Data driven sensor and actuator fault detection and isolation in wind turbine using classifier fusion

Renewable energy sources like wind energy are widely available without any limitation. Reliability of wind turbine is crucial in extracting the maximum amount of energy from the wind. Early fault detection, isolation and successful controller reconfiguration can considerably increase the performance in faulty conditions and prevent failures in the system. Along the same vein, fault diagnosis of wind turbine systems has received much attention in recent years. Fault detection methods based on time and frequency domain signal analysis without explicit mathematical model are state-of-the-art in complex processes. This paper investigates data-driven fault detection and isolation (FDI) design based on fusion of several classifiers for a wind turbine benchmark -second challenge. The proposed method is robust against different operational conditions and measurement errors. In fact, we develop a new data-driven FDI scheme, via analytical redundancy. Multi-Layer Perceptron (MLP), Radial Basis Function (RBF), Decision Tree, and K-Nearest Neighbor (KNN) classifiers are implemented in parallel, and fused together. Feature extraction from measurement signals enriches the information about wind turbine condition and improves decision making of proposed FDI scheme. Simulation results and Monte Carlo sensitivity analysis show the effectiveness of the proposed method.