Application of machine learning approaches to predict joint strength of friction stir welded aluminium alloy 7475 and PPS polymer hybrid joint

Abstract

Vehicle weight has been a critical concern in the aerospace and automobile industries for decades. Integrating dissimilar aluminium and polymer hybrid structures is beneficial for weight reduction without affecting structural performance. In the present work, aluminium alloy 7475 and polyphenylene sulfide (PPS) sheets were joined using the friction stir welding (FSW) technology in lap joint configuration. A series of FSW experiments have been performed by the design matrix developed using response surface methodology. Tensile lap shear strength (TLS) is calculated for each experimental run. In this study, an attempt has been made to assess the potential of machine learning algorithms to predict the TLS of the joint. It was found that the support vector machine (SVM) model with RBF kernel was the most effective for predicting the TLS. Furthermore, FSW process parameters are optimized by means of the desirability approach. The optimal set to attain maximum TLS is identified as the tilt angle of 2°, welding speed of 5.12 mm/min and tool rotational speed of 1185.92 r/min. Finally, a confirmation test was performed to validate the optimal set and the adequacy of the developed SVM model. From the confirmation test, the error percentage between experimental and prediction values is less than 5%. Metallographic analysis revealed that the joining mechanism is the macro and micromechanical interlocking assisted by chemical bonding.