Meta-Feature Encoding Based Mixed Domain Generalization Method for Machinery Fault Diagnosis Under Unseen Conditions

Recently, fault diagnosis methods based on domain generalization (DG) aim to improve the diagnosis performance in unseen target domains by multi-source domain knowledge transfer. However, existing methods mainly assume that the source domains are discrete and the domain labels are known a priori, which is difficult to satisfy in complex and changeable industrial systems. In addition, the gradient update conflict caused by source domains specific information leads to the decline of DG performance. Therefore, in this study, we relax the discrete domain assumption to the mixed domain setting and propose a novel meta-feature encoding based mixed domain generalization (ME-MDG) method for machinery fault diagnosis under unseen conditions. First, a domain feature-guided adaptive normalization module (ANM) is proposed to normalize the underlying distribution of multi-source domains and the mixed source domains are divided into potential domain clusters. Secondly, a novel meta-feature encoding (ME) method is proposed to explicitly encode the overall fault feature structure, which is used to learn the generalized fault feature representation. Finally, a novel gradient consistency update strategy (GCUS) is designed to reduce the impact of domain-specific differences on model generalization. The effectiveness and superiority of the proposed ME-MDG are verified on multiple DG diagnostic tasks in two datasets.