Numerical investigation and data-driven optimization of laser hardening process for U75V steel

Abstract

The laser hardening of U75V steel is governed by multiple process parameters, making effective process optimization challenging. To address this, a three-dimensional thermo-mechanical-metallurgical coupled finite element model was developed and integrated with a data-driven optimization method integrating a radial basis function (RBF) surrogate model and particle swarm optimization (PSO). Unlike conventional optimization approaches that mainly focus on hardened layer geometry or hardness-related indicators, the proposed strategy directly incorporates subsurface tensile residual stress as an optimization target, enabling coordinated control of hardening depth and residual stress state. The results demonstrate that the developed finite element model can reliably predict the evolution of the temperature field, microstructure, and residual stress distribution during laser hardening. The predicted hardened layer geometry agrees well with the experimental measurements, with agreement values of 91.9% for the surface hardened width and 92.3% for the maximum hardened depth. The simulated surface residual stress also shows good agreement with the x-ray diffraction measurements, with an average agreement of 91.6%. Based on the simulation dataset, the RBF surrogate model establishes an accurate nonlinear mapping between process parameters and response variables, while the PSO algorithm exhibits stable convergence behavior during the optimization process. The optimal process parameters were identified as a laser power of 1043.3 W and a scanning speed of 1.73 mm s−1. Under these conditions, a hardened layer depth comparable to that of the original process was maintained, while the maximum subsurface tensile residual stress was reduced by 17.1% in the X-direction and 4.6% in the Y-direction. To further improve the hardened layer uniformity in the initial scanning region, a 2 s laser preheating dwell was introduced at the scanning start position, improving the hardened layer depth in the initial region by 11%–22%, significantly enhancing the uniformity of the hardened layer along the scanning direction. This study provides a reliable and efficient simulation-driven approach for predicting and optimizing hardened layer depth and residual stress distribution in laser-hardened U75V steel components.