Reduced order models for thermal hydraulic transient analysis in nuclear engineering

Abstract

Reduced order modeling emerges as a valuable tool for addressing multi-query challenges in high-fidelity computational simulations, particularly in areas such as design optimization and uncertainty analysis. Given that computational costs pose a primary obstacle to leveraging multi-physics multi-scale integrated systems, achieving computational reduction without compromising calculation accuracy through reduced order modeling offers significant advantages. This study explores two hybrid approaches to reduced order modeling, specifically applied to nuclear thermal hydraulic accident analysis. The first approach involves constructing a reduced order basis through random sampling, followed by the development of surrogate models based on linear regression, second order polynomial regression, and artificial neural network to predict time dependent thermal hydraulic behaviors. The second approach utilizes the reduced basis to transform a solution module within a thermal hydraulic analysis code, assessing the feasibility of the intrusive reduced order modeling approach. A comparison between reduced order model predictions and full order model calculations yields promising results, elucidating applicable conditions and limitations of the proposed approaches.