Abstract:With the increasing demand for reliability and precision in manufacturing industry, timely and effective acquisition of rotating machinery fault information can ensure the normal operation of equipment.Deep LSTM residual network was used to complete the fault diagnosis of rotating machinery, which mainly consisted of three modules: initial data processing layer, SP-LSTM residual network signal diagnosis layer and GAP-ELM network fault classification layer. This method could realize deep feature mining of initial data and obtain subtle changes of fault data by using memory and forgetting gates in LSTM element. The GAP-ELM network could avoid the problem of low accuracy of traditional Softmax classification method, so as to complete fault diagnosis effectively. The CWRU set was used to complete the experimental comparison between the proposed method and methods in the literature. The results show that the proposed method has better robustness and is superior to methods in the literature in diagnosing normal signals, fault signals of rolling body and inner and outer ring. In addition, the method can be realized in fewer epochs, and with the increase of the epoch, the loss value of the method decreases.