A strategy for contact fatigue life prediction of polymer gears via an experimental-simulated hybrid data-driven model

A timely trend in gear transmission involves the replacement of steel with polymers. Nevertheless, the absence of fundamental durability data for polymer gears impedes their reliable application during power transmission. The expensive and time-consuming gear fatigue experiments make it impossible to rely merely on experimental data. In this study, a strategy for contact fatigue life prediction of polymer gears via an experimental-simulated hybrid data-driven model is presented. The hybrid data are established with a certain mixture ratio of experimental and simulation data and are augmented by the conditional tabular generative adversarial network (CTAB-GAN) algorithm. This specific algorithm was combined with the extreme gradient boosting (XGBoost) algorithm to predict the contact fatigue life of gears made from different polymer materials, with the prediction accuracy controlled within a 3-fold scatter band. Moreover, an empirical predictive formula for contact fatigue life was developed. The hybrid data-driven model, which merges experimental and simulated data, allows for efficient estimation of fatigue life and material selection strategies, generating insight into the anti-fatigue design of polymer gears.