Domain adaptation (DA) is crucial for effective bearing fault diagnosis, as it ensures reliable performance across varying operational conditions. Current multisource DA (MSDA) methods mainly focus on overall source domain alignment, neglecting individual fault type alignment, which leads to inadequate information extraction and suboptimal fault diagnosis accuracy. To address the aforementioned problems, this article innovatively proposes a novel MSDA strategy based on attention mechanism. By utilizing a multibranch adversarial fault diagnosis network with a partially shared structure, this study enables the simultaneous alignment of multiple source domains with the target domain. Additionally, the method uses a pseudo-label training strategy and an attention mechanism to align fault types between source and target domains, rather than applying an average weight across all fault types indiscriminately. This nuanced alignment significantly facilitates efficient fusion of multisource domains and fully leverages information from the source domains. The method’s effectiveness and superiority were validated through experimental analysis involving four distinct bearing conditions. Compared to the latest research methods, this approach achieves superior diagnostic performance and provides a new perspective for MSDA in fault diagnosis.
Qi et al. (Wed,) studied this question.