Research Of Train Bearing Wayside Acoustic Fault Diagnosis Based On Two-stage Learning Model

As one of the important transportation modes,railways play an important role in freight transportation and passenger transportation because of their low operating cost and large carrying capacity.As a key component of trains,train bearings have a significant impact on the normal operation of trains and the safety of passengers.The wayside acoustic monitoring system has the advantages of low cost,non-contact measurement and the ability to detect early failures,and has good application prospects.In recent years,artificial intelligence and machine learning have developed rapidly and have been successfully applied in the field of fault diagnosis.In this paper,considering the unique Doppler distortion characteristics of the wayside acoustic signal of the train bearing,based on the Safety Region(SR),Back Propagation Neural Network(BPNN)and Zero-shot Learning(ZSL)in machine learning,we studied a two-stage learning model,specific research content is as follows:(1)A two-stage learning model based on SR and BPNN is proposed.This method does not require Doppler correction and can solve the problem of insufficient fault samples.In the absence of a failure sample set in the early stage of the application,only healthy samples are used to establish the SR model.When the sample sets of different fault types are fully accumulated,BPNN is used to build a more accurate diagnosis model.When the learning model is established,the functional relationship between signal statistical characteristics,vehicle speed and diagnosis results is directly constructed.First,the basic theories of SR and BPNN are introduced;second,the Doppler effect and its impact on the establishment of the learning model are analyzed;then,a two-stage learning model based on SR and BPNN is proposed;finally,a comparative analysis through simulation and experiment The effectiveness of the proposed method is verified.(2)Aiming at the problem of fewer composite fault samples in the initial application stage,a train bearing composite fault diagnosis method based on ZSL is studied.Under the premise of lack of composite fault samples,this method can build an information bridge from single fault to composite fault through binary attributes,and use the existing single fault sample attribute classifier to realize the identification of composite faults.First,the basictheories of attribute learning and binary attribute learning are introduced.Second,the attribute classifier based on binary attributes is elaborated in detail.Then,a ZSL-based composite fault diagnosis method for train bearings is proposed.Finally,the effectiveness of the proposed method is verified by experimental signal analysis.