Multisource–Multitarget Partial Domain Adaptation for Bearing Fault Diagnosis

Most existing fault diagnosis methods focus on single-domain adaptation (DA), where fault labels in the training and test datasets are identical. However, fault diagnosis for multisource-multitarget domains, which is a critical yet underexplored area, presents unique challenges. Additionally, when the fault labels in the target domains form only a subset of those in the source domains, the task becomes more difficult due to interference from redundant fault classes. To address these challenges in partial DA for multisource-multitarget domains, this study proposes a novel bearing fault diagnosis method called source domain confusion-subdomain adaptive weighted alignment (SDC-SAWA). In this approach, a shared feature extractor is constructed during the SDC stage, and weighted similarity is incorporated in the SAWA stage. The similarity measure accounts for the impact of partial domains. The proposed method is validated using three bearing datasets and compared with several other methods.