Machine learning-based characterization of friction stir welding in aluminum alloys

This research presents a machine learning (ML) approach that integrates Bayesian theorem to predict residual stress, plastic deformation, and peak temperature in friction stir welding (FSW) of different types of aluminum alloys. The training data was acquired from extensive numerical simulations of FSW conducted across various input parameter configurations. The regression analysis exhibits strong predictive performance for residual stress, peak temperature, and plastic deformation, achieving determination coefficients of 0.969, 0.955, and 0.919, respectively. The high prediction performance of the developed model is primarily attributed to Bayesian optimization, distinguishing it from other conventional ML methods. The results further demonstrate the efficacy of the proposed model in assessing the relevance of input features to prediction efficiency and output targets. For instance, concerning processing parameters, the predictive performance is closely correlated with the significance of preheating, traverse speed, and rotational speed in FSW, compared to welding force and plate thickness. On the other hand, considering material properties, it was observed that the hardness value of aluminum alloys emerged as a crucial predictor in the model so that alloys with moderate hardness values showed greater welding consistency and significantly enhanced prediction accuracy.