In order to extract the fault information of rolling bearing vibration signals under strong noise coverage and improve the accuracy of fault diagnosis and classification, based on the theory of fuzzy entropy（FE）, a new method of complete ensemble empirical mode decomposition with adaptive noise（CEEMDAN） and convolutional neural network（CNN） is proposed, which make full use of the independence, relative consistency, and the advantages of fuzzy entropy and randomness. The fuzzy entropy of the original signal was obtained by cyclic sampling, decomposed by CEEMDAN method, and the optimal component group was screened by Pearson correlation coefficient. Finally, the optimal component group was input to CNN for fault diagnosis, and the t-SNE popular learning algorithm was used for clustering visualization. The results show that compared with EMD-Fuzzy Entropy and EEMD-Fuzzy Entropy under different working conditions, the proposed method has stronger robustness and generalization, and t-SNE visualization makes the results more intuitive.