Fault diagnosis of time-varying speed gearbox based on gated recurrent dropout attention unit

In response to the difficulty of fault diagnosis of gearbox under time-varying speed conditions, this paper presents a novel approach for diagnosing gearbox faults in time-varying speed, utilizing an improved gate recurrent unit (GRU), which adds attention gate mechanism and cyclic dropout learning strategies on the basis of the GRU, and constructs a new model named as gated recurrent dropout attention unit (GRDAU). By introducing attention gate mechanism to realize allocating weights dynamically, focusing on key features, and enhancing GRU’s ability to capture important information. In addition, the designed cyclic dropout learning strategy reduces excessive dependence on specific hidden states by randomly discarding some hidden state information. Finally, the robustness and excellent interference suppression ability of the proposed method were verified through case analysis of a gearbox under time-varying speed, and the diagnostic accuracy of the method is as high as 99.78 %. Comparative experiments were conducted to validate its superior performance and stronger generalization ability compared to existing advanced diagnostic methods.