Remaining useful life prediction of aircraft engine based on degradation pattern learning. (August 2017)

• Record Type: Journal Article
• Authors: Zhao, Zeqi; Bin Liang, ; Wang, Xueqian; Lu, Weining
• Is Part Of: Reliability engineering & system safety. Volume 164(2017)
• Page Start: 74
• Page End: 83
• Publication Date: 2017-08
• Subjects: Prognostic and health management -- Remaining useful life prediction -- Degradation pattern learning -- Neural network; Reliability (Engineering) -- Periodicals; System safety -- Periodicals; Industrial safety -- Periodicals; Fiabilité -- Périodiques; Sécurité des systèmes -- Périodiques; Sécurité du travail -- Périodiques; 620.00452
• DOI: 10.1016/j.ress.2017.02.007
• Languages: English
• ISSNs: 0951-8320

Abstract

Abstract: Prognostics, which usually means the prediction of the field reliability or the Remaining Useful Life (RUL), is the basis of Prognostic and Health Management (PHM). Research in this paper focuses on remaining useful life prediction of aircraft engine in the same gradual degradation mode. As the gradual degradation with same failure mechanism has some regularity in macro, there would be certain relation between an arbitrary point of the degradation process and the correspondent RUL. This paper tries to learn this certain relation via neural network and the learned network, which reflects the relation, can be partly perceived as degradation pattern. The main prognostic idea of degradation pattern learning is firstly proposed and illustrated. And then an improved back propagation neural network is designed and analyzed as the implementation technique, in whose loss function an adjacent difference item is added. Next details of implementation via adjacent difference neural network are elaborated. Finally, the proposed approach is validated by two experiments respectively using different aircraft engine degradation datasets. Results of the experiments show a relatively good prediction accuracy, which verifies the correctness, effectiveness and practicability of the idea. Highlights: Degradation pattern learning methodology based on mapping idea is proposed. Item of adjacent difference for neural network is proposed. Preprocessing framework for degradation data and a series of indicators are designed. Prediction of remaining useful life on basic and extensional datasets is promising.