J . Fluid Mwh. (1986). vol. 165, p p .

163-199 163
Printed in Great Britain

A theoretical and experimental study

of wall turbulence
By A. E. PERRY, S. HENBEST A N D M. S. C H O N G
University of Melbourne, Department of Mechanical Engineering,
Parkville, Victoria 3052, Australia

(Received 6 November 1984 and in revised form 5 October 1985)

In this paper the dimensional-analysis approach to wall turbulence of Perry & Abell
(1977) has been extended in a number of directions. Further recent developments of
the attached-eddy hypothesis of Townsend (1976) and the model of Perry & Chong
(1982) are given, for example, the incorporation of a Kolmogoroff (1941) spectral
region. These previous analyses were applicable only to the ‘wall region’ and are
extended here to include the whole turbulent region of the flow. The dimensional-
analysis approach and the detailed physical modelling are consistent with each other
and with new experimental data presented here.

1. Introduction
The study and description of wall-bounded shear flows has been an active and
challenging field for most of this century. By the end of the 1930s various
phenomenological theories had been proposed which were aimed mainly towards
describing the mean flow. These theories were unconvincing and little or no serious
attempt was made to extend them to include the fluctuating quantities. Izakson
(1937) and Millikan (1939) then developed a dimensional-analysis approach for the
mean flow, based on the existence of a ‘region of overlap’ in which two mean-flow
similarity laws are simultaneously valid, to establish the logarithmic law of the wall
and laws of skin friction in pipes, ducts and boundary layers. This approach is also
based on the study of how experimental data collapse when plotted with different
sets of scaling coordinates. These coordinates are determined from physical con-
siderations and dimensional analysis. Such an approach has little regard for the
detailed physical processes involved and relies only on physical assumptions of a
general nature. However, if successful, it establishes a functional framework for the
correlation of experimental data and further detailed physical modelling must be
consistent with this framework.
The application of this approach to the fluctuating quantities has not met with
the same success as it has with the mean flow, partly because of the difficulties
involved in obtaining accurate experimental data. Perry & Abell (1975) applied the
approach to their broadband streamwise-turbulence data obtained in smooth-walled,
fully turbulent pipe flow. They postulated an ‘inner-flow’ and an ‘outer-flow’ scaling
law for the $/q distribution (where $ is the mean square of the fluctuating
streamwise velocity and U, is the mean wall shear velocity) and deduced that q/q
is a universal constant in the region of overlap, where both laws are simultaneously
valid. This region coincides with that portion of the boundary layer in which the
mean-flow logarithmic law of the wall is valid and the Reynolds shear stress can be
considered to be constant, and will be referred to as the turbulent wall region.
164 A . E . Perry, S. Henbest and M . S. Chong
I n the light of further work Perry & Abell (1977) revised their ideas. By utilizing
their streamwise spectral data (measured in the turbulent wall region) and the physics
of Townsend’s (1976) attached-eddy hypothesis, they proposed certain spectral-
similarity laws, deduced their analytical form in the spectral regions of overlap and
from these laws predicted the distribution of $/U: in the turbulent wall region.
A physical model consistent with the above findings was attempted by Perry &
Chong (1982), hereinafter referred to as PC. This model was based on the flow-
visualization results of Head & Bandyopadhyay (1981) and the attached-eddy
hypothesis of Townsend (1976). They proposed that a wall shear layer is made up
of a forest of attached hairpin, h-shaped or horseshoe vortices inclined in the
downstream direction at approximately 45’ to the wall. Head & Bandyopadhyay
point out that the eddies maintain this angle for some distance as they are convected
downstream. Recent work by Smith (1984), Acarlar & Smith (1984), Moin & Kim
(1985) and Kim (1985) further supports the existence of hairpin vortices in wall
turbulence. For simplicity, PC confined their attention to h-shaped vortices.
Beginning with an isolated h-vortex they showed that because of its image in the
wall the vortex undergoes a stretching process in which the vortex height h increases
approximately uniformly with time and the distance A between the ‘legs’ of the
vortex at the wall decreases such that the product Ah remains constant (a plane-
strain-like motion in the plane of the vortex). They also showed that viscous diffusion
ultimately dominates the stretching process and proposed that when the legs of the
h-vortex eventually come together the vortex dies by vorticity cancellation. A
random array of h-shaped vortices, all a t different stages of stretching but with the
same circulation, was called a ‘hierarchy’. They found that, in order to obtain a
logarithmic mean-velocity distribution, a region of constant Reynolds shear stress
and the correct u, spectral behaviour in the turbulent wall region, it is necessary to

-
assume that a range of scales of geometrically similar hierarchies exist. The simplest
assumption is that all hierarchies have the same velocity scale ( U,)f and that their
lengthscale 6 varies. The probability density function (p.d.f.) of hierarchy scale 6
follows a geometric progression for a discrete distribution and an inverse power law
if a continuous distribution is assumed. The lengthscale of the hierarchies vary from

-
the smallest scale 6,,which is assumed to be proportional to the Kline et al. (1967)
scaling (i.e. 6, v/U,, where v is the kinematic viscosity), to the largest scale A,,
which is assumed to scale with the shear-layer thickness. How these hierarchies form
is a mystery but one possible explanation suggested by PC is vortex pairing: the
eddies in the smallest hierarchy form from a roll-up of viscous-sublayer material and,
although most of the eddies in this hierarchy die, some manage to pair to form eddies
belonging to the next hierarchy; most of the eddies in the next hierarchy die but some
manage to pair to form eddies belonging to the next hierarchy and so on. PC assumed
a discrete p.d.f. of hierarchy lengthscales that doubled from one hierarchy to the next,
and suggested that the effect of ‘ jitter’ or randomness about each discrete hierarchy
scale gives a continuous inverse-power-lawp.d.f. Their proposals are consistent with
those of Townsend (1976), who effectively replaced each hierarchy with a group of
identical ‘representative eddies ’ and assumed an inverse-power-law p.d.f. for the
representative eddy lengthscales in his broadband turbulence-intensity analysis.
Acarlar & Smith (1984) have shown that the speculated pairing process does in fact
occur, at least for hairpin-type vortices formed in the region behind rivet heads.
Whether such pairing occurs in turbulent boundary layers is an open question.
PC’s model was successful in linking the mean-flow similarity laws with the

Throughout this paper - means ‘proportional to’ or ‘scales with’.

 Theoretical and experimental study of wall turbulence 165
streamwise turbulence spectra and the broadband turbulence-intensity distributions
in the turbulent wall region. However, no account was taken of the possibility that
the attached eddies are surrounded by fine-scale isotropic eddies, which gives rise
to a Kolmogoroff (1941) spectral region, which includes an inertial subrange if the
Reynolds number of the flow is sufficiently large. Unfortunately, PC used the data
of Perry & Abell(l977) in which a Kolmogoroff region with an inertial subrange was
not obvious. Here we present strong experimental support for the existence of such
a region and attempt to explain its existence and how it fits in with PC’s model. We
begin in $2 by extending the spectral analysis of Perry & Abell (1977) to three
dimensions. In $3, the model of PC is improved and extended to include the ‘wake’
region (i.e. the region beyond the turbulent wall region). The analysis establishes a
link between low-wavenumber spectra of motions parallel to the wall in the turbulent
wall region and the mean-flow deviation from the logarithmic law of the wall in the
wake region. The outcome is a model which the authors consider to be applicable to
the whole turbulent region in zero-pressure-gradient boundary layers. With slight
modifications it should also be applicable in pipe and duct flow. Experimental spectral
and turbulence data obtained in smooth-walled fully developed pipe flow are
presented in $4 and these give strong support to the model of wall turbulence
presented here.

2. A dimensional-analysis approach to wall turbulence

We will consider here some examples of turbulent wall-shear flow over a smooth
surface. Much of the analysis is applicable to a slowly developing turbulent boundary
layer on a flat wall with zero streamwise pressure gradient, to fully developed
turbulent flow in a circular pipe and to flow in a high-aspect-ratio rectangular duct.
Let A , be a characteristic lengthscale of the outer part of the shear layer. This would
scale with the boundary-layer thickness, pipe radius or duct half-width. Let the
symbol represent the velocity at the edge of the boundary layer, the mean
velocity at the pipe axis or at the plane of symmetry of the duct. Let z be the
streamwise coordinate, y the cross-stream distance, z the distance normal to the wall
and let the corresponding velocity components be U,, U , and U,. Overbars will denote
mean values and lower-case letters will denote fluctuating quantities.
It is well known that in a region
- close to the wall, the mean flow follows Prandtl’s
law of the wall

The function f, is universal and independent of the large-scale flow geometry (i.e.
independent of whether we have boundary layer, pipe or duct flow). For the fully
turbulent region (i.e. the region of flow beyond the buffer zone) the mean flow follows
the velocity-defect law - -
u,
ulE-ul =fa[;]. (2)
The function f, is universal only for a given large-scale flow geometry. Millikan (1938)
assumed ‘that there is possibly a small but finite region near the wall in which both
(1) and (2) are valid . . . ’ and deduced that within the region of overlap
166 A . E . Perry, S. Henbest and M . S. Chong
by equating the velocity gradients given by (1) and (2). Here K and A are universal
constants and the constant B is dependent on the large-scale flow geometry.
Constants dependent on the large-scale flow geometry will henceforth be referred to
as ‘large-scale characteristic constants ’.
It will be assumed that the mean-flow vorticity and the energy-containing
turbulent motions, which include the Reynolds-shear-stress motions, are caused by
anisotropic coherent eddies attached to the wall in the sense of Townsend (1976,
pp. 152-3) and PC. Many definitions for the terms coherent structure or coherent eddy
exist (see e.g. Hussain 1982). The present authors regard coherent attached eddies as
having similar recognizable patterns with a fixed angle of inclination relative to the
wall (i.e. 9 = constant as shown in figure 1) that recur throughout the flow with a
range of lengthscales. These coherent eddies are assumed to be surrounded by a fluid
which contains fine-scale detached eddies. The motions in this surrounding fluid are
assumed to be statistically isotropic and statistically irrotational. These fine-scale
eddies contribute little to the broadband turbulence intensities and make no
contribution to the Reynolds shear stress but are responsible for most of the energy
dissipation in the flow. They are possibly the remainder of what were once attached
eddies that have been stretched, distorted and convected away from the near-wall
region by the more ‘active ’ attached eddies, and are therefore the ‘debris of dead-eddy
material ’ (see PC regarding eddy death). Many of the above physical assumptions
are not essential for the following dimensional arguments but are consistent with it
and will be used in a more detailed physical argument presented in $3.
The following analysis is applicable only to flow in the turbulent wall region (i.e.
v/ U, 4 z < A , ) . Figure 1 shows three different scales of attached eddies together with
the instantaneous streamline patterns that they generate relative to an observer
moving with the fluid in the far field of the eddy. These are identical with the
three-dimensional A-shaped vortices of PC (other eddy geometries are considered
later). Since we are considering the flow beyond the thin viscous sublayer, we will
assume that the flow is inviscid with finite slip at the boundary. Also shown in figure 1
is a probe situated at a distance z from the wall. It is not too difficult to see that
an eddy of scale S , = O(A,) will contribute to u, and u2at the probe location and
these contributions will be invariant with z for z 4 A,. This invariance of u1and of
u2with z is meant in the sense that, if we Taylor-series expand u, or u2with respect
to z about z = 0 for an eddy of scale S, then, for z 4 S, the zeroth-order term will
dominate over the higher-order terms; while for u3the first-order term will dominate.
It is clear that an eddy of scale 8, will contribute little to us.An eddy of scale 6, = O(z)
will contribute strongly to u,, u2and us and these motions will depend on z. Eddies
of scale 8, 4 z will not contribute to any motions at z because of the very small far-field
effect above the eddy (see the BioMavart-law calculations of PC). Thus, only eddies
of scale 6 = O(z) contribute to u3motions and all eddies of scale 6 >, O(z)contribute
to u, and u2motions at z.
Let @(,(k,) be the cross-power spectral density per unit streamwise wavenumber
k, for velocity fluctuations u2and ujand let this be normalized such that

joW @&l) dk, = “r? (5)

where i and j may equal 1, 2 or 3, and repeated indices do not denote a summation.
Let us consider the distribution of the u1 power-spectral density, @,,(k,), in the
turbulent wall region and only that range of wavenumbers in which motions are not
dependent explicitly on viscosity. This would cover most of the energy-containing
 Theoretical and experimental study of wall turbulence 167

LX View X-X

FIGURE
1. A sketch of three attached eddies of varying scales together with
the instantaneous streamline pattern generated by each.

region of the spectrum since the viscosity-dependent motions would occur only at
very high wavenumbers. Therefore in the energy-containing region the only variables
involved are U,, k,, z and A , (this is in accordance with Townsend’s 1976
Reynolds-number-similarity hypothesis). At z << A,, eddies of scale S = O ( A , ) will
contribute only t o the low-wavenumber motions, so a t these wavenumbers we would
expect an ‘ outer-flow ’ scaling law of the form

which from earlier discussions is known to be invariant with z for z << A , . Here
Gl1(kl AE) is the power-spectral density per unit non-dimensional wavenumber k, A,.
From now on the argument of a spectral function will denote the unit quantity over
which the energy density is measured. Eddies of scale 6 = O(z) will contribute to
motions a t moderate t o high wavenumbers, so a t these wavenumbers we would expect
an ‘inner-flow ’ scaling law of the form

A , is not involved in (7) since eddies of that scale make no contribution to this
wavenumber range provided that z << A , .
Let us now consider the very-high-wavenumber viscosity-dependent motions. We
would expect these motions t o be locally isotropic and the u1 spectra to follow the
classical Kolmogoroff (1941) viscosity-dependent scaling law

where 7 = (v3/e):and u = ( ~ 6 ) : . The quantities 7 and u are the Kolmogoroff length

and velocity scales respectively, and these depend only on the turbulent-energy
dissipation e and the kinematic viscosity v. Following Townsend (1961, 1976), a
reasonable assumption concerning the turbulent wall region is that turbulent energy
production p and dissipation are approximately in balance,? i.e.

-
t This contradicts the Townsend (1976) attached-eddy hypothesis. I t is simple to show for flow
consisting of attached eddies alone that p / e z+ in the turbulent wall region. The remainder of
the dissipation is carried by fine-scale detached eddies surrounding the attached eddies.
168
I n this region, -- A . E . Perry, S. Henbest and M . S. Chong
=
ships i t can be shown that
and Clq/az is obtained using (3). From these relation-

and ?#I=
K-3
- .

Figure 2 ( a )summarizes the various u, spectral regions given by (6),( 7 )and ( 8 )over
the range of wavenumber k,. Two regions of overlap are anticipated, shown as
‘overlap region I’ and ‘overlap region 11’.I n region of overlap I, (6) and ( 7 ) are
simultaneously valid, i.e. GlI(kl)= z q g 2 ( k l z ) = d, U;g,(k,dE) and therefore
gl(k,dE)/g2(klz)= z/d,. Hence g1 and g2 must be of inverse-power-law form as
given below :
(1la)

or

where A , is a universal constant. I n region of overlap 11, ( 7 )and ( 8 )are simultaneously

valid. Substituting (10a) and ( l o b ) into ( 8 ) and comparing this equation with ( 7 )
shows that v is not explicitly involved. The only functional form that will permit this
is

or

where KOis the universal Kolmogoroff constant. Region of overlap I1 is sometimes

called the inertial subrange. The functional forms given by ( 11) and (12) are the only
ones that simultaneously satisfy the two laws applicable in each region of overlap. The
boundaries of these regions of overlap are indicated in figure 2 ( a ) .The constants P ,
N and M are universal constants and F is a large-scale characteristic constant. The
relationships for these boundaries can be derived from (6), ( 7 ) , ( 8 )and ( l o b ) .It will
be shown later that there is encouraging experimental evidence for the existence of
these anticipated overlap regions. Equation (11) was deduced, using the above
dimensional argument, by Perry & Abell (1977).
The analysis for the u2motions yields a similar set of relations with an additional
set of universal constants, some of which can be related to the u1 spectral constants
in the Kolmogoroff region. For the us motions, one would expect from earlier
discussions that, no ‘outer-flow ’ scaling law should result, and the various u3spectral
regions are shown in figure 2 ( b ) with one region of overlap between an ‘inner-flow’
scaling law and the Kolmogoroff scaling law. Again, the ugspectral constants in the
Kolmogoroff region can be related to the u1 constants.
Sketches of the expected u1 spectral distribution plotted with ‘inner-flow ’ and
‘outer-flow ’-scaling coordinates are shown in figures 3 and 4 respectively. With
‘inner-flow’ scaling the spectra at low k , z should peel off’ at k , z = F z / A , from an
inverse-power-law region (11a ) ,and, a t very high k, z , peel off a t k, z = M ~ - f ( z +from
)f
a -5 power law, (12b). Here z+ = zU,/v. With ‘outer-flow’ scaling the spectrum of
the low-wavenumber motions, which are non-universal as regards to ‘inner-flow ’
 Theoretical and experimental study of wall turbulence 169
(4
-
r
Viscosity-
Motions that are independent of viscosity dependent
(i.e. Townsend’s Reynolds-number-similarityhypothesis)

T motions

-
(‘ Outer-flow’ scaling) (Kolmogoroff scaling)

a b
0 klAE=F klz= P k,z=N klq=M kl
or k, z = MK-+z\

t
L
region I region I1

(‘Inner-flow’ scaling)
(I=+ zu,/4

r
(b)
Viscosity-
Motions that are independent of viscosity dependent -----cc

-T
(i.e. Townsend‘s Reynolds-number-similarityhypothesis)

-- --
motions

@ d k l 7) - h3(kl7)
lJ=

(Kolmogoroff scaling)

(‘Inner-flow’ scaling)
FIGURE2. A summary of the various spectral regions for velocity fluctuations in the
turbulent wall region. (a)u1spectra. ( b ) us spectra..

scaling, should collapse to a universal region at low k, A,. At high k, A , the spectra
should peel off at k, A, = PA,/% from the inverse-power-law distribution. Figures 5
and 6 show the expected u3 spectral distribution plotted with ‘inner-flow’- and
‘ outer-flow ’-scalingcoordinates respectively. With ‘inner-flow ’ scaling the u3spectra
at low to high k , z should collapse to a universal region, which at high k , z follows
a -f power law from which the spectra peel off at k, z = M d ( z + ) f .‘ Pre-multiplied’
170 A . E . Perry, S . Henbest and M . S. Chong

M(z+)f

I
I
log P log N
log (kl z ) log ( k , z )
(4 (b)
FIQURE 3. Sketches of the expected distributions of ( a )u1 spectra and ( 6 ) pre-multiplied u1 spectra
when scaled with ‘inner-flow’-scalingcoordinates for varying values of z / A , and z+ within the
turbulent wall region.

j8
kl dE)
u:

spectra are also shown in the figures on semi-logarithmicplots ;such plots conveniently
show the non-dimensional energy contribution over any wavenumber range as an area
under the curve.
Returning to the u1 spectra, the broadband turbulence intensities can be found by
integrating over the various spectral regions. With reference to figure 2 ( a ) we have

In (13) the first integral equals a large-scale characteristic constant; the second is
evaluated using (11a ) ; the third equals a universal constant ; the fourth is evaluated
 Theoretical and experimental study of wall turbulence 171

FIGURE
5. Sketches of the expected distributions of (a)u3spectra and ( b ) pre-multiplied
u3spectra when scaled with ' inner-flow'-scaling coordinates.

FIQURE
6. Sketches of the expected distributions of (a)u3spectra and ( b ) pre-multiplied
u3spectra when scaled with 'outer-flow'-scaling coordinates.

using (12b) and the energy contributionfrom the last term is assumed to be negligible.
This yields -
U2
2 = &-A,
v [id
In - -C(z+)-i, (14)

where A , and Care universal constants and B, is a large-scalecharacteristicconstant.

Similarly, for the u2 motions,

where A , is a universal constant and B, is a large-scale characteristic constant ;while

for the u, motions -
U2
3 = A, -ic(z+)-i,
v
172 A. E. Perry, S. Henbest and M. S. Chong
where A, is a universal constant. It should be noted that C occurs in all of the above
equations and the factor { appears from the theory of isotropic turbulence applied
to the amplitudes of @, GZ2and @33 in the inertial subrange when expressed in terms
of the streamwise wavenumber k, (see Townsend 1976, p. 93; Batchelor 1956).
Equations (14), (15) and (16) are valid provided v / U , 4 z 4 A,, and in the limit as
z+-+ m the equations reduce to those arrived a t by Townsend (1976) who used
broadband-turbulence-intensity arguments and neglected the fine-scale motions
describable by (8).

3. A more detailed physical model for wall turbulence

3.1. Mean-flow distribution
Much of what follows is based on Townsend's (1976) attached eddy hypothesis and
the model of PC. Consider a representative eddy of scale S similar to those shown
in figure 1 with a characteristic velocity scale U,. Let an array of these eddies
distributed randomly in the (z,y)-plane, with average streamwise and cross-stream
spacings which scale with S,be representative of a hierarchy of scale 6,and let the
contribution that this hierarchy makes to the mean cross-stream vorticity at a
distance z from the wall be cH.
From dimensional analysis it can be shown that

5H = ?.f(;).

Then the mean cross-stream vorticity at a fixed z for a range of geometrically similar
hierarchies varying in scale from 6, to A, is

where pH(&) is the p.d.f. of hierarchy scales. Townsend (1976) assumed p H ( B )to be
continuous and of the form
JtY
pH(S) = S ! (19)

since this and the previous assumptions lead to a region of constant Reynolds shear
stress for S, 4 z 4 A,. Here A is a disposable universal constant. Using (17), (18)
and (19) we obtain
(20)
- -
where U g = ( UIE- U,)/U,, the non-dimensional mean-velocity defect. Also,
h = In (J/z); A , = In ( S , / Z ) ; A , = In (dE/z) and h(A) =f(z/S).
It can be shown that

By assuming that all the vorticity is confined to the vortex loop and that the
surrounding fluid is irrotational, PC were able to relate various forms of h(A)orf(z/S)
to the representative eddy geometry. We shall now examine various forms of h(h)
and compare the resulting U g distributions with the Hama (1956) velocity-defect-law
 Theoretical and experimental study of wall turbulence 173
distribution in a zero-pressure-gradient turbulent boundary layer on a flat plate. This
empirical law in terms of our logarithmic variables is

where the constants K and D are 9.6 and 2.309 respectively. Alternatively, we could
have used the Coles (1956) ‘law-of-the-wall’ and ‘law-of-the-wake’ formulation to
generate the velocity-defect law. Equation (22) is shown as one of the curves plotted
in figure 8.
Figure 7 shows ‘representative eddies’ of various geometries and their f(z/8) and
h(A)e-A distributions, and their resulting Ug distributions are shown in figure 8. All
cases shown lead to the expected logarithmic distribution in U s for A, sufficiently
large. The constant A in (19), (20) and (21) is evaluated using the condition
(dU;S/dAE)-+1 / for~ A, $ 0 and A, < 0. Case (a), a A -shaped vortex, where all the
cross-stream vorticity is confined to the top of the vortex loop, gives a logarithmic
distribution through the layer and its intercept is zero at A , = 0. Case ( b ) ,a h-shaped
vortex, where the cross-stream vorticity is distributed uniformly over the height of
the eddy, gives a negative intercept for the logarithmic distribution at A, = 0.
If we equate dzUg/dAk determined from (20) and (22), it is possible to find the
appropriate h(A) or f(z/8) which yields the Hama distribution in the region
0 < A, < -In (0.15).This gives the ‘bow-legged’parabolic eddy shown in figure 7 (c).
This eddy geometry seems unlikely.
To consider that a hierarchy can be represented by a single realizable eddy is a
simplification. Rather, in accordance with PC, it would be more realistic to regard
a hierarchy as consisting of an assemblage of eddies at different stages of stretching
as shown in figure 7 ( d , e ) . Case ( d ) shows a h-vortex undergoing a plane-strain-like
motion, as was indicated by the Biot-Savart-law calculations of PC. The eddy starts
at a height b8 and is stretched to a height 8,where it either pairs with another similar
eddy to form the shortest eddy ofthe next hierarchy or viscous diffusion and vorticity
cancellation causes eddy ‘death ’. Case ( e )shows an assemblage of n-shaped vortices
at different stages of stretching. The U;E distributions for these two cases for b = 0.5
are also shown in figure 8. For each case we obtain the correct logarithmic distribution
for A, large, but the extrapolated intercept is still negative. From a study of
experimental spectra (presented in 54.2.1) the authors propose that, instead of
varying the eddy geometry, the inverse-power-law p.d.f. should be modified by
increasing the weighting for the large-scale eddies, as shown in figure 9(a). The
large-scale eddies need not be geometrically similar to the smaller-scale attached
eddies, which are responsible for the logarithmic mean-velocity distribution in the
turbulent wall region. However, as a first approximation, it will be assumed here that
the large-scale eddies with this additional weighting are geometrically similar to the
smaller-scale ones. The type of modification shown in figure 9 ( a )will give the correct
type of behaviour for the Hama velocity-defect-law formulation (i.e. a positive
extrapolated intercept). A simplified modification is shown in figure 9 ( b ) and this is
equivalent to multiplying the inverse-power-law p.d.f. with a weighting function
W(d/d,), shown in figure 9(c),such that
174 A . E . Perry, S. Henbest and M . S. Chow

0
Dirac
delta
function

1 5
6
I
I
0
h(A)e-A

Dirac delta
function -
Wall

Y
8
0
‘
0
I 0 A

%1-10 -
I
-
Wall

t5
t
h(A)e-A

0 0 b I f
6 6

v 0 0 b 1 2 0 A
6 6

FIQURE 7. Projections in the (y, 2)-plane of various representative eddy geometries together with
their &/a) and h(A)e-* distributions. (a)n -eddy; ( b ) A-eddy; (c) bow-legged, parabolic eddy;
( d ) stretched A-eddy; (e) stretched n-eddy.
 Theoretical and experimental study of wall turbulence 175

IS I I I I

Ui,

10 -

0 1 2 3 4 5
AE

FIGURE8. The resulting U: distributions for each of the representative eddy geometries shown in
figure 7 compared with the Hama (1954) velocity-defect-law formulation. Caae (f) is for a stretched
n-eddy with a weighting function (a= 4.586, d = 0.588 and b = 0.263).

-
12
I
1
I
1
I I
I
!

0 l/a l/d 1 8
- -In(a) -In(d) 0 A-AE
A,
(d (4
FIGURE 9. (a)The type of modification to p H ( S )needed to obtain the Hama (1954) velocity-defect
law. ( b ) A simplified modification. (c) The corresponding weighting function, W(S/d,). (d)
w(A-AE) = W(S/d,).
176 A . E . Perry, S. Henbest and M . S. Chow
Figure 9 (d) shows this function with logarithmic variables, where
w(h-AE) = w(S/dE).
With this modification (20) becomes

:? 5:;:
--
-
A h ( A )eVnw ( A-AE) dh.

This is an integral equation similar t o those used for solving inverse-scattering

problems. Given dUg/dA, and w(A-A,), the problem is t o determine h(A).However,
since the precise form of w(A-A,) is unknown, we must resort to trial and error.
Figure 10 shows how the integral in (24) is carried out for trial distributions of h(A)ePh
and w(A-A,). The shaded area equals dUg/dA, and the width of the integration
, +
window, A, -A,, is fixed for a given Reynolds number since A, -A, = In ( A U J v ) &,
where & is a universal constant. I n (24) we must ensure that A, < 0, since, as PC
indicated, a special formulation is needed for the smallest hierarchy to include the
additional non-geometrically-similar eddies forming from the roll-up of viscous
sublayer material. For simplicity consider the case of an assemblage of stretched
fl -eddies as shown in figure 7(e). Values of a = 4.568 and d = 0.588 (defined in
figure 9) and b = 0.263 give a U;S distribution which fits the Hama formulation
reasonably well as seen in figure 8. This modification of the p.d.f. by a weighting
function has important implications in spectra and broadband turbulence intensities
as outlined in the following sections.
3.2. Turbulence spectra
Let the power-spectral density of uivelocity fluctuations, area-averaged in a plane
at a distance z from the wall, for a hierarchy of scale S be &(k, z, z/S). This will be
called the hierarchy spectral function. Then, the summed power-spectral density a t
a fixed z for a range of hierarchy scales varying from 8, t o A , is given by

To be consistent with the mean-flow work in $3.1,we shall assume that p H ( S )is given
by (23).Then pre-multiplying (25)by k , z enables this equation to be written in terms
of our logarithmic arguments, thus

where a = In ( k , S),a, = In (k,z ) , = k, Z @ ~and ~ /@iiV= k, z&. We shall call Pti

the pre-multiplied hierarchy spectral function. I n (26)we are integrating with respect
to 6 with all other parameters fixed, in particular k, z.
The functions $ii have been calculated for a hierarchy consisting of a random array
(in the (z,y)-plane) of A-shaped vortices of the same scale. For simplicity the effect
of stretching, as illustrated in figure 7(d, e ) , has not been included. Figure 11 shows
the proportions of the A-vortex used together with the image vortex. The vorticity
distribution in the vortex rods is assumed t o be Gaussian (see PC, $2) with a
characteristic lengthscale ro = 0.108. The function was computed by taking fast
Fourier transforms of ui velocity distributions, induced by one isolated A-vortex
and its image, along lines of constant y/S in a plane of constant 2/13. The velocity
distributions were calculated using the Biot-Savart law. The power-spectral density
 Theoretical and experimental study of wall turbulence 177

FIGURE
10.The evaluation of equation (24).

I" / /-'

FIQURE11. -vortex geometry used to calculate @4t/ q.

for a random array of such vortices was formed by ensemble averaging the square
of the moduli of the fast Fourier transform of the velocity signatures in the plane
z / S = constant for various y/6, assuming that the mean streamwise and cross-stream
densities of these vortices scale with 6. This result was pre-multiplied by k, z to give
+$$. Sketches of the distributions of @ll/q, $22/q and $'33/q are presented in the
form of ' contour maps' in figure 12 (a-c), respectively. It can be seen that the contours
of $11 and +22 are similar and considerably different from the $33 contours. For A
sufficiently large (i.e. z/d+O), the $11 and @22 contours run parallel to the /\-axis and
are thus invariant with h in this region, while the $33 contours asymptote to zero;
as long as the Townsend boundary condition (that at z = 0, u3 = 0 while u, and u2
remain finite) is upheld, the contours will behave in the manner described for h large,
no matter what shape of eddy or assemblage of eddies is used to represent a hierarchy.
For h sufficiently negative, all contours asymptote to zero, since the far-field effect
+
of the vortices in the hierarchy vanishes for z 6.
178 A . E. Perry, S. Henbest and M . S. Chng

h
, v,g contours

(4
12. Sketches of the contours of @&,h)/U:. (a)@ J V (;b ) @22/V;
FIGURE ( c ) @33/U:.
 Theoretical and experimental study of wall turbulence 179
In these figures, lines of constant k, z are given by
A = a-a,,
+
i.e. lines of slope 1 with an intercept of a, on the u-axis. Imagine a plane perpen-
dicular to the ( A , a)-plane and aligned along the line of constant k, z. This will cut
through the spectral ‘hill ’ $it(cx, A)/U: and a projection of this cut is made onto the
($$JT, A ) plane. Then weighting this projection with A w ( A - A , ) and integrating
between A, and A, gives the summed non-dimensional energy-density contribution
to 4% at a fixed k, z (i.e. fixed a,) for a fixed value of A, (i.e. fixed z / A , ) . This is shown
clearly in figure 13(a). In what follows, A, will be taken to be sufficiently negative
and hence (26) becomes independent of A,. Let a, vary from - 00 to +a with A,
fixed. Thus &(a,, AE) can be mapped out for various values of A,. Such a plot would
correspond to ‘inner-flow ’ scaling of the pre-multiplied spectrum of ui velocity
fluctuations. The resulting distributions of F,, and F2, are similar to the distribution
of k , z @ , , ( k , z ) / ~ shown in figure 3(b) and the distribution of F33 is similar to the
distribution of k, z@33(k,z ) / q shown in figure 5 (b); however, no Kolmogoroff region
exists nor does a viscous cutoff occur. Instead, at high k , z , F,,, 4, and F33 each
collapse to their universal curve which extends to infinite k, z.
In figure 12, lines of constant k, z also correspond to lines of constant k, A,. If
we let 01, = ln(k,AE), then the ‘outer-flow’-scaling plots of the pre-multiplied
spectra, Hii(aE,A,) = k, A , Gir(k, A,)/U;;, can be obtained. The distributions of H,,
and H,, are similar to the distribution of k, A , djii(kl/AE)/T shown in figure 4 ( b )and
the distribution of H33 is similar to the distribution of k,AE @ 3 3 ( k l A E ) / q shown in
figure 6 ( b ) .
Figures 13 ( a ) ,( b ) respectively show a comparison between $,,/T for the present
example of a A-vortex with its image and for the example used by PC. PC were
interested only in the u1spectra and used an infinite straight-line vortex of some fixed
orientation to generate the velocity signatures via the BioeSavart law. Such a model
does not have the correct boundary conditions for the u3 velocity fluctuations,
whereas the h-vortex with its image does. PC also assumed that $,,/U: dropped
suddenly to zero, for z exceeding S,as shown in figure 13 ( b ) .The problem of an isolated
vortex having infinite kinetic energy spread over the entire flow field at a fixed value
of z was overcome by the use of an artificial outer limit on the integral used for
obtaining the ensemble-averaged power-spectral density. In the case of the A-vortex
model, this problem does not arise since it has finite energy. Also, instead of using
a continuous inverse-power-law p.d.f., PC used a discrete p.d.f. where the hierarchy
scales went in a geometric progression with a factor of 2. This was thought to be
consistent with vortex pairing, where the lengthscales and the circulation doubled
from one hierarchy to the next. It will be seen later that the use of a continuous
inverse-law p.d.f. gives much the same answer. Hence we could equally well use a
geometric progression with a factor of 2lIr, where r is a resolution factor. The above
approach is equivalent to multiplying the integrand in (26) by a series of uniformly
spaced unit Dirac delta functions, which changes the integral into a summation. This
is equivalent to using a trapezoidal rule. In spite of the crudity of the PC analysis,
the results produced are remarkably similar to those of the more-refined analysis
presented here and are found to be insensitive to r provided r 2 1 . The PC analysis
for u1 spectra with a discrete p.d.f. with a resolution factor r will be used in the
analysis given in later sections because of its analytical simplicity.
180 A . E . Perry, S. Henbest and M . S. Chow

(b)
FIGURE13. (a)The distribution of @ll(a,
A)/u;L computed for the A-vortex case shown in
figure 11. (a) The distribution of @ll(a,A)/qfor the model of Perry BE Chong (1982).
 Theoretical and experimental study of wall turbulence 181

3.3. Turbulence intensities

Following Townsend (1976) and PC, the broadband turbulence-intensity distribution
for a range of scales of geometrically similar hierarchies is given by

where I t j ( z / S )is the Townsend eddy-intensity function for a hierarchy of scale 6.The

z/S+O, I,, - -
boundary conditions mentioned earlier given I,, and I,, constant and finite for
( z / S ) and I,, ( Z / S ) ~Townsend
. also assumed that, for z B 6,Ii3(z/S)
asymptotes to zero. Typical distributions of I i j ( z / S )shown in figure 14 highlight these
functional properties.
Following the previous mean-flow and spectral analyses ($83.1 and 3.2), it will be
assumed that pH(S)is given by (23). Then (27) can be rewritten in the form

where the argument of the eddy-intensity function has been changed to A. I&) can
be calculated from the pre-multiplied spectral hierarchy function $i3/ q using

Figure 15(a, b ) shows Ill(A) and 133(A) calculated from $$,/q obtained for the
h-vortex model used in $3.2. The distribution of I,,@) will be similar to that for
],,(A), in that they are both finite and constant for A B 0, and the distribution of
II3(A)will be similar to I,,(A), in that they both have finite area. Also shown in the
figure is a weighting function w(A-A,). It can be seen from figure 15(a, b ) and the
above comments that, for A , sufficiently large and A, sufficiently less that zero, (28)
must lead to

-
U2
= A,A,+B,,
e
-
U2
3 = A,,
v
A,, A,, A, and K13are universal constants. In fact, it can be shown that
A , = AIll ( z / S = 0 ) ,
A , = MIz2 ( z / S = 0 ) ,

A, = JOfflJlI3,(A)
dA,

K,, = ~ o f f l d I 1 3 (dA
A )= - 1,
 182 A . E . Perry, S . Henbest and M . S . Chong

---- Non-zero

0 1
(b)

Is,(z/8)
A I
I
I

0 1 218
(C)

14. Typical eddy-intensity functions. (a)Ill(z/8) ; ( b ) Z13(z/8); ( c ) Z3&/8),

FIGURE

I n (29),B, and B, are large-scale characteristic constants which depend on the form of
w(h-A,). The equations listed in (29) are the result obtained by Townsend (1976)
and by PC, and are the same as (14), (15) and (16) using the dimensional-analysis
arguments given in $ 2 if the fine-scale motions are neglected or if z++ 00.
The analysis presented in this detailed physical model of wall turbulence is
applicable only to flat-plate flow with a zero streamwise pressure gradient. Fully
developed flow in ducts and pipes would have the complicating feature of eddies
intruding from the opposite boundary, and in a pipe curvature effects would also
influence the large-scale motions. However, as indicated in the dimensional-analysis
arguments given earlier, the preceding analysis should be applicable in the turbulent
wall region in pipes and ducts. This is substantiated experimentally for pipes in the
next section. Much of the difference between the various flow geometries can be
accounted for by an appropriate modification t o the weighting function w ( h-h E ) .
This in turn will control the values of the large-scale characteristic constants given
 Theoretical and experimental study of wall turbulence 183

(b)
15. Distributions of ( a )Ill(A) and (b) Iaa(A) calculated from the computed distributions
FIGURE
of @ll/q and @aa/q for the A-vortex shown in figure 11.

in $2 and the distribution of low-wavenumber energy of the u1 and u2 spectra.

Furthermore, it is argued that, although there will be detailed differences between the
geometries for flow beyond the turbulent wall region, the general form of the scaling
laws will still be applicable to all flow geometries.

4. Experimental results
4.1. Apparatus and method
The results presented here were obtained in fully developed, turbulent flow in a
smooth-walled circular pipe. The pipe consisted of seven 6 m lengths of precision-
drawn brass tubing of 0.099 m internal diameter. Adjacent lengths were joined using
specially machined collars which ensured that the surface discontinuities at the joints
were minimal. The contraction used is described in Perry & Abell (1975) and a
sandpaper trip was used after the contraction to stabilize the transition of the flow
to turbulence.
All measurements were taken 398.5 diameters downstream from the pipe entrance
and hopefully this ensured that the flow was fully developed. The flow was examined
over the range of Reynolds numbers (Re = 2 d , q , / v ) of 75000 to 200000, which
corresponded to a Karman-number ( A E U J v ) range of 1610 to 3900. The wall-shear
velocity U, was determined from the static-pressure drop per unit length along the
pipe. All measurements, including the hot-wire results, were corrected for the
effect of the density variation of the air along the length of the pipe.
Temperature corrections were not necessary.
Turbulence measurements were obtained using constant-temperature hot-wire
184 A . E . Perry, S . Henbest and M . S . Chong
anemometers similar to those used by Perry & Morrison (1971). The hot-wire
filaments were made from 5 pm Wollaston wire and had nominal etched lengths of
0.90 mm. The wires were calibrated dynamically by following the method outlined in
88.6 of Perry (1982). The dynamic calibration facility used enabled the wires t o be
calibrated inside the pipe at the test section. During calibration the wires were
exposed to a turbulent free uniform stream by removing the length of the pipe
upstream from the test section and replacing it with a contraction with screens. The
dynamic calibration facility contained the traversing mechanism and this ensured
that the orientation of the wires during calibration was the same as that during
measurement.
The u1spectra were measured using uncalibrated normal wires and u3spectra using
dynamically matched, uncalibrated X-wires. The power-spectral density of the
pertinent anemometer signal was calculated digitally using a FFT algorithm. The
signal was low-pass filtered at half the digital sampling rate (using Krohn-Hite
analog filters model number 3323) t o ensure that no aliasing of the measured
spectrum occurred. The ensemble average of ten such power-spectral densities gave
sufficient convergence of the spectrum after a ‘smoothing window ’ had been applied.
The h a 1 spectrum covered a frequency range from 2 t o 10 kHz. The argument of
the spectrum was transformed from circular frequency to streamwise wavenumber
by applying Taylor’s (1938) hypothesis of frozen turbulence and assuming that all
eddies moved at a convection velocity equal to the local mean velocity a t the
measuring point. Zaman & Hussain (1981)from their investigation of the applicability
of the Taylor hypothesis were led to state that ‘the use of the local time-average
velocity in shear flows especially in the computation of wavenumber spectra and the
eduction of large scale structures is not acceptable’. The applicability of this
hypothesis is investigated further in $4.2.1.All spectra were normalized using (5) and
it was assumed for the purposes of normalization that a t low wavenumbers the
measured spectrum could be extrapolated with @(&k1) constant and a t high wave-
numbers with Qii(k1)cc k ~ ~ .
I n the turbulent wall region, the turbulent energy dissipation E was calculated using
(3), (9) and that for a pipe

I n the wake region, the assumption that production is in balance with dissipation
is no longer valid, so i t was necessary to resort t o

8 = 15v
r kt @ll(kl) dk,,
which assumes isotropy of the dissipating motions. It can be shown that (32) leads
to large errors in the value of E if 7 < 8 , where 8 is the smallest-scale motion that
the hot-wire(s) can resolve.
4.2. Experimental turbulence spectra
4.2.1. The turbulent wall regioni
(i) u1 spectra. Spectra of u1 velocity fluctuations measured in the turbulent wall
region are shown in figure 16(a) with ‘inner-flow ’-scaling coordinates for various
values of z / A E for normal wires of lengths 1 = 0.39 and 1.26 mm. The data appear
t For a pipe the turbulent wall region is tentatively defined here as 140v/U, < z < 0.144,.
 Theoretical and experimental study of wall turbulence 185

10 102

10' 10'

10' 100
@ll(klZ)
u:
100 10-1

10-1 lo-'

10-2 I0-J

=+ 5571

I I I I 0-4
10- 10-2 10-1 1 00 10'
klZ

108
- I I I I 10'

10' - - loo

100 -
@ll(kl
u:
10-1 -

lo-' -

10- -
0 14

I I I I
I 0-4 10-6
lo-= 10-1 1 00 10' I0
' 10 3
klA E
FIGURE 16. (a)u1spectra scaled with 'inner-flow'-scaling coordinates for varying values of z / A ,
within the turbulent wall region. Re = 200000. ( b ) The same scaled with 'outer-flow'-scaling
coordinates.
186 A . E . Perry, S. Henbest and M . S. Chong
to be consistent with the expected distributions shown in figure 3(a).The data are
shown scaled with ‘outer-flow’-scaling coordinates in figure 16(b)and appear to be
consistent with the expected distribution shown in figure 4 ( a ) .
Let us now examine the behaviour of the high-wavenumber peel-off of the spectra
from the -5-power-law distribution in figure 16 ( a ) .Earlier, we showed that the value
a t which the spectra peel off from the -5 power law owing to the effects of viscosity
is expected to follow
M
k , z = 7 (2,):. (33)
Kd

For u1 spectra measured using a normal wire of length 1 the peel-off from the -$
law may be caused by the spatial-resolution limit of the probe rather than by the
effects of viscosity. Such a peel-off is expected t o follow
Z
klZ * - (34)
1’
because Wyngaard (1968) has shown that the highest wavenumber resolvable by a
wire of length I is given by k , = O(l/l).
Let ( k , ~ be) ~the experimentally determined peel-off point. This point can be
calculated using the method illustrated in figures 17 ( a , b). Figure 17 ( c ) shows a plot
of log(k,z), versus log(z+) determined from u, spectra measured at the same
Reynolds number with wires of different length. Provided that z / l is sufficiently large
+
the data appear to follow a line of slope 9. However, as z decreases the data appear
+
to fall closer to the line of slope 1. It can be seen that, with the use of shorter wires,
evidence for the peel-off following (33) becomes more substantial. By extrapolating
the data in figure 17 ( c )to zero wire length, we obtain an estimate M = 0.085. When
the data shown in figure 16(a) are scaled with Kolmogoroff coordinates the extent
of the inertial subrange was greater for the spectra measured with the 0.39 mm wires
than that measured with 1.26 mm wires. Thus spatial resolution is a problem
whenever we are attempting to verify the existence of a Kolmogoroff region in
laboratory-produced flows. PC in their interpretation of the data of Perry & Abell
(1977) came to the conclusion that no Kolmogoroff region existed and that the spectra
could be explained solely in terms of attached coherent motions.
Let us examine the behaviour of the spectrum at low wavenumbers. I n figure
16 (a,b ) a small deviation above the inverse-power-law distribution is apparent at low
wavenumbers in each spectrum. It was seen earlier that such deviations can be
accounted for by the inclusion of a weighting function W ( d / A , ) . This deviation is more
apparent in the pre-multiplied spectra scaled with ‘ outer-flow ’-scaling coordinates
as shown in figure 18(a).A simulation was carried out to determine what effect the
inclusion of such a weighting function has on the predicted u1 spectral distribution.
For simplicity, this simulation used PC’s spectral model, a discrete p.d.f. of hierarchy
scales with a resolution factor r of 4, and the weighting function conjectured to be
appropriate for a pipe is given in figure 9 ( c ) with a = d = 2.0. Curve (i) in figure 18( b )
shows the spectral prediction without the inclusion of the weighting function and
curve (ii) that with the inclusion of the weighting function for a value z / A E
comparable with the values of z / A , for the data shown in figure 1 8 ( a ) .A similar
deviation at low wavenumbers is apparent in the streamwise spectral data of Perry
& Abell(l975)in their figure 11( a ) ,of figure 6 ( a )for y + = 215 of Bremhorst & Walker
(1976),and of figure 11 for y+ = 200 and 500 of Bullock, Cooper & Abernsthy (1978).
All of these spectra were measured in the turbulent wall region. Recently, a colleague
 Theoreticul and experimental study of wall turbulence 187

Quadratic curve-fit
extrapolating to zero q

log (kl a p = :gag(kl ZYI

FIGURE 17. (a,b ) The method used for determining ( k , ~ )(c)

~ Values
. of log,, ( k , z ) , versus
log,, (z+) obtained from u, spectra measured with normal wires of differing etched lengths 2.

K. L. Lim (private communication, 1984) has obtained u, spectral data for flow over
a smooth flat plate with a zero streamwise pressure gradient. A typical result is shown
in figure 18(c) and it shows a similar deviation but with a value of a of about 3.5.
Curve (iii) in figure 18 (b) shows the predicted spectral result for a = 3.5.This value
of a roughly corresponds to the value needed to obtain the Hama velocity-defect-law
formulation using stretched fl-eddies (see $3.1).
The incomplete collapse of the data at low k, A , in figure 16 (b) may be due to the
invalid use of Taylor's (1938) hypothesis, which utilizes one single convection velocity
for all eddy scales at a fixed point in the flow. It is suspected that the larger-scale
coherent attached eddies are convected downstream at a faster rate than the
smaller-scale coherent eddies; hence there is a spread in convection velocities for a
given wavenumber (e.g. see Wills 1964, who studied jets). A crude simulation was
carried out to see whether this lack of collapse of the data could be explained by
a spread in the convection velocity of the coherent attached eddies. The predicted
u1 spectral distribution from the model shown as curve (ii) in figure 18(b) is
reproduced as the heavy curve in figure 19. This was calculated by Fourier-
decomposing the spatial variation in velocity for each representative eddy or
hierarchy (giving the wavenumber directly) and is therefore independent of the
convection velocity of the eddies. This we will regard as the true spectrum. To
I PLM 165
188 A . E . Perry, S. Henbest and M . S. Chow

1 .oo

0.50

0.25

lo-* 10-1 100 10' 10s 103

kl 'E

(4
FIGURE 18. (a)Figure 16(b)presented as a pre-multiplied u1spectrum. (a) Computed pre-multiplied
u1 spectra using (i) no weighting function, (ii) weighting for a pipe and (iii) weighting function for
a boundary layer (a= 3.5). (c) pre-multiplied u1 spectra measured in a turbulent boundary layer,
z/AE = 0.0296, K A , / v = 130700, Lim (1984).

'True'

- k, Z@P,,(klZ)
u:

- log (42 )
FIGURE 19. The effect of including a spread in convection velocities of the A-shaped eddies on the
computed frequency spectrum which has been converted to wavenumber spectra using Taylor's
(1938) hypothesis.
 Theoretical and experimental study of wall turbulence 189
illustrate the effect of a spread in convection velocities of the eddies, this spectrum was
recalculated as follows: i t was assumed that the convection velocity for a hierarchy
of scale 6 is invariant with z and equal to the local mean velocity at z = calculated
using (3); the frequency spectrum as seen by a stationary probe at a distance z from
the wall was then calculated and the argument of this spectrum converted from
frequency to wavenumber using Taylor’s hypothesis with a single convection velocity
equal to the local mean velocity at z calculated using (3). The result is shown by the
dotted line in figure 19. It can be seen that the low-wavenumber bump or deviation
has been attenuated and the suggestion of another bump at high wavenumbers is
apparent. This curve bears a strong resemblance to that of the data of Perry & Abell
(1975) mentioned earlier. The low-wavenumber shift seen in figure 19 can be shown
to vary with z / A E and the variation, though not as large, is consistent with that of
the data shown in figure 16( b )and 18 (a).Hence the invalid use of Taylor’s hypothesis
has a measurable effect and prevents the data when plotted with outer-flow-scaling
coordinates from collapsing at low wavenumbers.
(ii) u3 spectra. The u3spectra measured in the turbulent wall region are shown in
figure 20 with ‘inner-flow’-scaling coordinates for varying values of z / A E a t six
different Reynolds numbers. The data at low and moderate k, z correlate reasonably
with the expected distribution shown in figure 5 ( a ) . More importantly, there is no
inverse-power-law region and no systematic peel-off of the spectra at low k , z as is
characteristic for the u1spectra (see figure 16a). This fact alone lends strong support
for the Townsend attached-eddy hypothesis and the model of PC. The slight spread
in the u3 spectra at low k , z cannot be explained by a spread in convection velocity
of the eddy scales, since a probe sensitive only to us motions only ‘sees’ eddies of
scale 6 = O(z). The authors contend that this spread may be caused by cross-
contamination of the X-wires from the u2 motions. This will occur if the X-wires are
rolled about the streamwise axis or if the hot-wire filaments are bowed. In figure 20
no inertial subrange is observed and the peel-offs at high k , z depend upon z / A E and
not z+; since the corresponding u1spectra (which were measured using normal wires)
showed a distinct inertial subrange, we suspect that the spatial-resolution limit of
the X-wires may dominate the form of the spectrum in this region. If this were the
case, the data when scaled with ‘outer-flow’-scaling coordinates should collapse at
high k, A,, since for these experiments the X-wire geometry and scale was fixed. Hence
the resolution limit of the probe happened to scale with A , since l / A E is fixed. This
is confirmed in figure 21, where the data are presented with ‘outer-flow’-scaling
coordinates. The spread in the data a t low k , A , with z/AE has increased significantly
and this is in agreement with the spectral proposal shown in figure 6(a).
4.2.2. The fully turbulent region
Here we will discuss the spectra which occur over the entire fully turbulent region,
which extends from the outer limit of the buffer zone to the pipe centreline. According
to the Townsend Reynolds-number-similarity hypothesis, the spectrum of the
energy-containing components of the ui velocity fluctuations in this region should
follow

where i = 1, 2 or 3 and qi is independent of viscosity. At high wavenumbers a

Kolmogoroff region should exist and provided that the Reynolds number of the flow
is sufficiently large this should include an inertial subrange. Of course this region can
only be derived from dimensional analysis and not from the attached-eddy model.
7-2
190 A . E . Perry, S . Henbest and M . S . Chow

u,P
10-1

10-2

10-3
10-3 10-2 10-1 100 10' 1o=
kl
FIGUBE 20. usspectra scaled with 'inner-flow'-scalingcoordinates for varying values of % / A , within
the turbulent wall region and for six Reynolds numbers equally spaced from 75000 to 200000.

10-8 10-1 100 10' 102 103

k,AE
FIGURE
21. uaspectra shown in figure 20 scaled with 'outer-flow'-scaling coordinates.

(i) u1 spectra. The u1spectra for six different Reynolds numbers, ranging from
75000 to 200000, are presented in figure 22(a-f). In each figure the wall distance
varies from the pipe centreline (z/dE= 1.0)to a value of z/dE close to the outer limit
of the buffer zone. Figure 24(a) shows a superimposed selection of data from
figure 22(a-f) and it is seen that (35) is applicable in the energy-containing region,
at least for the rarige of Reynolds numbers examined here. A t a fixed z/dE, the slight
spread in the spectra at low k l z with R e is thought to be due to a change in the
fractional spread of the convection velocities of the eddies as R e changes (see Perry
& Abell 1977).
That part of the u1spectral results that can be described by (35)are similar to those
predicted by PC, i.e. for the coherent attached eddies. Spectral distributions have
also been computed using the method detailed in $3.2 for a A-vortex without the
use of a weighting function. A Kolmogoroff region is not included in these calculations.
These are shown in figure 25(a) for a range of values z/dE. Even though these
 Theoretical and experimental 8tud9.JOf wall turbulence 191
103 I 103i i
10' 1o*
10' ndn - 10'
@ll(kl4
u: 100 100
10-1 10-1
--P
lo-' lo-'
10-8 10-8
lo-' 10-8 lo-*' 10-1 100 10' 10' lo-' 10-8 lo-' 10-1 loo 10' 10'

108 1 I 103 1

0.01
10' - 0.0
0
10'
10' 10'
100 100
10-1 10-1
10-1 lo-'

10-8 I 10-8
10-4 10- 10-3 10-1 100 101 10' 10-4 10-8 lo-* 10-1 100 101 10'

103, I 1 10'
10' 10'
100
10-1
10-2
lo-' 10-3
10-3 r0-4
\
10-4 10-6 I
10--I lo-' 10- 10-1 100 10' 10' 10-3 10-2 10-1 100 101 10' 103
kl k,
FIGURE22. u1 spectra spanning the whole turbulent region scaled with ' inner-flow'-scaling
coordinates at six Reynolds numbers for varying values of z/dE. (a)Re = 200000, U,= 1.188 m/s;
( b ) 175000, 1.041; (c) 150000, 0.920; ( d ) 125000, 0.777; (e) 100000, 0.635; (f) 75000, 0.487.

calculations are applicable only to flow over a smooth flat surface as mentioned
earlier, they show a distinct resemblance to the experimental pipe data shown in
figure 22. It is remarkable how small a value of z/d is required before a substantial
length of inverse-power-law region is apparent in the predicted u1spectra. This also
seems to be so for the experimental spectra.
192 A . E . Perry, S . Henbest and M . S . Chong
100

10-1

10-2

10-3

i
lo-' 10-3 10-2 10-1 100 101 102 103

100

10-1

10-2

10-3

lo-'
10-6 i
10-3 10-2 10-1 100 101 102 103
10-4

10-5 I

100

lo-'

10-2

10-3

lo-'
10-5
t
I
10-4

10-5
10-3 10-2 10-1 100 101 102 103
kl z
FIGURE23. u3 spectra spanning the whole turbulent region scaled with 'inner-flow'-scaling
coordinates at six Reynolds numbers for varying values of z/AE. Re and U, values as for
figure 22.

Let us now return to figure 22. A t the two highest Reynolds numbers, figure 22 (a,b ) ,
the data at high k , z collapse t o an inertial subrange. As the Reynolds number
of the flow is decreased, the effect of viscosity becomes more important in this
high-wavenumber region and the spectra peel off from the inertial subrange (( -5)-
power-law region) a t decreasing values of k, z as z / A , increases for a fixed Reynolds
number. This -5-power-law region evolves into a -%-power-law 'envelope ' a t the
lower Reynolds numbers.
 Theoretical and experimental study of wall turbulence 193
(ii) u3 spectra. The corresponding u3 spectra are presented in figures 23 (a-f ). A t
high k , z the data behave in a manner similar to the corresponding part of the u,
spectra. It can be seen that, for z sufficiently large, the spatial-resolution limit of the
X-wires in the Kolmogoroff region is no longer a problem, since the dissipation e is
small and so 7 is large. Figure 24(b) is a superposition of the data shown in
figure 23(a-f) and i t appears that (35) is upheld for the energy-containing u3
motions.
Figure 25(b) shows the u3 spectra computed using the h-vortex model for various
values of z/d,. These computed spectra show that, for z / A , sufficiently small, that
part of the spectrum due to the coherent attached eddies is universal for all
wavenumbers, as was shown in earlier sections.
A comparison of the computed u1 and u3spectra with the corresponding experi-
mental data in figure 24 ( a , b ) give encouraging support for the existence of a range
of scales of geometrically similar hierarchies, consisting of coherent attached eddies,
as was proposed by PC for wall-shear flow.

4.3. Broadband turbulence results

4.3.1. Turbulence results for u,
Distributions of $/q measured in the fully turbulent region are shown in
figure 26 for various values of A , U 7 / v . Also shown are the predicted distributions
of q/q using (14) for the highest and lowest experimental values of A , U7 / v .
These predictions are applicable only in the turbulent wall region and the values of
A,, B, and C have been estimated from the experimental spectra to be 0.90,2.67 and
6.06 respectively. In the turbulent wall region there is a systematic variation of the
data with A , U 7 / v at a fixed z / A , that is significantly greater than the predicted
variation. This variation of the data extends beyond the turbulent wall region and
still exists, though to a lesser extent, at the pipe centreline. The authors conjecture
that this variation may be due to the distortion of the geometry of the coherent
attached eddies in the smallest hierarchy by the circular boundary condition of the
pipe. We imagine that this distortion is transmitted to the eddies in the larger
hierarchies, which are formed in stages from the smallest hierarchy by a vortex-pairing
process. The ‘degree ’ of distortion will depend upon the ratio of the smallest hierarchy
scale (which scales with v / U 7 ) to the radius of curvature of the pipe. This ratio is
proportional to A , U J v , the Karman number. I n the limit as A , U7/v+cu, this
influence should vanish and the Townsend Reynolds-number-similarityhypothesis
should be upheld. This lack of Reynolds-number similarity, which is apparent even
in the largest-scale eddies, is difficult to see in the experimental spectra because of
the complicating effect of the spread in convection velocity of the eddy scales
discussed earlier. Perry & Abell (1977) also observed a lack of Reynolds-number
similarity even in the large scale motions and attributed this to an insufficient
flow-development length in their pipe. In the present study the development length
was approximately five times larger and the Reynolds-number effect is still present.
4.3.2. Turbulence results for u3
Figure 27 shows the distributions of measured in the fully turbulent region
for various values of A , U7/v.According to the analysis in $2, the distribution in the
turbulent wall region should be given by (16).However, the effect of the viscous cutoff
given by the second term in (16) is masked by an even stronger cutoff, namely the
spatial-resolution limit of the X-wires, as was confirmed by the u3spectra measured
194

10 I 1 1 1 1
,
(4
10

10
@ll(kl4
u14
10

10-

10-

10-

10-

10-4
10-3 10-2 10-1 100 10' 100 109

k1=

FIQURE
24. (a)A superposition of some of the u1spectral data shown in figure 22 (a-f ). ( b ) A
superposition of the usspectral data shown in figure 23a-f ).

in the turbulent wall region (see figure 20). There is no systematic variation of these
results with A , U J v . It was conjectured that the high-wavenumber cutoff of the u3
spectra was controlled by the scale of the X-wires. This means that the broadband
data should collapse with 'outer-flow' scaling, since Z/AE was held approximately
constant.
 Theoretical and experimental study of wall turbulence 195

10-3 10-2 10-1 10 0 101 1o=

-klz
(a)

100
0.0156. 0.0078.0.0039. 0.002

10-3
10-J lo-= 10-1 100 108 10'
-k,z
(b)

FIGURE25. Spectra computed using the A-vortex model for varying values of %/AE scaled
with 'inner-flow'-scaling coordinates. ( a )u1 spectra. (a) u3spectra.

The fact that the data collapse to a region of constant G/qin the turbulent wall
region is fortuitous and does not necessarily confirm the asymptotic prediction given
by (16)and by (29c). Figure 28 shows the results of a crude simulation which used
the eddy-intensity function I&) shown in figure 15 (b) and the weighting function
used earlier for pipes. The low cutoff A, shown in figure 15(b) was replaced by
A, = In (E/z) to simulate the spatial-resolution limit of the probe. Here, the value of
A, was allowed to be greater than zero and hence had an effect on the integral. For
1 = 2.00 mm, a flat region is predicted for values of z/AE comparable with those of
the data shown in figure 27 for the turbulent wall region.
196 A . E . Perry, S. Henbest and M . S. Chong

5.0 - (ii)(i) -
$
-
rr: 4.0 -

3.0 -
symbol Re W m l s )
AEU,/V
0 200000 1.188 3 900
g
2.0 - 0 175000 1.041 3 420 8
0 150000 0.920 3010 8
a 125000 0.777 2550
0 100000 0.635 2080
.O

0
0.10 1.00
z / AE
FIGURE 26. %/qdistributions for varying values of A, U J v . Curves (i) and (ii) are the predicted
distributions using equation (14) for the highest (3900) and lowest (1610) experimental values of
A E U , / u respectively (A, = 0.90, B, = 2.67 and C = 6.06).

I I
1.5 -

2
-
1.0 - -
u,P
0.5 - O%a -

0 . . * . . . I I

1.5 I I I I " ,
I I I 1 I m v ,

- 1.0
I=Omm
-
_-u:
u:
0.5 - -
0 1 1 , , , , I
0.02 0.10 1.OO
*Z/AE

FIGURE28. $/us" distribution calculated using the A-vortex model and the weighting function
for a pipe with a spatial-resolution cutoff determined by the effective X-wire.lengthscale 1. Here,
2 = 0 mm and 2.00 mm.
 Theoretical and experimental study of wall turbulence 197

5. Conclusions
From their analysis and measurements, the authors propose the following picture
of wall turbulence. Attached coherent eddies are formed from the viscous-sublayer
material and have a lengthscale corresponding to the Kline scaling and a velocity
scale equal to the wall-shear velocity. They stretch and grow with a fixed orientation
relative to the wall and are said to belong to the ‘first hierarchy’ of eddy scales. They
either die by viscous diffusion and vorticity cancellation or else pair or merge to
produce eddies of a larger lengthscale, and these eddies are said to belong to the
‘second hierarchy ’. This process repeats itself, giving a range of scales of geometrically
similar hierarchies. The p.d.f. of hierarchy lengthscales is assumed to be of inverse-
power-law form and all hierarchies have the same characteristic velocity scale. These
two assumptions lead to the logarithmic law of the wall, a region of constant Reynolds
shear stress and an inverse-power-law spectral region for the fluctuating velocity
components parallel to the wall in the turbulent wall region. These attached eddies
are responsible for the mean vorticity, Reynolds shear stress and most of the
energy-containing motions. It is conjectured that they do not contribute to the
turbulent energy dissipation, except perhaps for the two smallest hierarchies. In
$4.2.1spectral data were presented which showed support for the above picture of
wall turbulence and for the existence of a Kolmogoroff spectral region with an inertial
subrange. It has been shown by Perry & Chong that attached eddies alone cannot
explain the existence of a Kolmogoroff region. Here, we propose that the attached
eddies are surrounded by detached isotropic fine-scale eddies which are responsible
for a Kolmogoroff spectral region and for most of the turbulent energy dissipation.
The authors suggest that these eddies originate from the debris of dead attached-eddy
material which has been convected away from the wall and stretched and distorted
by the larger-scale attached eddies. From this model of wall turbulence it appears
that there is an energy flow to low wavenumber, due to eddy pairing, and an energy
flow to high wavenumber in the Kolmogoroff region. No physical model, such as a
cascade process, has been developed here and the existence of an inertial subrange
has been explained only by the conventional dimensional-analysis argument.
The dimensional-analysis approach of Perry & Abell (1977)to wall turbulence has
been extended to include all three components of velocity and the analytical
deductions are consistent with the detailed physical model developed here, which was
based on Townsend’s (1976) attached-eddy hypothesis and the model of Perry &
Chong (1982). The hypothesis of Townsend and the model of Perry & Chong
were only applicable in the turbulent wall region and have been extended here to
include the ‘wake ’ region. This involved modifying the inverse-power-law p.d.f.
proposed by Perry & Chong so that the model would give a mean-flow-velocity
distribution that follows the Hama (1954)velocity-defect law and the low-wavenumber
behaviour seen in experimental spectra of velocity components parallel to the wall
(streamwise and lateral) measured in the turbulent wall region. This required a
higher eddy population for the large-scale eddies and the type of modification to the
hierarchy p.d.f. needed depends on the type of flow being considered (i.e. boundary-
layer, pipe or duct flow).
Various eddy shapes and p.d.f.’s have been chosen in an attempt to correlate the
analysis with the experimental data. No definitive eddy shape or p.d.f. has emerged
and their exact form is still uncertain. Nevertheless, they lead to all the correct
analytical expressions for the mean flow and spectra. Thus a link exists between the
mean-flowdistribution in the Coles ‘wake ’ region and spectra of velocity fluctuations
198 A . E . Perry, S . Henbest and M . 8.Chong
parallel to the wall measured in the turbulent wall region. The broadband turbulence-
intensity measurements are also consistent with the analysis but are less definitive
in establishing the detailed structure of wall turbulence.
The link between the turbulence structure in the turbulent wall region and the
outer-flow ‘wake’ region might possibly lead to a closure model in the prediction of
a turbulent boundary layer in an adverse pressure gradient. Assuming that the
hierarchy spectral function is unaffected by the presence of a pressure gradient, then
a link exists between the Reynolds-shear-stress profile and the mean-velocity profile.
The link is the p.d.f. of hierarchy lengthscales, which when coupled with the mean-flow
momentum equation (Reynolds boundary-layer equation) might form the basis of
a prediction scheme.

The authors wish to acknowledge the financial assistance of the Australian

Research Grants Scheme.

REFERENCES
ACARLAR, M. S. & SMITH,C. R. 1984 An experimental study of hairpin-type vortices as a potential
flow structure of turbulent boundary layers. Rep. FM-5. Dept of ME/Mech., Lehigh University.
BATCHELOR, G. K. 1956 Theory of Homogeneous Turbulence. Cambridge University Press.
BREMHORST, K. & WALKER, T. B. 1973 Spectral measurements of turbulent momentum transfer
in fully developed pipe flow. J. Fluid Mech. 61, 173-186.
BULLOCK,K. J., COOPER,R. E. & ABERNATHY, F. H. 1978 Structural similarity in radial
correlations and spectra of longitudinal fluctuations in pipe flow. J. Fluid Mech. 88, 585-608.
COLES,D. E. 1956 The law of the wake in the turbulent boundary layer. J. Fluid Mech. 1,191-226.
HAMA, F. R. 1954 Boundary layer characteristics for smooth and rough surfaces. Trans. Soc. Naval
Arch. Mar. Engrs 62, 333-358.
HEAD,M. R. & BANDYOPADHYAY, P. 1981 New aspects of turbulent structure. J. Fluid Mech. 107,
297-337.
HUSSAIN,A. K. M. F. 1982 Coherent structures - reality or myth. Rep. FM-17. Dept. of Mech.
Eng., University of Houston.
ISAKSON,A. 1937 On the formula for the velocity distribution near walls. Tech. Phys. U.S.S.R.
IV,155.
KIM, J. 1985 Evolution of a vortical structure associated with the bursting event in a channel
flow. Fifth Symp. on Turbulent Shear Flows, Cornell University, Ithaca, New York, 9.23.
KLINE, S. J.,REYNOLDS, W. C., SCHRAUB, F. A. & RUNDSTADLER, P. W. 1967 The structure of
turbulent boundary layers. J. Fluid Mech. 30, 741-773.
KOLMOQOROFF, A. N. 1941 The local structure of turbulence in incompressible viscous fluid for
very large Reynolds numbers. C.R. Acad. Sci. U.R.S.S., 30, 301-305.
MILLIKAN, C. D. 1939 A critical discussion of turbulent flows in channels and circular tubes. I n
Proc. 5th Congress of Appl. Mech. Cambridge, Mass. (ed. J. P. Dentlartog & H. Peters),
pp. 386-392. Wiley.
MOIN, P. & KIM, J. 1985 The structure of the vorticity field in turbulent channel flow. Part 1.
Analysis of instantaneous fields and statistical correlators. J. Fluid Mech. 155, 441464.
PERRY, A. E. 1982 Hot- Wire Anemomdry. Clarendon.
PERRY, A. E. & ABELL,C. J. 1975 Scaling law for pipe flow turbulence. J. Fluid Mech. 67,257-271.
PERRY, A. E. & ABELL,C. J. 1977 Asymptotic similarity of turbulence structures in smooth- and
rough-walled pipes. J. Fluid Mech. 79, 785-799.
PERRY, A. E. & CHONQ,M. S. 1982 On the mechanism of wall turbulence. J. Fluid Mech. 119,
173-2 17.
PERRY, A. E. & MORRISON,G. L. 1971 A study of the constant temperature hot-wire anemometer.
J . Fluid Mech. 47, 577-599.
 Theoretical and experimental study of wall turbulence 199
SMITH,C. R. 1984 A synthesized model of near wall behaviour in turbulent boundary layers. In
Proc. of 8th Symp. on Turbuleme, Dept. of Chem. Eng., University of Missouri-Rolla.
TAYLOR, G. I. 1938 The spectrum of turbulence. Proc. R . Soc. L d .A164,476-490.
TOWNSEND, A. A. 1961 Equilibrium layers and wall turbulence. J . Fluid Mech. 11, 97-120.
TOWNSEND, A. A. 1976 The Structure of Turbulent Shear Flow (2nd ed.). Cambridge University
Press.
WILLS,J. A. B. 1964 On convection velocities in turbulent shear flow. J . Fluid Mech. 2 0 , 4 1 7 4 3 2 .
ZAMAN,K. B. M. Q. t HUSSAIN, A. K. M. F. 1981 Taylor hypothesis and large-scale coherent
structures. J . Fluid Mech. 112. 379-396.