Wind Turbine Fault Diagnosis with Generative-Temporal Convolutional Neural Network

Condition monitoring of wind turbines (WTs) has attracted a great deal of attention due to fast development of WTs. The inherent intermittence of wind energy and locating WTs in remote areas, makes designing proper fault diagnosing method difficult. To address this issue, we have proposed a two-block deep learning based method in this paper, which encapsulates two feature extraction and classification stages in an end-to-end architecture. In the designed method, we have utilized generative adversarial network (GAN) as feature extraction block and temporal convolutional neural network (TCNN) as fault classifier block. The proposed structure benefits from the leverage GAN and TCNN. The simulation results based on real-data from a 3 MGW WT in Ireland, which is obtained from supervisory control and data acquisition system (SCADA) demonstrates that it is a suitable alternative for WTs' fault classification. To show the superiority of the proposed method, the results are compared with support vector machine (SVM) and feed-forward neural network (FFNN).