Taguchi-Based Optimization and Artificial Neural Network Modeling of Post-Weld Heat Treatment Effects on the Microstructural and Mechanical Performance of Friction Stir Welded AA3103 Alloy

This investigation evaluates the combined influence of Taguchi-optimized friction stir welding (FSW) parameters and post-weld heat treatment (PWHT) on the microstructural evolution and tensile strength of AA3103 aluminium alloy joints, complemented by Artificial Neural Network (ANN) modeling for mechanical property prediction. A Taguchi L9 orthogonal array was implemented independently for three tool pin geometries (square, circular, and triangular), incorporating shoulder-to-pin diameter ratios (D/d = 3, 4, 5), spindle speeds (355, 710, 1400 RPM), and welding speeds (20, 40, 80 mm/min), resulting in 27 experiments. A two-stage PWHT was applied, consisting of annealing at 350 °C for 2 h to promote dislocation recovery and residual stress relaxation, followed by artificial ageing at 200–225 °C and water quenching. Tensile properties, evaluated according to ASTM E8/E8M-22, were statistically analysed using S/N ratios, ANOVA, regression model, and ANN. The D/d ratio emerged as the dominant factor governing strength retention, showing the highest S/N delta for ultimate tensile strength and yield strength, and representing the only statistically significant parameter for yield strength. Spindle speed primarily influenced ductility stability. The optimized parameter set (circular pin, D/d = 3, 1400, 40 mm/min) demonstrated superior strength and microstructure. PWHT reduced regression residual errors to within ±8% and promoted fine equiaxed grain stabilization in the stir zone with a moderated hardness gradient across the TMAZ and HAZ.