Health status assessment and remaining useful life prediction of aero-engine based on BiGRU and MMoE

Prognostics and health management (PHM) is a critical work to ensure the reliable operation of industrial equipment, in which health status (HS) assessment and remaining useful life (RUL) prediction are two key tasks. However, traditional PHM frameworks perform the two tasks separately, which ignore...

Celý popis

Uložené v:

Podrobná bibliografia
Vydané v:	Reliability engineering & system safety Ročník 220; s. 108263
Hlavní autori:	Zhang, Yong, Xin, Yuqi, Liu, Zhi-wei, Chi, Ming, Ma, Guijun
Médium:	Journal Article
Jazyk:	English
Vydavateľské údaje:	Barking Elsevier Ltd 01.04.2022 Elsevier BV
Predmet:	Aerospace engines Bidirectional gated recurrent unit Decision making Health status assessment Industrial equipment Information processing Life prediction Multi-gate mixture-of-experts Optimal tradeoff weight Predictions Reliability engineering Remaining useful life prediction Useful life Remaining useful life prediction Multi-gate mixture-of-experts Health status assessment Bidirectional gated recurrent unit Optimal tradeoff weight
ISSN:	0951-8320, 1879-0836
On-line prístup:	Získať plný text
Tagy:	Pridať tag Žiadne tagy, Buďte prvý, kto otaguje tento záznam!

Popis
Shrnutí:	Prognostics and health management (PHM) is a critical work to ensure the reliable operation of industrial equipment, in which health status (HS) assessment and remaining useful life (RUL) prediction are two key tasks. However, traditional PHM frameworks perform the two tasks separately, which ignore the internal relationship between the two tasks and reduce the efficiency of PHM. To solve the above issues, a dual-task network structure is proposed in this paper based on bidirectional gated recurrent unit (BiGRU) and multi-gate mixture-of-experts (MMoE), which simultaneously evaluates the HS and predict the RUL of industrial equipment. To be specific, BiGRU is used to bidirectionally extract shared information from sensor signals for HS and RUL, and MMoE structure is employed to adaptively differentiate between HS assessment and RUL prediction tasks and realizes a weighted decision making. Furthermore, a loss function based on homoscedastic uncertainty is adopted to learn optimal tradeoff weight between HS assessment loss and RUL prediction loss based on probabilistic modeling, which avoids a time-consuming manual weight tuning process. Experiments on C-MAPSS of aero-engines degradation dataset verify that the proposed method performs better than current popular models, and robustness of the proposed method is satisfactory. •A dual-task network structure is used to realize HS assessment and RUL prediction.•MMoE is proposed to improve the generalization of the dual-task network model.•A new loss function is presented to learn the tradeoff weight between the dual tasks.
Bibliografia:	ObjectType-Article-1 SourceType-Scholarly Journals-1 ObjectType-Feature-2 content type line 14
ISSN:	0951-8320 1879-0836
DOI:	10.1016/j.ress.2021.108263