Fault detection and diagnosis of the pitch system in a wind turbine using an adaptive Kalman filter and a hybrid deep learning model

This study presents a fault detection and diagnosis framework for the pitch system of a wind turbine. Fault detection is performed using an adaptive Kalman filter (AKF), which adaptively updates the process and measurement noise covariance matrices during operation, thereby improving the reliability and accuracy of state estimation under various fault conditions. For fault diagnosis, a hybrid convolutional neural network and long short-term memory (CNN-LSTM) model is developed to identify fault types by capturing both local temporal patterns and long-term dependencies from time-series data. This study presents the first integration of a hybrid CNN-LSTM network with a model-based AKF framework for wind turbine pitch system fault diagnosis. Four fault scenarios were considered: actuator stuck fault, pitch sensor fixed fault, pitch sensor bias fault, and pitch sensor scaling fault. The CNN-LSTM model was trained, tested, and validated using simulated data generated under various operating conditions to improve the multi-fault classification accuracy and robustness. Simulation results demonstrate that the proposed framework can reliably detect pitch system faults and accurately diagnose their types. Therefore, the combined AKF and CNN-LSTM framework provides an effective and reliable solution for real-time condition monitoring and predictive maintenance of wind turbine pitch systems.