Machine learning informed additive manufacturing of stainless steel 410 using cold metal transfer-based metal inert gas welding

This study investigates the use of machine learning to predict bead geometry in cold metal transfer (CMT)–based wire-arc additive manufacturing (WAAM) of SS410 martensitic stainless steel. A dataset comprising 50 experimentally deposited single beads was developed indigenously to model the relationship between key process parameters and the resulting bead aspect ratio. Multiple regression and advanced tree-based ensemble models, including Random Forest, XGBoost, Extra Trees Regressor and Cat-Boost Regressor (CBR), were implemented to capture the influence of wire feed rate, deposition rate (torch travel speed), current and voltage on bead morphology. Comparative evaluation of the models, supported by independent validation demonstrated that the CBR model provides the most accurate prediction of aspect ratio among other models investigated in this study. Feature importance analysis indicated that welding current is the dominant parameter governing bead geometry, followed by voltage. Microstructural characterisation of the thick walls produced using the optimal process parameters revealed a progressive increase in δ-ferrite content with build height, associated with heat accumulation during multilayer deposition. Correspondingly, the ultimate tensile strength and yield strength decrease by 14% and 12.7%, respectively. These findings highlight the potential of machine-learning-based frameworks for predicting and optimising process–geometry relationships in WAAM, while also indicating the need for future AI-assisted strategies to control phase evolution and mitigate the formation of detrimental microstructural constituents in martensitic stainless steels.