Modeling and simulation of friction stir welding process: A neural approach

Abstract

Friction Stir Welding (FSW) stands out as a groundbreaking method in solid-state joining for aluminum alloys, presenting an innovative way to achieve joints of exceptional quality. This research delves into the application of FSW for bonding, focusing on plates that are 6mm thick and made from aluminum alloys Al6063, Al5083, and AL6061, aiming to produce a variety of FSW joints. To evaluate the quality of these joints, the study compares mechanical properties such as tensile strength, safe bending strength, and bending toughness necessary for achieving a 90° bend. The investigation leverages welding data to formulate a neural model, starting with using a conventional feedforward neural model (CFNM). It tackles the limitations of CFNM, including its intensive training requirements and the challenge of dealing with unknown configurations, by proposing a new, more adaptable neural network model known as FNNM. When comparing the two models, it becomes evident that CFNM is constrained by a root mean square error (RMSE) of 7-15%, whereas FNNM marks a significant improvement with a minimal RMSE of 1-3%. This indicates that FNNM improves accuracy and effectively navigates the complexities of modeling with unknown parameters. Through this study, insightful contributions are made to understanding FSW in joining aluminum alloys and developing an advanced neural model capable of predicting the outcomes of welding with greater precision.