Fault Detection of Wind Turbine System Based on Deep Learning and System Identification

Early fault diagnosis in the Wind Turbine System (WTS) can reduce the maintenance cost and increase system reliability. However, due to its complexity, analysing WTS is a challenging task. Furthermore, impacts of component faults may be negligible and difficult to detect depending on their types and severity. This paper presents an integrated framework using the Long Short-term Memory and Mixture Density Network (LSTM-MDN) with real-time system identification to diagnose incipient and subtle faults which tend to modify the measurement or dynamics of the system. At first, the LSTM-MDN is constructed to estimate the blade and pitch sub-system outputs with the corresponding variance. Then system identification based on the Auto-Regressive Integrated Moving Average with eXogenous input (ARIMAX) is implemented to generate meta-data which can reveal any changes in system dynamics. In order to reduce the false alarm rate in the presence of uncertainties, we introduce an adaptive threshold using the MDN parameters and a pruning process for detecting different operation modes. Finally, a comprehensive comparison is conducted to evaluate the effectiveness of the proposed method.