International Journal of Heat and Fluid Flow 32 (2011) 192–200

Contents lists available at ScienceDirect

International Journal of Heat and Fluid Flow

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

Numerical simulation of the flow around a simplified vehicle model

with active flow control
Siniša Krajnović *, João Fernandes
Division of Fluid Dynamics, Department of Applied Mechanics, Chalmers University of Technology, SE-412 96 Göteborg, Sweden

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

Article history: Large-eddy simulation (LES) was used to study the influence and the resulting flow mechanisms of active
Received 12 November 2009 flow control applied to a two-dimensional vehicle geometry. The LES results were validated against exist-
Received in revised form 30 March 2010 ing Particle Image Velocimetry (PIV) and force measurement data. This was followed by an exploration of
Accepted 22 June 2010
the influence of flow actuation on the near-wake flow and resulting aerodynamic forces. Not only was
Available online 15 July 2010
good agreement found with the previous experimental study, but new knowledge was gained in the form
of a complex interaction of the actuation with the coherent flow structures. The resulting time-averaged
Keywords:
flow shows a strong influence of the extension of the actuation slots and the lateral solid walls on the
Active flow control
LES
near-wake flow structures and thereby on the resulting drag.
Ahmed body Ó 2010 Elsevier Inc. All rights reserved.
Drag reduction
Simplified vehicle

1. Introduction instabilities. Such instabilities rapidly produce two large 2D vorti-

Energy-efficient ground vehicles require careful geometrical
styling to decrease drag, primarily due to the formation of the
wake region. However, commercial vehicles such as trucks and
buses do not permit any great change of the rear part of the vehi-
cle. Instead of changing their geometry, attempts can be made to
influence the flow using flow control, which can be either active
or passive. Both strategies suffer from the inability to adjust to
the flow conditions if the flow control is not a function of the actual
flow and thus cannot adapt to the flow conditions.
The development of an efficient strategy for closed-loop active
flow control requires a better understanding of open-loop active
flow control and the flow mechanisms that are likely to produce
an increase in the base pressure of ground vehicles. The goal of
the present paper is twofold: to investigate the applicability of
the LES technique for the purpose of flow control and to improve
our understanding of the flow mechanisms acting in an active flow
control process.
The present study employed a two-dimensional bluff body with
a lateral shape similar to a so called Ahmed body (Ahmed et al.,
1984) used in the experimental study by Pastoor et al. (2008).
The interaction of the upper and lower shear layers after the trail-
ing edges of the 2D Ahmed body (Fig. 1) results in von Karman-like
* Corresponding author.
E-mail address: sinisa@chalmers.se (S. Krajnović).
URL: http://www.tfd.chalmers.se/~sinisa (S. Krajnović).
ces in alternating order. As the vortices are formed very early, the
near-wake separation bubble (the dead water) is short, producing a
low base pressure and large drag. An increase of the base pressure
can be achieved by an elongation of the near-wake region and sup-
pression or delay of the shear layer interaction. To achieve this
objective, the present work applies the strategy used in Pastoor
et al. (2008) to force symmetric vortex shedding and thereby delay
the wake instabilities.
2. Description of the model and numerical set-up
Both the natural and the controlled flows were studied at two
Reynolds numbers, Re = 2  103 and Re = 2  104. The lower
Reynolds number simulations were used for parametric studies
of the location of the actuation slots, the actuation amplitude
and actuation frequencies, as the low Re allows a large number
of wall-resolved LES simulations.
The only difference between the set-up of the low and high
Reynolds numbers is that the support rods were excluded in the
case of low Reynolds number to make the simulations even less
computer intensive. The model has the side shape of an Ahmed
body (Krajnović and Davidson, 2005a,b) with no slanted surface
at the rear end. It extends from the one to the other lateral wall,
forcing the flow above and below it. The geometry of the body is
shown in Fig. 1. It has a cross section from the side of a simplified
bus with chord length L = 262 [mm], height H = 72 [mm] and span-
wise width W = 550 [mm]. The front of the body has a radius of 25

0142-727X/$ - see front matter Ó 2010 Elsevier Inc. All rights reserved.
doi:10.1016/j.ijheatfluidflow.2010.06.007
 S. Krajnović, J. Fernandes / International Journal of Heat and Fluid Flow 32 (2011) 192–200 193

(b)

(a) (c)
Fig. 1. (a) Geometry of the 2D Ahmed body. (b) Zoom of the trip tape. (c) Zoom of
the actuation slots.

[mm]. Two trip tapes of a height of 0.8 [mm] and length of 5 [mm]
were placed on the upper and the lower face of the body 30 [mm]
downstream of the front.
The position of the body in a computational domain is shown in
Fig. 2. The model was centered in the computational domain with a
height of 7.7H. The distance from the model to the inlet and the
outlet was 10.25H and 20.83H, respectively.
It is worth mentioning here that there is a dependence of the
flow separation on the Reynolds number and curvature of the sur-
face at the front of the body (Cooper, 1985). In the flow studied
here at Re 6 2  104 and a radius of the front end of the body of
R = 25 [mm], the flow is expected to separate (see Cooper, 1985).
This means that no tripping of the transition is needed. However,
in the experimental study, trip tapes are placed on both the upper
and the lower sides of the body just after the curved part. Although
they are rather small (d1 = 8 mm and d2 = 5 mm), they were in-
cluded in the LES in order to remove sources of discrepancies be-
tween the experiments and the simulations. Two actuation slots
were placed at the rear edges of the body (Fig. 3). The slots in
the experimental study were inclined at a 45° angle. However, all Fig. 3. (a) Geometry of the body with the coordinate system and reference axis.
the LES used two pairs of slots around the upper and the lower (b) Three slot configurations.
edges, as shown in Fig. 3. The reason for this discrepancy is that
a better hexahedral computational grid was obtained with this slot

Fig. 2. Geometry of the body placed in the wind tunnel in the case of Re = 2  104.
194 S. Krajnović, J. Fernandes / International Journal of Heat and Fluid Flow 32 (2011) 192–200

set-up. Nevertheless, the effect of the actuation on the flow is kept 5. Parametric study of the slot location
the same in the LES as it is in the experiments by using the same
momentum coefficient and direction of the actuating flow. Three different configurations of the slot locations were investi-
gated to find the optimal slot configuration. The three configura-
tions of the slots, denoted A, B and C, are shown in Fig. 3b. All
3. Flow control strategy
three configurations are simulated using the optimal set of actua-
tion parameters, StA = 0.17 and Cl = 0.015, identified in the experi-
The boundary layers on the upper and lower surfaces separate
mental study (Henning et al., 2006, 2008). The statistics of the drag
at the rear end of the body, forming two shear layers that enclose
force signal are given in Table 1.
the wake flow. These two shear layers very quickly become unsta-
As shown in Table 1, slot configuration C, where the slots are
ble and start to interact with each other, resulting in von Karman-
placed on both the upper and the lower sides and on the base of
like vortex shedding. The shear layers are characterized by large
the body, is the most efficient in reducing drag. Slot configurations
energy dissipation that leads to a decrease in the pressure on the
A and B also result in a reduction in drag but are less efficient. It was
rear end surface. Interaction of the shear layers will in the mean
found that the influence of the actuation is minimal with slot con-
form a rather short separation bubble, producing low pressure on
figurations A and B, and the drag signal history is similar to that in
the rear end surface and thus high pressure drag. The idea of the
the natural flow (see Fernandes, 2007). Fig. 4 shows the drag force
flow control used in Henning (2008), Pastoor et al. (2008) and in
coefficient time history of configuration C. A harmonic function,
the present work is to delay the flow instabilities in the shear lay-
C d ðtÞ ¼ C d þ 0:1uA sinðxA tÞ, is plotted in the same figure to compare
ers and thereby the shear layer interaction, and thus to make the
the actuating and the actuated flows. The main frequency of the
mean separation bubble longer. This will redistribute the low pres-
drag force signal follows the actuation frequency, xA. Furthermore,
sure in the near wake and thus increase the rear end pressure and
the peak blowing and the peak suction are correlated with the max-
decrease the drag. The rear end pressure can be increased either by
imum and minimum local drag coefficient, respectively.
delaying or preventing the interaction of the two shear layers. If
the latter is achieved and the vortex shedding of the upper shear
layer is decoupled from the vortex shedding of the lower shear
6. Parametric study of the actuation amplitude and frequency
layer, less energy will be dissipated. The alternate blowing and
suction is used in the present work to delay the shear layer insta-
A parametric study of the actuation amplitude and frequency
bilities and thereby the interaction of the shear layers. As the actu-
was made in addition to the study of the influence of the slot
ating flow is characterized by two cycles, blowing and suction, an
set-up. The previous experimental work showed that no drag
additional actuating parameter, the actuating frequency, has to be
reduction was achieved for actuation frequencies of StA < 0.1 or
specified together with the amplitude of the actuating flow. The
for actuation amplitudes where Cl > 0.015. Thus, for the paramet-
actuating frequency and amplitude are characterized by Strouhal
ric study in the present work, several frequencies below StA < 0.1 in
number StA = fAH/U1 and momentum coefficient C l ¼ 4su2A =HU 21 .
combination with actuation velocities in the range 0.015 6 Cl 6
Here, fA is the frequency of the actuating flow, uA is the actuating
0.12 were tested. Only two of the investigated cases will be pre-
velocity, s is the width of one actuating slot and H is the height
sented here. The drag coefficient statistics for two actuation fre-
of the body. In addition to these actuation parameters, there are
quencies, StA = 0.045 and StA = 0.17, and the same actuation
two other critical parameters, i.e. the location of the actuating slots
amplitude, Cl = 0.03, are shown in Table 2. Although there is a
and the angle of the actuation flow stream. The present paper will
large difference between the two actuating frequencies, the statis-
consider the influence of the former three parameters on the con-
tics of the resulting drag signal are similar. On the other hand, the
trolled flow.
actuating velocity magnitude is the parameter that should have a
large impact on the resulting flow. A low actuating velocity magni-
4. Actuation slots and actuation parameters tude will not significantly influence the actuated flow, while an
actuation using a very high magnitude of the actuating velocity
Two pairs of slots (Fig. 3) extending in the spanwise direction will dominate the flow (Fig. 5). The influence of the low frequency
are used for actuation of the flow. The slots, of a width of 1 actuation (StA < 0.1) is presented in the upper part of Fig. 5. The
[mm] and a spanwise length of 250 [mm], were placed on the drag coefficient signal of this flow is similar to the natural flow sig-
upper and the lower face (one per face) and on the rear face of nal, and it contains other frequencies superimposed on the domi-
the model (two slots). This is the only difference in the set-up con- nant actuating frequency (dashed curve in Fig. 5). Thus the wide
figuration from the experiments, where only two slots were placed spectrum of wake instabilities of the natural flow is preserved in
at the upper and lower trailing edges with an angle of 45° relative the actuated flow at low actuating frequency. Increasing the actu-
to the streamwise direction. Two pairs of actuation slots are used ating frequency to StA = 0.17 leads to a signal that contains only one
at the rear edges of the body. This slot set-up was chosen in a para- dominant frequency, namely that of the actuating flow (the lower
metric study that used different combinations of the four slots figure in Fig. 5). Finally, it was concluded that the trends observed
from Fig. 3b. The parametric study was made using the optimal in Henning et al. (2006) for the influence of the actuating parame-
actuating parameters, i.e. the non-dimensional frequency StA and ters on the actuated flow at Re = 2  104 are similar to those ob-
the momentum coefficient Cl used in the experimental study served in the present parametric study of the flow at Re = 2  103.
(Henning et al., 2006, 2008).
An extensive LES study was performed to identify the optimal
Table 1
frequency and the amplitude of the actuating flow. This was done
Summary of the parametric study of the slot location influence.
at a Reynolds number of 2  103 as this low Reynolds number
enabled a large number of wall-resolved LES. It is important to Natural flow Slot config. A Slot config. B Slot config. C
mention that the flow is different between Re = 2  103 and Cd 0.807 0.798 0.8 0.786
Re = 2  104. However, as we found later in the work the optimal rms 0.0242 0.0382 0.03 0.0341
StA and Cl of the actuating flow at low Reynolds number flow Maximum 0.8738 0.9007 0.9147 0.8681
Minimum 0.7377 0.7158 0.7371 0.7030
was the same as the optimal actuation parameters of the high Rey-
% Reduction – 1.12 0.88 2.62
nolds number flow found in the experimental study.
 S. Krajnović, J. Fernandes / International Journal of Heat and Fluid Flow 32 (2011) 192–200 195

Fig. 4. Detail of the time history of C d for slot configuration C and the function C d ðtÞ ¼ C d þ 0:1uA sinðxA tÞ corresponding to actuation with StA = 0.17 and Cl = 0.015.

6.1. Boundary conditions and actuation

Table 2
Drag coefficient statistics for the actuation amplitude and frequency study.
The Reynolds number of the flow based on the inlet velocity and
Cl = 0.03, StA = 0.045 Cl = 0.03, StA = 0.17 the height of the model was ReH = 2  103 for the parametric study
Cd 0.7963 0.7949 and ReH = 2  104 for the final case. A constant velocity profile was
rms 0.0478 0.0478 applied at the inlet, while the homogeneous Neumann boundary
Maximum 0.8987 0.9246 condition was used at the outlet. A no-slip boundary condition
Minimum 0.6797 0.6696
was applied on the surface of the body and on the upper, lower
% Reduction 1.39 1.57
and lateral walls.

Controlled Flow: Cµ=0.03, St =0.045

A

0.85

0.8
C
d
0.75

0.7

260 280 300 320 340 360 380 400 420 440
time

Controlled Flow: C =0.03, St =0.17

µ A

0.85

0.8
C
d
0.75

0.7

250 300 350 400 450 500 550 600 650

time

Fig. 5. (Upper figure) Detail of the time history of Cd for the controlled flow with Cl = 0.03 and StA = 0.045. Dash-dotted curve shows a harmonic function with a frequency of
xA corresponding to StA = 0.045 and mean value C d . (Lower figure) Time history of Cd for the controlled flow using Cl = 0.03 and StA = 0.017.
196 S. Krajnović, J. Fernandes / International Journal of Heat and Fluid Flow 32 (2011) 192–200

Harmonic actuation was applied to each slot by imposing a

velocity on the boundary equal to ~ us ¼ uA sinðxA tÞðcosð/Þ~
iþ
sinð/Þ~jÞ, where ~
i and ~
j are the unit vectors in the x and y directions,
respectively, and the actuation angle, /, was 45°, in agreement
with the experiments. The actuation amplitude, uA, can be derived
from the expression for the momentum coefficient, C l ¼
4su2A =HU 1 . The actuation frequency and the moment coefficients
in the present study were StA = 0.17 and Cl = 0.015, respectively.

6.2. Numerical simulations

Large-eddy simulations using the standard Smagorinsky model

with the constant CS = 0.1 were made with a commercial finite vol-
ume solver, STAR-CD. A blend of a second-order central difference
and second-order upwind scheme in the form of the so called
MARS scheme (monotone advection and reconstruction scheme)
was used for the spatial discretization of the convective terms.
The blending factor in the MARS scheme was set to 1, which uses
a maximal portion of central differences. An implicit second-order
scheme (three-time-level scheme) was used for the temporal dis-
cretization. In this temporal scheme a quadratic variation between
three time levels is used to compute time derivative terms in the
momentum equations.
Structured computational grids containing a total of 2.8  106
and 5.5  106 nodes were used for the cases at Re = 2  103 and
Re = 2  104, respectively. The same computational grids were
used for both the natural and the controlled flow simulations. A
resolution expressed in the wall units for the low Reynolds number
case was y+  0.3, Dx+  6 and Dz+  35 (Fig. 6) in the wall-normal,
streamwise and spanwise directions, respectively. The same reso-
lution for the high Reynolds number case was approximately
y+ = 1, Dx+ = 14 and Dz+ = 30 (Fig. 7). A non-dimensional time step,
t* = tU1/H, was 0.0116 and 0.0058 in the low and the high Reynolds
number cases, respectively. These time steps resulted in maximum
CFL numbers of approximately 7.3 and 8.5 in the natural and con-
trolled flow simulations, respectively, in the case with low Rey-
nolds number. Here the CFL number for the controlled case refers
to the case with optimal actuation parameters. Corresponding
maximum CFL numbers for the high Reynolds number case were
0.3 and 8.5 in the natural and controlled flow simulations, respec-
tively. Furthermore, the CFL number is smaller than one in 96.5% of
the computational cells during the entire simulation. All simula-
tions were first run until the values of the aerodynamic coefficients
stabilized before averaging was initiated. The simulation with ac-
tive flow control was initiated from the natural flow after
t  215. The averaging for either the natural or the controlled flow
was made during t  230.
The number of computational cells in the actuation region is an
important parameter for accurate prediction of the actuated flow.
Previous numerical studies of actuated flows used very different
Fig. 7. Side view of the computational grid used for LES of the flow at Re = 2  104.
numbers of cells per slot (Ekaterinaris, 2004; Duvigneau and
Visonneau, 2006; Spalart et al., 2003). In the present work 5  9
computational cells were used in the streamwise/transverse and
spanwise directions for the resolution of the flow around each slot
(Fig. 6b).
7. Results
7.1. Validation of the LES results
The LES results were validated against the experimental data re-
ported in Henning (2008). The experimental data contain the Cd
values and the Particle Image Velocimetry (PIV) data. Although
the Reynolds number in PIV measurements (Henning, 2008) was
identical to the one used in the present LES, the Reynolds number
in the flow drag measurement experiments was some 4% higher
(i.e. Re = 2.3  104). To obtain the time-averaged flow, 2000 and
1600 instantaneous uncorrelated PIV time samples were used in
the natural and controlled flows, respectively. Regular statistical
averaging was done, i.e. no spurious replacement or interpolation
algorithms (common in PIV data post processing) were used in
the present work. Other details of the PIV investigation are given
in Henning (2008). A comparison of the velocity profiles in the
present LES with previous PIV data is shown in Figs. 8 and 9 for
the natural and the controlled flows respectively. The comparison
in Figs. 8 and 9 shows good agreement between the LES results
and the PIV data. However, the discrepancy between the LES and
PIV profiles is slightly larger for the controlled than for the natural
flow. The explanation for this is that the actuation applied in the
controlled flow increases the local velocity and, as the computa-
tional grid is identical in the natural and the controlled flow, the
resolution (expressed in wall units) is poorer in the LES of the con-
trolled flow. For example, it was found that the maximum CFL

Fig. 6. (a) Computational grid used for LES of the flow at Re = 2  103. (b) Zoom of the computational grid around the slots around the upper edge.
 S. Krajnović, J. Fernandes / International Journal of Heat and Fluid Flow 32 (2011) 192–200 197

1 1 1

0.5 0.5 0.5

0 0 0

−0.5 −0.5 −0.5

−1 −1 −1
−0.5 0 0.5 1 1.5 −0.5 0 0.5 1 1.5 −0.5 0 0.5 1 1.5

1 1 1

0.5 0.5 0.5

0 0 0

−0.5 −0.5 −0.5

−1 −1 −1
−0.5 0 0.5 −0.5 0 0.5 −0.5 0 0.5

Fig. 8. Natural flow. Comparison of LES (solid line) and PIV (+) from Henning (2008).

Fig. 9. Controlled flow. Comparison of LES (solid line) and PIV (+) from Henning (2008).

number was 9.7 in the cases at Cl = 0.03, and this represents an
increase of almost 15% compared to the maximum CFL obtained
at Cl = 0.015. Even more important is the fact that, although the
number of cells with higher CFL number, CFL > 7, constitutes only
0.1% of the total number of cells, all are located in the near-slot
region, one or two cell layers immediately after the slot surface.
This represents a problem in transient computations with active
flow control, since both extremely low spatial and time resolutions
are needed to resolve the flow around the actuation region without
a significant loss of accuracy. Although the computational grid was
198 S. Krajnović, J. Fernandes / International Journal of Heat and Fluid Flow 32 (2011) 192–200

not refined in the controlled flow compared to the natural one,
good agreement was obtained between the LES results and the
PIV data. This suggests that the resolution used in the present work
is sufficient for predictions of both natural and controlled flows.
Fig. 10a shows that the actuation dominates the near-wake
flow, resulting in a drag signal that is almost aligned with the sig-
nal of the actuation (Fig. 10b). Fig. 10b shows that blowing and
suction produce maximum and minimum drag, respectively. The
reason for this is found in the 2D coherent structures observed in
the natural flow that are also formed during the blowing sequence
of the actuation but are destroyed in the course of the suction
cycle.
7.2. Drag reduction using flow control
The LES of the natural and the controlled flows resulted in mean
Cd values of 1.02 and 0.92, i.e. a decrease of about 11% using flow
actuation. This can be compared with 0.98 and 0.83 obtained in
the experimental study at a slightly higher Reynolds number,
Re = 2.3  104 (Henning et al., 2006, 2008), for the natural and
the controlled flow, respectively. The source of the relatively large
difference (10%) in the drag coefficient for the controlled case be-
tween the experiment and the LES is not clear. There are two pos-
sible sources of error: the small difference in the slot set-up
configurations between the experiment and the LES and the use
of a blend of central difference and second-order upwind schemes
for the convective terms, which might be too dissipative for LES.
The difference in the drag results between the LES and measure-
ments of around 4% (C LES d ¼ 1:02 and C EXP
d ¼ 0:98) already exists
in the natural flow. (This can be compared with the measuring
accuracy in the experimental study of only ±0.35% (Pastoor et al.,
2008).) The question is whether the slot geometry is the origin of
the larger discrepancy between the drag results in LES and exper-
iments in the controlled flow (10%). There is no way to answer
this question without making LES with exactly the same slot con-
figuration used in the experimental study. At the present time it
can only be observed that the velocity profiles in the LES are in very
good agreement with the PIV data implying that the numerical
scheme is appropriate. One explanation for the increase in the dif-
ference in the drag coefficient between the present LES and the
measurements in the controlled flow could be the above men-
tioned increase in the local CFL number close to the slots in the
controlled flow (compared to the natural flow). However, such a
large difference in drag is not justified by the relatively small dif-
ference in the LES and the PIV velocity fields (see Fig. 9).
7.3. Flow mechanisms of flow control
A comparison of Figs. 11, 12A and C shows that blowing post-
pones the rollover of the shear layers and the formation of strong
2D alternating vortices. On the other hand, suction leads to strong
perturbations of the 2D structures and their breakdown into small
coherent structures (Fig. 12B and D). Note, however, that even
these structures are predominantly 2D. Figs. 11 and 12D indicate
that the actuation strategy used in the present work is efficient
in postponing the vortex formation. For example, it is obvious that,
during blowing, the length of the free shear layer in Fig. 12A is
approximately three times that of the natural flow case in

0.95
0.9
0.85
0.8
0.75

(a) (b)

Fig. 10. (a) Time history of the drag coefficient for the natural (upper curve) and the controlled flow (lower curve). (b) Partial time history of drag and lift coefficients for the
controlled flow. Positions A and C are locations of maximal blowing, while positions B and D are locations of maximal suction. E and F are intermediate points between
blowing and suction. Dash-dotted curve shows a harmonic function with a frequency of xA corresponding to StA = 0.17 and mean values C  d and C
l .

Fig. 11. Velocity vector field and vorticity, xz (left figure), and an isosurface of p = 12, colored with xz for natural flow (right figure). (For interpretation of the references to
colour in this figure legend, the reader is referred to the web version of this article.)
 S. Krajnović, J. Fernandes / International Journal of Heat and Fluid Flow 32 (2011) 192–200 199

Fig. 12. Velocity vector field and vorticity, xz (upper figures), and isosurface of p = 12, colored with xz for maximal (A and C) and minimal (B and D) values of drag
coefficient Cd in Fig. 10. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 11. Still, all large coherent structures are efficiently destroyed tween 0.1 < St < 0.3) which is in line with the break-up of large
during the suction phase (Fig. 12B and D). Both cycles together alternating vortices into a large number of smaller coherent struc-
thus synchronize shear layers and thereby influence wake dynam- tures with different characteristic frequencies discussed above.
ics such that the interaction of vortices rolling up from the two
shear layers is decreased or delayed. Both result in an increase of
the base pressure and a reduction in drag.
7.5. Discussion and conclusions

7.4. Dominant frequencies in the flow
Power density spectra of drag and lift signals were computed
for the natural and controlled flows. Fig. 13 shows the more inter-
esting spectral picture of the drag signal. A very clear dominant fre-
quency was found at Stnat = 0.31 in the natural flow case. This can
be compared with several dominant frequencies of St EXP nat ¼ 0:23
0:25 found in the experimental results of Henning et al. (2006),
Pastoor et al. (2008). However, in the controlled flow, the dominant
frequency was found at St = 0.17, and St = 0.31 was the second peak
frequency. This result is interesting as it shows that the actuation
(at Stact = 0.17) dominates the response and that the natural flow
shedding frequency remains superimposed on the drag signal
(note that the frequency of the natural flow, Stnat = 0.31, is identical
to the second peak frequency in the controlled flow). The con-
trolled flow signal transformed into the frequency space also
shows a considerably larger number of distinct frequencies (be-
Fig. 14 shows the vortex cores together with the streamlines
projected onto the symmetry plane and a plane in the vicinity of
one lateral wall of the domain. As seen in this figure, the actuation
has a greater impact on the spanwise part of the near wake where
the actuation was not applied, i.e. close to the lateral walls. Stream-
lines in Fig. 14b and e show that there is an elongation of the near-
wake separation bubble in the center plane. However, this effect is
much stronger close to the lateral wall (Fig. 14c and f). This is an
unexpected result and could be a consequence of the presence of
solid lateral walls. This raises questions about the efficiency of
the proposed actuation strategy in the case of a 3D Ahmed body
(Krajnović and Davidson, 2005a). It further suggests an elongation
of the spanwise length of the computational domain. An elongation
of the domain of this kind would reduce the influence of the
boundary layer and the coherent structures (horseshoe vortices
formed close to the lateral walls in the interaction of the boundary
layer with the body) on the actuated near wake.

(a) (b)

Fig. 13. Power spectral density of the drag signal for (a) natural flow and (b) controlled flow.
200 S. Krajnović, J. Fernandes / International Journal of Heat and Fluid Flow 32 (2011) 192–200
Fig. 14. Time-averaged flows for natural (a–c) and controlled flows (d–f). Figures (a) and (d) show a view from above the body and vortex cores and particle traces on the
body surface. The body is colored with the surface pressure. Figures (b), (c), (e) and (f) show streamlines projected on plane z = 0 (b and e) and z/H = 3.47 (c and f). (For
interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
The present work shows that the LES can be used to investigate
the active flow control for a reduction of drag in generic vehicles at
moderate Reynolds numbers. Although the increase of the local
velocity results in high CFL numbers in the regions of the actuation,
the LES results were found to be in good agreement with the
experimental data. LES practitioners often find that a pure central
differencing scheme leads to convergence problems and oscilla-
tions in the solution of bluff body flows. Thus a blend of central
differences with upwind schemes is often used in LES of bluff body
flows. The influence of the numerical diffusion from the hybrid
(blend of central differences and upwind schemes) numerical
scheme in the present work was found to be limited as long as
the largest portion of the scheme comes from the central differenc-
ing scheme (in a blend with an upwind scheme). This is a good
news, as the shear layer instabilities, which are important for the
formation of the wake flow, can be accurately predicted with a hy-
brid scheme.
Krajnović et al. (2009) recently made several LES of active flow
control of flow around a 3D Ahmed body with a rear end surface
slanted at a 25° angle. Several different actuators influencing both
the separation region on the slanted surface and the cone-like trail-
ing vortices were used. However, only one of the flow regions was
influenced at a time. This new study by Krajnović et al. (2009) was
only partly successful and no final consclusions could be drawn as
to the LES applicability for flow control around a 3D Ahmed body.
This is ongoing work, and one important issue that must be inves-
tigated is the resolution requirements in the flow regions near
actuators.
The present investigation used only open loop control, and the
natural step is to proceed with LES of closed loop control, similar
to that applied in Henning et al. (2006), Pastoor et al. (2008). The
challenge in a closed loop control LES is in simulating the time it
takes for the flow to adjust to the new actuating parameters (i.e.
the frequency and amplitude). If successful, such investigations
of closed loop control using LES rather than experiments will help
in a faster development of efficient means of active flow control for
a drag reduction in ground vehicles.
Acknowledgments
The work presented here was supported by Banverket (the
Swedish National Rail Administration). CD-adapco contributed li-
censes for the solver, STAR-CD. Computer time at SNIC (the Swed-
ish National Infrastructure for Computing) at the Center for
Scientific Computing at Chalmers (C3SE) is gratefully acknowl-
edged. The help of Dr. Lars Henning at the Measurement and Con-
trol Group at TU Berlin in providing the experimental data and
details about the experiments is gratefully acknowledged.
References
Ahmed, S.R., Ramm, G., Faltin, G., 1984. Some salient features of the time averaged
ground vehicle wake. SAE Paper No. 840300.
Cooper, K.R., 1985. The effect of front-edge rounding and rear edge shaping on the
aerodynamic drag of bluff vehicles in ground proximity. SAE Paper No. 850288.
Duvigneau, R., Visonneau, M., 2006. Optimization of a synthetic jet actuator for
aerodynamic stall control. Computers & Fluids 35, 624–638.
Ekaterinaris, J.A., 2004. Prediction of active flow control performance on airfoils and
wings. Aerospace Science and Technology 8, 401–410.
Fernandes, J., 2007. Large-eddy simulation of the flow around a vehicle model with
active drag control. Master’s thesis. rept. 2007:66, Dept. of Applied Mechanics,
Chalmers University of Technology, Gothenburg.
Henning, L., 2008. Regelung abgelöster scherschichten durch aktive beeinflussung.
phd thesis. Ph.D. thesis, Technischen Universität Berlin, Germany.
Henning, L., Pastoor, M., King, R., Noack, B., Tadmor, G., 2006. Feedback control
applied to the bluff body wake. In: King, R. (Ed.), Notes on Numerical Fluid
Mechanics and Multidisciplinary Design (NNFM), 1st Active Flow Control
Conference (AFC 2006). Springer, Berlin, Germany.
Krajnović, S., Davidson, L., 2005a. Flow around a simplified car, part 1: large eddy
simulation. ASME: Journal of Fluids Engineering 127, 907–918.
Krajnović, S., Davidson, L., 2005b. Flow around a simplified car, part 2:
understanding the flow. ASME: Journal of Fluids Engineering 127, 919–928.
Krajnović, S., Östh, J., Basara, B., 2009. LES of the flow control around an Ahmed body
with active flow control. In: Conference on Modelling Fluid Flow (CMFF’09), The
14th International Conference on Fluid Flow Technologies, Budapest, Hungary,
9–12 September.
Pastoor, M., Henning, L., Noack, B.R., King, R., Tadmor, G., 2008. Feedback shear
layer control for bluff body drag reduction. Journal of Fluid Mechanics 608,
161–196.
Spalart, P., Hedges, L., Shur, M., Travin, A., 2003. Simulation of active flow control on
a stalled airfoil. Flow, Turbulence and Combustion 71, 361–373.