Data-Driven Model-Based Fault Diagnosis in a Wind Turbine With Actuator Faults

Abstract

Wind turbines are renewable energy conversion devices that are being deployed in greater numbers. However, today’s wind turbines are still expensive to operate, and maintain. The reduction of operational and maintenance costs has become a key driver for applying low-cost, condition monitoring and diagnosis systems in wind turbines. Accurate and timely detection, isolation and diagnosis of faults in a wind turbine allow satisfactory accommodation of the faults and, in turn, enhancement of the reliability, availability and productivity of wind turbines. The so–called model-based Fault Detection and Diagnosis (FDD) approaches utilize system model to carry out FDD in real-time. However, wind turbine systems are driven by wind as a stochastic aerodynamic input, and essentially exhibit highly nonlinear dynamics. Accurate modeling of such systems to be suitable for use in FDD applications is a rather difficult task. Therefore, this paper presents a data-driven modeling approach based on artificial intelligence (AI) methods which have excellent capability in describing complex and uncertain systems. In particular, two data-driven dynamic models of wind turbine are developed based on Fuzzy Modeling and Identification (FMI) and Artificial Neural Network (ANN) methods. The developed models represent the normal operating performance of the wind turbine over a full range of operating conditions. Consequently, a model-based FDD scheme is developed and implemented based on each of the individual models. Finally, the FDD performance is evaluated and compared through a series of simulations on a well-known large offshore wind turbine benchmark in the presence of wind turbulences, measurement noises, and different realistic fault scenarios in the generator/converter torque actuator.