Research On Model Based Robust Fault Diagnosis Methods For Wind Turbine

At present the traditional regular inspection method is mainly used in the fault detection of the wind turbine which has low detection efficiency and can’t find the potential failure, so the fault detection method for wind turbine need to be further studied. The commonly used signal-based fault diagnosis method has some limitations when applied to the wind turbine for its complex structure. If the model-based fault diagnosis method can applied to wind turbines will effectively improve the efficiency of diagnosis for it can find potential failures and separate fault easily. The research object of this paper is the three-blade horizontal axis wind turbine and the main purpose is researching the technology of fault detection and isolation (FDI) for the wind turbine to improve the rapidity of fault detection and the level of intelligence. The research on model-based diagnosis technique is made both in theoretical and practical aspects to carry out a more systematic and in-depth in this paper, the main work is as follows:(1) The existing fault diagnosis method and the model-based fault diagnosis methods are reviewed, the commonly used residual generation method for deterministic system, the robust residual generation method for uncertainty system, the threshold selection method and the fault isolation method are introduced. Also, the state-space equation based mathematical model of the pitch system, drive train system and converter system is established according to the structure of wind turbine.(2) For the pitch system and converter system are mainly influenced by the external noise, the Kalman filter is used as residual generator to minimum the covariance of the residual signal. The generalized likelihood ratio is chosen for threshold setting after generate residual signal and the dual sensor redundancy is used to isolate the actuator fault and sensor fault of the pitch system and convertor system.(3) For the drive train system is not only influenced by the external noise but also influenced by the unknown input, so an optimal observer combined the advantage of unknown input observer and Kalman filter is designed as residual generator to decouple the interference and reduce the effect of noise. The generalized likelihood ratio is chosen for threshold setting after generate residual signal and the dual sensor redundancy is used to isolate the sensor fault of the drive train system.A complete set of model-based fault detection and isolation scheme is designed in this paper and its effectiveness has been proved both by theory and simulation. The conclusions of this study provide an important reference for the fault diagnosis of wind turbine system in engineering.