Direct Gaussian Process-Based Multi-Objective Bayesian Optimization of Selective Laser Melting Process Parameters

In selective laser melting (SLM), the relationships between process parameters and surface quality are highly nonlinear, while different quality indicators are often mutually conflicting, making multi-objective optimization under limited experimental budgets challenging. To address this issue, this study proposes a direct Gaussian process (GP)-based multi-objective Bayesian optimization framework for the efficient optimization of SLM process parameters. Targeted improvements in surrogate modeling and acquisition function design enable effective characterization of coupled multi-objective responses with a limited number of evaluations. Surface roughness ( S a ), defect area ratio, and macroscopic warpage root mean square were selected as optimization objectives based on laser confocal microscopy measurements, while laser power, scanning speed, and hatch spacing were treated as design variables. Comparative studies with the conventional GP-expected hypervolume improvement method and the evolutionary algorithm Non-dominated Sorting Genetic Algorithm II (NSGA-II) demonstrated that the proposed approach achieved faster convergence and superior Pareto front approximation under the same evaluation budget. From an engineering perspective, the proposed framework reduces reliance on extensive experiments and enables the identification of practically feasible parameter combinations, providing an efficient and transferable solution for intelligent SLM process optimization.