Gearbox Fault Diagnosis of Wind Turbines Based on Attention and Multi-scale CNN

Abstract: A gear box fault diagnosis method 1DMSCNN-AM is proposed, which integrates an attention mechanism with multi-scale convolutional neural network, aimed at improving accuracy in diagnosing fan gearbox faults in noisy environments. Firstly, preprocess the fault dataset that collects one-dimensional vibration signals of the gearbox; Secondly, the processed one-dimensional dataset is directly input into the multi-scale convolutional neural network model 1DMSCNN-AM with attention mechanism for training. During the training process, the model uses CBAM attention mechanism to weight different scale features, highlight key information to improve the representation ability of fault features and the model's noise resistance. In the data fusion layer, a multi head attention mechanism is adopted to capture the dependency relationships of different subspaces in the input features and enhance the feature representation ability; Finally, a fully connected layer is used to classify the gearbox fault dataset. Through experimental verification, the diagnostic accuracy of the method proposed in this article on the gearbox dataset can reach over 99.80%. Compared with other methods under noise interference of different signal-to-noise ratios, 1DMSCNN-AM still has strong fault diagnosis ability.