Research On Rolling Bearing Fault Diagnosis Method Based On Deep Learning

The research of rolling bearing condition monitoring and fault diagnosis can effectively guarantee the safe and stable operation of the equipment,avoid the occurrence of major accidents and reduce the loss of human and financial resources.Traditional fault diagnosis methods rely too much on signal processing technology and experts’ fault feature extraction experience,resulting in poor generalization ability of the model.Data driven deep learning technology relies on big data self-learning,self-adaptive and other characteristics,which can effectively solve the problem of rolling bearing fault diagnosis under complex and variable conditions.In this paper,the rolling bearing vibration signal is taken as the research object,combined with the deep learning theory,around the rolling bearing fault prediction and fault type identification.The research of the thesis is as follows:Traditional rolling bearing fault diagnosis methods are difficult to effectively extract fault feature information in complex and changeable multi working conditions.STFT can significantly enhance the time-frequency feature information of vibration signal.CNN is good at extracting nonlinear implicit correlation feature information of pictures.Therefore,a rolling bearing fault diagnosis method based on STFT-CNN is designed and implemented.After STFT transform of the original signal,the time-frequency spectrum of time and frequency is obtained,which is input into the improved lenet-5 model for fault type identification.The data selflearning of convolution layer and pooling layer is used to output the predicted fault type of the original signal.The effectiveness of the method is verified by the original vibration signals of rolling bearings with five different fault types under multiple working conditions.Rolling bearing defects will produce periodic impact or high-order harmonic vibration signal.Besides normal rotation and revolution,rolling element will also have sway and lateral vibration.So the time domain fault signal also has certain randomness.Therefore,a rolling bearing fault diagnosis method based on LSTM-SVM is proposed.In this method,continuous data segments in time domain are input into double-layer LSTM structure,and temporal and spatial correlation features of data are extracted adaptively.SVM multi classifier is used to obtain data prediction results.The effectiveness and generalization ability of this method are verified by processing field experimental data sets and CWRU open source data sets.The industrial working environment of rolling bearing is complex and harsh,and the collected vibration signal is annihilated by strong noise,which brings great challenges to fault feature extraction.Aiming at this problem,a rolling bearing fault diagnosis method based on EEMD-CNN is proposed.Firstly,the improved EEMD is used to decompose and denoise the original signal,and the CNN model is used to extract the high-dimensional features of different fault types of rolling bearing.The experimental results show that the model can effectively extract the data features under high noise,and has the characteristics of fast convergence,high accuracy of fault classification and strong universality.