The main drive system of CNC lathe is the core component of the machine tool,and its failure can cause machining quality and even operational safety problems.Digital twin technology can reduce the difficulty of fault diagnosis,but the current research still suffers from low efficiency of physical entity to virtual entity conversion and neural network overfitting problems.To solve the above problems,a fault diagnosis method based on digital twin and regularized BP neural network was proposed.A digital twin model of CNC lathe main drive system was established,and the exchange of twin data between physical and virtual entities was completed through OPC UA communication.Four regularization methods to improve the overfitting problem were compared and analyzed,and a fault diagnosis model was constructed based on regularized BP neural network by drop out method.By comparing the loss functions and prediction accuracy of BP neural network, DropOut-BPNN and convolutional neural network under different signal-to-noise ratios,the feasibility of the diagnostic model and the applicability of the algorithm are verified.