You are on page 1of 3

TMMV08 - Basic course in CFD

Extended Abstract, Assignment 4: Part II


18 March 2021, Linköping University, Sweden

Aerodynamic Simulations Over a Generic Car Model


- Turbulence Model Performance for Early Design Studies
Jana Caspari, Pijus Bernatavičius, Arturo Mateos Rodríguez

INTRODUCTION A Root-Mean Squared (RMS) residual target of 1e-4 was set. This
The present study investigates a simplified car model, known as the was reached for all simulations. At this occasion it is mentioned that
Ahmed-Body. Such geometrically simplified models are referred to the residuals were strongly oscillating after leveling out below the
as bluff bodies. Bluff bodies are models that reproduce the main residual target. This represents an expected behavior, as a highly
flow features, but reduce the computational costs derived from unsteady flow case is considered.
more complex and realistic geometries. The Ahmed body is first
reproduced within a simulation environment. Furthermore, the Table 1: Mesh Verification results
model is verified and validated to create a baseline for further design
investigations, that is the variation of the slant angle as well as the Cd Cl
variation of the aspect ratio of the car. The validation of the model
N1 2075238 2075238
is done based on the results presented in Ahmed et al. [1] for the
slant angle variation and Venning et al. [2] for the aspect ratio N2 947988 947988
variation. Furthermore, provided surface plots of Large Eddy N3 359907 359907
Simulations (LES) are used for the purpose of validation. r21 1.30 1.30
All analyses are conducted for a Reynolds number of 30000. r32 1.38 1.38
Regarding the simulation limitations, it must be said that the team phi1 0.453256 0.032821
was limited to make use of steady-state RANS simulations and the
Shear Stress Transport (SST) turbulence model, together with a phi2 0.432279 0.046583
maximum threshold of 512 000 mesh nodes. phi3 0.479491 0.126474
Considering these computational limitations, the aim of this study is GCI_fine 0.056910 0.182146
to investigate, if the known flow features around the Ahmed body GCI_coarse_3
0.099625 0.498685
can be reproduced by simulation and hence, if such computational 2
limited simulations may therefore be used as an alternative to
traditional wind tunnel testing for conducting aerodynamic analyses
of a car in early design stages. RESULTS & DISCUSSION
First, the simulation results were validated against provided surface
plots of skin friction- and pressure-coefficients. The latter is
METHODS showing high agreements for the front but fails representing the
For all simulations, the Ansys CFX solver using a High-Resolution flow properly over the slant, as can be seen in figure 1.
scheme was chosen. In addition to the mentioned SST model, low
turbulence intensity was applied to the flow. Only half of the Ahmed
body was modeled with the help of symmetry. For catching the
influence of the boundary layer, 20 inflation layers were used in the
vicinity of the wall. The first cell height was calculated based on Tips
and Tricks: estimating the first cell height for y+ [3] to place the first node
in the viscous sub-layer. The resulting y+ value lies in a range up to
two. To cope with the given maximum node limit, but at the same
time having a fine mesh around the analyzed model, a body of
influence was added around the simplified car shape. The final mesh
consisted of 494953 mesh nodes which resulted in 2075238
elements.
The mesh verification was conducted based on the approach given Figure 1: Validation of the simulated model (left) against provided
in Celik et al. [4], showing a final mesh related error of 6% for the pressure coefficient plots (right).
drag coefficient and 18% for the lift coefficient, observe table 1.
With respect to the node limitation given and the fact that the lift However, the skin friction is only poorly captured by the simulation
coefficient is a very aggregated variable, this value was seen to be as can be seen in figures 2 and 3. More precisely, the skin friction is
the best attainable after investigating different approaches regarding over predicted for the frontal area and under-predicted for the slant.
inflation layers and bodies of influence. In addition, figures 1 and 3 show, that the simulation is not able to
Here, N1-3 stands for the number of mesh elements, phi1-3 are the catch the vortices on the C-pillar.
results of Cl (Lift coefficient) or Cd (Drag coefficient) with The total lift coefficient is badly captured, showing underpredictions
corresponding mesh size, R21 and R32 for difference in mesh sizes of 88% compared to the provided data. However, this was expected
and finally GCI – Grid Convergence Index (compare Celik et al. as the mesh is not good enough for capturing this value. The total
[4]).
drag coefficient is overpredicted by 10%, lying in an acceptable slight decrease starting from 40° can be seen in figure 5. This is most
range. Table 2 represents these findings, underlining that especially probably rooted in the fact, that the vortices behind the car are not
for areas of high interest, i.e. the back and slant angle of the car, the sufficiently captured, as the pressure drop observed in the LES data
lift coefficient shows high deviations from the LES data. is mainly caused by the merge of C-pillar- and underbody-vortices,
compare Ahmed et al. [1].
Table 2: Cl and Cd results (component break-up)
Splitting up the drag in its components as shown in figure 5 reveals
Cl Cd that the simulation is underpredicting the pressure drag, especially
Front (LES) 0.0892604 0.1035130 over the slant angle. This is due to the fact that the SST model over-
predicts separation, causing a higher energy flow over the slant and
Front (Simulation) 0.0843072 0.140627
therefore underpredicts the pressure drag resulting at this location.
Top (LES) 0.6183200 0.0147397 It worth noting that the drag component in front of the car is over-
Top (Simulation) 0.420028 0.0127304 predicted. This might be related to the same reason as before, this
Side (LES) 0.00114029 0.0274277 is, the SST model assuming a higher energy flow than normal, which
Side (Simulation) -0.000570607 0.0248353 accounts for higher drag coefficient components in front of the car.
Bottom (LES) -0.731049 0.0154563 The mentioned shortcomings of the simulation are mainly a
Bottom (Simulation) -0.603616 0.0181861 problem of the SST model and not of the mesh. These findings
Slant (LES) 0.3073450 0.1444810 underline known problems regarding the SST model, that is, being
Slant (Simulation) 0.0608571 0.0608571 not well-suited to predict flow properties – especially skin-friction
Back (LES) 0.000655004 0.0982419 drag and the production of turbulent kinetic energy – in flows with
Back (Simulation) 0.000460768 0.0681235 high adverse pressure gradients. This is extensively discussed in
Jesus et al. [5].
Total (LES) 0.2857 0.4038596
Total (Simulation) 0.0328219 0.453256 The best, i.e. the lowest drag results are obtained for a slant angle of
10°, the worst for a flat geometry, i.e. a slant angle of 0° or 90°
regarding the point of view.
The adverse pressure gradient is lower for the small angles of the
slant and vice-versa for the higher angles. This tendency defines the
location of the flow separation. With low angled slants, separation
occurs just over the slant and the increasing slant angle moves the
separation of the flow towards the front of the body.
The higher the angle, the bigger wake is created. This phenomenon
can be observed in figure 6 where the Eddy Viscosity Ratio (EVR)
is visualized.
In conclusion it can be stated, that the obtained drag coefficient
values are in general reasonable compared to experimental results.
Unfortunately, the same cannot be said about the lift coefficient
Figure 2: Validation of the simulated model (left) against provided which was basically not captured at all due to the model limitations.
skin friction coefficient plots (right) for the frontal area.

Figure 3: Validation of the simulated model (left) against provided


skin friction coefficient plots (right) for the slant and back.

The simulation results of the slant angle investigations can be Fig 4: Variation of total drag coefficient for different slang angles.
extracted from figures 4 and 5. Here, the slant angle was varied from
5° to 40° in 5° steps. Comparing the results to the provided test data In the following, the influence of the aspect ratio on the drag is
as presented in Ahmed et al. [1], shows, that the overall behavior of examined. For this, the aspect ratio was varied from 1.05 to 2.28.
the total drag with angle variation is captured by the simulation, The results are validated against the data provided in Venning et al.
however, the behavior is not that distinct and therefore presented in [2]. One of the main findings of the mentioned paper is the fact,
a different scaling in figure 4 compared to figure 5. Additionally, the that the circulation at the slant decreases with increasing aspect ratio.
rapid decrease in drag coefficient at 30° is merely visible, only a This is reproduced in figure 7, as with lower recirculation, less
kinetic energy is introduced to the near-wall flow, leading to earlier However, for preliminary studies where it is more important to get
separation and higher drag results. This separation behaviour can be a feeling for the overall development of drag with geometric
easily seen in figure 8, showing a big wake behind a high aspect ratio variation, the method can be seen as sufficient.
car, but only very small eddy viscosity values for a small AR.

Figure 7: Component-wise split up of pressure drag coefficient for


Figure 5: Component-wise split up of pressure drag coefficient for different aspect ratios.
different slant angles.

Figure8: Comparison of EVR for AR=2.28 (top) and AR=1.05


Figure 6: Comparison of EVR for 90° (top) and 10° (bottom) slant (bottom).
angle.
REFERENCES
The paper gives no pressure data to compare the here presented
values to, but it can be assumed, that the here presented pressure [1] Ahmed, S.R. et al, Some Salient Features Of The Time_Averaged
values are in general too high as the interaction between the (not Grond Vehicle Wake, SAE Transactions , 473-503, 1984
caught) vortices around the C-pillar and from the underfloor have a [2] Venning, J et al., The effect of aspect ratio on the wake of the Ahmed
big influence on the result, especially for low aspect ratios, see Body, DOI:10.1007/s00348-015-1996-5, 2015
Venning et al. [2]. [3] TIPS & TRICKS: ESTIMATING THE FIRST CELL
HEIGHT FOR CORRECT Y+ 2020, accessed 11 March 2021,
Whereat for the slant angle variation, the slant was the main reason https://www.computationalfluiddynamics.com.au/tips-tricks-cfd-
for the change in pressure, here it is the frontal area (see component estimate-first-cell-height/
split up in figure 7, where the frontal area shows the steepest slope). [4] Celik, I. B., Z. N. Cehreli, and I. Yavuz, Index of resolution
quality for large eddy simulations, 949-958, 2005
It can be concluded that, under the given limitations, the SST model [5] Jesus, A.B. et al, Large Eddy Simulations Of Turbulent Flows With
is inapplicable for studies regarding the lift coefficient. In addition, Adverse Presssure Gradients, 29th Congress of the International
problematic for slant angle investigations, especially when the Council of the Aeronautical Sciences, St. Petersburg, Russia,
absolute drag values for the individual components are important. 2014.

You might also like