Nuclear data uncertainty reduction for advanced reactor modeling and simulation through Bayesian data assimilation
Open Access
- Author:
- Vidal Soares, Andre
- Graduate Program:
- Nuclear Engineering
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- June 10, 2024
- Committee Members:
- Dipanjan Pan, Professor in Charge/Director of Graduate Studies
Amanda Johnsen, Chair & Dissertation Advisor
Arthur Motta, Major Field Member
Elia Merzari, Major Field Member
William Walters, Special Member
Mauricio Nascimento, Outside Unit & Field Member - Keywords:
- molten salt reactor; nuclear data uncertainty; data assimilation; Bayesian Monte Carlo; nuclear material control and accountability; safeguards
- Abstract:
- Modeling and Simulation (M&S) are crucial for advancing nuclear technology and understanding nuclear systems throughout their lifespan. This research concentrates on molten salt reactors (MSRs) and the distinct challenges of managing and accounting for nuclear materials in liquid-fueled MSR designs. High-precision simulation tools are being created to incorporate online reprocessing mechanisms for more accurate nuclide inventory calculations. However, uncertainties in nuclear data and input data are fundamental aspects of M&S and pose significant challenges. This research examines the impact of using data assimilation through Bayesian statistics to reduce uncertainties in integral quantities for liquid-fuel thermal-spectrum MSR models based on the Molten Salt Demonstration Reactor (MSDR) and the Molten Salt Reactor Experiment (MSRE). Since actual measurement data are unavailable for the MSDR, hypothetical experimental data are generated. Notable reductions in the nuclear data uncertainty of nuclide concentration for specific isotopes were observed. For instance, a reduction of up to 30% in the nuclear data uncertainty of Pu-239 concentrations at the end of the depletion cycle was found. For the MSRE, existing data from operation reports were used to assimilate results from a simplified input model. Despite its simplicity, the model included a detailed depletion history for the reactor's power ramp in 1966. Results indicated that nuclear data uncertainties propagated to nuclide concentrations were insufficient to explain the large discrepancies between calculated and experimental results. Noble metal nuclides showed non-physical results, with differences around 100%. Simulation uncertainties (from input data parameters) were evaluated through variations in temperature, moderator density, power levels, and removal rates. The data assimilation results demonstrated that Bayesian statistics could significantly enhance the accuracy of nuclide concentration predictions. Agreement between posterior nuclide concentrations and experimental data improved (e.g., prior agreement of fission product Sr-92 improved from 28.57% to 10.35% in one of the study cases). This methodology has promising implications for the development of high-fidelity depletion calculations for advanced nuclear reactor technologies, particularly in reducing input data-induced uncertainty. The validation will help advance computational modeling tools for other MSR designs.