A dynamic weighted joint distribution domain adaptation network for cross-machine fault diagnosis of rolling bearings

Rolling bearings are fundamental components in machinery, and precise fault diagnosis is critical for assessing the overall reliability of mechanical systems. In cross-domain fault diagnosis, domain drift presents a primary challenge, often leading to performance degradation or model failure, particularly under cross-machine conditions characterized by pronounced distributional discrepancies. This study introduces an innovative framework for cross-machine fault diagnosis, termed the dynamical joint distribution domain adaptation (DJDA) network. In the DJDA framework, a dynamic weighting mechanism is employed to adaptively and quantitatively adjust both overall alignment and individual class discrimination between domains, while simultaneously learning marginal and conditional feature distributions. The multilayer-multiple universal-order moment matching technique is utilized to enhance global distribution alignment between domains, while multiclassifier-assisted adversarial domain adaptation (multiclassifier-GADA) automatically mitigates class discrepancies. Extensive experiments reveal that the proposed framework significantly outperforms state-of-the-art methods and demonstrates promising potential for application in industrial contexts.