Research On The Application Of Deep Learning Technology In The Bearing Fault Diagnosis Of Rail Transit

As an important mode of transportation,rail transit is in the peak period of large-scale line construction and large-scale opening,and its mileage of construction and operation is increasing greatly.As the carrier of rail transit,how to ensure the safe operation of rail transit vehicles,to avoid endangering the life safety of passengers,to avoid economic losses and adverse social impact,has become an indispensable part of rail transit development.Rolling bearing of rolling stock has been working under the working conditions of high speed and heavy load,which is one of the worst parts in the working environment of all equipment on the train,and it plays a key role in the safe operation of the train.Therefore,the diagnosis of bearing fault of rail transit vehicle is the most important part in the safe operation of rail transit vehicle.With the rapid development of the number and technology of rail transit vehicles,the traditional fault diagnosis technology of rail transit vehicles has not kept up with the requirements of the times.The advantages of deep learning in feature extraction and pattern recognition make it widely used in many fields,but it is still a new industry in complex industrial fault diagnosis;deep learning technology is still a new stage in rail transit fault diagnosis,and the in-depth study of deep learning technology will be of great significance for the research of rail transit bearing fault diagnosis.Based on this,this paper will study the bearing fault diagnosis method of rail transit vehicles.In this paper,the essence of the research on the rolling bearing of rail transit is firstly determined,that is,the research on the rolling bearing.The structure,vibration mechanism,failure form and the causes of the failure of the rolling bearing are analyzed,which provides the theoretical basis for the later research.Secondly,this paper analyzes several deep learning models applied in the field of bearing fault diagnosis,and through the analysis of the advantages and disadvantages of these models,it is concluded that the cyclic neural network(RNN)model based on time series is more suitable for bearing fault diagnosis.Because the traditional RNN model has defects,this paper aims to design a bearing fault diagnosis model based on time series.Next,the paper introduces a mature RNN optimization model based on the adjustment of RNN’s own structure,that is,the long-term and short-term memory(LSTM)neural network model.Then,on the basis of particle swarm optimization algorithm,a new optimization algorithm,dynamic particle swarm optimization(ad PSO),is proposed.Finally,ad PSO optimization algorithm is used to replace the gradient descent method used by traditional LSTM for LSTM model A new RNN optimization model,adpso-lstm,is obtained.The experiment is divided into three groups: traditional RNN model,LSTM neural network model based on gradient descent method and adpso-lstm model.From the test results,we can see that adpso-lstm has advantages over the other two models,but theaccuracy of the three models is not very high,and the robustness is not very good.Based on this,this paper uses the method of wavelet packet processing and model test to further improve the accuracy of fault diagnosis.The specific data processing method is to first denoise the data with wavelet packet,and then extract the energy characteristics of the denoised data.Through the analysis of the experimental results of several groups of comparative experiments,it can be seen that adpso-lstm model with wavelet packet processing has high efficiency,recognition accuracy and robustness in dealing with the problem of bearing fault diagnosis of rail transit.