A sensitive and easy-to-deploy condition monitoring method for main drive chain of large wind turbines

Effective condition monitoring of the main drive chain of wind turbines is crucial to reducing the operation and maintenance costs of wind farms. Based on the condition monitoring theory of Normal Behavior Model (NBM) of machine learning, this paper proposes a sensitive and easy-to-deploy condition monitoring method for main drive chain of large wind turbines. In order to better characterize the normal state of main drive chain, the process of selecting input and output variables for the NBM considers both the correlations among monitoring data and the working mechanism of main drive chain. The NBM, constructed based on the Informer network using the Transformer architecture, ProbSparse self-attention mechanism, and attention distillation mechanism, provides a better accuracy and requires fewer computing resources than traditional methods. In order to accurately and sensitively reflect the health conditions of main drive chain, the designed condition assessment index adopts a double-variable residual fusion mechanism and a historical memory elimination mechanism. The case studies show that the method is effective for condition monitoring and early warning for fault of main drive chain in an on-site environment. Further studies have found that the proposed method has strong transferability and is expected to be easily deployed at scale.