logo

SCIENCE CHINA Information Sciences, Volume 64 , Issue 1 : 112205(2021) https://doi.org/10.1007/s11432-019-2724-5

Online remaining-useful-life estimation with a Bayesian-updated expectation-conditional-maximization algorithm and a modified Bayesian-model-averaging method

More info
  • ReceivedAug 29, 2019
  • AcceptedNov 1, 2019
  • PublishedDec 18, 2020

Abstract

Online remaining-useful-life (RUL) estimation is an effective method with respect to ensuring the safety of complex-huge systems. Generally, current methods assume a specific degradation model when degradation values are observed in the initial degradation phase. However, this assumption may not always be robust enough owing to the often-ambiguous inherent incipient-degradation characteristic. Therefore, besides model-parameter uncertainty, the uncertainty of the degradation model is worth examining in online RUL estimations. In this paper, a Bayesian-updated expectation-conditional-maximization (ECM) algorithm is adopted to address the uncertainty of prior parameters, and a modified Bayesian-model-averaging method is used to deal with the uncertainty of the degradation model. Then, simulation studies are conducted to analyze the effectiveness of the proposed fusion algorithm. Results suggest that the Bayesian-updated ECM algorithm and modified Bayesian-model-averaging method effectively address the associated uncertainties of model parameters and the degradation model itself. Finally, we apply the proposed fusion algorithm to predict the RUL of a gyroscope.


Acknowledgment

This work was supported by National Key RD Program of China (Grant No. 2018YFB1306100) and National Natural Science Foundation of China (Grant Nos. 61922089, 61773386, 61833016, 61903376, 61673311).


References

[1] Zhang J X, Hu C H, He X. A Novel Lifetime Estimation Method for Two-Phase Degrading Systems. IEEE Trans Rel, 2019, 68: 689-709 CrossRef Google Scholar

[2] Si X, Li T, Zhang Q. A General Stochastic Degradation Modeling Approach for Prognostics of Degrading Systems With Surviving and Uncertain Measurements. IEEE Trans Rel, 2019, 68: 1080-1100 CrossRef Google Scholar

[3] Si X, Li T, Zhang Q. Prognostics for Linear Stochastic Degrading Systems With Survival Measurements. IEEE Trans Ind Electron, 2020, 67: 3202-3215 CrossRef Google Scholar

[4] Si X-S, Hu C H, Li T M, et al. A joint order-replacement policy for deteriorating components with reliability constraint. Sci China Inf Sci, doi: 10.1007/s11432-018-9804-9. Google Scholar

[5] Xi J, Wang C, Liu H. Completely Distributed Guaranteed-Performance Consensualization for High-Order Multiagent Systems With Switching Topologies. IEEE Trans Syst Man Cybern Syst, 2019, 49: 1338-1348 CrossRef Google Scholar

[6] Si X, Li T, Zhang Q. An Optimal Condition-Based Replacement Method for Systems With Observed Degradation Signals. IEEE Trans Rel, 2018, 67: 1281-1293 CrossRef Google Scholar

[7] Hu C H, Pei H, Si X S. A Prognostic Model Based on DBN and Diffusion Process for Degrading Bearing. IEEE Trans Ind Electron, 2019, : 1-1 CrossRef Google Scholar

[8] Zhang Z, Si X, Hu C. Degradation data analysis and remaining useful life estimation: A review on Wiener-process-based methods. Eur J Operational Res, 2018, 271: 775-796 CrossRef Google Scholar

[9] Hudak S J J, Saxena A, Bucci R J, et al. Development of Standard Methods of Testing and Analyzing Fatigue Crack Growth Rate Data. Westinghouse R & D Center Westinghouse Electric Corporation, 1978. Google Scholar

[10] Meeker W Q, Escobar L A. Statistical Methods for Reliability Data. New York: John Wiley & Sons, Inc., 1998. Google Scholar

[11] Ye Z S, Xie M, Tang L C. Semiparametric Estimation of Gamma Processes for Deteriorating Products. Technometrics, 2014, 56: 504-513 CrossRef Google Scholar

[12] Li Z, Deng Y, Mastrangelo C. Model selection for degradation-based Bayesian reliability analysis. J Manufacturing Syst, 2015, 37: 72-82 CrossRef Google Scholar

[13] Chien-Yu Peng , Sheng-Tsaing Tseng . Mis-Specification Analysis of Linear Degradation Models. IEEE Trans Rel, 2009, 58: 444-455 CrossRef Google Scholar

[14] Chien-Yu Peng , Sheng-Tsaing Tseng . Statistical Lifetime Inference With Skew-Wiener Linear Degradation Models. IEEE Trans Rel, 2013, 62: 338-350 CrossRef Google Scholar

[15] Asensio Ramos A. The Minimum Description Length Principle and Model Selection in Spectropolarimetry. Astrophys J, 2006, 646: 1445-1451 CrossRef Google Scholar

[16] Bozdogan H. Akaike's Information Criterion and Recent Developments in Information Complexity.. J Math Psychology, 2000, 44: 62-91 CrossRef PubMed Google Scholar

[17] Nguyen K T P, Fouladirad M, Grall A. Model selection for degradation modeling and prognosis with health monitoring data. Reliability Eng Syst Saf, 2018, 169: 105-116 CrossRef Google Scholar

[18] Ben Abdessalem A, Dervilis N, Wagg D. Model selection and parameter estimation in structural dynamics using approximate Bayesian computation. Mech Syst Signal Processing, 2018, 99: 306-325 CrossRef Google Scholar

[19] Si X S, Wang W, Chen M Y. A degradation path-dependent approach for remaining useful life estimation with an exact and closed-form solution. Eur J Operational Res, 2013, 226: 53-66 CrossRef Google Scholar

[20] Meng X L, Rubin D B. Maximum likelihood estimation via the ECM algorithm: A general framework. Biometrika, 1993, 80: 267-278 CrossRef Google Scholar

[21] Chen N, Tsui K L. Condition monitoring and remaining useful life prediction using degradation signals: revisited. IIE Trans, 2013, 45: 939-952 CrossRef Google Scholar

[22] Chakraborty S, Gebraeel N, Lawley M. Residual-life estimation for components with non-symmetric priors. IIE Trans, 2009, 41: 372-387 CrossRef Google Scholar

[23] Gebraeel N Z, Lawley M A, Li R. Residual-life distributions from component degradation signals: A Bayesian approach. IIE Trans, 2005, 37: 543-557 CrossRef Google Scholar

[24] Nitithumbundit T, Chan J S K. An ECM algorithm for skewed multivariate variance Gamma distribution in normal mean-variance representation. Statistics, 2015. Google Scholar

[25] van Dyk D A, Meng X, Rubin D B. Maximum likelihood estimation via the ECM algorithm: computing the asymptotic variance. Statistica Sinica, 1995,5(1):55-75 https://www.jstor.org/stable/24305553. Google Scholar

[26] Zhang X, Yan X. A new statistical precipitation downscaling method with Bayesian model averaging: a case study in China. Clim Dyn, 2015, 45: 2541-2555 CrossRef Google Scholar

[27] Volinsky C T, Raftery A E, Madigan D. David Draper and E. I. George, and a rejoinder by the authors. Statist Sci, 1999, 14: 382-417 CrossRef Google Scholar

[28] Gebraeel N, Elwany A, Jing Pan A. Residual Life Predictions in the Absence of Prior Degradation Knowledge. IEEE Trans Rel, 2009, 58: 106-117 CrossRef Google Scholar

[29] Si X S, Chen M Y, Wang W. Specifying measurement errors for required lifetime estimation performance. Eur J Operational Res, 2013, 231: 631-644 CrossRef Google Scholar

  • Figure 1

    The algorithm of RUL calculation for model ${M_c}$.

  • Figure 2

    (Color online) Results of Bayesian-updated ECM algorithm for model ${M_x}$. (a) Simulated trajectory with IID errors; (b) predicted RUL of model ${M_x}$; (c) estimated ${\mu~'_{x0}}$ of model ${M_x}$; (d) estimated ${\mu~'_{x1}}$ of model ${M_x}$.

  • Figure 3

    (Color online) Results of Bayesian-updated ECM algorithm for model ${M_y}$. (a) Simulated trajectory with Brownian errors; (b) predicted RUL of model ${M_y}$; (c) estimated ${\mu~'_{y0}}$ of model ${M_y}$; (d) estimated ${\mu~'_{y1}}$ of model ${M_y}$.

  • Figure 4

    (Color online) Effectiveness of modified Bayesian-model-averaging method for trajectory with IID errors. protectłinebreak (a) Modified Bayesian model averaging for trajectory with IID errors; (b) original Bayesian model averaging for trajectory with IID errors; (c) the obtained model probability of the modified Bayesian model averaging for model ${M_x}$; (d) the obtained model probability of the original Bayesian model averaging for model ${M_x}$.

  • Figure 5

    (Color online) Effectiveness of modified Bayesian model averaging for trajectory with Brownian errors. protectłinebreak (a) Modified Bayesian model averaging for trajectory with Brownian errors; (b) original Bayesian model averaging for trajectory Brownian errors; (c) obtained model probability of modified Bayesian model averaging for model ${M_y}$; (d) obtained model probability of original Bayesian model averaging for model ${M_y}$.

  • Figure 6

    Illustration of a deformed motor bearing.

  • Figure 7

    (Color online) Effectiveness of the proposed algorithm. (a) The observed drift data of the gyroscope; (b) estimated RUL of different prior parameters; (c) estimated RUL of different prior probabilities; (d) estimated RUL of our method.

  • Table 1  

    Table 1Distributions of the selected simulation parameters

    Simulation parameter Simulation distribution
    ${\theta'_x}$ $N(0.02,1~\times~{10^{~-~6}})$
    Exponential degradation with IID errors ${\beta'_x}$ $N(0.01,1~\times~{10^{~-~6}})$
    $\sigma_x^2$ 0.1
    ${\theta'_y}$ $N(0.02,1~\times~{10^{~-~6}})$
    Exponential degradation with Brownian errors ${\beta'_y}$ $N(0.01,1~\times~{10^{~-~6}})$
    $\sigma_y^2$ 0.0004
  • Table 2  

    Table 2Prior-distribution assumption for Bayesian-updated ECM algorithm

    Model Parameter Bayesian method [28] Our method
    Distribution 1 Distribution 2 Distribution 3
    ${\theta~'_x}$ $N(0.04,1~\times~{10^{~-~6}})$ $N(0.04,1~\times~{10^{~-~6}})$ $N(0.1,1~\times~{10^{~-~6}})$ $N(0.2,1~\times~{10^{~-~6}})$
    ${M_x}$ ${\beta~'_x}$ $N(0.02,1~\times~{10^{~-~6}})$ $N(0.02,1~\times~{10^{~-~6}})$ $N(0.05,1~\times~{10^{~-~6}})$ $N(0.1,1~\times~{10^{~-~6}})$
    $\sigma~_x^2$ 0.01 0.01 0.01 0.01
    ${\theta~'_y}$ $N(0.04,1~\times~{10^{~-~6}})$ $N(0.04,1~\times~{10^{~-~6}})$ $N(0.1,1~\times~{10^{~-~6}})$ $N(0.2,1~\times~{10^{~-~6}})$
    ${M_y}$ ${\beta~'_y}$ $N(0.02,1~\times~{10^{~-~6}})$ $N(0.02,1~\times~{10^{~-~6}})$ $N(0.05,1~\times~{10^{~-~6}})$ $N(0.1,1~\times~{10^{~-~6}})$
    $\sigma~_y^2$ 0.004 0.004 0.004 0.004
  • Table 3  

    Table 3Prior-distribution assumption for observed drift data

    Model中国 Parameter Distribution 1 Distribution 2 Distribution 3
    ${\theta~'_x}$ $N(0.0025,1~\times~{10^{~-~6}})$ $N(0.005,1~\times~{10^{~-~6}})$ $N(0.0075,1~\times~{10^{~-~6}})$
    ${M_x}$ ${\beta~'_x}$ $N(0.001,1~\times~{10^{~-~6}})$ $N(0.002,1~\times~{10^{~-~6}})$ $N(0.003,1~\times~{10^{~-~6}})$
    $\sigma~_x^2$ 0.01 0.01 0.01
    ${\theta~'_y}$ $N(0.0025,1~\times~{10^{~-~6}})$ $N(0.005,1~\times~{10^{~-~6}})$ $N(0.0075,1~\times~{10^{~-~6}})$
    ${M_y}$ ${\beta~'_y}$ $N(0.001,1~\times~{10^{~-~6}})$ $N(0.002,1~\times~{10^{~-~6}})$ $N(0.003,1~\times~{10^{~-~6}})$
    $\sigma~_y^2$ 0.00004 0.00004 0.00004