Research On Fault Diagnosis Of High-Speed Train Bearings Under Variable Working Conditions Based On CNN And Transfer Learning

Diagnosis and health management is an important step in the operation and maintenance system of rail transit.Rolling bearing is one of the core components of the high-speed train bogie,its reliability and safety directly affect the safe operation of trains.Therefore,research on bearing fault diagnosis is of great significance to guarantee the train safety.This paper takes the train bearing as the research object,focuses on the problems of variable working,strong noise and scarce label fault data in bearing fault diagnosis,proposes new fault diagnosis methods based on convolutional neural network and transfer learning.The main research contents are as follows:(1)A fault diagnosis method based on wavelet transform and convolution neural network is proposed to solve the problem of insufficient adaptability caused by the bearing fault diagnosis based on signal processing relying on expert knowledge.The method converts vibration time series information into time-frequency domain information by wavelet transform,and then uses convolution layer to construct fault feature extractor to automatically extract fault feature and classify them.The experimental results show that this method can effectively improve the accuracy of bearing diagnosis.(2)A fault diagnosis model based on multi-scale deep residual shrinkage network is proposed,which is used to solve the insufficient adaptability in the fault diagnosis of bearings under variable working conditions.On the one hand,residual learning and regularization are used to alleviate the over-fitting,which can improve the feature extraction ability of the model.On the other hand,in order to solve the problem if information loss in soft-threshold denoising,an improved soft-threshold is obtained by embedding automatic learning in the residual structure to improve the signal-noise ratio of samples.Last,in order to strengthen the adaptability of the model to different working conditions,attention mechanism is introduced to strengthen the important characteristics,weaken the interference characteristics.The experimental results show that the proposed method has good working condition adaptability and robustness to noise.(3)A transfer fault diagnosis method based on singular value decomposition(SVD)denoising and adaptive batch normalization(Ada BN)is proposed,which is used to solve the difficulty of feature extraction caused by strong noise.Firstly,the method improves the signal-noise ratio of signal through SVD.And then,it uses the Ada BN algorithm to reduce the domain gap between source domain and target domain,so that the two domains align on the overall distribution,and the robustness of the fault diagnosis model to the noise is enhanced.The experimental results show that this method has advantages in bearing diagnosis accuracy under strong noise.(4)A transfer fault diagnosis method based on improved domain-adversarial neural network(DANN)is proposed.On the one hand,this method improves the output diversity of DANN by batch nuclear-norm maximization(BNM),while ensuring the classification accuracy of the model for small samples.On the other hand,this method uses the multi-scale deep residua shrinkage network(M-DRSN)as the feature extractor of DANN to strengthen the feature extraction ability.The experimental results show that the method has good transfer diagnosis ability and solves the fault diagnosis of train bearings under variable working conditions and small samples condition in some degree.There are 57 pictures,26 tables and 68 references in the body.