Epistemic Uncertainty Quantification of RANS Turbulence Models

Reynolds Average Navier Stokes (RANS) turbulence models are widely used in industrialapplications. They have low computational cost; however, introduce model form uncertaintyinto simulation results. In contrast, Direct Numerical Simulation (DNS) provides highaccuracy results in a very expensive manner, that has kept it away from being applicable inindustrial uses. In this situation, quantification of structural uncertainty in turbulence modelsby using DNS data looks promising. This procedure makes the available uncertaintiesunderstandable and helps engineers to make better decision from turbulence models results.In this study, a new and efficient method to quantify this type of uncertainty is presented, inwhich RANS equations with and without modeling are balanced so that, the sources ofintroduced uncertainty in Reynolds stress tensor can be found. Thereafter, a tensor field foruncertainties is obtained and modeled using a Gaussian Random Field. Then, this field issampled using K-L expansion and RANS equations are solved to find the quantity of interestfor each sample as uncertainty propagation. Compared to previous methods, the presentedprocedure is more compatible with the physics of flow, more powerful in uncertaintypropagation, does not need adjoint method and is easy to extent to other flow configurations.

کلیدواژه ها:

RANS Turbulence Models ، Uncertainty Quantification ، Karhaunen-Loeve Expansion ، Direct Numerical Simulation