Dimensionality reducibility for multi-physics reduced order modeling

Highlights: •General algorithms for dimensionality reduction for single physics models. •General algorithms for dimensionality reduction for multi-physics models. •Development of surrogate models coupled with dimensionality reduction. •Development of reduced order models for RattleSnake-Bison code coupling. -- Abstract: Applications of reduced order modeling (ROM) to support analysis of complex reactor behavior using high fidelity simulations have developed rapidly in recent years. Reduction implies any computational approach aiming to reduce the cost of the simulation, especially for situations involving repeated executions such as probabilistic risk assessment and uncertainty quantification applications. This article presents a novel non-intrusive methodology to render reduction for multi-physics models by taking advantage of the combined reduction introduced by each sub-physics in the simulation. Next, a surrogate model is constructed in terms of the reduced dimensions. A key component of the proposed methodology is to upper-bound the errors resulting from the reduction to ensure its reliability for subsequent engineering applications. To implement and demonstrate the proposed ROM algorithm, the INL's MAMMOTH environment is employed to analyze the level of reduction in the coupled radiation-thermal transport modeling of a 2D quarter fuel pin in a light water reactor spectrum. MAMMOTH couples the neutronics model of Rattlesnake module and the fuel performance model of BISON module. Results show that the reduction obtained with coupled physics is more significant than that with individual sub-physics models.