Physics guided diffusion-based generative framework for bearing fault diagnosis of offshore wind turbine motors under multi-modal coupling and missing data

Fault diagnosis of offshore wind turbine motors under missing data remains challenging. Current studies, however, exhibit two shortcomings: (1) insufficient joint modeling of multi-sensor dynamics and cross-channel correlations, causing feature extraction to rely overly on local observations. (2) weak dynamic consistency constraints, which easily induce physical distortions in reconstructed signals. To address these issues, a physics-guided diffusion-based generative framework is proposed for bearing fault diagnosis of offshore wind turbine motors under multi-modal coupling and missing data. Specifically, a shared dynamics-prior-guided multi-sensor coupling strategy is developed to enable collaborative representation of multi-sensor signals. Furthermore, a physics-consistent synergistic diffusion mechanism is designed to achieve physically plausible reconstruction of missing data and enhanced fault feature extraction. Experimental results on multi-sensor data from offshore wind turbine motor fault simulations show that the proposed framework achieves superior reconstruction and diagnostic performance under various missing-data patterns, providing a reliable diagnostic framework for the health monitoring and maintenance of key components in offshore wind turbine motors.