A Data-Driven Residual-Based Method for Fault Diagnosis and Isolation in Wind Turbines

Abstract

In order to improve the reliability of wind turbines, avoid serious accidents, and reduce operation and maintenance costs, it is important to effectively detect early faults of wind turbines operating in harsh environment. This paper proposes a data-driven fault diagnosis and isolation method for wind turbines, which implements long short-term memory networks for residual generator and applies the random forest algorithm for decision making. The method has been evaluated in a wind turbine benchmark Simulink model, in comparison with four model-based algorithms and four data-driven methods, and the results have shown that the proposed method achieves the highest accuracy. Moreover, extensive evaluation has been conducted to analyze the robustness of proposed method, and the experimental results have verified the stability of the proposed method in diagnosis of wind turbine faults.