Remaining Useful Life Prediction of Aero-Engine Based on Transformer with Tendency Retainment

Abstract

One of the essential technologies for prognostics and health management of aero-engines is remaining useful life (RUL) prediction. Many deep learning models have recently been presented to extract features adaptively and forecast RUL end-to-end. However, it is still a challenging task to model data of long-life cycles and retain the degradation information when extracting features. To overcome the problem, we present a Transformer-based method with tendency retainment to predict RUL. Convolutional neural network is first used to fuse data from different sensors. Then, the long-life cycle data is encoded by Transformer encoder followed by long short-term memory neural network to extract features and finally RUL is predicted. Moreover, a tendency retainment module is designed based on contrastive learning to maintain the degradation information. The proposed method's performance is validated using NASA's C-MAPSS aero-engine dataset. The experimental results reveal that the proposed method outperforms other state-of-the-art methods in terms of prediction accuracy.