Research On Incipient Fault Diagnosis And Remaining Useful Life Prediction For High Speed Railway Traction System

With the scale expansion of high-speed railway in China,the operation safety has become more and more important.Traction system,as the core unit in the information control system of high-speed railway,provides reliable traction.However,slowly changing incipient faults such as the wear of motor bearings and the aging of various electronic components are common phenomenon in traction system.Over time,the faults will severe and then risk the system safety.Therefore,it is of great significance to realize the diagnosis of the slowly changing incipient fault and the remaining useful life prediction for the high-speed railway traction system.In this paper,the CRH2 high-speed train traction system is taken as the research object,to carry out the research on the diagnosis of slowly changing incipient fault and the remaining useful life prediction of the system.The main work are as follows.Firstly,the structure and the working mechanism of the traction system are presented.Based on this,the semi-physical simulation platform of the traction system is built to simulate the slowly changing incipient faults.After collecting data from the platform,the time domain and frequency domain features of each variable of the original data are extracted.Secondly,by analyzing the incipient fault data,the principal component analysis method is used to detect the incipient fault of the traction system.The verification results show that the method has better detection performance.Then,the neighborhood component analysis method is used to select effective features and the support vector machine(SVM)is used to diagnose the incipient fault.The verification results show that the method has a high accuracy rate.Then,a remaining useful life prediction method based on the similarity of degraded trajectories is proposed.This method uses the sequence-to-sequence structure based long short-term memory(LSTM-Seq2seq)model to extract the system’s performance degradation trajectory.After that,the remaining useful life of the current sample is obtained by matching the degradation trajectory of the current sample with that of the historical sample.The verification results show that this method is feasible for the remaining useful life prediction of the system.Nevertheless,the prediction is less robust considering the strong dependence on the quality of the sample library.Finally,aiming at the problems that the life prediction method above cannot achieve the degradation trend prediction of the system and the failure threshold is difficult to determine,a real-time health state assessment and remaining useful life prediction method based on SVM-LSTM is proposed.This method is first to detect and diagnose the incipient faults of the system.With the judged types of fault,one can employ the LSTM network to predict the trend of each variable and then assess the health state of the variables using the SVM model.Through the judgement of failure threshold associated with health state,the remaining useful life of the system can be accurately forecasted.Experimental results show the proposed method is effective and allows improving the prediction accuracy.