A fault detection method for wind turbines based on deep belief network

Fault detection (FD) is an essential and urgent task for modern wind turbines (WT). Numerous FD techniques developed in the literature and excellent performances achieved for this purpose. However, in real-world application due to sensitive to change in the mechanical system and environmental noise, the degradation in the achievements of fault detection methods remain a severe issue. In this paper, a fault detection procedure is presented for wind turbines to handle varying operating conditions. The process uses a deep belief network (DBN) to extract the nonlinear features and the invariant structures in the inputs signals for efficient and proper estimation. Then, the fault detection methodology is generated based on the obtained estimates. To verify the proposed method, the FD scheme is illustrated using a real wind speed sequence collected from the north of Morocco. Finally, the robust performance of the method is analyzed with a well-tuned neural network.