Rolling Bearing Fault Diagnosis Based on GWVD and Convolutional Neural Network

Abstract

In order to solve the problems that the traditional signal processing method cannot deal with non-Euclidian data, a graph Wigner-Ville distribution (GWVD) and convolutional neural network (CNN) based rolling bearing fault diagnosis method is proposed. GWVD combines graph signal processing with Wigner-Ville distribution and can effectively suppress the cross interference terms. GWVD based CNN has the advantages of extremely high energy aggregation and resolution, which improves the pattern recognition accuracy of convolutional neural network in noisy environment. The principle of GWVD and CNN based rolling bearing fault diagnosis are introduced. Firstly, one-dimensional bearing fault vibration signals are converted into graph signals according to the similarity between sampling points and then the completed data are divided proportionally. Secondly, the GWVD based pre-processing is used as the input layer of the convolutional neural network, and the bearing fault feature extraction and fault classification are realized by two-dimensional convolutional neural network. Finally, the Case Western Reserve University standard bearing data set is used for experimental verification and compared with the traditional Wigner-Ville distribution signal preprocessing method. The experimental results show that GWVD based CNN has higher energy aggregation and resolution, and is an effective intelligent bearing fault diagnosis method with better performance than that of traditional WVD based CNN.