Вы находитесь на странице: 1из 8

The Fifth International Symposium on Computational Wind Engineering (CWE2010)

Chapel Hill, North Carolina, USA May 23-27, 2010

Appropriate boundary conditions for computational wind


engineering models revisited
S. E. Norris a, P. J. Richards b
a

Mechanical Engineering, University of Auckland, New Zealand, s.norris@auckland.ac.nz


Mechanical Engineering, University of Auckland, New Zealand, pj.richards@auckland.ac.nz

ABSTRACT: The onset flow for a wind-engineering model can be idealized as a horizontallyhomogeneous turbulent boundary layer, with the flow being driven by a shear stress at the top
boundary. The inlet profiles and boundary conditions appropriate for modeling the flow using the
k-, k- and LRR QI turbulence models are derived. Means for their application within the
commercial CFD code CFX 12.0 are given.
Problems with excessive turbulence generation near the ground and the over-prediction of
stagnation pressures are discussed and possible solutions proposed.
1 INTRODUCTION
At the first Computational Wind Engineering Conference Richards and Hoxey (1993) recommended modeling the atmospheric surface layer as a horizontally-homogeneous turbulent boundary layer (HHTBL), which is one with constant properties in directions tangential to the ground
and hence the only variation is along the vertical axis. Since the pressure is constant the flow is
driven by a shear stress at the upper surface of the layer, and this is constant through the layer,
equaling the shear stress at the wall. Velocity and turbulence property profiles, together with the
associated boundary conditions, were proposed for CFD studies using the standard k- turbulence
model (Launder and Spalding, 1974) and were shown to satisfy horizontal homogeneity provided
the various constants satisfied particular relationships.
In this paper we further provide profiles and boundary conditions for HHTBLs modeled using
the k- turbulence model (Wilcox, 1993), and the Quasi-Isotropic LRR Reynolds-stress transport model (Launder et al., 1975). The implementation of these boundary conditions within the
commercial CFD package CFX 12.0 is discussed.
For a number of years eddy-viscosity turbulence models have dominated industry use of CFD;
as noted by Bradshaw (1999) it is so obvious that stress-transport models are more realistic in
principle than eddy viscosity models that the improvements they give are very disappointing.
However, it is shown that the use of an eddy-viscosity based model of a HHTBL can lead to spurious over-prediction of the pressure on windward faces of bodies in the boundary layer, whereas
a Reynolds-stress transport model does not exhibit this undesirable behavior, suggesting that use
of the latter deserves consideration.
2 BOUNDARY CONDITIONS FOR A HORIZONTALLY HOMOGENIOUS BOUNDARY
LAYER
Richards and Hoxey (1993) modeled a HHTBL by proposing velocity and turbulence property
profiles, together with the associated boundary conditions, for the standard k- turbulence model

The Fifth International Symposium on Computational Wind Engineering (CWE2010)


Chapel Hill, North Carolina, USA May 23-27, 2010

and showing that these satisfied horizontal homogeneity provided the model constants satisfied
particular relationships.
An alternative approach is to derive the profiles directly from the conservation and equilibrium equations associated with a particular turbulence model for a HHTBL. For example with
the standard k- model and a rough wall with U=0 at z=z0 these yield:

u* ln( z / z 0 )

U=

(C 2 C 1 ) C

, k=

u*2
C

, =

u*3
z (C 2 C 1 ) C

(1)

where u* is the friction velocity associated with the constant shear stress

xz = u*2

(2)

Comparison of Equation 1 with the more familiar form

u* ln( z / z 0 )
u2
u3
(3)
, k= * , = *

z
C
shows that the turbulence model has chosen its own value for von Krmns constant , such that
the usual k- turbulence model constants C1=1.44, C2= 1.92, C= 0.09 and =1.3 give
U=

(C 2 C 1 )

C = 0.4237 .

(4)

The form of Equation 3 only differs from that given by Richards and Hoxey in terms of the definition of the height at which the velocity is zero. The requirement that the shear stress xz is constant requires that the eddy viscosity varies linearly with the height
1
1
k2
= u* z = C 4 k 2 z
(5)

To implement such a profile the shear stress is imposed at the upper boundary of the domain, a
zero flux condition is set for k, and the flux of across the boundary is prescribed as,

T = C

u 4
T d
(6)
= *
z
dz
A similar analysis for the k- turbulence model (Wilcox, 1993) yields essentially the same
profiles for U and k, however with the standard constants C==0.09, =5/9, =0.075 and =2
the effective von Krmns constant is given by

( )

= 0.408

(7)

The specific dissipation rate, , profile is given by


u*
(8)
z
To implement the model similar boundary conditions are imposed at the upper boundary, with
the flux of being prescribed as
=

u*2
T d
=
dz
z

(9)

The Fifth International Symposium on Computational Wind Engineering (CWE2010)


Chapel Hill, North Carolina, USA May 23-27, 2010

The HHTBL can also be analyzed using the Quasi-Isotropic Reynolds stress transport model
of Launder et al. 1975 (LRR QI). For this model the Reynolds stresses are constant across the
layer and as implemented in CFX 12.0, with CS1=1.8 and C2=0.4, are
k=

1
2

165C S21
uu + vv + ww =
2
44C S1 + 10 60C 2 30C 2

2 2
u* = 2.975u*2

(10)

uu = 0.882 k , vv = 0.620 k , ww = 0.498k , uw = u*2 , uv = vw = 0 (11)


The LRR QI model also models , the profile being the same as Equation 3 and the flux through
the top boundary being
C , RS k 2
T d
(12)
=
, RS z
, RS dz
with C,RS=.1152 and ,RS=1.1. The effective value for von Krmns constant is related to
C1,RS=1.45, C2,RS= 1.9 and the other constants by
=

(C

2 , RS

C 1, RS ) , RS

2.975 3 C , RS

= 0.404

(13)

To implement these models in CFX 12.0 the boundary condition at the top of the domain can
be imposed as a free-slip wall, with a specified shear stress and a flux term for or as given in
Equations 6, 9 or 12 as appropriate. All other turbulence scalars have a zero gradient boundary
condition. Previous versions of CFX did not allow this shear stress boundary condition, but the
shear stress and gradient could be imposed by creating a sub-domain one cell thick at the top
boundary and prescribing a volume momentum source to drive the flow, and a volume sink term
for or .
For all three turbulence models a rough wall boundary condition needs to be imposed at the
ground. CFX models rough walls using a roughness length scale that is based on an equivalent
sand grain roughness zR, rather than the roughness length z0 used in Equation 3 above. The near
wall velocity profile is assume to be
z*

+C.
(14)
ln
1 + 0.3k *
Here z* and k* are the distance from the wall and the sand grain roughness expressed as Reynolds numbers,
U=

u*

u * z
u z
and k* = * R .
(15)

In most wind engineering cases k*>>1 and so by inspection it can be seen that the sand grain
roughness used in CFX is related to the roughness height z0 by
e C z 0
.
(16)
zR =
0.3
For = 0.41 and C = 5.2 this gives zR = 28.1 z0. Note that the location of the ground has to be
shifted vertically by one half of the sand grain roughness, so the bottom of the domain is located
at z = 0.5 zR. This relationship and the associated difficulties with implementing it through wall
functions in commercial codes have been discussed in detail by Blocken et al. (2007).
z* =

The Fifth International Symposium on Computational Wind Engineering (CWE2010)


Chapel Hill, North Carolina, USA May 23-27, 2010

3 TESTING THE BOUNDARY LAYER PROFILES


The boundary condition profiles derived above are for a horizontally-homogeneous boundary
layer. Therefore they (and the CFD code) can be tested by computing the flow through an empty
domain. The analytic profiles are specified at the inlet, the shear stress and or flux condition
imposed at the top boundary, and a rough wall specified at the bottom of the domain. Ideally the
outlet profiles should equal those at the inlet, and the value of the velocity and the turbulence
scalars should remain constant along planes of constant elevation.
This was tested using CFX 12.0 on a two-dimensional domain of 300 m length and 30 m
height, using a regular mesh of 30060 cells. The prescribed inlet flow had a velocity of 10 m/s
at 10 m elevation, with a roughness length of z0 = 0.02 m. The inlet and outlet profiles of U and k
calculated using the k- and k- turbulence models are plotted in Figures 1 and 2, whilst the profiles of U, k and the Reynolds stresses calculated using the LRR QI model are plotted in Figures
3 and 4. There is reasonable agreement between the inlet and outlet profiles of velocity and turbulence kinetic energy, excepting near the ground.
30

Inlet
Outlet

30

Inlet
Outlet
10
z (m)

20

z (m)

z (m)

20

Inlet
Outlet

10

10
1
0

6
8
U (m/s)

10

12

12

0.5

U (m/s)

1
1.5
k (m2/s2)

Figure 1. Profiles of velocity and turbulence kinetic energy calculated using the k- turbulence model.
30

Inlet
Outlet

30

Inlet
Outlet
10

z (m)

20

z (m)

z (m)

20

Inlet
Outlet

10

10
1
0

8
U (m/s)

12

8
U (m/s)

12

0.5

1
1.5
k (m2/s2)

Figure 2. Profiles of velocity and turbulence kinetic energy calculated using the k- turbulence model.

The profiles of the Reynolds stresses calculated using the LRR QI Reynolds stress model
(Figure 4) are again reasonably constant through the domain. Interestingly the vertical velocity
profile (shown in Figure 3) has a perturbation of a wavelength equal to the cell height, which is
also reflected in the pressure field (not shown). This odd-even checker boarding of the velocity
and pressure fields is a known issue with the implementation of Reynolds-stress transport models
in collocated finite volume solvers (Leschziner and Lien, 2002), the solution to which is the use
of an interpolation scheme incorporating forth-order smoothing, somewhat analogous to the
Rhie-Chow momentum interpolation scheme (Rhie and Chow, 1983).

The Fifth International Symposium on Computational Wind Engineering (CWE2010)


Chapel Hill, North Carolina, USA May 23-27, 2010

30

Inlet
Outlet

30

Inlet
Outlet
10

z (m)

20

z (m)

z (m)

20

10

10
1
0

12

U (m/s)

12

-0.06

U (m/s)

-0.04
-0.02
W (m/s)

Figure 3. Velocity profiles for a HHTBL calculated using the LRR QI model.
30

uu Inlet
uu Outlet
vv Inlet
vv Outlet
ww Inlet
ww Outlet

20

10

10

0.25
0.5
0.75
uu/k, vv/k, ww/k

Inlet
Outlet

20

z (m)

z (m)

20

30

Inlet
Outlet

z (m)

30

10

0.2

0.4

0.6

-uw/k

0.5
1
2 2
k (m /s )

1.5

Figure 4. Profiles of Reynolds stresses and turbulence kinetic energy calculated using the LRR QI model.

4 THE TURBULENCE KINETIC ENERGY ANOMALY


For a HHTBL, modeled using the k- turbulence model, the analytic solution for the layer is
given above in Equations 3. For such a layer k is constant, and so the turbulence production Pk
must equal the dissipation throughout the flow, since there is no vertical component of diffusion
or convection. Using the profile for velocity given in Equation 3, its derivative with respect to
height, the definition of T in Equation 5, and comparing with the profile of in Equation 3 we
see that the flow is indeed in equilibrium,
2

u 3
dU
Pk = T
= * =
z
dz

(17)

Whilst this is true for the analytic solution, it is not necessarily the case for its discrete form.
Hargreaves and Wright (2007) also noted a spike in turbulence kinetic energy in the second cell
above the ground and suggested it was a feature of the k- model. However, it appears that this
phenomenon is more related to the discretisation process used in calculating the production term
than the particular turbulence model. Conventionally Pk is calculated using the cell centered value for T, and cell centered differences for the gradient of U, whilst the shear stresses in the momentum equations are calculated using face centered differences.
Consider a finite volume cell P, in a mesh of constant spacing z. For a HHTBL layer xz is
constant, and so the stresses at the north and south faces of the cell can be used to give the velocities in the neighboring cells N and S,

The Fifth International Symposium on Computational Wind Engineering (CWE2010)


Chapel Hill, North Carolina, USA May 23-27, 2010

xz ,n = T ,n

du
dz

z u N u P

= u * z +

2 z

uN = uP +

u * z
( z + z 2 )

(18)

u * z
z u P u S
du

(19)
= u * z
uS = uP

2 z
( z z 2 )
dz s

These values can be substituted into the cell centered discretisation for dU/dz to find the estimate
for the turbulence production,
xz , s = T , s

u*3
1
u N uS
dU
, =

(20)
Pk = T

T
2
z
(
)
1

2
z
z

dz
2z

From this it can be seen that Pk does not equal as is required for equilibrium, but instead
exceeds it by a factor which increases as the ratio z/z increases. For a constant mesh spacing
is at a maximum approaching the wall, as will k. This is shown in Figure 5 (a), for a HHTBL calculated using the ALE code (Norris et al., 2010). For the near wall cell the values of Pk and are
fixed by the wall function, but for the second interior node where k is calculated using the conservation equation, k reaches a local maxima.
To prevent this anomalous spike in the value of k the production may be recast into a form
that uses the shear stresses at the north and south cell faces,
2
2

T ,n u N u P + T , s u P u S

z
z

Since xz is a constant across the boundary layer (and using Equation 5 for T), we have

xz2 ,n + xz2 , s
xz2
1

=
Pk =
T
2 T , P
2 T , P

(21)

2 2 u*4
u 3
(22)
= * =
2 u * z
z
and so the boundary layer can be seen to be in equilibrium.
This is demonstrated in Figure 5 (a). As noted above, the flow calculated using the cell centre
discretisation for Pk given in Equation 20 has an anomalous spike in the value of k near the wall.
By switching to the cell face discretisation given in Equation 21 the value of k is constant across
the layer, and no spike is seen.
Pk =

30

20

Eq 20
Eq 21

15
z (m)

z (m)

20
10

10
5
k-
LRR
0
1.3

1.4
1.5
k (m2/s2)

(a)

1.6

0
1
2
3
CP on Windward Centreline

(b)

Figure 5. Anomalies in the HHTBL. (a) Comparison of turbulence kinetic energy in a HHTBL calculated using a
cell centered discretisation for production (Equation 20) or a cell face discretisation (Equation 21). (b) Error in
windward face centerline CP calculated using k- model, compared with values calculated using LRR QI.

The Fifth International Symposium on Computational Wind Engineering (CWE2010)


Chapel Hill, North Carolina, USA May 23-27, 2010

5 THE EDDY VISCOSITY STAGNATION PRESSURE ANOMALY


The k- and k- models of a HHTBL are based on the k- model of Prandtl (Launder and Spalding, 1972), with the eddy viscosity increasing linearly with the distance from the wall (Equation
5). For wind engineering flow the values of k and z can be quite large, resulting in an excessively
large eddy viscosity. For instance, for the representative flow calculated above (10 m/s at 10 m
height, with z0 = 0.02 m), the turbulence kinetic energy is 1.45 m2/s2, and so the eddy viscosity is
T = 0.32 z kg/m.s.

(23)

At a height of 10 m this yields an effective viscosity of 3.2 kg/m.s, giving an effective Reynolds
number of 3.7 for the flow around a 1 m diameter cylinder, compared to the true value of
6.7105.
This high effective viscosity causes the flow to act in a manner akin to a Stokes flow, rather
than the high Reynolds number of the actual flow. Aside from incorrect flow structures, this may
lead to the pressure being over predicted. The total pressure defined as
ptot = p + 12 U 2

(24)

For a high Reynolds number flow the flow is expected to obey Bernoullis law, and the total
pressure should remain constant along a streamline (or possibly dropping through viscous effects). However, as noted by Issa (1995), for low Reynolds number flows the total pressure can
rise at a stagnation point. This has repercussions for wind-engineering flows calculated using an
eddy viscosity based turbulence model with a HHTBL upstream profile. Although the physical
flow has a high Reynolds number, and so the pressure at a stagnation point should equal the total
pressure of the upstream flow, the modeled flow has a low effective Reynolds number, and so
the pressure calculated at the stagnation point can exceed the total pressure of the onset flow.
Since the pressure forces typically dominate in wind loads, this means that the eddy viscosity
based models may over-predict wind loadings.
This effect is demonstrated in Figure 5 (b), which plots the pressure on the windward centerline of a chimney in a HHTBL calculated using the k- turbulence model. The pressures are nondimensionalized with respect to the dynamic pressure of the onset flow at that height, so the value of CP should not exceed 1. However the use of the eddy viscosity model causes the prediction
of excessive pressures at the stagnation point and values of CP reach approximately 1.8.

Figure 6. Contours of total pressure on the symmetry plane upstream of a chimney in a HHTBL. Flow is from left to
right. On the left the turbulence is modeled using the k- eddy viscosity model. On the right the flow is modeled using the LRR QI Reynolds stress model.

The Fifth International Symposium on Computational Wind Engineering (CWE2010)


Chapel Hill, North Carolina, USA May 23-27, 2010

The Reynolds-stress transport models do not use an eddy viscosity to model the constant
shear stress through the log-law boundary layer, instead modeling the Reynolds-stresses directly.
Therefore Reynolds-stress models would not be expected to exhibit this spurious behavior. Figure 5 (b) confirms this supposition, and for the chimney flow calculated using the LRR QI model
the pressure coefficient is 1 for most of the height of the chimney as expected. Figure 6 also illustrates this point where on the left with the k- model the total pressure increases as the chimney
is approached whereas on the right with the LRR QI model the contours remain horizontal up to
the chimney.
6 CONCLUSIONS
The onset flow for a wind-engineering model may be idealized as a horizontally-homogeneous
turbulent boundary layer, with the flow within the computational domain being driven by a shear
stress at the top boundary. The inlet profiles and free-stream and ground boundary conditions
appropriate for modeling the flow using the k-, k- and LRR QI turbulence models have been
derived, and their application within the commercial CFD code CFX 12.0 have been described.
Numerical calculations of a HHTBL using the k- turbulence model exhibit a characteristic
maxima in the value of k near the ground, rather than the constant value across the layer predicted by the analytic theory. This is shown to be due to the discretisation of the turbulence production, and an alternative discretisation is offered that does not exhibit this behavior.
The use of an eddy-viscosity turbulence model is shown to cause the over-prediction of pressure on the windward face for the flow around a chimney, bringing into question the validity of
using such a model in wind engineering flows. A Reynolds-stress transport model does not exhibit this spurious behavior, suggesting that these models should be explored for use in future wind
engineering computations.
7 REFERENCES
Bradshaw, P., 1999. The best turbulence models for engineers, in: Modeling of complex turbulent flows, 9-28. eds.
Salas, M.D., Hefner, J.N. and Sakell, L., Kluwer, Dordrecht.
Blocken, B., Stathopoulos, T. and Carmeliet, J., 2007. CFD simulation of the atmospheric boundary layer: wall
function problems. Atmospheric Environment, 41, 238-252.
Hargreaves, D.M. and Wright, N.G., 2007. On the use of the k- model in commercial CFD software to model the
neutral atmospheric boundary layer. Journal of Wind Engineering and Industrial Aerodynamics, 95, 355-369.
Issa, R.I., 1995. Rise of total pressure in frictional flow. AIAA Journal, 33, 772-774.
Launder, B. E., Reece, G.J. and Rodi, W., 1975. Progress in the development of a Reynolds-stress turbulence closure. Journal of Fluid Mechanics, 68, 537-566.
Launder, B.E. and Spalding, D.B., 1972. Lectures in mathematical models of turbulence, Academic, London.
Launder, B.E. and Spalding, D.B., 1974. The numerical computation of turbulent flows. Computer Methods in Applied Mechanics and Engineering, 3 269-289.
Leschziner, M.A. and Lien, F.-S., 2002. Numerical aspects of applying second-moment closure to complex flows,
in: Closure Strategies for Turbulent and Transitional Flows, 153-187. eds. Launder, B.E. and Sandham, N.D.,
Cambridge.
Norris, S.E., Were, C.J., Richards, P.J. and Mallinson, G.D., 2010. A Voronoi based ALE solver for the calculation
of incompressible flow on deforming unstructured meshes. International Journal for Numerical Methods in Fluids, DOI:10.1002/fld.2234.
Rhie, C.M. and Chow, W.L., 1983. Numerical study of the turbulent flow past an airfoil with trailing edge separation. AIAA Journal, 21 1525-1532.
Richards, P.J. and Hoxey, R.P., 1993. Appropriate boundary conditions for computational wind engineering models
using the k- turbulence model. Journal of Wind Engineering and Industrial Aerodynamics, 46 & 47, 145-153.
Wilcox, D.C., 1993. Turbulence modeling for CFD, DCW Industries, La Canada, California.

Вам также может понравиться