Journal of Wind Engineering & Industrial Aerodynamics 205 (2020) 104300

Contents lists available at ScienceDirect

Journal of Wind Engineering & Industrial Aerodynamics

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

Aerodynamics of curved underbody diffusers using CFD

Angel Huminic *, Gabriela Huminic
Aerodynamics Laboratory, Transilvania University of Brasov, Brasov, 500036, Romania

A R T I C L E I N F O A B S T R A C T

Keywords: This paper presents new results concerning aerodynamics of a bluff body fitted with non-flat underbody diffusers.
Aerodynamics For the current study a computational approach is used, taking the advantage that CFD has become a valuable and
Bluff body important tool in road vehicle aerodynamics. There are analyzed two basic shapes, namely circular and the
Curved underbody diffuser
elliptical surfaces. As in previous investigations performed, angle and length of the diffuser are the parameters
CFD
systematically varied within ranges relevant for a hatchback passenger car. Coefficients of lift and drag are
computed and compared with the values obtained for the flat underbody diffuser. In order to get a better over-
view, 3D maps of both lift and drag coefficients were generated using polynomial interpolation of their values.
The results reveal significant improvements concerning the lift generated by body fitted with curved diffusers,
which are emphasised by an average growth of the downforce larger than 20%. In addition, for small and
moderate angles, and small diffuser lengths, the curved diffusers lead also to smaller values of drag, compared to
the plane diffusers. Thus, short diffusers with curved sections work better from aerodynamic point of view. In
terms of vehicles design, short diffusers have also the advantage of small interferences with other underbody
components of cars, as rear wheelhouses and rear silencer. In addition, they offer a significant larger cargo space.

underbody, creating thus a depression, which is similar for both situa-

1. Introduction
Because aerodynamics contributes significantly to performances of
road vehicles, designers pay recently much attention to aero control
packages of cars. For high-speed vehicles, these could consist on various
aero elements, as wings, air splitters, flaps, side skirts, underbody dif-
fusers etc. The later ones are recently widely used also for mass pro-
duction cars due to the advantages they could generate, namely
reduction of both drag and lift, which have a significant impact on fuel
consumption, aeroacoustics, stability and handling of vehicles.
Concerning aero elements, most of studies performed have focused on
aerodynamics of cars bodies fitted with simple plane underbody dif-
fusers. As revealed in the recent review of Ehirim et al. (2018a), signif-
icant knowledge has been already acquired in this field regarding the
benefits induced by this device, also concerning the physics and phe-
nomena that describe the flow through diffusers. Summarizing the con-
clusions of previous studies, the air flow underneath cars with small
ground clearances may generates the Venturi effect, namely a pressure
fall in the constricted section between body and ground, similarly to the
flows through constricted tubes, as depicted in Fig. 1, which reveals cp
variation in the symmetry plane of a bluff body.
As shown, the flow accelerates in the first stage around the inlet of
tions. It follows a pressure recovery beneath the body, back to the free
stream pressure. This takes place continuously for the flat underbody,
and in two stages for the underbody with diffuser, due to a new accel-
eration of the flow around the diffuser’s first edge, which is marked by a
cusp point in the variation of pressure coefficient. A third acceleration of
the air flow occurs around the second edge of the diffuser, which is
shown by a small sudden drop of pressure coefficient, as revealed in the
detail of Fig. 1. In this way the depression and downforce generated in-
crease. Therefore, through the diffuser, a part of the kinetic energy of the
fluid turns into potential pressure energy. In addition, the pressure drop
at the rear end of the body is significantly smaller in the case of the
underbody with diffuser. This contributes to smaller wake and conse-
quently to diminution of drag. Hence, diffusers may be also used to
control boundary layer separation of flow underneath cars, in order to
avoid large turbulence phenomena behind them.
Several recent studies have proposed new solutions concerning the
geometry of these aero devices. Thus, Huminic et al. (2016) investigated
experimentally the effect of a combined plane-curved underbody diffuser
for a realistic car model, and reveal better performances compared to
those of the simple flat diffuser. Using a generic body of car on wheels,
Rossitto et al. (2017) investigated the afterbody rounding on the top part
of the body, and the effect of a curved underbody diffuser for two

* Corresponding author.
E-mail address: angel.h@unitbv.ro (A. Huminic).

https://doi.org/10.1016/j.jweia.2020.104300
Received 30 January 2020; Received in revised form 4 July 2020; Accepted 4 July 2020
Available online xxxx
0167-6105/© 2020 Elsevier Ltd. All rights reserved.
A. Huminic, G. Huminic Journal of Wind Engineering & Industrial Aerodynamics 205 (2020) 104300

Nomenclature p pressure, N/m2

p∞ free stream pressure, N/m2
A reference area of body, m2 T turbulence level
cD drag coefficient tair air temperature

cL lift coefficient Cv velocity, m/s
cp pressure coefficient v∞ free stream velocity, m/s
D drag x, y, z coordinates of the reference frame
Nk turbulence kinetic energy m yþ dimensionless wall coordinate
m2/s2 cD drag coefficient α upper slant angle

h height of the body δ deviation of CFD results from experiment
mL lift ε turbulence eddy dissipation, m2/s3
Nl length of the body ϕd angle of underbody diffuser

m lb length of the computational domain behind body μair air viscosity
m ld length of the underbody diffuser kg/(m∙s) ρair air density, kg/m3
m rd radius of the underbody diffuser ω turbulence eddy frequency, 1/s
m Re Reynolds number

Fig. 1. Variation of pressure coefficient (cp) beneath a bluff body, Huminic

et al. (2012).

configurations. Kekus and Angland (2018) also examined an Ahmed type Fig. 2. Ahmed body, 3D view and side view, plane (a), circular (b), and ellip-
tical underbody diffuser (c) (dimensions are in mm).
body fitted with two-steps underbody plane diffuser, and Ehirim et al.
(2018b) studied a plane underbody diffuser having a convex bump near
the end. Mair (1969), Grandemange et al. (2015) and Bonnavion and current study. Thus, several CFD (Computational Fluid Dynamics) ana-
Cadot (2019) also analyzed the effects of underbody diffusers with lyses were performed, as in previous investigation Huminic and Huminic
boat-tail shape on aerodynamics of Ahmed type bluff bodies. All (2017), using the facilities provided by ANSYS CFX. In the first stage, the
mentioned works highlight the positive effect of the underbody diffusers accuracy and validity of the adopted computational model and procedure
on the aerodynamics of the studied bodies. were checked using the experimental results of Strachan et al. (2007).
Following this basic idea, to increase the depression under cars, The main stages of the simulations are described in the following
current paper presents new results concerning the aerodynamics of a sections.
bluff body fitted with non-flat underbody diffusers. There are analyzed
two basic shapes, namely circular and elliptical surfaces. The angle (ϕd)
2.1. CAD and computational models
and the length of the diffuser (ld) are the parameters systematically
varied within ranges relevant for a hatchback passenger car. Coefficients
The simulations were carried out using the bluff body of Ahmed et al.
of lift (cL) and drag (cD) are compared with the values obtained for the
(1984), because its aerodynamics has been widely investigated in many
flat underbody diffuser, and the results reveal significant improvements
studies and it is very well known, as mentioned by Rao et al. (2018). It is
of the aerodynamic characteristics of the body fitted with curved dif-
1044 mm long, 389 mm wide and 288 mm height, and consists of three
fusers, mainly regarding the downforce generated. Thus, current work
sections: a short curved one in front, followed by a long straight middle
represents the first systematic study concerning curved underbody dif-
section and an upper angled (α) rear end. For current study, the angle α ¼
fusers in two well-defined configurations, which reveals not only their
35 was adopted because it corresponds to a low drag (cD ¼ 0.275) and
benefits as in already published research, but also their limitations.
stable aerodynamic configuration, as shown in Fig. 5. Thus, it represents
a generic 1:4 scale model of lower-medium size hatchback vehicles, as
2. Research methodology
depicted in Fig. 2, which show also the lateral view of the bodies
investigated in the current study. Dashed lines show the contours of
Based on amount of work and the large number of models involved
previous diffusers.
during investigation, a computational approach was adopted for the
For the first body, a circular sector of radius rd was taken into account

2
A. Huminic, G. Huminic Journal of Wind Engineering & Industrial Aerodynamics 205 (2020) 104300

The size of the first layer of cells around it was set to 5∙10-4 m in order to
fulfil the condition 30  yþ100 for computations using the logarithmic
wall function approach, and a growth rate of 1.1 was assigned. This
approach represents a good compromise between the accuracy and cost-
efficiency of CFD simulations, and produces consistent results at high
Reynolds numbers, as in road vehicle aerodynamics. Although a low y þ
approach (yþ1) is desired because it resolves almost whole spectrum of
eddies, the effort and computational resources needed are still too large
for many practical applications. In addition, the low y þ wall treatment
has a more restricted range of application, being suitable mainly for low
Reynolds turbulence models, since it assumes that the viscous sublayer is
properly resolved, as stated by Fu et al. (2018). Ten layers were gener-
ated also for the surface representing the ground. Since simulations are
performed using moving ground, as mentioned in the next paragraph, its
influence on the velocity profile is not significant, since the velocity of
ground equals the velocity of free stream. A detail of mesh around body
Fig. 3. Computational domain (not to scale) and Mesh detail around body. in in the symmetry plane is shown in Fig. 3. The number of elements for
the symmetry plane. entire computational domain is larger than 8600000, of which more than
190000 elements are on surfaces of the body with elliptical diffusers. The
for the underbody diffuser, having the beginning tangent (T) to the flat results concerning the independence of the solutions with the size of
surface of the underbody, computational domain and mesh are presented in the section devoted to
validation of the CFD procedure.
rd ¼ ld2þhd22hd, (1)

where hd is the height of the diffuser 2.2. Simulations setup and boundary conditions

hd ¼ ld tgϕd. (2) The used code solves the full Reynolds-Averaged Navier-Stokes
equations in their conservation form: continuity equation and mo-
For the second one, a quarter elliptical shape was considered. Thus,
mentum equation. Unsteady analyses (URANS method) were performed
both ends of the diffuser are tangent to the flat surfaces of the body with
for incompressible, isothermal and fully turbulent flows (T ¼ 0.2%) in
which they are in connection. Hence, the length ld and the height hd of
conditions of the International Standard Atmosphere: p∞ ¼ 101325 N/
the diffuser are the semi-minor axis of an ellipse. Therefore, shapes of
m2 p∞ ¼ 1 At.p∞ ¼ 1At.(reference pressure) and tair ¼ 15  C T∞ ¼
both curved diffusers can be expressed mathematically. The length and
288.15 KT∞ ¼ 288.15 K (reference temperature of the air). The solver
angle are considered in the same way as for the plane diffuser.
uses these parameters to compute density and viscosity of the air, ρair ¼
In the next step, CAD models were integrated into a rectangular
1.225 kg/m3 and μair ¼ 1.78938∙10-5 kg/(m∙s) respectively. Shear-
enclosure computational domain, as shown in Fig. 3. Generally, the size
Stress-Transport (SST) closure model of Menter (1994) was employed
of computational domains depends on many factors, as complexity of
to perform current simulations. Using a blending function to switch from
geometry and simulation, computational approach and code used to
k-ω model in the near-wall region to k-ε model in the far field of flow, it
simulate the studied phenomena, computational grid and concatenation
combines the advantages of both mentioned turbulence models. This
of elements in areas of interests. Even though there is no predefined
results in a major improvement concerning the prediction of the aero-
solution concerning this aspect, studies involving this bluff body, or
dynamic loads and flow separation. The performance of SST turbulence
similar to it, as those of Guilmineau (2008), Cheng et al. (2011) and Rao
model has been tested successfully in various studies involving fluid
et al. (2018), are performed using rectangular computational domains
dynamics phenomena, including external aerodynamics, as the studies
having dimensions within the following ranges: 2 l–3 l upstream the body
performed by Fu et al. (2018), Urquhart et al. (2018) and Mohottia et al.
and 3 l–6.8 l downstream, 0.7 l–1.8 l on both sides of the body and 1.1
(2019).
l–1.59 l above, relative to the length (l) of the body.
In order to accelerate the convergence of the analyses, steady state
These dimensions allowed blockage ratio smaller than 2%, which is
simulations were conducted in the first stage. Their solutions were used
within the recommended value of 5% in wind tunnel testing of road
in the next stage as initial conditions for transient analyses. These were
vehicles, as stated by Hucho and Sovran (1993). Based on studies
performed for a time of 1 s with constant time steps of 0.02 s. Thus, the
mentioned above, current computational domain has the following
duration of transient analyses is almost 2.5 times larger than physical
dimensions:
timescale of simulations (0.40455 s), which is the time taken for a point
in the flow to make its way through the fluid domain. It is computed as
- 2.5 l in front of the body,
length of the computational domain divided by the velocity of free
- 12 l behind the body,
stream. The time step was considered in such way that it is smaller than
- 1.5 l for top and lateral boundaries.
the time necessary for a point in the flow to cover a distance equal to the
characteristic length of the flow (length of the body) with the velocity v∞
According to previous works Huminic et al. (2012) and Huminic and
(0.0261 s). At each time step, flow variables were evaluated for cycles of
Huminic (2017), a larger length was considered behind the body, which
10 iterations. In order to achieve high precision results, residual target of
is placed 50 mm above the ground, as in the original study of Ahmed,
RMS was set up to 1e-006. In addition, history of both drag and lift co-
isolated insight the computational domain. In this way, no unnecessary
efficients was also monitored, and it was found that their variations
interference affects the flow around it. The blockage ratio for current
become insignificant during final stage of simulations. The results were
study is 1.7%, significantly smaller than in a wind tunnel test chamber, to
collected for the final step of each simulation.
avoid negative pressure effect between body and domain boundaries.
According to the frame in Fig. 3, the following boundary conditions
A multi-block scheme was used to generate the mesh of the compu-
were considered:
tational domain. It consists on tetrahedral and pentahedral elements in
- a uniform and constant velocity at inlet, vx ¼ v∞, vy ¼ vz ¼ 0, and
the proximity of the solid surfaces, in order to solve accurately the flow
also for ground, as for moving walls, in order to simulate the relative
inside the boundary layer, which consists of 30 layers around the body.
motion between vehicle and road; in this way, all the on-road boundary

3
A. Huminic, G. Huminic
Table 1
Change in drag and lift with size of the computational domain.
lb/l 6 9 12
ΔcD – 0.0008 0.0004
ΔcL – 0.001 0.0005
Table 2
Change in drag and lift with size of the mesh on surfaces of the body for lb/l ¼ 12.
No. of elements 81628 125784
ΔcD – 0.0004
ΔcL – 0.0002
Fig. 4. Current cL against experiments.
layer properties can be matched properly during simulations, this aspect
being essential to get appropriate values of the aerodynamic coefficients.
- a zero pressure condition, p ¼ 0p ¼ 0p ¼ 0, at the outlet boundary, as
for fully developed flow behind body,
- no slip conditions on the surfaces of the body, vx ¼ vy ¼ vz ¼ 0,
- free slip conditions for the top and lateral surfaces, as for fluid
boundaries.
The reference velocity of air was set to v∞ ¼ 40 m/s as in the original
study of Ahmed, and also in many studies concerning this bluff body, e.g.
the study of Guilmineau (2008). It corresponds to a Reynolds number Re
¼ 2.86∙106 based on the length of the body.
2.3. Validation of the CFD procedure
Accuracy of the results provided by the computational model and CFD
procedure used in the current study have been checked against the
experimental results provided Strachan et al. (2007) with moving ground
(MG). Thus, several simulations were performed in similar conditions
employed in these experiments. The independence of solutions with the
size of computational domain and mesh was checked for several di-
mensions of the length behind the Ahmed body with α ¼ 35 , and the
results are shown in Tables 1 and 2, where ΔcD ¼ cDi-cDi-1 and ΔcL ¼
cLi-cLi-1 are the changes in drag and lift due to variations of lb and mesh
size on the surfaces of body, which is the most influential parameter.
Sumantran and Sovran (1996) point out that CFD simulations must be
able to discriminate a change in drag as small as ΔcD ¼ 0.002, as in the
case of experiments in wind tunnels. In order to fulfil this requirement,
and make the variation of cD independent by changes of the
Journal of Wind Engineering & Industrial Aerodynamics 205 (2020) 104300
15
0.0002
0.0004
163424
0.0003
0.0001
Fig. 5. Current cD against experiments.
Fig. 6. Dimensionless velocity contours vz/v∞ on yOz plane, (x/l) ¼ 0 and α ¼
25 , adapted from Strachan et al. (2007).
computational domain parameters, a ten times smaller value ΔcDffi 10-4
was considered the limit between two CFD analyses in row. Based on
above results it was concluded that simulations do not depend with the
size and mesh of the computational domain, as mentioned in the para-
graph devoted to CAD and computational models.
The results concerning aerodynamic coefficients are depicted in
Figs. 4 and 5, for all angles (α) of the upper slanted rear surface of the
Ahmed body. Current CFD results concerning drag coefficient are also
compared with the values of Graysmith et al. (1994) and Ahmed et al.
(1984) obtained for stationary ground (SG).
As shown, current values of the lift coefficient estimate closely the
experiments of Strachan for angle up to α ¼ 25 and the experiments of
Graysmith for larger angles. The maximum relative deviation (δ) for all
experiments is smaller than 10%, and they are due to the presence of
sting in the experimental setups, which induces errors as proven by
Hetherington and Sims-Williams (2006).
4
A. Huminic, G. Huminic Journal of Wind Engineering & Industrial Aerodynamics 205 (2020) 104300

Fig. 7. Variation of vz/v∞ on (yOz) plane, (x/l) ¼ 0, (z/l) ¼ 0.26 and α ¼ 25 , Fig. 8. 3D maps of cL ðld ; ϕd Þ for plane (P), circular (C) and elliptical
adapted from Strachan et al. (2007). (E) diffusers.

δ ¼ 100cexp-cCFD/cexp %. (3)

Concerning the drag, there is a fine prediction of Strachan experiment

for α  30 , when the flow over the upper angled surface becomes fully
separated and the variations of aerodynamic forces are less significant.
For the case examined in the current investigations, α ¼ 35 , the devia-
tion from the experiments is smaller than 1%.
There were also examined the velocity variation in the wake, at the
end of the body, using results of LDA (Laser Doppler Anemometry) in-
vestigations provided by Strachan et al. (2007), as revealed in Fig. 6. The
study of Strachan et al. (2007) also includes the results of Lienhart et al.
(2002) for α ¼ 25 . This slant angle is one of the most studied due to its
strongly three-dimensional wake as stated by Guilmineau (2008).
As depicted, the results are similar for these investigations. All of
them emphasize the counter-rotating lateral vortices (blue areas) which
shed from the sides of the rear angled surface, which are revealed also by
studies of Ahmed et al. (1984) and Guilmineau (2008). As in the exper-
imental works, data has been extracted from these contours plots for a
direct comparison. Thus, vz/v∞ was evaluated for a line which goes Fig. 9. 3D maps of cD ðld ; ϕd Þ for plane (P), circular (C) and elliptical
approximately through the centres of the vortices, (z/l) ¼ 0.26, and these (E) diffusers.
results are shown in Fig. 7.
As revealed, the profiles of the velocity recorded in the vortex diffusers, ϕd ¼ 3 , 7 , to increase the accuracy of interpolation in this
structures are approximately the same, in a better correspondence with case. Lift (cL) and drag (cD) coefficients were computed as cL cLcL
the data provided by Strachan. There are differences in the proximity of
the symmetry plane of the body, and also concerning the minimum ve- L D
cL ¼ ρ 2 ; cD ¼ ρ 2 ; (4)
locity recorded, which show that the vortices recorded during experi- ∞ v∞ ∞ v∞
2
A 2
A
ments are more energetic. As previously mentioned, these can be
explained by the presence of the struts, which support the experimental where A ¼ 112032 mm2 is the reference area of the body.
models and influence the flow behind them. Based on the values of the aerodynamic coefficients, interpolation
Based on the data above, the computational model with respect to functions cL ðld ; ϕd Þ and cD ðld ; ϕd Þ were established, and 3D maps were
prediction of flow and aerodynamic loads proves to be accurate and all plotted, as depicted in Figs. 8 and 9. From the largest grid to the smallest
simulations were performed using a similar procedure. In this way, in- one, the 3D surfaces correspond to the cases of the plane diffuser (green
fluence of the diffusers parameters on drag and lift of the body was mesh), circular diffuser (red mesh), respectively to the elliptical diffuser
investigated for the following angles ϕd ¼ 2 , 4 , 6 , 8 over a range of (black mesh).
diffuser length ld¼(0.1, 0.2, 0.3, 0.4)l, taking into consideration that According to the ranges of the studied parameters, it was found that
diffuser height hdhdin road is limited due to the presence of rear bumper the lift coefficient is a third order polynomial function for all studied
(homologation requirements), or because it is the subject of constraints diffusers
concerning cargo space.
X
3
cL ðld Þ ¼ ai lid ; (5)
3. Results and discussions i¼0

In the first stage, the impact of plane diffuser was studied in order to where ai are the coefficients of the polynomial fitting function. In addi-
have a reference. Then, the effect of the curved underbody diffusers was tion, the lift coefficient varies linearly with the angle ϕd for the plane (P)
examined. Additional simulations have been carried out for elliptical and circular (C) diffusers, and it is a second order polynomial function for

5
A. Huminic, G. Huminic Journal of Wind Engineering & Industrial Aerodynamics 205 (2020) 104300

Table 3
Coefficients of cL ðld ; ϕd Þ interpolation function.
Plane diffuser (P) Circular diffuser (C) Elliptical diffuser (E)

a00 0.02427 0.03359 0.003527

a10 0.3261 0.4759 0.04249
a01 0.01557 0.01639 0.01379
a20 1.556 2.228 0.3725
a11 0.2891 0.3748 0.5787
a02 0 0 0.0006204
a30 1.841 2.614 0.6634
a21 0.331 0.4085 0.3965
a12 0 0 0.02193
RMSE 0.000897 0.001631 0.005228

Table 4
Coefficients of cD ðld ; ϕd Þ interpolation function.
Fig. 10. Variation of cD ðϕd Þ for ld ¼ 0:1 l
Plane diffuser (P) Circular diffuser (C) Elliptical diffuser (E)

b00 0.2925 0.3026 0.3242

b10 0.2143 0.3164 0.3547
b01 0.006253 0.01042 0.03614
b20 0.8416 1.184 0.856
b11 0.01437 0.002398 0.06699
b02 0.0004335 0.001034 0.006777
b30 1.036 1.409 0.666
b21 0.01268 0.03122 0.07373
b12 0.002238 0.003573 0.0007385
b03 1.122e-05 4.726e-05 0.00042
RMSE 0.0001453 0.0002547 0.0005832

the elliptical diffuser (E)

X
n
cL ðϕd Þ ¼ aj ϕjd ; (6)
j¼0

where nðPÞ ¼ nðCÞ ¼ 1, and nðEÞ ¼ 2.

Thus, cL ðld ; ϕd Þ is expressed as a third order linear interpolation
function

X
3
cL ðld ; ϕd Þ ¼ aij lid ϕjd ; ði þ j  3Þ; (7) Fig. 11. Variation of cD ðϕd Þ for ld ¼ 0:2 l
i;j¼0

where the values of the coefficients aij together with RMSE (Root Mean
Square Error) of the fitting functions are shown in Table 3.
As results reveal, there is a substantial growth of the downforce
generated by underbody diffusers, as concluded also in the review of
Ehirim et al. (2018a). In addition, the lift coefficients of body with curved
diffusers are significantly lower than that of body with plane diffusers.
Lift decreases continuously with both diffuser angle and length and
reaches the following minimum values for studied configurations: cLðPÞ ¼
 0:645, cLðCÞ ¼  0:814, and cLðEÞ ¼  0:941. The average relative
decreasing of lift for all configurations is ΔcLðCÞ ¼ 25 % for circular
diffusers, and ΔcLðEÞ ¼ 53 % for elliptical diffuser. References for the lift
variations are the values of the plane diffuser
cLðC; EÞ  cLðPÞ
ΔcLðC; EÞ ¼ 100 ½%: (8)
cLðPÞ

Concerning drag, the results show a more complex variation with the
parameters of diffusers. Thus, both cD ðld Þ and cD ðϕd Þ are third order
polynomial functions for all studied configurations

X
3
cD ðld Þ ¼ bi lid ; (9) Fig. 12. Variation of cD ðϕd Þ for ld ¼ 0:3 l
i¼0

X
3 and the interpolation function of cD ðld ; ϕd Þ is expressed similarly as in
cD ðϕd Þ ¼ bj ϕjd ; (10) Eq. (7)
j¼0

6
A. Huminic, G. Huminic Journal of Wind Engineering & Industrial Aerodynamics 205 (2020) 104300

Fig. 15. 2D map of. cD min

The values of the coefficients bij are shown in Table 4.

As Fig. 9 reveals, the curved diffusers could generates smaller values
of drag than the plane diffuser, similarly as for the lift force, as depicted
in Figs. 10–13, which show the variation of cD ðϕd Þ for studied diffuser
lengths. This behaviour occurs for small diffuser angles, from ϕd ¼ 2:6
Fig. 13. Variation of cD ðϕd Þ for ld ¼ 0:4 l to ϕd ¼ 4 , depending on the diffuser length. The upper limit of the
diffuser angle (ϕd ¼ 4 ) corresponds to the smallest diffuser length
(ld =l ¼ 0:1), as shown in Fig. 10, and if the diffuser length increase to ld =
l ¼ 0:3, then the upper limit of ϕd decreases to the value ϕd ¼ 2:6, as
shown in Fig. 12. The maximum variation of drag is ΔcLðEÞ ¼ 0:0053
(1.93%) for elliptical diffuser of ϕd ¼ 2 and ld =l ¼ 0:3. If ld =l ¼ 0:4,
Fig. 13, all studied bodies have similar values of drag when ϕd ¼ 2 , and
no significant decrease of drag for curved diffuser is revealed.
These figures present also the line for drag coefficient of the body
without diffuser, cDb ¼ 0:2747. Therefore, it is easy to apprehend the
benefit induced by underbody diffusers on aerodynamics of a car body.
Concerning the pattern of drag, it decreases in the first stage and
reaches a minimum value. This takes place faster for the elliptical setup.
Then, the drag starts to rise, significantly swift for the curved diffuser. A
similar trend of the drag variation was also reported by Bonnavion and
Cadot (2019) for a curved underbody diffuser of an Ahmed type bluff
body. For the elliptical setup, it even exceed the drag of the body without
diffuser if ϕd > 4 and ld > 0:1 l, and the gain on downforce generated
becomes less relevant.
The increase of drag is due to enlargement of the counter-rotating
vortices induced on the sides of the diffuser, which have a negative
impact on flow under the body, as revealed in previous study, Huminic
et al. (2012). This aspect is emphasised clearly in Fig. 14, which show the
contours in the down-stream, x ¼ 55 mm, of the dimensionless velocities
vz =v∞ and vx =v∞ for all the diffusers of ϕd ¼ 6 and ld ¼ 0:2 l. As shown,
these vortices increase in intensity for the curved diffusers. For elliptical
diffuser they are significantly larger, and in this configuration, it is
clearly shown that these counter-rotating vortices are turned towards the
symmetry plane of body. Thus, they also contribute to the increase of the
down force due to the pumping effect, since they draw more flow through
the underbody flow path, and a greater underbody flow rate leads to a
lower underbody pressure and larger downforce, as stated by Cooper
et al. (1998).
on yOz plane, x ¼ 55 mm
As previously mentioned, there are some particular configurations
when all diffusers have similar values of drag. For the shortest diffuser,
Fig. 14. Dimensionless velocity contours vz =v∞ , left and vx = v∞ , right, for ϕd ¼ this situation occurs when ϕd ¼ 4 . If ld increases, there is a continuous
6 and ld ¼ 0:2 l.
reduction of the diffuser angle corresponding to same drag coefficient,
until ϕd ¼ 2 .
X
3
Hence, the curved sections work better from aerodynamic point of
cD ðld ; ϕd Þ ¼ bij lid ϕjd ; ði þ j  3Þ: (11)
i;j¼0 view for small angles and short length of diffusers. For larger lengths and

7
A. Huminic, G. Huminic Journal of Wind Engineering & Industrial Aerodynamics 205 (2020) 104300

Table 5 an inflexion point. The pressure coefficient reaches the minimum value
Values of cD min ðld ; ϕd Þ more quickly for circular diffuser due to its faster increase of the curva-
cD min ld =l ½   ϕd ½  ture dz=dx.
The second acceleration of the air flow through the curved diffusers is
Plane diffuser 0.237 0.4 6
Circular diffuser 0.247 0.4 4 due to the natural tendency of fluids to adhere to surfaces of bodies on
Elliptical diffuser 0.251 0.3 2 which they flow. Thus, following the curvature of diffusers, the air ve-
locity increases mainly due to occurrence and development of the vertical
component vz , which is negligible in the flat section of the underbody.
This aspect is emphasised in Figs. 17 and 18, which show the profiles of
dimensionless velocity components vz =v∞ and vx =v∞ in five sections
along the diffusers, from inlet (x=l ¼ 0:8) to exit (x=l ¼ 1).
For small values of curvature, the vertical component growths as long
as the curvature increases. Because the elliptical diffuser has a significant
change of the curvature only at its end, it has a larger growth of vz . In
addition, as depicted in detail of Fig. 16, there is a third sudden accel-
eration of the flow around the edge at the underbody end for plane and
circular diffusers, which is emphasised also by vz =v∞ variation for x=l ¼
1:0. No acceleration of the air happens at the end of the elliptical diffuser
because its surface is tangent to the vertical end surface of the body.
Concerning horizontal component, the velocity profiles vx =v∞ are
similar at the inlet of diffuser for all configurations. After a slight growth
of vx for curved underbodies, it follows a continuous diminution, which
happens earlier for circular layout. For plane diffuser, vx only decreases
along the body. At the end of the body, vx =v∞ tends again to a similar
Fig. 16. Variation of cp in the symmetry plane for ϕd ¼ 6 and ld ¼ 0:2 l profile for all diffuser. Thus, due to the velocity growth, the depression
increases under body, which also leads to a larger downforce generated
small angles, the shape of diffuser has a smaller influence. by the curved diffusers.
The combinations of parameters for which each diffuser has a mini-
mum drag are depicted in Fig. 15, which represents the projection on 4. Conclusions
horizontal plane (ld ; ϕd ) of cD maps in Fig. 9. As shown, a small strip area
corresponding to cD min of the circular diffusers intercalates between the In this paper, a bluff body fitted with circular and the elliptical un-
sections of elliptical and plane diffusers. The values of minimum drag for derbody diffusers is studied, the angle and the length of diffuser being the
all studied cases are in Table 5. parameters which are systematically varied within ranges relevant for a
Pressure and velocity variations in the symmetry plane were also hatchback passenger car. The results are compared with those obtained
studied to gain insight of phenomena that lead to improvement of for the plane diffuser, and they reveal a significant reduction of the lift
aerodynamics for the bodies fitted with circular and elliptical diffusers. coefficients of body with curved diffusers due to the pair of counter-
As shown in Fig. 16, for ϕd ¼ 6 ld ¼ 0:2 l, there are two stages of rotating vortices induced on the sides of the diffusers, which contribute
pressure recovery beneath the body for curved diffusers, as for the plane to the increase of the pumping effect, and consequently to a larger
diffuser. Since there is no edge to generate a sudden growth of the air downforce, as stated by Cooper et al. (1998). The negative effect of these
velocity, as for plane diffusers, pressure variation takes place continu- vortices is the grow of drag for diffusers with large lengths and angles.
ously, and the previously mentioned cusp point of cp variation turns into The reduction of lift is also emphasised by the pressure coefficient

Fig. 17. Profiles of the dimensionless velocity vz =v∞ along the diffusers in the symmetry plane.

8
A. Huminic, G. Huminic
Fig. 18. Profiles of the dimensionless velocity vx =v∞ along the diffusers in the symmetry plane.
variation, which takes place continuously, with no cusp point as for plane
diffusers. The average reduction of lift for all configurations studied is Δ
cLðCÞ ¼ 25 % for circular diffusers and ΔcL(E) ¼ 53% for elliptical
diffuser.
Based on the values of aerodynamic coefficients, polynomial inter-
polation functions c(ld, ϕd) are established for both lift and drag. 3D
maps show the lift decreases continuously with both diffuser angle and
length for studied parameters. Concerning drag, the results show a more
complex variation with the parameters of diffusers. Thus, cD decreases in
the first stage and reaches a minimum value, and then it starts to rise,
significantly swift for the curved diffuser. For the elliptical setup, cD even
exceed the drag of the body without diffuser if ϕd>4 and ld > 0:1 l, and
the gain on downforce generated becomes less relevant. The curved
diffusers generate smaller values of drag than the plane diffuser for small
and moderate angles and short lengths of diffusers.
Hence, it is concluded that short diffusers with curved section works
better from aerodynamic point of view. In terms of vehicles design, short
diffusers have also the advantage of small interferences with other un-
derbody components of cars, as rear wheelhouses and rear silencer. In
addition. they offer a significant larger cargo space.
Credit author statement
The authors contributed equally to the work.
Declaration of competing interest
The authors declare that they have no known competing financial
interests or personal relationships that could have appeared to influence
the work reported in this paper.
References
Ahmed, S., Ramm, G., Faltin, G., 1984. Some Salient Features of the Time-Averaged
Ground Vehicle Wake. SAE Technical Paper 840300.
Bonnavion, G., Cadot, O., 2019. Boat-tail effects on the global wake dynamics of a flat-
backed body with rectangular section. J. Fluid Struct. 89, 61–71.
Cheng, S.Y., Tsubokura, M., Nakashima, T., Nouzawa, T., Okada, Y., 2011. A numerical
analysis of transient flow past road vehicles subjected to pitching oscillation. J. Wind
Eng. Ind. Aerod. 99, 511–522.
Cooper, K., Bertenyi, T., Dutil, G., Syms, J., 1998. The aerodynamic performance of
automotive underbody diffusers, developments in vehicle aerodynamics SAE SP-
1318, pp. 5–36.
Journal of Wind Engineering & Industrial Aerodynamics 205 (2020) 104300
Ehirim, O.H., Knowles, K., Saddington, A., 2018a. A review of ground-effect diffuser
aerodynamics. J. Fluid Eng. 141, 020801, 19 pages.
Ehirim, O.H., Knowles, K., Saddington, A.J., Finnis, M.V., 2018b. Aerodynamics of a
convex bump on a ground-effect diffuser. J. Fluid Eng. 140, 091102, 11 pages.
Fu, C., Uddin, M., Robinson, C.A., 2018. Turbulence modeling effects on the CFD
predictions of flow over a NASCAR Gen 6 racecar. J. Wind Eng. Ind. Aerod. 176,
98–111.
Grandemange, M., Cadot, O., Courbois, A., Herbert, V., Ricot, D., Ruiz, T., Vigneron, R.,
2015. A study of wake effects on the drag of Ahmed’ s squareback model at the
industrial scale. J. Wind Eng. Ind. Aerod. 145, 282–291.
Graysmith, J.L., Baxendale, A.J., Howell, J.P., Haynes, T., 1994. Comparisons between
CFD and Experimental Results for the Ahmed Reference Model. RAeS conference on
vehicle aerodynamics, Loughborough, pp. 30.1–30.11.
Guilmineau, E., 2008. Computational study of flow around a simplified car body. J. Wind
Eng. Ind. Aerod. 96, 1207–1217.
Hetherington, B., Sims-Williams, D., 2006. Support strut interference effects on passenger
and racing car wind tunnel models. SAE Technical Paper No. 2006-01-0565.
Hucho, W.-H., Sovran, G., 1993. Aerodynamics of road vehicles. Annu. Rev. Fluid Mech.
25, 485–537.
Huminic, A., Huminic, G., Soica, A., 2012. Study of aerodynamics for a simplified car
model with the underbody shaped as a Venturi nozzle. Int. J. Vehicle Des. 58, 15–32.
Huminic, A., Huminic, G., Carauleanu, M., Ciolofan, C., 2016. Comparative study on the
performances of aerodynamic devices used in decreasing of the automobiles lift force.
In: CONAT 2016 International Congress of Automotive and Transport Engineering,
pp. 48–54. https://doi.org/10.1007/978-3-319-45447-4_5.
Huminic, A., Huminic, G., 2017. Aerodynamic study of a generic car model with wheels
and underbody diffuser. Int. J. Automot. Technol. 18, 397–404.
Kekus, P., Angland, D., 2018. Automatic wind tunnel-based optimisation of an automotive
underbody diffuser. AIAA Aerospace Sciences Meeting. https://doi.org/10.2514/
6.2018-0045.
Lienhart, H., Soots, C., Becker, S., 2002. Flow and turbulence structures in the wake of a
simplified car model (Ahmed Model). In: Proceedings DGLR Fach. Symp. der AG
STAB Stuttgart University.
Mair, W.A., 1969. Reduction of base drag by boat-tailed afterbodies in low-speed flow.
Aeronaut. Q. 20 (4), 307–320.
Menter, F.R., 1994. Two-equation eddy-viscosity turbulence models for engineering
applications. AIAA J. 32, 1598–1605.
Mohottia, D., Wijesooriya, K., Dias-da-Costa, D., 2019. Comparison of Reynolds
Averaging Navier-Stokes (RANS) turbulent models in predicting wind pressure on tall
buildings. J. Build. Eng. 21, 1–17.
Rao, A., Minelli, G., Basara, B., Krajnovic, S., 2018. On the two flow states in the wake of a
hatchback Ahmed body. J. Wind Aerod. Ind. Aerod. 173, 262–278.
Rossitto, G., Sicot, C., Ferrand, V., Boree, J., Harambat, F., 2017. Aerodynamic
performances of rounded fastback vehicle. Proc. Inst. Mech. Eng. D: J. Automob. Eng.
231, 1211–1221.
Strachan, R., Knowles, K., Lawson, N., 2007. The vortex structure behind an Ahmed
reference model in the presence of a moving ground plane. J. Exp. Fluids 42,
659–669.
Sumantran, V., Sovran, G., 1996. Vehicle aerodynamics, PT-49. SAE International.
Urquhart, M., Sebben, S., Sterken, L., 2018. Numerical analysis of a vehicle wake with
tapered rear extensions under yaw conditions. J. Wind Eng. Ind. Aerod. 179,
308–318.