As a transmission system component with an integrated self-sensing unit, smart bearings can sense more accurate fault information. However, the randomness of the failure type and time of smart bearings makes it difficult to construct sufficient prior data sets for training accurate fault identification model. Aiming at this problem, an adaptive method of information and model based on digital twin is proposed for smart bearing fault diagnosis with small sample. Firstly, the bearing dynamics model is established based on the fault mechanism, and the simulation domain model data is obtained under the condition consistent with the real state. Then, the data information adaptation of different domains was carried out based on the proposed digital-twin driven information transfer method, the information distribution difference between the simulation domain data and the actual measurement data is reduced. Finally, the improved CNN model was used for digital-analog driven smart bearing fault identification, and the feature model is transmitted by the model transfer learning method. The process of model transfer learning further reduces the distribution difference between the features of the domain model. Experimental results indicate that the suggested approach diminishes the discrepancy between simulated and actual domain data, enhancing the accuracy in identifying faults in smart bearings.