Abstract: In order to solve the problems such as low accuracy of fan bearing fault diagnosis and inability of real-time con-dition monitoring due to insufficient training samples of newly built wind farms in new power system, a digital twin model of fan main bearing fault diagnosis based on the fusion of model knowledge and data drive is pro-posed in this paper. Firstly, the twin model of the main bearing of the fan is constructed by MATLAB/Simulink to generate simulation data under different working conditions, so as to simulate the fault data that is difficult to obtain in the actual operation of the main bearing of the fan, so as to solve the problem of insufficient data of the newly built wind farm and unable to generate fault data lossless. Then, a fault diagnosis model based on deep convolutional networks and improved residual blocks is proposed. By adding the BN layer and dropout layer, end-to-end real-time fault diagnosis of one-dimensional original vibration signals can be realized without data preprocessing. Finally, a large number of balanced data sets simulated by digital twinning technology were used for pre-training of the model, and the trained model was put into the actual operation of the fan main bearing for fault diagnosis through transfer learning.The experiment shows that taking CWRU data set as the real operation data of fan main bearing as an example, the feasibility of the proposed metho d is verified. The fault diagnosis accuracy of the proposed digital twin model method of wind turbine main bearing is 99.7%, marking a 22% improvement over the traditional CNN method without digital twin enhancement.