AIAA 2001-0423

A Single Element Wing in Ground

Effect; Comparisons of Experiments
and Computation
Jonathan Zerihan and Xin Zhang
School of Engineering Sciences,
University of Southampton,
Southampton SO17 1BJ, U.K.

39th AIAA Aerospace Sciences

Meeting and Exhibit
January 811, 2001/Reno, NV
For permission to copy or republish, contact the American Institute of Aeronautics and Astronautics
1801 Alexander Bell Drive, Suite 500, Reston, VA 201914344

A Single Element Wing in Ground Effect;

Comparisons of Experiments and Computation
Jonathan Zerihan and Xin Zhang
School of Engineering Sciences,
University of Southampton,
Southampton SO17 1BJ, U.K.
A computational study has been performed in order to model the flow around an
inverted aerofoil in ground effect. The method used is solutions of Reynolds-averaged
Navier-Stokes equations with turbulence modelled by Spalart-Allmaras model and k model. The results are compared to measured surface pressures and LDA results
taken at the centre of a wing in ground effect. Major features of the flow are captured.
The results yield good qualitative trends for the aerodynamic performance, using the
one-equation model when the surface pressures are compared at different heights. In
general, the wake thickness is predicted reasonably well in the region near to the trailing
edge. Further downstream, the wake is predicted to be thicker than that found in the
experiments, with reduced velocities. The ground boundary layer is predicted well using
the one-equation model, but is significantly too thick using the two-equation model.

Nomenclature
c
= wing chord; 223.4mm.
CL = three-dimensional downforce coefficient, L/q c;
force directed to ground.
Cl = sectional lift coefficient.
CP = pressure coefficient, p/q .
h
= height above ground; distance between suction
surface and ground plane at zero incidence.
udef = wake velocity deficit, ue um .
ue = edge velocity.
U = freestream velocity.
um = minimum velocity.
u, v = velocity components in x, y axes system.
x, y = Cartesian coordinates, x +ve downstream,
y +ve up, relative to wing leading edge.
Greek Symbols

top , bot
99

=
=
=
=

incidence.
top and bottom of the wake.
wake thickness defined by 0.99ue .
normal distance from suction surface
at trailing edge.

Introduction
HE front wing of a racing car acts in ground
effect, and contributes to about 25-30% of the
total downforce of the car, which, at top speed, may
approach three times the weight of the car. The downforce, or aerodynamic grip, is used in conjunction with

Formerly Ph.D Research Student, currently Aerodynamicist

British American Racing
Professor, School of Engineering Sciences. Senior Member
AIAA.

c 2001 by Xin Zhang. Published by the American

Copyright
Institute of Aeronautics and Astronautics, Inc. with permission.

the mechanical grip to improve the acceleration, braking, and cornering speed of the car. Of contemporary
interest is the level of downforce that the wing generates at different heights from the ground, not only
for optimising the height of the wing, but also for effects of suspension movements over bumps and as the
car accelerates and brakes. In addition to the aerodynamic performance of the front wing, another very
significant issue is the wake that it generates. The
flow to the undertray and diffuser in particular, but
also the radiators and rear wing, is severely effected
by the front wing because they all operate in the wake
from the wing.
Although numerous experimental and computational studies have been performed on aircraft type
wings in ground effect, i.e. with the pressure surface
nearest to the ground plane, for general aeronautical applications in addition to ground effect aircraft
(WIG), little data has been presented for racing car
type wings in ground effect, with the suction surface
lowermost.
Katz used a panel method to investigate the flow
around a downforce producing wing.13 Results from
a panel method4 and a Reynold-averaged Navier-Stoke
(RANS) solver5 modelling the entire car including the
front wing have also been published.
Knowles et al6 conducted experiments in a wind
tunnel with a moving ground facility. A single element GA(W)-1 wing was tested at a variety of incidences for a range of heights from out of ground
effect down to 0.12c from the ground. Force results
show that the downforce generated increases as the
ground is approached, for all incidences tested at. It
was observed that the lower surface suction increases
in ground effect, and that very close to the ground,

1 of 12
American Institute of Aeronautics and Astronautics Paper 2001-0423

stalled flow occurs over the rear portion of the wing.

In addition to their experimental results, a selection
of two-dimensional panel method results are presented
for comparison. Results compare reasonably at large
heights or low incidences, but are poor closer to the
ground.
Ranzenbach and Barlow have presented results from
experimental tests in a fixed ground wind tunnel comparing them to computation results using a RANS
solver.710 A single element aerofoil was tested at
different heights from the ground, at a single incidence. As in previous research, it was found that the
downforce increased as the height from the ground reduced. However, at a ground clearance of about 0.1c,
the downforce was found to reach a maximum, below
which the downforce dropped sharply. They called this
the force reduction phenomenon. Only overall force results are presented from their experiments; no surface
pressures or flowfield data is available. The force reduction phenomenon is modelled reasonably well in
the computational results close to the ground. Using vorticity plots, it is put forward that the force
reduction phenomenon is due to a merging of the wing
and ground boundary layers. The boundary layers are
shown to merge upstream of the trailing edge of the
aerofoil for a small ground height, below the force reduction phenomenon. The presence of the boundary
layers reduces the flow velocity between the suction
surface and the ground, leading to an increase in pressure, and hence a reduction in downforce. Extending
the computational study to the correct ground conditions with a moving ground yields the result that
the downforce produced at different ground heights
is qualitatively similar to the fixed ground case, although the magnitude of the downforce and also the
height at which the force reduction phenomenon occurs is greater for the moving ground case.
It is believed that tests with a fixed ground are of little practical use, and that this affects the results more
significantly than an aeronautical wing in ground effect, due to the fact that the suction surface is nearest
to the ground, which contributes to the downforce significantly more than the pressure surface.
As part of a detailed investigation into wings in
ground effect, Zerihan and Zhang1114 have used experimental tests in a wind tunnel with a moving ground
facility to further investigate the effect of the ground.
In addition to overall forces and surface pressures, experiments were also performed using laser Doppler
anemometry (LDA) and particle image velocimetry
(PIV) methods to examine the wake flow. This paper compares computations using a RANS solver with
the results from the experimental database.

Description of Experimental Results

Details of the aerodynamic performance of the wing
in ground effect are available in reference,11 together

with discussions of the mean and unsteady flow features of the wake at the semispan of the wing12 and
a further study describing the three-dimensional effects.13 Initial tests were performed with free transition. However, for the purpose of CFD modelling,
transition was tripped at x/c = 0.1 on both surfaces,
using 100 grit. The downforce at = 1 for different
heights is presented in Fig. 1, for both free and fixed
transition cases. The effect of the ground is to constrain the flow beneath the suction surface. At a large
height in ground effect, the flow is accelerated over
the suction surface to a slightly greater level than in
freestream, resulting in greater suctions on the suction surface. As the wing is brought closer to the
ground, the flow is accelerated to a higher degree, causing an increased peak suction, and associated pressure
recovery. At a height where the pressure recovery is
sufficiently steep, the boundary layer separates at the
trailing edge of the suction surface. For the transition
free case, the height at which boundary layer separation was first observed was h/c = 0.224. As the
height is reduced beyond this, the wing still generates more downforce, but the rate of increase slows,
and the downforce reaches a maximum, the downforce reduction phenomenon. Below this height the
downforce reduces. As the height is reduced from
the first height where flow separation was observed,
the separation point moves forward steadily. At the
maximum downforce, the boundary layer separates at
x/c 0.8, for the free transition case. Heights greater
than the maximum downforce are known as the force
enhancement region. Below the maximum downforce
is known as the force reduction region. Similarities
can be drawn comparing the reduction of the height of
a wing above the ground, with the increase of the incidence of a wing in freestream. In both cases, the
pressure recovery becomes steeper, eventually causing boundary layer separation, and the wing stalls.
The effect of fixing transition is to reduce the magnitude of the downforce, and to increase the height at
which the force reduction phenomenon occurs. It has
been shown that the maximum downforce occurs when
small gains in suction on some portions of the lower
surface and small reductions in suction on other portions of the lower surface, together with other small
reductions in pressure on the upper surface with a reduction in height, fail to contribute to an increase in
overall downforce.11, 13 Boundary layer merging was
not observed.13
The wind tunnel tests were performed using a single element wing with endplates, of aspect ratio approximately 5. A significant portion of quasi-twodimensional flow exists in the central portion of the
wing. Two-dimensional computations are used to simulate the flow at the semispan of the wing, comparing
pressures and LDA results obtained at the wing centre.
Details of the experimental results, together with the

2 of 12
American Institute of Aeronautics and Astronautics Paper 2001-0423

1.8

0.6

1.7

0.4

1.6
0.2

Transition fixed
Transition free

1.4

CL

y/c

1.5

1.3

-0.2

1.2

-0.4

1.1
-0.2

0.2

0.4

0.6

0.8

1.2

x/c

a) h/c = 0.179.
0.9
0.8
0

0.1

0.2

0.3

0.4

0.5

0.6

0.7
0

h/c
Fig. 1
Measured downforce for threedimensional wing.

y/c

-0.1

wing profile used can be seen in the previous work.

All wind tunnel tests were performed at a Reynolds
number of approximately 460000, with a freestream
velocity of 30m/s.

-0.2

-0.1

0.1

0.2

0.3

x/c

Computational Modelling

b) Leading edge region: h/c = 0.179.

Governing equations
0.1

y/c

The numerical model solves the three-dimensional

thin layer compressible Navier-Stokes equations in a
time dependent manner. The effect of turbulence was
modelled by either the the Spalart-Allmaras (S-A) one
equation model,16 and the Menter k model.17 The
code used was CFL3D, an implicit upwind code.15

-0.1

Grid strategy
-0.2

0.7

0.8

0.9

1.1

1.2

x/c

c) Trailing edge region: h/c = 0.179.

0.6

0.4

0.2

y/c

In order to generate the two-dimensional structured

grids at different heights from the ground, Gridgen was
used. Substantial grid refinement tests were performed
on the circumferential spacing at the aerofoil leading
edge, trailing edge, and the suctions and pressure surfaces, and on the normal spacing of the grid points at
the aerofoil surface and at the ground plane.
After preliminary tests on the gridding strategy were
performed, it was decided to use a solution that required neither grid patching nor grid overlaying. Although not all the features of the grid are ideal, this
was felt as the best compromise. The grid, for a height
of h/c = 0.179 can be seen in Fig. 2(a). This is shown
in more detail for the region near to the aerofoil leading and trailing edges in the same figure (Fig. 2(b)
and 2(c)), together with an example of another grid
at h/c = 0.313 (Fig. 2(d)). The upstream and downstream boundaries are located at 8c from the aerofoil.
The boundary above the wing is at 7c, which corresponds to the approximate location of the wind tunnel
roof. The computational domain contains about 30000
grid points in total, according to ground height.

-0.2

-0.4
-0.2

0.2

0.4

0.6

0.8

x/c

d) h/c = 0.313.
Fig. 2

Computational meshes.

3 of 12
American Institute of Aeronautics and Astronautics Paper 2001-0423

1.2

The grid contains five blocks in total. The first block

is a square C-type grid, around the aerofoil. The grid
cells at the corners of the grid are not ideal, but this
grid approach was the best compromise. A y + 1
was used for the first grid point normal to the aerofoil surface. The grid is clustered near to the aerofoil
leading and trailing edges. It was found that more
grid points were required on the suction surface compared to the pressure surface. Effort was made to force
a perpendicular grid near to the trailing edge region
of the suction surface. The aerofoil features a finite
trailing edge, corresponding to 0.007c. Block 2 is a
H-type grid covering the region from the trailing edge
to the downstream boundary; a closed wake type grid.
Block 3 is an H-type grid, extending from block 1 vertically to the top of the computational domain. Block
4 grids the upstream area from block 1 to the upstream
boundary, and is H-type. Block 5 is H-type, and covers the entire ground plane from the upstream to the
downstream boundaries. The reason that block 5 is
required is for the fine grid spacing necessary on the
ground plane; a y + 1 was also found to be required
for the first grid point from the ground.
In order to simulate the flow around an aircraft wing
at varying angles of attack, the same grid is generally
used, specifying a different incidence in the CFD input
datafile. For the configuration of a wing at different
heights in ground effect, a new grid has to be generated
for each height from the ground, which is a tedious process. The geometry in the database file was modified,
and blocks 1, 3, 4 and 5 had to be regenerated. For a
reduction in ground height, it was required to reduce
the circumferential size of the C-type grid around the
aerofoil, in addition to moving the ground grid vertically up relative to the aerofoil.
Boundary Conditions

The upstream and downstream and downstream

boundaries are modelled with the inflow/outflow
boundary condition. The top boundary uses an extrapolation simulation. The no-slip condition is applied to the surface of the aerofoil. The ground plane
has been modelled using a prescribed velocity equal to
freestream.
Turbulence Modelling

Initial tests at a single height and in freestream were

performed using two turbulence models; the SpalartAllmaras (S-A) one equation model,16 and the Menter
k model,17 with a view to establish the difference
in the performance and the flowfield of the models.
Previous studies have found the S-A model to predict
attached flows and two dimensional separated flows
well using CFL3D.18 The k model has been shown
to be very capable compared with the S-A model for
separated flows.19
The comparison of results at a single height in
ground effect were inconclusive as to which model per-

formed better, and due to the significant differences

found, it was decided to use both models for each
height.
Solution Process

It was found to be extremely difficult to obtain a

solution. When a steady solution was performed, this
yielded results with unsteady periodic characteristics.
However, an unsteady solution would generally converge on a single steady result.
The solution process used started with a steady
run. Mesh sequencing was employed, performing the
analysis on three grid levels, for convergence acceleration. Multigrid was then used for the remainder of
the steady and unsteady solutions. The output from
the steady solution was used as a restart for the next
stage, the unsteady solution. It was not possible to
start with an unsteady solution; the semi-converged
steady results were required to avoid the solution from
blowing up. The S-A model was found to be more robust than the k model, although this depended on
the exact characteristics of the grid and the flowfield.
For the k model, it was necessary to perform two
unsteady analyses consecutively; the first with a small
time step, and the second with a larger time step. The
S-A model only needed the run with the larger time
step.
Convergence problems using compressible solvers at
low Mach numbers are well documented.20 The combination of this, and the fact that a time-accurate
solution method is required, leads to long solution
times.

Results
The results of the computations are compared in
terms of the aerodynamic performance, and the flowfield results.
Aerodynamic Performance

Freestream
Both turbulence models give very similar results in
freestream, as can be seen by the pressure distributions
in Fig. 3.
The first point to make is that the computational
results show a large spike near to the leading edge of
the suction surface. The experimental results for this
case do not show this spike. This may either be due to
the spacing between the tappings, or because it is not
present for this particular experimental configuration.
Note that many other experimental results do show the
presence of this spike, for example the single element
wing at other incidences,11 the Gurney flap results,14
and more prominently in the double element results,
yet to be published. The magnitude of the spike is
sufficiently large that it is slightly greater than the
main suction peak, near to x/c = 0.16. The suction
peak is overpredicted a little by the computation. In
the pressure recovery region, however, from x/c 0.25

4 of 12
American Institute of Aeronautics and Astronautics Paper 2001-0423

0.5

CP

-0.5

-1

CFD k- model
CFD S-A model
Experiment

-1.5

-2
0

0.5

x/c

Fig. 3
Comparison of computational
and experimental surface pressures in
freestream.

to the trailing edge of the suction surface, the pressures

compare extremely well with the experimental values.
Near to the leading edge of the pressure surface,
stagnation is predicted well. The flow is then accelerated rapidly. At x/c = 0.03, the experimental results
show the flow to have reached a velocity, which remains approximately constant for a large portion of
the surface, before a gradual retardation starting at
x/c = 0.5. But, the computational results show a
small, but significant deceleration at x/c = 0.03 to
CP = 0.25. The pressures are then overpredicted by
the computation, with the difference decreasing in the
streamwise direction. From x/c 0.6, the difference is
small, and the remainder of the computational results
compare well with the experimental values.
The Cl for the aerofoil was found to be 0.885 for
the S-A model and 0.872 for the k model. An
estimation of the sectional downforce Cl from the experiments can be obtained by integrating the surface
pressures. However, it is stressed that this is only an
estimate, and will be used to compare the qualitative
trends between the experimental and computational
results, not as a quantitative comparison. The experimental results give Cl = 0.766. This is a difference
of approximately 16% between the computational and
the experimental results for the S-A model, and 14%
the k model. In addition to the difference in the
surface pressures at discrete points, this approximately
15% difference is also due to the leading edge suction
spike, and the linear interpolation of the results using
a large number of points, computationally about 200,
with the relatively coarse spacing of the 45 experimental results.

Ground Effect
Tabulated results of the downforce predicted by the
CFD, together with Cl from the integrated pressures
are given in Table 1 for all heights. The comparisons
are only to highlight the trends, not the outright values. Each of the surface pressure distributions will be
discussed below.
The distributions at h/c = 0.671 show the results
for the largest height in ground effect, Fig. 4(a) The
ground effect has increased the lower surface suctions,
in addition to a slight reduction in the upper surface pressures, which overall leads to an increase in
downforce. Again, the leading edge suction spike is of
a similar magnitude to the main suction peak. The
main peak is overpredicted by the computations, as
in freestream. However, the suctions throughout the
pressure recovery are also overpredicted by the computation, unlike the freestream results. Again on the
pressure surface, the pressures are too great from the
spike at x/c = 0.03 to about the mid-chord. Further
downstream, they compare better. The computations
predict Cl = 1.052 (S-A) and 1.030 (k ), compared
to the experimental value of 0.902, which is a similar
difference that in freestream, of 17% and 14%. The
very slightly greater downforce from the S-A model
compared to the k model can be seen to be due to
increments in the loading predicted on each surface.
The same general trends can be observed at h/c =
0.448, Fig. 4(b). The overprediction of the lower surface suctions has increased slightly, over the whole of
the surface. Although the suction at the leading edge
spike has increased, that at the main peak increases at
a greater rate such that it can be seen that the largest
suction is that at the mean peak. The slight increase
in the lower surface suction prediction is offset by a
small reduction in the overprediction in the pressure
surface results, resulting in computational results for
Cl are 16% and 13% greater than the experimental
value.
As the height is reduced to h/c = 0.313, not presented here, the previous trends of slight overprediction of lower surface suctions for the S-A model, and
underprediction of the k suctions continues. This
yields predictions of Cl that are 15% greater (S-A) 9%
greater (k) than the experimental results (Table 1).
Similarly for h/c = 0.224, Fig. 4(c), the overprediction in lower surface suctions remains approximately
constant for the S-A model, compared with the previous height. However, this reduces for the k
model, such that most of the suction surface results
are mapped very well for this example. For both models, nearly all pressure surface results are mapped well,
too. The effect is that the S-A model gives Cl = 1.475,
again 15% greater than the experimental prediction of
Cl = 1.286. The k model gives a closer result
of Cl = 1.352, just 5% greater than the experimental value. The same trend was found at h/c = 0.179,

5 of 12
American Institute of Aeronautics and Astronautics Paper 2001-0423

0.5
0.5
0
0

CP

CP

-0.5

-0.5

-1

-1.5
-1

CFD k- model
CFD S-A model
Experiment

CFD k- model
CFD S-A model
Experiment

-2

-1.5
-2.5

-2

-3
0

0.5

x/c

0.5

x/c

a) h/c = 0.671

b) h/c = 0.448
1

0.5

0.5

0
0
-0.5
-1

CP

CP

-0.5

-1

-1.5
-2

-1.5
CFD k- model
CFD S-A model
Experiment

-2

CFD k- model
CFD S-A model
Experiment

-2.5
-3

-2.5

-3.5

-3

-4
0

0.5

x/c

c) h/c = 0.224
Fig. 4

0.5

x/c

d) h/c = 0.134

Comparison of computational and experimental surface pressures.

where Cl for the two turbulence models are 15% (S-A)

and 2% (k ) greater than the experimental values.

5% underprediction in Cl from the S-A and the k

models respectively.

At h/c = 0.134, the greatest experimental Cl was

found. Again, the S-A model overpredicts the lower
surface suctions, (Fig. 4(d)) by a similar amount to the
previous height. The k model starts to underpredict the suctions very slightly, especially in the region
of the suction peak. On the pressure surface, the flow
is mapped well by both models, the S-A model giving
a slightly higher pressure near to the leading edge region. The S-A model gives Cl = 1.604, 16% greater
than the experimental value of Cl = 1.385, whereas
the k model predicts Cl = 1.386. Results at the
lowest height of h/c = 0.090 show the same trend, but
more extreme in nature; there is a similar overprediction of the lower surface suctions from the S-A model,
but the k model is underpredicting the suctions
more severely. This gives a 12% overprediction and a

Flowfield

Results from the computational database were extracted in order to compare with the experimental
wake and boundary layer surveys performed with the
LDA system.
Boundary layer
In obtaining the LDA results close to the wing surface, difficulties were experienced. The finite size of
the measurement volume coupled with flare from the
surface implied that it was impossible to obtain results
very close to the surface of the wing. For simplicity,
the surface, at which = 0, was defined as the last
point at which zero data was obtained. In this manner, the LDA results are a finite distance below their
true location. It is believed that the first 1mm approx-

6 of 12
American Institute of Aeronautics and Astronautics Paper 2001-0423

imately from the wall was void of any data. This is

a rough estimate. Hence, the LDA profiles should be
raised by, very approximately, 0.005c. This has not
been performed, due to the uncertainty in the size of
the correction required.
Fig. 5(a) shows the computational results with the
LDA results, for u/U at h/c = 0.224. The two turbulence models give different results within the boundary
layer. It can be seen that, very close to the surface,
the S-A model has a negative velocity, as the flow separated. Although this is not clear from the profile for
the k case, examination of the skin friction on the
wall gives a negative value also for the k case,
as the flow has also separated. Through the boundary layer, for a particular distance from the surface,
the velocity is consistently lower for the S-A model,
until it reaches the edge. Although within the boundary layer, the experimental results lie closer to those
for the k model, given the problem with obtaining experimental results near to the surface, and the
shape of the profile, it is difficult to say which model
gives a better match. Outside the edge of the boundary layer, u/U is approximately 0.03 greater for the
experimental results than for the computation.
At h/c = 0.134, Fig. 5(b), both computational results clearly show a separated boundary layer. In a
similar manner to the results at h/c = 0.224, the S-A
model gives a thicker boundary layer than the k
model. Again, it would appear that, although the experimental points lie closer to the k results, it
is difficult to say which model gives the best results.
Both models fail to predict the magnitude of the velocity in the recirculation region. Again, beyond the
edge of the boundary layer, both turbulence models
give results that are less than the experimental results,
by u/U 0.05.

0.12

0.1

/c

0.08

0.06

0.04

0.02

0
-0.2

0.2

0.4

0.6

0.8

1.2

0.8

1.2

u/U

a) h/c = 0.224.
0.12

0.1
CFD k- model
CFD S-A model
Experiment

/c

0.08

0.06

0.04

0.02

0
-0.2

0.2

0.4

0.6

u/U

b) h/c = 0.134.

Nearfield wake
To investigate the wake results, comparisons are
made at two streamwise locations. At x/c = 1.2, the
results are compared with velocities extracted from the
LDA grid results mapping the trailing edge region. At
x/c = 1.5, the results are compared with the LDA
wake surveys. Note that the dedicated wake surveys
included a finer distribution of points the direct wake
from the wing, and also additional points very close
to the ground, to map the ground boundary layer.
Tabulated results for the wake thickness from the computations and experiment are presented in Table 2 for
x/c = 1.5. Information on the ground boundary layer
is also given, in Table 3 at the same streamwise location.
At a height of h/c = 0.448, the wake surveys at
x/c = 1.2 are compared in Fig. 6(a). Outside of the
edge of the wake, the computations point to velocities
that are u/U 0.025 lower than the experimental
values. The results within the wake are difficult to

CFD k- model
CFD S-A model
Experiment

Fig. 5 Comparison of computational and

experimental boundary layer surveys at
suction surface trailing edge.

compare to a high degree of accuracy due to the coarse

spacing of the experimental points. The S-A model
gives a slightly greater minimum velocity at the centre of the wake. The wake thickness appears similar,
for the computations and the experiment. It is difficult
to be more precise, due to the coarse spacing of the experimental results, and the increased velocity outside
of the wake for the experimental results. The main
difference between the two models appears in their
ability to model the ground boundary layer. Although
few experimental points are available at this streamwise location, it is clear that the k model gives a
ground boundary layer significantly thicker than the
S-A model and the experiments.
Results are similar at x/c = 1.5 for the same height,

7 of 12
American Institute of Aeronautics and Astronautics Paper 2001-0423

0.2

0.2

0.1
0.1

0
0

-0.2

y/c

y/c

-0.1

CFD k- model
CFD S-A model
Experiment

-0.3

-0.1

CFD k- model
CFD S-A model
Experiment

-0.2

-0.4
-0.3

-0.5

-0.4

-0.6
0.5

0.6

0.7

0.8

0.9

1.1

0.3

1.2

0.4

0.5

0.6

0.7

0.8

u/U

u/U

a) x/c = 1.2.

a) x/c = 1.2.

0.9

1.1

1.2

0.2

0.2

0.1
0.1
0
0
CFD k- model
CFD S-A model
Experiment

-0.2

y/c

y/c

-0.1
CFD k- model
CFD S-A model
Experiment

-0.1

-0.3
-0.2
-0.4
-0.3
-0.5

-0.4

-0.6
0.7

0.8

0.9

0.6

1.1

0.7

0.8

0.9

1.1

u/U

u/U
b) x/c = 1.5.

b) x/c = 1.5.

Fig. 6 Comparison of computational and

experimental wake surveys at h/c = 0.448.

Fig. 7 Comparison of computational and

experimental wake surveys at h/c = 0.224.

Fig. 6(b). Outside of the wake, the velocities according

to CFD are u/U 0.03 lower than the experimental
results. The wake thickness is modelled well by the
computations (see Table 2). There is a difference in
the maximum velocity deficits of u/U 0.06, which
implies a little overprediction by the CFD in addition
to the difference found out of the wake. The results
for the k model show the ground boundary layer
to have grown compared to the previous streamwise
location (see Table 3). The thickness of this from the
S-A model is significantly closer to the experimental
values.
For the results at h/c = 0.224, Fig. 7(a) there is
a small overprediction of the wake thickness and the
maximum velocity deficit at x/c = 1.2. At x/c = 1.5,
the wake velocities are underpredicted more signifi-

cantly (see also Table 2). The differences between

the two turbulence models used are relatively small
compared to the difference between the computations
and the experiments in general. The ground boundary
layer for the k model is much thicker than at the
previous height, whilst the S-A model would appear to
give better results here regarding the thickness of the
layer (see Table 3). The minimum velocity, however,
is underpredicted a little, however this may partly be
due to the underprediction in the edge velocity.
For results at h/c = 0.134, which is at the maximum
downforce experimentally, and for the S-A model, but
below the maximum downforce for the k model, the
results are again similar, but more severe (see Fig. 8).
The main wake is now a little thicker than the experimental results at x/c = 1.2. The velocity at the

8 of 12
American Institute of Aeronautics and Astronautics Paper 2001-0423

0.2

the wake and the ground boundary layer. Results

are for the S-A turbulence model at the same heights
of h/c = 0.448, 0.224, 0.134. In addition, results at
h/c = 0.134 are also given for the k model, and
for the experimental LDA results. As the flow moves
downstream, it can be seen that, for a given height,
the wake thickens, and the velocities in the wake reduce due. A reduction in the height thickens the wake,
especially on the lower edge, where this is due to the increase of the suction surface boundary layer thickness.
The effect of using the k model is to massively overpredict the thickness of the ground boundary layer.
Both computational models give relatively similar results within the wake, however. Far from the trailing
edge, at x/c = 3.0, it can be seen that there is now a
significant difference in the velocities for the computational results compared with the experimental results.
Generally, the velocities within the wake are too low
for the computational results, and the thickness of the
wake is too large.

0.1

y/c

-0.1

CFD k- model
CFD S-A model
Experiment

-0.2

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1.1

1.2

u/U
a) x/c = 1.2.
0.2

Flow between wing and ground

The u/U velocity contours are also presented in
Fig. 10 for the flow between the wing and the ground,
for the same heights of h/c = 0.448, 0.224, 0.134 for the
S-A model, and h/c = 0.134 for the k model. The
reduction in the height of the wing from the ground results in the flow being accelerated to a greater extent,
as can be seen in the figure, and also the surface pressures earlier. Studying the results close to the ground
near to the peak in suction, it can be seen that the
ground boundary layer originates as the flow starts to
retard after the peak suction. Further analysis of the
results (not presented here) confirms this. The overprediction of the ground boundary layer thickness with
the k model is evident from near to its formation,
e.g. at x/c = 0.4. There is also a small difference in
using the k and S-A models in the velocity contours, which can be seen more clearly in the surface
pressure distributions. Using the k model gives
velocities which are lower than for the S-A model.

0.1

y/c

-0.1

CFD k- model
CFD S-A model
Experiment

-0.2

0.5

0.6

0.7

0.8

0.9

1.1

u/U
b) x/c = 1.5.
Fig. 8 Comparison of computational and
experimental wake surveys at h/c = 0.134.

wake centre is closer with the S-A model, but both

models show an incorrect upward shift in the wake,
compared to the LDA results. At x/c = 1.5, the velocities in the wake are significantly underpredicted.
The wake thickness is slightly underpredicted by both
models (see Table 2). For both streamwise locations,
k shows an increased ground boundary layer. From
the results at x/c = 1.5, it is apparent that the S-A
model gives a boundary layer that is a little to thin if
any, but which is massively overpredicts the velocity
deficits.
Farfield wake
A further comparison of the wake results is given in
Fig. 9, in which the u/U velocity contours are plotted for u/U < 0.99, to highlight the results within

Discussion
Examining the performance of the wing in
freestream shows that the pressure distributions are
modelled well, and the difference between the S-A and
the k models is insignificant. On the suction surface, the peak suction is overpredicted a little, but
the pressure recoveries compare extremely well. A
spike is prominent in the computational results, but
is not as apparent in the experimental results. On
the upper surface, the pressures are overpredicted a
little by the computations. Integrating the surface
pressures gives an experimental result of about 15%
less than the computations. Although this is partly
due to the small overpredictions in the pressures and
suctions over some portions of the upper and lower

9 of 12
American Institute of Aeronautics and Astronautics Paper 2001-0423

surfaces respectively, a significant contributory factor

is the coarser distribution of discrete points experimentally over which interpolation is applied, especially
considering the spikes that are present in the computations.
In ground effect, the lower surface suctions are constantly overpredicted a little by the S-A model. At
large heights, this is also the case for the k model.
However, the overprediction reduces as the height is reduced, and at h/c = 0.179, the predicted lower surface
suctions are too low. The underprediction of suction
increases still as the height is reduced. Generally, little
difference is found on the pressure surface, and results
compare well. The effect of this is that the S-A model
predicts the qualitative trend of the ground effect very
well. However, the k model does not show the correct trends, and at the closest height to the ground,
the overall loading is lower than the experiment.
The surface pressure distributions along with the
contour plots show that the difference using the two
models arises from the suction surface, and increases
as the height is reduced. The k model starts to give
reduced suctions compared to the S-A model, but still
greater than the experimental values, which are clear
from about h/c = 0.313. This effect is amplified such
that at small heights, the k results have reduced
suctions than the experimental results. This gives the
S-A results a greater adverse pressure gradient for the
pressure recovery region. The boundary layer profiles
seem to confirm this, with a thicker boundary layer for
the S-A results. It is difficult to confirm which model
generates the more realistic boundary layer profiles,
due to the problem in obtaining LDA results close to
the surface.
The major flaw with the k model is an overprediction of the ground boundary layer thickness. The
computational results confirm that this originates due
to the adverse pressure gradient after the peak suction from the wing. At a constant height close to the
ground, the k model gives a thick boundary layer,
and lower velocities in the region between the wing and
the ground. It would appear that the lower velocities
are a direct effect of the thicker boundary layer. It is
believed that this causes the difference in the predictive capabilities of the models.

Conclusion
Major physics of an aerofoil in ground effect are captured. Good qualitative results predicting the correct
trends have been obtained, in an attempt to model
a two-dimensional slice of a single element wing in
ground effect, using the S-A turbulence model. Comparing the results to integrated pressures from the
experimental study gives a constant difference, which
could be attributed to a number of factors in the model
tests. The pressure distributions appear accurate, the
differences partly due to the discrete spacing of exper-

imental points. There are deficiencies in modelling the

wake flow, and CFD predicts a larger wake, in terms
of thickness and velocity deficit as the streamwise distance is increased. The k model was shown not to
model the ground boundary layer correctly, resulting
in bad performance at low ground heights.

Acknowledgements
J. Zerihan is supported by an EPSRC studentship.
The authors would like to thank W. Toet of British
American Racing and C. Rumsey of NASA Langley
for their support and discussions.

References
1 J.

Katz, Calculation of the aerodynamic forces on automotive lifting surfaces, Transactions of the ASME: Journal of
Fluids Engineering, Vol.107, December 1985, pp.438-443.
2 J. Katz, Considerations pertinent to race-car wing design, Loughborough University Conference on Vehicle Aerodynamics, 1994, pp.23.1-23.7.
3 J. Katz, High-lift wing design for race-car applications,
SAE Publication 951976, 1995.
4 J. Katz, Aerodynamic model for wing-generated down
force on open-wheel-racing-car configurations, SAE Publication 860218, 1986.
5 J. Katz, H. Luo, E. Mestreau, J. Baum and R. L
ohner,
Viscous-flow simulation of an open-wheel race car, SAE Publication 983041, 1998.
6 K. Knowles, D. Donahue and M. Finnis, A study of wings
in ground effect, Loughborough University Conference on Vehicle Aerodynamics, 1994, pp.22.1-22.13.
7 R. Ranzenbach and J. Barlow, Two-dimensional airfoil in
ground effect, an experimental and computational study, SAE
Publication 942509, 1994.
8 R. Ranzenbach and J. Barlow, Cambered airfoil in ground
effect - wind tunnel and road conditions, AIAA Paper 95-1909,
1995.
9 R. Ranzenbach and J. Barlow, Cambered airfoil in ground
effect - an experimental and computational study, SAE Publication 960909, 1996.
10 R. Ranzenbach, J. Barlow and R. Diaz Multi-element airfoil in ground effect - an experimental and computational study,
AIAA Paper 97-2238, 1997.
11 J. Zerihan and X. Zhang, Aerodynamics of a single element wing in ground effect, AIAA Paper 2000-0650, AIAA
38th Aerospace Sciences Meeting, January 11-13, 2000. Also J.
Aircraft, No. 37, No. 6, November-December 2000, pp. 10581064.
12 J. Zerihan and X. Zhang, Turbulent wake behind a single
element wing in ground effect, 10th International Symposium
on Applications of Laser Techniques to Fluid Mechanics, Lisbon,
Portugal July 10-13, 2000.
13 J. Zerihan and X. Zhang, An experimental study of a
single element wing in ground effect, 3rd MIRA International
Vehicle Aerodynamics Conference, Rugby, UK October 18-19,
2000.
14 J. Zerihan and X. Zhang, Force enhancement of Gurney
flaps on a wing in ground effect, AIAA Paper 2000-2241, Fluids
2000, June 19-22, 2000.
15 J. Thomas, S. Krist and W. Anderson, Navier-Stokes computations of vortical flows over low aspect ratio wings, AIAA
Journal, Vol.28, No.2, Febraury 1990, pp.205-212.
16 P. Spalart, and S. Allmaras, A one-equation turbulence
model for aerodynamic flows, AIAA Paper 920439, 1992.
17 F. Menter, Zonal two equation k turbulence models
for aerodynamic flows, AIAA Paper 932906, 1993.

10 of 12
American Institute of Aeronautics and Astronautics Paper 2001-0423

0.2
0.1

y/c

0
-0.1
-0.2
-0.3

(a)

-0.4

0.2

0.4

0.6

0.8

1.2

1.4

1.6

1.8

-0.5
1

1.5

x/c 2

2.5

0.2

0.2
0.1

-0.2

y/c

y/c

0
-0.1

(b)

-0.2

-0.3
-0.4

1.5

x/c 2

2.5

-0.4
0.2

y/c

0.1
0
-0.1

(c)

0.2

0.4

1.5

x/c

2.5

0.6

0.8

1.2

1.2

x/c

-0.2
3

a) h/c = 0.448, S-A model.

0.2

y/c

0.1
0
-0.1

(d)

0.2

-0.2
1

1.5

x/c 2

2.5

3
0

y/c

0.2

y/c

0.1
0
-0.1

-0.2

(e)

-0.2
1

1.5

x/c 2

2.5

3
0

0.2

0.4

0.6

0.8

x/c
u/U:

0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95

b) h/c = 0.224, S-A model.

Fig. 9 u/U velocity contours within wake and
ground boundary layer; (a) h/c = 0.448, S-A. (b)
h/c = 0.224, S-A. (c) h/c = 0.134, S-A. (d) h/c = 0.134,
k . (e) h/c = 0.134 LDA measurements.
y/c

18 C. Rumsey and V. Vatsa, Comparison of the predictive

capabilities of several turbulence models, Journal of Aircraft,
Vol.32, No.3, May-June 1995, pp.510-514.
19 P. Godin, D. Zingg and T. Nelson, Highlift aerodynamic
computations with one- and two-equation turbulence models,
AIAA Journal, Vol.35, No.2, February 1997, pp.237-243.
20 Milholen, W.E., Chokani, N., and Al-Saadi, J., Performance of three-dimensional compressible Navier-Stokes codes
at low Mach numbers, AIAA Journal, Vol.34, No.7, July 1996,
pp.1356-1362.

0.2

-0.2

0.2

0.4

0.6

0.8

x/c

c) h/c = 0.134, S-A model.

y/c

0.2

-0.2

0.2

0.4

0.6

0.8

x/c

d) h/c = 0.134, k model.

Fig. 10 u/U contours within wake and
ground boundary layer.

11 of 12
American Institute of Aeronautics and Astronautics Paper 2001-0423

Table 1. Computational and experimental Cl

h/c
Cl expt Cl S-A Cl k ClSA Clk
f/s
0.766
0.885
0.872
16%
14%
0.671
0.902
1.052
1.030
17%
14%
0.448
1.009
1.174
1.136
16%
13%
0.313
1.145
1.320
1.253
15%
9%
0.224
1.286
1.475
1.352
15%
5%
0.179
1.362
1.565
1.393
15%
2%
0.134
1.385
1.604
1.386
16%
0%
0.090
1.371
1.539
1.297
12%
-5%

h/c
0.448

0.313

0.224

0.179

0.134

0.090

Table 2 Wake information at x/c = 1.5.

case
udef /U y at um y at top y at bot
exp
0.25
0.08
0.13
0.04
S-A
0.28
0.09
0.14
0.05
k
0.28
0.08
0.14
0.04
exp
0.25
0.08
0.13
0.04
S-A
0.31
0.08
0.13
0.03
k
0.32
0.08
0.13
0.03
exp
0.27
0.07
0.13
0.01
S-A
0.36
0.07
0.13
0.02
k
0.38
0.06
0.13
0.01
exp
0.28
0.05
0.12
-0.02
S-A
0.41
0.06
0.13
0.00
k
0.43
0.06
0.12
-0.01
exp
0.28
0.04
0.12
-0.05
S-A
0.45
0.05
0.13
-0.02
k
0.47
0.05
0.11
-0.03
exp
0.31
0.02
0.12
-0.10
S-A
0.55
0.03
0.13
-0.06
k
0.50
0.03
0.10
-0.07

99 /c
0.09
0.09
0.10
0.09
0.10
0.10
0.11
0.11
0.12
0.14
0.13
0.12
0.17
0.15
0.14
0.22
0.19
0.17

Table 3 Ground boundary layer information at x/c = 1.5.

h/c
case
um /U ue /U 99 /c
0.448
exp
0.98
1.09
0.006
S-A
0.93
1.06
0.007
k
0.94
1.06
0.028
0.313
exp
0.90
1.07
0.007
S-A
0.88
1.05
0.008
k
0.91
1.04
0.036
0.224
exp
0.86
1.07
0.009
S-A
0.82
1.04
0.009
k
0.88
1.04
0.044
0.179
exp
0.91
1.06
0.014
S-A
0.78
1.03
0.009
k
0.85
1.03
0.048
0.134
exp
0.94
1.05
0.019
S-A
0.73
1.03
0.009
k
0.84
1.02
0.054
0.090
exp
0.91
1.03
0.022
S-A
0.68
1.02
0.010
k
0.81
1.02
0.061

12 of 12
American Institute of Aeronautics and Astronautics Paper 2001-0423