Experimental data-driven calibration of the SST turbulence model: Validating physical interpretability via large Eddy simulation

Standard Reynolds-Averaged Navier-Stokes (RANS) models often struggle to predict complex film cooling flows due to deficiencies in their constitutive relations, particularly in capturing jet separation and anisotropic turbulent mixing. To overcome these limitations while maintaining computational efficiency, this paper presents a data-driven calibration framework for the Shear-Stress Transport (SST) model. By assimilating high-fidelity experimental data obtained from Pressure-Sensitive Paint (PSP) measurements, the Ensemble Kalman Filter (EnKF) algorithm is employed to optimize key model closure coefficients. Unlike traditional calibration methods that focus solely on output metrics, this study prioritizes the physical interpretability of the calibrated model. To rigorously validate this, Large Eddy Simulation (LES) is performed to provide a high-resolution benchmark of the spatial flow field. The comparison reveals that the experiment-driven model does not merely curve-fit the adiabatic effectiveness; it successfully reproduces the underlying flow physics, including the critical jet lift-off and shear layer development, which align closely with the LES predictions. Furthermore, the calibrated model demonstrates robust generalizability when applied to holes with manufacturing deviations (e.g., filleted corners), reducing prediction errors by over 90% compared to the standard SST model. This work establishes a physics-consistent, low-cost simulation strategy that bridges the gap between RANS efficiency and LES fidelity. • EnKF SST achieves <5% deviation vs. SST's 30 + % error. • LES-aligned velocity/temperature profiles validate physical interpretability. • Calibrated model shows robust generalizability to holes with realistic manufacturing deviations.