A large number of complex features need to be extracted for the fault diagnosis of wind turbine rolling bearings. A parallel bearing fault diagnosis model based on attention mechanism

ResNext network and long short-term memory (LSTM) network was proposed. Firstly

the collected one-dimensional vibration signal was preprocessed; then it was input into the model in two ways to extract features

and one of them was input into the ResNext module embedded in the attention mechanism. The attention mechanism can increase the weight of important features and reduce model operations. The other channel was input to the LSTM network to extract the dependence of the vibration signal on the time series. Finally

the two extracted features are fused and input to the Softmax layer for fault classification. The experimental results show that

compared with the current bearing fault diagnosis method based on deep learning

the proposed method performs better in bearing fault classification accuracy.