### 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, for example, the means, variances, distribution types, etc. However, in best-estimate system thermal-hydraulics codes such as TRACE, some parameters in empirical correlations may have large uncertainties which are unknown to code users. So, the uncertainties associated these parameters are simply ignored or described by “expert opinion”. Inverse Uncertainty Quantification (UQ) is performed in the current study to replace such ad-hoc expert judgment. Inverse UQ is the process of quantifying the uncertainties in input parameters given relevant experimental measurements. The purpose of inverse UQ, and this paper, is to seek statistical descriptions of the input model parameters that are consistent with the observed data. Bayesian analysis is used to formulate the inverse UQ problem given relevant experiment data. In this study, the steady-state PSBT benchmark void fraction data is used. Within the Bayesian framework we seek the posterior distributions of the uncertain TRACE modeling parameters, which is updated from our prior knowledge given measurement data. Markov Chain Monte Carlo (MCMC) method is used to explore the posterior distributions, and surrogate models of TRACE are used to alleviate the computational burden. Gaussian Process (GP) is used to construct the surrogate model which can reduce the simulation time significantly. The outcomes will be the posterior distributions of several modeling parameters that are significant to PSBT experiment. Results of inverse UQ can be used for future forward uncertainty propagation and validation analysis, which will be presented in a companion paper.

Original language | English (US) |
---|---|

State | Published - Jan 1 2016 |

Event | 17th International Topical Meeting on Nuclear Reactor Thermal Hydraulics, NURETH 2017 - Xi'an, Shaanxi, China Duration: Sep 3 2017 → Sep 8 2017 |

### Other

Other | 17th International Topical Meeting on Nuclear Reactor Thermal Hydraulics, NURETH 2017 |
---|---|

Country | China |

City | Xi'an, Shaanxi |

Period | 9/3/17 → 9/8/17 |

### Fingerprint

### Keywords

- Gaussian Process
- Inverse Uncertainty Quantification
- Markov Chain Monte Carlo
- Physical Model Uncertainty

### ASJC Scopus subject areas

- Nuclear Energy and Engineering
- Instrumentation

### Cite this

*Surrogate-based inverse uncertainty quantification of TRACE physical model parameters using steady-state PSBT void fraction data*. Paper presented at 17th International Topical Meeting on Nuclear Reactor Thermal Hydraulics, NURETH 2017, Xi'an, Shaanxi, China.

**Surrogate-based inverse uncertainty quantification of TRACE physical model parameters using steady-state PSBT void fraction data.** / Wang, Chen; Wu, Xu; Kozlowski, Tomasz.

Research output: Contribution to conference › Paper

}

TY - CONF

T1 - Surrogate-based inverse uncertainty quantification of TRACE physical model parameters using steady-state PSBT void fraction data

AU - Wang, Chen

AU - Wu, Xu

AU - Kozlowski, Tomasz

PY - 2016/1/1

Y1 - 2016/1/1

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, for example, the means, variances, distribution types, etc. However, in best-estimate system thermal-hydraulics codes such as TRACE, some parameters in empirical correlations may have large uncertainties which are unknown to code users. So, the uncertainties associated these parameters are simply ignored or described by “expert opinion”. Inverse Uncertainty Quantification (UQ) is performed in the current study to replace such ad-hoc expert judgment. Inverse UQ is the process of quantifying the uncertainties in input parameters given relevant experimental measurements. The purpose of inverse UQ, and this paper, is to seek statistical descriptions of the input model parameters that are consistent with the observed data. Bayesian analysis is used to formulate the inverse UQ problem given relevant experiment data. In this study, the steady-state PSBT benchmark void fraction data is used. Within the Bayesian framework we seek the posterior distributions of the uncertain TRACE modeling parameters, which is updated from our prior knowledge given measurement data. Markov Chain Monte Carlo (MCMC) method is used to explore the posterior distributions, and surrogate models of TRACE are used to alleviate the computational burden. Gaussian Process (GP) is used to construct the surrogate model which can reduce the simulation time significantly. The outcomes will be the posterior distributions of several modeling parameters that are significant to PSBT experiment. Results of inverse UQ can be used for future forward uncertainty propagation and validation analysis, which will be presented in a companion paper.

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, for example, the means, variances, distribution types, etc. However, in best-estimate system thermal-hydraulics codes such as TRACE, some parameters in empirical correlations may have large uncertainties which are unknown to code users. So, the uncertainties associated these parameters are simply ignored or described by “expert opinion”. Inverse Uncertainty Quantification (UQ) is performed in the current study to replace such ad-hoc expert judgment. Inverse UQ is the process of quantifying the uncertainties in input parameters given relevant experimental measurements. The purpose of inverse UQ, and this paper, is to seek statistical descriptions of the input model parameters that are consistent with the observed data. Bayesian analysis is used to formulate the inverse UQ problem given relevant experiment data. In this study, the steady-state PSBT benchmark void fraction data is used. Within the Bayesian framework we seek the posterior distributions of the uncertain TRACE modeling parameters, which is updated from our prior knowledge given measurement data. Markov Chain Monte Carlo (MCMC) method is used to explore the posterior distributions, and surrogate models of TRACE are used to alleviate the computational burden. Gaussian Process (GP) is used to construct the surrogate model which can reduce the simulation time significantly. The outcomes will be the posterior distributions of several modeling parameters that are significant to PSBT experiment. Results of inverse UQ can be used for future forward uncertainty propagation and validation analysis, which will be presented in a companion paper.

KW - Gaussian Process

KW - Inverse Uncertainty Quantification

KW - Markov Chain Monte Carlo

KW - Physical Model Uncertainty

UR - http://www.scopus.com/inward/record.url?scp=85051987392&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85051987392&partnerID=8YFLogxK

M3 - Paper

AN - SCOPUS:85051987392

ER -