Anomaly detection for wind turbine gearbox oil pressure difference based on SCADA data

Abstract

In this paper, a SCADA data-based fault detection method for gearbox oil pressure difference is proposed, and the core of the method is Spare Bayesian Learning (SBL) algorithm. The gearbox oil pressure difference probability estimation model can be constructed by training the historical normal operating based on SBL. Then, the probability distribution interval of the gearbox oil pressure difference can be estimated. The abnormal state of gearbox oil pressure difference can be judged by observing whether the actual value within the probability distribution interval. In addition, statistical hypothesis testing method is used to verify the reliability of the anomaly detection results. Through the method, the gearbox oil pressure difference anomaly detection problem can be transformed into a parameter estimation problem with low computational complexity. Case studies are conducted on two actual WTs with known gearbox oil pressure difference faults, and the results demonstrate the feasibility and effectiveness of the method.