You are on page 1of 7

Prandtl's Mixing Length Hypothesis

The general form of the Boussineq eddy viscosity model is given as



k
3
2
x
U
x
U
u u
ij
i
j
j
i
T j i

,
_

,
m m
u u
2
1
k , (1)

where
T
is the eddy viscosity. For thin shear layer, the relevant component of (1) may
be restated as


y
U
v u
T

. (2)

Prandtl assumed that
m T
~ l u , where u is a turbulent velocity scale and
m
l is referred
to as the mixing length. Furthermore, Prnadtl postulated that


y
U
~ u
m

l , (3)

and hence


y
U
2
m T

l ,
y
U
y
U
v u
2
m

l . (4)

The mixing length
m
l depends on the nature of the flow and, in general, is space
dependent.

For free shear flows,
m
l is proportional to the half-width of the shear layer l . For
different flows, the mixing lengths are given as

l l 09 . 0
m
, for plane jet,

l l 075 . 0
m
, for circular jet, (5)

l l 07 . 0
m
, for mixing layer.

For boundary layer flows, several different forms were suggested. For instance, Escudier
(1966, Imperial College Report) assumed

Variation of mean velocity and mixing length
in a turbulent boundary layer

'

0
0
0
m
y
y
y
l . (6)

where 41 . 0 and 09 . 0
0
.

1 U/U
o




/
m
l



/ y

/
o
1





For pipe flows, Nikuradse suggested the following expression:



4
0
2
0 0
m
r
y
1 06 . 0
r
y
1 08 . 0 14 . 0
r

,
_

,
_


l
, (7)

where
0
r is the radius of the pipe.









1
o
r / y





Variation of mixing length in a turbulent pipe
flow according to Nikuradse.
R
m
/r
o
R
m
=0.4 y

Eq. (7)


Shortcomings of Mixing Length Hypothesis

The eddy viscosity of the mixing length model as is given by equation (4) implies
that

0
T
for 0
y
U

. (8)

However, Equation (8) contradicts the experimental data in that
T
, in general, is not
zero when 0
y
U

. For example, it is observed that


max T T
| 8 . 0 at the center of a
pipe where
y
U

vanishes.

Note that equation (8) usually does not introduce significant errors for momentum
transport. This is because the turbulent stress approaches zero as 0
y
U

. However, the
mixing length hypothesis also implies that


y
U
m T T

l , (9)

where
T
is the heat or mass diffusivity. When 0
y
U

, equation (9) implies that


0
T
. This result, however, produces major errors in engineering heat transfer
calculations.

The figure shows an example of a heat exchanger where large resistance to heat
flow is predicted by the mixing length model but never observed experimentally.


Cold U
T


T
=
T
= 0


Hot




Variations of mean velocity and eddy
viscosity in a heat exchanger.

Another example of failure of the mixing length theory is illustrated in the
recirculating flow shown in the figure:




















Experimental data indicates that the maximum heat flux that occurs at the
reattachment point, while the mixing length model leads to 0
T T
(since 0
y
U

).

One-Equation Turbulence Models

Many of the one and multi-equation turbulence models are based on the Prandtl-
Kolmogorov equation given by

l
2
1
T
k , (10)

where
i i
u u
2
1
k is the kinetic energy of turbulence and l is a turbulent length scale. It
is expected that
2
1
k be a better representative of the turbulent velocity scale when
compared with
y
U

l .

The general equation for dynamics of k is given as


j j
2
j
i
j
i
j
i
j i
i i
j
j
x x
k
x
u
x
u
x
U
u u )
P
2
u u
( u
x
k
dt
d


1
]
1

. (11)
Reattachment Point
Schematics of backward facing step flows.

For a thin shear layer, equation (11) may be restated as

1
]
1


y
U
v u )
P
2
u u
( v
y dt
dk
i i
, (12)

where the viscous diffusion is neglected and is the dissipation. Equation (12) clearly
shows that the kinetic energy of turbulence is convected, diffused, produced, and
dissipated.


Modeling k-Equation

To close the k-equation, the unknown terms in equation (12) must be modeled.
We assume


y
U
v u
T

, (13)

where
T
is given by (10). The diffusion term in (12) is modeled by a gradient law. i.e.,


y
k P
u u
2
1
v
k
T
i i

,
_

+ , (14)

where
k
is a constant Prandtl number for k.

The dissipation is given as


l
2
3
D
k
c (15)

Therefore, the closed k-equation becomes


l
2
3
D
2
T
k
T
k
c
y
U
y
k
y dt
dk

,
_

,
_

. (16)

Here,
k
and
D
c are constants and l is a length scale distribution.





Near Wall Distribution

Near a wall, the convection and diffusion of turbulent kinetic energy may be
neglected. The production must then be balanced by the dissipation, and Equation (16)
then becomes


l
2
3
D
2
T
k
c
y
U

,
_

or
l
2
3
D
T
2
k
c
) v u (


(17)

Using (10), we find

2
2
D
k
) v u (
c

. (18)

From the experimental results, one finds 3 . 0 ~ 25 . 0
k
v u


.

Thus,

08 . 0 ~ 07 . 0 c
D
. (19)

For turbulence kinetic energy Prandtl number,

1
k
(20)

is usually used.


Distribution of Length Scale

The distribution of the length scale l is needed to complete the system of
governing equations. Near a wall, from (18), we find

k c v u
2
1
D
. (21)

Using (10) and (13), we find


y
U
c k
2
1
D
2
1


l ,
2
2
2
1
D
y
U
c v u

,
_



l (22)

Also, we know that in the inertial sublayer,

y
u
y
U
*

,
2
*
u v u . (23)

Using (23) in (22) and solving for l , we find


m
4
1
D
4
1
D
c y c l l (24)

For 08 . 0 c
D
, 4 . 0 , we find

y 2 . 0 l (25)

It was suggested that to use equation (24) in the entire flow domain with
m
l
given by an empirical mixing length expression. Nikuradse's equation given by (7) has
been used extensively. In immediate vicinity of the wall, further modification for viscous
effects are needed.


Achievements of the Model

Heat transfer in the heat exchanger and separated flows are predicted with
reasonable accuracy.


Shortcomings

Transport of the turbulent length scale is not accounted for.


The model offers little advantage over the mixing length model.

You might also like