Particle Filter Approach for Prognostics Using Exact Static Parameter Estimation and Consistent Prediction

##plugins.themes.bootstrap3.article.main##

##plugins.themes.bootstrap3.article.sidebar##

Published Jun 27, 2024
Kai Hencken Arthur Serres Giacomo Garegnani

Abstract

Particle filters are widely used in model-based prognostics. They estimate the future health state of an asset based on measurement data and an assumed degradation dynamics. Filters are in general applied to estimate only the states given a known dynamics of the process. In model-based prognostics, the dynamics is assumed to be known in an analytical form, but the parameters vary per device and need to be learned from the measurements as well. This is especially important for the calculation of the remaining useful life (RUL), as the prediction of the future evolution is needed.

There are commonly used approaches for this: Augmenting the state space with the parameter, together with assuming them to stay constant or adding an artificial diffusive evolution to them. The Liu–West filter improves on this by modifying the artificial evolution such that mean and standard deviation of the marginal parameter distribution are kept the same. Both approaches require to choose some tuning parameters, which might be difficult in practical applications. In addition, the model parameter is often assumed frozen for the prediction part, leading to an inconsistency. We propose how a modification of the parameter evolution in case of missing measurements can solve this in both cases.

More recently algorithms for combined state estimation and exact parameter estimation have been introduced, especially the Storvik filter, based on the usage of a sufficient statistic. We analyze how this can be applied to overcome difficulties with existing approaches, avoiding the need for tuning parameters. We also extend the Storvik filter in order to deal with time-steps with missing measurements. Two formally equivalent approaches are presented. These are applicable in all cases of missing measurements, coming either from irregular data acquisition, e.g. only during maintenance or inspection, or as part of the prediction step of the RUL calculation.

We study the different methods for two simple models in order to demonstrate potential issues with existing approaches and to explore the stability of the new one based on the Storvik filter. Finally we apply it to a practical application in the area of electrical distribution systems.

How to Cite

Hencken, K., Serres, A., & Garegnani, G. (2024). Particle Filter Approach for Prognostics Using Exact Static Parameter Estimation and Consistent Prediction. PHM Society European Conference, 8(1), 10. Retrieved from https://papers.phmsociety.org/index.php/phme/article/view/4009
Abstract 171 | PDF Downloads 102

##plugins.themes.bootstrap3.article.details##

Keywords

Prognostics; Particle Filter; Parameter Estimation; Liu-West filter; Storvik filter

References
An, D., Choi, J.-H., & Kim, N. H. (2013, July). Prognostics 101: A tutorial for particle filter-based prognostics algorithm using Matlab. Reliability Engineering & System Safety, 115, 161–169. doi: 10.1016/j.ress.2013.02.019 Barankin, E. W., & Maitra, A. P. (1963). Generalization of the Fisher-Darmois-Koopman-Pitman theorem on sufficient statistics. Sankhy¯a: The Indian Journal of Statistics, Series A, 217–244. Chopin, N., & Papaspiliopoulos, O. (2020). An Introduction to Sequential Monte Carlo. Cham: Springer International Publishing. doi: 10.1007/978-3-030-47845-2 Doucet, A., de Freitas, N., & Gordon, N. (2001). An introduction to Sequential Monte Carlo methods. In A. Doucet, N. de Freitas, & N. Gordon (Eds.), Sequential monte carlo methods in practice (pp. 3–14). New York, NY: Springer New York. Doucet, A., Godsill, S., & Andrieu, C. (2000). On Sequential Monte Carlo sampling methods for Bayesian filtering. Statistics and Computing, 10, 197 – 208. Erol, Y., Li, L., Ramsundar, B., & Russell, S. J. (2013). The extended parameter filter. Galar, D., Goebel, K., Sandborn, P., & Kumar, U. (2021). Prognostics and remaining useful life (rul) estimation: Predicting with confidence. CRC Press. Goebel, K., Daigle, M., Saxena, A., Sankararaman, S., Roychoudhury, I., & Celaya, J. (2017). Prognostics: The science of making predictions. Createspace Independent Pub. Hencken, K. (2021). Parameter estimation of a Wiener process of mechanical degradation through censored measurement of timings. Proceedings of the 31st European Safety and Reliability Conference, 317-324. Hu, Y., Baraldi, P., Di Maio, F., & Zio, E. (2015). A particle filtering and kernel smoothing-based approach for new design component prognostics. Reliability Engineering & System Safety, 134, 19-31. doi: https://doi.org/10.1016/j.ress.2014.10.003 Johannes, M. S., & Polson, N. G. (2006). Exact particle filtering and parameter learning.
Jouin, M., Gouriveau, R., Hissel, D., P´era, M.-C., & Zerhouni, N. (2016). Particle filter-based prognostics: Review, discussion and perspectives. Mechanical Systems and Signal Processing, 72, 2–31.

Joyce, P., & Marjoram, P. (2008). Approximately sufficient statistics and bayesian computation. Statistical applications in genetics and molecular biology, 7(1).

Liu, J., & West, M. (2001). Combined parameter and state estimation in simulation-based filtering. In A. Doucet,

N. de Freitas, & N. Gordon (Eds.), Sequential monte carlo methods in practice (pp. 197–223). New York, NY: Springer New York. Peng, W., Ye, Z.-S., & Chen, N. (2018). Joint online rul prediction for multivariate deteriorating systems. IEEE Transactions on Industrial Informatics, 15(5), 28702878.
Section
Technical Papers