Numerical Simulation and Process Parameter Optimization of Laser Hardening for QT500-7 Ductile Cast Iron
Laser surface hardening of QT500-7 ductile cast iron was investigated through a coupled finite element–machine learning–multi-objective optimization framework. A phase-transformation heat-transfer finite element model screened process windows for laser power (100–400 W), scanning speed (5–15 mm·s⁻¹), and overlap rate (60%–90%). A three-factor, three-level Box-Behnken design yielded hardened layer depth and fused layer depth as response variables. Four predictive architectures were benchmarked: Random Forest (RF), XGBoost, RF-XGBoost ensemble, and Bayesian-optimized RF-XGBoost (BO-RF-XGBoost). The BO-RF-XGBoost model achieved superior accuracy, with relative errors of 6.52% for hardened layer depth and 9.09% for fused layer depth. Multi-objective optimization compared Advantage Actor-Critic (A2C), Multi-Objective Particle Swarm Optimization (MOPSO), and Non-dominated Sorting Genetic Algorithm II (NSGA-II). A TOPSIS-entropy weight method ranked the Pareto front, identifying optimal parameters: laser power 230 W, scanning speed 14 mm·s⁻¹, overlap rate 75%. Experimental validation at these parameters produced a hardened layer depth of 230 μm and fused layer depth of 66 μm, with finite element model errors of 9.13% and 3.03%, respectively. Microhardness measurements showed the fused layer at 940 ± 40 HV0.5 and the hardened layer at 630 ± 30 HV0.5, both significantly exceeding the substrate hardness of 166 ± 15 HV0.5. The framework provides a reliable tool for parameter optimization in laser surface hardening of ductile cast iron.