This study proposes a fault diagnosis model that combines a digital twin with a multiscale parallel one-dimensional convolutional neural network. A digital twin model of the table feed system was first constructed and simulation experiments of various working conditions were conducted to obtain the missing fault data in the actual physical space.On this basis, the study utilizes the acquired signals to train the proposed migration model for diagnosis.The model extracts different types of fault features from the analog and real signals, respectively, through an intermediate multi-scale convolution algorithm. In addition, the model reduces the distributional disparities between the real and analog signals by using the Wasserstein distance as a regular term to impose constraints on the machine learning process.The study conducted simulation experiments, and the results indicated that the fault periods of the simulated and actual signals of bearing outer ring faults were 0.198s and 0.196s,respectively,with a relative error of only 1.02%.The average fault periods of the actual and simulated signals of the bearing inner ring faults were 0.199s and 0.197s, respectively, with a relative deviation of only 0.48%.In addition,the classification accuracy of the proposed model can be maintained above 95%. Thus,the proposed model has good practical value.