Research On Fault Warning And Diagnosis For Wind Turbine Bearings

Rolling bearing is an important part of the wind turbines, its status will influence directly whether or not the wind farm could run safety. Because of the bad environment, the bearing failure occurs extremely easily cause by wind speed fluctuations。And usually wind farms are far away from the Power plant control room, it would be very hard for us to discover and carry out the repair work if the rolling bearing has some defects. Ranging bearing scrap, heavy lead to more serious failures endangers wind turbine’s safety. Therefore, in order to guarantee the safe and economic operation of unit, having effective monitoring and fault diagnosis of the rolling bearings have important significance.Now, at home and abroad have carried out a lot research works about wind turbine monitoring and alarm, which also including the bearing parts. With the research works forward, some monitoring systems have been productization and applied to some wind farms. Because the research is not enough thorough, function is not perfect, it always result in misstatement and omission of fault events. The main reason is that the existing system is mainly to set alarm threshold, watching the data trend. The system couldn’t connect monitoring parameters with operating condition. And the results of diagnosis given by the system are only a preliminary diagnosis, without positioning to the right fault location and mode; it is not conducive to the implementation of the follow-up maintenance.In this paper, on the basis of study the theory of bearing failures, combining with the characteristics of the wind turbine bearing’s variable operation condition, summarizes the failure mode and the corresponding characteristics of Wind turbine bearing. There will generate amounts of data when the Wind turbines in monitoring process. Once every ten minutes to collect the data, this paper combined with the feature of bearing fault, found the bearing failure’s connotative characteristics in the huge data, proposed multiple parameter fitting deviation degree analysis of early warning methods. Selecting vibration signal of time domain characteristic, bearing temperature, power of wind turbine generator, these characteristic quantities to serve as the predictive failure’s pending parameters. IN this paper, through analyzing the characteristics of the temperature varied with the change of wind turbine generator’s power, the characteristics of different vibration feature parameter varied with the change of wind turbine generator’s power. Develop a reasonable warning rule to ensure the reliability of early warning.Early warning can only express bearing fault cannot accurately position to the fault. So it is necessary to further bearing’s diagnostic work. Characteristics of the vibration signal when bearing failure is cyclical shocks and the signal modulation. Since the unit variable-speed operation and large background noise characteristics make fault feature extraction become difficult. So the focus of the diagnosis is to eliminate the interference of variable conditions, the noise reduction and feature extraction. This paper put forward the order proportion sampling technology, the EMD (Empirical Mode decomposition) de-noising technology and marginal spectrum method to extract the fault information. The order proportion sampling technology solved the problem of the variable condition effectively, and the EMD de-noising technology rejected noise by using the cross correlation coefficient criterion and the kurtosis criterion. Having a significant noise reduction effect. The implementation steps of the method are to use order proportion sampling technology transformation of non-stationary time domain waveform smooth angular domain waveform. Then diagonal domain signal EMD decomposition, decomposition into the IMF, using the relationship between each other several criteria and kurtosis index selection criterion to the IMF, reconstruct the Angle domain signal, Diagonal square domain signal envelope or directly Hilbert changes of the Angle domain signal marginal spectrum, there’s a bug in the marginal spectrum order, through the fault order can accurately locate fault location.