2025 International Conference on Sensing, Measurement & Data Analytics in the era of Artificial Intelligence (ICSMD 2025)

Reliable equipment health monitoring and fault diagnosis technologies are crucial to ensuring the safe and efficient operation of high-end equipment. Cross-domain intelligent diagnosis technologies based on unsupervised domain adaptation have shown broad application prospects in scenarios such as cross-equipment and varying working conditions. However, such methods rely on specific prior assumptions regarding inter-domain label relationships and domain configurations, which limits the generalization and practicality of unsupervised domain adaptation technologies in actual industrial fault diagnosis scenarios. To address the above issues, this paper proposes a universal cross-domain fault diagnosis method applicable to diverse label and domain configurations. This method constructs a multi-scenario shared predictive class confusion (PCC) bias to guide cross-domain knowledge transfer, thereby adapting to various cross-domain fault diagnosis (CFD) scenarios. To measure the predictive class confusion bias more accurately, a prototype similarity-based fault discrimination method is proposed to enhance classification robustness, thus providing a reliable prediction distribution for estimating the PCC bias. In addition, a label smoothing-based probability calibration mechanism is designed for probability regularization to alleviate the underestimation of PCC bias caused by overconfident predictions. Comprehensive experiments are conducted on a planetary gearbox transmission system dataset, and the results show that the proposed method has universality in cross-domain diagnosis scenarios under four different label and domain configurations, and its performance is competitive scenario-specific comparison methods.