CHAPTER 6

Turbulent Flow and Heat

Transfer

6.1 INTRODUCTION
The following are the characteristics of turbulent motion
• Irregularity
Complex variations of velocity and temperature with space and time (fluc-
tuations) are the dominant characteristics of a turbulent flow. The irregular
motion is generated due to random fluctuations (Fig. 6.1). It is postulated
that the fluctuations inherently come from disturbances (such as, roughness
of the solid surface) and they may be either reduced due to viscous damping
or may grow by drawing energy from the free stream. At a Reynolds number
less than the critical, the kinetic energy of flow is not enough to sustain the
random fluctuations against the viscous damping and in such cases laminar
flow continues to exist. At a Reynolds number somewhat higher than critical,
the kinetic energy of flow supports the growth of fluctuations and transition
to turbulence is induced.
• Strong mixing
The fluctuating turbulent motion promotes higher level of transfer of momen-
tum, heat and mass; - is practically the most important feature. Turbulent
flows appear to be random. Turbulent flows are not always free of coherent
structures. The coherent structures component is periodic or at least repeat-
able.
• Three-dimensional turbulent motion
For a parallel flow, the axial velocity component is
′
u(y, t) = u(y) + u (Γ, t)

where y is the normal direction and Γ is any space variable.

Even if the bulk motion is parallel, the fluctuation u′ being random varies in
all directions. Now let us look at the continuity equation

171
172  Fundamentals of Convective Heat Transfer

(b)
(a)

Presence of turbulence is appreciated from the velocity signal (up ) at any

Figure 6.1
point P in the flow field

′
∂u ∂u ∂v ∂w
+ + + =0
∂x ∂x ∂y ∂z
′
Since ∂u
∂x
6= 0, the above equation depicts that y and z components of velocity
exist even for the parallel flow if the flow is turbulent. We can write
′
u(y, t) = u(y) + u (Γ, t)
′
v = 0 + v (Γ, t)
′
w = 0 + w (Γ, t)

Turbulence is a three-dimensional process in which vortex stretching and tilt-

ing are essential mechanisms for energy exchange between different scales of
motion.

• Turbulent motion carries vorticity and consists of interacting eddies

Wide spectrum of eddy sizes and corresponding fluctuation frequencies are
shown in Fig. 6.2. In a turbulent flow, energy is fed from large scales mainly
to smaller scales by nonlinear processes.

• The term homogeneous turbulence implies that the velocity fluctuations in the
system are random.

• The average turbulent characteristics are independent of the position of the

fluid, i.e., invariant to axis translation.

• If the velocity fluctuations are independent of the axis of reference (invariant

to axis rotation and reflection), the restriction leads to isotropic turbulence,
which by definition is always homogeneous.
Turbulence is generally damped by viscosity. One of the effects of viscosity
on turbulence is to make the flow more homogeneous and less dependent on
direction. If the turbulence has the same structure quantitatively in
 Turbulent Flow and Heat Transfer  173

Free jet
u signals
at 6 locations

Figure 6.2 Wide spectrum of eddy sizes and corresponding frequencies

all parts of the flow field, the turbulence is said to be homogeneous.

The turbulence is called isotropic if its statistical features have no
directional preference and perfect disorder persists. Its velocity fluc-
tuations are independent of the axis of reference, i.e., invariant to axis rotation
and reflection. Isotropic turbulence is by its definition always homogeneous.
In such a situation, the gradient of the mean velocity does not exist. The
mean velocity is either zero or constant throughout. However, when the mean
velocity has a gradient the turbulence is called anisotropic.
A little more discussion on homogeneous and isotropic turbulence is needed
at this stage. The term homogeneous turbulence implies that the velocity
fluctuations in the system are random. The average turbulent characteristic
are independent of the position in the field, i.e., invariant to axis translation.
Consider the root mean square (rms) velocity fluctuations:
p p p
u′ = u¯2 , v ′ = v¯2 , w′ = w¯2
In homogeneous turbulence, the rms values of u′ , v ′ and w′ can all be different,
but each value must be uniform over the entire turbulent field. Even if the
rms fluctuation of any component, say u′ s are constant over the entire field,
the instantaneous values of u may differ from point to point at any instant.
In addition to its homogeneous nature, if the velocity fluctuations are inde-
pendent of the axis of reference, i.e., invariant to axis rotation and reflection,
the situation leads to isotropic turbulence, which by definition as mentioned
earlier, is always homogeneous.
In isotropic turbulence, fluctuations are independent of the direction of refer-
ence and
p p p
u¯2 = v¯2 = w¯2
or
u′ = v ′ = w′
174  Fundamentals of Convective Heat Transfer

Again, it is of relevance to say that even if the rms fluctuations at any point
are same, their instantaneous values may differ from each other at any instant.
Turbulent flow is also diffusive. In general, turbulence brings about better
mixing of a fluid and produces an additional diffusive effect. The term eddy
diffusion is often used to distinguish this effect from molecular diffusion. The
effects caused by mixing are as if the viscosity is increased by a factor of
100 or more. At a large Reynolds number there exists a continuous transport
of energy from the free stream to the large eddies. Then smaller eddies are
formed continuously from the large eddies. Then from the large eddies smaller
eddies are continuously formed. Near the wall, the smallest eddies dissipate
energy and destroy themselves (Tennekes and Lumley [1]).

• Length scales of turbulence

The large scale motions with length scale L are unstable for large Reynolds
numbers. The scale L is commensurate with the dimension of the flow domain.
We shall try to understand smallest length scales of turbulent flows. At the
smallest length scale, viscosity can be effective in smoothing out the velocity
fluctuations. As mentioned, turbulence is generally damped by the viscosity
ν and the main contribution to energy dissipation ǫ can be attributed to very
small length scales of the order of the Kolmogorov-scale, ls given by (ν 3 /ǫ)1/4 .
The parameters that govern the small scale motions include dissipation rate
per unit mass, ǫ. The dimenssion of ǫ is (m2 s−3 ). The dimension of kinematic
viscosity, ν is (m2 /s). Making use of these parameters, it is possible to form
length scale, time scale and velocity scale as
 3 1/4
ν
ls = , ts = (ν/ǫ)1/2 and vs = (νǫ)1/4
ǫ
These are known as Kolmogorov scales of length, time and velocity. The
Reynolds number formed by ls and vs (ls vs /ν) = 1.
In the case of homogeneous turbulence, at high Reynolds numbers, the dissi-
pation rate, ǫ equals the production-rate of turbulence and is of the order of
u3k /Lk where uk is similar to root-mean-square velocity of turbulence and Lk
is the integral length scale of turbulence.
For the smaller length scales within the intertial range (where kinetic en-
ergy production rates due to external forcing and viscous dissipation rates
are small) the turbulent motion becomes more and more homogeneous and
isotropic even if the large scale motion is anisotropic. It is worth mentioning
that ratio of largest to smallest length scales is L/Ls , which is proportional
to (Re)3/4 .

6.2 CLASSICAL IDEALIZATION OF TURBULENT FLOWS

Since detailed descriptions of turbulent fluctuations are usually not of interest to
engineers we take a statistical approach by averaging turbulence data. Different
 Turbulent Flow and Heat Transfer  175

u
u

to to+∆t

t
Figure 6.3 Mean motion and fluctuations

trends of variation of the the mean and fluctuating components are shown in Figs.
6.3 and 6.4. Fig 6.4(b) reveals an unstready mean motion. The mean velocity has
a time period t2 and a large time period signifies low frequency oscillation of the
mean motion. The high frequency oscillations have a time period of t1 .
Statistical quantities may be calculated as
Z to+∆t
′ 1 ′
ui = ui + ui , p = p + p , u = udt (6.1)
∆t to

However the fluctuating components do not bring about the bulk displacement of
′
a fluid element. The instantaneous displacement is u dt and if that is indeed not
responsible for the bulk motion, we can conclude that

u = u + u' u = u + u'
u
Velocity

u
Velocity

v' t2>>t1
v' t1
t1 u' u'
t2

Time Time
(a) Steady mean motion (b) Unsteady mean motion

Figure 6.4 Steady and unsteady mean motions in a turbulent flow

Z to+∆t
′
u dt = 0 (6.2)
to
176  Fundamentals of Convective Heat Transfer

Due to the interaction of fluctuating components, macroscopic momentum trans-

port takes place. Therefore, interaction effect between two fluctuating components
over long period is nonzero and this yields
Z to+∆t
′ ′
u v dt 6= 0 (6.3)
to

We take time average of these two integrals and write

Z to+∆t
′ 1 ′
u = u dt = 0 (6.4)
∆t to

and
Z to+∆t
′ 1
′ ′ ′
uv = u v dt 6= 0 (6.5)
∆t to

Now, we can make a general statement with any two fluctuating parameters, say,
′ ′ ′ ′
with f and g as (f and g can be vectors or passive scalars)

′ ′ ∂f ′ ∂2f ′
f =g =0 = =0 (6.6)
∂s ∂s2
and

∂(f ′ g ′ )
f ′ g ′ 6= 0 and 6= 0 (6.7)
∂s
We shall state some rules of operation on mean time-averages herein. If f and g
are two dependent variables and if s denotes any one of the independent variables
x, y, z, t then
Z Z
∂f ∂f
= ; f ds = f ds (6.8)
∂s ∂s

The Reynolds decomposition may be expressed as

′ ′ ′ ′
u=u+u , v = v + v , w = w + w , p = p + p (6.9)

Plugging in continuity, we get

∇ · ui = 0 (6.10)
and
′
∇ · ui = 0 (6.11)
Introduction of Reynolds decomposition into the Navier-Stokes equations and sub-
sequent averaging and application of the laws of statistics leads to the appearance of
 Turbulent Flow and Heat Transfer  177

turbulence correlations (turbulent or Reynolds stresses). For example, if we perform

the aforesaid exercise on the x momentum equation, we obtain
 
∂u ∂(u u) ∂(u v) ∂(u w) ∂p
ρ + + + =−
∂t ∂x ∂y ∂z ∂x

" #
2 ∂u′ 2 ∂u′ v ′ ∂u′ w′
+µ∇ u − ρ + + (6.12)
∂x ∂y ∂z

Introducing simplifications arising out of continuity equation we shall obtain

 
∂u ∂u ∂u ∂u ∂p
ρ +ρ u +v +w =−
∂t ∂x ∂y ∂z ∂x

" #
∂u′ 2 ∂u′ v ′ ∂u′ w′
+µ∇2 u − ρ + + (6.13)
∂x ∂y ∂z

Performing a similar treatment on y and z momentum equations, we obtain the y

and z momentum equations in the form
 
∂v ∂v ∂v ∂v ∂p
ρ +ρ u +v +w =− + µ∇2 v
∂t ∂x ∂y ∂z ∂y

" #
∂u′ v ′ ∂v ′ 2 ∂v ′ w′
−ρ + + (6.14)
∂x ∂y ∂z

 
∂w ∂w ∂w ∂w ∂p
ρ +ρ u +v +w =− + µ∇2 w
∂t ∂x ∂y ∂z ∂z

" #
∂u′ w′ ∂v ′ w′ ∂w′ 2
−ρ + + (6.15)
∂x ∂y ∂z

It is to be noted that the terms containing prime symbols were not in the original
NS equations.
 ′ ′ ′   
σxx τxy τxz u′ 2 u′ v ′ u′ w′
′ ′ ′  
σT =  τxy σyy τyz  = −ρ  u′ v ′ v′ 2 v ′ w′  (6.16)
′ ′ ′ ′2
τxz τyz σzz ′
uw ′
vw ′ ′
w
′ ′
σT is the Reynolds stress tensor and written in compact form as −ρui uj

∂u
σxx = −p + 2µ − ρu′ 2 (6.17)
∂x
178  Fundamentals of Convective Heat Transfer
 
∂u ∂v
τxy = µ + − ρu′ v ′ (6.18)
∂y ∂x

The averaged equations may be written, in a compact form, as

 
∂ui ∂ui 1 ∂p ∂ ∂ui ′ ′
+ uj =− + ν − ui uj (6.19)
∂t ∂xj ρ ∂xi ∂xj ∂xj

∂ui
=0 (6.20)
∂xi
We have more unknowns than number of available equations. The modified system
of equations cannot be closed within itself unless empirical relations are supplied
from experiments to correlate the fluctuating components with the mean motion.
This is termed as the closure problem.
In solving the closure problem, the turbulent stresses should be determined by
using a turbulence model.
This section discusses only models for turbulent motions. The subgrid-scale
models for large-eddy simulations are not included in this discussion.
 
′ ′ ∂ui ∂uj 2
−ui uj = νt + − kδij (6.21)
∂xj ∂xi 3

The term νt is turbulent (eddy) viscosity. The term involving the Kronecker delta
δij in Eq. (6.21) is perhaps a somewhat unfamiliar addition to the eddy-viscosity
expression. It is necessary to make the expression applicable also to normal stresses
(when i = j). The first part of Eq. (6.21) involving the velocity gradients would
yield the normal stresses.
∂u ∂v ∂w
u′ 2 = −2νt , v ′ 2 = −2νt , w′ 2 = −2νt (6.22)
∂x ∂y ∂z
The sum of the stresses is zero because of the continuity equation. However, all
normal stresses are by definition positive quantities, and their sum is twice the
kinetic energy k of the fluctuating motion:
1  ′2 
k= u + v ′ 2 + w′ 2 (6.23)
2
Inclusion of the second part of the eddy viscosity expression (Eq. (6.21)) assures
that the sum of the normal stresses is equal to 2k. The normal stresses act like
pressure forces (i.e., perpendicular to the faces of a control volume). The second
part of Eq. (6.21) constitutes a pressure because, like pressure, energy k is a scalar
′ ′
quantity. Therefore, when Eq. (6.21) is used to eliminate ui uj in the momentum
equation, this second part can be absorbed by the pressure-gradient term so that
in effect the static pressure is replaced as unknown quantity by the pressure p + 32 k.
Therefore the appearance of k in Eq. (6.21) does not necessitate the determination
 Turbulent Flow and Heat Transfer  179

of k, it is the distribution of the eddy viscosity νt only that has to be determined.

In this context, k can be linked to the intensity of turbulence which is given by
r
1 ′2
I= (u + v ′ 2 + w′ 2 ) /U∞ (6.24)
3
We can also write non-dimensional kinetic energy as
k
2
= 1.5 I 2
U∞

6.3 REYNOLDS AVERAGED FORM OF ENERGY EQUATION

Let us consider a three-dimensional situation,
 2 
∂T ∂T ∂T ∂T ∂ T ∂2T ∂2T
+u +v +w =α + + (6.25)
∂t ∂x ∂y ∂z ∂x2 ∂y 2 ∂z 2

Using Reynolds decomposition of velocity and temperature

′ ′ ′ ′
u=u+u , v =v+v , w =w+w T =T +T (6.26)

we get
 
∂T ∂T ∂T ∂ ∂T ′ ′
u +v +w = α −u T
∂x ∂y ∂z ∂x ∂x

   
∂ ∂T ∂ ∂T
+ α − v′ T ′ + α − w′ T ′ (6.27)
∂y ∂y ∂z ∂z

The terms u′ T ′ , v ′ T ′ ,and w′ T ′ thus, cause additional heat flux in the x , y and
z directions respectively, due to turbulent motion. The total heat flux in the three
directions will therefore be given by
   
′′ ∂T ′ ′ ′′ ∂T ′ ′
qx = −ρcp α − u T ; qy = −ρcp α −v T ;
∂x ∂y

 
′′ ∂T ′ ′
qz = −ρcp α −w T (6.28)
∂z

As in the case of turbulent transport of momentum, it is convenient to define an

eddy viscosity or turbulent viscosity, to study the turbulent transport of thermal
energy, a turbulent thermal diffusivity αt can be defined.
The total heat flux in x, y and z directions can therefore be given as
′′ ′′ ′′
qx ∂T qy ∂T qz ∂T
= − (α + αt ) , = −(α + αt ) , = − (α + αt ) (6.29)
ρcp ∂x ρcp ∂y ρcp ∂z
180  Fundamentals of Convective Heat Transfer

Like eddy viscosity, αt is not a fluid property but depends on the state of turbulence.
In fact the Reynolds analogy between heat and momentum transport suggests
νt
αt = (6.30)
σt
The denominator σt is called turbulent Prandtl number. Experiments have shown
that σt varies very little across the flow. Many models make use of σt as a constant.
For the flow of air a value of 0.9 may be chosen.
Finally the transport equation (6.27)may be written as
 
∂T ∂T ∂T ∂ ∂T
u +ν +w = (α + αt )
∂x ∂y ∂z ∂x ∂x

   
∂ ∂T ∂ ∂T
+ (α + αt ) + (α + αt ) (6.31)
∂y ∂y ∂z ∂z

6.4 PRANDTL’S MIXING LENGTH

In the case of turbulent flow, velocity, pressure and temperature at a fixed point
in space do not remain constant with time but undergo very irregular fluctuations
of high frequency. The velocity of the fluid in turbulent flow can be represented by
two components

(i) mean motion

(ii) fluctuating component (or eddying motion)

The velocity of fluid is given by

′ ′ ′
u=u+u , v = v + v, w = w + w

Similarly,
′ ′
p = p + p, T = T + T

However, we must always consider that the fluctuating components do not bring
about the bulk displacement of a fluid element. Consider continuity equation in
′ ′
two dimensions: ∂u
∂x
+ ∂v
∂y
= 0. If we substitute u = u + u and v = v + v and then
perform time averaging, we shall obtain

∂(u + u′ ) ∂(v + v ′ )
+ =0
∂x ∂y
or
  " ′ #
∂u ∂v ∂u ∂v ′
+ + + =0 (6.32)
∂x ∂y ∂x ∂y
180  Fundamentals of Convective Heat Transfer

Like eddy viscosity, αt is not a fluid property but depends on the state of turbulence.
In fact the Reynolds analogy between heat and momentum transport suggests
νt
αt = (6.30)
σt
The denominator σt is called turbulent Prandtl number. Experiments have shown
that σt varies very little across the flow. Many models make use of σt as a constant.
For the flow of air a value of 0.9 may be chosen.
Finally the transport equation (6.27)may be written as
 
∂T ∂T ∂T ∂ ∂T
u +ν +w = (α + αt )
∂x ∂y ∂z ∂x ∂x

   
∂ ∂T ∂ ∂T
+ (α + αt ) + (α + αt ) (6.31)
∂y ∂y ∂z ∂z

6.4 PRANDTL’S MIXING LENGTH

In the case of turbulent flow, velocity, pressure and temperature at a fixed point
in space do not remain constant with time but undergo very irregular fluctuations
of high frequency. The velocity of the fluid in turbulent flow can be represented by
two components

(i) mean motion

(ii) fluctuating component (or eddying motion)

The velocity of fluid is given by

′ ′ ′
u=u+u , v = v + v, w = w + w

Similarly,
′ ′
p = p + p, T = T + T

However, we must always consider that the fluctuating components do not bring
about the bulk displacement of a fluid element. Consider continuity equation in
′ ′
two dimensions: ∂u
∂x
+ ∂v
∂y
= 0. If we substitute u = u + u and v = v + v and then
perform time averaging, we shall obtain

∂(u + u′ ) ∂(v + v ′ )
+ =0
∂x ∂y
or
  " ′ #
∂u ∂v ∂u ∂v ′
+ + + =0 (6.32)
∂x ∂y ∂x ∂y
 Turbulent Flow and Heat Transfer  181

v′

u′

Figure 6.5 Each dot represents a u′ v ′ pair at a given time

But we have seen that

∂u′ ∂v ′
u′ = v ′ = = =0
∂x ∂y
Invoking

∂u′ ∂v ′
= =0
∂x ∂y
∂u ∂v
in Eq. (6.32) we get ∂x
+ ∂y
= 0 which yields
′ ′
∂u ∂v
+ =0
∂x ∂y
we can also write ′ ′
∂u ∂v
=− (6.33)
∂x ∂y
If we consider momentum exchange between two adjacent layers, then on the basis
′
of above equation, it is postulated that if at any instant there is an increase in u
′
in the x direction it will be followed by an increase in v in the negative y direction.
In other words u′ v ′ is nonzero and negative. This is discerned in Fig. 6.5 which
shows a cloud of data points (sometimes called a scatter plot). The dots represent
′ ′
the instantaneous values of u v pair at different times.
For unit area of the plane PP (Fig. 6.6) the instantaneous turbulent mass trans-
′
port rate across the plane is ρv . Associated with this mass transport is a change
′
in the x component of velocity u . The net momentum flux per unit area, in the
x direction, represents the turbulent shear stress at the plane PP (Fig. 6.6) which
′ ′ ′
is ρu v . When a turbulent lump movement is in the upward direction (v > 0), it
182  Fundamentals of Convective Heat Transfer

y
u(y+l)
v l
P u(y)
P
u
u(y−l) l
y
fluid lump

x
Figure 6.6 Mixing length hypothesis

enters a region of higher u and is therefore likely to effect a slowing down fluctua-
′ ′ ′
tion in u , that is u < 0. A similar argument can be made for v < 0, so that the
average turbulent shear stress will be given as

τt = −ρu′ v ′ (6.34)

Again, let us imagine a turbulent fluid lump which is located a distance l above or
below the plane PP. These lumps of fluid move back and forth across the plane and
give rise to the eddy or turbulent-shear-stress effect.
At (y + l), the velocity would be
∂u
ū(y + l) = u (y) + l (6.35)
∂y
while at (y − l), we can write
∂u
ū(y − l) = u (y) − l (6.36)
∂y
Prandtl postulated that the turbulent fluctuation is proportional to the mean of
the above two quantities
′ 1
|u | ≈ (|∆u1 | + |∆u2 |)
2
or
′ ∂u ∂u
u ≈l = C1 l (6.37)
∂y ∂y
Here l is Prandtl’s mixing length. This is analogous to mean free path (average
distance a particle travels between collisions) in molecular transport problems.
′ ′
Prandtl also postulated that v is of the same order of magnitude as u so that
′ ∂u
v = C2 l (6.38)
∂y
 Turbulent Flow and Heat Transfer  183

Now, the turbulent shear stress could be written as

 2
′ ′ 2 ∂u ∂u
τt = −ρu v = ρl = ρνt (6.39)
∂y ∂y

The constants C1 and C2 can be included in still unknown mixing length, l. The
eddy viscosity thus becomes,

∂u
νt = l2 (6.40)
∂y

Consider the boundary layer on a flat plate. The shear stress due to laminar flow
is given by τl = µ ∂u
∂y
. The total shear stress is

∂u ∂u
τ = τl + τt = ρ ν + ρνt (6.41)
∂y ∂y
or
∂u
τ = ρ(ν + νt ) (6.42)
∂y
The turbulent viscosity, νt in Eq. (6.42) can be determined from Eq. (6.40). How-
ever, our problem is still not resolved. How do we determine the value of l, the
mixing length? Several correlations, using experimental results for τt have been
proposed to determine l. In the regime of isotropic turbulence, the most widely
used value of mixing length is
l=χy (6.43)
where y is the distance from the wall and χ is known as von Karman constant.
Experiments have shown that χ is approximately equal to 0.4 (Tennekes and Lumley
[1]).

6.5 UNIVERSAL VELOCITY PROFILE (ON FLAT PLATE)

The flow fields that are dominated by high frequency oscillations (small time peri-
ods), are isotropic and the mean velocity profiles in such flow fields are universal.
In this section we shall discuss the universal velocity profile.
A very thin layer next to the wall behaves like a near wall region of laminar
flow. The layer is known as laminar sub-layer and its velocities are such that the
viscous forces dominate over the inertia forces (Fig. 6.7). No turbulence exists in it.
We know that inertial effects are insignificant in the near wall region and can write

∂ 2 u ∂ u′ v ′
ν − =0 (6.44)
∂y 2 ∂y
which can be integrated as
∂u
ν − u′ v ′ = constant (6.45)
∂y
184  Fundamentals of Convective Heat Transfer

Turbulent zone
U

8
Velocity profile
+
y =70
Buffer zone
+ Us
y

δs Viscous sublayer

Figure 6.7 Different zones of a turbulent flow past a wall

Again, as we know that the fluctuating components vanish near the wall, the shear
stress on the wall is purely viscous and it follows that:
∂u τw
ν = (6.46)
∂y ρ
or
r
u−0 τw u2 τw
= = τ where uτ = (6.47)
y−0 ρν ν ρ

The quantity uτ is known as friction velocity, given by τw = ρ u2τ . From Eq. (6.47),
it is possible to write
uτ u
y = (6.48)
ν uτ

Hence a nondimensional coordinate may be defined as y + = yuτ /ν and we write

down the variation of the nondimensional velocity within the sublayer as

u+ = y + (6.49)

In the turbulent zone, the turbulent shear stress from Prandtl’s mixing length model
can be written as
 2
2 ∂u
τt = ρ l (6.50)
∂y

Where l is the mixing length = χ y and χ is a von Karman constant we write

 2
τt ∂u
= χ2 y 2 (6.51)
ρ ∂y
 Turbulent Flow and Heat Transfer  185

The shear stress in the turbulent region dictates the velocity Us at the edge of the
sublayer (Fig.6.7). The wall shear stress is estimated as τw = µUs /δs .
 2  2
∂u uτ
= [since τt = τw ] (6.52)
dy χy

uτ
u= ln y + constant (6.53)
χ
or
 
u 1 βν
= [ln y − ln y0 ] y0 being very small = (6.54)
uτ χ uτ
or
u 1 h yuτ i
= ln − ln β (6.55)
uτ χ ν
or
 
1
u+ = A1 ln y + + D1 put A1 = (6.56)
χ

These constants were determined from experiments. For smooth ducts, A1 has been
observed as A1 = 2.5 and D1 = 5.5 or

u+ = 2.5 ln y + + 5.5 (6.57)

or u+ = 2.5 ln [Ey + ] where E = 9.0 for smooth walls.

Finally the log-law is defined for the turbulent zone as
1
u+ = ln [Ey + ] (6.58)
χ
The coefficient χ is known as the von Karman constant.
The location where log profile and linear profile meet can be calculated as:
1
u+ = ln [Ey + ] (6.59)
χ
| {z }
log-law
or
1
y+ = ln [Ey + ] (from Eq. 6.49)
χ
or

y + = 11.63 (6.60)

As such, boundary layer thickness is around 1500 y + .

10
Turbulence Modeling

10.1 Introduction
The fluid flow in a domain can be completely described using

1. the continuity equation

2. the Navier-Stokes equation and

3. the associated boundary conditions.

For incompressible flows, the governing equations can be written as follows:

Continuity Equation
∂ui
=0 (10.1)
∂xi

Navier-Stokes Equation
∂ui ∂ui 1 ∂p∗ ∂ 2 ui
+ uj =− +ν (10.2)
∂t ∂xj ρ ∂xi ∂xj ∂xj
in which ui is the instantaneous velocity component in the xi direction, p∗ is
the sum of the pressure and the gravity body force, ρ is the density of the
fluid, ν is the kinematic viscosity of the fluid and the Einstein summation
convention applies to the repeated indices (i and j take a value of 1 to 3).
Equations 10.1 and 10.2 are the general quations for fluid flow and are valid

1
Turbulence Modeling 2

for any Reynolds number, i.e., even for the turbulent flows. Although numer-
ical procedures are available to solve these equations, limitations in capacity
and speed of the present day computers make it impossible to obtain a di-
rect solution of the equations for the practically relevant complex turbulent
flows of engineering interest. Also, engineers are usually not interested in the
complete description of the fluctuating motion of a turbulent flow but are
interested in studying the effect of turbulence on the mean flow behaviour.
Therefore, a statistical approach is usually taken (Reynolds, 1895) and the
equations are averaged over a time scale which is long compared with the
time scale of the turbulent motion but small compared with the time scale
of the unsteady mean flow.
In the approach suggested by Reynolds (1895), the velocity and pressure
are decomposed into the mean and the fluctuating quantities as follows:

ui = ui + u0i , p∗ = p + p0 (10.3)

in which the overbar denotes the mean quantity and the prime denotes the
fluctuating part. ui and p are defined as
Z t2 Z t2
1 1
ui = ui dt, pi = p∗ dt (10.4)
t2 − t1 t1 t2 − t1 t1
The following rules of averaging are valid.

u0i = 0, p0 = 0
ui uj = ui uj + u0i u0j
u0i uj = u0j ui = 0 (10.5)

Substitution of Equations 10.3 in 10.1 and 10.2 and subsequent averaging

according to Equations 10.4 and 10.5 yields the following equations.
∂ui
=0 (10.6)
∂xi
" #
∂ui ∂ui 1 ∂p ∂ ∂ui
+ uj =− + ν − u0i u0j (10.7)
∂t ∂xj ρ ∂xi ∂xj ∂xj
in which u0i u0j (= τij ) is the Reynolds-Stress tensor. Equations 10.6 and 10.7
are the governing equations for describing the mean flow characteristics of
Turbulence Modeling 3

a turbulent flow. However, they do not form a closed system since they
contain the unknown Reynolds-stess tensor with six unknowns. Mean val-
ues of pressure and velocity can be calculated only when these six unknown
Reynolds-Stresses are expressed in terms of the mean velocity field in a con-
sistent way. As shown in the next section, exact transport equations for u0i u0j
can be derived by manipulating the Navier-Stokes equation. However, they
contain correlations of next higher order, i.e., of order three and the closure
is still not achieved. Expressing turbulence correlations of a particular or-
der in terms of lower order correlations and/or mean flow characteristics in
order to obtain a closed system is termed as “Reynolds-Stress Closure Mod-
eling”. Several turbulence models are available for closing the system for use
in engineering applications. These are usually first-order closure models i.e.
they use closure equations for the second moment u0i u0j . Several second-order
closure models have been developed in the recent past for representing the
third moments. However, they require significant computational power for
their solution and are still in development stage as far as the engineering ap-
plications are concerned. A description of some of the first-order turbulence
models is presented in this chapter. Transport equations for the turbulent
kinetic energy and the turbulent dissipation rate constitute an essential part
of some of these models and therefore, their derivation is presented next.

10.2 Equations for Turbulent Fluctuations

Equations for turbulent fluctuations are obtained by subtracting the Reynolds
averaged equations from the governing Equations 10.1 and 10.2. Subtracting
Equation 10.6 from Equation 10.1 one obtains

∂u0i
=0 (10.8)
∂xi
Subtracting Equation 10.7 from 10.2
∂ ∂ ∂ui
[ui − ui ] + (uj + u0j ) (ui + u0i ) − uj
∂t ∂xj ∂xj
1 ∂ ∂
= − [p − p] + ν ∇2 (ui − ui ) + [τij ] (10.9)
ρ ∂xi ∂xj
Turbulence Modeling 4

or
∂u0i ∂ui ∂u0i ∂ui ∂u0i ∂ui
+ uj + uj + u0j + u0j − uj
∂t ∂xj ∂xj ∂xj xj ∂xj
0
1 ∂p ∂τij
= − + ν ∇2 u0i +
ρ ∂xi ∂xj
or
∂u0i ∂u0i ∂ui ∂u0i 1 ∂p0 ∂τij
+ uj = −u0j − u0j − + ν ∇2 u0i + (10.10)
∂t ∂xj ∂xj ∂xj ρ ∂xi ∂xj
Equations 10.8 and 10.10 represent the field equations for the fluctuating
components. These equations may be solved to obtain the fluctuating com-
ponents u0i . However, such a solution is very difficult to obtain since the
solution for u0i depends on the global history of the mean-velocity field with
an implicit dependence on its own initial and boundary conditions.

10.3 Reynolds-Stress Transport Equation

The Reynolds-Stress transport equation is obtained from the second moment
of Equation 10.10. Equation 10.10 for the ith component is multiplied by the
fluctuating velocity u0j , and the equation for the jth component is multiplied
by the velocity u0i . These two equations are added and the resulting equation
is time averaged to obtain the transport equation for τij . For the sake of
simplicity, the derivation for transport of τ12 is shown below.
∂u01 ∂u01 ∂u01 1 ∂p0
"
∂u1
u02 + uk + u0k + u0k +
∂t ∂xk ∂xk ∂xk ρ ∂x1
#
2 0 ∂
−ν ∇ u1 + τ1k
∂xk
∂u02 ∂u02 ∂u02 1 ∂p0
"
0 ∂u2
+u1 + uk + u0k + u0k +
∂t ∂xk ∂xk ∂xk ρ ∂x2
#
2 0 ∂
−ν ∇ u2 + τ2k = 0 (10.11)
∂xk
or
∂u01 ∂u02 ∂u01 ∂u02
! " #
∂u1
u02 + u01 + uk u02 + u01 + u02 u0k
∂t ∂t ∂xk ∂xk ∂xk
Turbulence Modeling 5

∂u01 ∂u02
( )
∂u2
+ u01 u0k+ u0k u02 + u01
∂xk ∂xk ∂xk
0 0
( )
1 ∂p ∂p
+ u02 + u01
ρ ∂x1 ∂x2
h i
− ν u02 ∇2 u01 + u01 ∇2 u02
∂ ∂
+ u02 [τ1k ] + u01 [τ2k ] = 0
∂xk ∂xk
or
∂ 0 0 ∂ 0 0 ∂ ∂
[u1 u2 ] + uk [u1 u2 ] + u02 u0k (u1 ) + u01 u0k (u2 )
∂t ∂xk ∂xk ∂xk
0 0
" #
0 ∂u1 u2 1 ∂ 0 0 ∂ 0 0
+ uk + (p u2 ) + (p u1 )
∂xk ρ ∂x1 ∂x2
p0 ∂u02 ∂u01
" #
− +
ρ ∂x1 ∂x2
∂ ∂
+ u02 [τ1k ] + u01 [τ2k ]
∂xk ∂xk
2 0 2 0 2 0
" ( )
∂ u ∂ u ∂ u
−ν u02 1
+ 1
+ 1
∂x21 ∂x22 ∂x23
∂ 2 u02 ∂ 2 u02 ∂ 2 u02
( )#
0
+ u1 + + =0
∂x21 ∂x22 ∂x23
or
" #
∂ 0 0 ∂ 0 0 ∂u1 ∂u2
[u1 u2 ] + uk [u1 u2 ] + u02 u0k + u01 u0k
∂t ∂xk ∂xk ∂xk
0 0 0 0
( )
∂u1 u2 uk 0 0 ∂uk 1 ∂ 0 0 ∂ 0 0
+ − u1 u2 + (p u2 ) + (p u1 )
∂xk ∂xk ρ ∂x1 ∂x2
p0 ∂u02 ∂u01
" #
− +
ρ ∂x1 ∂x2
∂τ1k ∂τ2k
+ u02 + u01
∂x ∂xk
" k
∂u01 ∂u02
!#
2 0 0
− ν ∇ (u1 u2 ) − 2 =0 (10.12)
∂xk ∂xk
Turbulence Modeling 6

Equation 10.12 is obtained by noting that

∂ 2 (u1 u2 ) ∂ 2 u2 ∂ 2 u1
!
∂ ∂ ∂u1 ∂u2
= (u1 u2 ) = u1 + 2 + u2
∂x1 ∂x1 ∂x1 ∂x21 ∂x1 ∂x1 ∂x21
or
∂ 2 (u1 u2 ) ∂ 2 u2 ∂u1 ∂u2 ∂ 2 u1
= u1 + 2 + u2
∂x2k ∂x2k ∂xk ∂xk ∂x2k
or
∂u1 ∂u2
∇2 (u1 u2 ) = u1 ∇2 u1 + u2 ∇2 u2 + 2 (10.13)
∂xk ∂xk
Substituting
∂u0k
= 0,
∂xk
∂τ1k ∂τ2k
u02 = u01 =0
xk ∂xk
and time averaging equation 10.12 over the time scale of turbulent fluctua-
tions, it can be shown that
∂τ12 ∂τ12
+ uk
∂t ∂xk
!
∂u1 ∂u2
= − τ2k + τ1k
∂xk ∂xk
∂u01 u02 u0k 1 ∂ n 0 0
" #
o
− + p u1 δ2k + p0 u02 δ1k
∂xk ρ ∂xk
p0 ∂u02 ∂u01
" #
+ +
ρ ∂x1 ∂x2
∂u01 ∂u02
!
−2ν
∂xk ∂xk
2
+ν∇ τ12 (10.14)

Equation 10.14 can be written in a genaral form for any i and j as given
below:
∂τij ∂τij
+ uk
∂t ∂xk
Turbulence Modeling 7

" #
∂uj ∂ui
= − τik + τjk
∂xk ∂xk
∂
+Πij − ij − Cijk + ν∇2 τij (10.15)
∂xk
in which
p0 ∂u0j ∂u0
" #
Πij = + i = pressure-strain correlation (10.16)
ρ ∂xi ∂xj
∂u0i ∂u0j
!
ij = 2ν = dissipation rate correlation (10.17)
∂xk ∂xk
and
1 1
Cijk = u0i u0j u0k + p0 u0i δjk + p0 u0j δik =
ρ ρ
third-order diffusion correlation (10.18)
Equation (10.15) is the Reynolds-stress transport equation from which the
transport equation for the turbulent kinetic energy can be obtained by a
simple contraction as shown in the next section.

10.4 Transport Equation for Turbulent

Kinetic Energy
The turbulent kinetic energy, k is defined by the following equation
τii 1
k≡ = [u02 + u02 02
2 + u3 ] (10.19)
2 2 1
A transport equation for k is obtained by substituting j = i in Equation 10.18
and then summing all the three equations for i = 1, 2 and 3
" #
∂τii ∂τii ∂ui ∂ 2
+ uk + 2τik + u0i u0i u0k + p0 u0k
∂t ∂xk ∂xk ∂xk ρ
p0 ∂u0i ∂u0 ∂u0i
− 2 + 2ν i − ν∇2 τii = 0 (10.20)
ρ ∂xi ∂xk ∂xk
Defining a scalar dissipation rate  as
∂u0i ∂u0i
!
1 1
 = ii = 2ν (10.21)
2 2 ∂xk ∂xk
Turbulence Modeling 8

substituting Equation 10.8 in Equation 10.20 and subsequently contracting

one can obtain
(1) (2) (3)
z}|{ z }| { z }| { (4)
∂k ∂k ∂ui z}|{
+ uk = − τik − 
∂t ∂xk ∂xk
!
∂ 1 0 0 0 1 0 0 2
− ui ui uk + p uk + ν∇ k} (10.22)
∂xk 2 ρ | {z
| {z } (6)
(5)

In Equation 10.22, term 1 is the rate of change of turbulent kinetic energy

k, term 2 is the convective transport of k, term 3 is the production by shear,
term 4 is the viscous dissipation and term 5 is the third-order turbulent dif-
fusive transport of k. The production term represents the transfer of kinetic
energy from the mean flow to the turbulent motion through the interaction
between the turbulent fluctuations and the mean-flow velocity gradients. The
viscous dissipation term transfers the kinetic energy to the internal energy
through viscous action.

10.5 Transport Equation for Turbulent

Dissipation Rate
The transport equation for the turbulent dissipation rate can be obtained by
taking the moment of Equation 10.10 as given below.
 
 0
∂u0 ∂  ∂ui ∂u0i 0
0 ∂ui 0 ∂ui 1 ∂p0 ∂τij 

2 0
2ν i ·  + uj + uj + uj + − ν∇ u −  =0
∂xk ∂xk 
∂t
|{z} ∂xj ∂xj ∂xj ρ ∂xi | {z }i ∂xj  
| {z } | {z } | {z } | {z } (6) | {z }
(1) (2) (3) (4) (5) (7)
(10.23)
Equation 10.23 is expanded term by term as shown below.

Term–1 :
∂u0 ∂u0i
!
∂
= 2ν i ·
∂xk ∂xk ∂t
Turbulence Modeling 9

∂u0 ∂ ∂u0i
!
= 2ν i
∂xk ∂t ∂xk
∂u0i ∂u0i
!
1 ∂ ∂ ii ∂
 
= 2ν = = (10.24)
2 ∂t ∂xk ∂xk ∂t 2 ∂t

Term–2 :
∂u0i ∂u0i
( )
∂
= 2ν · uj
∂xk ∂xk ∂xj
∂u0i ∂uj ∂u0i ∂u0i ∂u0i
( !)
∂
= 2ν · + uj
∂xk ∂xk ∂xj ∂xk ∂xk ∂xj
∂u0i ∂ ∂u0i ∂uj ∂u0i ∂u0i
!
= 2νuj + 2ν
∂xk ∂xj ∂xk ∂xk ∂xk ∂xj
∂ 1 ∂u0i ∂u0i ∂uj ∂u0i ∂u0i
" #
= 2νuj + 2ν
∂xj 2 ∂xk ∂xk ∂xk ∂xk ∂xj
∂ ∂ui ∂u0j ∂u0j
= uj + 2ν (10.25)
∂xj ∂xk ∂xk ∂xi
Note that the indices i and j are interchanged in the second term to confirm
to the format of the equation given in Speziale (1991).
Term–3 :
∂u0i ∂u0i
!
∂
= 2ν · u0j
∂xk ∂xk ∂xj
∂u0i ∂u0j ∂u0i ∂u0 ∂ ∂u0i
!
= 2ν + 2ν u0j i
∂xk ∂xk ∂xj ∂xk ∂xk ∂xj
∂u0i ∂u0k ∂u0i 1 ∂u0i ∂u0i
!
∂
= 2ν + 2νu0j
∂xm ∂xm ∂xk ∂xj 2 ∂xk ∂xk
∂u0i ∂u0i ∂u0k ∂u0i ∂u0i
!
∂ 1
= 2ν + 2ν u0j
∂xk ∂xm ∂xm ∂xj 2 ∂xk ∂xk
1 ∂u0i ∂u0i ∂u0j
!
−2ν
2 ∂xk ∂xk ∂xj
Note that index j is changed to k and k is changed to m in the first term.
Turbulence Modeling 10

Applying the continuity equation,

∂u0i ∂u0i ∂u0k ∂u0i ∂u0i

!
∂
Term–3 : = 2ν +ν u0j (10.26)
∂xk ∂xm ∂xm ∂xj ∂xk ∂xk

Term–4 :
∂u0i
!
∂ ∂ui
= 2ν · u0j
∂xk ∂xk ∂xj
∂u0i 0 ∂ ∂ui ∂u0i ∂u0j
!
∂ui
= 2ν u + 2ν
∂xk j ∂xk ∂xj ∂xj ∂xk ∂xk
∂u0i ∂ 2 ui ∂ui ∂u0i ∂u0j
= 2ν u0j + 2ν
∂xk ∂xk ∂xj ∂xj ∂xk ∂xk
∂u0 ∂ 2 ui ∂ui ∂u0i ∂u0k
= 2ν u0k i + 2ν (10.27)
∂xj ∂xj ∂xk ∂xk ∂xj ∂xj

Note that indices j and k are interchanged.

Term–5 :
∂u0i
!
∂ 1 ∂p0
= 2ν ·
∂xk ∂xk ρ ∂xi
∂u0i ∂
!
1 ∂p0
= 2ν ·
∂xk ∂xi ρ ∂xk
1 ∂p0 ∂u0i ∂u0i
! !
∂ ∂p0 ∂
= 2ν − 2ν
∂xi ρ ∂xk ∂xk ∂xk ∂xi ∂xk
1 ∂p0 ∂u0i ∂u0i
! !
∂ ∂p0 ∂
= 2ν − 2ν (10.28)
∂xi ρ ∂xk ∂xk ∂xk ∂xk ∂xi

Substituting the continuity equation and changing the index i to k and k to

m in the first term, we get

1 ∂p0 ∂u0k
!
∂
Term–5 : = 2ν (10.29)
∂xk ρ ∂xm ∂xm
Turbulence Modeling 11

∂u0i ∂
Term–6 : = 2ν · (−ν∇2 u0i )
∂xk ∂xk
Noting that
∂u0i
!
∂
(∇2 u0i ) = ∇2
∂xk ∂xk
∂u0i ∂u0i
!
2
Term–6 : = −2ν · ∇2 (10.30)
∂xk ∂xk
The following identity holds:

α2
! !
2 2∂α ∂α
∇ = α∇ α + (10.31)
2 ∂xm ∂xm

in which α is any variable. Substitution of Equation 10.31 in Equation 10.30

and subsequent simplification yields:

Term–6 :
1 ∂u0i ∂u0i ∂ 2 u0i ∂ 2 u0i
" #
2 2
= −2ν ∇ + 2ν 2
2 ∂xk ∂xk ∂xk ∂xm ∂xk ∂xm
∂ 2 u0i ∂ 2 u0i
= −ν∇2  + 2ν 2 (10.32)
∂xk ∂xm ∂xk ∂xm

Term–7 :
∂u0
!
∂ ∂τij
= −2ν i ·
∂xk ∂xk ∂xj
!
∂ ∂τij ∂ 0
= −2ν ui =0 (10.33)
∂xk ∂xj ∂xk

Adding Equations 10.24, 10.25, 10.26, 10.27, 10.29, 10.32 and 10.33 and then
rearranging the various terms we get

∂ ∂
+ ui =
∂t ∂xi
Turbulence Modeling 12


∂u0 ∂u0 ∂ui
−2ν ∂xji ∂xkj ∂x



k


∂u0 ∂u0 ∂u0


−2ν ∂xki ∂xmi ∂xmk


0 P
2u
0 ∂ui
−2ν ∂x∂k ∂x
i
j
u k ∂x



j

0

0
∂ui ∂ui ∂uk

−2ν ∂xk ∂xj ∂xj



  
0 0
∂ 0 ∂ui ∂ui
−ν ∂xk uk ∂xm ∂xm  


0
 D
∂p0 ∂uk
−2ν ∂x∂ k ρ1 ∂x


m ∂xm


∂ 2 u0i ∂ 2 u0i
!)
−2ν 2 Φ
∂xk ∂xm ∂xk ∂xm
+ν∇2  (10.34)
In Equation 10.34, the first four terms on the right hand side (denoted by
P ) represent the production of dissipation, the next two terms (denoted by
D ) represent the turbulent diffusion of dissipation and the seventh term
(denoted by Φ ) represents the turbulent destruction of dissipation. It can
be seen that Equation 10.34 for turbulent dissipation rate is analogous to
Equation 10.22 for the turbulent kinetic energy.

10.6 The Eddy Viscosity Concept

One of the most widely used concepts in the present day turbulence models
for practical engineering applications is the “eddy viscosity” concept. In
analogy to the viscous stresses in laminar flows, Boussinesq (1877) proposed
that the turbulent stresses are proportional to the mean-velocity gradients.
This concept can be generalized to express the Reynolds-stress tensor (for
incompressible flows) in the following form:
!
∂ui ∂uj 2
τij = −νt + + k δij (10.35)
∂xj ∂xi 3
in which νt is the turbulent or eddy viscosity. The second term on the right
hand side of Equation 10.35 does not appear in the original postulation. It is
added later on so that the equation is applicable to normal turbulent stresses
also. Inclusion of this term ensures that the sum of normal stresses is equal
to 2k.
Turbulence Modeling 13

It should be noted here that while the molecular viscosity, ν is a fluid

property, the turbulent viscosity νt depends on the flow, i.e. the state of
turbulence. νt may be different for different points in the flow and may vary
from flow to flow. In analogy to kinetic theory of gases (ν α average velocity
and mean free path of molecules), the turbulent viscosity may be expressed
as
νt α v0 l0 or νt α l02 /τ0 (10.36)
in which v0 is the velocity scale, l0 is the length scale and τ0 is the time scale
of turbulence.
The eddy viscosity concept inherently assumes that the Reynolds-stress
tensor can be characterized by a single length and time scales. However, it
is well known that the turbulent flows consist of a wide range of length and
time scales and this introduces a limitation to the applicability of the eddy
viscosity concept. Eddy viscosity concept is also criticised because of the
following facts:

1. Turbulent eddys are not rigid bodies like molecules and do not retain
their identity. The interaction among eddys is not elastic as is the case
for molecular interactions in kinetic theory of gases.

2. The large eddys are responsible for extracting the energy from the
mean motion and feeding it into the turbulent motion. The rate at
which the mean-flow energy is fed into the turbulent motion depends
on large scale motion although the final dissipation is affected by the
smallest eddys. Therefore, it is mainly the large scale turbulent motion
which affects the correlations u0i u0j and the length and the time scales l0
and τ0 in Equation 10.36 should correspond to the large-scale turbulent
motion. Even if one considers the effect of large eddys only, there is
not one single large eddy. Therefore, l0 and τ0 should correspond to
some kind of an effective large eddy called the “characteristic eddy”
of the flow, which determines the energy transfer from mean-flow to
turbulent motion. Kinetic gas theory for determining the molecular
viscosity requires that the mean free path is very small compared to the
flow domain. In a similar way, (Speziale 1991), for the eddy viscosity
concept to be valid, the length and the time scales of the “characteristic
eddys” should be small compared to the length and the time scales of
mean motion. However, for many turbulent flows, the length scale
Turbulence Modeling 14

of characterisitic eddys is not small compared with the flow domain

(example - boundary dominated flows) and the time scale ratio τ0 /T0
(T0 = time scale of the mean flow) can be of 0(1).

3. The eddy viscosity, νt as introduced in Equation 10.35 is a scalar quan-

tity. This may not be true even for simple turbulent shear flows. Also,
the eddy viscosity fails to distinguish between plane shear, plane strain
and rotating plane shear flows (Speziale 1991).

4. The eddy viscosity concept has been successful mainly for the pre-
diction of two-dimensional shear flows. The only Reynolds-stress of
importance in this case is u0 v 0 . The eddy viscosity is the coefficient of
proportionality between this stress and rate of shear du/dy. Good cor-
relations for νt are available in this case (Ferziger 1987). However, the
eddy viscosity concept may give erroneous results even for simple shear
flows such as wall jets and channel flows with varying wall roughness
(Rodi 1980).

The above discussion is meant to bring out the fact that the eddy vis-
cosity concept is not correct in general and should be used cautiously in
complex flows. Inspite of all the limitations, this concept is very wiidely used
in many popular turbulence models in use today. Eddy viscosity models are
not closed unless the length and time scales l0 and τ0 are specified. It is
the specification of these scales in a flow field depending on space and time
coordinates which constitutes the core of the majority of the popular turbu-
lence models for engineering applications. These models are discussed in the
following sections.

10.7 Zero-Equation Models

Zero-equation models are the simplest among the turbulence models using
the eddy viscosity concept. They do not employ the transport equations for
turbulent quantities like the turbulent kinetic energy. In these models, the
eddy viscosity, νt in Equation 10.35 is specified algebraically using empirical
information from experiments and past experience. Two of the commonly
used zero-equation models are presented below.
Turbulence Modeling 15

10.7.1 Constant Eddy Viscosity Model

In these models, νt is specified as a constant and its value is chosen depend-
ing on some experimental information and or a trial and error procedure.
The assumption of constant νt is a highly idealized one and the use of these
models results in grossly erroneous prediction of velocity fields, especially
near the boundaries. The main advantage of using constant eddy viscosity
models seems to be their simplicity with regards to numerical computations.
Also, they seem to yield saatisfactory results when applied to determine
temperature or concentration distributions in depth-averaged calculations
for free-surface flows. Even in these cases, the results are satisfactory only
for far field calculations. It should also be noted here that the constant eddy
viscosity models seem to give satisfactory results in some situations only be-
cause turbulent terms in the momentum equations are negligible, compared
to the convective and pressure gradient terms. The selection of νt value in
these modles, many times, is masked by factors such as (i) numerical diffu-
sion introduced by large grids and (ii) dispersive effects in depth-averaged
calculations that arise due to non-uniformity of the velocity in the vertical
direction. In some cases, the constant eddy viscosity model is merely used
to improve the numerical stability of a computational method. This should
not be confused with turbulence modeling. Because of the above reasons,
the constant eddy viscosity models have almost become obsolete these days
except for an occasional application in hydraulic engineering involving large
water bodies.

10.7.2 Mixing Length Models

Prandtl (1925) was the first person to propose a proper turbulence model
based on Equation 10.36. For a plane shear flow, he defined the mixing
length, lm as the distance travelled by a fluid lump in the transverse direc-
tion before the mean velocity changes by an amount equal to the transverse
fluctuation velocity. Prandtl postulated that

1. the turbulent length scale, l0 is equal to the mixing length, lm and

2. the velocity scale, v0 is equal to the mean velocity gradient times the
mixing length.
Turbulence Modeling 16

Figure 10.1: Mixing length in a turbulent boundary layer (after Ferziger,

1987).

When these are substituted into Equation 10.36 and the proportionality con-
stant is taken equal to one, we get

2 du
νt = lm (10.37)
dy

The turbulence model is closed once the mixing length, lm is specified by

simple empirical formulations. Experiments in shear flows indicate that far
from solid walls the largest turbulent length scales are some fraction of the
width of the shear layer. On the other hand, the size of the eddys near a solid
boundary is bounded by the distance to the surface. Therefore, the length
scale, lm in a turbulent boundary layer has a typical variation as shown in
Figure 10.1 (Ferziger 1987). The mixing length, lm is taken constant for free
shear flows.
The following empirical equations are generally used for estimating lm in
shear flows.
Turbulence Modeling 17

1. Wall boundary layers (Patankar and Spalding 1970)

lm = Ky (y/δ < λ/K) (10.38)

lm = λδ (y/δ > λ/K) (10.39)

in which δ = boundary layer thickness, y=distance from the wall, λ =
0.09 and K = Von-Karman Constant = 0.435.

2. Developed duct / pipe flows (Schlichting 1969)

2 4
lm y y
 
= 0.14 − 0.08 1 − − 0.06 1 − (10.40)
R R R
in which R = radius of the pipe or the half width of the duct.

3. Free shear flows

lm = C1 δ (10.41)
in which δ is the free shear layer width. C1 is a proportionality constant
which depends on the type of flow considered. For example, C1 = 0.075
for a round jet in stagnation surroundings and it is equal to 0.16 for a
plane wake.
For three-dimensional flows, the du dy
in Equation 10.37 is replaced by the
r.m.s. strain rate of the mean field as shown below.
 1/2
νt = `2m 2S ij S ij (10.42)

where !
1 ∂ui ∂uj
S ij = + (10.43)
2 ∂xj ∂xi
The main problems with the mixing length models are
1. they cannot switch from one type of region to another (e.g. from a
boundary layer to a free shear layer) within a single flow

2. they do not consider the process of convective or diffusive transport of

turbulence, i.e. the history effects on turbulence

3. they require adhoc prescription of lm in each problem that is considered

and
Turbulence Modeling 18

4. they do not compute any turbulent quantities, even the turbulent ki-
netic energy which is the most crucial measure of turbulence.
Although the assumptions behind the mixing-length models are open to ques-
tion, they have been successful in simulating a variety of simple flows. The
celebrated “law of the wall” for turbulent pipe and channel flows is one exam-
ple of the successful application of the mixing length theory. This law is used
to this day while formulating the boundary conditions for more sophisticated
models such as k- model for turbulence.

10.8 One-Equation Models

One-equation models overcome some of the limitations of the zero-equation
models by accounting for the nonlocal and history effects on turbulence in a
limited way. They compute the turbulent kinetic energy by solving the dif-
ferential transport equation (Equation 10.22).
√ In these models, the velocity
scale, v0 of the turbulence is taken as k since the turbulent energy, k is
mainly contained in the large-scale fluctuations. It should also be noted that
k is a direct measure of the intensity of turbulence fluctuations. With the
above postulation, the eddy viscosity can be written as (Kolmogorov 1942
and Prandtl 1945): √
νt = k l0 (10.44)
The turbulent kinetic energy, k can be determined by solving Equation 10.22.
However, this is not easy because this equation contains the higher-order cor-
relations for turbulent transport and dissipation. Therefore, model assump-
tions are made to obtain a closed set of equations. While postulating the
eddy-viscosity models, it has already been assumed that there is a clear-cut
separation of scales. Consistent with this assumption, the turbulent diffusion
flux of k is modeled by a gradient term.
" #
1 0 0 0 1 −νt ∂k
ui ui uk + p0 u0k = (10.45)
2 ρ σk ∂xk
in which σk is a dimensionless constant. As mentioned earlier, the rate of
dissipation,  is governed by the large-scale turbulent motion and therefore,
it is characterized by k and l0 i.e.
 = f (k, l0 ) (10.46)
Turbulence Modeling 19

Using the dimensional analysis, Equation 10.46 can be written as

k 3/2
 = CD (10.47)
l0
in which CD is an empirical constant. For simple flows, the suggested values
of σk and CD are 1.0 and 0.08 respectively. Substitution of Equations 10.35,
10.45 and 10.47 in Equation 10.22 leads to
k 3/2
" # " #
∂k ∂k ∂ui ∂uk ∂ui ∂ νt ∂k
+ uk = νt + + − CD + ν ∇2 k
∂t ∂xk ∂xk ∂xi ∂xk ∂xk σk ∂xk l0
(10.48)
Equations 10.6, 10.7, 10.35, 10.44 and 10.47 form a closed system of equations
for the determination of ui , p and k once the turbulence length scale, l0 is
specified. In One-Equation models, l0 is specified empirically. In simple shear
flows, l0 is determined in the same way the mixing length, lm is determined.
Bernard et al. (1974) and Gowain and Pritchett (1970) have proposed em-
pirical formulae for estimating l0 in general flows. The one-equation model
developed by Bernard et al. (1974) has been tested for wall boundary layers,
channel flows, free jets and wall jets. The model developed by Gowain and
Pritchett has been tested for two-dimensional channel flows, axysymmetric
jets and three-dimensional thermal plumes. However, the length scale de-
termination by these models is computationally very intensive. Also, these
models are not tested for more complex situations. For simple flows, the
mixing length models give as good results as one-equation models with much
lesser computational effort involved. Inspite of the above deficiencies, the
One-Equation Models have been popular in the field of aerodynamics (John-
son and King, 1984) owing to their simplicity in structure.

10.9 Two-Equation Models

Two-equation models are among the most popular turbulence models for
scientific and engineering calculations. In these models, two separate trans-
port equations are solved to determine the length and the velocity scales in
Equation 10.36 for the eddy-viscosity. By doing so, the necessity to specify
the length scale, l0 in an adhoc manner is avoided. From the above point of
view, the two-equation models were probably the first complete turbulence
models.
Turbulence Modeling 20

The transport equation for the velocity scale in the two-equation models is
usually the k-equation (Equation 10.48) as given for the One-equation mod-
els. However, the transport equation for the length scale need not necessarily
have the length scale itself as the dependent variable. Any combination of k
and l0 may be used as the dependent variable in the second transport equa-
tion. Kolmogorov√(1942) in his original work proposed a transport equation
for the frequency k/l0 . Rotta (1951) proposed an equation for the variable
kl0 while Spalding (1971) proposed a transport equation for the vorticity,
k/l02 . The celebrated k- model uses a transport equation for the turbulent
dissipaation, . The transport equation for any of the above quantities can
be expressed in a general form as shown below (Rodi 1980).
√ !
∂z ∂z ∂ kl0 ∂z
+ uk = {diffusion}
∂t ∂xk ∂xk σz ∂xk
z
+ Cz1 P {Production}
k√
k
− Cz2 z {diffusion}
l0
+ S (10.49)

in which, z = k m l0n , P =production of turbulent kinetic energy and S= a

secondary source term which depends on the choice of z. σz , Cz1 and Cz2
are empirical constants. Past experience indicates that k- model works
better than the other two-equation models based on Equation 10.49 because
(i) the -equation does not require the source term, S and (ii) the gradient
approximation which is used in the diffusion term seems to work better when
z = . It should also be noted here that Equation 10.49 has been originally
conceived heuristically. However, the -equation as used in the k- model can
be obtained from the complete -equation (Equation 10.34) using modeling
approximations for the higher-order correlations. A brief description of the
k- model is presented below.

k- Model
In the k- model, the length and the time scales are built up from the tur-
bulent kinergy, k and the turbulent dissipation rate,  using the dimensional
Turbulence Modeling 21

arguments √
k k k
l0 α and τ0 α (10.50)
 
Substitution of Equation 10.50 in Equation 10.36 yields

k2
νt = Cµ (10.51)

in which, Cµ is an empirical constant. The turbulent kinetic energy,k and the
dissipation rate,  are obtained by solving the modeled transport equations.
The transport equation for k in the k- model requires modeling of only the
turbulent transport term using the gradient approximation. This is given as
" #
∂k ∂k ∂ui ∂uk ∂ui
+ uk = νt +
∂t ∂xk ∂xk ∂xi ∂xk
!
∂ νt ∂k
+ −  + ν∇2 k (10.52)
∂xk σk ∂xk

The transport equation for  is a modeled version of the transport equation

given earlier. Equation 10.34 is rewritten as
∂ ∂
+ ui = ν ∇2  + P + D − φ (10.53)
∂t ∂xi
As mentioned earlier, P , D and φ represent the production of dissipation,
the turbulent diffusion of dissipation and the turbulent destruction of dissipa-
tion respectively. The diffusion term, D can be modeled using the gradient
approximation as is done in the case of modeled k-equation.
!
∂ νt ∂
D = (10.54)
∂xk σ ∂xk

in which σ is an empirical constant.

Analogous to modeling of  in the k-equation for the one-equation models,
the destruction of dissipation is assumed to depend only on the length and
the time scales. This can be written in a functional form as

φ = fφ (l0 , τ0 ) (10.55)
Turbulence Modeling 22

or using Equation 10.50,

φ = fφ (k, ) (10.56)
Dimensional analysis for φ requires that

φ α /τ0

or
φ α 2 /k
or
2
φ = C2 (10.57)
k
in which, C2 is an empirical constant.
The production of dissipation is assumed to be proportional to the pro-
duction of the turbulent kinetic energy, the length scale and the time scale
of turbulence. A simple dimensional analysis for P along with the above
assumption leads to

P = C1 P (10.58)
k
in which, C1 is a dimensionless constant. P is the production of the turbulent
kinetic energy and is given by
∂ui
P = −τik (10.59)
∂xk
(Note that P has the dimensions of /sec while P has the dimensions of
k/sec).
Substitution of Equation 10.59 in Equation 10.58 yields
 ∂ui
P = −C1 τik (10.60)
k ∂xk
Speziale (1991) showed that Equation 10.60 is strictly valid only when

1. there is a clear separation of time scales of turbulent fluctuations and

the mean velocity field and

2. the level of anisotropy is not very much.

Turbulence Modeling 23

Substitution of Equations 10.54, 10.57, 10.60 and 10.35 in Equation 10.43

leads to
!
∂ ∂ ∂ νt ∂
+ ui = ν ∇2  +
∂t ∂xi ∂xk σ ∂xk
2
!
 ∂ui ∂uk ∂ui
+ C1 νt + − C2 (10.61)
k ∂xk ∂xi ∂xk k

Equation 10.61 is the modeled version of the transport equation for the tur-
bulent dissipation rate. To summarize, the standard k- model comprises of
Equations 10.35, 10.51, 10.52 and 10.61. These equations along with Equa-
tions 10.6 and 10.7 form a closed system for determining ui , p, k and  in an
incompressible turbulent flow.
The standard k- model has five empirical constants Cµ , σk , σ , C1 and
C2 in its formulation. C2 is determined using the experiments on the decay
of k behind a grid and a value of C2 = 1.92 is suggested. The experiments
on the local-quilibrium-shear-layers suggested that Cµ ≈ 0.09. σk , σ and C1
are fixed using the computer optimization and their suggested values are 1.0,
1.3 and 1.44 respectively. It should be noted that the above values for the five
constants are not universal and the k- model may require some amount of
fine tuning in order to obtain correct results. Rodi (1980) lists several cases
where the k- model has been successfully applied using the above suggested
values of the five empirical constants.
Despite the enormous success of k- model in engineering applications,
caution is needed. This is especially so when the engineering applications
demand high accuracy. Some major problems encountered in the application
of the k- model are listed below.

1. k- model uses the Boussinesq eddy-viscosity concept (Equation 10.35)

which in turn assumes that the turbulent viscosity is isotropic. This
leads to equal normal Reynolds stresses in many flows. Therefore, it
can not reproduce secondary flows which arise due to unequal normal
Reynolds stresses (e.g. secondary flow in a non-circular duct).

2. As mentioned earlier, Boussinesq eddy-viscosity relationship cannot

simulate the effect on turbulence of the changes in the mean-flow di-
rection across the shear layers.
Turbulence Modeling 24

3. The production of turbulence in highly strained flows is over predicted

by k- model.
4. The k- model requires modifications when applied to flows with highly
curved stream lines (e.g. swirling flow).
5. The values of the empirical constants need modification when the k-
model is applied to axysymmetric jet flows and weak shear layer (Rodi
1980). It also needs modifications when applied to flows approaching
laminar conditions. It cannot be used near a solid wall; it requires wall
function implementation.
6. Speziale (1991) shows that the k- model does not strictly satisfy the
realizability constraint (τij > 0 for all i and j).
Most of the deficiencies of the two-equation models stem from the assump-
tion that there is a clear-cut separation of scales at the level of Reynolds-
stress tensor. Non-linear k- models (Speziale 1991) have been developed
in the past which account for the anisotropic turbulent viscosity and which
satisfy the realizability constraint. Secondary motions in non-circular ducts
could be simulated using these models. However, the current trend is to de-
velop second-order closure models based on the modeling approximations for
πij , ij and Cijk in the Reynolds-stress transport equation (Equation 10.15).

10.10 Closure
This chapter introduced the turbulence modeling based on the eddy-viscosity
concept. Constant eddy-viscosity, mixing-length, one-equation and two-
equation models have been described in detail. Limitations of all these mod-
els have been discussed. It has been shown how the standard k- model can
be developed using model approximations for the higher-order correlations in
the exact transport equation for turbulent kinetic energy, k and the turbulent
dissipation, .

References
Bernard, J. P., Denoyelle, J. F., Huffenus, J. P. and Peube, J. L., 1974, La
Simulation de la turbulence pour le calcul des e0 coulements de convec-
Turbulence Modeling 25

tion mixte0 , Bericht ALSTHOM Technique des Fluides, Grenoble.

Boussinesq, J., 1877, The’orie de le’coulement tourbilaant, Mem. Presente’s

paar divers savants Acad. Sci. Inst. Fr., Vol.23, pp. 46-50.

Ferziger, J. H., 1987, Simulation of incompressible turbulent flows, Journal

of Computational Physics, Vol.69, No. 1, pp. 1-48.

Gowain, T. H. and Pritchett, J. W., 1970, A unified heuristic model of fluid

turbulence, Journal of Computational Physics, Vol.5, pp. 383-405.

Johnson, D. A. and King, L. S., 1984, A new turbulence closure model

for boundary layer flows with strong adverse pressure gradients and
separation, AIAA paper No. 84-0175.

Kolmogorov, A.N., 1942, The equations of turbulent motion in an incom-

pressible fluid, Izv. Acad. Sci., USSR Phys., Vol.6, pp. 56-58. (Impe-
rial College Report No. N/6-1968).

Patankar, S. V. and Spalding, D. B., 1970, Heat and Mass Transfer in

Boundary Layers, Intertext, London.

Prandtl, L., 1925, Uber die ausgebildete turbulenz, ZAMM, Vol.5, pp. 136-
139.

Prandtl, L., 1945, Uber ein neuses formal-system fur die ausgebildete tur-
bulenz, Nachr. Akad. Wiss., Gottingen, Math-Phys, Kl. 1945, pp.
6-19.

Reynolds, O., 1895, On the dynamical theory of incompressible viscous

fluids and the determination of the criterion, Philos. Trans. of R. Soc.
London, Ser A-186, pp. 123-164.

Rodi, W., 1984, Turbulence Models and Their Application in Hydraulics—A

State of the Art Review, IAHR, Delft, The Netherlands, 2nd edition.

Rotta, J. C., 1951, Statistische Theorie Nichtomogener Turbulenz, Zeitschrift

f. and Physik, Bd. 192, pp 547-572, and Bd. 131, pp 51-57. (Im-
perial College, Department of Mechanical Engineering, Report No.
TWF/TN/38, 39-1968.
Turbulence Modeling 26

Schlichting, H., 1969, Boundary Layer Theory, McGraw-Hill, New York.

Spalding, D. B., 1971, The k − ω Model of turbulence, Imperial College,

Department of Mechanical Engineering, Report No. TM/TN/A/16.

Speziale, C. G., 1991, Analytical Methods for the development of Reynolds-

stress closures in turbulence, Annual Review of Fluid Mechanics, Vol.23,
pp. 107-157.