Cross-conditions fault diagnosis of rolling bearing based on reliable domain adversarial network

To address inter-class confusion and ambiguous decision boundaries in traditional rolling-bearing fault diagnosis under varying operating conditions, we propose a reliable domain adversarial network (RDAN) that enables more discriminative and robust cross-condition fault identification. Specifically, the RDAN is designed with a multi-modal and multi-channel feature extractor, which sufficiently captures domain-invariant features from both the frequency and time–frequency domains. In addition, a novel Dirichlet-evidence-based classification loss is introduced to jointly model classification accuracy and predictive uncertainty, by integrating evidential learning with a Kullback–Leibler divergence term. Furthermore, a new distribution alignment regularization method is proposed to minimize inter-class correlations and reinforce prediction consistency under various input perturbations, thereby effectively mitigating class confusion and enhancing the model’s robustness to signal variations. Finally, comparative experiments conducted on the Case Western Reserve University and drivetrain dynamics simulator (DDS) bearing fault datasets demonstrate the effectiveness and superiority of RDAN over other unsupervised domain adaptation methods in cross-domain bearing fault diagnosis.