A Wind Gearbox Fault Detection Method Based on a Lightweight Attention Network

It is of great importance to realize effective fault diagnosis of wind turbines gearbox in modern industrial systems. This paper proposes a fault diagnosis method that based on a lightweight attention network. This method uses a lightweight convolutional neural network MobileNetV3_L based on the Transformer to improve the new network RepViT network for wind gearbox fault diagnosis. The diagnosis process includes short-time Fourier transform processing of the laboratory artificial wind gearbox fault data set, and establishing a fault classification and recognition model based on a lightweight convolutional neural network. Finally, the model is used to perform fault diagnosis on a real wind gearbox dataset to verify its performance. Experimental research is conducted on the University of Macau gearbox dataset to validate the proposed method. The results demonstrate that the method exhibits superior stability and accuracy than other methods.