Optimization of laser welding process parameters based on Chaos–NRBO–KELM and URCMO

Laser welding is widely applied in manufacturing, but parameter selection is difficult due to nonlinear relationship between parameter and quality. This study proposes an optimization method for laser welding process parameters based on Kernel Extreme Learning Machine (KELM) and the Unconstrained-Reliable Constrained Multi-Objective Optimization (URCMO) algorithms. An <i>L</i>25(5³) orthogonal design (304 stainless steel) is used to train a KELM model tuned by a Newton-Raphson-based optimizer. URCMO generates Pareto solutions, and entropy-weight TOPSIS selects the best trade-off. The optimal parameters, <i>P</i>=798<i>W</i>, <i>V</i>=22.5<i>m</i><i>m</i>/<i>s</i>, and defocus amount 0.75 mm, achieved a tensile strength of 63.4 MPa with good mechanical performance. The results demonstrate that the proposed KELM-URCMO framework effectively enhances prediction accuracy and parameter optimization efficiency, providing valuable guidance for intelligent control in laser welding applications.