Experimental and Numerical Aerodynamic Analysis of A Passenger Car
Experimental and Numerical Aerodynamic Analysis of A Passenger Car
Experimental and Numerical Aerodynamic Analysis of A Passenger Car
Aerodynamic Analysis of a
Armagan Altinisik1
TOFASFIAT (Turkish Automotive
Passenger Car: Influence
Factory Incorporated Company),
Yeni Yalova Yolu Cad. No: 574,
of the Blockage Ratio on
Bursa 16369, Turkey
e-mail: [email protected] Drag Coefficient
Emre Kutukceken Experimental and numerical investigations were performed to determine the pressure dis-
ROKETSAN, tributions and the drag forces on a passenger car model. Experiments were carried out
Kemalpasa Mahallesi Sehit Yuzbas with 1/5th scale model FIAT Linea for 20% and 1% blockage ratios in the Uludag Uni-
Adem Kutlu Sokak No: 21 Elmadag, versity Wind Tunnel (UURT) and in the Ankara Wind Tunnel (ART), respectively. Com-
Ankara 06780, Turkey putational fluid dynamics (CFD) analysis for 1/5th scale model with 0%, 5%, and 20%
blockage ratios was performed to validate various blockage correction methods supple-
Habib Umur mentary to the experimental results. Three-dimensional, incompressible, and steady gov-
Mechanical Engineering Department, erning equations were solved by STAR-CCM code with realizable ke two-layer
Faculty of Engineering and Architecture, turbulence model. The calculated drag coefficients were in good agreement with the ex-
Uludag University, perimental results within 6%. Pressure coefficients on the model surfaces have shown
Gorukle Campus, similar trends in the experimental and numerical studies. Some of the existing blockage
Bursa 16059, Turkey correction methods were successfully compared in this study and predicted drag coeffi-
cients were within 6 5%. The authors propose the continuity and the Sykes blockage cor-
rection methods for passenger car models because they are very simple and practical and
they can be used economically for engineering applications. [DOI: 10.1115/1.4030183]
1 Introduction model, and DrivAer models have been widely used in experimen-
tal and numerical analysis due to the restricted access and short
The automotive industry changed their focus from maximizing
life span of real car models [710]. However, recently, aerody-
the power of cars to maximizing fuel efficiency as a result of high
namic studies on production cars can also be found in various en-
oil prices. Thus, the aerodynamic resistance of automobiles is
gineering journals and papers.
considered to be a crucial part of automotive design. The aerody-
Although wind tunnel experiments are still the most reliable
namic drag can be as high as 65% of the total drag for most pas-
and irreplaceable method for vehicle aerodynamics research [11],
senger vehicles at typical speed limit of 100 km/h [1]. Hence,
CFD is also widely accepted as a flexible and cost-efficient
reducing drag contributes significantly to the fuel economy of a
supplement to the traditional wind tunnel tests. Currently, a well-
car.
controlled CFD analysis is accurate to within 68% for the pre-
The aerodynamic drag coefficient is defined as follows:
dicted vehicle aerodynamic drag coefficient for cars [12]. It is also
2 known that the difference in drag coefficient values tested in
Cd Fd =0:5qU1 Av (1)
different wind tunnels for the same car could be up to 5% [13].
Most experimental aerodynamic research is undertaken in wind
In this equation, q represents the air density, U1 is the freestream tunnels under controlled test conditions with a very low level of
velocity, Av is the model frontal area, and Fd is the total drag force freestream turbulence. This provides the best environment for the
acting on the car projected on the longitudinal direction. acquisition of repeatable data but such ideal airflows are rarely
Pressure coefficient of the flow on vehicle surfaces is defined as found on the road. Watkins et al. [14] performed measurements
follows: demonstrating that atmospheric turbulence intensity lies typically
2
Cp Px P1 =0:5qU1 (2) in a range between 2% and 10%. Saunders and Mansour [15]
demonstrated that turbulence intensities may be as high as 20%
where P1 is the far field static pressure, and Px is the local surface when testing behind another vehicle with a separation of three car
pressure. lengths. Recently, turbulence generating systems have been used
There are various experimental and numerical studies con- to obtain real road conditions at wind tunnels. Cogotti [16] studied
ducted on drag optimizations in past literatures such as shape, the turbulence intensity effect on the aerodynamic coefficients for
style optimizations, and wake controls [26]. The generic car different passenger car models in Pininfarina wind tunnel and
models such as Ahmed model, SAE standard model, MIRA obtained small increase in drag coefficient with higher turbulence
intensities (DCd < 3% for I 68% at yaw angle 0).
Aerodynamic coefficients of Cp and Cd are changing noticeably
1
Corresponding author. by increasing blockage ratio. Blockage ratio, as defined in Eq. (3),
Contributed by the Fluids Engineering Division of ASME for publication in the
JOURNAL OF FLUIDS ENGINEERING. Manuscript received June 5, 2014; final manuscript
is one of the most important factors that determine the accuracy of
received March 22, 2015; published online April 29, 2015. Assoc. Editor: Francine aerodynamic measurements and suggested <7.5% at wind tunnel
Battaglia. tests [17].
Table 1 Reynolds dependence study performed by CFD Table 3 Test velocities at different test conditions
analysis
Velocity Static Drag
U1 (m/s) Scale Rex Cd profiles pressures forces
10 1/5th 6.26 105 0.265 Air velocity (m/s) Re1/5 scale UURT ART UURT ART UURT ART
20 1/5th 1.25 106 0.252
30 1/5th 1.88 106 0.248 3 l.88 105
40 1/5th 2.50 106 0.246 10 6.26 105
60 1/5th 3.76 106 0.245 20 l.25 106
40 1/1 1.25 107 0.245 30 l.88 106
calculations. The energy equation was excluded as there was no In the present study, the realizable two-layer ke model was
heat transfer or temperature change. Water-tight model was used for the turbulence model. Realizable ke model, developed
obtained by ANSA preprocessor. by Shih et al. [44], satisfies certain mathematical constraints on
The governing equations for computational fluid dynamics the Reynolds stresses that are consistent with the physics of turbu-
based on conservation of mass and momentum are given in Eqs. lent flows. Eddy viscosity model with Boussinesq approximation
(5) and (7)(9). was used in calculations. The realizable ke model is noticeably
more accurate than the other ke models in simulating the mean
@u @v @w velocity components. Initial studies have shown that the realizable
0 (5)
@x @y @z model provides the best performance of all the ke model versions
for several validations of separated flows and flows with complex
Velocity components are decomposed into the mean and fluctuat- secondary flow features [45]. The realizable ke turbulence model
ing components as is commonly used in external vehicle aerodynamics studies
[12,33,46,47]. Kang et al. [48] have chosen the realizable ke
u u u0 ; v v v0 ; w0
ww (6) model in their study in the vicinity of walls because it provides
superior performance for flows involving rotation, boundary
Introducing this fact to the NavierStokes equations leads to layers under adverse pressure gradients, separation, and recircula-
RANS with the additional Reynolds stress terms such as tion. Singh et al. [49] performed a validation of four turbulence
@=@x qu0 v0 models (SpalartAllmaras, standard ke, RNG ke, and realizable
ke) integrated in FLUENT and proved that a realizable ke model
gives the best match with the experimental results.
@ u @ u @ u
q u v w In this model, Cl is not a constant any more, but rather it is
@x @y @z computed according to
@ p @ @ @
lr2 u qu02 qu0 v0 qu0 w0 (7) 1
@x @x @y @z Cl (12)
U k
@ v @ v @ v A0 As
q u # w e
@x @y @z
@p @ 0 0 @ 02 @ 0 0 where A0 , As , and U* are the functions of velocity gradients.
lr2 v qu v qv qv w
@y @x @y @z The transport equations for both the turbulence kinetic energy
(8) k and the dissipation e are calculated as follows:
@w @w @w @kuj @ l @k
q u # w q l t Gk Gb qe YM Sk (13)
@x @y @z @xi @xi rk @xj
@ p @ 0 0 @ 0 0 @ 02
lr2 w qu w qv w qw @euj @ l @e e2
@z @x @y @z q l t qCe1 Se qCe2 p
(9) @xj @xj re @xj k #e
e
Ce1 Ce3 Gb (14)
In this approach to turbulence modeling, one puts Reynolds stress k
terms into the form of a viscous stress term with an unknown
coefficient, the turbulent viscosity in simple form where Ce1 ; Ce2 ; Ce3 ; rk ; and re are the constants and defined as in
Table 4. Sk and Se are the user defined source terms and S is the
@u modulus of the mean strain rate tensor. Gk and Gb are the turbu-
qu0 v0 lt (10) lence production terms and defined as in ke turbulence model.
@y
The turbulence viscosity lt is computed using turbulent kinetic Table 4 Realizable ke constants
energy (k) and its dissipation rate (e) as
Ce1 Ce2 Ce3 rk re
k2
lt Cl q (11) g S
k 1.9 1, Gb > 0 1 1.2
e Max 0:43; ;g
sg e 0, Gb < 0
where Cl is the turbulence viscosity constant for ke model.
Table 5 Various scale models and #meshes used in past CFD provides better resolutions in the flow wake region, however,
analysis more computational time and memory are needed [33]. The num-
ber of meshes in the solution domain was kept as much as possible
Scale Authors # Meshes Grid independency considering the hardware capacity under the light of past studies
(see Table 5). For the near-wall treatment, 16 prism layers were
1/3 Marklund and Lofdahl [53]
used starting from 0.3 mm to 30 mm with the stretch factor of 1.2.
1/1 Bagal and Mulemane [54] 14.3 106
1/5 Friedl and Watts [55] Eight volume sources were applied for the mesh generation. Two
1/20 Tsubokura et al. [56] 20 106 of them were for side mirrors, two of them for the regions close to
1/5 Akasaka and Ono [57] 3 106 the ground, and four of them for the wake and backside of the
Singh et al. [49] 0.42 106 Performed model. Figure 4 shows the sample midsection view of the domains
1/1 Kang et al. [48] 0.82 106 Performed and the dimensions of the solution domains for various blockage
Chu et al. [58] 0.55 106 Performed conditions.
Guilmineau et al. [59] 23 106 The numerical method was validated by the comparison of the
Guilmineau et al. [60] 10 106 predicted drag coefficient of 1/5th scale model blockage-free
1/12 Jindal et al. [61] 1.74 106
simulation and blockage-free experimental ART test. Obtained
1/5 Gohlke et al. [62]
Cheng et al. [63] 16 106 results were in very good agreement within 6%. Obtained CFD
1/1 Cheng et al. [64] drag coefficients for various blockage conditions were used in this
1/1 Song et al. [65] 4.4 106 Performed study to validate several blockage correction methods supplemen-
1/2.5 Jakirlic et al. [40] tary to the experimental results.
1/1 Basara et al. [66] 0.5 106 In our computational study, as boundary conditions, velocity
1/5 Present study 1316 106 inlet was used as uniform and outlet was considered as atmos-
pheric conditions by using pressure outlet. Reference density was
1.225 kg/m3, turbulence intensity was selected as 0.05 based on
UURT tunnel turbulence intensity level.
The two-layer approach was suggested first by Rodi [50]. In
this approach, the computation is divided into two layers. In the
layer adjacent to the wall, the turbulent dissipation rate e and the 4 Results and Discussion
turbulent viscosity lt are specified as functions of wall distance.
The values of e specified in the near-wall layer are blended 4.1 Flow Patterns and Velocity Profiles. Figure 5 shows the
smoothly with the values computed from solving the transport flow patterns obtained on the model surfaces with tuft test and
equation far from the wall. The coefficients in the models are CFD skin friction distribution. In the skin friction distribution,
identical, but model gains the added flexibility of an all y wall dark color () represents attached flow and () represents
treatment with the two-layer approach [51]. The two-layer formu- detached flow from the model surface. Flow separations (as indi-
lations make no assumption about how well the viscous sublayer cated - in the figure) have been observed on the model surfaces
is resolved. By using a blended wall law to estimate shear stress, mainly as: engine hoodwindscreen junction, front and rear tire
the result will be similar to the low y wall treatment if the mesh wake areas, upper part of the rear windscreen, and rear part of the
is fine enough. If the mesh is coarse enough (y > 30), the wall deck lid due to almost zero shear stress caused by adverse pres-
law is equivalent to a logarithmic profile [52]. Figure 3 shows sure gradient. The same flow structures were obtained in tuft test
that the y distributions on the model surfaces in this study are for 1/5th scale model. The tuft test demonstrates that the separa-
< 5 in the separation areas and 10 to 15 on the rest of the model tion on the slant surface starts upper part of the rear windscreen as
surfaces, demonstrating good near-wall resolution especially in in CFD skin friction distribution. There are two vortex lines
separation zones. The Wolfstein shear-driven formulation was observed on the side doors toward the roof, one from the external
used as default in two-layer calculations. mirror and the other one from the A-pillar zone. Fluid from the
CFD analysis was performed for 1/5th scale model supplemen- side doors mixes with the fluid from the upper roof causing small
tary to experimental results. Table 5 indicates that 1/5th scale right and left lateral swirls on the rear windscreen. The left swirl
model usage was preferred by some authors in CFD analysis. The is turning to the left while the right swirl is turning to the right.
fine mesh contains between 13 and 16 106 polyhedral cells for The unstable flow structure is observed at the inferior part of side
various blockage modelings used in the model. Polyhedral mesh doors (near to the ground) and tire wake areas. Flow is stable and
Fig. 5 Flow patterns: (a) 1/5th model tuft test at UURT (30 m/s) and (b) CFD skin friction
distribution
attached to the surfaces on the engine hood, front windscreen, and surface. The flow features at the rear end of the model are very
on the roof. complex and fully detached from the model surface. Boundary
Figure 6(a) presents selected normalized velocity profiles along layer thicknesses are increasing due to increasing pressure gra-
the model centerline (Y0). On the engine hood (point 1, dients through the points of 6 (x/L 0.810), 7 (x/L 0.841), and
x=L 0.118), front windscreen (point 3, x=L 0.324), and on the 8 (x/L 0.872). Figure 6(b) compares the normalized velocity
roof (point 4, x=L 0.561, point 5, x=L 0.748), the flow is stable profiles of U 3 m/s and 30 m/s at selected points (*). The nor-
and attached to the model surface. At the front wiper area (point malized velocity profiles at U 10 m/s and 20 m/s were similar to
2, x=L 0.217), flow is not stable anymore and detached from the U 30 m/s and not presented in this study. Velocity profiles at the
speed of 3 m/s are inaccurate in points of 2, 5, and 8 where lami- pressure gradient as in 30 m/s. It can be said that the flow is lami-
nar separation occurs and have significantly bigger boundary layer nar at 3 m/s while it is completely turbulent at 30 m/s. The CFD
thicknesses than at the speed 30 m/s. The comparison of the CFD and the experimental velocity profiles were matching very good in
and the experimental velocity profiles at points 6 (x/L 0.810) the rear wake as seen in Fig. 8(b).
and 10 (x/L 0.916) showed good agreement in Fig. 6(c) demon-
strating that STAR-CCM code with selected turbulence model
was successful in modeling the separation area. The CFD velocity 4.2 Pressure Distributions. Figure 9 shows 20% blockage
profiles at points of 6, 7, and 10 were also compared in Fig. 6(c). effect on static pressure distribution at model centerline and verti-
The separation on the slant surface starts at around point 7 in the cal sections. The highest positive pressure along the model
CFD analysis as in the experimental analysis. centerline was measured at the center of the front bumper (a).
Figure 7(a) compares the selected velocity profile with the This point is called a stagnation point. A local suction peak was
study of Koike et al. [4] on the roof before the separation point. obtained at the beginning of engine hood (b). After that point,
The z axis was normalized with the boundary layer thickness as static pressures increased progressively and reached a local peak
z/d0:995 . Koike et al. [4] obtained the boundary layer thickness as level at the bonnetwindscreen junction (c). A rapid reduction of
d 30 mm on 1/1 scale Mitsubishi Lancer Evolution VIII at the pressure followed, and a maximum suction peak occurred at
separation point, 100 mm upstream from the rear end of the roofwindscreen junction (d). Static pressure has slightly
roof. Gopal and Senthilkumar [67] measured the boundary layer increased to the point of (e) and remained almost same toward the
thickness as d 2 mm for 1/15th scale passenger car model before end of the roof. Static pressure again reached local negative peak
the separation point. In the present study, the boundary layer at roof trailing edge (f). Starting from the end of roof, sudden
thickness was found as d 6 mm for 1/5th scale model. pressure increase was observed until the initial part of the rear
[d0:995 =scale] were the same for all three studies. The velocity pro- windscreen (g). Pressure remained almost unchanged between the
files at the same point for different scale models were in good points of (g and h) due to the flow separation. Negative pressure
agreement. Figure 7(b) shows the comparison of the boundary increase was observed again at point (i), rear part of the deck lid.
layer thickness on the model roof from point 1 to point 2 with the On vertical section Y1, pressures were almost same from the
study of Jenkins [68] for 0.16 scale Ford C1 configuration. The centerline toward upper edge of the front door. But a strong suc-
external geometry of the models was different but the develop- tion peak on the upper edge of the door was observed at 20%
ment of the velocity profiles was almost same toward the rear end blockage effect. Sudden pressure increase was observed from the
of the model roof with increasing boundary layer thickness due to upper edge toward the bottom of the front door. The same suction
increasing pressure gradient. peak on the rear door upper edge was observed in section Y2.
Figure 8(a) shows the comparison of the normalized down- However on section Y3, the flow structure is very complex, from
stream velocity profiles in the near wake at 3 m/s and 30 m/s free- point (a) to (b) there is a sharp decrease in pressure and from (b)
stream velocities. Wake height is bigger at 3 m/s freestream to (c) there is again increase in pressure coefficient, this was due
velocity because fluid particles cannot resist the increasing to re-attachment of the detached fluid particles on the deck lid at
point (c). After point (c) toward (d), pressures gradually decreased represents negative pressure coefficients in the surface con-
and strong pressure fluctuations were observed at the bottom of tours. As seen from the contours, small pressure changes on the
the fender. model roof were observed at 5% blockage ratio compared to the
Significant static pressure variations were observed with 20% blockage-free analysis. But significant pressure changes were
blockage effect. Almost 40% suction pressure increase was obtained with 20% blockage ratio. The centerline and vertical sec-
obtained on the model roof due to 20% blockage effect. Variations tion CFD pressure distributions with increasing blockage effect
started from the middle of the front windscreen and became more demonstrated the same conclusion.
evident on the roof and vertical sections. At section Y1, suction Comparison of the centerline Cp distributions of various mod-
peak on front door upper edge is very evident with 20% blockage els in the literature are given in Fig. 13. Knowing that the
effect while no suction peak was measured with 1% blockage external shape of the model is very important for the pressure
ratio. distributions, the trends of the pressure distributions are similar
Figure 10 shows the comparison of the CFD Cp distributions in the studies of Islam et al. [69], Heft et al. [9], and Akasaka
with experimental UURT (20% blockage) and ART (blockage- and Ono [57]. The highest static pressure on the deck lid was
free) distributions at freestream velocity of 30 m/s. Pressure distri- found at x=L 0.936 in our study, while it was 0.937 in the
butions were in good agreement except for the separation area study of Jenkins [68].
that small deviations were observed. Akasaka and Ono [57] have
observed the similar deviations at the separation area on 1/5th
scale model passenger car (Fig. 11). 4.3 Drag Coefficients (Cd and Blockage Corrections. The
CFD surface pressure contours and distributions with increasing measured experimental and numerical drag coefficients of 1/5th
blockage effect are given in Fig. 12. represents positive and scale model at 30 m/s freestream velocity are summarized in
Table 6 for different blockage ratios. The experimental drag coef- Aerodynamic resistance forces and coefficients obtained
ficient for 20% blockage effect was measured as 0.431 in the directly from wind-tunnel tests (or from their numerical simula-
UURT tunnel with 63.2% deviation compared to blockage-free tions) require a correction for canceling the blockage effect. The
result of 0.264 in ART. The numerical drag coefficients for 20% vehicle model blocking the flow in wind-tunnel causes that the
and 5% blockage ratios were calculated as 0.409 and 0.2709 average air speed in the immediate vicinity of the model is
with 64.9% and 9.2% deviations, respectively, compared to the increased. The blockage can be additionally enlarged due to a
blockage-free result of 0.248. The numerical and the experimental wake behind the vehicle. Thus, the directly measured or computed
drag coefficients for different blockages ratios were in good agree- drag force and coefficient are overestimated and should be cor-
ment within 6%. rected by blockage correction methods.
Fig. 11 Comparison of centerline pressure distribution at rear part of the models with the
study of (b) Akasaka and Ono [57] and (a) present study
4.3.1 Various Blockage Correction Methods. The continuity Cowdrey [22] has rederived the Maskell [23] blockage correc-
blockage correction formula proposes the correction coefficient in tion and produced the correction that did not depend on the meas-
the below equation [18,19]: ured drag, although it still required an empirical constant that was
a function of body geometry. m0 is the semi-empirical constant
w 1 B2 (15) that must be determined by experiments.
Fig. 13 Comparison of 1/5th scale FIAT Linea centerline Cp distributions with various models
in the literature
Experimental Cd CFD Cd
Cd Uncorrected" Blockages % 20 5 20
Cd 0.431 0.271 0.409
DCd Cd Cd (0 blockage) 0.167 0.023 0.161
Model scale 1/5 1/5 1/5
Blockage corrections of Continuity method (Carr [18] and Gleason [19]) 0.276 0.245 0.261
drag coefficients Error % 4.5 1.6 5.2
Sykes [20] m 1.9 0.267 0.245 0.254
Error % 1.2 1.1 2.2
Stafford [21] m 1.22 0.326 0.254 0.309
Error % 23.4 2.6 24.7
Cowdrey [22] 0.264 0.234 0.250
Error % 0.2 5.7 0.8
Maskell III (Hackett and Cooper [24]) 0.250 0.232 0.236
Error % 5.5 6.3 4.7
Maskell III plus Thom (Cooper et al. [25]) 0.285 0.254 0.271
Error % 7.9 2.2 9.3
Area ratio method (Barlow et al. [26]) 0.391 0.264 0.371
Error % 48.1 6.6 49.6
Two variable method (Hackett and Cooper [24] 0.260 0.249 0.247
and Cooper et al. [25])
Error % 1.6 0.4 0.6
CFD correction method (Yang and Schenkel [12]) 0.270
Error % 3.1
s
Cdc Cdm =1 eT 2 (18) hCdm S=C
ew 1 hDCDM S=C 1 (24)
eT eS eW (19) 1 hCdm S=C
B H V
eS 0:36 (20)
H B C1:5 Cdm 1
s DCDM
1 hCdm S=C 2hS=C
1 DCDwi p
eW 1 h 1 (21) 1 1 4hCdm S=C (25)
1h Cdm
where h 0:96 1:94e0:06h=w )Cdm S=C, S and V are two times where h is the Maskell wake blockage constant and defined as in
the model frontal area and volume, and h/w is the model aspect ra- Eq. (21). S and C are the duplex model section and duplex test
tio. H is the width of duplex tunnel test section. B and C are twice section areas, respectively.
the actual test section height and area, respectively. Cdm is the Barlow et al. [26] have suggested the area ratio blockage cor-
uncorrected drag coefficient and DCDWi is wake drag increment rection method as given in the below equation:
and defined as 2
1
Cdc Cd 1 B (26)
1 1 p 4
DCDWi Cdm 1 1 4h (22)
1h 2h
The two variable blockage correction method obtained from the
Cooper et al. [25] have used Maskell III plus Thom blockage cor- force model developed by Hackett and Cooper [24] can be simpli-
rection method that is the modified version of Maskells analysis fied with the below formula for an automobile positioned midway
in their studies. This correction comprises four components as: a between the tunnel sidewalls, where the lateral interference veloc-
velocity increment due to wake blockage, a drag increment due to ity increment (d 0) and induced drag component (Cdi 0) [25].
wake blockage, a velocity increment due to solid blockage, and a
yaw angle change due to side force. Cdc Cd =1 e3 (27)
Cdm
Cdc DCDM (23) where e Dumx =U1 is longitudinal velocity interference incre-
1 es ew 2 ment and calculated as e5% 0.0286 and e20% 0.1838 in our
study from the CFD analysis selecting the maximum velocity
where es is defined as Eq. (20). ew and Maskell III drag coefficient increments of 0%, 5%, and 20% blockages. The cubic term in the
additive DCDM are defined in the below equations: denominator is not a square as is commonly expected. This results