Research On Fault Analysis And Diagnosis Of W Wind Turbine Doubly Fed Induction Generators

In our country, wind resource is mainly distributed in western areas and southeast coastal areas. Windfarms are usually built in the kind of environment where the temperature differential between day and night is very high, wind condition fluctuates frequently and violently or the climate is wet. So there are many kinds of fault modes which occur frequently. As the wind turbines are compact-sized and high-integrated, the entire device always has to be changed once a failure occurs. And the windfarms are often remote and on wind turbine is far away from each other, so the maintenance will cost high and cause great economic loss.More and more manufacturer focus on the early warning and diagnosis of wind turbines. Through installing different measuring points on different parts of the wind turbine, we can monitor the operating state of every part of the wind turbine in real time, figure out abnormal condition of the wind turbine in advance or in time, analyze and diagnosis the faults accordingly, decide the maintenance order and measures to guide the work.In this paper, a process of fault warning and diagnosis is proposed according to the feature of wind turbine doubly-fed induction generator. Take the Sinovel SL1500 wind turbine as the research object, build its model to deeply analyze its structure feature and function, and finally determine what is the typical fault modes in generator system in view of the real situation, including rotor unbalance, rotor misalignment, bearing failure, stator windings inter-turn short circuit, rotor windings inter-turn short circuit. Analyze the basic mechanism of every typical fault mode and establish the entire database of them, including how and why it occurs, develops, how it influences the component, the system and the wind turbine and how to fix it.Angular domain resampling is used to change nonstationary time-domain signal into stationary angular-domain signal, insuring the signal is in integer period. Introduce the no-dimensional factors into the fault early warning of the generator mechanical faults as no-dimensional factors are sensitive to faults. A fault process is proposed for doubly-fed induction generator on the basis of neural network, taking electric, vibration and SCADA parameters as input. Vibration parameters are the amplitude of the fault order in order spectrum. The normal range of SCADA parameters are got in different temperature-speed operating condition. And the abnormity rate is raised up as the input. For electric signals, the wavelet packet analysis method is used to extract the effective value of current feature band. Then use the input information to diagnosis the faults with neural network method, adjust the operating mode and make maintenance plans accordingly.At last the entire method is summarized, streamlined and become a part of the "Wind Power Equipment Condition Monitoring and Management System". The system has been completed and installed in the windfarm to demonstrate the method and improve it continuously, putting it into practice.