The Local Variational Multiscale Method For Turbulence Simulation PDF Download

"This work discusses the development, the verification and the validation of high-order (of accuracy) solvers for the simulation of turbulent fluid flows. In a first part, a methodology for the verification of high-order solvers is proposed which examines the implementation, in a given computer software, of mathematical models that represent the flow dynamics. The method of manufactured solutions is adopted which allows for the exact evaluation of discretization errors and observed orders of accuracy. The latter are compared to the theoretical orders of accuracy for a variety of increasingly complex problems, starting from unbounded (by walls) inviscid flows and gradually reaching the case of realistic turbulent boundary layers, modelled by the Reynolds-averaged Navier-Stokes equations, along with the original and the negative Spalart-Allmaras closure models. Each case serves to illustrate and discuss salient aspects of the proposed methodology.Code verification via manufactured solutions is furthermore distinguished from solution verification and the latter is extended to high-order frameworks by describing the approach for error estimation via extrapolation of high-order solutions which is applied, as an example, to the case of turbulent flow over a flat plate. This exercise enabled the exploration into the question of high-order grid convergence for various output quantities of interest as well as the question of uncertainty analysis. This pointed at the inadequacy of substituting solution verification to proper high-order solver verification, even for a relatively simple problem and an expert set of grids. It is therefore recommended to use the solutions of actual problems for engineering and science applications, only along with estimated numerical uncertainties, especially for lower orders which are more error-prone. In the second part of this work, a compact high-order variational multiscale method for the large-eddy simulation of turbulence is devised and validated. The existing multiscale method is thus expanded from modal frameworks to a large family of compact high-order nodal schemes, represented by the recent flux reconstruction discretization method. The potential of the proposed formulation is then assessed on a Taylor-Green vortex problem and its results are validated against the filtered data from direct numerical simulations. It is thus revealed that proper de-aliasing is mandatory to conserve the quality of high-order large-eddy simulation. Furthermore, reducing the numerical dissipation of Roe's Riemann solver is found to contribute to improving low-Mach flow solutions for variational multiscale methods. Finally, the proposed variational-multiscale formulation resulted in noticeable improvements over the baseline implicit as well as the classical large-eddy simulations." --