融合多尺度时空图和注意力的航空发动机 RUL 预测

Remaining Useful Life Prediction for Aircraft Engines Based on Multi-Scale Spatio-Temporal GNN and Fusion Attention

摘要

针对现有航空发动机剩余使用寿命（RUL）预测方法忽略时空融合以及局部与全局特征整合的不足，提出了一种基于多尺度时空图和融合注意力的航空发动机剩余寿命预测方法。在时间维度上，设计短期与长期因果卷积分支，结合注意力机制实现多尺度时间特征融合；在空间维度上，分别建模子系统内局部图与子系统间的超图结构，通过门控机制自适应融合局部与全局特征。最后，引入跨尺度时空注意力模块以强化关键时空依赖，提升预测精度。实验表明，该模型在 C-MAPSS 的多个子数据集上的性能均优于现有主流方法，展现出良好的预测准确性。

Abstract

Remaining Useful Life (RUL) prediction is a critical task in the health management of aircraft engines. To effectively model the complex temporal dynamics and spatial structural information, this paper proposes a novel RUL prediction method based on Multi-Scale Spatio-temporal Graphs with Fusion Attention (MSFA-STG). In the temporal dimension, we design parallel branches of short-term and long-term causal convolutions, which are fused using an attention mechanism to capture multi-scale temporal features. In the spatial dimension, both local graphs within subsystems and hypergraph structures across subsystems are constructed, and a gating mechanism is adopted to adaptively fuse local and global spatial features. Furthermore, a cross-scale spatio-temporal attention module is introduced to enhance the modeling of critical spatio-temporal dependencies, thereby improving prediction accuracy. Experimental results on the C-MAPSS dataset demonstrate that the proposed model outperforms mainstream methods, showing strong generalization and high predictive accuracy.

关键词

剩余使用寿命预测 / 时空图神经网络 / 多尺度融合 / 注意力机制 / 故障预测

Key words

remaining useful life prediction / spatio-temporal graph neural network / multi-scale fusion / attention mechanism / fault prognostics

参考文献

[1] GAURAV D, SEETARAM M. Fault diagnosis of machines using deep convolutional Beta—variational autoencoder[J].IEEE Transactions on Artificial Intelligence, 2021, 3(2):287-296.

[2] 郭津延, 郑红, 杜佳宇, 等. 基于双尺度自适应令牌注意力的交通流量预测 [J].华东理工大学学报 (自然科学版), 2025, 51(6):817-826.

[3] RAMI N K, ANGKOON P, ALI H A, et al. Recursive multi—signal temporal fusions with attention mechanism improves EMG feature extraction[J].IEEE Transactions Artificial Intelligence, 2020, 1(2):139-150.

[4] 孙俊静, 顾幸生. 基于注意力机制多尺度卷积神经网络的轴承故障诊断 [J].华东理工大学学报 (自然科学版), 2024, 50(2):247-256.

[5] ZHENG H C, MIN W, RUI Z, et al. Machine remaining useful life prediction via an attention—based deep learning approach[J].IEEE Transactions on Industrial Electronics, 2021, 68(3):2521-2531.

[6] ZHANG Y X, LI Y X, WANG Y L, et al. Adaptive spatio—temporal graph information fusion for remaining useful life prediction[J].IEEE Sensors Journal, 2022, 22(4):3334-3347.

[7] JIA Z Y, LIN Y F, WANG J, et al. Multi—view spatial—temporal graph convolutional networks with domain generalization for sleep stage classification[J].Computing Research Repository, 2021, 29:1977-1986.

[8] XIAO H, WANG J H, CENG M J, et al. A CNN—BiGRU predictive maintenance method for industrial equipment based on attention mechanism[C]//2025 7th International Conference on Artificial Intelligence Technologies and Applications (ICAITA). Wenzhou, Zhejiang, China: IEEE, 2025:88-91.

[9] CAO L A, LI Y F, CHEN J W, et al. Dual—channel degradation monitoring based on graph neural network for aero—engine remaining useful life prediction[J].IEEE Transactions on Instrumentation and Measurement, 2025, 74:1-19.

[10] ABHINAV S, KAI G, DON S, et al. Damage propagation modeling for aircraft engine run—to—failure simulation[C]//IEEE Conference on Prognostics and Health Management. Denver, CO, USA: IEEE, 2008:1-9.

[11] YU C W, MIN W, XIAO L L, et al. Multivariate time—series representation learning via hierarchical correlation pooling boosted graph neural network[J].IEEE Transactions on Artificial Intelligence, 2024, 5(1):321-333.

[12] THOMAS N, MAX W. Semi—supervised classification with graph convolutional networks [EB/OL]. (2017—02—22).https://doi.org/10.48550/arxiv.1609.02907.

[13] ZHOU H Y, ZHANG S H, PENG J Q, et al. Informer: Beyond efficient transformer for long sequence time—series forecasting [C]//AAAI Conference on Artificial Intelligence. California, USA: AAAI Press, 2021:11106-11115.

[14] YU T X, HONG S Y, YU D Z, et al. A dual—stage attention—based ConvLSTM network for spatio—temporal correlation and multivariate time series prediction[J].International Journal of Intelligent Systems, 2021, 36(5):2036-2057.

[15] LING C, DONG H C, ZONG J S, et al. Multi—scale adaptive graph neural network for multivariate time series forecasting[J].IEEE Transactions on Knowledge and Data Engineering, 2023, 35(10):10748-10761.

[16] HUI L, ZHEN Y L, WEI Q J, et al. Remaining useful life prediction using a novel feature—attention—based end—to—end approach[J].IEEE Transactions on Industrial Informatics, 2020, 17(2):1197-1207.

[17] XIANG L, QIAN D, JIAN Q S. Remaining useful life estimation in prognostics using deep convolution neural networks[J].Reliability Engineering & System Safety, 2017, 172:1-11.

[18] 袁烨, 黄虹, 程骋, 等. 基于特征注意力机制的 GRU—GAN 航空发动机剩余寿命预测 [J].中国科学: 技术科学, 2022, 52(1):198-212.

[19] ZHANG Z K, LI D, WU S Y, et al. BACE—RUL: A bidirectional adversarial network with covariate encoding for machine remaining useful life prediction[C]//Collaborative Computing: Networking, Applications and Worksharing, Cham: Springer, 2025:473-490.

[20] XUAN L L, YA W H, HANG W, et al. A closed—form continuous—depth neural—based hybrid difference features re—representation network for RUL prediction[J].Reliability Engineering & System Safety, 2025, 253:110540.

[21] ZHE L, BING L, CHANG Y F, et al. A Multi—scale cross—dimension interaction approach with adaptive dilated TCN for RUL prediction[J].Scientific Reports, 2025, 15(1):19229.

[22] JIN Y, MENG Z G, HU Q H, et al. Heterogeneous dynamic—aware GNN for RUL prediction of aero—engine[C]//IEEE Transactions on Instrumentation and Measurement. [s.l.]: IEEE,2025:1-14.

基金资助

上海市 2024 年度“科技创新行动计划”资助(24BC3200500, 24BC3200300)