Quantification of structural uncertainties in RANS turbulence models

Abstract

This thesis presents an approach for building a statistical model for the structural uncertainties in Reynolds averaged Navier-Stokes (RANS) turbulence models. This approach solves an inference problem by comparing the results of RANS calculations to direct numerical simulation. The adjoint method is used to efficiently solve an inverse problem to determine the RANS turbulent viscosity field that most accurately reproduces the mean flow field computed by direct numerical simulation. The discrepancy between the inferred turbulent viscosity and the turbulent viscosity predicted by RANS is modeled as a Gaussian random field. Finally, the uncertainty in the turbulent viscosity field is propagated to the quantities of interest. Results are first presented for turbulent flow through a straight channel. To model the uncertainty in more complex flows, the procedure is repeated for a collection of flows through randomly generated geometries.