Aiming at the problems of sample scarcity and model complexity when deep learning is applied to bearing fault diagnosis, this paper studied and proposed a fault diagnosis method combining data augmentation module AugModule and light weight classification network LightNet. STFT is used to convert the one-dimensional vibration signal of the device into a two-dimensional time-frequency map, and the data set is expanded by introducing spectral normalization and AdaptiveMix improved WGAN. The expanded data set is input into the LightNet network for image classification. LightNet is improved based on Shuf-flenetV2. It uses Scconv to improve accuracy, uses SMU activation function to replace Relu, proposes Channel rate and Channel crossing strategies to optimize performance, and resets the number of Block stacks in each Stage. To achieve the best balance between accuracy and efficiency. The experimental results show that the proposed method has higher accuracy and practical engineering significance while reducing the model parameters.