Integrating experiments and machine learning for high-temperature wear prediction in NiCrBSi/60WC laser claddings

Abstract This study investigates the high-temperature wear behavior of NiCrBSi/60WC coatings deposited on a steel substrate using laser cladding. The coated samples were evaluated using a high-temperature pin-on-disk tribometer under varying loads (20, 30, and 40 N), sliding speeds (2.5, 3.5, and 4.5 m s −1 ), and temperatures ranging from room temperature to 800 °C. At room temperature and 400 °C, the dominant wear mechanisms were oxidation and adhesion, whereas at 600 °C–800 °C, oxidative, abrasive, and adhesive wear modes were predominant. To predict the wear rate of the laser-cladded coatings on SS316L steel, several machine learning algorithms including decision tree, random forest, K-nearest neighbor, and extreme gradient boosting (XGB) were developed and assessed. Among these, the XGB model exhibited the best predictive performance, achieving the lowest root mean square error (RMSE) of 0.001 995. The XGB model was further optimized using grid search, random search, and Bayesian optimization (BO) techniques. The optimized XGB-BO model demonstrated the highest accuracy (94.30%) and the lowest RMSE (0.000 993), outperforming all other models in predicting the wear rate.