A surrogate-based Bayesian framework for parametric calibration of URANS turbulence models informed by anisotropic pressure fluctuations

Abstract

Flow-induced vibrations in wall-bounded flows remain a major challenge in many engineering applications. While fully predictive turbulence-induced vibration simulations generally require scale-resolving approaches, unsteady Reynolds-averaged Navier–Stokes (URANS) methods are employed in practice due to their favorable computational cost. However, their predictive reliability is limited by uncertainty in turbulence-model parameters, structural model–form error, and deficiencies in representing turbulent fluctuations. This work presents a surrogate-based Bayesian calibration framework for the SST \(k\)–\(\omega\) model, incorporating anisotropic pressure-fluctuation reconstruction to quantify and reduce model uncertainty. Global Sobol’ sensitivity analysis, surrogate modeling, and Bayesian inference are combined to identify influential parameters and assimilate high-fidelity reference data. The framework is applied to turbulent channel flow, turbulent annular flow, and a blunt-end cantilevered-rod configuration. Across all cases, the specific dissipation-rate coefficient \(\alpha_{\omega1}\) and the turbulent kinetic-energy dissipation coefficient \(\beta^*\) consistently dominate variability in mean-flow quantities and reconstructed fluctuation statistics. Bayesian calibration constrains these parameters to consistent regions of the admissible space, while less influential coefficients remain weakly informed. Gaussian process regression and polynomial chaos expansion surrogates yield nearly identical posterior and predictive distributions, demonstrating robustness to surrogate choice. Remaining discrepancies with reference data are attributed primarily to structural turbulence model limitations rather than parametric uncertainty.