Interpretable predictive modeling of laser cladding quality characteristics based on enhanced whale optimization algorithm-light gradient boosting machine and Shapley additive explanation

This study investigates the nonlinear relationship between laser cladding process parameters and the quality of resulting coatings. A full factorial experimental design was used for laser cladding 17-4PH alloy powder on 45# steel, with parameters including laser power, scanning speed, and powder feeding speed. Quality characteristics were evaluated based on cladding efficiency, dilution rate, molding coefficient, and cladding angle. A regression prediction model was constructed using a lightweight gradient lifter, and a multistrategy fusion enhanced whale optimization algorithm (EWOA) was developed to optimize light gradient boosting machine (LightGBM) hyperparameters. Results show that the EWOA-LightGBM model outperforms the unoptimized LightGBM model, with R² values of 0.9806, 0.9915, 0.9690, and 0.9676 for the quality property models. The Shapley additive explanation attribution analysis of the optimal model indicates the influence of various process parameters on the quality property model. In particular, the analysis reveals that the laser power exerts a positive influence on the quality model. The findings not only offer theoretical guidance and practical tools for predictive modeling of laser cladding but also provide novel solutions for the production of high-performance components in the manufacturing industry.