About this Abstract |
Meeting |
2020 TMS Annual Meeting & Exhibition
|
Symposium
|
Computational Materials Science and Engineering of Materials in Nuclear Reactors
|
Presentation Title |
Application of Variational Bayesian Monte Carlo Method for Improved Prediction of Doped UO2 Fuel Performance |
Author(s) |
Yifeng Che, Koroush Shirvan |
On-Site Speaker (Planned) |
Yifeng Che |
Abstract Scope |
Chromia-doped UO2 fuel has been shown to effectively promote grain growth and suppress fission gas release (FGR) under transient operating conditions. The most recent FGR model for the Chromia-doped UO2 fuel improves the FGR prediction, while is still not perfect due to missing physics as well as the inherent uncertain nature of the FGR process. This work intends to further improve the BISON FGR model for Chromia-doped UO2 fuel in Bayesian framework using the available Halden experimental data. Dimensionality reduction is first performed using principal component analysis to deal with the time-dependent FGR series data. Kriging is then used as metamodel of the computationally expensive code BISON. A novel optimization framework, Variational Bayesian Monte Carlo (VBMC), is utilized to improve the predictability of the most recent BISON FGR model. The performance of VBMC and the traditional statistical Markov Chain Monte Carlo sampling (MCMC) is compared and discussed. |
Proceedings Inclusion? |
Planned: Supplemental Proceedings volume |