Inverse uncertainty quantification by hierarchical Bayesian inference for TRACE physical model parameters based on BFBT benchmark

Chen Wang, Xu Wu, Tomasz Kozlowski

Research output: Contribution to conferencePaper

Abstract

In the framework of BEPU (Best Estimate plus Uncertainty) methodology, the uncertainties involved in simulations must be quantified to prove that the investigated design is reasonable and acceptable. The uncertainties in predictions are usually calculated by propagating input uncertainties through the simulation model, which requires prior knowledge of the model or code input uncertainties. However, in thermal-hydraulics (TH) best estimate codes like TRACE, some parameters called physical model parameters in empirical correlations may have large uncertainties and their distributions are unknown to code users. Inverse Uncertainty Quantification (IUQ), also called Bayesian Calibration or Data Assimilation, aims at inversely quantifying the uncertainties associated with input parameters that are consistent with experimental data, and thus replacing the ad-hoc “expert judgment” or “user self-assessment”. Previous IUQ work in the field of nuclear engineering mainly use Markov Chain Monte Carlo (MCMC) algorithms to calculate the posterior distributions of certain parameters, however, they are mostly limited to relatively small datasets. The calculated posterior distributions are only valid for the selected experimental cases and may vary given different datasets. This paper introduces a Hierarchical Bayesian Inference approach in the IUQ framework to account for the variability of TRACE physical model parameters in different experimental conditions. The hierarchical model provides a more stable framework so that outliers will not have significant influence on the calculation results. The hierarchical model introduces more parameters in the sampling process so a gradient-based sampling methods called No-U-Turn Sampler (NUTS) is used for this high-dimensional MCMC sampling problem. This framework is applied to TRACE physical model parameters based on steady-state void fraction data in BFBT benchmark and the posterior distributions obtained will be meaningful for future forward uncertainty analysis.

Original languageEnglish (US)
Pages4795-4807
Number of pages13
StatePublished - Jan 1 2019
Event18th International Topical Meeting on Nuclear Reactor Thermal Hydraulics, NURETH 2019 - Portland, United States
Duration: Aug 18 2019Aug 23 2019

Conference

Conference18th International Topical Meeting on Nuclear Reactor Thermal Hydraulics, NURETH 2019
CountryUnited States
CityPortland
Period8/18/198/23/19

Fingerprint

inference
Markov chains
sampling
Sampling
Markov processes
samplers
assimilation
estimates
Uncertainty
Nuclear engineering
hydraulics
voids
Uncertainty analysis
Void fraction
simulation
engineering
methodology
gradients
Hydraulics
predictions

Keywords

  • BFBT benchamrk
  • Hierarchical Bayesian inference
  • Inverse uncertainty quantification
  • MCMC

ASJC Scopus subject areas

  • Nuclear Energy and Engineering
  • Instrumentation

Cite this

Wang, C., Wu, X., & Kozlowski, T. (2019). Inverse uncertainty quantification by hierarchical Bayesian inference for TRACE physical model parameters based on BFBT benchmark. 4795-4807. Paper presented at 18th International Topical Meeting on Nuclear Reactor Thermal Hydraulics, NURETH 2019, Portland, United States.

Inverse uncertainty quantification by hierarchical Bayesian inference for TRACE physical model parameters based on BFBT benchmark. / Wang, Chen; Wu, Xu; Kozlowski, Tomasz.

2019. 4795-4807 Paper presented at 18th International Topical Meeting on Nuclear Reactor Thermal Hydraulics, NURETH 2019, Portland, United States.

Research output: Contribution to conferencePaper

Wang, C, Wu, X & Kozlowski, T 2019, 'Inverse uncertainty quantification by hierarchical Bayesian inference for TRACE physical model parameters based on BFBT benchmark' Paper presented at 18th International Topical Meeting on Nuclear Reactor Thermal Hydraulics, NURETH 2019, Portland, United States, 8/18/19 - 8/23/19, pp. 4795-4807.
Wang C, Wu X, Kozlowski T. Inverse uncertainty quantification by hierarchical Bayesian inference for TRACE physical model parameters based on BFBT benchmark. 2019. Paper presented at 18th International Topical Meeting on Nuclear Reactor Thermal Hydraulics, NURETH 2019, Portland, United States.
Wang, Chen ; Wu, Xu ; Kozlowski, Tomasz. / Inverse uncertainty quantification by hierarchical Bayesian inference for TRACE physical model parameters based on BFBT benchmark. Paper presented at 18th International Topical Meeting on Nuclear Reactor Thermal Hydraulics, NURETH 2019, Portland, United States.13 p.
@conference{c43f3405481b4c6ca641709d49990677,
title = "Inverse uncertainty quantification by hierarchical Bayesian inference for TRACE physical model parameters based on BFBT benchmark",
abstract = "In the framework of BEPU (Best Estimate plus Uncertainty) methodology, the uncertainties involved in simulations must be quantified to prove that the investigated design is reasonable and acceptable. The uncertainties in predictions are usually calculated by propagating input uncertainties through the simulation model, which requires prior knowledge of the model or code input uncertainties. However, in thermal-hydraulics (TH) best estimate codes like TRACE, some parameters called physical model parameters in empirical correlations may have large uncertainties and their distributions are unknown to code users. Inverse Uncertainty Quantification (IUQ), also called Bayesian Calibration or Data Assimilation, aims at inversely quantifying the uncertainties associated with input parameters that are consistent with experimental data, and thus replacing the ad-hoc “expert judgment” or “user self-assessment”. Previous IUQ work in the field of nuclear engineering mainly use Markov Chain Monte Carlo (MCMC) algorithms to calculate the posterior distributions of certain parameters, however, they are mostly limited to relatively small datasets. The calculated posterior distributions are only valid for the selected experimental cases and may vary given different datasets. This paper introduces a Hierarchical Bayesian Inference approach in the IUQ framework to account for the variability of TRACE physical model parameters in different experimental conditions. The hierarchical model provides a more stable framework so that outliers will not have significant influence on the calculation results. The hierarchical model introduces more parameters in the sampling process so a gradient-based sampling methods called No-U-Turn Sampler (NUTS) is used for this high-dimensional MCMC sampling problem. This framework is applied to TRACE physical model parameters based on steady-state void fraction data in BFBT benchmark and the posterior distributions obtained will be meaningful for future forward uncertainty analysis.",
keywords = "BFBT benchamrk, Hierarchical Bayesian inference, Inverse uncertainty quantification, MCMC",
author = "Chen Wang and Xu Wu and Tomasz Kozlowski",
year = "2019",
month = "1",
day = "1",
language = "English (US)",
pages = "4795--4807",
note = "18th International Topical Meeting on Nuclear Reactor Thermal Hydraulics, NURETH 2019 ; Conference date: 18-08-2019 Through 23-08-2019",

}

TY - CONF

T1 - Inverse uncertainty quantification by hierarchical Bayesian inference for TRACE physical model parameters based on BFBT benchmark

AU - Wang, Chen

AU - Wu, Xu

AU - Kozlowski, Tomasz

PY - 2019/1/1

Y1 - 2019/1/1

N2 - In the framework of BEPU (Best Estimate plus Uncertainty) methodology, the uncertainties involved in simulations must be quantified to prove that the investigated design is reasonable and acceptable. The uncertainties in predictions are usually calculated by propagating input uncertainties through the simulation model, which requires prior knowledge of the model or code input uncertainties. However, in thermal-hydraulics (TH) best estimate codes like TRACE, some parameters called physical model parameters in empirical correlations may have large uncertainties and their distributions are unknown to code users. Inverse Uncertainty Quantification (IUQ), also called Bayesian Calibration or Data Assimilation, aims at inversely quantifying the uncertainties associated with input parameters that are consistent with experimental data, and thus replacing the ad-hoc “expert judgment” or “user self-assessment”. Previous IUQ work in the field of nuclear engineering mainly use Markov Chain Monte Carlo (MCMC) algorithms to calculate the posterior distributions of certain parameters, however, they are mostly limited to relatively small datasets. The calculated posterior distributions are only valid for the selected experimental cases and may vary given different datasets. This paper introduces a Hierarchical Bayesian Inference approach in the IUQ framework to account for the variability of TRACE physical model parameters in different experimental conditions. The hierarchical model provides a more stable framework so that outliers will not have significant influence on the calculation results. The hierarchical model introduces more parameters in the sampling process so a gradient-based sampling methods called No-U-Turn Sampler (NUTS) is used for this high-dimensional MCMC sampling problem. This framework is applied to TRACE physical model parameters based on steady-state void fraction data in BFBT benchmark and the posterior distributions obtained will be meaningful for future forward uncertainty analysis.

AB - In the framework of BEPU (Best Estimate plus Uncertainty) methodology, the uncertainties involved in simulations must be quantified to prove that the investigated design is reasonable and acceptable. The uncertainties in predictions are usually calculated by propagating input uncertainties through the simulation model, which requires prior knowledge of the model or code input uncertainties. However, in thermal-hydraulics (TH) best estimate codes like TRACE, some parameters called physical model parameters in empirical correlations may have large uncertainties and their distributions are unknown to code users. Inverse Uncertainty Quantification (IUQ), also called Bayesian Calibration or Data Assimilation, aims at inversely quantifying the uncertainties associated with input parameters that are consistent with experimental data, and thus replacing the ad-hoc “expert judgment” or “user self-assessment”. Previous IUQ work in the field of nuclear engineering mainly use Markov Chain Monte Carlo (MCMC) algorithms to calculate the posterior distributions of certain parameters, however, they are mostly limited to relatively small datasets. The calculated posterior distributions are only valid for the selected experimental cases and may vary given different datasets. This paper introduces a Hierarchical Bayesian Inference approach in the IUQ framework to account for the variability of TRACE physical model parameters in different experimental conditions. The hierarchical model provides a more stable framework so that outliers will not have significant influence on the calculation results. The hierarchical model introduces more parameters in the sampling process so a gradient-based sampling methods called No-U-Turn Sampler (NUTS) is used for this high-dimensional MCMC sampling problem. This framework is applied to TRACE physical model parameters based on steady-state void fraction data in BFBT benchmark and the posterior distributions obtained will be meaningful for future forward uncertainty analysis.

KW - BFBT benchamrk

KW - Hierarchical Bayesian inference

KW - Inverse uncertainty quantification

KW - MCMC

UR - http://www.scopus.com/inward/record.url?scp=85073749986&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85073749986&partnerID=8YFLogxK

M3 - Paper

AN - SCOPUS:85073749986

SP - 4795

EP - 4807

ER -