Performance prediction and multi-objective optimization of 17-4PH stainless steel prepared by selective laser melting

Selective laser melting (SLM) process parameters critically influence the properties of fabricated components. However, establishing optimal parameters experimentally is challenging due to the complex nonlinear relationship between parameters and quality. To address this, this study proposes an intelligent parameter design method for 17-4PH stainless steel, integrating optimal Latin hypercube sampling (OLHS), machine learning, and multi-objective optimization. The variables considered are laser power, scanning speed, and scanning pitch. Comparisons with Newton-Raphson-based optimization (NBRO), Particle Swarm Optimization (PSO), and Gray Wolf Optimization (GWO) algorithms reveal that NBRO demonstrates superior predictive performance and convergence accuracy when automatically searching for optimal XGBoost hyperparameters. Its test set R 2 improved by 15.63%, 17.79%, and 16.38% for relative density, surface roughness, and microhardness predictions, respectively, compared to pre-optimization. Based on this, NSGA-III is employed for multi-objective optimization, combined with CRITIC-TOPSIS decision-making to obtain optimal process parameters. Experimental validation demonstrates that the prediction errors for relative density, surface roughness, and microhardness of the optimized specimens are only 0.20%, 5.45%, and 0.29%, respectively. This method significantly enhances process exploration efficiency and accuracy, providing a reliable solution for automated process design in additive manufacturing of high-performance material systems.