Abstract: To address the issue of scarce data in the fault diagnosis of rotor bar breaks in induction motors, this study designs a transfer fault diagnosis method that integrates wavelet transform and convolutional neural network(CNN) technology. Initially, Ansys software was used to establish models of healthy and rotor bar break fault induction motors, and three-phase current signals were collected. A Butterworth band-stop filter was designed to remove the fundamental wave. Then, the one-dimensional current signals were transformed into time-frequency feature maps through wavelet transform to extract features rich in fault information. On this basis, the optimal network structure parameters were determined through multiple experiments, and a CNN model was constructed. This model takes the time-frequency map as input, and the output layer uses a sigmoid classifier for rotor bar break fault diagnosis. To achieve model transfer, the feature maps obtained from simulation were used to train the CNN, and the trained model was directly applied to the fault diagnosis of actual motors without additional training data. Experimental results show that the proposed method can accurately identify different types of rotor bar break faults, with a diagnosis accuracy rate of 100%. It also has good applicability to actual induction motors, with a diagnosis accuracy rate of 100%, verifying the effectiveness of model transfer.