Transfer Residual Convolutional neural Network for Rotating Machine Fault Diagnosis under Different Working Conditions

Abstract

In recent years, due to the rise of deep learning, fault diagnosis theory has made great progress. However, it should be noted that the current fault diagnosis methods mainly concentrate on the fault identification of the machine under the same working condition, especially for rotating machinery. This means that the success of these fault diagnosis methods has an important premise, that is, the training samples and the test samples share the same data distribution. In order to solve the shortcomings of traditional fault diagnosis methods and the challenges of practical engineering issues, a transfer residual convolutional neural network is proposed in this paper. Compared with other traditional fault diagnosis methods, the proposed method can achieve accurate diagnosis of rotating machinery under different working conditions. Specially, multi-kernel maximum mean discrepancy (MK-MMD) is designed to the residual convolutional neural network (CNN) to extract the similar and common features of source domain and target domain. Then, the labeled source features and the unlabeled target features are input into the classifier to obtain the final diagnosis results. The comparison results demonstrate the effectiveness of the proposed method.