Global attention mechanism based deep learning for remaining useful life prediction of aero-engine

Aero-engine is one of the core components of aircraft. The accurate prediction of areo-engine Remaining Useful Life (RUL) is of great significance for ensuring the operation safety of aircraft. The emergence of attention mechanisms allows the deep learning model to effectively focus on important data features for RUL prediction tasks, which can improve the accuracy of RUL prediction for aero-engine. However, most mainstream dual attention mechanisms currently calculate attention weights for different dimensions separately, making it difficult to obtain global information. A novel global attention (GA) mechanism has been proposed in this paper that overcomes existing challenges and effectively identifies relevant data features for accurate RUL predictions. The structure design of GA adopts a very novel idea, requiring no internal recurrent neural network (RNN) or convolutional neural network (CNN) modules. Instead, it utilizes two pooling operations to extract features from different sensors and time steps, which are then calculated in parallel to obtain global attention coefficients. This enables the deep learning based model to adaptively learn the most important information of the input. Moreover, the parallel computation design avoids wastage of computational resources. On this basis, we combine GA with self-attention (SA) mechanism and temporal convolutional network (TCN) to propose an end-to-end deep learning RUL prediction method. To validate the effectiveness of the proposed method, the C-MAPSS aero-engine dataset is used. Experimental results demonstrate the superiority of our method compared to state-of-the-art RUL prediction methods.