Progress on coarse-mesh solidification modeling through an upscaling physics-based data-driven calibration

This research explores a fine-to-coarse mesh upscaling strategy for modeling solidification under turbulent flow conditions, with a specific focus on molten salt flows relevant to Generation-IV nuclear reactors. We propose modifications to the baseline solidification enthalpy–porosity coarse-mesh model employing a physics-informed approach that requires calibration from a high-resolution model. A data-driven approach is employed to calibrate the proposed model’s coefficients using fine-mesh data obtained from a validated CFD RANS solidification model. The performance of the new solidification coarse-mesh model is assessed for turbulent forced flow regimes with varying degrees of overcooling. In comparison to the resolved fine-mesh model, the trained coarse-mesh model significantly reduces computational time, demonstrating promising results in predicting various parameters such as pressure drop, velocity profiles, temperature profiles, outlet temperatures, and solid thickness distribution within the domain. •Development of a fine-to-coarse mesh upscaling strategy for modeling solidification under turbulent flow conditions.•Identification of baseline enthalpy–porosity model shortcomings in coarse meshes.•Data-driven approach to calibrate the physics-based coarse-mesh model coefficients.•The trained model reduces computational time while preserving accuracy.