Abstract: In order to improve the fault identification accuracy of rolling bearing, a fault diagnosis method based on adaptive local iterative filtering (ALIF) and time-shifted multi-scale fluctuation dispersion entropy(TMFDE) was proposed. Firstly, the vibration signal of rolling bearing was decomposed by ALIF to obtain a set of IMF components. Secondly, to obtain more integrated IMF components, the importance of each IMF component was evaluated based on the energy method, and the first three components were regarded as effective components. Then, the TMFDE was used to quantify the feature information in the effective component and construct the fault feature vector. Finally, the fault features were input into the extreme learning machine optimized by particle swarm optimization for fault identification. The method was tested by using the rolling bearing data of Southeast University. The results show that the method can accurately identify the type of fault, and compared with other methods, this method still has excellent stability when the amount of data is small.