Data-efficient optimization of LPBF 316 L stainless steel via domain knowledge-guided hybrid modeling

Parameter optimization in laser powder bed fusion (LPBF) remains challenging due to the high dimensionality of process variables and the cost of extensive experimentation. Here, a hybrid strategy synergizing physical insight with data-driven modelling is proposed to address the data scarcity. A central composite design (31 runs) with rotatable and symmetric characteristics was constructed and constrained within an intermediate laser volumetric energy density regime to fabricate 316 L stainless steel via LPBF. An iterative stepwise regression model with adaptive α-level tuning was developed to model the complex and nonlinear relationships between the key parameters (laser power, scanning speed, hatch spacing, layer thickness) and the resulting properties (relative density, tensile strength, yield strength, elongation), achieving R 2 values from 82.19% to 91.72%. The findings, derived from model analyses based on the specific experimental design range, unveil a transition in parameter dominance: from geometric factors alone governing density, to the set expanding to include energy inputs for mechanical properties. Experimental validation demonstrates optimized properties, including 99.96% relative density, 834 MPa tensile strength, 685 MPa yield strength, and 44% elongation. This work provides a resource-efficient and physically informed machine-learning framework for LPBF process optimization, offering practical guidance for high-performance additive manufacturing.