(A) study on the application of data driven modeling for nuclear system analysis code constitutive equation

Abstract

In nuclear engineering, 1-D thermal hydraulic codes are used for the analysis of large scale systems that cannot be analyzed through experiments in real scale. Most existing thermal hydraulic codes consist of governing equations and constitutive equations. The results of numerical analysis are obtained by approximating the mass, momentum, and energy conservation equations in one dimensional form based on a two-fluid flow model. In the process of approximating the conservation equations, the constitutive equations must be included to calculate the various coefficient values such as interfacial heat transfer coefficient, which greatly affects the accuracy of the code calculation results. However, despite numerous thermal hydraulic experiments and studies were undertaken in the last 40-50 years, code uncertainties and errors still exist in simulating reactor transients. This is because two phase flow phenomenon has complexity, and its model consists of only limited mathematical function forms. In addition, the model related to the interfacial term in the constitutive equation has a larger error because it is difficult to measure directly in the experiment compared to the near-wall terms. Furthermore, phasic interfacial terms also have additional variables for numerical stabilization. In this study, MARS-KS code results were compared by substituting original MARS-KS’s interfacial heat transfer model with feedforward network within a specific thermal hydraulic range to see the possibility of replacing the original constitutive equations to feedforward networks. In addition, experimental prediction accuracy of MARS-KS was evaluated by perturbing replaced feedforward network, and it is confirmed that how differently the original model and improved feedforward network predict interfacial heat transfer coefficient in a given thermal hydraulic range.