In view of the inconsistencies in the distribution of the bearing data training set and test set under variable working conditions， the generalization ability of the intelligent diagnosis model is insufficient， and the problem needs to be modeled separately for different tasks. The paper proposes an intelligent fault diagnosis method for rolling bearings based on multi-scale mixed domain feature extraction and domain adaptation. Firstly， the variational mode decomposition is used to extract the features of the multi-scale mixed domain which constructs a complete feature space. Secondly， the dimensionality reduction and optimization of features are realized through the random forest algorithm to eliminate redundant information. Finally， the optimized features are combined with the manifold embedded distribution alignment method to realize the distribution alignment of data in different fields and cross-domain diagnosis. Data sets under different working conditions are used for verification， and compared with traditional intelligent diagnosis methods and traditional transfer learning methods， the results show that the proposed method can accurately and effectively realize cross-domain diagnosis.