Abstract:On the actual working conditions, the operation of rolling bearing is not all the constant speed. Sometimes it also works in a variable, so studying on vibration signal of changing speed process is of great significance. In the process of its variable speed, vibration signal also contains a large number of characteristic information. If using the traditional spectrum analysis, the results appear serious phenomenon of“frequency smear”which is difficult to extract the fault information.So a method which studies on fault diagnosis based on the combination of order tracking and wavelet packet was proposed. First of all, the original timedomain signal which was sampled in ascending period was converted to angulardomain signal, then the angulardomain signal was decomposed and reconstructed by the wavelet packet .The reconstructed signal was analyzed by order spectrum analysis. Through the analysis of simulation signal and example, it can effectively diagnose the rolling bearing fault, which shows the effectiveness of the method.