Professional Documents
Culture Documents
Abstract
This paper is concerned with the major issues considered to be important for successful
application of computational #uid dynamics to environmental #ows. Among the issues of
primary concern in the present paper are meshing and turbulence modeling. As regards
meshing, we propose an approach that employs unstructured meshes in conjunction with
solution-adaptive mesh re"nement. Examples are presented to illustrate the e$cacy of the
unstructured mesh-based approach. The issue of turbulence modeling is discussed at length,
inasmuch as turbulence modeling determines the "delity of computational results for environ-
mental applications. Emphasis is laid upon the capability of engineering turbulence models to
capture the salient features of environmental #ows such as atmospheric boundary layer over
a smooth terrain and #ows around blu! bodies. 1999 Published by Elsevier Science Ltd. All
rights reserved.
Keywords: Unstructured mesh; Turbulence models; Blu! bodies; Separation; Vortex shedding
1. Introduction
The topics that fall in the realm of modern environmental #uid mechanics are
diverse. The examples are; indoor air #ows laden with contaminants, #ows around
in-land and o!-shore structures including buildings, bridges, stacks, towers, oil rigs,
and e%uents dispersion into the environment including rivers, lakes, estuaries, coastal
waters, and atmosphere. Among this broad range of topics, wind climate in urban
areas and its rami"cations to environmental, structural and architectural engineering
practices are of major interest to city planners, designers and builders. The whole issue
of wind engineering has numerous components to it, ranging from urban climatology
* Corresponding author.
E-mail address: sek@#uent.com (S.E. Kim)
0167-6105/99/$ - see front matter 1999 Published by Elsevier Science Ltd. All rights reserved.
PII: S 0 1 6 7 - 6 1 0 5 ( 9 9 ) 0 0 0 1 3 - 6
146 S.-E. Kim, F. Boysan / J. Wind Eng. Ind. Aerodyn. 81 (1999) 145}158
(LES) in lieu of the increasingly important role it plays these days in blu!-body
aerodynamics.
For the computations illustrated here, the commercial CFD software FLUENT
[1}3] was used.
Any numerical procedure to compute #uid #ows in the context of CFD requires
discretization of the #ow domain under consideration. Meshing the #ow domain
around an isolated building hardly poses a grand challenge. However, it is not an easy
task to design meshes for buildings groups or urban structures involving complex
topography. Especially when employing structured mesh, complex geometries with all
the signi"cant details frequently encountered in industrial applications often make
adequate meshing very di$cult, if not impossible, limiting the usability of CFD.
Furthermore, employing structured meshes with local clustering to resolve large
solution gradients often results in a considerable portion of the mesh being wasted in
areas where "ne mesh resolution is not really needed. These predicaments encoun-
tered in using structured meshes for complex geometries are carried over to building
aerodynamics.
In recent years, unstructured mesh technology has attracted a great deal of atten-
tion from the CFD community and has been quite successfully applied to industrial
#ows. It allows one to employ computational cell shapes of arbitrary topology
including quadrilaterals, hexahedra, triangles, tetrahedra, prisms, and combinations
of all these. One immediate advantage o!ered by unstructured meshes over structured
meshes is the #exibility in dealing with complicated #ow con"gurations which directly
translates into dramatic reduction in the amount of time to generate meshes.
Fig. 1 illustrates two unstructured hexahedral meshes generated for typical building
groups using commercial preprocessors. These meshes could be generated in a small
fraction of time that would have taken if block-structured meshes were adopted. The
signi"cant time saving is possible only because unstructured meshes are much more
amenable to automated meshing.
Another advantage of using unstructured meshes is that they naturally provide
a convenient framework for local mesh adaptation. This approach has a great
potential to bene"t CFD predictions of building aerodynamics, especially #ows over
building groups with a wide range of length scales to be resolved. The mesh adapta-
tion a!ects the mesh locally, yielding a mesh re"ned only where necessary. In the
adaptations using hanging nodes (locally embedded mesh), those computational cells
which satisfy a prescribed adaptation criterion are subdivided into smaller cells by
using midpoint nodes on edges and/or at cell and face centers. At the simplest level,
a region adaptation can be employed where the cells inside a prescribed region can be
marked for re"nement. More sophisticated adaptations make use of numerical solu-
tions themselves, yielding `solution-adapteda meshes. In the solution-adaptive re"ne-
ment, any solution variables or quantities derived therefrom, or the gradients of such
quantities can be used to specify adaptation criteria.
148 S.-E. Kim, F. Boysan / J. Wind Eng. Ind. Aerodyn. 81 (1999) 145}158
Fig. 2. Examples of mesh adaptation for a triangular bump mounted on a wall left: coarse mesh and
predicted streamlines right: adapted mesh and predicted streamline.
As an example of local mesh adaptation, let us consider here the #ow over
a triangular bump [4]. The incoming boundary layer #ow separates at the apex of the
ridge, reattaching the bottom plane at a distance in the downstream. A strong shear
layer is formed in the downstream of the apex between the recirculation region and
the freestream. From the #ow physics' point of view, the ideal computational mesh
would be the one that can resolve the incoming boundary layer and the recirculation
region, the shear layer and the redeveloping boundary layer after the reattachment, all
at the same time.
Fig. 2 shows two meshes (original and re"ned) and the corresponding streamlines
computed using the two meshes. The "ne mesh was generated using the original
coarse mesh by re"ning the computational cells where the vorticity magnitude is
larger than a prescribed value. We made use of the fact that vorticity magnitude is
usually large in boundary layers and wakes, in order to demarcate the region where
viscous e!ects are important, i.e. the boundary layer and the recirculation region in
this example. The reattachment points predicted with these two meshes are 8.4 and 9.0
bump heights, respectively. Still underpredicting the size of the recirculation (note that
the standard k}e model was used here), the result obtained with the locally re"ned
mesh matches the experimental data [4] (9.8 bump heights) quite closely.
S.-E. Kim, F. Boysan / J. Wind Eng. Ind. Aerodyn. 81 (1999) 145}158 149
Fig. 3. Flow over the curved two-dimensional hill } predictions using four di!erent turbulence models,
top-left: mesh, top-right: separation bubble, bottom-left: pressure distributions, bottom-right: skin-friction
distributions.
blu!-body aerodynamics. A typical symptom observed when using the standard k}e
model is the spuriously large generation of k (turbulent kinetic energy), which results
in, the framework of eddy-viscosity model, too large turbulent viscosity in the vicinity
of stagnation points. The spuriously large turbulent kinetic energy is convected
downstream, eventually a!ecting the mean #ow, suppressing or reducing #ow separ-
ation in the downstream. Fig. 4 aptly shows what typically happens in the vicinity of
the stagnation point of a blunt body when the standard k}e model is employed. This
con"guration was studied experimentally and numerically by Djilali et al. [16]. The
Reynolds number based on the body height is about Re "50 000. The "lled contours
in Fig. 4 depict the rate of production of turbulent kinetic energy given by the
standard k}e model (left) and the RSTM (right), respectively. It is clearly seen that the
standard k}e yields much larger generation of k than the RSTM that supposedly
predicts the production of k correctly.
It is worthwhile to examine this matter in some more detail. The exact expression
for the production of k includes a contribution from normal straining (*;/*x in this
example) which dominates the budget of k near the stagnation point. It can be written
as
*;
P"!(u!v) , (1)
I *x
152 S.-E. Kim, F. Boysan / J. Wind Eng. Ind. Aerodyn. 81 (1999) 145}158
Fig. 4. Production of turbulent kinetic energy over the front face a blunt rectangular section, left: standard
k}e model, right: RSTM.
where P is the production of k due to the normal straining, u and v are the #uctuating
I
velocity components in the x and y directions, respectively.
Note that this term does not require modeling when the RSTM model is employed
because the normal Reynolds stresses are known. However, in k}e models, the normal
stresses needs modeling (using the Boussinesq's hypothesis), and the modeled term
"nally reads
*;
P"4l . (2)
I R *x
Fig. 5. Predicted separation bubbles over a blunt rectangular section, left: standard k}e model, right:
RSTM.
o!er to blu!-body aerodynamics. It seems that this question has been rarely raised,
particularly in the building aerodynamics community. We employed here the RNG
k}e model [8] and the realizable k}e model [10] to see how they work for the blunt
rectangular body. Still adopting the isotropic eddy viscosity, the RNG k}e model and
the realizable k}e model cannot fundamentally resolve the predicament of anomal-
ously large generation of k. But both models have certain features in them which
greatly alleviate the predicament. In the RNG-based k}e model, an additional term in
the e equation augments the production of e in response to the rapid strain in the
neighborhood of the stagnation point, reducing spuriously high level of turbulent
kinetic energy. The realizable k}e model, on the other hand, is based on the idea that
the model constants, which were calibrated for simple equilibrium #ows, need to be
sensitized to the mean #ow (it rate of deformation) to ensure `realizabilitya of the
Reynolds stresses, i.e., positivity of normal Reynolds stresses and Schwarz' inequality
for shear stresses [10].
In Fig. 6 are shown the pressure and skin-friction distributions predicted by the
three di!erent k}e models. It is seen that the RNG k}e model and the realizable k}e
model signi"cantly improve the results. In particular, the realizable k}e model repro-
duces both the pressure and skin friction distributions remarkably well. This example
clearly demonstrates that there is a lot, indeed, that more recently proposed models
can o!er.
The #ows around actual buildings and structures in cities are far more complicated
than those in the examples shown thus far. In addition to all the features of blu!-body
aerodynamics, buildings are immersed in atmospheric boundary layer, being sub-
jected to ground or surfaces e!ects. The interaction between incoming atmospheric
boundary layer and buildings produces a highly complex #ow pattern in the up-
stream, around, and downstream of the body and on the ground surface. Fig. 7 depicts
the computed pressure distribution on the surface of a surface-mounted cube [17]. It
quite closely reproduces the characteristic features of the surface pressure distribution
observed in the experiment.
The #ows around buildings and structures in urban areas are inherently transient.
Besides atmospheric turbulence occurring in the form of gusts, buildings and struc-
tures themselves are responsible for contributing to #ow unsteadiness. In many #ows
around buildings, organized, large-scale motions are observed. One example is the
alternate shedding of vortices behind blu! bodies with a clearly identi"able frequency.
The periodic shedding of vortices is often mainly responsible for #uctuating wind
154 S.-E. Kim, F. Boysan / J. Wind Eng. Ind. Aerodyn. 81 (1999) 145}158
Fig. 6. Comparison of three di!erent k}e turbulence models for a blunt rectangular section, left:
C predictions, right: C predictions.
D
Fig. 7. Surface pressure distribution on a surface-mounted cube and the bottom wall.
loading on buildings and structures. Since the periodic vortex shedding has non-
turbulent origin (yet being a!ected by turbulence), numerical solutions based on the
ensemble-averaged Navier}Stokes equations with proper modeling of turbulence
should be able to reproduce it. Surveying various computational studies [18], how-
ever, indicates that the attempts to numerically simulate vortex shedding have not
been entirely successful.
S.-E. Kim, F. Boysan / J. Wind Eng. Ind. Aerodyn. 81 (1999) 145}158 155
Fig. 8. Periodic vortex-shedding in turbulent #ow over a square cylinder left: time history of C , right:
"
contours of instantaneous e!ective viscosity.
Here we present computational results for the #ow around a square cylinder
experimentally investigated by Lyn [19]. The Reynolds number (Re ) is approxim-
"
ately 21 400, based on the freestream velocity, ; , and the width of the square, D. The
experiment shows that the Strouhal number ( fD/;, where f is the shedding frequency)
is in the range of 0.13}0.14. The time-mean drag has not been reported, but Rodi [18]
quotes the range of 2.05}2.23 for the Reynolds number of 22 000.
A quadrilateral mesh with 13 600 cells was used for the computations. A second-
order accurate temporal discretization scheme was used along with a second-order
upwind scheme for convection term. The time step in dimensionless unit (*t; /D)
chosen for the computations is 0.04. The computations were made with the RNG k}e
model. Fig. 8 shows the results. As can be deducted from the time history of the drag
coe$cient (the left "gure in Fig. 8), the Strouhal number predicted by the computation
is approximately 0.135, falling in the range observed in the experiment [19]. The mean
drag is also found to agree fairly well with the values quoted by Rodi [18].
In cities where building groups are commonplace, aerodynamic interferences by
adjacent buildings are of great concern, inasmuch as the interference modi"es the
#ow "eld and eventually the wind loading on individual buildings. The most
common phenomenon observed around closely spaced buildings is the ampli"cation
of wind speed due to blockage e!ects. The concern for the consequences of the wind
ampli"cation e!ects has prompted investigations of the #ows around bodies in
proximity.
Two square cylinders placed side-by-side normal to a uniform stream can be
considered as an idealized case which we can learn from about the interaction of the
#ows around bodies in proximity. This #ow was the subject of the recent experimental
study by Kolar et al. [20]. The Reynolds number (Re ) based on the cylinder width
"
and the freestream velocity is about 23 100, which is quite close to the single cylinder
case [19]. The experiment reveals that, depending on the separation of the cylinders,
the interaction produces di!erent #ow regimes in terms of structure of the vortex
streets. In this paper, we considered the case where two square cylinders are separated
by 2.0 cylinder widths. According to the experimental study [20], the #ow for the
156 S.-E. Kim, F. Boysan / J. Wind Eng. Ind. Aerodyn. 81 (1999) 145}158
Fig. 9. Vortex-shedding behind two cylinders placed side-by-side. left: time history of C , right: contours of
"
instantaneous vorticity magnitude.
present Reynolds and the separation between the cylinders exhibits a stable `in-
antiphasea model of vortex-shedding pattern. The experiment also indicates that the
vortex-shedding frequency remains close to that of a single cylinder, although there
are some experimental evidence that indicates otherwise. Again the RNG k}e model
was employed for the computation. Fig. 9 shows the computational results. It was
found that the vortex-shedding frequency for the present double-cylinder case is
roughly 0.15, which is slightly larger than that for the single cylinder.
Despite their e$cacy for the simulation of practical engineering #ows, the phenom-
enological turbulence models based on Reynolds-averaged Navier}Strokes (RANS)
equations have not been completely satisfactory in dealing with blu!-body aerody-
namics. Partly being driven by the dissatisfaction with the RANS-based turbulence
models, yet perhaps more likely encouraged by the explosive increase in computing
power and its a!ordability, CFD practitioners, developers and users alike, have
come to seriously think about high-level simulation of turbulent #ows such as LES.
LES for blu!-body aerodynamics has been recently attempted by many investigators
[21,22].
The governing equations for LES contain terms that represent the e!ects of the
"ltered-out small eddies on the resolved eddies. These terms are called `subgrid scalea
(SGS) stresses, and should be modeled. The most popular and simplest SGS stress
model is the Smagorinsky model. In the Smagorinsky model, the SGS stresses are
related to the resolved scales via eddy-viscosity which is computed in terms of the
resolved velocity "eld (rate-of-strain) and the "lter width, which are all known. The
Smagorinsky model, despite its simple SGS stress modeling, has been successfully
applied to building aerodynamics. Recently, more elaborate SGS models have been
proposed. Particularly, worthy mentioning among others are the SGS eddy-viscosity
model derived by Yakhot and Orszag [23] and the dynamic model proposed by
Germano et al. [24] and its variants.
We present here a preliminary result for a surface-mounted cube [17]. The compu-
tation was made using the RNG-based SGS eddy viscosity model [23] on a 460 000
cell hexahedral mesh locally re"ned near the cube. Fig. 10 shows iso-contour of
S.-E. Kim, F. Boysan / J. Wind Eng. Ind. Aerodyn. 81 (1999) 145}158 157
Fig. 10. Iso-contours of instantaneous vorticity magnitude for the #ow around a surface-mounted cube
(LES result).
4. Closure
The major issues that determine successful application of CFD to building aerody-
namics were discussed. For complex geometries encountered in the modeling of
buildings and structures, unstructured mesh has a great potential to signi"cantly save
time and e!ort for mesh generation. It was demonstrated that locally re"ned meshes
can economically resolve the wide range of length and time scales. Some of the more
recently proposed turbulence models were shown to signi"cantly improve the accu-
racy of numerical solutions for turbulent #ows around blu! bodies. Finally, large eddy
simulation will play an increasingly more important role in the coming years,
especially in dealing with turbulence modeling issues that RANS equations-based
turbulence models cannot resolve.
References
[1] J.M. Weiss, W.A. Smith, Preconditioning applied to variable and constant density #ows, AIAA J. 33
(11) (1995).
[2] S.R. Mathur, J.Y. Murthy, A pressure-based method for unstructured meshes, Numer. Heat Transfer
31 (1997) 195}215.
[3] S.-E. Kim, S.R. Mathur, J.Y. Murthy, D. Choudhury, A Reynolds-averaged Navier}Stokes solver
using an unstructured mesh based "nite-volume scheme, AIAA-98-0231, 1998.
158 S.-E. Kim, F. Boysan / J. Wind Eng. Ind. Aerodyn. 81 (1999) 145}158
[4] D.K. Heist, F.C. Gouldin, Turbulent #ow normal to a triangular cylinder, J. Fluid Mech. 331 (1997)
107}125.
[5] B.E. Launder, D.B. Spalding, The numerical computation of turbulent #ows, Comput. Meth. Appl.
Mech. Eng 3 (1974) 269}289.
[6] D. Lakehal, Application of the k}e model to #ow over a building placed in di!erent roughness
sublayers, J. Wind Eng. Ind. Aerodyn. 49 (1998) 59}77.
[7] S. Murakami, A. Mochida, Y. Hayashi, Scrutinizing k}e EVM and ASM by means of LES and wind
tunnel for #ow"eld around a cube, Eighth Symposium on Turbulent Shear Flows, 17-1-1, 1991.
[8] V. Yakhot, S.A. Orazang, S. Thangam, T.B. Gatski, C.G. Speziale, Development of turbulence models
for shear #ows by a double expansion technique, Phys. Fluids A 4 (7) (1992) 1510}1520.
[9] D.C. Wilcox, Comparison of two-equationc turbulence models for boundary layers in pressure
gradient, AIAA J. 31 (8) (1993) 1414}1424.
[10] T.-H. Shih, W.W. Liou, A. Shabbir, J. Zhu, A new k}e eddy-viscosity model for high Reynolds number
turbulent #ows } model development and validation, Comput. Fluids 24 (3) 227}238.
[11] T.J. Craft, B.E. Launder, K. Suga, Extending the applicability of eddy viscosity models through the
use of deformation invariants and non-linear elements, Proc. Fifth International Symposium on
Re"ned Flow Modeling and Turbulence Measurements, Paris, 1993, pp. 125}132.
[12] T.B. Gatski, C.G. Speziale, On explicit algebraic stress models for complex turbulent #ows, J. Fluid
Mech. 254 (1993) 47}83.
[13] T.-H. Shih, J. Zhu, J.L. Lumely, A new Reynolds stress algebraic equation model, NASA Technical
Memorandum 106644, ICOMP-94-15; GMOTT-94-8, 1994.
[14] V. Baskaran, A.J. Smits, P.N. Joubert, A turbulent #ow over a curver hill } Part 1. Growth of an
internal boundary layer, J. Fluid Mech. 182 (1987) 47}83.
[15] M.M. Gibson, B.E. Launder, Ground e!ects on pressure #uctuations in the atmospheric boundary
layer, J. Fluid Mech. 86 (1978) 491}511.
[16] N. Djilali, I.S. Gartshore, M. Salcudean, An experimental and numerical study of the #ow around
a blunt rectangular section } a test case for computational method, Sixth Symposium on Turbulent
Shear Flows, 19-3-1, Toulouse, France, 1987.
[17] I.P. Castro, A.G. Robins, The #ow around a surface-mounted cube in uniform and turbulent streams,
J. Fluid Mech. 79 (Part 2) (1977) 307}335.
[18] W. Rodi, Experience with two-layer models combining the k}e model with a one-equation model near
the wall, AIAA Paper 91-0216, 1991.
[19] D.A. Lyn, Ensemble-averaged measurements in the turbulent wake of a square cylinder: a guide to the
data, Report No. CE-HSE-92-6, School of Civil Engineering, July 1992.
[20] V. Kolar, D.A. Lyn, W. Rodi, Ensemble-averaged measurements in the turbulent near wake of two
side-by-side square cylinders, J. Fluid Mech. 346 (1997) 201}238.
[21] S. Murakami, Blu! body aerodynamics and turbulence, J. Wind Eng. Ind. Aerodyn. 73 (1993) 65}78.
[22] T. Kogaki, T. Kobayashi, N. Taniguchi, Large eddy simulation of #ow around a rectangular cylinder,
Fluid Dyn. Res. 20 (1997) 11}24.
[23] V. Yakhot, S. Orszag, Renormalization group analysis of turbulence: I basic theory, J. Sci. Comput.
1 (1986) 1}64.
[24] M. Germano, U. Piomelli, P. Moin, W.H. Cabot, A dynamic subgrid scale eddy viscosity model, Phys.
Fluids A 3 (1991) 1760.
[25] A.K. Mikkelsel, F.M. Livesey, Evaluation of the use of numerical k}e model Kameleon II, for
predicting wind pressure on building surfaces, J. Wind Eng. Ind. Aerodyn. 57 (1995) 375}389.