A data-driven friction coefficient model and its application in meshing efficiency prediction of heavy-duty gears

Abstract

Heavy-duty gears are extensively utilized in high-power equipment such as helicopters, ships, and commercial vehicles, often leading to significant frictional power losses. Accurate friction prediction is essential for designing energy-efficient transmission systems. This study proposes a data-driven model to predict the friction coefficient and applies it to estimate the meshing efficiency of heavy-duty gears. By training on friction test data under various lubrication conditions, an extreme gradient boosting (XGBoost) model is developed to predict the friction coefficient, with hyperparameters optimized through grid search and cross-validation. The model’s decision mechanism is interpreted using Shapley additive explanations, highlighting the influence of speed, load, surface roughness, and lubricant viscosity on the friction coefficient. When applied to predict meshing efficiency, the model is experimentally validated, achieving a maximum prediction error of 0.211 % and an average error of 0.108 %. The effects of major operating and geometrical parameters are analyzed, showing that meshing efficiency increases with higher speeds, torque, pressure angles, tip relief length, and lower addendum coefficients. The results indicate that proper parameter optimization and the use of high-viscosity lubricants can enhance the energy efficiency of heavy-duty gears.