Wind turbine failure detection based on SCADA data and data mining method

WTFD (Wind Turbine Failure Detection) system is important for improving wind turbine reliability and reducing operation & maintenance cost. A WTFD method using only SCADA data and data mining algorithm is proposed. Firstly, select representative variables from a wide variety of SCADA variables using ARD (Automatic Relevance Determination) method. Then, the denoising autoencoders (DAE) model using the selected variables with sliding window is developed to capture nonlinear correlations among multiple variables and autocorrelation of each variable. Finally, using the Mahalanobs distance (MD) of reconstruction error to make the WTFD. Real world dataset is used to validate the efficiency of the proposed method. The results show that the proposed method can provide advanced failure alarm for wind turbines many hours before failure happens. The results of our proposed WTFD method can reduce operation & maintenance cost and improve wind turbine reliability.