Research On Bearing Life Prediction Of Doubly Fed Wind Power Generator Based On Data

Wind power generation technology is one of the major research of new energy power generation technology. The reliability research for the components of the wind turbine becomes an important branch of wind power generation technology. At present, doubly-fed wind power generation system is the mainstream of the wind turbine. Meanwhile, generator bearing is the component with the higher failure rate of doubly-fed wind power generation system.In recent years, bearing life prediction based on data is a new research direction in the field of big data, which becomes a hotspot research at home and abroad. Firstly, the bearing operate condition of the double fed induction generator(DFIG) is divided, and the characteristics which can reflect the bearing life are obtained by feature extraction. Then, three kinds of prediction methods are used to predict the bearing life of DFIG bearing. The simulation results show that the extracted features can be used to predict the bearing life by the appropriate prediction method.Bearing feature extraction is aiming to obtain the characteristics to reflect the bearing life by analyzing the monitoring data of DFIG bearing. In order to obtain vibration signal of the same conditions, K-means clustering is used to divide bearing operation condition by using generator power data, the bearing temperature and rotating speed of generator collected from monitoring, Because the existence of the noise in detected vibration signals, a wavelet packet method is used to de-noising vibration signal. Finally, the RMS which can reflect the bearing life is obtained by extracting features of de-noised vibration signal under the same conditions.The method of bearing life prediction based on Weibull distribution is using Weibull distribution function to analyze the trend of RMS of vibration signals. First, the RMS of vibration signals of bearing is transformed into the reliability function value of the Weibull distribution. And then, the parameters of the Weibull distribution are estimated by using least square method. Finally, the Weibull distribution parameters is substituted to the life expectancy calculation formula, after that, the expected life of bearing can be got.The bearing life prediction method based on ARIMA(Auto Regressive Integrated Moving Average Model) is using the time series analysis of the RMS to get the trend of RMS of vibration signal. Firstly, the RMS of vibration signal is processed with equal time interval. Then, the obtained RMS of vibration signal with equal time interval is treated with data smooth processing. Secondly, the criteria of model is used to determine the structure of ARIMA model, and MATLAB is used to determine ARIMA model parameters. Finally, the bearing life of DFIG is predicated by using the obtained ARIMA model.The bearing life prediction method based on Kalman filter is using the Kalman filter to optimize the change trend of the RMS of vibration signal by means of the state space description of RMS of vibration signal. First, the state space model of RMS is established by ARIMA model. Then, Kalman filter is used to realize the optimization of the observations of the ARIMA model. In the end, the more accurate prediction range of bearing life can be acquired.