Combining multiple deep learning algorithms for prognostic and health management of aircraft

被引：85
|
作者：
Che, Changchang [1 ]
Wang, Huawei [1 ]
Fu, Qiang [1 ]
Ni, Xiaomei [1 ]
机构：
[1] Nanjing Univ Aeronaut & Astronaut, Coll Civil Aviat, Nanjing, Jiangsu, Peoples R China
来源：
AEROSPACE SCIENCE AND TECHNOLOGY | 2019年 / 94卷
基金：
中国国家自然科学基金;
关键词：
Prognostic and health management (PHM); Long short-term memory (LSTM); Deep belief network (DBN); Condition assessment; Fault classification; Remaining useful life (RUL); SHORT-TERM-MEMORY; REMAINING USEFUL LIFE; FAULT-DIAGNOSIS; BELIEF NETWORK;
D O I：
10.1016/j.ast.2019.105423
中图分类号：
V [航空、航天];
学科分类号：
08 ; 0825 ;
摘要：
The development of airborne sensor monitoring and artificial intelligence technologies provides effective tools for precise prognostic and health management (PHM) of aircraft. This paper presents a PHM model which combines multiple deep learning algorithms for condition assessment, fault classification, sensor prediction, and remaining useful life (RUL) estimation of aircraft systems. A long short-term memory (LSTM) based recurrent network is used to predict multiple multivariate time series of sensors, and deep belief network (DBN) is applied to assess system condition and classify faults of aircraft systems. Then, the RUL can be estimated through the integration of condition assessment and sensor prediction. Finally, the proposed algorithm is validated experimentally using NASA's C-MAPSS dataset, and the results showed a lower error rate and deviation than traditional models. (C) 2019 Elsevier Masson SAS. All rights reserved.
引用
收藏
页数：9
相关论文
共 50 条
12345