of the variation of event arrival rate that follows the poisson distribution. the number of events arrived per hour can vary depending on the application. if the application demands a fast arrival of the event, then a faster arrival rate is needed. otherwise, for a certain application, where a change in the event occurs at a very slow rate, a slower arrival rate can be required. the effect of arrival rate on the lifetime and consumed power of the node is shown in table 3, wherein the arrival rates are varied from 1 to 10. it reflects that the number of events generated increases with the change in the mean of the distribution and affects the service time. thus, there is a variation in the utilization of the server. it can be observed that after the mean of 5, the probability of change detection is constant, indicating that there is not much deviation of values from the sensed and the stored value. table 4 also demonstrates the effect of the change in event arrival and the probability of change detection on performance parameters such as the lifetime and consumed power of the sensor node. the event arrival is varied in the range of 1 to 100, and α (the probability of change detection) is in the range of 0 to 1. the results show that the minimum value of the probability of change detection has a maximum lifetime as the probability of change detection indicates the difference between the sensed value and the necessity to be computed. this stresses the importance of the analyser state to detect the change if occurred and to process it before it communicates to the neighbouring state. the results also show that the second hypothesis is proved as true as the analyser state detects any changes in the event from the stored event and also reduces the need for multiple computing/processing. 5. conclusions to achieve dynamic power management, the present work proposes a wsn node model based on semi-markov model at the system level. compared to the previous work, this implementation is more suitable in the stochastic environment or application, and is more advantageous as it proposes an improvement in performance without any changes in the hardware design. the developed model is based on queuing theory as the considered event arrival is stochastic in nature. the model is analysed for the power consumption and lifetime based on the arrival of events and the probability of change detection. from the obtained results, it is observed that the lifetime of the node is affected by the arrival rate, and the power consumption is affected by the probability of change detection. in future work, the aim will be to analyse the proposed model considering other performance parameters which can affect the lifetime and power consumption of the node. the proposed work can hence be extended to the network level. conflicts of interest the authors declare no conflict of interest. references [1] m. s. obaidat and p. nicopolitidis, smart cities and homes: key enabling technologies, cambridge: morgan kaufmann, 2016. [2] o. mokrenko, “energy management of a wireless sensor network at application level,” ph.d. dissertation, département automatique, université toulouse iii paul sabatier, toulouse, 2015. [3] d. imededdin, a. salih, and h. medkour, “design and implementation of low power consumption wireless sensor node,” telkomnika (telecommunication, computing, electronics, and control), vol. 17, no. 6, pp. 2729-2734, december 2019. [4] e. popovici, m. magno, and s. marinkovic, “power management techniques for wireless sensor networks: a review,” 5th ieee international workshop on advances in sensors and interfaces, pp. 194-198, june 2013. [5] a. pughat and v. sharma, “a survey on dynamic power management approach in wireless sensor networks,” 6th ieee power india international conference, pp. 1-6, december 2014. [6] m. healy, t. newe, and e. lewis, “power management in operating systems for wireless sensor nodes,” ieee sensors applications symp., pp. 1-6, february 2007. [7] t. o. john, s. a. magaji, c. u. henry, and a. a. obiniyi, “dynamic power management in wireless sensor network,” innovative systems design and engineering, vol. 7, no. 4, pp. 99-113, 2016. 66 proceedings of engineering and technology innovation, vol. 20, 2022, pp. 57-67 [8] e. y. chung, l. benini, a. bogliolo, y. h. lu, and g. de micheli, “dynamic power management for nonstationary service requests,” ieee transactions on computers, vol. 51, no. 11, pp. 1345-1361, november 2002. [9] w. k. lee, s. lee, and w. siew, “hybrid model for dynamic power management, ” ieee transactions on consumer electronics, vol. 55, no. 2, pp. 656-664, may 2009. [10] c. arivalai and m. thenmozhi. “dynamic power management for improving sensor lifetime in internet of things based wireless sensor environments,” journal of computational and theoretical nanoscience, vol. 18, no. 3, pp. 913-921, march 2021. [11] a. pughat and v. sharma, “performance analysis of an improved dynamic power management model in wireless sensor node,” digital communications and networks, vol. 3, no. 1, pp. 19-29, february 2017. [12] a. yamawaki and s. serikawa, “battery life estimation of sensor node with zero standby power consumption,” ieee international conference on computational science and engineering, ieee international conference on embedded and ubiquitous computing, and 15th international symp. on distributed computing and applications for business engineering, pp. 166-172, august 2016. [13] a. pughat and v. sharma, energy-efficient wireless sensor networks, boca raton: crc press, 2017. [14] r. kallimani, “a survey of techniques for power management in embedded systems,” international journal of emerging technology in computer science and electronics, vol. 14, no. 2, pp. 461-464, april 2015. [15] a. pughat and v. sharma, “integrated fuzzy control to power management on event-based sensor node,” journal of information science and engineering, vol. 34, no. 4, pp. 835-849, 2018. [16] r. kallimani, k. pai, and k. rasane, “stochastic model of a sensor node,” international conference on computing in engineering and technology, pp. 242-250, january 2020. [17] e. ever, p. shah, l. mostarda, f. omondi, and o. gemikonakli, “on the performance, availability and energy consumption modelling of clustered iot systems,” computing, vol. 101, no. 12, pp. 1935-1970, may 2019. [18] m. s. kiran, v. subrahmanyam, and p. rajalakshmi, “novel power management scheme and effects of constrained on-node storage on performance of mac layer for industrial iot networks,” ieee transactions on industrial informatics, vol. 14, no. 5, pp. 2146-2158, may 2018. [19] a. a. lukman, j. agajo, k. j. gana, i. c. ogbole, and e. e. ataimo, “development of a low power consumption smart embedded wireless sensor network for the ubiquitous environmental monitoring using zigbee module,” atbu journal of science, technology, and education, vol. 5, no. 1, pp. 94-108, 2017. [20] h. m. jawad, r. nordin, s. k. gharghan, a. m. jawad, m. ismail, and m. j. abu-alshaeer, “power reduction with sleep/wake on redundant data (sword) in a wireless sensor network for energy-efficient precision agriculture,” sensors, vol. 18, no. 10, 3450, october 2018. [21] e. cañete, j. chen, m. díaz, b. rubio, and j. m. troya, “performance analysis of wireless sensor networks and priority queueing systems,” international journal of sensor networks, vol. 30, no. 2, pp. 126-139, may 2019. [22] t. phung-duc and k. i. kawanishi, “delay performance of data-center queue with setup policy and abandonment,” annals of operations research, vol. 293, no. 1, pp. 269-293, october 2020. [23] x. zhang, d. li, and y. zhang, “maximum throughput under admission control with unknown queue-length in wireless sensor networks,” ieee sensors journal, vol. 20, no. 19, pp. 11387-11399, october 2020. [24] e. de cuypere, k. de turck, and d. fiems, “a queuing model of an energy harvesting sensor node with data buffering,” telecommunication systems, vol. 67, no. 2, pp. 281-295, may 2018. [25] t. t. nguyen, w. w. lin, q. s. vo, and c. s. shieh, “delay aware routing based on queuing theory for wireless sensor networks,” data science and pattern recognition, vol. 5, no. 2, pp. 1-10, may 2021. [26] y. sun, h. zuo, y. chen, and b. wang, “queueing and channel access delay analysis in in-band full-duplex wireless networks,” international journal of distributed sensor networks, vol. 15, no. 4, pp. 1-15, april 2019. [27] c. li, l. sun, z. xu, x. wu, t. liang, and w. shi, “experimental investigation and error analysis of high precision fbg displacement sensor for structural health monitoring,” international journal of structural stability and dynamics, vol. 20, no. 6, 2040011, april 2020. [28] b. cao, j. zhao, p. yang, y. gu, k. muhammad, j. j. rodrigues, et al., “multiobjective 3-d topology optimization of next-generation wireless data center network,” ieee transactions on industrial informatics, vol. 16, no. 5, pp. 3597-3605, may 2020. [29] b. cao, j. zhao, y. gu, s. fan, and p. yang, “security-aware industrial wireless sensor network deployment optimization,” ieee transactions on industrial informatics, vol. 16, no. 8, pp. 5309-5316, august 2020. [30] m. škuta and d. macko, “automated integration of dynamic power management into fpga-based design,” ieee 22nd international symp. on design and diagnostics of electronic circuits and systems, pp. 1-4, april 2019. [31] m. u. younus, “analysis of the impact of different parameter settings on wireless sensor network lifetime,” international journal of advanced computer science and applications, vol. 9, no. 3, pp. 16-21, july 2018. [32] mathworks, “matlab & simulink,” https://uk.mathworks.com/products/simevents.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/). 67  proceedings of engineering and technology innovation , vol. 4, 2016, pp. 10 12 10 the innovative design of wheelchair with lifting and standing functions long-chang hsieh 1 , tzu-hsia chen 2,* , and shih-jie wei 3 1,2 department of power mechanical engineering, national formosa university, yunlin, taiwan . 3 machine tools technology center industrial technical research institute, taichung, taiwan . received 01 february 2016; received in revised form 27 february 2016; accept ed 06 march 2016 abstract traditionally, the mechanism of wheelchair with lifting and standing functions has 2 degrees of freedom, and used 2 power sources to perform these 2 motion function. the mechanis m p roposed for wheelchair with lifting and standing functions in this paper has 1 degree of freedom. hence, we can us e only 1 power source to drive the mechanism to achieve lifting and standing functions. for traditional standing wheelchair, its’ center of gravity moves forward when standing up and it needs 2 auxiliary wheels to prevent dumping. in this paper, by using the checklist method of osborn, the wheelchair with 1 dof is invented to perform lift ing and standing functions . the center of gravity of this new wheelchair a fter standing up still located between the front and rear wheels, no auxiliary wheels needed. the new design has the advantages of simple operation, more stability, and more safety. ke ywor ds : checklist method, innovative design, wheelchair with lift ing and standing functions 1. introduction wheelchair is an important assisted vehic le for the disabled person fo r body mov ing and partic ipat ing social a ctiv it ies. with the pro gress of sociality, the de mands of wheelcha ir are mo re and more dive rsified . hence , the standing up [1], lift ing up [ 2], ly ing down [3], tilting down [4], and mu lti-function [5] whee lchairs are invented to satisfy the d isabled person’s need. figs . 1(a )-1(d) show the whee lchairs with standing, lifting, ly ing, and tilting functions. th is paper focuses on the innovative design of wheelcha ir with lift ing and standing functions . currently, the mechanis m of wheelchair with lifting and standing functions has 2 degrees of freedom, and used 2 power sources to perform these 2 mot ion function. the 1 st purpose of this paper is to invent mechanis m with 1 degree of freedom (dof) to perfo rm lift ing and standing functions . hence, we can use only 1 powe r source to drive the mechanis m to achieve lifting and standing functions. according to fig. 1(a), for the traditional standing wheelchair, its’ center of gravity moves forwa rd when standing up and it needs 2 au xiliary wheels to prevent dumping. the 2 nd purpose of this paper is to improve the above problem. (a) standing (b) lifting (c) bed (d) tilting fig. 1 functional wheelchair by refe rring to such related patents [1-5], the studies on kinemat ics of mechanisms [6-8], and the studies on the creative design [8-10], we synthesize a mechanis m with 1 dog to perform the lifting and standing functions. the center of gravity of this new wheelchair a fter standing up and tilting down still located between the front and rear whee ls, no au xiliary wheels needed. hence, the new design has the advantages of simp le operation, more stability, and mo re safety. * corresponding aut hor. email: son.summer@msa.hinet.net proceedings of engineering and technology innovation , vol. 4, 2016, pp. 10 12 11 copyright © taeti 2. osborn’s check-list method the checklist is the simp lest method for problem solving. it is a method that provides hints by checking the ite ms on a prepared list against the problem. in design terms, a checklist is a set of questions that can be asked about any project in an attempt to stimulate ideas. osborn was called originator of modern creative science. he proposed many creative methods; one of them is check-list method. in his book “applied imagination”, he listed 83 questions that could be used to check for creativity. professor dior of harvard university summarized osborn’s 83 questions to 9 items. the common 9 ite ms are : 1. reverse, 2. transfer, 3. co mbine, 4. change, 5. extend, 6. en large, 7. reduce, 8. substitute, 9. rearrange. when we use of osborn check-list method to create new ideas and designs, we have to make sure that the creations could make advantages instead of just for invention. therefore, we must ma ke the innovation which can bring better performance than current status . 3. innovative design the 1 st purpose of this paper is to invent mechanism with 1 degree of freedom (dof) to perform lift ing and standing functions . according to fig. 1(a), for the traditional standing wheelchair, its’ center of gravity moves forward when standing up and it needs 2 auxiliary wheels to prevent dumping. the 2 nd purpose of this paper is to improve the above problems. in this paper, we use osborn’s check-list method to design the wheelchair having standing and tilting functions . 3.1. design requirements the require ments of the mechanis m for wheelchair with standing and tilting functions are: 1. the mechanism of wheelchair has 1 degree of freedom. 2. the lifting and standing functions can be achieved by using only 1 power source. 3. the center of gravity of this new wheelchair after standing up still located between the front and rear wheels, no auxiliary wheels needed. 3.2. chick -list method here, we adopt the transfer, co mb ine, and reduce methods of osborn check-list method to design the wheelchair with lift ing and standing functions. fig. 2 shows a mechanism with 1dog wh ich is the result of innovative design. this new mechanis m can perform the lifting and standing functions. fig. 2 ne w mechanis m for mu lti-function wheelchair the new mechanis m will have 9 links and 12 joints (inc luding 1 ca m pair j), i.e. j1=11 and j2=1, its’ degrees of freedom can be written as: f=3(n-1)-2j1-j2=3(9-1)-2(11)-1=1 (1) where f is the degree of freedoms, ji is the number of joints with i degree of freedom. 3.3. engineering design (a) low seated (b) high seated (c) standing p osition fig. 3 wheelchair with lifting and standing functions according to fig. 2, the corresponding 3d drawing of the wheelchair is completed by using the software “solid works .” figs . 3(a ), 3(b), and proceedings of engineering and technology innovation, vol. 4, 2016, pp. 10 12 12 copyright © taeti 3(c) show the corresponding diagrams of mu ltifunctional wheelcha ir at lower seat position, high seat position, and standing position. 4. conclusions traditionally, the mechanis m of wheelchair with lifting and standing functions has 2 degrees of freedom, and used 2 power sources to perform these 2 motion function. this invention “wheelchair with one degrees of freedo m to perform lifting and standing function s ” has two characteristics: 1.the mechanis m of wheelchair is a mechanism with 1 degree of freedom, 2. the wheelcha ir is driven by 1 power source to achieve lift ing and standing motion functions, 3. the center of gravity of this new wheelchair after standing up (or lift ing up) still located between the front and rear wheels, no auxilia ry wheels needed. the results of this paper can be used as a reference for engineering to design the functional wheelchair. acknowledgement the authors are grateful to the ministry of science and technology of the republic of china for the support of this research under most 104-2221-e-150-055. references [1] k. hunziker, stand-up unit for stand-up wheelchairs and chairs, particu larly therapy chairs, u.s. patent, 0207354 a1, 2010. [2] p. p. liu, l. h. cheng, w. t. hsieh, and k. w. zheng, the lift ing structure of seat for wheelchair, ta iwan patent, m 294,932, 2006. [3] y. c. yu, an electric wheelchair bed, taiwan patent, m492,731, 2015. [4] d. n. laslo, apparatus for tilting a wheelchair seat, u.s. patent, 5,322,741, 2012. [5] l. c. hsieh and j. b. huang, the electric wheelchair integrating s itting, standing, and tilt ing functions , taiwan patent, i495,461, 2015. [6] l. c. hsieh and t. h. chen, “optimal design of polishing trace for link-type optical fiber polisher,” journal of advanced materia ls research, vol. 328-330, pp. 900-905, 2011. [7] l. c. hsieh and t. h. chen, “the systematic design of link-type optical fiber polisher with s ingle flat,” journa l of advanced science letters, vol. 9, pp. 318-324, 2012. [8] a. f. osborn, applied imagination: principles and procedures of creative proble m solving, new yo rk: charles scribner's sons , 1953. [9] g. a ltshuller, and suddenly the inventor appeared – triz the theory of inventive problem solving, technical innovation center, inc., worcester, ma, 1996. [10] l. c. hsieh and h. c. tang, “the innovative design of automatic trans missions for electric motorcycles,” t ransactions of the canadian society for mechanical eng ineering, vol. 37, no. 3, pp. 741-753, 2013. http://twpat2.tipo.gov.tw/tipotwoc/tipotwkm?007ebfff0004010100000200001800a000000001000000000%5er%5e0%5ein:e7a88be694bfe7bea4204348454e2c204348454e47204348554e  proceedings of engineering and technology innovation , vol. 4, 2016, pp. 04 06 4 bending resistance capacity of steel pile caps strengthened with perfobond shear connectors dae-jin kim 1,* , young-ho kim 2 1 department of architectural engineering, kyung hee university, yongin, republic of korea. 2 institute of r&d, ji seung consultant co., ltc., seoul, republic of korea. received 31 march 2016; received in revised form 19 april 2016; accept ed 09 may 2016 abstract this study proposes a new perfobond shear connector, which can be used to strengthen the steel pile cap e mbedded into the structural foundation. the bending resistan ce capacity of the steel pile cap strengthened with the proposed perfobond connectors is evaluated by performing a test on seven specimens. test parameters include the type of perfobond shear connectors, infilled concrete depths and number of l-shaped steel plates. the lateral load versus displacement curve is plotted for all specimens, and their failure modes are identified. the effects of the test parameters on the bending resistance capacity of the specimen are e xa mined in this work. keywor ds : steel pipe pile, perfobond shear connector, concrete foundation, bending resistance 1. introduction at the foundation to support an upper structure, there e xists a connection between piles and concrete footing. the integration of these two structural co mponents is very important to guarantee the safe load transfer fro m the upper structure to the supporting layer of rocks. in general, this can be achieved by the use of steel reinforc ing bars e mbedded into the co ncrete footing. in this case, the bond between the steel rebars and concrete is activated by the mechanical bearing of bar ribs, and it is essential to retain a suffic ient level o f their bond strength. however, since their rib size is limited and they have relative ly s mall cross -sectional area, their bond strength is also limited to a small value as indicated in various concrete design code provisions such as aci building code [1, 2]. in order to handle this issue, we propose a new perfobond shear connector illustrated in fig. 1, wh ich can guarantee an excellent composite behavior among the structural components at the steel pile cap. th is type of perfobond shear connector can guarantee an e xce llent co mposite behavior among the structural components at the steel pile cap. this device is basically a steel plate with holes and embedded into concrete to ensure the perfect integration of steel me mbers and surrounding concrete. as external loads are applied, the dowel action of concrete inside the hole is activated and enables the load fro m the upper structure to be safely transferred to the steel pipe piles. (a) pe rfobond shear connector with closed holes (b) perfobond shear connector with open holes fig. 1 st rengthening of steel pile caps using the perfobond shear connectors this study experimentally investigates the structural performance of the steel pile cap strengthened with the proposed perfobond connectors under lateral loading. a bending test is conducted on seven specimens by considering the type of perfobond shear connectors, infilled concrete depth and number of l-shaped steel * corresponding aut hor. email: djkim@khu.ac.kr proceedings of engineering and technology innovation , vol. 4, 2016, pp. 04 06 5 copyright © taeti plates as test parameters. the failure characteristics of the test specimens are e xa mined, and their load-displacement curves are analyzed. 2. experimental program for this study, a total of seven specimens were manufactured and tested. among the seven specimens, six were strengthened with the newly proposed perfobond shear connectors, and the re main ing specimen with conventional steel reinforc ing bars. fig. 2 illustrates the details of typical test specimens with the perfobond connectors , including their geometrical configurations and infilled concrete depths. the pe rfobond shear connectors were connected to the outer surface of the steel pile cap by bolting. in addition, a number of l-shaped steel plates were connected to the inner surface of the steel pile cap by bolting. they play a role of shear keys and provide some additional shear resistance. the center of the steel p ile cap co incided with that of the concrete block in all specimens. fig. 2 details of specimens with perfobond connectors the list of the seven test specimens are provided in table 1. the main test parameters include the type of perfobond shear connectors, infilled concrete depths and number of l-shaped steel plates. two types of perfobond shear connectors were used in the test specimens. the first type (po) has open holes while the other (pc) closed holes . the reference specimen strengthened with conventional steel rebars (sb) had the infilled concrete depth of 700 mm, which is greater than the diameter of the steel pile. in contrast, the specimens strengthened with the perfobond shear connectors had two different infilled concrete depths corresponding to the diameter (d10) and half dia meter (d05) of the steel pile, respectively. three different numbers of l-shaped steel plates, which are 0 (nl), 3 (l3) and 6 (l6), were used in the specimens strengthened with the perfobond connectors . as shown in fig. 3, lateral load was applied to the test specimen by an actuator with ma ximu m capacity o f 500 kn, wh ich a llo ws displace ment-based control. the rate of loading was 0.5 mm/ min., and its magnitude was measured by the load cell attached at the end of the actuator. the load-versus -displacement data were recorded throughout the entire loading history using a computer-aided data acquisition system. fig. 3 test setup 3. results and discussion table 1 summarizes the test results such as peak loads and displacements at peak loads. from these results, it can be seen that the specimens strengthened with the open-hole type perfobond connectors have higher bending resistance than those with the closed-hole type connectors, and almost equal strength to that of the reference specimen. for e xa mp le, the ratios of the peak strength of the open-hole type perfobond specimens po_d05_l3 and po_d10_ l3 to that of the reference specimen are 93.3 % and 95.3 %, respectively. this is remarkable considering that the concrete infilled depths of these specimens are only in the range of 200 mm and 400 mm, while that of the reference specimen is 700 mm. 1 0 0 0 9 0 0 4 0 0 5 0 0 in fi ll e d c o n c re te d e p th 406 1400 lvdt 5 lvdt 4 lvdt 3 a-1 c-1 b-1 a-4 c-4 b-4 c-2 b-2 lvdt 1 lvdt 2 proceedings of engineering and technology innovation , vol. 4, 2016, pp. 04 06 6 copyright © taeti in addition, it can be noted from the test results that the peak strength of the specimen is increased with increasing concrete infilled depth. for example, in the case of pc_d10_l3 specimen, its peak strength is increased by 9.2 % due to the increase of concrete infilled depth from 200 to 400 mm, while that of po_d10_l3 specimen only by 2.2 %. therefore, the increase of peak strength due to increasing concrete infilled depth is more pronounced with closed-hole type perfobond specimens . table 1 summary of test results specimen p eak load displacement at peak load (mm) lat eral load (kn) bending moment (kn∙m) sb_d10_nl 342 308 50.0 p c_d05_nl 279 251 32.2 p c_d05_l3 262 236 33.2 p c_d10_l3 286 257 68.6 p c_d10_l6 303 273 48.0 p o_d05_l3 319 287 39.2 p o_d10_l3 326 293 43.6 4. conclusions in this study, we performed an e xpe rimental investigation on the structural performance of the steel pile cap strengthened with the proposed perfobond connectors under lateral loading. a bending test is conducted on seven specimens by considering the type of perfobond shear connectors, infilled concrete depth and number of l-shaped s teel plates as test parameters. fro m the test results, it was found that the specimens strengthened with the open-hole type perfobond connectors have higher bending resistance t han those with the closed-hole type connectors, and almost equal strength to that of the reference specimen. in addition, it can be noted fro m the test results that the peak strength of the specimen is increased with increasing concrete infilled depth. acknowledgement this work was supported by the basic science research program through the national research foundation of korea (nrf) funded by the ministry of science, ict & future planning (grant number: 2013r1a1a1076011). references [1] aci co mmittee 318. “building code require ments for structural concrete (aci 318-11) and commentary,” a me rican concrete institute. farmington hills: mi, 2011. [2] astm standard c39/c39m-11. “standard test method for co mpressive strength of cylindrical concrete specimens ,” ame rican society for testing and material, 2011.  proceedings of engineering and technology innovation, vol. 18, 2021, pp. 01-14 enhancing design features of asymmetric spur gears operating on a specified center distance using tooth sum altered gear geometry avil allwyn dsa 1,* , joseph gonsalvis 2 1 department of mechanical engineering, don bosco college of engineering, goa, india 2 department of mechanical engineering, rns institute of technology, bangalore, india received 28 january 2021; received in revised form 11 march 2021; accepted 12 march 2021 doi: https://doi.org/10.46604/peti.2021.7074 abstract asymmetric gears have evolved from the rising demand for power transmission drives with high load-carrying capacity, surface durability, and service life. direct design and s± profile shifted system are the most common approaches used for enhancing design features by geometry modification in asymmetric gears. this paper aims at establishing asymmetric gear geometry modification using tooth sum alteration for a family of gears running on a specified center distance as a feasible design approach. a complete mathematical treatment of the design approach is provided, and an in-house developed computer program is used for numerical simulation. the paper explores the influence of dynamic load factors, location factors for bending, specific sliding on load-bearing capacity, and surface durability on different tooth sum alterations. the study concludes that tooth sum altered asymmetric gear geometry can be employed as an effective design technique that offers designers flexibility in designing gears for surface wear, load-bearing, and tooth life. keywords: asymmetric gear, altered tooth sum, specific sliding, dynamic load, wear 1. introduction in recent years nonstandard asymmetric gear design has been researched for devising ways to enhance the performance of gear drives. the published research indicates attempts to improvise design features by varying the drive or coast side pressure angle for varying the tooth thickness and addendum height for contact ratio. these changes are brought about either by altering the cutter geometry or by profile shifting. the major limitation of these methods is the need to have custom-made tooling for the former and the need to trade off variation in the center distance for the latter. a method to alter the tooth geometry for enhanced design features without custom-made tooling or varying the center distance would indeed be novel. earlier, kapelevich [1] proposed a method for the design of gears with asymmetric teeth. several equations required for the design of asymmetric gears are developed for the synthesis of asymmetric gears. litvin, lian, and kapelevich [2] proposed gear geometry modification of asymmetric gears for the reduction in transmission error by using a combination of involute gears and double crowned pinion with a larger drive side pressure angle. karpat, ekwaro-osire, cavdar, and babalik [3] studied the dynamic characteristics of asymmetric spur gears by addendum modified gear geometry for the reduction in transmission error and dynamic factor. karpat and ekwaro-osire [4] proposed tooth profile tip relief modification as an effective measure to reduce dynamic loads and wear. sekar [5] made a comparative study of load sharing, fillet, and contact stresses in symmetric and asymmetric gear drives. sekar and muthuveerappan [6] estimated the tooth form factor and stress * corresponding author. e-mail address: avil.dsa@dbcegoa.ac.in tel.: +91832-2777344; mobile: +91-9657043438 proceedings of engineering and technology innovation, vol. 18, 2021, pp. 01-14 2 correction factor for asymmetric gears using standard iso-b methodology. gonsalvis and rayudu [7] reported the sensitivity of the contact ratio to tooth-sum alteration in symmetric gears, its influence on operating pressure angle, and profile shift coefficient. avil and gonsalvis [8] developed a methodology to evaluate dynamic load distribution in tooth-sum altered symmetric gears considering the altering root fillet geometry. gonsalvis and sachidananda [9] explored the possibility of altering the tooth-sum of symmetric gear-pair, to tailor the gear drive either for low, normal, or high contact ratio. while literature related to tooth sum altered symmetric gear geometry is scarce, however, the published literature reveals no reference on the benefits and problems of accommodating tooth sum altered profile shifted asymmetric gears running on a specified center distance. the present work attempts to explore the possibility of modifying the tooth geometry of an asymmetric spur gear-pair for design benefits by altering the tooth sum while holding the operating center distance constant, which also can be named as tooth-sum altered asymmetric gear (tsaag). the objective of this work is to establish mathematical expressions for the tsaag geometry, check for kinematic compatibility and design superiority of tsaag over normal contact ratio (ncr) asymmetric counterparts. a case study on altering the tooth sum of an asymmetric spur gear of a given module to obtain possible tsaag combinations is modeled as a single degree of freedom spring-mass-damper system (smd) [10-11] subjected to forced vibration, for evaluating the dynamic load factors, load-bearing ability and specific sliding characteristics with ncr asymmetric counterpart. 2. the geometry of tsaag fig. 1 tooth thickness distribution factor an asymmetric rack cutter of module m, with standard proportions, is shown for reference in fig. 1. the drive and the coast side pressure angles are denoted by 𝜙𝑝 𝑑 and 𝜙𝑝 𝑐 respectively. for convenience, the authors define drive side tooth thickness 𝑡𝑝 𝑑 , as a factor  of total tooth thickness 𝑡𝑝, along the pitch line. from the geometry of the rack we get d p pt t (1) 1 c p d p tan tan      (2) the general form of expression for tooth thickness 𝑡𝑥 [1] at any arbitrary location for profile shifted asymmetric spur gear of module m with total profile shift coefficient 𝑋𝑠 [12] and shift factor  is given by p x x x p t t r r           (3) pitch line t d p t c p t p m 1m 1.25m  d p  c p 0.25m proceedings of engineering and technology innovation, vol. 18, 2021, pp. 01-14 3  tan tan 2 d cc p s p p p t x m     (4)     d c x p x p xinv inv inv inv              (5) 2.1. total profile shift in tsaag system the need for profile shifting in tsaag arises due to tooth sum alteration and the constraint of constant center distance. the geometrical consequences of profile shifting in a tsaag system is either an increase or decrease in the tooth thickness, restoring physical contact between the teeth, and also an increase or decrease in the radius of the addendum circle maintaining the overall tooth height. for continual conjugate action, a calculated total profile shift coefficient 𝑋𝑠 is divided among the meshing gears. mathematically total profile shift coefficient 𝑋𝑠, for a tsaag system with design backlash b, operating on a working pressure angle 𝜙𝑤 𝑑 and radius 𝑟𝑤 with tooth thickness 𝑡𝑤 on the driver and driven is given as 1 1 2 1 2 w w w r t t b z     (6) applying eq. (3) for tooth thickness at working pitch circle, eq. (6) can be written as 1 2 1 1 2 1 2 1 2p p w w w w w p p t t r r r b r r z                       (7) for any profile shifted system the ratio of tooth numbers is equal to the ratio of the working pitch circle radii 2 2 1 1 a w w r z r z        (8) substituting eq. (8) in eq.(7), replacing (𝑧1 𝑎 +𝑧2 𝑎 ) = 𝑍𝑠 𝑎 and dividing throughout by (2𝑟𝑤1), eq. (7) can we rearranged as 1 1 2 1 1 2 2 a p p w s w bz t t m z r            (9) if the profile shift coefficient of the pinion is 𝑥1 = (𝜅𝑋𝑠), then the shift coefficient of the gear is 𝑥2 = (1 − 𝜅𝑋𝑠). using eq. (4), eq. (9) is written as   1 1 1 tan tan 2 2 d c a s p p w s w bz m x m z r                  (10) rearranging eq. (10), the general form of the expression for total profile shift coefficient can be expressed as   2 2 tan tan a w s s d c p p m z b x m           (11) the value of the total shift coefficient is constant for a tsaag gear pair as long as the sum of the drive side and coast side pressure angles is unchanged. the coordinates of the drive and coast-side involute for profile shifted asymmetric gear are given by proceedings of engineering and technology innovation, vol. 18, 2021, pp. 01-14 4 cos , sinx x x xx r y r   (12)    2 tan tan 2 d c d d x s p p p xx inv inv z                  (13) 2.2. pressure angles in tsaag system an asymmetric gear pair with reference tooth-sum 𝑍𝑠 𝑟 and drive side pressure angle 𝜙𝑝 𝑑𝑟, operating on a center distance 𝑐𝑠 𝑟 can be replaced by a tooth-sum altered 𝑍𝑠 𝑎 asymmetric gear-pair, operating at working pressure angle 𝜙𝑤 𝑑𝑎, keeping the center distance unchanged 𝑐𝑤 𝑑 = 𝑐𝑠 𝑟 (tsaag system). if the tooth sum is altered by a factor α, then the relation between the number of teeth and radius of the working pitch circle is given by 2 cos cos d w w d p r z m     (15) using the eq. (15), eq. (14) can be reframed as    1 2 1 2 cos cos a d a rw w wd p r r r r               (16) using the relation (𝑟𝑤1 + 𝑟𝑤2) = 𝑐𝑤 , (𝑟1 + 𝑟2) = 𝑐𝑠 and defining 𝑐𝑤 𝑎 = 𝛽𝑐𝑠 𝑟 , eq. (16) can be rearranged to express the operating pressure angles on the drive and the coast side. 1 1 cos cos cos cos d d w p c c w p                             (17) reframing eq. (16) to relate the working center distance of tsaag with its tooth sum we get cos 2cos a d r pa a w sd w m c z                (18) the standard center distance of the tsaag is related to its tooth sum by the relation 2 a a a s s m c z        (19) subtracting eq. (19) from eq. (18), assuming that the tsaag module remains unchanged   1 2 r a a w s s m c c z                  (20) 2.3. topping for tsaag system the tsaag is a profile shifted system on fixed center distance. for a given profile shift, the ratio of addendum radius gain to that of tooth thickness gain along the pressure line is always constant and is approximately equal to 3 (refer to fig. 2). this significant increase in addendum radius for attaining a nominal increase in tooth thickness along the pressure line may a r s sz z (14) proceedings of engineering and technology innovation, vol. 18, 2021, pp. 01-14 5 r 1 r w1 r w2 r 2 a w1 a w2 b w1 b w2 b 1 b 2 cause inappropriate root clearance. similarly, a small amount of negative shift results in a considerable reduction in the magnitude of the addendum circle radius. this issue can be overcome by topping the addendum with a calculated positive or negative value. to ensure appropriate root clearance, addendum radii are topped with tooth topping coefficient y. a tsaag generated by a rack cutter with an addendum of 1.25 module and dedendum of 1 module in mesh is shown in fig. 3. the radial root clearance 𝑟𝑐𝑙 is expressed as      1 2 1 2 1 22.25cl w wr b b m r r r r       (21) the radial distance between the root circle and the pitch circle is given by     1 2 1.25 1.25 1 s s b x m b x m             (22) using eq. (22), eq. (21) can be reframed as    0.25 a cl w s s m r c c x m     (23) using eq. (20) in eq. (23), and defining (0.25𝑚 − 𝑟𝑐𝑙 ) = 𝑌𝑚, an expression for tooth topping coefficient is given by 1 2 a s s z y x           (24) the root clearance for any gear pair is established to be at least 0.25m. tsaag system which is topped should satisfy the expression 0.25clr ym m  (25) tsaag system works on the concept of altering the tooth sum of the reference gear pair. the parameters α and β enables designers to opt for tooth sum altered tsaag system or center distance altered s± asymmetric profile shifted system. the eqs. (1)-(25) are the most general set of expressions for tsaag profile shifted system, s± profile shifted system, and standard spur fig. 2 profile shift fig. 3 topping for root clearance 0.87 2.62 1.74 0.6 0.58 0.89 proceedings of engineering and technology innovation, vol. 18, 2021, pp. 01-14 6  x t o o th c e n te r li n e (x c , y c ) d (x c , y c ) c (x d , y d ) r x f x gear system (sts). for a standard backlash-free gear-pair, α = β =1. for the tsaag system, 0.96< α <1.04 and β =1. for the s± profile shifted system, 0.96< β <1.04 and α =1. tsaag accommodates sums of various curvatures by keeping the center distance constant and changing operating pressure angles, whereas the s± asymmetric profile shifted system accommodates sums of various curvatures by keeping the sum of the base circle radii constant and changing operating pressure angles (refer to fig. 4). any two arbitrary gear pairs operate at the same working pressure angles as long as the ratio (α/β) is maintained constant. the magnitude of the total profile shift coefficient remains constant for any asymmetric gear pair as long as the sum of the drive side and coast side pressure angles are unchanged, referring to eq. (11). 2.4 contact ratio and interference avoidance criteria for tsaag it is possible to accommodate gear pairs with different modules and tooth sums on a specified center distance. operating higher module gears with lower tooth sums increases power transmission capability but reduces contact ratio and restricts speed range. gear ratios achievable with any tooth-sum should satisfy the interference avoidance criterion [13] which also establishes the limiting value for the module. the necessary and sufficient mathematical conditions are given as maxra f lr r    (26) and  1 2 ra rl z cr      (27) 3. safe load transmission capability of tsaag system fig. 5 load bearing in a tsaag system evaluation of power transmission capability based on material’s bending strength depends upon the tooth geometry. a tooth load intensity 𝑓𝑥 per unit face width along the pressure line at any arbitrary value of roll angle can be resolved into components along and perpendicular to the tooth centerline. the tangential component of the tooth force along the pressure fig. 4 difference between tsaag and s approach  a cos( d w ) a b c 1 ( d w ) a c abrb1+rb2 bc 1 + 2  a b c c 1 b 1 cos( d w ) a ( d w ) a (a) center distance altered gears (b) tooth sum altered gears (a) tooth subjected to a load (b) a gear pair in mesh proceedings of engineering and technology innovation, vol. 18, 2021, pp. 01-14 7 line causes tensile stresses on one side of the root fillet and compressive stress on the other side. the effect of the radial component of the tooth force is to cause compressive stress on either side of the tooth. the net effective fillet stress at the root is compressive on one side and tensile on the other. these cyclic tensile stresses are responsible for the initiation and propagation of fatigue cracks. the expression for safe load-bearing f [13] is modified to find the maximum static load which could be safely transmitted by a gear of a given module and material to avoid bending failure in any kind of spur gear system (refer to fig. 5). t dl f f m k x   (28)  md c x t f x y y tan k m k cos z a            (29) x sl eq x    (30) where 𝑘𝑡 is called the geometric location factor for bending, a and z are the cross-sectional area and the sectional modulus of the critical section. the cartesian coordinates of the critical section on the drive side and the coast side are (𝑥𝑐 𝑑 , 𝑦𝑐 𝑑 ) and (𝑥𝑐 𝑐 , 𝑦𝑐 𝑐 ) respectively. dynamic load factor 𝑥𝑑𝑙 is the ratio of location-based mesh stiffness 𝜆𝑥 to total mesh stiffness 𝜆𝑒𝑞 . 4. surface wear in tsaag system gear surface wear is an on-going process that results in loss of material along the tooth profile with an increasing number of load cycles. the profile deviation increases as wear progresses, the effects of which are manifested in the form of an increase in noise, vibration levels, and backlash. these deviations generate overloads on the teeth for which the gearing may not have been sufficiently dimensioned when designed, which could cause the onset of the other modes of failure in the spur gear tooth. surface wear is evaluated by the volume of material lost, and the degree of wear is described by wear rate, specific wear rate, or wear coefficient. archards wear equation [14] for sliding contacts is a mechanical wear model popular for its simplicity and its ability to characterize wear under a wide variety of conditions. flodin and anderson [15] modified the generalized archard’s equation to describe dry and boundary lubricated sliding surfaces of a gear pair in mesh. 1 2 , , 1 0 1 2 r rx n x n w fc r v v h h k a v     (31) observing term within the parenthesis [𝑣𝑟1 − 𝑣𝑟2 ∕ 𝑣𝑟1] = 𝑉𝑠𝑠 called as specific sliding, and 𝜎𝑓𝑐 𝛼0 = 2𝑓𝑑𝑥 𝜋⁄ [12], eq. (31) can be modified as , , 1 , , 1 4 4 * dx x n x n w ss x n x n dl ss w f h h k v f h h x v k             (32) where f is the static load transmitted per unit face width, ℎ𝑥𝑛 is the accumulated wear at any location after n cycles, 𝑓𝑑𝑥 is the dynamic load per unit face width, 𝑥𝑑𝑙 is the dynamic load factor at the contact location, 𝛼0 is the semi-contact width of the elliptical contact, and 𝑘𝑤 is the wear coefficient. the total profile error 𝜀𝑥 , dynamic load 𝑓𝑑𝑥 , and load factor 𝑥𝑑𝑙 at any location after n cycles is given by proceedings of engineering and technology innovation, vol. 18, 2021, pp. 01-14 8 1 2 xn xn xnh h   (33)  dx x r xf x   (34) dx dl f x f  (35) 5. implementation scheme and methodology for the case study, a gear pair of module 2 with a drive side pressure angle 20˚ and tooth thickness distribution factor τ = 0.438 is adopted for a center distance of 108mm. operating conditions created by four different tooth-sum alteration factors 0.96< α<1.04 and profile shift factors with β =1 are considered. the material properties, operating conditions, and gear geometry alterations described by eqs. (11), (17), and (24), are tabulated in table 1. table 1 tsaag geometry and material properties gr=1:1 α material properties and operating conditions α 0.96296 0.98148 1 1.01852 1.03703* density of steel 7700 kg/m 3 𝑍𝑠 𝑎 104 106 108 110 112 damping factor (𝜉) 0.17 𝜙𝑤 𝑎 0.4397 0.3968 0.349 0.294 0.2264 young’s modulus (𝛦) 206gpa β 0.144 0.144 0.144 0.144 0.144 fillet strength (𝜎𝑓 ) 413mpa for 10 6 cycles 𝑋𝑠 2.163 0.97 -0.1 -1.0273 -1.7885 wear coefficient (𝑘𝑤 ) 5×10 -16 m 2 n -1 y 0.1627 -0.0302 -0.1 -0.0272 0.2111 speed 1000 rpm 𝑡𝑝 4.715 3.847 3.069 2.392 1.839 *case of interference the product of dynamic load factor and location-based geometric factor determines the safe load-bearing ability of a gear, referring to eq. (28). the product of dynamic load factor and specific sliding is the deciding factor for accumulated wear depths over n cycles, referring to eq. (32). hence the determination of dynamic load factors is critical in the analysis of any gearing system. to evaluate dynamic load factors for tsaag, a gear pair in mesh is modeled as two rigid disks connected by a spring-damper set along the pressure line (refer to fig. 6). the single degree of freedom (dof) dynamic model considers the influence of mesh stiffness, damping forces 𝐹𝑐 , friction forces, static transmission error at the mesh interface, and are expressed as a time-varying function. the gear pairs are assumed to be of unit face width and free of tooth profile errors. the differential equations of motion (eom) for a single dof spring mass damper system can be expressed as eq r eq r eq rm x c x k x f   (36) 1 2 1 2 e e eq e e m m m m m   (37) fig. 6 dynamic model for spur gear i g j 1 i 1 j g j 2 i 2  1  2 r b1 r b2   1 , ω 1 {g}{1}{2}{3} i 4 j 4 {g} ω 2 t 2 j 3 i 3 proceedings of engineering and technology innovation, vol. 18, 2021, pp. 01-14 9    2 1 1 2 1 1 1 2 1 1 n n e i i e i i i i eq e e m m k m m              (38) 2eq eq eqc m k (39) where 𝑚𝑒𝑞, 𝑘𝑒𝑞, 𝑐𝑒𝑞 are mass, stiffness, and damping coefficient of the equivalent s-m-d system, 𝑚𝑒 is the effective mass, and 𝑥𝑟 is the relative displacement at the mesh interface. fem analysis is adopted to create gear tooth stiffness 𝑘𝑠 , for tsaag system as a function of its roll angle. the coefficients for the polynomials for various α values are tabulated (refer to fig. 7 and table 2). table 2 equation for stiffness of tsaag 𝑘𝑠(𝜃) = 𝑝1𝜃 3 + 𝑝2𝜃 2 + 𝑝3𝜃 + 𝑝4  p1 p2 p3 p4 0.96296 -7304.8 9601.2 -4348.9 725.24 0.98148 -3595.8 4013.4 -1631.5 284.15 1 -1780.7 1580.5 -597.23 131.89 1.01852 -937.43 536.73 -204.14 74.484 1.03703 -1049.9 427.48 -133.25 54.627 the equation of motion is numerically solved using the fourth-order runge-kutta linear iterative procedure. one mesh period is divided into a certain number of intervals and the solution procedure is started with a trial value of initial conditions. the values of the relative displacement and velocity after one mesh period are compared with the trail values. the solution is said to converge when the difference between the initial condition and the obtained values after one mesh cycle are within the preset tolerance. the value of relative displacement is used for finding dynamic load factors, referring to eq. (35). the safe static load transmission capability is evaluated using the ratio of location-based mesh stiffness to total mesh stiffness. a flow chart of the in-house developed code is provided in fig. 8. (a) α = 0.96296, κ = 0.5, β = 1 (b) α = 0.98148, κ = 0.5, β = 1 (c) α = 1, κ = 0.5, β = 1 (d) α = 1.01852, κ = 0.5, β = 1 (e) α = 1.03703, κ = 0.5, β = 1 (f) mesh stiffness of tsaag fig. 7 fem analysis for the stiffness of tsaag proceedings of engineering and technology innovation, vol. 18, 2021, pp. 01-14 10 fig. 8 flow chart for evaluation of (𝑥𝑑𝑙 × 𝑣𝑠𝑠 ) & (𝑥𝑑𝑙 × 𝑘𝑡 ) 6. results and discussion fig. 9 contact ratio vs. profile shift factor fig. 10 safe load vs. tooth sum alteration factor (α) contact ratio is a geometric parameter of much importance which is an indicator of the influence of gear geometry on its performance. since it is a pure geometric parameter, it is always a good starting point to begin an investigation that involves gear geometry modification. the investigation of the contact ratio provides information and direction in which the analysis needs to proceed. tsaag is a tooth sum altered, profile shifted, addendum topped system, and all of these modifications affect the contact ratio. a tsaag carrying positive profile shift exhibits a reduction in the contact ratio, owing to reduced tooth-sum, increased working pressure angle, and reduced radius of the addendum circle due to tooth topping. limiting the study to gears with a contact ratio above 1.3 (refer to fig. 9), it is observed that while the tooth-sum reduction by 4 is permissible with contact ratio falling to 1.36, a tooth-sum increase by 2 is permissible, without causing interference, raising the contact ratio above 2. it compute solving eom compute re-compute cumulaitive cumulative wear depth at each contact point n=n+1 end input compute cycles n=1, wear depths material properties start proceedings of engineering and technology innovation, vol. 18, 2021, pp. 01-14 11 is important to note that +4 tooth alteration without any design backlash results in undercutting of gears (refer to table 1) and would cause a certain part of the involute profile not to participate in load-bearing. this part of the involute profile would show an increase in the contact ratio mathematically but has no practical significance. hence positive tooth alterations are advised only up to a profile shift value at which the first sign of undercutting appears. similarly, the negative number of tooth alterations should be limited to a profile shift value at which the tooth becomes peaked. attempting higher negative alterations would result in an unreasonable reduction in contact ratio. introduction of design backlash reclaims profile shift to a certain degree, salvaging certain part of the addendum which would otherwise be topped. hence one can expect negative number of tooth alterations without backlash to lower the contact ratio, but introduction of backlash shows an improvement of the same. the preliminary investigation of tsaag based on contact ratio reveals the possibility of achieving high contact ratio (hcr) tsaag with positive values of tooth alterations (α > 1) and low contact ratio (lcr) tsaag, with negative values of tooth alterations (α < 1) and all gear pairs running on the same center distance. experimental results by kasuba [16] and kahraman-blankenship [17] have established that the dynamic loads decrease with increasing contact ratio in spur gearing. even though tsaag with hcr > 2 opens up the possibility of transmitting more power at low decibels without resorting to higher strength materials, it is important to interpret this result in the light of safe load transmission capability. the plot of safe load-bearing capability f vs α (refer to fig. 10) shows an increase in load-bearing capability on either side of unity but the interpretation varies. the increase in load-bearing ability for lcr tsaag with α < 1 is attributed to the increased tooth thickness and pressure angles. however, for hcr tsaag, the increase in load-bearing ability comes due to the contact ratio rising above 2 as seen in the case of α = 1.01852. further increasing the value of α, reduces load-bearing ability due to the reduction in tooth thickness. the product of dynamic load factor and location factor for bending 𝑥𝑑𝑙 × 𝑘𝑡, referring to figs. 11 (a)-(f), is the sole dimensionless parameter that decides the safe load-bearing ability of any gear. normal contact ratio standard tooth sum (ncr sts) asymmetric gears and lcr tsaag show one region of single tooth contact and two regions of double tooth contact. hcr tsaag shows three regions of triple teeth contact and two regions of double tooth contact, indicating an improvement in the load-bearing capacity. the maximum value of 𝑥𝑑𝑙 × 𝑘𝑡 in either of the gears decides the safe load transmission capability of any gear drive. dynamic load factor is speed-dependent and location factor for bending is geometry dependent. (a) α = 0.96296, κ = 0.5, β = 1 (b) α = 0.98148, κ = 0.5, β = 1 (c) α = 1, κ = 0.5, β = 1 (d) α = 1.01852, κ = 0.5, β = 1 (e) α = 1.3703, κ = 0.5, β = 1 (f) location factors 𝑘𝑡 for bending fig. 11 𝑥𝑑𝑙 × 𝑘𝑡 for 0.96296 ≤ α ≤ 1.3703 proceedings of engineering and technology innovation, vol. 18, 2021, pp. 01-14 12 the single tooth contact region is the largest for α = 0.96296 and starts reducing as alpha approaches unity. the dynamic load dominates the magnitude of 𝑥𝑑𝑙 × 𝑘𝑡 which is evident by observing the slope of the plots for α < 1. at any given speed, dynamic load in a gear system decreases with an increasing contact ratio. this is because of the narrow single or triple contact zone, which passes quickly, leaving no time for the system to respond. hence for hcr tsaag the value of 𝑘𝑡 dominates the magnitude of 𝑥𝑑𝑙 × 𝑘𝑡 . the plot of tsaag with α = 1.01852 yields the lowest of the maximum magnitude of 𝑥𝑑𝑙 × 𝑘𝑡 among tsaag, which gives a clear impression of having the largest load-bearing ability. also, hcr tsaag is expected to have lower decibel levels due to reduced dynamic loads. while it looks very encouraging to know that it is possible to have improved load transmission capability with hcr tsaag, it is also important to evaluate its surface durability. the product of dynamic load factor and specific sliding 𝑥𝑑𝑙 × 𝑉𝑠𝑠 is the sole dimensionless parameter that decides the wear accumulation, refering to eq. (32). the gear pairs are subjected to a load value to sustain 106 bending cycles. at κ=0.5, tsaag as well as ncr asymmetric gears, irrespective of the value of α, attain equal approach-recess action. the plots of 𝑥𝑑𝑙 × 𝑉𝑠𝑠 for tsaag operating under κ=0.5 with different α values, accumulated over 105 cycles are shown in figs. 12(a)-(f). for tsaag with α < 1 the slope of the curve is steeper in the region of single tooth contact owing to comparatively larger load and it reduces in the two teeth contact region. the region on either side of the pitch point covering a certain roll angle indicates a larger region of single tooth contact or lower contact ratio. the gears experience larger dynamic loads during the transition from single tooth to double tooth contact at lower speeds. the specific sliding in fig. 12(f) shows significant improvement for α < 1 in the addendum and dedendum regions of both pinion and the gear. the improvement in specific sliding offsets the higher dynamic loads, keeping the accumulation of 𝑥𝑑𝑙 × 𝑉𝑠𝑠 minimal up to 105 cycles. a comparison with asymmetric ncr gear indicates higher service life of tsaag lcr with higher decibel levels at lower speeds due to reduced contact ratio. (a) α = 0.96296, κ = 0.5, β = 1 (b) α = 0.98148, κ = 0.5, β = 1 (c) α = 1, κ = 0.5, β = 1 (d) α = 1.01852, κ = 0.5, β = 1 (e) α = 1.03703, κ = 0.5, β = 1 (f) specific sliding 𝑉𝑠𝑠 fig. 12 𝑥𝑑𝑙 × 𝑉𝑠𝑠 for 0.96296 ≤ α ≤ 1.3703 (continued) the accumulation of 𝑥𝑑𝑙 × 𝑉𝑠𝑠 for tsaag hcr as the value of α approaches 1.03703 is significant over 10 5 cycles at the region near the root and addendum circles, where the slope of specific sliding changes drastically. profile error introduced due to wear is accumulative over cycles of usage. the dynamic loads are lower as a result of a narrow region of three teeth contact proceedings of engineering and technology innovation, vol. 18, 2021, pp. 01-14 13 which passes quickly. the dynamic load factor increases with the accumulation of profile error. specific sliding considered constant for every cycle acts as an amplifier to the iterating dynamic load factors. hence the accumulation of 𝑥𝑑𝑙 × 𝑉𝑠𝑠 becomes large especially at locations having higher specific sliding. while the dynamic load in sts asymmetric gears and tsaag lcr with α < 1 does not grow as much owing to lower wear rates, tsaag hcr gears with α > 1, start with lower dynamic loads and build up to higher-value during the course of its life. these gears are expected to run with decibel levels lower than the sts asymmetric gears for an initial period, but the vibrations and noise levels increase over their service life. 7. conclusions altering the tooth sum of a reference ncr asymmetric gear pair offers a family of tsaag gears for the same center distance with improved performance characteristics. tooth sum is a macroscopic parameter that alters pressure angle and tooth thickness offering the flexibility of accommodating various tooth sums on the same center distance. increased tooth-sum coupled with reduced pressure angle and tooth topping results in hcr tsaag and decreasing tooth sum results in lcr tsaag. both lcr and hcr tsaag have improved load-bearing ability and mesh stiffness. the effects of tooth-sum alteration on surface wear are influenced by meshing parameters such as dynamic load factors, specific sliding, and material properties. wear rate increases for tsaag, as α moves away from unity either towards lesser or greater values within the permissible limits. the specific sliding and dynamic load factors play a significant role in determining the overall dynamic load, surface wear build-up, and service life. the rate of change of slope of the specific sliding curve is a dominant factor influencing the wear rate and tooth life. hcr tsaag offers up to 50% more load-bearing ability than ncr asymmetric gears, but higher wear rates increase dynamic loads which may reduce service life. lcr tsaag has up to 13% more load-bearing ability with better surface durability but suffers loss in contact ratio which may cause noisy operation. in conclusion, tooth sum modified asymmetric gear geometry (tsaag) is an effective design technique to introduce geometrical changes for design benefits while maintaining the center distance constant as they offer design benefits that are not achievable by ncr asymmetric gear design. nomenclature α tooth-sum alteration factor β design backlash ß center distance alteration factor 𝑐 center distance �̈�𝑟 relative acceleration at mesh interface 𝑐𝑒𝑞 equivalent damping coefficient �̇�𝑟 relative velocity at mesh interface 𝐸 modulus of elasticity κ profile shift factor 𝐹 total static load per unit face width λ mesh stiffness per unit face width 𝐹𝜆 spring force μ coefficient of friction 𝐹𝑐 damping force ϕ pressure angle 𝑓𝑑𝑥 location-based dynamic load per unit face width ρ radius of curvature ℎ accumulated wear 𝜎𝑓 allowable bending fatigue strength 𝑘𝑒𝑞 equivalent mesh stiffness 𝜃𝑟𝑎 roll angle of addendum radius 𝑘𝑡 tooth geometric factor for bending 𝜃𝑟𝑙 roll angle of radius of limiting circle 𝑘𝑤 wear coefficient ξ damping factor 𝑚 module ε tooth profile error 𝑟𝑏 radius of base circle 𝜎𝑓𝑐 contact pressure 𝑥 profile shift coefficient 𝑟𝑙 radius of limit circle 𝑥𝑠𝑙 static load factor 𝑟𝑓 radius of the fillet circle 𝑥𝑑𝑙 dynamic load factor 𝑟𝑥 radius of any point on the profile 𝑥𝑟 relative displacement at mesh interface 𝑇 total static torque per unit face width 𝑌 tooth topping coefficient 𝑣𝑟 rolling velocity 𝑧 number of teeth 𝑉𝑠𝑠 specific sliding 𝑍𝑠 tooth-sum of gear pair 𝑋𝑠 total profile shift coefficient proceedings of engineering and technology innovation, vol. 18, 2021, pp. 01-14 14 subscripts superscripts 1, 2 driver and driven gears a altered 𝑠 standard r reference 𝑤 working d drive side 𝑥 any location along the pressure line c coast side 𝑝 pitch circle abbreviations tsaag tooth sum altered asymmetric gear ncr normal contact ratio hcr high contact ratio rpm revolutions per minute lcr low contact ratio sts standard tooth-sum conflicts of interest the authors declare no conflict of interest. references [1] a. kapelevich, “geometry and design of involute spur gears with asymmetric teeth,” mechanism and machine theory, vol. 35, no. 1, pp. 117-130, january 2000. [2] f. l. litvin, q. lian, and a. l. kapelevich, “asymmetric modified spur gear drives: reduction of noise, localization of contact, simulation of meshing and stress analysis,” computer methods in applied mechanics & engineering, vol. 188, no. 1-3, pp. 363-390, july 2000. [3] f. karpat, s. ekwaro-osire, k. cavdar, and f. c. babalik, “dynamic analysis of involute spur gears with asymmetric teeth,” international journal of mechanical sciences, vol. 50, no. 12, pp. 1598-1610, december 2008. [4] f. karpat and s. ekwaro-osire, “influence of tip relief modification on the wear of spur gears with asymmetric teeth,” tribology transactions, vol. 51, no. 5, pp. 581-588, september 2008. [5] r. prabhu sekar, “performance enhancement of spur gear formed through asymmetric tooth,” proceedings of the institution of mechanical engineers, part j: journal of engineering tribology, vol. 233, no. 9, pp. 1361-1378, september 2019. [6] r. prabhu sekar and g. muthuveerappan, “estimation of tooth form factor for normal contact ratio asymmetric spur gear tooth,” mechanism & machine theory, vol. 90, pp. 187-218, august 2015. [7] j. gonsalvis and g. rayudu, “varying the tooth-sum of a gear pair operating on a specified center distance—its effects on contact ratio and gear ratio,” proceedings of the national conference of mechanisms and machines, pp. 139-146, 1990. [8] a. a. dsa and j. gonsalvis, “investigation on load distribution based on combined stiffness in altered tooth-sum gears,” international conference on mechanical power transmission, pp. 2-7, july 2019. [9] j. gonsalvis and h. sachidananda, “altered tooth-sum gearing for high contact ratio,” international journal of engineering research and applications, vol. 1, no. 3, pp. 1234-1241, octorber 2011. [10] k. ichimaru and h. fujio, “dynamic behavior of heavy-loaded spur gears,” journal of engineering for industry, pp. 373-381, 1974. [11] s. m. arikan, “dynamic load and contact stress analysis of spur gears,” american society of mechanical engineers, design engineering division (publication) de, vol. 32, no. pt 1, pp. 85-91, 1991. [12] g. maitra, handbook of gear design, new delhi: tata mcgraw hill, 1994. [13] j. r. colbourne, the geometry of involute gears, new york: springer verlag new york, 1987. [14] j. f. archard, “contact and rubbing of flat surfaces,” journal of applied physics, vol. 24, no. 8, pp. 981-988, august 1953. [15] a. flodin and s. andersson, “simulation of mild wear in spur gears,” wear, vol. 207, no. 1-2, pp. 16-23, june 1997. [16] r. kasuba, “dynamic loads in normal and high contact ratio spur gearing,” international symposium on gearing and power transmissions, 1981, pp. 49-54. [17] a. kahraman and g. w. blankenship, “effect of involute contact ratio on spur gear dynamics,” journal of mechanical design, vol. 121, no. 1, pp. 112-118, march 1999. 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/).  proceedings of engineering and technology innovation, vol. 1, 2015, pp. 36 39 36 cop y right © taeti using bpmn to model a patient safety promulgation service based on a clinical process way-ren huang 1 , ruo-chia tseng 2 , yung-fu hsu 1 , tso-ying lee 3 , woei-chyn chu 1,* 1 institute of biomedical engineering, national yang -ming university, taipei, taiwan. 2 department of molecular biology and human genetic, tzu chi university, taipei, taiwan . 3 nursing department, cheng hsin general hospital, taipei, taiwan. received 16 march 2015; received in revised form 20 april 2015; accept ed 02 may 2015 abstract in this paper, we used business process modeling standards to enhance the development of a patient-safety promulgation process with a web service. a web service that is related to a work process is typically in an organization that has been d esigned, built and stored in a web service repository. the pu rpose of using a web service is that it can be shared with mult iple functional units within an organization via an intranet. this paper further develops this approach and discusses the use of a visual modeling language to describe a clinical workflow clearly. the final outcome is business process re-engineering and is presented as a business process modeling notation. the workflow management we designed integrates a process for each functional unit in the organization to customize the workflow requirements that suit their needs. 748 cases had been notified among 13 months, the system play its role well even when notify proceeding had been changed. ke ywor ds : workflow management, web service, business process modeling notation 1. introduction in the early 20th century, we witnessed progress in modern manufacturing facilities, which could mass -produce and deliver products to large audiences in an unprecedented manner and speed. although this type of advance can be attributed to the development of telecommunication and advanced intelligent information technologies, establishing an optimal working environment and managing the required amount of work efficiently pose major challenges for business organizations. the reason for this challenge is because of the unstructured and unpredictable nature of demand and supply [5]. developments in e-commerce that are associated with the automation of the integration of heterogeneous systems and agent choreography can potentially answer some of the challenges in clinical workflow development. workflow management (wfm) or business process modeling notation (bpmn) has placed a large amount of attention on the need to improve the business process as well as the productivity of many organizations [6][9]. having this approach move through the work force and allowing people to work in an optimal environment to produce a maximu m output is increasingly important in business [3]. a business process is a series of business activities that describe how organizations produce and deliver special services to users, customers, or markets [8]. wfm is the way to manage the flow of activities or events, which then pass to appropriate partners to produce the required products or services. it is an effective way of decreasing the risks of time delays, costly amendments, insufficient resources and human errors. however, traditional wfm concepts are difficult to implement without the assistance of information technology [4]. we describe a traditional (before) and e-commerce (after) promulgation process in fig. 1. therefore, we use the bpmn to define, create and control a workflow by running it through a workflow engine. thus, the bpmn includes a process design, a process enactment and a system configuration [6]. in our study, we propose the use of a promulgation process of web services to support the workflow management. in the proposed promulgation workflow management (pwm) design, we describe promulgation architectures with a bpmn, and we use a business process execution language (bpel) to approach the web service. fig. 1 architectural diagram of the relationships of traditional and electronic patient safety architecture 2. methods design science is a technology-oriented paradigm that has its roots in engineering and the sciences of the artificial. whereas natural science attempts to understand reality, design science attempts to create things to meet certain desired goals [1]. generally accepted activities in design science are to build and to evaluate. build is the process of constructing an artifact to meet specific goals while evaluate concerns how well the constructions achieve the purpose. typical results in design science include construct, model, method and instantiation [2]. the following three languages have been considered to have adequate support for the workflow *corresponding aut hor. email: wchu@ym.edu.tw proceedings of engineering and technology innovation, vol. 1, 2015, pp. 36 39 37 cop y right © taeti modeling that is required in clinical settings: bpmn, un ified modeling language (uml) and the xml process description language (xpdl). according to our experience, we chose bpmn because of its simple and clear diagrams, its formalizations, its native support for web service invocations, and the availability of workflow engin es. we constructed promulgation workflow model (pwm) architecture (fig. 2) between a clinical process (fig. 3) and a bpmn in the build phase, and as expected, this architecture was quite straightforward. for example, the main process showed the general concept of the promulgation process; a sub-process or tasks may be used to model the clinical activity, the gateways for various clinical decision points and a pool or lane for clinical roles (e.g., nurse, head nurse, supervisor nurse, director associate, director, and quality manager (administrators)). fig. 2 the main process of the promulgation workflow model (pfm) fig. 3 this framework integrated the concept of a clinical pathway (based on the subject, object, action, and plan) and the workflow our system uses mssql as the database server and c # and business process execution language (bpel) as the web design languages. in this implementation, we attempt to utilize the current hardware resources of the organization. 3. results in the clinical promulgation process of a hospital scenario, when an accident event occurs, the duty nurse must write all of the contents of the accident event and notify her/his executive and then follow-up on the process or event. if the event occurs during the promulgation process, then there are several opinions that will be filed on the event by different people. finally, when the inner department’s promulgation process is completed, the administrator must decide on the end of the process or let it into the outer department’s promulgation process (fig. 3). it is difficult for medical nurses to accomplish these tasks, especially when they are in a health care environment. this difficulty will increase the time delays and human errors and decrease the quality of the medical care and the patient satisfaction. furthermore, it is difficult for those people who must file their opinions. however, the additional requirement for a nurse to work more carefully and precisely is likely to increase their work stress and, consequently, the numbers of mistakes made [7]. 3.1. architecture of bpmn for the above reasons, the design of the pwm architecture must be easy and clear to promulgate the accident events. for these reasons, we design a simple major process in the bpmn, which is shown in fig. 2. we only define three sub-processes, which are identify checking, work and notify. identify checking is the first sub-process, which is used to confirm the user identification and to take a user’s commission for the next sub-process (fig. 4). fig. 4 the sub-process of id checking then, the sub-process of work (user management) runs when the identification is successful upon checking. the sub-process of work classifies the users into three basic categories, which are based on their usages, their security clearance levels and the functional tasks of the process (fig. 5). fig. 5 the sub-process of work the first category (level 1) is the end-user (duty nurse), who initiates the promulgation process, builds the event details, and signs the event number (fig. 6). fig. 6 the sub-process of level 1 proceedings of engineering and technology innovation, vol. 1, 2015, pp. 36 39 38 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti the second category (level 2) is the director associate, supervisor, head of unit or department nurse. their respons ibilities are to manage the flow of the business function and to control the information and the accessibility of the system to each employee (fig. 7). fig. 7 the sub-process of level 2 finally, the third category (level 3) is the administration staff, who supports and maintains the pwm system. they can break/cancel the process at any time and can decide whether the process ends in the department or not (fig. 8). fig. 8 the sub-process of level 3 when the system checks whether the work process is complete, the promulgation process will begin the process of sending message services and then will notify those who should receive notification (fig. 9). fig. 9 the sub-process of sending a message service the user management policy of the pwm is designed to protect private information and the assignment of security clearance to any users. the goals of the design are to help end-users work faster with greater efficiency and to make fewer errors compared with the traditional promulgation of workflow arrangements. 3.2. interface design based on a user friendly approach (human-computer interaction), almost all of the user interfaces are comparable to the manual’s specifications. in the system, we include the following 11 forms of accident event: patient accident (1.1), remove tube (1.2), fall down (1.3), slash (1.4), suicide (1.5), violence (1.6), med err (1.7), burn (1.8), trans blood err (1.9), operate err (1.10), and else (1.11). we show the real screen at 1.1, 1.2, and 1.3 in fig. 10. fig. 10 three forms at 1.1, 1.2 and 1.3 the end-user can easily use this system without considering the details of the workflow. the supervisor, director associate, head of unit or department nurse can also clearly track the process instantly (fig. 11). fig. 11 process tracking and management interface. 4. conclusions after discussion, research, development and deployment, we expect that the concept of the design can achieve the six targets that are given below, in the hospital.  instantaneous notification: the reporting system works according to different levels and different types to notify those who should be notified by sending a message automatically.  record integration: under the system database management, the system can avoid writing the same information at the same time.  report integration: calculate the month / season / annual reports automatically.  process simplification: the e-process is easier than the traditional process.  saving time: the major benefits of information technology involve saving time.  improve satisfaction: good quality can improve the overall satisfaction. proceedings of engineering and technology innovation, vol. 1, 2015, pp. 36 39 39 cop y right © taeti we show how to remodel a clinical workflow with bpmn, and we execute the model. this approach encourages the promulgation guideline formation while shortening the mo deling and execution gap. although this paper focused in one specific application (the patient accident promulgation process), the proposed approach supports scalability and generalization. we deployed this system in its final vision in the hospital for all of the staff on jan-1-2011. acknowledgement this article accepts national yang ming university and cheng hsin general hospital research project grant (project number: 98f117cy17). references [1] a. hevner and s. chatterjee, design research in information systems, springer us, 2010. [2] a.r. hevner, s.t. march, j. park, and s. ram, “design research in information systems research,” mis quarterly, vol. 28, no. 1, pp 75-105, 2004. [3] b. reiner, e. siegel, and j. a. carrino, “workflow optimization: current trends and future directions,” journal of digital imaging, vol. 15, no. 3, pp. 141-152, 2002. [4] c. co mb i and g. po zzi, “architectures for a te mpora l workflow management system,” proceedings of a cm symposium on applied computing, nicosia, cyprus, march 2004. pp. 659-666. [5] j. chen, d. he, and w. anquan, “e-commerce and innovation business process reengineering,” portland international conference on management of eng ineering and technology (picmet '01), coll. of manage, vol. 1, pp. 163-164, 2001. [6] l. fischer, 2007 bpm and workflow handbook, lighthouse point, future strategies inc., florida, 2007. [7] r. eley, t. fa llon, j. soar, e. buikstra, and d. hegney, “barriers to use of information and computer technology by australia's nurses: a national survey.” journal of clinical nursing, vol. 18, pp. 1151-1158, 2009. [8] u. j. gelinas, s. g. sutton, and j. fedorowicz, “business process es and information technology,” cincinnati: south-western/thomson learning, 2004. [9] w. m. p. van der aalst, a. h. m. t. hofstede, and m. weske, “business process management: a survey,” lecture notes in computer science, vol. 2678, pp. 1-12, 2003.  proceedings of engineering and technology innovation, vol. 17, 2021, pp. 50-57 ftir analysis of biodegradation of polystyrene by intestinal bacteria isolated from zophobas morio and tenebrio molitor hong-han lin, hsiao-han liu * department of biological science & technology, i-shou university, kaohsiung, taiwan received 02 april 2020; received in revised form 03 september 2020; accepted 06 november 2020 doi: https://doi.org/10.46604/peti.2021.5450 abstract since the 1970s, polystyrene (ps) plastic marine pollution has become a global concern. tenebrio molitor and zophobas morio beetle larva actively respond to a diet of primarily polystyrene. the tantalizing evidence indicates that the gut bacteria of these beetle larva enables them not only to consume polystyrene, but to successfully biodegrade polystyrene. heretofore, data collection to verify polystyrene degradation by the gut bacteria of these larva has taken up to six months per test. our laboratory created a platform by dissolving ps into a liquid mineral culture medium to examine ps degradation by the gut bacteria. under investigation, ps works as main carbon source supporting the growth of gut bacteria. fourier-transform infrared spectroscopy (ftir) is an effective method which can provide relevant information on chemical changes. this study presents a methodology using ftir and visible light spectrometry as an efficient methodology to verify the physiological degradation of polystyrene. keywords: polystyrene, tenebrio molitor, zophobas morio, ftir 1. introduction polystyrene (ps) marine pollution had intensively increased since the 1970s [1]. plastic contributes to 60-90% of marine debris and 50% of debris belongs to the single-use plastics (sups), including polystyrene [2]. recent publications suggest that weathering processes promote further fragmentation of ps into nano-plastics, and show a metabolic toxicity which can change the behavior of marine creatures, or damage human health [3-4]. previous research successfully isolated two bacterial strains from the intestines of tenebrio molitor and zophobas morio larval worms. turbidity assay evidence suggested that these bacteria have the ability to digest polystyrene and successfully survive [5]. with two potentially ps-degrading enzymes identified in the foregoing research, the efficacy and enzymatic catalyzing mechanism were still under investigation. that is the research that this paper takes up. our laboratory adopted the naming conventions of “zm1”for the bacterial strains from zophobas morio and “tm1” for the strains taken from tenebrio molitor. 16srnas sequence of each revealed the taxonomic status of tm1 to be genus pantoea and that of zm1 to be genus klebsiella. zm1 has a 99% genetic similarity to klebsiella pneumoniae dsm 30104; tm1 has a 96% genetic similarity to pantoea agglomerans jcm 1236 [5]. with these potentially ps-degrading strains identified, they were cultured in a medium with polystyrene (ps) as the sole carbon source, making the tm1 and zm1 no option for sustenance other than polystyrene. according to our data, tm1 and zm1 began to synthesize bacterial enzymes that seem capable to successfully degrading the ps-medium [5]. fourier-transform * corresponding author. e-mail address: hansliu@isu.edu.tw proceedings of engineering and technology innovation, vol. 17, 2021, pp. 50-57 51 infrared spectroscopy (ftir) has been commonly applied in the field of biochemistry as a powerful technique to measure structural changes of organic and inorganic molecules and their functional groups [6-10]. in order to digest polystyrene as a carbon source, a secretion of the specific enzymes that can bind a polystyrene substrate which is required for the survival of the intestinal bacteria. therefore, if any enzymatic catalyzation of the carbon bonds in the ps-medium occurred, ftir could be a potential tool for measuring these molecular changes and demonstrating the sites where enzymes tend to activate [6]. 2. materials and methods 2.1. isolating polystyrene-degrading microbes the experiments began by acclimating the tenebrio molitor and zophobas morio larval worms to a polystyrene diet for a period of 21 days (fig. 1). ps-degrading bacterial strains tm1 and zm1 were then isolated from the intestines by selective ps agar plates [5]. (a) zophobas morio (b) tenebrio molitor fig. 1 two larval worms’ species subsisting on polystyrene 2.2. bacterial cultivation and ps-biodegradation experiments the intestinal bacterial strains zm1 and tm1 were inoculated and incubated within the ps-medium. a customized platform for polystyrene biodegradation testing of tm1 and zm1 from the previously cited research was used [5]. the ps-medium was 50 ml, with 45 ml of inorganic salt basal buffer, 5 ml of polystyrene emulsion, and 0.05 ml of 38% yeast extract. chloroform is used to dissolve polystyrene into emulsion, but after that it will be vaporized, so the final product doesn’t contain chloroform. at this stage, three tubes of ps-medium were loaded in 50 ml centrifuge tubes; two of them were inoculated with bacterial strains zm1 and tm1. each was cultivated for further ftir testing. the remaining tube of medium was designated as the un-inoculated control. the ftir cultivation courses were operated under a stable temperature at 37°c, 144 hours (6 days). 2.3. turbidity analysis by absorbance and transparency first, quantification and comparison of absorbance and transparency in ps-medium during cultivation course was conducted. second, the absorbance and transparency levels of the ps-medium were used as a measure of turbidity. turbidity is considered a quantitative indicator for ps degrading efficiency; analysis of this kind is also referred to as a turbidity assay [5, 11]. the absorbance of ps-medium was measured by microplate spectrophotometer (thermo scientific multiskan go), and 200 μl of sample were loaded for each measurement: from 230-900 nm. the od value was measured every 1 nm of wavelength. transparency was determined by spectrophotometer (metertech sp-830), and 1 ml of sample was loaded for each proceedings of engineering and technology innovation, vol. 17, 2021, pp. 50-57 52 test: from 400-700 nm. the percentage of transparency was measured at every 10 nm of wavelength. all measurements were performed with a time lapse at 0 hr, 24 hr and 48 hr. 2.4. ftir analysis ftir spectroscopy was applied to the ps-medium after the cultivation of two strains to determine if enzymatic changes consistent with degradation of carbon bonds took place [12]. to avoid the interference of non-specific organic molecules within the microbes’ outcome spectra, the samples (two inoculums and one control) were first centrifuged (4000 rpm, 25°c, 10 min) to remove debris and particles. after centrifugation, 3 ml of supernatant of the medium was dried in a drying oven. the dried sample was then mixed with kbr salt powder (0. 5 mg of sample plus 20 mg of kbr) and hand-pulverized by mortar; 10 mg of mixture were tableted for ftir test afterward. ftir spectroscopy (perkin-elmer ft-ir rx1) at 4000-400 cm -1 was performed and the samples were collected at day 0 and day 6 [13]. 3. result and discussion 3.1. absorbance and transparency determination the turbidity of ps-medium during cultivation was measured based on absorbance and transparency. the absorbance levels of the experimental groups gradually decreased during the cultivation time lapse (fig. 2), and the transparency rose (fig. 3), while the control remained stable. this data suggests that the activation of tm1 and zm1 caused ps to degrade in the medium. (a) absorbance with inoculation of strains zm1 (b) absorbance with inoculation of strains tm1 (c) absorbance of control (un-inoculated) fig. 2 turbidity assay based on absorbance to demonstrate biological degradation of polystyrene by the gut bacteria of tenebrio molitor and zophobas morio beetle larva this study performed turbidity assays based on absorbance and transparency on inoculated tm1 and zm1 gut bacteria 0 1 2 3 4 5 200 400 600 800 1000 o d 2 3 0 ~ 9 0 0 wavelength (nm) 0 hr 24 hr 48 hr 0 1 2 3 4 5 200 300 400 500 600 700 800 900 1000 o d 2 3 0 ~ 9 0 0 wavelength (nm) 0 hr 24 hr 48 hr 0 1 2 3 4 5 200 400 600 800 1000 o d 2 3 0 ~ 9 0 0 wavelength (nm) 0 hr 24 hr 48 hr proceedings of engineering and technology innovation, vol. 17, 2021, pp. 50-57 53 strains. the assay data reveal declining turbidity in the ps-medium across the full visible light spectrum, not only at the 600 nm point as the previous study had described [5]. (a) transparency with inoculation of strains zm1 (b) transparency with inoculation of strains tm1 (c) transparency of control (un-inoculated) fig. 3 turbidity assay based on transparency fig. 4 visualization of turbidity 0 20 40 60 80 100 350 400 450 500 550 600 650 700 750 t ra n sp a re n c y ( % ) wavelength (nm) 0 hr 24 hr 48 hr 0 5 10 15 20 25 30 350 400 450 500 550 600 650 700 750 t ra n sp a re n c y ( % ) wavelength (nm) 0 hr 24 hr 48 hr 0 10 20 30 350 400 450 500 550 600 650 700 750 t ra n sp a re n c y ( % ) wavelength (nm) 0 hr 24 hr 48 hr proceedings of engineering and technology innovation, vol. 17, 2021, pp. 50-57 54 fig. 5 bacterial cfu (measured by macconkey agar plate) fig. 6 bacterial growth curve a series of photos of strain tm1 and zm1 growing in ps-medium also shows the changes of turbidity. ps-medium inoculated with tm1 and zm1 become increasingly transparent from time at 0 hr to 96 hr, while the control group remains its originally opaque and turbid (fig. 4). figs. 5-6 shows the cfu data (measured by macconkey agar plate) represent bacterial concentration and growth. in cfu test, the dilution factor of zm1 at time 0, 24, 48 and 72 hr were 1, 10 7 , 10 6 and 5×10 6 , respectively; tm1 at time 0, 24, 48 and 72 hr were 10 2 , 2.5×10 6 , 10 6 and 10 7 , respectively. all colonies were yielded by plating 0.1 ml of diluted sample (fig. 5). 3.2. ftir analysis fig. 7 shows an ftir spectrum of the ps medium which is inoculated with tm1 and zm1. fig. 7(c) shows the un-inoculated control group for comparison. both the test and the control samples contained the same polystyrene and inorganic salts medium. hence, overlapping of peaks across these multiple salts and polystyrene substrate would be possible. according to the spectrum at day 0, the absorption peaks at 3400 cm -1 , 1044 cm -1 , 994 cm -1 , 926 cm -1 , and 856 cm -1 have appeared with initial inoculation of zm1 and tm1 (figs. 7(a) and 7(b)). by comparing the spectral data onto spectral database for organic compounds (sdbs), these peaks were regarded as signal of glycerol, the remaining solution from zm1 and tm1 bacterial stock (glycerol was used to preserve bacterial strain). at the initial stage, the medium of inoculated groups share the same constituent of the control group, except glycerol from the bacterial stock solution. after 6 days of the experiment, weakening on multiple peaks occurs to the inoculated test groups compared to the control group. the reduced peaks seen in the range of 3000 cm -1 to 3100 cm -1 (fig. 7) represent organic molecule functional groups including the band attributed to the aromatic c－h stretch. another absorption peak change in the test group occurred in the range of 2800 cm -1 to 3000 cm -1 , which is assigned to the aliphatic c－h stretch. additionally, changes seen occurring at 1654 cm -1 , 1494 cm -1 , and 1453 cm -1 are attributed to aromatic c = c stretch. and a c－h in-plane bending peak change occurred at 1109 cm -1 in the test group. aromatic rings and aliphatic c－h-rich backbones are two of the typical molecular features of polystyrene. 0 1 2 3 4 5 6 7 8 9 10 0 24 48 72 b a c te ri a l c o n c e n tr a ti o n (l o g 1 0 c f u /m l ) time (hours) cfu zm1 tm1 proceedings of engineering and technology innovation, vol. 17, 2021, pp. 50-57 55 compared with the spectra of the un-inoculated control, the ftir peak reductions seen in the carbon chain functional groups indicates that zm1 and tm1 have the ability to degrade carbon in polystyrene. for those polymers with long hydrocarbon backbones, it is possible that enzymes produced by tm1 and zm1 are degrading polystyrene from the side chains of the molecular backbone itself (fig. 8). (a) spectra with inoculation of strains zm1 (day 0: black; day 6: red) (b) spectra with inoculation of strains tm1 (day 0: black; day 6: green) (c) spectra of un-inoculated control (day 0: black; day 6: violet) (d) spectra of all three groups at day 6 (zm1: red; tm1: green; control: violet) fig. 7 ftir spectrum within the frequency range of 4000－400 cm -1 ; ① refers to aromatic c－h stretch; ② refers to aliphatic c－h stretch; ③ refers to aromatic c = c stretch; ④ refers to c－h in-plane bending proceedings of engineering and technology innovation, vol. 17, 2021, pp. 50-57 56 the alteration or replacement of these side groups could create a new functional group, which would be feasible for other enzymes to break down along the main hydrocarbon chain. for example, research indicates that benzene 1,2-dioxygenase can degrade benzene (an aromatic derivative). considering that the aromatic rings of ps were also degraded in our study, tm1 and zm1 may be shown to share a similar mechanism with benzene 1,2-dioxygenase degrading ps, by altering the benzene ring side chain. fig. 8 molecular structure of polystyrene (ps), with a benzene side chain and c－h rich backbone 4. conclusions to exclude the possibility of bacteria death causing the turbidity decrease, cfu tests were conducted and these showed no morbidity in the bacteria. with bacterial colony stable and turbidity decreasing, the inference is that the decrease in ps-medium turbidity was the result of digestion of ps emulsion. turbidity assay as a means of verifying plastic biodegradation, as well as a means to quantify the efficiency of plastic decomposition, has been cited in other studies [11, 14-15]. through ftir analysis, multiple carbon bonding changes were revealed. these changes at the aromatic c－h stretch; the aliphatic c－h stretch; the aromatic c=c stretch and the c－h in-plane bending offer further evidence to suggest that it is the digestive processes of the bacteria strains tm1 and zm1 in the ps-medium which were responsible for the alternations of the absorption peaks levels related to organic function seen [16-17]. other research has discovered a potential manner for enzymatic degradation of polystyrene, utilizing hydroquinone peroxidase produced by azotobacter beijerinckii hm121 [18]. hydroquinone peroxidase is an enzyme which can oxidatively catalyze phenols such as hydroquinone, a derivative of benzene rings [19]. as a polymer molecular structure, it is notable that polystyrene is a rich in benzene rings as well [20]. other studies [21-22] have mentioned the degradation ability of certain beetle larvae including the tenebrio molitor, however the identification of the specific microbes and their enzymatic degradation mechanisms were not yet be described in detail. in conclusion, our laboratory utilized the entire range of visible light wavelengths to scan turbidity via absorption and transparency modes to explore the optimal condition for future experiments. through ftir analysis our laboratory revealed structural changes of the polystyrene at the sites above as the tm1 and zm1 incubations which indicate biodegradation. further studies offers researchers the chances to isolate, identify and potentially clone the specific intestinal enzymes that enable bacterial strains such as tm1 and zm1 to break down polystyrene and expanded genetic engineering targets for this biodegrading enzymes. conflicts of interest the authors declare no conflict of interest. benzene ring; could be altered by benzene 1,2-dioxygenase (ec 1.14.12.3) proceedings of engineering and technology innovation, vol. 17, 2021, pp. 50-57 57 references [1] s. s. yang, a. m. brandon, d. f. xing, j. yang, j. w. pang, c. s. criddle, et al., “progresses in polystyrene biodegradation and prospects for solutions to plastic waste pollution,” iop conference series: earth and environmental science, vol. 150, pp. 1-10, may 2018. [2] r. e. schnurr, v. alboiu, m. chaudhary, r. a. corbett, m. e. quanz, k. sankar, et al., “reducing marine pollution from single-use alastics (sups): a review,” marine pollution bulletin, vol. 137, pp. 157-171, december 2018. [3] l. yin, h. liu, h. cui, b. chen, l. li, and f. wu, “impacts of polystyrene microplastics on the behavior and metabolism in a marine demersal teleost, black rockfish (sebastes schlegliei),” journal of hazardous materials, vol. 380, pp. 120861, december 2019. [4] m. revel, a. châtel, and c. mouneyrac, “micro (nano) plastics: a threat to human health?” current opinion in environmental science & health, vol. 1, pp. 17-23, february 2018. [5] z. l. tang, t. a. kuo, and h. h. liu, “the study of the microbes degraded polystyrene,” advanced technology innovation, vol. 2, pp. 13-17, 2017. [6] b. stuart, “ftir of biomolecules,” reviews in cell biology and molecular medicine, https://onlinelibrary.wiley.com/doi/abs/10.1002/3527600906.mcb.200300059, 2006. [7] s. bonhomme, a. cuer, a. delort, j. lemaire, m. sancelme, and g. scott, “environmental biodegradation of polyethylene,” polymer degradation and stability, vol. 81, no. 3, pp. 441-452, 2003. [8] g. chen, s. liu, s. chen, and z. qi, “ftir spectra, thermal properties, and dispersibility of a polystyrene/montmorillonite nanocomposite,” macromolecular chemistry and physics, vol. 202, no. 7, pp. 1189-1193, april 2001. [9] d. olmos, e. martin, and j. gonzalez-benito, “new molecular-scale information on polystyrene dynamics in ps and ps–batio 3 composites from ftir spectroscopy,” physical chemistry chemical physics, vol. 16, no. 44, pp. 24339-24349, 2014. [10] h. kundungal, m. gangarapu, s. sarangapani, a. patchaiyappan, and s. p. devipriya, “efficient biodegradation of polyethylene (hdpe) waste by the plastic-eating lesser waxworm (achroia grisella),” environmental science and pollution research, vol. 26, no. 18, pp. 18509-18519, may 2019. [11] yi-feng hsieh, “degradation of polyethylene succinate by thermophilic microbispora sp. bc35t-4,” master's dissertation, department of life science, fu-jen university, 2008. [12] h. k. chan, b. ongpipattanakul, and j. au. yeung, “aggregation of rhdnase occurred during the compression of kbr pellets used for ftir spectroscopy,” pharmaceutical research, vol. 13, no. 2, pp. 238-242, february 1996. [13] h. auta, c. emenike, and s. fauziah, “screening of bacillus strains isolated from mangrove ecosystems in peninsular malaysia for microplastic degradation,” environmental pollution, vol. 231, pp. 1552-1559, december 2017. [14] j. augusta, r. j. müller, and h. widdecke, “a rapid evaluation plate-test for the biodegradability of plastics,” applied microbiology and biotechnology, vol. 39, no. 4-5, pp. 673-678, july 1993. [15] m. timmins, d. gilmore, n. lotti, m. scandola, r. fuller, and r. lenz, “a spectrophotometric method for detection of enzymatic degradation of thin polymer films,” journal of environmental polymer degradation, vol. 5, no. 1, pp. 1-15, 1997. [16] m. sobkowiak and p. painter, “determination of the aliphatic and aromatic ch contents of coals by ft-ir: studies of coal extracts,” fuel, vol. 71, no. 10, pp. 1105-1125, october 1992. [17] a. n. a. el. hendawy, “variation in the ftir spectra of a biomass under impregnation, carbonization and oxidation conditions,” journal of analytical and applied pyrolysis, vol. 75, no. 2, pp. 159-166, march 2006. [18] k. nakamiya, g. sakasita, t. ooi, and s. kinoshita, “enzymatic degradation of polystyrene by hydroquinone peroxidase of azotobacter beijerinckii hm121,” journal of fermentation and bioengineering, vol. 84, no. 5, pp. 480-482, 1997. [19] r. s. yao, m. sun, c. l. wang, and s. s. deng, “degradation of phenolic compounds with hydrogen peroxide catalyzed by enzyme from serratia marcescens ab 90027,” water research, vol. 40, no. 16, pp. 3091-3098, september 2006. [20] m. d. faber, “microbial degradation of recalcitrant compounds and synthetic aromatic polymers,” enzyme and microbial technology, vol. 1, no. 4, pp. 226-232, october 1979. [21] b. y. peng, y. su, z. chen, j. chen, x. zhou, m. e. benbow, et al., “biodegradation of polystyrene by dark (tenebrio obscurus) and yellow (tenebrio ｍolitor) mealworms (coleoptera: tenebrionidae),” environmental science & technology, vol. 53, no. 9, pp. 5256-5265, april 2019. [22] y. yang, j. wang, and m. xia, “biodegradation and mineralization of polystyrene by plastic-eating superworms zophobas atratus,” science of the total environment, vol. 708, p. 135233, march 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/).  proceedings of engineering and technology innovation, vol. 10, 2018, pp. 01 07 development of a dryer with a heat exchanger to recover energy for drying agricultural produce jeng-liang lin * department of biomechatronic engineering, national chiayi university, chiayi, taiwan, roc received 12 march 2018; received in revised form 14 june 2018; accepted 24 july 2018 abstract the purpose of this study was to investigate the effect of a heat recovery by guiding the exhaust air through a heat exchanger in a hot air dryer. the temperature of inlet air could be increased by the developed heat recovery device to achieve energy saving. experimental results indicated that the heat recovery device could increase the inlet fresh temperature 11℃, the effectiveness ratio of the heat exchanger was up to 70-80%, and heating efficiency of the heater was 50% through the analysis of inlet and outlet temperature, airflow rate as well as fuel consumption of the heater. in addition, the maximum temperature difference in the drying chamber was in the range of 0.6-1.2℃, indicating that the airflow in the drying was uniform, thus the uniform quality of agricultural produce could be obtained. in order to analyse the optimum operation conditions, the effect of fan speed on the temperature and airflow distribution was investigated in the drying system. the results showed that the optimum fan speed should be faster during start-up period and lower during a steady stage. keywords: drying, heat exchanger, energy, agricultural produce 1. introduction drying is one of the most important processes for processing agricultural produce. there are three kinds of drying systems used in agricultural industry, they are hot air drying [1-2], dehumidification drying [3-4] and natural drying [5-6]. however, the hot air dryer is the most widely used among mechanical drying systems, especially when the weather is adverse or the automated production is used, both must rely on the mechanical drying system. in the processing of agricultural produce, in addition to the labor cost, the energy consumption of drying process is one of the major costs. therefore, the purpose of this study was to develop a set of drying system in which the condition of hot air state could be adjusted, design a heat exchanger to recover energy and select an appropriate fan. the major demand of the drying system was to design the airflow duct to achieve the airflow distribute uniformly in the drying chamber in association with the energy recovery technology [7-8]. the energy recovery was via the heat transfer of entering the fresh air and exhaust air to increase the intake air temperature and reduce the energy consumption of drying. 2. design and manufacture of the drying system the developed dryer was a hot air drying system with capable of recirculation, the schematic diagram is shown as fig. 1. considering the general processing capacity and reducing the heat loss, the #430 stainless steel insulated box was used for the dryer body and its physical dimension was w2300mm × l2000mm × h2250mm. a diesel burner was used as the heater, the * corresponding author. e-mail address: jllin@mail.ncyu.edu.tw tel.: +886-5-2717656; fax: +886-5-2717647 proceedings of engineering and technology innovation, vol. 10, 2018, pp. 01 07 copyright © taeti 2 switch of the heater was controlled by the setting temperature. a 36” fan with airflow rate of 60 m 3 /min was selected to drive hot air through the drying chamber. fig. 1 schematic diagram of the dryer with a heat exchanger the heat recovery device with dimension of w1900mm × l400mm was a cross flow heat exchanger, its schematic diagram is shown as fig. 2. the major component of the device was copper tubes with 440 mm in length, 20 mm in diameter and 1.5mm thickness. the tube bank of staggered tube rows consisted of odd-row and even-row in the heat exchanger having 40 and 41 tubes respectively. totally, the heat exchange had 8 rows and 324 of tubes. each tube with 120 of disc-type fins was used to enhance the heat recovery performance. the airflow inside the tube was the fresh air, the external airflow of the tube was the high humid hot air to be exhausted to the outside of the dryer after flowing through the agricultural produce; hence, the designed heat recovery device could achieve the purpose of pre-heating the fresh air. the airflow path in the dryer chamber was fresh air flowing through the tubes of the heat exchanger from the ambient, then roughly heating up to the set temperature by the heater, and then entering into the dryer chamber via the fan to flow through the agricultural product to absorb the moisture, and then passing through the airflow circulation duct to discharge from the drying system. there was one set of switching baffle in the airflow circulation duct which was used to control the relative humidity of drying chamber. when the relative humidity was within the operating range, the baffle was adjusted to the non-circulated position (horizontal) that could discharge the high moisture hot air directly flowing through the heat exchanger out of the dryer. when the relative humidity was too low, the baffle was adjusted to the circulated position (tilted) that could guide the high moisture hot air into the dryer chamber again to maintain the humidity of hot air in the drying chamber in order to avoid damaging the quality of agricultural produce due to the humidity being too low. fig. 2 schematic diagram of the heat recovery device proceedings of engineering and technology innovation, vol. 10, 2018, pp. 01 07 copyright © taeti 3 the control system was mainly to control the temperature setting for start-up and shutdown of the heater, and control the humidity setting of circulation baffle. as to the fan, a frequency converter was used to control the rotating speed, the effective range of its speed control was in the range of 75-100% of maximum speed. in addition to the operating knobs on the control panel, there were some numerical displays of temperature, humidity and fan speed. there was one set of electronic scale at the bottom of drying chamber which was able to display the total weight of drying material with the led, and a built-in rs-232 communication interface was used to monitor and record the material weight during the drying process in real-time. the electronics scale was a welded high-strength steel plate and four weight sensors which were used to neasure the weight range of 0~2000 kg with the accuracy of 0.2 kg. 3. material and method the uniformity of temperature distribution in the drying chamber and the performance of heat recovery were the two indexes for assessing the developed heat recovery dryer. in terms of temperature distribution, the drying chamber was divided into 12 equal segments, since both the heater and fan were installed at the central location of the dryer, consequently, it was expected that the temperature of drying chamber should be bilateral symmetrically. therefore, analyzing the temperature of right-half space in the chamber would be able to describe the temperature distribution of overall dryer chamber. six k-type thermocouples were installed on the upper and lower positions of front, middle as well as rear sections at the right side portion of dryer chamber. each thermocouple was encased by a stainless steel tube, and then inserted into approximately 575 mm from the side of the dryer to measure the temperature at the upper and lower positions of the front, middle as well as rear sections in the right-half space of dryer chamber. hence, there were temperatures at 6 locations used to analyze the uniformity of temperature distribution. as to analyze the performance of heat recovery device, there were 3 k-type thermocouples installed to measure the temperature before and after the fresh air flowing through heat exchanger, and the temperature of high-temperature humid air before flowing through the heat exchanger. in addition, a k-type thermocouple was installed to measure the intake air temperature at the bottom air intake. as to the measurement of airflow speed in the drying chamber, there were 3 hot wire anemometers installed on the upper and lower positions of drying chamber front section as well as the front-end of airflow circulation duct. a portable relative humidity meter in association with the output signal cable, which was placed at the inside of drying chamber door, was used to measure the relative humidity in the drying chamber. the installation locations of sensors are shown as fig. 1. the thermocouples, humidity meter, and anemometers were all connected to an automatic data acquisition system (advantech adam 5000), and then the genie graphic control software was used to record the variation of temperature, humidity and airflow speed during tests. the experiments for temperature distribution, relative humidity change, and airflow distribution were performed without drying material in the dryer chamber with setting temperature of 50℃. the experiments of the optimum speed for fan were to set the frequency converter at 45, 50, 55 and 60 hz, and 4 replications were performed for each speed, the duration of each test was approximately 40 minutes. 4. results and discussion 4.1. analysis of the airflow and heat recovery device during the tests of fan at the maximum speed, when the temperature of drying chamber reached the setting temperature after the system was a start-up, the heater stopped and the temperature of intake air gradually decreased until the temperature in the drying chamber was lower than the start-up temperature of the heater. then the heater started again to increase the temperature of intake air. this kind of cycle was able to achieve the purpose of temperature control (fig. 3). the experimental results showed that the time required by the drying system from the start-up to the stable state was approximately 400 seconds, proceedings of engineering and technology innovation, vol. 10, 2018, pp. 01 07 copyright © taeti 4 afterward, the heater was in the mode of stopping for 120 seconds and starting for 80 seconds to maintain the variation of air temperature in the drying chamber within 5℃ approximately. therefore, even this drying system used the simplest on-off control technique, it could still control the hot air temperature effectively. it could also be shown from the temperature values at 6 locations that the temperature inside the drying chamber was very uniform. at the stage of the transient heating process after start-up, the temperature difference inside the drying chamber was larger, about 1.2℃. after the system was stable, the temperature difference between the highest temperature and lowest temperature was around 0.6-0.8℃. the experimental results all showed that the highest temperature value appeared at the upper position of the rear section of the drying chamber, while the lowest temperature value appeared at the upper position of the middle section. theoretically, this result was reasonable since there should have the highest temperature value after just passing through the heater; however, due to the heat transfer of partition board above the middle section, the temperature could be lower than the front section. as to the relationship between the inlet and outlet temperature as well as temperature difference of heat exchanger, it is shown in fig. 4. the experimental results showed that the heat exchanger was able to increase the air temperature up to 11℃. it indicated that the performance of the developed heat recovery device was good. fig. 3 variation of temperatures at 6 different locations and maximum temperature difference in the dryer fig. 4 variation of inlet and outlet temperatures of the heat exchanger as well as its temperature increment in the dryer the results of relative humidity change in the drying chamber showed that the relative humidity gradually decreased from 63% to a relatively stable value of 18% after around 20 minutes (fig. 5). from the psychrometric chart, it was able to obtain that relative humidity of air was approximately 20% for air with the temperature of 30℃ and relative humidity of 60% at the initial state after the heating process of temperature up to 50℃. from this evidence, it was able to verify that the results of this proceedings of engineering and technology innovation, vol. 10, 2018, pp. 01 07 copyright © taeti 5 study were reasonable and the developed drying system had a good airtight chamber. consequently, it was able to reduce the energy consumption due to the heat loss. as obtained from the psychrometric chart, the enthalpy values of air before and after the heating were 71 and 92 kj/kg respectively. in association with the airflow rate 60 m 3 /min and the dry air specific volume 0.9 m 3 /kg approximately, it was able to obtain the heat effectively added to the intake air was 84 mj/hr. from the fuel amount measured before and after the tests, the average fuel consumption of diesel was 4.5 liter/hr. in addition, based on the specific gravity and heat value of diesel used by the heater were 0.85 and 45 mj/kg respectively [9], combustion heat of diesel would be 172 mj/hr. therefore, the heating efficiency of heater for this drying system was about 50%. fig. 5 variation of relative humidity in the dryer the experimental results showed that the heat recovery device was able to increase the intake air temperature from the atmospheric temperature of 35℃ to 46℃. when the performance of heat exchanger was analyzed, heat exchange effectiveness ratio could be used [10]. the ratio describes the effectiveness of the heat exchanger for heating the cold fluid to the entering temperature of the hot fluid. the definition of the ratio is temperature difference for airflow flowing through the heat exchanger with lower heat capacity and the maximum temperature difference in the heat exchanger. the maximum temperature difference flowing through the heat exchanger was the difference value between the temperature of airflow circulation duct and the atmospheric temperature. since there was partial air intake entering the drying chamber through the bottom of the heater used to avoid overheating the heater, the fresh air in the heat exchanger had lower heat capacity. applying the above stated approach to assess the effectiveness ratio of heat recovery system required to measure the temperature of airflow circulation duct, which was the temperature of hot and humid air before flowing through the heat exchanger. generally, its value was about the same as the average temperature in the drying chamber. if the hot and humid air temperature was 50℃ for performing the analysis, the theoretical temperature increase limit was 15℃. therefore, the effective ratio of the heat recovery device could achieve over 73%. 4.2. optimum operation of the fan speed after the drying system was start-up, the temperature of drying chamber increased rapidly to reach the set temperature, and then the on-off control was used to achieve the purpose of temperature control. the temperature increase rate after the drying system start-up is shown in fig. 6. the experimental result showed that the faster the fan speed, the faster the transient response. when the fan speed increased from 45 to 60 hz, the temperature increase rate could be increased from 7 to 8℃/min, especially, in the range from 50 to 55 hz. fig. 7 shows the average and standard deviation values of maximum temperature difference for the total 6 locations of upper and lower positions in the front, middle as well as rear sections of drying chamber during the stable period, which might be the maximum temperature difference in the drying chamber. the experimental results proceedings of engineering and technology innovation, vol. 10, 2018, pp. 01 07 copyright © taeti 6 showed that the temperature in the drying chamber became less uniform at the fan speed of 55 hz, however, its value was only around 1℃. therefore, if the fan was controlled within the speed range, the drying chamber could have a fairly uniform temperature and airflow. hence, it was able to achieve the prerequisite of uniform quality for the agricultural produce. fig. 6 variation of temperature increase rate with fan speed after start-up fig. 7 variation of the maximum temperature difference among six locations in the dryer with fan speed fig. 8 shows the average and standard deviation values of effectiveness ratio of the heat exchanger with fan speed during the stable period. the experimental results indicated that the effectiveness ratio was in the range of 70-80% and the lowest effectiveness ratio was at the fan speed of 55 hz, which was consistent with the fact that temperature to be less uniform at the fan speed of 55 hz. although higher fan speed should have higher heat transfer resulting in higher effectiveness ratio, there was not a clear relationship between effectiveness ratio and fan speed. the reason might be that airflow was interrupted by the switching baffle in the airflow circulation duct before passing through the heat transfer and fan speed in the operation range was not a critical factor for heat transfer. applying the large spiral fan in a small space, its airflow direction was not likely to be uniform. for the three selected measuring locations, the measured results of airflow speed in the drying chamber are shown in fig. 9. the location of the maximum average airflow speed was in the front end of airflow circulation duct which was reasonable since its cross area was the smallest, and airflow velocity increased with the increase of fan speed; however, the airflow speed of that location did not increase significantly as the frequency over 50 hz. the results also indicated that increase of fan speed did not increase the airflow rate of the other two locations. fig. 8 variation of the effectiveness ratio of the heat exchanger with fan speed fig. 9 variation of airflow velocity at three locations in the dryer with fan speed as summarized from the experimental results of fan speed, the analysis of temperature data showed that faster fan speed after the start-up shortened the time of reaching the setting temperature and lower fan speed was better for the temperature proceedings of engineering and technology innovation, vol. 10, 2018, pp. 01 07 copyright © taeti 7 uniformity in the drying chamber. therefore, the optimum operation mode of fan speed should be faster during the start-up period and lower after reaching the stable stage for improving the performance of dryer. 5. conclusions the developed drying system of this study was able to achieve the heat recovery and reduce the energy requirement for drying agricultural produce. the experimental results showed that the developed drying system had a good airtight chamber and the effectiveness ratio of the designed heat recovery device could reach 70-80%. through the analysis of inlet and outlet temperature, airflow rate, as well as fuel consumption of heater, the heating efficiency of heater was up to 50%. in addition, the maximum temperature difference in the drying chamber was in the range of 0.6-1.2℃, which indicated that the temperature distribution of drying system was fairly uniform. therefore, it was able to achieve the purpose of uniform quality for the drying agricultural produce. in addition, from the experiments of the effect of fan speed on drying system, the optimum operation for fan speed should be faster at the heating start-up stage and lower after reaching the stable stage. references [1] p. p. lewicki, “design of hot air drying for better foods,” trends in food science & technology, vol. 17, no. 4, pp. 153-163, april 2006. [2] p. k. wankhadea, r. s. sapkala, and v. s. sapkal, “drying characteristics of okra slices on drying in hot air dryer,” procedia engineering, vol. 51, pp. 371-374, 2013. [3] c. l. hii, c. l. law, m. cloke, and s. sharif, “improving malaysian cocoa quality through the use of dehumidified air under mild drying conditions,” journal of the science of food and agriculture, vol. 91, no. 2, pp. 239-246, january 2011. [4] m. djaenia and d. a. sari, “low temperature seaweed drying using dehumidified air,” procedia environmental sciences, vol. 23, pp. 2-10, 2015. [5] j. d. hill, “predicting the natural drying of hay,” agricultural meteorology, vol. 17, no. 3, pp. 195-204, september 1976. [6] t. filbakk, o. hoib, and j. nurmi, “modelling natural drying efficiency in covered and uncovered piles of whole broadleaf trees for energy use,” biomass and bioenergy, vol. 35, no. 1, pp. 454-463, january 2011. [7] b. golman and w. julklang, “analysis of heat recovery from a spray dryer by recirculation of exhaust air,” energy conversion and management, vol. 88, pp. 641-649, december 2014. [8] b. golman and w. julklang, “simulation of exhaust gas heat recovery from a spray dryer,” applied thermal engineering, vol. 73, no. 1, pp. 899-913, december 2014. [9] g. l. borman and k. w. ragland, combustion engineering, new york: mcgraw hill, 1998. [10] j. p. holman, heat transfer, new york: mcgraw hill, 2002.  proceedings of engineering and technology innovation, vol . 8, 2018, pp. 09 14 an intelligent manufacturing system for injection molding wen-chin chen 1,* , manh-hung nguyen 2 , pei-hao tai 1 1 department of industrial management, chung hua university, hsinchu, taiwan, roc. 2 ph.d. program of technology management, chung hua university, hsinchu, taiwan, roc. received 17 november 2017; received in revised form 28 november 2017; accept ed 10 december 2017 abstract in recent years, the great trends of industry 4.0, internet of things (iot), big data analytics, and cloud computing, the design and development of plastic injection molding (pim) products has been more requested to achieve the requirements of light, thin, short, small, multi-function, high-precision, energy-saving, and obliged to fulfill a large number of customized production. to tackle this arduous challenge, effectively developing a novel pim intelligent manufacturing system will play a crucial role. the aim of the proposed study is to carry on building an intelligent manufacturing system (ims) for pim industry, which is composed o f three subsystems: a multiple response optimization systems of pim, a database management system of process parameters, and a pim real-time monitoring and control system. firstly, the multiple response optimization systems present an intelligent optimization system to find optimal process parameters of multiple quality characteristics in the pim process . secondly, the database management system allows for saving the experimental data, pim process parameter settings and quality goals. the third is a pim real-time monitoring and control system, which establishes a graphic monitoring and control interface to real-time monitor the parameters of pim machine and the optimal process parameter settings . the proposed pim intelligent manufacturing systems enable the functions of real-time monitoring, process parameter optimization and database management, which can assure better pim product quality and yield rate, effectively reduce the manufacturing cost, and promote the competition of the pim industry in the future. keywords: pim, industry 4.0, iot, big data, cloud computing, ims, bpnn, modified pso-ga 1. introduction plastic injection molding (pim) is a very important process to produce plastic parts. pim is suitable to use for mass production of products because it is easy to convert raw material to be a plastic product in a single automation process. other advantages of utilization of pim are easy to produce light, corrosion resistance, shape, and low processing cost. there are several processes of injection molding to produce plastic parts: plastic, injection, packing, cooling, and ejection. even though most engineers argue that this is an easy process, but in the practice pim , process is more complex than it is thought. inappropriate material selection, process parameters, part and mold designs can affect the quality of plastic products. several defects that frequently occur in the pim process , for instance, warpage, shrinkage, sink marks, and weld lines. in view of current market trends of plastic products, injection molding machine can be the most crucial plastic processing machinery, accounting for 60%-85% in developed countries, and it finds application mainly in numerous fields such as electronic communications , automobile, plastic building materials, household appliances, food & beverage packaging, and medical electronic apparatus . * corresponding author. e-m ail address: wenchin@ chu.edu.tw tel.: +886-3-5186585; fax: +886-3-5186575 proceedings of engineering and technology innovation, vol. 8, 2018, pp. 09 14 copyright © taeti 10 however, in recent, the pim production faces copious critical challenges where the customers or ends users continuously require highly customized products in small batches and in a sh ort period [1]. the development of new generation of more flexible and intelligent pim manufacturing systems was desired to satisfy their requirements in terms of variety, response time and quality [2]. in recent years, with the vigorous and rapid development of the smartphone, panel, biotechnology, sports and medical wearable devices industry under the great trends of industry 4.0, internet of things (iot), big data analytics, and cloud computing, the design and development of 3c (communication, computer, and consumer electronics), car electronics, and medical electronics products become increasingly diversified and complicated. besides, these products have been more requested to achieve the requirements of light, thin, short, small, multi-function, high-precision, energy-saving, and obliged to fulfill a large number of customized production. to tackle this arduous situation, effectively creating a novel pim intelligent manufacturing system of real-time executing ,monitoring and supervision, self-adapting, and reconfiguring rapidly will play a crucial role. 2. industry 4.0 the first industrial revolution was introducing mechanical production facilities , which commenced in the second half of the 18th century and being consolidated throughout the entire 19th centu ry. in the second industrial revolution the electrification and the taylorism’s division of labor were developed from the 1870s. the third industrial revolution, also called “the digital revolution”, occurred in the 1970s, and promoted in the automation of production and advanced electronics , and information technology. the “industrie 4.0 working group” advocates that the “industry 4.0” integrating the internet of things (iot) and cyber-physical systems (cps) into the manufacturing process as a crucial enabler will be the fourth industrial revolution [3]. the term and an initiative named “industry 4.0” became known publicly in 2011 as an association of representatives from business, politics, and academia supported the concept as an approach to strengthening the competitiveness of the german manufacturing industry [4]. the german federal government declares that industry 4.0 will be an integral part of its “high-tech strategy 2020 for germany” initiative, attempts to reach technological innovation leadership of the german economy [3]. the “industrie 4.0 working group” developed first implementing recommendations published in april 2013, and they named three main components of industry 4.0: the internet of things (iot), cyber-physical systems (cps), and smart factories and consider the merge of iot into the manufacturing process for the fourth industrial revolution. moreover, the iot, cps can become the fusion of the physical and the virtual world, and realize integrations of computation and physical processes while embedded computers and networks monitoring and control them, normally with feedback loops [5-6]. a recent paradigm is called “the fourth industrial revolution” (industry 4.0) in which technologies are combined to integrate machines and humans to comp ose value chains of entities (i.e., manufacturing factories) in different geographical locations distributed manufacturing systems , which will provide services and products in an autonomous manner [7]. industry 4.0 considers the former paradigms compris ing reconfiguration and technological advances associated with cyber-physical systems and cloud computing environment. in fact, there is not a unique technique to overcome all the challenges; an alternative solution is a hybrid of complementary characteristics involving different techniques wherein the comprehensive use of holonic and multi-agent system (hmas) concepts can facilitate the development and integration of distributed heterogeneous systems combining hierarchical and heterarchical structures [8]. moreover, the number of scientific topics and the achievements in the hmas field is boosting and been explored to propose solutions for industry 4.0 [9]. 3. intelligent manufacturing system intelligent manufacturing systems (ims) are the novel generation of manufacturing systems. all ims subsystems include parts of so-called machine intelligence (sensor equipment). for the better understanding to term “intelligent” manufacturing proceedings of engineering and technology innovation, vol. 8, 2018, pp. 09 14 copyright © taeti 11 systems, it is the most suitable to compare its behavior with the classical (“non -intelligent”) automated flexible manufacturing system. there are three basic types of automated manufacturing systems [10]: (1) flexible manufacturing cell amounts to maximum three of the machine tools characterized by the highest level of flexibility. (2) the flexible manufacturing line can be characterized by the lowest level of flexibility, and its range of goods is narrow and within large batches of products . (3) flexible manufacturing system includes minimum three machines and more characterized by a lower level of flexibility. the intelligent manufacturing systems present systems , which owned capable of adaptation to unexpected changes , and they also consisted of software components using such techniques as parameter optimization, fuzzy logic, neural networks, expert s ystems and machine learning [11]. intelligent manufacturing, which combines multi-functional machines and mobile robots, can be achieved in three basic ways [2]: 1. existing manufacturing processes become more intelligent by monitoring and controlling the state of the manufacturing machine. 2. existing processes can be intelligent by adding sensors to monitoring and control the state of the processed product. 3. new processes designed intelligently to produce parts of desired quality without the need of sen sing and control of the process. this example comprises a part agentrunning in an industrial personal computer (ipc), three machine agents representing three computer numerical control (cnc) machines , and a transport agentrepresenting an automated guided vehicle (agv) and running in a programmable logic controller (plc) [12]. the production data are fell into three categories: product (or order), knowledge (nc program, processing time, and transportation time), resource (storage, machine, agv). the basis of closed-loop control for distributed production control system is the introduction of real-time information of the product as a feedback [13]. comparing with the client-server paradigm, the mobile agent will strengthen the system adaptability and flexibility [14]. 4. research methods, procedures, and progress the proposed intelligent manufacturing simulation system will be composed of three main subsystems: a database management system of process parameters, a multiple response optimization system of pim, and a real-time monitoring and control system. a schematic module for the pim intelligent manufacturing system can be s een in fig. 1 and its procedures are described in the following sections. process parameters database management system multiple response optimization system real-time monitoring and control system (gui interface) pim machine control system plastic injection molding machines rfid reader fig. 1 a schematic module for a pim intelligent manufacturing system 4.1. multiple response optimization system the study propos ed a two-stage multiple response optimization system to find optimal process parameters of multiple quality characteristics in the pim process. the taguchi method, back-propagation neural network (bpnn), multi-objective analysis of variance (anova), and modified hybrid genetic algorithms and particle swarm optimization genetic algorithms (modified pso-ga) were used to find optimum parameter settings. length and width characteristics were employed as the proceedings of engineering and technology innovation, vol. 8, 2018, pp. 09 14 copyright © taeti 12 product quality. the experimental work was conducted using the taguchi orthogonal array table. according to the result from the taguchi experiment, s/n ratios were calculated. the s/n ratio response factor cha rt of integrating into a single quality (i.e., total bias) and the main effects plot for s/n ratios of total bias were demonstrated. in addition, the s/n ratio predictor an d the quality predictor were constructed using bpnn. a modified pso-ga algorithm is proposed to improve numerical performance for a hybrid pso-ga algorithm, which was used to find initial weights of bpnn in conjunction with multilayer perceptron (mlp) and to reduce the training time of bpnn. in the first stage optimization, t he s/n ratio predictor and ga were used to reduce variance of the quality characteristic. then, in the second stage optimization, the s/n ratio predictor and the quality predictor with a modified pso-ga algorithm were employed to find optimal parameter settings for pro duct quality and stability of the process. finally, confirmation experiments were conducted to assess the effectiveness of the proposed system. 4.2. database management system of process parameters the study utilizes mysql 5.7 to construct the process parameters database, and uses the microsoft visual studio 2015-visual basic to develop an information platform (i.e., gui control interface) of the intelligent manufacturing system, which encompasses three subsystems: a database management system of process parameters, a multiple response optimization system, and a real-time monitoring and control system. the database management system by functions can be separated into two portions: the first part is the intelligent process parameter optimization database, which can access the implemented experimental data; the second part is optimal parameter settings generated by a multiple response optimization system, which can access the results created by multiple response optimization system. the above databases have different functions, and their column name and data types of database design and structure are quite distinct. the data type of data configuration of process details database can be seen in table 1, which represents the data type of the remain columns are s et “single” with respect to requiring higher precision figures and setting the product id as “varchar”. on the other hand, data type of database for results of the process parameters optimization system is in table 2. it suggests that the database is associated with a product id, which is beneficial to the follow-up relation query. the target of length and width will be read by the rfid reader and save them into a quality database. table 1 data type of database for the experimental data column name product id. it iv pp pt pt ave. length ave. width std. length std. width s/n length s/n width data type varchar single table 2 data type of database for results of the process parameters optimization system column name product id. length goal width goal it iv pp pt ct data type varchar single 4.3. real-time monitoring and control system the proposed monitoring and control system will provide the pim process demonstration, process parameter settings, records, and so on. it is also a communication channel and response interface with the multiple response optimization system and the database system. the flow chart of a real-time monitoring and control system can be represented in fig. 2, and its software utility and hardware appliance can be denoted below: software utility (1) process parameter database records all the pim process parameters and their quality characteristics. (2) monitoring and control system will provide the pim process demonstration, process parameter settings, records, and etc. the system is a communication channel and response interface with the multiple response optimization system. proceedings of engineering and technology innovation, vol. 8, 2018, pp. 09 14 copyright © taeti 13 (3) the multiple response optimization system offers the injection molding machines real-time adjusted control parameters and quality control details for determining the optimal process parameter settings, which can achieve the quality requirements. hardware appliance (1) the hardware communication interface is an interface which connects the workstation and plastic injection molding machines. the control interface of pim machines need to abide the communication paradigms such as rs-232c, 422, 485, ethernet, ethercat, … etc., owing to the professional use of the proposed pim machine control system. (2) the signal transformation module proceeds to supervise and control the alternative pa rameters, like temperature of ice water machine, which are not encompassed in the pim machines, or caters the required signals transformation for on -line quality inspection system. (3) the workstation of real-time monitoring and control system plays an essential core of the whole intelligent manufacturing system, which operates the multiple response optimization system, accesses process parameter database, and supervises and controls pim machines via hardware communication interface, signal transformation module, on-line quality inspection module, network control module,…etc. fig. 2 flowchart of a real-time monitoring and control system 5. conclusions this study proposed an intelligent manufacturing system (ims) composed of three subsystems: intelligent parameter optimization system of pim process, database management system, and real-time monitoring and control system. firstly the intelligent parameter optimization system uses taguchi method, anova, bnd mpnn aodified pso-ga methodologies to search for the optimal parameter setting. then the database management system is dedicated to accessing the experimental data and pim process parameter settings of fmm, which encompasses etching product id with quality targets. as to the pim real-time monitoring and control system, it will establish a graphic monitoring and control interface, whose monitoring scopes includes real-time monitoring the parameters of pim machine and the optimal process parameter settings created by the multiple response optimization system, and transferring the optimal process parameter settings into pim machines and simultaneously changing proceedings of engineering and technology innovation, vol. 8, 2018, pp. 09 14 copyright © taeti 14 their parameters. the proposed intelligent manufacturing system of a pim process will help the pim firms search for better parameter settings for new plastic products , and facilitate the pim firms easier to avoid product defects and build the sustainable competitive advantage over their competitors in the world. references [1] s. wang, j. wan, d. zhang, d. li, and c. zhang, “towards smart factory for industry 4.0: a self-organized multi-agent system with big data based feedback and coordination,” computer networks , vol. 101, no. 4, pp. 158-168, 2016. [2] a. boboli, j. okamoto, m. s. tsuzuki, t. c. martins, p. e. miyagi, and f. junqueira, “intelligent manufacturing system configuration and optimization considering mobile robots, multi-functional machines and human operators: new facilities and challenge for industrial engineering,” international federation of automatic control-papers on line, vol. 48, no. 3, pp. 1912-1917, 2015. [3] m. hermann, t. pentek, and b. otto, “design principles for industrie 4.0 scenarios ,” 49th hawaii international conf. system sciences (hicss), ieee press, january 2016, pp. 3928-3937. [4] h. kagermann, recommendations for implementing the strategic initiative industrie 4.0, national academy of science and engineering, berlin, forschungsunion, 2013. [5] e. a. lee, “cyber physical systems: design challenges,” 11th ieee symposium object oriented real-time distributed computing, may 2008, pp. 363-369. [6] h. k. agermann and w. d. lukas, “industrie 4.0: mit dem internet der dinge auf dem weg zur 4. industriellen revolution,” https://goo.gl/7c4l2w, april 13, 2014. [7] p. leitão, n. rodrigues, j. barbosa, c. turrin, and a. pagani, “intelligent products: the grace experience,” control engineering practice, vol. 42, pp. 95-105, 2015. [8] p. pujo, n. broissin, and f. ounnar, “prosis: an isoarchic structure for hms control,” engineering applications of artificial intelligence, vol. 22, no. 7, pp. 1034-1045, 2009. [9] v. mařík, a. schirrmann, d. trentesaux, and p. vrba, “industrial applications of holonic and multi-agent systems,” 7th international conf. industrial applications of holonic and multi-agent systems , september 2015. [10] n. danišová, k. velšek, and p. košťál, “automated tool changing system in the intelligent manufacturing and assembly cell,” international symposium on computing, communication, and control, october 2009, pp. 1-8. [11] š. horváth, e. hrušková, and a. mudriková, “areas in flexible manufacturing -assembly cell,” annals of faculty engineering hunedoara-journal of engineering, vol. 6, no. 3, pp. 123-127, 2008. [12] p. leitão, “agent-based distributed manufacturing control: a state-of-the-art survey,” engineering applications of artificial intelligence, vol. 22, no. 7, pp. 979-991, 2009. [13] h. zhang, d. tang, t. huang, and c. xu, “an agent based intelligent distributed control paradigm for manufacturing systems,” international federation of automatic control-papers on line, vol. 49, no. 12, pp. 1549-1554, 2016. [14] j. wang, l. zhang, l. duan, and r. x. gao, “a new paradigm of cloud -based predictive maintenance for intelligent manufacturing,” journal of intelligent manufacturing, vol. 28, no. 5, pp. 1125-1137, 2017. https://goo.gl/7c4l2w  proceedings of engineering and technology innovation , vol. 1, 2016, pp. 15 18 15 cop y right © taeti the application of target analysis in electricity demand-side management fung-fei chen 1 , seng-cho chou 2, * , chiao-yi wang 2 , tai-ken lu 3 1 taiwan power company research institute, taipei, 100, taiwan 2 department of information management, national taiwan university, taipei, 100, taiwan 3 department of electrical engineering, national taiwan ocean university, keelung, 202, taiwan received 13 april 2015; received in revised form 11 may 2015; accept ed 18 june 2015 abstract recently, target analysis combined with database technology and data mining has been widely used in industries such as marketing, finance, insurance, teleco mmunicat ions, advertising, and e-commerce. because of the unique comple xit ies of user behavior in e lectric ity de mand, e xa mples of target analysis applications have yet to be seen. considering the industry’s urgent need to enhance the efficiency of electric ity demand -side management, this study aims to build a min ing analysis model fo r potential target users of interruptible load that both fully reflects consumer behavior characteristics and serves as a rule for static comparisons. the results of a data mining analysis of the taiwan power co mpany (taipowe r)’s interruptible loads 1 to 6 show that the number of potential target users is 1669, which is 21% o f the original mining population. additionally, the target users who were classified to have “the most potential” for all categories of interruptible load only accounted for 0.76% of the total mining population (= 59/7814), verifying the mining effects. ke ywor ds : ta rget analysis, data mining, association rule, e lectric ity de mand-side management 1. introduction in manage ment, target analysis has always been crucial to enterprise operations because it enables full marketing e xploitation o f the 80/20 rule and increases the opportun ities for p roduct recommendation and cross selling. another aspect of target analysis that has attracted significant a ttention and extensive discussions in recent years is its ability to identify valuable potential target users. target analysis coupled with database technology and data mining, which e merged fro m analysis of significant a mounts of data, is currently applied e xtensively in the fie lds of marketing, finance, insurance, manufacturing, and medica l care [2, 4, 13, 16]. by e mp loying target analysis data mining, ma rketers can obtain hidden knowledge or identify the most co mmerc ia lly valuable information fro m significant amounts of unsorted data. when used appropriately, target analysis can provide significant overall benefits for the organization. most previous studies employed decision tree analysis or data mining through artific ial neural networks to identify and e xtract the characteristics of potential target users, before establishing comparison rules to enable companies to e xa mine potential customers as ma rket ing targets [3, 5, 6, 8, 10, 12, 19]. the purpose of these investigations was to determine the consumer behavior characteristics of e xis ting customers, and then use these characteristics to identify hidden potential target customers. however, for industries where consumer behavior is influenced by industry policies or other e xte rnal factors, because existing customers do not exh ibit the sa me consumer behavior as potential customers, establishing a comparison rule between the e xisting customers and potential customers is d ifficu lt, resulting in a lack of research in this area. this phenomenon is particularly co mmon in teleco mmunicat ion and utility industries, where marketing measures may a dversely affect the consumer behaviors of e xisting users, decreasing the objectivity of the consumer behavio r characteristics identified using target analysis, further hinde ring comparisons [1, 14, 18]. to address these issues and compensate for the disadvantages of current target marketing analysis methods, we conduct a case study of taipower’s de mand-side management data mining ana lysis of potential target users. this study also constructs a min ing analysis model for potential target users, which both fu lly reflects consumer behavior characteristics and serves as a rule for static comparisons. we hope that the target analysis mining model proposed in this study can provide a list of the most valuable potential users of an interruptible load. the analysis model is a lso e xpected to provide a useful reference for identifying p otential target users for other purposes related to electricity demand, thereby facilitating the demand-side management of electricity. 2. experiment 2.1. research design taipower’s demand-side management measures include time use rates, seasonal rates, interruptible loads, and air conditioning control; a mong which, interruptible load is the most effective for reducing peak powe r loads. since the introduction of taipower’s interruptib le load policy in 1987, despite the seven interruptible load categories, by the end of 2005, the number of interruptible load users was only 670. co mpared to the target population of over 10,000 qualified users, the number of actual users is min imal, presenting significant scope for promot ion. therefore , target analysis and mining analysis of significant a mounts of user data must be employed to identify potential target users and expand the scope of promotion [9]. target analysis generally includes customer data co llection, data analysis, the e xtraction of segmentation variables, and the establishment of segmentation market characteristics [7]. without identical consumer behavior data and uncertainty whether basic variables a re sufficient for ma rket segmentation, the consumer behavior of e xisting customers can be used as the foundation for segme ntation to identify the basic variable characteristics of each ma rket segment. however, applying the e xisting target ma rket ing analysis method, which selects the differing basic variables of the segmentation market fo r characteristic analysis, may lead to losses of consumer beha vior-related data or defic iencies in the descriptive variab le dimensions, thereby preventing the effective convergence *corresponding aut hor. email address: chou@ntu.edu.t w proceedings of engineering and technology innovation , vol. 1, 2016, pp. 15 18 16 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti of the target marketing scope. to solve this proble m, we determine the behavioral data of existing customers who have changed because of promotional measures, identify the basic variables that are not subject to influence but are related, and eventually establish potential target customer rational co mparison rules as the target analysis method in this study. the target market ing analysis model based on a lack of identica l consumer behavior data is shown in fig. 1. fig. 1 the target ma rket ing model used when there is insufficient information about equivalent consumption behavior 2.2. data description this study prima rily e xa mines the data of low-voltage users who qualify for ta ipower’s interruptible load policies, categorizing them as either usage data or basic data. designed to record the various electric ity co nsumption behaviors of e xisting users who adopt interruptible load policies, data usage includes 79 variables written in n ume rica l form, e xc luding “electric ity no.” and “industry.” to ensure the quality of this research, we first o mit a number of the insignificant colu mns based on the exper ience of experts. then, to facilitate subsequent data analysis, we categorize the re ma ining 51 fie lds into five e lectricity consumption factors according to their features and relevance. (1) “frequent and peak” demand or reading group: all 16 variables show users ’ frequent electricity consumption behavior during peak hours. (2) “off-peak” demand or reading group: all 8 variables show users’ electricity consumption behavior during off-peak hours. (3) “saturday semi-peak” demand or reading group: all 8 variables show users’ electricity consumption behaviors during semi-peak hours. (4) “comparative” demand or reading group: all 11 variables show users ’ varying electricity consumption behaviors in different periods. (5) “others ” group: the remaining 8 variables. additionally, of the 38 basic data variables, including both classifying and analysis variables, a number of the variables that we re irre levant to electricity consumption behavior were o mitted based on expert e xpe rience. data of the 7 re ma ining variables were used to develop comparison rules for identifying the potential users who may accept interruptible load policies, as shown below. table 1 the relevant variables of users ’ basic data by e xa min ing the electric ity consumption behaviors of e xisting users with various interruptible load polic ies and segmenting their object ive effectiveness or determining the amount they can generate, we a im to identify the customers with the most potential, the second most potential, and the least potential a mong the users of various interruptible loads. this study also correlates the basic files of e xisting users and extracts the unique characteristics from the basic data of this beneficia l user type. a co mparison rule fo r identifying potential target users is then developed to narrow the target ma rket ing scope and best employ the limited time and cost resources . 2.3. data mining process because of the require ments of the analysis model and the characteristics of relevant data adopted in this study, we e mploy the follo wing techniques for data mining analysis: principal co mponent analysis, data clustering, correlation ana lysis, association rule and manova tests. thus, a data min ing analysis process is established, as shown in fig. 2. 2.4. results the results of all data mining stages are presented b elow. 2.4.1. analysis of data dimension reduction to determine the most suitable linear equations, this study adopted principal co mponent analysis. by condensing the variables into a representative indicator, the research comple xity and dimensions were reduced effe c tively, facilitating the smooth operation of subsequent research processes. the princ ipal co mponent analysis equation is as follows: 　　 　　　　　　　　　　 pppppp pp xaxaxay xaxaxay   ... ... 2211 12121111 xa xa p 1  (1) where y1 to yp a re representative principa l co mponent indicators e xtracted fro m a certa in factor dimension. de pending on the difference of internal data, each factor dimension may possess 1 top (the number of variab les in that dimension) representative indicators. a suitable critical threshold (cu mulat ive e xp lanatory variance) can be determined for the selection. fig. 2 data analysis process after repeated verification, we found that principa l component analys is of dimensions 4 and 5 does not provide good results and fails to converge dimensions effectively. to ensure the credibility, reliability , and exp lanatory power of the data analysis results, only the first repproceedings of engineering and technology innovation , vol. 1, 2016, pp. 15 18 17 17 resentative princ ipal co mponent indicator (pc1 is the first representative principal co mponent indicator of dimension 1, pc2 is the first representative principal co mponent indicator of dimension 2, etc) of the first three usage data dimensions mentioned in section 2.4 were used as segmentation variables for future data co llect ion, as shown in table 2. table 2 cu mu lative e xplanatory variance of represent ative principal component indicators 2.4.2. cluster analysis the most common ly used data clustering methods include hierarch ical clustering, non -hierarch ical clustering, and artificia l neural networks[15]. a mong them, the k means method provides superior clustering results and higher effic iency compa red to hiera rchica l c lustering and artific ial neura l network clus tering [11, 15, 17]. using a two-dimensional and three-dimensional scatter plot and chi-square plot, we verified that the usage data in this study has norma l distribution. after assessments and experimentation, the k means method was employed for cluster analysis. tables 3 and 4 show the cluster analysis results of the data of existing users. 2.4.3. manova test a manova test was conducted in this study to d etermine whether the clustering results can differentiate. using wilk’s la mbda, all impact factors (i.e., seg mentation variables) were examined to table 3 cluster analysis results -the number of existing users in each cluster table 4 cluster analysis results – the number of inter ruptible load users belonging to more than one cluster determine whether they had a significant influence on the overall segmentation results. if the like lihood of the wilk’s la mbda value be ing sma lle r than a certain critica l chi-square distribution value is minimal (<0.0001), we can conclude that the impact factor has substantial influence (i.e ., discriminatory ability), as shown in eq. (2). fro m a quantitative analysis perspective, if each impact factor has significant influence, the credib ility of the c luster analysis results can be verified, as shown in table 5. table 5 manova test results for interruptible load 1 　　　 　　　　　 ＝　 　　　 　　　 ＊ ＊ )(ln 2 )1(1 )1( ,arg 0one: )1(0: 2 )1( 0 1 i1 210    pg resfac res g gp ngb ifhrejectsampleselfor sspssp ssp let leastath effectsfactornoh                (2) 2.4.4. correlation analysis because of the changing use behavior of existing interruptible load users, basic data were used to analyze user characteristics. to ensure that the basic variables afte r correlation reflect both the electricity consumption b ehavior of all users and the results of cluster analysis, co rrelation analysis of the three analytical variables in table 2 should be performed and the segmentation variables e mployed for cluster analysis (table 6). su itable variables identified in the analysis can be used for further mining analysis. an exa mp le of the test results for interruptible load 1 is shown below. table 6 correlation analysis results for interruptible load 1 of the analytical variab les of interruptible load 1, cs_ capacit y and cs_up_capa cit y a re the most relevant to users ’ consumption behaviors; thus, they can be included as the descriptive variables of interruptible load 1. 2.4.5. association rule analysis the association rule was e mp loyed to establish data correlation ru les. after co mputations to determine whether it satisfies the threshold limit of minimal support and minimal confidence, the ru le is matched with the apriori algorithm to select appropriate correlation rules. data analysis results : an exa mple o f the representative characteristic behaviors and correlation mining threshold of users with “the most potential” for interruptible load 1 is shown below.  number of existing users: 6  minimal support; 6  minimal confidence: 0.8  importance: 0.2  essential association rule: for users with an “a” contract type, the like lihood that they are fro m the top 20 industry types = 11 (basic meta l industries) is > 0.8. for three-s tage users in category 2, the like lihood that they are fro m the top 20 industry types = 11 (basic meta l industries) is > 0.8. for users whose usage type = 5, the like lihood that they are three-stage users in category 2 is > 0.8. for users whose usage type = 5, the likelihood that their contract type = 5 is > 0.8. when the necessary correlation rules for all clusters are condensed into rigorous comparison rules, the number o f potential target users and e xisting users of every cluster can be identified, as shown in table 7. the total nu mber of potential target users shown in table 7 is 4,687. however, because a number of the m overlap between clusters, the actual number o f potential target users is 3,770. additionally, some of the clusters proceedings of engineering and technology innovation , vol. 1, 2016, pp. 15 18 18 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti have few e xisting users (no more than 10, and some as few as 1 or 2); thus, the comparison rules established with these clusters are relative ly loose and can result in an overestimat ion of potential target users. by contrast, potential target users mined fro m c lusters with more e xisting users yield superior convergence. the potential target users from these clusters (exc luding c lusters with only 1 or 2 e xisting users) totals 2,058; e xcluding overlapped users, the actual number o f potential target users is 1,669, which is 21% o f the original min ing population. the number of ta rget users with the most potential fro m a ll interruptib le load policies is 0.76% of the overall mining population (= 59/7814). table 7 correlat ion rule co mparison analysis results for all clusters 2.4.6. generating a list of potential target users using the corre lation rules e xt racted fro m c lusters, we compile a list of potential target users from a mong the customers not yet using interruptible load polic ies. table 8 shows the first five results for users with the most potential in interruptible load 1. table 8 interruptible load 1users with the most potential 3. conclusions by applying target analysis to electricity demand -side manage ment and mining ana lysis to a list of potential target users, this study develops a mining analysis model for potential target users of interruptible load that both fully reflects consumer behavior characteristics and serves as a rule for static comparisons . this model was designed to compensate for the deficiencies of the current target ma rket ing analysis model. the results show that the min ing analysis model proposed in this study can effectively narrow the target marketing scope to 21% of the overall mining population, with a condensing capability of 79% , which facilitates the segmentation and differentiation o f customers with various potential benefits and value. the proportion of target users with the most potential for a ll interruptible load policies is 0.76% of the overall mining population (= 59/ 7814). each step of the min ing analysis process underwent thorough quantitative (theoretical verification) and qualitative testing (industry knowledge and e xperience-based assessments ). the practical significance of this study is that it provides the electricity industry with informat ion of the most potential and valuable target users. the min ing analysis model proposed in this study can be emp loyed by relevant marketing decision ma kers to identify potential ta rget customers for other de mand -side manage ment policies and would have a lot to be referenced for utility in the future. references [1] a. al-ghandoor, et al. "residential p ast and future energy consump tion: p otential savings and environmental imp act," renewable and sustainable energy reviews, 2008. [2] e. bay am, j. liebowitz, and w. agresti, "older drivers and accid ents: a meta-analy sis and d ata min in g ap p lication on traffic accident data," exp ert sy stems with app lications, vol. 29, p p . 598-629, 2005. [3] j. z. bloom, m arket segmentation. "a neural network ap p lication," annals of tourism research, vo l. 32, p p . 93-111, 2005. [4] r. j. brachman, t. khabaza, w. kloesgen, g. piatetsky -shap iro, and e. simoudis, "m ining business databases," communication of the acm , vol. 39, p p . 42-48, 1996. [5] s. w. ch an gchien, and t. c. lu, "m inin g association rules p rocedure to supp ort on-line recommendation by customers and p roducts fragmentation," exp ert sy stems with app lications, vol. 20, p p . 325-335, 2001. [6] s. daskalaki, i. kop anas, m . goudara, and n. avouris, "data minin g for decision supp ort on customer insolvency in telecommunications business," europ ean journal of op erational research, vol. 145, p p . 239-255, 2002. [7] s. dibb, and p. stern, "questioning the reliability of market segmentation techniques," omega international journal of m anagement science, vol. 23, p p . 625-636, 1995. [8] s. h. ha, and s. c. park, "app lication of data mining tools to hotel data mart on the intranet for database marketin g," exp ert sy stems with ap p lications, vol. 15, p p . 1-31, 1998. [9] http ://www.taip ower.com.tw, 2012. [10] s. y. hun g, d. c. yen, and h. y. wan g, "app ly ing data minin g to teleco m churn management," exp ert sy stems with ap p lications, vol. 31, p p . 515-524, 2006. [11] a. k. jain, m . n. m urty , and p. j. fly nn, "data clustering: a review," acm comp uting survey s, vol. 31, pp . 264-323, 1999. [12] t. s. lee, c. c. ch iu, y. c. chou, and c. j. lu, "m ining the customer credit usin g classification and regr ession tree and multivariate adap tive regression sp lines," comp utational statistics and data analy sis, vol. 50, p p . 1113-1130, 2006. [13] d. r. liu, and y. y. shih, "integr ating ahp and data min in g for p roduct recommendation based on customer lifetime value," information & m anagement, vol. 42, p p . 387-400, 2005. [14] s. roberts, "demograp hics, ener gy and our homes," energy policy , vol. 36, p p . 4630-4632, 2008. [15] s. s. shahap urkar, and m . k. sundareshan, "comp arison of self-organizin g map with k-means hierarch ical clustering for bioinformatics ap p lications," neural networks, proceedin gs. 2004 ieee international joing conf erence on, pp . 1221-1226, 2004. [16] w. e. sp angler, m . gal-or, and j. h. m ay , "using data minin g to p rofile tv viewers," communication of the acm, vol. 46, p p . 67-72, 2004. [17] a. ultsch, "self-or ganizin g neur al networks p erform different from statistical k-means clusterin g," proc. conf. soc. for information and classification, basel, 1995. [18] w. o. onuh, a. t. valerio, a. permalino jr., "residential demand for electricity dasmarinas, cavite, p hilipp ines," journal of glob al busin ess & economics, vol. 2, p p . 1-22, 2011. [19] c. h. wu, s. c. kao, y. y. su, and c. c. wu, "tar getin g customers via discovery knowledge for the insurance industry ," exp ert systems with app lications, vol. 29, pp . 291-299, 2005. http://www.taipower.com.tw/  proceedings of engineering and technology innovation, vol. 1, 2015, pp. 05 09 5 innovation management in services industries: the impacts of innovative capacity and transformative learning wen-hai chih 1 , ling-chu huang 1 , tsung-ju yang 2,* 1 department of business administration, national dong hwa university, hualien, taiwan, roc. 2 institute of service industries and management, minghsin university of science and technology, hsinchu, taiwan, roc. received 18 june 2015; received in revised form 11 july 2015; accepted 28 july 2015 abstract this study highlights the innovation management and learning perspective of a firm’s innovative capability. the model proposed in this study examines the relationships among competence exploitation, competence exploration, transformative learning, innovation, and performance. this study presents empirical results from 225 service industry in taiwan. first, the innovative capability factors positively affect transformative learning include competence exploitation and competence exploration. transformative learning in turn has positive effects both on innovation and performance. secondly, the rank order effects on innovation are competence exploitation, competence exploration, and transformative learning, respectively. the rank order effects on performance are competence exploration, competence exploitation, and transformative learning, respectively. finally, transformative learning is the mediating effect of competence exploitation and competence exploration on innovation and performance. keywords: competence exploitation, competence exploration, transformative learning, innovation, performance 1. introduction as the innovation management is associated with learning capacity. innovation is embedded in marketor learning-oriented firms. an innovative firm culture can guide an organization to develop new abilities (hurley & hult, 1998). ability to learn quickly and high internalization of knowledge are key competitive advantages in the firms (lee, liang, & liu, 2010). as a result, managing knowledge transfer has become a challenge for inter-firm (zahra & george, 2002). recently, firms have faced a strategic dilemma in developing new customized products or services. atuahene-gima (2005) showed that exploiting existing competencies may provide short-term benefits, but ultimately becomes a hindrance to the firm’s long-term viability by stifling the exploration of new competencies (levinthal & march, 1993). for example, many companies seek to develop their existing capabilities but hesitate to develop new abilities (o’reilly & tushman, 2004). this capability-rigidity paradox indicates that the exploitation of competence tends to crowd out the exploration of competence (leonard-barton, 1992). in general, learning-oriented firms are more likely to exhibit this element of firm culture (hult & ketchen, 2001). the rbv indicates that learning development and innovation is a distinctive capacity of organization (day, 1994). the innovation characteristics contribute to the creation of greater firm value and achieving better organizational performance (hurley & hult, 1998). however, data on learning-oriented firms suggests that innovation management is more complex than previously depicted (lichtenthaler, 2009). further investigation is required to compensate for a lack of previous research. previous studies have paid scant attention to services or how the mediating role of organizational learning influences innovation and performance. the purpose of this study was to discuss the relationships among competence exploitation, competence exploration, transformative learning, innovation and performance. innovative capability includes competence exploitation and competence exploration. the service innovation of taiwan was used as the background, to ensure compliance with innovation management approaches (hult & ketchen, 2001). secondly, data was collected on benchmark enterprises in service, to gain an understanding of how innovative capability drives transformative learning through innovation and performance. and explored in-depth the high-level learning process (slater & narver, 1995). thirdly, the integration of learning-oriented and innovation was supported by key references (hurley & hult, 1998). this study also believe that the absorptive capacity as the process-based using the service innovation (lichtenthaler, 2009). referring the viewpoint of transformative learning proposed by slater and narver (1995), the dual theory is applied to the study of innovation and performance 2. method 2.1. innovative capability firms have faced a strategic dilemma in developing new customized products or services. many companies seek to develop their existing capabilities but hesitate to develop new abilities (o’reilly & tushman, 2004). this type of innovative capability is related to the absorptive capacity proposed by cohen and levinthal (1990) (hurley & hult, 1998). absorptive capacity has a significantly positive influence on the internalization of knowledge (lee et al., 2010). absorptive capacity is the ability of a firm to assimilate and apply innovative information or elements (lichtenthaler, 2009). an innovative corporate culture promotes the combination of organizational resources and the internalization of knowledge to cultivate even greater innovative capacity. companies with greater innovative capacity are better able to develop and maintain competitive advantage, and achieve stronger organizational performance (day, 1994). based on the above, firms *corresponding author, email: morgan4117@gmail.com proceedings of engineering and technology innovation, vol. 1, 2015, pp. 05 09 6 copyright © taeti tend to enhance the two capabilities to increase the chances of simultaneous success for competence exploitation and competence exploration (atuahene-gima, 2005). 2.2. transformative learning organizational learning is vital to corporate success (day, 1994; hult & ketchen, 2001). maintaining learning capacity stronger than that of competitors is vital to maintaining competitive advantage (dickson, 1992). continual learning drives behavioral and operational changes, which in turn leads to improve firm performance (sinkula, 1994). a successful learning organization must be equipped with a suitable organizational structure and a high-level learning processes (argyris, 1977), in order to promote the development of knowledge. the development of new knowledge or insight during the process of organizational learning can potentially influence behavior (hurley & hult, 1998) and contribute to improve firm performance (slater & narver, 1995). based on the above, this study posited that transformative learning plays a mediating role in processing the flow of innovation information. 2.3. innovation innovativeness is embedded in marketor learning-oriented firms. the cultures of these firms are more exploratory, discovering the expressed and latent needs of customers (slater & narver, 1999). the deepest manifestations of market and learning orientations appear at the cultural level (schein, 1985). a firm with an innovative culture adopts or implements new ideas, products, and processes through management innovation. the goal of this approach is to achieve high performance for the firm (slater & narver, 1995). based on the above, this study defines innovation as a firm’s receptivity to new ideas and innovation as part of organizational competitive advantage (hurley & hult, 1998) 2.4. competence exploitation, competence exploration, transformative learning and innovation focus on the service industry, firms are increasingly relying on external knowledge to foster innovation and enhance their performance. hurley and hult’s (1998) organization and market driven innovation model points out the innovative capacity to competitive advantage and performance. innovative capacity is related to absorptive capacity (cohen & levinthal, 1990; lichtenthaler, 2009). due to the managerial challenges of inter-firm knowledge transfer, absorptive capacity is a major source of competitive advantage (zahra & george, 2002). such process-based capacity is a firm’s ability to utilize external knowledge through the sequential processes of exploitative, exploratory, and transformative learning (lane, koka, & pathak, 2006). therefore, competence exploitation involves the application of external knowledge through realized absorptive capacity. competence exploration is the acquisition of external knowledge through potential absorptive capacity (zahra & george, 2002). transformative learning links these two competences to maintaining knowledge over time (garud & nayyar, 1994). a process-based firm is better able to engage in innovation and achieve higher performance (day, 1994). based on the above, this study proposes hypotheses 1, 2, and 3: h1: competence exploitation has a positive effect on innovation. h2: competence exploration has a positive effect on innovation. h3: transformative learning has a positive effect on innovation. 2.5. competence exploitation, competence exploration, transformative learning and performance atuahene-gima (2005) stressed that innovative capacity is a measure of organizational capability and competitiveness. firms face a strategic dilemma in the development of customized new products or services. exploiting competence may provide short-term success, but it can become a hindrance to the firm’s long-term viability by stifling the exploration of new competencies (levinthal & march, 1993). the number of innovation organization is able to successfully adopt or implement can measure the definition of innovative capacity. the degree to which the culture within a firm is open to innovation, combined with its resources and other organizational characteristics, increases the capacity for innovation. firms develop a greater capacity for innovation, develop a competitive advantage, and can achieve higher levels of performance (day, 1994). based on the above, this study proposes hypotheses 4, 5, and 6: h4: competence exploitation has a positive effect on performance. h5: competence exploration has a positive effect on performance. h6: transformative learning has a positive effect on performance. 2.6. antecedents of transformative learning according to the capacity-based resource advantage is scarce, relatively non-transferrable, and difficult for competitors to understand or imitate (reed & defillippi, 1990). in cultivating innovative capacity, knowledge has become a management challenge for inter-firm, and rapid learning capacity is a key competitive advantage (zahra & george, 2002). innovative capacity of a firm is related to its learning ability (hurley & hult, 1998), strong firms tend to apply external knowledge to conduct processes of exploitation, exploration, and transformative learning (lane et al., 2006). the variability generated by the continuous learning and transformation process puts pressure on the organization and development (levitt, 1980). based on the above, this study proposes hypotheses 7 and 8: h7: competence exploitation has a positive effect on transformative learning. h8: competence exploration has a positive effect on transformative learning. 3. methods after the atlas of new designs is obtained, a detailed design can be carried out by selecting one from the atlas. 3.1. sample, pretest and data collection this study collected a sample from the china credit information service limited. (ccis) published by the top 5000 largest corporations in taiwan, which selects the top 2000 service. this study, according to lichtenthaler proceedings of engineering and technology innovation, vol. 1, 2015, pp. 05 09 7 copyright © taeti (2009), proposes absorptive capacity as the process-based with a focus on specific topics using the service innovation. general services finance, and information firms are typical representative of the service industry. this study mailed questionnaires to firms. to increase the return rate, the following steps were taken: 1) according the latest corporation directory published in 2010 by ccis, a postage-paid return envelope was included the questionnaire directly mailed to the general managers; 2) return letter to advertising; 3) research institutions, researchers and contacts were listed; 4) released in august to avoid releasing peak. before mailing the questionnaires, this study used convenient sampling to select 60 service firm managers and 60 emba students. a pretest was conducted in one month. there were 118 valid samples. the results of the reliability analysis of cronbach’s α were all higher than the standard value of 0.7 recommended by nunnally (1978) for each dimension initiated a large-scale release. 3.2. measures the respondents of this study consisted of general managers for each firm. except for the demographics (age, capital, employee, turnover, and listed/otc company), the questionnaire used a likert 7-point scale for the survey, with 1 indicating “strongly disagree” and 7 indicating “strongly agree”. this study operates as a multidimensional described as the following: the definition of innovative capacity is the organizations invest in service innovation and the pursuit of knowledge, skills, and processes, which converted to core competencies. two dimensions based on atuahene-gima (2005) included competence exploitation (five items) and competence exploration (five items). the definition of transformative learning is the firm with innovative capacity uses organizational learning to connect with the outside-in process. the items based on lichtenthaler (2009) included one dimension is transformative learning (eight items). the definition of innovation is the culture of a firm with innovativeness to implement new ideas, product, or processes successfully. the items based on hurley and hult (1998) included one dimension is innovation (four items). a firm evaluates the organizational performance by using subjective performance indicators to analyze firm-level performance. the items based on kirca, jayachandran, and bearden (2005) included one dimension is performance (three items). 4. generalization and number synthesis 4.1. sample, pretest and data collection this study mailed questionnaires to the top 2000 service in taiwan. 241 of these firms replied (12.0%) and after removing the invalid ones, a total of 225 valid questionnaires (11.2%) remained. non-response bias test was applied (armstrong & overton, 1977). the following is the basic data of this study. more than half of the service firms in the sample have been in operation for more than 21 years (52.4%), and have capital of under 500 million ntd (56.0%), turnover under 5 billion ntd (79.1%), less than 500 employees (76.9%). a smaller percentage was listed/otc companies (23.5%). these results reflect the status of the service in taiwan. the cronbach’s α of the competence exploitation, competence exploration, transformative learning, innovation, and performance are 0.797, 0.912, 0.916, 0.849, and 0.735, respectively, which were all higher than the standard of 0.7 suggested by nunnally (1978). this indicates that the internal consistency of measuring each dimension is good (table 1). table 1 the reliability of the variables examined variables items mean std. α cr ave reference cei 5 5.8 3.33 0.797 0.892 0.652 atuahene-gima (2005) cer 5 5.8 3.38 0.912 0.932 0.732 tl 8 5.8 5.41 0.916 0.938 0.655 lichtenthaler (2009) in 4 5.8 3.19 0.849 0.867 0.627 hurley & hult (1998) pm 3 5.7 2.37 0.753 0.785 0.552 kirca et al. (2005) note: cei=competence exploitation; cer=competence exploration; tl=transformative learning; in=innovation; pm=performance. std. =standard deviation; α=cronbach’s α; cr=composite reliability; ave=average variance extracted. table 2 matrix of the related coefficients variables 1 2 3 4 5 cei (0.807) cer 0.805* (0.856) tl 0.733* 0.744* (0.809) n 0.705* 0.684* 0.697* (0.792) pm 0.557* 0.541* 0.552* 0.435* (0.743) note: number in brackets is ave square values; *p<0.001 the measurement of the validity in this study refers to the development of literature for theoretical basis. the results indicated that the factor loadings of all items were significant, with composite reliability (cr) between 0.785~0.938, which was higher than 0.7, and average variance extracted (ave) between 0.552~0.732, which was higher than 0.5. this study showed convergent validity of measurement items (fornell & larcker, 1981). the variables were referenced with a number of indicators to measure fitness. the model showed good convergent validity. the discriminant validity is the measurement of the discriminant extent among dimensions suggested by fornell and larcker (1981). the result showed that the correlation between any two dimensions in this study was less than the ave square of each dimension, which means that there is discriminant validity among these dimensions (table 2). 4.2. hypotheses testing the structural equation modeling (sem) was conducted using maximum likelihood estimation (mle). the goodness of fit index of the whole model of this study is χ2=756.508, d.f.=260, χ2/d.f.=2.910, gfi=0.806, agfi=0.758, cfi=0.847, rmsr=0.082, rmsea=0.092, proceedings of engineering and technology innovation, vol. 1, 2015, pp. 05 09 8 copyright © taeti nfi=0.822，pnfi=0.712. a number of goodness-of-fit indexes fit the acceptable standard, which meant that the model fit was good (fig. 1 & table 3). fig. 1 the research model an empirical study of the service innovation in taiwan found that the effects of competence exploitation, competence exploration, and transformative learning were significant for innovation. the path coefficients were 0.320 (p<0.001), 0.294 (p<0.001), and 0.392 (p<0.001), respectively. hypotheses 1, 2, and 3 of this study were supported. next, the effects of competence exploitation, competence exploration, and transformative learning were significant for performance. the path coefficients were 0.268 (p<0.005), 0.275 (p<0.004), and 0.273 (p<0.016), respectively. hypotheses 4, 5, and 6 of this study were supported, respectively. finally, the effects of competence exploitation and competence exploration were significant for transformative learning. the path coefficient was 0.463 (p<0.001) and 0.511 (p<0.001). therefore, hypothesis 7 and 8was also supported. table 3 results of hypothesis testing hypotheses relationship path p-value results h1 cei → in 0.320 0.001 supported h2 cer → in 0.294 0.001 supported h3 tl → in 0.392 0.001 supported h4 cei → pm 0.268 0.005 supported h5 cer → pm 0.275 0.004 supported h6 tl → pm 0.273 0.016 supported h7 cei → tl 0.463 0.001 supported h8 cer → tl 0.511 0.001 supported table 4 impacts of variables variables direct impact indirect impact total impact in pm in pm in pm cei 0.320 0.268 0.181 0.126 0.501 0.394 cer 0.294 0.275 0.200 0.140 0.494 0.415 tl 0.392 0.273 0.392 0.273 note: cei=competence exploitation; cer=competence exploration; tl=transformative learning; in=innovation; pm=performance. this study analyzes the effects of each antecedent on innovation and performance (table 4), including direct and indirect effects. among all of the antecedents of innovation in the service, the total effect of competence exploitation has the greatest effect (0.501) on innovation, including direct effect (0.320) and indirect effect (0.181), respectively. next, the total effect of competence exploration has the greatest effect (0.415) on performance, including direct effect (0.275) and indirect effect (0.140), respectively. the results show that competence exploitation and competence exploration have the greatest effect on service innovation in the service. 5. conclusions and discussion this study gathered data on service in taiwan to understand the effects of the service innovation, and conducted theoretical and practical discussion of research results. first, the outcome of the analysis described above demonstrates that competence exploitation, competence exploration, and transformative learning each has significant and positive effect both on innovation and performance. this result corresponds to the concept of “absorptive capacity” advocated by lichtenthaler (2009). apart from proving that innovative capacity is an important factor in the development of innovation and adaptation to varying business environments, this result also reveals that transformative learning has significant explanatory power with regard to organizational performance. this study also shows that competence exploitation and competence exploration has significant and positive effect on transformative learning, respectively. this complies with the conceptual model proposed by hurley and hult (1998) and matches some aspects of the organizational learning mechanism proposed by bell, whitwell, and lukas (2002). secondly, many studies on innovation management promote the viewpoint that learning orientation positively contributes to firm performance. this study found that with regard to service, transformative learning plays an important role both in innovation and performance. the service industry has implicit subsequent costs (anderson, fornell, & rust, 1997), which indicates that apart from promoting the development of knowledge as a method of influencing behavior, firms should also emphasize innovative development and explorative capacity. transformative learning has direct effect both on innovation and performance; these results echo the concept of “absorptive capacity”, which is a point of particular emphasis in the service industry. thirdly, most services are intangible, heterogeneous, and inseparable (parasuraman, zeithaml, & berry, 1985). services need a flexible organizational learning to avoid the capability-rigidity. this study verifies the service innovation in taiwan show that transformative learning has significant and positive effect both on innovation and performance in service industries. although previous studies have mainly emphasized manufacturing industries, this study showed that elements of innovative capacity (including competence exploitation and competence exploration) are antecedents of transformative learning. the effect of competence exploitation on innovation is the largest, and the effect of competence exploration on performance is the largest. transformative learning plays an important mediating role in enhancing both innovation and performance. competence exploitation performance 0.392*** 0.463*** 0.511*** transformative learning competence exploration 0.268** 0.294*** 0.275** 0.320*** 0.273* innovation note: * p<0.05, ** p<0.01, *** p<0.001 proceedings of engineering and technology innovation, vol. 1, 2015, pp. 05 09 9 copyright © taeti 5.1. limitations and future research the results of this study may have the following bias. first, this paper conducted an empirical study of the service industries in taiwan, by mailing questionnaires to conduct surveys of firms. due to limited time, budget, and limited number of replies, the study results may not be generalizable for use in other industries or countries. future studies could analyze other industries or countries to make the results more generalization. second, the relationship of both competence exploitation and competence exploration on performance may differ in economic or competitive environments in service firms, and this study did not consider these environmental factors. future studies could also integrate political, economic, legal, and industrial factors. third, future empirical studies could examine the relative contribution of innovative capacity on the firm’s competitive advantage. factors such as innovativeness, capacity to innovate, learning orientation, and market orientation are all known to affect a firm’s competitive advantage (hult & ketchen, 2001; hurley & hult, 1998). fourth, transformative learning is an important mediator for achieving innovation and performance on service firm. future studies can continue to develop relations with the high-level learning process. fifth, this study was a cross-sectional research and future studies can use a longitudinal method to observe the long-term relationships among variables. references [1] a. parasuraman, v. a. zeithaml, and l. l. berry, “a conceptual model of service quality and its implications for future research,” journal of marketing, vol. 49, no. 4, pp. 41-50, 1985. [2] a. h. kirca, s. jayachandran, and w. o. bearden, “market orientation: a meta-analytic review and assessment of its antecedents and impact on performance,” journal of marketing, vol. 69, no. 2, pp. 24-41, 2005. [3] c. argyris, “double loop learning in organizations,” harvard business review, vol. 55, no. 5, pp. 115-25, 1977. [4] c. fornell and d. f. larcker, “evaluating structural equation models with unobservable variables and measurement error,” journal of marketing research, vol. 18, no. 1, pp. 39-50, 1981. [5] c. iii o’reilly and m. l. tushman, “the ambidextrous organization,” harvard business review, vol. 82, no. 4, pp. 74-81, 2004. [6] d. leonard-barton, “core capabilities and core rigidities: a paradox in managing new product development,” strategic management journal, vol. 13, special issue, pp. 111-125, 1992. [7] d. a. levinthal and j. g. march, “the myopia of learning,” strategic management journal, vol. 14, special issue, pp. 95-112, 1993. [8] e. h. schein, “organizational culture and leadership,” san francisco, 1985. [9] e. w. anderson, c. fornell, and r. t. rust, “customer satisfaction, productivity, and profitability: differences between goods and services,” marketing science, vol. 16, no. 2, pp. 129-145, 1997. [10] g. s. day, “the capabilities of market-driven organizations,” journal of marketing, vol. 58, no. 4, pp. 37-52, 1994. [11] g. t. m. hult and d. j. ketchen, “does market orientation matter? a test of the relationship between positional advantage and performance,” strategic management journal, vol. 22, no. 9, pp. 899-906, 2001. [12] j. c. nunnally, “psychometric theory,” new york, 1978. [13] j. m. sinkula, “market information processing and organizational learning,” journal of marketing, vol. 58, no. 1, pp. 35-45, 1994. [14] j. s. armstrong and t. s. overton, “estimating non-response bias in mail surveys,” journal of marketing research, vol. 14, no. 3, pp. 396-402, 1977. [15] k. atuahene-gima, “resolving the capability: rigidity paradox in new product innovation,” journal of marketing, vol. 69, no. 4, pp. 61-83, 2005. [16] p. j. lane, b. r. koka, and s. pathak, “the reification of absorptive capacity: a critical review and rejuvenation of the construct,” academy of management review, vol. 31, no. 4, pp. 833-863, 2006. [17] p. r. dickson, “toward a general theory of competitive rationality,” journal of marketing, vol. 56, no. 1, pp. 69-83, 1992. [18] r. garud and p. r. nayyar, “transformative capacity: continual structuring by intertemporal technology transfer,” strategic management journal, vol. 15, no. 5, pp. 365-385, 1994. [19] r. reed and r. defillippi, “causal ambiguity, barriers to imitation, and sustainable competitive advantage,” academy of management review, vol. 15, no. 1, pp. 88-102, 1990. [20] r. f. hurley and g. t. m. hult, “innovation, market orientation, and organizational learning: an integration and empirical examination,” journal of marketing, vol. 62, no. 3, pp. 42-54, 1998. [21] s. a. zahra and g. george, “absorptive capacity: a review, reconceptualization, and extension,” academy of management, vol. 27, no. 2, pp. 185-203, 2002. [22] s. c. lee, h. liang, and c. y. liu, “the effects of absorptive capacity, knowledge sourcing strategy and alliance forms on firm performance,” service industries journal, vol. 30, no. 14, pp. 2421-2440, 2010. [23] s. f. slater and j. c. narver, “market-oriented is more than being customer-led,” strategic management journal, vol. 20, no. 12, pp. 1165-1168, 1999. [24] s. f. slater and j. c. narver, “market orientation and the learning organization,” journal of marketing, vol. 59, no. 3, pp. 63-74, 1995. [25] s. j. bell, g. j. whitwell, and b. a. lukas, “schools of thought in organizational learning,” journal of the academy of marketing science, vol. 30, no. 1, pp. 70-86, 2002. [26] t. levitt, “marketing success through differentiation of anything,” harvard business review, vol. 58, no. 1, pp. 83-91, 1980. [27] u. lichtenthaler, “absorptive capacity, environmental turbulence, and the complementarity of organizational learning processes,” academy of management journal, vol. 52, no. 4, pp. 822-846, 2009. [28] w. m. cohen and d. a. levinthal, “absorptive capacity: a new perspective on learning and innovation,” administrative science quarterly, vol. 35, no. 1, pp. 128-152, 1990.  proceedings of engineering and technology innovation, vol. 9, 2018, pp. 25 31 message-efficient route planning based on comprehensive real-time traffic map in vanets chi-fu huang*, yi-hong chen department of computer science and information engineering, national chung-cheng university, chia-yi, taiwan, roc received 02 march 2018; received in revised form 05 april 2018; accepted 06 may 2018 abstract traffic congestion has become one of the major society issues in many urban areas around the world. it greatly increases the commuting time and fuel consumption. researchers have proposed many solutions in different aspects to alleviate this issue. in recent years, many works adopt vehicular ad hoc networks (vanets) to relieve the problem. vehicles with communication ability can receive traffic information from infrastructures or other vehicles. furthermore, drivers can select its driving path by avoiding congestion areas based on the real-time information. however, to speed up the procedures of information collections, many existing works suggest proactive messages to acquire traffic information from other vehicles, which incurs heavy communication overheads. in this work, we propose a real-time traffic collection mechanism, which adopt both proactive and passive schemes. vehicles passively collect traffic information in most of the time, and proactively broadcast the traffic data only in certain situations in order to speed up urgent information spreading. thus, vehicles can obtain real-time traffic information in a low-cost way. in addition, we propose a routing planning scheme, which considers travel time, reliability, expected traffic flow as well as other factors, to select reliable and fast routes for vehicles. the simulation results show that, comparing with previous works, the proposed schemes can find better driving paths for vehicles with lower communication cost. keywords: vanets, route planning, traffic congestion 1. introduction nowadays, traffic congestions have become serious issues in many urban areas. with the increasing number of vehicles, especially in rush hours of big cities, traffic congestion greatly increases the commuting time, fuel consumption, and air pollution. based on the estimate, the total cost of traffic congestion in the usa is roughly $160 billion annually, according to the urban mobility report [1] by the texas transportation institute (tti). therefore, there are many efforts focused on how to alleviate or avoid traffic congestion. in modern cities, sensors, such as cameras and/ or induction loop detectors, are deployed on the roads and the intersections to collect traffic information. by counting the number of vehicles passing through, the vehicular density and traffic condition of a certain road segment can be aware. this information is helpful to distribute traffic flower in a city. however, due to the deployment cost, in most cases only the main roads of a city can install sensors. as a result, only parts of traffic conditions of a city can be aware. it may cause secondary congestions in other neighboring roads. due to advances in wireless communication technologies, vehicles on the roads can assist each other in message exchange and form an interconnecting network, namely vehicular ad hoc networks (vanets) [2]. in a vanet, each vehicle is able to * corresponding author. e-mail address: cfhuang@cs.ccu.edu.tw tel.: +886-5-2720411; fax: +886-5-2720859 proceedings of engineering and technology innovation, vol. 9, 2018, pp. 25 31 copyright © taeti 26 communicate with nearby vehicles and share information through the network. with assistance of the appropriate applications, drivers can receive warning messages in front of an accident site and thus improve driving safety. moreover, collection and sharing of real-time traffic information based on vanets can also be useful to avoid traffic congestions. many other applications based on vanets can be found [3]. in recent years, there are many studies aimed to plan driving paths for vehicles to avoid congestion areas. the basic concept is to use the traffic information collected from road segments to calculate their estimated travel times. then minimum cost algorithm is utilized to derive the fastest rout for each vehicle. obviously, the immediacy of the traffic information affects the derived results of the planned paths a lot. in vanets, fast speeds of vehicles result rapid change of the network topology. outdated traffic information is useless for driving path planning since the road conditions may be greatly different. therefore, how to obtain real-time traffic information more efficiently is critical for vehicular rout planning. communications in vanets can be classified into two categories [4-5]: vehicle-to-infrastructure (v2i), and vehicle-to-vehicle (v2v), as shown in fig.1 and fig. 2. 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. vehicles can send and receive road conditions through rsus, and obtain wide range real-time traffic information. however, the drawback of v2i communication is that it needs huge deployment and maintenance cost. on the other hand, in v2v communication, vehicles form a self-organized network without any centralized system and fixed infrastructure. vehicles communicate with each other directly and achieve long-distance communications through multi-hop relays. due to the lower deployment cost, v2v communication has become a major trend of researching vanets. fig. 1 vehicle-to-infrastructure communications. fig. 2 vehicle-to-vehicle (v2v) communications. the existing researches using inter-vehicle communications to exchange traffic information can be divided into two categories: proactive [6] and passive [7]. in the proactive schemes, whenever traffic information regarding to a certain road segment is outdated, a request message is initiated towards the target road segment. the vehicle who owns the fresher newest information of the target road will reply the request. otherwise, the request will reach the road segment and ask a nearby vehicle to collect the last information. in the passive scheme, vehicles overhear periodical beacon messages from adjacent vehicle to estimate traffic condition of the road segment. vehicles exchange their own records with other vehicles when passing through intersections. information is then gradually distributed to the whole network. although the proactive schemes can achieve faster information distribution than the passive schemes, they suffer from the need of additional request messages, resulting in a larger packet overhead. the amount of messages of the passive schemes is relatively small since information exchanges only take place at intersections. however, the information obtained by passive schemes may lack of timeliness and only cover a limited range. in this paper, in order to collect traffic information efficiently with low cost, a hybrid scheme is proposed. moreover, we use the collected traffic information to avoid the congestion road and select routes with lower travel time for vehicle. the rest of this paper is organized as follows. in section 2, we describe the proposed solution. section 4 3 demonstrates the efficiency of the proposed schemes through simulation studies. section 6 4 discusses future works and concludes this paper. proceedings of engineering and technology innovation, vol. 9, 2018, pp. 25 31 copyright © taeti 27 2. the proposed solution the schemes proposed in this paper can be divided into two parts: (1) traffic information collection and (2) vehicle route planning. 2.1. traffic information collection in the traffic information collection, we adopt the v2v passive scheme, namely the comprehensive real-time traffic map (crt map) [7], to collect and establish a wide range of real-time traffic information map on each vehicle. crt map is established based on the concept of the crowd sensing [8]. vehicles driving on the roads collect the real-time traffic flow and share with each other. each vehicle keeps recording the traffic conditions of traveled roads into a designed digital map. then, the digital map would be exchanged with other vehicles when traveling to intersections. moreover, in this paper we improve the timeliness and the coverage range of traffic information. when encountering special situations, such as traffic jams or car accidents, vehicles proactively broadcast the road information to speed up the information distribution. each vehicle in the v2v network equip with a obu and periodically broadcast beacon messages, including the id, to adjacent vehicles. therefore, vehicles can observe how many and which vehicles on the same road and calculate the current traffic flow of this road. in order to get the accurate real-time traffic flow, a vehicle enters a road could ask the opposite lane of the vehicle about the traffic flow of the same direction. when the vehicle obtains new traffic information at the intersection, we can integrate the road information from different vehicles and update the traffic map in the vehicle. since periodically broadcasting beacon messages is a fundamental operation in the general v2v environment, we use this feature to calculate the traffic flow and send no additional packet, thereby reducing the packet overheads in the process of collecting traffic information. when vehicle travels on the road, it could estimate the current traffic flow and integrate with the existing data by the following formula (1), (2). 𝑊𝑖𝑗 𝑛 = { 1 1+∆𝑇𝑖𝑗 𝑛 , ∆𝑇𝑖𝑗 𝑛 ≤ 𝑇𝑇𝐻 1 1+𝑇𝑇𝐻 , ∆𝑇𝑖𝑗 𝑛 > 𝑇𝑇𝐻 (1) 𝑁′𝑖𝑗 = 𝑁𝑖𝑗 𝑛 = 𝑁𝑖𝑗 𝑘 = (𝑁𝑖𝑗 𝑛 ×𝑊𝑖𝑗 𝑛 )+(𝑁𝑖𝑗 𝑘 ×𝑊𝑖𝑗 𝑘 ) (𝑊𝑖𝑗 𝑛+𝑊𝑖𝑗 𝑘) (2) in order to spread out the traffic information, vehicles exchange their traffic map by inter-vehicle communication when they meet at the intersection. when vehicle vi arrives at an intersection, it first observes if there is an exchange process taking place or not. if so, vi joins this existing process and sends its traffic map to the leading vehicle. otherwise, vi initiates a new exchange process and broadcasts the request to collect neighbors’ traffic maps. then, vi sets a timer for receiving traffic map. after the neighbor vehicles receive the request, they transmit traffic map to vi by the random back off mechanism to avoid the collisions. when the timer runs out of time, vi uses the formula (1) and (2) to integrate traffic maps. then later vi disseminates the integrated traffic map to all neighbor vehicles, as indicated in fig. 3. if the neighbor vehicles have already been involved in the exchange process, they will update the traffic map after receiving the integrated map. on the other hand, if vehicles arrive at the intersection after the collection runs out of time, they can also receive the new traffic map. these vehicles integrate the new traffic map with the original map by themselves. the basic information exchange method described above allows the vehicle not only to collect traffic data by themselves, but also to exchange the information of other roads with the neighboring vehicles at the intersection. proceedings of engineering and technology innovation, vol. 9, 2018, pp. 25 31 copyright © taeti 28 fig. 3 information exchange process at an intersection in the above method, the exchange of information only performs at the intersections, thus it is less efficient to obtain traffic information in the following condition: (1) when the traffic jam occurs, the average speed on the road is reduced, so that the vehicle will spend more time traveling to the intersection. thus, the information spreading speed of congestion roads is also reduced. (2) if some records in the map are not updated for a long time, it reduces the effectiveness of the path planning. in particular, in the roads with low vehicular density, the information diffusion rate is relatively slow. however, these roads should be more preferred because lower vehicular density means higher driving speed. therefore, when the information of a road is outdated, we tend to speed up the access to its information. (3) as encountering a temporary condition, such as car accident and construction, it reduces the average speed of this road. therefore, to speed up the transmission of traffic information, we let the vehicles on that road proactively send a warning message to inform vehicles on the other roads. 2.2. vehicle route planning in the vehicle route planning, vehicles use the real-time information in the crt map to choose route. the basic factor is the travel time required for each road segment. in addition, we also consider the update time of the road information and regard this factor as the information reliability. next, we consider the future conditions of roads. the factor of expected traveling paths of vehicles is considered in the calculation of the popularities of road segments. the higher popularity of a road segment means there are more vehicles passing through the segment. thus, we tend to choose a lower popularity route. the collection and update of road popularities are integrated in the crt map operations. the information reliability of a road includes distance reliability and time reliability. the farther the road from the current position of a vehicle will be traveled in the longer future. the traffic condition on the road will continue changing. hence the information of the road is relatively less valuable to the current path planning. it means that, we would like to reduce the influence of those farther roads in the path selection. on the other side, time reliability is the difference between the update time of the record in the map and the current time. the longer the information is not updated, the less reliability of the information is. the popularity represents the number of vehicles expected to travel on a road. when the popularity of a certain road is high, it means that there may be more vehicles than expected when actually arriving at the road. thus, we tend to drive on the road with lower popularity. to calculate the popularity, we use a temporary popularity and a permanent popularity for each road segment. temporary popularity resets to zero when entering to a new road segment. the temporary popularity and the permanent popularity are integrated after arriving at an intersection. once a vehicle vi travels on a road, it can obtain the traveling paths of encountered vehicles through the beacon messages. vi increases the temporary popularity of corresponding road segments in the traveling paths by one. then, when vi travels to the intersection, it integrates the temporary popularity collected with the permanent popularity in the traffic map. when exchanging crt maps from different vehicles, popularity estimations by different vehicle can also be exchanged and integrated. proceedings of engineering and technology innovation, vol. 9, 2018, pp. 25 31 copyright © taeti 29 vehicle will calculate the cost of each road according to the above factors using the following equation (3) and select the route with the lowest cost to the destination. we assume that the following is the set of roads in one of the paths to the destination road𝑒𝑘 : {𝑒1, 𝑒2, … , 𝑒𝑘 |𝑒1 = 𝑒𝑛, 1 ≤ 𝑗 ≤ 𝑘} 𝐶𝑜𝑠𝑡(𝑒𝑛) = ∑ [(𝛼 × 𝑇𝑟𝑗 + 𝛽 × 𝑃𝑗 ) × 𝑊𝑇𝑗 × 𝑑𝑗 ] 𝑘 𝑗=1 (1) 𝑇𝑟𝑗 is the travel time of road 𝑒𝑗 . 𝑃𝑗 is the popularity of road 𝑒𝑗 , 𝑊𝑇𝑗 is the time reliability of 𝑒𝑗 and 𝑊𝑇𝑗 = 𝛾 ∗ ln((𝑇𝑐𝑢𝑟𝑟𝑒𝑛𝑡 − 𝑇𝑗 ) + 𝑒). 𝑑𝑗 is the distance reliability of𝑒𝑗 , which use the distance between the current road and road 𝑒𝑗 to divide by the average road length �̅�. 𝛼, 𝛽, 𝛾 are the weights of travel time, popularity and time reliability, respectively. 3. simulation results in this section, we assess the performance between the proposed scheme, dedr[9], and the shortest path. in dedr, the road segment is divided into multiple clusters. the cluster head is responsible for collecting the information of vehicles in the same cluster and then transmitting to other clusters. with the proactive information exchange between the clusters, you are able to obtain traffic information. in order to determine whether have to change the path, the authors in [9] also proposed the trust probability to predict the traffic condition of selected path. fig. 4 road map of the simulation in our simulation, we use sumo 0.30.0[10] and traci[11] to be our simulator. sumo is an open source vehicle simulator that can simulate the movement of vehicles. traci is library, which can communicate with sumo and control the behavior of vehicles in real-time. we use python as the implementation language of traci. furthermore, we use openstreetmap[12] to capture the map of kaohsiung as the road map used in our simulation, as shown in fig. 4. the size of our map is 4.5 km * 4.5 km. the total length of simulation time is 3000 seconds, and the vehicle depart rate is 1 ~ 7 vehicles per second, resulting a total number of 3000 ~ 21000 vehicles. we simulate two scenarios according to different vehicle generation patterns: (1) uniform mode, and (2) non-uniform mode. the uniform mode is to select 10 roads on each of the four boundaries of the map, in which we randomly select the source and destination from these 40 roads. in the non-uniform mode, in order to simulate the situation where vehicles are concentrated on certain roads, we randomly select the source and destination from 20 roads in two pairs of boundaries, where 10 roads are selected with the probability of 70%; the other 10 roads are 30%. we average the results over 100 times in each scenario. 3.1. travel time of vehicles in the comparison of traveling time, we compare our method with dedr and the vehicles traveling with shortest path. in fig. 5, when the vehicle depart rate is low, the three methods have slight differences in travel time. as the vehicle departs rate proceedings of engineering and technology innovation, vol. 9, 2018, pp. 25 31 copyright © taeti 30 increases, the vehicle density also becomes high. the average travel time of the three methods rise. moreover, the travel time of shortest path is much higher than other methods. it is because vehicles almost choose the same paths and will not change the path when traveling. comparing to dedr, the travel time of our method is much lower at different vehicle depart rates. approximately 18% travel time is reduced in the case of 7 vehicles produced per second. fig. 6 compares the travel time in the non-uniform mode. while comparing to the uniform mode, vehicles may be concentrated on some roads in the non-uniform mode, which increasing the chances of road congestion. therefore, the travel time of both three methods will rise slightly. however, the travel time of our method is still lower than dedr, especially in the case of high vehicular density. fig. 5 travel time vs. vehicle depart rate in uniform mode fig. 6 travel time vs. vehicle depart rate in non-uniform mode 3.2. packet overhead in fig. 7 and fig. 8, we compare the average packet overhead of different vehicle depart rates in two vehicle generation scenarios. as shown in the figures, dedr is a proactive method and proactively send traffic information to other vehicles. therefore, in both scenarios, the packet overhead of our method is less than dedr. our method can reduce up to about 70% of the packet overhead in the uniform mode, and 48% in non-uniform mode. in addition, because the probability of traffic congestions in non-uniform mode is higher than uniform mode, the packet overhead of our method in non-uniform mode is higher than uniform mode. besides, the information of roads with few vehicles driving on is hard to be updated due to the unbalanced vehicle distribution. our method will send additional packets for congested roads or outdated information roads, thus the packet overhead is higher in non-uniform mode. however, comparing to dedr, the packet overhead of our method is still lower. fig. 7 packet overhead vs. vehicle depart rate in uniform mode fig. 8 packet overhead vs. vehicle depart rate in non-uniform mode 4. conclusions in this paper, we utilize the passive traffic collection schemes through the inter-vehicle communication. in order to solve the problem of slow updating and the limited coverage range of coverage, we integrate proactive schemes to speed up information spreading for congestion roads, outdated roads, and temporary condition roads. we combine the proactive and proceedings of engineering and technology innovation, vol. 9, 2018, pp. 25 31 copyright © taeti 31 passive schemes, which allow vehicles collect extensive and real-time information with low packet overhead. moreover, we also propose a vehicle rout planning scheme that utilizes the traffic information collected by each vehicle and takes into account the expected route of other vehicles. the simulation results shows that vehicles using the proposed methods can reach their destinations with short travel time and obtain traffic information with low packet overhead. in the future, we will tend to combine the schedule of traffic lights to enable vehicles to select a faster and smoother route. acknowledgements the study is sponsored by most under grant no. 106-2221-e-194-002. references [1] d. schrank, b. eisele, t. lomax, and j. bak “2015 urban mobility report,” the texas a&m transportation institute and inrix, august 2015. [2] s. al-sultan, m. m. al-doori, a. h. al-bayatti, and h. zedan, “a comprehensive survey on vehicular ad hoc network,” journal of network and computer applications, vol. 37, pp. 380-392, january 2014. [3] c. englund, l. chen, a. vinel, and s. y. lin, “future applications of vanets,” in vehicular ad hoc networks springer international publishing, pp. 525-544, 2015. [4] m. j. booysen, s. zeadally, and g. j. van rooyen, “survey of media access control protocols for vehicular ad hoc networks,” iet communications, vol. 5, no. 11, pp. 1619-1631, 2011. [5] e. uhlemann, “introducing connected vehicles,” ieee vehicular technology magazine, vol. 10, no. 1, pp. 23-31, march 2015. [6] m. chaqfeh and a. lakas, “shortest-time route finding application using vehicular communication,” proc. ieee wireless communications and networking conference (wcnc), november 2014. [7] c. f. huang, y. f. chan, and r. h. hwang, “a comprehensive real-time traffic map for geographic routing in vanets,” applied sciences, vol. 7, no. 2, january 2017. [8] r. k. ganti, f. ye, and h. lei, “mobile crowdsensing: current state and future challenges,” ieee communications magazine, vol. 49, no. 11, november 2011. [9] j. lin, w. yu, x. yang, q. yang, x. fu, and w. zhao, “a real-time en-route route guidance decision scheme for transportation-based cyberphysical systems,” ieee transactions on vehicular technology, vol. 66, no. 3, march 2017. [10] d. krajzewicz, j. erdmann, m. behrisch, and l. bieker, “recent development and applications of sumo simulation of urban mobility,” international journal on advances in systems and measurements, vol. 5, no. 3-4, december 2012. [11] a. wegener, m. piorkowski, m. raya, h. hellbrück, s. fischer, and j. p. hubaux, “traci: a framework for coupling road traffic and network simulators,” proc. of the 11th communications and networking simulation symposium, ieee press, april 2008. [12] m. haklay and p. webber, “openstreetmap: user-generated street maps,” ieee pervasive computing, vol. 7, no. 4, october 2008. https://ieeexplore.ieee.org/xpl/mostrecentissue.jsp?punumber=6923023 https://ieeexplore.ieee.org/xpl/mostrecentissue.jsp?punumber=6923023 proceedings of engineering and technology innovation, vol. 8, 2018, pp. 32 39 comparative analysis of performance of r600a and r510a refrigerants in a retrofitted refrigerating system francis olusesi borokinni 1 , babatunde emmanuel olatunbosun 2,* 1 department of mechanical engineering, yaba college of technology, yaba , nigeria. 2 department of agricultural engineering, akwa ibom state university, ikot akpaden, nigeria. received 26 july 2017; received in revised form 07 august 2017; accept ed 17 august 2017 abstract hydrocarbons, in the environmentally friendly group have drawn much attention of scientists and researchers for applicat ion as refrigerants. in this work, the performance of t wo eco -friendly hydrocarbon refrigerants (r510a and r600a ) in a retro fitted vapour comp ression refrigeration system was investigated e xperimentally and co mpared with the baseline hydro fluorocarbon refrigerant (r134a). thermocouples and pressure gauges were fitted at the inlet and outlet of the compressor, condenser, and evaporato r to measure the temperature and pressure of the refrigerant at various stages of the re frigeration cycle . the results obtained showed that the coefficient of performance (cop) of the system using the two hydrocarbon refrigerants was higher than the coefficient of performance (cop) obtained using r134a as a working fluid. genera lly, the two hydrocarbon refrigerants performed better than r134a, but r510a gave the best overall performance in that it exhibited the lowest discharge pressure of 0.656mn/m2and the highest refrigerating effect of 253.29kj/kg. keywor ds: hydrocarbon, refrigerant, retrofitted, thermocouple 1. introduction most of the refrige ration and air-conditioning system operate on vapour compression refrigeration cycle [1] in which the refrigerant change phases fro m liquid to gas and gas to liquid in a closed cycle to generate cooling in the evaporator. refrigerants used in these systems are predominantly fro m a group of co mpounds called halocarbons (halogenated hydrocarbons) in 1974, sherwood rowland and mario molina predicted that chlorofluorocarbon (cfc) refrigerant gases would reach the high stratosphere and there damage the protective mantle of the oxygen allotrope, ozone. in 1985, with the discovery of the "ozone hole" over the antarctic, the prediction of rowland and molina's was proved correct. cfcs are non-toxic, non-fla mmab le are used extensively as refrige rants in refrigerat ion units, aerosol propellants, electronic c leaning solvents, and blowing agents [2]. over t ime, these cfcs get released into the a ir and often, strong winds carry the m into the stratosphere. when cfc molecules drift into the stratosphere, the uv-b and uv-c radiat ion fro m the sun releases their chlorine ato ms. co mple x che mica l reactions in the at mosphere result in the format ion of chlorine mono xide, which reacts with the ozone molecu le to form o xygen and regenerates more ch lorine atoms that carry on converting the ozone mo lecules. each ch lorine atom can destroy as many as 100,000 o zone molecules over 100 years. thus, even a small a mount of cfcs can cause tremendous damage to the ozone layer [3 -4]. therefore, they have been forbidden in developed countries since january, 1996. in 2010 production and usage of cfcs have been prohibited comp letely all over the world [5-6]. * corresponding author. email address: olat.tunde12@gmail.com proceedings of engineering and technology innovation, vol. 8, 2018, pp. 32 39 copyright © taeti 33 many studies have been carried out to find suitable rep lace ment for cfcs [7 -8] transitional alternative co mpounds, such as hydro-chlorofluorocarbons (hcfcs), wh ich are less harmful to the o zone layer, inc luding r22, r123 and r124 will be phased out internationally by 2020 and 2030 in developed and developing countries respectively, because they still contain ozone depleting chlorine though their ozone dep letion potentials (odps) are very sma ll and less than those of cfcs [9]. 2. material and method 2.1. properties of selected refrigerants two hydrocarbon (hcs) and one hfc refrigerants (r510a, r600a, and r134a) were selected and their performances in vapour compression refrigeration were investigated and analyzed. these are natural, ch lorine free refrigerants; therefore they are not harmful to the ozone layer and global warming. so me of the properties and environment impact of the selected environmental friendly refrigerants are as tabulated below. table 1 some properties and environmental impacts of selected alternative refrigerants physical and environmental characteristic of selected refrigerants properties r134a r600a r510a molecular weight (kg/kmol) 102.000 58.120 47.240 critical temperature ( o c) 101.300 134.660 128.100 critical pressure (mpa) 4.0069 3.630 5.120 density 225.500 268.560 boiling point ( o c) -26.07 -11.75 -34.40 odp 0 0 0 gwp 1300 3 3 2.2. experimental set-up the system was incorporated with two pressure gauges with accuracy of ± 0.5kpa at the inlet and outlet of the compressor for measuring the suction and discharge pressures. the temperature of the refrigerant at four diffe rent points was measured with digita l thermocouples with accuracy of ± 0.1 o c. the energy consumption of the refrigeration system was measured with energy meter with accuracy of ± 0.2kwh. data were co llected at diffe rent evaporator temperatures and the following performance para meters were obtained using equations 1 to 7: refrige rating effect (q evap), compressor work input (wc), condenser heat load (qcond) coeffic ient of performance (cop), vo lu metric cooling capacity (vcc) and pressure ratio (pr). c. 2.3. analysis of the heat transfer in the refrigeration system 2.3.1. evaporator the heat absorbed by the refrigerant in the evaporator or refrigerating effect (qevap, kj/kg) is expressed as: 1 4 ( ) evap q h h  (1) where, 1 h = specific enthalpy of refrigerant at the outlet of evaporator (kj/kg); and 4 h = specific enthalpy of refrigerant at the inlet of evaporator (kj/kg). 2.3.2. compressor the isentropic work input to compressor (ks/s) is expressed as: 2 1 ( ) cs r w m h h  (2) where h2 is the enthalpy of refrigerant at the outlet of compressor (kj/kg) proceedings of engineering and technology innovation, vol. 8, 2018, pp. 32 39 copyright © taeti 34 the actual compressor work (w e, kj/s) is given as / c cs w w ns (3) where ns is the isentropic efficiency 2.3.3. condenser the heat rejected by the condenser (qcondkj/s) to the atmosphere is given as 2 3 ( ) cond r q m h h  (4) where h3is the enthalpy of refrigerant at the outlet of condenser (kj/kg) 2.3.4. capillary tube in the capillary tube the enthalpy remains constant (isenthalpy process), therefore, 3 4 h h (5) fro m the first law of thermodynamic point of view, the measure of performance of the refrigeration cycle is the coefficient of performance (cop) and is the refrigerating effect produced per unit of work required it is expressed as: / evap c cop q w (6) the volumet ric cooling capacity ( vcc , kj/ m3) is the re frigerating effect per unit volu me flow rate at the in let to the compressor. it is expressed as / cc evap r s v q m v (7) where vs is the specific volume at inlet to the compressor (m 3 /kg) compressor pressure ratio (pr) is given as: / r dis sue p p p (8) where, dis p = re frigerant vapour pressure at the compressor discharge (kn/ m2) and dis p = refrigerant vapour pressure at the compressor section (kn/m 2 ). 2.4. retrofit procedures the e xisting refrigerat ing system used was designed to work with r134a refrigerants need to be retrofitted to use hydrocarbon refrigerants (r600a and r510a). the specific caution o f the refrigerator is shown in tab le 4. retrofitting means the modification of an e xisting refrigeration system, which was designed to operate on r134a refrige rant so that it can safely and effectively operate on hydrocarbon refrigerants. this is to ensure that existing equipment operates until the end of its economic life . r600a and r510a, wh ich is wide ly accepted as a substitute for r134a in refrigeration systems, is not a "drop-in" replacement for r134a 2.5. baseline test the refrigerator was tested using r134a as the baseline and performance data prior to retrofit were obtained. the system was evacuated with the he lp of a blue va c vacuum pump to re move the non -condensable particle fro m the system. the system was charged with the help of manifold gauge. the pressure gauge and thermocouples were connected to the system. the system was instrumented with two pressure gauges with accuracy of + 0.5kpa at the inlet and outlet of the compressor for measuring the suction and discharge pressures. proceedings of engineering and technology innovation, vol. 8, 2018, pp. 32 39 copyright © taeti 35 2.6. charging of refrigerant charging is the process of adding refrigerant to refrigeration system. the system may be charged with the refrigerant through the high or lo w sides. when charging is done on the high side, the refrigerant is introduced in liquid form but when charging in lo w side, the re frigerant is introduced in vapour form in order to prevent possible da mage to the co mp ressor. norma lly it is more convenient to add refrigerant to system through the low side on small units or when sma ll a mounts are to be added to large systems. the low side method of charging refrigerant into the system was employed in this work. ws 150 digital charging scale was used to charge the system. th is is an automatic d igital charging system that can charge the desired amount accurately. the charging system c onsists of platform, a processing lcd, an electronic controlled valve and charging line hose. the refrigerant cylinder is placed on the platform which measures the weight and also acts as a control panel. one charg ing hose is connected with the outlet of t he cylinder and in let of the electronic valve and another one is connected with the outlet of electronic valve and inlet of the service port. using this charging system, r600a and r510a were charged and tested one after the other. the systems being charged with r600a and r510a require a sma ller charge size than those using r134a. as reco mmended by [10], the charge of r600a and r510a was 90 percent by we ight of the original r134a charge with the optimized capillary tube system. fig. 1 schematic diagram of the existing refrigerator table 2 the specification of the refrigerator specifications value freezer capacity (litres) 130 fresh food compartment capacity (litres) 320 power rating (w) 60 current rating (a) 0.60 voltage (v) 220 frequency (hz) 50 no of door 1 refrigerant type r 134a freezer dimension width : 43 mm, diameter :500 mm and height :830mm 3. results the performance comparison of the investigated refrigerants (r600a, r510a and r134a) in the retrofitted vapour compression refrigerating system was carried out and the enthalpies of the system at various evaporating temperatures and pressures were obtained using refrigerant datab ase software known as refprop [11]. the performance para meters obtained using hydrocarbon refrigerants (r600a and r510a) in the system were analysed and compared with those obtained using the baseline refrigerant (r134a ). the results of effects of the evaporating te mperature on condenser heat load (qcond), volu metric condenser evaporator capillary tube compressor receiver dryer-filter 2 1 3 4 proceedings of engineering and technology innovation, vol. 8, 2018, pp. 32 39 copyright © taeti 36 cooling capacity (vcc) coeffic ient of performance (cop), refrigerat ing effect (qevap) distance temperature (tdis) and compressor work in put (wc) are investigated. table 3 the system enthalpies at varying evaporating temperature for r600a evaporating temperature enthalpy (kj/kg) ( o c) h1 h2 h3 -28 519.92 582.21 223.15 -29 518.44 591.71 227.87 -30 516.96 599.93 234.94 -32 514.08 604.92 244.49 -33 512.62 609.96 258.98 table 4 the system enthalpies at varying evaporating temperature for r510a evaporating temperature enthalpy (kj/kg) ( o c) h1 h2 h3 -26 476.36 528.12 89.46 -28 473.71 532.36 94.12 -30 471.08 535.22 106 -32 468.46 539.52 113.14 -34 465.84 546.76 122.76 table 5 the system enthalpy at varying evaporating temperature for r134a evaporating temperature enthalpy (kj/kg) ( o c) h1 h2 h3 -25 386.9 422.08 227.47 -27 385.39 428.73 230.3 -29 383.88 430.39 234.55 -30 383.11 432.07 237.42 -32 381.61 433.75 241.72 4. discussion 4.1. effect of evaporating temperature on discharge pressure the variations of the discharge pressure as a function of the evaporating temperature for the three r efrigerants are shown in fig. 2. as shown in the figure, the discharge pressure reduces as the evaporating temperature increases. r510a has the lowest pressure with average pressure of 13.40% lo wer than that of r134a. ho wever, r134a e xh ibited significantly high pressure as compared to r510a and r600a . the pressure of r510a was very close to that of r600a. re frigerant with low pressure is desirable in the system because the higher the pressure the weightier must be the equipment accessories and parts . 4.2. effect of evaporating temperature on pressure ratio fig. 2 variation of vapour pressure varying evaporator temperatures for r134a, r600a and r510a fig. 3 variation of pressure ratio with varying evaporator temperatures for r134a, r600a and r510a 0 2 4 6 8 10 12 -34 -29 -24 v a p o u r p re s s u re k j /k n evap oration temp erature (oc) r600a r510a r134a 0.50 1.00 1.50 2.00 2.50 3.00 3.50 4.00 -34 -29 -24 p r a ti o evap oration temp erature (oc) r600a r510a r134a proceedings of engineering and technology innovation, vol. 8, 2018, pp. 32 39 copyright © taeti 37 fig. 3 shows the variation of pressure ratio with varying evaporating te mperature for r600a, r510a and r134a. the figure shows that the pressure ratio decreases with increase in evaporating te mperature. the three re frigerants followed thes e similar trends. as reflected in the figure, the pressure ratios of r134a a re higher than those o f r510a and r600a. average pressure ratios obtained using r510a and r600a in the system were 6.2 and 4.2% lowe r than that of r134a respectively. therefore, same comp ressor is usable for both r510a and r600a, while a slightly heavy compressor for the same c apacity will be needed for r134a. 4.3. effect of evaporating temperature on condenser heat load the function of the condenser is to re move heat of the vapour refrigerant discharged from the compressor. heat is added to the refrigerant during evaporation in the evaporator by the compressor during the work o f co mpression. the heat fro m the refrige rant is moved by transferring heat to the wall of the condenser tu bes and then fro m the tubes to the condensing mediu m. the variat ions of the condenser heat load against evaporating temperature a re presented in figure 4 for the three refrigerants. the figure shows that the condenser heat load increases as the evaporating temperature increases. also, in the figure, r134a has the highest mean heat load. the average heat loads of r510a and r600a were 23.3 and 16.90% lower than that of r134a respectively. if the te mperature of the evaporator increases, the work of co mpression increases. as work of co mpressor increases the heat added to the refrigerant during co mpression increases so the condenser requires more heat to remove. fig. 4 variation of condenser heat load with varying evaporator temperatures for r134a, r600a and r510a 4.4. effect of evaporating temperature on coefficient of performance fig. 5 variation of coefficient of performance (cop) with varying evaporator temperatures for r134a, r600a and r510a 1.00 2.00 3.00 4.00 5.00 6.00 7.00 -34 -32 -30 -28 -26 -24 c o p evap oration temp erature (oc) r134a r510a r600a proceedings of engineering and technology innovation, vol. 8, 2018, pp. 32 39 copyright © taeti 38 fig. 5 shows the variat ion of coefficient of performance (cop) with varying evaporator te mperature for the three refrigerants. as shown in this figure, the cop increases as evaporator temperature decreases. the cop is the ratio of the refrigerating e ffect to the co mp ressor work. the increase in evaporating te mpe rature increases the refrigerating effect and decreases the compressor work, therefore increases the cop of the refrigerat ion system. the cop of r510a is the highest with average values of 14.6 and 13.7% higher than those of r134a and r600a respectively. 4.5. effect of evaporating temperature on refrigerating effect the refrigerating effect is the ma in purposes of the refrigeration system. the liquid re frigerant at the low pressure side enters the evaporator. as the liquid refrigerant passes through the evaporator coil, it continually absorbs latent heat of vapourization at constant temperature through the coil wa lls, fro m the mediu m being cooled and turn to vapour refr igerant. the refrigerat ing effect is the difference between the enthalpies of the refrigerant in the inlet and the outlet of the evapo rator. the variation of re frigerating effect with the inlet evaporating temperature is shown in fig. 6. fro m the figure it is evident that the refrigerating e ffect increases as the evaporating temperature increases. r134a has the lowest mean refrigerating effect. average re frigerating effect obtained for r510a and r600a were 26.0 and 16.5% higher than that of r134a, respectively. fig. 6 variation of refrigerating effect with varying evaporator temperatures for r134a, r600a and r510 5. conclusion hydrocarbons are environmentally friendly. they have zero ozone dep letion potential (odp) and negligible globa l warming potential (gw p). hydrocarbon is cheaper than the r134a wh ich is being used in the refrigerator at present. hydrocarbon is also easily av ailable. in this work the performances of three ozone friendly refrigerants, one hydro fluorocarbon (r134a) and two hydrocarbons (r510a and r600a) in a retrofitted e xisting vapour compression refrigeration system were investigated experimentally and compared. based on the results obtained, the following conclusions are drawn: (1) the coefficient of performance obtained for the two hydrocarbon refrigerants (r510a and r600a) were higher than that of baseline refrigerant (r134a) (2) the average pressure of r510a and r600a are 13.4% and 10.2% lower than that of r134a respectively. (3) the average compression work for r134a was 18.7% higher than that of r510a and r600a respectively. (4) the average pressure ratio of r134a is higher than those of r510a and r600a by 4.2 and 6.2%, respectively. (5) the condenser heat load increases as the evaporator temperature increases. r134a has the highest mean load. the mean heat load rates of r510a and r600a were 16.9 and 23.3% lower than that of r134a respectively generally, the t wo hydrocarbon refrigerants performed better than r134a and they can be used as retrofit substitutes for r134a in e xisting vapour compression refrigerating systems. the best performance was obtained from the use of r510a in the system. 50.00 100.00 150.00 200.00 250.00 300.00 350.00 400.00 -34 -32 -30 -28 -26 -24 r e fr ig e ra ti n g e ff e c t (k j /k g ) evap oration temp erature (oc) r134a r510a r600a proceedings of engineering and technology innovation, vol. 8, 2018, pp. 32 39 copyright © taeti 39 references [1] r. kumar and r. shekhar, “performance analysis of vapour compression refrigeration system by using azeotrops,” international journal for scientific research & development, vol. 3, no. 4, pp. 255-267, 2015. [2] f. bill, and l. duncan. “chlorofluorocarbons (cfcs),” the gale encyclopedia of science, http://link.galegroup.com/ap ps/doc/cv2644030480/scic?u=a lbertak 12&xid=c 35ecbf6, february 2018. [3] b. o. bolaji, a. o. akintaro, o. j. alamu, and t. m. a. olayanju, “design and performance evaluation of a cooler refrigerating system working with ozone friendly refrigerant,” the open thermodynamics journal, vol. 6, no. 1, pp. 2532, 2012. [4] a. b. agrawal and v. shrivastana, “ retrofitting of vapour compression refrigeration trainer by an eco-friendly refrigerant,” indian journal of science and technology, vol. 3, no. 4, pp. 455-458, april 2010. [5] y. kim, k. s. chang, and h. kim, “thermodynamic performance analysis of vapour compression system using alternative refrigerants based on a cycle simulation program,” journal of industrial engineering chemicals. vol. 13, no. 5, pp. 674-686, 2007. [6] j. fernandez-seara, f. j. uhia, r. diz, and j. a. dopazo, “vapour condensation of r22 retrofit substitutes, r417a, r422a and r422d on cuni turbo c tubes,” international journal of refrigeration, vol. 33, no. 1, pp. 148-157, 2010. [7] b. o. bolaji, m. a. akintunde, and t. o. falade, “comparative analysis of performance of three-ozone friends hfc refrigerants in a vapour compressor refrigerator,” journal of sustainable energy and environment, vol. 2, pp. 61-64, 2011. [8] d. del col, m. azzolin, s. bortolin, and c. zilio, “two phase pressure drop and condensation heat transfer of r32/r1234ze (e) non-azeotropic mixtures inside a single microchannel,” science and technology for the built environment, vol. 21, no. 5, pp. 595-606, june 2015. [9] b. o. bolaji and z. huan, “thermodynamic analysis of hydrocarbon refrigerants in a sub-cooling refrigeration system,” journal of engineering research, vol. 1, no. 1, pp. 317-3333, june 2013. [10] bitzer, refrigerant report, bitzer international, www.bitzer.de, february 2018. [11] e. w. lemmon, m. o. mclinden, and m. l. huber, “nist reference fluids thermodynamic and transport properties refprop 7.0,” national institute of standards and technology standard reference database, vol. 7, 2002. http://www.bitzer.de/  proceedings of engineering and technology innovation , vol. 4, 2016, pp. 37 39 37 nostril in rgb imaginary by using ni vision labview hendrick 1,2,* , aripriharta 1,3 , siang-kai chen 1 , meng-hsiung tung 1 , ta-chi chiang 1 , gwo-jia jong 1 1 department of electrical engineering, national kaohsiung university of applied sciences, kaohsiung, taiwan. 2 state polytechnic of padang (pnp), padang 21562, indonesia. 3 state university of malang (um), malang 65145, indonesia. received 01 april 2016; received in revised form 27 april 2016; accept ed 10 may 2016 abstract monitoring of breath function is needed for med ical science to evaluate condition of patient for some diseases. the patient will be observed continuously on face (nostril). face recognition is the most important stage to localize the measure ment. because of that need, therma l image is used to record changing temperature on patient face. therma l image has two type image, gray and rgb image. in this paper, we used rgb image as the raw data and trying to find face, eye and, nostril as breathing indicator. we divided the nostril recognition in three ma in steps: eyes detection, calculate the center of mass between two eyes, measured the center of mass distance to nose, get nostril location. we applied image processing algorithm such as gray image, b inary image and template matching . all those algorith m are progra mmed in ni vision lab view. the distance between subject and came ra is fixed in 1 meter. the e xpe rimental result showed that nostril a lways detected even though the subjects have some move ments du ring measurement. ke ywor ds : breath function, therma l image, face detection, nostril 1. introduction nowadays , vital signs monitoring of patient become important in modern medical observation. vital sign of human such as ecg, ppg, em g etc, are usually used wired device that placed on patient body. in other case of disease, patient should be in rela x position or feeling scared with all device that placed on their body. non-contact device is the solution to handle that problem, and use thermal image device to capture some of that. facial detection is the most important part in this field, because it contains eye, nose (nostril) and mouth of the patient [1]. first stage of facial recognition usually is focused of eye detection. eyes detection in thermal image is a little different from common eyes image captured from usb camera [2]. after eyes detected, the stage continues to nose detection by using center location between two eyes. detecting nose in facial d etection some time uses prediction to localize nose location [3]. in this research, our focus is on nostril of nose as our goal to monitoring some of human vital signs (breathing function) [4]. we devided into three stages until nostril detected: face detection, eyes detection, nose detection and nostril detection. flir thermal image is our device to capture human face and save into avi files. we also use above algorithm so that nostrils can be detected automatically. all stage is programmed in ni vision labview. 2. method 2.1. flir thermal camera and subject setup in this research, we used flir therma l came ra p384-20 that has resolution 384 x 288 pixe ls with 50 fra mes per second (fps) in co lor mode. fig. 1(a) shows our came ra flir therma l image camera. the distance between subjects and flir thermal ca mera is 1 meter with manual calibration on camera resolution. fig. 1(b) shows our experimental setup between camera and subject. the video of subject will be captured using flir thermal ca mera in 5 minutes with minor mov ement of their head and the files are converted to avi files. during video recording, the subjects will breathe normally. * corresponding aut hor. email: hendrickpnp77@gmail.com proceedings of engineering and technology innovation , vol. 4, 2016, pp. 37 39 38 copyright © taeti (a) (b) fig. 1 (a) flir therma l ca mera , (b) expe rimental setup 2.2. nostril track ing algorithm in order to detect nostril location, we p roposed 4 ma in steps such as: face detection, eyes detection, nose detection and nostril detection. all a lgorith ms are progra mmed with ni vision lab vie w 2015 by using virtual instrument (vi) in ni vision. the nostril t racking algorithm is e xp lained below: (a) (b) fig. 2 (a) gray image, (b) eyes template  step 1 – eyes detection: detecting eyes on this step through a few process to do, such as convert image to gray, set region of interest of eyes location and make eyes templete. the eyes template shows on fig. 2(b). in this research, we focus on the highest temperature in the corner of the eyes to make the eyes d etection faster to find. the ne xt step is to calculate the centre of mass (cof) between to eyes that will be used as the centre point to get the nose location.  step 2 – nose detection: the cof then is used as centre point to find nose location by using ratio 1/3 fro m chin of face[]. by applying that, we can set the roi of nose location.  step 3 – nostril detection: nostril is easly found on nose roi because nostril is ind icated as the highest temperature in roi location. 3. results and discussion fig. 3 shows block d iagra m of nostril detection in therma l image wh ich consist of load avi files, eyes -nose detection and nostril detection. fig. 3 lab view b lock diagra m of nostril de tection fig. 4 (a) gray image, (b) nostril not detected, (c) nostril, (d) nostril waveform cycle fig. 4 shows our results in nostril detection and location in therma l video. fig . 4(a) shows the center mass of eyes with blue rectangle and nostril location with green rectangle. fig 4 (b), (c) and (d) show nostril location and visualize nostril activity with graph. proceedings of engineering and technology innovation , vol. 4, 2016, pp. 37 39 39 copyright © taeti 4. conclusions the ma in objective of this research was to determine the location of the nostril on rgb therma l v ideo. the e xperimental result showed that nostril always detected even though the subjects have some move ments during measurement. acknowledgement this work was supported by dikti funded by ministry of research and technology and higher education, indonesia under contract number: 138.7/e4.4/ 2015. references [1] y. w. wu, h. liu, and h. b. zha, “modeling fac ial e xpression space for recogn ition,” ieee/ rsj international conference on intelligent robots and systems , pp. 1968-1973, 2005. [2] mehrube and l. m. pha m, “real-time eye tracking using a smart ca mera ,” proc. applied imagery pattern recognition workshop, november, 2014. [3] c. b. pere ira , x. yu, v. bla zek , and s. leonhardt, “robust remote monitoring of breathing function by using infrared thermography,” the 37th annual international conference of the ieee engineering in medicine and biology society (em bc ’15), pp. 4250-4253, 2015. [4] a. k. abbas, k. heimann, k. jergus, t. orlikowsky, and s. leonhardt, “neonatal non-contact respiratory monitoring based on real-time infra red thermography,” biomedica l engineering online, vol. 10, no. 93, pp. 1-17, 2011. n proceedings of engineering and technology innovation, vol. 6, 2017, pp. 36 39 on the analysis of product appropriateness principles nien-te liu * , yu-fen wu department of product design, shu-te university, kaohsiung, taiwan, roc. received 26 july 2016; received in revised form 05 august 2017; accept ed 17 august 2017 abstract the purpose of this study is to explore appropriateness of the various products presented in the market and the application of appropriateness principles to the product design. to obtain appropriateness principles , this study compares and analyzes a variety of products available in the market. after obtaining preliminary principles, many times convergence and analysis are operated for getting more accurate principles . then, ahp method is used for evaluating importance of every appropriateness principles and appropriateness principles are obtained. and, the appropriateness principles are used to construct the design process of appropriateness product. finally, some design cases are operated and appropriateness principles are evaluated for the products of design cases. the results of this study are obtained 10 appropriateness principles: freedom, demand, environmental protection, effectiveness, intuition, compliance, simplicity, optimal performance, simplicity, fault tolerance . and, weights of appropriateness principles get in this paper. keywords: appropriateness principle, ahp method, evaluation method 1. introduction more and more products are invented and manufactured in our daily life , but not any functions are used for users. therefore, to design the product with appropriate functions for users is an important duty. affordance is the theory for understanding user psychology [1], thus, designers could judge what are the important functions in development and design stage. norman use d the affordance theory into design filed and made a great impact for designers. norman thinks affordance is that the past knowledge and experience affect the user's feelings and judgments about the product [2]. norman’s perspectives are questioned by amant [3], who thinks affordance is s een as a user's perception of the environmental symbolic meaning, rather than the nature of the environment, contrary to the principle of “direct perception” of ecological perspectives. however, the above mentions are the discussion about the relationship between environment and users or the relationship between users and products. the affordance does not describe clearly about the function of product is suitable for user or not . the exhibition “product fitness 80” of muji proposed the concept of “product fitness 80,” and reflects muji's willingness to educate users by reviewing our own “adequacy” (fitness). the introduction of this exhibition has remarks “we constantly question if we have used excessive materials; whether products are overpackaged, or are t heir sizes and weights too much; can we reduce waste in the ordering, manufacturing or transportation of products? less is more.” [4] therefore, the purpose of this paper is proposed the theory about appropriateness, and the appropriateness design process. by following this process, designers could design the product with appropriate functions for users . * corresponding author. e-m ail address: ntliu@ stu.edu.tw tel.: +886-7-6158000#3824 proceedings of engineering and technology innovation, vol. 6, 2017, pp. 36 39 copyright © taeti 37 2. the construction of appropriateness principles 2.1. the analysis of existing cases for developing the theory of appropriateness , the product with appropriateness characteristics are collected and analyzed, and then the principles are summarized. for examples, fig. 1 is a tray made by rubber wood (recycled material), with arc side form and holes for holding easily [4]. fig. 1 rubber wood tray of muji the holes for holding provide convenience for product and reduce the load for user body. thus, the summary of appropriateness attribute for the tray is “reduction the load of user body.” here, 46 products are collected. every product is analyzed to describe the characteristics and to get the appropriateness attributes. after finishing analyzing 46 products, 31 appropriateness attributes are obtained. by classified process, 10 appropriateness principles are obtained. 2.2. list of appropriateness principles here the explanations of appropriateness principles are list in the following mention: (1) demand: whether the function provided by the product has reached the main demand of the user; (2) performance: whether the product provides the maximum functionality; (3) freedom: if the product can provide the best degree of freedom, when using the product, function can be changed by personal will; (4) environmental protection: selection of environmentally friendly products; (5) compliance: make sure to use the dimens ions for ease of use; (6) optimum: whether the material has reached the best effect; (7) easy operation: one hand can also be used; (8) simplicity: do not add extra features or decorations that you do not need to use; (9) simplicity: do not add extra features or decorations that you do not need to use; (10) fault tolerance: allow the scope of tolerance error. 3. weights of appropriateness principles for understanding the importance of appropriateness principles , this paper uses ahp method to get the weights [5]. the process of ahp method are following : proceedings of engineering and technology innovation, vol. 6, 2017, pp. 36 39 copyright © taeti 38 this study issued 243 questionnaires ; got effective questionnaires 197, after screening to get 174 questionnaires. and, we divided the 10 principles into 3 fields as shown as table 1. table 1 the fields of appropriateness principles fields no. a. function b. appearance c. additional feature 1 demand environmental protection optimum 2 performance compliance easy operation 3 freedom simplicity 4 intuition 5 fault tolerance in the next step, we make a couple comparing matrix for calculating the weights. here we need to compare the 3 fields with each other, and compare the principles with each other in every fields. after questionnaire and results analysis, table 2 shows the fields weights and table 3 shows the principle weights . table 2 the fields weights fields field weights function 0.52 appearance 0.27 additional feature 0.21 table 3 the principle weights principles no. principle principle weights rank a1 demand 0.24 2 a2 performance 0.07 4 a3 freedom 0.25 1 b1 environmental protection 0.19 3 b2 compliance 0.06 6 c1 optimum 0.03 8 c2 easy operation 0.02 9 c3 simplicity 0.05 7 c4 intuition 0.07 5 c5 fault tolerance 0.02 10 4. conclusions in this paper, the concept of appropriateness is propose d. and, the appropriateness principles are obtained by collecting and analyzing 46 appropriateness products. ahp method is used here for find the weights of principle. 10 principles are divid ed into 3 fields. the 3 field and their weights are function 0.52, appearance 0.27, and additional feature 0.21. 10 principles and their weights are demand 0.24, performance 0.07, freedom 0.25, environmental protection 0.19, compliance 0.06, optimum 0.03, easy operation 0.02, simplicity 0.05, intuition 0.07, and fault tolerance 0.02. acknowledgement the authors are grateful to the ministry of science and technology (taiwan, r.o.c) for supporting this research under grant most 104-2632-e-366-001 and most 105-2632-e-366-001. proceedings of engineering and technology innovation, vol. 6, 2017, pp. 36 39 copyright © taeti 39 references [1] j. gibson, the ecological approach to visual perception, london: houghton mifflin, 1979. [2] d. a. norman, the design of everyday things , new york: basic books, 1990. [3] r. s. amant, “user interface affordances in a planning representation ,” human-computer interaction, vol. 14, no. 3, pp. 317-354, september 1999. [4] “muji,” http://www.muji.net/lab/fitness80/en/, mar. 30, 2017. [5] t. l. saaty, the analytic hierarchy process, new york: mcgraw-hill, 1980.  proceedings of engineering and technology innovation , vol. 4, 2016, pp. 31 33 31 multi-band slot antenna tsang-yen hsieh * , jyh-liang wang, mei-hui wang department of electronic engineering, ming chi university of technology, new taipei city, taiwan . received 01 february 2016; received in revised form 08 march 2016; accept ed 03 april 2016 abstract a small-size printed antenna is developed for the application of mobile phones . the architecture of antenna bases on the concept of coplanar waveguide (cpw) feeding with slot design. the antenna is planned by an inverted triangular monopole structure with the overall size of 50 mm× 120 mm (shown in fig. 1). the proposed antenna was fabricated on pcb with fr4 d ielectric substrate. the substrate thickness is 1.6 mm with the relative permittiv ity of ~ 4.45. the slots are added to adjust resonance frequency and achieve characteristic of mu lti-band, which may cover the applications of gps, gsm (global system for mobile co mmunications), umts (universal mobile te lecommun ications system), and lt e (long term evolution). by the trial of simulat ion to tune the design of slots , an optimu m design of antenna can be obtained. fig.2 presents the s imu lated and measured return loss of the antenna. even though, there are few minor differences between the simulation and measured data. the measured and simulated results still can agree with each other. the achieved bandwidths are 0.54 to 1.01 ghz, 1.34 to 2.28 ghz, and 2.55 to 2.90 ghz, wh ile the return loss ( 11s ) is limited more than 6 d b. th is antenna is well suitable for the using of mobile phone applications. ke ywor ds : mult i-band, coplanar waveguide (cpw), printed circuit board (pcb), return loss , slot antenna 1. introduction the rapid development of wire less communication leads to the more de mands of the types, speed, and quality of wire less communication. the transmission and receiving antennas are the essential parts of wireless communicat ion systems (such as gsm, gps, umt s, ism , lt e, etc.). therefore , the quality of antenna will greatly affect the emission and reception performances of system. in recent years , a lmost all the mob ile devices are equipped with mu lti -band applications, so it requires multiple frequency characteristics of the antenna to transceiver the communication signals [1-2]. 2. method a small-size printed antenna was designed for the mult i-band require ments of mobile phones. the construction of antenna bases on the concept of coplanar waveguide (cpw) feeding with slot design. the antenna is planned by an inverted triangular monopole structure with the overall dimensions of 50 mm× 120 mm (shown in fig. 1). the proposed antenna was fabricated on pcb with fr4 d ielectric substrate. the substrate thickness is 1.6 mm with the re lative permittivity of ~ 4.45. fig. 1 multi-band slot antenna the origina l triangular structure of the antenna having a wideband characteristic, coupled with the use of the slots will destroy the wide band characteristic of the triangular structure, and then presenting a mu lti-frequency antenna characteristic. the slots were added to adjust * corresponding aut hor. email: t yhsieh@mail.mcut.edu.tw proceedings of engineering and technology innovation , vol. 4, 2016, pp. 31 33 32 copyright © taeti resonance frequency and achieve characteristic of multi-band, which may cover the applications of gps, gsm (global system for mobile co mmunicat ions), umts (universal mobile teleco mmunications system), and lt e (long term evolution). we used an electromagnetic simu lation software to predict various important parameters of this proposed antenna. an actual sample of this antenna was fabricated based on the simulation results. moreover, the measured data and simu lated results of this proposed antenna was compared. 3. simulation and result fig. 2 the measured and simulated data of s11 of the multi-band slot antenna fig. 3 the measured smith chart of the multi-band slot antenna fig. 2 shows the measured and simulated data of 11s of the mu lti-band slot antenna. there are few minor differences between the simu lation and measured data. the measured and simulated results still can agree with each other. the achieved bandwidths are 0.54 to 1.01 ghz, 1.34 to 2.28 ghz, and 2.55 to 2.90 ghz, while the return loss 11 s is limited more than 6 db. the measured smith chart of the multi-band slot antenna is presented in fig. 3. fro m wh ich it is apparent that the impedances of this antenna at 0.7, 1.8, and 2.6ghz band are very close to 50 ohms . the test results of bandwidths, ga ins and radiat ion e ffic ienc ies a re listed in tab le 1. the values of acco mp lished bandwidths are those we ment ioned be fore. via para mete r ad jus tment, the ga in o f slot antenna was g reater than -2.54d bi, rad iation efficiency was highe r than 25%. table 1 the test results of bandwidths, gains and radiation efficiencies of the mu lti-band slot antenna operat ing freq. (ghz) 0.8ghz band (0.7-0.96) 1.9ghz band (1.7-2.17 ) 2.6ghz band (2.56-2.7 ) simulat ed bw (gh z ) (s11 < -6db) 0.69 0.97 1.64 2.14 2.52 3.0 measured bw (gh z ) (s11 < -6db) 0.54 1.01 1.34 2.28 2.55 2.9 gain (dbi) -2.5 -0.3 2.3 3.9 1.8 3 efficiency (%) 25.2 46.9 52.5 77.5 44.8 47.3 4. conclusions a slot antenna used in 2g, 3g, 4g and gps frequency band is proposed. the applicat ions bands of the slot antenna cover gps, gsm 850 / 900/ 1800/ 1900, umt s, lt e700 / 800/ 850/ 900 and lt e 1700/ 1800/1900/2100/2600. the antenna was planned by an inverted t riangula r monopole structure with the design of slots , and the overall size is 50 mm× 120 mm. this an tenna was fabricated on pcb with fr4 d ie lec tric substrate. the substrate thic kness is 1.6 mm with the re lat ive permitt iv ity of ~ 4.45. the achieved bandwidths are 0.54 to 1.01 ghz, 1.34 to 2.28 ghz, and 2.55 to 2.90 ghz, wh ile the return loss 11s is limited mo re than 6 d b. via pa ra mete r ad justment, the ga in of slot a n tenna was greater than -2.54d bi, rad iat ion effic iency was highe r than 25%. this antenna is we ll suitab le for the using of mob ile phone applications . proceedings of engineering and technology innovation , vol. 4, 2016, pp. 31 33 33 copyright © taeti references [1] g. park, m. kim, t. yang, j. byun, and a. s. kim, “the compact quad-band mobile handset antenna for the lte700 mimo application,” antennas and propagation society international symposium, (apsursi '09), ieee press , pp. 1-4, june 2009. [2] c. j. park, d. h. park, k. s. m in, j. w . kim, and i. h. kim, “measure ment characteristics of lt e-mim o antenna fo r 4 g mob ile handy te rmina l,” inte rnationa l sy mposiu m on antennas and propagation (isap), pp. 523-526, 2012.  proceedings of engineering and technology innovation, vol. 11, 2019, pp. 12-20 on the wind reduction effect of windbreak nets in front of the simple greenhouse for planting fruits and vegetables cheng-chang lien 1,* , ching-hua ting 2 , zong-yi lin 1 1 department of biomechatronic engineering, national chiayi university, chiayi 600, taiwan, roc 2 department of mechanical and energy engineering, national chiayi university, chiayi 600, taiwan, roc received 02 march 2018; received in revised form 15 june 2018; accepted 26 july 2018 abstract the island of taiwan experiences frequent strong winds coming with typhoons in the summer and with northeast monsoons in the winter. the strong winds are hazardous to the structure of a simple greenhouse that is widely used for vegetable and fruit plantation. the damage may reduce agricultural yields. windbreak netting is a practical procedure for protecting a greenhouse from strong winds. the purpose of this study is to carry on the wind tunnel simulation test of windbreak nets in front of the simple greenhouse. windbreak nets of densities 50%, 40%, 30% and 20% had a uniform dimension of 10 cm (per unit 1 h) in height and 29.2 cm in width. a windbreak net was tested at tilt angles of 0 o , 5 o , 10 o and 15 o . the wind tunnel simulation test is carried on at 10 m/s of reference wind velocity. the wind velocities at different horizontal distances and vertical distances front and behind the windbreak net are measured. the relative wind velocity is calculated. the wind reduction effect at different horizontal distances behind the windbreak net is analyzed and discussed. from the test results, it is known that when the tilt angle of windbreak net is 0 o and the height of windbreak net is 1h, the corresponding wind reduction effect of windbreak nets with 20%, 30%, 40% and 50% porosity is 62%, 61%, 59% and 52% respectively at 2h horizontal distance. the corresponding wind reduction effect of windbreak nets with 20%, 30%, 40% and 50% porosity is 46%, 45%, 40% and 32% respectively at 10h horizontal distance. however, the increase of tilt angle of windbreak net does not have much influence on the wind reduction effect. keywords: windbreak net, porosity, tilt angle, wind tunnel test 1. introduction in order to prevent the cultivation of the crops influenced by external climate and pest etc., the simple greenhouse is often used to plant the fruits and vegetables. taiwan is located in the subtropical zone. it is often attached by typhoons in summer and there is strong monsoon in winter. whenever the strong wind is coming, it often causes serious wind damage. in order to reduce the damage, the windbreak equipment can be erected in front of the simple greenhouse to reduce the wind velocity, so that the simple greenhouse and fruits and vegetables can get suitable protection, and reduce the damage of crops from the wind disaster. the most common windbreak equipment is the windbreak net, which can reduce the direct impact of strong wind on crops and simple greenhouse, and increase the output and quality of the crops [1-2]. the mesh of windbreak net item is often revealed by the porosity. the porosity represents the percentage of porosity in a porous material and can be expressed as the ratio of the open area divided by the total area of the net sample. the common material of windbreak net is high-density polyethylene (hdpe). when windbreak net is set up, it is necessary to consider the * corresponding author. e-mail address: lanjc@mail.ncyu.edu.tw proceedings of engineering and technology innovation, vol. 11, 2019, pp. 12-20 13 porosity, height, width and distance between nets etc., in which the porosity has a higher impact on wind reduction effect [3-6]. in addition, the setup of windbreak net is mainly to reduce the wind velocity, but it shall be able to bear the wind pressure generated by strong wind. so the pillar material of windbreak nets must also be assessed and screened in detail, in order to improve its strong wind protection function and lengthen its service life. when the air current flows into the windbreak net, part of air current will flow through the windbreak net, and part of air current will flows upwards due to the obstruct of windbreak net. after the air passes over the windbreak net, the air current begins to move downwards, the wind velocity increases gradually as the distance away from the windbreak net. when part of air current passes through the windbreak net with porosity, the perturbation will be formed behind it, which can reduce the wind velocity, and reduce the damage of fruits and vegetables in the simple greenhouse [7]. a lot of papers were adopted in the wind tunnel simulation test to study the windbreak equipment, their results can be applied to the actual field, and can provide reliable references [8-11]. among them, the windbreak net has already been applied widely, such as being used for preventing wind disaster, sand and snow disaster. it has been proved more effective than windbreak forest. because the erection of windbreak net is convenient, cheap and fast, and can produce the windbreak effect at once, so it is the basic and convenient windbreak equipment [12].the purpose of this study is to carry on the wind tunnel simulation test for the windbreak net in front of the simple greenhouse for planting fruits and vegetables. under a constant wind velocity, let the air current passing through the windbreak net with different porosity and tilt angle, the wind velocities at different horizontal distances and vertical distances front and behind the windbreak net are measured. the relative wind velocity is calculated. the wind reduction effect at different horizontal distances behind the windbreak net is analyzed and discussed. 2. materials and method 2.1. experimental materials the wind tunnel machine used for simulation test is the blowdown-type. the fan of this wind tunnel equipment is driven by a motor with 5 hp, and a regulating valve is installed at the air inlet to adjust the wind velocity. the maximum wind velocity in wind tunnel can reach 15 m/s. the wind tunnel testing machine with the length of 10 m is made up of steel plate. the fan is the wind source, which produces the reference wind velocity for carrying on the test. the whole body can be divided into the contracting section, rectifying section, testing section and diffusing section. the cross-section of the testing section is 0.5 m in high, 0.3 m in wide and 4.8 m in long. in addition, there are two transparent observation windows in 0.5 m in high and 2.0 min wide at the side. the illustration of wind tunnel testing machine is shown as in fig. 1. fig. 1 illustration of wind tunnel testing machine proceedings of engineering and technology innovation, vol. 11, 2019, pp. 12-20 14 the net of the windbreak net adopts common nylon knitted net, as shown in fig. 2. the material of model pillar and the base is stainless steel. the thickness of pillar with height 10 cm is 0.5 cm. the width and length of the windbreak net model base is 30 cm and 50 cm. the windbreak net is fixed on the pillar to get the windbreak net model with different porosity. the tilt angle of windbreak net is changed by replacing the pillar with different tilt angle. the porosity of windbreak net is 50%, 40%, 30% and 20%. the tilt angle is changed at 0 o , 5 o , 10 o and 15 o at wind tunnel simulation test. (a) windbreak net sample with 50% porosity (b) windbreak net model with 0 o tilt angle for pillar fig. 2 windbreak net model with 0 o tilt angles and 50% porosity according to the statistics of central weather bureau in taiwan, the average wind velocity is about 5.1 m/s~20 m/s for the northeastern monsoon. the wind velocity of 10 m/s is selected as the reference wind velocity for testing after assessment. the 10 cm height of windbreak net model is referred to as 1h. the central surface of windbreak net model is used as the measurement position of wind velocity in wind tunnel. the horizontal distance x 20 cm in front of windbreak net model is regarded as -2 h. the horizontal distance at the place of the model is regarded as 0 h. there are 10 measurement points of horizontal distance at 2 h, 4 h, 6 h, 8 h, 10 h, 12 h, 16 h and 20 h behind the model. the vertical distance z from the base of windbreak net model is taken at the height of 0.1 h, 0.2 h, 0.3 h, 0.4 h, 0.5 h, 0.6 h, 0.7 h, 0.8 h, 0.9 h, 1.0 h, 1.15 h, 1.3 h, 1.45 h, 1.6 h, 1.75 h, 1.9 h, 2.1 h and 2.4 h. there are 18 measurement points of vertical distance, and 180 test points in total, as shown in fig. 3. fig. 3 measurement position of wind velocity at the central surface of windbreak net model and distribution position diagram of measurement points (where h is the height of windbreak net model, z is the vertical distance, x is the horizontal distance) 2.2. experimental method generally, outside the boundary layer, the velocity of the fluid is close to a definite value without changing with position, but inside the boundary layer, the velocity at the fixed surface is 0 m/s, and the velocity of fluid approaches a definite value far away from the fixed surface. before the wind tunnel simulation test, the inner air current state of wind tunnel should be tested first to understand the thickness of boundary layer. in order to increase the thickness of boundary layer for facilitating the wind tunnel simulation test, the spoiler and carpet are placed at the air inlet of rectifying section as the rectifying equipment. after proceedings of engineering and technology innovation, vol. 11, 2019, pp. 12-20 15 the wind tunnel test, the flow field approaches steady-state at about 60 cm behind the rectifying section, and the thickness of boundary layer is about 20 cm. so the windbreak net model is placed at 100 cm behind the rectifying section for carrying on the wind tunnel simulation test. the windbreak net model with specific porosity is placed in the wind tunnel testing machine at the position 210 cm away from the wind open of fan, as shown in fig. 4. the revolution of the fan was adjusted to maintain 10 m/s of wind velocity in front of the windbreak net model in wind tunnel before the simulation test. the wind velocity was measured according to the planned measurement points at a horizontal distance and vertical distance. the measurement points of horizontal distance are at 20 cm (-2 h) in front of windbreak net model and every 20 cm (2h) behind windbreak net model until 200 cm (20h). the measurement points of vertical distance are every 1 cm (0.1h) from the base of windbreak net model until 24 cm (2.4h). fig. 4 illustration for configuration position of windbreak net model in wind tunnel the measurement apparatus of wind velocity in wind tunnel is composed of an anemometer (dwyer, no. 400, america) and pitot tube. the wind velocity of every measurement point was measured by the pitot tube in wind tunnel. the pressure difference measured by pitot tube is transferred to the pressure converter (df103-14) through the pressure indicator (val-cd23). the data for that position are acquired by the signal acquisition device (cdaq-9174), and transferred to the computer for calculating and displaying its wind velocity finally. the triple repeated measurements are carried in wind tunnel test. the average value is taken as the value and record of wind velocity at the measurement point. the ratio of wind velocity (uz) measured at different vertical distances in wind tunnel and 10 m/s of reference wind velocity (uref) is defined as the relative wind velocity (uz/uref) as eq. (1). in the subsequent data analysis, the relative velocity is used as the correspondence value of wind reduction effect. the relative wind velocity = uz / uref (1) 3. result and discussion 3.1. empty flow field in wind tunnel test fig. 5 the relative wind velocity of empty flow field at every horizontal distance and vertical distance in the wind tunnel testing machine fig. 5 illustrates the relative wind velocity of the empty flow field at every horizontal distance and vertical distance in the wind tunnel testing machine without placing windbreak net model. it can be found out that under every horizontal distance, its corresponding relative wind velocity presents the curve distribution trend associated with the measurement point of vertical distance, and the relative wind velocity is increased with the height of vertical distance. though the ideal air current will generally be assumed to flow uniform in the field, but the flow field will be influenced by the wind tunnel wall of testing proceedings of engineering and technology innovation, vol. 11, 2019, pp. 12-20 16 section actually. when the air is flowing in the testing section, the wind tunnel wall will produce resistance to the air current, and the surface velocity on wind tunnel wall is 0. the velocity will increase slowly away from wind tunnel wall and approach a definite value. this will make the profile of wind velocity presenting curve distribution. 3.2. the relative wind velocity of windbreak net with 0 o tilt angle and different porosity (a) 20% of porosity (b) 30% of porosity (c) 40% of porosity (d) 50% of porosity fig. 6 the relative wind velocity of windbreak net with 0 o tilt angle and different porosity at every measurement point of horizontal distance and vertical distance fig. 6 illustrates the relative wind velocity of windbreak net with 0 o tilt angle and different porosity at every horizontal distance and vertical distance. fig. 6(a) illustrates the windbreak net with 20% porosity, at 0h horizontal distance, the relative wind velocity is lower below 1h vertical distance of windbreak net. this is because when the air current flows into the windbreak net, part of air will flow through the porosity of windbreak net, and part of air will flows upwards due to the squeezing action of windbreak net. after the air passes over the windbreak net, the air current begins to move downwards. comparing this with relative wind velocity of the empty flow field, the relative wind velocity at 0h horizontal distance is dropped significantly. this reveals the wind reduction effect behind the windbreak net with 20% porosity. in addition, at 0h horizontal distance, when the air current closes the windbreak net, part of air will flow through the windbreak net due to the porosity of windbreak net, its squeezing action is smaller, so that the rising phenomenon of air current is not obvious at the vertical distance above 1h of windbreak net. part of air current will flow through the porosity of windbreak net evenly, so the wind velocity still can be measured at the vertical distance behind the windbreak net. proceedings of engineering and technology innovation, vol. 11, 2019, pp. 12-20 17 at 2h, 4h and 6h horizontal distance, the relative wind velocity at 0h~1.0 h vertical distance is 0.20~0.40, 0.26~0.57 and 0.25~0.63 respectively. this reveals under the same horizontal distance, the relative wind velocity behind the windbreak net with 0 o tilt angle and 20% porosity will increase with the increase of vertical distance. at 12h horizontal distance, the relative wind velocity at 0h~1.0 h vertical distance is about 0.33~0.65. behind 12h horizontal distance, the variation curve of relative wind velocity is similar to the horizontal distance at -4h. its relative wind velocity will increase gradually with the increase of horizontal distance. this reveals the wind reduction effect decreases gradually with the increase of horizontal distance, and the windbreak net does not have the wind reduction effect behind 12h horizontal distance. fig. 6(d) illustrates the relative wind velocity of windbreak net with 0 o tilt angle and 50% porosity at every horizontal distance and vertical distance. at 2h horizontal distance, the relative wind velocity at 0h~1.0 h vertical distance is 0.28~0.48. under the same horizontal distance, the relative wind velocity of windbreak net with 50% porosity will increase with the increase of horizontal distance that similar to the windbreak net with 20% porosity. under the same horizontal distance, the relative wind velocity of windbreak net with 50% porosity measured from 0h~1.0 h will be higher than that of windbreak net with 20% porosity. 3.3. the wind reduction effect of windbreak net with different porosity table 1 the relative wind velocity of windbreak net with 0 o tilt angle and different porosity at different horizontal distance and 1h vertical distance the relative wind velocity (uz/uref) porosity (%) 50 40 30 20 horizontal distance measurement (h) 1 2 h 0.48±0.006 a 0.41±0.007 b 0.39±0.015 b 0.38±0.005 b 4 h 0.58±0.005 a 0.47±0.010 b 0.44±0.010 c 0.39±0.014 d 6 h 0.62±0.005 a 0.55±0.005 b 0.51±0.010 c 0.52±0.005 c 8 h 0.63±0.000 a 0.58±0.010 ab 0.53±0.011 b 0.55±0.037 b 10 h 0.68±0.015 a 0.60±0.015 b 0.55±0.005 c 0.54±0.009 c 12 h 0.66±0.006 a 0.62±0.005 b 0.58±0.013 c 0.55±0.005 d 16 h 0.68±0.010 a 0.62±0.000 b 0.60±0.000 bc 0.59±0.010 c 20 h 0.68±0.010 a 0.66±0.020 a 0.63±0.015 b 0.68±0.011 a 1 mean±std. with different superscripts in the same row are significantly different(p<0.05),using the scheffe test. fig. 7 the relative wind velocity of windbreak net with 0 o tilt angle and different porosity at different horizontal distance and 1h vertical distance the relative wind velocity of windbreak net with 0 o tilt angle and different porosity at different horizontal distance and 1h vertical distance is shown in table 1. from the table, it can find out that the relative wind velocity of windbreak net with different porosity is increased progressively with the increase of horizontal distance. at 2h horizontal distance, the relative wind velocity of windbreak net with 20% porosity does not have significant difference with respect to that of 30% and 40%, but have significant difference with respect to that of 50%. the relative wind velocity of the windbreak net with 20%, 30%, 40% and 50% porosity is 0.38, 0.39, 0.41 and 0.48 respectively, and the corresponding wind reduction effect is 62%, 61%, 59% and 52% respectively. at 6h horizontal distance, the relative wind velocity of windbreak net with 20% porosity has significant difference with respect to that of 50% and 40%, but does not have significant difference with respect to that of 30%. the relative wind velocity of the windbreak net with 20% and 30% porosity is 0.52 and 0.51 respectively, and the corresponding wind reduction effect is 48% and 49% respectively. the relative wind velocity of the windbreak net with 40% and 50% porosity is 0.55 and 0.62 respectively, and the corresponding wind reduction effect is 45% and 38% respectively. at proceedings of engineering and technology innovation, vol. 11, 2019, pp. 12-20 18 10h horizontal distance, the relative wind velocity of the windbreak net with 20%, 30%, 40% and 50% porosity is 0.54, 0.55, 0.60 and 0.68 respectively, and the corresponding wind reduction effect is 46%, 45%, 40% and 32% respectively. fig. 7 illustrates the relative wind velocity of windbreak net with 0 o tilt angle and different porosity at different horizontal distance and 1h vertical distance. it can find out that the relative wind velocity of windbreak net with 4 kinds of porosity is decreased significantly from 0h to 6h horizontal distance behind the windbreak net. it reveals that there is good wind reduction effect at the interval of 0h to 6h horizontal distance. at 2h horizontal distance, the windbreak net with 20%, 30% and 40% porosity has higher wind reduction effect. behind 6h horizontal distance, the wind reduction effect of windbreak net with 4 kinds of porosity is decreased progressively with the increase of horizontal distance. the windbreak net with 20% porosity has lower relative wind velocity, which represents its wind reduction effect is better. for the windbreak net with 50% porosity, the relative wind velocity behind 6h horizontal distance 6h is higher than that of 0h horizontal distance. then the higher relative wind velocity is maintained, so its wind reduction effect is poorer. 3.4. the wind reduction effect of windbreak net with the different tilt angle the relative wind velocity of windbreak net with 20% porosity and different tilt angle at different horizontal distance and 1h vertical distance is shown in table 2.it can find out that the relative wind velocity of windbreak net with different tilt angle is increased progressively with the increase of horizontal distance. at 2h and 4h horizontal distance, the relative wind velocity is 0.38, 0.37, 0.37, 0.37 and 0.38, 0.39, 0.40, 0.38 respectively for the windbreak net with 0 o , 5 o , 10 o , 15 o tilt angle. it reveals that the relative wind velocity of windbreak net with different tilt angle does not have significant difference each other. it means that the tilt angle does not influence the wind reduction effect of windbreak net within 4h horizontal distance behind the windbreak net. at 6h and 8h horizontal distance, the relative wind velocity of windbreak net with 0 o and 15 o tilt angle has significant difference each other, and the windbreak net with 15 o tilt angle has lower relative wind velocity. but behind 10h horizontal distance, the relative wind velocity of windbreak net with different tilt angle does not have significant difference each other. it means that at 10h horizontal distance behind the windbreak net, the tilt angle of windbreak net does not influence the wind reduction effect. table 2 the relative wind velocity of windbreak net with 20% porosity and different tilt angle at different horizontal distance and 1h vertical distance the relative wind velocity (uz/uref) tilt angle（°） 0 5 10 15 horizontal distance measurement (h) 1 2 h 0.38±0.010 a 0.37±0.015 a 0.37±0.020 a 0.37±0.021 a 4 h 0.38±0.005 a 0.39±0.020 a 0.40±0.015 a 0.38±0.010 a 6 h 0.51±0.011 a 0.49±0.015 ab 0.44±0.020 c 0.45±0.011 bc 8 h 0.52±0.005 a 0.51±0.005 ab 0.49±0.011 bc 0.48±0.010 c 10 h 0.54±0.010 ab 0.55±0.015 a 0.51±0.010 b 0.52±0.005 ab 12 h 0.54±0.000 a 0.54±0.011 a 0.55±0.028 a 0.56±0.010 a 16 h 0.59±0.020 a 0.58±0.015 a 0.62±0.023 a 0.60±0.010 a 20 h 0.67±0.020 a 0.64±0.005 a 0.66±0.032 a 0.62±0.011 a 1 mean±std. with different superscripts in the same row are significantly different(p<0.05),using the scheffe test. fig. 8 the relative wind velocity of windbreak net with 20% porosity and different tilt angle at different horizontal distance and 1h vertical distance proceedings of engineering and technology innovation, vol. 11, 2019, pp. 12-20 19 fig. 8 illustrates the relative wind velocity of windbreak net with 20% porosity and different tilt angle at different horizontal distance and 1h vertical distance. the windbreak net has lower relative wind velocity at the interval of 2h to 4h horizontal distance. however, the change of relative wind velocity is very small with the increase of tilt angle of windbreak net. it means that the influence of tilt angle of windbreak net on relative wind velocity is very low, namely the increase of tilt angle of windbreak net does not have much influence on the wind reduction effect. 4. conclusions erecting the windbreak net in front of the simple greenhouse can reduce the wind velocity. it can suitably protect the simple greenhouse and fruits and vegetables and reduce the damage of the crops from strong wind. the following results can be obtained by carrying on the wind tunnel simulation test of the windbreak net. for the windbreak net with 0 o tilt angle, at 1h vertical distance and 2h horizontal distance, the relative wind velocity of windbreak net with 20% porosity does not have significant difference with respect to that of 30% and 40%, but have significant difference with respect to that of 50%. the relative wind velocity of the windbreak net with 20%, 30%, 40% and 50% porosity is 0.38, 0.39, 0.41 and 0.48 respectively, and the corresponding wind reduction effect is 62%, 61%, 59% and 52% respectively. at 1h vertical distance and 6h horizontal distance, the relative wind velocity of the windbreak net with 20%, 30%, 40% and 50% porosity is 0.52, 0.51, 0.55 and 0.62 respectively, and the corresponding wind reduction effect is 48%, 49%, 45% and 38% respectively. at 1h vertical distance and 10h horizontal distance, the relative wind velocity of the windbreak net with 20%, 30%, 40% and 50% porosity is 0.54, 0.55, 0.60 and 0.68 respectively, and the corresponding wind reduction effect is 46%, 45%, 40% and 32% respectively. the increase of tilt angle of windbreak net does not have much influence on the wind reduction effect. conflicts of interest the authors declare no conflict of interest. references [1] l. m. huang, y. z. jiang, and s. m. kao, “study on anti-wind strength for the material of windbreak net pillar,” journal of agriculture and forestry, vol. 40, no. 2, pp. 11-28, june 1991. [2] y. s. tian, z. h. xiao, and j. y. zhang, “application of windbreak net and accommodation at the adverse circumstance of strong wind,” monthly agricultural magazine at taichung area, vol. 145, september 2011. [3] q. tang, y. sien, and b. h. li, “study on anti-wind distance for nylon windbreak net of sweetsop orchard at coastal region of taitung,” crop, environment and biological information, vol. 5, pp. 40-51, 2008. [4] y. z. jiang and l. m. huang, “study on the shelter effect for the erecting type of windbreak net,” journal of agriculture and forestry, vol. 39, no. 1, pp. 15-56, may 1990. [5] y. z. jiang and l. m. huang, “study on the optimal interval of windbreak net,” journal of agriculture and forestry, vol. 39, no. 2, pp. 121-144, june 1990. [6] l. m. huang, j. s. li, m. w. liu, and t. h. cheng, “study on wind tunnel test for the shelter effect of weaving type of windbreak net,” journal of agriculture and forestry, vol. 61, no. 1, pp. 65-83, may 2012. [7] j. l. santiago, f. martín, a. cuerva, n. bezdenejnykh, and a. sanz-andrés, “experimental and numerical study of wind flow behind windbreaks,” atmospheric environ. vol. 41, no. 30, pp. 6406-6420, september 2007. [8] w. m. cornelis and d. gabriels, “optimal windbreak design for wind-erosion control,” journal of arid environments, vol. 61, no. 2, pp. 315-332, april 2005. [9] z. dong, w. y. luo, g. q. qian, and h. t. wang, “a wind tunnel simulation of the mean velocity fields behind upright porous fences,” agricultural and forest meteorology, vol. 146, no. 1-2, pp. 82-93, september 2007. [10] j, p. bitog, i. b. lee, m. h. shin, s. w. hong, h. s. hwang, i. h. seo, j. i. yoo, k. s. kwon, y. h. kim, and j. w. han, “numerical simulation of an array of fences in saemangeum reclaimed land,” atmospheric environment, vol. 43, no. 30, pp. 4612-4621, september 2009. [11] m. du and t. maki, “effects of polyethylene windbreak net on drifting sand and microclimate,” j. agric. meteorol, vol. 52, no. 5, pp. 953-956, january 1997. proceedings of engineering and technology innovation, vol. 11, 2019, pp. 12-20 20 [12] j. r. brandle and h. d. nickerson, “windbreaks for snow management,” school of natural resources, paper 125, january 1996. 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/).  proceedings of engineering and technology innovation , vol. 4, 2016, pp. 49 51 49 using ground-penetrating radar to promote the investigating efficie ncy in mud pumping disaste r of railways chihping kuo 1,* , chihhao hsu 1,2 , yenyu chen 3 , chinwei wu 1 , hongyi wang 1 , dewei chen 1 , yixuan lin 1 1 department and institute of civil engineering and environmental informatics, minghsin university of science and technology, hsinchu, taiwan. 2 taipei engineering division, taiwan railways administration, motc, new taipei city, taiwan. 3 samho technology & engineering co., ltd, new taipei city, taiwan. received 20 january 2016; received in revised form 15 february 2016; accept ed 03 march 2016 abstract currently about 85% of the railway structure is constructed traditionally in taiwan, which means the foundation of railways is composed by in-situ soil materials and covered by ballast, sleepers and tracks. the rail is continued by fis hplates and then bolted. while train passes here, the deflection caused by repeated loads. the repeatedly force transfers through ballast to the saturated foundation, may create vacuum to draw phenomenon, called pump effect or mud pu mping. it could lead to serious train derailment capsized. present mud pumping detection method has to be performed during the non-operating time at night by visual. however, this approach may have omissions and shortcomings perspective concerns, and slow to find disasters during the rainy season. using non-destructive testing techniques (ground-penetrating radar) to inspect the quality of rail bed is widely in foreign. however, d etecting the distribution of mud beneath rail is an attractive subject here. by the usage of mention technology in this research, this technology is expected to be promoted. this study has been agreed by the official administration, taiwan railway bureau, to be carried out in nanwan branch in hsinchu by using ground-penetrating radar. co mparing with the visual inspection results, the mud pumping can be verified. where most serious mud pumping phenomenon observed was open to prove the function. the study is expected to launch the road bed structure further rehabilitation plans and preventative ma intenance engineering. furthermore, cost due to misjudgment is expected to be saved and traffic safety improved. keywor ds : mud pumping, railway bed, ground-penetrating radar, disaster 1. introduction 1.1. phenomena and effect of railway-bed mud-pumping the general structure of ra ilway bed is layout in fig. 1. the loading fro m tra in transfers to the subsoil of g round through rails, sleepers, ballasts , and subballasts. ballast plays important role in the loading transferring work. the loading has to be uniformly and equally transferred by ballasts. the rainfa ll requires to be drained rapidly by ballasts as well. when the ballasts were filled by subballasts , it would reduce the function of ballasts. subballasts generally is used for adjudging the required elevation of rails and composed by mixed soils. in most cases , fine materia ls like silts or clays are more welcome than granulated ones like sands or small gravels due to the budgets. fig. 2 shows the common phenomena observed in railway. the mud covers the sleepers and ballasts as pumping fro m subballasts . a serious issue is after the pumping occurred; the materia ls did not flow back to subballasts. therefore the cavities exist inside the subballasts and induce the differentia l settlements for sleepers and rails. a potential disaster consequently threatens the safety of train’s traffic. inspecting by naked vision is the traditional and now-used method for the inspectors in taiwan ra ilway bureau in ta iwan. besides, they can only discover the cases that already pumped out the ballasts. the distributions and * corresponding aut hor. email: picnic.kuo@must.edu.tw proceedings of engineering and technology innovation , vol. 4, 2016, pp. 49 51 50 copyright © taeti shapes of pumping or cavit ies are also difficult and impossible to be discovered in this way as illustrated in fig. 3. a more effic ient and convenient method has to be developed and proposed in this paper. fig. 1 general structure of railway bed fig. 2 mud pumping observed in the neiwan branch line in hsinchu (2016.01.05) fig. 3 how pu mped materia ls distribute and form inside the subballasts requires more efficient method to be investigated and inspected 2. method ground-penetrating radar (gpr) is a geophysical method that uses radar pulses to image the subsurface. this non-destructive method uses electromagnetic radiat ion in the mic rowave band (uhf/ vhf frequencies) of the radio spectrum, and detects the reflected signals from subsurface structures. gpr can have applica tions in a variety of media, including rock, soil, ice, fresh water, pavements and structures. a gpr transmitter e mits electromagnetic energy into the ground. when the energy encounters a buried object or a boundary between materials having different permittivit ies, it may be re flected or refracted or scattered back to the su rface. a receiv ing antenna can then record the variations in the return signal. dry sandy soils or massive dry materia ls such as granite, limestone, and concrete tend to be resistive rather than conductive, and the depth of penetration could be up to 15m. in moist and/or clay -laden soils and materia ls with high e lectrica l conductivity, penetration may be as little as a few centimetres. ground-penetrating radar antennas are generally in contact with the ground for the strongest signal strength. [1] generally gpr are adopted to check the elevation of subballasts and subsoils. however, the mothed is not popular in ta iwan. in this study, a gssi 400mhz antenna with sir-20 server were adopted to scan a 40m long railway. this section was selected for varies kinds of mud pu mping occurrences observed here. fig.4 shows the inspection work o f pu mping pe rformed by the authors. fig. 4 inspection for mud pumping in the neiwan branch line in hsinchu (2016.01.05) 3. results and discussion the results from gpr scan shows the distributions and shapes of pumped materia ls inside the ballasts and subballasts . the blue line indicates the boundaries of pumped materials . co mparing with the pu mping materia l surface that has pumped out the ballasts, the distribution of observed by vision and gpr are co incided. the depth of pumping material boundaries were also identical and proved after opening the ballasts as shown in fig. 6. using gpr to scan and inspect mud pumping is firstly performed in taiwan. the result shows the boundaries of pumping materials can be estimated. 縱向斷面 a c b section of traffic direction 橫向斷面 a c cross section hill-shape??? mushroom-shape??? proceedings of engineering and technology innovation , vol. 4, 2016, pp. 49 51 51 copyright © taeti fig. 5 the pattern of gpr scans and indicates the distributions and shapes of pumped materials inside the ballasts fig. 6 the boundaries of pumping materials were found by opening the ballasts acknowledgement financia l support fro m min istry of science and technology (most 104-2622-m-159-001 -cc3) is much apprec iated. study area was provided by taiwan ra ilways administration, mot c. re lated technique was provided by sa mho technology & engineering co., ltd. their kindly helps are very much appreciated. references [1] “ground-penetrating radar,” https://en.wikiped ia.org/wiki/ground-penetrating_radar, may 26, 2016.  proceedings of engineering and technology innovation, vol. 1, 2015, pp. 40 44 40 cop y right © taeti an aco-based routing algorithm for multimedia cdn networks jian-bo chen * , yui-lin wang department of information and telecommunications engineering, ming chuan university, taoyuan county, taiwan . received 19 march 2015; received in revised form 20 april 2015; accept ed 01 may 2015 abstract multimedia content delivery networks is used to distribute mu ltimed ia contents from orig in server to many replica servers , wh ich can reduce the loads and can improve the availability. the client request must be red irected to the most appropriate one a mong these replica servers. in this paper, we proposed ant colony optimiza tion (aco) based routing algorith m to solve this problem. the ants in aco-based routing algorithm will re lease pheromone when they go through the path. the routing decision is based on the cumulat ive phero mone of the path. the aco-based routing algorithm not only can find the most appropriate path but also can suit for dynamic t opology. when the topology changes frequently, the aco-based routing algorith m will a lways find a opt imized path. the simu lation results show that th e aco-based routing algorithm can achieve higher perfo rmance than the other mechanisms . keywor ds : content delivery network, ant colony optimization 1. introduction recently, the streaming mult imed ia contents play an important role in the internet. the popular mu ltimed ia web sites, such as youtube, have millions of mu ltimed ia contents could be accessible fro m the internet. these web sites usually do not put all the multimedia contents in a single server or a single p lace to serve the clients. instead, they distributed these multimedia contents into some replica servers in other places around the world. in this kind of network arch itecture, the loading of the single server can be shared by other servers. the architecture is also known as content delivery networks (cdn) architecture. the cdn arch itecture distributes the mu ltimed ia contents fro m the origin server to many replica servers, which can reduce the loading and can improve the availab ility. when a c lient issues a request to specific mu ltimed ia content, the cdn architecture will red irect this request to the most appropriate replica server. then this replica server serves the client request instead of the origin server, which can achieve better performance in the cdn networks. the bandwidth usage can be min imized, the cpu loading of servers can be reduced, and the connection number of specific link can be decreased. in other wo rds, the cdn architecture is especially suitable for mu ltimed ia content distribution and can reduce the user perceive latency. the cdn a rchitecture consists of several parts: origin server, replica servers, accounting system, distribution system, and request routing system [1-6]. the distribution system is used to distribute the mult imedia contents from origin server to other replica servers. the accounting system is in charge of the usage of mult imedia contents. in this paper, our focus is on the request routing system. the request routing system [7] includes two sub-system, server location system and request routing system. the server location sub-system is used to find the most appropriate replica server. when there are several replica servers contain the requested mult imed ia content, how to find the most appropriate server is the challenge of server location sub-system. once the most appropriate rep lica server is determined, how to redirect the client request to this re plica server is another challenge of request routing sub-system. in this paper, we will use ant colony optimization based routing algorith m (a ra) to solve the request routing problem. in the server location sub -system, the most appropriate replica server can be located according to the pheromone. the more phero mone the path contains, the better replica server can be determined. in the request routing sub-system, when the c lient passes through a sp ecific path, the phero mone will be reta ined in this path. all the pheromone will be accu mulated in this path. based on this ara algorith m, the most appropriate replica server will a lways be selected when the path contains more and more pheromone. the ara algorith m not only can find the best path but also can adapt with the dynamic routing environ ment. when the network topology changes or the link congests frequently, the ara algorithm can dynamica lly adjust its routing policy based on the amount of pheromone. the ara algorith m sends a lot of packets as artific ial ants to each replica server and stores the amount of pheromone for each path. the routing is decided by the phe romone function and a random value to generate a probability value. the replica servers send back the artific ial ants with their own status. when the artific ial ants go back to the origin server, the status information o f each replica server will be updated. in the streaming mult imed ia cdn arch itecture, when a client connects to a replica server and streams a multimedia content, the bandwidth of the path, the loading of the server and the connection number of the server will be changed. if the capacity, such as bandwidth availability, cannot afford to the c lient request, the ara algorith m will automatica lly chose another path to find out the other appropriate rep lica server. therefore , the a ra algorith m can enhance the accuracy of server locate and can improve the availability of servers. this paper is organized as follo ws. we will introduce the cdn architecture in sect ion 2 and introduce the aco in section 3. the ara algorith m will be described in section 4. our proposed method will be introduced in section 5. the simu lation results are shown in section 6. section 7 is the conclusions . *corresponding aut hor. email: jbchen@mail.mcu.edu.tw proceedings of engineering and technology innovation, vol. 1, 2015, pp. 40 44 41 cop y right © taeti 2. cdn architecture the cdn network is a content caching mechanism with low cost and high reliability. it can increase the efficiency of data transmission and reduce the loading of the origin server. the cdn can also provide service to clients around the world. the client request is served by the closest replica server which contains the requested multimedia contents. in general, the architecture of cdn consists of several parts, including client, replica servers, origin server, accounting system, distribution system and request routing system. the system architecture of cdn is shown in fig. 1. fig. 1 the cdn system architecture the request routing system has two important functions, server location and request routing. the function of server location is to determine the most appropriate replica server after receiving a request from client. the most appropriate replica server must have enough ability to serve the client and must have the multimedia content the client request. there are two mechanisms to obtain the capacity of the server [8]. one is server push mechanism [9, 10], which is an active way to collect the capacity of the server. the replica servers send their server capacities to the agent that manages all replica servers at a regular interval. the agents will summarize all the capacities and store them to the database. the other method is agent probe [11], which is a passive mechanism that an agent uses to query the capacities of servers at regular interval. the processes of cdn include several steps. figure 2 shows the procedures which is described as follows. step 1: the o rig in server sends an entry of mu ltimed ia content information to cdn by request routing system. step 2: the origin server t ransfers the mu ltimed ia content to distribution system. then the distribution system decides wh ich replica servers will store mu ltimedia content. step 3: the distribution system transfers the multimed ia content to the replica servers. step 4: the c lient sends a request to the request routing system. step 5: the request routing system determines the suitable replica server, and redirects the request to the specific replica server. step 6: the replica server ma kes a connection to the client and starts the streaming service. the design requirements and design constraints are su mmarized based on the characteristics of the mechanism. 2.1. design requirements (1) it has one degree of freedom, and therefore it has one input. (2) it is a planar mechanism with four links or more. (3) it has at least one cam pair to generate a non -uniform output speed. (4) it has at least one gear pair to change the uniform output speed. (5) it has a ground link to support or constrain other links. (6) it has one output link. request routing system orgin serverclient replica server distribution system 1 3 236 5 4 fig. 2 the cdn operation processes 3. aco algorithm the aco is a swarm intelligent algorithm [12] that simulates the processes of ants exploring environment and foraging for food. it is used to solve the optimization problem of static allocation. this algorithm was proposed by m. dorigo in 1992[13]. the aco uses the pheromone between ants to commun icate with each other. when ants reach a node, this refers to that the pheromone level of the node is more than the other node. the node contains higher pheromone level; the ants have higher probability to select it. the aco uses the feature of positive feedback [14] to solve the optimization problem. it also uses a random function to make the result variable. the characteristics of aco algorithm includes: (1) ants tend to choose the path with higher pheromones. (2) for shorter paths, the pheromones cumulative speed is more quickly. (3) ants communicate with each other indirect by release and stack the pheromones. nest 1 2 n-1 n 1 2 n-1 n ... ... 1 2 n-1 n 1 2 n-1 n ... ... . . . . . . . . . stage 1 stage 2 . . . stage m-1 stage m destination fig 3. the searching graph of aco proceedings of engineering and technology innovation, vol. 1, 2015, pp. 40 44 42 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti figure 3 shows the searching graph of aco. there are m stages from the nest of ants to destination. each stage will choose a value from n values by the pheromone function. the pheromone function will refer to the result of previous iterations and a random value to compute a probability value and decide the result of stage. after several iterations, the processes output a set of results with m values. 4. aco routing algorithm the ant colony optimization routing algorithm (ara) [15] is an enhanced algorithm from the ant colony optimization algorithm used for routing process. assume that the source node is treated as the nest of ants, and the destination node is treated as the food. 1 2 3 4 5 fig 4. the example of network topology the simple network topology is shown in fig. 4. in this e xa mple , we assume that node 1 is the nest of the ants and node 5 is the destination node. the ara algorith m is the processes of how the ants find the path from nest to destination. for instance, when an ant goes from node 1 to node 3, the pheromone in this lin k will be updated. if more ants pass through the link, then higher pheromone leve l will be accu mulated on this lin k. as a result, more ants will select this link as the most appropriate link. table 1 shows an e xa mple of the pheromone level for node 3. each node ma intains a table o f phero mone leve l to the destination nodes. this is also known as the routing table. in the routing table of node 3, if the destination is node 2, there are three neighbors which a re node 1, node 4, and node 5. the pheromone levels for these three neighbors are 0.4, 0.5, and 0.3. when the ant wants to select next node, it will choose the node with highest pheromone level as the ne xt node. in this example, it is node 4. table 1 the pheromone table of node 3 neighbor node d e st in a ti o n n o d e 1 4 5 1 0.5 0.3 0.2 2 0.4 0.5 0.3 4 0.2 0.5 0.3 5 0.2 0.3 0.5 5. proposed method this paper is focused on the aco-based routing algorith m for mu ltimed ia cdn network. the majo r objective is to enhance the accuracy of server location and to reduce the computation time in route decision. this algorith m also can solve the redirection fa ilu re proble m when the network is congested [16]. in cdn network, the request routing system is to select the most appropriate replica server and to redirect this client request to the replica server. assume that the replica server is chosen, but during the redirection process, the client cannot connect to the replica server. we ca ll this kind of situation as the red irection failure. we use the positive feedback from a ra algorithm to enhance the accuracy of server location calculation, and ara algorithm ma intains the updated data to increase the availability of servers. the process of ara algorith m includes six co mponents [17-19]. figure 5 shows the flow diagra m of our proposed method. the processes are co nstruction of the network topology, initia lizat ion, tour construction, data response and pheromone update . construction of the network topology initialization tour construction data response pheromone update reach server? no yes startstart fig. 5 the flow diagram of propose method the first step is to construct the network topology. before e xecuting the ara a lgorith m, the network topo logy, denote as g (n,e), must be constructed, where n is the set of nodes including origin server and replica servers, and e is the set of links between each node. after the step of network topology construction, each node must be initialized. that is, setting the initial pheromone levels and weights on each node. the initial pheromone levels and weights are determined by the management policy and corresponding situation. before this process, the ara algorithm sends lots of ants to construct the tours and to cumulate the pheromone level on each node. these ants select the forwarding path randomly. the ara algorith m uses the pheromone level and random value to generate a probability value. when these ants select the next nodes, each node will update its own pheromone table. the pheromone table records the pheromone level of each link that connects to this node. when these ants pass through a node and select a link, the pheromone level of this link will be updated and stored into the pheromone table. however, the pheromone level will count to infinite, so the pheromone levels are decreased over time. when an ant reaches to the replica server, the replica server will send back an ant to the origin server. the back ant carries the current status info rmation of the replica server. the back path is decided by the pheromone level so that this ant passes through the router directly, no calculation are needed. when the back ants reach to the origin server, the origin server updates the information what they carry. to keep the information newest, our proposed algorithm will be run every few minute. proceedings of engineering and technology innovation, vol. 1, 2015, pp. 40 44 43 cop y right © taeti the pheromone is the most important mechanism in ant colony optimizat ion algorith m. it allows the ants able to communicate with each other wh ile they arrive in d ifferent time. the prior ants store the results in nodes, and the posterior ants consult the results to make decision. in the beginning of algorithm e xecut ion, the distribution of pheromone is scattered. after some iterat ion, the better path has more phero mone and the pheromone will co ntinually increase. when pheromone level in a path is high enough, almost all ants will select this path and this path is become prima ry path. in this time , we call that the network is converged. although almost all ants select the primary path, there are still some ants selecting the different path because the prima ry path cannot be sure that it is always being the best. these ants try another path and find other feasible path as alternative path. when the primary path is more and more congested, the alternate path will replace the prima ry path. if a node or path failed, the pheromone is reset to zero immed iately. this means that no ants can go through this path. therefore, the ara algorith m can adapt to topology change in a short time. 6. simulation results the network topology used in our simu lation is shown in fig. 6. in this topology, one origin server, five replica servers and four routers are included. the replica servers contain some mult imed ia contents. in this simulation, we duplicate the mult imedia contents to replica server ra ndomly. the bandwidth between routers and replica servers is 100mbps, the bandwidth between routers is 1gbps, and the bandwidth between origin server and router is 100mbps. the bit rate and service t ime for mu ltimed ia contents is shown in table 2. for e xa mp le, the bit rate (transmission rate) of content2 is 768kbps and the service time (content length) is 700 seconds. the orig in server sends 20 ants to each replica server every 5 minutes. the update informat ion for each replica server inc ludes a timestamp field. when these ants come back fro m replica server to origin server, the origin server will update the newest data by the timestamp fie ld. to avoid routing loop, the ant visits all nodes will be dropped. to keep the better path with higher phero mone level, and to ensure another path with lower pheromone leve l, when an ant passes a node, it will increase 2 phero mone levels and decrease 1 pheromone every 30 second. the client request is distributed by erlangs distribution. the 10% erlangs means that the packet loss is 10%. each kind of netwo rk load is tested 10 t imes and the e xecution time for each test is 10 minutes . origin server replica server 1 replica server 4 replica server 2 replica server 3 replica server 5 content1 content 2 content 3 content1 content 3 content 4 content2 content 3 content 5 content1 content 4 content 5 content2 content 3 content 4 content 5 fig 6. the simulation network topology table 2 the information of multimedia contents bit rate (kbp s) service time (second) content 1 1024 400 content 2 768 700 content 3 1024 800 content 4 512 300 content 5 768 200 we use smpl simu lation tool in this simu lation. smpl is a set of c language functions. the hit rates of redirection with different schemes are co mpared in fig. 7. we compare our proposed ara algorith m with other two methods, routing table polling and network probing. the results show that our proposed ara algorithm has better performance in a congested and dynamic network. fig. 7 comparison of redirection hit rate 7. conclusion in this paper, we propose an aco-based routing algorithm for mu ltimed ia cdn network to enhance the accuracy of server location and the availability of cdn network. in this algorithm, nu merous artific ia l ants are sent to replica servers. the status information of replica servers is then sent back to the origin server to build a database for request routing system. we a lso propose a pheromone function for ants to select the next node in each decision. the a co-based routing algorith m also has the ability for load ba lancing that will reduce the overall drop rate. our simu lation results show that the algorithm can work in a congested network or dynamic network references [1] gang peng, cdn: content distribution network, technical report tr-125, experimental computer systems lab, stony brook university, 2003. [2] novella bartolini, emiliano casalicchio, and salvatore tucci, “a walk through content delivery networks,” lecture notes in co mputer science, vol. 2965, pp. 1-25, 2004. [3] turrini e., “an architecture for content dis tribution internetworking,” dept. of computer science, univ. of bologna, italy, march 2004. [4] pathan m. and buyya r., “a taxonomy and survey of content delivery networks,” univ. of melbourne, dept. of computer science and software engineering, technical report grids-tr-2007-4, australia, february 2007. 70 75 80 85 90 95 100 0% 10% 20% 30% 40% 50%t h e h it r a te o f r e d ir e ct io n (% ) ne twork load (erlangs) ara routing table polling network probing proceedings of engineering and technology innovation, vol. 1, 2015, pp. 40 44 44 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti [5] m. green, b. cain, g. tomlinson, and s. thomas , “cdn peering architectural overview,” network working group, internet-draft, november 2000. [6] m. day, b. cain, g. tomlinson, and p. rzewski, “a model for content internetworking,” network working group, category: informational, february 2003. [7] md. humayun kabir, eric g. manning, and gholamali c. shoja, “request-routing trends and techniques in content distribution network,” parallel, networking, distributed applications laboratory, dept. of computer science, univ. of victoria, canada, 2008. [8] james d. guyton and michael f. schwartz, “locating nearby copies of replicated internet servers,” proc. of applications, technologies, architectures, and protocols for computer communication, pp. 288-298, august 1995. [9] david r. boggs, internet broadcasting, ph.d. thesis, technical report csl-83-3, xerox palo alto research center, october 1993. [10] craig partridge, trevor mendez, and walter milliken, host anycasting service, ietf rfc 1546, november 1993. [11] e. w. m. wong and s. chan, “modeling of video-on-demand networks with server selection,” proc. global telecommunications, ieee, nov. 1998, pp. 8-12. [12] e. bonabeau, m. dorigo, and g. theraulaz, “swarm intelligence – from natural to artificial systems,” santa fe institute studies in the sciences of complexity, vol. 14, pp. 163-164, 2002. [13] dorigo, m. and ga mbardella l.m., “ant colony system: a cooperative learning approach to the traveling salesman problem,” ieee t rans. evolutionary computation, vol. 1, pp. 53-66, 1997. [14] dorigo, m., maniezzo, v. and colomi, a., “positive feedback as a search strategy,” technical report, no. 91-016, politecnico di milano, italy, 1991. [15] mesut g ünes, udo sorges, and imed bouazizi, “ara the ant-colony based routing algorithm for manets,” proc. parallel processing workshops, vancouver, bc, canada, august 2002. [16] gustavo sousa pavani and helio waldman, “traffic engineering and restoration in optical packet switching networks by means of ant colony optimization,” proc. broadband communications, networks and systems , san jose, california, october 2006. [17] eslam al maghayreh, “salam abu al-haija, faisal alkhateeb, and shadi aljawarneh, bees ants based routing algorithm,” proc. intelligent systems, modelling and simulation, liverpool, uk, january 2010. [18] dongming zhao, liang luo, and kai zhang, “an improved ant colony optimization for the communication network routing problem,” proc. bio-inspired computing, beijing, china, october 2009. [19] alireza abbasy, and seyed hamid hosseini, “ant colony optimization-based approach to optimal reactive power dispatch: a comparison of various ant systems,” proc. ieee pes powerafrica, johannesburg, south africa, july 2007.  proceedings of engineering and technology innovation , vol. 4, 2016, pp. 25 27 25 proactive management of water quality in aquifer storage transfer and recovery sang-il lee, hyon wook ji * department of civil and environmental engineering, dongguk university, seoul, south korea. received 18 february 2016; received in revised form 16 march 2016; accept ed 09 april 2016 abstract this paper reports the status of an ongoing effo rt to ensure the quality of water produced at an aquife r storage transfe r and recovery (ast r) site in south ko rea. ast r is one of the managed aquifer recha rge techn iques in which water is intentiona lly re charged to suit able aqu ife rs for subsequent recovery. surface water injected to an aquifer is converted into water of drin kab le qua lity by natu ral p rocess. in an ast r pro ject being developed in the densely populated de lta area of the nakdong river basin, a ne w fra me work is be ing imp le mented in wh ich water safety is addressed through the analysis and control of bio logica l, chemica l, and physica l ha zards fro m intake , pretreat ment, in jection and re covery, to d istribution and consumption of wate r. the task is done by the incorporation of ha zard analysis and critica l control po int (ha ccp) to the ast r process. originated fro m the food in dustry, ha ccp identifies, evaluates, and co n trols ha za rds to ensure the safety of th e product. princ iples o f ha ccp we re applied to the ast r water p roduction system. ha zard ana lysis revealed that most haza rdous events were due to the chemica ls transported in the basin. risk assessment was conducted to identify the sources of che micals wh ich pose a potentia l threat to the wate r quality o f the ast r site . such in formation will be va luable fo r the pro active manage ment of che mica ls at the basin scale. crit ica l control po ints (ccp) were d e termined by the decision tree method, and crit ica l limits fo r wate r qua lity pa ra mete rs at ccps were established subsequently . keywor ds : aquifer storage transfer and recovery, haccp, water quality, nakdong river, risk analysis 1. introduction the nakdong river , one of the four ma jor rivers in south korea, is a main source of drinking water for mo re than ten million people. the water quality of the na kdong river has been deteriorated due to both routine emission fro m cit ies and water pollution incidents [1]. according to a consumer survey in south korea, 32.5% of respondents do not drink tap water because they have concerns about raw water quality [2]. hence, there e xist imminent needs to improve drinking water quality and to protect raw water fro m pollution. to s atisfy these needs, two technologies are combined. first, aquifer storage transfer and recovery (astr) can be an alternative for quality improvement in drinking water production [3]. astr is a method to store surface water in an aquifer through an injection well and to recover the water through an extraction well (fig. 1). a pilot plant is being developed in the delta of the nakdong river. fig. 1 the process of drin king water supply using astr second, ha zard analysis and crit ical co ntrol point (ha ccp) is incorporated to ensure the water quality throughout the drinking water production process . haccp is a tool to assess hazards and establish control systems to guarantee the safety of food. recently, world health organization (who) and many countries including u.s and australia tend to adopt haccp * corresponding aut hor, email: mousehw@naver.com proceedings of engineering and technology innovation , vol. 4, 2016, pp. 25 27 26 copyright © taeti for drinking water supply [4]. 2. method the pilot study strictly fo llo ws the seven principles of haccp [4]. first, a co mprehensive list was created of all microbia l, che mical and physical haza rds having the potential to cause harm [5]. a unique feature of this study is that we conducted the risk analysis to obtain info rmation for proactive manage ment of chemical haza rds in the entire basin. potential risk can be e xpressed as a function of the amount of t ransfers, distance, and toxicity as in eq. 1. potential risk =f(amount of chemical transfers, distance, toxicity) (1) since the data of the a mount of transfers were insuffic ient for probabilistic ana lysis, synthetic data were generated by using the stochastic method. the generated data we re fitted to probability distributions. the potential risk for each c ity can be calculated and the most risky city becomes the target of special monitoring and manage ment to reduce chemical risk posing on the astr site. ‘determin ing the critica l control points ’ is a key issue in haccp [5]. ccp is a ma jor process to prevent and treat hazards efficiently. ccps were determined by the decision tree method in which four questions were asked [6]. then, critical limits for water quality parameters at ccp were established based on the literature and the legal regulations . critical limits can be considered as the minimu m baseline for which hazards do not affect human health. 3. results and discussion total 114 haza rdous events were identified fro m the entire process in producing drinking water using ast r. various chemical, physical and microbio logical ha zards re lated to those hazardous events exist. out of 114, nine ma jor hazardous events were particularly screened ; seven out of nine events originated fro m the catchment area. a ll these events involved chemical haza rds. if the haza rdous chemica l events occurring in the basin could be properly managed, the risk to the drinking water supply can be significantly reduced. the potential risk for 42 c ities in the basin was analyzed for the period of 2001 to 2012. as for the probability density function needed for synthetic data generation, the weibull d istribution (2-para meter) was the most frequently chosen, while the norma l and the gu mbe l distribution followed. the city of yangsan turned out to be the most potentially-risky city in terms of chemical ha zards . it means that if a plan to mitigate che mical ha zards in the na kdong river basin is to be imp le mented, yangsan should be the first city. ten processes for drinking water supply using astr were subject to the decision tree analysis . ccps turned out to be catchment area , pre-treat ment, in jection, treat ment, storage of treated water, and distribution to consumers . the process not classified as ccp is po int of attention (poa) [7]. a lthough its importance is relative ly lowe r, proper attention should be given to poas. critica l limits fo r water quality para meters at ccps we re developed to monitor and control microbial, chemical and physical hazards. 4. conclusions integration of astr and ha ccp was imple mented to ensure water quality on drin king water supply in a densely populated area. natural filt ration of low-quality surface water in an aquifer can enhance public trust in water supply among consumers. in addition, the applicat ion of haccp will reduce the risk and eventually improve the negative recognition on drin king water. although the pilot study is in its early phase, its influence to the water supply system in this region cannot be neglected. acknowledgement this research was supported by basic science research program through the national research foundation of korea (nrf) funded by the ministry of education (2015r1d1a1a 09060690). references [1] j. k. lee , t. o kim, and y. j. jung, “analysis of domestic water pollution accident and response management (in korean),” journal o f wetlands research, vo l. 15, no. 4, pp. 529-534, 2013. proceedings of engineering and technology innovation , vol. 4, 2016, pp. 25 27 27 copyright © taeti [2] s. h. cho, y. j. lee , and h. b. yoon, “korean’s perception and attitudes regarding their tap water (in korean),” korean public hea lth research, vol. 32, no. 2, pp. 130-140, 2006. [3] p. dillon, “ future manage ment of aquife r recharge,” hydrogeology journal, vol. 13, pp. 313-316, 2005. [4] a. tavasolifar, b. bina, m. m. a min, a. ebrah imi, and m. jala li, “imp le mentation of hazard analysis and critical control points in the drinking water supply system,” international journal of env iron mental health engineering, vol. 1, no. 3, pp. 1-7, 2013. [5] a. h. havelaar, “applicat ion of ha ccp to drinking water supply,” food control, vo l. 5, no. 3, pp. 145-152, 1994. [6] code x (code x alimentarius co mmission), “haza rd analysis and crit ical control po int (ha ccp) system and gu idelines for its application.” annex to cac/ rcp 1-1969, rev. 3, 1997. [7] t. dewettinc k, e. van houtte, d. geenens, k. van hege, and w. verstraete, “ha ccp (ha zard analysis and critica l control points) to guarantee safe water reuse and drinking water production a case study,” water science & technology, vol. 43, no. 12, pp. 31-38, 2001.  proceedings of engineering and technology innovation , vol. 4, 2016, pp. 16 18 16 photoluminescence study of hydrothermally grown zno nanostructures wei-you chen 1 , jiunn-chyi lee 2 , hung-pin hsu 1 , ya-fen wu 1,* 1 department of electronic engineering, ming chi university of technology, new taipei city, taiwan. 2 department of electrica l engineering, ta ipei city university of science and technology, ta ipei, ta iwan . received 15 january 2016; received in revised form 27 february 2016; accept ed 06 march 2016 abstract in th is work, the hydrotherma lly gro wn zno nanostructures were successfully prepa red using the aqueous solution of zinc nitride d e hydrate (zn(no3)2 ). the effect o f solution mo la rity on the opt ica l and structura l prope rties of zno nanostructures were studied by te mperature dependent photolu minescence (pl) measure ments. the intensity rat io of the u ltra violet e mission to the visib le e mission is calculated. it is found that the calcu lated results are a lso dependent on the solution mo larity and consistent with the c rystal qua lity of zn o nanostructures . ke ywor ds : zinc o xide, hydrotherma l, precursor concentration, photoluminescence 1. introduction zinc o xide (zn o) has been e xtensive ly in vestigated due to its unique propert ies such as wide direct bandgap energy (3.36 e v), la rge e xc iton binding ene rgy (60 me v), and therma l stability. these characte ristics ma ke th is mate ria l interesting for many app lications such as solar ce lls, optoelect ronic and nanowave de vices [1-3]. the ro le of the nanostructure size , impu rit ies , and mo rphology is very important to these applicat ions. various methods have been utilized to synthesize zno nanostructures, such as meta l-o rganic che mical vapor depos ition (m ocvd), sol-gel methods, and the lo w tempe rature hydrotherma l method [4-5]. the hydrothermal method has been widely used to zno structures on zn o-seeded substrate because of its advantages includ ing lo w-cost, low-te mpe rature , nanoe lectron ic co mpatib le , and suitable for la rge area substrates [6]. the properties o f zno produced by the hydrothermal method a re dependent upon p reparat ion para mete rs such as growth te mperatu re, pre cursor concentration, and gro wth time [7]. in this work, the hydrotherma lly gro wn zn o nanostructures were prepared with diffe rent aqueous solution mo la rities and diffe rent dep osition t ime . the effects of solut ion mo la rity on the optica l and structura l prope rties of zn o nanostructures were studied by te mperatu re dependent pl measure ments. it is found that the pl e mission intensity inc reases and then decreases with the increas ing solut ion mo la rity. 2. experiment a thin film of zinc acetate dihydrate (zn(ch3coo)2∙ 2h2o) was spin coated onto an indiu m-t in-o xide (ito) glass substrate and annealed at 300 °c in o xygen for 20 min. the hydrotherma l method was used to grow zno nanostructures on ito glass using an aqueous solution of zinc nitrate (zn(no3)2) and ammo nium hydro xide (nh3oh). the solution mo larity increases from 10 to 50 mm . pl measurements we re performed to e xa mine the structural and optical properties of the zno nanostructures. the te mperature dependent pl spectra we re measured using a he-cd laser operating at a wavelength of 325 n m and the average exc it ation intensity was 30 mw. the samples were mounted in a closed-cycle he cryostat where the temperature (t) was varied from 10 to 300 k. 3. results and discussion fig. 1 shows the xrd patterns of zn o nanostructures fabricated with solution-mola rity 20 mm and deposition time 1 hour. the detected xrd pea ks are characteristic of the zno with hexagonal wurt zite type crystalline structure. the strongest (002) pea k indicates that the fabricated zno is grown along the c-axis direction. * corresponding aut hor. email: yfwu@mail.mcut .edu.tw proceedings of engineering and technology innovation , vol. 4, 2016, pp. 16 18 17 copyright © taeti fig. 1 xrd patterns of zno nanostructures fig. 2 pl spectrum of zno nanostructures measured at room temperature the room te mperature pl spectrum of this sample is shown in fig. 2. observing the measured result, the pl spectrum shows a strong yellow-orange emission peak at ~610 n m in the visible region and a wea k pea k in the ultrav iolet (uv) reg ion at ~390 n m. room-te mperature pl spectra of zn o usually e xhib it a broad peak in the visible region. the origin of this e mission peak can be attributed to the defect in zno, such as oxygen vacancies located in zno surfaces [8]. the uv peak is attributed to free-e xc iton recombination corresponding to the near band-edge transition in zno. fig. 3 visib le e mission of pl spectra for the zno nanostructures with various solution molarities fig. 3 presents the visible e mission of pl spectra for the sa mples prepared at deposition time 3 hours, with various solution mola rit ies. it is clearly seen that the emission intensity increases as the solution mo larity increased fro m 10 to 30 mm, and then decreases as the solution mo larity is higher than 30 mm. we suppose that the increase in emission intensity could be associated with the increasing concentration of the oxygen vacancies [9]. the obtained results are similar for the sample with 1-hour deposition time. the relative pl intensity ratio of the emission in the uv region (iuv) to the emission in the visible region (ivis) is shown in fig. 4. it can be seen that the ratio increases with the molarit ies for the sa mples, and decreases as solution molarity greater than 50 mm. the iuvivis is one of the main factors that usually used to compare the optical properties of samples [10]. the higher pl intensity ratio means fewer structural defects in the sa mples. too higher solution mola rity leads to the disorder of zno structures , and deteriorate the crystal quality. fig. 4 variation of iuvivis values with different solution molarities 4. conclusions in this paper, the hydrothermally grown zno nanostructures were prepared with different aqueous solution mola rit ies and diffe rent dep osition time. the te mperature dependent pl measure ments were carried out to investigate the effects of solution molarit ies on the optical and structural properties of zno nanostructures . it is found that the visible emission intensity of pl spectra increases and then decreases with solution mola rit ies. moreover, the pl intensity ratio of the emission in the uv reg ion to the emission in the visib le region is calculated . the values of ratio increase as the mo larity varying fro m 10 to proceedings of engineering and technology innovation , vol. 4, 2016, pp. 16 18 18 copyright © taeti 40 mm , and decrease at mola rity 50 mm. it is concluded that the optical property and crystal quality of zno nanostructures are improved by an appropriately increasing solution molarity. however, too higher solution molarity leads to the disorder of zno structures, and deteriorate the crystal quality. references [1] w. water and s. y. chu, “physical and structural properties of zno sputtered films ,” mater. lett., vol. 55, pp. 67-72, 2002. [2] r. ferro, j. a. rodríguez, and p. bertrand, “peculia rit ies of nitrogen dio xide detection with sprayed undoped and indium-doped zinc oxide thin films,” thin so lid films , vol. 516, pp. 2225-2230, 2008. [3] t. szabó, j. németh, and i. dékány, “z inc oxide nanoparticles incorporated in u ltrathin layer silicate films and their photocatalytic properties,” co llo ids surf. a : physicochem. eng. aspects , vol. 230, pp. 23-35, 2004. [4] l. spanhel and m. a. anderson, “semiconductor clusters in the sol-gel process: quantized aggregation, gelation, and crystal growth in concentrated zno colloids,” j. am. che m. soc., vol. 113, pp. 2826-2833, 1991. [5] s. fay, l. fe itknecht, r. schlüchter, u. kroll, e. vallat-sauvain, and a. shah, “rough zno layers by lp-cvd process and their effect in improv ing performances of a morphous and microcrystalline silicon solar cells,” solar energy mater. solar ce lls, vol. 90, pp. 2960-2967, 2006. [6] h. zhang, d. yang, y. ji, x. ma, j. xu, and d. que, “ low te mperature synthesis of flowe r-like zno nanostructures by cetyltrimethyla mmon iu m bro mide-assisted hydrotherma l process,” j. phys. che m. b. , vol. 108, pp. 3955-3958, 2004. [7] e. pál, v. hornok, a. oszkó, and i. dékány, “hydrotherma l synthesis of prism-like and flowe r-like zn o and indium-doped zno structures ,” collo ids surf. a: physicochem. eng. aspects, vol. 340, pp. 1-9, 2009. [8] g. z. xing, b. yao, c. x. cong, t. yang, y. p. xie, b. h. li, and d. z. shen, “ effect of annealing on conductivity behavior of u ndoped zinc o xide prepared by rf magnetron sputtering,” j. alloys co mpd., vol. 457, pp. 36-41, 2008. [9] m. ghosh and a. k. raychaudhuri, “shape transition in zno nanostructures and its effect on blue-green photoluminescence,” nanotechnology, vol. 19, pp. 1-7, 2008.  proceedings of engineering and technology innovation, vol. 6, 2017, pp. 32 35 analysis of the correlation between the support performance of supported business for the regional industry dong-gu lee, je-doo ryu, keon-seok nam, kyoung-nam ha * korea institute of industrial technology, busan, korea. received 26 july 2017; received in revised form 07 august 2017; accept ed 17 august 2017 abstract reflecting the conditions and characteristics of korea, the region's main industry is being developed to expand new job creation in the region and enhance the competitiveness of local industries. the local main business is supported to 63 main industries selected by region. the main industries in busan include digital contents, intelligent machine parts, super precision convergence parts, mold heat treatment, and bio -health. in this study, the main industries of busan region in 2016 conducted an analysis on the companies that support the intelligent machine parts industry. for the analysis, we collected data on the sales of supported companies, sales after business support, and employment growth. the analysis tool utilized ibm spss statistics 18. as a result, we confirmed the relationship between employment and sales according to support business programs. based on the results of this study, it will be helpful for the task performing organization to identify the direction of the support program and the needs of the enterprise. keywords: supported business for the regional industry, correlation analysis, intelligent, mechanical parts 1. introduction since the 1960s, when the republic of korea spurred economic development, the economic scale and export scale as of 2016 have developed dramatically to ten world economic powers as of 2016, with rapid growth. in this process, the role of the government's ongoing r&d investment and support was important. in 2017, the size of r&d investment in the government accounted for 4.7% of the government budget at 19 trillion won, and it has increased by 2.5% each year. such government effor ts include support projects to support the growth of smes. protection and development of small and medium enterprises support policy for small and medium enterprises turned into active direction. policy efforts are being made to create innovative econ omies based on the improvement of technological innovation of smes . (castells & hall, 1994:westhead & storcy, 1994 : martin & scott, 2000). the republic of korea government formulated regional industry promotion policies and made efforts to develop small and medium enterprises. in 2013, the government integrated regional industry promotion plans, which were classified into strategic projects and municipal specialized industries, into a new region specialized industrial development policy. by doing so, we tried to expand local employment creation and maximize sales growth of regional companies. business support projects other than the regional specialized industrial development business intensively support the technical support, the commercialization support , the human resource development, etc. of specialized industrial fields selected for e ach region, to expand regional employment * corresponding author. e-m ail address: 0vincent@ kitech.re.kr proceedings of engineering and technology innovation, vol. 6, 2017, pp. 32 35 copyright © taeti 33 creation and to promote regional employment we are aiming to revitalize regional economies by increasing sales and other factors. in this study, we would like to analyze the enterprise support project of the busan area specialized industry in 2015 and analyze what kind of correlation the business support project has with the employment of the company and the increase in sale s. 2. pre vious re s e arch many researches have been made recently that present the direction of d esirable enterprise support projects through outcome analysis of existing enterprise support projects. lee, kim(2016) conducted an analysis of the efficiency of enterprise support project for the whole-south area specialized industry development business, and kim, won(2015) conducted an influential factor analysis on the outcome of support of enterprises by industry of regional industry promotion policy . lee jun-soo (2013) analyzed the problems of the enterprise support service project through research on improvement efficiency and improvement measures of the busan regional company support service project and presented a remedial measure. gim jaegun (2013) analyzed the satisfaction of tenant enterprises of corporate support services through analysis of th e functions and roles of techno park to support smes . 3. analysis of correlation in this study, pearson correlation analysis was conducted on the results of the “comprehensive package support project for intelligent machine parts industry”, a regional main industry development business enterprise support project conducted in 2015. in the prototype production support business of the intelligent machine parts industry, we selected the demand enterprise support program through the early enterprise demand survey. table 1 shows program frequency analysis. results of demand investigation ware requested in order of upgrading of products, supporting prototype production, certification, and patent support. the field of commercialization support was requested in order of b rand development support, exhibition support, marketing support, network support, and so on . using the ibm spss statistics 18, we conducted a frequency analysis of 138 selected companies. frequency analysis was conducted to confirm the frequency of programs meeting the needs of the demand enterprises and the results are shown in table 1. table 1 program frequency analysis program frequency percent effective percent cumulative percent domestic exhibition 32 23.2 23.2 23.2 brand 21 15.2 15.2 38.4 m arket analy sis 4 2.9 2.9 41.3 bottleneck technique 16 11.6 11.6 52.9 patent ap p lication 14 10.1 10.1 63.0 product quality 7 5.1 5.1 68.1 prototy p e 12 8.7 8.7 76.8 product p lanning 6 4.3 4.3 81.2 foreign exhibition 4 2.9 2.9 84.1 certified sup p ort 5 3.6 3.6 87.7 test analy sis 5 3.6 3.6 91.3 consulting 8 5.8 5.8 97.1 m arketing 4 2.9 2.9 100.0 sum 138 100.0 100.0 proceedings of engineering and technology innovation, vol. 6, 2017, pp. 32 35 copyright © taeti 34 as a result of the frequency analysis, the domestic exhibition support program was supported in order of the highest frequency 32 times support, brand support 21 times, bottleneck technical support 16 times, patent application 14 times, prototype 12 times, and so on. also, table 2 shows the distribution of support project costs, the highest among 108 companies with the highest under 10 million won. this seems to be due to support of domestic and foreign exhibitions, support for brands, bottlenecks technical support less than ten million won . table 2 frequency analysis of support project cost sup p ort p roject cost(million won) frequency less than 10 108.0 m ore than 10~ less than 20 13.0 m ore than 20~ less than 30 8.0 m ore than 30~ less than 40 3.0 m ore than 40~ less than 50 5.0 m ore than 50 1.0 table 3 also shows that employment with 117 programs equivalent to 85% was not triggered by 138 suppo rt programs for new employment. the creation of one or two employment occupied the highest proportion table 3 new employment frequency percent effective percent cumulative percent 0 117 84.8 84.8 84.8 1 9 6.5 6.5 91.3 2 8 5.8 5.8 97.1 3 2 1.4 1.4 98.6 4 1 .7 .7 99.3 5 1 .7 .7 100.0 sum 138 100.0 100.0 table 4 shows the results of descriptive statistical analysis, the average cost of corporate support was 83 million won, and the average sales of domestic commercialization was confirmed at 130 million won. table 4 descriptive statistics average standard deviation n sup p ort p roject cost (thousand won) 8300.02 10657.012 138 new emp loy ment .29 .803 138 sales (domestic) in commercialization 13658.12 53982.765 138 sales (overseas) in commercialization 1305.36 6692.743 138 table 5 descriptive statistics support project cost (t housand won) new employment sales of commercializat ion (domest ic) sales of commercializat ion (overseas) support project cost (t housand won) p earson correlat ion coefficient 1 . 416 ** .066 .016 significant probabilit y .000 .440 .849 n 138 138 138 138 new employment p earson correlat ion coefficient .416 ** 1 .118 -.009 significant probabilit y .000 .168 .915 n 138 138 138 138 sales of commercializat ion (domest ic) p earson correlat ion coefficient .066 .118 1 .453 ** significant probabilit y .440 .168 .000 n 138 138 138 138 sales of commercializat ion (overseas) p earson correlat ion coefficient .016 -.009 .453 ** 1 significant probabilit y .849 .915 .000 n 138 138 138 138 **. please note that the correlation coefficient is 0.01 level (both sides). proceedings of engineering and technology innovation, vol. 6, 2017, pp. 32 35 copyright © taeti 35 table 5 shows the results of correlation analysis. through correlation analysis, we analyzed the correlation between support project cost and new employment, sales (domestic) in commercialization, sales (overseas) in commercialization. the project cost shows that the correlation between new employment (r = 0.416) and quantity (+) is commercialized, while sales (domestic) shows correlation between sales (overseas) and commercial quantity (+) it was analyzed. 4. conclusions the results of empirical analysis of this study are as follows. first of all, the support cost of a company and new employment have a slightly high correlation, and it seems that the influence of direct and indire ct employment of a company is influenced according to the company support amount. but as a result of the frequency analysis of the overall employment, it is found tha t significantly lower employment has been carried out. in other words, it can be confirmed that there is a limit to increasing employment for project cost support. secondly, it is generally judged that there is a correlation between the support cost of a company and the sales (domestic) of commercialization. in fact, sales of commercialization were analyzed as having no significant correlation with the support cost of enterprises. in fact, the support cost for companies supporting prototype production is relatively high. however, there is a point that it is hard to cooperate with sales immediately. it is judged that it is desirable to evaluate prototype performance and sales collaboration rather than sales, as a result of prototype production support. thirdly, the sales of domestic commercialization and the sales of overseas commercialization hav e a slightly high correlation. it turns out that the sales of domestic commercialization and the sales of overseas commercialization positively have mutual influence. therefore, we can see how much employment is affected by corporate support project cost. however, employment frequency is low. there is also no significant correlation between corporate support project cost and sales of commercialization. when choosing the evaluation item of the enterprise support project, it is not simple employment and sales , it is necessary to find and present the result item desired by the company. in addition, it is judged that it is desirable to plan and evaluate to the consumer focus, not the manufacturer's center. references [1] s. s. lee and y. k. kim, “efficiency analysis of enterprise support project for nurturing specialized industries in jeonnam area,” 2016 korean institute of industrial engineers , symposium, pp. 34-40, november 2016. [2] y. h. won and g. h. kim, “an analysis on performance influence factors of enterprise s upport based upon the local industrial development policy: focused on the industry types of region -specialized industries,” korean association for policy science, june 2015. [3] s. w. lee, g. j. park, and j. d. kim, “correlation analysis on the performance of supported business for the robot in dustry,” korean society of precision engineering, may 2016. [4] j. k. kim, “evaluating roles of techno-park as a policy tool to support regio nal smes,” journal of the korean urban management association, vol. 26, no. 1, pp. 1-28, march 2013. [5] w. j. shim and e. s. kim, “korean national research and development project, multifaceted analysis of the output ratio of research expenses by government,” korea technology innovation society , march 2010.  proceedings of engineering and technology innovation, vol. 12, 2019, pp. 21-25 the basic study of underwater robot control for over actuated systems keon seok nam, dong gu lee, je doo ryu, kyoung nam ha * robot center, korea institute of industrial technology, busan, korea received 14 may 2018; received in revised form 30 may 2018; accepted 16 july 2018 abstract in general, only six individual thrusters are required to perform 6 dof (degree of freedom) motion. sometimes, however, more than six thrusters are used for 6 dof motion for a variety of purposes. the over actuated systems can transit quickly from transient state caused by disturbance to steady state through the over actuated input. so in the underwater robot, the purpose of over actuated system is mainly used to maintain stable motion and position. and that systems are performed to obtain the optimal efficiency through the thruster force distribution. in this paper, we designed and fabrication the rov (remotely operated vehicle) for the basic study of over actuated underwater robot. it has 4 horizontal thrusters and 4 vertical thrusters. using this system, we derived the thruster force distribution matrix for thruster allocation. also, to check the kinetic characteristics of over actuated underwater robot before applying various controller, we performed the basic motion performance test and motion control test using the pd controller in the indoor engineering water basin. keywords: over actuated system, rov, motion control, 6 dof 1. introduction various types of underwater robots are being used for exploration, inspection and operation in underwater environments. for efficient exploration, inspection, and operation, it is necessary to develop and use equipment that is suitable for the purpose. generally used underwater robot are divided into auv (autonomous underwater vehicle) capable of autonomous operation and rov (remotely operated vehicle) operated by operators. the auv shown in fig. 1, it is capable of performing specific tasks through autonomous navigation without having a separate tether and operator, but it can not acquire real-time captured images and underwater exploration information. the rov shown in fig. 2, it is operated by the operator through the tether in the separate surface ship, and has the feature to check the images and data acquired in real time. therefore, rov is generally used for inspection, construction and maintain for underwater structure. when an underwater robot is used to carry out specific tasks, one operator may operate the entire underwater robot, but this is very limited in various underwater tasks. it is mainly divided into two operators. one operate the underwater robot body and the other operators the underwater robot's manipulator. in the task, there are many cases in which the underwater robot body must maintain a specific position and attitude, and the operator can directly control the motion. however, because of the disturbance occurring in the underwater environment such as the current, it is difficult for unskilled operators to react immediately. in order to control the 6 dof motion of a general underwater robot, it is possible to control the translational and rotational motion of each axis by using six thrusters. however, there are cases where an over actuated system is constructed using several thrusters so that the motion control can be efficiently performed even if disturbance acts in the underwater environment. * corresponding author. e-mail address: 0vincent@kitech.re.kr proceedings of engineering and technology innovation, vol. 12, 2019, pp. 21-25 22 in this paper, we design rov of an over actuated system that controls 6 dof motion by using a number of the thruster, and it is a basic study to implement auto pilot function that can help operator 's convenience by using it. the configuration of the paper consists of the design and configuration of the underwater robot, the coordinate system setting of the underwater robot, the design of the attitude controller, the attitude control test in the engineering water basin, and the conclusion. fig. 1 torpedo type auv [1] fig. 2 observation class rov [2] 2. design and configuration (a) the outline of underwater robot (b) the thrusters and pressure vessels arrangement fig. 3 illustration of designed underwater robot fig. 4 the actual underwater robot the underwater robot consists of the sensors, external frame, inner frame, and pressure vessel. the shape of the underwater robot is designed as a droplet shape to reduce the resistance of the fluid. in addition, to minimize the weight and buoyant material size, carbon fiber was used except for the pressure vessel. the external shape of the underwater robot shown proceedings of engineering and technology innovation, vol. 12, 2019, pp. 21-25 23 in fig. 3(a). designed underwater robot, the maximum operational depth is 300 m and operation depth is 200 m. as shown in fig. 3(b), four vertical thrusters and four horizontal thrusters are used, and the arrangement of the vertical thruster is installed symmetrically on the left and right, front and rear. the horizontal thrusters are arranged at 45 degree angles to obtain the vector type thrust and two each are installed symmetrically on the front and rear section [3]. the types of sensors attached are dvl (doppler velocity log), ahrs (attitude and heading reference system), 2d image sonar, and hd-camera. the actual underwater robot is shown in fig. 4 3. coordinate system in order to present the attitude and position of the underwater robot, we set the coordinate system as in fig. 5. we used two kinds of coordinate system. one is the body fixed coordinate system and the other is earth fixed coordinate system. the origin of the body fixed coordinate system is located in the geometric center of the underwater robot body. the front of underwater robot body direction is defined as the x-axis and depth direction is defined as the z-axis. the y-axis direction is determined by the right-handed coordinate system. in addition, the earth fixed the coordinate system for expressing the attitude and position of the underwater robot is a ned coordinate system in which the north axis is the x-axis, the water depth direction is the z-axis, and the east axis is the y-axis [4]. fig. 5 coordinate system of underwater robot 4. motion controller the controller of over actuated system achieves the control purpose by finding an efficient thrust distribution method. generally, for thrust distribution, a thrust allocation matrix is used. it can be obtained via kinematic analysis. in this study, as a basic study before the thrust allocation control of over actuated system, the pitch angle and the depth position control were performed using symmetric thrusters to confirm the dynamic characteristic. the controller was used a pd controller and the control formula is as follows (1) [5]. d ( ) ( ) p de u t k e t k dt   (1) 5. test in the engineering water basin (a) wave basin (b) current generating basin fig. 6 the engineering water basin proceedings of engineering and technology innovation, vol. 12, 2019, pp. 21-25 24 the experiment was carried out in the engineering water basin to confirm the motion characteristics of the designed underwater robot. it was carried out at the marine robot center in the korea institute of industrial technology. the dimension of the wave basin is 5 m length, 2 m width, and 9 m depth. also, the dimension of the current generating basin is 22 m length, 3.5 m width and 3 m deep. the actual engineering water basin is shown in fig. 6. in case of inspection and operation of the general underwater robot, they usually work hovering at the specific depth. in order to confirm the motion characteristics at the hovering motion, using only the 4 vertical thrusters, the pitch angle and depth control were performed as shown in fig. 7. the pitch angle was controlled by two forward and rear thruster among the four vertical thruster, and the depth control was carried out using two thruster on the left and right. the desired value of the pitch angle control was set to 20 degrees, and the desired value of the depth control was set to maintain at 1 m. fig. 8 shows the attitude angle (roll, pitch, yaw) of the underwater robot, which is changed due to the rotation of each axis. also, fig. 9 shows the position in the depth direction. as can be seen in fig. 8 (b), it can be confirmed that the attitude was maintained about -17 degrees at 10 seconds. also, as shown in fig. 9, although the depth of 1 m was maintained, oscillation occurs at about 10 seconds, and after reaching the desired depth at 20 seconds, it is maintained about 0.8 m. fig. 7 the motion test of underwater robot (a) roll angle (b) pitch angle (c) yaw angle fig. 8 the attitude of underwater robot fig. 9 the z-position of underwater robot 6. conclusions we designed and fabricated an underwater robot used a number of the thruster for vector thrust. also, pitch and depth controls were performed to verify the dynamic characteristic before applying the thrust allocation matrix for the thrust proceedings of engineering and technology innovation, vol. 12, 2019, pp. 21-25 25 distribution of the multiple thrusters to generate the vector thrust. as shown in the previous section, the pitch control shows a difference of about 3 degrees from the desired value, and the depth control is maintained between 0.7 m and 1.0 m. this is expected to result in a steady state error resulting from the i controller being excluded from the pd controller. also, it can be seen that the oscillation occurring at about 10 seconds during the depth control is caused by the rapid change of the pitch angle, which appears to occur by using independent controllers that do not consider the relationship between pitch angle and depth control. in the future, it is planned to set the constraints for optimal control of the over actuated system using multiple thrusters, and to perform attitude / position control using the thrust allocation matrix. conflicts of interest the authors declare no conflict of interest. references [1] b. he, b. r. wang, t. h. yan, and y. y. han, “a distributed parallel motion control for the multi-thruster autonomous underwater vehicle,” mechanics based design of structures and machines, vol. 41, no. 2, pp. 236-257, february 2013. [2] r. d. christ and r. l. wemli sr, the rov manual:a user guide for remotely operated vehicles, 2nd ed. butterworth-heinemann, 2013. [3] t. i. fossen, t. a. johansen, and t. perez, a survey of control allocation methods for underwater vehicles, intech, pp. 109-128, 2009. [4] t. i. fossen, guidance and control of ocean vechicles, new york: john wiley & sons inc, september 1994. [5] d. h. kim, redundancy resolution and robust controller design for dynamic stability enhancement of underwater vehicle-manipulator systems, doctorate thesis, korea, graduate school of department of mechanical engineering, 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/).  proceedings of engineering and technology innovation, vol. 9, 2018, pp. 44 48 mf integrator: a module for converting metafont into outline font ammar ul hassan 1 , jaeyoung choi 1,* , geunho jeong 2 1 school of computer science and engineering, soongsil university, korea. 2 gensol soft, 99 jangseungbaegi-ro, dongjak-gu, seoul 06936, korea. received 12 march 2018; received in revised form 17 may 2018; accepted 25 may 2018 abstract metafont is a description language that generates tex-oriented bitmap fonts which are not popular outside the tex world. the number of fonts prepared with metafont are very small compared with outline fonts. font libraries like freetype, which are used to render fonts in outline and some bitmap formats doesn’t provide any support for metafont. a research, mfconfig module added an indirect support of metafont inside freetype rasterizer but it has problems of performance and dependency. in this paper, we have proposed a module named mf integrator for freetype rasterizer. the proposed module overcome the problems of performance and dependency of mfconfig module. authors tested proposed module with a client application which is used to display styled text on screen. users can now take the benefits of metafont by using our proposed mf integrator module. keywords: metafont, freetype, fontconfig, font processing 1. introduction fonts are the graphical representation of text in a specific style and size. these fonts are mainly categorized in two types; outline fonts and bitmap fonts. for creating a new font style for outline font, font designers have to design a new file every time. metafont is a programming language introduced by d. e. knuth [1], which allows users to generate different font styles easily. metafont files are different from outline font files. metafont file consists of functions for drawing characters and has parameters for different font styles. by changing the parameter values, various font styles can easily be generated. however, users are unable to use metafont on their pcs because the current font system doesn’t provide any direct use of metafont and tex bitmap fonts such as gf and pk fonts [2]. as it was developed in 1980’s, the pc was not fast enough to provide a real-time conversion of metafont fonts into its corresponding outline fonts. to use metafont in font engines like freetype [3], users need to convert metafont into its corresponding outline font. mfconfig [4] is a plug-in module of fontconfig [5] which provides users a way of using metafont inside current font system. this plugin-in module of fontconfig provides font information to freetype engine. it allows users to render metafont styled characters on screen. users can easily change parameter values and display various styled text using mfconfig module. even though by using mfconfig module variety of font styles can be generated from metafont, it has its own problems. the main problems of mfconfig module include performance and dependability. as it is plugged into the high-level of font system and not directly plugged in freetype engine, it makes its performance poor. it is also dependent on fontconfig library, therefore if there is no fontconfig in font environment this module cannot work . * corresponding author. e-mail address: choi@ssu.ac.kr proceedings of engineering and technology innovation, vol. 9, 2018, pp. 44 48 copyright © taeti 45 in this paper, mf integrator module is proposed which directly converts metafont into its corresponding outline font freetype engine. this module is implemented directly inside freetype so it has no dependency on other high-level font libraries like fontconfig and xft [7]. this module has overcome the problems of mfconfig module. by using mf integrator module, the conversion of metafont into its corresponding outline font is unnecessary. the mf integrator module has a better performance and no dependability issues as compared with mfconfig module. 2. mfconfig module and issues mfconfig is a plug-in module for fontconfig. it enables the use of metafont format on linux font system. it mainly consists of three sub-layers; communication layer, management layer, and conversion layer. communication layer is the starting point of mfconfig. it provides a medium for interaction between mfconfig and fontconfig. management layer is responsible for tasks like searching and other operations. and conversation layer is responsible for converting metafont file into its corresponding outline font file. fig. 1 shows the architecture of mfconfig module linked with fontconfig. fig. 1 mfconfig architecture linked with fontconfig although mfconfig supports metafont in linux font system, it still has its own problems. these problems are mainly related to performance and dependability. as mfconfig is plugged in the high-level of font system i.e. fontconfig and not directly plugged inside freetype, it makes its performance very slow compared to the font-specific driver modules supported by freetype. whenever a client application inputs a metafont file for rendering styled text, fontconfig communicates with mfconfig, performs operations, and then sends input to freetype for rendering text. this whole process becomes slow because of the high-level operations before freetype receives its input. other than the performance problem, mfconfig also has a dependability problem. as it is dependent on fontconfig library, this means that if there is a font environment where there is no fontconfig, then this module cannot be used. fontconfig is mainly used in linux font system for locating fonts and no other operating systems so, mfconfig cannot be supported in them. the proposed mf integrator module has tried to resolve these two problems of mfconfig. it can be used for processing metafont files directly inside freetype as like any other driver modules like truetype driver module, type1 driver module etc. 3. mf integrator module mf integrator module mainly consists of 3 sub-modules; linker module, administrator module, and transformation module as shown in fig. 2. linker module is the initial point of mf integrator module. any metafont file request to proceedings of engineering and technology innovation, vol. 9, 2018, pp. 44 48 copyright © taeti 46 freetype library is dealt by this module. linker module is mainly liable for linking metafont request to mf integrator module. the module provides a two-way linking, i.e. it links the metafont file request to mf integrator module and returns its respective outline font file. administrator module handles all the administrative related tasks like searching if the requested metafont is installed or not. storing the newly created outline font information in a list. this list indicates whether a font file is already prepared or not. managing tasks like verifying the list if the outline font file for the requested metafont file exists, sending metafont file to transformation module for converting it into outline font file, returning the newly prepared outline font file from transformation module to linker module. fig. 2 mf integrator module architecture the work of mf integrator module is perfectly compatible with any freetype version. this module can work as driver module for managing metafont font which is not supported in freetype by default. whenever a different metafont style is input, mf integrator module can easily display the resulting fonts on screen. as it is directly implemented inside freetype engine, it has a better performance and no dependability compared with mfconfig module 4. testing mf integrator module freetype library supports many font formats like truetype, type1, etc. these format specific font modules are called driver modules. based on the input, freetype decides which driver module to call for further operations. for example, if the input file is type1 format then freetype calls type1 driver module for processing it. freetype doesn’t support a driver module for processing metafont files. for checking the working of this new mf integrator module, a client application is used which is responsible for printing the styled font text on the screen. this client application takes a font file and text to be printed on the screen as input and passes this input to freetype for processing. after processing by freetype this application prints styled text on the screen. for testing that freetype now directly supports metafont file input using mf integrator module, a metafont called computer modern is input to the client application. this font file is passed to mf integrator module using the linker module. linker module then examines the request and passes this metafont file to administrator module. administrator module takes this request, performs management related tasks, and passes the input file to transformation module. transformation module converts this metafont input file to its corresponding outline font file. this newly created font file proceedings of engineering and technology innovation, vol. 9, 2018, pp. 44 48 copyright © taeti 47 is stored in the list for future reusability. this outline font file is then passed to freetype core processing layer for further operations. after default operations by freetype, it is returned to client application which displays the corresponding styled text on screen. fig. 3 shows the results of the printing of metafont “computer modern normal font” and “computer modern bold font”. fig. 3 normal and bold styled metafont using mf integrator module the mf integrator module is developed to overcome the problems of mfconfig module. it provides users to display various metafont styled fonts on screens just like any other fonts like truetype etc. by just changing the parameters of metafont source file users can display different styled fonts on screen. as this module is implemented directly inside freetype engine, its performance is very fast compared to the mfconfig module. freetype receives a metafont request from client application and process it directly just like any other default driver modules. mf integrator module is only dependent on freetype engine unlike mfconfig module which was dependent on fontconfig. this module provides reusability and portability features and can be used in any operating system which supports freetype engine. 5. conclusion in this paper, we proposed a module for freetype engine named mf integrator module. this module overcomes the problems like performance and dependability of mfconfig module. mf integrator module provides a direct usage of metafont format as it is implemented directly inside freetype. the research mfconfig module provided a way to use metafont format in current font system. but this module has its own problems; it is plugged in fontconfig and not directly in freetype, therefore if there is no fontconfig library in the operating system, this module cannot be used. mfconfig module performance is also slow because it is implemented in the high-level of font system. we tested proposed module with a client application which is used to display styled text on screen. users can now take the benefits of metafont by using our proposed mf integrator module. future work will consider about proposed module optimization and direct usage of bitmap fonts which is not supported by freetype. acknowledgment this work was supported by institute for information & communications technology promotion (iitp) grant funded by the korean government(msit) (no. r0117-17-0001, technology development project for information, communication, and broadcast). references [1] d. e. knuth, computers & typesetting volume c: the metafont book, 1st ed. boston: addison-wesley professional, 1986. [2] “the gftopk processor,” http://texdoc.net/texmf-dist/doc/generic/knuth/mfware/gftopk.pdf, january 06, 2014. proceedings of engineering and technology innovation, vol. 9, 2018, pp. 44 48 copyright © taeti 48 [3] d. turner, r. wilhelm, and w. lemberg, “freetype”, www.freetype.org, 2000. [4] j. y. choi, s. m. kim, h. j. lee, and g. h. jung, “mfconfig: a metafont plug-in module for freetype rasterizer,” tugboat, vol. 37, no. 2, pp. 163-170, 2016. [5] keith packard, “font configuration and customization for ppen source systems”, https://keithp.com/keithp/talks/guadec2002/fontconfig.pdf [6] k. packard, “the xft font library: architecture and users guide,” proc. the xfree86 technical conference, usenix press, november 2001, pp. 1-17. https://keithp.com/keithp/talks/guadec2002/fontconfig.pdf icati proceedings of engineering and technology innovation, vol. 10, 2018, pp. 42 46 the study of natural nano-composite filter for industrial wastewater treatment chin-ya kuo, hsiao-han liu * department of biological science & technology, i-shou university, kaohsiung 824, taiwan, roc received 02 march 2018; received in revised form 04 june 2018; accepted 24 july 2018 abstract due to the rapid development of industry, the amount of industrial wastewater and water pollution is a serious problem for environment. along with the economic development, the complications of wastewater treatment and drainage have been raised. for examples, textile printing and dyeing wastewater as model system, we utilized natural nano-composite filter adsorption test under different concentrations of dye liquid, different times with nano-composite powder, and observed the solutions with spectrophotometer. the manufacture of natural nano-composite filter was from a variety of mineral flour in accordance with the specific proportion of ore combination, and its inside appearance was like a sponge structure with many nano-sized pores, which increased its adsorption capacity. the production procedure of this natural nano-composite was much simpler than activated carbon, and the failure rate is relatively low, so the cost is lower twice than functional activated carbon. because it is made of natural mineral powder, it will not cause second pollution to the environment. the final results show that methylene blue and no.257 black dye adsorption capacity of natural nano-composite filter was more effective in high temperature 55℃. conclusively, we used the absorption curves to demonstrate the absorption kinetics. keywords: natural nano-composite filter, textile wastewater, industrial wastewater, adsorption test 1. introduction there are many industries that use dyes to stain their products and create a lot of industrial waste. these wastewaters lead to the decrease of transmissibility of rivers and affect the photosynthesis of aquatic organisms. they contain heavy metals and chemical components that promote the toxicity of water and further cause the biological damage to the entire biological chain [1]. a variety of industrial enterprises in the production process of wastewater discharge, including the process of water, machinery and equipment cooling water, flue gas washing water, equipment and site cleaning water and waste production. wastewater contained in the impurities includes waste, residue and some raw materials, semi-finished products, and by-products. the composition is extremely complex and the content of pollutants varies greatly. the strict classification of industrial wastewater is very difficult, because the same type of industry can release several different types of sewage at the same time. what more, sewage can have different substances and different pollution effects. therefore, the industrial wastewater can be divided into two categories according to the ingredients: (1) wastewater containing inorganic substances, including wastewater from metallurgy, building materials and chemical inorganic acid-base * corresponding author. e-mail address: hansliu@isu.edu.tw proceedings of engineering and technology innovation, vol. 10, 2018, pp. 42 46 copyright © taeti 43 production; (2) wastewater containing organic matter, including food industry, petrochemical industry, soymilk, coking, gas, pesticides, plastics, dyes and other factory drainage [2]. this study focuses on the treatment of textile dye wastewater; wastewater dye water quality is complex, according to the characteristics of wastewater and its emission requirements, the selection of appropriate treatment methods. removal of solid impurities and inorganic, coagulation and filtration can be used; decolorization can generally be coagulation and adsorption process consisting of composition; removal of toxic substances or organic matter, the main chemical oxidation, biological and reverse osmosis; removal of heavy metals, available ion exchange method, etc. chemical coagulation: acid, alkali, salt and organic matter of the low concentration of dye wastewater, the first by the average precipitation, adding an appropriate amount of alkali or acid, and then coagulant flocculation and precipitation. biological method: commonly used biological method activated sludge method, biofilm method, oxidation pond method and anaerobic biological method. adsorption method: commonly used adsorbents are activated carbon, activated diatomaceous earth, activated coal, and activated clay, lignite and polymer adsorbents. oxidation method: this method is through a strong oxidation, to destroy the organic matter in wastewater, to achieve a more complete bleaching, detoxification, deodorization and odor purposes [3, 4]. however, so far there is no single method for completely and successfully removing the color from wastewater. adsorption is the simplest method of removal. the most effective and most commonly used adsorbent is activated carbon, but its high manufacturing and recycling costs, for the large-scale use of the business, it is not cost-effective. [5, 6]. natural nano-composite filter is a kind of natural ore, after washing with sulfuric acid and baking at high temperature, it can be processed into granules. for example, by adding natural adhesive can make the powder into filter particles. the appearance of the ore powder is milky white powder, odorless, tasteless, non-toxic, strong adsorption, can absorb colored matter, organic matter; has a larger specific surface area and pore volume, with special adsorption capacity and ion exchange performance, more strong decolorization ability and activity, and good stability after decolorization. the advantage of using natural nano-composite filter is that its selective adsorption capacity, and production costs lower than activated carbon [7, 8]. the dyes used in this study are methylene blue and cationic dyestuffs no. 257 black; cationic dyestuffs does not contain acidic groups in the chemical structure and is cationic in water, also known as cationic dyes. dye is more stable in acidic solution, with the ph value is not stable, and even precipitation or decomposition. the advantages of good coloring effect, you can dye dark, bright colors, variety, etc [9]. 2. materials and methods 2.1. preparation of adsorbent fig. 1 natural nano-composite powder natural nano-composite filter is a kind of powder (fig. 1). in order to test the accuracy of the data, you need to go through the washing steps. weigh 10 g of nano-composite filter in a 50 ml centrifuge tube and add 40 ml of distilled water. mix well and centrifuge for 5 minutes (3000 rpm, 25℃). dispense the supernatant and add 10 ml of distilled water into a glass petri dish and finally oven baked at a high temperature (85℃) overnight. proceedings of engineering and technology innovation, vol. 10, 2018, pp. 42 46 copyright © taeti 44 2.2. dye solution preparation in this study, the cationic dye methylene blue was chosen as the control group. the cationic dyestuffs no. 257 black cationic dye was selected as the control group to compare the practicality of the nano-composite filter. a stock solution of 1000 ppm was obtained by dissolving a precisely weighed dye sample in distilled water. dilute the dye stock solution (1000ppm) to the desired concentration with distilled water if other concentrations are required. all the experiments were without any ph adjustment. 2.3. experimental method 2.3.1. adsorption balance time weigh 1 g washed with water nano-composite filter placed in 50 ml centrifuge tube and then add 50 ml 300 ppm concentration of the dye solution, mix well, the use of 125rpm operation of the temperature control vibrator bottle placed at 25 ± 2℃ , the sampling interval at different times from the shaker. after centrifugation (3000 rpm, 25℃, 5 min), the residual concentration of the dye solution was measured to use an elisa at a wavelength of 660 nm (methylene blue) and 600 nm (cationic dyestuffs no. 257 black), respectively. 2.3.2. effect of ph on adsorption in a series of 50 ml centrifuge tubes were added 300 ppm dye solution, respectively, adjust the ph, methylene blue to adjust the ph 2 ,3 ,4, 6, 7, 8 25d black cationic dyestuffs adjusted to ph 1, 2, 3, 6, 7, 8; and then add 1 g nano-composite filter, after mixing, using a temperature controlled vibrator operating at 125rpm bottle placed at 25 ± 2℃, shaking 3h. after centrifugation (3000 rpm, 25℃, 5 min), the residual concentration of the dye solution was measured at 660 nm and 600 nm using elisa( multislan go version 100.40) , respectively. 2.3.3. effect of different temperature on adsorption add 300 ppm of dye solution to a series of 50 ml centrifuge tubes, add 1 g of nano-composite filter, mix well, and place the bottle at 25℃, 30℃, 35℃ using a temperature controlled shaker operated at 125 rpm, 40℃, 45℃, 50℃, 55℃, shock 3h. after centrifugation (3000 rpm, 25℃, 5 min), the dye solution absorbance was measured at 660 nm and 600 nm by using elisa, respectively. 3. results and discussion the following experimental results for the control group of distilled water, absorbance value is 0.0393. 3.1. adsorption balance time fig. 2 adsorption balance time proceedings of engineering and technology innovation, vol. 10, 2018, pp. 42 46 copyright © taeti 45 according to fig. 2, the absorbance of the supernatant decreases with time and tends to be constant after 3 h. the results show that at room temperature nano-composite filter adsorbed about 3 h equilibrium, and the following experiment selected 3 h adsorption equilibrium times. 3.2. effect of ph on adsorption the initial ph of methylene blue solution is ph5. according to fig. 3, nano-composite filter shows better adsorption when the methylene blue solution ph is 4-6. fig. 3 effect of ph on adsorption by methylene blue fig. 4 effect of ph on adsorption by no.257 black the initial ph of cationic dyestuffs no.257 black solution is ph4. according to fig. 4, the nano-composite filter has the best adsorption when the cationic dyestuffs no.257 black solution is ph 2. the above results indicate that nano-composite filter for different dyes at different ph results are not the same. it will be different for different agents in different ph value of the adsorption effect has changed, for the methylene blue ph 4-6, nano-composite filter adsorption is the best; cationic dyestuffs no.257 black ph 2, nanocomposite filter adsorption effect is also the best, with the ph value increased to 6 the worse the adsorption effect, ph7-8 effect was significantly increased. 3.3. effect of different temperature on adsorption according to fig. 5, the result shows that the adsorption capacity of nano-composite filter increases with temperature after 45 ℃, and the best adsorption temperature is 55 ℃. however, the effect is good at room temperature (25 ℃). methylene blue is significantly more sensitive to temperature than cationic dyestuffs no.257 black. methylene blue has a particularly large line-up. in contrast, cationic dyestuffs no.257 black have different results with temperature, but the line undulations are not as big. fig. 5 effect of different temperature on adsorptio proceedings of engineering and technology innovation, vol. 10, 2018, pp. 42 46 copyright © taeti 46 4. conclusions the results show: (1) nano-composite filter adsorption equilibrium time of 3h. (2) nano-composite filter adsorption in acidic environment is better. (3) nano-composite filter at high temperature (> 45 ℃), the adsorption effect increased significantly. nano-composite filter manufacturing process is simple, the commodity plasticity is very high. in the future can be made into a filter module, the filter module is able to in the water pipe, put it for adsorption, decolorization; its water-insoluble features, can be added to clean products, to achieve a mild exfoliating effect. through another utilization of nano-composite filter adsorption capacity, we could use it into the middle part of a mask to achieve the effect of adsorption of polluted particles. this environmentally friendly material can be used to prevent air pollution of the pm2.5 particles. this product will be the best candidate to prevent pollutions when in the future, the world's water and air are seriously polluted. our next goal will be exploring the possibility of future applications of this nano-composite material in the field of air and environmental pollutions. references [1] m. c. shih, “kinetics of the batch adsorption of methylene blue from aqueous solutions onto rice husk: effect of acid-modified process and dye concentration,” desalination and water treatment, vol. 37, no. 1-3, pp. 200-214, january 2012. [2] a. sonune and r. ghate, “developments in wastewater treatment methods,” desalination, vol. 167, pp. 55-63, 2004. [3] c. i. pearce, j. r. lloyd, and j. t. guthrie, “the removal of colour from textile wastewater using whole bacterial cells: a review,” dyes and pigments, vol. 58, no. 3, pp. 179-196, 2003. [4] t. robinson, b. chandran, and p. nigam, “studies on desorption of individual textile dyes and a synthetic dye effluent from dye-adsorbed agricultural residues using solvents,” bioresource technology, vol. 84, no. 3, pp. 299-301, septemer 2002. [5] m. otero, f. rozada, l. calvo, a. garcıa, and a. moran, “kinetic and equilibrium modelling of the methylene blue removal from solution by adsorbent materials produced from sewage sludges,” biochemical engineering journal, vol. 15, no. 1, pp. 59-68, july 2003. [6] w. x. qi and h. yan, “study on adsorption of methylene blue (mb) on bamboo charcoal [j],” zhejiang chemical industry, vol. 3, p. 009, 2007. [7] d. ghosh and k. g. bhattacharyya, “adsorption of methylene blue on kaolinite,” applied clay science, vol. 20, no. 6, pp. 295-300, 2002. [8] m. s. el-geundi, h. m. ismail, and k. m. e. attyia, “activated clay as an adsorbent for cationic dyestuffs,” adsorption science & technology, vol. 12, no. 2, pp. 109-117, 1995. [9] a. brack and e. schmitt, cationic dyestuffs, u.s. patent no. 3,959,310, 1976.  proceedings of engineering and technology innovation , vol. 4, 2016, pp. 46 48 46 elementary school teachers’ conceptions of teaching mathematics with digital game-based teaching shih-hsuan wei * , jinn-min yang department of mathematics education, national taichung university of education, taichung, taiwan. received 02 may 2016; received in revised form 20 may 2016; accept ed 03 june 2016 abstract this study aimed to investigate the factors accounting for mathematics teaching self-efficacy and to exa mine the relationships among taiwanese teachers’ mathematics teaching self-efficacy, teaching and learning conceptions, technolog ical-pedagogical content knowledge for the digital learning, and attitudes toward game based learning. a total of 195 math teachers from 23 ele mentary schools in taiwan were invited to take part in the study. after ensuring the validity and reliability of each questionnaire, the results indicated that each measure had satisfactory validity and reliability. furthermore, the results revealed that teaching and learning conceptions, technological pedagogical content knowledge for the digital learning, and attitudes toward game-based learning mediated the relationship between teaching and learning conceptions and mathematics teaching self-efficacy, suggesting that knowledge of and attitudes toward digital game-based learning mediated the positive relationship between constructivist conceptions of teaching and learning and outcome expectancy; however, it mediated the negative correlations between traditional conceptions of teaching and learning and teaching efficacy. ke ywor ds : self-e fficacy, tpa ck, digita l ga me based learning 1. introduction with the advance and innovation of techno logy, internet based instruction has gained increasing attention among science educators and researchers in the field of educational technology. the demand for education and learning has been transformed fro m the traditional classroom into digital media pluralis m teaching methods . digital game-based learning has been received a growing attention. the majority of ele mentary schools in taiwan in terms of teaching practice situation are still facing the non-self-discipline e xpertise. in such case, elementary teachers (especially non-mathematical or scientific expertise in those years) whether to hold enough teaching digital game-based teaching self-belief and even whether the impact on their quality of education that is well worth the researchers’ attention. additionally, a strong sense of teaching efficacy could influence effective teaching and student achievement in the field of education [1]. however, it is probably more difficult to prepare the mathematics course units for elementary school teachers because they are required to be responsible for teaching the majority of subjects, including both their majors and non -majors, thus leading to a lower sense of self-efficacy related to teaching mathematics [2-3]. therefore, it is likely that a lower degree of self-efficacy may e xist when it comes to mathematics teaching in elementary schools. in this regard, exploring the sources of mathematics teaching self-efficacy among elementary teachers demands immediate attention. since [4] suggested that the ma jor factors in accounting for self-efficacy a re various kinds of personal e xperiences, teaching and learning conceptions, which are regarded as factors constructed from previous experience, should be taken into account. furthermore, researchers have asserted that internetbased instruction can meet learners’ needs for interactive and individualized lea rning [5-6]. accordingly, internet-based instruction is like ly to promote effec tive teaching, which in turn contributes to teacher self-efficacy. given this, the current study aimed to investigate factors in expla ining the sources of self-efficacy in teaching mathematics with digital game-based environment. * corresponding aut hor. email: scottiewei@mail.ntcu.edu.t w proceedings of engineering and technology innovation , vol. 4, 2016, pp. 46 48 47 2. method 2.1. participants a total of 195 mathe matics teachers (62.3% fe males) we re selected from 23 ele mentary schools in taiwan. of these teachers, 8.6% we re less than 30 years old, 38.3% were 31-40 years old, 45.8 we re 41-50 years old, 4.2% we re 51-60 years old, 1.3% we re more than 60 years old, and 1.8% did not report their age. their teaching e xperience ranged fro m 1 to 38 years, with an average of approximate ly 17 years. a lso, 39.4% of the partic ipants had a bachelor’s d egree, 58.2% had a master’s degree, 1.5% had a ph. d., and 0.9% did not report their degree. a lthough this sample could not be viewed as a national sample, the surveyed teachers came fro m a v ariety of schools in taiwan, across diffe rent demographic a reas and backgrounds, which may be said to be representative in taiwan. in this research, a survey with several d imensions to investigate teachers ’ self-efficacy beliefs towa rd their mathe matica l teaching in terms of mathe matics teaching self-e fficacy, the traditional or constructivist conceptions of teaching and learning, and technology pedagogical content knowledge towa rd ga mes was used. in which, the mathe matics teaching self-efficacy (mtse) scale was revised and translated from the sc ience teaching efficacy be lie f instrument (st ebi) developed by [2]. the teaching and lea rning conceptions (tlc) questionnaire was main ly modeled and translated fro m the scale developed by [7], inc luded two diffe rent conceptions of teaching/ learning: traditional and constructivist. furthermore, to streamline and simp lify the models, the survey also integrating some ite ms fro m tpa ck-g questionnaire, wh ich was developed by [8], was proposed to explore teacher’ technology pedagogical content knowledge toward games . 2.2. data analysis in order to determine the participant teachers, data analyzing with e xploratory factor analysis (efa) was used to reduce the number of ite ms and to ensure the structures of each measure were e mp loyed to test the relationships among teaching and learning conceptions, mathemat ics teaching self-efficacy, and tpck-g. 3. results and discussion table 1 descriptive statistics and correlations descript ive st at ist ics correlat ions mean sd 1 2 3 4 5 6 t lc 1. const ructivist 4.11 0.41 1.00 2. t radit ional mtse 3.86 0.45 0.09 1.00 3. t eaching efficacy 4.23 0.38 0.28** -0.20** 1.00 4. out come expectancy t p ack-g 3.6 0.42 0.21** 0.15 0.33* 1.00 5. t p ack-g 4.02 0.39 0.28** -0.25** 0.41** 0.31** 1.00 6. at t itudes 4.18 0.40 0.23** -0.21** 0.38** 0.22** 0.58** 1.00 not e: t lc, t eaching and learning conceptions; mtse; mathematics t eaching self-efficacy, t pack-g, t echnology pedagogical content knowledge t oward game. ∗p < .05. ∗∗p < .01. table 1 contains means, standard deviations and correlations for the major variables. as can been seen in table 1, constructivist conceptions and the two constructs of mathematics teaching self-efficacy were positively correlated (r = .28, p < .01 for teaching efficacy; r = .21, p < .01 for outcome expectancy), indicating that teachers with higher agreement with constructivist teaching conceptions displayed stronger teaching efficacy beliefs and outcome expectancy in mathematics teaching. also, traditional conceptions and teaching efficacy were negatively correlated (r = −.20 p < .01), indicating that teachers with more traditional teaching conceptions displayed lower teaching efficacy beliefs in mathematics teaching. 4. conclusions the educational and testing systems in taiwan may impede students’ creativity and intrinsic motivation to utilize deep approaches to learning mathematics , decreasing their self-efficacy for proceedings of engineering and technology innovation , vol. 4, 2016, pp. 46 48 48 copyright © taeti learning mathematics. according to the u. s. department of education, “technology is now considered by most educators and parents to be an integral part of providing a high -quality education”. while the use of computer-based technologies is becoming increasingly prevalent, students need to be prepared for the technology-driven workplace with practical skills that can be used in today’s innovative world . teaching and learning are comple x ph enomena. traditional teaching and learning models which focus more on me morization and less on higher level thinking skills are now considered inadequate for the needs of the 21st century classroom. successful mathematics teachers are acknowledged as being those who are equipp ed with adequate knowledge and who are capable of developing appropriate pedagogical strategies to accomplish teaching based on various levels of student prior knowledge. constructivist learning is considered as a dominant paradigm in the fie ld of education. given the importance of educational technology, many educators today find the constructivist approach in a technology-rich environment to be more feasible and effective [9-10]. technology has become an important part of inquiry -based learning. additionally, the tools of technology can contribute to learning environments in support of inquiry-based learning whereby students develop their information-processing and analysis skills through their own learning experience. references [1] r. k. henson, l. r. kogan, and v. h. tammi, “a reliability generalization study of the teacher efficacy scale and related instruments ,” educational and psychological measurement, vol. 61, no.3, pp. 404-420, 2001. [2] i. m. riggs and l. g. enochs, “toward the development of an elementary teacher's science teaching efficacy belief instrument,” science education, vol. 74, no. 6, pp. 625-638, 1990. [3] l. ra mey-gassert, m. g. shroyer, and j. r. staver, “a qualitative study of factors influencing science teaching self‐efficacy of ele mentary level teachers ,” science education, vol. 80, no. 3 pp. 283-315, 1996. [4] a. bandura, “self-efficacy: toward a unifying theory of behaviora l change,” psychological revie w, vol. 84, no. 2, pp. 191, 1977. [5] j. mcca rthy, j. mcca rthy, j. light, and d. mcnaughton, “the effects of internet-based instruction on the social proble m solving of young adults who use augmentative and alternative co mmunicat ion,” augmentative and alternative commun ication, vol. 23, no. 2, pp. 100-112, 2007. [6] w. nuffer and j. duke, “a comparison of live classroom instruction and internet-based lessons for a preparatory training course delivered to 4th year pharmacy students ,” journal of science education and technology, vol. 22, no. 4, pp. 522-528, 2013. [7] k. w. chan and r. g. elliott, “relat ional analysis of personal epistemo logy and conceptions about teaching and learning,” teaching and teacher education , vol. 20, no. 8, pp. 817-831, 2004. [8] c. y. hsu and c. s. chai, “ exp loring preschool teachers’ technological pedagogical content knowledge with educational games ,” workshop proceedings of the 20th inter-national conference on co mputers in education, pp. 302-308, november, 2012. [9] s. ka lyuga and j. sweller, “ rapid dynamic assessment of expertise to imp rove the effic iency of adaptive e-lea rning,” educational technology research and develo pment, vol. 53, no. 3, pp. 83-93, 2005. [10] m. l. connell, “technology in constru ctivist mathe matics classrooms ,” journal of co mputers in mathe matics and science teaching, vol. 17, no. 4, pp. 311-338, 1998.  proceedings of engineering and technology innovation, vol. 8, 2018, pp. 01 08 performance analysis of lms filter using barker and chaotic sequences for radar target detection jami venkata suman * department of ece, gmr institute of technology, rajam, india . received 05 sept ember 2017; received in revised form 24 oct ober 2017; accept ed 17 november 2017 abstract target detection and tracking play a very foremost role in space and underwater scenario. in this paper , we presented a new least mean square based radar target detection (lms-rtd) model is proposed to de-noising of continuous wave radar transmitted signal. in target detection, radio frequency (rf) energy is transmitted to and reflected from the reflecting object which has a status of target information. signal generation and target detection of radar model is designed and simulated using matlab 2017a simulink. the s tudy of performance analysis of signal generation model using binary chaos phase sequences using different chaotic maps, barker codes, combined barker code sequences and the de-noising algorithm using adaptive lms filter is also presented in this paper. implementation of an adaptive lms filter using verilog hdl and its analyses using xilinx fpga tool. in fpga analysis, number of lut’s and used flip flops are improved in proposed lms-rtd method than conventional methods . keywords: radar, chaotic maps, barker codes, matlab simulink, lms-rtd, xilinx fpga, rmse 1. introduction in recent days , radio detection and ranging (radar) has a major significance in the object detection and tracking in space applications. radar is a system used for detecting the presence, direction, distance, and speed of aircraft, ships and other objects by sending out pulses of high frequency electromagnetic waves that are reflected off the object back to the source. mainly it refers to the technique of using radio waves to detect the presence of objects in the atmosphere. the radar signal de-noising is the key function in target detection and target tracking systems. lms algorithm adjusts the filter coefficients to minimize the cost function. it is a class of adaptive filter used to mimic a desired filter by finding the filter coefficients that reveal to producing the least mean square of the error signal. stanford university professor bernard widrow has invented this lms algorithm in 1960. it is a stochastic gradient descent method in that the filter is only adapted based on the error at current time. the block diagram of adaptive lms filter is shown in fig. 1. fig.1 block diagram of adaptive lms filter * corresponding author. e-m ail address: j am i.venkatasum an@ gm ail.com https://en.wikipedia.org/wiki/adaptive_filter https://en.wikipedia.org/wiki/stanford_university https://en.wikipedia.org/wiki/bernard_widrow proceedings of engineering and technology innovation, vol. 8, 2018, pp. 01 08 copyright © taeti 2 where x(n)=input digital signal, y(n)=outp ut digital signal, d(n)=desired response and y e(n)= estimation error or error signal. eventually result can write in the form of three basic relations as follows : (1) filter output: y(n) w(n) * u(n) (1) (2) estimation error or error signal: e y (n) d(n) y(n)  (2) (3) tap weight adaptation: e w(n+1) w(n) mu * u(n) y *(n)  (3) in our surroundings , we have different kinds of signals such as natural, man -made, audio, video, speech, image, communication, geophysical, sonar, radar, medical and musical signals, while many are unwanted signals in a given situation. anything that carries information is a signal and an engineering contest is a function that “conveys information about the behaviour or attributes of some phenomena” and these signals might be necessary or unwanted that depends at that s ituation. signal processing is an operation meant for extracting, enhancing, storing, and transmitting useful information. hence it ten ds to be application dependent. adaptive filters are used in which we don’t have constant filter coefficients and no prio r information is known to us compared to conventional filters. chaos theory is a field of mathematics focused on the non linear deterministic dynamical systems behaviour. chaos is highly sensitive to initial conditions . this effect is popularly referred to as the butterfly effect. it is having applications in several disciplines including physics, engineering, economics, biology and philosophy. a very small disturbance in initial condition produces a drastically different final solution for the chaotic system and the long-term prediction is impossible. this happens even though these systems are deterministic, meaning that their future behaviour is fully determined by their initial conditi ons, with no random elements involved in this process. this behavior is known as deterministic chaos, or simply chaos [1]. a chaotic dynamical system is deterministic, bounded system that exhibits randomness behavior through its initial sensitive conditions. since, in practice initial conditions can never be specified with infinite precision. the random chaos sequences can be generated using various logistic, improved logistic, tent, cubic and quadratic chaotic maps [3-8]. in this paper an attempt is made to generate binary sequences using these maps and their properties at different lengths are studied for improving the performance of radar target detection. pitchaiah proposed the high-speed implementation of distributed arithmetic (da) based architecture for adaptive lms filter. it is not given precious results because of the required time and hardware complexity [9]. 2. lms rtd methodology in the recent days moving target(s) detection and tracking have become challenging. to increase the accuracy of the target detection in the free space environment, we have implemented an efficient de-noising algorithm in this lms–rtd method. in order to de-noise the signal, an algorithm is applied to the received noisy signal so that it can suppress the color noise, which is affected due to environmental disturbances occurred in the free space. t he target detection efficiency is directly proportional to the de-noising algorithm efficiency. in this paper, matlab 2017a simulink is used to model the radar transceiver and free space environment. the implementation of de-noising algorithm used for the lms is done by using verilog-hdl and it is synthesized by using xilinx fpga tool. the fig. 2 represents the block diagram of the proposed methodology, it has a signal generator system includes sinusoidal signal and barker codes, combined barker codes and b inary phase coded chaotic sequences using logistic, improved logistic, https://en.wikipedia.org/wiki/audio_signal https://en.wikipedia.org/wiki/video https://en.wikipedia.org/wiki/image https://en.wikipedia.org/wiki/mathematics https://en.wikipedia.org/wiki/dynamical_system https://en.wikipedia.org/wiki/initial_conditions proceedings of engineering and technology innovation, vol. 8, 2018, pp. 01 08 copyright © taeti 3 quadratic, cubic and tent maps. these barker codes have various lengths such as 2, 3, 4, 5, 7, 11 and 13. for improving the r adar target detection performance and increase the efficiency of operation, binary chaotic phase and barker sequences are used to modulate the sine wave. fig.2 block diagram of the proposed lms-rtd methodology the block diagram proposed method consists of the radio frequency section, switch, antenna, received signal, lms filter, convolution, and display. in this method, the modulator send s the sinusoidal signal through the rf section and then after over the antenna. as shown in fig. 2 the transmitter sends the radio signal through the antenna and then it is pas sed through the free space for detecting object(s). the radar pulse with high velocity has to hit the target and it is reflected back. depending o n the echo signal velocity, object velocity and its position from the base station are detected. the environme ntal disturbances are the cause for the suppression of the signal and this can be rectified by using adaptive lms filter. after the filtering process t he detected object parameters are processed, then analyzed according to that the position will be determined. 3. signal generation 3.1 binary phase code generation using chaotic maps a chaotic dynamical system is a bounded, deterministic system that exhibits random-like behavior through its sensitive dependence on its initial conditions. the use of chaotic s ignals in radar has received a lot of attention in the recent years because they behave like noise, have a wide band and are easy to generate. in mathematics , a chaotic map is a map (= evolution function) that exhibits some sort of chaotic behavior. maps may be parameterized by a discrete-time or a continuous -time parameter. discrete maps usually take the form of iterated functions . chaotic maps often occur in the study of dynamical systems . chaotic maps often generate fractals . although a fractal may be constructed by an iterative procedure, some fractals are studied in and of themselves, as sets rather than in terms of the map that generates them. this is often because there are several different iterative procedures to generate the same fractal. since, in practice init ial conditions can never be specified with infinite precision, the behavior of a chaotic system is unpredictable and therefore noise like. chaotic sequences can be generated usi ng different types of chaotic maps. some of the maps are a) logistic map b) improved logistic map c) cubic map d) tent map e) quadratic map http://en.wikipedia.org/wiki/mathematics http://en.wikipedia.org/wiki/chaotic_map http://en.wikipedia.org/wiki/map_(mathematics) http://en.wikipedia.org/wiki/discrete-time_dynamical_system http://en.wikipedia.org/wiki/chaotic_behavior http://en.wikipedia.org/wiki/iterated_function http://en.wikipedia.org/wiki/dynamical_system http://en.wikipedia.org/wiki/fractals http://en.wikipedia.org/wiki/set_(mathematics) proceedings of engineering and technology innovation, vol. 8, 2018, pp. 01 08 copyright © taeti 4 3.1.1. logistic map the logistic map is a polynomial mapping (equivalently, recurrence relation) of degree 2, often cited as an archetypal example of how complex, chaotic behavior can arise from very simple non-linear dynamical equations.as a part of discrete-time demographic model this logistic map was popularized in a seminar 1976 paper by the biologist robert may and this map is analogous to the logistic equation first created by pierre françois verhulst. mathematically, the logistic map is written : x(n+1) f * x(n) * x(n) *(1 x (n))µ   (4) where µ is constant and is called the bifurcation parameter [μ € (0, 4)] and [x € (0, 1)]. the behavior depends on the value of μ in the equation. when μ = 4 the logistic map exhibits chaotic behavior. 3.1.2. improved logistic map this map described by the equation . 2 x(n+1) f * x(n) 1 2* x (n)   (5) where x € (-1, 1) this equation shows the chaotic behavior of initial value x(0) varying from 0 to 1. when x (0) > 1 as n tends to infinity, x(n) also tends to infinite. 3.1.3. cubic map this map described by the equation . 3 x(n+1) f * x(n) * x (n) 3* x(n)   (6) where x € (-1, 1) and µ=4 3.1.4. tent map tent maps are widely used for chaos based applications because it is a discrete -time piecewise-affine input/output characteristics curve, such as true random number generation. this map described by the equation. 3 x(n+1) f * x(n) * x (n) [1 2 | x(n) 0.5]    (7) 3.1.5. quadratic map this map is described by the following equation . 2 x(n+1) f * x(n) 0.5 * x(n)   (8) where x(n) ∈ (−1, 1) and µ =4 3.2. binary phase code generation using bark er and combined bark er codes table 1 barker codes no of panels barker code 2 +1 -1 2 +1 +1 3 +1 +1 -1 4 +1 +1-1 +1 4 +1 +1+1 -1 5 +1 +1 +1 -1 +1 7 +1 +1 +1 -1 -1 +1 -1 11 +1 +1 +1 -1 -1 -1 +1 -1 -1 +1 -1 13 +1 +1 +1 +1 +1 -1 -1 +1 +1 -1 +1 -1+1 http://en.wikipedia.org/wiki/polynomial http://en.wikipedia.org/wiki/map_(mathematics) http://en.wikipedia.org/wiki/recurrence_relation http://en.wikipedia.org/wiki/quadratic_function http://en.wikipedia.org/wiki/chaos_theory http://en.wikipedia.org/wiki/non-linear http://en.wikipedia.org/wiki/robert_may,_baron_may_of_oxford http://en.wikipedia.org/wiki/pierre_fran%c3%a7ois_verhulst proceedings of engineering and technology innovation, vol. 8, 2018, pp. 01 08 copyright © taeti 5 the known nine barker codes are represented in table 1. combined barker code method involves phase coding with one barker code within each segment of another barker code. for example, the barker code of length 3, b3, may be combined with that of length 4, b4, to create a code of length 12: b43=b4⊗b3= (+1 +1-1 +1, +1 +1-1 +1, -1 -1 +1 -1). note that bnm=bn⊗bm is the kronecker product. note that bnn is not used because partial correlations are high [10]. 4. implementation of proposed method the proposed method was designed and simulated in matlab r2017a simulink to obtain the de -noised signal by using adaptive lms filter. xilinx ise tool is used to find the number of flip -flops, slices and luts. a number of iterations were performed to de-noising radar signal and to obtain the minimized values of rmse and better values of snr. 5. results and discussion the table 2 is the performance of the parameters such as number of slice, luts and flip flops for lms method. from this result table, we can easily understand that the nu mber of luts and flip-flop’s can be reduced in proposed lms method. due to the reduction of those parameters, the area can be optimized in lms filter. the lms performance is shown in fig. 3, which is getting from xilinx ise software. table 2 number of slices, luts, flips flops for lms algorithm using xilinx ise fpga tool filter order algorithms no. of slices no. of luts no. of flip flop 16 tap [9] a 732/28800 2714/28800 449/2997 16 tap [9] b 888/28800 1376/28800 681/1583 8 tap lms 208/5,472 272/10,944 288/10,944 the fig. 3 shows the hardware performance of lms method. it described the total number of used luts, slice and flip flop values of the proposed lms and existing methods. furthermore, the proposed lms method reduces number of luts and flip flop as compared with existing algorithms. fig. 3 performance of slices, luts and flip flop for different algorithms using bar graph the table 3 shows the study of performance analysis of rmse values using lms algorithm for different phase code sequences. as discussed earlier, the entire system is modelled by using matlab simulink environment. the simulink provides customizable block libraries, a graphical editor, and solvers for simulating and modelling dynamic systems . it is integrated with matlab, enabling to incorporate matlab algorithms into models then export simulation results to matlab for more analysis. the lms algorithm target detection tracking model consists of radar transmitter (sine wave, repeating square star, rate transition 1 and 2, and product) selector, switch, target, receiver part (buffer, mean, product1, and lms proceedings of engineering and technology innovation, vol. 8, 2018, pp. 01 08 copyright © taeti 6 filter). in this paper, the modulator will send the signal through the selector after this radar signal over the target as sh own in the fig. 7. it is transmitted towards the target to detect objects which are in space. the radar pulse with high velocity has hit to targ et and it is reflected back. table 3 rmse performance of lms algorithm bit length s nr = 27.8992201978764 7 bit 11 bit 13 bit 26 bit binary phase codes rms e rms e rms e rms e barker & combined codes b7,b11,b13, b13⊗b2 0.7059326171875 0.7058105468750 0.7060546875000 0.705810546875 logistic map 0.7059326171875 0.7059326171875 0.7059326171875 0.7059326171875 imp roved logistic map 0.7061767578125 0.7059326171875 0.7061767578125 0.7060546875000 cubic map 0.7059326171875 0.7058105468750 0.7056884765625 0.7056884765625 quadratic map 0.7064208984375 0.7054443359375 0.7059326171875 0.7059326171875 tent map 0.7059326171875 0.7059326171875 0.7059326171875 0.7059326171875 table 3 rmse performance of lms algorithm (continued) bit length s nr = 27.8992201978764 55 bit 65 bit 77 bit 143 bit 169 bit binary phase codes rms e rms e rms e rms e rms e barker codes b11⊗b5, b13⊗b5, b11⊗b7, b13⊗b11, b13⊗b13 0.705932617187 0.7056884765625 0.705810546875 0.7059326171875 0.7059326171875 logistic map 0.705932617187 0.7059326171875 0.7059326171875 0.7059326171875 0.7059326171875 imp roved logistic map 0.705688476562 0.7059326171875 0.705810546875 0.7059326171875 0.706054687500 cubic map 0.706054687500 0.7058105468750 0.7059326171875 0.706054687500 0.706054687500 quadratic map 0.705810546875 0.7061767578125 0.7061767578125 0.705810546875 0.706054687500 tent map 0.705932617187 0.7059326171875 0.7059326171875 0.7059326171875 0.7059326171875 fig. 4 transmit signal fig. 5 reflected signal fig. 6 detected target location proceedings of engineering and technology innovation, vol. 8, 2018, pp. 01 08 copyright © taeti 7 depending on the reflected or received signals velocity, target velocity from the base station ca n be measured. finally, the detected object parameters are processed, analyzed of the target and motion will be determined. the fig. 4 shows the modulated sine wave sequence which is used to transmit towards the target with respect to time. fig. 5 shows the noisy signal which is affected by additive white gaussian noise. fig. 6 shows the detected target location in terms of distance from transmitter the transmitter.. the rtl schematic of lms filter is shown in fig. 8. this schematic obtained from simplify pro by using verilog hdl code which is written for lms. fig. 7 proposed target detection tracking model using matlab simulink fig. 8 rtl schematic of lms filter proceedings of engineering and technology innovation, vol. 8, 2018, pp. 01 08 copyright © taeti 8 6. conclusions this paper presented a new efficient radar target detection using radar signal de-noising with an lms to suppress the noise which is added in free space. in order to validate the proposed lms-rtd design with existing designs, rmse and snr values are evaluated by using matlab simulink, number of slices, luts and flip-flops are evaluated using xilinx ise fpga tool. from the matlab simulation results, i concluded that the 7 bit binary chaotic quadratic phase code sequences is giving best rmse value as compared with barker codes and combined barker codes. from xilinx ise fpga simu lation results, proposed model details were estimated more precisely than other existing models with less hardware and computational complexity references [1] r. l. devaney, a first course in chaotic dynamical systems: theory and experiment (studies in nonline arity), westview press, 1992. [2] v. jagannaveen, t. prabakar , j. v. suman, and p. d. pradeep, “noise suppression in speech signals using adaptive algorithms ,” international journal of signal processing, image processing and pattern recognition, vol. 3, no. 3, pp. 87-96, september 2010. [3] j. b. seventline, d. e. rani, and k. r. rajeswari, “ternary chaotic pulse compression sequ ences ,” radioengineering, vol. 19, no. 3, pp. 415-420, 2010. [4] k. sridevi, j. b. seventline, and d. e. rani, “four phase pulse compression sequences generated using chaotic ma ps,” international journal of computer applications, vol. 48, no. 14, pp. 25-30, june 2012. [5] b. n. mahaseth and m. s. anuradha, “binary and ternary sequence generation using improved logistic map,” international journal of advanced research in computer and communication engineering, vol. 2, no. 8, pp. 3290-3294, august 2013. [6] a. ashtari, g. thomas , h. garces , and b. c. flores , “radar signal design using chaotic signals”, proc. international waveform design and diversity conference, pp. 353-357, october 2007. [7] g. heidari-bateni and c. d. mcgillen, “chaotic sequences for spread spectrum: an alternative to pn-sequences”, ieee conference on icstwc, pp. 437-440, 1992. [8] r. c. hilborn, “chaos and nonlinear dynamics: an introduction for scientists and engineers ,” canada: oxford university press, 1994. [9] t. pitchaiah and p. v. sridevi, “low power area efficient high speed implementation of lms adaptive filter using distributed arithmetic”, international journal of emerging trends in electrical and electronics, vol. 11, no. 4, pp. 69-75, 2015. [10] e. r. adams and p. losiewicz, “detection and characterization of phase-coded radar signals,” cranfield univ., research final report ada433120, pp. 4-5, 2004.  proceedings of engineering and technology innovation , vol. 4, 2016, pp. 13 15 13 improving the weldability of nickel-based superalloy by high frequency vibration ti-yuan wu 1,* , ming-tzer lin 1 , weite wu 2 1 institute of precision engineering, national chung hsing university, taichung, taiwan. 2 department of materia ls science and engineering, national chung hsing university, ta ichung , taiwan. received 22 february 2016; received in revised form 15 march 2016; accept ed 13 april 2016 abstract the objective of this paper is to discuss the weldability of mar-m -004 nic kel-based alloy by simu ltaneous proceeding vibration. three kinds of vibration modes were chosen to compare the results, including high frequency vibration, subresonant and without vibration . we used x-ray diffraction (xrd) to quantize the residual stress of each sample, a ls o the microstructure and crystal structure were investigated by using optical microscopes. the results showed that the grain size will get refined after vibrat ion we lding, especially in high frequency vibration. fro m xrd and microstructure results, by using the high frequency vibration method, there has a significant effect of having lowest residual stress and lowest stress rela xation; furthermore, the formation of cracks was also inhib ited and having the shortest crack length. ke ywor ds : vibratory stress relief, residual stress , welding crack, nickel-based superalloy, high frequency vibration 1. introduction nic ke l-based alloy has the properties of high strength and high corration resistance in high temperature [1]. ho wever, mar-m-004 n ickel-base superalloy contains high levels of titanium and alu minum. while cooling after we lding, a large nu mber of γ 'phase will be prec ipitated, and produce large shrinkage stress es, which will result in c racking and having poor weldabilities [2]. vibratory stress relief (vsr) is effective in reducing the residual stress, prevending defo rmat ion and increasing the materia l’s strength. furthermore, it has advantages of energy saving, without pollution, low cost, high efficiency and easy to operate [3]. besides , simultaneously apply the vibration force wh ile we lding, which can properly increase the atoms’ mobility in high te mperature, will have a better effect of stress relie f and refine the microstructure [4]. bonal corporation has developed a technique of subresonant vibration, which uses the frequency of 1/3 of resonant amplitude to do vsr. this d iscusses have got the world wide adoption,also been recognized in the papers published[5]. fig. 1 waveforms and spectrums of different vsr modes , (a) high frequency vibration (b) subresonant (c) resonant we have developed a new approach of high frequency vibratory stress relief, wh ich use a low-frequency wave to stimulate a h igh fre quency comple x waveform. thus, this technique has been proved to be more effective to promote local dislocation move ment and reduce the residual stress [6]. 2. method this e xperiment is the use of nd: ya g laser weld ing s imultaneously input vibration to fill (a) (b) (c) * corresponding aut hor. email: paxon1992911@gmail.com proceedings of engineering and technology innovation, vol. 4, 2016, pp. 13 15 14 copyright © taeti blind holes . both the filler and the base materia l are mar-m-004 nickel-base superalloy. as fig. 2, the vsr apparutus composed of a vibration motor input a vibration wave to the workp iece and scaning the operating frequency of different vsr modes. fig. 2 schematic of vibration welding process 2.1. processing steps (1) scanning the spectrum, subresonant and high frequency point of the workpiece. (2) using subresonant53.1hz, high frequency 51.7hz 50.4hz and without vibration to do simultaneous vibration welding. (3) for each vibration modes, fill two blind hole on the nickel-based alloy plate. 2.2. sample analysis (1) using cr-xdr to measure the residual stress value. (2) using om and sem to observe the mic rostructure. 3. results and discussion fro m fig. 3, the residual stress in weld bead is highest without vibration and lowest while using high frequency vibration , also, by cutting the weld bead to relax the stress . fig. 3 vibration modes -residual stress diagram fig. 4 shows the mic rostructure of the weld bead. while using high frequency vibration weld ing, there has the least cracks in fusion line and the shortest crack’s length in fusion zone. fig. 4 bead cross -section (a) no vibration 0hz (b) high frequency vibration 50.4hz (c) high frequency vibration 51.7hz (d) subresonant 53.1hz fig. 5 mic rostructure of weld bead (a) no vibration 0hz (b) high frequency vibration 50.4hz (c ) high frequency vibration 51.7hz (d) subresonant 53.1hz as the crystal structure in fig. 5, vibration weld ing can form equia xed dendritic g rains in contrast to the columnar dendrit ic grains without vibration. it represent that the solidification and cooling rate a re faster, and have the better gra in structures. furthermore, the high frequency vibration can form a optimu m structure as compare with other vibration methods. 4. conclusions summarize the total results, high frequency vibration we lding possessed grain refine ment, lowest residual stress and reduce the cracks , which can effective ly get a better mechanica l properties and weldability of nic kel-based super alloy. acknowledgement the authors are grateful to all colleagues and students who contributed to this study. vibration motor welding direction laser beam specimen weling wire a: subresonat 53.1hz b: wavelet 51.7hz c: wavelet 50.4hz d: no vibration 0hz a b c d 55 42 44 47 0 hz 50.4 hz 51.7 hz 53.1 hz 0 200 400 600 800 1000 1200 1400 1600 1800 s tre ss re la x tio n (% ) r e si d u a l st re ss ( m p a ) as welded before cutting  relaxtion (%) frequency(hz) 38 40 42 44 46 48 50 52 54 56 without vibration subresonant high frequency proceedings of engineering and technology innovation , vol. 4, 2016, pp. 13 15 15 copyright © taeti references [1] s. kou, welding metallurgy, 2nd ed., a john wiey & sons,inc., new jersey, pp. 263-300, 2003. [2] m. h. haafkens and g. h. matthey, “a new approach to the weldability of nickel-base as-cast and powder metallurgy superalloys,” welding journal, vol. 61, no. 1, pp. 25-30, 1982. [3] c. a. walker, a. j. waddell, and d. j. johnston, “vibratory stress relief an investigation of the underlying process,” process institution mechanical engineers, vol. 209, pp. 51-58, 1995. [4] s. weiss and s. baker, “ vibrational residual stress relief in a plain carbon steel weldment,” welding journal, vol. 55, 1976. [5] m. la x, “stress relief technology”, bonal technology, lnc, 1986. [6] w. wu, “method for relieving residual stress in an object,” us 2009 /0049912 al, 2009. 5__peti#7002__43-48 proceedings of engineering and technology innovation, vol. 18, 2021, pp. 43-48 hybrid energy harvester for medical sensor node toward real-time healthcare monitoring saeed mohsen * faculty of engineering, ain shams university, cairo, egypt received 14 january 2021; received in revised form 16 february 2021; accepted 04 march 2021 doi: https://doi.org/10.46604/peti.2021.7002 abstract in healthcare applications, the remote monitoring for moving patients depends on medical sensor nodes, which should be mobile. thus, the power mains of medical sensor nodes should be disconnected most of the time to monitor natural movements of patients. in this paper, a self-sustainable medical sensor node is proposed for healthcare monitoring applications. the node implementation consists of a microcontroller unit (mcu), a photo-plethysmography (ppg) sensor, a bluetooth low energy (ble) module, and a mpu module that includes a gyroscope with accelerometer. the power supply of the node is a hybrid energy harvester developed to provide a sustainable energy for the sensor node. the harvester is composed of a photovoltaic (pv) panel, a thermoelectric generator (teg) module, a dc-dc converter, and a super-capacitor. experimental results illustrate that the proposed node can monitor a physiological data on a mobile device using the ble terminal application. keywords: medical sensor node, healthcare, energy harvesting, super-capacitor 1. introduction in the last decade, sensor nodes have played an effective role in biomedical fields [1-3]. however, the power supply of sensor nodes is frequently batteries, which have limited lifetime [4-5]. energy harvesters can extend the battery lifetime ofsensor nodes [6-9]. photovoltaic and thermoelectric energy harvesters are used for the reason that they have high power density [10]. thus, a hybrid photovoltaic-thermoelectric energy harvester is implemented to overcome the limitation of using batteries. there are multiple sensor nodes presented for medical applications [11-13]. toh et al. [11] present an autonomous wearable sensor node with flexible photovoltaic energy harvesting. it utilizes an analog maximum power point tracking technique to manage the power of the node. lim et al. [12] introduce a solar energy harvester for a sensor device. it includes a photovoltaic panel (pv), a dc-dc boost converter, and a battery. tran and chung [13] present an autonomous sensor node, which consists of a photoplethysmogram (ppg) sensor and an electrocardiogram (ecg) sensor, to monitor a patient’s physiological data supplied by battery. the data of the sensor node is visualized by using the mymonitor android application. this paper presents an autonomous medical sensor node powered by a sustainable hybrid energy harvester. this harvester is developed based on a low-power dc-dc converter with two super-capacitors. these super-capacitors are used as backup power supply with overall capacity of 12.5 f. the harvested hybrid energy is sufficient to supply the medical sensor node. the data of the sensor node is collected by ppg, gyroscope and accelerometer sensors. the data is transmitted to the ble terminal android application via ble technology. * corresponding author. e-mail address: g17082131@eng.asu.edu.eg proceedings of engineering and technology innovation, vol. 18, 2021, pp. 43-48 the reminder of the paper is organized as follows: section 2 introduces the energy harvesting technology. section 3 presents the development of the medical sensor node with the hybrid energy harvester. the experimental results are illustrated in section 4. section 5 presents the conclusion of the paper. 2. energy harvesting technology energy harvesting is a technology that converts and stores an energy from an environmental source for supplying autonomous devices [14]. the generic block diagram of energy harvesting technology is illustrated in fig. 1. the block diagram consists of five blocks. the first one is environmental source, i.e. light, thermal, radio frequency (rf), vibration, wind, etc. energy harvester is the second block, which converts an environmental energy into electrical energy. the third block refers to conversion and energy management, which is used to maximize the output power of the energy harvester. the fourth block is energy storage, which stores the harvested energy as backup. energy storage can be rechargeable batteries or super-capacitors. there is a trade-off between them in terms of energy density and charging cycles. super-capacitors have unlimited charging cycles in comparison with batteries, while the batteries have higher energy density than super-capacitors [15]. finally, energy harvesting technology is employed to power a device, e.g. a medical sensor node. fig. 1 the generic block diagram of energy harvesting technology table 1 shows power densities of different energy sources [16]. an outdoor light has a power density of 15 mw/cm 2 , while the power density of an indoor light is 100 µ w/cm 2 . also, the power density from a thermal energy source is 60 µw/cm 2 , and the power density from a vibration energy source is 335 µ w/cm 2 . in addition, the radio frequency (rf) can generate a power in the range of 0.1 to 1 µ w/cm 2 . finally, an air flow energy source has a power density of 7600 µ w/cm 2 at a velocity of 5 m/s. it is clear that the highest energy source is outdoor light. on the other hand, the thermal energy provides a convenient energy source with medical sensor nodes due to the fact that body heat is an inherent characteristic of human body. therefore, the combination of light and thermal energy sources is used in the proposed system. table 1 power densities of different energy sources energy source power density (µ w/cm 2 ) light (outdoors) 15000 light (indoors) 100 thermal 60 vibration 335 radio frequency (rf) 0.1–1 air flow 7600 (at 5 m/s) 3. development of medical sensor node with hybrid energy harvester 3.1. architecture fig. 2 demonstrates the proposed architecture of the medical sensor node with hybrid energy harvester. this architecture is composed of a ppg sensor to measure the heartbeats of a patient, a mpu module that include an accelerometer to sense the patient’s linear accelerations and a gyroscope to detect the angular velocity, a ble module that send the sensors data to a smartphone, and a microcontroller unit (mcu) that collect these sensors data for processing. this sensor node is powered by a hybrid energy harvester, which contains a thermoelectric generator (teg) module to generate an electric power of the thermal energy of patient body, a pv panel to convert the light into an electric power, and a dc-dc converter to fix the voltage of pv panel and teg module to charge a small-size super-capacitor. 44 proceedings of engineering and technology innovation, vol. 18, 2021, pp. 43-48 fig. 2 the architecture of the medical sensor node with the hybrid energy harvester 3.2. hardware and software implementation the hardware implementation of the medical sensor node with hybrid energy harvester is shown in fig. 3. it includes a flexible pv panel (mpt3.6-75) with an area of 7.2 × 6.0 cm 2 and a teg module (sp1848-27145). it has a small-size of 4 cm × 4 cm × 0.4 cm. the pv panel and the teg module are connected in parallel. the dc-dc converter is based on a ltc3105 chip from linear technology ® . this converter has high efficiency with input voltage from 0.2 to 5 v. the energy storages are two super-capacitors (hv1625-2r7256r) that have capacity of 25 f and voltage rate of 2.7 v. these super-capacitors are connected in series with voltage rate of 5.4 v. the microcontroller unit (mcu) is a low-power at mega-328p manufactured by atmel. this unit has speed of 16 mhz with operating voltage of 5 v. the ppg sensor is selected sen-11574. it is a high range from 0 to 200 bpm. the mpu module is chosen 6050 with accelerometer resolution in the range of -2 to +2 g. its gyroscope angular velocity is between -250 to +250 degree/second [17]. the used ble module is bt05 that has the range of 100 meters. fig. 4 describes the software flowchart of the medical sensor node. the sensor node is operating in two modes: active and sleep. firstly, the microcontroller is configured in active mode. secondly, it continuously measures the data of the sensors. thirdly, the sensor data are sent for a duration of 3 seconds using the ble module. fourthly, the sensors are turned off. fifthly, the ble module is slept and then, the microcontroller enters in sleep mode for a period of 597 seconds to save the power consumption of the node. finally, the previous steps are continuously repeated in a loop [18]. fig. 3 hardware implementation of the medical sensor node with the energy harvester fig. 4 software flowchart of the medical sensor node 4. experimental results fig. 5 describes experimentally the i-v-p electrical characteristics of the pv panel under illumination of 1000 w/m 2 . the orange and blue curves are the photovoltaic power (ppv) and current (ipv) respectively versus the photovoltaic voltage (vpv). the 45 proceedings of engineering and technology innovation, vol. 18, 2021, pp. 43-48 fig. 5 i-v-p electrical characteristics of the pv panel at 1000 w/m 2 fig. 6 r-v and r-p characteristic curves of the teg at ∆t = 5 °c maximum power reaches to 207 mw at voltage of 4.6v. fig. 6 shows the r-v and r-p characteristic curves of the teg module at ∆t = 5 °c. the maximum output power (pout) is measured at 10.5 ω load resistance. also, the dashed blue curve shows the maximum output voltage (vout) of 105 mv with a 60 ω load resistance. fig. 7 shows the charging curve of the 12.5 f super-capacitor of the hybrid energy harvester. the voltage of the super-capacitor is increased from 2.6 v to 5 v in 165 minutes. table 2 demonstrates the charging rate of the 12.5 f super-capacitor. the voltage measurements of the super-capacitor are taken via the excel multi-meter. when the pv panel generates the power higher than the teg module during the sunlight, the dc-dc boost converter starts to charge the super-capacitor. under the darkness case, the teg module supplies power to charge the super-capacitor. fig. 7 charging curve of the 12.5 f super-capacitor table 2 charging rate of the 12.5 f super-capacitor time intervals (min) charging rate (mv/s) voltage (v) 0–26 0.25 2.6–3.0 26–88 0.24 3.0–3.9 88–165 0.23 3.9–5.0 table 3 power consumption of the sensor node mode current (ma) time (s) power (mw) active 20 3 100 sleep 1.5 597 7.5 table 3 illustrates the measured power consumption of the sensor node for each mode. in active mode, the node consumes current of 50 ma. also, the measured current consumption in the sleep mode is 29 ma. according to the flowchart of the node, the total period (t) of the sensor node is 600 seconds (10 minutes) and the operating voltage of the sensor node is 5 v. hence, the average power consumption (pavg) of the sensor node is calculated using eq. (1). therefore, the average power consumption is equal to 7.96 mw. avg active active sleep sleep p p t p t( ) / t= × + × (1) vpv (v) load resistance (ω) v o u t (m v ) p o u t (μ w ) ip v ( m a ) p p v ( m w ) time (minutes) s u p p e rc a p v o lt a g e ( v ) 46 proceedings of engineering and technology innovation, vol. 18, 2021, pp. 43-48 fig. 8 shows the ppg signal that is captured from a patient’s finger in a sitting position. the ppg signal is filtered by a digital low pass filter and recorded by using the serial monitor program. during the experiment, the node samples the pulse oximeter sensor at a sampling frequency of 100 hz. it is continuously monitored for 3 seconds. fig. 9 illustrates the monitored physiological data of the patient on the ble terminal android application. in this figure, the reading of a heartbeat is monitored with 60 bpm and is stored in a mobile phone for displaying. the readings of the patient’s body acceleration are displayed with -1 g in x, y, and z directions. also, the readings of the angular velocity of the patient body are 180 degree/sec in x, y directions and 45 degree/sec in z direction. thereby, these physiological data are an indication of patient status. in other words, a doctor can detect the status via a transmitted data from the patient through the bluetooth low energy (ble) of the node to a mobile phone of the doctor. fig. 8 the captured ppg signal from a patient’s finger fig. 9 screen shot of the monitored physiological data finally, table 4 illustrates a comparison between the proposed system and previous studies. in this work, the proposed sensor node has three physiological parameters, which are more than the nodes parameters in [11-13]. the bluetooth low energy (ble) wireless technology offers a higher range than the technologies in [11, 13]. on the other hand, the utilized energy harvester in the proposed system is a combination of two energy sources, which are “flexible photovoltaic (pv) panel and thermoelectric generator (teg)”, while only one energy source “pv panel” was used in [11-13]. table 4 comparison between the proposed system and previous studies reference [11] [12] [13] this work sensors temperature sensor temperature sensor ppg senor and ecg senor ppg sensor and mpu module number of physiological parameters one (temperature) one (temperature) two (ppg and ecg) three (heartbeat, acceleration, and angular velocity) wireless technology rf transceiver 75 m zigbee 100 m hc-06 9 m ble 100 m energy harvester pv panel pv panel pv panel flexible pv panel and teg module 5. conclusions this paper presents a self-sustainable medical sensor node for real-time healthcare monitoring. the node is powered by a hybrid energy harvester, which is a sustainable supply source during the running time of the node. the architecture of the harvester consists of a flexible photovoltaic panel to sense the irradiance of the sunlight with a thermoelectric generator (teg) module to detect the temperature of human body. the flexible pv panel and teg module are connected in parallel. the combination of the pv panel and teg module are exploited to harvest the fast, sufficient, and more stable energy from the sunlight and the source of body heat. the proposed node continuously measures the heartbeats, accelerations and directions of a patient’s body. finally, a physiological data of the sensor node is monitored on the ble terminal android application. in the future, multiple medical sensor nodes will be developed for a wireless body area network implementation. 47 proceedings of engineering and technology innovation, vol. 18, 2021, pp. 43-48 conflicts of interest the authors declare no conflict of interest. references [1] m. magno, g. a. salvatore, p. jokic, and l. benini, “self-sustainable smart ring for long-term monitoring of blood oxygenation,” ieee access, vol. 7, pp. 115400-115408, 2019. [2] t. wu, f. wu, j. m. redoute, and m. r. yuce, “an autonomous wireless body area network implementation towards iot connected,” ieee access, vol. 5, pp. 11413-11422, 2017. [3] s. mohsen, a. zekry, k. youssef, and m. abouelatta, “an autonomous wearable sensor node for long-term healthcare monitoring powered by a photovoltaic energy harvesting system,” international journal of electronics and telecommunications, vol. 66, no. 2, pp. 267-272, 2020. [4] a. dionisi, d. marioli, e. sardini, e. sardini, and m. serpelloni “autonomous wearable system for vital signs measurement with energy-harvesting module,” ieee transactions on instrumentation and measurement, vol. 65, no. 6, pp. 1423-1434, june 2016. [5] a. decker, “solar energy harvesting for autonomous field devices,” iet wireless sensor systems, vol. 4, no. 1, pp. 1-8, march 2014. [6] t. wu, m. s. arefin, j. m. redouté, and m. r. yuce, “a solar energy harvester with an improved mppt circuit for wearable iot applications,” proceedings of the 11th eai international conference on body area networks, december 2016, pp. 166-170. [7] t. wu, m. s. arefin, d. shmilovitz, j. m. redoute, and m. r. yuce, “a flexible and wearable energy harvester with an efficient and fast-converging analog mppt,” ieee biomedical circuits and systems conference (biocas), october 2016, pp. 336-339. [8] s. barker, d. brennan, n. g. wright, and a. b. horsfall, “piezoelectric-powered wireless sensor system with regenerative transmit mode,” iet wireless sensor systems, vol. 1, no. 1, pp. 31-38, march 2011. [9] r. hamid and m. r. yuce, “a wearable energy harvester unit using piezoelectric electromagnetic hybrid technique,” sensors and actuators a: physical, vol. 257, pp. 198-207, april 2017. [10] f. wu, c. rudiger, and m. r. yuce, “design and field test of an autonomous iot wsn platform for environmental monitoring,” 27th international telecommunication networks and applications conference (itnac), november 2017, pp. 1-6. [11] w. y. toh, y. k. tan, w. s. koh, and l. siek, “autonomous wearable sensor nodes with flexible energy harvesting,” ieee sensors journal, vol. 14, no. 7, pp. 2299-2306, july 2014. [12] l. j. chien, m. drieberg, p. sebastian, and l. h. hiung, “a simple solar energy harvester for wireless sensor networks,” 6th international conference on intelligent and advanced systems (icias), august 2016, pp. 1-6. [13] t. v. tran and w. y. chung, “highefficient energy harvester with flexible solar panel for a wearable sensor device,” ieee sensors journal, vol. 16, no. 24, pp. 9021-9028, october 2016. [14] k. v. naveen and s. s. manjunath, “a reliable ultracapacitor based solar energy harvesting system for wireless sensor network enabled intelligent buildings,” 2nd international conference on intelligent agent and multi-agent systems (iama), september 2011, pp. 20-25. [15] s. mohsen, a. zekry, k. youssef, and m. abouelatta, “a self-powered wearable wireless sensor system powered by a hybrid energy harvester for healthcare applications,” wireless personal communications, vol. 116, no. 4, pp. 3143-3164, september 2020. [16] a. s. weddell and m. magno, “energy harvesting for smart city applications,” international symposium on power electronics, electrical drives, automation and motion, june 2018, pp. 111-117. [17] j. k. lee, k. kim, and s. lee, “stretchable, patch-type, wireless, 6-axis inertial measurement unit for mobile health monitoring,” proceedings of engineering and technology innovation, vol. 14, pp. 16-21, january 2020. [18] n. aphiratsakun and s. liwsakphaiboon, “path control irobot create2 based sensors,” proceedings of engineering and technology innovation, vol. 16, pp. 45-51, 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/). 48  proceedings of engineering and technology innovation, vol. 15, 2020, pp. 01 07 estimation of the pcb production process using a neural network kuo-hsien hsia 1 , jr-hung guo 2,* 1 department of computer science and information engineering, far east university, tainan, taiwan 2 department of electrical engineering, national yunlin university of science & technology, yunlin, taiwan received 16 april 2019; received in revised form 16 may 2019; accepted 23 june 2019 doi: https://doi.org/10.46604/peti.2020.4265 abstract printed circuit boards (pcb) are an integral part of all electronic products, and the production process for printed circuit boards is quite complex. as the life cycle of electronic products becomes shorter and shorter, and the precision and signal bandwidth of electronic products become higher and higher, the manufacturing process of printed circuit boards is further complicated. therefore, how to pre-evaluate the production difficulty before starting the production will effectively increase the efficiency and quality of printed circuit board production. gerber file is the most commonly used data format for the printed circuit board industry. this file contains most of the parameters required for the manufacture of printed circuit boards. therefore, this study uses a neural network to evaluate new pcb products before they are produced through the production parameters that are more influential in the pcb manufacturing process. this makes it possible to evaluate the difficulty and the required production process before the new pcb product is produced. this will be very beneficial for the pcb production schedule, quality control, and cost. keywords: printed circuit board (pcb), neural network, gerber file, production schedule 1. introduction printed circuit boards (pcb) are an essential component of all electronic products. however, due to the complexity of its manufacturing process and the shortened life cycle of electronic products, the pcb industry must adjust production equipment and parameters at any time according to the needs of products or customers. in addition, due to the demand for equipment and processing, the pcb industry often needs to outsource the process or equipment shortage. this makes the entire production process control, production data collection, quality management, raw materials management, etc., face many problems. therefore, how to make these problems and processes be managed and properly arranged will be an important issue for the pcb industry to continue to survive and grow under international competition. in previous studies, such as [1], the hybrid taguchi genetic algorithm (htga) was used to optimize the drilling path of the pcb. [2] used the modified artificial bee colony algorithm (mabca) for the automatic nozzle changer (anc) assignment problem, the nozzle assignment problem, and the component pick-and-place sequence problem three problems are optimized. [3] used image and embedded systems to do pcb defect detection. [4] used a gerber file as the detection of pad on the pcb. from previous studies, we found that there is not much research on pcb production forecasting, mainly because the previous pcb production mode is mass production. due to rapid market changes and diversified customer demand, the manufacturing pattern of printed circuit board products has gradually changed to a “small variety” and “short delivery” production method. faced with more and more requests for temporary rush orders and insert orders, how to quickly replace the * corresponding author. e-mail address: g9710801@yuntech.edu.tw proceedings of engineering and technology innovation, vol. 15, 2020, pp. 01 07 2 production materials and product models under the pressure of limited delivery time. and let the on-site personnel maintain the quality consistency under the frequent conditions of changing the line, and the production management personnel can quickly arrange the production line, in line with the customer delivery period, which will be the key to the success of the enterprise transformation industry 4.0. in order to solve these problems, many manufacturers have introduced mes (manufacturing execution system) and aps (advanced planning and scheduling), but there are not many successful cases. the main reason is that product information cannot be collected in real time. the problem of not being able to collect production information in real time is that the production equipment is complicated, or the product information is collected manually, and the data collection cannot be instantaneous or even wrong. therefore, if we can obtain the pcb specification data, we can pre-evaluate its difficulty, even the required time and raw materials, which will greatly help the pcb industry. the pcb is produced in the pcb factory by the gerber [5] format file. because of its simple data format and composition, and almost every layout software such as altium designer, pads, allegro, etc. can output this format, it is widely used in the circuit board industry. gerber file is a series of data consisting of x, y coordinates. these coordinates can indicate the starting point, end point, pad position or other special position on the electronic circuit diagram. the gerber file was originally a data file exported from the pcb cad software as an optical drawing language. the format developed by a us company called gerber scientific's professional plotter in the 1960s is now owned by ucamco. almost all developments related to pcb cad systems are based on this format, and the gerber file can be directly inputted into the plotter to draw the negatives required for each station on the board. therefore, the gerber file has become recognized in the electronics industry standard. usually, a gerber file defines a single line layer image, so the pcb manufacturer can get the information needed for production from the gerber file of each layer. the gerber format commonly used in the pcb industry is the rs-274x gerber format, also known as the extended gerber format or x-gerber, which is a two-dimensional vector graphics description format. the rs-274x is a readable ascii format consisting of a series of commands and coordinates. the rs-274x document contains a complete description of the layers of the board, with all the elements needed for board graphics, and can define any shape apertures. it can be automatically imported and processed, making it ideal for fast and secure data conversion and reliable automated workflows. therefore, this study hopes to be able to analyze the production difficulty through the neural network when obtaining the gerber file and even calculate the precise production time and the required raw materials. because the production process of pcbs is very long, this study will focus on the exposure process in the etching process. the reason why this process will be selected as the research goal is that there are more and more layers of pcbs, and the pcb will appear as shown in fig. 1 due to design considerations. (a) pcb examples 1 (b) pcb examples 2 fig. 1 two examples of pcbs that are difficult to make the reason why this process will be selected as the research goal is that there are more and more layers of printed circuit boards, and the pcb will appear as shown in fig. 1 due to design considerations. under normal circumstances, in the etching process, the larger the area of the remaining copper foil, the shorter the time required for exposure, so the exposure time required for fig. 1(b) should be relatively short under normal conditions. however, as in the position indicated by the arrow in the area of the red frame of fig. 1(a) and fig. 1(b), we can find that there are many lines with a small interval so that the proceedings of engineering and technology innovation, vol. 15, 2020, pp. 01 07 3 exposure time needs to be shortened during exposure. in order to avoid the area where the interval is small, it is enlarged or completely etched during the etching operation, causing a problem of quality. so how to correctly estimate the exposure time is very important. although gerber file is a two-dimensional vector graphic description format, table 1 is a small gerber file sample, and fig. 2 is the actual drawing result. from this example, we can see that the format is only describing the categories of apertures as well as the actions and coordinates, not the actual drawing objects. therefore, it is difficult to know the design of the pcb directly from the gerber file. therefore, this study draws the gerber file graphically and then uses convolutional neural networks (cnn) to find the number of smaller pitch areas in the pcb design. the results obtained through such calculations and analysis can be used to estimate the time required for production. in terms of time estimation, we also use the convolutional neural networks method, and the parameters used as input are sample image type, sample image size, time, and pcb size. table 1 gerber file code sample [6] g54d16* <-select aperture 6 (20 mil circle trace) g01x90y85d02* <-move the shutter to 90,85 y85d01* <-open the shutter and draw to 90,85 x190* <-while shutter remains open, draw to 90,185 y85* <-draw to 190,85 x90* <-draw to 90,85 g54d45* <-select aperture 45 (60 mil circle flash) x140y135d02* <-close shutter and move to 140,135 d03* <-flash expose x0y0d02*m02* <-return to the origin and stop fig. 2 gerber file code sample draw result [6] 2. convolutional neural networks method fig. 3 a simple convolution net [7] in this study, we used the convolutional neural networks method, which is commonly used in many types of neural networks for image recognition and image processing. the advantage is that it can be converted from the point in the image to the comparison of the regions, and then through the judgment of the feature, the stepwise comparison results are superimposed to find the position with the feature. this comparison process is shown in fig. 3. how was the convolution layer achieved? convolution +relu convolution +relu pooling pooling fully connected fully connected output predictions proceedings of engineering and technology innovation, vol. 15, 2020, pp. 01 07 4 basically, centering on each point in the image, then taking n x n points to form an area, where n is called kernel size. the matrix weight of n x n is called the convolution kernel. because the general image is based on rgb, it is necessary to separately process the rgb three layers separately, and then superimpose the results of the three layers to obtain the recognized result. the whole process is shown in fig. 4. fig. 4 simple convolutional neural networks [8] however, convolution has a large amount of computation, so we must first pool the graphics we want to compare. pooling is a way to compress images and preserve important information. pooling selects different windows on the image and selects a maximum value in the window range. after the original image is pooling, the number of pixels it contains will be reduced to one-fourth of the original. however, because the pooling image contains the maximum value of each range in the original image, it still retains the degree of conformity between each range and each feature. in other words, the pooling information is more focused on whether there are matching features in the image, not where the features exist in the image. this can help cnns determine whether a feature is included in the image without having to be distracted by the location of the feature. in addition, we also added a rectified linear unit (relu), which is used to convert all negative numbers on the picture in the convolution operation to 0. this technique can prevent the result of cnns from approaching 0 or infinity. although this method can avoid the interference of the noise in the image and the sharpening action of the image. however, there are still several problems when actually performing image recognition with the convolutional neural networks method. the first is how the features of the image should be obtained. this part can be solved by backpropagation, that is, we can prepare some identification samples in advance. then, after a lot of training, the convolutional neural networks method can accurately find out whether the feature is in the image and its position. table 2 shows the sample images we selected. these 6 samples are more common in pcb design with small line widths, small line spacing, and special designs. in addition, in order to speed up the training speed and accuracy of the convolutional neural networks method, we produce sample images in 45 degrees for each image. finally, in the resolution part of the sample image, since we ha ve processed the gerber file into an image and processed it, and the resolution of the image we draw is 200 dpi, the sampling of all sample images is also obtained under the same resolution. in terms of time estimation, we use the long short-term memory network (lstm network), and the parameters used as input are the sample image type, the number of sample images, the time, and the pcb board size. then use the previous pcb production time in the exposure process and the gerber file and other materials for training. proceedings of engineering and technology innovation, vol. 15, 2020, pp. 01 07 5 table 2 6 pcb shape sample images (a1) s1 0° (a2) s2 0° (a3) s3 0° (a4) s4 0° (a5) s5 0° (a6) s6 0° (b1) s1 45° (b2) s2 45° (b4) s3 45° (b4) s4 45° (b5) s5 45° (b6) s6 45° (c1) s1 90° (c2) s2 90° (c3) s3 45° (c4) s4 90° (c5) s5 90° (c6) s1 90° (d1) s1 135° (d2) s2 135° (d3) s3 135° (d4) s4 135° (d5) s5 135° (d6) s6 135° (e1) s1 180° (e2) s2 180° (e3) s3 180° (e4) s4 180° (e5) s5 180° (e6) s6 180° (f1) s1 225° (f2) s2 225° (f3) s3 225° (f4) s4 225° (f5) s5 225° (f6) s6 225° (g1) s1 270° (g2) s2 270° (g3)s3 270° (g4) s4 270° (g5) s5 270° (g6) s6 270° (h1) s1 315° (h2) s2 315° (h3) s3 315° (h4) s4 315° (h5) s5 315° (h6) s6 315° 3. experimental results since we need to plot the gerber file, we refer to the gerberview [9] project to draw the gerber file, and then use c# to develop the neural network analysis software. and using emgucv [10] to process the painted gerber file into image files and sample images. the interface of this software is shown in fig. 5. proceedings of engineering and technology innovation, vol. 15, 2020, pp. 01 07 6 after the entire software was completed, we used 1000 gerber files as training samples. we also record the number and location of six sample images for each gerber file while training. after training, we found that we set the weight of each sample in the six sample images to be the same at first, but after training, we found that the sample image weight of the 45-degree multiple will increase. in addition, if there is less than 3 mil design in the gerber file, there is a slight influence on the recognition rate. after the identification and statistics of the images, we used the data and the results of the previous production process to make time prediction training. then use another 50 gerber files to estimate the time and compare it with the actual production results. the results are shown in fig. 6. from the figure, we can see that the average correct rate is still more than 90%. fig. 5 convolutional neural networks and gerberview software fig. 6 comparison of the estimated time and the actual time 4. conclusions in this study, we used a gerber file as the basis for pcb production time estimates. because the gerber file is a step-by-step description of the objects on the pcb, rather than a complete representation of each object with 2d data. therefore, in this study, the gerber file is first imaged, and six types of sample images that are difficult in pcb production are selected, and then identified by cnn. then, the results of the identification will be trained with the previous production materials, and the pcb data that have not yet been produced will be used for estimation. after the actual test, the estimated time and the actual time correct rate is on average more than 90%. therefore, it can be confirmed that this method is effective. however, during the experiment, some pcb special designs were still found, which would have a bad effect on the results. in the future, we will continue to use the cnn method to directly analyze the gerber file, hoping to improve the results of the entire identification and prediction. and let this result extend to more pcb processes, and even the production quality and other information will be added in the future. and let this result extend to more pcb processes, and even the production quality and other information will be added in the future. proceedings of engineering and technology innovation, vol. 15, 2020, pp. 01 07 7 conflicts of interest the authors declare no conflict of interest. references [1] d. h. a. janan and t. k. liu, “the optimization of cnc pcb drilling using hybrid taguchi genetic algorithm,” kybernetes, vol. 45, no. 1, pp. 107-125, 2016. [2] c. j. lin and m. l. huang, “modified artificial bee colony algorithm for scheduling optimization for printed circuit board production,” journal of manufacturing systems, vol. 44, pp. 1-11, july 2017. [3] n. dave, v. tambade, b. pandhare, and s. saurav, “pcb defect detection using image processing and embedded system,” international research journal of engineering and technology, vol. 3, no. 5, pp. 897-1901, 2016. [4] v. raghuvanshi, a. burman, p. p. bartakke, and s. deshpande, “pcb solder pad inspection mechanism using gerber file,” 2016 international conference on communication and signal processing (iccsp), ieee, april 2016. [5] “gerber formats,” https://www.ucamco.com/en/guest/downloads, february 8, 2019. [6] “gerber translator,” http://literature.cdn.keysight.com/litweb/pdf/ads2008/trans/ads2008/gerber_translator.html, march 26, 2019. [7] “an intuitive explanation of convolutional neural networks,” https://ujjwalkarn.me/2016/08/11/intuitive-explanation-convnets/, march 26, 2019. [8] “convolutional networks,” http://cs231n.github.io/convolutional-networks/, march 26, 2019. [9] “gerber view,” https://sourceforge.net/projects/gerberview/, march 2, 2019. [10] “emgu cv,” http://www.emgu.com/wiki/index.php/main_page, january 8, 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/).  proceedings of engineering and technology innovation , vol. 4, 2016, pp. 07 09 7 performance improvement of a feedback control system using an accelerometer-enhanced velocity observer yu-sheng lu * , chung-heng lee department of mechatronic engineering, national taiwan normal university, taipei 106, taiwan. received 22 february 2016; received in revised form 19 march 2016; accept ed 16 april 2016 abstract the paper shows how an acce lero meter enhanced velocity estimator can be used to improve the tracking performance of a feedback control system. in contrast to conventional v elocity estimators that use positional info rmation only, the accelero meter -enhanced velocity estimator fuses the position sensor and the accelerometer together to produce an improved velocity estimation. expe rimental results are pre sented to show the effectiveness of the accelerometer-enhanced velocity estimator on improving the tracking performance of a linear motion stage. keywor ds : accelerometer, feedback control system, linear motion stage, velocity estimator, velocity observer 1. introduction a linear mot ion stage, shown in fig. 1, is the e xperimental system, whose schematic is shown in fig. 2. in the e xperimental system, a perma nent-magnet synchronous ac motor is driven by a regulator current converter that receives a torque-producing command in the form of an alog voltage fro m the dac interface of a co ntrolle r core . the controlle r core is a dsp/ fpga -based system with dio, adc and dac interfaces. the fpga is configured to interface with an optical linear encoder for p osition counting and velocity detection. here, the velocity detection imple mented in the fpga is based on the so-called inverse-time method (itm) [1] that estimates velocity by measuring the time elapsed during two consecutive rising edges of a quadrature encoder signal. moreover, the adc interface in the fpga acquires acceleration from an accelerometer. fig. 1 photo of the experimental system fig. 2 schematic of the experimental system the dsp reads feedback informat ion on the acceleration, position and velocity (when the itm is used) through the dsp interface in the fpga, calcu lates velocity estimation and co ntrol algorith ms, and sends the control effort to the regulator current converter through the dac interface. the output shaft of the motor is co nnected to a ball screw that translates rotational motion of the rotor to linear motion of the pa yload, on wh ich the accelero meter is mounted. in this paper, the position control of the payload is considered. * corresponding aut hor. email: luys@ntnu.edu.tw proceedings of engineering and technology innovation , vol. 4, 2016, pp. 07 09 8 copyright © taeti 2. method the linear motion stage is used to evalu ate the performance of two veloc ity observers: one is an accelero meter-enhanced velocity estimator, the dcvo [2], and the other is the conventional, popular estimator using the inverse-time method (itm). the plant is modeled as a second -order system described by  dubxaxax  12  (1) in which 857.4 2 a , 01 a , 11432b , x and u denote the plant’s output and input, respectively, and d denotes an uncertain input disturbance. the feedback controller is designed based on the integral variable structure control (ivsc) law [3]. define a switching function zceces 10   (2) where n c 2 0  , 2 1 n c  , and 0 0 zedtz t   . fig. 3 step response with the itm fig. 4 step response with the dcvo here, )( 000 1 10 ececz    and 30 n   . let the ivsc law be     0 1 2 0 1 ˆˆ sgn( ) u x r c e c e c s x d r c e c e s                   (3) in wh ich n c  2 , 1ˆ b   , ˆ0.4   , 2 1ˆ ab   , 0 , and 1.0d . 3. results and discussion fig. 5 output error with a resolution of 20 m fig. 6 output error with a resolution of 5 m two references are used in the following e xperiments: one is a step reference of 10 mm, and the other is a sinusoidal reference of 0.25 hz. figs . 3 and 4 show the step responses with the itm and the dcvo, respectively. it is seen that the switching frequency of the control input associated with the dcvo is much faster than with the itm, meaning that the dcvo enables faster correction of output errors than the itm. the upper subplot of fig. 5 shows the error responses to the step reference, whereas the lower sub plot shows the error responses to the sinusoidal reference. it is seen that compared with the itm, the 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 5 10 15 m m position 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 -0.1 0 0.1 m m error 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 100 200 m m /s error dot 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 -0.5 0 0.5 1 1.5 time (s) v control 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 5 10 15 m m position 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 -0.1 0 0.1 m m error 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 100 200 m m /s error dot 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 -0.5 0 0.5 1 1.5 time (s) v control 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 -0.1 -0.05 0 0.05 0.1 time (s) m m itm dcvo 0 0.5 1 1.5 2 2.5 3 3.5 4 -0.1 -0.05 0 0.05 0.1 time (s) m m itm dcvo 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 -0.1 -0.05 0 0.05 0.1 time (s) m m itm dcvo 0 0.5 1 1.5 2 2.5 3 3.5 4 -0.1 -0.05 0 0.05 0.1 time (s) m m itm dcvo proceedings of engineering and technology innovation, vol. 4, 2016, pp. 07 09 9 copyright © taeti dcvo reduces the output error with the aid of an accelerometer. moreover, in the case of regulation control, output precision is limited by the sensor resolution. in the subsequent experiments, the resolution of the pos itional sensor is reduced to 5 m. like wise; the upper subplot of fig. 6 shows the error responses to the step reference, whereas the lower subplot shows the error responses to the sinusoidal reference. it can be clearly seen that the dcvo outperforms the itm. because of the advance of mems technology, accelerometers become cheap and ubiquitous, ma king the accelerometer-enhanced velocity estimator a cost-effective scheme with improved performance. 4. conclusions in this paper, two velocity estimation schemes have been experimentally evaluated using the same ivsc law. the e xperimental results show that the dcvo enables higher-speed error correction and leads to better tracking precision than the conventional estimator using the itm. acknowledgement the authors would like to thank the ministry of sc ience and technology for their support of this research under grants no. m ost 104-2221 -e-003-011-m y2. references [1] j. s. farrokh, h. vincent, and c. s. j. chen, “discrete-time adaptive windowing for velocity estimation,” ieee trans . contr. syst. technol., vol. 8, no. 6, pp. 1003-1009, 2000. [2] y. s. lu and s. h. liu, “the design and imple mentation of an accelerometer-assisted velocity observer,” isa transactions, vol. 59, pp. 418-423, nov. 2015. [3] j. h. lee, “highly robust position control of blddsm using an improved integral variable structure system,” automatica, vol. 42, pp. 929-935, 2006. proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx non-facial video spatiotemporal forensic analysis using deep learning techniques premanand ghadekar, vaibhavi shetty * , prapti maheshwari, raj shah, anish shaha, vaishnav sonawane department of information technology, vishwakarma institute of technology, pune, india received 11 july 2022; received in revised form 15 september 2022; accepted 04 october 2022 doi: https://doi.org/10.46604/peti.2022.10290 abstract digital content manipulation software is working as a boon for people to edit recorded video or audio content. to prevent the unethical use of such readily available altering tools, digital multimedia forensics is becoming increasingly important. hence, this study aims to identify whether the video and audio of the given digital content are fake or real. for temporal video forgery detection, the convolutional 3d layers are used to build a model which can identify temporal forgeries with an average accuracy of 85% on the validation dataset. also, the identification of audio forgery, using a resnet-34 pre-trained model and the transfer learning approach, has been achieved. the proposed model achieves an accuracy of 99% with 0.3% validation loss on the validation part of the logical access dataset, which is better than earlier models in the range of 90-95% accuracy on the validation set. keywords: transfer learning, mel-spectrogram, forgery, data augmentation 1. introduction surveillance cameras are now found in almost every location, such as banks and businesses, where the recordings are used to reduce crime. however, due to the availability of video editing software like adobe, the process of video editing has become simple [1]. currently, videos are commonly considered the most extensively utilized communication and entertainment medium. hence, this kind of vogue surely emphasizes the utilization of automated perusal and video content understanding using technology. this is referred to as the major goal of computer vision [2]. several methods have been developed for detecting image forgeries, most of which rely on the extraction of specific image modifications in the output image or examination of discrepancies compared to a regular camera pipeline [3]. based on the modification domain, these adjustments can be classified as intra-frame or inter-frame [4]. this study concentrated on forgeries in the video along directions of inter-frame which are ubiquitous in surveillance videos and difficult to detect. a significant gap has been found in existing work in direction of inter-frame forgeries due to the lack of a temporal video forgery dataset. a dataset that consists of various temporal forgeries is created and published on kaggle [5]. the main objective of the proposed research is to come up with a model trained on this dataset that can identify temporal forgeries with an average accuracy of 85% on the validation data. this study also helps identify false audio, reduce the spread of rumors and hate speech, make better-informed decisions, and master the art of fake audio detection. digital authentication and forensics are the conformation and examination of audio for validation of its uniqueness (identify forgery, if any), and it also has a lot of applications [6]. copy-move, deletion, insertion, replacement, and splicing are all methods for audio forgery [7]. for the audio forgery detection of text-to-speech * corresponding author. e-mail address: vaibhavi.shetty19@vit.edu 2 proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx (tts) and voice-conversion (vc) frauds, a resnet-34 using the transfer learning approach is implemented. after successful training of the model, the prediction of audio files and their classification into three categories: real, spoof_tts, and spoof_vc have been done. the asvspoof 2019 dataset has data balance problems [8]. hence, the proposed model introduces a framework to fill that gap, leading to help the models generalize to a wider range of inputs. 2. literature review richard and roussev [9] discussed many digital forensic image/video analysis techniques, incorporating deep learning object identification structure using the yolo method, and chromatic and pattern techniques for object recognition approaches. to accomplish digital video analysis in a forensic context, their study not only covers various forensic visual data analysis issues and resolutions but also describes several unique graphic data analytic techniques. several experimental results for picture enhancing techniques and object recognition methods are shown, demonstrating how yolo in particular may be used to find numerous criminal suspects and crime scene objects, then establish a link between some of them. “yolov3: an incremental improvement,” [10] made transparent yolo upgrades. the findings of multistage and singlestage object detectors are compared in this article. in terms of speed and accuracy, the numbers confirmed that the yolov3 object detector outperforms other object detectors. the goal of virtual/digital media forensics is to establish systems that can automatically assess visual integrity. in the literature, feature-based [11-12] and convolutional neural network (cnn)-based [13-14] integrity analysis approaches had been investigated. most of the proposed techniques for video-based digital forensics attempt to identify computationally cheap alterations, such as dropped or duplicated frames or copy-move operations [15]. ways that differentiate computerfound faces from genuine faces are used to detect face-based interventions. and a two-stream network was proposed to identify two distinct face-swapping manipulations [16]. a new dataset by rossler et al. [17] was especially relevant to practitioners, which has around half a million modified photographs created via feature-based face editing. hinton et al.[18] talked about the limits of cnns for inverse graphics applications, laying the groundwork for a more vigorous “capsule” design. however, due to the absence of an optimization algorithm and the limits of technology at the time, this complicated architecture could not be executed properly. instead, cnns that are simple to create have become popular. sharma and singh [19] proposed a combined technique of image classification that employs transfer learning for feature selection and principal component analysis (pca) for feature reduction. capsule networks have now been created with impressive early results due to the introduction of the expectation-maximizing routing algorithm along with the dynamic routing algorithm [20]. according to sabour et al. [20], stratified pose relationships amongst the pieces of objects are well characterized using the output of a dynamic routing algorithm, i.e., the accordance between capsules. many machine learning algorithms have been specially designed for video forgery detection. to discover counterfeiting, saddique et al. [21] proposed adopting discrete texture analysis in successive frames. christlein et al. [22] analyzed the effectiveness of characteristics for copy-move watermarking that used a multitude of conventional feature sets, including scale-invariant feature transform (sift) and speed-up robust features (surf), and block-based features such as pca, discrete wavelet transform (dwt), discrete cosine transform (dct), and kernel principal component analysis (kpca). using the concept of a near-neighbor-dense field, d’amiano et al. [23] suggested a patch-match-based copy-move detection approach. however, while confronted with huge amounts of data, this method failed miserably. wu et al. [24] suggested a method for detecting frame duplication and frame deletion in vector flow picture sequences by observing velocity and discontinuity peaks. in the moving picture expert group (mpeg) [25] videos, it presented a video forgery copymove detection algorithm. to determine the optical flow coefficient for each region, their method separates each video frame proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 3 into suspected cleared sections. when an uncommon trend in the optical flow coefficient object is found, it indicates counterfeiting. singh and singh [26] proposed a passive blind approach that uses the correlation coefficient and coefficient of variation to detect duplicate frames. wang et al. [27] offered discrete wavelet packet deconstruction and singular point analysis of speech data, to identify audio tampering of time-domain such as audio recognition, addition, replacement, and slicing. it provides a technique for measuring reverberation length for identifying indicators of tampering in audio tapes. they compared pitch to format sequences to detect copy-move forgeries in audio recordings. to identify places of copy-move fraud in a video, a histogram is computed using lbp and a comparison technique is applied. detailed analysis of image and video forgery along with fake video datasets used for tampering has been demonstrated [28]. 3. dataset and attributes this part describes the dataset used for both video and audio forgery detection. the attributes of the datasets such as the number of files, description of files, etc. are mentioned. the representation technique of the audio files is researched and an explanation for choosing the mel-spectrogram has been given. (a) frame augmentation before insertion (b) inserting augmented frames in a random position (c) deleting the original frame from a random position fig. 1 temporal forgery techniques implemented in dataset creation (1) video forgery: a custom dataset for temporal forgery detection has been developed by modifying the dashcam dataset containing 1544 videos [29]. (2) creating a custom dataset: the creation of a dataset for forgery detection has been achieved by introducing seven types of important temporal forgeries in the dashcam dataset, like insertion, deletion, duplication, flipping, rotations, and zooming forgery. fig. 1 shows some of the forgery techniques implemented during dataset creation. training data contains 9448 videos of the dashcam dataset containing non-tampered and tampered videos that are forged by the mentioned forgery techniques. test data contains 2904 videos of the same type. (3) audio forgery: asvspoof 2019 was created for the third automatic speaker verification spoofing and countermeasures challenge. table 1 shows the number of audio files present in the asvspoof 2019 logical access (la) dataset according to their labels. table 1 asvspoof 2019 data distributions train dev eval bonafide 2580 2548 7355 spoof 22800 22296 63882 total 25380 24844 71237 rotation zooming flipping 4 proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 3.1. audio representation technique 3.1.1. spectrogram fourier transform is used to build spectrograms from sound sources. the fourier transform displays the amplitude of each fundamental frequency after dividing a signal into its fundamental frequencies. a spectrogram breaks the length of a sound source into tiny window segments, which are then subjected to the fourier transform to detect the frequencies contained within each window. next, all of those windows’ fourier transforms are then integrated into a single plot. it plots frequency (y-axis) vs. time (x-axis) and uses different colors to show each frequency amplitude. the brightness of the color that represents the signal is proportional to its energy. 3.1.2. chroma features chroma-based characteristics, also known as “pitch class profiles,” are a useful tool for analyzing music with usefully categorized pitches (typically into twelve categories) and tuning which approximates the equal-temperament scale. chromatic and melodic aspects of music are captured by chroma features, which are resistant to changes in timbre and instrumentation. 3.1.3. mel-spectrograms a mel-spectrogram happens to a spectrogram where the frequencies exist convinced to the mel scale. it remaps the principles in hertz to the mel scale as shown in fig. 2. the mel scale can be termed as the scale of pitches perceived by listeners to have the same distance from one another. general frequency measurement has a common reference point which can be defined by equating a 1000 hz tone, with a pitch of 1000 mel and 40 db greater than the listener’s threshold. fig. 2 hertz vs. mel scale representation the formula to convert frequency from hertz into mel scale can be expressed by: 10 2595 log 1 100 f m        (1) where m represents mel on the mel scale and f represents frequency in hertz. a mel-spectrogram forms two influential changes relating to a normal spectrogram that plots frequency intersection time. it uses the mel scale as a suggestion of correction frequency in contact with the y-point around which something revolves. and the decibel scale is used as a suggestion of correction amplitude to signify banners. the proposed research uses the mel-spectrogram-based dataset and it is considered better than other audio representation techniques. the reason has also been demonstrated and experimented on in the experimental section. m e l sc a le hertz scale proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 5 3.2. data augmentation the training, development, and assessment datasets of asvspoof 2019 have a lot of data imbalance, as shown in table 1. as a result, this research shows that developing an augmentation framework that can generate mel-spectrograms from the current datasets while also addressing the dataset’s data imbalance problem, and allowing the network to learn more valuable features. specaugment changes the spectrogram by distorting it in time, masking blocks of successive frequency channels, and masking blocks of utterances in time. time shifting, time masking, and frequency masking are the three primary methods for augmenting data. 3.2.1. time shifting in time shifting, the audio is moved linearly from the left or right with a random second in time shifting. here, the audio is fast-forwarded by a certain interval of x seconds, the first of these x seconds is marked as 0, i.e., silence. then, the shift of the audio to the right (backward) for x seconds again, and the last x seconds are marked as 0, i.e., silence. fig. 3 shows the spectrogram of the original audio and time-shifted audio. (a) spectrogram of original audio (b) spectrogram of time shifted audio fig. 3 spectrogram of time shifted audio 3.2.2. frequency masking masked frequency channels are [f0, f0 + f]. here f0 is selected from (0, v-f), where v denotes the number of frequency channels, and the selection of f is made from a uniform distribution ranging from 0 to masking parameter f. fig. 4 shows the masked mel-spectrogram using frequency masking. 3.2.3. time masking as shown in fig. 5, while doing time masking, the masking of t sequential steps of time [t0, t0 + t] is obtained. the t is selected within a uniform distribution ranging from 0 to a masking parameter t, and from [0, τ – t] t0 is selected. here, τ is the length of the audio file. fig. 4 frequency masking fig. 5 time masking frequency mask time mask 6 proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 4. methodology 4.1. video forgery fig. 6 proposed methodology flow diagram step 1: the proposed model as shown in fig. 6 takes data in the form of multiple sequences of images from videos. the dashcam dataset consists of 1181 videos for training and 363 videos for testing. as shown in fig. 1, after applying the forgery techniques to the existing dashcam dataset, a new dataset with a total of 9448 videos for training and 2904 videos for testing is produced. it is used for video forgery analysis, model training, and validation. step 2: for training, the model takes data in the form of video and extracts some sequences of frames from the video. these multiple sequences of videos are labeled with the category of forged and original, depending upon the type of video from which these frames are extracted. step 3: for making the operation of frame extraction and sequencing faster, pre-extracted frames from the videos are kept in storage. the starting point of the sequence of frames is randomly chosen to avoid overfitting on specific time instances. the length of the sequence or clip is a data-dependent hyperparameter (depends upon video length). the labeled data created in this procedure is passed to the model for training and validation purposes. step 4: the model contains multiple convolutional 3d layers which convolve the sequence of frames to a 3d volume of features as shown in fig. 7. from the convolved output, the model chooses important features using max pooling 3d layers. the dropout layers are added in between the series of max pooling 3d and convolutional 3d layers to avoid model overfitting. the model contains relu as an activation function for the neurons. the output layer consists of two neurons that contribute two classes named forged and original. as the model is a classification problem, it uses a categorical cross-entropy loss function. a stochastic gradient descent optimizer is used to overcome the overhead of the gradient descent algorithm and a learning rate scheduler is used to decay the learning rate with an increase in the number of epochs. the tesla k80 gpu is used for training this model. step 5: table 2 shows the observed metrics while training the model. the accuracy and loss are 85.55% and 30.74% respectively. the precision (positive predictive value) and recall (sensitivity) values are 86.51% and 74.75% respectively. table 3 shows the validation metrics for the model. the accuracy and loss are 82.17% and 35.87% respectively. temporal video forgery dataset generation video data input pre-processing  extracting frames  inserting forgery  creating forged videos prediction on evaluation dataset model training on original and forged dataset proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 7 fig. 7 proposed model of cnn architecture with hyperparameters table 2 training metrics table 3 validation metrics accuracy 0.8555 accuracy 0.8217 loss 0.3074 loss 0.3587 precision 0.8651 recall 0.7475 the formulas for calculating accuracy and loss are shown below: tp tn accuracy fp fn tp tn      (2) fp fn loss fp fn tp tn      (3) a = filter size: 3×3×3, layer: convolution 3d, input channel: 3 relu(a) activation function kernel size: 1×2×2, layer: maxpooling 3d b = filter size: 3×3×3, layer: convolution 3d, input channels: 64 relu(b) activation function kernel size: 2×2×2, layer: maxpooling 3d c = filter size: 3×3×3, layer: convolution 3d, input channel: 128 relu(c) activation function d = filter size: 3×3×3, layer: convolution3d, input channel: 256 relu(d) activation function kernel size: 2×2×2, layer: maxpooling 3d e = filter size: 3×3×3, layer: convolution 3d, input channel: 256 relu(e) activation function f = filter size: 3×3×3, layer: convolution3d, input channel: 512 relu(f) activation function kernel size: 2×2×2, layer: maxpooling 3d j = input channels: 8192, layer: linear, output channels: 4096 k = relu(j) activation function dropout(k) 50% g = filter size: 3×3×3, layer: convolution 3d, input channel: 512 relu(g) activation function h = filter size: 3×3×3, layer: convolution 3d, input channel: 512 relu(h) activation function kernel size: 2×2×2, layer: maxpooling 3d l = input channels: 4096, layer: linear, output channels: 4096 m = relu(l) activation function dropout(m) 50% input channels: 4096, layer: linear, output channels: 2 8 proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx the terms false positives (fp), false negatives (fn), true positives (tp), and true negatives (tn) are derived from the confusion matrix in fig. 8. fig. 8 confusion matrix of the proposed model prediction on the validation dataset 4.2. audio forgery the proposed study takes the training audio data files from the asvspoof 2019 la dataset section and develops a melspectrogram representation of each audio file. data augmentation has been used to address the problem of data imbalance in the dataset. next, the resnet-34 model applies transfer learning to the dataset and supplemented data. the data is then divided into three categories: real, spoofed_tts, and spoofed_vc. the proposed methodology for the research is described in the next section. step 1: this study proposes a voice classifier based on deep convolutional neural networks for detecting spoofing attempts with the help of the asvspoof 2019 dataset and required pre-processing. step 2: in the suggested technique, the train folder is used from the asvspoof 2019 dataset and an audio time-frequency model of power spectral densities on the mel frequency scale (mel-spectrogram). step 3: for deeper residual training, (80-20) train and validation split (for transfer learning on resnet-34 architecture) are designed. the fastai package and the tesla k80 gpu are used to implement this transfer learning approach. the proposed methodology is shown in fig. 9. transfer learning: given the significant computing and time resources required to create neural network models for this challenge, and the significant improvements in the skill that they provide on related problems. it helps in improving the dl models using pre-trained models as a preliminary step in computer vision. a residual network, or resnet for short: it is an artificial neural network that uses skip connections or shortcuts to bypass some layers in the creation of a deeper neural network. skipping enables the creation of deeper network layers without having to deal with vanishing gradients. step 4: on the validation split formed on the train folder files, the first epoch provides an accuracy of 92.47%. a total of 12 epochs have been executed for audio forgery detection on the asvspoof dataset. after every 4 epochs, the learning rate is identified to get minimum loss and then changed for the further epochs. step 5: better accuracy of 99% with 0.03% validation loss on the validation set is achieved after performing 12 epochs and fine-tuning the learning rate of the proposed model. fig. 10 depicts the relationship between loss and learning rate. the final model metrics are shown in fig. 11. original forged o ri g in a l f o rg e d proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 9 fig. 9 proposed methodology flow diagram fig. 10 loss versus learning rate graph (for fine-tuning) fig. 11 metrics of the final model preparation 5. experiments 5.1. video forgery (1) error level analysis (ela) ela is used to recognize parts of the picture with varying compression rates. one of the major drawbacks of ela is that it provides inaccurate recognition when low-quality jpeg images and recoloring are considered [30]. using the ela technique, ela-processed images for the input images (casia2 dataset) are generated as shown in fig. 12. these newly generated images which undergo ela processing are passed to a 2d convolutional neural network with labels attached as “original” and “forged”. the model comes up with an accuracy of 90.28% for the validation part. this experiment is done to see the performance changes in the classification of forged videos frame-by-frame. the outcome is that the performance of this ela metric-based model is not as efficient as the proposed model which works on convolutional audio data input and pre-processing mel-spectrogram generation transfer learning over resnet-34 training on generated spectrograms evaluation on dev and validation dataset tesla k80 gpu 1e-06 1e-05 1e-04 1e-03 1e-02 0.40 0.35 0.25 0.30 0.20 0.15 learning rate l o s s 10 proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 3d cnn layers. here, casia 2 dataset has been used. by using ela, forgery in the spatial domain is detected. this approach can be used to detect a spatial forgery in videos by separating the frames and doing ela analysis frame-by-frame. fig. 12 input image versus forged image versus image after ela processing (2) discrete cosine transform (dct) the dct is a mathematical modification that is essential to the jpeg standard compaction. the primary objective of these procedures is to change a signal from one sort of interpretation to another. the dct may be used to transform the signal (intra-frame information) into quantitative data (“frequency” or “spectral” information), allowing the image to be quantified and compressed. casia 2 dataset is used here. dct is applied to the frame for compression. the difference between the original and compressed frame is obtained to identify the spatial forgery in the frame. dct coefficients are used to identify irregularities due to the spatial domain analysis caused by superimposing an image over another one. (3) using image processing (a non-ai approach) the structural similarity index measure (ssim) checks the similarity between two images by the standard deviation of pixel values of the image. these become the factors that can be used to detect some types of forgeries such as insertion, duplication, copy-moving, and removal of the region of a frame in a video. ssim is a perception-based model that analyzes image deterioration as a perceived change in structural information, as well as crucial perceptual appearances, such as contrast and luminance masking. structural information means the assumption that pixels have a lot of interdependencies, especially when they’re close together in a space. in the proposed approach, ssim and the standard deviation have been used to detect and analyze forgeries that can be embedded between consecutive video frames. this has been done by computing and analyzing the ssim value between two consecutive frames of a video, along with calculating the difference between the standard deviation of pixel values of these two frames. a sudden change in standard deviation values of a frame in a video sequence with a very low similarity index between consecutive frames depicts a high probability of forgery. meanwhile, a frame window is proportionally divided into several segments. ssim and the standard deviations are calculated for each segment and compared with corresponding segments in the previous frame. this results in more accurate forgery detection and localization of forgery. fig. 13 shows the entire flow diagram of the process. forged portion original image forged image ela processed image proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 11 fig. 13 non-ai video forgery detection flow diagram (a) frame 1: untampered video frame segment (b) frame 2: tampered video frame segments in next consecutive frame fig. 14 non-ai approach illustration to identify video forgery in fig. 14, “a” represents the difference between the standard deviation values of the current frame segment and the corresponding segment of the frame previous to it. “b” represents the ssim value between the consecutive frame segments. identification of tampered frame segments is made by setting a minimum threshold to ssim values and a maximum threshold to the difference of standard deviation values between consecutive frame segments. the red highlighted segments indicate forgery in a particular section of the video. by using this technique, temporal forgery analysis of any video can be done. 5.2. audio forgery table 4 shows the comparison of the metrics such as accuracy and f1 scores with other chart types (spectrogram, chroma stft). the proposed model achieves a better performance under experimental conditions. by observing the chart type comparison, chroma stft gives the least accuracy and f1 scores, whereas the mel-spectrogram gives the best accuracy and f1 scores. fig. 15 shows the various methods to convert the visual and audio media transmitted via radio wave signals to an image. it can be seen that the reason behind choosing the mel-spectrogram over the differing present methods is that the spectrogram gives a short “snapshot” of visual and audio media transmitted via radio waves. therefore, it is suitable to recommend cnn-located architectures grown for management representation. fig. 14 shows the confusion matrix generated on the validation part of the asvspoof 2019 la dataset. video data input segmenting video frame applying video metrix:  structure similarity index measure (ssim)  standard deviation (contrast) detection and localization of forgery 12 proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx the counts of fp and fn are quite less than tp and tn. the wrong predictions inform that type 1 errors are more than type 2. however, the error rate is very less on the whole and the errors observed in some samples were because a lot of tweaking/augmentation was done on the testing set. table 4 asvspoof 2019 validation set accuracy versus graphing approach chart type accuracy f1 score classes mel-spectrogram (proposed) 99% 99% real 99% spoof_tts 98% spoof_vc spectrogram 90.33% 90.52% real 95.83% spoof_tts 86.34% spoof_vc chroma stft 82.50% 73.32% real 93.52% spoof_tts 80.10% spoof_vc fig. 15 confusion matrix of the proposed model prediction on the validation dataset 6. conclusions to improve the accuracy and quality of video and audio forgery identifications, two models for their detection are proposed. the experiments lead to the following conclusions: (1) this research on video temporal forgery identification fills the gap in existing work on inter-frame forgery detection due to the lack of temporal video forgery detection. it proposes the use of convolutional 3d layer model architecture with an accuracy of 85.55%. also, a non-ai technique has been developed using metrics like ssim and the standard deviations of the video frame segments to identify runtime temporal forgery. a comprehensive dataset for temporal forgery identification has been created for future research. (2) in audio forgery identification, asvspoof 2019 dataset using transfer learning is proposed. moreover, it proposes a comparative study on various audio representation techniques and a study on why the mel-spectrogram is efficient for audio data. augmentation of data has been done to handle the data imbalance problem. (3) the computational complexity in cnn models utilized in the audio and video forgery algorithms, the number of parameters in each feature map is limited to a constant (usually less than 1) multiplied by the input pixels n. convolving a fixed length filter over an image with n pixels requires o(n) time because each output is just the sum-product of k pixels in the image and k weights in the filter, and k is constant with n. similarly, every max or avg pooling operation takes no more than linear time in terms of input size. hence, the entire runtime remains linear o(n). confusion matrix real spoof_tts spoof_vc predicted s p o o f_ t t s s p o o f_ v c r e a l a c tu a l proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 13 (4) both the video and audio forgery models do not incur any computational overhead. all the processing is done as the requirements of the proposed model. overall, the proposed models achieved the optimal accuracy performance of 99% on the validation dataset with minimal loss. future work of this research can be directed to combining the video and audio forgery detection works. one way of doing this is by extracting audio and visual parts of video and feeding them to respective models. outputs of both models can be combined to generate the final result. conflicts of interest the authors declare no conflict of interest. references [1] s. fadl, q. han, and q. li, “cnn spatiotemporal features and fusion for surveillance video forgery detection,” signal processing: image communication, vol. 90, article no. 116066, january 2021. [2] y. b. deshmukh and s. k. korde, “forensic video/image analytics – a deep learning approach,” international journal of creative research thoughts (ijcrt), vol. 8, no. 9, pp. 411-418, september 2020. [3] d. afchar, v. nozick, j. yamagishi, and i. echizen, “mesonet: a compact facial video forgery detection network,’’ ieee international workshop on information forensics and security (wifs), article no. 8630761, december 2018. [4] j. xiao, s. li, and q. xu, “video-based evidence analysis and extraction in digital forensic investigation,” ieee access, vol. 7, pp. 55432-55442, april 2019. [5] p. ghadekar, p. maheshwari, r. shah, a. shaha, v. sonawane, and v. shetty, “video forgery dataset,” https://www.kaggle.com/datasets/rajshah1/video-forgery-dataset, september 10, 2022. [6] h. malik and h. farid, “audio forensics from acoustic reverberation,” ieee international conference on acoustics, speech and signal processing, pp. 1710-1713, march 2010. [7] c. kraetzer, a. oermann, j. dittmann, and a. lang, “digital audio forensics: a first practical evaluation on microphone and environment classification,” mm&sec '07: proceedings of the 9th workshop on multimedia & security, pp. 63-74, september 2007. [8] j. yamagishi, m. todisco, m. sahidullah, h. delgado, x. wang, n. evans, et al., “asvspoof 2019: the 3rd automatic speaker verification spoofing and countermeasures challenge database,” [sound]. university of edinburgh. the centre for speech technology research (cstr), https://doi.org/10.7488/ds/2555. [9] i. i. i. richard and v. roussev, “digital forensic tools: the next generation,” digital crime and forensic science in cyberspace, igi global, pp. 75-90, april 2006. [10] j. redmon and a. farhadi, “yolov3: an incremental improvement,” university of washington, technical report, article no. 1804.02767, april 2018. [11] h. farid, photo forensics, the mit press, february 2019. [12] d. güera, y. wang, l. bondi, p. bestagini, s. tubaro, and e. j. delp, “a counter-forensic method for cnn-based camera model identification,” ieee conference on computer vision and pattern recognition workshops (cvprw), pp. 1840-1847, july 2017. [13] d. güera, f. zhu, s. k. yarlagadda, s. tubaro, p. bestagini, and e. j. delp, “reliability map estimation for cnn-based camera model attribution,” ieee winter conference on applications of computer vision (wacv), pp. 964-973, march 2018. [14] p. bestagini, s. milani, m. tagliasacchi, and s. tubaro, “local tampering detection in video sequences,” ieee 15th international workshop on multimedia signal processing (mmsp), pp. 488-493, september-october 2013. [15] d. graupe, “principles of artificial neural networks,” advanced series in circuits and systems, vol. 7, world scientific, 2013. [16] p. zhou, x. han, v. i. morariu, and l. s. davis, “two-stream neural networks for tampered face detection,” ieee conference on computer vision and pattern recognition workshops (cvprw), pp. 1831-1839, july 2017. [17] a. rössler, d. cozzolino, l. verdoliva, c. riess, j. thies, and m. nießner, “faceforensics: a large-scale video dataset for forgery detection in human faces,” arxiv preprint, article no. 1803.09179, march 2018. [18] g. e. hinton, a. krizhevsky, and s. d. wang, “transforming auto-encoders,” artificial neural networks and machine learning – icann 2011, lecture notes in computer science, vol. 6791, pp. 44-51, 2011. 14 proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx [19] r. sharma and a. singh, “an integrated approach towards efficient image classification using deep cnn with transfer learning and pca,” advances in technology innovation, vol. 7, no. 2, pp. 105-117, april 2022. [20] s. sabour, n. frosst, and g. e. hinton, “dynamic routing between capsules,” proceedings of the 31st international conference on neural information processing systems (nips'17), pp. 3859-3869, december 2017. [21] m. saddique, k. asghar, u. i. bajwa, m. hussain, and z. habib, “spatial video forgery detection and localization using texture analysis of consecutive frames,” advances in electrical and computer engineering, vol. 19, no. 3, pp. 97-108, 2019. [22] v. christlein, c. riess, j. jordan, c. riess, and e. angelopoulou, “an evaluation of popular copy-move forgery detection approaches,” ieee transactions on information forensics and security, vol. 7, no. 6, pp. 1841-1854, december 2012. [23] l. d’amiano, d. cozzolino, g. poggi, and l. verdoliva, “a patchmatch-based dense-field algorithm for video copy–move detection and localization,” ieee transactions on circuits and systems for video technology, vol. 29, no. 3, pp. 669-682, march 2019. [24] y. wu, x. jiang, t. sun, and w. wang, “exposing video inter-frame forgery based on velocity field consistency,” ieee international conference on acoustics, speech and signal processing (icassp), pp. 2674-2678, may 2014. [25] g. ulutas, b. ustubioglu, m. ulutas, and v. v. nabiyev, “frame duplication detection based on bow model,” multimedia systems, vol. 24, no. 5, pp. 549-567, october 2018. [26] g. singh and k. singh, “video frame and region duplication forgery detection based on correlation coefficient and coefficient of variation,” multimedia tools and applications, vol. 78, no. 9, pp. 11527-11562, may 2019. [27] z. wang, y. yang, c. zeng, s. kong, s. feng, and n. zhao, “shallow and deep feature fusion for digital audio tampering detection,” eurasip journal on advances in signal processing, vol. 2022, article no. 69, 2022. https://doi.org/10.1186/s13634-022-00900-4 [28] f. h. chan, y. t. chen, y. xiang, and m. sun, “anticipating accidents in dashcam videos,” computer vision – accv 2016, vol. 10114, pp 136-153, 2016. [29] s. tyagi and d. yadav, “a detailed analysis of image and video forgery detection techniques,” the visual computer, 2022, in press. https://doi.org/10.1007/s00371-021-02347-4 [30] i. b. k. sudiatmika, f. rahman, t. trisno, and s. suyoto, “image forgery detection using error level analysis and deep learning,” telecommunication computing electronics and control (telkomnika), vol. 17, no. 2, pp. 653-659, 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/). https://doi.org/10.1186/s13634-022-00900-4 https://doi.org/10.1007/s00371-021-02347-4  proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx development of a new ground motion model for a peninsular indian rock site ravi kiran akella * , mohan kumar agrawal, jayanta chattopadhyay reactor safety division, bhabha atomic research centre, mumbai, india received 20 july 2022; received in revised form 29 september 2022; accepted 04 october 2022 doi: https://doi.org/10.46604/peti.2022.10526 abstract the ground motion model (gmm) plays a vital role in the generation of seismic design basis ground motion parameters. even though many intra-plate gmms are available, very few of them are based on peninsular india (pi) region-specific seismological parameters. hence, it is imperative to develop a gmm using seismological parameters derived from earthquakes in the peninsular indian region. in this study, a new gmm is developed for a pi rock site. due to the scarcity of real earthquakes, artificial earthquake records are simulated to generate a new gmm for pi. the accelerograms of these artificial earthquakes are obtained from the stochastic finite fault simulation technique. region-specific seismological parameters are obtained from the available pi earthquakes. the generated gmm is compared with other intra-plate gmms for different earthquake magnitudes. also, the generated gmm is validated with the koyna earthquake record and it is observed that the gmm’s predictions are closer to the record. keywords: ground motion model, anelastic attenuation factor, high-frequency decay parameter, stress-drop 1. introduction the seismic design of engineering structures requires the evaluation of ground motion parameters during a seismic event such as peak ground acceleration (pga) or spectral acceleration (sa). the ground motion model (gmm) is a crucial element in seismic hazard assessment to evaluate ground motion parameters due to seismic events. gmm provides ground motion parameters, and pga or sa are functions of earthquake magnitude and distance. gmm can be developed using regression from actual earthquake records or using artificial earthquakes. method of regression from actual earthquake records can be used for inter-plate regions, where plenty of real earthquake records are available. for intra-plate regions, seismicity rates are less and hence the method of regression from artificial earthquakes is more suitable for the generation of gmm. intra-plate regions have relatively less fractured crust than inter-plate regions due to less seismicity rates. therefore, the seismic ground motion characteristics of intra-plate earthquakes are different from inter-plate earthquakes. intra-plate earthquakes have less attenuation rate due to less fractured crust. seismo-tectonically, india is divided into two broad regions [1]. the first region is a seismo-tectonically active inter-plate region, which includes the himalayan region, the south tibet plateau, andaman–sumatra, etc. the second region is peninsular india (pi), in which seismic activity is relatively less in comparison with the first region. it has been designated as a stable continental region (scr) and is considered an intra-plate region. sharma et al. [2] provided gmm based on earthquake data for the himalayan region, which is an active inter-plate region. ramkrishnan et al. [3] generated gmm for the north and central himalayan region using actual earthquake record data. bajaj and anbazhagan [4] provided attenuation relations for the active * corresponding author. e-mail address: arkiran@barc.gov.in tel.: +91-22-25993546 proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 2 region using limited strong ground motion data. raghu kanth and iyengar [5] and ndma [6] provided a gmm for pi. also, intra-plate gmms generated for eastern north america (ena) may be applicable for pi due to the similarity in seismo-tectonic features [7]. atkinson and boore [8] provided attenuation relations for ena hard-rock sites in 2011. this is a modification for the earlier ground motion relation by atkinson and boore [9]. gmm provided by toro et al. [10] is derived from stochastic simulation and is valid for central north america and ena. the hybrid empirical gmm by pezeshk et al. [11] is also applicable to ena and termed “peza”. the gmm provided by campbell [12] is applicable for hard-rock sites of ena. this was generated using a hybrid approach of gmm generation for ena from western north america (wna) data. silva et al. [13] provided gmm for ena using regression analysis of simulations. a list of abbreviations used for all gmms is given in table 1. table 1 abbreviations used for gmms no. ref. abbreviation used 1 this study present-gmm 2 kanth and iyengar [5] ri-2007 3 ndma [6] ndma-2010 4 atkinson and boore [8] ab-2011 5 toro et al. [10] toro-1997 6 pezeshk et al. [11] peza-2011 7 campbell [12] cb-2003 8 silva et al. [13] silva-2002 the major limitation of these gmms [5-13] is that they are either based on seismological parameters of ena and wna or several assumptions in consideration of seismological parameters. hence, it is imperative to develop a gmm using seismological parameters derived from earthquakes in the peninsular indian region. in india, many nuclear power plants and nuclear facilities are located in the peninsular indian region. as very limited earthquake data is available for this intra-plate region, regression from artificial earthquakes is considered to be suitable for the generation of gmm for a pi site. the stochastic finite fault simulation technique is used to generate artificial earthquakes. a seismic hazard assessment is carried out for a proposed nuclear power plant site in pi [14-15]. the purpose of seismic hazard assessment is to generate seismic design basis parameters for the nuclear power plant site [14-15] using probabilistic seismic hazard analysis (psha). in this study, the logic tree approach was considered in which all available gmms [5-13] were considered. however, it was observed that elsewhere gmms [7-13] are based on seismological parameters of ena and wna and two gmms [5-6] of the indian region predict very high sas than expected in the region [14-15]. hence, in the present study, a new gmm is developed using seismological parameters derived from earthquakes in the peninsular indian region. a typical hard-rock site that has a shear wave velocity (vs30) of 2.9 km/s (as per soil profile type classification of the national earthquake hazards reduction program (nehrp)) [14-15] is considered for the present study. the region-specific seismological characteristics, such as the high-frequency decay parameter, anelastic attenuation factor, and stress parameter considered for the present study are obtained from literature on pi earthquakes. the details of various seismological parameters are given in subsequent sections. 2. stochastic finite fault model for ground motion simulation the generation of artificial earthquake records is done by using the stochastic finite fault modeling technique [16]. in this technique, the fault plane is modeled as m × n sub-faults. for 𝑖𝑗th sub-fault, the acceleration spectrum 𝐴𝑖𝑗 (𝑓) is given in eq. (1). proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 3         2 0 2 0 2 exp exp / 1 ij ij ij ij ij cm f fr a f f r qf                (1) where 𝑀0𝑖𝑗 , 𝑅𝑖𝑗 , and 𝑓0𝑖𝑗 are seismic moment, hypo-central distance, and corner frequency for 𝑖𝑗th sub-fault, kappa (k) is the high-frequency decay parameter and q is the quality factor. as shown in eq. (1), exp(−𝜋𝑓𝑘) models the high-frequency de-amplification from near-surface materials and is characterized by k, known as the high-frequency decay parameter. another term exp ( −𝜋𝑓𝑅𝑖𝑗 𝑄𝛽 ) /𝑅𝑖𝑗 models the path effect, which includes geometric attenuation (1/r) and anelastic attenuation. the anelastic attenuation is inversely related to q. corner frequency, 𝑓0𝑖𝑗 , is related to the stress drop and is given by: 1/3 0 0 4.9 6 ij ij f e m             (2) where ∆σ is the stress drop in bars, 𝑀0𝑖𝑗 is a seismic moment in dyne/cm and is given by log 𝑀0 = 1.5𝑀 + 16.1 and  is shear wave velocity in km/s. 3. seismological parameters for simulation of artificial earthquakes pi has very limited real earthquake data. seismological parameters such as high-frequency decay parameter, anelastic attenuation factor and stress drop obtained from pi earthquake data are used for the simulation of artificial earthquakes. the details of seismological parameters for the simulation of artificial earthquakes are given below. 3.1. kappa (k) for peninsular india (pi) as discussed, exp (−𝜋𝑓𝑘) models a high-frequency de-amplification from near-surface materials. the k, known as a high-frequency decay parameter, characterizes the high-cut filter. the k is usually obtained from the slope of the fourier acceleration spectrum at the source or near-field [16]. high-frequency de-amplification effects are represented by two different approaches in the literature. the first approach was given by anderson and hough [17] and also used by boore [18]. and it can be expressed by:  ( ) expp f f   (3) the second approach given by hanks [19] is provided by: (1/ 2) 8 ( ) 1 m f p f f             (4) where 𝑓𝑚 is a cut-off frequency. high-frequency de-amplification given by the two approaches will be the same if the following equation is satisfied [19]. it can be expressed by: 1 m f   (5) proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 4 the jabalpur earthquake occurred on may 21st, 1997 [20], and its magnitude (mw) is 5.8. the sa of this earthquake record is observed to fall beyond a frequency of 32 hz. hence, singh et al. [20] considered cut-off frequency (𝑓𝑚) as 35 hz. for the same high-frequency de-amplification, the corresponding value of k is about 0.01 using eq. (5). on october 16th, 2000 [21], another jabalpur earthquake occurred with mw 4.7. in this earthquake record, the sa is observed to fall beyond 17 hz. for the same high-frequency de-amplification, the corresponding value of k is about 0.0187 using eq. (5). for the jaitapur site [22], the range of k is from 0.01 to 0.016 with a median of 0.013. in the present work, k of 0.01, 0.014, and 0.018 are used for artificial earthquake generation. 3.2. quality factor (q) for pi the anelastic attenuation is inversely related to q. the term exp ( −𝜋𝑓𝑅𝑖𝑗 𝑄𝛽 ) /𝑅𝑖𝑗 models the path effect, which includes geometric attenuation (1/r) and anelastic attenuation. mandal and rastogi [23] suggested the frequency-dependent quality factor, q as 169𝑓 0.77 for the koyna region. as the koyna region has a low q-value, it remains a tectonically active area. q-value obtained from the indian shield from the 1997 jabalpur earthquake is 508𝑓 0.48 [20]. the same value is obtained from the 2000 jabalpur earthquake [21]. for the jaitapur site [22], the chosen q-values are 84𝑓 0.65, 118𝑓 0.65, and 152𝑓 0.65. it is to be noted that the q0-values of the indian shield region are between 200 and 600 [23]. hence, the q-factor of 508𝑓 0.48 obtained from the pi earthquake is considered due to proximity and geotechnical similarity. 3.3. stress parameter (∆σ) for pi it is defined as the reduction in the value of stress across a fault during a seismic event. singh et al. [20-21] recommended using 100 and 300 bar as stress drop based on the spectral analysis of the 1997 jabalpur earthquake [20] and 2000 jabalpur earthquakes [21]. for the 1967 koyna earthquake (mw = 6.3), the predicted ground motion matches the record for a stress drop of 100 bar [20]. the range of 100-300 bar is considered by raghu kanth and iyengar [5] for pi. for the jaitapur site [22], 140, 170, and 200 bar values are used for the generation of seismic design parameters. hence, for the simulation of artificial earthquakes, 100 bar, 200 bar, and 300 bar are chosen for stress drop. 3.4. rupture width and length wells and coppersmith [24] provided the empirical relations for rupture width and rupture length. these relations were obtained from the analysis of various historical earthquakes. the equations used are as follows: log( )a a b m   (6) log( )w c d m   (7) where m is the magnitude, a is the rupture area (km 2 ), w is the rupture width (km), a = -3.49 and b = 0.91 (from 148 earthquake records), and c = -1.01 and d = 0.32 (from 153 earthquake records) [24]. 3.5. site amplification site amplification occurs when seismic waves travel from a rupture source of high shear wave velocity to a surface of a lower value. this is caused by the change in seismic impedance due to the shear wave velocity gradient. in the present study, site amplification for a hard-rock site with a shear wave velocity of 2.9 km/s [25] is used. the site amplification used is given in table 2. proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 5 table 2 site amplification with frequency for a hard-rock site with 𝛽 = 2.9 km/s frequency (hz) sa 1e-2 1.0 0.1 1.02 0.2 1.03 0.3 1.05 0.5 1.07 0.9 1.09 1.25 1.11 1.8 1.12 3.0 1.13 5.3 1.14 8.0 1.15 14.0 1.15 100.0 1.15 4. simulation of artificial earthquake records parameters for the generation of artificial earthquake records are given in table 3. based on the maximum magnitude of the peninsular indian region, a magnitude range of 4.5-7.5 is considered. for a generation of design basis ground motion, an area of a 300 km radius around the site is usually considered. hence, 5-300 km is considered for the epi-central distance range. depth of focus is varied based on a study for seismic hazard assessment for a typical peninsular indian site [14]. the generation is carried out using finite-fault stochastic simulation methodology [16]. a total 12,852 number of artificial earthquake records are generated, for various combinations of parameters given in table 3. from these 12,852 records, response spectra for 5% damping are obtained as shown in fig. 1. it can be seen that the pga (acceleration corresponding to the frequency of 100 hz) of these 12,852 response spectra has a range of 0.5-1440 gals. table 3 parameters for generation of artificial earthquake records no. parameter median value range/value 1 k 0.014 0.01-0.018 2 q 508𝑓 0.48 3 ∆σ 200 100-300 4 mw 6.0 4.5, 5.0, 5.5, 6.0, 6.5, 7.0, 7.5 5 epi-central distance (km) 5, 10, 20,30, 40, 50, 60, 70, 80, 90, 100, 125, 150,175, 200, 250, 300 6 depth of focus (km) d = 5 km (mw ≤ 5.5) d = 10 km (5.5 < mw  6.5) d = 15 km (6.5 ≤ mw  7.0) d = 20 km (mw ≥7.0) fig. 1 response spectra (5% damping) of all artificial records 0.1 1 10 100 1e-3 0.01 0.1 1 10 100 1000 10000 1e-3 0.01 0.1 1 10 100 1000 10000 present-gmm (m = 7.5, rh = 20.615 km) a c c e le ra ti o n ( g a l) frequency (hz) present-gmm (m = 4.5, rh = 300.04 km) proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 6 5. development of the present-gmm the form of the present-gmm is similar to that of an attenuation relation for the intra-plate region [9]. the form of the present-gmm is given in eq. (8).      210 1 2 3 4 5 2 6 7 3 8 9 1 10log bry c c m c m c c m a c c m a c c m a c r          (8) where, 𝑎1 = max [log ( 10 𝑅 ) , 0], 𝑎2 = max[log(𝑅), log70], and 𝑎3 = max [log ( 𝑅 130 , 0)]. it is to be noted that m and r in eq. (8) correspond to moment magnitude and hypo-central distance respectively. coefficients of the above frequency-dependent attenuation relation are given in table 4. table 4 coefficients of present frequency-dependent gmm frequency (hz) c1 c2 c3 c4 c5 c6 c7 c8 c9 c10 r-square 0.1 -3.5055 0.9682 -0.0814 -5.3108 0.6170 -2.3235 0.1638 4.5892 -1.1207 0.0009 0.972 0.3 -6.8364 2.0656 -0.1658 -4.1702 0.5041 -1.4664 0.0508 6.5860 -1.3171 0.0004 0.968 0.5 -8.6835 2.6196 -0.1999 -3.1674 0.3563 -1.1284 0.0517 9.7760 -1.8752 -0.0001 0.965 0.7 -8.6839 2.6375 -0.2008 -2.6751 0.2952 -0.9429 0.0856 11.2086 -2.2262 -0.0009 0.958 1 -8.1259 2.5228 -0.1914 -2.3588 0.2447 -0.9917 0.0937 8.6866 -1.7789 -0.0010 0.949 1.5 -7.1146 2.2310 -0.1669 -1.9500 0.1858 -0.6576 0.0737 3.8247 -0.8046 -0.0016 0.936 2 -5.7435 1.9008 -0.1437 -2.0452 0.1856 -1.0270 0.0750 -1.3918 0.1522 -0.0012 0.929 2.5 -5.0697 1.7303 -0.1324 -2.0022 0.1826 -0.7988 0.0708 -1.3740 0.1002 -0.0016 0.922 3 -4.5260 1.5970 -0.1226 -2.0131 0.1778 -1.3084 0.0974 -1.0230 0.0340 -0.0012 0.915 3.3 -4.2627 1.5382 -0.1194 -2.0350 0.1829 -1.1641 0.0819 -0.9120 -0.0348 -0.0014 0.915 3.6 -4.0726 1.4774 -0.1141 -1.9689 0.1732 -1.0838 0.0702 -1.3838 0.0529 -0.0015 0.911 4 -3.4808 1.3282 -0.1051 -2.0671 0.1908 -0.8624 0.0627 -3.2702 0.4343 -0.0018 0.909 5 -3.1783 1.2320 -0.0960 -1.9337 0.1687 -1.1496 0.1079 -3.4047 0.5771 -0.0020 0.907 6 -2.6408 1.1218 -0.0905 -2.0650 0.1878 -0.9777 0.0924 -3.2664 0.4486 -0.0023 0.909 8 -2.4356 1.0841 -0.0879 -2.1003 0.1878 -1.0015 0.0761 -2.5430 0.2836 -0.0022 0.908 9 -2.1751 1.0188 -0.0825 -2.0834 0.1809 -1.3408 0.1207 -0.1316 -0.2071 -0.0023 0.908 10 -1.8273 0.9393 -0.0783 -2.1663 0.1938 -1.2833 0.1236 1.5041 -0.5826 -0.0025 0.908 12.5 -1.6384 0.9151 -0.0787 -2.2552 0.2077 -1.2673 0.1265 2.5458 -0.8271 -0.0027 0.909 15 -1.7204 0.9197 -0.0763 -2.1271 0.1850 -1.4519 0.1621 6.2029 -1.4411 -0.0030 0.910 18 -1.4749 0.8703 -0.0747 -2.2080 0.1992 -1.6288 0.2086 8.4817 -1.9191 -0.0033 0.907 20 -1.2144 0.8106 -0.0720 -2.2910 0.2126 -1.5401 0.2107 5.8524 -1.4275 -0.0036 0.909 22 -1.2975 0.8303 -0.0726 -2.2539 0.2044 -1.8178 0.2593 4.0384 -1.0584 -0.0037 0.906 25 -1.3511 0.8235 -0.0702 -2.1772 0.1899 -1.8758 0.2633 4.5023 -1.1098 -0.0036 0.903 28 -1.1998 0.8011 -0.0708 -2.3045 0.2099 -1.8330 0.2738 6.6242 -1.5658 -0.0038 0.901 31 -1.3720 0.8413 -0.0729 -2.2679 0.2029 -1.8068 0.2831 8.2891 -1.8609 -0.0039 0.901 34 -1.2753 0.8333 -0.0740 -2.3614 0.2149 -1.8382 0.2942 8.8317 -1.9756 -0.0039 0.901 37 -1.2287 0.8351 -0.0752 -2.4346 0.2224 -1.8425 0.2877 7.1788 -1.6717 -0.0037 0.901 40 -1.1890 0.8242 -0.0753 -2.4789 0.2313 -1.7675 0.2905 6.3510 -1.5040 -0.0038 0.901 45 -1.1364 0.8124 -0.0758 -2.5290 0.2416 -1.7204 0.2976 6.2781 -1.4935 -0.0039 0.901 50 -1.1441 0.8153 -0.0773 -2.5934 0.2530 -1.6469 0.2851 5.5864 -1.3394 -0.0038 0.903 100 -1.2010 0.8210 -0.0771 -2.5804 0.2491 -1.7180 0.2882 4.7007 -1.1373 -0.0036 0.911 proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 7 the standard deviation log10𝑦𝑏𝑟 at all frequencies is 0.55. the r-square value for each frequency is provided in table 4. it can be seen that r-square values are more than 0.9 for all frequencies, which indicates that a good fit is chosen for regression. it is a standard practice to plot the regression residuals to verify whether the correct fit is chosen to reproduce the simulations. if there is no significant residual trend with distance or magnitude, it can be inferred that a good fit is chosen for regression. regression residuals versus distance are plotted for mw 5.0 as shown in fig. 2. fig. 3 shows the plot of regression residuals with a distance of mw 6.0. the plot of regression residuals for mw 7.0 is shown in fig. 4. it can be seen that no considerable residual trend is observed from these figures. fig. 2 plot of residuals for mw 5.0 fig. 3 plot of residuals for mw 6.0 fig. 4 plot of residuals for mw 7.0 the present-gmm is valid for a magnitude range of 4.5-7.5 and an epi-central distance range of 5-300 km. the lowest possible spectral acceleration (sa) by present-gmm corresponds to a magnitude of 4.5 and an epi-central distance of 300 km. as the depth of focus considered for a magnitude of 4.5 is 5 km, the corresponding hypo-central distance is 300.04 km. the highest possible sa by present-gmm corresponds to a magnitude of 7.5 and an epi-central distance of 5 km. as the depth of focus considered for a magnitude of 7.5 is 20 km, the corresponding hypo-central distance is 20.615 km. the lowest and highest possible sas by present-gmm (eq. (8)) are shown in fig. 1. it can be seen that the pga (acceleration corresponding to the frequency of 100 hz) predicted by gmm has a range of 1-300 gals, which is lying in the range of individual records. 6. comparison of predictions of present-gmm with other intra-plate ground motion relations for hard-rock the comparison of predictions of present-gmm, with other intra-plate ground motion relations for hard-rock for mw 5.0 is shown in fig. 5. the prediction of various gmms for pga corresponding to various distances (hypo-central distance) and 10 100 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 l o g r e s id u a l (f o r p g a ) r (km) 10 100 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 l o g r e s id u a l (f o r p g a ) r (km) 10 100 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 l o g r e s id u a l (f o r p g a ) r (km) proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 8 mw 5.0 is shown in fig. 5(a). it can be seen that ndma-2010 and ri-2007 predict much higher pga values compared to other gmms. peza-2011 predicts higher pga values for distances up to 15 km. the present-gmm’s predictions are lying in the band of predictions by other gmms. various gmm predictions of 5% damped response spectra for mw 5.0 and rh = 50 km are given in fig. 5(b). (a) pga of mw 5.0 (b) 5% damped response spectra for mw 5.0 and rh = 50 km fig. 5 comparison of other hard-rock ground motion relations with the present-gmm for mw 5.0 it can be seen that ndma-2010 and ri-2007 predict higher sas compared to other gmms. the response spectrum predicted by the present-gmm lies in the band of predictions by other gmms. also, the sas predicted by the present-gmm are in the range of predictions of other international gmms. however, the frequency band is relatively lower than those of other gmms. the comparison for mw 6.0 is shown in fig. 6. the pga comparison for various distances is shown in fig. 6(a). response spectral comparison for mw 6.0 and rh = 50 km is given in fig. 6(b). it can be seen that ndma-2010 and ri-2007 have given higher predicted pga and sa values in comparison with other gmms. it can be observed that predictions by present-gmm are lying in the range of predictions of other gmms. (a) pga of mw 6.0 (b) 5% damped response spectra for mw 6.0 and rh = 50 km fig. 6 comparison of other hard-rock ground motion relations with the present-gmm for mw 6.0 the comparison for mw 7.0 is shown in fig. 7. the pga comparison for various distances is shown in fig. 7(a). response spectral comparison for mw 7.0 and rh = 50 km is given in fig. 7(b). it can be seen that ndma-2010 and ri-2007 have given higher predicted pga and sa values when compared with other gmms. however, for short distances (up to 30 km), ndma-2010 predictions are relatively lower than ri-2007. it can be noticed that predictions by the present-gmm are lying in the range of predictions of other intra-plate ground motion relations for hard-rock. 10 100 0.0 0.1 0.2 0.3 0.4 present-gmm ab-2011 toro-1997 peza-2011 cb-2003 silva-2002 ndma-2010 ri-2007 p g a ( g ) distance (km) 0 20 40 60 80 100 0.00 0.03 0.06 0.09 0.12 0.15 present-gmm ab-2011 toro-1997 peza-2011 cb-2003 silva-2002 ndma-2010 ri-2007 s a ( g ) frequency (hz) 10 100 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 present-gmm ab-2011 toro-1997 peza-2011 cb-2003 silva-2002 ndma-2010 ri-2007 p g a ( g ) distance (km) 0 20 40 60 80 100 0.0 0.1 0.2 0.3 0.4 present-gmm ab-2011 toro-1997 peza-2011 cb-2003 silva-2002 ndma-2010 ri-2007 s a ( g ) frequency (hz) present-gmm ab-2011 toro-1997 peza-2011 cb-2003 silva-2002 ndma-2010 ri-2007 present-gmm ab-2011 toro-1997 peza-2011 cb-2003 silva-2002 ndma-2010 ri-2007 proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 9 (a) pga of mw 7.0 (b) 5% damped response spectra for mw 7.0 and rh = 50 km fig. 7 comparison of other hard-rock ground motion relations with the present-gmm for mw 7.0 7. comparison of present-gmm predictions with strong ground motion data of koyna earthquake strong ground motion availability for the peninsular indian region is scarce. the koyna earthquake is one of the few earthquakes that occurred in the peninsular indian region. an mw 6.5 earthquake struck on december 10th, 1967 in koyna [26]. this is one of the major seismic events for which instrumental strong ground motion data is available. the horizontal acceleration record of this earthquake at an epi-central distance of 12.74 km is shown in fig. 8. corresponding response spectrum along with estimated spectra from various gmms are shown in fig. 9. a comparison of the frequency content of the koyna earthquake with other gmms is shown in fig. 10. fig. 8 accelerogram of koyna earthquake it is observed that the response spectrum predicted by the present-gmm compares well with the actual earthquake record. from the frequency content shown in fig. 9, it is observed that response spectra of other internationally available gmms contain relatively higher frequency content as compared to the present-gmm and koyna earthquake records. this can be attributed to the high-cut filter term in the ground motion simulations. it may also be due to the relatively smaller value of k = 0.005 for ena [9] as compared to the relatively higher value of k = 0.01-0.018 for pi. 10 100 0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00 present-gmm ab-2011 toro-1997 peza-2011 cb-2003 silva-2002 ndma-2010 ri-2007 p g a ( g ) distance (km) 0 20 40 60 80 100 0.00 0.25 0.50 0.75 1.00 present-gmm ab-2011 toro-1997 peza-2011 cb-2003 silva-2002 ndma-2010 ri-2007 s a ( g ) frequency (hz) 0 5 10 15 20 -0.5 -0.4 -0.3 -0.2 -0.1 0.0 0.1 0.2 0.3 0.4 0.5 a c c e le ra ti o n ( g ) time (s) present-gmm ab-2011 toro-1997 peza-2011 cb-2003 silva-2002 ndma-2010 ri-2007 s a (g ) proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 10 fig. 9 comparison of response spectrum of koyna earthquake with other hard-rock ground motion relations and present-gmm fig. 10 comparison of the frequency content of the koyna earthquake with otherhard-rock ground motion relations and present-gmm 8. results and discussion seismological parameters derived from earthquakes in the peninsular indian region are used to generate 12,852 artificial earthquake records. constants of the present-gmm are obtained by regression of response spectra of all these records. the present-gmm is applicable for mw 4.5-7.5 and an epi-central distance range of 5-300 km. the predictions of present-gmm are compared with other intra-plate ground motion relations for hard-rock. it is observed that two other earlier gmms of pi have predicted higher pga and sa values when compared with other gmms. it is also observed that predictions by present-gmm are lying in the range of predictions of other gmms. the sas predicted by the present-gmm compare well with that of the actual koyna earthquake. sas of other internationally available gmms contain relatively higher frequency content as compared to the present-gmm and the koyna earthquake record. this can be attributed to the relatively higher value of the spectral decay parameter for pi when compared with that of ena. the present-gmm can be further improved by refining the seismological parameters using future earthquake records if any. 9. conclusions the present-gmm based on the regression of simulated accelerograms was generated for a pi hard-rock site. the following conclusions are drawn from the present study: (1) as artificial earthquakes are simulated using seismological parameters from actual earthquakes of pi, the present-gmm is applicable. (2) as r-square values are more than 0.9 for all frequencies, it can be confirmed that a good fit is chosen for regression. (3) present-gmm is compared with other intra-plate ground motion relations for hard-rock. the comparison shows that the present-gmm’s predictions are well within the band of predictions by other intra-plate ground motion relations. (4) finally, the present-gmm is validated with the record of the koyna earthquake of pi. it is observed that the present gmm’s predictions are closer to the actual earthquake record of the peninsular indian region as compared to other gmms. therefore, it can be concluded that the present-gmm can be used to evaluate ground motion parameters for the peninsular indian hard-rock site. 0.1 1 10 100 1e-3 0.01 0.1 1 koyna record mean (present-gmm) mean+sigma (present-gmm) mean-sigma (presentgmm) ab-2011 toro-1997 peza-2011 cb-2003 silva-2002 ndma-2010 ri-2007 s a ( g ) frequency (hz) 0 20 40 60 80 100 0 1 2 3 koyna record present-gmm ab-2011 toro-1997 peza-2011 cb-2003 silva-2002 ndma-2010 ri-2007 d a f frequency (hz) proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 11 conflicts of interest the authors declare no conflict of interest. references [1] s. k. nath and k. k. s. thingbaijam, “peak ground motion predictions in india: an appraisal for rock sites,” journal of seismology, vol. 15, no. 2, pp. 295-315, april 2011. [2] m. l. sharma, j. douglas, h. bungum, and j. kotadia, “ground-motion prediction equations based on data from the himalayan and zagros regions,” journal of earthquake engineering, vol. 13, no. 8, pp. 1191-1210, 2009. [3] r. ramkrishnan, k. sreevalsa, and t. g. sitharam, “development of new ground motion prediction equation for the north and central himalayas using recorded strong motion data,” journal of earthquake engineering, vol. 25, no. 10, pp. 1903-1926, 2021. [4] k. bajaj and p. anbazhagan, “determination of gmpe functional form for an active region with limited strong motion data: application to the himalayan region,” journal of seismology, vol. 22, no. 1, pp. 161-185, january 2018. [5] s. t. g. raghu kanth and r. n. iyengar, “estimation of seismic spectral acceleration in peninsular india,” journal of earth system science, vol. 116, no. 3, pp. 199-214, june 2007. [6] working committee of experts (wce), “development of probabilistic seismic hazard map of india,” ndma (national disaster management authority), government of india, technical report, 2010. [7] e. schweig, j. gomberg, m. d. petersen, m. ellis, p. bodin, l. mayrose, et al., “the mw 7.7 bhuj earthquake: global lessons for earthquake hazard in intra-plate regions,” journal of the geological society of india, vol. 61, no. 3, pp. 277-282, 2003. [8] g. m. atkinson and d. m. boore, “modifications to existing ground-motion prediction equations in light of new data,” bulletin of the seismological society of america, vol. 101, no. 3, pp. 1121-1135, june 2011. [9] g. m. atkinson and d. m. boore, “earthquake ground-motion prediction equations for eastern north america,” bulletin of the seismological society of america, vol. 96, no. 6, pp. 2181-2205, december 2006. [10] g. r. toro, n. a. abrahamson, and j. f. schneider, “model of strong ground motions from earthquakes in central and eastern north america: best estimates and uncertainties,” seismological research letters, vol. 68, no. 1, pp. 41-57, january/february 1997. [11] s. pezeshk, a. zandieh, and b. tavakoli, “hybrid empirical ground-motion prediction equations for eastern north america using nga models and updated seismological parameters,” bulletin of the seismological society of america, vol. 101, no. 4, pp. 1859-1870, august 2011. [12] k. w. campbell, “prediction of strong ground motion using the hybrid empirical method and its use in the development of ground-motion (attenuation) relations in eastern north america,” bulletin of the seismological society of america, vol. 93, no. 3, pp. 1012-1033, june 2003. [13] w. silva, n. gregor, and r. darragh, “development of regional hard rock attenuation relations for central and eastern north america,” pacific engineering and analysis, https://www.nrc.gov/docs/ml0423/ml042310569.pdf, november 2002. [14] a. ravi kiran, m. h. prasad, m. k. agrawal, g. vinod, and g. r. reddy, “seismic hazard assessment and uniform hazard response spectrum (uhrs) generation for tarapur site,” ref. ahwr/dbgm/01900/r0/190513, technical report, 2019. [15] a. ravi kiran, s. bandopadhyay, m. k. agrawal, g. r. reddy, and r. k. singh, “generation of uniform hazard response spectrum for a peninsular indian site,” international journal of earth sciences and engineering, vol. 8, no. 2, pp. 332-335, april 2015. [16] d. motazedian and g. m. atkinson, “stochastic finite-fault modeling based on a dynamic corner frequency,” bulletin of the seismological society of america, vol. 95, no. 3, pp. 995-1010, june 2005. [17] j. g. anderson and s. e. hough, “a model for the shape of the fourier amplitude spectrum of acceleration at high frequencies,” bulletin of the seismological society of america, vol. 74, no. 5, pp. 1969-1993, october 1984. [18] d. m. boore, “short-period pand s-wave radiation from large earthquakes: implications for spectral scaling relations,” bulletin of the seismological society of america, vol. 76, no. 1, pp. 43-64, february 1986. [19] t. c. hanks, “fmax” bulletin of the seismological society of america, vol. 72, no. 6a, pp. 1867-1879, december 1982. [20] s. k. singh, m. ordaz, r. s. dattatrayam, and h. k. gupta, “a spectral analysis of the 21 may 1997, jabalpur, india, earthquake (mw = 5.8) and estimation of ground motion from future earthquakes in the indian shield region,” bulletin of the seismological society of america, vol. 89, no. 6, pp. 1620-1630, december 1999. proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 12 [21] s. k. singh, s. d. pimprikar, b. k. bansal, j. f. pacheco, r. s. dattatrayam, and g. suresh, “an analysis of the mw 4.7 jabalpur, india, earthquake of 16 october 2000: toward ground-motion estimation in the region from future events,” bulletin of the seismological society of america, vol. 97, no. 5, pp. 1475-1485, october 2007. [22] b. k. rastogi and a. g. chhatre, “seismotectonic investigations for siting of nuclear power plants, assessment of design basis ground motion and tsunami hazard,” journal of indian society of earthquake science, vol. 1, pp. 93-109, december 2014. [23] p. mandal and b. k. rastogi, “a frequency-dependent relation of coda qc for koyna-warna region, india,” pure and applied geophysics, vol. 153, no. 1 pp. 163-177, november 1998. [24] d. l. wells and k. j. coppersmith, “new empirical relationships among magnitude, rupture length, rupture width, rupture area, and surface displacement,” bulletin of the seismological society of america, vol. 84, no. 4, pp. 974-1002, august 1994. [25] d. m. boore and w. b. joyner, “site amplifications for generic rock sites,” bulletin of the seismological society of america, vol. 87, no. 2, pp. 327-341, april 1997. [26] a. n. tandon and h. m. chaudhury, “koyna earthquake of december 1967,” india meteorological department, scientific report, no. 59, 1968. 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 (http://creativecommons.org/licenses/by/4.0/).  proceedings of engineering and technology innovation , vol. 4, 2016, pp. 40 42 40 earth dam monitoring by using infrared thermography detection chien-yuan chen 1,* , su-chin chen 2 , kuan-hua chen 1 1 department of civil and water resources engineering, national chiayi university, chiayi, taiwan. 2 department of soil and water conservation, national chung -hsing university, taichung, taiwan. received 13 january 2016; received in revised form 10 february 2016; accept ed 03 march 2016 abstract infrared thermography is applied for artificial earth dam surface monitoring. using an infrared thermal imager is a nondestructive testing method for determining internal material changes by examining surface changes in radiation temperature. this study was conducted at the artificial earth dam experimental test site located in landao creek in huisun forest, nantou county, and central taiwan. infrared thermography analysis found different zones with larger radiation temperature changes. the seepage caused the earth dam soil to be wet, as can be reflected by thermography. the seepage failure zone was found to coincide with dramatic changes in radiation temperature recorded using thermography. this study found that dam surface areas with large radiation temperature changes could be failure areas, and that the potential earth dam failure mode could be identified . ke ywor ds : infra red thermography, earth da m, monitoring 1. introduction infrared thermography detectors are co mmonly used for military purposes. their applications also extend to industry, with the devices used for electric equipment maintenance and industrial inspections. using the detectors is a type of nondestructive testing method for monitoring internal material changes of a large area. landslide dams (e.g., earth dams) are attributed to topographic and geologic conditions. dams can cause water to pond upstream and inundate inhabited areas downstream if breached [1]. most natural dams are attributed to rainfall or earthquake-induced landslides, riverbank slope slips, or debris flows damming the river underneath. typhoon morakot made landfall in taiwan in 2009, bringing torrential rainfall to southern taiwan. the storm created 18 natural dams attributed to landslides and debris flows [2]. these natural dams are inaccessible, thus they are unable to be surveyed in the first stage for their safety evaluation. evaluating the stability and potential failure mode of these dams would be valuable for disaster prevention and mitigation; therefore, a long-distance nondestructive test for conducting such evaluation would be highly desirable. infrared thermography has been used as a nondestructive testing methodology for shotcrete slopes. the infrared temperature changes of shotcrete slope surfaces reflect the internal differences. the caves inside shotcrete slopes exhibit large differences in infrared temperature [3]. infrared thermography has been applied for mapping open fractures in deep-seated rockslides and unstable cliffs [4]. it has revealed the existence of open cracks, loosened areas, and nonkarst caves. optimal conditions for infrared thermography include an environment temperature lower than the average surficial rock temperature, a favorable field of vision, to avoid direct solarization, no heavy snow cover, and good for bold rock surface [4]. the failure mode of natural a dam depends on the characteristics of dam material, mechanics strength of the dam, soil permeability, and upstream discharge conditions. dam failure modes include overtopping erosion with peak flow, progressive erosion by seepage, and slope slide [5-7]. the most landslide dam failures (more than 50%) are attributed to overtopping, followed by piping and slope failure [6]. 2. study area and methodology the field experiment site was located in landao creek of huisun forest in nanotu county, central taiwan (fig. 1). the test site was a modi*corresponding aut hor. email: chienyuc@mail.ncyu.edu.t w proceedings of engineering and technology innovation , vol. 4, 2016, pp. 40 42 41 copyright © taeti fied experimental sediment discharge station used for observing the dam breach process [8]. the creek is one of the potential debris flows in taiwan. at the time of the study, the creek was 2,792 m long and had a slope between 44° and 55°. the creek was the site of a debris flow in 2001 during typhoon toraji and in 2004 during typhoon mindulle. fig. 1 site location of the field test at landao creek in middle taiwan an artific ial earth da m was constructed to model a landslide-induced earth dam (fig. 2). the fie ld test was located at the downstream of the creek. the artific ial earth dam was co nstructed to be 25 m long and 2 m high. the discharge was released upstream toward the earth dam. the da m fa ilure p rocess was mo nitored using the infrared thermography detector. fig. 2 size of the artificial earth dam 3. results and discussion preliminary infra red thermography detection conducted shortly after construction of the earth dam is shown in fig. 3. because of the topographic characteristics, the dam surface temperature appeared to be lowe r on the left side of the upstream and highly susceptible to ponding . the average radiation te mperature was 32.5 °c on the left side, lowe r than the 33.5 °c recorded on the right side; this difference is attributed to ponding-induced high water content. the infrared thermography revealed that the ponding area had a lower radiation te mperature than did the other areas. as the water level increased with the upstream discharge, the soil water content increased and the infrared temperature decreased (fig. 4). even after the water level increased to its highest point, the dam was not breached, but seepage occurred at the foot of the dam, as reflected in the infra red thermography (fig. 5). the fa ilure mode of progressive erosion by seepage was observed in the field test. fig. 3 upstream ponding of the earth dam fig. 4 rising of water level of the earth dam: (a) (b) upstream face and (c)(d) da m top view fig. 5 downstream seepage of the earth dam 4. conclusions the infrared thermography detector is capable of conducting the long-distance and large-area nondestructive motoring of objects. it is capable landao creek nankon river (a) (b) (c) (d) seepage in downstream proceedings of engineering and technology innovation, vol. 4, 2016, pp. 40 42 42 copyright © taeti of identifying potentially unstable areas by examining short-term temperature changes. dam soil with high water content exhibits dramatic radiation temperature changes and is prone to fail because of seepage. the seepage failure zones are indicated by dramatic changes in radiation temperature, as measured using infrared thermography. the potential failure mode of an earth dam can be predicted after the failure zone has been identified. the failure mode of progressive erosion by seepage was observed in the field test. references [1] o. korup, “ geo morphic ha zard assessment of landslide da ms in south westland, new zealand: fundamental proble ms and approaches ,” geo mo rphology, vol. 66, pp. 167-188, 2005. [2] forestry bureau, council of agriculture, executive yuan. ne ws release http://www.f orest.gov.tw/mp.asp?mp=204 (retrieve date 2012/11/21). (in chinese) [3] s. c. fang, “alishan highway slope failure characteristics and failure potential evalu ation,” national cheng kung un iv., dept. of civil engineering, doctoral thesis , 2009. (in chinese) [4] i. ba roň, d. beč kovský, and i. l. m íča, “application of infrared thermography for mapping open fractures in deep -seated rockslides and unstable cliffs,” landslides, vol. 36, pp. 265-275, 2012. [5] r. l. schuster, “landslide dams in the western united state,” proceedings of ivth international conference and fie ld workshop on landslides, tokyo, pp. 411-418, 1985. [6] r. l. schuster and j. e. costa, “a perspective on landslide da ms,” landslide da ms, processes, risk and mit igation, geotechnical spec ial publicat ion, no. 3, asce, pp. 1-2, 1986. [7] i. miyagi, t . m izuyama , and k. inoue, natural disasters, tokyo: ko kon, 2002. (in japanese) [8] s. c. chen, s. p. an, t . y. hsu, and c. wang, “high prec ision and adjusted discharge sediment, the e xperimental s tation in landao creek, huisun forest,” journal of chinese soil and water conservation, vol. 46, no. 1, pp. 1-6, 2015. (in chinese)  proceedings of engineering and technology innovation, vol. 13, 2019, pp. 01-09 weighted random forests for evaluating financial credit risk tzu-tsung wong * , shang-jung yeh institute of information management, national cheng kung university, tainan, taiwan received 13 may 2019; received in revised form 12 june 2019; accepted 06 july 2019 abstract credit evaluation of customers is a critical issue in financial organizations. classification algorithms have been proposed for credit evaluation in recent years, and the class distribution in the financial data for those studies are not skewed. however, only a small proportion of customers will be the cases for bad credit. financial records should be considered as an imbalanced data set for analyzing credit risk. ensemble algorithms that make predictions by group decisions generally have relatively high accuracy than the ones inducing only one model from data. this study introduces a mechanism based on the weighted random forest to improve the prediction accuracy on the records with bad credit. this mechanism is tested on two financial data sets to demonstrate that it can achieve relatively high performance in evaluating credit risk and that the number of decision trees in a forest is not helpful. critical attributes are also identified to provide practical meanings for credit risk analysis. keywords: credit risk analysis, decision tree, imbalanced data, random forest 1. introduction credit risk evaluation is getting important since the global financial crisis. the credit evaluation of customers is critical to financial organizations. it will be extremely helpful if the data of customers can be used to build models for supporting decisions. support vector machine and neural network are two popular classification algorithms for credit evaluation [1-11]. feature selection is a possible way to improve the performance of financial risk analysis [12-14]. ensemble algorithms induce several prediction models to make group decisions, and hence they generally have a better performance than the ones that learn only one model from data. for example, the random forest is an ensemble algorithm proposed by breiman [15], that grows multiple decision trees to take a majority vote for class prediction. studies have demonstrated that a random forest is a proper tool for credit risk analysis [16-20]. only a small proportion of the customers in a financial organization will have troubles in paying their loans. this implies that the data for credit risk analysis will be imbalanced, while none of the previous studies considered this to process data and to interpret the learning results. there are generally two class values in an imbalanced data set, and the one with fewer instances is called positive, and another is called negative. when over 90% of the instances in a data set are negative, the prediction accuracy of a classification model will be larger than 0.90 if it predicts a negative class value for all instances. this indicates that pursuing higher prediction accuracy is not appropriate for processing imbalanced data sets, while most classification algorithms are designed to have high accuracy. the learning procedure of traditional classification algorithms is generally designed to achieve a high probability of correct prediction. there are two popular ways in processing imbalanced data sets: change the class distribution in data sets, or modify the learning procedure of a classification algorithm [21]. resampling is a popular approach to adjust the class * corresponding author. e-mail address: tzutsung@mail.ncku.edu.tw tel.: +886-6-2757575; fax: +886-6-2362162 proceedings of engineering and technology innovation, vol. 13, 2019, pp. 01-09 2 distribution in a data set [22-23]. for example, smote is an over-sampling method for generating synthetic instances to balance positive and negative ones in a data set [24]. it is also possible to apply under-sampling methods for removing negative instances [25]. however, the model induced from the revised data set can be overfitting and difficult to interpret [26-27]. barandela et al. [28] consider weights to calculate distances for k-nearest neighbors in processing imbalanced data. an ensemble-based classifier that collects the predictions from multiple classification models to make group decisions is also an effective way for improving the performance on imbalanced data sets [29-30]. this study will employ the weighted random forest proposed by chen et al. [31] to analyze customer credit and propose a way to rank attribute for credit evaluation. the experimental results will demonstrate that the weighted random forest is more effective and suitable for discovering important attributes in credit risk analysis. the purpose of this paper is to consider the data for credit evaluation as an imbalanced set. ensemble method weighted random forest will be employed to analyze this kind of data sets not only for improving the prediction accuracy on positive instances but also for discovering critical attributes in evaluating the credit of customers. the data set after discretization will also be used to test ordinary random forest for demonstrating that the weighted random forest is a better tool for imbalanced data analysis. the implications of the critical attributes will be discussed. the whole research procedure is shown in fig. 1. fig. 1 the research procedure 2. weighted random forest a forest is composed of a set of decision trees, and every decision tree has two kinds of nodes: internal nodes and leaf nodes. every internal node is marked by an attribute to indicate the paths for branching, and there is a class value inside every leaf node for prediction. training data will be used to calculate impurity measure, such as information value and gini index, to determine the branching attributes in internal nodes. the random forest proposed by breiman [15] iteratively performs two steps to find a decision tree until the number of decision trees reaches a pre-specified threshold. let a data set have n instances composed of m attributes. the first step is to randomly select n instances with replacement from the data set. the instances that are not chosen for training are called out-of-bag data. they will be used to validate the decision tree built from the training data. when n is large, about 36.8% of the instances in the data set will play the role of validation. after the training set is determined, the second step is to randomly select m attributes for growing a decision tree; i.e. only the chosen attributes can be considered in growing a decision tree. since the instances and the attributes for growing a decision tree are both random, the decision trees built in this way will be diverse for making group decisions in classifying new instances. ordinary decision tree learning has a pruning step to avoid the occurrence of overfitting. group decisions are not sensitive to noise, and hence every decision tree in a forest need not be pruned. the gini index of a candidate attributes a is calculated as 2 1)( ii pagini  , where pi is the proportion of the instances with a class value i. the one with the largest gini index will be the branching attribute. this implies that the misclassification cost of all instances are the same regardless of their class values. a decision tree grown by this way is unlikely to have a high prediction accuracy on the positive instances in an imbalanced data set. this is the main reason why the random forest is not a proper tool for classifying imbalanced data sets. the weighted random forest proposed by chen et al. [31] considers the proceedings of engineering and technology innovation, vol. 13, 2019, pp. 01-09 3 misclassification costs of various class values to determine branching attributes. let c+ and cbe the misclassification cost of positive and negative instances, respectively. then the weight of positive class value is w = c+/(c++c-), and the weighted gini index of attribute a is calculated as 22 )1(1)(   pwwpagini . weight w is also considered in classifying new instances. let y+ and ybe the respective numbers of positive and negative training instances reaching the leaf node for classing a new instance. then the weighted votes for positive and negative class values from this decision tree are wy+ and (1-w)y-, respectively. these weighted votes are aggregated over decision trees to determine the predicted class value of the new instance. weighted random forest is, therefore, more suitable for processing imbalanced data sets. 3. performance evaluation true positives and false negatives represent the numbers of correct and wrong predictions on positive instances, respectively, and similarly for true negatives and false positives on negative instances. let tp, fp, tn, and fp be true positives, false positives, true negatives, and false negatives, respectively in a confusion matrix as shown in table 1. then accuracy is calculated as (tp+tn)/n, where n = tp+fp+tn+fp is the number of testing instances. since most of the instances are negative, fp+tn is far larger than tp+fn. accuracy will be large when a negative value is assigned to all instances. that is the reason why accuracy is an improper measure for performance evaluation when data sets are imbalanced. f-measure is a harmonic mean of recall = tp/(tp+fn) and precision = tp/(tp+fp), and it can be expressed as eq. (1): )/(2 precisionrecallprecisionrecallmeasuref  (1) recall shows the percentage of the positive instances that are correctly predicted, and precision indicates the accuracy of positive predictions. when the values of recall and precision are largely different, f-measure will be close to the smaller one. an algorithm that has a high value on only one of recall and precision cannot achieve a large f-measure. table 1 the confusion matrix for an imbalanced data set predicted class value yes no actual class value yes true positive (tp) false negative (fn) no false positive (fp) true negative (tn) auc represents the area under the roc curve determined by the value of recall and fp-rate = fp/(fp+tn). the roc curve represents the achievement of positive instances with respect to the number of misclassifications on negative instances. the maximal possible value of auc is one when an algorithm makes correct predictions on all positive and negative instances. both f-measure and auc can provide information about the prediction accuracy on positive instances, and hence they are popular measures in analyzing imbalanced data sets. knowing the importance of attributes in credit evaluation is extremely helpful for decision makers. it is easy to interpret the result implied by a single decision tree. however, every prediction of weighted random forest is made by the weighted majority vote of its induced decision trees. to rank attributes about their importance becomes relatively complex. this study proposes the following method composed of three steps to evaluate the critical level of each attribute. step 1. calculate the precision of each decision tree from its out-of-bag data, denoted as q1. step 2. consider attributing a as noise to recalculate the precision of each decision tree from its out-of-bag data, denoted as q2. the way of changing attribute a into a noisy one is to randomly permute the values of this attribute. step 3. calculate difference q1-q2 that represents the benefits of introducing attribute a for this decision tree, and average the differences of all decision trees to be the critical level of this attribute. proceedings of engineering and technology innovation, vol. 13, 2019, pp. 01-09 4 a larger critical level for an attribute means that this attribute is more useful in identifying positive instances. attributes are thus sorted in a descending order based on their critical levels to filter important ones for credit evaluation. 4. experimental study two data sets for credit risk evaluation are download from the internet to test whether weighted random forest can have a better performance on imbalanced data sets than random forest. the basic characteristics of the two data sets are given in table 2. note that the instances with missing values have been removed from the data sets. calculating the gini index is relatively straightforward for discrete attributes. the entropy-based discretization proposed by fayyad and irani [32] is employed to transform continuous attributes into discrete ones. this discretization method recursively bipartitions an interval into two subintervals based on information gain. the maximal description length principle is employed to derive a threshold for stopping the bipartition. the evaluation method is k-fold cross-validation with k = 5. this kind of cross-validation method first divides a data set into k folds. every fold will be in turn used to test the learning results induced from the other k-1 folds. the values of tp, fn, fp, and tn are aggregated over the testing instances for computing f-measure and auc in each iteration of k-fold cross-validation. the mean values of the metrics over the k folds will be compared to investigate the effectiveness of weighted random forest. the number of trees in both classification algorithms will be ranged from 100 to 500 with step size 100 to investigate the impact of forest size. table 2 basic characteristics of the two data sets for credit evaluation data set credit card scoring give me some credit source https://github.com/gastonstat/creditscoring https://www.kaggle.com/c/givemesomecredit number of positives 1,217 8,357 number of negative 3,160 111,912 class credit status serious past-due delinquency attributes job seniority revolving utilization of unsecured lines type of home ownership age time of requested loan number of times 30-59 days past due client’s age debt ratio marital status monthly income existence of records number of open credit lines and loans type of job number of times 90 days late amount of expenses number real estate loans or lines amount of income number of times 60-89 days past due amount of assets number of dependents amount of debt amount requested of loan price of good 4.1. data set “credit card scoring” this data set has 4,454 instances, and every instance has 13 attributes. the proportion of the instances with bad credit is less than 30%. the prediction accuracy is higher than 0.7 when all instances are classified as negatives. it is not a typical imbalanced data set, while can still be used to test the applicability of weighted random forest. the respective misclassification costs for positive and negative class values are 3160/1217 and 1 based on their proportions in the data set. the testing results of random forest and weighted random forest are summarized in table 3. note that all values in table 3 are the averages over five folds. consider the case for the random forest with 100 decision trees. the five pairs of recall and precision for folds 1 through 5 are (0.3770, 0.6835), (0.4052, 0.6225), (0.4783, 0.6395), (0.4570, 0.7222), and (0.4575, 0.6848). the f-measures calculated from the five pairs are 0.4859, 0.4909, 0.5473, 0.5598, and 0.5485, and their average is 0.5265 as shown in the second row of table 3(a). in this case, the mean recall and mean precision are 0.4350 and 0.6705, respectively, and the harmonic mean of these two values is 0.5277 that is not equal to 0.5265. proceedings of engineering and technology innovation, vol. 13, 2019, pp. 01-09 5 for the sake of clarity, the values of recall, precision, and f-measure for algorithms random forest (rf) and weighted random forest (wrf) are depicted in fig. 2. regardless of the number of trees, precision is larger than recall. this suggests that increasing the value of recall will have a larger chance to improve f-measure. the precisions for random forest and weighted random forest are close for any given forest size. since weighted random forest has a larger recall, it also has a larger f-measure. this data set is not highly skew, and hence the improvement on f-measure is not large. though every auc in table 3(b) is larger than its corresponding value in table 3(a), their difference is less than 0.01 in all cases. weighted random forest almost has no improvement on auc for this data set. table 3 and fig. 2 also show that increasing the number of trees is not an effective way to improve the performance of both classification algorithms. table 3 the experimental results of (a) random forest and (b) weighted random forest for data set credit card scoring (a) no. of trees recall precision f-measure auc 100 0.4350 0.6705 0.5265 0.8291 200 0.4412 0.6772 0.5335 0.8316 300 0.4415 0.6810 0.5344 0.8331 400 0.4443 0.6819 0.5370 0.8336 500 0.4514 0.6814 0.5416 0.8336 (b) no. of trees recall precision f-measure auc 100 0.4634 0.6594 0.5430 0.8322 200 0.4753 0.6677 0.5534 0.8338 300 0.4711 0.6642 0.5499 0.8337 400 0.4852 0.6749 0.5630 0.8339 500 0.4812 0.6661 0.5572 0.8343 fig. 2 the line charts of recall, precision, and f-measure for data set credit card scoring 4.2. data set “give me some credit” there are 120269 instances in this data set, and every instance has 8 attributes. the proportion of the instances with bad credit is approximately 7%. the prediction accuracy is higher than 0.9 when all instances are classified as negatives, and hence it is an imbalanced data set. the respective misclassification costs for positive and negative class values are 111912/8357 and 1 based on their proportions in the data set. the testing results of random forest and weighted random forest are summarized in table 4 and depicted in fig. 3. similar to the previous data set, every precision is larger than its corresponding recall. every recall is less than 0.2, which implies less than 20% of positive instances can be predicted correctly. this suggests that f-measure can be greatly improved by increasing recall. this is the reason why weighted random forest that has a slightly smaller precision can still achieve a higher f-measure than random forest. similar to the previous data set, weighted random forest achieves a slightly larger auc than a random forest, while their difference is even smaller. this suggests that auc may not be a proper measure for evaluating the performance of classification algorithms on financial credit data. again, the performance of both algorithms is not sensitive to the increasing forest size. since computational cost is higher for larger forest size, increasing the number of decision trees should not b e necessary. the experimental results of these two data sets suggest that increasing the number of decision trees is not beneficial for improving the performance of the two classification algorithms. 0.3 0.4 0.5 0.6 0.7 0.8 100 200 300 400 500 number of trees recall(rf) precision(rf) f-measure(rf) recall(wrf) proceedings of engineering and technology innovation, vol. 13, 2019, pp. 01-09 6 table 4 the experimental results of (a) random forest and (b) weighted random forest for data set “give me some credit” (a) no. of trees recall precision f-measure auc 100 0.1407 0.5604 0.2247 0.8250 200 0.1422 0.5649 0.2269 0.8307 300 0.1405 0.5616 0.2245 0.8324 400 0.1407 0.5605 0.2247 0.8335 500 0.1408 0.5670 0.2252 0.8327 (b) no. of trees recall precision f-measure auc 100 0.1796 0.5507 0.2705 0.8289 200 0.1769 0.5464 0.2670 0.8335 300 0.1828 0.5606 0.2753 0.8353 400 0.1819 0.5536 0.2736 0.8377 500 0.1800 0.5544 0.2715 0.8373 fig.3 the line charts of recall, precision, and f-measure for data set “give me some credit” 4.3. attribute ranks by using the critical level introduced in section 3 to rank attributes, the results are summarized in table 5. in data set “credit card scoring,” the top three important attributes filtered by weighted random forest are “amount of income,” “job seniority,” and “price of good.” all of these three attributes can imply whether customers can consistently pay their bills. marital status is the most useless attribute in this case. it is surprising that the rank of attribute “amount of debt” is 12. this implies that the amount of debts is not very helpful in evaluating the financial status of a customer. table 5 attribute ranks for both data sets rank credit card scoring give me some credit 1 amount of income revolving utilization of unsecured lines 2 job seniority debt ratio 3 price of good monthly income 4 amount requested of loan age 5 client’s age number of times 90 days late 6 amount of assets number of open credit lines and loans 7 amount of expenses number of times 30-59 days past due 8 existence of records number of time 60-89 days past due 9 type of job number of dependents 10 time of requested loan number real estate loans or lines 11 type of home ownership 12 amount of debt 13 marital status the top three important attributes filtered by weighted random forest for data set “given me some credit” is “revolving utilization of unsecured lines,” “debt ratio,” and “monthly income.” it is surprising that the three attributes (number of ti mes 30-59, 60-89, and 90 days past due) relevant to whether a customer has a late payment are not critical to the evaluation of bad credit. this is not reasonable and could be the major reason why most positive instances cannot be corrected classified, as indicated by the small recalls given in table 4. 0.1 0.2 0.3 0.4 0.5 0.6 100 200 300 400 500 number of trees recall(rf) precision(rf) f-measure(rf) recall(wrf) precision(wrf) f-measure(wrf) proceedings of engineering and technology innovation, vol. 13, 2019, pp. 01-09 7 attribute “amount of income” has a high rank in both data sets, and hence it is absolutely a necessary attribute in evaluating customer credit. the age of a customer should also be considered in approving a loan. the job of a customer is relatively important to the information about his/her family. the results of these two data sets are not consistent with the attributes relevant to debt. this can be an interesting question to explore further. since data sets for credit risk analysis may be collected from various resources, feature selection should be helpful for improving the performance of weighted random forest. 5. conclusions most data sets for credit evaluation are imbalanced, and hence random forest may not be a proper tool for this purpose. in this paper, the weighted random forest is introduced to analyze this kind of imbalanced data sets, and a way for filtering critical attributes from the learning results of weighted random forest is proposed. the testing results of two credit data sets show that weighted random forest outperforms random forest in terms of f-measure because the misclassification costs considered in the weighted random forest can increase the probability of correctly identifying positive instances. area under the roc curve may not be proper measure in determining superior classification algorithms for processing imbalanced data sets. more decision trees are generally not helpful for performance improvement, while more computational effort is needed. the critical attributes filtered from the two data sets also provide useful implications for understanding the credit of customers. those implications can provide practical suggestions in reviewing the applications of loans. the experimental results on two data sets for credit evaluation indicate that identifying positive instances is generally more difficult than having a high prediction precision on positive instances. this suggests that a classification algorithm will be a better tool for credit risk analysis if its learning procedure can be adjusted to improve the capability in identifying positive instances. this can shed some lights on designing the learning mechanism of classification algorithms for processing imbalanced data sets. conflicts of interest the authors declare no conflict of interest. acknowledgement this research was supported by the center for innovative fintech business models of national cheng kung university and the ministry of science and technology in taiwan under grant no. 107-2410-h-006-045-my3. references [1] b. baesens, t. van gestel, s. viaene, m. stepanova, j. suykens, and j. vanthienen, “benchmarking state-of-the-art classification algorithms for credit scoring,” journal of the operational research society, vol. 54, no. 6, pp. 627-635, june 2003. [2] c. l. huang, m. c. chen, and c. j. wang, “credit scoring with a data mining approach based on support vector machines,” expert systems with applications, vol. 33, no. 4, pp. 847-856, november 2007. [3] p. danenas and g. garsva, “selection of support vector machines based classifiers for credit risk domain,” expert systems with applications, vol. 42, no. 6, pp. 3194-3204, april 2015. [4] n. mohammadi and m. zangeneh, “customer credit risk assessment using artificial neural networks,” international journal of information technology and computer science, vol. 8, pp. 58-66, march 2016. [5] j. yao and c. lian, “a new ensemble model based support vector machine for credit assessing,” international journal of grid and distributed computing, vol. 9, no. 6, pp. 159-168, 2016. [6] a. aghaeirad, n. chen, and b. ribeiro, “improve credit scoring using transfer of learned knowledge from self-organizing map,” neural computing and applications, vol. 28, no. 6, pp. 1329-1342, june 2017. proceedings of engineering and technology innovation, vol. 13, 2019, pp. 01-09 8 [7] q. zhang, j. wang, a. lu, s. wang, and j. ma, “an improved smo algorithm for financial credit risk assessment evidence from china’s banking,” neurocomputing, vol. 272, pp. 314-325, january 2018. [8] d. k. gupta and s. goyal, “credit risk prediction using artificial neural network algorithm,” international journal of modern education and computer science, vol. 10, no. 5, pp. 9-16, may 2018. [9] s. ben jabeur, a. sadaaoui, a. sghaier, and r. aloui, “machine learning models and cost-sensitive decision trees for bond rating prediction,” journal of the operational research society, doi: 10.1080/01605682.2019.1581405, april 2019. [10] l. munkhdalai, t. munkhdalai, o. e. namsrai, j. y. lee, and k. h. ryu, “an empirical comparison of machine-learning methods on bank client credit assessments,” sustainability, vol. 11, pp. 1-23, january 2019. [11] m. z. abedin, c. guotai, fahmida-e-moula, a. s. m. s. azad, and m. s. u. khan, “topological applications of multilayer perceptrons and support vector machines in financial decision support systems,” international journal of finance and economics, vol. 24, pp.474-507, january 2019. [12] d. liang, c. f. tsai, and h. t. wu, “the effect of feature selection on financial distress prediction,” knowledge-based systems, vol. 73, pp. 289-297, january 2015. [13] s. dahiya, s. s. handa, and n. p. singh, “a rank aggregation algorithm for ensemble of multiple feature selection techniques in credit risk evaluation,” international journal of advanced research in artificial intelligence, vol. 5, pp. 1-8, october 2016. [14] f. n. koutanaei, h. sajedi, and m. khanbabaei, “a hybrid data mining model of feature selection algorithms and ensemble learning classifiers for credit scoring,” journal of retailing and consumer services, vol. 27, pp. 11-23, november 2015. [15] l. breiman, “random forests,” machine learning, vol. 45, no. 1, pp. 5-32, october 2001. [16] n. ghatasheh, “business analytics using random forest trees for credit risk prediction: a comparison study,” international journal of advanced science and technology, vol. 72, pp. 19-30, 2014. [17] m. malekipirbazari and v. aksakalli, “risk assessment in social lending via random forests,” expert systems with applications, vol. 42, no. 10, pp. 4621-4631, june 2015. [18] v. garcia, a. i. marques, and j. s. sanchez, “exploring the synergetic effects of sample types on the performance of ensembles for credit risk and corporate bankruptcy prediction,” information fusion, vol. 47, pp. 88-101, may 2019. [19] a. coser, m. m. maer-matei, and c. albu, “predictive models for loan default risk assessment,” economic computation and economic cybernetics studies and research, vol. 53, pp. 149-165, june 2019. [20] m. oskarsdottir, c. bravo, c. sarraute, j. vanthienen, and b. baesens, “the value of big data for credit scoring: enhancing financial inclusion using mobile phone data and social network analytics,” applied soft computing, vol. 74, pp. 26-39, january 2019. [21] v. lópez, a. fernández, s. garcía, v. palade, and f. herrera, “an insight into classification with imbalanced data: empirical results and current trends on using data intrinsic characteristics,” information sciences, vol. 250, pp. 113-141, november 2013. [22] g. e. batista, r. c. prati, and m. c. monard “a study of the behavior of several methods for balancing machine learning training data,” acm sigkdd explorations, vol. 6, pp. 20-29, 2004. [23] a. estabrooks, t. jo, and n. japkowicz, “a multiple resampling method for learning from imbalanced data sets,” computational intelligence, vol. 20, pp. 18-36, 2004. [24] n. v. chawla, k. w. bowyer, l. o. hall, and w. p. kegelmeyer, “smote: synthetic minority over-sampling technique,” journal of artificial intelligence research, vol. 16, pp. 321-357, 2002. [25] m. a. tahir, j. kittler, k. mikolajczyk, and f. yan, “a multiple expert approach to the class imbalance problem using inverse random under sampling,” proceedings of the eighth international workshop on multiple classifier systems, pp. 82-91, 2009. [26] y. li, h. guo, x. liu , y. li, and j. li, “adapted ensemble classification algorithm based on multiple classifier system and feature selection for classifying multi-class imbalanced data,” knowledge-based systems, vol. 94, pp. 88-104, february 2016. [27] z. sun, q. song, x. zhu, h. sun, b. xu, and y. zhou, “a novel ensemble method for classifying imbalanced data,” pattern recognition, vol. 48, no. 5, pp.1623-1637, may 2015. [28] r. barandela, j. s. sánchez, v. garcía, and e. rangel, “strategies for learning in class imbalance problems,” pattern recognition, vol. 36, pp. 849-851, 2003. [29] b. krawczyk, “learning from imbalanced data: open challenges and future directions,” progress in artificial intelligence, vol. 5, pp. 221-232, 2016. https://www.researchgate.net/journal/2165-4050_international_journal_of_advanced_research_in_artificial_intelligence proceedings of engineering and technology innovation, vol. 13, 2019, pp. 01-09 9 [30] l. i. kuncheva and j. j. rodríguez, “a weighted voting framework for classifiers ensembles,” knowledge and information systems, vol. 38, no. 2, pp. 259-275, february 2014. [31] c. chen, “using random forest to learn imbalanced data,” ph.d. thesis, department of statistics, university of california, berkeley, july 2004. [32] u. m. fayyad and k. b. irani, “multi-interval discretization of continuous valued attributes for classification learning,” proceedings of the thirteenth international joint conference on artificial intelligence, 1993, pp. 1022-1027. 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/). proceedings of engineering and technology innovation, vol. 10, 2018, pp. 36 41 a context retrieval method for context-awareness using ontologybased approach in internet of things environments yoosang park, yeonseung yoo, jonghyeok mun, jongsun choi, and jaeyoung choi * school of computer science and engineering, soongsil university, seoul 06978, korea received 12 march 2018; received in revised form 15 april 2018; accepted 24 may 2018 abstract a context-aware system is required for providing context-aware services to users in the internet of things (iot) environment. it consists of three primary tasks: gathering context data, abstracting the collected data, and providing services to users. in iot environments, context data are generated by a large number of sensors, and this context data are part of the context-awareness services provided to the users. when the context-aware system provides context-aware services with their service descriptions, it is necessary to process the data gathered by abstracting and contextualizing them. generally, the context-aware system encounters a problem in that the context representations between contextualized sensor data and their related service descriptions do not match well. we herein propose a context retrieval method that facilitates in obtaining context information for the context-aware system in iot environments. the context-aware system communicates with an ontology module that handles input data described in a set of universal resource identifiers, as a triplet. the ontology module resolves each representation request from the context-aware system. the proposed method provides an ontology-based mapping procedure for the context representation problem described above. keywords: context-aware, sensor data abstraction, internet of things (iot), ontology 1. introduction with the recent advances in the internet of things (iot) technology, investigations on context-aware techniques in the iot environments have been conducted [1-2]. context-aware techniques aim to understand information about the users, their locations, and provide services based on user needs through the context-aware system. we call this information around users as context or surroundings. context evaluates the state information that refers to the user’s environment. in iot environments, as devices have become key items that can be deployed anywhere, processing their sensor data is an important issue. we consider this set of sensor data as “massive data” because devices in the iot environments transmit data in real time; therefore, contextualizing them becomes an issue [3-4]. owing to the numerous sensor data generated, the contextaware system should process a large amount of data to provide context-aware services to the user. although the context aware system manages this large data, the services may not be provided to the users timely because of the large data volume. typically, the engine for processing context abstraction consists of three primary functionalities for contextualizing sensor data: collecting sensor data from iot devices, abstracting the collected data as part of the context, and evaluating and transmitting contexts to service engine for invoking certain services [5]. this context engine is also considered as part of the context-aware system, and focuses on handling context information. for context engine to understand this context * corresponding author. e-mail address: choi@ssu.ac.kr tel.: +82-2-820-0684; fax: +82-2-827-0684 proceedings of engineering and technology innovation, vol. 10, 2018, pp. 36 41 copyright © taeti 37 information, context representation is important because iot-based devices can transmit different types of sensed data that is provided to the user as a service. however, this context representation can be different in most scenarios. for example, if the users require a temperature control service, its service document may be described as follows: “when room a’s temperature is low, turn on the a/c service to attain the warm state.” to manage the information from the service document and the sensor data, the concept of ontology is used. details for the concept of ontology and methods to represent contexts are described in the preliminary, section 2. ontology primarily offers a functionality called semantic reasoning. however, owing to the large amount of data generated, semantic reasoning process becomes slow. we herein propose a method for retrieving context information for the service engine. the proposed method aims to link a statement of the contextualized sensor data and a condition from the service document. it contains two primary features. first, updating the interface for contextualized sensor data to the ontology for the context engine. next, requesting the interface to query conditions on the service document for the service engine. with this proposed method, the service engine can query descriptions on service document along with the generated context information in iot environments. 2. preliminary in general, ontology represents knowledge in the semantic web field that supports information or resource descriptions in specific service domains. this information or resource descriptions are regarded as an entity containing attributes or values within the graph data structure. when one entity has one attribute or a value with a relation, it is called a triplet. a triplet consists of three parts: subject, predicate, and object. although various types of triplets exist, we focus only on generalized triplet forms web ontology language (owl), and resource description framework (rdf) from the world wide web consortium (w3c) [6-7]. a subject represents the contents of a statement. a predicate is a relation that refers to the statement about its subject, and an object refers to its value, attribute, or statement. for example, when we define a type of sensor device as tempsensor, it would be <tempsensor, typesof, device>. several useful applications exist: apache jena, neo4j, graphdb, virtuoso, and marklogic to store the above-mentioned triplets [8]. to use these triplets as a parameter to the application, we use the simple protocol and rdf query language (sparql). sparql supports connections for a target ontology, an ontology name, namespaces, and resources for entity [9]. 3. related work context-aware systems such as service engines, or context engines are primarily related to ontology in iot environments owing to the issues for representing context information in each engine, and bridging context representations between service description and contextualized sensor data. featured service domains that collaborate on both context-aware system and ontology are as follows: ambient assisted living (aal), including healthcare, business processes, and smart homes [10-12]. context-aware systems collect the atomic information of the surroundings and interact with the users to provide application services. for such service domains, each semantic processing procedure between the collected data and service description varies. when the context-aware system is integrated with ontology, it may be difficult to explore the entire semantic reasoning from the sensor data level to the application service level. in ontology, surroundings are represented as triplets, and their semantic reasoning procedure generally searches all entities and handles complex triplets by consuming a long time. moreover, the primary problem that bridges the relevant contexts can be difficult to solve. first, in aal, research is conducted that aims to alert for situations where objects such as unauthorized people who try to enter the restricted area. they categorized the contextualizing procedure in three parts: context perception (acquisition), situation management, and semantic reasoning. the reasoning procedure occurs in every step of the contextualization [10]. next, another research is conducted that aims to build business processes and descriptions in iot environments with additional attributes before the reasoning procedure executes for the contextualization [11]. subsequently, in smart homes, a research proceedings of engineering and technology innovation, vol. 10, 2018, pp. 36 41 copyright © taeti 38 was conducted to represent the user’s activity pattern with wave signals. in this research, they defined rules for each part of the triplets called the semantic web rule language (swrl). the advantage of using the swrl is that additional rules can be processed during semantic reasoning. from the abovementioned investigations, semantic reasoning is typically used for linking different context representations [12]. our past study [13] aimed to build the metadata for each object and collecting sensor data through unified attributes, and transmitting context information as a keyword that is also considered as an entity. 4. proposed method in section 4, we outlined the context-aware system and its primary features, the proposed method, and its procedures. fig. 1 a layered architecture in sensor data abstraction in iot environments 4.1. context-aware system overview fig. 1 shows a layered architecture that consists of three primary layers: physical layer, context-aware layer, and application layer. each layer consists of its own task. the physical layer is a bottom layer where sensor and robotic devices are deployed to generate sensor data. sensor devices operate in iot environments whereas robotic devices represent states of an executing service, or a sensor list that is attached with the robot. the context-aware layer is a middle layer that collects sensor data and contextualizes them in the triplet form. an application layer is a top layer that provides context-aware services such as temperature control, turning lights on or off, activating robot service nodes, etc. the input data for contextaware system consists of two engines: service engine and context engine. the input data for the service engine is a set of service documents that contains service flow information, situation list, and conditions for context. the context engine provides the interface that supports sensor device connection management for data acquisition, collect their sensor data, and contextualize them into the context. as we described in section 1, semantic mismatches that are referred to the same context information exists. for example, given two sensors deployed at the same place, the temperature values of each sensor will be different. for this scenario, the conditions in the service document can be “if the temperature of rooma is low,” “subsequently, turn on the a/c service to attain a warm state.” for the context engine, sensor data representations are described as follows “tempsensor1 is a type of device,” “tempsensor1 has a value of 22 °c,” and “tempsensor1 is located in rooma.” table 1 lists the details for the set of triplet descriptions described above. in the context-aware layer, it is necessary to resolve the triplet <rooma, isnow, cold>. service engine confirms the triplet validation, and verifies whether the same triplet in the ontology is stored. to provide the service of verifying the triplet, <tempsensor1, isnow, cold> and <tempsensor2, isnow, cold> should result in <rooma, isnow, cold>. however, reducing the locations of “tempsensor1” and “tempsensor2” to “rooma” requires additional querying to compose certain triplets. in this context, additional information to combine these two triplets <tempsensor1, isnow, cold> and proceedings of engineering and technology innovation, vol. 10, 2018, pp. 36 41 copyright © taeti 39 <tempsensor2, isnow, cold> into <rooma, isnow, cold> is required. further, it is redundant that the context engine resolving these triplets during other sensor data is generated in real time. moreover, as the number of sensor devices is increased, the reducing procedure mentioned above will operate slowly, and thus would not be able to respond to the resulting condition of the service engine timely. therefore, the proposed method is necessary to map these triplets. table 1 context representation scenario: temperature control service based on temperature sensor data layer context representation in triplet form physical layer <tempsensor1, typesof, device> <tempsensor1, isvalueof, 22> <tempsensor1, islocatedin, rooma> <tempsensor2, typesof, device> <tempsensor2, isvalueof, 23> <tempsensor2, islocatedin, rooma> context-aware layer <tempsensor1, isnow, cold> <tempsensor2, isnow, cold> application layer if <rooma, isnow, cold> then, turn on <ac_service, warm> 4.2. modules for context retrieval method the proposed context retrieval method aims to solve the problem described in section 4.1. the context retrieval method consists of two primary interfaces. first, update the interface that the context engine calls and transmit the context information actively. next, request the interface that the service engine calls to verify whether the conditions are currently in the true states. the modified architecture with the proposed method as a context retrieval module is shown in fig. 2. fig. 2 modified layered architecture with proposed method. context retrieval and ontology are included fig. 3 conceptual view of sensor data abstraction the context retrieval module serves as a middle-man between the service engine and context engine in the contextaware layer. fig. 3 shows the conceptual design of the proposed method. the context retrieval module consists of three submodules: update manager, request manager, and sparql manager. each manager must input at least three parameters to process the whole triplet input. at first, we describe the ontology design and subsequently each manager, respectively. fig. 4 shows the data representation formed with the triplets for each engine and subontologies. the ontology consists of four subontologies: ontology index, history, stateinfo, and description. the ontology index indexes the subontologies. the history ontology stores the generated sensor data logged by the context engine. the stateinfo ontology stores the contextualized triplets that are described by concatenating the subject and object with an underscore from the service description. the stateinfo ontology only functions when the condition of service description is required. the description proceedings of engineering and technology innovation, vol. 10, 2018, pp. 36 41 copyright © taeti 40 ontology is for the context engine to determine the condition triplets for service description. four subontologies are established to avoid writing duplicate data or data loss. triplets are stored within asynchronous timing. furthermore, it can respond to requests from the service engine timely. fig. 4 details for context retrieving method in context-aware layer updating the manager pushes the triplets into the history and stateinfo subontologies. the context engine calls the update manager whenever the context information is ready to use. triplets, e.g., <rooma_cold, isvalueof, emptystate> is in the stateinfo subontology. fig. 4 is followed by the concatenation of subject and object with the underscore from the description ontology and initialized by “emptystate.” the context engine updates this “emptystate” to “true” when the request manager is to be verified. the request manager queries the stateinfo and description subontologies and is called by the service engine. when the service engine needs to resolve the triplet <rooma, isnow, cold>, it calls the request manager with <rooma_cold, isvalueof, true> in the stateinfo subontology. the ontology index contains the subontology and hints “stateinfo” with predicate querying <isnow, goesto, _?>. however, it is necessary to query the triplets into the target ontology, and format them into the proper target ontology, namespace, international resource identifiers (iri), and triple description. procedures for context retrieval module consist of three phases: initialize, contextualize, and context retrieval. phase 1. initialize in the initialize phase, the service engine processes the input data that is introduced as a service document and verifies the given triplets on the document. at this point, the service engine calls the request manager to query whether triplets exist in the description subontology and subsequently updates the initialized triplets in the form of <subject_object, isvalueof, emptystate> into the stateinfo subontology. phase 2. contextualize in the contextualize phase, the context engine collects the sensor data and profiles, and updates them into the history subontology by calling the update manager. when the service engine requests the subjected triplet to request a manager, it directly calls the update manager to obtain the current state. after the context engine verifies the current state, the context engine calls the update manager to update the triplets into the stateinfo subontology. therefore, to avoid the context engine from losing data owing to unexpected termination, the context engine always searches for the recent data in the history subontology. proceedings of engineering and technology innovation, vol. 10, 2018, pp. 36 41 copyright © taeti 41 phase 3. context retrieval in the context retrieval phase, the triplets in the stateinfo subontology are transmitted when the service engine requests for the context information in the stateinfo subontology. consequently, context mismatches between the service engine and context engine are resolved with the proposed method. the key idea is to represent the triplets with concatenation, and bridging different aspects of the context representation in the context-aware system. 5. conclusions we herein proposed a context retrieval method for the context-aware system. this method operated with the module that handled each context input data in the ontology and returned the contextualized result to the context-aware system. finally, the context-aware system could provide context-aware services to the users with the context representation that we described above. acknowledgements this work was supported by the industrial convergence core technology development program (no. 10048474) funded by the ministry of trade, industry & energy (motie), korea. references [1] b. morin, n. harrand, and f. fleurey, “model-based software engineering to tame the iot jungle,” ieee software, vol. 34, no. 1, pp. 30-36, january 2017. [2] f. zambonelli, “key abstractions for iot-oriented software engineering,” ieee software, vol. 34, no. 1, pp. 38-45, january 2017. [3] m. a. razzaque, m. milojevic-jevric, a. palade, and s. clarke, “middleware for internet of things: a survey,” ieee internet of things journal, vol. 3, no. 1, pp. 70-95, february 2016. [4] s. breiner, e. subrahmanian, and r. d. sriram, “modeling the internet of things: a foundational approach,” proc. of the seventh international workshop on the web of things, november 2016, pp. 38-41. [5] a. ikram, a. anjum, r. hill, n. antonopoulos, l. liu, and s. sotiriadis, “approaching the internet of things: a modelling, analysis and abstraction framework,” concurrency and computation: practice and experience, vol. 27, no. 8, pp. 1966-1984, october 2013. [6] x. h. wang, d. q. zhang, t. gu, and h. k. pung, “ontology based context modeling and reasoning using owl,” proc. of the second ieee annual conference on pervasive computing and communications workshops, pp. 18-22, 2004. [7] s. bechhofer, “owl: web ontology language,” encyclopedia of database systems, pp. 2008-2009, 2009. [8] m. junghanns, a. petermann, m. neumann, and e. rahm, management and analysis of big graph data: current systems and open challenges, handbook of big data technologies, pp. 457-505, 2017. [9] s. harris, a. seaborne, and e. prud'hommeaux, “sparql 1.1 query language,” w3c recommendation, vol. 21, no. 10, 2013. [10] d. calvaresi, d. cesarini, p. sernani, m. marinoni, a. f. dragoni, and a. sturm, “exploring the ambient assisted living domain: a systematic review,” journal of ambient intelligence and humanized computing, vol. 8, no. 2, pp. 239-257, april 2017. [11] k. suri, w. gaaloul, a. cuccuru, and s. gerard, “semantic framework for internet of things-aware business process development,” proc. ieee 26th international conf. on enabling technologies: infrastructure for collaborative enterprises(wetice), ieee press, pp. 214-219, 2017. [12] m. alirezaie, j. renoux, u. köckemann, a. kristoffersson, l. karlsson, e. blomqvist, n. tsiftes, t. voigt, and a. loutfi, “an ontology-based context-aware system for smart homes: e-care@home,” sensors, vol. 17, no. 7, pp.1586, 2017. [13] y. park, j. choi, j. choi, and y. cho, “conceptual metadata model for sensor data abstraction in iot environments,” international conf. on advanced control, february 2018.  proceedings of engineering and technology innovation, vol. 10, 2018, pp. 13 18 robust control of delayed fin stabilizer stochastic systems of a ship cheung-chieh ku * , chin-yuan hsu department of marine engineering, national taiwan ocean university, keelung, taiwan, roc received 02 march 2018; received in revised form 24 june 2018; accepted 29 july 2018 abstract in this paper, the robust control problem of delayed fin stabilizer stochastic system of a ship with uncertainty is discussed and investigated. to describe the system, linear parameter varying (lpv) modelling approach and multiplicative noise term are used to establish the corresponding polynomial model. for simulating the general operating environment, the delay effect is considered as time-varying case. moreover, the gain-scheduled control scheme is employed to discuss the delay-dependent stabilization problem and to design the corresponding controller. moreover, a novel lyapunov-kravoskii function is proposed by using parameter-dependent matrix and integral lyapunov function to reduce the conservatism of the derived stability conditions. in order to apply the convex optimization algorithm, the derived conditions are converted into linear matrix inequality (lmi) form. by solving the conditions, some feasible solutions can be obtained to establish the controller to guarantee robust stability of the delayed fin stabilizer stochastic system of a ship in the mean square. keywords: lpv system, stochastic behavior, lmi, fin stabilizer system of a ship 1. introduction generally, the comfort of passengers is deeply affected by sea waves according to violent rolling. therefore, an important issue of reducing the roll motion of the ship is usually discussed for building a liner or a passenger ship. besides, a large amplitude rolling motion may cause damage to the cargoes and vessels during the shipping process. for the above reasons, an effectiveness of fin stabilizer was investigated in [1-7] to reduce the rolling motions. generally, a pair of fins consists a ship fin stabilizer which locates approximately amidships on the bilge of the hull. those fins can be used to control changes of ship roll angle and its rate. through changing the angle [5-6], the hydrodynamic forces are induced on the fins to produce a moment that can reduce the wave induced roll motion. furthermore, the state feedback control method [3-4] is usually used to deal with the fin stabilizer system. however, it is difficult for reaching control target of the fin stabilizer system because the uncertainties appear in calculating the fin lift from fin angle. for the reason, lpv system is applied to describe the dynamics of the fin stabilizer system in this paper and to discuss the stabilization problem. lpv system [8-13] has been built by several linear systems and a specific weighting function to represent uncertain systems or nonlinear systems. according to the structure of lpv system, a general description for uncertain systems can be proposed to represent complex uncertainties. in [12-13], a gain-schedule (gs) control scheme has been applied to deal with the stabilization problem of lpv system. because the structure of gs controller is similar to lpv system, the robustness of the described system can be increased. it means that the gs scheme is vary suitable control scheme for systems described by numerous sub-systems. therefore, many robust stability criteria [10-13] have been proposed for lpv systems via applying the gs design scheme. furthermore, many practical robust control applications [9, 13] have been achieved via lpv system and * corresponding author. e-mail address: ccku@mail.ntou.edu.tw tel.: +886-2-24622192; fax: +886-2-24620724 proceedings of engineering and technology innovation, vol. 10, 2018, pp. 13 18 copyright © taeti 14 gs control scheme. in this paper, the lpv system is applied to represent the uncertain fin stabilizer system and the gs control scheme is used to discuss the stabilization problem of the system. in addition to uncertainty, effect of time delay always exists in dynamic process such as propagation/transportation of material, information or energy. the similar effect of time delay appears in the fin stabilizer system due to signal transportation. generally, delay-dependent criterion [11, 14-16] is discussed for general effect of time delay due to conservatism caused by constant delay. besides, some relaxed technologies [15-16] for delay-dependent criteria are developed to reduce conservatism of stability conditions. moreover, lyapunov-krasovskii function is often applied to derive sufficient conditions to analyze the stability of delayed system. thus, time delay effecting on state is also concerned with the considered control problem of the uncertain fin stabilizer system. moreover, stochastic behaviour [17-20] caused by high sea states often exists in operation during the shipping process. however, stochastic behaviour cannot be measured and predicted in the practical operation. referring to [17], stochastic differential equation provides a powerful tool to describe the stochastic behaviours. in the stochastic differential equation, stochastic behaviour is modelled as multiplicative noise term which is consisted by system states and white noise. hence, the stochastic difference equation has been applied in many control fields to extend design methods from deterministic systems to stochastic systems. based on the stochastic differential equation, many stability criteria have been developed to deal with stability issue of stochastic systems. from [21-22], one can find that the lpv system with multiplicative noise term can be successfully employed to represent uncertain stochastic systems. therefore, a stability criterion of delayed uncertain fin stabilizer stochastic system of a ship is discussed in this paper. to discuss the considered criterion, a delayed lpv stochastic system is built to describe practical fin stabilizer system of a ship with time delay, time-varying parameters and stochastic behaviors. based on the modelling approach and stochastic differential equation, the linear systems with multiplicative noise term in lpv stochastic system can be modelled by setting varying range and number of time-varying parameters. in order to analyze stability of the system, some relaxed sufficient conditions are derived via lyapunov-krasovskii function and jensen inequality [14]. in order to apply convex optimization algorithm, those stability conditions are converted into linear matrix inequality (lmi) form [23]. through solving the conditions, the feasible solutions can be obtained to build gs controller such that the considered fin stabilizer system of a ship is robust stable. based on the simulation result, the fin stabilizer system of a ship can be stabilized with the added delay effect and stochastic behavior. 2. system description and problem formulation referring to [4], the dynamics of fin stabilizer system of a ship can be represented as fig. 1. based on fig. 1, the dynamic equation of fin stabilizer system can be described as follows:   3 51 2 1 3 5x x d si i b b c c c m m              (1) where  denotes the ship’s rolling angle, x i and x i denote the mass inertia moment and affixing mass inertia moment relative to the vertical axes of ship, d m denotes the disturbance moment of sea waves, s m denotes the stable moment of lift feedback fins. besides, 1 c , 3 c , 5 c , 1 b , 2 b are constants and 1 c dh , where d denotes the ship’s tonnage and h denotes the height of steady center of ship’s rolling. in this paper, the parameters of a certain ship are assumed as 1457.26d t , 1.15h m , 6 3.4383 10 x x i i   , 6 3 2.097 10c   , 6 5 4.814 10c   , 6 1 0.636 10b   , 6 2 0.79 10b   .let t d s m m m  , the nonlinear model of this ship’s rolling can be described as follows: proceedings of engineering and technology innovation, vol. 10, 2018, pp. 13 18 copyright © taeti 15 3 5 7 0.185 0.23 0.4874 0.61 1.4 2.9084 10 t m               (2) to consider eq. (2) operated around 0 0 , the nonlinear fin stabilizer system can be rewritten as a linear system. moreover, the delay effect, uncertainties and stochastic behavior are added to describe the natural unstable source. furthermore, the following delayed lpv stochastic systems is provided to represent (2) with the considered effects.                  2 1 i i ti i i i dx t t x t x t d t u t dt x t d t        a a b e (3) where 𝐀𝟏 = [ 0 1 −0.458156 −0.185 ] , 𝐀𝟐 = [ 0 1 −0.516644 −0.185 ] , 𝐀𝒕𝟏 = [ 0 1 −0.0195 0 ] , 𝐀𝒕𝟐 = [ 0 0 −0.0195 0 ] , 𝐁𝟏 = [ 0 2.9084 × 10−7 ] , 𝐁𝟐 = [ 0 2.9084 × 10−7 ] , 𝐄𝟏 = [ 0 0 0 0.0698 ] , 𝐄𝟐 = [ 0 0 0 0.0698 ] , 𝑎1(𝑡) = |sin⁡(𝑡)| and 𝑎2(𝑡) = 1 − |sin(𝑡)| ∙ ⁡0 ≤ 𝑑(𝑡) ≤ 𝑑𝑀, and 𝑑(𝑡) ≤ 𝜀 < 1. 𝑑𝛽(𝑡) is scalar continuous type brownian motion satisfying the independent increment property [18]. fig. 1 dynamic of fin stabilizer system to deal with stabilization problem of eq. (3), the following gain-scheduled controller is designed via state-feedback control concept.        2 1 j j j u t t x t    f (4) substituting eq. (4) into eq. (3), the following closed-loop system can be inferred:                    2 2 1 1 i j i i j ti i i j dx t t t x t x t d t dt x t d t          a b f a e                 2 2 1 1 i j ij ti i i j t t x t x t d t dt x t d t         g a e (5) where ij i i j  g a b f . 2.1. lemma 1 for any constant matrix 0m , scalars 1r and 2r satisfying 2 1 r r , a vector function    : 0 xnx t ,  such that the integrals concerned are well defined, the following inequality holds. proceedings of engineering and technology innovation, vol. 10, 2018, pp. 13 18 copyright © taeti 16             2 2 2 1 1 1 t 2 1 r r r r r r x s ds x s ds r r x s x s ds   m m in the following section, a stability criterion is proposed to deal with the stability issue of the close-loop system (5). based on the proposed stability criterion, the robust stability of the closed-loop system (5) can be guaranteed. 3. delay-dependent stability criterion in this section, the stability criterion subject to time-varying delay performances for closed-loop system eq. (5) is developed. some sufficient conditions are derived via applying lyapunov-krasovskii function and itô’s formula in the following theorem. 3.1. theorem 1 for given scalars m d and  , the closed-loop system (5) is asymptotically stable if there exists t 0 i i  p p , 0q , and 0r such that the following conditions are satisfied.     t t 0 1 0 0 1 i ij ij i i i i m i ti m d * * * d                        p g g p e p e q r p a q r for 1 2i, j , (6) in order to limitation of pages, the proof of this theorem is omitted. although the condition in theorem 1 belongs to bilinear matrix inequality, the similar conversion and algorithm in [21] can be used to convert eq.(6) into lmi problem. in the following section, the numerical simulation is proposed. 4. simulation results to apply the proposed design method, 10 m d  and 0.7  are set. moreover, the following feasible solutions can be obtained. fig. 2 responses of 𝜒1(𝑡) fig. 3 responses of 𝜒2(𝑡) 𝐏𝟏 = [ 0.9271 0.8592 0.8592 6.1864 ] × 107, 𝐏𝟐 = [ 0.3741 0.8592 0.8592 6.6161 ] × 107, 𝐅1 = |−1.2315 3.1461| × 10 6, 𝐅2 = |−1.2315 3.1461| × 10 6, 𝐐 = [ 0.2910 0.1170 0.1170 7.5274 ] × 106 , 𝐑 = [ 0.0151 0.1212 0.1212 7.6452 ] × 106 , and 𝛼 =⁡−0.1287 with the above feasible solutions, the gs proceedings of engineering and technology innovation, vol. 10, 2018, pp. 13 18 copyright © taeti 17 controller (4) can be designed. based on the designed gs controller, the responses of (3) are stated in fig. 2-3 with initial condition 𝑥(0) = [ 𝜋 18 0] 𝑇 . from figs. (2)-(3), the state of eq. (3) are converged to zero. according to simulation result, the proposed design method can be used to stabilize the uncertain fin stabilizer stochastic system with time delay. 5. conclusions in this paper, a gs controller design method was proposed to deal with stabilization problem of uncertain fin stabilizer stochastic system with time delay. based on the modelling approaches, a lpv system with multiplicative noise was provided to describe the considered system. for the system, some conditions were derived via lyapunov-krasovskii function and jensen inequality. by solving the derived conditions, the feasible solutions can be obtained to build gs controller such that the closed-loop system is robust stability. finally, a simulation result has been provided to demonstrate the effectiveness and usefulness of the proposed design method. acknowledgement this work was supported by the ministry of science and technology, under contract most 106-2221-e-019-006. references [1] h. jin, y. gu, b. wang, and x. hou, “variable parameters pid controller design of fin stabilizer,” ship engineering, vol. 4, pp. 55-60, 1994. (in chinese) [2] h. jin, y. luo, and z. wang, “design of sliding mode variable structure controller for electric servo system of fin stabilizer,” proc. of the sixth world congress on intelligent control and automation, vol. 1, 2006, pp. 2086-2090. [3] y. h. liang, h. z. jin, and l. h. liang, “fuzzy-pid controlled lift feedback fin stabilizer,” journal of marine science and application, vol. 7, no. 2, pp. 127-134, june 2008. [4] t. perez and g. c. goodwin, “constrained predictive control of ship fin stabilizers to prevent dynamic stall,” control engineering practice, vol. 16, no. 4, pp. 482-494, april 2008. [5] c. y. tzeng and c. y. wu, “on the design and analysis of ship stabilizing fin controller,” journal of marine science and technology, vol. 8, no. 2, pp. 117-124, january 2000. [6] n. wang, x. meng, and q. xu, “fuzzy control system design and stability analysis for ship lift feedback fin stabilizer,” proc. of 7th world congress on intelligent control and automation, ieee press, august 2008, pp. 1223-1228. [7] z. h. xiu and g. ren, “fuzzy controller design and stability analysis for ship's lift-feedback-fin stabilizer,” proc. of the 2003 ieee international conf. intelligent transportation systems, ieee press, december 2003, pp. 1692-1697. [8] f. blanchini, d. casagrande, s. miani, and u. viaro, “stable lpv realization of parametric transfer functions and its application to gain-scheduling control design,” ieee transactions on automatic control, vol. 55, no. 10, pp. 2271-2281, october 2010. [9] l. liu, x. wei, and x. liu, “lpv control for the air path system of diesel engines,” ieee international conf. control and automation, november 2007, pp. 873-878. [10] j. daafouz, j. bernussou, and j. c. geromel, “on inexact lpv control design of continuous-time polytopic systems,” ieee transactions on automatic control, vol. 53, no. 7, pp. 1674-1678, august 2008. [11] f. wu and k. m. grigoriadis, “lpv systems with parameter-varying time delays: analysis and control,” automatica, vol. 37, no. 2, pp. 221-229, february 2001. [12] m. g. yoon, v. a. ugrinovskii, and m. pszczel, “gain-scheduling of minimax optimal state-feedback controllers for uncertain lpv systems,” ieee transactions on automatic control, vol. 52, no. 2, pp. 311-317, february 2007. [13] t. yue, l. wang, and j. ai, “gain self-scheduled 𝐻∞control for morphing aircraft in the wing transition process based on an lpv model,” chinese journal of aeronautics, vol. 26, no. 4, pp. 909-917, august 2013. [14] x. l. zhu and g. h. yang, “jensen integral inequality approach to stability analysis of continuous-time systems with time-varying delay,” iet control theory and applications, vol. 2, pp. 524-534, 2008. [15] b. chen, x. liu, and s. tong, “new delay-dependent stabilization conditions of t-s fuzzy systems with constant delay,” fuzzy sets and systems, vol. 158, no. 20, pp. 2209-2224, october 2007. proceedings of engineering and technology innovation, vol. 10, 2018, pp. 13 18 copyright © taeti 18 [16] w. j. chang, c. c. ku, and p. h. huang, “robust fuzzy control for uncertain stochastic time-delay takagi-sugeno fuzzy models for achieving passivity,” fuzzy sets and systems, vol. 161, no. 15, pp. 2012-2032, august 2010. [17] l. e. ghaoui, “state-feedback control of systems with multiplicative noise via linear matrix inequalities,” systems & control letters, vol. 24, no. 3, pp. 223-228, february 1995. [18] e. gershon, u. shaked, and i. yaesh, 𝐻∞ control and estimation of state-multiplicative linear systems, london: springer, 2005. [19] y. y. cao, j. lam, and l. hu, “delay-dependent stochastic stability and 𝐻∞ analysis for time-delay systems with markovian jumping parameters,” journal of the franklin institute, vol. 340, no. 6-7, pp. 423-434, september-november 2003. [20] c. c. ku, “robust mixed 𝐻2/𝑃𝑎𝑠𝑠𝑖𝑣𝑖𝑡𝑦 performance controller design for uncertain drum-boiler system,” journal of marine science and technology, vol. 23, no. 5, pp. 1003-1010, october 2016. [21] c. c. ku and g. w. chen, “gain-scheduled controller design for discrete-time linear parameter varying systems with multiplicative noises,” international journal of control, automation and systems, vol. 13, no. 6, pp. 1382-1390, december 2015. [22] c. c. ku and c. i wu, “gain-scheduled 𝐻∞ control for linear parameter varying stochastic systems,” asme, journal of dynamic systems, measurement and control, vol. 137, no. 11, pp. 111012-1-111012-12, august 2015. [23] s. boyd, l. e. ghaoui, e. feron, and v. balakrishnan, linear matrix inequalities in system and control theory, philadelphia: siam, pa, 1994. [24] w. j. chang, y. c. chang, and c. c. ku, “passive fuzzy control via fuzzy integral lyapunov function for nonlinear ship drum-boiler systems,” asme, journal of dynamic systems, measurement and control, vol. 137, no. 4, pp. 041008-1-041008-10, april 2015.  proceedings of engineering and technology innovation , vol. 4, 2016, pp. 34 36 34 a circuit device of detecting the bearing deflection chien-yu lu 1 , shuai-feng lin 2,* , alvin chang 3 1 department of industrial education and technology, national changua university of education, changhua, taiwan. 2, 3 department of industrial education and technology, national changua university of education, changhua, taiwan. received 17 march 2016; received in revised form 10 april 2016; accept ed 23 april 2016 abstract this paper investigates the operation of the motor caused by the bearing deflection, we will design detection circu it to detect bearing deflection, the circuit includes coil, rectifier circuit and voltage measuring device. in the interior of the motor, coil has been designed in gap between the stator silicon steel, and then connect a rectifie r c ircuit to an e xternal motor, the rectifie r circuit is connected to the voltage measurement terminal. when bearings close to stator silicon steel in the operation of the motor, the induced voltage of the coil will rise and transmit the voltage to rectifier c ircuit instantly. in this case, the voltage rectifier circuit will convert the dc voltage, the voltage delivered to the voltage measure ment termina l, at this time, users can know the motor bearings have deflected phenomenon based on the dc voltage , and can immed iately imp rove motor and maintenance operations. ke ywor ds : coil, bearing deflection, rectifie r circuit, voltage measurement 1. introduction motors have been widely used in conventional industries, manufacturing industries and technology industries. the inverter and a c servo drives are the fastest updating products, ma king it e xt re me ly competit ive in performance and cost. ac drive technology is also one of the core technologies . [1] this paper aims to e xp lore the device capable of detecting deflect ion of the motor bearing; the reason behind this is the deflection of the motor shaft when rotating at high speeds, [2] causing frict ion between the shaft and the motor that generates loud noises which lead to damag ing the motors ’ durability and the loss of precision of the shaft, [3] further effecting accuracy of the machine. the device to ensure orderly functions of the motor shaft after repair, and to decide if further steps are required are as shown in fig. 1. fig. 1 implementation of equipment fig. 2 non-deflection motor shaft principle : when the motor shaft is rotating at high speeds (no matter forward or reverse), the shaft will be met with ev ident deflection. as the shaft nears the stator silicon sheet, [4] there will be a significant rise of voltage fe lt by the inductor, after the voltage will be sent to the re c* corresponding author. email: powerm4vm0@gmail.com http://terms.naer.edu.tw/detail/112073/?index=2 http://terms.naer.edu.tw/detail/112073/?index=2 proceedings of engineering and technology innovation , vol. 4, 2016, pp. 34 36 35 copyright © taeti tifier c ircuit, there it converts the ac voltage is dc voltage; the voltage is then measured for a voltage value data. this voltage value [5] is used for the measurement of the distance between the shaft and the stator silicon sheet which is proof of deflection happening to the motor shaft, and conduct immediate motor ma intenance. deflec tion and non-deflection are as shown in fig. 2 and 3. fig. 3 deflection of motor shaft this paper investigates electronically lin king a rectifie r c ircuit to a monitoring device. an alarm is set to go off when the voltage value pre-set by the user in the rectifier c ircuit is reached, and to remind the user to repair the shaft deflection. installment of monitoring device is in fig. 4. fig. 4 installment of monitoring device 2. method this design is aimed towards a device that can detect and measure shaft deflection, prima rily consisting of the inductor unit, rectifie r circuit and voltage measurement. 2.1. design requirements (1) inductor unit: placed in the gap of the stator silicon sheet inside the motor, the inductance coil. [5] (2) rectifie r circuit: placed on the exterio r of the motor, and electronica lly linked to the inductor unit. its function is to convert av voltage fro m the inductor unit into dc voltage. the rect ifier circuit inc ludes electronically linked circuit co mponents and electronic co mponents. circuit co mponents are capacitance and electronic components are diodes. (3) vo ltage measurement: placed on the exterior of the motor and electronically lin ked to the rectifier c ircuit. its function is to convert the voltage into voltage value data. it is a three meter unit and is electronica lly linked to a monitoring device. (4) monitoring device: it is electronica lly linked to the metering unit. the ala rm to re mind the user is set to go off when the pre-set voltage value is reached, it serves as a maintenance reminder for the shaft. 3. results and discussion this design is based off existing motor technology, but designed for detecting shaft deflection, to create new techniques for lowering damages to the motor when running . 4. conclusions the motor is widely used in the manufacturing industry. any machine that produces force is run by a motor, making the debate for this device’s usage un-necessary. the advantage of applying this device is that it lowers damages upon the motor when running, predicting d eflection happenings on the motor for the user to conduct immediate ma intenance, and to ensure orderly functions . acknowledgement this is research is directed by dr. chien-yu lu, who lead this research team to continue improving and finishing this project. without dr. lu, this research cannot complete. in addition, i need to thank mr. alvin chang who supports searching information abroad. finally, all of the researchers and faculties in the department worth the honors going to be earned by this project. thank you all for the diligent efforts . proceedings of engineering and technology innovation , vol. 4, 2016, pp. 34 36 36 copyright © taeti references [1] i. y. shao, “ introduction to motor control power e lectronics laboratory,” national jiao tung univ., dept. of electronic engineering, technical report, hsinchu county, taiwan. [2] r. to mović , “ca lculat ion of the boundary values of rolling bearing deflection in rela tion to the number of active rolling ele ments,” mechanism and machine theory, vol. 47, pp. 74-88, 2012. [3] c. y. shih “ vehic le stability control for an electric vehic le with dual wheel motors,” unpublished. [4] j. s. jhang, y. h. tsai, and s. m. yang, “optimization and e xpe rimental verific ation of an a xial magnetic bearing,” 31 st conference of electrical engineering , de cember 2010. [5] r. abdollahi, “ha rmonic mit igation using 36-pu lse ac-dc converter for d irect torque controlled induction motor drives,” journal of applied research and technology, vol. 13, pp. 135-144, february 2015. microsoft word 3-peti#6580 21-31.docx proceedings of engineering and technology innovation, vol. 17, 2021, pp. 21-31 free vibration of thick fgm plates under tsdt and thermal environment chih-chiang hong * department of mechanical engineering, hsiuping university of science and technology, taichung, taiwan received 15 october 2020; received in revised form 30 november 2020; accepted 30 december 2020 doi: https://doi.org/10.46604/peti.2021.6580 abstract three parameters of thermal environment, varied calculated shear correction, and third-order shear deformation theory (tsdt) of displacement are important in the frequency study. these three effects have been studied on the non-dimensional and dimensional frequencies of thick fgm plates. an additional c1 displacement term in nonlinear coefficient of tsdt is used to present the frequency of vibration into the simply homogeneous equation of thick fgm plates. the determinant of the coefficient matrix containing the c1 displacement term in dynamic differential equilibrium equations can be derived into the five degree polynomial free vibration equation. the non-dimensional and dimensional of natural frequency can be obtained. the effects of plate thickness, temperature of environment and power law index of fgm on the non-dimensional and dimensional frequency of fgm plates are investigated. keywords: tsdt, fgm, frequency, nonlinear, homogeneous 1. introduction recently, many vibration papers devoted to the studies of mechanics of homogeneous, functionally graded material (fgm) porous beams/plates at micro, macro and nanoscales, diverse plates/beams theories, and linear/nonlinear thickness changes. in 2020, żur et al. [1] presented the free vibration by using the nonlocal sinusoidal higher-order shear deformation theory (hsdt) for the magneto-electro-elastic fgm nanoplates. in 2020, ouakad et al. [2] presented the nonlinear vibration by using the stress-driven nonlocal integral elasticity for the actuated hybrid nanotubes. in 2020, ghobadi et al. [3] presented the nonlinear free vibration by using the galerkin’s and perturbation methods for the fgm flexoelectric nano-plates. in 2020, ghobadi et al. [4] presented the nonlinear vibration by using the galerkin’s and multiple scale methods for the fgm thermo‐ electro‐elastic nanostructures. in 2021, guo et al. [5] presented the non-dimensional frequency by using the element-free imls-ritz method for the graphene nanoplatelets reinforced composite (gplrc) plates with matrix cracks. there are some non-dimensional and dimensional frequency of vibration presentations in the composite structure of plates. in 2020, babaei et al. [6] used the 3d elasticity theory of structures to investigate the non-dimensional natural frequency of saturated porous annular sector in thick fgm plates. some numerical solution of finite element method (fem) are obtained. in 2019, alaimo et al. [7] used the analytical procedure of navier to investigate the frequency of composite plates embedding viscoelastic layers. some numerical results of fem are obtained. in 2019, safaei et al. [8] used the mesh-free method and hsdt to investigate the frequency of fgm carbon nanotube (cnt) plate under thermo-mechanical loads. some numerical results of dynamic behaviors are obtained. in 2019, sun and wei [9] used the singular boundary method (sbm) to investigate the frequency of thin elastic plates. some numerical results of harmonic behaviors are obtained. in 2019, vinyas et al. [10] used the hsdt of displacements to investigate the frequency of three-phase smart magneto-electro-elastic (tps-mee) composite plates. some numerical results of fem are obtained. in 2017, rezaei et al. [11] investigated the non-dimensional frequency of * corresponding author. e-mail address: cchong@hust.edu.tw tel.: +886-919037599; fax: +886-4-24961187 proceedings of engineering and technology innovation, vol. 17, 2021, pp. 21-31 22 fgm plates with porosities by using linear model of displacements with analytical method. some numerical results of non-dimensional frequencies are obtained. in 2017, yao et al. [12] used von karman large deformation theory and hooke law to investigate the frequency responses of the bistable piezoelectric plate. some numerical results of frequencies are obtained. jha et al. investigated the numerical solutions of thick fgm with the higher order shear/shear-normal deformations theories (hosts/hosnts) in 2013, the numerical frequency solutions for free vibration of simply supported fgm plates are obtained [13]. in 2017, duc et al. [14] used the fsdt displacement method to present the analytical solutions for static response and free vibration of carbon nanotube reinforced composite (cntrc) fgm plates. several numerical investigations have been studied in the thick and thin fgms. hong [15] used the nonlinear model tsdt to obtain the vibration solutions for the thermal stress and deflection of the thick fgm plates in 2019. it would be interesting to study the non-dimensional and dimensional natural frequency containing the c1 displacement term in the tsdt displacement model and calculated values of shear correction for the thick fgm plates under free vibration. length to thickness ratio, temperature of environment and power law index of fgm three parametric effects on the non-dimensional and dimensional natural frequency of thick fgm plates are also studied. 2. procedure of formulations the typically two-material plates respectively in constituent material 1 and constituent material 2 of fgms are shown in fig. 1 with thickness h1 and h2 are considered for the derivation of formulation under the thermal influence of environment. usually, power-law type of materials function are selected to study the numerical solutions for the two-material. these standard forms of properties e.g. young’s modulus of fgm efgm can be denoted by efgm = (e2-e1)[(z+h * /2)/h * ] rn , in which rn is power index, e1 and e2 are young’s modulus of constituent material 1 and 2, respectively, for the other properties can be simplified into the mean expressions e.g. poisson’s ratios of fgm is νfgm = (ν1+ν2)/2, in which ν1 and ν2 are poisson’s ratios of constituent material 1 and 2, respectively [16]. also the constituent material properties pi are denoted and found in the variation of temperature t of environment i.e. pi = p0(p-1t -1 +1+p1t+p2t 2 +p3t 3 ), in which p0, p-1, p1, p2 and p3 are temperature coefficients. displacements in thick plates with variables u, v and w of fgm are in dependent of time. these displacements can be denoted and assumed in the additional nonlinear coefficient c1 term for c1 = 4/[3(h * ) 2 ] presented in tsdt mode [17], where h * is the summation of two-material thickness. fig. 1 two-material fgm plate under thermal environment the nonlinear reddy’s tsdt displacement field, strain-displacement relations and dynamic equilibrium differential equations of motion governing the fgm plates under consideration of formulations can be referred to the published paper in 2019 by hong [15]. the inter-laminar stresses within thick fgm plate under the influence of temperature difference △t for the k th constituent layer can be derived and given [18-19]. the tsdt term of displacement modes in thick fgm plate contained in the dynamic equilibrium equations can be derived and given [20] with k th constituent ply density ρ(k) integration parameters ii, ji and k2, in the subscript i = 0, 1, 2,…, 6. von karman type of strain-displacement relation equations can be derived and used with displacements u 0 and v 0 are in the axes direction x and y respectively, with transverse displacement w is in the axis direction z within the middle plate of thick fgm, also with the shear rotations ψx and ψy are in the x and y direction of axes respectively. by substituting stress and proceedings of engineering and technology innovation, vol. 17, 2021, pp. 21-31 23 strain-displacement equations into dynamic equations of motion, thus the dynamic differential equations of equilibrium contained tsdt terms of thick fgm plates can be obtained in matrix forms in terms of partial derivatives of displacements, shear rotations, mechanical loads (p1 and p2 are external loads in axes direction x and y respectively, q is pressure load), thermal loads and inertia terms. these calculated correction values of shear coefficient kαhare typically only in functions of parameters t, rn and h * [15]. stiffness parameters are integrated and given as follows. * * / 2 2 3 4 6 / 2 ( , , , , , ) (1, , , , , ) s s s s s s s s s s s s s s h i j i j i j i j i j i j h i j a b d e f h q z z z z z dz − = ∫ (1) * * * * * * * * * * * * * * * * / 2 2 3 4 / 2 5 ( , , , , , ) (1, , , , , ) h i j i j i j i j i j i j h i j a b d e f h k q z z z z z dz − = ∫ α (2) where ������ and ���∗�∗ are transformed reduced stiffness of fgm can be simplified in 2007 by shen [21], in which i s , j s = 1, 2, 6 and i * , j * = 4, 5. 3. numerical results and discussion the two-material fgm thick plates with same direction of constituents are selected to investigate the non-dimensional and dimensional frequency of numerical results under free vibration without thermal forces, distributed forces and pressure forces (△t = 0, p1 = p2 = 0, q = 0). the dimensional frequency ωmn (the subscripts m and n are mode shape numbers in the x and y direction of axes) can be simplified (i1 = i3 = j1 = 0, bij = eij = 0, d16 = d26 = a16 = a26 = 0 and a45 = d45 = f45 = 0) and derived for the following displacements and shear rotations in simply supported boundary equations, 0 ( / ) ( / ) ( ) mn mn u a cos m x a sin n y b sin t= π π ω (3) 0 ( / ) ( / ) ( ) mn mn b sin m x a cos n y b sin tv = π π ω (4) ( / ) ( / ) ( ) mn mn c sin m x a sin n y b sinw t= π π ω (5) ( / ) ( / ) ( ) x mn mn d s m x a sin n y b sio n tc=ψ π π ω (6) ( / ) ( / ) ( ) y mn mn e sin m x a cos n y b sin t=ψ π π ω (7) in which amn, bmn, cmn, dmn, and emn are amplitudes, t is the time, a is the length and b is the width of thick fgm plates. the major difference in the present free vibration work of simply homogeneous equation with respect to fully homogeneous equation is the calculated simplification in the matrix element terms. after eqs. (3)-(7) have been substituted into equations of equilibrium with no external loads (f1 = f2 =...= f5 = 0), and by conveniently assuming that matrix element terms fh13 = fh14 = fh15 = fh23 = fh24 = fh25 = 0 and i3 = j1 = i6 = j4 = 0 in the fully homogeneous equation, thus the simply homogeneous equation for the dimensional frequency ωmn contained the c1 terms of tsdt can be obtained as follows [15]. 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 − − − = − −                                                  λ λ λ λ λ (8a) proceedings of engineering and technology innovation, vol. 17, 2021, pp. 21-31 24 where 0 2 mn mn i=λ ω (8b) 2 2 11 11 66 ( / ) ( / )fh a m a a n b= +π π (8c) 1 612 62 )( ( / )( / )fh a a m a n b= + π π (8d) 22 66 2 2 2 2 ( / ) ( / )fh a m a a n b= +π π (8e) 2 2 2 4 2 2 2 2 33 55 44 1 11 1 1 66 2 4 2 2 22 1 55 1 55 1 44 1 44 ( ) ( ) ( / ) (2 4 )( / ) ( / ) ( / ) 3 (2 3 / / )( / ) 3 (2 3 )( / ) fh a m a a n b c h m a c c h m a n b c h n b c d c f m a c d c f n b = + + + + + − − − − π π π π π π π π (8f) in the derivation of mathematics, the algebraic determinant of eq. (8a) could be presented in the following equation in terms of λmn, thus the ωmn can be calculated. 5 4 3 2 (1) (2) (3) (4) (5) (6) 0 mn mn mn mn mn a a a a a a+ + + + + =λ λ λ λ λ (9a) where (1)a sd= − (9b) 11 22 (2) ( )a fh fh sd sc+= + (9c) 22 1211 1212 11 (3) [( ) ( ) ]a fh fh fh fh sd fh fh sc sb− −= ++ + (9d) 22 111 12 1 212 2 (4) ( ) ( )a fh fh fh fh sc fh fh sb sa−= + + + (9e) 22 1211 22 21 11 (5) [( ) ( ) ]a fh fh fh fh sb fh fh sa+− += − (9f) 22 11 12 21 (6) ( )a fh fh fh fh sa−= (9g) in which 2 2 0 ( / )sd k i= (9h) 33 44 2 0 /sdsc fh fh k i+= (9i) 33 55 44 55 33 44 35 35 34 34 452 0 45 ( /)sb fh fh fh fh fh fh fh fh fh fh k i fh fh= −+ + − − (9j) 33 44 55 44 34 35 35 34 45 35 35 44 34 34 55 45 45 33 sa fh fh fh fh fh fh fh fh fh fh fh fh fh fh fh fh fh fh + + − −= − (9k) a numerical method used to calculate and solve the polynomial eq. (9a) in terms of fifth-order of � , thus the natural frequency ωmn can be determined. a fortran program coded in the algorithm of newton's method by conte and de boor in 1980 [22] is used to find the solution of eq. (9a). the accuracy of tolerance is 1e-06 and used in the program to meet the condition of convergence then find the root of eq. (9a). the composited fgm two-material sus304/si3n4 is implemented for the computation results under temperature of environment t. the material 1 sus304 (stainless steel) is located at lower proceedings of engineering and technology innovation, vol. 17, 2021, pp. 21-31 25 position, the material 2 si3n4 (silicon nitride) is located at upper position of the fgm thickness, respectively. in the preliminary and fundamental calculations, it did not consider the c1 effect of tsdt onto the derivative of shear coefficient kα. for example a/b = 1, h1 = h2, h * = 0.12mm-12mm, the kα values under t = 300k, 600k and 1000k are shown in table 1(a), table 1(b) and table 1(c), respectively. when the values of kα are calculated under the condition of cases only for the corresponding t, rn and h * , then they are used in getting results for figs. 2-5. table 1(a) varied kα vs. rn and h * under t = 300k c1(1/mm 2 ) h * (mm) kα rn: 0.1 0.2 0.5 1 2 5 10 92.592598 0.12 0.098323 0.127625 0.266187 0.638005 1.007910 0.932667 0.843506 0.925925 1.2 0.079956 0.084882 0.102678 0.138574 0.217517 0.396486 0.492255 0.231481 2.4 0.075063 0.074789 0.074832 0.074112 0.064424 0.021065 0.001152 0.037037 6 0.069011 0.063127 0.048652 0.030937 0.010870 0.000220 0.000000 0.009259 12 0.064733 0.055449 0.034895 0.015697 0.002738 0.000006 0.000000 table 1(b) varied kα vs. rn and h * under t = 600k c1(1/mm 2 ) h * (mm) kα rn: 0.1 0.2 0.5 1 2 5 10 92.592598 0.12 0.085699 0.111597 0.236756 0.596251 0.995481 0.923808 0.834206 0.925925 1.2 0.069478 0.073722 0.089181 0.120708 0.191371 0.359433 0.456324 0.231481 2.4 0.065173 0.064854 0.064737 0.063996 0.055506 0.018026 0.000982 0.037037 6 0.059859 0.054641 0.041932 0.026560 0.009297 0.000188 0.000000 0.009259 12 0.056109 0.047937 0.030016 0.012979 0.002340 0.000005 0.000000 table 1(c) varied kα vs. rn and h * under t = 1000k c1(1/mm 2 ) h * (mm) kα rn: 0.1 0.2 0.5 1 2 5 10 92.592598 0.12 0.071316 0.091135 0.192482 0.562116 1.236624 1.094820 0.917769 0.925925 1.2 0.057519 0.059382 0.067907 0.088108 0.137812 0.270674 0.355043 0.231481 2.4 0.053882 0.052072 0.048781 0.045488 0.037321 0.011122 0.000565 0.037037 6 0.049404 0.043712 0.031294 0.018557 0.006113 0.000115 0.000000 0.009259 12 0.046254 0.038257 0.022289 0.009340 0.001533 0.000003 0.000000 the non-dimensional frequency parameters f * , ω*, ω in terms of fundamental dimensional frequency ω11 with mode shape numbers in the subscripts m = n = 1 are used for sus304/si3n4 thick plate under free vibration and given as follows, 2 * * 11 2 /f h e= ρω (10) * 1 2 2 1 / ) /( s s b i d= ωω π (11) * 2 1 111 2 1 / ) (1 /( )a h e−ω = ρ νω (12) * * / 2 1 / 2 h h s i dz − = ∫ ρ (13) * * / 2 2 1 / 2 h s h d z dzq − = ∫ (14) 2 1 1 1 / (1 )q e= −ν (15) in which ω11 is the fundamental dimensional frequency, ρ2 and ρ1 are the density of fgm material 2 and material 1, respectively. in the tables 2-4, the present results of non-dimensional f * , ω* and ω are compared with available published paper. in table 2, the present results of f * vs. h * are shown for a/h * =10, 300k, kα and c1 values, we found f * = 0.080832 at h * = 900mm, proceedings of engineering and technology innovation, vol. 17, 2021, pp. 21-31 26 rn = 0.5 is in close to f * = 0.0839 for al/zro2 by jha et al. in 2013 [13] with hosnt12 in which 12 degrees of freedom considered for the model. in table 3, the present results of ω* vs. h * are shown for a/h * =10, 300k, kα and c1 values, we found ω * = 4.094445 at h * = 600mm, rn = 2 is in close to ω * = 4.1165 with h * = 200mm forced vibration by kim in 2005 [23]. table 2 comparison of non-dimensional f * c1(1/mm 2 ) h * (mm) f * present results, a/ h * = 10, t = 300k, varied kα, for sus304/si3n4 jha et al. 2013, for al/zro2, rn = 0.5 rn = 0.5 rn = 1 rn = 2 0.925925 1.2 0.003713 0.002326 0.002412 0.333333 2 0.008001 0.004982 0.005163 0.000033 200 0.017724 0.017726 0.017708 0.000014 300 0.026743 0.026744 0.026727 0.000003 600 0.053789 0.053789 0.053773 0.000001 900 0.080832 0.080832 0.080816 0.0839 table 3 comparison of non-dimensional ω* c1(1/mm 2 ) h * (mm) ω* present results, a/ h * = 10, t = 300k, varied kα kim 2005, forced vibration, h * = 200mm, rn = 2 duc et al. 2017 cntrc, fsdt rn: 0.5 1 2 0.925925 1.2 0.282727 0.177156 0.183727 0.333333 2 0.609222 0.379389 0.393123 0.000033 200 1.349598 1.349716 1.348404 0.000014 300 2.036305 2.036380 2.035098 0.000003 600 4.095676 4.095697 4.094445 4.1165 3.99244 (ud) 0.000001 900 6.154796 6.154793 6.153549 also, compare the present results with the analytical fsdt results of uniform distribution (ud) in cntrc fgm plates by duc et al. in 2017 [14]. in table 4, the present results of ω vs. h * are shown for a/h * = 10, 700k, kα and c1 values, we found ω = 5.1314993 at h * = 250mm, rn = 1 is in close to ω = 5.359 with forced vibration under temperature rise (△t = 400k) by ungbhakorn & wattanasakulpong in 2013 [24]. table 4 non-dimensional ω comparison for sus304/si3n4 c1(1/mm 2 ) h * (mm) ω present results, a/h * = 10, t = 700k, varied kα ungbhakorn & wattanasakulpong 2013, rn = 1 rn: 0.5 1 2 0.925925 1.2 0.548323 0.568358 0.983723 0.333333 2 1.966345 1.217748 1.266043 0.000033 200 4.089171 4.089357 4.084728 0.000021 250 5.131368 5.131493 5.126937 5.359 0.000014 300 6.173302 6.173382 6.168875 the dimensional ωmn (unit 1/s) under no temperature difference (△t = 0) of two directional mode in the subscripts m, n are investigated. in fig. 2, the present results of ω1n vs. rn are shown for a/h * = 5 and 10, respectively for t = 300k, kα and c1 values. for the cases of rn = 0.5, 1 and 10, the present results of ω1n are oscillating and converging to nearly 0.005/s with n (in the subscript n = 1-9). the greatest value of ω11 = 0.054676/s is found, then decreasing to value ω19 = 0.005694/s for a/h * = 5, rn = 10. the greatest value of ω11 = 0.067990/s is found, then decreasing to value ω19 = 0.005715/s for a/h * = 10, rn = 10. when n = 9, the frequencies are almost meeting at a point, e.g. ω19 = 0.005694/s for a/h * = 5, rn = 0.5, 1 and 10, also ω19 = 0.005715/s for a/h * = 10, rn = 0.5 and 1, but not for rn = 10. the frequencies would be meeting closely at a point for all the value of rn when the values of n are greater more than 9. proceedings of engineering and technology innovation, vol. 17, 2021, pp. 21-31 27 (a) a/h * =5 (b) a/h * =10 fig. 2 ω1n (unit 1/s) vs. rn in fig. 3, the present results of ω1n vs. t are shown for a/h * = 5 and 10, respectively for rn = 0.5, kα and c1 values. generally the values of ω1n are oscillating and converging to nearly 0.005/s with n (in the subscript n = 1-9) for t = 300k, 600k and 1000k. the greatest value of ω12 = 0.059699/s is found, then decreasing to value ω19 = 0.001000/s for a/h * = 5, t = 600k. the greatest value of ω11 = 0.060500/s is found, then decreasing to value ω19 = 0.010078/s for a/h * = 10, t = 600k. when n = 9, the frequencies are almost meeting at a point, e.g. ω19 = 0.001000/s for a/h * = 5, t = 600k and 1000k, but not for t = 300k, also ω19 = 0.010078/s for a/h * = 10, t = 300k, 600k and 1000k. the frequencies would be meeting closely at a point for all the value of t when the values of n are greater more than 9. (a) a/h * = 5 (b) a/h * = 10 fig. 3 ω1n (unit 1/s) vs. t the influence of gradation of material properties e.g. rn values and thermal effect e.g. t values used on the vibration can be described in figs. 2-3 as follows. the present results of ω1n are oscillating and converging with the rn values. also the present results of ω1n are oscillating and converging with the t values. the trend of the curves will not change for the non-dimensional and dimensional frequencies. fig. 4 ω9n (unit 1/s) vs. n proceedings of engineering and technology innovation, vol. 17, 2021, pp. 21-31 28 in fig. 4, the present results of ω9n vs. n are shown for a/h * = 5 and 10, respectively for t = 300k, rn = 0.5, kα and c1 values. in general, the values ω9n are oscillating with n (in the subscript n = 1-9) for a/h * = 10. the values ω9n are linearly decreasing with n for a/h * = 5. the greatest value of ω94 = 0.011816/s is found, then decreasing to value ω97 = 0.001000/s for a/h * = 10. the greatest value of ω91 = 0.005274/s is found, then decreasing to value ω99 = 0.003356/s for a/h * = 5. for tsdt coefficient c1 = 0, the results can be added as shown in fig. 5 and recently can't be compared with pretty rare published data. in fig. 5(a), the present results of ω1n vs. rn are shown for a/h * = 5, t=300k, varied kα and c1 = 0. the values of ω1n are decreasing smoothly with n = 1-6, but have a jump at n = 7, then also decreasing smoothly with n = 8-9. the greatest value of ω1n is 0.019021/s for rn = 10 at n = 1, it is almost three times smaller than that 0.054676/s in the nonzero of c1 case. in fig. 5(b), the present results of ω1n vs. t are shown for a/h * = 5, rn = 0.5, varied kα and c1 = 0. the values of ω1n are decreasing smoothly with n = 1-6 for t = 300k and 600k, but have a jump at n = 7, then also decreasing smoothly with n = 8-9. the values of ω1n are decreasing smoothly with n = 1-5 for t = 1000k, but have a jump at n = 6, then also decreasing smoothly with n = 7-9. the greatest value of ω1n is 0.0188063/s for t = 1000k at n = 1, it is a little smaller than that 0.020803/s in the nonzero of c1 case. in fig. 5(c), the present results of ω9n vs. n are shown for a/h * = 5, t = 300k, rn = 0.5, varied kα and c1 = 0. the values of ω9n are decreasing linearly with n = 1-9. the greatest value of ω9n is 0.007045/s at n = 1, it is a little greater than that 0.005274/s in the nonzero of c1 case. the varied kα and c1 value of tsdt have great effect on the values of natural frequencies. (a) ω1n (unit 1/s) vs. rn for t = 300k (b) ω1n (unit 1/s) vs. t for rn = 0.5 (c) ω9n (unit 1/s) vs. n for t=300k, rn = 0.5 fig. 5 c1=0 for a/h * = 5 and varied kα similar way for keeping n = 1 and varying m from 1 to 9, the ωm1 can also be plotted and results can be included and discussed for varied kα and c1 values as shown in figs. 6-7. in fig. 6, the present results of ωm1 vs. rn are shown for a/h * = 5 and 10, respectively for t = 300k, kα and c1 values. the cases of rn = 0.5, 1 and 10 in the present results of ωm1 are small oscillating and converging to nearly 0.005/s with m (in the subscript m = 1-9). the greatest value of ω11 = 0.054676/s is found, then decreasing to value ω91 = 0.005969/s for a/h * = 5, rn = 10. the greatest value of ω11 = 0.067990/s is found, then decreasing to value ω91 = 0.004257/s for a/h * = 10, rn = 10. when m = 9, the frequencies are almost meeting at a point, e.g. ω91 = 0.005969/s proceedings of engineering and technology innovation, vol. 17, 2021, pp. 21-31 29 for a/h * = 5, rn = 0.5, 1 and 10, also ω91 = 0.006950/s for a/h * = 10, rn = 0.5 and 1, but not for rn = 10. the frequencies would be meeting closely at a point for all the value of rn when the values of m are greater more than 9. in fig. 7, the present results of ωm1 vs. t are shown for a/h * = 5 and 10, respectively for rn = 0.5, kα and c1 values. the cases of t = 300k, 600k and 1000k in the present results of ωm1 are small oscillating and converging to nearly 0.005/s with m (in the subscript m = 1-9). the greatest value of ω11 = 0.031320/s is found, then decreasing to value ω91 = 0.005513/s for a/h * = 5, t = 600k. the greatest value of ω11 = 0.098299/s is found, then decreasing to value ω91 = 0.007040/s for a/h * = 10, t = 1000k. when m = 9, the frequencies are almost meeting at a point, e.g. ω91 = 0.005513/s for a/h * = 5, t = 300k and 600k, but not for t = 1000k, also ω91 = 0.007040/s for a/h * = 10, t = 300k, 600k and 1000k. the frequencies would be meeting closely at a point for all the value of t when the values of m are greater more than 9. (a) a/h * = 5 (b) a/h * = 10 fig. 6 ωm1 (unit 1/s) vs. rn (a) a/h * = 5 (b) a/h * = 10 fig. 7 ωm1 (unit 1/s) vs. t the results of non-dimensional and dimensional natural frequency are obtained in free vibration of two-material thick fgm plates with power law index by considering the tsdt model, varied shear coefficient and temperature of environment. some of the present results of non-dimensional f * , ω* and ω are compared with available published paper and found in close data. it is fundamental to study the effects of varied shear coefficient values and additional c1 displacement term in nonlinear coefficient of tsdt on the frequency calculations under free vibration. the varied kα and c1 value of tsdt have great effect on the values of natural frequencies. for the cases of rn = 0.5, 1 and 10, the present results ω1n are oscillating and converging to nearly 0.005/s with n (in the subscript n = 1-9). for the cases of t = 300k, 600k and 1000k, the present results ω1n are oscillating and converging to nearly 0.005/s with n (in the subscript n = 1-9). similar way for keeping n = 1 and varying m from 1 to 9, the cases of rn = 0.5, 1 and 10 in the results of ωm1 are small oscillating and converging to nearly 0.005/s with m (in the subscript m = 1-9), the cases of t = 300k, 600k and 1000k in the results of ωm1 are small oscillating and converging to nearly 0.005/s with m (in the subscript m = 1-9). proceedings of engineering and technology innovation, vol. 17, 2021, pp. 21-31 30 4. conclusions the natural frequency of free vibration in two-material thick fgm plates are great effected by power law index, the tsdt model, varied shear coefficient and temperature of environment. it is fundamental and new to study the additional effect of nonlinear displacement tsdt c1 term on the frequency for two-material thick fgm plates. the present natural frequency results are oscillating and converging with respect to mode shapes by simultaneously including x and y directions. these natural frequency data of free vibration can provide a base reference and used further in the thermal vibration study in the future. nomenclature h1 thickness of constituent material 1 h2 thickness of constituent material 2 h * summation of two-material thickness a length of thick fgm plates b width of thick fgm plates x x direction of cartesian axes y y direction of cartesian axes z z direction of cartesian axes t temperature of environment rn power index in power-law type of fgm materials function efgm young’s modulus of fgm e1 young’s modulus of constituent material 1 e2 young’s modulus of constituent material 2 νfgm poisson’s ratios of fgm ν1 poisson’s ratios of constituent material 1 ν2 poisson’s ratios of constituent material 2 ρ1 density of constituent material 1 ρ2 density of constituent material 2 pi constituent material properties p0, p-1, p1, p2 and p3 temperature coefficients of constituent materials u, v and w displacements in x, y and z directions of cartesian axes u 0 and v 0 displacements in the axis directions x and y within the middle plate of thick fgm c1 nonlinear reddy’s tsdt displacement field c1 = 4/[3(h * ) 2 ] △t temperature difference k th k th constituent layer ρ(k) kth constituent ply density ii, ji and k2 density ρ (k) integration parameters, in the subscript i = 0, 1, 2,…, 6 ψx and ψy shear rotations in the x and y direction of axes p1 and p2 external loads in axes direction x and y q pressure load kα shear coefficient ������ and ���∗�∗ transformed reduced stiffness of fgm with i s , j s = 1, 2, 6 and i * , j * = 4, 5 ωmn dimensional frequency with the subscripts m and n are mode shape numbers in the x and y direction of axes ω11 fundamental dimensional frequency with m = n = 1 f * , ω*, ω non-dimensional frequency parameters t time f1, f2,..., f5 external loads λmn frequencies parameter in λmn = i0ωmn 2 conflicts of interest the authors declare no conflict of interest. proceedings of engineering and technology innovation, vol. 17, 2021, pp. 21-31 31 references [1] k. k. żur, m. arefi, j. kim, and j. n. reddy, “free vibration and buckling analyses of magneto-electro-elastic fgm nanoplates based on nonlocal modified higher-order sinusoidal shear deformation theory,” composites part b, vol. 182, 107601, february 2020. [2] h. m. ouakad, a. valipour, k. k. żur, h. m. sedighi, and j. n. reddy, “on the nonlinear vibration and static deflection problems of actuated hybrid nanotubes based on the stress-driven nonlocal integral elasticity,” mechanics of materials, vol. 148, 103532, september 2020. [3] a. ghobadi, h. golestanian, y. t. beni, and k. k. żur, “on the size-dependent nonlinear thermo-electro-mechanical free vibration analysis of functionally graded flexoelectric nano-plate,” communications in nonlinear science and numerical simulation, in press. [4] a. ghobadi, y. t. beni, and k. k. żur, “porosity distribution effect on stress, electric field and nonlinear vibration of functionally graded nanostructures with direct and inverse flexoelectric phenomenon,” composite structures, in press. [5] h. guo, t. yang, k. k. żur, and j. n. reddy, “on the flutter of matrix cracked laminated composite plates reinforced with graphene nanoplatelets,” thin-walled structures, vol. 158, 107161, january 2021. [6] m. babaei, m. h. hajmohammad, and k. asemi, “natural frequency and dynamic analyses of functionally graded saturated porous annular sector plate and cylindrical panel based on 3d elasticity,” aerospace science and technology, vol. 96, 105524, january 2020. [7] a. alaimo, c. orlando, and s. valvano, “analytical frequency response solution for composite plates embedding viscoelastic layers,” aerospace science and technology, vol. 92, pp. 429-445, september 2019. [8] b. safaei, r. moradi-dastjerdi, z. qin, and f. chu, “frequency-dependent forced vibration analysis of nanocomposite sandwich plate under thermo-mechanical loads,” composites part b, vol. 161, pp. 44-54, march 2019. [9] l. sun and x. wei, “a frequency domain formulation of the singular boundary method for dynamic analysis of thin elastic plate,” engineering analysis with boundary elements, vol. 98, pp. 77-87, january 2019. [10] m. vinyas, k. k. sunny, d. harursampath, t. nguyen-thoi, and m. a. r. loja, “influence of interphase on the multi-physics coupled frequency of three-phase smart magneto-electro-elastic composite plates,” composite structures, vol. 226, 111254, october 2019. [11] a. s. rezaei, a. r. saidi, m. abrishamdari, and m. h. pour mohammadi, “natural frequencies of functionally graded plates with porosities via a simple four variable plate theory: an analytical approach,” thin-walled structures, vol. 120, pp. 366-377, november 2017. [12] m. yao, w. hu, and w. zhang, “nonlinear frequency responses of the bistable piezoelectric plate,” procedia iutam, vol. 22, pp. 208-215, 2017. [13] d. k. jha, t. kant, and r. k. singh, “free vibration response of functionally graded thick plates with shear and normal deformations effects,” composite structures, vol. 96, pp. 799-823, february 2013. [14] n. d. duc, j. lee, t. nguyen-thoi, and p. t. thang, “static response and free vibration of functionally graded carbon nanotube-reinforced composite rectangular plates resting on winkler–pasternak elastic foundations,” aerospace science and technology, vol. 68, pp. 391-402, september 2017. [15] c. c. hong, “gdq computation for thermal vibration of thick fgm plates by using fully homogeneous equation and tsdt,” thin-walled structures, vol. 135, pp. 78-88, february 2019. [16] 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. [17] 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, no. 5-6, pp. 463-483, march 2004. [18] 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, no. 7, pp. 971-985, september 2005. [19] j. m. whitney, “structural analysis of laminated anisotropic plates,” u.s.a: crc press, 1987. [20] j. n. reddy, “energy principles and variational methods in applied mechanics,” new york: wiley, 2017. [21] h. s. shen, “nonlinear thermal bending response of fgm plates due to heat condition,” composites part b: engineering, vol. 38, pp. 201-215, 2007. [22] s. d. conte and c. de boor, “elementary numerical analysis, an algorithmic approach,” 3rd ed. u.s.a.: mcgraw-hill book company, 1980. [23] y. w. kim, “temperature dependent vibration analysis of functionally graded rectangular plates,” journal of sound and vibration, vol. 284, no. 3-5, pp. 531-549, june 2005. [24] v. ungbhakorn and n. wattanasakulpong, “thermo-elastic vibration analysis of third-order shear deformable functionally graded plates with distributed patch mass under thermal environment,” applied acoustics, vol. 74, no. 9, pp. 1045-1059, september 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 (cc by-nc) license (https://creativecommons.org/licenses/by-nc/4.0/).  proceedings of engineering and technology innovation, vol. 4, 2016, pp. 28 30 28 advertisement-clic k prediction based on mobile big data from hyxen adlocus yu-xiang fei 1,* , ji-ying chen 1 , shih-hau fang 1 , yu tsao 2 , jen-wei huang 3 , bo-wei liang 4 1 department of electrical engineering and innovation center for big data and digital convergence , yuan ze university, taoyuan, taiwan. 2 research center for information technology innovation, academia sinica, taipei, taiwan. 3 department of electrical engineering, national cheng kung university, tainan, taiwan. 4 hyxen technology limitedliability company, new taipei city, taiwan. received 30 january 2016; received in revised form 25 february 2016; accept ed 09 march 2016 abstract the popularity of internet has made adve rtisement ma rketing gone virtualized and location-based mobile advertising successful in recent years. adlocus, an app developed by hyxen technology, is one good exa mple to achieve this. this advertising software can tailor to the campaign needs and target users within a dia meter o f 1 km. however, the question is that is it possible to predict whether the user is willing to clic k on the advertisement. this paper adopts many ways to analy ze how these rela tions influence in different kinds of mobile advertisement. a co mprehensive performance comparison of different models is provided, and the analysis of different factors is also discussed, including c lic k t ime, advertise ment category, language, and mobile phone manufacturers . keywor ds : advertisement-click prediction, mobile devices , audience targeting. 1. introduction the popular network has recently increased. when watching advertisements on television, we fee l a waste of time, then turning to other programs or doing other things so that adve rtisements are limited on television. however, in order to increasingly ma ke an impression on users for advertisements, hyxen technology focused on location-based services to solve the problem. this company offered a service called adlocus which uses positioning technology to provide mobile advertisement. the comme rcia l value of advertise ments on the web depends on whether users click on the advertisements. many issues, such as users’ intent analysis and advertisement selection may affect the c lic k probability of advertise ments [1]. hence, users who installed adlocus will be provided a large nu mber of advertisements of in-app. meanwhile, advertisements of push which will sent notifications. a mong these a dvertisements, such as sponsored search advertising, contextual advertising, d isplay advertising, and real-t ime b idding auctions, have all relied heavily on the ability of learned models to predict ad clic k-through rates (ctr) accurately, quickly, and reliably [2]. the paper use collected adlocus data to analyze user’s behavior and predict how they hit ads. the data, which has been collected through hyxen for the entire april-2015, containing over one hundred thousand samples per day. there are 21 data attributes, including advertisement category, type of connection, type of device, playing time , advertisers, carrier, user equipment, software version, language, city, and the like. we select seven out of these features to predict the click behaviors [3]. the machine learning a lgorith ms used in this research includ e support vector machine, decision tree , and rusboost. this research can help advertising companies better perform audience targeting, connecting the advertisement with the right audience at the right time and the right place, and even stimulate interactions and have influence on them. * corresponding aut hor. email: s1044638@mail.yzu.edu.tw proceedings of engineering and technology innovation , vol. 4, 2016, pp. 28 30 29 copyright © taeti 2. results and analysis 2.1. database there are two main different advertising patterns on adlocus: 1. in -app (when open the app, banner style ads to appear on mobile phone) 2. push (send ads to your phone's notice board). this e xpe riment using the push ads, collected 30 days a month of data. info rmation ite ms closer to six million, wh ich contains a lot of text and character portions. the clicked rate is less than 5%. 2.2. data analysis to predict the click-through rate and what factors most re levant, so analyzed 21 data attributes. the study picks out seven features of data attributes having great influence: adve rtisement category, type of device, playing time , os system of phone, phone used language, ca rrie r, and city. this paper used the word embedding approach to transform the features to the numerical va lues. fig. 1 shows the hourly ct r during 30 days. ct r is the whole day statistics per second. this figure shows that ct r is higher at 3:00 a.m. to 4:00 a .m., and the ma ximu m ct r is 4.73 percent at 3:27. ct r is lo wer at 5:00 a.m. to 5:30 a.m. the total average is 2.4 percent. fig. 1 cumulative ctr per second fig. 2 shows ctr according to different categories of ads. this figure shows that game/app shows the higher ct r, while trave l shows the lowest ctr. fig. 2 ctr of advertisement category fig. 3 shows ctr according to different mobile phone language, where zh -tw represents traditional chinese and zh -cn represents simp lified ch inese. this figure shows that the highest ct r is unknown. this is because many phones are unknown language labels. the lowest ct r is simplified chinese. fig. 3 ctr of mobile phone user’s language fig. 4 shows ctr according to different mobile phone manufacturers. this figure shows that xiaomi shows the higher ctr, while apple shows the lowest ctr. fig. 4 ctr of mobile phone manufacturers 2.3. prediction this e xperiment adopted 4 types machine learning, includ ing support vector machine (svm ), k-nearest neighbor (knn), decision tree (dt ), and rusboost. to predict the click based on training data (april-15 wednesday) and testing data (april-1 wednesday). table 1 shows the results . table 1 prediction accuracy algorithm accuracy (%) dt(complex) 94.9 dt(medium) 95.0 knn(weight) 95.6 knn(fine) 94.2 svm 96.4 rusboosted 73.5 the table shows svm p rovides a higher accuracy rate. however, the ct r of the testing data only is 3.6%, ma king the baseline accuracy achieves 96.4%. proceedings of engineering and technology innovation, vol. 4, 2016, pp. 28 30 30 copyright © taeti 3. conclusions this paper adopts many ways to analyze how these relations influence in different kinds of mobile advertisement. a comprehensive perfo rmance comparison of different models is provided, and the analysis of different factors is also discussed, including click time, advertisement category, language, and mobile phone manufa cturers. this study uses different methods to predict ctr. results show that svm shows the best accuracy. acknowledgement the authors would like to thank the financial support provided by national science counil nsc102-2221-e-155-006, and hyxen technology co., ltd, taiwan. references [1] c. j. wang and h. h. chen, “ learning user behaviors for advertisements click predic tion,” proceedings of the 34rd international acm sigir conference on research and development in information retrieval workshop on internet advertising, 2011. [2] h. b. mcmahan, et. a l., “ad c lick predict ion: a view fro m the trenches,” proceedings of the 19th acm sigkdd international conference on knowledge discovery and data mining, 2013. [3] x. he, et al., “pract ical lessons from pre dicting clic ks on ads at facebook,” proceedings of the eighth international workshop on data m ining for online advertising, acm, 2014.  proceedings of engineering and technology innovation, vol. 1, 201 5, pp. 32 35 32 cop y right © taeti a critical review into the evolution of japanese project management: a comparative approach foon-siang low * , heap-yih chong faculty of engineering & science, universiti tunku abdul rahman, 53300 kuala lumpur, malaysia. received 28 april 2015; received in revised form 15 may 2015; accept ed 01 june 2015 abs tract the success or failure of a company relies on how a particular project is managed, and the proper use of project management improves management performance. this article emph asizes the importance of discovering more about project management with the help of two globally renowned project management methods, namely the united states (us) method (project management body of knowledge (pmbok)) and the japanese method (project & program management (p2m) and kaikaku (reform) project management (kpm)). the objectives of this article are (a) to identify the features and essence of p2m/kpm; and (b) to compare them with those of pmbok. features and essence of p2m/kpm are innovative reformations, value creation, flexibility, mission-driven approach, 3kkakusin (innovation), kaizen (improvement) and kaihatsu (development) and 3sscheme, system and service project models. the comparative results show that pmbok and p2m/kpm both have different philosophies in terms of scope, managerial approach and problem-solving methods. identifying suitable criteria will provoke interest in organizations to implement the most suitable method in their project management processes. keywor ds : p2m, kpm, pmbok, comparison 1. introduction the proper use of project management radically improves management effectiveness and performance [1]. although project managers in different countries run projects of a similar nature, they approach them in different ways in order to bridge the cultural differences of international stakeholders [2-3]. the most common and well-established united states (us) method, the project management body of knowledge (pmbok), has been adopted in many organizations around the world, including in japanese organizations. thus, it is interesting to point out the different project management methods that are more or less influenced by different cultures. on the other hand, although more than 2000 japanese have obtained the project management institute’s (pmi) project management professional qualification [4], the japanese also generated their own approach to project management in 2002, namely project & program management (p2m). although p2m is still being put into practical use both internationally and in japan [5], an improved paradigm called kaikaku (reforms or innovative reforms) project management (kpm) has been introduced [6]. the application and effectiveness of p2m and kpm need to be addressed, as many studies have shown an increased interest in and appreciation for japanese management principles and practices in recent years [7]. this research thus stresses the relevance of identifying the features and essence of two globally renowned project management methods, pmbok and p2m/kpm. a critical review of both these management methods was carried out and by clarifying the pros and cons of each method; a project manager can apply them when managing projects in order to obtain the ult imate results. therefore, the objectives of this article are (a) to identify the features and essence of p2m/kpm methods, and then, (b) to compare them with those of pmbok. 2. p2m and kpm japan’s economy declined in international competitiveness, dropping from number one in 1993 to number 30 in 2002. fu rthermore, it was threatened after china became the world’s factory, by producing improved qualities of various manufacturing goods at low costs [4]. these factors inspired the japanese to develop their own project management methods. p2m is the first japanese project and program management for enterprise innovation developed by professor shigenobu ohara in 20016 [8]. the p2m model aims at creating a strategic framework of innovation to improve corporate values in project management methodologies [4] and to create a way for japanese enterprises to develop more innovative approaches to ensure that their bus inesses can compete in the global business environment [9]. the p2m model has a combination of entry-level project management, program management, and 11 segment management frames [8]. the essence of p2m is focused on the profiling ideas of complexity to implementation and findings solutions to complex issues [10]. subsequently, to survive and regain their global competitiveness, the japanese looked for solutions in the kaikaku (reforms) of business management, organization, and technology [11]. although not all companies could accept reformation in order to keep up with the economic crisis, it was found that those who were successful at reforming were those who had utilized the intellectual property of the entire organization rather than those who had only focused on technological abilities [6, 11]. the p2m/kpm adopts mission-driven approach in managing their projects [6]. kpm consists of three significant elements for successful performance: 3k-kakusin (innovation), kaihatsu (development) and kaizen (improvement). kakusin is anything to do with creation of new ideas, devices or processes based on combination of new knowledge; kaihatsu is the challenge to acquire the latest knowledge and information; and kaizen is the continuing efforts for improvement at the work-floor level [12]. in p2m, there was no classification according to 3ks. however, the kpm method concentrates on the innovation, development, and improvement of japanese management methods using the foundations of p2m. thus, it takes into account the whole lifecycle of the project from idea, planning, execution, investment, and recuperation to creating value for the future. the kpm method promotes the creation of future value by implementing a number of reform projects linked to strategy, thus providing a body of knowledge to train core leaders, whose responsibility is to recoup the investment, and propose a methodology for avoiding the risks of failure and resistance in an organization that solves complex issues[6]. japanese organizations place an emphasis on the flexibility to adapt to environmental changes, and their models are created based on this concept. the strategies and methodologies of kpm have proven to be effective and successful in providing learning opportunities in companies, enhancing participation, and motivating the consensus and awareness of core leaders [12]. modeling idea is part of p2m/kpm which integrates interdisciplinary knowledge, methodologies and approaches [11]. there are three project models, namely 3sscheme, system and service, whereby these models represent a generic lifecycle of project combination from mission to capital recovery [13]. companies that construct *corresponding aut hor. email: lowfs@ut ar.edu.my t el.: +603-4107-9802; fax: +603-4107-9803 proceedings of engineering and technology innovation, vol. 1, 2015, pp. 32 35 33 cop y right © taeti their organizations with the elements of kpm, while being aware of the 3s project models, will have a project management system that functions well [14].overall, the summarized features and essences of p2m/kpm can be depicted in fig. 1. fig. 1 feat ures and essence of p 2m/kp m 3. pmbok the pmi (project management institute) attained pmbok standardization in 1983 and released its first edition of pmbok in 1996, with updated editions in 2000, 2004 and 200815. pmbok describes generally accepted knowledge and practices that are applicable to most projects most of the time [9, 16]. this standard describes the project management processes, tools, and techniques utilized to manage projects toward a successful outcome, and it divides projects into nine knowledge areas: project integration management, project scope management, project time management, project cost management, project quality management, project human resource management, project communications management, project risk management, and project procurement management [15]. not all key areas are applicable on all types of projects and it is the responsibility of the project management team to determine the appropriateness in a given project [9]. the main feature of pmbok is to acquire an expected performance in accordance with stakeholder needs by operational process [17]. operational process involves planning, executing and controlling. pmbok aims at achieving three requirements: time, cost, and project scope [18]. pmbok is strictly focused on bringing activities in line with a plan [19] and does not manage resources to achieve objectives [20]. managing projects with pmbok is found to be beneficial to various industries. a good project management method should be adopted to ensure the organization achieves its long-term goals. 4. me thodology an exploratory study has been carried out to examine project management theories by extracting relevant information from japanese and western project management guidebooks as well as from other major journals, articles, conference proceedings, and published books. two major processes were involved in this critical review. firstly, the principles of p2m and kpm were su mmarized and analyzed. subsequently, their essence or distinct features were evaluated by comparing them with those of pmbok. 5. comparative study the pmbok method is narrower in conceptual breadth and scope compared with p2m [21]. before, it deals mainly with projects, whereas p2m handles program management as well as project management; therefore, evaluations are based on the entire mission rather than on just a specific mission as practiced in project management [4]. however, four years after the first release of p2m, pmi has published standards for program management as well [22, 23]. however, there is limited information on program management in the latest pmbok version [15]. in view of social behavior in management, japanese people tend to practice collectivism and humanism, while the west emphasizes individualism [24]. individualism means work or job tasks are clearly defined and divided for each employee, and appraisals or evaluations are based on individual performance. in the p2m method, collaboration with colleagues or cooperation among team members is common. rather than individual goals, they look more into achieving group goals[25]. having a common goal facilitates good teamwork and eventually, enhances communication among project members. bad communications will result in communication conflicts that can provoke misunderstandings [26]. therefore, communication management is vital in managing projects. relationship in the japanese community is important for bringing together or connecting a group of people, such as the associations among project team members or departmental staff. in p2m, the relationship management framework is one of the 11 individual management domains that are not specifically featured in pmbok as one of the main knowledge areas. japanese coworkers have stronger friendships at work compared with their us counterparts [27], because they emphasize human relations and mutual trust [28]. trust will lead to the formation of long-term relationships [17]. for example, longer-term alliances and informal relational contracts have been applied in information technology (it) related projects in japan compared with the us approach [29]. when good relationships are built, employees feel obligated to perform at their best for their companies. based on the conventional japanese way of management, lifetime or lifelong employment is common in japanese organizations, which is based on an unwritten employment contract and thus it relies on mutual trust [28, 30]. lifetime employment encourages loyalty31. therefore, workers are rarely dismissed due to business downturns [32]. this is different from the american-style layoff [33]. culturally, the japanese take better care of their workers compared with the west [34]. however, when the economy turns sour in the 1990s for japan, their conventional model has to be reformed in order to survive the recession. restructuring, which includes work force lay-offs, production and inventory reductions, and plant closures, was applied [35]. for example, due to the above-mentioned scenario, the toyota group has revised its conventional lifelong employment system by accepting contract workers and temporary employees extensively [28]. at the same time, kaikaku or innovative (kakusin) reformations also encompass joint venture activities with foreign companies to overcome the recession [35, 36]. in addition, p2m/kpm also incorporated knowledge and intellectual properties into the kaihatsu and reform processes [23]. new inventions, development of new ideas, r&d and technologies know-how progressions play an important role in a company to secure the company’s competitiveness and to increase the value of intellectual property rights [28]. kaihatsu is not limited to development of technology alone, but also covers business, product, process, and even market as well [12]. just-in-time (jit) production is another innovative approach to regain profits [37], as it aims to reduce inventory to minimal. jit system only produces the necessary items in the required quantities when needed [38, 39]. p2m/kpm emphasizes flexibility and adaptability and proposes how organizations can achieve total optimization [6]. rather than focusing on mere partial optimization when economic crises occurred in the 1980s and 1990s, toyota maintained its market share by using the flexible kaizen philosophy. this philosophy p2m/kpm: features & essence project and program management innovative reformation mission-driven approach flexibility to adapt to environmental hanges 3k – kakusin, kaihatsu, kaizen 3s – scheme, system, services value creation integration management core philosophy of ‘solutions to complex issues’ proceedings of engineering and technology innovation, vol. 1, 201 5, pp. 32 35 34 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti applied to manufacturing and production processes [40, 41] that strive to eliminate waste and problems and to improve quality through persistent efforts [42]. other kaizen activities include continuous improvement of products, business operations and management systems, achieving highly flexible approaches to rapid changes, and learning from accumulated knowledge [8]. in pmbok, however, there is inflexibility in terms of project schedule, costs, and quality [9]. adaptability to environmental changes or flexibility in these aspects is not very much emphasized under the pmbok method. table 1 comparison between p2m/kpm and pmbok p2m/kpm pmbok s cope: -handles p rograms & p rojects 6 . -broader in scop e 21 . -evaluations based on the entire mission 4 . -collectivism 24 s cope: -deals mainly with p roject level 15 . -narrower in scop e 21 . -evaluations made on sp ecific missions 4 . -individualism 24 managerial approach: -concurrent development and integration management 23, 29, 44 . -lifelong emp loy ment and loy alty 31 . -do not p ractice layoff system in conventional model 28, 32 . -partially adopts the lay off system 28, 35 . managerial approach: -phased develop ment concep t 29 . -practice lay off system 33 . problem-solving methods: -flexibility ap proach 42 -m ission-oriented: focus on not only clients’ goals, but also how they are achieved, taking into account op inions and suggestions 13 . -ambiguous, uncertain 24 . -decision making outcomes favor inter-human relations 46 . problem-solving methods: -fixed approach on schedule, cost and quality 9 -goal-oriented: focus on delivering whatever is required by clients without considering other factors as long as the result is achieved 13 . -definiteness, clarity 24 . -decision making outcomes favor p erformance oriented 45 and recognition of effort 46 . another difference between the pmbok and p2m methods is that the former is relatively goal-oriented, whereas the latter is mission-driven [13, 43]. in order to accomplish a mission, p2m takes into account not only each process that comes along the way but also the detailed content of each process. it uses clear and measureable success measures for each project. missiondriven approach enables solving of complex problems by transforming strategic aims into value creation operations and capital recovery through the 3s project models [11]. by contrast, pmbok is prone to place emphasis on meeting the project objectives through initiating, planning, executing, monitoring and controlling, and closing [15]. in the construction industry, for instance, there is a need to go through certain stages, such as design, planning, construction, commissioning, and maintenance. these stages of the entire project follow a phased development concept, whereas parallel or concurrent development concept applies to japanese project management [29]. furthermore, p2m is the only standard that furnishes an integration management model across programs and portfolios of projects at enterprise level [23, 44]. last but not least, another point of difference between these two approaches is their styles of decision making, thinking and mindsets. in p2m/kpm, the japanese tend to be uncertain and ambiguous, whereas pmbok practices definiteness and clarity [24]. in the mindset of the western, logical thinking is emphasized and this eventually removes all ambiguities. things are defined and made clear. in terms of decision making, american business leaders tend to be more performance oriented [45]. they make decisions that will create opportunities for their efforts to be recognized, thus reflects a comparatively higher need for achievement [46]. in contrast to this, the japanese people give priority to inter-human relations [24], thus creating an indecisive environment. when making a decision, japanese business leaders tend to favor outcomes that preserve already established relationships or that could assist in cultivating new ones [46]. they need to consider the other party’s concerns and conditions before finalizing a conclusion or making a decis ion. conclusions are often made based on a group verdict. overall, a comparison of the differences between p2m/kpm and pmbok is summarized in table 1. 6. conclus ions project management is a special management technique applied globally in organizations, and its practice has proven to be successful in certain industries. the goal-oriented pmbok claims that excellent results are accomplished by focusing on time, cost, and quality and it promotes the idea of standardized project management. pmbok is prone to place emphasis on meeting the project objectives through the five process groups: initiating, planning, executing, monitoring and controlling, and closing, whereas the mission-oriented p2m/kpm takes into account the detailed content of each process and responds well to environmental changes. in other words, p2m/kpm emphasizes flexibility and adapting to environmental changes in order to overcome crises and takes into account other factors rather than only concentrating on the end result. the p2m/kpm method focuses on solving complications that occur during the process, concentrating on measures of success, customer satisfaction, and communication management. the japanese method of management also stresses loyalty and trust. in any industry, an organization requires good project management skills to increase the success rate in projects. improper management will lead to problems and trouble in the long run. thus, by highlighting and extracting the features, essences, and applications of pmbok and p2m/kpm, we get a clearer picture on how the east and west manage their projects in general. nevertheless, these two management styles are found to be beneficial to industries such as manufacturing, construction and it in their own ways, and the execution of projects under these management methods have proven to be successful. this article, however, has some limitations. it only touches on p2m/kpm and pmbok, briefly discussing their main features and essences. further in-depth investigations should be completed on each of the frameworks and features to understand in greater detail their essence, importance, and appropriateness in order to successfully apply them in a project. since at present, there is no such universal project management body of knowledge for any industry, selecting and combining the best features of pmbok and p2m/kpm might create a well-rounded project management body of knowledge that may work not only regionally but also for the rest of the industry. acknowle dge me nt it is an honor for us to express deep appreciation to p2m’s founder professor shigenobu ohara, a guest professor of tokyo university of agriculture and technology and vice president of international association of project and program management (ia-p2m) for his patience, guidance and resources. we would also like to extend our gratitude to the sumitomo foundation for granting us financial support under the sumitomo research grant (vote number: 108410). re fe re nce s [1] g. qiu, project management-engineering method and p ractice, beijing: science publication, 2001. proceedings of engineering and technology innovation, vol. 1, 2015, pp. 32 35 35 cop y right © taeti [2] m. saludin, practices and outcomes of p articip ative management in jap anese subsidiary companies in malaysia, phd dissertation, univ. putra malaysia, malaysia, 2005. [3] e. fisher, “what practitioners consider to be the skills and behaviours of an effective people project manager,” international journal of project management, vol. 29, no. 8, pp. 994-1002, december 2011. [4] s. ohara, what’s p2m, project management association of jap an, tokyo, 2003. [5] t. kinoshita, revitalizing japanese economy/firms and app ropriateness of p2m approach, project management association of japan, tokyo, 2005. [6] j. c. lee and d. mccalman, japanese management approaches: the fit for project m anagement , international journal of m anagement, vol. 25, no. 3, p . 584 2008. [7] s. ohara, p2m: a guidebook of project and program management for enterprise innovation (vol. 1), project management professionals certification center, tokyo, 2005. [8] p. dinsmore and j. cabanis-brewin, the ama handbook of p roject management 2nd ed., american management association, new york, 2006. [9] s. ohara, p2m-the japanese version of complex p roject management for enterprise innovation in turbulent environment, proc. 17th ipma international congress, moscow, 2003. [10] s. ohara and t. asada, japanese project management: kpm -innovation, development and improvement, singap ore: world scientific, pp. 313-337, singapore, 2009. [11] s. ohara and t. asada, japanese project management: kpm -innovation, development and improvement, singap ore: world scientific, pp. 5-23, singapore, 2009. [12] s. ohara, “mission driven approach (mda) of managing comp lex projects,” journal of the international association of project & program management, vol. 1, no. 1, pp. 61-70, 2006 [13] t. taketomi, s. ohara, and t. asada, japanese project management: kpm-innovation, development and improvement, singap ore: world scientific, pp. 45-59, 2009. [14] a guide to the project management body of knowledge (pm bok® guide), 4th ed., project management institute publications, usa, 2008. [15] i. wirth and d. tryloff, “preliminary comparison of six efforts to document the project-management body of knowledge,” international journal of project management, vol. 13, no. 2, pp. 109-118, april 1995. [16] t. asada, “a study of multip le value measures applied into strategic project m anagement,” international association of project & program management , pp. 46-62. (in japanese). [17] p. gao, j. w. feng, and h. t. wang, “develop ment and comp arative analysis of the project management bodies of knowledge,” management science and engineering, vol. 1, no. 1, p p . 106-111, sep tember 2007. [18] t. williams, “assessing and moving on from the dominant project management discourse in the light of project overruns,” ieee technology management council, vol. 52, no. 4, pp. 497-508, nov. 2005. [19] s. colly er and c. warren, project management approaches for dy namic environment, international journal of project management, vol. 27, no. 4, pp. 355-364, 2009. [20] p.w.g. morris, l. crawford, d. hodgson, m.m. shep herd, and j.thomas, “exploring the role of formal bodies of knowledge in defining a p rofession – the case of project management,” international journal of project management, vol. 24, no. 8, pp. 710-721, 2006. [21] the standard for program management, project management institute, newtown square, pa, 2006. [22] l. crawford, s. ohara and t. asada, japanese project management: kpm-innovation, development and improvement, singap ore: world scientific, pp.381-402, 2009. [23] s. ohara, “the fourth generation type japanese project management: sy nergetic integration of innovation, develop ment and imp rovement,” journal of international association of project & program management, vol. 2, no. 1, pp. 75-84, 2010. [in japanese] [24] j. liker, m . hoseus, and t.c. organizations, toyota culture: the heart and soul of the toyota way, new york: m cgraw-hill, 2008. [25] s. ohara, “reality research to multi stakeholder conflicts issue in p roject businessreframing agenda by psychology , standards, organization based triad approach,” proc. 12th national congress international association of project & program management, tokyo, japan, 2011. [26] j. r. lincoln, k. mcbride, “jap anese industrial organization in comp arative persp ective,” annual review of sociology , vol. 13, no. 1, p p. 289-312. 1987. [27] t. kinoshita, s. ohara, and t. asada, japanese project management: kpm-innovation, development and improvement, singap ore: world scientific, pp. 83-104, 2009. [28] a. tiwana and a. bush, “a comparison of transaction cost, agency , and knowledge-based predictors of it outsourcing decisions: a u.s.-japan cross-cultural field study,” journal of m anagement infornation systems, vol. 24, no. 1, pp. 259-300, 2007. [29] h. inohara, human resource development in japanese comp anies, tokyo: asian productivity organization, 1990. [30] n. oliver and b. wilkinson, the jap anization of british industry, uk: basil blackwell ltd, pp. 335-6, 1992. [31] a. rashid, k. jusoff, a. zalena and y. takahashi, “the japanese influence in m alaysian automotive industry: human resources m anagement and development practices,” journal of management science and engineering, vol. 3, no. 4, page 59, 2009. [32] c.b. m eek, “ganbatte: understanding the japanese employee,” business horizons, vol. 42, no. 1, pp. 27-36, 1999. [33] s. othman, ai-habshi, and a.h. ghazali, islamic values and m anagement, institute of islamic understanding m alaysia, ikl\1 publication, 2004. [34] r. schonberger, japanese production management: an evolutionwith mixed success, journal of operations management, vol. 25, no. 2, pp. 403-419, 2007. [35] d. robertson, how renault saved nissan’s bacon, sunday times business times, south africa, june 2004. [36] p. williamson, winning in asia: strategies for competing in the new millennium, harvard business school press, june 2004. [37] t. ohno, toyota production system beyond large-scale production, productivity press, kraus productivity, tokyo, 1988. [38] y. m atsui, an emp irical analysis of just-in time production in jap anese manufacturing companies, international journal of production economics, vol. 108, no. 1-2, pp. 153-164, 2007. [39] k. m urata and h. katayama, “development of kaizen case-base for effective technology transfer-a case of visual m anagement technology ,” international journal of production research, vol. 48, no. 16, pp. 4901-4917, 2010. [40] i. kato and a. smalley, toyota kaizen methods: six steps to imp rovement, new york: productivity press, 2011. [41] w. j. glover, j. a. farris, e. m. vanaken, and t. l. doolen, “critical success factors for the sustainability of kaizen event human resource outcomes: an empirical study,” international journal of production economics, vol. 132, no. 2, pp. 197-213, 2011. [42] g. f. jergeas , “fundamentals of project management,” http://www.ucalgary.ca/uofc/faculties/eng/projectmanagement/ jergeas/apegga-fundamentals1a-2004.pdf [43] s. ohara, “project management and qualification system in jap an —exp ectation for p2m and challenges,” proc. of international project management congress: project management develop ment in the asia-pacific region in the new century, toky o, japan, november 16-21, 2001. [44] r. house, p. hanges. m. javidan. p.dorfman and v. gupta, culture, leadership and organizations: the globe study of 62 societies, thousand oaks: sage publications, 818 pages, 2004. [45] m . m artinsons and r. davison, “strategic decision making and sup port systems: comparing american, japanese and chinese management,” decision support systems, vol. 43, no. 1, pp. 284-300, 2007.  proceedings of engineering and technology innovation, vol. 11, 2019, pp. 38-45 a point memory state observer with adjustable parameters for a class of uncertain linear systems with state delays shunya nagai 1 , hidetoshi oya 2,* , and tsuyoshi matsuki 3 1 department of information systems creation, kanagawa university, kanagawa, japan 2 department of computer science, tokyo city university, tokyo, japan 3 department of electronics and control engineering, national institute of technology, niihama college, ehime, japan received 10 may 2018; received in revised form 26 november 2018; accepted 07 december 2018 abstract in this paper, we present a point memory robust state observer with time-varying adjustable parameters for a class of uncertain linear systems with state delays. the point memory robust state observer proposed in this paper consists of fixed observer gain matrices and time-varying adjustable parameters, which are determined by updating rules. sufficient conditions for the existence of the proposed point memory robust state observer can be reduced to solvability of lmis. finally, simple numerical examples are included to illustrate the effectiveness of the proposed robust state observer. keywords: point memory robust state observer, time-varying adjustable parameters, state delays, lmis 1. introduction time delays are a phenomenon that occurs in physical systems such as manufacturing, long transmission lines in networked control systems and so on. when designing control systems, time delays is usually the reason that generates oscillation, poor performance, or instability of underlying control systems. namely, dealing with time delays is an important issue of controller design, and designers have to manage time delays so as to avoid the negative effects on the performance of control systems. therefore, various control strategies for time-delay systems have been widely studied (see. [1] and references therein). in particular, [2] has adopted a point memory feedback strategy and presented an lmi-based design method of lq regulator for a class of time-delay systems. additionally, for uncertain time delay systems, lots of existing results for robust controller design methods have been shown (e.g. [3-4]). particularly, in the work of [5], a guaranteed cost controller for a class of uncertain time-delay systems has been suggested. furthermore, a variable gain robust controller for a class of uncertain linear systems with state delays has also been proposed [6]. by the way, in the control theory, the concept of state variable feedback is the most fundamental strategy for controlling dynamical systems. in particular, for linear systems, when the liner system is controllable (e.g. [7-8]) and references therein), the closed-loop poles arbitrary can be assigned. however, due to physical, technical and/or economic reasons, not all variables can be measured; designers have to be able to deduce information on the state of the dynamical system by means of observations and measurements or a priori knowledge of the system structure. namely, the so-called state observers must be designed. if the estimate of the state variable can be constructed, by using the estimate can derive from a feedback control system based on the state observer. it is well-known that the state observer was first presented and developed by [9-10], and further studied by many researchers (e.g. [11]). the observer design only aimed to linear time-invariant systems with complete * corresponding author. e-mail address: hide@tcu.ac.jp proceedings of engineering and technology innovation, vol. 11, 2019, pp. 38-45 39 knowledge of system parameters, also the input and output signals. however, it is inevitable to include uncertain parameters and parameter variations; at the same time, the controlled systems are affected by unknown disturbances. therefore, state observers which guarantee the exactness of state estimation in the presence of unknown parameters are required. thus, the design problem of robust state observers for uncertain dynamical systems has been well studied, and in order to deal with this problem, a large number of existing results of robust state observer design have been presented [12-13]. in the work of wang et al.[14], a design method of an optimal observer of uncertain linear systems has been proposed. note that these robust state observers have fixed observer gains. furthermore, a robust state observer for linear systems with perturbations has been shown [15]. the [15] have introduced time-varying complementary variables, and based on a riccati equation approach a design method of the robust state observer has been suggested. additionally, the relation between solvability of the derived riccati equation and h∞-norm performance of a transfer function has been analyzed. for their work, a comment and authors' reply have also been presented [16-17]. however, the robust state observer given in [15] is incomplete. in this paper, on the basis of the existing result [6], we present a point memory robust state observer with time-varying adjustable parameters for a class of uncertain linear systems. the proposed point memory robust state observer consists of a time-varying adjustable parameter and a number of fixed gain parameters. in this paper, we derive complete sufficient conditions of the existence of the proposed variable gain robust state observer, and the sufficient conditions are given in terms of lmis. this paper is organized as follows; notations and useful lemmas, which are used in this paper, are shown in section 2. in section 3, we show the class of uncertain linear systems under consideration and an observer with fixed and time-varying adjustable parameters. section 4 is the main result in this paper. the design method of the proposed point memory robust state observer with time-varying adjustable parameters is presented. finally, we show simple illustrative examples to show the effectiveness of the robust state observer developed in this paper. 2. preliminaries in this section, notations, well-known and useful lemmas (see [18] for details) which are used in this paper are presented. the following notations are used in this paper. for a matrix ,s the inverse of matrix ,s and its transpose are denoted by 1 ,  s and t s , , respectively. moreover,  seh and ni mean t ss  and n -dimensional identity matrix, respectively, and for real symmetric matrices w and z , zw  (resp. zw  ) means that zw  is positive (resp. nonnegative) definite matrix. for a vector , n w  w denotes standard euclidian norm and for a matrix w , w represents its induced norm. the symbols “  ” and “  ” mean equality by definition and symmetric blocks in matrix inequalities, respectively. lemma 1 (schur complement formula [18]) : for a given constant real symmetric matrix  , the following items are equivalent: (i) ψ = ( ψ11 ψ12 ∗ ψ22 ) > 0, (ii) ψ11 > 0 and ψ22 − ψ12 t ψ11 −1ψ12 > 0, (iii) ψ22 > 0 and ψ11 − ψ12ψ22 −1ψ12 t > 0. 3. problem formulation consider the uncertain time delay system described as the following state equation;     ),()( ),()()()()()( tcxty tbuhtxetcatxetcatx dt d hh t h t   (1) where n tx )( , m tu )( and l ty )( are the vectors of the state, the control input and the measured output, respectively. in eq. (1), matrices ,a b and c represent the nominal systems parameters with appropriate dimensions, and proceedings of engineering and technology innovation, vol. 11, 2019, pp. 38-45 40 nq e   and n h q h e   denote the structure of unknown parameters. moreover ql t   )( and h ql h t   )( are unknown time-varying parameters which satisfy the relations 0.1| |)(| |  t and 0.1| |)(| |  t h , respectively. now the design problem under consideration is how to design a variable gain robust state observer with the measured output l ty )( and the control input m tu )( . the estimation error between the state variable and the estimate based on the observer can converge asymptotically to 0. in this paper, we introduce the following point memory robust state observer; ),()())()(())()(()()()( ttbuhtxchty h ltxctylhtx h atxatx dt d           (2) where n tx   )( is the estimate of the state variable )(tx , and ln l   and ln h l   are fixed observer gain parameters. moreover n t )( in eq. (2) is a compensation input with time-varying adjustable parameters [6]. then we obtain the following estimation error system; ),()()()()()()()()()( thtx h et h t ctext t chtec h l h atelcate dt d  (3) where )(te is an estimation error vector defined as )()()( txtxte   , from the above, our control objective in this paper is to construct the point memory robust state observer of eq. (2), for the uncertain time-delay system of eq. (1). that is to derive the fixed gain matrices ln l   and ln h l   , and the compensation input n t )( such that the estimation error system of eq. (3) is asymptotically stable. 4. main results in this section, we derive an lmi-based design procedure for the proposed point memory robust state observer with time-varying adjustable parameters. next theorem gives a sufficient condition for the existence of the proposed robust state observer. theorem 1: consider the uncertain time delay system of eq. (1) and the robust state observer of eq. (2). if the symmetric positive matrices exist, nn p   , nn h p   and ll x   , matrices ln w   and ln h w   and positive scalars , ,h  and h which satisfy the lmis   ,0 000 0000                           l i h l i h l i l i lnlnlnlnh e t h e h p t pc t pc t pc t pcc h w h pa h pe t ewcp t aeh       (4) ,0          l i t pcxc t c (5) then the fixed gain matrices ln l   and ln h l   are determined as wpl 1  and hh wpl 1  , respectively and the compensation input n t )( is designed as: note that if 0)( 2/1 tcex , then ),()(  tt  where )(lim 0    tt [19]. proceedings of engineering and technology innovation, vol. 11, 2019, pp. 38-45 41 proof: in order to prove theorem 1, we introduce the following function: .)( 1 2 )( 2/1 2 )( 2 )( )( txce t cp tcex htx h e h txe t         (6) by using the point memory robust state observer with the fixed gain matrices ln l   and ln h l   , and the compensation input n t )( , the estimation error )(te converges asymptotically to 0. . 0 )()()()(),(    h dte h pt t etpet t etev  (7) the time derivative of the function ),( tev of eq. (7) along with the trajectory of the estimation error system of eq. (4) is given by  ( , ) ( ) ( ) ( ) 2 ( ) ( ) ( ) 2 ( ) ( ) ( ) 2 ( ) ( ) ( ) 2 ( ) ( ) ( ) ( ) ( ) ( ). d t t t v e t e t h p a lc e t e t pc t ex t edt t t t e t p a l c e t h e t pc t e x t h h h h h t t t e t p t e t p e t e t h p e t h h h                    (8) if the relation 0)()( tcpepcte tt holds, then it is to be seen from eq. (6) and the relation 0)()(2  tpte t  that the following inequality is satisfied;  ( , ) ( ) ( ) ( ) 2 ( ) ( ) ( ) ( ) ( ) ( ) ( ) d t t v e t e t h p a lc e t e t p a l c e t h e h hdt t t e t p e t e t h p e t h h h             (9) the inequality of eq. (9) can be written as:   . )( )()()( )( )( ),(                       hte te h p c h l h ap h plcapeh t hte te tev dt d (10) thus the matrix inequality:   ,0 )()(          h p c h l h ap h plcapeh (11) is satisfied, then the following relation holds; ( , ) 0, ( ) 0 ( ) 0. d v e t e t and e t h dt       (12) next we consider the case of 0)()( tcpepcte tt . additionally, we introduce the relation: ),()()()( txce t ct t etcpe t pct t e  (13) where ll x   is a symmetric positive definite matrix. note that if the condition 0)()( txcecte tt holds then 0)()( tcpepcte tt is also satisfied. we can see that the relation of eq. (13) means the following inequality; proceedings of engineering and technology innovation, vol. 11, 2019, pp. 38-45 42 ,cp t pcxc t c  (14) by applying lemma 1(schur complement formula) to eq. (14), we obtain .0          l i t pcxc t c (15) it is obvious that if the condition of eq. (10) holds, then we have 0)()( txcecte tt in the case of .0)()( tcpepcte tt from the definition of the estimation error vector, the relations )()()( txtetx   and )()()( htxhtehtx    holds. therefore, the time derivative of the function ),( tev can be rewritten as     ( , ) ( ) ( ) ( ) 2 ( ) ( ) ( ( )) ( )) 2 ( ) ( ) ( ) 2 ( ) ( ) ( ( ) ( )) 2 ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) ( ) 2 ( ) ( c) ( ) d t t t v e t e t h p a lc e t e t pc t e e t x t edt t t t e t p a l c e t h e t pc t e e t h x t h h h h h t t t e t p t e t p e t e t h p e t h h h t t e t h p a lc e t e t p a l e t e h h h                                 1 ( ) ( ) ( ) ( ) 1 ( ) ( ) ( ) ( ) 1 ( ) ( ) ( ) ( ) 1 ( ) ( ) ( ) ( ) 2 ( ) ( ) ( ) ( ) ( ) ( ). t t t t e t pc cpe t e t e ee t t t t t e t pc cpe t x t e e x t t t t t e t pc cpe t e t h e e e t h h h h t t t t e t pc cpe t h x t h e e x t h h h h t t t t p t e t p e t e t h p e t h h h                              (16) here we used the relation )()()( txtxte   and the well-known inequality b t ba t ab t a   1 2  (17) for any vectors a and b with appropriate dimensions and any positive constant . by substituting the compensation input of eq. (6) to eq. (16) and some algebraic manipulations gives . )( )( ),,,,,( )( )( ),(                 hte te hhh pp t hte te tev dt d  (18) in eq. (15), nn hhh pp 22 ),,,,,(    is the matrix given by          h p c h l h ap hhh pp hhh pp )(),,,,.( 11),,,,,(   (19)   h pe t ecp t pc hh lcapehhhh pp             1111 )(),,,,,( 11 (20) therefore, the inequality condition of eq.(12) is satisfied provided that the following condition holds, .0),,,,,(  hhh pp  (21) proceedings of engineering and technology innovation, vol. 11, 2019, pp. 38-45 43 additionally, one can clearly see that the matrix inequality of eq. (21) is a sufficient condition of eq. (11), i.e. if the matrix inequality of eq. (21) holds then the condition of eq. (11) is also satisfied. finally, we consider the matrix inequality of eq. (21). by applying lemma 1 (schur complement formula) to eq. (21), we can obtain the lmi of eq. (4). thus, the proof of theorem 1 is accomplished. 5. numerical example this section shows a simple numerical example to demonstrate the proposed point memory robust state observer. in this example, we consider the uncertain time delay system given by   1.0 1.0 1.0 0.5 0.0 ( ) ( ) ( ) 0.0 1.0 1.0 0.0 0.0 0.5 0.0 1.0 0.0 0.0 0.0 ( ) ( 0.5) ( ), 0.0 0.5 1.0 0.0 0.5 1.0 ( ) 1.0 1.0 ( ). h d x t t x t dt t x t u t y t x t                                                              　 　 　 (22) firstly, by solving lmis of eq. (4) and eq. (5), we have symmetric positive definite matrices 2 2p  , 2 2 h p  and 1 1x  , matrices 2 1w  and 2 1 h w  and positive scalars , ,h  and h as 3 2 3 3 3 3 2 2 2 2 2.1570 2.1208 5.6711 10 1.3509 10 10 , , 6.4073 102.1570 4.2335 5.0857 4.0718 10 , 10 , 10 , 1.7941 4.9957 5.4931 10 , 5.4351 10 , 5.4737 10 , 5.4632 10 h h h h p p x w w                                              　 　 　 　 　 = 　 2. (23) thus the fixed gain matrices 2 1l  and 2 1 h l  can be calculated as 1 1 3 8.3413 2.4208 10 10 , . 8.2843 6.4078 10 h l l                  　 (24) in this example, initial values for the uncertain linear system of eq. (22) and the proposed variable gain robust state observer are selected as  (0) 1.0 0.5 t x   and  (0) 0.0 0.0 t x   , respectively. furthermore, unknown parameters are given as  ( ) 0.0 cos(5 )t t  and, respectively. moreover, we assume that the control input is given by ( ) sin( )u t t . the simulation result of this numerical example is shown in figs. 1-4. in these figures, xl, xextl and el (l=1,2) denotes the l-th element of the state, the estimate of the state variable and the estimation error. from these figures, we can see that the proposed point memory robust state observer can estimate the state variables of the uncertain time delay system of eq. (22) and the estimation error for the proposed point memory robust state observer converges to 0. thus we have shown the effectiveness of the proposed point memory robust state observer. fig. 1 time histories of the first element of ( )x t and ( )tx  fig. 2 time histories of the second element of ( )x t and ( )tx  proceedings of engineering and technology innovation, vol. 11, 2019, pp. 38-45 44 fig. 3 time histories of the first ( )e t fig. 4 time histories of ( )e t 6. conclusions in this paper, on the basis of the work of endo et al.[6], we have proposed a new point memory robust state observer for a class of uncertain time delay systems. the sufficient condition for the existence of the proposed point memory robust state observer is reduced to lmis, and thus the proposed robust state observer can easily be obtained. future research subjects are going to extend the proposed point memory robust state observer to uncertain large-scale interconnected systems with state delays, uncertain discrete-time systems and so on. furthermore, we will evaluate the conservativeness of the proposed observer design. conflicts of interest the authors declare no conflict of interest. references [1] k. gu, v. l. kharitonov, and j. chen, “stability of time-delay systems,” birkhauser, 2003. [2] t. kubo, “lq regulator of systems with time-delay in the states by the point memory feedback,” ieice trans. fundamentals of electronics, communications and computer sciences, vol. j87-a, no. 4, pp. 577-579, 2004. [3] x. li and c. e. d. souza, “delay-dependent robust stability and stabilization of uncertain linear delay systems: a linear matrix inequality approach,” ieee transactions on automatic control, vol. 42, no. 8, pp. 1144-1148, august 1997. [4] x. li and c. e. d. souza, “criteria for robust stability and stabilization of uncertain linear systems with state delay,” automatica, vol. 33, no. 9, pp. 1657-1662, september 1997. [5] 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, june 1999. [6] k. endo, h. oya, t. kubo, and t. matsuki, “synthesis of variable gain robust controllers based on point memory lq regulator for a class of uncertain time-delay systems,” 2015 10th asian control conference, ieee press, pp. 2957-2961, september 2015. [7] m. norton, “modern control engineering,” pergamon press, 1972. [8] b. d. o. anderson and j. b. moore, “optimal control-linear quadratic method,” prentice hall, 1990. [9] d. g. luenberger, “observing the state of a linear system,” ieee transactions on military electronics, vol. 8, no. 2, pp. 74-80, 1964. [10] d. g. luenberger, “observer for multivariable systems,” ieee transactions on automatic control, vol. 11, no. 2, pp. 190-197, april 1966. [11] b. gopinath, “on the control of linear multiple input-output systems,” the bell system technical journal, vol. 50, no. 3, pp. 1063-1081, 1971. [12] j. c. doyle and g. stein, “robustness with observers,” ieee transactions on automatic control, vol. 24, no. 4, pp. 607-611, 1979. [13] b. l. walcott and s. h. zak, “observation of dynamical systems in the presence of bounded nonlinearities/uncertainties,” 25th ieee conference on decision and control, ieee press, april 2007. [14] s. d. wang, t. s. kuo, and c. f. hsu, “optimal-observer design for linear dynamical systems with uncertain parameters,” international journal of control, vol. 45, no. 2, pp. 701-711, march 2007. proceedings of engineering and technology innovation, vol. 11, 2019, pp. 38-45 45 [15] d. w. gu and f. w. poon, “a robust state observer scheme,” ieee transactions on automatic control, vol. 46, no. 12, pp. 1958-1963, december 2001. [16] m. boutayab and m. darouach, “comments of a robust state observer scheme,” ieee transactions on automatic control, vol. 48, no. 7, pp. 1292-1293, 2003. [17] d. w. gu and f. w. poon, “authors’ reply,” ieee transactions on automatic control, vol. 48, no. 7, p. 1293, 2003. [18] s. boyd, l. e. ghaoui, e. feron, and v. balakrishnan, “linear matrix inequalities in system and control theory,” siam, 1994. [19] m. maki and k. hagino, “robust control with adaptation mechanism for improving transient behaviour,” international journal of control, vol. 72, no. 13, pp. 1218-1226, 1999. 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/).  proceedings of engineering and technology innovation, vol. 6, 2017, pp. 23 27 flow analysis of underwater water jet system using computational fluid dynamics and particle image velocimetry je-doo ryu * , kyoung-nam ha, dong-gu lee, keon-seok nam korea institute of industrial technology, 474, ilgwang-ro, ilgwang-myeon, gijang-gun, busan, korea. received 26 july 2017; received in revised form 12 august 2017; accepted 27 august 2017 abstract the width and depth of excavation are important for the development of a water jet system for underwater surface excavation in the water. this is determined by the pressure distribution affecting the ground. therefore, it is necessary to analyze the flow of the water jet, but it is difficult to grasp visually, and generally used sensors are difficult to use in water. to overcome this problem, the flow analysis was simulated by using computational fluid dynamics (cfd), particle image velocimetry (piv) experiment was performed to confirm the accuracy of cfd and to visualize the nozzle flow. the cfd was simulated by using midas nfx and various pressures of nozzle injection, velocity distribution and pressure distribution were analyzed. for the piv experiment, the water jet system is installed at l22m x w3.5m x d3m water tank, and a 5w nd-yag laser was installed at the surface of the water, and a high-speed camera was installed to capture the flow in the side observation field. and particles of 20 micrometer size were mixed in the water. in the results of the cfd simulations, the nozzle injection pressure, which is the measurable velocity of the piv system was selected, the nozzle flow was visualized and the velocity distribution was measured by using the piv experiment. the velocity distribution measured by piv is compared with the cfd analyzed the velocity distribution under the same conditions. keywords: computational fluid dynamics (cfd), particle image velocimetry (piv), water jet 1. introduction the ocean takes up two-thirds of the earth's surface, marine technology development is essential and being developed. as a result of many developments such as offshore plant and ocean exploration and offshore wind power, one of the necessary technologies is the technology of digging the sea-bed. it uses mechanical plows, drilling and cutter methods, and waterjet methods to dig the seafloor. the method of digging the sea-bed by direct force such as plowing, drilling, and cutter method is complicated in structure, and power and operation vessel problem as water depth becomes deeper, a waterjet-based excavation method is started to be used due to waterproofing problem. the excavation method using waterjet has a disadvantage of relatively simple structure compared to other mechanical excavation methods, but it is difficult to predict the desired excavation area. in this paper, we try to grasp the excavation area and the depth of the nozzle arrangement by assuming that the excavation method using waterjet has a great influence on the desired excavation area. * corresponding author. e-mail address: ryujd88@kitech.re.kr proceedings of engineering and technology innovation, vol. 6, 2017, pp. 23 27 copyright © taeti 24 this paper is expected to have a great influence on the nozzle placement on the desired excavation area in the water jet excavation method. in the underwater environment, we cannot use the sensors that are commonly used, and because it is difficult to quantitatively measure through experiments, we go through a lot of computational fluid dynamics (cfd). in this paper, we investigate the reliability of cfd by comparing particle image velocimetry (piv) experiment with the cfd results before the cfd analysis by the nozzle. 2. experimental conditions to verify the validity of the cfd before simulating the nozzle arrangement using cfd, the nozzle injection speed and flow field are measured using the piv system. 2.1. piv (particle image velocimetry) the nozzle spraying system and the laser system were installed in a water tank of length 22 m width 3.5 m depth 3 m. the nozzle injection system consists of a pump (maximum output pressure 13 bar), a solenoid valve, a pressure sensor, a flow sensor, and a nozzle. the laser system consists of a nd-yag 3w, a laser with a wavelength of 532 nm and a width of 2 mm or less, and a wave plate to allow the laser to penetrate below the water surface. a high-speed camera capable of shooting at up to 2000 fps was installed for measurement on the side. fig. 1 piv system particles of 20μm or less were dissolved in the water tank, and the pump sucks the water and discharges it to the nozzle. piv experiments were conducted while finding the nozzle injection pressure at which the particles were measured while changing the nozzle injection pressure. the order of the piv experiment is as follows: (1) position the laser at the desired position and align the laser sheet with the center of the nozzle. (2) set the high speed camera shooting position to the position where you want to measure. (3) control the pump rpm to discharge the desired pressure. (4) open the solenoid valve when the desired pressure is reached and the piv shooting is ready. (5) take a picture with a high-speed camera. (6) make sure that sufficient amount of particles are taken clearly in the picture. (7) after shooting is completed, process the piv after the shot. proceedings of engineering and technology innovation, vol. 6, 2017, pp. 23 27 copyright © taeti 25 (a) nozzle injection (b) piv flow visualization fig. 2 nozzle injection experiment experimental results show that the particle velocity is measured at 0.1 bar and the measurement speed of the high speed camera is 100 fps. the results obtained by piv analysis are shown in the figure. (a) water jet streamline of piv result (b) water jet flow velocity of piv result fig. 3 piv result 2.2. computational fluid dynamics (cfd) to compare the cfd results with the piv results, the experiment conditions that were tested in piv were applied to the cfd. the cfd used nfx 2015 from madis. based on the design data of the nozzle shape, we made 3d shape as shown in the figure and made 229,000 elements in 1mx 1mx1m space. the pressure at the nozzle inlet end was set to 0.1 bar. proceedings of engineering and technology innovation, vol. 6, 2017, pp. 23 27 copyright © taeti 26 fig. 4 nozzle 3d shape fig. 5 mesh generation result fig. 6 water jet flow velocity of cfd analysis result fig. 7 water jet flow velocity of cfd analysis result 2.3. comparison of cfd and piv results in order to compare cfd and piv analysis results, the flow velocity was compared with the height of the nozzle injection center at a distance of 20 cm from the nozzle outlet end. proceedings of engineering and technology innovation, vol. 6, 2017, pp. 23 27 copyright © taeti 27 the piv measurement results and the cfd analysis results were similar to each other, with the nozzle injection center velocity being the fastest, and the velocity being sharply decreasing with increasing distance from the center. however, the difference between the two maximum flow velocity is large. the maximum flow velocity the piv measurement is 0.27m/s and the maximum flow velocity the cfd analysis result is 0.93m/s. and velocity difference was large at 0.05m radius from the nozzle injection center. fig. 8 water jet flow velocity of piv and cfd result at 0.2m 3. conclusions in this paper, piv measurement results and cfd analysis results were compared in order to analyze the flow of underwater waterjet. it was thought that the validity of the cfd could be verified through the piv, but the results of two experiments showed a great difference. the maximum velocity of the cfd analysis result is much higher than that of piv. this is due to hardware limitations of the piv measurement system. the measurement speed of the high speed camera changes according to the brightness of the laser light and the actual 30cm size is measured at 100fps with 1024 x 1024 image. then the actual measurable speed is 0.3m / s or less. it is expected that additional piv experiments should be conducted by changing the laser or high-speed camera. thus, when the verification of cfd is completed first, the effect of waterjet excavation according to nozzle spacing and arrangement with cfd will be further studied. references [1] h. a. kim, k. m. kwon, h. i. park, w. d. jung, and h. w. kwak, “a basic study on nozzle design of the sub-sea trenching machine,” the korean society for marine environment & energy, pp. 387-392, november 2006. [2] h. s. choi, h. j. kim, c. k. yang, and k. y. hong, “a study on the excavation mechanism of sea bottom by water jetting,” the korean society for marine environment & energy, pp. 31-36, may 2000.  proceedings of engineering and technology innovation, vol. 11, 2019, pp. 46-50 the design and prototype manufacture of a planetary gear reducer 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 02 may 2018; received in revised form 28 may 2018; accepted 05 august 2018 abstract the gear reducer of the power system can be ordinary spur gear reducer, worm and worm gear reducer, or planetary gear reducer. due to the compact size, light weight, and multi-degrees of freedom; planetary gear reducers are commonly used in various transmissions. one of its applications is used as the gear reducer for industrial purpose. the reduction ratios of 1-stage planetary gear reducers are limited to 3 ~ 10. this paper focused on the design and prototype manufacture of a 1-stage planetary gear reducer. first, according to the concept of train value equation, the planetary gear reducer with reduction ratio 4 is proposed. based on the involute theorem, the gear data of the planetary gear reducer are obtained. finally, based on the results of kinematic design and meshing efficiency analysis, the integrated design of the planetary gear reducer was carried out and the prototype was manufactured to verify the design theorem. the results of this paper can be used as a reference for engineers to design the gear reducers for industrial purpose. keywords: engineering design, planetary gear reducer, prototype manufacture, train value equation 1. introduction traditionally, the reduction ratio of an ordinary spurs gear pair, worm and worm gear pair, and 1-stage planetary gear train is limited to 4~7, 10~100, and 3 ~ 10, respectively. figs. 1 (a)-(c) show the ordinary spur gear reducer, worm and worm gear reducer, and 1-stage planetary gear reducer, respectively. due to the reason of the compact size, light weight, and multi-degrees of freedom; planetary gear reducers are commonly used in various transmissions. one of its applications is used as the gear reducer for industrial purpose. (a) ordinary spur gear reducer (b) worm and worm gear reducer (c) 1-stage planetary gear reducer fig. 1 gear reducers in the past, planetary gear trains were the subjects of intensive researches directed at kinematic analysis [1-5], kinematic design [6-13], and meshing efficiency analysis [14-19]. the purpose of this paper is to propose a planetary gear reducer with reduction ratio rr=4 for industry purpose. the results of this paper can be used as a reference for engineers to design all kinds of planetary gear reducers. the research result can also provide the application experiences for the transmission design practice course. * corresponding author. e-mail address: summer34134@gmail.com tel.: +886-3-559-3142 ext. 3023; fax: +886-3-557-3797 proceedings of engineering and technology innovation, vol. 11, 2019, pp. 46-50 47 2. train value equation for a planetary gear train, shown in fig. 2, we denoted the first sun gear as i, the last sun gear as j, and the carrier (arm) as k, respectively. the train circuit can be denoted as (i, j; k); the relationship among ωi, ωj, and ωk can be expressed as: fig. 2 planetary gear train 0)1(  kjijjii  (1) here ξji is the train value of sun gear j to sun gear i. the sum of the coefficients of eq. (1) is equal to 0, i.e., 0)1(1  jiji  . for a planetary gear train with the single train circuit, there are three members (i, j, and k) can be adjacent to input shaft, output shaft, and frame. if sun gear j is adjacent to the frame (ωj=0), and sun gear i is adjacent to the input shaft (ωi=ωin), and carrier k is adjacent to the output shaft (ωk=ωout), according to eq. (1), its reduction ratio (rr) can be written as:  1 ( 1) 1 jiin i r ji out k r             (2) for a planetary gear train with ξji < 0, the reduction ratio rr > 1. for the planetary gear reducer shown in fig. 3, let z2=24, z5=24, and z3=72, then there is the train value ξ32=-3. since the sun gear 3 is adjacent to the frame (ω3=0), and the sun gear 2 is adjacent to the input shaft (ω2=ωin); the carrier 4 is adjacent to the output shaft (ω4=ωout), according to eq. (2), its reduction ratio rr = 4. fig. 3 planetary gear reducer with reduction ratio rr = 4 3. latent power theorem and meshing efficiency for a planetary gear train with sun gear i, ring gear j, and carrier k. ti, tj and tk are the torques of members i, j and k, respectively. since they rotate around the same axis, and according to the torque balance, we have: 0 kji ttt (3) furthermore, the latent powers of sun gear i and sun gear j relative to carrier k ( 𝑃𝑖 𝑘 and 𝑃𝑗 𝑘 )are defined as :   kii k i tp   (4)   kjj k j tp   (5) 𝜂𝑖𝑗 𝑘 ( 𝜂𝑗𝑖 𝑘 ) is the meshing efficiency of planetary gear train; if the latent power is transmitted from sun gear i (sun gear j) to sun gear j (sun gear i), when carrier k is relatively fixed. based on the concept of latent power [14-19], the relationship between k i p and k j p can be expressed as: proceedings of engineering and technology innovation, vol. 11, 2019, pp. 46-50 48 0p and 0 pfor pp k j k i k ij k i k j   (6) 0p and 0 pfor pp k j k i k ji k j k i   (7) the meshing efficiency of the planetary gear train is defined as the ratio of output power to input power, i.e., ( ) out m pgt in p p    (8) according to eqs. (3) ~ (8), the meshing efficiency of planetary gear reducer can be expressed as follows:     ( ) 1 1 k ji ij m pgt ji        (9) where 𝜂𝑖𝑗 𝑘 ( 𝜂𝑗𝑖 𝑘 ) is the meshing efficiency of planetary gear train; if the latent power is transmitted from sun gear i (sun gear j) to sun gear j (sun gear i) when carrier k is relatively fixed. the meshing efficiency of ordinary external gear pair (m(ex)) and internal gear pair(m(in)) can be expressed as [20-22]: 2 2 ( ) 1 1 1 1 1 1 / 1 [ ] 2 2 ar r m ex a r a r fr f           (10) 2 2 ( ) 1 1 1 1 1 1 / 1 [ ] 2 2 ar r m in a r a r fr f           (11) where rr is the reduction ratio of gear pair, βa1 and βa2 are the angles of approach of driving and driven gears, fa and fr are the average friction coefficient of approach and recess. for external (internal) gear pair, the meshing efficiencies are between 99.2% ~ 99.6% (99.7% ~ 99.9%). for the planetary gear reducer in fig. 3, if the meshing efficiencies of external and internal gear pairs are 0.994 and 0.998, then the meshing efficiency of 𝜂𝑖𝑗 𝑘 is 0.994 0.998 0.992 k ij     . then, according to eq. (9), the meshing efficiency of the planetary gear reducer is 99.4%. based on the above results, we concluded that the meshing efficiencies of planetary gear reducers are very good. 4. engineering design and prototype in the following paragraph, the planetary gear reducer, shown in fig. 3, is the example to illustrate how to design the planetary gear reducer. table 1 shows some important gear data of this feasible design example. table 1 gear data of the planetary gear reducer shown in fig. 3 external gear pair (2, 5) internal gear pair (5, 3) sun gear 2 planet gear 5 ring gear 3 teeth number 24 24 72 normal module (mm) 1.25 1.25 1.25 helical angle 0⁰ 0⁰ 0⁰ shift coefficient 0 0 0.3 pitch dia (mm) 30 30 90 addendum dia. (mm) 32.5 32.5 88.25 base dia. (mm) 28.19078 28.19078 84.57234 center distance (mm) 30.2 30.2 pressure angle (deg.) 21d 1m 4s 21d 1m 4s contact ratio (approach/ recess) 1.44721 (0.72361 / 0.72361) 1.71049 (0.98689 / 0.72361) according to the gear data in table 1, the engineering design drawings shown in figs. 4(a), 4(b), 4(c),and 4(d) are sun gear 2 (z2=24), planet gear 5 (z5=24), ring gear 3 (z3=72), and planetary gear set of planetary gear reducer. figs. 4(e) and 4(f) show its corresponding 1/2 section view and isometric view. proceedings of engineering and technology innovation, vol. 11, 2019, pp. 46-50 49 (a) sun gear (z2=24) (b) planet gear (z5=24) (c) rnig gear (z3=72) (d) planetary gear set (e) sectional view (f) isogonal view fig. 4 engineering drawing of planetary gear reducer figs. 5(a), 5(b), 5(c), and 5(d) show the corresponding prototype photo of sun gear 2 (z2=24), planet gear 5 (z5=24), ring gear 3 (z3=72), and planetary gear set of planetary gear reducer. figs. 5(e) and 5(f) show its corresponding prototype photo of 1/2 section view and isometric view. (a) sun gear (z2=24) (b) planet gear (z5=24) (c) rnig gear (z3=72) (d) planetary gear set (e) sectional view (f) isogonal view fig. 5 prototype photos of planetary gear reducer 5. conclusions the reduction ratios of 1-stage planetary gear reducers are limited to 3 ~ 10. this paper focuses on the integrated design and prototype manufacture of the planetary gear reducer (with reduction ratio rr=4) for industry purpose. according to the kinematic design, a planetary gear reducer with rr=4 is synthesized. according to the latent power theorem, its corresponding meshing efficiency is calculated to be 99. 4%. finally, the engineering design of the planetary gear reducer is accomplished and its corresponding prototype is manufactured to verify the design theorem. the results of this paper can be used as reference for engineering to design the gear reducer for industrial purpose and will enhance the competitiveness of gear industrial. acknowledgements the authors are grateful to the ministry of of the republic of china for the support of this science and technology research under most 106-2221-e-150-015. conflicts of interest the authors declare no conflict of interest. proceedings of engineering and technology innovation, vol. 11, 2019, pp. 46-50 50 references [1] l. w. tsai, “an algorithm for the kinematic analysis of epicyclic gear trains,” proc. 9th applied mechanisms conf, 1985, pp. 1-5. [2] l. c. hsieh, h. s. yan, and l. i. wu, “an algorithm for the automatic kinematic analysis of planetary gear trains by fundamental circuit method,” journal of the chinese society of mechanical engineers, vol. 10, no. 2, pp. 153-158, 1989. [3] c.h. hsu, “an analytic methodology for the kinematic synthesis of epicyclic gear mechanisms,” journal of mechanical design, vol. 124, no.3, pp. 574-576, 2002. [4] m. c. tsai and c. c. huang, “kinematic analysis of planetary gear systems using block diagrams,” journal of mechanical design, vol. 132, no. 6, 2010. [5] y. c. wu, p. w. ren, and l. a. chen, “kinematic analysis of an 8-speed bicycle transmission hub,” applied mechanics and materials, vol. 479-480, pp. 234-238, 2014. [6] l. c hsieh, j. y. liu, and m. h. hsu, “systematic design of south pointing chariots with planetary gear trains,” transaction of canadian society for mechanical engineering, vol. 20, no. 4, pp. 421-435, 1996. [7] l. c hsieh, “an efficient method for the kinematic design of coupled gear differentials for automobiles,” journal of applied mechanisms and robotics, vol. 4, no. 1, pp. 7-12, 1997. [8] w. m. hwang and y. l. huang, “configuration design of six-speed automatic transmissions with two-degree-of-freedom planetary gear trains,” transactions of the canadian society for mechanical engineering, vol. 29, no. 1, pp. 41-55, 2005. [9] l. c. hsieh, h. s. lee, and t. h. chen, “an algorithm for the kinematic design of gear transmissions with high reduction ratio,” materials science forum, vol. 505-507, pp. 1003-1008, 2006. [10] l. c. hsieh, and t. h. chen, “the systematic design of multi-speed internal gear hub for a bicycle,” advanced materials research, vol. 199-200, pp. 431-435, 2011. [11] l. c. hsieh, and h. c. tang, “the kinematics of eight-speed planetary gear hubs for bicycles,” journal of applied mechanics and materials, vol. 232, pp. 955-960, 2012. [12] l. c. hsieh, h. c. tang, t. h. chen, and j. h. gao, “the kinematic design of 2k type planetary gear reducers with high reduction ratio” journal of applied mechanics and materials, vol. 421, pp. 40-45, 2013. [13] y. c. wu and l. a. chen, “design of an 8-speed internal gear hub with a rotary control mechanism for bicycles,” tehnicki vjesnik, vol. 22, no. 4, pp. 865-871, 2015. [14] g. white, “derivation of high efficiency two-stage epicyclic gears,” mechanism and machine theory, vol. 38, no. 2, pp. 149-159, 2003. [15] c. chen and j. angeles, “virtual power flow and mechanical gear mesh power losses of epicyclic gear trains,” journal of mechanical design, vol. 129, pp. 107-113, 2007. [16] l. c. hsieh and t. h. chen, “on the meshing efficiency of 3k-type planetary simple gear reducer,” journal of advanced science letters, vol. 12, pp. 34-39, 2012. [17] l. c. hsieh and t. h. chen, “the kinematic and efficiency analysis of planetary gear trains with single train circuit,” information-an international interdisciplinary journal, vol. 16, pp. 7025-7032, 2013. [18] l. c. hsieh and h. c. tang, “on the meshing efficiency of 2k-2h type planetary gear,” journal of advances in mechanical engineering, vol. 5, 2013. [19] y. c. wu, l. a. chen and c. w, chang, “prediction of efficiency of an existing 14-speed bicycle internal drive hub,” journal of mechanical engineering, vol. 2, no. 8, pp. 266-273, 2014. [20] t. h. chen, “the equation derivation and analysis of meshing efficiency for straight spur gear pair,” transactions of the canadian society for mechanical engineering, vol. 39, no. 3, pp. 467-477, 2015. [21] l. c. hsieh, t. h. chen, and h. c. tang, “the design and meshing efficiency analysis of helical spur gear reducer with single gear pair for electric scooter,” transactions of the canadian society for mechanical engineering, vol. 39, no. 3, pp. 455-465, 2015. [22] l. c. hsieh and t. h. chen, “the engineering design and transmission efficiency verification of helical spur gear reducer with single gear pair,” transactions of the canadian society for mechanical engineering, vol. 40, no. 5, pp. 981-993, 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/).  proceedings of engineering and technology innovation, vol. 9, 2018, pp. 38 43 on a touch-activated wearable device with automated location sending capability conrad a. populi * , mary emmanuelle huela, john paulo ilagan, cezarina noble department of electronic engineering, technological institute of the philippines, aurora blvd., quezon city, philippines received 02 march 2018; received in revised form 08 may 2018; accepted 28 may 2018 abstract this study proposes a device that enables a person in distress to remotely seek for help. it is a wearable device consisting of a microcontroller and a gps-gsm module that has the capability to determine the closest police stations based on the current location of the wearer. when being triggered, it sends the current coordinates of the wearer via sms messages to the corresponding nearest police station, and a pre-defined contact number. afterwards, the device sounds off an alarm to fend off any threat, or to serve as an alert of your location. keywords: gps, gsm, sms, wearable, coordinates, closest police station, alarm, program 1. introduction according to data released by the philippine star on march 6, 2016, every 53 minutes, a woman or a child is raped. [1] furthermore, this claim has been supported by the philippine national police (pnp) on their submitted annual national report for fy 2015 which implicates that rape is the 5th most prevalent crime in the philippines. moreover, quezon city, davao city and manila are the top three cities with the highest number of rape cases nationwide. the project aims to design and develop a prototype of a device that will enable women to seek for help in case of distress or apparent threat. the goals of the project shall be accomplished through analyzing existing similar devices in the market, researching wireless protocols whether are available and applicable to the project, establishing the block diagram of the whole system, and finally developing the design and prototype of the device. similar existing devices are already in the market abroad, and there are others that are being developed. these are wearable devices in the form of pendants, rings, bracelets, undergarments, and belts. some just serve as an alarm while some have complex functions. “suraksha” is designed to send the exact location of the victim to the police, and it is actuated by voice, switch and shock. [2] “femme”, a bluetooth device, is triggered by pressing the emergency button. depending on how the button is pressed, the device can send distress messages and the location of the person to the police, can record the audio of the incident and can initiate a call. [3] in the paper entitled, “smart girls security system”, a device was designed in the form of a belt. this device sends the location of the person to the police control room emits an alarm and generates an electric shock to injure the attacker. [4] the paper entitled “smart security solution for women based on internet of things (iot)” presented a wearable band that is connected to the internet through a smartphone. the device is preprogrammed to record and assess the wearer’s behavior and reactions, and to send the results to the smartphone. if it receives an emergency signal, the software application sends help requests to the nearest police station and people in the near radius of the application. [5] the difference with the proposed design and these devices is that the proposed design is a standalone device that has the capability to know which police stations are the closest. then, it can send a message to a registered contact number and the * corresponding author. e-mail address: cpopuli.ece@tip.edu.ph proceedings of engineering and technology innovation, vol. 9, 2018, pp. 38 43 copyright © taeti 39 closest police station informing the coordinates of the user and then emit an alarm. this is what the automated location sending capability of the design means. 2. block diagram fig. 1 shows the proposed system-architecture of the study. the 3.7v battery serves as the power supply to the microcontroller unit (mcu). when the device is touched, the sensor-based switch turns to on state. this event triggers the alarm system. also, the gps acquires the coordinates (longitude and latitude) of the position of the wearer from the gps satellite and transfers it to the mcu. then, the mcu processes this information and locate the nearest mobile number as to where the sms text will be sent to using gsm. the mobile number of the possible recipients is determined using the great-circle formula between two points algorithm, which is encoded as a program in the microcontroller. fig. 1 system architecture 3. system flowchart the system flowchart presented in fig. 2 shows how the device operates. first, the device is activated by a sensor-based switch, particularly a capacitive touch sensor. if the switch is activated for at least 2 seconds, the alarm trigger causes a loud noise to be emitted. the 2-second allowance is configured to avoid false triggering. afterwards, the device acquires the wearer’s position from a gps satellite. given the present location of the user, the microcontroller calculates the nearest predetermined location of where the device will send an sms to. the content of the sms is the acquired coordinates of the wearer. fig. 2 system flowchart of the design proceedings of engineering and technology innovation, vol. 9, 2018, pp. 38 43 copyright © taeti 40 4. design model the 3d model of the prototype, as illustrated in fig. 3, consists of the mechanical dimensions and the shape of the device. in this project, the point is on the core wherein the device is encased. fig. 3 design of the prototype the dimension of the proposed prototype will be approximately 4.6cm x 4.6cm x 2cm while the length is especially designed for the average women’s size of wrist, which is typically around 16.5cm to 20.3cm; according to the claim by klingenberg. in terms of the straps, the said prototype used nylon kernmantle rope as its primary material because of its comfortable-to-wear quality and it’s good for the hard grip of its main part. besides, with this kind of non-metal bracelet, the end-users can easily adjust it depending on their most comfortable fit. moreover, the core is made of acrylic, ideal for physical protection of the internal part of the prototype against the possible external damage. fig. 4 3d model side view fig. 5 3d model top view fig. 6 3d model back view fig. 7 3d model front view fig. 4 shows the side view of the device. the height of the device is approximately 20mm. however, there are no switches or buttons on the sides of the device. fig. 5 shows the top view of the device. at the center is a 10x10mm square allotment for proceedings of engineering and technology innovation, vol. 9, 2018, pp. 38 43 copyright © taeti 41 the electrode used in the touch sensor. at the top of the square is a set of holes in a circular pattern for the vent of the alarm. fig. 6 shows the back view of the device. the width of the device seen here is approximately 40mm not including the strap attaching portion. fig. 7 shows the front view of the core of the device. the dimensions are the same as above but the only difference is that there is an allotment for the micro usb port. the purpose of the micro usb is mainly for the charging of the device since gadgets nowadays generally use the micro-b usb for general purposes. 5. technical specification fig. 8 circuit digram of the prototype fig. 8 shows the circuit diagram of all the components in the prototype. the atmega168 in the middle represents the microcontroller uhex and its pins are connected to the other main components specifically the gps/gsm module fona 808, the capacitive touch controller at42qt1010, the buzzer cmt-4023s, and the 3.7v battery. the connection of the fona 808 starts with its vcc and the rx and tx pins. the buzzer is connected to a general purpose input output (gpio) pin in the microcontroller. another connection thru the general purpose input output pin is with the output of the capacitive touch sensor. 6. discussion table 1 selected police station in marikina city point station location 1 pcp 1 kalumpang 14.623368, 121.090450 2 pcp 3 sta. elena 14.633126, 121.095471 3 pcp 8 marikina heights 14.649818, 121.117315 4 pcp 9 parang 14.659243, 121.111994 5 pcp 6 concepcion uno 14.653430, 121.102467 proceedings of engineering and technology innovation, vol. 9, 2018, pp. 38 43 copyright © taeti 42 to test the device in general, the proponents selected five police stations in marikina city listed in table 1. the coordinates of the locations of these police stations were gathered in terms of latitude and longitude. the testing was done in an open area because one of the limitations of gps is that it is unable to work effectively indoors. a serial monitor was included in the program in order to validate if the device computes for the nearest location and sends it to the default receiver and nearest police station. the proponents provided a comparison of the measured distances with the measured distances from google earth. table 2 shows the actual location of the user and the computed location of the police stations, while fig. 9 shows the information flow diagram of the sms. table 2 test data for the first location point station location location of the user measured distance from the device distance from google earth percent difference 1 pcp 1 kalumpang 14.623368 121.090450 14.6255846023, 121.0621185302 3057.81762 m 3.06km 0.07132% 2 pcp 3 sta. elena 14.633126 121.095471 3685.41650 m 3.69 km 0.12421% 3 pcp 8 marikina heights 14.649818 121.117315 6521.36669 m 6.52 km 0.02096% 4 pcp 9 parang 14.659243 121.111994 6541.88232 m 6.54 km 0.02878% 5 pcp 6 concepcion uno 14.653430 121.102467 5332.29345 m 5.33 km 0.04303% fig. 9 information flow diagram 7. conclusions the proponents were able to design and prototype a sensor-based switch that will trigger the device to turn on. through a profound understanding with the language of c++, the proponents were also able to program the microcontroller so as to calculate the nearest predetermined location of where the device will send an sms to. the sms contains the exact location of proceedings of engineering and technology innovation, vol. 9, 2018, pp. 38 43 copyright © taeti 43 the user, which was determined through the global positioning system in terms of latitude and longitude. once the proponents reach the held objectives, the testing procedures followed and applied in the area of marikina city. 8. recommendations to the future designers, this design uses just one module for both the gsm/gps. since other pins were left unused in this project, it would be better if the future designers would make a module that would incorporate only the pins they needed. also, it would be best to use different approach in debugging or testing the prototype in order to get the desired results or outcomes. this design is limited by the module it uses. the gps and gsm module employs an antenna. to further boost the signal being captured by the said module, it is recommended to design a new antenna instead of using the antenna bought with the module. in order to not stray from the objectives plus the condition of being time-bound, the proponents decided to not include designing an antenna in this design project. references [1] r. villanueva, “one woman or child is raped every 53 minutes,” the philippine star, march 5, 2016. [2] n.bhardwaj and n. aggarwal, “design and development of ‘suraksha’ a women safety device,” international journal of information & computation technology, vol. 4, no. 8, pp. 787-792, february 2014. [3] d. monisha, m. monisha, g. pavithra, and r. subhashini, “women safety device and application – femme,” indian journal of science and technology, vol. 9, no. 10, pp. 1-6, march 2016. [4] b. chougula, a. naik, m. monu, p. patil, and p. das, “smart girls security system,” international journal of application or innovation in engineering & management, pp. 281-284, april 2014. [5] g. c. harikiran, k. menasinkai, and s. shirol, “smart security solution for women based on internet of things (iot),” international conference on electrical, electronics, and optimization techniques, ieee press, march 2016, pp. 3551-3554.  proceedings of engineering and technology innovation, vol. 10, 2018, pp. 19 28 investigation of peel resistance of adhesives materials: a review yuvraj j. awalekar ,* , atul s. takalkar, sachin s. shinde department of mechanical engineering, shivaji university, k.i.t’s college of engineering, kolhapur, maharashtra, india received 08 january 2018; received in revised form 09 may 2018; accepted 10 september 2018 abstract different adhesives are used for joining the different types of materials. the different types of peel tests available for investigating the adhesive strength are 90º, 135º, 180º and t-peel test. this paper reviews the different peel tests available as well influence of other factors for determining of peel resistance between two materials. the investigation of adhesive strength helps to select a proper adhesive material for joining of two different materials. the paper briefs the effect of peel angle, different types of peel tests such as 90º and t-peel test, temperature, moisture, thickness, the rate of change of peeling rate on peel strength. keywords: adhesive bonding, de-lamination, interface adhesion, peeling test 1. introduction adhesive bonding is used on a large scale in different industries. with the invention of new adhesive materials, it can be a challenge to figure out a way to join two different materials while retaining their individual properties. peeling force is an important property to characterize adhesion of materials. this force is defined as the force required to remove the adhesive film from a substrate. the peel strength is the measure of the average force to part two bonded materials like tape, labels, textile or plastic films. the strength is calculated during a peel test at a constant speed rate divided by the average force (the peel force) required during the test the unit width of the bonded samples. depending on materials, norms, products, the tests can be done with different angles 90° and 180° are commonly used. the adhesive bonding is used where extremely small parts need to be bonded securely within a unit. the use of adhesive bonding is important in manufacturing products which are rugged, light-weight and transparent. in such case, the manufacturer should know about the adhesive strength of the adhesive. different types of peel tests can be used to determine the adhesive strength between two bonded materials. the applications of peeling and its role in adhesion are widespread in both engineering and biology. a specific problem that has captivated a diverse group of scientists, particularly technicians, in the area of peeling and adhesion for the last decade is the enhancement of the adhesive properties of materials through studying analogous behavior in nature. the investigations focused on naturally occurring reversible adhesion in biological systems can be applied to the development of engineered reversible adhesive systems, which have a wide range of engineering applications including biomedical devices, microelectronic storage and packaging, and surgical robotics. at the core of the problem, how a natural reversible adhesion process works in both micro and macro scale mechanics should be known. the failure of adhesive joints is commonly said to be either adhesive failure or cohesive failure. the combination of these two is also possible. an adhesive joint failure occurs between the interface of adhesives and adherent. the cohesive failure is a failure that occurs in the largest part of the adhesive layer and usually desired mode of failure. if the adhesives and adherents have the different thermal-mechanical properties, it is observed to fail without additional load. the surface of the structure also affects the location failure, either smooth surface or a rough surface [1]. * corresponding author. e-mail address: yuvrajawalekar@gmail.com tel.: +91-9049096312 proceedings of engineering and technology innovation, vol. 10, 2018, pp. 19 28 copyright © taeti 20 the mechanics of the peel test can be examined through experiments, finite element simulations, and theoretical approach with the aim of developing, governing relations to describe the role of fracture in the peel test for elastic adhesive tapes. an inverse formulation was developed to extract a cohesive zone law from a set of experimental peel tests using a theoretical framework based on non-linear beam theory. 2. literature in the most of the literature, the peel test is preferred to determine the adhesive properties by performing theoretical calculation and experiment tests. the adhesion properties are evaluated with help of peel resistances which implicates the peel strength. the forces acting per unit width of the laminates measures the required energy for the peeling of the laminates. the peel test is also used to assess the knowledge about the effects of parameters like peel angle, a number of ingrained adhesion effects in between the laminates, thickness of the laminates, and the rate of the test. the temperature also plays a major role while performing the analysis. following table shows authors and their respective work is done with materials used during the peel test. table 1 authors, material used and work done name of author materials work done ulrich eitner, li c. rendler [2] ribbons. peel test on ribbons s. giannis, r.d. adams, l.j. clark, m.a. taylor [3] aluminium, titanium. peel test on aircraft fuel tank sealants jinhyeok jang, minchang sung, sungjin han, woong-ryeol yu [4] steel, nylon the delamination of steel-polymer composites idris k. mohammeda, anthony j. kinlocha, maria n. charalambidesa [5] polyethylene peeling behavior of soft adhesives z. l. peng, s.h. chen, a. k. soh [6] bio-mimetic nano-films peeling behavior of a bio-inspired nano-film stephan marzi, anders biel, olaf hesebeck [7] steel peel test in a rotary impact device. f. nihal tuzun, m. safak tunalıoglu [8] steel, composites investigated the effect of five different fillers by using t-peel. michael nase, mirko rennert, konstantin naumenko, victor a. eremeyev [9] self-adhesive polymers self-adhesive polymeric films. g hinopoulos and w r broughton [10] cr1 mild steel, 6al-4v titanium, 5251 aluminium fea study carried out on a t-peel joint. h. hadavinia, l. kawashita, a. j. kinloch, d.r. moore, j.g. williams [11] hyperelastic thin films adhesive fracture energy. gregory polyzois, panagiotis lagouvardos, spiroszinelis, maryfrangou [12] titanium (cpti) investigation of bond strength weizeng, weixing sun, nicola bowler, simon laflammea [13] aluminum, steel, concrete, and fiberglass peel resistance of adhesive joints with elastomer–carbon black. yasuhiro tanimoto, hiroyuki saeki, suguru kimoto, tsuyoshi nishiwaki c, norihiro nishiyama [14] gc reline soft, gc reline extra soft investigation of adhesive properties a. nick, b. nick, f. j. wortmann [15] fibre-reinforced epoxy composite peel test between temperature -55°c to 80°c sofia teixeira de freitas, jos sinke [16] fiber metal laminate (fml), carbon fiber reinforced polymer. peel test between temperature -55°c, room temperature (+22°c) and 100°c g hinopoulos and w r broughton [10] cr1 mild steel, 6al-4v titanium, 5251 aluminum peel test for substrate thickness 1.5 to 5 mm zhilong peng, cong wang, lei chen, shaohua chen [17] glass, viscoelastic polyvinylchloride peel test for film thickness 1.5 to 5 mm 0.13, 0.26, 0.39 2.1. peel angle peel angle influences the peel force in the peeling test. the literature shows that as the peeling angle increases the peeling force decreases to some angles. ulrich eitner and li c. rendler [2] carried out a peel test for soldered or glued ribbons on solar cell metallizations. it was found that higher forces were required for decreasing peeling angles less than 90° and weakest forces required for 135°. the work also includes the adhesive fracture patterns. it was found that with the change of peeling angles proceedings of engineering and technology innovation, vol. 10, 2018, pp. 19 28 copyright © taeti 21 from 90° to 180° helps to obtain adhesive or cohesive fracture patterns instead use of silicon cell fracture. the relation between peel force and adhesive fracture energy was proposed. the fig. 1 shows the relation between peel forces and adhesive fracture energies for 180° peeling angle. from results, it is observed that for the given force the adhesive fracture energy is maximum for the soft ribbon. s. giannis, r.d. adams, l.j. clark, m.a. taylor [3] carried out peel test on aircraft fuel tank sealants and observed results for different sealants are shown in fig. 2. the obtained results show that for the same thickness and the same sealant the peel resistance increases as the peel angle was increased from 135° to 180°. also, jinhyeok jang, minchang sung, sungjin han, woong-ryeol yu [4] carried out peel tests of steel polymer composite (nylon) at various angles and obtained results are shown in fig. 3. fig. 1 the relation between peel forces and adhesive fracture energies for 180° peeling angle [2] fig. 2 the relation between peel resistance and the peel angle [3] (a) 0–90 degrees (b) 90–180 degrees fig. 3 comparison of experimental and predicted peel strengths for different peel angles [4] fig. 4 experimental (green marked) and numerical peel forces at different angles [5] proceedings of engineering and technology innovation, vol. 10, 2018, pp. 19 28 copyright © taeti 22 the results were obtained for different materials which shows that as the peel angle increases the peel force decreases. the materials used for peel tests were eg steel–nylon, pl steel–nylon, gi steel–nylon, ga steel–nylon, posmac steel–nylon from which combinations of eg steel-nylon shows lesser deviation in experimental and predicted results. also, the peel test was carried out by idris k. mohammeda, anthony j. kinlocha, maria n. charalambidesa [5] for pharmaceutical drug patches which consist of a polyester backing membrane supporting an acrylic pressure-sensitive adhesive (with and without anti-fungal drug) adhered to a polyethylene substrate. the results obtained through experimentation are shown in fig. 4 with respect to peel angle. a numerical peeling model of nano-films with a finite length in adhesive contact with a rigid substrate is investigated by z. l. peng, s.h. chen, a. k. soh [6]. it was found that the given peeling angle, peeling-off force remains constant if the length of the nano-film is larger than an effective adhesive length. the literature summary for peel angle is shown in table 1. specialization is to assign specific types of members and joints to every available generalized chain subject to certain design requirements to obtain the specialized chains. design requirements are determined based on the concluded topological structure of the original designs. 2.2. t-peel test fig. 5 schematic diagram of t-peel test [7] it is a peel test in which both materials are pulled simultaneously as shown in fig. 5. the one end of the joint is tightly held in the jaws and the other end is pulled in the vertical direction by holding in another jaw. by measuring the peel forces the peel resistance was determined for the adhesive. such type of peel test is used when both adherents are flexible. stephan marzi, anders biel, and olaf hesebeck [7] also performed the t-peel test on a rotary impact device where the material is not supported but clamped stiffly. the deformation of the specimen and its clamping is analyzed by three-dimensional digital image correlation system. fig. 6 schematic of the adhesive fillet geometries for the t-peel joint [10] also, f. nihal tuzun, m. safak tunalıoglu [8] investigated the effect of five different fillers on the adhesion strength of the metal parts. michael nase, mirko rennert, konstantin naumenko, and victor a. eremeyev [9] used t type peel test for self-adhesive polymeric films. whereas g hinopoulos and w r broughton [10] used a t type peel test to investigate the effect of adherend properties, geometric parameters, and environmental conditioning on joint performance. the finite element analysis was carried out for steel joints with the adhesive fillet occupying either half or all of the space between the steel substrates. the schematic diagram of the adhesive fillet geometries for the t-peel joint is shown in fig. 6. it was observed that the joint with the 50% adhesive fillet deformed substantially more than the joint with the 100% fillet under the same tensile load. the von mises equivalent stress at the end of the overlap for the specimen with 100% fillet was reduced by 200% compared with the 50% fillet specimen [10]. also, h. hadavinia, l. kawashita, a. j. kinloch, d.r. moore, and j.g. williams [11] determined the adhesive fracture energy with the help of a peel test. the literature summary for the t-peel test is shown in table 1. proceedings of engineering and technology innovation, vol. 10, 2018, pp. 19 28 copyright © taeti 23 2.3. 90º peel test in 90º peel test, bottom substrate is made rigid and the upper substrate is exposed so as to obtain a smooth pull of peel arm as shown in the fig. 7. the bottom substrate is rigid by means of grips and the peel arm of the upper substrate is pulled in order to measure the peel force. this test is preferred when the upper adherend to be pulled is having less thickness than that of the adherend which at the bottom. fig. 7 schematic representation of the experimental set up of 90 degree peel test [12] fig. 8 survival graph of the four pdms bonded to acrylic substrate with primer [12] polyzois, panagiotis lagouvardos, spiroszinelis, and maryfrangou [12] investigated the bond strength of four silicone (pdms) pro gregory sthetic elastomers such as vst-50, a-588, techsil 25 and silbione used for the construction of official prostheses to cpti and acrylic resin as control through a primer and bonding enhancer application. the peel bond test at 90° was conducted as per iso 813:97 and the obtained results were analyzed and interpreted using anova and weibull regression analysis as shown in fig. 8. from the analysis, it is observed that only one of the materials techsil 25 has a high probability of surviving a stress. also, h. hadavinia, l. kawashita, a. j. kinloch, d.r. moore, and j.g. williams [11] determined the adhesive fracture energy using a single arm 90-degree peel test as shown in fig. 9. fig. 9 the single-arm 90° peel test [11] for the single-arm peel joints, aluminum-alloy sheets of 180 mm x 20 mm x 10 mm served as the rigid base plates. the dimensions of the peel arms were measured prior to joint preparation. the single-arm peel tests were carried out by maintaining the peel angle of 90º. all peel tests were carried out at the same temperature, humidity (21°c, 55% rh), and crack growth rate (5 mm/min). also, weizeng, weixing sun, nicola bowler, and simon laflammea [13] carried out peel test for four different adhesives which were j-b weld, 3m scotch-weld dp125gy, loctite pl premium (3x) construction adhesive, and henkel hysol ea 9394 and evaluated the results based on adhesive costs. the characteristics of adhesive properties proceedings of engineering and technology innovation, vol. 10, 2018, pp. 19 28 copyright © taeti 24 between a denture base and resilient denture liner were investigated by yasuhiro tanimoto, hiroyuki saeki, suguru kimoto, tsuyoshi nishiwaki c, and norihiro nishiyama [14] using 90° peel test. the literature summary for 90º peel test is shown in table 1. 2.4. temperature the temperature parameter has a major impact on adhesive strength. with the change in temperature changes the properties of the adhesive. a. nick, b. nick, and f.-j. wortmann [15] studied the effect of temperature on four different types of adhesives as shown in table 2. the peel tests were performed by keeping the same relative humidity for all samples and temperature ranges between -55°c to 80°c. the observed results are shown in fig. 10. fig. 10 relation between maximum force and temperature [15] table 2 different types of adhesives used for peel test type curing system specification a cold curing epoxy/polyamine-system (scotch-weld dp 190) b hot curing 3-component epoxy-system (araldit ly 556, hy 917, dy 070) c cold curing polyurethane/isocyanate-system (scotch-weld 3532 b/a) d cold curing epoxy/polyamine-system, different in composition than type a (scotch-weld 9323 b/a) fig. 11 the relation between max load and temperature of adhesives af163 and ea9696 [16] the obtained peel test results show significant differences in peel strength of different adhesives. the uniform dependency of the peel strength of the temperature is observed. for the cold curing systems, the peel strength increases with increasing temperature up to the glass transition temperature of the adhesives. at the higher temperatures, the peel strength decreases. hence, a maximum of peel strength is observed near room temperature for adhesive type a and c. the adhesive d shows a nearly linear increase in peel strength up to high temperatures at a relatively high level. the consequent type d can be used for applications over a broad temperature range. whereas sofia teixeira de freitas and jos sinke [16] studied the effect of temperature on two different adhesives. the tests were carried out at the three different temperatures -55°c, room temperature (+22°c), and 100°c under quasi-static loading. the obtained results are shown in fig. 11 where it is observed that for both adhesives as the temperature was increased the peel load was also increasing as a consequence of enhancement in adhesive strength. the literature summary at different temperatures for the peel test is shown in table 1. proceedings of engineering and technology innovation, vol. 10, 2018, pp. 19 28 copyright © taeti 25 every feasible specialized chain is particularized into its corresponding mechanical device in a skeleton drawing. therefore, the last step is to identify all non-existing designs from the atlas of designs as the new designs. 2.5. moisture and thickness: the moisture content in the adhesive as well as around the adhesive affects the adhesive strength. also, the adhesive film thickness has a major impact on peel resistance. g hinopoulos and w r broughton [10] carried out the effect of environmental conditions on the performance of mild steel joints. this was investigated using a sequentially coupled mechanical-diffusion finite element model, which incorporated continuously varying adhesive material properties. table 3 summary of parametric study [10] property moisture content adherend stiffness adhesive fillet adherend thickness joint stiffness    equivalent stress     peel stress     axial stress    shear stress   plastic strain     the results revealed that the distributions of stresses become more uniform along the adhesive layer when the adhesive contains an increased amount of moisture. the peel stresses at the edge of the adhesive fillet were observed to decrease with increasing moisture content. these changes were due to moisture plasticising the adhesive which results in lower adhesive stiffness. the von mises equivalent stress distribution along the center line of the adhesive layer is shown in fig. 12 and a summary of parametric study are shown in table 3. fig. 12 mises equivalent stress distribution along the center line of the adhesive layer [10] fig. 13 the influence of film thickness on the peel-off force with a determined peeling rate ( peeling: angles 60° and 90°) [17] proceedings of engineering and technology innovation, vol. 10, 2018, pp. 19 28 copyright © taeti 26 the study implies that the stress distributions are sensitive to adherend material properties, adherend thickness, and flange radius with the effect of changes in flange radius being less pronounced. also, zhilong peng, cong wang, lei chen, and shaohua chen [17] carried out a peel test and investigated the effect of thickness on the peel-off force. the results obtained from the experimental test are shown in fig. 13 where it indicated that, as the adhesive film thickness was increased the peel-off force increased. the tests were carried out for two different angles such as 60º and 90º, which shows that as the peel angle is increased from 60º to 90º peel-off force decreases. the summary of peel testing of materials with the different thickness of adherend and film is shown in table 1. 2.6. fracture mechanism during the peel test, the failure may be cohesive or adhesive. when the adhesive bond is stronger than the adherend used then cohesive failure occurs. sometimes the combination of both cohesive and adhesive failure occurs at the interface. cohesive failure leads to fracture of the adherend. sofia teixeira de freitas and jos sinke [18] has carried out peel test on composite and aluminium adherend. during the peel test failure occurs at the interface and pictures of the fracture surfaces of two adhesives scanned by the electron microscope are shown below. fig. 14 peel test carried at room temperature (rt) and high temperature (ht) [18] peel tests carried out were frpt (floating roller peel test) and cpt (composite peel test) at room temperature (rt). af 163-2 is the adhesive used during the test. the pictures above shows ‘a’ as an adhesive failure, ‘c’ as a cohesive failure and ‘il’ as an intra-laminar failure. intra-laminar failure of the composite also indicates good adhesion, since the failure is cohesive within the composite adherend and not at the interface. furthermore, this type of failure also indicates that the intra-laminar strength of the composite adherend is lower than the debonding strength of the adhesive. the above peel tests also concluded that in most cases of good adhesion. increasing the temperature favors cohesive failure of the adhesive in detriment of intra-laminar failure of the composite. the fracture mechanism of a cohesive failure is independent of the peeling-off adherend (composite or aluminium). 2.7. effect of peeling rate on peeling strength figure 15 shows the relationship between the peeling rates and the peeling strength. in figure 15, the unhardened test pieces with a peeling angle of 135° is plotted on the black line with dot marks, while the unhardened test pieces with a peeling angle of 80° is plotted in the red line with triangle marks and the hardened test pieces with a peeling angle of 135° is plotted in blue line with diamond marks. proceedings of engineering and technology innovation, vol. 10, 2018, pp. 19 28 copyright © taeti 27 increasing peeling strength leads to increasing peeling rate. peeling rates are positively correlated to adhesive energy, as the adhesive energy required is larger when the cracking velocity increases. compared to the black curve and the red curve, it is found that increasing the peeling angle leads to decreasing peeling strength. comparison of the black and blue curves indicated that hardened test pieces gave rise to higher peeling strength values due to greater elastic modulus values. fig. 15 relationship between peeling rate and peeling strength. (peeling angle: 135°or 80°) [19] 3. discussions and conclusions the t-peel test can be performed for materials like mild steel, titanium, aluminum and composites for various types of adhesive material used for joining. the selected specimen should be flexible in order to obtain a uniform bending. from the literature, it is found that the selection of peel test is based on types of adhesives, types of materials to be joints, and the flexibility of materials. the 90º peel test is used when the upper adherend is flexible and adherend at the bottom is rigid. also, it can be used when both adherends are flexible. the adherend to be pulled upward must be flexible so that uniform vertical force can be applied to get an accurate reading. the 90º peel test can be used for the adhesives such as epoxy, al2o3. in case of 135º peel test, the same types of adhesives and materials can be used. for 180º peel tests, the material to be peeled must more flexible so that there is a uniform bending while pulling action is performed. peel test can be used successfully to assess the interface adhesion in aged and non-aged conditions and can be used as a fast, easy, and reliable test to study the long-term durability of composite-metal bonded joints. the peel resistance increases as the peel angle changes from 135 0 to 180 0 and this tendency seem to be independent of the adhesive layer thickness. there is a degradation in chemical bonding because of moisture content in the adhesive. adhesive bond becomes stronger as the adhesive layer thickness is increased. during the experimental determination of adhesion strength through a peel test, the deformation energy of the peeled foil, plus frictional losses may constitute a significant portion of the total peel energy. when the true adhesion strength is desired, the energy dissipated in bending of the foil has to be found, whether experimentally or analytically, and deducted from the total peel energy to get the net adhesion strength. while carrying out peel test cracks are formed at the adhesive film and substrate interface. such cracks are responsible for inducing the film by the formation of crack tip at the interface. when the film is thin then transverse shear is dominating for a crack at the interface. for thick film, the crack will induce the three films only if bending being entirely elastic [20]. references [1] c. r. dundeti and p. r. punnam, “a finite element analysis of crack propagation in interface of aluminium foil ldpe laminate during fixed arm peel test,” master’s degree thesis, dept. of mechanical engineering, blekinge institute of technology, karlskrona, 2017. [2] u. eitner and l. c. rendler, “the impact of ribbon properties on measured peel forces,” energy procedia, vol. 92, pp. 500 504, 2016 proceedings of engineering and technology innovation, vol. 10, 2018, pp. 19 28 copyright © taeti 28 [3] s. giannis, r.d. adams, l.j. clark, and m.a. taylor, “the use of a modified peel specimen to assess the peel resistance of aircraft fuel tank sealants,” international journal of adhesion & adhesives, vol. 28, pp. 158-175, september 2007. [4] j. jang, m. sung, s. han, and w.r. yu, “prediction of delamination of steel-polymer composites using cohesive zone model and peeling tests,” composite structures, vol. 160, pp. 118-127, october 2016. [5] i. k. mohammed, a. j. kinloch, and m. n. charalambides, “modelling the peeling behavior of soft adhesives” procedia structural integrity, vol. 2, pp. 326-333, june 2016. [6] z. l. peng, s. h. chen, and a.k. soh, “peeling behavior of a bio-inspired nano-film on a substrate,” international journal of solids and structures, vol. 47, pp. 1952-1960, april 2010. [7] s. marzi, a.biel, and o. hesebeck, “3d optical displacement measurements on dynamically loaded adhesively bonded t-peel specimens,” international journal of adhesion & adhesives, vol. 56, pp. 41-45, july 2014. [8] f. n. tuzun and m. s. t. lu, “the effect of finely-divided fillers on the adhesion strengths of epoxy-based adhesives,” composite structures, vol. 121, pp. 296-303, november 2014. [9] m. nase, m. rennert, k. naumenko, and v. a. eremeyev, “identifying traction-separation behavior of self-adhesive polymeric films from in situ digital images under t-peeling,” journal of the mechanics and physics of solids, 2 march 2016. [10] g. hinopoulos and w. r broughton, “evaluation of the t-peel joint using finite element analysis,” the materials centre, national physical laboratory teddington, middlesex, technical report, tw11 0lw, uk. [11] h. hadavinia, l. kawashita, a. j. kinloch, d.r. moore, and j.g. williams, “a numerical analysis of the elastic-plastic peel test,” engineering fracture mechanics, vol. 73, pp. 2324-2335, june 2006. [12] g. polyzois, p. lagouvardos, s. zinelis, and m. frangou, “peel testing of bond between pdms prosthetic elastomers and cpti: a weibull analysis approach,” international journal of adhesion & adhesives, vol. 30, pp. 500-504, may 2010. [13] w. zeng, w. sun, n. bowler, and s. laflamme, “peel resistance of adhesive joints with elastomer–carbon black composite as surface sensing membranes,” international journal of adhesion & adhesives, vol. 58, pp. 28-33, january 2015. [14] y. tanimoto, h. saeki, s. kimoto, t. nishiwaki, and n. nishiyama, “evaluation of adhesive properties of three resilient denture liners by the modified peel test method,” acta biomaterialia, vol. 5, pp.764-769, september 2008. [15] a. nick, b. nick, and f.-j. wortmann, “peel testing of adhesively bonded joints of carbon fibre,” deutsches wollforschungsinstitut an der rwth aachen e. v., veltmanplatz 8,technical report, 52062 aachen. [16] s. t. de freitas and jos sinke, “failure analysis of adhesively-bonded metal-skin-to-composite-stiffener: effect of temperature and cyclic loading,” composite structures, vol. 166, pp. 27-37, january 2017. [17] z. peng, c. wang, l. chen, and s. chen, “peeling behavior of a viscoelastic thin-film on a rigid substrate,” international journal of solids and structures, vol. 51, pp. 4596-4603, october 2014. [18] s. t. de freitas and j. sinke, “test method to assess interface adhesion in composite bonding,” applied adhesion science, vol. 3, pp. 1-13, march 2015. [19] y. wang, “the effect of peeling rate and peeling angle on the peeling strength,” master’s degree thesis, university of akron, may 2014. [20] m. d. thouless and q. d. yang, “a parametric study of the peel test,” international jounal of adhesion & adhesives, vol,28,pp.176-184, september 2007.  proceedings of engineering and technology innovation, vol. 9, 2018, pp. 17 24 a study on unmanned surface vehicle combined with remotely operated vehicle system dong-wook jung 1 , seung-min hong 2 , jae-heon lee 1 , hyun-joon cho 1 , hyeung-sik choi 1,* , mai-the vu 1 1 department of mechanical engineering, korea marigime and oceon university, busan, korea 2 department marine equipment engineering, korea marigime and oceon university, busan, korea received 02 march 2018; received in revised form 04 may 2018; accepted 02 july 2018 abstract in this paper, we proposed a new hybrid system that combines an usv and the rov. it is designed to overcome the cumulative navigation error and the battery problem for long time operation. the usv is connected to the underwater robot with a tether cable that enables gps data transmission and stable power supply. in addition, by installing a winch system on the usv, it is possible to move and to retrieve the rov for long distance travel and for depth control. the rov equipped with underwater sensors which can acquire underwater image capture and undersea topography information and can transmit it to the land through the usv in real time was designed. line-of-sight navigation and dynamic positioning algorithm were studied for the usv to perform autonomous navigation and keeping-position while rov is doing underwater operations. good performance of the usv was verified by simulation and real sea experiment using the developed hybrid usv and rov. keywords: unmanned surface vehicle, usv, remotely operated vehicle, rov, hybrid system, dynamic positioning 1. introduction recently, exploration and development of marine resources have become an answer to the problem of energy shortages. due to the nature of the marine environments, it is important to operate a underwater robot to minimize loss of life during the work. also, the recent discovery of marine resources around the world is becoming a major issue when it comes to using limited resources in the face of energy shortages. the sea has an infinite amount of resources, but currently only some parts of the ocean are utilized. for exploiting underwater resources, underwater robot such as the autonomous underwater vehicle (auv) is very useful and essential. in this reason, auv platforms have been developed from abroad, such as the isimi and remus series. [1, 2] the auv has features of getting underwater terrain information and underwater object search. for high-speed / precision motion performance and long operation of the auv, expensive navigation sensors and high-capacity batteries are required. in addition, in order to retrieve the information acquired under the water, the auv should be recovered from the sea or should rise to the surface for sending the stored data through wi-fi or iridium communication. rov, on the other hand, is a submersible that is remotely controlled by cables connected to the operating system. due to the limitation of the cable length, it is not possible to perform a wide range of surveying, but it is suitable for underwater work using manipulators and cameras. in the case of the usv, studies have been conducted to extend mission roles by increasing the communication distance, and to expand the range of surveillance and reconnaissance missions by installing various sensors. [3] maritime boundary, * corresponding author. email address: hchoi@kmou.ac.kr proceedings of engineering and technology innovation, vol. 9, 2018, pp. 17 24 copyright © taeti 18 reconnaissance, and ocean and oceanographic topography are important tasks of unmanned aerial vehicles. in order to carry out these missions, large amounts of information obtained from cameras, radar and sonar are needed. in order to send such information to the control station in real time, a fast wireless communication speed is required. [4] nowadays, thanks to advanced researches, the availability of usv has increased in the area of autonomous navigation and remote control and its technology is becoming more advanced. [5] therefore, it is necessary to develop a new concept of the platform that can receive underwater information in real time with a long operating duration and high speed performance. the concept of a new platform can be confirmed in fig. 1. in this paper, we introduce the development of a new hybrid unmanned ship and underwater robot (hsuv) that combines auv, rov and usv technologies to improve existing underwater robot performance. the hybrid usv and rov consist of rov, launch / recovery winch, and usv that can move fast to a target location with the rov mounted. the usv performs position and dp control using gps and ahrs sensors and three thrusters, while the winch system controls the depth of the rov. fig. 1 concept of usv-rov system the rov is equipped with ahrs, us bl, and depth sensors to accurately follow usv's movements and to control depth. also equipped with an underwater search camera and a multi-beam sensor, underwater information can be transmitted to the ground in real time. by controlling the tether cable length of rov according to the water depth, scanning seabed exploration can be conducted with a constant resolution even if the depth changes. this concept is shown in fig. 2. fig. 2 concept of rov depth control the position of usv is determined based on gps, and the relative position between usv and rov is measured using usbl mounted on the rov. the exact position of the rov can be calculated by gps and usbl data. as a result, more accurate positioning results can be obtained at a lower cost than usv / auv combined platforms. [6] proceedings of engineering and technology innovation, vol. 9, 2018, pp. 17 24 copyright © taeti 19 2. usv system 2.1. structure of usv the usv platform studied in this paper uses a rubber boat as a hull considering shock and buoyancy. three propellers are installed to control yaw and surge motion. two propellers are mounted horizontally in the direction of the track and one propeller is mounted vertically to perform dynamic positioning functions. in addition, a docking station for rov docking was installed under the hull. the communication between usv and operator is made up of rf and lte, and the communication between usv and rov is made by serial communication through tether cable. the structure is shown in fig. 3. use gps to measure the position of usv, and use usbl to determine the relative position between rov and usb. in addition, real-time monitoring is possible on the ground with rf communication. the usv consists of the main control system for position, attitude and speed control, a winch system for controlling the depth of rov, and a communication system for data transmission / reception and rov control. fig. 3 system structure of hybrid system 2.2. control system of usv the control system consists of a control system for the usv operation and a winch control system for controlling the depth of the rov. this will be described in detail as follows. the usv's control system is shown in fig. 4. ahrs and gps sensor is used to control usv position and attitude and control the path by los method. rov is controlled by serial communication through tether cable. also, by installing a winch system, it is possible to control the depth of the rov and to launch and recover the rov. fig. 4 control system of usv proceedings of engineering and technology innovation, vol. 9, 2018, pp. 17 24 copyright © taeti 20 2.3. winch control system the winch system was installed in usv to control the depth of rov steadily. a tether cable is designed with stainless steel wire to withstand the tension caused by the weight and wave of rov. fig. 5 shows the winch control system mounted on the usv. fig. 5 winch system the control system for driving the winch system is shown in fig. 6. the main controller receives rov depth information and controls the rotation of the winch to reach the target depth. fig. 6 winch control system 3. design of rov the rov is mounted under the hull to acquire underwater data. rov is connected with usv to receives gps information and power, transmits image information acquired by underwater camera and multi-beam sensor, undersea topography information, and position and attitude information of underwater robot. 3.1. rov structure the shape of rov is shown in fig. 7. by using three thrusters, it can control surge, sway and yaw. fig. 7 3d modeling & feature of rov proceedings of engineering and technology innovation, vol. 9, 2018, pp. 17 24 copyright © taeti 21 rov is designed to be docked to usv to reduce power consumption and fluid resistance when moving. rov is docked shown as fig. 8. fig. 8 feature of docking station 3.2. control system of rov the rov control system is based on a dsp control board. the dsp performs rov driving, sensor data acquisition, arithmetic processing, and communication functions. the control system rov is equipped with usbl, depth and ahrs sensors to determine position and orientation. acquired images and undersea features were processed with a separate embedded pc to improve storage and processing efficiency. fig. 9 shows the rov control system. fig. 9 control system of rov fig. 10 is a control program designed for control of usv and rov simultaneously. usv gps position information is displayed on the bottom right corner, and rov relative position is displayed on the bottom center. in addition, the attitude of usv and rov can be displayed in 3d, and the waypoint and control gain can be set. rov dock rov multi-beam camera proceedings of engineering and technology innovation, vol. 9, 2018, pp. 17 24 copyright © taeti 22 fig. 10 control console program of hsuv 4. hybrid usv-rov system 4.1. control system of usv-rov hybrid system the control system of hsuv is shown in fig. 11. rf communication can be used to transfer route path and winch control commands to the usv or to receive position and attitude information of the hybrid system. underwater images and undersea topography information can be received using lte network and monitored in real time. fig. 11 control system of usv-rov hybrid system 5. performance of hybrid system 5.1. navigation algorithm the usv must be able to move to a specific location and maintain its position for underwater work. therefore, the navigation algorithm was developed for calculates the target heading of the usv and move to the target point by using the los (line of sight) method. 𝜓𝑈 = 𝑡𝑎𝑛 −1 [ 𝑌𝑘−𝑌(𝑡) 𝑋𝑘−𝑋(𝑡) ] 𝜌2(𝑡) = [𝑋𝑘 − 𝑋(𝑡)] 2 + [𝑌𝑘 − 𝑌(𝑡)] 2 < 𝜌𝑐 2 (1) in (1), [y (t), x (t)] symbolizes the position of the unmanned line, and [x_k, y_k] symbolizes the position of the waypoint. 𝜌𝑐 symbolizes the radius of the waypoint. if the usv enters the radius of the waypoint, the heading angle is calculated to the x and y positions of the usv using equation 1 so that the target does not deviate from the radius. also, it was designed so as not to deviate from the predetermined radius by using the pid controller as in equation 2 and equation 3 . proceedings of engineering and technology innovation, vol. 9, 2018, pp. 17 24 copyright © taeti 23 𝑢𝑝𝑜𝑠𝑖𝑡𝑖𝑜𝑛 = 𝐾𝑝 ∙ 𝑒𝑝𝑜𝑠𝑖𝑡𝑖𝑜𝑛 + 𝐾𝑖 ∙ ∫ 𝑒𝑝𝑜𝑠𝑖𝑡𝑖𝑜𝑛𝑑𝑡 + 𝐾𝑑 ∙ �̇�𝑝𝑜𝑠𝑖𝑡𝑖𝑜𝑛 (2) 𝑢𝜓 = 𝐾𝑝 ∙ 𝑒𝜓 + 𝐾𝑖 ∙ ∫ 𝑒𝜓𝑑𝑡 + 𝐾𝑑 ∙ �̇�𝜓 (3) 5.2. performance test of navigation fig. 12 is the test result of the waypoint driving performance of the hsuv when the rov is launched by 0.5m. the darkly marked point is the target point, and the softly marked is the actual travel path of the hybrid system. it seems to be pushed in one direction due to disturbance such as wind, but it returned to the target point within a short time. fig. 12 test result of waypoint navigation test 6. conclusions in this paper, we proposed a new hybrid system that combines the usv and the rov to overcome the cumulative navigation error problem of the underwater robot and the battery problem for long time operation. the usv is connected to the underwater robot with a tether cable that enables gps data transmission and stable power supply. the rov equipped with underwater sensors which can acquire underwater image capture and undersea topography information and can transmit it to the land through the usv in real time was constructed. line-of-sight navigation and dynamic positioning algorithm were developed for the usv to perform autonomous navigation and keeping-position with rov’s underwater operations. performance test of the usv was done at real sea using the developed hybrid usv-rov. the navigation algorithm was applied using the gps and ahrs sensor for tracking the trajectory composed of ten circular way-points. despite the disturbance waves and tides, the hsuv performed tracking all the waypoints successfully. the experimental results showed that the maximum distance between hsuv and waypoints was under 1.5m, which is the radius of the waypoint and the position error between usv and rov was under 1.5m, according to usbl data. depth error of rov was about 0.5m, which is considered to be occurred by waves. acknowledgement this research was supported by unmanned vehicles advanced core technology research and development program through the national research foundation of korea(nrf), unmanned vehicle advanced research center(uvarc) funded by the ministry of science and ict, the republic of korea(no. 2017m1b3a1a03094006), and this research is a part of the project national research foundation of korea (nrf-2016r1a2b4011875) references [1] b. h. jun, j. y. park, p. m. lee, f. y. lee, and j.h. oh, “development and tank of an autonomous underwater vehicle 'isimi',” journal of ocean engineering and technology, vol. 21, no. 2, pp. 67-74, april 2007. proceedings of engineering and technology innovation, vol. 9, 2018, pp. 17 24 copyright © taeti 24 [2] a. kukulya, a. plueddemann, t. austin, r. stokey, m. purcell, b. allen, r. littlefield, l. freitag, p. koski, e. gallimore, j. kemp, k. newhall, and j. pietro, “under-ice operations with a remus-100 auv in the arctic,” autonomous underwater vehiciles(auv), ieee/oes, pp.1-8, september 2010. [3] p. mahecek, c. a. kitts, and i. mas, “dynamic guarding of marine assets through cluster control of automated surface vessel fleets,” ieee/asme transactions on mechatronics, vol. 17, no. 1, pp.65-75, february 2012. [4] s. p. hong, j. w. jeong, c. w. lee, h. s. lee, h. w. choi, and i. h. park, “a study on the environment of usv wireless communication,” journal of ocean engineering and technology, vol. 23, no. 2, pp. 53-57, april 2009. [5] j.e. manley, “unmanned surface vehicles, 15 years of development,” oceans 2008, ieee, pp. 1-4, september 2008. [6] d. pearson, e. an, m. dhanak, k. v. ellenrieder, and p. beaujean, “high-level fuzzy logic guidance system for an unmanned surface vehicle (usv) tasked to perform autonomous launch and recovery (alr) of an autonomous underwater vehicle (auv),” autonomous underwater vehiciles, ieee/oes, october 2014. [7] r. j. yan, s. pang, h. b. sun, and y. j. pang, “development and missions of unmanned surface vehicle,” journal of marine science and application, vol. 9, no. 4, pp. 451-457, december 2010. [8] edoardo i. sarda and manhar r. dhanak. “a usv-based automated launch and recovery system for auvs,” ieee journal of oceanic engineering, vol. 42, no. 1, pp. 37-55, january 2017. [9] m. miranda ii, “mobile docking of remus-100 equipped with usbl-aps to an unmanned surface vehicle: a performance feasibility study,” florida atlantic university, florida, 2014.  proceedings of engineering and technology innovation, vol. 11, 2019, pp. 01-11 half/full toroidal, single/double roller, cvt based transmission for a super-turbo-charger albert boretti * independent scientist, bundoora, australia received 30 september 2018; received in revised form 02 october 2018; accepted 05 ocotber 2018 abstract turbocharging dramatically improves the power density of internal combustion engines. however, when the energy to turbine is either smaller or larger than what is needed by the compressor for a specific steady state or transient point, there are downfalls, such as turbo-lag, lack of boost, or wasted exhaust energy. here we propose a super-turbo-charger, where the turbo-charger shaft is connected to the crankshaft through a continuously variable transmission (cvt) and gears’ pairs. different designs are considered, half-toroidal and full-toroidal, single-roller or double-roller, single-cvt or two-cvt-in-series. energy is drawn from the crankshaft or delivered to the crankshaft to improve the steady or transient operation. in this paper, a six-cylinder, dual fuel diesel injection, ignition engine is super-turbocharged. engine performance simulations show the opportunity to achieve a nearly flat maximum torque all over the range of engine speeds and very high efficiencies from one fourth of the load up-wards. keywords: supercharging, turbocharging, dual fuel, diesel injection ignition, lng 1. introduction here we report the simulated performances of a dual fuel double direct injector diesel-lng engine featuring a super-turbo charger [1] and [2]. while in a turbocharger the compressor energy is obtained from the exhaust gases, in a super-charger the compressor energy is extracted from the engine crankshaft. in the design proposed in [1] and [2], an over-sized turbo-charger is connected to the crankshaft through a continuously variable transmission (cvt) based transmission to provide a large variation of the speed ratio between the engine and the turbocharger. a clutch may also be added to also permit the balanced operation of the turbocharger decoupled from the speed of the engine when deemed appropriate [3]. the opportunity to run the turbocharger at a speed different from the equilibrium speed may improve steady and transient performances, either as fuel conversion efficiency and power/torque output. power can be delivered to or drawn from the crankshaft. improved power and torque are expected, especially at low speed. additionally, the excess exhaust energy may be recovered at high speed. the super-turbo also helps with transient, either accelerations, removing the turbo-lag, or deceleration, permitting, in case of a vehicle equipped with a kinetic energy recovery system, also the recovery of the exhaust energy. the optimum operating point of the engine and turbo-charger may differ from a point of balanced works of compressor and turbine, and the additional degree of freedom in the optimization is beneficial. as the extra work from the turbine can be collected at the crankshaft, the turbo-charger can be selected larger than in a traditional installation, because the crankshaft can collect the extra energy from the turbine. the control of the turbo-charger speed, and therefore of the flow of power to or from the crankshaft, and the pressure boost, is achieved by controlling the speed ratio across the mechanism. this is done through the cvt based mechanism. the cvt may be based on half-toroidal or full-toroidal design. * corresponding author. e-mail address: a.a.boretti@gmail.com proceedings of engineering and technology innovation, vol. 11, 2019, pp. 01-11 2 natural gas can be either turned into a drop-in diesel, liquid fuel by using gas-to-liquids (gtl) technologies or used compressed (cng) or liquified (lng) in a dual fuel conversion of the diesel engine. either way, there are advantages for emissions. gtl diesel is colorless and odorless. it contains less impurities, mostly sulphur, aromatics and nitrogen, than diesel from crude oil. in the gtl process, synthesis gas (hydrogen and carbon monoxide), is produced from natural gas by partial oxidation. impurities are removed. then the synthesis gas is turned into liquid hydrocarbons using a catalyst. cracking and isomerization is finally applied to tune the molecular chains needed in the desired product. the most successful gtl plant to date, produces a drop-in diesel fuel (gt-fuel), plus kerosene, paraffin, various fluids and solvents, oils, naphtha, waxes and phase change materials [4]. this gtl fuel consists mostly of straight chain normal-paraffins and branched iso-paraffins, has a much higher cetane number, a higher mass calorific value, lower levels of sulphur and aromatics, and a lower density. these different properties may be used in a certain extent to reduce emissions of particulate matter, nox, hydrocarbons and carbon monoxide. test results have been promising, however, with benefits that are reducing in engines satisfying more strict emission standards. the option to use cng or lng is more complicated for the engine and vehicle hardware and software, as major changes are needed. however, the advantages in terms of emissions may be even much larger, being natural gas much cleaner to burn than any one of the liquid fuels produced by refining crude oil. westport has provided so far, the most brilliant and widespread dual fuel conversion of diesel heavy duty engines with their hpdi concept [5]. the hpdi is based on the direct injection of both the diesel and the lng, with a pre/pilot injection of the diesel, and then a main injection of the lng, from a same dual fuel injector. we proposed an enhanced version of the hpdi concept, where two injectors were used for every cylinder, one for the diesel, and one of the alternative fuel, to deliver improved combustion modes working on the relative timings of diesel and alternative fuel injection, and therefore on the premixed and diffusion combustion of the alternative fuel [6-9]. 2. cvt based super-turbo-charger the toroidal cvt based transmission here proposed is a variant of the one anticipated by torotrack for their variable-speed supercharging technology [10], hu et al. [11]. the torotrak v-charge supercharger is a centrifugal compressor linked to the crankshaft through a cvt based mechanism. we use the toroidal cvt based transmission to connect the crankshaft to a turbine and compressor sharing same shaft rather than a compressor. the torotrack v-charge comprises a traction drive variator and a traction drive epicyclic. the traction drive variator changes the drive ratio from 0.355: 1 (underdrive) up to 2.820: 1 (overdrive, 2.820 = 1/0.355). before the traction drive variator, there is a 3:1 overdrive from the engine. after the traction drive variator, there is a 12.67:1 overdrive through the traction drive epicyclic. the cvt is full toroidal. the external overdrive is 312.67 = 38:1. the spread of the cvt speed ratios is 2.820/0.355 = 7.930. different variants are being considered to make the design simpler and better cover the speed ratios of turbocharger and engine. the total overdrive may be changed, as well as the splitting of the total overdrive before and after the traction drive toroidal cvt variator. by increasing the first overdrive, the traction drive toroidal cvt variator operates at higher speed and reduced torque for same power transmitted. the spread of the cvt speed ratios is the most difficult part to change. with a full-toroidal design, the spread of the cvt speed ratios is limited to around 8. with a half-toroidal design, the spread of the cvt speed ratios is limited to around 4. opposite to the gasoline engine application of [3], where both the engine and the turbocharger have large spreads of speed, in this case, where the spread of speed is reduced in both the engine and the turbocharger, a single full-toroidal cvt may be used. opposite to [3] where the minimum speed is 1,000 rpm, but the maximum speed of the engine is 5,500 rpm and the maximum speed of the turbocharger is 200,000 rpm, in the application here considered the maximum speed of the engine is proceedings of engineering and technology innovation, vol. 11, 2019, pp. 01-11 3 4,500 rpm and the maximum speed of the turbocharger is 150,000 rpm. to achieve larger cvt speed ratios, [3] use two half-toroidal cvt in series, giving a cvt speed ratios of about 16, or a full-toroidal cvt giving a cvt speed ratio of about 8 plus a clutch. the clutch prevents the engine crankshaft to run the turbocharger at excessive speeds reducing the loads. a cvt based crankshaft to turboshaft connection, with the cvt mechanism composed of two half-toroidal cvt in the series, is shown in fig. 1. the cvt mechanism is composed of two half-toroidal cvt in the series. alternatively, it is composed of one full-toroidal cvt. n1 is the first overdrive. n3 is the last overdrive. the cvt delivers a speed ratio from n2:1 to 1/n2:1. this cvt model is reproduced, modified after [12]. with one half-toroidal cvt, at the most n2=2, for a spread of speeds ratio across the cvt of n2 2 =4. with two half-toroidal cvt in series, at the most n2=4, for a spread of speeds ratio across the cvt of n2 2 =16. with one toroidal cvt, at the most n2=2.828, for a spread of speeds ratio across the cvt of n2 2 =8. fig. 1 cvt based crankshaft to turboshaft connection. (reproduced and modified after kamel [12]) 3. efficiency of the cvt after applications of half-toroidal and full-toroidal single-roller cvt are covered in [10,13-14]. design and analysis of half-toroidal and full-toroidal single-roller cvt are covered in verbelen et al. [15], zhanget al. [16], zou et al. [17], carbone et al. [18], delkhosh et al. [19], verbelen et al. [20] and verbelen et al. [21]. while these designs are based on multiple single-roller, the opportunity to use multiple double-roller is considered in carbone et al. [22], de novellis et al. [23]. verbelen et al. [20] discusses in a synthetic way the efficiency of a cvt of half-toroidal design. the efficiency η is expressed as a function of input speed, load torque, speed ratio and clamping force. the impact of clamping force can be eliminated if a proper slip controller is considered. thus, the efficiency may be expressed as a function of speed ratio, load torque and input speed. verbelen et al. [20] show that after having computed (or measured) an efficiency map 𝜂𝑟𝑒𝑓(𝜏, 𝑇1), with 𝜏 the speed ratio across the cvt and t1 the input torque, then there is a linear relation between efficiency and input speed. hence, verbelen et al. [20] then rescale the efficiency map computed at a certain input speed to a different input speed proportional to the difference of input speed. in a half-toroidal cvt, power is transmitted from the input disc in the output disc through a system of rollers. the geometrical speed ratio, is the ratio of 𝑟1 and 𝑟2. the contact points can be changed by changing the tilting angle 𝛾 of the rollers. to avoid wear, a traction fluid is used. to limit the slip in the device, the clamping force f is applied at the input disc. this is shown in fig. 2. verbelen et al. [20] model of a half-toroidal cvt by using the approach detailed in carbone et al. [18] and verbelen et al. [21] here briefly summarized. only steady state results are considered. proceedings of engineering and technology innovation, vol. 11, 2019, pp. 01-11 4 fig. 2 half-toroidal single-roller cvt geometry the model consists of a contact model, a mechanical model, a simple load and a slip controller. in the contact model the traction conditions are determined based on the clamping force f, the tilting angle 𝛾, the input speed n1 and output speed n2 of the cvt. the traction conditions are used to calculate the generated torque at the input 𝑇1 and output 𝑇2. slip is inevitable, needs control, and it can be calculated as: slip = 1 − n1 n2 r1 r2 (1) the speed ratio range is defined by geometrical parameters and is limited by the input speed. based on the geometrical parameters the maximum speed ratio is: τmax = 1 + α 1 + α − cos (2θ − π 2 ) (2) where α is the aspect ratio, equal to r0/rr and 𝜃 is the half cone-angle. the minimum speed ratio is then the inverse of the maximum speed ratio, and the spread of speeds across the cvt is the maximum speed ratio squared. if the input speed increases, the roller speed increases. the heavy loading of the rollers limits the maximum permissible speed. the maximum load torque 𝑇𝑚𝑎𝑥 then depends on the position of the roller 𝛾 which is related to the speed ratio 𝜏, the maximum allowable pressure 𝑝𝑚𝑎𝑥 imposed by the clamping force 𝐹 and the geometrical parameters of the cvt. the steady state efficiency of the cvt is then: η = n2 ∙ t2 n1 ∙ t1 (3) this efficiency is a function of the four input parameters input speed, load torque, speed ratio and clamping force, but if unstable conditions are excluded from proper slip control through 𝐹, then only three parameters may be considered, input speed, load torque, speed ratio. verbelen et al. [20] proposes an efficiency map 𝜂𝑟𝑒𝑓(𝜏, 𝑇1) as a function of load torque 𝑇1 and speed ratio 𝜏, for a specific reference speed n1,ref=3,000 rpm. their efficiency increases with the load torque, and it reduces with the speed ratio, even if there is a limiting line on top of the operating range from the maximum allowable torque. minimum efficiencies are found for load torque zero. verbelen et al. [20] then observed a quasi linear relation between input speed and efficiency. their speed ratio 𝜏 vary between 0.5 and 2. their maximum load torque is about 400 n·m. they propose to take: 𝜂(𝜏, 𝑇1, n1) = 𝑘(𝜏, 𝑇1)·(n1 − n𝑟𝑒𝑓) + 𝜂𝑟𝑒𝑓(𝜏, 𝑇1) (4) proceedings of engineering and technology innovation, vol. 11, 2019, pp. 01-11 5 with n𝑟𝑒𝑓 the reference speed and 𝜂𝑟𝑒𝑓 the efficiency at this of reference speed. for a given speed ratio, the efficiency increases with speed, in addition to increase with the load torque.1 by using the approach of verbelen et al. [20], we may then obtain an efficiency map at the reference speed characterized by high efficiencies 91-94% in the high load torque range for every speed ratio, but efficiencies 82-88% in the low load torque range for every speed ratio, and efficiencies as low as 60% for zero load torque. the efficiencies slightly increase with the speed ratio. these efficiencies, then slightly increase or reduce by increasing or decreasing the speed. the efficiencies of verbelen et al. [20] are referred to specific values of input torque t1 and speed n, for a specific geometry, of giving dimensions. by changing the design of the cvt, dimensions, and input torque and speed, the efficiency values will change. discussing of the different design opportunities, we note that: (1) reduced efficiencies are obtained by using a full, rather than a half, toroidal cvt for the same spread of speeds ratio. (2) better efficiencies are obtained adopting a double-roller vs. a single-roller design. (3) efficiencies reduce with two cvt in the series, being the total efficiency the product of the two efficiencies. (4) the ac load is inductive: r=64.81ω; l=0.488h; cos (φ) =0.38. (5) for the specific application requiring a large spread of speeds ratio, the best option for what concerns efficiency is the full toroidal design, as two half-toroidal cvts in series are needed to deliver same, a large spread of speeds ratio. because verbelen et al. [20] is practically ignoring clamping control, they get very low efficiency figures when the module of the load torque reduces. the efficiencies of verbelen et al. [20] may generally be improved if the clamping force is managed to cause the traction coefficient to always sit around 0.07. in this case, the efficiency can be maintained above 94% for the double-roller variants, and above 88% for the single-roller variants, almost regardless of input speed and torque. by assuming similarity between designs, i.e. the non-dimensional maps: x = η𝑟𝑒𝑓 ( 𝜏 𝜏𝑚𝑎𝑥 − 𝜏𝑚𝑖𝑛 , 𝑇1 𝑇1,𝑚𝑎𝑥 ) (5) 𝑦 = k ( 𝜏 𝜏𝑚𝑎𝑥 − 𝜏𝑚𝑖𝑛 , 𝑇1 𝑇1,𝑚𝑎𝑥 ) 𝑁1,𝑟𝑒𝑓 (6) are about the same, apart from a multiplying technology factor χ larger than unity accounting for the full-toroidal and double-roller advantages over half-toroidal and single -roller designs, following verbelen et al. [20], we may define for the single-roller, full-toroidal cvt the efficiency map 𝜂𝑟𝑒𝑓(𝜏, 𝑇1) of fig. 3, and the speed correction factor map 𝑘(𝜏, 𝑇1) of fig. 4. for sake of simplicity, we also assume that same 𝜂 apply for transmission of power to, and from, the super-turbocharger, obviously with entry conditions on one side or the other of the single-roller full-toroidal cvt. the proposed efficiency map is only preliminary. while a good design with enhanced clamping control may certainly deliver an area of operating points well, will above 90%, albeit with efficiencies still dropping below 90% when seal and parasitic losses will dramatically increase at very low torques, this novel application of the cvt certainly needs specific proceedings of engineering and technology innovation, vol. 11, 2019, pp. 01-11 6 research and development with prototyping and testing a critical aspect. the design, and consequently the efficiency map, is very application specific, not generic. fig. 3 reference efficiency map 𝜂𝑟𝑒𝑓(𝜏, 𝑇1) of the cvt based transmission fig. 4 speed coefficient map 𝑘(𝜏, 𝑇1) of the cvt based transmission 4. performance of the engine we consider here a total external overdrive of 37.6:1, and a single full-toroidal cvt of overdrive 2.820 and underdrive 0.355 = 1/2.820, for a spread of the cvt speed ratios 2.820/0.355 = 7.930. opposite to boretti and castelletto [3], a clutch is not essential with a full-toroidal cvt. fig. 5 cvt entry power in kw vs. cvt entry speed in rpm and cvt speed ratio fig. 6 cvt entry torque in n·m vs. cvt entry speed in rpm and cvt speed ratio fig. 5 presents the cvt entry power, side of the crankshaft, versus the cvt entry speed, side of the crankshaft, and the cvt speed ratio. the power is in kw and the speed is in rpm. external overdrive is 37.6:1. the spread of cvt speed ratios is 2.8202 = 7.93. there is only one single-roller, full-toroidal cvt. the overdrive before the cvt is 3:1. the power is positive if delivered to the cvt (extra power to run the compressor). the overdrive before the cvt is 3:1, as it in the design proposed by torotrack for the v-charge. fig. 6 presents the cvt entry torque, side of crankshaft, versus the cvt entry speed, side of the crankshaft, and the cvt speed ratio. the torque is in n·m and the speed is in rpm. in the engine performance simulations, the efficiency of the cvt transmission, only weights on the computation of the power to and from the turbocharger, i.e. only on the difference in between the compressor and turbine powers, when these two parameters are unbalanced. we only consider the steady operation of a turbocharged diesel injection, ignition dual fuel diesel-lng engine featuring this super-turbocharger in very preliminary results. the transient operation is not considered, but the turbo-lag is expected to be completely cancelled. the diesel is assumed to have carbon atoms per molecule 13.50, hydrogen atoms per molecule 23.60, lhv of 43.25 mj/kg. the lng (methane) is assumed to have carbon atoms per molecule 1, hydrogen atoms per molecule 4, and lhv of 50 mj/kg. the model is of a dual-fuel, diesel lng injection ignition engine, with each cylinder supplied with two independent injectors for the diesel and the lng. the turbo-charger is oversized. also oversized are the port areas and the valves’ diameters proceedings of engineering and technology innovation, vol. 11, 2019, pp. 01-11 7 and lifts. the turbo-charger is intended to be used for a maximum speed-times-displacement that is 30% larger than the present 4,500 rpm-times-3.8 liters. maximum speed of the turbo-charger is 150,000 rpm. minimum speed is 13,000 rpm. minimum speed of the engine is 1,000 rpm and maximum speed 4,500 rpm. (a) at 1,500 rpm (b) at 3,000 rpm (c) at 4,500 rpm (d) at different speeds fig. 7 lng to diesel ratio by mass and by energy as a function of the bmep in bar fig. 7 finally presents the ratio of lng (methane) to diesel by mass or by energy as a function of the bmep in bar at different speeds. the diesel injected is basically the fuel needed in the diesel only operation during the pilot and pre-injections, while the lng injected is basically the fuel needed in the main injection during the diesel only operation. at minimum loads, the pilot and pre-injection diesel represents the most part of the fuel. at higher loads, the main-injection lng dramatically increases. while the c/h ratio of the diesel is 13.50/23.60 = 0.57, the c/h ratio of the lng is 0.25. hence, because of the lpg fuel replacement of the main injection diesel, the equivalent c/h ratio of the fuel changes with the load from about 0.56 at minimum loads to about 0.30 at maximum loads. hence, approaching top loads, the co2 emission is almost halved. similarly, at top loads, the pm production within the cylinder is expected to be minimal, as it should reduce more than proportionally to the diesel fuel replacement by mass. fig. 8 total engine power in kw, as a function of engine speed and brake mean effective pressure (bmep) in bar fig. 9 compressor power in kw, as a function of engine speed and brake mean effective pressure (bmep) in bar proceedings of engineering and technology innovation, vol. 11, 2019, pp. 01-11 8 fig. 10 turbine power in kw, as a function of engine speed and brake mean effective pressure (bmep) in bar fig. 11 difference between turbine and compressor power in kw, as a function of engine speed and brake mean effective pressure (bmep) in bar fig. 12 fuel conversion efficiency η * , ratio of the total power to the fuel flow power, in %, vs. engine speed in rpm and brake mean effective pressure (bmep) in bar fig. 13 brake specific co2 emission, in g/kw-h, vs. engine speed in rpm and brake mean effective pressure (bmep) in bar fig. 14 apparent equivalent air-to-fuel ratio λ vs. engine speed in rpm and brake mean effective pressure (bmep) in bar figs. 8 to 11 present the total engine power, compressor power, turbine power, and difference between turbine and compressor power in kw, as a function of engine speed in rpm and brake mean effective pressure (bmep) in bar. fig. 12 presents the fuel conversion efficiency η * , now ratio of the total power, from the pistons and the turbocharger at the crankshaft, to the fuel flow power, vs. the speed of the engine in rpm and the brake mean effective pressure (bmep) in bar. fig. 13 presents the brake specific co2 emission, in g/kw-h, vs. engine speed in rpm and brake mean effective pressure (bmep) in bar. figure 14 finally presents the apparent equivalent air-to-fuel ratio λ vs. engine speed in rpm and brake mean effective pressure (bmep) in bar. the turbo-charger continues to receive energy at low speeds, mid-high loads, while at high speeds, mid-high loads, the turbocharger usually provides energy. the low speed torque (bmep) is increased up to the medium speed values. with the super turbocharger, the maximum torque and the maximum power are both increased. proceedings of engineering and technology innovation, vol. 11, 2019, pp. 01-11 9 the efficiency is about constant 2,000 to 4,000 rpm, and one third to full load, only slightly but mostly improved vs. the baseline turbocharged configuration. at a maximum load, the engine runs lean, however with λ values much smaller than in a diesel. this may require some lng to be introduced before the diesel is injected. above 4,000 rpm, the efficiency reduces mostly because of the combustion becoming more difficult. 5. discussion the torotrak v-charge super-charger was a production-ready cvt based supercharger. the technology was developed to the point that the concept was proven and ready to commercialization after prototyping and testing in laboratory experiments and on the road [11, 24-26]. the v-charge was installed on a small ford focus titanium 1.0t ecoboost. power was risen from 92 to 118 kw and torque was up from 169 to 249 n m. the improvements in fuel economy were significant, rated at 12% better fuel economy versus the naturally aspirated larger engine. only the recently announced freeze in the development of new internal combustion engines across the world, due to the possible dismissal of the internal combustion engine in favor of the electric mobility promote by european lawmakers within two decades, has halted the project. the proposed innovation differs from the v-charge only for the connection to a turbocharger rather than a compressor. the additional packaging issues are minimal, and the system may well fit the engine compartment of even small cars such as the ford focus, as it was done for the v-charge. the dynamic model of the mechanical system, as well as of the engine, is missing. in this preliminary phase, we only consider the steady state operation. while the dynamic models will be needed in a better proof of concept and obviously product development, the static model is here used to highlight some of the advantages of the proposed innovation. the proposed efficiency map is a reference map obtained from load and speed extrapolation from an existing design. as the efficiency of the cvt only weights on the difference between turbine and compressor power, the influence on engine efficiency of small variations in cvt efficiency are negligible. only a steady state model of the cvt is used. while it is correct to assume that the efficiency depends on input speed, load torque and speed ratio, however, during transient operation, there is always an increase in slip and therefore a drop -in efficiency. 6. conclusions this work follows the prior works by the author on dual fuel diesel injection, ignition with traditional turbochargers, and the super turbocharging of direct injection diesel engines and direct injection jet ignition gasoline engines. the super turbocharger connecting the turbocharger shaft to the crankshaft through a cvt and a gear is an option to improve both fuel conversion efficiency and power/torque output, either steady state and transient. it is expected that the proposed innovation will drastically improve the transient operation cancellation of the turbo-lag. there is no reason why the proposed innovation should not deliver same improvements, under this criterion, of f1 turbocharged engines with mgu-h, or the ford focus titanium 1.0t ecoboost with the v-charge. the preliminary results proposed here are specific for a dual fuel diesel-lng diesel injection ignition engine. the engine delivers high boost at any speed, plus high fuel conversion efficiency, above 40%, over the most part of the load-speed map. albeit the results are qualitative, the trends are trustworthy, as demonstrated by the latest f1 experience, where a turbo-charger similarly operated through a motor-generator unit connected to the turbo-charger shaft, and the use of direct injection and jet ignition, translate into strongly improved fuel economies running lean of stoichiometry stratified with minimal cyclic variability, no turbo-lag, and dramatically improved low speed torque. proceedings of engineering and technology innovation, vol. 11, 2019, pp. 01-11 10 acknowledgements this preliminary simulation work was performed in 2013/2014 to understand the areas where to focus further research and development in f1 and le mans lmp1-h powertrains, within the limits of the rules of the time, and on the hypothesis of relaxed constraints. the use of high pressure (500 bar) direct injection and jet ignition, coupled to a carefully redesigned, oversized, turbocharger connected to a motor-generator unit (mgu), was the proposition for f1 gasoline applications, as well as for lmp1-h gasoline powertrains. similarly, an oversized turbocharger fitted to a motor-generator unit and extremely high pressure direct injection (3000 bar) was the suggestion for lmp1-h diesel powertrains. the author received no funding for this work. conflicts of interest the author declares no conflict of interest. references [1] a. boretti, “super turbocharging the direct injection diesel engine,” nonlinear engineering, vol. 7, pp. 17-27, 2018. [2] a. boretti and a. ordys, “super-turbocharging the dual fuel diesel injection ignition engine,” sae technical paper, july, 2018. [3] a. boretti and s. castelletto, “super-turbocharged direct injection jet ignition gasoline engine,” paper accepted for presentation to the fisita world congress 2018, chennai, india, october 2018. [4] shell, “pearl gtl overview,” https://www.shell.com/about-us/major-projects/pearl-gtl/pearl-gtl-an-overview.html, october 2018. [5] westport, “westport hpdi 2.0,” https://www.westport.com/is/core-technologies/hpdi-2, october 2018. [6] a. boretti, “advantages of the direct injection of both diesel and hydrogen in dual fuel h2ice,” international journal of hydrogen energy, vol. 36, pp. 9312-9317, 2011. [7] a. boretti, “novel heavy duty engine concept for operation dual fuel h2-nh3,” international journal of hydrogen energy, vol. 37, pp. 7869-7876, 2011. [8] a. boretti, “numerical study of the substitutional diesel fuel energy in a dual fuel diesel-lpg engine with two direct injectors per cylinder,” fuel processing technology, vol. 161, pp. 41-51, 2017. [9] a. boretti, “advances in combustion systems for gas engines,” sae technical paper, september 2013. [10] m. magda, “inside look: variable-speed supercharging technology,” https://www.enginelabs.com/news/inside-look-variable-speed-supercharging-technology/, january 2016. [11] b. hu, s. akehurst, a. g. lewis, p. lu, d. millwood, c. copel, e. chappell, a. d. freitas, j. shawe, and d. burtt, “experimental analysis of the v-charge variable drive supercharger system on a 1.0 l gtdi engine.” proceedings of the institution of mechanical engineers, part d: journal of automobile engineering, vol. 232, pp. 449-465, 2018. [12] h. kamel, “toroidal cvt transmission,” https://grabcad.com/library/torodial-cvt-transmission, december 2012. [13] c. brockbank, “application of a variable drive to supercharger and turbo compounder applications,” sae technical paper, april, 2009. [14] d. cross and c. brockbank, “mechanical hybrid system comprising a flywheel and cvt for motorsport and mainstream automotive applications,” sae technical paper, april, 2009. [15] f. verbelen, s. derammelaere, p. sergeant, and k. stockman, “a comparison of the full and half-toroidal continuously variable transmissions in terms of dynamics of ratio variation and efficiency,” mechanism and machine theory, vol. 121, pp. 299-316, 2018. [16] y. zhang, x. zhang, and w. tobler, “a systematic model for the analysis of contact, side slip, and traction of toroidal drives,” journal of mechanical design, vol. 122, pp. 523-528, 2000. [17] z. zou, y. zhang, x. zhang, and w. tobler, “modeling and simulation of traction drive dynamics and control,” journal of mechanical design, vol. 123, pp. 556-561, 2001. [18] g. carbone, l. mangialardi, and g. mantriota, “a comparison of the performances of full and half-toroidal traction drives,” mechanism and machine theory, vol. 39, pp. 921-942, 2004. [19] m. delkhosh, m. s. foumani, m. boroushaki, m. ekhtiari, and m. dehghani, “geometrical optimization of half-toroidal continuously variable transmission using particle swarm optimization,” scientia iranica, vol. 18, pp.1126-1132, 2011. proceedings of engineering and technology innovation, vol. 11, 2019, pp. 01-11 11 [20] f. verbelen, s. derammelaere, p. sergeant, and k. stockman, “visualizing the efficiency of a continuously variable transmission,” in energy efficiency in motor driven systems eemods, in press. [21] f. verbelen, s. derammelaere, p. sergeant, and k. stockman, “half-toroidal continuously variable trans-mission: trade-off between dynamics of ratio variation and efficiency,” mech. mach. theory, vol. 107, pp. 183-196, 2017. [22] g. carbone, f. bottiglione, l. d. novellis, l. mangialardi, and g. mantriota, “the double-roller full-toroidal variator: a promising solution for kers technology,” in proceedings of the fisita 2012 world automotive congress, 2013, pp. 241-250. [23] l. d. novellis, g. carbone, and l. mangialardi, “traction and efficiency performance of the double-roller full-toroidal variator: a comparison with half-and full-toroidal drives,” journal of mechanical design, vol. 134, 2012. [24] s. akehurst, “torotrak v-charge cvt driven superchargera downsizing enabler?” www.eng.ox.ac.uk/engines/uniceg-presentations/v-charge, april 2017. [25] j. crosse, “torotrak readies v-charge supercharger for downsized engines,” https://www.autocar.co.uk/car-news/industry/torotrak-readies-v-charge-supercharger-downsized-engines, february 2017. [26] green car congress, “testing confirms torotrak v-charge variable drive mechanical supercharger enables more ambitious downsizing; 1.0l demo at aachen,” https://www.greencarcongress.com/2016/10/20161010-torotrak.html, october 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/). microsoft word 2-peti#5515 13-22.docx proceedings of engineering and technology innovation, vol. 16, 2020, pp. 13 22 appraisal of hybrid foundations on loose granular soil of western saudi arabia qazi umar farooq * , ahmed b. shaath department of civil engineering, islamic university of madinah, saudi arabia received 11 april 2020; received in revised form 28 june 2020; accepted 10 july 2020 doi: https://doi.org/10.46604/peti.2020.5515 abstract western saudi arabia is the most important geographic location for the islamic world. millions of pilgrims annually visit the region. to accommodate this huge number of pilgrims, various tall buildings are constructed, yet there is a need for many more. the demolition of old buildings is also in progress. the soil supporting of the foundations mostly comprises sand with varying relative density. most of the buildings are supported by conventional foundations, which require deep excavation for low relative densities. the combinations of pile and raft or hybrid foundations have been effectively used in tall buildings around the globe. in this research, comsol multiphysics fem analysis has been adopted to investigate the performance of a hybrid foundations with different pile combinations supported on loose granular soil. the results reflect the effective utilization of the hybrid foundation on loose sand. the arrangement of piles in the foundation system can affect the economy and safety of the structure. keywords: loose sand, hybrid foundation, pile arrangement, foundation settlement 1. introduction the western geographic region of the kingdom of saudi arabia is an extremely vital geopolitical zone. it not only contains the two holiest cities of islam but also has an important port city jeddah. the kingdom’s 2030 vision for the development of the tourism sector will likely boost the inflow of the tourists and the pilgrims [1]. the government is also planning to build a mega tourist destination named neom which is full-fledged smart city with various tourist attractions [2]. hence, there is a huge potential for infrastructure expansion in the region. the region is surrounded by various mountain ranges. the structure supporting soil mainly comprises sand with varying degrees of relative density [3]. foundation settlement in loose sand is a universal phenomenon [4]. during the service life of a structure, foundations supported on sands, which have lower relative density may experience large settlements due to their low shear strength and high compressibility [5]. under inadequate geotechnical conditions, conventional isolated footings or even raft may not perform very well. the shallow foundations supported by weak soil may cause undesirable distress in the building structure. the circumstances can become worse if there is a seismic activity [6]. the classical pile foundation system has been effectively used for vital projects under unfavorable soil conditions; however, it has several disadvantages as well. there are numerous issues that can arise during the foundation construction [7]. possible problems are categorized as follows: pile formation problems, concrete failing, reinforcement installation issues, * corresponding author. e-mail address: qaziumar@gmail.com tel.: +966-54-8453411; fax: +966-14-845-6139 proceedings of engineering and technology innovation, vol. 16, 2020, pp. 13 22 14 working load failing, and foundation failure due to the settlement [8]. deep foundations often intercept groundwater tables and require expensive dewatering. the deep foundations are subjected to high lateral loading and perform poorly if liquefaction occurs followed by the seismic activity [9-10]. the foundation performance of the raft can be enhanced by adding the piles. the ideal bearing capacity and controlled settlement circumstances can be achieved by properly designing the pile placement [11]. the pile-raft foundation system had been economically and effectively used in various parts of the world [12-13]. a parametric study conducted by using ‘plaxis’ reveals that the disconnected piled raft (dpr) foundation is more efficient for soft soils. the system performed better for heavy structures [14]. the piled raft foundation problem can be solved effectively through a combination of structural responses and geotechnical characteristics without a complex model of the soil and foundation. applying the pile ‘load-cut off’ procedure within piles capacity, the piles are dominant in terms of the piled raft behavior. in case the load is beyond the piles-capacity, the raft behavior is dominant [15]. various studies have shown that the addition of limited piles in length and number may serve a purpose. an experimental study conducted to analyze the behavior of pile–raft system showed that the small number of piles within the group had a low contribution to the overall load carrying capacity. the failure load for a piled raft is greater than standing pile group containing the same number of piles. this difference was increased by increasing the number of piles [16]. whilst most of the above-referred studies focus on the design procedures and performance of pile raft system on varying soil types, the evaluation of the pile placement and quantity under the raft, for particular geotechnical conditions has not been well understood. most of the experimental and numerical models have considered symmetric distribution of piles under the raft, which can be an economical and over-engineering option. in this research, the single term “hybrid foundation” is coined to designate the pile-raft, discontinued pile-raft, and other similar combined foundation systems. this paper discusses the philosophy of piles using the arrangement for the hybrid foundations under varying loading conditions. the research specifically targets the loose granular soils of western saudi arabia. however, results can be utilized for the design of hybrid foundations on other similar geotechnical conditions. 2. soil properties the study mainly focuses on the loose sand conditions of western saudi arabia. the soil samples were collected from the western city, al madinah. the samples were collected from an open pit at 1-m depth from the ground surface. the index properties of the soil were determined in laboratory. as per the unified soil classification system (uscs), the soil is classified as “poorly graded sand” with symbol ‘sp’ [17]. it is non-plastic in nature (liquid limit = 0, plastic limit = 0, plasticity index = 0). the sand is located at 7.5 yr 5/4 (brown) on the munsell soil color chart [18]. the grain size distribution curve of the considered soil is shown in fig. 1. the uniformity coefficient (cu) of the sand was = 3.64. fig. 1 particle-size distribution curve of al madinah sand 0 20 40 60 80 100 0.01 0.1 1 10 (% ) f in e r particle size (mm) particle-size distribution curve cu= 3.64 proceedings of engineering and technology innovation, vol. 16, 2020, pp. 13 22 15 the maximum and minimum void ratios (emax and emin) of the sand were also determined in the laboratory and summed up in table 1. in order to represent the loose sand conditions, a low relative density value of dr. = 30% has been selected in the study. the fundamental soil parameters required for the numerical study were then determine by using standard empirical correlations [19]. the soil parameters used in this study are shown in table 2. table 1 minimum and maximum void ratio limit unit weight γ � kn/m� void ratio e �% maximum 15.99 54.60 % minimum 13.79 79.27 % table 2 soil parameters for numerical modelling relative density d� (%) void ratio e (%) angle of internal friction ∅ �rad angle of friction δ �rad unit weight γ �kn/m� the angle of dilation ψ (rad) 30 71.87 0.531 0.398 14.32 0.007 3. methodology the strategic objective of this research is to evaluate the performance of the hybrid foundation under varying loading conditions, with different pile placement (pile layout). the pile placement is varied in terms of the number and location. the fem base numerical analysis has been adopted in this research. the numerical modeling has been done by using the solid mechanic’s module of comsol multiphysics software [20]. it was possible with the help of software to model the pile and soil together. three (03) cases of hybrid foundations were considered in this study. each case differs in terms of the pile layout. the 20-m by 20-m layout was considered in the model. the soil depth of 15-m was adopted. the hybrid foundations consisted of 1-m thick raft, along with 12-m long piles. the pile diameter was 0.5-m in all cases. there were six (06) piles used in case 1, the pile layout was straight. only three (03) piles were used in the staggered layout presented in case 2. however, case 3 had nine (09) piles placed in three (03) groups each comprises three piles. the pile positioning and geometry of the hybrid foundation models for cases 1, 2, and 3 are shown respectively in figs. 2(a)-2(c). (a) straight layout (case 1) (b) staggered layout (case 2) (c) grouped pile layout (case 3) fig. 2 hybrid foundation cases the soil properties defined in section 2 were incorporated into the model as the loose granular material. the soil is modelled as mohr-coulomb’s material [21], with the following governing equations: the function of normal and shear stress: f τ f (σ)= (1) proceedings of engineering and technology innovation, vol. 16, 2020, pp. 13 22 16 the linear function: f τ c σ tan φ= + (2) the linear function in terms of effective stress: , f τ c σ tan φ′ ′= + (3) where τf is the shear stress, σ is the normal stress, c is defined as material cohesion, φ is the frictional angle of the soil. c’ and σ� are respective values in terms of effective stresses. the hybrid foundation material is assumed to be reinforced concrete. the concrete is considered as linear elastic material with typical values of density poisson‘s ratio and elasticity modulus. 3.1. loadings fig. 3 typical floor plan of the building the typical floor plan of the building is indicated in fig. 3. the floor area of 19.5-m x 19.5-m is adopted to overlay the 20-m x 20-m raft of hybrid foundation. building design loads are taken as per building code requirements for structural concrete (aci 318m-11) [22]. live load (ll)=10kn/m 2 was applied to all floors. the self-weight of the structure has been considered in the analysis by using cosol multiphysics ‘body load’ built-in module. it can accurately apply the self-weight of the object, by using its geometry, material density, and gravity. table 3 shows the number of stories and the estimated design load: table 3 estimate design load/stress number of stories estimate design load (kpa) 1 54.4 2 108.8 3 163.2 4 217.6 5 272 3.2. numerical modeling the surface load is applied on top of the raft as per table 3. each case has been tested with the above five loading conditions. the sides of the raft were kept free. two built-in material models have been utilized in the fem analysis. the first material model is the concrete, which has been adopted for hybrid foundation material and uses linear elastic equations. the second material model is the elastoplastic mohr-coulomb’s model used for the 15-m thick layer of loose granular soil, supporting the hybrid foundation and its influence zone. whereas, the natural ground strata, supporting the overall system (hybrid foundation and surrounding soil) is considered as a spring foundation. the spring constant of the soil was taken as 100mn/m.m 2 . the analysis was done considering the stationary condition. the system has been analyzed by using a flexible generalized minimum residual (fgmers) iterative solver. the convergence plot for the solution of case 2 with 272 kpa load is presented in fig. 4. proceedings of engineering and technology innovation, vol. 16, 2020, pp. 13 22 17 fig. 4 convergence plot for case 2 4. analysis results the results obtained from the fem based stress-strain study are presented in this section. each case has been tested for all the five designed loads. the results are plotted for total displacement counter, stress or pressure distribution, and total displacement at critical sections. the plots for each case against the above mention parameters are represented separately in the following subsections: 4.1. straight pile layout (a) displacement contour (b) stress distribution (c) displacement at critical sections (y-z plane) (d) displacement at critical sections (x-z plane) fig. 5 displacement and pressure distribution for straight pile layout (case 1) the fem analysis results for straight pile layout (case 1) are presented in fig. 5. the results are plotted for the maximum applied load of 272 kpa. the top left diagram shows the total displacement contour, it can be seen that the maximum displacement occurs at the center of the raft. the maximum stress developed at the intersection of pile and raft (top right proceedings of engineering and technology innovation, vol. 16, 2020, pp. 13 22 18 diagram). the diagrams at the bottom of the figure display the displacement at sections located at the center line of piles in z-x and z-y planes. in this case, the maximum foundation displacement is 18.0 mm and the maximum pressure is 16.32 mpa. the pressure plot represents the mean stress value at any given location. mathematically, it is 1/3 of the first invariant of the cauchy or true stress tensor. it can be obtained by: 11 22 33 π (σ σ σ ) / 3= + + (4) where π is the pressure (mean stress) in mpa, σ11, σ22, and σ33, are principal stresses. 4.2. staggered piles the case 2 presents the staggered pile layout only three piles were used. the results of the numerical analysis at 272 kpa load are shown in fig. 6. similar to case 1, the displacement and pressure counters are presented in the top two diagrams. the bottom diagrams show displacement at sections, located on the centerline of the piles. the maximum displacement, in case 2, was 19.5-mm positioned at the center of the raft. the maximum stress was 14.35-mpa located at the pile-raft joint. like case 1 noticeable stresses were observed throughout the pile length. (a) displacement contour (b) displacement at critical sections (y-z plane) (c) stress distribution fig. 6 displacement and pressure distribution for staggered pile layout (case 2) 4.3. group piles the last case presents the group piles. in this case, nine (09) piles were used in the three groups at a staggered position. the displacement and stress contours generated from fem results are shown in fig. 7. the maximum displacement was 29.36 mm, while the maximum pressure was just 0.24 mpa. proceedings of engineering and technology innovation, vol. 16, 2020, pp. 13 22 19 (a) displacement contour (b) stress distribution fig. 7 displacement and pressure distribution for group pile layout (case 3) 5. discussions fig. 8 summary of fem results (displacement) fig. 9 summary of fem results (stresses) in all of the three cases, the soil stresses are very low, i.e., they are well above the bearing capacity of the soil. the maximum displacement data obtained from all the foundation cases under five different loading conditions is plotted in fig. 8. the foundation system with the straight layout and six piles has revealed the best resistance against the overall settlement of the foundation system. the hybrid foundation with three staggered piles has a consistent settlement with slightly higher values 0.00 5.00 10.00 15.00 20.00 25.00 30.00 35.00 0 50 100 150 200 250 300 d is p la c e m e n t u ( m m ) applied load (kpa) maximum foundation displacement (u max) straight staggered group 0.003 0.007 0.019 0.051 0.137 0.370 1.000 2.700 7.290 19.683 0 50 100 150 200 250 300 s tr e ss p ( m p a ) applied load (kpa) peak stress at pile-raft joint (p max) straight staggered group proceedings of engineering and technology innovation, vol. 16, 2020, pp. 13 22 20 than that of straight pile layout. the group pile system has shown the maximum settlement. the load settlement curve has the same pattern for all three cases. the primary reason for a higher settlement in the group pile system is the self-weight of the system. in the hybrid foundation system, the pile raft joint plays a very significant role, stresses at this location are very critical and governs the overall safety of the building. the maximum stress developed at the pile-raft joint in the foundation system is selected for each case. these stresses are then plotted for all three cases against five loadings in fig. 9. the stresses in the group pile system were very low as compared to the staggered and straight foundation systems. therefore, stresses (dependent variable) in fig. 9 were plotted on a natural logarithmic scale. if staggered and straight systems compared to the joint stresses in later are slightly higher, the low stresses in the group pile system are due larger supporting cross-sectional areas. the hybrid foundation system with group piles (case-3) has displayed the maximum settlement. the deformed shape of the foundation system is presented in fig. 10. the distortion (sagging) of the raft and pile bending can be clearly seen in the fig. 10 fig. 10 deformed shape for case 3 the volume of reinforced concrete used in the foundation system plays a very important part in the overall project economy. the volume of concrete required for straight, staggered, and group foundation system was respectively 456.55, 428.27 and 484.82 m 3 in fig. 11. since the pile diameter and length is the same in all cases, the volume difference is the numerical reflection of the pile number. the hybrid foundation system with staggered piles is the most economical one, whereas the foundation system having group piles is relatively extravagant. fig. 11 concrete volume use in different foundation systems 6. conclusions the hybrid foundation system can be effectively applied to the buildings resting on loose granular soil. the system distributes the load over the larger surface area of the soil and utilizes friction, bearing capacity, and tip resistance. the system requires fewer piles, as compared to the conventional pile foundation system. various pile layouts can be adopted in the hybrid 400 425 450 475 500 1 2 3 v o lu m e o f c o n c re te ( m 3 ) cases concrete volume for hybrid foundation system straight staggered group proceedings of engineering and technology innovation, vol. 16, 2020, pp. 13 22 21 foundation system; however, the staggered pile layout is most effective and economic. pile foundations placed in subgroups can be applied to reduce the stresses in the pile raft joint. nevertheless, this type of combination is not only uneconomical but is relatively risky. the combination can be used for heavily loaded structures, where stresses at the pile raft joint are critical. the excessive settlements can be controlled either by soil improvement or foundation design modifications. this research was carried out on loose granular soil of madinah saudi arabia. on the other hand, results can be applied to other comparable geographic regions. the performance of hybrid foundation, under fluctuating groundwater conditions, will be investigated as a part of future research plan, precise modeling of partly saturated conditions will be a potential research challenge and laboratory investigations will be planned to reinforce the numerical study. the effect of pile material, cross-sectional and longitudinal geometry on pile quantity of a hybrid foundation system is suggested as a prospective research topic. conflicts of interest the authors declare no conflict of interest. references [1] m. abuhjeeleh, “rethinking tourism in saudi arabia: royal vision 2030 perspective,” african journal of hospitality, tourism, and leisure, vol. 8, no. 5, pp. 1-16, 2019. [2] a. a. farag, “the story of neom city: opportunities and challenges,” new cities and community extensions in egypt and the middle east, springer press, june 2018, pp. 35-49. [3] g. f. brown, d. l. schmidt, and a. c. huffman jr., “geology of the arabian peninsula: shield area of western saudi arabia,” us geological survey, 1989. [4] j. b. burland, m. c. burbidge, e. j. wilson, p. r. vaughan, c. r. i. clayton, p. r. filho, et al., “discussion. settlement of foundations on sand and gravel.,” proc. of the institution of civil engineers, december 1986, pp. 1625-1648. [5] j. he, j. chu, and h. liu, “undrained shear strength of desaturated loose sand under monotonic shearing,” soils and foundations, vol. 54, no. 4, pp. 910-916, july 2014. [6] m. t. naqash, q. u. farooq, and o. harireche, “seismic evaluation of steel moment resisting frames (mrfs)—supported by loose granular soil,” open journal of earthquake research, vol. 8, no. 2, pp. 37-51, may 2019. [7] j. berrill and s. yasuda, “liquefaction and piled foundations: some issues,” journal of earthquake engineering, vol. 6, pp. 1-41, 2002. [8] g. madabhushi, j. knappett, and s. haigh, design of pile foundations in liquefiable soils, world scientific, 2009. [9] t. thavaraj, w. l. finn, and g. wu, “seismic response analysis of pile foundations,” geotechnical and geological engineering, vol. 28, no. 3, pp. 275-286, march 2010. [10] w. l. finn, “a study of piles during earthquakes: issues of design and analysis,” bulletin of earthquake engineering, vol. 3, no. 2, pp. 141, january 2005. [11] h. g. poulos, “piled raft foundations: design and applications,” geotechnique, vol. 51, no. 2, pp. 95-113, march 2001. [12] h. g. poulos, j. c. small, and h. chow, “piled raft foundations for tall buildings,” geotechnical engineering journal of the seags & agssea, vol. 42, no. 2, pp. 78-84, june 2011. [13] m. majima and t. nagao, 14. behaviour of piled raft foundation for tall building in japan, thomas telford, 2000. [14] b. hor, m. j. song, m. h. jung, y. h. song, and y. h. park, “a 3d fem analysis on the performance of disconnected piled raft foundation,” japanese geotechnical society special publication, vol. 2, no. 34, pp. 1238-1243, 2016. [15] d. d. c. nguyen, s. b. jo, and d. s. kim, “design method of piled-raft foundations under vertical load considering interaction effects,” computers and geotechnics, vol. 47, pp. 16-27, august 2012. [16] m. y. fattah, m. a. yousif, and s. m. al-tameemi, “effect of pile group geometry on bearing capacity of piled raft foundations,” structural engineering and mechanics, vol. 54, no. 5, pp. 829-853, 2015. [17] astm committee d-18 on soil and rock, standard practice for classification of soils for engineering purposes (unified soil classification system), astm international, 2006. [18] m. color, munsell soil color charts , 1975. proceedings of engineering and technology innovation, vol. 16, 2020, pp. 13 22 22 [19] m. cubrinovski and k. ishihara, “maximum and minimum void ratio characteristics of sands,” soils and foundations, vol. 42, no. 6, pp. 65-78, 2002. [20] “comsol,” https://www.comsol.com/documentation, 2014. [21] j. f. labuz and a. zang, “mohr–coulomb failure criterion,” rock mechanics and rock engineering, vol. 45, pp. 975-979, july 2012. [22] american concrete institute, building code requirements for structural concrete (aci 318-95) and commentary (aci 318r-95), the institute, 1995. 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/).  proceedings of engineering and technology innovation, vol. 1, 2015, pp. 45 49 45 cop y right © taeti the study on the strategies of choosing music schools from the viewpoint of kansei engineering tian-wei sheu1, ching-tzu chen1, jia-tzer hsu2,*, nagai masatake1 1 graduate institute of educational measurement and statistics, national taichung university of education, taichung, taiwan . 2 department of computer science and information engineering, hung kuang university, taiwan. received 01 april 2015; received in revised form 10 may 2016; accept ed 03 june 2015 abstract this research is to apply the kansei engineering to the study of choosing music schools. the purpose is to help parents select the perceptual characteristics and priorities of env ironment elements which are suitable for children to learn music. in this study, the combinations of nagai method (5w1h), ahp (analytical hierarchy process), ism (interpretive structural modeling) and gra (grey relational analysis) are used to clarify the development characteristics of music schools, and establish feasible options and assessment models. it is hoped to provide good references for the parents who wish to let their children engaged in music education. ke ywor ds : kansei engineering, music school, nagai method, ahp, ism, gra 1. introduction music education has always been regarded as the basis of the character education. the considerations of choosing music schools are issues of concern to students and parents, because music schools focus on the cultivation of musical talents and the enhancement of music literacy, and offer age-appropriate and talent-appropriate teachings to those students having interests or development potentials in the stages of elementary, junior, and high school in order to lay a good foundation for domestic music gifted talents and facilitate a broader space for development. the establishment of the music classes is able to not only maintain the requirements of general school music education, but also provide the high-quality learning environment of music classes for children to achieve the age-appropriate and talent-appropriate development potentials and develop the ability to enjoy music. the main purposes of this research are how to improve the quality and value of high-quality music schools and how to find the most appropriate selection which best meets the children’s interests and development potentials from those music schools. 2. research methods music education will often take two different stances, one focuses on nurturing play-designed performers or creation-designed creators; the other is to advocate children’s preferences and develop their music literacy mainly along their interests. in this study, the combination of ahp and ism is used to investigate the characteristic elements of the music schools, provide the selection strategy references for parents, and finally work out the possible options. 2.1. first stage: collection for the basic elements of vocabularies of the music schools we first conduct the search and collection of music-school-related vocabularies and kansei-engineering-related literatures, then conduct initial screening for such vocabularies using the kj method and the 5w1h method. 2.2. second stage: extraction and convergence for t he basic elements of vocabularies of the music schools after comparing the basic elements of vocabularies of the high-quality music schools with those of the general music schools, we then extract the more important kansei elements of vocabularies. 2.3. third stage: establishment of the correspondent rel ations among the basic elements of vocabularies of the music schools we obtain the main selection criteria using the cluster analysis approach. when a number of criteria are produced, we use the ahp method to conduct the assessment of the importance of the criteria, and then find the sorting of the design criteria [8]. 2.4. fourth stage: establishment of the strategies for selection of the music schools through the calculation of the ism software, we can get the hierarchical diagram of the ism structure model [9]. as shown in figure 1, the vocabularies located at the top are the highest goals for development of the music schools, while those located at the bottom are the basic elements of vocab ularies for development of the music schools. 3. results according to the cluster states of the basic elements of vocabularies of the music schools and the relations among those vocabularies, the hierarchical diagram in fig . 1 can be divided into four hierarchies. they are described as follows: 3.1. hierarchy 1: resource properties of the music schools this hierarchy is to mainly assess the affected software and hardware structure of the music schools because of the high-quality environmental learning space, resources available and other conditions, i.e., the basic level that is affected by the factor of resource properties. *corresponding aut hor, email: jt h@sunrise.hk.edu.tw proceedings of engineering and technology innovation, vol. 1, 2015, pp. 45 49 46 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti fig. 1 the hierarchical diagram of the ism structure model 3.2. hierarchy 2: sk ill of the music education this hierarchy is to mainly illustrate that the basic skills of the music education will affect the performance of external competitions. 3.3. hierarchy 3: appreciation ability of the music educ ation because of the learning of skills of the music education, the music education is no longer limited to the skill practice, while the music appreciation is integrated into the music curriculum design which emphasizes on listening to music, appreciating the importance of music, and encouraging students to have more exposure to music and participate in music show. those are to prompt the students to have the music appreciation capabilities, and then have a more beautiful lifestyle and music literacy. 3.4. hierarchy 4: creation of the music education the integration of this hierarchy will be able to highlight the significance and value of the music school history, and emphasize on the effectiveness of the music cultivation education and the forward-looking of the school overall development. according to the bottom-up structure shown in fig. 1, this paper prepares three design paths which are feasible options to choose music schools. as shown in fig. 2, the three paths are encoded from left to right as path 1, path 2, and path 3. in order to explore the best selection strategy (path) to choose music schools, we first conduct structure analysis for the lexical content of these three paths, and then understand the existing structural meaning of each chosen path. the analytic explan ations for the three paths are described as follows: 3.5. case i: on the left path of fig. 2 (path 1) this option is defined as “the structure type of cultivation of music performers or musical creative talents.” it is a long way to cultivate the music experts and creative talents. in addition to students with the advantages of congenital development potential, it also needs good teachers to guide the practice of music skills, as well as to guide students' musical performance economy convenience performance resource diversity originality performance variability cooperation comfort effectiveness forward-looking creation diversification hierarchy 1 hierarchy 2 hier archy 3 hierarchy 4 generality timing space appreciation fluency skill teacher http://cdict.info/wwwcdict.cgi?word=generality proceedings of engineering and technology innovation, vol. 1, 2015, pp. 45 49 47 cop y right © taeti ability and appreciation of the music. these elements are indispensable conditions. particularly, the music creation of the lyricist and composer needs even more inspiration and a solid musical literacy skill in order to create a classic work. to promote the cultivation of music creative talents and highlight the important characteristics of the selection strategy from the structure, teacher, diversity, skill, fluency and appreciation become the important demands of the high -quality music schools. from the interpretation of the vocabulary structure of this chosen path, the high-quality music schools can attract more students who will be engaged in music creation by emphasizing high-quality teachers. through the influence of music appreciation, it is hoped to cultivate creative musicians or performers and push the uniqueness of the music schools to its value target. 3.6. case ii: on the middle path of fig . 2 (path 2) this option is defined as "the structure type of using the resources of music classes to show musical performance ability." the high-quality music schools have great relations with the school environment and space resources. this option emphasizes on adequate equipments and the comfort of learning environment, and is mainly to stress the material factors. 3.7. case iii: on the right path of fig . 2 (path 3) this option is defined as "the structural type of musical literacy including both personal preference and team cooperation." the music schools, which focus on the music education cultivation, emphasize on personal preference, learning interest and the importance of team cooperation. this option is mainly to stress the human factors. according to the four divided hierarchies shown in fig . 1, this paper carries out the expert assessment of the design criteria and identifies the sorting of important criteria. in accordance with the sorting of hierarchies one to four, they are coded as follows: (a), (b), (c), (d), then the results of the assessment information are written to obtain the matrix shown in table 1. the weight value (weight) is obtained from the calculation of the ahp software, and the c.i. value is less than 0.1. to co n fig. 2 the path chart of the selection strategy of choosing music schools economy convenience performance resource diversity originality performance variability cooperation comfort hierarchy 1 hierarchy 2 hierarchy 3 hierarchy 4 generality timing space diversification appreciation fluency skill teacher effectiveness forward-looking creation path 1 path 2 path 3 http://cdict.info/wwwcdict.cgi?word=generality proceedings of engineering and technology innovation, vol. 1, 2015, pp. 45 49 48 cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti cop y right © taeti firm the validity of the information, the consistency test is passed. the sorting of the four hierarchies criteria are (d) > (c) > (b) > (a). in 1989 deng proposed the grey system theory. this study aims to use grey relational analysis by applying the data into matlab grey system for the establishment of standard sequence and comparative and sequence calculate the weight values and the sequence in new product design [1] [2]. the grey relational analysis theory can be used to manage the uncertain, multi-dimensional, discrete, and incomplete data. the main function of gra is to calculate the discrete data and quantify the factors. through the ordinal process, the info rmation can be translated and there are many successful researches [3] [4] [5] [6]. the researchers further select path 1 to 3 to conduct computation of the gray associated degree matrix, that is, the best-fit selection strategy associated computing matrix. the assessment results at this stage are shown in tables 2 to 5. the c.i. values in these tables are all less than 0.1, passing the consistency test of assessment survey. the assessment matrix in tables 2 to 5 are then used to conduct the grey relational degree calculation, and the best-fit option operational matrix is shown in table 6.the study results show that case ii, the structure type of using the resources of music classes to show musical performance ability, is the best-fit selection strategy in this study. 4. conclusion in this study, the combinations of nagai method (5w1h), ahp, ism and gra are used to establish selection programs and assessment models for choosing music schools. during the study, we use the expert survey assessment method which becomes an indicator for the research structure model, because the efficiency is high in clarifying the survey data through the analysis of the data information. through the difference of lgra coefficient values, we find the best option is "the structure type of using the resources of music classes to show musical performance ability." the results of this study can provide parents with more comprehensive considerations to choose music schools . table 1 selection criteria paired comparison matrix (a) (b) (c) (d) weight (a) resource 1 1 1/3 1/5 0.11 (b) skill 1 1 1 1/3 0.17 (c) ap p reciation 3 1 1 1 0.29 (d) creation 5 3 1 1 0.43 c.i. = 0.062 table 2 assessment table for (a) resource criterion (a) resource criterion case (1) case (2) case (3) gamma val u e larger-the-better 3 1 5 (1) music p erformers or musical creative talents 1 1/3 1 0.0046 (2) using the resources of music classes to show musical p erformance ability 3 1 5 1.0000 (3) musical literacy includin g both p ersonal p reference and team coop eration 1 1/5 1 0 c.i.=0.015 table 3 assessment table for (b) skill criterion (b) s kill criterion case (1) case (2) case (3) gamma val u e larger-the-better 1 3 5 (1) music p erformers or musical creative talents 1 3 5 1.0000 (2) using the resources of music classes to show musical p erformance ability 1/3 1 2 0.2335 (3) musical literacy includ in g both p ersonal p reference and team coop eration 1/5 1/2 1 0 c.i.= 0.002 table 4 assessment table for (c) appreciation criterion (c) appreciation criterion case (1) case (2) case (3) gamma val u e larger-the-better 3 2 1 (1) music p erformers or musical creative talents 1 1/2 1/3 0 (2) using the resources of music classes to show musical p erformance ability 2 1 1/2 0.4205 (3) musical literacy includin g both p ersonal p reference and team coop eration 3 2 1 1.0000 c.i.= 0.005 table 5 assessment table for (d) creation criterion (d) creation criterion case (1) case (2) case (3) gamma val u e larger-the-better 1 5 3 (1) music p erformers or musical creative talents 1 5 3 1.0000 (2) using the resources of music classes to show musical p erformance ability 1/5 1 1/2 0 (3) musical literacy includin g both p ersonal p reference and team coop eration 1/3 2 1 0.2335 c.i.= 0.002 table 6 the best-fit option operational matrix c ri te ri a (a) (b) (c ) (d) lgra val u e larger-t he-bett er 1 1 1 1 (1) case i (path 1) 0.0046 1.0000 0 1.0000 0.8912 (2) case ii (path 2) 1.0000 0.2335 0.4205 0 1.0000 (3) case iii(path 3) 0 0 1.0000 0.2335 0 s orti n g re su l t (2)＞(1)＞(3) references [1] d. yamaguchi, g. d. li and m. nagai., “new grey relational analysis for finding the invariable structure and its applications,” journal of grey system, vol. 8, no. 2, pp. 167-178, 2005. [2] d. yamaguchi, g. d. li and m. nagai., “verification of effectiveness for grey relational analysis models,” journal of grey system, vol. 10, no. 3, pp. 169-182, 2007. [3] j. c. liang, j. r. wang and l. h. wang, “kansei product design based on the personal’s hair image,” international journal of gray system, vol. 14, no. 1, pp. 29-40, 2011. [4] j. c. liang, t. w. sheu, b. t. wang, j. w. tzeng and n. masatake, “the study of product structure integrates kansei design evaluation identification on creation of proceedings of engineering and technology innovation, vol. 1, 2015, pp. 45 49 49 cop y right © taeti new products,” international journal of taiwan kansei information, vol. 2, no. 1, pp. 27-38, 2011. [5] j. c. liang, y. l. lee and h. j. weng, “design strat egies of household tea tables with glass based on gra,” international journal of gray system, vol. 13, no. 3, pp. 91-96, 2010. [6] j. c. liang, y. l. lee and s. f. liu, “strategic kansei design for a nice doorplate based on gra,” international journal of gray system, vol. 12, no.4, p p. 177-184, 2009. [7] j. l. deng., “introduction of grey system theory,” journal of grey system theory, vol. 1, no. 1, pp. 1-24, 1989. [8] t. l. saaty, the analytic hierarchy process, new york: mcgraw-hill, 1980. [9] warfield j. n., interpretive structural modeling (ism) group planning & problem solving methods in engineering, new york: wiley, 1982.  proceedings of engineering and technology innovation, vol. 10, 2018, pp. 08 12 a study on the development of game-based mind wandering judgment model in video lecture-based education jaechoon jo, yeongwook yang, heuiseok lim * department of computer science and engineering, korea university, seoul, korea received 02 march 2018; received in revised form 30 may 2018; accepted 16 july 2018 abstract although video lecture materials are very efficient learning materials, they are likely to be unilateral learning materials by the lecturer. it is easily degraded to be one-sided learning, which has been considered as a problem of online education, and it is difficult to judge whether learners are actually learning. therefore, in this paper, a minimum learning activity judgment model that can automatically determine if they actually learn through mind wandering judgment was proposed to overcome the limitations of previous learning materials, and educational effect verification experiment was performed. experiment results show that the video lecture class using the minimum learning activity judgment system was effective in improving the academic achievement. keywords: online education, e-assessment, flipped classroom, judge system, minimum learning activity 1. introduction education has taken a path of computerization, as industry and technology have developed and the society has been computerized. beginning with the remote education in the 1900s, e-learning using electronic media emerged in 2000 as the internet developed. e-learning has made many changes in the target, place, time, medium, and method of education. people can receive the desired education regardless of their age or status. they can have education at the place and time and with the medium they want. e-learning has increased the demand and satisfaction of many learners in a revolutionary way without the limitation of time and space. with the development of information technology, even college classes can be taken online. it has led to the creation of an environment like ‘massive open online courses (mooc)’ [1]. mooc is an online learning process aiming at the environment where anyone who wants to study can participate and openly access the web. mooc provides students with the opportunity to earn regular credits or degrees through e-learning lectures as well as to acquire the necessary knowledge [2]. however, the environment of e-learning can make a lack of communication between the teacher and leaners because it is a non-real-time and indirect learning environment comparing to face-to-face classes [3]. of course, there has been much effort in mooc to enforce communication between the teacher and learners stepping further from the unilateral remote education. however, in many online learning programs including mooc, the problems such as insufficient concentration, necessity to compose a self-study environment and lack real-time feedback still remain [3-5]. to overcome such problems, teachers have proposed online quizzes or tasks to learners in online education environment to induce concentration and judge if they have actually learned. however, such methods make teachers use more time and money and learners have additional burden. another automatic method to check if the learners have actually learned is video playback time. however, this method is only * corresponding author. e-mail address: limhseok@korea.ac.kr tel.: +82-2-3290-2396 proceedings of engineering and technology innovation, vol. 10, 2018, pp. 08 12 copyright © taeti 9 based on the playback time of the video lecture, so it is difficult to judge whether the learner is really learning or he/she has concentrated on the lecture [6]. many online education systems currently use pop-up events or online quizzes in the middle of video lectures to determine whether learners are watching the video lecture, or allow the learners to go to the next lecture only when the special event in the middle of the lecture is clicked to verify if they have actually participated in the learning process. however, those existing methods may interrupt the learning flow of learners, or may create ineffective events even in the parts which are not necessary to be learned. although video lecture materials are very efficient materials, they are likely to be unilateral learning materials by the instructor. it is because the video lecture is not based on the communication with the lecturer but a unilateral lecture of the lecturer. the lecturer has no means to check if the learning contents have been successfully delivered to leaners. the lecturer can check the learning status of students only when he/she asks questions to each student about the contents of the lecture or give them a quiz. however, it is very inefficient if the lecturer has to take pains to make questions and to mark in order to check whether they have learned from the video lecture. therefore, such judgment on learning status is rarely made. the concerns relating to online education such as likelihood to be unilateral learning and difficulty in the judgement of learning status should be supplemented using computing technology. accordingly, in this paper, the minimum learning activity judgment model to check the status of learning of learners was proposed to overcome the limitations of existing methods and an educational effectiveness test was conducted. 2. minimum learning activity judgment model in this paper, a minimum learning activity judgment model to automatically determine the learning status of learners through mind wandering judgment in video lecture based online education was proposed. the minimum learning activity is defined as the activity to watch a video without mind wandering. it is called minimum learning, as it does not matter how much the learners have understood what they watched. minimum learning activities focus more on learning efforts (learning activities) than on understanding or understanding the level of the learning contents. in order to automatically determine the minimum learning activity, it should be judged if there was mind wandering while they watched the video lecture. therefore, the word frequency effect theory of language linguistics is applied to determine mind wandering. the word frequency effect is a theory that there is a difference in cognitive responses between high frequency words and low frequency words, and there is also difference in cognitive responses between words that people actually know and words they do not know. the minimum learning activity judgment model determines the minimum learning activity by calculating the difference in the word-recognition reactions of the learners. to do this, a word game was developed. in other words, a minimum learning activity judgment system applying word game was developed. fig. 1 structure of word game fig. 1 shows the word game structure of the system. the word game is an o/x type game. it automatically extracts high frequency words from a video lecture that a learner has watched and mixes them with words that have never appeared in the video lecture. the learner will choose ‘o’ for the word that appeared in the video lecture, and ‘x’ for the word that has not appeared in the lecture. the system determines whether the minimum learning activity has occurred by calculating the number of correct answers and the recognition response time. proceedings of engineering and technology innovation, vol. 10, 2018, pp. 08 12 copyright © taeti 10 the minimum learning activity is judged as pass or fail through the number of correct answers and recognition response time. according to language recognition discipline, it takes 0.7 to 1.2 seconds when people see a word that they know. this was applied to the response time to judge the recognition. as the word game in this paper was implemented in a web environment, +0.4~0.5 second was given as tolerance considering the web loading environment. accordingly, 0.3~1.7 seconds was set as recognition response time. in other words, if a leaner chose the right answer within the response time, it is considered that the learner has watched the video lecture without mind wandering. in order to set the standard score to judge the minimum learning activity, the minimum learning standard score was obtained through an experiment. sixty domestic college students participated in the experiment. participants watched the video lecture and checked whether they were focused on a 5-point scale. after watching the video lecture, they were divided into two groups using word game results. those who had 11 correct answers out of 14 (80% correct answers) were classified as the group that focused on the video lecture, and those who had 7 correct answers out of 14 (50% correct answer) were classified as non-focused group. in other words, the hyper-parameter of the judgment standard score was set as 0.8 based on the experimental result. 3. experiments and results to investigate the educational effect, this experiment was conducted on 123 5th grade elementary school students for 5 weeks. the experimental group consisted of 62 students (30 males and 32 females) and the control group consisted of 61 students (30 males and 31 females). the homogeneity of two groups was investigated through the result of the pre-test comparing academic achievements of students in two groups. the results of four students in the control group who did not take the academic achievement test were excluded. to compare the academic ability of the experimental group and the control group, korean, english and math questions were presented and they were asked to take the test under the same conditions. in order to verify the difference of academic ability between groups, the independent t test was conducted. in the t-test, there was no significant difference between groups in korean (t = -0.602, df = 113, p = 0.548), mathematics (t = 0.239, df = 113, p = 0.812) and english. (t = 1.963, df = 82, p = 0.053. in english, the equal variance assumption was significant, but it was not significant with the corrected t value (t=1.963, df = 82, p = 0.053). the total score of the three subjects, which can be regarded as the average academic ability, was not statistically different between the experimental group and the control group (t=0.799, df = 113, p = 0.426). as shown in table 1. before the experiment, they took the academic achievement test. after the pre-test, the system using the minimum learning activity judgment was administered to the experimental group and the conventional learning system using the traditional online video lecture was administered to the control group (no word game, no learning judgment function). in order to investigate the change in academic achievement during the experimental period for 5 weeks, they had interim and post-test at the third week and the fifth week. as an academic achievement test tool, diagnostic evaluation of the elementary school curriculum was used. table 1 result of academic achievement group mean sd n pre test experimental 67.290 14.8880 62 control 67.123 16.8820 57 total 67.210 15.8065 119 mid test experimental 85.774 12.3081 62 control 76.193 13.2953 57 total 81.185 13.6127 119 post test experimental 86.484 11.8719 62 control 68.298 16.5583 57 total 77.773 16.9178 119 proceedings of engineering and technology innovation, vol. 10, 2018, pp. 08 12 copyright © taeti 11 there was a statistically significant difference in academic achievement between the groups (f(1,117)=16.997, p<0.001). there was no statistically significant difference between the control group and the experimental group in the pre-test (f(1,117)=0.003, p=0.954), but both groups showed statistically significant difference in interim and post-test (f(1,117)=16.665, p<0.01, f(1,117)=47.977, p<0.01). in the experimental group, there was a statistically significant difference between the preand interim and between preand post-test (f(1,117)=16.665, p<0.01, f(1,117)=47.977, p<0.01), but there was no statistically significant difference between interim and post-test (f(1,61)=0.393, p=0.533). in control group, there was a statistically significant difference between the preand interim and between interim (f(1,56)=25.766, p<0.01, f(1,56)=17.101, p<0.01), but there was no statistically significant difference between preand post-test (f(1,56)=0.278, p=0.60). 4. conclusions this study proposed a minimum learning activity judgment model to judge the learning status of learners through mind wandering in the video lecture-based education. to administer a comparative experiment of educational effectiveness, video lecture learning was carried out using the system developed for the 5th grade of elementary school students for 5 weeks. additionally, academic achievement tests were given as pre -, interim, and post-test. experimental results showed that the video lecture class using the minimum learning activity judgment system was effective on the improvement of the academic achievement. in the case of the experimental group, the improvement between the pre-test and the interim test, and between the pre-test and post-test was statistically significant. in the case of the control group, the improvement between pre-test and interim test and between interim test and post-test was statistically significant, but the improvement between the pre-test and the post-test was not significant. the academic achievement of the two groups was similar in the pre-test but there became significant difference from the interim test (the achievement result at the 3rd week evaluation). especially, in the academic achievement tests, the control group showed score drop between the interim test and the post test. on the other hand, the experimental group showed a continuous increase. given that the difficulty of learning increases over time from the pre-test to post-test, it can be considered that the minimum learning activity system has not only short-term and simple effect but also long-term effect on learners. in addition, according to the interviews with learners, the experimental group students who used the minimum learning activity judgment system did focus more on the video lecture for the word game. and they had interest and tension thinking of their ranking in the game after playing the game. it seems that these interests and tensions have positive effects on learning, as keller's arcs theory. this concentration leads to the concentration of learning time. the result of judgment if students have actually learned enables the lecturer to give them fast and efficient feedback. this can be thought of as enhancing learning achievement. in the control group, they showed improvement until the interim test as the introduction of the new method made them concentrate at the beginning of the experiment. however, in the long term, the effect of the method administered to the control group is negligible. in other words, learners can concentrate on video lecture and the class contents more, when a video lecture is applied with follow-up treatment rather than a simple video lecture without such treatment. it can be considered that the minimum learning activity judgment system can give learners satisfaction with learning by changing the characteristic video lecture from unilateral delivery to be-directional interaction. acknowledgement this research is supported by ministry of culture, sport and tourism(mcst) and korea creative content agency(kocca) in the culture technology(ct) research & development program 2018. (no. r2016030031) proceedings of engineering and technology innovation, vol. 10, 2018, pp. 08 12 copyright © taeti 12 references [1] d. f. o. onah and j. sinclair, “learners expectations and motivations using content analysis in a mooc,” edmedia 2015 world conference on educational media and technology, vol. 2015, no. 1, june 2015, pp. 185-194. [2] c. j. bonk, m. m. lee, t. c. reeves, and t. h. reynolds, moocs and open education around the world, united kingdom: routledge, 2015. [3] j. jang, i. ko, and c. jeong, “the impact of e-learning system characteristics on learner commitment and satisfaction,” journal of information technology applications & management, vol. 15, no. 1, pp. 99-116, 2008. [4] d. s. zhang, l. zhou, r. o. briggs, and j. f. nunamaker jr. “instructional video in e-learning: assessing the impact of interactive video on learning effectiveness,” information & management, vol. 43, no. 1, pp. 15-27, january 2006. [5] m. t. h. chi, “active‐constructive‐interactive: a conceptual framework for differentiating learning activities,” topics in cognitive science, no.1, no. 1, pp. 73-105, january 2009. [6] p. pham and j. t. wang, “attentivelearner: improving mobile mooc learning via implicit heart rate tracking,” proc. international conf. artificial intelligence in education, springer press, june 2015, pp. 367-376.  proceedings of engineering and technology innovation , vol. 4, 2016, pp. 19 21 19 a way to introduce innovative approach in the field of microelectronics and nanotechnologies in the chinese education system olivier bonnaud 1,2,3,* , lei wei 3 1 department of microelectronics and microsensors, ietr, university of rennes 1, rennes , france. 2 national coordination for education in microelectronics and nanotechnologies, grenoble, france . 3 department of electrical engineering, south -east university, nanjing (jiangsu), china. received 22 february 2016; received in revised form 21 march 2016; accept ed 10 april 2016 abstract the innovation is more and more useful in order to develop new products such as the connecting objects. the field of microe lectronics is particularly concerned by them since it co mbines the fast evolution of the technology, and their application to many domains. the educ ative system in china is ma inly based on me morization o f knowledge: the students are lac king of methodology, understanding and practice. in the same time, the pedagogy in higher education moves progressively towards a numerical approach. in this context, the ma in challenge is to give to the students, the methodology, the know-how, and thus an innovative behaviour. in this way, several suggestions are given: introduction of much more works on applications of the theoretical knowledge, tra ining on the lin king of knowledge, and imp rovement of the practice by laboratory works on dedicated pla tforms. keywor ds : innovation, higher education, microelectronics , practice and labworks. 1. introduction the economica l world encourages more and more innovation in order to develop new products that should answer to the future societal needs, more especially the connecting objects. the fie ld of microelectronics is particularly concerned because these objects combine the strong improvement of the microe lectronics technologies , and the application to many d oma ins. in the sa me time, the pedagogy in higher education shifts towards a numerica l world by involving tools based on massive open online courses (moocs), internet sites, modelling and simu lation. the ma in challenge is to g ive to the future engineers, the methodology and the know-how, wh ich is not enough provided nowadays, and thus an innovative approach. the pedagogical approach in ch ina, fro m the prima ry school to the university, is based on the learning of knowledge, main ly. the evaluation and the selection are based on the me morization as well. several weaknesses are revealed. this paper highlights these points and suggests several improve ments in order to give to the future engineers an innovative behavior in the fie ld of microelectronics . 2. situation of the higher education in china the previous experience in china of the authors shows that the basic education presents several weakness es in the pedagogical methodology. on the one hand, in the learning of basic knowledge since the primary school, the pupils, then the schoolers, and finally the students are spending a lot of time learning by heart a difficult language, and, with the same approach, the scientific background (mathemat ics, physics, chemistry, etc.). on the other hand, they have a low a mount of practice, inc luding applications of theories through resolution of problems, and a very lo w a mount of laboratory works. but this practice is revealed very effic ient to assimilate more the methodology and the logical de mo nstrations than the theoretical knowledge. if the me mo rizat ion by heart is efficient to g ive to the * corresponding aut hor. email: olivier.bonnaud@univ-rennes1.fr proceedings of engineering and technology innovation , vol. 4, 2016, pp. 19 21 20 copyright © taeti students the good capacities of communication in their national language, however the students lack a general approach that allows a deep u nderstanding of a scientific fie ld, and also the link of knowledge between several fields. in the fra me of 1000 ta lents program of china’s govern ment, an analysis of the situation in the fie ld of m icroe lectronics raises additional weaknesses, specific to this field. indeed, due the very e xpensive equipment for devices and circuits, the students have in average a very poor know-how. in addition, because this domain has a permanent and very fast evolution, the practice can be always obsolete if an innovative strategy is not applied by the institutions and their ed ucational departments. these weaknesses give rise to some suggestions that can imp rove the efficiency of the acquisition of skills in this domain. 3. evolution of the technologies the very fast evolution of the microelec tronics integration since more than forty years is well known and is called the moore’s law [1]. this heuristic indicates the decrease of the transistor size as a time function thanks to the tremendous improvement of the technological fabrication processes and to the huge increase of the performances of the co mputer tools and associated software. these improve ments have later opened the doors to the system integration, called the “more than moore" evolution [ 2]. it corresponds today to an increase of the co mple xity by combining several technologies and the packaging capabilities , by mixing other functions than electrical ones , and by designing the ele mentary devices and the integrated circuits in three dimensions . this evolution induced new generic technologies such as mech atronics, optoelectronics, bioelectronics, systems -on-chips, large area electronics, etc. that combine skills in several fie lds and that are intrinsically mu ltidiscip linary [3]. indeed, these new concepts are already applied main ly to the sectors of health, environ ment, security, energy, agriculture, transport, and commun ications, which a re considered as the first priority for our future technological societies. thanks to the concomitant strong evolution of the commun ications, through internet, through wireless communicat ions such as wifi, zig bee, bluetooth, 3g, 4g, cloud, etc., it is predicted a fast evolution towards connecting objects in a ll the mentioned applications. many reports made by international e xpe rts highlight this evolution for the next ten years [4]. 4. innovation approach linked to microelectronics for the design of these connecting objects , the engineers in research and development centers must combine the knowledge and the know-how of several fields and must be innovative. this innovation may co me by the cap abilities of the designers to analyze the economical and societal needs , to make links between technologies, to extract the concepts of existing applications, to apply the selected concepts of one field to another one. all these qualities and know-how should be initiated during the studies . in other words, this means a high knowledge and the know-how that allows the control of the limitat ions of the systems, the definit ion of the adapted mission profiles, the control of the security of the equipment and of the effective application, and the reliability as well. this supposes during the studies a very good background in scientific disciplines, a good training including pract ice and laboratory works, and a mult idisciplinary approach including the capacity to establish links between the different fie lds (commun ications between engineers, understanding of the concepts, etc.). 5. expected evolution of the higher education many governments of developed countries try to develop the numerica l approach of many branches of the economy, and more especially the education one. thanks to the huge increase of the performances of the dig ital tools associated to the powerful co mputing capabilities of the hardware (microprocessors, dsp, etc.), to the size of the software, new educative tools can be created. they can contain texts, pictures, animations , videos , but also simu lations and modeling. if these tools can be really useful to ma ke easier the co mprehension of the basic knowledge, they appear insuffic ient to understand the methodology, to make links between knowledge and above all they rema in virtual. the functioning of connecting objects is often very different fro m the one of virtual objects . the main reason is the comple xity of the env ironment that can affect the functioning of a proceedings of engineering and technology innovation , vol. 4, 2016, pp. 19 21 21 copyright © taeti product, for e xa mple , effects of the temperature, the pressure, the chemica l environ ment, etc. the natural environment is so much comple x that it is very difficult to guarantee that a simulator takes all the para meters into account! thus a comple mentary practice is mandatory in order to well understand the limitat ions , the range of the physical para meters, and the mission profile . this practice is also a way to better e xpla in the fundamental background, to give a know-how in real situations that is a key approach to develop the skills of analysis, of synthesis, of ma king lin ks between several disciplines. this approach allo ws the building of an innovative behavior of the students. 6. suggestion of evolution of the chinese higher education many suggestions can be made in order to improve the innovative approach [5]. we mention those that can be the most efficient in the present situation. the first concerns the methodology of teaching that should not be mainly based on me morization but that could include problem solving, cases studies of real objects and their limitations of functioning. the second point is focused on the need of practice. in the field of microelectronics, due to the high costs of the software and of the technological facilities, the set-up of common platforms should give a solution. this approach was developed in france since many years in the frame of a national network [6-7]. t welve national centers receive more than 14.000 students per year in technological cleanrooms, on benches of physical and electrical characterizations, on platforms of modeling, simulation, and computer-aided design. moreover, an innovative strategy of the network is applied through annual projects in order to maintain an up-to-date training. 7. conclusions several weakness es were observed in the chinese education system: they seem to have an important impact on the learning of the mic roelectronics discipline. the introduction of new approaches that include much mo re analyses, understanding and acquisition of know-how should lead to a mo re e ffic ient lea rning adapted to the permanent innovation needed in this domain. acknowledgement this work is supported by the french ministry of education, the ch inese 1000 talents program, and the french excellence init iative program idefi (a nr-11-idfi-0017 (2012)). a special thanks to l. chagoya-ga rzon, secretary of gip-cnfm for her fru itful advice for the redaction of this paper. references [1] g. e. moore, “cramming more components onto integrated circuits ,” electronics magazine, vol. 38, no. 8, pp.114-117, 1965. [2] w. arden, et al. editors “more-thanmoore,” international roadmap committee of itrs, white paper, 2010. [3] o. bonnaud, l. fesquet, “multidisciplinary topics for the innovative education in microelectronics and its applications ,” proc. ieee intern. conf. on information technology based higher education and training (ithet’15), 11-13 june, 2015, pp. 1-5. [4] j. dokic, b. müller, g. meyer, “strategic res earch agenda of eposs,” http://www.smart-s ystems -integration.org/ public/documents/pub lications , april 2015. [5] o. bonnaud, “difference of pedagogical approaches for chinese and french master students in a french-chinese microelectronics joint master diploma,” proc. ieee intern. conf. on information technology based higher education and training (ithet’15), 11-13 june 2015, pp. 1-4. [6] o. bonnaud, p. gentil, et al., “gip-cnfm: a french education network moving from microelectronics to nanotechnologies ,” proc. ieee global engineering education conference (educon’11), 3-6 april 2011, pp. 122-127. [7] “national coordination for education in microelectronics and nanotechnologies,” http://w ww.cnfm.fr, may 2016.  proceedings of engineering and technology innovation, vol. 14, 2020, pp. 01 08 robotic control by teleoperation and delay time issues using the internet anuntachai machim * , murray john lawn, ikuo yamamoto department of mechanical science graduate school of engineering nagasaki university, nagasaki, japan received 14 january 2019; received in revised form 13 february 2019; accepted 14 march 2019 doi: https://doi.org/10.46604/peti.2020.3931 abstract internet usage has become an essential part of our daily lives, it has become universal. the aim of this research is to consider existing internet-based networks with regard to their support for the remote operation of robotic technology, in particular, looking at transmission speed and delay time as it relates to teleoperation. currently dedicated connections are typically used when a high quality of service (qos) is required for mission critical services or safety-critical systems (scs), however such connections are very expensive to set up and maintain. therefore, this study focuses on modifying existing internet networks in a manner to provide a better qos with little additional cost. using a number of microcontrollers, computers, and routers, two different protocols were used to control a robotic device remotely. use of the point-to-point tunnelling protocol (pptp) used for implementing a virtual private network (vpn) was found to provide higher average transmission speeds compared to the using a uniform resource locator (url) which is a regular internet connection. keywords: robotics, iot, internet, teleoperation, robot remote control, delay time, virtual private network, vpn 1. introduction from the infancy of the internet on october 29, 1969, the first ever electronic message was sent between two computers that were literally side by side with their only physical connection being through the arpanet, one of the first computer networks of the world [1-2]. the first message on the internet was “lo”, as in lo and behold! this technology has led to a technology explosion which continues to accelerate in many directions. one application of the internet is as a medium through which one can control things for example, service mobile robots are controlled remotely over any tcp/ip enabled network [3], some research has shown the control of devices (servo motors) via internet, with accuracy between 97% to 99% [4], some researchers have used field-programmable gate arrays (fpga) to interface robotic devices via the internet [5], a vpn has been used to prevent robot control data from being accessible by other computers on the internet, creating secure connections between the robot operation system and authorized remote clients [6]. a representational state transfer (rest) architecture has been used to provide interoperability between computer systems via the internet [7] in a similar manner to the use of php, sql in this study. this type of control using the internet has come to be generally referred to as the internet of things (iot). however, in regard to the sending and receiving of information, and the control of machines over the internet network in real time is problematic. in the case of controlling a robotic mechanism in real time, the issue of encountering loss of or the delay of data transmission is significant. this research focuses on correctin g and/or improving the efficiency of existing internet connections to minimize this data loss and delay. in this research, we used * corresponding author. e-mail address: anuntachai.nut@gmail.com proceedings of engineering and technology innovation, vol. 14, 2020, pp. 01 08 2 hardware and software that are current, commonly used and relatively easy to use, specifically the authors used a linux operating system (os), python, and arduinos. these are all open source ensuring good development support. transmission control protocol (tcp) internet protocol version 4 (ipv4) was used to transmit and receive data between the device's over the internet. 2. methods 2.1. hardware design regarding hardware design and the associated programming for such as a server, optimization of data handling to ensure efficient transfer of information or throughput is central. 2.1.1. communication between computer, server and microprocessor raspberry pi 3 model b (rpi3b) the authors chose the ssh (secure shell) protocol for communication between the computer, server and microprocessor. in addition, the authors used php: hypertext preprocessor language and mysql as an open source relational database management system (rdbms) for checking the status of the devices, connected by lan (local area network) and wan (wide area network) networks through a router. 2.1.2. communication between microprocessor and microcontroller arduino pro mini and micropython an isp (in-circuit serial programmer) was used to communicate with arduino pro minis to upload programs and i²c (inter-integrated circuit) for communicating with the raspberry pi and the arduino raspberry pi circuit schematics are shown in fig. 1 and 2. micropython was used to implement the uart bus to communicating with the raspberry pi. fig. 1 circuit for the data receiver 2.2. programming 2.2.1. server a “linus-unix” based operating system using debian 9 for programming was employed. ubuntu 18.04 lts (long term support) was used for the server, supported by linux mint, linux kali, and raspbian functionality. proceedings of engineering and technology innovation, vol. 14, 2020, pp. 01 08 3 2.2.2 control robot the main program for the arduino pro mini is written in arduino, with other support programs such as shell script, php, python3, and mysql. python2 allowed communication with arduino using the i 2 c and spi bus for data transmission between the rpi3b and arduino pro mini. the main micropython program was written in python3 to communicate with the raspberry pi using python2 for uart communication. fig. 2 circuit for data transmission fig. 3 overall system communications schematic 2.3. communication between microprocessor and microcontroller devices communication between the raspberry pi and arduino pro mini enables the raspberry pi to upload or edit arduino programs using the spi bus (fig 1 and 2). it communicates via the i²c bus, using arduino for data transmission between the arduino pro mini and the raspberry pi, using an ethernet port (rj45 port) to connect between the rpi3b and router for data transmission over the internet. using python 2 on the raspberry pi for sending data between the two devices using “import smbus” and “#include <wire.h>” function libraries respectively. two micropython uart ports were used. one for writing and editing programs and another one for data transmission. the raspberry pi used python 2 “import serial” and micropython used the “from machine import uart” command in python 3. 2.4. communication between the microprocessors using a raspberry pi 3 model b microprocessor, the authors used python to communicate between the two devices. by sending data via lan cable using “import socket” from the python2 library for transmission data. both devices use the same proceedings of engineering and technology innovation, vol. 14, 2020, pp. 01 08 4 data port but are assigned as a receiver and messenger as required. the receiver identifies itself as the recipient, (the device as shown in fig. 1 schematic and fig. 4 raspberry pi) by specifying tcp ip as localhost or specifying the ip address of the device received from router x, as shown in fig. 3. and specifing port as 5005 (fig. 5 receivers). the messenger will send the message to the target ip, or in this case the ip of the receiver (fig. 5 messengers). in this case, the messenger has successfully connected to the vpn. the ip of the receiver was 192.168.12.198. the ip of the messenger was 192.168.12.x, where x was the number assigned by the pptp vpn server [8] , so it is as if both are on the same lan. from this point, both can communicate, despite being remotely located. fig. 4 testing equipment messengers receivers (a) sending data (b) receiving data fig. 5 the main programs for sending and receiving data from both microprocessors sending information on the url using php and python2 is illustrated in fig. 6. both microprocessors use python2 “import urllib” and the “urllib.urlopen” to link the url, then use php to connect the data to the database (sql server). here, it was necessary for microprocessor x to read the data in a continuous loop. fig. 6 data transmission test on the url and sql server 2.5. system test the system test configuration schematic is shown in fig. 3. where flex sensor data is generated by microcontroller y, processed by microprocessor y and transmitted to the internet via router y, the circuit is shown in detail in fig. 2. router x has a public ipv4 or ddns. data reception is provided by router x for processing by microprocessor x for actuator control proceedings of engineering and technology innovation, vol. 14, 2020, pp. 01 08 5 via microcontroller x, of which a detailed schematic is shown in fig. 1. data storage and monitoring are provided by the sql server, microprocessors x and y send data to the database via the url using php onto the mysql database [9-11] and the vpn server functionality is provided. first, both microprocessors x and y will send ipv4 values to the sql server to check the device's operating status [12]. secondly, microprocessor y will change the ipv4 status to emulate the same gateway as microprocessor x at the same time microprocessor x will check its status. finally, both microprocessors will connect as if they were on the same gateway and will be able to communicate using a shell script. to test the system (remotely controlled robotic hand) the authors used a mac os terminal to simulate microprocessors x and y. the authors used a web browser on the mac that is compatible with the “raspberry pi zero w”. the raspberry pi was equipped with a “raspberry pi camera” and used the “motion” program running on the native raspbian os. the authors used the “terminal” on the mac os system to execute and check the data of both devices (fig. 7). here, microprocessors x and y functionality is simulated by a single physical device (mac os), however, routers x and y are separate routers connecting independently to different public ipv4 addresses (wan). fig. 7 remotely controlled robotic hand test 3. results (a) mean delay time (b) minimum delay time (c) maximum delay time (d) standard deviation fig. 8 transmission data delay time between devices fig. 8. shows the transmission data delay between devices, clearly, in the case of a shared internet connection significant variation of speed will occur based on usage and is reflected in the maximum and minimum delay time variation. however, the proceedings of engineering and technology innovation, vol. 14, 2020, pp. 01 08 6 standard deviation values are small, indicating the minimal variation of the mean delay time. the minimum delay time indicates the maximum data transmission speed possible. from this data transmission test thailand to japan, it can be seen that the vpn has an average speed that is faster compared with transmission via php, sql in the magnitude of 1.45 times mean delay time and increased qos [13]. but php, sql does have the ability to send data faster based on the minimum delay times, in the case of sending information, from thailand to japan (for php,sql the server is in thailand which uses ipv4 unlike microprocessors x and y). the transmission times were measured by setting up databases in both the sending and receiving microprocessors (rpi3b), the above times were based on network time protocol (ntp) of rpi3b (debian.pool.ntp.org). data trials were carried out by transmitting 24 bits at one minute intervals for one week. (a) information received from the device located in thailand (b) information sent from the device located in thailand fig. 9 random data transmission experiments using vpn from fig. 9, the data sent from thailand to japan (time zone variation of 2 hours), shows that the random data sent was received without any errors. fig. 10 shows how bandwidth affects transmission delays. clearly, reduced bandwidth will result in greater data transmission delays. however, it can be seen that a bandwidth of more than 60 kbps (upload and download) did not result in further significant reductions of delay time. regarding bandwidth on a shared network, while maximum transmission speeds may be engineered, however, on account of the shared nature of the networks, actual bandwidth performance will be somewhat random. this is also most likely the reason for the lack of correlation between minimum delay time and bandwidth in the fig. 10 minimum delay (instantaneous peak) values. while there are numerous factors that effect delay times. for example, some routers may check regular url traffic but pass vpn traffic unchecked. based on the above results, communication between devices is more efficient as vpn usage results in faster average data transmission speeds of 1.45 times, reduced mean delay time thailand to japan and minimal average data transmission delays. using vpn also offers many other advantages, such as making it simpler to edit programs and provide for remote system management. however, data transmission by vpn connection and url are both affected by network distance and bandwidth. as shown in fig. 9, the delay time is directly affected by the type of data transmission, that is vpn or url. fig. 10 shows that the bandwidth directly affects the data transmission speed between devices. proceedings of engineering and technology innovation, vol. 14, 2020, pp. 01 08 7 (a) mean delay time (b) maximum delay time (c) minimum delay time (d) standard deviation fig. 10. bandwidth controlled data transmission test 4. conclusion this research focused on the selection of a protocol that would optimally suit remote control of such as robotic devices. based on the results of the experiment, it can be concluded that using a vpn provided higher average data transmission rates, although using a regular url resulted in minimum (instantaneous peak) delay time. factors that impact internet transmission speeds are many and varied, such as the pathway taken by specific packets on an ip network from source to destination, quality of service (qos), bandwidth management, specific operating systems, programming and algorithms used. conflicts of interest the authors declare no conflict of interest. references [1] r. cohen-almagor, “internet history,” international journal of technoethics, university of hull, vol. 2, no. 2, pp. 45-64, 2011. [2] b. m. leiner, v. g. cerf, d. d. clark, r. e. kahn, l. kleinrock, d. c. lynch, j. postel, l. g. roberts, and s. s. wolff, “the past and future history of the internet,” communications of the acm, vol. 40, no. 2, february 1997. [3] n. chivarov, y. paunski, v. ivanova, v. vladimirov, g. angelov, d. radev, and n. shivarov, “intelligent modular service mobile robot controllable via internet,” ifac proceedings volumes, vol. 45, no. 10, 2012, pp. 149-153. [4] w. m. h. w. kadira, r. e. saminb, and b. s. k. ibrahim, “internet controlled robotic arm,” sciverse sciencedirect, procedia engineering, vol. 41, 2012, pp. 1065-1071. [5] r. dhanabal, bh. s. r. p. varma, and v. bharathi, “remote access of fpga robot via internet,” 2014 international conference on electronics and communication system (icecs 14), february 2014. [6] d. portugal, s. pereira, and m. s. couceiro, “the role of security in human-robot shared environments: a case study in ros-based surveillance robots,” 26th ieee international symposium on robot and human interactive communication (ro-man), august 2017 proceedings of engineering and technology innovation, vol. 14, 2020, pp. 01 08 8 [7] c. xia, y. zhang, l. wanga, s. coleman, and y. liu, “microservice-based cloud robotics system for intelligent space,” robotics and autonomous systems, no. 110, pp. 139-150, october 2018. [8] a. a. jaha, f. b. shatwan, and m. ashibani, “proper virtual private network (vpn) solution,” ieee xplore, january 2009 [9] a. veglis, m. leclercq, v. quema, and j. b. stefani, “php and sql made simple,” ieee distributed systems online, vol. 6, no. 8, pp. 1541-4922, august 2005 [10] h. martínez-garcía, “implementation of a remote laboratory for distance training in robotic applications,” ieee 13th international conference on industrial informatics (indin), july 2015. [11] s. h. masovic, m. h. saračević, p. stanimirovic, and p. v. krtolica, “computing triangulations of the convex polygon php/mysql environment,” series mathematics informatics, vol. 34, no. 1, pp. 137-147, 2019. [12] o. goldstain, i. ben-gal, and y. bukchin, “remote learning for the manipulation and control of robotic cells,” european journal of engineering education, vol. 32, no. 4, pp. 481-494, august 2007. [13] s. h. masovic, m. h. saračević, p. stanimirovic, p. v. krtolica, and a. r. gallon, “qos-classifier for vpn and non-vpn traffic based on time-related features,” sciencedirect, computer networks, vol. 144, pp. 271-279, 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/).  proceedings of engineering and technology innovation, vol. 14, 2020, pp. 09 15 the autonomous shopping-guide robot in cashier-less convenience stores po-han lin 1,* , cheng-yi lin 1 , chen-ting hung 1 , jen-jee chen 1 , jia-ming liang 2 1 deptpartement of electrical engineering, national university of tainan, tainan, taiwan 2 department of computer science and information engineering, chang gung university, taoyuan, taiwan received 19 march 2019; received in revised form 13 may 2019; accepted 08 july 2019 doi: https://doi.org/10.46604/peti.2020.3961 abstract in recent years, several cashier-less convenience stores have appeared, including amazon’s. a store without cashiers will become a trend in the near future. to substitute the employee in traditional stores, this research proposes and designs an autonomous shopping cart robot to guide customers’ purchase according to the requested shopping list to enhance their shopping experience. the core techniques of the system are the autonomous driving robot, the traffic control center for robots and the dynamic route planning algorithm. the robot is a self-propelled vehicle developed by ros (robot operating system), in which we achieve the automatic driving via image recognition, april tag identification and driving direction guidance from the path planning and traffic control services. this enables the robot to lead customers to find their commodities following the preplanned route. in conjunction with the vocal service, the robot can notify the customer when arriving at each commodity, he or she plans to buy. we also design a light app for customers to easily set up and manage their shopping list, call for the robotic shopping cart’s help, and interact with the shopping cart robot. to enhance the shopping experience of customers, we design the dynamic route planning genetic algorithm to dynamically plan the shopping route according to the customer’s request and the traffic condition. experiments show that our genetic algorithm can provide the most stable performance and always get efficient shopping route planning in a limited time compared to other methods. keywords: ros (robot operating system), wireless networks, navigation, autonomous driving, cashier-less convenience store, route planning, genetic algorithm 1. introduction the development of artificial intelligence and automation technologies has been advancing by leaps and bounds over the recent years. this also helps several emerging applications make big progress, including the autonomous driving vehicle. to reduce the setup cost for researchers and in school students and help them to test their designed or learned autonomous driving algorithms in a quick manner, massachusetts institute of technology computer science and artificial intelligence laboratory (csail) designs an open source platform, called duckietown [1]. this helps more researchers and students to be involved in the field of autonomous vehicle technology [2] and has been confirmed from the actual case of education [3]. in this paper, we aim to improve the quality of shopping in cashier-less convenience stores with autonomous vehicle-based robots. this expands the possibilities of autonomous vehicle technology to the indoor application, not only the common outdoor car application. in [4], a map-based navigation method with indoor vision is proposed. to improve the adaptability of the robot, a small vehicle is equipped with a camera to capture the visual information of the environment which helps the robot to sense the * corresponding author. e-mail address: cliffabc2004@gmail.com mailto:cliffabc2004@gmail.com proceedings of engineering and technology innovation, vol. 14, 2020, pp. 09 15 10 environment changes so that the vehicle could smoothly navigate along the planned route. on the other hand, autonomous driving is an emerging and hot research area [5]. it is a complex system, so the functional modules are highly dependent. the stability of the system is very important. there are many researchers using ros for the implementation of automatic driving and prove that this is feasible [6]. the authors in [6] demonstrate that ros is suitable for developing autonomous driving applications in various configurations from the perspectives of ros characteristics, communication overhead, and advantages and disadvantages comparisons. amazon’s cashier-less convenience store [7-8] has been debated since last year. a store without any cashier will become a trend in the near future. to guide customers’ purchase in the cashier-less convenience store and enhance their shopping experience, we propose to design an automatic shopping cart robot to substitute for the employee in traditional stores. to realize such an auxiliary robot helping customers’ shopping in stores, a novel framework which integrates the autonomous driving system for navigation, dynamic route planning genetic algorithm and app-based human-machine interaction interface is presented. for dynamic route planning, we set up a cloud server to calculate the shortest route according to the customer’s requested shopping list by using a genetic-based algorithm. in this work, our contributions are threefold: (1) considering the trend of more cashier-less stores being built, our system provides a brand-new way to enhance customers’ shopping quality and experience. the proposed framework is a general solution such that it can work in most cases. (2) a prototyping system is completed and developed. in the prototype, the hardware setup is based on the low-cost raspberry pi2 while the software is totally open source. even the prototyping cart is made by using 3d printing. the setup cost and the maintenance fee are acceptable for most organizations and groups. new functions, algorithms, mechanisms, and structures can be enhanced or added on it without difficulties, which guarantees the scalability of our system. (3) to realize the autonomous driving and navigation functions over the robotic shopping cart, we implement image recognition capability at the local cart and develop dynamic route planning genetic algorithm at a central server. the former allows the robotic cart to respond to road conditions and emergencies immediately; while the latter enables the best route navigation to be provided to the customer by considering the global situations of the store. experiments show that the system works correctly and simulation results demonstrate that our proposed algorithm is stable and efficient in time complexity, route efficiency, and performance stability. 2. system architecture and proposed methods in this section, we will first introduce the system architecture of the whole system. then, the operation procedure of the ros-based robotic shopping cart will be described. finally, the dynamic route planning genetic algorithm will be illustrated. 2.1. system architecture fig. 1 system illustration proceedings of engineering and technology innovation, vol. 14, 2020, pp. 09 15 11 in the proposed system, as shown in fig. 1, customers have to connect with the shopping cart by scanning the qr code on the cart with our developed app. after connecting to the cart robot, the customer can browse the commodities via the app and create his or her shopping list. upon receiving the shopping list of the customer, the central server will plan the most efficient route for the customer to get all his/her checked commodities. then, the shopping cart robot will start to navigate its customers. each kind of commodity in the store is tagged with an apriltag. once the cart robot identifies an apriltag that belongs to one of the unordered commodities of the customer, it will stop in front of the commodities and notify the customer with voice. when the user is ready for the next commodity, the robotic shopping cart will continue its navigation. moreover, our app allows the customer to interact with the cart robot or issue a command to the robot. furthermore, the system administrator can see what the robot sees via the remote monitoring subsystem to check the status of each robot. the remote monitoring subsystem can also provide the information of robot distribution in the store for the administrator. fig. 2 shows the system architecture, which includes the app, central cloud server, shopping cart robot, and the remote monitoring subsystem. in next subsection, we will describe the system operation procedure in more detail. fig. 2 system architecture 2.2. operation procedure the robot is equipped with a fisheye camera, which keeps recording the road image and then passes the images to the lane control module and voice control module. as shown in fig. 2(3), the lane control module, fsm control module, and wheel moter control module are used to realize the lane-following function of the robot. on getting the compressed image from the camera node, the line detector node uses the opencv library to identify the desired color on the road and then passes the image to the ground projection node. the ground projection node marks and converts the image to compute exact coordinates of the image in the space. afterwards, the information enters the lane filter node and then calculates the driving direction according to the identified lane and coordinates. the driving direction is sent to the lane control node and the node will convert the direction into the dedicated car_cmd. the operation procedure then enters the fsm control module. upon the receipt of the car_cmd from the lane control module, the fsm node switches or maintains the state accordingly. inside the module, the fsm and car cmd switch nodes communicate with each other to keep the fsm status in the car cmd switch node up to date. specifically, the car cmd switch node is responsible to drive the wheel driver node in the wheel motor control module to manipulate the motors according to the state of fsm. on the other hand, upon the receipt of the recording image from the camera node, the apriltags node in the voice control module processes the image and recognizes whether there is any apriltag in the image. if yes, the id of the tag will be conducted and then be forwarded to the commodity_node. on receiving the tag information from the apriltag node, the commodity_node will check whether the tag is for a commodity or an intersection. if it is a tag of a commodity and the commodity is in the customer’s shopping list, the robotic cart will no tify the customer and wait for the command of the customer. on the contrary, if the tag means an intersection, the robot will send proceedings of engineering and technology innovation, vol. 14, 2020, pp. 09 15 12 the content of the tag to the central server and waits for the instruction from the server. possible instructions include turning right, turning left, going straight and making a u-turn. 2.3. dynamic route planning to let the customer be able to find all his or her requested commodities in short time. we design a dynamic route planning genetic algorithm to plan the shortest route in the cashier-less convenience store for the user according to his or her selected commodities in app. the route planning problem can be modeled by using a graph g = {v, a}, where v = {v1, ..., vn} is the set of vertices, each vi ∈ v is either an intersection or a commodity, and a = {(vi, vj): vi, vj ∈ v, i != j}, each arc (vi, vj) ∈ a has a weight (or length) cij ≥ 0. the problem is similar to but not completely equal to the travelling salesman problem (tsp), which is a well-known np-hard problem [9]. but we still can refer to those previous methods which solve the tsp problems. since the response time is important to the customers, our designed algorithm has to be completed in limited time and the planned route can be sub-optimal but close to the optimal solution. to achieve this, we have designed a dynamic route planning scheme based on genetic algorithm. 2.4. dynamic route planning genetic algorithm the genetic algorithm models a process of evolution of a population of individuals. each individual gk is a feasible solution to the route planning problem. the components of an individual are called genes. here, in this work, each gene gk(l), l = 1..m, is one of the m selected commodities of the customer. for any two neighbor genes, gk(l) and gk(l+1), we adopt the shortest length route, such that the cost cgk(l),gk(l+1) of moving from commodity gk(l) to gk(l+1) is the shortest route length from gk(l) to gk(l+1). so, the total cost of individual gk is 1 0 ( ), ( 1)( ) m tot k l gk l gk lc g c    (1) where gk(0) and gk(m+1) are the starting point (i.e., entrance of the store) and ending point (i.e., checkout counter), respectively. the steps of our dynamic route planning genetic scheme are described as below. (1) create initial populations gk, k = 1..k. set t = 1. (2) repeat steps 2.1-2.5 until t = t: (2-1) remove the individuals with ctot > cth. (2-2) choose two parent individuals gx and gy from the remaining populations. note that for each gk, the chosen probability is proportional to 1/ctot(gk). (2-3) apply crossover to gx and gy and then obtain new individuals g’x and g’y. repeat steps 2.2 and 2.3 until the total amount of parent individuals and offspring individuals being equal to or greater then k. (2-4) for each individual gk, each of its gene gk(l), l = 1..m, has a probability of pm to apply mutation. (2-5) set t = t + 1. (3) choose the individual with the least route length among the remaining populations as the solution. 3. prototype implementation and experimental results in our designed cashier-less convenience store environment, user can browse the commodities by app and request the duckiebot to find the selected commodities. duckiebots follow the central server planned route to guide users in the store. once the duckiebot finds an apriltag of commodities, it stops and uses the voice function to notify the customer. the user can use app (as shown in fig. 3) to control the state of shopping cart. proceedings of engineering and technology innovation, vol. 14, 2020, pp. 09 15 13 we use duckiebot, a device that serves as the primary client for autonomous driving in duckietown [4], as the basic architecture for navigation robots. our main hardware and software are based on the raspberry pi 2 b+ ubuntu operating system. other hardware includes a adafruit dc & stepper motor hat, adafruit 16-channel pwm/servo hat, fisheye camera, car module, mobile power and the internal system of the robot is open-source ros (as shown in fig. 4). a demo video is uploaded and the link is as below: https://ez2o.co/67hh5. fig. 3 our developed app (a) robotic shopping cart model (b) shopping cart with machine hand fig. 4 the hardware prototyping of our robotic shpping carts in the experiments, we compare our route planning genetic algorithm (ga) with other methods. the methods include exhaustive search algorithm (esa), simulated annealing algorithm (saa), greedy method and tabu search method (ts). as shown in fig. 5 and fig. 6, esa always gets the shortest route but spends the longest execution time. the execution time increases exponentially as the number of selected commodities increases. the lengths of the derived routes in ga, saa, and ts are close to esa while the execution time of them are much less than esa. fig. 5 the effect of the number of selected commodities on the execution time 0 10 20 30 40 50 60 7 8 9 10 11 12 13 14 se c number of selected commodities ga esa saa greedy ts proceedings of engineering and technology innovation, vol. 14, 2020, pp. 09 15 14 fig. 6 the effect of the number of selected commodities on the route length as shown in fig. 7, we can see ga always get the shortest route length compared to other methods. compared to ga, the performance of other methods is unstable. saa and ts can get almost the same performance when the number of selected commodities is less than 20, but the outcome becomes unstable when the number of selected commodities are more than 20. as shown in fig. 8, the more the generations we use in ga, the better the solutions to route planning we get. we can see that the performance converges when the running generations are more than 200. the improvement of route planning results are very limited, even if we spend more time on calculation. so, we choose to set the number of generations to 200 in ga. fig. 7 the effect of the number of selected commodities on the normalized route length (the normalized route length of each method = (route length of each method/route length of ga) * 100% fig. 8 the effect of running generations of ga on the route length 4. conclusions in this paper, to improve the shopping quality of customers in cashier-less convenience stores, we design an automatic robotic shopping cart which integrates image recognition, dynamic route planning, and app as the human-machine interface. 0 10000 20000 30000 40000 7 8 9 10 11 12 13 14 le n g th number of selected commodities ga esa saa greedy ts 90 100 110 120 130 140 150 11 16 21 26 31 36 n o r m a li z e d r o u te le n g th (% ) number of selected commodities ga saa greedy ts 0 10000 20000 30000 40000 50000 60000 11 16 21 26 31 36 le n g th number of selected commodities 50 100 150 200 250 proceedings of engineering and technology innovation, vol. 14, 2020, pp. 09 15 15 the shopping cart robot is a self-propelled vehicle developed by open source software ros, in which the automatic driving is achieved via image recognition. the proposed dynamic route planning genetic algorithm gets the shortest route compared to other methods in limited time. the robot can guide customers to find their interested commodities following the planned route. in conjunction with the vocal service, the robot is able to notify the customers. the customers are allowed to interact with the robot by the app. conflicts of interest the authors declare no conflict of interest. acknowledgment this research is co-sponsored by most 107-2221-e-024-001-my3, 106-2221-e-182-015-my3, itri, moe, and chang gung memorial hospital, taoyuan. references [1] what is duckietown? online available: http://duckietown.mit.edu/ [2] l. paull, j. tani, h. ahn, et al, “duckietown: an open, inexpensive and flexible and capable platform for autonomy education and research,” 2017 ieee international conference on robotics and automation (icra), 2017, pp. 1497-1504. [3] j. tani, l. paull, m. t. zuber, d. rus, j. how, j. leonard, and a. censi, “duckietown: an innovative way to teach autonomy,” international conference edurobotics 2016, 2016, pp. 104-121. [4] chi-shian lin, “study on map-based indoor mobile robot vision navigation,” matster’s thesis of department of electrical engineering, national cheng kung university, tainan, taiwan, 2009. [5] ö. ş. taş, f. kuhnt, j. m. zöllner, and c. stiller, “functional system architectures towards fully automated driving,” 2016 ieee intelligent vehicles symposium (iv), june 2016, pp.304-309. [6] a. hellmund, s.wirges, ö. ş. taş, c. bandera, and n. o. salscheider “robot operating system: a modular software framework for automated driving,” 2016 ieee 19th international conference on intelligent transportation systems (itsc), 2016. [7] inside amazon go, a store of the future online available: https://www.nytimes.com/2018/01/21/technology/inside-amazon-go-a-store-of-the-future.html [8] d. grewal, a. l. roggeveen, and j. nordfält, “the future of retailing,” journal of retailing, vol. 93, no. 1, pp. 1-6, 2017. [9] m. r. garey and d. s. johnson, “computers and intractability: a guide to the theory of np-completeness,” w. h. freeman and company, san francisco, 1979. 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/).  proceedings of engineering and technology innovation, vol. 13, 2019, pp. 26-31 hydrothermally grown zno nanoflowers on a template-assisted ordered seed array fang-hsing wang * , ching-shan wang, han-wen liu department of electrical engineering and graduate institute of optoelectronic engineering, national chung hsing university, taichung, taiwan received 12 april 2019; received in revised form 15 may 2019; accepted 21 june 2019 abstract 1-d nanostructure arrays recently attract much attention because of their unique optical, structural, and electronic properties in the field of materials science, microelectronics, and optoelectronic engineering. zinc oxide (zno) nanoflowers were synthesized by a facile hydrothermal method on a template-assisted deposited zno seed array. the zno thin film was prepared via a sol-gel spin-coating process on a concave sapphire substrate first. then an al layer and an optically clear resin film were individually deposited on the zno thin film. after a lift-off process, a convex zno seed array film was formed. finally, zno nanoflowers were synthesized by the hydrothermal method at 90 o c for 10 to 60 min. the structural, morphological and optical properties of the zno nanoflowers were investigated. the xrd results indicate that the zno nanoflowers were polycrystalline with a hexagonal wurtzite-type structure with a (002) preferential orientation. the fe-sem micrographs exhibited the diameter and length of zno nanorods increased with the increasing growth time from 10 to 60 min. the 105-nm diameter and 1150-nm length nanorods were obtained with 60-min growth time. photoluminescence spectra showed a sharp emission peak (iuv) at approximately 380 nm and its intensity increased with the growth time. a weak emission band (ivis) at 450–550 nm was also observed and the iuv/ivis increased with the growth time. this result indicates that the defects were reduced and the crystal quality was enhanced with the growth time. the prepared zno nanoflowers can be applied to various optoelectronic and sensing devices. keywords: zinc oxide, nanoflower, template-assisted, lift-off, hydrothermal, sol-gel 1. introduction zinc oxide (zno) is an ii-vi compound semiconductor with a wide direct bandgap of 3.3 ev and a large exciton binding energy of 60 mev at room temperature. zno has the advantages of high thermal and chemical stability, easy processing by wet chemical etching and can be fabricated in various nanostructured forms and low-cost-low-temperature processes [1-2]. zno has been extensively used in fabricating transparent conducting electrodes, uv lasers and light-emitting diodes, solar cells, photodetectors, and gas sensors [3-7]. zno-based devices come out on the basis of bulk, thin film or nanostructured materials. nanostructure materials are comprised of an array of zero-dimensional, one-dimensional (1-d), two-dimensional, and three-dimensional structures. among them, 1-d nanostructure array materials generally have unique optical, structural, and electronic properties, and play an important role in the field of materials science, microelectronics, and optoelectronic engineering, especially for light harvesting and energy conversion/storage, environmental sensing and monitoring, and catalysis. the common representation of 1-d nanomaterials includes nanotubes, nanorods, nanowires, nanofibers, nanoflowers, etc. nanoflowers are newly developed one of the nanomaterials that exhibit a structure similar to flowers. because of its simple * corresponding author. e-mail address: fansen@dragon.nchu.edu.tw proceedings of engineering and technology innovation, vol. 13, 2019, pp. 26-31 27 preparation method, high stability, and improved efficiency, it has attracted much attention [8]. [9] firstly proposed an organic-free hydrothermal process to prepare flower-like zno nanostructures in 2004. [10] presented that substrate played a crucial role in preparing zno nanostructures and the flower-like zno exhibited better optoelectronic performance than the block-like and rod-like ones. [11] reported that different concentrations of oh in the solution strongly affected the nucleation rate and crystal growth the morphology of flower-like zno. this work develops a novel approach to synthesize flower-like zno nanostructures on ordered seeded silicon substrates. regardless of complicated equipment and expensive high-temperature vapor-phase methods, we adopt a low-cost aqueous chemical growth method, hydrothermal, to grow flower-like zno nanostructures, which are synthesized on the ordered zno seed arrays prepared via a template-assisted sol-gel spin-coating process. the structural, morphological and optical properties of the flower-like zno nanostructures were characterized. 2. experimental details fig. 1 flowchart for the synthesis of zno nanoflowers zinc oxide (zno) nanoflowers were synthesized by a facile hydrothermal method with a template-assisted deposited zno seed array. fig. 1 shows the flowchart for the approach. the process consists of seven steps: (1) concave sapphire substrates were cleaned with acetone, isopropyl alcohol, and d.i. water for 10 min each with an ultrasonic cleaner (delta, dc-300) and then dried under nitrogen flow. (2) a sacrificial aluminum layer was deposited by thermal evaporation. (3) a zno seed layer with a thickness of approximately 190 nm was deposited via a sol-gel spin-coating process and then annealed in ar ambient at 500°c for 1 h. (4) an optical grade silicone a-b glue was coated and baked at 100°c for 1 h, as fig. 2 (a) shows. (5) the sacrificial aluminum was etched in a mixed solution with k3fe(cn)6：koh：h2o = 10 g：1 g：100 ml. the etching rate was approximately 10 nm/s. fig. 2 (b) and (c) exhibit the samples in the etching solution and after the etching process, separately. (6) the zno seed layer/silicone a-b glue was lift-off and transferred to silicon substrates. a convex zno seed array film was formed. (7) finally, zno nanorods were synthesized by the hydrothermal method at 90 o c for 10 to 60 min with the precursors of zinc acetate and hexamethylenetetramine in d. i. water (200 ml). the structural, morphological and optical properties of the flower-like zno nanostructures were investigated by x-ray diffraction (xrd) (panalytical, 18-kw rotating anode x-ray generator, japan) with cu-kα radiation (λ = 0.154056 nm) in θ–2θ scan mode, field emission scanning electron microscopy (fesem) (3.0 kv; jeol jsm-6700f, japan), and photoluminescence (pl) spectrometry (horiba jobin yvon ihr550 fluorescence spectrometer) with a 325 nm he-cd laser. proceedings of engineering and technology innovation, vol. 13, 2019, pp. 26-31 28 (a) after optical glue coating (b) in the etching solution (c) after the etching process fig. 2 photographs of the samples 3. results and discussion 3.1. characterization of zno seed array (a) plan-view (b) 45 o tilt view fig. 3 fesem images of the zno seed film. the scale bar represents 1 m. fig. 3 (a) and (b) shows plan-view and 45 o tilt view fesem images of the zno seed film on the concave substrates. from fig. 3(a), it was found that the zno seed film formed a hexagonal pattern array and the pitch between the zno seed dots was about 1.35±0.2 nm. the surface particle size of the zno seed film was about 65±10 nm. from fig. 3 (b), it was also found that the zno seed film exhibited a convex mountain-like array and the thickness of the zno seed film was 184 nm. the energy dispersive x-ray spectroscopy (eds) analysis was performed and we found si, zn, o, and a small amount of al and c elements on the samples. the appearance of al may be caused by the deposition of aluminum impurities (such as aluminum hydroxide) in the etching solution for a long time. fig. 4 shows xrd patterns of zno seed film on silicon substrates. the zno seed film has three weak diffraction peaks of (002), (100), and (101), and (002) orientation was slightly stronger than the others. this result suggests that zno seed film exhibits a wurtzite structure with polycrystalline structure. 20 30 40 50 60 (101) (002) in te n si ty ( a r b . u n it ) 2(degree) (100) zno seed layer fig. 4 xrd pattern of the zno seed film on silicon substrates proceedings of engineering and technology innovation, vol. 13, 2019, pp. 26-31 29 3.2. characterization of flower-like zno nanorod array (a) 10 min (b) 20 min (c) 30 min (d) 60 min fig. 5 plan-view fesem images of the flower-like znra. the scale bar represents 1 m. fig. 5 displays plan-view fesem images of the flower-like zno nanorod array with the growth time (a) 10 min, (b) 20 min, (c) 30 min, and (d) 60 min. the length and diameter of the zno nanorods increased with the growth time from 10 to 60 min. the diameter of 105 nm of the zno nanorods was obtained for 60 min-growth times. for the sample with 10 min-growth times, the nanorods were short. it may be attributed to excess impurities on the seed film surface. fig. 6 shows 45 o tilt-view fib-sem image of the flower-like zno nanorod array with a growth time of 60 min. the nanorod length was approximately 1125 nm after subtracting the thickness of the seed layer by 200 nm. the aspect ratio and density of the nanorods were calculated as 11.0 and 20 m -2 for the sample with 60 min growth time. fig. 6 45 o tilt-view fib-sem image of the znra. the scale bar represents 1 m fig. 7 shows xrd patterns of the flower-like zno nanorod array (znra) grown by the hydrothermal method with growth times of 10, 20, 30, and 60 min. it was obvious that the intensities of the detected diffraction peaks increased with increasing growth time. peaks with (002), (101), (102), (100), and (110) orientations were found, which suggests that the zno nanoflowers have a polycrystalline structure. the intensity of the diffraction peak in the (002) direction located at approximately 34 ̊ was higher than the others, which indicates that the zno nanorods had a preferred orientation in the c-axis direction. this result conforms to the wurtzite structure of zno according to the jcpds database. fig. 8 shows proceedings of engineering and technology innovation, vol. 13, 2019, pp. 26-31 30 photoluminescence (pl) spectra of the flower-like zno nanorods with growth times of 10 to 60 min. the pl spectrum showed that the relative intensity of ultraviolet (uv) and visible light (450-550 nm) depends on the growth time of zno nanorods. the peak of photoluminescence of uv band approximately 380 nm was strongly enhanced when the growth time of zno nanorods was increased, while the visible bands decreased. the intensities of the uv band and visible band of pl spectra of the flower-like zno nanorods are listed in table 1. the increase in uv intensity is related to the surface to volume ratios of the zno nanostructure, while visible luminescence (approximately 521-543 nm) caused by oxygen defects [12-13]. the decreased ivis/iuv ratio with the increasing growth time indicates that the surface to volume ratio and crystal quality of zno nanorods can be enhanced by increasing the growth time. the calculated surface to volume ratio for the 60 min-sample was 3.90×10 -2 . the improved crystal quality of the flower-like zno nanorod array consists of the result of xrd analysis. fig. 7 xrd patterns of the znra grown by the hydrothermal method 350 400 450 500 550 600 650 in te n si ty ( a r b . u n it ) wavelength (nm) zno nanorods 60 min 30 min 20 min 10 min fig. 8 photoluminescence (pl) spectra of the flower-like znra with growth times of 10-60 min table 1 intensities of uv band and a visible band of pl spectra of the flower-like znra with different growth times time (min) iuv ivis iuv/ivis 10 594.4 169.80 3.50 20 1073.6 118.61 9.05 30 1597.0 116.23 13.7 60 2068.4 83.22 24.9 4. conclusions the zno nanoflower array has been fabricated by a facile hydrothermal method via a template-assisted prepared convex zno seed array. the diameter and length of the flower-like zno nanorods both increased with increasing growth time. the diameter, length, aspect ratio, surface-to-volume ratio, and density of the zno nanorods were 105 nm, 1125 nm, 11.0, proceedings of engineering and technology innovation, vol. 13, 2019, pp. 26-31 31 3.90×10 -2 , and 20 m -2 , respectively, for the growth time of 60 min. the xrd analysis indicates that the zno nanoflowers were polycrystalline with a hexagonal wurtzite-type structure with a (002) preferential orientation. photoluminescence spectra exhibited a strong uv peak at approximately 380 nm and a weak visible band (at approximately 450-550 nm). the increased iuv/ivis ratio revealed that the defects were reduced and the crystal quality was enhanced with the increasing growth time. the developed zno nanoflowers array can be applied to various optoelectronic and sensing devices. conflicts of interest the authors declare no conflict of interest. acknowledgments we thank to ministry of science and technology of taiwan under the grant most 106-2221-e-005-057 and most 107-2221-e-005-048 for financial support. references [1] ü . özgür, y. i. alivov, c. liu, a. teke, m. a. reshchikov, s. doğan, v. avrutin, s. j. cho, and h. morkoç, “a comprehensive review of zno materials and devices,” journal of applied physics, vol. 98, 041301(1-103), 2005. [2] a. b. djurisic and y. h. leung, “optical properties of zno nanostructures,” small, vol. 2, pp. 944-961, 2006. [3] v. a. coleman and c. jagadish, “zinc oxide bulk, thin films, and nanostructures: processing, properties and applications,” 1st ed. oxford: elsevier, pp. 1-20, 2006. [4] y. song, f. chen, y. zhang, s. zhang, f. liu, p. sun, x. yan, and g. lu, “fabrication of highly sensitive and selective room-temperature nitrogen dioxide sensors based on the zno nanoflowers,” sensors and actuators b: chemical, vol. 287, pp. 191-198, may 2019. [5] d. acharyya, “influence of rgo cladding in improving the sensitivity and selectivity of zno nanoflowers-based alcohol sensor,” ieee sensors journal, vol. 18, pp. 1820-1827, march 2018. [6] f. h. wang and c. e. tsai, “zno nanorod-based ultraviolet photodetector prepared on patterned sapphire substrates,” ieee journal of selected topics in quantum electronics, vol. 23, no. 2, p. 3800606, april 2017. [7] f. h. wang and c. l. chang, “effect of substrate temperature on transparent conducting al and fco-doped zno thin films prepared by rf magnetron sputtering,” applied surface science, vol. 370, pp. 83-91, 2016. [8] p. shende, p. kasture, and r. s. gaud, “nanoflowers: the future trend of nanotechnology for multi-applications,” artificial cells, nanomedicine, and biotechnology, vol. 46, no. s1, pp. s413-s422, january 2018. [9] h. zhang, d. yang, x. ma, y. ji, j. xu, and d. que, “synthesis of flower-like zno nanostructures by an organic-free hydrothermal process,” nanotechnology, vol. 15, pp. 622-626, march 2004. [10] q. liang, f. qiao, x. cui, and x. hou, “controlling the morphology of zno structures via low temperature hydrothermal method and their optoelectronic application,” materials science in semiconductor processing, vol. 89, pp. 154-160, 2019 [11] r. shi, p. yang, x. dong, q. ma, and a. zhang, “growth of flower-like zno on zno nanorod arrays created on zinc substrate through low-temperature hydrothermal synthesis,” applied surface science, vol. 264, pp. 162-170, 2013. [12] q. li, j. bian, j. sun, j. wang, y. luo, k. sun, and d. yu, “controllable growth of well-aligned zno nanorod arrays by low-temperature wet chemical bath deposition method,” applied surface science, vol. 256, pp. 1698-1702, 2010. [13] b. lin, z. fu, y. jia, and g. liao “defect photoluminescence of undoping zno films and its dependence on annealing conditions”, journal of the electrochemical society, vol. 148, pp. g110-g113, 2001. 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/).  proceedings of engineering and technology innovation, vol. 1, 2015, pp. 10 14 10 a study on the mode of introducing independent maintenance into traditional industries that rely on manual work chi-chih shen * department of industrial engineering and management, taiwan shou-fu university, tainan 721, taiwan, roc. received 20 june 2015; received in revised form 17 july 2015; accept ed 31 july 2015 abstract the vast majority of taiwan's enterprises are small and mediu m-sized enterprises, which account for 98% and mostly are built up by the traditional industry. but along with the technical progress and development of production technology, the traditional industry is faced with survival bottleneck because of high cost and various competit ors. therefore this research aims to explore the mode of introducing independent maintenance to traditional industries, and to develop a suitable one for today's traditional industries. research results will be provided to the domestic traditional industries for reference in order to reduce the cost and difficulty they may have in importing independent maintenances. keywor ds : traditional industry, independent maintenance, competitive power 1. introduction the economic construction of taiwan is led by the development of industry, which mainly relies on manufa cturing sector. in terms of the production specifications, manufacturing sector can be divided into traditional industries, basic industries, science and technology-intensive industries. the vast majority of taiwan's enterprises are small and mediu m-sized enterprises, which account for 98% and mostly are built up by the traditional industry. nowadays with the rapid development of society, techno logy and information communication, in order to protect the traditional industry from being impacted, and remain its role in the economic development, we must firstly understand the characteristics of the traditional industries and then take proper measures accordingly. in order to enable them fully play their roles, a joint cooperation between the industry, the official sector and academic institutions is required, so that the traditional industry can be led into a higher level of development. the whole staff’s awareness of and support for independent maintenance and their dedication is also important for the continuous development of enterprises’ operating structure. only with these efforts can the trad itional industries of taiwan can be transformed and the competitiveness of enterprises enhanced. the first systematic concepts and methods of maintenance were developed in the late 1960s when airlines grew interested in equipment ma intenance because of the issue of improving the work efficiency of jetliner ma intenance. and then the method of reliability centered maintenance (rcm) was developed, which became widely used among the aviation community. in the early 1970s, the military also used this method in military aircraft and warships. rcm method is designed ma inly for diverse and comple x equipment ma intenance. it ma kes reliability as the core consideration and prioritizes the maintenance work coordinated with failure mode and effect analysis (failure mode and effect analysis fmea) so as to make the reliability of the equipment system achieve the best state. there are six stages in the evolution of the tpm: 1. the era of break-down maintenance. (break-down maintenance; before the bm-1950) [gera1d j. hahn, and sa mual s. shapiro]; the maintenance work begins after the equipment fails to work or the performance deteriorates. the maint enance features the ex-post rush repair of the repair personnel, who usually have no awareness of equipment management. 2. the era of preventive maintenance (preventive the maintenance; pm-1951's) [edward a, si1ver and claude-nico1as fiechter]; periodic inspections and scheduled checks for a variety of devices are carried out so that the poor operating condition of equipments, if any, can be discovered as soon as possible and so the failure can be prevented. (routine maintenance, roving maintenance, periodic maintenance, preventive maintenance, and repair) therefore preventive maintenance is the inspection and check work conducted mainly by the maintenance personnel. 3. the era of productive maintenance (productive the maintenance; pm-1954): it is advocated by the u.s. general electric company, which means that in consideration of the economic value of the equipments, the factory carries out scheduled checks, lubrication and minor repairs in order to reduce the cost caused by downtime and substantial ma intenance work. it is the integration of ex-post maintenance and preventive maintenance and aims to achieve the purpose of economic production. 4. the era of corrective maintenance (corrective maintenance; cm -1957~) [von alven, wh: it means upgrading the quality of equipments and relevant parts in order to improve their re liability and durability. 5. the era of maintenance prevention (the maintenance prevention; mp-1960~) the [sa mue1 s shapiro, and a1an j. gross]: improve the project plan so that the devices do not need maintenance or only requ ires minima l and simple ma intenance. 6. the era of total productive maintenance (total productive maintenance; the tpm-1970~) [seiichi nakajima]: a co mprehensive and universal system which focuses on integrating inter-departmental and within-departmental communication and cooperation. that is, let all departments of a company from the management to operators; let all the staff take an active part in the plan and maintenance of equipment so as to establish a comprehensive production and maintenance system. developed by bin and others, the dynamic preventive ma intenance system (fa ilure and scheduled maintenance analysis --fasma), which is human-centered, is designed for a manufacturing environment that is cellular based, system-oriented, and that emphasizes comprehensive qual *corresponding aut hor, email: chi-chih@tsu.edu.t w proceedings of engineering and technology innovation, vol. 1, 2015, pp. 10 14 11 cop y right © taeti ity management. this system adopts a deductive database with dynamic interface, and use foregoing rcm method to continuously improve preventive maintenance plan so as to minimize downtime. this system infers the category of fault and the time to maintain only from the historical data and fault knowledge, not including experiential maintenance knowledge and real-time status information of equipments. okogbaa and others [okogbaa, g., hliarlg , j. and she11 r l.] ca me up with a preventive maintenance system (ppms) for the automated manufacturing systems, stressing the concept of opportunistic maintenance. when imple menting preventive maintenance, this system tells the distribution of the failure based on statistical analysis of historical data collected by equipments and uses simulation analysis to predict failure and derives the best timing to maintain, which is very helpful for the reduction of production costs. mcfadden [mcfadder1, rh] e xplored the process of establishing a database system with reliability, operability, and maintainability. he also analyzed in detail the data types, as well as the design of the database. the pintelonm and wassenhove [pintelonm, l. and wassenhove, lv] put forward a control panel that includes various performance indicators and detailed reports. the performance indicators include a comprehensive indicator for ma intenance ma nagement. the range of measurement covers fields like cost, equipment, manpower, materials, and work. these perfo rmance indicators help managers to evaluate maintenance effectiveness; while early warning information provided by cb and detailed information by dr is useful for dec ision-making. the above literature is the overview of precedented research on equipment maintenance. because in the past years the maintenance management of equipment didn’t draw sufficient attention, there are few researches on equipment maintenance or tpm in ta iwan. .[shen ji zhi, ling ben taro, qiu yuanxi, etc.]. in foreign countries, excepts japan(sasaki, terumi, ling ben taro, marcelo rodrigues, kazuo hatakeyama, amir.khanlari, kaveh mohammadi), researches in this field has also been valued in recent years.(the bin wu and jonathan j m. seddon, edward a, si1ver and claude-nico1as fiechter, okogbaa, g., hliarlg, j ., and she11, r. l., pintelonm, l. and wassenhove, lv, kristy o.cua, kathleen e.mclone, roger g.schroeder, g. .. chand, b. shirvani) 2. method 2.1. the basic concept of independent maintenance and its implementation steps the so-called independent maintenance means that with the purpose of taking care of one’s own equipment, each staff me mber carries out routine point inspections on their own equipment, lubricate, replace parts, and detect abnorma lities early and correct precision so as to prevent a serious fault from happening. in order to imple ment the independent maintenance, workers must be trained into masters in equipment. equipped with operating skills, worke rs also must have the ability of detecting abnormalities, of improving them and then the ability of understanding the relationship between equipment with quality of products. it often thought that the failure or the deterioration of equipment function is due to human misconducts. so if this kind of ideas can be changed and people can have the concept of "zero defect" and "zero failure", gradually they will change the equipment and on-site. the systematic stimulation of the company will help ensure the full participation of the whole staff in the staged imp le mentation of independent maintenance. the steps of imple menting independent maintenance are as follows: 1.the initial cleaning: cleaning equipment is helpful for understanding the equipment and finding whether the device is normal or not. in other word, cleaning is point inspection. 2. the solution for the source of failure and difficult parts: find the source of pollutants in equipments and working environment in order to prevent pollution. as for those parts on which the daily maintenance work like cleaning, point inspection and lubrication is hard to carry out, measures also should be taken to improve them so that the maintenance work can be done easily.3. develop interim benchmarks for independent maintenance: base on experience gained from the first two steps, suggestions should be made on issues like which parts needs point inspection, lubrication and cleaning, the corresponding methods and also the period.4. the overall inspection: in order to seek the highest efficiency of the device, workers must understand the structure, function, principles of equipments and should know the status of the equipment, so a comprehensive inspection is a must in order to find out the potential defects and restore the original status which the equipment is supposed to maintain.5. independent inspection: the purposes of the first four steps are to maintain the device and improve its degradation. if the reliability, ma intenance and quality of the equipment is about to be further improved, then the benchmark made previously for cleaning, lubrication and overall inspection must be re-e xa mined so as to improve the efficiency of point inspection, making no inspection mistakes. a summary of the benchmarks of independent maintenance will then be achieved as a result. 6. standardization: step one to five center on devices, but a standardized set of practice is needed in order to administer and imp le ment such activities like the preparedness of the basic conditions and daily point inspection. the standardized practice will enable a person to do the job well alone no matter who the person is and where he is, achieving self-management. 7. the thorough imp lementation of self-management : this step is to summarize the activities promoted in the foregoing six steps, and show improved results by changing the equipment, personnel, and on-site, which brings more self-confidence and courage to challenge more ideal targets like zero defect, zero breakdown, zero disasters and zero short-term downtime. 2.2. the basic concept of individual improvement the so-called individual improvement is the activity of exc luding the loss caused by personnel, materials and equipment (such as the loss brought about by the operating mistakes of workers, by the equipment failure or short downtime and by lost work time due to unbalanced production line). that is, to maximize productivity and aim at real effect. before the improvement activities begin, personnel, materia ls, equipment, and other various types of loss projects should be mastered along with corresponding data, so that benchmark values for each loss projects can be set to server as the comparison with the improved situation in the future. following are the ten steps to carry out individual ma intenance: 1. select the equipment, production line and project for demons tration:(1) those defective production lines and projects and equipments that wear a lot in the proceedings of engineering and technology innovation, vol. 1, 2015, pp. 10 14 cop y right © taeti 12 individual improvement plans, (2) horizontally lay out those ones with great importance. (3) the equipment, projects and equipments that are the same with those in independent maintenance activities. 2. form an ad hoc group: me mbers of the ad hoc group, in addition to the project they are in charge of, should also offer technical analysis and support for others’ loss projects, bringing into full play the role of interpersonal relationship. (1) the managers of d epartments become leaders (leaders of demonstrative d epartments become managers and leaders of demonstrative section become section supervisor). (2) invite the assistant units of technology, design and produ ction to join in. whenever loss occurs, it should be borne by the unit which is in charge for the corresponding loss unit and which causes the loss. (3) the promoting units designate members of the group and directs. 3. grasp information about the current loss: (1) check and know about the loss. (2) when there is not enough information about loss, collect them. 4. set the theme and the goal of improvement:(1) set the theme of improvement according to the findings of research on the current situation.(2) set the goal and period of improvement based on the concept of zero-loss.(3) determine the person in charge of the countermeasures for loss. 5. work out plans for improve ment:(1) complete a plan including analysis, countermeasures and procedures of improvemen t. make a schedule to promote imple mentation of the plan.(2) carry out high-end diagnosis. 6. analyze and develop strategies and assessment:(1) utilize a ll the skills and methods like analysis, investigation and experiments. establish the improvement case and develop the assessment method.(2) make efforts continuously until the goal is achieved.(3) carry out high-end diagnosis (looking for better ways in the form of announcement campaign) for the purpose of e nrichment. 7. put into to effect measures for improvement t: imple ment necessary budget management, and make i mprovements. 8. confirm the effect: to confirm the effect of improving various loss projects. 9. standardization:(1)standardize the standards for the manufacturing, operating, material-choosing and maintenance and make countermeasures for the prevention of the recurrence of loss.(2) make the manual for horizontal imple mentation..(3) make high-end diagnosis.10.horizontal lay-out: lay out the same production lines, projects, equipment horizontally . 3. the process of introducing independent maintenance into traditional industries 3.1. the plan of importing independent maintenance pilla r the master plan is a rough plan, so when the industry is about to promote the independent maintenance comprehensively, it should develop a more detailed overall plan of importing the pillar based on the benchmark of master plan, and then import the overall plan to make a three -year schedule and a promotion plan for the current year. the promotion plan should be more detailed than the three-year plan. although there are many plans during the promotion process, a complete plan is needed in order to carry out all works smoothly. the following is about how make the plan of importing the pillar of independent maintenance: this s tudy divides the independent maintenance plan into two parts for illustration. the first part is the overall plan of importing the pillar of independent maintenance, whose content is as shown in table 1. the second part is the plan of importing the four steps of independent maintenance, whose main work item and order are as shown in table 2. a gantt chart, namely the plan for the importation of each steps, can be made according to the items and order listed in table 2. table 1 overa ll plan of import ing independent maintenance pillar it em abst ract explanat ion 1 members and managers of t he organizat ion for plan promotion ‧the struct ure of independent maint enance branches, members doing various jobs and the units they belong to. ‧t he manager of each position 2 object ive ‧t he overall goal for the coming three years and the goal for the current year should be included.(choose from the project s of efficiency measurement indicat ors) 3 p roject s ‧list all the project s t o be accomplished in the current year and next three years. each project should have a manager. 4 schedule ‧every important project should be included in the schedule of promotion in order to facilitate the management. the schedule should specify the beginning date and finishing date of every project t hrough gantt chart. 5 p lan of promot ion for each st ep ‧express each st ep in order through gantt chart , making t he plan of promotion. 6 st at ist ic dat a of various achievement indicators ‧ every clique, group, sect ions and department should have it s own st at ist ic data and collect , analyze and renew t hem monthly. 7 list of all cliques ‧t he name of each clique and the unit s t hey belong t o. (concluded by this research) after the overall plan is done, the goal values should be allocated to all departments. each department should make its own plan of promoting independent maintenance according to the goal and the activities in the overall plan. the finished plan then will be further imple mented in sections and cliques. every unit should have its own plan for implementation. table 2 plan of each step in importing independent maintenance implement at ion order major work in t he first st ep major work in t he second st ep major work in t he t hird st ep major work in t he fourt h st ep 1 t he first st ep educat ion t raining t he second st ep educat ion t raining t he t hird st ep educat ion t raining t he fourt h st ep educat ion t raining 2 t he first st ep make import at ion plan t he second st ep make import at ion plan t he t hird st ep make import at ion plan t he fourt h st ep make import at ion plan 3 clean equipment s t horoughly make t emporary benchmarks for cleaning, lubricat ion and point inspect ion make checking form for overall inspect ion set benchmarks for reflect ions on cleaning, lubricat ion and point inspect ion 4 marking out defect ive part s wit h t ags implement independent diagnosis implement overall inspect ion and mark out defect ive point s make independent maint enance benchmarks proceedings of engineering and technology innovation, vol. 1, 2015, pp. 10 14 13 cop y right © taeti table 2 plan of each step in importing independent maintenance (continued) 5 t ake off t he t ags aft er t he first -t ime improvement is done implement depart ment al diagnosis or high-end diagnosis analyze and improve t he defect ive project s make t he schedule for independent maint enance 6 add lubricant t o equipment s and t ight en screws make t emporary benchmarks for overall inspect ion implement independent maint enance(cleani ng, lubricat ing and point inspect ion) 7 confirm t he source of problems and t he difficult part s, mark t hem out and draw count ermeasures(at t ach t ags t o t he problemat ic part s) implement independent diagnosis implement independent diagnosis 8 improve t he source of problems and t he difficult part s(remove t he t ags where t he improvement job is done) implement depart mental diagnosis or high-end diagnosis 9 confirm and record t he part s t hat have been cleaned, lubricat ed and point -inspect ed(includes t he benchmark met hod, dealing met hod, period, t ime and t he managers) 10 implement independent diagnosis 11 implement depart mental diagnosis or high-end diagnosis (concluded by this research) 3.2. the steps of introducing pillar of independent maintenance into traditional industries that mainly rely on manual work because the products and industrial characteristics of each industry are different, they rely on equipment to different degrees. for exa mple, traditional industries that operate artificia lly have low re liance on equipments. some industries use simple equipments, which don’t break down easily and can restore from failure quickly even if a break down happens. this kind of industries includes assembly industry, electronic plug-in industry, woodwork and bamboo wares manufacturing industry, all of wh ich are based on manual labor. what this kind of industries wants to pursue is how to reduce the loss of manpower and increas e work efficiency is. the following is to elaborate the steps of introducing independent maintenance into traditional industries. 1. initial cleaning and improvement of sources of problems: the ma jor activities take place at on -site. all redundant items will be d isposed; the original material, semi-finished products, parts and smelting moulds will be cleared up; a thorough and comprehensive cleaning of on-site will be done in order to remove all dust and dirt: strategies will be made to solve problems regarding needless items, source of dust and parts that fail cleaning and point inspections. this belongs to the first stage and the estimated time needed is six months to one year. 2. make the contemporary benchmark for the imp le mentation of 5 s: ma ke action benchmark for effective ma intenance and cleaning carried out in short time. the imp rovement should be easy for the imp le mentation of point inspection and visual management. this belongs to the first stage and the needed time is estimated to be six months to one year. 3. overall inspection and the improvement of the problematic points: understand the structures and functions of products and provisions about quality assurance, find out the loss projects that lower efficiency of the personnel and improve them. (loss projects include loss caused by operating mistakes of workers, by operation, by logistics, by testing adjustment and so on). inspect and confirm the condition of moulds, smelting equipments and measuring instruments. set interim benchmarks for the daily overall inspection. this belongs to the second stage and the needed time is estimated to be six months to one year. 4. independent point inspection and standardized operations: review and revise the contemporary benchmarks for the imple mentation of 5 s and the interim benchmarks for the daily overall inspection in order to improve efficiency. train versatile workers and ensure the same quality and output even when there are changes in the operating staff. set operation standards and let it guide operations. this is to belong to the second stage and estimate import time six months to a year. 3.2. individual improvement of steps of importing pillar maintenance and the explanation for activities the purpose of individual improvement is to thoroughly eliminate the 16 kinds of loss caused by equipment, personnel and the raw materia l in order to ma ximize the productivity, and improve and make use of the technical and analysis ability of relevant personnel through the activity. eliminate losses. for e xa mple, for industries which rely on equipments, to eliminate losses is to bring down to zero the eight kinds of losses that lower the efficiency of equipments; and for those relying on artificial operation, that is to eradicate the loss that reduces workers’ efficiency. table 3 is the illustration for individual improvement of steps of introducing pillar ma intenance into traditional industries and the activities. table 3 the steps and activities of the indiv idual i mprovement of the pillar import st eps names main act ivit ies 1 select t he equipment , and project for demonst ration: ‧demonst rat ive equipment should mainly be chosen from those bott leneck project s in operat ion, equipment that wear a lot, important equipment and equipment or product ion lines t hat have bad horizontal expansion effect. proceedings of engineering and technology innovation, vol. 1, 2015, pp. 10 14 cop y right © taeti 14 table 3 the steps and activities of the indiv idual i mprovement of the pillar import (continued) 2 form an ad hoc group ‧the manager of the depart ment t hat this improvement case belongs to become the leader. the larger the case is, the higher the level should be ‧t he group should consist of managers, techn icians or designers and indirect personnel. ‧t he special case should get regist ered and reflect on the job schedule periodically 3 check and know about the current loss ‧affirm and mast er t he loss project s of the equipment and some relevant informat ion. when there is not enough informat ion, make the best effort s to collect them. (loss project s includes16 kinds of loss like failure, t emporary halt , replacement of mould and lines, amen dment of defect ive product s, loss caused by operat ing mist akes of workers, by operat ion, and by t est ing adjust ment. 4 set the t heme of improvement and t he goal ‧set t he t heme of improvement according to the findings of analysis (greater loss project s should get improved first ). ‧set the goal and period of improvement based on t he concept of zero-loss 5 work out plans for improvement ‧complet e a plan including analysis, count ermeasures and procedures of improvement . designat e managers of every operation 6 analyze problems and draw up count ermeasures ‧use every analyt ical method t o find out t he real reason t hat cause various kinds of losses and t hen draw up feasible resolut ion. 7 p ut int o effect improvement plans and check t he result s ‧carry out indispensable budge management and improve it. ‧aft er the improvement , check t he result s. keep improving those met hods that fail to work out well unt il t he goal is achieved. 8 st andardize the pract ice of improvement ‧st andardize the st andards for the manufact uring, operat ing, material-choosing and maintenance and make countermeasures for t he prevent ion of the recurrence of loss. make the manual for horizontal implementation. (concluded by this research) 4. conclusions this study draws up the mode of introducing independent maintenance into the traditional industries, which nowadays mainly rely on manual work. the mode includes contents like importing the pillar of independent maint enance, imple mentation steps, estimated schedule, indicators for performance measurement, managers of organizations for promoting independent maintenance, training courses needed, the production of import plan for independent ma intenance, the way of propaganda, diagnostic procedures of independent maintenance, the form summarizing the collected data and matters that should be noted in the importing process. research results will be provided to the domestic traditional industries for reference in order to reduce the cost and difficulty they have in importing independent maintenance and help them import tpm smoothly. so that the industrial structure of traditional industries in taiwan can be upgraded with improved competitiveness and profitability. with the researching results, this study hopes to achieve the following objectives: firstly, ma ke the traditional industries understand the definition, philosophy and obje ctives of independent maintenance. secondly, make the traditional industries understand the theory and practice of independent maintenance. thirdly, understand the current situation of the domestic traditional industries, and analyze their bottlenecks and the plight they are facing. fourthly, analyze the feasibility of introducing independent maint enance into domestic traditional industries and develop the import mode. fifth, analyze the estimated results of introducing independent maintenance into domestic traditional industries and the bottlenecks that may appear in the process. finally, it is hoped that the independent main tenance can be successfully imported into traditional industries, improving the struggling situation they have nowadays in the competitive environment. references [1] a. m. smith , re liab ility -cente red ma int enanc e , mcgraw-hill, 1993. [2] b. wu and j. seddon, “an anthropocentric approach to knowledge-based preventive maintenance,” journal of intelligent manufacturing, vol. 5, pp. 389-397, 1994. [3] e. a. si1ver and claude-nicolas fiechter, “preventive maintenance with limited historical data,” european journal of operational research, vol. 82, no.1, pp. 125-144, 1995. [4] g. j. hahn and s. s. shapiro, statistical model in eng ineering, wiley, 1994. [5] r. h. mcfadden, “developing a database for a reliability, availability, and maintainability improvement program for an industrial plant or commercial building,” ieee transactions on industry applications ,vol. 26, no. 4, 1990. [6] g. okogbaa, j. huang, and r. l. shell, “database design for predictive preventive maintenance system of automated manufacturing system,” co mputers and industrial engineering, vol. 23, no.1-4, pp. 7-10, 1992. [7] l. pintelonm, and l. v. wassenhove, “a maintenance management tool,” omega, vol. 18, no. 1, pp. 59-70, 1990. [8] s. s. shapiro and a. j. gross, “statistical modeling techniques,” dekker, 1981. [9] s. nakajima, “tpm: introduction to tpm, total productive maintenance,” p. 14, cambridge, mass productivity press, 1988. [10] v. alven, h. william, “reliability engineering (arinc research corporation),” prentice-hill, englewood cliffs, nj, 1964. [11] k. o. cua, k. e. mclone, r. g. schroeder, “relationships between imple mentation of tqm, jit, and tpm and manufacturing performance,” journal of operation management, pp. 675-694, 2001. [12] g. chand, b. shirvani, “implementation of in cellular manufacture,” journal of materials processing technology, pp. 149-154, 2000. [13] m. rodrigues , k. hatakeyama, “analysis of the fall of tpm in companies,” materials processing technology, pp. 276-279, 2006. [14] p. katila,”applying total productive maintenance-tpm principles in the flexible manufacturing systems,” technical report, lulea tekniska university, 2000: 23, 2000. [15] a. khanlari, k. mohammadi, b. sohrabi, “prioritizing equipments for preventive maintenance (pm) activities using fuzzy rules,” computers & industrial engineering, pp. 01-16, 2000. [16] l. x. ma, s. c. dong, y. q. yong, g. w. yu, “study on application of tpm in small and median -sized enterprises,” international conference on management science and industrial engineering (msie), pp. 678-681, 2001.  proceedings of engineering and technology innovation, vol. 12, 2019, pp. 39-46 secure intrusion detection system for manets using triple-des algorithm abhiram shashank n 1, * , bharathi c 2 1 department of computer science and engineering 2 saveetha school of engineering, saveetha university, thandalam, india received 23 september 2017; received in revised form 30 november 2017; accepted 03 february 2018 abstract networking is the process of accessing, exchanging or sharing the information. packet switching plays a vital role in data transfer. compared to wired network, wireless network has more data transferring. manets and wsn are the most common forms of wireless media; in manets nodes are deployed or distributed in ad-hoc way and they are communicating or exchange message using wireless transmission. security is a measure concern in mobile ad-hoc network because manets have wide distribution of node and open medium; therefore, it becomes vulnerable and is easy for malicious hackers to attack. keywords: mobile adhoc network (manet), intrusion detection system (ids) 1. introduction networks are a group of system networks linking together. there are two main classifications, peer to peer and client/server. networks can be characterized by their size and purpose. the size of the network is explained by geographical area that they occupy and the number of links that are part of the network. some networks based on size are; lan (local area network), wan (wide area network), man (metropolitan area network), pan (personal area network). some networks based on their purpose are; san (storage area network), vpn (virtual private network), manet (mobile ad-hoc network). in this paper we discussed about the intrusion detection systems on manets. mobile ad-hoc network or manet is an infrastructure-less ip based on network of mobile and wireless machine nodes. it is a type of ad-hoc network that can change locations and configure itself while moving. in manet each node act as a “router” to transmit the traffic to other specific node in the network. manets is very successive, attractive, and pervasive technology in wireless network. to maintain the mobility is an important task done by manets. manets is much more easier to be affected by different types of attack because it provides distributed architecture, volatile network topology, limited bandwidth of single hop and multi hope. in single hope, the entire node is in the defined coverage area, if there is intermediate node used for communication between two nodes is been called multi hop network. in manets there are two types of attack possible one is active attack and the other is passive attack. manets is used in emergency requirements because it allows easy deployment, minimal configuration, and low cost. however, it has restricted the battery power and resources. the aim of networking is to facilitate the exchange of data such as audio, text or video between various points across the world. for the delivery of data, various types of switching techniques are used in networking. the various types of switching techniques are packet switching, message switching and circuit switching. in this paper, we are dealt with packet switching. * corresponding author. e-mail address: abhiramsashank@gmail.com, cbharathi26@gmail.com proceedings of engineering and technology innovation, vol. 12, 2019, pp. 39-46 40 packet switching is a method of transferring the data in a form of packets to a network. to transfer the information fast and efficient, the data are broken into small pieces of variable length, called packet. the receiver has to reassemble these packets into a same file to retrieve the information. each packet carries the information of source and destination address, using them to travel independently through the network. the packets belonging to same files may or may not travel through same path. there are two types of packet switching; connection-oriented packet switching (virtual circuit) and connectionless packet switching (datagram). in this paper, we are dealt with connectionless packet switching (datagram). each packets in the datagram contains all necessary addressing information, such as source address, destination address and port numbers etc. in datagram each packet acts independently. packets that belong to same file may take different routes because of the dynamic routing technique; therefore, the packets arrived to destination are out of order. in fig.1, the recei ver send the packets to different transmission path and reach to different destination. in connectionless setup, packet delivery is not guaranteed; therefore, the receiver has to provide delivery acknowledgement using additional protocols and request for missing packets. packet switching is cost beneficial because the devices do not need massive amount of secondary storage. the delays in packet switching are: transmission delay, propagation delay, queuing delay, and processing delay. transmission delay is simply time required to put data bits on the wire/communication medium, and it depends on length of packet and bandwidth of network. propagation delay is time taken by the first bit to travel from sender to receiver and it depends on distance and propagation speed. queuing delay is time where a process waits in a queue until it can be executed. it is the time difference between the arrival of packet and the start of the data process. processing delay is the time required to process the packet’s header and determine where to direct the packet. 2. internet attacks networks are often subjected to attack without any security measures, example-internet. there are two types of attacks;  active attack  passive attack active attack is a network attack that the intruder tries to alter the information on data that has being transferred. passive attack is a network attack in which the information is not altered but monitored or overheard while transmission. the both active and passive attacks are further classified. fig. 1 packet switching – connectionless (datagram) the types of active attacks:  masquerade  replay  denial of service attack  modification of messages the types of passive attacks: proceedings of engineering and technology innovation, vol. 12, 2019, pp. 39-46 41  traffic analysis  eavesdropping  man in the middle attack 2.1. active attacks 2.1.1 masquerade attack it is an attack that uses fake identity. the attackers pretends to be an authorized user of a system to gain access to personal computer information, such as stolen ids and passwords. if the authorization is not fully protected, then it became vulnerable to masquerade attack. once the attackers entry, they have a full access. this attack occurs more in an organisation where they modify data or software, and may also alter the network configuration and routing information. 2.1.2 replay attack it is a network attack that the intruder detects data transmission and make the transmission maliciously delayed or repeated, i.e. a replay is an attack on security protocol that replays the data transmission from a sender to the intended receiver, so as to fool the participants to believe they successfully completed the transmission of data. the message is encrypted correctly so that the receiver mistake it as a correct request and does the actions provided by the intruder. 2.1.3 denial of service attack it is a type of cyber attack which the intruder makes the network resource temporarily or indefinitely unavailable the user's service or indefinitely disrupting service. it prevents the users from accessing the specified computer systems or networks. denial of service attack floods the servers with traffic, in order to make it difficult for the users to connect them. attackers mostly attack web servers of high-profile organisations. dos attacks do not led to the loss of information, but the the users have to deal with money and time complexity. 2.1.4 modification of messages it is an attack which the intruder alters packet header address, to direct the packets to different destination intended by the intruder or change the data and send them to to the targeted machine. this also includes deletion of data, alteration or insertion in an unauthorized manner. 2.2. passive attacks 2.2.1 traffic analysis attack it is a type of attack against encryption. the common way to secure the contents is encryption. the intruders could observe the traffic patterns of these messages. the attackers’ posses some basic information about network, i.e. the activity of the network, protocol being used and also access points of the network. the attackers can determine the location, the host and also the frequency and length of messages being exchanged. traffic analysis can determine what type of data is being communicated even if the data is scrambled or encrypted. the size of packets being exchanged between two hosts can also be valuable information for an attacker, even if the contents can’t be viewed. traffic analysis is very difficult to detect because they do not involve any change in the data. 2.2.2 eavesdropping attack real-time interception of private communication is unauthorized, i.e. phone call, instant message, fax, video call. it also weakened between client and server that allows intruder to send traffic to it. this type of network attack is due to lack of encryption services. attackers can install network monitoring software on a computer to carryout eavesdropping and intercept data. using firewalls, antivirus software, and virtual private networks and avoid sending sensitive information on public networks prevent eavesdropping attacks. 2.2.3 eavesdropping attack it is a type of cyber attack which the intrudergains information on data between two parties by eavesdropping or proceedings of engineering and technology innovation, vol. 12, 2019, pp. 39-46 42 impersonating. mitm attack is a serious threat to online security because the attacker have is able to capture and manipulate sensitive information in real-time. mitm attacks distributes malware that provides the attackers access user’s web browser and data that are exchanged. online banking and e-commerce are sites that mitm attackers target. 3. acknowledgment schemas there are numerous ids (intrusion detection system) produced for giving security. the various approaches to find ids are, 3.1. ack (acknowledgement) schema in this schema, the receiver sends an ack in form of signal to the sender as a receipt that the data is transmitted to destination successfully. the ack is received after the transmission of every data block. the sender sends the next block of data only after it receives ack from previous block of data. it is an indication that the data been transmitted has been received uncorrupted and there’s no errors of receiving station. 3.2. nack (negative acknowledgement) schema in this schema, the sender receives n-ack or rej (rejection) when the file received by the receiver is corrupted. an alternate signal arq (automatic request for retransmission) is automatically generated when there is n-ack. when source receives nak or rej signal from the destination, the source either re-transmits the data or stops the transmission process entirely. 3.3. watchdog schema this schema detects the malicious node which is unsafe to establish connection. it is has two operations: phase1-watchog, phase2-pathrater. watchdog uses next hops transmission for detecting malicious attack. however, it increases the failure if the next node fails to establish transmission within intended time interval. pathrater make sure the malicious node been detected by watchdog is not been used later or for future purpose. 3.4. twoack schema in this schema, the twoack uses three consecutive nodes to check whether the packet is transmitted successfully. it is achieved by acknowledging the data over every three consecutive nodes. in other hand, manets allows the node to self organize for functioning, the performing nodes consumes energy and other resources. therefore, the network nodes may decide to be non co-operative with other nodes to perform well individually which affects other nodes. this kind of nodes is said to be selfish nodes or misbehaving nodes. the schema makes sure that nodes are not used later or future purpose. 3.5. aack (adaptive acknowledgement) schema this schema is the combination of twoack and ack. it has end-end transmission. source sends the packets through nodes to the destination. once received, the receiver sends an ack in the reverse order. this process should be done within predefined interval; if not, then it will switch to twoack process. 4. cryptography algorithms cryptography algorithms can be fundamentally divided into  symmetric  asymmetric  hash functions in this paper we are dealing with symmetric algorithms. 4.1. symmetric algorithm proceedings of engineering and technology innovation, vol. 12, 2019, pp. 39-46 43 symmetric algorithm encrypts and decrypts a data using the same key. anybody who holds a key can exchange message with anybody else holding the same key. once intruder get the key, they can access the message; therefore, the key must be shared secretly. the most commonly used symmetric algorithm is des. 4.4.1. data encryption standard (des) fig. 2 data encryption standard it is a symmetric-key block cipher. it is an implementation of a feistel cipher. the block size is 64-bit but the key length is 56 bits, since 8 of the 64 bits key is not used by the encryption algorithm as shown in fig. 2. the process or steps involved in fiestal cipher are:  round function  key generation fig. 3 enhanced adaptive acknowledgement (eaack) round function: the cipher follows des function. the des function applies 48-bits key to the rightmost 32-bits to produce a 32-bit output. in round function p-boxes and s-boxes are used, also known as initial and final permutation. permutation-box (p-box) is a method of bit-shuffling used to transpose bits across s-boxes inputs. p-boxes are used for expansion, compression and straight accordingly with respect to number of output bits that is less than, greater than or equal to the number of input bits. substitution-box (s-box) is a vital component of symmetric key algorithms which performs substitution. s-box takes m number of inputs and transforms them into n number of outputs, where n is not necessarily equal to m. key generation: des input key size is 64 which contain 56 bit key and 8 bit parity. parity bits are 8th bit of every 8 bit (one byte). therefore, they are the multiple of eight: 8, 16, 24, 32, 40, 48, 56 and 64. permuted choice pc-1 is used to remove proceedings of engineering and technology innovation, vol. 12, 2019, pp. 39-46 44 these bits from the 64 bit input key. so pc-1 gives 56 bits as output. 5. mechanism the mechanisms used in this paper are  eaack  hybrid cryptography  triple des algorithm 5.1. eaack  acknowledgement (ack)  s-acknowledgement (s-ack)  misbehaviour report authentication (mra) 5.1.1 acknowledgement (ack) refer to fig.3. here the ack schema acts as a part of the hybrid scheme in eaack to reduce network overhead. 5.1.2 s-acknowledgement (s-ack) in this acknowledgement schema, the receiver can inform the sender about all segments of data that has been received successfully. the sender understands the drop of packets when the particular segment is not acknowledged. thus the sender only needs to re-transmit the segments that has been lost. 5.1.3 misbehavior report authentication (mra) this schema is able to detect malicious nodes which produce false misbehavior report. false behavior report is generated by intruders when they break down some nodes and divert the network. mra verify whether the report on false misbehavior is received by different node. when the destination node receives mra packet, it searches through its local knowledge base and compare whether the report packet is received by another node. if so, then it is understood that this false misbehaviour report and node is marked malicious. 5.2. hybrid cryptography fig. 4 hybrid cryptography a hybrid cryptosystem is the one which combines the convenience of a public-key cryptosystem with the efficiency of a symmetric-key cryptosystem. in this project, we used triple des algorithm as a symmetric-key cipher, which applies the des cipher algorithm three times to each data block as shown in fig. 4. it is very easy to modify the existing software to use triple des. there is also the advantage of reliability and longer key length that eliminates many of the attacks that can be used to reduce the amount of time it takes to break des. triple des provides a relatively simple method of increasing the key size of des to protect against such attacks, without the need to design a completely new block cipher algorithm. 5.2.1 triple des algorithm triple des is based on the des (data encryption standard) algorithm. des has proven reliability and a longer key length proceedings of engineering and technology innovation, vol. 12, 2019, pp. 39-46 45 to prevent many of the internet attacks. however even this algorithm is not strong enough to secure the data very long. thus triple des was endorsed as a replacement for des. the triple des uses the key of size 56 which runs the algorithm in succession with three different keys. the result of the key size will be 168 bits. the triple des follows encryption-decryption process as follows  encrypt the plaintext using single des-key k1  then decrypt the output of step 1 using single des-key k2  finally encrypt the output of step 2 using single des-key k3  the final result is the cipher text  decryption process is just reverse of encryption process 5.3. benefits  eaack gives better performance than other schema.  eaack uses digital signature it causes routing overhead.  hybrid cryptography is used to reduce routing overhead by detecting malicious path. using shared key source node and destination node authenticate to transfer data packet. 6. conclusions ids (intrusion detection system) is important for providing security to manets it use acknowledgment schema such as twoack, ack, aack, eaack to avoid the defect. eaack gives better performance than other schema. eaack uses digital signature and causes routing overhead. therefore, this paper proposes techniques to reduce routing overhead by detecting malicious path. using shared key source node and destination node transfer data packet. conflicts of interest the authors declare no conflict of interest. references [1] p. joshi, p. nande, a. pawar, p. shinde, and r. umbrae, “eaacksecure intrusion detection and prevention system for mantes,” in pervasive computing (icpc), international conference on, january 2015. [2] s. s. jathe, v. dhamdhere “hybrid cryptography for secure superior malicious behavior detection and prevention system for manet’s,” international journal of innovative research in science, engineering and technology (an iso 3297:2007 certified organization) vol. 4, july 2015. [3] r. meher, s. ladhe “review paper on flooding attack in manet,” ruchita meher et al int. journal of engineering research and applications, vol. 4, pp. 39-46, january 2014. [4] p. d. nikam, v. raut “attacks prevention and detection techniques in manet: a survey,” pranjali d. nikam int. journal of engineering research and applications, vol. 4, pp. 15-19, november 2014. [5] b. suruthi, n. v. r. kumar, and m. tech. “an enhanced intrusion detection system for manets using hybrid key cryptography,” vol. 5, 2014. [6] a. p. tapaswi, p. p. nashikkar, r. s. barge, a. m. shinde, and p. p. patil “extended eaack an secure intrusion detection system with detection and localization of multiple spoofing attackers in manet,” vol. 4, pp. 39-46, january 2014. [7] h. khatri, a. gupta, and d. pal “mitigation of http-get flood attack” international journal for research in applied science & engineering,” vol. 2 november 2014. [8] ramya k, beaulah david and shaheen h, “hybrid cryptography algorithms for enhanced adaptive acknowledgement secure in manet,” vol. 16, february 2014. [9] e. m. shakshuki, n. kang, and t. r. sheltami, “eaack a secure intrusion-detection system for manets,” ieee transactions on industrial electronics, vol. 60, march 2013. proceedings of engineering and technology innovation, vol. 12, 2019, pp. 39-46 46 [10] s. m. lakshmi, s. bhavana, t. sujata, “enhancement of security levels using a secure intrusion detection system in manets,” k. anusha, and g.v. rajyalakshmi,” secure adaptive acknowledgement algorithm for intrusion detection system” int, vol. 16, may 2014. [11] b. thanikaivel, b. pranisa, department of computer science and engineering, sona college of technology salem, tamil nadu, india, “fast and secure data transmission in manet,” 2012 international conference on computer communication and informatics (iccci 2012), coimbatore, india, january 2012. [12] d. sandhiya, k. sangeetha, r. s. latha, “adaptive acknowledgement technique with key exchange mechanism for manet,” 2012. [13] n. kang, e. shakshuki, and t. r. sheltami, “detecting forged acknowledgements in manets,” in proc. ieee 25th int. conf.aina, biopolis, singapore, pp. 488-494, march 2011. [14] r. ferdous, v. muthukkumarasamy, and a. sattar, institute for integrated and intelligent systems griffith university, australia, “trust management scheme for mobile ad-hoc networks” 10th ieee international conference on computer and information technology, 2010. [15] b. pahlevanzadeh, s. a. hosseini seno, t. c. wan, r. buldiarto, and m. m. kadhum, “a cluster based distributed hierarchical ids for manets,” international conference on network applications, protocols and services, pp. 1-7, november 2008. [16] k. liu, j. deng, p. k. varshney, and k. balakrishnan, “an acknowledgment-based approach for the detection of routing misbehavior in manets,” ieee trans. mobile computer, vol. 6, pp. 536-550, may 2007. [17] b. pahlevanzadeh, and a. samsudin, “distributed hierarchical ids for manet over aodv” ieee international conference on telecommunications and communications, pp. 99-104, may 2007. [18] g. jayakumar and g. gopinath, “ad hoc mobile wireless networks routing protocol-a review,” j. comput. sci., vol. 3, no. 8, pp. 574-582, 2007. [19] s. marti, t. j. giuli, k. lai and m. baker, “mitigating routing misbehavior in ad-hoc networks,” pp. 255-265, 2000. [20] t. anantvalee and j. wu, “a survey on intrusion detection in mobile ad hoc networks,” in wireless/mobile security, new york, 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/). microsoft word 6-peti#4130 45-51.docx proceedings of engineering and technology innovation, vol. 16, 2020, pp. 45 51 path control irobot create2 based sensors narong aphiratsakun 1,* , santi liwsakphaiboon 2 1 department of mechatronic engineering; 2 department of computer engineering, assumption university, samut prakan, thailand received 13 january 2020; received in revised form 06 may 2020; accepted 10 august 2020 doi: https://doi.org/10.46604/peti.2020.4130 abstract robots become a part of humans’ daily lives. they work faster, more accurate, and tireless compared to humans. as technology developes, robots are more capable of serving people in more tasks than before. the objective of this study is to explain the methodology of controlling irobot create2 by using arduino uno to command irobot create2 to use operation codes. instead of using a brute-force way to control the robot motion by commands, which may cause the whole path to go wrong if only one command is deflected. however, with the sensors, the robot can adjust the direction path despite the deviation. the arduino uno communicates with irobot through the serial communication using softwareserial. arduino requests a sensors data from irobot and calculates deviations in order to adjust the path. the study includes the general knowledge of arduino control irobot create2 based on the robot’s sensors and the detailed process of completing the first round of irobot competition. keywords: arduino, coding, control, irobot create2, sensors 1. introduction irobot create2 [1] is an affordable stem resource for educators, students, and developers to learn robotic control. it is a vacuum cleaning robot that is programmable and is used to grasp the fundamentals of robotics, computer science, and engineering [2-3]. some previous projects about irobot controlling are camera bot [4] pairing your create2 with a microcomputer, a camera then driving around and seeing what your robot sees. dj create [5] is a roaming of the robotic dj for your party by integrating irobot with a bluetooth module. however, most of the programs, like dj create, cannot control the path of irobot effectively. irobot company set the robot controlling competition [6], the assumption team took part in and passed the first round, also qualified for the final round of this competition. this study shares their experience and details in the robot controlling and the method of how to achieve. 2. system 2.1. irobot create2 irobot create 2 in fig. 1 is explicitly designed for a development of the robotics. the irobot create 2 replaces its roomba predecessor’s vacuum cleaner hardware with a cargo bay that also houses a db-25 port to provide a serial * corresponding author. e-mail address: oofpokemon@gmail.com tel.: +6687-349-3482, +4176-223-0109 proceedings of engineering and technology innovation, vol. 16, 2020, pp. 45 51 46 communication, digital input & output, analog input & output, and an electric power supply. the irobot create2 also has a 7-pin mini-din serial port connector which sensor datas can be read and motor commands can be issued using the irobot roomba open interface (roi) protocol. fig. 1. irobot create2 2.2. mini-din connector to use the oi, a processor which is capable of generating serial commands such as a pc or a microcontroller must be connected to the external mini-din in fig 2 connector on roomba. this connector provided two-way, serial communication at ttl (0-5v) levels. the connector also provided an unregulated direct connection to roomba’s battery, which can be used to power the oi applications in table 1. fig. 2 mini-din connector table 1 mini-din pin table pin name description 1 vpwr roomba battery + (unregulated) 2 vpwr roomba battery + (unregulated) 3 rxd 0 – 5v serial input to roomba 4 txd 0 – 5v serial output from roomba 5 brc baud rate change 6 gnd roomba battery ground 7 gnd roomba battery ground 2.3. controller: arduino uno fig. 3 arduino uno r3 proceedings of engineering and technology innovation, vol. 16, 2020, pp. 45 51 47 arduino uno microcontroller board in fig. 3 is based on the microchip technology atmega328 8-bit microcontroller (mcu). arduino uno features 14 digital input/output pins (six of which can be used as pwm outputs), six analog inputs, and a16mhz quartz crystal. the atmega328 on the uno comes preprogrammed with a bootloader that allows the user to upload a new code to the mcu without the use of an external hardware programmer. 3. task fig. 4 stage 3-d model the competition’s rules and qualifications are as followed: 1. the robot used in the competition has to be irobot create2 which is given by the organizer at the beginning, then the robot must not be modified and/or added any sensor. 2. only arduino uno rev 3 is allowed to be used as a controller board. 3. the size and the position of objects (including micro bricks) in the stage are known before the competition day. (figs. 4-5) fig. 5 position of micro bricks the task is divided into two stages: the first stage is for testing simple robot control skills and the second stage requires more control skills to precisely pick up yellow micro bricks and dodge brown micro bricks. each stage’s rules are listed below. the first stage rules: (1) when the robot reaches the first checkpoint, it gains 150 scores. (2) one chance for retrying if the robot falls from the stage cliff. start b zone a start a zone b check point a check point b proceedings of engineering and technology innovation, vol. 16, 2020, pp. 45 51 48 (3) the robot hits the wall of the stage, it loses 25 scores per hit. maximum loss 100 scores. (4) two minutes is the limit for the first stage. if exceeds, 0 score for first stage and can proceed to the second stage. (5) the robot can either continue moving right after reaches the first checkpoint or stop. the second stage rules: (1) the robot reaches the destination area and the whole parts of the robot exits from the stage, it gains 100 scores. (2) the robot hits the wall of the stage, it loses 25 scores per hit. maximum loss 100 scores. (3) three minutes is the limit for the second stage. if exceeds, 0 score for the second stage. (4) gain 50 scores per yellow micro brick (fig. 6(a)) is collected (total 5 yellow micro bricks in stage). (5) lose 50 scores per brown micro brick (fig. 6(b)) is collected (total 5 brown micro bricks in stage). (6) if it does not use retry chance in the first stage, it can be used in the second stage instead. the full score is 500 scores, the ranking is determined from the highest to lowest and fastest to slowest if the scores are equivalent. (a)yellow micro bricks (b)brown micro bricks fig. 6 micro brick 4. control method fig. 7 serial communication serial communication is the process of sending a data one bit at a time, sequentially, over a communication channel or computer bus. it is available in arduino uno (known as uart) to be used for the communication between the arduino board and other devices. fig. 8 arduino-create2 connection arduino uno r3 serial communication irobot create2 proceedings of engineering and technology innovation, vol. 16, 2020, pp. 45 51 49 the methodology of design a path control for irobot used sensors to check the position, fig. 7 used sensors of irobot to receive data then sended to arduino, and arduino made the decision which direction to go by sending opcode to irobot through wires (fig. 8). 5. control program the program in arduino was divided into 2 part for 2 stages and used the button on irobot to run the program (fig. 9). fig. 9 irobot control flowchart below is the flowchart and explanation for each program. 5.1. the first stage flowchart (fig. 10) (1) after the program starts, first it resets the flag, sets distance and radius for irobot. (2) next, arduino checks the value from the wheels encoder if irobot reaches the distance, goes to check flag state. (3) if irobot doesn’t reach the distance set, arduino requests a wall sensor and a cliff sensor from irobot. (4) if detects the sensor of a wall or cliff, arduino sends command to irobot to adjust the direction angle to left or right. (5) if doesn’t detect the sensor of a wall or cliff, irobot keeps driving straight. (6) if reaches the distance, arduino will check the flag status. (7) if a flag is not set, turns irobot in radius and set the flag. (8) if a flag is set, stops irobot and exits the program. 5.2. the second stage flowchart (fig. 11) (1) once program starts, irobot moves out from the start point, turns the counter-clockwise, then initializes all variables. (2) request the wheels encoders to compute the traveled distance and check if it equals to the distance set or not. (3) if doesn’t reach the distance, checks for wall, here irobot will keep moving along the wall. otherwise, adjust the direction. (4) when reaches the distance set, irobot moves in a sequence of the actions to collect the last micro brick as it position is not near the wall. proceedings of engineering and technology innovation, vol. 16, 2020, pp. 45 51 50 fig. 10 stage 1 program flowchart fig. 11 stage 2 program flowchart 6. experiment and result the programs were tested 30 times per program. for the first stage (fig. 12), the accuracy of irobot moving path is 67% (20 success out of 30 times). the success here means irobot can move from the start area to the checkpoint area without hitting the wall or falling from the cliff. most of the errors occur is that the robot does not move in the expected direction. sometimes it does not adjust the angle and hit the wall. fig. 12 first stage testing fig. 13 second stage testing the root cause of this error happened when a duty cycle of the motor speed is more than 58%, irobot receiving the data speed is slow down. the problem can be solved in 2 ways. the first method is by using arduino leonardo which has 2 internal modules uart (rs-232). the first uart connects to usb and the second one connects to pin0 and pin1 of the board. this way, arduino can communicate with irobot without using the softwareserial. the second method is simply to adjust a duty cycle of the motors below 58%. proceedings of engineering and technology innovation, vol. 16, 2020, pp. 45 51 51 for the second stage (fig. 13), the test result is that the robot can collect all yellow micro bricks 18 times out of 30 times without collect any brown micro brick. collect 4 out of 5 yellow micro bricks (0 time)and collect 1 out of 5 yellow bricks (3 times). 7. conclusion this study explained the method of controlling irobot create 2 using arduino uno command and opcode. the control flowchart was used to qualify for the final round of irobot competition. in the competition, irobot create 2 passed the first stage with a full score (plus 150 marks) and managed to collect 4 yellow micro bricks (plus 200 marks) as robot hit the wall 4 times in the second stage, 100 scores are deducted for the penalty, resulting in having 350 scores. total time using for stage1 and stage2 is 125.30 seconds. using the sensors and wheels encoders is efficient but will encounter a serial communication problem that it is different between the arduino uno’s speed of receiving the sensor data speed and irobot’s sending speed. for further improvement, the control program is needed to be improved in order to increase the efficiency of the robot. acknowledgment we wish to thank smith simargool and mustafa dadarananda for their assistance with the testing statistic used in this paper. we would like to acknowledge the support provided by assumption university during the preparation before competition. conflicts of interest the authors declare no conflict of interest. references [1] irobot, “irobot® create® 2 open interface (oi) specification based on the irobot® roomba® 600,” https://www.irobotweb.com/-/media/mainsite/files/about/stem/create/2018-07-19_irobot_roomba_600_open_inte rface_spec.pdf. [2] irobot, “create 2 programmable robot,” https://edu.irobot.com/what-we-offer/create-robot. [3] “irobot create,” http://en.wikipedia.org/wiki/irobot_create. [4] irobot, “camera bot,” https://www.irobotweb.com/-/media/mainsite/pdfs/about/stem/create/camerabot.pdf. [5] irobot, “dirt cheap dj create® 2 bin,” https://www.irobotweb.com/-/media/mainsite/pdfs/about/stem/create/djcreate.pdf. [6] n. aphiratsakun, “irobot create – the next chapter of smart living,” pp. 19, 22-23, 2018. [7] throbotics, irobot rules of the competition of round 1, 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/). microsoft word 6-v24(2023)-peti#10200(53-62).docx proceedings of engineering and technology innovation, vol. 24, 2023, pp. 53-62 design optimization of a capacitive sensor for mass measurement of nanometer-sized exhaust carbon particles vaishali sanjay kulkarni1,*, suvarna sandip chorage2 1department of electronics and telecommunication engineering, all india shri shivaji memorial society’s institute of information technology, savitribai phule pune university, pune, india 2department of electronics and telecommunication engineering, bharti vidyapeeth’s college of engineering for women, savitribai phule pune university, pune, india received 29 may 2022; received in revised form 22 february 2023; accepted 27 february 2023 doi: https://doi.org/10.46604/peti.2023.10200 abstract nanometer-sized carbon particulates generated by incomplete combustion in heavy-duty vehicles are harmful to human health. a high-resolution technique is needed to detect and measure these pollutants. this study aims to optimize a capacitive sensor design for detecting and measuring particulates. firstly, the effect of design parameters on particulate detection and sensor compliance sensitivity is investigated by using the finite element method. by comparing the simulation results with literature findings for performance validation, the sensor structure is optimized to detect lower particulate concentrations. the simulation result shows that particulate detection sensitivity has linear variations with changes in particulate mass. with optimum electrode spacing and top insulation layer thickness of 5 µ m, the sensor can detect a particulate deposition of 0.033 mg/min and generate a maximum capacitance of 581 pf. since the optimized design can measure particulate deposition at a lower range and with higher sensitivity, it is suitable to be applied to detect nanometer-sized carbon particulates. keywords: capacitive sensing, nanometer-sized particulate mass, capacitive sensor, mass measurement 1. introduction carbon particulates are a major contributor to environmental pollution and ozone layer damage, ranging in size from 20 to 100 nanometers. these particulates may include soot particles, particulate matter, or carbon black, and pose a severe health risk to individuals. a particulate filter is used in vehicle systems to reduce emissions of this pollutant. exhaust systems detect carbon particles using pressure sensors which measure pressure differences due to particle accumulation. a pressure difference is, therefore, an indication of particulate filter failure. an alternative manual method of measuring particulate mass uses nanometer-sized filter paper in smoke collection machines to collect particulates. this method estimates particulate mass manually by identifying the presence of particulates on the filter paper. the existing particulate detection method uses resistance-based sensors that serve only as a detection facility but cannot measure the particulate mass [1]. resistance-based sensing involves either an accumulative electrode or a charge-based deposition method [2], with large-sized particle tracking being complex. it is, therefore, vital to consider the lagrangian and eulerian modeling methods for gas particulate modeling [3]. while eulerian modeling is less complicated, it is more timeconsuming than lagrangian modeling [4]. the literature mainly uses electromechanical, interdigitated electrode systems for gas sensing due to their minute dimensions, with these electrodes resembling the interlocking fingers of two clapped hands * corresponding author. e-mail address: vaishsan2000@gmail.com proceedings of engineering and technology innovation, vol. 24, 2023, pp. 53-62 54 and having an electric field used to build capacitance [5-6]. capacitive sensors are ideal for their low static power consumption [7-9] and find use in various fields, such as biomedical, automotive, robotics, and gas sensing. capacitive sensors can also detect and measure volatile organic compounds [10-12]. microsensors have been widely used in sensing since the 1960s [13], indicating a need for a systematic study of sensor models, features, and design issues. further research needs to also explore the factors that influence sensitivity distribution and the impact of conductivity on the performance of capacitive sensors. the electrode geometry significantly affects the sensor’s sensitivity [14]. an electronic nasal pod (enp) is a wearable device capable of preventing allergies and sensing air pollutants. to fabricate a 3d-printed enp, a high-efficiency particulate air (hepa) filter is integrated for sieving pollutants at the first stage. the enp uses three electrodes to sense oxygen electrochemically using a potentiostat [15]. a novel pressure sensor has also been developed to perform sensing based on changes in the overlapping areas of electrodes rather than their separation. this sensor, which has a square membrane of 2.5 mm × 2.5 mm, can provide a total capacitance change of 0.365 pf for an applied pressure of 0-100 kpa [16]. insights into the trade-off between sensor design parameters and sensor requirements were revealed in a study that focused on optimizing capacitive sensors [17]. an innovative capacitive moisture sensor was designed for polymers to demonstrate the relationship between the sensor’s geometrical parameters and performance [18]. despite the limitations associated with some commonly used microelectromechanical systems (mems) design methods, evolutionary computations have recently become famous for designing and optimizing mems [19]. a model based on a mems capacitive sensor demonstrated that finite element analysis is the most accurate technique for estimating sensor capacitance [20]. interdigitated capacitive sensors were used in mining environments to monitor real-time nanoscale particle concentrations, with simulations indicating that the sensor response is proportional to the number of particles accumulated on the sensor [21]. comsol, a tool for modeling and simulating sno2/rgo gas sensors, was used to detect and characterize no2 [22]. a complete mass sensing module was achieved by integrating a tpos mems oscillator, mems, circuit, and air flow technologies to develop a pm2.5 sensor [23]. a thin film circular diaphragm was developed for mems resonant mass sensors that detect airborne particles, demonstrating real-time detection of standard silica microspheres [24]. to measure pm2.5 and pm10, an air pollution monitoring device with a particle matter sensor was used [25]. a study which investigated low-cost pm sensors proposed methods to correct the error factors caused by the optical scattering method [26]. a piezoelectric resonating membrane operates the gravimetric-based particulate matter sensor for portable applications, and future work includes implementing a thermophoresis mechanism to remove residual particles deposited on the sensor during measurements [27]. while few modeling and simulation reports exist on sensors designed for detecting and measuring carbon particulates, there is ample scope for developing micro-level devices to accomplish this. this study shows the result of the design optimization of the proposed particulate sensor and evaluates its effectiveness by interpreting the results. the study focuses on predicting particle deposition and the corresponding electric field strength of particulate sensors through computational fluid dynamics (cfd) and electric field simulations. the work is novel because it optimizes sensor design parameters and measures the particulate mass with high precision to a minimum value. however, there were several challenges, such as determining the period of particulate deposition and integrating the deposited mass with electrical simulations. much of the effort was devoted to the mathematical modeling of cfd and the execution of electric field simulations. modeling the nanoparticle sensor in both the electrical and mechanical domains was challenging, especially due to the fragile top insulation layer, which had a thickness of 5 µm. however, this layer is crucial for achieving high sensitivity. the results of the sensor model development, simulation method, simulation cases, and particulate detection are presented in detail. proceedings of engineering and technology innovation, vol. 24, 2023, pp. 53-62 55 2. an optimized sensor design dimensions fig. 1 presents the sensor layout with dimensions, cross-section, and materials. the dimensions are optimized through iterations to obtain capacitance with good resolution. the sensor substrate size is mm × 10 mm, while the sensor structure is an interdigitated electrode with an electrode length of 4300 µ m, spacing of 5µ m, and width of 100 µ m. the sensor uses 38 electrodes. fig.1 (b) shows a cross-section of the optimized sensor with a substrate and top insulation layer of si3n4. (a) optimized dimensions (µm) (b) cross-section of idc fig. 1 the layout of interdigitated electrode capacitor (idc) [2] 2.1. methodology, material properties, and boundary definitions fig. 2 shows a two-dimensional (2d) model of a single-unit cell. physical materials and boundaries used for simulations are defined as shown in fig. 3. for the design requirements and a temperature of 473℃, the relative permittivity values are 3.6 for silicon dioxide, 9.5 for silicon nitride, and 1 for air. the analysis considers deposition in accumulative mode with the particulate diameters ranging from 0.01 mm to 100 mm. the simulations use a gas flow velocity of 38 m/s to 47 m/s and a particulate density of 1150 kg/m3. according to the calculations, the estimated density of the mixture (air plus soot) is 0.48109 kg/m3. the soot concentrations in terms of volume and mass are 8.0 × 10-06 and 0.019123, respectively. fig. 2 2d model of one unit cell for analysis and simulation of electrostatic fields fig. 3 model of a sensor unit cell with a particulate layer for analysis with defined boundaries the static analysis applies a boundary condition of zero charges to all external edges of the sensor. the voltages at sensor terminals 1 and 2 are 0 v and 12 v, respectively. the parametric sweep function in comsol allows the particulate layer to gradually increase from 0.1 µm to 2 µ m with a step size of 0.1 µ m and particulate diameters between 10 µ m and 100 µm. as a result, the electric field and capacitance values are subject to variation. proceedings of engineering and technology innovation, vol. 24, 2023, pp. 53-62 56 2.2. design requirements (1) optimize the sensor design dimensions to extract the maximum capacitance value. (2) obtain high sensitivity to low particulate deposits. (3) detect and estimate time intervals of particulate deposition. (4) withstand temperatures up to 400℃. (5) optimize the design of experiments (doe) study to predict change in the capacitance for particulate disposition. 3. equations governing physics the two components of the analysis are the cfd and the electrical. the cfd component analyzes the flow characteristics to predict the mass of particulates at the sensor surface. the electrical component evaluates the sensor’s effectiveness to predict its electrical properties. the electrostatics domain and cfd model are closely coupled due to the influence of particulate mass. ( ) ( ) ( ) ( ) ( ) 2 3 t u u pl t vu u t u l f   × ∇ = ∇ − + + + ∇ − + ∇ × +    ρ µ µ µ µ (1) 0p u∇ × = (2) ( ) t u k k pk µ ρ µ ρ σ ∈    × ∇ = ∇ + ∇ + − ∈       (3) ( ) 2 1 2 t u k c pk c k k µ ρ µ ρ σ ∈ ∈ ∈    ∈ ∈ × ∇ = ∇ + ∇ ∈ + −       (4) e p∈= × (5) 2 t u k c= × ∈ µ ρ (6) ( ) ( ) 22 2 : 3 3 t t pk u u u u k uµ ρ   = ∆ ∆ + ∆ − ∆ × − ∆ ×    (7) � = density (kg/m3 ), t = absolute temperature (k), p = pressure (pa), u = velocity vector (m/s), � = dynamic viscosity (pa.s), f = volume force vector (n/m3), v = velocity (m/s), ep = turbulent dissipation rate, q = heat flux by conduction (w/m2), k = thermal conductivity (w/m.k) (heat transfer), k = turbulent kinetic energy (turbulent flow), cp = specific heat (j/kg.k). the equations presented in this section describe the physics underlying the computational models. the equations solved are the reynolds-averaged navier-stokes (rans) equations for the conservation of momentum and the continuity equation for the conservation of mass [23]. the solver calculated that a single particulate would take 5.7 ms to reach the sensor. therefore, the transient solver’s period is set to 0-15 ms, with a time step of 0.5 ms. the modeling uses an electrostatic (es) interface to calculate the electric field, displacement field, and distribution of the dielectric potential. for in-plane 2d modeling, the electrostatic interface assumes symmetry, and the electric field, e, is tangential to the xy plane. e v= −∇ (8) the electric field, v, is defined by eq. (8) in static conditions. adding eq. (8) to the constitutive relationship � = ��� + . proceedings of engineering and technology innovation, vol. 24, 2023, pp. 53-62 57 ( )0 v p−∇ × ∇ − =ε ρ (9) the modeling presents gauss’ law as a relationship between the electric displacement, d, and the electric field, e, given by eq. (9). the permittivity of vacuum is ε0 with si units of f/m, the vector of electric field polarization is p, with si units of c/m2, and the density of space charges is ρ, with si units of c/m3. the equation describes the electrostatic field of a dielectric material. for in-plane 2d modeling, the electrostatic interface assumes symmetry, and the electric field, e, is tangential to the xy plane. ( )0d v p−∇ × ∇ − =ε ρ (10) the thickness of the z-direction given by d and the physics interface solves this eq. (10). 4. particulate mass estimation fig. 4 the deposited mass on the sensor face (0 to 15 ms) fig. 5 results of doe s-curve behavior a steady-state analysis is performed on the sensor model to determine its electric potential field and capacitance. an electrostatic node is added to the model to solve electrical field and capacitance physics. fig. 4 shows a plot of the particulate mass and derived mass values from 0 to 15 ms. the highest gas velocity reaches 48.5 m/s between 0 and 15 ms, propagating from the sensor surface. the fluid pressure gradually increases to 1.02 × 10-05 at 6.5 ms and remains constant until 15 ms. the correlation between the deposited mass and sensor capacitance results in a total deposition of 0.033 mg over 60 seconds. a proceedings of engineering and technology innovation, vol. 24, 2023, pp. 53-62 58 maximum mass of 24.8 mg/s can be calculated based on the mass volume fraction of the 8.0 × 10-06 sensor location, with gases flowing at 24.8 mg/s to the sensor surface and exiting at the outlet into the atmosphere. the finite element analysis reveals a particulate deposition rate of 0.0005 g/s on the sensor surface between 0 and 15 ms, which remains constant for one second. thus, it is possible to estimate the deposition rate over an extended period. the analysis also proves that the deposited mass on the sensor face (0 to 15 ms) is proportional to the gas flow. fig. 5 shows that particulate mass accumulation is crucial to the outcome. designed experiments to broaden the scope of work for flow rate/velocity and particulate concentration. the flow velocity, v1, ranges from 35 m/s to 41 m/s, and the particulate concentration from 6.4 × 10-05 to 1.0 × 10-06. a doe study indicates that the total mass deposited ranges from 0.000069 mg at v1 = 41 m/s and c = 1 × 10-06 to 0.0044 mg at v1 = 35 m/s and c = 6.4 × 10-05 during the 15 ms period. as noted in the sensor’s response time and sensitivity, the sensor shows a fast response time of 4.8 milliseconds to low-mass depositions. 5. modeling of incremental particulate layer capacitance the incremental particulate layer sensor models the electric potential distribution and the resulting capacitance. fig. 6 presents the capacitance values as extracted from comsol contour plots. the magnitudes of electric potential at various layer thicknesses (0.1 µ m to 2 µ m in 0.5 µ m steps) in the sensor domain are extracted and plotted in fig. 7. as the results show, the capacitance of the sensor increases from 566 pf to 568 pf, revealing a linear relationship between the detection time and the deposited particulates on the sensor surface. fig. 6 electrical potential (v) versus (tc-soot layer thickness) at 0.1 µ m and 2 µ m fig. 7 sensor capacitance (soot layer thickness tc = 0.1 µ m to 2 µ m) proceedings of engineering and technology innovation, vol. 24, 2023, pp. 53-62 59 an electrical capacitance in pf is measured through electrical simulations to determine the particulate deposition in mg. the sensor showed a linear response within the low particulate deposition ranges. fig. 8 illustrates the capacitance extracted for the corresponding thicknesses of the deposited particulate layer. accordingly, in fig. 9, the masses are plotted on the xaxis against the corresponding extracted capacitance on the y-axis, with the maximum capacitance of 581 pf corresponding to a mass of 0.05 mg. fig. 8 sensor capacitance for mass-deposited (mg) fig. 9 the mass plot (tc = 1 to 1000 µ m) after 0.05 mg fig. 10 the effect of increasing the thickness of the top insulation layer the analysis and simulation results indicate that the sensor operates linearly as the particulate layer thickness increases from 0.1 µ m to 2 µ m. fig. 8 illustrates the corresponding increase in capacitance. there is a micro-level increment in capacitance as the thickness of the deposited mass increases, the formation of agglomerates being responsible for this micro proceedings of engineering and technology innovation, vol. 24, 2023, pp. 53-62 60 level increment. also, the mass plot (tc = 1 to 1000 µ m) exhibits saturated behavior after 0.05 mg, as shown in fig. 8. by using capacitance values derived from cfd simulations, the incremental layer thickness values are converted to particulate masses, measured in mg. modeling incremental particulate layer capacitance with an increased thickness of the top insulation layer. the sensor’s dimensions are optimized to reflect linear capacitance variations with a top insulation thickness of 5 µ m. the sensor capacitance measurement has a mass detection limit of 0.05 mg. results show that the thickness of the insulation layer on the top of electrodes directly affects the sensor’s sensitivity. fig. 10 illustrates the effect of increasing the thickness of the insulation layer to 0.5 mm. as the thickness of the top insulation layer increases, the sensor's sensitivity decreases, resulting in a constant capacitance and no significant change in the sensor's output capacitance. 6. comparison of sensor performance with literature-based benchmarks hagen et al. [2] highlighted the experimental dimensions and developed a 2d cad model of the sensor with electrodes using comsol. the dimensions of the device measure equal electrode width and spacing of 100 µ m, an electrode thickness of 4 µ m, and an electrode length of 48 µ m. the reference device has a substrate thickness of 635 meters and an insulation layer of 8 meters, with 29 electrodes. the predicted capacitance values in the reference sensor agree with the experimental data, with an error margin of ±5%. however, the measured capacitance values (ranging from a minimum of 8.40 pf to a maximum of 50.49 pf) are too small to be used for mass measurement. as a result, it is necessary to estimate the maximum electrostatic field by optimizing the design dimensions. a subsequent study provided insights that enabled the optimization of the sensor’s design. this optimization enhanced the sensor’s selectivity and sensitivity for detecting and measuring particulates, as presented in this paper. a 2d model of the sensor was developed in comsol, which reduced both computational time and costs. the software module developed for the sensor analysis can serve as a proven design tool for screening various idc layouts. furthermore, this simulation module could be applied to estimate the expected sensor’s sensitivity. hagen et al. [2] also discussed how the findings of an experiment with a sensor that could not measure mass contributed to the development of an improved sensor. the optimized sensor proves its superiority to the experimental one in [2], as it can measure particulate deposition at a lower range and with higher sensitivity. 7. conclusions this study presented an electrostatic simulation of an optimized capacitive sensor design that can detect nanometer-sized carbon particulates at the lowest possible level. cfd simulations were conducted to model the deposition of particulate mass on the sensor surface. the amount of particulate mass based on time and mixture flow was predicted, and the contour plots of electric potential distribution and sensor capacitance values for the optimized capacitive sensor design were illustrated. the predicted sensor capacitance increases with the rising of the particulate deposition. the study also compared the optimized sensor with the benchmark design [2] and illustrated the performance improvements achieved with the optimized design. given the forthcoming emission norms worldwide, it is imperative to address mass estimation promptly. a thresholdbased corrective action approach based on the results can significantly reduce environmental emissions. it would be helpful to assess the performance of the sensor before prototyping begins. the results indicate that particulate detection sensitivity increases with decreasing insulating layer thickness, and the optimized sensor produces ten times more output than the benchmark sensor. in addition, the presented sensor design incorporates interdigitated electrodes, making it convenient, durable, and reliable to operate at room temperature. proceedings of engineering and technology innovation, vol. 24, 2023, pp. 53-62 61 however, due to high levels of deposition, the sensor becomes saturated, requiring regeneration and calibration to accommodate dynamic conditions. accordingly, future work should focus on optimizing sensor location, designing sensor mounts that minimize back pressure, and optimizing electromagnetic parameters for optimum sensing performance. conflicts of interest the authors declare no conflict of interest. references [1] h. husted, g. roth, s. nelson, l. hocken, g. fulks, and d. racine, “sensing of particulate matter for on-board diagnosis of particulate filters,” sae international journal of engines, vol. 5, no. 2, pp. 235-247, may 2012. [2] g. hagen, m. feulner, r. werner, m. schubert, a. muller, g. riess, et al., “capacitive soot sensor for diesel exhausts,” sensors and actuators b: chemical, vol. 236, pp. 1020-1027, november 2016. [3] b. zhao, c. chen, x. yang, and a. c. k. lai, “comparison of three approaches to model particle penetration coefficient through a single straight crack in a building envelope,” aerosol science and technology, vol. 44, no. 6, pp. 405-416, april 2010. [4] p. sydenham and r. thorn, handbook of measuring system design, vol. 2, uk: john wiley & sons, 2005. [5] a. v. mamishev, k. sundara-rajan, f. yang, y. du, and m. zahn, “interdigital sensors and transducers,” proceedings of the ieee, vol. 92, no. 5, pp. 808-845, may 2004. [6] c. sapsanis, h. omran, v. chernikova, o. shekhah, y. belmabkhout, u. buttner, et al., “insights on capacitive interdigitated electrodes coated with mof thin films: humidity and vocs sensing as a case study,” sensors, vol. 15, no. 8, pp. 18153-18166, august 2015. [7] c. sapsanis, s. sivashankar, h. omran, u. buttner, and k. n. salama, “capacitive immunosensor for c-reactive protein quantification,” ieee 58th international midwest symposium on circuits and systems (mwscas), august 2015. [8] h. omran, m. arsalan, and k. n. salama, “a robust parasitic-insensitive successive approximation capacitance-todigital converter,” proceedings of the ieee 2014 custom integrated circuits conference, september 2014. [9] h. omran, m. arsalan, and k. n. salama, “7.9 pj/step energy-efficient multi-slope 13-bit capacitance-to-digital converter,” ieee transactions on circuits and systems ii: express briefs, vol. 61, no. 8, pp. 589-593, august 2014. [10] j. w. gardner, p. k. guha, f. udrea, and j. a. covington, “cmos interfacing for integrated gas sensors: a review,” ieee sensors journal, vol. 10, no. 12, pp. 1833-1848, december 2010. [11] p. w. oluwasanya, g. rughoobur, and l. g. occhipinti, “design, modeling and simulation of a capacitive sizediscriminating particulate matter sensor for personal air quality monitoring,” ieee sensors journal, vol. 20, no. 4, pp. 1971-1979, february 2020. [12] r. s. jachowicz and s. d. senturia, “a thin-film capacitance humidity sensor,” sensors and actuators, vol. 2, pp. 171-186, 1981-1982. [13] x. b. li, s. d. larson, a. s. zyuzin, and a. v. mamishev, “design principles for multichannel fringing electric field sensors,” ieee sensors journal, vol. 6, no. 2, pp. 434-440, april 2006. [14] x. hu and w. yang, “planar capacitive sensors – designs and applications,” sensor review, vol. 30, no. 1, pp. 24-39, 2010. [15] a. kothuru and s. goel, “electronic nasal pod: a 3d printed device to filter and electrochemically detect pollutants,” ieee sensors, october 2020. [16] s. zargari, s. falaki, and h. veladi, “design and finite element analysis of a mems based capacitive pressure sensor using cnt/pdms nanocomposite electrodes,” 24th iranian conference on electrical engineering, may 2016. [17] y. huang, z. zhan, and n. bowler, “optimization of the coplanar interdigital capacitive sensor,” aip conference proceedings, vol. 1806, no. 1, article no. 110017, february 2017. [18] r. m. d. santos, j. m. sallese, m. mattavelli, a. s. nunes, c. dehollain, and d. barrettino, “high precision capacitive moisture sensor for polymers: modeling and experiments,” ieee sensors journal, vol. 20, no. 6, pp. 3032-3039, march 2020. [19] p. wang, q. lu, and z. fan, “evolutionary design optimization of mems: a review of its history and state-of-theart,” cluster computing, vol. 22, no. 4, pp. 9105-9111, july 2019. proceedings of engineering and technology innovation, vol. 24, 2023, pp. 53-62 62 [20] p. w. oluwasanya, g. rughoobur, and l. g. occhipinti, “comparison of analytical and numerical methods of obtaining coplanar capacitance of microelectrodes for particulate matter detection,” ieee sensors journal, vol. 20, no. 15, pp. 8631-8636, august 2020. [21] d. back, d. theisen, w. seo, c. s. j. tsai, and d. b. janes, “development of interdigitated capacitive sensor for realtime monitoring of sub-micron and nanoscale particulate matters in personal sampling device for mining environment,” ieee sensors journal, vol. 20, no. 19, pp. 11158-11597, october 2020. [22] f. y. niyat and m. h. s. abadi, “comsol-based modeling and simulation of sno2/rgo gas sensor for detection of no2,” scientific reports, vol. 8, article no. 2149, february 2018. [23] c. h. weng, g. pillai, and s. s. li, “a pm2.5 sensor module based on a tpos mems oscillator and an aerosol impactor,” ieee sensors journal, vol. 20, no. 24, pp. 14722-14731, december 2020. [24] f. foncellino, l. barretta, e. massera, and a. corigliano, “piezoelectric mems for microparticles detection,” ieee sensors, october-november 2021. [25] p. das, s. ghosh, s. chatterjee, and s. de, “a low cost outdoor air pollution monitoring device with power controlled built-in pm sensor,” ieee sensors journal, vol. 22, no. 13, pp. 13682-13695, july 2022. [26] h. cho and y. baek, “practical particulate matter sensing and accurate calibration system using low-cost commercial sensors,” sensors, vol. 21, no. 18, article no. 6162, september 2021. [27] n. singh, m. y. elsayed, and m. n. el-gamal, “towards the world’s smallest gravimetric particulate matter sensor: a miniaturized virtual impactor with a folded design,” sensors, vol. 22, no. 5, article no. 1727, march 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 (http://creativecommons.org/licenses/by/4.0/). nst 2007 extended summary proceedings of engineering and technology innovation, vol. 5, 2017, pp. 07 12 the design and implementation of scrambling wavelength hopping with awg router and optical switch over secure wdm network yao-tang chang 1,* , hsu-chih cheng 2 , yih-chuan lin 3 , yu-xiang zheng 2 1 department of information technology kao yuan university kaohsiung, taiwan . 2 department of electro-optical engineering, national formosa university, yunlin, taiwan . 3 department of computer science and information engineering, national formosa university, yunlin, taiwan. received 26 july 2017; received in revised form 13 august 2017; accept ed 17 august 2017 abstract in current e xperimental design, by e xchanging the optical switch configured in front of arrayed waveguide grating (awg) router, the wave length hopping configuration was imple mented over wavelength division mu ltip le xing (w dm) network. in addit ion, in order to enhance the variab le wavelength hopping pattern of proposed scheme, the transceiver oscillators were used to trigger the optical switch by a series of random and un predictable electrical signal. the e xpe riment results proved possible solution of secure wdm network that the wavelength hopping effect was monitored by optical spectrum analyze r (osa) in optical do main. by u sing the oscilloscope for monitoring, the results showed the transmitted analogy signal of 10m hzwas e xtracted correct ly by photo-detector while the wavelength hopping is happened. ke ywor ds: a rrayed waveguide grating, wave length hopping, wavelength division mu ltiple xing, optical switch, photo-detector 1. introduction improving the confidentiality of access networks has emerged as a crit ical p roble m in recent years. it has been shown that optical code-division multip le-access (ocdma) schemes significantly improve the robustness of optical networks toward ma lic ious attack [1-8]. gene rally speaking in confidentia l topic, the wdm network was vie wed as without capability against eavesdroppers’ attacking. in order to achieve the confidential ability in wavelengt h division multip le xing (wdm) network, the a rrayed waveguide grating -based (awg-based) wavelength hopping was presented by one of current authors [9-10]. unfortunately, the proposed scheme had never been imp le mented in e xpe riment stage. here, the possible and feasible configuration was demonstrated in wired w dm network to achieve similarly security mechanism of frequency hopping in wireless communications such as those executed over wifi network. 2. system architecture and principles in current e xperimental design, the proposed of wavelength hopping configuration for scra mbling modulated signal was shown in fig. 1. for co mmerc ia l components in practical e xperiment, the 8×8 aw g router and 2× 2 optical switch were e xploited. actually, the only one input port of optical switch was applied and then below-mentioned optical switch could be called 1×2 optical switch. the sld light source was applied to prov ide the three wave lengths of λ1=1556.3n m, λ2=1557.1n m, and λ8=1562.1n m, respectively. connecting after the sld light source, the 1×2 optica l switch which triggered by transmitted oscillator was * corresponding author. e-mail address: t10066@cc.kyu.edu.tw tel.: +886-7-6077138; fax: +886-7-6077959 proceedings of engineering and technology innovation, vol. 5, 2017, pp. 07 12 copyright © taeti 8 used to exchange the optical path to enter arbitrary input port of aw g router (e.g., port #1 – #8 as seen in fig. 1) wh ile cros s (on) and bar (off) state was happened. by emp loying inherent cyclic and fsr property of awg router, the awg router enables realizing n×n routers, in which n input wavelengths (#1 – #n) were de-multiplexed to n output ports. fig. 1 the proposed wavelength hopping configuration using awg router and optical switch on the transmitter here, the different wavelengths appear in the output port of awg router as user’s carrier and transfer to electro -optics modulator (eom). the individual wave length carrier was modulated by the analog signal of 10m hz for each user and coupled into fiber to received end. finally, the wavelength hopping phenomenon of modulated signal was monitored by optical spectrum analy zer (osa) and achieved the varying analog signal of 10mhz co ming fro m the diffe rent fg# 1(user #1) and fg#2(user#2). following the cyclic property of awg, the path of laser source was changed by optical switch fro m bar (off) to cross (on) state and entered into the different input port of aw g. hence, the various wavelengths were appeared in origina l output port of awg to create the wavelength changing. for changing the wavelength carrier o f each user, each user corresponded to perform wave length hopping and then protect ed against the un-authorized attacking. in addit ion, in order to enhance the variable wavelength hopping pattern of proposed scheme, the optical switch could be triggered by transceiver oscillators creating from a series of random and aperiodic electrical signal. in current practical e xperiment, the wavelength hopping of fg#1(user#1) was created fro m λ1=1556.3n m to λ8=1562.1n m while the optical switch state was converted by transmitter oscillator fro m bar to cross state. simu ltaneously, the wavelength hopping of fg # 2(user #2) was created fro m λ2=1557.1n m to λ1=1556.3n m wh ile the optical switch state was converted by transmitter oscillator from bar to cross state. fig. 2 the proposed wavelength hopping configuration using awg router and optical switch and on the receiver proceedings of engineering and technology innovation, vol. 5, 2017, pp. 07 12 copyright © taeti 9 on the receiver shown in fig. 2, the 1× 2 optical switch was trigged by received oscillator which had the same frequency as transmitter and decided the state of switch off (bar) or switch on (c ross), depending on transmitted end. finally, the diffe rent optical signals appearing in the output port of awg router were detected through photo-detector to retrieve the diffe rent analog signal fro m fg#1(user #1) and fg# 2(user#2). the oscilloscope (osc) connected the photo-detector to observe the optical signal modulated by analog signal of 10mhz. 3. experimental result in current e xperiment, all the type and specification of using components is summarized in table 1. in order to guarantee the imp le mentation of wavelength hopping correctly, the monitoring architecture configured with optical spectrum analyzer (osa) and os cilloscope (osc) was shown as fig . 3 on the transmitter. he re, the osa was used to measure the wavelength hopping happening when the optical switch was triggered by specific sequence (i.e., hopping pattern). simu ltaneously, the osc was used to measure the analog signal of 10mhz when the a mplitude of desired signal was retrieved by photo-detector from the specific wavelength. fig. 3 the monitoring arch itecture configured with optical spectrum analy zer (osa) and oscilloscope (osc) on the transmitter there were two states of optical switch to be imple mented in fig. 1. the bar (off) and cross (on) of 1× 2 optical switch were designed to verify the wavelengths hopping phenomenon, respectively. table 1 all the type and specification of using components in current experiment production specification transmitter laser source nxtar sld-2200 optical switch new port spsn-9-22-fcapc awg dwdm-g-100g-8-8-fc-apc analog signal(10mhz/1mhz) twintex tfg-3510 eom lucent zx86-12g-s+ coupler cp-s-a-2x2-1550-50/ 50-900-1-1-3x54mm-fc/ upc fiber fc/upc-fc/apc-sm-sx-0.9mm-3m receiver optical switch thorlabs gmbh osw22-1310e awg dwdm-g-100g-8-8-fc-apc photo-detector new focus model 1611 measuring equipment optical spectrum analyzer (osa) anritsu ms9710c oscilloscope (osc) tektronix tds2012b proceedings of engineering and technology innovation, vol. 5, 2017, pp. 07 12 copyright © taeti 10 fig. 4(a) and fig. 5(a) presented the wavelength spectra on the switch bar (off) state to focus only on the output port #1 and #2 of aw g router. the λ1=1556.3n m, λ2=1557.1n m and λ8=1562.1n m entered into the input port #1 of aw g router and appeared a pair of (output port #1, output port #2) wavelengths corresponding to (λ1=1556.3n m, λ2=1557.1n m). actually, the wavelength of λ8=1562.1nm appear in the input port #8 of awg router. (a) λ1=1556.3n m on the switch bar (off) state (b) λ8=1562.1nm on the switch cross (on) state fig. 4 appearing wavelength in output port #1 of awg router similarly, as seen from fig. 4(b) and fig. 5(b ), converting to the switch cross (on) state, the same wavelength of sld changed the entrance into the input port #2 of awg router and appear a pa ir of (output port #1, output port #2) wavelengths corresponding to (λ8=1562.1n m, λ1=1556.3n m). actually, the wave length of λ2=1557.1n m appear in the input port #2 of awg router. by mon itoring the switch changing fro m bar (off) to cross (on) state, the wavelength hopping of fg# 1(user#1) was imple mented fro m λ1=1556.3n m to λ 8=1562.1n m. similarly, the wavelength hopping of fg# 2(user#2) was imp le mented from λ2=1557.1n m to λ1=1556.3nm. (a) λ2=1557.1n m on the switch bar (off) state (b) λ1=1556.3nm on the switch cross (on) state fig. 5 appearing wavelength in output port #2 of awg router by measuring the appearing output port #1 of awg router modulated by eom #1, figure 6 presented the different wavelengths as carriers which were modulated by the analog signal of 10mhz on the switch bar (o ff) state or cross (on) state. the oscilloscope (osc) showed that the individual optical signal could be modulated by the analog signal through electrooptics modulator (eom) to make the wavelength hopping possible. similarly, by measuring the appearing output port #2 of awg router modulated by eom #2 , the figure 7 presented the different wavelengths as carriers which were modulated by the analog signal of 10mhz. proceedings of engineering and technology innovation, vol. 5, 2017, pp. 07 12 copyright © taeti 11 (a). λ1=1556.3n m on the switch bar (off) state (b) λ8=1562.1nm on the switch cross (on) state fig. 6 the modulated analog signal of 10mhz appearing in output port #1 of awg router (a) λ2=1557.1n m on the switch bar (off) state (b) λ1=1556.3nmon the switch cross (on) state fig. 7 the modulated analog signal of 10mhz appearing in output port #2 of awg router 4. conclusion in current study, the possible configuration in the wavelength hopping was experimented to achieve the secure modulated analog signal of 10mhz by electrical/optical modulator. the e xperiment results proved possible solution of secure wdm network that the wavelength hopping effect was monitored by optical spectrum analy zer (osa) in optical domain. in addition, by using the oscilloscope (osc) for monitoring the original signal, the results showed the transmitted analog signal of 10mhz was e xtracted successfully by photo -detector while the wavelength hopping was happened. in near future works, by replac ing 10m hz analo g signal with video strea m, the desired converting interface of video and binary data bit (i.e ., e ither bit 0 or 1) will be installed. the e xperiment will be investigated to extract and recovery the video signal correctly. furthermore , a secure wavelength hopping scheme will be imp le mented and verified to apply in mult imedia application. acknowledgment this study was supported under grant nos. most 105-2221-e-244-004 and most 104-2221-e-150-043. references [1] t. h. shake, “security performance of optical cdma against eavesdropping,” journal of lightwave technology, vol. 23, no. 2, pp. 655-670, february 2005. [2] t. h. shake, “confidentiality performance of spectral-phase-encoded optical cdma,” journal of lightwave technology, vol. 23, no. 4, pp. 1652-1663, april 2005. [3] s. p. tseng and j. wu, “the sac ocdma pon with super perfect difference code,” proc. ieee international conference on communications , june 2011, pp. 1-4. [4] l. tancevski, i. andonovic, and j. budin, “secure optical network architectures utilizing wavelength hopping/time spreading codes,” ieee photonics technology letters, vol. 7, no. 5, pp. 573-575, may 1995. proceedings of engineering and technology innovation, vol. 5, 2017, pp. 07 12 copyright © taeti 12 [5] y. t. chang, c. c. sue, and j. f. huang, “robust design for reconfigurable coder/decoders to protect against eavesdropping in spectral amplitude coding optical cdma networks,” journal of lightwave technology, vol. 25, no. 8, pp. 1931-1948, august 2007. [6] y. t. chang and y. c. lin, “dynamic reconfigurable encryption and decryption with chaos/m-s equence mapping algorithm for secure h.264/avc video streaming over ocdma passive optical network,” multimedia tools and application, vol. 75, no. 16, pp. 9837-9859, august 2016. [7] y. t. chang and c. w. tsailin, “dynamic scrambling scheme of arrayed -waveguide grating-based encryptors and decryptors for protection against eavesdropping,” computers and electrical engineering, vol. 49, pp. 184-197, january 2016. [8] c. t. yen and c. y. liu, “hybrid ofdm/ook modulations in ocdma scheme for free space optics,” proc. engineering and technology innovation, vol. 3, 2016, pp. 1-3. [9] y. t. chang and y. c. lin, “the scrambling cryptography implemented with chaotic sequence trigger optical switch algorithm in wdm passive optical network,” proc. of the 49th annual ieee international carnahan conference on security technology, september 2015, pp. 259-262. [10] y. t. chang, j. f. huang, y. c. huang , and y. t. liou, “security enhancement on reconfiguring coded wavelength with tunable wavelength filter array triggered chaotic seque nces,” proc. engineering and technology inno vation, vol. 3, 2016, pp. 4-6.  proceedings of engineering and technology innovation, vol. 14, 2020, pp. 16 21 stretchable, patch-type, wireless, 6-axis inertial measurement unit for mobile health monitoring jae keun lee, kangil kim, sangmin lee * department of biomedical engineering, kyung hee university, yongin, korea received 16 may 2019; received in revised form 08 july 2019; accepted 09 september 2019 doi: https://doi.org/10.46604/peti.2020.4272 abstract wearable devices which measure and transfer signals from the human body can provide useful biometric data for various biomedical applications. in this paper, we present an implementation of the advanced inertial measurement unit (imu) with wireless communication technology for mobile health monitoring. the device consists of rigid silicon-based components on a flexible/stretchable substrate for applications in epidermal electronic devices to collect precise data from the human body. using the bluetooth low energy (ble) system-on-a-chip (soc), the device can be miniaturized and portable, and the collected data can be processed with low power consumption. the dimensions of the implemented system are approximately 40 mm × 40 mm × 100 m. also, the device can be attached closely to human skin, which results in minimized signal distortion due to body movements or skin deformations. in order to achieve device flexibility and stretch ability, the interconnection wires are designed as serpentine-shaped structures on a stretchable substrate. the previously reported “cut-and-paste” method is utilized to fabricate the device that produces complex, twisty interconnections with thin metal sheets. the implemented patch-type, wireless, 6-axis imu is expected to have potential in various applications, such as health monitoring, dependency care, and daily lifelogging. keywords: bluetooth low energy (ble), inertial measurement unit (imu), epidermal electronics, serpentine-shaped interconnection 1. introduction recently reported researches have shown that the highly-reliable wearable devices which provide continuous physiological biometric information can be used for applications in mobile health [1-3], physical training [4], prosthetics [5] and human-machine interfaces (hmi) [6-8]. especially, skin-like wearable electronic modules have soft form-factors to achieve close contact on the human body, which results in the significantly improved performance, such as the signal-to-noise ratio [9]. from the inherent skin-like characteristics of flexible and stretchable substrates, recent developments focus on the implementation of highly-scalable and modularized epidermal electronic devices for biomedical applications. most commercialized wearable devices are implemented on a rigid substrate consist of silicon-based integrated circuit (ic) components to provide incomparable performance for signal processing with low-power consumption [10-11]. however, there is a large mechanical mismatch in the interface between the rigid circuit-board based devices and elastic human skin, which may reduce the accuracy of the measurement data. thus, accurate biometric information can be measured using a hybrid integration scheme, where stretchable interconnection wires are used for elastic bio-interfaces and rigid silicon-based electronic ics are used for calculations, signal processing and data transmission. in order to improve the fidelity between the system and the skin, the implementation of the robust interconnection part is a fundamental technology to develop the system-level based epidermal electronic devices. several groups have reported the * corresponding author. e-mail address: sangmlee@khu.ac.kr proceedings of engineering and technology innovation, vol. 14, 2020, pp. 16 21 17 use of semiconductor microfabrication processes for implementing the serpentine-structured metal interconnects. this is achieved by fabricating a metal interconnection wire with a serpentine structure on the substrate using a semiconductor process, and then attaching the rigid electronic sensor with the ics. due to the serpentine-structured metal interconnects, the stress applied on the wires during elongation can be minimized. although, microfabrication facilities for semiconductors are well-known to achieve precise micro patterns and minimized footprint for highly-miniaturized system, it is highly-expensive and time-consuming processes. in this paper, an epidermal, patch-type 6-axis inertial measurement unit (imu) is presented for biomedical applications, such as gait analysis, musculoskeletal disorders, and monitoring daily routines [12]. the proposed system is implemented using the packaged silicon-based ics soldered on the bonding pads, which the metal interconnect between the ics have serpentine-structures. also, the whole device is implemented on a soft and stretchable platform to be attached on any part of human skin with high conformity. each functional block of the system is fully-modular and reconfigurable, and this system is configured by a commercialized 6-axis accelerometer, a bluetooth low energy (ble) system-on-a-chip (soc), several passive components and a chip antenna to measure the continuous biometric information with ease. for implementation of the proposed device, a simple, cleanroom-free manufacturing method is used. the detailed design schematic, fabrication process and performance evaluations will be discussed in the following sections. as a result, the system and the proposed method in this paper can be extended to the development of an advanced, multi-sensing, biometric platforms. 2. implementation process the implemented epidermal, patch-type 6-axis imu on a soft substrate is shown in fig. 1. sensors and devices used to construct the proposed system include 6-axis inertial sensor (mpu6050, invernesses, san jose, ca, usa), ble socs (nrf52832, nordic semiconductor asa, norway), and several passive components. the serpentine-structured copper (cu) wires are used to mount and integrate rigid electronic elements, which allows it to have elongation properties of the system under diverse deformation. the whole system is encapsulated using the stretchable adhesive film (tegaderm, 3m, mn, usa), which can be attached to the human body and protects the system from moisture and dust. the system is implemented based on the modular concept, allowing the functional building blocks to be reconfigurable depending on the applications. by maintaining the signal processing parts, additional functions can be integrated by adding multiple sensors, such as electrodes, strain gauges or temperature sensors. fig. 1 implemented epidermal, 6-axis imu on stretchable platform in fig. 2, the block diagram of the epidermal, path-type 6-axis imu is shown. the 6-axis inertial sensor and mcu transmit and receive data using the two-wire serial interface (twi), which uses two bus signals (scl and sda) to exchange signals. the signals from the 6-axis inertial sensor are converted by the analog-to-digital converter (adc), which is embedded in the ble soc, and the digitized data are spread by the universal synchronous/asynchronous receiver/transmitter (usart) structure in the signal conditioning unit. 6-axis sensor ble soc bluetooth antenna power lines serpentine interconnection proceedings of engineering and technology innovation, vol. 14, 2020, pp. 16 21 18 fig. 2 block diagram of the epidermal, 6-axis imu using wireless data communication (a) lamination (b) patterning (c) soldering (d) detachment and encapsulation fig. 3 implementation process using “cut-and-paste” method the implementation process to fabricate the patch-type 6-axis imu on a flexible/stretchable polymeric surface is shown in fig. 3. this process is used to design miniaturized systems on substrates with various form-factors using computer-aided design (cad) program. to implement the proposed system, serpentine-structured interconnects and soldering pads for silicon ic assembly are designed. after designing the footprints of the system using cad tool, a programmable cutter using sharp metal blade (silhouette cameo® , silhouette america, usa) is equipped to define the interconnects and soldering pads. an 18-μm-thick cu foil (copper 110 annealed, online metals, usa) is laminated on the thermal release tape (trt), and the cu foil-laminated trt is patterned according to the designed patterns programmed by the cad tool. the blade depth of the cutting equipment is controlled only to cut the cu foil layer. after cutting the cu foil-laminated trt, whole residual area except for designed patterns of cu foil is removed. the patterned cu foil laminated trt layer is adhered with a kapton film, water-soluble tape (wst) layer, which is bonded onto a glass substrate. and then, the whole bonded substrate is heated on a hot plate at 120 °c for 30 seconds to remove the trt layer. after removing the trt layer, a lead-free solder paste (chipquik, ny, usa) is used to assemble packaged ics on the patterned cu soldering pads. the electronic components, such as 6-axis inertial sensor, ble soc, chip antenna, passive components, and bridge interconnects are aligned on the soldering pads and heated on the hot plate at 230 °c for 3 minutes for fixation. then, the soft adhesive film is covered on the circuit for encapsulation. the wst layer and the kapton tape are removed to leave only electronic components and the trt cu foil trt cu foil cu foil laminated on trt cu foil patterning trt cu foil wst trt cu foil wst kapton tape remove cu foil residuals & wst lamination kapton tape lamination cu foil wst kapton tape cu foil wst kapton tape tegaderm remove trt & chip soldering laminate stretchable layer cu foil tegaderm tegaderm wst/kapton tape removal & encapsulation proceedings of engineering and technology innovation, vol. 14, 2020, pp. 16 21 19 serpentine-structured cu interconnects. after covering the bottom layer of the device, the implementation of the patch-type 6-axis imu on a stretchable platform is finalized. the overall dimensions of the proposed system are approximately 40 mm × 40 mm, and the thickness is less than 100 m (excluding the ics). 3. performance evaluation the implementation results of the epidermal, 6-axis imu are shown in figs. (4)-(7), respectively. the stress applied to the interconnection wire during elongation can be minimized by its serpentine structure. therefore, the modulus of the implemented device depends on the stiffness of the tegaderm (approximately 7.4 mpa), which is similar to that of human skin (0.32~4 mpa). in fig. 4, the connection to the wireless terminal using bluetooth is shown. after the connection, short impact and tilting motions in a single direction are applied and measured wirelessly. (a) bluetooth connection (b) with short impact (c) tilting in x-axis (d) tilting in y-axis (e) continuous no monitoring fig. 4 wireless connection and continuous signal monitoring results fig. 5 performance evaluation of the wireless, epidermal, 6-axis imu for rotational motion also, experiments are conducted with applied rotational motion (fig. 5) and arbitrary motion (fig. 6) to the patch-type, 6-axis imu attached to a human arm. in fig. 5, the performance comparison results between the implemented patch-type device and wear notch (notch interfaces, inc., ny, usa) are shown. the maximum root-mean-square (rms) error is about several tens of degrees. however, this appears to be an error due to the alignment mismatch and will be corrected through future calibration. the results show that real-time, continuous biometric motion data of the human body can be successfully 90° 1 2 x z y r = 1.00 rmse=9.4° r = 0.99 rmse=14.4 ° r = 0.03 rmse=17.6 ° proceedings of engineering and technology innovation, vol. 14, 2020, pp. 16 21 20 monitored. in fig. 7, the implemented patch-type, 6-axis imu is attached to the arm to calculate the precise joint angle of the forearm. the flexibility and stretch ability of the implemented device allow it be utilized on the human body including the wrists, the ankles, the neck, and the abdominal region, which are common parts for monitoring the motion for dependency care and clinical diagnosis. (a) device attached on forearm (b) lateral translation (c) rotation fig. 6 application result in 3-dimensional simulator control (a) device attachment (b) anterior view (c) lateral view fig. 7 joint angle analysis using the wireless, epidermal, 6-axis imu 4. conclusions this paper presents an epidermal, patch-type 6-axis inertial measurement unit (imu) for applications in wearable biomedical devices. the proposed device is implemented using the conventional circuit components on a soft adhesive substrate, and by firmly attached to the target, it is possible to collect accurate data from the human body. the planar dimensions of the implemented system are approximately 40 mm × 40 mm, and the overall thickness of the system, including the ics, is less than 100 m. a serpentine-structured interconnects are adopted between the circuit elements to achieve flexible and stretchable characteristics of the device. the preliminary implementation and experimental results of this paper demonstrated the potential for future applicability in various areas, such as dependency care and monitoring of human motion. conflicts of interest the authors declare no conflict of interest. acknowledgments this work was supported by the national research foundation of korea (nrf) grant funded by the korea government (msip; ministry of science, ict & future planning (no. nrf-2017r1c1b5017561). this work was also supported by the proceedings of engineering and technology innovation, vol. 14, 2020, pp. 16 21 21 national research foundation of korea (nrf) grant funded by the ministry of science, ict & future planning for convergent research in human plus (no, nrf-2019078037). references [1] d. kim, n. lu, r. ma, y. kim, r. kim, s. wang, et al., “epidermal electronics,” science, vol. 333, no. 6044, pp. 838-843, august 2011. [2] h. lee, t. k. choi, y. b. lee, h. r. cho, r. ghaffari, l. wang, et al., “a graphene-based electrochemical device with thermoresponsive microneedles for diabetes monitoring and therapy,” nature nanotechnology, vol. 11, no. 6, pp. 566-572, 2016. [3] a. miyamoto, s. lee, n. f. cooray, s. lee, m. mori, n. matsuhisa, et al., “inflammation-free, gas-permeable, lightweight, stretchable on-skin electronics with nanomeshes,” nature nanotechnology, vol. 12, no. 9, pp. 907-913, 2017. [4] p. salvo, a. pingitore, a. barbini, and f. di francesco, “a wearable sweat rate sensor to monitor the athletes’ performance during training,” science & sports, vol. 33, no. 2, pp. e51-e58, october 2017. [5] w. gao, h. ota, d. kiriya, k. takei, and a. javey, “flexible electronics toward wearable sensing,” accounts of chemical research, vol. 52, no. 3, pp. 523-533, 2019. [6] y. liu, j. j. s. norton, r. qazi, z. zou, k. r. ammann, h. liu, et al., “epidermal mechano-acoustic sensing electronics for cardiovascular diagnostics and human-machine interfaces,” science advances, vol. 2, no. 11, pp. e1601185, november 2016. [7] s. huang, y. liu, y. zhao, z. ren, and c. f. guo, “flexible electronics: stretchable electrodes and their future,” advanced functional materials, vol. 29, no. 6, pp. 1805924, 2019. [8] s. lim, d. son, j. kim, y. b. lee, j. song, s. choi, et al., “wearable electronics: transparent and stretchable interactive human machine interface based on patterned graphene heterostructures,” advanced functional materials, vol. 25, no. 3, pp. 375-383, november 2014. [9] m. kaltenbrunner, t. sekitani, j. reeder, t. yokota, k. kuribara, t. tokuhara, et al., “an ultra-lightweight design for imperceptible plastic electronics,” nature, vol. 499, no. 7459, pp. 458-463, july 2013. [10] q. cao, h. s. kim, n. pimparkar, j. p. kulkarni, c. wang, m. shim, et al., “medium-scale carbon nanotube thin-film integrated circuits on flexible plastic substrates,” nature, vol. 454, no. 7203, pp. 495-500, july 2008. [11] b. crone, a. dodabalapur, y. y. lin, r. filas, z. bao, a. laduca, et al., “large-scale complementary integrated circuits based on organic transistors,” nature, vol. 403, no. 6769, pp. 521-523, 2000. [12] w. teufl, m. lorenz, m. miezal, b. taetz, m. frohlich, and g. bleser, “towards inertial sensor based mobile gait analysis: event-detection and spatio-temporal parameters,” sensors, vol. 19, no. 1, pp. 38, 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/).  proceedings of engineering and technology innovation, vol. 11, 2019, pp. 21-29 research on traffic signal timing method based on ant colony algorithm and fuzzy control theory linlin li 1 , zhiyi fang 1,* , yunhan ma 2 , baiqi wang 1 , hongliang dong 3 , zhanyang zhang 4 1 college of computer science and technology, jilin university, china 2 department of statistics, college of mathematics, jilin university, china 3 banine technologies ltd. data & info research lab, china 4 computer science department, graduate center, city university of new york, usa received 03 may 2018; received in revised form 27 october 2018; accepted 20 december 2018 abstract the number of private cars has a blowout growth with the development of economics, which leads to the existing limited traffic resources cannot meet the normal traffic demand. the emergence of intelligent traffic has improved this phenomenon. using intelligent traffic technology to conduct intersection vehicles can alleviate the congestion effectively. traffic signal timing method plays an important role in intelligent traffic research. an independent intersection dynamic timing method combined with fuzzy control theory and improved ant colony algorithm is proposed in this paper. according to the characteristics of traffic flow distribution, the timing period is obtained with the improved webster algorithm. through the optimal solution obtained by ant colony algorithm and the added delay of traffic signal calculated by fuzzy control method, the dynamic timing period of the traffic signal is obtained. the validity of the proposed method is proved by comparing with the original time period and the traditional algorithm. keywords: intelligent traffic, dynamic timing, fuzzy control, ant colony 1. introduction in recent years, road congestion results in an increasing number of traffic accidents and serious traffic problems. many metropolises, at the beginning of the design, did not take the rapid growth of vehicles into account, so the urban road design could only cope with the number of vehicles at that time. however, the process of urbanization is swift and the number of vehicles is growing at a geometric level, the scale of roads has not kept pace with the growth in the number of vehicles. traffic congestion will lead to chaos in the traffic order, which wastes energy and causes economic losses. urban emerging vehicles such as metro can shorten the time of travel, but fail to solve the traffic congestion fundamentally. mitigation of urban road congestion can be considered from three aspects: widening the road, reducing the intersection and optimizing the traffic signal control system at the intersection. the investment in road broaden is huge and the effect is not obvious in a short time. the establishment of the viaduct can reduce road intersection to achieve effective shunt, but the cost is high. therefore, optimizing the traffic signal control system is the most effective approach. real-time traffic flow can be obtained through monitoring facilities. adjusting the traffic signal period, according to the traffic flow to reduce the stopping time of road intersection and alleviate the traffic congestion effectively. at present, more and more attention has been received to the traffic signal timing method. * corresponding author. e-mail address: fangzy@jlu.edu.cn tel.: +86 138-4311-8180 proceedings of engineering and technology innovation, vol. 11, 2019, pp. 21-29 22 optimizing the traffic signal control system is an efficient way to relieve traffic congestion. however, how to optimize efficiently is the key issue we need to study. the past crossroads are exactly the same for the opposite red and green time. the current traffic structure has changed a lot, the viaduct, fork road, and traffic flow is gradually increasing. the time of the traffic signal on opposite direction is different now, even the traffic signal period of two adjacent intersections is different, which stems from the measures taken to alleviate the traffic congestion by analyzing the traffic flow and calculating the traffic signal period. at present, numerous cities optimize the traffic signal timing method in various degrees. through the acquisition of mass data and optimization algorithm to enhance the coordination ability of traffic lights on the arterial road. the application of the intelligent signal timing system is mainly considered the fixed length of traffic signal, and it is still not good enough to match the space-time resources with the traffic demand. the purpose of this paper is to use automatic control technology, information technology and computer technology to dynamically adjust the traffic signals at intersections and efficiently alleviate traffic congestion, which has a wide range of practical and broad prospects for development. 2. signal control principles 2.1. plane intersection this paper studies the plane intersection, which means all roads intersect in the same plane, but do not involve multiple paths intersecting in different planes. common plane intersection, including crossroads, t-junctions, y-junctions and roundabout. the different directions pointed in the intersection lead to the intersection as a traffic accident prone point, which is called the conflict point. in order to reduce the conflict points, the intersection of multiple junctions is often set up as a roundabout. with fewer vehicles, there are no conflict points at the earliest roundabout. generally, the roundabout is suitable for the junction with smaller traffic flow; however, as the number of vehicles increased, the roundabout is exposed to the disadvantages of large area but low traffic capacity. this paper studies the crossroad and the conflict point are aimed at motor vehicles. there are usually 16 conflict points at a crossroads. 2.2. basic parameters (1) phase: at the signal control intersection, each control state, which is the combination of different light colors shown in different directions of each inlet, is called a signal phase. (2) timing period: the combined time of red lights, green lights and yellow lights in one intersection. (3) effective green time: the total period minus the red light time and the time lost when the vehicle starts. (4) green ratio: green ratio refers to the proportion of vehicle passing time and period in the traffic signal period. (5) traffic flow: the sum of all vehicles passed by a certain section of a unit time. (6) queue length: queue length refers to the total length of vehicles parked within the actual line of a certain phase intersection. (7) delay time: delay time is the time difference between the actual taken time to cross the intersection, and the time should be taken when crossing the intersection. (8) parking rate: parking rate is the number of times a car has been parked from one intersection to another. (9) traffic capacity: traffic capacity refers to the maximum number of vehicles that passed at one intersection within a unit of time. (10) green wave belt: according to the time that the vehicle needs to pass a certain section and coordinates the traffic signals at each intersection, then the vehicle can continuously obtain the green light when passing. proceedings of engineering and technology innovation, vol. 11, 2019, pp. 21-29 23 2.3. signal timing method 2.3.1. webster algorithm webster algorithm is the most classic single-intersection traffic signal timing algorithm. this algorithm takes the minimum traffic flow delay as the standard and reasonably sets the green time [1]. webster algorithm has the advantage that: when arterial road is in a heavy traffic flow situation and secondary road is in a less traffic flow situation, this algorithm can guarantee the arterial road traffic and avoids wasting the green time of the secondary road. the disadvantage is that the algorithm's timing effect is almost indistinguishable from the timing control when the road is oversaturated. since webster algorithm only considers the delay time, more optimization algorithms are proposed to add more indicators to the model. such as adding coefficient of parking compensation [2] or traffic capacity [3] into webster algorithm and combined delay time with parking rate to evaluate the signal timing. or combined these two indicators into a nonlinear function model and introduced genetic algorithm and simulated annealing algorithm [4]. 2.3.2. ant colony algorithm an ant colony algorithm is a probabilistic algorithm based on the evolution of bionics. its main purpose is to solve optimization problems. in this paper, the optimization model of ant colony algorithm is proposed to optimize the signal timing of intersection. the main idea of ant colony algorithm is to treat the path of each ant from the origin to the destination as a solution. the set of these solutions constitutes the solution space for the problem we want to optimize. according to the characteristics searching for ant colony, more ants will choose the better path [6], which can be the optimal solution of the problem. 2.3.3. fuzzy control method with the advent of big data technologies, intelligent traffic equipment such as speed measuring equipment and traffic flow measuring equipment are now installed in many urban intersections, which brought great convenience to traffic signal timing research. fuzzy controller is not an accurate calculator, it relies on the expert experience to fuzzy theory. so it is difficult to adjust the membership function and the fuzzy rule in the absence of experience and knowledge. based on the detected real-time traffic flow data of intersections, the fuzzy control theory can be used to optimize the traffic signal timing method dynamically, which is better than webster algorithm [5]. although ignoring important indicators such as queue length, it is a novel idea to apply fuzzy control theory to the dynamic timing of signal lights. 3. traffic signal timing method comprehensive consideration of above algorithms, this paper presents a signal dynamic timing method for intersection with fuzzy control system and improve the ant colony algorithm. in the optimization model of ant colony algorithm, the average delay, the parking rate and the traffic capacity are used as the evaluation criteria to obtain the improved traffic signal period. in the fuzzy control system, a fuzzy controller with double input and single output is adopted to calculate the traffic signal delay, which use the queue length as the evaluation criterion. the experimental results confirm that this approach has superior regulation performance on the intersection vehicles. 3.1. an improved model of ant colony algorithm 3.1.1. basic principle of ant colony algorithm an ant colony algorithm is a heuristic algorithm based on the development of bionics [7]. all paths from the initial point to the terminal point represent all solutions of the problem. according to the process of searching optimal solution by the information exchange of pheromones can be seen that the pheromone concentration is maximum on the optimal path, which causes more ants to choose this path. the positive feedback mechanism applied to the computer is a continuous iteration process from disorder to order [8]. proceedings of engineering and technology innovation, vol. 11, 2019, pp. 21-29 24 for a given directed graph g(v, e), v represents the set of nodes and e represents the set of edges among nodes. the state transfer probability of ant k from the initial point i to the terminal point j at t time is 𝑝𝑖𝑗 𝑘 (𝑡), as shown in eq. (1). 𝛼 and 𝛽 are heuristic factors. 𝜏𝑖𝑗 (𝑡) represents the pheromone concentration of the path from node i to node j at t time. the heuristic function represented by 𝜇𝑖𝑗 (t) is shown in eq. (2), where 𝑑𝑖𝑗 represents the distance between node i and node j. 𝑎𝑙𝑙𝑜𝑤𝑒𝑑𝑘 represents the next optional target node set, as shown in eq. (3), where 𝑡𝑎𝑏𝑢𝑘 represents the tabu list of paths.           , 0, k ij ij k is iss allowedk ij t t j allowed t t p t else                   (1)   1ij ij t d   (2) k k allowed v tabu  (3) the pheromone formula needs to be updated as eq. (4) when all ants pass through n paths. 𝜌 represents the residual coefficient of pheromone. 𝜏𝑖𝑗 𝑘 (𝑡) represents the pheromone left by the k ant [9]. 𝑄 is the mass coefficient and 𝑄 > 0. 𝐿𝑘 indicates the path length of k ant.       1 m k ij ij ijk t n t t         (4)    , , 0, kk ij q if the ant passes by i j l t else         (5) 3.1.2. improved ant colony algorithm the total number of experimental ants is m and each ant correspond to a path in the algorithm, which is a solution to the problem. the updated state transfer probability is defined as eq. (6). 𝜏𝑖𝑗 (𝑡) indicates the amount of residual information on edge (i, j) at t time.       1 k ij mij irr t p t t      (6) at 𝑡 + 𝑛 time, the pheromone concentration of edge (i, j) is updated as eqs. (7) and (8). f is the objective function value at t time, 𝑓𝑚𝑎𝑥 and 𝑓𝑚𝑖𝑛 are the maximum and minimum values of the current objective function.      ij ij ijt n t q t        (7)       max max min ij f f t f f     (8) the algorithm steps are as follows: (1) initialize the number of iterations n = 0 and set up nmax. randomly release ants and set the quantity to m. assign initial values to τij(t) and ∆τij(t). proceedings of engineering and technology innovation, vol. 11, 2019, pp. 21-29 25 (2) n ← n + 1 (3) k ← k + 1 (4) all initial nodes are stored in tabu list and local search is conducted according to the state transfer probability. when the target node is determined, it will be stored in the tabu list until the tabu list is full [10]. (5) update the pheromone of the path and calculate the pheromone increment of the path in the tabu list. (6) update all path pheromones according to eq. (7). find the shortest length of the path and empty the tabu list, then loop through the step (4). (7) to iterate continuously and calculate the optimal solution that satisfies the condition. 3.2. an additional delay method based on fuzzy control system (a) input data (b) output data fig.1 membership function considering the influence of the vehicle queue length of traffic signal timing, this section uses fuzzy control method to calculate the additional delay of traffic signals through the vehicle queue length. the “double input-single output” fuzzy controller is adopted to calculate the additional green light time, according to the queue length of the current phase and the next phase. seven fuzzy have been divided from very short to very long and have been set up for two sets of input data and one been set of output data respectively. and the gauss function is selected as the membership function as shown in figs. 1(a) and (b). the purpose of fuzzy control is to reduce the average delay time. the flow diagram of the simulated intersection is shown in fig. 2. based on the above data, 49 input-output fuzzy rules are established as table 1. table 1 fuzzy rules vs s ls m ll l vl vs vs vs vs vs s s s s vs vs vs s s s s ls s s s ls ls ls ls m ls ls ls m m m m ll m m m m ll ll ll l ll ll ll l l l l vl l l vl vl vl vl vl proceedings of engineering and technology innovation, vol. 11, 2019, pp. 21-29 26 fig. 2 simulated intersection flow diagram the implication relationship of the system is as follows: 49 1 2 48 49 1 ii r r r r r r     (9) finally, the output of additional delay is as table 2. table 2 additional delay delay 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 0 1 1 1 1 1 1 1 1 4 5 5 5 5 5 5 1 1 1 1 1 1 1 2 2 4 5 5 5 5 5 5 2 1 1 1 1 1 2 4 5 5 5 5 5 5 5 5 3 2 2 2 2 2 2 4 5 5 5 5 5 5 5 5 4 4 4 4 4 4 4 6 7 8 8 8 8 8 8 8 5 5 5 5 5 5 5 8 10 10 10 10 10 10 10 10 6 8 8 8 8 8 8 10 12 13 13 13 13 13 13 13 7 10 10 10 10 10 10 12 13 13 13 13 13 13 13 13 8 13 13 13 13 13 13 13 15 17 18 18 18 18 18 18 9 15 15 15 15 15 15 15 16 18 20 20 20 20 20 20 10 18 18 18 18 18 18 20 20 20 22 23 23 23 23 23 11 20 20 20 20 20 20 22 25 25 25 25 25 25 25 25 12 20 20 20 20 21 21 23 25 25 25 25 25 25 25 25 13 25 25 25 25 26 27 28 28 28 28 28 28 28 28 28 14 25 25 25 25 26 29 29 29 29 29 29 29 29 29 29 3.3. experiments 3.3.1. evaluation metrics the delay time, parking rate and traffic capacity are selected as the evaluation metrics criteria for multi-objective optimization [11]. (1) delay time reducing vehicle delay time is conducive to the release of road resources and is an important indicator of road service level. eqs. (10) and (11) is used to calculate the delay time. di represents the average delay time of vehicle in phase i. qi proceedings of engineering and technology innovation, vol. 11, 2019, pp. 21-29 27 represents the actual traffic flow of i phase. c denotes the signal period and λi represents the green ratio of i phase. xi and yi represent the road saturation and traffic intensity of the i phase respectively. y is the sum of the maximum flow ratio of ea ch phase; l stands for the lost time.     2 1 2 1 n i n ii i i i ii n n i ii i c q d q x d q q           (10)  i i y c l yc    (11) (2) parking rate parking rate is also an index to be optimized in this paper. hi is the average parking number of the i phase and h indicates the average parking number of intersections.  0.9 1 1 n i n ii i ii i n n i ii i c q h q y h q q         (12) (3) parking rate the multi-objective optimization algorithm mainly optimizes the traffic capacity of intersection. the total capacity of the intersection is as eq. (13), in which si represents the saturation volume of phase i. n n i i ii i q q s    (13) 3.3.2. model selection choose the intersection of an arterial road in d city. the three phases are selected as follows: the straight line and the right turn line in the direction of east and west are the first phase; the left turn line in the direction of east and west is the second phase; the straight line and the right line in the direction of north and south are the third phase. according to the delay time and the parking rate is the decreasing function and the traffic capacity is the increasing function, the combination objective function can be established. 𝐷0, 𝐻0, 𝑄0 represent the initial values. 𝛼, 𝛽 and 𝛾 respectively represent the weight of delay time, parking rate and traffic capacity. the specific formula is as follows. 0 0 0 min d h q f d h q      (14) 1 1800 y c   (15)  1iy y   (16) cy  (17) table 3 arterial road intersection data entrance traffic flow east west south north left straight 1 straight 2 right straight left straight right straight left straight right total lane flow 198 497 457 440 6 580 304 186 111 phase flow 198 1394 6 580 490 111 proceedings of engineering and technology innovation, vol. 11, 2019, pp. 21-29 28 the experimental data selection is shown in table 3. based on the data can calculate the initial average delay is 29s, the initial parking time is 85 and the initial traffic capacity is 4901 𝑝𝑐𝑢 ℎ⁄ . set the number of ants to 30 and the iterations number to 100. 𝛽 is 0.9 and 𝑄 is 0.8. the optimal solution that obtained by the ant colony optimization algorithm will be combined with the additional delay time that calculated by the fuzzy control method. and the final solution c=149 seconds are obtained. compared with the original period and the traditional algorithm as follows. table 4 comparison of experimental results timing method period green time / green ratio delay time parking rate traffic capacity phase 1 phase 2 phase 3 original 114 57/0.5 23/0.2 25/0.22 29 85 4901 webster algorithm 72 34/0.47 14/0.19 15/0.21 20 57 4448 improved algorithm 149 79/0.53 29/0.19 32/0.22 25 109 5259 as can be seen from the table 4, compared with the current traffic timing scheme of d city, the webster algorithm decreases the average delay time at the expense of reducing traffic capacity. although the method proposed in this paper has increased the parking rate, the traffic capacity has been greatly improved while the average delay time decreases. when the traffic flow is heavy, this improved algorithm will have a significant effect on the traffic diversion at the intersection. 4. conclusions this research puts forward an optimization method of traffic signal timing, which combined the improved ant colony algorithm with fuzzy control method for single intersection signal timing. through experimental verification, the proposed method is confirmed to improve the traffic capacity of intersection vehicles and reduce the delay time of vehicles. traffic signal control system is not only a way to improve the efficiency of the vehicle in a certain intersection, it can also control the traffic flow of the whole traffic network after the gradual optimization of the system, by this way can the urban road capacity achieve to the best. the traffic signal timing system needs to be continued optimize with the increasing number of vehicles. although this research made improvements on the traffic signal timing method, there are still many problems have not been ponderer. to sum up, firstly, this research considered only a case of cross intersection. in actual traffic, there are a lot of traffic nodes in various forms. how to control traffic reasonably according to the different intersections is the main subject of current research. secondly, this paper considered only the traffic condition of motor vehicles when simulating the traffic signal system, and does not take pedestrians and non-motor vehicles into account, since the large number of pedestrians and non-motor vehicles are considered, the difficulty of control will be greatly increased. how to improve the traffic efficiency of vehicles while ensuring the safety of pedestrians and non-motor vehicles through intersections will be the key issue of future research. in addition, the priority of buses is not considered in this paper; signal priority to give the bus green light permissions also increased the difficulty of crossroads control. conflicts of interest the authors declare no conflict of interest. references [1] a. j. miller, “settings for fixed-cycle traffic signals,” journal of the operational research society, vol. 14, no. 4, pp. 373-386, december 1963. [2] m. d. foy, r. f. benekohal, and d. e. goldberg, “signal timing determination using genetic algorithms,” transportation research record, no. 1365, pp. 108-115, 1992. [3] b. park, n. rouphail, and j. sacks, “assessment of stochastic signal optimization method using microsimulation,” transportation research record: journal of the transportation research board, vol. 1748, pp. 40-45, 2001. proceedings of engineering and technology innovation, vol. 11, 2019, pp. 21-29 29 [4] c. p. pappis and e. h. mamdani, “a fuzzy logic controller for a trafc junction,” ieee transactions on systems man & cybernetics, vol. 7, no. 10, pp. 707-717, october 1977. [5] b. abdulhai and r. pringle, “machine learning-based adaptive signal control using autonomous q-learning agents,” proc. the iasted international conf. intelligent systems and control, pp. 14-16,august 2000. [6] j. l. duan, “research on dynamic timing optimization algorithm for urban traffic lights,” m.s. dissertation, shenyang univ., 2014. [7] m. dorigo, v. maniezzo, and a. colorni, “introduzione agli algoritmi naturali,” rivista di informatica, vol. 3, pp. 179-197, 1994. [8] h. p. lu and r. m. li, “development status and trend of urban intelligent transportation system,” engineering research-engineering in the interdisciplinary perspective, vol. 1, pp. 6-19, 2014. [9] y. k. huang, r. z. ding, and z. w. zhao, “urban main road traffic light optimization and simulation program,” shandong industrial technology, vol. 9, pp. 55-57, 2014. [10] s. q. liao and y. l. wang, “traffic signal timing model and genetic algorithm at single intersection,” highway and transportation, vol. 3, pp. 45-48, 2012. [11] x. liang, y. h. yin, and y. wu, “dynamic optimization of signal timing in single intersection,” highway traffic technology, vol. 2, pp. 125-127, 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/). microsoft word 1-v24(2023)-peti#10307(15-21).docx proceedings of engineering and technology innovation, vol. 24, 2023, pp. 15-21 a convolutional neural network for automatic brain tumor detection saeed mohsen1,2,*, wael mohamed fawaz abdel-rehim2,3, ahmed emam2,4, hossam mohamed kasem5 1department of electronics and communications engineering, al-madinah higher institute for engineering and technology, giza, egypt 2department of artificial intelligence engineering, faculty of computer science and engineering, king salman international university (ksiu), south sinai, egypt 3department of computer science, faculty of computers and information, suez university, suez, egypt 4computer science and math department, faculty of science, menoufia university, menoufia, egypt 5electronics and electrical communication engineering department, faculty of engineering, tanta university, tanta, egypt received 12 june 2022; received in revised form 11 december 2022; accepted 22 december 2022 doi: https://doi.org/10.46604/peti.2023.10307 abstract magnetic resonance imaging (mri) combined with artificial intelligence (ai) algorithms to detect brain tumors is one of the important medical applications. in this study, a convolutional neural network (cnn) model is proposed to detect meningioma and pituitary, which was tested with a dataset consisting of two categories of tumors with 1,800 mri images from several persons. the cnn model is trained via a python library, namely tensorflow, with an automatic tuning approach to obtain the highest testing accuracy of tumor detection. the cnn model used python programming language in google colab to detect sensitivity, precision, the area under the pr and receiver operating characteristic (roc), error matrix, and accuracy. the results show that the proposed cnn model has a high performance in the detection of brain tumors. it achieves an accuracy of 95.78% and a weighted average precision of 95.82%. keywords: artificial intelligence, mri, convolutional neural network, brain tumors 1. introduction one of the vital topics in the medical field is brain tumor detection [1-2]. therefore, it is necessary to evaluate the tumor and select a proper treatment decision according to the classes. classification of brain tumors is a challenging problem for physicians because of the heterogeneous nature of tumor cells. physicians contribute to the diagnosis of brain tumor diseases; however, they need an efficient and helpful tool for their diagnosis. computer-aided devices are widely used for brain tumor explorations-based tomography ct scan images, but these devices have low-accuracy diagnoses [3]. in recent years, a robust solution for computer-aided diagnosis devices has been introduced for diagnosing brain tumors via magnetic resonance imaging (mri) [4]. mri is the most useful technique in the detection of brain tumors as it has high image quality [5]. furthermore, artificial intelligence (ai) algorithms are considered a key enabler in assisting brain tumor detection, especially the advance of high-performance deep learning (dl). the dl algorithms with high levels of testing accuracy are a significant phase-oriented and high-precision approach to detecting brain tumor patients. the study aims to provide an accurate and efficient method to differentiate between meningioma and pituitary cases. it helps radiologists and doctors in screening brain tumor cases as well. * corresponding author. e-mail address: g17082131@eng.asu.edu.eg proceedings of engineering and technology innovation, vol. 24, 2023, pp. 15-21 16 machine learning (ml) and dl algorithms are used to classify brain tumors [6]. however, the low accuracy of existing algorithms needs to be improved [7]. convolutional neural network (cnn) has played an effective role in diagnosing brain tumors. it classifies the tumor type through medical imaging of the human brain. mri is a good instance that it has discovered many patients with brain tumors. dl algorithms were presented to classify brain tumor patients based on brain mri images, but a sufficient testing accuracy level was not achieved. afshar et al. [8] proposed a capsnet algorithm to detect brain tumor diseases via a capsule neural network approach. this algorithm was applied to a 3,064-image dataset and had a maximum accuracy of 86.56%. anaraki et al. [9] implemented a cnn algorithm to recognize brain tumors. this algorithm was trained on a dataset of 600 various images and reached an overall testing accuracy of 94.2%. saxena et al. [10] examined vgg-16, inception-v3, and resnet-50 algorithms with a 253-image dataset split into 183, 50, and 20 for training, validation, and testing. the maximum accuracy of inception-v3 and vgg-16 were 55% and 90%, respectively, while the resnet-50 reached a total accuracy of 95%. zhou et al. [11] developed a combination of lstm-densenet algorithms to identify brain tumors that were tested on a dataset of 930 pituitary, 708 meningioma, and 1,426 glioma patients with a maximum accuracy of 92.13%. dl and ml architectures have been introduced to predict brain tumors in patients. cheng et al. [12] proposed a support vector machine (svm) algorithm that reached a total precision of 91.28% with a dataset of 930 pituitaries and 1,426 glioma images. a cnn algorithm was utilized, which reached a testing accuracy of 84.19% with 3,064 mri images [13]. kaplan et al. [14] implemented a k-nearest neighbor (knn) algorithm with the nlbp feature extraction approach, which achieved a 95.56% testing accuracy. pashaei et al. [15] presented a cnn algorithm with an extreme learning machine (elm) technique to train 70% of a 3,064 brain tumor cases dataset, then utilized the algorithm to assess the rest 30%, reaching a testing accuracy of 93.68%. zacharaki et al. [16] employed svm-knn algorithms to classify brain tumors, which were applied to 102 mri images and reached a maximum testing accuracy of 85%. in this study, a cnn model is developed to detect brain tumors. the cnn performance is assessed via several assessing metrics, i.e., sensitivity, precision, the area under the pr and receiver operating characteristic (roc), error matrix, and accuracy. the tumors are categorized into two groups: meningioma and pituitary. the accuracy of the cnn reached 95.78%, while the sensitivity and f1-score were 95.78%. a hyper-parameter tuning approach is utilized to achieve the best performance of the cnn. the paper is organized as follows: the methodology is explained in section 2. the results are presented with the discussion of the paper in section 3. finally, section 4 concludes the study. 2. methodology in this paper, a cnn model is implemented using a tensorflow library. the cnn is built to classify two types of brain tumors: meningioma and pituitary. this model is utilized due to its proper with images dataset where it captures spatial features of these images. all layers of the cnn are tuned via its hyperparameters, and the activation functions used for the cnn are relu and sigmoid. 2.1. the proposed model this study proposed a cnn model for classifying mri brain tumors because the advantage of cnn architecture is the convenience of capturing spatial data. fig. 1 demonstrates the cnn architecture, which has four convolutional layers, three max-pooling layers, four dropout layers, a flattened layer, a fully connected layer, and an output layer. the input shape of the mri images has a height of 224 pixels and a width of 224 pixels. a two-dimensional convolutional layer has been used for each mri image with a relu activation function, which is used to extract spatial features. the layers used in the step are described as follows: proceedings of engineering and technology innovation, vol. 24, 2023, pp. 15-21 17 (1) the convolutional layer includes 32 filters and a 3 × 3 kernel; the other has 64 filters with a relu function. a maxpooling layer includes a 2 × 2 pool, which minimizes the convolutional output complexity to perform a down-sampling process. a dropout layer is used to tackle the over-fitting problem [17]. (2) a 2-d convolutional layer has 64 filters and a 3 × 3 kernel, which is implemented to enable the cnn to classify more features that are lost in the previous convolutional layers. a max-pooling layer has a 2 × 2 pool and a dropout layer with a 25% drop rate from neurons. (3) a 2-d convolutional layer has 128 filters, a 3 × 3 kernel, a max-pooling layer, and a dropout layer with a 0.25 drop rate. (4) a flattened layer is utilized to transform the data into one dimension. the fully connected layer is initialized with 64 neurons with a relu activation function. a dropout layer with a drop rate of 50%, the output layer is configured with a sigmoid function to minimize the output to 2 classes. the proposed model uses a cross-entropy loss function to calculate the error among the actual and prediction values, and the optimizer uses the adam algorithm [18]. the cnn batch size and training iterations are 32 and 50, respectively. the total number of parameters for the implemented cnn is 5,668,162. table 1 shows the used layers of the proposed cnn. fig. 1 the structure of the cnn model table 1 the layers of the cnn layer type kernel size output shape trainable parameters 1 conv2d 3 × 3 222 × 222 × 32 896 2 conv2d 3 × 3 220 × 220 × 64 18496 3 maxpooling2d 2 × 2 110 × 110 × 64 0 4 dropout 110 × 110 × 64 0 5 conv2d 3 × 3 108 × 108 × 64 36928 6 maxpooling2d 2 × 2 54 × 54 × 64 0 7 dropout 54 × 54 × 64 0 8 conv2d 3 × 3 52 × 52 × 128 73856 9 maxpooling2d 2 × 2 26 × 26 × 128 0 10 dropout 26 × 26 × 128 0 11 flatten 86528 0 12 fully connected 64 5537856 13 dropout 64 0 14 output 2 130 2.2. dataset description the mri dataset collected from the repository of kaggle [19] includes 900 images of the meningioma and 900 images of the pituitary. the proposed model is evaluated through the mri dataset collected for testing and training. 1,800 mris are proceedings of engineering and technology innovation, vol. 24, 2023, pp. 15-21 18 divided into 20% of the dataset utilized for testing, while 80% remaining is used for training, where the purpose of the testing is to assess the cnn model, while the purpose of the training is to learn the cnn model features of the mri images. the difference between meningioma and pituitary images is the intensity of their pixels or their features. 3. results and discussion the cnn model is implemented by running python programming language in google colab. the cnn is trained using a high-speed laptop with intel core i3, two central processing units (cpu), and 4gb of ram. fig. 2 shows cnn’s training and testing accuracy curves. at 50 epochs, the accuracy percentage for training is 93.6%; the accuracy percentage for testing is 95.78%. these values mean that the cnn performs better based on the testing and training dataset at 50 training epochs. fig. 3 demonstrates learning curves for the loss of cnn. at 50 epochs, the loss rate for training is 0.1734; the loss rate for testing is 0.1476. analyzing the performance of the model can be described as [20]: tp tn accuracy tp tn fp fn + = + + + (1) tp precision tp fp = + (2) = tp recall tp fn+ (3) 12 precision recall f score precision + recall × = × (4) fig. 2 accuracy curves for the cnn fig. 3 loss curves for the cnn training accuracy testing accuracy training loss testing loss proceedings of engineering and technology innovation, vol. 24, 2023, pp. 15-21 19 table 2 presents a classification report of the brain tumors with evaluation metrics of the recall, precision, and f1-score for the cnn model. in this work, the weighted average of precision, sensitivity, and f1-score are 95.82%, 95.78%, and 95.78%, respectively. these values mean that the cnn performs better based on the testing and training dataset at 50 training epochs. these results are estimated from the confusion matrix that is used to assess the performance of the cnn model. table 2 classification report for the cnn model class precision recall f1-score meningioma 0.9440 0.9733 0.9584 pituitary 0.9725 0.9422 0.9571 macro average 0.9582 0.9578 0.9578 weighted average 0.9582 0.9578 0.9578 fig. 4 describes the confusion matrix determined based on the cnn. in this part, the diagonal values are the testing accuracy to classify the mri images. the matrix lists data of the true positives of meningioma as 219; the true positives of the pituitary are 212. these values mean that the cnn performs better for the meningioma class than the pituitary class. the normalized confusion matrix of the cnn is introduced in fig. 5. the proposed cnn for the accuracy of meningioma is 0.97; the accuracy of the pituitary is 0.94. therefore, classification errors of 0.058 and 0.027 for meningioma and pituitary, respectively. these values mean that the cnn achieves fewer error results for the meningioma class than the pituitary class. fig. 4 confusion matrix for cnn fig. 5 normalized confusion matrix for cnn fig. 6 illustrates the pr curves for each class. the pr curve of class meningioma has the lowest area of 0.987, while the pr curve of class pituitary has the highest area of 0.989. thereby, the micro-average pr curve has an area of 0.998. fig. 7 presents the roc curves, where the area values of the meningioma and pituitary tumor classes are 0.99. the micro-average area for the roc curve is 0.99; thus, it proves that cnn achieves high performance based on these curves. the roc curve is described by the true positive rate (tpr) to the false positive rate (fpr). the pr curve is the ratio between the precision and the recall. fig. 6 pr curves of the cnn predicted label meningioma t ru e l a b e l m e n in g io m a p it u it a ry p it u it a ry m e n in g io m a t ru e l a b e l predicted label meningioma pituitary pituitary pr curve of class meningioma pr curve of class pituitary p re c is io n recall pr curve of class meningioma (area = 0.987) pr curve of class pituitary (area = 0.989) micro-average pr curve (area = 0.988) proceedings of engineering and technology innovation, vol. 24, 2023, pp. 15-21 20 fig. 7 area under roc curves for the cnn the pr curves, normalized confusion matrices, and roc curves show that the accurate classification of brain tumor patients is achieved by using the cnn model. additionally, cnn had a low loss rate and high accuracy when testing and training. table 3 introduces a comparison of the accuracy achieved by previously published works [8-16] and this work. in this study, the cnn achieved a maximum accuracy of 95.78%, which is better than the previous models. table 3 comparison between previous works and the proposed work reference model testing accuracy (%) [8] capsnet 86.56 [9] cnn + ga 94.2 [10] vgg-16 inception-v3 resnet-50 90 55 95 [11] densenet + lstm 91.28 [12] svm 91.28 [13] cnn 84.19 [14] knn + nlbp 95.56 [15] cnn + kelm 93.68 [16] svm 85 proposed work cnn 95.78 4. conclusions this paper has proposed a new cnn model to classify two brain tumors “meningioma” and “pituitary” from mri image patients. it obtains a high testing accuracy of 95.78% in the image dataset of 900 meningioma images and 900 pituitary images. the normalized confusion matrix, recall, precision, f1-score, the area under the pr, and roc curves evaluate the efficiency of the cnn. the cnn has a precision of 95.82%, a recall of 95.78%, and an f1-score of 95.78%. the pr curves with an area of 98.7% for the meningioma class and 98.9% for the pituitary class; the average area under the roc curve is 99% for the meningioma and pituitary classes for cnn. the results show that the new cnn model can improve the detection of brain tumors. conflicts of interest the author declares no conflict of interest. references [1] h. h. sultan, n. m. salem, and w. al-atabany, “multi-classification of brain tumor images using deep neural network,” ieee access, vol. 7, pp. 69215-69225, 2019. roc curve of class meningioma roc curve of class pituitary t ru e p o si ti v e r a te false positive rate roc curve of class meningioma (area = 0.99) roc curve of class pituitary (area = 0.99) micro-average roc curve (area = 0.99) proceedings of engineering and technology innovation, vol. 24, 2023, pp. 15-21 21 [2] m. b. naceur, m. akil, r. saouli, and r. kachouri, “fully automatic brain tumor segmentation with deep learningbased selective attention using overlapping patches and multi-class weighted cross-entropy,” medical image analysis, vol. 63, article no. 101692, july 2020. [3] h. mzoughi, i. njeh, m. b. slima, and a. b. hamida, “review of computer aided-diagnosis (cad) systems for mri gliomas brain tumors explorations based on machine learning and deep learning,” 6th international conference on advanced technologies for signal and image processing, pp. 1-6, may 2022. [4] s. bauer, r. wiest, l. p. nolte, and m. reyes, “a survey of mri-based medical image analysis for brain tumor studies,” physics in medicine & biology, vol. 58, no. 13, pp. r97-r129, july 2013. [5] m. pantoja, m. weyrich, and g. fernández-escribano, “acceleration of mri analysis using multicore and manycore paradigms,” the journal of supercomputing, vol. 76, no. 11, pp. 8679-8690, november 2020. [6] q. t. ostrom, h. gittleman, g. truitt, a. boscia, c. kruchko, and j. s. barnholtz-sloan, “cbtrus statistical report: primary brain and other central nervous system tumors diagnosed in the united states in 2011–2015,” neurooncology, vol. 20, no. suppl_4, pp. iv1-iv86, october 2018. [7] s. mohsen, a. elkaseer, and s. g. scholz, “human activity recognition using k-nearest neighbor machine learning algorithm,” proceedings of the international conference on sustainable design and manufacturing (kes-sdm 2021), smart innovation, systems and technologies, vol, 262, pp. 304-313, september 2021. [8] p. afshar, a. mohammadi, and k. n. plataniotis, “brain tumor type classification via capsule networks,” 25th ieee international conference on image processing, pp. 3129-3133, october 2018. [9] a. k. anaraki, m. ayati, and f. kazemi, “magnetic resonance imaging-based brain tumor grades classification and grading via convolutional neural networks and genetic algorithms,” biocybernetics and biomedical engineering, vol. 39, no. 1, pp. 63-74, january-march 2019. [10] p. saxena, a. maheshwari, s. tayal, and m. saumil, “predictive modeling of brain tumor: a deep learning approach,” computer vision and pattern recognition, november 2019. https://arxiv.org/abs/1911.02265 [11] y. zhou, z. li, h. zhu, c. chen, m. gao, k. xu, et al., “holistic brain tumor screening and classification based on densenet and recurrent neural network,” international miccai brainlesion workshop, brainlesion: glioma, multiple sclerosis, stroke and traumatic brain injuries, lecture notes in computer science, vol. 11383, pp. 208-217, january 2019. [12] j. cheng, w. huang, s. cao, r. yang, w. yang, z. yun, et al., “enhanced performance of brain tumor classification via tumor region augmentation and partition,” plos one, vol. 10, no. 10, article no. e0140381, october 2015. [13] n. abiwinanda, m. hanif, s. t. hesaputra, a. handayani, and t. r. mengko, “brain tumor classification using convolutional neural network,” world congress on medical physics and biomedical engineering 2018, vol. 68/1, pp. 183-189, may 2018. [14] k. kaplan, y. kaya, m. kuncan, and h. m. ertunç, “brain tumor classification using modified local binary patterns (lbp) feature extraction methods,” medical hypotheses, vol. 139, article no. 109696, june 2020. [15] a. pashaei, h. sajedi, and n. jazayeri, “brain tumor classification via convolutional neural network and extreme learning machines,” 8th international conference on computer and knowledge engineering, pp. 314-319, october 2018. [16] e. i. zacharaki, s. wang, s. chawla, d. s. yoo, r. wolf, e. r. melhem, et al., “classification of brain tumor type and grade using mri texture and shape in a machine learning scheme,” magnetic resonance in medicine, vol. 62, no. 6, pp. 1609-1618, december 2009. [17] j. xiong, k. zhang, and h. zhang, “a vibrating mechanism to prevent neural networks from overfitting,” 15th international wireless communications & mobile computing conference, pp. 1737-1742, june 2019. [18] d. p. kingma and j. ba, “adam: a method for stochastic optimization,” 3rd international conference for learning representations, may 2015. https://doi.org/10.48550/arxiv.1412.6980 [19] “brain tumor datasets,” https://www.kaggle.com/datasets/drsaeedmohsen/googledrive, december 15, 2022. [20] s. mohsen, a. elkaseer, and s. g. scholz, “industry 4.0-oriented deep learning models for human activity recognition,” ieee access, vol. 9, pp. 150508-150521, 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/). proceedings of engineering and technology innovation, vol. 8, 2018, pp. 46-59 mfo ptimized fractional order based controller on power system stability bidyadhar rout * , bibhuti bhusan pati department of electrical engineering, veer surendra sai university of technology, burla, odisha, india. received 05 sept ember 2017; received in revised form 06 sept ember 2017; accept ed 17 november 2017 abstract this paper presents a novel idea of designing the fractional-order pid (fopid) type static synchronous series compensator (sssc). a power system stabilize r( pss) is installed to enhance the system transient stability by damping the oscillat ions . also the superiority of the propos ed method is verified by comparing with conventional pi, pi-pd and pid controllers. the determination of the controlle r para meters has been considered as an optimizat ion problem using moth fly optimization (mfo). it is shown that mfo is more effective as well as giving robust response than differentia l evo lution (de) optimizat ion . the superiority of the controlle r is tested on single-machine infinite-bus (smib) power system at various operating conditions and fault locations . keywor ds: moth fly optimization, fractional order controller, pid controller, power system, transient stability 1. introduction recently, the us ing of mfo optimization technique is based on the motion of a moth in a transverse orientation for navigation. it keeps a fixed angle with respect to the moon or the fla me. the superiority of the moth fly optimization (m fo) algorith m over fire fly a lgorith m (fa ) and other algorith ms have been reported in many researches [1]. th is work has focussed on the application of the mfo algorithm to a ty pical smib power system. in a power system, the synchronous series compensator (sssc), is very effective in damp ing the electromechanica l oscillations along with power flow control. it consists of voltage source self-co mmutated switching converters which synthesises the three phase voltages (in quadrature) with line current to establish the compensation of the power system voltage imbalance [2-6]. in dynamic state; sssc ma inly controls da mping of oscillations by injecting th e series voltage to the line [3]. to imp rove the system dynamic performance, an e xte rnal control loop is added to sssc which consists of a controller changing the series injected voltage during transient period [4-6]. such a stabiliser is co mmonly called as lead lag type sssc da mp ing controller. the synchronizing torque, damp ing torque an d transient stability limit fo r both small signal as well as for transient stability are successfully improved with rea l power in lead-lag (ll) based sssc da mping controller and pss design [6-8]. it is well known that the conventional controller design methods are not attractive enough for robust stability, due to computational burden, more t ime consuming, and slow convergence and moreover, the controller para meters are trapped to their local min ima and not optima l. therefore, various optimizat ion methods such as particle swarm optimization (pso), diffe rential evolution (de) etc.[2, 9-11] are e xtensively used for tuning pss and au xiliary controlle r based sssc controller parameters[2, 11-13]. * corresponding author. email address: bdr23@rediffmail.com, bbpati_ee@vssut.ac.in mailto:bdr23@rediffmail.com proceedings of engineering and technology innovation, vol. 8, 2018, pp. 46-59 copyright © taeti 47 because of its robustness and simple structure, the proportional integral derivative (pid) controlle rs have been largely imple mented. [5-7, 14-15]. since, there are difficult ies in finding the mathematica l model as well as determination of parameters of pid controlle r in co mple x/nonlinear higher o rder power systems, artific ia l intelligence techniques have been used [14-17]. in this work mfo optimized controller parameters have been analyzed. in this investigation, the fractional pid type sssc controller will be used and the results will be compared with other conventional controlle rs such as pi type, pi-pd type, and pid type sssc controllers to show the robustness of the da mping at same and different operating conditions for the proposed smib powe r system. the para meters of fractional pid type sssc and pss controllers will be simultaneously optimized by the proposed mfo algorithm [1, 16-18]. 2. system investigated the robustness of damp ing performances has been assessed for the co-ord inately designed controllers and is tested in a smib system as shown in fig. 1 [4-9, 19]. this setup contains a generator connected to an infin ite-bus through a double circuit (dc) t ransmission line. the generator is provided with the sssc, and e xc itation system a long with pss. the sssc is connected in series in between bus-1 and bus -2 through coupling transforme r. a line transformer t lin ks to the generator and bus-1 where as bus-2 and bus-3 are connected through dc trans mission lines. vt and vb are the voltages at the generator terminal and inf inite-bus respectively. v1 and v2 are the bus voltages at bus-1 and bus-2 respectively as shown. current i is the line current, pl1 is the rea l powe r flowing in one of the dc transmission line, where the three phase fault is to be created and pl is the t ie line active powe r flo wing in a transmission line. fig. 1 shows the model power system is developed by using matlab/simu lin k. a ll the re levant parameters of this system are can be found in [13]. the use of pss can also be described in a similar way[5-7]. fig. 1 the single machine infinite bus (smib) power system with a sssc[2-3] 2.1. system modelling the smib nonlinear dynamic mode l deals with the transient analysis [1-5]. the speed deviation is the input to the controller. it is the remote signal fro m the fault location. the following mechanica l dynamics are taken for the proposed system analysis . 1 ( ) r e r m d p b p dt j     (1) r dt d   (2) where 𝜔𝑟 and 𝜃represent the angular speed and the rotor angle as state variables of the generator, pe and pm are the electrical output and mechanical input power respectively, j and b represent inertia and the coefficient of viscous friction of the rotor respectively. detail mathematical expression interlinking the system state variables are described in [9-14]. sssc infinit e-bus double tr. line vs c bus1 bus2 bus3 b v3v2v t v 1l p l p dc v q v t i 1 v ac v c b     c b c b c b lo ad 2l p cnv v generat or proceedings of engineering and technology innovation, vol. 8, 2018, pp. 46-59 copyright © taeti 48 2.2. structure of sssc and control system as discussed, sssc is composed of a three-phase voltage source converter (vsc), a series coupling transformer, a dc capacitor vdc , and a c and dc voltage regulators[2-7]. the objective o f using vsc is to convert a dc voltage into three phase ac voltage with fundamental frequency which is to be fed into line and in phase quadrature (independent fro m line current) with the line current i. the injected a c voltage vq changes in its magnitude due to variable fictit ious capacitive or inductive reactance during transient conditions. this voltage controls the active power flow efficiently and da mps out the power swings. in capacitive mode, vcnv is greater than vac and it supplies active and reactive power to power system and in inductive mode, vcnv is lower than vac. the control device ma intains the voltage profile of transmission line unchanged by controlling the converter voltage[2, 5, 19]. 2.3. fractional order pid (fopid) controller widespread interest in fopid controlle r has attracted many researchers in power system to provide insight into the transient stability study. it is represented by pi d   which provides added degree of freedom for design ing controller gains (kp, ki, kd) where the orders of integral and derivative are real nu mbers not necessarily only integers . the fopid controller has the merit of provid ing an e xtra degree of freedom and is less sensitive to para meter variation co mpared to a c lassical pid controller [17]. the transfer function of such fopid controller: ( ) ( )i c p d k g s k k s s      (3) 3. the structures of various controllers the structures of various types of controllers e.g. pi type, pi-pd type, pid type and the fopid controller have been described in the following section for both the sssc as well as pss installation in the example system. 3.1 pi type sssc controller structure fig. 2 shows the block diagra m o f pi type sssc controlle r structure. here kp is the proportional ga in and ki is the integral ga in. the pi block is connected to a washout block(for high pass filter action) followed by a two stage lead lag compensation block. the wash out block with time constant tw ma ke it sure that there is no steady state error of the voltage reference due to the speed deviation ∆𝑤. time constant tw is not critica l and may be in the range of 1 to 20 sec [12]. he re, tw = 10s is taken into consideration. as the input signal (i.e. ∆𝑤) to the controller is remote signal, the sensor time with transmission time delay ttd = 65ms is provided at the input to the controller. 31 2 4 111 ( )( )( )( )( ) 0.065 1 1 1 1 w ii i ssscpi p w i i st stk st u k y s s st st st        (4) fig. 2 structure of pi type sssc damping controller 3.2 pi type pss controller structure 311 11 1 21 41 111 ( )( )( )( )( ) 0.015 1 1 1 1 w ii i psspi p w i i st stk st u k y s s st st st        (5)   washout 2-st age lead-lag w w st st 1 1 2 1 1 i i st st   3 4 1 1 i i st st   q max v q min vsensor wit h t ransmission time delay out put i k p k    qref v q v qcnv v  p roportional gain 1 1 td st integral gain  input signal proceedings of engineering and technology innovation, vol. 8, 2018, pp. 46-59 copyright © taeti 49 fig. 3 structure of pi type pss damping controller the controller gains and time constant parameters of pi type sssc and pss controlle r must be tuned for achieving best performances. a first order sensor with sensor time de lay ttd = 15ms is chosen by this controller for sensing the low frequency speed deviation ∆𝑤 during disturbances [8]. 3.3. pi-pd type sssc controller structure improved system performance is e xpected with the cascade control system. fig. 4 shows a cascade pi-pd controller fo r sssc controller [7-10]. the performances are improved by putting a first order filter with tuning pole and filter constant n = 100 similar pi-pd type pss controller structure can be used for analysing the behaviour of pss damping controller. 32 1 2 3 3 2 4 111 ( )( )( )( )( )( ) 0.065 1 1 1 1 w idi id ssscpi pd p d p w id id st stk stsn u k k k y s s s n st st st           (6) fig. 4 structure of pi-pd type sssc damping controller[19] 3.4 pid type sssc controller structure conventional pid type sssc controlle r structure is taken for co mparing the responses with the previous controller structures. for the conventional pid type sssc damping controller, the output is : 1 3 2 4 1 11 ( )( )( )( )( ) 0.065 1 1 1 1 ic w c c ssscpid dc pc w c c k st st stsn u k k y s s s n st st st           (7) fig. 5 structure of conventional pid type sssc controller similar pid type pss controller structure can be used for analysing the behavious of pss damping controller s. here, kpc, kic kdc, kp1c, ki1c, kd1c, t1c, t2c, t3c, t4c, t11c, t21c, t31c, t41c are gains and time constant parameters of pid type sssc and its pid type pss controller that are to be tuned for achieving best performances [20].    input signal washout 2-st age lead-lag w w st st 1 1 2 1 1 c c st st   3 4 1 1 c c st st   q max v q min vsensor wit h t ransmission t ime delay out put ic k pc k    qref v q v qcnv v dc k  derivat ive gain p roportional gain 1 1 td st  n int egral gain s 1     input signal washout 2-st age lead-lag w w st st 1 1 2 1 1 id id st st   3 4 1 1 id id st st   q max v q min v sensor wit h t ransmission t ime delay out put    qref v q v qcnv v  p roportional gain 1 1 td st  2i k  1 s 2p k    3d k 3p k n 1 s  p roportional gain derivat ive gain p roportional gain   input signal washout 2-st age lead-lag 1 w w st st 11 21 1 1 i i st st   31 41 1 1 i i st st   f max v f min vsensort imedelay out put 1i k 1p k    qref v q v f v  p roportional gain 1 1 td st int egral gain  proceedings of engineering and technology innovation, vol. 8, 2018, pp. 46-59 copyright © taeti 50 3.5 fopid type sssc controller structure for fractional type sssc controller, the output is: 31 2 4 111 ( )( )( )( )( ) 0.065 1 1 1 1 wi ssscfpid d p w st stk st u k s k y s s st st st           (8) the proposed fractional pid type sssc is a supplementary da mp ing controlle r whose structure is shown in fig. 6, is to modulate the sssc injected voltage .qv similar fopid type pss controlle r structure has been applied to analyse the damping behaviour. a first order sensor with sensor time de lay 𝑇𝑡𝑑 =15ms is taken with this controller for sensing the low frequency speed deviation w during disturbances. the controller gain 𝐾𝑝 ,𝐾𝐼 𝐾𝑑 ,𝐾𝑝1 ,𝐾𝐼1 , 𝐾𝑑1 and the time constant values like𝑇1 , 𝑇2, 𝑇3 , 𝑇4, 𝑇11 , 𝑇21 , 𝑇31, 𝑇41 , , , 1 and 𝜇 1are to be tuned optimally [21]. fig. 6 structure of fopid type sssc damping controller 3.6. control objectives power system after passing through a large disturbance exhib it oscillation which can be minimized or quenched by application of sssc and this improves the stability. these oscillations are observed in terms of powe r angle, rotor speed and line power. the objective is to reduce any one or all of these deviations . in the present study, an integral time absolute error (ita e) of the speed signals corresponding to the remote modes of oscillat ions is taken as the objective function. the objective function is expressed as dttwj simtt t .. 0     (9) where tsim is the simulat ion time period. the percentage overshoot and the settling time can be imp roved by minimizing this objective function. the constraints here are the parameters of sssc controlle r. the optimisation problem can be as the following optimization problem: minimizing j, and subject to for pi type sssc damping controller min max min max , p p p i i i k k k k k k    (10) min max min max min max min max 1 1 1 2 2 2 3 3 3 4 4 4 , , , i i i i i i i i i i i i t t t t t t t t t t t t        (11) for pi type pss controller min max min max 1 1 1 1 1 1 , p p p i i i k k k k k k    (12) min max min max min max min max 11 11 11 21 21 21 31 31 31 41 41 41 , , , i i i i i i i i i i i i t t t t t t t t t t t t        (13) for pi-pd type sssc damping controller  1 s     input signal washout 2-st age lead-lag w w st st 1 1 2 1 1 st st   3 4 1 1 st st   q max v q min vsensor wit h t ransmission t ime delay out put 1p k    qref v q v q v1dk s   int egral gain derivat ive gain p roportional gain 1 1 td st 1i k proceedings of engineering and technology innovation, vol. 8, 2018, pp. 46-59 copyright © taeti 51 min max min max min max min max 2 2 2 2 2 2 3 3 3 3 3 3 , , , p p p i i i p p p d d d k k k k k k k k k k k k        (14) min max min max min max min max 1 1 1 2 2 2 3 3 3 4 4 4 , , , id id id id id id id id id id id id t t t t t t t t t t t t        (15) for pi-pd type pss controller min max min max min max min max 12 12 12 12 12 12 13 13 13 13 13 13 , , , p p p i i i p p p d d d k k k k k k k k k k k k        (16) min max min max min max min max 11 11 11 21 21 21 3 31 31 41 41 41 , , , id id id id id id id id id id id id t t t t t t t t t t t t        (17) for pid type sssc damping controller min max min max min max , , pc pc pc ic ic ic dc dc dc k k k k k k k k k      (18) min max min max min max min max 1 1 1 2 2 2 3 3 3 4 4 4 , , , c c c c c c c c c c c c t t t t t t t t t t t t        (19) for pid type pss controller min max min max min max 1 1 1 1 1 1 1 1 1 , , p c p c p c i c i c i c d c d c d c k k k k k k k k k      (20) min max min max min max min max 11 11 11 21 21 21 31 31 31 41 41 41 , , , c c c c c c c c c c c c t t t t t t t t t t t t        (21) for fopid type sssc damping controller min max min max min max , , p p p i i i d d d k k k k k k k k k      (22) max 44 min 4 max 33 min 3 max 22 min 2 max 11 min 1 ,,, tttttttttttt  (23) maxmin   (24) for fopid type pss controller min max min max min max 1 1 1 1 1 1 1 1 1 , , p p p i i i d d d k k k k k k k k k      (25) min max min max min max min max 11 11 11 21 21 21 31 31 31 41 41 41 , , ,t t t t t t t t t t t t        (26) min max 1 1 1     (27) in this investigation de and mfo techniques are applied to search for optima l set of sssc-based damping controlle r parameters. a brief introduction of moth-flame optimization(mfo) technique is described in the next section. 4. overview of mfo this has been wide ly used in many such important applications and giving very pro mising results. it is brie fly d iscussed in this section to make the paper self-content. the behavior of the fly is mathe matica lly modeled by an algorithm ca lled mfo algorith m, assuming the candidate solutions are moths and the proble m’s variab les are the position of moths in the space. this algorith m is a population -based algorithm, where the set of moths is represented in a matrix m consisting of the array as their corresponding fitness values. the fla me matrix is represented by where the array of f also stores the fitness values. the init ial point, final po int of the fla me and range of fluctuation of spira l in the search space are required for the spira l movement of the moth which is expressed as ( , ) i i j m s m f (28) proceedings of engineering and technology innovation, vol. 8, 2018, pp. 46-59 copyright © taeti 52 where mi indicates the i th moth position, fi indicates the j th fla me position. knowing the init ial point, final point of the flame and range of fluctuation of spiral in the search space, the spiral movement of the moth is expressed as j bt iji ftcosedfms  )2(..),(  (29) where t is a random number [-1, 1], and 𝐷𝑖 is the distance of the i th moth for the j th flame. and is calculated as: iji mfd  (30) where mi indicates the i th moth, fj indicates the j th fla me, and di indicates the distance of the i th moth for the j th flame wh ich is to be minimized. the next position of a moth is defined with respect to a fla me following eq. (29). the t-para meter in the equation which closeness to flame equals to -1 and 1 imp lies farther to the fla me. due to change of order of fla mes and revision of moth’s position at each iteration, the e xp loitation of the pro misingly best solutions is obtained through an adaptive mechanism proposed by using the number of flames. ) 1 *(_ t n lnroundnoflame   (31) where n and t represents ma ximu m no. of fla mes , and iterat ions respectively, l is the current nu mber of iteration . the algorithm flow chart is shown in fig. 7. fig. 7 flow chart of the proposed mfo algorithm[1] proceedings of engineering and technology innovation, vol. 8, 2018, pp. 46-59 copyright © taeti 53 5. simulation results and analysis the system described in fig. 1 is simulated in matla b under three phase disturbance at different locations . the objective function is min imised by tuning the fractional pid type sssc and pss controller para meters with m fo and compared with de algorith m as shown in tab le 4. the result is compared with pi type sssc and pss (table 1), pi-pd type sssc and pss (table 2), pid type sssc and pss da mp ing controllers (tab le 3). the robustness and effectiveness of the proposed fpid type sssc and pss is verified at va rious generator loadings. after hundred iterations, the optimized parameters a long with the performance indices of itae are noted down fro m both de and mfo a lgorithms. the simu lation results for all controllers are shown in figs. 8-11. 5.1. case-1: pi type sssc-pss controller the nomina l loading 0 0(0.8 . , 48.48 )p u   , is the self-clea red between bus -2 and bus -3. a three phase line fault is created at one of the double section line between bus -2 and bus -3 at t = 1s in smib power system for nominal loading and self-clea red for5 cycles. the post fault oscillat ion are da mped out through mfo and de optimised pi type sssc and pss controller and the effective results are found through mfo optimized para meters. the para meters comparison is shown in table-1 and the simulat ion responses are shown in fig s. 8-11with the legend “pi type sssc (mfo)” with a solid green line and “pi type sssc (de)” with dotted pink line. table 1 de and mfo tuned optimal parameter of pi type sssc and pss controller at nominal loading 0 0 (0.8 . , 48.48 )p u   optimisation controller controller parameters itae *10e-3 de pi type sssc kp ki t1i t2i t3i t4i 3.8 195.403 12.9818 0.5881 0.6896 1.4308 1.9145 pi type pss kp1 ki1 t11i t21i t31i t41i 21.0723 11.0937 1.2773 1.6716 0.0146 1.9139 mfo pi type sssc kp ki t1i t2i t3i t4i 3.7 158.3171 6.3069 1.7456 1.9974 0.4081 0.8803 pi type pss kp1 ki1 t11i t21i t31i t41i 1.7652 11.4863 1.1977 0.0010 1.7735 1.6014 fig. 8 speed deviation w in p.u fig. 9 rotor angle in in degree fig. 10 the line active power l p in mw fig. 11 sssc-injected voltage q v in p.u proceedings of engineering and technology innovation, vol. 8, 2018, pp. 46-59 copyright © taeti 54 5.2. case-2:pi-pd type sssc and pss controller the nominal loading 0 0(0.8 . , 48.48 )p u   , is the self-clea red at between bus -2 and bus -3. table 2 shows the optimised parameters with both de and mfo for pi-pd type sssc and pss controller during three phase bus -2 and bus -3 and selfcleared condition fo r 5 cyc les . the effective responses are analysed in fig.9. it is observed in fig. 12-15 for speed deviation, tie line power and termina l voltage. the improve ment of settling time and decrement of overshot of transient responses are observed in this mfo optimised pi-pd type sssc and pss controller more effectively than controller. table 2 de and mfo tuned optimal parameter of pi-pd type sssc and pss controller with nominal loading 0 0 (0.8 . , 48.48 )p u   line outage op timisation controller controller parameters itae *10e-3 de pi-pd ty p e sssc kp2 ki2 kp3 kd3 t1id t2id t3id t4id 4.2 8.9062 7.7024 91.2532 32.1083 0.9168 1.7648 0.2907 0.7114 pi-pd ty p e pss kp12 ki12 kp13 kd13 t11id t21id t31id t41id 55.8208 17.3684 15.7298 12.5812 0.8535 1.4744 0.2200 1.4390 m fo pi-pd ty p e sssc kp2 ki2 kp3 kd3 t1id t2id t3id t4id 2.9 0.9990 8.2677 102.8586 92.6227 1.0339 0.6070 0.9069 1.6722 pi-pd ty p e pss kp12 ki12 kp13 kd13 t11id t21id t31id t41id 0.9990 0.0000 99.87 9.3844 0.0001 1.9974 0.0001 1.8503 fig. 12 speed deviation fig. 13 rotor angle in degree fig. 14 tie line active power fig. 15 terminal voltage 5.3. case-3: pid type sssc and pss controller the nominal loading at ( 0 0 0.8, 48.4 e p   ), is the line outage bus -2 and bus -3. pid type sssc and pss controlle r has also been tested with this smib system at line outage disturbance condition at same fault location between bus -2 and bus -3.the three phase line outage disturbance is done at t = 1s for 5 cycles and the e ffective responses are analysed in figs. 16-18 following optimized controlle r para meters in table -3. it is found that, the responses for pid type sssc and pss controller are better as co mpared to pi and pi-pd type sssc and pss controlle r. the m fo optimized controller para meters of pid type sssc and pss g ives better responses than the de tuned pid type sssc and pss controlle r. the settling time in pid type sssc is significantly improved than pi and pi-pd type sssc controllers. proceedings of engineering and technology innovation, vol. 8, 2018, pp. 46-59 copyright © taeti 55 fig. 16 speed deviation fig. 17 the line active power fig. 18 sssc-injected voltage table 3 de and mfo tuned optimal parameter of pid type sssc and pss controller with nominal loading (0.8p 0 0 (0.8 . , 48.48 )p u   , line outage optimisation controller controller parameters itae *10e-3 de pid type sssc kpc kic kdc t1c t2c t3c t4c 3.7 134.1158 13.9487 12.3751 1.7451 1.5531 0.2802 0.5056 pid type pss kp1c ki1c kd1c t11c t21c t31c t41c 111.8589 10.1969 48.8007 1.559 0.4859 0.1774 1.6069 mfo pid type sssc kpc kic kdc t1c t2c t3c t4c 3.2 469.7830 3.2563 18.4041 1.4773 1.6225 1.1597 1.7668 pid type pss kp1c ki1c kd1c t11c t21c t31c t41c 0.9990 40.4269 2.4261 0.7406 1.8584 1.8608 1.8763 5.4. case-4: fractional pid type sssc and pss controller the nominal loading at ( 0 0 0.8, 48.4 e p   ) line outage disturbance is at bus-1. the proposed fractional pid type sssc and pss controllers have been imple mented in this work to show their better quality over all the controllers. the transient analysis is done for the line outage disturbances of smib test system at no minal loading. the d isturbance is create d at 1 sec at one of the double circuit transmission line for 5 cycles and simulat ion results are studied from the fig. 19-22 for speed deviation, tie line active power, terminal voltage and sssc injected voltage , respectively. the settling time, overshoot and undershoot of the transient responses are improved effect ively with this proposed controller as shown in table 4. the time de lay of 50 m sec is introduced between the fopid type sssc and pss. optimized with m fo algorith m, the proposed controller significantly improves the transient response as compared to de optimized fopid type sssc and pss controller and all other controllers. table 4 case-4: de and mfo tuned optimal parameter of fopid type sssc and pss controller with nominal loading 0 0 (0.8 . , 48.48 )p u   , line outage proceedings of engineering and technology innovation, vol. 8, 2018, pp. 46-59 copyright © taeti 56 fig. 19 speed deviation fig. 20 terminal voltage fig. 21 tie line active power fig. 22 sssc-injected voltage controller 5.5 case-5: fractional pid type sssc and pss controlle r the nomina l loading at 0 0(0.8 . , 48.48 )p u   line outage disturbance is between bus -2 and bus -3. the mfo optimized speed responses of all the controllers are taken together in fig. 23 and revealed that, the fractional pid with fractional number of integral ga in and derivative gain e ffective ly improves the dynamic responses as compared to rest of the controllers. the zoo m portion of speed deviation shown in fig. 24 c learly shows the speed response has no second overshoot, undershoot is less and it takes 2 sec to settle for damping out the complete oscillation s. op timisation controller controller parameters itae *10e-3 de fpid ty p e sssc kp ki kd t1 t2 t3 t4  μ 3.4 294.7739 54.5911 16.8004 0.5892 1.1404 1.6048 0.7965 0.6887 0.9087 fpid ty p e pss kp1 ki1 kd1 t11 t21 t31 t41 1  μ1 57.7629 11.4368 31.4137 0.2451 1.3925 0.2039 0.9830 0.4103 0.9008 m fo fpid ty p e sssc kp ki kd t1 t2 t3 t4  μ 2.9 100.2096 37.6333 8.6803 1.8305 1.9974 1.5018 1.7566 1.0185 0.9987 fpid ty p e pss kp1 ki1 kd1 t11 t21 t31 t41 1  μ1 161.2293 94.0954 27.9560 0.0920 0.0001 0.5633 1.4125 1.1644 1.1644 fig. 23 speed deviation for different controllers compared with proposed fopid type sssc and pss controller fig. 24 speed deviation in zoom version proceedings of engineering and technology innovation, vol. 8, 2018, pp. 46-59 copyright © taeti 57 5.6. case-6: fopid type sssc and pss controller nominal loading, permanent line tripping under severe disturbance condition, the proposed controller is verified at nominal loading. a three phase fault for 5 cycles is created at the mid-point of the overhead line of bus -2 and bus -3 and the fault is clea red by indefin ite tripping of the faulted section. the responses of the system are shown in the figs. 25-27. the speed deviation is shown in fig. 25 in p.u, the rotor angle in deg in fig. 26, and the healthy line power 2lp in mw in fig. 27, respectively. the system responses are unstable without the controller and becomes stable by fopid type sssc type pss da mping controller. this controlle r brings the initia l operating point of the speed deviation at 3 sec and the rotor angle in fig. 26 is shifted to another operating point at 60° from its initial angle of 48.48°after 2.6 sec. fig. 25 speed deviation in p.u. fig. 26 rotor angle in degree fig. 27 power at healthy line 5.7. case-7: fpid type sssc and pss controller with light loading 0(0.4 . , 22.85 )p u and self-clearing, light load perfo rmances are a lso performed by changing the generator loading and the stability of the smib system is checked with the proposed fpid type sssc and pss da mping controller. the responses are shown in the fig. 28-30 with three phase fault for 5 cyc les at middle of the double transmission line. the rotor angle in fig. 29 reduces 0 22.85 (without controller) to 0 22.5 (with controller) and settling time is achieved at 3 sec. the e ffect iveness of this proposed controller is also verified in fig s . 28, 30 and 31 by showing the da mping of the oscillations and reduced overshoot and undershoot after the first swing. fig. 28 speed deviation proceedings of engineering and technology innovation, vol. 8, 2018, pp. 46-59 copyright © taeti 58 fig. 29 rotor angle fig. 30 tie line powerin mw 6. conclusions the main purpose here is to damp the powe r system oscillat ions and to enhance the system performance under disturbances. the coordinated pi, pi-pd, pid and fpid type sssc and pss da mping controllers are designed and imple mented in smib power system. the perfo rmance of fractional pid type sssc and pss has been compared with other controllers by its itae based on speed deviation of the test system and found minimu m ita e as objective function . with optimizing the controller para meters by mfo, the robustness and effectiveness are verified under various contingencies. the system damping has also been compared by using proposed controller with de, and has concluded that the mfo technique yields better ita e value and better dynamic response. the simu lation work has been done in matla b environ ment by running several times with adjusting controlling variables f= 0.5-0.8, cr= 0.5-0.8, strategy=1-3, iteration, populations and the range of other controller para meters within its limit in de technique. results obtained are good. however, in the proposed controller, 50 iterations, 10 population, f=0.8, cr=0.8, strategy=3 are co nsidered and much better result are obtained. similarly, 5 agents and 50 iterat ions are taken for mfo technique for still better results as co mpared to de in thi s proposed test systems. the results indicate the e ffectiveness of the proposed designs in prov iding good damp ing characterist ics to power system oscillat ions. this also helps in enhancement of dynamic performance. this approach can be imp le mented with mult i machine grid connected power systems. references [1] s. mirja lili, “moth-fla me optimization algorithm: a novel nature -inspired heuristic paradigm,” knowledge-based system, vol. 89, pp. 228-249, november 2015. [2] s. m. h. hosseini, h. samad zadeh, j. ola mae i, and m. fa rsadi, “ssr mit igation with sssc thanks to fuzzy control,” turkish journal of electrical engineering & computer science s, vol. 21, pp. 2294-2306, january 2013. [3] m. klein, g. j. rogers, and p. kundur, “a fundamental study of inter-area oscillation in powe r systems,” ieee press, august 1991, pp. 914-921. [4] c. liu, g. ca i, and d. yang, “design nonlinear robust damping controlle r for static synchronous series compensator based on objective holographic feedback-h,” journal of advances in mechanical engineering (sage journal), vol. 8, no. 6, pp. 1-11, june 2016. [5] m. bongiorno, j. svensson, and l. angquist, “on control of static synchronous series compensator for ssr mitigation,” ieee transactions on power electronics, ieee press, june 2007, pp. 735-743. [6] l. bangjun and f. shu min, “a brand new nonlinear robust control design of sssc for t ransient stability and damp ing improve ment of mu lti-machine power systems via pseudo -generalized ha miltonian theory,” control engineering practice, vol. 29, pp. 147-157, august 2014. [7] u. q. sun, y. k. sun, and x. x. liu, “ h2/hn cost -guaranteed control for the static synchronous compensator,” iet control theory, vol. 69, pp. 641-646, 2009. http://journals.sagepub.com/author/liu%2c+cheng http://journals.sagepub.com/author/cai%2c+guowei http://journals.sagepub.com/author/yang%2c+deyou proceedings of engineering and technology innovation, vol. 8, 2018, pp. 46-59 copyright © taeti 59 [8] m. s. castro, h. m. ayres, v. f. da costa, and l. c. p. da silva, “impacts of the sssc control modes on small-signal and transient stability of a power system,” electrical power system research, vol. 77, no. 1, pp. 1-9, january 2007. [9] m. a. abido, “ pole p lace ment technique for pss and t csc based stabilizer design using simulated annealing,” international journal of electrical power systems research, vol. 22, no. 8, pp. 543-554, november 2000. [10] m. e. about-ela, a. a. sa lla m, j. d. mc ca lley, and a. a. fouad, “damp ing controller design for power system oscillations using global signals,” ieee transactions on power systems, ieee press, may 1996, pp. 767-773. [11] m. a. abido, “ optima l design of power system stabilizers using particle swa rm optimization,” ieee trans energy convers, ieee press, november 2002, pp. 406-413. [12] z. l. ga ing, “a partic le swarm optimization approach for optimu m design of pid controller in a vr system,” ieee trans energy convers 19, ieee press, may 2004, pp. 384-391. [13] s. panda, “robust coordinated design of mult iple and mu lti-type damping controller using differentia l evolution algorithm,” international journal of electrical power & energy systems, vol. 33, no. 4, pp. 1018-1030, may 2011. [14] h. e. mostafa, m. a. el-sharkawy, a. a. emary, and k. yassin, “design and a lloca tion of powe r system stabilize rs using the particle swarm optimization technique for an interconnected power system,” int ernational journal electrical power energy systems ., vol. 34, no. 1, pp. 57-65, january 2012. [15] t. t. nguyen and r. gianto, “application of optimization method for control co -ordination of psss and facts devices to enhance small-disturbance stability,” proc. ieee pes transmission & distribution conf. may da llas, ieee press, may 2006, pp. 21-24. [16] y. li and k. h. ang, “pid control system analy sis and design,” ieee control systems . magazines , ieee press, june 2005, pp. 559-576. [17] c. h. lee and f. k. chang, “fractional -order pid controlle r optimization v ia improved electro magnetis m-like algorithm,” expert systems with application, vol. 37, no. 12, pp. 8871-8878, december 2010. [18] f. padula and a. visioli, “tuning rules for optima l pid and fract ional-order pid controllers,” journal of process control, vol. 21, no. 1, pp. 69-81, january 2011. [19] r. k. khadanga and j. k. satapathy, “time delay approach for pss and sssc based coordinated controller design using hybrid pso-gsa algorith m,” international journal of electrica l powe r and energy systems, vol. 71, pp. 262 -273, october 2015. [20] s. s. moha med, a. e. mansour, and m. a. abdel ghany, “ design of fractional order pid controller for smib power system with upfc tuned by mu lti-object ives genetic algorith m,” proc. 16 th int. conf. on aerospace sciences & aviation technology, pp. 26-28, may 2015. [21] m. r. faieghi and a. ne mati, “ on fract ional-order pid design, applications of matlab in science and engineering,” book edited by prof. tadeusz michalowski, pp. 237-292, september 2011.  proceedings of engineering and technology innovation, vol. 13, 2019, pp. 32-40 0d/2d p-doped znxcd1-xs/g-c3n4 heterojunctions towards highly efficient photocatalytic hydrogen evolution hongfei yin, xiaoheng liu * key laboratory of education ministry for soft chemistry and functional materials, nanjing university of science and technology, nanjing, china received 02 march 2019; received in revised form 24 april 2019; accepted 29 may 2019 abstract solar-driven water splitting over semiconductor photocatalysts is a promising approach for converting solar energy into renewable and storable chemical energy. cds, an efficient photocatalyst for hydrogen evolution, suffers from high recombination of photogenerated electron-hole pairs and high photocorrosion in aqueous media, which limit its practical applications. doping and the formation of heterojunctions are two efficient methods to improve the photocatalytic activities of photocatalysts. herein, we rationally designed and fabricated p-doped 0d/2d znxcd1-xs/g-c3n4 nanocomposites by in-situ immobilizing znxcd1-xs onto the surface of g-c3n4 nanosheets in a hydrothermal environment, followed by a phosphorization process. the as-prepared p-doped znxcd1-xs/g-c3n4 nanocomposites were systematically characterized by analyzing the phase structure, chemical components, electronic and optical properties and separation of charge carriers. more importantly, these p-doped znxcd1-xs/g-c3n4 heterostructures have been proven to be highly efficient visible light responsive photocatalysts for hydrogen evolution, and meanwhile exhibit excellent photo-stability during recycling runs. the sufficient evidence exhibit that the significantly improved photocatalytic performance is mainly attributed to the prolonged lifetime of charge carriers and the improved separation efficiency of photogenerated electron-hole pairs. keywords: p-doped znxcd1-xs/g-c3n4, heterojunctions, photocatalysis, hydrogen evolution 1. introduction since the discovery of photocatalytic water splitting on tio2 electrodes by fujishima and honda in 1972, the utilization of solar energy for the alleviation of steadily worsening environmental issues and energy crisis has attracted a lot of research interests, and a large number of photocatalytic materials, including metal oxide, metal sulfide, organic compounds, and composites, have been developed for various applications such as water splitting, co2 reduction, wastewater treatment and nitrogen fixation [1-5]. however, pure tio2 demonstrated extremely low photocatalytic activity due to its relatively large band gap (3.2 ev for the anatase phase and 3.0 ev for the rutile phase). the limited uv-responsive activity of tio2 largely inhibits its large scale application because uv light makes up just a small part of the total solar spectrum reaching the earth surface. as in so doing, to explore and design visible light driven photocatalysts is very important to improve the photocatalytic efficiency towards practical applications [6–8]. recently, much research has been focused on the development of visible-light-responsive photocatalysts to take advantage of the solar light resources more effectively because visible light constitutes a larger proportion than uv light in solar light. in fact, metal chalcogenides due to appropriate band gap width and band edge position have elicited more and more * corresponding author. e-mail address: xhliu@njust.edu.cn tel.: +025-84315943; fax: +025-84315943 33 proceedings of engineering and technology innovation, vol. 13, 2019, pp. 32-40 attention in photocatalytic h2 evolution [9]. especially, cds with suitable band edge positions and appropriate band gap of 2.4 ev has been proven to be an efficient semiconductor h2-production photocatalyst and has been extensively studied [10]. however, there are several obvious problems that restrict the wide application of cds to a large extent. in detail, quick recombination of photogenerated charge carriers and instability during the photocatalytic reaction owing to photocorrosion, where s 2 in cds can be oxidized by photogenerated holes accompanied by the release of cd 2+ [11]. these drawbacks can be overcome by elemental doping into cds lattices, such as ni, mn, zn and p [12-15], and constructing type-ii semiconductor heterostructures [16-19], due to its effectiveness for spatially separating the photogenerated electron-hole pairs through the band alignment between two semiconductors [20, 21]. znxcd1−xs solid solution, as a hydrogen-producing photocatalyst with superior performance, has been extensively studied due to its tunable band structure, which makes it possible to balance the redox ability of photo-induced charge carries and absorption of visible light. to date, znxcd1−xs solid solution is the most efficient sulfide photocatalyst for hydrogen production without any co-catalyst [22, 23]. unfortunately, the catalytic activity of znxcd1−xs solid solution is still limited due to the high recombination rate of electron-hole pairs. metal-free polymeric graphitic carbon-nitrogen(g-c3n4) with suitable band edge positions and appropriate band gap has emerged as appealing visible light driven photocatalyst and widely applied in the field of photocatalysis due to its simple syntheses, nontoxicity, earth-abundant, and excellent structural stability in the thermal and chemical environment [16, 17]. however, the low specific surface area and high recombination rate of photogenerated electron-hole pairs originate from the bulk structure and the presence of internal grain boundary, lead to the low photocatalytic activity, largely limiting the practical application of bulk g-c3n4 [24, 25]. based on the band structure of znxcd1−xs and g-c3n4, guo et al. combined g-c3n4 and cdxzn1−xs to construct heterojunctions by hydrothermal method, the activity of the composite is 1.95 times than that of pure cd0.5zn0.5s [26]. jin et al. reported that water splitting performance of the binary cd0.2zn0.8s/g-c3n4 composite photocatalyst is about 1.5 times higher than pure cd0.2zn0.8s [27]. in the present work, we rationally designed and fabricated p-doped 0d/2d znxcd1-xs/g-c3n4 nanocomposites by in-situ immobilizing znxcd1-xs onto the surface of g-c3n4 nanosheets in a hydrothermal environment, followed by a phosphorization process. we anticipate that g-c3n4 could act as substrates to develop a suitable heterogeneous interface to promote photo-induced interfacial charge transfer. the as-fabricated p-doped znxcd1-xs/g-c3n4 nanocomposites and the enhancement of photocatalytic activities were systematically characterized by xrd, fesem, tem, and electrochemical measurements. 2. experimental melamine (c6h6n6), zinc acetate dehydrates (zn(ac)2·2h2o), cadmium acetate dehydrates (cd(ac)2·2h2o) and thioacetamide (taa) were bought from sinopharm chemical reagent co., ltd. (shanghai, china). nah2po2·h2o was bought from macklin (shanghai, china). all reagents were of analytic purity, used without any purification. 2.1. preparation of bulk g-c3n4 melamine powder was put into an alumina crucible with a cover and heated to 550°c at a heating rate of 2.3°c/ min and maintained at this temperature for 4h. the resulted yellow product was collected and ground into powder for further use. 2.2. preparation of g-c3n4 nanosheets the as-prepared bulk g-c3n4 (1g) was mixed with 10 ml of h2so4 and stirred overnight at room temperature. then the mixture was slowly poured into 150 ml of deionized water and sonicated for exfoliation. the obtained suspension was centrifuged, washed thoroughly with deionized water to remove the residual acid, and finally dried at 80°c in air overnight. 2.3. preparation of zn0.5cd0.5s/g-c3n4 50 mg g-c3n4 nanosheets were dispersed completely into a mixed solvent consisting of 50 ml of deionized water by ultrasonication, then 1mmol zinc acetate dihydrate (zn(ac)2·2h2o), and 1mmol cadmium acetate dihydrate (cd(ac)2·2h2o) https://www.sciencedirect.com/topics/chemistry/solid-solution https://www.sciencedirect.com/topics/chemistry/hydrogen-production https://www.sciencedirect.com/topics/physics-and-astronomy/hydrothermal-synthesis https://www.sciencedirect.com/science/article/pii/s0169433218310031?via=ihub#b0155 https://www.sciencedirect.com/science/article/pii/s0169433218310031?via=ihub#b0160 proceedings of engineering and technology innovation, vol. 13, 2019, pp. 32-40 34 were dissolved into the above suspension, the mixture was stirred magnetically for 2 hours. subsequently, naoh aqueous solution was added dropwise followed by the addition of 2mmol thioacetamide(taa), then stirring for 4h at room temperature. then the mixture was transferred to a 100 ml polytetrafluoroethylene-lined stainless autoclave at 180 °c for 12 h under autogenous pressure in an electric oven. after cooling down to room temperature naturally, the product was collected, washed by water and ethanol for several times, and fully dried at 60 °c to obtain the final zn0.5cd0.5s/g-c3n4 nanocomposites, denoted as gs. the pure zn0.5cd0.5s was prepared with the same method in the absence of g-c3n4. 2.4. preparation of p-doped znxcd1-xs/g-c3n4 the p-doped zn0.5cd0.5s/g-c3n4 sample was synthesized using sodium hypophosphite (nah2po2·h2o) as the p precursor. typically, zn0.5cd0.5s/g-c3n4 and nah2po2·h2o were mixed together and finely grind with a mortar. then, the mixture was heated at 300 °c in an n2 atmosphere (a ramp rate of 2 °c/min). after cooling down to room temperature naturally, the product was collected, washed by water and ethanol several times, and fully dried at 60 °c overnight, and the samples were denoted as pgs. 2.5. catalysts characterization the crystal structures and phase states of the photocatalysts were analyzed by x-ray diffractometry using a bruker d8 advance diffractometer with cu kα radiation(λ= 0.15418 nm) at an operating voltage of 40kv and an operating current of 40ma. the morphologies of the samples were observed by field-emission scanning electron microscope (fesem, hitachi s-4800) with an acceleration voltage of 15.0kv and transmission electron microscope (tem, jeol jem-2100) at an acceleration voltage of 200kv. uv–vis diffused reflectance spectra were recorded on an evolution 220 uv-vis spectrophotometer (thermo fisher, america) from 200 to 800nm. 2.6. electrochemical measurements electrochemical impedance spectroscopy (eis) and photocurrent measurements were performed in a conventional three-electrode on a chi-760e electrochemical workstation. ag/agcl and platinum wire were used as the reference electrode and counter electrode, respectively. a 0.5 m na2so4 aqueous solution was used as the electrolyte. a 300w xe lamp equipped with a 420nm cut-off filter was employed as a light source. the working electrode was prepared as follow: 4 mg of the as-obtained sample and 10μl nafion solution (5 wt%) were dispersed in 1 ml water-ethanol solution with a volume ratio of 3:1 by sonicating for 1 h to form a homogeneous slurry. then 100μ l of the dispersion was loaded onto a 1cm×3cm ito-coated glass substrate with a coating area of 1cm 2 . 2.7. catalysts activity evaluations the photocatalytic activities of as-fabricated p-doped zn0.5cd0.5s/g-c3n4 nanocomposites were estimated by the degradation of rhodamine b (rhb) dye and hydrogen evolution from water splitting with the exposure of visible light irradiation. a 300 w xenon lamp equipped with a cut-off filter of 420 nm was employed as the light source for the photocatalytic reaction. during the photodegradation reaction, 20 mg of photocatalysts was totally dispersed into 50 ml aqueous solution of rhb dye with an initial concentration of 10 mg/l. prior to irradiation, the suspension was drastically stirred in the dark for 1 h to achieve adsorption-desorption equilibrium between the photocatalysts and rhb. at an interval of 10 min, 3 ml aliquots were collected, centrifuged, and then analyzed by recording variations at the wavelength of maximal absorption (λ= 554 nm) on a uv-vis spectrometer (uv-1201). as for hydrogen generation, 50 mg of the prepared zn0.5cd 0.5s, gs and pgs were used for measuring the hydrogen evolution from 200 ml 0.35 m na2s and 0.25 m na2so3 aqueous solution under visible light irradiation. for g-c3n4, 50mg sample was dispersed into 200 ml aqueous solution containing triethanolamine (10 vol%) as a sacrificial electron donor. the photocatalytic system for hydrogen evolution (plx-10a) was evacuated several times to remove the air inside the reaction system prior to irradiation. the evolved hydrogen was analyzed by on-line gas chromatography (gc-1690, jiedao, tcd, ar carrier). 35 proceedings of engineering and technology innovation, vol. 13, 2019, pp. 32-40 3. results and discussion the typical xrd patterns of g-c3n4 bulk, g-c3n4 nanosheets, zn0.5cd0.5s/g-c3n4, and p-doped zn0.5cd0.5s/g-c3n4 nanocomposites were illustrated in fig. 1(a) and (b). the g-c3n4-bulk exhibit two obvious characteristic diffraction peaks located at 13.1° and 27.4°, which could be ascribed to (100) and (002) diffraction plane of graphitic carbon nitrogen, separately representing the in-plane structural packing motif and inter-planar stacking peak of aromatic systems [28]. it is found that the (002) diffraction at around 27.4°relating to the characteristic interlayer stacking structure is shifted to 27.9°, indicating that the bulk g-c3n4 was exfoliated into nanosheet. the xrd patterns of zn0.5cd0.5s/g-c3n4 and p-doped zn0.5cd0.5s/g-c3n4 nanocomposites display similar characteristic diffraction peaks, it is noticeable that the main three diffraction peaks of p-doped zn0.5cd0.5s in the range of 24.5-30°slightly shift to smaller diffraction angles after p doping, implying the appearance of lattice expansion after p doping. (a) bulk g-c3n4 and g-c3n4 nanosheet (b) gs and pgs fig. 1 xrd patterns of the as-prepared samples the in-situ growth strategy for preparing the composite is schematically illustrated in fig.2. a previous electrophoresis experiment illustrated that g-c3n4 is negatively charged when it is dispersed in water because of the deprotonation of amine groups on the g-c3n4 surface [29]. when cd(ac)2 and zn(ac)2 are introduced into the suspension, zn 2+ and cd 2+ can be tightly adsorbed onto the g-c3n4 nanosheets through electrostatic interaction. upon the addition of naoh and taa, zn 2+ and cd 2+ react with s 2− to generate zn0.5cd0.5s nps. as a result, the zn0.5cd0.5s particles are well deposited on g-c3n4 nanosheets, and a zn0.5cd0.5s/g-c3n4 composite is obtained. the p-doped zn0.5cd0.5s/g-c3n4 nanocomposites can be obtained by the phosphorization of zn0.5cd0.5s/g-c3n4 using nah2po2 as p source. fig. 2 schematic of the in-situ fabrication of the p doped zn0.5cd0.5s/g-c3n4 composite the morphology of the as-prepared samples was investigated by sem and tem as shown in fig. 3. fig. 3(b) is a typical tem image of pure zn0.5cd0.5s nanoparticles with a size of about 20nm. fig. 3(c) shows the tem pattern of zn0.5cd0.5s/g-c3n4, which exhibits a 0d/2d nanostructure and g-c3n4 with length and width of 400nm and 800nm, respectively. fig. 3(a) and fig. 3(d) are sem and tem images of p-doped zn0.5cd0.5s/g-c3n4, respectively. a fluffy structure can be clearly observed. these features could provide an appropriate location for developing 0d/2d heterostructure, similar to zn0.5cd0.5s/g-c3n4, indicating that the doping of p didn’t change the morphology of the original zn0.5cd0.5s/g-c3n4. proceedings of engineering and technology innovation, vol. 13, 2019, pp. 32-40 36 (a) sem of p-doped zn0.5cd0.5s/g-c3n4 (b) tem of zn0.5cd0.5s (c) tem of zn0.5cd0.5s/g-c3n4 (d) tem of p-doped zn0.5cd0.5s/g-c3n4 fig. 3 morphology of the as-prepared samples the optical properties of as-obtained zn0.5cd0.5s, g-c3n4 nanosheets, zn0.5cd0.5s/g-c3n4 as well as p-doped zn0.5cd0.5s/g-c3n4 nanocomposites were characterized by uv-vis diffuse reflectance spectra (uv-vis drs) and displayed in fig. 4. it is clearly seen that pure zn0.5cd0.5s possess strong ability towards light absorption from the uv range to the visible light region up to 531 nm, while the absorption threshold of pure g-c3n4 nanosheets is only about 422 nm. the introduction of g-c3n4 nanosheets has an obvious effect on the optical property of zn0.5cd0.5s. remarkably, the obtained binary catalyst shows a small absorption blue-shift as a comparison to pure zn0.5cd0.5s. such result may be ascribed to the interaction between the zn0.5cd0.5s and g-c3n4 nanosheets and improved light scattering resulted from the smaller particle sizes of zn0.5cd0.5s in zn0.5cd0.5s/g-c3n4 nanocomposites [30]. after the doping of p, the absorption edge almost shifts to 680nm, the possible reason is that the heterojunction formed between zn0.5cd0.5s and g-c3n4 nanosheets may effectively decrease contact barrier and strengthen electronic coupling of the semiconductors, thus resulting in the enhanced visible light absorption, leading to the enhancement of photocatalytic activities. fig. 4 uv-vis diffuse reflectance spectra of the as-obtained samples the photocatalytic activity of the as-prepared samples was firstly evaluated by degradation of rhb pollutant under visible light irradiation. as shown in fig. 5,we can find that negligible photodegradation is detected in the blank experiment in the absence of photocatalyst, indicating the high stability of rhb under this experimental condition. therefore, we can rule out the effect of dye self-degradation on evaluating the photocatalytic ability. after visible light irradiation for 60min, only 15% and 50% of rhb can be removed by g-c3n4 nanosheets and zn0.5cd0.5s, respectively. however, under the same conditions, about 70% rhb can be degraded by zn0.5cd0.5s/g-c3n4, and nearly 97% of rhb can be removed by p-doped zn0.5cd0.5s/g-c3n4 37 proceedings of engineering and technology innovation, vol. 13, 2019, pp. 32-40 within 30min. the obvious enhancement of photocatalytic activity may be attributed to the heterojunction formed between zn0.5cd0.5s and g-c3n4 as well as the doping of p. fig. 5 the photocatalytic degradation of rhb by the as-prepared samples moreover, the photocatalytic activity for hydrogen generation with different samples from an aqueous solution containing 0.35 m na2s and 0.25 m na2so3 was evaluated under visible light irradiation. as can be seen from fig. 6, the bare g-c3n4 nss shows a very low photocatalytic activity towards water splitting and corresponding h2 evolution rate is only 17.51 μmol g -1 h -1 . the low photocatalytic activity may be ascribed to the faster recombination of photo-generated holes and electrons due to the presence of interior defects created during the high-temperature preparation route [24]. compared with bare g-c3n4 nss, benefiting from the narrow bandgap and appropriate band edges, the pure zn0.5cd0.5s nps showed excellent photocatalytic activity with an h2 production rate of 2174.36 μmol g -1 h -1 . interestingly, both zn0.5cd0.5s/g-c3n4 and p-doped zn0.5cd0.5s/g-c3n4 exhibit improved photocatalytic activity for hydrogen generation. their corresponding h2 production rates (fig. 6b) are determined to be 2660.82 μmol g -1 h -1 and 2880.83 μmol g -1 h -1 for zn0.5cd0.5s/g-c3n4 and p-doped zn0.5cd0.5s/g-c3n4, respectively. the significant enhancement of photocatalytic activities for photocatalytic water splitting could be ascribed to the synergetic effects of p doping and a close heterogeneous interface constructed between zn0.5cd0.5s nps and g-c3n4 nss, which induce the faster separation of photo-generated electrons and holes, thus improving the photocatalytic performances. (a) photocatalytic hydrogen evolution (b) hydrogen production rate over different catalysts fig. 6 photocatalytic activity for hydrogen evolution generally, phase structure, morphology, light absorption intensity, the separation efficiency of photo-generated carrier charge pairs, etc. play important roles in deciding the photocatalytic activities [31-35]. however, the similar phase structure and morphology, the decreased light absorption indicate that these factors are not responsible for the significant enhancemen t of the photocatalytic performances. fig. 7(a) shows the photocurrent response of the different samples with the exposure of discontinuous visible light. it can be clearly seen that pgs exhibits the highest photocurrent density among all the as-prepared samples, which implies that the photoinduced charge carriers can be separated efficiently after phosphorization. the efficient charge transfer was further confirmed by the electrochemical impedance spectra (eis) analysis as shown in fig. 7(b). the eis nyquist plots reveal that the prepared pgs possesses a smaller semicircle in comparison with those of its counterparts, suggesting that the phosphorization process can enhance the separation and transfer efficiency of the charge carrier pairs. proceedings of engineering and technology innovation, vol. 13, 2019, pp. 32-40 38 (a) transient photocurrent–time (i–t) curves (b) eis nyquist plots of different samples fig. 7 photoelectrochemical tests of the as-prepared samples 4. conclusions in summary, the novel p-doped zn0.5cd0.5s/g-c3n4 heterostructured photocatalysts have been synthesized by in-situ deposition of zn0.5cd0.5s nanoparticles on the surface of g-c3n4 nanosheets, followed by a phosphorization process. the as-prepared p-doped zn0.5cd0.5s/g-c3n4 composites in this work have been proved to be a better photocatalytic system than zn0.5cd0.5s/g-c3n4, and exhibit enhanced photocatalytic activity for rhb degradation and hydrogen evolution under visible light irradiation. the enhancement of photocatalytic activity, on the one hand, is mainly attributed to the unique property of g-c3n4 nanosheets, such as large surface areas and improved electron transportability. on the other hand, it also results from the synergetic effect of p doping and closes heterogeneous interface constructed between zn0.5cd0.5s nps and g-c3n4 nss, which induce the faster separation of photo-generated electrons and holes. this work may provide a new possibility to further improve the performances of g-c3n4-based photocatalysts, and facilitates their applications in solar energy utilization and conversion. acknowledgement this work was supported by the national natural science foundation of china [no. 51572126] and [no. 51872141]. the author would like to thank anonymous reviewers for their pertinent comments. conflicts of interest the authors declare no conflict of interest. references [1] a. fujishima and k. honda, “electrochemical photolysis of water at a semiconductor electrode,” nature, vol. 238, pp. 37-38, july 1972. [2] d. q. zeng, p. y. wu, w. j. ong, b. s. tang, m. d. wu, h. f. zheng, y. z. chen, and d. l. peng, “construction of network-like and flower-like 2h-mose2 nanostructures coupled with porous g-c3n4 for noble-metal-free photocatalytic h2 evolution under visible light,” applied catalysis b: environment, vol. 233, pp. 26-34, october 2018. [3] j. r. ran, m. jaroniec, and s. z. qiao, “cocatalysts in semiconductor-based photocatalytic co2 reduction: achievements, challenges, and opportunities,” advanced materials, vol. 30, pp. 1704649, february 2018. [4] w. chen, z. c. he, g. b. huang, c. l. wu, w. f. chen, and x. h. liu, “direct z-scheme 2d/2d mnin2s4/g-c3n4 architectures with highly efficient photocatalytic activities towards treatment of pharmaceutical wastewater and hydrogen evolution,” chemical engineering journal, vol. 359, pp. 244-253, march 2019. [5] x. z. chen, n. li, z. z. kong, w. j. ong, and x. j. zhao, “photocatalytic fixation of nitrogen to ammonia: state-of-the-art advancements and future prospects,” materials horizons, vol. 5, pp. 9-27, january 2018. [6] r. b. we, z. l. huang, g. h. gu, z. wang, l. x. zeng, y. b. chen, and z. q. liu, “dual-cocatalysts decorated rimous cds spheres advancing highly-efficient visible-light photocatalytic hydrogen production,” applied catalysis b: environment, vol. 231, pp. 101-107, september 2018. https://www.sciencedirect.com/science/article/pii/s0926337318302042#! https://www.sciencedirect.com/science/article/pii/s0926337318302042#! https://www.sciencedirect.com/science/article/pii/s0926337318302042#! https://www.sciencedirect.com/science/article/pii/s0926337318302042#! https://www.sciencedirect.com/science/article/pii/s0926337318302042#! https://www.sciencedirect.com/science/article/pii/s0926337318302042#! https://www.sciencedirect.com/science/article/pii/s0926337318302042#! http://apps.webofknowledge.com/full_record.do?product=ua&search_mode=generalsearch&qid=9&sid=7ftkuzwwxoa6z3hvskv&page=1&doc=5&cacheurlfromrightclick=no http://apps.webofknowledge.com/full_record.do?product=ua&search_mode=generalsearch&qid=9&sid=7ftkuzwwxoa6z3hvskv&page=1&doc=5&cacheurlfromrightclick=no 39 proceedings of engineering and technology innovation, vol. 13, 2019, pp. 32-40 [7] s. guo, z. deng, m. li, b. jiang, c. tian, q. pan, and h. fu, “phosphorus-doped carbon nitride tubes with a layered micro-nanostructure for enhanced visible-light photocatalytic hydrogen evolution,” angewandte chemie international edition, vol. 55 pp. 1830-1834, december 2015. [8] l. zhao, l. zhang, h. lin, q. nong, m. cui, y. wu, and y. he, “fabrication and characterization of hollow cdmoo4 coupled g-c3n4 heterojunction with enhanced photocatalytic activity,” journal of hazardous material, vol. 299, pp. 333-342, december 2015. [9] j. zhang, j. g. yu, m. jaroniec, and j. r. gong, “noble metal-free reduced graphene oxide-znxcd1–xs nanocomposite with enhanced solar photocatalytic h2-production performance, ” nano letter, vol. 12, pp. 4584-4589, august 2012. [10] q. j. xiang, b. cheng, and j. g. yu, “hierarchical porous cds nanosheet-assembled flowers with enhanced visible-light photocatalytic h2-production performance,” applied catalysis b: environment, vol. 138-139, pp. 299-303, july 2013. [11] a. kudo and y. miseki, “heterogeneous photocatalyst materials for water splitting,” chemical society reviews, vol. 38, pp. 253-278, november 2008. [12] w. chen, g. r. duan, t. y. liu, z. m. jia, x. h. liu, s. m. chen, and x. j. yang, “synthesis of homogeneous one-dimensional nixcd1-xs nanorods with enhanced visible-light response by ethanediamine-assisted decomposition of complex precursors,” journal of material science, vol. 50, pp. 3920-3928, june 2015. [13] m. liu, l. zhang, x. he, b. zhang, h. song, s. li, and w. you, “l-cystine-assisted hydrothermal synthesis of mn1-xcdxs solid solutions with hexagonal wurtzite structure for efficient photocatalytic hydrogen evolution under visible light irradiation,” journal of material chemistry a, vol. 2, pp. 4619-4626, january 2014. [14] s. xie, x. lu, t. zhai, j. gan, w. li, m. xu, m. yu, y. m. zhang, and y. tong, “controllable synthesis of znxcd1-xs@zno core@shell nanorods with enhanced photocatalytic activity,” langmuir, vol. 28, pp. 10558-10564, june 2012. [15] r. shi, h. f. ye, f. liang, z. wang, k. li, y. x. weng, z. s. lin, w. f. fu, c. m. che, and y. chen, “interstitial p-doped cds with long-lived photogenerated electrons for photocatalytic water splitting without sacrificial agents,” advanced materials, vol. 30, 1705941, december 2017. [16] z. wu, g. zhao, y. zhang, h. tian, and d. li, “enhanced photocurrent responses and antiphotocorrosion performance of cds hybrid derived from triple heterojunction,” the journal of physical chemistry c, vol. 116, pp. 12829-12835, may 2012. [17] n. n. hewa-kasakarage, p. z. el-khoury, a. n. tarnovsky, m. kirsanova, i. nemitz, a. nemchinov, and m. zamkov, “ultrafast carrier dynamics in type ii znse/cds/znse nanobarbells,” acs nano, vol. 4, pp. 1837-1844, march 2010. [18] y. tak, h. kim, d. lee, and k. yong, “type-ii cds nanoparticle-zno nanowire heterostructure arrays fabricated by a solution process: enhanced photocatalytic activity,” chemical communication, pp. 4585-4587, august 2008. [19] s. liu, n. zhang, z. r. tang, and y. j. xu, “synthesis of one-dimensional cds@tio2 core-shell nanocomposites photocatalyst for selective redox: the dual role of tio2 shell,” acs applied materials & interfaces, vol. 4, pp. 6378-6385, november 2012. [20] h. wang, l. zhang, z. chen, j. hu, s. li, z. wang, j. liu, and x. wang, “semiconductor heterojunction photocatalysts: design, construction, and photocatalytic performances,” chemical society reviews, vol. 43, pp. 5234-5244, may 2014. [21] s. j. moniz, s. a. shevlin, d. j. martin, z. x. guo, and j. tang, “visible-light driven heterojunction photocatalysts for water splitting: a critical review,” energy & environmental science, vol. 8, pp. 731-759, january 2015. [22] m. c. liu, d. w. jing, z. h. zhou, and l. j. guo, “twin-induced one-dimensional homojunctions yield high quantum efficiency for solar hydrogen generation,” nature communications, september 2013. [23] j. g. song, h. t. zhao, r. r. sun, x. y. li, and d. j. sun, “an efficient hydrogen evolution catalyst composed of palladium phosphorous sulfide (pdp~0.33s∼1.67) and twin nanocrystal zn0.5cd0.5s solid solution with both homo and hetero-junctions,” energy & environmental science, vol. 10, pp. 225-235, november 2016. [24] q. tay, p. kanhere, c. f. ng, s. chen, s. chakraborty, a. c. h. huan, t. c. sum, r. ahuja, and z. chen, “defect engineered g-c3n4 for efficient visible light photocatalytic hydrogen production,” chemistry of materials, vol. 27, pp. 4930-4933, july 2015. [25] q. lin, l. li, s. liang, m. liu, j. bi, and l. wu, “efficient synthesis of monolayer carbon nitride 2d nanosheets with tunable concentration and enhanced visible-light photocatalytic activities,” applied catalysis b: environment, vol. 163, pp. 135-142, february 2015. [26] x. x. wang, j. chen, x. j. guan, and l. j. guo, “enhanced efficiency and stability for visible light driven water splitting hydrogen production over cd0.5zn0.5s/g-c3n4 composite photocatalyst,” international journal of hydrogen energy, vol. 40, pp. 7546-7552, june 2015. proceedings of engineering and technology innovation, vol. 13, 2019, pp. 32-40 40 [27] h. liu, z. t. jin, and z. z. xu, “hybridization of cd0.2zn0.8s with g-c3n4 nanosheets: a visible-light-driven photocatalyst for h2 evolution from water and degradation of organic pollutants,” dalton transaction, vol.44, pp. 14368-14375, july 2015. [28] j. luo, x. zhou, l. ma, and x. xu, “enhancing visible-light photocatalytic activity of g-c3n4 by doping phosphorus and coupling with ceo2 for the degradation of methyl orange under visible light irradiation,” rsc advances, vol. 5, pp. 68728-68735, august 2015. [29] b. c. zhu, p. f. xia, w. k. ho, and j. g. yu, “isoelectric point and adsorption activity of porous g-c3n4,” applied surface science, vol. 344, pp. 188-195, july 2015. [30] j. zhang, y. hu, x. jiang, s. chen, s. meng, and x. fu, “design of a direct z-scheme photocatalyst: preparation and characterization of bi2o3/g-c3n4 with high visible light activity,” journal of hazardous materials, vol. 280, pp. 713-722, september 2014. [31] x. j. bai, l. wang, and y. f. zhu, “visible photocatalytic activity enhancement of znwo4 by graphene hybridization,” acs catalysis, vol. 2, pp. 2769-2778, november 2012. [32] n. zhang, s. q. liu, x. z. fu, and y. j. xu, “synthesis of m@tio2(m = au, pd, pt) core-shell nanocomposites with tunable photoreactivity,” the journal of physical chemistry c, vol. 115, pp. 9136-9145, april 2011. [33] h. x. xong, t. zhao, g. d. zhou, r. f. qian, t. feng, and j. h. pan, “revisiting structural and photocatalytic properties of g-c3n4/tio2: is surface modification of tio2 by calcination with urea an effective route to solar photocatalyst ?” catalysis today, december 2018. [34] r. f. qian, h. x. zong, j. schneider, g. d. zhou, t. zhao, y. l. li, j. yang, b. w. bahnemann, and j. h. pan, “charge carrier trapping, recombination and transfer during tio2 photocatalysis: an overview,” catalysis today, october 2018. [35] x. z. jiang, m. manawan, t. feng, r. f. qian, t. zhao, g. d. zhou, f. t. kong, q. wang, s. y. dai, and j. h. pan, “anatase and rutile in evonik aeroxide p25: heterojunctioned or individual nanoparticles?” vol. 300, pp. 12-17, february 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/). template encit2010 proceedings of engineering and technology innovation, vol. 13, 2019, pp. 41-45 simulation of a 0.6 v wideband cmos lna design using forward body bias jian-ming wu * , chih-yu huang, cheng-shong hong, you-ruei jhuang department of electronic engineering, national kaohsiung normal university, kaohsiung, taiwan received 15 may 2019; received in revised form 5 june 2019; accepted 1 july 2019 abstract a 0.6-v wideband low-noise amplifier (lna) design using a 0.18 μm cmos standard process is presented in this paper. an important goal for the design is to achieve wideband under a low supply voltage. the presented lna design is a three-stage amplifier, which is based on a common gate stage and two common source stages. the common gate stage provides a wideband input matching. the two common source stages provide a flat gain. a forward body bias technique is used to reduce the supply voltage of the wideband lna. the forward body bias, that is, the body-source voltage (vbs) is greater than 0 v, decreases the threshold voltage of the nmos transistor, achieving the low voltage lna. the supply voltage that is applied to the wideband lna is only 0.6 v and the power consumption (pdc) is 9.24 mw. simulation results indicate that the gain exceeds 10 db and the noise figure (nf) is below 6.5 db from 1.62 ghz to 7.02 ghz. the main advantage of the lna is it's wideband with a low supply voltage. keywords: low-noise amplifier (lna), forward body bias, low voltage, wideband 1. introduction the fifth generation (5g) wireless system will be the major mobile communication technology for information delivery in the near future. the 5g wireless system is a fast network, providing high data rates above 1g bit/s [1]. the 5g wireless system uses a massive multi-input multi-output (mimo) technique to achieve high data rates. high data rates require a wide bandwidth. the 5g wireless system uses a carrier aggregation technique to obtain broadband. a massive mimo technique is used to increase the data rates of the 5g wireless system. a massive mimo system consists of multiple antennae phased arrays with a beamforming technique. a massive mimo system using a beamforming technique scans multiple users at the same time to increase the throughputs of the 5g wireless system. a massive mimo technique provides a spatial multiplexing and a spatial diversity against multiple path attenuations and reduces noise interferences, which raise the data rates of the 5g wireless system. in general, a wireless system has a less continuous bandwidth. therefore, a carrier aggregation technique is used to aggregates discontinuous or continuous carrier components that have different frequency bands to achieve broadband, meeting the requirements of the 5g wireless system. for instance, a discontinuous carrier component has a 20-mhz bandwidth and then aggregating five discontinuous carrier components can obtain a bandwidth of 100 mhz. the characteristic of a carrier aggregation technique is to use discontinuous carrier components that have different frequency bands, implying that a carrier aggregation technique can use fragmented frequency bands to obtain a wide bandwidth. * corresponding author. e-mail address: jianmingwu@nknucc.nknu.edu.tw tel.: +886-7-7172930 ext. 7914; fax: +886-7-6051330 proceedings of engineering and technology innovation, vol. 13, 2019, pp. 41-45 42 broadband considerations require that an rf receiver of the 5g wireless system must have a wide bandwidth. an rf receiver in the 5g wireless system must also have low power consumption (pdc) to support long standby times of mobile devices. operating an rf component at a low supply voltage can substantially reduce the power consumed by 5g receivers. a low-noise amplifier (lna) is the first component in an rf receiver and is used to amplify an rf signal. an lna is the first stage in an rf receiver so its noise figure (nf) dominates the noise characteristics of the overall receiver. proposed approaches to design a wideband lna include 1) common gate amplifier [2-3], 2) resistive shunt-feedback technique [4], 3) distributed amplifier [5], and 4) balanced amplifier [6]. the common gate amplifier can be done with simplicity but has a poor frequency response. the resistive shunt-feedback technique can greatly increase bandwidth but is not efficient in power consumption. the distributed amplifier provides a good input and output impedance matching, but a large chip area is needed. the balanced amplifier can provide a high gain but is complicated to implement. methods have been reported to design an lna with a low supply voltage. they are summarized to include the: 1) forward body bias technique [7-8], 2) folded cascade technique [9], and 3) current reused technique [10]. the forward body bias technique can be simple to implement but has a leakage current in the body-source junction of a transistor. the folded cascade technique provides high linearity and good isolation but is done with complexity. the current reuse technique is good at the decrease in power consumption but requires a high supply voltage. the purpose of this work is to design a wideband lna using the tsmc 0.18 μm cmos standard process for 5g applications. the common gate technique is used to achieve wideband due to its simplicity. the forward body bias technique is used to reduce a supply voltage because of its simplicity. in this paper, section 2 presents the circuit design of the wideband lna. the simulated results are demonstrated in section 3. section 4 shows the conclusion of this paper. 2. experimental details fig. 1 schematically depicts the circuit of the 0.6-v wideband lna. the presented lna design uses a common gate technique and a forward body bias technique [2]. the lna is a three-stage amplifier. the first stage is a common gate stage, which consists of a transistor m1, a source inductance ls, and an inductance lc. the transistor m1 is the main active element of the first stage. since ls has a small reactance at a low frequency, the lna has a small input impedance at a low frequency. at a high frequency, the input impedance of the lna approximates to 1/gm1 where gm1 is the transconductance of the transistor m1 [2]. lc can block rf signals from the source of the transistor m2. the function of the first stage is to provide a wideband input matching. the second stage is a common source amplifier. the transistor m2 is the main amplified element of the second stage. lg and cg consist of a series resonator. the inductance ld is used to increase the gain. rg and r1 are the biasing resistances of the transistors m1 and m2. cb is a bypass capacitance. the third stage is also a common source amplifier, which consists of a transistor m3, an inductance lb, and a drain inductance lo. r1 is also the biasing resistances of the transistor m3. the transistor m3 is the main amplified element of the third stage. lb and the gate-source capacitance of the transistor m3 form a series resonator. lo at the drain of the transistor m3 is used to increase the gain. the second and third stage provides a flat gain. a forward body bias technique is used to reduce the supply voltage of the wideband lna. by referring to [11], the threshold voltage (vth) is expressed as 0 ( 2 2 ) (v) th t f bs f v v v     (1) proceedings of engineering and technology innovation, vol. 13, 2019, pp. 41-45 43 where vbs presents a body-source voltage, vt0 is a threshold voltage when the vbs is equaled to 0 v, γ denotes a fabricationprocess parameter, and f presents physical fermi potential. a forward body bias, that is, vbs is greater than 0 v, reduces the threshold voltage of the nmos transistor, achieving a low supply voltage of the lna. fig. 1 circuit schematic of 0.6-v wideband lna 3. layout and simulation the wideband lna circuit is designed using the tsmc 0.18 μm cmos foundry process. fig. 2 shows the chip layout of the lna rfic. the chip area is 1.2 mm × 1.2 mm. it is considered that mutual coupling between the components cause s signal interferences. on the periphery of each component, a guard ring is connected to the ground to prevent signals from interfering with each other. a short trace reduces the parasitic elements including resistance, inductance, and capacitance. a mitered trace has a small impedance variation, reducing signal reflection. fig. 2 chip layout of wideband lna the wideband lna rfic is now being fabricated. a supply voltage is 0.6 v and the current consumption is 15.4 ma. power consumption is calculated by supply voltage multiplying by current consumption, which equals to 9.24 mw. fig. 3 shows the simulation results of the gain. fig. 3 shows that the gain exceeds 10 db from 1.62 ghz to 7.02 ghz. fig. 4 plots the simulated noise figure. it can be seen that the noise figure is below 6.5 db from 1.62 ghz to 7.02 ghz. fig. 5 plots the simulated magnitude of the reflection coefficient seen looking into the rfin port. fig. 5 shows that the reflection coefficient vdd rg lg cg r1 ld lc ls cb lb lo vbb vbb vbb m2 m1 m3 vgg rfin rfout rfin gnd gnd vbb rfout vgg vbb vdd vbb gndgnd proceedings of engineering and technology innovation, vol. 13, 2019, pp. 41-45 44 magnitude is less than -5.36 db from 1.62 ghz to 7.02 ghz. fig. 6 plots the simulated magnitude of s12, corresponding isolation. fig. 6 reveals that the magnitude of s12 is below -25.94 db from 1.62 ghz to 7.02 ghz. table 1 summarizes the performance merits of this work and compares it with previous proposed wideband lna designs in [12-18]. the comparison shows that the presented wideband lna has the lowest supply voltage among the works listed in table 1. fig. 3 simulated gain of wideband lna fig. 4 simulated noise figure of wideband lna fig. 5 simulated magnitude of the reflection coefficient seen looking into the rfin port of wideband lna fig. 6 simulated magnitude of s12 of wideband lna table 1 performance summary for this work and comparison with prior works specification this work * [12] [13] [14] [15] [16] [17] * [18] * frequency (ghz) 1.62−7.02 1.3−12.1 2.3−9.2 2−7 0.4−10 3.4−11.4 3.1−10.6 2−10 maximum gain (db) 20.13 8.15 9.3 13 12.4 16 11.037 12.2 minimum nf (db) 4.65 2.5 4 4.1 4.4 3.1 1.811 3.7 s11 (db) < -5.36 < -17.5 < -9.9 < -17 < -10 < -8 < -1.390 < -13 s12 (db) < -25.94 < -25.8 < -43 < -40 n/a < -40 < -14.265 n/a supply voltage (v) 0.6 1.2 1.8 2.5 1.8 1.8 1.2 1.8 pdc (mw) 9.24 10.68 9 9.3 12 11.9 n/a 48 * simulated results. 4. conclusions a 0.6-v wideband lna with a forward body bias technique has been designed using the tsmc 0.18 µ m standard cmos process. the wideband lna rfic achieves that the gain exceeds 10 db and the noise figure is below 6.5 db from 1.62 ghz to 7.02 ghz when the supply voltage is only 0.6 v. the main advantage of the lna is it's wideband with a low supply voltage. therefore, the lna can be applied to 5g wireless systems, internet of things (iot), wireless sensor networks (wsn), and automobile electronics. 1.62 2.22 2.82 3.42 4.02 4.62 5.22 5.82 6.42 7.02 frequency (ghz) 10 12 14 16 18 20 22 g a in ( d b ) 1.62 2.22 2.82 3.42 4.02 4.62 5.22 5.82 6.42 7.02 frequency (ghz) 4 4.5 5 5.5 6 6.5 7 n f ( d b ) 1.62 2.22 2.82 3.42 4.02 4.62 5.22 5.82 6.42 7.02 frequency (ghz) -35 -30 -25 -20 -15 -10 -5 m a g n it u d e o f s 1 1 ( d b ) 1.62 2.22 2.82 3.42 4.02 4.62 5.22 5.82 6.42 7.02 frequency (ghz) -80 -70 -60 -50 -40 -30 -20 m a g n it u d e o f s 1 2 ( d b ) proceedings of engineering and technology innovation, vol. 13, 2019, pp. 41-45 45 acknowledgment this work was supported by the ministry of science and technology, taiwan, under grant 107-2221-e-017-004. the authors wish to thank the taiwan semiconductor research institute, taiwan, for providing the tsmc 0.18 μm cmos foundry service. conflicts of interest the authors declare no conflict of interest. references [1] m. elkholy, s. shakib, j. dunworth, v. aparin, and k. entesari, “a wideband variable gain lna with high oip3 for 5g using 40-nm bulk cmos,” ieee microwave and wireless components letters, vol. 28, no. 1, pp. 64-66, january 2018. [2] a. dehqan, e. kargaran, k. mafinezhad, and h. nabovati, “an ultra-low-voltage ultra-low power cmos uwb lna using forward body biasing,” ieee international midwest symposium on circuits and systems, 2012, pp. 266-269. [3] j. y. lee, h. k. park, h. j. chang, and t. y. yun, “low-power uwb lna with common-gate and current-reuse techniques,” iet microwaves, antennas & propagation, vol. 6, no. 7, pp. 793-799, may 2012. [4] y. c. lin, m. l. yeh, w. r. liou, and c. c. chang, “a tunable 3.9-7.1 ghz cmos lna for ultra-wideband wireless communication,” international conference on communications, circuits, and systems, 2007, pp. 1226-1229. [5] s. c. shin, c. s. lin, m. d. tsai, k. y. lin, and h. wang, “a low-voltage and variable-gain distributed amplifier for 3.110.6 ghz uwb systems,” ieee microwave and wireless components letters, vol. 16, no. 4, pp. 179-181, april 2006. [6] d. m. pozar, microwave engineering, 2nd ed. new york: john willey & sons, 1998. [7] j. m. wu, y. t. lin, and y. s. tsai, “a sub-mw cmos lna for wsn applications,” asia pacific microwave conference, 2012, pp. 899-901. [8] d. wu, r. huang, w. wong, and y. wang, “a 0.4-v low noise amplifier using forward body bias technology for 5 ghz application,” ieee microwave and wireless components letters, vol. 17, no. 7, pp. 543-545, july 2007. [9] h. h. hsieh, j. h. wang, and l. h. lu, “gain-enhancement techniques for cmos folded cascade lnas at low-voltage operations,” ieee transactions microwave theory and techniques, vol. 56, no. 8, pp. 1807-1816, august 2008. [10] r. m. weng, c. y. liu, and p. c. lin, “a low-power full-band low-noise amplifier for ultra-wideband receivers,” ieee transactions microwave theory and techniques, vol. 58, no. 8, pp. 2077-2083, august 2010. [11] a. s. sedra and k. c. smith, microelectronic circuits, 5th ed. new york: oxford university press, 2004. [12] h. y. yang, y. s. lin, and c. c. chen, “2.5 db nf 3.1-10.6 ghz cmos uwb lna with small group-delay variation,” electronics letters, vol. 44, no. 8, pp. 528-529, april 2008. [13] a. bevilacqua and a. m. niknejad, “an ultrawideband cmos low noise amplifier for 3.1-10.6 ghz wireless receivers,” ieee journal of solid-state circuits, vol. 39, no. 12, pp. 2259-2268, december 2004. [14] t. k. k. tsang, k. y. lin, and m. n. el-gamal, “design techniques of cmos uwb amplifier for multistandard communications,” ieee transactions on circuits and systems ii: express briefs, vol. 55, pp. 214-218, march 2008. [15] k. h. chen, j. h. lu, b. j. chen, and s. i. liu, “an ultra-wide-band 0.4-10 ghz lna in 0.18-μm cmos,” ieee transactions on circuits and systems ii: express briefs, vol. 54, pp. 217-221, march 2007. [16] y. j. lin, s. h. hsu, j. d. jin, and c. y. chan, “a 3.1-10.6 ghz ultra-wideband cmos low noise amplifier with currentreused technique,” ieee microwave and wireless components letters, vol. 17, pp. 232-234, march 2007. [17] aditi and m. bansal, “a high linearity and low noise shunt resistive feedback uwb lna,” conference on information and communication technology, november 2017, pp. 1-5. [18] s. mubashir and v. singh, “a 2-10 ghz ultra-wideband common-gate low noise amplifier using body bias technique in 0.18 μm cmos,” devices for integrated circuit, pp. 541-545, march 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/). microsoft word 5-peti#6551 42-49.docx proceedings of engineering and technology innovation, vol. 17, 2021, pp. 42-49 instability risk and beam profile variation in optical ring resonator due to thin gradient index lens khalid ali khan 1,* , suleyman malikmyradovich nokerov 2 1 department of electrical and computer engineering, mettu university, mettu, ethiopia 2 oguz han engineering and technology university of turkmenistan, city ashgabat, turkmenistan received 15 july 2020; received in revised form 09 october 2020; accepted 15 december 2020 doi: https://doi.org/10.46604/peti.2021.6551 abstract in this study, a simple ring resonator model in presence of thin gradient index (grin) lens is investigated to characterize the optical beam maginification quality beyond its traditional modalities. this model allow us to vary and control the limit of resonator stability more significantly.it consist of two folding arms and each arm can be realized by its cavity components. insertation of thin grin lens ( thickness < 9.3mm) in ring resonator, mainly in between first folding range gives the magnified output beams and meets the beam expander feature for the laser. variation of grin lens thickness (l) is an emphatic and influencing parameter than its refractive index (n) to disturb the resonator stability. resonator stability in tangential (t) plane is relatively more sensitive than sagittal (s) plane. vigorous magnification in optical beam size at the end of output range in a cavity is the noticeable consequences because of grin lens. keywords: optical ring resonator, grin lens, resonator stability, abcd matrix, beam propagation microstrip 1. introduction actually, optical resonator or cavity with different structure along with their gain is globally known as the integral part of laser and light interferometer. wide application of resonator such as in quantum electrodynamic [1-2], enhancement of circulating power in resonator [3], in high-sensitivity laser spectroscopy and sensing [4-7], optical switching [8], photonic biosensor [9], in communication system [10] is gaining research popularity in physicists and engineers omnidirectionally. john e. heebner et al. [11] has also developed the ring resonator theory for the measurement of optical transmission characteristic.but, historically, fabry-perot interferometer was the first optical resonator [12-14]. other than it, boyd et al. [15-16] modified and modelled the fabry-perot resonator with spherical mirrors first. eventhough, several decades earlier, explanation and mathematical expression for gaussian laser beam propagation has been given by fox [17], pierce [18], goubau [19], kogelnik [20] and by others. however, complete derivation for laser resonator to achieve the stability is available in [21]. the light rays that bounce back and forth in between the resonator shows the focusing mechanism. this paper presents a modified version of optical ring resonator supported by thin gradient index (grin). the presence of grin lens improves output performance of the resonator including laser pumping, beam magnification and focusing mechanism under its proper parametric adjustment. table 1 lists formulated abcd matrix for used elements [22-23]. the application of grin lens draws the novel anxiety into the possibilities and limitations of resonator stability and on its beam * corresponding author. e-mail address: sts.khalid@gmail.com tel.: +251-954591921 proceedings of engineering and technology innovation, vol. 17, 2021, pp. 42-49 43 magnification property [24]. abcd matrix has been also used to analyze the cavity. resonator stability is decided if the condition is satisfied: 1 1 2 a d+ − ≤ ≤ + (1) consequently, light ray trajectory is close and bounded to optical axis after multiple number of round-trips. in ring resonator, round trip matrix can be calculated by multiplying the elemental abcd matrix from reference element to first element of the model then after from last element to an element next to the reference. table 1 elements and its abcd matrix element abcd matrix descriptions empty space �� � �� � � � �1 �0 1 l=empty length mt=abcd matrix in t plane ms=abcd matrix in s plane spherical mirror �� � 1 0� � ∙��� � 1�, �� � 1 0� �∙��� � 1� r=radius of curvature of mirror �=beamincidence angle brewster crystal plate �� � 1 �∙�������0 1 �, �� � 1 �∙�������0 1 � l=thickness of crystal n=refractive index of crystal grin lens �� � �� � �� �� �� cos "#$�$% & ' 1�$�$% sin "# $�$% & ' ��$�$%sin "#$�$% & ' cos "#$�$% & ' *+ ++ +, here, refractive index(n) as a function of radial coordinate r is as $ � $% � �� $�-� [22] .� / 0� � -� l=thicknes of grin lens, n0=refractive index of grin at center n=n(r)=radial var iationin refractive index 2. modelling of ring resonator the major components (elements) and schematic model of simple ring resonator has been shown in fig. 1. it is made up of two spherical mirrors (m1, m2), one brewster crystal plate (cr1) and with a single piece of radial gradient index (grin) lens (f1). here, thin cr1 and thin grin lens (f1) has been used to maintain the resonator stability.in fact, total length {1300(l0)+75(d1)+1.262·1.3(cr1)+50(d2)+0.1·1.5(f1)+35(l1)=1460.80mm}is known as the resonator or cavity length and found to be 1460.80mm. mathematically, this length of cavity is the sum of indivisual elemental length. by assumption, beam characteristics are evaulauted and tested at 1550 nanometer (nm) wavelength by software simulation process. the main objective of the cr1 in the resonator is to compensate the astigmatism, whereas, addition of grin gives optical beam magnification that is justified by its performance. the technical specifications of all components are listed in table 2. fig. 1 the schematic model of ring resonator with grin lens table 2 the technical specifications of components s. no elements specifications 1 empty space “l0” (output radius) l0=1300mm 2 spherical mirror “m1” and m2 radius of curvature (r)=150nm, α=10° 3 empty space “d1”, “d2” and “l1” d1=74mm, d2=50mm, and l1=35mm 4 brewster crystal plate (cr1) crystal width (l)=1mm, refractive index (n)=1.3 5 grin lens (f1) thickness=0.1mm, refractive index (n)=1.5 proceedings of engineering and technology innovation, vol. 17, 2021, pp. 42-49 44 proper inclination without any deviation is given to the spherical mirrors m1 and m2 so that it is matched with beam incidence angle (α) to ensure maximum capturing of beam. therefore, the complete round trip matrix (m0) for the proposed cavity model (by considering l0 as reference point) will be as: 0 0 2 1 1 2 1 1 1m =l m l f d cr d m⋅ ⋅ ⋅ ⋅ ⋅ ⋅ ⋅ (2) where l0, l1, m1, m2, f1, d1, d2, cr1 represent the elemental abcd matrix that has been calculated as per its specific values and recorded in table 3. table 3 elemental abcd matrix as per mentioned specification elements t-plane matrix s-plane matrix l0 �1 13000 1 �1 13000 1 m2 � 1 0�0.014 1 � 1 0�0.013 1 l1 �1 350 1 �1 350 1 f1 � 1 0.067�0.001 1 � 1 0.067�0.001 1 d2 �1 500 1 �1 500 1 cr1 �1 0.5740 1 �1 0.970 1 d1 �1 740 1 �1 740 1 m1 � 1 0�0.014 1 � 1 0�0.013 1 hence, by using eq. (1) and recorded value of individual elemental abcd matrix as table 3, the complete round trip matrix in terms of abcd matrix for proposed resonator in tangential (t) and sagittal (s) plane are respectively (l0 has been taken as reference). 16.77 1323.23 0.201 15.83 t m − = − (3) 15.29 1246.96 0.189 15.36 s m − = − (4) where, mt = round trip matrix in t plane, ms = round trip matrix in s plane 3. simulated results and analysis 3.1. effect of gradient index (grin) lens on stability normal stability condition (eq. (1)) is considered to decide the stability region in tangential (t) as well as in sagittal plane (s). plotted stability graph for respective grin lens-parameters and cr1 plate parameters confine the restricted boundary condition under which back forth optical pulse oscillation takes place without outward bouncing in a cavity. simulated graph as in fig. 2 illustrates that resonator stability is very sensitive toward grin lens thickness. a small increment in thickness variation may lose the system stability. for proposed model, thickness should be less than 0.38mm in order to sustain stability. fig. 3 (a) and fig. 3(b) are the stability variation plot with respect to refractive index(n) variation of grin lens. this can be easily understood that refractive index variation of grin lens does not alter the state of stability in both plane observation proceedings of engineering and technology innovation, vol. 17, 2021, pp. 42-49 45 fig. 2 stability dependency on grin length (l) (a) stability dependency on grin refractive index in t plane (b) stability dependency on grin refractive index in s plane fig. 3 stability variation graph with variation of grin refractive index(n) (a) stability under cr1 length (stability continues, if cr1 length<9.3mm) in presence of grin (b) stability under cr1 length (stability continues,if cr1 length<13.43mm) in absence of grin fig. 4 impact of brewster crystal thickness (l) on stability variation (t, s). because slight and unpredictable stability variation is found in t plane (from 0.746 to 0.741) as well as in s plane (from -0.030 to -0.035) with respect to large variation in refractive index starting from 1 to 2.5. however, cr1 thickness play a vital role in stability governance. here, it is more interesting to note that cr1 shows the satisfactory stability performance for both planes, if and only if, 1 9.32 ( )thickness of cr mm in presence of grin≤ (5) 1 13.43 ( )thickness of cr mm in absence of grin≤ (6) otherwise, resonator may lose stability. therefore, present of grin lens in modelled resonator brings the resonator close to instability region. fig. 4(a) and fig. 4(b) illustrates the cr1 thickness limitation to support stability of resonator. furthermore, as far as the refractive index (n) of cr1 is concerned, its variation does not change the state of stability more significantly, either in the presence or the absence of grin. almost similar plot as drawn in fig. 5(a) and fig. 5(b) reveals that proceedings of engineering and technology innovation, vol. 17, 2021, pp. 42-49 46 a slight change in stability is observed if refractive index (n) changes and goes only up to 2.5, afterwards, stability becomes constant in both plane (t, s). it is noteworthy to clear that stability region in between -0.1 to +0.5 in presence of grin lens or in between -0.30 to +0.30 in absence grin shows that change in refractive index (n) of cr1 does not affect seriously. (a) stability dependency on cr1 refractive index (n) in presence of grin lens (b) stability dependency on cr1 refractive index (n) in absence of grin lens fig. 5 impact of refractive index (n) of brewster crystal on stability variation 3.2. gaussian beam propagation and beam magnification in previous section, it was proved that stability of the cavity in both plane (t and s) does not depend upon refractive index (n) of grin lens(f1) and refractive index (n) of brewster crystal plate (cr1). the abcd matrix for respective, spherical mirrors, brewster plate and grin lens are independent to its position inside the cavity except empty space. however. this abcd matrix is useful to calculate the beam parameter throughout the cavity. in most cases, it pre-assumed that laser beam propagation has ideal gaussian intensity profile, even-though it is not true in real time. however, we will assume theoretical properties of gaussian beam with quality factor, m 2 =1 (it is known as “m squared” factor). the beam radius variation in entire cavity in presence and in absence grin lens are shown in fig. 6(a) and fig. 6(b) respectively. (a) beam radius throughout the cavity in presence of grin lens (b) beam radius throughout the cavity in absence of grin lens fig. 6 beam radius vartiaon throughout the cavity here, it is more interesting to comment that there is abrupt contraction in beam radius in between both spherical mirrors (m1, m2) in both cases. larger beam radius is found at the interface of spherical mirrors and beam exit end (l0=0) of the cavity. additionally, sudden beam increment can be also seen at the brewster crystal surface in tangential plane. because when beam passes through it, refracted beam size in tangential plane increased due to increase in refractive index. therefore, it is clear from fig. 6(a) and fig. 6(b) that inclusion of grin lens in between brewster crystal plate and m2 (it is also called first folding range) in ring resonator provides the magnified beam to the respective output range (l0). referring to fig. 6(a), it is found that minimum beam radius across l0 is 243 micrometer (um) that is found at 577.77 mm location. hereafter, a small increment in beam radius exists up to 722.22 mm distance and finally it becomes 295.43 um. proceedings of engineering and technology innovation, vol. 17, 2021, pp. 42-49 47 nevertheless, in absence of grin lens, the beam variation in between 577.77 mm to 722.22 mm length across l0 is almost zero and found constant. however, this segment (577.77 mm to 722.22 mm) shows a drastic change in wavefront curvature radius in s plane. thus, recorded data shows that (in s plane in presence of grin) 0 577.77 2396.46 at l mm wave front radius mm = = − (7) 0 722.22 1537.95 at l mm wave front radius mm = = + (8) therefore, fig. 7(a) and fig. 7(b) reveals that in s plane, abrupt and large transition in wavefront radius between two point predicts slow and small radius variation at that point and vice versa.it must be also recorded that the contraction of beam radius yields negative and smaller wavefront curvature radius, whereas beam radius expansion generates positive and larger wavefront curvature radius. (a) wavefront radius variation throughout the cavity in presence of grin lens (b) wavefront radius variation throughout the cavity in absence of grin lens fig. 7 wavefront radius variation throughout the cavity alternatively, sudden and unexpected wavefront radius transition can be also observed at l0=1300 mm (in s plane) at which wavefront radius of +860.81 mm becomes to -83.54 mm. relatively, cavity without grin lens that produces smaller beam spot across l0, generates larger wavefront curvature radius corresponding to it. for example, beam wavefront radius (in s plane) 0 577.77 2794 87 at l mm wave front radius . mm = = + (9) 0 722.22 2794 87 at l mm wave front radius . mm = = + (10) similarly, at the interface of m1 (l0=1300 mm) another abrupt change in wavefront radius is found from +952.51 mm to -82.77 mm. 4. conclusions in this paper, we have investigated the impact of grin lens on resonator stability and beam characteristics in optical ring resonator. designed resonator has the cavity length and inter-mode-beat frequency of 1460.80 mm and 205.226 mhz respectively. inter-mode-beat frequency reflects the pulse repetition rate of 205.226 mhz that estimates the theoretical round-trip distance of 1461.80 mm inside the cavity. a concise comparison with the commonly used ring resonator with same cavity length is also performed. referring to fig. 3 and fig. 4, it can be concluded that placing of thin grin lens in first folding range (in between cr1 to m2) achieves the beam size magnification across output range (l0) without altering the stability of ring resonator. larger beam proceedings of engineering and technology innovation, vol. 17, 2021, pp. 42-49 48 magnification (around 1.7 times) is recorded at the end of l0 in t plane. thickness variation of grin lens disturb the resonator stability more vigorously in t plane if it is not designed properly. nevertheless, its application as beam expander may correspond easy portability with other optical devices rather than keplerian and galilean beam expander (keplerian and galilean beam expander are unportable for laser beam) conflicts of interest the authors declare no conflict of interest. references [1] y. jiang, y. mei, y. zou, y. zuo, and s. du, “intracavity cold atomic ensemble with high optical depth,” review of scientific instruments, vol. 90, no. 1, pp. 013105–1-013105–7, january 2019. [2] j. ye, c. j. hood, t. lynn, h. mabuchi, d. w. vernooy, and h. j. kimble, “quantum manipulation and measurement of single atoms in optical cavity qed,” ieee transactions on instrumentation and measurement, vol. 48, no. 2, pp. 608-612, may 1999. [3] a. mills, s. zhdanovich, m. x. na, f. boschini, e. razzoli, m. michiardi, et al., “cavity-enhanced high harmonic generation for extreme ultraviolet time and angle-resolved photoemission spectroscopy,” review of scientific instruments, vol. 90, no. 8, pp. 083001–1-083001–9, august 2019. [4] b. liu, h. zhou, l. liu, x. wang, m. shan, p. jin, and z. zhong, “an optical fiber fabry-perot microphone based on corrugated silver diaphragm,” ieee transactions on instrumentation and measurement, vol. 67, no. 8, pp. 1994-2000, august 2018. [5] c. zhu, y. zhuang, y. chen, and j. huang, “a liquid-level sensor based on a hollow coaxial cable fabry-perot resonator with micrometer resolution,” ieee transactions on instrumentation and measurement, vol. 67, no. 12, pp. 2892-2897, december 2018. [6] a. cygan, p. wcisło, s. wojtewicz, g. kowzan, m. zaborowski, d. charczun, et al., “high-accuracy and wide dynamic range frequency-based dispersion spectroscopy in an optical cavity,” optics express, vol. 27, no. 15, pp. 21810-21821, july 2019. [7] g. gagliardi and h.-p. loock, cavity-enhanced spectroscopy and sensing. springer series in optical sciences 179, heidelberg: springer-verlag berlin heidelberg, 2014. [8] v. van, t. a. ibrahim, k. ritter, p. p. absil, f. g. johnson, r. grover, et al., “all optical nonlinear switching in gaas-algaas microring resonators,” ieee photonics technology letters., vol. 14, no. 1, pp. 74-76, january 2002. [9] r. w. boyd and j. e. heebner, “sensitive disk-resonator photonic biosensor,” applied optics, vol. 40, no. 31, pp. 5742-5747, december 2001. [10] b. e. little, s. t. chu, h. a. haus, j. foresi, and j.-p. laine, “micro-ring resonator channel dropping filters,” journal of lightwave technology, vol. 15, no. 6, pp. 998-1005, april 1997. [11] j. e. heebner, v. wong, a. schweinsberg, r. w. boyd, and d. j. jackson, “optical transmission characteristics of fiber ring resonators,” ieee journal of quantum electronics, vol. 40, no. 6, pp. 726-730, june 2004. [12] r. h. dicke, molecular amplification and generation systems and methods, u. s. patent, 2 851 652, september 9, 1958. [13] a. m. prokhorov, “molecular amplifier and generator for submillimeter waves,” journal of experimental and theoretical physics, vol. 7, no. 6, pp.1658-1659, june 1958. [14] a. l. schawlow and c. h. townes, “infrared and optical masers,” physical review., vol. 112, no. 6, pp.1940-1949, december 1958. [15] g. d. boyd and j. p. gordon, “confocal multimode resonator for milli meter through optical wavelength masers,” the bell system technical journal, vol. 40, no. 2, pp. 489-508, march 1961. [16] g. d. boyd and h. kogelnik, “generalized confocal resonator theory,” the bell system technical journal., vol. 41, no. 4, pp. 1347-1369, july 1962. [17] a. g. fox and t. li,“resonant modes in a maser interferometer,” the bell system technical journal., vol. 40, no. 2, pp. 453-488, march 1961. [18] j. r. pierce, “modes in sequences of lenses,” proceeding of the national academy of sciences of the united states of america, , national academy of sciences press, november 1961, pp. 1808-1813. [19] g. goubau, “optical relations for coherent wave beams,” electromagnetic theory and antennas, new york: mac millan, june 1962, pp. 907-918. proceedings of engineering and technology innovation, vol. 17, 2021, pp. 42-49 49 [20] h. kogelnik, “imaging of optical modes-resonators with internal lenses,” the bell system technical journal, vol. 44, no. 3, pp. 455-494, march 1965. [21] h. kogelnik and t. li, “laser beams and resonators,” applied optics, vol. 5, no. 10, pp.1550-1557, october 1966. [22] “rezonator application,” www.rezonator.orion-project.org, june 5, 2020. [23] s. n. b. tushar, s. d. nath, j. akhtar, and m. i. reja, “modelling and analysis of zfolded solid state laser cavity with two curved mirrors,” international journal of microwave and optical technology, vol.13, no.3, pp. 244-253, may 2018. [24] k. a. khan and s. m. nokerov, “analysis of gaussian beam profile and its amplification in ring resonator with thin gradient-index lens,” modern means of communications, materials of the 25 th international scientific and technical conference, belarusian state academy of communications press, october 2020, pp. 42-44. 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 proceedings of engineering and technology innovation, vol. 14, 2020, pp. 29 44 applications of membrane hydrocyclone recovery of silicon carbide powder yen-cheng tsai, chen-yang hsu, rome-ming wu * department of chemical and materials engineering, tamkang university, new taipei city, taiwan received 07 april 2019; received in revised form 07 may 2019; accepted 23 june 2019 doi: https://doi.org/10.46604/peti.2020.4205 abstract this work investigated the possibility of using membrane hydrocyclone as a separator to recover the silicon carbide powder. the weight percentage of silicon carbide powder is 62.9%. classifying these powders by a 2.5 cm-diameter hydrocyclone that equips membrane tube in the center position, overflow and underflow are obtained. the results of the primary separation show that the overflow and underflow are about half of the proportion of solid, but silicon carbide powder content in the overflow is higher. therefore, it is necessary to consider the recovery of silicon carbide powder contained in the overflow. keywords: separation, recovery, membrane, hydrocyclone, classification 1. introduction today there is much-classifying equipment [1], and one of the most popular classifying equipment is hydrocyclone [2]. in the past, people made efforts to design new hydrocyclones with different structures in changing fluid characteristics and were committed to developing hydrocyclones with high separation efficiency. for example, assisted by electricity, [3] presented an electrical hydrocyclone with a central electrode rod. it's cut size decreases when the diameter and length of the electrode rod increase. [4] conceived of blocking the boundary layer flow, developing water injected hydrocyclone, so that small particles are no longer exist in the underflow. [5] put forward a filtering hydrocyclone which conical portion is replaced by the filter material, which enhances the classification efficiency. [6] put forward a hydrocyclone with an inlet, one underflow outlet, and two overflow outlets. the main functions of the two overflow outlets are to prevent large particles flow into the overflow by short-circuiting and to accelerate the prior separation of the particles in inner and outer overflow. the velocity profiles of hydrocyclones were first measured by [7], he found that the tangential velocity increased to a maximum from hydrocyclone wall towards the center, and then decreased rapidly. some studies agree on the same qualitative flow behavior [8-9]. owing to a strong air core in the center, tangential and axial velocities become difficult to measure. however, with the development of science and technology, researchers focus on simulations by computational fluid dynamics (cfd) techniques [10]. the performance of a hydrocyclone can be displayed in the following indexes: (1) flow split of a hydrocyclone is defined as the ratio of the volumetric flow rate of underflow to overflow. (2) separation efficiency es: u u s f f g c e g c  (1) * corresponding author. e-mail address: romeman@mail.tku.edu.tw tel.: +886-2-26215656-3286; fax: +886-2-26209887 proceedings of engineering and technology innovation, vol. 14, 2020, pp. 29 44 40 where gu and gf are mass flow rates of underflow and feed, respectively. cu and cf are weight concentrations of solids in underflow and feed, respectively. (3) reduced separation efficiency s e 1 s uf s uf e r e r    (2) where ruf is the ratio of volumetric flow rate of underflow to feed. (4) cut size d50: the certain particle size that 50% particles flow out through underflow, and 50% particles flow out through overflow. (5) partition curve: the probability that the particles with a certain size could be separated into underflow from the feed of hydrocyclone. ( ) ( ) ( ) u u p p pf f g c d p d g c d  (3) recently, [11] used cfd studies to simulate the separation of microorganisms and mammalian cells with hydrocyclones. some works on the simulation of hydrocyclones using the incompressible navier-stokes equations, supplemented by a suitable turbulence model, have proven to be appropriate for modeling the flow in a hydrocyclone [12-13]. 2. experimental details 2.1. materials silicon carbide powder ranging from 2~23 μm diameter was used in this study. the particle size distribution was shown in fig. 1. the average particle diameter is 16.5 μm, and the specific gravity of particles is 3.2. the adopted composite ceramic tubular membrane is made by tami. the support layer of the tubular membrane is made of titania, the filter layer is made of titania and zirconia, with a pore size of 0.8 μm. the tubular membrane is with diameter 10 mm, length 600 mm, single channel, and a surface area of 0.008 m 2 . laser particle size distribution analyzer, model no. horrible la-30 was used to measure particle size distribution from 0.1 μm~600 μm. by the light scattering theory, the light source with 405 nm of helium-neon led was used to measure the sample particle size distribution on overflow and underflow. fig. 1 psd of silicon carbide powder 2.2. experimental setups and operation the geometry of the hydrocyclone is displayed in fig. 2. the diameters of feed, overflow, underflow, and hydrocyclone are 5, 20, 15, and 25 mm, respectively. the length of the cylindrical part is 40 mm, while the length of the cone part is 110 mm, making an overall 2.5 o cone angle. the experimental apparatus in this study is shown in fig. 3. first, make sure the 0 0.2 0.4 0.6 0.8 1 2 20 f particle size (𝜇m) proceedings of engineering and technology innovation, vol. 14, 2020, pp. 29 44 41 bypass valve (v3) is fully opened. then fully open the overflow valve (v4) and underflow valve (v2), adjust the valve of feeding end (v1) and observe the feeding gauge pressure (p1) until the pressure is reached the one required. record inlet pressure, the pressure at the overflow (p2) and the pressure at underflow (p3). measure and sample the dilute liquid from overflow and the concentrated liquid stream from underflow, and record the amount of clear liquid filtrate at the outlet end of the outside-in filtration. particles distribution, concentration, and separation efficiency were obtained. 2.3. simulation the simulation of the high-turbulent flow in hydrocyclone requires the basic equations of fluid dynamics combined with an adequate turbulence model. the flow pattern in hydrocyclone was modeled by newtonian water flow and by reynolds stress turbulence model (rsm). the reynolds-averaged navier–stokes equations are described as follows: ( ) 0i i v t x          (4) 2 ( ) ( ) [ ( )] ( ) 3 j ji i i j ij i j j i j j i j j v vvp v v v v v t x x x x x x x                              (5) the exact transport equations for the transport of the reynolds stresses, ji vv  , may be written as follows: ( ) ( ) [ ( )] [ ] ( ) ( ) ( ) 2 i j j i j i j i j i j i jk k kj ik k k k k k ji i i j i j j ik j ik v v v v v v v v v v v p v v v v t x x x x x vv v v v v g v g v p x x x                                                           j k k v x x     (6) where the left-hand side is local time derivative and convection terms, the right-hand side are turbulent diffusion, molecular diffusion, stress production, buoyancy production, pressure strain, and dissipation terms, respectively. a velocity inlet boundary condition was applied at the inlet: . @u const inlet pipe (7) the overflow and underflow used pressure outlet boundary conditions. the outlet fluid is moving under absolute pressure 1 atm therefore the gauge pressure at the overflow and underflow are zero. the boundary conditions are: 0 @p underflow (8) 0 @p overflow (9) the computational fluid dynamics program fluent 6.2 (fluent inc., usa) solved the governing equations, eqs. (4-7), together with the associated boundary conditions eqs. (7-9). simulations were carried out for about 50,000 incremental steps where in general a preset value of convergence criteria 1 × 10 -4 was achieved. fluent predicts the trajectory of a discrete phase particle by integrating the force balance on the particle. this force balance can be written as (for x direction): ( ) ( ) p x p pd p du g f u u dt        (10) 2 re18 24 d d p p c f d    (11) where fd(u–up) is the drag force per unit particle mass. proceedings of engineering and technology innovation, vol. 14, 2020, pp. 29 44 42 where u is the fluid phase velocity, up is the particle velocity,  is the molecular viscosity of the fluid,  is the fluid density, p is the density of the particle, and dp is the particle diameter. re is the relative reynolds number, which is defined as re p pd u u    (12) fig. 2 geometry of membrane hydrocyclone fig. 3 creative design methodology 3. result the meaning of the classification efficiency curve is the rate of a particle of a certain size flowing from the feeding inlet of the hydrocyclone and leaving from the underflow of the hydrocyclone. figure 4 is the velocity distribution under different operating pressures. fig. 5 is the classification efficiency curve of the separators under the same feeding conditions. fig. 4 velocity distribution for 0.5 bar, 1.0 bar, 1.5 bar, and 2.0 bar the separation efficiency is enhanced by a series of hydrocyclone separator in order to recover silicon carbide of higher purity. the separation system does not have to completely replace the membrane hydrocyclone with a conventional separator, but rather consider its application location. finally, the membrane hydrocyclone in order to enhance the filtrate production to show its characteristics, need to explore the parameters of the experiment, such as raising the inlet pressure, change the membrane tube, adjust the concentration of the liquid and temperature. although the overall separation efficiency is slightly less than tube without membrane, the single membrane hydrocyclone has a good impact on separation efficiency. proceedings of engineering and technology innovation, vol. 14, 2020, pp. 29 44 43 a dual feed hydrocyclones is proposed for separating an aqueous slurry of particles into a top stream which contains the light particles and a bottom stream which contains large particles [18]. a silicon containing material can be processed for production of a silicon-rich composition and can be used to form a wafer-saw cutting fluid [19]. the invention of hydrocyclone system for recovering silicon and silicon-containing compounds from slurry that is generated during wafer cutting or sawing operations in the photovoltaic (pv) and microelectronic (me) industries. the system can recover at least 60% of the silicon contained in a silicon-cutting slurry. a study on recycling silicon and silicon carbide powders from wiresaw slurry containing silicon carbide, silicon kerfs, polyethylene glycol (peg), and iron fragments was conducted [20]. after the first stage treatment of silicon carbide powders on hydrocyclones, 85 wt% silicon carbide powders can be collected at underflow. in this study, the weight percentage of original silicon carbide powder is 62.9%. after separation, it can concentrate to 83.4 wt%. in addition, it can get 38.2 l/hr-m2 clear peg filtrate. fig. 5 plots of separation efficiency versus particle size on different operations (blue curve: equips membrane tube; red curve: tube without membrane) a dual feed hydrocyclones is proposed for separating an aqueous slurry of particles into a top stream which contains the light particles and a bottom stream which contains large particles [18]. silicon-containing material can be processed for the production of a silicon-rich composition and can be used to form a wafer-saw cutting fluid [19]. the invention of hydrocyclone system for recovering silicon and silicon-containing compounds from the slurry that is generated during wafer cutting or sawing operations in the photovoltaic (pv) and microelectronic (me) industries. the system can recover at least 60% of the silicon contained in a silicon-cutting slurry. a study on recycling silicon and silicon carbide powders from wiresaw slurry containing silicon carbide, silicon kerfs, polyethylene glycol (peg), and iron fragments was conducted [20]. after the first stage treatment of silicon carbide powders on hydrocyclones, 85 wt% silicon carbide powders can be collected at underflow. in this study, the weight percentage of original silicon carbide powder is 62.9%. after separation, it can concentrate on 83.4 wt%. in addition, it can get 38.2 l/hr-m 2 clear peg filtrate. 4. conclusions in this study, silicon carbide particles were used to implement hydrocyclone experiments. the air core area which is replaced by ceramic tubular membrane was discussed, and the effects of separation phenomena of different hydrocyclones were explored. by inserting a tubular membrane, the hydrocyclone becomes a new-type hydrocyclone with three outlets. they are respectively the dilute liquid of overflow, the concentrated liquid of underflow, and the clear filtrate. it is as well equipped with both functions of classification and filtration. comparing with conventional hydrocyclone, it has equivalent classification function, but the overflow concentration is relatively dilute, the underflow concentration is relatively concentrated, and clear filtrate can be obtained at the same time. conflicts of interest the authors declare no conflict of interest. 0 0.2 0.4 0.6 0.8 1 2 20 η particle size (𝜇m) proceedings of engineering and technology innovation, vol. 14, 2020, pp. 29 44 44 acknowledgment the authors would like to thank the national science council of the republic of china, taiwan, for financially supporting this research under contract no. nsc 97-2221-e-032-030-my3. references [1] j. margraf and j. werther, “continuous classification in a centrifugal fluidized bed apparatus,” journal of the taiwan institute of chemical engineers, vol. 40, pp. 669-681, 2009. [2] k. nageswararao, “a critical analysis of the fish hook effect in hydrocyclone classifiers,” chemical engineering journal, vol. 80, pp. 251-256, 2000. [3] h. yoshida, s. taniguchi, and k. fukui, “effect of apex cone on particle classification performance of cyclone separator, journal of the taiwan institute of chemical engineers, vol. 35, pp. 41-46, 2004. [4] d. d. patil and t. c. rao, “classification evaluation of water injected hydrocyclone,” minerals engineering, vol. 12, pp. 1527-1532, 1999. [5] l. g. m. vieira, e. a. barbosa, j. j. r. damasceno, and m. a. s. barrozo, “performance analysis and design of filtering hydrocyclones,” brazilian journal of chemical engineering, vol. 22, pp. 143-152, 2005. [6] a. mainza, m. s. powell, and b. knopjes, “differential classification of dense material in a three-product cyclone,” minerals engineering, vol. 17, pp. 573-579, 2004. [7] d. f. kelsall, “a study of the motion of solid particles in a hydraulic cyclone,” trans instn chem engrs., vol. 30, pp. 87-108, 1952. [8] j. bergstrom and h. vomhoff, “experimental hydrocyclone flow field studies,” separation and purification technology, vol. 53, pp. 8-20, 2007. [9] j. bergstrom, h. vomhoff, and d. soderberg, “tangential velocity measurements in a conical hydrocyclone operated with a fibre suspension,” minerals engineering, vol. 20, pp. 407-413, 2007. [10] c. y. hsu and r. m. wu, “hot zone in a hydrocyclone for particles escape from overflow,” drying technology, vol. 26, pp. 1011-1017, 2008. [11] r. a. medronho, j. schuetze, and w. d. deckwer, “numerical simulation of hydrocyclones for cell separation,” latin american applied research, vol. 35, pp. 1-8, 2005. [12] b. wang, k. w. chu, and a. b. yu, “numerical study of particle-fluid flow in a hydrocyclone,” ind eng chem res., vol. 46, pp. 4695-4705, 2007. [13] b. wang and a. b. yu, “numerical study of the gas-liquid-solid flow in hdrocyclones with different configuration of vortex finder,” chemical engineering journal, vol. 135, pp. 33-42, 2008. [14] l. y. chu, w. m. chen, and x. z. lee, “enhancement of hydrocyclone performance by controlling the inside turbulence structure,” chemical engineering science, vol. 57, pp. 207-212, 2002. [15] b. li, m. huang, t. fu, l. pan, w. yao, and l. guo, “microfiltration process by inorganic membranes for clarification of tongbi liquor,” molecules, vol. 17, pp. 1319-1334, 2012. [16] m. kim, j. h. kim, a. s. syed, y. m. kim, k. k. choe, and c. y. kim, “application of centrifugal partition chromatography for bioactivity-guided purification of antioxidant-response-element-inducing constituents from atractylodis rhizoma alba,” molecules, vol. 23, pp. 2274-2283, 2018. [17] r. m. wu, “hydrocyclone separator,” us patent, us 8182684 b1, may 2012. [18] f. g. miller, “dual feed hydrocyclone and method of separating aqueous slurry,” us patent, us 4652363, march 1987. [19] j. a. fallavollita, “methods and apparatus for recovery of silicon and silicon carbide from the spent wafer-sawing slurry,” us patent, us 2011/0059002 a1, march 2011. [20] s. a. sergiienko, b. v. pogorelov, and v. b. daniliuk, “silicon and silicon carbide powders recycling technology from wire-saw cutting waste in slicing process of silicon ingots,” separation and purification technology, vol. 133, pp. 16-21, 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/). microsoft word 4-peti#6082 32-41.docx proceedings of engineering and technology innovation, vol. 17, 2021, pp. 32-41 mixed noise removal by processing of patches rithu james 1,* , harsha appukuttan 1 , liza annie joseph 2 1 department of electronics and communication engineering; 2 department of applied electronics and instrumentation engineering, rajagiri school of engineering and technology, india received 20 july 2020; received in revised form 09 october 2020; accepted 07 december 2020 doi: https://doi.org/10.46604/peti.2021.6082 abstract sonar images are degraded by mixed noise which has an adverse impact on detection and classification of underwater objects. existing denoising methods of sonar images remove either additive noise or multiplicative noise. in this study, the mixed noise in sonar images, the additive gaussian noise and the multiplicative speckle effect are handled by the data adaptive methods. a patch based denoising is applied in two phases to remove the additive gaussian and multiplicative speckle noises. in the first phase, the adaptive processing of local patches is used to remove the additive gaussian noise by exploiting the sonar image local sparsity. the pca and svd methods are used for denoising the noisy image patches and blocks of patches. in the second phase, the weighted maximum likelihood denoising of the nonlocal patches reduces the speckle effect by exploiting the non-local similarity in a probability distribution. experiments on side scan sonar images are conducted and the results show the importance of removing both the additive and multiplicative components from the sonar images. keywords: speckle, patches, blind denoising, blocks 1. introduction imaging sonars [1] is a point source that emanates acoustic plane wave of frequency in the range of 100 khz to 2 mhz. sonar images have rare edges as it is produced by the coherent nature of the scattering phenomenon and is influencedby a multiplicative speckle [2] component. the degradation model in [3] describes the multiple noise affected sonar images and the fully developed speckle amplitude has a multiplicative rayleigh distribution [4]. the multiple noises in sonar images are the additive gaussian noise and the speckle noise. due to the distinct characteristics of mixed noise, it is not easy to handle both the degradation process simultaneously. the methods developed for one noise cannot effectively suppress the other noise effect. the design of classical filters is for the additive noise removal only. in some of the speckle suppression methods for a sonar image, the multiplicative noise model is converted into an additive one by considering the logarithm of the image, which is called the homomorphic approach and it assumes the noise is gaussian. in the proposed method, a framework for the mixed noise removal in sonar images was done. the methodology assumes that the sonar images are degraded by multiple noises which hurt detection and classification performance. wavelet transform uses a fixed basis for the representation of sonar images. in this work, both locally and globally adaptive basis were used for the sonar image sparse representation. in the second phase patch based on the mixed denoising algorithm, the additive white gaussian noise removal is by the sparse approximation of local image patches in the first phase and despeckling is done on non-local patches in the second phase. * corresponding author. e-mail address: rithujames@gmail.com tel.: +919387211328 proceedings of engineering and technology innovation, vol. 17, 2021, pp. 32-41 33 2. various filtering techniques wiener filter is a classical filter designed for additive noise removal. some of the speckle suppression methods take a homomorphic approach which converts the multiplicative noise into additive noise. but such a transformation reduces the contrast in the image and introduces a bias [5]. adaptive filters are usually employed to reduce speckle noise in homogeneous areas of the sonar images while preserving texture and high frequency information in heterogeneous areas. available adaptive classical speckle filters are the lee, the kuan, the frost filters, the iterative speckle reducing anisotropic diffusion (srad) filter, and the multi-scale wavelet decomposition based on filters. a deep learning approach to directly estimate the speckle noise in the logarithmic domain according to the convolutional neural network (cnn) is established in [6-7]. in [8] the restoration of images was corrupted by impulse noise via the distribution-transformed network (dtn), which utilizes the convolutional neural network to learn pixel-distribution features from noisy images. however, to the author’s knowledge, very few publications on the removal of the multiple noises in sonar are available. the total variation image intensity is used for gaussian and poisson noise removal by than et al. [9]. a mixed noise removal method utilizing direction weighted median filter and block matching 3d filtering method is used by takuro et al [10]. 3. methodology first given a side scan sonar image, the unknown standard deviation �� of the image is estimated using the noise level estimation algorithm. the estimation of the noise level in the image and the non-blind denoising are the two processes of the blind denoising in sonar images. for a non-blind denoising algorithm, the noise level ��is assumed to be a known parameter. nevertheless, for the blind denoising algorithm, the noise level �� need to be estimated together with the denoising process. here a blind denoising method is used. fig. 1 depicts the methodology adopted. fig. 1 methodology first the methodology adopted is to estimate the noise variance of the given sonar image, then patch based processing is done for both the additive gaussian noise removal and the multiplicative speckle removal. the adaptive processing of local patches was used to eliminate the additive gaussian noise. this uses principal component analysis (pca) and singular value decomposition (svd) methods by processing the noisy image in a single patch and blocks of patches. the multiplicative speckle noise is then removed by processing of non-local patches. this uses the weighted maximum likelihood denoising of the non-local patches for reducing the speckle effect. simulations on sonar images were conducted and the results show the sonar image estimate the noise variance of the image denoising by processing of local patches non local patch processing for despeckling denoised image proceedings of engineering and technology innovation, vol. 17, 2021, pp. 32-41 34 effectiveness of mixed noise removal in sonar images. the different methods were compared by using different without-reference image performance evaluation criteria. 3.1. noise level estimation for noise variance estimation, the sparse nature of sonar image is taken into consideration. due to the natural sonar image redundancy, only low-dimensional subspace is spanned by it. the image is made up of several patches. if the data of patches��� � ∈ � × spans a subspace whose dimension is smaller than � × �, then such patches are called low-rank patches. in the proposed algorithm, low rank patches are selected. this avoids high frequency components in the noisy image. by doing this, a texture strength metric based on the local image gradient matrix and its statistical properties is used. the principal component analysis [11] is used to estimate the noise level of the additive gaussian noise from the selected patches. an input noisy image is decomposed into several patches in a raster scan form in the patch-based approaches [12]. the patch with the smallest standard deviation will have the least change of intensity among all the decomposed patches. the patches’ data model is: 1, 2, 3, , , = i i i my z n i= + … (1) where m represents the number of patches, �� denotes the �� patch which is noiseless with size � × �, and each patch is defined by its center pixel. �� is the observed patch represented as a vector. the observed patch is corrupted by noise vector �� assumed to be of zero-mean and variance�� �. the minimum variance direction is obtained as in [13]. this is the eigenvector corresponding to the minimum eigen value of the covariance matrix defined by: 1 1 m t y i i i y y m = ∑ = ∑ (2) where for the noisy patch y, ∑� represents its covariance matrix. 2 ( ) ( ) min y min z n ∑ = ∑ +λ λ σ (3) for the patch �� , ∑� is the covariance matrix obtained after removing the additive noise, and for the matrix∑, ���� (∑) is the minimum eigen value. in the method, the minimum eigen value of the covariance matrix ���� (∑) is taken as zero. the noise variance is estimated from the subspace which is spanned by the eigenvectors with zero eigen values of the covariance matrix ∑�. as a result it is assumed that the gaussian noise has the same power in all direction and eigen values are the same. 2 ( ) n min y = ∑σ λ (4) the non blind denoising is performed with the variance obtained for the image. 3.2. denoising by processing of local patches the patch based on the estimation method described above is used for getting the additive noise variance �� � for blind denoising. the noisy image is decomposed into patches using principal components [14] coefficients or svd coefficients. proceedings of engineering and technology innovation, vol. 17, 2021, pp. 32-41 35 these coefficients help to represent the noisy image using fewer values and the clean coefficients are estimated using the lmmse estimator. the procedure is to estimate the noise variance of the principal component coefficients, �� � for each patch using the maximum likelihood estimator. 2 22 1 0, [ ] i nn im max y −= ∑ σσ (5) the coefficients are denoised using lmmse estimator. 2 2 2 ( )l i i i n z y= × +σ σ (6) where �� is clean coefficient or the denoised image coefficient obtained after the removal of the additive gaussian noise. first, divide the image into non overlapping patches [15] of size � × �. next patches are transformed to one dimensional training vector. a matrix of training vector s of size � � × � is made by stacking these training vectors as columns. m is the number of patches. the noise free image is generated by denoising separately all the patches of the image. the eigenvectors of (�� � )(��) are taken as the pc basis functions. the principal components of the corresponding training vectors are represented by each column in the matrix. by projecting the training vectors in � onto the pc basis functions, the pc coefficients �( , !) for = 1, ⋯ , � � and ! = 1, ⋯ , � is got. the noisy image is represented by a fewer values using pc coefficients. for svd based on the denoising obtain the singular values of (�� � )(��). svd coefficients are calculated by projecting the training vectors in s onto corresponding basis function. the compression is obtained as svd, converging the image information into fewer coefficients. the requirement of complex calculations is less and the memory needed is also reduced. relevant data resides with the higher singular values. the left singular vectors are taken as the basis function in the process. to get back the denoised image, the entire decomposition process is reversed. the patches are arranged properly after the denoised training vectors were transformed back to patches for image reconstruction.in the approach of blocks of patch processing, denoising is performed in blocks [16]. a block matching technique is used to find the training samples of the local patch. this is by selecting a group of similar patches in the neighborhood by exploiting the images’ self-similarity. the entire s matrix is divided into blocks of the fixed size. the selection of the block size is critical. block sizes that are too small results in the complexity of computations. on the other hand, block sizes that are too large results in dissimilar patches getting grouped together and high calculation errors. the block size selection should be such that keeps a tradeoff between the number of similar patches and the computation speed. find the pc and the singular value coefficients after the block formation. then denoise the image at the block level and then reconstruct the final image. 3.3. despeckling by processing of non local patches in the second phase of denoising, the sonar image is assumed to be degraded by the rayleigh distributed speckle noise model. s s s z u f= (7) proceedings of engineering and technology innovation, vol. 17, 2021, pp. 32-41 36 where �% is the observed noisy value at site s and &% its underlying noise-free value. the denoising process is a weighted maximum likelihood (wmle) estimation problem [17] with the estimate &' % defined by: max logˆ ( , ) ( ) s s t t s u arg pu w s t z u= ∑ (8) where t is a pixel index and ((), *) is a data-driven weight [18] depending on the similarity between pixels with index s and t. due to the distribution characteristics of the speckle effected sonar images, the non-local patch based on the filter is very efficient. the pixel values far apart can be averaged together according to the weight values or neighboring patch similarity. the new measure of similarity between pixels will better reflect the image content. for the non-additive noise model, the weight ((), *). 1/ ( , ) ( ) h s t w s t p u u v ∆ ∆ =≜ (9) where &∆, and &∆are the sub-image extracts from the image u in the respective windows ∆% and ∆� , and ℎ > 0 is a scalar parameter. the pixel amplitudes 1% are modeled as independent and identically distributed according to the following rayleigh distribution: 2 2 12ˆ ˆ ˆ( ) ( ) ( ) l l s s s sl s s lal p a r a exp l r r − γ = − (10) where �2 is the given reflectivity image and l the equivalent number of looks and the noise-free amplitude image 13 is the square root of the reflectivity image �2 . the following estimation is from the first order optimality condition, 2 ( , )ˆ ( , ) t t s t w s t a r w s t ∑ = ∑ (11) it must be held to maximize the wmle according to weights derivation for multiplicative noise: 2 1 , , , , , , 2 2 , , , ˆ ˆ( , ) , l s k t k s k t k s k t k s k t k a a p a a r r a a a −   −      (12) with the final weights as: , , , , 2 1 ( , ) s k t k k t k s k a al w s t exp log a ah −    = −∑ +          (13) which can be used for the processing. the size of the number of significant weights to control the amount of filtering is related to h parameter in the nl means algorithm. the h parameter counter balances the invalidity of the patch independence assumption. for the processing side scan sonar image l is the mean to standard deviation square and h is the α quantile of the centered similarity criterion with α = 0.88 set in the experiments. proceedings of engineering and technology innovation, vol. 17, 2021, pp. 32-41 37 4. performance evaluation the images were taken from www.edgetech.com. the proposed method is assessed using the visual quality of the image for the detection of targets and by the quantitative measures. the different assessment parameters used are given below 4.1. equivalent number of looks (enl) in the sonar image intensity data, the speckle variance is proportional to the mean intensity squared. to quantify the quality of despeckled sonar images, the metric equivalent number of looks (enl) is commonly used. with a smaller spread of values due to speckle, the larger will be enl. the calculation of enl is: 2ˆ ˆ ( )x x enl = µ σ (14) where 78' and �8' are mean and standard deviation respectively of a homogeneous area, 9' in the despeckled image. the performance of despeckling technique is better if the enl value is high. 4.2. speckle suppression index (ssi) let x and 9' be original and the speckle reduced sonar images respectively. the speckle suppression index (ssi) is denoted as the ratio of the ssi coefficient of variance (ratio of standard deviation to mean). ˆ( ) ( ) ˆ( ) ( ) var x mean x ssl mean x var x = (15) smaller values of ssi, if more speckle suppression. 4.3. correlation coefficient (cc) the correlation coefficient (cc) for reflectivity image x and the denoised image 9' is calculated as follows: ˆ ˆ, ˆ ˆ[( )( )] x x x x x x e x x− − = µ µ µ σ σ (16) where 78 and 78' are mean values, �8 and �8' are standard deviations of reflectivity and denoised sonar images respectively. it can be found that the higher cc the better. 4.4. speckle suppression and mean preservation index (smpi) therefore, smpi is used for simultaneous estimation of speckle suppression and mean preservation capabilities of despeckling technique. it can be found that the lower smpi the better. ˆ( ) ˆ( ) var x smpi q var x = (17) where ˆ( ) ( )q k mean x mean x= + − (18) proceedings of engineering and technology innovation, vol. 17, 2021, pp. 32-41 38 ˆ( ( ) ( )) ( ) k max mean x min x mean x = − (19) 4.5. coefficient of variation the coefficient of variation (cv) is the ratio of the standard deviation to the mean. the higher the coefficient of variation, the greater the level of dispersion around the mean [19]. it is generally expressed as a percentage. the lower the value of the coefficient of variation, the more precise the estimation is cv= standard deviation/mean (20) 5. results and discussion in this research, a patch based blind denoising procedure is followed to estimate the noise level in the sonar images which results in an improved denoising performance. the patch based denoising is applied to additive gaussian and multiplicative speckle noises simultaneously. additive noise removal helps to smoothen the image while speckle removal helps to get a better image with details preserved. the method enforces smoothness while preserving edges. additive gaussian noise is removed using the pca and svd basedlocal patch processing method and then the nonlocal patch processing is done on the same image to remove the speckle noise. after a number of simulations, for the additive noise level estimation, a patch of size 8x8 was used. the local patch based processing in the method uses a patch size of 8x8. in the block based processing, 8 adjacent patches are combined to form a block. for the nonlocal patch processing, a search window of size 24x24 and a similarity window of size 8x8 are used. the denoising capability of such mixed noise removal will help in avoiding false detection of targets, but there can be undetected ones. for the nonlocal patch processing which is used for the multiplicative noise removal, the window width is varied and the despeckling results are also compared. the visual quality performance comparison is shown in fig. 2. it shows that the proposed method denoises without distorting useful image information as well as the important image edges. the additive denoising is approached using the sparse approximation of local image patches which is a characteristic of sonar images. for the multiplicative noise removal non-homomorphic approach was followed, so that the unnecessary introduction of the bias in the image was avoided. (a) real image (b) wiener filter (c) wavelet (d) srad (e) svd 1 st phase (f) svd 2 nd phase (g) pca 1 st phase 1 (h) pca 2 nd phase fig. 2 demonstrating the visual quality difference in different methods applied on sss image proceedings of engineering and technology innovation, vol. 17, 2021, pp. 32-41 39 fig. 2 shows that the proposed method denoise without distorting useful image information as well as the important image edges and table 1 depicts the numerical comparison and the graphical comparison in fig. 3. table 1 quantitative comparison of the denoising technique with some of the existing methods for a real side scan sonar image of enl 5.8422 metric wiener srad wavelet svd 1 st phase pca 1 st phase svd 2 st phase pca 2 st phase enl 6.9486 6.1805 5.5116 7.3797 7.3781 7.3800 7.3845 ssi 0.9171 0.9724 1.0297 0.8899 0.8900 0.8899 0.8896 cc 0.9646 0.9898 0.9933 0.9467 0.9467 0.9466 0.9465 smpi 0.2023 0.9552 3.0623 0.1093 0.1102 0.1143 0.1139 cov 0.3794 0.4022 0.4260 0.3681 0.3682 0.3681 0.3680 fig. 3 comparison by metric values (on vertical axis) to grade the performance of the proposed method the discarding of the lower coefficients results in the elimination of some noise associated with sensors along with the advantage of dimensionality reduction. as svd eliminates noise components associated with both the sensors and the pings, it is more effective in denoising than pca. the method is having the advantage of adaptively determining the local basis functions from the local image patches unlike the fixed ones for the entire image as in wavelet based technique. these basis functions are then being applied adaptively to local image patches. the metric value comparisons for the varying window width are depicted in the table 2 and the graphical comparison in fig. 4. the higher despeckling capability at the lower window width is at the cost of increased computational complexity. the improvement in the enl is a clear indication of the speckle reduction in the second stage. the smpi values are of small metric values to be clearly displayed on the selected scale of the graphical representation. fig. 4 the metric values for varying window width table 2 quantitative comparison of denoising techniques metric values for varying window size for a real side scan sonar image of enl 5.8422 window 2 3 4 5 6 7 8 enl 9.1561 9.1584 8.7726 8.257 7.9164 7.7095 7.5502 ssi 0.79893 0.79883 0.8162 0.8413 0.85921 0.87066 0.8798 cc 0.8989 0.89473 0.90658 0.92205 0.93147 0.93715 0.94143 smpi 0.033575 0.00942 0.016907 0.046955 0.012042 0.025944 0.039381 cov 0.33048 0.33044 0.33763 0.34801 0.35542 0.36015 0.36393 0 2 4 6 8 enl ssi cc smpi cov wiener srad wavelet svd 1st phase pca 1st phase svd 2st phase pca 2st phase 0 2 4 6 8 10 enl ssi cc smpi cov 2x2 3x3 4x4 5x5 6x6 7x7 8x8 proceedings of engineering and technology innovation, vol. 17, 2021, pp. 32-41 40 6. conclusions the traditional filtering method on sonar images can only remove one kind of noise and the details of the image are blurred in various degrees. mixed noise needs to be removed from the sonar images for further applications such as object detection. in this research, the sonar image local sparsity and its nonlocal similarity in a probability distribution are simultaneously exploited in removing the mixed additive gaussian and multiplicative speckle noise respectively. experimental results have demonstrated its better performance compared with other algorithms that remove only one type of noise. the performance can further be enhanced by patch size adaptation. creating blocks of similar patches rather than adjacent patches can give good improvement on a statistical basis in the block adaptive method. the despeckling effects can be improved if the adaptation of the pca and svd are for the selected similar patches or having them applied in a non-local means filtered data. the proposed method in the research deals with only locally adaptive basis, but globally adaptive basis can also be analyzed as future research. the future research also includes the extensions on the potential of such denoised sonar images to be used for better sonar image segmentation and registration. conflicts of interest the authors declare no conflict of interest. references [1] p. blondel, the handbook of sidescan sonar, 1st ed. uk: springer praxis publishing, 2009. [2] j. w. goodman, “some fundamental properties of speckle,” journal of optical society of america, vol. 66, no. 11, pp. 1145-1150, 1976. [3] a. k. jain, fundamentals of digital processing, usa: prentice-hall, inc, 1989. [4] f. maussang, j. chanussot, a. htet, and m. amate, “mean-standard deviation representation of sonar images for echo detection: application to sas images,” ieee journal of oceanic engineering, vol. 32, no. 4, october 2007. [5] a. achim, a. bezerianos, and p. tsakalides, “novel bayesian multiscale method for speckle removal in medical ultrasound images,” ieee transacton medical imaging, vol. 20, no. 8, pp. 772-783, august 2001. [6] r. w. liu, y. lu, f. chen, and l. xie, “learning a deep convolutional network for speckle noise reduction in underwater sonar images,” proceedings of the 2019 11th international conference on machine learning and computing, february 2019, pp. 445-450. [7] u. anithaa, s. malarkkanb, g. d. anbarasi jebaselvia, and r. narmadhaa, “sonar image segmentation and quality assessment using prominent image processing techniques,” applied acoustics, vol. 148, pp. 300-307, may 2019. [8] g. li, f. zhang, and q. liu, “distribution-transformed network for impulse noise removal,” journal of shanghai jiaotong university (science), june 2020. [9] n. h. t. dang, s. d. dvoenko, and v. s. dinh, “a mixed noise removal method based on total variation,” informatica, vol. 40, no. 2, pp. 159-167, july 2016. [10] t. yamaguchi, a. suzuki, and m. ikehara, “detail preserving mixed noise removal by dwm filter and bm3d,” ieice transactions on fundamentals of electronics, communications and computer sciences, vol. ei00-a, no. 11, pp. 2451-2457, november 2017. [11] x. liu, m. tanaka, and m. okutomi, “single-image noise level estimation for blind denoising,” ieee transactions on image processing,” vol. 22, no. 12, pp. 5226-5237, december 2013. [12] s. pyatykh, j. hesser, and l. zheng “image noise level estimation by principal component analysis,” ieee transactions on image processing, vol. 22, no. 2, pp. 687-699, february 2013. [13] c. m. bishop, pattern recognition and machine learning, new york: springer, 2006. [14] v. p. s. naidu and j. r. raol, “pixel-level image fusion using wavelets and principal component analysis,” defense science journal, vol. 58, no. 3, pp. 338-352, march 2008. [15] r. james and m.h. supriya, “sonar image denoising using adaptive processing of local patches,” 2015 international symposium on ocean electronics (sympol), ieee press, november 2015, pp. 1-7. [16] y. he, t. gan, w. chen, and h. wang, “adaptive denoising by singular value decomposition,” ieee signal processing letters, vol. 18, no. 4, pp. 215-218, april 2011. proceedings of engineering and technology innovation, vol. 17, 2021, pp. 32-41 41 [17] c. a. deledalle, l. denis, and f. tupin, “iterative weighted maximum likelihood denoising with probabilistic patch-based weights,” ieee transactions on image processing, vol. 18, no. 12, pp. 2661-2672, december 2009. [18] a. buades, b. coll, and j. morel “a non-local algorithm for image denoising,” 2005 ieee computer society conference on computer vision and pattern recognition (cvpr'05), ieee press, july 2005, pp. 60-65. [19] c. kervrann, p. perez, and j. boulanger, “bayesian non-local means image redundancy and adaptive estimation for image representation and applications,” siam conference on imaging science, july 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/).  proceedings of engineering and technology innovation, vol. 6, 2017, pp. 01 05 intelligent wireless remote control system design for a biped robot chih-hsuan chen 1 , chih-min lin 1, 2,* 1 department of electrical engineering, yuan ze university, taoyuan, taiwan, roc. 2 school of informatics, xiamen university, xiamen, fujian, china . received 30 july 2017; received in revised form 15 august 2017; accept ed 26 august 2017 abstract this paper presents the design and implementation of an intelligent wireless remote control system for a biped robot. a recurrent cerebellar model neural network (rcmnn) is introduced, then it is used for the walking control of a biped robot. furthermore, the remote communication module is designed for the remote control of a robot. finally, experimental results show that the developed system can achieve satisfactory control performance for the walking control of a high-order nonlinear biped robot. keywords: biped robot, wireless remote control, cerebellar model neural network 1. introduction biped robots are a favorite topic in the field of robotics [1-3]. many dynamics analysis and system control researches have been presented for the biped robots [4-5]. in order to achieve adaptive control of a biped robot to face unknown environment, an intelligent control algorithm will be proposed in this study . cerebellar model neural network (cmnn), first proposed by albus, has been applied to control a robot manipulator [6]. cmnn has been applied for the control of complex dynamic systems due to its good generalizing capability and fast learning property. in this paper, a recurrent cmnn (rcmnn) is introduced; then, an adaptive rcmnn-based control system is designed for the walking control of a biped robot. remote control has been frequently used in a lot of auto -matic control systems. recently, the wireless remote control robots have been developed to dictate a biped robot to move according to the user’s commands . the proposed walking command system can be applied to the remote medical robot, the exploration robot, the home security robot, and so on [7]. in the deve loped biped robot, the zigbee wireless communication has been designed for the remote communication between the monitoring computer and the robot. 2. remote control of the biped robot in order to achieve remote wireless control, a wireless module is used for the communication between the robot and personal computer (pc), as shown in fig. 1. the wireless module is also used to monitor the motion of the robot. the zigbee-100, which has 2.4 ghz bandwidth, is designed for robot communication. this peer-to-peer communication module is not easy to be interfered by other equipment or bands . * corresponding author. e-m ail address: cm l@ saturn.y zu.edu.tw proceedings of engineering and technology innovation, vol. 6, 2017, pp. 01 05 copyright © taeti 2 fig. 1 block diagram of remote communication robot 3. adaptive rcmnn-based control system 3.1. cerebellar model neural network a recurrent cmnn (rcmnn) is introduced as shown in fig. 2. the basic functions of rcmnn are described as follows. input space receptive -field space weight memory space association memory space recurrent unit k1  nk    output space kow     kpw   an p 1p k1  kna  k ikr tikr t on o 1o si sa sr sw so input space receptive -field space weight memory space association memory space recurrent unit k1  nk    output space kow     kpw   an p 1p k1  kna  k ikr tikr tikr ttikr t on o 1o si sa sr sw so fig. 2 architecture of an rcmnn (1) input space s i : for a given a a nt n ppp  ],,,[ 21 p , where a n is the number of input state variables, each input state variable i p is assumed to be quantized into e n discrete regions (called “neurons”) according to a given control space. (2) association memory space s a : several elements , fn , can be accumulated as a “block”. in this space, each block performs a receptive-field basis function, and the gaussian function is adopted here as the receptive-field basis function, which can be represented as 2 2 ( ) , 1, 2, rik ik ik d ik p c exp for k n v            (1) where ik  represents the output of the k -th receptive-field basis function for the i-th input with the mean ik c and variance . ik v in addition, the input of this block can be represented as )()()( ttrtptp ikikirik   (2) where ikr is the recurrent weight, and )( ttik  denotes the value of ik through delay time t . it is clear that the input of this block contains the memory term )( tt ik  , which stores the past information on the network and presents a dynamic mapping. (3) receptive-field space s r : the k -th multidimensional receptive-field function is defined as 2 2 11 ( ) ( , , , ) , 1, 2, a a n n rik ik k k k k ik d ii ik p c exp for k n v               p c v r (3) computer terminal peripheral circuits zigbee zigbee rs232 wireless communicate computer proceedings of engineering and technology innovation, vol. 6, 2017, pp. 01 05 copyright © taeti 3 where a a nt knkkk ccc  ],,,[ 21 c , a a nt knkkk vvv  ],,,[ 21 v and a a nt knkkk rrr  ],,,[ 21 r . the multi-dimensional receptive-field functions can be expressed in vector form as t nk d ],,,,[),,,( 1  rvcpφ (4) where da d nntt n t k t  ],,,,[ 1 cccc  , da d nntt n t k t  ],,,,[ 1 vvvv  and da d nntt n t k t  ],,,,[ 1 rrrr  . (4) weight memory space s w : each location of s r to a particular adjustable value in the weight memory space can be expressed as 11 1 1 11 1 [ , , , , ] o d o oo d d d o p n n n k kp knp n n n p n n w w w w w w w w w                    w w w w (5) where d d nt pnkppp www  ],,,[ 1 w , and kp w denotes the connecting weight value of the p-th output associated with the k -th receptive-field. (5) output space s o : the output of rcmnn is the algebraic sum of the activated weights in the weight memory and is expressed as    dn k kkp t pp wo 1 φw (6) the outputs of cmnn can be expressed in a vector not ation as φwo tt np o ooo  ],,,,[ 1  (7) the architecture of rcmnn is designed to have the advantages of simple structure and dynamic characteristics. the role of the recurrent loops is to consider the past value of the receptive-field basis function in the association memory space. thus, this rcmnn has dynamic characteristics. 3.2. adaptive rcmnn-based control system the biped robot system is highly nonlinear. to effectively control a biped robot, an adaptive rcmnn-based control is introduced in this subsection. the configuration of the proposed adaptive rcmnn control system of the biped robot system is illustrated in fig. 3. there are two inputs for the adaptive rcmnn controller: first is the tracking error, wh ich is the difference between the reference trajectory m d and actual trajectory d derived from accelerometer; and second is the angular velocity  obtained from gyroscope. the output of rcmnn is the control signal hipu , which adjusts the hip joint to achieve a stable posture. the robot needs this on -line adjusting technique to adjust its original traject ories as it encounters unexpected situations. another controller is the zmp compensator, which is also an adaptive rcmnn for auto-tuning the reference trajectories. the two inputs of the zmp compensator are: the tracking error between the reference zmp position ref zmp p and the actual zmp position zmp p derived from (8); and the change of this zmp tracking error. the out put of this rcmnn is the zmp control effort zmp u . by combining the pre-scheduled walking pattern p with the output of zmp compensator zmp u , and after inverse kinematics, the computed control torque 0  can be obtained for the biped robot. by adding the computed torque 0  with the hip compensating torque, hipu , the actual driving torque  will be applied to drive the biped robot. proceedings of engineering and technology innovation, vol. 6, 2017, pp. 01 05 copyright © taeti 4 fig. 3 block diagram of the adaptive rcmnn control biped robot system 4. experimental results 0 2 4 6 8 10 12 14 16 -5 0 5 10 15 20 25 30 z m p x -c o o rd in a te (c m ) time(sec) stable region (a) 0 2 4 6 8 10 12 14 16 -5 0 5 10 15 20 25 30 z m p x -c o o rd in a te (c m ) time(sec) stable region 0 2 4 6 8 10 12 14 16 -5 0 5 10 15 20 25 30 z m p x -c o o rd in a te (c m ) time(sec) stable region (a) (a) x-direction 0 2 4 6 8 10 12 14 16 -8 -6 -4 -2 0 2 4 6 8 z m p y -c o o rd in a te (c m ) time(sec) stable region (b) 0 2 4 6 8 10 12 14 16 -8 -6 -4 -2 0 2 4 6 8 z m p y -c o o rd in a te (c m ) time(sec) stable region 0 2 4 6 8 10 12 14 16 -8 -6 -4 -2 0 2 4 6 8 z m p y -c o o rd in a te (c m ) time(sec) stable region (b) 0 2 4 6 8 10 12 14 16 -8 -6 -4 -2 0 2 4 6 8 zmp x-coordinate(cm) z m p y -c o o rd in a te (c m ) (c) 0 2 4 6 8 10 12 14 16 -8 -6 -4 -2 0 2 4 6 8 zmp x-coordinate(cm) z m p y -c o o rd in a te (c m ) (c) (b) y-direction (c) x-y plane fig. 4 actual zmp of biped robot proceedings of engineering and technology innovation, vol. 6, 2017, pp. 01 05 copyright © taeti 5 the rcmnn control system is accomplished by the proposed control system, whose structure is depicted in fig. 2. the rcmnn used here is characterized as 9  e n , 4fn . the initial learning-rates are selected as 01.0zwη and .001.0 zrzvzm ηηη the initial values of weights are chosen as 0 kp w and 0 ik r . the initial values of parameters of means and variances are chosen as 22 ik v and 1 2 3 4 5 6 7 8[ , , , , , , , ] [ 49, 35, 21, 7, 7, 21, 35, 49]i i i i i i i im m m m m m m m      for all i and k . the receptive-fields are selected to cover the input space ]}35,35][35,35{[  along each input dimension, respectively. these parameters have been chosen through trials to achieve satisfactory con trol performance. the actual zero moment points (zmps) are shown in fig. 4. the experimental results show that the robot ’s zmp has been maintained within a stable region to achieve stable walking. 5. conclusions this paper has successfully developed and implemented an intelligent wireless remote control system for a biped robot. the rcmnn-based control system is developed. this adaptive control system has been successfully implemented on a biped robot system. the experimental results demonstrate the ability to achieve satisfactory control performance for stable walking for a high nonlinear biped robot. acknowledgement the support of the nation science council of republic of china under grant nsc 101-2221-e-155-026-my3 is gratefully acknowledged. references [1] y. sakagami, r. watanabe, c. aoyama, s. matsunaga, n. higaki, and k. fujimura, “the intelligent asimo: system overview and integration,” ieee/rsj international conf. on intelligent robots and systems, september-october 2002, vol. 3, pp. 2478-2483. [2] j. h. kim and j. h. oh, “walking control of the humanoid platform khr-1 based on torque feedback control,” proc. ieee international conf. on robotics and automation, april-may 2004, vol. 1, pp. 623-628, 2004. [3] y. d. hong, y. h. kim, j. h. han, j. k. yoo, and j. h. kim, “evolutionary multiobjective footstep planning for humanoid robots,” ieee transactions on systems, man, and cybernetics, part c (applications and reviews), vol. 41, no. 4, pp. 520-532, july 2011. [4] k. hosoda, t. takuma, a. nakamoto, and s. hayashi, “biped robot design powered by antagonistic pneumatic actuators for multi-modal locomotion,” robotics and autonomous systems, vol. 56, no. 1, pp. 46-53, january 2008. [5] y. s. cha, , k. g. kim, j. y. lee, j. j. lee, m. j. choi, m. h. jeong, et al., “mahru-m: a mobile humanoid robot platform based on a dual-network control system and coordinated task execution,” robotics and autonomous systems, vol. 59, no. 6, pp. 354-366, june 2011. [6] j. s. albus, “a new approach to manipulator control: the cerebellar model articulatio n controller (cmac),” journal of dynamic systems, measurement and control, vol. 97, no. 3, pp. 220-227, september 1975. [7] t. kim, e. kim, and j. w. kim, “development of a humanoid walking command system using a wireless haptic controller,” international conference on control, automation and systems , ieee press, october 2008, pp. 1178-1183.  proceedings of engineering and technology innovation, vol. 5, 2017, pp. 13 18 long-term structural performance of simplified slab system with steel deck-plate and sfrc geonho hong * , seungkoo hwang department of architectural engineering, hoseo university, asan si, korea. received 02 july 2017; received in revised form 08 august 2017; accept ed 13 august 2017 abstract slab is one of the most important members in building because it covers the largest amount of total structural members. the structural type and construction method of a slab can have a large influence on the construction pe riod and the amount of structural materials. this study investigates a simplified slab system with simple construction procedures. the simplified slab system consists of steel deck plate and steel fiber reinforced concrete (sfrc) without conventional reinforcing bar placing in construction field. the purpose of this study is to evaluate the long -term structural performance of a simplified slab system that replaces the form work and tensile reinforcement using structural deck-plate and replaces the temperatu re reinforcement using steel fiber reinforced concrete. two continuous -span slab specimens were tested and analyzed in the view-point of long-term serviceability such as deflection and crack, and evaluated the structural capacity and behavior. keywords: slab system, deck-plate, sfrc, long-term performance 1. introduction slab is one of the most important members in building structures because it covers the largest area of the building. the structural type and construction method of the slab can have a large influence on the construction period and the amount of structural materials.[1-2] this study investigated a simplified slab system with simple construction procedures. the simplified slab system consists of a composite steel deck plate and steel fiber reinforced concrete (sfrc) without conventional steel reinforcing bar. placing of the reinforcing bars is not required in the construction procedure of the slab. in this system, the composite steel deck plate replaces the tensile reinforcement, temporary concrete forms and shores, which results in simpler construction. the steel fibers can replace the temperature reinforcement of the slab, and reduce cracks.[3] previous research results indicated that the overall flexural performance of the simplified slab sy stem was suitable in slab members, even though reinforcing bars were not used. it has higher flexural capacity and stiffness than normal rc slab in the simply supported slab specimens, and improves the crack resistance performance. [4] the purpose of this study is to evaluate the long-term structural performance of a simplified continuous slab system because short-term characteristics of simply supported it have been confirmed in previous studies. serviceability such as deflection and crack development was tested and evaluated in the viewpoint of long -term loading conditions. * corresponding author. e-m ail address: honggh@ hoseo.edu tel.: +82-41-5405774; fax: +82-41-5405778 proceedings of engineering and technology innovation, vol. 5, 2017, pp. 13 18 copyright © taeti 14 2. experimental program 2.1. test plan and method two continuous slab specimens were planned and tested for evaluating their long -term structural characteristics. the details of tes t specimens are summarized in table 1 and figure 1. all slab specimens were designed as a two continuous span slab to be supported by a beam in the middle span. the first specimen was a conventional one-way rc slab with top and bottom reinforcing bars. yield strength of reinforcing bars was 400 mpa. the second one was a simplified slab system with steel deck-plate and sfrc without reinforcing bars. test was performed in three stages . the first stage was to fabricate the specimen and to observe the crack propagation while curing until the concrete compressive strength was developed. in the second stage, crack developments and deflections were observed under the long-term loading condition by concrete blocks over a period of 100 days as shown in figure 2. at the final stage, the structural recovery was observed after removing the long -term load. table 1 detail of test specimens sp ecimen length (mm) width (mm) thickness (mm) fck (m pa) steel fiber (kg/m 3 ) re-bar (bottom/top ) deck-p late rc 7100 1200 180 24 none 6-d13/6-d10 none ss 15.7 none 1mm thk. (a) rc (b) ss fig. 1 test specimen proceedings of engineering and technology innovation, vol. 5, 2017, pp. 13 18 copyright © taeti 15 fig. 2 long term loading concept diagram 2.2. materials the most important materials in simplified slab system are deck-plate and steel fiber. the role of deck-plate is to replace the temporary form, shore and bottom reinforcement. structural composite steel deck plate in this study was made by sgc400 in ks d 3506 which has 295n/mm 2 yield strength, and 400n/mm 2 tensile strength. section properties of deck-plate are shown in table 2. the steel fiber was an end hooked type with an aspect ratio 80. table 3 shows the mechanical properties of steel fiber, and fig. 3 shows the shape of the constituent materials . table 2 material properties of deck-plate standard m echanical p rop erty thickness (mm) weight (kg/m 2 ) section p rop erty (1m) yield strength (n/mm 2 ) tensile strength (n/mm 2 ) elongation (%) area (mm 2 ) neutral axis (mm) m oment of inertia (mm 4 ) ks d 3506 295 400 18 1.0 15.45 1,947 17.72 948,000 table 3 mechanical properties of end-hook type steel fiber diameter (mm) length (mm) asp ect ratio (l/d) density (g/cm 3 ) elastic modulus (× 10 3 n/mm 2 ) tensile strength (n/mm 2 ) 0.75 60 80 7.85 200 the average over 500 (a) deck-plate (b) steel fiber fig. 3 materials 3. results and discussion the experimental results were mainly evaluated in terms of cracks and deflection in the slab specimens. in the first stage, cracks due to drying shrinkage and self-weight were measured by the crack measuring instrument at the upper surface of the specimen after curing was carried out. in the second stage, concrete block corresponding to half of the live load and additional dead load was loaded at the center of the slab after 28 days of concrete curing, and the progress of deflection and cracking was observed. fig. 4 is a photograph showing a long -term loading test, and fig. 5 is a photograph of crack measurement. proceedings of engineering and technology innovation, vol. 5, 2017, pp. 13 18 copyright © taeti 16 fig. 4 long-term loading test fig. 5 crack measurement fig. 6 shows the schematic diagram of cracking behavior just before and after long -term loading test. as shown in the figure, it can be seen that the number of cracks in ss specimen is remarkably reduced compared with that in conventional rc specimen. and the crack width in the negative moment area which is the upper part of the center supporting beam was the largest in both specimens. (a) age 27 (b) age 30 fig. 6 crack pattern and development fig. 7 shows the number of cracks caused by shrinkage and dead load in each specimen before the long -term loading experiment. the number of cracks in normal rc slab was 37, and the number of it in ss slab was 6, showing about 16% cracks compared to rc slab cracks. the number of cracks in the two specimens increased sharply after the 11th day of age, which is consistent with the fact that the strain of steel fiber reinforced concrete changes rapidly around 10 days in choi et al. [5] since then, the development of crack numbers has grown slowly until the 28 th day. fig. 7 number of cracks fig. 8 shows the crack width change in negative moment region according to the loading conditions. all specimens were loaded with concrete blocks at 28 days of age, and the load was removed at 138 days. as a result of observing the crack with the largest crack width in the negative moment area, the crack width of the ss specimen was only about 65% of the rc specimen crack width before additional load was applied. after the concrete blocks were loaded, the crack width of the negative moment area of ss specimen increased rapidly, and the crack width was larger than that of rc specimen after 55 days of age. however, number proceedings of engineering and technology innovation, vol. 5, 2017, pp. 13 18 copyright © taeti 17 of cracks and crack width other than negative moment area of ss specimen were evaluated to have fewer cracks and crack width than normal rc specimen. in other words, the load resistance of the ss specimen in positive moment area was evaluated to be superior to that of the rc specimen due to the deck-plate effect. from these results, it can be concluded that the steel fiber reinforced concrete deck-plate slab itself without reinforcement is not expected to have efficient resistance against the load in the negative moment area, but the resistance against drying shrinkage is excellent. fig. 8 crack width development in negative moment area fig. 9 long-term deflection fig. 9 shows the measured long-term deflection of the two slab specimens. in the viewpoint of overall deflection behavior, the deflection of both slab specimens continuously increased until the concrete blocks were was removed. the overall shape of deflection pattern was similar. due to the stiffness effect of deck-plate, deflection of the ss specimen was less overall than those of the rc specimen. the amount of elastic deflection of ss specimen was only about 50% of that of the rc specimen. the amount of deflection increase due to the long-term loading showed a similar tendency, but the increase of long-term deflection in ss specimen was slightly smaller than that of rc. 4. conclusions the conclusions of the two continuous slab long-term loading tests are as follows. (1) the cracks generated in the ss specimen before long-term loading were less than half of the rc specimen and the crack width was also smaller than rc slab. this result indicates that the crack resistance of the steel fiber reinforced concrete was superior to that of the normal reinforced concrete slab. (2) the steel fiber reinforced concrete deck-plate slab itself without reinforcement is not expected to have efficient resistance against the load in the negative moment area, but the resistance against drying shrinkage is excellent than normal reinforced concrete slab. proceedings of engineering and technology innovation, vol. 5, 2017, pp. 13 18 copyright © taeti 18 (3) the overall shape of long-term deflection pattern was similar in two specimens . but, the amount of deflection in the ss specimen was less overall than that of the rc specimen because of the stiffness effect of deck-plate. the amount of elastic deflection of ss specimen was only about 50% of that of the rc specimen. and, the amount of deflection increase due to the long-term loading showed a similar tendency, but the increase of long-term deflection in ss specimen was slightly smaller than that of rc. acknowledgement this work was supported by the national research foundation of korea (nrf) grant funded by the korean government(moe) no. 2016r1d1a3a03918164. references [1] h. sim, g. yang, e. lee, and h. chung, “the experimental study of flexural behavio ur of continuous composite slab reinforced with structural steel deck plate,” proceedings of architectural institute of korea, 2004. [2] s. lee and y. lee, “an evaluation of structural safety for steel wire-integrated deck plate system,” journal of the architectural institute of korea, vol. 23, no. 11, pp. 43-50, december 2007. [3] architectural institute of korea, a study on the design of steel fiber reinforced concrete slabs on grade. aik, korea , 2007. [4] g. hong, “flexural performance evaluation of composite steel deck one-way slab system with sfrc”, journal of asian concrete federation, vol. 1, no. 1, pp. 22-27, september 2015. [5] h. choi, k. choi, o. choi, and y. hwang, “a prediction model of shrinkage cracking of steel fiber reinforced concrete,” journal of the architectural institute of korea structure & construction , vol. 28, no. 6, pp. 59-66, november 2012.  proceedings of engineering and technology innovation, vol. 6, 2017, pp. 12 16 coal fired power plant with chemical looping combustion gajanan dattarao surywanshi, venkata suresh patnaikuni * department of chemical engineering, national institute of technology warangal, warangal, india. received 30 july 2017; received in revised form 17 august 2017; accept ed 30 august 2017 abstract around 54.7% of the total energy in india is produced from coal-fired power plants. in order to reduce the emission of co2 into the atmosphere from these power plants, the effluent co2 needs to be captured and sequestered or utilized. chemical looping combustion (clc) is an inherent co2 capture technique in which coal is reacted with oxygen in the form of metal oxide instead of air and produces concentrated co2 as a mixture of co2 and steam. co2 from this mixture can be easily separated and captured by condensing the steam. in this paper, a clc integrated coal fired power plant with fe2o3/al2o3 as oxygen carrier was simulated using aspen plus ® simulation tool and the energy analysis was presented. high ash indian coal was used in the present work. results showed that clc integrated coal fired power plant (cfpp) is more energy efficient with co2 capture. keywords: chemical looping combustion, power plant, coal, co2 capture 1. introduction the carbon dioxide level in the atmosphere is considered as a main environmental problem to be solved immediately. this level of carbon dioxide in the atmosphere has been increased at a tremendous pace for last decades [1]. since carbon dioxide is the most powerful greenhous e gas [2], it causes global warming and subsequent rise in atmospheric temperature. average global temperature is raised from14.25°c in 2000 to 14.6°c in 2010 [3]. power and industrial sectors are the main sources of carbon dioxide [4]. among the power pla nts, fossil fuel combustion accounts for about 40% of co2 emissions to the atmosphere. at the same time, more than 50% of the energy demand in the world is satisfied by the fossil fuels. carbon dioxide capture and sequestration is the best method to get co ncentrated co2 stream from power plants to store it safe and permanently. there are three methods of carbon dioxide capture; pre -combustion, post combustion and oxy-fuel combustion methods [5]. in precombustion method, fuel is partially oxidized to co and h2 (syngas), co formed can be converted to co2 to capture prior to the combustion step using physical absorption, adsorption etc. post combustion method separates the co2 from the flue gas of combustion, i.e. after the combustion step. in oxy -fuel combustion method, an additional air separation unit is provided to give pure oxygen as oxidant for combustion. so the combustion produc ts are pure co2 and h2o that eases the separation of carbon dioxide. but both gas separation unit and air separation unit a re energy-intensive that reduces the efficiency of the plant. chemical looping combustion (clc) is a novel technology for combustion that enables the co 2 capture unit integration in an igcc power plant without compromising the efficiency of it as in conven tional co2 capture methods. * corresponding author. e-m ail address: pvsuresh@ nitw.ac.in tel.: +91-870-2462628 proceedings of engineering and technology innovation, vol. 6, 2017, pp. 12 16 copyright © taeti 13 in clc, combustion occurs in two reactors air reactor and fuel reactor. solid materials (metal oxides) called oxygen carriers circulate in between these reactors to transfer oxygen required for combustion. air reactor and fuel reactor are fluidised by air and fuel respectively. fuel takes the oxygen from metal oxides for combustion and converts the metal oxides to reduced form. thes e reduced metal oxides are further oxidised in air reactor, as it utilise the oxygen from air to form metal oxides. this loop continues and combustion occurs in fuel reactor without any contact between air and fuel. also the combustion products are pure co 2 and h2o and co2 can be separated by a simple condensation of water. reaction in air reactor is exothermic and that in fuel reactor is endothermic (most often). so the heat required in fuel reactor can be carried from air reactor by oxygen carriers. therefore, the total heat evolved from the system remains same as that of normal combustion. as the coal consumption will be increasing by coming years, the countries like india having a great coal reserves can achieve the future energy demand with cheaper resource. in this work, indian coal having high ash content has been taken as t he fuel and checked the feasibility of integration of clc. the software used for the simulation studies is aspen plus . 2. process model forclc integrated coal direct power plant with combined cycle in this case solid fuel can be directly sent to the fuel reactor of clc system in gas turbine section. the gasification and reduction of metal oxides take place in a single physical system (fuel reactor). so there is no need of a separate gasificati on section to produce syngas. therefore, an air separation unit to provide pure oxyge n for gasification and gas treatment processes for separation of co2 from the other gases are not required. this reduces the capital cost as well as the size of the plant. being one of the most energy consuming processes, elimination of asu enhances the th ermal efficiency of the plant. flow sheet of the plant is given in fig. 1. fig. 1 aspen flow sheet of ig-clc with combined cycle ryield block converts the coal into its constituent elements and they are sent to fuel reactor. the fuel reactor is simulated using two rgibbs reactors allowing for gasification at the bottom section and reduction reaction at the top. metal oxide part icles coming from air reactor enter the fuel reactor from the top and flow down under gravity. meanwhile coal is introduced at the bottom section along with the gasifying agent steam. metal oxides (fe2o3) react with the gasified products in the top section while going down and they are reduced to fe 3o4. the gaseous products rise up and come out from the top section. reduced metal oxides along with char and ash come out from the bottom. metal oxides are separated from ash and char using a component separator block ‘sep2’ and recycled to air reactor for further oxidation. air reactor is maintained in isothermal condition w ith 1100°c. fuel reactor is kept at adiabatic conditions. sensitivity analysis has been carried out to study the effect of solids on char conversion by varying the steam requirement. proceedings of engineering and technology innovation, vol. 6, 2017, pp. 12 16 copyright © taeti 14 coal is defined as a ncpsd sub stream as its state is not included in the software database. ncpsd sub streams are solids with a particle size distribution that has to be specified. the heat of combustion is specified in dry basis as 3850 kcal/kg in these simulations, wet coal mass flow rate of 15.5 kg/sec corresponding to 250 mw of heat input energy has been used. the composition analysis of the coal considered for the present study is given in table 1. table 1 component attribute analysis of coal proximate analysis (weight %) moisture 7.72 volatile matter 22.35 ash content 38.39 fixed carbon 31.54 ultimate analysis (weight %) carbon 42.39 hydrogen 3.33 nitrogen 2.49 oxygen 5.28 sulphur 0.4 hhv value (kcal/kg) 3850 sulfur analysis (weight %) pyritic 0.2 sulfate .16 organic .04 3. results and discussion a detailed study of in-situ clc is required to improve its performance. as oxygen carriers have a direct effect on conversion, higher carbon conversion can be achieved in this case. it is found that the steam requirement for this case is high. 0.43 kmo l/s of steam is used, which is relatively high. so the sensible use of oxygen carriers to reduce steam consumption would boost the net power production. 3.1. effect of oxygen carriers flow rate on overall conversion of fuel fig. 2 effect of fe2o3 flow rate on overall conversion of fuel overall conversion of fuel is calculated on the basis of carbon present in the coal. so the conversion can be defined as; 2 (%) co out overall coversion cin  (1) where, co2 out-moles of co2 present in fuel reactor outlet and cin -moles of carbon in the coal input proceedings of engineering and technology innovation, vol. 6, 2017, pp. 12 16 copyright © taeti 15 split fraction of steam from hrsg is kept at 0.5 (0.43 kmol/s). the variation of conversion with flow rate of fe 2o3 is given in fig. 2. it can be observed from the figure that the fuel conversion can be increased by increasing the solids flow rate. since higher conversion helps in extracting more co2 and results in more power production, further simulations are carried out for 97% of carbon conversion. 3.2. role of oxygen carriers on gasification of coal in fuel reactor fig. 3 effect of solids on char conversion for fe2o3 the oxygen carriers, which are unconverted in the top section of reduction reaction, will be flowing down to the gasification section. fig. 3 represents the variation in the amount of unconverted carbon with respec t to the oxygen carrier flow. solid flow rate is represented by the ratio of solids supplied to solids required for the complete coal conversion. with the help of above sensitivity analysis for various parameters on the clc process, the plant is simulated and the results are summarised in table 2. it contains total energy production, total energy consumption and co 2 capture efficiency. table 2 simulation results of the clc integrated coal direct power plant energy (mw) value coal inp ut 250 gas turbine outp ut 35 steam turbine outp ut 56.3 total outp ut 91.3 total p ower consump tion air sep aration unit ----- o2 comp ression ----- co2 cap ture unit 4.2 net p ower p roduction 87.1 overall efficiency 34.84 % % co2 cap ture 99.94% 4. conclusions a clc integrated coal fired power plant with fe 2o3/al2o3 as oxygen carrier and high ash indian coal as the fuel was simulated using aspen plus ® simulation tool and the energy analysis was presented. results showed that clc integrated coal fired power plant (cfpp) has an overall net thermal efficiency of 34.84% with a co2 capture efficiency of 99.94%. this level of thermal efficiency is higher than the conventional coal fired power plant with post combustion co 2 capture. references [1] c. song, “global challenges and strategies for control, conversion and utilization of co2 for sustainable development involving energy, catalysis, adsorption and chemical processing ,” catalysis today, vol. 115, no. 1-4, pp. 2-32, june 2006. proceedings of engineering and technology innovation, vol. 6, 2017, pp. 12 16 copyright © taeti 16 [2] j. adanez, a. abad, f. l. garcia, p. gayan, and l. f. de diego, “progress in chemical-looping combustion and reforming technologies,” progress in energy and combustion science, vol. 38, no. 2, pp. 215-282, april 2012. [3] p. nema, s. nema, p. roy, “an overview of global climate changing in current scenario and mitigation action,” renewable and sustainable energy reviews, vol. 16, no. 4, pp. 2329-2336, may 2012. [4] ipcc special report on carbon dioxide capture and storage, chapter 2, 2005 [5] m. k. mondal, h. k. balsora, and p. varshney, “progress and trends in co2 capture/separation technologies: a review,” energy, vol. 46, no. 1, pp. 431-441, october 2012.  proceedings of engineering and technology innovation, vol. 17, 2021, pp. 13-20 numerical study of geosynthetic-reinforced soil wall subjected to static footing loading ananya srivastava, sagar jaiswal, vinay bhushan chauhan * department of civil engineering, madan mohan malaviya university of technology, gorakhpur, india received 13 november 2020; received in revised form 30 december 2020; accepted 15 january 2021 doi: https://doi.org/10.46604/peti.2021.6693 abstract this study intends to examine the behavior of a grs wall with static footing loading above it, while varying the positions of the footing. for the study of behavior of such complex structure, finite element modeling is handy and enables to look into the various stress/strain developed in the numerical model. in view of the above, a series of finite element (fem) simulations using a software (optum g2) is performed for the analysis of the grs wall. the governing parameters, such as footing width (b), reinforcement length (l), offset distance (d), are evaluated and the effect of these factors on the ultimate bearing capacity (q) and settlement (s) of the footing is presented in this study. the results depict that the settlement of the footing substantially reduced in the range of 36% and its ultimate bearing capacity is increased to 42% more than the conventional retaining walls. keywords: grs wall, static footing loading, fem, ultimate load-bearing capacity 1. introduction geosynthetic-reinforced soil (grs) walls are now preferred more than the traditional gravity retaining walls as such walls can withstand much larger surcharge loads than conventional retaining walls owing to their increased stability and load-bearing capacity. in case of bridge abutment, the wall usually used are bulky gravity walls that have to be supported by the piles/group of piles. however, grs walls when implied as bridge abutments such conditions does not arise that is why grs wall is preferred for such a construction. this also aids in reducing the bridge bumps at the edge of an approaching embankment and an adjacent bridge, thus, creating a more economic and safer solution. sometimes the backfill is used as a construction bed for building structure by laying the shallow foundation. it is therefore necessary to understand the behavior of footing resting on the backfill of grs wall. the previously available literature attained through various field observations, laboratory and numerical simulations also suggested that the ultimate bearing capacity and the settlement characteristics of the footing situated on the backfill of the wall were enhanced by adopting suitable arrangement of the reinforcements in the soil mass [1-4]. the outcomes from a limited number of field tests and laboratory model tests highlight the fact that these grs walls when subjected to static footing loads perform far better than the traditional gravity walls in terms wall deformations, ultimate load-bearing capacity and settlement of the footing [5]. however, when exposed to ultra-violet rays, the polymeric reinforcement may degrade its original strength. according to the aashto [6] and the fhwa [7] design guidelines, the stability of the reinforced soil retaining walls is guaranteed by three components namely, the retained soil mass, reinforced soil mass, and foundation soil. the retained soil mass is restrained by the reinforced soil mass, the active zone of the reinforced soil mass is restrained by the reinforcement * corresponding author. e-mail address: chauhan.vinaybhushan@gmail.com proceedings of engineering and technology innovation, vol. 17, 2021, pp. 13-20 14 layers which are embedded in the passive zone and both the retained and reinforced soil masses are supported by the foundation soil. when a structure, is planned close to bridge abutment such as a footing is situated on the reinforced wall, its stability is influenced by the load which is supported by the footing. structural symmetry does not exist for footing, as on one side the facing of the grs wall works as a boundary while on the other side there lies an infinite soil mass, thus making its analysis quite complex and highlighting the necessity to study its behavior thoroughly. moreover, as the position of the footing varies (i.e., near or far from the wall facing), the lateral displacement of the wall (z), load-bearing capacity (q) and the settlement of the footing (s) varies accordingly and the analysis become a tedious task. researchers have previously conducted many numerical analyses on the behavior of the grs walls carrying a strip footing considering two dimensions (2d) of the proposed model under the plane strain conditions [8-14]. the results so obtained reestablished the superiority of the reinforced earth structures over the traditional earth retaining structures by their excellent serviceability and load-bearing capacity. however, very limited studies, conducted so far, can provide a satisfactory result that can clearly define the relationship between the important controlling parameters (position of the footing on the grs wall) and the aberration in the behavior and performance of the grs wall under such fluctuating conditions [3, 15-17]. the previously conducted studies fall short in providing a unanimous opinion regarding the optimum values of the essential parameters being used in the finite element modeling of the above-mentioned structure. only a few studies have been able to establish a concrete load (q)-settlement (s) behavior of footings positioned on the grs walls. limited work has yet been done to explore the performance of the grs walls while varying the location of the footing. nevertheless, an exhaustive performance report is yet to be made which can provide an extensive database for the optimal construction and efficient design suitable for a grs wall bearing the footing load. henceforth, this paper presents fem based numerical study, conducted through a sequence of numerical model simulation to assess the effects of the governing parameters such as offset distance of the static footing load, the width of the footing, and length of geosynthetic reinforcement. the effect of these factors on the bearing capacity of the footing load on the grs wall and performance of the overall structure is evaluated based on criterion namely, lateral deflection of the wall facing, backfill surface settlement of grs wall, and the vertical settlement of the footing. moreover, a comparative analysis has been performed on a gravity wall and the grs wall of similar dimensions and loading conditions to investigate the load-bearing capacity and settlement characteristics of both the walls. this study also aims to inspect the effect caused by the modes of connection between the wall facing and the geosynthetic reinforcement and to identify the probable modes of failure of the wall. 2. methods 2.1. numerical modeling in the present study the numerical analysis in the present study has been performed by a numerical computational tool optum g2 [18], a finite element program dedicated to geotechnical deformation and stability analysis under plane strain conditions. this software empowers to use the graphical user interface and select the appropriate elements for the better accuracy of the load-bearing capacity characteristics while iterating the numerical simulations. the grs wall of height, h = 6m is considered in the current study is reinforced using a geogrid element and having a rigid facing (as blocks of concrete) and is subjected to static footing loading. a gravity wall of similar height (6m) is also constructed under the same guidelines to provide comparative results in this study. the configuration of the grs wall numerical model is constructed according to the fhwa [7] design recommendations. figs. 1 and 2 show the grs wall with footing and gravity wall with footing considered in this study respectively. for the grs wall and footing system, a leveling pad which is a rigid material (concrete) carrying measurements as 2.0 m × 0.2 m (in altitude) is rested on the soil used in the foundation. above this, a modular block concrete facing with measurements as 1.5m × 0.6m (in proceedings of engineering and technology innovation, vol. 17, 2021, pp. 13-20 15 altitude) is positioned, over which another layer of backfill soil is deposited which provides a lift of 0.6m. a geosynthetic layer of the preferred length (l = 0.7h to recreate the field construction procedure) is positioned on the backfill soil before placing the subsequent modular block facing. the preceding lift is still maintained such that a section of geosynthetic reinforcement is sealed between the consecutive concrete facings. then backfill layer is again placed over the geosynthetic layer and this process is reiterated until the required wall height is accomplished. the above process shows the real stress path experienced by the total wall system. normal supports were utilized to fix the lateral boundaries, as it inhibits the movement of the soil in the lateral direction. the soil at the bottom of the structure was bound by fixed supports to inhibit both vertical and horizontal movement of the soil. the rigid footing (concrete) with dimensions 1.0m × 0.5m is positioned on the wall with varying offset distances (d), which is defined as the distance of the footing from the rigid concrete facing block, as shown in fig. 1. fig. 1 geosynthetic-reinforced soil (grs) wall considered in the present study fig. 2 gravity wall considered in the present study the offset distance is varied in the study and the effect of this variation on the grs wall and the gravity wall is investigated. 2.2. material parameters the concrete facing blocks, footing, leveling pad, are modeled as rigid materials with unit weight as 25 kn/m 3 . the backfill soil is considered as an elastoplastic material that follows the mohr-coulomb failure criterion following associated flow rule used by various researchers due to its proficiency in reasonably simulating the soil behavior in grs walls [19-23] having cohesion, c = 0 kpa and internal friction angle, ϕ = 34°, and dry unit weight, γdry = 19 kn/m 3 . the reinforcement using geogrid element is considered as an elastic material able to take tensile load only with elastic stiffness (ea) of 1000 kn/m [24]. proceedings of engineering and technology innovation, vol. 17, 2021, pp. 13-20 16 the mechanism utilized for the transfer of forces between two dissimilar materials is simulated using an elastoplastic interface, bearing the interface coefficient values like 0.7 [24] and 0.8 [25] for geosynthetic-soil and geosynthetic-block interfaces, respectively. these are allocated at all the interfaces in this grs wall and footing system. such simulations are carried out using the multiplier elastoplastic analysis in which a set of multiplier loads (shown in figs. 1 and 2) are incremented in a sequence of steps until collapse load is determined [24, 26]. the optimum number of elements present in the mesh primarily governs the efficiency of the numerical model. moreover, taking a large number of elements for a given mesh becomes time taking procedure for the simulation. on account of the above, a sensitivity analysis is conducted to acquire the ideal number of elements desired for the mesh taken in the present analysis [24-25, 27-28]. the overall number of elements is ranged from 2000 to 6000 with a successive increment of 1000 elements in each simulation. it was found that 4000 elements are optimal for the mesh taken in the current analysis. the selected number of elements produces a very fine mesh for the considered geometry and found to be significant for achieving reasonable performance. 3. results and discussion the present study performs a comparative analysis of the performance of the unreinforced gravity wall and the grs wall under identical loading conditions. the footing was placed at an offset distance d = 2m, where the footing is neither too distant from the wall facing nor very near to it and has an even distribution of the reinforcing members beneath the surface. the results obtained from the numerical simulations are presented through a graph of settlement vs incremental load derived by applying the load on the footing in fig. 3. fig. 3 maximum settlement of the footing (s) vs applied incremental load on footing resting over a grs wall and unreinforced gravity wall at offset distance (d) of 2m the graph noticeably shows that the settlement (s) of the footing in the grs wall is lower in magnitude than the settlement obtained in the unreinforced gravity wall as the load increases. the footing situated on the grs wall experiences a maximum settlement of 63% with an incremental loading whereas the footing on the unreinforced wall suffers a settlement of 74.94% with the same offset distance d = 2m in both cases. the load-bearing capacity (q) in the case of a grs wall is higher than the load-bearing capacity (q) of the unreinforced gravity wall. the difference in the magnitude of settlement characteristics of the footing between grs wall and conventional retaining wall increases as the height of the wall rises, which suggests the supremacy of grs walls over the conventional gravity walls. henceforth, to examine the performance of the grs wall with a footing load, an extended study was performed to analyze the behavior of the grs wall and footing under static loading system, while varying the offset distance of the footing. in order to choose the appropriate elastic stiffness (ea) of the geosynthetic, the numerical simulations were performed at various values of ranging from 500 kn/m to 2000 kn/m at an interval of 500 kn/m each to determine the optimum value of ea to economize the design. through the outcomes of the proceedings of engineering and technology innovation, vol. 17, 2021, pp. 13-20 17 simulations, it was evident that as the elastic stiffness of the reinforcing material increases, the lateral displacement of the wall facing decreases correspondingly [29]. furthermore, the result indicated the value of elastic stiffness is kept constant at 1 000 kn/m throughout the study. the settlement of the footing is also affected by footing’s offset distance, which is the distance between the wall facing and the position of the footing. fig. 4 below shows the variation of the d and depicts the effect it produces on the settlement of the footing, w.r.t the incremental load applied due to the footing. as shown in fig. 4, the d/b ratio has a significant impact on the settlement characteristics of the footing load situated on the grs wall. the footing experiences maximum settlement when the d/b ratio is least i.e., 0.5. this can be attributed to the fact that when d/b = 0.5, on one side of the footing there exists a rigid wall facing while on the other side there is a presence of extended reinforcement member, causing uneven reinforcement, due to which the total axial load cannot be endured, leading to the failure of the wall and excessive settlement of the static footing. as the d/b ratios increase from 1.0 to 6.0, again at those positions of the footing where the reinforcements under the soil are almost equally distributed, the settlement values are low in magnitude. on the other hand, when d/b ratio increases and the footing is placed beyond the influence of reinforcement members, the settlement is higher. the footing settlement increases by 53% as the position of footing is varied from d/b = 0.5 to 6.0. this indicates that the footing must be placed at such a position where the reinforcement is equally distributed with minimum influencing factors, to receive maximum load-bearing capacity (q) and least settlement (s). the results of this study demonstrates that the footing that rests over the backfill of the grs wall should neither be very close to the rigid block nor very far away. for better results, the footing must be placed in the range of d/b = 2.0-4.0. the results of the load and the normalized settlement of the footing (s/b) are presented in the form of the plot (fig. 4), which follows the same variation trends as depicted by xiao et al. [30]. fig. 4 variation of load vs normalized settlement of the footing for the various d/b ratios fig. 5 variation of load-bearing capacity vs normalized lateral displacement of the wall facing for various d/b ratio the lateral displacement of the wall facing is a prominent serviceability criterion as it may lead to the successive backfill settlement, due to which the surrounding structure resting nearby/backfill surface may be susceptible to failure or distress. here again, d is known to play a major role in the lateral displacement of the wall facing as load coming through footing pushes the wall outwards to move away from its original position. a comparison of the lateral displacement of the wall facing (z) normalized with the height of the wall (h) versus its load-bearing capacity (q) is made for various d/b values and shown in fig. 5. for d/b = 0.5, the possible failure is being intercepted at the upper part of the wall facing, due to which the movement of the upper block is rapid, however, the movement of the lower portion of the wall is not much high. for d/b = 6, the footing is far away from the wall, and behaves like footing in general cases, whose influence is not greatly observed on the lateral displacement of the wall. also, at d/b = 1, the lateral displacement is much higher which can lead in serviceability issues. despite, the z/h is lowest at a given load for d/b = 2-4, at which the displacement of the footing is also the minimum. therefore, this study suggests the optimum value of d/b = 2-4 must be used, depending upon the importance of the structure. proceedings of engineering and technology innovation, vol. 17, 2021, pp. 13-20 18 a similar variation trend is noticed in the result reported by rahmouni et al. [31] for d/b = 1.0. this can be attributed to the fact that, when the footing is situated closest to the wall, it is surrounded by the rigid wall facing and strong reinforcement beneath the ground. thus, the load so exerted is being resisted (failure plane intercept) by both the rigid wall facing and reinforcement which is evenly situated causing lesser displacement of the wall facing. also, the lower wall facing suffers greater displacements than the upper wall facings which are in line with the results of zheng et al. [32]. when the d/b ratio increases, the influence of reinforcement beneath the footing is reduced due to uneven distribution of reinforcement on the both sides of the footing. due to this uneven distribution of the load-bearing members, the footing cannot bear the axial load, thus eventually leading to its failure in the form of displacement of the wall facing. when the d/b ratio is varied from 0.5 to 6.0, the settlement of the footing undergoes a minor raise of approximately 1.6% in magnitude, but the ultimate bearing capacity of the footing increases by 21%. a comparison is made for footing resting on the reinforced zone of grs wall and beyond the reinforcement and depicted in fig. 6. the present study shows two types of failures, a failure above the topmost layer of reinforcement and general failure within the reinforced zone. when geosynthetic reinforcement distorts locally due to applied stress below the base of the footing, it generates tension that carries a vertical resistance component, i.e., an uplift force as shown in the fig. 6. in fig. 6(a), the heaving of the soil on the left of the footing is found to be lesser when compared to the right due to obvious reason, as a rigid wall is present on the left side of the. usually, in case of shallow foundation, the zone of shear failure extends up to a depth of 2b below the footing. the first layer of reinforcement is at a depth of 0.6m from the ground level, in this case the failure plane/wedge does not extend below the reinforcement provided and gets confined in the local zone of soil resting above the first reinforcement layer. the laterally outward displacement coordinates signify the wall deformation caused as the footing sinks into the ground when the stress exceeds the bearing capacity. also, a severe stress zone is getting developed beneath the footing which is limited to depth of 1.8b, however, no interference between the path of the failure plane and the reinforcement occurs as the tensile reinforcement is carrying the load by the footing (fig. 6(b)). (a) generation of failure zone when the footing lies above the reinforcement (b) generation of failure zone when the footing lies beyond the reinforcement fig. 6 the potential failure pattern of the footing 4. conclusions and recommendations in the present study, the performance of a 6m high grs wall under static footing load was assessed and the influence of the governing factors namely the reinforcement length (l), offset distance (d), and the width of the footing (b), on the ultimate load-bearing capacity (q), the settlement of the footing (s) and horizontal deformation of the facing (z) was evaluated. a comparative study was performed between the grs wall and gravity wall of similar dimensions, to investigate the variance in their performance and behavior under analogous loading conditions. this study also demonstrates that there exists an optimum proceedings of engineering and technology innovation, vol. 17, 2021, pp. 13-20 19 footing-facing distance (d) which gives the maximum value of load-bearing capacity and least settlement; however, the determination of ultimate load-bearing capacity is wisely based on the backfill surface settlement induced from the footing load which may influence the serviceability of the nearby existing structure. based on this study, the following conclusions can be drawn: the lateral deformation of the wall facing was lesser than 5% when the footing load was nearest to the facing of the wall, at a reinforcement length of 0.7h. as the offset distance increased from 1.0m to 6.0m the deformation of the wall facing increased by 26% than the initial conditions. the grs walls had the maximum load-bearing capacity q occurring at d/b = 0.5 and 6.0. but these ratios are not feasible for the safe design of the footing placed over the grs wall as at these ratios, the footing shows excessive settlement. the settlement of the footing load increases, when the location of the footing goes beyond the influence of the reinforcement zone, due to the uneven distribution of the reinforcing members. therefore, this study suggests an optimal ratio of d/b = 2-4 for the placement of the footing above the grs wall. there occurs an increment of 53% in the settlement when the d/b ratio is varied from 0.5 to 6.0. conflicts of interest the authors declare no conflict of interest. references [1] k. kazimierowicz-frankowska, “a case study of a geosynthetic reinforced wall with wrap-around facing,” geotextiles and geomembranes, vol. 23, no. 1, pp. 107-115, february 2005. [2] k. hatami and r. j. bathurst, “numerical model for reinforced soil segmental walls under surcharge loading,” journal of geotechnical and geoenvironmental engineering, vol. 132, no. 6, pp. 673-684, june 2006. [3] d. t. bergado, j. c. chai, and n. miura, “fe analysis of grid reinforced embankment system on soft bangkok clay,” computers and geotechnics, vol. 17, no. 4, pp. 447-471, 1995. [4] m. s. won and y. s. kim, “internal deformation behavior of geosynthetic-reinforced soil walls,” geotextiles and geomembranes, vol. 25, no. 1, pp. 10-22, february 2007. [5] m. t. adams and j. g. collin, “large model spread footing load tests on geosynthetic reinforced soil foundations,” journal of geotechnical and geoenvironmental engineering, vol. 123, no. 1, pp. 66-72, january 1997. [6] aashto, lrfd bridge design specifications, customary u.s. unit. american association of state highway and transportation officials, washington, dc, usa, 2012. [7] r. r. berg, b. r. christopher, and n. c. samtani, “design of mechanically stabilized earth walls and reinforced soil slopes,” fhwa, washington, dc, rep. no. fhwanhi-10-024, 2009. [8] y. zheng and p. j. fox, “numerical investigation of geosynthetic-reinforced soil bridge abutments under static loading,” journal of geotechnical and geoenvironmental engineering, vol. 142, no. 5, 04016004, may 2016. [9] h. i. ling and d. leshchinsky, “finite element parametric study of the behavior of segmental block reinforced-soil retaining walls”, geosynthetics international, vol. 10, no. 3, pp. 77-94, june 2003. [10] k. hatami and r. j. bathurst, “development and verification of a numerical model for the analysis of geosynthetic reinforced soil segmental walls under working stress conditions,” canadian geotechnical journal, vol. 42, no. 4, pp. 1066-1085, august 2005. [11] e. guler, m. hamderi, and m. m. demirkan, “numerical analysis of reinforced soil-retaining wall structures with cohesive and granular backfills,” geosynthetics international, vol. 14, no. 6, pp. 330-345, december 2007. [12] b. huang, r. j. bathurst, and k. hatami, “numerical study of reinforced soil segmental walls using three different constitutive soil models,” journal of geotechnical and geoenvironmental engineering, vol. 135, no. 10, pp. 1486-1498, october 2009. [13] b. huang, r. j. bathurst, k. hatami, and t. m. allen, “influence of toe restraint on reinforced soil segmental walls,” canadian geotechnical journal, vol. 47, no. 8, pp. 885-904, august 2010. [14] s. h. mirmoradi and m. ehrlich, “numerical evaluation of the behavior of grs walls with segmental block facing under working stress conditions,” journal of geotechnical and geoenvironmental engineering, vol. 141, no. 3, march 2015. proceedings of engineering and technology innovation, vol. 17, 2021, pp. 13-20 20 [15] d. t. bergado and c. teerawattanasuk, “2d and 3d numerical simulations of reinforced embankments on soft ground,” geotextiles and geomembranes, vol. 26, no. 1, pp. 39-55, february 2008. [16] j. han, principles and practice of ground improvement. john wiley & sons, 2015. [17] r. k. rowe and s. k. ho, “horizontal deformation in reinforced soil walls,” canadian geotechnical journal, vol. 35, no. 2, pp. 312-327, april 1998. [18] “optum g2,” http://www.optumce.com/, 2020. [19] d. leshchinsky, y. hu, and j. han, “limited reinforced space in segmental retaining walls,” geotextiles and geomembranes, vol. 22, no. 6, pp. 543-553, december 2004. [20] j. huang, r. l. parsons, j. han, and m. pierson, “numerical analysis of a laterally loaded shaft constructed within an mse wall,” geotextiles and geomembranes, vol. 29, no. 3, pp. 233-241, june 2011. [21] k. h. yang and c. n. liu, “finite element analysis of earth pressures for narrow retaining walls,” journal of geoengineering, vol. 2, no. 2, pp. 43-52, august 2007. [22] j. han and d. leshchinsky, “analysis of back-to-back mechanically stabilized earth walls,” geotextiles and geomembranes, vol. 28, no. 3, pp. 262-267, june 2010. [23] y. yu, r. j. bathurst, and t. m. allen, “numerical modelling of the sr-18 geogrid reinforced modular block retaining walls,” journal of geotechnical and geoenvironmental engineering, vol. 14, no. 5, 04016003, may 2016. [24] s. jaiswal, a. srivastava, and v. b. chauhan, “performance of strip footing on sand bed reinforced with multilayer geotextile with wraparound ends,” proceedings of indian geotechnical conference, 2020, pp. 192-202. [25] a. srivastava and v. b. chauhan, “numerical studies on two-tiered mse walls under seismic loading,” sn applied sciences, vol. 2, no. 10, pp. 1-7, september 2020. [26] k. krabbenhoft, a. lyamin, and j. krabbenhoft, “optumg2: theory, optum computational engineering,” 2017. [27] a. pandey and v. b. chauhan, “evaluation of pull-out capacity of helical anchors in clay using finite element analysis,” geo-congress 2020: modeling, geomaterials, and site characterization, pp. 60-68, february 2020. [28] s. jaiswal, a. srivastava, and v. b. chauhan, “numerical modeling of soil nailed slope using drucker-prager model,” proceedings of geo-science and geo-structures, 2020. [29] y. zheng and p. j. fox, “numerical investigation of the geosynthetic reinforced soil–integrated bridge system under static loading,” journal of geotechnical and geoenvironmental engineering, vol. 143, no. 6, 04017008, june 2017. [30] c. xiao, j. han, and z. zhang, “experimental study on performance of geosynthetic-reinforced soil model walls on rigid foundations subjected to static footing loading,” geotextiles and geomembranes, vol. 44, no. 1, pp. 81-94, february 2016. [31] o. rahmouni, a. mabrouki, and d. benmeddour, “numerical study of the bearing capacity of a strip footing near of geosynthetic-reinforced soil retaining walls,” 4th international conference on new developments in soil mechanics and geotechnical engineering, june 2016, pp. 633-638. [32] y. zheng, p. j. fox, and p. benson shing, “numerical study of deformation behavior for a geosynthetic-reinforced soil bridge abutment under static loading,” international foundations conference and equipment expo, pp. 1503-1512, 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/). peti#10101_inpress20220809 proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx evaluation of local interpretable model-agnostic explanation and shapley additive explanation for chronic heart disease detection tsehay admassu assegie * department of computer science, injibara university, injibara, ethiopia received 15 december 2021; received in revised form 13 may 2022; accepted 14 may 2022 doi: https://doi.org/10.46604/peti.2022.10101 abstract this study aims to investigate the effectiveness of local interpretable model-agnostic explanation (lime) and shapley additive explanation (shap) approaches for chronic heart disease detection. the efficiency of lime and shap are evaluated by analyzing the diagnostic results of the xgboost model and the stability and quality of counterfactual explanations. firstly, 1025 heart disease samples are collected from the university of california irvine. then, the performance of lime and shap is compared by using the xgboost model with various measures, such as consistency and proximity. finally, python 3.7 programming language with jupyter notebook integrated development environment is used for simulation. the simulation result shows that the xgboost model achieves 99.79% accuracy, indicating that the counterfactual explanation of the xgboost model describes the smallest changes in the feature values for changing the diagnosis outcome to the predefined output. keywords shap, lime, xgboost, counterfactual explanation 1. introduction ensemble-based automated chronic heart disease detection systems have achieved promising performance in automated decision-making over the past few years. the smartness and maximized performance of ensemble methods in automated medical decision-making are due to the ever-growing power of computing devices and chronic heart disease datasets. although the performance of ensemble methods is much better than simple linear methods, it is difficult to explain the decision-making process of an ensemble model to a medical expert, because of the inherent complexity in the design of the ensemble model [1]. thus, explaining the decision-making process is crucial to increase the trust of medical experts in using ensemble models (e.g., the xgboost model) for making significant decisions such as heart disease diagnosis; failure to accurately identify a disease will result in danger to life. currently, complex ensemble-based models that rely on incomprehensible inferences are not an option for existing cardiac diagnostic systems. transparency is one of the main reasons why the adoption of ensemble-based complex models and automated diagnosis systems in the healthcare industry requires more caution than in other domains such as the e-commerce and entertainment industry [2]. over the past few years, the use of complex models, such as deep learning and ensemble methods, has become common for disease detection [3]. toğaçar et al. [4] proposed a support vector machine-based chronic heart disease diagnosis system. however, they found that much research effort is needed to evaluate the automated diagnosis output by using model explanation techniques [4]. they concluded that model explanation techniques such as local interpretable model-agnostic explanation (lime) * corresponding author.tsehayadmassu2006@gmail.com tel.: +251-9-21114923 proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 2 and shapley additive explanation (shap) have not been adequately and formally practiced in medical applications. interpretable ensemble-based models can explain the models’ diagnostics while creating efficient and accurate diagnostics. thus, an ensemble-based diagnosis system is critical for assisting physicians in making heart disease diagnoses [5]. rudin et al. [6] evaluated the shap interaction value of an xgboost-based gold price prediction model. the shap interaction value represents the feature importance score of the prediction accuracy of the xgboost model for gold price prediction. moreover, the shap summary plot is analyzed to explain the impact of features on the predictive outcome of the xgboost model for gold price prediction. the experimental result shows that the shap value provides insights into the prediction process of the model, and the features that influence the outcome of the xgboost model are identified. gramegna et al. [7] and cavaliere et al. [8] proposed an explainable model for diabetes mellitus prediction. the performance of the model was evaluated using the area under the curve (auc) (receiver operating characteristics curve (roc)) as the performance metric and shap as the model explainer. the researchers employed light gradient boosting (lightgbm) for model development. experimental results show that shap has the potential to explain the predicted probabilities based on the baseline lightgbm model. furthermore, the experiment reveals that the feature with the largest average absolute shap value is the most important since that influences the diagnosis outcome. the shap value is calculated and averaged across all samples and is ranked and plotted from the highest to the lowest. this study aims to evaluate the effectiveness of lime and shap in explaining how the xgboost model obtains the prediction outcome of chronic heart disease. to achieve the goal, a chronic heart disease dataset is employed. diverse counterfactual explanations (dice) are also used. 2. literature review machine learning (ml) is widely used to assist medical professionals in making decisions for heart disease diagnosis. however, the complexity of ensemble models (e.g., the xgboost model) is hindering the wider applicability of automated decision-making in the medical field, where a decision is critical to a patient’s life [9]. to interpret the ensemble models, researchers have conducted several studies on the model explanation techniques. bhagwat et al. [10] found that the use of complex models (such as ensemble and deep learning models) enabled data scientists to achieve the best possible accuracy in heart disease diagnosis. their study was conducted using benchmark datasets such as the heart disease data repository of the university of california irvine (uci). however, a better-performing model is too complex to interpret and explain. thus, providing a human-understandable explanation for such a complicated model helps domain experts trust the diagnosis outcome. interpretable ml is an area of research devoted to interpreting diagnostics produced by complex models. wang et al. [11] classified explainable ml problems into three categories, i.e., model interpretation, result checking, and transparent box design. this study focuses on outcome interpretation and model explanation, and explores the answers to the following research questions: (1) how do the explanation techniques explain the xgboost model’s diagnostic outcome? (2) which explanation technique is more appropriate for explaining the xgboost model’s diagnostic output to a medical expert? (3) what is the feature weight value of the feature that alters the positive diagnosis output into the desired diagnostic output? 3. methodology this study is conducted by using a heart disease dataset obtained from the uci repository. to develop an automated chronic heart disease diagnosis model, xgboost is employed. for simulation, python 3.7 programming language with jupyter proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 3 notebook integrated development environment is employed. to evaluate the effectiveness of lime and shap for describing the xgboost model diagnosis output, heart disease positive and heart disease negative instances are randomly selected from the testing set to simulate local or instance level explanation. the schematic diagram of the proposed model is illustrated in fig. 1. fig. 1 schematic diagram for model explanation 3.1. model explanation problem the problem of ml model interpretation can be defined as finding an explanation e. e is in a human-understandable domain by an interpretable model g. e represents explanation. an interpretable model g produces the prediction p’ = g(x) for a given data point x and its function c = f(x). the interpretable model globally mimics the prediction behavior of a complex model g. the explanation e ∈ e is given by interpretable models such as e = α(c, x) for some training points x and their prediction α [12]. 3.2. lime lime is employed to describe and interpret diagnosis outcomes of the ml model at the individual level for each predicted instance in a testing set [13]. thus, lime is a local model explanation approach because it explains single output at a local or instance level. this study employs lime to present the diagnosis probability of the xgboost model at the instance level. the explanation of the xgboost model diagnosis outcome at the local level helps healthcare experts make important decisions. mathematically, lime is defined as follows: ( ) ( , , ) ( )explanation x arg min g g l f g x g= ∈ + ωπ (1) where the model g explains the instance x. g measures the distance between the explanation and prediction of the original xgboost model, while the xgboost model complexity ω (g) is minimized (by considering fewer features). g is the family of possible explanations. the proximity πx defines the neighborhood distance around the instance x. the neighborhood distance is the distance between the randomly generated feature and the original feature in the dataset. 3.3. shap shap describes the model output based on feature interaction values [14]. feature interaction is a technique from the coalitional game theory, which is used to determine a fair pay-out distribution, based on the contributions of each player in a coalition. shap values are used to calculate each related feature’s contribution to the predicted outcome. the shap values of a feature are computed based on the mean marginal difference with and without the feature in question. shap applies the feature interaction value to estimate the model output. the technical definition of shap values is “the average marginal influence of a feature value over all possible combinations”. in other words, shap values consider all possible predictions for observation using all possible combinations between input features to each feature and the output of the xgboost model [15]. training set (70%) extreme boosting explainer, lime, shap train-test split ensemble learning original dataset testing set (30%) result explanation the trained model proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 4 3.4. data description in this study, the uci heart disease dataset is used for training and testing the xgboost model to predict if a person has heart disease or not. the dataset consists of 1025 instances of which 499 samples are healthy and 526 samples are patients. the dataset is divided into two sets: a training set and a testing set. the standard method of the train-test split is followed to divide the dataset into training and testing sets using the ratio of 70% by 30% respectively. the training set consists of 717 samples and the test set consists of 308 samples. for model training, the training set is used, and the test set is used for model testing. the researcher presents the diagnosis outcome made by the developed model using randomly selected test samples from the testing set. then, lime and shap explanations are generated based on the diagnosis outcome of the model. table 1 illustrates some of the data samples used in the simulation. the features of heart disease include age (age of patient), sex (gender of a patient), chest pain (cp), resting blood pressure (trestbps), cholesterol level (chol), fasting blood sugar (fbs), heart rate (thalach), angina induced by exercise (exang), oldpeak (depression), slope (slope of the heart), number of vessels (ca), thallium scan (thal), and target. table 1 samples of heart disease dataset no. age sex cp trestbps chol fbs restecg thalach exang oldpeak slope ca thal target 293 58 1 0 128 259 0 0 130 1 3 1 2 3 0 697 74 0 1 120 269 0 0 121 1 0.2 2 1 2 1 353 57 1 0 110 20 0 1 126 1 1.5 1 0 1 1 481 63 0 0 150 407 0 0 154 0 4 1 3 3 0 823 54 1 0 140 239 0 1 160 0 1.2 2 0 2 1 462 52 1 0 118 186 0 0 190 0 0 1 0 1 1 536 50 0 2 130 219 0 1 179 0 0 2 0 2 1 438 47 1 2 130 253 0 1 179 0 0 2 0 2 1 347 54 0 2 108 267 0 0 167 0 0 2 0 2 1 148 59 1 0 164 176 1 0 90 0 1.0 1 2 1 0 4. simulation results and discussion for the simulation of lime and shap, two randomly selected samples are considered. one is a healthy person or with a “heart disease negative” prediction outcome. the other is a patient or with a “heart disease positive” prediction outcome. two simulations (case 1 and case 2) are conducted on each predicted instance by using lime and shap. in the first simulation (case 1), the explanation generated by lime for each instance is analyzed. in the second simulation (case 2), the explanation generated by shap for each instance is analyzed. overall, true positive and true negative instances are used in the simulations to analyze how lime and shap are used to explain the xgboost model’s diagnosis output. 4.1. case 1 in this simulation (case 1), a random sample of the patient (instance id-1) is considered to analyze the explanation generated by lime on the xgboost model’s diagnosis outcome. the details of the instance employed in this simulation are shown in table 2. the xgboost model diagnosis probability for the patient class or target 1 is 99.35, and the probability for the healthy class or target 0 is 0.006. as shown in table 2, the xgboost model detects the presence of heart disease with a probability of 1.00. instance id-1 has a target value of 1 (as shown in table 2), which shows that the instance belongs to the patient class. the model correctly predicts instance id-1 with a diagnosis probability of 1.00 for the patient class and 0.00 for the heart disease negative class. fig. 2 summarizes the lime explanation for the diagnosis outcome of the xgboost model. the model’s diagnosis outcome is “patient” or “heart disease positive”. the sex feature affects the model’s output negatively while other features have a positive impact on the diagnosis outcome. as illustrated in fig. 2, the reasons why the xgboost model reaches a positive diagnosis proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 5 outcome are that the number of vessels (ca) is less than 0.00 and that the thallium scan is also less than 2.0. moreover, the heart rate is greater than 165 mmhg, influencing the model’s prediction as heart disease positive. the feature with the highest impact on the model’s output is the number of vessels. table 2 instance id-1 no. age sex cp trestbps chol fbs restecg thalach exang oldpeak slope ca thal target 293 59 1 2 150 212 1 1 157 0 1.6 2 0 2 1 fig. 2 explanation generated by lime for instance id-1 4.2. case 2 in this simulation (case 2), two heart disease negative instances (instance id-1 and instance id-2) are randomly selected from the test set. then, the diagnosis outcome of the xgboost model is explained using lime. table 3 summarizes the details of the instance used in the experiment (i.e., the xgboost model diagnosis outcome of the instance considered for the simulation). as shown in table 3, the xgboost model predicts instance id-2 as heart disease negative. the model correctly predicts instance id-2 with a diagnosis probability of 0.99. fig. 3 illustrates the explanation generated by lime, which also shows that the diagnosis probability of instance id-2 is 0.99. the features contributing to this diagnosis outcome include thallium scan (thal), oldpeak, sex, exercise-induced angina, slope, heart rate (thalach), and total resting blood pressure (trestbps) ordered by the magnitude of impact on the diagnosis outcome as illustrated in fig. 3. in contrast, features such as the cholesterol level and the number of vessels do not affect the model output. a person with a cholesterol level ≤200.00 and vessels ≤0.00 has a lower probability of getting heart disease. along with the lime explanation demonstrated in fig. 3, the influence of the features on the model output is given in table 4 for instance id-2. table 4 summarizes the contribution of heart disease features to the xgboost model prediction outcome using instance id-2. as illustrated in table 4, chest pain (cp) contributes more to the model’s prediction output being “heart disease positive” or “patient”, as compared to other features. in table 5, the shap values for instance id-1 are demonstrated. the instance id-1 is a positive instance considered for simulation. as demonstrated in table 5, the features, such as heart rate (thalach) and slope, add bias to the diagnosis made by the model. fig. 4 illustrates the effect of heart disease features on the xgboost model output using shap. as illustrated in fig. 4, chest pain has the highest impact on the xgboost model output. thus, chest pain is an indicator of the presence of heart disease. moreover, heart rate has the second highest impact on the xgboost model. a patient with a higher heart rate has a higher risk proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 6 of getting the disease compared to a patient with a lower heart rate. the third feature influencing model’s output is the number of vessels colored by x-ray. in contrast, thallium (thal), resting electrocardiogram, and slope have a lower effect on the output of the xgboost model. table 3 instance id-2 no. age sex cp trestbps chol fbs restecg thalach exang oldpeak slope ca thal target 49 60 1 0 117 230 1 1 160 1 14 2 2 3 0 fig. 3 explanation generated by lime for instance id-1 table 4 lime explanation for the features of instance id-2 (probability = 0.99) feature value impact on the xgboost model output chest pain (cp) 2.000 +0.14 thalassemia (thal) 2.000 +0.21 number of major vessels (ca) 0.000 +0.19 oldpeak 0.000 +0.09 sex 0.000 (female) +0.08 age 39 +0.027 exercise-induced angina (exang) 0.000 (no angina due to exercise) +0.08 resting electrocardiogram (restecg) 1.000 +0.03 maximum heart rate achieved (thalach) 133.000 -0.03 slope 1.000 (upward sloping) -0.09 table 5 shap explanation for the features of instance id-1 (e [f(x)] = 0.513) feature value impact on the xgboost model output chest pain (cp) 1 +0.16 number of major vessels (ca) 0 +0.12 thalassemia (thal) 2 +0.07 sex 0 (female) +0.03 slope 2 +0.03 age 34 +0.02 cholesterol (chol) 210 +0.02 total resting blood pressure (trestbps) 118 +0.02 exercise-induced angina (exang) 0 (no angina due to exercise) +0.01 maximum heart rate (thalach) 192 +0.01 resting electrocardiogram (restecg) 1 +0.01 fasting blood sugar (fbs) 0.000 (<120 mm/dl) 0.0 oldpeak 7 0.0 proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 7 fig. 4 effect of heart disease features on the xgboost model output using shap 4.3. shap explanation the shap value for instance id-1 is illustrated in fig. 5. as shown in fig. 5, blood pressure, thallium scan, and electrocardiogram have a negative impact on the xgboost model diagnosis outcome. in contrast, features such as the number of blood vessels, chest pain, cholesterol, age, and sex heart rate affect the xgboost model output. the features with the greatest effect on the model’s output include the number of blood vessels stained using x-ray, ca = 2 (showing that the vessels are not blocking the blood flow). the second most influencing feature for the model to make the negative diagnosis is chest pain, cp = 0 (showing that the patient is experiencing angina, e.g., symptomatic or asymptomatic angina pain). -0.15 -0.1 -0.05 0 0.05 0.1 0.15 0.2 0.25 ch thal ca oldpeak sex age exang restecg trestbps thalach slope f e a tu re i m p a c t o n t h e m o d e l o u tp u t feature shap explanation for the positive outcome -0.1 -0.05 0 0.05 0.1 0.15 0.2 ca chol sex slope oldpeak restecg f e a tu re i m p a c t o n t h e m o d e l o u tp u t feature shap explanation for the negative outcome fig. 5 shap explanation of the positive diagnosis outcome (instance id-1) fig. 6 shap explanation of the negative diagnosis outcome (instance id-2) proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 8 in addition, as illustrated in fig. 5, the instance has a chest pain value of 0, indicating the existence of a-typical angina. as a result, chest pain influences the model’s diagnosis outcome showing “heart disease patient”. furthermore, instance id-1 has high serum cholesterol (chol = 212 as illustrated in fig. 5). the patient also has a thallium scan value of 2 indicating a fixed defect, and the st segment is downward sloping (slope = 2). moreover, the patient has a heart rate of 157 mmhg, influencing the model’s positive diagnosis outcome. thus, a counterfactual (cf) explanation for this patient should be “to reduce cholesterol level and heart rate”. the change in cholesterol and heart rate could result in reduced heart disease risk. the feature impacting the xgboost model’s diagnosis outcome is the number of blood vessels (ca = 2, showing that the blood vessels are the worst cases blocking the blood flow). the second significant feature influencing the xgboost model’s output is chest pain, cp = 2. this indicates that the patient is experiencing symptomatic or asymptomatic pain, leading the model to show that the patient is suffering from heart disease. the third influential feature is the patient’s cholesterol level (chol = 212 mm/dl), which is higher than the normal cholesterol level of a healthy individual (170 mm/dl). a high cholesterol level could lead to the blockage of blood vessels. fig. 6 illustrates the shap explanation for xgboost model diagnosis outcome on instance id-2. as demonstrated in fig. 6, age and sex do not influence the xgboost model diagnosis outcome. chest pain, number of blood vessels, and cholesterol level contribute a lot to the model’s diagnosis outcome. 4.4. comparison of shap and lime explanations for further comparison and analysis, the xgboost model is tested on the heart disease positive instance using lime and shap. the diagnosis outcome of the xgboost model is explained using the shap value and explanation generated by lime. fig. 7 illustrates the effect of heart disease features on the xgboost model output. as demonstrated in fig. 7, the impact of the features on the xgboost model’s diagnosis outcome changes in a similar direction although the magnitude of feature impact of shap value is greater than the explanation generated by lime. the xgboost model decides that the given instance is a heart disease patient due to the higher impact of heart rate and chest pain type as illustrated in fig. 7. fig. 7 comparison of the feature impact on the xgboost model output using shap and lime 4.5. roc for the xgboost model roc is a performance measure that provides a comprehensive evaluation of classification models such as xgboost [16]. roc considers the confusion matrices in all threshold operations and combines the confusion matrices to obtain a performance result. fig. 8 demonstrates the roc for the developed xgboost model, showing that the area under the roc (auc) of the proposed model is scored as 0.99. -0.15 -0.1 -0.05 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 cp thal ca oldpeak sex age exang restecg thalach slope f e a tu re i m p a c t o n t h e m o d e l o u tp u t feature feature influence on the model output (lime vs shap) shap lime proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 9 fig. 8 roc of the xgboost model 4.6. counterfactual explanation when an ml model is applied to real-world chronic heart disease data, along with the reason for its diagnosis outcome, it is also crucial to identify the variation required in feature magnitude that alters the diagnosis outcome. the cf explanation is a model-agnostic explanation procedure that provides the smallest alterations required in feature values for the diagnosis outcome to be changed to the predefined outcome [17-18]. in other words, if x is the feature and y is an output label, the cf explanation shows the effect on y due to small changes in the value x. thus, cf helps determine the number of changes that need to be done in x to change the outcome from y to y’ (e.g., from “heart disease patient” to “not heart disease patient”). it gives the what-if explanations for the model. cf plays a great role in showing what change is required to the feature values of a patient suffering from heart disease from being patient to non-patient. to provide the proof of cf and explore the effect of feature magnitude on the diagnosis outcome of the xgboost model, three instances (instances id-3, 4, and 5) are randomly taken from the heart disease dataset, as shown in table 6. two heart disease positive instances and one heart disease negative instance are shown in the target column in table 6. these instances are considered an input to the cf explanation. with the positive instances as the input, two different cfs are generated. the cfs show the minimum changes required for the feature values to change the target or label of the instance, i.e., change the target from “patient” (1) to “no disease” (0) or vice-versa. the following are the observations regarding the output to instance id-3. in the cf output, sex, age, and type of chest pain (cp) are kept constant. therefore, in each of the cfs, those features are unvaried. the cfs for the instance are shown in table 7. as shown in table 7, the increase in oldpeak by the value of 0.5 and the change from thallium scan (thal = 2, showing the fixed defect) to the reversible defect (thal = 3) lead to a change in the intensity of heart disease. furthermore, the reduction of heart rate (thalach) by 38 mmhg, i.e., from 165 mmhg to 127 mmhg, leads to a change in the heart disease intensity. thus, the cf explanation is significantly important to determine the change in the values of input features to change the diagnosis outcome of the ml model. table 6 instances considered for the cf explanation instance id age sex cp trestbps chol fbs restecg thalach exang oldpeak slope ca thal target 3 43 0 2 122 213 0 1 165 0 0.2 1 0 2 1 4 74 0 1 120 269 0 0 121 1 0.2 2 1 2 1 5 59 1 0 164 176 1 0 90 0 1.0 1 2 1 0 table 7 cf explanation for instance id-3 instance id age sex cp trestbps chol fbs restecg thalach exang oldpeak slope ca thal target 3 43 0 2 122 213 0 1 165 0 0.2 1 0 2 1 new_cfs 43 0 2 122 213 0 1 127 0 2.6 1 0 3 0 proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 10 table 8 cf explanation for instance id-4 instance id age sex cp trestbps chol fbs restecg thalach exang oldpeak slope ca thal target 4 74 0 1 120 269 0 0 121 1 0.2 2 1 2 1 new_cfs 74 0 1 120 269 0 0 121 1 6.0 0 1 2 0 table 9 cf explanation for instance id-5 instance id age sex cp trestbps chol fbs restecg thalach exang oldpeak slope ca thal target 5 59 1 0 164 176 1 0 90 0 1.0 1 2 1 0 new_cfs 59 1 2 164 176 1 0 90 0 1.0 1 0 1 1 as shown in table 8, the increase in oldpeak by the value of 0.4 and the change in slope from upward sloping (slope = 2) to flat (slope = 0) leads to a change in the intensity of heart disease. thus, changing upward sloping to flat changes the output value of the xgboost model. as shown in table 9, the change in slope from flat (slope = 0) to upward sloping (slope = 2) leads to a change in the intensity of heart disease. thus, the slope value of heart disease changes the value of the xgboost model output from “patient” to “not patient” or “healthy class”. cf can show the changes required in the values of the input feature for changing the diagnostic outcome of the xgboost model to the desired outcome. hence, an explanation of the xgboost model diagnosis output provides insights into what can be carried out to change the diagnostic outcome. for example, what can the patient do to reduce the heart disease risk? the results in tables 7-9 show that a small change in the value of the original input feature produces the desired diagnostic outcome. as shown in table 9, only a change in the number of vessels colored by x-ray (ca) and chest pain (cp) could produce positive diagnostic outcomes while keeping other feature values constant. this shows that the cf explanation provided above is diverse, as the changes in more than one feature, namely the number of vessels colored by x-ray (ca) and chest pain (cp), have changed the outcome of the xgboost model. the quality of cfs is described in terms of the proximity of the cf value to the original feature, which is the euclidean distance between the cf-generated and the original feature value. for instance, in table 7, changing the heart rate value from 165 mmhg to 127 mmhg based on the cf explanation changes the model output from 1 to 0. thus, the proximity or distance between the original heart rate value and the cf-generated heart rate value is 6.164414 mmhg. the cf-generated heart rate value is feasible as reducing the patient’s heart rate by a value of 6.164414 mmhg is achievable. 4.7. comparison of the existing work and current study this section compares the existing system and the current heart disease detection model. for comparison, accuracy is considered a performance measure. the comparative result shown in table 10 illustrates that lime and shap are effective in selecting the most influential feature to simultaneously improve the model performance and make the trade-off between the model interpretability and accuracy. overall, this study shows that lime and shap are significantly important to improve the model accuracy and explain the heart disease diagnosis outcome of the ml model. table 10 comparison of the existing and current systems study publication year algorithm employed model explainer method accuracy achieved (%) [1] 2020 xgboost shap 84.98% [3] 2021 cnn shap 99.62% [6] 2021 xgboost shap 99.4% [9] 2020 svm grammar based 96.31% [10] 2021 rf feature selection 99.3% [11] 2021 cnn feature selection 99.74% this study 2022 xgboost shap and lime 99.79% proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 11 5. conclusions in this study, the xgboost model is proposed for heart disease detection. moreover, lime and shap are used to explain the diagnosis outcome of the proposed model. the simulation shows that the proposed model has promising performance as it diagnoses heart disease with an auc score of 0.99. moreover, the study identifies 10 influential features closely related to the risk of heart disease with the help of shap and lime. with shap and lime, the diagnostic outcome of xgboost can be explained to clinicians to better understand the reason behind the model’s diagnosis outcome. the result shows that the performance of the xgboost model improves while making the trade-off between performance and interpretability, which makes the model suitable for clinical practice. the simulation results also show that by applying lime and shap, better xgboost models with reliable results and better performance can be developed. lime provides only local instance-based explanations while shap can explain the model’s prediction outcome at the local and global level. in the future, the researcher plans to test the performance and explainability of the xgboost model by using other clinical datasets. the explainability will be evaluated with the results provided by lime, shap, and dice. conflicts of interest the author declares that there are no conflicts of interest. statement of ethical approval for this type of study, a statement of human rights is not required. statement of informed consent for this type of study, informed consent is not required. references [1] y. meng, et al., “what makes an online review more helpful: an interpretation framework using xgboost and shap values,” journal of theoretical and applied electronic commerce research, vol. 16, no. 3, pp. 466-490, june 2021. [2] s. m. lundberg, et al., “from local explanations to global understanding with explainable ai for trees,” nature machine intelligence, vol. 2, pp. 56-67, january 2020. [3] s. m. lundberg, et al., “explainable machine-learning predictions for the prevention of hypoxemia during surgery,” nature biomedical engineering, vol. 2, no. 10, pp. 749-760, october 2018. [4] m. toğaçar, et al., “detection of covid-19 findings by the local interpretable model-agnostic explanations method of types-based activations extracted from cnns,” biomedical signal processing and control, vol. 71, article no. 103128, january 2022. [5] p. xie, et al., “an explainable machine learning model for predicting in-hospital amputation rate of patients with diabetic foot ulcer,” international wound journal, vol. 19, no. 4, pp. 910-918, may 2022. [6] c. rudin, et al., “interpretable machine learning: fundamental principles and 10 grand challenges,” https://arxiv.org/pdf/2103.11251.pdf, july 2021. [7] a. gramegna, et al., “shap and lime: an evaluation of discriminative power in credit risk,” frontiers in artificial intelligence, vol. 4, article no. 752558, 2021. [8] f. cavaliere, et al., “parkinson’s disease diagnosis: towards grammar-based explainable artificial intelligence,” ieee symposium on computers and communications, pp. 1-6, july 2020. [9] s. j. sushma, et al., “an improved feature selection approach for chronic heart disease detection,” bulletin of electrical engineering and informatics, vol. 10, no. 6, pp. 3501-3506, december 2021. [10] r. bhagwat, et al., “a framework for crop disease detection using feature fusion method,” international journal of engineering and technology innovation, vol. 11, no. 3, pp. 216-228, june 2021. [11] k. wang, et al., “interpretable prediction of 3-year all-cause mortality in patients with heart failure caused by coronary heart disease based on machine learning and shap,” computers in biology and medicine, vol. 137, article no. 104313, october 2021. proceedings of engineering and technology innovation, vol. x, no. x, 20xx, pp. xx-xx 12 [12] t. suresh, et al., “a hybrid approach to medical decision-making: diagnosis of heart disease with machine-learning model,” international journal of electrical and computer engineering, vol. 12, no. 2, pp. 1831-1838, april 2022. [13] r. bharti, et al., “prediction of heart disease using a combination of machine learning and deep learning,” computational intelligence and neuroscience, vol. 2021, article no. 8387680, 2021. [14] s. rana, et al., “slow learner prediction using multi-variate naïve bayes classification algorithm,” international journal of engineering and technology innovation, vol. 7, no. 1, pp. 11-23, january 2017. [15] g. vilone, et al., “notions of explainability and evaluation approaches for explainable artificial intelligence,” information fusion, vol. 76, pp. 89-106, december 2021. [16] m. toğaçar, et al., “intelligent skin cancer detection applying autoencoder, mobilenetv2 and spiking neural networks mesut,” chaos, solitons, and fractals, vol. 144, article no. 110714, march 2021. [17] a. f. anderson, “case study: nhtsa’s denial of dr raghavan’s petition to investigate sudden acceleration in toyota vehicles fitted with electronic throttles,” ieee access, vol. 4, pp. 1417-1433, 2016. [18] t. a. assegie, “evaluation of the shapley additive explanation (shap) technique for ensemble learning methods,” proceedings of engineering and technology innovation, vol. 21, pp. 20-26, april 2022. copyright© by the author. 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/).  proceedings of engineering and technology innovation, vol. 11, 2019, pp. 30-37 feasibility study on using low grade silicon carbide and stone sludge to fabricate eco bricks min-hsin liu * , cyuan-fu tsai, yi-le hsu department of environmental engineering and management, chaoyang university of technology, taichung, taiwan, roc received 28 may 2018; received in revised form 06 june 2018; accepted 06 july 2018 abstract the aim of this study was to introduce 5% up to 40% of low grade silicon carbide (lgsc), with construction excavated soil before the soil being reused as raw material in eco brickmaking. furthermore, stone sludge was added to mix with lgsc to fabricate eco bricks. the results indicated that the use of 10% granite sludge, 30% lgsc and 5% brick waste clinkers could produce grade-2 eco bricks in sintering temperature 1,050°c, however 30% granite sludge was substituted to obtain the same grade eco bricks. water absorption of eco bricks has increased during the addition ratio of lgsc increased. however, the compressive strength increased while water absorption decreased during the operation temperature increased from 1,020°c. to 1,050°c. keywords: silicon carbide, granite sludge, eco bricks, lgsc 1. introduction the increasing of environmental awareness has led to the rise of the solar energy industries. in order to maintain the lubrication of the cutting operations in the production of silicon chips in solar panels and cutting fluids. the mixture of cutting oils and silicon carbide powder, are added during the cutting process. to maintain the lubrication function of the cutting operations, the cutting fluid must be replaced regularly, thus generating large amounts of waste cutting fluid. the waste cutting fluid can become reused waste. through physical separation methods, the waste cutting fluid can be separated from the solid waste. they are mainly consisting of silicon carbide, silicon, and a small amount of diethylene glycol, also known as low grade silicon carbide (lgsc). the silicon carbide particles in the lgsc are relatively stable and do not undergo any changes or deterioration in quality during the cutting process; they account for 46.56% among of the lgsc [1]. element silicon is the main impurity of the lgsc, together with other impurities such as iron, copper, and zinc [2]. to purify and extract the silicon carbide in lgsc, the particles are first filtered through centrifugation and then mixed with the cleaning fluid in hydro cyclone, using the ion-exchange resin method to remove the dissolved solids. however, this method requires several steps, chemicals and produces a large amount of waste water; it also requires the regular cleaning of the ion-exchange resin and other components and thus have a relatively high cost. the main rocks used in the stone processing industry are granite and marble. during the cutting, trimming, grinding, polishing and wastewater treatment process, the stone waste slabs and slag has been produced. the stone wasteful plates and sludge hold similar compositions of the original stone, and the output quantity is large, however the final disposal of the sludge is burial or reuse. whereas stone wasteful plates and sludge can be reused as cement raw materials, concrete raw materials, gravel raw materials, chemical raw materials and building materials. due to the poor sales, the number of resource recycling products is limited. however, lokeshwari and jagadish had used granite fines waste in eco-friendly building blocks [3], and munir et al. demostrated waste marble sludge to incorporate thermally efficient fired clay bricks [4]. * corresponding author. e-mail address: jliu@cyut.edu.tw tel.: +886-910-599767; fax: +886-4-23742365 proceedings of engineering and technology innovation, vol. 11, 2019, pp. 30-37 31 eco bricks are extensively made by the sintering of industrial waste [5], thus it is encouraged to recycle and reuse the industrial waste. in the eco bricks certified by the renewable resources and green product review [6], the added industrial wastes includes ceramic waste, inorganic sludge, construction waste, water purification sludge, and glass waste. pundienė et al. represented the effect of different preparation methods of silicon carbide aggregate on the properties of refractory concrete with cenospheres [7]. however, the current handling methods of lgsc consist only incineration and heat treating, used as a composite deoxidizing agent or a heating agent in steelmaking processes. therefore, the main aims of this research is to study the optimal addition of lgsc. also, the stone wasteful plates/sludge is to construct the excavated soil sintering processes and to make eco bricks for utilizing the characteristics of the silicon carbide; and silicon has improved the quality of such bricks, in turn leading to widespread application of lgsc eco bricks in buildings. the concept of this study using lgsc and stone sludge to fabricate eco-friendly bricks was iluustrated in fig. 1. the grade determination of the bricks complies with the specifications of ordinary bricks pursuant to chinese national standard cns382 in taiwan. in general buildings, grade 3 bricks should have a water absorption lower than 15% and compressive strength higher than 150 kgf/cm 2 ; grade 2 bricks should have a water absorption lower than 13% and compressive strength higher than 200 kgf/cm 2 , and grade 1 bricks should have a water absorption lower than 10% and compressive strength higher than 300 kgf/cm 2 . fig. 1 the concept of this study using lgsc and stone sludge to fabricate eco-friendly bricks 2. materials and methods four phases of experiments were conducted in this study, two kinds of construction excavated soil were selected for the tests. in phase i during the soil characteristics analysis, one of the soil was identified as clay loam (31.5% sand separate, 6.1% silt separate and 62.4% clay separate), the other was classified as silt loam (20% sand separate, 77% silt separate and 3% clay separate). the sources for lgsc were obtained from the solid waste extracted from the waste cutting fluid recycled from two solar wafer plants in taiwan, and were abbreviated as t plant and m plant. inductively coupled plasma optical emission spectrometry (icp-oes, thermo scientific) was then conducted to analyze the metal composition of the lgsc, while the elemental analyzer (elementar vario el iii) was conducted to analyze the element composition of the lgsc. in phase ii, the addition, proportion of the lgsc was then added to the two types of construction excavated soil in the proportions of 5%, 10%, 15%, 20%, 25%, 30%, 35%, and 40%, respectively. after mixing and air drying, the soils with lgsc were then sintered in high temperature ovens to make brick embryos of 5 cm x 5 cm x 1 cm size, the grade of brick embryos was determined to examine the quality. in phase iii, a fixed 5% of waste clinkers from brick factory was added in the process, and the sintering temperature has been increased from 1,020ºc to 1,050ºc to reduce the porosity of the brick embryo samples [8] and increase the strength and grade quality of the bricks. in phase iv of this study, the proportions of stone wasteful plates and sludge 10%, 20%, 30%, 40%, 50% were added to substitute construction excavated soil for brick making, the grades of brick embryos were then determined. https://www.sciencedirect.com/topics/materials-science/silicon-carbide proceedings of engineering and technology innovation, vol. 11, 2019, pp. 30-37 32 3. results and discussion 3.1. analytical results of lgsc (phase i) the results of the metal composition analysis of two plant sources of lgsc are shown in table 1, and the date of element composition analysis is listed in table 2. the major composition of lgsc are silicon and iron, however the aluminum composition of lgsc of m plant (3,560 mg/kg) is much higher than that from t plant (122 mg/kg), which affects the brick quality in a sintering process due to the high reactivity of aluminum. in addition, the carbon composition of lgsc of m plant (13.06%) is higher than that from t plant (1.9%), the results indicate the portion silicon carbide (sic) in m plant is higher than that in t plant. because of the hardness and low thermal expansion coefficient of sic, the compressive strength of bricks is increased. table 1 metal analysis result of lgsc from two plant sources analytical items t plant (mg/kg) m plant (mg/kg) si 537,000 286,000 fe 77,300 40,300 ca 845 1,090 mg 291 296 na 1,310 1,130 k 487 1,030 al 122 3,560 mn 237 291 cd 2.36 1.56 cr 20.1 22.1 cu 1,510 810 ni 41 41 pb 11.3 2.32 zn 366 527 table 2 element analysis result of lgsc from two plant sources element items t plant (%) m plant (%) c 1.9 13.06 n 0.02 0.14 h 0.05 1.69 o 0.53 5.75 s 0.25 0.7 3.2. results by adding lgsc to construction excavated soil (phase ii) fig. 2 grade distribution chart of sintered bricks with different percentages of the lgsc from t plant the lgsc recycled from the two plants are added to the construction excavated soil at percentages of 5%, 10%, 15%, 20%, 25%, 30%, 35%, and 40%. the grade distribution chart of the bricks made for the addition of lgsc from t plant to the clay loam and silt loam is shown in fig. 2 in the experiments, only lgsc were added at percentages of 0%, 5%, and 10% conform to the standards of grade 3 bricks. the bricks without addition of lgsc have a water absorption of 10.9% and compressive strength of 250 kgf/cm 2 and are classified as grade 2 bricks. bricks added with 5% and 10% of lgsc have a proceedings of engineering and technology innovation, vol. 11, 2019, pp. 30-37 33 water absorption of 12.6% and 13.7%, respectively, and both possess a compressive strength of 240 kgf/cm 2 ; thus, they are classified as grade 3 bricks. bricks added with 15% to 40% of lgsc have water absorption that exceeds the grade 3 specifications but the compressive strengths remain within the scope of grade 3 standards. the results show that when lgsc is added to clay loam, as the quantity addition increases, the water absorption increases, and the compressive strength decreases. fig. 2 also shows the experiment results for the addition of lgsc recycled from t plant to silt loam bricks. after the addition of lgsc, the bricks do not meet the standards of grade 3 bricks. the results show that although the bricks have compressive strengths within the scope of the grade 2 standards of 200 kgf/cm 2 , the water absorption data are greater than the 15% standard required by grade 3; therefore, the groups do not conform to the scope of the standards. when the lgsc recycled from t plant is added to the two different construction excavated soils, the bricks show signs of poor water absorption performance, while the bricks sintered from silt loam have better compressive strength than those from clay loam. the grade distribution chart of the bricks made for the addition of lgsc of m plant to the construction excavated soil is shown in fig. 3 when lgsc from m plant is added to the bricks made from clay loam, the bricks show compressive strengths higher than 200 kgf/cm 2 , which is within the scope of class 2 bricks, but poorer water absorption performance. the water absorption of the bricks added with 0%, 5%, and 10% lgsc is 10%, 12.7%, and 11.8% respectively, which is within the scope of grade 2 bricks. the water absorption of bricks added with 15%, 20%, and 30% lgsc is within the scope of grade 3 bricks. bricks with higher percentages of lgsc added do not fall within the scope of grade 3 bricks. when the percentage of lgsc from m plant added to clay loam is higher, the resulting bricks have poorer water absorption performance, but when added with 25% of lgsc, the bricks show a significant increase in compressive strength. when the lgsc from m plant is added to the silt loam, after sintering, only the water absorption data of the four groups of bricks with 0%, 5%, 10%, and 15% lgsc added conform to the specifications of grade 2 and grade 3, while the others fail to meet the standards. however, the compressive strength of the bricks from all groups is higher than the grade 1 standard of 300 kgf/cm 2 , so bricks made from silt loam show a significant advantage in compressive strength than bricks made from clay loam. fig. 3 grade distribution chart of sintered bricks with different percentages of the lgsc from m plant 3.3. results by adding lgsc and brick waste clinker to construction excavated soil (phase iii) the results of the addition of lgsc from t plant and m plant to clay loam and silt loam show that the addition of lgsc increases the water absorption of the brick embryo samples in all the experiments. therefore, in the third phase experiment, a combination addition of 10%-25% of lgsc from t plant or 25%-40% of lgsc from m plant were added, together with 5% brick waste clinkers (bwc), to the construction excavated soil for brick sintering. as the bwc contain defects after sintering, the bwc is first crushed and pulverized into particles, after the particles with a diameter of 0.4 mm or less were added in the brick sintering process. during low temperature sintering, the bwc has low compactness, and the increase in pores results in an increased water absorption. however, at high temperature sintering, the waste clinkers have higher compactness, causing the water absorption to decrease. therefore, the temperature of the brick sintering process in phase ii has increased from 1,020ºc to 1,050ºc. proceedings of engineering and technology innovation, vol. 11, 2019, pp. 30-37 34 the grade distribution chart of the brick embryo samples made for the addition of lgsc from t plant and 5% bwc to the construction excavated soil is shown in fig. 4 the graph shows that the brick samples made from clay loam have a significant drop in water absorptions and an increase in compressive strength. the addition of bwc increases the grades of the bricks to the class 2 standards. the brick embryo samples made from silt loam, also shown a sharp increase in grade levels, with the optimal condition being the addition of 15% lgsc and 5% bwc to the silt loam. the resulting brick is elevated to meet class 1 standards. fig. 4 grade distribution chart of sintered bricks with different percentages of lgsc from t plant and 5% of bwc added to the construction excavated soil the grade distribution chart of the brick embryo samples made by the addition of lgsc from m plant and 5% bwc to construction excavated soil is shown in fig. 5. the graph shows that the bricks embryo made from clay loam exhibit a significant drop in water absorptions while the compressive strengths are not significantly affected. the bricks embryo made with the addition of the bwc belongs to class 2. the bricks embryo made from silt loam shown a drastic drop in water absorption while the compressive strengths are not significantly affected. the bricks embryo made with the addition of 25%, 30%, and 35% lgsc belongs to class 2, while the bricks embryo made with the addition of 40% lgsc meet the standards of class 3 bricks. the results show that the addition of 5% bwc can reduce the water absorption of the bricks embryo made from the silt loam group. although the increase in compressive strength was minor, the grade of the resulting bricks embryo has increased. fig. 5 grade distribution chart of sintered bricks with different percentages of lgsc from m plant and 5% of bwc added to construction excavated soil 3.4. results by adding lgsc and stone sludge to construction excavated soil (phase iv) granite is a common type of felsic intrusive igneous rock that is granular and phaneritic in texture, granite is an igneous rock with between 20% and 60% quartz by volume, and at least 35% of the total feldspar consisting of alkali feldspar. granite is nearly always massive (lacking any internal structures), hard and tough. the average density of granite is between 2.65 and https://en.wikipedia.org/wiki/felsic https://en.wikipedia.org/wiki/intrusive_rock https://en.wikipedia.org/wiki/igneous_rock https://en.wikipedia.org/wiki/phaneritic https://en.wikipedia.org/wiki/quartz https://en.wikipedia.org/wiki/alkali_feldspar https://en.wikipedia.org/wiki/density proceedings of engineering and technology innovation, vol. 11, 2019, pp. 30-37 35 2.75 g/cm 3 , its compressive strength usually lies above 2,000 kgf/cm 2 . herein, granite sludge has been selected to fabricate eco bricks in this study. the composition ratio of 10%-30% lgsc from m plant, 10%-50% granite sludge and the other portions (clay loam and 5% bwc) are mixed to fabricate eco bricks at sintering temperature 1,050ºc. the water absorption and compressive strength of eco bricks are listed in table 3. the grade distribution chart of sintered bricks with 30% lgsc and various percentage of granite sludge without adding bwc and adding 5% bwc are shown in fig. 6 and fig. 7, respectively. the bricks do not meet the standards of grade 3 bricks when the composition ratio of granite sludge is above 30%. the comparison of fig. 6 and fig. 7 indicate that the addition of 5% bwc is insignificant increase of brick grade when granite sludge is adopted. table 3 water absorption and compressive strength by adding lgsc and granite sludge to fabricate bricks composition ratio (%) water absorption (%) compressive strength (kgf/cm 2 ) granite sludge 10%+lgsc 10%+ clay loam 80% 7.48 267.98 granite sludge 10%+lgsc 10%+clay loam 75%+bwc 5% 4.94 230.01 granite sludge 10%+lgsc 20%+clay loam 70% 7.02 240.96 granite sludge 10%+lgsc 20%+clay loam 65%+bwc 5% 7.13 239.04 granite sludge 10%+lgsc 30%+clay loam 60% 11.13 208.18 granite sludge 10%+lgsc 30%+clay loam 55%+bwc 5% 9.22 212.71 granite sludge 20%+lgsc 10%+clay loam 70% 7.53 200.80 granite sludge 20%+lgsc 10%+clay loam 65%+bwc 5% 10.14 199.60 granite sludge 20%+lgsc 20%+clay loam 60% 8.08 254.43 granite sludge 20%+lgsc 20%+clay loam 55%+bwc 5% 7.55 218.08 granite sludge 20%+lgsc 30%+clay loam 50% 8.93 213.55 granite sludge 20%+lgsc 30%+clay loam 45%+bwc 5% 7.23 214.39 granite sludge 30%+lgsc 10%+clay loam 60% 9.16 227.72 granite sludge 30%+lgsc 10%+clay loam 55%+bwc 5% 10.20 217.39 granite sludge 30%+lgsc 20%+clay loam 50% 7.14 215.69 granite sludge 30%+lgsc 20%+clay loam 45%+bwc 5% 12.45 213.97 granite sludge 30%+lgsc 30%+clay loam 40% 9.06 220.00 granite sludge 30%+lgsc 30%+clay loam 35%+bwc 5% 10.82 191.11 granite sludge 40%+lgsc 10%+clay loam 50% 15.82 232.96 granite sludge 40%+lgsc 10%+clay loam 45%+bwc 5% 13.75 257.93 granite sludge 40%+lgsc 20%+clay loam 40% 16.65 209.81 granite sludge 40%+lgsc 20%+clay loam 35%+bwc 5% 14.06 228.17 granite sludge 40%+lgsc 30%+clay loam 30% 17.06 191.86 granite sludge 40%+lgsc 30%+clay loam 25%+bwc 5% 17.03 182.26 granite sludge 50%+lgsc 10%+clay loam 40% 13.75 213.40 granite sludge 50%+lgsc 10%+clay loam 35%+bwc 5% 15.98 210.84 granite sludge 50%+lgsc 20%+clay loam 30% 15.93 196.45 granite sludge 50%+lgsc 20%+clay loam 25%+bwc 5% 15.60 196.84 granite sludge 50%+lgsc 30%+clay loam 20% 17.83 201.85 granite sludge 50%+lgsc 30%+clay loam 15%+bwc 5% 16.93 203.44 fig. 6 the grade distribution chart of sintered bricks with 30% lgsc and various percentage of granite sludge (without adding bwc) proceedings of engineering and technology innovation, vol. 11, 2019, pp. 30-37 36 fig. 7 the grade distribution chart of sintered bricks with 30% lgsc and various percentage of granite sludge (adding 5% bwc) 4. conclusions this feasibility study demonstrates that the lgsc from the solar industry and granite sludge can be used to substitute clay in brick fabrication. when lgsc and granite sludge are introduced in bricks fabrication, the quantity of lgsc addition increases, the water absorption of brick increases, and the compressive strength decreases. however the bricks sintered from silt loam show a significant advantage in compressive strength than bricks made from clay loam. the addition of bwc increase the grades of the bricks to the class 2 standards. the brick samples made from silt loam also show a sharp increase in grade levels. when granite sludge is adopted by brick making, the bricks do not meet the standards of grade 3 bricks when the composition ratio of granite sludge is above 30%, the addition of 5% bwc is increased insignificantly of brick grade when granite sludge is adopted. acknowledgement this study was funded by the ministry of science and technology of taiwan, r.o.c. the views or opinions expressed in this article are those of the writers and should not be construed as the opinions of the ministry of science and technology. mention of trade names, vendor names, or commercial products does not constitute endorsement or recommendation by ministry of science and technology. conflicts of interest the authors declare no conflict of interest. references [1] h. s. gopala and k. murthy, “evolution and present status of silicon carbide slurry recovery in silicon wire sawing,” resources conservation and recycling, vol. 104, part. a, pp. 194-205, 2015. [2] a. bidiville, i. neulist, k. wasmer and c. ballif, “effect of debris on the silicon wafering for solar cells,” solar energy materials and solar cells, vol. 95, no. 8, pp. 2490-2496, 2011. [3] m. lokeshwari m. and k.s. jagadishb, “eco-friendly use of granite fines waste in building blocks,” procedia environmental sciences, vol. 35, pp. 618-623, 2016. [4] m. j. munir, s. m. s. kazmi, y.-f. wu, a. hanif, and m. u. a. khan, “thermally efficient fired clay bricks incorporating waste marble sludge: an industrial-scale study,” journal of cleaner production, vol. 174, pp. 1122-1135, 2018. [5] w. li, g. lei, y. xu, and q. huang, “the properties and formation mechanisms of eco-friendly brick building materials fabricated from low-silicon iron ore tailings,” journal of cleaner production, vol. 204, pp. 685-692, 2018. [6] a. o. babatunde and y. q. zhao, “constructive approaches toward water treatment works sludge management: an international review of beneficial reuses,” environmental science & technology, vol. 37, no. 2, pp. 129-164, 2007. proceedings of engineering and technology innovation, vol. 11, 2019, pp. 30-37 37 [7] i. pundienė, a. korjakins, j. pranckevičienė, and m. kligys, “effect of silicon carbide aggregate, prepared by different methods, on the properties of refractory concrete with cenospheres,” ceramics international, vol. 44, pp. 15944-15953, 2018. [8] q. huang, and z. jin, “the high temperature oxidation behavior of reaction-bonded silicon carbide,” journal of materials processing technology, vol. 110, no. 2, pp. 142-145, 2001. 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.sciencedirect.com/science/journal/02728842  proceedings of engineering and technology innovation, vol. 8, 2018, pp. 23 31 the study of tibicos fermentation product as pilot model of healthy drinks hsiao-han liu 1,* , bo-shun tu 1 , yew-loom chen 2 , heng-long wang 3 1 department of biological science & technology, i-shou university, kaohsiung, taiwan, roc. 2 department of hospitality management, tajen university, pintung, taiwan, roc. 3 department of life sciences, national university of kaohsiung, kaohsiung, taiwan, roc. received 31 july 2017; received in revised form 25 august 2017; accept ed 10 sept ember 2017 abstract sugar kefir grains (tibicos) are a symbiotic culture of bacteria and yeasts which were assembled by various strains of microbes attached polysaccharide bracket composed of white transparent particles. initially we used chemical colorimetric method to measure that of carb ohydrate concentrations, and brad ford method for that of protein concentrations. we utilized black sugar as carbohydrate source and medium to cultivate tibicos, and collected media after 0 hr., 24 hr., and 48 hr. different time points. these collected tib icos grains and fermentation broth were measured with the above methods . in addition, we used gram stain to observe the inside microbial populations of tibicos. the results showed that the weight of cultured grains with brown sugar solution at 25 ℃ and sealed for 48 hr increased to 45% as that of original grains. the components of tibicos fermentation broth were analyzed by hplc for ions as following: lactic acid, chloride ion, malate, sulfate, oxalic acid, phosphate, citrate, and other minor ingredients . furthermore, we used the gram stain to observe the microbial composition within tibicos grains. the more detail identification of microbial population in tibicos was done with bacteria and yeasts. the bacteria parts used pcr amplification with the 16s rdna primer sets (533r and 341fgc), gel purification, fragments sequencing, and alignment with 16s ribosomal rna sequences in ncbi database (16s rdna of bacteria and archaea). the result sequences were assigned as bacillus strain as following: (a) bacillus circulans, (b) bacillus eiseniae, (c) bacillus oceanisediminis, (d) bacillus atrophaeus, (e) bacillus siralis, (f) bacillus massiliosenegalensis. the yeast parts were first isolated as single colony, purified the chromosomal dna, and amplifie d by pcr method with 18 s rdna primer set (fr1 and ns1), then gel purified, dna sequenced, and aligned in ncbi database (nucleotide collection). the identified yeast strains as following: (a) sporobolomyces k oalae, (b) meyerozyma guilliermondii, (c) aureobasidium pullulans. keywords: tibicos , health drink, fermentation broth, 18s rdna sequence, 16s rdna sequence 1. introduction sugar kefir grains (tibicos) are assembled by various strains of microbes attached polysaccharide bracket composed of white transparent particles. these microbes included bacteria and yeast, which generated the organic acids and some intergradient that could improve the heath state of people [1]. there are tibicos all over the world, but because o f the weather * corresponding author. e-m ail address: hansliu@ isu.edu.tw proceedings of engineering and technology innovation, vol. 8, 2018, pp. 23 31 copyright © taeti 24 and climate in different regions of the latitude of their differentiated; however, tibicos are not all the same, either which found that there are two of the same area with the same culture conditions. tibicos could adapt to the environment and become diverse that mixed bacteria were highly viable bacteria. there are some folktales that tibicos fermentation broth (brown sugar kefir) can prevent insomnia and reduce high blood pressure, high fat, weight, but we try to investigate these legends . and there is little literature to investigate tibicos; we found that there were some literatures about western kefir cultured in milk that could be a good parallel model to explore. for example: some western kefir strain are a combination of lactobacillus, pediococcus, streptococcus, leuconostoc bacteria, and from yeast, candida, kloeck era [2]. and other possible yeast strain also exist in water kefir, such as candida, kloeck era. so, it is interesting whether that taiwan common tibicos strains are like the combination of western kefir strains or not? generally, tibicos fermentation broth are made from yeast, lactic acid bacteria, brewer's yeast etc., and tibicos from differ ent taiwan area can be made from many fermentation broth which contain different sugar-containing liquids (such as brown sugar, sucrose, glucose, etc.), and these different carbohydrates are used to supply nutrients to feed it to produce lactic acid, al cohol (ethanol) and carbon dioxide gas as a carbonate-containing drink. with the change of the times, the more developed of science and technology, people's lives are more and more comfortable, and have multiple choice of food and drinks, but have the more risk of choice with nutrient imbalance. the health drinks may be the good s olution for this imbalance of food choice [3-4]. furthermore, could we find the same effect on tibicos’ drink? to develop a pilot model for health drink with tibicos worth our attention to this novel and tradition health drink. since ancient times, tibicos has been fermented to drink for a long time, so far, some scholars have been suspected that tibicos ingredients on the human body has no practical help, as well as the spread by internet and the effect on the street stalls . however, we try to explore with scientific experiments to confirm the results with the tibicos fermentation broth composition analysis and identification of microbes. we can have better understanding of tibicos and the scientific data that prove the value of tibicos. it is worthy our attention that the future development of health drinks with tibicos will be a great mission for us to investigate further functions of these healthy drinks . 2. method 2.1. the culture, preparation and gram stain of tibicos to prepare 15 gram brown sugar and to add this distilled water 200 ml into a wide open glass bottle, the brown sugar water must be mixed evenly, and put 35 gram of tibicos grains into glass bottle; do not stir. in further experiment, it fermented 0h, 24h, 48h, and collected metabolic solution. change the brown sugar water every day; to stop the proliferation of tibicos, just to rinse this grains with the distilled water, and placed in the container into the refrigerator. use the mesh cloth to keep this tibicos with air. do not use metal containers. to crush tibicos grains after smear on the slides, air dry and fix the microbes. stain with crystal purple staining, add to t he slide with appropriate (covered with bacteria coated) crystal violet dye solution for 1 minute. wash and dip the dyeing solution and rinse carefully with water. drop iodine solution for 1min. wash the iodine solution with water. decolor the slide with 95 % ethanol for 20-25s to the effluent without color, wash immediately. drop the safranin dye 5min. and wash the dirt on the smear with water. dry the stained smear slides and put the air to dry or dry with absorbent paper. observe under microscopic examination with low magnification, and then high -power, and finally observed with oil mirror, and determine the morphology of the microbes [5]. proceedings of engineering and technology innovation, vol. 8, 2018, pp. 23 31 copyright © taeti 25 2.2. gel electrophoresis of the microbes’ genomic dna pcr fragment weigh out the appropriate mass of agarose into an erlenmeyer flask. agarose gels are prepared using a w/v percentage solution. the concentration of agarose in a gel will depend on the sizes of the dna fragments to be separated, with most gels ranging between 0.5%-2%. the volume of the buffer should not be greater than 1/3 of the capacity of the flask. add running buffer to the agarose-containing flask. swirl to mix. the most common gel running buffers are tae (40 mm tris -acetate, 1 mm edta). melt the agarose/buffer mixture. this is most commonly done by heating in a microwave, but can also be done over a bunsen flame. at 30 s intervals, remove the flask and swirl the contents to mix well. repeat until the agarose has completely dissolved. add ethidium bromide (etbr) to a concentration of 0.5 μg/ml. alternatively, the gel may also be stained after electrophoresis in running buffer con taining 0.5 μg/ml etbr for 15-30 min, followed by de-staining in running buffer for an equal length of time. allow the agarose to cool either on the benchtop or by incubation in a 65 °c water bath. place the gel tray into the casting apparatus. alternatively, one may also tape the open edges of a gel tray to create a mold. place an appropriate comb into the gel mold to create the wells. pour the molten agarose into the gel mold. allow the agarose to set at room temperature. re move the comb and place the gel in the gel box. then setup the apparatus and electrophoresis, and observe the result with a gel documentation system [6]. 2.3. the molecular analysis of bacterial and yeast dna in tibicos the 16 s rdna of the tibicos’ bacteria was amplified by pcr usin g 533r and 341fgc primer set. the pcr results were separated by the dgge method and dna gel electrophoresis. after sequencing, the target gene was amplified by pcr using 533r and 341f again. send the pcr results to the sequence. compare the results in the ncbi database [7]. isolate and purify fungal colonies in tibicos’ grain. stain colonies and identification. isolate single colony and extract chromosomal dna. the 18s rdna of the cells was amplified by pcr using fr1 and ns1 primers. collect and purify th e pcr dna products from the dna gel electrophoresis. after sequencing, compare the results in the ncbi database [8]. 2.4. the carbohydrate and protein analysis of tibicos’ metabolite in time course dilute the tibicos solution to 1 % (0 hr., 24 hr., and 48 hr.). take 50μl of the test solution and add 500μl of 95% sulfuric acid, shake and mix well. add 50 μl of 5% phenol, shake and mix well. take 200μl, measured od492. take the value into the equation; get the value from the standard curve. these samples were measured and converted into accurate units [9]. take 200 μl of the tibicos solution (0 hr., 24 hr., 48 hr) with 200 μl bradford reagent for 10 min. observe under od595 absorbance and convert the od value with the protein standard curve to the protein con centration [10]. 2.5. the hplc analysis of tibicos’ metabolite follow the dionex ionpac® as11 manual 5.7 for gradient separation of krebs cycle acids. the time course samples were injected using the autosampling system (spectra system as3500, thermo, fin nigan) by means of an ion liquid chromatography ionpac® as11 2 × 250 mm column, dionex. the organic acid was separated using a 5-100 mm naoh gradient eluent, and the anion concentration in the culture medium was finally analyzed with conductivity analysis [11]. 3. results and discussion since tbicos were traditional drinks long time ago[1], we established the platform to develop it into health drinks. in fig. 1, we can observe the tibicos drinks in fig. 1(c) and kefir grains in figs. 1(a), (b), and (d). we test the tibicos under different temperature and carbohydrates, for example: glucose, sugar and black sugar (data not shown). to compare the tibicos and kefer proceedings of engineering and technology innovation, vol. 8, 2018, pp. 23 31 copyright © taeti 26 grains, we can find similar morphology of grains. however, the black sugar and room temperature were the best condition for the culture of the tibicos . (a) general tibicos grains (b) the grains under microscope (c) a tradition tibicos drinks (d) the general kefir grains which were grown in milk fig. 1 the tibicos, drinks and kefir grains from the observation the smearing of the grain under microscope and gram stain, we can find that the grains form the tibicos were gram positive cocci (figs. 2(a), (b), and (c)), gram negative rod-like bacilli (fig. 2(d)), gram negative cocci (fig. 2(e)), and gram positive bacilli (fig. 2(f)). these results will be the first observation of the tibicos grains by gram stain, and there were gram positive cocci and bacilli, and gram negative cocci and bacilli. we can compare the similar results in kerfir grains [12]. (a) gram positive cocci (b) gram positive cocci (c) gram positive cocci (d) gram negative rod-like bacilli (e) gram negative cocci (f) gram positive bacilli fig. 2 by gram staining, the tibicos grains are showed both gram negative and positive stain mix population proceedings of engineering and technology innovation, vol. 8, 2018, pp. 23 31 copyright © taeti 27 we used molecular method to identify the different strains of bacteria (fig. 3(a)) and yeast (fig. 3(b), (c)), and the primer sets used in this experiment were 341 to 533r for bacteria 16s rdna sequences , ns-1 to fr1, fr3 for yeast 18s rdna sequences. the gel electrophoresis for pcr results show that near 200 base pairs fragment in all six bacteria strains (fig. 3(a)), and yeast 1.6 kb fragments of t1, t2 (fig. 3(b)), t3 (fig. 3(c)). the specific sites of sequences in primer sets were shown in fig. 3(d). (a) pcr fragment of 6 bacteria strains with 553r to 341f primer set (b) yeast ns1 to fr1 pcr fragments with dna marker, isolate t3, t2,t1 (c) isolate t3, t2, blank, dna marker (d) the 16s rrna bacteria genome map and pcr primer 553r to 341f sites and the yeast 18s rrna ns1 (17-36), fr1 (1664-1649), fr2 (1747-1727) primers fig. 3 results of gel electrophoresis the results of molecular analysis for bacteria (table 1) and yeast (table 2) were compared with the ncbi databank of 16s rdna of bacteria and archaea and ncbi database (nucleotide collection) respectively. the bacteria parts used pcr amplification with the 16s rdna primer sets (533r and 341fgc), gel purification, fragments sequencing, and alignment with 16s ribosomal rna sequences. the result sequences were assigned as bacillus strain as following: (a) bacillus circulans, (b) bacillus eiseniae, (c) bacillus oceanisediminis, (d) bacillus atrophaeus, (e) bacillus siralis, (f) bacillus massiliosenegalensis. the yeast parts were first isolated as single colony, purified the chromosomal dna, and amplified by pcr method with 18 s rdna primer set (fr1 and ns1), then gel purified, dna sequenced, and aligned in ncbi database. the identified yeast strains as following: (a) sporobolomyces k oalae, (b) meyerozyma guilliermondii, (c) aureobasidium pullulans. table 1 descript max score total score query cover ident a bacillus circulans strain atcc 4513 16s ribosomal rna gene, partial sequence 268 268 96% 96% b bacillus eiseniae strain a12 16s ribosomal rna gene, partial sequence 281 281 97% 98% c bacillus oceanisediminis strain h2 16s ribosomal rna gene, partial sequence 281 281 97% 98% d bacillus atrophaeus 1942 strain 1942 16s ribosomal rna gene, partial sequence 283 283 97% 98% e bacillus siralis strain 171544 16s ribosomal rna gene, partial sequence 287 287 95% 98% f bacillus massiliosenegalensis strain jc6 16s ribosomal rna gene, partial sequence 281 281 90% 100% proceedings of engineering and technology innovation, vol. 8, 2018, pp. 23 31 copyright © taeti 28 table 2 descript max score total score query cover ident a sporidiobolus salmonicolor gene for 18s ribosomal rna, partial sequence 217 217 60% 95% b meyerozyma guilliermondii strain 15h4-po-p5-2 18s ribosomal rna, gene, partial sequence 710 710 100% 97% c aureobasidium pullulans gene for 18s rrna, partial sequence, strain 486 486 99% 99% the results of tibico metabolites’ carbohydrate and protein analysis in time course were shown in fig. 4. from these results, we could notice that after 24 hours, the carbohydrate in metabolite increased about 49% to that of 0 hr and decreased 28 % to that of 0 hr after 48 hours. the protein content in metabolite increased 16.6% (24 hr) and 18.7% (48 hr) to that of 0 hr. (a) the carbohydrate content change in tibicos’ metabolite with time course (b) the protein content change in tibicos’ metabolite with time course fig. 4 results of the carbohydrate and protein analysis in tibicos’ metabolite with time course the results of the hplc analysis in tibicos’ metabolite with time course were shown respectively in following: the profile of lactic acid concentration in fig. 5(a), the profile of cl, no3, malate, sulfate, oxalic, phosphate, citrate concentration in fig. 5(b), the profile of lactic acid, cl, sulphate concentration in fig. 5(c), the profile of tibicos ’ metabolite in 24 hr. with lactic, cl, malate, sulfate, oxalic, phosphate, citrate concentration in fig. 5(d), the profile of tibicos ’ metabolite in 48 hr. with lactic, cl, malate, sulfate, oxalic, phosphate, citrate concentration in fig. 5(e). the final time courses were listed in the table of fig. 5(f). we find that the concentration of lactic acid increased 4.1 fold after 24 hours, and 4.3 fold after 48 hours to that of 0 hr, the concentration of chloride ion increased 2.4 fold after 24 hours, and 2.5 fold after 48 hours to that of 0 hr, the concentration of malate, oxalic acid, citrate and phosphate was newly produced after 24 and 48 hours, the concentration of sulphate has a little change. (a) profile of lactic acid (b) profile of cl, no3, malate, sulfate, oxalic, phosphate, citrate fig. 5 results of the hplc analysis in tibicos’ metabolite with time course proceedings of engineering and technology innovation, vol. 8, 2018, pp. 23 31 copyright © taeti 29 (c) profile of lactic acid, cl, sulfate (d) profile of tibicos ’ metabolite in 24 hr with lactic, cl, malate, sulfate, oxalic, phosphate, citrate (e) profile of tibicos ’ metabolite in 48 hr. with lactic, cl, malate, sulfate, oxalic, phosphate, citrate (f) table of organic acid contents with time course fig. 5 results of the hplc analysis in tibicos’ metabolite with time course (continued) 4. conclusions in this paper, the content of lactic acid, malate, phosphate and citrate in the metabolic solution of tibicos was discus sed. the concentration of protein and organic acid was increased obviously, which may play an important role in the health drink development. in addition to the kefir culture study, kefir was found to be cultured in milk or cultured in cocoa, while the environment of kefir was cultivated in milk [12], and compared to our tibicos for environmental testing. in this study of sucrose, brown sugar, and glucose, it is found that tibicos is s uitable for culturing in brown sugar, and this environmental optimization would be related to local different species of tibicos or kefir. the original brown sugar water contained only 4.123 mg / μl lactic acid, while adding tibicos in this solution, after 24hr. t his solution increased to 17.044 mg / μl lactic acid. and after 48 hr., the concentration of lactic acid in tibicos metabolic solution was 17.728 mg / μl. the long-term drinking of tibicos metabolic solution can allow the human body to ingest some lactic acid, this could give our digestive system with benefits [13]. no citric acid was detected in original brown sugar solution, but after adding tibicos grains for 24 hr., this solution contained 0.011 mg / μl citric acid, and after 48 hr. of fermentation, this citric concentration of tibicos solution increased to 0.013 mg /μl. furthermore, literature showed that the body intake of the citric acid would h elp the body to degrade the body fat by inhibiting the accumulation of tissue fat. this probiotics solution can also help in the growth of the body which would promote the body metabolism [14]. on the market, there are also a lot of citric acid derived health food, and stimulate the human body with this nutritional content [15]. no malic acid was found in original brown sugar solution, after proceedings of engineering and technology innovation, vol. 8, 2018, pp. 23 31 copyright © taeti 30 adding tibicos grains for 24 hr., this metabolites contained 0.052 mg /μl malic acid, after fermentation for 48 hr., the concentration of malic acid in tibicos metabolites increased to 0.056 mg/μl. malic acid would activate human pulmonary artery and improve t o lower the body high blood pressure [16]. in this study, the aureobasidium pullulans were isolated from yeast in the tibicos grains. this strain was isolated from the postharvest pathogens and was belonged to valuable fungi [17]. we isolated bacillus circulans from tibicos grains, however, this bacillus circulans would increase the plasmin production and help to prevent thrombosis -related diseases [18]. we also isolated bacillus oceanisediminis from tibicos, and this bacillus belonged to the gram-positive bacilli oceanisediminis, which was first discovered in the south korean coastal marine sediments [19]. meyerozyma guilliermondii was also isolated form tibicos, and this yeast has the spore germination which would have antibacterial activity, and often used in bread preservative s that will extend the shelf life of bread [20]. this yeast was also added in the soy sauce which extend the freshness of its flavor [21]. references [1] t. c. shiu, “the pilot study of tibicos as a model to develop healthy drinks,” master thesis, dept. of biological science and technology, i-shou university, taiwan, 2012. [2] a. g. j. stadie, m. wenning, m. a. ehrmann, and r. f. vogel, “the microbial diversity of water kefir,” international journal of food microbiology, vol. 151, no. 3, pp. 284-288, december 2011. [3] a. g. friques, c. m. arpini, i. c. kalil, a. l. gava, m. a. leal, m. l. porto, et al., “chronic administration of the probiotic kefir improves the endothelial function in spontaneously hypertensive ra ts,” journal of translational medicine, vol. 13, no. 1, p. 390, december 2015. [4] f. r. maciel, g. r. punaro, a. m. rodrigues, c. s. bogsan, m. m. rogero, m. n. oliveira , et al., “immunomodulation and nitric oxide restoration by a probiotic and its activity in gut and peritonea l macrophages in diabetic rats,” clin nutr, vol. 35, no. 5, pp. 1066-1072, october 2016. [5] c. c. kuo and c. m. chao, “gram stain for vibrio species,” the american journal of emergency medicine, vol. 31, no. 5, pp. 877-878, may 2013. [6] r. freitag, encyclopedia of separation science, i. d. wilson, 1st ed., oxford: academic press, 2000, pp. 1201-1208. [7] c. schabereiter-gurtner, s. maca, s. rölleke, k. nigl, j. lukas, a. hirschl, et al., “16s rdna-based identification of bacteria from conjunctival swabs by pcr and dgge fingerprinting,” investigative ophthalmology & visual science, vol. 42, no. 6, pp. 1164-1171, may 2001. [8] y. t. hoshino and s. morimoto, “comparison of 18s rdna primers for estimating fungal diversity in agricultural soils using polymerase chain reaction-denaturing gradient gel electrophoresis,” soil science and plant nutrition, vol. 54, no. 5, pp. 701-710, october 2008. [9] l. s. magwaza and u. l. opara, “analytical methods for determination of sugars and sweetness of h orticultural products -a review,” scientia horticulturae, vol. 184, pp. 179-192, march 2015. [10] t. zor and z. selinger, “linearization of the bradford protein assay increases its sensitivity: theoretical and experimental studies ,” analytical biochemistry, vol. 236, no. 2, pp. 302-308, may 1996. [11] a. c. m. de sena aquino, m. s. azevedo, d. h. b. ribeiro, a. c. o. costa, and e. r. amante, “validation of hplc and ce methods for determination of organic acids in sour cassava starch wastewater,” food chemistry, vol. 172, pp. 725-730, april 2015. [12] c. puerari, k. t. magalhães, and r. f. schwan, “new cocoa pulp-based kefir beverages: microbiological, chemical composition and sensory analysis,” food research international, vol. 48, no. 2, pp. 634-640, october 2012. [13] a. m. o. leite, m. a. l. miguel, r. s. peixoto, p. ruas -madiedo, v. m. f. paschoalin, b. mayo, et al., “probiotic potential of selected lactic acid bacteria strains isolated from brazilian kefir grains,” journal of dairy science, vol. 98, no. 6, pp. 3622-3632, june 2015. [14] r. barczynska, k. jochym, k. slizewska, j. kapusniak, and z. libudzisz, “the effect of citric acid-modified enzyme-resistant dextrin on growth and metabolism of selected strains of probiotic and other intestinal bacteria, ” journal of functional foods, vol. 2, no. 2, pp. 126-133, april 2010. [15] b. van de wier, j. m. balk, g. r. m. m. haenen, d. giamouridis, j. a. bakker, b. c. bast, et al., “elevated citrate levels in non-alcoholic fatty liver disease: the potential of citrate to promote radical production,” febs letters, vol. 587, no. 15, pp. 2461-2466, august 2013. proceedings of engineering and technology innovation, vol. 8, 2018, pp. 23 31 copyright © taeti 31 [16] b. sharif kashani, p. tahmaseb pour, m. malekmohammad, n. behzadnia, f. sheybani-afshar, m. fakhri, et al., “oral l-citrulline malate in patients with idiopathic pulmonary arterial hypertension and eisenmen ger syndrome: a clinical trial,” journal of cardiology, vol. 64, no. 3, pp. 231-235, september 2014. [17] a. di francesco, l. ugolini, l. lazzeri, and m. mari, “production of volatile organic compounds by aureobasidium pullulans as a potential mechanism of action against postharvest fruit pathogens,” biological control, vol. 81, pp. 8-14, february 2015. [18] d. yogesh and p. m. halami, “a fibrin degrading serine metallo protease of bacillus circulans with α -chain specificity,” food bioscience, vol. 11, pp. 72-78, september 2015. [19] h. j. kim, y. j. lee, h. jeong, d. w. lee, and s. j. lee, “insights into diversity and s pecificity of heavy metal resistance and efflux systems in bacillus oceanisediminis 2691,” new biotechnology, vol. 31, supplement, p. s154, july 2014. [20] r. coda, c. g. rizzello, r. di cagno, a. trani, g. cardinali, and m. gobbetti, “antifungal activity of meyerozyma guilliermondii: identification of active compounds synthesized during dough fermentation and their effect on long -term storage of wheat bread,” food microbiology, vol. 33, no. 2, pp. 243-251, october 2013. [21] p. aryuman, s. lertsiri, w. visessanguan, n. niamsiri, a. bhumiratana, and a. assavanig, “glutaminase-producing meyerozyma (pichia) guilliermondii isolated from thai soy sauce fermentation,” international journal of food microbiology, vol. 192, pp. 7-12, january 2015. proceedings of engineering and technology innovation, vol. 12, 2019, pp. 35-38 time series analysis of outcomes for small and medium enterprises’ support of regional industry dong-gu lee, je-doo ryu, keon-seok nam, kyoung-nam ha * korea institute of industrial technology, busan, korea received 14 may 2018; received in revised form 26 june 2018; accepted 26 july 2018 abstract continued investment in research and development (r&d) by the government and corporate support played a major role in the rapid growth of the republic of korea. in 2017, the amount of r&d investment by the government accounted for 19.7 trillion won (krw), which is 4.7 % of the government’s budget. the government’s r&d budget is increasing by 2.5 % each year. we analyzed the outcomes of the projects, supported by the busan regional companies conducted in 2016 to 2017. for a time series analysis, we gathered companies support amount by year, sales after company support, and material for employment. we used ibm’s statistical package for the social sciences (spss) 18 for correlation analysis. we confirmed the relationship between employment and the sales of the annual company support program. keywords: time series analysis, regional industry, support smes, spss statistics 1. introduction the industrial technology is recognized as an important element of national competitiveness and the korean government has implemented various policies for corporate support. in particular, it supports the growth of multidimensional companies through r&d and supports smes. prior corporate support policies prioritized securing technology competitiveness of companies based on r&d. however, as a final goal of recent corporate support projects, commercialization support is expanding. the republic of korea started a regional industrial promotion policy in 1999 with the aim of revitalizing the regional economy. to nurture regional strategic industries, the government promoted "policies that intensively support smes belonging to regional strategic industries". recently, it focused on policies that induce cooperation and competition among regions. the forms of corporate support for regional industry support projects are divided into infrastructure, technology, technology development (r&d), technical support, commercialization support, and human resources development (non-r&d). results of r&d projects have clear technical development results, but it is difficult to evaluate the outcome of the company support project as employment and sales. in this paper, we analyzed the business support of busan area specialized industry in 2015 and 2016, and try to analyze what correlation the enterprise support project has with increasing employment and sales of enterprise. 2. previous research many researchers present the direction of desirable enterprise support projects through outcome analyses of existing enterprise support projects. [1] conducted an analysis of the efficiency of enterprise support projects for the whole southern * corresponding author. e-mail address: 0vincent@kitech.re.kr proceedings of engineering and technology innovation, vol. 12, 2019, pp. 35-38 36 area’s specialized industry development business. [2] conducted an influential factor analysis on the outcome of the support of enterprises by industry of the regional industry promotion policy. [3] demonstrated that the effectiveness of the corporate support project is being implemented in results at a time lag through efficiency analysis of the government enterprise support project through dea analysis. 3. analysis of correlation pearson's correlation analyses were conducted on the results of the enterprise support project of the intelligent machine parts industry, which is a non-r&d business in the region's main industry development project that was conducted in 2015 and 2016. in 2015 support projects, there were multiple programs, such as prototype creation, product luxury, exhibition support, brand, bottleneck technology, and certification for package support. support program in 2016 consisted of prototype creation, product upgrades, exhibition support, and marketing support. using ibm spss 18, we analyzed the data of the two years and defined the variables shown in table 1. table 1 definitions of variables variable definition of variable unit support project costs through support projects, the supported amount thousand won new employment new employment during the support period people domestic sales in commercialization domestic sales generated during the support period thousand won overseas sales in commercialization overseas sales generated during the support period thousand won fig. 1 frequency analysis of support project costs table 2 new employment new employment 2015 2016 frequency percent frequency percent 0 117 84.8 48 53.53 1 9 6.5 22 24.4 2 8 5.8 12 13.3 3 2 1.4 2 2.2 4 1 0.7 2 2.2 5 1 0.7 1 1.1 6 0 0 3 3.3 total 138 100 90 100 proceedings of engineering and technology innovation, vol. 12, 2019, pp. 35-38 37 first, we analyzed the frequency of the support project costs in each fiscal year. as a result of the frequency analysis, the support project cost was under 10 million won because the exhibition and marketing supports were relatively frequent. table 2 shows the results of frequency analysis of support projects. table 2 shows the frequency analysis of new employment, 40 new employees were hired in 2015, and 83 new employees were hired in 2016. however, more than 50 % were not hired. table 3 summarizes the average value of the support costs, new employment, domestic, and overseas sales in commercialization. the average cost of the support projects in 2015 was 8.3 million won and the average domestic sales in commercialization were confirmed at 13 million won. the average cost in 2016 was 9.9 million won, and the average sales of commercialization were confirmed at 28 million won. table 3 descriptive statistics 2015 2016 total average n total average n support project costs (million won) 1,145 8.3 138 891.3 9.9 90 new employment 40 0.29 138 83 0.92 90 domestic sales in commercialization 1,884 13.658 138 2,522 28 90 overseas sales in commercialization 180 1.305 138 1,139 12.6 90 tables 4-5 show the results of the correlation analysis. by correlation analysis, we analyzed the correlation between yearly support project cost and new employment, sales of domestic in commercialization, and commercialization of overseas. table 4 2015 correlation analysis support project costs new employment domestic sales in commercialization overseas sales in commercialization support project costs pearson correlation coefficient 1 .416 .066 .016 significant probability .000 .440 .849 n 138 138 138 138 new employment pearson correlation coefficient .416** 1 .118 -.009 significant probability .000 .168 .915 n 138 138 138 138 domestic sales in commercialization pearson correlation coefficient .066 .118 1 .453** significant probability .440 .168 .000 n 138 138 138 138 overseas sales in commercialization pearson correlation coefficient .016 -.009 .453** 1 significant probability .849 9.15 .000 n 138 138 138 138 ** please note that the correlation coefficient is 0.01 level (both sides). table 5 2016 correlation analysis support project costs new employment domestic sales in commercialization overseas sales in commercialization support project costs pearson correlation coefficient 1 .303** .068 -1.26 significant probability 0.004 .524 .238 n 90 90 90 90 new employment pearson correlation coefficient .303** 1 .092 .103 significant probability .004 .389 .333 n 90 90 90 90 domestic sales in commercialization pearson correlation coefficient .068 .092 1 -0.56 significant probability .524 .389 .602 n 90 90 90 90 overseas sales in commercialization pearson correlation coefficient -.126 .103 -.056 1 significant probability .238 .333 .602 n 90 90 90 90 ** please note that the correlation coefficient is 0.01 level (both sides). proceedings of engineering and technology innovation, vol. 12, 2019, pp. 35-38 38 in 2015, the support costs seemed to correlate positively (+) with new employment (r = 0.416). in addition, the domestic and overseas sales in commercialization showed a positive (+) correlation with quantity. in 2016, the support costs had a positive correlation (+) with new employment (r = 0.303), but the domestic and overseas sales in commercialization did not have significant correlations. 4. conclusions in this paper, we analyzed the correlations between sales and employment of the support amount of the company support business for two years. the analysis of the outcome of the company support business for two years, the support costs of the company and new employment have a high correlation. it was confirmed that the direct and indirect employment of the company is done according to the company support amount. however, the frequency analysis of the overall employment shows that considerably lower employment was being conducted. we confirmed that it was still difficult to increase employment, just only by the supporting project costs. generally, it was judged that there was a correlation between the support costs of a company and the domestic sales in commercialization. however, in 2015, it was confirmed that the 2016 data analysis results of the sales in commercialization were not strongly correlated with the support costs of the company. it is difficult for the actual results of the prototype production support to occur in that year. we analyzed the results of the same support project for two years. the results show that it is necessary to have a performance index that could grasp the substantial influence on the company's ability and productivity, rather than being evaluated on the basis on financial performance indicators, such as employment and sales. conflicts of interest the authors declare no conflict of interest. references [1] s. s. lee and y. e. kim, “efficiency analysis of enterprise support project for nurturing specialized industries in jeonnam area,” 2016 korean institute of industrial engineers, symposium, 2016, pp. 34-40. (in korean) [2] g. h. kim and y. h. won, “an analysis of performance influence factors of enterprise support based upon the local industrial development policy: focused on the industry types of region-specialized industries,” korean association for policy science, 2015. (in korean) [3] y. s. kang, s, m, sohn, s. j. im, and b, y, cho, “a study with dea analysis on the efficiency of the government support for small and medium enterprises,” korea academic society of business administration, 2014, pp. 1891-1907. (in korean) 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/). proceedings of engineering and technology innovation, vol. 7, 2017, pp. 08 13 the influence of density of the materials and body mass index on the elderly’s sleep efficiency wichan torobrum * , nandh thavarungkul, preecha termsuksawad, nakorn srisukhumbowornchai, purit thanakijkasem division of materials technology, school of energy, environment and materials , king mongkut’s university of technology thonburi, bangkok, thailand . received 11 july 2017; received in revised form 25 august 2017; accept ed 30 august 2017 abstract insomnia is a co mmon proble m a mong the elders and it d irectly a ffect s their health. the mattress is one factors influencing sleeping e fficiency. the developed pressure at the sleeping surface during sleeping will to cause body discomfort, leading to sleep problem. the developed pressure depends on the density of bedding materia ls and body mass. the present research aims to study the effects of the mattress density and body mass on the elderly sleep efficiency. the sleep data was collected from 6 v olunteers: 3 male and 3 fe ma le at the age group of 60 – 79 years. the e xpe riment was done on mattresses with densities of 24.6 and 60 kg/ m 3 and 4-inch thic kness. the volunteers randomly slept on each mattress for three nights. sleep data was recorded by sleep trac ker (jawbone up24), calibrated with polysomnography. the study revealed that sleep efficiency on the 60 kg/ m 3 density mattress was more than 90 percent while those on the 24.6 kg/m 3 density mattress were in the range of 78 84 percent. moreover, sleep effic iency of the volunteers with the body mass index of 22.22 23.74 kg/cm 2 was significantly diffe rent fro m those of volunteers with the body mass index of 24.61 -24.92 kg/c m 2 when they slept on the 24.6 kg/ m 3 density mattress. additionally, the pressure test revealed that a pressure at shoulder may affect sleep efficiency. keywor ds : pressure, sleep efficiency, elderly, body mass index 1. introduction undeniably, ind ividuals’ hygiene and health is related to a balance between rela xation and daytime activ ities. most of us spend one-third of our life in sleeping [1-2]. one factor that effects on our health is sleep quality, which can be indicated by sleep efficiency: a ratio of total sleep time to total time in bed. there are many factors affecting sleep effic iency such as physical, mental, e mot ional, and environ mental factors [3]. it is found that 7 percent of american people, especially e lders, have sleep problems, resulted by their mattresses [4]. the reason why elders have the sleep problem is deterioration and ma lfunction of cells and organs , like backbone [5]. the inappropriate sleep postures, pressure on sleeping surface and body mass of obese elder apparently cause shorter sleep time, lead ing to low sleep effic iency. conversely, those with norma l body mass index would have higher sleep effic iency and slow wave sleep (sws) than those of the obese ones [6 -7]. in addit ion, pressure on sleeping surface, depending on mattress density, is believed to be re lated to lo w bac k d iscomfort, pain, or stiffness and shoulder pain [8]. according to the previous study, pressure between the sleeping surface and skin of 30 mmhg is suggested to be the acceptable upp er limit to assure that capillary blood perfusion to soft tissue can be maintained [9-10]. * corresponding author. e-mail address: wichan1001@hotmail.com tel.: +668-8-5225586 proceedings of engineering and technology innovation, vol. 7, 2017, pp. 08 13 copyright © taeti 9 9 however, there is no in-depth study on the relationships between the pressure on the sleeping surface and sleep effic iency. the pressure on the sleeping surface depends on the mattress density and body mass laid on the mattresss. according to the previous studies, a med iu m-firm mattress in which the layers of materia l had been laid is recommended to obtain good sleep quality [11-14]. however, the mediu m-firm mattress is not clearly defined. it depends on types of material and individual physical characteristics. this work, therefore, aims to study the effects of the mattress density and body mass on the sleep efficiency, and relate these factors to the pressure on t he sleeping surface. 2. methods the equipments used in this work we re pressure mapping fsa version 4.0, vista medica l co mpany and sleep tracker (jawbone up 24). the pressure mapping is used to measure pressure on the sleeping surface during sleeping. the jawbone up 24 is used to measure sleep data such as sound sleep time , light sleeping time, t ime in bed (tib), number of a waken ings and total sleep time (tst ). the data obtain s fro m the ja wbone up 24 is co mpared with those obtain ed from polysomnography(psg) using sample size of 5 and the relationship between these data is shown by linear regression as shown in figure 1. fig. 1 relationship between sleep efficiency obatined from jawbone up and psg fig 1 shows the sleep efficiency obtained from ja wbowe up 24 co mpared with that obtained from psg. th is figure shows e xcellent linear re lationship between these value. the regression equation shows that sleep efficiency measured by jawbone up is about 6.0 percent lower than that obtained from psg. the mattress es used in this work are 24.6 kg/m 3 density polyurethane foa m and the 60 kg/ m 3 density reborn foa m. the sizes of both mattresses are 90 x 200 square centimeter and 4-inch thickness. six e lderly volunteers who have no sleep problems and have normal body mass are chosen in this test. to conduct the sleep test, the mattresses would be covered with thin fabric before conducting the test. all volunteers would randomly sleep in their houses with different mattress es for 3 nights each. prior to e xperiments, each volunteer will be asked about his/her sleep behavior. on the test day, volunteers must avoid any caffeine, a lcoholic drinks and perfu mes. the time to sleep is scheduled as the same as their norma l sleep time. the volunteers are requested to take a bath and to wear casual comfort dresses as their pleasure. the data collected are sleep onset latency, time in bed, number of a wakenings and total sleep t ime . the sleep effic iency is calcu lated fro m the following equation: 100 total sleep time sleep efficiency time in bed   (1) for pressure measurement, the pressure is measured three times for each material. all volunteers will lie on two mattresses with a pressure mapping in between for both supine and lateral positions. the highest pressure at the shoulders and hip has been chosen to find their relationships with sleep efficiency. proceedings of engineering and technology innovation, vol. 7, 2017, pp. 08 13 copyright © taeti 10 3. results 3.1. effects of mattress density on sleep efficiency fig 2 demonstrated the sleep efficiency of six volunteers sleeping on two mattresses with different density. fig. 2 sleep efficiency measured by jawbone up 24 fro m the figure, the sleep effic iencies on the 60 kg/ m 3 density mattress are higher than those on the 24.6 kg/ m 3 density mattress. using the pair t-test, the sleep efficiency on 60 kg/ m 3 mattress is higher than that on 24.6 kg/ m 3 mattress significantly with significant level o f 0.05. the higher sleep effic ienc y on the higher density mattress may be exp lained by the fact that the higher density mattress cause less body sink, which results in less spine bending. the spine bending causes back muscle tautness and an increase of lu mba r spine pressure. these phenomen a leads to back pain and resulting in low total sleep time. as indicated by gabriel et.al, sleeping on the soft sleeping surface is proven to cause low back pain [15]. it should be pointed out that sleep efficienc ies on the 60 kg/ m 3 mattress are about 90 percent or higher. as mention earlier, sleep e ffic iency measured by jawbone up is about 6 percent less than that measured by psg; there fore, the sleep efficiency on 60 kg/ m 3 density mattress is about 96-97 percent if measured by psg. this value is very hig h. therefore, the 60 kg/m 3 density mattress would be recommended for elders to gain good sleep efficiency during sleeping. fig. 3 number of awakening of volunteers on different mattresses fig. 3 shows number of awa kening of 6 volunteers who had slept on the 24.6 and 60 kg/ m 3 density mattresses. form the figure, they have woken up 3 to 5 times per night when they sleep on the 24.6 kg/m 3 density mattress and range of 0 to 3 times per night when they sleep on the 60 kg/ m 3 density mattress. the result confirms that sleeping on the 24.6 kg/ m 3 density mattress causes more discomfort than sleeping on the 60 kg/m 3 density mattress. proceedings of engineering and technology innovation, vol. 7, 2017, pp. 08 13 copyright © taeti 11 11 3.2. effects of pressure on the sleeping surface on sleep efficiency to study the relationship between sleep efficiency and pressure, the pressure mapping is used to measure pressure between the sleeping surface and skin in both supine and lateral positions. an e xa mple of pressure profile during sleeping in lateral position is shown in fig 4. the profile shows both number and color code to indicate the magnitude of pressure on the sleeping surface during sleeping. the pressure, used to relate with sleep effic iency, is the ma ximu m p ressure measured with their individual preferred position. fig. 4 pressure profile measured during sleeping with lateral positioning. fig. 5 pressure at shoulder and sleep efficiency of volunteers sleeping on the 24.6 kg/m 3 and 60 kg/m 3 density mattresses according to fig. 5, the pressure at shoulder might be related to sleep effic iency. apparently, the sleep effic iency is 90 percent or higher when pressure at shoulder is at least 17.6 mmhg . co mpared with the p revious suggestion, which states that the pressure of approximate ly 30 mmhg is the upper pressure limit that blood could norma lly flo w and nourish organs and tissues of cells, fig. 5 indicates that pressure at shoulder, can increase up to 35 mmhg, wh ich is more than the suggested upper pressure limit. fig. 6 pressure at hip and sleep efficiency of volunteer sleeping on the 24.6 kg/m 3 and 60 kg/m 3 density mattress proceedings of engineering and technology innovation, vol. 7, 2017, pp. 08 13 copyright © taeti 12 in addition, the result shows that for pressure at shoulder of lowe r than 17.6 mmhg , the lo wer the pressure, and the lower is the sleep efficiency. it should be noted that pressure at shoulder of lo wer than 17.6 mmhg only occurs when volunteers sleep on the 24.6 kg/ m 3 density mattress. it can be concluded that pressure at shoulder on the low density mattress is low and resulting in low sleep effic iency. this means that pressure at shoulder may effected on human discomfort during sleeping. in contrast to fig. 5, pressure at hip does not show any relationship with sleep efficiency as shown in fig. 6. 3.3. effect of body mass index on sleep efficiency the sleep efficiency might be effected by other factors such as age, weight and height, [10]. the refore, the relat ionships among body mass index (bm i), mattress density and the sleep efficiency was studied. f or this reason, the volunteers is classified into two groups: those of the bmi of 22.22 -23.74 kg/cm 2 and of 24.61-24.92 kg/c m 2 , respectively. the relat ionship among body mass inde x, mattress density and sleep effic iency occuring on the 24.6 and the 60 kg/ m 3 density mattress is shown in fig 7. fig. 7 bmi and sleep efficiency of volunteer sleeping on the 24.6 kg/m 3 and 60 kg/m 3 density mattresses the independent t-test shows that bmi related to sleep effic iency when sleeping on the 24.6 kg/ m 3 density mattress but not on the 60 kg/m 3 density mattress with significance level of 0.05. the statement means that bmi is less related to sleep effciency when sleeping on the high density mattress. according to definition of bmi, it represents the ratio of mass to square of height. it is used to define body fatness. for the low density mattress, a sma ll increase of body mass index or bod y fatness causes more body sink into the mattress. as a result, body spine bends and potenially lead to back pain and discomfort. the figure also shows that for low density matteress, the higher the bmi o f volunteer, the lowe r is the sleep efficiency. the result is in an agreement with the study by vorona et. al. [16], who indicates that the total sleep time of one who has high body mass index is lower than the one whose body mass index is normal. 4. conclusions the present study demonstrated that mattress density and body mass index e ffected on sleep efficiency. sleeping on the 60 kg/ m 3 density mattress provided higher sleep efficiency than on the 24.6 kg/ m 3 density mattress for all volunteers. the body mass inde x was found to affect sleep e ffic iency only when volunteers slept on the 24.6 kg/ m 3 mattress. the study also showed that pressure at shoulder may a ffect sleep effic iency when volunteers slept on low density mattress but on high density mattress. references [1] d. k. fontaine, “sleep and the critically ill patient,” in : m. r. kinney, d. r. packa, and s. b. dunbar, aacn’s clinical reference for critical care nursing, 3rd ed. pp. 351-364, st. louis: mosby, 1993. proceedings of engineering and technology innovation, vol. 7, 2017, pp. 08 13 copyright © taeti 13 13 [2] l. a. hodgson, “why do we need sleep : relating theory to nursing practice,” journal of advanced nursing, vol. 16, no. 2, pp. 1503-1510, 1991. [3] c. a. miller, “nursing care of olderly adult : theory and practices”, 3 rd ed. philadelphia: lippincott. [4] r. addison, m. j. thorpy, and t. roth, “a survey of the united states public concerning the quality of sleep,” sleep research, vol. 16, p. 244, 1986. [5] l. ayalon and s. ancoil-islael, sleep: a comprehensive handbook, new jersey: jonh wisley & sons, 2006. [6] w. moraes, d. poyares, i. zalcman, et al., “association between body mass index and sleep duration assessed by objective methods in a representative sample of the adult population,” sleep med icine, vol. 14, no. 4, pp. 312-318, april 2013. [7] t. shochat, g. shefer-hilel, and a. zisberg, “relationships between body mass index and sleep quality and duration in adults 70 years and older,” sleep health , vol. 2, no. 4, pp. 266-271, december 2016. [8] s. sigman and j. richmond, “office diagnosis of shoulder disorder,” physician sports med icine, vol. 23, no. 7, pp. 2531, 1995. [9] r. gardner and w. i. grossman, “normal motor patterns in sleep in man ,” in: e. d. weitzman, “advances in sleep research,” spectrum publications, vol. 2, pp. 67-107, 1975. [10] e. j. bridges, o. s. joseph, and s. mazer, “skin interface pressure on the nato litter,” military medicine, vol. 168, no. 4, pp. 280-286, 2003. [11] b. h. jacobson, h. a. gemmell, b. m. hayes, and t. s. altena, “effectiveness of a selected bedding system on quality of sleep, low back pain, shoulder pain, and spine stiffness ,” journal of manipulative and physiological therapeutics, vol. 25, no. 2, pp. 88-92, 2002. [12] b. h. jacobson, t. wallace, and h. gemmell, “subjective rating of perceived back pain, stiffness and sleep quality following introduction of medium-firm bedding systems,” journal of chiropractic medicine, vol. 5, no. 4, pp.128-134, 2006. [13] b. h. jacobson, a. boolani, and d. b. smith, “changes in back pain, sleep quality, and perceived stress after introduction of new bedding systems ,” journal of chiropractic medicine, vol. 8, no. 1, pp. 1-8, march 2009. [14] b. h. jacobson, a. boolani, g. dunklee, a. shepardson, and h. acharya, “effect of prescribed sleep surfaces on back pain and sleep quality in patients diagnosed with low back and shoulder pain,” applied ergonomics, vol. 42, no. 1, pp. 91-97, december 2010. [15] g. o. ayeni, o. olaniyi, and o. a. ayeni, “sleep habits and occurrence of lowback pain among craftsmen,” international journal of public health science, vol. 3, no. 2, pp.101-106, june 2014. [16] r. d. vorona, m. p. winn, t. w. babineau, b. p. eng, h. r. feldman, and j. c. ware, “overweight and obese patients in a primary care population report less sleep than patients with a normal body mass index,” archives of internal medicine, vol. 165, no. 1, pp. 25-30, january 2005. microsoft word 5-peti#3938 39-44.docx proceedings of engineering and technology innovation, vol. 16, 2020, pp. 39 44 densification behavior, microstructure, and property of zno with multi-oxides glass addition at different sintering temperatures boen houng 1,* , wei-chueh chien 1 , chen-kai sun 2 1 department of materials science and engineering, i-shou university, kaohsiung, taiwan 2 emerging display technologies corporation, kaohsiung, taiwan received 15 march 2020; received in revised form 31 may 2020; accepted 23 june 2020 doi: https://doi.org/10.46604/peti.2020.3938 abstract zinc oxide (zno) mixed with various amount of glasses were sintered at different temperatures. the densification behavior of zinc oxide with glass addition and its microstructure and dielectric constant sintered from 900° to 1200°c have been investigated. a unique glass composition contained geo2, moo3, and v2o5 (gmv) was designed to act as the sintering aid to enhance the densification and to adjust the dielectric constant of zno. the effect of sintering temperature on the densification behavior and dielectrics properties of zno was investigated by dilatometer, x-ray diffractometer, scanning electron microscopy and lcr meter. the glass additive formed a thin continuous liquid phase and rearranged zno particles into a dense microstructure at relatively low temperature. the dielectric constants of glass added zno ceramics were found to vary with the glass concentration and sintering temperature. keywords: zno, glasses, densification behavior, dielectric constant 1. introduction zno has been extensively studied because of its remarkable chemical, optical and electronic properties which give a large perspective to its practical applications such as gas sensors, surface acoustic wave devices, optical waveguides blue/uv light emitting devices as well as photocatalysts [1-7]. in addition, zno with high dielectric constants and low dielectric loss have received special attention due to the rapid developments in microwave telecommunications, satellite broadcasting, discrete and multilayer capacitors, dynamic random access memories and low-loss substrates for microwave integrated circuits. [8-9]. moreover, the development of the low-temperature co-fired ceramic technology (ltcc) becomes an important fabricating technology that allows the incorporation of the multilayer structure into a monolithic bulk module with minimal processing steps, thus leading to thin and compact communication devices. therefore, the ltcc with high electrical conductivity metallization such as silver has been identified to be a feasible solution for applications in the area of wireless communications [10]. large numbers of materials have been studied with excellent dielectric properties, however, most of them need to be sintered at a higher sintering temperature greater than 1200°c and longer soaking time to achieve high enough density. to solve this problem, low-melting oxides such as bi2o3, moo3, cuo, and mno2-cr2o3-sb2o3-co3o4 are generally mixed with dielectric ceramics to reduce the sintering temperature [11-15]. on the other hand, by using nano-sized particles as starting ceramic materials has also shown an improvement in reducing sintering temperature [16-17]. however, these * corresponding author. e-mail address: boyen@isu.edu.tw proceedings of engineering and technology innovation, vol. 16, 2020, pp. 39 44 40 approaches are far from the requirement for ltcc that a firing temperature below 900°c is essential. since the metal ag, melted at 961°c, has been used as a conductive electrode in ltcc, thus it is required to develop low-firing temperature ceramics with desirable dielectric properties. applying a liquid-phase sintering process, refractory ceramics can be sintered at a lower temperature by forming thin continuous or semicontinuous liquid phases at grain boundaries [18-19]. during sintering, the ceramic particles are soluble in the liquid. this solubility causes the liquid to wet the solid, leading a capillary force that pulls the particles together. at the same time, the high temperature softens the solids, further promoting densification. high-diffusion rates are also associated with liquids, resulting in fast sintering or lower sintering temperatures. in addition, zno with high dielectric constants and low dielectric loss have received special attention due to the rapid developments in microwave telecommunications, satellite broadcasting, discrete and multilayer capacitors, dynamic random access memories and low-loss substrates for microwave integrated circuits. in this work, we have designed a unique glass composition containing geo2, moo3, and v2o5 as a sintering aid that allows the zno to densify at many low temperatures by viscous flow and liquid-phase sintering mechanisms. on the other hand, this glass additive can also adjust the dielectric constant of zno by varying the concentration of glass additive. the effect of the glass concentration on the densification behavior, microstructure and dielectric properties of zno were investigated. 2. experimental the glass composition used as a sintering aid was made up of geo2, moo3, v2o5 in 1:1:3 molar proportion (all with purity ≥99.99%, aldrich, st. louis, usa). the traditional twin-roller quencher is used to prepare glass additives. the starting powders were first mixed to form a 60 g batch and then melted in a 90% pt-10% rh crucible at 1000°c for 30 minutes using an electric furnace. the melts were stirred to ensure the homogeneity of the glass composition. the molten glasses were subsequently quenched into a twinned roller, yielding thin ribbons of approximately 0.2 mm thickness. the resulting glasses were fully amorphous, as was confirmed by x-ray diffractometry. the ribbon glasses were then ground into powders (<325 mesh) and mixed with zno powders in the proportion of 1 to 10 wt % using ethanol solvent and zirconia milling media for 24 h. after drying, the mixtures were die-pressed at 80 mpa to yield several disk type pellets (12 mm in diameter and 3 mm in thickness). the pellets were then sintered at 800, 900, 1000, and 1100°c for 4 h with a heating rate of 5°c/min. the bulk density of samples was measured by the archimedes method. an x-ray diffractometer (xrd, panalyical, x’pert pro) with cu kα radiation (λ= 0.1542 nm) was used to characterize the crystallization of the sintered ceramics. fractured surfaces of the sintered samples were examined using field emission scanning electron microscopy (fesem, philips, xl-40feg). x-ray energy dispersive spectroscopy (eds) was used to obtain the chemical compositions of the zno grain. shrinkage behavior of the samples during heating from room temperature to 1300°c was measured using a horizontal loading dilatometer with alumina rams and boats (dil-402c, netzsch instrument). the sintered samples were electroded with dc-sputtered films of gold on both sides for dielectric property measurement using an impendence analyzer (agilent, 4263b, palo alto, usa). 3. results and discussion the shrinkage history of zno various gmv glass concentrations as a sintering aid was examined by dilatometer as shown in fig. 1. it can be seen from the diagram that as the number of glass additives increased shrinkage curves were shifted towards much lower temperatures than the typical sintering temperature of pure zno. pure zno appears to start shrinkage at a slow rate approximately 1200°c. nevertheless, zno ceramics with 1 and 5 wt % glass additives showed the apparent shrinkage above 900°c. as the number of glass additives increased to 10 wt%, zno exhibited a large shrinkage of proceedings of engineering and technology innovation, vol. 16, 2020, pp. 39 44 41 approximately 8 % at 900°c. the results suggest that gmv glass is an effective sintering aid for zno. the differential thermal analysis (dta) curve revealed that the gmv glass softened at 435°c and melted at 608°c, respectively. the decrease of sintering temperature was caused by the viscous liquid phase effect of gmv glass and served as a bond for the zno body, therefore the densification of zno occurred at much lower temperatures. fig. 1 the shrinkage behavior of the zno with and without gmv glass additives fig. 2 x-ray diffraction patterns of zno with different gmv glass concentrations xrd patterns of zno with different concentrations of gmv glass is shown in fig. 2. it is found that the main crystal phase of the sample is wurtzite zno at a glass doping concentration of 1 wt %. however, in addition to the wurtzite zno phase, zn3v2o8 phase was also detected as glass additives increased to 3 wt% according to jcpds 34-378. the formation of zn3v2o8 might be attributed to the reaction between the zno matrix and the liquid glass phase. because of the composition of glass contained a greater amount of v2o5, zno matrix reacted with liquid glass to precipitate a stable zn3v2o8 phase during sintering processing. fig. 3 is the cross-sectional fesem micrographs of zno ceramics sintered at 900°c for 1 hour with 0, 1, 5, and 10 wt% gmv glass. the sample without glass additives shows an obvious porous microstructure with the smallest grain size. increasing the glass additives greatly promoted the densification and the grain size of zno. the bulk density and the grain size of zno as a function of glass concentrations is illustrated in fig. 4. it can be seen that the density increased from 4.45 to 5.05 g/cm 3 and the grain size increased from 11 to 17 µm as glass concentration increased from 0 to 10 wt %. fig. 3 fesem images of zno with gmv glass concentrations at 0, 1, 5, and 10 wt % fig. 4 the bulk density and the grain size of zno as a function of gmv glass concentration proceedings of engineering and technology innovation, vol. 16, 2020, pp. 39 44 42 in addition, the sintering temperature also has a significant effect on the microstructure development of zno, especially at higher gmv glass concentrations. fig. 5 are fesem images of zno with 10 wt % gmv glass sintered from 800 to 1100°c for 4 h. it can be seen that the morphologies of the zno are homogeneous without any abnormal grain growth. the grain size increased from approximately 9-10 µm at 800°c to 30-40 µm at 1100°c. this increase in densification and grain size is attributed to the dissolution of smaller particles and the growth of larger particles during sintering by material transfer through glass liquid phase. fig. 6 is quantitative energy dispersive x-ray spectrometer (eds) analyses of the zno crystalline phase. the results revealed the presence of ge, mo, and v ions in zno crystalline phase, indicating that the zno shows a certain limited solubility in the liquid at the sintering temperature; the essential part of such lower sintering process is the dissolution and re-precipitation of zno particles to give increased grain size and densification. fig. 5 fesem images of zno with 10 wt% of gmv glass concentrations sintered at (a) 800, (b) 900, (c) 1000, and (d) 1100°c fig. 6 the energy dispersive x-ray spectra (eds) of zno crystalline phases table 1 the dielectric properties of zno doped with different gmv glass concentrations gmv glass concentration (wt%) εγ (theoretical) εγ (measured) tan δ (dielectric loss) 0 19 12 0.096 1 25 20 0.085 5 29 23 0.066 10 27 25 0.035 the theoretical and the measured dielectric properties of the zno with different concentrations of gmv glass sintered at 900 °c are summarized in table1. the theoretical dielectric constant, εγ, was calculated using the logarithmic mixing rule [20] of eq. (1) with the data of zno (ε1 = 38), zn3v2o8 (ε2 = 31), and gmv glass (ε3 = 12). 1 1 2 2 3 3log log log logε ε ε εγ ν ν ν+ += (1) where v1, v2, and v3 represent the volume fractions of zno, zn3v2o8, and glass phase in the mixture, respectively. whereas, the dielectric constant of gmv glass was measured after the as-quenched ribbon was heat-treated at 900°c. the volume fractions of zno, zn3v2o8, and glass phase in the samples were determined using the method developed by chipera1 et al. according to the crystalline peak intensity present in the xrd pattern [21]. the measured dielectric constant of zno was found to agree well with the theoretical values. as the gmv glass concentration increased to 10 wt %, the dielectric constant showed an increase to approximately 25. the increase of the dielectric constant in the sample with gmv glass concentrations is due to the improvement of densification of zno ceramics. the presence of glass phases in high gmv concentration samples decreased the dielectric constant of zno. while the dielectric loss almost linearly decreased with the proceedings of engineering and technology innovation, vol. 16, 2020, pp. 39 44 43 increase of gmv glass concentrations. as well known, the value of the dielectric loss is mainly caused by the extrinsic effect that is related to the microstructure of materials such as porosities, grain boundaries, grain sizes and impurities [22]. in this work, zno ceramics doped with a high concentration of gmv glass showed a better microstructure in term of homogeneous grain size and less porosity, therefore resulting in lower dielectric losses. 4. conclusions this research demonstrated that the glass additives containing geo2, moo3, and v2o5 has an effect on lowering the sintering temperature of zno from 1100 to 800°c as a combination of viscous flow and liquid phase sintering effects. this allows the ltcc processing of zno ceramic for telecommunications application to be feasible. the densification and the grain size of zno were found to increase with the glass concentration and sintering temperature as a result of the mass transfer through the liquid phase. the dielectric constants of zno decreased to 25 with the glass concentration increased to 10 wt % because of the presence of the lower dielectric constant of glass phase. while the dielectric loss of zno decreased with the glass concentration due to the denser and homogenous microstructure of zno. in this study, the glass additive not only can be used as the sintering aid to enhance the densification of zno but also be able to modify the dielectric constant of zno by varying the glass concentration. in summary, the multiple oxides glass-doped zno ceramics have attained optimal properties at glass concentration of 10 wt% with a dielectric constant of 25 and dielectric loss of 3.5% as a result of high density and larger grain size associated with less porosity. conflicts of interest the authors declare no conflict of interest. references [1] d. peinsipp and a. h. pecina, “temperature behavior of zno varistors before and after post sintering heat treatment,” materials letters, vol. 43, no. 5-6, pp. 295-302, may 2000. [2] r. kumar, o. al-dossary, g. kumar, and a. umar, “zinc oxide nanostructures for no2 gas-sensor applications: a review,” nano-micro letters, vol. 7, no. 2, pp. 97-120, 2015. [3] a. m. lerer, p. e. timoshenko, and t. y. chernikova, “dispersion characteristics of optical waveguides based on zno nanorods,” journal of communications technology and electronics, vol. 61, no. 6, pp. 619-623, june 2016. [4] j. bao, m. a. zimmler, and f. capasso, “broadband zno single-nanowire light-emitting diode,” nano letters, vol. 6, pp. 1719-722, 2006. [5] y. h. chiu, k. d. chang, and y. j. hsu, “plasmon-mediated charge dynamic and photoactivity enhancement for audecorated zno nanocrystals,” journal of materials chemistry a, vol. 6, pp. 4286-4296, 2018. [6] j. m. li, y. t. wang, and y. j. hsu, “a more accurate, reliable method to evaluate the photoelectrochemical performance of semiconductor electrode without under/over estimation,” electrochimica acta, vol. 267, pp. 141-149, march 2018. [7] j. m. li, h. y. cheng, y. h. chiu, and y. j. hsu, “zno-au-sno2 z-scheme photoanodes for remarkable photoelectrochemical water splitting,” nanoscale, vol. 8, pp. 15720-15729, 2016. [8] y. j. cheng, l. bai, g. yu, and x. zhang, “effect of particles size on dielectric properties of nano-zno/ldpe composites,” materials, vol. 12, pp. 1-14, january 2019. [9] m. d. p. ahmad, a. v. rao, k. s. babu, and g. n. rao, “effect of particles size on dielectric properties of nanozno/ldpe composites,” materials chemistry and physics, vol. 224, pp. 79-84, 2019. [10] s. nishgaki and j. fukuta, “low-temperature cofirable multilayered ceramic bearing pure ag conductors and their sintering behavior,” advanced in ceramics. vol. 26, pp. 199-215, 1989. [11] s. bernik, l. cheng, m. podogar, and g. li, “low-temperature sintering of zno-bi2o3 based varistor ceramics for enhanced microstructure development and current-voltage characteristics,” ceramics-silikáty, vol. 62, pp. 8-14, 2018. [12] s. yu, b. tang, s. zhang, and x. zhou, “effect of zno ratio on sintering behavior and microwave dielectric properties of bao-zno-tio2 ceramics,” journal of alloys and compound, vol. 505, pp. 814-817, 2010. proceedings of engineering and technology innovation, vol. 16, 2020, pp. 39 44 44 [13] l. x. pang, h. wang, and d. z. yao, “a new temperature stable microwave dielectric with low-firing temperature in bi2moo6-tio2 system,” journal of alloys and compound, vol. 493, pp. 626-629, 2010. [14] c. l. huang, r. j. lin, and h. l. chen, “microwave dielectric properties and microstructures of cuoand zno-doped laalo3 ceramics,” material research bulletin, vol. 37, pp. 449-457, 2002. [15] a. bouchekhlal and farida hobar, “effect of sintering temperature on microstructure and nonlinear electric characteristics of zno varistor,” journal of advanced dielectrics, vol. 8, pp. 1850014-1-1850014-5, 2018. [16] e. savary, s. marinel, h. colder, c. harnois, f. x. lefevre, and r. retoux, “microwave sintering of nano-sized zno synthesized by a liquid route,” powder technology, vol. 208, pp. 521-525, 2011. [17] m. mazaheri, a. m. zahedi, and s. k. sadrnezhaad, “microwave dielectric properties and microstructures of cuoand zno-doped laalo3 ceramics, two‐step sintering of nanocrystalline zno compacts: effect of temperature on densification and grain growth,” journal of the american ceramic society, vol. 91, pp. 56-63, 2008. [18] y. q. wua, y. f. zhang, g. pezzotti, and j. k. guo, “effect of glass additives on the strength and toughness of polycrystalline alumina,” journal of european ceramic society, vol. 22, no. 2, pp. 159-164, february 2002. [19] e. dudnik, a. shevchenko, a. ruban, v. redko, and l. lopato, “low-temperature synthesis of α-al2o3,” powder metallurgy, vol. 47, pp. 379-383, 2008. [20] s. k. kim, s. h. shim, s. kim, and s. o. yoon, “low temperature and microwave dielectric properties of tio2/zbs glass composites,” ceramic international, vol. 36, no. 5, pp. 1571-1575, july 2010. [21] s. j. chipera and d. l. bish, “fitting full x-ray diffraction patterns for quantitative analysis: a method for readily quantifying crystalline and disordered phases,” advances in materials physics and chemistry, vol. 3, pp. 47-53, 2013. [22] r. c. pullar, s. j. penn, x. wang, i. m. reaney, and n. m. alford, “dielectric loss caused by oxygen vacancies in titania ceramics,” journal of european ceramic society, vol. 29, no. 3, pp. 419-424, february 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/).  proceedings of engineering and technology innovation, vol. 13, 2019, pp. 20-25 vision technology based traffic safety analysis using signal data wonho suh * , jung ryul kim, joongmin cho department of transportation and logistics engineering, hanyang university erica campus, ansan, korea received 26 april 2019; received in revised form 26 may 2019; accepted 26 june 2019 abstract recent vision technology allows traffic engineers to analyze traffic safety based on image processing applications. vehicle trajectory data including vehicle position and speed are extracted and converted into traffic conflicts related variables. these variables are analyzed to find correlations with crash data and geometric design variables and included in the statistical analysis process to evaluate the safety of intersections. in this paper, signal timing information data is included in the analysis process to investigate if there are any correlations with traffic conflict data. for example, traffic conflicts happen more frequently in a certain movement at certain signal time phase. the goal of this paper is to develop a method for vision technology-based traffic safety analysis process using traffic signal data to identify crash prone movement and signal to time at a given intersection. the proposed technique is demonstrated in real-world video data collected in an intersection. this paper is expected to provide more insight and technique in traffic safety evaluation. keywords: traffic safety, image processing, vision technology, vehicle detecting, safety evaluation 1. introduction while the majority of traffic accidents are caused by drivers' error, traffic accidents occur for various reasons including driver-related factors, vehicles related factors, and environmental related factors including traffic conditions. a significant number of traffic accidents have been found to occur at intersections [1-3]. an intersection is a point where traffic flows running in different directions. vehicles merge, diverge, and intersect each other at an intersection. usually, an intersection with heavy traffic is controlled by a signal. the signalized intersection has the advantage of smoothly controlling the vehicles approaching from different access roads through the signaling device. conventional road safety assessments evaluated safety based on historical accident data. however, in order to collect accident data for statistical analysis, significant time, and cost are required [4]. therefore, safety evaluation based on conflict data has been developed and utilized in many countries. recent image detection technology has been developed to obtain vehicle trajectory information based on image data at an intersection [5-7]. vehicle trajectory data including vehicle position and speed are extracted and converted into traffic conflicts related variables. these variables are analyzed to find correlations with crash data and geometric design variables and included in the statistical analysis process to evaluate the safety of intersections. in this paper, signal timing information data is included in the analysis process to investigate if there are any correlations with traffic conflict data. for example, traffic conflicts happen more frequently in a certain movement at certain signal time phase. the goal of this paper is to develop a method for vision technology-based traffic safety analysis process using traffic signal data to identify crash prone movement and signal to time at a given intersection. * corresponding author. e-mail address: wonhosuh@hanyang.ac.kr tel.: +82-31-400-5154; fax: +82-31-436-8147 proceedings of engineering and technology innovation, vol. 13, 2019, pp. 20-25 21 2. data collection and analysis video data were collected from an intersection in korea and vehicle trajectory data were processed and extracted (fig. 1). right turn on red is permitted in this intersection and cycle length of the intersection was 140 seconds. fig. 1 data collection at an intersection different types of conflicts were identified. in this study, merging conflicts with right turn vehicles (fig. 2) and conflicts between vehicles moving in the same direction (fig. 3) were collected and analyzed. the severity of conflicts was calculated based on the vehicle position and speed data. also, the conflict data were analyzed based on the timing data of the intersection signal to find the correlations between the conflict and signal timing. fig. 4 demonstrates the workflow in this research. fig. 2 example of conflicts in merging traffic fig. 3 example of conflicts between vehicles moving in the same direction proceedings of engineering and technology innovation, vol. 13, 2019, pp. 20-25 22 fig. 4 research workflow 3. results conflicts between the vehicles moving in different direction were analyzed as shown in fig. 4. it was found that a significant number of conflicts occurred in signal phase two and four. most of the conflicts were conflicts between left-turning vehicles and through vehicles (fig. 6). however, conflicts between right-turning vehicles and through vehicles were relatively spread out in the entire signal timing period. fig. 5 conflicts between the vehicles moving in a different direction a total of 69 conflicts occurred between vehicles. among those 69 conflicts, there were 46 conflicts between left turning and right turning vehicles, and 23 conflicts between through moving vehicles and right turning vehicles. of the 46 conflicts that occurred between the left turning and the right turning vehicles, the largest number of conflicts, 18 cases occurred during phase 2, representing about 39%. subsequently, 13 and 11 conflicts occurred in phase 4 and phase 1, respectively, and the lowest level of conflict occurred in 4 cases in phase 3. proceedings of engineering and technology innovation, vol. 13, 2019, pp. 20-25 23 a total of 23 conflicts were found between through moving vehicles and right turning vehicles. this number is only 33 percent of the conflicts occurred between the left turning and the right turning vehicles. considering the total number of through moving vehicles, it is clearly seen that conflicts occurred more often between the left turning and the right turning vehicles compared to conflicts between the through moving vehicles and the right turning vehicles. fig. 5 demonstrates conflicts between the vehicles moving in the same direction. it is clearly seen that most of the conflicts occurred in phase 3 while there were almost no conflicts during phase 1 and phase 2. also, more conflicts occurred in the middle and later part of phase 3 which implies that there might be correlations between the conflict and signal phase and timing. it has been known that vehicle speed and its standard deviation have correlations with accident and traffic conflict. it is believed that the speed of the vehicle affects the accident severity as well in the event of a collision. the greater the variation in the speed of the vehicle, the greater the probability of an accident. it was assumed that the difference was coming from the speed difference of vehicles. for example, vehicles passing the intersection during the latter half of phase tend to have higher speed than vehicles passing the intersection during the first half of phase since vehicles in the queue at the intersection were gradually dissipated during the first half of phase, therefore vehicles have lower speed during the first half of phase meaning less chance of having conflicts. to verify the hypothesis average vehicle speed and its standard deviation were calculated for the vehicles passing the intersection in the first half and the latter half of each phase. if the average speed of vehicles passing the intersection during the latter half of phase is statistically higher than the speed of vehicles passing the intersection during the first half of phase, it is likely that it may affect the frequency and the severity of the conflict. fig. 6 conflicts between the vehicles moving in the same direction 4. conclusions and future work recent vision technology allows traffic engineers to analyze traffic safety based on image processing applications. vehicle trajectory data including vehicle position and speed are extracted and converted into traffic conflicts related variables. these variables are analyzed to find correlations with crash data and geometric design variables and included in the statistical analysis process to evaluate the safety of intersections. in this paper, signal timing information data is included in the analysis proceedings of engineering and technology innovation, vol. 13, 2019, pp. 20-25 24 process to investigate if there are any correlations with traffic conflict data. for example, traffic conflicts happen more frequently in a certain movement at certain signal time phase. in this paper, signal timing information data is included in the analysis process to investigate if there are any correlations with traffic conflict data. also, vehicle speed and its standard deviation are analyzed based on the assumption that traffic characteristics might be different between vehicles passing the intersection during the latter half of phase and vehicles passing the intersection during the first half of phase. first, the number of conflicts were compared based on the directional signal phase. it was found that the number of traffic conflicts by directional signal phase varies depending on the directional signal phase. secondly, the characteristics of conflicts were compared by conflict type. the results show that conflicts between vehicles running in different directions have different characteristics by conflict type. thirdly, the characteristics of conflicts were analyzed by the first and second half of directional signal phase for the left turn and right turn vehicle conflicts. it was found that the conflict characteristics do not differ between the first half and the second half of the phase. finally, the characteristics of conflicts were analyzed by the first and second half of directional signal phase for through and right turn vehicle conflicts. the results show that the conflict characteristics differ according to the first half and the second half of the phase. underlying assumption on this analysis was that vehicles passing the intersection during the latter half of phase tend to have higher speed than vehicles passing the intersection during the first half of phase since vehicles in the queue at the intersection were gradually dissipated during the first half of phase, therefore vehicles have lower speed during the first half of phase meaning less chance of having conflicts. conflicts between the vehicles moving in a different direction at the selected intersection were analyzed. also, conflicts between the vehicles moving in the same direction at the selected intersection were analyzed as well. it was found that there was a significant difference between the two conflicts. additionally, the conflicts were different between the directional movements of the vehicle. it was found that a significant number of conflicts occurred in signal phase two and four. most of the conflicts were conflicts between left-turning vehicles and through vehicles. however, conflicts between right-turning vehicles and through vehicles were relatively spread out in the entire signal timing period. also, it was found that conflicts occurred more often between the left turning and the right turning vehicles compared to conflicts between the through moving vehicles and the right turning vehicles. additionally, more conflicts occurred in the middle and later part of phase which implies that there might be correlations between the conflict and signal phase and timing. as shown in this paper, traffic conflicts happen more frequently in a certain movement at certain signal time phase. the goal of this paper is to develop a method for vision technology-based traffic safety analysis process using traffic signal data to identify crash prone movement and signal to time at a given intersection. more robust statistical analysis is required to investigate the correlations between traffic conflicts and signal timing. as limitations of this study, video data were collected at only one intersection. it is necessary to analyze more video data collected from multiple intersections to compare the results from this study and compare with the results from other intersections to verify the conclusion. also, it is necessary to analyze additional factors affecting the characteristics of conflicts in addition to the factors considered in this paper. in this study, the proposed image detection method extracts the vehicle trajectory using the background subtraction method. however, this method generates false data when the target vehicle is covered from other objects such as shadows, signal poles, and adjacent vehicles. therefore, it is necessary to develop a video detection method and a data analysis method that handles these problems more efficiently. the preliminary experimental results from this study show that the proposed method has the potentials to be utilized as a part of a vision-based traffic safety analysis tool. also, it has the potentials to identify possible factors and characteristics in proceedings of engineering and technology innovation, vol. 13, 2019, pp. 20-25 25 traffic accidents which have been not analyzed in the conventional traffic safety analysis. since the outcome of the paper is based on the limited data set, the proposed method is not proven to be applied to other intersections. however, the proposed method is believed to provide some insight into more efficient safety analysis. more extensive studies may be conducted in the future to demonstrate the effectiveness of the proposed method in much greater testing scale. conflicts of interest the authors declare no conflict of interest. acknowledgments this work was supported by nrf2017r1d1a1a09000606 of the national research foundation of korea from the korean government. references [1] k. ismail, t. sayed, and n. saunier, “n. automated analysis of pedestrian-vehicle conflicts: context for before-and-after studies,” journal of the transportation research board, vol. 2198, no. 1, pp. 52-64, january 2010. [2] a. laureshyn, “superpixel based road user tracker,” workshop on the comparison of surrogate measures of safety extracted from video data, transportation research board 93rd annual meeting, november 2013, pp. 1-5. [3] t. madsen, c. bahnsen, h. lahrmann, and t. b. moeslund, “automatic detection of conflicts at signalized intersections,” workshop on the comparison of surrogate measures of safety extracted from video data, transportation research board 93rd annual meeting. january 2014. [4] w. suh, j. park, and j. kim, “traffic safety evaluation based on vision and signal timing data,” proceedings of engineering and technology innovation, vol. 7, pp. 37-40, 2017. [5] n. smaoui, k. kim, o. gnawali, y. lee, and w. suh, “respirable dust monitoring in construction sites and visualization in building information modeling using real-time sensor data,” sensors and materials, vol. 30, pp. 1775-1786, 2018. [6] w. suh, j. anderson, a. guin, and m. hunter, “evaluation of video detection system as a traffic data collection method,” scientia iranica, vol. 22, pp. 2092-2102, may 2015. [7] m. park, j. kim, y. lee, j. park, and w. suh, “vision-based surveillance system for monitoring traffic conditions,” multimedia tools and applications, vol. 76, no. 23, pp. 25343-25367, 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/). proceedings of engineering and technology innovation, vol. 5, 2017, pp. 01 06 innovative security technology for optical fiber data transmission using optical vortex marek życzkowski * , konrad dominik brewczyński, mateusz karol institute of optoelectronics, military university of technology, warsaw, poland . received 10 july 2017; received in revised form 11 august 2017; accept ed 13 august 2017 abstract this article presents system concept of the use of an optical vortex phenomenon for secure data exchange in the optical fiber line. optical vortices are obtained in free space, and then they are introduced into the optical fiber. their p roperties are e xa mined in the world [1-2], so far, wh ich directly shows the ability to increase the bandwidth of optical fiber. given the structure and characteristics of the optical vorte x, the authors propose to set a secure optical link. such link can be provided in two ways . the first method involves coding an information on optical vorte x. the data in the optical fiber line can be coded by time-position modeling of the optical properties of the vortex. in addition, changing the topology of the optical vortex and the use of mode dependencies also offer the possibility of information coding. the second way to use the vortex in the optical fiber is setting of optical fiber sensor, sensitive to the disturbance of fiber-optic trans mission line. it can be achieved by propagation both coded informat ion in basic mode and an optical vorte x in a mic rostructural fiber. in the case of physical impact on optical fiber the vorte x disturbs the flow of data, forming the information noise on the output of the fiber. the artic le p resents the setup of the generation of optical vorte x, for the te leco mmunications bandwidth, developed at the institute of optoelectronics, in the security systems group and initial tests of the setup . keywor ds: optical vortex, fiber optic, safe communication 1. introduction the work on obtaining and using optical vortices in teleco mmun ications fiber optic networks, undertaken at the institute of optoelectronics, is the result of global de mand for increasing the speed of informat ion flow and for increasing the security of data transmission. each day the internet traffic increases. according to mints the interpolated growth of demand for data is linear since about 2007 and is projected to rema in so [3]. according to r.j. essiambre and r.w. t kach we are inevitably approaching the limit of the capacity of optical fibers. the only solution is to search for new methods of data encoding and new possibilit ies of denser packing da ta. the solution may be optical vortices, which can propagate in telecommun ications optical fiber parallel to data coded current technical solutions. space division multiple xing allows increasing the capacity of an optical fiber. there must be maintained appropriate conditions for the propagation of optical vortices. first of all optical vort ices propagate only multi-core optica l fibers [4]. another proble m with working with optical vortices is their instability over long distances. currently the world record to maintain a stable optical vortex is its propagation on distance 1.1 km [5]. * corresponding author. e-mail address: marek.zyczkowski@wat.edu.pl proceedings of engineering and technology innovation, vol. 5, 2017, pp. 01 06 copyright © taeti 2 2. optical vortex forming setup to create an optical vorte x in the optical fiber, it is necessary to create it first in free space. for this purpose, the gaussian beam must be used of a wavelength 𝜆 = 1550 𝑛𝑚 . in order to generate a proper focused gaussian beam we used: a laser (o f a wavelength 𝜆 = 1550 𝑛𝑚 ), spatial filter and a pinhole. this bea m is introduced to the optical fiber with a length of 0.5 m by an fc-pc connector. the beam a fter leaving the fiber goes to the spatial filter. the spatial filter used in the setup is a lowpass filter and this means that peripheral parts of the beam, which contain high frequencies, are being cut. the principle of operation of a spatial filter is based on diffraction, where in the side bands are filtered off on pinhole, and only a ma in mod is transmitted further. the spatial filter has an aperture of adjustable dia meter and the dia meter used in the setup is 5 μm. (a) scheme (b) photo of the setup fig. 1 the setup for an optical vortex forming there have been added a polarisation controlle r to the setup of generating the gaussian beam. po larised gaussian beam is designed to fit the phase plate. consequently, a beam is polarised in the horizontal direction. the beam, a fter passing through the phase plate, is delayed in phase from 0 do 2𝜋 . we used the cell with the nematic liquid crystal 5cb in the setup forming the optical vorte x. it is div ided into 16 pieces, each of which is supp lied by a diffe rent voltage. this imp lies that each of the 16 pieces has a different refract ive inde x. thanks to this the gaussian beam, which propagates by the phase plate is gradually delayed, diffe rently for each piece of the plate. depending on the number of pieces and the voltage applied to them the topology of the optical vortex changes [6]. proceedings of engineering and technology innovation, vol. 5, 2017, pp. 01 06 copyright © taeti 3 fig. 1 shows the setup for optical vorte x forming which is set together at the institute of optoelectronics. the setup consists of (in order):  laser, 𝜆 = 1550 nm,  fiber optic polarisation controller,  lens,  spatial filter,  lens,  phase plate,  infrared camera,  laptop (which visualises the camera view). fig. 2 shows the optical vortex gained at the institute of optoelectronics at military university of technology . fig. 2 the optical vortex observed at infrared camer 3. mathematics of nematic liquid crystals fig. 3 shows the 5cb ne matic molecu le. the e lectrical permeab ility for such ne matic crystal is different for paralle l direction 𝜀∥ to director �̆� and different for perpendicular direction 𝜀⊥ to director �̆�. fig. 3 nematic 5cb molecule with the marked vectors: parallel and perpendicular to the director 𝑛 the electrical permeability anisotrophy is given by eq. (1): 𝜀𝑎 = 𝜀∥ − 𝜀⊥ (1) where: 𝜀⊥ – electrical permeability perpendicular to the molecules director, 𝜀∥ – electrical permeability parallel to the molecules director. proceedings of engineering and technology innovation, vol. 5, 2017, pp. 01 06 copyright © taeti 4 these permeabilities are factors of the continuous med iu m, not of a single mo lecule . if 𝜀𝑎 > 0, then we have a positive liquid crystal anisotropy; if 𝜀𝑎 < 0, the crystal is negative. the ne matic c rystal which is used to form the optical vorte x is a positive liquid crystal. the resultant refractive index of the light is given by eq. (2): 𝑛𝑎 = 𝑛𝑒 − 𝑛0 (2) where: 𝑛0 – the ordinary refractive index of the light, 𝑛𝑒 – the extraordinary refractive index of the light , 𝑛𝑎 – the resultant refractive index of the light. the density of the energy of the interaction of the nematic crystal with the electric field is made by eq. (3) [7]: 𝑓𝐸 = − 1 2 𝐸 ⋅ 𝐷 = − 1 2 (𝐸∥ , 𝐸⊥) ⋅ (𝜀0 𝜀∥ 𝐸∥ , 𝜀0 𝜀⊥ 𝐸⊥) = − 1 2 𝜀0 (𝜀∥ 𝐸∥ 2 + 𝜀⊥ 𝐸⊥ 2 ) (3) where: 𝐸 – external electrical field, 𝐷 – electrical induction. the function 𝑓𝐸 depends on the orientation fie ld 𝑛 and its first deriat ive, the electrica l fie ld e is e xc ited in the ne matic crystal by the electrical induction d. the phase shift between the ordinary and extraordinary rays of the light at the exit of ne matic c rystal can be described by the following eq. (4): 𝜑 = 2𝜋 𝜆 𝑑 [ 1 𝑑 ∫ 𝑛𝑒 𝑛𝑜 √𝑛𝑒 2 ∙ 𝑠𝑖𝑛2 𝜗(𝑧) + 𝑛𝑜 2 ∙ 𝑐𝑜𝑠 2 𝜗(𝑧) 𝑑𝑧 − 𝑛𝑜 𝑑 0 ] (4) for 𝜗(𝑧) = 0 𝜑 = 2𝜋 𝜆 (𝑛𝑒 − 𝑛𝑜 )𝜆 for 𝜗(𝑧) = 𝜋 2 𝜑 = 0 we can see, from the above, that if the liquid c rystal molecules are o riented parallel to the phase plate covers, ie. 𝜗(𝑧) = 0, then the propagation of light through the liquid crystal is impossible, because the ordinary ray and the e xtraordinary cannot interfe re with each other. when the directors of the liquid crystal mo lecules are angled to the the covers of phase plate c lose to 𝜗(𝑧) = 𝜋 2 , then propagation of light through the liquid crystal is possible, because the phase shift between the ordinary and the extraordinary ray of light is zero, so the rays can interfere with each other. depending on the deviation angle 𝜗(𝑧) of the liquid crystal mo lecules fro m the covers of phase plate the light phase is delayed from 0 to 2𝜋 , forming the optical vortex this way. 4. the concept of using the optical vortex so far, the ioe wat group performed numerica l analysis of the forming and carrying the optical vortex in the optical fiber. on that basis the vortex forming setup was made. it a llo wed the study of stable spatial structure of optical vorte x in the optical telecommunications . proceedings of engineering and technology innovation, vol. 5, 2017, pp. 01 06 copyright © taeti 5 in cooperation with the tea m of dr. pa weł mergo fro m um cs in lub lin, we made mic rostructural fibers appropriate (according to calculations) to carry an optical vortex. currently the tec hnical study is focused on the correct introducing vortex to a mic rostructural fiber and obtaining a stable and long distance propagation in it. the idea of ma intaining the optical vorte x in a teleco mmun ication fiber is to use it to secure data e xchange. according to our conceptual analysis optical vortices can be used for data protection. th is applicat ion can be done in various ways by using the specific properties of optical vortices. adequately to standard fiber optic systems, we reco mmend the use of light intensity changes of an optical vortex as a fiber optic sensor. in addit ion, there is a possibility of modulating the optica l vorte x to falsify positions of bits in space and time so that the reproducing of the information is not possible. fig. 4 shows schematic concept of using the optical vortex for building safe data transfer system. (a) the optical fiber sensor (b) an information carrier fig. 4 a simplified diagra m of the concept of using an optical vortex as: a. the optical fiber sensor, b. an information carrier for an optical encoding; blue color is for the information flow, red is for sensor element in the first case, according to the assumptions, the information b it (h ighlited with a blue line on fig. 4 a .) propagates in the ma in core of the optical fiber. the optical vorte x (h ighlited with a red line on fig. 4 a .) p ropagates in the rest of the cores. the ma in core of the fiber is now the main co mmunicat ion channel. the e xte rnal optical fiber cores carry constant or free ly modulated optical vortex light. in the situation of physical violation, the optical fiber the optical vorte x is lea king and mixing with the encoded informat ion. it starts a flow of light energy, which in effect fa lsifies the wiretapping detector. in addition, at the time of disturbance on the output of the mic rostructural fiber, there is being c reated an information noise that goes to main rece ivi ng ele ment. on the basis of this intensity effect, the alarm of attempt of wiretapping the information transmitted in the optical fiber can be generated. in the second case, we use the optical vorte x modulation. fig. 4(b). shows the concept of using the optical vortex as a modulated information carrier. optical vorte x, according to the assumptions, can be spatially modulated or time modulated by changing its topology and adjusting its phase delay. this modulation a llo ws encoding information bit on the optical vortex. spatial-time change of the position of the individual optical bits prevents correct reading it – without building a proper detection system and without access to the en tire structure of the core. wiretapping with a tradit ional method, eg. by using a clip-on coupler is impossible. co mpared to known methods of securing the informat ion in the fiber optic trac k, such as interference sensors, qkd systems, these systems bring new types of functional characteristics and they can be used to build secure tracks of classified/confidential data exchange . proceedings of engineering and technology innovation, vol. 5, 2017, pp. 01 06 copyright © taeti 6 5. summary we briefly described the idea of the formation and operation of the optical vortex obtained at the institute of optoelectronics at military university of technology. we a lso showed physics phenomena and mathemat ical description of the changes taking place in the nematic liquid crystals, which are responsible for the formation of an optical vortex. we introduced the concept of using an optical vorte x both to encode information in the optica l fiber trac k and to protect data which is transmitted in this track. as part of the work, we achieved a stable distribution of optical vortex for optical telecommun ications range of the gen erated light 1550 n m. in addition, we developed and made microstructural optical fiber that is suitable for carry ing vortices and allows imple mentation of the described setup concepts. our further studies are aimed on a deeper understanding and improving systems using optical vortices to safe information carrying . references [1] n. bozinovic, “orbital angular momentum in optical fibers ,” ph.d. dissertation, college of eng., boston univ., united states , 2013. [2] w. fraczek, “interferometric measurement of optical field phase distribution using the phase discontinuity ,” ph.d. dissertation, lower silesian digital library, 2008. [3] r. j. essiambre and r. w. tkach, “capacity trends and limits of optical communication networks ,” proc. of the ieee, vol. 100, no. 5, pp.10351055, may 2012. [4] c. brunet and l. a. rusch, “optical fibers for the transmission of orbital angular momentum modes ,” optical fiber technology, vol. 31, pp. 172-177, september 2016 [5] n. bozinovic, y. yue, y. ren, m. tur, p. kristensen, h. huang, et al., “terabit-scale orbital angular momentum mode division multiplexing in fibers ,” science, vol. 340, no. 6140, pp. 1545-1548, 2013. [6] j. albero, p. garcia-martinez, n. bennis, e. oton, b. cerrolaza, i. moreno, et al., “liquid crystal devices for the reconfigurable generation of optical vortices,” journal of lightwave technology, vol. 30, no. 18, pp. 3055-3060, 2012. [7] j. kędzierski, m. a. kojdecki, z. raszewski, j. zieliński, “frederiks transitions and their role in determining material constants of liquid crystals,” chapter of monograph: complementary methods in study of condensed phases, p p. 242-271, 2008.  proceedings of engineering and technology innovation, vol. 9, 2018, pp. 09 16 fresh properties and strength development of cement mortar using nitrite-based accelerator and chemical admixtures masumi inoue 1,* , heesup choi 1 , yuhji sudoh 2 1 school of regional innovation and social design engineering, kitami institute of technology, hokkaido, japan 2 chemicals division basic chemicals department, nissan chemical industries, ltd., tokyo, japan received 21 march 2018; received in revised form 01 may 2018; accepted 17 july 2018 abstract a nonfreezing agent prevents initial frost damage by simplified curing and also promotes early strength development. in general, a nitrite-based accelerator is used as the main component of this nonfreezing agent. when a large proportion of a conventional accelerator is used, the slump loss increases because the setting and hardening processes are strongly enhanced. in addition, these conventional accelerators may not be sufficiently effective at temperatures of -10 °c or less. therefore, the fresh properties and strength development of cement mortars when using a nitrite-based accelerator and chemical admixtures in a low-temperature environment are examined. the objective is to develop a high-performance nonfreezing agent. this paper presents the contributions of the nitrite-based accelerator and two types of high-range water-reducing agents, used in combination. from the investigation, it was confirmed that the change in the mortar flow can be suppressed by combining two types of high-range water-reducing agents with a nitrite-based accelerator, even after the passage of 60 min after mixing. furthermore, the mix was found to contribute to not only the initial strength but also toward maintaining the good strength for a longer time. keywords: nitrite-based accelerator, high-range water-reducing agent, mortar flow, compressive strength, cold-weather concrete 1. introduction in the case of concrete constructions in cold weather, it is necessary to control the temperature of heat curing using a temporary enclosure and heater to prevent initial frost damage. if the installation of the temporary enclosure is difficult, owing to a steep slope, narrow working space, and strong winds, a nonfreezing agent is commonly used. this nonfreezing agent, by simplified curing, prevents initial frost damage and also secures initial strength development. furthermore, the function of this agent is to reduce the freezing temperature of the water in the concrete. the characteristics of concrete, using a nitrite-based accelerator and chemical admixtures in a low-temperature environment, are investigated in this study with the aim to develop a high-performance nonfreezing agent [1-4]. in general, a nitrite-based accelerator is used as the main component of this nonfreezing agent [5]. calcium nitrite chiefly accelerates the hydration process by increasing the solubility of c3s (alite, 3cao·sio2) and (belite, 2cao·sio2). in addition, it also reacts with c3a (aluminate phase, 3cao·al2o3) in cement to generate hydronitroaluminate, 3cao•al2o3•ca(no2)2•10h2o. furthermore, calcium nitrite also reacts with calcium hydroxide in concrete to generate calcium hydroxynitrate, cao•ca(no2)2•3h2o. these cement hydrates lead to refinement in cement hardening, resulting in the improvement of the mechanical properties of concrete in low-temperature environments [6]. it is * corresponding author. e-mail address: m-inoue@mail.kitami-it.ac.jp tel.: +81-157-26-9513; fax: +81-157-26-9513 proceedings of engineering and technology innovation, vol. 9, 2018, pp. 09 16 copyright © taeti 10 assumed that the early strength development of concrete is promoted by increasing the proportion of the added accelerator. however, conventional nonfreezing agents might not induce the process leading to the early strength development of concrete at temperatures of –10 °c or lower. to promote early strength development at these low temperatures, it is necessary to increase the proportion of the accelerator added in the concrete. however, if the accelerator is added in excess, a strong hydration reaction will occur. as a result, the slump loss of fresh concrete after mixing will become too large, and the workability of the fresh concrete is impaired [7]. in this study, the fresh properties and strength development of cement mortar, when using a nitrite-based accelerator and chemical admixtures, in a low-temperature environment are examined. the objective is to develop a high-performance nonfreezing agent. this paper presents the contributions of the nitrite-based accelerator and various high-range water-reducing agents used in combination. 2. evaluation of fresh properties 2.1. outline of the experiment according to previous studies, the initial slump loss can be suppressed by combining a melamine-based water-reducing agent and polycarboxylic acid-based water-reducing agent in a normal temperature environment (+20 °c) [1, 8]. initially, the fresh properties of cement mortar using a large proportion of a nitrite-based accelerator and two types of high-range water-reducing agents, in a low-temperature environment, are examined. the aim is to clarify the compatibility of the accelerator and chemical admixture. 2.2. materials used the properties of the materials used in this study are listed in table 1. normal portland cement was used, and number 5 grade silica sand was used as a fine aggregate. a 45% water solution, primarily composed of calcium nitrite (cn) was used as the accelerator. a melamine-based high-range water-reducing agent (me) and a polycarboxylic acid-based high-range water-reducing agent (pc) were also used. table 1 properties of the materials used material(code) properties cement (c) normal portland cement, density: 3.16 g/cm 3 , specific surface area: 3280 cm 2 /g fine aggregate (s) no.5 silica sand, absolute dry density: 2.61 g/cm 3 , water absorption: 0.26%, fineness modulus: 2.16 accelerator (cn) main component: calcium nitrite; other accelerating components (45% water solution), density: 1.42–1.44 g/cm 3 melamine-based water reducing agent (me) main component: amidosulfonic-acid-modified melamine condensate, density: 1.17–1.19 g/cm 3 polycarboxylic acid-based water reducing agent (pc) main component: polycarboxylic acid ether compound, density: 1.03–1.07 g/cm 3 2.3. mix proportions the mix proportions of the cement mortar used in the experiment are listed in table 2. the water-to-cement ratio used was 50%. the mass ratio of the aggregate (s) to the cement (c) used was 2.50. in general, the processes leading to the early strength development of concrete increases with an increase in the amount of added accelerator. the standard replacement ratio of conventional accelerators is approximately 4% to 7% of the unit content of cement (3-5 l per 100 kg of cement). in this study, the replacement ratios of cn used were 0%, 7%, 11%, and 15% as recommended by previous studies on high-performance non-freezing agents [1-4]. in addition, the replacement ratios of the melamine-based water-reducing agent (me) to the polycarboxylic acid-based high-range water-reducing agent (pc) used were 1:0, 0:1, 1:2, and 2:1. the total amounts proceedings of engineering and technology innovation, vol. 9, 2018, pp. 09 16 copyright © taeti 11 of me and pc were determined to satisfy the target value of mortar flow. the target value of the mortar flow was 200±20 mm immediately after mixing. 2.4. test method table 2 mix proportions of cement mortar symbol w/c (%) s/c unit content (kg/m 3 ) admixture (c×%) w c s cn me pc cn0 50 2.5 282 564 1409 0 0 0 cn7 7 cn11 11 cn15 15 cn7-d 7 0.4 0.2 cn11-a 11 1 0 cn11-b 0 1 cn11-c 0.33 0.67 cn11-d 0.67 0.33 cn15-a 15 3 0 cn15-b 3 0 cn15-c 1 2 cn15-d 2 1 the temperature of concrete at the time of unloading is specified as 10 °c to 20 °c by the architectural institute of japan in the recommendation for practice of cold weather concreting [9]. in this study, the mixing and all tests of the mortar were performed in a temperature and humidity controlled testing room (10 ± 1 °c and 85% ± 5% relative humidity) to obtain the desired temperature of the mortar. the mortar flow was measured immediately after mixing the mortar and also after mixing with a shovel for 30 min and 60 min, assuming that cast-in-place concrete is placed in a construction field within 60 min. the initial and final setting times were measured according to jis a 1147 (method of testing for the time of setting of concrete mixtures by penetration resistance). the inside temperature of the mortar (φ50×100mm) was measured immediately after casting using a thermocouple. 2.5. results and discussion (1) mortar flow the results of the flow test of mortar are presented in fig. 1. when the accelerator (cn) was only used, the change in mortar flow of mix cn0 (without cn) was hardly observed after mixing to after 60 min. the change in the mortar flow of mix cn7 (with cn 7%) was relatively small over the same period. however, the changes in the mortar flow of mixes cn11 and cn15, which contained a large proportion of cn became was large during the first 30 min and then almost constant for the rest of the time. this is because the hydration reaction of fresh mortar proceeded rapidly owing to the presence of large proportions of the accelerator. from these initial results, more mortar flow tests of mixes, which combined two types of high-range water-reducing agents, were conducted to control the changes in the mortar flow of mixes cn11 and cn15. fig. 1 mortar flow 90 110 130 150 170 190 210 230 250 0 30 60 m o rt a r fl o w time (min) cn0 cn7 cn11 cn15 90 110 130 150 170 190 210 230 250 0 30 60 m o rt a r fl o w time (min) cn11-a cn11-b cn11-c cn11-d 90 110 130 150 170 190 210 230 250 0 30 60 m o rt a r fl o w time (min) cn15-a cn15-b cn15-c cn15-d proceedings of engineering and technology innovation, vol. 9, 2018, pp. 09 16 copyright © taeti 12 when the accelerator and high-range water-reducing agents were used, the changes in mortar flows of mixes cn11-a (with only me) and cn11-c (with me:pc=1:2) were larger after mixing, when compared with those of the other mixes. however, the changes in mortar flows of mixes cn11-b (with the only pc) and cn11-d (with me:pc=2:1) became smaller, and these mixes had good fluidity even after the passage of 60 min after mixing. on the other hand, mixes cn15-a, cn15-b, and cn15c had no fluidity after the passage of 30 min after mixing. however, the change in the mortar flow of mix cn15-d (with me:pc=1:2) was relatively small but had no the fluidity, after the passage of 60 min after mixing. it can be concluded from these results that it is difficult to secure fresh properties of mix cn15 at real site, even if the high-range water-reducing agents were to be used. (2) setting time the results of the setting tests are presented in fig. 2. when the accelerator (cn) was only used, the initial and final setting times decreased with an increase in the amount of accelerator. however, when the accelerator and different high-range water-reducing agents were used, there was no significant difference in the set times of these mixes. on the other hand, the set time of these mixes when using cn11 and high-range water-reducing agents was slightly longer when compared with that of mix cn11. however, there were no significant differences in the set times. therefore, it can be concluded that the effect of high-range water-reducing agents on setting and hardening times is relatively small. fig. 2 setting time (3) inside temperature of mortar the changes in the inside temperature of the mortar are presented in fig. 3. the temperature changes in mixes cn11 and cn15, which contained a large proportion of accelerator (cn), are larger when compared with the temperature changes of mix cn0. the time to reach a temperature peak for mixes cn11 and cn15 were 15 min and 20 min, respectively. after these respective periods, the temperature of these mixes decreased rapidly. this is because the hydration reaction proceeded rapidly after mixing with a high proportion of accelerator. therefore, it can be concluded that the changes in mortar flow become very large remarkably because of mixing with a high proportion of accelerator. fig. 3 changes in the inside temperature of mortar 9:03 8:16 6:19 3:28 6:45 6:15 6:33 6:36 13:18 12:46 8:56 6:51 9:27 9:32 9:40 9:34 2 4 6 8 10 12 14 cn0 cn7 cn11 cn15 cn11-a cn11-b cn11-c cn11-d s e tt in g t im e ( h o u rs ) upper:final setting time lower:initial setting time 8 10 12 14 16 18 20 0 1 2 3 4 5 in si d e t e m p e ra tu re ( ℃ ) time (hour) cn0 cn7 cn7-d cn11 cn11-d cn15 cn15-d proceedings of engineering and technology innovation, vol. 9, 2018, pp. 09 16 copyright © taeti 13 on the other hand, the temperature changes of mixes cn7-d, cn11-d, and cn15-d, which contained the accelerator and high-range water-reducing agents (with me:pc=1:2), were smaller when compared with those of mixes cn7, cn11, and cn15, which contained only the accelerator. furthermore, the time required to reach a temperature peak for mixes cn11-d and cn15-d was shorter when compared with that required for mixes cn11 and cn15. in particular, the peak temperature for mix cn11-d was smaller when compared with that of cn11. this is because the hydration reaction of mix cn11-d was suppressed as a result of using two high-range water-reducing agents from the time of initial mixing up until 60 min after mixing. 3. evaluation of strength development 3.1. outline of the experiment in this study, experiments were conducted using cement mortar to study the influence of a nitrite-based accelerator and various high-range water-reducing agents, used in various combinations, on the strength development process of the cement mortar. the compressive strength, expansion strain, and change in pore structure were determined. 3.2. materials used and mix proportions the same materials and mix proportions were used, as mentioned in section 2. the replacement ratios of the accelerator (cn) were 0%, 7%, 11%, and 15%. the replacement ratio of the melamine-based water-reducing agent (me) to the polycarboxylic acid-based high-range water-reducing agent (pc) was 2:1, which had a controlling effect on the mortar flow, as mentioned earlier in section 2. 3.3. test method the mixing was performed in a temperature and humidity controlled testing room (10 ± 1 °c and 85% ± 5% relative humidity) to obtain the required temperature of the mortar. the compressive strength (φ50 × 100 mm) was measured after 1, 3, 7, 14, and 28 d of curing in the testing room (10 ± 1 °c and 85% ± 5% relative humidity). in the unrestrained deformation test, the cement mortar was cast in a mold (100×100×400 mm). a polytetrafluoroethylene sheet and styrene board were installed on the inside of the mold to prevent restraint of volume change of mortar and water dispersion of mortar. an embedded gauge was also installed in the center position of the mold. the sealed specimens were then cured in the temperature and humidity controlled testing room (10 ± 1 °c and 85% ± 5% relative humidity). the expansion and shrinkage strains of the test specimen for each case were measured immediately after casting. the change in pore structure was measured using mercury porosimetry. initially, the mortar specimens were immersed for 4 h in acetone to stop the hydration of the mortar. later, the specimens were dried in a drying oven of 105 °c for 24 h. 3.4. results and discussion (1) compressive strength the results of the compressive strength test of mortar are presented in fig. 4. when the only accelerator was used, the compressive strength measured at 1 day of curing, became larger with an increase in the amount of added accelerator. mixes cn11 and cn15 showed compressive strengths 6.6 mpa and 8.0 mpa, respectively, which exceeded the 5.0 mpa value of the compressive strength required for preventing the initial frost damage. however, after 3, 7, 14, and 28 d of curing, the compressive strength decreased with an increase in the amount of added accelerator. this is because the hydration reaction was sufficiently enhanced during the curing process, at a temperature of +10 °c. in addition, when a large proportion of the nitrite-based accelerator is used, calcium nitrite chiefly accelerates the hydration process by increasing the solubility of c3s and c2s, and it also reacts with c3a. in particular, the reaction of c3a generates a large amount of ettringite, and this reaction also helps in the expansion of concrete, from the initial stage [6, 10]. it is assumed that the formation of these rough hydrated phases by the generation of ettringite affected the strength development process of the cement mortar. proceedings of engineering and technology innovation, vol. 9, 2018, pp. 09 16 copyright © taeti 14 when the accelerator and two types of high-range water-reducing agents (me:pc=2:1) were used, mixes cn11-d and cn15-d showed compressive strengths 6.0 mpa and 7.5 mpa, respectively, which exceeded the critical value of 5.0 mpa. after 3 and 7 d of curing, the compressive strengths became larger with an increase in the amount of added accelerator. even after 14 and 28 d, mixes cn11-d and cn15-d showed good strength development. therefore, it can be concluded that a large proportion of the added accelerator not only helps in obtaining the initial strength for preventing the initial frost damage but also maintains good strength for a long time. fig. 4 compressive strength (2) unrestrained expansion and shrinkage strain the expansion and shrinkage strain behavior of each specimen under unrestrained deformation from the onset of casting to 4 h after casting is presented in fig. 5. when only the accelerator (cn) was used, the expansion strain immediately after casting increased with an increase in the amount of added accelerator. it is assumed that mixes cn11 and cn15 showed a large expansion strain because the reaction rate of c3a was very high from the initial stage of casting and also because of the enhanced hydration of c3s and c2s [6, 10]. however, the expansion strain of mixes using the accelerator and high-range water-reducing agents was smaller than that of mixes using only the added accelerator. furthermore, the time to reach the peak temperature peak was longer by approximately 40 min. these results support the argument that the strength development of mixes is increased by using high-range water-reducing agents. fig. 5 changes of expansive strain (3) changes in the pore structure the pore volume distributions of mixes cn0, cn11, and cn-d are presented in fig. 6, and the cumulative pore volume distributions of each case are presented in fig. 7. when the specimens were cured for 1 d, the pore volume of mixes cn11 and cn11-d became larger when compared with that of mix cn0 without the accelerator, and the total pore volume became smaller when compared with that of mix cn0. this is because the hydration reaction was sufficiently enhanced because of using a large proportion of the nitrite-based accelerator. on the other hand, when the specimens were cured for 7 and 28 d, 0 10 20 30 40 50 cn0 cn7 cn11 cn15 cn7-d cn11-d cn15-d c o m p re ss iv e s tr e n g th ( n /m m 2 ) 1day 3days 7days 14days 28days 0 30 60 90 120 150 0 1 2 3 4 e x p a n si v e s tr a in ( μ ) time (hour) cn7 cn7-d cn11 cn11-d cn15 cn15-d proceedings of engineering and technology innovation, vol. 9, 2018, pp. 09 16 copyright © taeti 15 there were no significant differences in the pore distributions between mixes cn11 and cn11-d. however, the total pore volume of mix cn11-d became smaller when compared with that of mix cn11 without the high-range water-reducing agents. when a large proportion of the accelerator only was used, it was assumed that the total pore volume increased owing to the formation of the rough hydrated phases during the initial hydration process. (a) cn0 (b) cn11 (c) cn11-d fig. 6 pore volume distribution 4. conclusions the aim of this study was to develop a high-performance nonfreezing agent for use in the construction of cold-weather concrete. tests to determine the fresh properties and strength development processes of mortar specimens containing a nonfreezing agent were performed, and the contributions of a nitrite-based accelerator and high-range water-reducing agents used in combination were examined. the following conclusions were obtained from the investigation. (1) the changes in the mortar flow of mix cn11-d, which contained a large proportion of accelerator and two types of high-range water-reducing agents, became smaller, and this mix had good fluidity even after the passage of 60 min after the completion of mixing. (2) when a large proportion of the accelerator only was used, the compressive strength at 1 d increased with an increase in the amount of added accelerator. however, after 3, 7, 14, and 28 d of curing, the compressive strength decreased with an increase in the amount of added accelerator. (3) mix cn11-d (with me:pc=2:1) showed not only the initial strength for preventing initial frost damage but also maintained good strength for a long time. (4) the expansion strain of mix cn11-d was smaller than that of mix cn11without high-range water-reducing agents. furthermore, the total pore volume of mix cn11-d after 28 d of curing was smaller when compared with that of mix cn11. 0.0 0.5 1.0 1.5 2.0 2.5 0.001 0.01 0.1 1 10 100 1000 p o re v o lu m e (v o l. % ) pore diameter (μm) cn0 1day 7days 28days 0.0 0.5 1.0 1.5 2.0 2.5 0.001 0.01 0.1 1 10 100 1000 p o re v o lu m e ( v o l. % ) pore diameter (μm) cn11 1day 7days 28days 0.0 0.5 1.0 1.5 2.0 2.5 0.001 0.01 0.1 1 10 100 1000 p o re v o lu m e (v o l. % ) pore diameter (μm) cn11-d 1day 7days 28days proceedings of engineering and technology innovation, vol. 9, 2018, pp. 09 16 copyright © taeti 16 references [1] t. aakama, m. inoue, y. sudoh, and s. mikami, “fresh properties and early strength development of concrete using calcium nitrite and water-reducing agents,” proc. of the japan concrete institute, vol. 34, pp. 155-159,july 2012. (in japanese) [2] m. iwasawa, m. inoue, h. choi, y. sudoh, and k. ayuta, “characteristics of concrete using nitrite-based accelerator and chemical admixtures in low-temperature environments,” proc. of the 7th international conference of asian concrete federation, pp. 1-8, october 2016. [3] m. iwasawa, m. inoue, h.choi, and y. sudoh, “study on deformation bahavior of concrete large amount of using nitrite-based accelerator,” proc. of the japan concrete institute, vol.39, pp.115-120, june 2017. (in japanese) [4] m. iwasawa, m. inoue, h. choi, and y. sudoh, “the fundamental study on the expansion and shrinkage behaviors of concrete using nitrite-based accelerator,” proc. of the 2017 world congress on advances in structural engineering and mechanics, p.1-11, august 2017. [5] y. hama and e. kamada, “ the properties of concrete containing a frost-resistance accelerator,” concrete journal, vol.37, no.1, pp.3-8, november 1999. (in japanese) [6] v. s. ramachanran, concrete admixture handbook, u.s.a.: noyes publications, 1995. [7] concrete institute of japan in hokkaido branch, research committee report on the latest technology of concrete admixtures, pp. 76-79, 2011. (in japanese) [8] s. takii, s. hiraoka, t. okamamoto, n. masuyama, y. nagakawa, y. sudoh, m. inoue, and d. atarashi, “search for mix proportions of concrete for pavement in early traffic opening is possible and its practicality,” journal of japan society of civil engineers, vol. 70, no. 1, pp. 1-10, 2014. (in japanese) [9] practical guideline for investigation, recommendation for practice of cold weather concreting, architectural institute of japan, 2010. (in japanese) [10] y. hama and e. kamada, “effects on protection fresh concrete against frost damage by non-chloride and non-alkali type antifreezing admixtures,” concrete research and technology, vol. 7, no. 1, pp. 113-122, 1996. (in japanese)  proceedings of engineering and technology innovation, vol. 12, 2019, pp. 01-08 cae analysis of primary shaft systems in great five-axis turning-milling complex cnc machine chih-chiang hong * , cheng-long chang, nai-rui ou, chien-yu lin department of mechanical engineering, hsiuping university of science and technology, taichung, taiwan received 26 july 2018; received in revised form 13 august 2018; accepted 14 september 2018 abstract the design and analysis of primary shaft systems by using the commercial software of computer aided engineering (cae) in the heavy industry field are novel. the research purpose is to provide the computational results for the construction of primary shaft systems in the five-axis turning-milling complex machine. the cae with commercial software is used to analyze the linear static construction, stress and deformation for primary shaft systems in the great five-axis turning-milling complex computer numerical control (cnc) machine. it is desirable to compute and find the most external loads of primary shaft systems in the cnc machine which can be used to operate in safety condition and under its yield stress value of materials. the linear computational results of static stresses and displacements in primary shaft systems are obtained and investigated with the commercial solidworks® 2014 simulation module. keywords: cae, static, primary shaft, stress, cnc 1. introduction there are many researches for the primary shaft in the fields of the gear system, rotor system, pump regulator system, propulsion system and hooke’s joint system. in 2018, chowdhury and yedavalli [1] presented the study and investigation of natural frequencies in the coupled spinning gear shaft system. in 2018, eftekhari et al. [2] presented the effects nonlinear electromagnetic forces on the steady state amplitude responses for a simply supported rotor system. in 2018, zhang et al. [3] presented the numerical and experimental results of tilt angles in the high-speed shaft for electro-hydrostatic pump’s regulator system. in 2017, chowdhury and yedavalli [4] presented the low speed study of coupled spur geared shaft system to investigate the natural frequencies and dynamic responses. in 2017, huang et al. [5] presented the optimization study of marine propulsion shaft system to investigate the vibration transmission. in 2016, zou et al. [6] presented the numerical results of natural frequencies and amplitude responses in the marine propulsion shafting system. in 2014, bulut [7] presented the numerical results of finite element method (fem) to investigate the dynamic stability of a shaft hooke’s joint system. in 2014, han et al. [8] presented the numerical results of direct spectral method (dsm) to investigate the force responses of a slant cracked shaft in a geared rotor system. in 2013, lahriri and santos [9] presented the numerical and experimental results of contact forces of the shaft and the bearing in a rotor–stator system. in 2012, adekunle et al. [10] presented the computer aided design (cad) technique to study and design the shaft under various severest loadings. in 2012, shahgholi and khadem [11] presented the static results of multiple scales method to investigate the steady state response of the rotating shaft. in 2010, lin and chu [12] presented the investigation of frequency and transversal responses of the slant cracked shaft in a rotor system. in 2009, hosseini and khadem [13] presented the vibration results of perturbation method to investigate the time responses of the * corresponding author. e-mail address: cchong@mail.hust.edu.tw tel.: +886-919037599; fax: +886-4-24961187 proceedings of engineering and technology innovation, vol. 12, 2019, pp. 01-08 2 rotating shaft. in 1995, bert and kim [14] presented the studies and investigations of bending and buckling for the long, powerful driveshaft with laminated composite materials. the author has some presentations on the application in commercial computer aided engineering (cae) software for the turning-milling complex computer numerical control (cnc) machine. in 2018, hong et al. [15] presented the cae static analysis of secondary shaft systems of the great five-axis cnc machine by using solidworks® module. usually there are seven steps used in the processes of simulation with the software solidworks® simulation module. in 2017, hong et al. [16] presented the cae dynamic structural analysis of the great five-axis cnc machine by using solidworks® module. in 2016, hong et al. [17] presented the cae static structural analysis of the great five-axis cnc machine by using solidworks® module. the purpose of the research is to provide some computational results to design the primary shaft system in the five-axis turning-milling complex cnc machine. the linear static results of stresses and displacements of the primary shaft systems are obtained and investigated with the solidworks® 2014 simulation module. the maximum values of stress and displacement are obtained to give a safety prediction for the design of primary shaft systems due to external axial pressure loads and torsion loads. the novelties are to present the designs of primary shaft systems in cnc machines by using the commercial cae for engineers and to investigate the simulation data for the primary shaft systems under external pressure and torsion loads. 2. method of simulations 2.1. mathematical model and software in the linear static structural analysis, a general mathematical model in matrix form would be given and used in the computer program to solve for stress and displacement results. generally, there are some steps used in the method of simulation with the software solidworks® simulation module to find the results of displacement and stress [15]. assumptions involved in the study are for the inner part in gear trains by using the simplify shapes, for the geometry of the cut tool by using the upper right end shaft to simulate the milling. 2.2. assembling 3d parts (a) dimensions of primary shaft systems (b) assembling 3d parts of primary shaft systems (c) shafts in primary shaft systems (d) upper pairs gear train in primary shaft systems fig. 1 dimensions and assembling 3d parts of primary shaft systems proceedings of engineering and technology innovation, vol. 12, 2019, pp. 01-08 3 to use the commercial cae software and run the static results for complex cnc machine, it is necessary to prepare the assembling 3d parts of primary shaft systems as shown in fig. 1. the dimensions of the main parts are provided for primary shaft system with 1190.5 mm, 940 mm and 860 mm as shown in fig. 1(a). the main components of the primary shaft system are geared trains, shafts and box as shown in figs. 1(b)-1(d). the dimensions of interior parts of the gearbox like shaft and gears are upper shaft diameter is 209 mm, upper right end shaft diameter is 110 mm, upper big gear diameter is 440 mm, lower shaft diameter is 20 mm, lower small gear diameter is 40 mm, lower big gear diameter is 120 mm, respectively. there are upper pairs and lower pairs, gear trains composed the gear train system to transmit the rotation speed for the shafts as shown in the fig. 1(c). the tool is fixed on the end of gear shaft to provide drilling and milling functions. the primary shaft system can be rotated by the base in rotational motion with respect to y axis. 2.3. materials of 3d parts it is necessary to define the individual material of assembling 3d parts for great five-axis turning-milling complex cnc machine. the materials of the main parts of the primary shaft systems are given with cast iron and steel alloy. the yield stress of cast iron material for the primary shaft system is 275mpa. the yield stress of the steel alloy material for the gear train system is 620mpa. to prevent failure in the cnc machine, the value of working stress in each material of components should be under its yield stress value. 2.4. boundary conditions of primary shaft systems the boundary conditions of primary shaft systems are used to compute and analyzed corresponding to external loads locate at position with 100 mm apart along the x axis of the head of gear shaft. boundary conditions of primary shaft systems are shown in fig. 2. clamp boundary conditions of the base in the box of the primary shaft system are shown in fig. 2(a). external loads (torsion) locate at position typically on the right end of the primary shaft system is also shown in fig. 2(a). when the detail structural design of the gear train study is needed, the boundary conditions of upper pairs, gear train in the primary shaft system is shown in fig. 2(b). the clamp boundary conditions of box, shaft and gear train are specified in green color arrows in fig. 2 of the works, respectively. external loads (axial pressure) are typically located on the right end of the gear is also shown in fig. 2(b). (a) boundary conditions in the base of box (b) boundary conditions in upper pairs gear train fig. 2 boundary conditions of primary shaft systems 2.5. properly parameter of meshes it is critical to gain a mesh convergence and a good element before the cae simulation. for the convenience, grid-mesh with proper mesh controlled based on the size of parts, medium mesh density, and used to generate a proper grid-mesh. when the values of the results, e.g. maximum displacement changed less than or equal to 1e-06, thus the mesh would be in the good convergence condition, and this mesh of elements can be used for the further analysis of the cae simulations. a typical grid-mesh for the primary shaft systems is shown in fig. 3. with proper parameter element length based on standard mesh is used to generate a proper grid-mesh in the computation and analyses. grid-mesh in primary shaft system for gear shaft is shown in fig. 3(a). grid-mesh in the upper pairs and lower pairs, gear trains system are shown in figs. 3(b) and 3(c), respectively. proceedings of engineering and technology innovation, vol. 12, 2019, pp. 01-08 4 (a) grid-mesh in primary shaft systems (b) grid-mesh in the upper pairs gear train (c) grid-mesh in the lower pairs gear train fig. 3 typical grid-mesh for the primary shaft systems 3. results and discussion 3.1. static results due to external loads (a) stress for the shaft system due to axial 10mpa is parallel to x axis (b) displacement for the shaft system due to axial 10mpa is parallel to x axis (c) stress for the shaft system due to axial 10mpa is parallel to y axis (d) displacement for the shaft system due to axial 10mpa is parallel to y axis (e) stress for the shaft system due to axial 10mpa is parallel to z axis (f) displacement for the shaft system due to axial 10mpa is parallel to z axis fig. 4 stress and displacement for the shaft system due to axial 10mpa external loads proceedings of engineering and technology innovation, vol. 12, 2019, pp. 01-08 5 firstly, use the solidworks® 2014 simulation module to obtain the stresses and displacements of static results due to external axial pressure loads (10mpa) on the right end head of primary shaft systems. static stress and displacement results of the primary shaft due to axial pressure value 10mpa is perpendicular to y-z plane and parallel to x axis are shown in figs. 4(a) and 4(b), respectively. the maximum value 168mpa of stresses is found in the left pitch area of lower pairs, gear train and the maximum value 0.03352 mm of displacements is found in the top area of shaft head. static stress and displacement results of the primary shaft due to axial pressure value 10mpa is perpendicular to x-z plane and parallel to y axis are shown in figs. 4(c) and 4(d), respectively. the maximum value 19mpa of stresses is found in the right pitch area of lower pairs, gear train and the maximum value 0.008606 mm of displacements is found in the top area of shaft head. static stress and displacement results of the primary shaft due to axial pressure value 10mpa is perpendicular to x-y plane and parallel to z axis are shown in figs. 4(e) and 4(f), respectively. the maximum value 33mpa of stresses is found in the right pitch area of upper pairs gear train and the maximum value 0.009433 mm of displacements is found in the top area of shaft head. table 1 maximum stresses and displacements for the primary shaft external loads maximum stress maximum displacement when axial 10mpa is parallel to x axis 168mpa 0.03352mm when axial 10mpa is parallel to y axis 19mpa 0.008606mm when axial 10mpa is parallel to z axis 33mpa 0.009433mm when torsion 10000nm is parallel to x axis 72mpa 0.02728mm secondly, to obtain the static results of stresses and displacements due to external torsion loads (10000nm) on the right end head of primary shaft systems. static stress and displacement results of primary shaft systems due to torsion value 10000nm is parallel to x axis are shown in figs. 5(a) and 5(b), respectively. the maximum value 72mpa of stresses is found in the area of shaft head and the maximum value 0.02728 mm of displacements is found in the top area of shaft head. maximum stresses and displacements for the primary shaft due to external loads are listed in the table 1. the maximum stresses are found in the values smaller than the yield stress value 275mpa of cast iron material for the primary shaft system. it is suggested that the primary shaft can stand greater and over than external loads with pressure value 10mpa and torsion value 10000nm. it is also necessary to find the more detailed data for the design of lower pairs gear train systems, since the dominated, maximum value 168mpa of stresses is found in the left pitch area of the lower pairs gear train due to pressure value 10mpa and torsion value 10000nm. factor of safety in the design value is more than or equal to 1, it can also be considered as the static work by using the ratio value of maximum stress vs. external load. e.g. for the maximum stress value is 168mpa in the table 1: if the factor of safety value 1.5 has been used, then the value of allowed external load is 168mpa/ 1.5 and that is where 112mpa can be loaded; if the higher value of the factor of safety 2.0 has been used, then the smaller value of allowed external load is 168mpa/ 2.0 and that is where 84mpa can be loaded on the gear train systems. (a) stress for the shaft system due to torsion (10000nm) is parallel to x axis (b) displacement for the shaft system due to torsion (10000nm) is parallel to x axis fig. 5 stress and displacement for the shaft system due to torsion (10000nm) is parallel to x axis proceedings of engineering and technology innovation, vol. 12, 2019, pp. 01-08 6 3.2. static results for the gear train systems (a) stress for the upper pairs gear train due to axial 10mpa is parallel to x axis (b) displacement for the upper pairs gear train due to axial 10mpa is parallel to x axis (c) stress for the upper pairs gear train due to torsion (10000nm) is parallel to x axis (d) displacement for the upper pairs gear train due to torsion (10000nm) is parallel to x axis (e) stress for the lower pairs gear train due to axial 10mpa is parallel to x axis (f) displacement for the lower pairs gear train due to axial 10mpa is parallel to x axis (g) stress for the lower pairs gear train due to torsion (10000nm) is parallel to x axis (h) displacement for the lower pairs gear train due to torsion (10000nm) is parallel to x axis fig. 6 stress and displacement for the upper and lower pairs gear train due to external loads proceedings of engineering and technology innovation, vol. 12, 2019, pp. 01-08 7 for the precision and more detailed design of lower pairs gear train systems, it can also be used in solidworks® 2014 simulation module to obtain the stresses and displacements of static results due to external axial pressure loads (10mpa) and torsion loads (10000nm) on the gear train systems. static stress and displacement results of upper pairs gear train due to axial pressure value 10mpa is parallel to x axis are shown in figs. 6(a) and 6(b), respectively. the maximum value 19.8mpa of stresses are found in the upper bearing area of upper pairs gear train and the maximum value 0.00694 mm of displacements is found in the inner position of the upper bearing area of upper pairs gear train. static stress and displacement results of upper pairs gear train due to axial torsion value 10000nm is parallel to x axis are shown in figs. 6(c) and 6(d), respectively. the maximum value 58.7mpa of stresses are found in the inner position of the upper bearing area of upper pairs gear train and the maximum value 0.007076 mm of displacements is found in the upper area of upper pairs gear train. static stress and displacement results of lower pairs gear train due to axial pressure value 10mpa is parallel to x axis are shown in figs. 6(e) and 6(f), respectively. the maximum value 32.9mpa of stresses is found in the upper bearing area of lower pairs gear train and the maximum value 0.005053 mm of displacements is found in the outer position of the upper bearing area of lower pairs gear train. static stress and displacement results of lower pairs gear train due to axial torsion value 10000nm is parallel to x axis are shown in figs. 6(g) and 6(h), respectively. the maximum value 1327mpa of stresses is found in the upper bearing area of lower pairs gear train and the maximum value 3.198 mm of displacements is found in the outer position of the upper gear area of the lower pairs gear train. the maximum stress value 1327mpa is found in the value greater than the yield stress value 620mpa of steel alloy material for the gear train systems. it is suggested that the gear train system cannot stand greater and over than external loads with torsion value 10000nm. in the linear value of calculation under the torsion value 4672nm for the gear train system of the cnc machine can be considered in safety condition. 4. conclusions the static results of stresses and displacements of a primary shaft system of cnc machines are obtained with the solidworks® 2014 simulation module. the static maximum stress and displacement for the types of clamp boundary conditions of primary shaft and gear trains are studied. it is desired to compute and find the most of the external load of primary shaft systems in the cnc machine, which can be used to operate in safety condition under its yield stress value of materials. the maximum values of stress due to the effect of typical external loads e.g. axial pressure and axial torsion are all smaller than yield stress value, so the cnc machine can be considered in safety condition. the combined effect of axial and torsion loads can be considered in the linear work by summing the corresponding results of divisional load. factor of safety in the design value is more than or equal to 1 can also be considered in the static work by using the ratio value of maximum stress vs. external load. to provide some highlight results in detailed as follows, (1) it is helpful for engineers to investigate great five-axis turning-milling complex cnc machine data under cae analysis, (2) commercial cae solution for the primary shaft systems under external pressure and torsion loads is provided, (3) the linear results are provided by using the simulation module of solidworks® . conflicts of interest the authors declare no conflict of interest. references [1] s. chowdhury and r. k. yedavalli, “vibration of high speed helical geared shaft systems mounted on rigid bearings,” international journal of mechanical sciences, vol. 142-143, pp. 176-190, 2018. [2] m. eftekhari, a. d. rahmatabadi, and a. mazidi, “nonlinear vibration of in-extensional rotating shaft under electromagnetic load,” mechanism and machine theory, vol. 121, pp. 42-58, 2018. [3] j. zhang, y. chen, b. xu, m. pan, and q. chao, “effects of splined shaft bending rigidity on cylinder tilt behaviour for high-speed electro-hydrostatic actuator pumps,” chinese journal of aeronautics, in press. proceedings of engineering and technology innovation, vol. 12, 2019, pp. 01-08 8 [4] s. chowdhury and r. k. yedavalli, “dynamics of low speed geared shaft systems mounted on rigid bearings,” mechanism and machine theory, vol. 112, pp. 123-144, 2017. [5] x. huang, z. ni, z. zhang, and h. hua, “stiffness optimization of marine propulsion shafting system by frf-based substructuring method and sensitivity analysis,” ocean engineering, vol. 144, pp. 243-256, 2017. [6] d. zou, l. liu, z. rao, and n. ta, “coupled longitudinal–transverse dynamics of a marine propulsion shafting under primary and internal resonances,” journal of sound and vibration, vol. 372, pp. 299-316, 2016. [7] g. bulut, “dynamic stability analysis of torsional vibrations of a shaft system connected by a hooke's joint through a continuous system model,” journal of sound and vibration, vol. 333, pp. 3691-3701, 2014. [8] q. han, j. zhao, w. lu, z. peng, and f. chu, “steady-state response of a geared rotor system with slant cracked shaft and time-varying mesh stiffness,” communications in nonlinear science and numerical simulation, vol.19, pp. 1156-1174, 2014. [9] s. lahriri and i. f. santos, “theoretical modelling, analysis and validation of the shaft motion and dynamic forces during rotor– stator contact,” journal of sound and vibration, vol. 332, pp. 6359-6376, 2013. [10] a. a. adekunle, s. b. adejuyigbe, and o. t. arulogun, “development of cad software for shaft under various loading conditions,” procedia engineering 38, 2012, pp. 1962-1983. [11] m. shahgholi and s. e. khadem, “primary and parametric resonances of asymmetrical rotating shafts with stretching nonlinearity,” mechanism and machine theory, vol. 51, pp. 131-144, 2012. [12] y. lin and f. chu, “the dynamic behavior of a rotor system with a slant crack on the shaft,” mechanical systems and signal processing, vol. 24, pp. 522-545, 2010. [13] s. a. a. hosseini and s. e. khadem, “combination resonances in a rotating shaft,” mechanism and machine theory, vol. 44, pp. 1535-1547, 2009. [14] c. w. bert and c. d. kim, “analysis of buckling of hollow laminated composite drive shafts,” composites science and technology, vol. 53, pp. 343-351, 1995. [15] c. c. hong, c. l. chang, c. c. huang, c. c. yang, and c. y. lin, “cae analysis of secondary shaft systems in great five-axis turning-milling complex cnc machine,” advance in technology innovation, vol. 3, pp. 43-50, 2018. [16] c. c. hong, c. l. chang, and c. y. lin, “dynamic structural analysis of great five-axis turning-milling complex cnc machine,” global journal of researches in engineering: a, mechanical and mechanics engineering, vol. 17, pp. 1-8, 2017. [17] 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, pp. 1971-1984, 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/).  proceedings of engineering and technology innovation, vol. 15, 2020, pp. 35 41 ultra-violet treatment for fermenting low-salt soya sauce sophia ferng 1 , i-hui wu 1 , ang-yen li 2 , cheng-kuang hsu 1 , robin yih-yuan chiou 1 , ching-hua ting 2,* 1 department of food science, national chiayi university, chiayi, taiwan 2 department of mechanical and energy engineering, national chiayi university, chiayi, taiwan received 22 april 2019; received in revised form 17 may 2019; accepted 26 august 2019 doi: https://doi.org/10.46604/peti.2020.4117 abstract low-salt soya sauce has become a market trend due to consumers’ demand for a low sodium diet life. in tradition, a low-salt soya sauce (with salt concentration below 14.4%) is made from a high-salt one (18% salt concentration) through diluting or reducing the sodium content. the post-processing deteriorates the quality of the soya sauce produce as some specific, beneficial chemical components are inevitably removed. in the production of a native-born low-salt soya sauce, a key problem encountered is possible microbial contamination that easily develops in a low salt environment. in this study, we evaluated the effect of ultra-violet (uvc 254 nm) irradiation on soya mash of 12% salt concentration fermented at 35°c. the ultra-violet treatment could effectively prevent the soya mash from microbial contamination. keywords: soya sauce, law salt, ultra-violet treatment, native-born 1. introduction soya sauce was invented by the chinese about 3500 years ago. it has become a popular condiment on a table or seasoning in cooking in the oriental countries, as it not only provides a delicious flavor to the food but also promotes digestion to the body [1]. it can be made from a mixture of soya beans and wheat using a well-established two-step fermentation process [2], i.e. koji and moromi fermentation. in moromi fermentation, the mash can be easily spoiled because of microbial contamination by unexpected microorganisms that are carried by the materials or mostly come from the atmosphere [3]. to avoid microbial contamination during fermentation, the mash fermentation is normally conducted in 18-20% brine. the resulting broth contains a very high concentration of salt. this leads to a soya sauce produce rich of sodium that may incur cardiovascular and kidney diseases to the human body after serving. for health, soya sauce with a salt content less than 130g/l is recommended [4]. the regular industrial procedure of making low-salt soya sauce is to make a high-salt (18%) broth and then dilute the broth or reduce the sodium content with certain physical or chemical operations [4]. both approaches will inevitably deteriorate the quality of the low-salt soya sauce produce as the flavor chemicals and nutrient components are inevitably removed in post-processing. in tradition, a moromi pot is placed outdoors for fermentation that may take four months to one year [5-6], depending on the climate. the moromi placed outdoors is vulnerable to microbial contamination. outdoor fermentation is unsuitable for the production of low-salt soya sauce as unexpected microorganisms can grow easily in a low-salt environment. also, this lengthy fermentation time occupies a big portion of the cost in soya sauce production. the industry has devoted efforts in shortening * corresponding author. e-mail address: cting@mail.ncyu.edu.tw tel.: +886-5-2717642; fax: +886-5-2717561 proceedings of engineering and technology innovation, vol. 15, 2020, pp. 35 41 36 the time of fermentation. chemical soya sauce is a typical example in which, the amino acid, an important component of soya sauce is hydrolyzed from soya bean using hydrochloric acid. amino acid is then conditioned with flavoring compounds and browned with chemicals to produce soya sauce [7]. humidity and temperature are two climatic factors that dominate the enzymatic activation and microbial metabolism in moromi fermentation. moromi is sunk in brine, an uncontrollable, high humid environment. the temperature is a parameter sensible to the ambiance which varies hour by hour, day by day, and season by season. the enzymatic activation and microbial metabolism at different phases of fermentation could progress efficiently and nicely under a stable, adequate fermentation temperature profile [8]. the fermentation of moromi could be shortened to three to four months with less or no unexpected microbial contamination under a temperature-controlled environment [9-10]. a temperature-controlled environment has demonstrated its superiority in preventing hazardous microbial contamination [10]. this is because the enzymatic conversion of proteins may act rigorously in the first few days under an adequate fermentation temperature [11]. most microbes do not survive in a high-salt environment. however, the surface of moromi can be easily contaminated by ambient bacteria as the surface is not sunk in brine and is exposed to the air [3]. to prevent contamination by the microbes from the air, in this study, ultraviolet irradiation at the wavelength 254 nm (i.e. uvc) was used to eradicate possible microorganisms on the moromi surface. this is beneficial as ultra-violet light does not penetrate into the surface of moromi other than clear water [12]. to evaluate the efficacy of uvc treatment on the production of low-salt soya sauce, moromi of 18% and 12% salt concentrations were used as a control group while another 12% moromi treated with uvc irradiation as the experimental group. the counts of yeast, total viable bacteria, and bacillus spp. bacteria were examined every two weeks during three months of fermentation to justify the performance of uvc treatment on production of low-salt soya sauce. 2. materials and methods 2.1. raw materials and chemicals fresh soya bean and wheat once bought were immediately stored in a refrigerator at 4°c prior to using. the soya bean (glycine max shi-shi) was cultivated by national chiayi university, taiwan and the wheat was imported from canada. the koji mold was aspergillus oryzae (chuan feng microbe, taiwan). all chemicals used were of analytical grade. 2.2. koji fermentation raw soya bean was first washed and soaked in 4°c water for 24 hours and then steamed at 121°c for 20 minutes. wheat was roasted at 140°c for 30 minutes and then cracked into four to five pieces per kernel that may accompany with small particles (i.e. wheat flour). roasted wheat and steamed soya bean were mixed at a ratio of 1:2 (w/w). the mixture was then inoculated with 0.1% (w/w) of aspergillus oryzae (ca. 108 spores g/l), 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 in thickness and incubated at 30 ± 3°c. koji was harvested when the culture turns into greenish-yellow in color. this process may take 60-72 hours. the koji produce was stored at 4°c before subsequent moromi fermentation. 2.3. moromi fermentation fig. 1 illustrates the setup of the temperature-controlled fermentation system [10]. both experimental and control groups used 20-l pottery tanks for moromi fermentation. moromi tanks were bathed in a stainless tank (diameter 84 cm and height 65 cm) full of circulating water for temperature control. temperatures inside and outside of the pottery tank were measured for on-line fermentation temperature control and off-line analysis. our previous experimentation showed no temperature gradient in the mash at steady state. hence the mash had only its central temperature being measured. the circulating water was heated to 45°c with a heat pump. the fermentation temperature was instructed to maintain at 35°c, a compromised temperature proceedings of engineering and technology innovation, vol. 15, 2020, pp. 35 41 37 heat pump plc ipc solenoid valve pump temp module water soya mash temp-controlled tank pot water storage tank s s atm temp soya mash temp logger fig. 1 setup of the moromi fermentation system [10] suitable for yeast, lactic acid bacteria, and aspergillus oryzae, but not for hazardous bacteria or film yeast (pichia spp.) [9, 13]. a programmable logic controller (plc) performs controls and an industrial personal computer (ipc) executes system monitoring and data acquisition. fermented koji was put in the pottery tanks and then mixed with 18% or 12% brine at a weight ratio of 1:2. the tanks were covered in compliance with industrial practice except allowing the insertion of thermocouples for temperature measuring. the temperature-controlled system was instructed to maintain the central temperature of the mash at 35±1°c. during the first week of fermentation, the mash was stirred for three minutes once every day and then once a week for the remaining period of fermentation. both groups had a microbial examination every two weeks till the fermentation completed three months after [10]. fig. 2 illustrates the setup of moromi fermentation under uvc treatment. moromi in 12% brine was placed in a 20 -l pottery tank and the tank is covered with a stainless lid, on that, a uvc lamp (16w fam/10x25box, philips, holland) and a timer controller were mounted. the timer controls the timing and the duration of uvc irradiation. uvc dosage was therefore properly administered. the distance between the surface of the soya mash and the uvc lamp was well adjusted to allow the moromi surface with uniform uvc irradiation. (a) uvc setup (b) tank lid with a uvc lamp and a temperature sensor mounted (c) complete setup fig. 2 setup of soya mash fermentation under uvc treatment 2.4. microbial analysis 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 or below, at which yeast starts to propagate [14]. unexpected extra bacteria, mainly bacillus spp., may intrude from the ambience in moromi fermentation, form molds on the mash surface, and lead therefore to a stinking broth. thus, counting the growth facts of the total viable bacteria, the yeast, and the bacillus spp. bacteria unveils the progress of microbial metabolism and the efficacy of the uvc treatment in fermentation [15]. microbial analysis was operated every two weeks. the 10-fold-concentrated cells were serially diluted (10 0 to 10 -6 ) in a 0.85% sterile saline solution. to count the total viable bacteria, 100 µ l of the properly diluted sample was spread on plate count agar (pca), incubated at 30°c for 48 hours, proceedings of engineering and technology innovation, vol. 15, 2020, pp. 35 41 38 and then counted. for yeast counting, 100 µ l of the properly diluted sample was spread on yma (yeast mold agar), incubated at 30°c for 48 hours, and then counted. to count the bacillus spp. bacteria, 100 µ l of the properly diluted sample were spread on myp (mannitol egg yolk polymyxin agar), incubated at 30°c for 24-48 hours, and then counted. the colonies formed (25-250) were calculated as log cfu per milliliter of soya sauce [16]. 3. results 3.1. visual inspection of fermented moromi fig. 3 shows visual inspection of different mature moromi with and without uvc treatment. the moromi with 18% brine, an ordinary industrial formula, had the darkest color and no mold development on the surface and had the stickiest broth among all groups. the 12% moromi without uvc treatment had unwanted, hazardous, odorous, grey-greenish molds developed on the surface. moromi produces with such quality is considered failed. the 12% moromi with uvc treatment did not have mold development on the surface. the color was less dark and the broth was less sticky than the 18% one. mature moromi with such a quality can be processed for further soya sauce production. (a) the temperature-controlled tank (b) 18% brine (c) 12% brine (d) 12% brine with uvc fig. 3 visual inspection of different mature moromi 3.2. optimum uvc dosage there is no existing guideline for a proper uvc dosage on the application of soya sauce fermentation. hence several treatment schemes were evaluated to arrive at optimum dosage. the moromi was exposed to uvc irradiation for one hour every 24 hours. this scheme resulted in a browning reaction on the moromi surface but no mold development was observed on the surface. the browning reaction was unwanted as the mash was “burned” by the uvc. the treatment was adjusted to 30 min irradiation every 12 hours. there was a bit mold development on the moromi surface, but no more browning reaction observed. the dosage was further adjusted to 30 minutes irradiation every six hours. this scheme was acceptable as neither surface contamination nor browning reaction was observed. 3.3. microbial analysis fig. (4)-(6) show the changes in yeast, total viable bacteria, and bacillus spp. bacteria counts in the course of moromi fermentation. the bacteria were counted every two weeks and broth samples were taken on the surface (suffixed with “s”) and in the center of the moromi (suffixed with “c”). the yeast counts (fig. 4) and the total viable bacteria counts (fig. 5) decreased slightly at the beginning of fermentation and then saturated. this trend was attributed to the inhibitory effects of salt on microbial growth and the low ph induced by lactic acid bacteria [17]. the yeast counts did not have a significant difference between the surface and the center positions. besides, there is no significant difference among all fermentation formula. in general, there were no significant differences in yeast, total viable bacteria, and bacillus supp. bacteria counts between sampling places and among fermentation schemes. the uvc treatment did not influence the development of known microorganisms in moromi fermentation. as shown in fig. 5, each fermentation schemes and sampling places did not have significant differences in the total viable bacteria counts. the surface of the 12% moromi was contaminated with hazardous bacteria and this lead therefor to visible proceedings of engineering and technology innovation, vol. 15, 2020, pp. 35 41 39 mold development and odorous smell. however, the contamination did not affect the total viable bacteria counts. this might be due to that the salt in the 12% formula, although much less than the 18% one, can effectively prevent the contaminating bacteria from further penetrating into the moromi. uvc irradiation does not penetrate the non-transparent object [12]. hence the viable bacteria in the mash were not affected by the uvc. fig. 6 shows the developments of bacillus spp. bacteria. the bacillus spp. counts in all schemes increased slightly at the beginning of fermentation and then saturated. there were no significant differences among each moromi schemes. bacillus spp. bacteria is probiotic, especially beneficial to the digestive system [18]. however, too much bacillus spp. bacteria in moromi produce disgusting flavor, reduces the utilization of nitrogen, and deteriorates the unique aroma of soya sauce produce. fig. 4 development of yeast counts in moromi fermentation fig. 5 development of total viable bacteria counts in moromi fermentation fig. 6 development of bacillus spp. counts in moromi fermentation 5.0 5.2 5.4 5.6 5.8 6.0 6.2 6.4 6.6 6.8 7.0 0 week 4 weeks 6 weeks 8 weeks 10 weeks 12 weeks y e a st ( lo g c f u /m l) fermentation time (week) 18%s 18%c 12%s 12%c 12%+uvc s 12%+uvc c 5.00 5.20 5.40 5.60 5.80 6.00 6.20 6.40 6.60 6.80 7.00 0 week 4 weeks 6 weeks 8 weeks 10 weeks 12 weeks t o ta l v ia b le b a c te ri a ( lo g c f u /m l) fermentation tme (week) 18%s 18%c 12%s 12%c 12%+uvc s 12%+uvc c 4.00 4.50 5.00 5.50 6.00 6.50 7.00 2 weeks 4 weeks 6 weeks 8 weeks 10 weeks 12 weeks b a c il lu s su b ti li s (l o g c f u /m l) fermentation time (week) 18%s 18%c 12%s 12%c 12%+uvc s 12%+uvc c proceedings of engineering and technology innovation, vol. 15, 2020, pp. 35 41 40 4. conclusions the purpose of using 18% bine in moromi fermentation is to inhibit possible hazardous bacteria development as most unwanted bacteria are intolerable to a salt-rich solution. reducing the salt concentration from 18% to 12% incurs the risk of bacterial contamination. the 12% moromi had serious contamination on the surface, which was unusable in the production of soya sauce. when another 12% moromi was treated with uvc irradiation, microbial contamination was effectively diminished. microbial analysis showed that moromi fermentation with uvc treatment did not have a side effect on microbial growth. the optimized uvc dosage (30 minutes irradiation every six hours) could effectively inhibit bacteria development on the surface of moromi and did not induce browning reaction. this dosage can be used for a future application for production of low-salt soya sauce with uvc treatment. conflicts of interest the authors declare no conflict of interest. acknowledgment this work was sponsored by the ministry of science and technology of the roc (taiwan) government under the grant most 105-2221-e-415-030. references [1] d. fukushima, “soy proteins for foods centering around soya sauce and tofu,” journal of the american oil chemists society, vol. 58, no. 3, pp. 346-354, march 1981. [2] t. yokotsuka, “soya sauce biochemistry,” advances in food research, vol. 30, pp. 195-329, 1986. [3] w. h. huang, “influences of a temperature controlled system powered by solar energy and an uv treatment on the quality of soy sauce,” master dissertation, department of food science, national chiayi university, chiayi, taiwan, 2016. [4] j. luo, l. ding, x. chen, and y. wan, “desalination of soy sauce by nanofiltration,” separation and purification technology, vol. 66, no. 3, 429-437, may 2009. [5] 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. [6] y. zhu and j. tramper, “koji-where east meets west in fermentation,” biotechnology advances, vol. 31, no. 8, pp. 1448-1457, december 2013. [7] r. schueller, the soya sauce handbook: a reference manual for the food manufacturer, kikkoman corporation, japan, 1996. [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 soya sauce koji,” journal of agricultural and food chemistry, vol. 53, no. 5, pp. 1521-1525, april 2005. [9] f. m. yong and b. j. b. wood, “microbiology and biochemistry of soya sauce fermentation,” advances in applied microbiology, vol. 17, pp. 157-194, 1974. [10] s. ferng, w. h. huang, c. p. wu, y. t. lu, c. k. hsu, r. y. y. chiou, and c. h. ting, “a controlled fermentation environment for producing quality soya sauce,” advances in technology innovation, vol. 4, no. 1, pp. 11-20, january 2019. [11] n. x. hoang, s. ferng, c. h. ting, y. c. lu, y. f. yeh, y. r. lai, and c. k. hsu, “effect of initial 5 days fermentation under low salt condition on the quality of soy sauce,” lwt-food science and technology, vol. 92, pp. 234-241, june 2018. [12] j. a. guerrero-beltrán and g. v. barbosa-cánovas, “review: advantages and limitations on processing foods by uv light,” food science and technology international, vol. 10, pp. 137-147, june 2004. [13] 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 soya sauce,” lwt-food science and technology, vol. 74, pp. 242-250, december 2016. [14] r. y. cui, j. zheng, j. 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. proceedings of engineering and technology innovation, vol. 15, 2020, pp. 35 41 41 [15] x. l. gao, c. cui, h. f. zhao, m. m. zhao, l. yang, and j. y. 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, august 2010. [16] m. h. kang, j. s. kim, s. j. park, and t. shibamoto, “a study on biological and chemical changes in fermented soybeans (meju), grown under organic vs non-organic conditions, during storage,” journal of agriculture and biological sciences, vol. 3, no. 2, pp. 278-286, march 2012. [17] y. yan, y. qian, f. ji, j. chen, and b. han, “microbial composition during chinese soya sauce koji-making based on culture dependent and independent methods,” food microbiology, vol. 34, no. 1, pp. 189-195, may 2013. [18] s. lee, j. lee, y. i. jin, j. c. jeong, y. h. chang, y. lee, y. jeong, and m. kim, “probiotic characteristics of bacillus strains isolated from korean traditional soy sauce,” lwt-food science and technology, vol. 79, pp. 518-524, june 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/).  proceedings of engineering and technology innovation, vol. 12, 2019, pp. 09-20 performance characterization of surface quality and tool wear of wet/dry drilling on steels by using coated drill bits a. bovas herbert bejaxhin 1,* , g. paulraj 2 1 department of mechanical engineering, srm trp engineering college, irungalur, trichy, india 2 department of mechanical engineering, vel tech owned by r.s. trust, avadi, chennai, tamil nadu, india received 15 june 2018; received in revised form 05august 2018; accepted 10 september 2018 abstract the creation of holes by the drilling operation is a normal process which we were followed from the earlier days. especially the drill bits’ usage is an important consideration among the different drilling operations. here we considered the coated form of drill bits for the purpose of improving the surface quality and to overcome the tool wear by the particulate deposition over the surface of the specimen. the drilling features are to find out over the specimen en 8 alloy steel and mild steel which are experimentally verified. in the view of differentiating the eminence of surface roughness and tool wear an appearance of alloy steel and mild steel specimens, the wet and dry conditioned drilling operations are performed by the manifestation of coolants. the importance of pertura and latuma tool coated drill bits developed by the physical vapour deposition (pvd) method are used for this operation. based on the drilling conditions whether usage or non-usage of coolants is majorly influenced on surface roughness, and the tool coatings are politely influenced on the surface roughness of each specimen have been found out. likewise, the tool coatings and the rotational speed influenced more on machining time has been recognized. tool wear patterns are categorized and compared with the simulation data of drilling in deform 3d. effective stress is identified and related to the value of surface coarseness in both conditions of drilling on different steel specimens. keywords: drilling, coating, surface roughness, tool wear 1. introduction in recent days, the tool wear and surface quality attainment are the challengeable task with an optimum level of satisfaction. to overcome this basic problem among the drilling operation, we can easily utilize some innovative ideas with less effective manner. this effective and dynamic outcome over the drilling operation with the coated tools has been implemented through the effects of drilling parameters which is utilized by taguchi's optimization. it was found that the feed and speed are important process parameters to control surface roughness, tool wear, metal removal rate, and hole diameter error. the frictional contact was initiated with the specimen while the tool has been rotated with the feed rate, the cutting profile over the specimen was placed closely. thus, the suitable combination of rotational speed and feed rate can be affected by the quality of the holes and a number of factors also plays an important role in hole quality [1]. deformation damage of gfrp during high speed and ultrasonic assisted drilling were discussed by the researchers for finding the most influential factor on peel up and push out the delamination of gfrp. the push out delamination is more severe than that of peel up by the contribution of thrust force [2]. for medium largely sized drill bits influenced least by the spindle speed and feed on gfrp composites. the influence of cutting parameters on delamination in drilling and the feed rate of the highly influential *corresponding author. e-mail address: herbert.mech2007@gmail.com tel.: +91 9842427770, 9952831242 proceedings of engineering and technology innovation, vol. 12, 2019, pp. 09-20 10 parameter on the delamination in drilling of sandwiches composites. medium level of feed and speed were recommended [3]. in the drilling operation of an al5005, the effect of drilling parameters on surface roughness and built up edge whereas the surface roughness was decreased with the increase of rotational speed and point angle while it is increased with the increase of feed rate and drill diameter [4]. the discussion was about the surface roughness and roundness error was measured on the holes drilled with the cryogenically treated drill bits on aisi austenitic stainless-steel specimen. this treated tool gets hardened more in its physical characteristics and influenced on the cutting constraints. however, the travel distance of each cutting point of the tooltips is closer over the specimen surface without any frictional effects on the tool surface. the cutting speed has a significant effect on surface roughness as well as the feed rate also influenced effects on roundness error [5]. the effects of the drilling parameter to get the lowest torque value, the thickness of the specimen and drill bit size are the significant parameters influencing torque were discussed [6]. the burr size of the drilling of al-sic composites, the most significant parameters of feed rate, point angle, and concentrations of reinforcements were identified the burr height and thickness reduced by 25-40% of the increase of point angle. thus, the higher the point angle is desirable for minimizing the burns produced during the drilling of al-sic composites [7]. sem images clearly indicate the matrix delamination and microcracks. the effect of drilling parameters on thrust force and surface roughness, tool wear patterns are discovered at al-sic composite. thrust force calculated by using tools dynamometer and the tool wears also compared with the experimental data [8]. by comparing the drilling performance of titanium nitrate (tin) and titanium aluminium nitrate (tialn) coated high-speed steel (hss) drill bit for the machining on en8 material under dry machining conditions. the better tool life was achieved because of high hardness coating with the reduced wear rate during the work tool interaction. here tool life was proved better in this tialn coated drill bit compared with tin and uncoated drills for the constant tribology conditions [9]. the effects of cutting parameters speed, feed, depth of cut on metal removal rate and temperature of en8 carbon steel was discussed. rsm method of optimization helps to increase the quality and reduce the manual effort, production cost and machining time [10]. the experimental verification showed that the influence of drill parameters point angle, clearance angle, cutting speed, feed and drill diameters on the thrust force and torque in drilling. holes drilled on al6061 alloy, the cutting speed, and point angle are the significant factor for thrust force similarly for the torque [11]. the analysis on the effect of machining parameters on thrust force and surface roughness of drilled cfrp composite specimens were discussed. it is evaluated that the reduction in thrust force reduces the induced delamination, surface roughness, and the drilling quality has improved. better quality can be achieved by using high-speed feed and point angle [12]. high speed and feed with small diameter drill type used to obtain good ra and mrr. the effects of drilling parameter were accomplished the burr height and surface roughness was evaluated using three different drill types, the less surface roughness were recorded as by using tin and carbide drill bits at the low feed rate and higher cutting speed. less burr height was obtained in high rotational speed and carbide drill type [13]. the fatigue stress concentration factor (kf) of the machined surfaces determined from experiments and its predictions for the effective fatigue stress concentration using the arola-ramulu model was within 2% of the apparent fatigue stress concentration factors estimated from experimental results. if the fatigue stress concentration factor (kf) is increasing the surface roughness also increased [14]. the temperature interaction in between the tool and the work area was improved by using the carbon-oriented coatings on the drilling tools which contains graphite particles or a diamond. they showed that the performance behavior of the coated drills without the cutting fluids was better to compare with the uncoated drill bits. the machining was affected without metal removal fluids (mrf). but in this case, the surface was not affected with or without mrf the dlc and blend coated drill bits performed well in the heat affected zone during drilling with less power consumption [15]. the carbon fiber reinforcement plastic (cfrp) specimen drilled with the composite coated diamond films deposited over the drill bits by the cvd method. here they confirmed that the shape of the tool acts as a shaping factor that influence d proceedings of engineering and technology innovation, vol. 12, 2019, pp. 09-20 11 more on drilling force and hole quality for coated as well as uncoated tools. the hard coatings improved the tool life and constancy of drill holes [16]. the performance evaluation has been compared among the diamond coated and uncoated carbide drills on the drilling of aluminum silicon alloys (a356). the usage of lubrication has to be minimized and it can be verified through the experimentation. here they showed that the minimal lubricant usage cannot be improved in hole quality, wear rates etc. there was no change has been identified in the drilling quality excluding the irregular wear can be occurred on the surface of the diamond coated drill bits [17]. moreover the various developments in the drilling operations were obtained now a day. some advancements were utilized in the mining operations with automated drilling processes. especially, auto mine rotary drilling and the surface drilling are the advance cum familiar process used in the field of mining which is having the key benefits of more efficiency, productivity, fast and reliable control, utilization and safety. the operator is exactly out if hazardous atmosphere, reduced from the noise and vibration problems. the complete automation increased the productivity and the stress relief over the machines. the safety dovetails avert some accidental or hazardous actions. the accurate drill positioning was maintained here. multi drill control also has been incorporated with the single operating console. the purpose of this paper is to perform the drilling operation for various speeds at different machining conditions. the drilling operation has performed by using the coated drill bits and evaluate the surface finish and tool wear rates. the betterment of results has been declared based on the optimization tools. the tool characterization helps to identify the reasons for tool wear and its effects on surface morphology. the possibility of new methods will be demonstrated and confirmed by the machining simulations from deform 3d. 2. materials and methods drilling is an important operation in the assemblage of automobiles structural frames and aircraft. the lifespan of the joint can be critically affected by the quality of the drilled holes. especially borehole instability and hole deviation will create huge problems during the assemblage of various parts in the industry sector. to overcome these surface-oriented problems, material selection and the usage of types of drill tool to be more concentrated. here the drilling operation was performed in universal radial drilling machine tuv-nabcb-qm002 (ser25) powered 1 hp with the high rotational speed range of 1440 rpm. for the different rotational speeds, the drilling operation has been performed on both steel specimens. during wet conditions, the coolant is used and it is not during the dry cutting operation. the machining conditions are shown in table 1. the drilling operation was performed on alloy steel and mild steel rectangular shaped bar of cross section 300 x 50 x 50 mm as shown in fig. 2. the 10 mm diameter high-speed steel (hss) drill bits (jis-skd61) as shown in figs.1(a) and (b). the two conditions were to be followed (i.e) wet and dry cutting for the drilling operation. the coolant oil used for the machining process is called wet conditioned machining and when not using the coolant means its dry conditioned drilling. (a) coated drill bits (b) drilled specimen fig. 1 photography of tool and the specimens proceedings of engineering and technology innovation, vol. 12, 2019, pp. 09-20 12 table 1 machining conditions of the drilling operation sl.no machining conditions 1 condition wet, dry 2 speed 500 rpm, 1000 rpm 3 feed rate 0.2 mm/rev (constant) 4 drill tool material ϕ10mm hss drill bit 5 coatings & thickness pertura & latuma & t= 4μm work material en8 alloy steel, mild steel bar table 2 composition of en 8 alloy steel material (en8/aisi1040) c mn p s si fe percentage (%) 0.36-0.44 0.60-1.00 0.05 0.005 0.10-0.40 remains table 3 chemical composition of mild steel material (ms/aisi1018) c cu mn p si s fe percentage (%) 0.25-0.29 0.20 1.03 0.04 0.28 0.05 98.0 the new tool coatings were introduced over the drill bits called as latuma and pertura. this well bonded physical vapour deposition method of coatings was deposited over the tool cutting profiles with the thickness of 4μm were performed in oerlikon balzers coating india pvt. ltd. the pertura and latuma coating are name wise under the brands of pvd coating which is developed over the tools and the components. the latuma coating is the newest source of technology which contains a high amount of aluminum with excellent oxidation resistance and hot hardness. it is having excellent resistance of crater wear and also used for balancing the residual stress and coating hardness. its productivity is good with high variables of machining conditions. it is well suited for the drilling with carbide and hss drills. pertura is a nano-layer structure promises the best balancing of residual stress, hardness, and fracture toughness and thus dependably stops crack propagation. this allows higher cutting speeds than with other coatings as well as protracted tool life. it is used for deep hole drilling by using minimum quantity lubrication. latuma and pertura coating have been applied on the surface of the drilling tools with the standard coating thickness within the ranges of 3-6 μm for the 10 mm diameter high-speed steel (hss) drilling tools which were maintained and performed in oerlikon balzers coating india pvt. ltd. the drilling operation was performed on the specimen with the low and high cutting speed of 500 rpm and 1000 rpm at constant feed rate as 0.2 mm/rev is mentioned in table 1. the chemical composition of en 8 alloy steel and mild steel are mentioned here below in tables 2 and 3. the percentage of composition showed clearly the manganese and silicon particulates are involved more in both ms steel and alloy steels. the carbon percentage is low in mild steel and high in alloy steel based on its density and volume contribution. the machinability was good in both alloy and mild steel components. here in this work, the surface roughness and tool wear have to be calculated and compared with the predictive values of dynamic simulation results. 3. experimental methods the experiments were carried out on a universal radial drilling machine tuv-nabcb-qm002 made from siddhapua enterprise. by using coated drill tools, the through all holes were performed with the uniform gap of 20 mm in the specimens as shown in figs. 2(a), (b) and (c). the uniform feed rate is maintained for all the experiments. the chip formation is good and it is in the form of continuous chips without any damages. the results of experimentation were evaluated after drilling in terms of the following measured machining performance: surface roughness (ra), burr height (h). the surface roughness of drilled holes was measured by using a mitutoyo sj-201 surface roughness tester as shown below in fig. 3. the instrument was set to a required length and the numbers of specimens were measured. four directional measurements of surface roughness were taken perpendicular and parallel to the whole circumference. proceedings of engineering and technology innovation, vol. 12, 2019, pp. 09-20 13 (a) universal radial drilling machine (b) drilling without coolants (c) drilling with coolants fig. 2 drilling operation on mild steel and alloy steel with coated drill bits the surface roughness values given in this work are the mathematical average of four measurements taken from inside the same hole surface. the optical profile projector was used for the burr height measurements for some delaminated holes. the optical systems used in this projector used to produce the fringe patterns of the hole burr outline surface. the dark fringes were identified and denoted the height of the burrs in the edges of the drilled holes. fig. 3 surface roughness measurement on drilled holes by mitutoyo sj-201 the tool wear was identified optically using a tool maker’s microscope. the tip of the drill bits wears out mildly due to the dry cutting without using coolant. this wear length was measured by using the micrometer provided in the x and y axis of the tool maker’s microscope. various tip tool wear images were captured and showed in figs. 7(a) and(b). also here in surface roughness and machining time from the main effects plot at figs. 4(a) and (c), the machining conditions either wet or dry and the tool coatings are the primary and secondary influencing factors on surface roughness. the minimum value is the better option from the surface roughness responses were well predicted the dominant factors for the betterment of surface quality. the responses were spread over the residual v.s. fits chart for roughness and machining time as shown in figs. 4(b) and (d). proceedings of engineering and technology innovation, vol. 12, 2019, pp. 09-20 14 (a) the main effects plot for means of machining time (b) residual vs fits for machining time (c) the main effects plot for means of surface roughness (d) residual vs fits for surface roughness fig. 4 means and normal probability chart of machining time (minutes) and surface roughness (μm) 4. results and discussion after the completion of drilling on mild steel and alloy steel by using coated drill bits, the required output responses were analyzed and measured the surface roughness and machining time. the main effects plots were identified from the optimization techniques. the fits verses residual has been plotted based on the array formation in taguchi’s method. the four variable material, machining condition, rotational speeds, and the coating types were considered for finding the mean values of the machining time and surface roughness. the most dominant parameters have easily identified these observations from the figs. 4(a), (b), (c) and (d). here the machining time saved due to the rotational speed and the tool coating as the foremost parameter. especially the latuma coated drill bit performed well at the machining time at the higher rotational speed of 1000 rpm. speed and tool coating are the dominant factors in machining time were identified. the variables are lying nearby the boundary level and evenly spread over in residual response by the machining time in the normal probability chart. also here in surface roughness, from the main effects plot in fig. 4(c),the machining conditions and the tool coatings are the primary, and secondary influencing factors on surface roughness. the minimum value is the better option from the surface roughness responses were well predicted the dominant factors for the betterment of surface quality. the responses were spread over the residual v.s. fits chart. in this experiment, only two types of tool coatings were compared for the purpose of identifying the performance during drilling of different hardness materials like alloy steel and mild steel. so we have conducted the experiments based on the two variations of tool materials, speed, condition, and coating types. it is enough for the findings through l8 orthogonal array formation which helps in identify the significant effects on output response. proceedings of engineering and technology innovation, vol. 12, 2019, pp. 09-20 15 4.1. prediction of drilling performance the drilling dynamic performance was accomplished by using deform 3d dynamic simulator tool. some predicted values of stresses and temperatures for the above-mentioned machining conditions by using coated drill bits. simulated results from deform 3d were utilized for the confirmation of the experimental readings of stresses and displacements. the mesh generation was done on both the drilling tool and the specimen in the preprocessor stage machining. all the required fields of input parameters were entered in the preliminary steps of machining simulation of deform platform. (a) tool mesh generation (b) specimen mesh generation (c) positioned view of drilling before the simulation fig. 5 tool and workpiece mesh generation in deform 3d simulator in drilling machining the tool mesh generated by 10697 elements and 2830 nodes and for the workpiece mesh size of 13590 elements and 3161 nodes. the boundary conditions were applied to the workpiece as fixed and the drilling tool is rotated about the z axis as shown in fig. 5(c). the material properties were identified as aisi 1013 from the material library. the simulation controls are specified as a number of simulation steps and increments for the required drill depth of 5 mm. higher stresses were acted in between tool and workpiece due to the interaction effects. this was predicted thorough deform 3d simulator as mentioned in figs. 6(a), (b), (c) and (d). based on these predicted values wherever effective stress becomes more, the surface roughness will get poor outcomes. instead of using uncoated inserts, the coated inserts help to improve the surface properties and machining time in a very effective manner. the simulation results showed the effective stress of the tool work interaction and the temperature of the cutting profiles of both alloy and mild steel drilling operation. these predicted values of σeff and t clearly stated that wherever effective stress is more, the surface roughness become less in the drilled profiles. here the drilling of alloy steel gives the surface roughness of 0.841 μm is more than the ra of mild steel of 0.734 μm at the same rotational speeds of 500 rpm. the temperature distribution also becomes minimum as 30.7°c during the wet conditioned drilling of mild steel. consecutive through all holes were drilled along the surface of both alloy and mild steels with a uniform gap maintained between each hole. the fewer roughness values that mean good surface finish which had happened during the wet conditioned machining, especially with pertura, coated drilling tool achieved a minimum surface roughness of 0.734 μm at minimum speed and 1.368 μm at both low and high rotational speeds for the mild steel specimen. proceedings of engineering and technology innovation, vol. 12, 2019, pp. 09-20 16 (a) effective stress prediction (pertura coated tool) (b) temperature prediction (pertura coated tool) (c) effective stress prediction (latuma coated tool) (d) temperature prediction (latuma coated tool) fig. 6 deform 3d simulation results of drilling operation at 500 rpm proceedings of engineering and technology innovation, vol. 12, 2019, pp. 09-20 17 table 4 experimental results of machining time and surface roughness of mild steel material condition speed (rpm) coating machining time (min) surface roughness (μm) mild steel wet 500 pertura 3.41 0.734 wet 500 latuma 2.06 1.575 wet 1000 pertura 2.11 1.368 wet 1000 latuma 1.37 2.366 dry 500 pertura 2.53 2.128 dry 500 latuma 2.33 1.811 dry 1000 pertura 2.03 2.652 dry 1000 latuma 2.25 2.150 average 2.26 1.848 table 5 experimental results of machining time and surface roughness of alloy steel material condition speed (rpm) coating machining time (min) surface roughness (μm) alloy steel wet 500 pertura 3.36 1.640 wet 500 latuma 1.52 0.841 wet 1000 pertura 1.56 1.893 wet 1000 latuma 1.38 1.548 dry 500 pertura 3.11 3.859 dry 500 latuma 1.40 1.713 dry 1000 pertura 1.56 1.369 dry 1000 latuma 1.41 1.498 average 1.91 1.795 latuma coated drills is not good for dry conditioned machining on mild steel as the specimen. similarly, for the drilling of alloy steel, good surface finishes 0.841 μm was obtained during wet conditioned machining with latuma coated drill bits. the surface roughness of the fluted area in the coated tooltips were recorded around 0182 μm to 0.324 μm. here also the surface quality was achieved well at lower speeds. the machining condition and the rotational speeds are the major dominant factor here for surface roughness which was identified from figs. 4(a) and (c). comparatively the uncoated drill bits were verified with the coated hss drilling tools that they were given the higher value of surface roughness ranges from 2.868 μm to 3.417 μm for the drilled holes of both alloy and mild steel. thus, the hole inner surface quality becomes poor when it will be compared with the holes drilled by the coated tools. 4.2. effect of coating on surface roughness the drilling operation was performed by the pertura and latuma coated tools at different cutting conditions. the results revealed that the less machining time was obtained for the drilling of an alloy steel material at higher rotational speeds. latuma coating on drill bits plays an important role in the saving of machining time by the smooth interaction of tool and the specimen. the surface roughness values are measured by using mitutoyo sj-201 tester and mentioned in tables 4 and 5. consecutive through all holes were drilled along the surface of both alloy and mild steels with a uniform gap maintained between each hole. the fewer roughness values that mean good surface finish which had happened during the wet conditioned machining, especially with pertura, coated drilling tool achieved a minimum surface roughness of 0.734 μm at minimum speed and 1.368 μm at both low and high rotational speeds for the mild steel specimen. latuma coated drills is not good for dry conditio ned machining on mild steel as the specimen. similarly, for the drilling of alloy steel, good surface finishes 0.841 μm was obtained during wet conditioned machining with latuma coated drill bits. here also the surface quality was achieved well at lower speeds. the machining condition and the rotational speeds are the major dominant factor here for surface roughness which was identified from figs. 4 (c) and (d). 4.3. effect of coating on tool wear after the dry and wet conditioned drilling was performed on the alloy and mild steel specimens, the tool wear was captured on the different coated high-speed steel (hss) drill bits by using the tool maker’s microscope. the micrometer x and y readings used to show the tiny wear span of the tooltips and edges of hss drill bits as shown in figs. 7(a) and (b). proceedings of engineering and technology innovation, vol. 12, 2019, pp. 09-20 18 (a) wear length of latuma coated drill bit (b) wear length of pertura coated drill bit fig. 7 the span of tool wear images of coated high-speed steel drill bits by tool maker’s microscope the latuma and pertura pvd coatings were deposited warmly along the entire flute length of the drill bits. the margin wear was identified below the chisel edge of the pertura coated drill bits. it was measured vertically below the lip of the drill bit by using micrometer read markings in the tool maker’s microscope. the wearing length of 5.36 mm was observed along the rake face of the pertura coated drill bit as margin wear. similarly, the tiny margin wear on the rake face of the latuma coated drill bit was measured as 0.96 mm. the comparison was made among these two coated drilling tools, the coated drill bits achieved less in tool wear when it was involved wet and dry cutting conditions. there were no such wear markings were identified on latuma coated drill bits. it was measured vertically below the lip of the drill bit by using micrometer read markings in the tool maker’s microscope. the wearing length of 5.36 mm was observed along the rake face of the pertura coated drill bit as margin wear. similarly, the tiny margin wear on the rake face of the latuma coated drill bit was measured as 0.96 mm. the comparison was made among these two coated drilling tools, the coated drill bits achieved less in tool wear when it was involved in wet and dry cutting conditions. there were no such wear markings were identified on latuma coated drill bits. 4.4. interpretation of results the surface roughness becomes good in alloy steel because it is having naturally 0.36-0.44% of carbon is relatively high and also the silicon participation is maximum up to 0.4% compared with the mild steel. this will lead to achieving less surface roughness value. at the same time, coated drill bits were deposited by the pvd method with the additive of al and cr combined form particulates. so the chemical interaction and temperature distribution were occurred in between the coated tool and the inner surface of the hole. the slight deposition of al and cr particles from latuma and pertura coated drill bits leads to change the surface integrity with respect to the machining conditions. less tool wear was created from the latuma coated drill bits compare with the pertura coated drill bit. margin wear marks were identified in the flute rake face of the pertura coated drill bit as shown in fig.7(b). the comparison was made accordingly among the surface roughness of the specimen and the tool wear level of the drill bits has been demonstrated by the fig. 8. fig. 8 comparison of surface roughness and machining time 0 1 2 3 surface roughness (μm) machining time (min) 1.848 2.26 1.795 1.91 comparison of results mild steel proceedings of engineering and technology innovation, vol. 12, 2019, pp. 09-20 19 5. conclusions in this work, the surface quality was achieved with the contribution of latuma and pertura coated drill bits effectively. especially the holes were made on both alloy and mild steel during the wet conditioned drilling operation was very successful. the surface roughness decreased at a low rotational speed of 500 rpm in the drilling of mild steel with pertura coating. simultaneously the latuma coated tool achieved well during the wet machining of ally steel material. here the rotational speed and the tool coatings are the more dominant thing on surface roughness and machining time. the optimization results can be made the things very easy to identify the influencing elements through the sn ratio and main effects plots. the tool wear was analyzed on those latuma and pertura coated drill bits. very few wear images were obtained on latuma coated compare with the pertura coated drill bits. the wear span of latuma coated drill tip edge is very small rather than the pertura coated one. the dynamic simulation helps to predict and adopt a strong understanding of the stresses and temperatures of machining interactions. conflicts of interest the authors declare no conflict of interest. references [1] j. pradeep kumar and p. packiaraj, “effect of drilling parameters on surface roughness, tool wear, material removal rate, and hole diameter error in drilling of ohns,” international journal of advanced engineering research and studies, vol. 1, no. 3, pp. 150-154, june 2012. [2] b. v. kavada, a. b. pandey, m.v. tadavi, and h. c. jakharia, “a review paper on effects of drilling on glass fiber reinforced plastic,” iciame 2014, procedia technology, vol. 14, pp. 457-464, august 2014. [3] m. a. j. bosco, k. palanikumar, b. durga prasad, and a. velayudham, “influence of machining parameters on delamination in drilling of gfrparmour steel sandwich composites,” nuicone 2012, procedia engineering, vol. 51, pp. 758-763, january 2013. [4] e. bahce and c. ozel, “experimental investigation of the effect of machining parameters on the surface roughness and the formation of built up edge (bue) in the drilling of al 5005,” tribology in engineering, hasim pihtili, intechopen, chapter 2, pp. 15-28, may 2013. [5] cicek a. kıvak, t. samtas, g. adem cicek, turgay kıvak, and gurcan samtas, “application of taguchi method for surface roughness and roundness error in drilling of aisi 316 stainless steel,” journal of mechanical engineering, vol. 58, no. 3, pp. 165-174, january 2012. [6] n. s. mohan and s. m. kulkarni, “influence of drilling parameters on torque during drilling of gfrp composites using response surface methodology,” the 2nd international joint conference on science and technology (ijcst) pp. 1-7, september 2017. [7] a. a. thakre, “modelling of burr size in drilling of aluminium silicon carbide composites using response surface methodology,” an international journal of engineering science and technology, vol. 19, pp. 1199-1205, march 2016.. [8] x. chen, l. xie, x. nan, j. tian, and w. zhao, “experimental study of small hole drilling characteristics of sicp/al composites,” procedia cirp, vol. 46, march 2016 pp. 319-322. [9] puneeth. v and smitha. b, “studies on tool life and cutting forces for drilling operation using uncoated and coated hss tool,” international research journal of engineering and technology (irjet), vol. 4, no. 6, pp. 1949-1954, june 2017. [10] b. pradeep kumar, dr. n. v. n. indra kiran, and s. phani kumar. “effect of cutting parameters in drilling of en8 (080m40) carbon steel to obtain maximum mrr and minimum temperature by using rsm (under dry condition)” international journal of engineering and management, vol. 7, no. 2, pp. 533-539, april 2017. [11] r. sreenivasulu and c. s. rao, “effect of drilling parameters on thrust force and torque during drilling of aluminium 6061 alloy based on taguchi design of experiments,” journal of mechanical engineering, no. 46, pp. 41-48, december 2016. [12] d. shetty s, n. shetty, a. rajat, g. shetty, and p. patil, “characterization of machining parameters on thrust force and surface roughness in drilling of 40-60 wt. % bd cfrp composite,” international journal of applied engineering research, vol. 12, no. 16, pp. 5570-5577, september 2017. [13] n. tosun, “determination of optimum parameters for multi-performance characteristics in drilling by using grey relational analysis,” international journal of advanced manufacturing technology, vol. 28, pp. 450-455, march 2006. proceedings of engineering and technology innovation, vol. 12, 2019, pp. 09-20 20 [14] d. arola and c.l. williams, “estimating the fatigue stress concentration factor of machined surfaces,” international journal of fatigue, vol. 24, pp. 923-930, september 2002. [15] j. m. dasch, c. c. ang, c. a. wong, y. t. cheng, a. m. weiner, l. c. lev, and e. konca, “a comparison of five categories of carbon-based tool coatings for dry drilling of aluminum,” surface and coatings technology, vol. 200, pp. 2970-2977, february 2006. [16] x. wang, x. shen, c. zeng, and f. sun, “combined influences of tool shape and as-deposited diamond film on cutting performance of drills for cfrp machining,” surface and coatings technology, vol. 347, pp. 390-397, may 2018. [17] d. u. braga, a. e. diniz, g. w. a. miranda, and n. l. coppini, “using a minimum quantity of lubricant (mql) and a diamond coated tool in the drilling of aluminum-silicon alloys,” journal of materials processing technology, vol. 122, pp. 127-138, march 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/).  proceedings of engineering and technology innovation, vol. 5, 2017, pp. 41 44 segmented compression molding for composite manufacture chih-yuan chang * department of mechanical and automation engineering, kao yuan university, kaohsiung, taiwan, roc. received 02 july 2017; received in revised form 06 august 2017; accept ed 10 august 2017 abstract in the present study a new technique of vacuum assisted resin transfer molding (vartm), called segmented compression molding (scm), is proposed to improve filling process. in the method, a double -bag is placed between the upper mold and the cavity, and each bag is controlled individually. through the vacuum pulling the double-bag upward, resin is easily infused into the loose preform. after the completion of resin infusion, the resin inlet is closed. the vacuum on the bags are sequentially released to atmospheric pressure, the inflated bags squeeze the surplus resin from the wetted preform toward dry loose preform and the filling process is thus reduced. this study is to introduce the scm, model the filling process and explore the surface quality o f the scm parts. results show that the scm filling process is decreased by using a high initial cavity thickness. as compared with transfer resin transfer molding (rtm) and typical vartm, scm can reduce the filling process in the work. the surface roughnes s of scm part approximates that of vartm, but it is much inferior to that of rtm. keywords: vacuum assisted resin transfer molding, resin transfer molding, surface roughness 1. introduction polymer composites have been utilized in various industries such as aerospace and automobile. some typical processes are resin transfer molding (rtm), compression rtm (crtm) and vacuum assisted rtm (vartm) also known as resin infusion. for the sake of reducing filling time, various innovative methods have been developed recently. allende et al. [1] proposed the fast remotely actuated channels (fastrac) system consisting of primary, secondary bags and a metal plate with machined channels placed between two bags. a vacuum applied in the channel space to pull the primary bag away from the preform and resin was easily infused into the preform. alms and advani [2] proposed the flow flooding chamber (ffc) method which used a rigid chamber to replace the metal plate and secondary vacuum bag in fastrac. alms et al. [3] further presented the vacuum induced preform relaxation method which utilized a small vacuum chamber for guiding the liquid flow. heider et al. [4] proposed an injection design methodology in which multiple high -permeability fabrics were integrated periodically betwe en the thick-section and low-permeability preform. chang [5] proposed the vacuum-assisted crtm (vacrtm) process which was similar to ffc method. chang and li [6] utilized an innovative vacuum bag consisting of a film and several air cushions. the distribution channels were created between air cushions during infusion. chang and chen [7] proposed the progressive compression method (pcm), using a concept of the segmented compression to improve infusion time and material waste. * corresponding author. e-m ail address: y uan@ cc.ky u.edu.tw tel.: +886-7-6077028; fax: +886-7-6077112 proceedings of engineering and technology innovation, vol. 5, 2017, pp. 41 44 copyright © taeti 42 a new technique of vartm, called segmented compression molding (scm), is proposed to reduce the filling time. the technique shares the common concept with the double bag [1], and segmented compression [ 7]. fig. 1 illustrates the schematic diagram of scm. the technique provides the advantages in term of needless cleaning of upper mold and short filling time. the potential for low void content in the part is expected due to the cavity with vacuum assistance. in the present study, a one-dimensional filling model coupled with preform deformation is employed to investigate the filling process . the initiating timing of the primary bag compression is also discuss ed. (a) resin infusion (b) the secondary bag compression (c) the primary bag compression fig. 1 schematic diagram of scm 2. theory in the filling process of scm, the resin flow through a fiber preform is modeled as flow through a deformable porous media. the governing equations of resin flow are written as follows.     0      vh xt h   (1) x pk v     (2) where v , p, μ are resin velocity, pressure and viscosity; φ, h, k are the preform porosity, thickness and permeability, respectively. note that k varies with the preform deformation and the kozeny-carman model is employed to relate k with the preform porosity. during preform compaction, the resin pressure obeys the terzaghi’s law. an empirical power model is utilized to provide the relationship between φ and compression pressure exerted on the preform (pcom ). b compf v f v  0 (3) where vf=1-φ; vf0 and b are the material constants. proceedings of engineering and technology innovation, vol. 5, 2017, pp. 41 44 copyright © taeti 43 in the present numerical analysis , a finite difference method is utilized to solve the governing equations of resin flow. the process parameters are used: ambient pressure of 101.3 kpa, infusion pressure of 95 kpa, vent pressure of 1.3 kpa, mold cavity length of 1 m and resin viscosity of 0.2 pa-s. the properties of the preform are listed: kozeny-carman constant of 71.8× 10 -12 m 2 , filament density of 2540 kg/m 3 , vf0 of 0.11 and b of 0.126. 3. results and discussion three scm cases are used to investigate the scm filling process as listed in table 1. fig. 2 shows the filling time for various cases. evidently, a high initial cavity height can reduce the filling process by comparing case 1 with case 2. in scm, the premature initiation of the primary bag compression is necessary since the late compression of secondary bag is relatively slow. a comparison between case 2 and case 3 give an explanation that the appropriately premature initiation of the primary bag compression is preferable. as compared with rtm and typical vartm, case 3 can reduce the filling process by 13.4% and 50.1% respectively in the present study. table 1 process parameter of three scm cases, rtm and vartm -3 initial cavity height 1.3x10 mscm case 1 initiating timing of primary bag compression -31.5x10 m -31.5x10 m preform at the inlet is compacted to -30.95x10 m preform at the inlet is compacted to -30.95x10 m preform at the inlet is compacted to -31.3x10 m rtm -30.838x10 m final cavity height -30.838x10 m -30.838x10 m -30.838x10 m -30.838x10 m vartm -30.838x10 m scm case 2 scm case 3 case 1 case2 case3 rtm vartm f il li n g ti m e (s ) 0 2000 4000 6000 8000 10000 12000 14000 16000 i: needed time of resin infusion s: needed time of secondary bag compression p: needed time of primary bag compression t: total filling time t s t t p i p s i p s i i i p o s tin fu s io n fig. 2 filling time for various cases (a) scm part (b) vartm part fig. 3 surface quality of scm and vartm. proceedings of engineering and technology innovation, vol. 5, 2017, pp. 41 44 copyright © taeti 44 the surface quality of the scm part made from four layers of bi-directional mats (tgfw -600) and unsaturated polyester resin (2597pt-6) is evaluated. as expected, the surface flatness of the scm part is improved relative to pcm due to the primary bag flattening entire preform. fig. 3 shows the surface roughness of scm part supplied by mold. through the visual observation, it is found that the surface finish of scm supplied by the bag is not as good as the mold side surface due to the flexible bag. the phenomenon of the “print through” is obvious and needed to be improved for both surfaces . through the measurement of the surface roughness tester, the surface roughness of scm part approximates that of vartm but is much inferior to that of rtm. acknowledgements this work was supported by the ministry of science and technology of republic of china under grant most 105-2221-e-244 -006. references [1] m. allende, r. v. mohan, and s. m. walsh, “experimental and numerical analysis of flow behavior in the fastrac liquid composite manufacturing process ,” polymer compos ites, vol. 25, no. 4, pp. 384-396, july 2004. [2] j. alms and s. g. advani, “simulation and experimental validation of flow flooding chamber method of resin delivery in liquid composite molding,” composites part a: applied science and manufacturing , vol. 38, no. 10, pp. 2131-2141, october 2007. [3] j. b. alms, a. catry, j. l. glancey, and s. g. advani, “flow modification process for vacuum infusion using port -based resin flow control,” sampe journal, vol. 45, no. 2, pp. 54-63, march 2009. [4] d. heider, p. simacek, a. dominauskas, et al. “infusion design methodology for thick-section, low-permeability preforms using inter-laminar flow media,” composites part a: applied science and manufacturing, vol. 38, no. 2, pp. 525-534, february 2007. [5] c. y. chang, “experimental analysis of mold filling in vacuum assisted compression resin transfer molding ,” journal of reinforced plastics and composites , vol. 31, no. 23, pp. 1630-1637, october 2012. [6] c. y. chang and s. y. li, “effect of process parameters on the surface quality of air cushion method ,” journal of reinforced plastics and composites , vol. 34, no. 3, pp. 257-266, january 2015. [7] c. y. chang and w. r. chen, “experimental s tudy of progressive compression method of resin delivery in liquid composite molding,” proceedings of engineering and technology innovation, vol. 2, pp. 23-25, 2016. microsoft word 4-peti#4599 30-38.docx proceedings of engineering and technology innovation, vol. 16, 2020, pp. 30 38 development of maize stalk cellulose fiber reinforced calcined kaolinite clay geopolymer composite addisu workiye 1,* , eyassu woldsenbet 2 1 school of mechanical and industrial engineering, addis ababa institute of technology, addis ababa, ethiopia 2 addis ababa institute of technology, addis ababa university, addis ababa, ethiopia received 21 july 2019; received in revised form 02 november 2019; accepted 14 january 2020 doi: https://doi.org/10.46604/peti.2020.4599 abstract in recent years the popularity of ecological and renewable materials has grown. aware of the availability of local resources and economic grounds, the attention is paid to the development geopolymer composite. geopolymers are inorganic polymers that are formed by the polymerization reaction of silico-aluminate and silico-oxide. the aim of the research was to develop single maize stalk cellulose fiber reinforced calcined kaolinte caly based geopolymer composite. kaolinite clay was characterized for its chemical composition from two different sites located in ethiopia to use as precursor material. in addition to that, single maize stalk fiber was extracted from the maize stalk local variety by using a retting process, sodium hydroxide purity 98 % used for chemical treatment of the fiber for 30 minutes and its tensile strength 1184 mpa and young modulus 16 gpa were determined based on ast m d3822 to use as reinforcement. sodium hydroxide and sodium silicate in appropriate ratio were used as an alkaline activator. therefore, the geopolymer composite developed from chemical treated short single maize stalk cellulose fiber and calcined kaolinite clay based geopolymer matrix activated by sodium water glass. geopolymer composites reinforced with 0 weight%, 0.1 weight%, 0.2 weight%, 0.6 weight%, and 1 weight% maize stalk single cellulose were prepared and tested for compression strength according to ast m c 1424. measured compression strength ranged from 16 mpa up to 27 mpa. the result shows that the appropriate addition of single maize stalk cellulose fibers can improve the compression strength of a calcined kaolinite based geopolymer keywords: geopolymer, maize, fiber, kaolin, composite 1. introduction the agricultural residue becomes a major source of fibers in the development of the building and packaging industries. nature has given immense sources of natural fibers to humankind, especially for plant fibers, cellulose, and vegetable fibers are the main sources in the building and packaging. natural fibers possess a lot of advantages over synthetic fibers in terms of cost, density, bio-degrading, good acoustic, mechanical properties, abundance, renewable, ease of separation, and carbon dioxide sequestration. unlike synthetic fibers, natural fibers are flexible and less likely to be fractured during composite processing. this enables the fibers to maintain the appropriate aspect ratios to provide good composite reinforcements [1]. maize is the global leading cereal in terms of production with 1,016 million metric tons produced on 184 million hectares globally based on the food and agriculture corporate statistical database 2013. the increasing demand for maize and its global advance implies that by 2023 the maize will account for the greatest share (34 %) of the total crop area harvested * corresponding author. e-mail address: natanium27@gmail.com tel.: +251-11-232414 proceedings of engineering and technology innovation, vol. 16, 2020, pp. 30 38 31 according to the organization for economic co-operation and development and the food and agriculture organization of the united nations. corn stalks consist of a pithy core with an outer layer of long fibers. currently, the corn stalks are chopped and used for forage, left on the field, or baled for animal bedding [2]. maize husk and stalk were assessed for their fiber characteristics and suitability for pulp and paper production. corn stover typically consists of about 50 % stalk, 23 % leaves, 15 % cobs, and 14 % husk. the stalks contain an inner pith and outer rind which is the source for fibers [3]. tensile properties measurement of the stalk rind was carried out experimentally, the rind specimens were subjected to uniaxial tensile loading at a slow rate of 3 mm/min. from the recorded load elongation curves tensile stresses and modulus of elasticity were determined [4]. the alkaline treatment removed the lignin, cellulose, hemicellulose, and residues of the fiber’s surface. maize fibers are treated with 15 % diluted sodium hydroxide solutions at room temperature for 72 hours, then the fibers were washed several times with the distilled water to remove the excess sodium hydroxide from its surface. next, the fibers neutralized with 2.5 % diluted hydrochloric [5]. natural cellulose fibers are extracted from lignocellulose byproducts using bacteria, fungi, mechanical, and chemical methods [6]. retting, the traditional process to extract fibers, uses bacteria and fungi in the environment to remove lignin, pectin, and other substances [7]. however, sodium hydroxide is the most commonly used chemical for fiber extraction [7]. the chemical concentration, temperature, and duration of treatment are the main factors for determining the quality of chemically extracted fibers [8]. the fundamental physical and mechanical properties of the maize fibers could be summarized as follows: a density of 1.5 g/cm 3 , an average diameter of 145 µm with a range of 60-220 µm, average tensile strength, the elastic modulus of 141 mpa, and 6.3 gpa, respectively [9]. also, the maize stalk fibers were tested for tensile strength; the fibers of uniform size were selected and tested in the hounsfield tensiometer machine. the ultimate strength of the fibers was 152 mpa with an average young’s modulus of 8582 mpa [10]. in many practical situations, the cross-section is assumed to be circular for natural fibers and the projected width of the fiber is used to calculate a cross-sectional area. kaolinite is a white hydrated aluminum silicate crystalline mineral. it is a soft powder that has a fine particle and plate-like in appearance. large volumes of kaolin clays are used for the production of cement, ceramics, bricks, and porcelain [11]; the greatest demand for kaolinite is in the paper industry to produce a high-quality paper. it is also used as a filler for paint, rubber, and plastics [12]. the rough geology of clay is a sedimentary material with grains smaller than 0.002 millimeters in diameter [13]. the clay is in its most reactive state when the calcining temperature leads to lose hydroxides and results in a collapsed and disarranged structure [14]. the clay used for the study was calcined at a temperature ranging from 550 °c to 1100 °c and was characterized by x-ray diffraction, infrared spectroscopy, and thermal analysis. after the clay was calcined, the clay was activated by sodium hydroxide and potassium hydroxide solutions at concentrations of 5, 8, 10, 13, 15, or 18 m. the compression strength and the water absorption of the hardened geopolymer cement paste samples were determined after 28 days [15]. geopolymers are produced from the reaction of solid alumino-silicate precursors with the alkaline solution at room or elevated temperature. geopolymeric cement is a new type of calcium oxide free inorganic alumino-silicate material, which has received a great deal of attention across the world in recent years [16]. the abaca fiber reinforced the foamed fly ash geopolymer. the measured compression and flexural strength ranged from 19.56 mpa to 36.84 mpa and 2.41 mpa to 6.25 mpa, respectively. the results suggest an enhancement of the compression strength by the abaca reinforcement and elevated temperature curing [17]. the use of long sisal fibers improved the mechanical properties of the composites, increasing the modulus of rupture compared to the non-reinforced matrix. unreacted products were found in the matrices with a higher amount of activator and/or shorter curing times, demonstrating that, it may influence the final mechanical strength value [18]. a novel bio-composite panel made with recycled wood chips was presented and proceedings of engineering and technology innovation, vol. 16, 2020, pp. 30 38 32 assessed. the bio-composite binder was a natural hydrated lime with the addition of metakaolin [19]. another researcher attempted to improve brittle behavior by reinforcing the geopolymer matrix with bagasse cellulose fibers [20]. the study investigates the development of the calcined kaolin geopolymer based on composites reinforced by maize stalk fiber up to 1 weight%. maize stalk fibers are extracted by retting process and the sodium hydroxide chemical treatment done in order to enhance its hydrophobic property. moreover, the characterization of maize stalk single fibers was conducted to the calcined kaolin based on the geopolymer composite reinforced by maize stalk fiber which was developed and its compression strength and bulk density characterized. 2. material and method 2.1. kaolin deposit in this study, the raw materials are planned to be explored from two different sites. one site is ketchene that were found in the gulele sub-city of addis ababa where pottery making is a dominant practice in this area. the other site is burayu town, where most of the brick factories around addis ababa were found. 2.2. characterization table 1 shows that the laboratory test results on the sampled soils. the chemical or oxide compositions of the four sampled soils are tested in the ethiopian geological survey laboratory. the result indicates the availability of enough sio2 and al2o3 are used as a precursor for geopolymer. from the chemical composition analysis result, chefe red clay soil were selected the material for the preparation of geopolymer paste since it has more sio2 and al2o3. the calcination process in kaolin modification enhanced the properties of the final material obtained. the calcination of kaolin removed the crystalline bound water of hydration which is usually 14 % by weight. the calcined kaolin can serve as a binding agent, filler, fixing agent, heat transfer agent, catalyst support, and processing aid in a broad range of applications. table 1 chemical composition of clay soil samples (%) soils type sio2 al2o3 fe2o3 cao mgo na2o k2o tio2 mno p2o5 burayu-white clay soil one 66.76 17.86 9.53 0.40 0.84 0.22 0.84 0.53 0.08 0.04 burayu-white clay two 67.55 13.66 9.62 0.32 0.73 2.59 2.56 0.47 0.26 0.03 chefe red clay soil 48.72 32.10 12.03 0.30 1.13 0.21 1.15 0.94 0.05 0.08 gel 68.34 15.95 4.93 0.91 1.59 1.76 2.98 0.61 0.11 0.04 fig. 1 calcined kaolinite clay the calcination was conducted by kilns and furnace ltd england thermo computer tc 60 type fl350 and its maximum temperature was 1300 °c found in kombolcha polytechnic college. as fig. 1 shows that the kaolin material was crushed and ground using the machine and sieved using sieve 75 μm, then inserted into the furnace using a clay plate. next, the machine was adjusted to calcination temperature from room temperature to a 500 °c rate of 7.9 °c/min from 500 °c to 950 °c proceedings of engineering and technology innovation, vol. 16, 2020, pp. 30 38 33 rate of 7.9 °c/min for the duration of 2 hours and cooled inside the furnace to room temperature at a rate of 5.11 °c/min. due to the calcination temperature rate and cooling rate slow, the kaolinite clay has enough time to remove the crystalline bound water of hydration from the entire material in the furnace. the clay is in its most reactive state when the calcination temperature leads to losing of hydroxides and results in a collapsed and disarranged structure as indicated on: ( )2 2 5 2 2 7 24 2al si o oh al si o h o→ + (1) 2.3. maize stalk celluolse fiber extraction the maize stalk fiber was selected due to the abundance, accessibility, mechanical property, and low cost. the maize stalk was collected from a research plantation area at ethiopia national agricultural research institute kulumsa center. its main chemical constitutes are 30 % 40 % cellulose, 7 % 21 % lignin, and 28 % hemicellulose. the water retting natural fiber extraction process is used in the research to gain fibers of the greater uniformity and higher quality, the duration of the retting process was 14 days. the maize stalks were harvested on local variety april 12 th , 2017, and air-dried for 15 days, then was cut in the internodes as indicated in fig. 2. each node was split into two pieces and immersed in the solution of 2.5 g/10 l volume together with its husk as shown in fig. 3. after 14 days, the lignin was dissolved and cellulose fiber was easily separated from the hemicellulose by hand, then was washed by water several times. the leafy outer shell covering of an ear of maize is referred to as corn husk. fig. 2 raw maize stalk cut in to the internodes fig. 3 maize stalk and husk in retting process the alkalization is a common preprocessing technique used on natural fiber to remove hemicelluloses, fats, and waxes that may reduce the interfacial strength when processed into the composite form. therefore, the extracted cellulose fiber was treated by 98 % purity sodium hydroxide for 30 minutes as shown in fig. 4. fig. 4 extracted sodium hydroxide treated fiber 2.4. fiber tensile strength test single maize stalk cellulose fiber tensile strength test was conducted in the research laboratory of ethiopia textile, garment, and fashion institute. fiber tensile strength test of 10 sodium hydroxide treated single maize stalk cellulose fiber conducted using fiber strength test machine. 10 n load cells, standard gauge length 25 mm, and speed 100 mm/min was used for the tensile test of a single fiber with ast md 3822. measured average breaking force 2.49 n and average breaking elongation of 2.21 % were determined. proceedings of engineering and technology innovation, vol. 16, 2020, pp. 30 38 34 2.5. fiber cross section a fiber cross-section of single maize stalk cellulose fiber was measured by a biological microscope at the ethiopia institute of agricultural research kulumsa agricultural research center laboratory. sodium hydroxide treated fiber shows a clear surface with no impurities the natural substances on the fiber surface contribute to ineffective fiber-matrix bonding and poor surface wet-out. fig. 5 fiber cross section imagecaptured by biological microscope image of sodium hydroxide treated maize stalk fiber of 41 samples were captured by biological microscope and its cross-section measured randomly at the top, middle, and the bottom section from the image by ts view software as shown in fig. 5. the measured average diameter were 44 μm, this is in good agreement with recent results from a study of jute fiber diameter by (virk) who suggested the use of a fiber area correction factor of 1.42 in the measurement of fiber modulus and strength to account the overestimation of the fiber cross-section area by the diameter method [21]. fig. 6 shows maize stalk single cellulose fiber cross-sectional diameter ranges between 25 μm to 60 μm and an average of 44 μm. however, 1.42 cross sectional area correction factors considered during calculation based on the previous researcher. fig. 6 maize stalk fiber cross section 2.6. tensile strength result of maize stalk single fiber fig. 7 tensile strength and young modulus of single maize stalk cellulose fiber treated by sodium hydroxide purity 98 % for 30 minutes 0 10 20 30 40 0 10 20 30 40 50 r a d iu s µ m sample number stalk fibre 0 10000 20000 30000 0 500 1000 1500 2000 1 2 3 4 5 6 7 8 9 10 y o u n g m o d u lu s (m p a ) t e n si le s tr e n g th (m p a ) number of tested fiber maize stalk fiber sample tensile strength young modulus proceedings of engineering and technology innovation, vol. 16, 2020, pp. 30 38 35 breaking force and elongation result of ten different samples and cross-section diameter of fiber 44 μm, gauge length 25 mm were taken to determine the tensile strength and young modulus as shown in fig. 7 which the result achieved was 1184 mpa and 16 gpa respectively. the research finding of n. reddy et al. specifically on breaking elongation and young modulus were comparable to the result achieved [22]. since the maize stalk abundantly available, biodegradable, and cheap, its sodium hydroxide chemically treated fiber has good mechanical properties possible to use as natural fiber reinforcement. 3. geopolymer composite processing 3.1. binding agent polymerization is formed by mixing of sodium silicate and sodium hydroxide solution to form 12 m sodium hydroxide solution, the sodium hydroxide flakes of 489 grams dissolve in deionized water to require the concentration. during the formation of concentration, it liberated a large amount of heat to form polymerization. the mixed sodium silicate and 12 m sodium hydroxide solutions are allowed to keep at room temperature for a minimum of 24 hours and this mixed solution acts as a binding agent as shown in fig. 8. maize stalk fibers were chopped into 5 mm size and soaked into the sodium hydroxide solution purity 98 % for 30 minutes for the chemical treatment. after 30 minutes, the treated single maize stalk cellulose fiber was washed repeatedly and allowed to dry as shown in fig. 9. fig. 8 bindig agent fig. 9 chopped treated fiber 3.2. mixing, casting and curing the maize stalk cellulose fiber-reinforced geopolymer composites were intended to replace conventional cements and concretes, using low carbon emission and producing sustainable green materials. geopolymer was prepared at the construction material laboratory, sieved calcined kaolinite clay (75 μm), and the single cellulose maize stalk, chopped 5 mm fiber length (1 weight%). first, dry mixed in a high shear mixer for homogeneous and uniform distribution and sodium water glass was slowly added to the precursor and mixed until complete integration occurred. the geopolymer composite slurry was poured into a high strength plastic mold attached to a vibration table for a more uniform distribution of maize stalk cellulose fiber and fewer voids formation. the filled mold was wrapped in a plastic wrap to prevent water loss during setting and curing. next, cured for 3 hours in a 50 °c oven. then, after 3 days, the geopolymer composite specimen was demolded and set to dry at room temperature for 28 days. 3.3. geopolymer composite compression strength universal compression /flexural tester was used to measure the compression strength of the samples. the compression strength test was performed following ast m c1424-10. the compression strength is a measure of the structure’s resistance proceedings of engineering and technology innovation, vol. 16, 2020, pp. 30 38 36 or ability to withstand the compression stress. fig. 10 shows a cylindrical specimen with 25 mm diameter and 50 mm height and were tested in accordance with ast m c1424-10. a test was carried out on a universal testing machine with 300 kn load cells at the addis ababa institute of technology construction material testing laboratory. three replicated treatments were prepared for the test, samples were subjected to gradual load with the rate of 0.28340 mpa/s then the maximum load were recorded at failure. three replicates were tested for each treatment, the mean ultimate load was taken as the compression strength. the resulting volumetric weight of each treatment was taken as the average of the three cylindrical samples from each mix design and the dimensions of each sample were measured using a caliper and weighed subsequently. fig. 10 cylinderical specimen 25mm diameter and height 50mm table 2 shows that the curing method has an effect on the compression strength with the same weight percent of fiber 0.1 weight% oven-cured sample has more compression strength than the ambient cured sample. the result of 1 weight% maize stalk fiber geopolymer 27 mpa comparable to the research finding k. korniejenko et al. geopolymer composite with 1 weight% cotton fiber reinforcement which was 28 mpa [23]. table 2 compressionstrength result fber loading(gram) force (kn) area(mm 2 ) compression strength (m pa) curing method density g/cm 3 0 4.79 314 15.27 ambient 2.453 0.1wt% 7.5 314 23.77 oven cured 0.1wt% 5.58 314 17.78 ambient 2.46 0.2wt% 9.37 490 19.09 ambient 2.64 0.6wt% 10.66 490 21.73 ambient 2.44 1wt% 13.41 490 27.32 ambient 2.32 4. result and discussion an average breaking force 2.49 n and average breaking elongation 2.21 % result of 10 different maize stalk cellulose fiber samples, and average cross-section diameter of the fiber 44 μm, gauge length 25 mm were taken to determine the tensile strength. young modulus average result was 1184 mpa and 16 gpa respectively. breaking elongation 2.21 % and young modulus 16 gpa results are comparable to the research finding of n. reddy et al [22]. the maize stalk single fiber tensile strength test results are in the range of the tensile strength of flax fiber and its young modulus result in the range of young modulus of abaca and sisal fiber [24]. table 2 shows the geopolymer composite compression strength average values of 25-50 mm tested specimens for 28 days curing time. compared to pure geopolymer, there was an increment of 10 % in compression strength of the composite made from 1 % sodium hydroxide treated single maize stalk cellulose fibers. additionally, it revealed the increment of 6 % in compression strength of the composite made from 0.1 % sodium hydroxide treated single maize stalk cellulose fiber, oven-cured at 60 °c for 4 hours and left to dry for 28 days at the ambient condition when compared to cured by the ambient proceedings of engineering and technology innovation, vol. 16, 2020, pp. 30 38 37 condition. the average of the compression strengths of three specimens was taken as the measured value for each run geopolymer. the measured samples indicate compression strength 16 mpa to 27 mpa from pure geopolymer up to 1 weight% fiber loading as shown in fig. 11. compression strength requirements are at least 17 mpa and 28 mpa for residential and commercial concrete respectively [25]. the density of pure geopolymer samples 2.45 g/cm 3 and 1 weight% fiber loading 2.32 g/cm 3 were determined. fig. 11 compression strength of geopolymer reinforced by maize stalk fiber 5. conclusions the clay mineral is abundantly available in ethiopia at a different location and the sample clay material was taken from burayu white clay, chefe red clay soil, and gel. their chemical composition analysis test was conducted on the ethiopian geological laboratory center. according to the results, chefe red clay soil has good sio2 and al2o3 content than other samples. the kaolinite clay was calcined in the laboratory from room temperature to a 500 °c rate of 7.9 °c/min, from 500 °c to 950 °c rate of 7.9 °c/min for two hours to obtain the precursor material. the chemical treatment is done on the extracted the maize stalk and husk fiber using the sodium hydroxide purity 98 % for 30 minutes, then the tensile strength testing carried out based on ast m d 3822 standard. tensile strength 1184 mpa and young modulus 16 gpa achieved for single maize stalk cellulose fiber. the average of the compression strengths of three specimens was taken as the measured value for each run geopolymer sample. the masured compression strength of the pure geopolymer was 16 mpa and for 1 weight % fiber loading 27 mpa, which approach with 28 mpa compression strength of geopolymer from fly ash 1 weight % cotton fiber loading that cured for 28 days. since the maize stalk cellulose fiber density is significantly lower than that of calcined kaolinite clay, it is reasonable for the geopolymer composite density to decrease with maize stalk cellulose fiber loading. maize stalk cellulose fiber reinforced calcined kaolinite clay geopolymer composite can be used for the finishing material for interior wall application. conflicts of interest the authors declare no conflict of interest. references [1] a. k. mohanty, m. misra, and l. t. drzal, natural fibers,biopolymers,and biocomposites, 1st ed. new york: crc press, 2005. [2] j. a. youngquist, b. e. english, h. spelter, and p. chow, “agricultural fibers in composition penels,” proceedings of the 27th international particleboard /composite materials symposium, 1993, pp. 133-152. 15.6 17.78 23.77 19.09 21.72 27.33 0 5 10 15 20 25 30 0% 0.10% 0.10% 0.20% 0.60% 1% c o m p re ss io n s tr e n g th ( m p a ) fiber weight% geopolymer composite proceedings of engineering and technology innovation, vol. 16, 2020, pp. 30 38 38 [3] d. o. ekhuemelo, “assessment of fiber characteristics and suitability of maize husk and stalk for pulp and paper production,” journal of research in forestry, wildlife and environmental, vol. 5, no. 1, pp. 41-49, march 2013. [4] l. zhang, z. yang, and h. guo, “tensile properties of maize stalk rind,” bioresource, vol. 11, no. 3, pp. 6151-6161, may 2016. [5] p. baranitharan and g. mahesh, “alkali treated maize fiber reinforced with epoxy polymer matrix composites,” international journal of innovative science and modern engineering, vol. 2, no. 5, pp. 1-7, april 2014. [6] n. reddy and y. yang, “biofibers from agricultural byproducts for industrial applications,” trends in biotechnology, vol. 23, no. 1, pp. 22-27, january 2005. [7] g. henrikson, d. e. akin, r. t. hanlin, c. rodriguez, d. d. archibald, l. l. rigsby, and k. l. eriksson, “identification and retting efficiencies of fungi isolated from dew-retted flax in the united states and europe,” applied and environmental micropbiology, vol. 63, no. 10, pp. 3950-3956, october 1997. [8] i. doraiswamy and p. chellamani, pineapple leaf fibers, 1st ed. north carolina: north carolina state university, 1993. [9] m. dauda, m. yoshiba, k. miura, and s. takahashi, “processing and mechanical properties of long maize fibers reinforced polypropylene composites,” transactions of the materials research society of japan, vol. 26, no. 4, pp. 1083-1090, january 2001. [10] s. bavan and m. k. g. channabasappa, “examination of tensile strength and ftir features of maize fibers reinforced polymer composite,” international journal on recent trends in engineering and technology, vol. 5, no. 4, pp. 27-30, march 2011. [11] h. h. murray, “overview clay mineral application,” applied clay science, vol. 5, no. 5-6, pp. 379-395, march 1991. [12] h. h. murray, “traditional and new applications for kaolin,smectite and palygorskite:a general over view,” applied clay science, vol. 17, no. 5-6, pp. 207-221, march 2000. [13] g. bedassa, “characterization of the genesis of belessa kaolin occurrences, hosaina area.” master thesis, deptment resource geology, addis ababa university, addis ababa, june 2017. [14] b. b. sabir, s. wild, and j. bai, “metakaolin and calcined clays as pozzolans for concrete,” cement and concrete composites, vol. 23, no. 6, pp. 441-454, december 2001. [15] a. nmir, n. hamdi, o. yazoghli-marzoul, m. duc, and e. srasra, “synthesis and characterization of kaolinite based geopolymer: alkaline activation effect on calcined kaolinitic clay at different temperature,” journal of materials and environmental science, vol. 8, no. 2, pp. 676-690, 2017. [16] j. davidovits, “properties of geopolymer cement,” proceedings first international conference on alkaline cements and concretes, 1994, pp. 131-149. [17] s. n. janne paulin and b. p. michael angelo, “development of abaca fiber reinforced foamed fly ash geopolymer,” matec web of conferences, vol. 156, pp. 1-8, 2018. [18] l. c. silva, r. a. dos reis ferreira, et al., “optimization of metakaolin based geopolymer composite using sisal fibers,response surface methodology and canonical analysis,” international journal of advanced engineering research science(ijaers), vol. 6, no. 4, pp. 32-44, april 2019. [19] m. ravi, “biocomposite panels from recycled wood chips for sustainable building applications,” master thesis, deptartment civil engineering, university of british columbia, vancouver, august 2019. [20] d. wattanasiriwech, t. munmueangkham, and s. wattanasiriwech, “impact strength and physical properties of geopolymer composites reinforced with bagasse cellulose fibers,” 2018 world congress on advances in civil, environmental & materials research (acem18), august 2018, pp. 1-7. [21] a. s. virk, “numerical models for natural fiber composites with stochastic properties,” ph.d. dessertation, school of marine science and engineering, university of plymouth, plymouth, 2010. [22] n. reddy and y. yang, innovative biofibers from renewable resources, 1st ed. verlag berlin heidelberg: springer, 2015. [23] k. korniejenko, e. fraczek, e. pytlak, and m. adamski, “mechanical properties of geopolymer composites reinforced with natural fibers,” international conference on ecology and new building material products, 2016, pp. 388-393. [24] m. c. symington, w. m. banks, and o. d. west, “tensile testing of cellulose based natural fibers for structural composite applications,” journal of composite material, vol. 43, no. 9, pp. 1083-1108, 2009. [25] national ready mixed concrete association(nrmca) “concrete in practice” http://welschreadymix.com/wp content/uploads/2016/04/cip35testing_compresive_strength_of_concrete.pdf, june 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/).  proceedings of engineering and technology innovation, vol. 14, 2020, pp. 22 28 personalized rehabilitation recognition model upon anfis yao-chiang kan 1 , yu-chieh kuo 2 , hsueh-chun lin 2,* 1 department of communication engineering, yuan ze university, taoyuan, taiwan 2 institute of health service administrations, china medical university, taoyuan, taiwan received 13 march 2019; received in revised form 19 may 2019; accepted 11 july 2019 doi: https://doi.org/10.46604/peti.2020.3912 abstract this study applied the adaptive neuro-fuzzy inference system (anfis) to design a recognition model of personalized rehabilitation. in the model, the user may take a wearable sensor and follow the assigned joint-relax exercise to measure the motions of the upper limbs. the sensor that is embedded with the chips of accelerometer, gyroscope, and inclinometer produced the sample datasets due to the exercise schedule of physiotherapy assignment. all motion samples were labeled by arbitrary numbers, which can be identified to the specific motion, for the data training process. a fuzzy inference system (fis) was initially designed by the steps of data pre-processing, featuring, fuzzifying, and ruling fuzzy logics according to the sample datasets. the fis was then trained by the anfis for optimization by tuning parameters of the features. in testing, the accomplished fis could recognize the motion features by the defuzzifier that infers the label corresponding to the motion. as a result, the average recognition rate was higher than 90% when the testing motions followed the sampling schedule of the physiotherapy assignment. the model can be applied in the ubiquitous healthcare measurement for health services. the professionals can assess whether the subject obeyed the assigned program or not based on detail motions of the exercise. this approach can be enabled on the trackable interface for the physiatrists to screen the motions of routine rehabilitation. keywords: anfis, physiotherapy, rehabilitation recognition, sensor-enabled wristband, ubiquitous healthcare 1. introduction the physiotherapy provides the rehabilitation program for the patients who are weakened on motions. the therapeutic exercise is thus designed to flex the joints of their injured limbs. the physical therapists scheme the rehabilitation progra ms including the joint kinematics for those disabled parts, e.g. flexion, extension, abduction, adduction, pronation, supination, rotation, and deviation [1]. many clinical studies of healthcare services advised that the routine exercises with a scheduled rehabilitation period are helpful in healing [2]. the physiatrists request daily practice at home, or with facilities in the hospital, for the patients and observe their performance to adjust the health promotion program of joint motions. the modern healthcare services with internet technique offered ubiquitous healthcare measurement (uhm) for patients to support self-management for homely rehabilitation. the physiatrists are able to screen the daily records of the patients via the uhm for evaluating whether the assigned exercise needs progressive revision or not [3]. the uhm enables rehabilitation in a wireless area network by wearing non-imaging monitoring devices to record the motion data of patients with privacy [4-5]. therefore, a wristband with accelerometer and gyroscope was widely used to detect human activities such as housekeeping, falling, walking, and sporting in recent years [6]. with bluetooth technology, the sensor wristband can efficiently transport the measurement data to a smartphone for the uhm. the personalized rehabilitation data can be measured by a bluetooth sensor and be recognized by the proper algorithms for management [7-8]. * corresponding author. e-mail address: snowlin@mail.cmu.edu.tw proceedings of engineering and technology innovation, vol. 14, 2020, pp. 22 28 23 for the study of activity recognition, the algorithms involving artificial neural network (ann), backward propagation neural network (bpnn), fuzzy logic theory, etc. are popular for computation. the rule‐ based fuzzy inference system (fis) is quite appropriate for the tendency of stationary motions (e.g. rehabilitation) [9]. a fuzzy set involves membership funcions (mfs) to fuzzify the input and output features; then, the logic rules based on if‐ then syntax are scripted to defuzzify the mfs for inference (i.e. defuzzification) [10-11]. in the fis, the mamdani and sugeno models perform two mf types, which the mfs of output features are respectively formatted by geometrical-shape functions and linear-expression crisps [12-13]. the adaptive neuro‐ fuzzy inference system (anfis) enables a data training process to adjust the fis parameters for improving the inference capability [14]. in general, the fis was suggested to recognize the inferable features of the rehabilitation exercises in the uhm. this study suggested a prototype of the rehabilitation model, which applied anfis to recognize the personalized motions. a wearable sensor compliant with bluetooth protocol was employed to collect data. the anfis enables an adaptive data training process based on the fuzzy logic rules to achieve personalized recognition. 2. methodology the proposed model requests the anfis to recognize the rehabilitation exercises. the user wears a bluetooth-compliant sensor jy-61 that embeds the chips of accelerometer, gyroscope, and inclinometer into the device to detect 3-axis components of accelerations, angular velocities, and tilt angles, respectively. (a) the wearable jy-61 sensor, embedded chips, and wearing location on the wrist (b) definition of the local and global coordinates for wearing the sensor, and the joint-relaxing motions of abduction-adduction and external-internal rotation fig. 1 prepared instrument and exercises for the personalized rehabilitation recognition 2.1. exercise design we took three rehabilitation exercises as an example, in which the user wore the sensor on the wrist in the measurement. the sensor was able to send 100 datasets of signals per second. as shown in fig. 1, the neutral position of the human body is the global coordinate, and the center of the sensor is the local coordinate. the motions of the exercise were designed by global coordinates, but the sensor signals were measured in local coordinates. when the exercise was activating, the sensor data were corresponding to the initial position of the sensor. two exercises including abduction, adduction, and rotation of the shoulder and elbow were applied to measure the range of motion regarding the upper-limb joints. for the exercises addressed below, the sensor was worn on the wrist and each exercise cycle was repeated five times. proceedings of engineering and technology innovation, vol. 14, 2020, pp. 22 28 24 (1) shoulder ex. with horizontal abduction and adduction (abd-add) motions the arm initially straightens downward and the palm faces inward: keep the initial state for two seconds  raise the arm up laterally to shoulder level (abduction) in four seconds  hold the limb at shoulder level for two seconds  drop the arm down to the initial state in four seconds. (2) elbow ex. with external and internal rotation (ext-int rot) motions – the elbow bends to 90° and the palm faces inward the abdomen, and the upper arm must close to the body: keep the initial state for two seconds  rotate the arm away from facing inward toward the abdomen to facing outward in four seconds  hold the limb there for two seconds  rotate the arm toward the abdomen in four seconds. 2.2. anfis process a typical anfis involves five parts in a cyclical process: (a) data preprocessing, (b) feature visualizing, (c) fuzzifying, (d) data training, and (e) defuzzifying; in which, the process trains the fis to obtain the optimized parameters based on the fuzzy logic rules for recognizing the personalized rehabilitation motions. the flowchart of the anfis process is shown in fig. 2. fig. 2 machine learning flowchart of anfis for recognizing the rehabilitation exercise (a) the user needed to exactly obey the motion schedule of the designed exercise. the sample datasets were sent by the wearable sensor and we were able to define the motion label with the arbitrary numbers in a specific range. for the example of the shoulder exercise, the motions were labeled by random numbers in [4, 5), [5, 6), [6, 7), and [7, 8), respectively. then, the user practiced the same schedule to prepare a test dataset for evaluation. in this study, the possible features includ e relative acceleration, angular velocity, and tilt angle of the motion with respect to the initial sensor position (i.e. racc, rangv, and rang for starting moving, in moving, and limb position, respectively). (b) we observed the accessible features according to the diagrams of senor data variation as shown in fig. 3; whereas, the time-domain data distribution was transformed into the frequency-domain spectrum. the spectra of the available features are related to the membership function (mf) of the corresponding motions in fuzzy computing. for example, we selected rangx and rangvx (i.e. the components in x-axis) as the features to perform the still and moving motions for the abd-add exercise. (c) we created the fis by choosing the proper mfs for the fuzzy set that presents a featuring process before data training. as shown in fig. 3 for example, we adopted the triangle and trapezoid mfs for the input features, and four triangle mfs for the output features according to the spectra of rangvx and rangx variables in the abd-add motions. the mfs of output feature is corresponding to the labels defined in the previous four-step motions. combination of the various input features raw data: acceleration, angular velocity, tilt angle  derived features: relative acceleration, angular velocity, angle, etc. sensoring membership function design scheduled motions initial fis fuzzifying anfis tuning tuning optimizing initializing exercising collecting labeling designingdefining data training inferring controlling evaluating data preprocessing feature visualizing fuzzy logic rules defuzzifying featuring feature set producing producing motion label guide reference spectrum fuzzy set sampling data testing data recognition output trained fis proceedings of engineering and technology innovation, vol. 14, 2020, pp. 22 28 25 with the proper mfs can yield a fuzzy set available for the same output motions. the mfs of output feature, which is initially designed as the mamdani-type function, must be converted to the sugeno-type crisp set before the training process in the anfis. besides, the amount of the output crisps for the sugeno-type fuzzy set must be a multiplication of those in the input features. (a) the feaure rangvx in time-domain distribution (b) the feature rangvs in frequency-domain spectrum (c) the feaure rangx in time-domain distribution (d) the feature rangs in frequency-domain spectrum (e) the mfs of rangvx feature (f) the mfs of rangx feature (g) the mfs of abd-add output feature fig. 3 example of data transform between time and frequency domains for the features of the shoulder exercise (d) we defined the if-then-based fuzzy rules to control the logic of input and output features in each cyclic epoch. with the rules, the input mfs indicate the consistent output mfs, while the number of the output mfs and the rules must be identical. for an example of the fis structure of the abd-add motions, we created five dummy rules that represent five virtual motions proceedings of engineering and technology innovation, vol. 14, 2020, pp. 22 28 26 and designed four rules that satisfy four real motions. the initial fis will be tuned by the anfis, which applies the method of root mean square errors (rmse) in a data training process, to optimize the mf parameters of the fis for each epoch. (e) we defuzzified the trained fis with the capable mfs and the fuzzy rules to infer the motion labels in recognition. the process of defuzzification computes participation degrees of the input mfs based on the rules. the test datasets were employed for evaluation. the inferred motion labels were compared with the assigned schedule to evaluate if they are indexed in the expected motion range. we proposed a quartile motion-label scheme to verify the recognition rate of the fis. the scheme splits the data distribution of motion into four sets (i.e. cut at 25%, 50%, 75%), and it evaluates the accuracy of each quartile set. the sets of 25% and 75% are usually counted as changing the motion, while the data set around 50% (i.e. the middle portion) can be a confidence interval of the reliable motion labels. (a) shoulder ex. with abduction and adduction (b) elbow ex. with external and internal rotation fig. 4 the inference diagrams of the test data versus the sampling schedule 3. results and discussion the trained fis models of the exampled exercises were evaluated by applying the test datasets. the diagrams of both exercises as shown in fig. 4 compare the labeled test data with the inferred motion labels. the inferred motion labels are observed to fit within the defined range if the motion cycles follow the schedules. the inference diagram displays a visualized chart of both inferred and scheduled motions to track the measurement. when the inferred motion labels were not shown in the range of scheduled motion labels, it would report the outlier movements with respect to the assigned motions. the user is able to obtain a personalized fis model that is trained by the sample dataset based on the guideline for recognition in measurement. the healthcare professionals can assess rehabilitation performance for improving the therapeutic procedure. table 1 the recognition rates based on quartile scheme for the test data of scheduled exercises shoulder ex with abduction-adduction motion 25% 50% 75% rest_low 1 0.97 1 abd 1 1 0.8 rest_half 1 1 1 add 0.87 1 1 average 0.97 0.99 0.95 elbow ex with external-internal rotation motion 25% 50% 75% rest_inside 1 1 1 er 1 0.41 0.03 rest_outside 1 0.38 0.03 ir 0.71 1 0.87 average 0.939 0.70 0.48 the recognition accuracy of the chosen motions was evaluated by the quartile schemes as shown in table 1. the outlier data were aware of moving too fast, which may lead to the angular velocity over the maximum of sample data. the average abduction-adduction exercise external-internal rotation exercise proceedings of engineering and technology innovation, vol. 14, 2020, pp. 22 28 27 rates of recognition for the shoulder ex, which performed steady motions by the schedule, were about 0.99 due to the middle (50%) of the quartile scheme. the rates of the elbow ex, which did not move stably from external to internal rotation and probably caused data shifts, were ranged from 0.93, 0.70, and 0.48 with respect to 25%, 50%, and 75% portions. this evaluation suggests that the middle of motion data is acceptable for recognition. we further verified classification performance by combining both exercises and retrieving the middle section of motion data in every second. the subject can randomly choose either shoulder of elbow exercises, but the movement should follow the guide as the best for validation. the sensitivity, specificity, and accuracy of the shoulder exercise are calculated as 0.942, 0.998, and 0.951, respectively; while they reached 0.995, 0.614, and 0.703 for the elbow exercise. 3.1. benefits the proposed model follows the process of the supervised algorithm in machine learning to label the motion data based on the scheduled exercise. with the anfis, it is available for generating a personalized model of rehabilitation recognition. the proposed bluetooth-compliant sensor is able to provide plentiful sample datasets in the training procedure. the personal movement features can be chosen to initialize the mfs of fis with the fuzzy rules for data training in the anfis process. our approach suggests a confidence interval of the motion to extract proper data, e.g. the middle of each scheduled motion period. the model is achievable for recognizing rehabilitation exercises in healthcare, which is enough for practical application [15]. 3.2. limitations the same feature can measure a variety of exercises within the same range of motions, e.g. the features “rangvx” and “rangx” for both of the shoulder and elbow exercises in this study. however, they cannot be combined in an exercise by a unique fis. therefore, the user needs to wear more sensors for measuring this complex exercise. in addition, the sensor usually detects unstable noises for irregular motions, which the features are ambiguous in recognition. if the user does not follow the exercise schedule and the tracking data are probably shifted, the measurement is not reliable for the report. in rehabilitation, the physiotherapist used to design simple exercises for the patient. 4. conclusions we applied the anfis to establish a recognition prototype of personalized rehabilitation model for ubiquitous healthcare measurement in this study. the bluetooth-compliant sensor embedded with the chips of the accelerometer, gyroscope, and inclinometer can be worn to measure the motion data of the rehabilitation exercises. the model presents two exercises with joint-relax motions for the shoulder and elbow. in which, the user wore the sensor on the wrist and obeyed the motion schedule to move the limbs. each motion dataset is labeled by an arbitrary number within a defined range. an initial fis with the fuzzy logic rules was created by the sample datasets; then, the steps of data preprocessing, feature visualizing, fuzzifying, data training, and defuzzifying were processed in the anfis. the anfis repeated the data training cycles to accomplish the trained fis, which can defuzzify the test data upon rule-based fuzzy logics to infer the motion labels and recognize the exercise. in approach, the average recognition rate due to the quartile motion-labels scheme was higher than 90% if the user faithfully followed the assigned schedule. this approach can be promoted to the uhm for comprehensive health services in the future. conflicts of interest the authors declare no conflict of interest. acknowledgement this work was supported under the grant most 105-2119-m-039-003, 106-2119-m-039-002, 107-2119-m-039-002 by the ministry of science and technology, taiwan, r.o.c., and the grant cmu106-s-29 by china medical university. proceedings of engineering and technology innovation, vol. 14, 2020, pp. 22 28 28 references [1] m. jackson, b. michaud, p. tétreault, and m. begona, “improvements in measuring shoulder joint kinematics,” journal of biomechanics, vol. 45, no. 12, pp. 2180-2183, june 2012. [2] y. j. chang, w. y. han, and y. c. tsai, “a kinect-based upper limb rehabilitation system to assist people with cerebral palsy,” research in developmental disabilities, vol. 34, no. 11, pp. 3654-3659, september 2013. [3] darwish and a. e. hassanien, “wearable and implantable wireless sensor network solutions for healthcare monitoring,” sensors, vol. 11, no. 6, pp. 5561-5595, may 2011. [4] m. chen, s. gonzalez, a. vasilakos, h. cao, and v. leung, “body area networks: a survey,” mobile networks and applications (monet). springer netherlands. vol. 16, no. 2, pp. 171-193, april 2011. [5] m. singh and s. k. soni, “spatial correlation-based clustering in wireless sensor network,” international journal of engineering and technology innovation, vol. 8, no. 4, pp. 294-306, january 2018. [6] p. mandal, k. tank, t. mondal, c. h. chen, and m. j. deen, “predictive walking-age health analyzer,” ieee journal biomedical and health informatics, vol. 22, no. 2, pp. 363-374, march 2018. [7] r. r. fletcher, m. z. poh, and h. eydgahi, “wearable sensors: opportunities and challenges for low-cost health care,” in 2010 annual international conference of the ieee engineering in medicine and biology, buenos aires, argentina, september 2010, pp. 1763-1766. [8] k. nguyen-huu, c. g. song, and s. w. lee, “smartwatch/smartphone cooperative indoor lifelogging system,” international journal of engineering and technology innovation, vol. 8, no. 4, pp. 261-273, september 2018. [9] c. c. lee, “fuzzy logic in control systems: fuzzy logic controller: part ii,” ieee trans. on systems, man, and cybernetics, vol. 20, no. 2, pp. 419-435, 1990. [10] l. a. zadeh, “outline of a new approach to the analysis of complex systems and decision processes,” ieee transactions on systems, man, and cybernetics, vol. 3, no. 1, pp. 28-44, january 1973. [11] l. li, y. ma, b. wang, h. dong, and z. zhang, “research on traffic signal timing method based on ant colony algorithm and fuzzy control theory,” in proceedings of engineering and technology innovation, vol. 11, pp. 21-29, 2019. [12] e. h. mamdani and s. assilian, “an experiment in linguistic synthesis with a fuzzy logic controller,” international journal of man-machine studies, vol. 7, no. 1, pp. 1-13, january 1975. [13] t. takagi and m. sugeno, “derivation of fuzzy control rules from human operator’s control actions,” proc. of the ifac symp. on fuzzy information, knowledge representation, and decision analysis, marseilles, france, july 1983, pp. 55-60. [14] j. s. r. jang, “anfis: adaptive-network-based fuzzy inference system,” ieee transaction on systems, man, and cybernetics, vol. 23, no. 3, pp. 665-685, 1993. [15] f. attal, s. mohammed, m. dedabrishvili, f. chamroukhi, l. oukhellou, and y. amirat, “physical human activity recognition using wearable sensors,” sensors (basel), vol. 15, no. 12, pp. 31314-31338, december 2015. doi: 10.3390/s151229858 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/).  proceedings of engineering and technology innovation, vol. 14, 2020, pp. 29 38 the emergence and growth path of shanzhai firms jason shay 1 , peter shieh 2,* , ming-piao tsai 2 , hsiang-jui cheng 3 1 boeing global services, dallas, tx, usa 2 department of automation engineering, national formosa university, yunlin, taiwan 3 taiwan semiconductor manufacturing company, ltd., taichung, taiwan received 11 april 2019; received in revised form 09 may 2019; accepted 26 june 2019 doi: https://doi.org/10.46604/peti.2020.4064 abstract shanzhai refers to a set of dynamic strategies for firms to grow out of counterfeiting. shanzhai firms have traditionally been perceived as counterfeiters from the emerging economies producing cheap and low-quality products. this trend has changed in recent years when shanzhai firms have grown into market leaders and have introduced leading technologies and applications. this paper focuses on the growth of shanzhai firms by exploring how shanzhai firms and shanzhai industries emerged. we further investigate the context that supports the emergence of shanzhai firms into market leaders. we leverage two case studies using the growth path of two market leaders in their respective industries to illustrate the two paths leading to the emergence of shanzhai firms. we attempt to show that although shanzhai firms start out with low legitimacy, these firms grow their legitimacy over time by building their core competitiveness in their respective industries. keywords: shanzhai, emerging economies, growth strategies, counterfeits 1. introduction although imitation is considered as a type of strategy that firms often employee, counterfeit (imitations that infringe intellectual properties) is not generally accepted as a legitimate set of strategies for firms to practice [1-2]. however, it is intriguing that several firms considered as counterfeiters not too long ago have grown out of the emerging economies into big multinational firms and even market leaders. in this research, we attempt to address the concept of shanzhai and the development path of shanzhai firms leading them to become major players in their respective industries. originating from china, the concept of shanzhai bloomed in the last two decades. shanzhai refers to a set of dynamic strategies for firms to grow out of counterfeiting. shanzhai is borrowed from the historical description of rural and unregulated villages in the mountains with their own leaders, armed forces, governance structures, and rules that are outside of the main country government regulations [3]. thus, the term shanzhai describes chinese firms’ practices of either counterfeiting or manufacturing closely imitated products and/or brands, evading laws and regulations. these firms are originally referred to as shanzhai firms. public perceptions of shanzhai products are lookalike products with substandard quality that sell at lower prices [3]. overall, public perceptions of shanzhai have been negative over the past two decades. yet the growth of shanzhai firms has been unmistakable. although research on the growth of the firm in emerging economies has been expanding since the 1990s [4], the term shanzhai had not appeared in the management literature until 2009 [5]. until now, most literature focuses on distinguishing shanzhai from counterfeit, or on investigating marketing strategies or consumer preferences. by definition, counterfeit is the * corresponding author. e-mail address: pimshieh@nfu.edu.tw tel.: +1-469-353-3647 proceedings of engineering and technology innovation, vol. 14, 2020, pp. 29 38 30 practice to capitalize on external competences without proper authorizations. if a firm develops own competences out of counterfeiting, it is a shanzhai firm as opposed to a counterfeit firm. in the management literature, however, how shanzhai firms develop and how such development change over time remains unclear. therefore, we ask the following three interesting but underexplored questions: how do firms evolve into shanzhai firms? what is the development path of shanzhai firms? how do shanzhai strategies help firms grow? we argue that the distinguishing factor between shanzhai firms and counterfeiters lies in whether the firms ultimately build up their own core competencies to explore and exploit further business opportunities and markets or not. three contributions emerge. first, contributing to the shanzhai literature, this research is among the first effort attempting to examine the emergence and impact of shanzhai strategies. second, we contribute to the emerging economies literature by examines the strategies help late-mover firms from emerging economies tap into mainstream technological developments and even become marketleading firms. third, contributing to the innovation literature, this research shows how shanzhai brings about another form of the innovative system. 2. original equipment manufacturers since shanzhai was brought into the management literature in 2009, current literature considers shanzhai as a more dynamic strategy. in earlier stages of an industry lifecycle, many of the current shanzhai firms started out as original equipment manufacturers (oems) that build products designed and ordered by branding firms [3]. during this period, some oem firms would sell—to separate markets and customers—substandard products and/or excess products intentionally produced [1]. as compared to non-oem late-comer firms, these oem firms would be at an advantage due to two reasons. first, these firms gained technological advantages by learning from branding firms. some of these oem firms even enjoyed financial and/or technological resources from branding firms to perform research and development, and to fulfill production tasks by branding firms. second, oem firms can defer such fixed costs as machinery on to products sold to branding firms, giving them pricing advantages when selling shanzhai products on the markets. more recently, after decades of outsourcing practices, multinational branding firms began the practice of multiple-sourcing [6]. while single-sourcing aims at sourcing some parts of production processes that are outside of firms’ competences, multiple-sourcing allows branding firms to further reduce costs by allocating and reallocating procurement orders among multiple oem firms. from branding firms’ perspectives, multiple-sourcing defers a portion of transaction costs between branding firms and oem firms, onto oem firms by ensuring competition among oem firms. however, from oem firms’ perspectives, multiple-sourcing dramatically increases oem firms’ costs and risks. for instance, if oem firms lose orders, it is not likely they could quickly find alternative orders to fill their idle capacities that oem firms prepared for the branding firms. due to the competition among oem firms, many oem firms are already equipped with product design capabilities and sufficient technological capabilities and production capacities to produce the final products that are transforming into original design manufacturers (odm) firm (oem firms that also design for the branding firms). when oem and odm firms face over-competition or seek to find alternative methods to lower their dependence on branding firms, some oem and odm firms choose to cooperate with other branding firms. in addition, certain oem and odm firms begin introducing their own products into the market either through lookalike products or with new brand names owned by the oem and odm firms. the process of using the capabilities originally to serve the branding firms to divert into their own brands is shanzhai. fig. 1 illustrates the evolution of oem and odm firms into shanzhai firms and then into branding firms. in the case of shanzhai, both routes exist. the first route begins with some oem firms producing mass quantities of “white-tagged” products, which are products without a logo printed on them. when branding firms (which could be proceedings of engineering and technology innovation, vol. 14, 2020, pp. 29 38 31 well-established firms or small new firms) come to these oem firms, branding firms can order from these existing products. the oem firms then print the branding firms’ logos onto the products for sale. the second route begins when oem firms introduce their own brands. these oem firms would be likely to design their own products based on the designs that were originally used to produce the products of their clients. no matter which route the oem firms choose, up to this point, these oem firms could be doing the shanzhai business in the grey area of counterfeit. fig. 1 the evolution of oem and odm firms into shanzhai firms and then into branding firms traditional viewpoint on counterfeiting firms’ points to the direction that counterfeiting firms do not really possess their own competences [7]. these counterfeiting firms merely imitate products and/or services on the market. more recently, scholars have challenged this perspective by highlighting that counterfeiting is a significantly big industry to be viewed as homogeneous [7]. scholars try to introduce a more heterogeneous viewpoint in an attempt to account for different counterfeiting practices. scholars were able to ascertain how counterfeiting firms specialize in different forms of counterfeiting strategies. shanzhai firms would use some strategies that are classified as counterfeit (such as producing knockoffs or look-alike products) but shanzhai firms also show capabilities to develop their own competencies [3, 8]. in sum, counterfeiting firms demonstrate the behavior to capitalize on other firms’ competences. while it is observed that shanzhai firms also demonstrate the same behavior, the difference between shanzhai firms and counterfeit firms lies in the development of a firm’s own competences. if a firm develops own competences out of counterfeiting, it is a shanzhai firm as opposed to a counterfeit firm. in the case of oem firms’ transformation, some oem firms begin as almost counterfeiters, producing products similar to their branding-firm clients. with capabilities in both product and manufacturing processes, some of these oem firms become shanzhai firms when they deviate from the prevalent designs and begin introducing new products and/or serving different market niches as the prevalent products do. taiwanese bicycle manufacturing industry started out as oem manufacturers that produce metal-based products, such as screws, bolts, and automotive parts. with these competencies in metal processing, companies first stepped into bicycle oem industry in the late 1970s when the u.s.-based bicycle branding companies started placing bicycle orders with these oems. since then, oems worked its way from pure oem make-to-order to odm that not only manufacture but also design bicycles for branding firms. the most important difference between oems and an odms lies in whether sourcing firms rely on the manufacturers in designing products (i.e. products are designed by sourcing firms or by the manufacturer). during those years as oems and odms, taiwanese firms strengthened many of their competencies such as metal processing and manufacturing processes, and also gained many competencies such as metal structuring and material technologies. in the later years of oem/odm and branding collaboration, branding firms had multiple oem firms and odm firms working competitively at the same time. due to intense competition among branding firms’ oem and odm partners, oem and odm firms need not only to fulfill procurement orders ahead of schedule, but for oem and odm firms had to do much more at their own expense, both in design and manufacturing aspects. for example, oem and odm firms had to prepare excess production capacities (free, of course) to what they needed to fulfill branding firms’ orders. furthermore, oem and odm firms had to perform research and development at their own costs on many aspects, such as improving production speed and quality and developing new materials and designing new structures for branding firms to develop its next products. when all of these competitive pressures kicked in, oem and odm firms that are out-competed would either be left out of the next proceedings of engineering and technology innovation, vol. 14, 2020, pp. 29 38 32 procurement or receive less order from branding firm, and the oem and odm firms will be left with hefty fixed costs to allocate. therefore, these oem and odm firms had to find other ways to cover these costs. with all the production and technological edge at hand, some oem and odm firms realized their risk of competition among their counterparts, and decided that there was a need to begin developing their own “bikes” and “brand” from what their learned from branding firms and from the capabilities it developed originally to serve branding firms [9]. from the case, we suggest that some shanzhai firms and industries are actually fostered by branding firms themselves. many suppliers began by serving mainstream branding companies. however, strategies by those branding firms, such as multiple-sourcing, caused many producers to be left out of the supply chain with full research and development (r&d), technological, and production capabilities. these oem firms have to come up with ways to fill their capacities when branding firms decide to the source of competing for oem firms. shanzhai has emerged as a major way to enable such growth. 3. technological modularization for firms without an oem background, past research finds a second path leading to shanzhai. researchers find that the ability in modularizing products and technologies in the value chain is one important factor distinguishing shanzhai firms from pure counterfeiting firms [8, 10]. technology management literature provides modularization with a formal definition that modularization is a strategy used to efficiently organize complex products and processes [11]. modularized systems are designed in such a manner that independent components are bundled up and work as an integrated whole unit [12]. according to campagnolo et al. [13], the basic concept behind the modular design is to minimize interdependence between modules while maximizing interdependence within the modules. scholars find that modular design is constructive in promoting new designs with a shorter production lead-time [3, 14-15]. in industries that are highly modularized, shanzhai firms evolve along two paths: (1) shanzhai firms assisting the formation of modularized technologies and (2) shanzhai firms exploiting the benefits of modularized technologies. modularization helps shanzhai firms by providing them with flexibilities to change, such as adopting different technologies and adopting new technological developments. scholars believe that shanzhai firms could find and tap into a position in the already established value chains and that shanzhai firms could bring new values to existing industries and value chains. scholars have identified that modularization is an important common ground for thriving shanzhai industries [8, 10]. shanzhai firms are likely to thrive around industries or value chains that are highly modularized. modularization in industries or value chains often implies that the industries or value chains are highly specialized, allowing firms to find specific positions to tap into, rather than needing to have the whole set of know-how in the products or services. for example, during the early 2000s when smartphones were introduced, only firms that owned the whole set of technologies are able to design smartphones, such as apple and samsung. approximately ten years later, smartphone components are modularized into sub-sections, such as camera modules, antenna modules, and touchscreen modules. nowadays, relatively small and young companies entering the final product market can all procure components and/or modules sold by specialized module and/or component producers and integrate those components and modules into the prototypes of final products that are ready to be outsourced to professional manufacturers for mass production. through modularization, technological entry barriers to the industry would be lowered that even small firms with much fewer resources and technological capabilities can now design smartphones and have the chances to actually introduce their products to the market [16]. in addition, the modularization also brings opportunities to shanzhai firms by providing flexibilities for firms to produce final products. technological modularization provides firms with more options in choosing and switching components. for example, the practice of pin-to-pin substitution (meaning that the shanzhai product perfectly follows the prevalent designs for client firms to use) is a well-known strategy promoted by shanzhai firms to urge branding firms to switch the components used proceedings of engineering and technology innovation, vol. 14, 2020, pp. 29 38 33 in their products. subsequent to modularization took place, unified standards and/or protocols would likely emerge, and the component manufacturers could follow the leading standards and/or protocols. therefore, downstream firms would have more flexibility in switching among suppliers. in sum, modularization provides both existing and new firms with the flexibilities to introduce new designs cheaper and faster [15, 17-18]. so why do industries and value chains modularize? the current literature points toward two main reasons, technological complexity, and economies of scale [8, 10, 17]. take the cell-phone camera module as an example. during the first couple of generations of apple’s iphone, apple designs everything ranging from processor chips to cameras, speakers, antennas, and so on. over the years, iphone designs evolve into being lighter and thinner and are more advanced. the components used in iphones also evolve to be considerably more complicated than the designs and technologies behind the components used in iphones developed beyond apple’s capabilities to handle. nowadays, apple depends on specialized firms to produce modularized components for apple to put into its iphone designs. while apple is an expert in designing smartphones, it would be challenging for apple to be specialized in all the technologies required to design entire smartphones. at the same time, the economy of scale could be demonstrated by knowing that when smartphones become smaller, the manufacturing and assembling processes become increasingly complicated. it would take several significantly expensive and specialized machinery (that may not be useful in producing any other products) to produce camera modules or to assemble components into the final products. it is practically impossible for apple to own all the technologies and machinery to make iphone end-products. therefore, ultimately apple would realize that there is a need to design their products based on modularized components that they can outsource to specialized manufacturers for production. this is also the first source of how shanzhai firms emerge out of technological modularization, by assisting technological modularization in the industries. in this section, we demonstrate the second type of shanzhai using the case of some taiwanese and chinese ic designing companies. these companies start out using counterfeit strategies, imitating the market leaders’ designs, but some of them are now emerging as a competitor to leaders in low-end smartphone chipsets markets. most of these latecomers were ic designer and ic service provider under ic oem firm. these firms were firms that are very closely working with different streams of ic design technologies and is very flexible in making adjustments according to their clients’ designs. figure 2 illustrates how technological modularization induces the formation of shanzhai firms. technology complexity increase core competence development branding firm competence oursourced tochnology modularized shanzhai branding fig. 2 technological modularization induces the formation of shanzhai firms after leaders began to dominate the smartphone processor market, latecomers began to follow leaders from generation, structure, to specifications that leaders put on their processor chips. latecomers introduce these processor chips not only to small smartphone companies but also to big companies as a cheap alternative for their lower-end smartphones. furthermore, latecomers began packaging complementary components relating to processor chips into chipsets. this way, latecomers further helped smaller companies by reducing the complexities for firms to design smartphones. since processor chipsets ar e the “hearts” to smartphones, by taking the burden to design processor chips out of smartphone designers, latecomers dramatically lowered entry barriers for firms to design smartphones. more recently, latecomers work closely with small smartphone companies in customizing specialized processor chips. [3, 8, 10, 19]. this way, these latecomers gain their market by further reduce entry barriers to late-coming smartphone designers. from the example of ic designers, it would not be hard to tell the differences between pure counterfeit firms from shanzhai firms. while pure counterfeit firms only copy and sell branding firms’ products, shanzhai firms may start out similar proceedings of engineering and technology innovation, vol. 14, 2020, pp. 29 38 34 to counterfeit firms, but shanzhai firms would also open up other routes to serve underserved market segments. as scholars pointed out, shanzhai firms’ development would broaden up as they invest more resources into their respective industries [20]. for example, with shanzhai firms’ capabilities in integrating new technologies or introducing technologies from other fields, or by integrating the needs to serve different market segments, shanzhai firms are able to open up market opportunities. furthermore, while most counterfeiters may continuously look for different products to counterfeit, shanzhai firms actually move between counterfeiting and creating. even until today, undeniably that late-coming ic designers are still closely following the development of market leaders, however, we can also see latecomers actively involving in creating very genuine products and solutions to the market and serve market niches that leaders may have overlooked. figure 3 illustrates the role of core competency development in counterfeiters’ transition into shanzhai firms. counterfeit counterfeit shanzhai core competence development yes no fig. 3 the role of core competency development in counterfeiters’ transition into shanzhai firms 4. inputs for determination of jmf there are many advantages for firms to practice shanzhai strategies. first, shanzhai strategies provide firms with a lowered entry barrier into specific industries and in a faster way [20-21]. for some industries such as smartphones or processor chips, it is very expensive for firms to enter the industries in a completely genuine way and also very hard for firms to establish their market positions. for example, while leading firms spend on average between $100-200 million developing one generation of processor chips [22], this amount would be a huge burden for latecomers to invest at early stages. by first imitating or counterfeiting established firms, shanzhai firms could accumulate resources early on and then decide whether they would depart from counterfeiting existing technologies and products and begin developing their genuine technologies and products. we see many firms started out making almost identical phones with almost identical technological specifications to leading brands. in recent years, these two brands began making their own technological advancements. for example, a smaller and younger smartphone company from asia introduced its very first processor chip design, which cost it more than 100 million dollars in research and development [23]. without the capital resources it accumulated in the past by selling shanzhai smartphones, it would not be able to achieve this goal. second, the innovation path in shanzhai innovation networks broadens up the deeper the firms are involved in the industry, as opposed to traditional innovative networks where the innovation path would narrow down after a period of time [3]. scholars argued that the reason behind this finding is that shanzhai firms would become more flexible in their capabilities to integrate new concepts, new designs, and new technologies into their own products and designs. for example, smartphone designers have been incorporating graphics processing technologies to strengthen the photographic and retouching aspects of smartphones to become specialized in targeted product markets. in a traditional network, however, firms that are specialized in developing cutting-edge technologies will be likely to get stuck into the technological and product paradigms that got them successful in the first place. proceedings of engineering and technology innovation, vol. 14, 2020, pp. 29 38 35 on the other hand, the current literature also points out some liabilities that shanzhai firms would encounter. first, public perception and brand image on shanzhai firms and shanzhai products tend to be more on the negative sides. for example, many people still believe that shanzhai smartphones are inferior to name-brand smartphones, although third-party testing has shown that some shanzhai smartphones actually perform better. while many benchmark testings show that shanzhai smartphones are either at par with or exceeding performances of high-end branding models. also, many of shanzhai smartphones now have genuine designs and components. however, public perception would perceive the shanzhai firm as producing substandard products or well overpriced if selling at competitive prices to high-end branding models. second, shanzhai firms also have a higher risk of being sued. as profitability for branding firms becomes lower, many branding firms blame shanzhai firms for depriving branding firms’ profits. therefore, branding firms have been raising intensive patenting wars on shanzhai firms. many shanzhai firms have been feeling this heat. for example, many famous smartphones designers started out as shanzhai firms have been sued by branding firms such as apple, qualcomm, and microsoft over patent infringement. some products even get banned from selling in the european union or the united states. lee et al. [18] pointed out that actions such as counterfeit or imitation commonly observed in shanzhai firms could become potentially problematic due to the very fine line between being legal or illegal [24]. the related the question to be asked here is: why do we see more shanzhai activities in emerging economies? although there are still many liabilities to be shanzhai firms, we still see shanzhai strategies flourish in emerging economies. current literature provides us some clues to the reason why. to date, most firms from emerging economies could be considered as latecomers to their industries. as latecomers, it is crucial for these firms both to utilize the resources they have at hand and to find the fastest way to catch up [20]. first, while many firms from developed economies use emerging economies as their production bases, due to lowered costs and regulations, emerging economies are packed with technologies and capacities to produce cutting-edge products. these technologies and capacities, in turn, become resources for the late-comer firms from emerging economies to work with. second, using the metaphor from formula one, when taking over racers in the front, one will first have to closely follow the racer in the front. by closely following the lead racer in the front, there are many advantages such as smoother aerodynamics and knowledge about road conditions. using similar logic, when latecomers from emerging economies try to follow the development paths of market leaders, latecomers enjoy advantages such as reduced research and development costs and more developed markets [17, 20-21]. third, as market needs and consumer preferences differ across countries and that most branding firms develop their products based on needs and preferences from their home countries, many market niches from emerging markets remained to be filled [3, 19]. these market niches leave rooms for shanzhai firms to work with. shanzhai firms can learn from branding firms, modify branding firms’ products according to local needs and preferences, and better serve these underserved markets. 5. discussion and contributions three contributions emerge from this research. first, contributing to the shanzhai literature, this research is among the first effort attempting to qualitatively examine the emergence and the impact of shanzhai strategies. we used two case studies to demonstrate how a firm with a background as an oem firm utilizes shanzhai strategies to penetrate new markets and to become a renowned brand. we also demonstrated how shanzhai firms capitalize on modularized technologies to lower entry barriers and to innovate and introduce new products. second, contributing to the emerging economies literature, this research examines how strategies help late-moving firms from emerging economies tap into mainstream technological developments and even become market-leading firms. third, contributing to the innovation literature, this research shows how shanzhai brings about another form of the innovative system. as shown in the previous sections, shanzhai firms have now evolved from counterfeiting firms into distinctive industries and networks that are able to produce their own competences and innovations. therefore, shanzhai could become another innovation system that is distinct from those innovation systems that we are familiar with. proceedings of engineering and technology innovation, vol. 14, 2020, pp. 29 38 36 6. future research shanzhai has been widely utilized and has been effective for some firms not only to gain market positions but to become market leaders. it is safe to say that shanzhai may be an effective way for latecomers to catch up with leading firms. however, when firms grow stronger and even become leaders, some questions still await to be answered. first, when shanzhai firms with inferior technologies develop into being more sophisticated, their technological capabilities may be as strong as or even stronger than firms that they used to imitate. for example, many of the applications by alibaba (taobao), tencent (wechat), and zte corporation are exceeding their counterparts from developed economies. under this situation, their options to imitate others would be very limited. how will their growth strategies evolve? second, when shanzhai firms become leaders in their industries, what will they do to secure their positions? compared to leading firms, shanzhai firms may not be used to defend their intellectual properties [1-2]. when these shanzhai firms grow stronger, their intellectual properties would increase that there would be firms out there trying to imitate them. how will these shanzhai firms use and protect their resources? for example, how will shanzhai firms use (or support) the concepts of open contents versus traditional copyrights? third, since public perceptions are likely to link shanzhai firms more with inferior technologies and brand imitators when shanzhai firms develop into innovative products, how do shanzhai firms manage public perceptions and their brand images? fourth, if “shanzhai” refers latecomers from less-developed economies imitating market-leading firms from developed economies, “reverse-shanzhai” is the imitation behavior by market-leaders on latecomers. this is now becoming a noticeable phenomenon that market-leading firms (such as facebook from the united states) are now actually “reverse-shanzhai” on their counterparts from less-developed economies, such as wechat by tencent. facebook’s founder and ceo mark zuckerberg publically admit facebook’s desire to learn from wechat and that what's functions and business models could help fuel facebook’s growth [25-27]. with the rise of reverse-shanzhai, questions yet to be answered raise from both sides. for example, how will the intellectual battlefield transform while both parties are protecting, exploiting, and exploring all at the same time? how will the inter-company relationship become between companies on the two ends of the spectrum? finally, although shanzhai as a term is chinese in origin, shanzhai strategies are not limited to firms in emerging economies in general. numerous firms in today’s developed economies such as britain, canada, germany, japan, and the united states undertook what can be called shanzhai strategies when their economies were taking off [28]. how these strategies are phased out as ipr protection tightens around the world and as these firms’ capabilities to go after high-end markets develop will remain a fascinating area for future research. 7. conclusions this paper explores how shanzhai firms emerge. we demonstrated the two contexts that foster shanzhai firms, and why the shanzhai phenomenon bloomed from emerging economies and continue to be widespread in emerging markets. shanzhai firms begin as counterfeiters and some of them have successfully evolved into innovative firms with their own competences that even become global leaders in their respective industries. while the majority of the shanzhai behaviors are observed on firms from the emerging economies nowadays, firms with more advanced technological capabilities from developed economies have been observed to reverse-shanzhai on firms from the emerging economies. therefore, the shanzhai phenomenon is not specific to the emerging economies, but shanzhai can and will be a type of growth strategies that companies use to grow their competencies even further. conflicts of interest the authors declare no conflict of interest. proceedings of engineering and technology innovation, vol. 14, 2020, pp. 29 38 37 references [1] b. berman, “strategies to detect and reduce counterfeiting activity,” business horizons, vol. 51, no. 3, pp. 191-199, 2008. [2] k. k. y. lai and j. l. zaichkowsky, “brand imitation: do the chinese have different views?” asia pacific journal of management, vol. 16, no. 2, pp. 179-192, august 1999. [3] k. rong, z. liu, and y. shi, “reshaping the business ecosystem in china: case studies and implications,” journal of science and technology policy in china, vol. 2, no. 2, pp. 171-192, 2011. [4] m. w. peng and p. s. heath, “the growth of the firm in planned economies in transition: institutions, organizations, and strategic choice,” academy of management review, vol. 21, no. 2, pp. 492-528, april 1996. [5] y. hu, “why shanzhai mobile phones thrive?” management & technology of sme, vol. 18, pp. 13-114, 2009. [6] m. treleven and s. b. schweikhart, “a risk/benefit analysis of sourcing strategies: single vs. multiple sourcing,” journal of operations management, vol. 7, no. 4, pp. 93-114, december 1988. [7] t. staake, f. thiesse, and e. fleisch, “business strategies in the counterfeit market,” journal of business research, vol. 65, no. 5, pp. 658-665, 2012. [8] x. liu, y. xie, and m. wu, “how latecomers innovate through technology modularization: evidence from china’s shanzhai industry,” innovation, vol. 17, no. 2, pp. 266-280, 2015. [9] j. chen, k. g. chang, and y. s. su, “innovation management: winning competitive advantage in the globe,” taipei, taiwan: best-wise publishing, 2015. [10] x. liu, y. wu, and g. lai, “strategic transformation of shanzhai firms: evidence from a chinese case,” international journal of technology, policy and management, vol. 15, no. 1, pp. 77-94, 2015. [11] f. kodama, “measuring emerging categories of innovation: modularity and business model,” technological forecasting and social change, vol. 71, no. 6, pp. 623-633, 2004. [12] j. k. gershenson, g. j. prasad, and y. zhang, “product modularity: definitions and benefits,” journal of engineering design, vol. 14, no. 3, pp. 295-313, 2003. [13] d. campagnolo and a. camuffo, “the concept of modularity in management studies: a literature review,” international journal of management reviews, vol. 12, no. 3, pp. 259-283, 2010. [14] c. y. baldwin and k. b. clark, “managing in an age of modularity,” managing in the modular age: architectures, networks, and organizations, pp. 84-93, 2003. [15] r. n. langlois, “external economies and economic progress: the case of the microcomputer industry,” business history review, vol. 66, no. 1, pp. 1-50, 1992. [16] j. zhou, y. tong, and j. li, “disruptive innovation for the base of the pyramid market: a case study on china’s shanzhai cell phone industry,” international journal of innovation and sustainable development, vol. 6, no. 4, pp. 392-419, 2012. [17] c. y. baldwin and k. b. clark, “design rules: the power of modularity,” cambridge, ma: mit press, vol. 1, 2000. [18] c. k. lee and s. c. hung, “institutional entrepreneurship in the informal economy: china's shan‐ zhai mobile phones,” strategic entrepreneurship journal, vol. 8, no. 1, pp. 16-36, 2014. [19] s. zhu and y. shi, “shanzhai manufacturing: an alternative innovation phenomenon in china: its value chain and implications for chinese science and technology policies,” journal of science and technology policy in china, vol. 1, no. 1, pp. 29-49, 2010. [20] m. a. schilling and h. k. steensma, “the use of modular organizational forms: an industry-level analysis,” academy of management journal, vol. 44, no. 6, pp. 1149-1168, 2001. [21] m. g. harvey and i. a. ronkainen, “international counterfeiters-marketing success without the cost and the risk,” columbia journal of world business, vol. 20, no. 3, pp. 37-45, 1985. [22] a. shilov, “amd: development of zen cpus is our largest r&d spending now,” https://www.kitguru.net/components/cpu/anton-shilov/amd-development-of-zen-process or-is-our-largest-rd-spending-now/, june 2017. [23] h. l. chen, “xiaomi joins apple and samsung, developing its own processors,” https://http://www.ithome.com.tw/news/112404, june 2017. [24] j. mitra and s. chunlin, “innovation, entrepreneurship, and governance: the shanzhai handset business,” icsb world conference proceedings, international council for small business (icsb), january 2010, p. 1. [25] s. adams, “mark zuckerberg regrets not learning from wechat sooner,” https://radiichina.com/mark-zuckerberg-regrets-not-learning-from-wechat-sooner/, march 2019. [26] n. statt and s. liao, “facebook wants to be wechat,” https://www.theverge.com/2019/3/8/18256226/facebook-wechat-messaging-zuckerberg-strategy, march 2019. proceedings of engineering and technology innovation, vol. 14, 2020, pp. 29 38 38 [27] i. deng, “facebook is changing, and this chinese app is the model,” https://www.inkstonenews.com/tech/zuckerberg-regrets-not-learning-wechat-facebook-turns-private-chats/article/30012 18, 2019, march 2019. [28] m. w. peng, d. ahlstrom, s. m. carraher, and w. s. shi, “an institution-based view of global ipr history,” journal of international business studies, pp. 1-15, 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/).  proceedings of engineering and technology innovation, vol. 24, 2023, pp. 73-86 an effective supervised machine learning approach for indian native chicken’s gender and breed classification thavamani subramani*, vijayakumar jeganathan, sruthi kunkuma balasubramanian department of electronics and instrumentation, bharathiar university, coimbatore, india received 04 april 2023; received in revised form 07 april 2023; accepted 11 april 2023 doi: https://doi.org/10.46604/peti.2023.11361 abstract this study proposes a computer vision and machine learning (ml)-based approach to classify gender and breed in native chicken production industries with minimal human intervention. the supervised ml and feature extraction algorithms are utilized to classify eleven indian chicken breeds, with 17,600 training samples and 4,400 testing samples (80:20 ratio). the gray-level co-occurrence matrix (glcm) algorithm is applied for feature extraction, and the principle component analysis (pca) algorithm is used for feature selection. among the tested 27 classifiers, the fg-svm, f-knn, and w-knn classifiers obtain more than 90% accuracy, with individual accuracies of 90.1%, 99.1%, and 99.1%. the bt classifier performs well in gender and breed classification work, achieving accuracy, precision, sensitivity, and f-scores of 99.3%, 90.2%, 99.4%, and 99.5%, respectively, and a mean absolute error of 0.7. keywords: native chicken breed classification, gender classification, machine learning algorithms, glcm, pca 1. introduction gender and breed predictions for native chickens are important tasks in large -scale poultry production, which require human intervention. chicken gender identification plays a major role in chicken ratio management, and breed identification is essential in large-scale production to avoid farm-level cross-breeding. with the advancement of machine learning (ml) approaches, ml can be implemented in poultry farms to perform identification activit ies. this section discusses the types and importance of native chickens, the classification of gender and breed, and the feasible approaches for poultry management. 1.1. importance of native chicken production native chickens have good survival capabilities, and they obtain good yields without any special diet. the native breed can survive with kitchen wastes, insects, and greens [1]. because of good survivability, harsh environmental adoption, and simple housing facility, rural poultry farmers prefer native breeds’ meat and egg. the main motive for native poultry farming is the high monetary value of their meat and eggs. the tamil nadu (tn) government also encourages open-source poultry farming with native chicken breeds to maximize income and provide nutritious food to underdeveloped villages. the medium-scale native chicken production is also a key to empowering women and unemployed youths in rural regions of tn [2]. less production creates demand for native chicken meat and egg in the markets. therefore, native chicken meat and egg are more expensive than other poultry meat and egg to date. owing to the benefits for health, consumers are willing to spend more on native breeds of india [3]. large-scale native-breed poultry farming was introduced to meet the growing demand. in large-scale production houses, the breed selection is based on individual characteristics, including growth period, laying, and * corresponding author. e-mail address: thavamani.ei@buc.edu.in proceedings of engineering and technology innovation, vol. 24, 2023, pp. 73-86 74 hatching performances. native chicken breeds have the advantages of well environmental adaptability, low financial input, suboptimal rearing conditions, abundant protein nutrients in meat and eggs, and genetic diversity. moreover, native breed hens are good sitters and have good hatching capability and broodiness. [4-5]. 1.2. types of native chicken breeds the chicken breed is defined according to individual characteristics and features. in india, nineteen varieties of native chicken breeds are registered in the icar-national bureau of animal genetic resources. the following names are the registered chicken breed: ankaleshwar, aseel, busra, chittagong, danki, daothigir, ghagus, harringhata black, kadaknath, kalasthi, kashmir favorolla, miri, nicobari, punjab brown, tellichery, mewari, kaunayen, hansli, and uttara [4]. 1.3. difficulties of identify the chicken gender and breed in commercially produced large-scale native chicken meat and egg production industries, gender and breed classification is one of the time-consuming tasks for farmers [6]. in manually operated production farms, humans may be easily affected during the poultry pathogens’ period [7]. in modernized poultry production industries, these tasks are handled by machines. since many researchers proposed the artificial intelligence (ai) approach for various management works, this study adopted the ml approach to change the classification of chicken gender and breed from human vision to machine vision. 1.4. different ml approaches for poultry industry management the need for human resources in task management will vary based on the poultry production rate, including small, medium, and large scales of native chicken breeds’ meat and egg production industries. backyard farming is unsuitable for large-scale production as it requires more area and human resources for management. in this automated world, different technologies offer reasonable solutions to poultry sector management [8]. ml-based image processing or computer vision approach allows the user to recognize and determine the particular tasks that rely upon the specified tools such as surveillance cameras, sensors, and audio devices. in general, ml methods have developed from the learning process, and these methodologies need to learn through experiences to perform a particular task [9]. visual data has trained ml approaches to perform tasks, such as analysis, processing, and recognizing; moreover, various key algorithms are also used for further recognition, object detection, and classification works. in recent years, the poultry industry has started to utilize ml concepts for different purposes, including chicken gender classification [6], chicken floor distribution monitoring [10], and behavior monitoring. these works can be thoroughly distinguished as chicken lameness prediction [11-12], laying hen behavior monitoring [13], vision-based farm management [14], meat safety management [15-17], woody severity identification, and poultry data analysis. furthermore, earlier chicken disease detection has been implemented to restrict feeding behavior, broiler chicken growth monitoring, and health prediction [18-20]. according to previous studies, the ml approach is suitable for the classification of chicken gender and breed. the proposed ml approach in this study achieves an acceptable accuracy rate through cpu systems. 1.5. proposed system and contributions technological advances have revolutionized the poultry industry’s role and structure in india. it became one of the most specialized enterprises in many parts of the country. many researchers have reported that ml approaches are more suitable and cost-effective for medium and large-scale poultry production industries [6, 10-19]. in both medium and large-scale native chicken meat and egg production farms, the gender and breed identification tasks are done manually. in general, the farmers analyze the gender and breed based on the following features of appearance such as feather color, comb structure, beak shape, skin color, size of the body, and wing. this proposed work is to extract texture and color features from the dataset using the gray-level co-occurrence matrix (glcm) algorithm to complete this classification work. proceedings of engineering and technology innovation, vol. 24, 2023, pp. 73-86 75 2. literature review this section introduces the materials and methods using ml approaches developed by previous researchers. there are different approaches available for poultry task management, such as the classification of chicken gender, monitoring and analysis of chicken activities, and earlier prediction of chicken health, etc. support vector machine (svm) and decision tree (dt) classifiers were used to classify the meat sample based on nir spectral information, ph value, water holding capacity, and meat color [18]. compared with the svm classifier, dt obtained effective results of true positive (tp) 0.778, false positive (fp) 0.199, precision 0.747, recall 0.778, and f-measure 0.741 for meat classification. the feeding and drinking behavior of chicken breeds “rhode island red” and “barred plymouth rock” were analyzed using ethovision xt 10 software, and the chickens' videos were recorded using infrared-equipped and closed-circuit tv dome cameras with a density of 1.45 pixels/cm2 or 0.688 cm2 per pixel [19]. 3d vision and ml approach was developed to detect sick and dead chickens with zigbee protocol ad-hoc network, and 40,000 mortality sample data was collected through a foot ring sensor [20]. this model measured the 3d displacement of the chicken using a foot ring and calculates the variation through their activity. ml classifiers such as svm, k-nearest neighbour (knn), naive bayes (nb), dt, and back propagation (bp) were used to analyze the health status of chickens and determine whether they were active, listless, sick, or dead. in this work, bp performed better than the remaining algorithms, for recognition accuracy of active 99.9%, listless 95.2%, sick 95.6%, and dead 100%. the earlier prediction of chicken lameness using the linear regression (lr) method is necessary to avoid severe chicken injuries [21]. the experiment was carried out on 250 cocks that were 35 days old. the 3d kinetic camera used in this experiment obtained 640 × 480 pixels at a 2 m distance from the object. a double threshold algorithm was used to segment the chicken images with the values of minimum 2,000 and maximum 2,800. the lr classifier obtained an accuracy of 93% in classifying the lame chickens. a genetic marker with the minimum amount of single nucleotide polymorphism (snp) was used to identify the korean native chicken breeds by analyzing the linkage disequilibrium (ld) and genome-wide association study (gwas) relationship [22]. this work used 283 samples from 20 native korean chickens. principle component analysis (pca) selected the genetic features, and the result of eight ml classifiers was compared. the classifier model adaboost (ab) in this research achieved a maximum accuracy of 99.6%. mx control center v2.4 mobutix ag software and linear real-time model-based chicken health and growth rate monitoring system were developed to identify sick broilers early [23]. the eyenamic data prediction algorithm was trained and tested using the images obtained from an internet protocol (ip) camera with 1280 by 960 pixels and a 0.5 hz frame rate for monitoring the chicken. more than five million chicken images were used for training and testing the prediction algorithm. as per the reports, the eyenamic system could predict sick chickens based on their feeding and drinking behavior and obtained 95.24% accuracy. chicken feeding and drinking behavior determine the growth rate [24]. to monitor chickens’ feeding activity, the general linear model was used to assess the relationship between the variables such as occupied area, activity index, and total birds present in an area. the images were analyzed within a rectangular bounding box of 180 × 170 pixels with 16 birds/m2. the dataset consists of 14,000 broilers aged between 17 to 24 days and obtained 86.4% of prediction accuracy for effective eating. in feeding behavior analysis, faster region-based convolutional neural network (faster r-cnn) based object detection and svm-based behavior classification were used [25]. this work used 9,040, 2,260, and 2,826 chicken images for training, validating, and testing. adjacent frames range from 20 to 150 pixels with five pixels intervals. faster r-cnn-based behavior analyzer obtained 100% precision, recall, and f-score for object detection with 3.1 to 4.2mm root mean square errors (rmses). the svm classifier obtained a value greater than 98% of precision, recall, and f-score with 19 mm rmses. proceedings of engineering and technology innovation, vol. 24, 2023, pp. 73-86 76 li et al. [26] followed the faster r-cnn method to examine the spatial and temporal broiler stretching behavior. for this work, 580 × 680 pixels 9,600 images were utilized, including 6,144 for training, 1,536 for validation, and 1,920 for testing the classifier. two sets of the dataset were used, and the optimal label method with a five-fold cross-validation method was also used to evaluate the detector. the experimental results obtained 99%, 99%, 94%, and 92% accuracy, specificity, recall, and precision in this proposed method. 3. proposed methodology this study extends the investigation using 27 supervised ml classifiers, feature extraction, and feature optimization algorithms for indian native chicken gender and breed classification work. fig. 1 depicts six steps for the overall workflow of the proposed work. (1) dataset preparation, (2 and 3) pre-processing, (4) segmentation (5) feature extraction, (6) pca for feature optimization, training, classifiers training and testing, and performance evaluation. fig. 1 classification framework 3.1. dataset (a) hen (b) cock fig. 2 common features comb ear lobe beak wings tail proceedings of engineering and technology innovation, vol. 24, 2023, pp. 73-86 77 cock hen (a) peela cock hen cock hen (b) yakub (c) nuri cock hen cock hen (d) kagar (e) chitta cock hen cock hen (f) sabja (g) parrot beak cock hen cock hen (h) chittagong (i) kadaknath cock hen cock hen (j) hitcari (k) new hampshire fig. 3 sample images of the native breed hardware and software requirements proceedings of engineering and technology innovation, vol. 24, 2023, pp. 73-86 78 for this investigation, the applied dataset consists of eleven varieties of indian native chicken images, including chitta, chittagong, hitcari, kadaknath, kagar, new hampshire, nuri, parrot beak aseel, peela, sabja, and yakub from a rural poultry production houses of dharmapuri district in tn state. the data collection was done through a smartphone camera, and this proposed work used a 22,000 images dataset that consists of manually labeled chicken gender and breed, including cock and hen in each class. this dataset has 1,000 cock and 1,000 hen images for each breed, 17,600 images for training, and 4,400 for testing (ratio of 80:20). figs. 2(a) and (b) indicate common features of hen and cock. fig. 3 shows the eleven types of native indian chicken breeds. each figure mentions the chicken gender: cocks and hens.the classifications of chicken gender and breed initially followed the video-to-image conversion, namely video to frame. the conversion was completed by using free video to jpg converter software (10 frames/sec), and this work was carried out using matlab 2021a on a desktop 3g17ucc with intel® core ™ i7-7700 processor and 64-bit os × 64 processors. 3.2. pre-processing the raw dataset usually contains various noise factors, like irregular intensity distribution resulting from different camera positions during video collection. to avoid diminishing system performance, pre-processing is mandatory after converting the frame from the video files. in fig. 1, steps 1 to 4 are important steps in this proposed work. in step 1, the chicken gender and breed dataset is converted from a video file to images (10 frames/sec) indicated. the actual size of the images is different; therefore, optimizing the storage size and minimizing the computational time for image resizing is important. steps 2 and 3 are contrast enhancement. step 4 mentions using k-means clustering algorithms to segment the chicken images from the background. in step 5, feature extraction from rgb to grayscale conversion and noise removal techniques are applied. 3.3. video-to-image conversion frame extraction is used to segment a long video sequence into the smaller unit for further processing. a smartphone camera was used for data collection, then the dataset was separated manually based on breed features. here, the joint photographic experts group converter software application was used for the video file-to-image conversion (10 frames/sec). 3.4. image resizing in image data, pixel dimension, color depth, and image format will determine the file’s size, which refers to the number of pixels in an image. image resizing is an essential step in ml. in contrast, image resizing makes the image larger or smaller without removing anything from the original data by changing an image’s height, width, or size. this image-resizing process helps the ml models to train significantly faster. the previous chicken breed images were different in size; in this present dataset, the images are resized in 256 × 256 pixel ranges, and nearest neighbor algorithms were used to resize the images. as the size of the extracted image for texture and color features increased, the values of all features also increased proportionally. therefore, compared to the size of 126 × 126 pixel range, the optimum size of 256 × 256 used for extraction is better resolution and minimum loss of information. 3.5. contrast enhancement when resizing the images causes low contrast or blur portions, contrast enhancement techniques are used to overcome them. partial spatial starching algorithms are applied to enhance resized images by stretching and compressing the image without any information loss. this algorithm mapped and stretched the new pixels from a lower threshold value to an upper threshold value, and the remaining pixels were compressed. proceedings of engineering and technology innovation, vol. 24, 2023, pp. 73-86 79 3.6. k-means clustering the k-means clustering algorithm is to cluster the dataset into k-number groups. this clustering function has two stages, the first algorithm calculates the k-centroid, and the next part clusters each point into the nearest centroid points from the dataset. the euclidean distance method was used to define the nearest centroid (minimum euclidean distance) point of the given dataset. subsequently, the k-means clustering algorithm was used to reduce the sum of the distance from the chicken image to its cluster centroid of all clusters. this process was repeated until getting less error value. finally, clustered pixels were converted into an image [27]. the clustering images are shown in step 4 in fig. 1. 3.7. rgb to grayscale conversion and contrast enhancement in a pre-processing step, the images were converted from rgb to grayscale, which can be completed by removing the hue and saturation information while remaining the luminance. improving an image’s visibility by adjusting the relative brightness and darkness of the object is called contrast enhancement. the contrast and tone can be modified from the image by mapping the grayscale and adding new values through a grayscale transform. to enhance the grayscale image contrast, the contrast-limited adaptive histogram equalization (clahe) algorithm was utilized. the functionality and efficiency of this algorithm were based on the following parameters, which were the tiles value and clip limits presented in an image. the clahe method contrast enhancement was divided into three steps, tile generation, histogram equalization, and interpolation. the image's local contrast was enhanced by the adaptive method. initially, the algorithm split the images into clear blocks and equalized the histogram. then, improving the local contrast and edges of all blocks was completed. the main merit of this method is avoiding noise amplification [28]. 3.8. noise removal images are distorted due to different types of noise, such as gaussian noise, salt & pepper noise, poisson noise, and speckle noise. it may happen during data acquisition, transmission, processing, or coding. the salt & pepper noise will affec t the quality of an image. to overcome this issue, the denoising filter can effectively restore appearance in terms of speed, quality, and low complexity [28]. the conversion of images from video files sometimes does simple blurring, embossing, outlining, and sharpening effects; hence, spatial filtering can be applied to identify and clean the object from an image. it replaces every element with a weighted average of its neighborhood pixels. the spatial filtering changes the pixel intensities according to the neighboring pixel intensities for sharpening and smoothing the image. based on function, the spatial filter is classified into a smoothing spatial filter for image blurring and noise reduction; and a sharpening filter for removing blurring and highlighting the edges in an image. this filter response is based on neighboring pixels, which is more suitable for removing the impulse or salt & pepper noise from an image. image restoration follows the order statistics filter, and this spatial filter function is based on the neighborhood pixels operated by a median filter. the median filter replaces a pixel’s value with the gray level median in the neighborhood pixels . this filter has excellent noise-reduction capabilities and effectively reduces salt & pepper and impulse noise without smoothing effects. in this proposed work, the median spatial filter was used for noise removal. the salt & pepper noise reduction method using a median spatial filter can be expressed as: ( ) ( ) ( )ˆ , , ,=    f x y medium s t sxy g s t (1) where the (x, y) is the spatial domain of image information. proceedings of engineering and technology innovation, vol. 24, 2023, pp. 73-86 80 3.9. glcm algorithm for feature extraction the features must be derived from the image, created, and stored in a signal format with time and frequency domains. the frequency domain conveys more information about the morphological nature of the image than the spatial domain characteristics [28]. for further work, feature extraction is essential. this work followed the texture and color feature extraction method. the glcm algorithms extracted the second-order statistical information from the grayscale level between neighboring pixels representing an image. this feature extraction algorithm is helpful in texture feature analysis by calculating the pairs of pixels with declared values and the spatial relationship between the pictures. the glcm was constructed by n = 18, 16, and 32 levels of quantization. the image contained objects of a variety of shapes and sizes that were arranged in horizontal and vertical directions. the example specified a set of horizontal offsets that only vary in distance. second-order metrics analyzed the relationship between pixel pairs. a co-occurrence matrix was used to calculate the texture value presents in images. this matrix is a function of both the angular relationship and distance between two neighboring pixels. it shows the number of occurrences of the relationship between a pixel and its specified neighbor. glcm has proved to be a popular statistical method of extracting textural features from images. in the 1973s, haralick et al. defined the possibilities of the following 14 features that can extract from a single image, including contrast, correlation, energy, homogeneity, sum of squares and variance, entropy, sum average, sum entropy, sum variance, difference variance, difference entropy, information measure of correlation 1, information measure of correlation 2, and maximum correlation. the glcm is the function that can only be performed to extract features on grayscale images. this work extracted the following four features: contrast, correlation, energy, and homogeneity. table 1 mentions the values of the above-mentioned features for each breed. table 1 four values of glcm feature samples for each breed contrast correlation energy homogeneity breed name 0.0813 0.9856 0.2625 0.9712 kagar cock 0.1441 0.9874 0.1266 0.9357 kagar hen 0.7335 0.8845 0.0576 0.7652 nuri cock 0.0597 0.9882 0.2404 0.9702 nuri hen 0.0591 0.9857 0.1916 0.9705 peela cock 0.1282 0.9682 0.2147 0.9474 peela hen 0.1569 0.9576 0.1537 0.9225 yakub cock 0.1242 0.9707 0.1687 0.9407 yakub hen 0.1394 0.9573 0.1678 0.9305 chitta cock 0.2716 0.892 0.2147 0.888 chitta hen 0.0994 0.9645 0.197 0.9526 chittagong cock 0.2267 0.9622 0.1016 0.8925 chittagong hen 0.1998 0.9752 0.109 0.9071 hitcari cock 0.1432 0.9531 0.2762 0.931 hitcari hen 0.1617 0.9796 0.101 0.9217 sabja cock 0.3202 0.9314 0.1148 0.8561 sabja hen 0.1279 0.9894 0.1229 0.9378 kadaknath cock 0.0707 0.993 0.1735 0.9655 kadakanath hen 0.1806 0.9835 0.2167 0.9208 parrot beak aseel cock 0.3056 0.9498 0.0995 0.8593 parrot beak aseel hen 0.1671 0.979 0.2328 0.9665 new hampshire cock 0.0398 0.9908 0.5868 0.9811 new hampshire hen contrast: measures intensity between a pixel and its neighboring pixels presented all over the image. the inertia and variance are called contrast properties. the contrast range of the image is measured in the field between [0 (size) (glcm,1)-1^2]. the calculation of contrast can be obtained by: proceedings of engineering and technology innovation, vol. 24, 2023, pp. 73-86 81 ( ) 2 , , ,=  m ncontrast m n p m n (2) where p(m,n) is the spatial relationship between the images. correlation: how a pixel related to its neighboring pixels can be obtained using correlation. in the image, the correlation ranges between -1 to 1. the correlation calculation between the pixels presented in an image can be expressed as: ( )( ) ( ) , , , ,    =   m n m m n m p m n correlation m n (3) where -1 indicates that variables are inversely proportional to each other, and 1 demonstrates that variables are directly proportional to each other. energy: in the constant image, if the energy value is 1, the properties and angular second moment are uniform. the energy from the image can be calculated as: ( ) 2 , ,=  m nenergy p m n (4) homogeneity: the returns values that measure the proximity of the distribution of elements in the glcm to the glcm diagonal. in images, the homogeneity ranges between 0 and 1, and its value is 1 for a diagonal glcm. the proximity distribution can be found as: ( ) , , 1 , = +  m n p m n homogeneity m n (5) 3.10. pca model for glcm extracted feature selection in ml approaches, the pca model is used for dimensionality reduction [29]. with the help of orthogonal transformation, the pca converts the observations of correlated features into linearly uncorrelated features, which are called the principal components. this technique reduces the data variance from the given dataset and draws a strong pattern. in this work, the purpose of the pca model is to select the glcm extracted features (contrast, correlation, energy, and homogeneity). according to the experimental results of the proposed work, glcm with four features achieved 99.1% of classification accuracy for bt classifiers. however, the enabled pca with glcm features (correlation and homogeneity) achieved a better result of 99.3%. cross-validation is used to check the ability of the proposed algorithm for new data prediction, which protects against overfitting by portioning the dataset into folders and assessing the accuracy of every folder individually, and this work followed 10-fold cross-validation. 4. performance evaluation in ml techniques, performance evaluation matrices are available for classification works. the balanced dataset used accuracy, recall, and precision matrix for performance evaluation, and the imbalanced dataset used auc-roc and gini coefficient to evaluate the performance. the following matrix evaluated the chicken gender and breed classification work based on accuracy, precision, recall, and f-score. from the classification results, the performance was evaluated based on tp, true negative (tn), fp, and false negative (fn), the evaluation of the classifier’s performance can be expressed as: 100 + =  + + + tn tp accuracy fn tn tp fp (6) the accuracy calculation is to identify the appropriate categorization of the classifier by the given samples, and this can be calculated as: proceedings of engineering and technology innovation, vol. 24, 2023, pp. 73-86 82 100=  + tp precision tp fp (7) the ability of optimistic sample prediction by the classifier can be calculated as: 100=  + tp recall tp fn (8) the recall calculation is to measure the positive sample detection ability of the proposed model. the ratio between the number of optimistic predictions in the total number of positive samples can be calculated as: 2 -s 100   =   recall precision f core recall precision (9) 5. experimental results and discussion 5.1. classification results supervised ml algorithms were used to recognize the chicken gender and breed with texture and color features extracted by glcm algorithms. this proposed work performance was tested with two modes, including glcm with disabled pca and glcm with enabled pca. when pca is disabled, glcm defaults to selecting all the glcm features; when pca is enabled, the extracted features are selected, optimized, and trained by the classification algorithms. the feature-based homogeneity (used to find similar pixels in a given dataset) and correlation (measures the relationship between the given dataset) obtained high accuracy. this experiment tested 27 types of classifiers, based on classification results out of 8 class classifier models achieved more than 90% of accuracy. the outcome of the proposed work is that the bt classifier obtained higher performance than the remaining algorithms. the bt classifier obtained 99.1% of accuracy without pca, and the bt classifier obtained 99.3% accuracy with correlation and homogeneity features enabled pca. fig. 4 classification accuracy of 27 types of classifiers based on the classification performance with enabled pca, the following seven classifiers achieved lower performance metrics than the bt classifiers, including fine gaussian support vector machine (fg-svm), fuzzy k-nearest neighbor (f-knn), multidimensional k-nearest neighbor (m-knn), subspace-knn, weighted k-nearest neighbor (w-knn), cosine-knn, and cubic-knn. the fg-svm obtained 92.4%, 91%, 92%, and 90.1% for accuracy, precision, sensitivity, and f-score. proceedings of engineering and technology innovation, vol. 24, 2023, pp. 73-86 83 the m-knn model achieved 99.3%, 99.1%, 98.3%, and 98.7% of accuracy, precision, sensitivity, and f-score. f-knn classifier achieved 99.3% accuracy, 99.4% precision, 99% sensitivity, and 99.2% f-score value. the subspace knn model obtained 90.5% accuracy, 99.2 % precision, 99% sensitivity, and 99.1% f-score. w-knn achieved 99.3%, 99.4%, 99.1%, and 99.2 % of accuracy, precision, sensitivity, and f-score. cosine-knn obtained the performance evaluation values of 98.3%, 98.2, 98.2%, and 98.4 for accuracy, precision, sensitivity, and f-score. the cubic-knn model achieved 98.4% accuracy, 98.5% precision, 98.1% sensitivity, and 98.2% f-score. fig. 5 native chicken gender and breed classification accuracy of eight types of classifiers fig. 6. ppv and fdr percentage of bt classifier proceedings of engineering and technology innovation, vol. 24, 2023, pp. 73-86 84 the remaining algorithms performed poorly for the chicken gender and breed classification. fig. 4 shows the classification accuracy of 27 classifiers glcm features with an enabled pca for feature selection. fig. 5 shows the overall classification accuracy of eleven native breeds. fig. 6 shows the positive prediction value (ppv) and false detection rate (fdr) of the bt classifier with eleven native chicken breed datasets. compared to other classifiers, the bt classifier is the best-performed for all gender and breed classification work. table 2 shows the type of native chicken gender and breed classification. bt classifier obtained 100% accuracy, precision, sensitivity, and f-score for both cock and hen prediction of peela, yakub, nuri, kagar, parrot beak aseel, chittagong, kadaknath, hitcari, and new hampshire breeds. in the chitta classification, the bt classifier obtained 96.2% of accuracy and precision and 100% of sensitivity and f-score. the sabja classification obtained 96.9% of accuracy and precision, 94.3% of sensitivity, and 95.5% of f-score. table 2 the overall classification performance of the bt classifier chicken breed accuracy precision sensitivity f-score peela 100% 100% 100% 100% yakub 100% 100% 100% 100% nuri 100% 100% 100% 100% kagar 100% 100% 100% 100% chitta 96.2% 96.2% 100% 100% sabja 96.9% 96.9% 94.3% 95.5% parrot beak aseel 100% 100% 100% 100% chittagong 100% 100% 100% 100% kadaknath 100% 100% 100% 100% hitcari 100% 100% 100% 100% new hampshire 100% 100% 100% 100% overall performance 99.3% 90.2% 99.4% 99.5% 5.2. the accuracy comparison of the model with and without pca this section compares the classification performances of glcm features and enabled pca (homogeneity and correlation) models. the chart in fig. 7 indicates the percentage of accuracy with and without the pca model for the bt classifier. accuracy is considered the medium of measuring performance. this proposed method used glcm for texture and color feature extraction (contrast, correlation, energy, and homogeneity). the glcm also influenced pca by selecting extracted features, which enhanced the proposed approach's performance. bt classifier obtained a classification accuracy of 99.1% for glcm features; however, the enabled pca with correlation and homogeneity features significantly enhanced the proposed model classification accuracy of 99.3%. yao et al. [29] proposed a deep learning approach with computer vision, and this work achieved 96.85% accuracy for chicken gender classification. from the results, the proposed model obtained high accuracy for gender and breed classification work. fig. 7 the obtained accuracy rate of the bt classifier with and without the pca model proceedings of engineering and technology innovation, vol. 24, 2023, pp. 73-86 85 6. conclusions a supervised ml approach helps minimize human resources when classifying chicken’s gender and breed in the poultry meat and egg production industry. cock and hen ratio management is an important task in the poultry industry . on the other hand, breed identification can avoid the farm level of cross-breeding. in this study, a k-means clustering algorithm was used to classify chicken’s gender and breed via centroid-based image segmentation, which separated the chicken images from the background. the glcm approach was employed to extract the second-order statistical feature, including contrast correlation, energy, and homogeneity. after that, the pca model was utilized to reduce the optimal subset of the above-mentioned features. the combination of glcm and pca algorithms can effectively enhance the accuracy of classification. in addition, the bt classifier achieved accuracy, precision, sensitivity, and f-scores of respective 99.3%, 90.2%, 99.4%, and 99.5% with under 0.7 mean absolute error. the results show that the implementation of ml is useful in the modern large-scale native chicken production industry by utilizing glcm features with the pca model and the bt classifier. the classification of chicken's gender and breed with reduced human resources can be accomplished with the proposed work. this proposed work proves that supervised ml techniques with glcm futures are suitable for classifying the gender and breed of native chickens. in the future, improving the classification performance without the texture and color features of native chickens can be researched. acknowledgment the authors would like to thank the department of science and technology (bharathiar universitygrant number: 12019/if190264) for funding this project. conflicts of interest the authors declare no conflict of interest. statement of ethical approval all procedures performed in studies involving animals were in accordance with the ethical standards of the institution or practice at which the studies were conducted. statement of informed consent for this type of study, informed consent is not required. references [1] m. k. padhi, “importance of indigenous breeds of chicken for rural economy and their improvements for higher production performance,” scientifica, vol. 2016, article no. 2604685, 2016. [2] m. kumar, s. p. dahiya, and p. ratwan, “backyard poultry farming in india: a tool for nutritional security and women empowerment,” biological rhythm research, vol. 52, no. 10, pp. 1476-1491, 2021. [3] r. t. wilson, “an overview of traditional small-scale poultry production in low-income, food-deficit countries,” annals of agricultural & crop sciences, vol. 6, no. 3, article no. 1077, 2021. [4] s. haunshi and u. rajkumar, “native chicken production in india: present status and challenges,” livestock research for rural development, vol. 32, no. 11, article no. 181, november 2020. [5] m. kanakachari, h. rahman, r. n. chatterjee, and t. k. bhattacharya, “signature of indian native chicken breeds: a perspective,” world’s poultry science journal, vol. 78, no. 2, pp. 421-445, 2022. proceedings of engineering and technology innovation, vol. 24, 2023, pp. 73-86 86 [6] s. s. thavamani, j. vijayakumar, and k. sruthi, “glcm and k-means based chicken gender classification,” smart technologies, communication and robotics, pp. 1-5, october 2021. [7] h. m. hafez and y. a. attia, “challenges to the poultry industry: current perspectives and strategic future after the covid-19 outbreak,” frontiers in veterinary science, vol. 7, article no. 516, august 2020. [8] d. k. dittoe, e. g. olson, and s. c. ricke, “impact of the gastrointestinal microbiome and fermentation metabolites on broiler performance,” poultry science, vol. 101, no. 5, article no. 101786, may 2022. [9] c. okinda, i. nyalala, t. korohou, c. okinda, j. wang, t. achieng, et al., “a review on computer vision systems in monitoring of poultry: a welfare perspective,” artificial intelligence in agriculture, vol. 4, pp. 184-208, 2020. [10] y. guo, l. chai, s. e. aggrey, a. oladeinde, j. johnson, and g. zock, “a machine vision-based method for monitoring broiler chicken floor distribution,” sensors, vol. 20, no. 11, article no. 3179, june 2020. [11] x. yang, y. zhao, g. m. street, y. huang, s. d. filip to, and j. l. purswell, “classification of broiler behaviours using triaxial accelerometer and machine learning,” animal, vol. 15, no. 7, article no. 100269, july 2021. [12] p. pons, j. jaen, and a. catala, “assessing machine learning classifiers for the detection of animals’ behavior using depth-based tracking,” expert systems with applications, vol. 86, pp. 235-246, november 2017. [13] s. m. derakhshani, m. overduin, t. g. c. m. van niekerk, and p. w. g. groot koerkamp, “implementation of inertia sensor and machine learning technologies for analyzing the behavior of individual laying hens,” animals, vol. 12, no. 5, article no. 536, march 2022. [14] n. sharma, r. sharma, and n. jindal, “machine learning and deep learning applications-a vision,” global transitions proceedings, vol. 2, no. 1, pp. 24-28, june 2021. [15] a. siddique, s. shirzaei, a. e. smith, j. valenta, l. j. garner, and a. morey, “acceptability of artificial intelligence in poultry processing and classification efficiencies of different classification models in the categorisation of breast fillet myopathies,” frontiers in physiology, vol. 12, article no. 712649, 2021. [16] s. barbon, a. p. a. da costa barbon, r. g. mantovani, and d. f. barbin, “machine learning applied to near-infrared spectra for chicken meat classification,” journal of spectroscopy, vol. 2018, article no. 8949741, 2018. [17] m. pitesky, j. gendreau, t. bond, and r. carrasco-medanic, “data challenges and practical aspects of machine learning-based statistical methods for the analyses of poultry data to improve food safety and production efficiency,” cabi reviews, vol. 2020, https://doi.org/10.1079/pavsnnr202015049, january 2020. [18] x. zhuang, m. bi, j. guo, s. wu, and t. zhang, “development of an early warning algorithm to detect sick broilers,” computers and electronics in agriculture, vol. 144, pp. 102-113, january 2018. [19] b. milosevic, s. ciric, n. lalic, v. milanovic, z. savic, i. omerovic, et al., “machine learning application in growth and health prediction of broiler chickens,” world’s poultry science journal, vol. 75, no. 3, pp. 401-410, september 2019. [20] i. elawady and c. özcan, “a new effective denoising filter for high density impulse noise reduction,” turkish journal of electrical engineering and computer sciences, vol. 30, no. 4, article no. 15, 2022. [21] t. babu, t. singh, d. gupta, and s. hameed, “optimized cancer detection on various magnified histopathological colon imagesbased on dwt features and fcm clustering,” turkish journal of electrical engineering and computer sciences, vol. 30, no. 1, article no. 1, 2022. [22] p. k. bhagat, p. choudhary, and kh. manglem singh, “a comparative study for brain tumor detection in mri images using texture features,” sensors for health monitoring, vol. 5, pp. 259-287, 2019. [23] b. t. jijo and a. m. abdulazeez, “classification based on decision tree algorithm for machine learning,” journal of applied science and technology trends, vol. 2, no. 01, pp. 20-28, march 2021. [24] y. s. can, “stressed or just running? differentiation of mental stress and physical activityby using machine learning,” turkish journal of electrical engineering and computer sciences, vol. 30, no. 1, article no. 21, 2022. [25] g. li, x. hui, z. chen, g. d. chesser jr, and y. zhao, “development and evaluation of a method to detect broilers continuously walking around feeder as an indication of restricted feeding behaviors,” computers and electronics in agriculture, vol. 181, article no. 105982, february 2021. [26] g. li, y. zhao, z. porter, and j. l. purswell, “automated measurement of broiler stretching behaviours under four stocking densities via faster region-based convolutional neural network,” animal, vol. 15, no. 1, article no. 100059, january 2021. [27] n. dhanachandra, k. manglem, and y. j. chanu, “image segmentation using k -means clustering algorithm and subtractive clustering algorithm,” procedia computer science, vol. 54, pp. 764-771, 2015. [28] priyanka and d. kumar, “feature extraction and selection of kidney ultrasound images using glcm and pca,” procedia computer science, vol. 167, pp. 1722-1731, 2020. [29] y. yao, h. yu, j. mu, j. li, and h. pu, “estimation of the gender ratio of chickens based on computer vision: dataset and exploration,” entropy, vol. 22, no. 7, article no. 719, july 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 (http://creativecommons.org/licenses/by/4.0/).  proceedings of engineering and technology innovation, vol. 6, 2017, pp. 06 11 a study on the burn pattern experiments in structure fire gi-dong song 1 , man-sung hur 2,* , hyun-joo lee 1 1 department of scientific investigation, chungnam national university, daejeon, korea . 2 faculty of fire and safety, woosong college, korea. received 30 july 2017; received in revised form 10 august 2017; accept ed 12 august 2017 abstract the fire has created tremendous damage, yet skilled fire investigators can read the fire patterns left by the fire. through the fire patterns interpretation and analysis, fire investigators should be able to determine exactly how and where the fire or explosion started. fire patterns can help the fire investigator trace fire movement back to the point of origin. the origin and cause investigation for is attempting to determine the first material ignited, the source of ignition, and how the two brought together to cause the ignition. the investigators should read the fire scene to establish the area of origin. in this study, through burn pattern experiments, fire behavior in the compartment, fire pattern formation mechanism, fire patterns reproducibility, fire patterns persistence through flashover and full room involvement condition, fire scene reconstruction for fire origin determination, and other fire pattern’s temperatur e will be discussed and analyzed in depth. burn test was conducted to recognize and analyze fire patterns created in the phase of pre flashover or post flashover in the compartment (container house). in addition, fire pattern experiments are designed to test combustible material, including polyurethane foam with a lower thermal inertia or wooden chair with a higher thermal inertia. during these experiments, through the thermocouple fire pattern’s temperature will be measured in the stage of pre flashover or post flashover. keywords: burn pattern experiments, fire patterns persistence and reproducibility, thermocouple, pre and post-flashover 1. introduction fire investigators have relied on burn patterns as their basis for determining the fire origin. a burn pattern is the visible, measurable physical change or identifiable shapes formed by a fire effect, or a group of fire effects. fire origin determination is largely a matter of fire pattern recognition and analysis [1]. this burn pattern experiment is to s how a new paradigm of fire investigation through burn pattern analysis on a basis of scientific method which has been a tool to define the origin and cause of a fire. in structure fire, burn pattern tests are to reproduce the burn patterns , such as v shaped pattern, u shaped pattern, inverted cone pattern, clean burn, alligator char pattern, protected area and heat shadowing that were observed and identified in fire scenes. in addition, burn pattern tests are to analysis fire behavior and combustion charact eristics of material in stages of fire growth in compartment fire. soot and clean burn are closely related to carbon from incomplete combustion and flame plume from complete combustion[2]. * corresponding author. e-m ail address: m shur@ wsi.ac.kr proceedings of engineering and technology innovation, vol. 6, 2017, pp. 06 11 copyright © taeti 7 these tests confirmed that fire patterns such as clean burn and v-shaped pattern persist during post flashover conditions, as well as providing evidence of the evolution of these fire patterns, such as u-shaped pattern and inverted cone pattern. the most important finding from these burn tests were that patterns can provide substantial evidence for to determine accurately the correct origin of a fire. 2. experimental design 2.1. burn pattern experiment process this burn pattern experiments were conducted in a container house at industrial complex located in the suburbs of the metropolitan city of daejeon, south korea. fig. 1 the floor plan of compartment 2.2. container house materials container steel panels are mostly composed of ‘sm20c’(carbon steel for machine structure) and container compartment is 4,000mm wide, 2,400mm long, 2,400mm high. 2.3. surroundings conditions outside container house surroundings conditions outside container house include 24℃ of temperature, 42% of humidity, 3 m/s of wind speed, and wind direction from sw to ne. 2.4. experimental details for the thermocouple measurement, fire investigator established 7 points and 29 points of temperature measurement (fig. 1). the materials for a fire source include the sofa and chair at the center of container house, the flat tv set on the wall, the book shelves, the table of the kitchen sink, the refrigerator in the corner and the small size wastebasket on the floor of the compartment. during tes t fire investigator uses a disposable lighter and liquid accelerant paint thinner for lighting . proceedings of engineering and technology innovation, vol. 6, 2017, pp. 06 11 copyright © taeti 8 3. burn pattern formation and its analysis 3.1. specific burn pattern created during fire test go through the initiation phase (fig. 2) and go to the fully-developed phase (fig. 3) in the clean burn. fig. 2 soot development phase fig. 3 post flashover and full room involvement fig. 4 unidentifiable u-shaped pattern fig. 5 v-shaped pattern persists fig. 6 heat shadowing identified in the pro cess of fire scene reconstruction 3.2. burn pattern analysis in pre-flashover phase u-shaped pattern and inverted cone pattern were observed. in post -flashover phase v-shaped pattern, clean burn, heat shadowing and protected area, and alligator char, po ur pattern, drop down pattern were formed. 3.2.1. u-shaped pattern u-shaped patterns are created by the effects of radiant heat energy on the vertical surfaces , such as a wall. actually, u-shaped pattern can be formed by ignition occurred while mainly co oking or frying, reaching the ignition temperature. therefore , it is closely related to human errors. the ignition of cooking oils plays an important role in the problem of residential fir es. proceedings of engineering and technology innovation, vol. 6, 2017, pp. 06 11 copyright © taeti 9 during this test, the u shaped pattern was observed on the interior wall of container house. from cooking oil ignition and growth, the fire plume caused its upward current to create a u shaped pattern in the stage of pre flashover. but in the phase of post flashover or full room involvement, it should be noted to make it difficult or impossible for fire investigators to interpret and identify the u shaped pattern. (fig. 4) 3.2.2. inverted cone pattern inverted cones are commonly caused by the vertical flame plumes of the burning volatile fuels not reaching the ceil ing. inverted cone patterns are characterized by relatively short -lived fires and low hrr fires that do not fully evolve into floor-to-ceiling flame plumes or flame plumes that are not vertically restricted by ceilings. at this experimental setting, in itial inverted cone pattern cannot be well found and identified because initial burn pattern can be almost destroyed by the effects of intense radiant heat through the post flashover or the full room involvement. 3.2.3. v pattern the appearance of the v-shaped pattern is created by flames, convective or radiated heat from hot gases, and smoke within the fire plume. the v-shaped pattern often appears as demarcation lines defining the borders of the fire plume and less heated areas outside the fire plume. at this experiment, v-shaped pattern persists in the presence of an ignitable liquid through flashover and full room involvement fire. (fig. 5) 3.2.4. clean burn pattern clean burn is found on noncombustible surfaces where there has been direct flame contact which burns off soot or smoke deposits. smoke deposits on surfaces are subject to oxidation. the carbon will be oxidized to gases and disappear from the surface. 3.2.5. drop down(fall down) pattern the fire investigators should keep in mind that during t he growth and spread of a fire, burning debris often falls to lower levels, such as the floor and then burns upward from there. this is known as drop down (fall down). fall down can ignite other combustibles producing low burn patterns that can easily be confused with the area of fire origin. 3.2.6. pour pattern the pour pattern by liquid accelerant is the typical burn pattern identified in the arson fire scene. during burn test paint thinners have been poured on the floor and wooden chairs in this compa rtment. gasoline, kerosene, and paint thinner are the most common fuels as an accelerant. 3.2.7. alligator char pattern as the burning continues and more volatiles are driven out of the wood, the char layer becomes deeper and splits and cracks forming a char surface often described as 'alligatoring' that looks like the scales on an alligator's back. in some case alligator ’s pattern (large rolling blisters) was caused by the rapid intense movement of heat and flame. this condition may be associated with the use of an accelerant. at experiment, paint thinner have been poured at wooden chairs on the floor of container house and ignited. proceedings of engineering and technology innovation, vol. 6, 2017, pp. 06 11 copyright © taeti 10 3.2.8. heat shadowing heat shadowing results from an object blocking the travel of radiated heat, convected heat, or direc t flame impingement from its source of the material on which the pattern is produced. conducted heat, however, does not produce heat shadowing. this heat shadowing (wall clock) was identified in the process of reconstructing the fire scene. (fig. 6) it is believed that the wall clock fell to the floor of the compartment due to the effects of flashover and full room involvement fire. 3.2.9. protected area closely related in appearance to the resulting pattern of heat shadowing is a protected area. a protec ted area results from an object preventing the combustion products from depositing on the material. the patterns produced by protected areas may assist the fire investigators in the process of fire scene reconstruction during the origin determination by in dicating the location of objects in their pre-fire locations. therefore, heat shadows and protected areas may both provide valuable information or clues to fire investigators. 3.2.10. 'tear pattern’ tear pattern was created during combustion experiments for the pattern penetration of horizontal surfaces in a upward direction. 4. dis cus s ion and re s ult 4.1. fire behavior in the compartment the compartments or enclosures , such as container house, can represent any confined space which limits air supply. the fire development stages of structure fire include ignition period, growth period, flashover, fully developed period, and decay period [3]. usually the fire growth period is defined as the pre-flashover stage, and the post-flashover phase includes the fully developed period and the decay period. in most structure fires , it is only during the early growth stages that fuel is the limiting factor in the development and spread of the fire and during this stage the fire is defined as fuel -controlled [4]. in ventilation-controlled period, usually the stages of flashover or fully developed fire, fire growth is predominantly limited by air (oxygen) available in the compartment. on the decay stage, the fire would be changed from ventilatio n controlled to fuel controlled [5]. 4.2. fire patterns reproducibility and persistence this experiment demonstrated that nine of the burn patterns can be created during the burn test in the container house. the burn patterns reproduced during the combustion experiments include v p attern, u-shaped pattern, inverted cone pattern, heat shadowing, protected area, alligator char pattern, clean burn, pour pattern, drop down(fall down) and unique 'tear pattern'. the fire patterns created at the early stages of fire growth may or may not p ersist through flashover and full room involvement. usually a v pattern persists in the presence of ignitable liquid as arson fire accelerant, even during post flashover and full room involvement. flashover sometimes creates new fire patterns . it should be noted that the flashover often destroy and distorts or obscures the fire patterns at or near the area(point) of fire origin, making it difficult or impossible to identify the fire patterns [6]. proceedings of engineering and technology innovation, vol. 6, 2017, pp. 06 11 copyright © taeti 11 4.3. fire scene reconstruction protected areas have been shielded from combustion product, such as char and soot. heat shadowing results from an object blocking the travel of radiated heat, convected heat, or direct flame impingement from its source of the material on which the pattern is produced. conducted heat does not create heat shadowing. heat shadowing can change, mask, or prohibit the production of identifiable demarcation lines that may have appeared in that material. but patterns produced by the heat shadowing may assist the fire investigators in the proc ess of fire scene reconstruction during the fire origin determination. heat shadowing and protected areas can help the fire investigators reconstruct the pre -fire location of contents in the room. 5. conclusions the fire patterns reproduced from this combustion experiments in compartment include nine of fire patterns , such as a u shaped pattern, protected area that can actually be identified at a fire scene. in the phase of the pre flashover u-shaped pattern, inverted cone pattern were created, however, in the stage of post-flashover v-shaped pattern, clean burn, fall down, heat shadowing and protected area, alligatoring, and pour pattern were produce d. this burn test demonstrated that the fire pattern , such as v pattern persisting through flashover and full room involvement condition in a compartment fire. on the other hand, u-shaped pattern no longer persists through post flashover and full room involvement. once the area and possibly the point of origin is identified, the fire investigator should pinpoint the cause of fire or explosion. the created specific fire patterns' temperatures measured by the thermocouple range from 300℃ below at a u-shaped pattern, inverted cone pattern to 876℃ ~ 930℃ on v-shaped pattern, clean burn, fall down, heat shadowing and protected area, and to 903℃ at alligatoring, and pour pattern. references [1] national fire protection association, nfpa 921 guide for fire and explosion investigations, 2011. [2] international fire service training association (ifsta), introduction to fire origin and cause, 3nd ed. ifsta committee, 2003. [3] d. d. h. john, kirk’s fire investigation, 6th ed., milwaukee: brady, 2007. [4] c. l. tien, k. y. lee, and a. j. stretton, the sfpe handbook of fire protection engineering, 4nd ed. quincy: nfpa inc., 2008. [5] v. babrauskas, “the sfpe handbook of fire protection engineering,” 4nd ed. quincy: nfpa inc., pp. 12, 2008. [6] g. e. gorbett, b. meacham, and c. wood, “development and assessment of a decision support framework for enhancing the forensic analysis and interpretation of fire patterns,” diss. dissertation, worcester polytechnic institute, 2015.  proceedings of engineering and technology innovation, vol. 5, 2017, pp. 19 24 pumped-storage scheduling using glowworm swarm algorithm po-hung chen 1,* , chun-yao lee 2 and horng-lin shieh 3 1,3 department of electrical engineering, st. john’s university, new taipei city, taiwan, roc. 2 department of electrical engineering, chung yuan christian university, chung li city, taiwan , roc. received 27 july 2017; received in revised form 17 august 2017; accept ed 21 august 2017 abstract this paper presents new solution methods and results based on a glowworm swarm algorithm for solving the 24-hour pumped-storage generation scheduling problem. complete solution algorithms and encoding/decoding techniques are proposed in the paper. the optimal schedules of both pumped -storage and thermal units are concurrently obtained within the evolutionary process of evaluation functions. significantly, no hydro -thermal iteration is needed. the proposed approach is applied with success to an actual utility system, which consists of four pumped-storage units and 34 thermal units. the results indicate the attractive properties of the glowworm swarm algorithm in practical application, namely, a highly optimal solution cost and more robust convergence behaviour . keywords: pumped-storage, glowworm swarm algorithm, hydro -thermal iteration. 1. introduction the exact optimal solution to the p/s scheduling problem can be obtained by exhaustive enumeration of all p/s and thermal unit combinations at each time period. however, the burden of computation makes it unacceptable for realistic applications. conventional methods for solving the p/s scheduling problem are based on decomposition approaches that involve a hydro and a thermal subproblem [1-3]. these two subproblems are usually coordinated by lagrange multipliers, and then the optimal generation schedules of both p/s and thermal units are obtained via repetitive hydro -thermal iterations. a well-recognized difficulty is that the solutions to these two subproblems may oscillate between maximum and minimum generations with slight changes of the multipliers [4]. as a result, the solution cost usually gets stuck at a local optimum rather than at the global optimum. however, the optimality of solution is very important to electric utility. even a small reduction in percentage production co st may lead to large saving of money. obviously, a complete and efficient algorithm for solving the p/s scheduling problem is still in demand. recently, a global optimization technique known as glowworm swarm algorithm (gsa) has emerged as a candidate for many optimization applications due to its flexibility and efficiency [5-6]. gsa has been successfully applied in various areas such as economic dispatch [7], multimodal functions optimization [8], optimal power flow [9], and so on. in this paper, a new gsa approach is developed for solving the p/s scheduling problem for the coming 24 hours. one of the advantages of the new approach is the use of stochastic operators rather than deterministic rules to obtain the global optimu m in order to escape from local optimums where other methods might land. a representative test example based on the actual taipower system is presented and analyzed to illustrate the capability of the proposed approach in practical applications . * corresponding author. e-m ail address: phchen@ m ail.sj u.edu.tw proceedings of engineering and technology innovation, vol. 5, 2017, pp. 19 24 copyright © taeti 20 2. problem formulation 2.1. list of symbols psi t p ower generation of thermal unit i in hour t t jv water volume of the up p er reservoir of p lant j at the ending of hour t ph j t p ower generating (p ositive) or p ump ing (negative) of p/s p lant j in hour t t ljv , water volume of the lower reservoir of p lant j at the ending of hour t f pi t si t ( ) p roduction cost for psi t i j t natural inflow into reservoir j in hour t t number of scheduling hours q j t water discharge of p/s p lant j in hour t nh number of p/s p lants t pjq , water p ump ing of p/s p lant j in hour t n s number of thermal units s j t water sp illage of p/s p lant j in hour t pl t sy stem load demand in hour t ursi up ramp rate limit of thermal unit i ploss t sy stem transmission network losses in hour t drsi down ramp rate limit of thermal unit i 2.2. modeling of a pumped-storage plant a p/s plant, which consists of an upper and a lower reservoir, is designed to save fuel costs by generating during peak load with water in the upper reservoir, which would be pumped up during light load hours . the equivalent plant model can b e obtained using an off-line mathematical procedure which maximizes the total plant generation output under different water discharge rates [2]. the generation output of an equivalent hydro plant is a function of the water discharge through the turbine and the net head (or the content of reservoir). the general form is expressed by : p f q vhj t j t j t   ( , ) 1 (1) the quadratic discharge-generation function to be used in this paper as a good approximation of the hydro plant generation characteristics, considering the net head effect, is given below: p q qhj t j t j t j t j t j t          1 1 1 2 (2) where coefficients  j t1 ,  j t1 , and  j t1 depend on the content of the upper reservoir at the ending of hour t-1. in this work, the read-in data include five groups of   , , coefficients that relate to different storage volumes, from minimum to maximum, for the upper reservoir. then, the corresponding coefficients for any reservoir volume are calculated by using a linear interpola tion [3] between the two closest volumes, as shown in the first quadrant of fig. 2. in pumping mode, since all p/s units of taipower are designed for constant power pumping, the characteristic function of a p/s plant is a discrete distribution as shown in the third quadrant of fig. 1. discharge (cubic meter per second) output (mw) p max vmin vmax input (mw) p umping (cubic meter per second) fig. 1 typical input-output characteristic for a p/s plant proceedings of engineering and technology innovation, vol. 5, 2017, pp. 19 24 copyright © taeti 21 2.3. objective function and constraints the scheduling of p/s units deals with the problem of obtaining the optimal generations both for p/s and thermal units. it aims to minimize the production costs of thermal units while satis fying various constraints. with discretization of the total scheduling time into a set of shorter time intervals (say, one hour as one time interval), the scheduling of p/s units can be mathematically formulated as a constrained nonlinear optimization prob lem as follows : problem: minimize f pi t si t i n t t s ( )   11 (3) subject to the following constraints: system power balance p p p psi t i n hj t j n l t loss ts h         1 1 0 (4) water dynamic balance t j t pj t j t j t j t j sqqivv   , 1 (5) t j t pj t j t lj t lj sqqvv   , 1 ,, (6) thermal generation and ramp rate limits max p p dr p min p p ursi si t si si t si si t si( , ) ( , )       1 1 (7) water discharge limits q q qj j t j  (8) water pumping limits pj t pjpj qqq ,,,  (9) reservoir limits j t jj vvv  (10) lj t ljlj vvv ,,,  (11) 3. glowworm swarm algorithm solution methodology 3.1. review of glowworm swarm algorithm the basic glowworm swarm algorithm (gsa) technique was firstly introduced by krishnanand and ghose in 2009 [5, 6]. the gsa is a swarm intelligence optimization technique based on the behavior of glowworms. the behavior of glowworms which is used for gsa is the apparent capability of the glowworms to change the intensity of the luciferin emission and thus appear to glow at different intensities [5]. in the gsa, glowworms can automatically subdivide into subgroups and thus can find multiple global solutions simultaneously, and thus gsa is very suitab le for multimodal problems. however, in gsa, there is no sufficient number or neighbors limit and there is no perception limit based on distance, but it can still have a cognitive limit which allows swarms of glowworms to split into sub-groups and converge to high function value points. this property of gsa allows it to be used to identify multiple peaks of a multi-modal function [6]. proceedings of engineering and technology innovation, vol. 5, 2017, pp. 19 24 copyright © taeti 22 3.2. solution methodology for ease of exposition, consider a p/s plant consisting of four units. the encoding scheme that translates the encoded parameter-water discharges of each plant into their binary representation is shown in fig. 2. using a plant's water discharge, instead of the plant's generation output, the encoded parameter is more beneficial for dealing with the difficult water balance constraints. each chromosome string contains 24 genes to represent the solution for the hourly discharge/pumping schedules of the p/s plant in a 24-hour period. each gene is assigned by the same number of five bits. the first bit is used to identify whether the plant is in generating or in pumping mode. the other four bits are used to represent a normalized water discharge q j t in generating mode, or to represent the number of pumping units in pumping mode. the resolution is equal to 1/24 of the discharge difference from minimum to maximum in generating mode. hour 1 2 ..... 24 1 0 0 0 1 0 0 0 1 1 ..... 1 0 0 1 0 fig. 2 the encoding scheme for a p/s plant with four units 3.3. decoding evaluation of a chromosome is acco mplished by decoding the encoded chromosome string and computing the chromosome's fitness value using the decoded parameter. the detailed decoding procedure is summarized in the following steps : (1) decode the first bit to identify whether the plant is in gene rating or pumping mode. (2) if in pumping mode, go to step 3, else in generating mode, then go to step 6. (3) decode the other four bits of the gene to calculate the number of pumping units and the total volumes of water pumping . (4) calculate the upper boundary of th e water pumping. (5) calculate the mw power for pumping . (6) decode the other four bits of the gene to obtain the normalized discharge in decimal values . (7) calculate the upper boundary of the discharge. (8) translate the normalized value to the actual value. (9) calculate the generation output using (2). (10) calculate the remaining thermal load. (11) continue the computation of the above 10 steps from hour 1 to hour 24. (12) do thermal unit commitment for the remaining thermal load profile, and return the corresponding thermal cost to the main program. (13) translate the corresponding thermal cost to the fitness function. (14) repeat the above 13 steps from the first chromosome to the last chromosome. 4. test results the proposed gsa approach was implemented in software and tested on a portion of the taipower generation system, which consists of 34 thermal units and the ming -hu p/s plant with four units. this software package was written in matlab and executed on a personal computer. proceedings of engineering and technology innovation, vol. 5, 2017, pp. 19 24 copyright © taeti 23 detailed characteristic data of the ming -hu p/s plant are given in table 1. the thermal system consists of 34 thermal units involving six large coal-fired units, eight small coal-fired units, seven oil-fired units, ten gas turbine units, and three combined cycle units. for data on the characteristics of the 34-unit thermal system please refer to [10]. the software is tested on a summer weekday . the load profile is obtained by subtracting the expected generation output of other hydro plants and nuclear units from the actual system load profile. the optimal schedules of both p /s units and thermal units are obtained within 5 minutes, well satisfied the taipower's requirement. test results are schematically shown in fig. 3. fig. 3 shows the total generating/pumping profile created by the proposed approach . table 1. characteristics of the ming-hu p/s plant. installed capacity maximal discharge (m3/s) maximal pumping (m3/s) lower reservoir efficiency maximal storage (k×m3) minimal storage (k×m3) 250mw×4 380 249 9,756 1,478 0.74 fig. 3 hourly mw generating/pumping schedules 5. conclusions this paper presents a new methodology based on a gsa for solving the p/s units scheduling problem in the daily hydro-thermal coordination. one of the advantages of the proposed approach is the flexibility of gsa for modeling various constraints. the difficult water dynamic balance constraints are embedded and satisfied throughout the proposed encoding and decoding algorithms. the effect of net head was also considered. numerical results from an actual utility system indicate the attractive properties of the gsa approach in practical application, which are a highly optimal solution and more robust convergence behavior. references [1] a. j. wood and b. f. wollenberg, power generation, operation, and con trol, 3rd ed., john wiley & sons, new york, 2013. [2] m. e. el-hawary, “an overview of scheduling functions in hydro -thermal electric power systems,” proc. the 1991 iasted international conference, pp. 1-37, march 1991. [3] s. ruzic, , et al., “a flexible approach to short-term hydro-thermal coordination,” ieee transactions on power systems, vol. 11, no. 3, pp. 1564-1578, august 1996. [4] p. h.chen, “two-level hierarchical approach to unit commitment using expert system and elite pso,” ieee transactions on power systems , vol. 27, no. 2, pp. 780-789, may 2012. proceedings of engineering and technology innovation, vol. 5, 2017, pp. 19 24 copyright © taeti 24 [5] k. n. krishnanand and d. ghose, “glowworm swarm optimization: a new method for optimizing mufti-model functions,” international journal of computational intelligence studies , vol. 1, no. 1, pp. 93-119, may 2009. [6] k. n. krishnanand and d. ghose, “glowworm swarm optimization for simultaneous capture of multiple local optima of multimodal functions,” swarm intelligence, vol. 3, no. 2, pp.87-124, june 2009. [7] d. n. jayakumar and p. venkatesh. “glowworm swarm optimization algorithm with topsis for solving multiple objective environmental economic dispatch problem,” applied soft computing, vol. 23, pp. 375-386, october 2014. [8] y. q. zhou, g. zhou, and j. l. zhang, “a hybrid glowworm swarm optimization algorithm to solve constrained multimodal functions optimization,” optimization, vol. 64, no. 4, pp. 1057-1080, 2015. [9] s. s. reddy and c. s. rathnam, “optimal power flow using glowworm swarm optimization ,” international journal of electrical power & energy systems , vol. 80, pp. 128-139, september 2016. [10] p. h. chen, “pumped-storage scheduling using evolutionary particle swarm optimization,” ieee transactions on energy conversion, vol. 23, no. 1, pp. 294-301, march 2008. nst 2007 extended summary proceedings of engineering and technology innov ation, vol. 8, 2018, pp. 40 45 applying project management perspective for erp implementation: a case study li-man liao * , ching-jen huang, xiang-yu lin department of industrial engineering and management, national chin-yi university of technology, taichung, taiwan. received 31 july 2017; received in revised form 17 november 2017; accept ed 24 november 2017 abstract the case-study company is a well-known enterprise whose primary product is precision locknuts in taiwan. with the rise of sales performance, the legacy enterprise resource planning (erp) system does not provide the degree of re liab ility and maintainability the case comp any needed. in order to meet the growing needs of enterprise, the case company faces to the challenge of application integration within the enterprise. the erp imp le mentation success depends on the rigor of the project management processes. in this researc h, firstly, case study approach is adopted to investigate how poor project manage ment imperil the imp le mentation of an erp system. hav ing learned the lessons from the failure of its first erp imple mentation, the case company applies the theory of business process reengineering (bpr) to e xp lore and rationalize the business processes, and design its project manage ment practices to successfully carry out its second erp imp le mentation. and the goal directed project manage ment (gdpm) princip les are applied to de velop a proper erp imple mentation plan. the summary descriptions of the important aspects of erp imple mentation are stated that the sponsor or the client of the project, the goals, and limitations on the project, and time a re included. th is study exp lores and identifies the critica l elements of project management to provide an effective guidance for the first implementation stage . keywor ds: erp, project management, bpr 1. introduction the ma rket trend of industrial nuts and bolts tends to positive growth, the value of industrial nuts and bolts within taiwan has achieved almost 731 billion nt dolla rs in 2016. the case company is a well-known enterprise and specializing in production of precision locknut in taiwan. diffe rentiated capability becomes a basis for survival within a globally competitive ma rketplace where various capabilities compete intensely. recently, it has devoted to the ‘small-volu me production of a wide range of diffe rent products’ transformation, and confronts the challenge of information integration to carry out self-require ment within the enterprise. a suitable enterprise resource planning (erp) system can integrate the processes of internal organization and provide man y benefits, including imp roving productivity, enhancing competitive advantage and satisfying customer de mands. despite the advantages of erp system, failure cases of erp imp le mentation occur frequently. therefore, it is crit ical how to e xecute and manage t he erp imple mentation to avoid the lack of oversight on any mistakes of systems implementation stage lengthy and over budget . the ma rket trend of industrial nuts and bolts tends to positive growth, the value of industrial nuts and bolts within taiwan has achieved almost 731 billion nt dolla rs in 2016. the case company is a well-known enterprise and specializing in production of precision locknut in taiwan. diffe rentiated capability becomes a basis for survival within a globally * corresponding author. e-mail address: t10066@cc.kyu.edu.tw tel.: +886-7-6077138; fax: +886-7-6077959 proceedings of engineering and technology innovation, vol. 8, 2018, pp. 40 45 copyright © taeti 41 competitive ma rketplace where various capabilities compete intensely. recently, it has devoted to the ‘small -volu me production of a wide range of diffe rent products’ transformation, and confronts the challenge of information integration to carry out self-require ment within the enterprise. a suitable enterprise resource planning (erp) system can integrate the processes of internal organization and provide man y benefits, including imp roving productivity, enhancing competitive advantage and satisfying customer de mands. despite the advantages of erp system, failure cases of erp imp le mentation occur frequently. therefore, it is crit ical how to e xecute and manage t he erp imple mentation to avoid the lack of oversight on any mistakes of systems implementation stage lengthy and over budget . the erp imp le mentation inc ludes are three phases which are evaluation and selection, installation and transformation of a new erp system, and inspection and adjustment. this study focuses on the first phase: evaluation and selection and refers to the goal directed project management (gdpm) relates to many principles to develop the erp project p lan [7]. there are 11 steps are developed in the first phase. in order to achieve the company’s e xpectation what the project plan is completed in may 2017. firstly, the company constructs a project team cross -functional cooperation, reviews the current processes to aggregate and finds out the critical bottleneck. secondly, a gantt chart is developed which is project schedule for the estimate t ime of e xecution of each activity. in the meantime, the pro ject tea m co llects in formation of erp system vendors, contacts with erp system vendors, provide a rfp (request for proposal) of erp system evaluation as refe rence for those candidate venders. and the candidate venders present their system function, technique supporting capability, and competitive advantage. finally, a satisfaction questionnaires surv ey is made by me mbe rs of the project team and crit ical participants, and obtain the quotation of an information system from the first priority candidate of the erp system vender. 2. system architecture and principles 2.1. enterprise resource planning (erp) erp is an exce llent business application in modern business operations which is designed to coordinate information system and integrate business processes [9]. despite the advantages of erp system, the failure rate of erp imp le mentation is still high [2, 10]. nu merous authors have identified the critica l factors of a successful erp imp le mentation to avoid company trapping into a loss control failure situation. a structured and disciplined approach is applied to greatly fac ilitat e the erp imp le mentation. th e most pro minent factors include a clea r understanding of strategic goals, e xcellent imple mentation project management, a great imple mentation tea m, top manage ment support, e xtensive education and training as we ll as organizational change manage ment, etc. sa ini et al. [10] proposed their hypothetical success factors model which was validated fro m the data collected by questionnaire. they a lso found the business culture to impact the erp imple mentation in organizations. umb le et al.[11] co mp iled a list of 11 impo rtant steps for successful imple mentation fro m four e xisted works, and described the critica l factors for successful erp imp le mentation and system selection guidelines in a company. 2.2. project management of erp system ple x, manage ment plays an important role in organizations and has a major impact. successful erp imple mentation requires an exce llent project manage ment while enterprise must have an effective business strategy and tactic. the ma in phases of the project life cycle are varied depending upon the industry involved. it is important to ensure that the project life cycle used on the project is appropriate and the detail with in each of these distinct manageable phases. a project manage ment should have implic it trade-off a mong time , budget, and function. gray and la rson [3] described the common phases of the whole length on project life wh ich are defin ition, p lanning, e xecution and delivery phases and shown as fig. 1. so fa r the case co mpany has been imp le menting for ten months. the erp definit ion phase has done which y ields the business processes to imple ment the actual erp system, and the p lanning phase has started since six months ago and the works have been progressing. proceedings of engineering and technology innovation, vol. 8, 2018, pp. 40 45 copyright © taeti 42 the project governance requires a set of “check and balances” to ensure that the right decisions were made [6]. metaxiotis et al. [7] presented an overview of goal directed project management (gdpm) which aims at the facilitation and effective guidance fo r the erp systems installation process. the gdpm princip les indicate the three parts of p lanning, organization as well as control, wh ich support the imple mentation and optima l adaption (ioa) of erp system. they also proposed five steps in the life cycle of erp project to offer a progressive and smooth passage by erp system. che n et al. [2] e xplored and identified crit ical e le ments of project manage ment, which connected with “what” and “how” aspects of the erp project failure, and adopted or avoided them for the second erp implementation. according to the technique report of cisco systems, inc. [1], an erp system must meet business needs that would require heavy involvement fro m the business community. it is critica l that select the very best people to build the imple mentation team who includes enterprise’s key-men, vendor of erp software and consultants. the imp le mentation team is constructed and provides inter-departmental communicat ion that can bring the centripetal force of the entire enterprise. it is responsible for creating the initia l, detailed project p lan or overa ll sched ule for the entire project, assigning responsibilities for various activities and determining due dates. saini et a l. [10] described that consultant selection could share knowledge and e xperience for the company. umb le et al. [11] refer to the request for proposal (rfp) which contains the feature and function list, create an initial of a ll imp le mentation schedule for the project. the planning and execution stages consume most of time of erp imple mentation, therefore enhancing the development of pro ject sche dule is crit ical. the stage provides a forecasted duration for the project activities to avoid high of the overall project budget and lengthy execution time for project stakeholders . execution definition planning delivery time start finish t h e d e g r e e o f e ff o r t now fig. 1 project life cycle [3] planning stagedefinition stage 1.define the project strategic goal 2.project team composition 3.review the current processes 6. execute presentation of erp system vendors 7.write a request for proposal (rfp) 8.interview the candidate vendors 5.collect information of erp system vendors 9.develop a satisfaction questionnaire 4.develop a project schedule 10.hold a formal presentation 11.quote and contract signing fig. 2 the erp system introducing evaluation framework proceedings of engineering and technology innovation, vol. 8, 2018, pp. 40 45 copyright © taeti 43 2.3. business process re-engineering (bpr) as the developing trend for modern business moves toward a collaborative mode l, the o wn business practices and procedures should be improved to re ma in its competit ive. business processes are actions that firm involves in accomp lishing each business task to achieve business purpose. many erp imp le mentation fa ilures have been due to the lack of focus on the business process and change manage ment. managers must describe the current business processes and rationalize those processes to develop a new best practices model fro m the erp application. business process re -engineering (bpr) is closely tied to erp imp le mentation, a la rge a moun t of re-engineering should take place iterative ly and usually has the significant challenge of erp systems. ke mp and lo w [4] focused on organizat ional behavior which imp le mented erp system with re engineering of the re lated business processes incorporating change manage ment. the bpr inc ludes the change manage ment, organization culture as we ll as operation processes to re -planning and carry out. umble et a l. [11] mentioned that organizational change manage ment is one of the critica l factors for successful erp imple mentation. for sustaining the effectiveness of the reengineering efforts, redesigned processes must correspond with realignment in organizational control. organizational culture is defined as the deeply rooted values, beliefs and assumption that ar e shared by personnel in an organization. it involves all part icipants of each department and the process of the whole enterprise. a business with e xcellent culture could sustain the long -term co mpetitive advantage. an erp system has been credited with rescuing the idea of bpr which focused to redefine and redesign the workflows to fit the new software [9]. clearly, this is a good chance for enterprises to imple ment bpr as the new erp system is introduced. in order to obtain an e xce llent bpr, the organizati on must have the objective of regulating the whole fra mewo rk, e xecutor power fo r decision ma king, commit ment by top management, and the great channel of communication, etc . 3. developing the project planning a project is undertaken for a specific purpose with tangible value in a g iven duration. th is section main ly describes how to imple ment project manage ment of erp system for the case co mpany. the erp imple mentation is d ivided into three phases: evaluation and selection, setup and transformation of a new erp system, and check phases. the first phase has two stages which are definition and planning and shown as fig. 2. this case study describes the first erp upgrade project at the dead sea works (dsw) chemical company. the case company has an imp le mented e xperience fro m the e xisting erp. a project tea m consists of top managers and key me mbers wh ich is responsible for creat ing the initia l, detailed project planning and making all necessary demands will be available on the second erp system. the case company has a consultancy, corporate synergy develop ment center, which prov ides essential assistance to identify the important e le ments of the erp pro ject. therefore, this study proposes the erp imple mentation project p lan via literatures review and the consultant’s suggestions. because the top managers have a great vision and determination to overcome all the obstacles to erp imp le mentation; moreover, almost e mployees are young and have a strong will to participate th e changeover of the new erp system. fig. 3 the gantt chart of evaluation and selection phase at erp project 91 60 91 15 35 4 17 60 2016/5/1 2016/7/2 2016/9/2 2016/11/3 2017/1/4 2017/3/7 2017/5/8 review the current process collect information of erp system vendors execute presentation of erp system vendors write a request for proposal (rfp) interview the candidate vendors develop a satisfaction questionnaire hold a formal presentation quote and contract signing proceedings of engineering and technology innovation, vol. 8, 2018, pp. 40 45 copyright © taeti 44 according to the goal directed project management (gdpm ) principles by meta xiotis et al. [7], the sponsor must have a description of the critica l aspects of the it pro ject. in may 2016, the information depart ment started up the erp project p la n and named it “the pro ject manage ment of case co mpany’s erp system rep lace ment assessment”. the d irector of informat ion department is the sponsor who forms the erp project tea m. firstly, the current processes are reviewed to find out the key process bottlenecks, and rationalize the internal processes of each department and eliminate the obstacles of across department boundaries. the pro ject must be constructed with t ime schedule and all the activit ies have to identify and be performed. therefore, the project tea m develops a project schedule to estimate the input resources for the e xecution of each activity. the every activity is recorded into calendar to ensure that the ultimate goals of the project are achieved. secondly, the director of informat ion depart ment contacts with erp software vendors and collects data relat ive to erp imple mentation, and the qualified vendors are sifted out. each q ualified vendor must present his own system functions, available resources and previous successful cases. third ly, a request for proposal (rfp) of the case company is proposed. the rfp contains the purpose of the project planning, current operational proce sses, and demands of each department for the erp system. the contents make the staffs of the case company and the venders understand the project goal and imp le ment procedures. after the vendors’ presentation takes place, the project team and partic ipants evaluate the fitness of erp software v ia questionnaire survey as a preliminary assessment for the vendors. finally, the pro ject tea m determines the first priority vendor who provides the quotation and signs the contract. the contract had done in may 2017. the objective of this study is the assistance that progressive and smooth passage of the enterprise is provided by the new erp environment . 4. conclusion this study ma inly identified the c rit ical ele ments and procedures of evaluation and selection stage of er p project imple mentation. owing to learn ing the lesson of the first erp system, the case company pays attention to the new erp imple mentation. the overall internal business processes and reporting sheets are reviewed and analyzed, and then are rationalized and aggregated via business process reengineering (bpr) approach. the fu zzy ahp approach is applied to evaluate three erp systems, and vikor method is developed to determine the best -fit erp system for the case company. the executive steering committee (esc) tea m co mprises the top executives, the leader of each functional department, staffs of information department of the case company, and the consultant. the esc tea m focuses on the processes of each department to find out major bottlenecks, and integrate s all demands as a reference of the new erp system. the erp system is designed to integrate all information processing, so the case company imp le ment s business process reengineering (bpr) to identify the suitable operations for each department by erp imp le mentation. besides, this study constructs an imp le mentation fra me work of the erp project, and develop a project schedule as a direction of the evaluation and selection stage of project manage ment. although the project imple mentation ever is id ling about tw o months because of some proble ms cannot proceed immed iately. fortunately, the case co mpany found a solution to overcome the proble ms. the case company has contracted with the first priority erp vendor in may 2017, and the project proceeded to the next pha se. acknowledgment thi sresearch is funded by the ministry of science and technology of the republic of ch ina under grants most104 2622-e-167-018-cc3. and special thanks to all who have helped to make this study, especially a ll the people of case company. references [1] r. d. austin, l. n. richard, and j. c. mark, cisco systems, inc.: implementing erp, harvard business review, pp.118, september 1999. [2] c. c. chen, c. c. h. law, and s. c. yang, “managing erp implementation failure: a project management perspective,” transactions on engineering management, vol. 56, no. 1, pp. 157-170, february 2009. proceedings of engineering and technology innovation, vol. 8, 2018, pp. 40 45 copyright © taeti 45 [3] c. f. gray, e. w. larson, project management: the management process, 4th ed. new york: mcgraw-hill, 2008. [4] m. j. kemp and g. g. low, “erp innovation implementation model incorporating change management,” journal of the business process management, vol. 14, no. 2, pp. 228-242, 2008. [5] l. m. liao, c. j. huang, and p. c. kang, “applying fuzzy-ahp and vikor method to evaluate the fitness of erp systems,” proc. the 2016 international conf. industrial engineering and operations management (ieom2016), jw marriott hotel, kuala lumpur, malaysia, march, 2016. [6] j. t. marchewka, information technology project management: providing measurable organizational value, 5th ed. new york: wiley, 2015. [7] k. metaxiotis, i. zafeiropoulos, k. nikolinakou, and j. psarras, “goal directed project management methodology for the support of erp imp le mentation and optima l adaptation procedure,” information management & co mputer security, vol. 13, no. 1, pp. 55-71, 2005. [8] ministry of economic affairs (moea), industry technology commentary: discussion about upgrading of nuts and bolts fastener industry, 2016. [9] d. l. olson, managerial issues of enterprise resource planning systems, new york: mcgraw-hill, 2004. [10] s. saini, s. nigam, and s. c. misra, “success factors for implementing erp in smes in india: a conceptual model,” international conference on information management and engineering (icime), ieee press, june 2010. [11] e. j. umble, r. r. haft, and m. m. umble, “enterprise resource planning: implementation procedures and critical success factors,” european journal of operational research, vol. 146, no. 2, pp. 241-257, april 2003.  proceedings of engineering and technology innovation, vol. 9, 2018, pp. 32 37 long pinions for alternative transmission mechanizations madhusudan raghavan * propulsion systems research lab, gm r&d center, pontiac, michigan, usa received 27 february 2018; received in revised form 17 may 2018; accepted 19 july 2018 abstract multi-speed transmissions are used in the automotive industry. the design of such transmissions may be framed as an algebraic/graph theory problem. such an approach is particularly useful for designs with multiple planetary gear sets. some automotive manufacturers use “long pinions” or “shared pinions” as building block elements within these gear sets. the long pinion is an interesting architectural arrangement that combines certain packaging and functional attributes. in the present article, the algebraic use of the long pinion is demonstrated in the creation of vastly different transmission architectures from the same skeleton diagram. keywords: automotive transmissions, shared pinion, long pinion 1. introduction the 2016 international energy outlook [1], offers the following predictions. global population grows 23% from 7.3 billion in 2015 to 9.0 billion in 2040. total world marketed energy grows by 42% from 575 quads in 2015 to 815 quads in 2040 (quad = quadrillion btu). fossil fuels still supply over 75% of world energy in 2040. petroleum remains the world’s largest energy source and worldwide energy-related co2 emissions rise from 34 billion metric tons per year in 2015 to 43 billion metric tons per year in 2040, a 29% increase. against this backdrop of increased population, urbanization, and increased energy usage and emissions, we continue to explore novel automotive propulsion architectures with improved efficiency. the problem of designing ultra-compact transmissions is a valuable aspect of this research as it frees up room for the insertion of electrification components within the packaging envelopes of existing propulsion systems. a vehicle transmission delivers mechanical power from the engine to the remainder of the drive system, such as fixed final drive gearing, axles and wheels. the mechanical transmission enables some freedom in engine operation, usually through alternate selection of multiple drive ratios, a neutral selection that allows the engine to operate accessories with the vehicle stationary, and clutches and a torque converter for smooth transitions between driving ratios and to start the vehicle from rest with the engine turning. an electrically variable transmission (evt) is a mechanical transmission augmented by one or more electric motor/generators. an evt uses differential gearing to send a fraction of its transmitted power through an electric path to the final drive. the remainder of its power flows through another, parallel path that is all mechanical and direct, of fixed ratio, or alternatively selectable. 2. background and prior work the theory of multi-speed transmission kinematic operation has been traditionally handled via lever diagrams [2]. these are convenient graphical representations of transmission gear ratios in terms of the relative dimensions of ring, sun, and planetary gear sets. since the automative industry is generally moving in the direction of larger numbers of fixed speed ratios * corresponding author. e-mail address: madhu.raghavan@gm.com tel.: +1-248-930-5248 proceedings of engineering and technology innovation, vol. 9, 2018, pp. 32 37 copyright © taeti 33 as well as hybrids, we briefly review recent work on multi-speed transmissions and evts. lewis and bollwahn [3] present the general motors hydra-matic/ford six-speed fwd automatic transmission family. designed-in modularity requires only changes to the second and third axis and case housings to achieve various torque requirements as stipulated by the specific vehicle application. singh and olenzek [4], discuss variants of general motors’ family of small front wheel drive six speed automatic transmissions including the 6t40 and 6t45, which cover a range of vehicles from small and compact cars to small suvs and handle engines torque capacities up to 240 n-m gas (280 n-m diesel) & 315 n-m gas (380 n-m diesel), respectively. clark et al., [5], describe general motors’ front wheel drive seven speed dry dual clutch automatic transmission introduced in 2014. the 250 n-m input torque rated gear box was designed and engineered for a global market in both front wheel drive and all-wheel drive configurations. raghavan [6], introduces the eight speed rear wheel drive transmission family recently launched by general motors. this set of designs yields a 7.0 overall ratio spread, enabling improved launch capability because of a deeper first gear ratio and better fuel economy due to lower top gear n/v capability. grewe et al., [7], describe the gm two-mode hybrid transmission for full-size, full-utility suvs. this system integrates two electromechanical power-split operating modes with four fixed gear ratios and provides fuel savings from electric assist, regenerative braking and low-speed electric vehicle operation. miller et al., [8] describe the voltec 4et50 multi-mode electric transaxle, which introduces a unique two-motor ev driving mode that allows both the driving motor and the generator to simultaneously provide tractive effort while reducing electric motor speeds and the total associated electric motor losses. in the present paper, we show how such planetary gear set based designs could potentially be made more compact via the use of long pinions. this work is a natural extension of the results first presented by raghavan [9]. 3. lever diagrams and planetary geartrains the primary building block of a planetary gear train is the planetary gear set shown in fig. 1. it is comprised of a sun gear, a ring gear, and a set of planet gears (also known as pinion gears). fig. 1(a) shows a “simple” planetary gear set and fig. 1(b) shows a “double-pinion” planetary gear set. the planet gears are carried on a carrier member. the entire system of fig. 1(a) or 1(b) may be represented by an edge-vertex graph (a vertical straight line segment with 3 nodes on it) with the nodes labeled as ring r, carrier c, and sun s, respectively (see fig. 1(c)). such a graph is also known as a “lever,” because the relative rotational speeds of ring, carrier, and sun may be computed by treating rotational speeds as forces acting on the lever, and taking moments about appropriate nodes on the lever [2]. fig. 1 simple planetary and double pinion planetary gear sets an alternative to this geometric representation of levers is the following algebraic representation. let ,x ,y and z be the speeds of the ring, the carrier, and the sun in fig. 1(a). then these speeds are related by the following equation: proceedings of engineering and technology innovation, vol. 9, 2018, pp. 32 37 copyright © taeti 34 x a y a z 1 ) 1 1(  (1) this relationship holds true by virtue of the mechanical interconnections and gear interactions in the planetary gear set (see shigley and uicker [10]). eq. (1) contains a parameter a , which is equal to the ratio r s n n , sn and rn being the numbers of teeth on the sun gear and the ring gear, respectively. for double-pinion planetary gear sets (see fig. 1(b)) the governing equation is: 𝑧 = (1 − 1 𝑎 ) 𝑦 + 1 𝑎 𝑥. 4. long pinion arrangements given 2 adjacent planetary gear sets, we say that they use a “long pinion” if they share a common carrier and an integral pinion member. we can write expressions for the long-pinion arrangement as follows. case 1: for a combination of a simple planetary gear set (henceforth abbreviated as pg) connected to another simple pg via a long-pinion carrier, the constraint equations are: 11111 )1( ykxkz  , (first planetary) (2) 22222 )1( ykxkz  , (second planetary) (3) 12 yy  , (common carrier) (4) where 𝑘𝑖 = 1 𝑎𝑖 , i=1,2. additionally, the pinion speed of the first pg, ω1 = (𝑥1−𝑦1)2𝑘1 (𝑘1−1) , is set equal to the pinion speed of the second pg, ω2 = (𝑥2−𝑦2)2𝑘2 (𝑘2−1) , because the pinion is shared between the 2 pgs. the reader is referred to the work of shigley and uicker [10] for more details on how to derive the kinematic equations relating the speeds of the ring, carrier, sun, and pinion gears. )1( 2)( )1( 2)( 2 222 1 111      k kyx k kyx . (common pinion) (5) by using eqs. (4) and (5) to eliminate the variables 2x and 2y from eq. (3) in terms of 1x and 1y , we get the following equation 1 1 22 1 1 22 2 1 y k x k z                       . (second planetary) (6) where  ii a 1 , .2,1i eqs. (2) and (6) serve as constraints relating the rotational speeds of the 4 independent nodes },,,{ 2111 zzyx of the 2-planetary system comprised of pg1 and pg2 together with the common carrier and pinion. case 2(a): pg1 is simple and pg2 is of the double-pinion type 11111 )1( ykxkz  , (first planetary) (7) 22222 )1( ykxkz  , (second planetary) (8) 12 yy  . (common carrier) (9) the pinion speed of the first pg is to be set equal to the pinion speed of the second pg. however, the second pg has a double pinion arrangement and so we have a choice of either pinion 1 or pinion 2 (see fig. 1(b)). if we use pinion 1, its speed proceedings of engineering and technology innovation, vol. 9, 2018, pp. 32 37 copyright © taeti 35 in terms of other parameters of pg2 is 𝑅2 𝑃2 (𝑥2 − 𝑦2) where 2p and 2r are respectively, the pinion diameter and the ring gear diameter on pg2. we can see that for pinion 1, 𝑃 = 𝜉 𝑅−𝑆 2 , where ξ < 1. by using this result, the speed of pinion 1 may be expressed as (𝑥2−𝑦2) 𝜉 2 (1−𝑎2) . setting this equal to the speed of one of the pinions on pg1, we get    )1( 2)( 1 111 k kyx    2 22 1 a yx    (10) upon rearrangement of terms in eq. (10), we get 1 1 2 1 1 2 2 1 yxx                         (11) substituting this expression for 2x in eq. (8), and using eq. (9) to eliminate 2y , and then, re-arranging terms, we get 1 1 2 21 1 2 22 1 ykxkz                         (12) eqs. (7) and (12) serve as constraints relating the rotational speeds of the 4 independent nodes },,,{ 2111 zzyx of the 2-planetary system comprised of pg1 and pg2 together with the common carrier and common pinion fixed interconnection between the simple planetary gear set pinion and pinion 1 on the double pinion planetary gear set. similar expressions can be worked out for the cases listed table 1 . table 1 gearset combinations 2(b) pg1 is simple; pg2 is double pinion pinion-pinion2 3(a) pg1 is double pinion; pg2 is simple pinion 1-pinion 3(b) pg1 is double pinion; pg2 is simple pinion 2-pinion 4(a) pg1 is double pinion; pg2 is double pinion pinion 1-pinion 1 4(b) pg1 is double pinion; pg2 is double pinion pinion 1-pinion 2 4(c) pg1 is double pinion; pg2 is double pinion pinion 2-pinion 1 4(d) pg1 is double pinion; pg2 is double pinion pinion 2-pinion 2 5. long pinion transformations example let’s take a look at a 3 planetary gear set design with 2 brakes and 3 clutches yielding 6 forward speeds and 1 reverse speed as shown in fig. 2 along with the clutching table. in the following we use a long pinion transformation to create a more compact alternative to the design of fig. 2. if we focus on just the first 2 planetary gearsets with their fixed interconnections, the sub-system is as shown in fig. 3(a). the associated equations are: 11111 )1( ykxkz  , (first planetary) (13) 22222 )1( ykxkz  , (second planetary) (14) 21 yx  , (fixed interconnection 1) (15) 21 xy  . (fixed interconnection 2) (16) substituting for 2y from eq. (15) and for 2x from eq. (16) into eq. (14), we get      12122 111 ykxkz  (17) proceedings of engineering and technology innovation, vol. 9, 2018, pp. 32 37 copyright © taeti 36 eqs. (13) and (17) taken together represent the physical system shown in fig. 3(b), featuring a simple planetary gear set coupled to a double pinion planetary gear set via fixed interconnections, ring to ring, and carrier to carrier. going a step further we may compare eq. (17) with eq. (12) which represents the governing equation of a double pinion planetary gear set which shares a 1p pinion and carrier with an adjacent simple planetary gear set. equating corresponding coefficients, we get lp lp lplporiginal kk 1 2 221    (18) where the subscripts original and lp apply respectively to the original gear sets of fig 3(a) and the long pinion gear sets that we wish to transform them into. using the numerical values for the 𝑅𝑖𝑛𝑔 𝑆𝑢𝑛 ratios from fig. 3(a), we set 9.0,37.2,57.1 21  lporiginaloriginal kk  . further, setting originallp kk 11  , we may compute lp k2 from eq. (18). its value is 2.35. by this process, the arrangement in fig. 3(b) with ring-ring and carrier-carrier fixed interconnections, transforms into the long pinion arrangement of fig. 3(c), with ( 𝑅𝑖𝑛𝑔 𝑆𝑢𝑛 ) 1 =1.57, ( 𝑅𝑖𝑛𝑔 𝑆𝑢𝑛 ) 2 =2.35. fig. 2 three planetary six speed transmission (a) original gearsets (b) {ring-ring, carrier-carrier} (c) long pinion fig. 3 transformation to long pinion proceedings of engineering and technology innovation, vol. 9, 2018, pp. 32 37 copyright © taeti 37 kinematically, the 3 arrangements of fig. 3 are equivalent. therefore, we may substitute the arrangement of fig. 3(c) in place of the first 2 planetary gearsets in fig. 2 to get fig. 4. note that the speed ratios and clutch sequences are identical for figs. 2 and 4. the arrangement of fig. 4 can be implemented with a shorter axial length relative to fig. 2 due to the integration of the first 2 pgs via the long pinion transformation. fig. 4 six speed transmission with long pinion 6. conclusions in the preceding sections, we have demonstrated the algebraic use of long pinion arrangements in the creation of alternative architectures for automotive transmissions. once a powerflow with favorable characteristics is established, there are many ways in which it can be mechanized. we have shown the use of long pinion arrangements to replace (ring-carrier, carrier-ring) connections in adjacent planetary gear sets, while retaining functional equivalence. various other permutations and combinations of fixed interconnections may be worked out to achieve ultra-compact designs. references [1] “international energy outlook 2016,” https://www.eia.gov/outlooks/ieo/pdf/0484(2016).pdf, may, 2016. [2] h. benford and m. leising, “the lever analogy: a new tool in transmission analysis,” society of automotive engineers technical paper, february 01, 1981. [3] c. lewis and b. bollwahn, “general motors hydra-matic & ford new fwd six-speed automatic transmission family,” society of automotive engineers technical paper, april 16, 2007. [4] t. singh and r. olenzek, “general motors small front wheel drive six speed automatic transmission family,” society of automotive engineers technical paper, april 12, 2010. [5] k. clark, t. singh, r. buffa, j. gayney, et al., “general motors front wheel drive seven speed dry dual clutch automatic transmission,” sae international journal of engines, vol. 8, vol. 3, pp. 1379-1390, april 2015. [6] m. raghavan “synthesis of transmissions with four planetary gearsets,” proc. of the 14th iftomm world congress, october 2010. [7] t. grewe, b. conlon, and a. holmes, “defining the general motors 2-mode hybrid transmission,” society of automotive engineers technical paper, april 16, 2007. [8] m. miller, a. holmes, b. conlon, and p. savagian, “the gm voltec 4et50 multi-mode electric transaxle,” sae international journal of engines, vol. 4, no. 1, pp. 1102-1114, 2011. [9] v. kumar, j. schmiedeler, s. sreenivasan, hj su, advances in mechanisms, robotics and design education and research, vol. 14, heidelberg: springer, 2013. [10] j. e. shigley and j. j. uicker, theory of machines and mechanisms, new york: mcgraw-hill, 1980. https://www.eia.gov/outlooks/ieo/pdf/0484(2016).pdf proceedings of engineering and technology innovation, vol. 7, 2017, pp. 25 30 reversible data hiding over facebook ran lyu * , jian-tao zhou, wei-wei sun, li dong department of computer and information science, university of macau, macau, china. received 08 july 2017; received in revised form 15 august 2017; accept ed 28 august 2017 abstract facebook, the most popular online social network (osn), could be used as a platform to share secret messages through jpeg images online. however, due to the various lossy operations conducted over facebook, the data embedded into the jpeg images can be easily destroyed, making the data extraction infeasible. more importantly, all these operations are carried out without users’ interference. in this paper, we first perform an in -depth investigation of the various lossy operations that facebook applie s to uploaded images. based upon such prior knowledge, we propose a dct-domain data hiding scheme that can effectively embed a large amount of data and successfully extract them out from the downloaded images, defeating the uncontrolled lossy operations. compared with the state-of-the-art techniques, the proposed method offers much higher embedding capacity, and can extract the data successfully with very high probability. furthermore, the restored image upon data extraction is of high quality, and the file size expansion is negligible. extensive experimental results are provided to validate our findings. keywords: data hiding, jpeg, reversible data hiding, facebook 1. introduction facebook, the most popular online social network (osn), could be used as a pla tform to share secret messages through jpeg images online. the operations conducted by facebook (including resizing, jpeg compression, etc) are much more complex than other social networks [1]. although the idea of sharing the hidden message over facebook is attractive, it faces many inevitable challenges. huang [2] proposed a very robust steganography for pure jpeg. however, huang's method is not robust to overflowing caused by re-compression on facebook. hiney [3] tested nine steganography methods on face book. the experimental results have shown that these methods work well offline, but perform poorly on facebook. castiglione analyzed how some prevalent osns processed the uploaded images [1] and proposed two steganography schemes [4] with low embedding capacity on facebook. nagaraja [5] proposed stegobot, using the yass [6] specifically, which has the limitation of the embedding capacity. ning [7] proposed an approach 2-lsb inspired by lsb [8]. the drawbacks are two-fold: lower embedding capacity compared to lsb; lacks of discussion about visual quality. amsden [9] proposed to transmit hidden information using jphide and jpseek [10] over a cover photo on facebook (failed over album images). in this work, we first perform an in-depth investigation of the various lossy operations that facebook applies to uploaded images. based upon such prior knowledge, we propose a dct-domain data hiding scheme. compared with the state-of-the-art techniques, the proposed method offers much higher embedding capacity, and can e xtract the data successfully with very high probability. the rest of the paper is organized as follows. the investigation of facebook is given in section 2. the proposed scheme and system are described in section 3. experimental results are shown in section 4. finally, the conclusion is given in the section 5. * corresponding author. e-m ail address: m b45442@ um ac.m o proceedings of engineering and technology innovation, vol. 7, 2017, pp. 25 30 copyright © taeti 26 2. analysis of the operations conducted over facebook we upload 1338 gray uncompressed images from ucid-v2 [11] to facebook, and then download them. the published images are compressed using quality factor (qf) chosen by facebook. the quantization tables are extracted from the downloaded images. it is observed that all the quantization tables match the international jpeg group standard [1 2]. it turns out that the employed qf varies from 71 to 92. we then make a hypothesis for the re-compressed image of facebook: the qf chosen by facebook should be smaller than or equal to the one in uploaded image. to verify this assumption, we compress images with qfs vary from 50 to 100 and upload them to facebook. then we download them and extract the quantization tables from the downloaded images. the statistical results agree with our assumption. we have an important observation: for jpeg images compressed by qf≤71, the qfs of downloaded image are all consistent with 71. 3. facebook message sharing system (fmss) to clearly describe the facebook message sharing system (fmss), the related terms are defined: cover image, the original image; stego image, the embedded image; fbstego image, the published stego image over facebook; restored image, the image restored from the fbstego image after extraction ; qfembed, the qf used in stego image offline; qffb, the qf used in the fbstego image on facebook. the following roles are involved in the fmss: sender, facebook and receiver. sender compresses the cover image with qfembed=71 before embedding which forces facebook chooses qfembed=qffb in re-compression. in fmss, we fully trust the sender and receiver that they can act exactly as we wish. facebook might damage the data hiding scheme accidentally. however, facebook will follow a fixed process during re-compression. therefore, we take the facebook as semi-trusted. 3.1. block selection algorithm even we make qfembed=qffb, the inverse and forward discrete cosine transform (idct and fdct) in re-compression of facebook s till has a chance to cause the failure of data hiding. given i’ as the embedded image before the pixel operations in jpeg. the pixel operations include truncation and rounding, which cause the overflowing err: i’-round(truncate(i’)). since overflowing phenomenon is inevitable in jpeg algorithm, we check the embedded block whether it's overflow with large err value or not. not all blocks in an image have large err of overflowing phenomenon and we call them stable blocks re ferring stable qf. the new selection method of stable blocks needs a map for each image to identity positions. it's need to reduce the additional storage introduced by map. we call embeddable blocks qualified block (qb) and others unqualified block (ub). a flag bit bf{1,0} needs to be inserted to determine the qbs and ubs . in this solution, the burden is only related to the number of embedded blocks, but not the size of the image. to achieve higher capacity in the embedding process, we adopt the same strategy proposed by huang [2]: embedding bits in block with more zero coefficients. 3.2. embedding algorithm by spanning qb in zigzag, we build a qualified embedding coefficients vector q and shifting coefficients vector s with size of n and m, respectively. each secret message bit can be represented as b∈{1,0}. the proposed embedding algorithm for qb is as eq. (1)-(3): qi’=qi+sign(qi)·b i=1,…,n-1 qi∈{ ac coefficients of each qb and | qi |=1} (1) sj’=sj+sign(sj) j=1,…,m sj∈{ ac coefficients of each qb and | sj |=2} (2) qi’=qi+sign(qi)·bf i=n qi∈{ ac coefficients of each qb and | qi |=1} (3) proceedings of engineering and technology innovation, vol. 7, 2017, pp. 25 30 copyright © taeti 27 the flag bit bf=1 is embedded into the last qi by eq. (3). to make sure of efficiency, there should be at least one valid message bit and the flag bit in a qb, so the length of q should be greater than 1, i.e., n≤2. the blocks with low capacity and overflowing phenomenon can be set as ubs. by spanning each ub in zigzag sequence, the first coefficient with the value of ±1 or ±2 counting backwards is u. for ub, we only need to embed the flag bit 0 by : u'=sign(u). if there's no u in current ub, we can just do nothing and move to next block. sender scans the blocks to check they're qbs or ubs by proposing block selection. then message bits and flag bits are embedded following the embedding algorithm ab ove. not all blocks need to be scanned while embedding since the embedding process might finish before scanning all blocks. a reasonable choice for visual quality is to s top embedding scanning and keep the following blocks unchanged. 3.3. data extraction and image restoration algorithm given the secret message m with the length of l. l can be represented by l bits. here l is a fixed and predefined number. therefore, sender embeds l and m with l+l bits totally. in the experiment, we set l=16. the receiver downloads the fbstego image from facebook and extracts the secret message bits from it. the receiver scans the blocks to check if they're qbs or ubs by extracting the flag bits. the receiver can calculate the length of the embedding bits l by first l extracted bits. specifically, in message extraction part, the vector qi * is built in each block and the restored ac coefficients qk * ' can be calculated by eq. (4). qk * '=sign(qk * ) k =1,…,p qk *∈{ ac coefficients that | qk * |=1 or | qk * |=2} (4) if p = 0 or qp * =1 (it indicates bf=0), the block is ub and we move to next block. otherwise, take the block as qb. each extracted bit b' in qbs is calculated by: b'=0 if |qk * |=1 or b'=1 if |qk * |=2. for each qb, the restored ac coefficients qk * ' can be calculated by eq. (5): qk *'=sign(qk *) (5) the scanning procedure continues. note that after the extraction of l + l bits, the rest blocks won't be scanned and restored because there's no change in them during embedding process. 4. experimental results we select test images from ucid [11] and convert these color images into gray. the size of images in experiments is either 384·512 or 512·384. in our experiments, facebook does not resize those images. we implement four competing methods: huang's method [2]; lsb, 2-lsb and lsb+2-lsb based on [7]. note that lsb+2-lsb are called mixlsb in our experiments. for fair comparison, these four competing methods (huang's method [2], lsb[7], 2lsb[7] and mixlsb[7]) and our proposed method are all compressed with qfembed=71 before embedding. the embedding bits are randomly generated and all cover images are compressed by jpeg standard [12]. 4.1. data extraction accuracy the 1338 images in ucid [11] are all used for comparing accuracy. we use bit error rate (ber) to describe the performance of accuracy for each method, which is defined in eq. (6). ber=#(error)/#(embedded) (6) where #(·) is the function of calculating the number of bits. clearly, lower ber indicates better performance. as we can see from the table 1, even excluding the influence of qfembed, our method is still much better than other methods in terms of ber. the four compared methods are totally invalid when it comes to facebook. note that the random guess for each bit is about 0.5 and proceedings of engineering and technology innovation, vol. 7, 2017, pp. 25 30 copyright © taeti 28 four compared methods are all very close to t his rate. huang's method is designed for pure jpeg and not robustness to overflowing. ning's methods are failed on the current facebook system. one of potential reasons might be the updating of facebook. the superior performance of our method is attributed to the proposed block selection algorithm. the selected blocks are robustness to re-compression of jpeg and other unknown functions on facebook. table 1 comparison of bit error rate (ber) of 1338 images in ucid for each method method proposed huang[2] lsb[7] 2lsb[7] mixlsb[7] ber 0.18% 44% 46% 48% 46% 4.2. embedding capacity since the successfully embedded bits are almost zero for four competing methods, it has no need to discuss their embedding capacity performance. the 1338 images in ucid [11] are all used to evaluate the capacity of the proposed scheme. the average embedding capacity for our method is about 12064 bits in 384·512 image, which achieves a high embedding rate: 6.1%.the capacity of our method is related to the number qualified blocks (qbs ), which has a lot to do with the content of the image itself. in other words, an image with more qbs is likely to achieve higher capacity. therefore, sorting out high -capacity image based on the number of qbs can improve the average embedding capacity. we sort 1338 cover images from largest capacity to lowest and embed bits in high-capacity image. the refined average capacity is about 20000 bits in 200 images. 4.3. file size expansion in order to evaluate the performance of file size expansion, we randomly select 100 images from ucid [11]. we only choose the size values of fbstego images in each method. we compress cover images with qfembed=71 in jpeg standard and call it as jpeg71. we upload jpeg71 on facebook and download to obtain fbjpeg71. we add jpeg71 images and fbjpeg71 images in comparison, because the ideal size is not only small, but also close to the same image without embedding rates. the variation of file size for different payloads is measured by bytes and the results are shown in fig. 1. as can be seen, lsb, 2lsb and mixlsb have no increment in file size and are almost the same size. this is because these three methods use vli coding [13] which won't affect the entropy codes of jpeg. jpeg71 and fbjpeg71 are unchanged for no embedding bits. co mpared to huang, we have smaller average file size and slower trend of growth. and most of file size values in proposed method are located in the spac e between fbjpeg71 to jpeg71, which contributes to covertness. fig. 1 file size (bytes) of fbstego image 4.4. image visual quality to evaluate the performance of visual quality, we randomly select 100 images from ucid [11]. the performance of visual quality is evaluated by peak signal-to-noise ratio (psnr) and structural similarity index (ssim). one example of the proposed method is shown in fig. 2. these three types of images are also used in the four competing methods. the payloads in stego vary proceedings of engineering and technology innovation, vol. 7, 2017, pp. 25 30 copyright © taeti 29 from 1000 bits to 20000 bits. we use mutual information (mi) to describe the mutual dependence between the two random variables: x and y. x is the real embedded sequence and y is the sequence extracted by steganography. the mutual information of x and y is donated by i(x;y) and h(·) is the function of calculating the entropy of a random sequence. mi can be calculated by eq. (7): mi=i(x;y)=h(x)+h(y)-h(x,y) (7) (a) cover image (b) steego image (embedded 10000 bits) (c) jpeg71 image (d) restored image (embedded 10000 bits) fig. 2 one example of proposed method (a) original image in steganography (b) the image embedded with 10000 bits by proposed method (c) the image restored form fbstego (d)the image compressed by jpeg in quality factor 71 (a) ssim (b) psnr fig. 4 visual quality of stego images vs cover images (a) ssim (b) psnr fig. 5 visual quality of stego images vs jpeg71 images (a) ssim (b) psnr fig. 6 visual quality of restored images vs cover images the results of visual quality described by ssim and psnr for five methods are shown in fig s. 4-6. the y-axis is the multiply of mi and payload: mi·payload. we use ssim and psnr as the x-axis to show for same visual quality, which method can achieve larger useful capacity. this evaluation of combin ing the influence of visual quality, the accuracy and the capacity. since only our method and huang's method are reversible, the other three methods are not shown for restoring images . the mi and useful capacity of our method outperforms huang’s method while the value of visual quality is similar. from the res ults, one can proceedings of engineering and technology innovation, vol. 7, 2017, pp. 25 30 copyright © taeti 30 see that the proposed method has much better mi·payload than four competing methods with same visual quality. the visual quality of our proposed method is mainly concentrated in bigger value of ssim and psnr, which outperforms four competing methods. with the increment of payload, our proposed method shows a more outstanding and stable performance of the mi and visual quality than these four methods. our method won't be influenced by the unstable blocks. thus, mi and visual quality are stable with the payload growing. in conclusion, our method has great and stable performance on visual quality and mi. 5. conclusions many previous literatures noticed that facebook had huge potential on sharing secret messages. in this work, we performed an in-depth investigation of the various lossy operations on facebook based on the experimental data. based upon such prior knowledge, we built a facebook message sharing system called fmss that can effectively embed a large amount of data and successfully extract them out from the downloaded images, defeating the uncontrolled lossy operations on facebook. the experiments revealed that the proposed method offers much higher embedding capacity, and can extract the data successfully with very high probability compared with t he state-of-the-art techniques. furthermore, the restored image upon data extraction is of high quality, and the file size expansion is negligible. references [1] a. castiglione, g. cattaneo, and a. de santis, “a forensic analysis of images on online social networks,” international conf. intelligent networking and collaborative systems (incos), ieee press , january 2011. [2] f. huang, x. qu, h. j. kim, and j. huang, “reversible data hiding in jpeg images,” ieee transactions on circuits and systems for video technology, vol. 26, no. 9, pp. 1610-1621, september 2016. [3] j. hiney, t. dakve, k. szczypiorski, and k. gaj, “using facebook for image steganography,” international confer . availability, reliability and security (ares), ieee press , august 2015. [4] a. castiglione, b. d'alessio, and a. de san tis, “steganography and secure communication on online social networks and online photo sharing,” international conf. broadband and wireless computing, communication and applications (bwcca), ieee press , october 2011. [5] s. nagaraja, a. houmansadr, p. piyawongwisal, v. singh, p. agarwal, and n. borisov, “stegobot: a covert social network botnet,” international workshop on information hiding , springer berlin heidelberg, 2011. [6] k. solanki, a. sarkar, and b. manjunath, “yass: yet another steganographic scheme that resists blind steganalysis,” international workshop on information hiding. springer berlin heidelberg, pp. 16-31, 2007. [7] j. ning, i. singh, h. v. madhyastha, s. v. krishnamurthy, g. cao, and p. mohapatra, “secret message sharing using online social media,” ieee conf. communications and network security (cns), ieee press, october 2014. [8] n. f. johnson and s. jajodia, “exploring steganography: seeing the unseen,” computer, vol. 31, no. 2, pp. 26-34, february 1998. [9] n. d. amsden, l. chen, and x. yuan, “transmitting hidden information using steganography via facebook,” international conf. computing, communication and networking tech nologies (icccnt), ieee press , july 2014. [10] a. latham, “steganography,” http://linux01.gwdg.de/alatham/stego.html [online]. accessed 11, october 2016. [11] g. schaefer and m. stich, “ucid an uncompressed colour image database,” proceedings of spie, vol. 5307, pp. 472-480, 2003. [12] g. k. wallace, “the jpeg still picture compression standard,” ieee transactions on consumer electronics, vol. 38, no. 1, pp. xviii-xxxiv, february 1992. [13] t. nakahashi and t. kinoshita, variable length image coding system, u.s. patent, 5,319,457, june 07, 1994,  proceedings of engineering and technology innovation, vol. 10, 2018, pp. 29 35 battery management system for 24-v battery-powered electric wheelchair tzu-hsiang tseng * , liang-rui chen, yu-jia zhang, bo-rui xu, jin-an li department of electical engineering, national changhua university of education, changhua, taiwan, roc received 25 april 2018; received in revised form 30 may 2018; accepted 13 july 2018 abstract this paper describes a simplified battery-management system using a digital signal processor for a 24-v battery-powered electric wheelchair to modulate and protect the battery from operating outside its safe operating area, such as over-temperature, under-temperature, over-voltage, under-voltage, over-current discharge, or over-current charging. using the buck-boost circuit topology, we achieved and demonstrated both battery balance and protection in a prototype 12-v/40-ah series-connected two-battery module. simulation results show that the proposed battery management system is feasible for a 24-v battery-powered electric wheelchair application. keywords: battery management system (bms), digital signal processor, battery balance, battery protection 1. introduction with the rapid development of science and technology, traditional hand-propelled wheelchairs have gradually evolved into electrically powered wheelchairs that can be operated by individuals with mobility impairments [1-2]. recently, a wide variety of electric wheelchairs have emerged, including solar hybrid wheelchairs [3] and four-wheel-drive wheelchairs [4]. as various forms of electric wheelchairs continue to develop, energy sources, cost-effectiveness, and safety is also becoming increasingly more important topics of discussion [5-7]. electric wheelchairs malfunctions have been reportedly causing potential traffic concerns. power system diagnosis showed the majority of the incidents and damages to the system were caused by imbalanced or unprotected batteries. in this study, we developed a battery-management system (bms) for electric wheelchairs. the electric motors of commercial electric wheelchairs are commonly powered by batteries. a single battery can only supply limited voltage and capacity and has a shorter battery life under a high-power state compared to multiple batteries. to operate in a high-voltage and high-power state, serially connected batteries are often used. however, balancing the capacity in serially connected batteries becomes a challenge as the batteries are often not identical, with capacity variance resulting from the manufacturing process, battery usage, such as repeated charging and discharging, or extended applications in a high-power state causing over-discharge. this results in an imbalanced capacity and voltage among serially connected batteries, which affects the overall lifespan and energy loss of the batteries. many research articles [8-20] have proposed possible solutions to improve a series-connected two-battery module. the protection of energy storage in electric wheelchair batteries guarantees the safety of its users. 2. system specification we developed a block diagram of the battery-balancing system and battery-protection circuit, as shown in fig. 1. through a voltage-detection circuit, current-detection circuit, and battery-balance controller, the battery-balancing system can passively balance the voltage of two batteries by extracting voltage from the higher-potential battery to the lower-potential battery. * corresponding author. e-mail address: bob84425@gmail.com proceedings of engineering and technology innovation, vol. 10, 2018, pp. 29 35 copyright © taeti 30 a digital signal processor (dsp) was used to measure, monitor, and convert the received battery voltage through an internal analog-to-digital converter and transmit the signal to the double-active switch of the buck-boost circuit. the pwm signal was modulated to control the on and off states of the switch to balance the current and voltage. the battery protection circuit was composed of a temperature-detection circuit, current-detection circuit, and battery-protection control circuit. during the charging or discharging stage of the series-connected two-battery module, the battery-protection circuit detected and transmitted voltage signals collected from the parameters of the surrounding environment, such as the charging current, discharging current, and temperature, to the dsp. these voltage signals were then converted through the internal analog-to-digital converter and transmitted to the corresponding switches for modulation and control. voltage detection circuit temperature detection circuit voltage detection circuit temperature detection circuit current detection circuit controller b a t t e r y v o l t a g e r e g u l a t o r b1 v b1 t b2 v b2 v b i 1 s 2 s fig. 1 bms block diagram l s2 s1+ + vb1 vb2 il fig. 2 bidirectional buck-boost circuit diagram in this paper, we focused on the development of the dsp to understand the operating state of the battery based on the digital signals and to trigger the protection and balancing actions when necessary. using a bidirectional buck-boost circuit, the battery voltage in serially connected batteries can be regulated, as shown in fig. 2. proceedings of engineering and technology innovation, vol. 10, 2018, pp. 29 35 copyright © taeti 31 the bidirectional buck-boost circuit was composed of two power transistors s1, s2 and an inductor l to form a synchronous rectified, bidirectional power-transfer buck-boost circuit. thus, the battery voltage can be regulated by transferring power from the higher-voltage battery to the lower-voltage battery. starting with battery b1 as the basis, the operation was divided into two modes: buck mode and boost mode. buck mode lowered the voltage of battery b1, whereas boost mode increased the voltage of battery b1. when the battery voltage vb1 was higher than vb2, the upper switch s1 was turned on while the lower switch s2 was turned off, forming a discharge path from battery b1 to inductor l, as shown in fig. 3(a). this discharge current flowed through and stored power in inductor l. next, the upper switch s1 was turned off while the lower switch s2 was turned on, forming a charging path from inductor l to battery b2, as shown in fig. 3(b). this inductor current il flowed into battery b2, charging the battery. in this operating mode, battery b1 was discharged while battery b2 was charged, effectively reducing the voltage of b1 and increasing the voltage of b2. (a) s1 on, s2 off (b) s1 off, s2 on fig. 3 buck mode circuit diagram (a) (b) (a) s2 on, s1 off (b) s2 off, s1 on fig. 4 boost mode circuit diagram (a) (b) on the other hand, when the battery voltage vb1 was lower than vb2, the lower switch s2 was turned on while the upper switch s1 was turned off, forming a discharge path from battery b2 to inductor l, as shown in fig. 4(a). this discharge current flowed through and stored power in inductor l. next, the lower switch s2 was turned off while the upper switch s1 was turned on, forming a charging path from inductor l to battery b1, as shown in fig. 4(b). this inductor current il flowed into battery b1, charging the battery. in this operating mode, battery b1 was charged while battery b2 was discharged, effectively increasing the battery voltage of b1 and reducing the battery voltage of b2. proceedings of engineering and technology innovation, vol. 10, 2018, pp. 29 35 copyright © taeti 32 3. simulation results the circuit operation timing simulation in buck mode is shown in fig. 5, where vgs1 and vgs2 were the operating gate voltages of switch s1 and s2 respectively, ib1 was the discharging current of battery b1, and ib2 was the charging current of battery b2. similarly, the circuit operating timing simulation in boost mode is shown in fig. 6, where vgs1 and vgs2 were the operating gate voltage of switch s1 and s2 respectively, ib1 was the charging current of battery b1, and ib2 was the discharging current of battery b2. 0.0585 0.05855 time(s) 1 0 0.4 0.2 0.6 0.8 1 0 0.4 0.2 0.6 0.8 0 0 0 -1 1 2 3 -2 -3 1 -1 -2 -3 -4 0.058650.05845 0.0586 0.0587 fig. 5 buck-mode timing simulation of a buck-boost circuit 1 0 0.4 0.2 0.6 0.8 0.0524 time(s) 15 0 0.4 0.2 0.6 0.8 -10 -5 0 5 -5 0 5 -15 10 15 -5 0 5 10 0.0525 0.0526 0.05280.0527 fig. 6 boost-mode timing simulation of a buck-boost circuit 25 20 15 10 5 -5 0 200 400 600 800 10000 time(s) 25 fig. 7 simulation result of the battery-balancing operation proceedings of engineering and technology innovation, vol. 10, 2018, pp. 29 35 copyright © taeti 33 fig. 7 shows the simulation result of battery-balancing operations. the b1 and b2 voltages were 12.7 v and 12.08 v, respectively, with a 0.72-v delta. the battery voltage difference was clearly reduced to within 0.1 v during the measurement timeframe. fig. 8 shows the simulation result of the battery’s voltage protection. when the battery voltage was higher than 30 v, the charging switch was turned on to stop the external voltage from charging the battery. as the battery voltage recovered to 24 v, the protection was switched off to allow normal operation in a closed circuit. when the battery voltage was lower than 21 v, the discharging switch was turned on to stop the battery from discharging to the load. again, as the battery voltage recovered to 24 v, the protection was switched off to allow normal operation in a closed circuit. fig. 9 shows the simulation result of the battery’s temperature protection. when the temperature voltage was higher than 4.43 v (lower than -1 °c) or lower than 2.5 v (higher than 43 °c), the charging and discharging switch were turned on to form an open circuit, activating the battery’s temperature-protection mechanism. if the temperature was within the operational range, both switches were turned off for normal operation in a closed circuit. battery voltage discharge switch 30v 21v charge switch 1 0 0.4 0.2 0.6 0.8 1 0 0.4 0.2 0.6 0.8 0.02 0.04 0.06 0.08 0.10 time(s) 18 20 22 24 26 28 30 32 fig. 8 simulation result of the battery’s voltage protection charge switch discharge switch 4.43v 2.5v temperature voltage 5 4.5 3.5 2.5 4 3 1 0 0.4 0.2 0.6 0.8 1 0 0.4 0.2 0.6 0.8 0.02 0.04 0.06 0.08 0.10 time(s) fig. 9 simulation result of the battery’s temperature protection 4. conclusion this paper successfully developed a bms with a voltage-balancing and -protection mechanism, to achieve a voltage-balancing modulation, voltage monitor, current monitor, and temperature monitor for the battery. the description of these applications is as follows: battery voltage balancing modulation: monitor and modulate the charging and discharging current of each battery in serially connected batteries. effectively increase battery’s lifespan via voltage regulation by balancing the batteries to within a 0.1-v range. proceedings of engineering and technology innovation, vol. 10, 2018, pp. 29 35 copyright © taeti 34 battery voltage monitor: when the battery voltage was higher than 30 v, the charging switch was turned on to form an open circuit to protect the battery from over-voltage charging. when the battery voltage was lower than 21 v, the discharging switch was turned on to form an open circuit to protect the battery from over-voltage discharging. battery current monitor: when the battery current was higher than 25 a for longer than 1 s, the charging switch was turned on to form an open circuit to protect the battery from over-current charging. when the battery current was higher than 60 a for longer than 1 s, the discharging switch was turned on to form an open circuit to protect the battery from over-current discharging. battery temperature monitor: when the battery temperature was higher than 43 °c or lower than -1 °c, the charging and discharging switches were turned on to form an open circuit, protecting the battery from operating outside its safe operating region. 5. outlook the voltage-balancing and battery-protection mechanisms for a battery were developed successfully in this study. however, the safety parameters, electromagnetic compatibility (emc), and tolerance need to be improved for practical applications. the description of these aspects is as follows: safety parameters include settings for over-voltage, over-current, and over-temperature operations. if the parameters were too loose, batteries would be damaged before the protection mechanism was activated. on the other hand, if the parameters were too tight, the protection mechanism would activate within the safe operating region, interrupting the normal operation of electric wheelchairs. thus, the definition of the safety parameters is critical. electromagnetic interference (emi) refers to the interference of the designed battery-balancing circuit and the battery-protection circuit with other electronic systems. design in accordance with available emi and emc standards for power converters is suggested. finally, tolerance refers to the operation of the designed battery-balancing circuit and the battery-protection circuit outside its safe operating region. design in accordance with ul or ce standards for power converters is suggested. references [1] m. wada and f. kameda, “a joystick car drive system with seating in a wheelchair,” 2009 35th annual conference of ieee industrial electronics, ieee press, november 2009, pp. 2163-2168. [2] y. k. kim, y. h. cho, n. c. park, s. h. kim, and h. s. mok, “in-wheel motor drive system using 2-phase pmsm,” 2009 ieee 6th international power electronics and motion control conference, wuhan, may 2009, pp. 1875-1879. [3] p. kishore, m. ananth, s. chidambaram, b. vivekanandan, and n. a. gounden, “solar based hybrid electric powered wheelchair,” 2013 texas instruments india educators' conference, ieee press, april 2013, pp. 18-24. [4] m. wada, “holonomic and omnidirectional wheelchairs with synchronized 4wd mechanism,” 2007 ieee/rsj international conference on intelligent robots and systems, ieee press, november 2007, pp. 1196-1202. [5] r. h. putra, a. g. w. rahman, e. s. ningrum, and d. s. purnomo, “design and stress analysis on electric standing wheelchair,” 2017 international electronics symposium on engineering technology and applications (ies-eta), ieee press, september 2017, pp. 112-117. [6] k. vitols and a. podgornovs, “concept of cost-effective power-assist wheelchair's electrical subsystem,” 2017 5th ieee workshop on advances in information, electronic and electrical engineering (aieee), ieee press, november 2017, pp. 1-4. [7] k. k. ayten, a. dumlu, and a. kaleli, “real-time trajectory tracking control for electric-powered wheelchairs using model-based multivariable sliding mode control,” 2017 5th international symposium on electrical and electronics engineering (iseee), ieee press, october 2017, pp. 1-6. [8] m. j. rana and m. a. abido, “energy management in dc microgrid with energy storage and model predictive controlled ac–dc converter,” iet generation, transmission & distribution, vol. 11, no. 15, pp. 3694-3702, november 2017. proceedings of engineering and technology innovation, vol. 10, 2018, pp. 29 35 copyright © taeti 35 [9] q. xu et al., “a decentralized dynamic power sharing strategy for hybrid energy storage system in autonomous dc microgrid,” in ieee transactions on industrial electronics, vol. 64, no. 7, pp. 5930-5941, july 2017. [10] d. kumar, f. zare, and a. ghosh, “dc microgrid technology: system architectures, ac grid interfaces, grounding schemes, power quality, communication networks, applications, and standardizations aspects,” in ieee access, vol. 5, pp. 12230-12256, june 2017. [11] a. merabet, k. tawfique ahmed, h. ibrahim, r. beguenane, and a. m. y. m. ghias, “energy management and control system for laboratory scale microgrid based wind-pv-battery,” ieee transactions on sustainable energy, vol. 8, no. 1, pp. 145-154, jan. 2017. [12] m. farrokhabadi, s. könig, c. a. cañizares, k. bhattacharya, and t. leibfried, “battery energy storage system models for microgrid stability analysis and dynamic simulation,” ieee transactions on power systems, vol. 33, no. 2, pp. 2301-2312, march 2018. [13] s. sharma, s. bhattacharjee, and a. bhattacharya, “grey wolf optimisation for optimal sizing of battery energy storage device to minimise operation cost of microgrid,” iet generation, transmission & distribution, vol. 10, no. 3, pp. 625-637, march 2016. [14] b. xie, y. liu, y. ji, and j. wang, “two-stage battery energy storage system (bess) in ac microgrids with balanced state-of-charge and guaranteed small-signal stability,” energies, vol. 11, no. 2, pp. 1-14, february 2018. [15] b. v. mbuwir, f. ruelens, f. spiessens, and g. deconinck, “battery energy management in a microgrid using batch reinforcement learning,” energies, vol. 10, no. 11, pp. 1846-1-1846-19, november 2017. [16] a. hussain, v. h. bui, and h. m. kim, “impact analysis of demand response intensity and energy storage size on operation of networked microgrids,” energies, vol. 10, no. 7, pp. 882-1 882-19, june 2017. [17] a. hussain, v. h. bui, and h. m. kim, “fuzzy logic-based operation of battery energy storage systems (besss) for enhancing the resiliency of hybrid microgrids,” energies, vol. 10, no. 3, pp. 271-1 271-19, february 2017. [18] a. hussain, v. h. bui, and h. m. kim, “an energy-based control strategy for battery energy storage systems: a case study on microgrid applications,” energies, vol. 10, no. 5, pp. 215-1-215-20, february 2017. [19] s. y. yu, h. j. kim, j. h. kim, and b. m. han, “soc-based output voltage control for bess with a lithium-ion battery in a stand-alone dc microgrid,” energies, vol. 9, no. 11, pp. 924-1 924-15, november 2016. [20] y. s. kim, c. s. hwang, e. s. kim, and c. cho, “state of charged-based active power sharing method in a stad alone microgrid with high penetration level of renewable energy sources,” energies, vol. 9, no. 7, pp. 480-1 480-13, june 2016.  proceedings of engineering and technology innovation, vol. 15, 2020, pp. 16 23 natural nanoporous filter material as a new high-efficiency natural adsorbent to remove textile dyes chin-ya kuo, hsiao-han liu * department of biological science & technology, i-shou university, kaohsiung, taiwan received 14 auguest 2019; received in revised form 27 november 2019; accepted 25 march 2020 doi: https://doi.org/10.46604/peti.2020.4668 abstract the objective of this paper is to perform the innovation design of removing most textile dyes that are harmful to the environment and might induce cancer. however, many methodologies had been developed for various chemical and physical processes to remove different dyes, such as ozone oxidation, electrochemical methods, chemical coagulation, hypochlorite oxidation, and adsorption to remove dyes from wastewater. a novel nanoporous filter methodology and mathematic simulations for adsorption were established as an effective medium for removing dyes from wastewater which was compared to other expensive treatments. the different concentrations of the dye liquid are used as a tester, and the different concentrations of nanoporous adsorbent were added in a uniform distribution and were tested with different time courses and under different temperatures. the final readings were measured by a spectrophotometer and fit into a mathematic model. the result indicates that this nanoporous and natural adsorbents are very good at cleaning the dyes in this system. the fit-in mathematic models could be applied in these tests which can be used in the industrial conditions for a low cost without secondary dye pollutions. keywords: natural nanoporous filter, textile wastewater, adsorption test, mathematic simulation model 1. introduction most of the ready-to-wear market and the pursuit of fashion in today's society have resulted in large garment factories around the world that use large amounts of water, chemicals, dyes, etc. these factories will produce a large amount of wastewater. if they didn’t discharge into the river or the sea, they would directly cause environmental pollution. indirectly, they would affect human health and even death due to bioaccumulation [1-4]. dyes are aromatic molecular structures that are expected to be strong and stable. therefore, they are difficult to degrade (making them more difficult to biodegrade). they are very toxic and harmful, however, they may cause a large amount of environmentally toxic waste or pose a serious hazard to public health. scientists have developed various chemical and physical processes to remove different dyes. for example, dyes are removed from wastewater by ozone oxidation, chemical coagulation, electrochemical methods, hypochlorite oxidation, and adsorption [2, 5-8]. adsorption is an effective method for removing dyes from wastewater, and it is an alternative to other expensive processing techniques. some studies in the literature have reported the use of natural adsorbents to remove textile dyes because of their economic appeal and advantages, primarily because they are inexpensive and available. however, we need to find effective natural adsorbents with high adsorption capacity. furthermore, they can be easily separated and set up standard procedures for recycling and regeneration [6, 9-11]. * corresponding author. e-mail address: hansliu@isu.edu.tw tel.: +886-7-6151100 ext.7318; fax: +886-7-6151959 proceedings of engineering and technology innovation, vol. 15, 2020, pp. 16 23 17 among natural materials, natural clay minerals are ideal adsorbents due to their low cost, abundant resources, strong adsorption, and ion exchange capacity. the adsorption capacity of these clay minerals is derived from the negative charge of their surface structure and can adsorb positively charged ions in the sewage. the adsorption performance is produced by a large specific surface area and a good porosity. an in-depth study of the interaction between pollutants and clay particles are needed, such as bentonite and diatomaceous earth. clay minerals are effective in adsorbing inorganic and organic pollutants. the most widely used adsorbents for wastewater is activated carbon because of its high removal capacity. however, its use is subject to regeneration problems, high-cost production, phase separation difficulties, and poor mechanical properties [7, 12]. the natural nano-composite filter is a kind of natural mineral, which is obtained by washing with sulfuric acid and baking at high temperatures. it can be processed into particles. for example, adding natural adhesive can make the mineral powder into filter particles of smoke. the manufacture of natural nano-composite filter material is from a variety of mineral flour with the specific proportion of ore combination, and their inside appearances are like a sponge structure with many nano-sized pores, which increased its adsorption capacity [13]. the production procedure of this natural nano-composite is simpler than activated carbon, so the cost is lower twice than functional activated carbon. because it is made of natural mineral powder, it would not cause the second pollution into the environment. this study focuses on the experimental conditions of nano-composite adsorbent with different dyes in different rpm (shaking speed), ppm (solute dye concentration). moreover, this study would investigate the formula in which their characteristics of dye-binding and their emission requirements by the selection of appropriate treatment methods such as the effects of binding time, different dye concentrations, different shaking speeds. the dyes in this study were methylene blue (mb) and cationic dyestuffs no. 257 black (bd). mb is commonly used in absorption test, while the bd dye does not contain acidic groups and exhibits catholicity in water, also known as cationic dyes. 2. materials and methods 2.1. preparation of adsorbent the natural nano-composite filter is powdered and requires a water wash step for the accuracy of the experimental data. washing step, take 10 g nano-composite filter in a 50 ml centrifuge tube and add 40 ml of distilled water, mix well, centrifuge for 5 minutes (3000 rpm, 25°c), pour off the supernatant, add 10 ml. the distilled water was then poured into a glass petri dish and finally baked overnight (90°c) in an oven. 2.2. dye solution preparation in this study, the cationic dye “methylene blue” was selected as the control group; the cationic dye “cationic dyestuffs no. 257 black” was selected as the control group to compare the practicability of the nano-composite filter. a stock solution having a concentration of 1000 ppm has been obtained by dissolving a precisely weighed dye sample in distilled water. the dye stock solution (1000 ppm) was diluted with distilled water to the desired concentration, while other concentrations were required. all solutions were not subjected to any ph adjustment. the chart of figs. 1(a) and 1(b) showed the full length scanning of these two dyes, and in the chart that showed the unit of the y axis as od reading, and the unit of the x-axis as nm. 2.3. adsorption balance time weigh 1g washed with water nano-composite filter placed in 50 ml centrifuge tube and then added 50 ml 300 ppm concentration of the dye solution, mixed well, the use of 125rpm operation of the temperature control vibrator bottle placed at 25 ± 2°c , the sampling interval at different times from the shaker. after centrifugation (3000 rpm, 25°c, 5 min), the residual proceedings of engineering and technology innovation, vol. 15, 2020, pp. 16 23 18 concentration of the dye solution was measured using an elisa at a wavelength of 660 nm (methylene blue) and 600 nm (cationic dyestuffs no. 257 black), respectively (figs. 2(a) and 2(b)). (a) methylene blue (b) cationic dyestuffs no. 257 black fig. 1 absorption full-length scaninng charts (a) adsorption balance time (b) adsorption effect fig. 2 absorption experiment of the dyes on the 96-well plate 2.4. effect of different concentrations on adsorption add 300 ppm dye solution to a series of 50 ml centrifuge tubes, add 1 g nano-composite filter, mix well, and place the bottle at 25°c, 30°c, 35°c using a temperature-controlled vibrator operating at 125 rpm. c, 40°c, 45°c, 50°c, 55°c, shaking for 3 h. after centrifugation (3000 rpm, 25°c, 5 min), the absorbance of the dye solution was measured at 660 nm and 610 nm using an elisa, respectively. 2.5. effect of different temperatures on adsorption add 300 ppm dye solution to a series of 50 ml centrifuge tubes, add 1 g nano-composite filter, mix well, and place the bottle at 25°c, 30°c, 35°c using a temperature-controlled vibrator operating at 125 rpm. c, 40°c, 45°c, 50°c, 55°c, shaking for 3 h. after centrifugation (3000 rpm, 25°c, 5 min), the absorbance of the dye solution was measured at 660 nm and 610 nm using an elisa, respectively. 2.6. effect of different rpm on adsorption add 300 ppm dye solution to a series of 50 ml centrifuge tubes, add 1 g nano-composite filter, mix well, and place the bottle at 25°c, 30°c using a temperature-controlled vibrator operating at 100 rpm, 125 rpm, 150 rpm. c, 35°c, 40°c, 45°c, 50°c, 55°c, shaking for 3 h. after centrifugation (3000 rpm, 25°c, 5 min), the absorbance of the dye solution was measured at 660 nm and 610 nm using an elisa, respectively. proceedings of engineering and technology innovation, vol. 15, 2020, pp. 16 23 19 2.7. adsorption kinetics adsorption kinetics is the main method to explore the adsorbent adsorption machine. the kinetic models for adsorption are pseudo-first order and pseudo-second order. the expressions are: (1) pseudo-first order 1 1         e t q in k t q (1) (2) pseudo-second order 2 1 1    e t e k q q q (2) where qt is the adsorption amount per unit mass of adsorbent at time t (mg/g), qe is the adsorption amount of adsorption equilibrium adsorbent (mg/g), k1 is the adsorption rate constant of the pseudo-first-order equation kinetic model (min -1 ), and k2 is the adsorption rate constant of the quasi-second-order equation dynamics model (g．mg -1 ．min -1 ) [14-17]. 3. results and discussion 3.1. dye full-length scanning chart the following charts were the full-length visible light scanning of the two dyes at 660 nm (methylene blue, mb) and 610 nm (cationic dyestuffs no. 257 black, bd). the charts showed the lambda max of the two dyes from full wavelength scanning (visible wavelength: 380 nm 700 nm). the following experimental results for the control group of distilled water, the absorbance reading is 0.0393. 3.2. adsorption balance time according to fig. 2(a), the absorbance of the supernatant decreases with time and tends to be constant after 3 h. the results showed that at room temperature nano-composite filter adsorbed about 3 h equilibrium, and the following experiment selected 3 h adsorption equilibrium times. 3.3. effect of different concentrations on adsorption according to fig. 3, the adsorption effects of different concentrations were determined under fixed rpm. the three images clearly show that the bd is almost completely absorbed in the concentration of 400 ppm at 100 ppm, and the supernatant can be completely clarified after 3 h; the concentration is 600 ppm, and the supernatant cannot be clarified after 3 h of adsorption equilibrium time. (a) 100 rpm fig. 3 adsorption effect at different concentrations of bd proceedings of engineering and technology innovation, vol. 15, 2020, pp. 16 23 20 (b) 125 rpm (c) 150 rpm at 25°c fig. 3 adsorption effect at different concentrations of bd (continued) according to fig. 4, the adsorption concentration of mb concentration of 300 ppm and 350 ppm can be clarified after adsorption. the data shows that the maximum adsorption concentration of the material for mb is below 550 ppm. (a) 100 rpm of mb at 25°c (b) 125rpm of mb at 25°c (c) 150 rpm of mb at 25°c fig. 4 adsorption effect at different concentrations 3.4. effect of different rpm on adsorption according to fig. 5, the adsorption effect of the same concentration of bd under different rpms is shown. several data showed that 125 rpm is better than 150 rpm. for this result, because a larger rpm will be studied later, the reason will be discussed after the data comes out. according to fig. 6, the adsorption effect of the same concentration of mb at different rpms is shown. the results at a concentration of 300 ppm to 350 ppm showed that the effect of 150 rpm was better than that of 125 rpm; the concentration of 400 ppm 500 ppm showed that the effect of 125 rpm was better than 150 rpm. for this result, because a larger rpm will be studied later, the reason will be discussed after the data comes out. proceedings of engineering and technology innovation, vol. 15, 2020, pp. 16 23 21 (a) 400 ppm of bd (b) 450 ppm of bd (c) 500 ppm of bd (d) 550 ppm of bd (e) 600 ppm of bd fig. 5 adsorption effects at different rpm (a) 300 ppm of mb fig. 6 the adsorption effect at different rpm proceedings of engineering and technology innovation, vol. 15, 2020, pp. 16 23 22 (b) 350 ppm of mb (c) 400 ppm of mb (d) 450 ppm of mb (e) 500 ppm of mb fig. 6 adsorption effect at different rpm (continued) 3.5. adsorption kinetics finally, insert the pseudo-second order. the formula for the two dyes is shown in table 1. the formula quantifies the amount of adsorption. table 1 the result of calculation in simulation formula k2(g．mg -1 ．min -1 ) qe (mg/g) cationic dyestuffs no.257 black (bd) 2.157x10 -1 4.523 methylene blue (mb) 2.310x10 -1 2.643 4. conclusions the obtained shows that the commercially available natural nanoporous filter material in this study leaves dye-polluted wastewater 100% dye-free. to reach the dye-free condition for methylene blue our study shows that the ideal set of variables is a concentration of up to 550 ppm at 25°c and a shaking speed at 100 rpm for 3 hours. to reach the dye-free condition for cationic dyestuffs no. 257 black (bd) the optimum adsorption condition is a concentration of 550 ppm at 45°c and a shaking speed at 150 rpm for 3 hours. besides, using pseudo-second-order adsorption kinetics, we created a 3-parameter model to design a schedule of nano-composite concentrations for different scales of application. these formula parameters will enable the industry to apply the findings to a range of production scales at the lowest cost and with no secondary pollutions. conflicts of interest the authors declare no conflict of interest. proceedings of engineering and technology innovation, vol. 15, 2020, pp. 16 23 23 references [1] m. mohammed, a. shitu, and a. ibrahim, “removal of methylene blue using low cost adsorbent: a review,” research journal of chemical sciences, vol. 4, no. 1, pp. 91-102, january 2014. [2] r. kant, “textile dyeing industry an environmental hazard,” natural science, vol. 4, no. 1, pp. 22-26, december 2011. [3] m. s. chiou, p. y. ho, and h. y. li, “adsorption of anionic dyes in acid solutions using chemically cross-linked chitosan beads,” dyes pigments, vol. 60, no. 1, pp. 69-84, january 2004. [4] m. c. j. d. shih and w. treatment, “kinetics of the batch adsorption of methylene blue from aqueous solutions onto rice husk: effect of acid-modified process and dye concentration,” desalination water treatment, vol. 37, no. 1-3, pp. 200-214, january 2012. [5] a. demirbas, “agricultural based activated carbons for the removal of dyes from aqueous solutions: a review,” journal of hazardous materials, vol. 167, no. 1-3, pp. 1-9, auguest 2009. [6] s. rakass, a. mohmoud, h. oudghiri hassani, m. abboudi, f. kooli, and f. al wadaani, “modified nigella sativa seeds as a novel efficient natural adsorbent for removal of methylene blue dye,” molecules, vol. 23, no. 8, pp. 1950, auguest 2018. [7] j. mo, q. yang, n. zhang, w. zhang, y. zheng, and z. zhang, “a review on agro-industrial waste (aiw) derived adsorbents for water and wastewater treatment,” journal of environmental management, vol. 227, no. 1, pp. 395-405, december 2018. [8] b. h. tan, t. t. teng, and a. m. omar, “removal of dyes and industrial dye wastes by magnesium chloride,” water research, vol. 34, no. 2, pp. 597-601, february 2000. [9] k. p. singh, s. gupta, a. k. singh, and s. sinha, “ optimizing adsorption of crystal violet dye from water by magnetic nanocomposite using response surface modeling approach,” journal of hazardous materials, vol. 186, no. 2-3, pp. 1462-1473, february 2011. [10] v. c. srivastava, i. d. mall, and i. m. j. c. e. j. mishra, “adsorption thermodynamics and isosteric heat of adsorption of toxic metal ions onto bagasse fly ash (bfa) and rice husk ash (rha),” chemical engineering journal, vol. 132, no. 1-3, pp. 267-278, auguest 2007. [11] j. o. amode, j. h. santos, z. m. alam, a. h. mirza, and c. c. mei, “adsorption of methylene blue from aqueous solution using untreated and treated (metroxylon spp.) waste adsorbent: equilibrium and kinetics studies,” international journal of industrial chemistry, vol. 7, no. 3, pp. 333-345, september 2016. [12] s. tahir and n. rauf, “removal of a cationic dye from aqueous solutions by adsorption onto bentonite clay,” chemosphere, vol. 63, no. 11, pp. 1842-1848, june 2006. [13] c. y. kuo and l. hsiao-han, “the study of natural nano-composite filter for industrial wastewater treatment,” proceedings of engineering and technology innovation, vol. 10, p. 42, december 2018. [14] m. c. ncibi, b. mahjoub, and m. seffen, “kinetic and equilibrium studies of methylene blue biosorption by posidonia oceanica (l.) fibres,” journal of hazardous materials, vol. 139, no. 2, pp. 280-285, january 2007. [15] y. t. huang, l. c. lee, and m. c. shih, “a study on the pseudo-second-order kinetic equation for the adsorption of methylene blue onto nitric acid-treated rice husk: comparison of linear methods,” international journal of scientific research publications desalination, vol. 8, no. 6, pp. 509-515, june 2018. [16] p. e. meissner, g. mandi, b. coulibaly, s. witte, t. tapsoba, and u. mansmann, et al., “methylene blue for malaria in africa: results from a dose-finding study in combination with chloroquine,” malaria journal, vol. 5, pp. 84, october 2006. [17] y. wong, y. szeto, w. cheung, and g. mckay, “equilibrium studies for acid dye adsorption onto chitosan,” langmuir, vol. 19, no. 19, pp. 7888-7894, auguest 2003. 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 3-v24(2023)-peti#11150(22-28).docx proceedings of engineering and technology innovation, vol. 24, 2023, pp. 22-28 development of a chaff dispense program for target tracking radar deception min-joon choi1, je-hong park2, min-seok jie2, won-hyuk choi2,* 1avionics software development center, republic of korea air force, seosan-si, korea 2department of avionics engineering, hanseo university, seosan-si, korea received 10 november 2022; received in revised form 27 january 2023; accepted 14 february 2023 doi: https://doi.org/10.46604/peti.2023.11150 abstract this study aims to develop an appropriate chaff dispensing program to deceive the target tracking radar (ttr) effectively. chaff is a countermeasure commonly used by fighter aircraft to deceive ttr. however, there has been a lack of methodology for calculating chaff dispense programs that take into account the specific characteristics of the fighter, chaff, and ttr. this study proposes a methodology that considers these variables to calculate chaff dispense programs and addresses this gap. the proposed method is demonstrated through tess engagement, which shows its effectiveness in various engagement situations. keywords: chaff, ttr, chaff dispense program, electronic warfare, tess 1. introduction chaff is a popular passive countermeasure that enables fighter aircraft to against target tracking radar (ttr) [1]. it is composed of a bundle of conductive wires that creates a larger radar cross section (rcs) than the rcs of a fighter, misleading the ttr into tracking the chaff cloud instead of a fighter as a target [2]. to deceive the ttr, some chaff must be burst to create an rcs larger than the fighter’s rcs before the fighter passes through a radar resolution cell (rrc) of the ttr [3]. several articles have been published on effective chaff dispense program calculation. for example, žák et al. [4] present an appropriate interval for bursting chaff to deceive radar in certain example situations. however, the lack of specification of the correlations among the variables used to calculate the interval makes it challenging to apply this method to other engagement situations. wu et al. [5] present a model for calculating the chaff dispense program based on missile approach time, airplane movement, and chaff cloud movement. nevertheless, this model does not consider variables such as the rcs of airplanes and chaff, or the variables used in the ttr to calculate rrc. shengliang et al. [6] use the characteristics of the chaff cloud and radar to calculate the chaff dispense program, but this method relies on ship-based calculations and may not be suitable for fighter engagement situations. as a result, there is currently no methodology for calculating the chaff dispense program that takes into account the characteristics of the fighter, chaff, ttr, and engagement situation. this study presents a methodology for calculating the chaff dispense program that considers variables such as the fighter and engagement situation. in addition, this study analyzes the deceptive effect of the calculated chaff dispense program against the ttr through engagement simulations. * corresponding author. e-mail address: choiwh@hanseo.ac.kr proceedings of engineering and technology innovation, vol. 24, 2023, pp. 22-28 23 2. methodology for chaff dispense program the following assumptions are required for the calculation of a chaff dispense program. when a fighter dispenses chaff while passing through an rrc, the first chaff dispensed creates a larger rcs than the later chaff. as a result, it can be concluded that minimizing the interval between bursts or dispensing multiple chaff simultaneously is more effective in creating a larger rcs before the fighter passes through the rrc. however, electromagnetic coupling and shielding of chaff limit the increase in rcs from dispensing chaff at very short intervals or multiple chaff bursts compared to a single chaff cartridge [7-8]. for this study, it is assumed that the chaff rcs remains constant regardless of the chaff dispense sequence. to adjust for the discrepancy between the actual rcs of chaff and the estimated rcs, a coefficient value of 0.5 is applied to the chaff rcs [9]. a chaff dispense program is designed to maximize the effectiveness of chaff in deceiving ttr during engagements. this program takes into account various factors, such as the rcs of the fighter, the rcs of the chaff, the ttr, and the engagement situation. 2 a cbc   = σ σ (1) the burst count (bc) equation is derived to enhance the efficacy of chaff. it takes into consideration the fighter’s rcs (σa), the rcs of single chaff (σc), and the number of chaffs to be dispensed. eq. (1) calculates the number of chaffs to be dispensed by rounding up to the nearest integer the result of multiplying the fighter’s rcs (σa) by the coefficient of 2 and then dividing that value by the rcs of single chaff (σc). the coefficient of 2 represents the minimum ratio of chaff rcs to fighter rcs required to deceive ttr [10]. 180 90 180 80 180 80 180 90 170 90 170 80 170 80 170 90 170 90 170 80 170 80 170 90 180 90 180 80 180 80 180 90 a a e a e a e a e a e a e a e a e a e a e a e a e a e a e a e a e − − − − − − − − − − − − − − − −                 = ⋯ ⋯ ⋮ ⋮ ⋮ ⋮ ⋮ ⋯ ⋯ σ (2) the fighter’s rcs (σa) is represented as a two-dimensional matrix, with azimuth angles (a) ranging from -180 to 180 degrees and elevation angles (e) ranging from -90 to 90 degrees, as shown in eq. (2). however, when calculating the burst count (bc) using eq. (1), only the exposed rcs value is necessary. this value is adjusted based on the fighter’s yaw angle (y), pitch angle (p), and slant angle (s) with respect to the ttr. ( ) ( )180 10, 90 10a a y s p  = − − −σ σ (3) eq. (3) represents the formula for the exposed rcs (σa) in the two-dimensional matrix as a function of the fighter’s yaw angle (y), pitch angle (p), and slant angle (s) between the ttr and the fighter. eq. (3) provides information on how the fighter's orientation affects the rcs, covering a range of -180 to 180 degrees for azimuth angles (a) and -90 to 90 degrees for elevation angles (e). additionally, the equation also takes the adjustments made to the rcs based on the yaw, pitch, and slant angles of the fighter. ( )0.5c cmaxrcs br t= × ×σ σ (4) the rcs of chaff (σc) is determined by multiplying the maximum rcs (σmaxrcs) of chaff for each frequency band by the chaff's bloom rate (br) over the time (tc) it takes for the fighter to pass through the rrc, as shown in eq. (4). to account for the difference in polarization between the chaff and the ttr’s antenna, a coefficient factor of 0.5 is applied [9]. this equation considers various factors, such as the maximum rcs of chaff, the chaff’s bloom rate, and the time it takes for the fighter to pass through the rrc, to calculate the rcs of chaff. proceedings of engineering and technology innovation, vol. 24, 2023, pp. 22-28 24 20.17 maxrcs n=σ λ (5) the maximum rcs of chaff (σmaxrcs) can be calculated by multiplying the square of the wavelength, the number of dipoles (n), and the coefficient factor 0.17 [11]. fig. 1 shows the change in bloom rate (br) of chaff over time, with the curved graph plotting the percentage of the maximum rcs achieved by chaff against time in seconds since being burst. the graph shows that after 0.5 seconds of being burst, chaff has a bloom rate of 60% compared to its maximum rcs [12]. fig. 1 the bloom rate of chaff ct l v= (6) tc is the time it takes for a fighter to pass through the length (l) of the rrc, calculated by dividing l by the speed (v) of the fighter. the calculation of l depends on whether the angle at which the fighter will pass through rrc (φ) is greater than or less than the slant angle of rrc (φr), assuming the fighter passes through the central point of the rrc. fig. 2 shows the rrc, l, φ, φr, and the fighter passing through rrc depending on the engagement situation. fig. 2 engagement situation ( )1cos cos cos cos sin sins y p s p−φ = − (7) ( )1tanr e d −φ = (8) the angle at which a fighter will pass through rrc (φ) is determined by the slant angle (s) between the ttr and fighter, the yaw angle (y), and the pitch angle (p). the slant angle of rrc (φr) is determined by the cylindrical section (e) of rrc and the range gate width (d) of the ttr. the range gate width of the ttr (d) is typically 1.2 times the pulse width. ( ) ( ) 2 22 2cos sin az el azel sr sr e sr sr × = + θ θ θ θ θ θ (9) proceedings of engineering and technology innovation, vol. 24, 2023, pp. 22-28 25 eq. (9) represents the formula for the cylindrical section (e) of rrc, and it includes variables such as the slant range (sr) between the ttr and fighter, the azimuth/elevation beamwidth (θaz, θel) of ttr’s antenna. other variables used in the equation are calculated based on the pitch angle (p), yaw angle (y), and slant angle (s) of the fighter [13]. detailed information about these variables can be found in table 1. table 1 variables needed for calculating the cylinder section variables calculating cos�� �cos�sin �� �cos� sin �� − �cos sin� + sin cos�cos �� sin�� �cos sin� + sin cos�cos �� �cos� sin �� − �cos sin� + sin cos�cos �� to calculate the length (l) of the rrc that a fighter will pass through, two cases are considered by comparing φ and φr. the first case occurs when a fighter enters the base of the rrc and passes through the opposite base. the second case occurs when a fighter enters the rrc pillar and passes through the opposite pillar. case 1: 0° ≤ φ ≤ φ� or 180 ° − φ� ≤ φ ≤ 180 ° 2 cos c d l × = φ (10) c is the speed of light. case 2: φ� ≤ φ ≤ 180 ° − φ� sin e l = φ (11) cbi t bc= (12) the burst interval (bi) is calculated by dividing the time required for the fighter to pass through the rrc (tc) by the bc. 3. simulation and results in this paper, the deception effectiveness of the chaff dispense program against ttr is confirmed using leonardo’s tactical engagement simulation software (tess). tess is a high-resolution modeling software based on matlab and simulink that simulates the engagement situation between a fighter and a weapon system, including ttr [14]. by considering various factors such as the engagement environment, ecm, fighter’s maneuverability, and miss distance between the fighter and missile, tess provides the probability of kill for the fighter [15]. since tess does not provide the tracking information during the simulation, the chaff dispensing program’s deception effectiveness is evaluated by counting the number of scenarios in which the miss distance between the fighter and the missile is within a range of 25 m [16]. the miss distance of 25 m is twice the blast radius of the warhead carried by the sa-3 missile. initially, the sa-8 missile’s warhead blast radius was considered as the basis for the modeled weapon system in tess, but it was later decided to use the sa-3’s warhead instead due to the sa-8’s warhead blast radius of only 5 meters, which was considered too short to produce meaningful results. [17]. the weapon system that engages the fighter is “s-cgn-ha-r”, provided as a basic model in tess. it tracks the target through ttr and guides a missile to the target using command guidance radar. the waveform of ttr in “s-cgn-ha-r” is composed of pulses of electromagnetic waves with a frequency of 6 ghz, transmitted at intervals of 100 μs and a duration of proceedings of engineering and technology innovation, vol. 24, 2023, pp. 22-28 26 1.5 μs. the beam width of the electromagnetic waves transmitted by the ttr is 3°. the target tracked by ttr is “fighter”, a basic model provided in tess. the rcs of “fighter” is specified in table 2. each row and column in table 2 shows the size of the rcs reflected when tracking the fighter at the exposed angle. table 2 fighter rcs as dbm2 for 6ghz el az -90 -80 -70 -60 -50 -40 -30 -20 -10 0 10 20 30 40 50 60 70 80 90 -180 38.42 34.87 13.27 0.89 6.33 8.66 9.63 7.69 7.08 21.54 9.94 2.55 0.94 -0.4 2.38 4.63 14.14 37.04 40.32 -170 36.11 20.82 2.7 0.8 13.05 10.96 8.91 2.91 0.33 8.03 1.88 -3.21 -6.57 -5.06 -2.39 -2.7 3.52 22.55 37.82 -160 36.14 19.23 0.25 -0.2 12.67 7.32 -1.31 -1.65 -6.04 1.11 -3.36 -4.39 -4.92 -9 -4.9 -3.57 1.33 21.04 37.81 -150 36.28 21.49 -0.08 -2.85 -2.09 -2.52 -4.88 -2.72 -7.85 -3.71 -5.73 -4.79 -4.97 -4.18 -7.11 -3.06 8.07 19.34 38.11 -140 36.4 15.1 5.75 -6.07 -5.88 -4.65 -3.98 -5.24 -7.07 -1.13 -8.24 -9.9 -8.56 -5.68 -5.99 -1.85 8.42 11.85 37.82 -130 36.14 18.26 9.98 0.13 -3.94 -8.2 -3.84 -6.54 -6.51 0.3 -8.35 -8.27 -6.01 -6.13 -6.19 -4.36 9.77 12.58 37.75 -120 36.1 18.83 13.13 13.24 2.09 -3.51 -1.83 -5.74 -5.83 2.36 -3.75 -8.58 -6.9 -5.57 -2.03 5.39 13.74 12.54 37.72 -110 36.14 20.79 19.11 15.31 12.31 6.53 5.8 0.54 1.94 10.89 4.4 0.48 0.62 8.55 10.99 13.39 14.47 13.83 37.7 -100 36.39 26.24 22.35 21.23 18.97 12.34 17.08 11.83 12.96 17.45 10.96 10.05 9.11 14.54 17.48 19.93 18.31 22.42 37.82 -90 36.38 29.44 28.97 24 27.33 28.89 29.22 25.85 25.12 26.45 25.96 26.87 27.35 26.82 27.65 27.69 26.9 24.88 37.9 -80 36.39 23.3 20.87 12.14 14.85 12.22 10.93 8.67 9.68 15.6 14.27 16.39 15.22 12.46 18.31 18.96 16.74 20.76 37.82 -70 36.14 17.09 10.08 11.98 8.34 10.06 9.98 6.45 2.7 6.11 10.63 10.6 7.67 11.56 14.85 11.92 12.49 16.8 37.7 -60 36.1 12.32 4.41 6.37 3.93 4.11 0.76 2.74 4.55 7 4.16 5.22 4.05 4.88 11.77 13.71 6.32 17.08 37.72 -50 36.14 13.61 7.35 2.72 2.76 3.42 -0.7 -1.05 -3.19 0.3 -7.66 0.09 3.19 1.85 4.32 10.03 9.56 18.82 37.75 -40 36.4 25.2 8.2 0.78 1.94 -2.63 -4.57 -4.53 -3.89 -3.37 -6.29 -1.68 -4.18 2.39 4.46 6.93 8.98 22.42 37.82 -30 36.28 21.07 5.13 3.68 2.31 -4.27 -5.21 -8.08 -4.84 -5.92 -6 -9.94 -1.71 1.41 5.22 4.8 10.9 27.71 38.11 -20 36.14 9.37 4.14 3.63 -3.4 -5.99 -3.99 -5.94 -5.1 -4.76 -4.76 -2.84 0.82 6.94 15.58 5.83 9.74 19.35 37.81 -10 36.11 8.51 7.17 5.36 -2.46 -8.02 -6.62 -4.6 -0.95 -2.96 -4.04 -3.77 8.91 11.1 15.85 6.7 13.14 16.89 37.82 0 38.42 17.71 8.59 8.18 1.85 0.25 -3.34 -1.16 2.2 6.38 -0.64 -3.49 9.61 9.84 4.63 9.35 16.5 19.73 40.32 10 36.11 8.4 7.15 5.36 -2.35 -7.57 -6.45 -4.36 -0.76 -3.14 -4.25 -3.97 8.93 11.09 15.86 6.68 13.1 16.9 37.82 20 36.14 9.57 4.11 3.51 -3.46 -5.69 -3.79 -5.83 -5.18 -4.89 -4.83 -2.81 0.78 6.9 15.58 5.77 9.83 19.34 37.81 30 36.28 21.02 5.16 3.72 2.34 -4.31 -5.47 -7.77 -4.62 -6.06 -6.21 -10 -2.17 1.37 5.18 4.84 10.89 27.72 38.11 40 36.4 25.2 8.26 0.79 2.16 -2.69 -5.33 -4.4 -3.63 -4.19 -7.61 -2.1 -4.9 2.17 4.35 6.87 8.92 22.42 37.82 50 36.14 13.7 7.38 2.64 2.61 3.55 -1.32 -0.64 -2.7 -0.52 -9 0.18 3.65 2.07 4.49 10.09 9.57 18.78 37.75 60 36.1 12.41 4.45 6.35 3.88 4.05 0.77 2.97 4.44 6.43 4.27 5.27 4.08 5 11.69 13.66 6.33 17.09 37.72 70 36.14 17.03 9.92 12.06 8.34 10.04 9.95 6.64 1.34 6.14 10.63 10.66 7.71 11.56 14.85 11.86 12.44 16.91 37.7 80 36.39 23.33 20.92 12.37 14.09 12.59 11.04 8.6 9.79 15.78 14.44 16.52 15.35 12.19 18.36 18.96 16.88 20.83 37.82 90 36.38 29.49 28.97 24.09 27.26 28.92 29.45 25.86 25.11 26.46 25.85 26.89 27.32 26.85 27.84 28.01 27.02 24.86 37.9 100 36.39 26.23 22.4 21.3 18.99 12.32 16.98 11.75 12.85 17.46 10.98 9.84 8.93 15.26 17.49 20.05 18.34 22.47 37.82 110 36.14 20.78 19.1 15.3 12.47 6.46 5.58 0.39 2.04 10.69 4.42 0.26 0.91 8.4 10.78 13.45 14.53 13.89 37.7 120 36.1 18.87 13.18 13.28 2.36 -2.66 -2.14 -5.64 -5.29 2.3 -3.59 -8.13 -6.49 -5.78 -2.02 5.34 13.71 12.64 37.72 130 36.14 18.26 9.99 0.1 -3.34 -7.97 -4.51 -6.12 -7.44 0.18 -8.66 -8.29 -5.86 -6.38 -6.46 -4.27 9.76 12.58 37.75 140 36.4 15.11 5.75 -5.79 -5.54 -4.79 -3.97 -4.85 -7.7 -1.34 -7.76 -8.45 -7.69 -5.67 -6.09 -1.9 8.46 11.88 37.82 150 36.28 21.47 0.01 -2.51 -2.21 -2.92 -5.98 -2.92 -8.09 -3.68 -5.7 -5.06 -4.79 -3.88 -6.8 -3.1 8.08 19.34 38.11 160 36.14 19.23 0.3 -0.21 12.65 7.31 -1.41 -1.68 -6.18 1.08 -3.33 -4.45 -4.92 -9.2 -4.72 -3.64 1.35 21.03 37.81 170 36.11 20.82 2.71 0.77 13.04 10.98 8.98 3.14 0.56 7.92 1.87 -3.06 -6.53 -5.2 -2.38 -2.72 3.53 22.53 37.82 180 38.42 34.87 13.27 0.89 6.33 8.66 9.63 7.69 7.08 21.54 9.94 2.55 0.94 -0.4 2.38 4.63 14.14 37.04 40.32 for each engagement simulation, the yaw angle (y), pitch angle (p), altitude (a), speed of fighter (v), and distance (r) between the weapon system and fighter are varied. the chaff dispensed from the fighter is the ale-40, provided as a basic model in tess. the ale-40 employs three sets of dipoles listed in table 3 and has a σmaxrcs of 325.5m2, as calculated using eq. 5. default values are used for other variables in the weapon system, target, and chaff. table 3 three sets of dipoles resonant at 6, 8, and 10 ghz [18] dipole length resonant frequency number of diploes 24.1 mm 6.0 ghz 767,000 18.0 mm 8.0 ghz 1,534,000 14.4 mm 10.0 ghz 2,301,000 proceedings of engineering and technology innovation, vol. 24, 2023, pp. 22-28 27 the engagement simulations varied the values of y, p, a, v, and r for each scenario, with their minimum, interval, and maximum values shown in table 4. the simulation was repeated 8,400 times until all variables reached their maximum values with an increasing interval from the minimum values. in addition, the fighter dispensed chaff dispense program calculated using eqs. (1)-(12) with each variable in each simulation. in some scenarios, the chaff dispense program’s bc was calculated as a single burst using the variables because the exposed fighter rcs was lower than one chaff’s rcs. in such case, 2 chaffs were applied as the bc, and half of tc was applied as the bi. table 4 the variables in engagement simulation y (yaw angle) p (pitch angle) a (altitude) r (distance) v (speed of fighter) min 0º -30º 1000 m 2000 m 150 m/s interval 20º 10º 500 m 1000 m 50 m/s max 180º 30º 3000 m 7000 m 300 m/s as a result of the simulation, it was confirmed that in 41 out of the total 8,400 engagement scenarios, the missile and the fighter had a miss distance within 25m. the miss distance for each engagement scenario is shown in fig. 3. the x-axis of the graph represents the engagement simulation number, and the y-axis represents the miss distance. fig. 3 miss distance of each scenario in the 41 cases where the miss distance was within 25m, 39 of them were identified as having a yaw angle (y) of 0 degrees and a pitch angle (p) lower than 0 degrees. further analysis of these 39 cases revealed that the missile was moving towards the chaff cloud and the ttr was also tracking the chaff cloud. however, simulations were terminated due to tess’s algorithm determining that the missile could not close in on the target, as the fighter had overlapped with the missile’s trajectory. as a result, it can be concluded that the chaff dispense program was successful in deceiving ttr in most engagement scenarios. 4. conclusions the study developed a methodology for calculating a chaff dispense program for deceiving ttr by considering various factors such as the fighter, chaff, ttr, and engagement situations. the effectiveness of the chaff dispense program was confirmed through simulations using tess. the results show that the proposed method can effectively deceive ttrs during engagements. however, the study also identified limitations in the calculation methodology, such as the exclusion of rolling angle from the 3-d angle of aircraft maneuvering for fighter rcs calculation and the lack of consideration for the difference in chaff rcs based on the dispensing order. the calculation results also varied greatly depending on the variables, making it difficult for pilots to calculate and dispense chaff effectively during an engagement. therefore, further research is needed to optimize the chaff dispense program and address these limitations to improve its effectiveness in various engagement situations and variables. proceedings of engineering and technology innovation, vol. 24, 2023, pp. 22-28 28 acknowledgments this work was supported by a grant from hanseo university in 2022. conflicts of interest the authors declare no conflict of interest. references [1] y. li, s. quan, d. xiang, w. wang, c. hu, y. liu, et al., “ship recognition from chaff clouds with sophisticated polarimetric decomposition,” remote sensing, vol. 12, no. 11, article no. 1813, june 2020. [2] b. r. mahafza, s. c. winton, and a. z. elsherbeni, handbook of radar signal analysis (advances in applied mathematics), 1st ed., new york: chapman and hall/crc, 2021. [3] m. siddiq, “deployment of chaff in centroid mode against anti-ship missiles using a variable azimuth and elevation launcher,” ph.d. dissertation, department system engineering, naval postgraduate school, monterey, california, september 1989. [4] j. žák, m. vach, and f. dvořáček, “advanced chaff usage in modern ew,” ieee radar methods and systems workshop, pp. 56-59, september 2016. [5] x. wu, z. qi, and t. long, “research on applications of chaff,” 8th international conference on signal processing, vol. 4, november 2006. [6] h. u. shengliang, w. u. lingang, j. zhang, and j. xu, “research on chaff jamming recognition technology of antiship missile based on radar target characteristics,” 12th international conference on intelligent computation technology and automation, pp. 222-226, october 2019. [7] d. y. lee, j. s. kim, and d. w. seo, “analysis of effect of coherent and incoherent components on rcs of chaff cloud,” journal of advanced marine engineering and technology, vol. 46, no. 3, pp. 128-134, june 2022. (in korean) [8] g. s. chae, j. s. lim, and y. h. kim, “a rcs investigation of multiple chaff clouds using probability distribution characteristics,” journal of convergence for information technology, vol. 7, no. 2, pp. 37-42, april 2017. (in korean) [9] l. b. v. brunt, applied ecm: volume 1, 1st ed., madison: ew engineering, 1978. [10] r. e. fitts, “the strategy of electromagnetic conflict,” air force academy colo, february 01, 1979. [11] w. yu, m. kang, x. wang, j. luo, and y. sun, “modeling and analysis of fire control radar capabilities against passive jamming,” man–machine–environment system engineering: proceedings of the 19th international conference on mmese, pp. 485-491, august 2019. [12] handbook for chaff, leonardo, 2022. [13] handbook for employment of chaff, tracor. [14] d. c. shin, c. h. lee, d. h. kim, y. l. rhie, and y. h. choi, “a study on the integration method of threat environment engineering models in electronic warfare sils system,” proceedings of symposium of the korean institute of communications and information sciences, pp. 295-296, february 2020. (in korean) [15] c. h. park, “development of an electromagnetic wave service and modeling of an electromagnetic wave transmitting/receiving environment for open architecture radar electronic warfare simulations,” m.s. dissertation, department aerospace engineering, sejong university, seoul, february 2019. (in korean) [16] j. i. jo, j. h. lee, and j. m. ahn, “a study on the jamming technique for monopulse missile using tess(tactical engagement simulation software),” proceedings of symposium of the korean institute of communications and information sciences, pp. 707-708, june 2017. (in korean) [17] o. widlund, s. nais, and j. hawkes, janes land warfare platform-artillery & air defence 2022-2023, coulsdon, jane’s information group, 2022. [18] b. c. f. butters, “chaff,” iee proceedings f: communications radar and signal processing, vol. 129, no. 3, pp. 197201, june 1982. 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 (http://creativecommons.org/licenses/by/4.0/). 2__peti#6452__15-24 proceedings of engineering and technology innovation, vol. 18, 2021, pp. 15-24 optimum combinations of ggm and gdem models for precise national geoid modelling essam mohamed al-karargy*, gomaa mohamed dawod department of physical geodesy, survey research institute, national water research center, cairo, egypt received 21 september 2020; received in revised form 17 december 2020; accepted 27 january 2021 doi: https://doi.org/10.46604/peti.2021.6452 abstract this study aims to develop a local geoid model (lgm) for egypt to determine the optimal combinations of global models with global navigation satellite systems (gnss/levelling) data. a precise national geodetic dataset, four global geopotential models (ggms), and three global digital elevation models (dems) have been utilized. hence, twelve gravimetric lgms have been developed using the least-square collocation (lsc) method fitted to gnss/levelling data and judged over 100 checkpoints. results revealed that improvements in local geoid accuracy are attributed mainly to ggms models representing the long wavelength of the earth's gravitational field. regarding dems, the accuracy of lgms does not significantly depend on the utilized dem. based on the available data, the attained optimum geoid of egypt has been developed with a standard deviation, equals 0.129 m. keywords: ggm, dem, gnss, geoid, egypt 1. introduction gnss play a vital role in surveying geodetic activities worldwide. it produces geodetic or ellipsoidal heights reference to the surface of an ellipsoid, while most surveying and mapping activities depend on the orthometric heights referenced to the mean sea level (msl) datum. hence, height conversion of gnss heights to msl heights necessitates a geoid model to define the vertical separation, geoid heights, between those two surfaces over a specific region. geoid modelling, on a national scale, has been a crucial task for geodetic communities in several countries worldwide in the last few decades. for example, the u.s national geodetic survey (ngs) is currently developing a time-dependant geoid model for the entire usa to be completed by 2022 [1]. similarly, an updated geoid model for the united kingdom (uk) and ireland has been developed to increase accuracy and homogeneity of height transformation across this region [2]. additionally, the french institut national de l’information géographique (ign) is developing a new precise geoid model for france that could achieve a millimeter level of accuracy in height transformation [3]. more recently, the geospatial information authority (gia) of japan is developing a new refined gravimetric geoid model on a 1x1.5 arc-second grid [4]. ggms constitute a major factor in geoid modelling since they provide the long and medium wavelengths of the earth's gravitational field. since the mid-1960s, numerous ggms have been developed and being applied in geoid modelling. consequently, the evaluation of ggms in a specific region has been extensively investigated by geodetic studies in the last few decades. for example, fernádez et al. [5] and el-ashquer et al. [6] have utilized free-air gravity anomalies and gnss/levelling datasets to study the selection of optimal ggm in costa rica and kuwait respectively. additionally, gomez et al. [7] have utilized mean water level data in a lake in argentina as a validation tool for ggm assessment. since the earth geopotential model 2008 (egm2008) is the most well-known ggm worldwide, its precision performance has been * corresponding author. e-mail address: essamalkrargy@gmail.com tel.: +002-10-66361763; fax: +002-02-33933536 proceedings of engineering and technology innovation, vol. 18, 2021, pp. 15-24 investigated in several countries such as egypt [8] and sweden [9]. on the other hand, global dems (gdems) have been utilized in geoid modelling for the evaluation of the terrain effects. so, the optimum dem is also significant in such a task particularly in countries with no published precise dem. in egypt, several geoid development tryouts have been carried out in the last few decades. alnaggar [10] has developed the first pioneer national-scale geoid. dawod [11] has developed a national geoid model based on the data of the egyptian national standardization network of 1997 (engsn97) along with gps/levelling data. also, abd-elmotaal [12] has developed a gravimetric geoid model utilizing high-degree tailored reference geopotential model. recently, dawod and abdel-aziz [13] have studied the utilization of a geographic information systems (gis) technique in geoid modelling. the traditional manner of investigating the accuracy and reliability of ggms and dems is the assumed point-wise comparison. in this approach, a specific ggm is evaluated over checkpoints with known geoidal undulations and the statistical indicators provide a measure for its reliability in geoid modelling [6]. similarly, a particular dem is judged over known stations with available orthometric heights. a novel approach is proposed in the current study for deciding which ggm and dem models are optimum for utilization in developing a national geoid model. accordingly, this study aims at investigating the optimum combination of some selected ggms and gdems models in developing a precise local geoid model of egypt based on the available local heterogeneous geodetic datasets. 2. methodology the current study employs four ggms models and three gdems models in developing twelve 5'x5' local geoid models for egypt. gdems models have been utilized since there is no local precise dem published for egypt [14]. the utilized ggms models include xgm2019e_2159, egm2008, eigen-6c4, and geco. these models have been selected since they are the only (to date) high-resolution ggms with a degree of 2190. concise descriptions of those models are presented in the following subsections: 2.1. global geopotential models (ggms) (1) xgm2019e_2159: a recent ggm combines 15' terrestrial gravity anomalies datasets over land, 1' mean sea surface database over oceans, 5' goco06s ggm, and the earth2014 topographic model. it has been developed to degree 2190 (utilized herein) and later extended to degree 5400 [15]. (2) geco: a global gravity model utilizes the goce satellite-based gravity data to improve the accuracy of the egm2008 ggm in low and medium frequencies. geco is developed, in 2015, up to degree and order 2190 [16]. geco is the most-recent published high -resolution ggm up till 2017. (3) eigen-6c4: a model released in 2014, that utilizes satellite tracking data (from the laser geodynamics satellite: lageos, the gravity recovery and climate experiment: grace, and the: gravity-field and steady-state ocean circulation explorer: goce missions) along with a global surface gravity anomaly grid and altimetry data. the model is up to 2190 degree, developed by both the germany german research center for geosciences (gfz) and the french centre national d'études spatiales (cnes) [17]. (4) egm2008: an integrated ggm developed in 2008 by the u.s. national geospatial-intelligence agency (nga) up to 2190 degree, based on grace-based satellite tracking data, terrestrial gravity data, and altimetry data. it was a milestone in ggm development since its preceding model did not exceed 360 maximum degrees [18]. table 1 shows the characteristics of those selected global ggms. 16 proceedings of engineering and technology innovation, vol. 18, 2021, pp. 15-24 table 1 characteristics of selected ggms ggm year degree utilized data average accuracy over 12035 gnss/levelling stations (m) xgm2019e_2159 2019 2190 a, g, s(goco06s), t 0.236 geco 2015 2190 egm2008, s(goce) 0.237 eigen-6c4 2014 2190 a, g, s (goce, grace, lageos) 0.236 egm2008 2008 2190 a, g, s(grace) 0.240 where: a = altimetry data, g = ground geodetic data, s = satellite data, t = topographic models (after [19]) 2.2. global digital elevation models (gdem) on the other hand, the utilized gdems models, herein, include srtm1, aster, and ace2 models. table 2 summarizes the characteristics of those selected global dems, including: (1) srtm: the shuttle radar topography mission (srtm) is a global dem, that has two versions: srtm1 with a spatial resolution of 1 arc second, i.e., approximately 30 meters, and srtm3 model with a 3 arc second resolution. srtmgl1 v. 3 [20] has been utilized herein. (download from e.g. https://earthexplorer.usgs.gov/). (2) ace2: the altimeter corrected elevations, version 2 is a global digital elevation model created by synergistically merging the srtm data set with satellite radar altimetry within the region bounded by 60°n and 60°s. ace2 was developed at resolutions of 3, 9 and 30 arc-seconds, and 5 arc-minutes [21]. the 3" ace2, utilized herein (downloaded from https://sedac.ciesin.columbia.edu/data/set/dedc-ace-v2). (3) aster: the advanced spaceborne thermal emission and reflection radiometer (aster) is a 1 arc-second global dem [22]. the aster version 3 has been publically released in 2019 (download from e.g. https://asterweb.jpl.nasa.gov/gdem.asp). table 2 shows the characteristics of those selected global dems. table 2 characteristics of the utilized global dems dem released year spatial resolution arc second m aster v.3 2019 1" ∼ 30 ace2 2019 3" ∼ 90 srtmgl1 v.3 2019 1" ∼ 30 the topography of egypt, as derived from ace2 gdem for example, is depicted in fig. 1. it can be noticed that the egyptian terrain heights range from -137 m to +2603 m with an average of 302 m. it can be realized that the mountainous areas are located only in the middle of sinai peninsula, eastern deserts along the red sea, and in southwestern borders. the flat topography generally reduces the effects of terrain corrections in geoid modelling. for each combination, a 5'x5' gravimetric geoid is developed first, and then fitted to gnss/levelling dataset. fig. 1 topography of egypt 17 proceedings of engineering and technology innovation, vol. 18, 2021, pp. 15-24 the least-square collocation is one of the most imperative tools for gravity field modelling due to its high capability of dealing with heterogeneous input features. the basic mathematical form of lsc, in a vector form, could be written as [23]: x au s n= + + (1) where � is the vector of measurements, � is the vector of the noise component in the measurements, � is the vector of signal components in the measurements to be predicted, and � is the vector of unknown parameters to be estimated. thus, lsc combines adjustment, filtering and prediction of the input earth's gravity field parameters to estimate geoidal heights and other geodetic related values [ibid]. solving that model requires constructing a covariance matrix that can be estimated based on the variances of the required noise and unknown parameters. the accuracy of lsc solution, as represented by the error covariance matrix, depends on the original covariance matrices or, in other words, on the accuracy of the input datasets. the processing, in the current study, has been carried out using the lsc-based gravsoft package [24]. with this software, the effects of a ggm and a dem are first removed from the input datasets of the earth's gravitational field to obtain a residual field, second computations of predicted quantities and error estimates are carried out, and finally re-adding the effects of both ggm and dem to the attained residual parameters [23]. additionally, an empirical covariance function is estimated based on the residual gravity anomaly data. the package decomposes the geoid undulations (n), in the remove-compute-restore processing strategy, into three components as [24]: ggm g tc n n n n∆= + + (2) where ���� provides the long to the medium wavelength of the gravitational field of the earth as computed, �∆ represents the medium to short local wavelength deduced from local datasets, and � � represents the terrain corrections or the topography contribution as represented by a certain dem. the three components of the geoidal undulation, in the last equation, are computed by the following equations [23]: max 2 0 ( ) ( ) (( cos sin ) cos ) n n n ggm nm nm nm n m gm a n c m s m p r r − = = = +∑ ∑ λ λ ϕ γ (3) ( )( ) 4 g ggm r n s g g d∆ = ψ ∆ − ∆∫∫ σ ο σ πγ (4) 3 ( , , )( ) ( , , )p tc h p h p p p n x y z h z g dxdydz r x x y y h z = − − − −∫∫∫σ ρ (5) where � is the degree of the ggm model, max-n is the maximum degree of the ggm model, m is the maximum order of the model, is the normal gravity of the reference ellipsoid, r is the geocentric radius of the computation point, g is the newtonian gravitational constant, m is the mass of the earth, r is the mean radius of the earth, a is the semi-major axis, � is the geocentric latitude, � is the geocentric longitude, ∆� is the gravimetric observations given as free-air gravity anomalies, ∆���� is the gravity anomalies computed by a specific ggm, �(ψ) is the stokes' function,σ is a differential surface area on a unit sphere, σ° is the differential surface area on the computational cap, ��̅� and ��̅� are the fully normalized harmonic coefficients, �° is the surface atmosphere potential, �° is the sea-level radius of the earth, �(�, �, �) is the topographical density at running point, ℎ! and ℎ are the computational and running point respectively, and "�� is the fully normalized associated legendre polynomial. 18 proceedings of engineering and technology innovation, vol. 18, 2021, pp. 15-24 next, all geoid combinations have been investigated using the arcgis 10 software, to interpolate the attained geoidal heights at the 100 checkpoints and compare them to their corresponding known values. to compare the performance of each attained lgm, three statistical measures have been used: standard deviation, average, and range of the discrepancies between the known and interpolated geoid undulations. three descending ranks (r1, r2, r3) are given to each lgm over a scale of ten, where ten represents the minimum value of the statistical measure and one represents the maximum value. weights of those ranks have been selected as 4, 3, and 3 respectively. hence, a weighted mean rank (r) is computed for each lgm, using its corresponding three individual rank measures (ri) and their assigned weights (wi) to represent a unique indicator of its accuracy and to serve as a base of comparison between the twelve accomplished lgms, as: i i i r w r w = ∑ ∑ (6) fig. 2 presents the overall steps utilized in the data processing of the proposed methodology. fig. 2 flow chart of data processing 19 proceedings of engineering and technology innovation, vol. 18, 2021, pp. 15-24 3. available data regarding the available terrestrial data over egypt, the study utilized two main types: terrestrial gravity data and gnss/levelling data (fig. 3). the first-order egyptian national gravity networks of both 1997 and 1977 contain 247 measured gravity points [11]; and 1100 gnss/levelling stations all over egypt, that were observed by the survey research institute (sri) in various projects over the last ten years [25]. it is worth mentioning that the average accuracy of the egyptian national gravity standardization network of 1997 (engsn97) is ±0.02 mgal, while the corresponding value of the national gravity standard base network of 1977 (ngsbn77) is ±0.08 mgal [11]. on the other hand, the utilized gnss/levelling points have been surveyed using the first-order levelling and the first-order gnss geodetic network standards, and their average accuracy could be estimated as ±3 4 cm [25]. ten percent of the known gnss/levelling stations have been reserved as checkpoints to estimate the external accuracy of each developed lgm. it should be noticed that such local datasets may not adequate in number and they are not homogeneously distributed over egypt. however, such data are the most-recent accurate geodetic measurements existing over the country [ibid]. fig. 3 available terrestrial data in egypt 4. results and discussion table 3 overall accuracy of different ggms and dems over checkpoints (m) (a) overall accuracy of lgms based on different ggms over checkpoints: ggm average standard deviation (m) improvements % xgm2019e_2159 ±0.137 17% egm2008 ±0.140 geco ±0.159 eigen-6c4 ±0.157 (b) overall accuracy of lgms based on different dems over checkpoints: dem average standard deviation (m) improvements % ace2 ±0.146 2% aster v.3 ±0.148 srtmgl1 v.3 ±0.150 the first investigation step has been carried out in the overall perspective to get the big picture about the contribution of ggms and dems in geoid modelling for egypt. hence, the average standard deviation over checkpoints for each ggm-based lgm, averaging its performance using all dems, has been computed. similarly, the mean standard deviation over check points for each dem-based lgm, averaging its performance using all ggms, has been also computed. table 3 presents the accomplished findings that reveal two imperative remarks. first, local geoid accuracy levels range from ±0.137 m using the 20 proceedings of engineering and technology innovation, vol. 18, 2021, pp. 15-24 xgm2019e_2159 model to ±0.159 m using the geco ggm. it emphasizes that significant improvements of 17% approximately in local geoid accuracy are attributed to the utilization of ggms models representing the long wavelength of the earth gravitational field. regarding the contribution of dems, it can be noticed from that table that the overall accuracy of lgms does not significantly depend on the utilized dem. that could be attributed to the almost flat topography of egypt as depicted in fig. 1, which reduces the effect of the topographic correction in the geoid modelling process. furthermore, the statistical characteristics of the developed twelve lgm of egypt are tabulated in tables 4-7. several imperative remarks could be perceived from those tables. first, it can be recognized, from the first column in all tables, that the table 4 chrematistics of xgm2019e_2159 ggm differences over checkpoints (m) measure/lgm no. utilized gdem ace2 aster v.3 srtmgl1 v.3 1 2 3 minimum -0.276 -0.279 -0.275 maximum 0.271 0.430 0.429 range 0.547 0.710 0.704 average 0.003 0.005 0.006 standard deviation ±0.129 ±0.140 ±0.141 table 5 chrematistics of geco ggm differences over checkpoints (m) measure/lgm no. utilized gdem ace2 aster v.3 srtmgl1 v.3 4 5 6 minimum -0.292 -0.301 -0.293 maximum 0.649 0.607 0.663 range 0.940 0.908 0.956 average 0.016 0.015 0.017 standard deviation ±0.161 ±0.156 ±0.162 table 6 chrematistics of eigen-6c4 ggm differences over checkpoints (m) measure/lgm no. utilized gdem ace2 aster v.3 srtmgl1 v.3 7 8 9 minimum -0.621 -0.633 -0.633 maximum 0.369 0.375 0.376 range 0.989 1.008 1.008 average 0.010 0.010 0.010 standard deviation ±0.156 ±0.157 ±0.157 table 7 chrematistics of egm2008 ggm differences over checkpoints (m) measure/lgm no. utilized gdem ace2 aster v.3 srtmgl1 v.3 10 11 12 minimum -0.295 -0.317 -0.320 maximum 0.322 0.327 0.328 range 0.618 0.644 0.647 average 0.019 0.018 0.019 standard deviation ±0.140 ±0.140 ±0.140 21 proceedings of engineering and technology innovation, vol. 18, 2021, pp. 15-24 aec2 gdem generally produce the minimum range of variations over checkpoints. so, it might be concluded that aec2 performs a little bit better than the other two gdems over egypt, at least in geoid modelling applications. moreover, it can be revealed that the standard deviation, as an accuracy indicator, ranges from ±0.129 m for lgm 1 to ±0.162 m for lgm 6. for the range of discrepancies over known checkpoints, the developed lgms perform diversely and their variations range from a minimum of 0.547 m for lgm 1 to 1.008 m for lgm 8 and 9. however, all these levels of accuracy are not accurate enough for high-precision gnss-based geodetic and mapping applications in egypt. consequently, this signifies the critical need for collecting more measured geodetic datasets in terms of number and homogenous distribution over the country. to come up with the optimum lgm for egypt, the twelve lgms have been compared in terms of standard deviation, average, and ranges of their geoid undulations' variations over checkpoints. as mentioned earlier, a descending rank is assigned to each lgm, on a scale of ten, for each statistical measure. subsequently, eq. 2 is applied to compute the weighted mean rank for each model. the attained results are tabulated in table 8. it can be realized that the overall rank of the lgms, on a scale of ten, varies from 2.9 for lgm 6 to 10.0 for lgm 1. consequently, based on the currently available local geodetic datasets, lgm 1 is the optimum model for egypt. it has been developed based on the xgm2019e_2159 ggm and the ace2 gdem. its geoid undulations (fig. 4) range from 8.948 m to 21.088 m with a mean of 14.106 m. it is recommended to use this local geoid, for the time being, for gnss-based surveying and mapping applications in egypt. table 8 ranks of developed lgms for egypt lgm no. r1 r2 r3 r 1 10 10 10 10.0 2 9 9 1 7.2 3 8 8 2 6.8 4 3 5 5 4.0 5 7 6 4 6.1 6 1 4 6 2.9 7 6 7 3 5.7 8 4 7 3 4.7 9 4 7 3 4.7 10 9 2 8 6.7 11 9 3 7 6.8 12 9 2 9 6.9 where r1, r2, r3 are the individual ranks for standard deviation, average, and range respectively. r is the weighted overall rank. fig. 4 developed optimum local geoid of egypt 22 proceedings of engineering and technology innovation, vol. 18, 2021, pp. 15-24 5. conclusions local geoid modelling involves a local geodetic dataset, an optimum ggm, and a gdem. a precise geodetic dataset, comprising 247 terrestrial gravity points and 1000 gnss/levelling stations, has been utilized in the current research study using the least-square collocation geoid modelling method. rather than the classical point-wise method of investigating the accuracy of ggms and dems, the current research proposed a new approach for deciding which ggm and dem models are optimum for utilization in developing a national geoid model. thus, four ggms (namely xgm2019e_2159, egm2008, eigen-6c4, and geco) and three gdems (aster v.3, ace2, and srtmgl1 v.3) have been utilized and twelve 5'x5' gravimetric geoid models have been developed for egypt. such models have been, then, fitted to gnss/levelling stations and then judged over 100 checkpoints with known geoidal heights. accomplished findings revealed that the utilized gdems provide comparable results from a precision perspective, which could be due to the flat topography of egypt. thus, it can be concluded that the terrain effects in geoid modelling are quite small. additionally, it has been found that the aec2 model performs relatively better than the other two investigated gdems over egypt. furthermore, it has been found that the standard deviation of the twelve lgm ranges from ±0.129 m to ±0.162 m. on the other hand, the undulations discrepancies over known checkpoints vary from 0.547 m to 1.008 m. concerning the general performance of all investigated ggms, it has been concluded that the xgm2019e_2159 produces the best standard deviation (±0.129 m) while the geco ggm produces the worst value ±0.162 m). nevertheless, all these levels of accuracy cannot be as accurate as needed for high-precision gnss-based geodetic and mapping applications in egypt. as a result, there is a vital need for collecting extra local geodetic datasets in terms of number and homogenous distribution over the country. all specialized governmental, military and private organizations have to cooperate together to establish a recent integrated national dataset for geodetic and environmental applications. based on statistical ranking, it has been found that the lgm developed using the xgm2019e_2159 ggm and the ace2 gdem, could be considered as the optimum model for egypt for the time being. therefore, it is recommended to use this local geoid for gnss-based surveying and mapping applications in egypt. acknowledgement the authors appreciated the revision of the manuscript and the valuable comments given by the late prof. dalal s. alnaggar who passed away during the development of this research study. may god blesses her soul. conflicts of interest the authors declare no conflict of interest. references [1] k. m. ahlgren, s. a. holmes, x. li, y. m. wang, and m. a. youngman, “geoid modeling at noaa’ s national geodetic survey as 2022 approaches,” fig working week, helsinki, finland, 2017. [2] m. greaves, p. downie, and k. fitzpatrick, “osgm15 and ostn15: updated transformations for uk and ireland,” geomatics world, no. 32/33, pp. 1-5, 2016. [3] f. duquenne and a. coulomb, “the french approach to modernize its vertical reference,” finance information group working week, may 2015, pp. 17-21. [4] k. matsuo and y. kuroishi, “refinement of a gravimetric geoid model for japan using goce and an updated regional gravity ffield model,” earth, planets and space, vol. 72, no. 1, pp. 1-18, march 2020. [5] a. v. fernádez, o. l. castro, and j. g. león, “geoid heights in costa rica, case of study: central pacific zone,” revista ingenieria, vol. 30, no. 1, pp. 1-20, november 2020. 23 proceedings of engineering and technology innovation, vol. 18, 2021, pp. 15-24 [6] m. el-ashquer, h. al-ajami, a. zaki, and m. rabah, “study on the selection of optimal global geopotential models for geoid determination in kuwait,” survey review, vol. 52, no. 373, pp. 373-382, may 2019. [7] m. e. gomez, r. perdomo, and d. del cogliano, “validation of recent geopotential models in tierra del fuego,” acta geophysica, vol. 65, no. 5, pp. 931-943, september 2017. [8] a. a. saad and m. s. elsayed, “evaluation of goce satellite only models along the river nile,” scientific and academic publishing, vol. 8, no. 3, pp. 131-140, 2019. [9] m. eshagh and s. zoghi, “local error calibration of egm08 geoid using gnss/levelling data,” journal of applied geophysics, vol. 130, pp. 209-217, july 2016. [10] d. s. alnaggar, “determination of the geoid in egypt using heterogeneous geodetic data,” phd dissertation, faculty of engineering, cairo university, 1986. [11] g. m. dawod, “a national gravity standardization network for egypt,” phd dissertation, faculty of engineering at shoubra, zagazig university, 1998. [12] h. abd-elmotaal, “gravimetric geoid for egypt using high-degree tailored reference geopotential model,” national research institute of astronomy and geophysics journal of geophysics. special issue, pp. 507-531, december 2008. [13] g. dawod, w. resrach, and t. m. abdelaziz, “utilization of geographically weighted regression for geoid modelling in egypt,” journal of applied geodesy, vol. 14, no. 1, october, 2019. [14] t. abdel-aziz, g. dawod, and h. ebaid, “dems and reliable sea level rise risk monitoring in nile delta, egypt,” discover sustainability, vol. 1, no. 1, pp. 1-11, december 2020. [15] p. zingerle, r. pail, and t. gruber, “high-resolution combined global gravity field modelling--towards a combined d/o 10800 model,” geophysical research abstracts, vol. 21, january 2019. [16] m. gilardoni, m. reguzzoni, and d. sampietro, “geco: a global gravity model by locally combining goce data and egm2008,” studia geophysica et geodaetica, vol. 60, no. 2, pp. 228-247, march 2016. [17] c. förste, s. bruinsma, o. abrikosov, f. flechtner, j. c. marty, j. m. lemoine, et al., “eigen-6c4-the latest combined global gravity field model including goce data up to degree and order 1949 of gfz potsdam and grgs toulouse,” eguga, p. 3707, may 2014. [18] n. k. pavlis, s. a. holmes, s. c. kenyon, and j. k. factor, “the development and evaluation of the earth gravitational model 2008 (egm2008),” journal of geophysics research solid earth, vol. 117, no. b4, april 2012. [19] “the international center for global earth model,” http://icgem.gfz-potsdam.de/tom_longtime, october 12, 2020. [20] “srtmgl1 v003,” the us geological survey, https://lpdaac.usgs.gov/products/srtmgl1v003/, 2020. [21] r. smith and p. berry, “altimeter corrected elevations, version 2,” https://sedac.ciesin.columbia.edu/data/set/dedc-ace-v2, 2019. [22] “aster digital elevation model v003,” https://catalog.data.gov/dataset/aster-digital-elevation-model-v003, july 17, 2020. [23] b. hofmann-wellenhof, “moritz, h.(2005) physical geodesy”, new york: springer, 1967. [24] r. forsberg and c. c. tscherning, gravsoft: geodetic gravity field modelling programs manual, 2nd ed. denmark: national space institute, 2008. [25] e. m. al. krargy, “development of a national geoid for egypt using recent surveying data,” phd dissertation, faculty of engineering, minufiya university, egypt, 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/). 24  proceedings of engineering and technology innovation, vol. 17, 2021, pp. 01-12 application of image processing techniques for autonomous cars shaun fernandes, dhruv duseja, raja muthalagu * department of computer science, birla institute of technology and science pilani, dubai campus, dubai, uae received 16 july 2020; received in revised form 28 october 2020; accepted 17 december 2020 doi: https://doi.org/10.46604/peti.2021.6074 abstract this paper aims to implement different image processing techniques that will help to control an autonomous car. a multistage pre-processing technique is used to detect the lanes, street signs, and obstacles accurately. the images captured from the autonomous car are processed by the proposed system which is used to control the autonomous vehicle. canny edge detection was applied to the captured image for detecting the edges, also, hough transform was used to detect and mark the lanes immediately to the left and right of the car. this work attempts to highlight the importance of autonomous cars which drastically increase road safety and improve the efficiency of driving compared to human drivers. the performance of the proposed system is observed by the implementation of the autonomous car that is able to detect and classify the stop signs and other vehicles. keywords: image processing, autonomous cars, self-driving, object detection 1. introduction image processing is a field in a computer that deals with processing and extracting meaningful and useful data from images. since images represent unstructured data, they are significantly more difficult than processing structured data such as tables and forms. depending on the application, image processing can be applied in numerous fields, and via many different methods and algorithms. it is a large and growing branch that is currently in use in multiple areas such as the medical field, video processing, pattern recognition, machine/robot vision, driverless cars, etc. in a recent paper by viswanathan and hussein [1-2], much study was done in the field of image detection for autonomous cars. features such as lane detection, traffic signal detection, and speed bump detection were studied and the method of implementation also were reported in by many researchers around the world. the lane detection system prescribed here talks about a few major steps involved. selecting the region of interest (roi) and discarding unwanted data, then image pre-processing, canny edge detection, and hough transforms applied on the final output. as mentioned, these were methods originally used in lane detection, but in more recent times, spiking neural networks have become popular, being able to detect edges faster and more efficiently than traditional methods. traffic signal detection was achieved by looking for a certain shape in the image, and then looking at the color in that image if it was found. the paper also proposes a steer by wire implementation in cars, where instead of having the drive shaft directly connected to the axil, it added an electronic control unit (ecu) in between. this ecu would then take user steer as input and based on external conditions decide whether to execute the command to steer or not. using these methods would prove extremely helpful in future implementations of autonomous cars. an interesting collision detection system is proposed by guha and the team [3-4]. this system aims to focuses on other objects along the road instead of looking for lanes, such as stop signs, traffic lights, and taillights of cars ahead are looked for. upon seeing those, instructions are given to the car to stop. it has been implemented using harr cascades, which provide light * corresponding author. e-mail address: raja.m@dubai.bits-pilani.ac.in tel.: +971-562249227886-5 proceedings of engineering and technology innovation, vol. 17, 2021, pp. 01-12 2 weight and efficient solution to the problem of object detection. using multiple images of positive examples of their objectand many more negative examples, they were able to accurately classify these important objects and react accordingly. after calibrating the camera, calculated the camera tilt angle, and considered its height off of the ground, they were also able to predict the distance of the object with a fair amount of accuracy. shah, adhikari, and maheta [5-6] were also able to produce similar results while attempting to tackle the same problem. edge detection was implemented for lane detection. but for street signs like left, and right arrows, matrix matching algorithms were used instead of harr cascades, and similar results were shown. this goes to show that different algorithms may produce similar output and that each should be tested, in order toselect the best on a case by case basis. xue li [7] and the team worked on drastically improving the efficiency of the edge detection algorithm. instead of using the conventional means of canny edge detection, which many claim to be slower and at times inaccurate as a means of highlighting edges in the image, the team decided to use a spiking neural network. spiking neural networks are built to emulate real neural networks more precisely than simple mlp’s. they do this by not letting all neurons fire at the same time but instead firing only when a certain threshold is met, and they also incorporate time into the model. the output of the edges from the snn was passed through a hough transform and using the snn, xue li was able to gain accuracy and efficiency far greater than conventional methods of canny edge detection. consequently improve efficiency of the lane detection algorithm overall. once the input images are preprocessed, features that define a lane like colors and edges can be extracted using color space exploration and edge detection algorithms like canny edge detector [8]. the hough transform algorithm [8], is one of the most widely used algorithms for lane detection in previous studies. conventional vision-based lane detection techniques can be roughly divided into two main classes, feature-based techniques [9] and model-based techniques [9]. feature-based techniques mainly rely on the detection of features describing lane markings such as lane color, texture, and edges. most of the previous works used a technique like sobel filters or canny edge detectors to detect the lanes using pixel intensities and edge information with lane markings. the lane markings are being extracted using different preprocessing steps. instead of using the rgb color images directly, the researchers have converted the rgb images to other color formats. the general idea of color space transformation being that the colors most used for lane markings, yellow and white, are more distinct in other color domains. once the edges are detected, then they are combined using a lane hypothesis which can be used to filter out outliers. in general, feature-based methods have better computational efficiency and can accurately detect lanes when lane markings are clear, but quickly lose robustness depending on many factors. a weakly-supervised adversarial domain adaptation is implemented to improve the segmentation performance from synthetic data to real scenes, which consists of three deep neural networks [10]. the detection and segmentation (ds) model focus on detecting objects and predicting a segmentation map; a pixel-level domain classifier (pdc) tries to distinguish the image features from which domains; and an object-level domain classifier (odc) discriminates the objects from which domains and predicts object classes [11]. siamese fully convolutional networks (fcns) (named “s-fcn-loc”), which can consider rgb-channel images, semantic contours, and location priors simultaneously to segment the road region elaborately [11]. vision technology is implemented that allows traffic engineers to analyze traffic safety based on image processing applications [12-13]. in this paper, a system for image processing in autonomous cars has been implemented by applying a multistage preprocessing to the images for lane detection, the lanes can be detected quickly and accurately. the implementation of the lane detection model has been created and was in a working state but was not integrated with the current model. it was designed to be used as an aid for a future autonomous car model that run on real roads. the output of the model could be passed as a secondary input to the neural network to aid it in detecting lanes and making decisions proceedings of engineering and technology innovation, vol. 17, 2021, pp. 01-12 3 the remainder of the paper is organized as follows. section 2 describes the problem formulation and objective. section 3 describes design methodology. section 4 describes the lane detection by using image processing techniques. experimental results of the object detection are presented in section 5. finally, section 6 concludes our work. 2. problem formulation and objective one of the greatest causes of death, besides diseases, is car accidents. according to a recent survey [14], 1 out of 80 people, or approximately 1.2%, die in a car accident. most of these accidents can be attributed to human error. some of the major causes of road deaths are distraction, over speeding, drunk driving, and recklessness. autonomous cars aim to provide a solution to this massive problem of loss of life from car accidents. since these systems are programmed to drive efficiently and safely, they minimize, if not, eliminate the need for a human driver, they would also eliminate the aforementioned human errors. this would lead to a significantly safer driving experience and would hence decrease the road accident rate. this project aims to take a step in the right direction and move from traditional human-driven cars to fully autonomous cars that make driving much safer. this paper aims at multiple objectives. first, all hardware components of the wheel-e car have to be tested and calibrated to make sure that everything runs smoothly. the wheel-e, which is being interfaced via an arduino, had to communicate with the computer via bluetooth and a program written to give different driving instructions to the car from t he computer. this would then be tested to see if all controls sent from the computer were properly received and executed by the car. having the controls sent directly from the computer would later make it easier to gather training data for the neural network that will control the car. unfortunately, due to limitations on the processing and storage capabilities of the arduino, the idea of using the wheel-e was canceled. instead, a simple remote-controlled car was used for this paper. the remote controller and the internal wiring have to be turned on to be able to control by an arduino. wires has to be soldered onto the controller’s circuit board and then connected to the arduino. the arduino has to be interfaced to the computer via a python program, allowing controls to be sent from the computer and directly executed by the r/c car. lane detection is a very important step as the car moves along the road, it will need to know exactly where the lanes are, so it knows the boundaries and can obtain information on how to steer the car. to do this, the images sent from the car had to be cropped to display only the region of interest (roi), preprocessed to increase accuracy, canny edge detection applied on the image to only display the edges, and finally, hough transform used to detect and mark the lanes immediately to the left and right of the car. besides lanes, other objects also has to be recognized using image processing techniques: stop signs which would tell the car to wait for a few seconds. other cars in front, which pose as obstacles, also needed to be detected to avoid collisions. an algorithm to measure the distance to each of these objects. all these objects were classified using haar cascades, which is fast and efficient, and can detect objects no matter how big or small and no matter where they are placed in the image. the haar cascades need to be trained on many training images from each class, and the images has to be taken from many different points of view. a lot of images of negative examples has also been provided to properly train the classifier. the distance measurement of the classified objects was done using monocular methods. having the camera calibrated previously, and measured the width of the objects used, distance to these objects could be measured and could provide useful information for the autonomous car. the image classifiation need to be integrated along with the car. in this final stage, the autonomous car would be set up with a camera that would constantly stream live videos of the road to a computer with the image processing software installed. the lane detection system would aid the car, staying in lane and steering in the right direction. upon detecting a stop sign at a proceedings of engineering and technology innovation, vol. 17, 2021, pp. 01-12 4 certain distance the car will be told to stop for a few seconds before proceeding with caution. and detection of cars ahead would tell the car to slow down or stop completely to avoid collisions. using these images processing tools provide much-needed aid to autonomous cars in making decisions. 3. design methodology 3.1. hardware implementation of wheel-e this work initially made use of a wheel-e as its base. the wheel-e is an arduino controlled model car. the model is composed of a large chassis base, 2 dc motors, 2 large wheels, optical interrupt sensors, a small castor wheel at the front, a battery holder for 4 batteries, breadboard, dual h-bridge motor driver, arduino camera, bluetooth module, and an arduino uno as shown in fig. 1. fig. 1 arduino based wheel-e car the chassis base is a large precut acrylic base that is specifically designed with the wheel-e in mind. the 6-volt dc motors are the powerhouses of the system, they drive the car forwards. since no two motors are identical, these motors spin at different speeds even when supplied with the same amount of power, and this is a problem that needs to be solved later. these motors attach to 2 large plastic wheels, and each wheel has an optical encoder wheel. by reading how many times the slots of the optical encoder wheel have passed by the optical interrupt sensor, we can determine how much the wheel has rotated and how much distance has been covered by the car. the castor wheel is not electronically powered that means it cannot be rotated. it is simply used to provide support and stability to the car. the battery holder is capable of holding 4x 1.5-volt batteries, supplying the motors and the arduino with up to 6-volts of power. the l298n dual h-bridge motor driver is used to interface the arduino with the motors. it provides the dc motors with power, receives and executes commands for speed and pwm sent from the arduino. the main function of the h-bridge is to provide the ability to switch polarities of the wheel input and hence change the direction of rotation of the wheel from forwards to backward. the arduino camera is one of the most important parts of this project. it is used to get a live feed of images taken from in front of the wheel-e. the bluetooth module is used to relay images back to the computer and to receive directions on where and when to turn from the computer. 3.2. rectifying steering problems of wheel-e we had initially started with only the model of the wheel-e and the base code. the first thing to be done was to learn how to use and program the arduino. once the basics of arduino coding were understood, many changes need to be made. the existing model had a few flaws that have to be corrected to be able to properly work with the model in the later stages of the project. the first stage of work was to set up the hardware properly and to ensure that it work correctly. the main thing that need to be corrected was the turning of the arduino. since no two motors are alike, and many factors such as the thickness of connecting wires, build of the motors, the strength of the motor magnets, etc., affect their speed. proceedings of engineering and technology innovation, vol. 17, 2021, pp. 01-12 5 because of this, the motors were never spinning at the same speed. this posed a problem since the car would tilt to the side when it was supposed to go straight and would make sharp right turns and very wide left turns. these irregularities in turnin g would affect the car later when it was being controlled autonomously since the control algorithm would assume a perfectly stable and symmetrical turning system. to address this problem the optical interrupt sensors need to be used. since every time the wheel turned, it also turned the optical encoder wheels which had 20 holes in it, the optical interrupt sensor was able to read every time a new hole passed by the sensor. it would then send an interrupt to the arduino which would count the number of pulses it received from the sensor. the following is the general formula for calculating rpm: 1 . min no of pulses rpm times in utes pulses per turns   (1) but since the arduino provides time in milliseconds, and the number of pulses per turn is 20 (20 holes in the encoder wheel), the modified formula bellow was uses instead: 60 1000 . min 20 no of pulses rpm times in utes    (2) now being able to accurately calculate the speed that each wheel is spinning at, we could proceed to adjust the power sent to the motors. this would be used to compensate for one motor is faster than the other. this was done by lowering the pwm for the faster motor (b) till it match the speed of the slower one (a). we used a proportional method for adjusting the speed of the motors. the proportion of the speed of one motor to another was calculated. (assume a to be the slower motor) a b b b rpm rpm percentage difference rpm   (3) the pwm of the slower motor was kept at its maximum, and pwm of the faster motor was reduced at first, after that, it constantly readjusted proportional to the difference between the two motor speeds to try and make it match the exact speed of the slower motor. ( ) b b b b new pwm pwm pwm percentage difference   (4) by implementing this function, we were able to adjust the motor speed of the car to match each other with some accuracy. this function is constantly being called, hence, the car is always constantly, dynamically adjusting its speed to be able to drive better than it did before. this solution is unfortunately not enough to get the car to drive straight. the car would instead of following a straight line, veer off to one side or the other. even slight differences in a rotation of one wheel would set th e car oversteering to the opposite direction. once the speeds for the motors were adjusted and calibrated, the next step was to create a turning function. this function simply took in as arguments which direction the car should turn in and how much it should turn. turning was then achieved by having both wheels moving forward but having one wheel move a certain degree faster than the other wheel. so, if we want a sharp left turn, we make the right wheel spin much faster than the left wheel, on the other hand, if we want a wide turn right, we would make the left wheel spin lightly faster than the right wheel. even though the steering had been slightly improved, there were other unforeseen problems. these came mostly from the camera part. as the arduino is a very small and low powered microcontroller, there was a severe lack of required memory and processing power. it was unable to handle taking and storing many images, as it lacked the memory needed. since the processor was also lacking power, any pictures that were taken would take time to be processed and sent over bluetooth. this proceedings of engineering and technology innovation, vol. 17, 2021, pp. 01-12 6 caused a large lag over half a second, which would not allow for real-time image processing and hence was the largest bottleneck in the control pipeline. another drawback is that the arduino would often not respond to the commands sent over bluetooth. the commands would be passed but it was noticed that for the backward and forward command, the arduino would not respond until it was restarted. without the ability to reliably send steering commands to the car, the entire project would be brought to a standstill. hence the idea of using the arduino-based wheel-e was dropped and instead a remote-controlled car was used. 3.3. hardware implementation of remote-controlled car as the wheel-e had proven to be a failure for the purposes of this project, the r/c car was selected as its replacement. the project required the car to be controlled remotely from the computer, and so many changes would have to be made to be able to achieve such control. firstly, the remote control of the r/c car was opened. inside was a circuit board (fig. 2) that consisted of its ic, antenna, and 4 x 4-pin push buttons, one button each for forward, backward, left and right [15]. fig. 2 arduino and remote controller after soldering wires using the arduino and an led for testing, all 4 pins of all 4 push buttons were tested and the different kinds of pins were found. on an average 4 pin push button, 2 adjacent pins on one side are connected as are the 2 pins opposite to them. current only flows from one side to the other when the button is held down. after finding which pins were always on, and which pins would only turn on when the button was held down, it was realized that sending a high signal to one of those pins would act the same as physically pressing the button. so, 4 wires were soldered, and each one on the side of the button that turned high upon button press and was low otherwise. this allowed simulating button presses for the controller’s forward, backward, left and right commands simply by providing a high signal from an arduino to the corresponding pin of the controller. 3.4. interfacing the car with the computer after setting up the hardware connections from the arduino to the controller as in fig. 3, the software side of things needed to be implemented. the arduino program was straightforward, simply responding to certain input in the serial monitor and performing the corresponding task (‘f’ for forward, ‘l’ for left, ‘s’ for stop, and so on). once the arduino was set up with code to steer the car based on commands received, the whole setup had to be interfaced via python, and optimized there. to be able to control the car, the program needed to be able to read multiple key presses simultaneously, remember which keys are currently being held down, and send an appropriate message to the arduino which drives the car. to get multiple keypresses at once, the pynput python keyboard input library was used. multiple threads were created for each function, so they could all run efficiently side by side. this also allowed the program to keep track of multiple distinct key presses simultaneously. the commands were then sent to the arduino using the python serial library. proceedings of engineering and technology innovation, vol. 17, 2021, pp. 01-12 7 fig. 3 proposed design architecture after having completed the arduino setup, the next task is to stream images from the car to the computer, which would then be processed, sending drive instruction to the car. to get these images, an android phone was used. a mobile application allowed a video stream to be created on the local network, so as long as the computer and the phone were connected to the same network, they could communicate with each other. the pc would then connect to this stream using the python opencv library. this same library could also convert the video stream into individual frames of images, a necessary step for image processing. 4. lane detection by using image processing techniques lane detection is one of the most important aspects of an autonomous car. a camera, usually installed in the center of the dashboard behind the windshield, provides a constant stream of images of the road ahead. lane detection is necessary, as it provides the car with information about whether it is maintaining lane discipline, and the angle of the lanes can tell it about upcoming turns. it is responsible for keeping the vehicle in the middle of the lane at all times. the process implemented goes through a few steps:  gray scale conversion  yellow and white filter  gaussian blur  canny edge detection  region of interest  hough transform 4.1. gray scale conversion the image provided by the camera will likely be in color, but most of the information about the lanes can be found from a gray image itself. as shown in fig. 4, the original image will have 3 dimensions of information: red, green, and blue. but a grayscale image contains only a single dimension: intensity as in fig. 5. this reduces the size of the data being worked with a third of its original size. such a large amount of improvements in efficiency is achieved by losing small amount of information lost by choosing grayscale image. fig. 4 original image used as an input fig. 5 gray scale converted image proceedings of engineering and technology innovation, vol. 17, 2021, pp. 01-12 8 4.2. yellow and white filter roads all over the world employ a standard color scheme for lane marking. all lanes are either pure white or bright yellow. this simplifies the process of lane detection. instead of looking at the entire image, only those parts that are yellow or white can be filtered out and further analyzed. to get the yellow regions of the image, it is first converted to hsv (hue, saturation, value) format. this is done because it is difficult to obtain the correct range for bright yellow when using the red, green, blue format. once converted, a mask is created containing only yellow portions of the hsv image, and another mask containing only white portions. these masks contain the original image where their respective colors are present, and the rest of the pixels of the image are set to black. these two masks are then combined by using a bitwise and operator. finally, the combined mask is applied on the grayscale image and all that remains are white and bright yellow parts of the image as in fig. 6. fig. 6 output image after applied white and yellow filter 4.3. gaussian blur before applying edge detection on the image, it needs to be processed further. at this stage, the image is at a fairly high resolution, so small flaws in otherwise straight lines would be recognized by the algorithms. hence these images need to have some amount of detail taken away, which is achieved by a gaussian blur. the gaussian blur algorithm reduces the resolution of the image by setting the value of a pixel as the average of the value of all the pixels around it. this helps suppress noise that may be present in the image, and smooth out lines and make them appear straighter as in fig. 7. fig. 7 output image after applied gaussian blur 4.4. canny edge detection now that the image has been completely pre-processed, edge detection can be applied to it. even at this stage, the image has a little too much data to use the hough transform on as shown in fig. 8. so, to optimize it even further, only the edges of the image will be taken. an edge is any point where the average intensity of pixels on one side is drastically different when compared to the average intensity on the other side. the canny edge detection algorithm works by applying a few filters on the image and then calculating the directional gradient of each pixel and checking whether it satisfies the criteria for an edge or not. compared to other strictly defined methods, canny seems to be one of the best when both speed and accuracy are concerned. proceedings of engineering and technology innovation, vol. 17, 2021, pp. 01-12 9 fig. 8 output image after applied canny edge detection 4.5. spiking neural networks for edge detection while canny edge detection is one of the most popular algorithmically defined methods of edge detection, there is still room for improvement. using a spiking neural network is a much more efficient method of edge detection. spiking neural networks are built to emulate real neural networks more closely than simple mlp’s. they do this by not having all neurons fire at the same time but instead firing only when a certain threshold is met, and they also incorporate time into the model. xue li and team [4] trained a spiking neural network and were able to achieve very high accuracy and speed while a snn was not used for the purpose of this work, it still remains one of the best methods to improve the currently standing algorithms and is highly recommended for future works. 4.6. region of interest while the image provided by the camera is necessary for lane detection, a large region of the image is unimportant. the only region that is required is the road straight ahead, and the immediate flanking lanes as shown in fig. 9. the rest of the image consists of vehicles ahead, the horizon and the sky above it, and lanes other than the one the car is currently driving on. discarding all this unhelpful data from the image, at an early stage helps save a lot of computation time for the onboard computer, which can then be used by other important processes. also, it helps the further algorithms pinpoint the location of the lanes instead of looking at other unnecessary parts of the image. fig. 9 region of interest 4.7. hough transform a hough transform is used in the end to get the two lanes out of the image of edges. the hough transform is a mathematical function that is used to find the lines in an image. it transforms all lines into points in the hough space, and points into lines, and these points are calculated by the following equation. cos sinr x y   (5) proceedings of engineering and technology innovation, vol. 17, 2021, pp. 01-12 10 observing the curves produced in the hough space will show that there are 2 major clusters of points of intersection in the hough space, representing all the lines that comprise the 2 major lanes on either side of the road. these lines have been plotted onto the original image as illustrated in fig. 10. fig. 10. hough transform lines fig. 11. lanes output from hough transform the image shows about 6 lines on the left lane and about 7 on the right. each of these groups needs to be considered separately, but the number of lines must be reduced to one each. this can be achieved by taking the average of each set of hough lines produced. this leaves a final image as illustrated in fig. 11 that considers the average length and direction of each line. these lines may be extrapolated to account for areas of the road where there are no lanes. 5. results and discussion 5.1. detection of stop sign and distance measurement while lane detection is important for an autonomous car to be able to drive according to the rules and maintain lane discipline, it is also important for them to be able to recognize street signs. street signs have been around for decades as they provide an easy, fast, and convenient way for a driver to look at instructions and warnings even when driving at high speeds. such signs often provide very valuable information to the driver and could similarly aid an autonomous car as it drives, especially along smaller, unknown roads. as a proof of concept, this project implements the detection and response to a stop sign. as the r/c car moves along the specified path, its attached camera keeps a lookout for such signs and obstacles. upon seeing a stop sign, it calculates the distance to the sign, based on previously done calibration, and stops if it is within a certain distance of said sign. after stopping for a few seconds, it then proceeds with caution. fig. 12 stop sign detection with distance measurement to train the stop sign classifier, images were taken of a small, cutout stop sign as shown in fig. 12. these were taken from many different angles to make the classifier robust to changes in camera angles. they were then cropped to only contain the desired region. many negative samples were also provided to the classifier, so it knew what and what not to look for. the classifier was based on haar cascades. the haar cascade classifier learns by superimposing the positive samples over the negative samples and generating features out of the superimposition. the training algorithm creates thousands of features and proceedings of engineering and technology innovation, vol. 17, 2021, pp. 01-12 11 then applies each of these features over every pixel of the image. only those features that have a low loss value when classifying the given set of images are retained, while the rest are discarded. in the end, those features are combined in the form of a cascade. larger, general features are kept higher up and checked sooner, while smaller, specific features are low in the tree and checked only if the ones above them proved successful. along with being able to classify a stop sign for a given image, the distance to the stop sign also needs to be calculated. only once the car came within a certain radius of the stop sign was it supposed to stop; hence distance was necessary. to calculate the distance, a monocular method was implored as 2 cameras could not be fit onto the car. there is a direct correlation between actual size, perceived size, the focal length of the camera, and the distance. simply using geometric triangle similarity, it is known that the ratios of the sides of two similar triangles are equal [16]. during calibration, the focal length was found by obtaining concurrent measurements using the following equation. tanwidth in pixels dis ce from camera focal length actual width of objects   (6) once the focal length was obtained, all that need to be done was rearrange the formula so that the distance could be calculated as follows tan focul length actual width dis ce from camera width in pixels   (7) while this method is not perfect, it is able to provide a rough estimate of the distance of the object. this method is also resistant to changes in the angle of images, as long as the horizontal angle is not changed by much. 5.2. vehicle detection with distance measurement besides stop signs, the algorithm was also trained to recognize vehicles. in this case it was trained on one specific car that would be used as an obstacle for the autonomous car. the algorithm again made use of haar cascades as its classifier. using many positive examples of the vehicle, and a set of negative examples without the vehicle. the classifier was trained to recognize the car with high accuracy, using the same monocular vision method of triangular similarity, the distance to the car was also measured, and the camera had to be recalibrated on the new object. the result is illustrated in fig. 13. fig. 13 vehicle detection with distance measurement 6. conclusions a system for image processing in autonomous cars has been implemented. by applying a multistage pre-processing to the images for lane detection, the lanes can be detected quickly and accurately. selecting the region of interest as the first step saves at least 2-3 times the amount of processing required. selecting only white and yellow regions further reduces processing proceedings of engineering and technology innovation, vol. 17, 2021, pp. 01-12 12 time and drastically improves the detection of the relevant edges. classification of stop signs and detection of vehicles ahead was also achieved. this would be greatly useful for recognizing and responding to not only road signs but also vehicles ahead. not only were these objects recognized, but their distance from the camera was also calculated. the algorithm responds to a stop sign by stopping the car for a few seconds if it is within a certain distance from the stop sign. after the wait time had elapsed, it continues along. it also responds to vehicles ahead by stopping completely if it gets too close. these implementations ensure that the autonomous car follows the rules of the road, and does its best to prevent accidents, preventing loss of life. in the future, the proposed algorithm can be further trained to recognize and respond to many more road signs. conflicts of interest the authors declare no conflict of interest. references [1] v. viswanathan and r. hussein, “applications of image processing and real-time embedded systems in autonomous cars: a short review,” international journal of image processing, vol. 11, no. 2, pp. 35-49, april 2017. [2] j. v. brummelen, m. o’brien, d. gruyer, and h. najjaran, “autonomous vehicle perception: the technology of today and tomorrow,” transportation research part c: emerging technologies, vol. 89, pp 384-406, april 2018. [3] j. joel, g. fatma, b. aseem, and g. andreas, “computer vision for autonomous vehicles: problems, datasets and state of the art,” foundations and trends® in computer graphics and vision, vol. 12, no. 1-3, pp. 1-308, july 2020. [4] y. xing, c. lv, l. chen, h. wang, h. wang, d. cao, et al., “advances in vision-based lane detection: algorithms, integration, assessment, and perspectives on acp-based parallel vision,” ieee/caa journal of automatica sinica, vol. 5, no. 3, pp. 645-661, may 2018. [5] a. jain, h. reddy, and s. dubey, “automated driving vehicle using image processing,” international journal of computer science and engine and engineering, vol. 2, no. 4, pp. 138-140, may 2014. [6] r. muthalagu, a. bolimera, and v. kalaichelvi, “lane detection technique based on perspective transformation and histogram analysis for self-driving cars,” computers & electrical engineering, vol. 85, 106653, july 2020. [7] x. li, q. wu, y. kou, l. hou, and h. yang, “lane detection based on spiking neural network and hough transform,” 8 th international congress on image and signal processing (cisp), october 2015. [8] y. huang, y. li, x. hu, and w. ci, “lane detection based on inverse perspective transformation and kalman filter,” ksii transactions on internet and information systems, vol. 12, no. 2, pp. 643-661, february 2018. [9] j. m. yang and s. h. wei, “spatial and spectral nonparametric linear feature extraction method for hyperspectral image classification,” advances in technology innovation, vol. 2, no. 3, pp. 68-72, july 2017. [10] q. wang, j. gao, and x. li, “weakly supervised adversarial domain adaptation for semantic segmentation in urban scenes,” ieee transactions on image processing, vol. 28, no. 9, pp. 4376-4386, september 2019. [11] q. wang, j. gao, and y. yuan, “embedding structured contour and location prior in siamesed fully convolutional networks for road detection,” ieee transactions on intelligent transportation systems, vol. 19, no. 1, pp. 230-241, january 2018. [12] w. suh, j. r. kim, and j. cho, “vision technology based traffic safety analysis using signal data,” proceedings of engineering and technology innovation, vol. 13, pp. 20-25, july 2019. [13] w. h. suh, j. w. park, and j. r. kim, “traffic safety evaluation based on vision and signal timing data,” proceedings of engineering and technology innovation, vol. 7, pp. 37-40, december 2017. [14] injury facts, “historical fatality trends,” https://injuryfacts.nsc.org/motor-vehicle/historical-fatality-trends/deaths-and-rates/, october 23, 2018. [15] j. wan, “working of a 4-pin push button,” http://www.jobiwan.net/xfp/button.png, december 10, 2018. [16] t. emara, “pinhole camera distance measurement,” http://emaraic.com/assets/img/posts/computer-vision/distance-measurement/pinhole-camera.gif, december 10, 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/). 4___peti#9477___30-39 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 30-39 experimental and analytical study of silica particles on self-healing concrete kamasani chandrasekhar reddy * , krishnaiah gari hemanth kumar department of civil engineering, siddharth institute of engineering & technology, puttur, india received 15 february 2022; received in revised form 19 june 2022; accepted 20 june 2022 doi: https://doi.org/10.46604/peti.2022.9477 abstract this study aims to investigate the properties of green concrete made with ground granulated blast-furnace slag (ggbs), robo sand (rs), and coconut shell (cs). ggbs is the mineral admixture used to replace cement. nano-silica particles (nsps) and cs are used as coarse aggregates, and rs is the fine aggregate used to replace river sand. the workability, mechanical properties, and durability properties of green concrete are investigated and compared with those of conventional concrete (cc). test results show that the cement replaced with 30% ggbs and 3% nsps exhibits superior strength. the compressive and splitting tensile strengths are increased by 24.03% and 42.32% after 28 days of curing, respectively. the workability is improved by 12.22% (slump) and 13.25% (compaction factor) after 28 days of curing. the sorptivity of hm3 (3.26%) is lower than that of cc due to the uniform distribution between particles. microstructure evolution is carried out to identify concrete mix behavior. keywords: ggbs, nano-silica, coconut shell, robo sand, durability properties 1. introduction concrete is an artificial matrix with water, cement, fine aggregates (fas), and coarse aggregates (cas). nowadays, the manufacturing of concrete encounters challenges because of the less accessibility of cement, fas, and cas in the construction field. many cement-based self-healing materials are utilized to replace the common ingredients in conventional concrete (cc) to minimize cracking and shrinkage in concrete. hence, researchers have developed alternatives for fas, cas, and cement [1]. however, those developed alternatives may result in environmental pollution. to avoid the pollution, many researchers have focused on developing new alternatives with renewable resources such as organic by-products and waste products for the construction industry. eco-friendly materials such as agro-based materials are used. also, with recycling processes, solid waste materials are used as an alternative to aggregates and cement in the construction field, which benefits the environment and human beings at the same time. mathew et al. [2] studied the change in concrete properties after replacing cement with coconut shell (cs) ash. the embodied carbon, workability, and mechanical properties of concrete were tested. they found higher strength in the case of 10% replacement and a 15% reduction in embedded carbon in the case of 20% replacement, as compared to the control mix. palanisamy et al. [3] utilized the cs powder to partially replace fas and examined the performance of physical and mechanical properties. finally, from the outcomes of compressive and flexural tests, they concluded that the use of cs decreases density and increases compressive and flexural strengths. with 30% replacement of fas, the cost of concrete was reduced, as compared with the control mix. * corresponding author. e-mail address: kamasani.kcr@gmail.com proceedings of engineering and technology innovation, vol. 22, 2022, pp. 30-39 ikponmwosa et al. [4] used the rice husk ash (rha) with the cs ash to replace 30% cement, and found that this combination gave better results than simply using cc. the presence of waste particles increases the strength and reduces the cost of the material. the strength development rate of normal concrete is also found during the first 14 days of curing. azunna et al. [5] studied the difference between the cc pipes and the concrete pipes made with cs. from three-edge test results, they found that both pipes are within permissible limits as per indian standard (is) codes. they concluded that the use of cs should be encouraged in concrete construction. overall, the prior studies highlighted the need to use admixtures to enhance the features of concrete. they also tried to replace ingredients of concrete with waste materials. the current research aims to assess the mechanical and durability properties of green concrete made with the combination of ground granulated blast-furnace slag (ggbs), robo sand (rs), and cs. the contribution of this study includes the following. (1) this study investigates the effect of ggbs, nano-silica particles (nsps), and cs on concrete by analyzing the fresh properties, mechanical properties, and durability properties of various designated mixes. (2) this study compares the experimental results of various designated mixes with cc to identify the optimum mix. 2. methodology green concrete specimens are cast and cured after being synthesized with various compositions, i.e., mix – i, mix – ii, mix – iii, mix – iv, and mix – v. the workability, mechanical properties, and durability properties of the specimens are tested. the effect of the addition of dispersive particles is identified using surface morphology analysis. the entire process is shown in fig. 1. fig. 1 methodology of this study 2.1. materials the elements adopted in this research are cement, fas (natural river sand and rs), cas, ggbs, cs, nsps, and potable water. the following section contains information about the materials utilized. fig. 2 shows the raw materials of ggbs, cs, and nano-silica powder. methodology synthesis of green concrete specimens with various compositions curing process of all samples effect of the addition of dispersive particles tests of workability, mechanical properties, and durability properties surface morphology analysis conclusions mix – ii (hm1) ggbs – 10%, nsps – 1%, cs – 5% mix – iii (hm2) ggbs – 20%, nsps – 2%, cs – 10% mix – iv (hm3) ggbs – 30%, nsps – 3%, cs – 15% mix – v (hm4) ggbs – 40%, nsps – 4%, cs – 20% mix – i (hm0) conventional concrete (cc) 31 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 30-39 (a) ggbs (b) coconut shell (c) nano-silica powder fig. 2 raw materials for the preparation of concrete 2.1.1. cement cement is a binding material in concrete. in this project, the opc 53 ultratech brand is used. tests according to is 12269-2013 for cement are conducted and found suitable [6]. the chemical and physical characteristics of cement are placed in tables 1 and 2. table 1 chemical properties of cement elements sio2 al2o3 fe2o3 cao so3 mgo na2o k2o loi composition (%) 21.36 4.16 4.32 60.58 3.26 1.70 0.28 0.36 2.18 table 2 physical properties of cement test particulars results obtained fineness 6.3% initial and final setting time (minutes) 35 & 410 mins specific gravity 3.09 normal consistency (%) 31% compressive strength (n/mm 2 ) 51 2.1.2. fine aggregate (fa) natural river sand is the fa used to cast cc. the crushed stone aggregate (rs), which passes through a 0.075 mm sieve and retains the size of 4.75 mm, is considered the alternative for natural river sand. rs is clean and free from contaminants. the rs used in this study is obtained from silica mines & minerals, bangalore, india. tests according to is 2386.1-1963 are carried out and found suitable [7]. 2.1.3. coarse aggregate (ca) a local granite crushed stone (with the size of 20 mm) is used as a ca and is obtained from sri srinivas granites, tirupati, india. it assembles solid and hard stacks of concrete with cement and sand, and supplies mass to concrete. tests according to is 2386.1-1963 are conducted and found suitable [8]. 2.1.4. ground granulated blast-furnace slag (ggbs) ggbs is obtained when making iron in blast furnaces. it has cementitious properties and is obtained from astrra chemicals, chennai, india. the chemical and physical characteristics of ggbs are represented in tables 3 and 4, as shown below. table 3 chemical properties of ggbs elements sio2 al2o3 fe2o3 mno cao na2o loi composition (%) 27-38 7-15 0.2-1.6 0.15-0.76 34-43 0.2-0.48 bal table 4 physical properties of ggbs test particulars results obtained specific gravity 2.88 surface area (g/cm 2 ) 1200 bulk density (kg/m 3 ) 1195 optimum moisture content (%) 32 32 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 30-39 2.1.5. nano-silica particles (nsps) the nsps in the cementing process can accelerate the early hydration of concrete, which is beneficial for strengthening the early strength of concrete. the nsps used in this study are obtained from astrra chemicals, chennai, india. the chemical and physical properties of nsps are represented in tables 5 and 6. table 5 chemical properties of nsps elements sio2 na2o al2o3 so3 fe ca cu loi composition (%) 99.4 0.45 0.075 0.05 0.25 0.12 0.16 bal table 6 physical properties of nsps test particulars values density 2.17-2.66 gr/cm 3 melting point ±1700°c boiling point 2300°c color white bulk density 0.011 gr/ml 2.1.6. coconut shell (cs) cs is an agricultural by-product widely used in india. it is beneficial for strengthening concrete and is obtained from srp global exports, chennai, india. the cs’s chemical and physical properties are represented in tables 7 and 8 as shown below. table 7 chemical properties of cs element cellulose hemi celluloses lignin pectin moisture content microfibrillar angle (degree) wt.% 32-43 0.15-0.25 40-45 3-4 8 30-49 table 8 physical properties of cs test particulars results obtained specific gravity 1.33 bulk density (kg/m 3 ) 800 shell thickness (mm) 3.09 water absorption capacity (%) 15.17 fine modulus 3.19 2.2. preparation of coconut shell into ash the cs waste is the agro-waste producing 12.3 million tons of coconuts annually in india. the cs used in this study is collected from a local resource, tirupati, india. the cs ash is suitable for making carbon black due to its excellent natural structure and low ash quantity. initially, cs is thoroughly cleaned with distilled water and dried at room temperature for 24 hrs. later, it is transferred into a tubular furnace and heated at a temperature of 550°c. the cs ash is sieved through 200 sieves. 2.3. mix design the design m30 mix is carried out for cc as per is 10262-2019 [9], and the same design mix is used to prepare the concrete by replacing its ingredients with ggbs, nsps, rs, and cs. the hybrid mix (hm) ratios of m30 grade concrete for a cubic meter are listed in table 9. 33 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 30-39 table 9 mixed proportion of biohybrid blended concrete mixture serial no. designation cement (%) river sand (fa) (%) ca (%) replacement (%) ggbs nsps rs cs 1 hm0 100 100 100 0 0 0 0 2 hm1 89 0 95 10 1 100 5 3 hm2 78 0 90 20 2 100 10 4 hm3 67 0 85 30 3 100 15 5 hm4 56 0 80 40 4 100 20 3. tests the compressive strength of concrete with different mixes is tested using a universal testing machine (utm), and the test results are depicted as a pictorial diagram. the specimens are prepared by mixing and filling the concrete in the molds with a suitable layer of 50 mm and compacting the concrete in a cube (150 × 150 × 150 mm) [10]. the splitting tensile samples are prepared by mixing and filling the concrete in the specimens with a suitable layer of 50 mm and compacting the concrete in a cylinder with a length of 300 mm and a diameter of 150 mm using utm. the samples of the cylinders are tested and cured for 7, 14, 21, and 28 days. tests according to is 5816-1999 are carried out [11]. various tests are conducted on the specific gravity of cement, and the dry sieving method is conducted to test the consistency of cement, fineness of cement, soundness, as well as initial and final setting times. similarly, several tests are conducted on ggbs to determine the specific gravity, strength, initial setting time, and fineness. in this study, the tests conducted on aggregates are sieve analysis, specific gravity, bulking of sand, and water absorption tests. the preparation and testing samples are shown in fig. 3 and fig. 4. after preparing the mix with various proportions, the workability tests are conducted on fresh concrete, and the mechanical tests are conducted on hardened concrete. the compressive, splitting tensile, and flexural strength tests are performed on the specimens (cube, cylinder, and beam). (a) accelerated chamber (b) concrete mixing (c) preparation of samples fig. 3 mixing and sample preparation (a) compression test (b) tensile test (c) slump test fig. 4 testing apparatus 34 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 30-39 4. results and discussion 4.1. workability of concrete the slump cone and compaction factor tests on green concrete are shown in table 10. the slump cone test is conducted according to is 1199-1959 [12]. in the compacting factor test, an apparatus with two upper and lower hoppers and a cylinder is placed one by one according to the procedure. the test is carried out according to is 1199-1959 [13]. the composition factor is calculated by using eq. (1). 1 2 2 w w compaction factor w w − = − (1) where w is the empty cylinder weight, w1 is the weight of the cylinder and free fall concrete, and w2 is the weight of the cylinder and hand compacted concrete. with the combination of 30% ggbs, 3% nsps, and 15% cs, the workability is increased, and the slump flow is later decreased. the sulphonated naphthalene-based superplasticizers are mixed in the concrete at a dose of 1.5% of the cement weight to improve the workability. it is found that, in the case of hard particles (hm3), the slump increases from 90 to 101 mm (12.22%) and the compaction factor increases from 0.83 to 0.94 (13.25%), compared to the case of cc. hence, hm3 is considered the optimum mix for better performance. amin et al. [14] concluded that using 15% of cs and nsps in concrete would cause the decrease in slump and the delay in initial and final setting time. the slump slightly increases to 95 (hm4) because the high mix does not dissolve in concrete. hence, the addition of nsps and cs is a significant effect illustrating good workability, reduced heat of hydration, and cost reduction in concrete production. table 10 workability of hm hybrid mix id slump (mm) compaction factor hm0 90 0.83 hm1 92 0.87 hm2 97 0.91 hm3 101 0.94 hm4 95 0.89 4.2. mechanical studies 4.2.1. compressive strength compressive strength values are attained by examining cubes made with cc and different hms, as shown in fig. 5. all the mixes contain a strength above 30 mpa, which is the anticipated strength. all the mixes show an increase in compressive strength at the 28th day of curing. the compressive strength improves with the use of cs and ggbs with nsps, later decreasing at a certain level, as shown in fig. 5. in the case of using 15% cs and 30% ggbs with 3% nsps (hm3), it is found that the compressive strength increases after 7 days from 20.67 to 24.17 n/mm 2 (16.93%), after 14 days from 26.72 to 30.68 n/mm 2 (14.82%), after 21 days from 33.68 to 39.34 n/mm 2 (16.81%), and after 28 days from 36.75 to 45.58 n/mm 2 (24.03%), as compared to the case of cc. hence, the hm3 is considered the optimum mix for better performance, tends to fill the pores, and increases the compressive strength of concrete. according to gao et al. [15], substituting 25% of the cc with a nano combination resulted in more than 45 n/mm 2 after 28 days. furthermore, filler components enhanced the compressive strengths of cc combinations at an early age. test results show that the ggbs with nsps and cs shows superior mechanical properties. the compressive strength is reduced in the case of hm4 due to the balling effect of hybrid content. 35 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 30-39 fig. 5 compressive strength of various mixes 4.2.2. splitting tensile strength the tensile strength of mixes is attained by performing a split tensile test on standard cylindrical specimens. incorporating hms into cc raises the strength by 32% after 28 days curing, as shown in fig. 6. the splitting tensile strength increases with the use of 3% nsps and 15% cs. in the case of ggbs with 3% nsps (hm3), it is found that the splitting tensile strength increases after 7 days from 3.05 to 4.23 n/mm 2 (38.69%), after 14 days from 4.08 to 5.73 n/mm 2 (40.44%), after 21 days from 4.51 to 6.37 n/mm 2 (41.24%), and after 28 days from 4.75 to 6.76 n/mm 2 (42.32%), compared to the case of cc. hence, hm3 is considered the optimum mix for better performance. the hydration of cc enhances the mechanical properties of samples at later ages by decreasing macro cracking size and increasing cs/nsps ratio. nandhini et al. [16] conducted cracking control of traditional concrete and cc modified with nsps and fas. they suggested that the amount of paste in the mix be increased in strength so that fragments can be dispersed more evenly, ensuring a high level of workability. the reduction in tensile strength observed in the case of hm4 is attributed to an increase in porosity and distribution of pores. therefore, the presence of nsps and cs tends to fill the pores and control the crack growth, increasing the mechanical strength of concrete. fig. 6 splitting tensile strength of various mixes 4.3. durability studies 4.3.1. permeable void test the permeable voids are determined as per astm c 642-82 on cured samples with cubes (150 × 150 × 150 mm) [17]. the surface of the cubes is cleaned with a dry cloth, and the weights of the saturated surface dry cubes are noted. these cubes are kept in a hot oven at 105°c for a long time to ensure a consistent weight. constant weight is reached after the oven-dried cubes are soaked in water, and an increase in weight is found at frequent periods. the percentage of permeable spaces of the specimens is calculated using eq. (2). 36 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 30-39 (%) 100 a b percentage of permeable voids v − = × (2) here, a is the mass of the surface-dried saturated specimen, b is the mass of the oven-dried sample, and v is the volume of the samples. the astm c 642 [18] classifies the volume of permeable voids (vpv) into categories: the vpv less than 14% represents the excellent quality of concrete, the 14-16% vpv is good, the 17-19% vpv is marginal, and the 19% vpv represents poor quality of concrete. the results of penetrable voids for cc and hm are exhibited in table 11. it is identified that a lower vpv is found (11.21%) in the case of hm3 compared to the normal mix. this may be attributed to calcium-silicate hydrate (csh) gel filling up the cracks and pits present in concrete and improving material stability. high compressive strength cubes that are less porous will improve the permeable void ratio. the strength of the concrete is enhanced with the respective usage of nsps and cs in normal concrete. further addition of hm4 shows increasing permeable voids (12.68%). it may appear due to the higher paste content and dosage of superplasticizer adopted for producing cc. this leads to an increase in porosity suggested by chen et al. [19]. table 11 permeable void test mix id surface saturated specimens (kg) oven-dried specimens (kg) permeable voids (%) hm0 7.95 7.44 15.07 hm1 8.2 7.71 14.53 hm2 8.04 7.64 13.96 hm3 7.96 7.49 11.21 hm4 7.99 7.52 12.68 4.3.2. sorptivity test in unsaturated specimens, sorptivity is an indication of moisture activity. here, it is identified as a vital measure for assessing the long-term durability of concrete. as a result, the testing method used for its resolution is a good indicator of the quality of close-to-surface concrete. it resembles the way that water and other corrosive agents will pass through many concretes [20]. a concrete sorptivity coefficient is critical for estimating the framework’s mortality rate and improving its performance. measuring the capillary rise absorption rate on a uniform product can determine the sorptivity. it is dried in an oven at 105°c. it is kept in water at a depth of 5 mm from the bottom of the sample, and the flow from the outside of the sample is restricted through non-absorbent covering. later, the weight of the specimen is used to measure the amount of water absorbed for a few minutes. within 30 seconds of each procedure, wet cells are used to wipe away any surface water accumulated. permeable materials can absorb and transmit water through the capillary action, referred to as the sorptivity coefficient (k). the average time denominator increases the total collective water absorption (at each inflow surface location) (t). the sorptivity values are explored for cc and hms, as presented in table 12. table 12 shows that the sorptivity of hm2 (5.08%), hm3 (3.26%), and hm4 (4.21%) is lower than that of cc due to uniform distribution between particles. praveenkumar and vijayalakshmi [21] confirmed that 20% of ggbs with nsps samples showed an 8% reduction in sorptivity after 150 days. hence, it is attributed to the effect of these pozzolans and the decrease in permeability. table 12 sorptivity test serial no. designation sorptivity (%) 90 days 120 days 150 days 1 hm0 7.26 7.19 7.04 2 hm1 6.82 6.36 6.22 3 hm2 5.45 5.21 5.08 4 hm3 3.76 3.47 3.26 37 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 30-39 5. conclusions the usage of waste materials for raw materials is beneficial to the environment and minimizes material costs. based on the experiment in this study, it is found that the combination of using ggbs and nsps to replace 33% cement, using rs to replace 100% river sand, and using cs (hm3) to replace 15% cas is the optimum combination to be the alternative for cc. the following conclusions are drawn. (1) in the case of hard particles (hm3), the slump increases from 90 to 101 (12.22%), and the compaction factor increases from 0.83 to 0.94 (13.25%), compared to the case of cc. it is because of the addition of the superplasticizer in the hm. (2) the compressive strength of cc is 45.58 n/mm 2 after 28 days of curing, while the addition of cs and ggbs with nsps is used as a binary blended concrete resulting in increased strength for all grades. with the optimum mix (hm3) of 15% cs with 3% nsps for cement replacement, the strength obtained is 24.03% higher than with cc. (3) the tensile strength achieved is 6.76 n/mm 2 in the case of the hm3 mix after 28 days, which is 42.32% higher than in the case of cc due to the uniform distribution between particles and high hardness. further addition of hm4 to blended concrete reduces the mechanical strength due to the increasing porosity level and distribution of pores. (4) the vpv in concrete is 11.21% (less than 14%, excellent) in the case of hm3, which is better than in the case of cc. (5) among all the mixes, hm3 shows the maximum reduction (3.26%) of sorptivity value after 150 days. from the rapid chloride permeability test (rcpt), the resistance to chloride ion penetration in the case of hm3 is 32% higher than cc after 150 days. (6) the results show that it is suitable to use ggbs (with nsps), rs, and cs to replace cement, river sand, and cas in the concrete, respectively. hm3 is considered the optimum mix for better performance, tends to fill the pores, and increases the mechanical strength of concrete. adding reinforcement particles will delay beams’ damage and improve their durability. conflicts of interest the authors declare no conflicts of interest. references [1] s. ahmad, et al., “fibre-reinforced self-compacting concrete: a review,” iop conference series: materials science and engineering, vol. 377, no. 1, article no. 012117, december 2017. [2] s. p. mathew, et al., “experimental study of thermal properties of concrete with partial replacement of coarse aggregate by coconut shell,” materials today: proceedings, vol. 27, pp. 415-420, june 2020. [3] m. palanisamy, et al., “permeability properties of lightweight self-consolidating concrete made with coconut shell aggregate,” journal of materials research and technology, vol. 9, no. 3, pp. 3547-57, june 2020. [4] e. e. ikponmwosa, et al., “performance of coconut shell alkali-activated concrete: experimental investigation and statistical modelling,” silicon, vol. 13, no. 2, pp. 335-340, march 2021. [5] s. u. azunna, et al., “characterization of lightweight cement concrete with partial replacement of coconut shell fine aggregate,” sn applied science, vol. 1, no. 6, article no. 649, may 2019. [6] c. a. okeh, et al., “behaviour of hybrid steel fibre reinforced self compacting concrete using innovative hooked-end steel fibres under tensile stress,” construction and building materials, vol. 202, pp. 753-761, march 2019. [7] l. xu, et al., “effects of coarse aggregate and steel fibre contents on mechanical properties of high performance concrete,” construction and building materials, vol. 206, pp. 97-110, may 2019. [8] m. h. de medeiros, et al., “partial replacement and addition of fly ash in portland cement: influences on carbonation and alkaline reserve,” journal of building pathology and rehabilitation, vol. 2, no. 1, pp. 1-9, march 2017. [9] h. sasanipour, et al., “effect of specimen shape, silica fume, and curing age on durability properties of self-compacting concrete incorporating coarse recycled concrete aggregates,” construction and building materials, vol. 228, article no. 117054, december 2019. 38 proceedings of engineering and technology innovation, vol. 22, 2022, pp. 30-39 [10] s. k. mezzal, et al., “effect of discarded steel fibers on impact resistance, flexural toughness and fracture energy of high-strength self-compacting concrete exposed to elevated temperatures,” fire safety journal, vol. 121, article no. 103271, may 2021. [11] y. gao, et al., “duan preparation and flexural fatigue resistance of self-compacting road concrete incorporating nano-silica particles,” construction and building materials, vol. 278, article no. 122380, april 2021. [12] p. murthi, et al., “saravanan enhancing the strength properties of high-performance concrete using ternary blended cement: opc, nano-silica, bagasse ash,” silicon, vol. 21, no. 8, pp. 1949-1956, august 2019. [13] m. oghabi, et al., “the laboratory experiment of the effect of quantity and length of plastic fiber on compressive strength and tensile resistance of self-compacting concrete,” ksce journal of civil engineering, vol. 24, no. 8, pp. 2477-2484, august 2020. [14] m. amin, et al., “effects of nano cotton stalk and palm leaf ashes on ultrahigh-performance concrete properties incorporating recycled concrete aggregates,” construction and building materials, vol. 302, article no. 124196, october 2021. [15] b. y. lee, et al., “evaluation of time to shrinkage-induced crack initiation in opc and slag cement matrices incorporating circulating fluidized bed combustion bottom ash,” construction and building materials, vol. 257, article no. 119507, october 2020. [16] k. nandhini, et al., “effect of blending micro and nano silica on the mechanical and durability properties of self-compacting concrete,” silicon, vol. 13, no. 3, pp. 687-695, march 2021. [17] l. s. reddy, et al., “thermal conductivity and strength properties of nanosilica and ggbs incorporated concrete specimens,” silicon, vol. 14, pp. 145-151, 2022. [18] i. rahmani, et al., “a laboratory study on the flexural and shear behavior of rc beams retrofitted with steel fiber-reinforced self-compacting concrete jacket,” iranian journal of science and technology, vol. 45, pp. 2359-2375, 2021. [19] w. chen, et al., “dynamic compressive properties of high volume fly ash (hvfa) concrete with nano silica,” construction and building materials, vol. 301, article no. 124352, september 2021. [20] r. hari, et al., “mechanical and durability properties of sisal-nylon 6 hybrid fibre reinforced high strength scc,” construction and building materials, vol. 204, pp. 479-491, april 2019. [21] t. r. praveenkumar, et al., “microstructural properties of nano-rice husk ash concrete,” nanotechnology for environmental engineering, vol. 4, no. 1, article no. 9, december 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/). 39  proceedings of engineering and technology innovation, vol. 15, 2020, pp. 24 34 effect of acceleration on wrap faced reinforced soil retaining wall on soft clay by performing shaking table test ripon hore 1,* , sudipta chakraborty 2 , ayaz mahmud shuvon 2 , mehedi ahmed ansary 1 1 department of civil engineering, bangladesh university of engineering and technology, dhaka, bangladesh 2 bangladesh network office for urban safety (bnus), bangladesh university of engineering and technology, dhaka, bangladesh received 21 june 2019; received in revised form 03 october 2019; accepted 03 january 2020 doi: https://doi.org/10.46604/peti.2020.4485 abstract this research incorporates shaking table testing of scale wrap faced soil wall models to evaluate the seismic response of embankment. currently the seismic designs of highway or railway embankment rely on little or no empirical data for calibrating numerical simulations. this research is working towards filling that empirical data gap. the specific purpose of the study was to evaluate the seismic response of constructed embankment model regarding the different input base accelerations with fixed frequency. a series of one-dimensional (1d) shaking table tests (0.05g, 0.1g, 0.15g and 0.2g), were performed on a 0.4 meters high wrap faced reinforced-soil wall model. additionally, it was placed over 0.3 meters high soft clayey foundation. predominantly, the influence of the base acceleration on the seismic response was studied in this paper. the physical models were subjected to harmonic sinusoidal input motions at a fixed frequency of 1 hz, in order to assess the seismic behavior. the effects of parameters such as acceleration amplitudes and surcharge pressures on the seismic response of the model walls were considered. the relative density of the backfill material was kept fixed at 60%. the results of this study reveal that input accelerations and surcharge load had significant influence on the model wall, pore water pressure, and changes along the elevation. acceleration response advances with the increase in base acceleration, so the difference being more perceptible at higher elevations. the pore water pressures were found to be high for high base shaking and low surcharge pressures at higher elevations. the results obtained from this study are helpful in understanding the relative performance of reinforced soil retaining wall under different test conditions resting on soft clay. keywords: reinforced embankment, soft clayey soil foundation, input acceleration, sinusoidal input motions, input frequency, surcharge load, pore water pressure, shaking table test 1. introduction highways are essential lifelines that should be in operation even after disastrous events. for that reason, during and after earthquakes, earth structures such as highway embankments should be stable enough to provide required safety and emergency needs. the soil-foundation composed of clayey soil has become the focus of earthquake engineering in more cases. however, it is challenging to study the characteristic of pore water pressure in such a clayey soil-foundations under an earthquake action. although such a soil-foundations of higher clay content, a rise of excess pore water pressure triggered by earthquakes and a drop of effective stress may lead to the soil softening. moreover, soil-foundation softening may cause damage to the embankment. thus, engineering judgment and application are required to reinforce the embankment to acquire the stability conditions. geosynthetics have proven to be useful and cost-effective site improvement materials. they have become essential materials in civil engineering and environmental geotechnics. application of the geosynthetics in highway embankments provides an additional tensile strength and durability to the structures under horizontal and vertical loads, so it enables engineers to construct more stable and earthquake resistant highway embankments [1-2]. the researches in geotechnical * corresponding author. e-mail address: riponhore@gmail.com tel.: +88-02-8181162; fax: +88-02-9136751 proceedings of engineering and technology innovation, vol. 15, 2020, pp. 24 34 25 engineering fields witnessed distinguishable developments in the methods and techniques of studying the engineering problems concerning dynamic effects and seismic performance of structural elements problems. physical modeling is on one of these techniques, which principles were widely used on different small-scale models shaking table test demonstrating different prototypes scale around the worlds [3]. different sizes of shaking table were designed and used throughout the world to solve problems under different dynamic motions and analyze its effect [4]. design, development, calibration and performance of this equipment are very important [5]. humaish et al. [6] and chau et al. [7] were investigated different geotechnical and engineering problems which deals with designing, manufacturing, and testing of small shaking table. worldwide many researchers [8-9] conducted shaking table tests on geosynthetics reinforced soil wall in different times. srilatha et al. [10] and edinçliler and toksoy [11] were performed to reduce scale shaking table test on model embankment to compare the behavior of reinforced and unreinforced embankment under both earthquake with uniform cyclic loading. yazdandoust [12] conducted a series of one gravitational shaking table tests under harmonic as well as seismic loading on 0.9 meters high reinforced-soil wall models to assess the behavior of the models. zhou et al. [13] presented shaking table tests on three aeolian sand high embankment models. moreover, shaking is an effective technique to study the seismic response of underground structures embedded in soft soil sites. goktepeet al. [14] observed that the local soil properties have considerably amplified the earthquake response of the structures in comparison to the fixed base structure. ç elebi et al. [15] specified a realistic geometric scaling coefficient for test model to be used in a small-capacity shaking table. the scaling factor addressed in this study involves not only geometric similarity but also kinematic and dynamic similarity with the real system. the dynamic parameters for scaled model of a single layer soil are restricted with base-rock which has been compared numerically with the proposed laminar soil container to provide a good agreement [16]. many researchers [17-19] performed shaking table test, to study the seismic responses of underground structures embedded in soft soil. moreover, it is necessary to consider the influence of pore pressure in soft soil sites. it is understood from the above literature that, the reinforced soil retaining wall on soft clay is prevalent in all over the world. moreover, it is very important to consider seismic loading in design phase. keep everything in concern, a 1/10 scaled geosynthetics wrap faced embankment model was designed for this study in order to investigate the effect of different input motions with differences in amplitude and fixed frequency. twelve shaking table test were carried out on this model embankment. during the experiments, accelerations moving through the embankment as well as the vertical and horizontal displacement values for each motion were obtained. the total method of this study is presented thorough a flowchart in figure 1. the effect of acceleration on the model wall was studied. moreover, the effect of acceleration on surcharge and pore water pressure were also investigated. fig. 1 flow chart of research methodology proceedings of engineering and technology innovation, vol. 15, 2020, pp. 24 34 26 the objectives of this study were to (a) conduct a series of shaking tests effectively under different base acceleration with fixed frequency, (b) fix a suitable scaling for constructing a dummy embankment, (c) observe the behavior of wrap -faced reinforced soil wall, (d) observe the seismic response of the scaled model embankment due to acceleration of base shaking and (e) determine the layer by layer response of soil wall due to these shakings. the specific purpose of the study was to evaluate the seismic response of constructed embankment model regarding the different input base accelerations with fixed frequency. 2. testing equipment’s 2.1. shaking table a single degree shaking table was used to simulate the horizontal shaking action, associated with seismic and other vibration conditions. the testing platform was made of steel (2 m by 2 m), had 1350 kg of payload capacity as shown in fig. 2. shaking was provided by a digitally controlled servo-hydraulic actuator. it had an acceleration range from 0.01g to 4g and frequency range from 0.05 hz to 50 hz with maximum amplitude of ±200 mm. fig. 2 laminar box mounted on the shaking table 2.2. used container the ideal container was one that gave a seismic response of the soil model identical to that obtained in the prototype. in this study, embankment with soft clayey soil models was constructed in a laminar box to reduce boundary effects as far as practicable. the laminar shear box developed at the laboratory was composed of 24 hollow aluminum layers of frames is presented in fig. 1. each layer consisted of an inner frame with inside dimensions of 915 mm × 1220 mm × 1220 mm. the gap between the successive layers was 2 mm, and the most bottom layer was rigidly connected to the solid aluminum base plate of size 915 mm x 1220 mm in plan and 15 mm thickness. 2.3. air pluviation technique in order to achieve the same density characteristics for the sandy soil specimen, three common techniques are obtained vibration, taming and pluviation. in this study, a portable traveling pluviator was used for achieving desired relative density, which was developed by hossain and ansary [20]. desired relative density could be achieved by changing the height of fall. 3. testing materials 3.1. clayey soil the clayey soil sample was collected at a depth of 1.5 m below the existing ground level from nearby campus of bangladesh university of engineering and technology, dhaka city of bangladesh. the liquid and plastic limits are used internationally for soil identification, classification and strength correlations. the liquid limit was determined by performing trials in which a portion of the sample is spread in a brass cup, divided in two by a grooving tool, and then allowed to flow proceedings of engineering and technology innovation, vol. 15, 2020, pp. 24 34 27 together from the shocks caused by repeatedly dropping the cup in a standard mechanical device. the plastic limit was determined by alternately pressing together and rolling into a 3.2-mm (1/8 in) diameter thread a small portion of plastic soil until its water content was reduced to a point at which the thread crumbles and can no longer be pressed together and rerolled. test procedures for obtaining the liquid and plastic limits are detailed in astm d 4318standard test methods for liquid limit, plastic limit and plasticity index of soils. three samples were prepared for measuring liquid limit (ll) and plastic limit (pl). three liquid limit (ll) and plastic limit (pl) were 36.5%, 37.0%, 39.5% and 13.5%, 14.0%, 16.5%, respectively. average liquid limit (ll) and plastic limit (pl) were established to be 38% and 15%, respectively. moreover, the average plasticity index was 23. the soil was defined as lean clay (cl), as per as uscs classification. the geotechnical property of the clayey soil is summarized in table 1. table 1 properties of clayey soils soil γ*unsat (kn/m 3 ) γsat (kn/m 3 ) c (kpa) ø (°) specific gravity (gs) ll pl permeability coefficient (ms −1 ) soil porosity clayey 15 18 26 20 2.62 38% 15% 3 × 10 −7 0.6 3.2. remolded clayey soil in this study, clayey soil was reconstituted by thoroughly mixing the oven-dried clayey soil powder with an initial water content equal to the ll [21]. the thorough mixing of the slurry was attained with the aid of a ‘hobart’ rotary mixer. with this slurry, a 300 mm reconstituted clayey soil layer was at first constructed in the laminar shear box; and then consolidated under the drained isotropic condition at loads: 15kpa, 20kpa, 25kpa, 30kpa, 40kpa, 60kpa, 80kpa and, 100kpa, respectively. the consolidation process was observed through the settlement of clayey soil versus time graph, plotted with the aid of calibrated mechanical dial gauges, instrumented on either side of the laminar shear box. the average settlement curve is shown in fig. 3. the value of effective stress (c') was 26 kpa obtained from the unconsolidated undrained (uu) triaxial compression test. fig. 3 settlement curve 3.3. sandy soil the sand material used during the experiments is named as “sylhet sand” which is locally found and it is widely used for highway embankments. the coefficient of uniformity (cu) and the coefficient of curvature (cc) were estimated as 2.16 and 0.66, respectively from this curve and the results indicating a poorly graded, medium to fine sand (sp). moreover, fineness modulus was determined to be at 2.66. other physical parameters such as specific gravity (gs) and maximum dry density (ɤd) were determined to be 2.64 and 16.8 kn/m 3 , respectively. strength parameters were determined by using the direct shear test. the test establishes an internal angle of friction ɸ of 32°. the properties of the target sandy soil is presented table 2. table 2 properties of the sandy soils soil γ*unsat (kn/m 3 ) γsat (kn/m 3 ) emax emin cu cc c (kpa) ø (°) specific gravity (gs) sandy 16.8 20 0.875 0.550 2.16 0.66 0 32 2.64 proceedings of engineering and technology innovation, vol. 15, 2020, pp. 24 34 28 3.4. reinforcing material a non-woven polypropylene multifilament geotextile (df50) was used for reinforcing the sand in the tests. the ultimate tensile strength of the geotextile was 15.5 kn/m. moreover, the ultimate tensile strength at 5% strain of the geotextile was 16.57kn/m. tensile properties of geotextile were determined by using the universal tensile testing machine. tensile test (astm d 4595) 1500 scale was used for wide width strip [22]. elongation of specimens with respect to gradually increased load was measured manually. the average initial modulus (ei) and secant modulus (es) in both directions (x and y) were 0.039 kpa, 0.1407 kpa, and 0.0467 kpa, 0.1230 kpa, respectively. several researchers [23-25] have investigated the degree of improvement achieved with different reinforcement lengths. as seen from the studies in the literature, there is no unique value proposed for the reinforcement length. furthermore, the reinforcement must have sufficient length and stiffness [26]. in this study the geotextile reinforcement length was 700 mm, to wrap each sand layer properly. 4. model design considerations the embankment model was properly scaled to achieve similitude with the prototype since it is not possible to predict the accurate behavior of the prototype with comparison to reinforced model wall in 1 g gravitational field [27-28]. the scaling factors for this study was exponent of confining pressure (α) = 0.5 for sand [22] considering the prototype to model scale being λ = 10 as presented in table 3. keeping in mind, the container size and the capacity of the shaking table, the value of λ was considered. the reliability of model tests depends on whether the model can represent the real behavior of the prototype system. model size in this study was fixed according to the facilities available for this test. precise scaling of soil walls and reinforcement properties is needed for considering the stress-dependent behavior of the soil and boundary conditions. table 3 scale relations and ratio of model system description parameter scale factor scale factor m/p scale factor p/m acceleration a 1 1 1 density ρ 1 1 1 length l 1/ λ 0.10 10 stress σ 1/ λ 0.10 10 strain g 1/ λ 1-α 0.32 3.125 displacement d 1/ λ 2-α 0.031 32.25 frequency f λ 1-α/2 5.62 0.18 time t 1/ λ 1-α/2 0.178 5.62 *p-prototype; m-model 4.1. soft soil foundations in the southern and central part (dhaka city) of bangladesh, the clay layer thickness is comparatively higher than others part of the country [29]. this clay layer thickness affects the foundation design. according to this, dhaka soft clayey soil was considered as the target prototype of the foundation soil. the present study was conducted with a thickness of 300 mm (12 inch) clayey soil layer foundation above which a 50 mm and the sand blanket was provided as shown in fig. 4. approximately 1 m2 geotextile was placed between the clayey soil foundation and sand blanket. considering the prototype to model scale being λ =10 and scale factor 1/ λ, the height of the prototype was 3 meters. 4.2. model embankment wall height is a significant factor of governing scale effects and the reaction of the model in contrast with the prototype. the higher height of sand wall provided better result. the average height of traditional walls usually ranges from 4.0 to 5 m [23]. many shaking table tests, on the different height of the sand wall such as 0.5 m [30], 0.6 m [22], 1 m [31], and 2.8 m [32] were performed at various time. a 400 mm high model with scale factor of 10 was constructed for this current study. the schematic geometry of the experimental model for this study is shown in fig. 5. the sand wall consisted of 4 (four) layers and proceedings of engineering and technology innovation, vol. 15, 2020, pp. 24 34 29 each layer was 100 mm. wrap-around type wall facing was used with flexible geotextile in the present study as shown in fig. 4. each individual facing was shaped by wrapping each layer with soft flexible geotextile element. however, in each individual test 60% relative density was kept by maintaining 150 mm height of fall. concrete slab was used as surcharge load at the top of the upper layer of the model wall is presented in fig. 4. three different types of 0.5 kpa, 1.0 kpa and 1.5 kpa surcharge loa ds were employed in this study. fig. 4 wrap-faced wall with lvdt 5. shaking table test 5.1. instrumentation fifteen sensors were used for this shaking table study; there are six accelerometers and four strain gauges, two pore pressure sensors and three displacement sensors. accelerometers, strain gauges, pore pressure and displacement sensors are notated with a, sg, p and lvdt, respectively. a1 was positioned on the shaking table to measure the input base motion. a2 was located on the clayey soil layer; a3, a4, a5 and a6 were placed inside the sand wall and located linearly at the different level. both the pore pressure sensors (p1 and p2) were placed on clayey soil layer. the four strain gauges were placed inside the embankment and attachment with geotextile. three displacement transducers (lvdt1, lvdt2, and lvdt3) were placed at different elevations from the sand blanket along the wall facing to measure horizontal displacement. a schematic view of the studied model with instrumentation plan is depicted in fig. 5. fig. 5 experimental setup design (all dimensions are in mm) proceedings of engineering and technology innovation, vol. 15, 2020, pp. 24 34 30 5.2. input motions this current study covers determination of the effects of acceleration amplitude, predominant surcharge and frequency of the motion on the seismic performance of reinforced embankment models. only 1 hz frequency was used in the present study which is less than the natural frequency (68 hz) of the embankment model. in order to do so, the applied dynamic motions cannot create any resonance. input dynamic motions consist of sinusoidal base motions of four acceleration amplitudes (0.05g, 0.10g, 0.15g and 0.20g) with three predominant surcharges (0.5 kpa, 1.0 kpa and 1.5kpa) and 1 hz frequency. different researchers [23] mentioned that this harmonic sinusoidal base acceleration is more aggressive than an archetypal earthquake with the same predominant frequency and amplitude. total 12 numbers of harmonic sinusoidal shaking were performed with using the designed laminar shear box for this research is presented in table 4. table 4 loading schemes of the shaking table tests test name acceleration, g frequency, hz surcharge load, kpa relative density, % rt1 0.05 1 0.5 60 rt2 0.1 1 rt3 0.15 1 rt4 0.2 1 rt5 0.05 1 1.0 rt6 0.1 1 rt7 0.15 1 rt8 0.2 1 rt9 0.05 1 1.5 rt10 0.1 1 rt11 0.15 1 rt12 0.2 1 6. results and discussions results were obtained from several shaking table tests on the embankment with soft clayey soil model is discussed in this section. reinforced-soil wall was constructed by using sand upon the clayey soil layer in equal lifts (el) of 100 mm (4 inch) to achieve a total wall height (h) of 400 mm. the length (l) of the geotextile reinforcement at the interface of the sand layers was kept the same in all tests as 510 mm (20 inch). after accomplishing the full setup, each wall model was subjected to 30 cycle’s sinusoidal motions of shaking at predetermined frequency, acceleration, amplitude and surcharge load as mentioned in table 3. 6.1. acceleration amplification response figs. (6)-(8) compared the acceleration amplification profile with the height of the wall for different surcharge (0.5, 1.0 and 1.5 kpa), base motion (0.05g, 0.1g, 0.15g and 0.2g) with fixed frequency (1 hz) and 60% relative density after each test. here, the elevation (s) was represented in non-dimensional form after normalizing by the full wall height (h). fig. 6 showed the acceleration amplifications along the height of the wall for different base accelerations of 0.05, 0.1, 0.15 and 0.2g from rt1 (0.05g, 1 hz and 0.5 kpa), rt2 (0.1g, 1hz and 0.5 kpa), rt3 (0.15g, 1hz and 0.5 kpa) and rt4 (0.2g, 1 hz and 0.5 kpa) model tests, respectively, which were conducted at 1 hz frequency, 0.5 kpa surcharge. maximum acceleration amplification was observed at the top of the wall in all the tests. fig. 7 showed the acceleration amplifications along the height of the wall for different base accelerations of 0.05, 0.1, 0.15 and 0.2g from rt1 (0.05g, 1 hz and 1.0 kpa), rt2 (0.1g, 1hz and 1.0 kpa), rt3 (0.15g, 1hz and 1.0 kpa) and rt4 (0.2g, 1 hz and 1.0 kpa) model tests, respectively, which were conducted at 1 hz frequency, 1.0 kpa surcharge. besides, fig. 8 showed the acceleration amplifications along the height of the wall for different base accelerations of 0.05, 0.1, 0.15 and 0.2g from rt1 (0.05g, 1 hz and 1.5 kpa), rt2 (0.1g, 1hz and 1.5 kpa), rt3 (0.15g, 1hz and 1.5 kpa) and rt4 (0.2g, 1 hz and 1.5 kpa) model tests, respectively, which were conducted at 1 hz frequency, 1.5 kpa surcharge. from the figures, it is clear that maximum acceleration amplification was observed at the top of the wall in the tests. moreover, acceleration amplifications were increased with increase of base acceleration in soil layer. within the range of tests proceedings of engineering and technology innovation, vol. 15, 2020, pp. 24 34 31 conducted, acceleration amplifications at the top of the wall for 0.2g base accelerations were 1.52, 1.41 and 1.29 for the surcharge 0.5, 1.0 and 1.5 kpa, respectively. this observation is in concurrence with the results of physical tests reported by krishna et al. [22]. fig. 6 effect of base acceleration on acceleration amplification (surcharge 0.5 kpa) fig. 7 effect of base acceleration on acceleration amplification (surcharge 1.0 kpa) fig. 8 effect of base acceleration on acceleration amplification (surcharge 1.5 kpa) 6.2. effect of acceleration on face displacement horizontal face displacement along the height of the wall was monitored by using three lvdts positioned is shown in figs. 4, (9)-(11) present the displaced face profiles from numerous shaking table test analysis. here elevation (s) and horizontal displacements (δh) are presented in non-dimensional from after normalizing them by the height of the wall (h). the normalized displacement profile for different base accelerations of 0.05g, 0.1g, 0.15g and 0.2g from tests rt1, rt2, rt3 and rt4 respectively, which were conducted at 1 hz frequency and 0.5 kpa surcharge pressure, is shown in fig. 9. and, the normalized displacement profile for different base accelerations of 0.05g, 0.1g, 0.15g and 0.2g from tests rt5, rt6, rt7 and rt8 respectively, which were conducted at 1 hz frequency and 1.0 kpa surcharge pressure, is shown in fig. 10. also, the normalized displacement profile for different base accelerations of 0.05g, 0.1g, 0.15g and 0.2g from tests rt9, rt10, rt11 and rt12 respectively, which were conducted at 1 hz frequency and 1.5 kpa surcharge pressure, is shown in fig. 11. the figures depict that the normalized displacements were relatively high at higher base accelerations at the normalized elevatio n of s/h= 0.85. a maximum horizontal displacement of 2.50% of the total wall height (h), for 0.5 kpa surcharge, was observed compared with 2.2% for 1.0 kpa surcharge and 1.5% for 1.5 kpa. however, this observation is also in concurrence with the results of physical tests conducted by krishna et al. [22]. 6.3. effect of acceleration on pore water pressure typical pore water pressure variations obtained from the tests are presented in figs. (12)-(14). the height of clayey soil layer (h) was taken as 300 mm in the case of pore water pressure. the variations of the pore water pressure from model tests 0.7 aa modify.grf acceleration amplification n o rm a li se d e le v a ti o n , z /h 1 1.1 1.2 1.3 1.4 1.5 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 test no. acce. rt1 0.05g rt2 0.1g rt3 0.15g rt4 0.2g 1.12aa.grf acceleration amplification n o rm a li se d e le v a ti o n , z /h 1 1.05 1.1 1.15 1.2 1.25 1.3 1.35 1.4 1.45 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 test no. acce. rt5 0.05g rt6 0.1g rt7 0.15g rt8 0.2g 1.72aa.grf acceleration amplification n o rm a li se d e le v a ti o n , z /h 0.99 1.02 1.05 1.08 1.11 1.14 1.17 1.2 1.23 1.26 1.29 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 test no. acce. rt9 0.05g rt10 0.1g rt11 0.15g rt12 0.2g proceedings of engineering and technology innovation, vol. 15, 2020, pp. 24 34 32 rt1, rt2, rt3 and rt4 with base accelerations 0.01g, 0.1g, 0.15g and 0.2g respectively for 1hz frequenc y and surcharge load of 0.50 kpa are shown in fig. 12. and, the variations of the pore water pressure from model tests rt5, rt6, rt7 and rt8 with base accelerations 0.01g, 0.1g, 0.15g and 0.2g respectively for 1hz frequency and surcharge load of 1.0 kpa is shown in fig. 13. besides, the variations of the pore water pressure from model tests rt9, rt10, rt11 and rt12 with base accelerations 0.01g, 0.1g, 0.15g and 0.2g respectively for 1 hz frequency and surcharge load of 1.5 kpa is shown in fig. 14. the pore water pressure response against base acceleration variation was directly proportional at s/h=0.85 as can be observed from the figures. the figures depict that pore water pressure increases with the increasing of acceleration. the maximum pore water pressure was 0.30, 0.29 and 0.27 kpa at base acceleration of 0.2g for the surcharge load 0.50 kpa, 1.0 kpa and 1.5 kpa respectively. moreover, this finding is similar to the results of zhang et al. [33]. fig. 9 effect of base acceleration on face displacement (surcharge 0.5 kpa) fig. 10 effect of base acceleration on face displacement (surcharge 1.0 kpa) fig. 11 effect of base acceleration on face displacement (surcharge 1.5 kpa) fig. 12 effect of base acceleration on pore water pressure (surcharge 0.50 kpa) fig. 13 effect of base acceleration on pore water pressure (surcharge 1.0 kpa) fig. 14 effect of base acceleration on pore water pressure (surcharge 1.5 kpa) 0.7dd.grf normalised face displacement, h/h (%) n o rm a li s e d e le v a ti o n , z /h 0 0.25 0.5 0.75 1 1.25 1.5 1.75 2 2.25 2.5 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 test no. acce. rt1 0.05g rt2 0.1g rt3 0.15g rt4 0.2g 1.12dd.grf normalised face displacement, h/h (%) n o rm a li s e d e le v a ti o n , z /h 0 0.25 0.5 0.75 1 1.25 1.5 1.75 2 2.25 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 test no. acce. rt5 0.05g rt6 0.1g rt7 0.15g rt8 0.2g 1.72dd.grf normalised face displacement, h/h (%) n o rm a li s e d e le v a ti o n , z /h 0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 test no. acce. rt9 0.05g rt10 0.1g rt11 0.15g rt12 0.2g pp 0.7kpa.grf pore water pressure (kpa) n o rm a li s e d e le v a ti o n , z /h 0.05 0.075 0.1 0.125 0.15 0.175 0.2 0.225 0.25 0.275 0.3 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 test no. acce. rt1 0.05g rt2 0.1g rt3 0.15g rt4 0.2g pp 1.12kpa.grf pore water pressure (kpa) n o rm a li s e d e le v a ti o n , z /h 0.025 0.075 0.125 0.175 0.225 0.275 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 test no. acce. rt5 0.05g rt6 0.1g rt7 0.15g rt8 0.2g pp 1.72 kpa.grf pore water pressure (kpa) n o rm a li s e d e le v a ti o n , z /h 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0 0.08 0.16 0.24 0.32 0.4 0.48 0.56 0.64 0.72 0.8 test no. acce. rt9 0.05g rt10 0.1g rt11 0.15g rt12 0.2g proceedings of engineering and technology innovation, vol. 15, 2020, pp. 24 34 33 7. conclusions in this paper, foremost efforts were devoted to constructing a small-scale embankment (1/10 reduced scale) and experimenting of simple shaking table test to study the seismic response, especially the influence of input acceleration. shaking table experiments were performed under twelve distinct dynamic motions under twelve single frequency motions (i.e. sine wave form) having four acceleration amplitudes, one fixed frequency and three predominant surcharge loads (0.50 kpa, 1.0 kpa and 1.50 kpa). based on the tests in this study, the following conclusions can be drawn: 1. acceleration response gets intensified with increase in base acceleration, the difference being more noticeable at higher elevations. however, as the base accelerations were increased, the acceleration amplifications were continued to increase. in all tests amplification of acceleration to maximum was observed at the top of the wall. moreover, in clayey soil layer amplification of acceleration escalated with escalation of base acceleration. acceleration amplifications at the top of the wall for 0.2g base accelerations were 1.55, 1.45 and 1.37 for the surcharge 0.50 kpa, 1.0 kpa and 1.50 kpa, respectively. in general, accelerations were amplified both for sand and clayey soil layer, at higher elevations and with high base acceleration. 2. however, the face deformations were high for high base acceleration at the top of the reinforcing layer and impact was insignificant at the other layers. normalized displacements were comparatively high at higher base accelerations at the normalized elevation of s/h= 0.85. a maximum horizontal displacement of 2.20% of the total wall height (h) for 0.50 kpa surcharge was observed and compared with 2.0% total wall height (h) for 1.0 kpa surcharge and that of 1.7% for 1.50kpa surcharge. generally, it can be said that accelerations were amplified (both in sand and clayey soil layer) at higher elevations and with low surcharge pressures. in general, the face deformations are high for low surcharge pressures. 3. the figures also illustrate that pore water pressure augmentation is proportional with the acceleration increment. the maximum pore water pressure was 0.30, 0.29 and 0.27 kpa at base acceleration of 0.2g for the surcharge load 0.50 kpa, 1.0 kpa and 1.50 kpa, respectively. however, the pore water pressures were high for high base shaking, and low surcharge pressures at higher elevations. although this study had limitations and performed with small scale, the results from this study provide useful directions in future regarding relative performance of reinforced soil wall on soft clay soil under various test conditions and provide guideline for further research. conflicts of interest the authors declare no conflict of interest. references [1] s. denine, n. della, m. r dlawar, f. sadok, j. canou, and j. c. dupla, “effect of geotextile reinforcement on shear strength of sandy soil: laboratory study,” studia geotechnica et mechanica, vol. 38, no. 4, pp. 3-13, 2016. [2] h. aldeeky, o. al hattamleh, and b. a. alfoul, “effect of sand placement method on the interface friction of sand and geotextile,” international journal of civil engineering, vol. 14, no. 2, pp. 133-138, march 2016. [3] a. joghataie and m. s. dizaji, “transforming results from model to prototype of concrete gravity dams using neural networks,” journal of engineering mechanics, vol. 137, no. 7, pp. 484-496, february 2012. [4] t. s. ueng, m. h. wang, m. h. chen, c. h. chen, and l. h. peng, “a large biaxial shear box for shaking table test on saturated sand,” geotechnical testing journal, vol. 29, no. 1, pp. 1-8, june 2005. [5] a. turan, s. d hinchbergerand, and h. el naggar, “design and commissioning of a laminar soil container for use on small shaking tables,” soil dynamics and earthquake engineering, vol. 29, no. 2, pp. 404-414, february 2009. [6] a. h.humaish, m. s. shamkhi, and t. k. al-hachami, “design, manufacturing and testing of small shaking table,” international journal of engineering & technology, vol. 7, no. 4.20, pp. 426-430, 2018. [7] k. t. chau, c. y. shen, and x. guo, “nonlinear seismic soil-pile-structure interactions: shaking table tests and fem analyses,” soil dynamics and earthquake engineering, vol. 29, no. 2, pp. 300-310, february 2009. proceedings of engineering and technology innovation, vol. 15, 2020, pp. 24 34 34 [8] m. yazdandoust, “study on seismic performance of reinforced soil walls to modify the pseudo static method,” international journal of civil and environmental engineering, vol. 9, no. 9, pp. 1248-1259, 2015. [9] h. i. ling, y. mohri, d. leshchinsky, b.christopher, k.matsushima, and h. liu, “large-scale shaking table tests on modular-block reinforced soil retaining walls,” journal of geotechnical and geoenvironmental engineering, asce, vol. 131, no. 4, pp. 465-476, april 2005. [10] n. srilatha, g. m.latha, and c. g. puttappa, “effect of frequency on seismic response of reinforced soil slopes in shaking table tests,” geotextiles and geomembranes, vol. 36, pp. 27-32, february 2013. [11] a. edinçliler and y. s. toksoy, “shake table tests to measure the dynamic performance of geotextile-reinforced embankment,” periodicapolytechnica civil engineering, vol. 61, no. 4, pp. 803-814, february 2017. [12] m. yazdandoust, “investigation on the seismic performance of steel-strip reinforced-soil retaining walls using shaking table test,” soil dynamics and earthquake engineering, vol. 97, pp. 216-232, june 2017. [13] z. zhou, j. lei, s. shi, and t. liu, “seismic response of aeolian sand high embankment slopes in shaking table tests,” applied sciences, vol. 9, no. 8, p. 1677, april 2019. [14] f. goktepe, e. celebi, and a. j. omid, “numerical and experimental study on scaled soil-structure model for small shaking table tests,” soil dynamics and earthquake engineering, vol. 119, pp. 308-319, april 2019. [15] e. ç elebi, f. göktepe, and a. j. omid, “seismic soil response of scaled geotechnical test model on small shaking table,” arabian journal of geosciences, vol. 12, no. 2, pp. 44, january 2019. [16] f goktepe, aj omid, and e celebi, “scaled soil-structure interaction model for shaking table testing,” acta physica polonica a, vol. 132, no. 3, pp.588-590, 2017. [17] j. chen, x. shi and j. li, “shaking table test of utility tunnel under non-uniform earthquake wave excitation,” soil dynamics and earthquake engineering, vol. 30, no. 11, pp. 1400-1416, november 2010. [18] w. cai, b. yu, and s. kaewunruen, “shaking table tests of suspended structures equipped with viscous dampers,” applied sciences, vol. 9, no. 13, pp. 1-16, june 2019. [19] j. ye and l. jiang, “simplified analytical model and shaking table test validation for seismic analysis of mid-rise cold-formed steel composite shear wall building,” sustainability, vol. 10, no. 9, pp. 1-18, september 2018. [20] m. z. hossain and m. a. ansary, “development of a portable traveling pluviator device and its performance to prepare uniform sand specimens,” innovative infrastructure solutions, vol. 3, no. 1, p. 53, june 2018. [21] n. z. m. yunus, d. wanatowski, n. a. hassan, and a. marto, “shear strength and compressibility behavior of lime-treated organic clay,” ksce journal of civil engineering, vol. 20, no. 5, pp. 1721-1727, july 2019. [22] a. m. krishna and g. m. latha, “seismic response of wrap-faced reinforced soil retaining wall models using shaking table tests,” geosynthetics international, vol. 14, no. 6, pp. 355-364, september 2007. [23] m. el sawwaf and a. k. nazir, “behavior of repeatedly loaded rectangular footings resting on reinforced sand,” alexandria engineering journal, vol. 49, no. 4, pp. 349-356, december 2010. [24] e. cicek, e. guler, and t. yetimoglu, “effect of reinforcement length for different geosynthetic reinforcements on strip footing on sand soil,” soils and foundations, vol. 55, no. 4, pp. 661-677, august 2015 [25] m. zarringol, and m. zarringol, “study on the impact of sand-clay bond in geo-grid and geo-textile on bearing capacity,” journal of sustainable development, vol. 9, no. 6, pp. 83-95, november 2016. [26] e. cicek, e. guler, and t. yetimoglu, “effects of the first reinforcement depth on different types of geosynthetics,” scientia iranica, vol. 26, no. 1, pp. 167-177, 2019. [27] m. sabermahani, a. ghalandarzadeh, and a. fakher, “experimental study on seismic deformation modes of reinforced-soil walls,” geotextiles and geomembranes, vol. 27, no. 2, pp. 121-136, april 2009. [28] g. m. latha and a. m. krishna, “seismic response of reinforced soil retaining wall models: influence of backfill relative density,” geotextiles and geomembranes, vol. 26, no. 4, pp. 335-349, august 2008. [29] r. hore, m. r. arefin, and m. a. ansary, “development of zonation map based on soft clay for bangladesh,” journal of engineering, vol. 10, no. 1, pp. 13-18, march 2019. [30] k. watanabe, y. munaf, j. koseki, m. tateyama, and k. kojima, “behavior of several types of model retaining walls subjected to irregular excitation,” soils and foundations, vol. 43, no. 5, pp. 13-27, 2003. [31] m. m. el-emam, and r. j. bathurst, “influence of reinforcement parameters on the seismic response of reduced-scale reinforced soil retaining walls,” geotextiles and geomembranes, vol. 25, no. 1, pp. 33-49, february 2007. [32] h. i. ling, y. mohri, d. leshchinsky, c. burke, k. matsushima, and h. liu, “large scale shaking table tests on modular-block reinforced soil retaining walls,” journal of geotechnical and geo environmental engineering, vol. 131, no. 4, pp. 465-476, april 2005. [33] z. zhang, c. cho, q. pan, and x. lu, “experimental investigation on excess pore water pressure in soft soil-foundations under minor shocks,” international journal of engineering and applied sciences, vol. 5, no. 4, pp. 259-263, 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 7-v24(2023)-peti#11525(63-72).docx proceedings of engineering and technology innovation, vol. 24, 2023, pp. 63-72 development of mixing and pressing processes of split-gill mushroom spawn blocks worapong boonchouytan1,*, jaknarin chatthong1, nantapong pongpiriyadecha2 1department of industrial engineering, faculty of engineering, rajamankala university of technology srvijaya, songkhal, thailand 2department of industrial, rattaphum college, rajamankala university of technology srvijaya, songkhal, thailand received 09 february 2023; received in revised form 17 april 2023; accepted 24 april 2023 doi: https://doi.org/10.46604/peti.2023.11525 abstract this study aims to develop the mixing and pressing processes of split-gill mushroom spawn blocks through the development and construction of a semi-automatic mushroom spawn mixing. the developed machine uses a 0.5 hp motor to drive the mixing tank and the press cylinder, which are connected to a 1:60 reduction gear. the results show that the semi-automatic mushroom spawns mixing and pressing machine developed in this study are within the standard ranges, that the split-gill mushroom spawn blocks with an average weight of 598 g, an average height of 10.2 cm, and an average density of 0.33 g/cm3. as for production capacity, manual pressing produced 40 mushroom spawn blocks per hour while the developed machine produced 112 mushroom spawn blocks per hour, which is 2.8 times faster. keywords: machine, split-gill mushroom, mixing process, mushroom spawn block pressing 1. introduction fig. 1 shows split-gill mushroom, also known as schizophyllum commune, which is popular and widely distributed in many asian countries such as thailand, laos, myanmar, and northeast india [1]. it grows on dead or fallen broadleaf trees. its characteristic features are pale yellow to brown gills with dense white hairs and no stem [2]. fig. 1 split-gill mushroom in addition, this mushroom is a medicinal herb [3] with nutritional values, especially in terms of its high carbohydrate and protein content and low-fat content [4]. it is also high in schizophyllan, a compound with many biological functions [5] such as immunomodulatory, anti-inflammatory, antioxidant, anti-cancer, anti-viral, and anti-fungal, including prebiotic activities [6-9]. * corresponding author. e-mail address: worapong.b@rmutsv.ac.th proceedings of engineering and technology innovation, vol. 24, 2023, pp. 63-72 64 the production of split-gill mushroom spawn blocks starts from the mixing of raw materials. water is then added to the mixture to about 65% moisture content. the suitability of moisture content can be checked by holding and squeezing the sawdust substrate by hand. if the substrate remains globular when you open your hand, the moisture level is ideal. but if the substrate falls apart when opening your hand, the substrate is still too dry and some more water is needed to be added. about 600 g of the substrate is then packed into each bag with a folded bottom. the bag is lifted off and pressed against the ground lightly to make the mushroom substrate denser or press down with the hands. a bottleneck is put on the opening at the top of each bag. the opening of the bag is pulled downward and tied with a rubber band at the bottleneck to maintain its shape for further mushroom inoculation. a wooden stick is used to make a hole about 5 cm deep. the hole is then covered with cotton and paper and fastened with rubber, or you can use a cap instead of paper to prevent the cotton from getting wet while steaming. according to the production standards, the split-gill mushroom spawn blocks have an average weight of 590-610 g, with an average height of 9.5-10.5 cm, and an average density of 0.30-0.34 g/cm3 [10]. several studies have been conducted on the production of mushroom spawn blocks. for example, modestus et al. [11] a briquette machine was designed using a screw-type extruder to convert the processed water hyacinth plant and wastepaper into a solid briquette for domestic consumption. the designed machine has a production efficiency of 84% and also produced smoke-free wh briquettes with high resistance to mechanical action, better handling, and efficient fuel characteristics for household use. the briquettes are developed by utilizing the remaining bag logs of white oyster mushrooms (waste) to become charcoal briquettes as the alternative energy [12]. briquettes were made from a blend of corncobs and the bark of an oil palm trunk using a manual press. the significance of this study lies in the provision of a piece of baseline information to encourage local bio-energy development and serve as a guide for stakeholders in nigeria with a potential interest in investing in briquette technology [13]. preparation of biomass briquettes using durian peel char and spent mushroom compost char by using the screw press machine is utilized to form the briquette. the combined chars are fed into the extruder. the cylindrical briquette has an inner diameter of 5 cm and a height of 9 cm [14]. briquetting machine with vertical ring die. the briquetting machine was driven by an electric motor, so the spindle speed of the briquetting machine was controlled by adjusting the motor speed with a frequency converter [15]. therefore, to solve such problems, the researchers developed a semi-automatic mushroom spawn mixing and pressing machine covering the operations from mixing the mushroom substrate in the mixing tank, filling the mushroom substrate into bags, and pressing to obtain the mushroom spawn blocks. the mushroom spawn blocks obtained from the semi-automatic mushroom spawn mixing and pressing machine did not contain the split-gill mushroom spawn. a mushroom inoculation step was required. the technology of the semi-automatic mushroom spawn block mixing and the pressing machine was subsequently transferred to the split-gill mushroom farming community enterprise in tha kham sub-district, hat yai district, songkhla province. 2. research methodology 2.1. machine design and construction fig. 2 shows the diagram of the semi-automatic split-gill mushroom spawn mixing and pressing machine developed in this study [16]. this machine was made of angle steel as the main material and had a dimension of 925 × 1125 × 1500 mm. inside the machine frame (1), there was a motor and a gear reducer for driving the mixing tank and the press cylinder. the raw materials of mushroom substrate were put and mixed in the mixing tank (6). the substrate was discharged into a bag placed proceedings of engineering and technology innovation, vol. 24, 2023, pp. 63-72 65 over the cylindrical outlet (7) under the mixing tank. for the pressing unit, there was a long crankshaft (3) moving up and down simultaneously on each side, installed in front of the machine. this crankshaft was supported by a short rod (4). the height of the mushroom spawn block could be adjusted (5). the control box (14) for the motor was mounted on the side of the machine frame (1). fig. 3 shows the exploded diagram of the semi-automatic split-gill mushroom spawn mixing and pressing machine developed in this study. it consisted of a machine frame (1) and a motor (8) and belts (8a and 8b) inside this frame. the belt (8b) drove the reduction gear (10) to transmit power to the bevel gear (12). the bevel gear transmitted power to the crankshaft (13) with a gear ratio of 1:23 and the radius of the crankshaft dish was 10 cm. the power was then transmitted to the crankshaft made of a long steel rod (3) and to the support rod made of a short steel rod (4). the height of the spawn block could be adjusted using an adjustable screw (5) which was installed in front of the machine. another belt (8a) drove the reduction gear (9) to transmit power to the mixing tank (6). fig. 2 drawing of semi-automatic split-gill mushroom spawn mixing and pressing machine fig. 3 exploded diagram of semi-automatic split-gill mushroom spawn mixing and pressing machine fig. 4 shows the exploded diagram of the power unit of the semi-automatic split-gill mushroom spawn mixing and pressing machine developed in this study. the power unit was driven by a motor (8), which drives 2 belts. the first belt (8a) drove the reduction gear (9) with a gear ratio of 1:60 and transmitted power to the coupling (9a) for the impellers. the second belt (8b) drove the reduction gear (10) with a gear ratio of 1:60 and transmitted power to the shaft (11) driving the bevel gear (12) with a gear ratio of 1:1. fig. 4 power unit of semi-automatic split-gill mushroom spawn mixing and pressing machine fig. 5 shows the pressing unit of the semi-automatic split-gill mushroom spawn mixing and pressing machine developed in this study. the pressing unit consisted of a bevel gear (12), in which both arms of the bevel gear were connected to the crankshaft (13) with a gear ratio of 1:23 and the radius of the crankshaft dish was 10 cm. this section was connected to the crankshaft made of a long steel rod (3) and to the support rod made of a short steel rod (4). the height of the spawn block could be adjusted using an adjustable screw (5) which was installed in front of the machine. proceedings of engineering and technology innovation, vol. 24, 2023, pp. 63-72 66 fig. 5 pressing unit of semi-automatic split-gill mushroom spawn mixing and pressing machine fig. 6 shows the mixing unit of the semi-automatic split-gill mushroom spawn mixing and pressing machine developed in this study. the unit consisted of a reduction gear (9) with a gear ratio of 1:60 connected to a coupling (9a). the coupling was connected to the impellers (17) inside the mixing tank (6). after mixing, the substrate was discharged into a bag placed over the cylindrical outlet (7) under the mixing tank. fig. 6 mixing unit of semi-automatic split-gill mushroom spawn mixing and pressing machine 2.2. machine performance test the performance test of the split-gill mushroom spawn block mixing and the pressing machine was performed by comparing the weight, height, and density of the mushroom spawn blocks obtained from the conventional machine, manual pressing, and the developed machine. the details are as follows: (1) conventional mushroom spawn pressing machine is a standard machine for producing mushroom spawn blocks. (2) manual pressing is the traditional method for producing mushroom spawn blocks using human labor. the data and experimental results were also collected from a group of farmers who participated in this study. (3) developed machine is the machine designed and constructed in this study. the data and experimental results were collected for a group of farmers who participated in this study for further practical use as shown in fig. 7. (4) raw materials used as mushroom substrates are shown in table 1. (5) raw materials used as mushroom substrates were mixed in the mixing tank for machine testing with a moisture content of 65-75%. (6) a bag was then placed over the cylindrical outlet under the mixing tank. (7) the mushroom substrate in the bag was pressed using the press cylinder in front of the machine. (8) after pressing, the mushroom spawn block was removed from the machine and a cap was put on the opening at the top of the bag as shown in fig. 8. the data such as weight, height, and time were recorded in the table. proceedings of engineering and technology innovation, vol. 24, 2023, pp. 63-72 67 fig. 7 semi-automatic split-gill mushroom spawn mixing and pressing machine table 1 graph representations no. category volume (kg.) 1 sawdust 100.0 2 fine rice bran 7.0 3 sugar 1.0 4 gypsum 1.0 5 lime 1.0 6 epsom salt 0.2 7 water 45.0 (a) putting a cap on the opening at the top of the bag (b) split-gill mushroom spawn block obtained fig. 8 split-gill mushroom spawn block 3. results and discussion performance test of semi-automatic mushroom spawn block mixing and the pressing machine was performed by filling mushroom substrate with an average weight of 570-600 g into the bag and then pressing. the mushroom spawn blocks were obtained from the conventional machine, manual pressing, and developed machine. the data of the mushroom spawn blocks obtained from manual pressing is shown in table 2 and data of the mushroom spawn blocks obtained from semi-automatic mushroom spawn mixing and pressing is shown in table 3. 3.1. comparison of weight of mushroom spawn blocks a comparison of the weight of split-gill mushroom spawn blocks obtained from the conventional machine, manual pressing, and semi-automatic split-gill mushroom spawn mixing and pressing machine is shown in fig. 9. the average weight of split-gill mushroom spawn blocks obtained from the conventional machine, manual pressing, and semi-automatic split-gill mushroom spawn mixing and pressing machine was 588, 596, and 598 g, respectively. according to the production standards, proceedings of engineering and technology innovation, vol. 24, 2023, pp. 63-72 68 the weight of the split-gill mushroom spawn blocks ranges between 590-610 g [10], and the average weight of produced split-gill mushroom spawn blocks is within the standard range. when considering the weight of split-gill mushroom spawn blocks obtained from the conventional machine, it was found that the weight of split-gill mushroom spawn blocks was the lowest due to insufficient flow of raw materials. when observing the work of manual pressing, it was found that the weight was closest to the average due to the expertise of the work. fig. 9 comparison of weight of split-gill mushroom spawn blocks 3.2. comparison of height of mushroom spawn blocks the comparison of the height (cm) of mushroom spawn blocks obtained from the conventional machine, manual pressing, and semi-automatic split-gill mushroom spawn mixing and pressing machine is shown in fig. 10. the average height of split-gill mushroom spawn blocks obtained from the conventional machine, manual pressing, and semi-automatic split-gill mushroom spawn mixing and pressing machine was 10.8, 10.1, and 10.2 cm, respectively. according to the production standards, the height of the split-gill mushroom spawn blocks ranges between 9.5-10.5 cm [10], and the average height of produced split-gill mushroom spawn blocks is within the standard range. fig. 10 comparison of height of split-gill mushroom spawn blocks when considering the height of split-gill mushroom spawn blocks compressed by conventional machines, it was found that the height of the split-gill mushroom spawn blocks was the highest and exceeded the standard value of the mushroom mass because conventional machines could not adjust the distance of the compression shaft. when observing the operation of the semi-automatic split-gill mushroom spawn mixing and pressing machine, it was found that the weight value was closest to the average due to being able to adjust the distance of the compressed axis. proceedings of engineering and technology innovation, vol. 24, 2023, pp. 63-72 69 3.3. comparison of density of mushroom spawn blocks the comparison of the density (g/cm3) of mushroom spawn blocks obtained from the conventional machine, manual pressing, and semi-automatic split-gill mushroom spawn mixing and pressing machine is shown in fig. 11. an example of calculation for a mushroom spawn block with a diameter of 15 cm, a height of sawdust of 10.9 cm, and a weight of 597 g. volume can be calculated by 2 v r hπ= (1) when r is the radius of the circular cross-section, h is height. substitute the known values in the above equation can be derived � � � � 7.52 � 10.9 � 1.926.18 cm3. density can be calculated by m vρ = (2) when ρ is the density of the material, m i is the mass of the material, v is the bulk density of the material. substitute the known values in the above equation can be derived � � 597 1926.18⁄ � 0.31 g/cm3. fig. 11 comparison of density of split-gill mushroom spawn blocks the average density of split-gill mushroom spawn blocks obtained from the conventional machine, manual pressing, and semi-automatic split-gill mushroom spawn mixing and pressing machine was 0.31, 0.33, and 0.33 g/cm3, respectively. according to the production standards, the density of the split-gill mushroom spawn blocks ranges between 0.30-0.34 g/cm3 [10], and the average density of produced split-gill mushroom spawn blocks is within the standard range. 3.4. comparison of production capacity per hour from table 2, it was found that manual pressing resulted in the split-gill mushroom substrate blocks with an average weight of 596 g, and an average height of 10.12 cm and took about 1.48 min per block. the production capacity was 40.55 blocks or about 40 blocks per hour. from table 3, it was found that the semi-automatic split-gill mushroom spawn mixing and pressing machine resulted in the split-gill mushroom substrate blocks with an average weight of 598 g, and an average height of 10.19 cm, and took about 1.07 min per block. the production capacity was 56.07 blocks per hour. however, as the semi-automatic machine could produce 2 blocks at a time, the production capacity was 112 blocks per hour as shown in fig. 12. considering manual pressing by 1 worker, excluding time spent in sawdust mixing, it can be seen that manual pressing could produce 40 mushroom spawn blocks per hour while the developed machine could produce 112 mushroom spawn blocks per hour, which is 2.8 times faster. proceedings of engineering and technology innovation, vol. 24, 2023, pp. 63-72 70 table 2 data of the mushroom spawn blocks obtained from manual pressing no. weight of split-gill mushroom spawn block (g) time (min) height of split-gill mushroom spawn block (cm) 1 585 1.49.33 9.80 2 595 1.42.15 9.75 3 590 1.46.00 10.27 4 605 1.46.43 9.85 5 610 1.55.30 10.46 6 595 1.54.31 10.50 7 580 1.51.91 10.50 8 590 1.47.90 9.80 9 595 1.49.92 10.50 10 610 1.49.54 9.76 average 596 1.48.00 10.12 table 3 data of the mushroom spawn blocks obtained from the semi-automatic split-gill mushroom spawn mixing and pressing machine no. weight of split-gill mushroom spawn block (g) time (min) height of split-gill mushroom spawn block (cm) 1 605 1.12.27 10.2 2 600 1.00.75 10.3 3 600 1.03.26 10.2 4 595 1.05.17 10.3 5 605 1.09.85 10.2 6 585 1.07.48 10.1 7 590 1.06.50 10.2 8 595 1.08.68 10.1 9 600 1.10.93 10.2 10 600 1.09.07 10.1 average 598 1.07.00 10.19 fig. 12 comparison of production capacity per hour 3.5. economic analysis in the production of mushroom spawn blocks, labor cost for manual pressing of mushroom spawn block is charged per block and the productivity is not enough to meet the demand. therefore, in this study, a semi-automatic mushroom spawn block pressing machine was developed and constructed to reduce the time and steps in mushroom spawn block production. it was found that the developed mushroom spawn pressing machine reduced production time and increased the production capacity to 2.8 times faster compared to manual pressing. the details of the cost comparison between the production of mushroom spawn blocks by manual pressing and semi-automatic mushroom spawn pressing machine and the payback period when labor cost per block is 0.27 baht are as follows. this semi-automatic mushroom spawn block pressing machine uses a 40 112 1.48 1.07 0 2 4 6 8 10 12 14 0 20 40 60 80 100 120 140 manual pressing developed machine t im e ( m in ) c a p a c it y ( b lo c k s) treatments capacity time proceedings of engineering and technology innovation, vol. 24, 2023, pp. 63-72 71 single 0.5 hp motor to drive the entire mechanism of the machine which consumes a total of 1,650 w (7.5 a) of power, if the semi-automatic mushroom spawn block pressing machine operates 4 h a day, and 24 days a month. (note: 1 unit of electricity meter equals 1,000 w.) in the above calculation, one worker working 4 hours a day can produce 720 mushroom spawn blocks a day, so the wage is 194.4 baht per day. the semi-automatic mushroom spawn block pressing machine requires 1 worker to work 4 hours a day to control the machine and it can produce 2,720 mushroom spawn blocks per day. after deducting electricity costs, overhead costs, labor costs, and depreciation of the machine, the cost is 530.73 baht per day and the payback period of the semi-automatic mushroom spawn pressing machine is 1 month and 25 days. 3.6. assessment of machine performance the results of the survey with 60 mushroom spawn block producers in the community using a questionnaire to rate the performance of the developed machine (when 5 = most satisfied, 4 = very satisfied, 3 = moderately satisfied, 2 = less satisfied, and 1 = least satisfied) are shown in table 4. table 4 summary of a survey of performance of the developed machine with a group of mushroom spawn block producers no. area average range 1 continuous production of the mushroom spawn blocks 5.00 1 2 density of the mushroom spawn blocks 4.30 7 3 safety in the mushroom spawn block production 4.50 5 4 production capacity of the machine 4.45 6 5 control of the automatic mushroom spawn pressing machine 4.70 3 6 ease of use of the machine 4.85 2 7 power consumption of the automatic mushroom spawn pressing machine 4.56 4 average 4.62 from table 4, the continuous production of the mushroom spawn blocks showed the highest score with an average score of 5, followed by the ease of use of the automatic mushroom spawn block pressing machine with an average score of 4.85, control of automatic mushroom spawn press machine with an average score of 4.70, power consumption of automatic mushroom spawn pressing machine with an average score of 4.56, safety in mushroom spawn block production with an average score of 4.50, the production capacity of the machine with an average of 4.45 and the density of the mushroom spawn block with an average score of 4.30. overall, the group of mushroom spawn block producers showed the highest level of satisfaction with the performance of the automatic mushroom spawn block pressing machine developed in this study. 4. conclusions this study developed the mixing s through the construction of a semi-automatic machine, the performance test was also performed. based on the experimental results, the following conclusions are made: (1) the developed machine produced the split-gill mushroom spawn blocks with an average weight of 598 g which is between 590-610 g of the production standard. (2) the developed machine produced the split-gill mushroom spawn blocks with an average height of 598 cm which is between 9.5-10.5 cm of the production standard. (3) the developed machine produced the split-gill mushroom spawn blocks with an average density of 0.33 g/cm3 which is between 0.30-0.34 g/cm3 of the production standard. (4) as for production capacity, manual pressing produced 40 mushroom spawn blocks per hour while the developed machine produced 112 mushroom spawn blocks per hour, which is 2.8 times faster. (5) the developed machine showed a payback period of 1 month and 25 days. proceedings of engineering and technology innovation, vol. 24, 2023, pp. 63-72 72 acknowledgments this research was funded by the science, research and innovation promotion fund for the year 2022 under the strategic plan, research and innovation at rajamangala university of technology srivijaya. conflicts of interest the authors declare no conflict of interest. references [1] k. yelithao, u. surayot, c. lee, s. palanisamy, n. m. prabhu, j. lee, et al., “studies on structural properties and immune-enhancing activities of glycomannans from schizophyllum commune,” carbohydrate polymers, vol. 218, pp. 37-45, august 2019. [2] c. hobbs, “the chemistry, nutritional value, immunopharmacology, and safety of the traditional food of medicinal split-gill fugus schizophyllum commune fr.:fr. (schizophyllaceae). a literature review,” international journal of medicinal mushrooms, vol. 7, no. 1&2, pp. 127-140, january 2005. [3] a. klaus, m. kozarski, m. niksic, d. jakovljevic, n. todorovic, and l. j. van griensven, “antioxidative activities and chemical characterization of polysaccharides extracted from the basidiomycete schizophyllum commune,” lwt-food science and technology, vol. 44, no. 10, pp. 2005-2011, december 2011. [4] v. basso, c. schiavenin, s. mendonça, f. g. de siqueira, m. salvador, and m. camassola, “chemical features and antioxidant profile by schizophyllum commune produced on different agroindustrial wastes and byproducts of biodiesel production,” food chemistry, vol. 329, article no. 127089, november 2020. [5] s. lee and c. s. ki, “inflammatory responses of macrophage-like raw264.7 cells in a 3d hydrogel matrix to ultrasonicated schizophyllan,” carbohydrate polymers, vol. 229, article no. 115555, february 2020. [6] d. smirnou, p. knotek, k. nesporova, d. smejkalova, v. pavlik, l. franke, et al., “ultrasound-assisted production of highly-purified β-glucan schizophyllan and characterization of its immune properties,” process biochemistry, vol. 58, pp. 313-319, july 2017. [7] b. du, y. yang, z. bian, and b. xu, “characterization and anti-inflammatory potential of an exopolysaccharide from submerged mycelial culture of schizophyllum commune,” frontiers in pharmacology, vol. 8, article no. 252, may 2017. [8] n. k. chandrawanshi, d. k. tandia, and s. k. jadhav, “nutraceutical properties evaluation of schizophyllum commune,” indian journal of science and research, vol. 13, no. 2, pp. 57-62, 2017. [9] k. zhong, l. tong, l. liu, x. zhou, x. liu, q. zhang, et al., “immunoregulatory and antitumor activity of schizophyllan under ultrasonic treatment,” international journal of biological macromolecules, vol. 80, pp. 302-308, september 2015. [10] n. saetang, r. ramaraj, p. pumisutapon, and y. unpaprom, “influence of hot water treatment on alternative mycoprotein from schizophyllum commune,” proceeding of the 3rd international conference on renewable energy, sustainable environmental & agri-technological and innovation, pp. 1-9, december 2021. [11] o. o. modestus, a. i. olabisi, t. e. boye, and e. o. benjamin, “development of water hyacinth briquetting machine,” international journal of scientific research in information systems and engineering, vol. 2, no. 1, pp. 43-47, april 2016. [12] a. hufad, purnomo, n. sutarni, a. rahmat, and j. sulistyo, “economic analysis of oyster mushroom waste utilization for alternative energy in community empowerment program,” test engineering and management, vol. 83, pp. 8680-8690, may/june 2020. [13] s. y. kpalo, m. f. zainuddin, l. a. manaf, a. m. roslan, and n. n. r. nik ab rahim, “techno-economic viability assessment of a household scale agricultural residue composite briquette project for rural communities in nigeria,” sustainability, vol. 14, no. 15, article no. 9399, august 2022. [14] k. suppalakpanya, r. nikhom, and s. nikhom, “preparation of biomass briquettes using durian peel char and spent mushroom compost char,” international energy journal, vol. 20, no. 4, pp. 621-628, december 2020. [15] s. chen, y. zhao, z. tang, h. ding, z. su, and z. ding, “structural model of straw briquetting machine with vertical ring die and optimization of briquetting performance,” agriculture, vol. 12, no. 5, article no. 736, may 2022. [16] w. boonchouytan and j. chatthong, semi-automatic mushroom briquette mixing and pressing machine, thai petty patent, 2203000949, april 21, 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 (http://creativecommons.org/licenses/by/4.0/).