Virginia Tech
Development of RAPID-AI, Physics-Based AI-Driven Algorithms for Pin-wise Burnup Dependent Fission Neutron Distribution and Ex-core Reactor Monitoring
Abstract
dc:description.abstractThis dissertation develops advanced neutronics formulations and associated algorithms for nuclear reactor core physics and monitoring. Reactor monitoring calculations require determination of the changing fission neutron source and the response of detector systems in real time. Whole-core fuel burnup calculations require repeated neutron transport and material updates because irradiation changes the material composition and fission neutron source distribution. This dissertation extends burnup-dependent Real-time Analysis for Particle Transport and In-situ Detection (RAPID) from its previous research-reactor application to a full-scale, pin-wise pressurized water reactor calculation. RAPID is based on Multistage Response Function Transport (MRT) methodology and uses the combined fission matrix formulation. This hybrid deterministic and Monte Carlo approach uses precalculated combined fission matrix (CFM) coefficients that vary with burnup. RAPID-AI is developed by optimizing the RAPID calculation and integrating neural-network predictions of material composition and CFM coefficients. The optimized calculation reduces computing time while maintaining agreement with the original RAPID results. A material model predicts the next pin-wise composition and the fission-source quantities needed for the reactor calculation. A CFM coefficient model predicts neutron production between fuel regions from the current composition and source. The models are trained using Serpent Monte Carlo calculations and evaluated through comparisons of the predicted quantities, whole-core eigenvalue, and fission neutron distribution. The integrated burnup calculation (bRAPID) uses both models to calculate pin-wise power, advance burnup, and update the material composition through successive reactor states. The RAPID-AI solution required 8.5 s on one core, compared to 7.6 days for the Serpent calculation on 40 cores. The material prediction model took 3.4 s to accurately predict the fission neutron source, atomic densities, and nuclide dependent energy recovered and ν values for 10 assembly types, compared to 10 hours for the Serpent burnup calculations. Using the RAPID-AI CFM neural network took 45 s to generate the full set of coefficients compared to 40 hours to generate the coefficients using Serpent fixed-source calculations. For the Watts Bar test benchmark, RAPID-AI calculations with predicted and Serpent-generated CFM coefficients gave a relative eigenvalue difference of −105.6 pcm and a fission neutron distribution mean relative difference of −0.04%. For ex-core monitoring, a combined Current Response Function and Detector Response Function methodology is developed enabling neutron transport from the reactor to ex-core neutron detectors beyond the confinement building. The method combines precalculated responses for the reactor structures, shielding, and detector. The methodology is demonstrated using a fresh-core neutron source and a computational model of the miniCHANDLER multi-modal neutron and anti-neutrino detection system. The calculated neutron-capture response is in excellent agreement with the direct Serpent calculation within its reported statistical uncertainty. The integration of these calculation algorithms, the neural network models, and optimization of RAPID-AI is pioneering work for the development of real-time reactor simulations incorporating changing core conditions.
Degree
thesis:*- Name thesis:degree_name
- Doctor of Philosophy
- Level thesis:degree_level
- doctoral
- Discipline thesis:degree_discipline
- Nuclear Engineering
- Department dc:contributor.department
- Mechanical Engineering
- Grantor dc:publisher
- Virginia Tech
- Year dc:date.issued
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Stroh, Brian
- Chair dc:contributor.committeechair
-
- Haghighat, Alireza
- Committee members dc:contributor.committeemember
-
- Liu, Yang
- Snoj, Luka
- Sjoden, Glenn Eric
- Link, Jonathan Marion
Subjects
dc:subject × 5Rights
dc:rights- Statement dc:rights
-
- In Copyright
- Licence dc:rights.uri
- Language dc:language.iso
- en
Identifiers
dc:identifier.*- Dc Identifier Other
- vt_gsexam:47866