Quantification of the uncertainty within a SAS-SST simulation caused by the unknown high-wavenumber damping factor

• Published in: Nuclear engineering and design Vol. 381; p. 111307
• Main Authors: Duan, Yu, Ahn, Ji Soo, Eaton, Matthew D., Bluck, Michael J.
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 01.09.2021; Elsevier BV
• Subjects: Conditional probability; Damping; Finite element method; Gaussian process; Gaussian processes/Kriging; Mathematical analysis; Mathematical models; Numerical models; Probability density functions; SAS-SST simulation; Simulation; Surrogate modelling; Turbulence models; Uncertainty; Uncertainty quantification; Wavelengths; Uncertainty quantification; SAS-SST simulation; Surrogate modelling; Gaussian processes/Kriging
• ISSN: 0029-5493; 1872-759X
• DOI: 10.1016/j.nucengdes.2021.111307

•Response of different QoIs to the changing Cs at the same location are different.•The prior pdf of Cs determines the profiles of statistical quantities of the QoIs.•Altering Cs causes rebalance of the resolved and modeled turbulent Reynolds stresses.•Altering Cs has different effects on recreating the different instability mechanisms.•The uncertainty quantification can offer more confidence to the CFD results. This paper aims to quantify the uncertainty in the SAS-SST simulation of a prism bluff-body flow due to varying the higher-wavenumber damping factor (Cs). Instead of performing the uncertainty quantification on the CFD simulation directly, a surrogate modelling approach is adopted. The mesh sensitivity is first studied and the numerical error due to the mesh is approximated accordingly. The Gaussian processes/Kriging method is used to generate surrogate models for quantities of interest (QoIs). The suitability of the surrogate models is assessed using the leave-one-out cross-validation tests (LOO-CV). The stochastic tests are then performed using the cross-validated surrogate models to quantify the uncertainty of QoIs by varying Cs. Four prior probability density functions (such as U0,1, N0.5,0.12, Beta2,2 and Beta5,1.5) of Cs are considered. It is demonstrated in this study that the uncertainty of a predicted QoI due to varying Cs is regionally dependent. The flow statistics in the near wake of the prism body are subject to larger variance due to the uncertainty in Cs. The influence of Cs rapidly decays as the location moves downstream. The response of different QoIs to the changing Cs varies greatly. Therefore, the calibration of Cs only using observations of one variable may bias the results. Last but not least, it is important to consider different sources of uncertainties within the numerical model when scrutinising a turbulence model, as ignoring the contributions to the total error may lead to biased conclusions.