You are on page 1of 23

CFD Calculation of Turbulent Flow with Arbitrary

Wall Roughness
David Apsley
Received: 30 March 2006 / Accepted: 2 November 2006 /
Published online: 31 January 2007
#
Springer Science + Business Media B.V. 2007
Abstract Wall functions are used in CFD calculations of turbulent flow to handle the no-
slip boundary condition at solid surfaces without an unacceptably-large number of grid
cells. Recent work in the Turbulence Mechanics Group at the University of Manchester has
been aimed at developing wall functions suitable for non-equilibrium flows primarily for
smooth walls. For environmental fluid-mechanics problems, however, there is a need to
extend the new wall-function concepts to arbitrary wall roughness. Suga, Craft and
Iacovides (Extending an analytical wall function for turbulent flows over rough walls, 6th
International Symposium on Engineering Turbulence Modelling and Measurements,
Sardinia, 2005) have shown how this may be achieved by making the zero-eddy-viscosity
height, y
n
, a function of roughness. Their function is, however, largely empirical, and this
paper goes further by demonstrating that the zero-eddy-viscosity height can be more
satisfactorily determined by appealing to higher-order consistency with the log law.
Roughness-dependent changes are also made to the near-wall prescription of the turbulent
kinetic energy dissipation rate, e. The wall function for arbitrarily-rough surfaces is
developed in detail and its ability to reproduce the Darcy friction factor in rough-walled
pipes is tested. Some computational results for the more complex application which
stimulated this work sediment transport and sandbank morphodynamics are then given.
The application of the wall function to the Reynolds-stress transport equations and
modifications for streamwise pressure gradients are also discussed.
Key words computational fluid dynamics
.
wall functions
.
rough-wall boundary layers
1 Introduction
In modelling turbulent flow past solid surfaces computational fluid dynamics (CFD) has to
accommodate the thin region of large gradients, anisotropic turbulence and influence of
Flow Turbulence Combust (2007) 78:153175
DOI 10.1007/s10494-006-9059-x
D. Apsley (*)
School of Mechanical, Aerospace and Civil Engineering, University of Manchester,
P.O. Box 88, Manchester M60 1QD, UK
e-mail: d.apsley@manchester.ac.uk
molecular viscosity in the vicinity of the boundary. Two alternative approaches may be
used. The first accepts the computational expense of resolving this region fully and employs
so-called low-Reynolds-number turbulence models. The second adopts a high-
Reynolds-number or wall-function approach, whereby the computational mesh is
comparatively coarse and the behaviour of variables at distances out to the near-wall node
is assumed from values at that node on the basis of universal functions (often based on
the log law).
Low-Reynolds-number turbulence modelling strives to use grids capable of resolving the
near-wall sublayer down to y
+
values of 1 or less, incorporating viscosity-dependent terms
in the eddy-viscosity formula and turbulence-transport equations in order to reproduce the
theoretical near-wall behaviour of turbulence variables (in particular: turbulence kinetic
energy k y
2
, dissipation rate " ~ 2nk
_
y
2
constant and eddy viscosity n
t
= O y
3
( ) as
wall-normal distance y0). The writer has been as guilty as any in employing
computational cells finer than any practical manufacturing smoothness. Many such models
use functions based on the gradients of k or the Reynolds stresses in order to avoid the
appearance of wall-distance and wall-normal-direction parameters in the low-Re terms.
However, although wall-normal direction may be ambiguous, the distance to the nearest
wall is well-defined and continuous (even in corners, where a point may be equidistant
from two or more wall segments), so the appearance of gradient functions, which are often
numerically-destabilising, is not easily justified.
Wall functions have been widely criticised in the past for their assumption of standard
profiles which have little theoretical foundation in non-equilibrium flow, particularly near
flow-separation and impingement points. A second undesirable restriction has been the
need to locate the near-wall node in the fully-turbulent region a requirement that limits
grid refinement. For environmental flows, however, and, in general, for flow over surfaces
which cannot be regarded as hydraulically smooth (the original motivation for this work
was an erodible bed), a wall-function approach is the only practical option, and the
challenge is to devise methods which can meet the following criteria:
& Satisfactory behaviour in non-equilibrium boundary layers;
& Insensitivity to the position of the near-wall node and removal of the need to place it in
the fully-turbulent region;
& Seamless coverage of the transition from hydraulically smooth to hydraulically rough.
In recent years wall-function methods have been revitalised by research in the
Turbulence Mechanics Group at UMIST (now part of the new University of Manchester)
through the analytical-wall-function approach [3, 4]. For smooth-wall boundary layers
this goes a long way to meeting the first two criteria; namely, better behaviour in non-
equilibrium boundary layers and insensitivity to the size of the near-wall cell. Two key
elements of this approach are the following.
(1) As originally proposed by Chieng and Launder [2], a turbulent velocity scale
dependent on the turbulent kinetic energy at the near-wall node, k
P
:
u
t
= C
1=4
m
k
1=2
P
This is equal to the actual friction velocity
u
t
=

t
w
=r
_
154 Flow Turbulence Combust (2007) 78:153175
in the log layer (and leads to a value of 0.09 for the constant C
m
in the standard k-e
model), but the prescription based on k behaves more sensibly near separation/
impingement points where the wall shear stress
w
vanishes.
(2) An assumed mean-velocity profile determined by a continuous total-viscosity profile
incorporating both sublayer and fully-turbulent limits:
n
total
= n max 0; ku
t
y y
n
( )
In non-dimensional wall units this can be written
n
total
n
= 1 max 0; k y

n
_ _ _ _
where y

= yu
t
=n, the tilde signifying that the velocity scale is u
t
rather than u
t
.
Both of these elements will be retained in the present work, although a different value of
y

n
(based on stronger asymptotic adherence to the log law) is adopted, compared to that in
Craft et al. [3, 4]. Note also that results in this paper are generally formulated in the more-
familiar + form of dimensionless wall units (y

= u
t
y=n) rather than the * form used in
the Craft et al. sequence of papers (y* = k
1=2
p
y
_
n). When the tilde version is employed,
the two dimensionless heights are proportional: y

= C
1=4
m
y*.
Recently, Suga et al. [11] have extended the analytical wall function to rough
boundaries. Roughness enters the wall function solely through y

n
, which is made an
empirical function of

k

s
(equivalent sand roughness in wall units). In this paper it is shown
in Section 2 below that y

n
can be determined in a less empirical way by enforcing a higher-
order consistency with the log law:
U

=
1
k
ln y

B k

s
_ _
;
retaining not just the slope constant (1/) but the offset B and its dependence on roughness.
In addition, roughness-dependent changes are made to the prescription for near-wall
dissipation rate e; such changes are found necessary to maintain u
t
as a close
approximation to u
t
, which is implicit in the derivation of the eddy-viscosity profile.
Section 2 provides a detailed derivation of the wall function for rough boundaries,
including its use with Reynolds-stress-transport as well as eddy-viscosity models. In Section 3
the wall-function methodology is judged on the proposed criteria outlined above and its
ability to reproduce the Darcy friction factor for flow in rough pipes. Its utility in providing
the surface shear stress, which is crucial in predicting the morphodynamics of erodible beds,
is described in Section 4. Finally, Section 5 summarises the main elements of the model.
2 Wall-function Specification
2.1 Output required from a wall function
In a cell-centred, finite-volume calculation with the k- turbulence model a wall function
must provide the following:
& For the mean-velocity components, the wall shear stress
w
;
& For the turbulent-kinetic-energy and Reynolds-stress equations, the cell-averaged
production and dissipation, P
k ( )
av
and "
an
;
Flow Turbulence Combust (2007) 78:153175 155
& For the dissipation (e) equation, the value of e at the near-wall node (used as a
boundary condition).
These are to be deduced from the (wall-tangential component of) mean velocity, U,
turbulent kinetic energy, k, and distance of the near-wall node from the boundary, y
p
. A
collocated arrangement is assumed, with all field variables stored at cell centre (Fig. 1).
2.2 Mean velocity and shear stress
The principal assumptions used to derive the analytical mean-velocity profile are as
follows.
& The shear stress is either constant (and equal to its value
w
at the wall) or varies
linearly with distance y from the wall according to the streamwise pressure gradient:
t = t
w

dP
dx
y (1)
& The mean velocity is determined from the shear stress and a total effective viscosity
n
total
= n n
t
= n max 0; ku
t
y y
n
( ) (2)
where the zero-eddy-viscosity height y
3
is to be specified as a function of roughness.
& The rate of dissipation of turbulent kinetic energy, e, is given by a continuous profile
" =
u
3
t
k y y
d
( )
; y _ y
"
"
w
; y _ y
"
:
_
_
_
(3)
where, again, y
"
and y
d
are to be specified as a function of roughness.
Sketches of the assumed profiles of 3
t
, U and e are given in Fig. 2.
In Craft et al. [3, 4] equation (1) is modified to include the fluid acceleration UU/x.
However, since this is not constant but varies in a complex manner with distance from the
wall, it has not been included here.
For simplicity, the formulae developed below assume a constant shear stress (=
w
).
The inclusion of a streamwise pressure gradient is straightforward, but the algebra is
considerably more complex, and so the theoretical formulae have been relegated to an
Appendix. For pipe flow (Section 3) the effect of including a streamwise pressure gradient
in the wall function is negligible, whilst for general 2-d flow calculations with fixed
boundaries its inclusion presents no great difficulties. However, in sediment morphody-
namics (Section 4), where prediction of surface evolution relies on a well-behaved
distribution of bed shear stress, the inclusion of a pressure gradient gave rise to significant
k
p
p

w

U
y
p
near-wall node
Fig. 1 Near-wall cell
156 Flow Turbulence Combust (2007) 78:153175
numerical difficulties in the bed-height equation. Since this paper focuses on modifications
for wall roughness, and because the work was originally motivated by sediment
morphodynamics, pressure gradients have been excluded from the wall function in the
results presented here.
With the total-viscosity formula (2) the shear stress and mean-velocity profile are related by
t
r
=
n
@U
@y
; y _ y
n
n ku
t
y y
n
( ) [ [
@U
@y
; y _ y
n
:
_
(4)
where U=0 on y=0 and U is continuous at y = y
n
.
If =
w
and y
n
_ 0 this can be integrated to give (in wall units):

T y

T y

n

1
k
ln 1 k y

n
_ _ _ _ _
;
y

_ y

n
y

_ y

n
:
_
(5)
where

=
U
u
t
;

T =
t
w
ru
2
t
(6)
The tilde makes the notation slightly clumsy, but it is used here to emphasise that the
wall units are based on the equivalent velocity scale u
t
derived from the turbulent kinetic
energy, rather than the actual friction velocity u
t
based on the (as-yet-unknown) wall shear
stress
w
. In the log layer, u
t
= u
t
and

T = 1.
For sufficiently rough walls, it will be shown that it is necessary to admit cases where
y

n
_ 0. Here, there is no viscous sublayer and direct integration of (4) with U=0 at y=0
produces

=

T
1
k
ln
1 k y

n
_ _
1 ky

n
_ _
(7)
For a general y

n
, (5) and (7) can be summarised collectively as

T y

; y

_ y

T y

n0

1
k
ln
1k y

n
( )
1k y

n0
y

n ( )
_ _ _ _
; y

_ y

n
:
_
_
_
(8)
y




t
y

y y
U
y


y

w

av

Fig. 2 Assumed profiles for eddy viscosity (3


t
), tangential mean velocity (U) and dissipation rate ()
Flow Turbulence Combust (2007) 78:153175 157
where
y

n0
= max y

n
; 0
_ _
(9)
The required output from the wall function is the wall shear stress in terms of the near-
wall mean velocity, not vice versa. Moreover, many CFD codes will see the velocity
gradient at the wall as U
p
/ y
p
. To this end, (8) can be rearranged (and implemented in
computer code) in terms of an effective wall viscosity 3
eff,wall
such that
t
w
= rn
eff ;wall
U
p
y
p
(10)
where
n
eff ;wall
= n
1 y

P
_ y

n
y

P
y

n0

1
k
ln [
1k

y

P
y

n
( )
1k y

n0
y

n
( )
[
; y

P
_ y

n
:
_

_
(11)
It remains to specify the matching height y

n
. The method for doing so constitutes a
novel and important element of this paper.
2.3 The zero-eddy-viscosity height y

n
Key to the wall-function formulation is the zero-eddy-viscosity height (in wall units), y

n
.
The presumptions here are that:
(1) As in Suga et al. [11], it should be a function of the non-dimensional roughness height

s
=
k
s

u
t
n
;
(2) It should be chosen so as to provide the best asymptotic matching to an equilibrium
mean-velocity profile (i.e., log law, with roughness dependence) as y

.
Consider a zero-pressure-gradient equilibrium turbulent boundary layer. Here one may
drop the tilde and note that

T = t
w
=ru
2
t
= 1.
Consider first the case y

n
> 0. Then, in the turbulent region, the mean-velocity profile
(5) is
U

= y

n

1
k
ln 1 k y

n
_ _ _
Write this as
U

= y

n

1
k
ln [ky

(1
1 ky

n
ky

)[
and expand, using the laws of logs (ln ABC = ln A ln B ln C) and series expansion
(ln 1 z ( ) = z
z
2
2

z
3
3
). Collecting terms of similar order in y
+
:
U

=
1
k
ln y

n

1
k
ln k
_ _
O(
1
y

)
Comparing this with the usual log law:
U

=
1
k
ln y

B
158 Flow Turbulence Combust (2007) 78:153175
gives, from the constant term,
y

n
= B
1
k
ln k (12)
Typical smooth-wall values are k=0.41, B=5.2, giving
y

n
= 7:37
This is slightly different from the value of 5.9 (corresponding to y*=10.8) in Craft et al.
[4]. The important thing, however, is that (12) can automatically be used withany accepted
roughness dependence of B. For fully-rough walls a well-attested expression is [9, p527]:
B = 8:0
1
k
ln k

(13)
whence
y

n
= 10:17
1
k
ln k

(14)
This formula indicates that y

n
will become negative for sufficiently large k
+
. When this
is the case one must use the alternative velocity profile given by (7). Expanding in similar
fashion one finds
U

=
1
k
ln
1k y

n
( )
1ky

n
_ _
=
1
k
ln ky

1 1ky

n
( )=
ky

1ky

n
_ _ _ _
=
1
k
ln y

1
k
ln k ln 1 ky

n
_ _ _
O
1
y

_ _
Again, comparing with the log law:
U

=
1
k
ln y

B k

s
_ _
gives, after some rearrangement,
y

n
=
1
k
1 e
k B
1
k
ln k ( )
_ _
(15)
Equations (12) and (15) provide a continuous prescription for y

n
as B varies. In
principle, one could use any acceptable roughness-dependent function B. However,
calculations for fully-developed pipe flow (Section 3) leads to the following suggested
specification for y

n
:
y

n
=
B
1
k
ln k if B
1
k
ln k _ 0
1
k
1 e
k B
1
k
ln k ( )
_ _
; if B
1
k
ln k _ 0
:
_

_
(16)
where the following prescription for B admits both smooth- and rough-wall limits:
B = 8
1
k
ln k

s
3:152
_ _
(17)
with k=0.41. (The constant within the logarithm yields the smooth-wall value B=5.2 when
k

s
= 0).
Flow Turbulence Combust (2007) 78:153175 159
The functional variation of y

n
with

k

s
is plotted in Fig. 3. Note that, according to
(16), the smooth-walled value is y

n
= 7:37 and the viscous sublayer is destroyed (i.e.,
y

n
= 0) when

k

s
= 62.
2.4 Production and dissipation
Cell-averaged values for the rate of production and dissipation of turbulent kinetic energy,
P
k ( )
av
and "
av
, are obtained by integrating assumed profiles for P
(k)
and e:
P
k ( )
= uv
@U
@y
= n
eff
n
_ _
@U
@y
_ _
2
(18)
" =
u
3
t
k y y
d
( )
; y _ y
"
"
w
; y _ y
"
:
_
_
_
(19)
The presumption is that e tends to a constant, e
w
, as the wall is approached. The value is
such as to make the profile continuous at y = y
"
. Note that a constant height offset, y
d
, has
been introduced to the upper part of the profile, for reasons which will be explained below.
This does not change the large-y behaviour of e, which still falls off as 1/y. Heights y
d
and
y
"
are specified after the presentation of cell-averaged quantities below.
Averaging over the depth of a cell, :
P
k ( )
av
=
1
D
_
D
0
P
k ( )
dy =

T
2
u
3
t
kD
ln
s
2
s
1

1
s
2

1
s
1
_ _
(20)
"
av
=
1
D
_
D
0
" dy =
u
3
t
kD
ln

d
y

"
y

"
y

"
y

d
_ _
(21)
where s
1
and s
2
are non-dimensional mixing lengths:
s
1
= 1 k y

n0
y

n
_ _
s
2
= 1 k

D

n
_ _
Equation (20) assumes that the cell depth exceeds y
n
; if not, P
k ( )
av
is simply zero.
Similarly, equation (21) assumes that exceeds y
"
; if not, "
an
is equal to e
w
. Note that, for
cell-centred storage,

D

= 2y

P
.
It is necessary to specify y
d
and y
"
. For smooth walls the turbulent kinetic energy
equation implies that
" ~
2nk
y
2
constant y 0 ( ) (22)
since k varies quadratically as y0. It had originally been intended, as in Craft et al.
[3, 4], to take y
d
=0 and y
"
as the height at which the log-layer profile for e from (19) and
the sublayer profile from (22) coincide ... assuming that k can be replaced by k
p
. This gives
y

"
= 2:73. However, there is little relationship between the quadratic behaviour of k in the
160 Flow Turbulence Combust (2007) 78:153175
sublayer and the constant value k
P
assumed in the log layer. Moreover, as has already been
noted, the viscous sublayer vanishes for sufficiently-large roughness. Most important of all,
however, is that with such a small value of y

"
the cell-averaged dissipation tends to be
relatively high and, consequently, the turbulence-energy-based scale u
t
often differs
significantly from the stress-derived scale u
t
even in near-equilibrium boundary layers. This
is inconsistent with a main premise of the wall function. Instead, it is necessary to tie cell-
averaged dissipation and production more closely, and inspection of the bracketed terms in
(20) and (21) show that this can be achieved, with differences of O(1/
+
), by taking
y

d
= y

n

1
k
(23)
y

"
= y

d
E exp
y

d
y

"
y

d
_ _
(24)
where
E =
1
k
s
1
exp 1
1
s
1
_ _
For arbitrary roughness, equation (24) must be solved iteratively. However, this
converges very rapidly starting from the smooth-wall values
y

d
= 4:9; y

"
= 27:4 smooth wall ( ) (25)
Finally, the value of e
P
from equation (19) is used to set the actual value of e at the near-
wall node; in the e equation this constitutes a boundary condition and no finite-volume
equations are solved for the near-wall cells.
-6
-4
-2
0
2
4
6
8
0 20 40 60 80 100 120 140 160 180 200
k
s
+
y

+
Re=5E4
Re=1E5
Re=5E5
Re=1E6
Fit
Fig. 3 Zero-eddy-viscosity height y

3
as a function of roughness
Flow Turbulence Combust (2007) 78:153175 161
2.5 Stress-transport equations
The stress-transport equations take the form (e.g. [6]):
D
Dt
u
i
u
j
=
@
@x
k
d
ijk
P
ij

ij
"
ij
where the LHS is the rate of change following the flow and the terms on the RHS are,
respectively, diffusion due to spatial inhomogeneity, production by mean strain,
redistribution by pressure fluctuations and dissipation by viscosity. For a simple shear
flow, local equilibrium (P
ij

ij
"
ij
= 0), together with a linear model for the pressure-
strain correlation
ij
, make it possible to determine the structure functions u
i
u
j
_
k in terms
of the constants in the pressure-strain model. Together with knowledge of k, this can be
used to fix the values of the stresses at the near-wall node. In general, the local-equilibrium
assumption predicts the values of stresses in wall-oriented coordinates (t,n,b):
u
2
t
; u
2
n
; u
2
b
; u
t
u
n
where t is the tangential coordinate (defined by the direction of the tangential component of
near-wall velocity), n is normal to the wall and b completes the right-handed triad.
Cartesian components of stress can then be obtained by rotation.
This procedure works well in attached, 2-dimensional flows, where the direction of the
mean velocity is fixed. Major numerical problems arise, however, near separation or
reattachment points because of the discontinuity in the direction of the unit tangent vector t.
This can result in large changes in the stresses from iteration to iteration and prevent
numerical convergence. The problem is exacerbated in three dimensions where the
direction of t is indeterminate when the near-wall velocity vanishes.
An alternative which is more in accordance with the treatment of the k equation is to
use cell-averaged production (and dissipation) for the stress equations also. For a simple
shear flow one has, in wall-oriented coordinates:
P
tt
= 2u
t
u
n
@U
t
@n
= 2P
k ( )
; P
nn
= P
bb
= 0
P
tn
= u
2
n
@U
t
@n
=
u
2
n
u
t
u
n
P
k ( )
; P
nb
= P
bt
= 0
Then, provided that we assume a typical value (here we use 0.83, or 0:248
_
C
1=2
m
) for the
ratio of normal stresses to shear stresses, u
2
n
_
u
t
u
n
, in an equilibrium boundary layer, we
may relate the cell-averaged production of the stresses to the cell-averaged production of k;
the non-zero components are:
P
tt;av
= 2P
k ( )
av
; P
tn;av
= 0:83P
k ( )
av
(26)
with the actual Cartesian components being obtained by rotation; for example:
P
11;av
= P
tt;av
t
1
t
1
P
tn;av
t
1
n
1
(27)
where (t
i
) and (n
i
) are the Cartesian components of the tangential and normal vectors t and
n, respectively. Although t is discontinuous where the mean velocity vanishes, the cell-
averaged production terms are not discontinuous, because they are proportional to the
magnitude of the near-wall mean velocity.
162 Flow Turbulence Combust (2007) 78:153175
3 Pipe Flow
3.1 Calculation of the Darcy friction factor
An important parameter in fully-developed pipe flow is the friction factor l, relating
frictional head loss to the dynamic head based on the bulk velocity. If the pipe has length L
and diameter D and the bulk velocity is U
an
then the frictional head loss h is given by
Dh = l
L
D
U
2
av
2g
(28)
(There are other definitions of friction factor in general use. l here corresponds to, and is
computed as, four times the skin-friction coefficient. The definition adopted here simply
ensures that the numerical constant in the head-loss formula is unity.)
Values of l for arbitrary roughness and Reynolds number can be determined from the
implicit ColebrookWhite equation:
1

l
_ = 2:0 log
10
k
s
3:7D

2:51
Re

l
_
_ _
(29)
where Re is the Reynolds number based on bulk velocity and diameter. Values are typically
plotted on a Moody chart (l vs Re for different values of relative roughness, k
s
/D).
Computations of fully-developed pipe flow were performed with an in-house research
code (STREAM). Details of the code are deferred to Section 4, but a brief description of
how a fully-developed flow may be computed over a short fetch is as follows.
(1) A 2-d (axisymmetric) finite-volume mesh is generated, with a suitable distribution of
grid lines in the cross-stream (r) direction and (e.g.) three columns of cells in the
streamwise (x) direction.
(2) A finite-volume calculation is initiated, with top-hat profiles for all variables.
(3) For each variable except pressure a downstream boundary condition of zero
streamwise gradient is applied (/x=0). Pressure is linearly extrapolated from the
interior nodes; (a non-zero pressure gradient is needed to drive the flow).
(4) At the end of each iteration the downstream velocity U is rescaled to give the desired
bulk velocity (=1 in non-dimensional units) and then the values of all transported
variables are mapped back to the inflow for the next iteration.
Since the advective derivative vanishes in the fully-developed flow, a simple first-order
upwind-differencing scheme is sufficient. Although the cross-stream velocity V should also
vanish in the final solution, a faster convergence to the fully-developed state (and a check
on numerical implementation) is achieved by retaining this equation in the solution
procedure.
3.2 Friction factor as a function of roughness and Reynolds number
In a first series of numerical experiments, fully-developed pipe flow was computed using
the standard k-e turbulence model on a uniform mesh with constant cross-stream grid
spacing Dr = 0:02D, at pipe Reynolds numbers Re = U
av
D=n between 510
4
and 10
6
and
for both smooth walls and a range of relative roughness values k
s
/D between 0 and 0.02.
For each (Re, k
s
/D) pair the value of y

n
was varied in order to reproduce the solution of
(29) exactly. The resulting optimal values are plotted against

k

s
in Fig. 3, along with
Flow Turbulence Combust (2007) 78:153175 163
the fitted value of y

n
according to equations (16) and (17). (For smooth walls the near-wall
value y

p
varies from 26 at Re=510
4
to 381 at Re=10
6
; for each Reynolds number it
increases gradually with roughness.)
In a second set of numerical experiments, equations (16) and (17) were used throughout
and fully-developed pipe flow was computed for Reynolds numbers between 510
4
and
10
6
and relative roughness k
s
/D ranging from 0.0 (i.e., smooth wall) to 0.03. The meshes
used had 25 cells in the radial direction, but the near-wall depth r
wall
was varied so as to
ensure a value of y
+
at the near-wall node of about 50 in the smooth-walled flow; (values of
y

p
increase with increasing roughness). The resulting friction factors for k-e and Reynolds-
stress-transport models are shown in Fig. 4. For the k-e model there is universally
excellent agreement with the ColebrookWhite formula. The values of l for the Reynolds-
stress-transport model show the correct trend with increasing roughness but depart more
from the ColebrookWhite values for larger relative roughness.
3.3 Sensitivity to mesh parameters
In order to determine the sensitivity of the wall-function to mesh size, a single Reynolds
number (Re=10
5
) was employed and the friction factor l computed as a function of relative
roughness for a series of uniform meshes (cell depth r, number of cells in the radial
direction N
r
):
Mesh A: r/D=0.005, N
r
=100
Mesh B: r/D=0.01, N
r
=50
Mesh C: r/D=0.02, N
r
=25
Mesh D: r/D=0.033, N
r
=15
The results are shown in Fig. 5. Values of l are insensitive to the near-wall cell size,
except for the finest grid (mesh A). In that case the near-wall node value y

p
is only 13.2
and both the whole near-wall cell and the next layer of cells lie in a region where mean
velocity and turbulence variables change rapidly. On this mesh roughness actually improves
matters, partly because y

p
is greater, but also because the flow is well-mixed closer to the
wall and hence gradients are smaller and more accurately resolved beyond the near-wall
node.
Computations were also performed on an expanding mesh (B), with r
min
=0.01 at the
pipe wall, but only N
r
=20 cells. The resulting friction factors were graphically
indistinguishable from those for mesh B and have not been shown. This implies
insensitivity to mesh expansion factors.
3.4 Velocity and turbulent-kinetic-energy profiles
Comparison has been made with the pipe-flow data contained in the AGARD [1] database
of test cases for the validation of LES calculations (Chapter 5, case PCH02). Figure 6
shows the mean-velocity profile (in wall units) for a smooth wall and for a rough wall with
k

s
= 11:77. The pipe Reynolds number is 10
5
. The plot, on log-linear axes, clearly shows
the log-law region and the effect of roughness in displacing the velocity profile.
The variation of turbulent kinetic energy is shown as a function of distance from the pipe
axis in Fig. 7. Turbulent kinetic energy k has been non-dimensionalised in this instance
using the bulk velocity in the pipe. Note, however, that the increase in turbulence energy
due to roughness is proportional to the increase in wall stress; if k is non-dimensionalised
164 Flow Turbulence Combust (2007) 78:153175
(a)
0.01
0.10
1.0E+04 1.0E+05 1.0E+06
Reynolds Number, Re
F
r
i
c
t
i
o
n

f
a
c
t
o
r
,


k
s
/D=0
k
s
/D=0.0003
k
s
/D=0.001
k
s
/D=0.003
k
s
/D=0.01
k
s
/D=0.03
(b)
0.01
0.10
1.0E+04 1.0E+05 1.0E+06
Reynolds Number, Re
F
r
i
c
t
i
o
n

f
a
c
t
o
r
,


k
s
/D=0
k
s
/D=0.0003
k
s
/D=0.001
k
s
/D=0.003
k
s
/D=0.01
k
s
/D=0.03
Fig. 4 Variation of friction factor with roughness and Reynolds number for (a) k- model; (b) Reynolds-
stress transport model. Symbols computations; solid curves solution of the ColebrookWhite equation
Flow Turbulence Combust (2007) 78:153175 165
using the friction velocity then smooth- and rough-wall profiles are coincident, with a
maximum value for k
_
u
2
t
of 3.38 (corresponding almost exactly to 1
_
C
m
_
).
The shear stress varies linearly across the pipe (as it must in fully-developed flow) and is
not shown here.
4 Flow Over an Erodible Bed
The application which prompted this work is the computation of flow around, and
consequent evolution of, sandbanks. Specifically, the objective is to simulate sediment
transport and morphodynamics in both constant-current and tidal flows in a series of
experiments conducted at HR Wallingford in 20042005. The size of the sediment particles
used is such that bed load rather than suspended load is the dominant mechanism of
sediment transport. A careful treatment of wall roughness is needed because the lower
boundary can range from hydraulically-smooth to fully-rough over the course of a tidal
cycle and, in the case where the sand mound is built on a non-erodible concrete bed, at
different positions in the flow domain.
To illustrate the application, Figs. 8, 9, 10 show the predicted evolution of a sand cone
in steady and tidal flows. The cone is initially axisymmetric, with a slope of 1 in 3 and
height of 0.15 m, and is formed from sand of density 2,650 kg m
3
and median diameter
454 m on a non-erodible (smooth) bed. The local surface roughness k
s
is set to the lesser
of the particle diameter and the local bed height. The water depth is 0.2 m. The bulk
velocity for the constant-current case is 0.37 m s
1
, whilst the maximum bulk velocity in
Fig. 5 Sensitivity to the mesh spacing (mesh parameters are given in the text)
166 Flow Turbulence Combust (2007) 78:153175
each direction for the tidal case is 0.39 m s
1
and varies sinusoidally with a period of
10 min. In a constant current the cone steepens (due to increased stress at the summit) and
gradually moves downstream. Flow separation and recirculation wash sand back upstream
in the lee of the bank and lead to the formation of a characteristic crescent shape. Flow
reversal in the second half of a tidal cycle sweeps sand in the opposite direction and the
result after several cycles is an enhanced cross-stream spread.
Flow calculations were performed with an in-house research code, STREAM, which
uses the finite-volume method to solve the Reynolds-averaged NavierStokes (RANS)
equations on multi-block curvilinear meshes. Variable storage is cell-centred and co-located
and the code uses the standard SIMPLE pressure-correction algorithm. The code features an
extensive repertoire of turbulence models [7], but only the standard k- model [5] has been
used in this application. The code permits the use of a (bed- and free-surface-fitting)
moving mesh but, for simplicity, the free surface has here been treated as a rigid-lid, the
mesh evolving slowly to conform to the moving sand.
The flow calculation is coupled to a bed-evolution equation. More complete numerical
details will be given in a separate paper devoted to morphodynamics, but a brief description
follows. Conservation of sand leads to an equation for the time variation of the bed height
z
b
(x,y):
1 p ( )
@z
b
@t
= \
h
v q
h
0
5
10
15
20
25
30
10 100 1000 10000
y+
U
+
Smooth - experiment
Smooth - computed
Rough - experiment
Rough - computed
Fig. 6 Mean-velocity profiles
Flow Turbulence Combust (2007) 78:153175 167
where p is the porosity (here taken as 0.4). q
h
is the horizontal rate of transport of sand per
unit horizontal length, and is different from the actual bed-flux vector q
b
when there are
slopes because: (1) the actual length of surface exceeds the projected horizontal length; and
(2) only the horizontal component of the bed flux transports sand out of a particular vertical
column. In finite-volume form the morphodynamic equation takes the form
1 p ( )A
h
Dz
b
Dt
=
_
@A
q
b
v
ds . n
where the integral is taken around the boundary of a cell face on the bed. Here, A
h
is the
horizontal projected area of a cell face, z
b
is the height of a cell-face-centre control point, ds
is an element of cell edge and n is a unit vector normal to the surface. Cell vertices on the
erodible surface are determined by 4th-order interpolation from cell-face-centre control
points in the adjacent cells.
Bed-load transport and consequent morphodynamics is determined by the
relationship between the bed flux vector q
b
and the surface shear stress
w
. For this, the
popular Meyer-Peter and Mller [8] model has been used. This is conveniently written in
dimensionless form:
q
*
= 8
*

*
crit
_ _
3=2
(30)
0.000
0.002
0.004
0.006
0.008
0.010
0.012
0.0 0.1 0.2 0.3 0.4 0.5
r/D
k
/
U
b
2
Smooth wall
Rough wall
Fig. 7 Computed turbulent-kinetic-energy profiles
168 Flow Turbulence Combust (2007) 78:153175
Fig. 9 Evolution of a sand cone in a constant current height contours
Fig. 8 Evolution of a sand cone in a constant current: (a) surface mesh (vertical height exaggerated);
(b) streamlines
Flow Turbulence Combust (2007) 78:153175 169
where
q
*
=
q
b

s 1 ( )gd
3
_ dimensionless bed flux ( )

*
=

w
s 1 ( )gd
dimensionless shear stress = Shields stress ( )
d
*
= d
s 1 ( )g
3
2
_ _
1=3
dimensionless particle diameter ( )
d is the median particle diameter and s is the ratio of particle to fluid densities. The critical
shear stress, at which sediment motion is just initiated, is obtained from the parameter-
isation of Soulsby [10]:

*
crit
=
0:30
1 1:2d
*
0:055 1 exp 0:020d
*
_ _
_
(31)
The direction of q
b
on horizontal surfaces is that of
w
. Modification of (30) and (31) for
slopes (specifically, the replacement of * by an effective stress combining * with the
downslope component of weight, and the reduction in
crit
* in proportion to the component
of gravity normal to the surface) will be reported in more detail in the separate paper on
sandbank morphodynamics.
Although no detailed velocity measurements were made in these experiments some
comparison of the sediment morphodynamics may be undertaken. Figure 11 shows a
comparison between computed and observed planforms after the first half cycle of the tidal
flow over the sand cone. (In subsequent cycles the small volume of sand used in the
experiments tended to break up, making detailed comparison difficult.)
Fig. 10 Evolution of a sand cone in a tidal flow height contours
170 Flow Turbulence Combust (2007) 78:153175
Figures 12 and 13 show the simulated evolution of an initially-Gaussian-shaped sand
mound in a steady current over a real-time period of 2 h. The mound has initial height
0.15 m and a maximum slope of 1 in 10. The sediment has the same properties as the cone
discussed above. The free surface is treated as a rigid lid, with undisturbed depth 0.2 m, and
the bulk velocity is 0.5 m s
1
. The computational domain has the same cross-stream
dimensions as the experimental flume (9 m). Figure 12 shows the time-evolution of the
mound on a non-erodible (smooth concrete) bed, whilst Fig. 13 shows similar development
on a mobile bed composed of the same type of sediment.
For the mobile-bed case, since the quantity of sediment in one region is governed by a
dynamic balance between the rates at which sediment is transported in and out, it is vital
that inflow profiles should be those of a fully-developed flow over surface of the same
roughness to avoid the development of bedforms due to the adjustment of the boundary
Fig. 11 Planform of the sandbank after 1/2 cycle: (a) CFD (delimited by 1% of initial height); (b) laboratory
experiments
Fig. 12 Evolution of a Gaussian sand mound in a constant current non-erodible floor
Flow Turbulence Combust (2007) 78:153175 171
layer. For this reason, the inflow profile is determined by an initial fully-developed-flow
calculation (with the correct bulk velocity and bed roughness) in similar fashion to the pipe-
flow calculations.
The evolution of the Gaussian sand mound is similar to that of the cone; that is, the
mound initially steepens and subsequently develops a crescent shape. The major differences
in sandbank evolution for a mobile, as opposed to fixed, bed are: (1) the development of
significant scour holes; and (2) the tendency of the tail of the sandbank to move very
slightly upstream (because of the imbalance between sediment transport into and out of this
region). Downstream of the sand mound the recirculating flow delves a pair of scour holes
(of depth around 20% of the initial height of the mound). Due to the finite lateral dimensions
of the flume, the flow is accelerated around the sides of the mound and the preferential
removal of sediment from these regions leads to the development of secondary scour holes at
the side (and subsequent deposition to form a more two-dimensional sand bar).
5 Summary and Conclusions
In this paper the analytical wall function of Craft et al. [4], which is based on a continuous
total-viscosity profile, is extended to arbitrarily-rough boundaries. The paper makes the
premise (as in [11]) that this may be accomplished by making the non-dimensional zero-
eddy-viscosity height y

n
a function of

k

s
(the equivalent-sand-roughness height k
s
in wall
units). In contrast to that paper, however, the functional variation with

k

s
is here deduced
more satisfactorily from asymptotic comparison with the log law. A different prescription
for the near-wall dissipation rate is also employed; this is necessary to maintain the close
relationship between the turbulence-energy-based scale u
t
and the stress-derived friction
velocity u in equilibrium boundary layers.
The wall function can be implemented in a cell-centred finite-volume code by relating
wall shear stress to cell-centre velocity by an effective wall viscosity. Formulae for cell-
averaged rates of production and dissipation of turbulent kinetic energy are given, and it is
explained how to use the wall function for Reynolds-stress-transport as well as eddy-
viscosity models. Appendix summarises the relatively minor changes which would be
Fig. 13 Evolution of a Gaussian sand mound in a constant current mobile floor
172 Flow Turbulence Combust (2007) 78:153175
needed to incorporate a streamwise pressure gradient (which leads to the shear stress
varying linearly with height).
The key formulae for implementation of the wall function are:
& Specification of the zero-eddy-viscosity height as a function of roughness equations
(16) and (17);
& Relationship between near-wall velocity and wall stress either equation (8) or
equations (10) and (11);
& Cell-averaged production and dissipation equations (20), (21), (23), (24);
& Value of e at the near-wall node equation (19);
& For Reynolds-stress equations, the cell-averaged production of (wall-oriented) stresses
equation (26).
The wall function has been tested against standard pipe-flow friction formulae. The wall
function is able to reproduce the Darcy friction factor well over a range of realistic
Reynolds numbers and relative roughness.
Use of the wall function in complex flows has been demonstrated by application to the
flow-driven morphodynamics of sand mounds, a situation where accurate prediction of the
surface stress is vital, since it provides the primary motive force for the sand particles.
Acknowledgements The author is indebted to colleagues in the School of Mechanical, Aerospace and Civil
Engineering; in particular, to Dr Tim Craft, Prof. Brian Launder and Dr Alexey Gerasimov for discussions
about the analytical wall-function approach.
Appendix
Inclusion of a streamwise pressure gradient
The formulae in Section 2 were derived assuming that the shear stress was constant with
height and equal to that at the wall. This can be extended to include a linear variation with
height due to a pressure gradient, as in equation (1).
The inclusion of a pressure gradient is not without its difficulties, however. Numerically,
the surface shear stress can become significantly decoupled from the mean velocity at the
near-wall node, causing instability. In experiments, strong streamwise pressure gradients are
known to produce marked deviations from the log law on which equation (2) is based.
Finally, in three dimensions there is no necessity for the pressure gradient and near-wall
mean velocity to be aligned. Some of these issues are discussed at the end of this Appendix.
Assume that
t = t
w

dP
dx
y = rn
total
@U
@y
(32)
with the total viscosity given by equation (2). In 3 dimensions dP/dx is the component of
pressure gradient in the direction of the near-surface flow. Integration yields a
dimensionless relationship between mean velocity, wall shear stress and pressure gradient:
U
u

=

w
u
2

_
y

0
dy

1 max 0; .(y

3
)
_ _

3
u
3

dP
dx
_
y

0
y

d y

1 max 0; .(y

3
)
_ _
Flow Turbulence Combust (2007) 78:153175 173
i.e.,

=

TT
fac


GG
fac
(33)
where

=
U
u
t
;

T =
t
w
ru
2
t
;

G =
n
ru
3
t
dP
dx
(34)
and the non-dimensional integral factors arising from shear-stress and pressure-gradient
contributions are
T
fac
=
y

; if y

_ y

3
y

30

1
.
ln
1. y

3
( )
1. y

30
y

3 ( )
_ _
; if y

_ y

3
_
_
_
(35)
G
fac
=
1
2
y

2; if y

_ y

3
1
2
y

30
2
1
.
y

30
_ _

1
.
2
.y

3
1
_ _
ln
1. y

3
( )
1. y

30
y

3 ( )
_ _
; if y

_ y

3
_
_
_
(36)
The wall shear stress can then be found from values at the near-wall node P:

T =

GG
fac;P
T
fac;P
(37)
The wall-shear-stress/near-wall-velocity relationship can also be written in terms of an
effective wall viscosity 3
eff,wall
and pressure-gradient correction
G
:
t
w
= rn
eff ;wall
U
p
y
p
t
G
(38)
where:
n
eff ;wall
= n
y

P
T
fac;P
; t
G
= ru
2
t

G
G
fac;P
T
fac;P
(39)
The cell-averaged rate of production of turbulent kinetic energy is
P
k ( )
av
=
1
D
_
D
0
uv
@U
@y
_ _
dy =
1
D
_
D
y
n0
n
eff
n
_ _
@U
@y
_ _
2
dy
and, after considerable algebra, this can be evaluated (assuming D > y
n0
) as
P
k ( )
av
=
u
3

T
2
I
1
I
2
( )
2

G
1
.
I
0
.y

3
2
_ _
I
1
.y

3
1
_ _
I
2
_

G
2
1
.
2
I
1
2.y

3
3
_ _
I
0
.y

3
1
_ _
.y

3
3
_ _
I
1
.y

3
1
_ _
2
I
2
_ _
_

_
_

_
(40)
174 Flow Turbulence Combust (2007) 78:153175
where the integrals are
I
1
=
_
s
2
s
1
s ds =
1
2
s
2
2
s
2
1
_ _
; I
0
=
_
s
2
s
1
ds = s
2
s
1
;
I
1
=
_
s
2
s
1
1
s
ds = ln
s
2
s
1
; I
2
=
_
s
2
s
1
1
s
2
ds =
1
s
2

1
s
1
_ _
and the limits of integration are the non-dimensional mixing lengths
s
1
= 1 k y

n0
y

n
_ _
s
2
= 1 k

D

n
_ _
No modification is required for the cell-averaged dissipation rate.
In summary, a linear variation of shear stress with height due to a streamwise pressure
gradient can be accommodated in the wall function by:
(1) An additional contribution
G
to wall stress (equation (38));
(2) A modified cell-averaged-production term (equation (40)).
The modifications come with two warnings. Firstly, in three dimensions there is no
reason why the pressure gradient should be flow-aligned; the above expressions use the
streamwise component of pressure gradient, but this will be discontinuous (in
magnitude and direction) near points where U=0. Secondly, a sufficiently strong
pressure gradient may yield a wall shear stress in direct opposition to the near-wall
velocity; this can have a detrimental and destabilising effect on the numerical procedure
for sediment morphodynamics.
References
1. AGARD: A selection of test cases for the validation of large-eddy simulations of turbulent flows.
Technical Report, AGARD-AR-345, NATO (1998)
2. Chieng, C.C., Launder, B.E.: On the calculation of turbulent heat transport downstream from an abrupt
pipe expansion. Numer. Heat Transf. 3, 189207 (1980)
3. Craft, T.J., Gerasimov, A.V., Iacovides, H., Launder, B.E.: Progress in the generalisation of wall-function
treatments. Int. J. Heat Fluid Flow 23, 148160 (2002)
4. Craft, T.J., Gerasimov, A.V., Iacovides, H., Kidger, J.W., Launder, B.E.: The negatively buoyant
turbulent wall jet: performance of alternative options in RANS modelling. Int. J. Heat Fluid Flow 25,
809823 (2004)
5. Launder, B.E., Spalding, D.B.: The numerical computation of turbulent flows. Comput. Methods Appl.
Mech. Eng. 3, 269289 (1974)
6. Launder, B.E., Reece, G.J., Rodi, W.: Progress in the development of a Reynolds-stress turbulence
closure. J. Fluid Mech. 68, 537566 (1975)
7. Leschziner, M.A., Apsley, D.D.: Investigation of advanced turbulence models for the flow in a generic
wing-body junction. Flow Turbul. Combust. 67, 2555 (2001)
8. Meyer-Peter, E., Mller, R.: Formulas for bed-load transport. Rept 2nd Meeting Int. Assoc. Hydraul.
Struct. Res., pp. 3964, Stockholm, Sweden (1948)
9. Schlichting, H., Gersten, K.: Boundary-layer Theory (8th Ed.), Springer, Berlin Heidelberg New York
(2000)
10. Soulsby, R.: Marine Sands. Thomas Telford, London, UK (1997)
11. Suga, K., Craft, T.J., Iacovides, H.: An analytical wallfunction for burbulent flows and heat transfer
over rough walls, Int. J. Heat Fluid Flow, 27, 852866 (2006)
Flow Turbulence Combust (2007) 78:153175 175

You might also like