Fracture analysis of friction stir spot welded acrylonitrile butadiene styrene sheet in butt configuration

The friction stir spot welding (FSSW) process is a novel technique that overcomes the limitation of resistance spot welding. Recently, FSSW used for welding of polymers which are difficult to be joined by traditional welding processes. The demand for Acrylonitrile Butadiene Styrene (ABS) for industrial applications has increased in recent years. However, to employ this technique the challenge is to get optimal FSSW parameters setting to achieve the best weld strength during the welding of ABS sheets. To achieve this, in the present work, full factorial experimental design layout was employed to investigate the effect of process parameters on weld strength i.e., ultimate tensile strength (UTS) and percentage elongation during FSSW of ABS-ABS sheet in butt configuration. To predict the UTS and percentage elongation, machine learning regression namely, linear, polynomial, support vector machine, and decision tree was used. Further, the study includes the identification of the fracture patterns post tensile test specimens based on the topography of the fracture surface under scanning electron microscopy. It was found that plunge depth is the most significant parameter followed by spindle speed and dwell time. The optimal setting of process parameters i.e., spindle speed of 1000 rpm, plunge depth of 1 mm, and dwell time of 40 s resulted in maximum UTS of 7.849 MPa. The maximum value of percentage elongation obtained was 5 at the parameter setting of spindle speed of 1000 rpm, plunge depth of 0.8 mm, and dwell time of 40 s. Polynomial regression outperformed in the prediction of UTS and percentage elongation with an R-square of 0.99.


Introduction
Polymer joining methods include mechanical joining, adhesive bonding, and welding. Five types of welding exist, including solid-state welding. Solid-state welding can be friction welding, ultrasonic welding, or diffusion welding. Friction Stir Spot Welding (FSSW) uses friction between the base material and the tool to produce the heat required to soften the joint material [1]. In ferrous and nonferrous metals, and in plastics, FSSW is a newly developed solid-phase welding technique developed by The Welding Institute (TWI). A number of industries are increasingly using lightweight materials like Al alloys due to their formability, strength-to-weight ratio, corrosion resistance, and numerous other characteristics [2]. Friction-based welding processes produce highstrength welds, require less energy, and have a low impact on the environment. Another advantage of FSSW is its high reliability and lack of required pre-processing of the substrate. As the heating region is restricted FSSW there is low distortion in the material due to heat gradient and the mechanical behavior achieved is close to the base material. For FSSW welding, quality depends on stirring and upsetting effects and temperature distribution. The amount of frictional heat transferred to the material depends on the amount of frictional heat determined by the rotation speed of the tool, immersion speed, preheating time, and stirring time [3]. Many studies deal with friction stir welding (FSW) and FSSW of metal parts, but few focus on FSW and FSSW of polymers. FSW and FSSW on polymeric materials require appropriate processing conditions because metals and thermoplastics highest accuracy of 0.89, followed by KNN with an accuracy of 0.67, and AdaBoost resulted in the lowest accuracy of 0.57. Fleming et al [24] worked on the identification of faults such as tool imbalance and too much flash. An SVM-based model was used to indicate the existence of discrepancies and find out the depth of the gaps. The predicted results had an accuracy of 100% for each training and testing system. This demonstrates the effectiveness and accuracy of this technique, which can be used in a wide range of other FSW fault detection conditions. Adaptive Neuro-Fuzzy Inference System (ANFIS) has been successfully applied to predict the effectiveness of manufacturing processes. Roshan et al [25] report that there were very few studies on the optimization of FSW process using the ANFIS model, regardless of the fact that ANFIS is a valuable tool for prediction and optimization of many applications, including industrial production, machining, acoustics, and materials processing. Shehabeldeen et al [26] used ANFIS in tandem with Harris Hawk Optimiser (HHO) for the prediction of tensile strength. Using the HHO strategy, which predicts the FSW parameters and derives the optimal UTS, the experimental data, and ANFIS-HHO predictions correlate much better than ANFIS predictions. Elatharasan and Kumar [27] used the quadratic regression to assess the UTS, yield strength, and displacement of the FSW joint. Multi-objective optimization using the regression model is a beneficial method for improving the FSW parameters to achieve the best UTS, yield strength, and displacement of a joint at a 95% confidence level. When regression and ANN were compared, it was discovered that ANN is much more accurate and robust than regression models. Chekilil et al [28] made a quadratic regression model to predict yield strength, yield strength strain, UTS, ultimate tensile strength strain, rupture strength, and elongation. Their model performed with a 95% confidence level. Analysis of coefficients reveals increasing rotational speed induces a decrease in yield strength. Increasing rotation speed by 100% reduces the yield strength approximately by 9%. After crossing a critical value, an increase in feed rate increases the elastic limit of the joint.
Motivation behind the present study; Friction Stir Welding (FSW) was initially developed with the aim to join metallic materials that were difficult or impossible to weld using conventional fusion welding techniques. The numerous advantages of this process, alongside the increase of industrial demand for lightweight designed structures, naturally led to studying the possibility of using FSW for welding non-metallic materials. The commercially available joining techniques for joining polymers are usually limited to specific applications, customized design and materials. Hence, there is a scope in FSSW of polymer material particularly ABS material. The above listed studies have mostly focused on the FSW process for joining similar and dissimilar metals. The researchers working with polymers have most commonly used polypropylene, high density polypropylene, or polyethylene. Few studies have ventured into FSSW of ABS sheets in butt configuration. Therefore, this study aims to explore the application of FSSW of ABS sheets, and utilizing ML techniques to predict mechanical properties of the joint.
Following are the main contributions made by this paper: 1. This paper demonstrate weldability of ABS-ABS by FSSW.
2. The effect of three process parameters namely spindle speed, plunge depth, and dwell time on UTS and percentage elongation.
3. Fracture analysis of specimens ruptured during tensile testing.
4. Prediction of UTS and percentage elongation by regression and classification ML techniques.

Materials and methods
Present study work methodology includes performing FSSW of Acrylonitrile butadiene styrene (ABS) sheets on universal milling machine. ABS sheets are procured from Polestar Polymers, Mumbai, India. The material properties of ABS are shown in table 1. Next the FSSW joint strength was performed in terms of ultimate tensile strength (UTS) and percentage elongation by employing tensile test. ASTM D638 is employed for tensile test in the present study. Followed by fracture mode identification of tensile tested specimen using stereo images and scanning electron microscopy (SEM). Then prediction of UTS and percentage elongation by employing machine learning regression, as depicted in figures 1 and 2.  Figure 2 elaborates step by step methodology adopted to achieve the set objectives of the study. A universal milling machine of BFW make and UF-1 model was used to perform the process. The UF-1 has a table size of 590 l × 270 W × 340 H, and vertical head quill moment of 70 mm. The machine provides 12 speeds ranging from 45 rpm to 12000 rpm and also has 18 traverse speeds starting at 16 mm min −1 to 88 mm min −1 . Figure 3 shows the machine, the workpiece clamped ready for FSSW and the FSSW tool. An Aluminum, cylindrical pin tool was used to perform FSSW. The pin had a height of 1.2 mm and 3 mm diameter.
In this study 2 mm thick and 70 mm by 50 mm pieces of ABS were used to form butt welds. These butt welds were formed by performing double pass FSSW, the first pass had five spot welds and the second pass comprised 4 spot welds. These welds were equidistant from one another and ensured to lie at the center of the gap between adjacent spot welds on the other side of the pieces, refer figure 4.
The standard deviation and repeatability of a dataset were calculated. The dataset included input parameters such as spindle speed (rpm), plunge depth (mm), and dwell time (secs), and the output parameter was the ultimate tensile strength (MPa) shown in table 2. The standard deviation was calculated for each combination of input parameters, excluding samples where data was missing. The resulting standard deviation values ranged from 0.057 to 3.558 MPa. The average standard deviation across all input parameter combinations was 0.735 MPa, which indicates the average variability of the ultimate tensile strength measurement in this dataset. Figure 4(a) depicts the specimens obtained based on pilot study and figure 4(b) shows the specimens obtained based on main experiments trail conditions. However, trial conditions 18 (1400 rpm, 1 mm, 80 s), and   figure 5(b) were discarded from analysis due to no weld forming and being too weak for testing. Tensile testing was done on the universal testing machine (UTM) of VEEKAY TESTLAB (refer figure 6) at 10 mm min −1 speed and a load versus displacement graph was plotted. Based on these plots, the UTS and percentage elongation was calculated and tabulated in table 2. Table 2 also depicts the weld quality for each trail conditions. Weld quality of each samples are derived based on visual examination and stereoscopic images. A full factorial design was made with three parameters varying on three levels, the levels were decided based on pilot study, literature review, and machine capabilities. These include spindle speed (1000 rpm, 1400 rpm, 2000 rpm), plunge depth (0.8 mm, 0.9 mm, 1 mm), and dwell time (40 s, 60 s, 80 s). The resulting 27 trial conditions  were considered for FSSW of ABS-ABS sheet on the vertical milling machine with a cylindrical pin tool maintaining the plunge pattern described earlier. For initial trial conditions the joint strength was observed to improve with increase in plunge depth. This is because with deeper plunging of the tool the size of the heat affected region increases, thereby resulting in more material mixing in the stir zone.
To understand the fracture morphology of tensile tested specimens, stereoscopy and scanning electron microscopy was employed. SEM (model-SIGMA SEM-EDS and Zeiss make) was employed for capturing the fracture morphology of tensile fracture surfaces. Fracture morphology studies help in determining the cause of failure. The fracture surface exhibits topographies that corresponds to a fracture-mode. Fracture morphology studies is a method of identification of fracture mode using fracture surface topographies. Normally there are two type of fracture that we come across namely, ductile fracture and brittle fracture. Ductile fracture is correlated to presence to fine or moderate or elongated dimples at the fracture surfaces. Striations or microstructure-dependent fracture surfaces indicate the fatigue fractures. If cleavages or ridges or micro-voids or bright appearance are present at the fracture surface, this indicates the brittle fracture. In case of hydrogen embrittlement, quasi cleavages or inter-granular cracking's are observed.
The tensile test data points were used to build regression models for predicting UTS and percentage elongation. Linear regression (LR), Polynomial regression (PR), Decision tree regression (DTR), and Support Vector Regression (SVR) were considered as machine learning regressor for prediction. Figure 7 depicts the flowchart of machine learning regression model development. The first step starts with correlation analysis between FSSW process parameter with output responses namely, UTS and percentage elongation. Second step is to set the primary and secondary variables of the study. Third step is to divide the dataset into training and testing dataset. In present study 70% dataset is considered as training dataset and 30% dataset is considered as testing dataset. To get required accuracy and precision of model prediction, hyper parameters are tuned. Once the model is trained and required accuracy and precision is obtained, the model is tested with the training dataset. Then the model's performance is evaluated in terms of mean absolute error (MAE), mean square error (MSE), root mean squared error (RMSE) and R-sq.

Results and discussion
In the Friction Stir Spot Welding (FSSW) process tool is only plunged in to the material instead of moving the tool along the joint. This process includes three stages namely, tool pin penetration, melt pool mixing and indentation mark when tool pin exit. The process begins with the rotating tool slowly penetrate the sheets until the too shoulder touches the sheets. This results into frictional heat generation and it increases as the dwell time is increased. The frictional heat increases the temperature at the joint which softens the material. The tool pin rotation into the joint results into material flow in the radial and axial directions. The axial force applied by the tool shoulder results into the formation of solid weld. This allows the sheets to join. After the mixing stage the tool pin is withdrawn quickly from the joint. The recoil of tool pin is decided based on either the tool pin reached the final depth of penetration or once the dwell time is reached. Once the tool pin is withdrawn an exit hole is left. During the welding process, various zones are formed namely, stir zone, heat affected zone, thermos-mechanical affected zone and unaffected based material zone. New grain growth takes place in these zones. Welding process parameters namely, plunge depth, spindle speed and dwell time affects the heat input and mechanical deformation. These process parameters affect the mechanical behavior of ABS material.
In present work, pilot study was conducted to get an experience for the process and understand how different parameters affect the process. Spindle speed, plunge depth, and dwell time were the main focus process parameter of this study. Another variable considered in the pilot study was the pattern of spot welds, a joint with more spot welds will have better strength than a joint with fewer spot welds. The pilot experiments were done in two sets, in the first group of experiments, spindle speed (1400 rpm), plunge depth (1 mm), and dwell time (40 s) was constant. The spot weld pattern was varied with 4-3, 4-5, and 7-6 welds in the first and second pass respectively. In double pass FSSW the overlapping region gets heated and stirred twice resulting in least microhardness as shown in [29], however performing a second pass was necessary to increase overall strength of the joint. In the second group spindle speed (2000 rpm), plunge depth (1 mm), and weld pattern (4-5) were  constant and dwell time varied from 40 s to 60 s. It was observed that as the number of spot welds increases, the thermally affected regions start to overlap and result in material sticking to the tool pin. This is due to the already broken polymer chains being broken further because of agitation and heat generated by friction. In the second set of experiments it was observed that higher dwell resulted in better welds. Figure 4(a) shows the specimens welded of this pilot study for both the groups.

Tensile test
Tensile test was performed at room temperature with quasi-static condition. The load and corresponding displacements are captured simultaneously during the test. Figure 8 shows load versus displacement graphs generated from tensile test for specimen 7 with highest UTS, specimen 12 having strength closest to mean UTS and specimen 20 with lowest UTS. Trial conditions 7, 8, and 9 having a spindle speed of 1000 rpm, 1 mm plunge depth, and 40 s to 80 s dwell time resulted in best tensile strength of 7.849 MPa, 6.280 MPa, and 3.009 MPa respectively. Sample 20 with parameters 2000 rpm, 0.8 mm plunge depth, and 60 s dwell resulted in the weakest weld with strength of 0.153 MPa. Similarly, highest elongation observed was 5.0% for specimen 1, 3.92% for specimen 8, and 3.23% for specimen 7. The fact that change in the temperature level due to frictional heat affected the strain at stir zone and it was observed that the initiation and crack growth is located at the stir zone. Figure 8 depicts the load drop zone on the load versus displacement curve of FSSW ABS-ABS sheet. This may be possible due to various phases of the crack initiation at the stir zone and crack propagation in the joints during the tensile tests. To examine the phenomenon further investigations are carried out.
In present work three process parameters namely spindle speed, plunge depth and dwell time were consider for the FSSW of ABS sheet. The effect of these parameters on weld strength was studied using statistically analysis of obtained experimental UTS and percentage elongation values. Figure 9(a) shows the scatter plot matrix of FSSW parameters, UTS and percentage elongation. It shows the correlation between spindle speed, plunge depth, & dwell time with UTS and percentage elongation. It can be observed that the degree of correlation between UTS and dwell time is low. And the type of correlation is negative between UTS and dwell time. The degree of correlation between UTS and plunge depth is high. And the type of correlation is positive between UTS and plunge depth. The degree of correlation between UTS and spindle speed is low. And the type of correlation is negative between UTS and spindle speed. The similar type of inference is drawn for percentage elongation based on the scatter plot matrix. The degree of correlation between percentage elongation and dwell time is low. And the type of correlation is negative between percentage elongation and dwell time. The degree of correlation between percentage elongation and plunge depth is high. And the type of correlation is positive between percentage elongation and plunge depth. The degree of correlation between percentage elongation and spindle  speed is low. And the type of correlation is negative between percentage elongation and spindle speed. The similar conclusion is obtained based from tables 3 and 4 for UTS and percentage elongation, respectively. From table 3 it is found that the most significant parameter that has positive effect on UTS value is plunge depth based on t-stat & p-value. Whereas, spindle speed and dwell time has negative effect on UTS value based on t-stat & p-value. From table 4 it is found that the most significant parameter that has positive effect on percentage elongation value is plunge depth based on t-stat & p-value. Whereas, spindle speed and dwell time has negative effect on percentage elongation based on t-stat & p-value. Among the designed experimental layout for the study 1000 rpm spindle speed resulted in strong weld strength by generating sufficient frictional heat for ABS-ABS weld specimens. Excessive spindle speed leads to overheating of the tool, high heat input, higher inertial forces and grain growth in stir zone, thereby resulting in more welding defects, similar observation by [9][10][11][12]. This condition decreases the weld quality and weld strength. During FSSW, the increase in frictional heat generation and nugget size depends upon the dwell time. With lower dwell time the frictional heat generated will be insufficient which result in small nugget size thereby lower weld strength. On the other hand, higher dwell time generates higher frictional heat, grain growth in the stir zone & thermo mechanical affected zone which leads to more mechanical strain of the stir zone & thermo mechanical affected zone, thus lowering the weld strength, similar observation by [13,14]. The optimum value of dwell time among the designed experiments which resulted in higher UTS is 40 s. Plunge depth increased the frictional heat generation, increase in temperature, material softening around tool pin, lowering the material viscosity and increase in pressure exertion by the tool shoulder. This enhances the material flow during FSSW increasing the width of stir zone (nugget size) thereby, increasing the weld strength, similar observation by [2,3]. For present study the optimum value of plunge depth which resulted in maximum UTS value is 1 mm. In figure 9(b) the blue colour ball size indicates values of UTS. And the maximum UTS value of 7.849 MPa was obtained at 1000 rpm spindle speed, 1 mm plunge depth and 40 s dwell time. In figure 9(c) the red colour ball size indicates values of percentage elongation. And the maximum percentage elongation value of 5 was obtained at 1000 rpm spindle speed, 0.8 mm plunge depth and 40 s dwell time.
As per ASM Handbook Vol 11-Failure Analysis and Prevention, optical stereo microscopy can be used to examine and characterize fracture surface features as an initial examination [30]. The stereoscopic imaging technique is an optical microscopy method designed for low magnification. The facture surface topography is captured using stereo microcopy in conjunction with macro photography (CCD camera). The important feature present in the failed specimen is the topography of a fracture surface. This feature helps in determining the mode of failure of specimens. Table 5 depicts the optical stereoscopic images of the fracture surfaces after tensile test. Column I of the table 5 shows the failed specimens after tensile testing. Column 2 shows the cross section of the specimen after tensile test indicating the fracture surface. Column 3 indicates the stereo images of fracture surface for all specimens. These images depicts the macrography of the fracture surfaces. The fracture patterns observed from stereoscopic images includes mirror zone, mist zone and deformation zone. Mirror zone have features namely, smooth and flat surface having small crazes around the crack origin point. It is characterized by slow crack growth and crack size is inversely proportional to the square of the stress at fracture. The feature of mist zone is that it's having flat smooth area having slight change in surface texture in comparison to mirror zone. Mist zone has characteristics of slow to fast crack growth. Deformation zone has features namely, hackles, beach marks and striations. It has characteristics where the area of zone is directly proportional to the type of loading and the applied stress. Scanning electron microscopy (SEM) images are utilized to understand the topography of fracture in FSSW joint after tensile test. For studying the topography of fracture three specimens were considered namely, specimen no.7, specimen no.12 and specimen no.20 based on the UTS value. Figure 10 shows the fracture surface morphology of specimen no.7 at 10.00 K X magnification, where various spherical dimples were observed. The break of these dimples was the sign of fracture happened with degree of ductility of ABS sheet. The presence of dimples indicates the sound welding joints. The ductility nature resulted in the higher tensile strength of the joint and for specimen no.7 UTS value was 7.849 MPa. Figure 11 represents the fracture surface morphology of specimen no. 12 at 10.00 KX magnification, where mixture of dimples, cleavage and ridge fracture are observed. The ridge like structures are due to larger forces acting upon ABS sheets during FSSW. No dimples are observed at the fracture surface and fewer dimples in the joint section, indicating weaker bond strength and showing brittle nature of fracture. The UTS value for specimen no. 12 was 1.281 MPa. Figure 12 shows the fracture surface morphology of specimen no. 20 at 10.00    KX magnification. Here the fractograph is characterized by mixture of fan like shape pattern, micro voids, dimples, and brittle appearance. This indicates the step by step propagation of cracks during tensile test. The fracture surface with brittle appearance results in weak bond strength. Also no dimples are observed at the fracture surface which indicates the brittle nature of fracture. The UTS value for specimen no. 20 was 0.153 MPa.

Machine learning regressions
The tensile test results were combined in a data frame for analysis. A correlation analysis as seen from figures 13 (a) and (b) which is in line with relevant literature reveals plunge depth have highest correlation with both UTS and elongation of FSSW joints followed by dwell time and spindle speed. Based on these findings regression algorithms were applied for predictions of UTS and percentage elongation of FSSW joints. The regression models developed were-LR, PR, SVR, and DTR.
Linear regression finds a linear relationship between the dependent variable and the independent variables. The linear regression model provides a sloped straight line representing the relationship between the variables. Linear regressor proved to be adept at predicting elongation however failed to give satisfactory results for UTS as can be seen from quantile-quantile (Q-Q) plot in figure 14.
Polynomial regression just like linear regression gives relation of dependent and independent variables, however the resulting equation has a degree higher than one. Non-linear regression analysis shows promising prediction for UTS and percentage elongation, as seen from quantile-quantile (Q-Q) plot in figure 15. A degree 5 polynomial resulted in 0.999 R 2 and root mean square error (RMSE) of 0.00095. This degree was identified by comparing RMSE values of testing and training datasets of polynomials from second degree to 10th degree. Figure 16 shows test-train RMSE plots for both UTS and elongation. In both the cases training RMSE and testing RMSE show little difference after crossing 4th degree. Similar process was performed for elongation and lowest RMSE of 0.0025 with an R 2 of 0.884 for 5th degree polynomial.
SVM algorithm is used to find a hyperplane in an N-dimensional space that classifies the data points. The dimensionality of space is dependent on the number of features being considered. A hyperplane is a decision boundary with different categories falling on either side of it. The algorithm tries to generate the hyperplane with maximum margin, higher the margin with higher is the confidence in classification being correct. Figure 17 shows the QQ plot for UTS and percentage elongation, around 60% of data points are lying around the straight line, indicating the performance of the model. For SVR increasing gamma from 0.1 to 0.15 at an epsilon value of 0.25 drastically improved results taking R 2 from 0.378 to 0.416, the RMSE value also showed a slight improvement of 0.037 for UTS prediction. In the SVR model for elongation reducing gamma value was found to improve the fit, and change in epsilon value had very little effect on the results. The final R 2 and RMSE for both parameters were 0.416 and 1.197 for UTS and 0.665 and 0.641 for elongation. A deeper investigation is needed to understand the effect of hyper parameters on the model performance and finding the best values.
Decision tree generates a sequence of rules that can be used to classify data depending on data of attributes and classes. Even though it can solve regression, the decision tree is most commonly used for classification problems. A decision tree starts at a root node and branches before terminating with leaf nodes. As one moves through the tree categories start getting more closely related to one another. Complex decision trees do not generalize well and can be unstable resulting in vastly different trees due to small variations in data. DTR tends to over fit the data if no cap is put on the maximum depth of the tree. With a maximum depth of 5 the UTS model has an R 2 of 0.903 and RMSE of 0.486, and for elongation R 2 of 0.919 and RMSE is 0.315. Figure 18 depicts the QQ plot where maximum datasets are lying on the straight line. This shows that the model can predicts UTS and percentage elongation around 90% R 2 . Figure 19 and table 6 show comparison of performance measures of four regression models considered for the study. Based on the values of performance measures mean absolute error, mean squared error, root mean squared error and R-square. Absolut error is the difference between measured value and true value. The average of all absolute error is the mean absolute error. The least MAE obtained for UTS and percentage elongation is by polynomial regression model of 0.0005 and 0.0016, respectively. Followed by decision tree regression, support vector regression, and linear regression. Mean squared error measures the amount of error present in the models. It's the average squared distance between the measured value and predicted value. It's also known as mean square deviation, similar to variance. When the error in model is zero, it indicates the MSE is zero. For present study the least MSE obtained is for polynomial regression followed by DTR, SVR and LR. Root mean squared error is the standard deviation of the prediction errors or residuals. RMSE is the measure of spread out of residuals. Meaning how the data is lying around the line of best fit. For present study the model having least value of RMSE is polynomial regression followed by DTR, SVR and LR.       study the polynomial regression having fifth degree outperformed than other machine learning models for predicting UTS and percentage elongation of FSSW ABS-ABS sheet butt joint. The present study scope includes the weldability of ABS-ABS by FSSW process. Followed by studying the effect of process parameters namely spindle speed, plunge depth, and dwell time on UTS and percentage elongation. Fracture analysis of specimens ruptured during tensile testing. Prediction of UTS and percentage elongation by regression and classification ML techniques. In future the study will be extended for improving the weld strength by varying the FSSW tool profile. Further the optimization and prediction of UTS will be done by employing nature based optimization coupled machine learning techniques.

Conclusions
This work employed a novel friction stir spot welding technique with multi-fold objective while welding ABS-ABS sheet in butt configuration. Based on the obtained results following conclusions drawn: • The employed full factorial experimental design consisting of twenty seven trail condition having three process parameters of three levels resulted in successful FSSW of ABS-ABS sheets.
• Based on the scatter plot matrix, t-stat, p-values and correlation heat maps, it was found that plunge depth is most significant parameter followed by spindle speed and dwell time that affects the UTS and percentage elongation.
• The optimal setting of process parameters i.e., spindle speed of 1000 rpm, plunge depth of 1 mm, and dwell time of 40 s resulted into maximum UTS of 7.849 MPa. Maximum vale of percentage elongation obtained was 5 at parameter setting of spindle speed of 1000 rpm, plunge depth of 0.8 mm, and dwell time of 40 s.
• Machine learning regression was successfully implement while FSSW of ABS-ABS sheet for prediction of UTS and percentage elongation. Among the considered four regression model, polynomial regression of fifth degree outperformed in prediction of UTS and percentage elongation with R 2 of 0.99.
• It was found that 85% specimens (23 specimens out of 27 specimen numbers) had good weld quality as per visual examination and stereomicroscopy images.
• Scanning electron microscopic analysis revealed the facture patterns based on the topography of facture surface post tensile test. Specimen having maximum UTS value had failed due to ductile fracture which was confirmed by the presence of dimples on the fracture surface. Specimen having minimum UTS value had failed due to brittle fracture which was confirmed by the absence of dimples on the fracture surface, rather the facture surfaces had cleavages and ridges.
Thus, the process parameters spindle speed, plunge depth and dwell time decides the weld joint strength of FSSW of ABS-ABS sheet. Present study revealed that the heat input and processing temperature is controlled for FSSW of ABS-ABS sheet by decreasing the spindle speed & dwell time and increasing plunge depth to achieve maximum weld strength and good weld quality.

Data availability statement
All data that support the findings of this study are included within the article (and any supplementary files).