You are on page 1of 22

ARTICLE IN PRESS

Journal of Wind Engineering and Industrial Aerodynamics 96 (2008) 1390 1411

Contents lists available at ScienceDirect

Journal of Wind Engineering


and Industrial Aerodynamics
journal homepage: www.elsevier.com/locate/jweia

A CFD investigation into the transient aerodynamic


forces on overtaking road vehicle models
R.J. Corin , L. He, R.G. Dominy
School of Engineering, University of Durham, South Road, Durham, DH1 3LE, UK

a r t i c l e in f o

a b s t r a c t

Article history:
Received 5 September 2006
Received in revised form
28 March 2007
Accepted 14 March 2008
Available online 13 May 2008

The transient aerodynamic forces occurring when one road vehicle


overtakes another were investigated using two-dimensional (2D)
computational uid dynamics. The relative velocity of the vehicles
was varied to allow comparison of the quasi-steady and unsteady
modelling approaches. The quasi-steady approach captured the
pseudo-periodic variation in the aerodynamic forces during the
overtaking manoeuvre. However, at vehicle velocities typical of
motorway driving conditions, it did not adequately predict the
magnitude of these forces. The study also identied signicant
dynamic ow features occurring during overtaking manoeuvres in a
crosswind. The dynamic variation in the aerodynamic forces was up
to 400% greater than that predicted using quasi-steady analysis,
indicating that the quasi-steady approach is totally unsuitable for
modelling overtaking manoeuvres in a crosswind. These dynamic
effects are likely to have a considerable impact on the stability of the
vehicles involved. With little existing work on passing manoeuvres
in a crosswind, these results highlight the importance of dynamic
effects and the need for further investigation into the problem.
& 2008 Elsevier Ltd. All rights reserved.

Keywords:
Overtaking vehicles
Passing manoeuvres
Crosswind
Quasi-steady modelling
Unsteady modelling
CFD

1. Introduction
As one vehicle passes another during an overtaking manoeuvre the ow elds around the two
vehicles interact generating transient aerodynamic forces. These forces can have an adverse effect on
vehicle handling and stability. Practical problems regarding the relative movement of wind tunnel
models and data acquisition under unsteady conditions have meant wind tunnel simulations of
overtaking manoeuvres have traditionally been completed using the quasi-steady approach. The

 Corresponding author. Tel.: +44 7813079699.

E-mail address: robertcorin@gmail.com (R.J. Corin).


0167-6105/$ - see front matter & 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jweia.2008.03.006

ARTICLE IN PRESS
R.J. Corin et al. / J. Wind Eng. Ind. Aerodyn. 96 (2008) 13901411

Nomenclature
Cd
Cp
Cs
Cym
D
k
L
M
P
PN
Re

aerodynamic drag coefcient


surface static pressure coefcient
aerodynamic side force coefcient
aerodynamic yawing moment coefcient
drag force (N)
ratio of relative velocity to overtaken vehicle velocity k Vr/V
length
of
overtaken
model
(PARAD1) (m)
yawing moment (Nm)
static pressure (N/m2)
free stream static pressure (N/m2)
Reynolds Number

SF
V
Vcw
Vr
Vres
W
x
y
b
Dx
Dy
r

1391

side force (N)


velocity of overtaken vehicle (m/s)
crosswind Velocity (m/s)
relative velocity between overtaking and overtaken vehicle (m/s)
resultant velocity, vector summation of V, Vr and Vcw (m/s).
width
of
overtaken
model
(PARAD1) (m)
lateral position on model (m)
longitudinal position on model (m)
yaw angle (degrees)
lateral spacing between models (m)
longitudinal spacing between models (m)
air density (kg/m3)

quasi-steady approach assumes that the relative velocity of the vehicles has a negligible effect and
the ow conditions remain approximately steady during the overtaking manoeuvre. As such, the
overtaking manoeuvre is analysed with the vehicles at a number of discrete static positions.
The validity of the quasi-steady approach and the impact of the ratio between the relative velocity
and the overtaken vehicle velocity (k Vr/V) is an area of some debate. Yamamoto and Nakagawa
(1997) conducted a dynamic study on 1/10th scale models and concluded that, for a velocity ratio of
ko0.25, dynamic effects could be neglected and the problem could be modelled statically. Telionis et
al. (1979) conducted tow tank analysis on a greater range of velocity ratios and suggested that
dynamic effects were insignicant up to ko0.4.
The unsteady computational uid dynamics (CFD) analysis of Okumura and Kuriyama (1997) was
conducted at higher velocity ratios and concluded that dynamic effects were important. For a
velocity ratio of k 0.5 the peak yawing moment of the vehicle being passed was almost 100% higher
than that predicted using the quasi-steady approach.
The experimental study of Gillieron and Noger (2004) combined analytical methods and CFD with
use of the new experimental overtaking test bench at the Institute Aerotechnique de Saint Cyr lEcole.
The relative velocity range tested was 010 m/s (022 mph) on 1/5th scale Ahmed models with a free
stream velocity of 30 m/s (68 mph). Results indicated that the side force at a relative velocity of 10 m/
s (22 mph, k 0.32) was 60120% higher than that predicted by the quasi-steady analysis. The latest
report of Noger et al. (2005) suggests that the aerodynamic forces on overtaking road vehicles
become velocity dependent when the ratio of the relative velocity to the overtaken vehicle velocity
(k) is greater than 0.2. Data presented for a velocity ratio of kE0.5 showed aerodynamic behaviour
differing signicantly from that predicted using quasi-steady analysis.
It is widely recognised that crosswinds can adversely inuence vehicle handling and stability and
create potentially hazardous driving conditions. However, little work has been completed on the
inuence of a crosswind on overtaking manoeuvres. Most existing studies concentrate on vehicles in
isolation such as Doctons (1996) study on the impact of transient crosswinds on an individual
passenger vehicle. Docton found that ow separation from the leeward side of the vehicle took a
nite amount of time to develop as the vehicle moved into the crosswind. This delay in separation
was accompanied by exaggeration of low-pressure regions, resulting in aerodynamic forces
exceeding those predicted using steady analysis by 4060%.
Telionis et al. (1979) gave some consideration to the impact of a crosswind on the aerodynamic
forces encountered by a small car passing a large truck. Their study on crosswinds at angles up to
b 241 indicated increases in the drag and side force of 100150% as the car emerged into the
crosswind from the leeward wake of the truck. In the recent experimental study by Noger et al. (2005) it

ARTICLE IN PRESS
1392

R.J. Corin et al. / J. Wind Eng. Ind. Aerodyn. 96 (2008) 13901411

was observed that for yaw angles of b 7101 there was considerable interaction between the
overtaking vehicle and the wake of the overtaken vehicle. The combination of crosswind ow structures
and the transient ow conditions occurring during overtaking manoeuvres may therefore signicantly
amplify the aerodynamic forces on the vehicles and further reduce vehicle stability. However, at present
the limited existing work means these ow conditions are not sufciently understood.
In summary, although the impact of relative velocity on the aerodynamic forces on overtaking
road vehicles has been investigated on a number of occasions the results remain inconclusive. There
has also been little work investigating the fundamental dynamic effects associated with overtaking
in a crosswind. The present work therefore uses two-dimensional (2D) CFD to overcome the practical
problems associated with wind tunnel testing and provide a visualisation of the changes occurring in
the ow-eld during overtaking manoeuvres. The CFD method was used to compare the quasi-steady
and the unsteady simulations to clarify the signicance of dynamic effects on the overtaking
manoeuvre and identify the limitations of the quasi-steady modelling approach. The scope of this
project extends beyond existing work to analyse the transient aerodynamic effects associated with a
crosswind and provide further fundamental understanding of their impact on passing manoeuvres.
2. Experimental apparatus
Wind tunnel testing was conducted to allow validation of the CFD solver for ow around the
chosen model geometries. The vehicle models available for this testing were used by Sims-Williams
(2001) and were selected to promote 2D ow and provide relatively simple geometries for
computational modelling. Each model was tted with pressure tapings at the model mid-height
(Fig. 1a and b). Aluminium end plates (250 mm  670 mm  3 mm) with chamfered ends perpendicular to the ow were attached to the top and bottom of each model to reduce ow threedimensionality. The models were securely attached to the tunnel oor on four 60 mm long 10 mm
diameter cylindrical spacers to avoid the tunnel oor boundary layer.
All experimental wind tunnel testing was conducted in the University of Durham 2 m2 open-jet
open-return wind tunnel at the maximum tunnel velocity of 29 m/s, which equates to a Reynolds
number based on model length of Re(L) 9.8  105. The non-dimensionalisation of pressures was
performed using readings from a pitot-static probe at the start of the working section.

3. Coordinate system and non-dimensional coefcients


For meaningful comparison between datasets the pressure, force and moment coefcients were
specied according to Fig. 2 and non-dimensionalised according to Eqs. (1)(4), where D, SF and M
were the drag force, side force and yawing moment obtained by integrating the pressure distribution
around the model. Fig. 2 also displays the lateral and longitudinal model spacing convention adopted.
Note that the lateral spacing (Dx) was non-dimensionalised using the PARAD1 model width (W) and
the longitudinal spacing (Dy) was non-dimensionalised using the PARAD1 model length (L). Details
of the velocity vector sign convention used for a crosswind are given in Fig. 3
C p s

Cd

Cs

ps  p1
1=2rV 2res
D=span

1=2rV 2res W
SF=span
1=2rV 2res L

C ym

M=span
1=2rV 2res L2

(1)

(2)

(3)

(4)

ARTICLE IN PRESS
R.J. Corin et al. / J. Wind Eng. Ind. Aerodyn. 96 (2008) 13901411

1393

Fig. 1. (a) Dimensions of the main body of the PARAD1 vehicle model: (b) dimensions of the main body of the PARAD2 vehicle
model.

Fig. 2. Coordinate system for aerodynamic forces and moments, and lateral and longitudinal model spacing.

4. Computational simulation approach


The CFD package used for this study was developed by Professor Li He of the University of Durham
primarily for the investigation of unsteady statorrotor interaction in axial ow turbines. The
computational solver adopts a cell-centred nite volume approach for which the temporal

ARTICLE IN PRESS
1394

R.J. Corin et al. / J. Wind Eng. Ind. Aerodyn. 96 (2008) 13901411

Fig. 3. Velocity vector triangle for a crosswind.

integration of the discretised equations is completed using the four-step RungeKutta time-marching
scheme. A time-consistent multi-grid technique is used to accelerate the unsteady calculations
(further information is provided in He, 2000).
4.1. Computational mesh design
The size of the computational domain for the overtaking simulation was set at 2 L either side of
the models and 3 L upstream and downstream of the models. This domain size was consistent with
that used by Okumura and Kuriyama (1997) and was deemed adequate to capture the changes in the
ow eld during the interaction while giving sufcient distance between the models and the
boundaries.
The computational method employed used multi-block structured meshes with 9 mesh blocks for
each model. Enlargement ratios between successive cells were set between 1.03 and 1.08 to give
appropriate mesh expansion in the far eld. This conguration ensured acceptable step changes in
the cell sizes, particularly at the interfaces between mesh blocks (Fig. 4a and b). Computational space
and calculation time limitations restricted the number of points on the model surface to 250. The
total number of mesh cells in the computational domain for the two models was approximately
5  104.
4.2. Boundary conditions
The outer boundaries of the computational domain were linked to the unbounded far eld
boundary conditions for which stagnation pressure, stagnation temperature and ow speed and
direction were set.
At the boundaries between xed blocks the solver determined the ow variables from the
corresponding boundary points in the adjacent block using linear interpolation. For blocks
undergoing relative motion the solver used a second-order interpolation and correction method to
transfer instantaneous information directly across the interface (see He, 2000). For the majority of
the overtaking manoeuvres the xed and moving mesh blocks were not aligned. To account for this a
periodic boundary condition was imposed as illustrated in Fig. 5.
At the surface of the body the mesh density was insufcient to adequately resolve the boundary
layer. Therefore, a slip wall condition was used to determine the velocity while a logarithmic law was
used to determine the surface shear stresses. For separated ow regions with reversed ow, the solid
surface treatment is equivalent to a non-slip wall.
4.3. Solution method and time step
For a steady ow solution time consistency was unimportant and local time steps meeting the
local stability requirements could be employed in each block to reduce computation time. The steady
solution was deemed to have converged when the percentage change in velocity between time steps
dropped below 0.05%. The time step history for the drag and side forces was then checked to verify
convergence.

ARTICLE IN PRESS
R.J. Corin et al. / J. Wind Eng. Ind. Aerodyn. 96 (2008) 13901411

1395

Fig. 4. (a) 9-Block computational mesh. The values on the diagram refer to the number of grid points in the x and y direction
for each block (position shown Dy/L 0, Dx/W 0.5). (b) Detail of mesh density at front of models. Note that for the blocks
surrounding the models the values refer to the number of points around the model surface and the number of points
comprising the block thickness, respectively.

ARTICLE IN PRESS
1396

R.J. Corin et al. / J. Wind Eng. Ind. Aerodyn. 96 (2008) 13901411

Fig. 5. Periodic boundary conditions. Boundary ab boundary cd.

For the dynamic simulation the full unsteady time-consistent multi-block solver with a four-level
multi-grid was employed. The time step dependence of the solution was tested using speed up
factors of 15 and 30 times the time step dictated by the CFL stability limit. A speed up factor of 30
was deemed to provide an adequate compromise between solution accuracy and computation time.
The majority of steady solutions converged within 10,000 time steps while the unsteady solutions
required 10,00020,000 steps and the crosswind simulations required up to 30,000 steps.

4.4. Turbulence modelling


The one-equation SpalartAllmaras (SA) model was employed and the recirculating near wake
described by Bearman (1980) was observed at the rear of the isolated PARAD1 model.
For many bluff body problems the position of separation is governed by the geometry of the body.
However, for cases without a clear, geometry-dened separation point the problem is more
complicated. While details about the boundary layer thickness and pressure gradients can be
obtained from the Reynolds Number and body geometry, the point of separation is strongly
dependent on the ow structures and turbulence production within the buffer region of the
boundary layer (Krajnovic and Davidson, 2004). Detailed modelling of turbulence development
within the boundary layer is therefore required to produce accurate predictions regarding separation.
As a consequence, the CFD separation prediction is heavily reliant on the mesh density and the
numerical turbulence and separation modelling techniques adopted. Unfortunately, further increases

ARTICLE IN PRESS
R.J. Corin et al. / J. Wind Eng. Ind. Aerodyn. 96 (2008) 13901411

1397

in mesh density in the boundary layer and at the rear of the models were prohibited by the
computational processing limits within the solver. While errors in the separation prediction were
undesirable, it was accepted that computational limitations prohibited any signicant improvements
in this area.
4.5. Computational analysis procedure
Initial geometry validation trials were conducted for the PARAD1 and PARAD2 models in isolation
using the steady solver under conditions matching the Sims-Williams (2001) experimental
conditions (Re 8.9  105).
The model scale was then increased by a factor of 10 to represent more accurately a full-scale
road vehicle (Vehicle length L 5.22 m) and to maintain the appropriate unsteady frequencies
and timescales associated with the overtaking manoeuvre. The Reynolds number was also increased
to Re(L) 9.4  106 to match standard motorway driving conditions of V 27 m/s (60 mph).
The quasi-steady analysis was followed by unsteady analyses at relative vehicle velocities of
Vr 2.2, 4.5, 8.9, 13.4 m/s (5, 10, 20, 30 mph). These relative velocities equate to velocity ratios of
k 0.083, 0.17, 0.33, 0.5, respectively. Note that the unsteady runs were started from a converged
steady solution.
Finally quasi-steady and unsteady simulations were performed in a b 201 crosswind. For both
simulations the PARAD1 model velocity was V 27 m/s (60 mph) with a crosswind of Vcw 9.8 m/s
(22 mph). The relative velocity for the unsteady simulation was Vr 8.9 m/s (20 mph, k 0.33).

5. Results
5.1. Single model geometry validation
Fig. 6 displays experimental and computational pressure coefcients as a function of the
streamwise position along the PARAD1 model. The plot includes data recorded by Sims-Williams
(2001) for the same PARAD1 model obtained in the University of Durham 0.85 m  0.55 m open-jet,
open-return wind tunnel.
The kink in the pressure distribution at y/LE0.04 for both the experimental results and SimsWilliams (2001) data indicated the formation of a small separation bubble at the leading corner of
the PARAD1 model (Fig. 6). The experimental variability in the extent of leading edge separation and
the links to the Reynolds number pose considerable problems when modelling this type of ow
structure using CFD. The greatest discrepancy between the computational and experimental data
occurred at the rear of the model where the computational simulation struggled to predict the
position of ow separation. It was recognised that separation prediction was one of the limitations of
the computational model. Similar results were obtained for the PARAD2 model, with the
computational simulation showing generally good agreement with the experimental results.
A further point to note was the under-prediction of the pressure along the sides of the model. This
feature was observed by Sims-Williams (2001) and was attributed to blockage effects in an open jet
wind tunnel. Over-deection of the outer streamlines of the jet results in lower velocities and higher
pressures at the model surface when compared to that of an innite test section. It is understood that
the area blockage ratio for the wind tunnel used by Sims-Williams was greater than the 4% blockage
ratio for the tunnel used in this study. The slightly lower pressures recorded in this study relative to
Sims-Williams results were therefore consistent with this difference in blockage ratio.
5.2. Validation of unsteady solver
The computational code for modelling unsteady ow has been extensively validated for the case
of rotorstator interaction in axial ow turbines. Further details of the unsteady solver validation are
given in He (2000).

ARTICLE IN PRESS
1398

R.J. Corin et al. / J. Wind Eng. Ind. Aerodyn. 96 (2008) 13901411

Fig. 6. Comparison of experimental and CFD pressure distributions for PARAD1 model in isolation.

5.3. Effect of relative velocity on overtaken model (PARAD1)


Fig. 7ac illustrates the impact of increasing the relative velocity of the vehicles from 0 to 13.4 m/s
on the aerodynamic forces on the overtaken vehicle (PARAD1). The plots highlight signicant underprediction of the aerodynamic forces by the quasi-steady approach. Whilst continuous drag and side
force data was obtained from the solver, the yawing moment in Fig. 7c had to be calculated from the
surface pressure readings at discrete points in the overtaking manoeuvre, and required the unsteady
simulation to be run from the start position to each point of interest. Due to the time constraints
on the project this was only completed for one example unsteady case (V 27 m/s (60 mph),
Vr 8.9 m/s (20 mph)). It should be noted that the rapid oscillations in the drag and side forces at Dy/
LE2 were due to the stabilisation of the unsteady solver and were not attributable to genuine
aerodynamic effects.
5.4. Effect of relative velocity on overtaking model (PARAD2)
Fig. 8ac illustrates the impact of increasing the relative velocity of the vehicles from 0 to 13.4 m/s
on the aerodynamic forces on the overtaking vehicle (PARAD2). In these plots the quasi-steady
approach is shown to over-predict the aerodynamic forces on the overtaking vehicle. The coefcients
for the overtaking vehicle (PARAD2) were non-dimensionalised using the total PARAD2 velocity
(V+Vr) as the reference velocity.
5.5. Effect of a crosswind on the overtaking manoeuvre
5.5.1. Comparison of unsteady and quasi-steady analysis ow structures
The introduction of a crosswind further complicated the overtaking process and presented
additional questions concerning the applicability of the quasi-steady analysis approach to such situations. The crosswind study at a yaw angle of b 201 was conducted using both the quasi-steady

ARTICLE IN PRESS
R.J. Corin et al. / J. Wind Eng. Ind. Aerodyn. 96 (2008) 13901411

1399

Fig. 7. Effect of relative velocity on the variation in the aerodynamic coefcients during an overtaking manoeuvre. Data
plotted for the overtaken model (PARAD1) for a lateral spacing of Dx/W 0.5 and a Reynolds number of 9.4  106: (a) variation
in drag coefcient, (b) variation in side force coefcient and (c) variation in yawing moment coefcient.

approach and the example unsteady conguration of V 27 m/s (60 mph), Vr 8.9 m/s (20 mph,
k 0.33). Figs. 9 and 10 illustrate the pressure eld and ow streamlines obtained using the two
methods.
The surface static pressure coefcients for the overtaken and overtaking models are shown in Figs.
11 and 12. Note that the reference velocity used to calculate the non-dimensional pressure
coefcients was the resultant velocity Vres. For the overtaking model (PARAD2) this resultant velocity
was the vector summation of V, Vr, and Vcw.

5.5.2. Effect of a crosswind on the overtaken model (PARAD1)


The striking differences between the quasi-steady and unsteady ow structures in Figs. 9 and 10
had a considerable effect on the aerodynamic forces. Fig. 13ac illustrates the drag, side force and
yawing moment coefcients for the overtaken vehicle (PARAD1). Note that the coefcients were nondimensionalised using the resultant velocity Vres.

ARTICLE IN PRESS
1400

R.J. Corin et al. / J. Wind Eng. Ind. Aerodyn. 96 (2008) 13901411

Fig. 8. Effect of relative velocity on the variation in the aerodynamic coefcients during an overtaking manoeuvre. Data
plotted for the overtaking model (PARAD2) for a lateral spacing of Dx/W 0.5 and a Reynolds number of 9.4  106: (a)
variation in drag coefcient, (b) variation in side force coefcient and (c) variation in yawing moment coefcient.

5.5.3. Effect of a crosswind on the overtaking model (PARAD2)


The variations in the aerodynamic coefcients for the overtaking model (PARAD2) are illustrated
in Fig. 14ac. Note that the forces and moments were non-dimensionalised using the resultant
velocity Vres. For the PARAD2 model this resultant velocity was based on the vector summation of V,
Vr, and Vcw.

6. Discussion
Although both the unsteady and quasi-steady analysis methods show similar pseudo-periodic
trends the magnitudes of the aerodynamic forces differ signicantly (Figs. 7 and 8). One dynamic
consideration is the change in pressure around the overtaking model as the relative velocity is
increased. It can be assumed that the variation in pressure coefcient (Cp) around the overtaking
model in isolation is independent of the model velocity. Since the pressure coefcient at a position s

ARTICLE IN PRESS
R.J. Corin et al. / J. Wind Eng. Ind. Aerodyn. 96 (2008) 13901411

1401

Fig. 9. Pressure elds and ow streamlines for the quasi-steady analysis of overtaking manoeuvre in a b 201 yaw
crosswind.

Fig. 10. Pressure elds and ow streamlines for the UNSTEADY analysis of overtaking manoeuvre in a b 201 yaw
crosswind.

ARTICLE IN PRESS
1402

R.J. Corin et al. / J. Wind Eng. Ind. Aerodyn. 96 (2008) 13901411

Fig. 11. Comparison of surface static pressure coefcients for quasi-steady and unsteady simulations of the overtaking
manoeuvre in a b 201 yaw crosswind. Overtaken model (PARAD1).

is dened by Cp(s) (P(s)PN/(1/2)rV2) any increase in the free stream velocity (V) results in an
increase in the magnitude of the static pressure (P(s)). Therefore, as the overall velocity of the
overtaking vehicle is increased (i.e. Vr is increased) the magnitudes of the static pressures around the
model are also increased.

6.1. Effect of relative velocity on the overtaken model (PARAD1)


As the magnitudes of the static pressures around the overtaking vehicle (PARAD2) are increased
the effect of the overtaking manoeuvre on the PARAD1 vehicle is also increased. The peak-to-peak
drag coefcient variation for the overtaken vehicle (PARAD1) increased almost linearly by
approximately 12% for every m/s increase in the relative velocity (or 5% per mph increase in Vr).
For a relative velocity of 13.4 m/s (30 mph, k 0.5) the peak-to-peak drag variation was 150% greater
than that predicted by the quasi-steady simulation (Fig. 7a).
Interestingly the side force variation for the overtaken vehicle (PARAD1) shown in (Fig. 7b) does
not show such dependence on the relative velocity of the vehicles. For Dy/Lo0.25 the side force on
the overtaken vehicle was predominantly due to the positive pressure region at the front of the
overtaking vehicle. As the degree of vehicle overlap increased, the low-pressure region between the
models became the dominant factor and the direction of the side force reversed (Fig. 7b).
Although the strength of the positive pressure eld at the front of the overtaking model increased
with relative velocity, this positive pressure increase was balanced by the reduction in pressure in the
region of overlap. These opposing dynamic effects appear to explain the limited impact of the relative
velocity on the side force on the PARAD1 model.

ARTICLE IN PRESS
R.J. Corin et al. / J. Wind Eng. Ind. Aerodyn. 96 (2008) 13901411

1403

Fig. 12. Comparison of surface static pressure coefcients for quasi-steady and unsteady simulations of the overtaking
manoeuvre in a b 201 yaw crosswind. Overtaking model (PARAD2).

The lack of a clear relationship between the relative velocity and side force is inconsistent with
the ndings of Gillieron and Noger (2004) who observed increases in peak Cs with increasing Vr.
Having said this, dynamic effects had an impact for all the relative velocities and resulted in a peak
side force up to 80% greater than that predicted by the quasi-steady simulation (Fig. 7b). This is
consistent with the 60120% increases in Cs recorded by Gillieron and Noger (2004) for a velocity
ratio of k 0.32.
The changing pressure elds during the overtaking manoeuvre resulted in a yawing
moment away from the overtaking model for Dy/Lo0 and towards the overtaking model for
Dy/L40 (Fig. 7c). The main impact of the increased relative velocity of the vehicles was to
increase the strength of the low-pressure regions around front and rear corners of the overtaken
vehicle (PARAD1). These pressure changes resulted in a 70% increase in the yawing moment
for a relative velocity of 8.9 m/s (20 mph, k 0.33) compared to that predicted by the quasisteady analysis (Fig. 7c). This increase in peak yawing moment is consistent with the 100%
increase observed by Okumura and Kuriyama (1997) for a 3D passing manoeuvre at a velocity ratio of
k 0.5.
6.2. Effect of relative velocity on the overtaking model (PARAD2)
Fig. 8a illustrates that the drag coefcient for the overtaking vehicle (PARAD2) decreased with
increasing relative velocity particularly during the second half of the manoeuvre (Dy/L40). At the

ARTICLE IN PRESS
1404

R.J. Corin et al. / J. Wind Eng. Ind. Aerodyn. 96 (2008) 13901411

Fig. 13. Effect of a crosswind on the variation in the aerodynamic coefcients during an overtaking manoeuvre. Data plotted
for the overtaken model (PARAD1) for a lateral spacing of Dx/W 0.5 and a Reynolds number of 1 107. The unsteady analysis
conditions were V 27 m/s (60 mph), Vr 8.9 m/s (20 mph), Vcw 9.8 m/s (22 mph): (a) variation in drag coefcient, (b)
variation in side force coefcient and (c) variation in yawing moment coefcient.

relative velocity of 8.9 m/s (20 mph) the peak-to-peak drag variation was only 90% of that predicted
by the quasi-steady analysis.
As the velocity of the overtaking vehicle increases, the relative strength of the pressure eld
around the overtaken model (PARAD1) decreases. Thus, the impact of the overtaken model on the
overtaking model decreases with increasing relative velocity. Fig. 15 illustrates the reduced effect of
the PARAD1 model on the PARAD2 model and the impact on the pressure distribution.
This argument is supported by reductions in the side force and yawing moment coefcient
variation with increasing relative velocity (Fig. 8ac). The peak-to-peak variations in side force and
yawing moment for a relative velocity of 8.9 m/s (20 mph, k 0.33) were only 60% and 65%,
respectively of those predicted using the quasi-steady approach. The ndings are in agreement with

ARTICLE IN PRESS
R.J. Corin et al. / J. Wind Eng. Ind. Aerodyn. 96 (2008) 13901411

1405

Fig. 14. Effect of a crosswind on the variation in the aerodynamic coefcients during an overtaking manoeuvre. Data plotted
for the overtaking model (PARAD2) for a lateral spacing of Dx/W 0.5 and a Reynolds number of 1 107. The unsteady analysis
conditions were V 27 m/s (60 mph), Vr 8.9 m/s (20 mph), Vcw 9.8 m/s (22 mph): (a) variation in drag coefcient, (b)
variation in side force coefcient and (c) variation in yawing moment coefcient.

those of Okumura and Kuriyama (1997) who observed decreases in the forces on the passing model
with increasing Vr.
6.3. Effect of a crosswind on the overtaking manoeuvre
The crosswind study provided considerable insight into the differences between the dynamic and
quasi-steady modelling methods. For the standard overtaking manoeuvre discussed in Section 6.1
inviscid effects, such as the ow acceleration between the models, primarily governed the
aerodynamic force variation. However, at signicant yaw angles the effect of ow separation
becomes more important as illustrated in Figs. 9 and 10.

ARTICLE IN PRESS
1406

R.J. Corin et al. / J. Wind Eng. Ind. Aerodyn. 96 (2008) 13901411

Fig. 15. Features of the unsteady and quasi-steady pressure distributions for the overtaking model (PARAD2) (Dy/L 0.25).

6.4. Dynamic separation delay and suction spike growth


At Dy/L 1.5 both the quasi-steady and unsteady analyses (Figs. 9a and 10a) show a
recirculating separation region on the leeward side of the PARAD1 model. (The larger corner radius
for the PARAD2 model appeared to reduce ow separation from this model).
For the quasi-steady case this separation feature extended to the PARAD2 model as it entered the
recirculating wake region (Dy/L 1.0, Fig. 9b). For positions of signicant model overlap (Dy/
L 0.5, 0.0, 0.5) the proximity of the PARAD2 model suppressed ow separation from the PARAD1
model. Then, when the overtaking model was beyond the front of the PARAD1 model (Dy/L41.0), the
recirculating separation region on the leeward side of the PARAD1 model was re-established and at
Dy/L 1.5 the ow structure closely matched that at Dy/L 1.5 (Fig. 9a and g). The variation in ow
separation from the PARAD1 model is shown in Fig. 16.
Qualitatively similar ow separation and separation suppression structures were observed
for the dynamic simulation (Fig. 10ac). However, the re-establishment of the separation region
at Dy/L 1.0 that had been observed for the quasi-steady analysis did not occur. Fig. 10dg
clearly shows the persistence of the ow turning caused by the PARAD2 model, with the
streamlines on the leeward side of the PARAD1 model almost parallel to the model surface.
The variation in ow separation from the PARAD1 model for the unsteady case is summarised
in Fig. 17.
Unfortunately pressure eld data was not collected for more advanced positions in the overtaking
manoeuvre, and it was not possible to determine whether separation reoccurred at Dy/L41.5.
However, it appears that the relative motion of the two models resulted in some form of dynamic
delay in ow separation from the PARAD1 model.
This dynamic effect is similar to that encountered by Docton (1996) who observed a delay in
ow separation during the transient analysis of a vehicle entering a 301 crosswind. Doctons
(1996) suggestion that the phenomenon of separation requires a nite amount of time to develop at a
surface appears to be applicable to this problem. While it was not possible to determine how long it
took for the separation region to become re-established without further analysis, the results of this
study suggest that the separation delay is a signicant feature of overtaking manoeuvres in a
crosswind.

ARTICLE IN PRESS
R.J. Corin et al. / J. Wind Eng. Ind. Aerodyn. 96 (2008) 13901411

1407

Fig. 16. Flow separation from PARAD1 model in a crosswind (QUASI-STEADY simulation).

Fig. 17. Flow separation from PARAD1 model in a crosswind (UNSTEADY simulation).

6.5. Pressure eld variation for the overtaken model (PARAD1)


The notable differences between the quasi-steady and unsteady ow structures discussed above
had striking effects on the pressure elds around the models. Fig. 11f shows the peak suction
pressure for the dynamic simulation was 2.2 times greater than that predicted by the quasi-steady
simulation. Inspection of the ow structure plots in Figs. 9f and 10f suggests that this was due to the
dynamic delay in separation. The lack of separation magnied the degree of ow acceleration around
the front of the model, generating a strong suction spike. In contrast, separation in the quasi-steady
case greatly reduced the degree of ow turning and a far lower suction peak was generated.
This dynamic effect is also similar to that observed by Docton (1996) who recorded a dynamic
suction peak over 3 times greater than that predicted by the steady analysis. Docton also attributed
this suction peak exaggeration to the nite time taken for separation to develop at a surface.
The extreme exaggeration of the suction spike for the dynamic simulation also appeared to cause
an anticlockwise movement of the stagnation point on the PARAD1 model. This placed a positive
pressure on the far side of the PARAD1 model, reducing the degree of ow turning around the lower
far corner, and thereby reducing the magnitude of the suction spike in this region (Fig. 11f).

6.6. Pressure eld variation for the overtaking model (PARAD2)


The key differences between the quasi-steady and dynamic pressure distributions occurred in the
latter stages of the manoeuvre (Dy/L40). In contrast to the quasi-steady simulation, the dynamic
simulation does not show ow separation from the PARAD2 model for this stage of the overtaking
manoeuvre. Instead, the dynamic exaggeration of the suction spike at the front of the model was
observed in the same fashion as was observed for the PARAD1 model. At Dy/L 1.5 the dynamic
suction peak was over 2 times greater than that predicted by the quasi-steady analysis (Fig. 12(g).
It should be noted that, for the quasi-steady case, the yaw angle on both models is identical,
however, as the relative velocity of the models is increased, the effective yaw angle on the overtaking
model (PARAD2) is reduced. At a relative velocity of Vr 8.9 m/s (20 mph) the yaw angle for PARAD2
reduces from 201 to 15.31. The velocity triangle for the overtaking model is given in Fig. 18. For this
kind of bluff body the ow incidence at which separation occurs is unpredictable, and it is possible

ARTICLE IN PRESS
1408

R.J. Corin et al. / J. Wind Eng. Ind. Aerodyn. 96 (2008) 13901411

Fig. 18. Velocity triangle for the overtaking model (PARAD2) in a crosswind.

that this incidence reduction explains the absence of separation from the PARAD2 model for the
duration of the dynamic simulation. The fully attached ow around the dynamic PARAD2 model
undergoes greater ow turning and acceleration around the model corners than for the separated
ow case. It is therefore unsurprising that the magnitude of the suction spike is greater for the
dynamic case than it is for the quasi-steady simulation. It is acknowledged that differences in the
numerical implementation of the steady and unsteady solvers also affect the separation prediction.
Whilst the difference in yaw angle requires care to be taken when comparing the quasi-steady and
unsteady results for the overtaking model (PARAD2), the yaw angle for the overtaken model
(PARAD1) remained the same for both the quasi-steady and unsteady cases. However, it is clear that
during the overtaking manoeuvre the ow elds around the two models interact. Therefore a single
quasi-steady simulation will never be able to truly represent an unsteady overtaking manoeuvre for
which, even for identical models, the difference in yaw angle, due to the relative velocity of the
models, results in a different ow eld around the overtaking model. It is possible that the simple
difference in ow elds around the PARAD2 model due to the different yaw angles, and the impact of
the different ow elds on the overtaken model, had a comparable impact on the overtaking
manoeuvre to the dynamic effects discussed above. Unfortunately time constraints prevented the
analysis of the quasi-steady case at the lower yaw angle. Further studies in this area would provide
valuable insight into the impact of yaw angle related changes in the ow eld on the forces occurring
during overtaking manoeuvres.

6.7. Impact of dynamic effects on aerodynamic force variation


Pressure plots were not obtained beyond Dy/L 1.5, and it was not possible to determine
precisely whether the exaggerated suction spikes decreased with time or whether ow separation
reoccurred. However, Figs. 13a and 14a show that the drag coefcient for both the PARAD1 and
PARAD2 models steadily increased for Dy/L41.5. This indirectly suggests that the front corner
suction spikes on both models were strongly time dependant and gradually decreased as the vehicles
moved further apart resulting in a considerable extension to the interaction distance. The interaction
distance of at least 4L indicated in Figs. 13 and 14 was 3060% greater than that observed for the
overtaking manoeuvre without a crosswind.
The impact of the dynamic ow features discussed in previous sections meant that the quasisteady simulation failed to capture the oscillations and the magnitudes of the aerodynamic forces for
both the PARAD1 and PARAD2 models. In general, the quasi-steady simulation under-predicted the
peak-to-peak aerodynamic force variation on the PARAD1 model and, at times, failed completely to
predict the direction of the forces. For example, at Dy/L 1 the quasi-steady analysis predicted a
large positive drag coefcient while the unsteady approach predicted an equally large negative drag
coefcient (Fig. 13a). The PARAD1 peak-to-peak drag, side force and yawing moment variations for
the unsteady analysis were 400%, 120% and 200% greater than those predicted by the quasi-steady
analysis, respectively.
The overall drag and side force trends for the PARAD1 model were similar to those described for
the zero crosswind case. Unsurprisingly, the increased yaw angle resulted in an increase in the mean
side force in the direction of the crosswind, while the dynamic separation delay and suction peak
exaggeration magnied the drag variation.
Fig. 14 for the PARAD2 model illustrate the dramatic variation in aerodynamic forces occurring as
a result of plunging into the unsteady wake of the PARAD1 model. As the PARAD2 model emerged

ARTICLE IN PRESS
R.J. Corin et al. / J. Wind Eng. Ind. Aerodyn. 96 (2008) 13901411

1409

Fig. 19. Dynamic formation of a recirculation region in a crosswind.

into the crosswind from the wake there was a sharp rise in the side force which peaked at Dy/LE1.0
(Fig. 14b). This side force decreased as the longitudinal gap between the models grew indicating a
gradual reduction in the magnitude of the low-pressure spike. Similar features have been observed
for dynamic crosswind tests and results suggest that the dynamic effects decay after 23 vehicle
lengths (Barnard, 1996).
6.8. Other dynamic crosswind features
One further dynamic feature of the interaction is illustrated in Fig. 10b and c. At this position the
motion of the overtaking model draws uid forward from the rear of the overtaken model. As the
overtaking model begins to overlap the overtaken model this uid motion is opposed by uid
attempting to pass between the models and a recirculation region is established (Fig. 19).
It is possible that the uctuations in this dynamic recirculation region behind the PARAD1
model resulted in the oscillations in the PARAD1 drag coefcient between Dy/L 1.0 and Dy/L 0.0
(Fig. 13a).
6.9. Implications for vehicle comfort and stability
The dynamic stability of a vehicle is dependent on a combination of the transient aerodynamic
forces, driver response and vehicle chassis dynamics. While the aerodynamic coefcients calculated
during this study provide valuable indicators for areas of future work, it would be unwise to use them
as the basis for rm conclusions about dynamic vehicle stability.
One of the notable features of the changing pressure elds during an overtaking manoeuvre is the
impact of the venturi effect that results from the proximity of the models. It should be noted that;
whilst for the 2D case uid is forced to accelerate between the models resulting in considerable
reductions in static pressure (Fig. 12e), for true 3D vehicle geometries air is free to spill over and the
vehicle, thus alleviating the venturi effect and reducing the forces on the vehicles.
The dynamic effect likely to have the greatest impact on vehicle stability was the delay in ow
separation and the amplication of the low-pressure suction spike at the front corner of the
overtaken model in a crosswind.
Previous studies have acknowledged that it is the growth of the negative pressure region on the
leeward side of a vehicle that has the greatest inuence on aerodynamic stability in a crosswind
(Hucho, 1998).
The strength of this negative pressure region and the subsequent aerodynamic forces can be
greatly diminished if the vehicle geometry is designed to induce ow separation on the leeward side
of the vehicle (Hucho, 1998). This project illustrated an unsteady side force (with no separation) 10
times greater than that predicted by the quasi-steady simulation (with separation) for the PARAD1
model at Dy/L 1.0 (Fig. 13b). This nding suggests that appropriately designed devices promoting

ARTICLE IN PRESS
1410

R.J. Corin et al. / J. Wind Eng. Ind. Aerodyn. 96 (2008) 13901411

ow separation could signicantly reduce the impact of dynamic effects on the aerodynamic forces
on the vehicle. It should be noted that inducing separation has the adverse effect of increasing the
aerodynamic drag (as can be seen in Fig. 13a) and wind noise. In addition, sudden separation results
in rapid changes in the side forces and yawing moment and is likely to pose control problems for the
driver (Barnard, 1996).
The cabin noise associated with the external ow around a vehicle is one issue affecting passenger
comfort, with a dominant component resulting from eddy vortex creation in the region of the front
windscreen. For more realistic 3D vehicle shapes it is possible that the dynamic amplication of the
low-pressure spike around the front corners of the model in a crosswind (Fig. 11f) could act to
increase vortex generation in this region and therefore increase cabin noise.

7. Conclusions
Computational uid dynamics (CFD) methods were successfully developed and applied to the
study of the aerodynamic forces on overtaking road vehicles. The study involved simplied 2D
representations of road vehicles and focused on the effect of the relative velocity of the vehicles and
the inuence of a crosswind on the transient aerodynamic forces. Experimental wind tunnel testing
was conducted statically on each of the models in isolation to provide test data to validate the
computational solver. The experimental results highlighted the continuing problem of accurately
predicting boundary layer separation and complex turbulent ow structures using CFD.
The CFD investigation into the impact of relative velocity highlighted the dynamic amplication of
the strength of the pressure elds around the vehicles involved in the overtaking manoeuvre. The results
conrmed that the quasi-steady approach captures the overall pseudo-periodic trends in the force
variation during an overtaking manoeuvre. However, dynamic effects occurring at relative velocities
typical of motorway driving conditions prevent the quasi-steady method from accurately predicting the
peak magnitudes of these aerodynamic forces. For typical motorway vehicle velocities the dynamic
forces were up to 150% greater than the quasi-steady forces for the overtaken vehicle and were as little as
60% of the quasi-steady forces for the overtaking vehicle. The inadequacy of the quasi-steady approach
was most noticeable for velocity ratios of k40.2. It should be noted that for 3D vehicles the dynamic
pressure amplication would not be as signicant as observed for the 2D case as a result of 3D ow over
and under the vehicles reducing the strength of the Venturi effect between the models.
The computational study into overtaking manoeuvres in a crosswind identied signicant
dynamic effects. Dynamic delay in boundary layer separation from the overtaken vehicle resulted in
greatly amplied low pressure regions around the vehicles and produced aerodynamic force
variation up to 400% greater than that predicted using the quasi-steady approach. This dynamic
effect is likely to have a detrimental effect on the dynamic stability of road vehicles during overtaking
manoeuvres in crosswinds. It is acknowledged that an evaluation on representative 3D automobile
models and the use of measured or computed dynamic forces in a vehicle-dynamics simulation
programme would be required to accurately determine the impact of these dynamic effects on the
stability of real vehicles. However, the 2D results presented provide both an important rst step
towards full 3D simulation and a useful reference point for full understanding and identication of
truly 3D unsteady effects.
The signicance of these dynamic features means the quasi-steady approach is totally unsuitable
for analysis of overtaking manoeuvres in a crosswind. The result supports the conclusion that
transient simulations are required to accurately predict the main features of overtaking manoeuvres.
To the authors knowledge there is no existing work detailing these dynamic features in the context
of passing manoeuvres, and therefore, these ndings provides considerable interest and scope for
future investigation.

Acknowledgements
The author wishes to acknowledge the help of the Durham University School of Engineering.

ARTICLE IN PRESS
R.J. Corin et al. / J. Wind Eng. Ind. Aerodyn. 96 (2008) 13901411

1411

References
Barnard, R.H., 1996. Road Vehicle Aerodynamic Design. Longman, New York.
Bearman, P.W., 1980. Bluff body ows applicable to vehicle aerodynamics. J. Fluids Eng. 102, 265274.
Docton, M.K.R., 1996. The simulation of transient crosswinds on passenger vehicles. Ph.D. Thesis, University of Durham.
Gillieron, P., Noger, C., 2004. Contribution to the analysis of transient aerodynamic effects acting on vehicles. SAE Paper No.
2004-01-1311.
He, L., 2000. Three-dimensional unsteady NavierStokes analysis of stator-rotor interaction in axial-ow turbines. Proc.
IMechE 214 (Part A).
Hucho, W.-H. (Ed.), 1998. Aerodynamics of Road Vehicles: From Fluid Mechanics to Vehicle Engineering, fourth ed. SAE Inc.
Krajnovic, S., Davidson, L., 2004. Large-eddy simulation of the ow around a simplied car model. SAE Paper No. 2004-010227.
Noger, C., Regardin, C., Szechenyi, E., 2005. Investigation of the transient aerodynamic phenomena associated with passing
manoeuvres. J. Fluids Struct. 21, 231241.
Okumura, K., Kuriyama, T., 1997. Transient aerodynamic simulation in crosswind and passing an automobile. SAE Paper No.
970404.
Sims-Williams, D.B., 2001. Self-excited aerodynamic unsteadiness associated with passenger cars. Ph.D. Thesis, University of
Durham.
Telionis, F.P., Fahrner, C.J., Jones, G.S., 1979. An experimental study of highway aerodynamic interferences. J. Wind Eng. Ind.
Aerodyn. Aerodyn. Transportation, 113125.
Yamamoto, S., Nakagawa, K., 1997. Aerodynamic inuence of a passing vehicle on the stability of the other vehicles. JSAE Rev.
18 (1), 3944.

You might also like