Abstracts

Reactor agnostic multi-group cross section generationfor fine-mesh deterministic neutron transportsimulations; Reactor agnostic MGXC generation for fine-meshdeterministic neutron transport simulations

by William Robert Boyd




Institution: MIT
Department:
Year: 2017
Keywords: Nuclear Science and Engineering.
Posted: 02/01/2018
Record ID: 2191267
Full text PDF: http://hdl.handle.net/1721.1/112525


Abstract

A key challenge for full-core transport methods isreactor agnostic multi-group cross section (MGXS) generation. MonteCarlo (MC) presents the most accurate method for MGXS generationsince it does not require any approximations to the neutron flux.This thesis develops novel methods that use MC to generate thefine-spatial mesh MGXS that are needed by high-fidelity transportcodes. These methods employ either engineering-based or statisticalclustering algorithms to accelerate the convergence of MGXS talliedon fine, heterogeneous spatial meshes by Monte Carlo. Thetraditional multi-level approach to MGXS generation is replaced byfull-core MC calculations that generate MGXS for multi-groupdeterministic transport codes. Two pinwise spatial homogenizationschemes are introduced to model the clustering of pin-wise MGXS dueto spatial self-shielding spectral effects. The Local NeighborSymmetry (LNS) scheme uses a nearest neighbor-like analysis of areactor geometry to determine which fuel pins should be assignedthe same MGXS. The inferential MGXS (iMGXS) scheme appliesunsupervised machine learning algorithms to "noisy" MC tally datato identify clustering of pin-wise MGXS without any knowledge ofthe reactor geometry. Both schemes simultaneously account forspatial self-shielding effects while also accelerating theconvergence of the MC tallies used to generate MGXS. The LNS andiMGXS schemes were used to model MGXS clustering from radialgeometric heterogeneities in a suite of 2D PWR benchmarks. Bothschemes reduced U-238 capture rate errors by up to a factor of fourwith respect to schemes which neglect to model MGXS clustering. Inaddition, the schemes required an order of magnitude fewer MCparticle histories to converge MGXS for multi-group deterministiccalculations than a reference MC calculation. These resultsdemonstrate the potential for single-step MC simulations of thecomplete heterogeneous geometry as a means to generate reactoragnostic MGXS for deterministic transport codes. The LNS and iMGXSschemes may be valuable for reactor physics analyses of advancedLWR core designs and next generation reactors with spatialheterogeneities that are poorly modeled by the engineeringapproximations in today's methods for MGXSgeneration.Advisors/Committee Members: Kord Smith, Benoit Forget (advisor).