The sound signal has the advantage of non-contact measurement when collecting, but it is easy to be interfered by ambient noise, resulting in a low signal-to-noise ratio, which is not conducive to the acquisition of feature information.In order to extract effective feature information from the sound data of rolling bearings and realize accurate fault identification, a sound signal fault diagnosis strategy based on the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and hierarchical fluctuation dispersion entropy (HFDE) was proposed.In this strategy,CEEMDAN was used to alleviate the mode confusion defect of ensemble empirical mode decomposition (EEMD);aiming at the defect that the high-frequency information of time series could not be considered in traditional multi-scale fluctuation dispersion entropy (MFDE),a hierarchical fluctuation dispersion entropy (HFDE) nonlinear dynamic index based on hierarchical processing was proposed.The proposed strategy was applied to the fault identification of rolling bearings, and the sound data under different fault conditions could be detected.Through numerical simulation and analysis of rolling bearing experimental data, the proposed method was compared with CEEMDAN-MFDE, EEMD-HFDE, EEMD-MFDE, HFDE and MFDE.The results show that the accuracy rate of the proposed method reaches 100%, and the average recognition accuracy rate of multiple experiments also reaches 995%, which is higher than the comparison methods, thus the effectiveness and superiority of the strategy are verified.