Research On Fault Diagnosis Method Of Railway Freight Train Wheelset Bearing Based On Cyclic-Correntropy And Transfer Learning

In recent years,with the continuous expansion of the scale of China's railway freight transport,the operational safety issues have gradually attracted widespread attention.As a carrier and core of railway freight transportation,security risks of railway freight train are likely to further expand into serious safety accidents.However,as one of the key components of railway freight train,the wheelset bearings of freight train affect the operation safety of railway freight train directly.Therefore,an in-depth study of the fault diagnosis technology of railway freight train wheelset bearings is of great significance for improving the safety guarantee level and scientific operation and maintenance of railway freight train.In this thesis,taking railway freight wheel bearing as the research object,has researched on impulsive noise and small sample issues in fault diagnosis of railway freight train wheelset bearings in different application scenarios.Correspondingly,two bearing fault diagnosis methods have been proposed.This thesis mainly completed the following works:(1)Aiming at the issue of impulsive noise in online monitoring application scenarios,this thesis proposes a bearing fault diagnosis method based on empirical mode decomposition and cyclic correntropy.Based on the analysis of the suppression effect of correntropy on impulsive noise,this thesis using the cyclic correntropy spectrum of the signal which is reconstructed by empirical mode decomposition and the original signal to identify the type of bearing fault in the impulsive noise environment.Then using a simulation signal and two experimental signals to verify the effectiveness of the method.At last,a fault identification effect evaluation indexKR_s is proposed to quantify the fault identification effects of different methods.The comparison with common methods such as spectral kurtosis and spectral correlation proves that the method proposed in this thesis has better identification effect.(2)Aiming at the issue of small samples with few fault samples in the offline analysis application scenario,this thesis proposes a small sample bearing fault diagnosis method based on improved joint distribution adaptation.Based on the study of the transfer learning method based on joint distribution adaptation,the basic classifier KNN of the joint distribution adaptation is improved by the correntropy induced metric.Combining the 23 fault feature indexes extracted from the three aspects of time domain,frequency domain and entropy domain,a rolling bearing transfer diagnosis model under a small sample is constructed.Using the rich bearing data of Case Western Reserve University laboratory realize the transfer fault diagnosis task of the actual railway freight wheelset bearing data.To some extent,it solves the issue of small samples of railway freight wheelset bearing.(3)In order to convert the two proposed fault diagnosis method theories into practical applications,this thesis develop a prototype system for fault diagnosis of freight train wheelset bearing based on the above research results.Using the App Designer tool in the MATLAB platform,the signal processing fault diagnosis module and the intelligent classification fault diagnosis module were designed,which realized the functions of fault diagnosis,data management and diagnosis report in the prototype system.The functional test was completed using actual data signals,and the feasibility of the prototype system was verified,which laid the foundation for the further development of the on-site freight train wheelset bearing fault diagnosis system.