Parameter estimation and uncertainty quantification in turbulence modeling

Abstract

We consider the task of calibrating turbulence models against reference data, with particular reference to the estimation of parameters of Reynolds-averaged Navier-Stokes closure models. Traditional calibration methods against canonical flows are discussed, and we introduce the framework of Bayesian probability to generalize this basic approach. A major benefit of Bayes is that we obtain uncertainty information indicating the level of confidence in the calibration results; which we can use to subsequently estimate the uncertainty in predictions due to the model. Numerical methods for solving the challenging computational problems involved are discussed. Finally we apply the same Bayesian framework to the data-assimilation problem of indirectly identifying turbulence anisotropy fields given experimental data pressure data.