Research On CNN Fault Diagnosis Method Based On Multi-Scale And Two-Stream Feature Fusion

Rotating machinery plays a unique role in modern industrial production.If it fails at work,it will bring a certain degree of economic loss and may cause casualties in severe cases.Therefore,it is very important to accurately identify the faults of different rotating machinery.In recent years,with the rapid development of deep learning,many scholars have produced new ideas for the research of rotating machinery fault diagnosis methods.Based on the above reasons,this paper takes rotating machinery as the research object,and proposes three fault diagnosis methods based on convolutional neural network.(1)Aiming at the problems that traditional fault diagnosis methods require manual feature extraction,deep structure of convolutional neural network and insufficient classification ability of classifier,a fault diagnosis method based on one-dimensional convolutional neural network and support vector machine is proposed.This method not only makes full use of the powerful feature extraction ability of convolutional neural network,but also has the excellent classification ability of support vector machine.The experimental results of the axial piston pump fault data set show that the fault recognition accuracy of this method is high,and it can better complete the required fault diagnosis classification task.(2)Aiming at the problem of insufficient feature extraction of convolutional neural network fault information in single channel,a one-dimensional convolutional neural network fault diagnosis method based on multi-scale is proposed.By using multi-scale convolution structure for feature fusion,the integrity of fault features is ensured.In order to reduce the training parameters of the network model,the global average pooling layer is introduced into the model structure.In the experimental stage,this method is applied to the fault data sets of motor and axial piston pump respectively,and the optimization algorithm and learning rate that affect the performance of the network model are analyzed.The method is compared with other fault diagnosis methods,which further proves that the network model has higher fault recognition accuracy.(3)Aiming at the problem that one-dimensional data structure will lose some receptive fields when analyzing data,which will affect the performance of fault diagnosis,a fault diagnosis method based on two-stream feature fusion convolutional neural network is proposed.The network model inputs two different types of data,one of which is onedimensional time series data,and the other is a two-dimensional wavelet time-frequency diagram corresponding to one-dimensional data.Two fully connected layers are added after the convergence layer of the network model,which increases the representation ability of the model and makes sufficient preparations for the dimension transformation of the final output results.In the process of method verification,the effectiveness of the proposed method is verified by using the open data set of the drive end bearing of Case Western Reserve University and the fault data set of the axial piston pump under different working conditions,and the method is compared with other fault diagnosis methods and the methods mentioned above.The experimental results show that the fault recognition accuracy of this method is higher.