Bayesian Optimization of FSW Process Parameters for Minimized Residual Stress

This study used deep learning to reduce residual stress in friction stir welded aluminum alloys 6061-T6 and 6063-T6. Residual stress was predicted superbly using a neural network model with Bayesian optimization, lowering it by 6.5% from an unoptimized scenario. The key variables that influence residual stress were tool plunge depth, at 38.3% influence, followed by tool rotation speed at 34.2%, and welding speed at 27.5%. The process parameters were optimized to 973 RPM, 74 mm/min, and 2.7 mm plunge depth, respectively, and a residual stress of 88.2 MPa was achieved. Data-driven process parameter strategies enhance the efficiency, quality, and cost of welding process and are appropriate for aerospace, automobile, and shipbuilding industries where structural integrity is paramount.