Experiments in Fluids 29 (2000) 553±563 Ó Springer-Verlag 2000
Experimental study of flow past a square cylinder at high Reynolds numbers
A. K. Saha, K. Muralidhar, G. Biswas
Abstract Measurements of two-components of velocity in
the wake of a square cylinder using a hot-wire anemometer
are reported. Two Reynolds numbers, namely 8700 and
17,625, have been considered. The measurements were
carried out in a low-speed, low-turbulence wind tunnel.
Benchmark experiments at much lower Reynolds numbers
show good agreement between the present experiments
and published results. At higher Reynolds numbers, the
experimental data reveal anticipated trends in terms of
wake recovery and turbulence decay. Both velocity and
velocity ¯uctuations show symmetry about the wake axis.
The experimental data have been compared with the large
eddy simulation (LES) calculation reported by Wang et al.
[University of Illinois at Urbana ± Champaign (1996) Re-
port CFD 96-03] and LDV measurements of Lyn et al.
[J Fluid Mech (1995) 304: 285±319]. The agreement among
the three sets is generally acceptable in terms of the time-
averaged velocity components, but not the velocity ¯uc-
tuations. The turbulence ¯uctuations in the present ex-
periments are seen to be lower than in the referred work.
The differences have been traced to factors such as the
aspect ratio, blockage ratio and upstream turbulence.
Experiments with increased upstream turbulence did show
a reduction in the discrepancy between the present ex-
periments and the published data. An assessment of the
experimental data in terms of physical mechanisms re-
vealed that (a) streamwise normal stresses were correlated
with the vortex centers, and (b) the turbulence kinetic
energy pro®les are similar to the turbulence shear stress.
Spectral analysis of the velocity signals was carried out in
the present work. Energy transfer from the mean ¯ow to
the streamwise velocity ¯uctuation was con®rmed in the
near wake. A redistribution of the kinetic energy between
the streamwise and transverse components of velocity over
a longer distance downstream was subsequently observed.
List of symbols
B width of the square cylinder, m
2 observed objects through signature analysis and the
CD drag coef®cient, D/(0.5 q uav B)
Received: 17 May 1999/Accepted: 29 December 1999
A. K. Saha, K. Muralidhar, G. Biswas (&)
Department of Mechanical Engineering
Indian Institute of Technology
Kanpur, UP ± 208 016, India
This investigation has been supported by the Aeronautics R & D higher Reynolds numbers. These must be computed using
Board, New Delhi, grant number AERO/RD-134/100/10/98-99/1032. the unsteady three-dimensional Navier±Stokes equations
553
D drag per unit length of the square cylinder
f frequency of vortex shedding, Hz
k dimensionless turbulence kinetic energy
Re Reynolds number, uavB/l
S Strouhal number, f B/uav
u, v streamwise and transverse components of velocity
(dimensionless)
uav average incoming velocity, also used as a velocity
scale
; v
u time-averaged streamwise and transverse
components of velocity
u¢, v¢ streamwise and transverse components of velocity
¯uctuations
u02 †1=2 time average of streamwise ¯uctuation (rms)
v02 †1=2 time average of transverse ¯uctuation (rms)
u0 v0 † time-averaged shear stress
x, y streamwise and transverse coordinates measured
from the centre of the cylinder (dimensionless)
Greek symbols
l dynamic viscosity, kg/m-s
m kinematic viscosity, m2/s
q density, kg/m3
x dimensionless spanwise vorticity
1
Introduction
Motion of non-aerodynamic objects through ¯uids is
encountered extensively in engineering. Submarines,
ships, passenger aircraft and cars are examples where
the object is in motion through a stationary ¯uid me-
dium. High-rise buildings, chimneys and tube banks in
heat exchangers are examples where the ¯uid is in
motion. In each class, the wake of the bluff object
predominantly determines the distribution of forces on
it as well as ¯ow-induced vibration and heat transfer
rates. Recent applications such as the detection of un-
dispersion of pollution have further intensi®ed the need
for understanding wake dynamics. The present study is
concerned with high Reynolds number ¯ow past a
square cylinder.
Flow past bluff bodies has been simulated numerically
by different researchers. There exist two as well as three-
dimensional computations (Mukhopadhyay et al. 1992;
Tamura et al. 1990) on the ¯ow past a square cylinder.
Two-dimensional calculations are not appropriate at
 on ®ne grids. Engineering applications have, however, been
computed at high Reynolds number in two-dimensions
using various eddy viscosity models of turbulence. Owing
to certain drawbacks associated with the eddy viscosity
models, the bluff body ¯ows are presently being computed
using large eddy simulation (LES) (Wang et al. 1996).
Okajima (1982) carried out an experimental study of
¯ow past a square as well as a rectangular cylinder for a
wide range of Reynolds numbers, namely 70±(2 ´ 104).
This study established the variation in Strouhal number,
that is the non-dimensional vortex-shedding frequency,
554
with Reynolds number. These experiments have shown
that there is an abrupt change in Strouhal number when
the aspect ratio of the cylinder is reduced to the range 2±3.
This is re¯ected as a corresponding change in the wake
¯ow pattern. DuraÄo et al. (1988) have conducted laser
Doppler velocimetry (LDV) measurements for ¯ow past a
square cylinder in a water tunnel at a Reynolds number of
14,000. They have separated the periodic and random
components of velocity ¯uctuations. These measurements
show that the kinetic energy associated with the random
components is about 40% of the total. Lyn et al. (1995)
have reported an LDV study of turbulent ¯ow past a
square cylinder with emphasis on the ensemble-averaged
characteristics of the ¯ow behaviour. The Reynolds num-
ber considered in their study was 21,400. The experiments
were carried out in a closed and constant head water
tunnel. Their results showed a relationship between the
¯ow topology and the turbulence distribution. Vorticity
saddles and streamlines saddles were clearly distinguished.
A distinction was identi®ed between the ¯ow in the base
region and the near wake. Differences in the length and
velocity scales, and celerity of the vortices of ¯ows around
a circular cylinder and a square cylinder have also been
compared.
In the present investigation, an experimental study of
¯ow in the near wake of a square cylinder at Reynolds
numbers of 8700 and 17,625 was carried out using a hot-
wire anemometer. Measurements were conducted along
the mid-plane of the cylinder and across the wake. Both
streamwise and transverse components of velocity were
measured. Time±mean velocity as well as ¯uctuations were
recorded in the experiments. The scope of the present
work is to map the near wake of the cylinder and compare
it with the numerical results of Wang et al. (1996) and the
experiments of Lyn et al. (1995).
The present experiments were carried out in an open
circuit wind tunnel with a nominal cross-section of
40 ´ 40 cm. A 3-m-long test section of the wind tunnel was
available for experiments. Since the divergence angle of the
test section was 4°, a fair constancy of pressure was
observed. The wind tunnel fan was electronically controlled
and stable velocities in the range 5±20 m/s were realizable in
the experiments. The speed control was formally rated at
‹0.5% but over a 3-h period, no noticeable change was ob-
served in the approach velocity. Free stream turbulence in
the approach ¯ow was damped by an array of mesh screens.
In all the experiments, turbulence was found to be less than
the background noise of the anemometer, being equivalent
to 0.01% at an approach velocity of 10 m/s. The inlet velocity
pro®le was uniform to within ‹0.5% over 90% of the tunnel
cross-section. The square cylinder used was made of
Perspex, 25 ´ 25 mm in cross-section. The use of Perspex
was advantageous since (1) good quality smooth surfaces
were attainable over the entire cylinder, and (2) precise
sharpness of the cylinder edges could be maintained. The
cylinder length was 40 cm, thus spanning the entire width of
the wind tunnel. Upstream velocities of 5.2 and 10.53 m/s
were utilized to produce Reynolds numbers of 8700 and
17,625, respectively, in the experiments. The room tem-
perature during the all experiments was 26(‹2) °C. A few
selected low Reynolds number experiments were carried out
for benchmarking the instrumentation in a smaller but
similar facility. Here, lower velocities and a smaller cylinder
could be employed.
Velocity was measured using a two-channel hot-wire
anemometer along with an X-wire probe. The X was
formed in the vertical plane, with the cylinder placed in a
horizontal position. The probe was mounted on a tra-
versing mechanism that facilitated all three orthogonal
movements, to a positional accuracy in the most signi®-
cant direction, namely the vertical at ‹0.1 mm. The
commercially available Dantec anemometer and probes
were employed in the present work. The two wires of the
probe were calibrated in the wind tunnel itself. Small
changes in room temperature (‹1 °C) were compensated
through the use of a correction formula that assumes a
constant heat transfer coef®cient. The probe was recali-
brated for larger changes in temperature. Both wires were
operated at 200 °C, and their calibration curve were seen
to be almost identical. The assumption of equal sensitivity
coef®cients of the two wires was occasionally employed
during data reduction. The calibration curves were
smoothed using a ®fth-order polynomial. During calibra-
tion, a pitot tube connected to a Furnes Controls
19.99 mm of H2O digital manometer was utilized. Both DC
and rms values of voltages were recorded using ``true''
voltmeters supplied by Dantec. Integration times of typi-
cally 300 s were used to obtain all time-averaged quanti-
ties. The cross-correlation between the voltage ¯uctuations
was determined by collecting long time traces through an
Advantest spectrum analyser, followed by a numerical
integration of the product of the two signals (over 10 s).
The frequency of vortex shedding was measured by
the spectrum analyser using the FFT algorithm. The
time-averaged drag acting on the cylinder was deter-
mined by the wake survey method. Momentum loss
calculation was carried out at the mid-plane of the cyl-
inder at a dimensionless distance of x = 15. A consid-
erable variation in the static pressure was seen within
the wake and so the measurements were directed indi-
vidually at the total and static pressures. Some variation
in the tunnel static over a distance of 15 units was also
accommodated for. The conversion of voltages to ve-
locities was accomplished through an explicit non-
real-time method developed by Chew and Simpson
(1988). This technique has the advantage of determining
the mean and the ¯uctuating components of velocity as
well as their cross-correlations with very few assump-
tions regarding probe parity and turbulence level. Details
of the apparatus and instrumentation have been reported
elsewhere (Saha 1999).
 Velocity and velocity ¯uctuations in the wake of the
square cylinder were measured beyond a dimensionless
distance of 1.5 units. In view of the insensitivity of the hot-
wire probe to reversed ¯ow, measurements in the recir-
culation region of the cylinder are questionable. Earlier
studies show that the size of the recirculation region at
high Reynolds numbers is in the range 1.3±1.4 (DuraÄo
et al. 1988; Lyn et al. 1995). The recirculation zone was
established in these studies through alternative techniques
such as the LDV and the ¯ying hot wire. Hence, the choice
of the ®rst measurement station for the time-averaged
velocity as 1.5 can be taken as an acceptable starting point.
In experiments with low as well as high velocities, the
velocity ¯uctuations close to the cylinder can be expected
to be large. Under these conditions, the hot-wire mea-
surements using the ®xed operating point approach is
questionable. In the present experiments, the ®rst mea-
surement location was at x = 1.5, and a de®nite positive
value of the mean velocity was recorded. Hence, on no
occasion was the calibration curve utilized close to zero
velocity. Secondly, the instantaneous voltage signal of the
anemometer was converted to pointwise velocity and
subsequently, the time-averaged values were computed.
The difference between this strategy as against the con-
ventional operating point approach was found to be neg-
ligible, particularly in the high Reynolds number
experiments. This observation also ®nds support in the
work of Swaminathan et al. (1986). These authors have
shown a turbulence correction factor of 10% at a turbu-
lence intensity as high as 45% with respect to the mean
¯ow. The corresponding probability of occurrence of re-
versed ¯ow was found to be less than 1%. The above
discussion improves the degree of con®dence that can be
placed on the hot-wire data.
The uncertainty in the present set of experiments was
found to be unexpectedly low. This observation was drawn
from the fact that the dimensionless time-averaged ve-
locity pro®les at Reynolds numbers of 8700 and 17,625
were practically identical. The clear peak in the near-wake
spectra ensured negligible uncertainty in the Strouhal
number. The uncertainty in the drag coef®cient assessed
using the pilot tube and hot-wire data was found to be
within ‹5% with 95% con®dence.
2 of 14,000 and 21,400, respectively. The reasons for this
Results and discussion
The results of the present work have been organized as
follows: (1) drag coef®cient and Strouhal number; (2)
time-averaged velocity and velocity ¯uctuations; (3) Effect
of inlet turbulence; and (4) turbulence spectra.
555
Fig. 1. Comparison of measured Strouhal number with published
data
Reynolds number is shown in Fig. 2. The measured values
follow the pattern predicted by the numerical computa-
tion. The drag coef®cient shows initially an increasing
trend with Reynolds number and later becomes almost
asymptotic. It was not possible to measure drag below a
Reynolds number of 400 by the wake survey method,
owing to diffusive and wall effects. Measured values of
drag at low Reynolds numbers have not been reported in
the literature.
The experimental results of the present study are re-
ported for Re = 8700 and 17,625. The blockage ratio based
on the wind tunnel height is 6.25% and the length-to-width
ratio of the cylinder is 16. To a ®rst approximation, this
represents the cylinder in an in®nite medium; blockage
and aspect ratio effect are thus expected to be of secondary
importance (West and Apelt 1982).
At the Reynolds numbers studied, the drag coef®cients
were found to be 2.13 and 2.2, respectively. These values
are comparable with the numerical estimate of 2.07±2.67 at
a Reynolds number of 21,400 by Wang et al. (1996). It is in
closer agreement with the experiments of Cheng et al.
(1992), who have presented the drag coef®cient as 2.1.
Based on the vortex-shedding frequencies, the respec-
tive Strouhal numbers have been found in the present
experiments to be 0.144 and 0.142 at Re = 8700 and
17,625. These values differ from those reported by DuraÄo
et al. (1988) as well as Lyn et al. (1995), who present the
Strouhal number as 0.138 and 0.132 at Reynolds numbers

2.1
Drag coefficient and Strouhal number
To benchmark the measurement technique against pub-
lished data, selected experiments were carried out at low
Reynolds numbers. Figure 1 depicts such a comparison for
the Strouhal number as a function of Reynolds number.
Data available in the published literature are also included.
The data points of the present work lie between the nu-
merical results of Davis et al. (1984) and the experiments
of Okajima (1982). The drag coef®cient as a function of Fig. 2. Comparison of measured drag with published data
 discrepancy is possibly a high inlet turbulence level of
2±6% in the referred experiments. The numerical simula-
tion of Wang et al. (1996) shows a Strouhal number of 0.13
at a Reynolds number of 21,400. The discrepancy here is
probably related to a smaller blockage in the simulation of
Wang et al. (1996).
2.2
Time-averaged velocity and velocity fluctuations
A detailed comparison of the measured pro®les of velocity
556 and velocity ¯uctuations with the LDV measurements of
Lyn et al. (1995) and the LES calculations of Wang et al.
(1996) is now presented. Time-averaged velocities and
velocity ¯uctuations were non-dimensionalized using uav
the average incoming velocity. The turbulent shear stress
was non-dimensionalized using uav2.
The numerical calculations of Wang et al. (1996) em-
ploy a Reynolds number of 21,400 as stated above. Besides
this, the cylinder length is p, the blockage ratio is 0.05 and
periodicity boundary conditions have been applied in the
spanwise direction. In the LDV experiments, the Reynolds
number is 21,400, the cylinder length is 9.75 and the
blockage ratio is 7%. The inlet turbulence level in the
present experiments was 0.01% at 10 m/s (Re = 17,500)
and 0.05% at 5 m/s (Re = 8750). In the water tunnel ex-
periments of Lyn et al. (1995), the inlet turbulence level
was around 4% at Re = 21,400. In the numerical simula-
tion of Wang et al. (1996), the inlet turbulence level was
not speci®ed; one can expect it to be negligible. In view of
these differences, one cannot expect a complete match
among LDV measurements, numerical calculations and
the present hot-wire measurements.
Figure 3a and b shows the time-averaged streamwise
velocity pro®les at three locations downstream of the
cylinder at Reynolds numbers of 8700 and 17,625, re-
spectively. The numerical results are represented on the
same graph by solid lines. A remarkable aspect to be no-
ticed in Fig. 3 is the invariance of the dimensionless ve-
locity distribution with respect to both Reynolds numbers.
This is also an indication of a low uncertainty level in the
high Reynolds number experiments.
Figure 4 is a comparable plot for the transverse com-
ponent of velocity. Both Figs. 3 and 4 reveal good sym-
metry (and antisymmetry) in velocities with respect to the
wake axis. The measurement of a negative transverse ve-
locity is a check on the correctness of the data-reduction
procedure. The agreement between experimental and nu-
merical data is uniformly satisfactory.  
Figures 5 and 6 compare
 the streamwise u02 †1=2 and
transverse v02 †1=2 velocity ¯uctuations between experi-
ments and numerical simulation. The agreement here is
less satisfactory. Signi®cant differences can be seen at the
®rst measurement station. Both experiments as well as
calculations reveal a double-peaked pro®le in u02 , while
single peak is seen in v02 with a maximum occurring along
the axis. The differences between the experiments and
simulations tend to diminish at larger streamwise locations.
Figure 7 compares the turbulent shear stress from the
present experiments with the LDV measurements and the
numerical computation. The comparison is seen to be
closer in these ®gures compared with v02 . The change of
sign in u0 v0 † about the wake centreline is clearly revealed.
Once again, the agreement improves in the downstream
direction.
The centreline recovery of the time-averaged stream-
wise velocity is shown in Fig. 8a and b for the Reynolds
numbers under discussion. The present measurements
show that the recovery is faster compared with the ex-
periments of Lyn et al. (1995) and LES results of Wang
et al. (1996).
In summary, the numerical results match experiments
at high Reynolds number when compared in terms of the
time-averaged velocities, the streamwise velocity ¯uctua-
tions and the turbulent shear stress. Numerical predictions
of the transverse velocity ¯uctuations are signi®cantly
higher than the measured values. The possible reasons for
the discrepancy are the short spanwise dimension (=p) in
computations as against a value of 16 used in experiments,
Fig. 3a, b. Comparison of
measured time-averaged
streamwise velocity pro®les: a
Re = 8700; b Re = 17,625
 557

Fig. 4a, b. Comparison of

measured time-averaged trans-
verse velocity pro®les: a
Re = 8700; b Re = 17,625

Fig. 5a, b. Comparison of

measured time-averaged pro-
®les of streamwise velocity
¯uctuations: a Re = 8700; b
Re = 17,625

Fig. 6a, b. Comparison of

measured time-averaged pro-
®les of transverse velocity
¯uctuations: a Re = 8700; b
Re = 17,625
558

Fig. 7a, b. Comparison of

measured time-averaged pro-
®les of turbulent shear stress: a
Re = 8700; b Re = 17,625

stream. A minimum in this velocity is attained close to the

wake centreline. As seen in Fig. 3, the minimum velocity
increases in the downstream direction. Simultaneously,
there is an increase in the wake size. The two processes can
happen only when ¯uid is entrained from the edges of the
wake on each side of the centreline. A measure of the
entrainment of the ¯uid can be derived from the trans-
verse velocity pro®les in Fig. 4. It can now be seen that
entrainment, and hence the velocity recovery, are both
very high for small distances behind the cylinder. Both of
them asymptotically reach their respective limits for
higher downstream distances.
The variation in the turbulent ¯uctuations in the wake
requires greater elaboration. The mechanism involved is
not only dissipation but also the substantial energy
transfer from the core to the surrounding ¯uid. Hence, the
¯uctuations decay rapidly in the near wake, and their
transverse variation become progressively uniform.

2.3
Effect of inlet turbulence on the wake
Experiments with hot wire were in partial agreement with
LDV measurements and LES. Signi®cant differences were
seen in the pro®les of transverse velocity ¯uctuations. One
factor that has not been matched among the LDV exper-
iments, numerical simulation and the present experiments
is the turbulence level in the incoming ¯ow. The present
Fig. 8a, b. Comparison of recovery of measured time-averaged section is concerned with the study of inlet turbulence on
streamwise velocity along the wake centreline: a Re = 8700; the velocity ¯uctuations in the wake.
b Re = 17,625 The effect of inlet turbulence has been studied by
placing a circular cylinder centrally at a distance of 150
diameters (of the circular) upstream of the square cylin-
differences in the blockage ratio and the upstream tur- der. Consequently, the approximate turbulence level
bulence level. measured just upstream of the square cylinder (y = )2.0
A physical discussion on the results obtained can be to 2.0) was about 3.5%. The measurements at two different
conducted in the following manner. The separation of ¯ow streamwise locations, namely x = 1.5 and 2.5 were con-
over a bluff object causes a pressure drop across its sur- ducted at a Reynolds number of 12,000. The time-averaged
face, thus leading to a non-zero form drag. Equivalently, velocity pro®les at x = 1.5 (Fig. 9a) show the in¯uence of
this is manifested as loss of momentum of the ¯uid in the inlet turbulence on the recovery rate. The increase in the
wake. Thus, the time-averaged streamwise velocity is turbulence level leads to a slower recovery of the velocity
smaller within the wake compared with its value in the free de®cit. Simultaneously, the measured pro®les approach

Fig. 9a±c. Effect of upstream turbulence on time-averaged pro-
®les of: a streamwise velocity; b transverse velocity; and
c streamwise velocity ¯uctuations at x = 1.5 for a Reynolds
number of 12,000
those of Lyn et al. (1995). The effect of inlet turbulence on
the transverse velocity as depicted in Fig. 9b is seen to be
small. The time-averaged streamwise ¯uctuations at
x = 1.5 are presented in Fig. 9c. The ®gure shows that
there is a decrease in velocity ¯uctuations due to an in-
crease in inlet turbulence. Figure 10 shows the effect of
inlet turbulence on the u, v and u02 †1=2 at a location
x = 2.5 over one half of the wake. These pro®les are greatly
in¯uenced and approach the data of Lyn et al. (1995).
Thus an increase in the upstream turbulence in the present
experiments shows only a minor in¯uence at x = 1.5 but
an improved comparison with the published measure-
ments at a distance of x = 2.5.
The role of upstream turbulence in modifying the wake
can be explained as follows. External turbulence increases
the entrainment of the ¯uid into the free shear layer from
the recirculation zone and hence increases their curvature
559
Fig. 10a±c. Effect of upstream turbulence of time-averaged pro-
®les of: a streamwise velocity; b transverse velocity; and c
streamwise velocity ¯uctuations at x = 2.5 for a Reynolds number
of 12,000
(Huot et al. 1986). When the entrainment becomes very
large, the free shear layers eventually reattach to the sides
of the obstacle. Consequently, separation is reinitiated at
the rear corners. The entrainment of the ¯uid from the
freestream to the wake is then slow and the recovery rate is
small. Figure 10b shows that the transverse entrainment in
the present measurements is marginally higher at a loca-
tion x = 2.5 compared with Lyn et al. (1995). The higher
entrainment in the present experiments with low upstream
turbulence is possibly the cause for a higher recovery rate
compared with the published results.
2.4
Turbulence spectra
Since hot-wire experiments reveal local pointwise infor-
mation, it is usually cumbersome to map the ¯ow ®eld in
all respects. Instead, the power spectra of x and y com-
 ponents of velocity determined at selected locations in the
wake can shed light on the wake dynamics. With these
objectives, the u and the v spectra were recorded both
along the centreline (y = 0) and an offset position
(y = 1.0) at a Reynolds number of 17,625. Figure 11 shows
the u spectra at locations x = 2.5, 5.0, 10.0 and 30.0 for two
y planes in the cylinder wake. Figure 12 is a comparable
plot for the v spectra. In these ®gures, the abscissa is the
dimensionless frequency, while the ordinate is plotted in
560
Fig. 11. u Spectra at various streamwise locations
consistent units. (Note: At low Reynolds numbers, the
near-wake spectra showed dominant peaks. The far-wake
spectra were not recorded owing to the potential in¯uence
of the bounding walls).
Figure 11 clearly brings out a dominant peak at the
shedding frequency, particularly when the probe is away
from the centreline. The spectra along the centreline do
not reveal a clear trend. At y = 0, the second harmonic is
seen to be excited at x = 2.5, signifying the in¯uence of

Fig. 12. v Spectra at various streamwise locations
vortices shed from both halves of the cylinder on the
probe. At higher x locations the u spectra tend to become
broader, with a peak developing close to the origin. Thus,
¯ow near the centreline shows an established cascade with
energy transfer from the time-averaged streamwise veloc-
ity to the smaller ¯uctuations.
The peak in the u- spectra seen at the offset location
diminishes in strength as compared with that for the mean
561
¯ow for x- locations greater than 10.0. At x = 30.0, a
broadband spectrum in u is seen to be established.
In contrast to the u velocity, clear peaks are visible in
the v spectra at y = 0 as well as at y = 1.0. The peak is
observable all the way upto x = 30.0, though it is stronger
at the offset location. The broadening of the v spectra at
x = 30.0 indicates the loss of coherence in the ¯ow ®eld
and an evolution towards a fully developed state.
 An interesting result can be discerned by comparing the
peak values of u and v spectra between x = 2.5 and 5.0
(y = 1.0). One can see an increase in the peak values in the
downstream direction, followed by decay at farther dis-
tances. This result suggests an intensi®cation of the vor-
tices in the near wake, followed by their decay.
Two new results that emerge from the spectral plots are:
1. The shift in the peak of the u spectra towards the origin
of the frequency axis at x = 2.5 con®rms energy transfer
from the mean ¯ow to the streamwise velocity ¯uctua-
562 tion close to the cylinder. The redistribution of kinetic
energy between u and v occurs over a much longer
length, and is not complete even at x = 30.0.
2. The initial growth of the spectral peaks followed by a
decay indicates a strengthening of the vortices outside the
vortex formation length, namely 1.0±1.5. The underlying
mechanism is the transport of transverse momentum
from the core of the wake to the free shear layer.
Fig. 13a, b. Vorticity variation in the near wake of a square
2.5 cylinder: a Re = 8700; b Re = 17,625
Identifiable physical mechanisms
A physical interpretation of the results presented above is
Figure 13 shows that the positions of the vorticity
taken up in this section. The discussion is in terms of the
maxima and minima upto x = 4 are essentially unchanged.
spanwise vorticity in the time-averaged ¯ow ®eld and the
This is proof that the vortices move parallel to the mid-
turbulence kinetic energy. During the measurement pro-
plane in the near wake. The authors have computed ¯ow
gram, the traversing mechanism was set up to resolve the
past a square cylinder at high Reynold numbers using the
vertical (y) direction transverse to the wake. The stream-
k±e model (Saha 1999). Their numerical prediction also
wise (x) direction was resolved to a lesser extent. From the
shows vortices originating on the top and bottom side of
de®nition
the square cylinder. Farther downstream, the vortex cen-
ov ou ters are closer to the midplane, as opposed to the vortex
xˆ
ox oy shedding patterns for a circular cylinder.
the spanwise vorticity was hence approximated as The turbulence kinetic energy pro®les in the wake for the
two Reynolds numbers are shown in Fig. 14. A comparison
ou with Fig. 7 shows that k is similar to the turbulence shear
xˆ
oy stress u¢v¢ at all x locations, except that the latter changes
This is not a serious limitation, since v as well as its sign. The change of sign is quite prominent at the ®rst
streamwise derivative are relatively small outside the base x location where turbulence is produced by the shear in the
region of the cylinder. The turbulent kinetic energy (k) was mean velocity pro®le. At subsequent locations, turbulence is
computed using the formula
3 02
kˆ u ‡ v02 †
4
where the approximation
w02 ˆ 0:5 u02 ‡ v02 †
was used (Hadid et al. 1992).
The vorticity plots using the de®nition above are shown
in Fig. 13 for both Reynolds numbers of the present study.
A comparison of this data at x = 1.5 with Figs. 5 and 6
shows that large streamwise normal stresses are correlated
with the vortex centres the vorticity reaches a maximum.
This correspondence is not seen either in the transverse
normal stress component or at other x locations. This
result can be explained as follows: At the ®rst x location,
turbulence production (primarily the streamwise normal
stress) originates in the shear present in the time-averaged
¯ow and thus correlates with the vorticity distribution. At
other downstream points, turbulence is governed by the
transport processes in the wake and its association with Fig. 14a, b. Kinetic energy variation in the near wake of a square
the time-averaged velocity gradient is weakened. cylinder: a Re = 8700; b Re = 17,625
dominated by transport (to and from a point) including
transport from the core to the edge of the wake. Conse-
quently, the k and u¢v¢ pro®les are much ¯atter.
3 Chew YT; Simpson RL (1988) An explicit non-real time data
Conclusions
Hot-wire measurements in the wake of a square cylinder
were reported at Reynolds numbers of 8700 and 17,625.
The high Reynolds number experiments were compared
with LDV measurements and LES. The primary conclu-
sions of this study are as follows:
1. The agreement among the measured and simulated
data is satisfactory in terms of the time-averaged velocity
components, streamwise velocity and the turbulent shear
stress. The numerical prediction of transverse velocity
¯uctuations is substantially higher than the measured
values. This discrepancy can be attributed to (1) a small
spanwise length in the simulation in comparison to the
experiments, (2) differences in the blockage ratio, and (3)
a low upstream turbulence level in the present experi-
ments.
2. A fresh set of experiments with a higher turbulence
level in the approach ¯ow has shown that the comparison
improves with the published experimental data.
3. The spectra of the streamwise velocity component at
the near wake con®rms that the transfer of energy takes
place from the mean motion of the streamwise velocity
¯uctuation and the redistribution of kinetic energy be-
tween the two components of velocity occurs over a longer
streamwise direction downstream. The spectra also show
the transport of transverse momentum from the core of
the wake to the free shear layer.
4. At the ®rst measurement station x = 1.5, there was a
good association between the locations of the vortex
maxima and the turbulence shear stress. The correspon-
dence was lost at other downstream points. The turbulence
kinetic energy and the turbulence shear stress showed
similar trends in the wake. However, the shear stress
changed sign at the ®rst measurement point, revealing that
turbulence production at this site was shear-dominated.
References
Cheng CM; Lu PC; Chen RH (1992) Wind loads on square
cylinder in homogeneous turbulent ¯ows. J Wind Eng Ind
Aerodyn 41: 739±749
reduction method of triple sensors hot-wire anemometer in
three-dimensional ¯ow. J Fluids Eng 110: 110±119
Davis RW; Moore EF; Purtell LP (1984) A numerical-experi-
mental study of con®ned ¯ow around rectangular cylinders.
Phys Fluids 23: 46±59
DuraÄo DFG; Heitor MV; Pereira JCF (1988) Measurements of
turbulent and periodic ¯ows around a square cross-section
cylinder. Exp Fluids 6: 298±304 563
Hadid AH; Sindir MM; Issa RI (1992) Numerical study of two
dimensional vortex shedding from rectangular cylinders. Comp
Fluid Dyn J 2: 207±214
Huot JP; Rey C; Arbey H (1986) Experimental analysis of the
pressure ®eld induced on a square cylinder by a turbulent ¯ow.
J Fluid Mech 162: 283±298
Lyn DA; Einav S; Rodi W; Park J-H (1995) A laser-Doppler ve-
locimetry study of ensemble-averaged characteristics of tur-
bulent near wake of a square cylinder. J Fluid Mech 304:
285±319
Mukhopadhyay A; Biswas G; Sundararajan S (1992) Numerical
investigation of con®ned wakes behind a square cylinder in a
channel. Int J Numer Meth Fluids 14: 1473±1484
Okajima A (1982) Strouhal numbers of rectangular cylinders.
J Fluid Mech 123: 379±398
Saha AK (1999) Dynamical characteristics of the wake of a square
cylinder at low and high Reynolds numbers. Ph.D. Thesis,
Indian Institute of Technology Kanpur, India
Swaminathan MK; Rankin GW; Sridhar K (1986) Evaluation of
the basic systems of equation for turbulence measurements
using the Monte Carlo technique. J Fluid Mech 170: 1±19
Tamura T; Ohta I; Kuwahara K (1990) On the reliability of two-
dimensional simulation for unsteady ¯ows around a cylinder-
type structure. J Wind Eng Ind Aerodyn 35: 275±298
Wang G; Choudhuri PG; Bhandarkar MA; Vanka SP (1996)
Report, Large eddy simulations of bluff body wakes on parallel
computers, CFD 96-03, University of Illinois at Urbana ±
Champaign, Contract No. 0014-92-J-1334 of Of®ce of Naval
Research
West GS; Apelt CJ (1982) The effect of tunnel blockage and as-
pect ratio on the mean ¯ow past a circular cylinder with
Reynolds numbers between 104 and 105. J Fluid Mech 114:
361±377