Simulation research on traction transmission gear modification based on machine learning

During the operation of high-speed power car transmission gears, issues such as uneven load distribution on the tooth flank and gear mesh impact may occur, which potentially lead to gear failure. To address these issues, a three-dimensional model of the gear transmission system, including the axle and bearing components, was developed in Romax software using the CRH3 as a case study. Based on the recursive feature elimination-extreme gradient boosting (RFE-XGBoost) feature selection model, the optimal combination of modification parameters was selected from the nine commonly adopted modification parameters. With the selected optimal modification parameters as input variables and the unit length normal load as the response variable, surrogate models were constructed based on the back propagation (BP) neural network, random forest regression and XGBoost algorithm respectively, and then compared and selected. Finally, the particle swarm optimization (PSO) algorithm was adopted to call the surrogate model, and the optimization objective was set as the minimum of the unit-length normal load. The optimization calculation was carried out for the parameter values of gear modification. The optimization results indicate that the best gear modification effect is achieved when the helix angle modification is 13.6 μm, and the tip relief length is 18.5 mm and the tip relief amount is 4.4 μm. Through Romax simulation calculation verification, the unit length normal load decreased by 24.36%, the transmission error value decreased by 69.91%, effectively improving the performance of gear transmission and alleviating the phenomenon of uneven load distribution on the tooth surface. The research results can provide references for the selection of gear modification schemes and the optimization design.