The accurate extraction of the abnormal vibration characteristics of generator drive bearing is the basis of realizing the mechanical running state detection and prediction of the fatigue damage of motor bearing. At present, the feature extraction method of the abnormal vibration of generator drive bearing is obtained by using the method of power spectrum density feature extraction. Because of the characteristic of power spectral density, the following performance is not good for the abnormal vibration of generator bearing transmission bearing under the alternating temperature. An abnormal vibration spectrum feature extraction algorithm based on Hilbert transform for generator bearing is proposed. The dynamic model of a multi parameter generator is established, and the vibration analysis model of bearing is built.The abnormal vibration of generator is decomposed by empirical mode decomposition and Hilbert spectrum. The Hilbert spectrum of the abnormal vibration of generator is extracted. Based on the structure and dynamic parameters of the drive system, the gear mesh modification is carried out. Simulation results show that the method is stable, reliable, and performance is superior, and the state monitoring and damage prediction ability of generator is improved.