Exploring Acquisition-Pathway Analysis of a Generic Molten-Salt Fast Reactor using Multiphysics Informed Diversion Signatures

Molten Salt Reactors represent a fundamentally unique class of advanced reactor that use a liquid fuel form, making nuclear material accounting and control of these systems a new challenge for international safeguards. To mitigate special fissionable material diversion and inform future nuclear material accounting and control needs for safeguards considerations, a comprehensive understanding of possible diversion indicators and necessary measurements is required. One method to provide a first principles informed approach uses multiphysics modeling of an integral molten-salt reactor system, including neutronics, thermal-hydraulics, and thermochemistry. Pairing this deterministic multiphysics framework along with stochastic methods allows for the generation of large multiphysics data sets that characterize nominal reactor behavior into probabilistic envelopes for specific indicators of diversion. Next, machine learning (ML) algorithms are applied to these large datasets of isotopics and gamma spectra across various measurement locations to determine the detectability of extraction one significant quantity of plutonium and uranium-235 for multiple diversion scenarios. These diversion scenarios include slow-drip diversion, gaseous plutonium extraction, and solid uranium plating within the reactor system or associated fuel salt processing system. Statistical Analyses of these datasets identified differentiable isotopics between nominal and diversion scenarios when combined with isotopic characteristics, such as half-life, thus reducing future ML data requirements. ML analysis demonstrated near-perfect (99.9%) accuracy in discerning nominal from diversion operations using the full operational lifetime isotopic composition, with typical diversion identification within 1 year. Spectral data posed a more challenging detection problem, motivating future work to integrate sensitivity and statistical analysis insights to enhance ML detection for spectral data and to identify implications uncertainty propagations of nuclear data.