On dynamic subgrid-scale modeling for large-eddy simulation of compressible turbulent flows
Abstract
Recently, through the advent of the dynamic subgrid-scale model, Large-Eddy Simulation (LES) has shown great promise towards the accurate simulation of complex turbulent flows. The main objective of this work is to advance the LES method for analysis of compressible turbulent flows. The model is employed in simulations of compressible decaying isotropic turbulence, compressible homogeneous turbulent shear flow and of a spatially evolving supersonic turbulent boundary layer flow, and evaluated against results from direct numerical simulations (DNS) and experiments. Results suggest the model captures compressibility effects well in decaying isotropic turbulence. The use of spatial filters for use in inhomogeneous flow simulations is also examined. Implicit filters perform much better than explicit filters, but are more expensive to employ. Results also suggest a great insensitivity of the model on the filter width ratio. A modification of the convective terms in the momentum and energy equations reduces the effects of aliasing errors. Two formulations of the energy equation are examined and the nonconservative form is found more accurate. Testing of a finite difference scheme shows inadequate capture of smaller resolved scales. This is alleviated by explicitly filtering the fields during the simulation. In simulations of homogeneous turbulent shear flow, the dynamic model captures qualitatively the reduced rate in turbulence growth due to compressibility. However, it predicts somewhat lower growth rates than these found by DNS. A spatially evolving supersonic boundary layer is simulated using a high-order-accurate finite difference scheme and the dynamic model. A parametric study suggests that the upwind-biased scheme has a detrimental effect on the resolution of the smaller scales due to excessive numerical dissipation from the spatial differencing. Also, since the dynamic model uses the smaller resolved-scale eddies to determine the model coefficients, the predicted coefficients do not have the appropriate values. The use of higher-order schemes is found to better capture the smaller resolved scales and substantially improve the quality of the results. The effect of descretization errors on LES needs to be addressed further before proceeding with LES of flows of engineering interest.
Degree
Ph.D.
Advisors
Blaisdell, Purdue University.
Subject Area
Aerospace materials|Mechanical engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.