Data-based Fault Diagnosis And Fault Tolerant Control For Wind Turbines

With the in-depth development of economic globalization and increasingly intensified market competition pressure in the 21 st century, the human is becoming more and more depend on energy, and energy crisis is becoming more and more severe deterioration. As a clean, green and renewable energy, wind power has received attention around the world. Besides, wind turbines, as a n effective tool for utilizing wind power, have got considerable development, such as, vaster stand-alone capacity of wind turbines, higher degree of automation and so on. Interest in advanced controllers for normal operation has expanded in recent years, but fault detection and fault tolerant control for wind tu rbines is a less well-developed area of interest. In term of wind turbines, abnormal condition forecast, fault detection and fault tolerant control technology play a crucial role in cost-effective, quality and reliability operation. Mechanism model based performance monitoring and fault diagnosis methods have been researched for several decades, and a relatively mature theoretical system have been established. However, it is difficult to obtain precise models for the increasing complexity wind turbines. Hence, it is significant to construct a real-time fault diagnosis and fault tolerant control system via process measurements for wind turbines.This paper proposes a holonomic data based fault diagnosis and fault tolerant control system for wind turbines on the basis of previous research. Firstly, design a partial least squares based optimal variable algorithm to select several variables which are most useful for detecting fault from large number of sensor variables after briefly introducing wind turbines model. Secondly, combine DO observer, parity space and subspace identification methods, structure residual generators directly from the input and output measurements, and utilize generalized likelihood ratio processing residual signals to realize real-time fault detection. After faults have been detected, possibilistic C-means algorithm is chosen to hand the faults classification question, which has a lower error rate than ot her classification methods when never occurred faults appear. Then, data based soft sensor and Expanded Internal Model Controller based on Youla parameterization, co-prime decomposition theory are designed for sensor faults and actuator faults, respectively. Finally, several typical wind turbine failures are designed to demonstrate the fault diagnosis and fault tolerant control scheme proposed in this paper under SIMULINK circumstance. The results show that the scheme can not only accurately monitor system performance and detect wind turbine faults. But also guarantee the stability of the system to some extent when sensor or actuator faults occur.