Rolling bearing fault diagnosis method based on SOA-BiLSTM

Abstract

To address the problem that the effectiveness of bearing fault diagnosis in long short-term memory (LSTM) networks depends on the combination of model hyperparameters, a method based on Snake Optimizer Algorithm (SOA) with Addictive Attention is proposed to search the global optimal hyperparameters of LSTM is proposed. First, SOA is used to find the optimal hyperparameter combinations of the LSTM, then the data are input to the LSTM under the optimal parameter combinations in forward and inverse order, respectively, and finally the output is stitched as the final diagnosis result. The experimental results show that SOA can search for the most suitable hyperparameters of LSTM, can effectively improve the diagnostic results of LSTM and make LSTM have stronger fault diagnosis ability.