Abstract: In view of the dependence of the current migration diagnosis algorithm on the size and quality of training samples and the difficulty of collecting and labeling bearing fault data in the special working condition environment, a bearing fault diagnosis method based on the co-migration of multi-mode simulation data is proposed. First, the bearing fault dynamics model embedded with actual working conditions parameters is used to generate various fault simulation signals, which solves the problem of insufficient actual fault samples and missing labels. Then, multiple sub-source domains are established based on the analysis of the migratable modes of the simulation data, and the unsupervised iterative migration of each sub-source domain is introduced by the geometric statistical joint alignment method, which overcomes the negative migration problems caused by insufficient information of single mode migration and excessive differences in cross-domain features. Finally, an optimized fuzzy integral decision fusion method is used to collaboratively assign the pseudo-labels of multi-mode features in the migration iterations to gradually improve the credibility of the target domain labels and the domain adaptation capability of the migration model. The experimental results show that the proposed method is driven by fault simulation data and can achieve the accurate identification of various bearing faults without the guided migration of the measured label data. The method is robust to changes in working conditions and target domain sample size and has good application prospects in the field of high-end bearing fault diagnosis supported by non-complete data.