Abstract:Aiming at the problem that the equipment runs in normal operation state for a long time in actual industrial scenarios,the fault samples are not easy to obtain and the types of samples are not balanced,which lead to performance degradation of data-driven deep intelligent diagnostic model,a two-stage processing model based on Wasserstein-divergence generative adversarial networks(WGAN-div) and deep convolutional neural networks DLA was proposed.WGAN-div was used to generate fault samples to achieve class balance among samples,and the balanced data set was fed into DLA34 network for feature extraction and fault classification.DLA34 could integrate semantic and spatial information of each layer with its special aggregation structure to achieve deeper information sharing.Finally,the bearing failure dataset of Case Western Reserve University was used for verification.The experimental results show that WGAN-div in the proposed model can generate samples that are highly similar to the original samples,and the data balancing effect is better than that of the current mainstream GAN,WGAN and DCGAN.The accuracy of fault recognition completed by DLA34 can reach 100% on the set of data sets.