Abstract: In the motor bearing fault diagnosis based on the deep learning, generally adopt data augmentation method based on generative adversarial networks (GANs) in order to get sufficient fault data, thereby ensuring the outstanding performance of the model. Problems such as poor data generation quality, slow network training speed and tedious training may occur in the data augmentation of 1D time-domain signal. To solve these problems, this paper proposed a data augmentation method based on 2D gray pixel images and auxiliary classifier generative adversarial networks (2D-ACGANs). Firstly, the raw time-domain signal was converted into 2D gray pixel images to obtain the input data suitable for the two-dimensional convolutional neural network. On this basis, combining with the ACGANs, the label of the raw data was used as the input of the network for data augmentation. Compared with the 1D data augmentation method, this method can effectively reduce the quantity of network training parameters, and at the same time solved the problems of tedious training and loss of label information in the traditional method. Finally, the proposed method was applied to the real data of motor bearings fault diagnosis for comparison and verification. The results show that the improved 2D-ACGANs algorithm can generate higher quality data, effectively improve the fault recognition accuracy and the network training speed, and has good engineering application feasibility.