A multi-representation transfer adversarial network for intelligent fault diagnosis of rotating machinery

Abstract

Fault diagnosis of rolling bearings is among the most crucial links in the prognostic and health management of bearings. To solve the problem that cross-domain fault diagnosis cannot be performed due to the distribution differences between different working conditions, a transfer diagnosis method based on multi-representation adversarial neural network is proposed. First, the multi-representation neural network is applied to extract multiscale features. Second, the domain adversarial network is utilized to set the gradient inversion layer and extract the domain invariant features in the multiscale features. In terms of the loss function, the Wasserstein function and cross-entropy loss function are utilized to measure the distance between the source domain and the target domain. The experimental case of rolling bearing supports the effectiveness and superiority of the proposed method.