A tuning-free POD-FV-ROM with automatic boundary enforcement for practical thermal-hydraulic applications

Abstract

Reduced order models (ROMs) have been widely adopted to accelerate high-fidelity simulations while retaining essential predictive accuracy. However, two gaps persist for POD-FV-ROMs in the context of thermal-hydraulic problems. First, when Dirichlet and Neumann boundaries coexist, a robust and effective strategy to enforce boundary constraints in the ROM is still missing. Second, the ROM performance has not been convincingly demonstrated on practical three-dimensional cases with complex geometries. To close these gaps, a ROM combining POD-Galerkin projection with supremizer stabilization and POD-RBF interpolation was investigated, together with a penalty formulation for boundary treatment. Two novel a priori estimators are proposed to determine penalty factors (PFs) without manual tuning, one based on residuals relevant to projected snapshots and one derived from optimization to a domain-wise error. The entire ROM framework was assessed on an 2 × 2 helical cruciform fuel assembly discretized into around 24 million cells with parameterized boundary conditions. The results demonstrated that, under appropriate PFs, the ROM delivered satisfying accuracy while achieving a speed up of five orders of magnitude.