Aero Engines Fault Diagnosis Method Based on Convolutional Neural Network Using Multiple Attention Mechanism

At present, the aero engines fault diagnosis method based on convolutional neural network can only deal with the features of local region and it was difficult to effectively capture the long-range dependencies which leads to the problem of low accuracy. Attention mechanism can solve this problem to some extent now. However, the typical attention blocks don't have sufficient interaction for input features in a certain dimension. To solve it, we propose multiple attentional mechanism, which is implemented in the form of multiple attention block. In this block, we introduce channel attention mechanism for the problem that non-local attention block includes all features into the calculation of attention matrix but channel feature interaction is still insufficient, so that it can make full use of spatial features, consider the interaction of channel features and finally further improve the interaction of global features. The input and output dimensions of this method are consistent, so it can be freely embedded into the structure of the convolutional neural network. The experimental results based on CFM56-7B engines of an airline showed that the proposed mechanism could effectively improve the recognition accuracy of the model, and was higher than other neural network structures, which proved that the multiple attention mechanism has strong generalization, robustness and competitiveness.