Weak Fault Mechanism Of Bearings For Rail Vehicles And Study On Fault Diagnosis Method Of KPCA-DBN

The progress of urbanization has accelerated the pace of urban construction in China.Rail transit has gradually become an artery running through a city and a bond between regions.With shortcut,safety,reliability,low environmental impact,etc,rail transit has made a major dent in the problems of traffic congestion,single means of transport and severe air pollution becoming the preferred way people travel.Therefore,ensuring the safe operation of rail vehicle,which is closely related to the safety of life and property of passengers,is not only a responsibility but also a mission.In this regard,higher requirements have been put forward for both the effectiveness of fault diagnosis methods and the timeliness of maintenance work even for the reliability of equipment manufacturing.Improve quality in production and construction to reduce the frequency of faults and timely detect faults in process of operation and maintenance,only then can the safety and reliability of rail vehicle be fully ensured.This paper focuses on the analysis of mechanism of weak faults for rolling bearing and the study of fault diagnosis method.Firstly,it is shown that the weak faults of rolling bearing are mostly surface damage related to lubrication by the research actuality in fault mechanism.A non-Newtonian elastohydrodynamic simulation model about a double-row tapered roller bearing is established for the axle box bearing of CRH3.The parameters of the simulation model are changed according to the actual working conditions of CRH3.Based on the results,the phase of starting or braking at the maximum acceleration running at a low speed tends to cause mixed lubrication.Under the working condition,the influence of surface morphology on the film-forming properties of grease is analyzed.As a result,recommendations of surface optimization for bearing processing are provided to improve the mixed lubrication and reduce the occurrence of weak faults.In detail,surface roughness S_q is from 0.03μm to 0.1μm.Texture direction should be parallel to the rotation direction of rolling element.Kurtosis is 3 and skewness is a negative value from-1 to-0.5.Secondly,the simulated vibration signals of rolling bearing for rail vehicle are denoised by the method of VMD—Sample entropy—Wavelet threshold denoising.The intelligent fault diagnosis model based on KPCA-DBN is established for identification of 12 faults including weak faults and the average test accuracy reaches 99.1%and the training time is 28.3s which is half(28.3/58.7)as long as directly processing by DBN.The combination of KPCA and DBN simplifies the DBN network structure and the weight of the visual layer is initialized by the contribution rate of principal components to improve the convergence speed.Accurate identification of weak faults is achieved without manual feature extraction.Finally,the testbed collecting and transmitting vibration signals is built.Based on the block diagram program in LabVIEW,the user can control the motor speed and realize acquisition and transmission of vibration signals by visualized operations.The validity of KPCA-DBN in practical application is verified by processing the real vibration signals of rolling bearing collected from traction motor in subway.It takes 31s to train the network for classifying four categories,which includes the identification of fault locations,and the accuracy rate reaches 96.1%.The fault diagnosis results are guidance in maintenance and repair work.The intelligent approach for fault diagnosis based on vibration signals of rolling bearing is also established.