Abstract: [Objectives] The generator bearings of wind turbines are prone to damage during operation, and traditional spectrum analysis, which relies on human experience, often fails to accurately and promptly detect faults. To reduce operational risks and minimize economic losses caused by fault-induced downtime, a data processing and analysis method based on the BERT large model is developed for the condition monitoring system (CMS) for wind turbine generator bearings, which can be used for precise fault data identification. [Methods] Using CMS data collected from an actual wind farm, 17-dimensional features containing bearing outer race characteristic frequencies, along with time-domain and frequency-domain information, are generated. These features are structurally processed and input into the BERT large model. A fault data identification model is obtained through data training. [Results] The model’s fault data identification capability is tested using 240 sets of actual data from other wind turbines in the same wind farm, achieving an identification accuracy of 98.8%. [Conclusions] The proposed method can be effectively used for precise analysis and identification of CMS data under complex operating conditions and high noise environments, providing valuable guidance and support for enhancing the intelligence and precision of wind turbine condition monitoring.

Key words: wind turbine, generator bearing, condition monitoring system, BERT model, fault diagnosis