No label defined (Q4337613)

Starting point of this comprehensive research report (132 pages, including 11 figures and 526 references) is an analysis of the general and application oriented demands concerning LES (large eddy simulation)-specific average operators and fine structure models. Shortcomings and performance limits of classical SGS (subgrid scale) models of eddy viscosity/Smagorinsky-type are pointed out, and motivations for ``dynamic'' models are given. Besides, the authors present potential advantages of the coefficient determination in dynamic models, based on the Germano identity. In particular, general and model-specific problems concerning mathematical-physical foundations and numerical conversion of ``dynamic'' modeling are discussed, together with the solution by means of hypotheses introduced in the literature. An overview is given of essential concepts of ``dynamic'' modeling and ``dynamic'' subgrid parametrization (closing hypotheses for the SGS stress tensor in the filtered hydrodynamic basic equations of incompressible fluids). In addition to the original modeling concept of Germano, a modified approach is included based on a statistical formulation, as well as an alternative concept based on an asymptotic expansion. Then a further development of ``dynamic'' SGS model invariants is suggested, using the Germano identity and relations derived thereof. These invariants are mainly of algebraic Smagorinsky , scale-similarity- and combined type with local tensorial, component related or scalar model coefficients. Some model properties are discussed. In connection with the model proposals, statements regarding compatibilities with filters and discretization are cited. Numerical studies are carried out for a fully developed turbulent pipe flow at Reynolds number 50.000, referred to the pipe diameter and the axial velocity, to test the Smagorinsky-type model and two ``dynamic'' invariants, combined with two discrete filters. For the numerical experiments the authors develop a software based on a finite difference discretization on a three-dimensional staggered grid, on a split-up procedure for the treatment of the pressure-velocity coupling, and on the domain decomposition parallelization strategy. Results of simulation tests for 11 combinations of discrete filters, subgrid scale model invariants and grids show the dependence of the SGS turbulent viscosity and the calculated statistical one-point turbulence characteristics on the subgrid parametrization, the discrete filter approximation, and on the grid density (four distinctly fine discretizations with a maximum of about one million nodes). In connection with the use of empirical substitute boundary conditions on the walls and the relatively coarse decomposition of zones near the walls, the authors note in the immediate neighborhood of the wall some deviations from data reported in the literature. The ``dynamic'' fine-structure model with invariant of Lilly and the model with rectangular filter approximated by means of the trapezoidal rule yield results which agree as well as possible with experimentally obtained flow data, in contrast with the classical Smagorinsky-type model.