Scholarlink

We perform multilevel Bayesian calibration of a model for fission gas diffusivity in \({\textrm{UO}_{2}}\) nuclear fuel. Specifically, we use a two-level delayed acceptance method that couples a machine learning surrogate for Xe and U diffusivities with a high-fidelity cluster dynamics model, to improve upon surrogate-assisted, single-level Bayesian calibration that does not account for residual error in the fitted surrogate. We integrate multilevel delayed acceptance with parallel prefetching techniques to improve the computational throughput of the calibration process. Our results indicate an average speed-up of more than 48 \(\times \) in terms of the computation time per effective sample.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Bayesian Calibration of Fission Gas Diffusivity in Nuclear Fuels Using Multilevel Delayed Acceptance MCMC

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Pieterjan Robbe,
  • Tiernan A. Casey,
  • Michael W. D. Cooper,
  • Christopher Matthews,
  • Khachik Sargsyan,
  • David A. Andersson,
  • Habib N. Najm

摘要

We perform multilevel Bayesian calibration of a model for fission gas diffusivity in \({\textrm{UO}_{2}}\) nuclear fuel. Specifically, we use a two-level delayed acceptance method that couples a machine learning surrogate for Xe and U diffusivities with a high-fidelity cluster dynamics model, to improve upon surrogate-assisted, single-level Bayesian calibration that does not account for residual error in the fitted surrogate. We integrate multilevel delayed acceptance with parallel prefetching techniques to improve the computational throughput of the calibration process. Our results indicate an average speed-up of more than 48 \(\times \) in terms of the computation time per effective sample.