Applications
- Vehicle Aerodynamics
Models active in this Benchmark case
- Incompressible flow
- Detached Eddy
- SST Formulation
- Double Precision
Objective
To compare the performance of Envenio’s EXN/AERO manycore CFD solver against a leading commercial CFD tools on a cost-normalized basis.
From a field of candidates, Envenio selected an engineer to run a CFD simulation using two commercial software packages. These included Fluent and EXN/Aero. While the engineer was required to be knowledgeable in Fluent, experience with EXN/Aero was not required and the engineer was provided training from Envenio in this regard.
Apart from knowledge of CFD software packages, the tester was required to:
- have experience with unsteady turbulent flow simulation including URANS and LES/DES approaches,
- be comfortable navigating & managing data within a linux environment,
- have access to a linux machine for post-processing results.
The engineer was required to run a ~30 million element finite volume simulation using each solver. For the Fluent solution, the tester created a cut-cell mesh in the Ansys meshing utility that included progressive mesh refinement near the body using hanging nodes. For EXN/Aero, a fully Cartesian CGNS mesh file (with structred data type) was created in Pointwise by Envenio engineers and provided to the tester.
Upon completion of the simulations, the engineer prepared a short report that described the simulation results in terms of performance and accuracy. Specific metrics are listed in the ‘outcomes’ section.
Model Description – Ahmed Body
The Ahmed body described originally by Ahmed et al. in 1984 [1] is shown Fig. 1. The configuration with a slant angle of has been studied extensively both experimentally and computationally. A review of recent experimental drag data over a range of Reynolds numbers is provided by Bello-Millan et al. (2016). The reference shows a wide range of experimental data, , which depends critically on the sharpness of the model edges and Reynolds number. A comprehensive computational study using both LES and DES approaches by Serre et al. (2011) reports lift and drag results for several subgrid and turbulence models including that employed by EXN/Aero. Meille et al. (2011) provide RSM RANS simulation and experimental results.
Figure 1: Ahmed body general dimensions
Mesh Description
Domain:
Shape Rectangular box
Length 9.2 m
Width 2.0 m
Height 1.4 m
Position of model:
Upwind face relative to inlet: 2.1 m
CL offset relative to domain CL 0.0 m
Bottom face above ground plane 0.05 m
Mesh Resolution in the vicinity of the Ahmed Body:
CV Thickness Top 2.3e-5 m
CV Thickness Side 3.7e-5 m
CV Thickness Bottom 1.3e-3 m
CV Thickness Wall 3.7e-5 m
CV Thickness Slant 1.5e-5 m
Streamwise Length Top 0.02 m (max)
Streamwise Length Side 0.02 m (max)
Streamwise Length Bottom 0.01 m – 0.015 m
Streamwise Length Wall 0.01 m – 0.015 m
Streamwise Length Slant 0.01 m – 0.015 m
Horizontal Aspect Ratio (all) ~10 (max)
Vertical Aspect Ratio (all) ~1000 (max)
Simulation Completion Criteria
The simulation should be run for at least one ‘wash through’ of initial conditions in order to obtain accurate lift and drag numbers. The number of iterations N is calculated using the equation below:
where d is distance from inlet to outlet along the mean velocity vector, is the mean velocity magnitude and is the time step duration. Nominally, N will be the same for the Fluent, StarCCM and EXN/Aero simulations.
For performance measurement purposes, the engineer should monitor the ‘CPU time’ column in the CGNS file. When it has reached a statistically steady value in time, the engineer should run for 50 iterations and then stop.
Simulation Setup
Solver Control
Time Step 0.00005 s (note EXN/Aero can actually run at 1e-4 sec)
EXN GPU Allocation 2
EXN CPU Allocation 4
X-axis orientation Positive downstream
Y-axis orientation Positive to the left of the body, looking downstream
Z-axis orientation Positive upward, normal to ground plane
Ansys / StarCCM
CPU Allocation At the discretion of the engineer
Boundary Conditions
Velocity [x,y,z] [40 , 0, 0]
Kinetic Energy 0.0001
K Dissipation 0.0003
Wall model Smooth wall
Outlet Constant Pressure
Domain Settings
Initial Velocity [40 , 0, 0]
Initial Kinetic Energy 0.0001
Initial K Dissipation 0.0003
Turbulence Model Detached Eddy, SST formulation
Flow type Incompressible
Near-body Precision Double-precision. Use single precision for outer blocks if possible
Constant Density 1.203 kg/m^{3}
Constant Viscosity 1.05 x 10^{-5}
Other notes
Wash through time 0.23 seconds (4600 time steps)
Mesh Topology Hexahedral, written in CGNS format
Calculation Methodology
Simulation Results
The table below shows a comparison of the present simulation results to published numerical and experimental values. The Reynolds number is defined as , the drag coefficient is and the lift coefficient is . L is the model length and A is the frontal area. The EXN/Aero results and the Serre et al. (2011) DES results most closely match the recent experiments of Bello-Millan et al. (2016) where they fall between the sharp edge model and smoothed model results. The Fluent results using both LES and RANS match the earlier measurements of Ahmed et al. (1984) and those of Meile et al. (2011). A characteristic of higher drag coefficients in either experiment or simulation is separated flow at the edge along the rear roof line.
Simulation Timing and External Factors
^{Cost-Normalized Specific Performance Figures }
Reporting Item
Locally Hosted |
Fluent | EXN/Aero |
# CPUs | 4 | 4 |
# GPUs | 0 | 2 |
Cost | $5,000.00 | $25,000.00 |
Capital & Operating / hour | $0.19 | $0.95 |
Annual License | $30,000 | $30,000 |
License / hour | $3.42 | $3.42 |
Compute Cost / hour | $3.61 | $4.38 |
Compute Cost | $16,654 | $1,437 |
Timesteps | 22,030 | 22,050 |
MSM (timestep / $) | 1.33 | 15.34 |
Annual Timestep Max (MSM) | 168,577 | 588,322 |
Reporting Item
Cloud Hosted |
Fluent | EXN/Aero |
# CPUs | 4 | 16 |
# GPUs | 0 | 2 |
Cost | $3.50 | $1.05 |
Hosting Fee | $14.00 | $2.75 |
Annual License | $0.00 | $30,000 |
License / hour | $0.00 | $3.42 |
Compute Cost / hour | $14.00 | $6.17 |
Compute Cost | $64,152 | $2,028 |
Timesteps | 22,030 | 22,050 |
MSM (timestep / $) | 0.34 | 10.87 |
Annual Timestep Max (MSM) | 168,577 | 588,322 |
^{Impacts}
References
- R. Ahmed, G. Ramm, Faltin, G. Some Salient Features of the Time-Averaged Ground Vehicle Wake, Automobile Aerodynamics: Wakes, Wind Effect, Vehicle Development, SAE SP569, 1984.
- Bello-Millan, F.J., Tl Makela, L. Parras, C. del Pino, C.Ferrera, Experimental study on Ahmed’s body drag coefficient for different yaw angles, J. Wind Eng. Ind. Aerodynamics, Vol. 157, pp. 140-144, 2016.
- Serre, E. M. Minguez, Pasquetti, R., Guilmineau, E., Deng, G. B., Kornhass, M., Schafer, M., Frohlich, J., Hinterberger, C. , Rodi, R., On simulating the turbulent flow around the Ahmed body: A French-German Collaborative evaluation of LES and DES, Computers & Fluids, v78 n9, pp. 10-23, 2011.
- Meile, W., Brenn, G., Rappenhagen, A., Lehner, B., Fuchs, A. Experiments and numerical simuations on the aerodynamics of the Ahmed body, CFD Letters, Vol. 3, No. 1, pp. 32-39, 2011.