1 Sustainable Marine Structures | Volume 05 | Issue 02 | September 2023 Sustainable Marine Structures https://ojs.nassg.org/index.php/sms Copyright © 2023 by the author(s). Published by Nan Yang Academy of Sciences Pte Ltd. This is an open access article under the Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0) License. (https://creativecommons.org/licenses/ by-nc/4.0/). DOI: http://dx.doi.org/10.36956/sms.v5i3.875 ARTICLE Enhancing Friction Stir Welding in Fishing Boat Construction through Deep Learning-Based Optimization Erfan Maleki1 Okan Unal2,3 Seyed Mahmoud Seyedi Sahebari4 Kazem Reza Kashyzadeh5* Nima Amiri6 1. Mechanical Engineering Department, Politecnico di Milano, Milan, 20156, Italy 2. Mechanical Engineering Department, Karabuk University, Karabuk, 78050, Turkey 3. Modern Surface Engineering Laboratory, Karabuk University, Karabuk, 78050, Turkey 4. Department of Mechanical and Manufacturing Engineering, Ontario Tech University, L1G 0C5ON, Oshawa, Canada 5. Department of Transport, Academy of Engineering, RUDN University, 6 Miklukho-Maklaya Street, Moscow, 117198, Russian Federation 6. Paul M. Rady Department of Mechanical Engineering, University of Colorado, Boulder, CO, 80309, United States ARTICLE INFO ABSTRACT Article history Received: 17 June 2023 Revised: 10 August 2023 Accepted: 18 August 2023 Published Online: 19 August 2023 In the present study, the authors have attempted to present a novel approach for the prediction, analysis, and optimization of the Friction Stir Welding (FSW) process based on the Deep Neural Network (DNN) model. To ob- tain the DNN structure with high accuracy, the most focus has been on the number of hidden layers and the activation functions. The DNN was devel- oped by a small database containing results of tensile and hardness tests of welded 7075-T6 aluminum alloy. This material and the production method were selected based on the application in the construction of fishing boat flooring, because on the one hand, it faces the corrosion caused by prox- imity to sea water and on the other hand, due to direct contact with human food, i.e., fish etc., antibacterial issues should be considered. All the major parameters of the FSW process, including axial force, rotational speed, and traverse speed as well as tool diameter and tool hardness, were consid- ered to investigate their correspondence effects on the tensile strength and hardness of welded zone. The most important achievement of this research showed that by using SAE for pre-training of neural networks, higher accu- racy can be obtained in predicting responses. Finally, the optimal values for various welding parameters were reported as rotational speed: 1600 rpm, traverse speed: 65 mm/min, axial force: 8 KN, shoulder and pin diameters: 15.5 and 5.75 mm, and tool hardness: 50 HRC. Keywords: Simulation and modelling Welding Friction stir welding Deep neural network Optimization *Corresponding Author: Kazem Reza Kashyzadeh, Department of Transport, Academy of Engineering, RUDN University, 6 Miklukho-Maklaya Street, Moscow, 117198, Russian Federation; Email: reza-kashi-zade-ka@rudn.ru https://creativecommons.org/licenses/by-nc/4.0/ https://creativecommons.org/licenses/by-nc/4.0/ http://dx.doi.org/10.36956/sms.v5i3.875 https://orcid.org/0000-0003-0552-9950 2 Sustainable Marine Structures | Volume 05 | Issue 02 | September 2023 1. Introduction One of the oldest methods of non-separable con- nections, which is widely used in various industries, is welding. In fact, large structures cannot be produced in one piece, such as the body of ships, airplanes, trains, subways, and ground vehicles [1,2]. Therefore, it is nec- essary to connect different parts to each other and create an assembly. Hence, depending on the application of the connection, different types of welding are used in various industries, each of which has advantages and disadvan- tages compared to the others [3]. For example, the biggest challenge for all types of welding methods is the creation of tensile residual stress in the welding area [4] or micro- structure change in the Heat-Affected Zone (HAZ), both of which lead to a decrease in joint strength [5,6]. In other words, these two areas are prone to damage and cracks. However, some additional disadvantages arise when an intermediate material is used for welding. Therefore, in large industries, it is usually tried to use welding methods that do not require an electrode or additional material, such as Resistance Spot Welding (RSW) in the automotive industry [7]. Moreover, the FSW process which was first presented by The Welding Institute (TWI) in 1991 [8], has emerged as an effective alternative to traditional Metal Inert Gas (MIG) welding for use in marine applications, particularly as the industry moves towards increased use of aluminum alloys. Based on the TWI report in 2007, 171 large organizations and companies received licenses to manufacture shipbuilding from aluminum extrusion by FSW process, especially, Al7075 is used to manufacture aluminum panels for deep freezing of fishing boats. In this regard, one of the biggest challenges and concerns in the manufacturing is to optimize the strength of welded joints via the welding process. The powerful combination of reduced weight from aluminum and increased strength of FSW welds can yield spectacular benefits for marine designs. Generally, FSW is well suited for marine appli- cations because of the nature of the welds [9]. During the last decades, the selection of the favorable parameters of this process to obtain the optimal properties of the welded parts is remained challenging and widely considered by many researchers. In this regard, the goal is to increase the quality and improve the strength of the connection, and for this purpose, various techniques have been used. The number of publications in this field is very large and a comprehensive study is required for a detailed review, which is beyond the scope of the present research. How- ever, some of the efforts made to optimize aluminum parts welded in this way are collected in Table 1. As presented in Table 1, scientists paid the most at- tention to data mining methods such as Taguchi Meth- od (TM), Linear Regression Method (LRM), Analysis of Variance (ANOVA), and Response Surface Method (RSM) to optimize different parameters of FSW process. In addition, the main goal was to improve the tensile strength and hardness of the connection. In recent years, attention has been paid to machine learning methods such as Artificial Intelligence (AI) for modeling and optimiz- ing friction stir welding of aluminum alloys. Neverthe- less, the present article includes the most comprehensive laboratory results and six input parameters. Also, the deep machine learning method was used to provide a new approach in optimizing the friction stir welding process of aluminum parts. Recently, the advantages and accu- racy of using Deep Neural Network (DNN) technique compared to Artificial Neural Network (ANN) have been evident and proven in various fields of engineering [28,29]. Padhy et al. have published a comprehensive review on the FSW technology and the effects of process param- eters on the material characterization and metallurgical properties like microstructure [30]. In addition, Gangwar and Ramulu have focused on the titanium alloys joints by FSW [31]. Also, to improve the quality of this type of welded joint, the effects of two parameters of the tool and the thickness of the raw material sheets have been evaluated. After that, microstructure, material properties (i.e., hardness), mechanical properties (i.e., tensile, fa- tigue, and fracture toughness), residual stress field, and temperature distribution have been studied. Mohanty et al. have utilized RSM to examine the effects of tool shoulder and probe profile geometries simultaneously on the FSW of aluminum sheets [32]. In this research, three parameters of tool type, tool probe diameter, and shoulder flat surface, each of them at three different lev- els, were considered as input to the RSM analysis. Also, connection strength, weld cross section area, and grain size in both welded area and HAZ were considered as output. The results showed that straight cylindrical FSW tool with the minimum level of probe diameter provided better strength of welded joint. Ahmed et al. have opti- mized FSW process parameters to achieve the maximum tensile strength and hardness of welded 5451Al sheets by performing Taguchi sensitivity analysis [33]. They stated that the pin profile of the tool is the most effective fac- tor in both outputs with more than 60% effectiveness. Moreover, research on the connection of thick aluminum plates through FSW has also been carried out [34]. Before this, in most articles, the connection of thin sheets up to 5 or 6 mm has been considered. In this regard, Kumar et al. have proposed a trapezoidal pin of tool to reach a free-defect connection of 12 mm thick aluminum sheets. 3 Sustainable Marine Structures | Volume 05 | Issue 02 | September 2023 In general, many parameters are important in industrial constructions, which are ultimately directly or indirectly related to the economic field and cost. For example, in the construction of a fishing boat, it is possible to refer to the initial costs including the preparation of the raw material i.e., aluminum alloy, the construction design in- cluding the dimensions and thicknesses of the sheets for the construction of the ship, and finally the construction method and the type of connection of the sheets to each other. On the other hand, considering the working con- ditions in the vicinity of sea water and the corrosion as a result, and on the other hand, in direct connection with fish and antibacterial issues, it is necessary to choose the type of alloy and heat treatment correctly. Finally, in or- der to reduce production time and ultimately reduce the cost of free-defect production, in order to minimize re- pairs, it is necessary to optimize the production process. Apart from these cases, just the free-defect construction is not enough, and it must have enough strength so that it does not have problems after some time of use [35]. For this purpose, in addition to examining the tensile strength, it is better to study the hardness and microstruc- ture in the welded area. Therefore, the results of such research can satisfy part of the demands stated above. Table 1. A summary of the research conducted to optimize the friction stir welding of aluminum parts. Reference Year Material Method Parameters Objective [10] 2015 AA6082-T6 Taguchi Joint geometries: butt, lap, and T-shaped Ultimate tensile strength [11] 2010 AA1050 Grey relation analysis and Taguchi Rotating speed, welding speed, and shoulder diameter Tensile strength and elongation [12] 2015 AA8011 Taguchi-Based Grey Relational Analysis Tool shoulder diameter, rotational speed, welding speed, and axial load Tensile strength and microhardness [13] 2016 Cast AA7075/SiCp Composite Response surface methodology, regression model, and Fuzzy grey relational analysis approach Spindle speed, travelling speed, downward force, and percentage of SiC added to AA7075 Ultimate tensile strength and percentage elongation [14] 2019 AA6082/SiC/10P composite Taguchi approach and analysis of variance (ANOVA) Tool rotation speed, welding speed, and tool tilt angle Ultimate tensile strength [15] 2018 Dissimilar alloys (AA6082/AA5083) Taguchi method, Grey relational method, weight method, and analysis of variance (ANOVA) Tool rotation speed, welding speed, tool pin profile, and shoulder diameter Ultimate tensile strength and elongation [16] 2008 RDE-40 aluminium alloy Taguchi technique and analysis of variance (ANOVA) Rotational speed, traverse speed, and axial force Tensile strength [17] 2010 AA7075-T6 Response surface methodology and analysis of variance (ANOVA) Tool rotational speed, welding speed, axial force, tool shoulder diameter, pin diameter, and tool hardness Tensile strength [18] 2017 Al5052-H32 alloy Response surface methodology Tool profile, rotational speed, welding speed, and tool tilt angle Tensile strength and elongation [19] 2012 Dissimilar alloy: AA6061-T6 and AA7075-T6 Response surface methodology Rotational speed, welding speed, and axial force Ultimate tensile strength, yield strength, and displacement [20] 2009 Cast aluminum alloy A319 Taguchi method and analysis of variance (ANOVA) Tool rotation speed, welding speed, and axial force Tensile strength [21] 2021 Dissimilar aluminum alloys 6061 and 5083 Response surface methodology Tool pin profile, tool rotation speed, feed rate, and tool tilt angle Ultimate tensile strength, yield strength, and microhardness [22] 2021 Dissimilar AA7075-T651 and AA6061 Taguchi technique and analysis of variance (ANOVA) Tool offset, pin profile of tool, and tilt angle of tool Tensile strength [23] 2022 Armor-grade aluminum alloys AA5083 Response surface methodology, regression, and analysis of variance (ANOVA) Shoulder diameter, shoulder flatness, pin profile, and welding speed Ultimate and yield tensile strength and elongation 4 Sustainable Marine Structures | Volume 05 | Issue 02 | September 2023 In the FSW, a tool moves along the joint line of two plates (similar or dissimilar) that simultaneously rotates and therefore it creates frictional heat that mechanically intermixes the metals and forges the hot and softened metal by the applied axial force. Figure 1 depicts the schematic of the FSW process and its effective parame- ters such as axial force, rotational and traverse speeds as well as tool geometry and hardness. As mentioned, in the FSW both mechanical and thermal processes are involved which show their effects in the welding zone and its sur- rounding regions and divide it into four major parts of Weld Nugget (WN), Thermo-Mechanically Affected Zone (TMAZ), Heat Affected Zone (HAZ), and Base Material (BM) that can be described as fully recrystallized region, area that plastically deformed without recrystallization, thermal affected, and region of original properties, respec- tively [36]. Because of the variety of physical phenomena in the FSW process, its analysis and optimization have become very complicated. Therefore, scholars have tried to solve the problems in this field by considering different alternative approaches of experiments like modeling and optimization methods [37]. In this regard, chu et al. have performed mechanical and microstructural optimization in the FSW joint of Al-Li alloy [38]. They used RSM and Box-Behnken experimental design to maximize static strength (i.e., tensile and shear stresses). Moreover, Elec- tron Backscattering Diffraction (EBSD) and Differential Scanning Calorimetry (DSC) observations have been utilized to optimize hardness and reach the ultrafine grains. Sreenivasan et al. have optimized FSW process parameters for joining composite materials (i.e., AA7075- SiC) [39]. They considered different parameters, including friction pressure, spindle speed, burn-off length, and upset pressure, in three levels, as the input variables in the opti- mization algorithms. Also, they tried to optimize hardness and Ultimate Tensile Strength (UTS). To achieve this goal, they employed two methods of linear regression and genetic algorithm. In a similar study, FSW process param- eters have been optimized to obtain the maximum joint strength and the highest elongation [40]. Also, Heidarza- deh et al. have employed ANOVA and RSM analysis to present mathematical relationships between the strength, elongation, and hardness of the connection in terms of FSW process parameters, including rotation speed, trav- erse speed, and axial force. They stated that the hardness reduces by raising the rotation speed and axial force and decreasing the traverse speed at the same time. Moreover, rotation speed and axial force are the most important fac- tors that affect the strength and elongation values, respec- tively. After that, Heidarzadeh continued his research in this field with a focus on material properties and imaging observations, including EBSD and TEM [41]. He et al. have published a review report on the numerical simulation of FSW process [42]. They discussed different techniques for process simulation and presented the results obtained from them. However, methods based on artificial intelli- gence such as neural networks are remarkably applied in different aspects of science and engineering [43-46] as well as their application for modeling of FSW process [47-49]. In general, a neural network has three major layers of input, hidden, and output [50,51]. A literature review conducted in this research shows that shallow Neural Network (SNN), as the primary generation of artificial neural networks, has been mostly used in the simulation of FSW. In fact, SNNs have 1 or 2 hidden layers which are generally trained by Back-Propagation (BP) algorithm [46]. These networks be- sides their beneficial applications have some limitations. The most important limitation of SNNs is that a large number of data sets are required for their development. Recently, by considering the improvements achieved in the training of neural networks by deep learning presented by Hinton et al. [52]; it is feasible to develop deep neural network with higher efficiency by employing small data set [53]. As an innovation, the current paper aims to show the capability of the DNN for prediction, analysis, and op- timization of the FSW process for the first time by using Reference Year Material Method Parameters Objective [24] 2021 Dissimilar aluminum alloys of AA 7075-O and AA 5052-O grade Taguchi approach and analysis of variance (ANOVA) Tool pin profile, tool rotational speed, feed rate, and tool offset Tensile strength and microhardness [25] 2018 Dissimilar AA5083-O and AA6063-T6 aluminum alloys Artificial intelligence and genetic algorithm Tool rotational speed, welding speed, shoulder diameter, and pin diameter Tensile strength, microhardness, and grain size [26] 2021 Dissimilar AA3103 and AA7075 aluminum alloys Taguchi method Tool rotation speed, feed rate, and tool pin profile Hardness, tensile strength, and impact strength [27] 2018 Armor-marine grade AA7039 Desirability approach: RSM & ANOVA Rotational speed, feed rate, and tilt angle Ultimate tensile strength and tensile elongation Table 1 continued 5 Sustainable Marine Structures | Volume 05 | Issue 02 | September 2023 small data set. In the following, the authors have attempt- ed to assess strength and hardness in the welded zone in terms of different process parameters as well as pin and shoulder features (i.e., hardness and diameter). 2. Experimental Data The data used in this research were extracted from the paper published by Rajakumar et al. [54]. In order to con- duct experiments, they prepared rolled sheets of 7075-T6 aluminum alloy with a thickness of 5 mm and dimensions of 150 × 300 mm. The ultimate and yield tensile strengths of the sheets are 485 and 410 MPa, respectively. Table 2 also presents the chemical composition of this material. Next, friction stir welding operation perpendicular to the rolling was performed as a single pulse and with non-con- sumable tools. The details of this process are shown in Figure 2. In the next step, tensile test samples were fabricated ac- cording to the ASTM standard. They prepared two types of smooth and notched samples. All tests were performed at room temperature and at a speed of 0.5 mm/min. How- ever, in the present study, only the laboratory results for smooth samples were used. Moreover, Vickers microhard- ness in the welded zone was measured by applying a force of 50 gr. Figure 1. Schematic illustration of the FSW process and parameters affecting welded joint quality. Table 2. Chemical composition of prepared sheets. Element Mg Mn Zn Fe Cu Si Al Value (%) 2.10 0.12 5.10 0.35 1.2 0.58 90.55 Figure 2. Dimensional details of sheets, sheets rolling and welding process directions. 6 Sustainable Marine Structures | Volume 05 | Issue 02 | September 2023 3. DNN Developing A DNN with four hidden layers was developed to mod- el the FSW process by using experimental data described in the previous section. This experimental dataset was selected because it includes high amounts of effective parameters of FSW process. Nevertheless, the data used for this simulation is given in the Appendix (Table A1). Also, the first author has used these data previously for providing a model based on the common SNN with BP method [48]. Hence, the innovation of the present work with the previous one is to present a novel modified neu- ral network based on deep learning. Parameters of axial force, rotational and traverse speeds as well as shoulder and pin diameters, and tool hardness were considered to investigate their related influences on the strength and hardness of welded zone. To this end, 25 and 5 data were used for the training and testing steps, respectively. In addition, in order to develop DNN, as the dataset is small, a Stacked Auto-Encoder (SAE) was used for pre-training of DNN. SAE is a specific hidden layer SNN that has the same input and output layers and also has the same num- ber of neurons in each layer of it with respect to the main DNN architecture. Used DNN along with linking to SAE is presented in Figure 3. Also, the accuracy of predicted results was determined via correlation coefficient (R2). The obtained accuracies for the developed DNN with and without SAE for both training and testing steps based on Equation (1) are reported in Table 3. common SNN with BP method [48]. Hence, the innovation of the present work with the previous one is to present a novel modified neural network based on deep learning. Parameters of axial force, rotational and traverse speeds as well as shoulder and pin diameters, and tool hardness were considered to investigate their related influences on the strength and hardness of welded zone. To this end, 25 and 5 data were used for the training and testing steps, respectively. In addition, in order to develop DNN, as the dataset is small, a Stacked Auto-Encoder (SAE) was used for pre-training of DNN. SAE is a specific hidden layer SNN that has the same input and output layers and also has the same number of neurons in each layer of it with respect to the main DNN architecture. Used DNN along with linking to SAE is presented in Figure 3. Also, the accuracy of predicted results was determined via correlation coefficient (2). The obtained accuracies for the developed DNN with and without SAE for both training and testing steps based on Equation (1) are reported in Table 3. Figure 3. Used DNN along with linking to SAE.  = =1  (,−)(,−) =1  (,−)2(,−)2 (1) Table 3. Obtained values of accuracies for developed DNN. Developed DNN Accuracy Without SAE-train 0.982 Without SAE-test 0.974 With SAE-train 0.996 (1) Figure 3. Used DNN along with linking to SAE. Table 3. Obtained values of accuracies for developed DNN. Developed DNN Accuracy Without SAE-train 0.982 Without SAE-test 0.974 With SAE-train 0.996 With SAE-test 0.993 7 Sustainable Marine Structures | Volume 05 | Issue 02 | September 2023 4. Results and Discussions Based on the results of the determination of accuracies that show the DNN was developed well, parametric anal- ysis was performed. In Figure 4 and Figure 5, the effects of welding process parameters and tool features on the welded zone strength and hardness are continuously re- vealed respectively in a general condition and interval of each parameter for the applied analyses are shown at the bottom of each contour. The region that has ≤ 95% of the considered experimental value was specified in each 2D contour by a black line. It can be seen that, in the same graph whole prediction, analysis, and optimization can be carried out with accuracy close to 100% (R2 ≈ 1). From Figure 4, as the rotation speed increases, the strength of the welded zone always increases, which is consistent with the results of Ahmed et al. in the study of 2022 [33]. Also, with the increase of the axial force, the strength im- proves, but its changes are not significant compared to the rotational speed parameter and it depends on the values of other parameters. Ref No. 33 also deals with the opti- mization of process parameters. The authors showed that the tensile strength of FSW joint considering the feed rate of 18 mm/min is much higher than the tensile strength of FSW joint with feed rates of 16 and 20 mm/min. In other words, the tensile strength of the connection increases with the increase of the feed rate in the range of 16 to 18 mm/min, and it decreases with the increase of the feed rate in the range of 18 to 20 mm/min. Therefore, this interpre- tation is exactly in accordance with the results presented in Figure 4. Moreover, the contour presented in Figure 4b indicates that if the rotational speed is less than 1000 rpm, changes in axial force do not affect the strength. However, all the results presented in this section are considering the welding conditions in this study and different results may be obtained in other conditions. Therefore, more studies are needed to generalize the results to other conditions, which is on the agenda of this research group for its future studies. Furthermore, by focusing on the hardness in the welded zone as a response, it is seen that the hardness increases with increasing rotational speed. But in interac- tion with other parameters, i.e., axial force and transverse speed, there is an intermediate area where the highest hardness is obtained within this specific area and the low- est hardness is obtained outside this area. For example, it is clear from Figure 4f that setting the rotational speed in the range of 1500 to 1700 rpm and also the axial force be- tween 7 and 9 KN can result in the highest hardness in the welded zone. As shown in Figure 5, no specific trends can be found between the system responses (i.e., strength and hardness in the welded zone) and tool characteristics, including pin and shoulder diameters and their hardness. In other words, the interaction between various parameters of this section must be checked with more precision and laboratory data and it is very difficult and perhaps unlikely to be able to independently declare the effect of one of these parame- ters on the output, because depending on the conditions of other parameters, there is a possibility of changing the state of the response. It is clearly evident in the contours with yellow color theme (i.e., Figure 5d, e, and f) that there are different circular and elliptical layers, therefore, there is a specific area in each contour that should be tried to select the characteristics of the tool in such a way that the responses be placed within these areas. In summary, according to all the achievements present- ed above, in order to obtain the optimal parameters, all the contours should be placed on each other at the same time and the common space between the areas marked by black lines should be selected. In this way, optimal values of rotational speed, traverse speed, axial force, shoulder diameter, pin diameter, and tool hardness are specified as 1600 rpm, 65 mm/min, 8 KN, 15.5 mm, 5.75 mm, and 50 HRC, respectively. 8 Sustainable Marine Structures | Volume 05 | Issue 02 | September 2023 Figure 4. Parametric analysis of the effects of axial force, rotational and traverse speeds on strength (a-c) and hardness (d-f) of welded zone. 9 Sustainable Marine Structures | Volume 05 | Issue 02 | September 2023 Figure 5. Parametric analysis of the effects of shoulder diameter, pin diameter, and tool hardness on strength (a-c) and hardness (d-f) of welded zone. 10 Sustainable Marine Structures | Volume 05 | Issue 02 | September 2023 5. Conclusions The application of DNN to model the FSW process was investigated in this study. The obtained results revealed that by using SAE for pre-training of neural networks higher accuracy can be obtained. In addition, it can be concluded that by applying DNN on small dataset with discrete values, continuous values for whole considered intervals can be pre- dicted for parametric analysis of FSW. Moreover, these 2D contours with the accuracy of close to 100% can be easily used for further analysis and optimization. However, one of the challenges in this research was that the small dataset was used to estimate the tensile strength, but more studies are needed to optimize the hardness, and no specific trend was found for hardness changes in the welded area according to the investigated parameters. Therefore, in future research, the authors seek to perform more tests using design of ex- periment (DOE) techniques such as Taguchi method and considering more input parameters. In addition, the authors agree with the opinion of many reports that the main causes of failures of mechanical parts are the fatigue phenomenon [55]. Therefore, in the future research series, this research team seeks to model, analyse, and optimize the friction stir weld- ing process of aluminium sheets with the aim of improving the fatigue life of the joint. For this, they will use various techniques including data mining, artificial intelligence, deep learning, etc. to make a comprehensive study. Also, the ac- curacy of the methods will be compared with each other and the advantages and disadvantages of each of them will be discussed. Author Contributions Conceptualization, E.M., O.U., and K.R.K.; method- ology, E.M., O.U., S.M.S.S., and K.R.K.; software, E.M., O.U., and S.M.S.S.; validation, E.M., and O.U.; formal analysis, E.M., O.U., and K.R.K.; investigation, E.M., O.U., S.M.S.S., K.R.K., and N.A.; resources, E.M., O.U., and K.R.K.; data curation, E.M., and O.U.; writing—orig- inal draft preparation, E.M., O.U., S.M.S.S., and K.R.K.; writing—review and editing, K.R.K.; visualization, E.M., O.U., and K.R.K.; supervision, E.M., O.U., and K.R.K.; project administration, E.M., and K.R.K.; funding acqui- sition, E.M., O.U., and K.R.K., All authors have read and agreed to the published version of the manuscript. Funding This research received no external funding. Acknowledgement This paper has been supported by the RUDN Universi- ty Strategic Academic Leadership Program. Data Availability The data that support the findings of this study are available from the corresponding author upon reasonable request. Conflict of Interest The authors declare no conflict of interest. References [1] Gite, R.A., Loharkar, P.K., Shimpi, R., 2019. Friction stir welding parameters and application: A review. Materials Today: Proceedings. 19, 361-365. DOI: https://doi.org/10.1016/j.matpr.2019.07.613 [2] Reza Kashyzadeh, K., Ghorbani, S., 2023. High-cy- cle fatigue behavior and chemical composition empirical relationship of low carbon three-sheet spot-welded joint: An application in automotive in- dustry. Journal of Design Against Fatigue. 2(1), 1-8. [3] Amiri, N., Farrahi, G.H., Kashyzadeh, K.R., et al., 2020. Applications of ultrasonic testing and machine learning methods to predict the static & fatigue be- havior of spot-welded joints. Journal of Manufactur- ing Processes. 52, 26-34. DOI: https://doi.org/10.1016/j.jmapro.2020.01.047 [4] Reza Kashyzadeh, K., Farrahi, G., Ahmadi, A., et al., 2023. Fatigue life analysis in the residual stress field due to resistance spot welding process considering different sheet thicknesses and dissimilar electrode geometries. Proceedings of the Institution of Me- chanical Engineers, Part L: Journal of Materials: De- sign and Applications. 237(1), 33-51. DOI: https://doi.org/10.1177/14644207221101069 [5] Farrahi, G.H., Ahmadi, A., Kasyzadeh, K.R., 2020. Simulation of vehicle body spot weld failures due to fatigue by considering road roughness and vehicle velocity. Simulation Modelling Practice and Theory. 105, 102168. DOI: https://doi.org/10.1016/j.simpat.2020.102168 [6] Farrahi, G.H., Ahmadi, A., Kashyzadeh, K.R., et al., 2020. A comparative study on the fatigue life of the vehicle body spot welds using different numerical techniques: Inertia relief and Modal dynamic analy- ses. Frattura ed Integrità Strutturale. 14(52), 67-81. DOI: https://doi.org/10.3221/IGF-ESIS.52.06 [7] Farrahi, G.H., Kashyzadeh, K.R., Minaei, M., et al., 2020. Analysis of resistance spot welding process parameters effect on the weld quality of three-steel sheets used in automotive industry: Experimental and https://doi.org/10.1016/j.matpr.2019.07.613 https://doi.org/10.1016/j.jmapro.2020.01.047 https://doi.org/10.1177/14644207221101069 https://doi.org/10.1016/j.simpat.2020.102168 https://doi.org/10.3221/IGF-ESIS.52.06 11 Sustainable Marine Structures | Volume 05 | Issue 02 | September 2023 finite element simulation. International Journal of Engineering. 33(1), 148-157. DOI: https://doi.org/10.5829/ije.2020.33.01a.17 [8] Friction Stir Butt Welding, International Patent Ap- plication No. PCT/GB92 [Internet]. International Patent Application No. PCT/GB92. GB Patent Appli- cation No. 9125978.8. Available from: https://cir.nii. ac.jp/crid/1570291224176390272 [9] Roldo, L., Vulić, N., 2019. Friction stir welding for marine applications: mechanical behaviour and mi- crostructural characteristics of Al-Mg-Si-Cu plates. Transactions on Maritime Science. 8(1), 75-83. DOI: https://doi.org/10.7225/toms.v08.n01.008 [10] Silva, A.C., Braga, D.F., De Figueiredo, M.A.V., et al., 2015. Ultimate tensile strength optimization of different FSW aluminium alloy joints. The Interna- tional Journal of Advanced Manufacturing Technolo- gy. 79, 805-814. DOI: https://doi.org/10.1007/s00170-015-6871-2 [11] Aydin, H., Bayram, A., Esme, U., et al., 2010. Appli- cation of grey relation analysis (GRA) and Taguchi method for the parametric optimization of friction stir welding (FSW) process. Materials and Technolo- gy. 44(4), 205-211. [12] Ghetiya, N.D., Patel, K.M., Kavar, A.J., 2016. Multi-objective optimization of FSW process param- eters of aluminium alloy using Taguchi-based grey relational analysis. Transactions of the Indian Insti- tute of Metals. 69, 917-923. DOI: https://doi.org/10.1007/s12666-015-0581-1 [13] Deepandurai, K., Parameshwaran, R., 2016. Multi- response optimization of FSW parameters for cast AA7075/SiCp composite. Materials and Manufactur- ing Processes. 31(10), 1333-1341. DOI: https://doi.org/10.1080/10426914.2015.1117628 [14] Bhushan, R.K., Sharma, D., 2019. Optimization of FSW parameters for maximum UTS of AA6082/ SiC/10P composites. Advanced Composites Letters. 28, 0963693519867707. DOI: https://doi.org/10.1177/0963693519867707 [15] Jain, S., Sharma, N., Gupta, R., 2018. Dissimilar alloys (AA6082/AA5083) joining by FSW and para- metric optimization using Taguchi, grey relational and weight method. Engineering Solid Mechanics. 6(1), 51-66. DOI: http://dx.doi.org/10.5267/j.esm.2017.10.003 [16] Lakshminarayanan, A.K., Balasubramanian, V., 2008. Process parameters optimization for friction stir welding of RDE-40 aluminium alloy using Tagu- chi technique. Transactions of Nonferrous Metals Society of China. 18(3), 548-554. DOI: https://doi.org/10.1016/S1003-6326(08)60096-5 [17] Rajakumar, S., Muralidharan, C., Balasubramanian, V., 2010. Optimization of the friction-stir-welding process and tool parameters to attain a maximum tensile strength of AA7075-T6 aluminium alloy. Pro- ceedings of the Institution of Mechanical Engineers, Part B: Journal of Engineering Manufacture. 224(8), 1175-1191. DOI: https://doi.org/10.1243/09544054JEM1802 [18] Shanavas, S., Dhas, J.E.R., 2017. Parametric optimi- zation of friction stir welding parameters of marine grade aluminium alloy using response surface meth- odology. Transactions of Nonferrous Metals Society of China. 27(11), 2334-2344. DOI: https://doi.org/10.1016/S1003-6326(17)60259-0 [19] Elatharasan, G., Kumar, V.S., 2012. Modelling and optimization of friction stir welding parameters for dissimilar aluminium alloys using RSM. Procedia Engineering. 38, 3477-3481. DOI: https://doi.org/10.1016/j.proeng.2012.06.401 [20] Jayaraman, M., Sivasubramanian, R., Balasubra- manian, V., et al., 2009. Optimization of Process Parameters for Friction Stir Welding of Cast Al- uminium Alloy A319 by Taguchi Method [Inter- net]. Available from: http://nopr.niscpr.res.in/han- dle/123456789/2786 [21] Verma, S., Kumar, V., 2021. Optimization of friction stir welding parameters of dissimilar aluminium alloys 6061 and 5083 by using response surface methodology. Proceedings of the Institution of Me- chanical Engineers, Part C: Journal of Mechanical Engineering Science. 235(23), 7009-7020. DOI: https://doi.org/10.1177/09544062211005804 [22] Yuvaraj, K.P., Varthanan, P.A., Haribabu, L., et al., 2021. Optimization of FSW tool parameters for join- ing dissimilar AA7075-T651 and AA6061 aluminium alloys using Taguchi Technique. Materials Today: Proceedings. 45, 919-925. DOI: https://doi.org/10.1016/j.matpr.2020.02.942 [23] Ghangas, G., Singhal, S., Dixit, S., et al., 2022. Mathematical modeling and optimization of friction stir welding process parameters for armor-grade al- uminium alloy. International Journal on Interactive Design and Manufacturing (IJIDeM). 1-18. DOI: https://doi.org/10.1007/s12008-022-01000-1 [24] Ramesha, K., Sudersanan, P.D., Santhosh, N., et al., 2021. Design and optimization of the process param- eters for friction stir welding of dissimilar aluminium alloys. Engineering and Applied Science Research. 48(3), 257-267. [25] Gupta, S.K., Pandey, K.N., Kumar, R., 2018. Artifi- https://doi.org/10.5829/ije.2020.33.01a.17 https://cir.nii.ac.jp/crid/1570291224176390272 https://cir.nii.ac.jp/crid/1570291224176390272 https://doi.org/10.7225/toms.v08.n01.008 https://doi.org/10.1007/s00170-015-6871-2 https://doi.org/10.1007/s12666-015-0581-1 https://doi.org/10.1080/10426914.2015.1117628 https://doi.org/10.1177/0963693519867707 http://dx.doi.org/10.5267/j.esm.2017.10.003 https://doi.org/10.1016/S1003-6326(08)60096-5 https://doi.org/10.1243/09544054JEM1802 https://doi.org/10.1016/S1003-6326(17)60259-0 https://doi.org/10.1016/j.proeng.2012.06.401 http://nopr.niscpr.res.in/handle/123456789/2786 http://nopr.niscpr.res.in/handle/123456789/2786 https://doi.org/10.1177/09544062211005804 https://doi.org/10.1016/j.matpr.2020.02.942 https://doi.org/10.1007/s12008-022-01000-1 12 Sustainable Marine Structures | Volume 05 | Issue 02 | September 2023 cial intelligence-based modelling and multi-objective optimization of friction stir welding of dissimilar AA5083-O and AA6063-T6 aluminium alloys. Pro- ceedings of the Institution of Mechanical Engineers, Part L: Journal of Materials: Design and Applica- tions. 232(4), 333-342. DOI: https://doi.org/10.1177/1464420715627293 [26] Raj, A., Kumar, J.P., Rego, A.M., et al., 2021. Opti- mization of friction stir welding parameters during joining of AA3103 and AA7075 aluminium alloys using Taguchi method. Materials Today: Proceed- ings. 46, 7733-7739. DOI: https://doi.org/10.1016/j.matpr.2021.02.246 [27] Verma, S., Gupta, M., Misra, J.P., 2018. Optimization of process parameters in friction stir welding of ar- mor-marine grade aluminium alloy using desirability approach. Materials Research Express. 6(2), 026505. DOI: https://doi.org/10.1088/2053-1591/aaea01 [28] Maleki, E., Unal, O., Seyedi Sahebari, S.M., et al., 2023. A novel approach for analyzing the effects of almen intensity on the residual stress and hardness of shot-peened (TiB+ TiC)/Ti–6Al–4V composite: Deep learning. Materials. 16(13), 4693. DOI: https://doi.org/10.3390/ma16134693 [29] Maleki, E., Unal, O., Seyedi Sahebari, S.M., et al., 2022. Application of deep neural network to predict the high-cycle fatigue life of AISI 1045 steel coated by industrial coatings. Journal of Marine Science and Engineering. 10(2), 128. DOI: https://doi.org/10.3390/jmse10020128 [30] Padhy, G.K., Wu, C.S., Gao, S, 2018. Friction stir based welding and processing technologies-process- es, parameters, microstructures and applications: A review. Journal of Materials Science & Technology. 34(1), 1-38. DOI: https://doi.org/10.1016/j.jmst.2017.11.029 [31] Gangwar, K., Ramulu, M., 2018. Friction stir weld- ing of titanium alloys: A review. Materials & Design. 141, 230-255. DOI: https://doi.org/10.1016/j.matdes.2017.12.033 [32] Mohanty, H.K., Mahapatra, M.M., Kumar, P., et al., 2012. Modeling the effects of tool shoulder and probe profile geometries on friction stirred aluminum welds using response surface methodology. Journal of Marine Science and Application. 11, 493-503. DOI: https://doi.org/10.1007/s11804-012-1160-z [33] Ahmed, S., Rahman, R.A.U., Awan, A., et al., 2020. Optimization of process parameters in friction stir welding of aluminum 5451 in marine applications. Journal of Marine Science and Engineering. 10(10), 1539. DOI: https://doi.org/10.3390/jmse10101539 [34] Kumar, D.A., Biswas, P., Tikader, S., et al., 2013. A study on friction stir welding of 12mm thick alu- minum alloy plates. Journal of Marine Science and Application. 12, 493-499. DOI: https://doi.org/10.1007/s11804-013-1221-y [35] Ganjabi, M.A., Farrahi, G., Reza Kashyzadeh, K., et al., 2023. Effects of various strength defects of spot weld on the connection strength under both static and cyclic loading conditions: empirical and numerical investigation. The International Journal of Advanced Manufacturing Technology. 127, 5665-5678. DOI: https://doi.org/10.1007/s00170-023-11923-y [36] Boukraa, M., Lebaal, N., Mataoui, A., et al., 2018. Friction stir welding process improvement through coupling an optimization procedure and three-di- mensional transient heat transfer numerical analysis. Journal of Manufacturing Processes. 34, 566-578. DOI: https://doi.org/10.1016/j.jmapro.2018.07.002 [37] Heidarzadeh, A., Chabok, A., Pei, Y., 2019. Friction stir welding of Monel alloy at different heat input conditions: Microstructural mechanisms and tensile behavior. Materials Letters. 245, 94-97. DOI: https://doi.org/10.1016/j.matlet.2019.02.108 [38] Chu, Q., Li, W.Y., Yang, X.W., et al., 2018. Micro- structure and mechanical optimization of probeless friction stir spot welded joint of an Al-Li alloy. Journal of Materials Science & Technology. 34(10), 1739-1746. DOI: https://doi.org/10.1016/j.jmst.2018.03.009 [39] Sreenivasan, K.S., Kumar, S.S., Katiravan, J., 2019. Genetic algorithm based optimization of friction welding process parameters on AA7075-SiC com- posite. Engineering Science and Technology, an In- ternational Journal. 22(4), 1136-1148. DOI: https://doi.org/10.1016/j.jestch.2019.02.010 [40] Heidarzadeh, A., Barenji, R.V., Khalili, V., et al., 2019. Optimizing the friction stir welding of the α/β brass plates to obtain the highest strength and elonga- tion. Vacuum. 159, 152-160. DOI: https://doi.org/10.1016/j.vacuum.2018.10.036 [41] Heidarzadeh, A., 2019. Tensile behavior, micro- structure, and substructure of the friction stir welded 70/30 brass joints: RSM, EBSD, and TEM study. Ar- chives of Civil and Mechanical Engineering. 19(1), 137-146. DOI: https://doi.org/10.1016/j.acme.2018.09.009 [42] He, X., Gu, F., Ball, A., 2014. A review of numerical analysis of friction stir welding. Progress in Materi- als Science. 65, 1-66. DOI: https://doi.org/10.1016/j.pmatsci.2014.03.003 https://doi.org/10.1177/1464420715627293 https://doi.org/10.1016/j.matpr.2021.02.246 https://doi.org/10.1088/2053-1591/aaea01 https://doi.org/10.3390/ma16134693 https://doi.org/10.3390/jmse10020128 https://doi.org/10.1016/j.jmst.2017.11.029 https://doi.org/10.1016/j.matdes.2017.12.033 https://doi.org/10.1007/s11804-012-1160-z https://doi.org/10.3390/jmse10101539 https://doi.org/10.1007/s11804-013-1221-y https://doi.org/10.1007/s00170-023-11923-y https://doi.org/10.1016/j.jmapro.2018.07.002 https://doi.org/10.1016/j.matlet.2019.02.108 https://doi.org/10.1016/j.jmst.2018.03.009 https://doi.org/10.1016/j.jestch.2019.02.010 https://doi.org/10.1016/j.vacuum.2018.10.036 https://doi.org/10.1016/j.acme.2018.09.009 https://doi.org/10.1016/j.pmatsci.2014.03.003 13 Sustainable Marine Structures | Volume 05 | Issue 02 | September 2023 [43] Maleki, E., Bagherifard, S., Razavi, S.M.J., et al., 2022. On the efficiency of machine learning for fa- tigue assessment of post-processed additively manu- factured AlSi10Mg. International Journal of Fatigue. 160, 106841. DOI: https://doi.org/10.1016/j.ijfatigue.2022.106841 [44] Maleki, E., Unal, O., Kashyzadeh, K.R., 2019. Sur- face layer nanocrystallization of carbon steels sub- jected to severe shot peening: Analysis and optimiza- tion. Materials Characterization. 157, 109877. DOI: https://doi.org/10.1016/j.matchar.2019.109877 [45] Maleki, E., Unal, O., Kashyzadeh, K.R., 2018. Fa- tigue behavior prediction and analysis of shot peened mild carbon steels. International Journal of Fatigue. 116, 48-67. DOI: https://doi.org/10.1016/j.ijfatigue.2018.06.004 [46] Reza Kashyzadeh, K., Amiri, N., Ghorbani, S., et al., 2022. Prediction of concrete compressive strength using a back-propagation neural network optimized by a genetic algorithm and response surface analysis considering the appearance of aggregates and curing conditions. Buildings. 12(4), 438. DOI: https://doi.org/10.3390/buildings12040438 [47] Dewan, M.W., Huggett, D.J., Liao, T.W., et al., 2016. Prediction of tensile strength of friction stir weld joints with adaptive neuro-fuzzy inference system (ANFIS) and neural network. Materials & Design. 92, 288-299. DOI: https://doi.org/10.1016/j.matdes.2015.12.005 [48] Maleki, E., 2015. Artificial neural networks applica- tion for modeling of friction stir welding effects on mechanical properties of 7075-T6 aluminum alloy. IOP Conference Series: Materials Science and Engi- neering. 103(1), 012034. DOI: https://doi.org/10.1088/1757-899X/103/1/012034 [49] Shojaeefard, M.H., Behnagh, R.A., Akbari, M., et al., 2013. Modelling and Pareto optimization of me- chanical properties of friction stir welded AA7075/ AA5083 butt joints using neural network and particle swarm algorithm. Materials & Design. 44, 190-198. DOI: https://doi.org/10.1016/j.matdes.2012.07.025 [50] Maleki, E., Bagherifard, S., Guagliano, M., 2022. Application of artificial intelligence to optimize the process parameters effects on tensile properties of Ti-6Al-4V fabricated by laser powder-bed fusion. International Journal of Mechanics and Materials in Design. 18(1), 199-222. DOI: https://doi.org/10.1007/s10999-021-09570-w [51] Kashyzadeh, K.R., Ghorbani, S., 2023. New neural network-based algorithm for predicting fatigue life of aluminum alloys in terms of machining parameters. Engineering Failure Analysis. 146, 107128. DOI: https://doi.org/10.1016/j.engfailanal.2023.107128 [52] Hinton, G.E., Osindero, S., Teh, Y.W., 2006. A fast learning algorithm for deep belief nets. Neural Com- putation. 18(7), 1527-1554. DOI: https://doi.org/10.1162/neco.2006.18.7.1527 [53] Feng, S., Zhou, H., Dong, H., 2019. Using deep neural network with small dataset to predict material defects. Materials & Design. 162, 300-310. DOI: https://doi.org/10.1016/j.matdes.2018.11.060 [54] Rajakumar, S., Muralidharan, C., Balasubramani- an, V., 2011. Influence of friction stir welding pro- cess and tool parameters on strength properties of AA7075-T6 aluminium alloy joints. Materials & Design. 32(2), 535-549. DOI: https://doi.org/10.1016/j.matdes.2010.08.025 [55] Arghavan, A., Reza Kashyzadeh, K., Asfarjani, A.A., 2011. Investigating effect of industrial coatings on fatigue damage. Applied Mechanics and Materials. 87, 230-237. DOI: https://doi.org/10.4028/www.scientific.net/ AMM.87.230 Appendix A The data used in this research, including various weld- Table A1. The data used in the current study, including welding conditions and testing results [48]. Case No. FSW process parameters Tool features Responses in welded zone Rotational speed Transverse speed Axial force Shoulder diameter Pin diameter Tool hardness Strength Hardness (rpm) (mm/min) (KN) (mm) (mm) (HRc) (MPa) (VHN) 1 1400 60 8 15 5 45 314 203 2 1800 60 8 15 5 45 310 185 3 1400 40 8 15 5 45 279 194 4 1400 60 8 15 5 45 310 198 5 1400 80 8 15 5 45 308 197 6 1400 60 7 15 5 45 282 180 7 1400 60 8 15 5 45 314 199 https://doi.org/10.1016/j.ijfatigue.2022.106841 https://doi.org/10.1016/j.matchar.2019.109877 https://doi.org/10.1016/j.ijfatigue.2018.06.004 https://doi.org/10.3390/buildings12040438 https://doi.org/10.1016/j.matdes.2015.12.005 https://doi.org/10.1088/1757-899X/103/1/012034 https://doi.org/10.1016/j.matdes.2012.07.025 https://doi.org/10.1007/s10999-021-09570-w https://doi.org/10.1016/j.engfailanal.2023.107128 https://doi.org/10.1162/neco.2006.18.7.1527 https://doi.org/10.1016/j.matdes.2018.11.060 https://doi.org/10.1016/j.matdes.2010.08.025 https://doi.org/10.4028/www.scientific.net/AMM.87.230 https://doi.org/10.4028/www.scientific.net/AMM.87.230 14 Sustainable Marine Structures | Volume 05 | Issue 02 | September 2023 ing conditions and laboratory results of tensile and micro- hardness tests, are given in Table A1 [48]. Case No. FSW process parameters Tool features Responses in welded zone Rotational speed Transverse speed Axial force Shoulder diameter Pin diameter Tool hardness Strength Hardness (rpm) (mm/min) (KN) (mm) (mm) (HRc) (MPa) (VHN) 8 1400 60 8 12 5 45 280 193 9 1400 60 8 15 5 45 310 198 10 1400 60 8 18 5 45 256 197 11 1400 60 8 15 4 45 292 194 12 1400 60 8 15 5 45 310 198 13 1400 60 8 15 6 45 300 197 14 1400 60 8 15 5 40 261 186 15 1400 60 8 15 5 45 313 198 16 900 60 8 15 5 45 245 175 17 1200 60 8 15 5 45 290 191 18 1400 20 8 15 5 45 255 180 19 1400 100 8 15 5 45 245 179 20 1400 60 6 15 5 45 263 173 21 1400 60 10 15 5 45 285 171 22 1400 60 8 9 5 45 242 178 23 1400 60 8 21 5 45 296 187 24 1400 60 8 15 3 45 264 181 25 1400 60 8 15 7 45 284 178 26 1400 60 8 15 5 33 271 178 27 1400 60 8 15 5 56 282 178 28 1400 60 9 15 5 45 301 190 29 1400 60 8 15 5 50 310 192 30 1600 60 8 15 5 45 314 202 Table A1 continued