Fault Identification And Remaining Life Prediction Of Traction Motor Bearings In Urban Rail Transit

The running of urban rail transit train needs to meet the requirements of safety,reliability and comfort,and it cannot do without timely maintenance.Traditional traction motor test bench often focuses on testing the motor rotor coil circuit and insulation,and rarely involve testing the health of the traction motor bearings.The failure and life degradation of traction bearings should be tested on the traction motor test bench.The purpose of this research is to propose a health management plan based on traction motor bearing vibration data,including fault identification and remaining life prediction of traction motor bearings,so as to provide a feasible plan for the improvement project of the traction motor test bed.It includes four parts: bearing fault and life related theory,long and short-term memory network and particle filter algorithm,fault identification and remaining life prediction.The identification of bearing fault types and the prediction of remaining bearing life belong to the research hotspot of PHM(Prognostics and Health Management)equipment health management technology.The research focuses on fault identification and remaining life prediction:(1)Bearing fault identification.For bearing fault signals sampled under variable speed conditions,it is impossible to use Long Short-Term Memory(LSTM)to directly identify faults.Therefore,this paper adopts the idea of re-sampling the sampled signal in the re-sampling of the angle domain after shaping under variable speed conditions,and then recognizing the LSTM network,which has achieved certain results.In order to further improve the accuracy of fault recognition,this paper proposes the CEEMD-LSTM method,which extracts the intrinsic mode(IMF)of the angular domain resampling signal through CEEMD(Complementary Set Empirical Mode),and filters out IMFs with low correlation coefficients.Compared with EEMD-LSTM,EMD-LSTM and LSTM,CEEMD-LSTM has a significant improvement in recognition accuracy.(2)Prediction of remaining bearing life.By comparing the multiple characteristic indexes of vibration signals of the bearing degradation data set,the effective value index can describe the degradation characteristics of the bearing more ideally,and then use LSTM to predict the remaining life of the bearing,which can roughly predict the degradation trend of the effective value.In order to further improve the accuracy of life prediction,the proposed PF-LSTM algorithm essentially uses the predicted result of the LSTM algorithm as the "observed value" of the particle filter algorithm(Particle Filtering,PF),and then uses the state transition equation to correct the prediction result.In fact,It is based on the integration of data-driven and empirical models.Through experimental comparison,the prediction result has a higher correlation coefficient than the LSTM algorithm and a smaller RSME,which can achieve a more accurate prediction of the remaining life of the bearing.In summary,the theory of bearing fault identification based on CEEMDLSTM-based variable-speed operating conditions and LSTM-PF-based bearing remaining life prediction theory proposed in this paper is feasible,which can provide theoretical guidance for the transformation of bearing traction motor test benches.