An integrated RSM-ANN approach for predicting mechanical properties in bobbin tool friction stir welded dissimilar AA5052 and AA6061 alloys

This investigation employs response surface methodology (RSM) and artificial neural network (ANN) model to develop an integrated optimization and prediction for the mechanical characteristics of bobbin tool friction stir welded dissimilar AA5052-H32 and AA6061-T6 alloys. The effects of rotational speed (Rs: 450–650 rpm), travel speed (Ts: 80–100 mm/min), and tool shoulder diameter (Td: 6–8 mm) on ultimate tensile strength (UTS) and microhardness (MH) were investigated through central composite design. The quadratic regression models developed using RSM exhibited excellent accuracy, with R2 values of 99.96% for UTS and 99.97% for MH. The parameters Rs: 650 rpm, Ts: 80 mm/min, and Td: 8 mm produced maximum UTS of 219.12 MPa and MH of 122.12 HV, attributed to enhanced frictional heat, grain refinement, and dynamic recrystallization within the stir zone. ANN modeling further validated the framework, showing strong correlations (R = 0.983) between experimental and predicted values. Microstructural examination under optimal conditions confirmed improved metallurgical bonding, lamellar material flow, and homogeneous precipitate redistribution. Overall, the integrated RSM–ANN framework provides robust tool for process optimization, enabling the defect-free joining of dissimilar alloys with excellent performance for automotive, maritime, and aerospace fields.