Sei sulla pagina 1di 15

A Comparison of Experimental Investigations and

Numerical Simulations around Two-box form


Models
Sylvain Parpais (+), Jean Farce (++), Olivier Bailly (+), Herve Genty (*)
Technocentre Renault, La Ruche, 4e etage, 1 Avenue du Golf, F-78288 Guyancourt Cedex, France
+ Technocentre Renault, ++ Formerly Technocentre Renault, now Renault F1,
* Renault Wind tunnel team
Swapan Mallick, Alain Belanger, Satheesh Kandasamy
Exa Corporation, 450 Bedford Street, Lexington, MA 02420

Synopsis
The effect of back-end and underbody geometry variations on lift and drag for twobox form shapes has been investigated using Renaults test suite models. For
these geometries, detailed new experimental data has been obtained including
centreline pressure data, off-centreline pressure data, and extensive wake
visualisations, as well as aerodynamic lift and drag. Further, numerical simulations
of these geometries have been performed using the Digital Physics based
software PowerFLOW.
The objective of this study is to present the results of experimental investigations
and to assess the ability of PowerFLOW to predict the flow-field around two-box
form cars. The effects of geometry variations on both experiment and numerical
simulation are discussed and compared.
It is concluded that good agreement between experiment and PowerFLOW is
observed. Some deviations from the wind tunnel investigations are noted and
discussed.
1. Introduction
The equations of fluid flow are recovered using the Lattice Boltzmann Method [1].
Further, the recovery of isolated flow field structures in simple geometries and a
demonstration of the recovery of interacting three dimensional aerodynamic
structures were demonstrated by comparing predicted drag for 30 vehicles with
experiment [2]. Additionally, these vehicles were subdivided into classes to
demonstrate the prediction of rank ordering. An extensive validation of
aerodynamic prediction of the class of stylised three-box form geometries has been
previously demonstrated [3,4].
In this paper, we specifically consider two-box form models, and consider two types
of geometric changes: modifications of backlight angle (where the flow regime

changes between 25 degrees and 37 degrees), and modifications to underbody


geometry. In general, decreasing complexity of underbody design leads to a
reduction in drag, as a result of reduced protrusion drag. Whilst this is not always
the case, in the series of test cases presented, drag does indeed follow this trend.
More importantly the effect on lift can be quite dramatic since small changes in
geometry can have a marked effect on the underbody lifting surface. Comparisons
between numerical and experimental flow structures are presented.
2. Mathematical Models
The underlying computational method, Digital Physics is based on the Lattice
Boltzmann Method. Details of this method have been given in a number of sources
including [1, 5]. A description of the modeling strategy employed for both wall and
turbulence models have also been previously described [6]. A number of
sophisticated improvements to both the wall and turbulence models have been
recently incorporated in PowerFLOW, which have improved predictions of transient
flow structures [7]. These have been shown to maintain good aerodynamic
predictions [3] whilst facilitating sophisticated aero-acoustics predictions via
spectral filtering of aerodynamic noise sources [8].
3. Test cases
3.1 Changing backlight angle
The effect of changing the backlight angle may be considered in terms of
competing two-dimensional and three-dimensional separation. Two-dimensional
separation occurs when flow separates on edges perpendicular to the flow
direction, leading to vortices with axis lines perpendicular to the flow. These may
either reattach on the vehicle itself such as for leading edge hood vortices, vortices
on the edges of the bumpers, vortices in the stagnation zone at the base of the
windshield, or subtle separations in adverse pressure gradients on the backlight.
Or they may separate without reattachment at points where the flow is subject to a
local expansion such as at the end of the roof, or the end of the vehicle - this leads
to a dead water region [9].
Three-dimensional separation occurs at edges, which are partially parallel to the
flow direction: flow separates from the edges rolling up to form highly energetic,
free, trailing vortices fed directly from the kinetic energy of the free-stream flow they are conical in shape and are similar to those which occur on delta wings. Welldefined vortex lines in horseshoe pattern may be defined when these separations
are symmetric about the centerline, and are typically observed on A and C-pillars.
Both types of separations may be characterised for the three basic car shapes:
fastback, notchback and squareback [10]. Squarebacks may be considered the
limiting geometry of a fastback shape, and two box form models are of either
fastback or squareback configuration. In general, squarebacks simply have twodimensional separations. However, for fastbacks there is an interplay between the
C-pillar vortices and the two-dimensional roof separations from the end of the roof
dependent on backlight angle.

For shallow backlight angles, there may be a short roof separation, which
reattaches on the slant edge, and a pair of weak C-pillar vortices. As the angle is
increased, the reattachment point of the roof separation moves downstream,
reducing the pressure within the confined recirculation zone. The C-pillar vortices
also become stronger since the free-stream flow feeds directly onto the vortex
more easily. Additionally, the change in recirculation pressure also causes the Cpillar vortices to be drawn inward, also contributing to an overall reduction in base
pressure [10]. This leads to an increase in drag. At some critical angle, the roof
separation is no longer able to reattach on the length of the slant, and the C-pillar
vortices burst, leading to a sudden increase in base pressure and a marked
reduction in drag. At this point, the wake is essentially two-dimensional with no Cpillar vortices observed, and looks much like a squareback wake.
These two flow regimes, high drag and low drag, have been well documented for
axi-symmetric bodies with a slanted backlight (with no ground) [11], where the
critical angle is between 50 and 70 degrees, and simplistic vehicle shapes (with
ground present), where the critical angle is 30 degrees [11,12]. Both papers show
backlight pressure distributions corresponding to each flow regime. In the Morel
body, there is as much as 0.4 difference in Cp between the flow regimes, along the
centreline. The characterisation of this phenomenon on realistic car shapes has
also been reported, with a critical angle at 28 degrees, which is quite close to the
critical angle reported by Morel for simplistic geometries.
3.1.1 Vehicle geometries
Figure 1 shows the baseline geometry. The models have some degree of
complexity with detailed wheels and wheel housings. Four geometries are used:
modules 7, 1, 9 and 5 which correspond respectively to backlight angles, , of 0,
16, 25 and 37 degrees (measured from horizontal). Wake surveys are captured at
three positions, 0.010m, 0.280m and 0.560m downstream of geometry. Pressure
taps cover the full backlight on one side.
3.1.2 Experimental results
The experimental results have been obtained using a two fifth scale model. Figure
2 shows backlight pressures for module 9 (=25) and module 5 (=37), showing the
difference between the attached and separated flow regimes. Module 9 (=25)
shows fully attached pressure recovery, on the full span of the backlight. On the far
outboard sides of the backlight, the presence of C-pillar vortices is marked by a
significant reduction on pressure. Module 5 (=37) shows fully separated flow
across the span of the backlight. Oil streamlines are shown in Figure 3, for
modules 1 (=16), 9 (=25) and 5 (=37). Fully attached flow is shown for module 1
(=16) and 9 (=25). There appears to be large region of slow moving, attached
fluid in module 9 (=25) which is not picked up by the visualisation technique).
Module 5 (=37) shows a fully separated flow pattern.
Figures 4-7 show wake surveys of total pressure deficit of each configuration, at
the locations marked in Figure 1. In Figure 4, the squareback shape of the module

7 (=0) shows a plane of two-dimensional separation from the roof and upper
sides, which recovers downstream. In Figure 5, when a backlight angle is
introduced in module 1 (=16) the presence of a pair C-pillar vortices is shown
this is barely noticeable at position (a), but becomes increasingly pronounced
downstream. The downwash created in the middle of these vortices drags the
wake down from the roofline this can be seen by comparing Figure 4(c) with
Figure 5(c). The centre of the vortices stays approximately the same height above
the ground in the wake.
In Figure 6, (module 9, =25), as the backlight angle is increased, the lobes
resulting from the C-pillar vortices are much more pronounced, dominating most of
the wake flow. However, in Figure 7 (module 5, =37), we see the flow fully
separate with a very weak C-pillar vortex signature present. This is quite similar to
the wake flow of the squareback, module 7 (=0), in that the wake largely consists
of one flow profile in the upper section compare for example, Figure 4(c) with
Figure 7(c). There is an asymmetry in the flow field, which is evident in Figure 7(c)
it is not clear if this is coming from the incoming flow or the geometry itself it
can also be seen in Figure 5(c). Figure 8 shows measured forces. The squareback
shows a considerably higher drag than any of the other configurations. As backlight
angle increase, from modules 1 (=16) to module 9 (=25) drag increases this is
comparable with the drag increase observed in stylised geometries in a high drag
flow regime as backlight angle increases. Once separation occurs, in module 5
(=37), drag drops indicating a change in flow regime. The change in drag is quite
subtle, but the effect of the change in flow regime is pronounced in lift.
From module 7 (=0) to module 1(=16), there is an increase in lift caused by the
large area on the upper surface with low pressure which is created by the presence
of an angled backlight. As backlight angle increases, base pressure drops along
the slant, giving an increase in lift as well as a drag increase. For the module 5
(=37), the base pressure increases leading to a drop in lift.
3.1.3 Numerical analysis
The PowerFLOW cases were set up to match experiment. Surface meshes were
created for each geometry from CAD data, which are then imported into the wind
tunnel template provided with PowerFLOW. This template contains a series of
nested grid regions that are automatically defined. These are used during the
calculation to generate automatically a volumetric grid of cubic cells - the length of
each cubic cell changes by a factor of two across each grid region, permitting rapid
changes in cell size. Additional resolution requirements are added to the geometry
so that each local resolution is increased using a consistent case setup
methodology.
3.1.4 Comparisons
PowerFLOWs results are shown in Figures 2-8. The pressure profiles of the
attached and separated flow regimes of module 9 (=25) and module 5 (=37)
respectively is correctly predicted. This is also shown in the oil streamlines in
Figure 3. The C-pillar vortices for module 9 (=25) are captured but the details of

the structure would require increased resolution. Separated flow is clear for the
module 5 (=37), and the separation line appears to be correct.
Good comparisons with wake flow structure can be seen in Figures 4-7, showing
total pressure deficit at three wake locations shown in Figure 1. In Figure 4, the
squareback shape of module 7 (=0) separates cleanly along the upper sides and
roof at position (a). The position of the wheel wakes matches, and there is good
agreement in the underbody flow, as evidenced by the envelope of low total
pressure deficit. As the wake develops the signature of the hip profile of the
geometry is seen both numerically and in experiment this is very pronounced in
Figure 4(c).
When a backlight angle is introduced in module 1 (=16), C-pillar vortices form. In
Figure 5(b), there is a very slight difference in the strength of these vortices
comparing the simulation to experiment. This is more pronounced in Figure 5(c),
however the general shape matches experiment well. Additionally, the
development of the wheel wake from Figure 5(a) to 5(c) shows good agreement.
As the backlight angle increases, in module 9 (=25), the interaction between the
roof separation and the C-pillar vortices increases. This is shown in Figure 6(b),
where the upper lobes of the C-pillar vortices are just slightly apart, distanced by
the roof separation. In Figure 6(c) the downwash from the C-pillar vortices reduces
the signature of the roof separation. In module 5 (=37), two-dimension separation
but the separation from the outboard edges can still be seen, in Figure 7(a), (b).
The yaw from experiment is not captured in the flow field predicted numerically,
since this was not accounted for in the simulation.
Predicted lift and drag are compared with experiment in Figure 8. Both qualitative
and quantitative values match well.
3.2 Modifications to underbody detail
We now consider the effect of modifying the underbody for a fixed upperbody twobox form model. In the model selected, the backlight angle is high and twodimensional separation occurs from both the roof and C-pillars. The benefit of
optimising the underbody during aerodynamic development can be significant,
allowing for potential savings of, perhaps 20-30 counts (i.e. 0.020-0.030 in drag
coefficient). Additionally, the underbody can have a significant effect on the lift
profile. However, since obtaining detailed experimental flow structure information in
the underbody region can be awkward, CFD can be of great assistance in
understanding these local flow structures, and in determining what changes might
be considered to meet design criteria.
3.2.1 Vehicle geometries
The underbody model variants are for the Renault CLIO with production level
complexity, including wing mirrors, inset glass, detailed bumpers and body styling.
Figure 9 shows the baseline geometry and the variations in underbody for each of
the models, ranging from highly detailed to completely smooth. Clio A has the most
realistic geometry with detailed wheels and suspension, wheel housings, a full

exhaust line, fuel tank and spare tyre. Clio B has flat underbody panels that do not
cover the exhaust line, and Clio C has the smoothest underbody geometry.
3.2.2 Experimental results
The experimental results have been obtained using a two fifth scale model. Figure
10 shows smoke visualisations of Clio A. Clear separation is shown on the upper
surface, and a clean separation line might be expected on the backlight at this
angle, as is demonstrated in the CFD visualisation. Figure 11 shows experimental
drag and lift. A clear drop in drag is observed as underbody complexity is reduced
suggesting little change in upper body flow structure due to the differing underbody
designs. Changes in the lift are more pronounced. However, Clio Bs lift is lower
than Clio A, due to reduced rear axle lift caused by the smoother geometry around
the rear axles. Clio C has the lowest lift of and this is again due to the smoother
surface reducing underbody pressures.
3.2.3 Comparisons to CFD
All three Clios configurations have been simulated using PowerFLOW. Figure 11
shows the comparison of experiment with prediction. The trend in both drag and lift
is correctly captured. Numerical analysis shows that despite the close absolute
values of drag for Clio B and Clio C, the force integration plot of Figure 12 shows
that Clio B has significant differences from Clio C in the way in which the total drag
of the vehicle is created.
Initially, flow stagnates on the nose of the geometries, leading to a sudden large
increase in drag; this can be observed between x=-1.5 and x=-1.4. At the point of
flow acceleration around the leading edge of the bonnet, extremely low pressures
lead to a suction which causes the integrated drag to fall. At the point of the front
wheels, geometry differences in the wheel housing of Clio B lead to a significantly
higher drag than the smooth underbody geometry of Clio C. More detailed analysis
of this section of the flow would lead to potential design changes which could
reduce the overall drag. For example, rounding the wheel housings would reduce
separations immediately downstream of the wheel. The extra drag caused by the
wheel housings is marked as a constant difference between the cumulative drag of
Clio B and Clio C along much of the remaining body length, till x=0.75.
A few features may be noted in this section. Firstly, cumulative drag drops at the
top of the windscreen, again because of pressure suction from accelerated flow.
Additionally, cumulative drag increases only slightly between x=-0.25 and x=0.75,
even though Clio B has some considerably complex geometry in this region. This
indicates, perhaps counter-intuitively, that design modifications in this section are
likely to be of little benefit. At x=0.75, the pressure recovery begins. A significant
discontinuity occurs just upstream of the rear wheel, at x=0, which is possibly
caused by end of the exhaust channel where flow is stagnating. This has a
downstream effect which, though minimised, appears to lead to the increased drag
in Clio B compared with Clio C.
The lift of Clio A is considerably different to that of Clio B and Clio C. The force
integration plot in Figure 12(b) indicates that the difference comes purely from the

backend at x=1.0. At this point Clio A has a fuel tank, which has a guard
immediately upstream of it. It appears that flow is stagnating on this guard and
generating considerable lift.
4 Conclusions
Good agreement is shown between experiment and PowerFLOW. The bulk flow
characteristics in the flow field of two-box models in Renaults test suite are
captured well, across a range of different backlight angles. PowerFLOW is seen to
capture the balance between relative strength of C-pillar vortices with the strength
of the roof separation. Additionally drag and lift trends for changes in underbody
detail are predicted.
5 Acknowledgements
Exa would like to thank SGI for their assistance in the preparation of this paper.
References
[1] Chen, .H., Teixeira, C. and Molvig, K., Digital Physics Approach to
Computational Fluid Dynamics: Some Basic Theoretical Features. International
Journal of Modern Physics C, (1994), 8, 4, 675-684.
[2] Mallick, S., Remondi, S., Chen, H., Pervaiz, M., Developing CFD through
Validation of Turbulence Models, MIRA Vehicle Aerodynamics Conference (2000).
[3] Lietz, R., Mallick, S., Kandasamy, S., and Chen, H., Exterior Airflow
Simulations Using a Lattice Boltzmann Approach, SAE 01-0596 (2002).
[4] Lietz, R., Pien, W. and Remondi, S., A CFD Validation Study for Automotive
Aerodynamics, SAE 00PC-229, (2000).
[5] Chen, H., Teixeira C., Molvig, K., Realization of Fluid Boundary Conditions via
Boltzmann Dynamics, Int. J. Mod. Phys. C, (1998), 9, 8, 1281-1292.
[6] Teixiera, C.M., Pervaiz M.M., Two Equation Turbulence Modeling with the
Lattice Boltzmann Method, Proc. ASME PVP Division Conf., 2nd Int. Symp.
Computational Techniques for Fluid/Thermal/Chemical Systems with Industrial
Applications, Aug 1-5, 1999 Boston, MA, USA.
[7] Yakhot, V., et al, A New Approach to Modelling Strongly Non-Equilibrium,
Time-Dependent Turbulent Flows (Internal Document).
[8] Duncan, B.D., Sengupta, R., Mallick, S., Shock, R., Sim-Williams, D.B.,
Numerical Simulation and Spectral Analysis of Pressure Fluctuations in Vehicle
Aerodynamic Noise Generation, SAE 01-0597 (2002).
[9] Hucho W.H., Aerodynamics of Road Vehicles, 4th ed., Butterworth & Co.,
1987.
[10] Ahmed, S.R., Ramm, G., Faltin, G., Some Salient Features of the TimeAveraged Ground Vehicle Wake, SAE 840300, (1984).
[11] Morel T., Aerodynamic Drag of Bluff Body Shapes Characteristic of HatchBack Cars, SAE 780267 (1978).
[12] Bearman, P.W., Near Wake Flows Behind Two-Dinemsional and ThreeDimensional Bluff Bodies, J. Wind. Eng. 69-71 pp33-54, (1997).

(a)

(b)

(c)

Figure 1: Baseline geometry (with wake survey locations at 0.010m, 0.280m and
0.560m downstream of geometry). The four angles tested are shown. Positions of
experimental probes are also shown.

Figure 2: Experimental and computational Cp results for module 9 (=25, left) and
5 (=37, right). C-pillar vortices in left image are indicated by drop in outboard
pressure. Fully separated flow in right image shows little variation across span.

Figure 3: Oil streamlines for modules 1 (=16, top), 9 (=25, middle) and 5 (=37,
bottom). Top and middle images show fully attached flow. Bottom image shows full
separation.

(a)

(b)

0.1

(c)

1.0

Note: Three sets of images are shown


for each position (a), (b), and (c) as
defined in Figure 1. Each set of images
shows experiment (upper) and
PowerFLOW (lower). This squareback
geometry shows separations (on the
upper body) which are relatively clean.
The influence of the hip is evident at
position (c).

Figure 4: Module 7 (=0) wake survey, at three positions.

(a)

(b)

0.1

(c)

1.0

Note: For a small backlight angle, a


weak C-pillar vortex forms. This is
shown at position (b) and there is a
good match between computation and
experiment. There is a slight difference
between the strength of the vortex in
experiment and computationally, evident
at position (c).
Figure 5: Module 1 (=16) wake survey

(a)

(b)

0.1

1.0

Note: Interaction between roof


separation and C-pillar vortices is
stronger for the module 9 (=25) than
the module 1 (=16). This is evident at
position (b), where the upper lobes of
the C-pillar vortices are slightly apart.
There is good match between numerical
and experimental wake surveys.
Figure 6: Module 9 (=25) wake survey

(a)

(b)

0.1

1.0

Note: The backlight angle is now past


the critical angle, and the flow fully
separates. The yaw found in experiment
is not captured in the computation
solution, since this was not modelled.

(c)

Figure 7: Module 5 (=37) wake survey

0.4

0.2

0.302
0.294

0.502
0.281

0.4

0.099

0.5

0.68
0.51

0.581
0.621

0.591
0.626

0.545
0.548

0.646
0.613

0.6
ACd(Expt-new)
ACd(3.4)

ACl

0.3
Mod7

Mod1

Mod9

Mod5

-0.2

Geometry

Mod 7

Mod 1

-0.073

ACd

0.6

0.8

Mod 9

Mod 5

Geometry

Figure 8: Modules Forces comparisons between experiment and PowerFLOW.


Drag factor comparison is in left image, lift factor comparison is on the right.

Clio A

Clio B

Clio C

Figure 9: Clio baseline geometry and underbody configurations. Clio A has the
most detailed underbody, and Clio C is the least detailed.

Figure 10: Upper body flow field of Clio

ACl (Exptnew)
ACl (3.4)

ACd(3.4)

0.400

0.300

ACl

ACd

ACd(Expt-new)

0.500

0.200
0.100

0.300

0.127
0.090

0.400

0.187
0.160

0.500

0 298
0 385

0.555
0.588

0.600

0.577
0.600

0.632
0.625

0.700

ACl(Expt-new)
ACl(3.4)

0.000

ClioA

ClioB
Geometry

ClioC

ClioA

ClioB

ClioC

Geometry

Figure 11: Force comparisons for Clio. Drag factor is in left image, lift factor is on
right.

(a)

(b)

Figure 12: Forces integration for Clio configurations. Left image shows the
integration of drag factor for the three cases. Clio B and C have similar total
drag factor, but this is generated from different mechanisms. Right image
shows integration of lift factor. Clio A has a very different lift to that of Clio B
or C. This comes from a component located at x=0.8, which turns out to be a
guard for the fuel tank.

Potrebbero piacerti anche