International Journal of Heat and Fluid Flow 57 (2016) 34–47

Contents lists available at ScienceDirect

International Journal of Heat and Fluid Flow

journal homepage: www.elsevier.com/locate/ijheatfluidflow

Non-symmetric bi-stable flow around the Ahmed body

W. Meile a, T. Ladinek a, G. Brenn a,∗, A. Reppenhagen b, A. Fuchs b
a
Institute of Fluid Mechanics and Heat Transfer (ISW), Graz University of Technology, Inffeldgasse 25/F, 8010 Graz, Austria
b
The Virtual Vehicle Research Center, Inffeldgasse 21a, 8010 Graz, Austria

a r t i c l e i n f o a b s t r a c t

Article history: The flow around the Ahmed body at varying Reynolds numbers under yawing conditions is investigated ex-
Received 19 June 2015 perimentally. The body geometry belongs to a regime subject to spanwise flow instability identified in sym-
Revised 16 September 2015
metric flow by Cadot and co-workers (Grandemange et al., 2013b). Our experiments cover the two slant angles
Accepted 3 November 2015
25° and 35° and Reynolds numbers up to 2.784 × 106 . Special emphasis lies on the aerodynamics under side
Available online 7 December 2015
wind influence. For the 35° slant angle, forces and moments change significantly with the yawing angle in
Keywords: the range 10° ≤ |β | ≤ 15°. The lift and the pitching moment exhibit strong fluctuations due to bi-stable flow
Ahmed body around a critical angle β of ±12.5°, where the pitching moment changes sign. Time series of the forces and
Bi-stable flow moments are studied and explained by PIV measurements in the flow field near the rear of the body.
Non-symmetric flow © 2015 Elsevier Inc. All rights reserved.
Aerodynamic loads

1. Introduction
The Ahmed body is a bluff model body with basic aerodynamic
properties of a vehicle, which was developed for investigating the in-
fluence of the slant angle at the back on the flow field and on the
resulting aerodynamic forces, with suppressed interactions between
the front and the rear parts (Ahmed et al., 1984). The Ahmed body
has been the subject of many experimental investigations and simu-
lations, in most cases in symmetric flow. The study of Le Good and
Garry (2004) shows that the use of generic models is a good way
for determining fundamental aerodynamic characteristics of cars, in-
cluding side wind effects, although most models do not represent all
aspects of the flow around car bodies equally well. Baxendale et al.
(1994) measured drag and lift forces of 8 slant configurations under
symmetric conditions in a wind tunnel. Due to the use of a moving
belt and a low Reynolds number, the force coefficients differ from
Ahmed et al.’s results. Further measurements of the pressure distri-
bution, flow velocity, and turbulence properties of the Ahmed body
in symmetric flow were carried out by Lienhart et al. (2000) and
Lienhart and Becker (2003). The air velocity in those experiments
was 40 m/s. The data were made available for the 9th ERCOFTAC/IAHR
Workshop on Refined Turbulence Modelling, Darmstadt, 2001 – Case
9.4, and the 10th Joint ERCOFTAC (SIG-15)/IAHR/QNET-CFD Workshop
on Refined Turbulence Modelling, Poitiers, 2002 – Case 9.4 of the Eu-
ropean Research Community on Flow, Turbulence, and Combustion
∗
Corresponding author. Tel.: +43 316 873 7341; fax: +43 316 873 7356.
E-mail address: brenn@fluidmech.tu-graz.ac.at, guenter.brenn@tugraz.at
(G. Brenn).
(ERCOFTAC 2001, 2002). This database is valuable for many quantita-
tive comparisons. Conan et al. (2011) measured forces on the Ahmed
body in open and closed wind tunnel test-sections and investigated
the wake flow with the critical slant angle with PIV. In general, the
drag was found to be larger than in Ahmed et al.’s data, for the 25°
slant by 20%. A recent study of Thacker et al. (2012) addressed the
suppession of the 3D separation on the rear slant of the Ahmed body
with 25° slant angle by rounding of the roof/slant junction. A posi-
tive influence of this measure was found. However, during the force
measurements in the sharp edge configuration, the drag was overes-
timated by roughly 30% (compared to Ahmed et al., 1984). The high
sensitivity of the separation to the sharpness of the roof/slant junc-
tions was identified as one reason.
Modern numerical techniques may provide deep insight into de-
tails of flows. Minguez et al. (2008) applied large eddy simula-
tion (LES) with a spectral vanishing viscosity (SVV) technique to
the Ahmed body with a 25° slanted back in symmetric flow at the
Reynolds number of 7.68 × 105 based on the height of the model.
Results were in good agreement with experimental data of Lienhart
et al. (2000). A group around Serre studied three numerical tech-
niques for simulating the flow around the same body at the same
Reynolds number (Serre et al., 2013). The authors conclude that the
simulations may provide good overall agreement with experiments.
Vortical structures are well predicted, but the partial detachment of
the mean flow over the slant was captured only by one method. In
essence, LES for high Reynolds number flows remains challenging.
The wake determining aerodynamic forces on bodies may be
asymmetric at sufficiently high Reynolds number, even for symmet-
ric bodies. For axisymmetric bluff bodies, the axisymmetry breaks
above a critical Reynolds number, which is 210 for the sphere

http://dx.doi.org/10.1016/j.ijheatfluidflow.2015.11.002
S0142-727X(15)00132-0/© 2015 Elsevier Inc. All rights reserved.
 W. Meile et al. / International Journal of Heat and Fluid Flow 57 (2016) 34–47 35

(Grandemange et al., 2013b). The Ahmed body exhibits a similar

symmetry breaking between two steady wake states at Re = 340
(formed with the height H of the body base) in the laminar regime
(Grandemange et al., 2012). Increasing Reynolds number makes the
wake unsteady. At Re ∼ 105 , at long time scales, the flow is bi-
stable, with the recirculation region switching in a random man-
ner between two preferred reflectional symmetry breaking positions
(Grandemange et al., 2013a). As a consequence, the aerodynamic
forces acting on the body are unsteady. This phenomenon is reported
to be independent of the Reynolds number up to Re ∼ 3 × 106 . The
work referenced shows that the formation of the wake instability
depends on the aspect ratio H/W and the ground clearance C/W of
the body, where W is the width of the body. Small values of C/W
< 0.05 prevent the underbody flow by viscous resistance, large val-
ues C/W > 0.09 allow for that flow without separation and with
similar shear layers from the top and bottom of the base. The in- Fig. 1. Geometry of the Ahmed body used, together with the coordinate systems.
stability in the spanwise direction, which may lead to the bi-stable
wake, forms for C/W > 0.1 and for aspect ratios between 0.6 and 1.3.
One state may be preferred in cases of a residual yaw angle or wall
effect.
Investigations in non-symmetric flow are of interest both for sci-
ence and industry for the importance of the side wind influence in car
aerodynamics. In the real situation, asymmetric conditions may orig-
inate from different sources—unsteady atmospheric flows, transient
crosswinds, strong lateral gusts, and from the traffic environment
(passing, crossing manoeuvers). These real crosswind conditions,
which may influence some conclusions drawn from experimental and
numerical studies, are difficult to reproduce both in experiment and
simulation. Non-symmetric flow is subject to instability which may
infer important influences on the forces resulting from the flow, sim-
ilar to what was reported by Grandemange et al. (2013a). An adapted,
simplified Ahmed body geometry was used by Docton (1996) for
experimental investigation and validation of a test rig for transient
vehicle side wind simulation. Surface pressures as well as drag and
lift forces were measured at full scale Reynolds numbers by Bayraktar
et al. (2001). Their model was 4.7 times larger than the original
Ahmed body, and the investigations included three different slant
Fig. 2. Location of the pressure taps in the left rear part of the body with ϕ = 25°.
angles below the critical value of 30°, where the C-pillar vortices dis-
appear, as well as yawing angles up to ±15°. Further investigations on
side wind effects on a simplified generic passenger car shape (Willy experiments. The paper ends with the summary and conclusions in
model) were performed by e.g. Gohlke et al. (2007) and Guilmineau Section 6.
and Chometon (2009). As a continuation of the work of Krajnović
and Davidson (2004) on the flow around the Ahmed body, the Willy 2. Experimental setup and procedure
model was used to investigate crosswind impact on the basis of LES
(Krajnović and Sarmast, 2010). Tsubokura et al. (2010) quantified side 2.1. The Ahmed body
wind effects on a vehicle model with large eddy simulation (LES)
based methods, referring to findings from the Ahmed body. Experi- The geometry of the Ahmed body investigated is depicted with
mental investigations by Ferrand (2014), exposing the Willy model its original dimensions in Fig. 1. The model was built from wooden
to harmonic crosswind, aimed at better understanding of the flow materials, with an interchangeable rear part, so as to realise the two
physics and quantifying the vehicle’s sensitivity to unsteady cross- slant angles investigated with the same front part of the body. The
wind. Experiments on various car models were discussed by Drage model was connected to the balance below the test section floor by
et al. (2008). The experiments on the Ahmed body included force cylindrical aluminium stilts (d = 30 mm), which were fixed by screws
and pressure measurements. A comparison of experimental and on a special mounting support, so that the model could be turned
numerical data is given in Meile et al. (2011). In these experiments, with the balance platform. The stilts reach through a turntable disk
a significant jump of the rear axle lift and pitching moments was in the test section floor without mechanical contact. The body was
observed with the 35° slant at yawing angles near ±12° to 13°. The directionally aligned with the geometrical jet axis of the wind tunnel.
aim of the present study is to investigate experimentally the non- The symmetry lines marked on the body were adjusted by means of
symmetric flow around the Ahmed body with sub- and supercritical measurement angles so that they coincided with appropriate lines on
slant angles. Special emphasis is placed on the influence of the slant the bottom plate of the test section. This alignment was accurate to
angle in non-symmetric flow. Our model has a ground clearance within ±0.1°.
C/W of 0.13 and the aspect ratio 0.74 and is therefore subject to the The locations of the pressure taps were chosen according to the
instability addressed by Grandemange and co-workers. Our paper measurements of Lienhart et al. (2000) and are sketched in Fig. 2.
is organised as follows: we first introduce the Ahmed body together The exact coordinates are given in Table 1. Metering points (MP) 1–
with the experimental setup and procedure used in the study. Section 10 are located in the symmetry plane (y = 0 mm), MP 11–17 in the
3 validates the measurements with existing data for symmetric flow. plane y = –90 mm, and MP 18–24 in the plane y = –180 mm. There-
In Sections 4 and 5 we present and discuss the results from the fore, in Table 1 only the x and z coordinates of the MP are given. For
36 W. Meile et al. / International Journal of Heat and Fluid Flow 57 (2016) 34–47

Table 1 of ±0.5% of the measured value were achieved. Errors due to the res-
Pressure tap locations.
olution of the signal conditioner are negligible. To ensure accurate
ϕ = 25° ϕ = 35° steady-state measurements, data were sampled at the highest pos-
MP x [mm] z [mm] x [mm] z [mm] sible rate (different for the various sensors, but of the order of sev-
eral kHz). One-second mean values were deduced, and these values
1, 11, 18 –215 338 –215 338 were recorded and averaged over periods of at least 15 s to produce
2, 12, 19 –181.01 328.59 –164,28 325.69
steady mean values. With the cross sectional area of the model given
3, 13, 20 –101.26 291.40 –92.19 275.22
4, 14, 21 –21.50 254.20 –20.11 224.74 in Section 2.1, the blockage ratio of the test section is between 3.8%
5, 15, 22 0 214 (216.18∗ ) 0 165 (163∗ ) and 7.5% for 0° ≤ β ≤ 25°. This is acceptable for achieving meaningful
6, 16, 23 0 125 (128∗ ) 0 125 (133∗ ) results.
7, 17, 24 0 65 (68∗ ) 0 65 (73∗ ) The static pressures on the model surface and the ambient pres-
8 –21 50 –21 50
9 –128.45 50 –116.77 50
sure p0 in the hall, used as the reference pressure, were measured
10 –215 50 –215 50 with transducers type HCLA 12 × 5 EB from SensorTechnics with a
range of ± 1250 Pa. The typical error is 0.05% fso. The pressure taps
with a diameter of 0.5 mm were connected to the sensors by 0.4 m
manufacturing reasons, the z coordinates at the vertical backplane long plastic tubings (di = 0.8 mm). Prior to the measurements, the
differ slightly from Lienhart et al. (2000). The corresponding data of sensors were calibrated with a low pressure calibrator type WIKA
the original work are included and indicated by asterisks. CPC 2090. Data were acquired with a CompactDAQ-9178 system
and a C-series 9205 analog input module from National Instruments
2.2. Wind tunnel and measuring techniques (NI). The software package LabView (NI) was used for data record-
ing and processing. The samples were taken over periods of 60 s or
The present investigations were performed in the 2 m low-speed 600 s with a sampling frequency of 100 Hz in parallel with the force
aerodynamic wind tunnel of the ISW at Graz University of Technol- measurements.
ogy. This wind tunnel, depicted in Fig. 3, is of the Göttingen type with For measuring the flow field in the wake of the body, particle-
a ¾-open test section and closed return. A detailed description was image velocimetry (PIV) was used. A pulsed dual-cavity Nd:YAG laser
given by Gretler and Meile (1993). The nozzle outlet cross-section system Gemini PIV from New Wave Research was applied. The wave-
is 2 m x 1.46 m, the maximum wind speed is vmax ≈ 41 m/s. The length is 532 nm, the pulse duration 5 ns, and the released energy
non-uniformity of the flow velocity in the ∼0.7 m wide central part 200 mJ per pulse. The images were recorded with two cameras of
of the test section is below ± 0.25% of the mean value. Larger devi- type High Sense MkII (Dantec Dynamics) with 60 mm fixed-focus
ations of at maximum 1% occur near the open jet boundaries only. Nikon lenses. The CCD chips have 1344 × 1024 active pixels at 12
The longitudinal degree of turbulence in the test section centre is bit depth. The seeding was generated with a smoke generator Fog-
≈ 0.2 %. The static pressure is constant throughout the empty test smoker 2F using Slow Fog Universal fluid (Ehle) and released in the
section from about 0.8 m downstream from the nozzle exit. The test settling chamber of the wind tunnel.
section pressure is only a few Pascal below the ambient pressure in
the wind tunnel hall, corresponding to a pressure coefficient cP ≈ – 2.3. Measurement and evaluation of forces and moments
0.004 at the free-stream velocity of 26 m/s.
The dynamic pressure in the test section (i.e. the reference for the In a first step, the forces were measured at β = 0°, each with in-
undisturbed velocity far upstream from the model) is measured in- creasing and with decreasing air velocities, in order to determine the
directly according to the plenum’s method (Ehlers and Röser 2006). influence of the Reynolds number and to test for possible hysteresis.
The pressure sensor used is type DMP343i from BD sensors with an In order to achieve reliably steady mean values, three measurements
effective range of 1000 Pa and an accuracy of ± 0.1% fso. were carried out for each air velocity, each for 15 s. In order to ensure
The aerodynamic forces on the body were measured with the six- that forces acting on the model are measured alone, as it is done in
component platform balance of the wind tunnel. All load cells are earlier studies in the literature, the tare loads of the stilts were deter-
type Z6-H3 of HBM (200 kg) with an accuracy of 0.05% fso and op- mined in separate measurement series and subtracted from the full
erated with an HBM MGCPlus signal conditioner system controlled loads. Fig. 4 depicts the separate model suspension and the stilts from
by the HBM software CATMAN 5.0. This conditioner system resolves the model through the turntable disk.
forces down to ± 0.02 N. The balance was calibrated following the The readings from the six load cells of the balance allow for the
procedures suggested by Barlow et al. (1999) and Panchenko (1998). derivation of three forces and three moments with respect to the ref-
Due to the calibration, maximum deviations in drag, lift, and moment erence coordinate system chosen in accordance with Hucho (2005).

Fig. 3. The 2 m wind tunnel at the ISW: settling chamber (DVK), nozzle (D), test section (MS), collector (AD), 4 corners (EI – EIV), return section split into three identical parts—3
fans (G), 3 diffusers (DIFF).
 W. Meile et al. / International Journal of Heat and Fluid Flow 57 (2016) 34–47
Fig. 4. The model in the test section supported for tare load measurements. View
downstream into the collector of the wind tunnel.
Fig. 5. Definition of the yawing angle together with the reference coordinate system
(x , y , z ) and the body-fixed coordinate system (x1 , y1 , z1 ).
The origin is located on the test section floor, in the centre between
the stilts, and the vertical axis is aligned with the axis of the balance
(see Fig. 5). This position of the model implies 0.9 m distance of the
front part from the nozzle exit. All aerodynamic loads are measured
in a wind-fixed system, while, in cases β = 0°, the forces and mo-
ments in a model-fixed system are determined computationally. The
yawing angle of the body relative to the wind direction is sketched in
Fig. 5. The primes distinguish the wind-fixed coordinate system for
force measurements from the system related to wake measurements,
as given in Fig. 1, and the coordinates with subscript “1” denote the
according body-fixed coordinate system with respect to the yawing
angle. The loads with respect to the model-fixed coordinate system
are depicted in Fig. 6. The drag, lift, and side forces, as well as the
pitching, yawing, and rolling moments are determined. The force and
moment coefficients are defined as
Fi Mj
ci = cM j = (1a and b)
(ρ U∞
2 /2) A
re f (ρ U∞
2 /2) A
re f lre f
where i = D, L and S for the drag, lift and side forces and j = R, P and
Y for the rolling, pitching and yawing moments. The reference length
lref is the track width of 0.327 m for the rolling moment coefficient
and the wheelbase of 0.47 m for the two other moment coefficients.
The reference area ARef = 0.112 m2 is the cross sectional area of the
model perpendicular to its longitudinal axis. The reference velocity
is the free stream velocity U∞ in the wind tunnel test section, and
37
Fig. 6. Relationship between forces in the wind-fixed (x , y ) and in the model-fixed
(x1 , y1 ) coordinate systems.
Fig. 7. Overview of the measurement planes.
ρ denotes the air density. The lift forces on the front and rear axles
(FL, f , FL, r ) are determined from the overall lift force FL and the pitch-
ing moment MP .
The corresponding force coefficients are defined as in Eq. (1a). The
pressure coefficient is defined with the static pressure p at the model
surface and the reference pressure p0 as
p − p0
cP = (2)
ρ U∞
2 /2
2.4. Flow field measurements in the wake
The flow field in the wake of the Ahmed body was investigated
with PIV for relating the aerodynamic loads to properties of the flow
field. The PIV measurements were designed to reveal the flow struc-
ture qualitatively, without detailing the turbulent dynamics of the
wake, which was beyond the scope of our work. A more detailed anal-
ysis of the wake could follow the work by Grandemange et al. (2013a).
In the experiments on the wake flow, the free stream velocity in the
test section was 30 m/s, which was operable with our wind tunnel
over long periods of time and representative of the flow phenomena
studied.
Both cameras of the PIV system were mounted upright, one above
the other, with an overlap of about 12 mm in the image plane,
and operated simultaneously. This resulted in an image section of
161 × 412 mm suitable for our study. The camera setup was traversed
three times by one image width in the main flow direction, resulting
in a 644 mm long region covered. In the cross-stream direction, 10–
12 planes with distances y = 40 mm were placed, as depicted in
Fig. 7. The reference coordinate system (x, y, z) is specified in Fig. 1.
After traversing the system three times in the x-direction from the
initial position, 8 single images had to be composed to yield the final
image of the measurement planes. The recording frequency was 5 Hz,
the recording time was set to 20 s, resulting in 100 double frames to
38 W. Meile et al. / International Journal of Heat and Fluid Flow 57 (2016) 34–47

Table 2
Data processing methods and parameters in Dantec Dynamic Studio.

Method Parameter Value

Calibration Scale factor 24.425

Image stitch Row × column 2×1
Y offset 74 pixels
Image arithmetic Subtract image min
Adaptive correlation Final interrogation area size 64 × 64 pixels
Number of refinement steps 1
Overlap 75%
Peak validation Relative to peak 2 1.1
Range validation Velocity range min 0 m/s
Velocity range max 40 m/s
Moving average validation Averaging 5×5
Acceptance factor 0.1
Iterations 3
Vector statistic Include all valid, non-substituted

Fig. 8. Drag coefficient versus Reynolds number (ϕ = 25°, β = 0°).

represent the mean velocity distribution, which is sufficient for reli-

able mean velocities (Thacker et al., 2012). The program package Dy-
namic Studio 2.10.86 served for data acquisition. For the data process-
ing, the newer version 3.14.35 (both from Dantec Dynamics) was used.
Processing parameters are listed in Table 2. In order to avoid loss of in-
formation in the correlation procedure, the simultaneously recorded
pairs of images were combined with “Image Stitch”. The images were
correlated with the algorithm “Adaptive Correlation”. For validation,
the three available routines were applied in the sequence specified
in Table 2. Nonvalidated vectors were rejected, but not replaced. No
filters were applied. The field of mean velocity vectors was calculated
by “vector statistics” and stored with “numeric export”. All further
data processing (such as vector stitching, time averaging of veloci-
ties, further post-processing, etc.) was completed using MATLAB. The
interrogation area size of 64 × 64 pixels, together with an overlap of
75%, yielded a spatial resolution of 2.5 mm.
The mean number N (averaged over the whole image area) of val-
idated vectors may serve as an indicator for the overall measurement
performance. In symmetric flow and at moderate yawing conditions
N > 90 was achieved in all measurement planes y = const. At larger
yawing angles, N > 80 prevails in ≥ 75% of the measurement planes.
Low seeding concentration is a well-known problem in regions near
solid walls, and also borders and interfaces of images may introduce
some error. In such regions the quality is reduced. The maximum tur-
bulence intensity derived from the present measurements was about
35%. For 100 pairs of images and a 95% level of confidence, this leads
to an RMS error of the time averaged velocity of 6.9% in the worst
case (Benedict and Gould, 1996). Higher turbulence intensities (up
to 45%) were found only in regions with less validated vectors, sug-
gesting an RMS error of 9.8%. With respect to the aim of the present
investigations, uncertainties of the specified order of magnitude are
acceptable.
In order to check a dependence of the mean velocity on the num-
ber of image pairs recorded, the results were evaluated by averag-
ing the raw data over 50–100 image pairs in steps of 10 pairs for
selected locations in the near wake. In symmetric flow conditions,
the positions 173 mm ≤ x ≤ 239 mm, y = 0 and various vertical
distances were considered, where a comparison with LDA data of
Lienhart et al. (2000) reveals very good agreement (see Section 3.2
below for a quantitative comparison). For reference in flow fields
with higher dynamics, the data for ϕ = 35°, β = –12.5° in the two
states “1” and “2” found in the bi-stable flow in the planes y = 160
and 200 mm were evaluated (see Fig. 7). Various vertical positions at
46 mm ≤ x ≤ 364 mm were verified. In the considered locations, the
standard deviation of the mean velocities for numbers of image pairs
between 50 and 100 does not exceed 10% of the final mean value,
thus confirming that 100 image pairs are sufficient for the present
investigation.
2.5. Experimental program
The aim of the experiments was to investigate the dynamics of the
forces and the bi-stable flow field on the Ahmed body. Steady mean
values of forces and moments were measured over periods of 15 s at
• The two slant angles ϕ = 25°, 35°,
• Yawing angles in the range –25° ≤ β ≤ 25° in 2.5° steps,
• The yawing angles β = ±11°, ±12°, ±13°, ±14° with ϕ = 35°,
• The velocities U∞ = 10, 20, 30, 40 m/s—corresponding to Reynolds
numbers 0.69 × 106 ≤ Re ≤ 2.73 × 106 .
Time resolved measurements of forces and moments over periods of
600 s, with an averaging time of 1 s, were performed at
• The slant angle ϕ = 35°,
• The yawing angles β = ±11°, ±12°, ±13°, ±14°, ±15°
• The velocities U∞ = 30 and 40 m/s.
The pressure distribution was measured for both slant angles in
parallel to the corresponding force measurements under the follow-
ing conditions:
• Yawing angles β = 0°, ±5°, ±7.5°, ±10°, ±12.5°, ±15° (with
ϕ = 25°); 60 s periods
• Yawing angles β = 0°, ±5°, ±10°, ±11°, ±12°, ±12.5°, ±13°, ±14°
±15° (with ϕ = 35°), 60 s periods
• Yawing angles β = ±10°, ±11°, ±12°, ±12.5°, ±13°, ±14° ±15°
(with ϕ = 35°); over periods of 600 s (time resolved measure-
ments)
• The velocities U∞ = 30 and 40 m/s.
Velocity measurements in the wake were performed with PIV at the
constant free-stream velocity of 30 m/s with both slant angles for the
yawing angles β = +5°, 0°, –5°, –10°, –12.5°, and –15°. These measure-
ments will be described in more detail in Section 4 below.
3. Symmetric flow—data validation
3.1. Drag and lift coefficients
We validate our wind tunnel measurements by comparison with
corresponding data from the literature, starting from two force coef-
ficients. Fig. 8 displays the drag coefficient of the Ahmed body as a
function of the Reynolds number for the slant angle ϕ = 25° in sym-
metric flow (β = 0°). The maximum velocity (40 m/s) corresponds to
the Reynolds number Re = 2.73 × 106 . We compare our data to those
of Ahmed et al. (1984) and Bayraktar et al. (2001) at Re = 4.29 × 106 .
The curve fit through the present data (Meile et al., 2011) includes the
 W. Meile et al. / International Journal of Heat and Fluid Flow 57 (2016) 34–47
Fig. 9. Lift coefficient versus Reynolds number (ϕ = 25°, β = 0°).
values reported in Ahmed et al. (1984) and Bayraktar et al. (2001) and
so extends over one order of magnitude in the Reynolds number. The
data show that the present measurements reproduce the reference
data accurately.
In the lower Reynolds number regime 0.69 × 106 ≤ Re ≤ 2.73 × 106
covered by the present measurements, the drag coefficient varies by
about 13%, while for larger Re this dependency is less pronounced.
Bayraktar et al. (2001) report a decrease of cD by about 3.5% over the
investigated range 4.29 × 106 ≤ Re ≤ 1.32 × 107 . Their data show the
major part of the decrease up to Re ≈ 9 × 106 , and a much less pro-
nounced variation above this value. The data of Thacker et al. (2012)
for the sharp-edge configuration are included for comparison. The
values are much larger and tend towards a constant value at the end
of the tested Re regime.
The lift coefficient as a function of the Reynolds number is shown
in Fig. 9 for the slant angle ϕ = 25°. Since Ahmed et al. (1984) did
not provide lift data, we included reference data of Bayraktar et al.
(2001), Baxendale et al. (1994), and Morel (1978), although the test-
ing conditions were different. The latter author investigated the influ-
 ground
ence of the ground clearance on lift and drag. In his work, the
clearance C was related to the equivalent diameter Deq = 4W H/π ,
where W denotes the body width and H the body height. Investiga-
tions with C/Deq = 0.82 and 0.12 led Morel (1978) to the conclusion
that the drag coefficient is only weakly influenced, while the lift coef-
ficient is reduced significantly by a nearly constant value of cL ≈ 0.2
in the whole regime of slant angles.
In the present experiments, C/Deq = 0.13 and there is a slight
Reynolds number dependence in the lower regime. The lift data
Fig. 10. Velocity profiles u/U∞ in the symmetry plane for symmetric flow at U∞ = 30 m/s. (a) ϕ = 25°, (b) ϕ = 35°.
39
of Bayraktar et al. (2001), obtained at a relative ground clearance
of C/Deq = 0.48, are much larger and constant in the investigated
Reynolds number regime. However, a comparison with Morel’s data
(C/Deq = 0.82) indicates the crucial influence of the ground clearance
on lift and suggests that ground effects were already noticeable in
the experiments of Bayraktar et al. (2001). In contrast, the lift coeffi-
cient provided by Baxendale et al. (1994) is lower than in the present
work, although the relative ground clearance was identical. Their ex-
periments were performed with a moving belt and without support-
ing stilts, which may reduce the lift coefficient by up to 10% against
the value with fixed floor (Bearman et al., 1988). This closely corre-
sponds to the difference between the present experiments and those
of Baxendale et al. (1994). Compared to the results of Baxendale et
al. and Morel (C/Deq = 0.12), our present experiments are reasonable.
The data measured in the study of Thacker et al. (2012) show an in-
crease of lift in the lower Re regime, similar to the present data. Nev-
ertheless, their values are much higher.
3.2. Flow field in the wake
For a quantitative validation of our PIV velocity measurements,
the velocity profiles in several planes were compared with LDA data
of Lienhart et al. (2000) available from Ercoftac (2001, 2002). Fig. 10
depicts the normalized magnitude of the mean flow velocity u/U∞
in the symmetry plane for both slant configurations. The abscissa is
the x-position of the profiles. In order to compare directly with the
reference data, the profiles from the present experiments were de-
duced by linear interpolation between pairs of x-positions spaced by
2.5 mm.
The agreement is very good over wide parts of the wake. As a mea-
sure of the quality, the difference between the normalized velocities
u/U∞ = |(u/U∞ )PIV − (u/U∞ )LDA |instead of relative values is used.
On average over the region 88 mm ≤ x ≤ 438 mm behind the 25°
slant a value of u/U∞ = 0.034 results. The average over the region
17 mm ≤ x ≤ 438 mm behind the 35° slant is u/U∞ = 0.046.
More pronounced deviations are found near the model walls (es-
pecially for ϕ = 25°) which may be explained by light reflections in
the PIV measurements. A larger deviation for the 35° model is seen
in the shear layer between the wake and the outer flow. The devia-
tions in this region averaged over the same x-range, but considering
only points 0.208 ≤ z ≤ 0.348, results in u/U∞ = 0.06. This may
be explained by the lower Reynolds number in the present experi-
ments and the fact that Lienhart et al. (2000) tripped the flow to as-
sure fully turbulent conditions. For all other planes y = const. = 0, the
comparison with data of Lienhart et al. (2000) shows even bet-
ter agreement and can be found in Wanker (2011). We conclude
from this positive validation that our PIV velocity measurements are
40 W. Meile et al. / International Journal of Heat and Fluid Flow 57 (2016) 34–47
Fig. 11. The wakes in symmetric flow at (a) ϕ = 25°, (b) ϕ = 35°.
accurate enough to allow for an analysis of the spatial structure of the
flow field in the near wake of the body, which is an important aim of
our investigation.
4. Velocity field in the wake
Since recently, an increased body of large eddy simulations (LES)
on flow around the Ahmed body is available (e.g., Krajnović and
Davidson, 2004, 2005a, 2005b). These data may enhance the under-
standing of the flow. Data from both experiments and simulations for
yawing conditions, however, are still limited.
We present the flow fields measured at U∞ = 30 m/s in vari-
ous x,z-planes (see Fig. 7) with the
local velocity magnitude normal-
ized by the free-stream velocity u2 +w2 /U∞ displayed in colours,
and the streamlines drawn as solid lines. Furthermore, the isotachs
u/U∞= 0 (white line), as the border of the region with reversed flow,
and u2 +w2 /U∞ = 0.5 (red line) indicating the extent of the wake are
shown. In 2D pictures, the full recorded domainis presented, while
in 3D illustrations only the flow inside the line u2 +w2 /U∞ = 0.5 is
displayed for better visualisation of the wake.
4.1. The flow field in symmetric conditions
The wake of the body in symmetric flow was investigated for val-
idation of the measurement
 procedure. A comparison of the near
wakes (inside the red line u2 +w2 /U∞ = 0.5) is shown in Fig 11. With
the subcritical slant angle ϕ = 25°, a thin separation bubble appears
in the central region of the slant, also found by Thacker et al. (2012)
for their sharp-edge configuration. The otherwise attached flow sep-
arates at the upper edge of the base. This results in a relatively short
near wake (xNW ∼ 195 mm in the symmetry plane) with the typi-
cal pair of counter-rotating vortices (data not shown here). The de-
tected length of the wake agrees with the value of xNW = 187 mm by
Krajnović and Davidson (2005a) to within 4%, although the Reynolds
number was much smaller in the LES simulations. The authors claim
relevance for higher Reynolds numbers as well.
In contrast to the highly three-dimensional nature of the wake
with ϕ = 25°, the wake structure with ϕ = 35° is nearly two-
dimensional, since no C-pillar vortices occur in symmetric flow. The
flow separates at the upper edge of the slant and, therefore, a rela-
tively long wake (xNW ∼ 350 mm) is formed. The wake exhibits again
Fig. 12. The wake at ϕ = 35°, β = –10°.
two counter-rotating vortices, with the upper one extending over the
slant and into a large region behind the body.
4.2. Yawing conditions
4.2.1. Slant angle ϕ = 25°
The wake flow for the subcritical slant with ϕ = 25° is not of big
interest here, so that we address it only briefly. Further discussions
may be found in Meile et al. (2012). Under yawing for small angles
β = ±5°, the flow is attached over the slant and the wake is short, as
in symmetric flow. With increasing yawing angle (β = –10°, –12.5°,
–15°), the flow field turns into three-dimensional, with small differ-
ences between these three angles (Meile et al., 2012). The longitu-
dinal vortex at the windward side is much larger than observed for
smaller yawing angles. A large region is influenced by this vortex up-
stream above the slant and can be identified in several inner mea-
surement planes. The vortex must therefore be initiated upstream
at the prismatic part of the Ahmed body. Indeed, the formation of
longitudinal vortices at the upper long edges of prismatic bodies un-
der yawing conditions is known from other situations like flat-roofed
buildings, containers, tanks, etc. At the windward side of the Ahmed
body, this vortex rotates in the same direction as the C-pillar vor-
tex, and it may be assumed that the vortices merge. At the lee side,
the roof-edge vortex rotates in opposite direction and, hence, the
C-pillar vortex is reduced or even vanishes at larger yawing angles.
The pressure profile depicted in Fig. 21c below clearly supports this
assumption.
4.2.2. Slant angle ϕ = 35°
With ϕ = 35° at small yawing angles, the flow separates at the up-
per edge of the slant, and the wake is long. Even at smaller angles, the
wake appears deformed in the rear portion of the measurement re-
gion when compared to its vertical extent in symmetric flow. This can
be ascribed to longitudinal vortices emerging from the upper edges
of the prismatic part of the body, as with the 25° slant.
A significant change of the wake is evident at β = –10°. The wind-
ward roof-edge vortex causes a downwash towards the slant: the
wake is about 100 mm shorter than in symmetric conditions, and the
deformed region has propagated towards the body. This phenomenon
is seen in the left part of Fig. 12. A comparison with the symmetric
flow field in Fig. 11b highlights these differences.
The bi-stable state at the yawing angle β ≈ –12.5° is of particu-
lar interest. In order to clearly distinguish between the two states,
two separate series of velocity measurements were recorded, while
observing the balance readings without tripping the flow. In several
previous investigations (e.g. Ahmed et al., 1984), the flow was specifi-
cally manipulated to maintain one of the two possible states for mea-
suring over longer periods of time. The different wake structures are
compared in Fig. 13.
The flow fields in the two cases are significantly different. In
the following we refer to the two structures as state “1” and state
 W. Meile et al. / International Journal of Heat and Fluid Flow 57 (2016) 34–47
Fig. 13. The wake at ϕ = 35°, β = –12.5°. (a) state “1”, (b) state “2”.
Fig. 14. Velocity field and streamlines in various planes y = const. (values in mm) at
ϕ = 35°, β = –12.5°, state “1”.
“2” in accordance with the description of pressures and forces in
Section 5.3. The velocity distribution and streamlines correspond-
ing to states “1” and “2” are depicted in Figs. 14 and 15, respec-
tively. In both cases we restrict the presentation to the outer planes
at 120 mm ≤ y ≤ 320 mm.
The flow fields of state “1” (Fig. 14) are largely identical with the
observations for the smaller β = –10°. The vortex emerging from
the windward roof edge is located above the wake and is identi-
fied in the planes y = 160 mm and y = 200 mm. The sense of rota-
tion follows from the flow directions—downward in the inner region
(y ≤ 160 mm) and upward in the outer region. The wake region is
41
Fig. 15. Velocity field and streamlines in various planes y = const. (values in mm) at
ϕ = 35°, β = –12.5°, state “2”.
≈ 200 mm long. In state “2” (Fig. 15), the wake is significantly shorter,
comparable to the wake at ϕ = 25° for the same yawing angle. The
flow separates at the upper edge of the slant, but it is reattached to
the slanted surface, thereby forming a separation bubble. As a pro-
cess leading to this state we may find that the roof-edge vortex is bent
down and so forms a C-pillar vortex. A possible reason may be a span-
wise switching of the near wake between two preferred states (as re-
ported by Grandemange et al., 2013a, b). If the structure is shifted to-
wards the windward edge, this could initiate a strong interaction with
the roof-edge vortex. The result is evident in the planes y = 280 mm
and y = 320 mm.
The velocity measurements for a given free-stream velocity and
placement of the body show big differences between the flow fields
of states “1” and “2”. From the above results sketches of the flow fields
representing the dominant vortices in the two states were derived.
State “1” is sketched in Fig. 16a, and state “2” in Fig. 16b.
The formation of the roof-edge vortices is identified as the reason
for the observed change of state. Both the intensity of these vortices
and the downwash effect of the windward vortex increase with the
yawing angle, leading to reattachment of the flow on the slant. The
instability of the flow may be caused by an unsteadiness induced by
the interaction of the longitudinal vortices with the wake, which may
be initiated by the spanwise switching of the near wake. The corre-
sponding pressure fluctuations will be shown in Figs. 22, 23 and 29
of the next section. These fluctuations cause a change of the vortex
position near the slanted surface in time.
The results for the largest yawing angle β = –15° show a behaviour
similar to state “2” described above. The data show that a change of
state occurs in the range 11° ≤ |β | ≤ 14° of the yawing angle, and that
the change is finished at |β | = 15°.
5. Forces, moments and pressure under yawing conditions
The forces and moments acting on the Ahmed body and their
potential changes with the yawing angle are important for assess-
ing side wind effects. In the following we describe results for the
Reynolds number Re ≈ 2.1 × 106 (for U∞ = 30 m/s) which we relate
to the flow field data from the wake acquired with PIV.
42 W. Meile et al. / International Journal of Heat and Fluid Flow 57 (2016) 34–47
Fig. 16. Sketch of the dominant vortex structure at ϕ = 35°, β = –12.5°. (a) state “1”, (b) state “2”.
Fig. 17. Force coefficients versus yawing angle at ϕ = 25°, U∞ = 30 m/s.
Fig. 18. Force coefficients versus yawing angle at ϕ = 35°, U∞ = 30 m/s.
5.1. Time-average forces and moments
Figs. 17 and 18 show the coefficients of forces determined for the
slant angles ϕ = 25° and ϕ = 35°, respectively, averaged over 15 s.
For the slant angle of 25°, the drag coefficient varies only slightly
with the yawing angle. The drag increase by cD ≈ 0.03 in the range
0° ≤ β ≤ 15° equals the results of Bayraktar et al. (2001) at Reynolds
numbers Re ≥ 8.8 × 106 (the corresponding values of cD at β = 0°
are displayed in Fig. 8). For β > 15° the drag decreases slightly. For
the present slant angle, the lift forces are positive in the whole range
of yawing angles. In symmetric flow (β = 0°), the lift force at the
front axle is very close to zero. The overall lift is generated at the rear
part of the body. With increasing yawing angle, all lift forces increase,
which is more pronounced from β ≈ 9° onwards. The front axle lift
Fig. 19. Moment coefficients versus yawing angle at ϕ = 25°, U∞ = 30 m/s.
also increases, which may be ascribed to flow separation in this re-
gion. As can be expected, the side force is zero in symmetric flow and
increases continuously with β . The change is moderate up to β = 7.5°
and more pronounced for larger β due to the flow separation in the
front part.
The measurements with ϕ = 35°, in contrast, exhibit a discontin-
uous curve shape of the forces with a significant change in the range
10° ≤ β ≤ 15°. In symmetric flow, the drag is smaller than with the
subcritical slant angle (which is a well-known behaviour), but much
larger at yawing angles > 15°. The overall lift is positive on the whole
regime of β , but notably smaller than with ϕ = 25°, and close to
zero in symmetric flow. The front axle lift shows only minor varia-
tions throughout the range of yawing angles, but the rear axle lift is
negative for β < 10°, with a minimum near β ≈ 5°, reaches zero at
β = 10°, and jumps to large positive values in the range 10° ≤ β ≤ 15°.
The overall lift is therefore mainly dominated by the changes near the
rear axle. The front axle lift increases for 15° ≤ β ≤ 25°, which indi-
cates flow separation in the front part. In the same range of β , the lift
in the rear part decreases and the overall lift is slightly reduced. The
side force increases again nearly linearly with β , and the changes seen
in the other forces are reflected by a kink in the slope of this curve. In
the regime where drag and lift change sharply, the side force grows
less, followed by a regime with a steeper profile. This may be ascribed
to enhanced separation near the front part at larger yawing angles.
The moment coefficients are depicted in Figs. 19 and 20 for
ϕ = 25° and 35°, respectively. The dependence of the moments on
the yawing angle is clearly determined by the relation between the
forces. For both slant angles, the rolling and pitching moments are
positive at positive yawing angles and vice versa, and both moments
increase smoothly, which is in accordance with the side force.
With the subcritical slant angle ϕ = 25°, the pitching moment
is always negative (towards the front) due to the distribution of lift
between the axles. With the supercritical slant angle ϕ = 35°, the
 W. Meile et al. / International Journal of Heat and Fluid Flow 57 (2016) 34–47 43

Fig. 20. Moment coefficients versus yawing angle at ϕ = 35°, U∞ = 30 m/s.

evolutions of the rolling and yawing moments are similar, but the
pitching moment behaves completely different. For |β | ≤ ∼ 12° this
moment is positive. However, the significant change of the rear axle
lift near |β | ≈ 12° to 13° causes the pitching moment to change its
sign. Further increase of β causes a reduction of the pitching moment,
but the sign remains negative. The observed variations of forces and
pitching moment may have a significant influence on the driving dy-
namics of the vehicle, or may cause unexpected load distributions on
other structures.
It is seen in the diagrams above that the symmetry with respect to
β = 0° is good, although not perfect everywhere. Inexplicable asym-
metry of data was also reported by Bayraktar et al. (2001).
The absolute values of force and moment coefficients at the high-
est velocity of 40 m/s are not identical to the values determined at
30 m/s, which is caused by the Reynolds number dependence, but
the evolution with the yawing angle follows the same trends. With
the subcritical slant angle, the slope of the coefficients is continuous,
while the supercritical slant angle leads to jumps in drag, lift, and Fig. 21. Pressure coefficients at ϕ = 25°, U∞ = 30 m/s. (a) MP 1–10, (b) MP 11–17, (c)
MP 18–24.
pitching moment. This behaviour with changes in the Reynolds num-
ber was used in measuring the wake at a velocity of 30 m/s without
any risk of losing important information.
The behaviour at the two low velocities investigated is different bars indicate the RMS values of the pressure fluctuations calculated
from the higher velocities. We do not show those data here. For the from the fully resolved signal and are drawn symmetrically around
subcritical slant angle, the evolutions of the force coefficients are the mean value. Only moderate pressure fluctuations are seen.
similar to the 30 m/s case, but drag and lift coefficients are in gen- For symmetric flow, the pressure distribution compares very well
eral larger and the side force is smaller. The change of state seen for with the data of Lienhart et al. (2000) obtained with a velocity of
U∞ = 30 m/s at ϕ = 35° was detected also with U∞ = 20 m/s, but 40 m/s, but with an offset of cP = –0.041 ± 0.01 throughout the
at positive yawing angles the jump is shifted to β ≈ 14°, while for field. The reasons for this offset are unclear. One possible reason may
negative β the slope shows a kink in the according range of β . At the be a difference in the reference pressures used in the two sets of
lowest velocity, no jump of coefficients could be detected at ϕ = 35°. experiments.
A comparable Re-dependent behaviour of the change of state was re- In the symmetry plane y = 0, the pressure near the roof edge is
ported by Conan et al. (2011) for the critical slant angle ϕ = 30°. These low, reaches a distinct minimum at MP 2 on the slant, then increases
data clearly indicate the crucial Reynolds number influence on the bi- downstream, and reaches a nearly constant high level cp = –0.1 at the
stable state also. backplane and the underbody. The data of Thacker et al. (2012) for
The reason for the observed sharp changes of the forces and mo- U∞ = 33 m/s exhibit a similar increase along the slant, except that
ments with the supercritical slant may be seen in changes of the flow their measurements did not resolve the minimum at our MP 2, while
field. In search for an explanation of the force profiles, temporal mean it is clearly shown in the data of Lienhart et al. (2000) in symmetric
data may not be enough. We therefore analyse time-resolved data be- flow. Their pressure distribution at y = –90 mm is close to ours, with
low to improve the view on this behaviour. a constant deviation explainable by the location of pressure measure-
ment in their experiment different from ours.
5.2. Time-average pressures The pressure distribution along the outermost line in symmet-
ric flow shows a minimum at MP 20 near the middle of the slant.
The pressure distributions along all 3 lines of metering points This is known to be due to the C-pillar vortex. For positive yaw-
(MP) with the subcritical slant angle ϕ = 25°, measured at the ve- ing angles, the pressure at this windward edge is significantly re-
locity of 30 m/s and represented by the pressure coefficient, are de- duced with increasing yawing angle and the minimum is shifted to-
picted in Fig. 21. All the data represent averages over 60 s. The error wards MP 19. This indicates that the vortex may be generated farther
44 W. Meile et al. / International Journal of Heat and Fluid Flow 57 (2016) 34–47
Fig. 22. Pressure coefficients at ϕ = 35°, U∞ = 30 m/s, |β | = 0°, 5°, 10°, 12.5°, 15°. (a)
MP 1–10, (b) MP 11–17, (c) MP 18–24.
upstream than with smaller yawing angles. For negative yawing an-
gles, the pressure at this leeward edge continuously increases on the
slant, and the minimum—present in MP 20 at low yawing angles—
vanishes. From these observations we conclude that, under yawing
conditions, the C-pillar vortex is enhanced at the windward edge,
while it is reduced at the leeward edge. This conclusion was already
addressed in Section 4.2.1 based on the corresponding velocity field.
The pressure distributions measured at 30 m/s along correspond-
ing lines with the supercritical slant angle are shown in Figs. 22 and
23. The latter figure displays the data for 11° ≤ |β | ≤ 14° in a separate
diagram to avoid data overload. The data for 11° ≤ |β | ≤ 14° are aver-
ages over 600 s, while the averaging period was 60 s for all the other
yawing angles.
With ϕ = 35°, the pressure level at the slant is much higher than
with the smaller slant angle. In the uppermost row of MPs on the
roof just before the sharp edge, a level –0.4 ≤ cP ≤ –0.2 is reached,
and even the minima at y = –180 mm do not significantly exceed the
value of cP = –1 as with ϕ = 25°. The pressure at the backplane is of
the same order of magnitude as with the lower slant angle.
In the symmetry plane, the pressure level is high and a mini-
mum develops between MP 2 and 3 for larger yawing angles. Prac-
tically no effect of the orientation of the slant relative to the incom-
ing wind can be detected. The data points for corresponding yawing
angles are practically equal (max. differences < 3.8% at MP 4), ex-
cept β = ± 12.5°. At MP 11–17 in the central line we see again that
Fig. 23. Pressure coefficients at ϕ = 35°, U∞ = 30 m/s, 11° ≤ |β | ≤ 14°. (a) MP 1–10, (b)
MP 11–17, (c) MP 18–24.
minima occur at larger angles β , where the pressure is lower than at
corresponding points in the symmetry plane.
The effect of the orientation is best visible at the outermost line.
The pressure minimum is shifted to MP 19 when the edge is wind-
ward, but to MP 20 in leeward orientation. The error bars clearly show
that the pressure fluctuations are much larger than with the subcriti-
cal slant. Due to the bi-stability of the flow discussed in the next sec-
tion, the time-average pressure data for 11° ≤ |β | ≤ 14° may lead to
erroneous interpretations, as was pointed out by Grandemange et al.
(2013a). This is particularly true for the variance which was not con-
ditionally determined in Figs. 22 and 23. A conditional analysis is
given in the next section.
5.3. The bi-stable state—time-resolved pressure and forces
The time-resolved measurements yielding data over periods of
600 s were carried out in order to study the temporal flow behaviour
over a long period of time. We present the data for 30 m/s only.
Figs. 24 and 25 depict force data for positive and negative values of
β . Here we show the lift coefficient only, since its variation is largest.
We discuss the data for the yawing angles β = ± 12.5° for which
we presented velocity data in Section 4 above (Fig. 13). The two dif-
ferent states with low and high force coefficients are marked in the
diagrams with “1” and “2”, respectively. Both states may remain for
several minutes, but they might as well disappear after a few seconds,
 W. Meile et al. / International Journal of Heat and Fluid Flow 57 (2016) 34–47
Fig. 24. Time series of the lift coefficient at ϕ = 35°, U∞ = 30 m/s, 11° ≤ β ≤ 15°.
Fig. 25. Time series of the lift coefficient at ϕ = 35°, U∞ = 30 m/s, –15° ≤ β ≤ –11°.
as can be seen for values of β different from ±12.5°. The changes ap-
pear completely random without a distinct frequency. This leads to
the hypothesis that there exists a bi-stable state as a limiting case
between two flow patterns. In the following we concentrate on nega-
tive yawing angles. The variations of drag, lift, side force and pitching
moment with time are depicted in Fig. 26.
In state “1”, drag and lift are low. The rear axle lift is nearly zero,
causing a positive pitching moment. This corresponds to observations
at the smaller angle β = –11° in the diagrams: starting from a given
low level of the side force and the related yawing and rolling mo-
ments (data not shown here) in state “1”, a change of the flow into
state “2” makes the overall lift increase to a large value. This lift in-
crease is due to the rear part, since the rear axle lift increases from
nearly zero to a large positive value. Simultaneously, the front axle
lift is reduced and, therefore, the pitching moment switches to a neg-
ative value.
The variation of the drag is smaller in magnitude, but always syn-
chronous to the change of the lift. An increase of drag and rear axle
lift indicates lower pressure at the rear part of the model. This, in
turn, allows for the hypothesis that a new inward rotating C-pillar
vortex is generated at the windward edge of the slant, as known from
the subcritical slant angle. The pressure reduction may be ascribed
to the altered flow field caused by this vortex. For this asymmetric
flow field one could expect that the negative rolling moment should
also increase in magnitude. However, the rolling moment decreases
45
Fig. 26. Time series of force and moment coefficients at ϕ = 35°, U∞ = 30 m/s, β = –
12.5°.
by ∼3%. This can be explained by the interaction between the lift and
side forces.
At the investigated yawing angles, the absolute value of the side
force is much larger than the lift. Therefore, the side force has more
influence on the rolling moment. The side force is reduced during the
change of state, and so is the rolling moment. This is due to the reduc-
tion of the windage because of the shorter wake, which, in turn, is a
result from the vortex formation as known from the subcritical slant.
Further measurements at the Reynolds number for U∞ = 40 m/s, not
shown here, revealed that the bi-stable state occurs with similar force
fluctuations in the same range of β .
The aforementioned hypotheses are confirmed by the pressure
measurements performed at the same time as the force measure-
ments. Figs. 27 (β >0°, pressure taps windward) and 28 (β < 0°,
pressure taps leeward) depict the temporal behaviour of pressure by
means of 1 s mean values for MP 1–7 located in the symmetry plane.
These data confirm the random change of states between different
plateaus over certain periods of time found in the force measure-
ments. Comparing the data to Figs. 24 and 25, states “1” and “2” oc-
cur in the same time windows. The pressure levels in one state are
practically equal at all the metering points for both yawing angles, as
may be expected for the symmetry plane. At β = 12.5° state “1” with
low forces and high pressure prevails. At the corresponding negative
yawing angle, state “2” prevails. This difference is clearly visible also
in the mean values depicted in Fig. 22c.
Mean pressures in both states were determined by a conditional
analysis, with the state “1” or “2” as the condition. The distribution
along lines y = constant is presented in Fig. 29. The error bars rep-
resent the standard deviations conditional to each state. Even in the
symmetry plane (y = 0) the differences between the two states are
evident. The orientation of the body in the flow has no influence
on the pressure there. This influence increases towards the outer
lines of taps. The under-pressure on the slant is stronger in state “2”.
The contrary is true for the vertical back, particularly pronounced in
the symmetry plane (including the underbody, MP 8–10). The strong
under-pressure on the slant in state “2”, together with the higher
46 W. Meile et al. / International Journal of Heat and Fluid Flow 57 (2016) 34–47
Fig. 27. Time series of pressure coefficient at ϕ = 35°, U∞ = 30 m/s, β = 12.5°, MP 1–7.
Fig. 28. Time series of pressure coefficient at ϕ = 35°, U∞ = 30 m/s, β = –12.5°, MP
1–7.
pressure on the underbody, explains the strong increase of lift on the
rear axle. The increase of drag results from the same reason, where
the much smaller amount is due to the small projected streamwise
area of the slant and the high pressure on the vertical back.
The outermost line of metering points MP 18–24 at y = –180 mm
is of particular interest because the minima in the pressure distribu-
tion in state “2” indicate longitudinal vortices for both orientations,
but the minimum pressure location is different for the windward and
the leeward orientations of the slant edge (see Fig. 29c). For the wind-
ward orientation, the forces measured suggested a C-pillar like vortex
formed. However, the observed pressure distribution indicates that a
vortical structure must exist also in leeward orientation. This is con-
firmed by the flow field measured and discussed in Section 4.
6. Summary and conclusions
The bi-stable flow around the Ahmed body in non-symmetric
flow and the resulting aerodynamic forces were investigated. The
influence of the yawing and slant angles on the flow around the
body is studied. Aerodynamic loads and wall pressures were mea-
sured simultaneously, and the velocity field in the wake in a sep-
arate series. The flow Reynolds number was varied in the range
0.69 × 106 ≤ Re ≤ 2.73 × 106 .
With the lower slant angle ϕ = 25°, the time average forces, mo-
ments and pressure vary continuously with the yawing angle. The
Fig. 29. Pressure coefficients at ϕ = 35°, U∞ = 30 m/s, β = ±12.5°. (a) MP 1–10, (b) MP
11–17, (c) MP 18–24.
drag coefficient was found to decrease with increasing Re number,
while the lift coefficient slightly increases. The pressure distribution
and the velocity fields indicate that the vertical dimension of the typ-
ical longitudinal vortices increases or decreases with the windward
or leeward orientation of the C-pillars. With the larger slant angle
of 35°, the behaviour at small yawing angles |β | ≤ 5° is similar as
in symmetric flow. The drag is lower than with ϕ = 25°, and the
overall lift is nearly zero, which is due to the much lower pressure
level at the slant. However, except at the lowest Reynolds number,
the forces, moments and pressures change randomly between two
levels for the yawing angles 11° ≤ |β | ≤ 14°. This switching is most
pronounced near |β | ≈ 12.5° and accompanied by strong pressure
fluctuations. Such vortices are identified already at |β | = 10° above
the near-wake. The spatial extension of these vortices in the vertical
direction increases with the yawing angle. The inward rotating vor-
tex at the windward roof edge causes downwash, which reattaches
the flow to the slanted surface. The vortex is sucked down and acts
like a C-pillar vortex, leading to low pressures at the slant. On the lee-
ward edge, the pressure distribution is similar, but the corresponding
vortex rotates outward. The instability of the flow field against the
pressure fluctuations in the range 11° ≤ |β | ≤ 14°, associated to the
instability of the underbody flow for our H/W = 0.72 and C/W = 0.13
by Grandemange et al. (2013a), leads to the observed change of the
vortex positions in time and, consequently, to the changes between
states “1” and “2”.
The present investigations show that the flow under yawing con-
ditions is strongly influenced by the slant angle in a “critical” regime
where the flow pattern may change significantly between two states,
leading to a significant change of the aerodynamic forces in time.
In well-defined laboratory conditions, the switching between the
 W. Meile et al. / International Journal of Heat and Fluid Flow 57 (2016) 34–47
two states could simply be initiated by introducing thread probes
for visualization. In realistic conditions this bi-stable behaviour of
hatchback cars may even be forced by disturbances through gusty
conditions.
As a measure against this behaviour, the rounding of edges may
have a positive influence. The flow is then better attached, as shown
by Thacker et al. (2012), and pressure fluctuations as well as the un-
steadiness of the vortices may be suppressed.
Acknowledgements
The authors acknowledge the financial support of the “COMET K2
- Competence Centres for Excellent Technologies Programme” of the
Austrian Federal Ministry for Transport, Innovation and Technology
(BMVIT), the Austrian Federal Ministry of Science, Research and Econ-
omy (BMWFW), the Austrian Research Promotion Agency (FFG), the
Province of Styria and the Styrian Business Promotion Agency (SFG).
Furthermore, the support from the industrial partner MAGNA STEYR
Fahrzeugtechnik AG & Co KG (Graz) of a part of the experiments is
gratefully acknowledged. The authors gratefully acknowledge sup-
port from the Fachhochschulstudiengänge Burgenland GmbH, Aus-
tria, who provided the PIV system for the present work. Furthermore,
we are grateful to Ao. Univ.-Prof. Dr. Jakob Woisetschläger for his con-
tinuous support of the PIV measurements.
References
Ahmed S.R., Ramm G., Faltin G., 1984. Some salient features of the time-averaged
ground vehicle wake, SAE paper 840300.
Barlow, J.B., Rae jr., W.H., Pope, A., 1999. Low-Speed Wind Tunnel Testing. John Wiley
& Sons, New York.
Baxendale, A.J., Graysmith, J.L., Howell, J., Haynes, T., 1994. Comparisons between CFD
and experimental results for the Ahmed reference model. In: RAeS Conference on
Vehicle Aerodynamics. Loughborough, UK, pp. 30.1–30.11.
Bayraktar I., Landman D., Baysal O., 2001. Experimental and computational investiga-
tion of Ahmed body for ground vehicle aerodynamics, SAE paper 2001-01-2742.
Bearman P.W., De Beer D., Hamidy E., Harvey J.K., 1988. The effect of a moving floor on
wind-tunnel simulation of road vehicles. SAE paper 880245.
Benedict, L.H., Gould, R.D., 1996. Towards better uncertainty estimates for turbulence
statistics. Exp. Fluids 22, 129–136.
Conan, B., Anthoine, J., Planquart, P., 2011. Experimental aerodynamic study of a car-
type bluff body. Exp. Fluids 50, 1273–1284.
Docton, M.K.R., 1996. The Simulation of Transient Cross Winds on Passenger Vehicles.
Durham theses, Durham University.
Drage, P., Hörmann, T., Meile, W., Gabriel, A., Brenn, G., Lindbichler, G., 2008. Efficient
use of computational fluid dynamics for the aerodynamic development process in
the automotive industry. In: 26th AIAA Applied Aerodynamics Conference, 18-21
August 2008, Honolulu, Hawaii, and AIAA 2008-6735.
Ehlers M.T., Röser P., 2006. Bestimmung und Einstellung der Windgeschwindigkeit im
Windkanal der Fa. Modine, 7. Karlsruher Fahrzeugklima-Symposium, January 26,
URL: http://www.twk-karlsruhe.de/de/w_symposium_07.asp.
ERCOFTAC - European Research Community on Flow, Turbulence and Combus-
tion, 2001, 2002. http://tmdb.ws.tn.tudelft.nl/workshop9/workshop01.html
http://tmdb.ws.tn.tudelft.nl/workshop10/case9.4/case9.4.html http://www.
ercoftac.org/fileadmin/user_upload/bigfiles/sig15/database/index.html.
47
Ferrand, V., 2014. Forces and flow structures on a simplified car model exposed to an
unsteady harmonic crosswind. J. Fluids Eng 136, 011101.
Gohlke, M., Beaudoin, J.F., Amielh, M., Anselmet, F., 2007. Experimental analysis of flow
structures and forces on a 3D-bluff-body in constant cross-wind. Exp. Fluids 43,
579–594.
Grandemange, M., Cadot, O., Gohlke, M., 2012. Reflectional symmetry breaking of the
separated flow over three-dimensional bluff bodies. Phys. Rev. E 86, 035302(R).
Grandemange, M., Gohlke, M., Cadot, O., 2013. Turbulent wake past a three-
dimensional blunt body. Part 1. Global modes and bi-stability. J. Fluid Mech. 722,
51–84.
Grandemange, M., Gohlke, M., Cadot, O., 2013. Bi-stability in the turbulent wake past
parallelepiped bodies with various aspect ratios and wall effects. Phys. Fluids 25,
095103.
Gretler, W., Meile, W., 1993. Der 2m-Windkanal am Institut für Strömungslehre und
Gasdynamik der Technischen Universität Graz. ÖIAZ 138 (3), 90–96.
Guilmineau, E., Chometon, F., 2009. Effect of side wind on a simplified car model: ex-
perimental and numerical analysis. ASME J. Fluids Eng. 131, 021104.
Hucho, W.H., 2005. Aerodynamik des Automobils, fifth ed. Vieweg„ Braunschweig –
Wiesbaden.
Krajnović S., Davidson L., 2004. Large-eddy simulation of the flow around simplified
car model, SAE paper 2004-01-0227.
Krajnović, S., Davidson, L., 2005. Flow around a simplified car, Part 1: Large eddy simu-
lation. ASME J. Fluids Eng. 127, 907–918.
Krajnović, S., Davidson, L., 2005. Flow around a simplified car, Part 2: understanding
the flow. ASME J. Fluids Eng. 127, 919–928.
Krajnović, S., Sarmast, S., 2010. Numerical investigation of the influence of side winds
on a simplified car at various yaw angles. In: Proceedings of the ASME 2010 3rd
Joint US-European Fluids Engineering Summer Meeting and 8th International Con-
ference on Nanochannels, Microchannels, and Minichannels, August 1–5, 2010.
Montreal, Canada Paper Nr. FEDSM-ICNMM2010-30766.
Le Good G.M., Garry K.P., 2004. On the use of reference models in automotive aerody-
namics, SAE-paper 2004-01-1308.
Lienhart H., Stoots C., Becker S., 2000. Flow and turbulence structures in the wake of
a simplified car model (Ahmed model), DGLR Fach-Symp. AG STAB, University of
Stuttgart, November 15–17.
Lienhart H., Becker S., 2003. Flow and turbulence structures in the wake of a simplified
car model, SAE paper 2003-01-0656.
Meile, W., Brenn, G., Reppenhagen, A., Lechner, B., Fuchs, A., 2011. Experiments and
numerical simulations on the aerodynamics of the Ahmed body. CFD Lett. 3 (1),
32–39.
Meile, W., Wanker, T., Brenn, G., 2012. The unsymmetric flow around the Ahmed body.
In: Proceedings of the Conference on Modelling Fluid Flow (CMFF’12). Budapest,
Hungary, pp. 239–246. September 4-7, 2012.
Minguez, M., Pasquetti, R., Serre, E., 2008. High-order large-eddy simulation of flow
over the “Ahmed body” car model. Phys. Fluids 20, 095101.
Morel, T., 1978. The effect of base slant on the flow pattern and drag of three-
dimensional bodies with blunt ends. In: Sovran, G., Morel, T., Mason, W.T. (Eds.),
Aerodynamic Drag Mechanisms Of Bluff Bodies And Road Vehicles. Plenum Press,
New York, pp. 191–226.
Panchenko I.N., 1998. Investigation and calibration of aerodynamic balances. Rus-
sian Aerospace Science and Technology Series, Central Aerohydrodynamic Institute
(TsAGI), Zhukovsky (Russia).
Serre, E., Minguez, M., Pasquetti, R., Guilmineau, E., Deng, G.B., Kornhaas, M.,
Schäfer, M., Fröhlich, J., Hinterberger, C., Rodi, W., 2013. On simulating the turbu-
lent flow around the Ahmed body: A French-German collaborative evaluation of
LES and DNS. Comp. Fluids 78, 10–23.
Thacker, A., Aubrun, S., Leroy, A., Devinant, P., 2012. Effects of suppressing the 3D sepa-
ration on the rear slant on the flow structures around an Ahmed body. J. Wind Eng.
Ind. Aerodyn. 107-108, 237–243.
Tsubokura, M., Nakashima, T., Kitayama, M., Ikawa, Y., Doh, D.H., Kobayashi, T., 2010.
Large eddy simulation on the unsteady aerodynamic response of a road vehicle in
transient crosswinds. Int. J. Heat Fluid Flow 31, 1075–1086.
Wanker T., 2011. Der Ahmed-Körper unter Schräganströmung. Diploma Thesis. Insti-
tute of Fluid Mechanics and Heat Transfer, Graz University of Technology.