Evaluation of Distributionally Robust Uncertainty Quantification Methods for Aero-engine Remaining Useful Life Prediction

Yu WANG; Zhonghua CHENG; Wei ZHANG; Enzhi DONG; Yuqi JIANG

doi:10.12382/bgxb.2025.0188

您当前的位置：

首页 >

文章列表页 >

Evaluation of Distributionally Robust Uncertainty Quantification Methods for Aero-engine Remaining Useful Life Prediction

更新时间：2026-02-10

- Evaluation of Distributionally Robust Uncertainty Quantification Methods for Aero-engine Remaining Useful Life Prediction
- Acta Armamentarii Vol. 46, Issue S2, Pages: 250188(2025)
- 作者机构：
  
  1. 陆军工程大学石家庄校区, 河北石家庄 050003
  2. 河北省机械装备状态监测与评估重点实验室, 河北石家庄 050003
  3. 中国兵器科学研究院, 北京 100089
- 作者简介：
- 基金信息：
- DOI：10.12382/bgxb.2025.0188
  CLC：
- Received：19 March 2025，
  
  Online First：03 February 2026，
  
  Published：2025
- 稿件说明：
移动端阅览
Yu WANG, Zhonghua CHENG, Wei ZHANG, et al. Evaluation of Distributionally Robust Uncertainty Quantification Methods for Aero-engine Remaining Useful Life Prediction[J]. Acta Armamentarii, 2025, 46(S2): 250188.
DOI：

Yu WANG, Zhonghua CHENG, Wei ZHANG, et al. Evaluation of Distributionally Robust Uncertainty Quantification Methods for Aero-engine Remaining Useful Life Prediction[J]. Acta Armamentarii, 2025, 46(S2): 250188. DOI： 10.12382/bgxb.2025.0188.

摘要

针对航空发动机剩余使用寿命预测中不确定性量化方法的不同情况下适用性问题

在Inception架构的基础上提出一种面向同分布与分布外数据的系统性评估框架。基于美国国家航空航天局商用模块化航空推进系统仿真(NASA Commercial Modular Aero-Propulsion System Simulation

N-CMAPSS)数据集

通过集成12种不确定性量化方法并构建多维评价指标体系

揭示不同方法在复杂退化建模中的性能边界与优化方向。实验结果表明:在同分布场景下

深度核学习以最优预测精度(

MAE

=3.18

RMSE

=3.93)满足高精度需求

而基于变分推理的证据下界贝叶斯神经网络凭借最低校准误差(

MACE

=0.059

RMSCE

=0.062)更适合可靠性敏感任务;深度集成通过负对数似然(

NLL

=1.89)验证了其概率建模优势

并实现认知、偶然不确定性的有效解耦。面对分布外数据

基于变分推理的证据下界贝叶斯神经网络在

MAE

(分布外:5.12)和

RMSCE

(分布外:0.083)上表现稳健

而传统方法的RMSE在分布外场景下激增42%

揭示了其分布鲁棒性缺陷。研究进一步提出动态策略推荐机制

针对不同场景差异化选择最优模型

为航空发动机健康管理的不确定性量化模型的选择提供决策。新提出的多维评估范式与不确定性分析方法

为复杂装备智能运维系统提供了理论支撑

为航空发动机预测与健康管理的可信赖、可解释研究奠定了基础。

Abstract

This study addresses the challenge of scenario-specific applicability of uncertainty quantification (UQ) methods in the prediction of aero-engine remaining useful life (RUL).A systematic evaluation framework for both in-distribution (IID) and out-of-distribution (OOD) scenarios is proposed based on the Inception architecture.In order to reveal the performance boundaries and optimization pathways of different evaluation methods for complex degradation modeling

12 UQ methods are integrated and a multi-dimensional evaluation metric system is established based on the NASA commercial modular aero-propulsion system simulation (N-CMAPSS) dataset.Experimental results demonstrate that

under IID conditions

deep kernel learning (DKL) achieves superior prediction accuracy (MAE=3.18

RMSE=3.93)

making it ideal for precision-critical applications

while Bayesian neural networks with variational inference using evidence lower bound (BNN-VI-ELBO) exhibit the lowest calibration errors (MACE=0.059

RMSCE=0.062)

prioritizing reliability-sensitive tasks.Deep ensemble (DE) validates its probabilistic modeling advantage through negative log-likelihood (NLL=1.89) and successfully decouples the epistemic and aleatoric uncertainties.For OOD data

BNN-VI-ELBO maintains robustness with MAE=5.12 and RMSCE=0.083

whereas the RMSE of the traditional methods (e.g.

SWAG) is increased by 42%

exposing their distributional robustness limitations.A dynamic strategy recommendation mechanism is further developed to adaptively select the optimal models for different scenarios

providing a decision-making methodology for aero-engine health management.The proposed multi-dimensional evaluation paradigm and uncertainty decomposition framework provide theoretical foundations for intelligent maintenance systems

advancing trustworthy and interpretable prognostics and health management (PHM) in aviation engineering.

关键词

Keywords

references

LIU N , ZHAO W J , WANG Y , et al. Research on the navigation method of large-scale differential tail-control improvised guided munitions based on rotational speed constraints [J ] . Defence Technology , 2024 , 38 ( 8 ): 155 - 170 .

CHEN Q , SHENG H L , ZHANG J , et al. Novel high-safety aeroengine performance predictive control method based on adaptive tracking weight [J ] . Chinese Journal of Aeronautics , 2024 , 37 ( 7 ): 352 - 374 .

KONG J Z , LIU J , ZHU J Z , et al. Review on lithium-ion battery PHM from the perspective of key PHM steps [J ] . Chinese Journal of Mechanical Engineering , 2024 , 37 ( 4 ): 14 - 35 .

LI H Q , ZHANG Z X , SI T M , et al. A review on physics-informed data-driven remaining useful life prediction:challenges and opportunities [J ] . Mechanical Systems & Signal Processing , 2024 , 209 : 111120 .

WANG B , LEI Y G , LI N B , et al. A hybrid prognostics approach for estimating remaining useful life of rolling element bearings [J ] . IEEE Transactions on Reliability , 2020 , 69 ( 1 ): 401 - 412 .

ZHENG S W , LIU J , CHEN Y X , et al. Causal graph-based spatial-temporal attention network for RUL prediction of complex systems [J ] . Computers & Industrial Engineering , 2025 , 201 : 110892 .

CHAE G S , BAE J S , OH Y K . State-of-health estimation and remaining useful life prediction of lithium-ion batteries using DnCNN-CNN [J ] . Journal of Energy Storage , 2025 , 106 : 114826 .

NEMANI V , BIGGIO L , XUN H , et al. Uncertainty quantification in machine learning for engineering design and health prognostics:a tutorial [J ] . Mechanical Systems and Signal Processing , 2023 , 205 : 110796 .

BAI R , LI Y B , KHANDAKER N , et al. Diversity entropy-based Bayesian deep learning method for uncertainty quantification of remaining useful life prediction in rolling bearings [J ] . Journal of Vibration and Control , 2023 , 29 ( 21/22 ): 5053 - 5066 .

YANIV O , EMILY F , REN J , et al. Can you trust your model’s uncertainty? evaluating predictive uncertainty under dataset shift [J ] . Advances in Neural Information Processing Systems , 2019 , 12 : 14003 - 14014 .

WANG J , GAO R X , YUAN Z , et al. A joint particle filter and expectation maximization approach to machine condition prognosis [J ] . Journal of Intelligent Manufacturing , 2019 , 30 ( 2 ): 605 - 621 .

KANG R , GONG W J , CHEN Y X . Model-driven degradation modeling approaches:Investigation and review [J ] . Chinese Journal of Aeronautics , 2020 , 33 ( 4 ): 1137 - 1153 .

DONG Y L , ZHOU Z , DAI H , et al. Reliability evaluation of LED lamp beads considering multi-stage wiener degradation process under generalized coupled accelerated stress [J ] . Electronics , 2024 , 13 ( 23 ): 4724 .

KUZIO D , ZIMROZ R , WYLOMA SKA A. A modified gamma process for RUL prediction based on data with time-varying heavy-tailed distribution [J ] . Information Sciences , 2025 , 690 : 121603 .

李敬一 , 苏翔 . GWO优化CNN-BiLSTM-Attenion的轴承剩余寿命预测方法 [J ] . 振动与冲击 , 2025 , 44 ( 2 ): 321 - 332 .

LI J Y , SU X . GWO Optimization of CNN-BiLSTM-Attenion bearing remaining life prediction method [J ] . Journal of Vibration and Shock , 2025 , 44 ( 2 ): 321 - 332 . (in Chinese)

GUO J Y , WANG Z Y , YANG Y L , et al. A dual-channel transferable RUL prediction method integrated with Bayesian deep learning and domain adaptation for rolling bearings [J ] . Quality and Reliability Engineering International , 2024 , 40 ( 5 ): 2348 - 2366 .

VAHIDE B . A comprehensive analysis for classification and regression of surface points based on geodesics and machine learning algorithms [J ] . Engineering Computations , 2023 , 40 ( 9/10 ): 2270 - 2287 .

LI Y P , GOODRICH M J , PETERSON E K , et al. Uncertainty quantification in epigenetic clocks via conformalized quantile regression [J/OL ] . medRxiv:the preprint server for health sciences , 2025 ( 2025-2-11 )[ 2025-3-1 ] . https://doi.org/10.1101/2024.09.06.24313192 https://doi.org/10.1101/2024.09.06.24313192 .

SUKRITI S , MIGUEL J L H . Deep Kernel learning for reaction outcome prediction and optimization [J ] . Communications Chemistry , 2024 , 7 ( 1 ): 136 .

ZHANG D W , LIU T C , KANG J . Density regression and uncertainty quantification with Bayesian deep noise neural networks [J ] . Stat , 2023 , 12 ( 1 ): 324 - 330 .

CHI J W . Tourism development and income inequality in OECD countries:new insights from method of moments quantile regression [J ] . Tourism Economics , 2024 , 30 ( 3 ): 767 - 784 .

MORIMOTO M , FUKAMI K , MAULIK R , et al. Assessments of epistemic uncertainty using Gaussian stochastic weight averaging for fluid-flow regression [J ] . Physica D:Nonlinear Phenomena , 2019 , 440 : 133454 .

DEPEWEG S , HERNÁNDEZ-LOBATO J M , DOSHI-VELEZ F , et al. Decomposition of uncertainty in Bayesian deep learning for efficient and risk-sensitive learning [J ] . Advances in Neural Information Processing Systems , 2017 , 80 : 1184 - 1193 .

HERNÁNDEZ-LOBATO J M , LI Y Z , HOLMES D , et al. Black-box α-divergence minimization:arXiv:1511.03243v1 [R ] . Ithaca,NY , US : Cornell University , 2015 : 1511 .03243v1.

GAL Y , GHAHRAMANI Z . Dropout as a Bayesian approximation:representing model uncertainty in deep learning:arXiv:1506.02142 [R ] . Ithaca,NY , US : Cornell University , 2015 : 1506 .02142.

ARIAS C M , KULKARNI C , GOEBEL K , et al. Aircraft engine run-to-failure dataset under real flight conditions for prognostics and diagnostics [J ] . Data , 2021 , 6 ( 1 ): 5 .

HYUNHO M , GIOVANNI I . Multi-objective optimization of extreme learning machine for remaining useful life prediction [J ] . Applications of Evolutionary Computation , 2022 , 13224 : 191 - 206 .

Views

下载量

CNKI被引量

Alert me when the article has been cited

提交

Tools

Publicity Resources

A Gas Path Fault Diagnosis Method for Aero-engine Based on TCN-LGBM Model

An Aero-Engine Quality Assessment Method Based on Particle Swarm Optimized Fermatean Fuzzy Structural Equation Model

Remaining Useful Life Prediction Method of Rolling Bearing Based on Frequency-domain TCAE-Informer

RUL Prediction of Wet Clutches based on Bivariate Inverse Gaussian Degradation Process with Common Random Variable

Effect of ³⁷Cl enrichment on neutrons in a Molten Chloride Salt Fast Reactor

Related Author

Chenfeng SUN

Likun REN

Jie ZHAO

Yongqiang LI

Weimin LÜ

TANG Xilang

SHAN Zhifeng

YANG Qianjun

Related Institution

Naval Aviation University

Equipment Management and UAV Engineering College, Air Force Engineering University

National Key Laboratory of Unmanned Aerial Vehicle Technology

Unit 95866 of the PLA

陆军工程大学石家庄校区

Postal code：100089
Tel：010-68963060/68962718 Fax：010-68963025 Email：bgxb@cos.org.cn
Technical support is provided by Beijing Founder electronics co., LTD 京ICP备05059581号-4 京公网安备11010802024360号
It is recommended to read the content of this site in Chrome&IE9+. Please switch to extreme mode in browser 360.
Cookies We use cookies to help provide and enhance our service and tailor content. By continuing, you agree to the use of cookies.
Total Visits：49245 Visits Today：42

⁰