Effects of turbulence and subgrid-scale modeling on transient free-surface dynamics predicted by a coupled volume-of-fluid method and discrete phase model in a gas-blown continuous casting mold

A transient multiphase model coupling the volume-of-fluid method and discrete phase model was evaluated against a 1:2 gas-blown water-model experiment. Five turbulence configurations—standard k–ε, shear-stress transport k–ω, large-eddy simulation (LES), k–ε-based stress-blended eddy simulation (SBES), and k–ω-based SBES—were compared using jet trajectory, near-surface velocity, wave-height statistics, oil-layer exposure, visible oil entrainment, and computational cost. On the common 1.80 × 106 cell mesh, k–ε-based SBES and LES gave three-point mean absolute relative velocity errors of 8.32% and 10.60%, respectively. SBES required 44.1% of the LES wall-clock time. The SBES velocity RMSE was 0.009 03 m/s. Its mean absolute wave height and temporal standard deviation were 0.79 and 0.59 mm, respectively, compared with 0.87 and 0.61 mm experimentally. LES gave smaller common-mesh errors in both wave-height statistics and exposure variability, whereas SBES gave a slightly smaller mean-exposure error. For visible entrainment, SBES predicted a maximum penetration depth of 107.51 mm vs 110.86 mm experimentally, but mean visible droplet counts were 6 and 3, respectively. Within k–ε-based SBES, the wall-adapting local eddy-viscosity closure was closest to the experimental point estimates in all four retained interfacial statistics. Model performance therefore depends on the observable, and selection should consider complementary measurements and computational cost within the investigated water-model condition.