Research On Bearing Fault Diagnosis Based On Deep Learning

The rolling bearing is one of the key parts of rotating machinery.This paper takes the rolling bearing as the research object,uses the spectrum analysis traditional method,and focuses on the problem of extracting the frequency characteristic of the fault bearing.At the same time,deep learning intelligent diagnosis method is used to carry out research on pattern recognition,and an effective rolling bearing fault diagnosis method is realized.The new method is used to explore the life prediction of turbofan engine,which is a preparation for the application of deep learning in bearing life prediction.The main work of this paper is as follows:(1)Research on bearing fault diagnosis based on MCKD-EWT methodIn this paper,the MCKD(Maximum Correlated Kurtosis Deconvolution)method and EWT(Empirical Wavelet Transform)method to fault diagnosis of rolling bearing.In this paper,the method is verified by the bearing vibration signals collected from the laboratory of Mechanical Engineering College of Inner Mongolia University of Science and Technology,and the generalization of the method is verified by the bearing vibration signal set of Luoyang Bearing Research Institute.(2)Research on bearing fault diagnosis based on improved CNN-LSTM networkIn this paper,the original bearing vibration signal as the input to the bearing fault classification,feature extraction from the Network structure and optimize the Network types,this paper improves the traditional CNN(Convolutional Neural Network)and add new feature extraction module,to improve the bearing fault classification accuracy and improve the Network generalization.From the aspect of network structure optimization,an improved CNN network is proposed in this paper.Experiments show that,compared with traditional CNN,the accuracy of bearing fault classification is greatly improved.From the aspect of optimization of network feature extraction types,LSTM(Long Short Term Memory)was added on the basis of the improved CNN network.LSTM was used to extract the global features of vibration signals to make up for the defect that the improved CNN could only extract the local features of vibration signals,so as to increase the diagnostic accuracy and enhance the network generalization.The improved CNN and the improved CNN-LSTM network were applied to the vibration signals of the bearing fault collected by the Bearing Fault laboratory of Inner Mongolia University of Science and Technology to classify the bearing faults,and the data set of the bearing vibration signals of Western Reserve University was used for further verification.The results show that the improved CNN and the improved CNN-LSTM network have better effects than the traditional CNN,and the Improved CNN-LSTM network has stronger generalization.Compared with the traditional signal processing method,it has the advantages of simple input,few parameters and accurate results.(3)Application of improved CNN-LSTM network in the prediction of remaining service life of turbofan engineIn order to further explore the suitability of the proposed improved CNN-LSTM network,the remaining life prediction of turbofan engine was studied.Experiments show that the network is still applicable to the remaining service life of turbofan engines.It is also ready for the application of the Improved CNN-LSTM network in bearing life prediction.