Application of Uncertainty Quantification Techniques to Studies of Wall-Bounded Turbulent Flows
- Location: 2446, ITC, Lägerhyddsvägen 2, hus 2, Uppsala
- Doctoral student: Rezaeiravesh, Saleh
- About the dissertation
- Organiser: Avdelningen för beräkningsvetenskap
- Contact person: Rezaeiravesh, Saleh
Wall-bounded turbulent flows occur in many engineering applications. The quantities of interest (QoIs) of these flows can be accurately obtained through experimental measurements and scale-resolving numerical approaches, such as large eddy simulation (LES). However, due to the prohibitive computational costs imposed by the turbulent boundary layers (TBL) involved in these flows, the use of a standard wall-resolving (WR)LES is limited to low Reynolds (Re-) numbers. As an alternative, wall-modeled (WM)LES can be employed, in which the near-wall region of the TBL is modeled.
This thesis evaluates the uncertainties involved in the measured QoIs of a set of experiments on TBLs, and also, investigates the predictive accuracy and sensitivity of LES, both wall-resolving and wall-modeled. For these purposes, different uncertainty quantification (UQ) techniques are employed.
In particular, such techniques are applied to the forward (uncertainty propagation) and inverse (parameter estimation) problems involved in the measurement of mean velocity and wall shear stress using hot-wire anemometry and oil-film interferometry, respectively. The possibility of reducing epistemic uncertainties by a more detailed analysis is demonstrated. The metamodels constructed by combining non-intrusive generalized polynomial chaos expansion with the stochastic-collocation method are employed to investigate the sensitivity of WRLES of turbulent channel flow to grid resolution. This research further provides a set of recommendations for grid resolution. Through the use of a systematic simulation campaign, the predictive accuracy and sensitivity of WMLES of the same flow is investigated with respect to several influential factors. The metamodel technique is also used to explore the sensitivity to the grid anisotropy and wall model parameters. Based on this study, a set of best practice guidelines is obtained for WMLES of turbulent channel flow, the validity of which is confirmed in a wide range of Re-numbers. For all the UQ-based studies, variance-based sensitivity analysis is also performed.
For WMLES, this thesis also introduces several developments in wall-stress modeling. The performance of algebraic wall-stress models is investigated in an a-priori framework, using accurate WRLES data. Two novel approaches based on integrating the wall model and dynamically adjusting its parameters are proposed and tested. This thesis also contributes to the development of an open-source library for WMLES based on OpenFOAM, which is used in the afore-mentioned systematic study for channel flow.