Heat-affected zone (HAZ) softening and corrosion degradation remain major obstacles to achieving property equivalence between welded joints and the base metal (BM) in thin 5052 aluminum alloy sheets. To overcome these limitations, a data-driven optimization strategy for cold metal transfer (CMT) welding was developed by coupling a Random Forest (RF) surrogate model with Bayesian active learning using an Expected Improvement (EI) acquisition function. This ML-guided approach successfully identified a set of optimal low–heat-input parameters. The optimized welded joints achieved tensile strengths exceeding 93% of that of the BM while maintaining comparable ductility and exhibiting nearly identical electrochemical behavior in acidic media. Integrated microstructural and compositional analyses reveal that weld metal grain refinement, spatial confinement of HAZ softening, and reduced compositional gradients collectively enable the simultaneous attainment of mechanical and corrosion property equivalence. This work provides a transferable machine-learning-assisted framework for parameter design and performance enhancement in welded thin aluminum alloy structures.
Wang et al. (Wed,) studied this question.