Monte Carlo Variance Reduction in MOOSE Stochastic Tools Module: Accelerating the Failure Analysis of Nuclear Reactor Technologies

Modeling and simulation using the MOOSE open-source software has been considerably benefiting the design and safety assessment of nuclear reactor technologies. Given the myriad uncertainties in the modeling process, not only does safety needs to be characterized in a probabilistic fashion, but also, probabilistic failure analysis needs to be fast and capable of considering high-fidelity models. This paper presents some recent developments to the MOOSE's stochastic tools module for accelerating the probabilistic failure analysis with adaptive sampling-based variance reduction methods. These methods guarantee higher confidence in the probabilistic failure assessments compared to a Monte Carlo for the same or lesser number of model evaluations. In adaptive sampling, a new sample is proposed, the model is evaluated, a decision is made of whether or not to accept the proposed sample, and this decision-information is further used for proposing the next sample. We describe the software objects in MOOSE such as Sampler, MultiApp, and Reporter that support adaptive sampling-based variance reduction. We further describe the MOOSE input format for using adaptive sampling. Finally, we demonstrate these adaptive sampling capabilities in MOOSE using two applications: high temperature creep response of a nuclear alloy and probabilistic failure analysis of the advanced nuclear fuel TRISO.