|Principal Investigator(s)||Prof.Dr. Claus-Dieter Munz|
The costs for the direct simulation of aeroacoustic problems can become very high if not insurmountable due to a multi-scale problem: Noise producing small-scale structures such as vortices as well as the propagating acoustics with large wavelengths and small amplitudes need to be resolved highly accurate at the same time. On the other hand, direct simulations require the least modelling and include the retroaction of the acoustics to the flow field, a feature that other hybrid models do not offer.
In order to reduce the computational effort of direct simulations, the calculation domain is divided into subdomains, where the grids, methods, time steps, orders of accuracy, etc. are adapted to the local physical requirements. By doing so, the optimal solver is used for each domain. To give an example, an unstructured mesh is only employed in the vicinity of a complicated geometry (e.g. a nozzle), while Cartesian grids are used farther away from an obstacle, delivering faster and more accurate results.
Moreover, each solver should run on the most suitable processor: High-order methods for unstructured grids, such as the Discontinuous Galerkin (DG) and Finite Volume (FV) methods are efficient on cache-based machines with scalar CPUs. On the other hand, Finite Volume and Finite Difference schemes for structured grids show good vectorization properties and can therefore especially exploit vector architectures such as the NEC-SX8.
The IAG in-house code KOP3D is a framework which provides different numerical methods and a well-tested coupling procedure. While being a completely stand-alone code, it uses PACX-MPI to organize the data distribution between different system architectures.
In the past, KOP3D has successfully demonstrated the efficiency of the domain decomposition approach for a sphere scattering benchmark example and shall now prove its applicability to real life 3D calculations. For this purpose, a supersonic nozzle flow with a free-jet (Ma=1.4, Re=30000) is calculated in 3D.