Distance self-attention network method for remaining useful life estimation of aeroengine with parallel computing

Abstract

Remaining useful life (RUL) estimation of aeroengine is significant in the health monitoring, operation and maintenance of aircrafts. Traditional deep learning methods fail to consider the degradation rules of aeroengine and have low computational efficiency for RUL estimation. Therefore, a novel deep learning architecture called distance self-attention network (DSAN) is developed based on self-attention and parallel computing on time series. In the proposed DSAN method, a distance function is developed to improve the matching ability of self-attentions and optimize the feature extraction capability, and the fusion layer inspired by the computation of recurrent neural network (RNN) is developed to fuse historical information and real-time data. The effectiveness of the DSAN method for RUL estimation is validated by utilizing the Commercial Modular Aero Propulsion System Simulation (C-MAPSS) data provided by NASA. It is revealed that the DSAN method is superior to the typical methods such as convolutional neural network (CNN) and long-short term memory (LSTM), because the root mean square error (RMSE) decreased by 7.3%∼ 25.3%, and the Score reduced by 28% ∼51.8%. The efforts of this paper provide a promising method for aeroengine RUL estimation, which has the potential to support the health monitoring and predictive maintenance of multi-sensor systems.