1.Key Laboratory of Modern Measurement & Control Technology of Ministry of Education,Beijing Information Science and Technology University,Beijing,100192
2.Mechanical Electrical Engineering School,Beijing Information Science and Technology University,Beijing,100192
3.Safety Production Department,Sinovel Wind Group Co. ,Ltd. ,Beijing,100000
To address the challenges of accuracy and uncertainty quantification in RUL prediction of high-end rotating machinery, a prediction method was proposed based on VDGP. The method achieved recursive uncertainty quantification by constructing a deep Gaussian process update model, and enhanced large-scale data processing capability through the use of inducing points and variational inference. Experiments on the C-MAPSS and wind turbine planetary gearbox datasets demonstrate that VDGP achieves higher prediction accuracy and narrower confidence intervals compared to the standard Gaussian process methods. On the C-MAPSS FD002 dataset, the root mean square error and scoring function are reduced by 0.21% and 45.3% respectively, relative to the best baseline method.
针对上述问题,本文构建了变分深度高斯过程(variational deep Gaussian process, VDGP)模型,用于实现高端旋转机械及其关键部件的RUL预测,并提供预测结果的不确定性量化,提高寿命预测的准确度。该方法通过构建不确定量分布函数的深度高斯过程(DGP)来更新模型,随着新监测信息的不断获取,持续更新并修正不确定量的分布函数,逐步精确量化预测结果的不确定性。为解决高斯过程(GP)在大规模监测数据下计算成本高的问题,在DGP中引入变分推断,并采用诱导点技术降低计算复杂度,提高RUL的预测精度。本文还对比了不同核函数和模型层数下的性能表现,为所提方法在工业场景下的应用提供参考。
1 高斯过程
高斯过程(GP)是一种非参数贝叶斯模型,用于对潜在函数f( x )的先验分布建模,在给定输入集合时诱导出输出的联合高斯分布。输入 x 与输出y的关系为
y=f( x )+ε
式中:f( x )表示噪声去除后的真实映射;ε是均值为0、方差为的高斯噪声。
的定义为,其中,m( x )为均值函数,k( x, x ′)为核函数, x 、 x ′为输入变量。使用较为广泛的RBF内核
训练DGP模型等价于最大化输出 Y 的边缘似然。为简化符号表示,设 F(l)为对应 X(l)(l=1,2,…,L1)、 Y 的无噪声变量即 X(l)= F(l)+ ε(l)(l=1,2,…,L1)和 Y = F(L)+ ε(L)。由于DGP每层的输出仅依赖其上一层,因此具有条件独立性性质。基于条件独立性属性,隐藏层 Y 的条件概率密度
传统的高斯模型计算成本高。对于旋转机械长时间运行监测的大规模数据集,矩阵求逆会导致计算复杂度迅速增加。为降低计算复杂度并使后续推断可行,引入诱导点近似来逼近原始GP。选取较少的诱导点能显著降低计算量,从而提高大规模数据集下的计算效率。在该近似框架下,每层 GP 可由诱导点 Z(l)及其诱导变量 u(l)表示:
式中:为输出与诱导点集合 Z(l)的协方差矩阵;为诱导点之间的协方差矩阵。
基于诱导点近似的GP采用变分推断来近似真实后验分布,并通过最大化证据下界(evidence lower bound, ELBO)来优化变分参数和模型超参数。记 ft 是与 yt 对应的无噪声函数值,为变分分布,其中,为对诱导变量的变分后验分布。为便于推导,符号中省略了输入 Xt 与诱导点 Z 的显式表示(隐含在协方差矩阵的计算中),则对数边际似然为
通过应用詹森不等式得到变分下界:
式中:N(*)为高斯分布的概率密度函数; KNN 为p( fd | X )的协方差矩阵; KNM 为 fd 和 ud 之间的协方差矩阵,下标d表示多输出(多维)高斯过程中的第d个输出维度; KMM 为诱导点集合上的协方差矩阵;为变分分布对应的变分参数集合。
此处使用GP诱导变量对数似然积分定理,即在稀疏高斯过程(SGP)框架下,给定一个GP映射 Y ~GP( X ),该GP的诱导变量为 U,诱导点为 Z,对于特定的输出维度d,有
每个高斯层不仅输出预测的均值和方差,还将不确定性通过层与层之间传递,使得最终输出层能反映整个网络中逐层传播的不确定性。VDGP的第一层会根据输入 x 计算出对应的预测分布。第二层将该分布作为输入并进行推理,输出新的预测分布。通过多层递归计算,VDGP能捕捉并传播每层的不确定性,使最终输出的预测结果不仅具有更高的准确性,也能全面量化整体的不确定性。
表5中的4种VDGP变体模型在C-MAPSS发动机数据集上的应用效果已被验证,但这些模型在新数据集上的泛化能力有待验证。因此,选择LSTM[28]、MLP[42]、CNN[27]、LSTM with attention[29]、Bi-level LSTM Scheme[35]、CBLSTM[40]、LSTM-Z[41]与4种VDGP变体模型进行对比分析,它们的性能结果见表9。
GAOPengjie, WANGJunliang, SHIZiqi, et al. Long-term Temporal Attention Neural Network with Adaptive Stage Division for Remaining Useful Life Prediction of Rolling Bearings[J]. Reliability Engineering & System Safety, 2024, 251: 110218.
[2]
LINRuiguan, YUYaowei, WANGHuawei, et al. Remaining Useful Life Prediction in Prognostics Using Multi-scale Sequence and Long Short-term Memory Network[J]. Journal of Computational Science, 2022, 57: 101508.
[3]
CHENChuang, LUNingyun, JIANGBin, et al. Prediction Interval Estimation of Aeroengine Remaining Useful Life Based on Bidirectional Long Short-term Memory Network[J]. IEEE Transactions on Instrumentation and Measurement, 2021, 70: 3527213.
LIUXiaofeng, KANGYingying, BOLin. Self-driven Independent Degradation Trajectory Construction and Remaining Life Gray Prediction for Bearings[J]. China Mechanical Engineering, 2024, 35(9): 1613-1621.
[6]
CHENZuoyi, HUANGHong-Zhong, DENGZhongwei, et al. Shrinkage Mamba Relation Network with Out-of-distribution Data Augmentation for Rotating Machinery Fault Detection and Localization under Zero-faulty Data[J]. Mechanical Systems and Signal Processing, 2025, 224: 112145.
[7]
ZHUTing, CHENZhen, ZHOUDi, et al. Adaptive Staged Remaining Useful Life Prediction of Roller in a Hot Strip Mill Based on Multi-scale LSTM with Multi-head Attention[J]. Reliability Engineering & System Safety, 2024, 248: 110161.
[8]
CAOWei, MENGZong, LIJimeng, et al. A Remaining Useful Life Prediction Method for Rolling Bearing Based on TCN-transformer[J]. IEEE Transactions on Instrumentation and Measurement, 2024, 74: 3501309.
[9]
PENGPeng, LIYonghua, GUOZhongyi. High-performance Remaining Useful Life Prediction for Aeroengine Based on Combining Health States and Trajectory Similarity[J]. Engineering Applications of Artificial Intelligence, 2025, 141: 109799.
[10]
CHENZesheng, TUXiaotong, HUYue, et al. Real-time Bearing Remaining Useful Life Estimation Based on the Frozen Convolutional and Activated Memory Neural Network[J]. IEEE Access, 2019, 7: 96583-96593.
XIARan, SUChun. Remaining Useful Life Prediction for Lithium-ion Batteries Based on Health Indicators and Hybrid Bi-LSTM-NAR Model[J]. China Mechanical Engineering, 2024, 35(5): 851-859.
[13]
WANGZhe, YANGLechang, FANGXiaolei, et al. Image-based Remaining Useful Life Prediction through Adaptation from Simulation to Experimental Domain[J]. Reliability Engineering & System Safety, 2025, 255: 110668.
[14]
LUOPeien, YINZhonggang, ZHANGYanqing, et al. A Novel Whale Optimization Algorithm Based on Music Theory Knowledge for RUL Prediction of Motor Bearing[J]. IEEE Transactions on Instrumentation and Measurement, 2023, 72: 3534211.
CAOXiangang, YEYu, ZHAOYoujun, et al. Remaining Useful Life Prediction of Rotating Machinery Based on KPCA-LSTM[J]. Journal of Vibration and Shock, 2023, 42(24): 81-91.
[17]
LIBiao, TANGBaoping, DENGLei, et al. Self-attention ConvLSTM and Its Application in RUL Prediction of Rolling Bearings[J]. IEEE Transactions on Instrumentation and Measurement, 2021, 70: 3518811.
LIUHuakai, DINGKang, HEGuolin, et al. Research on the Prediction Method of Remaining Useful Life of Gears by Combining Fault Mechanism and Convolutional Neural Network[J]. Journal of Mechanical Engineering, 2024, 60(12): 116-125.
[20]
LIUXiaofei, LEIYaguo, LINaipeng, et al. RUL Prediction of Machinery Using Convolutional-vector Fusion Network through Multi-feature Dynamic Weighting[J]. Mechanical Systems and Signal Processing, 2023, 185: 109788.
[21]
WANGBiao, LEIYaguo, LINaipeng, et al. Multiscale Convolutional Attention Network for Predicting Remaining Useful Life of Machinery[J]. IEEE Transactions on Industrial Electronics, 2020, 68(8): 7496-7504.
[22]
DUVENAUDD. Automatic model construction with Gaussian processes[D]. Cambridge, UK: University of Cambridge, 2014.
[23]
LIYanhui, YANGPengfei, BAILu, et al. BRDF Modeling and Optimization of a Target Surface Based on the Gradient Descent Algorithm[J]. Applied Optics, 2023, 62(36): 9486-9492.
[24]
HEIMESF O. Recurrent Neural Networks for Remaining Useful Life Estimation[C]∥2008 International Conference on Prognostics and Health Management. Denver: IEEE, 2008: 1-6..
[25]
SAXENAA, GOEBELK, SIMOND, et al. Damage Propagation Modeling for Aircraft Engine Run-to-failure Simulation[C]∥2008 International Conference on Prognostics and Health Management. Denver: IEEE, 2008: 1-9.
[26]
LIHan, ZHAOWei, ZHANGYuxi, et al. Remaining Useful Life Prediction Using Multi-scale Deep Convolutional Neural Network[J]. Applied Soft Computing, 2020, 89: 106113.
[27]
LIUHui, LIUZhenyu, JIAWeiqiang, et al. A Novel Deep Learning-based Encoder-decoder Model for Remaining Useful Life Prediction[C]∥2019 International Joint Conference on Neural Networks (IJCNN). Budapest: IEEE, 2019: 1-8.
[28]
THANGB, HERNANDEZ-LOBATOD, HERNANDEZ-LOBATOJ, et al. Deep Gaussian Processes for Regression using Approximate Expectation Propagation [C]∥Proceedings of the 33rd International Conference on Machine Learning (ICML). New York, 2016: 1472-1481.
[29]
DAMIANOUA. Deep Gaussian Processes and Variational Propagation of Uncertainty[D]. Sheffield: University of Sheffield, 2015.
[30]
XIEYucen, ZOUJianxiao, PENGChao, et al. A Novel PEM Fuel Cell Remaining Useful Life Prediction Method Based on Singular Spectrum Analysis and Deep Gaussian Processes[J]. International Journal of Hydrogen Energy, 2020, 45(55): 30942-30956.
[31]
SATEESH BABUG, ZHAOPeilin, LIXiaoli. Deep Convolutional Neural Network Based Regression Approach for Estimation of Remaining Useful Life[C]∥Database Systems for Advanced Applications. Cham: Springer, 2016: 214-228.
[32]
ZHENGShuai, RISTOVSKIK, FARAHATA, et al. Long Short-term Memory Network for Remaining Useful Life Estimation[C]∥2017 IEEE International Conference on Prognostics and Health Management (ICPHM). Dallas: IEEE, 2017: 88-95.
[33]
CHENZhenghua, WUMin, ZHAORui, et al. Machine Remaining Useful Life Prediction via an Attention-based Deep Learning Approach[J]. IEEE Transactions on Industrial Electronics, 2020, 68(3): 2521-2531.
[34]
BENKERM, FURTNERL, SEMMT, et al. Utilizing Uncertainty Information in Remaining Useful Life Estimation via Bayesian Neural Networks and Hamiltonian Monte Carlo[J]. Journal of Manufacturing Systems, 2021, 61: 799-807.
[35]
GAOHui, LIYibin, ZHAOYing, et al. Dual Channel Feature Attention-based Approach for RUL Prediction Considering the Spatiotemporal Difference of Multisensor Data[J]. IEEE Sensors Journal, 2023, 23(8): 8514-8525.
[36]
ElDALIM, KUMARK D. Fault Diagnosis and Prognosis of Aerospace Systems Using Growing Recurrent Neural Networks and LSTM[C]∥2021 IEEE Aerospace Conference (50100. SkyBig: IEEE, 2021: 1-20.
[37]
DUANYuhang, LIHonghui, HEMengqi, et al. A BiGRU Autoencoder Remaining Useful Life Prediction Scheme with Attention Mechanism and Skip Connection[J]. IEEE Sensors Journal, 2021, 21(9): 10905-10914.
[38]
XIATangbin, SHUJunqing, XUYuhui, et al. Multiscale Similarity Ensemble Framework for Remaining Useful Life Prediction[J]. Measurement, 2022, 188: 110565.
[39]
SONGTao, LIUChao, WURui, et al. A Hierarchical Scheme for Remaining Useful Life Prediction with Long Short-term Memory Networks[J]. Neurocomputing, 2022, 487: 22-33.
[40]
LINLin, WUJinlei, FUSong, et al. Channel Attention & Temporal Attention Based Temporal Convolutional Network: a Dual Attention Framework for Remaining Useful Life Prediction of the Aircraft Engines[J]. Advanced Engineering Informatics, 2024, 60: 102372.
[41]
LIYasong, ZHOUZheng, SUNChuang, et al. Life-cycle Modeling Driven by Coupling Competition Degradation for Remaining Useful Life Prediction[J]. Reliability Engineering & System Safety, 2023, 238: 109480.
YANXiaojia, LIANGWeige, ZHANGGang, et al. Prediction Method for Mechanical Equipment Based on RCNN-ABiLSTM[J]. Systems Engineering and Electronics, 2023, 45(3): 931-940.
[44]
WANGHuan, LIUZhiliang, PENGDandan, et al. Understanding and Learning Discriminant Features Based on Multiattention 1DCNN for Wheelset Bearing Fault Diagnosis[J]. IEEE Transactions on Industrial Informatics, 2020, 16(9): 5735-5745.
[45]
YOUDazhang, CHENLinbo, LIUFei, et al. Intelligent Fault Diagnosis of Bearing Based on Convolutional Neural Network and Bidirectional Long Short-term Memory[J]. Shock and Vibration, 2021, 2021(1): 7346352.
ZHUANGYuxuan, LIQi, YANGBingru, et al. An End-to-end Approach for Bearing Fault Diagnosis Based on LSTM[J]. Noise and Vibration Control, 2019, 39(6): 187-193.