基于 LSTM-贝叶斯融合模型的器件退化识别与寿命预测

Device degradation identification and lifetime prediction based on an LSTM-Bayesian fusion model

摘要

面对高可靠性器件退化数据及失效样本稀缺、寿命预测不确定性难以量化的问题，本文提出一种融合长短期记忆（LSTM，long short-term memory）网络、Bootstrap集合学习与蒙特卡洛dropout（MCD，Monte-Carlo dropout）的小样本退化寿命预测方法首先，将退化序列转化为滑动窗口样本，并采用Bootstrap训练模型集合其次，在预测阶段，结合MCD采样，综合表征数据与模型的不确定性，获得点预测以及85%、90%和95%预测区间，利用早期观测信息进行退化分型并分别训练子模型，以提高长期外推的稳定性并减小预测区间宽度最后，采用80 ℃条件下20只钨丝灯泡的加速退化试验（ADT，accelerated degradation test）照度序列对本文所提方法进行验证结果表明，该方法能够稳定识别退化类型，准确预测退化趋势，并给出可靠的剩余寿命区间，可为高可靠性器件的可靠性评估与寿命论证提供量化依据。

Abstract

To address the scarcity of degradation data and failure samples and the difficulty in quantifying lifetime-prediction uncertainty for high-reliability devices, this paper proposes a small-sample degradation lifetime prediction method. The method integrates a long short-term memory (LSTM) network, Bootstrap ensemble learning and Monte Carlo dropout (MCD). Firstly, the degradation sequences are transformed into sliding-window samples, and Bootstrap is used to train a model ensemble. Secondly, in the prediction stage, by incorporating MCD sampling, the data uncertainty and model uncertainty are jointly characterized, and point predictions as well as the 85%, 90%, and 95% prediction intervals are obtained. Early observations are then used for degradation typing, and sub-models are trained separately to improve the stability of long-term extrapolation and to reduce prediction interval width. Finally, the proposed method is validated using the illuminance sequences of 20 tungsten filament bulbs obtained from an accelerated degradation test (ADT) at 80 ℃. The results show that the proposed method can stably identify degradation types, accurately predict degradation trends and provide reliable remaining useful life intervals. It thus offers a quantitative basis for the reliability assessment and life substantiation of highly reliable devices.

关键词

加速退化试验（ADT） / 长短期记忆（LSTM）网络 / 小样本退化数据 / 置信区间 / 早期分型

Key words

accelerated degradation test (ADT) / long short-term memory (LSTM) network / small-sample degradation data / confidence interval / early typing

参考文献

[1] Liu Yishi, Hong Sheng, Zio E, et al. Integrated fault estimation and fault-tolerant control for a flexible regional aircraft[J]. Chinese Journal of Aeronautics, 2022, 35(3): 390-399.

[2] Liu Guihong, Guan Qiang, Tang Yincai, et al. Interval modeling for Gamma process degradation model[J]. Symmetry, 2022, 14(5): 954.

[3] Zhou Huaxi, Zhou Changguang, Wang Xiaoyi, et al. Degradation reliability modeling for two-stage degradation ball screws[J]. Precision Engineering, 2022, 73: 347-362.

[4] Shen Lijuan, Wang Yudong, Zhai Qingqing, et al. Degradation modeling using stochastic processes with random initial degradation[J]. IEEE Transactions on Reliability, 2019, 68(4): 1320-1329.

[5] Lin Yanhui, Ding Zeqi. An integrated degradation modeling framework considering model uncertainty and calibration[J]. Mechanical Systems and Signal Processing, 2022, 166: 108389.

[6] Tang Shufeng, Xie Xuesong, Zhang Xiaoling. Reliability prediction of semiconductor devices based on GRU-LSTM neural network[J]. Proceedings of SPIE, 2024, 13066: 6.

[7] Li Jiayi, Chang Yuanjiang, Xu Liangbin, et al. A data-driven LSTM-based management and control approach for fatigue life of subsea wellhead system[J]. Ocean Engineering, 2024, 313: 119335.

[8] Liang Tiantian, Wang Runze, Zhang Xuxiu, et al. Predicting reliability and remaining useful life of rolling bearings based on optimized neural networks[J]. Structural Durability & Health Monitoring, 2023, 17(5): 433-455.

[9] Nguyen K T P, Medjaher K, Gogu C. Probabilistic deep learning methodology for uncertainty quantification of remaining useful lifetime of multi-component systems[J]. Reliability Engineering & System Safety, 2022, 222: 108383.

[10] Chen Xiaowu, Liu Zhen. A long short-term memory neural network based Wiener process model for remaining useful life prediction[J]. Reliability Engineering & System Safety, 2022, 226: 108651.

[11] Bae J, Xi Zhimin. Learning of physical health timestep using the LSTM network for remaining useful life estimation[J]. Reliability Engineering & System Safety, 2022, 226: 108717.

[12] Shao Xiaoyan, Cai Baoping, Gao Lei, et al. Data-model-linked remaining useful life prediction method with small sample data: a case of subsea valve[J]. Reliability Engineering & System Safety, 2024, 250: 110323.

[13] Xiao Dengyu, Qin Chengjin, Ge Jianwen, et al. Self-attention-based adaptive remaining useful life prediction for IGBT with Monte Carlo dropout[J]. Knowledge-Based Systems, 2022, 239: 107902.

[14] Liu Yao, Wang Yashun, Fan Zhengwei, et al. Reliability modeling and a statistical inference method of accelerated degradation testing with multiple stresses and dependent competing failure processes[J]. Reliability Engineering & System Safety, 2021, 213: 107648.

[15] 段俊峰, 李宁, 唐求, 等. 基于 CK-GPR 的多应力环境智能电表剩余寿命预测[J]. 仪器仪表学报, 2021, 42(4): 102-110.

[16] Basora L, Viens A, Chao M A, et al. A benchmark on uncertainty quantification for deep learning prognostics[J]. Reliability Engineering & System Safety, 2025, 253: 110513.

[17] Lee J, Mitici M. Deep reinforcement learning for predictive aircraft maintenance using probabilistic remaining-useful-life prognostics[J]. Reliability Engineering & System Safety, 2023, 230: 108908.

[18] Liu Yitong, Wen Jiarui, Wang Guoqiang. A comprehensive overview of remaining useful life prediction: from traditional literature review to scientometric analysis[J]. Machine Learning with Applications, 2025, 21: 100704.

[19] Yu Wennian, Tu Wenbing, Kim I Y, et al. A nonlinear-drift-driven Wiener process model for remaining useful life estimation considering three sources of variability[J]. Reliability Engineering & System Safety, 2021, 212: 107631.

[20] Ferreira C, Gonçalves G. Remaining useful life prediction and challenges: a literature review on the use of machine learning methods[J]. Journal of Manufacturing Systems, 2022, 63: 550-562.

[21] Li Huiqin, Zhang Zhengxin, Li Tianmei, et al. A review on physics-informed data-driven remaining useful life prediction: challenges and opportunities[J]. Mechanical Systems and Signal Processing, 2024, 209: 111120.

[22] Lin Yanhui, Yan Pengcheng, Zio E. Recent advances in uncertainty analysis for prognostics and remaining useful life prediction: a review[J]. Reliability Engineering & System Safety, 2026, 269: 112110.

[23] Efron B. Bootstrap methods: another look at the jackknife[J]. The Annals of Statistics, 1979, 7(1): 1-26.

[24] Rubin D B. The Bayesian bootstrap[J]. The Annals of Statistics, 1981, 9(1): 130-134.

基金资助

中央高校基本科研业务费项目(3122026018)