Session: 05-07 Advanced Diagnostics & Data Analytics I

Paper Number: 124893

124893 - Aero-Engine Remaining Useful Life Prediction via Physics-Informed Self-Attention Encoder 

Prediction of aero-engine remaining useful life (RUL) is crucial for the aeronautic industrial. In recent years, the deep learning enhanced methods for aero-engine RUL prediction has attracted more and more attention by the community. However, how to correctly process information, extract features and carry out feature fusion from cluttered sensors is still a common problem. In this paper, a hybrid framework containing a method of data reduction and feature extraction based physics-informed self-attention encoder is proposed to predict the RUL of aero-engines. The self-attention mechanism is used to process the data. In the part of feature extraction, the encoder-decoder architecture is applied to reduce the data dimensions and extract the features. In the encoder section, data features are embedded in vectors of low dimensions. Then, a neural network based on the self-attention mechanism is constructed to make predictions of RUL. Physical information is embed into the training process. The proposed algorithm framework is experimentally verified on the Commercial Modular Aeronautical Propulsion System Simulation (C-MAPSS) dataset. Results show that the algorithm framework can effectively improve the prediction accuracy of the remaining service life by information processing and feature extraction. In addition, the proposed algorithm framework improves the interpretability in the process of data processing and feature extraction, and has high applicability in accurate predictive maintenance.

Presenting Author: Xuanwu Zhang Zhejiang University

Presenting Author Biography: Ph.D student of Zhejiang University

Authors:

Xuanwu Zhang Zhejiang University
Jiayue Lou Zhejiang University
Jifa Zhang Zhejiang University
Yao Zheng Zhejiang University
Yifan Xia Zhejiang University