In view of the unstable and nonlinear characteristics of the rolling bearing vibration signal of rotating electrical machines, the traditional time?frequency analysis method and wavelet packet decomposition method have energy leakage and poor adaptive ability in the signal decomposition process, and the EMD decomposition method has modal aliasing and other problems. In order to improve the fault diagnosis accuracy of rolling bearings, CEEMDAN combined with energy moment method is proposed to extract the original vibration signal features. The weight and offset of ELM hidden layer are optimized by SPSO algorithm, and the CEEMDAN?SPSO?ELM method is used to analyze and diagnose single and multiple damage faults of rolling bearings.The effectiveness of the algorithm and the improvement of diagnosis accuracy are verified by comparative experiments.