Fault Diagnosis Based on C-DCGAN for Rolling Bearing

Abstract

In actual engineering, the rolling bearing fault samples are small and non-balanced, when the bearing data is unbalanced, the classification of the trained diagnostic model is often inclined to the majority class, which greatly affects the diagnostic accuracy of the minority class. Aiming at the above problems, this paper proposes a fault diagnosis method for generating adversarial network based on conditional deep convolution. Firstly, the bearing vibration signal is converted into a two-dimensional image by using the gram angular field, and then the distribution of the fault data is learned by combining the characteristics of the deep convolutional neural generation adversarial network and the conditional generation adversarial network, and more labeled fault data is generated for the expansion of the fault datasets, and finally the expanded datasets are input into the CNN-SVM diagnostic model. Experimental results show that compared with CGAN, CNN-SVM and other fault diagnosis algorithms, the proposed algorithm can classify bearing faults more accurately.