Hybrid FEM–ML framework for multi-objective optimization of mechanical properties and surface quality in dissimilar AA6061-T6/AA6082-T6 friction stir welding

Friction stir welding (FSW) offers a high-quality solid-state joining method for aluminium alloys, but optimizing mechanical properties and surface quality in dissimilar AA6061-T6/AA6082-T6 joints remains challenging. This study introduces a hybrid FEM–ML framework for predicting and optimizing tensile strength, hardness retention, and surface roughness in dissimilar FSW. Physics-based FEM simulations generate thermal descriptors (peak temperature, cooling rate, thermal gradients) that are combined with process parameters to train machine learning models. Compared to parameter-only models, the hybrid approach improves prediction accuracy, achieving R² values of 0.99 for minimum hardness, 0.97 for surface roughness, and 0.90 for tensile strength. Multi-objective optimization identifies a robust, defect-free process window that yields an ultimate tensile strength of ~235 MPa, a minimum hardness of ~63 HV, and a surface roughness (Ra) < 5 µm.