Machine learning-assisted multi-objective optimization of direct laser metal deposition single tracks for thin-walled SS316L–Inconel 625 multi-material fabrication

Abstract

Direct laser metal deposition is an advanced additive manufacturing technique for producing multi-material components through localized energy input and controlled powder delivery. However, producing multi-material functionally graded thin-walled structures remains challenging because single-track geometry is strongly affected by both processing parameters and material composition. This research developed a machine-learning-assisted, composition-dependent, multi-objective optimization framework for geometrically consistent direct laser metal deposition of functionally graded structures ranging from stainless steel 316L to Inconel 625. A total of 162 single-track experiments were performed using six powder compositions, ranging from stainless steel 316L to Inconel 625, to evaluate the effects of laser power, scanning speed, and powder feed rate on single-track geometry and effective energy. Gaussian process regression provided the most consistent predictive performance among the investigated surrogate models and was integrated into a multi-objective optimization and decision-making framework to determine composition-specific process parameters. The optimization aimed to maintain bead width and deposited height near their target values while minimizing penetration depth and effective energy, with dilution screening used to ensure a metallurgically acceptable range. Experimental validation confirmed good agreement between the predicted and measured responses, with bead widths of 0.93–0.99 mm and deposited heights of 0.152–0.170 mm under the machine-adjusted optimized parameter sets. The validated parameters were then transferred to multilayer thin-walled fabrication. The fabricated stainless steel 316L and Inconel 625 graded wall exhibited widths of 1.253–1.385 mm and low longitudinal dimensional variation of 0.24%–1.31%, together with stable layer stacking, sound interlayer bonding, and a smooth microstructural and compositional transition. Energy-dispersive X-ray spectroscopy line scans and microhardness measurements further confirmed gradual elemental redistribution and a continuous increase in hardness from approximately 206.5 to 319.8 HV across the graded wall. These results demonstrate that the proposed framework provides a practical route for transferring optimized single-track parameters to stable thin-walled multi-material fabrication.