Prediction of the remaining service life of a rolling bearing based on transfer learning
To address the issue of low generalization ability in predicting the remaining useful life(RUL)of bear-ings using the prediction model,and difficulties in accurately predicting the RUL of untrained bearings,this paper proposes a method to transfer knowledge of bearing conditions for RUL prediction.By utilizing computed time and frequency-domain features,and fuzzy entropy as predictive features,the bearing's entire life cycle is divided into normal and degraded stages employing the'3σ'criterion to achieve prediction of the RUL during the degradation stage.A bearing's RUL prediction model based on Gated Recurrent Units is constructed,trained on the full life cycle data of a particular bearing to enable the model to learn the state information of new bearings.Research indi-cates that,compared to methods not employing transfer learning,the root mean square error in predicting the RUL of all bearings decreased by 52.53%,while the average percentage error reduced by 68.87%.The proposed meth-od effectively and accurately predicts the RUL of bearings.
国家科技期刊平台
NSTL
万方数据
维普
