For the past few years, In the research of intelligent monitoring of industrial equipment, deep learning is becoming a method that get the widespread concern of researchers. In general, the collection of a great quantity of typical data that has been labeled makes the deep learning approach a great success. However, it is often limited by real fault data samples, and the generalization ability of the established model is poor. A novel fault diagnostic method that we assemble data generation and fault diagnosis called ASM1D-GAN is proposed to address these problems. This method is composed of 1D convolutional neural network, Generative Adversarial Networks (GANs), and fault classifier. We assemble the data generation and fault diagnosis procedures together. Through a new antagonistic machine learning mechanism, the ASM1D-GAN model is optimized, so as to achieve higher sample quality and fault mode classification ability. This novel method can extract fault features from natural fault samples and generate effective new ones. The experimental results demonstrate the effective fault feature generation ability and the superior fault diagnostic ability of the proposed method.
