Mutual Information-Guided Domain-Shared Feature Learning for Bearing Fault Diagnosis Under Unknown Conditions

With the support of unsupervised domain adaptation (UDA) techniques, variable condition-bearing fault diagnosis has achieved considerable progress. Nevertheless, the prerequisite of obtaining target data in advance limits the practical application of these diagnostic models in real-world scenarios. For bearing fault diagnosis under unknown conditions, domain generalization-based methods show great promise, and acquiring domain-invariant knowledge is crucial to enhance generalization capability. However, working condition (WC)-related information is often coupled with health state (HS)-related information, which makes it challenging to obtain purely domain-invariant and discriminative HS-related features. To address this issue, this article proposes a mutual information-guided domain-shared feature learning (MI-DSFL) algorithm. MI-DSFL designs an HS diagnosis branch and a WC identification branch to directly extract HS-related features and WC-related features, respectively. Through the interaction between the two branches and the mutual influence of the corresponding classifiers, domain-invariant HS-related features and WC-related features are ultimately decoupled. In addition, by minimizing the mutual information between HS-related features and WC-related features, the HS diagnosis branch further enhances its ability to capture purer HS-related features. Finally, a cross-domain soft triplet loss (CS_tri) is designed to guide the final embedding space, further improving the generalization ability. Extensive experiments on cross-domain fault diagnosis tasks demonstrate the effectiveness of the proposed method.