Personal tools
You are here: Home Science & Projects Deisa Extreme Computing Initiative Projects 2009 - 2010 Computing Backbone for Unstructured LES and DNS

Computing Backbone for Unstructured LES and DNS

Research Area Engineering
Principal Investigator(s) Koen Hillewaert
  • Cenaero, CFD and mutiphysics group, Gosselies, Belgium
  • Katholieke Universiteit Leuven, Belgium
  • University of Liege, LTAS, Li├Ęge, Belgium
  • Clarkson University, Dept of Mechanical and Aeronauticaul Engineering, Postdam, USA


Bluff body flows and flows at low to moderate Reynolds numbers are often dominated by large scale separation and instabilities, such that RANS simulations are no longer sufficient to even predict time-average forces correctly, and more direct methods such as DNS/LES or DES are necessary. These approaches however require extremely low dispersion and dissipation errors, in order to avoid contamination of the modeling with numeric error.
As numerical simulation perfuses more and more domains related to fluids engineering, LES and DES are applied to ever more complicated geometries. In these cases it is impossible to generate the (high-quality) structured meshes used for the classical spectral or high-order finite difference methods, hence we need methods that allow for unstructured, low-quality meshes.
Current state of the art unstructured DNS/LES/DES technology consists of secondorder kinetic-energy conserving finite volume schemes (ke-FV). Due to the low order of accuracy there is a growing concern however this technology will require too high resolution and mesh quality.
Discontinuous Galerkin methods provide a natural way to provide a very high order of accuracy on unstructured meshes whilst retaining computational simplicity and efficiency. On the downside, these methods cannot provide exact numerical conservation of kinetic energy. Due to the locality of the datastructure, DGM codes can be implemented very efficiently (measured global performance 6 to 10 Gflops on SSE architectures, running at 2.5GHz).
Both ke-FV and DGM technologies are being developed jointly by the different partners participating to this project. At this stage of development it is desirable to perform a quantitive comparison of both methodologies on a number of selected benchmark DNS computations. The results will provide guidelines concerning necessary resolution and choice of method. Subsidiary to this research the impact of the Power architecture on DGM efficiency will be assessed.

Document Actions