Prediction of the Turbulent Flow Field Around a Ducted Axial Fan

Institute of Aerodynamics, RWTH Aachen University

Principal Investigator: Wolfgang Schröder, Institute of Aerodynamics, RWTH Aachen University

Exploiting the available computing capacities of supercomputer Hornet of the High Performance Computing Center Stuttgart, researchers from the Institute of Aerodynamics (AIA) of the RWTH Aachen University conducted a large-scale simulation run in their efforts to tackle the prediction of the acoustic field of a low pressure axial fan using computational aeroacoustics (CAA) methods. Goal of this project, which scaled to 92,000 compute cores of the HPC system Hornet, was to achieve a better understanding of the development of vortical flow structures and the turbulence intensity in the tip-gap of a ducted axial fan.

Key Facts

  • 92,000 compute cores
  • 110 machine hours
  • 80 TB of data
  • 1,000 mill. grid points
  • 320,000 time steps

Axial fans can be found in many technical applications, from computer CPUs to automotive engines and industrial air-conditioning systems. In addition to the performance, noise reduction has become one of the major issues for engineers in recent years. Therefore, efficient numerical methods which are able to accurately predict the acoustic field are required.
In a joint university-industry project, the prediction of the acoustic field of a low pressure axial fan using computational aeroacoustics (CAA) methods is tackled. The source distribution of the CAA analysis, however, requires highly resolved instantaneous flow field data. Since Reynolds averaged Navier-Stokes (RANS) computations strongly depend on the chosen turbulence model and are not always reliable due to, e.g., the strong streamline curvature dominating the flow structures inside the tip-gap region, the subsonic flow field is predicted by large-eddy simulation (LES). For this purpose, the unstructured flow solver ZFS (zonal flow solver), which solves the Navier-Stokes equations for unsteady and compressible flows in the rotating frame of reference on Cartesian meshes, is used. The overall accuracy of the flow solver in space and time is of second order.
Simulations were performed at a Reynolds number of Re=9.36×105 based on the outer casing diameter and the rotational velocity of the casing wall. The current mesh has approx. 1 billion cells, resolving only a 72° segment of the axial fan, i.e., one out of five blades, to reduce the high computational costs. This high resolution is necessary to accurately resolve the vortical flow structures and their development in the tip-gap region.

Simulations were performed on the CRAY HPC system HORNET of the High Performance Computing Center Stuttgart (HLRS). The flow solver has been fully parallelized using the message passing interface (MPI) so that different number of cores can be used for the flow simulation. The minimum number of cores required for a simulation using 1 billion cells is approx. 10,000. However, up to 92,000 cores have already been used.

The required time per simulation, i.e. four full rotations of the rotor using 92,000 cores, is about 110 machine hours.

To obtain accurate statistical data from the turbulent flow field like, e.g., the Reynolds stress tensor and two-point correlations of the velocity components, a large number of samples of the instantaneous field is required. The statistical data requires about 80 TB of disk space.

Project Team and Scientific Contact

Dr.-Ing. Matthias Meinke, Alexej Pogorelov M. Sc.,
Prof. Dr.-Ing. Wolfgang Schröder (PI)

Institute of Aerodynamics, RWTH Aachen University
Wüllnerstraße 5a, D-52062 Aachen, Germany

Fig. 1: Instantaneous contours of the Q-criterion, coloured by the relative Mach number, show the vortical structures of the flow field (Institute of Aerodynamics, RWTH Aachen University).