The rebirth of the Cartesian grid
[ jeudi 17-10-2013 15:00
| français ]The rebirth of the Cartesian grid
Turbulence, Mixing and Flow Control Group
Department of Aeronautics
Imperial College London
Simulating and understanding turbulent flows remains one of the most challenging problems in mechanics. Significant progress has been made recently using high performance computing, and computational fluid dynamics (CFD) is now a critical complement to experiments and theories in order to understand turbulent flows and how to apply them in various engineering contexts.
Only very few codes for Direct and Large Eddy Simulations (DNS/LES) are capable of undertaking massive simulations with several billion mesh nodes on thousands of computational cores. Most of them are simulating idealized homogeneous, isotropic turbulence, using spectral methods with periodic boundary conditions in at least two spatial directions. Engineering problems have more complex geometries and full spectral approaches are not practical. In conventional CFD, especially in an industrial context, complex geometries are usually treated using non-structured element meshes, requiring low-order schemes and sophisticated tools for the generation of highly distorted meshes. The resulting accuracy is most of the time incompatible with a detailed analysis of engineering problems.
Note that the spectral element method seems to be a very promising strategy to undertake complex problems with the spectral accuracy. However, using this technique on thousands of computational cores is a challenging task that requires important numerical developments to conciliate accuracy, efficiency and scalability. In this talk, I am presenting an innovative numerical method which can reconcile accuracy, efficiency, versatility and scalability using a Cartesian grid. It allows accurate simulations of a wide range of engineering configurations. Some examples will be shown in this talk.