You are on page 1of 21

Introduction to Turbulence

by
Håkan Gustavsson
Division of Fluid Mechanics
Luleå University of Technology
Foreword

The intention with these pages is to present the student with the basic theoretical concepts of
turbulence and derive exact relations from the governing equations. The idea is to show that
despite the complexity of turbulent flows, some general properties can be educed from the
equations. Hopefully, this will remove some of the mystique that has surrounded turbulence
as a topic in undergraduate courses. The material is used as lecture notes in the course MTM
162 (Advanced Fluid Mechanics) given in the last year of undergraduate studies at Ltu.

Luleå,
October 2006

Håkan Gustavsson
Division of Fluid Mechanics
Contents Page

1. Introduction 1
2. Reynolds’ decomposition 3
3. Equations for the mean flow 4
3.1 Continuity 4
3.2 Momentum 4
3.3 Kinetic energy 6
4. Equations for the turbulent fluctuations 7
4.1 Momentum 7
4.2 Kinetic energy 8
5. Turbulent channel flow 9
5.1 Momentum equation 9
5.2 Turbulence production 13
5.3 Mean velocity profile 14
6. Kolmogorov microscales 16
7. Turbulence structure 17

References 18
1

1. Introduction
Turbulence is generally considered one of the unresolved phenomena of physics. This means
that there is not one model that describes the appearance and maintenance of turbulence in all
situations where it appears. Because of the technical importance of turbulence, models based
on correlations of particular experimental data have been developed to a large extent.
The task to develop a general turbulence model is challenging since turbulence appears almost
everywhere: Flows in rivers, oceans and the atmosphere are large scale examples. Flows in
pipes, pumps, turbines, combustion processes, in the wake of cars, airplanes and trains are
some technical examples. Even the blood flow in the aorta is occasionally turbulent. In fact,
one can say that turbulence is the general flow type on medium and large scales whereas
laminar flows appear on small scales, and where the viscosity is high. For example, the flow
of lubricating oils in bearings is laminar.
Before we discuss the technical aspects of turbulence it is necessary to state its main
kinematic characteristics. In the list below, some flows may have one or two features but
turbulence has all three.

Irregularity. Observing structures in the flow from a smoke stack, or measuring the velocity
in a pipe flow, show that any particular pattern never repeats itself. This randomness suggests
that a statistical treatment of turbulence is worthwhile. In fact, statistical quantities such as
mean values, correlations etc. are generally repeatable and make statistical theories attractive.

Mixing. A case of randomness is the deflection of a water surface due to wind. However, in
this flow fluid particles stay largely in one place which they do not in a stirred cup of coffee.
Thus, mixing is a prominent feature of turbulence and involves mixing of particles and all
physical quantities related to particles i.e. heat, momentum etc.

Three-dimensional vorticity fluctuations. Turbulent flows always exhibit high levels of


vorticity. The maintenance of turbulence requires the process of vortex stretching which
occurs only in 3D. A word of caution: 2D turbulence may occur if magnetic fields control the
flow.

In addition to these kinematic characteristics, we may also add some features of turbulence
that help to decide when to expect its appearance and judge its physical significance.
2

Large Reynolds number. Typical for all turbulent flows is that a relevant Reynolds number
(UL/ν) for the flow is large. It is the objective for stability theory to determine the critical
Reynolds number over which turbulence may appear. This is a part of the turbulence enigma.

Dissipation. Turbulent flows loose mechanical energy due to the action of shear stresses
(dissipation) at a much larger rate than laminar flows. The flow losses are much larger. The
energy is converted into internal energy and thus shows up as an increase of temperature.
Without a continuous supply of energy, turbulence decays (cf. the stirred coffee in a cup when
the spoon is removed).

Continuum. The smallest scales of turbulence are generally far larger than molecular length
scales. Turbulence is therefore a continuum phenomenon, and should be possible to describe
by the equations of motion of fluid mechanics.

Turbulent flows are flows. Turbulence is a feature of the flow and not of the fluid. Thus,
different fluids show the same properties given the (non-dimensional) flow parameters are the
same.

For the analysis of turbulence phenomena we use the equations of motion (continuity,
momentum and energy) but the added complexity of randomness makes statistical tools
necessary. Mathematically, we will make frequent use of tensor notation in writing the
equations of motion. This simplifies the derivations and reduces the writing considerably. A
powerful tool is also dimensional analysis which relates the ingredient variables through their
dimensions. With simple assumptions about the flow very far-reaching conclusions can be
drawn with dimensional arguments. We will use this technique mainly when deriving the
smallest dissipative scales in turbulence, the Kolmogorov microscales.

In choosing specific flows to analyse, we pick wall bounded turbulent flows (channel flow)
for which analytical results may be derived and some crucial concepts appear. The next step
would be to treat ‘free’ turbulence (wakes, jets and free shear layers) where just a modest
level of turbulence modelling (eddy viscosity) leads to surprisingly useful results. The further
steps of turbulence modelling and the modern approach of direct numerical simulation of
turbulence (DNS) are treated at the graduate level.
3

2. Reynolds’ decomposition

Typical signals from measurements of velocity components near a solid wall are shown in
figure 2.1. The data illustrate the randomness of the signal but one can generally produce a

Figure 2.1: Near-wall data of u- v- and uv signals and the corresponding short-time variances
of u and v for T=10t*.(From Alfredsson & Johansson 1984, JFM 139)

time-average defined as
T
1
f = F = lim ∫ f (t )dt ,
T →∞ T 0
(2.1)

where f can be any of the fluid mechanical variables of interest (velocity component, pressure
etc.). We use the over-bar notation, or capital letter, for the time-average. Using this
definition, we split each instantaneous component into its time-average and a time-dependent
fluctuation. For the velocity and the pressure this gives

ui = Ui + ui′ (Note that u i ′ = 0, by definition)

p = P + p′ (2.2a,b)

This splitting of a turbulent signal is denoted Reynolds′ decomposition of a turbulent flow.


4

3. Equations for the mean flow


To describe the average and dynamic properties of turbulent flows, use is made of the basic
conservation laws of fluid mechanics, i.e. the equations for mass, momentum and energy.
These relations have in general to be complemented with a state condition for the fluid.
In this text we will not treat energy through the thermodynamic energy equation but restrict
the study to the kinetic energy. This can be done without invoking thermodynamics but rather
by multiplying the momentum equation with a suitable velocity component.

3.1 Continuity
In this presentation we will assume the flow to be incompressible i.e.
div u = ui,i = 0 (3.1)
Using the decomposition in (2.2a), it is deduced that the following relations apply for the
mean flow and the fluctuations:
Ui,i = 0 (3.2)
and u′i,i = 0 (3.3)
Thus, both the mean flow and the fluctuations satisfy the incompressible condition.

3.2 Momentum
We first formulate the averaging process for the general momentum equation and then
specialize to the Navier-Stokes equations. Thus, we have

Du i ⎛ ∂u ∂u ⎞ ∂σ (3.4)
ρ = ρ⎜⎜ i + u j i ⎟⎟ = ρFi + ij
Dt ⎝ ∂t ∂x j ⎠ ∂x j

where Fi is a mass force (typically g) and σij is the stress tensor.


Using the decomposition in (2.2a) for ui and applying the averaging process (2.1) the left hand
side of (3.4) becomes

∂u i ∂ (U i + u ′i ) ⎛ ∂U i ∂u ′ ⎞ (3.5)
ρuj = ρ (U j + u ′j ) = ρ ⎜⎜ U j + u′j i ⎟⎟
∂x j ∂x j ⎝ ∂x j ∂x j ⎠
In the averaging process, the two terms involving only one primed quantity become zero so
only the mean term and the doubly primed term remain. Using continuity, the last term can
also be written as

∂u′i ∂ (ρu′i u′j )


ρ u′j = (3.6)
∂x j ∂x j
5

For the stress tensor we have σ ij = Σ ij + σ′ij , where Σ is the time-average. The average of the

fluctuating part depends on the particular (rheological) model that is used to describe the
relation between the stress and the velocity field. As we will consider only Newtonian fluids,
the stress is linearly related to the velocity gradient. Thus, averaging gives
σ′ij = 0. (3.7)

Combining (3.4)-(3.7) then result in the following equation for the average quantities
∂U i ∂ (Σ ij − ρu′i u ′ j ) (3.8)
ρU j = ρFi +
∂x j ∂x j
where the term (3.6) has been moved to the right hand side so that a physical interpretation of
its significance can be given. (3.8) is denoted the Reynolds’ equation for a turbulent (mean)
flow and may be interpreted as if the stress is given an extra contribution due to the turbulent
fluctuations, − ρu ′i u ′j . This extra stress is denoted the Reynolds’ stress in honour of Osborne

Reynolds (1842-1912) who was the first to identify this contribution of the turbulent
fluctuations. It should be noted that the Reynolds stress has its origin in the non-linear
(advective) term in the momentum equation and is thus a property of the flow and not of the
fluid. It is also noted that the Reynolds’ stress is a second rank tensor and properly should be
denoted the Reynolds’ stress tensor. It is the objective of turbulence modelling to connect
the Reynolds’ stress to other flow quantities.
For the particular case of an incompressible Newtonian fluid, the stress depends on the
pressure and the velocity gradient through
σij = −pδij + μ(u i , j + u j,i ) (3.9)

where μ is the dynamic viscosity. For the mean flow this reduces to
Σ ij = − Pδij + μ(U i , j + U j,i ) (3.10)

and (3.8) becomes

∂U i ∂ (ρu ′i u ′ j ) (3.8)′
ρU j = −P,i + ρFi + μU i , jj −
∂x j ∂x j
Dividing through by ρ, we obtain the momentum equation for a Newtonian fluid that will be
used frequently in the sequel.

∂U i 1 ∂ (u′i u′ j ) (3.11)
Uj = − P,i + Fi + νU i , jj −
∂x j ρ ∂x j
Often, the mass force is neglected.
6

3.3 Kinetic energy


If we multiply (3.11) by Ui , neglecting Fi and using continuity, the following terms are
obtained
∂U i ∂ (U i U i / 2 ) (3.12)
Ui U j = Uj
∂x j ∂x j
1 1 ∂
U i P,i = (U i P ) (3.13)
ρ ρ ∂x i
∂ (U i , j U i ) (3.14)
U i νU i , jj = ν − νU i , j U i , j
∂x j
∂ (u ′i u ′ j ) ∂ (U i u ′i u ′j ) (3.15)
Ui = − u ′i u ′j U i , j
∂x j ∂x j
Collecting the terms (3.12)-(3.15) we obtain
∂ (U i U i /2 ) ∂
Uj
∂x j
=
∂x j
( )
− U j P/ρ + νU i, j U i − U i u ′i u ′j − νU i, j U i, j + u ′i u ′j U i, j (3.16)

The terms in (3.16) have been arranged so that a physical interpretation is possible. The left
hand side is the advected variation of the kinetic energy. On the right hand side, the first term
is denoted a transport term since an application of Gauss’ theorem shows that it is only the
changes at the surfaces that contribute to this term. The second term contains only squared
terms and because of the minus sign, it represents a loss of kinetic energy due to viscous
dissipation. The last term represents either a gain or a loss of kinetic energy and shows how
the Reynolds stresses, together with the mean shear, act to change the kinetic energy of the
mean flow.
7

4. Equations for the turbulent fluctuations


Since the Reynolds’ stresses contain the turbulent fluctuations, it is necessary to derive
equations also for the development of these quantities. We will do this in a general form and
in the next chapter specialize to plane channel flow.

4.1 Momentum
The instantaneous velocity and pressure fields (2.2) must satisfy the momentum equation so
(3.4) becomes
∂u′i
+ (U j + u′j ) =
(
∂ (U i + u′i ) 1 ∂ Σ ij + σ′ij ) (4.1)
∂t ∂x j ρ ∂x j

Here, the average (capital letters) satisfy (3.8) so a subtraction of this equation gives the
equation for the turbulent quantities:

∂u′i ∂u′
+ U j i + u′j =
(
∂U i 1 ∂ σ′ij + ρu′i u′j − ρu′i u′j ) (4.2)
∂t ∂x j ∂x j ρ ∂x j

For a Newtonian medium


σ′ij = −p′δij + μ (u′i, j + u′j,i )

and (4.2) reduces to

∂u′i ∂u′
+ U j i + u′j
∂U i
=−
1 ∂p′

∂ 2 u′i
+
(
∂ u′i u′j − u′i u′j ) (4.3)
∂t ∂x j ∂x j ρ ∂x i ∂x j∂x j ∂x j

This is the momentum equation for the turbulent fluctuations. The last term on the right hand
side is non-linear and represents the deviation from the mean of the Reynolds stress and may
therefore be denoted the fluctuating Reynolds’ stress. (4.3) illustrates the problem with the
analysis of turbulent flows: in order to determine the turbulent fluctuations we need to know
their average expressed as the Reynolds stress. It is therefore necessary to go to higher order
whereby new unknown quantities appear. Ending this sequence of equations is the closure
problem of turbulence and is so far unresolved. Because of this difficulty, it has become
necessary to find simpler models for turbulence. Despite the introduction of large scale
simulations for turbulence flows, turbulence modelling is still a very active research area.
8

4.2 Kinetic energy


If we multiply eq. (4.3) by u′i , we can derive an equation for the kinetic energy of the
fluctuation field. Averaging over time (carry out the details!), result in the following
expression:

∂ ⎡ ⎛ ∂u′ ∂u′ ⎞⎤
2
∂q ν ⎛ ∂u′ ∂u′ ⎞ (4.4)
Uj =− ⎢u′j (p′/ρ + q) − νu′i ⎜⎜ i + j ⎟⎟⎥ − u′i u′j U i, j − ⎜⎜ i + j ⎟⎟
∂x j ∂x j ⎢ ⎝ ∂x j ∂x i ⎠⎥⎦ 2 ⎝ ∂x j ∂x i ⎠

where q = u′i u′i /2 .
The first term on the right hand side is interpreted as due to turbulent transport; its
contribution is zero if integrated between solid walls. The second term is turbulent production
and should be compared with the similar term in (3.16). Just note the different signs! What
appears as a gain in energy for the fluctuations is seen as a loss of energy for the mean flow.
This gives a physical significance to the Reynolds stresses: they act together with the mean
shear (Ui,j) to transfer kinetic energy between the mean flow and the turbulent fluctuations.
The last term in (4.4) is always negative and represents viscous dissipation, generally denoted
ε.
9

5. Turbulent channel flow


In this section we will write the momentum equation in the simple (but important!) geometry
of a plane channel flow. The mean flow is driven by a pressure gradient in the x-direction (cf
figure 5.1) and after a certain distance from the channel entrance we assume the flow
characteristics to be independent of x (‘fully developed flow’) and z (2D).
y=h
U(y) y

y1
Figure 5.1: Geometry for turbulent channel
y=-h flow, driven by a pressure gradient.
Mathematically, this is written as
U = U(y), V = 0 (5.1)
∂ ∂ (5.2)
= 0 (except for P) and = 0
∂x ∂z
The condition V = 0 follows from continuity assuming U independent of x.

5.1 Momentum equation


The two components of Reynolds’ equation for the mean flow (3.11) become,

x-dir: 0=−
1 ∂P
ρ ∂x
d2U d
+ν 2 −
dy dy
u′v′ ( ) (5.3)

y-dir: 0=−
1 ∂P d 2

ρ ∂y dy
v′ ( ) (5.4)

Here, (5.4) can be integrated in y leading to


P P (x)
+ v′2 = 0 (5.5)
ρ ρ
where P0 may be seen as an integration constant but it also shows that the pressure across a
boundary layer may change due to turbulent fluctuations. Eliminating P between (5.3) and
(5.5) gives

0=−
1 dP0
ρ dx
d2U d
+ν 2 −
dy dy
u′v′ ( ) (5.6)
10

(5.6) can be integrated in y and choosing the integration interval ∫ ...dy we obtain
−h

⎛ dU dU ⎞
0=−
1 dP0
ρ dx
(y + h) + ν⎜ −
⎜ dy dy ⎟
(
⎟ − u′v′ − 0 ) , (5.7)
⎝ y=−h ⎠

where we have used the fact that the Reynolds stress vanishes on the wall. The evaluation of
the mean shear on the wall may be expressed in terms of the wall shear stress since we have
dU 1 dU
ν = μ = τ w /ρ ≡ u ∗2 (5.8)
dy y = − h ρ dy y = − h

Here, we have defined a characteristic velocity, u ∗ , which is denoted the friction velocity (or
wall velocity). It has a fundamental role when scaling the velocity close to the wall.

The relation of u ∗ to other velocities in a flow can be derived using the local friction

coefficient Cf,x ≡
τ w , where U is a typical (mean) velocity. Substituting for u and
m ∗
ρU2m / 2

knowing that Cf,x is dependent on the Reynolds number, one obtains u ∗ /Um = Cf ,x / 2 . For a
turbulent boundary layer Cf,x = 0.059Rex-1/5 and in a turbulent pipe flow Cf,x = 0.079Re-1/4,
respectively. Thus the ratio u ∗ /Um will vary weakly with the Reynolds number.

To see the use of u ∗ in the scaling of the momentum equation we first eliminate P0 by putting

y = 0 in (5.7), using that U is symmetric and u′v′ = 0 there. This yields


1 dP0
0=− h − u *2 (5.9)
ρ dx
Combining with (5.7) gives
y dU
0 = u *2 +ν − u′v′ (5.10)
h dy

(5.10) is valid in the whole interval y ≤ h but the different terms balance each other

dependent on where in the interval we are. As the wall proximity is of most interest, it is
useful to introduce the distance from the wall as a new variable, y1 = y + h. In terms of y1,
(5.10) thus becomes
⎛ y ⎞ dU
0 = − u *2 ⎜1 − 1 ⎟ + ν − u′v′ (5.11)
⎝ h⎠ dy1
11

(5.11) is of considerable interest since it couples the Reynolds’ stress to the mean flow.
However, the balance between the different terms depend on the y1 position and to elucidate
this it is necessary to scale the equation. This can be done in (at least) two ways, one using h
as the characteristic length, the other u ∗ /ν, denoted the wall length. The velocity is scaled by

u ∗ in both cases.

Outer scaling: Scaling with h (and u ∗ )


(5.11) reduces to
y1 ν d(U/u ∗ ) u′v′
0 = −1 + + − 2
h u ∗ h d(y1/h) u*

u ∗h
Introducing the Reynolds no., Re∗ = , this expression may also be written as
ν
y1 1 d(U/u ∗ ) u′v′
0 = −1 + + − 2 (5.12)
h Re∗ d(y1/h) u*

Since Re ∗ in general is a large number, (5.12) is reduced in the center portion of the channel

y1 u′v′
to 0 = −1 + − 2 (5.13)
h u*
which shows that the Reynolds’ stress varies linearly with y1 in this region. This result has
been verified experimentally (cf. fig 5.2).

Figure 5.2: Variation of Reynolds stress


across a channel (from Reichardt)

It is observed that the linear relationship (5.13) holds in almost 70% of the channel and is thus
a surprisingly good approximation to (5.12). Closer to the wall, however, another balance
12

exists between the terms in (5.12). In particular, the viscous term must be important and it can
be incorporated by another scaling of (5.12).

Inner scaling: Scaling with ν / u ∗ (and u ∗ ).


The second scaling that may be done is such that the viscous term is weighted by unity. The
proper length scale is then ν/ u ∗ and (5.11) then becomes

y1 u ∗ 1 d(U/u ∗ ) u ′v′ (5.14)


0 = −1 + ⋅ + −
ν Re ∗ d(y1u ∗ /ν ) u *2
It is customary to introduce the new wall distance
y1 u ∗ (5.15)
y+ =
ν
and the new velocity
U (5.16)
U+ =
u∗
(5.14) can then be written as
y + dU + u ′v′ (5.14)’
0 = −1 + + − 2
Re ∗ dy + u*
With a large value of the Reynolds no., (5.14)’ reduces to
dU + u ′v′ (5.17)
0 = −1 + − 2
dy + u*
The distribution of uv close to the wall can been measured in detail with LDV and the results
are shown in figure 5.3. It is observed that the uv-value is almost constant in a large portion of
the wall layer.

Figure 5.3: Distribution of Reynolds


stress close to a wall. (from Karlsson et
al.)
13

5.2 Turbulence production


According to (4.4), the production of turbulent kinetic energy is given by − u ′i u ′j U i, j . For the

channel flow here, this reduces to


dU (5.18)
P = −u′v′
dy
The minus-sign in (5.18) needs a comment. If we consider the lower half of figure 5.1 and let
a fluid particle move upwards, its v′ component is > 0. But, since the particle starts from a
region with lower mean velocity it will cause a negative u′ where it ends up. Thus, the product

u′v′ is negative and, since dU > 0, there will be a production of turbulent energy due to this
dy
motion. A similar argument for a particle moving downward also gives a positive production.
We can now use the results of the inner scaling to estimate where the maximum turbulent
production is to be expected. Close to the wall, (5.11) reduces to

⎛ dU ⎞ (5.19)
0 = − u *2 + νU′ − u′v′ ⎜⎜ U′ = ⎟
⎝ dy1 ⎟⎠
Multiplying (5.19) by U′ we get

P = −u′v′ U′ = (u ∗2 − νU′) ⋅ U′ (5.20)

Seen as a function of U′, (5.20) can be differentiated to give optimum (maximum) for P. We

have dP = u 2 − νU′ − νU′ = u 2 − 2νU′


dU ′
∗ ∗

Thus, at
U′ = u ∗2 / 2ν (5.21)

there is maximum in P which becomes


Pmax = ( u ∗2 / 2ν )2 (5.22)

Since u ∗2 / ν is the wall-value of U′, (5.21) shows that Pmax occurs where this value is halved.

The only thing that must be asserted with this analysis is that the maximum point actually lies
within the wall region so that (5.19) is valid. Before this can be asserted, some knowledge of
the velocity profile is necessary.
14

5.3 Mean velocity profile


The information gained from the two types of scaling leads to a way to estimate the turbuelent
velocity profile. First, (5.17) indicates that close to the wall both the mean velocity and the
Reynolds’ stress are functions of y+, only. Thus,
U+ = f(y+)
and − u′v′ / u ∗2 = g(y+) (5.23)
This is denoted a wall-law.
In the outer region, where (5.13) applies, there is no information on U since it is absent in the
equation. However, some information can be obtained from the energy equation for the
fluctuations (4.4). In the outer region, it reduces to
dU d ⎛1 ⎞ (5.24)
− u′v′ = ⎜ v′p′ + v′q ⎟ + ε
dy1 dy1 ⎝ ρ ⎠
(5.13) shows that u′v′ is of the order u ∗2 which we can expect also q and p′/ρ to be. Thus we

can write dU u ∗ dF (5.25)


=
dy1 h dη
Integrating this relation from the center of the channel gives
U − U0 (5.26)
= F(η)
u∗
This type of relation is denoted a velocity defect law and is valid in the center portion of the
channel. Closer to the wall, the wall law (5.23) applies and it is natural to ask what happens in
the intermediate region. This can be given some input by considering the velocity derivative
dU/dy1 derived from the two relations (5.23) and (5.26).

Law of the wall (5.23)⇒ dU = u ⋅ df ⋅ dy + = u ∗ ⋅ df


2
(5.27)
ν dy +

dy1 dy + dy1

Law of the wake (5.26) ⇒ dU = u ⋅ dF ⋅ dη = u ∗ ⋅ dF (5.28)


dη dy1 h dη

dy1
In the overlap region both relations must hold. Multiplying both sides with y1 / u ∗ gives then

df dF (5.29)
y+ =η
dy + dη
Here, the left hand side is a function of y+ and the left hand side a function of η. Thus they
must be a constant. This constant is denoted von Karmáns constant (κ) and is roughly 0.41.
Integrating the two relations then give
15

1
f (y+ ) = ln y + + constant (5.30)
κ
and similarly for F. The presence of a logarithmic velocity profile is a prominent character of
turbulent boundary layers. In figure 5.4 we show an example; note that the y+-coordinate is
logarithmic. This makes the wall region exaggerated.

Figure 5.4: Turbulent velocity profile close


to a wall. Scaling in wall variables; note the
logarithmic distance. This exaggerates the
near wall region. (From Karlsson et al.)

In section 5.2 we derived the condition for maximum turbulence production assuming this to
be in the near-wall region and showed this to be where U′ = u ∗2 / 2ν . Expressed in wall

variables this may also be written as dU+/dy+ = 1/2. For the linear profile closest to the wall,
dU+/dy+ = 1 and in the logarithmic region dU+/dy+ = 2.44/y+. Thus we conclude that the value
1/2 is obtained in a point between these two regions confirming the assumption that the
maximum turbulence production is obtained in the near wall region. It turns out that
maximum turbulence production occurs at y+ ≈12-15.
16

6. Kolmogorov microscales
Turbulence is generated at fairly large scales but due to non-linear interactions smaller and
smaller scales are involved. At the smallest level, the energy is lost to viscous dissipation and
it was originally suggested that this process was isotropic, i.e. does not have a preferred
orientation. At this level, the scales should be governed by viscosity (ν) and the dissipation
rate (ε) only. These quantities have dimensions m2/s and m2/s3, respectively, so the design of
length, time and velocity scales based on these quantities is straight forward and yield the
following set:
η = (ν 3 / ε )
1/ 4
Length scale:

τ = (ν / ε )
1/ 2
Time scale: (6.1a-c)

ϑ = (νε )
1/ 4
Velocity scale:
These are called the Kolmogorov microscales. The length scale is the smallest scale in a
turbulent flow.
A couple of points related to these scales are of interest.
• The Reynolds number based on the microscales is equal to unity.
• For turbulence in balance, the dissipation rate must be equal to the energy transfer
from the largest scales, where energy is put in. By knowing the energy input in e.g. a
stirring process it is thus possible to determine the microscales.
17

7. Turbulence structures
Starting in the 1960s, much research in turbulence has been directed to find structures and
events that have some deterministic features. One such process is the so-called bursting
process in the near-wall region. During this event most of the turbulence energy is produced.
The sequence is depicted in figure 9.1. The starting point is the appearance of longitudinal
(low-speed) streaks on which develops 3D and highly unstable structures (hairpin vortices).
An underlying idea is that the appearing distortion of the mean profile produces an inflexion
in the profile which, from linear stability theory, is known to be unstable. However, later
stability research has established that other mechanisms may be active; one is the process of
transient growth which is able to explain the appearance of the longitudinal streaks.

Figure 9.1: The bursting sequence; from


Kline et al. (1967)
18

References
There is a multitude of books on turbulence of which the list below is just a short example.
Despite its age, the book by Tennekes and Lumley is still a quite useful opener to the subject
and may be recommended. A more modern book, where turbulence modelling and
simulations are treated, is the book by Pope.

Bradshaw P. 1985, An Introduction to Turbulence and its Measurement, Pergamon Press


Durbin P. and Pettersson Reif B.A. 2001, Statistical Theory and modelling for Turbulent Flows, Wiley.
Landahl M.T. and Mollo-Christensen E. 1986, Turbulence and random processes in fluid mechanics,
Cambridge University Press.
Lesieur M. 1990, Turbulence in Fluids, Kluwer.
McComb W.D. 1990, The Physics of Fluid Turbulence, Oxford Science Publications.
Pope S.B. 2000, Turbulent Flows, Cambridge University Press
Rotta J.C. 1972, Turbulente Strömungen, B.G. Teubner, Stuttgart
Schlichting H. 1979, Boundary layer theory, 7th edition, McGraw Hill
Tennekes H. and Lumley J.L. 1972, A first course in turbulence, MIT Press.
Townsend A.A. 1980, The structure of turbulent shear flow, 2nd edition, Cambridge University Press.
Wilcox D.C. 1993, Turbulence Modeling for CFD, DCW Industries.