# School of Mechanical Aerospace and Civil Engineering

**MSc Fluid Mechanics
**

Near-Wall Modelling

T. J. Craft

George Begg Building, C41

Contents:

◮ Navier-Stokes equations

◮ Inviscid ﬂows

◮ Creeping Flows

◮ Boundary layers

◮ Transition, Reynolds averaging

◮ Turbulent heat transport

◮ Mixing-length models of turbulence

◮ Turbulent kinetic energy equation

◮ One- and Two-equation models

◮ Turbulent heat transport

◮ Near-Wall Modelling

◮ Flow management

Reading:

F.M. White, Fluid Mechanics

J. Mathieu, J. Scott, An Introduction to Turbulent Flow

P.A. Libby, Introduction to Turbulence

P. Bernard, J. Wallace, Turbulent Flow: Analysis Mea-

surement & Prediction

S.B. Pope, Turbulent Flows

D. Wilcox, Turbulence Modelling for CFD

Notes: http://cfd.mace.manchester.ac.uk/tmcfd

- People - T. Craft - Online Teaching Material

Introduction

◮ The near-wall, viscosity-affected, layer of a turbulent ﬂuid ﬂow poses a

number of challenges, from both numerical and modelling viewpoints.

◮ Viscous effects on the turbulence need to be accounted for – often by

including ‘near-wall damping’ terms and other source terms in the

modelled transport equations.

◮ Since very steep gradients of velocity (and turbulence statistics) occur

across the viscous layer, very ﬁne grids are needed to provide adequate

numerical resolution.

◮ In this lecture we brieﬂy consider two alternative strategies for modelling

near-wall ﬂow regions:

◮

Low-Reynolds-number modelling, where the viscous layer is resolved

numerically, and viscous effects included in the turbulence model.

◮

Wall functions, where the viscous layer is not resolved, but

approximations are introduced to account for the ﬂow behaviour

across it.

Near-Wall Modelling 2010/11 2 / 18

Low-Reynolds-Number Modelling

◮ In mixing-length and one-equation models, a lengthscale increasing

linearly with wall-distance ensured a logarithmic velocity proﬁle in the

fully-turbulent near-wall region.

◮ Damping terms were introduced to

reduce the lengthscale (and hence

the turbulent viscosity) across the

viscous sublayer.

◮ In the mixing-length model the

Van-Driest damping function was

introduced:

l

m

= Dmin(κy, λδ) (1)

with D = [1−exp(−y

+

/A

+

)] and

A

+

= 26.

lm

y

y

Region of

U

Near-Wall Modelling 2010/11 3 / 18

Low-Reynolds-Number Two-Equation Models

◮ When studying two-equation models we considered how to tune the

various model coefﬁcients to a number of high-Reynolds-number ﬂows

(ie. situations found in fully turbulent ﬂow regions).

◮ Adaptations are typically required to obtain accuracy across the

viscosity-affected near-wall layer.

◮ These often include near-wall damping terms in the turbulent viscosity

and other model coefﬁcients (dependent on wall-distance or turbulent

Reynolds number, R

t

= k

2

/(εν))

◮ Additional near-wall source terms are sometimes also included in the

modelled lengthscale governing equation.

◮ Some ε-based models employ a modiﬁed ε equation or variable, in order

to simplify boundary conditions, and improve near-wall stability.

Near-Wall Modelling 2010/11 4 / 18

◮ As an example, the low-Reynolds-number Launder-Sharma (1974) k-ε

model uses

D˜ ε

Dt

= c

ε1

˜ εP

k

k

−c

ε2

˜ ε

2

k

+

∂

∂x

j

_

(ν +ν

t

/σ

ε

)

∂ ˜ ε

∂x

j

_

+2νν

t

_

∂

2

U

i

∂x

j

∂x

k

_

2

(2)

where

˜ ε = ε −2ν

_

∂k

1/2

∂x

j

_

2

(3)

The turbulent viscosity is taken as

ν

t

= c

µ

f

µ

k

2

˜ ε

(4)

where c

µ

= 0.09 and

f

µ

= exp

_

−3.4

(1+R

t

/50)

2

_

(5)

with R

t

= k

2

/(˜ εν).

Near-Wall Modelling 2010/11 5 / 18

◮ Near-wall k

proﬁles from

several

low-Re-number

EVM’s.

◮ Note some

models are tuned

to give correct

near-wall

asymptotes in

these ﬂows.

Channel Flow

Boundary Layer

Near-Wall Modelling 2010/11 6 / 18

Wall Functions

◮ The above modelling reﬁnements allow us to apply the models across the

thin viscous sublayer.

◮ However, the near-wall grid needs to be ﬁne enough to provide an

accurate resolution of the ﬂow variables across this layer.

◮ Typically, this requires ensuring the near-wall node lies at a

non-dimensional distance of around y

+

< 1.

◮ In large 3-D calculations, employing such ﬁne grids to resolve the very

near-wall regions of the ﬂow can be extremely expensive.

◮ We thus consider an alternative method of treating the thin near-wall

region where viscous effects are inﬂuential.

◮ The treatment is designed to reduce the computational cost associated

with resolving this thin layer, and is based on the ideas of local

equilibrium simple shear ﬂows already discussed.

Near-Wall Modelling 2010/11 7 / 18

The Need for Wall Functions

◮ In the immediate wall

vicinity, even averaged

turbulent ﬂow properties

undergo extremely abrupt

changes.

◮ In ﬁnite volume, or other numerical schemes, the variation of quantities

between nodes is approximated using linear, or sometimes higher order,

polynomial interpolation.

◮ Since there are steep wall-normal gradients across the viscosity-affected

layer, one must normally use a much ﬁner grid in the direction normal to

the wall than in other regions of the ﬂow.

◮ As an example, 7–10 nodes might be needed to cover a laminar

boundary layer; a turbulent one might require 10–12 nodes for the fully

turbulent region, plus an additional 7–20 nodes to cover the viscous

sublayer (depending on ﬂow and closure model).

Near-Wall Modelling 2010/11 8 / 18

◮ The resulting very elongated computational cells can adversely affect

convergence rates, so numerical stability may require one to use a

reasonably ﬁne grid in other directions also.

◮ Example near-wall grids

around a compressor

blade.

Near-Wall Modelling 2010/11 9 / 18

◮ Three-dimensional grid around a wing-tip geometry:

◮ It is quite possible to spend 70% or more of the execution time for a

complex 3-D computation in simply resolving the near-wall sublayer.

◮ An approach removing the need for such a ﬁnely meshed region would

thus result in considerable savings in computer resources.

Near-Wall Modelling 2010/11 10 / 18

Development of Wall Functions

◮ In a wall function one aims to obtain approximate analytical expressions

for the mean velocity distribution across the near-wall layer, thus avoiding

the need for a detailed numerical resolution of the sublayer.

◮ In a ﬁnite-volume method, integrating the streamwise momentum

equation over a cell leads to a balance between net inﬂow of momentum,

pressure forces, and the shear stresses on the faces of the cell.

◮ Shear stresses at the cell faces are then

approximated assuming a linear variation

of velocity between nodes.

◮ In the near-wall cell shown the shear

stress at the north face of the near-wall cell

is evaluated as τ

n

= µ(U

3

−U

2

)/(y

3

−y

2

),

and the wall shear stress as

τ

w

= µ(U

2

−U

1

)/(y

2

−y

1

).

y

x

δx

1

2

3

1

2

3

y

2

3

◮ This gives a reasonable approximation of τ

w

if the node spacing is

sufﬁciently small.

Near-Wall Modelling 2010/11 11 / 18

◮ To see how a wall function could be developed to give a reasonable

estimate for τ

w

even if the near-wall cell is relatively large, we ﬁrst

consider a laminar near-wall ﬂow.

◮ If we assume velocity variations in the y direction are much larger than in

other directions, and convective transport is negligible, integrating the

streamwise momentum equation twice results in

U =

_

τ

w

µ

_

y +

y

2

2µ

∂P

∂x

(6)

◮ For reasonably small y, or if the pressure gradient is small, the

approximation noted earlier,

τ

w

= µU

2

/y

2

(7)

(taking U

1

and y

1

as zero) is thus reasonably accurate.

◮ The wall shear stress force (τ

w

δxδz) opposing the ﬂuid’s motion can thus

be expressed in terms of velocity differences between adjacent nodes.

◮ For larger cell sizes, we might use equation (6) to give

τ

w

=

_

µU

2

−(y

2

2

/2)∂P/∂x

_

/y

2

(8)

Near-Wall Modelling 2010/11 12 / 18

◮ In a turbulent ﬂow, the velocity proﬁle will only approximate equation (6)

across a small part of the laminar sublayer. It will certainly not hold in the

fully turbulent region.

◮ For turbulent ﬂow, we thus consider how to devise an expression allowing

an estimate of the wall shear stress in terms of the velocity at the

near-wall node, even if the grid is not ﬁne enough to properly resolve the

velocity proﬁle between the wall and near-wall nodes.

◮ In a local equilibrium turbulent ﬂow, with l

m

= κy, we found that the

velocity in the fully turbulent region satisﬁes the log-law:

U

+

=

1

κ

log(Ey

+

) (9)

where κ and E are usually taken as

0.41 and 9 respectively, and

U

+

= U/(τ

w

/ρ)

1/2

y

+

= y(τ

w

/ρ)

1/2

/ν

Near-Wall Modelling 2010/11 13 / 18

◮ Rearranging this expression, we can obtain:

κU

log(Ey(τ

w

/ρ)

1/2

/ν)

=

_

τ

w

ρ

_

1/2

(10)

or, if we assume the conditions hold at the near-wall node,

τ

w

= ρ

_

κU

2

log(Ey

2

(τ

w

/ρ)

1/2

/ν)

_

2

(11)

◮ Consequently, in a turbulent ﬂow one could simply use equation (11) in

place of equation (7).

◮ Since the log-law is valid in the fully turbulent ﬂow region, the near-wall

node (node 2) should now be located not in the viscous layer, where

gradients are very steep and rapidly changing, but in the fully turbulent

region (where gradients will be signiﬁcantly smaller).

◮ In terms of non-dimensional wall-distance, this typically amounts to

having y

+

greater than 30 or so at the near-wall node.

◮ With the above approximation, one thus avoids the need to resolve the

sublayer with a very ﬁne grid.

Near-Wall Modelling 2010/11 14 / 18

◮ For a numerical solution one also needs the effective viscosity at the

near-wall node, in order to interpolate its value to the north cell face.

◮ At node 2, we can obtain the effective viscosity by taking the derivative of

equation (9)

∂U

∂y

=

(τ

w

/ρ)

1/2

κy

Since τ = τ

w

= µ

t

∂U/∂y in the fully turbulent region, then

µ

t

=

τ

w

∂U/∂y

= ρκy(τ

w

/ρ)

1/2

(12)

◮ Thus, we can place the near-wall node beyond the viscous sublayer and

use equation (11) – in an iterative mode, since it is an implicit equation for

τ

w

– to obtain the wall shear stress, and then equation (12) to obtain the

turbulent viscosity.

◮ This approach avoids the need for a very ﬁne near-wall grid, and means

that in principle one need only use the high-Reynolds-number form of

whatever turbulence model is being employed (since the model is not

actually applied in the viscosity-affected region).

Near-Wall Modelling 2010/11 15 / 18

Improved Wall Functions

◮ The above form of wall function can be used with any of the turbulence

models examined so far.

◮ However, it does have a number of weaknesses.

◮ One is that equation (12) implies the turbulent viscosity vanishes when

the wall shear stress is zero.

◮ It will thus give essentially zero turbulent viscosity near ﬂow reattachment

or impingement points despite the fact that, in reality, turbulence levels

will be quite high there.

◮ If using at least a one-equation model, where we have values for the

turbulent kinetic energy k, a somewhat improved wall function formulation

can be adopted.

◮ This improved form is based on broadly the same strategy, but with a

slight reformulation of the log-law.

Near-Wall Modelling 2010/11 16 / 18

◮ In a local equilibrium boundary layer we had

|uv | = τ

w

/ρ and |uv |/k = c

1/2

µ

◮ It follows that k = (τ

w

/ρ)/c

1/2

µ

, and we could use k

1/2

instead of

(τ

w

/ρ)

1/2

as a velocity scale in non-dimensionalizing the mean velocity

and wall-distance.

◮ The modiﬁed non-dimensional velocity and wall-distance are usually

written as

U

∗

=

Uk

1/2

(τ

w

/ρ)

and y

∗

= yk

1/2

/ν (13)

and the corresponding log-law becomes

U

∗

=

1

κ

∗

log(E

∗

y

∗

) (14)

with κ

∗

= c

1/4

µ

κ and E

∗

= c

1/4

µ

E.

◮ The above form is, in fact, more widely used in general-purpose software

than the mixing length form of log-law considered earlier.

Near-Wall Modelling 2010/11 17 / 18

◮ With the above alternative formulation, the effective viscosity µ

t

at the

near-wall node is now given by µ

t

= ρκ

∗

yk

1/2

, which does not vanish

when τ

w

does.

◮ In practice, the value of k appearing in the above expressions is usually

simply taken as the value at the near wall node, k

2

.

◮ There are further improvements that can be made:

◮

Some schemes try to use a value of k that will be less dependent on

the position of the near-wall node.

◮

Others introduce additional terms to account for situations where the

near-wall ﬂow may not be expected to obey the log-law.

◮ Details of these more advanced schemes are not, however, covered here.

Near-Wall Modelling 2010/11 18 / 18

or other numerical schemes. Since there are steep wall-normal gradients across the viscosity-affected layer. In ﬁnite volume. one must normally use a much ﬁner grid in the direction normal to the wall than in other regions of the ﬂow. the near-wall grid needs to be ﬁne enough to provide an accurate resolution of the ﬂow variables across this layer.4 (1 + Rt /50)2
Note some models are tuned to give correct near-wall asymptotes in these ﬂows. even averaged turbulent ﬂow properties undergo extremely abrupt changes.
2010/11 8 / 18
◮
◮
◮
◮ ◮
We thus consider an alternative method of treating the thin near-wall region where viscous effects are inﬂuential. or sometimes higher order.◮
As an example. As an example. and is based on the ideas of local equilibrium simple shear ﬂows already discussed. a turbulent one might require 10–12 nodes for the fully turbulent region. this requires ensuring the near-wall node lies at a non-dimensional distance of around y + < 1. Typically.
νt = cµ fµ
where c µ = 0. the variation of quantities between nodes is approximated using linear. The treatment is designed to reduce the computational cost associated with resolving this thin layer. polynomial interpolation. However.
2010/11 7 / 18
◮
◮
Near-Wall Modelling
Near-Wall Modelling
.
Near-Wall Modelling 2010/11 5 / 18 Near-Wall Modelling 2010/11 6 / 18
Wall Functions
◮
The Need for Wall Functions
◮
The above modelling reﬁnements allow us to apply the models across the thin viscous sublayer.09 and fµ = exp
k2 ˜ ε
(4)
−3.
◮
In the immediate wall vicinity. 7–10 nodes might be needed to cover a laminar boundary layer. the low-Reynolds-number Launder-Sharma (1974) k -ε model uses ˜ ˜ ˜ ˜ Dε εP ε2 ∂ ∂ε (ν + νt /σε ) + 2ννt = cε 1 k − cε 2 + Dt k k ∂ xj ∂ xj where ˜ ε = ε − 2ν The turbulent viscosity is taken as
◮
Channel Flow
Boundary Layer
∂ 2 Ui ∂ xj ∂ xk
2
(2)
◮
∂k ∂ xj
1/2
2
(3)
Near-wall k proﬁles from several low-Re-number EVM’s. plus an additional 7–20 nodes to cover the viscous sublayer (depending on ﬂow and closure model). employing such ﬁne grids to resolve the very near-wall regions of the ﬂow can be extremely expensive. In large 3-D calculations.
(5)
ε with Rt = k 2 /(˜ ν ).

An approach removing the need for such a ﬁnely meshed region would thus result in considerable savings in computer resources. and convective transport is negligible. or if the pressure gradient is small.
◮
It is quite possible to spend 70% or more of the execution time for a complex 3-D computation in simply resolving the near-wall sublayer. For larger cell sizes. and the shear stresses on the faces of the cell. and the wall shear stress as τw = µ (U2 − U1 )/(y2 − y1 ). integrating the streamwise momentum equation twice results in U=
◮
◮
τw µ
y+
y2 ∂P 2µ ∂ x
(6)
◮
◮
3
For reasonably small y . so numerical stability may require one to use a reasonably ﬁne grid in other directions also. we ﬁrst consider a laminar near-wall ﬂow. In the near-wall cell shown the shear stress at the north face of the near-wall cell is evaluated as τn = µ (U3 − U2 )/(y3 − y2 ). thus avoiding the need for a detailed numerical resolution of the sublayer. we might use equation (6) to give
2 τw = µ U2 − (y2 /2)∂ P/∂ x /y2
◮
This gives a reasonable approximation of τw if the node spacing is sufﬁciently small. If we assume velocity variations in the y direction are much larger than in other directions. pressure forces.
2010/11 10 / 18
◮
Near-Wall Modelling
2010/11
9 / 18
Near-Wall Modelling
Development of Wall Functions
◮
◮
In a wall function one aims to obtain approximate analytical expressions for the mean velocity distribution across the near-wall layer. the approximation noted earlier.
◮
Three-dimensional grid around a wing-tip geometry:
◮
Example near-wall grids around a compressor blade. Shear stresses at the cell faces are then approximated assuming a linear variation of velocity between nodes.◮
The resulting very elongated computational cells can adversely affect convergence rates.
τw = µ U2 /y2
y
(7)
◮
(taking U1 and y1 as zero) is thus reasonably accurate.
3
2010/11 11 / 18
(8)
2010/11 12 / 18
Near-Wall Modelling
Near-Wall Modelling
.
To see how a wall function could be developed to give a reasonable estimate for τw even if the near-wall cell is relatively large. integrating the streamwise momentum equation over a cell leads to a balance between net inﬂow of momentum. In a ﬁnite-volume method.
◮
2 1 δx x
◮
The wall shear stress force (τw δ x δ z) opposing the ﬂuid’s motion can thus be expressed in terms of velocity differences between adjacent nodes.

a somewhat improved wall function formulation can be adopted. we thus consider how to devise an expression allowing an estimate of the wall shear stress in terms of the velocity at the near-wall node.
τw = ρ
◮
κ U2 log(Ey2 (τw /ρ )1/2 /ν )
2
(11)
◮
Consequently. If using at least a one-equation model. the near-wall node (node 2) should now be located not in the viscous layer.◮
In a turbulent ﬂow.41 and 9 respectively. but with a slight reformulation of the log-law. in order to interpolate its value to the north cell face. but in the fully turbulent region (where gradients will be signiﬁcantly smaller). This improved form is based on broadly the same strategy. since it is an implicit equation for τw – to obtain the wall shear stress. With the above approximation. This approach avoids the need for a very ﬁne near-wall grid. It will thus give essentially zero turbulent viscosity near ﬂow reattachment or impingement points despite the fact that. with lm = κ y . we can obtain the effective viscosity by taking the derivative of equation (9) (τw /ρ )1/2 ∂U = ∂y κy Since τ = τw = µt ∂ U/∂ y in the fully turbulent region. we found that the velocity in the fully turbulent region satisﬁes the log-law: 1 U + = log(Ey + ) (9) κ where κ and E are usually taken as 0. One is that equation (12) implies the turbulent viscosity vanishes when the wall shear stress is zero. For turbulent ﬂow. and U + = U/(τw /ρ )1/2 y + = y (τw /ρ )1/2 /ν
◮
Rearranging this expression. one thus avoids the need to resolve the sublayer with a very ﬁne grid. However. At node 2. in reality. It will certainly not hold in the fully turbulent region.
2010/11 14 / 18
◮
◮
◮
Near-Wall Modelling
2010/11
13 / 18
Near-Wall Modelling
◮
For a numerical solution one also needs the effective viscosity at the near-wall node. turbulence levels will be quite high there. in a turbulent ﬂow one could simply use equation (11) in place of equation (7). the velocity proﬁle will only approximate equation (6) across a small part of the laminar sublayer. then
Improved Wall Functions
◮
◮
The above form of wall function can be used with any of the turbulence models examined so far. this typically amounts to having y + greater than 30 or so at the near-wall node. even if the grid is not ﬁne enough to properly resolve the velocity proﬁle between the wall and near-wall nodes. Since the log-law is valid in the fully turbulent ﬂow region.
2010/11 16 / 18
◮
◮
µt =
◮
τw = ρκ y (τw /ρ )1/2 ∂ U/∂ y
(12)
◮
Thus. In terms of non-dimensional wall-distance. In a local equilibrium turbulent ﬂow. we can obtain:
κU = log(Ey (τw /ρ )1/2 /ν )
τw ρ
1/2
(10)
◮
or. where gradients are very steep and rapidly changing. and means that in principle one need only use the high-Reynolds-number form of whatever turbulence model is being employed (since the model is not actually applied in the viscosity-affected region). if we assume the conditions hold at the near-wall node. it does have a number of weaknesses. where we have values for the turbulent kinetic energy k . we can place the near-wall node beyond the viscous sublayer and use equation (11) – in an iterative mode. and then equation (12) to obtain the turbulent viscosity.
2010/11 15 / 18
◮
◮
◮
Near-Wall Modelling
Near-Wall Modelling
.

◮
◮
U∗ = with κ ∗ = cµ κ and E ∗ = c µ E. which does not vanish when τw does. There are further improvements that can be made:
◮
◮
It follows that k = (τw /ρ )/c µ . k2 . In practice. the value of k appearing in the above expressions is usually simply taken as the value at the near wall node. covered here.
(14)
1/4
1/4
The above form is.◮
In a local equilibrium boundary layer we had |uv | = τw /ρ and |uv |/k = c µ
1/2
◮
With the above alternative formulation. The modiﬁed non-dimensional velocity and wall-distance are usually written as Uk 1/2 and y ∗ = yk 1/2 /ν U∗ = (13) (τw /ρ ) and the corresponding log-law becomes
1/2
◮
◮
◮
Some schemes try to use a value of k that will be less dependent on the position of the near-wall node. however. in fact. and we could use k 1/2 instead of (τw /ρ )1/2 as a velocity scale in non-dimensionalizing the mean velocity and wall-distance. more widely used in general-purpose software than the mixing length form of log-law considered earlier. the effective viscosity µt at the near-wall node is now given by µt = ρκ ∗ yk 1/2 . Others introduce additional terms to account for situations where the near-wall ﬂow may not be expected to obey the log-law.
2010/11 17 / 18 Near-Wall Modelling 2010/11 18 / 18
Near-Wall Modelling
.
◮
1 log(E ∗ y ∗ ) κ∗
Details of these more advanced schemes are not.