Abstract:In order to deal with non-stationary vibration signal of wind turbines gearbox under time-varying conditions, the fault feature extraction method combining spectral kurtosis and Vold-Kalman Filter Based Order Tracking (VKF-OT) is put forward. By the method, after set rotation and meshing frequency as extracting frequency of VKF-OT, the order component with speed changes was extracted, then, the vibration amplitude and phase could be obtained directly from the complex envelop of each order component. The experimental analysis was carried out to prove that this method could retain the transient information of gearbox. Spectral kurtosis of two frequencies components was calculated, and the ratio of frequency band energy which corresponding Maximum spectral kurtosis and total energy of the original order signal was extracted as fault feature. Finally, the features of 150 groups of vibration signal from wind turbine gear box under different conditions was described by using Gaussian mixture model, and Maximum Bayesian classifier was used to achieve failure recognition. The amount of recognition rate indicates that this method can identify local early weak fault of gear in arbitrary time-varying conditions effectively.