TOPIC 3

TURBULENT FLOW

TURBULENCE

“We know what a turbulent flow is, when we see it!

It is characterized by disorder, vorticity, and mixing.”
Moshe Matalon
University of Illinois at Urbana-Champaign

1
 Objectives (part 1)

1. Understand physical and mathematical description

of turbulent flow
2. Examine turbulent flow over a flat plate

Ref: “A first course in Turbulence”, Tennekes and Lumley, 2001, MIT Press 2
“Turbulent Flows”, Pope
 Properties of Turbulent flow
Always occur at ‘high’ Reynolds numbers.
The instabilities that lead to randomness are related to the interaction
of the viscous terms and the nonlinear inertia terms in the Navier-
Stokes Equations.
Disturbances in the flow, due to instabilities, may develop large
velocity gradients. At 𝑅𝑒 ≪ 1, viscous dissipation prevents the
development of these large velocity gradients. But at 𝑅𝑒 ≫ 1 viscous
dissipation is unable to smear out the large developing gradients.

1. Chaotic
One characteristic of turbulent flows is their irregularity or randomness.
A full deterministic approach is very difficult. Turbulent flows are usually
described statistically. Turbulent flows are always chaotic. But not all
chaotic flows are turbulent. The flow consists of a spectrum of different
3
scales.
 Properties of Turbulent flow
2. Diffusivity
The diffusivity of turbulence causes rapid mixing and increased rates
of momentum, heat, and mass transfer. A flow that looks random
but does not exhibit the spreading of velocity fluctuations through
the surrounding fluid is not turbulent. If a flow is chaotic, but not
diffusive, it is not turbulent.

3. Three-Dimensional
Turbulent flow is always three-dimensional and unsteady.
However, when the equations are time-averaged, we can treat the
flow as two-dimensional (if the geometry is two-dimensional).

4
 Properties of Turbulent flow

4. Dissipative
Kinetic energy gets converted into heat due to viscous shear
stresses. Turbulent flows die out quickly when no energy is
supplied. Random motions that have insignificant viscous losses,
such as random sound waves, are not turbulent.

5. Rotational
Turbulent flows have non-zero vorticity. Mechanisms such as the
stretching of three-dimensional vortices play a key role in turbulence.
Turbulence has high levels of ‘vorticity’ / ‘eddies’.
5
 Properties of Turbulent flow

Continuum
– Scale of fluctuations are larger than molecular scales.
– Navier Stokes equations apply and can predict turbulence,
ex, Direct Numerical Simulation (DNS) of turbulence.

Not a feature of the fluid but of the flow

– If Reynolds number is large enough then the major
characteristics of turbulent flows are not controlled by the
viscosity (molecular properties) of the fluid.

Can’t predict detailed motion but it is possible to

predict mean quantities (more or less).
Turbulence can be thought of as a spatially varying mean
flow with super imposed, 3D, random fluctuations.
6
7
Taylor Couette flow between concentric
cylinders

Stable secondary Flow Wavy Taylor vortices

Turbulent Taylor
vortices

8
Coherent structures in a turbulent channel flow

9
 Statistical Description of Turbulent Flows
Turbulent eddies create fluctuations in velocity.
The aim is to describe the fluctuating velocity and scalar field in terms of their
statistical distributions.

T is large compared to scale of fluctuations

Stationary: If independent of t.

10
 Statistical Description of Turbulent Flows

Can now define the velocity as a sum of the mean velocity and the time
varying, fluctuating velocity:
u = u +u'
Flow Spatially Time
varying varying

11
 Statistical Description of Turbulent Flows
Spatial Average
x +
u ( x) = 
1
 x ud 

Isotropic turbulence

Turbulence Intensity

12
Turbulent stresses

• In turbulent flow, the experiments showed that the shear

stress is much larger, due to the turbulent fluctuations
• Turbulent stress acts as an additional source of diffusion
of the momentum

tturb – friction between fluctuating fluid particles and the fluid

body

tlam – friction between layers in the flow direction

Next step: find the expression of the turbulent shear stress

using Navier-Stokes equations 13
Equations of motion for turbulent flow
Remember the Navier-Stokes equations for incompressible flow
Continuity:
u v w
+ + =0
x y z
Momentum:
DV  V 
 =  + V V  =  g − P +  2V
Dt  t 

Note: we can save some time up ahead if we re-write the above into a slightly
different form:
u    p
+ (  uu ) + (  uv ) + (  uw ) = +  g x − +  2u
t x y z x

14
Reynolds averaged Navier Stokes equations
(RANS)
We want to derive equations for the mean velocity field. So, we
substitute the following into the above equations.

u = u +u' 𝑃 = 𝑃ത + 𝑃′
v = v + v' Remember that the mean
w = w + w' of a mean is a mean and
u’ is a zero mean random
And average the Navier-Stokes equations. variable
First the continuity equation:
u =u
 ( u + u ')  ( v + v ')  ( w + w ')
+ + =0 u' = 0
x y z
(
 u + u' ) +  ( v + v ') +  ( w + w ') = 0
x y z
15
 Reynolds averaged Navier Stokes equations
(RANS)

This results in the following equation of mean continuity:

u v w
+ + =0
x y z

Similarly for momentum…

X-component:
Sub in and average the equation:
 ( u + u ')    p
t
+
x
(  ( u + u ')( u + u ') ) + (  ( u + u ') (v + v ') ) + (  ( u + u ' ) ( w + w ' ) = +  g x − +  2 ( u + u ' )
y z x

Note that:

uu ' = 0
, etc.
16
Reynolds averaged Navier Stokes equations
(RANS)
• The x-component averaged becomes:
    p
 u +  uu +  uv +  uw =  g x − +  2u
t x y z x
   
−  u 'u ' + u 'v ' + u ' w '
 x y z 
The mean momentum equation has three additional terms that are
unknown
These correlations are called Reynolds Stress terms: Total of 6 unknown terms.
This means that we have a
t uu =  u ' u ' t uv =  u ' v ' t uw =  u ' w ' Reynolds stress tensor. In
turbulent flows these terms
The other 2 momentum equations generate are never negligible and our
similar terms source of difficulty in
analyzing mean fluid flows.
−  v '2 , −  w '2 , −  v ' w ' 17
• Reynolds stress tensor

 u 'u ' u ' v ' u ' w ' 

 
t turbulent =   v 'u ' v ' v ' v ' w ' 
 w ' u ' w ' v ' w ' w '
 

• Viscous stress tensor

 u  u v   u w  
 2
x  y + x   z + x  
    
  v u  v  v w  
t laminar =   +  2  +  
  x y  y  z y  
 w u  w v  w 
 + 

 y + z  2
 x z    z  18
Reynolds averaged Navier Stokes equations
(RANS)
We can now write the Reynolds Averaged Navier Stokes (RANS) Equations as:


DV
Dt
(
=  g − P +   V −   τ (t ) )
Note : there is a combination of the laminar u stresses and turbulent
‘Reynolds’ stresses.

In order to calculate the mean velocity profile using some kind of computation
algorithm, we need to know these terms. However, they are a function of the
details of the random fluctuations… ie, we don’t know them!

Now we have more unknowns than we have equations. Need to develop an

equation for the Reynolds stresses. These models are referred to as ‘closure’, i.e.,
they close the set of required equations.

How do we estimate/calculate what these stresses are? We use semi-empirical,

experimental based estimates !! 19
Modeling turbulent viscosity
1.Boussinesq’s ‘Eddy Viscosity Hypothesis’

u
t yx = t t - turbulent viscosity
( Turbulent )
y
Random eddy motion of groups of particles resembles the random
motion of molecules in a gas , colliding with each other after travelling
a certain distance and exchanging momentum in the process.

Momentum transport by eddies in turbulent flow is analogous to the

molecular momentum diffusion.
The turbulent viscosity is almost always significantly larger than the
molecular viscosity.

The turbulent viscosity is usually strongly dependent on the fluid flow.

Therefore, we need to be able to estimate the viscosity.
20
How we model turbulent viscosity ?
Prandtl mixing length theory

From dimensional analysis we know that ‘eddy’ viscosity should scale

with the scale of the velocity fluctuations and the length over which
they move by the following:
t * *
=l u

l* is a length scale and u* is a velocity fluctuations scale.

By considering simple shear flow Prandtl used the Kinetic Theory of

gas to derive the following:
u
u  l
*

y
Mixing length is related to the average size of eddies that are primarily
responsible for mixing.
21
Prandtl mixing length theory

2
 u 
t yx (Turbulent ) = l  
2 Near wall l  = ky
 y 

This seems to work pretty well for the special case of unbounded simple shear.
However, it doesn’t work at all for the near wall conditions.
22
Flow near a surface (flat plate)

• There are 3 flow regimes near a flat plate:

• Wall layer: Viscous shear dominates the flow

• Outer Layer: Turbulent shear dominates
• Overlap Layer: Both types are important

23
Wall Layer
Dimensional analysis:
What would the mean velocity be a function of? u = f (t w ,  ,  , y )
Non-dimensional velocity:

1/ 2
+ u tw 
u = * Where: u = 
*

u 
And non dimensional position is

+ y
y = * Where y = *
*

y u
Very close to wall ( y+  5) experiment says that

+ +
u =y LAW OF THE WALL
24
Outer Layer

Dimensional analysis suggests that

U −u  y
= f  VELOCITY-DEFECT LAW
 
*
u

That is, the velocity defect is a function of the height of the

boundary layer, , and the flow velocity, U, at the outer layer …

25
Overlap region

Near the wall we have, from ‘Mixing length’ model and eddy
viscosity:

2
 u 
t yx (Turbulent ) =  y    t w
2 2

 y 
If we integrate, yields to get the ‘LOG-LAW’:

u = ln ( y + ) + B
+1
k

Empirically/experimentally … k=0.41 and B=5.0

26
 Summary - Turbulence

u
t yx = t
( Turbulent )
y
From analogy between eddies and molecules in a low-density gas
t * * u
=l u u  l
*

 2
y
 u 
t yx (Turbulent ) = l   2

 y 
Flow over a plate
1/ 2
+ + tw  
u =y u = 
*
y = *
*


Wall Law
u

u = ln ( y + ) + B
+ 1
Log Law
k

U −u  y
Defect Law = f  27
 
*
u
CFD Turbulent Models
• Two Equations Models: k-e model
The turbulent kinetic energy

k= 1
2 ( u ' u ' + v ' v ' + w ' w ') ‘Eddy viscosity’ based

The equation for the dissipation rate, e (empirical)

De  t
=  
 Pe e2 t k2
 + Ce 1 − Ce 1 = C
Dt   e  k k  e

CFD codes would then be used to solve the following equations:

- 1 Continuity
- 3 Momentum equations
- 1 Kinetic energy equation
-1 Dissipation rate equation
• Other similar ‘Eddy viscosity’ based models include k-w (vorticity) 28
model
Other models

• Reynolds Stress Models

You can also model each of the Reynolds stresses. This is used when you have
anisotropic turbulence. Model equations are generated for each of the
correlations and for the dissipation. This means we don’t need the turbulent
viscosity hypothesis.
• Large Eddy Simulation
You assume an eddy viscosity for just the smallest scales of turbulence, which
are also the most isotropic and easiest to model. Then you resolve the largest
eddies in a time varying manner.
Still too computationally intensive for practical application.
Direct Numerical Simulation
• Simply solve the time varying Navier-stokes equations and resolve all the
turbulence. Not practical for even the simplest problems.

29
 Application
Two infinite plates a distance h apart are parallel to the xz plane with the upper plate
moving at speed V. There is a fluid of viscosity  and constant pressure between the
plates. Neglecting gravity and assuming incompressible turbulent flow u(y) between
the plates, use the logarithmic law and appropriate boundary conditions to derive a
formula for dimensionless wall shear stress versus dimensionless plate velocity.
Sketch a typical shape of the profile u(y).

30
31
32