Friction torque prediction of tapered roller bearings under different assembly conditions based on XGBoost and SHAP

Friction torque serves as a critical metric for assessing the dynamic performance of tapered roller bearings (TRBs), influenced by factors such as input speed, temperature, preload force, radial load, and fit tolerance. Fit tolerances are design variables, while preload force, radial load, and speed are operational variables, with temperature acting as a response variable. This study proposes a friction torque prediction method for TRBs using the XGBoost algorithm enhanced by Bayesian optimization (BO). Additionally, other ensemble learning algorithms, including random forest, AdaBoost, CatBoost, LightGBM, and NGBoost, are employed for comparison. Performance is evaluated using root mean square error (RMSE), mean absolute error (MAE), and R -squared ( R 2 ). Experimental results indicate that XGBoost outperforms others during training. However, the differences of prediction performance among random forest, CatBoost, LightGBM, NGBoost, and XGBoost are minimal. When combined with SHapley Additive exPlanations (SHAP), speed is identified as the most significant factor affecting friction torque, followed by temperature.