Data-fusion-based variable-fidelity reduced-order model for accurate thermal–hydraulic behavior prediction in a 6 × 6 rod bundle

Abstract

Achieving fast and low-cost computation of the thermal–hydraulic flow field inside a nuclear reactor is of great significance for understanding the reactor’s thermal–hydraulic characteristics and ensuring its operational safety. In this study, a variable-fidelity reduced-order model (ROM) is proposed, which integrates data from both low-fidelity and high-fidelity simulations. The low-fidelity model—characterized by a coarse mesh, standard k-ε turbulence model, Semi-Implicit Method for Pressure Linked Equations (SIMPLE) algorithm, and first-order upwind scheme—is employed to capture the global trend of the flow field of a 6 × 6 rod bundle. Due to its low computational cost, a large number of low-fidelity samples can be generated efficiently. In contrast, the high-fidelity model—using a refined mesh, SST k-ω turbulence model, coupled solver, and second-order upwind scheme—is utilized to correct the flow fields obtained from the low-fidelity model. However, the high computational cost limits the number of high-fidelity samples. To bridge the fidelity gap, a bridge function is constructed to fuse the data from both fidelities. Proper Orthogonal Decomposition (POD) is applied to extract POD modes and corresponding POD coefficients from both low-fidelity and high-fidelity snapshots. The difference between the POD coefficients of the two fidelities is modeled as a function of the input design variables using a Radial Basis Function (RBF) surrogate. For any new design variables, the differences in POD coefficients can be rapidly predicted and added to the corresponding low-fidelity POD coefficients. These corrected POD coefficients are then combined with the high-fidelity POD modes to predict the flow fields. Compared to conventional ROMs, the proposed data-fusion-based variable-fidelity model offers improved computational efficiency and prediction accuracy. The results of this study contribute to the efficient and accurate thermal–hydraulic analysis and optimization design of nuclear reactors.