Downloaded from https://www.cambridge.org/core.

National Cheng Kung University, on 19 Oct 2021 at 09:31:19, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S002211208300230X

J . Fluid Mech. (1983), id.137, p p . 83-113 83

Printed i n Ghat Britain

Structure of a turbulent separation bubble

By M A S A R U KIYA AND K Y U R O SASAKI
Faculty of Engineering, Hokkaido University, Sapporo, 060, Japan

(Received 12 January 1983 and in revised form 6 ,July 1983)

Flow in the separation bubble formed along the sides of a blunt flat plate with
right-angled corners has been studied in terms of extensive single- and two-point
measurements of velocity and surfacc-pressure fluctuations. The cross-correlations
between the surface-pressure and velocity fluctuations are found to be useful for the
study of large-scale vortex structure in the bubble. Large-scale vortices are shed
downstream from the separation bubble with a frequency of about 0.6U,/xR, where
Urnis the approaching velocity and xR is the time-mean length of the bubble. On
top of this regular vortex shedding, there exists a large-scale unsteadiness in the
bubble. Vortices which are much larger than the regular vortices are shed with
frequencies less than about o . 2 ~ T m / x RThe
. large-scale unsteadiness is accompanied
by enlargement and shrinkage of the bubble and also by a flapping motion of the
shear layer near the separation line. The intermittent nature of the flow in the bubble
is clarified in some detail. The distributions of the cross-correlations between the
pressure and velocity fluctuations demonstrate the vortex structure in the reattaching
zone. The longitudinal distance between the vortices is estimated to be ( 0 . 7 4 . 8 )x R
and their convection velocity is about 0.517, near the reattachment line. The
cross-correlations also suggest the existence of a longitudinal counter-rotating system
in the bubble. The distance between the axes of the rotation is of the order of 0 . 6 ~ ~ .
Variations of timescales, lengthscales and phase velocities of the vortices are
presented and discussed.

1. Introduction
Flows a t high Reynolds number with separation and reattachment have long been
a subject of many studies. Reviews on these flows, together with lists of references,
are included in articles by Fernholz (1978) and Johnston (1978). I n this paper, we
are concerned with a two-dimensional separation-reattachment flow (hereinafter
called a ‘separation bubble’) where the separation occurs from a sharp corner. An
excellent critical review of previous studies in this category is given by Bradshaw
& Wong (1972).
Very little is still known of details of the flow structure in the reattaching zone
where the separated shear layer reattaches onto a solid surface. A fundamental work
in this direction is Wood & Bradshaw’s (1982)experimental study of a turbulent plane
mixing layer constrained by a solid surface. The behaviour of large-scale vortices in
the reattaching zone plays a decisive role in determining flow propert>iesnot only in
the separation bubble but in the developing region downstream of the bubble. An
extensive measurement of turbulence in the developing region downstream of a
downward-facing step is reported by Chandrsuda & Bradshaw (1981). Bradshaw &
Wong (1972) argue that the large vortices are torn virtually into two in the
reattaching zone and that there is thus an abrupt change in the lengthscales of the
Downloaded from https://www.cambridge.org/core. National Cheng Kung University, on 19 Oct 2021 at 09:31:19, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S002211208300230X

84 M . Kiya and K. Sasaki

vortices. To the best of our knowledge, however, such changes in lengthscales have
not yet been confirmed experimentally.
We consider in this paper a separation bubble formed along a side of a blunt flat
plate with right-angled corners (figure 1 ) . The plate is aligned in such a way that
the sides are parallel to an approaching uniform flow. The plate is long enough for
the interaction between the separation bubbles on both sides to be negligible, so that
attention can be confined to one of the separation bubbles. This particular separation
bubble was chosen in this study in view of the following reasons.
(i) The thickness S of the boundary layer on the separation line is much smaller
than the height 2H of the plate when the Reynolds number is sufficiently high. The
thin shear layer enables us to omit the parameter S/2H from the interpretation of
the experimental results.
(ii) The solid surface downstream of the separation line belongs to one of the
simplest shapes. Since the shape of an afterbody affects the separation bubble, it
should be as simple as possible in order to obtain essential features of the bubble.
The time-mean velocity and pressure, turbulence intensities and Reynolds shear
stress in this separation bubble are reported by several authors (Ota & Itasaka 1976;
Ota & Narita 1978; Kiya, Sasaki & Arie 1982, hereinafter referred to as KSA).
Measurements are also made of the r.m.s. value, power spectrum, integral timescale
and cross-correlation between the surface-pressure and velocity fluctuations (Hillier
& Cherry 1981 a , b ; KSA). The last reference is mainly addressed to a discrete-vortex
simulation of the separation bubble. Observations of vortices in the bubble are
reported a t low and moderate Reynolds numbers by Lane & Loehrke (1978) and Ota,
Asano & Okawa (1981).
Hillier & Cherry (1981a ) showed the existence of a low-frequency unsteadiness in
the separation bubble, some of its basic phenomena, the shedding of vortical structure
from the reattaching region, and the fact that the unsteady motion is significantly
three-dimensional. I n the present paper, these features are confirmed and extended
by more detailed measurements. The low-frequency unsteadiness is also observed in
the separation bubble behind a downward-facing step (Eaton & Johnston 1982). They
suggest that the likely cause of the low-frequency motion is an instantaneous
imbalance between the entrainment rate from the recirculation zone and the
reinjection near the reattachment line.
The information obtained from the abovementioned studies, however, is not
adequate enough to clarify the structure of large eddies in the separation bubble,
especially in the reattaching zone. A good deal of work is needed to understand the
separation-bubble flows well enough to predict their behaviour in general. The
purpose of this paper is to present some new measurements which demonstrate the
phenomenon of the separated shear layer and the structure of large-scale vortices in
the bubble more clearly than the previous studies. The velocity and surface-pressure
fluctuations and their cross-correlations were extensively measured in the separation
bubble. The pressure fluctuations along the surface of the plate were found to be
indicative of the large-scale vortices, so that they were used as reference signals to
sort out the velocity fields associated with the vortices. The velocity-pressure
correlations were utilized to find coherent vortices in turbulent boundary layers
(Willmarth & Wooldridge 1963).
Figure 1 shows the definition of symbols. The Cartesian coordinates x, y, z are
defined in such a way that the x-axis is in the longitudinal direction, the y-axis is
vertically outwards from the side, and the z-axis is normal to the x- and y-axes so
as t o form a right-handed system. The origin is located a t an edge of the plate whose
Downloaded from https://www.cambridge.org/core. National Cheng Kung University, on 19 Oct 2021 at 09:31:19, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S002211208300230X

Structure of a turbulent separation bubble 85

’t I w +
t v+v’
w ’ L u+u’

us.p , H XR
___c --------
H

-------
‘\
i
/
I’
_ / - -

FIGURE
1. Configuration of flow and definition of symbols.

thickness is denoted by 2H. The time-mean components of velocity in the x-,y-,

z-directions and surface pressure are denoted by U , V , W and P respectively, and the
corresponding fluctuating components by u’, v’, w fand p’. The root-mean-square
values of the fluctuating quantities will be denoted by the suffix rms. Since the time-
mean flow is in the (x,y)-plane, the velocity component W is practically zero. The
velocity a t infinity upstream is denoted by U , and the pressure there by P,. The
time-mean reattachment line is written as xR.
I n $ 2 the experimental apparatus is briefly discussed. I n $ 3 the results are
presented and discussed. Finally a summary of the results is given in $4.

2. Experimental apparatus and methods

Experimental apparatus and method will be only briefly described here because
they were mostly the same as those employed in KSA. Experiments were performed
in an open-return low-speed air tunnel with a working section 40 cm high, 20 cm wide
and 1 m long. The freestream turbulence, which was originally 0.4 %, was reduced
to 0.2 yo a t the speed of 20 m/s by enclosing the intake of the tunnel by glass-wool
screens. The reduced freestream turbulence yielded a 5 yo longer separation bubble
than in KSA.
The blunt plate tested was 2 cm thick, 50 cm long and 20 cm wide. Its leading edge
was right-angled and located 50 cm downstream of the beginning of the test section.
The plate spanned the tunnel horizontally along its centreline, thus having an aspect
ratio of 10. No endplates were used. A number of pressure taps of 0.7 mm diameter
were fitted along the lower side of the plate a t the midspan. An exact alignment of
the plate was found by matching pressures recorded on tappings located a t the upper
and lower surfaces of the plate 5.5, 11.0 and 27.0 cm from the leading edge.
The Reynolds number was based on the freestream velocity and the height of the
plate was 2.6 x lo4 throughout the present experiment.
The tunnel-wall blockage, which originally amounted to 5 Yo, was corrected by
fitting false boundaries along the ceiling and floor of the test section. The false
boundaries were curved according t o the blockage-free streamline, flared by an
amount equal to the displacement thickness of the boundary layer along the tunnel
walls, and joined smoothly to the bell entrance of the test section.
The time-mean and fluctuating velocities were measured by linearized constant-
temperature hot-wire anemometers using single and X-wire probes and split-film
probes. The hot wires were tungsten wires of 5 pm diameter with a working length
Downloaded from https://www.cambridge.org/core. National Cheng Kung University, on 19 Oct 2021 at 09:31:19, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S002211208300230X

86 M . Kiya and K . Sasaki

of about 1 mm. The X-wires were inclined by +45O with respect to the main-flow
direction. The hot wires were calibrated statically in the undisturbed part of the flow
in the test section by assuming a cosine-law response to yaw. The overheat ratio of
the hot wires was about 1.5. Square-wave tests showed the maximum frequency
response to be about 10 kHz. No corrections were made for the effects of high
turbulence intensities on the time-mean and fluctuating velocities except when
otherwise stated.
Two kinds of split-film sensors of diameter 0.132 mm (produced by Thermo-Systems
Inc.) were employed. One had the plane of the split normal to the main flow, and
thus was able to detect reversals of the local flow direction. This was used to measure
U , u’ and the reverse-flow intermittency in the whole region of the bubble. The other,
which had the plane of the split parallel t o the main flow, was used to measure u‘
and u’ in regions where the reverse-flow intermittency was lower than 0.1. I n regions
of high turbulence with low time-mean velocity, u’ fluctuations measured by the
former are more reliable than those measured by the hot-wire probes. The split-film
sensors were calibrated statically in the undisturbed flow according to Technical
Bulletin TB20 TSI Split Film Sensor Calibration and Applications in a slightly
modified manner. A response of the split-film probes was checked by comparing
measured u’ spectra with those obtained by the single hot-wire probe in a grid
turbulence of U ~ ~ J U=, 0.058 and U , = 10 m/s. The spectra were practically
identical for frequencies less than about 1.5 kHz, the energy in high frequencies being
higher for the hot wire than for the split films.
Hot-wire results of velocity are likely to be reliable only if the ratio uirns/U.say,
is less than about 0.3, while if the relative intensity exceeds 0.5 hot-wire results are
likely to be highly unreliable (Chandrsuda & Bradshaw 1981).Results within the band
of 0.3 < ui,,/U < 0.5 should thus be treated with caution.
The hot-wire and split-film probes were mounted on a traversing mechanism
outside the tunnel. This allowed the positioning of the probe to be adjusted with an
accuracy of 0.1 mm. Pressure fluctuation on the plate surface was detected by a
semiconductor strain-gauged transducer, which was mounted inside the plate with
a small cavity between a pressure t a p of 0.7 mm diameter and the diaphragm of the
transducer. The gain factor of the transducer was 1 k 0 . 0 6 up to 530 Hz, with
negligibly small phase shift. Most of the measurements were made in the midspan
plane.
The velocity and pressure fluctuations were recorded on two or three channels of
an analogue tape recorder and later analysed by a computer (HITAC M200) or a
digital signal processor (SANE1 7T07A) to obtain various statistical properties. No
sensible phase lag was found to exist between the pressure transducer and the hot-wire
anemometers and between the hot-wire anemometers.
The flow was called ‘turbulent’ and the intermittency function was set to unity

-
if the short-time averaged energy

,@ = - [+4AtU;l(t)dt,
At t - i A t
wasgreater than a threshold value. Here uk was obtained by high-pass filtering (4kHz)
the longitudinal velocity fluctuation u’. The cut-off frequency 4 kHz was chosen
because the dissipation spectrumf2Eu<(f) attained a maximum approximately a t this
frequency. It is reasonable to assume that the flow is turbulent in regions where this
high-frequency fluctuation is detected. The high-frequency flucatuation has small
amplitudes so that u;.”is not influenced markedly in character by reversals of the flow
Downloaded from https://www.cambridge.org/core. National Cheng Kung University, on 19 Oct 2021 at 09:31:19, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S002211208300230X

Lb-ucture of a turbulent separation bubble 87

direction. The time interval At of the integration was chosen to be 0.8 ms, which is
about one-tenth of the vortex-shedding period (see 53.3). The threshold value was
selected as 3 x 10-5U&by comparing simultaneously the waveforms of u;." and the
intermittency function. The high-frequency energy in the potential field was assumed
to be well below the threshold level. This method of obtaining the intermittency
function is similar to that of Fiedler & Head (1966).
Fluid in the separation bubble moves intermittently back and forth corresponding
to the formation and subsequent downstream shedding of large-scale vortices. The
fraction of time during which the flow is in the upstream direction is defined as the
reverse-flow intermittency I,.. Measurements of I,. were made by the split-film probe.
The split-film output u ( t ) was averaged during a short time interval At, i.e.

C(t)= - J+'At u ( t )dt,

At t-tAt

where At was chosen to be 0.8 ms. The signal G(t)was subsequently used t o calculate
Ir.

3. Results and discussion

3.1. Two-dimensionality of the $ow and the end eflects
As mentioned in $2, no endplates were employed in this experiment. One may argue
that the endplates should be used to improve the two-dimensionality by modifying
the flow in the junction between the tunnel wall and the plate. It should be noted,
however, that the use of the endplates cannot necessarily suppress a vortex system
in the junction, i.e. a horseshoe vortex and longitudinal trailing vortices. Hillisr &
Cherry (1981a) correctly state that the endplates are unlikely to improve the
two-dimensionality of the unsteady flow field in the separation bubble. The use of
the endplates cannot be regarded as compulsory.
The spanwise uniformity of the time-mean and r.m.s. longitudinal velocities was
checked in the time-mean reattachment section, where the flow was suspected to be
most disturbed by the end effects. The uniformity was found to be excellent over a
region & 3.5H on both sides of the midspan, the maximum variations in U and u;ms
in this section being less than 1.5 yo and 1 yo respectively of the freestream velocity.
I n regions lz/HI 2 4.0, however, the two-dimensionality decreased as lz/HI increased.
The surface-flow visualization (see figure 9 of KSA) showed a fairly two-dimensional
pattern in the region lz/HI 5 4.This extent of the spanwise uniformity was conjectured
to be tolerably large to ensure that the separation-bubble properties which were
measured in the midspan plane were representative of the infinite-span case. The
present aspect ratio of 10 is partially justified by the fact that the reduction of the
aspect ratio of 13.2-9.33 produces no systematic effect on the bubble length or mean
and fluctuating quantities (Hillier & Cherry 1981a ) . Some features of the separation
bubble to be presented in the following are possibly contaminated to a certain extent
by the vortex system in the junction. We feel that such contamination is not serious,
because, as will be demonstrated later, some of the present results do agree with
Hillier & Cherry's ( 1 9 8 1 ~data
) taken a t the aspect ratio of 13.2, which they claim
to be sufficiently large. The vortex system is supposed to be fairly steady, so that
it is unlikely to produce significant fluctuating quantities in the separation bubble.
Downloaded from https://www.cambridge.org/core. National Cheng Kung University, on 19 Oct 2021 at 09:31:19, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S002211208300230X

88 M . Kiya and K . SasaEi

0 0.5 0 0.5 0 0.5 0

UIU-
FIGURE 2. Distribution of time-mean longitudinal velocity in and around the separation bubble :
0, split-film result; 0 , X-wire result corrected for high turbulence intensities; ---, dividing
streamline (solid lines for visual aid only).

3.2. Mean-$ow properties

The time-mean reattachment line of the separated shear layer was found to be a t
x / H = 10.1 on the basis of the reverse-flow intermitteney (see figure lo), i.e.
xR = 10.1H. Profiles of the longitudinal mean velocity ZJ measured by the split-film
probe are shown in figure 2, together with those obtained by the X-wire probe. The
hot-wire results were corrected for the effect of high turbulence intensities by noting
that the addition,of time-mean outputs of the X-wire probes yields
2iU(1+ w;",,/u2 + higher-order terms).
The term w&,,luZ was corrected by iteration in terms of measured w&,,.The hot-wire
results are seen to be reliable in regions where U / U , is greater than about 0.4.
I n order to show the development of the separated shear layer, it is convenient
to introduce a few representative parameters. A position where uims attains a
maximum value can be interpreted as the centre of the shear layer because the lateral
distributions of IT have a point of inflexion in the vicinity of this position. The normal
distance between this position and the plate surface is denoted by yc. The time-mean
edge of the shear layer y6 was defined as a position where U amounts t o 99% of a
maximum velocity. Another edge y B was defined as a position at which an
intermittently turbulent signal first appeared when a hot-wire probe was traversed
vertically from outside of the separation bubble. Although one may anticipate a
considerable uncertainty in determining yB, this position is believed to be obtained
with the same accuracy as ys. The results are shown in figure 3. The position yc
changes only slightly in the region x/xR 3 0.6, where yc/xR is approximately equal
to 0.14. The positions y s and yB, on the other hand, continue to increase with
increasing x approximately in proportion to xi.
The time-mean surface pressure in the separation bubble was in fairly good
agreement with the results of KSA and Hillier & Cherry (1981b ) .

3.3. Velocity Jluctuations

Since distributions of the r.m.s. velocities u:,,/U, and zlims/Uas and the Reynolds
shear stress -m/Vm were generally the same as those given in RSA, only the
Downloaded from https://www.cambridge.org/core. National Cheng Kung University, on 19 Oct 2021 at 09:31:19, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S002211208300230X

Structure of a turbulent separation bubble 89

XIXR

3. Three representative positions of the separation-reattachment flow: 0,

FIGURE yc; a,y6;
0 ,yR (solid lines for visual aid only).

&* /U,
4.Distributions of r.m.s. longitudinal velocity u;,,/U, : 0,
FIGURE split-film result; 0 ,X-wire
result; ---, dividing streamline (solid lines for visual aid only).

split-film results will be presented. The transverse r.m.s. velocity w;,,/U, was also
measured, but will not be presented.
The distributions of u;,,/U, are shown in figure 4, which compares two sets of
data measured by the split-film and X-wire probes. They are in fairly good agreement
in regions where u;,,/U, 5 0.5. Figure 5 compares the lateral r.m.s. velocity v:,,/U,
measured by the split-film probe with the X-wire result in the reattachment section
x/xR = 1.O. The latter is seen to be reliable only in the region y/xR 2 0.15, where the
reverse-flow intermittency (see $3.4)is zero.
I n the neighbourhood of the dividing streamline, the level of the r.m.s. velocities
Downloaded from https://www.cambridge.org/core. National Cheng Kung University, on 19 Oct 2021 at 09:31:19, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S002211208300230X

90 M . Kiya and K . Xasaki

0.4

0.3

0.2

0.1

0 0.1 0.2 0.3

FIGURE 5. Distributions of r.m.s. lateral velocity z1ims/t79 in the reattachment section

x / x R = 1.0: A, split-film result; 0 , X-wire result (solid line for visual aid only).

were in the order uims,w;,, and u ~ ~ This, . order is the same as that in the turbulent
plane mixing layer between a uniform stream and a still fluid (Wygnanski & Fiedler
1970). Maximum values of u:ms/ITm,wgm,/UoOand t~~ms/lTco were roughly 0.20, 0.20
and 0.13 respectively for x/xR 5 0.4, and 0.25,0.22and 0.17 for x/xR 2 0.6,while the
corresponding maximum r.m.s. velocities in the plane mixing layer are 0.18,0.15and
0.14 (Wygnanski & Fiedler 1970). Since the maximum velocity difference UM in the
longitudinal velocity component across the shear layer is larger than the freestream
valuc IT,, the comparison with the data of the plane mixing layer may be made more
appropriately by employing ITM as a reference velocity than ITo0. The maximum
values of &,,,/cTM, w&/~’M and v:m,/uM were on the average 0.14, 0.15 and 0.09
respectively for x/xR 5 0.4, and 0.21, 0.18 and 0.14 for “T/zR 2 0.6. The somewhat
smaller values of the r.m.s. velocities in the upstream region x/xR 5 0.4 than in the
plane mixing layer are possibly associated with the stabilizing effect of the convex
curvature of the separated shear layer (Johnston 1978). Flow-visualization studies
(Lane & Loehrke 1980; Ota et al. 1981) and the discrete-vortex simulation of KSA
suggest that the rolled-up vortices in the shear layer contact with the surface roughly
a t x/xR = 0.5; the behaviour of the large vortices and thus the velocity fluctuations
in the region ./YE >, 0.5 are possibly affected by the presence of the solid surface.
The Reynolds shear stress in the reattachment section x/xR = 1 .O is shown in figure
6. The split-film result for the maximum value of -m
is larger by about 30% than
the hot-wire result. Since the hot wire was aligned parallel t o the plate surface, its
result is probably affected by a relatively large longitudinal curvature of time-mean
streamlines. The measured -m is known to be sensitive to a small misalignment
of the probe with respect to the inclination of the time-mean streamlines. The splibfilm
Downloaded from https://www.cambridge.org/core. National Cheng Kung University, on 19 Oct 2021 at 09:31:19, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S002211208300230X

Structure of a turbulent separation bubble 91

0.4 I I

0.:

-
Y
xR
0.;

0. I

0 0.01 - 0.02 0.03

-u’v’luz,
6. Distribution of Reynolds shear stress in the reattachment section x/xE
FIGURE = 1.0: A,
split-film result; 0,
X-wire result (solid line for visual aid only).

data are supposed to be more reliable because this probe can detect the u’ and v’
fluctuations irrespective of the mean-flow inclinations in regions where the reverse-flow
intermittency is zero.
Figure 7 shows the power spectrum Eu(of u’together with the surface-pressure
spectrum Eprin the reattachment section x/xR = 1.0. The spectra are defined by

where f is the frequency. The velocity spectrum a t ys has a broad maximum a t the
frequencyfxR/ Uoox 0.6, whichisthesameasthepeakfrequencyofthesurface-pressure
spectrum. The disappearance of the peak a t the centre yc implies that the velocity
fluctuations associated with large-scale coherent vortices were contaminated by those
due to small-scale eddies. This suggests that the large-scale vortex structure in the
separation bubble is reflected in the velocity fluctuations in the outer part of the
bubble and in the surface-pressure fluctuations.
For a turbulent boundary layer with zero pressure gradient, Bradshaw (1967)shows
that the peak frequency of the u’ spectra in the irrotational field becomes lower as
the distance from the turbulent region increases, and that the peak frequency tends
to that of the surface-pressure spectrum as the hot-wire position approaches the edge
of the turbulent region. Since these properties are believed to be independent of the
turbulent structure, the u’ spectra in the present case probably depend on the dis-
tance from the turbulent region. The velocity fluctuations in the irrotational field
are driven by the large-scale vortices: it is reasonable to assume that the longitudinal
motion of the large vortices is well reflected in the u’spectra near the edge of the
turbulent region. This is the reason why the spectra were not measured beyond ys.
4 FLX 137
Downloaded from https://www.cambridge.org/core. National Cheng Kung University, on 19 Oct 2021 at 09:31:19, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S002211208300230X

92 M . Kiya and K . Sasaki

FIGURE 7. Power spectra of longitudinal-velocity and surface-pressure fluctuations in the reattach-

ment section x/xE = 1.0: 0, E r f / ~ & sat ys; Cj, E,,/u& a t yc; A, Epf/p&ms.
Velocity fluctuations
were detected by a single hot wire. (Solid lines for visual aid only.)

By the use of the unsteady form of Bernoulli equation and Taylor’s hypothesis,
Bradshaw (1967) demonstrates that in the potential field the pressure-fluctuation
spectrum is nearly the same shape as the u‘ spectrum. He also argues that the
low-frequency contributions to the pressure fluctuation could not differ markedly in
character between the irrotational field and the surface, so that the surface-pressure
spectrum has nearly the same shape as the u‘ spectrum near the edge of the turbulent
region. The low-frequency range well includes the peak frequencies. A similarity in
the u‘ spectrum a t y8 and the surface-pressure spectrum (figure 7) may suggest that
Bradshaw’s argument also applies t o this more complex flow.
The cross-spectrum IEUfv.land the phase #., of the velocity fluctuations u’, v’
measured a t the positions y6 and yc in the reattachment section are shown in figure
8. The cross-spectra attain a maximum a t the frequency fxR/Um x 0.5, which is near
to the peak frequency of Eupand E p f .It is reasonable to assume that these peaks
of the spectra correspond to the shedding of large-scale vortices from the separation
bubble. Such vortices are clearly observed in flow-visualization studies (Hillier &
Cherry 1981b ; Ota et al. 1981). From figures 7 and 8 and from the cross-spectra and
coherence of surface-pressure and velocity fluctuations (figure 29), the typical
vortex-shedding frequency is estimated to be 0 . 6 u , / X ~ . This agrees fairly well with
the value 0.7Um/sRobtained by Hillier & Cherry (1981a). An examination of Eaton
& Johnston’s (1982)results for a separation bubble downstream of a two-dimensional
sudden expansion, the expansion ratio being 3 : 5 , reveals that the u’ spectra in the
reattaching zone have a broad but clear peak a t a frequency fxR/U, = 0.5, where
Urn is taken as the main-flow velocity upstream of the expansion. The aspect ratio
in their study is 12. The peak frequency is close to that for the leading-edge separation
bubble. This feature strongly suggests auniversal nature oftwo-dimensional separation
bubbles downstream of a sharp corner.
Downloaded from https://www.cambridge.org/core. National Cheng Kung University, on 19 Oct 2021 at 09:31:19, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S002211208300230X

Structure of a turbulent separation bubble 93

6"V In
-0.5
0
0;s 1;o
fxRfum
1;s

-0.5
oj 0;s
fxR/um
2;o I
-1 .o -1 .o
-1.5 -1.5
FIGURE8. Cross-spectrum IE,.,,l and phase $,., of the velocity fluctuations u', v' in the
reattachment section x/xR = 1.0: ( a )y = ys; ( b ) y = yc. Velocity fluctuations were detected by a
X-wire probe. The vertical scales for (E,,,,(: are linear but arbitrary.

It should be noted in passing that the phase is about ~ I Cand -IC at ys and
yc respectively. This is consistent with the fact that the velocity fluctuations u', v'
a t ys contribute very little to the Reynolds shear stress --, while those a t yc
produce the Reynolds shear stress most effectively at the frequencyfxR/Um x 0.5.
Figure 9 shows the intermittency factor of turbulence yt plotted against the
normalized distance y/ys, together with the result for the downward-facing step
(Chandrsuda & Bradshaw 1981). The results are similar for the two cases, except that
the maximum of yt is somewhat higher in the present case than in the downward-facing
step. It is also interesting to note that yt never attains t o unity through the
reattachment section and that yt significantly decreases towards the wall.
The burst frequency ft was defined as the number of turbulent bursts per unit time :
ft is zero if the flow a t a fixed point is always laminar or turbulent. This was
determined in terms of the intermittency function mentioned above. As shown in
figure 9, the burst frequency becomes a maximum a t the height where yt = 0.5, and,
unlike turbulent boundary layers, increases towards the wall. These features suggest
an intimate connection between flows in the outer and inner parts of the reattaching
zone.
3.4. Reverse-$ow interrnittency
Figure 10 shows the longitudinal distribution of I , measured a t y/xR = 0.005,
together with the time-mean longitudinal velocity Us a t the same height. The position
of I , = 0.5 coincides exactly with that of Us = 0. The time-mean reattachment line
of the shear layer can be defined as the line where I , = 0.5. It is interesting t o note
that I , never becomes unity, even in the middle of the bubble. A sharp decrease of
I , towards the edge of the plate suggests the formation of a secondary separation
bubble in the downstream vicinity of the separation line. These features of I , are well
represented by the discrete-vortex simulation of KSA. If the reattaching zone is
defined rather arbitrarily as a region of 0.1 < I , < 0.9, its width is about 0.42,.
The distribution of the intermittency across the separation bubble is shown in figure
11. The intermittency seems to attain a maximum near the plate surface.
4-2
Downloaded from https://www.cambridge.org/core. National Cheng Kung University, on 19 Oct 2021 at 09:31:19, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S002211208300230X

94 M . Kiya and K. Sasaki

7t
FIGURE 9. Turbulence intermittency factor yt and burst frequency ft in the reattachment section
x / x R = 1.0: 0,yt (present result, hot-wire measurement); A, yt (Chandrsuda & Bradshaw 1981,
downward-facing step); 0,
ft (present result) (solid lines for visual aid only).

1 .oo I I I I 0.50
p-'-u-uO-0---6-o-o-\'

9
0 -
A
0 6 A

- \
0.75 - 0.25
"\
*'.I

.-- \ ________-_
*/J us
-
0'

A
*I*'
-----------
o.so-\.

' 0 '0 ./*/.

.B I 0

0.25 - '0
'0,
0'0, 0,.
0
'
-0.25
.-*-.-*-.O ,\O
0
'
'
0
0 I I -0.50

Figure 12 presents the distribution of the frequency fr, which is half of the average
number of changes in the flow direction during unit time :fr is zero if the flow is always
in one direction. The frequency f, was obtained in terms of the short-time-averaged
velocity zX(t). The non-dimensional frequency fr zR/l T , takes a maximum value of
about 0.7 along the line 1, = 0.5. This maximum value is not far from the frequency
of vortex shedding from the separation bubble.
Downloaded from https://www.cambridge.org/core. National Cheng Kung University, on 19 Oct 2021 at 09:31:19, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S002211208300230X

Structure of a turbulent separation bubble 95

4
FIQURE 11. Distribution of reverse-flow intermittency I , in the separation bubble ; measurement
by a split-film sensor: ---, I, = 0.1 ; ---, 0.5; - - - -0.9
, (solid lines for visual aid only).

fr X R I u,
FIQURE 12. Distribution of frequency f,.,which is half of the average number of change in the flow
direction during unit time; measurement by a split-film sensor: ---, I , = 0.1 ; ---, 0.5; - - - -, 0.9
(solid lines for visual aid only).

3.5. Integral timescales of velocity fluctuations

The integral timescales Tur,T,. and T,. of the velocity fluctuations u', v' and UI'
respectively were obtained from the corresponding autocorrelation coefficients. Their
distributions in the lateral direction are plotted in figure 13 for the reattachment
section x/xR = 1.0. The timescale T,. takes a minimum value in the vicinity of ys,
whereas T,.and !Iare ,#constant in the region y Q t(ys+yc). The ratios T,./T,.
!!fairly
and Tu,/Twf
-are roughly two in the inner region y 5 yc and roughly unity in the outer
region y ys. Similar results were also obtained in the sections x/xR = 0.6 and 1.4.
Antonia & Van Atta (1977) observed the same feature in two-dimensional turbulent
boundary layers.
Figure 14 shows the longitudinal distributions of the timescales T,!,T,.and T,.
measured at ys and denoted by Tufs, TVfsand T,, respectively. Also included is the
timescale Tqr6obtained by a single hot wire parallel to the z-axis. This timescale was
presented in order to give a measure of the timescales near the separation line
where the spatial resolution of the X-wire probes is poor owing to extremely small
Downloaded from https://www.cambridge.org/core. National Cheng Kung University, on 19 Oct 2021 at 09:31:19, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S002211208300230X

96 M . Kiya and K.Sasaki

0.30 I I I I

0.25 -

& 0.20-
-tt
-
s”
(-3 0.15 -
ha

0.10- -

0.05 -

L I 1
0
0. I 0.2 0.3 0.4 0.5

.‘ I

0.2
I

0.4
I
0.6 0.8
1 I
1.0
I
1.2
1
1.4
I

1.6 1.8
x/xR
FIQURE 14. Change of integral timescales a t y6 in the longitudinal direction; measurement by
single and X-wire probes: 0, A,T,,,; 0,
Tufa; T,,,; 0 ,TQz6(solid line for visual aid only).

shear-layer thickness. The four timescales are roughly the same within the
experimental uncertainties. They increase approximately linearly with increasing x
up to the time-mean reattachment line, downstream of which the timescales are
almost unchanged. An increase in the timescales can be interpreted as an increase
in the size of coherent vortices under the assumption that the convection velocity
of the vortices does not change significantly in the longitudinal direction. Nearly
constant timescales downstream of the reattachment line are thus conjectured to
result from hampered growth of the vortices due to the presence of the plate surface.
The timescales T,.,, Turcand Twrcof u’, v’ and wfrespectively were also measured
along the centre y c (not presented here). The timescales TVtcand TWfcwere almost
the same and changed in the longitudinal direction in the same way as the timescales
Downloaded from https://www.cambridge.org/core. National Cheng Kung University, on 19 Oct 2021 at 09:31:19, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S002211208300230X

Structure of a turbulent separation, bubble 97

I I I I I I I 1 1

0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8

XIXR

FIQURE 15. Integral timescale Tprand longitudinal lengthscale L,,, of surface-pressure fluctua-
tion: 0, T,,; a,L,,, (present result); 0,L I P ,(Hillier & Cherry 1981a) (solid lines for visual
aid only).

a t ys. The timescale Turcwas larger, very roughly by a factor of 2, than Tt,rcand Twrc
in the vicinity and downstream of the reattachment line, whereas it was much larger
than the latter near the separation line. This feature suggests that the u’ velocity
along ye has contributions not only from the rolled-up vortices but also from another
agent which had a much longer timescale than the vortices.
3.6. Surface-pressure jluctuations
The r.m.s. surface-pressure distribution was in fairly good agreement with the results
of KSA and Hillier & Cherry (1981b). A broad maximum of about 0.12(4pU2,)
appeared a little upstream of the reattachment line. The maximum was so broad that
the r.m.s. pressure changed only insignificantly in the reattaching zone.
Figure 15 shows the integral timescale Tpj of the pressure fluctuations plotted
against the longitudinal distance. The timescale is fairly constant in the downstream
half of the bubble, whereas in its upstream half Tpfincreases sharply towards the
separation line. The latter feature is conjectured t o be associated with a low-frequency
lateral oscillation of the shear layer near the separation line, which will hereinafter
be referred to as ‘flapping’ motion of the shear layer. The flapping is perhaps
responsible for much larger Tuft than 1Lfc and Tu,rcshortly downstream of the
separation line.
The timescale Tp,,together with the phase velocity Ucprof the pressure fluctuation,
yields the longitudinal lengthscale Lrpr(= CTcpp Tp,).The result is also plotted in figure
15. The lengthscale has a fairly constant value of 0. 112, from z/zRz 0.6 to 1.7,which
Downloaded from https://www.cambridge.org/core. National Cheng Kung University, on 19 Oct 2021 at 09:31:19, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S002211208300230X

98 31. Kiya a,nd K . Xasaki

fXR/U..

FIGURE
16. Surface-pressurespectrum E,. at the section x / x E = 0.2; vertical scale is linear but
arbitrary (solid line for visual aid only).

is the most distant section where the measurements were made. The value of
L x p ,= 0 . 1 1 X R is almost the same as the lengthscale Lzu,of the longitudinal velocity
fluctuation in the reattachment section (see figure 18). Accordingly i t is reasonable
to assume that the surface-pressure fluctuation is produced by the same large-scale
vortices as those which are responsible for the velocity fluctuation u’, a t least in the
vicinity of the reattachment section.
The pressure spectrum Epra t the reattachment section has been presented in figure
7. The peak frequency fxR/Um x 0.6 in figure 7 was also observed in the pressure
spectra measured in the region 0.6 < x/xR < 1.7 and in Hillier & Cherry’s ( 1 9 8 1 ~ )
spectra in the range 0.55 < x/xR < 1.35. As shown in figure 16, the spectrum a t
x/xR = 0.2 has a clear peak a tfxR/Um x 0.12. The same peak frequency was observed
a t the position x/xR = 0.3 (not given here). Moreover, the pressure spectra presented
by Hillier & Cherry have a peak at the same frequency in the region
0.026 < x/xR < 0.25. As has been pointed out by Hillier & Cherry, this feature is a
result of the flapping of the shear layer near the separation line, which will be
described in detail in $3.8.Since the pressure spectrum in figure 16 takes a minimum
a t fxR/Um x 0.2, the flapping frequency is perhaps distributed in the range
fXR/ET, < 0.2, with a central frequency fxR/Um = 0.12.
I n their downward-facing step flow, Eaton & Johnston (1982) observed a low-
frequency unsteadiness which probably accompanies the flapping of the shear layer.
Although the resolution of their u’ spectra was not sufficient to identify any spectral
peak in the low-frequency range, more than 30 yo of the measured u& occurred a t
frequencies belowfxR/ x 0.15.This frequency is comparable t o the central flapping
frequency fxR/Um = 0.12 for the present case. Since the vortex systems in the ends
of the plate are supposed to be markedly different from those in the ends of the
expansion, almost the same value of the non-dimensional frequency suggests that the
flapping is not directly connected to the flows in the ends.
The skewness of the surface-pressure fluctuation was positive in the reattaching
zone, attaining a maximum near the reattachment line and being negative upstream
Downloaded from https://www.cambridge.org/core. National Cheng Kung University, on 19 Oct 2021 at 09:31:19, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S002211208300230X

Xtructure of a turbulent separation bubble 99

0 L 0.2 0.4A-0 0.6 0.8 1.0

cum#,
U,.,lU,
FIGURE17. Distribution of phase velocities Ucuf and U , , in the reattachment section
x/xR = 1.0: 0, t l c p . ; A, U , , (Hillier &, Cherry 1981~);
ticu,;0, - - - ,y b ; --, yc (solid line
for visual aid only).

and downstream of the reattaching zone. From the discrete-vortex simulation of KSA
and also from Komatsu & Kobayashi's (1980) experiment, among others, the pressure
fluctuation at a fixed position is found to be negative beneath large-scale vortices and
positive when no such vortices exist above that position. The positive skewness
implies that the pressure waveforms are positivciy spiky. The positive spikes are
probably produced by the inrush of irrotatiorul fluid of high total pressure towards
the plate surface.

3.7. Phase oelocity and integral lengthscales of zrelocity and

surface-pressure Jluctuations
Figure 17 shows the distribution of the phase velocity tTcur of u' in the reattachment
section. Two single hot-wire probes were mounted on a traversing mechanism in such
a way that they were always a t the same height ylx, from the plate surface. The
longitudinal coordinate of one wire was x/xR = 0.8, while that of the other was
x/xR = 1.2. The cross-correlation of u Ja t the two positions gave a time lag which
was used to estimate the phase velocity in the reattachment section X/xR = 1 .O. The
phase velocities are approximately between 0.5U, and 0.7Um,attaining a maximum
O.7U, a t a height a little below the edge ys of the separation bubble.
The phase velocity Ucpl of the surface-pressure fluctuation p' a t a particular
position x was obtained by measuring the time lag between the outputs of the two
pressure transducers placed a t r+0.2xR. The value of Ucp.on the reattachment line
is 0.5U,, and seems to be joined smoothly to Ucu,(see figure 17).
Figure 18 shows the distributions (in the reattachment section) of the longitudinal
integral lengthscale Lxu,(= Ucu,Tur) and the spanwise lengthscale Lzulobtained from
the cross-correlation of u' at two positions separated in the spanwise direction. The
lengthscale Lzuj remains nearly constant at 0.122, across the bubble, whereas the
spanwise length Lzurincreases significantly with increasing y. The latter suggests that,
Downloaded from https://www.cambridge.org/core. National Cheng Kung University, on 19 Oct 2021 at 09:31:19, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S002211208300230X

100 M . Kiya and K . Sasaki

(Lxli, L,**
LZU5bP.) I X R
FIGURE 18. Longitudinal and spanwise integral lengthscales of longitudinal-velocity and surface-
pressure fluctuations in the reattachment section x/xR = 1.0: A >L,,,; A ,L,,,; 0, Lz,,; 0 ,Lzp,
(Hillier & Cherry 1 9 8 1 ~ )-;- --,ys;-, yc (other lines for visual aid only).

in the central part of the bubble, coherent velocity fluctuations associated with large
vortices are contaminated by small eddies which are highly three-dimensional. Such
small eddies are expected to reduce the spanwise length Lzu,.The almost-constant
longitudinal scale Lxu,is conjectured to be brought about by a combined effect of
the small eddies and the low-frequency flapping of the separated shear layer (see fj3.8).
A perusal of the autocorrelation of u' indicated that a drop of the correlation near
zero time lag was greater a t yc than a t ys. This can produce a shorter timescale Tur
a t yc than a t ys. On the other hand, the flapping manifested itself as a somewhat
slower decay of the autocorrelation for large time lag a t yc than a t ys. This is because
the vertical velocity gradient aU/ay is the greatest near yc : the flapping is felt much
more strongly a t yc than at ys. Such slow decay of the autocorrelation tends to
increase the timescale T,.. A combined effect of the small eddies and the flapping
happened t o be a slight increase of Tu. as y decreases from ys t o yc (see figure 13).
This increase of T,. is offset by a decrease of the phase velocity Ucu,t o produce almost
constant lengthscale Lxu,across the bubble.
The ratio Lzuy/Lzu, is about 0.5 at the centre yc and about 2.6 at the edge of the
bubble. It may be noted that the value Ltuf/xR= 0.12 on ys in the reattachment
section is about 20% greater than that of the plane mixing layer, where x R should
be interpreted as the distance measured from the origin of mixing. The spanwise
length of the surface-pressure fluctuations L,,, along the reattachment line is
approximately equal to Lzu,in the reattachment section.
The spanwise integral lengthscales Lzupalong ys and yc are plotted in figure 19
against the longitudinal distance. The lengthscales Lzurare fairly constant downstream
of the reattarhment line, as are the timescales of the velocity fluctuations (see figure
14). An increase of L,,, with increasing x up to the reattachment line is perhaps
associated with the amalgamation of rolled-up vortices, which produces larger and
larger vortices in the downstream direction. Vortices of larger cross-section are
Downloaded from https://www.cambridge.org/core. National Cheng Kung University, on 19 Oct 2021 at 09:31:19, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S002211208300230X

Structure of a turbulent separation bubble 101

0'4

0,
e,
19. Spanwise integral lengthscales of u' along yc and ya plotted against X / X R :
FIGURE Lzu,a;
Lzufc(solid lines for visual aid only).

i
1.o

\
0.8 \
\
\
R u'u' \

20. Autocorrelation of longitudinal-velocity fluctuations Rusu8near the centre of the

FIGURE
separated shear layer: -, x/xR = 0.02; ---, 0.4.

expected to yield velocity fluctuations that are more correlated in the spanwise
direction.

3.8. Flupping of the shear layer and large-scale unsteadiness i n the bubble
The autocorrelation coefficients of the longitudinal velocity fluctuation Ruflu,near the
centre of the shear layer y c are shown in figure 20 for two positions shortly
downstream of the separation line. The autocorrelations have a long tail which was
clearly observed up to the section x/xR = 0.5. Such a tail was gradually lowered as
Downloaded from https://www.cambridge.org/core. National Cheng Kung University, on 19 Oct 2021 at 09:31:19, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S002211208300230X

102 M.Kiya and K.Sasaki

FIGURE 21. Cross-correlation of longitudinal-velocity and surface-pressure fluctuations. The signals

were low-pass filtered (0.25U,/xR = 50 Hz) before processing. The vertical arrow $ shows the
location of the fixed surface-pressure transducer whose output is denoted by p i . --- > yc; - - - - - ,
ys (other lines for visual aid only).

x/xR increased further, but a trace of the tail was still obvious even in the
reattachment section. At x/xR = 0.02, a high-frequency periodic wave is superposed
on the tail, the non-dimensional frequency fxR/ U , being approximately 30. This
high-frequency component can be interpreted as the frequency at which the rolled-up
vortices in the shear layer pass through the position of a fixed hot-wire probe. We
assume that the tails of the autocorrelation curves are associated with the low-
frequency flapping motion of the shear layer near the separation line. The flapping
is probably produced by a large-scale unsteadiness in the bubble, the origin and nature
of which are still unclear. The gradual decay of the tails in the more-downstream
sections x/xR > 0.5 may be explained by the cross-sectional dimensions of the
rolled-up vortices, which are much larger than the spatial extent of the flapping of
the shear layer. It is possible that the flapping can change the centre of the large
vortices in the lateral direction, but the velocity fluctuations associated with such
change of the vortex centre are weaker than or comparable t o those caused by the
vortex itself. This is conjectured to be the reason why the flapping is observed most
clearly near the separation line, where the cross-sectional dimensions of the rolled-up
vortices are a t least comparable to the spatial extent of the flapping.
I n order to get an idea about the spatial extent of the flapping, the cross-correlation
between u' and p' was measured in the (x,y)-plane a t many points in the upstream
part of the separation bubble, the surface-pressure transducer being fixed at the
position x/xR = 0.2. Since the flapping frequency is estimated to be less than about
0.2u,/xR, the velocity and pressure signals were passed through a low-pass filter
(O.25um/xR = 50 Hz) before being fed to the signal processor. The result is shown
in figure 21. The cross-correlation is negative and the constant-correlation curves are
approximately parallel to yc. These features are conjectured to be further indirect
evidences of the flapping motion. I n particular, the fact that the cross-correlation is
negative can be shown t o be consistent with the flapping motion of the shear layer
if the quasi-steady approach is employed. When the shear layer moves from its
time-mean position towards the plate surface, the longitudinal velocity a t the edge
Downloaded from https://www.cambridge.org/core. National Cheng Kung University, on 19 Oct 2021 at 09:31:19, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S002211208300230X

Structure of a turbulent separation bubble 103

-0.4L

FIQURE 22. Cross-correlation of surface-pressure fluctuations a t x/xE = 0.2 and 1.0 ( a ) 50 Hz

( = 0.25Um/x,) low-pass filtered; ( b ) 50 Hz high-pass filtered before processing. p i and p k are
pressure fluctuations a t x/xR = 0.2 and 1.0 respectively.

increases because of an increased curvature of the shear layer near the separation line.
Since the pressure on the plate surface can be assumed t o be approximately equal
to the pressure a t the edge ofthe shear layer, the surface pressure then becomes lower
than the time-mean value. Thus we have u‘ > 0 and p’ < 0. On the other hand, when
the shear layer moves outwards from its time-mean position, a decrease in the
curvature of the shear layer leads to u’ > 0 and p’ < 0.
The negative cross-correlation can be interpreted in another way. I n the irrotational
region the unsteady form of Bernoulli’s equation yields
plUl= --p(lJm-Uc)P,
where higher-order terms are neglected and U, is a representative convection velocity
of u’ (Bradshaw 1967). If we assume with Bradshaw that the surface-pressure
fluctuation did not differ markedly in character from the pressure fluctuation near
the edge of the turbulent region, the above equation predicts negative cross-correlation
between u’ and the surface-pressure fluctuation. The result of figure 21 is consistent
with this prediction. It should be emphasized, however, that the source of the
fluctuations in this case is the flapping of the separated shear layer.
Figure 22 shows the cross-correlation of the surface-pressure fluctuations at two
largely separated sections x/xR = 0.2 and 1.0; the signals were low-pass filtered
(50 Hz) before processing. The cross-correlation is negative over a wide range of the
time lag. The phase # p ; p k of the pressure fluctuations p; and p k at x/xR = 0.2 and
1 .O respectively are presented in figure 23, which shows that the phase is - 7t in the
frequencyrange fxR/UooQ 0.2.This corresponds to the fact t h a t p i a n d p k are mostly
of the opposite sign. A steplike change of the phase a tfxR/ U , z 0.2 may again indicate
that the flapping of the shear layer occurs with frequencies less than about
fxR/ u, z 0.2. Moreover, the cross-correlations between u’ at a fixed position
Downloaded from https://www.cambridge.org/core. National Cheng Kung University, on 19 Oct 2021 at 09:31:19, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S002211208300230X

104 M . Kiya and K . Sasaki

fXRIU..

0 0.25 0.50 0.75 1 .oo 1.25

of I I I I I

FIGURE
23. Phase of surface-pressure fluctuations p t and p k at x/xR = 0.2 and 1.0 respectively
(solid line for visual aid only).

( x / x R = 0.2,y/xR = 0.15) near the separation line and p’ a t various longitudinal

positions were negative for x/xR 5 0.6 and positive for 0.6 5 x / x R 5 1.2.
These features suggest the existence of a large-scale unsteadiness in the separation
bubble. As mentioned above, positive p’ at the position x / x R = 0.2 implies that the
shear layer has moved outwards from its time-mean position. The negative cross-
correlation shows that this is accompanied by negative p‘ at a distant position
x / x R = 1 .O. On the other hand, the inward movement of the shear layer is accompanied
by negative and positive p’ at X / X ~= 0.2 and 1.0 respectively. Employing the
quasi-steady approach, the position x / x R = 1.0 can be said to be located upstream
of the instantaneous reattachment line or within the instantaneous separation bubble
when p’ at the same position is negative. If this position is located downstream of
the instantaneous reattachment line or outside the instantaneous separation bubble,
then we expect positive p’ at x / x R = 1 .O. Accordingly, the flapping of the shear layer
is closely related to the shrinkage and enlargement of the separation bubble. It should
be emphasized that this large-scale unsteadiness is different from the smaller-scale
unsteadiness caused by the previously mentioned ‘regular ’ vortex shedding from the
bubble; the representative frequency of the former is much lower than that of the
latter by a factor of roughly 6.
We imagine that very weak regular vortex shedding occurs from the main body
of the separation bubble, and as a result a large amount of vorticity is accumulated
within the bubble. This is accompanied by a considerable increase of the bubble length
and the outward motion of the shear layer. When a sufficient amount of vorticity
is accumulated, an extremely large vortex is eventually shed from the bubble; the
bubble then rapidly shrinks and the shear layer moves inwards. The same feature
has recently been observed in the discrete-vortex simulation of the separation bubble
(see figure 24) and also in Hillier & Cherry’s (1981b) smoke photograph (figure 2 of
their paper).
Downloaded from https://www.cambridge.org/core. National Cheng Kung University, on 19 Oct 2021 at 09:31:19, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S002211208300230X

Structure of a turbulent separation bubble 105

0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0
x/xR

FIGURE 24. Shedding of an extremely large vortex from the separation bubble simulated by the
discrete-vortex model (Kiya et al. 1982). Line segments show paths of elemental vortices during
a short time interval.

X/xR

FIGURE 25. Cross-correlation of surface pressures Rp,pfand that of surface pressure and longitudinal
velocity R,,,,:a, R,.,. (Hillier & Cherry 1981a); 0, - R,,,, (present result). The vertical arrow
& shows the longitudinal coordinate of the fixed pressure transducer and hot-wire probe. The y
coordinate of the hot-wire probe is y/xR = 0.4.

3.9. Cross-correlations between surfuce-pressure and velocity

fluctuations ;further measurements
The surface-pressure fluctuations beneath the separation bubble are closely related
to the velocity fluctuations associated with the large-scale vortices existing above the
surface. The cross-correlations between the surface-pressure and velocity fluctuations
can thus give some aspects of the large-scale vortices in the separation bubble.
I n order t o show supporting evidence for this conjecture, figure 25 compares the
cross-correlation between the pressure fluctuations at two positions separated in the
longitudinal direction (Hillier & Cherry 1981a) with that between u’a t a fixed point
in the outer part of the bubble and the pressure fluctuations a t various longitudinal
positions. The x-coordinates of the fixed pressure transducer in the former case and
the hot-wire probe in the latter were chosen as x/xR = 1.0, the y-coordinate of the
hot wire being yB. Since Rpfu,is negative when u’is taken in the outer part of the
bubble (see figure 26), - R p f u , is presented in figure 25. The fact that the two
correlation curves are fairly similar demonstrates that the correlation Rpfufcan be
used to find the large-scale vortex structure in the separation bubble, probably except
the vicinity of the separation line where the rolled-up vortices are somewhat remote
from the surface.
The profile of R,.,, gives an idea about an ‘averaged’ spatial waveform of the
fluctuating surface pressure. The reasoning is as follows. When u’ measured a t the
position (xR,yB) (temporarily denoted by uLR) is positive there exists a large vortex
beneath ye as demonstrated by KSA. At the same instant the surface-pressure
Downloaded from https://www.cambridge.org/core. National Cheng Kung University, on 19 Oct 2021 at 09:31:19, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S002211208300230X

106 M . Kiya and K . Xasaki

0.z -9 v A - 4 -

0.
Y
XR

0.

0.

R p'u', R p y
FIGURE 26. Cross-correlation coefficients of surface-pressure and velocity fluctuations measured a t
the reattachment section x/xE = 1.0: 0, R,,,,(split-film result); 0 , R,,,,(hot-wire result); 0,
R,,,,(Hillier & Cherry 1981a , hot-wire result); A, R,,,,(split-film result); A,R,.,.(hot-wire result)
(solid lines for visual aid only).

fluctuationp' is negative a t x,. In the statistical sense, when uLRattains a maximum,

it is reasonable to assume that the position (xR,yB) is just above the vortex centre,
so that p' a t X, takes a negative minimum. p' is positive on both sides of the vortex,
and thus a maximum of p' is expected to appear a t the middle of two consecutive
vortices. Accordingly the cross-correlation Rpfurbetween ukR and p' at various
longitudinal positions takes a negative minimum when p' is measured beneath the
vortex centre and a positive maximum when p' is taken a t the middle of the two
consecutive vortices. I n this sense the cross-correlation Rpfurgives an information
about the spatial waveform of the surface-pressure fluctuation, which is the average
over a large number of realizations.
The correlation is negative beneath the position where u' is measured, and becomes
positive on both sides of the negative region. The distance between the peak and
valleys of - RPlulindicates a rather large distance between the large-scale vortices.
A typical distance is approximately O.7xR, which is close to the distance 0:8xR
evaluated from the phase velocity Uep1and the frequency O . ~ U , / X R at which the
vortices pass through a fixed position in the reattaching zone. Hillier & Cherry (1981a )
also estimated the distance to be 0.62,.
The cross-correlation R,!,.becomes positive when u' is measured in the lower part
of the bubble as shown in figure 26. The fact that the correlation is negative in the
outer part of the bubble can be explained in terms of the unsteady form of Bernoulli's
equation: see the discussion of figure 21. The cross-correlation Rprurbetween the
surface-pressure and lateral-velocity fluctuations is also included in this figure. The
signs of Rp,ufand R,.,, are generally such that the Reynolds shear stress - p U " is
positive.
We can construct from the correlations Rprufand R,,,, a vector whose x- and
y-components are the former and the latter respectively. The result is shown in figure
27, where the pressure transducer was fixed a t x/xR = 1.O and the X-wire probe was
Downloaded from https://www.cambridge.org/core. National Cheng Kung University, on 19 Oct 2021 at 09:31:19, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S002211208300230X

Structure of a turbulent separation bubble 107

0'5 i

I I . . . . .
"0.7 O k Ol9 !I IIl 1(2 113 1(4 115 l%
X/XR

FIQURE 21. Vector fields of the cross-correlation coefficients R,.,, and R,.,.. The x- and y-components
of the vector are the former and latter respectively. Thick arrow shows the position of fixed pressure
transducers, which is x / x R = 1 .O.

traversed in its vicinity. The vector fields can give an idea about an 'averaged ' field
of the velocity vector (u', w') in the large vortices by the same reasoning as explained
before. The direction of arrows in figure 27 is arranged in such a way that the
surface-pressure fluctuations are negative. The vector field suggests the existence of
an inclined vortex rotating in the clockwise sense, whose downstream side is nearer
to the surface than its upstream side. The same measurement was also made by fixing
the pressure transducer a t x / x R = 0.6. A comparison between this result and figure
27 revealed that the height of the centre of the vortices and the longitudinal distance
between them were a little larger in the downstream region than in the upstream
region.
Figure 28 shows the loci of Rplu< = 0 and Rplv,= 0 in the reattaching zone. These
loci can be interpreted as the lines on which the velocities u'and v' respectively are
zero on the average. Tangents were drawn to the curves a t the point of intersection,
and ellipses of the same eccentricity were fitted in such a way that tangents to the
ellipses a t points of intersection between the former tangents and the ellipses were
parallel to the x- and y-axes. I n doing this, we assumed that the centre of the ellipses
coincides with the vortex centre and that their axes are the bisectors of the first
tangents. The centre of the ellipses may not exactly be the vortex centre owing to
the effect of image reflection in the surface. This is not serious because we are
interested only in a rough picture of the vortex. The resulting ellipses are included
in figure 28. The major axis of the ellipses inclines by an angle of about 4 5 O to the
longitudinal direction, the minor radius being about 0.7 times the major radius.
Figure 29 shows the cross-spectrum, phase and coherence of the surface-pressure
and velocity fluctuations in the reattachment section. The height of the hot wire from
the surface was fixed a t y6.The cross-spectrum and coherence have a somewhat broad
but clear peak at a frequency near f x R / U m = 0.6. This frequency is again the
frequency a t which the large-scale vortices pass through the reattachment section.
3.10. A three-dimensional aspect of the separation bubble
The flow in the bubble has thus far been described from two-dimensional aspects.
Since the surface-pressure fluctuations are well correlated with the large-scale
Downloaded from https://www.cambridge.org/core. National Cheng Kung University, on 19 Oct 2021 at 09:31:19, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S002211208300230X

108 M . Kiya and K . 8asaki

0
0.80 0.85 0.90 0.95 1.00 1.05 1.10 1.15 1.20
XJXR

FIGURE 28. Qualitative picture of large-scale vortices in the reattaching zone. Ellipses give an idea
about streamlines in the large-scale vortices. ---, tangents to R,!,,= 0 and R,,,,= 0; ---,
bisector of the tangents (major axis of the ellipses). The vertical arrow J. shows the location of
the pressure transducer.

vortices, it is expected that the cross-correlation coefficient Rpfwfof the surface-

pressure and transverse-velocity fluctuations can yield some information about
three-dimensional features of the vortices in the separation bubble. Figure 30 shows
the cross-correlation measured along several longitudinal lines in the plane
z/zR = 0.35. The pressure fluctuation was always taken a t x/zR = 1.0, and is thus
denoted by p k in this figure. The distributions of the correlation along the corres-
ponding lines in the opposite plane z / x R = - 0.35 were approximately antisymmetrical
to those shown in figure 30. The sections z/xR = k0.35 were chosen because the
correlation RpkWr, which was measured along several spanwise lines in the reattaching
zone by fixing the pressure transducer a t the midspan, took a maximum or a
minimum a t z/xR = k 0.35, such as shown in figure 31. The location where Rpbw,= 0
approximately falls on straight lines L, and L, which have an inclination of about
100' and 56' respectively with respect to the longitudinal direction. The loci L, and
L, were also found when the same measurements were made in the plane z/xR = 0.25
with the pressure transducer fixed a t x/xR = 0.6: the inclination was about 110' for
I,, and 83' for L,. The sections z/xR = kO.25 were chosen by the same reason as for
Z / X R= k0.35.
The cross-correlation measurements were also made along three typical lines A , B
and C (see figure 30) parallel to the z-axis. The three lines lie in a plane normal to
the straight line L,. The result is presented in figure 31. The correlation is almost zero
Downloaded from https://www.cambridge.org/core. National Cheng Kung University, on 19 Oct 2021 at 09:31:19, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S002211208300230X

Structure of a turbulent separation bubble 109

”
1.5 2.0 0.5 1.0 1.5 2.0

fX,lU.. fXRIU..

FIGURE 29. Cross-sPectrum lEp,RU’Ir lEp,Rd,Phase $ p , , u,,

$prRV,and coherence coh2pCRU,,cob;, uI
of surface pressure and velocity fluctuations in the reattachment section x/xE = 1.0. h e
y-coordinate of the hot wire is ys. The vertical scales for IEPrRu$and lEPrR,,$ are linear but
arbitrary.

along the line B but of the opposite sign along the lines A and C. Moreover, its
distributions along A and C are approximately antisymmetrical with respect to the
midspan. These features suggest the existence of a counter-rotating system shown
in figure 31 : the spanwise distance between centres of the system is estimated to be
0 . 6 ~The
~ . distance was roughly 0 . 5 when ~ ~ the surface-pressure fluctuation was
measured a t = 0.6.
It should be mentioned that, since the aspect ratio of the separation bubble was
2.0xR in this experiment, the distributions of the cross-correlation RpSwfwere
possibly affected by the sidewalls of the air tunnel. Nevertheless we conjecture that
the general feature of its distributions is well represented by the present result.
At present we are unable to establish interrelations of the structure suggested by
figures 30 and 31 and that of figures 27 and 28. Studies of the three-dimensional
feature of large vortices in the separation bubble are now in progress in our laboratory
by means of flow-visualization techniques and conditional-sampling measurements.
A result is presented in figure 32, which is a flow-visualization photograph taken a t
a Reynolds number of 630 with an aspect ratio of 20. Fairly periodic spanwise
structures are observed, a t a Reynolds number lower, by two orders of magnitude,
than that of the air-tunnel experiment. We suggest that the structures of large
vortices do not differ markedly in character in the two cases. The spanwise distance
Downloaded from https://www.cambridge.org/core. National Cheng Kung University, on 19 Oct 2021 at 09:31:19, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S002211208300230X

110 M . Kiya and K. Xasaki

.FIGI~W30. Cross-correlation between surface-pressure and transverse-velocity fluctuations mea-

sured in the plane z / x n = 0.35 along several lines parallel to t>hez-axis. The surface pressure pk
was measured a.t z/xR = 1.0, as shown by the longest arrow. 0, y/xR = 0.4;0 , 0.3; 0, 0 . 2 ; A,
0.1, Coordinates of A , B , and C (denoted by short arrows) are (x/zR,y/xn) = (1.20,0.30),(1.35,0.20)
and (1.50.0.10)respectively. ---, lines approximately passing through points of R p k w ,= 0.

0.4 1 I 1 1 I I
-0.6 -0.4 -0.2 0 0.2 0.4 I 6
zlx,
FIN-RE31. Cross-correlation between surface-pressure and transverse-velocity fluctuations mea-
sured along lines parallel t o the z-axis which pass through the points A , B and C in figure 31 : 0,
A ; A, B ; 0 ,C. The circles are schematic representations of the structure in the plane which includes
A , R and C, viewed from above. The direction of rotation corresponds to an instant when p k is
negative.

between two neighbouring structures was roughly 0 . 7 in~ the ~ reattaching zone when
an average was taken for several similar photographs. This is close to the distance
0.62, estimated from figure 31. The phase coherence of the spanwise structure is
p r h a p s lost in the air-tunnel experiment by the effects of freestream turbulence and
rloir;r in initial conditions at separation.
Downloaded from https://www.cambridge.org/core. National Cheng Kung University, on 19 Oct 2021 at 09:31:19, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S002211208300230X

Structure of a turbulent separation bubble 111

FIQURE
t
32. Photograph of the flow in the (x,2)-plane. Flow is from left t o right. Visualization by
hydrogen bubbles, which were produced by a thin platinum wire placed on the plate surface just
downstream of the separation line. f,time-mean reattachment position; Reynolds number = 630;
aspect ratio = 20.

4. Conclusions
The flow in the separation bubble formed along the sides of a blunt flat plate with
right-angled corners has been described in terms of various statistical properties of
the surface-pressure and velocity fluctuations.
Large-scale vortices are shed downstream from the separation bubble with A
frequency of about 0 . 6 U , / x R . On top of this ‘regular’vortex shedding, there exists
a large-scale unsteadiness in the bubble ; vortices much larger than the regular
vortices are shed with frequencies less than about 0.2U,/xR, the central frequency
being about 0.12U,/xR. The large-scale unsteadiness in the bubble is accompanied
by a flapping motion of the shear layer near the separation line. The authors
conjecture that the regular vortex shedding is hindered by still unknown mechanisms,
and as a result a large amount of vorticity is accumulated within the bubble. This
accumulation will lead to a large increase of the bubble length. The accumulated
vorticity is eventually shed downstream as an extremely large vortex to suddenly
reduce the bubble length to a great extent. The formation and shedding of such a
large vortex are observed in Hillier & Cherry’s (1981b ) smoke photograph and also
in the discrete-vortex simulation. The detailed nature of this large-scale unsteadiness
in the bubble well deserves a further investigation.
The cross-correlations between the surface-pressure and velocity fluctuations were
found to be useful means for the study of large-scale vortex structure in the bubble.
This is because the vortices are strong enough to produce large pressure fluctuations,
which can serve as a ‘reference signal‘ to sort out the velocity fluctuations associated
with the large-scale vortices. This situation in turn means that the velocity
fluctuations a t appropriate positions in the bubble can be used to obtain some
information about the instantaneous distribution of the fluctuating surface pressure.
The distance between the large-scale vortices estimated from such velocity-pressure
cross-correlations is in fair agreement with the distance 0 . 8 evaluated
~ ~ from the
phase velocity of the vortices and the frequency a t which the vortices pass through
a fixed position.
Downloaded from https://www.cambridge.org/core. National Cheng Kung University, on 19 Oct 2021 at 09:31:19, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S002211208300230X

112 M . Kiya and K . Sasaki

The spatial distributions of the cross-correlation coefficients Rpfa,and Rprv( give
an idea about the vortex structure in the reattaching zone, where the average centre
of the vortices is estimated to be a t about 0.2xR from the surface and the average
longitudinal distance between the vortices is (0.7-0.8) xR.The loci on which Rpfur and
R,.,. are zero can be interpreted as those on which u’ and v‘ respectively are zero
on the average. The position where they intersect can be taken as the vortex centre.
By fitting ellipses of the same eccentricity to the loci, it was found that the major
axis of the ellipses makes an angle of about 45O with the longitudinal direction in such
a way that the downstream side of the ellipses is nearer to the plate surface than their
upstream side. Moreover, the minor diameter of the ellipses is about 0.7 times the
major diameter.
An aspect of the unsteadiness in the separation bubble was obtained in terms of
the spatial distributions of the reverse-flow intermittency and the frequency of
switching of the flow direction. The switching frequency is maximum along the line
on which the reverse-flow intermittency takes the value of 0.5.
The approximately linear increase of the integral timescales along the edge of the
separated shear layer indicates that the coalescence or amalgamation of rolled-up
vortices occurs successively upstream of the reattaching zone. The timescales are
almost constant downstream of the reattachment line ; this suggests that the shed
vortices travel downstream without significantly changing their structure within a
certain distance after the reattachment.
The spanwise distributions of the cross-correlation between the surface-pressure
and transverse-velocity fluctuations suggested the existence of a longitudinal counter-
rotating system of fluid. The spanwise distance between the axes of the rotation is
estimated to be roughly 0 . 6 ~The~ . velocity fields conditionally averaged on the basis
of the surface-pressure waveforms can produce more detailed flow structure in the
separation bubble. Studies are now in progress and will be published in a forthcoming
paper.

This work was financially supported by the Grant-in-Aid for Scientific Studies from
the Ministry of Education, Science and Culture of Japan. We also thank Mr
E. P. Sutton, Dr J. C. R. Hunt (University of Cambridge, United Kingdom), Mr
H. Tamura and Dr Y. Suzuki for their valuable comments during the course of this
work. We are grateful for referees’ helpful comments.

REFERENCES
ANTONIA,R. A. & VAN ATTA, C. W. 1977 Statistical characteristics of Reynolds stresses in a
turbulent boundary layer. AIAA J . 15, 71-75.
BRADSHAW, P. 1967 Irrotational fluctuations near a turbulent boundary layer. J . Fluid Mech. 27,
209-230.
BRADSHAW, P. & WON@, F. Y. F. 1972 The reattachment and relaxation of a turbulent shear layer.
J . Fluid Mech. 52, 113-135.
CHANDRSUDA,C. & BRADSHAW, P. 1981 Turbulence structure of a reattaching mixing layer. J .
FZuid Meeh. 110, 171-194.
EATON, ,J. K. & JORNSTON, J. P 1982 Low frequency unsteadiness of a reattaching turbulent shear
layer. In Turbulent Shear E’lows 3 (ed. L. J. S. Bradbury, F. Durst, B. E. Launder, F. W.
Schmidt & J. H. Whitelaw), pp. 162-170. Springer.
FERNHOLZ, H H . 1978 External flows. In Turbulence, 2nd edn (ed. P. Bradshaw), p. 45. Springer.
FIEDLER, H. & HEAD,M . R. 1966 Intermittency measurements in the turbulent boundary layer.
J. Fluid &frrh. 25. 719-735.
Downloaded from https://www.cambridge.org/core. National Cheng Kung University, on 19 Oct 2021 at 09:31:19, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S002211208300230X

Structure of a turbulent separation bubble 113

HILLIER, R. & CHERRY, N. J. 1981 a Pressure fluctuations under a turbulent shear layer. In R o c .
3rd Turbulent Shear Flow Symp., Davis, California, 9-11 September, pp. 16.23-16.28.
HILLIER, R. & CHERRY, N. J. 1981 b The effects of stream turbulence on separation bubbles. J .
Wind Engng & Ind. Aerodyn. 8, 49-58.
JOHNSTON, J. P. 1978 Internal flows. In Turbulence, 2nd edn (ed. P. Bradshaw), p. 109. Springer.
KIYA,M., SASAKI, K. & ARIE,M. 1982 Discrete-vortex simulation of a turbulent separation bubble.
J . Fluid Mech. 120, 219-244.
KOMATSU, S. & KOBAYASHI, H. 1980 Vortex-induced oscillation of bluff cylinder. J . Wind Engng
& Ind. Aerodyn. 6, 33s-362.
LANE,J. C. & LOEHRKE, R . I . 1980 Leading edge separation from a blunt plate at low Reynolds
number. Trans. ASME I: J . Fluids Engng 102, 494496.
OTA,T., ASANO,Y. & OKAWA, J. 1981 Reattachment length and transition of the separated flow
over blunt flat plates. Bull. J S M E 24, 941-947.
OTA, T. & ITASAKA, M. 1976 A separated and reattached flow over a blunt flat plate. Trans.
A S M E I : J . Fluid Engng 98, 7%86.
OTA,T. & NARITA, M. 1978 Turbulence measurements in a separated and reattached flow over
a blunt flat plate. Trans. A S M E I : J . Fluids Engng 100, 224-228.
WILLMARTH, W. w. & WOOLDRIDGE, c. E. 1963 Measurements of the correlation between the
fluctuating velocities and the fluctuating wail pressure in a thick turbulent boundary layer.
A G A R D Rep. 456.
WOOD,D. H . & BRADSHAW, P. 1982 A turbulent mixing layer constrained by a solid surface. Part
1 . Measurements before reaching the surface. J . Fluid Mech. 122, 57-89.
WYCNANSKI, I . & FIEDLER, H. E. 1970 The two-dimensional mixing layer. J . Fluid Mech. 41,
327-361.