You are on page 1of 77

3A1 Incompressible Flow

BOUNDARY LAYER THEORY


T. B. Nickels 2007
1
Background
1 History
The eld we now call uid mechanics was, before 1905, split into two separate elds:Theoretical
hydrodynamics, which involved the theoretical analysis of inviscid uids (which we have dis-
cussed in the rst part of 3A1), and Hydraulics, which was basically a set of empirical
relations used by engineers in the design of uid systems.
M. J. Lighthill described the rst as The study of phenomena which could be proved, but
not observed and the latter as The study of phenomena which could be observed but not
proved. In other words, Theoretical Hydrodynamics was mathematically rigorous but did
not describe real phenomena and Hydraulics described real phenomena but had little if any
theoretical foundation.
One particular and obvious failing of Theoretical hydrodynamics (potential ow), for exam-
ple, is that it predicts zero drag on bodies such as spheres, cylinders and aerofoils.
In 1905 Ludwig Prandtl united these two elds by introducing his famous BOUNDARY
LAYER HYPOTHESIS. The essence of this hypothesis is that FAR from solid boundaries
the ow is well approximated by inviscid theory (the EULER EQUATIONS) and it is only in
a thin region near the boundary (the BOUNDARY LAYER) that viscosity becomes impor-
tant. He then developed BOUNDARY LAYER THEORY by making various assumptions
to simplify the full set of equations for this region close to the boundary.
2 Characterising the boundary layer
In IB you did an experiment in the small blue wind-tunnels in the hydraulics lab (it was
a very long time ago). In that experiment you measured the streamwise velocity near the
surface of a at plate mounted in the wind-tunnel for two dierent cases: one in which
the boundary layer was laminar and one in which it was turbulent. The latter case was
made turbulent by the addition of a trip to the nose of the plate (essentially just a thin
rod attached to the surface perpendicular to the oncoming ow). Some results are shown in
gure 1. There are some obvious features of these results.
The velocity drops from the free-stream value as the plate is approached down to zero
at the plate (of course you can never measure the speed right at the plate but you can
safely extrapolate the trend down to zero).
The region where the velocity is changing is fairly thin relative to the size of the
wind-tunnel.
The laminar and turbulent velocity proles are signicantly dierent - the turbulent
one has much larger gradients near the wall and extends further into the ow (is
thicker).
The curve labelled BLASIUS refers to an exact solution for the laminar boundary layer
that we will derive later in this part of the course.
Now before we go on to look at the boundary layer equations and solutions we might rst
look at how we can characterise, and parameterise the measured proles.
2
Figure 1: IB Lab results
2.1 The skin-friction
Probably the most important boundary layer parameter for an engineer is the drag it exerts
on the wall. This is readily calculated using concepts from IB thermouids. The shear stress
in a Newtonian uid is given by
=
u
y
. (1)
Now if we want to know the shear-stress acting on the wall (or, equivalently, that the wall
exerts on the uid) then we consider the value of the shear stress at the wall,

w
=
u
y

y=0
. (2)
From this we can also come up with a non-dimensional skin-friction coecient
C

f
/2 =
w
/(U
2
1
) (3)
where the dash is there for historical reasons and just means the local skin friction coecient
at some point along the plate as opposed to the overall skin friction coecient C
f
for the
whole plate (i.e. integrated over the whole plate length). If you look at gure 1 you can
see that the turbulent boundary layer has a much larger shear-stress at the wall than the
laminar one (look at the gradient at y = 0).
2.2 How thick is the boundary layer?
Since in some of the potential ow analysis we assume the solutions are valid outside the
boundary layer then we might need to know the thickness of the boundary layer. The eect
3
y
u
Missing mass, momentum, energy
(compared with the inviscid case)
U
1
Inviscid solution
slip at the wall
U
1
-u(y)
u(y)
Figure 2: Boundary layer prole showing missing mass, momentum and energy
of viscosity on the mean velocity drops o but it is hard to say where it becomes negligible.
One crude measure of thickness sometimes used in experiments is the 99% thickness. This
is the distance from the wall at which the velocity reaches 99% of its far-eld or free-stream
value. Whilst this value is useful it is also a bit arbitrary (why not the 98% thickness?) and
it is also tricky to measure accurately since the velocity at this point changes very slowly
with distance from the wall.
Another way to characterise the boundary layer is to compare the real ow with the ow we
would have in the inviscid case. In this situation there would be slip at the wall and hence no
prole. In comparison with this ideal ow we see that the mass ux is reduced (the velocities
are less) and the momentum ux is also reduced (for the same reason). Further we could
say that the kinetic energy ux is also reduced (we could extend this to other quantities).
So we might say that in the boundary layer we have a mass ow decit, a momentum
decit and an energy decit (when compared with the inviscid ow). The amount of these
decits is related to the thickness of the boundary in some way. We have already discussed
displacement thickness (related to the mass ow decit) in the rst part of the course but
we will re-cap here.
Mass ux - displacement thickness. If we want the inviscid ow to have the same mass ux
as the real ow then we need to remove a layer of uid from the inviscid ow near the wall.
We can work out the thickness of this layer (which we will call

) by assuming that its mass


ux is equal to the amount missing when we compare the real ow and the ideal
U
1

=
_

0
(U
1
u)dy (4)
and hence

=
_

0
_
1
u
U
1
_
dy (5)
Note that this is an integral denition and so it is also much less sensitive to experimental
error than the 99% thickness. We can also do the same for the momentum (we call this
thickness ) and the energy (
E
) i.e.
=
_

0
(U
1
u)u
U
2
1
dy =
_

0
_
1
u
U
1
_
u
U
1
dy (6)
4
and

E
=
_

0
(U
2
1
u
2
)u
U
3
1
dy =
_

0
_
1
_
u
U
1
_
2
_
u
U
1
dy (7)
Note that the momentum thickness and the energy thickness might be more appropriately
called the missing-momentum thickness and the missing-energy thickness since larger
values imply that there is more of the quantity missing relative to the inviscid ow. Since
there is force acting on the uid opposite to the ow (the skin-friction drag) then we might
expect the momentum thickness to increase since momentum must be lost due to this external
force (similarly energy is usually lost in the streamwise direction and so this thickness will
usually increase). Note that, in the case of ows with pressure-gradients, there are additional
forces acting which can also change these thicknesses and they are not always acting in
the same direction as the skin-friction (just think about this for later in the lectures).
2.3 Shape factors
Another useful non-dimensional way of looking at boundary layer proles is to consider
the ratios of these lengths. The most common parameter used is simply the ratio of the
displacement thickness to the momentum thickness and is called the shape-factor (or form-
factor), H. So H =

/. The reason it is called shape factor is that if one considers the


denitions of the thicknesses given above then it may be noted that the momentum thickness
gives more weight to ow near the wall than the displacement thickness. Hence boundary
layers that have smaller velocities closer to the wall end up having a large shape-factor.
In particular boundary layers close to separation have large shape-factors and hence this
parameter is often used as an indicator of whether separation is imminent.
5
Question 1.
An experimentalist has measured the boundary layer velocity prole on a wing and found
that a useful curve-t to the data is given by u/U
1
= 1 e
5y/
where y is the distance from
the wall and is the 99% thickness of the boundary layer as measured.
a) Find the displacement, momentum and energy thicknesses of the boundary layer at this
point in terms of .
b) She makes another measurement further downstream. Consider each thickness and state
whether it you think will be larger or smaller downstream - giving sensible reasons. Consider
how this will depend on where the measurements are taken on the aerofoil.
c) She decides a better curvet for the ow near the wall is u/U
1
= 16
0.926
15, where
= y/. She realises immediately that there is a diculty with the behaviour of this function
as y since it is not bounded. She also realises that the way to x this is to integrate
from 0 to , rather than 0 to since it is a good t in this range . Work out expressions for
the displacement and momentum thickness in this case. Why is this procedure (integrating
only up to ) reasonable? Compare the results with those from part (a) to see the errors
involved in this approximation.
Note that this is a very common procedure since it allows for the use of functional forms that
are easy to analyse mathematically and only results in very small errors in the computed
quantities.
Answers: (a)/5, /10, /6, (b) All increase-faster in APG, (c) 0.193, 0.095
Question 2.
In another experiment a researcher decides that a better t to the velocity prole, in his
ow, is one in which the velocity gradient at the edge of the layer (y =
99
) is zero and
U/U
1
= 1 at y/
99
= 1. This is often a reasonable approximation since the velocity typically
asymptotes to the free-stream value extremely quickly. A suitable t to the experimental
results is U/U
1
= a
0
(y/) a
1
(y/)
2
a
2
(y/)
4
.
a) Apply the boundary conditions in order to relate a
2
and a
1
to a
0
. Write down the expres-
sion for U/U
1
. Note that the shape of the prole is now determined by only one parameter,
a
0
which can be varied to produce dierent prole shapes - try plotting the function for
various values of 0 < a
0
< 2.67 (this restriction gives physically plausible shapes).
b) Consider the case where a
0
= 2. This looks something like a zero pressure-gradient
boundary layer (actually the shape is not a great t to known solutions but that doesnt
matter here). Find the displacement thickness and the momentum thickness in terms of .
Find the shape-factor.
c) Consider the case where a
0
= 0. Find the displacement thickness and the momentum
thickness in terms of . Find the shape-factor. What physical situation does this prole
represent (consider the shear stress at the wall)?
Answers: (a) a
0
(3a
0
/2 2)
2
(1 a
0
/2)
4
, (b)

= /3, = 2/15, H = 5/2, (c)

= 8/15, = 8/63, H = 63/15 = 4.2, this is a separating (or separated) prole since the
wall shear-stress is zero.
6
3 The evolution, or growth of boundary layers
Now we have some way of characterising a boundary layer given its velocity prole (either
from experiment or theory) then we might ask how it develops along the surface of a body.
Does it get thicker or thinner? How does the shear-stress at the wall (and hence the skin-
friction) change and, on what does it depend. To answer this we need to look at the
equations of motion, in particular the momentum and continuity equations are quite useful
in this regard. In general for a viscous ow these are represented by the Navier-Stokes
equations and continuity. We are only interested in incompressible ow in this course so we
will assume constant density.
4 The Navier-Stokes Equations
see KUETHE & SCHETZE, Appendix B.8
Liepman & Roshko 13.113.7
Prandtl & Tietjens Applied Aero-Mechanics
Schlichting Boundary Layer Theory
Whilst we have already derived these equations in the rst part of 3A1, for completeness and
consistency we present them here again. Inviscid analysis suggests that a body in an inviscid
uid has zero drag. In order to resolve this paradox it is necessary to include viscous terms
in the equations. This was rst done by Navier in 1827 followed by other researchers but
the most elegant and complete form of the equations was derived by Stokes in 1845. This
set of equations for the MOMENTUM of a uid are known as the Navier-Stokes equations
and may be written
u
t
+u
u
x
+v
u
y
+w
u
z
=
1

p
x
+
_

2
u
x
2
+

2
u
y
2
+

2
u
z
2
_
v
t
+u
v
x
+v
v
y
+w
v
z
=
1

p
y
+
_

2
v
x
2
+

2
v
y
2
+

2
v
z
2
_
w
t
+u
w
x
+v
w
y
+w
w
z
=
1

p
z
+
_

2
w
x
2
+

2
w
y
2
+

2
w
z
2
_
(8)
In addition to these equations we also have the equation for conservation of mass known in
uids as the CONTINUITY EQUATION which is given by
u
x
+
v
y
+
w
z
= 0 (9)
It is often convenient to rewrite these equations in a more compact form using either VEC-
TOR notation or TENSOR notation. In vector notation
u
t
+ (u.)u =
1

p +
2
u
.u = 0 (10)
7
and in tensor notation
u
i
t
+u
j
u
i
x
j
=
1

p
x
i
+

2
u
i
x
j
x
j
u
i
x
i
= 0 (11)
where a repeated sux denotes a summation over all values of the sux (Einsteins con-
vention). These equations are generally accepted to be sucient to fully describe the ow
of any Newtonian uid. It should be pointed out that while these equations appear simple
they are extremely dicult to solve since they are NON-LINEAR.
4.1 Two-dimensional ow
In most of the following analysis we will restrict ourselves to two-dimensional ows. In this
case the Navier-Stokes equations become
u
t
+u
u
x
+v
u
y
=
1

p
x
+
_

2
u
x
2
+

2
u
y
2
_
v
t
+u
v
x
+v
v
y
=
1

p
y
+
_

2
v
x
2
+

2
v
y
2
_
u
x
+
v
y
= 0 (12)
ASIDE: If 0 then these reduce to the Euler equations. If we also specify STEADY
FLOW (i.e. ow which does not change with time) then the time derivatives are zero (i.e.
u/t = 0 and v/t = 0) and if all boundary conditions are irrotational this reduces to
POTENTIAL FLOW. We can then dene a stream function and we have a solution of
Laplaces equation
2
= 0 as we found in the rst part of 3A1.
We are, however, interested in the case when viscosity is important.
Consider rst the ow around a thin streamlined body, for example a thin aerofoil. Away
from the body the Euler equations apply and for steady ow the solution of Laplaces equa-
tion will give good results (and may be sucient in many cases). In the region close to the
surface a boundary layer exists and we need to include the viscous terms. If the curvature
of the surface is small then we can neglect it and just use the full two-dimensional equations
as given above in x and y. Now if the boundary layer is thin we can calculate the velocity
on the wall streamline, u
1
(x) using Euler and assume it is the same as the velocity at the
outer edge of the boundary layer.
Hence the boundary conditions for the solution of the equations are
y = 0 u = v = 0
y = u = u
1
(x) (13)
and in the free stream we know from Euler
u
1
u
1
x
=
1

p
x
(14)
8

l
y
x
Figure 3: Schematic diagram of the boundary layer
(cf. Bernoulli
1
2
u
2
1
+p
1
/ = constant).
So now we have set up the boundary conditions we need to see if we can simplify the equations
any further. In order to do this we rst consider Prandtls ORDER OF MAGNITUDE argu-
ment which may be used to simplify the full Navier-Stokes equations to a more manageable
form in the case where the boundary layer is thin.
5 The boundary layer equations - Order of Magnitude
argument.
The essential part of this argument is to recognize that boundary layers are (in general) thin
in comparison to their length of development (except perhaps right at the start of the body).
Hence /l is small, where is the thickness of the boundary layer and l is the length over
which it develops (see gure 3).
Now we will examine the order of magnitude of all of the terms in the two-dimensional
Navier-Stokes equation in the region close to the boundary. By order of magnitude we mean
the size of the terms, which we will represent by O(), i.e. this means is of the order of
magnitude of. This sort of argument is often called a scaling argument. In essence what
we are doing is looking at how the various terms in the equation change when we change
the primary ow variables (such as the mean velocity, the size of the object we are studying,
the viscosity of the uid). To say that one variable scales with another quantity simply
means that we expect it to increase proportionally when we increase that variable. To take
an example, the skin friction drag on a body is proportional to its surface area so we might
say that the skin friction drag scales with the surface area. This approach is very useful to a
scientist or engineer since in this way we can determine which terms of an equation are likely
to be important under certain conditions and then simplify the equation (by dropping those
terms that are likely to be insignicant). It should be noted that we do this all the time,
often almost unconsciously, when, for example we ignore relativistic eects in the equations
of motion of a body, or neglect quantum eects. We expect these terms to be very small
in the situations we encounter most of the time. These are extreme examples (where the
extra terms that would appear in the equations are extremely small) but in other situations
it is not as obvious which terms we can neglect. The order-of-magnitude analysis provides a
formal procedure (in the less obvious cases) by which we can simplify the equations. Now we
return to the two-dimensional Navier-Stokes equations and examine the order of magnitude
of the terms so we can consider in what circumstances we can neglect some. In particular
we are interested in the situation of a boundary layer developing on a wall. Starting from
9
left to right in the equations we nd,
u
1
(x) = O(U

)
u
x
= O
_
U

l
_
u
y
= O
_
U

_
(15)
Now near the surface the slope of the streamlines must be O(/l) and therefore
v
U

= O
_

l
_
v = O
_
U

l
_
(16)
(or we can get the same result using the continuity equation), show this. Also the second
order terms are

2
u
x
2
= O
_
U

l
2
_

2
u
y
2
= O
_
U

2
_
(17)
and the order of the other terms can be found by similar means.
Substituting these into the momentum equations gives
O
_
U
2

l
_
+ O
_
U
2

l
_
= O
_
1

p
x
_
+O
_

_
U

l
2
+
U

2
__
O
_
U
2

l
2
_
+O
_
U
2

l
2
_
= O
_
1

p
y
_
+O
_

_
U

l
3
+
U

l
__
(18)
Now in order to non-dimensionalise things we divide through by U
2

/l This leads to
O(1) +O(1) = O
_

(p/U
2

)
(x/l)
_
+O
_
1
Ul

_
+O
_
_
1
Ul

_
l

_
2
_
_
O
_

l
_
+O
_

l
_
. .
INERTIA FORCES
= O
_
1

(p/U
2

)
(y/l)
_
. .
PRESSURE GRADIENT FORCES
+O
_

l
1
Ul

_
+O
_
l

1
Ul

_
. .
V ISCOUS STRESS FORCES
(19)
Now, following Prandtl, we will examine what happens when U

l/ . Consider the
rst of the two momentum equations, bearing in mind that according to the boundary layer
hypothesis << l and hence (l/)
2
. The term in 1/Re goes to zero and the question
is what happens to the term
1
Re
_
l

_
2
(20)
as Re ??
10
Prandtl considered three cases:
CASE (a)
1
Re
_
l

_
2
>> O(1) (21)
This leads to a balance between the pressure gradient forces and the viscous forces. This
sort of balance occurs in low Reynolds number ows where the inertia is small. HOWEVER,
we are considering the case where Re . One possible case where this might occur is in
the fully developed ow in a parallel pipe or duct. BUT there is a problem with this. Since
the inertia terms have dropped out, then Reynolds number cannot be a ratio of inertia to
viscous forces so we have a paradox. This case doesnt make sense.
CASE (b)
1
Re
_
l

_
2
<< O(1) (22)
Here we have a balance between pressure gradient forces and inertia forces. This case is
O.K. but it simply represents the Euler equations. It suggests that viscous forces are not
important, but we know that they must be in a boundary layer. So this solution is not
useful for our purposes. This is where the classical theorists went wrong. They assumed
that their inviscid solutions (eg. potential ow) should become more accurate at higher
Reynolds number and perhaps exact in the innite Reynolds number limit. Experiments
showed that this was not the case at all (bodies did not seem to approach a situation where
they had zero drag, for example)
CASE (c)
1
Re
_
l

_
2
= O(1) (23)
In this case inertia, pressure gradient and viscous forces are ALL equally important together.
Consideration of this case led to the breakthrough of Prandtl. He recognised that, even in
ows at extremely high Reynolds numbers there would be a region near the wall where the
velocity dropped to zero and hence, in this region, the local Reynolds number of the ow
would be low and so viscous forces could not be neglected. So, regardless of the external
Reynolds number (eg. for a cylinder U

D/, there is some region close to the surface,


however small, where viscosity is important.
Rearranging the previous equation leads to

l
= O
_
1

Re
_
. (24)
Note that this tells us how the boundary layer thickness varies with distance along a plate
(how it grows). Substituting x, the distance from the leading edge for the length l and
multiplying across we nd
= O
_
x
U

_
. (25)
Since it is always of this order then we can say that the boundary layer thickness scales in
this way - we will nd the same result from an exact solution of the boundary layer equations
11
later on. Note also that (24) justies our original assumption that /l is small - at least at
reasonable Reynolds numbers. Note also that this Reynolds number is based on distance
along the plate so we would expect this assumption to break down very near the leading
edge where l is small and hence so is Re = U

l/.
We can substitute this back into our momentum equations to examine the consequences.
O(1) +O(1) = O
_

(p/U
2

)
(x/l)
_
+O
_
1
Re
_
+O(1)
O
_
1

Re
_
+O
_
1

Re
_
= O
_
(p/U
2

)
(y/l)
_
+O
_
1
Re
3/2
_
+O
_
1

Re
_
(26)
Now all the terms in inverse powers of Re go to zero as Re and hence p/y = 0. We
can now simplify the equations to
u
u
x
+v
u
y
=
1

p
x
+

2
u
y
2
0 =
p
y
(27)
The second equation suggests that p = f(y) and hence p = p
1
(x)
1
so to summarize and
include this we nally have
u
u
x
+v
u
y
=
1

dp
1
dx
+

2
u
y
2
u
x
+
v
y
= 0 (28)
These equations are known as PRANDTLS BOUNDARY LAYER EQUATIONS and they
become asymptotically correct as Re . To solve them we need boundary conditions
and these are
y = 0 , u = v = 0
y = , u = u
1
(x)
u
1
du
1
dx
=
1

dp
1
dx
(29)
These equations can be solved numerically or by series expansion for various cases and in this
course we will consider some exact solutions for laminar boundary layers - but a little later.
1
This is useful when we consider using the potential ow solution for the outer ow to determine the
pressure along the wall in order to calculate the boundary layer. Since the pressure does not vary through
the boundary layer then we can use the pressure-gradient at the wall from the potential ow as the pressure-
gradient in the boundary layer equations
12
Question 3.

U

L
H
h
s
u
Let us consider the order of magnitude argument for a much simpler equation. Suppose
we are examining the ow over a hill on the oor of a wind-tunnel. We are interested in
the pressure gradient along the surface of the hill and decide to use Bernoullis equation in
its dierentiated form. There is a potential energy term in the equation and we can write
Bernoulli as
p +
1
2
u
2
+gh = constant. (30)
Since we are interested in the pressure gradient then we take the derivative with respect to
s, the distance along the surface (say, just to be careful, outside the boundary layer). This
gives
p
s
+u
u
s
+g
h
s
= 0, (31)
or

p
s
= u
u
s
+g
h
s
, (32)
which is, of course the Euler equation. Now the question is Is it safe to neglect the grav-
ity term? - if we can the problem becomes much simpler. We might also be interested in
whether we can do that in the real case we are modelling which might be a large hill in a eld.
We will consider the case where the hill is long and low so H/L is quite small (H/L << 1).
The simplest way to decide on the importance of the terms is to estimate their size, or their
order of magnitude.
Perform an order of magnitude analysis of the equation and hence nd under what condi-
tions the contribution of the gravity term to the pressure-gradient may be neglected.
Answer: It depends on the non-dimensional number gH/U
2

, if this is small we can neglect


the gravity term. Note that this number is a variation on the Froude number.
Before proceeding to nd precise solutions of the equations we will look at the integral
constraints of the boundary layer as a whole. In other words we ignore the details of the
velocity proles and instead integrate them up and see what we can say about the general
growth of the boundary layer based on the fact that the rate-of-change of momentum must
be equal to the sum of the external forces. This should say something about, for example,
how the (missing) momentum thickness changes along a plate. There are two, equivalent
approaches from slightly dierent perspectives - one proceeding from the boundary layer
equations and the other from a straightforward control volume approach.
13
6 The Momentum Integral Equation
6.1 Integrating the boundary layer equations
There are many computational approaches which are based on approximations which yield
quick solutions of reasonable accuracy using the INTEGRATED form of the boundary layer
equations (see Schichting).
If we integrate the boundary layer equations from y = 0 to some point outside the boundary
layer y = h where h > and u = U
1
at y = h then we get
_
h
0
(u
u
x
+v
u
y
U
1
dU
1
dx
)dy =

(33)
and using continuity we can put
v =
_
y
0
u
x
dy (34)
which leads to
_
h
0
(u
u
x

u
y
_
y
0
u
x
dy U
1
dU
1
dx
)dy =

(35)
Now if we integrate BY PARTS and the second term becomes
_
h
0
_
u
y
_
y
0
u
x
dy
_
dy = U
1
_
h
0
u
x
dy
_
h
0
u
u
x
dy
therefore,
_
h
0
_
2u
u
x
U
1
u
x
U
1
dU
1
dx
_
dy =

o

_
h
0

x
[u(U
1
u)] dy +
dU
1
dx
_
h
0
(U
1
u)dy =

o

(36)
Now the displacement and momentum thicknesses as derived earlier are,

=
_

0
_
1
u
U
1
_
dy
=
_

0
u
U
1
_
1
u
U
1
_
dy (37)
Now let h

o

=
d
dx
(U
2
1
) +

U
1
dU
1
dx
(38)
and dividing through by U
2
1
and rearranging it is not dicult to show that we nd
d
dx
..
inertia forces
+
H + 2
U
1

dU
1
dx
. .
Pressure gradient forces
=
C

f
2
..
Skin friction forces
(39)
where
H =

= shape factor (40)


The boundary layer is often assumed to separate when H 3 to 4 from empirical data.
This equation can also be derived by considering forces on a small control volume.
14
6.2 The control volume approach (ab initio)
In order to nd the momentum integral equation we consider forces on a small control volume
which is part of the boundary layer. The rate of change of momentum of this control volume
must be balanced by the forces on the control volume (i.e. the wall stress and the pressure
gradient forces).
Consider a control volume
y
x
A
B
C
D
E F
h
u
p
u+u
p+p

w
x
Edge of boundary layer
Figure 4: Control volume for derivation of momentum equation
MASS FLOW BALANCE Mass ow in through AE into control volume AEFD is

_
h
0
udy (41)
Mass ow out of volume through DF is

_
h
0
udy +
d
dx
_

_
h
0
udy
_
.dx +O((dx)
2
) (42)
So the nett mass ow OUT OF the control volume AEFD is
d
dx
_

_
h
0
udy
_
.dx +O((dx)
2
) (43)
So now we can calculate the mass-ow through the top, i.e. EF If V
h
is the average velocity
over EF then the rate of mass ow through EF is
V
h
dx (44)
and therefore, by continuity,
V
h
=
d
dx
_

_
h
0
udy
_
dx +O(dx) (45)
So now we know the mass-ow through the sides of the control volume. We can now work
out the momentum ux.
15
MOMENTUM BALANCE We can work out the net ux of x-momentum through the control
volume and balance this with the x-direction forces acting on the control volume.
(

M
x
)
DF
(

M
x
)
AE
=
d
dx
_
_
h
0
u
2
dy
_
dx +O((dx)
2
) (46)
(

M
x
)
EF
= V
h
U
1
dx = U
1
d
dx
_

_
h
0
udy
_
dx +O((dx)
2
) (47)
Forces
Pressure force is x-direction on AEFD is h.dp
Pressure force = h
dp
dx
dx +O((dx)
2
) (48)
Skin friction force is
o
.dx
Hence balancing the net forces with the net eux of momentum
d
dx
_
_
h
0
u
2
dy
_
dx U
1
d
dx
_

_
h
0
udy
_
dx = h
_
dp
dx
_
dx
o
dx +O((dx)
2
) (49)
Now we divide through by dx and then let dx 0 and hence the term O((dx)
2
) 0 and
the equation becomes
d
dx
_
_
h
0
u
2
dy
_
U
1
d
dx
_

_
h
0
udy
_
= h
_
dp
dx
_

o
(50)
Now it only remains to rearrange this so we can express things in terms of

and where
as stated earlier

=
_

0
(1
u
U
1
)dy (51)
and
=
_

0
U
U
1
(1
u
U
1
)dy (52)
hence we get
d
dx
_
_
h
0
u(u U
1
)dy
_
+
dU
1
dx
_
_
h
0
udy
_
= hU
1
dU
1
dx

o
(53)
and rearranging this
d
dx
_
_
h
0
u(u U
1
)dy
_
+
dU
1
dx
_
_
h
0
(U
1
u)dy
_
=
o
(54)
which leads to
U
2
1
d
dx
+ 2U
1

dU
1
dx
+U
1
H
dU
1
dx
=
o
(55)
and nally we get the von Karman momentum integral equation
d
dx
+
H + 2
U
1

dU
1
dx
=

o
U
2
1
=
C

f
2
(56)
Some points to note are that
16
This equation is valid for both laminar and turbulent ow
It forms a good basis for approximate methods for solving boundary layer problems
It also always serves as a good check on both calculations and experimental data
Question 4.
A boundary layer develops with zero pressure-gradient. Its prole is measured at a
streamwise location and it is found that a good t to the function is given by U/U
1
=
1.54 (y/) 0.61 (y/)
4
for 0 < y < where is, say, the 99% thickness which is found
in this case to be 10mm. Another measurement is made 20mm downstream and the 99%
thickness there is found to be 12mm. The same function also ts the prole well at this
point.
Use the momentum integral equation to estimate the value of the skin-friction coecient
(C

f
) at this location. Comment on the likely accuracy of this estimate.
Answer: C

f
= 0.0257
Question 5.
Uniform suction at the plate is sometimes used to control boundary layers - in some cases to
maintain laminar ow and in others to reduce the thickness of the boundary layer. Consider
the ow of a boundary layer over a porous surface through which ow is sucked such that
the velocity normal to the plate is uniform and equal to V
o
.
We will assume that there is no streamwise pressure-gradient in order to simplify the problem.
Derive the momentum integral equation for this ow. Note that the boundary layer equations
are unchanged but the boundary condition at the wall is now dierent. In this case the
continuity equation may be written as
v =
_
y
0
u
x
dy V
o
(57)
which ensures that at the wall the vertical velocity component is equal to the wall suction
velocity. Note that if we dierentiate both sides with respect to y we get back the usual
continuity equation since V
o
is constant everywhere.
You can proceed either by integrating the boundary layer equations or by applying a control
volume analysis. If you are keen do the problem using both methods.
Answer: d/dx +V
o
/U
1
= C

f
/2
17
Question 6.
The boundary layer equations apply to any long thin ow - that is any ow in which
gradients across the ow are much larger than those along the ow. There was no explicit
mention of a wall in the derivation (the wall enters through the application of boundary
conditions when solving the equations). As a result it is possible to apply them to other
ows. Consider the 2D laminar wake behind a body (eg. an aerofoil) as shown. The wake
must have some decit in velocity since some momentum is lost from the uid due to the
drag of the body.

U
1
u(y)
x
y
flow
The equations are exactly the same as for the 2D boundary layer as given in (269). The
denition of the momentum thickness is the same here as for the boundary layer except that
the integration is from y = and y = +.
a) Derive the integral momentum equation in this case by integrating the boundary layer
equation with respect to y between these limits and hence nd an expression for d/dx. Note
that you can follow the basic procedure in the notes above. The end result is rather simple
and quite dierent from that for the boundary layer.
b)Could you have predicted this result without doing any analysis - if so what is the reason-
ing?
c) Repeat the process using the control volume method (the ab initio approach).
Answers: (a) d/dx = 0, (b) Yes
Laminar boundary layers
7 Similarity solutions
In this section some special exact solutions of the boundary layer equations will be considered
in which the shape of the proles are invariant as the ow develops. Before we derive these
solutions we will rst see what we can nd out about general constraints on the shape of the
velocity proles that are given by the equations.
18
7.1 The shape of the velocity prole
In order to examine some constraints on the shape of the velocity prole we will assume it
can be expressed as a Taylor series expansion in the vicinity of y = 0 i.e. at the wall. Hence
we may write,
U(x, y) = a
0
(x)y +a
1
(x)y
2
+a
2
(x)y
3
+a
3
(x)y
4
+... (58)
where the constants are functions of x. The rst constant term (in y
0
)is obviously zero since
the velocity goes to zero at the wall and so it has been dropped. Now we can substitute this
expansion into the boundary layer equation and equate the coecients of like powers of y.
We will retain only terms of order lower than y
3
- to simplify the mathematics and since it
turns out that the higher order terms do not give much useful information( it helps to know
the answer beforehand!)
a
0
da
0
dx
y
2
(1/2)a
0
da
0
dx
y
2
=
1

p
x
+(2a
1
+ 6a
2
y + 12a
3
y
2
) (59)
Now equating the coecients of like powers of y we nd
a
1
=
1
2
p
x
(60)
a
2
= 0 (61)
a
3
=
1
24
a
0
da
0
dx
. (62)
We also know that the shear stress at the wall is proportional to the velocity gradient at the
wall i.e.

w
=
U
y

y=0
= a
0
(63)
If we non-dimensionalise y with the local boundary layer thickness, and substitute back
into the expansion, then perform a little manipulation to put things in non-dimensional form
we nd
U
U
1
=
1
2
C

f
Re

+
1
2
Re

U
2
1
dp
dx

2
+
1
96
C

f
Re
3

dC

f
dx
2C

U
2
1
dp
dx
_

4
+... (64)
where = y/ and Re

= U
1
/ is the local Reynolds number based on the thickness.
The main things that can be learned from this process are;
There is no cubic term in the expansion and in zero pressure-gradient no quadratic
term
The second derivative is balanced by the pressure-gradient
The fourth order term depends on the gradient of the skin-friction - not just the local
skin friction
An adverse pressure-gradient reduces the velocity near the wall and a favourable does
the opposite (not unexpected) - note you need to consider the term of order
4
to see
this clearly.
Increasing Reynolds number increases the velocity near the wall
19
Another point that may be noted is that if all the coecients of powers of are constants
(i.e. they do not vary with x) then the shape of the prole stays the same for all x - this
special case is known as a similarity solution and this constraint on the coecients allows us
to nd some useful solutions of the boundary layer equations. Of course, with this limited
expansion, ensuring constant coecients does not necessarily guarantee that the shape is
invariant since higher order terms might vary with x - however it turns out that this expan-
sion is sucient. We examine true similarity solutions in the next two subsections of the
notes where we start by assuming that the complete shape does not vary with x.
Question 7.
Using the results for the Taylor-series expansion (only the terms given) derive the conditions
necessary for the shape of the prole given by U/U
1
= f(y/) to be invariant with stream-
wise distance x.
a) Use these conditions to nd the variation of and C

f
with x for the zero pressure-gradient
case.
b)In the case with pressure-gradient it is possible to do the same but the analysis is a
little tricky. The actual variations depend on the pressure-gradient. Instead show that
(U
1
/U
o
) = a(x/L)
m
and /
o
= b(x/L)
n
give a constant shape when m and n are related
in a particular way (U
o
and
o
are just some reference values used to make the expressions
non-dimensional - they might be the values at some upstream location. L is also just a
(constant) length used for non-dimensionalising the equations. a and b are constants). State
the appropriate relationship between m and n (check that it is consistent with the zero
pressure-gradient case). Note that the rst two terms of the expansion are sucient for part
(b).
You may use the result that U
1
dU
1
/dx = (1/)dp/dx (from Bernoulli outside the layer) to
relate the variation in U
1
to the pressure-gradient.
Answer:(a)
_
x/U
1
, C

f
1/
_
U
1
x/, (b) m + 2n 1 = 0
7.2 The zero pressure-gradient boundary layer
Now that we have derived the integral constraints on the equations and we have some idea
of the factors aecting the overall growth of the boundary layer, we can look more closely
about the actual shape of the boundary layer proles. We already know something about
the shape from the Taylor series expansion, but here we will nd the whole shape and the
precise variation of the integral parameters with streamwise distance. In order to do this
we need to solve the boundary layer equations to nd velocity proles that satisfy these
dynamical equations. This is not that easy in general since we are solving a set of PDEs and
there are a wide variety of possibilities. There are various numerical techniques for doing so
but before considering them we will look at some analytical solutions which can lead us to
a better understanding of the behaviour of the boundary layer and it develops and evolves.
One way of simplifying the treatment of PDEs is to look for similarity solutions in which the
boundary layer proles develop in such a way that they maintain their shape whilst growing.
These are the solutions we will consider in this section. The rst and simplest case is the
20
U
y
U/U
1
y/
1
4
x=1
U
1
=4
x=2
U
1
=5
x=3
U
1
=6
Figure 5: Collapse of experimental data using similarity scaling
zero pressure-gradient boundary layer.
In this case the pressure gradient is zero, so the equations we need to solve are
u
u
x
+v
u
y
=

2
u
y
2
u
x
+
v
y
= 0 (65)
with boundary conditions
u = v = 0 at y = 0
u = U
1
at y >>
u = U
1
at x = 0 (66)
where is a length scale which is proportional to and which will be dened later.
Prandtl suggested that a SIMILARITY SOLUTION may be valid i.e. we assume that
u
U
1
= F
_
y

_
(67)
in other words the shape of the proles stay the same when scaled with U
1
and this so
F(y/) is not a function of x. Now if we want to check this assumption we can plot u/U
1
versus y/ for dierent x positions and dierent free-stream velocities: we nd that they
do indeed seem to fall on a universal curve independent of U
1
, x and as shown in gure 5.
This then reduces our PDE to an ODE since we are looking at a quantity, or quantities
which are only functions of one variable (y/ in this case).
Now we can work out the terms in the boundary layer equations in terms of F. We will put
= y/ for convenience, and hence
u = U
1
F() (68)
and we note that the function F must have the following properties
F = 0 at = 0
F = 1 at = (69)
21
Now we can work out the terms in the equations in terms of F
u
x
= U
1
F

x
(70)
where

x
=
y

2
d
dx
. (71)
Note that the dash denotes dierentiation of this function with respect to the one variable
on which it depends, (similarly two dashes represent dierentiation twice - and so on).
Now to nd the terms in v we use continuity
v =
_
y
0
u
x
dy
=
_
y
0
d
dx
U
1

dy
= U
1
d
dx
_

0
F

d
= U
1
d
dx
_
[F]

__
Fd
_

0
_
(72)
which leads to
v = U
1
d
dx
_
F
_

0
Fd
_
(73)
Now in order to get rid of the integrals lets change the variable so that we only have D.E.s
rather than integro-dierential equations. We can do this by writing
_

0
Fd =
f
2
or F =
f

2
(74)
and hence we nd that the terms are
u = U
1
f

2
u
x
=
U
1

d
dx

f

2
v = U
1
d
dx
_

2

f
2
_
u
y
=
U
1
f

2
u
y
2
=
U
1
f

2
2
(75)
22
Now we can substitute these terms into the boundary layer equations to give;

U
2
1

d
dx

f

4
+
U
2
1

d
dx

f

4

U
2
1

d
dx
ff

4
=
U
1

2
f

2
(76)
and hence

U
1


d
dx
. .
function of x
=
2f

ff

. .
function of
(77)
Now since x and are independent variables then this equation is not possible unless both
sides are equal to a constant. By inspection this constant must be negative and hence we
reduce this to two O.D.E.s

U
1


d
dx
=
2f

ff

= (78)
where is a positive constant. The value of is arbitrary and its value only aects the
denition of which we have not dened yet.
What is IMPORTANT to note is that by assuming a SIMILARITY SOLUTION we have
changed the original P.D.E. into a pair of O.D.E.s.
For convenience and for HISTORICAL reasons we use = 2.
Now subsituting this in to the rst equation we nd
U
1


d
dx
= 2
_

0
d =
2
U
1
_
x
0
dx

2
2
=
2x
U
1
(79)
and hence we nd an expression for i.e.
= 2

x
U
1
(80)
which we actually indirectly found earlier by the order of magnitude argument. So we now
know the way in which the boundary layer grows with streamwise distance, x.
Now substituting = 2 into the second equation leads to
f

+ff

= 0, (81)
which is called the Blasius equation. It is a 3rd order non-linear D.E. with boundary condi-
tions
f

= 0 at = 0
f

= 2 at =
f = 0 at = 0 (82)
23
The third condition can be derived from v = 0 using continuity. Now, as engineers, what we
really want is the local skin-friction coecient since from this we can calculate drag. This
is dened as
C

f
=

o
1
2
U
2
1
(83)
and we know that

=
u
y
=
U
1
2
f

()
from which we nd that at the wall ( = 0)

=
U
1
f

(0)
2
=
U
1
4

U
1
x
f

(0)
=
1
4

1/2
U
3/2
1
x
1/2
f

(0) (84)
and substituting into the expression for C

f
we nd
C

f
=
f

(0)
2
1
(Re
x
)
1/2
(85)
Now since f is a universal function of then
f

(0)
2
= a universal constant (86)
The equation we have just derived for the local skin friction coecient is known as the
Blasius Law of Skin Friction for Laminar Flow.
The only thing we do not know in this expression is the value of the universal constant
f

(0)/2. In order to nd this we must solve the Blasius D.E.


To do this we can use a series expansion method. Assume:
f = C
o
+C
1
+C
2

2
2!
+C
3

3
3!
+...
f

= C
1
+ 2C
2

2!
+ 3C
3

2
3!
+... (87)
Now we know from boundary conditions that f(0) = 0 and f

(0) = 0 and so C
o
= 0 and
C
1
= 0 immediately. Now we can substitute these expressions into the Blasius D.E. and we
nd an expression as follows
C
3
+C
4
+ (
C
2
2
2!
+
C
5
2!
)
2
+... = 0 (88)
24
This series must converge for all i.e. the left-hand side must be equal to zero for all and
hence all the coecients of this series must be equal to ZERO. So
C
3
= 0
C
4
= 0
C
2
2
2!
+
C
5
2!
= 0 (89)
and so on. When this is done we nd that all non-zero coecients in the expression for f()
can be expressed in terms of C
2
and hence we get
f =
C
2

2
2!

C
2
2

5
5!
+
11C
3
2

8
8!

375C
4
2

11
11!
+... (90)
so the problem reduces to nding a value of C
2
which satises the outer boundary condition
as .
One simple approach is the SHOOTING METHOD. Here trial values of C
2
which satisfy the
boundary condtions are launched at = 0 and their behaviour for large is examined.
The discrepancy from the desired solution is noted and the value is adjusted and a new trial
is LAUNCHED.
C
2
=1

Correct
boundary condition
f (inf) =2
f
2
C
2
=2
C
2
=1.328
Figure 6: Shooting method
The wiggles in the solution at very large come from truncating the series. Blasius found
(by hand!) that C
2
= 1.32824 and since f

(0) = C
2
then this leads to
C

f
= 0.664(Re
x
)
1/2
(91)
and hence we know how the skin friction coecient varies with streamwise distance, x. Now
we havent dened in a simple physical way. Examining the Blasius solution found above
we nd that if is the 99% thickness of the layer then 2.5. The solution for the shape
of the prole is shown in gure 7.
25
0.0 1.0 2.0 3.0 4.0

0.0
0.5
1.0
1.5
U/U
1
Figure 7: The Blasius solution
REFERENCES: Duncan, Thom and Young, Mechanics of Fluids, p.261 Schlichting, Bounday
Layer Theory, p.121 Knudsen and Katz, Fluid Dynamics and Heat Transfer, p.253
NOTE: Schlichting uses a slightly dierent denition of i.e. = y
_
U
1
/x. Now we can
look at various cases of dierent pressure gradient and dierent wall shear stress.
7.3 The laminar boundary layer with pressure-gradient
In this case we use the full boundary layer equation
u
u
x
+v
u
y
=
1

dp
1
dx
+

2
u
y
2
u
x
+
v
y
= 0 (92)
Now to get a feel for the eect of pressure-gradient at the wall we consider what happens as
y 0. The inertia terms disappear since both u and v go to zero and hence we get
0 =
1

dp
dx
+

2
u
y
2
(93)
and we also know by denition that

=
u
y
(94)
(strictly there is a v/x term in this equation which is small according to the boundary
layer hypothesis.) and hence we nd
dp
dx
=
_

2
u
y
2
_
y=0
=
_

y
_
y=0
(95)
26
and of course

o
=
_
u
y
_
y=0
(96)
Falkner and Skan Solution Now then suppose we have a pressure gradient acting. Falkner
and Skan asked the question: What sort of pressure gradient will lead to a similarity solution?
The answer they found is there is a similarity solution for the case where the free-stream
velocity varies as a simple power law of the form
U
1
= ax
m
(97)
This applies to ow over a wedge
2
. Now the equations we want to solve are
u
u
x
+v
u
y
= U
1
dU
1
dx
+

2
u
y
2
u
x
+
v
y
= 0 (98)
and we proceed as before for the Blasius solution.
Put u/U
1
= F() where = y/ hence u = U
1
F() but now U
1
= (x).
u
x
=
dU
1
dx
F F

U
1

d
dx
(99)
Then we have to look at continuity, as before, to nd v
v =
_

0
Fd.
dU
1
dx
+U
1
d
dx
_

0
F

d (100)
and, as before, we put
_

0
Fd = f (101)
and hence
u
x
=
dU
1
dx
f

U
1

d
dx
(102)
and
v =
dU
1
dx
f +U
1
d
dx
[f

f] (103)
and as before
u
y
=
U
1
f

(104)
and

2
u
y
2
=
U
1
f

2
(105)
Then we can substitute this into the boundary layer equation leading to the complicated
expression
U
1
f

_
dU
1
dx
f

U
1

d
dx
_
+
_

dU
1
dx
f +U
1
d
dx
[f

f]
__
U
1
f

_
= U
1
dU
1
dx
+
U
1

2
f

(106)
2
You can convince yourself of this by working out V
1
from the continuity equation and examining V
1
/U
1
27
Now expanding this
U
1
dU
1
dx
(f

)
2

U
2
1

d
dx
f

U
1
dU
1
dx
ff

+
U
2
1

d
dx
f

U
2
1

d
dx
ff

= U
1
dU
1
dx
+
U
1

2
f

(107)
which becomes

dU
1
dx
(f

)
2

dU
1
dx
ff

U
1

d
dx
ff

=

2

dU
1
dx
+f

(108)
Now we want f to be a universal function of alone. This will be true if the coecients of
the terms in f are UNIVERSAL CONSTANTS. i.e. if

dU
1
dx
=
1
and
U
1

d
dx
=
2
(109)
where
1
and
2
are some universal constants.
We will rst examine what happens if we assume that the rst coecient is a universal
constant and see if this is consistent with the second also being a universal constant.
We return to the assumed form for the mean velocity variation
U
1
= ax
m
dU
1
dx
= amx
m1
therefore

.amx
m1
=
1
=
_

1

amx
m1
(110)
We can now nd d/dx
d
dx
=
_

am
_
1 m
2
_
x
1m
2
1
(111)
and hence
U
1

d
dx
=
1
1 m
2m
(112)
which is a constant for a given value of m as required. Now substituting back into the
boundary layer equation

1
(f

)
2

1
ff

1
1 m
2m
ff

=
1
+f

(113)

1
(f

)
2
ff

1
2m+ 1 m
2m
=
1
+f

(114)

1
(f

)
2
ff

1
m + 1
2m
ff

=
1
+f

(115)
f

+
1
ff

m+ 1
2m
+
1
[1 (f

)
2
] = 0 (116)
28
Now since
1
is arbitrary then let

1
_
m+ 1
2m
_
= 1 (117)
now the chosen value of
1
will have an eect on the denition of as before. Also for
historical reasons we will write
1
= Hence we nd
f

+ff

+[1 (f

)
2
] = 0 (118)
with boundary conditions
= 0 , f

= f = 0
= , f

= 1 (119)
This is a third order ordinary dierential equation which can be solved numerically or by a
series expansion method.
The ow U
1
= ax
m
corresponds to the ow over the surface of a wedge of semi-angle
m
m+1
or

2
for m positive. When m = 0 we recover the Blasius solution i.e. a at-plate with zero

Figure 8: Flow corresponding to Falkner-Skan solution


pressure gradient. Figure 9 shows solutions for several values of m from strong favourable
pressure gradient (m = 1) to the case in which the boundary layer is on the verge of
separation (m = 0.091). Separation will be discussed in the next section but we note here
that the point of separation (where the boundary is on the verge of separation, or has just
started to separate) corresponds with zero skin-friction. Since the wall-shear stress is given
by
w
= (u/y)|
y=0
then this occurs when the gradient of the graph in the gure is zero
(as it appears to be in the separating case). Comparing the gradients at the wall then it
is obvious that the strong favourable pressure-gradient case has a larger wall shear-stress
than the other two (and hence a larger skin friction). A further obvious feature is that the
favourable pressure-gradient case has a larger area under the graph than the other two. In
aerodynamics jargon we would say that the prole is fuller or more full. This is just
another way to say that the skin friction is larger and a fuller prole will be more resistant
to separation as a result (it can go further before the skin friction is reduced to zero).
The m = 1 case is a special case of a favourable pressure-gradient where the ow corresponds
with a plane stagnation point ow. In this case the boundary layer thickness is constant
29
0 1 2 3 4 5

0
0.2
0.4
0.6
0.8
1
u/U
1
m=-0.091 (separating)
m=0
m=1 (Stag)
m=0 .33
m=-0.065
Figure 9: Falkner-Skan solutions for various pressure-gradients (plots courtesy of David
Dennis)
with x. This case was discussed in the rst part of 3A1 where it was used to explain the idea
of the displacement thickness of the boundary layer. It is an EXACT SOLUTION of the
Navier-Stokes equations and is also useful as a starting point for numerical schemes since
this ow occurs at the nose of bodies where the boundary layer assumptions are not valid.
This ow pattern is shown in gure 10 and the boundary layer proles are shown in gure 11.
Figure 10: Plane stagnation point ow
7.4 The boundary layer as vorticity
Now that we have solved the boundary layer equations for some cases we can go on to look
at the ows in a slightly dierent manner. The vorticity in a two-dimensional ow is given
30
Figure 11: Boundary layer proles in plane stagnation point ow
by
=
v
x

u
y
. (120)
The boundary layer approximation suggests that the rst term is very small relative to the
second and so we will ignore it. If we now examine the Blasius solution we can dierentiate
the velocity prole to nd the vorticity. Simple inspection of the prole suggests what this
will look like but it is plotted in gure 12.
0 1 2 3 4 5
y/
0
0.1
0.2
0.3
0.4
0.5

/
U
1
Figure 12: The Blasius vorticity distribution
We see that the boundary layer is simply a layer of vorticity near the wall (it dies o rather
rapidly at large distances from the wall). So an alternative way of looking at a boundary
layer is as a layer of vorticity. To examine how this evolves (in this laminar case) we can use
31
the vorticity equation which is, in 2D
3
,
D
Dt
=
_

x
2
+

2

y
2
_
. (121)
The rst term on the right is very small under the boundary layer approximation and we
can consider the material derivative on the left as being the rate of change with time when
moving with the uid - in this case we can just consider it a normal derivative, hence

t
=

y
2
. (122)
The solution to this equation is a Gaussian vorticity prole of the form
=
o
e
y
2
/4t
. (123)
(see the examples question in which you can derive this). A measure of the thickness of the
layer is
vort
=

4t (which is the standard deviation of the Gaussian). Finally we ask how


we relate time in the moving frame to our xed frame - moving at a velocity U
1
in a time,
t, we cover a distance of x = U
1
t metres and hence we can relate the time of the moving
observer to the distance along the plate t = x/U
1
. So then the thickness of this layer should
grow with distance as
vort
=
_
4x/U
1
- which is exactly the growth rate we worked out
for the Blasius layer using the similarity solution. Note that the actual shape of the Blasius
vorticity prole is not quite Gaussian due to the terms we have neglected. Nevertheless it is a
reasonable approximation for our purposes. This approach clearly illustrates the boundary
layer as a convection/diusion problem - convection along the wall due to the ow and
diusion out by the gradients.
What is the eect of Reynolds number then on the boundary layer thickness? If we slow
down the ow (for example) then for a given distance the layer has a longer time to diuse
and hence we expect thicker boundary layers at low Reynolds number (similarly if we increase
the viscosity then the vorticity diuses more quickly and we nd the same result).
We can extend this conceptual model to give a qualitative idea of the eect of pressure-
gradients. Consider for example the case of an adverse pressure-gradient (APG). In this
case the velocity along the wall is dropping with streamwise distance. Our vorticity still
diuses according to =

4t but when we want to convert this time into distance then


we need to use a decreasing convection velocity and hence in APGs the layer grows faster
with respect to x (when compared with the ZPG case). A similar argument suggests that
boundary layers in favourable pressure-gradients grow more slowly with respect to x. It is
even possible to push these simple ideas a little further. Consider that the vorticity moves
with the free-stream velocity (or some fraction of it). Then at a given x location it has been
travelling for a time
t =
_
x
0
1
U
1
(x)
dx (124)
and hence the thickness should be
=

4
_
x
0
1
U
1
(x)
dx. (125)
3
bear in mind that the equation in three dimensions has vortex stretching and tilting terms not present
here
32
If we consider a power-law variation of external velocity (U x
m
as for the Falkner-Skan
solutions) then it is not dicult to show that
=

4
a(1 m)x
m1
. (126)
which is the same dependence on x as was found from the similarity solutions (the constants
are a little dierent). Note that the case of m = 1 (stagnation point) is dicult for this
simplied model since if we start from the origin then it takes innite time to move away
due to the zero velocity. Bear in mind that in this approach we have neglected various
additional eects, such as the eect of the vertical component of velocity which can be very
important. For example in the case of suction the dominant convection of vorticity is suction
towards the wall which is balance by the diusion of vorticity away from the wall. Also in
the stagnation ow the boundary layer equations break down near the origin.
So we have shown that the laminar boundary layer is a layer of vorticity at the wall. This
concept is often very useful when trying to understand the behaviour of boundary layers and
is also used to model them. One way of doing this is to consider the boundary layer as a row
of line vortices along the surface - they represent, in a discrete fashion the continuous layer
of vorticity. In particular the concept of separation is more intuitive when we consider the
layer of vorticity separating from the wall and in a model we might allow the line vortices to
leave the surface and evolve freely in the outer ow (where they will move under the veloc-
ities induced by the other line vortices in the ow. Note that there are more complications
in correctly modelling a boundary layer in this manner but the concept is essentially the same.
33
Question 8.
As we noted above the boundary layer can be considered to be a layer of vorticity near the
wall. In this question we consider a layer of vorticity in an otherwise still uid (apart from the
velocity induced by the vortex layer) - away from any boundary. We can use the similarity
method above to work out how the layer develops (we have already discussed the solution in
the notes but here we will derive it as an exercise in nding similarity solutions). Consider
a thin layer of vorticity in an otherwise irrotational ow. We consider that it extends to
innity in the positive and negative x-directions. In this case it is indeed very long and thin
and so the boundary layer approximation applies (exactly) since if it is uniform along its
length there are no gradients in that direction (they are not just small as they are for the
Blasius solution). In this case the solution is governed by the vorticity equation of this form

t
=

y
2
. (127)
Now we look for a similarity solution where the shape of the vorticity prole remains the
same with time but it spreads out and the peak vorticity drops in some fashion i.e.
=
o
f(y/) =
o
f(), (128)
where is some measure of the thickness of the layer and
o
is, say the peak vorticity at the
centre of the layer. We have introduced = y/ just to simplify the notation. Note that
this is an unsteady problem and hence
o
and are both functions of time. Note also that
since they are single values they are not functions of position, y. Now you can substitute
this function into the equation and nd an ODE for the function f(y/). Note as a hint
that since f depends on then f/t is not zero - you must use the chain rule eg. f/t
=(f/)(d/dt=(f/)(d/d)(d/dt) etc.
a) Find the ODE for the function f
b) Find the function f either by solving the ODE (not too hard) or by guessing the solution
and showing it satises the ODE.
c) Find expressions for the variation of and
o
with time, t
Answers: see notes
34
Question 9.
A two dimensional laminar jet issues from a slot into a still uid. At some point downstream
the centreline velocity is U
o
and the width of the jet is (this might be measured by the
point at which the velocity drops to half U
o
for example). Assume that a similarity solution
of the form
U = U
o
f
_
y

_
(129)
is applicable.
a) Find the ODE for the velocity prole in this case. Note that the boundary layer equations
hold for this ow.
b) Find the condition on the variation of U
o
and such that the shape of the prole is
invariant with streamwise distance.
c) Using the result from (b) and the momentum integral equation (which is the same as for
the laminar wake in question 6) nd the functional variation of U
o
and with x. (Note there
is no streamwise pressure-gradient in this problem).
Answers: (a) f

= a(f
2
I
1
f

) bI
1
f

, where a = (
2
/) dU
o
/dx, b = (U
o
/) d/dx,
I
1
=
_

0
fd (b) a = constant, (c) U
o
1/x
1/3
, x
2/3
Question 10.
Consider the laminar wake from question 6. Assume a self-similar distribution of the velocity
given by
U = U
1
U
o
(x)f(), (130)
where = y/.
It can be shown that for the zero pressure-gradient case (in which we are interested) a
similarity solution can only apply in the limit where U
1
/U
o
. Since the wake spreads
and the decit velocity drops o then this corresponds to a long way downstream. Consider
this case. (Note you cannot just say U = U
1
since this corresponds to no wake at all).
a) Find the ODE for the velocity prole f()(do not solve it) - retain only the largest terms
in the limit.
b) Find the variation of the thickness with x by assuming similarity applies.
c) Use the momentum integral equation (derived earlier) with the result from (b) to nd the
variation of U
o
with x.
Answers: (a) f

+ af

bf = 0 where a = (U
1
/) d/dx, b = (U
1
/)(/U
o
) dU
o
/dx,
(b)

x, (c)U
o
1/

x
7.5 The thermal and concentration boundary layers
The previous subsection considered the boundary layer as a layer of vorticity governed by
the vorticity equation, which, after application of the boundary layer approximations in 2D
is
D
Dt
=

y
2
. (131)
This is interesting since the same equation applies for the diusion of heat or of a passive
scalar (a passive scalar is something that does not change the ow and is not a vector eg.
concentration of dye in water) except that we replace with the appropriate diusivity for
the quantity of interest. Hence we also have thermal boundary layers and concentration
35
boundary layers governed by the same equations. The equation is essentially a convection-
diusion equation and the evolution of all these three quantities is determined by how they
are moved around by the ow (convection) and how they diuse out due to gradients (dif-
fusion). So the equation for concentration of a passive scalar after applying the boundary
layer approximations is
Dc
Dt
= D
o

2
c
y
2
, (132)
where D
o
is the diusivity of the substance. In the case of the thermal boundary layer we
consider the case of ow over a heated plate. We will make the additional assumption that
the temperature dierences are not too large and so we dont have to worry about buoyancy
forces - we could make this more rigorous but for now just accept the assumption. In this
case then we can write an equation for the convection/diusion of heat
DT
Dt
=

2
T
y
2
, (133)
where T is the local temperature of the uid and = /c
p
is the thermal diusivity (
is the conductivity). We could use the same method of similarity to nd solutions of these
equations but since the form of these equations is identical we can jump straight to the
solution. Then the thickness of the thermal boundary layer in laminar ow and developing
in a zero pressure-gradient is simply given by

T
=
_
4x/U
1
(134)
and we see immediately that the ratio of the thickness of the momentum boundary layer (or
layer of vorticity) to the thermal boundary layer is simply
_
/. The quantity under the
square-root is the Prandtl number. So we see that for high Prandtl number ows (such as
liquids or oils) the thermal boundary layer is much thinner than the momentum boundary
layer and hence its behaviour is govern by the ow very near the wall (at very high Prandtl
number it is immersed in the linear region right at the wall). In very low Prandtl number
ows it extends well beyond the momentum boundary layer and into the region of uniform
ow outside. In air Prandtl number is of order one and so the layers are of comparable
thickness.
In the case of the concentration boundary layer the ratio of the momentum boundary layer
thickness to its thickness,
c
is
_
/D
o
. The quantity here under the square-root is called
the Schmidt number. In water it has a value of about 1000 which means that momentum
(and vorticity) diuse about 30 times faster than dye in water.
The thermal boundary layer equation can also be derived by considering the transport of
enthalpy. The left-hand-side represents the heat transferred into a small control volume by
the ow of uid and the right-hand-side represents heat transfer from/to the small control
volume due to conduction. Near the wall the convective terms become small (eg. UT/x)
since the velocity goes to zero and hence the heat transfer (per unit area) from the wall is
simply given by Q = T/y|
y=0
. The analogy with the shear-stress at the wall is clear and
in that case we could think of the transfer of momentum from the wall as being due to the
gradient of momentum (U/y). There are two usual wall boundary conditions considered
when solving this equation which are constant heat ux and constant temperature. The
adiabatic wall is one example of the rst boundary condition in which the heat ux is zero.
36
y

T y ( )

T


T
w

T y ( )

T


T
w

T y ( )

T


T
w
Hot wall Cold wall Adiabatic wall
gradient
gradient gradient
Figure 13: Examples of thermal boundary layers
If it is not zero the wall can either be hot or cold and hence the heat transfer can be either
to or from the wall - this implies that the gradient of the temperature at the wall can be
of either sign. Some examples are shown in gure 13 to give some feel for what a thermal
boundary layer might look like. Note the temperature gradients at the wall in each case.
One further note regarding the thermal boundary layer equation is that it is linear and hence
it is possible to produce additional solutions from known solutions by superposition. That
is not true of the boundary layer equation (the momentum boundary layer).
8 Boundary layer separation
The Falkner-Skan solution produced some solutions for boundary layers subjected to pressure-
gradients and we saw that, in one case (m = 0.091), the prole corresponded to a bound-
ary layer on the verge of separation. This case has zero skin-friction and the point of
zero skin-friction is dened as point of separation. If we were to proceed to larger adverse
pressure-gradients then the ow near the wall would reverse and hence we would get negative
skin-friction. The problem with trying to analyse ows past separation is that the bound-
ary layer approximation is no longer valid and hence the boundary layer equations are not
either. Once past separation we are forced back to using the Navier-Stokes equation since
gradients in both directions can be signicant. It is also worth noting that separation in
practical applications also often leads to three-dimensional ow patterns in which even our
two-dimensional assumptions can break down (it is possible to have a purely two-dimensional
separation in some cases but it is not common in real situations).
Despite the complexity of separation we can, at least, examine some of the mechanisms
involved and the features of the ow. First, for interest, return to the idea of the boundary
layer as a layer of vorticity. We have already touched on how this relates to separation
earlier and here we just make some additional points. Consider the separating prole shown
in gure 7.3. Without any calculations we can see that near the wall the gradient of the mean
velocity, U/y is close to zero and hence the vorticity is almost zero right near the wall. It
obviously increases as we move further from the wall and then dies to zero at large distances.
Hence in this case we have a layer of vorticity that is separated from the wall by a region
of zero (or at least very small) vorticity. If the ow reverses near the wall then we have
opposite-signed vorticity right near the wall, then a region of almost zero vorticity and then
37
y
u
U
1
u(y)
y

+
-
+ - - - - - - - - - - - + +
-
-
-
Conceptual model using small vortices
Figure 14: Separated boundary layer prole showing vorticity and a simple conceptual model
the vorticity from our original layer further out. It is often useful to keep this conceptual
model in mind when trying to understand boundary layer behaviour. The conceptual model
given is not complete but it can provide a useful mental picture of separation and boundary
layers in general.
8.1 Separation criteria
When calculating boundary layers or just attempting to predict their behaviour it is useful
to have a criterion to determine where separation is likely. The obvious and most rigorous
is to dene the point where the skin-friction goes to zero but this is sometimes hard to deal
with mathematically or computationally. As a result various researchers have proposed other
criteria. One simple one is based on the shape-factor. This increases in adverse pressure-
gradients and so it is possible to propose a single value at which separation occurs. A value
of around 3.7 is often used for laminar boundary layers (see Thwaites method later in the
notes). There are also other criteria which provide ways of estimating the separation point
from the external ow behaviour. None of these criteria is perfect since the separation point
depends on the shape and the history of the boundary as it develops.
9 Arbitrary Pressure Gradients
The previous analytical solutions were for special cases where an analytical solution could be
found. In the case of arbitrary pressure gradients then it is necessary to solve the equations
numerically or by more complicated series methods. There are also approximate methods
that make some assumptions about the proles and reduce to much simpler solutions. We will
rst discuss an approximate method and then discuss the more general numerical approach.
The approximate methods were derived at a time before widespread digital computing and
were designed so that hand calculations (albeit lengthy) are possible. They still have some
nice features which are of use in understanding and dealing with simple problems. The aim of
38
such methods is not to give a detailed solution but to work out how the integral parameters
of the ow (such as skin-friction, boundary layer thickness) vary as the ow develops.
9.1 Thwaites Method
(see Duncan, Thom and Young pp.273278)
In the general case of an arbitrary pressure-gradient we can no longer assume that the
velocity prole remains self-similar - the shape may change as the ow develops. If we are
only interested in the integral properties of the boundary layer (eg. how thick it is, the
value of the skin-friction etc.) then we might instead make the next simplest assumption
that is that the shape is no longer invariant but can be expressed in terms of one extra
parameter - hence by giving a value to this parameter we determine the shape. This type
of method is classed as an integral method, since the aim is to solve the momentum integral
equation for the variation of the integral parameters of the ow as it develops. Recall that
the momentum integral equation relates three basic parameters (the momentum thickness, ,
the shape parameter, H, and the skin friction coecient C

f
) hence we need three equations
in total: the momentum integral equation provides one.
In Thwaites method he dened a parameter, m, related to the shape of the prole as
m =

2
U
1
_

2
u
y
2
_
w
(135)
where the subscript w means at the wall (y = 0). This is directly related to the applied
pressure-gradient since at the wall

2
u
y
2
_
w
=
1

dp
dx
= U
1
dU
1
dx
. (136)
so we can also write m as,
m =

dU
1
dx
. (137)
He also dened a non-dimensional skin-friction parameter as
l =

U
1
_
u
y
_
w
(138)
which is the thing that we might wish to nd. So the situation is that we have a form
parameter, m which in some way uniquely determines the shape of the velocity prole and a
skin-friction parameter, l that is related to the skin-friction and must be a function only of
m since the skin friction depends only on the shape of the prole and the prole is uniquely
determined, in some fashion, by m. Note also that the shape-factor H that appears in the
momentum-integral equation must also just be a function of m since it is determined by the
shape of the prole. Now if we use the momentum-integral equation we can nd an equation
that relates l to m and H and the growth of the momentum thickness. If we knew l as a
function of m and H as a function of m then we could simply integrate the equation and
nd how the momentum thickness and skin-friction vary with x for any pressure-gradient.
The way Thwaites proceeded originally take the momentum integral equation and using the
above denitions with a little manipulation derive this equation
U
1
d
dx
(
2
) = 2[m(H + 2) +l] = L(m), (139)
39
where the L(m) = 2[m(H +2) +l] which is, of course only a function of m. So we now nd
we have an equation with an expression for the change of momentum thickness on the left (
actually of
2
- but we can nd the growth rate from this) and on the right something that
is a function only of m. So if we can nd the function on the RHS we can then go on to
predict the growth of the boundary layer. To nd this function Thwaites took the results
for various exact similarity solutions and plotted L(m) versus the value of m - he found that
a good t was given by L(m) = 0.45 + 6m - although there was some scatter especially for
adverse pressure gradients (the reason for the scatter is that, in reality, the proles cannot
be expressed simply in terms of one parameter, m - but its not a bad approximation to the
truth). With this (or any other) relationship that gives L in terms of m then the momentum
integral equation simplies and , in the case of this t, we nd
U
1
d
dx
(
2
) 0.45 + 6
dU
1
dx

2
= 0 (140)
which can be rearranged
d
dx
(U
6
1

2
) = 0.45U
5
1
(141)
and solved
[
2
]
x
1
=
0.45
(U
6
1
)
x
1
_
x
1
0
U
5
1
dx (142)
where the particular quantities are evaluated at some point along the ow where x = x
1
.
Now the way we might use this is to specify the variation of U
1
with x in which we are
interested. This might come from solving the potential ow equations as we did in the rst
part of the course - this would prescribe the ow outside the boundary layer and hence the
variation of U
1
. Then we can substitute that into the above equation and integrate to nd
the momentum thickness at any point of interest.
Once the variation of has been evaluated then an additional equation is needed to solve
for the skin-friction variation and the variation of the shape factor (if required). Thwaites
proposed additional empirical relationships but perhaps the simplest are
H = 2.62 + 3.75m+ 5.24m
2
0.1 < m < 0 (143)
H = 2.472 +
0.0147
0.107 m
0 < m < 0.1
These combined with the equation above provide sucient information to solve for all the
relevant parameters given an applied pressure-gradient or, equivalently, variation of external
velocity(remember that we needed two additional equations which are the t of L(m) and
the relationships above for H(m)). It is as well at this point to note the limitations of this
sort of method. It assumes that all of the appropriate local parameters (such as skin-friction
(l), momentum and shape factor) depend only on the local value of the pressure-gradient
parameter m. This ignores the history of the boundary layer which may have undergone a
variety of changes in external ow before reaching the point of interest. Only the integral of
that history is taken into account. This is a fairly reasonable approximation for ows that
are slowly and smoothly developing as may be typical over normal aerofoil shapes.
Note that the curve-t of L(m) versus m was not precise and, as noted, showed some
scatter. In order to provide a better t, for the case of velocity proles over aerofoils,
Thwaites considered only the exact solutions that are most relevant to typical aerofoil ows
and tabulated the values of m, l(m) and H(m) which produce more accurate results than the
40
simple linear t above (L(m) = 0.45+6m). A table of these values may be found in Duncan,
Thom and Young, P. 275. The discussion and analysis in this section is based closely on
that provided in this reference.
Question 11.
Suppose that for your ow of interest over the rear of an aerofoil of chord length C, say,(strong
adverse pressure-gradients) a reasonably good t to the known exact solutions for use in
Thwaites method is L(m) = 10m (its really not bad for 0.06 < m < 0.08). Suppose also
that you have a measurement of the boundary layer at the start of the streamwise region of
interest (x = x
o
) where the velocity outside the layer, U
1
(x
o
), is known as is the momentum
thickness, (x
o
).
a) What value of the shape factor, H does this t for L(m) versus m predict will occur at
separation?
b) Derive an expression for the momentum thickness, (x) in terms of the variation of the
outer ow, U
1
(x) and the initial conditions ((x
o
), U
1
(x
o
)).
c) Assume that the external ow varies as U
1
(x) = U
1
(x
o
)((x x
o
)/C)
1/6
. Find the varia-
tion of (x) as a function of x and hence nd the variation of the skin-friction coecient as
a function of x x
0
/C and H.
d) Using the result of part (c) and assuming a much simplied curve-t for H i.e. H = 1+32m
nd the location at which separation is predicted. Note the curve-ts here are chosen for
simplicity of analysis rather than accuracy. In order to achieve a concrete answer assume
that the initial Reynolds number U
1
(x
0
)(x
o
)/ = 100 and (x
o
)/C = 0.01.
Answers: (a) 3, (b)m =
1
6
(xo)U
1
(xo)

(xo)
C
_
xxo
C
_
1/2
, (c)C

f
=
1
3
(xo)
C
_
xxo
C
_
1/6
(3 H),
(d)(x x
o
)/C = 0.1406
9.2 Karman-Pohlhausen method
We will discuss the numerical approach (see Schlichting). According to Schlichting the
boundary layer momentum equations with the the continuity equation can be put in the
form
u
x
= (
u
y
_
y
0
1
u
2
_

2
u
y
2
f(x)
_
dy +
1
u
_

2
u
y
2
f(x)
_
) (144)
where
1

dp
dx
= f(x) (145)
f(x) can be thought of as a forcing function. We imagine that f(x) is known by solving
the Laplace equation for ow about the body, ignoring the boundary layers. We are mostly
interested in the cases where this forcing is due to the pressure-gradient but in other situ-
ations it might be due to some other force acting (eg. coriolis or electromagnetic or some
other imposed forcing).
Given this, the right hand side of the equation involves only derivatives and integrals with
respect to y. The left hand side has only derivatives of x.
Assume curvature of body is suciently small so that we can atten out the bodys surface.
First we start with an initial prole which must be given. Usually it is expressed in terms
of a series expansion. Now since we know the initial prole (i.e. we know u as a function of
y) and we know (1/)dp/dx from potential ow theory then we know u/x.
41
y
x x
o

u y
u
x
x y x
o
( ) ( , ) +


u y ( )
Figure 15: Abitrary pressure gradient calculation
Now for the next prole at x
o
+ x then
u(x
o
+ x, y) = U(x
o
, y) +
u
x
(x
o
, y)x (146)
and so we can proceed along the wall and compute the rest of the boundary layer. Note: we
must be careful since

2
u
y
2
f(x)
u
2
(147)
must remain bounded for y 0. Since
lim
y0
_
_

2
u
y
2
f(x)
u
2
_
_
=
0
0
. (148)
Note that this comes from the fact that the pressure-gradient at the wall is balanced by

2
u/y
2
- consider the boundary layer equations as the wall is approached where the inertial
terms go to zero since the velocity goes to zero. The problem is avoided as long as the
numerator goes to zero faster than the denominator.
This approach is good as long as the boundary layer stays attached to the body (NO SEP-
ARATION).
Now for this rst approximation we have assumed the boundary layer thickness was neg-
liglible when calculating 1/dp/dx from potential ow. But now we know the boundary layer
displacement thickness which eectively alters the shape of the body as seen by a potential
ow calculation. Hence we can now iterate and nd the pressure gradient from potential
ow for our new body shape which is the body plus the displacement thickness of the
boundary layer. Hence we can keep iterating until the solution converges to the required
degree of accuracy.
This approach is O.K. as long as the boundary layers remain attached. The good thing is
that if they remain attached in the computation then they will remain attached in real life
so we know when the method breaks down.
Once they do separate we must use the full Navier-Stokes equations for the separated region
and this region is usually unsteady for
Re =
U
1
l

> 100 (149)


42
In order to compute this separated region we need to use one of the worlds largest super-
computers at least and at high Reynolds number even this may not be sucient. Fortunately
a lot of Heat Transfer problems occur at low Reynolds number and then we can use this
method (eg chip cooling problems).
43
10 Turbulence
10.1 Transition to turbulence in boundary layers
Let us return to the IB boundary layer experiment. In that experiment we started with
a smooth at plate and measured the boundary layer at a certain speed (around 11 m/s).
The boundary layer was laminar and the velocity prole matched the Blasius solution quite
well. Then a very small rod was added across the ow near the leading edge and the mean
velocity prole was measured again at the same speed. Despite the fact that the outer ow
was the same and the boundary conditions at the plate (in the region of interest) were the
same the prole was very dierent. What does this mean? Applying a small perturbation
(or kick) to our Blasius boundary layer changed the solution completely. So then the
Blasius boundary layer in the experiment was unstable to this perturbation. If it were stable
it would have recovered and returned to the same solution. The resulting solution after the
perturbation was a turbulent boundary layer which is very dierent in shape and is com-
plicated and unsteady with the velocity uctuating signicantly with time (we measured
that indirectly by listening to the ow with the stethoscope - the velocity uctuations result
in pressure uctuations which can be heard as sound). Despite the unsteady uctuations
we could measure the time-averaged (mean) velocity prole since the time response of the
pitot-tube and manometer are so slow they average out these uctuations.
This instability of laminar boundary layers is quite common at higher Reynolds numbers
and so turbulent boundary layers are quite common - so we need to understand something
about them.
First consider a rough argument for what is going on. When we add a perturbation to a
laminar boundary layer then, mostly, the viscous stresses act to damp out this perturbation
4
,
while the non-linear terms (the inertia forces) tend to increase it. As the Reynolds number
is increased then, by denition, the inertia forces become larger relative to the viscous forces
and hence the damping of the system is reduced. At some Reynolds number then the damp-
ing will be not be sucient to damp the perturbation and it will grow.
It is worth noting here that the Reynolds number of the boundary layer depends on the
distance along the plate (as well as the free-stream velocity). The reason for this is that,
as the boundary layer gets thicker the velocity gradients are reduced - given a prole that
maintains the same shape. As the velocity gradients are reduced the viscous stresses are re-
duced and so is the damping. The result of this is that a boundary layer that starts laminar
will eventually become unstable at some distance down the plate and become turbulent -
this process is known as transition. Figure 16 is a schematic diagram of this process showing
the critical length at which the boundary layer becomes turbulent.
Note also that it is dicult to precisely dene this length since the transition starts with
a very small perturbation (which might come from uctuations in the outer ow, vibra-
tions, surface imperfections) which grows to a point where it breaks down and becomes a
complicated three-dimensional motion. We could then put the transition point where the
4
This is a simplication:viscosity can lead to destabilising inuences but for our purposes this argument
is suciently accurate
44
x
x
trans
Tollmein-Schlichting waves
Turbulence
U
1
U
1
x
trans
/ 5 x 10
5

Figure 16: Transition length for a boundary layer


perturbation rst starts to grow (very dicult to measure), where it reaches a certain am-
plitude or where the ow seems to be fully turbulent (dicult to dene). Analytical studies
can dene the location where the perturbation rst starts to grow, whereas, for the engineer
it is more useful to know where the ow is fully turbulent (or perhaps where the drag has
increased signicantly as it does when going from a laminar to a turbulent boundary layer).
In order to nd the location of the transition then it is necessary to examine the stability of
laminar boundary layers. Fortunately we have some solutions of the boundary layer equa-
tions already such as the Blasius solution and the Falkner-Skan solutions (there are others).
All we then need to do is to add a small perturbation to these solutions and see what hap-
pens. If the perturbation is very small then we can linearise the non-linear boundary layer
equations and nd an equation for the evolution of the perturbation (does it grow or die
away?).
This approach is known as the linearised theory of hydrodynamic stability which assumes
very small (strictly innitessimal) two-dimensional disturbances (Orr-Sommerfeld equation).
Whilst we cannot do the subject justice here the approach is conceptually straightforward.
Take the boundary layer equations for a laminar boundary layer (eg. the Blasius solu-
tion we derived earlier) and add a very, very small disturbance of the form, for example
u

= F(y) e
i((xct))
. Where = 2/l and l is the wavelength of the disturbance. So adding
this disturbance is like adding a wave onto the boundary layer with an amplitude that de-
pends on time. Note that c is in general a complex number: the real part corresponds to the
velocity of the wave and the imaginary part to the growth rate of the disturbance. (Note
that we generally add a u

and v

perturbation by using a perturbation stream-function -


this ensures continuity is satised - this is more complicated and we only want to get the
gist of the idea here )
We add this on to the original solution and substitute it into the boundary layer equations.
Since the perturbation is small it becomes possible to linearise the equations (basically, we
throw out terms that are higher order in our small perturbation) and extract an equation
that describes the growth of the disturbance for any given wavelength and Reynolds number.
If the imaginary part of c (c
i
) found from solving this equation is greater than zero then the
disturbance grows exponentially, if it is less than zero it dies away.
45
Figure 17: Stability diagram for a boundary layer
Figure 17 shows contours of the growth rate of disturbances versus wavenumber ( the in-
verse of the wavelength) and Reynolds number:this is for a zero pressure-gradient boundary
layer. The outermost contour is the neutral stability curve - i.e. the one for which the distur-
bance neither grows nor decays. Any point inside this contour corresponds with an unstable
case - i.e. where the disturbance grows. The leftmost point on this contour corresponds to
the minimum stable Reynolds number, Re
1
= 520 (in this case the Reynolds number on the
diagram is based on
1
=

, the displacement thickness, and the free-stream velocity). The


wavenumber of this point - of about 0.3

corresponds with the most unstable wavelength


and hence the wave that starts the transition. Note that we have assumed that the boundary
layer is perturbed by innitesimal waves of all wavelengths and the ow stability picks out
the rst one with a non-negative growth rate. Also note that the growth rate of this rst
unstable wave is very small (nominally zero) and so this point will be dicult to detect. In
practical situations it is of more interest to know the point at which the ow is rst turbulent
which corresponds with waves with large growth rates and a range of wavelengths - this will
occur at higher Reynolds number. The numbers on the contours (ranging from 0 to 1.96)
correspond to the amplication rates of the disturbances.
This analysis is analogous to the sort of analysis that is done in control theory where a sys-
tem is perturbed with a given frequency disturbance and the response is noted. The classical
example of this problem is the spring mass system with damping that becomes unstable for
zero damping and at the resonant frequency. The boundary layer is a much more compli-
cated system and is non-linear and hence the analysis is more complicated. Nevertheless, in
simplistic terms, the Reynolds number may be considered to be an inverse damping param-
eter since the viscous terms usually tend to damp out perturbations which is why the low
Reynolds number cases are stable.
46
The rst type of disturbance that just starts to grow is a two-dimensional wave on the bound-
ary layer - these waves are known as Tollmein-Schlichting waves. Once they grow suciently
non-linear eects kick-in and the whole ow becomes complicated and nally turbulent.
In reality the disturbances are usually non-linear (i.e. they are not very small) and three-
dimensional and in this case the behaviour is more complicated since the ow bypasses
the usual route to transition. This is referred to as Bypass transition.
It is also of interest to note that laminar pipe ow and channel ow (Poiseulle ow) turn out
to be stable according to linear stability analysis whereas in practice they become turbulent
at high enough Reynolds numbers. The reason that they become turbulent is that they are
unstable to nite disturbances (but stable to innitesimal disturbances). A simple analogy
to this situation is to compare trying to stand a sharpened pencil on its point (unstable to
any disturbance however small) and standing it on its other at end. In the second case it
is stable to small disturbances but unstable to larger ones. This is really what happened
in the IB experiment. The laminar boundary layer in that case is stable for innitesimal
perturbations (otherwise it would become turbulent on its own) but it was unstable to the
larger perturbation given by the trip-wire (the rod at the start).
Question 12.
Figure 17 shows the stability plot for a zero pressure-gradient boundary layer. The Reynolds
number for transition is shown as U
1

/ = 520.
a) Use the results from the Blasius analysis given earlier in the notes to express this as a
Reynolds number based on x (Re
x
= U
1
x/). Note:

/ = 0.86 for the Blasius prole.


b) The usual engineering rule-of-thumb for zero pressure-gradient boundary layer transition
is Re
x
= 5 10
5
. Compare this with the result from (a) and comment on the dierence.
Also convert the engineering value back to a value for

U
1
/ and look it up on the stability
plot.
c) A laminar boundary layer develops in a zero pressure-gradient ow in a wind-tunnel which
has been designed to be extremely low turbulence, acoustic noise and vibration (essentially so
there are no perturbations at all). A perturbation is added to the boundary layer such that
the wavelength of the perturbation is l = 63

. At approximately what Reynolds number


will the boundary layer rst become unstable. What Reynolds number would be a good en-
gineering approximation for when the ow becomes turbulent (consider amplication rates)
Answers: (a) 9.14 10
4
(b) It is larger since it is hard to detect the very start of transition
where the growth rate of the disturbances is close to zero - the engineering value corresponds
to the point where the ow is denitely turbulent (rather than the initiation of the growth)
- the point on the stability curves corresponds to a large growth rate (about 1).(c) About
Re
delta
1
= 3000. To notice the transition then we need a decent amplication rate of say
about 1 which makes the Reynolds number roughly 10,000.
47
10.1.1 Factors aecting transition
The area of transition and of stability theory for boundary layers is a broad eld and is
beyond the scope of this course. We have touched on the ideas here so the basic approach is
understood. What is more important from the point of an engineer is to understand what
factors aect the transition and in what direction. The linear stability analysis provides a
maximum Reynolds number for which the ow can remain laminar: in real ows transition
often occurs well before this since real ows always contain small but nite perturbations.
Due to the limited number of lectures available here we will merely list a few common factors
and their eects.
Pressure-gradient; Adverse pressure-gradients make the b/l less stable and favourable
the opposite
Free-stream turbulence; Leads to early transition (adds nite perturbations)
Surface roughness - leads to early transition - tripping (again adding nite perturba-
tions)
Wall suction; tends to stabilise the boundary layer leading to later transition
10.2 Fully turbulent boundary layers
Turbulent ow is complex unsteady and three-dimensional and hence we must include time-
dependence and three-dimensionality into the equations of motion.
If we dene the total derivative as
D
Dt
=

t
+u

x
+v

y
+w

z
(150)
then the Navier-Stokes equations can be written
Du
Dt
=
1

p
x
+
2
u
Dv
Dt
=
1

p
y
+
2
v
Dw
Dt
=
1

p
z
+
2
w (151)
These non-linear equations are extremely dicult to solve even on a computer. The reason
is that turbulent ows consist of a wide range of dierent sizes and frequencies of motions
(large swirling motions to very small swirling motions). The ratio of the size of the largest
motion to the smallest increases with the Reynolds number. If we solve these on a grid
then it must be ne enough to resolve the smallest motions whilst the domain must be large
enough to resolve the largest motions. Hence the computational cost increases with the
Reynolds number and in fact approximately with the Reynolds number cubed.
In the case of boundary layers the highest Reynolds number that has so far been computed
on a super-computer is about R

= U
1
/ = 1410 (this was a zero pressure-gradient layer
48
t
U(t)
U
u(t)
Figure 18: Denition of U
with simple boundary conditions). For comparison the turbulent boundary layer on an air-
craft wing corresponds to about R

= 50000 - it is nowhere near feasible to computer even


a zero pressure-gradient boundary layer at this Reynolds number using the biggest and best
super-computers available in the world today - it is also unlikely to be feasible in the near
future. In the case of the atmospheric boundary layer the Reynolds number is approximately
R

= 500000.
Faced with this diculty of solving the exact equations (when research on turbulent bound-
ary layers rst started the use of computers to solve the equations was not even plausible,
let alone feasible), researchers turned to statistical methods. The kinetic theory of gases
had provided appropriate tools and had been very successful. It is not possible to solve the
equations for the behaviour of the many billions of molecules in a volume of gas but it is
possible to predict their averaged behaviour in terms of, for example, the mean pressure and
temperature. The behaviour of turbulent ow is at least analogous in the sense that we can
think of it as a huge number small parcels of uid moving around in an apparently random
fashion and we are interested in predicting some mean quantities (eg. the average drag on a
plate caused by a turbulent boundary layer).
So instead of asking what is the instantaneous value of the velocity at some point in the
ow we might instead ask what is the time-average value of the velocity at that point and
what is the standard deviation of the velocity from this average value. To proceed in this
manner we consider splitting the variables into mean and uctuating parts. We can do this
since, in our ows of interest the turbulence is a random, STATIONARY process (its mean
values are independent of time) and hence we can break up the velocity components into a
mean value and a uctuating component i.e.
u = U +u

(t) (152)
where U is the time average of u over a suciently long period so that it converges. If it is
a stationary process then U is independent of time.
U =
1
T
_
t+T
t
udt (153)
49
where T is suciently large (strictly T ). It is a stationary process if we do the same
averaging starting at another t and get the same result.
So if we assume that the turbulence is a stationary random process we can write all the
variables in this form i.e.
u = U +u

(t)
v = V +v

(t)
w = W +w

(t)
p = P +p

(t)
= +

(t)
(154)
where we usually ignore the term and assumed is a constant for incompressible, low
Mach number ow since

M
2
(155)
If we substitute these terms into the Navier-Stokes equations and take time averages of the
equations we can nd the behaviour of the means and uctuating components. We must
observe the following rules:
u
x
=
U
x
_
uds =
_
Uds
u +v = u +v
u = U +u

= U +u

= u +u

(156)
and hence u

= 0.
Also additional rules
u.v = u.v = u.v
uu

= u.u

= 0 (157)
But it is important to note that while u

= 0 and v

= 0 u

is not in general zero (this can


occur if u

and v

are 90
o
out of phase).
50
After we do this decompostion and then average the Navier-Stokes equations we get the
REYNOLDS AVERAGED EQUATIONS
u
j
u
i
x
j
=
1

p
x
i
+

2
u
i
x
j
x
j

i
u

j
x
j
u
i
x
i
= 0
u

i
x
i
= 0 (158)
Note here that a repeated index denotes a summation hence, for example,
u

i
x
i
=
u

x
+
v

y
+
w

z
(159)
where the three components of the velocity vector are u
1
, u
2
, u
3
or u, v, w. Now for LAMINAR
ow we have
u
j
u
i
x
j
=
1

p
x
i
+

2
u
i
x
j
x
j
u
i
x
i
= 0 (160)
and these equations are the same provided that the turbulence is stationary and the laminar
ow is steady and we put
u
i
= u
i
(161)
except for the extra term

i
u

j
x
j
(162)
which represents the gradients of stresses due to the random turbulent motion. This term
is a nine component tensor which is symmetric.
Question 13.
Using the continuity equation and splitting the velocity components into mean and uc-
tuation parts show that for a turbulent ow both the mean velocity components and the
uctuating components satisfy the continuity equation separately as shown in (158) above.
Hinze and Schlichting have applied this to jets and boundary layers and have used an order
of magnitude analysis to nd for boundary layers
u
u
x
+v
u
y
=
1

dp
dx

u

y
+

2
u
y
2
u
x
+
v
y
= 0 (163)
whereas for laminar ow
u
u
x
+v
u
y
=
dp
1
dx
+

2
u
y
2
u
x
+
v
y
= 0 (164)
51
Which is the same if u u and v v except for the extra turbulent stress term. If we dene
a new stress by

= u

+
u
y
(165)
then the equations become
u
u
x
+v
u
y
=
1

dp
dx
+

y
_

_
u
x
+
v
y
= 0

= u

+
u
y
(166)
Unfortunately now we have 3 equations and 4 unknowns u, v, / and u

, since we have
averaged the equations and hence lost some information. In order to solve the equations we
must CLOSE them by nding an extra CLOSURE EQUATION. We will come back to this
in a later section.
Before going into this consider a physical explanation of this mysterious extra stress. Con-
sider what u

means. If this is negative then we have two cases; u

is positive when v

is
negative on average or u

is negative when v

is positive on average. Now if u

is positive then
a parcel of uid is a bit faster than the mean and hence it has higher momentum. So when
u

is negative this means that, on average higher momentum uid moves down towards the
wall and lower momentum uid moves up away from the wall (it is best to think about this
for a while and try waving your hands about to see what is going on).
Now why a stress? Consider for simplicity a control volume with the wall as its bottom
boundary. The eect of this negative u

is to increase the momentum in this control vol-


ume by adding high momentum uid and throwing out (the top) low momentum uid. This
increase is eectively the same as would happen if we applied a net force to the control
volume in the streamwise direction.
This is a simple (but correct) explanation. It also explains why the proles for laminar and
turbulent boundary layers look as they do. Consider the proles from the IB experiment.
In the turbulent case there is more momentum near the wall due to the redistribution of
momentum by this Reynolds shear stress term. This also explains why turbulent boundary
layers are less likely to separate than laminar ones - the Reynolds shear stress redistributes
the momentum such that it is increased near the wall which compensates to some extent for
the drop in streamwise momentum near the wall due to the action of an adverse pressure-
gradient. Put simply the pressure-gradient applies a backward force which reduces the
momentum near the wall and the Reynolds stress constantly brings in new (positive) mo-
mentum from the outer part of the ow.
52
Question 14.
Integrate the boundary layer equation for the turbulent boundary layer (as shown in (163))
from the wall to outside the layer and show that the momentum integral equation is the
same for a turbulent boundary layer as it is for a laminar boundary layer. The method is
the same as given earlier in the notes. If you are smart you might note that the equation
is the same as for the laminar case except for the one extra term added on the right-hand
side that needs to be integrated up - so you might save some time and eort. Note that the
Reynolds shear stress goes to zero in the outer ow - outside the boundary layer there is no
turbulence (also note that it must go to zero at the wall since the no-slip and no-penetration
conditions mean the uctuations must go to zero - and the mean velocities too, of course!).
10.3 The mean velocity prole
The variation of the mean velocity is very important since once we have it we can work out
the momentum thickness, the shape factor and using the momentum-integral equation we
can then work out the wall shear stress and hence the skin-friction. This is a very important
quantity for an engineer - perhaps the most important.
The earliest researchers in the eld simply used empirical curve-ts to experimental data as
a way of solving the problem. The most famous is the one-seventh power-law prole where
the mean velocity was assumed to vary as y
1/7
. This is a reasonable t over most of the
prole but has the wrong behaviour very near the wall and at large distances from the wall
(where it continues to increase without bound). The advantage of such a curve-t is that
it is easy to deal with analytically and since it is integrated when nding the skin-friction
small errors dont make a lot of dierence to the answer. It is possible to derive the variation
of the skin-friction with distance along the plate using this assumption - the full derivation
is given in an Appendix to the notes.
One problem with the power-law is that there is no physics in its derivation and this be-
comes obvious when we note that the best power law t depends on the Reynolds number
(sometimes 1/8 is better, sometimes 1/6 and so on). That indicates that there is something
wrong somewhere.
In order to get a better understanding of the true shape we can apply some simple physical
arguments which we will then use for dimensional analysis so as to get some quantitative
results.
Consider what happens very close to the wall. At the wall the velocity parallel to the wall
must go to zero due to the no-slip condition and the component normal to the wall must also
go to zero due to the no-penetration condition. This is true for both laminar and turbulent
boundary layers. This means that the uctuating velocities must also go to zero at the wall
and hence very close to the wall the Reynolds shear stress must go to zero (in fact it is not
dicult to show that it must vary as y
3
just by considering the Taylor series expansion of the
uctuations). The viscous stress, on the other hand depends only on the velocity gradient
which does not go to zero at the wall but instead increases up to some nite value. Hence
there will be a region very close to the wall where the viscosity dominates and the equation
53
is the same as for a laminar boundary layer. This region is known as the laminar sublayer
(or sometimes as the viscous sublayer). Its thickness depends on the Reynolds number and
it is very thin at high Reynolds number.
As we move out from the wall then we would expect the Reynolds shear-stress (the turbu-
lence stress) to increase while the viscous stress drops (the mean velocity gradient drops)
and hence there will be some region where they are both important - this region is called
the buer region.
Further out as the viscous stress drops even further and the turbulence uctuations increase
then we might expect a region where the viscous stress becomes negligible and hence the
mean ow might be independent of viscosity.
As we move still further out the Reynolds shear stress drops and we might expect the outer
variables such as the layer thickness and pressure-gradients to become important.
10.4 Some dimensional analysis and hand-waving
Using the physical arguments we have just discussed and some others we can do some di-
mensional analysis in order to work out something about the shape of the mean velocity
prole.
Before making any assumptions we might look at the things on which the velocity prole
might depend. We will consider zero pressure-gradient ow in order to make the analysis
simpler. So just consider all the relevant independent (or imposed) variables that aect the
ow
u = f
1
(y,
o
, , , , U
1
), (167)
where f
1
is some unknown function. Dimensional analysis would then lead us to four non-
dimensional groups - which is not that helpful for our purposes. Now in order to simplify it
we look at the dierent regions discussed above and see what variables we can throw away.
Before starting, to simplify things note that the quantity
_

o
/ has the dimensions of a
velocity and since we are interested in velocities we can use this in place of
o
. The ve-
locity produced is known as the wall-shear velocity and here we will give it the symbol U

hence U

=
_

o
/ (in some literature the symbol used is u

). In this way we also remove


direct eects of the density since it only enters in modifying the kinematic viscosity and the
wall-shear velocity.
Now consider a region quite close to the wall. If we are far enough away from the outer
edge then we might guess that the local ow does not know anything directly about the
free-stream velocity or the boundary layer thickness (except in so far as they aect the shear
stresses etc.). In this case we can write
u = f
2
(y, U

, ), (168)
and we nd only two non-dimensional quantities so this can be re-written as
u
U

= f
_
yU

_
, (169)
54
10 100 1000 10000
z
+
5
10
15
20
25
30
35
U
+
10m/s (R

=2,583 &

=1,047)
10m/s (R

=9,256 &

=3,436)
10m/s (R

=20,013 &

=7,611)
20m/s (R

=37,542 &

=14,379)
30m/s (R

=52,103 &

=19,956)
40m/s (R

=68,151 &

=26,376)
Log-law (=0.41 & =5.0)
Figure 19: Law-of-the-wall scaling: Boundary layers scaled on wall-variables showing collapse
in the inner region (note here z
+
= y
+
) for a range of Reynolds numbers: zero pressure-
gradient ow
where all the f functions are unknown and I am just using subscripts (f
1
, f
2
etc.) to point out
that they are dierent. This famous expression was rst derived by Prandtl and is known as
the Law-of-the-wall. This suggests that, close enough to the wall, all turbulent boundary
layer velocity proles will collapse when scaled like this. It is an amazing result that is very
well supported by experimental data. Figure 191 shows proles from a zero pressure-gradient
boundary layer over a range of Reynolds numbers plotted in law-of-the-wall co-ordinates -
the collapse is very good in the inner part but they start to diverge towards the outer edge
due to the eects of the outer ow variables (in particular ).
The region where it applies covers the laminar sublayer, the buer region and a signicant
part of the ow further out. The region of applicability varies for dierent ows but it covers
at least about the lower 10% of the thickness of the layer where all the signicant action
is. As an aside, note that in the literature when something is non-dimensionalised using
wall-variables(eg. U

, ) then the quantity is superscripted with a plus. Hence the law of


the wall might be written as u
+
= f(y
+
).
Next lets go very close to the wall into the region of the viscous sublayer. To simplify this
further return to the boundary layer equation.
u
u
x
+v
u
y
=

y
_

_
(170)
we have dropped the pressure-gradient term and based on our earlier argument the shear
stress is just the viscous part - the Reynolds shear stress is negligible. To proceed further
55
note that as we go towards the wall the mean velocity components also go to zero and so
the left-hand side of the equation becomes negligible. We are then just left with

2
u
y
2
= 0. (171)
Integrating once we nd

u
y
= constant, (172)
and at y = 0 this constant is just the wall-shear stress by denition. Integrating again we
nd the prole is linear and may be written
u
U

=
yU

(173)
which is, of course, consistent with the law-of-the-wall as given in (191) but we have now
dened the form of the function in some region very close to the wall. Empirically this
applies well for a region of approximately y
+
< 5. We sometimes say for y is less than ve
wall-units. The quantity /U

has dimensions of a length and is often called a wall unit.


The physical size of this quantity depends on the Reynolds number but just to give an idea,
in a windtunnel with a ow of 30 m/s (not that fast) and say ow developing over a metre
then one wall-unit 15m.
We will skip over the buer region since in that region all the terms are important (the mean
momentum, the viscous stress and the Reynolds shear stress) and so it is hard to simplify
things further. Beyond this, however, but still within the region where the law-of-the-wall
applies, we expect that the viscous stress term becomes negligible in comparison with the
Reynolds stress (and the mean momentum). In this part then we would expect the ow
to be independent of viscosity. Note that we need to be a bit careful. The velocity in this
region cannot be independent of viscosity since the velocity at a given y relative to the wall
depends on the change in velocity that occurs across the viscous sublayer and buer region
plus the changes in velocity that occur locally which may be independent of viscosity. Since
the change in velocity across the viscous sublayer and buer layer depend on viscosity then
the actual value of velocity further out must depend on viscosity.
We might seem to be stuck but re-reading the previous paragraph suggests a way forward.
Whilst the actual velocity in this region depends on the viscosity (due to the increase very
near the wall) we might suggest that the changes that occur in the velocity within this region
are independent of the viscosity (since these are determined by local stresses). So then
u
y
= f
3
(y, U

) (174)
this is consistent with the law-of-the-wall still but we have eliminated viscosity in the deriva-
tive. We can proceed from here in two equivalent ways. The rst and simplest is to note that
we now have three variables and two fundamental quantities and so only one non-dimensional
number (Pi-group) which must therefore be constant
y
U

u
y
= A, (175)
56
where A is just some constant. Integrating gives
u
U

= Aln y +B (176)
where the constant B will depend on changes that occur in the viscous sublayer and buer
layer. In order to x the constant pick any point within the region with some xed value
of yU

/ and therefore xed u/U

and we nd B = a
1
Aln(/U

) where again a
1
is just
some constant and so this can be written as
u
U

= Aln
_
yU

_
+C (177)
The other, perhaps neater but equivalent way is to note that, from the law-of-the-wall
u
y
=
U
2

_
yU

_
, (178)
where the dash denotes dierentiation with respect to the argument. This is then substituted
into 175 and integrated to nd the function f in the law-of-the-wall:the same result follows.
Note that, for historical reasons the constant A is usually replaced by 1/ where the constant
is referred to as the von-Karman constant. von Karman derived the log law in a dierent
manner which resulted in the constant coming out in this manner. So we would usually see
the log-law written as
u
U

=
1

ln
_
yU

_
+C (179)
where is a universal constant with a value of about 0.4.
This expression is known as the logarithmic law for the mean velocity or, more aectionately
as the log-law. Whilst there have been a variety of challenges to this in the past it is strongly
supported by data (see for example gure 191) and is at the very least an excellent t to
data over a wide range of Reynolds numbers (which has a sensible physical basis, as opposed
to, say, the 1/7 power law). It applies in some part of the ow not too close to the wall, as
we have discussed, so we are beyond the viscous sublayer and buer region - opinions dier
as to the number but a universal value of around y
+
= yU

/ > 50 is a reasonable estimate.


It also only applies when we are far enough from the outer edge of the layer and so for some
universal value of y/ which is usually taken as approximately y/ < 0.1. The reason it
is hard to precisely pin these numbers down is that we are considering a region where the
viscous stresses are small enough to neglect and where the outer inuences are also small
enough. These eects are always present and never exactly zero. As a nal note since we
now have an estimate for where the log-law starts (y
+
= 50) and where it ends (y/ = 0.1)
then we can work out the length of the log-law. The ratio of the outer limit to the inner
is y
outer
/y
inner
= 0.1/(50/U

) = 0.002 U

/. So it depends on the Reynolds number


and gets longer for higher Reynolds number. Figure 191 shows this trend. On the gure
K

= U

/.
Now we turn to the outer part of the ow near the edge of the boundary layer. We postulate
that the ow out here does not know about the existence of the wall directly and what is
important out here is the dierence of the velocity in the layer from the free-stream velocity
57
(U
1
u(y)). We assume that the wall shear stress is still important since it determines the
stresses further out from the wall (we would expect that if we doubled the wall shear-stress
then the stresses further out would also increase - it is only in this indirect sense that the
ow out here feels the eect of the wall). This is an important assumption - the outer
ow only feels the eect of the wall through the transmitted shear stress (we will return
to this later). Now we also not that out here the boundary layer thickness must enter
as an important variable since the velocity will depend on how far we are from the edge.
Dimensional analysis then gives
U
1
u(y) = f
4
(y, , U

, U
1
) (180)
which leads to
U
1
u(y)
U

= f
4
(y/, U
1
/U

) (181)
but it is usual to drop the dependence on U
1
/U

since this is weak in zero pressure-gradient


ow (this quantity varies only slowly with Reynolds number, or equivalently x)
U
1
u(y)
U

= f
4
(y/) (182)
This expression is usually called the velocity defect law and is the outer ows equivalent of
the law-of-the-wall. It is interesting that we can also derive the log-law by working from this
expression and moving in. Recall to derive it before we started at the wall and moved out
until we could neglect viscosity. In the outer part viscosity is already negligible (think about
the velocity gradient) but the boundary layer thickness is important. We might imagine that
as we move toward the wall and away from the edge then, far away from the edge (but still
where the viscous stress is neglible) then the boundary layer thickness would not be relevant
any more - at least in terms of the changes in velocity so as before we could write
y
U

u
y
=
1

, (183)
and from (180)
u
y
=
U

4
(y/) (184)
then substituting and integrating to nd f
4
(y/) gives
U
1
u(y)
U

=
1

ln (y/) +D (185)
where D is some constant of integration. So we nd we have another log law!! Also some-
where in the middle not too close to the wall and also far from the edge. You may have
guessed that it is the same log-law just expressed in dierent variables. To show this take
(179) and let y = - at this point then u = U
1
so you have an expression for U
1
/U

. Subtract
(179) from this and you have an expression for the defect. Figure 20 shows the collapse of
proles with velocity defect scaling. They collapse best in the outer region but then diverge
in the inner region as viscous eects become important.
Question 15.
Show that the log-law for the outer ow is consistent with the log-law for the inner ow as
discussed in the text above.
58
10
-3
10
-2
10
-1
10
0

0
5
10
15
20
(
U
1
-
U
)
/
U

10m/s (R

=2,583 &

=1,047)
10m/s (R

=9,256 &

=3,436)
10m/s (R

=20,013 &

=7,611)
20m/s (R

=37,542 &

=14,379))
30m/s (R

=52,103 &

=19,956)
40m/s (R

=68,151 &

=26,376)
Log_Law (=0.41 & =2.309)
Figure 20: Defect-law scaling: Boundary layers showing collapse in the outer region (note
here = y/) for a range of Reynolds numbers: zero pressure-gradient ow
Finally we consider how to deal with the outer ow in combination with the law-of-the-
wall. We have shown that the law-of-the-wall applies from the wall up to some value of
y/ 0.1 where the eect of the outer ow and, in particular becomes important. It
becomes logarithmic far enough from the wall but below this value. Beyond this point the
wall-scaling no longer applies and the mean velocity deviates from the inner log-law solution.
Coles(1956) suggested a simple way to account for this is to assume that this deviation is a
function of y/ and hence that the whole prole could be written as
u(y)
U

= f
_
yU

_
+
_
y

_
. (186)
He called this deviation () the wake function since its shape is similar to that of the wake
of a body. The dierent regions of the boundary layer are shown in gure 21. There is a
lot of information on the gure but dont be deterred. Figure 191 shows velocity proles for
zero pressure-gradient boundary layers across a range of Reynolds numbers.
Note that from this expression we can say something about the variation of the all-important
skin-friction coecient. It is not dicult to show that C

f
/2 = (U

/U
1
)
2
. Consider (189)
evaluated at the edge of the layer
U
1
U

= f
_
U

_
+(1). (187)
This can be rewritten as

_
2
C

f
= f
_
_
U
1

f
2
_
_
+(1). (188)
59
1 10 100 1000 10000
yU

/
0
10
20
30
U/U

Boundary layer velocity profile


U/U

=yU

/
Logarithmic law
Wake
laminar sublayer
logarithmic region
Buffer region
y/ = 0.1
10% of boundary layer thickness
Figure 21: Dierent regions of a turbulent boundary layer
which in the logarithmic part of the ow is

_
2
C

f
=
1

ln
_
_
U
1

f
2
_
_
+C +(1). (189)
which is unfortunately and implicit expression for C

f
and is in terms of the local Reynolds
number. It is possible to relate to the streamwise distance x through the momentum
equation and after some simplication this leads to

_
1
C

f
= a
1
+a
2
ln(R
x
C

f
) (190)
where R
x
= U
1
x/ and the a
1
and a
2
are just constants determined from experiments.
This expression is sometimes called the Karman-Schoenherr law for skin-friction. The full
derivation is given in an appendix to these notes. It is not too dicult but takes a number
of steps. A comparison of this prediction from the above analysis with a large variety of
data is shown in gure 22.
10.4.1 The eect of pressure-gradient
Since many ows of interest have an applied pressure-gradient then we need to consider how
things change with pressure-gradient. In this case there is an extra term in the boundary
layer equation. We will consider only moderate pressure-gradients which might be typical of,
say, aerofoils and other common ows. We will also consider only moderately high Reynolds
numbers.
60
Figure 22: Recent data showing C

f
versus R

In the inner part of the ow the pressure-gradient introduces and extra parameter which is
the non-dimensional pressure-gradient (non-dimensionalised on wall-variables) hence
u
U

= f
_
yU

,

U
3

dp
dx
_
. (191)
It can be shown, however that at reasonable Reynolds number that this extra term becomes
very small due the being divided by U
2

and so for moderate to high Reynolds numbers and


moderate pressure-gradients this term is usually neglected. Hence the same law-of-the-wall
applies and all of the analysis above is still valid (we still get a log-law of the standard form
and the inner region is not signicantly modied). It is small relative to the other terms in
the equation (in particular the stresses).
In the outer region, however, this is not so. The viscous stress is negligible in the outer
region and the Reynolds shear stress is falling towards zero and so the pressure-gradient can
be important. Hence the defect law must be modied and we nd
U
1
u(y)
U

= f
4
_
y

,

U
2

dp
dx
_
. (192)
Note that if we keep the non-dimensional pressure-gradient parameter constant then the
defect proles will still collapse as for zero pressure-gradient layers but the shape will be
dierent. In the general case, however, the proles still collapse in the inner region (using
law-of-the-wall scaling) but not in defect form.
Using the same logic as before Coles(1954) suggested then that this could be accounted for
by allowing the wake function to vary with pressure-gradient. For simplicity (and based on
data) he suggested that the functional form would be the same but the strength of the wake
part (relative to the inner part) would change depending on the pressure-gradient this is
61
1 10 100 1000 10000 1e+05
yU

/
0
10
20
30
40
50
U/U

Zero pressure-gradient
Adverse pressure-gradient
Favourable Pressure-gradient
Figure 23: The eect of pressure-gradients on the boundary layer prole
usually written as
u(y)
U

= f
_
yU

_
+
(x)

W
_
y

_
, (193)
where W(y/) is a universal function and is known as Coles wake factor which varies
with pressure-gradient (and distance along the plate if pressure-gradient is varying) - the
fact that it is divided by is only a mathematical convenience for some other analysis.
Coles found that the function W(y/) = 1 cos(y/) although this is only a curve-t and
obviously only works up to the edge of the layer (y/ = 1) and not beyond. The eect of
pressure-gradients on the proles are shown in gure 23 where it can be seen that the scaling
still applies in the inner region but the outer wake region is strongly aected. The value
of increases for adverse pressure-gradients and decreases for favourable ones. In strongly
favourable pressure-gradients there is virtually no wake at all and the log law applies almost
to the edge of the layer.
Note that the prole shape now depends on an extra parameter and so it is not possible
in general to nd the skin-friction variation with x unless we know the function (x) - this
can only come from measurements. In general we would know the pressure-gradient but
if we could relate to the pressure-gradient parameter then we could solve the problem.
There are various attempts to do this but all contain a large amount of empiricism and
work only in certain cases. In reality depends on a number of factors including the
history of the boundary (what pressure-gradients it has undergone upstream of the location
of interest). Hence methods of evaluating the skin-friction in this manner are always just
good approximations for a particular situation.
62
Question 16.
(a) A boundary layer has a prole given by
U
1
u(y)
U

= f() (194)
where = y/. (a) Show that

/ = C
1
U

/U
1
, where

is the displacement thickness and


C
1
=
_

0
f()d (195)
(b) Show that / == C
1
U

/U
1
C
2
(U

/U
1
)
2
where is the momentum thickness. Note that
the denitions of momentum and displacement thickness are exactly the same for turbulent
boundary layers as they are for laminar ones.
C
2
=
_

0
(f())
2
d (196)
(c) nd an expression for the shape factor, H in terms of C
1
, C
2
, U

/U
1
.
(d) Consider the law-of-the wall expression evaluated at y = and show that this implies that
the shape factor depends on Reynolds number (you need not derive the exact expression).
10.4.2 The eect of wall roughness
In many, if not most, engineering applications the surface over which the boundary layer
develops is not perfectly smooth and may be very rough indeed. Now all real surfaces have
some roughness if you look close enough so the rst question to consider is when the wall
may be considered smooth. It turns out that if the height of the protrusions on the surface
is well within the laminar sublayer then the uid ows smoothly over them and there is no
signicant eect on the overall ow. It is found that this occurs when the height of the
roughness is nominally less than about 5 wall units. If we denote the roughness height by
h then when h
+
< 5. Note that the actual acceptable height will depend on the thickness
of the laminar sublayer which varies with Reynolds number so a ow that appears smooth
to low Reynolds number boundary layer may well appear rough to one at higher Reynolds
number.
Once the roughness is larger than the laminar sublayer then we would expect it to introduce
perturbations into the near wall ow and modify the region very near the wall signicantly.
This messes up our scalings very near the wall but we have noted already that the ow
further out does not know what is going on at the wall except in-so-far as it detects a change
in the stresses (which scale with the wall-shear stress). This prompted Townsend(1956) to
propose his wall-similarity hypotheses. Consider the defect scaling. In this there are no
extra parameters introduced by what is going on at the wall and hence the defect law will
be unchanged - the actual wall shear-stress increases but since this is accounted for in the
scaling then the defect law still works perfectly well. Experiments conrm that this is the
case as long as the roughness elements are not so big as to extend well into the ow.
What about the near wall scaling? Our law-of-the-wall. The height, h, of the elements
63
imposes a new length scale and so we might write
U
U

= f
_
yU

,
hU

_
(197)
We can also follow the same arguments as before to derive a log-law assuming that the height
of the roughness elements becomes unimportant far enough away from the wall (in terms of
the velocity gradient), as is the viscosity. This has the form
U
U

=
1

ln
_
y
h
_
+C(h
+
) +(y/) (198)
where the is the same as before but the C will depend on the ow near the wall hence
C is a function of hU

/ = h
+
since the wall shear stress will depend on the drag of the
elements which is a combination of viscous drag and form drag. The wake function should
be unaected since the outer ow is unaware of the details of the near wall ow.
If the roughness elements become larger then the roughness elements become like small blu
bodies. Just like the disk in the 3A1a experiment then their drag becomes independent of
the Reynolds number and so even near the wall the viscous stresses become unimportant
(relative to the form drag on the elements and the Reynolds shear stress). Hence our wall-
unit length-scale /U

is no longer important. In this case then the law-of-the-wall becomes


U
U

= f
_
y
h
_
(199)
and the log-law has the same form but in this case we would expect the constant, C, to be
just a universal constant (since there is no other non-dimensional parameter on which it can
depend). In this case the wall-shear stress just depends on the form drag of the elements.
We can again use a simple argument to look at the variation of the skin-friction coecient.
Again consider the log-law at the outer edge of the ow and we nd
U
1
U

=
_
2/C

f
=
1

ln
_

h
_
+C
_
hU
1

f
2
_
+(1). (200)
Consider a xed height of roughness and increase the Reynolds number by increasing the
speed of the ow. The thickness of the boundary layer is only a weak function of Reynolds
number (it gets thinner as Re increases but we will neglect that to rst order). As the
Reynolds number increases the sublayer gets thinner and thinner and so the roughness el-
ements protrude further into the ow - h
+
increases. Beyond a certain point, however, C
ceases to depend on h
+
and hence with /h constant the local skin-friction coecient be-
comes (roughly) constant. This is the situation that applies if we measure the drag of a plate
of xed length L and vary the Reynolds number - if we use a number of dierent roughness
heights then we can plot the drag of the plate as shown in gure 24.
Now consider increasing the Reynolds number by keeping the velocity constant but moving
along the plate. In this case the boundary layer gets thicker and so the relative roughness
gets smaller as we move along the plate so the local skin friction coecient drops for a xed
height of roughness.
64
C
D
U
1
L/
Large roughness
Decreasing roughness
Turbulent smooth curve
Laminar
Figure 24: Drag coecient for a at plate of length L (at zero incidence) with dierent levels
of roughness
10.5 Turbulent pipe ow
Since pipes are used quite frequently in engineering then turbulent pipe ow is of particular
interest. Flow in a pipe develops from the start of the pipe where a boundary layer forms
on the wall and grows with distance along the pipe, this might be initially turbulent or
laminar depending on the speed of the ow. After some distance the boundary layer grows
until it reaches the centre and cannot grow anymore and so cannot develop with streamwise
distance. At some distance after this point the ow settles down to an internal equilibrium
and it becomes independent of streamwise distance - the velocity prole (and the proles
of the other quantities eg. Reynolds stresses). This is known as fully-developed ow and,
typically occurs after about 100 diameters from the start. This is the ow we will consider
here.
All of our previous arguments and dimensional analysis for the turbulent boundary layer
still apply in this case since none specically referred to the geometry of the ow. The only
dierence is that in the pipe the thickness of the layer is xed and equal to the radius of
the pipe, for which we use the symbol a. Also the outer ow velocity U
1
is now the velocity
at the centre of the pipe. Hence, as before for boundary layers,
U
U

= f
_
yU

_
(201)
near the wall and
U
1
U
U

=
_
y
a
_
(202)
in the outer part - near the centre-line. And we can use the same arguments again to derive
the logarithmic law both in law-of-the-wall co-ordinates and in outer co-ordinates. These
have exactly the same form as before if we substitute a for and take U
1
as the centre-line
velocity. Note that in this ow there is still a wake component but it must be independent
of x since the mean velocity prole is independent of x. The wake component is, however,
quite small since there is a favourable pressure-gradient acting in a pipe due to the friction
loss. To see this more clearly consider a control volume across the pipe. The momentum in
must equal the momentum out since the velocity prole does not change. Hence the sum
65
of the external forces is zero and so the force due dierence in pressure across the control
volume is balance by the force exerted by the wall shear-stress i.e. for a control volume of
length dx

o
2adx =
dP
dx
dxa
2
(203)
or, using the same denition of U

as before
a
2U
2

dP
dx
= constant. (204)
We should then not be surprised that the wake doesnt change since the non-dimensional
pressure-gradient parameter based on the outer ow is a (negative) constant. Note also
that since U

does not vary with x


5
(and neither do a or ) then this implies a constant
pressure-gradient - in other words the pressure varies linearly along the pipe.
11 The friction factor in a pipe
We can now determine the friction factor in a turbulent pipe ow. In this derivation the
pipe ow is assumed to be fully-developed which is to say that the wall boundary layers
have had enough development length to grow across the pipe and hence the velocity prole
is self-similar (usually about 60 to 100 diameters). In a pipe we nd that if the velocity
proles are similar the problem reduces to a balance between the skin-friction forces and the
pressure-gradient forces. Let us consider a cylindrical slug of uid in the pipe.
There is no inertia then

o
2L = (p
1
p
2
)a
2
(205)
hence

o
=
(p
1
p
2
)
L
_
a
2
_
(206)
It is usual in pipe ow to use the bulk velocity, U
b
which can be obtained using a stop-watch
and a bucket. It is the average velocity over the pipe area i.e.
U
b
=
1
A
_
A
udA (207)
and from the denition of friction factor ,f,
(p
1
p
2
)
L
d = f
1
2
U
2
b
(208)
This leads to
f
8
=
U
2

U
2
b
(209)
for a pipe. We can compare this to a boundary layer where
C

f
2
=
U
2

U
2
1
(210)
5
The wall shear stress depends on the gradient of the velocity at the wall which depends on the velocity
prole. In fully developed ow this does not change
66
and combining the two suggests that for a pipe ow
f
4
=

o
1
2
U
2
b
(211)
Now we also know that
U
1
U
U

=
_
y
a
_
(212)
and this leads to
1
A
_
A
U
1
U

dA
1
A
_
A
U
U

dA =
1
A
_
A

_
y
a
_
dA (213)
U
b
U

U
1
U

=
1
A
_
A

_
y
a
_
dA (214)
Now we can integrate across the pipe using annular strips. If we do this then
a
2
=
_
a
0
2(y a)d(y a) (215)
so
U
b
U

U
1
U

=
1
a
2
_
a
0
2a(1
y
a
)(
y
a
)d(a(1
y
a
) (216)
where (y/a) is given by
(
y
a
) =
1

ln(
y
a
) W(1) +W(
y
a
) (217)
and hence
U
b
U

U
1
U

= 2
_
1
0
(1
y
a
)(
y
a
)d(
y
a
) = universal constant (218)
since the terms in the integral are all functions of y/a and the integral is denite. If W(1)
and W(y/a) are considered to be small then it can be shown that
U
b
U

U
1
U

=
3
2
(219)
HINT: to do this integration you must know lim
x0
xln(x) which may be found from
lhospitals rule.
Now we want to nd the friction factor so we must consider the region nearer the wall. We
know that U = U
1
at y = a and
U
1
U

=
1

ln
_
aU

_
+A+W(1) (220)
and substituting this back we nd
U
b
U

ln
_
aU

_
AW(1) =
3
2
(221)
(retaining the H(1) term here). We also have a realtionship for the friction factor which is
U
b
U

8
f
(222)
67
1000 10000 1e+05 1e+06 1e+07 1e+08
Re
0.01
0.02
0.03
0.04
0.05
f
Oregon Data - low temperature helium
Princeton Data - very high pressure pipe
Curvefit using theory
Figure 25: Friction factor relation from the recent paper of Smits et al.
hence

8
f
=
1

ln
_
U
b
d

.
U

U
b
_
+A+W(1)
3
2

1

ln(2) (223)
where d is the pipe diameter and U
b
d/ is the Reynolds number. Now substituting again for
U
b
/U

in terms of the friction factor we get

8
f
=
1

ln
_
Re.

8
f
_
+A+W(1)
3
2

1

ln(2) (224)
which may be rewritten as

8
f
=
1

ln
_
Re.

8
f
_
+D (225)
where D is a universal constant. Schlichting gives this formula as
1

f
= 2.0 log
10
(Re
_
f) 0.8 (226)
which is called Prandtls smooth tube formula. Recent research using the combined re-
sources of a 25 tonne pressurised (200 atmospheres) super-pipe at Princeton with a 30g
liquid helium rig in Oregon have produced better estimates for the constants which gives
an excellent t to the data over nearly four orders of magnitude in Reynolds number (from
approximately Re=2000 to Re=20,000,000). They nd
1

f
= 1.93 log
10
(Re
_
f) 0.537 (227)
see gure 25
The constants used generally are
= 0.41
A = 5.1. (228)
68
but note that the new constants given above suggest a value of = 0.422 for the super-pipe
data. Note that this is still an active area of research. There is still some question as to
the correct values of the constants and whether they have the same asymptotic values in
dierent ows (eg. pipes versus boundary layers). From a general engineering perspective
in most cases the errors incurred due to errors in are much smaller than the other errors
in design and manufacture (eg. pipe roughness, losses in imperfect joins etc.).
11.1 Rough pipe ow
Since most pipes in general use have some roughness then it is useful to consider the behaviour
of rough pipes. Again the same arguments used for rough boundary layer ow and the same
scalings apply when we replace by a and take U
1
as the centreline velocity. It is also
possible to calculate the skin friction for the rough pipe. Here we will consider just the case
for large roughness since it is easier. Using the arguments as before that when the elements
extend well beyond the sublayer their drag becomes independent of Reynolds number we
nd the log law
u
U

=
1

ln(y/h) +C +W(y/) (229)


At the centreline this becomes
U
1
U

=
1

ln(a/h) +C +W(1) (230)


where the term on the left is directly related to the inverse of the friction factor. Since,
for a given pipe, everything on the right of the equation is constant then the friction-factor
f becomes independent of Reynolds number. We already know how it behaves for smooth
pipes from the above and in between the two we have some transition region between that
behaviour and the constant value here. We can also note that increasing the relative rough-
ness (h/a) reduces the right hand side and hence increases the friction factor (which is not
too surprising). This information is summarised on a diagram known as the Moody diagram
(gure 26) which shows the friction factor of a pipe versus Reynolds number for a range
of dierent relative roughness values. Note also that the Reynolds number at which the
friction factor becomes constant varies with relative roughness. This is because the impor-
tant parameter determining when we can neglect the viscous drag depends on the ratio of
the roughness height to the sublayer thickness. At higher Reynolds number the sublayer is
thinner and hence
12 Closure Hypotheses
So, as noted, we nd ourselves short of one equation. The problem is to relate our new
Reynolds stress term somehow to the mean velocities - if we can do that then we can nd
solutions. This is the Closure Problem much discussed in turbulence circles - in particular
among people trying to compute turbulent ows in practical situations. There are many
dierent closure hypotheses which have been used to solve the REYNOLDS AVERAGED
EQUATIONS.
The simplest is the Boussinesq hypothesis. By analogy with LAMINAR ow Boussinesq
suggested the following relation.
69
2 4 6 8 2 4 6 8 2 4 6 8 2 4 6 8 2 4 6 8 10
3
10
4
10
5
10
6
10
7 8
10
0.01
0.012
0.100
0.090
0.080
0.070
0.060
0.050
0.040
0.035
0.030
0.025
0.018
0.016
0.014
0.020
g
u
d
L
h
f
2
2

d u
= Re
pipe friction chart
applicable to circular pipes running full
Glasgow College of Nautical Studies Faculty of Engineering
GCNS runs distance learning courses for the
Engineering Council Graduate Diploma.
website: www.glasgow-nautical.ac.uk
e-mail: engineering@glasgow-nautical.ac.uk
Figure 26: Moody diagram

=
u
y
(231)
where
T
is the turbulent shear stress given by
T
= u

. The constant is called the


EDDY VISCOSITY. In free shear ows such as jets and wakes is assumed to be a constant
just as is a constant - this gives results that are roughly right. In reality however is a
property of the ow and not of the uid and so its value can vary. In particular in the case of
boundary layers it is found that must vary with the distance from the wall to give results
consistent with experiments.
The log-law has been derived by Squire(1947) using an eddy viscosity assumption.

= U
2

= u

=
U
y
(232)
Now the law of the wall suggests that is only a function of wall variables close to the wall
and furthermore beyond the viscous sub-layer then we would expect the viscous terms to be
negliglible and hence should not be important, hence
= f(y, U

) (233)
Now dimensional analysis tells us that there is only one PI product and so

yU

= constant = (234)
and substituing back into the equation for shear stress we nd
U
2

= yU

U
y
(235)
70
and then we have an expression for U/y which can be integrated and gives the same result
as before ( a log-law). In the outer part of the ow then we expect that the eddy viscosity
should depend on rather than y and hence

= constant = B (236)
and thus the eddy viscosity is constant in the outer part of the layer. This approach has had
some limited success. Its advantage is that it is mathematically and conceptually simple.
Its disadvantage is that the physics is somewhat dubious - there is no reason to assume that
the Reynolds shear-stress is proportional to the mean velocity gradient but it turns out that
they are roughly similar. The Reynolds shear stress is large just outside the laminar sublayer
(where it is approximately equal to the wall shear-stress and drops to zero outside the layer -
just as the velocity gradient does. This does not, however, imply that they are proportional
to each other.
There are a variety of other turbulence models in use. They all attempt to relate the
turbulence quantities to the mean quantities in some fashion. They allow engineers to solve
the boundary layer equations using CFD methods. It is important to remember, however,
that they are all based on extra assumptions about the physics which are no better than our
dimensional analysis and hand-waving arguments. As such they should be treated with care
and not trusted unless they can be checked against experimental results.
71
13 Appendix:Power law prole used to calculate skin
friction - zero pressure-gradient ow.
Although we have found that a sensible physical functional form for the boundary layer
velocity prole is logarithmic, before proceeding to use this fact to derive a skin-friction law,
we shall consider an older and simpler approach to the problem. It was found quite early on
in research that a reasonable empirical curve-t to data is a 1/7th power law i.e.
u
U

= C
1
_
yU

_
1/7
, (237)
where the constant C
1
actually varies with the Reynolds number but, lets say, has a value of
around 8.7 for this 1/7th power-law form at typical or reasonable Reynolds numbers and is
a constant. There have been various variations on the value of the power but essentially it is
found that a power-law form like this ts data reasonably well
6
. This power-law form is very
convenient mathematically since we can easily integrate or dierentiate it where necessary.
Here we will derive the skin-friction variation for a general power-law prole of this form i.e.
u
U

= C
1
_
yU

_
1/n
, (238)
where for the 1/7th power-law the n would be 7. Note that the best t to a typical prole will
produce a particular value of C
1
for a given choice of n. It has been found empirically that
the choice of n giving a good t to data depends on the Reynolds number of the boundary
layer: higher Reynolds numbers are better tted by larger values of n (for n = 7, C
1
= 8.74,
for n = 9 C
1
= 10.6. This is a reection of the fact that the proles change with Reynolds
numbers and we are trying to t a universal function to them which is not appropriate -
nevertheless the approach gives roughly reasonable results and is simple to analyse. It also
shows - in a straightforward manner - the way in which skin-friction and boundary layer
growth can be derived from an assumed form of boundary layer prole. At the edge of the
layer this becomes
U
1
U

= C
1
_
U

_
1/n
, (239)
Now consider the momentum equation
d
dx
=
C

f
2
=
U
2

U
2
1
. (240)
Now we need the momentum thickness and so we integrate the prole up (from 0 to ) and
nd that

=
n
(1 +n)(2 +n)
, (241)
which can then be substituted into the momentum integral equation along with (239 giving
n
(1 +n)(2 +n)
d
dx
= (C
1
)
2n/(n+1)
_
U
1

_
2/(n+1)
. (242)
6
note, of course, that it does not have the correct asymptotic behaviour as y but that is OK as long
as we avoid that region. Remember in question 1 in the notes where you showed that the integral properties
are not aected much as long as you cut o the integrals at some nite below which the t to the proles
is good
72
which we can now separate and integrate since we have only and x in the equation (and
some constants). We will assume that = 0 at x = 0. This then integrates to

x
= C

2
R
2/(n+3)
x
, (243)
where the constant C

2
is given by
C

2
=
_
(2 +n)(n + 3)
n
C
2n/(n+1)
1
_
(n+1)/(n+3)
, (244)
and nally dierentiating (243) and substituting into the momentum integral equation we
can nd
C

f
= C

3
R
2/(n+3)
x
, (245)
where
C

3
= C

2
(2n/[(2 +n)(n + 3)]) . (246)
Whilst the mathematics is a little tedious is it not horribly complicated. With the choice
of n = 7 then we nd C

f
= 0.0592R
1/5
x
and /x = 0.37R
1/5
x
. These are useful rough
expressions which give reasonable estimates of the skin-friction and boundary layer growth
for zero pressure-gradient turbulent boundary layers.
73
Appendix 2:Skin friction derivation for a boundary layer
from the log law
SEE DUNCAN, THOM AND YOUNG P.320 to 323 The power-law t for the velocity
prole of a turbulent boundary layer is, at best, a crude approximation. There is no physics
behind the shape or scaling based on that approach. We have seen in previous sections
that there are sound scaling arguments that apply to the turbulent boundary layer and
these result in a logarithmic behaviour. The scaling arguments also show that there are
two appropriate scaling regions in the ow and hence the analysis based on this theory is
necessarily more complicated. It is, however, more sound and also gives excellent agreement
with good experimental data. This is the only semi-theoretical solution known based on
similarity laws. In the zero pressure-gradient case it is straightforward (if a little tricky) but
in the case with pressure-gradient things become more dicult. It gives us an expression for
how the skin-friction varies based on the momentum integral equation and using the velocity
prole given earlier (i.e. the log-law plus a deviation function).
So we have
d
dx
=
C

f
2
(247)
U
U

=
1

ln
_
yU

_
+A +

w
_
y

_
(248)
U
1
U
U

=
1

ln
_
y

_
+

(w(1) w(y/)) (249)


hence we can write
U
1
U
U

= f(y/) (250)
and we will dene a new useful variable by
S =
U
1
U

(251)
Now let us examine the displacement thickness

=
_
1
0
(1 U/U
1
)d
=
U

U
1
_
1
0
U
1
U
U

d
=
1
S
_
1
o
f()d
=
C
1
S
(252)
where as previously = y/. This formula can also be used to give a precise denition of
since

is know accurately (from an integral), hence


=

S
C
1
(253)
74
and by a similar approach we can show that

=
C
1
S

C
2
S
2
(254)
where
C
1
=
_
1
0
f()d
C
2
=
_
1
0
f
2
()d (255)
Now from the momentum integral equation
d
dx
=
C

f
2
(256)
which can be written non-dimensionally as
dR

dR
x
=
C

f
2
=
1
S
2
(257)
where R

= U
1
/ and R
x
= U
1
x/. Now at the edge of the layer we can write
[S A

w(1)] = ln
_
U

_
(258)
or
U

= e
[SA

w(1)]
(259)
and this leads to
R

= S
U

(
C
1
S

C
2
S
2
) (260)
and substituting
R

= (C
1

C
2
S
)e
[SA

w(1)]
(261)
But we also know that
dR

dR
x
=
1
S
2
(262)
from the momentum equation. For convenience we can put
A +

w(1) = (1) (263)


and taking a derivative w.r.t. S
dR

dS
= (
C
2
S
2
+C
1

C
2
S
)e
(S(1))
(264)
Now apply the chain-rule
dR
x
dS
=
dR
x
dR

.
dR

dS
(265)
and nally substituting in we nd
75
dR
x
dS
= (C
2
C
2
S +C
1
S
2
)e
(S(1))
(266)
integrating from the leading edge of the plate where x = 0 and S = 0 then we nd
R
x
= [C
1
S
2
(
2C
1

+C
2
)S +
2

(
C
1

+C
2
)]e
(S(1))
(267)
S is usually large so it is common to retain only the S
2
terms and from this it is not dicult
to show that the nal skin friction law is

_
1
C

f
= A

+B

log
10
(R
x
C

f
) (268)
where A

= 1.7, B

= 4.15.
76
Appendix 3:Datacard
Module 3A1
Boundary Layer Theory Data Card
Displacement thickness;

=
_

0
_
1
u
U
1
_
dy
Momentum thickness;
=
_

0
(U
1
u)u
U
2
1
dy =
_

0
_
1
u
U
1
_
u
U
1
dy
Energy thickness;

E
=
_

0
(U
2
1
u
2
)u
U
3
1
dy =
_

0
_
1
_
u
U
1
_
2
_
u
U
1
dy
H =

Prandtls boundary layer equations (laminar ow);


u
u
x
+v
u
y
=
1

dp
1
dx
+

2
u
y
2
u
x
+
v
y
= 0
von Karman momentum integral equation;
d
dx
+
H + 2
U
1

dU
1
dx
=

o
U
2
1
=
C

f
2
Boundary layer equations for turbulent ow;
u
u
x
+v
u
y
=
1

dp
dx

u

y
+

2
u
y
2
u
x
+
v
y
= 0
77

You might also like