A probabilistic anomaly detection approach for data-driven wind turbine condition monitoring

Abstract

Continuous monitoring of wind turbine (WT) operation can improve the reliability of the wind turbine and lower the operation and maintenance costs. To improve the condition monitoring (CM) and fault detection performance on WTs, this paper proposes an artificial intelligence-based probabilistic anomaly detection approach that can not only provide a deterministic estimation of the WT condition but also evaluate the uncertainties associated with the estimation. An abnormal WT condition is detected based on the evaluated uncertainties, to provide a noise-free incipient fault indication. Compared to the conventional deterministic CM approaches with a residual-based anomaly detection criterion, the proposed probabilistic approach tends to accurately detect the faults earlier, which allows more time for maintenance scheduling to prevent WT component failure. The early fault detection ability of the proposed approach was verified on an operational WT in China.