You are on page 1of 12

ORIGINALS

R. J. Pattenden Æ S. R. Turnock Æ X. Zhang
Measurements of the flow over a low-aspect-ratio
cylinder mounted on a ground plane
Received: 23 July 2004 / Revised: 2 February 2005 / Accepted: 2 February 2005 / Published online: 12 May 2005
Ó Springer-Verlag 2005
Abstract The flow over a finite-height cylinder of aspect
ratio 1, with one end mounted on a ground plane and
the other end free, has been studied by means of surface
flow visualisation, particle image velocimetry (PIV) and
surface pressure measurements. The diameter-based
Reynolds number was 200,000. The mean flow topology
has been identified in three areas: the horseshoe vortex
system, the separated flow over the free-end and the
wake region. Evidence is shown for the existence of a
three-horseshoe vortex system, while the mean flow over
the free-end consists of an arch vortex with its bases on
the forward half of the free-end. There are two tip vor-
tices coming off the free-end. The wake region is found
to be highly unsteady, with considerable variation from
the mean flow.
1 Introduction
The circular cylinder has proved to be a fruitful area of
fluid dynamics research, due to its combination of a
simple geometry and complex, unsteady flow features.
Most of the work has concentrated on the two-dimen-
sional cross flow around cylinders of infinite length or
enclosed between end-plates. Above a Reynolds number
(Re), based on diameter d, of 50, a Ka´ rma´ n vortex
shedding pattern, in which vortices are shed alternately
from each side of the cylinder at a particular frequency,
can be observed. In many real-life situations, the cylin-
der is not infinitely long or enclosed, but instead having
at least one end over which the fluid can flow. Examples
of such structures range from large-scale buildings and
cooling towers, down to the trip studs used in experi-
mental models. Similar generic geometries can be found
on aircraft landing gear.
This paper presents the results of a series of experi-
ments on the flow over a cylinder of height/diameter
ratio (h/d) equal to 1. This represents the case of the
cylinder with a free-end, where the whole height of the
cylinder is subject to the downwash from the free-end.
The work was performed as part of a combined exper-
imental and numerical study (Pattenden 2004), the aim
being to investigate the physics of the flow in the sepa-
ration regions and to assess the validity of large-eddy
simulations of this flow. The experiments reported here
were intended to provide a sufficient set of data to val-
idate the computational fluid dynamics (CFD) code, as
well as providing valuable insight into the flow physics.
Various measurements were made to get a full picture
of the flow. These included surface flow visualisation,
particle image velocimetry (PIV) measurements in a
number of planes and unsteady surface pressure mea-
surements. This paper will first present the background
to the work, including a summary of previous work in
the field. The experimental methods will then be de-
scribed before the results are presented by studying each
part of the flow field separately.
2 Background
The amount of available work on low-aspect-ratio cyl-
inders is limited. Table 1 lists previous experiments on
low-aspect-ratio finite cylinder flows, along with the
types of measurements. A number of authors have
studied the effect of reducing the aspect ratio on vortex
shedding in the wake and on the drag coefficient. These
include Farivar (1981), who conducted experiments on
cylinders of 2.78<h/d<15, which identified three layers
of vortex shedding, with the shedding frequency in-
R. J. Pattenden Æ S. R. Turnock (&) Æ X. Zhang
School of Engineering Sciences,
University of Southampton, Southampton,
Highfield, SO17 1BJ, UK
E-mail: steve@ship.soton.ac.uk
Present address: R. J. Pattenden
School of Civil Engineering and the Environment,
University of Southampton, Southampton, UK
Experiments in Fluids (2005) 39: 10–21
DOI 10.1007/s00348-005-0949-9
creasing toward the two-dimensional value nearer the
base. Ayoub and Karemcheti (1982) measured surface
pressures and wake velocity fluctuations, which showed
a shedding region near the tip different to that over the
rest of the length. Slaouti and Gerrard (1981) looked at
the effect of varying the end conditions on the form of
the vortex shedding pattern on a high-aspect-ratio cyl-
inder. Okamoto and Yagita (1973) took surface pressure
measurements on cylinders with aspect ratios between 1
and 12.5, and found that the drag coefficient decreased
with decreasing aspect ratio. They also noted that the
vortex shedding pattern does not exist at h/d £ 6, as the
effects of the free-end reach the base. Sakamoto and Arie
(1983) investigated the effects of aspect ratio and
boundary layer thickness on the vortex shedding fre-
quency behind the cylinders. They found that the shed-
ding frequency is related to the aspect ratio by a power
law, and that the shedding changes from Ka´ rma´ n type
to symmetric arch type at an aspect ratio of 2.5.
Okamoto and Sunabashiri (1992) also observed the
change from asymmetric to symmetric shedding at an
aspect ratio of 4. Kawamura et al. (1984) tested cylinders
of 1<h/d<8, performing flow visualisation, hot-wire
measurements and pressure measurements. They at-
tributed the reduced drag coefficient, relative to an in-
finite cylinder, to the reduced separation velocity, due to
separation occurring earlier. This earlier separation is
attributed to the increased base pressure due to the
downwash from the free end. Uematsu et al. (1990)
studied the effect of transition on the flow around cyl-
inders of aspect ratios 1–4. The separation point was
found to move back slowly through the critical Reynolds
number range. Leder (2003) measured the wake behind a
cylinder of aspect ratio 2, with two different end shapes;
flat and hemispherical. This showed the change in vor-
ticity caused by the change in end shape. Sumner et al.
(2004) measured the wake behind cylinders of aspect
ratios 3, 5 and 9, showing the development of the tip and
base vortex structures in the wake. The shedding pattern
was found to be different at h/d=3.
While a number of investigators have studied the
nature of the wake region, few people have studied the
flow over the free-end. Kawamura et al. (1984) made
surface flow visualisations on the free-end, which
showed two swirl patterns in the forward half of the tip.
They attribute this to the reversed flow along the centre
being entrained by the obliquely separated flow from the
leading edge. They describe the flow structure on the
free-end as being similar to the ‘‘mushroom’’ vortex seen
by Winkelmann and Barlow (1980) on an aerofoil after
stall. This means that the swirls are the bases of a vortex
which arches over the free-end through the region of
separated flow. Roh and Park (2003) also visualised the
flow on the free-end. This also showed the two swirl
patterns on the forward half of the surface. Based on
their smoke visualisations, however, they suggest that
the swirls are the bases of streamwise vortices trailing
downstream.
3 Description of experiments
3.1 Wind tunnel and model
The experiments were performed in a 0.9·0.6 m open-
circuit wind tunnel, operated at a flow speed U
¥
of
Table 1 Summary of previous experiments on cylinders with a free-end mounted on a ground plane
Author(s) Year h/d Re·10
À4
Measurements
P F HW 7PP LDA PIV OFV SFV
Okamoto and
Yagita (1973)
1973 1 fi 12.5 1.3
4 4
Farivar (1981) 1981 2.78 fi 15 7.0
4 4
Sarode et al. (1981) 1981 1.1 fi 10 2.2
4 4
Taniguchi et al. (1981b) 1981 0.75 fi 5 2.1
4 4 4
Taniguchi et al. (1981a) 1981 1 5.0
4
Okamoto (1982) 1982 1 4.7
4 4 4
Sakamoto and Arie (1983) 1983 1 fi 8 0.027 fi 0.092
4 4
Kawamura et al. (1984) 1984 1 fi 8 3.2
4 4 4
Sin and So (1987) 1987 0.5 fi 2.5 4.8
4 4
Uematsu et al. (1990) 1990 1 fi 4 3.2 fi 15.1
4 4 4
Baban and So (1991a, b) 1991 1 fi 2 4.6
4 4
Okamoto and
Sunabashiri (1992)
1992 0.5 fi 24 2.5 fi 4.7
4 4 4 4
Luo et al. (1996) 1996 4 fi 8 3.3
4
Kappler (2002) 2000 2 fi 5 2.2 fi 5.4
4 4
Park and Lee (2000, 2002) 2000 6 fi 13 2.0
4 4 4
Leder (2003) 2003 2 20.0
4
Roh and Park (2003) 2003 1.25, 4.25 0.6,14.8
4 4
Sumner et al. (2004) 2004 3, 5, 9 6.0
4 4 4
Present work 2004 1 20.0
4 4 4
Key to measurement types: P surface pressures, F force measurements, HW hot-wire anemometry, 7PP seven-hole pressure probe, LDA
laser Doppler anemometry, PIV particle image velocimetry, OFV oil flow visualisation, SFV smoke flow visualisation
11
20 m/s. The cylinder model had an aspect ratio h/d=1,
with height and diameter of 150 mm (Fig. 1). These
dimensions give a Reynolds number based on diameter
of 1.96·10
5
, which is just below the critical value for
transition in the boundary layer on a circular cylinder
(Zdravkovich 1997). The x, y and z axes are in the
streamwise, transverse and spanwise directions, respec-
tively, while the respective velocity components are de-
noted as u, v and w.
The model was mounted on a ground plate, which
had an elliptic leading edge 0.6 m upstream of the axis of
the cylinder. The boundary layer on the ground plate
was allowed to develop naturally. Its profile was mea-
sured using a hot-wire without the model being present
at x/d=À0.5. This indicated a turbulent boundary layer
with a displacement thickness of 0.1d at 20 m/s. Table 2
shows further details of the boundary layer. The
boundary layer is defined by its thickness d, the dis-
placement thickness d
*
, the momentum thickness h, the
Reynolds number based on momentum thickness, Re
h
and the shape factor g=d
*
/h. The flow at the leading
edge of the ground plate was uniform across the span of
the tunnel to within 0.5%, while the turbulence intensity
was 0.3%.
It is necessary to consider the effects of the walls of
the tunnel on the flow around the body. This is assessed
by means of a blockage coefficient or, in the case of a
three-dimensional body, three ratios. These are the lat-
eral blockage ratio B
l
=d/W
T
, the vertical blockage ratio
B
v
=d/H
T
and the total blockage ratio B
t
=B
l
B
v
, where
W
T
and H
T
are the width and height of the wind tunnel
working section above the ground plate, respectively.
In this case, the ratios are B
l
=0.167, B
v
=0.275
and B
t
=0.042. For a bluff body, there is no simple
relationship describing the effect of blockage, but studies
have been carried out by West and Apelt (1982) on
cylinders spanning the tunnel, and by Farell et al. (1977)
on cooling tower models. According to the work of
Farell et al. (1977), the base pressure coefficient (C
Pb
) in
these measurements can be expected to be 35% higher
than in unconfined flow. The change in the pressure
gradient induced by a tunnel blockage can also have an
effect on the flow topology due to changes in the sepa-
ration and transition points.
3.2 Surface flow visualisation
Oil flow visualisation gives insight into the topology of
the flow. It shows the lines of separation and attach-
ment, as well as the bases of vortices on the surface. To
achieve this, all the surfaces of the ground plate and
model were covered with black self-adhesive PVC. A
mixture of titanium dioxide, paraffin and oleic acid was
then applied to all the relevant surfaces. The tunnel was
run for about 30 min so that the mixture dried com-
pletely. The flow pattern was then photographed using a
standard 35 mm camera, with the model in place in the
tunnel.
3.3 Particle image velocimetry
The PIV measurements were carried out using a Dantec
FlowMap system. A 120-mJ Nd:Yag dual-cavity laser
was mounted on the roof of the wind tunnel, above a
slot, to form a 1-mm-thick light sheet. This allowed
measurement in either the longitudinal, x–z, plane or the
transverse, y–z, plane. In the longitudinal planes, u and
w velocities were measured, while in the transverse
planes, v and w were measured. The 80C60 HiSense
CCD camera was mounted either at the side of the
tunnel, viewing through a glass window, for the longi-
tudinal planes, or inside the tunnel, viewing upstream,
for the transverse planes. Details of the image locations
and fields of view are given in Table 3. The seeding was
provided by a water-based fog generator, placed outside
the tunnel so that the fog would diffuse before entering
the tunnel. This avoided large concentrations of particles
entering the image plane.
The cross-correlation method is used to calculate the
velocities from the images. The images are divided into
interrogation areas of 32·32 pixels, with an overlap of
50%. The physical size of these areas are given in
Table 2 Details of the boundary layer on the ground plane at the
position of the leading edge of the model (x/d=À0.5, y/d=0)
U
¥
(m/s) d/d d
*
/d h/d Re
h
g
10 0.20 0.022 0.018 1,761 1.21
15 0.14 0.016 0.013 1,942 1.25
20 0.10 0.013 0.010 2,037 1.30
55mm
600mm
2400mm
1200mm
545mm 150mm
150mm
Cylinder model
Ground plate
(b) Side view of model in working section
Contraction
Working section
0.9m
0.6m
2.4m
Diffuser
4.0m
6.4m
(a) Arrangement of wind tunnel
x=0
U
Fig. 1 Diagram of wind tunnel and model arrangement (not to
scale)
12
Table 3, denoted as D
PIV
. In the longitudinal planes, the
dominant velocity is parallel to the plane and of the
same order of magnitude as the free-stream velocity. At
20 m/s, the time for a particle to cross the interrogation
area of the coarse-grid longitudinal planes is 365 ls.
This means that the time between consecutive frames
should be less than 90 ls, according to experience, which
suggests that the time should be less than 25% of the
time to cross the interrogation area. In practice, a time
of 60 ls was found to give the most reliable measure-
ments.
In the transverse planes, the dominant flow is nor-
mal to the plane, with the in-plane components being
relatively small compared to the free-stream velocity.
This can cause poor correlation if the particles are
crossing the light sheet too quickly, so the time between
images generally needs to be small when the flow is
normal to the plane. With a light sheet thickness of
1 mm, the particles would take 50 ls to cross the light
sheet, although in reality, the thickness of the light
sheet and its intensity distribution are not known ex-
actly. Also, the flow in the turbulent recirculation re-
gion varies greatly across the plane, so it is difficult to
determine what the time between frames should be. In
practice, a time of 15 ls was found to give the maxi-
mum number of valid correlations. The erroneous
vectors caused by poor correlation were removed using
a range validation, with the upper limit typically set to
25 m/s for the longitudinal planes. The discarded vec-
tors were replaced with interpolated values from the
surrounding points.
The accuracy of the velocity measurements is esti-
mated by assuming an accuracy in the correlation of
0.1-pixel displacement (Raffel et al. 1998). This corre-
sponds to a velocity accuracy of between 0.5% for the
closeup views and 6% for the transverse planes. There
is an additional statistical error due to the finite number
of samples. By looking at the mean after a certain
number of samples, it was found that, after 500 sam-
ples, the mean had converged to within 1% of the value
after 1,000 samples. In the longitudinal planes, the
deviations from the mean flow were smaller than in the
transverse planes, so only 500 samples were needed in
these planes.
3.4 Surface pressure measurements
The pressure on the surface of the model was measured
using a set of five pressure transducers set in the sides
of the cylinder. The sensors used were piezoresistive
gauge pressure transducers (Endevco Model 8507C-2)
with a 2-mm-diameter silicon diaphragm and a range
of 13,790 N/m
2
. The five transducers were equally
spaced along a generator of the cylinder (z/d=0.17 to
0.83 inclusive) and the cylinder was rotated on the
dynamometer base through 360° at 5° intervals, so that
a full picture of the surface pressures could be gath-
ered.
It was found to be necessary to calibrate the
transducers in the model, even when fixed using a
flexible adhesive to minimise the radial loading on the
barrel of the transducer. This was achieved by mea-
suring the surface pressures at two angles using pres-
sure tappings in the transducer holes which were
connected to a Furness FC012 digital micromanometer
with a known calibration. The frequency response is
flat up to 20% of the resonant frequency, 70,000 Hz,
which covers the range of interest. The total uncer-
tainty to a 95% confidence level in the pressure mea-
surements is ±1.4%.
4 Results and discussion
4.1 Overview
The flow considered here is complex, with a number of
constituent features, as shown in the schematic diagram
in Fig. 2. The dominant features of the time-averaged
flow are the horseshoe vortex, which is formed on the
ground when the upstream flow separates due to the
Table 3 Details of PIV measurement planes
Plane x/d y/d z/d D
PIV
(mm) Samples U
¥
(m/s)
L1 À1.45 fi 0.50 0 0 fi 1.55 7.3 1,000 20
L2 À0.71 fi 1.24 0 0 fi 1.55 7.3 1,000 20
L3 0.66 fi 2.60 0 0 fi 1.55 7.3 1,000 20
L4 0.46 fi 1.01 0 0 fi 0.44 2.1 500 10,20
L5 À0.60 fi 0.57 0 0.62 fi 1.56 4.4 500 10,20
L6 À1.28 fi À0.34 0 0 fi 0.75 3.5 500 10,15,20
L7 À1.03 fi À0.48 0 0 fi 0.44 2.1 500 10,15,20
T0 0.50 À0.75 fi 0.75 0 fi 1.20 5.6 1,000 20
T1 0.67 À0.75 fi 0.75 0 fi 1.20 5.6 1,000 20
T2 0.83 À0.75 fi 0.75 0 fi 1.20 5.6 1,000 20
T3 1.00 À0.75 fi 0.75 0 fi 1.20 5.6 1,000 20
T4 1.50 À0.75 fi 0.75 0 fi 1.20 5.6 1,000 20
T5 2.50 À0.75 fi 0.75 0 fi 1.20 5.6 1,000 20
13
adverse pressure gradient; the vortex system on the free-
end, inside the separated flow; the arch vortex in the rear
recirculation region; and the trailing vortices down-
stream of the reattachment.
4.2 Pressures on the cylinder
Figure 3 shows the mean surface pressure coefficient C
P
,
around the circumference of the cylinder at different
heights above the floor, at U
¥
=20 m/s. The angle u is
measured from the upstream face of the cylinder. The
pressures are seen to be constant over most of the span
of the cylinder, except near the top and bottom. At
u=0°, the pressure coefficient near the top is reduced to
0.9, due to the upwash over the tip. At separation, the
flow is again constant over the height of the cylinder,
except near the tip, where separation is delayed by 5°.
This is consistent with the flow visualisation (Fig. 5)
which shows the separation line (S
S
) sweeping back to-
wards the tip. At the back of the cylinder, the pressure
coefficient rises near the base from À0.57 at z/d=0.5 to
À0.45 at z/d=0.17. This is caused by the reversed flow
impinging on the surface of the cylinder.
These pressures agree closely with those measured by
Okamoto and Sunabashiri (1992) and Kawamura et al.
(1984), despite the higher Reynolds number as shown in
Fig. 3. The rise from the minimum is slightly quicker
than the others, suggesting that separation occurs a few
degrees earlier than in the other experiments. The local
drag coefficient found by integrating the pressure dis-
tribution over the surface at z/d=0.5 (Eq. 1) is 0.79,
compared to 0.7 for Okamoto’s and 0.75 for Kawam-
ura’s data:
C
D
¼
Z
p
0
C
P
cos /d/ ð1Þ
The plot of power spectral density function E(f) in
Fig. 4 shows a peak in the energy spectrum of the
pressure signal at 12 Hz, which corresponds to a
Strouhal number, fd/U
¥
, of 0.09.
4.3 Horseshoe vortex
The signature of the horseshoe vortex on the ground can
be seen in the flow visualisation picture in Fig. 5. The
flow features in the oil flow visualisation are interpreted
according to Tobak and Peake (1982) and Perry and
Chong (2000). The primary separation point (S
1
) is at
0 50 100 150
5
0
0.5
1
φ (degrees)
C
P
z/d=0.83
z/d=0.67
z/d=0.50
z/d=0.33
z/d=0.17
0 50 100 150
5
0
0.5
1
φ (degrees)
C
P
Current work
Okamoto (1992)
Kawamura (1984)
(a)
(b)
Fig. 3a, b Pressure distributions around the cylinder. a Distribu-
tion of pressure coefficient around cylinder at different heights, at
U
¥
=20 m/s. b Distribution of pressure coefficient around cylinder
compared to other authors
0 0.1 0.2 0.3 0.4 0.5
0
0.02
0.04
fd/U

E
(
f
)
Fig. 4 Power spectral density function of pressure measurements at
z/h=0.5, u=180°, U
¥
=20 m/s
Separation
Horseshoe vortex
Arch vortex
z
x
y
d
U

Tip vortex
Trailing vortex
Fig. 2 Schematic diagram of the time-averaged flow over a
truncated cylinder (not to scale)
14
x/d=À1. Line C is the line of converging streamlines at
the upstream edge of the primary vortex. This line has a
thickness of about 0.04d, which is likely to be the width
of the vortex 1’ in Fig. 6. This means that the separation
point S
2
is at the downstream edge of this line, while the
attachment point A
1
is at the upstream edge. The
streamlines on the inside of this line are created by the
outward flow under the primary vortex and are swept
back by the spiral motion of the vortex. There is another
line of diverging streamlines (D) inside this line,
extending from the leading edge of the cylinder. This is
the inner edge of the horseshoe, which moves away from
the cylinder as it moves downstream. The inward facing
streamlines inside this line are due to a small vortex
which is formed in the corner, and rotates in the oppo-
site sense to the primary vortex. Figure 5 shows a view
of the front of the cylinder, illustrating the upwash near
the free-end and the downwash near the ground.
Figure 7 shows the time-averaged streamlines on the
symmetry plane in front of the cylinder, which indicate a
dominant vortex. Its centre is at x/d=À0.67, z/d=0.04.
Extrapolation of a dividing streamline to the ground
suggests that there is a separation point on the ground
plane in front of the horseshoe vortex at x/d=À0.76
when U
¥
=20 m/s. This corresponds to the position of
line C, under vortex 1’, in the oil flow visualisation
(Fig. 5), and so, there should be a reattachment at this
point. Indeed, on closer inspection of the vector plots
(Fig. 7), it can be seen that the vectors closest to the
ground are actually pointing downwards, indicating that
this point is a reattachment point (A
1
), as proposed by
S A S
1 1’
2
0
2 1 1
Fig. 6 Topology of horseshoe vortex system inferred from present
work and based on Baker (1980). The vortices are numbered 0, 1, 1’
and 2, with vortex 1 being the called the primary vortex and 2 the
secondary vortex. The separation and attachment points S
1
, S
2
and
A
1
are also marked. (Drawing not to scale)
Fig. 5a–d Surface flow
visualisation images,
U
¥
=20 m/s. a Floor of tunnel
(flow from left to right). b Top
of cylinder (flow from left to
right). c Side of cylinder (flow
from left to right). d Front of
cylinder (flow into page)
15
Baker (1980). The implication of this is that there are
recirculating vortices either side of this point that are not
visible in these measurements, due to the resolution of
the PIV system of 0.013d. The reflection of the light off
the ground makes it impractical to measure any closer to
the ground than this. The height of the saddle point
between these vortices is z/d=0.02 or 0.2d.
Figure 8 shows plots of the time-averaged streamwise
velocity, u, found by averaging the ensemble of instan-
taneous realisations through the centre of the vortex on
the symmetry plane. This shows that there is little dif-
ference in the mean position of the centre of the vortex,
x/d=À0.67, z/d=0.04, at the two different speeds.
While the discussion above relates to the time-aver-
aged flow, found by averaging all the PIV images, the
horseshoe vortex system actually displays considerable
unsteadiness. This can be seen in Fig. 9a, which shows
four consecutive realisations of the flow, 0.25 s apart. In
the first step, there is a large vortex close to the cylinder.
This appears to decay in the next instant when a vortex
forms upstream. This new vortex then translates down-
stream as the old vortex either decays or merges with the
new one. Finally, there is a single large vortex at the
downstream location and the cycle repeats. Figure 9
shows the positions of the centre of the main vortex in
each of the 500 samples. The location of the centre of the
vortex can be found by searching for the minimum
vorticity magnitude in the region of the horseshoe vortex
system. Although, in general, the maximum vorticity
does not necessarily occur at the centre of a vortex, in
this situation, it was found to be a reliable measure.
There is no clear evidence of the bi-modal behaviour as
reported by Devenport and Simpson (1990). This could
be due to the lower Re
h
in the present work, or the fact
that their model’s nose was elliptical rather than round.
The higher Reynolds number may increase the insta-
bility which causes this behaviour.
4.4 Free-end flow
The flow over the free-end is characterised by the sepa-
ration from the sharp leading edge. This forms a com-
plex three-dimensional recirculation region on top of the
cylinder. Figure 5 shows the flow on the surface of the
free-end of the cylinder. The most prominent features
are the two foci (F
T
) at x/dÀ0.25. There is a saddle
point of separation (S
T
) between them at x/dÀ0.4. This
is where the reversed flow along the centreline of the tip
separates from the surface to form a small vortex near
the leading edge, and where the reversed flow turns
outwards towards the sides. Flow also enters the foci
from the sides behind the separation line. There is a
saddle point of reattachment (R
T
) where the flow reat-
taches to the free-end.
The PIV velocity vectors of the mean flow in the
streamwise symmetry plane on top of the cylinder
(Fig. 10) show the region of recirculating flow. The
9 8 7 6 5
0
0.05
0.1
0.15
0.2
x/d
z
/
d
8 7
0
0.02
0.04
x/d
z
/
d
U/U

=0.1
(a)
(b)
Fig. 7 a Time-averaged streamlines on centreline (y/d=0) up-
stream of cylinder, showing the horseshoe vortex. b An exploded
view of the area marked by the dashed rectangle in (a), showing the
attachment point with time-averaged vectors of velocity
(U
¥
=20 m/s)
9 8 7 6 5
1
0
0.1
0.2
0.3
0.4
0.5
x/d
U
/
U

U

=10 m/s
U

=20 m/s
0 0.1 0.2 0.3 0.4 0.5
4
2
0
0.2
0.4
0.6
z/d
U
/
U

U

=10 m/s
U

=20 m/s
(a)
(b)
Fig. 8a, b Normalised time-averaged streamwise velocity through
the centre of horseshoe vortex. a z/d=0.04. b x/d=À0.66
16
mean flow shows a vortex structure on the centreline,
formed by the reversed flow along the tip. The location
of the reattachment point, R
T
, at x/d=0.17 can be seen.
There is an uncertainty of ±0.01d in this measurement,
due to the resolution of the PIV. Looking at the
instantaneous vorticity plots in Fig. 10, the shear layer
over the top of the cylinder is visible. This remains fairly
steady for the first 0.2d, but quickly becomes fully tur-
bulent, forming the coherent structures visible down-
stream of x/d=À0.3. Figure 11 shows the flow
immediately behind the trailing edge of the cylinder. The
tip vortices formed by the flow up over the edge of the
free-end are visible at the corners of the free-end.
One recent paper on the subject of the free-end flowby
Roh and Park (2003) proposed a flow topology consist-
ing of two pairs of counter-rotating streamwise vortices,
with the inner pair being rooted on the swirl patterns on
the tip. These measurements were carried out at Rey-
nolds numbers of 5.92·10
3
and 1.48·10
5
, with aspect
ratios of 1.25 and 4.25; the high Re, low aspect ratio case
should correspond quite closely to the present study.
Their light-sheet visualisation appears to show these
vortices inside the two tip vortices and trailing down-
stream of the cylinder. However, there is no evidence of
this inner pair of vortices in the current results. While the
flow topology on the surface of the free-end agrees well
with the present findings, the flow topology on the
symmetry plane does not. The PIV measurements of the
flowon the symmetry plane show that the flowreattaches
at the saddle point R
T
, whereas Roh and Park show a
separation here with no reattachment on the free-end.
The PIV on the symmetry plane shows a vortex core in
the recirculation region, which must be attached to a
surface. The most likely explanation is that this vortex is
attached to the free-end at the two swirl patterns seen in
the oil flow visualisation. This would agree with the
hypothesis of Kawamura et al. (1984), that a mushroom
vortex exists here. If the smoke visualisation of Roh and
Park is showing this vortex, the core would have to cross
5 0 0.5
1
1.2
x/d
z
/
d
U/U

=1
5 0 0.5
1
1.2
x/d
z
/
d
0.5 0 0.5
1
1.2
x/d
z
/
d
(a)
(b)
Fig. 10a, b The flow in the symmetry plane (y/d=0) on top of the
cylinder showing time-averaged velocity vectors and two instanta-
neous vorticity plots (U
¥
=20 m/s). a Velocity vectors of the
time-averaged flow. b Vorticity contours at two instantaneous
realisations of the flow
0
0.05
0.1
0.15
0.2
z
/
d
0
0.05
0.1
0.15
0.2
z
/
d
0
0.05
0.1
0.15
0.2
z
/
d
9 8 7 6 5
0
0.05
0.1
0.15
0.2
z
/
d
x/d
8 7 6 5
0
0.05
0.1
x/d
z
/
d
(a)
(b)
Fig. 9a, b Unsteadiness in the horseshoe vortex system at
U
¥
=20 m/s. a Sequence of images showing motion of horseshoe
vortex system. b Scattering of horseshoe vortex centre on y/d=0 in
each of 500 PIV images
17
the symmetry plane downstream of the cylinder, which
would imply a much larger separation bubble.
4.5 Separation and flow structure of the wake
The flow separates from the cylinder surface at an angle
of approximately 70° from the leading edge. This can be
seen from the flow visualisation in Fig. 5, which shows a
side view of the cylinder. The oil flow is attached on the
forward part of the cylinder before forming a line (S
S
),
where the flow separates. This line bends backwards at
the top and bottom of the cylinder as the separation is
delayed by about 5°, due to the oblique flow over the
end. At the sub-critical Reynolds number of 1.96·10
5
, at
which these experiments were conducted, the boundary
layer over the surface of the cylinder is laminar, with
transition presumably occurring in the shear layer
immediately after separation. This corresponds to the
transition in shear layers regime, as reported by Zdrav-
kovich (1997) for the infinite cylinder case at Reynolds
numbers up to 2·10
5
. On an infinite cylinder, separation
would be expected at 80° in this regime; the difference
being attributed to the higher pressure behind the cyl-
inder due to the flow over the free-end entering the re-
circulation region (Okamoto and Yagita 1973).
To see the effect of transition on the separation point,
the boundary layer was tripped to turbulence using a
0.5-mm wire placed on the forward surface of the
cylinder. The surface flow was visualised in this condi-
tion, and it was seen that separation is delayed until 80°
0
0.2
0.4
0.6
0.8
1
x/d=0.50
z
/
d
x/d=0.67
0
0.2
0.4
0.6
0.8
1
x/d=0.83
z
/
d
x/d=1.0
5 0 0.5
0
0.2
0.4
0.6
0.8
1
x/d=1.5
z
/
d
y/d
0.5 0 0.5
x/d=2.5
y/d U/U

=1
Fig. 12 Streamwise evolution
of the flow in the y–z plane,
showing time-averaged velocity
vectors viewed from
downstream, U
¥
=20 m/s
6 4 2 0 0.2 0.4 0.6
0.8
1
1.2
y/d
z
/
d
V/U

=1
Fig. 11 Time-averaged velocity vectors at x/d=0.5 (U
¥
=20 m/s)
18
from the leading edge. This is further forward than the
100° angle that would be expected for an infinite cylinder
in the turbulent boundary layer regime. The later sepa-
ration point appeared to make little difference to the
topology of the flow, and so, other measurements were
not repeated in this condition.
Downstream of separation, the detached shear layer
around the cylinder encloses a recirculation region. This
extends to x/d=1.6, where the flow descending from the
free-end attaches to the ground plane, marked by R
F
in
Fig. 5. For comparison, Okamoto and Sunabashiri
(1992) found a value of x/d=2.9 for this reattachment
position, discovered by searching for the line of zero
dynamic pressure using a Pitot tube. The more recent
work of Leder (2003) gives x/d=2.2 for an aspect ratio
of 2, using LDA, and Sumner et al. (2004) give x/d=3
for an aspect ratio of 3, found using a seven-hole
pressure probe. Okamoto and Sunabashiri did find that
the recirculation length reduced with aspect ratio, so it
seems that the present work is in line with Leder and
Sumner et al., but not with Okamoto and Sunabashiri.
This may be due to the different measurement technique
of Okamoto and Sunabashiri.
Between the reattachment point and the cylinder,
there is a region of backflow along the ground, which
wraps into the arch vortex. The bases of this vortex can
be seen in the flow visualisation in Fig. 5, marked as F
F
.
The shear layers at the sides and over the top interact
and form a pair of tip vortices aligned in the streamwise
direction. The development of these vortices can be seen
in the PIV measurement planes in Figs. 12 and 13.
At the first plane (x/d=0.5), two counter-rotating tip
vortices can be seen at the free-end of the cylinder. They
are most clearly shown by the area of high vorticity, ra-
ther than the velocity vectors, as these are influenced by
the global flow around the back of the cylinder. These
vortices remain at a similar location in the plane until x/
d=1.0, where they start to dilate and descend towards
0
0.2
0.4
0.6
0.8
1
x/d=0.50
z
/
d
x/d=0.67
0
0.2
0.4
0.6
0.8
1
x/d=0.83
z
/
d
x/d=1.0
5 0 0.5
0
0.2
0.4
0.6
0.8
1
x/d=1.5
z
/
d
y/d
0.5 0 0.5
x/d=2.5
y/d
0 2
Fig. 13 Streamwise evolution
of the flow in the y–z plane,
showing time-averaged
contours of vorticity magnitude
viewed from downstream,
U
¥
=20 m/s. (Vorticity
magnitude normalised by
velocity and diameter)
19
the ground plane. They appear to be pushed down by the
downwash behind the cylinder. Moving downstream, the
tip vortices expand as they descend to the ground and the
vorticity merges with that produced by the converging
flow outside the separation bubble.
At x/d=0.5, two areas of high vorticity are observed
either side at the base of the cylinder. These are created
by the reversed flow impinging on the curved rear face of
the cylinder. This is also visible on the surface flow vi-
sualisation in Fig. 5, where there are noticeable swirls on
the surface marked F
B
. They disappear downstream
when the downwash becomes dominant. The main
horseshoe vortex is likely to be outside the extents of
these planes. As has been seen from the surface flow
visualisation, the horseshoe vortex extends to y/d=±1
at x/d=0.5. The inside edge of the horseshoe is dis-
cernible from the velocity vectors at the sides of the plot,
where the vectors at the outside lower corner are
pointing downwards and outwards.
Further downstream beyond x/d=1.5, the flow
turning in from the sides of the cylinder combined
with the downwash from the top causes the flow to
converge on the centreline, where it impinges on the
ground plane. At this point, the flow on the ground is
forced outwards, giving rise to vorticity, which devel-
ops downstream to form the two trailing vortices, as
seen at x/d=2.5.
The time-averaged form of the wake within the re-
circulation bubble has been shown by Okamoto (1982) to
be an arch vortex, which is attached to the ground behind
the cylinder. It has been suggested that this flow structure
is shed at a certain Strouhal number in a symmetrical
manner, as opposed to the antisymmetric shedding
experienced behind a higher aspect ratio cylinder. In fact,
Okamoto and Sunabashiri (1992) found that the peak in
the frequency spectrumwas very broad for the lowaspect
ratio cylinder, suggesting that the shedding was not
strongly coherent. Indeed, instantaneous realisations of
the flow field, taken at an interval of 12.5 s, as shown in
Fig. 14, show an irregular pattern of vortices, with no
discernible pattern to the flow, which is consistent with
the turbulent nature of the flow rather than a regular
vortex shedding pattern. This is the same both close to
the cylinder at x/d=0.5 and further downstream at x/
d=2.5. At the downstream location, there is some
asymmetry in the instantaneous flow field, which may
indicate some large-scale oscillation. The difference be-
tween the flow at any instant and the mean flow raises
important questions as to the validity of applying time-
averaged modelling methods to this class of flows.
5 Conclusions
This series of experiments has provided a large amount
of data on the flow over a low-aspect-ratio cylinder.
Surface flow visualisation has been conducted on all
surfaces, showing clearly the footprints of the vortex
structures found in the mean flow. Measurements have
been made of the velocity field in a number of planes
using particle image velocimetry (PIV), which shows the
mean and instantaneous flow structures. Pressure mea-
0
0.2
0.4
0.6
0.8
1
z
/
d
x/d=0.5 x/d=0.5
5 0 0.5
0
0.2
0.4
0.6
0.8
1
x/d=2.5
0.5 0 0.5
x/d=2.5
0 10
Fig. 14 Instantaneous
realisations of vorticity
magnitude at x/d=0.5 and
x/d=2.5 viewed from
downstream. U
¥
=20 m/s. Left
and right images are separated
by a time interval of 12.5 s.
Vorticity magnitude normalised
by velocity and diameter
20
surements have also been made. The principal features
found are summarised below.
A horseshoe vortex is formed at the junction of the
cylinder with the ground plane. The secondary separa-
tion point, S
2
, seen in both the PIV data and flow vi-
sualisation is at x/d=À0.76, while the centre of the
vortex is at x/d=À0.67, z/d=0.04. The primary sepa-
ration point, S
1
, in the flow visualisation is at x/d=À1.
The pair of vortices required to fit this pattern appear to
be too close to the wall to resolve with the PIV, although
the saddle point at z/d=0.02 is visible. These findings
support Baker’s hypothesis of the three-vortex structure,
but with very thin secondary vortices.
The instantaneous images of the flow reveal an
oscillatory pattern of vortex creation and merging, with
a vortex formed at an upstream location translating
downstream while growing, and then subsiding or
merging with the new upstream vortex.
The flow over the free-end separates from the leading
edge, and then attaches on the top of the cylinder at x/
d=0.17. Two swirls are visible on the surface of the free-
end, which appear to be the bases of vortices. It appears
that these are the bases of an arch vortex that passes
through the centre of the separation bubble. The present
findings appear not to support the hypothesis of Roh
and Park that there are two trailing vortices emanating
from these swirls.
The transverse PIV planes show longitudinal vortices
originating from the sides of the free-end and at the base
after separation. Moving downstream, the vorticity in
the shear layers forms two large counter-rotating vorti-
ces, which trail far downstream.
Acknowledgements The work described in this paper was sup-
ported by a PhD studentship from the School of Engineering Sci-
ences, University of Southampton, United Kingdom.
References
Ayoub A, Karemcheti K (1982) An experiment on the flow past a
finite circular cylinder at high subcritical and supercritical
Reynolds numbers. J Fluid Mech 118:1–26
Baban M, So R (1991a) Aspect ratio effect on flow-induced
forces on circular cylinders in a cross-flow. Exp Fluids
10:313–321
Baban M, So R (1991b) Recirculating flow behind and unsteady
forces on finite-span circular cylinders in a cross-flow. J Fluid
Struct 5:185–206
Baker C (1980) The turbulent horseshoe vortex. J Wind Eng Ind
Aerod 6:9–23
Devenport W, Simpson R (1990) Time-dependent and time-aver-
aged turbulence structure near the nose of a wing-body junc-
tion. J Fluid Mech 210:23–55
Farell C, Carrasquel S, Guven O, Patel V (1977) Effect of wind-
tunnel walls on the flow past circular cylinders and cooling
tower models. J Fluid Eng T ASME 99:470–479
Farivar D (1981) Turbulent uniform flow around cylinders of finite
length. AIAA J 19:275–281
Kappler M (2002) Experimentelle Untersuchung der Umstro¨ mung
von Kreiszylinder mit ausgepra¨ gt dreidimensionalen Effekten.
PhD thesis, Institute for Hydromechanics, University of Kar-
lsruhe, Germany
Kawamura T, Hiwada M, Hibino T, Mabuchi I, Kumuda M
(1984) Cylinder height greater than turbulent boundary layer
thickness—flow around a finite circular cylinder on a flat plate.
B JSME 27(232):2142–2151
Leder A (2003) 3D-flow structures behind truncated circular cyl-
inders. In: Proceedings of the 4th ASME/JSME joint fluids
engineering conference, Honolulu, Hawaii, July 2003
Luo S, Gan T, Chew Y (1996) Uniform flow past one (or two in
tandem) finite length circular cylinder(s). J Wind Eng Ind
Aerod 59(1):69–93
Okamoto S (1982) Turbulent shear flow behind hemisphere-cylin-
der placed on a ground plane. In: Turbulent shear flows, vol 3.
Springer, Berlin Heidelberg New York, pp 171–185
Okamoto S, Sunabashiri Y (1992) Vortex shedding from a circular
cylinder of finite length placed on a ground plane. J Fluid Eng T
ASME 114:512–521
Okamoto T, Yagita M (1973) The experimental investigation on
the flow past a circular cylinder of finite length. B JSME
16:805–814
Park C-W, Lee S-J (2000) Free end effects on the near wake flow
structure behind a finite circular cylinder. J Wind Eng Ind
Aerod 88:231–246
Park C-W, Lee S-J (2002) Flow structure around a finite circular
cylinder embedded in various atmospheric boundary layers.
Fluid Dyn Res 30:197–215
Pattenden R (2004) An investigation of the flow around a truncated
cylinder. PhD thesis, University of Southampton, UK
Perry A, Chong M (2000) Interpretation of flow visualisation. In:
Smits A, Lim T (eds) Flow visualisation: techniques and
examples. Imperial College Press, London
Raffel M, Willert C, Kompenhans J (1998) Particle image veloci-
metry: a practical guide (experimental Fluid Mechanics).
Springer, Berlin Heidelberg New York
Roh S, Park S (2003) Vortical flowover the free end surface of a finite
circular cylinder mounted on a flat plate. Exp Fluids 34:63–67
Sakamoto H, Arie M (1983) Vortex shedding from a rectangular
prism and a circular cylinder placed vertically in a turbulent
boundary layer. J Fluid Mech 126:147–165
Sarode R, Gai S, Ramesh C (1981) Flow around circular- and
square-section models of finite height in a turbulent shear flow.
J Wind Eng Ind Aerod 8:223
Sin V, So R (1987) Local force measurements on finite-span cyl-
inders in a cross-flow. J Fluid Eng T ASME 109(2):136–143
Slaouti A, Gerrard J (1981) An experimental investigation of the
end effects on the wake of a circular cylinder towed through
water at low Reynolds numbers. J Fluid Mech 112:297–314
Sumner D, Heseltine J, DansereauO(2004) Wake structure of a finite
circular cylinder of small aspect ratio. Exp Fluids 37:720–730
Taniguchi S, Sakamoto H, Arie M (1981a) Flow around a circular
cylinder vertically mounted in a turbulent boundary layer.
B JSME 24(193):1130–1136
Taniguchi S, Sakamoto H, Arie M (1981b) Flow around circular
cylinders of finite height placed vertically in turbulent boundary
layers. B JSME 24(187):37–44
Tobak M, Peake D (1982) Topology of 3-dimensional separated
flows. Annu Rev Fluid Mech 14:61–85
Uematsu Y, Yamada M, Ishii K (1990) Some effects of free-stream
turbulence on the flow past a cantilevered circular cylinder.
J Wind Eng Ind Aerod 33(1–2):43–52
West G, Apelt C (1982) The effects of tunnel blockage and aspect
ratio on the mean flow past a circular cylinder with Reynolds
numbers between 10
4
and 10
5
. J Fluid Mech 114:361–377
Winkelmann A, Barlow J (1980) Flowfield model for a rectangular
planform wing beyond stall. AIAA J 18:1006–1008
Zdravkovich M (1997) Flow around circular cylinders, vol 1: fun-
damentals. Oxford University Press, New York
21