Generative Adversarial Networks Enhanced Extreme Learning Machine to Classify Faults in Rolling Bearings

Abstract

Complete faulty data is significance in fault classification. However, in engineering applications, the data we obtained is usually incomplete and unbalanced. In the present, a new generative adversarial networks (GANs) enhanced extreme learning machine (ELM) is proposed. Using GANs, numerous new samples similar to the raw signal are generated and further combined with the raw signal to obtain synthetic samples. ELM is activated by the synthetic samples as training samples to detect unknown faulty signals (test samples). Experiments show that the proposed method has better performance than ELM to overcome the shortcomings of lower classification accuracy.