A Lightweight and Robust Model for Engineering Cross-Domain Fault Diagnosis Via Feature Fusion-Based Unsupervised Adversarial Learning

Abstract

​The cross-domain fault diagnosis models of rolling bearings generally have weaknesses such as large size, complex calculation and weak anti-noise ability. Hence, a lightweight and robust model via feature fusion-based unsupervised adversarial learning (LRFFUAL) is proposed, which could be special benefit for practical engineering applications. Firstly, a feature fusion block consisted of depthwise separable convolution and average pooling layers is designed to reduce the volume and computation of the model. Secondly, a channel residual strategy is proposed to enhance the robustness of the model, which applies residual techniques between the original features and some channels with weak feature information to achieve data augmentation. Thirdly, a new adversarial learning strategy is proposed to improve the convergence speed and generalization performance of the model by inputting marginal features into the discriminator. The experimental results show that LRFFUAL has the advantages of smaller size, less computation, and stronger robustness compared with SOTA methods.

Keywords: lightweight and robust, feature fusion, adversarial learning, channel residual