Research On Rolling Bearing Fault Diagnosis Method Based On Multi-scale Convolution Autoencoder And Its Application

Deep learning has become a research hotspot in the field of faults in recent years.Its deep network structure has a strong ability of feature learning for complex problems,which makes it have a strong ability of fault data classification.In this paper,bearing fault diagnosis in rotating machinery is the main research content.Aiming at the problem that the fault features of rotating machinery need to be extracted manually,the fault diagnosis technology based on deep learning method is studied deeply.Based on the convolutional auto-encoder,a onedimensional multi-scale convolutional auto-encoder(1DMSCAE)is proposed,and around the network a series of performance tests are carried out with single channel and multi-channel fault data,and the application of deep learning network in practical fault diagnosis is explored.The main work of this paper is as follows:(1)verifies the feature extraction ability and data classification ability of convolutional auto-encoder with fault simulation signal.It is found that the fault diagnosis accuracy of convolutional auto-encoder for simulation signal is still low.(2)Aiming at the problem that the traditional convolutional auto-encoder can not extract the features on multi-scale,which leads to low diagnostic accuracy,a one-dimensional multiscale convolutional auto-encoder with parallel multi-scale convolutional kernel structure is proposed.This network innovatively replaces the single-scale convolution structure of the traditional convolutional auto-encoder with the convolutional kernel network structure of multiple parallel convolutions of different scales.In the performance verification of onedimensional multi-scale convolutional auto-encoder,through the analysis of Case Western Reserve University bearing fault data and the data collected by the laboratory ABLT-1A bearing strengthening test machine,the following conclusions are obtained: compared with the traditional machine learning method and the general deep learning method,the one-dimensional multi-scale convolutional auto-encoder proposed in this paper has great advantages in terms of reconstruction error,network convergence speed during training,applicability on multiple data sets,and diagnostic accuracy.(3)Aiming at the problem of insufficient credibility of the traditional method of deep learning and D-S evidence theory,a combined model of 1DMSCAE and weighted D-S evidence theory is proposed.The combination model first uses the data of different sensors to train different 1DMSCAE networks,and then the results obtained by weighting the network output according to the formula are used as basic probability assignments.The D-S evidence theory method is used for decision layer fusion,and the fusion result is used as the final diagnostic result.The combined model was verified by using the double-channel fault data of the drive and fan sides of Case Western Reserve University.The results show that the reliability of the method is generally higher than 0.99 and the diagnostic accuracy is higher than that of the ordinary D-S evidence theory method.At the same time,it also proves that the model has good noise resistance.(4)A fault diagnosis system based on 1DMSCAE model is designed and developed.The overall function of the system is designed into four functional modules: data monitoring,realtime fault diagnosis,fault information management and user information management.The fault diagnosis system adopts C/S architecture,takes Tomcat server as the server,develops the APP client which can be loaded on the Android platform intelligent device,and realizes the function of obtaining the real-time data status of the device according to the 1DMSCAE diagnosis model and visualizing it on the client.