Degradation prediction for complex systems such as inertial measurement units (IMUs) is challenged by the need for nonlinear modeling, limited physical interpretability, and multi-source heterogeneous data fusion. To address these issues, a degradation prediction model integrating physics-informed neural ordinary differential equations (PINN-NODE) and heterogeneous graph neural networks (HGNN) was proposed. A parallel dual-branch architecture was used. In the temporal dimension, continuous-time dynamics under physical constraints were modeled by PINN-NODE. Irregularly sampled data were handled. In the spatial dimension, coupling relationships between sensors and environmental factors were captured by HGNN. Environmental stresses were reconstructed using dynamic edge weights. Spatiotemporal features were fused by a multilayer perceptron (MLP) to obtain an end-to-end predictor. On the IMU degradation dataset, RMSE is reduced by 56.43%, and MAE is reduced by 58.65%, relative to an LSTM baseline. The proposed model is shown to provide improved prediction accuracy with enhanced physical interpretability.
YU Jin-he, ZHAO Jian-yin.
Complex System Degradation Prediction Method Based on PINN-NODE and HGNN[J].
Science Technology and Engineering, 2026, 26(29): 12618-12627 DOI:10.12404/j.issn.1671-1815.2507689
惯性测量单元(inertial measurement unit, IMU)是现代导航、控制和定位系统的中心器件,广泛应用于航空航天、自动驾驶及无人平台等领域[1]。IMU在长期使用过程中会持续受到器件老化、温度变化、机械振动等因素的影响,其零偏、比例因子等关键性能参数将发生漂移,进而导致测量精度下降,严重时甚至危害系统可靠性与运行安全[2]。因此,对IMU性能退化过程进行准确建模与趋势预测,是实现剩余寿命评估和预测性维护的重要基础,对保障系统安全具有重要意义。
LiuZhenyu, HuiZeyu, GuoXu, et al. Tightly coupled simultaneous localization and mapping algorithm for LiDAR inertial measurement unit based on sliding window optimization[J]. Science Technology and Engineering, 2022, 22(21): 9167-9175.
[3]
MiJ, WangQ, HanX. Low-cost MEMS gyroscope performance improvement under unknown disturbances through deep learning-based array[J]. Sensors and Actuators A: Physical, 2024, 368: 115086.
NingShaohui, RongYouzhi, DongZhencai. Residual life prediction of rolling bearings based on PCA and DRSN-TCN[J]. Science Technology and Engineering, 2025, 25(27): 15096-15104.
[6]
HeZ Y, LüW M, HuW L. Residual life prediction of rate gyroscope based on Wiener process and Kalman correction[J]. Journal of Ordnance Equipment Engineering, 2019, 40(8): 53-58.
[7]
TanT T, ZhangS Y, YangH Y, et al. Storage life evaluation of MEMS gyroscope based on Wiener process[J]. Equipment Environmental Engineering, 2024, 21(2): 59-64.
WangXin, MengTianyu, ZhouJunxi. Remaining useful life prediction for aircraft engine using attention based LSTM[J]. Science Technology and Engineering, 2022, 22(7): 2784-2792.
[10]
LiH W, HaoS, NiY Q, et al. Hybrid structural analysis integrating physical model and continuous-time state-space neural network model[J]. Computer-Aided Civil and Infrastructure Engineering, 2025, 40(2): 166-180.
[11]
HuY, WeiR, YangY, et al. Performance degradation prediction using LSTM with optimized parameters[J]. Sensors, 2022, 22(6): 2407.
[12]
Najera-FloresD A, HuZ, ChadhaM, et al. A physics-constrained Bayesian neural network for battery remaining useful life prediction[J]. Applied Mathematical Modelling, 2023, 122: 42-59.
[13]
YanR, ZhouZ, ShangZ, et al. Knowledge driven machine learning towards interpretable intelligent prognostics and health management: review and case study[J]. Chinese Journal of Mechanical Engineering, 2025, 38(1): 5.
[14]
ChaoM A, KulkarniC, GoebelK, et al. Fusing physics-based and deep learning models for prognostics[J]. Reliability Engineering & System Safety, 2022, 217: 107961.
[15]
RaissiM, PerdikarisP, KarniadakisG E. Physics-informed neural networks: a deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations[J]. Journal of Computational Physics, 2019, 378: 686-707.
[16]
NageshC K, SankaranarayananS, KaurR, et al. Taylor-model physics-informed neural networks (PINNs) for ordinary differential equations[C]// Proceedings of the International Conference on Neuro-symbolic Systems. Philadelphia, PA: PMLR, 2025: 621-642.
[17]
CuomoS, Di ColaV S, GiampaoloF, et al. Scientific machine learning through physics-informed neural networks: Where we are and what's next?[J]. Journal of Scientific Computing, 2022, 92(3): 88.
[18]
LiX E, WangJ, SunY, et al. Physics-informed hybrid neural architecture for coupled degradation modeling and remaining useful life prediction of LiFePO4 batteries[J]. Journal of the Electrochemical Society, 2025, 172(6): 060505.
[19]
FalasS, AsprouM, KonstantinouC, et al. Robust power system state estimation using physics-informed neural networks[J]. IEEE Transactions on Industrial Informatics, 2025, 21(10): 8057-8067.
[20]
ChenR T Q, RubanovaY, BettencourtJ, et al. Neural ordinary differential equations[J]. Advances in Neural Information Processing Systems, 2018, 31: 6572-6583.
[21]
KarniadakisG E, KevrekidisI G, LuL, et al. Physics-informed machine learning[J]. Nature Reviews Physics, 2021, 3(6): 422-440.
[22]
RubanovaY, ChenR T Q, DuvenaudD. Latent ODEs for irregularly-sampled time series[C]// Proceedings of the 33rd International Conference on Neural Information Processing Systems. Vancouver, BC: Neural Information Processing Systems, 2019: 5320-5330.
[23]
BonnafféW, SheldonB C, CoulsonT. Neural ordinary differential equations for ecological and evolutionary time-series analysis[J]. Methods in Ecology and Evolution, 2021, 12(7): 1301-1315.
[24]
FangZ, LongQ, SongG, et al. Spatial-temporal graph ode networks for traffic flow forecasting[C]// Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining. New York: Association for Computing Machinery (ACM), 2021: 364-373.
HuDengyu, WangBaohua, LiuJinhong. Fault diagnosis in distribution networks with distributed generation based on improved graph neural network[J]. Science Technology and Engineering, 2025, 25(21): 8936-8944.
[27]
WangX, JiH, ShiC, et al. Heterogeneous graph attention network[C]// The world Wide Web Conference. San Francisco, CA: ACM, 2019: 2022-2032.
[28]
ZhouZ, BaskerR, YeungD Y. Graph neural networks for multivariate time-series forecasting via learning hierarchical spatiotemporal dependencies[J]. Engineering Applications of Artificial Intelligence, 2025, 147: 110304.
[29]
JinM, KohH Y, WenQ, et al. A survey on graph neural networks for time series: forecasting, classification, imputation, and anomaly detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2024, 46(12): 10466-10485.
[30]
BingR, YuanG, ZhuM, et al. Heterogeneous graph neural networks analysis: a survey of techniques, evaluations and applications[J]. Artificial Intelligence Review, 2023, 56(8): 8003-8042.
[31]
NgoQ H, NguyenB L H, VuT V, et al. Physics-informed graphical neural network for power system state estimation[J]. Applied Energy, 2024, 358: 122602.
[32]
WangX, BoD, ShiC, et al. A survey on heterogeneous graph embedding: methods, techniques, applications and sources[J]. IEEE Transactions on Big Data, 2022, 9(2): 415-436.