Research On Fault Diagnosis Method For Rolling Bearing Of EMU Based On Improved Convolutional Neural Network

As one of the key rotating parts of the high-speed EMU running part,the rolling bearing directly affects the running quality and driving safety of the EMU.Therefore,it is of great significance to carry out research on the fault diagnosis technology of the EMU rolling bearing.At present,the signal processing-based vibration analysis and diagnosis method and the intelligent diagnosis method based on machine learning shallow model widely used in the rolling bearing fault field of rail vehicles still have certain deficiencies in the versatility and intelligence,and research is more general and intelligent diagnosis.The method has a positive role in promoting the intelligent operation and maintenance of key components of the EMU.In view of the traditional intelligent diagnosis method relying on signal processing technology and expert diagnosis experience to extract fault features,the deep learning model is only used as a classifier.The diagnostic model identification accuracy is not high,the generalization is not strong,and the modeling analysis of the diagnostic model is not Considering the data imbalance between the "normal state" sample and the "fault state" sample in the actual fault diagnosis,this paper mainly carries out the following four aspects of research work:(1)Firstly,using CWRU bearing standard data set for modeling analysis,verifying that the convolutional neural network extracts fault features from the original one-dimensional vibration signal layer-by-layer convolution pooling,and realizes the algorithm of accurate classification of different types and degrees of faults.Then,by introducing algorithms such as Dropout,BN,SVM,etc.,the convergence speed,recognition accuracy and generalization ability of the diagnostic model are improved.Finally,the recognition accuracy under noise interference and the statistical indicators P,R,FI are objectively evaluated.Classification performance of CNN-BN-SVM model.(2)The bearing fault detection system of the high-speed EMU running part was developed.According to the common fault type of the EMU,the EDM machining fault bearing was used.The experimental data of the EMU axlebox bearing fault data acquisition based on the high-speed rolling vibration test bench was designed and utilized.The detection system carries out bearing vibration data acquisition,and provides experimental data support for the subsequent establishment of the fault diagnosis model for the EMU rolling bearing.(3)Firstly,based on the EMU rolling bearing fault data set,the CNN-BN-SVM diagnostic model is established,and the influence of the data imbalance between the normal state sample and the fault state sample on the performance of the classification model is analyzed.Then,by introducing data resampling and cost sensitivity The learning strategy enhances the ability of the CNN model to classify unbalanced samples.Finally,the superior performance of the CNN-CSSVM model in terms of noise immunity and unbalanced data is illustrated by comparison with several classic traditional intelligent diagnostic methods.(4)The CNN-CSSVM diagnostic model studied in this paper is deployed to the high-speed EMU travel bearing fault detection system as the support algorithm of its intelligent identification module.The deployment plan of the diagnostic model is designed based on C#and open source frameworks such as TensorflowSharp and Libsvm.