Machine learning application for evaluating the friction stir processing behavior of dissimilar aluminium alloys joint

Abstract

This paper reports on the employment of the machine learning (ML) techniques, namely support vector machine (SVM), artificial neural networks (ANN), and random forest (RF), for predicting the tensile behavior of friction stir processed (FSP) dissimilar aluminium alloys joints (6083-T651 and 8011-H14). The dissimilar aluminium joints are fabricated using the friction stir welding (FSW) process. After that, the friction-stir welded joints are subjected to the FSP procedure at different combinations of process parameters. The rotational speed, traverse speed, and tilt angle are used as the input parameters, while tensile strength and grain size are used as the output parameters. In addition, three performance characteristics (i.e., coefficient of correlation (CC), mean absolute error (MAE), and root mean square error (RMSE)) are used to check the adequacy of the developed model of ML techniques. It is observed that support vector machine_radial basis function kernel is the most accurate modeling technique for predicting the tensile behavior of processes samples. Furthermore, the optical microscope is also utilized to check the grain size of the nugget zone (NZ) of the weld bead for FSP. It is found that the minimum grain size (i.e., 5.06 µm) is obtained for the FSP sample and this grain size corresponded to the high ultimate tensile strength (UTS). Moreover, the fractographic analysis showed the ductile behavior of FSW and FSP samples.