See discussions, stats, and author profiles for this publication at: [Link]

net/publication/228727571

Cross Comparisons of CFD Results of Wind Environment at Pedestrian Level

around a High-rise Building and within a Building Complex

Article in Journal of Asian Architecture and Building Engineering · May 2004

DOI: 10.3130/jaabe.3.63

CITATIONS READS

108 877

7 authors, including:

Yoshihide Tominaga Akashi Mochida

Niigata Institute of Technology Tohoku University
203 PUBLICATIONS 10,273 CITATIONS 291 PUBLICATIONS 10,744 CITATIONS

SEE PROFILE SEE PROFILE

Hiroto Kataoka Kazuyoshi Harimoto

Obayashi Corporation Independent Researcher
38 PUBLICATIONS 3,350 CITATIONS 4 PUBLICATIONS 621 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Yoshihide Tominaga on 27 May 2014.

The user has requested enhancement of the downloaded file.

 Cross Comparisons of CFD Results of Wind Environment at Pedestrian Level
around a High-rise Building and within a Building Complex

Yoshihide Tominaga*1, Akashi Mochida2, Taichi Shirasawa3

Ryuichiro Yoshie4, Hiroto Kataoka5, Kazuyoshi Harimoto6, Tsuyoshi Nozu7

1
Professor, Niigata Institute of Technology, Japan
2
Assoc. Professor, Graduate School of Engineering, Tohoku University,Japan
3
Graduate student, Graduate School of Engineering, Tohoku University,Japan
4
Deputy General Manager, Technical Research Institute, Maeda Corp.,Japan
5
Chief Research Engineer, Technical Research Institute, Obayashi Corp.,Japan
6
Research Engineer, Technology Center, Taisei Corp.,Japan
7
Research Engineer, Institute of Technology, Shimizu Corp.,Japan

Abstract
Recently, prediction of the wind environment around a high-rise building using Computational Fluid Dynamics
(CFD) has been carried out at the practical design stage. However, very few studies have examined the accuracy
of CFD including the velocity distribution at pedestrian level. Thus, a working group for CFD prediction of the
wind environment around a building was organized by the Architectural Institute of Japan (AIJ). This group
consisted of researchers from several universities and private companies. In the first stage of the project, the
working group planned to carry out cross comparison of CFD results of flow around a single high-rise building
model placed within the surface boundary layer and flow within a building complex in an actual urban area
obtained from various numerical methods. This was done in order to clarify the major factors affecting prediction
accuracy. This paper presents the results of this comparison.

Keywords: CFD; wind environment assessment; cross comparison; revised k-ε models; actual urban area

Introduction numerical methods, in order to clarify the major factors

Recently, prediction of the wind environment around affecting prediction accuracy. The first part of this paper
a high-rise building using Computational Fluid compares results of CFD prediction of flow around a
Dynamics (CFD) has been carried out at the practical 2:1:1 shaped building model and a 4:4:1 shaped building
design stage. The performance of CFD prediction of flow model placed within the surface boundary layer using
around a bluff body based on various turbulence models various turbulence models. The latter part describes the
has been investigated by many authors [1-5]. However, cross comparison of results of the wind environment at
these previous researches focused mainly on the pedestrian level within a building complex in an actual
prediction accuracy of the separating flow and pressure urban area using different grid systems.
distribution around the roof. Few have examined the
accuracy of CFD prediction of the velocity distribution 2 Outline of cross comparisons
at pedestrian level. Thus, a working group for CFD 2.1 Flowfields tested
prediction of the wind environment around a building 1) Test Case A (2:1:1 shaped building model)
was organized by the Architectural Institute of Japan Test Case A is the flowfield around a high-rise building
(AIJ). This group consists of researchers from several model with the scale ratio of 2:1:1 placed within a surface
universities and private companies [Note]. boundary layer (Fig.1a). For this flowfield, detailed
At the first stage of the project, the working group measurement was reported by Ishihara & Hibi [6]. The
planned to carry out cross comparison of CFD results Reynolds number based on H (building height) and U0
of flow around a high rise building predicted by various (inflow velocity at z=H) was 2.4×104.
2) Test Case B (4:4:1 shaped building model)
For Test Case B, the flowfield around a building model
Contact Author: Yoshihide Tominaga, Niigata Institute of with the scale ratio of 4:4:1 (Fig.1b) was selected. A
Technology, 1719, Fujihashi, Kashiwazaki-shi, Niigata, 945- wind tunnel experiment was carried out by the present
1195, Japan authors to obtain the experimental data for assessing the
Tel & Fax:+81-257-22-8176 accuracy of CFD results. The Reynolds number based
E-mail:tominaga@[Link] on H (building height=4b) and U0 (inflow velocity at
(Received November 1, 2003 ; accepted Februry 1, 2004 ) z=H=4b) was 7.2×104.

Journal of Asian Architecture and Building Engineering/May 2004/8 1

3) Test Case C (a building complex in an actual urban measured by non-directivity thermistor anemometers for
area) case C.
The target for Test Case C was the flowfield within a 2.2 Specified Conditions
building complex in an actual urban area (Fig.1(c)). A In order to assess the performance of turbulence
wind tunnel experiment was carried out by the present models, the results should be compared under the same
authors. computational conditions. Special attention was paid to
In the experiments for cases A and B, the wind velocity this point in this study. The computational conditions,
was measured by a split fiber type anemometer that could i.e., grid arrangements, boundary conditions, etc., were
monitor each component of an instantaneous velocity specified by the organizers of the cross comparison, and
vector. On the other hand, the mean wind velocity was is summarized in the Appendix 1 and Table 4. The

Fig.1. Flowfields tested in this study

Table 1. Computed cases for 2:1:1 shaped building model(Test CaseA)

2 JAABE vol.3 no.1 May. 2004 Yoshihide Tominaga

contributors were requested to follow the given
conditions.

3. Results and discussion

3.1 Test Case A (2:1:1 shaped building model)
The computed cases are outlined in Table 1. Nine
groups have submitted a total of eighteen datasets of
results. The performance of the standard k-ε and five
types of revised k-ε models was examined. Furthermore,
Differential Stress Model (DSM)[7] and Direct
Numerical Simulation (DNS) with third-order upwind Fig.2. Lateral distribution of <u> along lateral direction (y)
near ground surface at z=1/16H height
scheme [8] and Large Eddy Simulation (LES) using the
Smagorinsky subgrid-scale model [9] were also included compared here except for LES1. It is surprising to see
for comparison. The computational conditions in this that there are significant differences between the XF
test case are described in Appendix 1 and Table 4. values of the standard k-ε model. As is already noted,
1) Reattachment lengths the grid arrangements and boundary conditions were set
The predicted reattachment lengths on the roof, XR, to be identical in all cases, and QUICK scheme was used
and that behind the building, XF, are given for all cases for convection terms in many cases. The reason for the
in Table 1. As shown by the results of the standard k-e difference in XF values predicted by the standard k-ε
(KE1~8), the reverse flow on the roof, which is clearly models is not clear, but it may be partly due to differences
observed in the experiment, is not reproduced. This was in some details of the numerical conditions, e.g. the
pointed out in previous researches by the present authors convergence condition, etc. The results of the revised k-
[1,2]. On the other hand, the reverse flow on the roof ε models except for the Durbinís model are in the
appears in the results for all revised k-ε models (LK1, tendency to evaluate XF larger than the standard k-ε
RNG1, MMK1, RNG1, LK2, LK3, MMK2, DBN), model. This discrepancy is improved in the LES and DNS
although it becomes a little larger than that in the computations. On the other hand, DSM greatly
experiment. In the DSM result, the predicted separated overestimates XF. The overestimation of reattachment
flow from a windward corner is too large, and does not length behind a three-dimensional obstacle was also
reattach to the roof. The result of LES without inflow reported by Lakehal and Rodi [5]. In ref. [5], predicted
turbulence (LES1) can reproduce the reattachment on results of flow around a surface mounted cube obtained
the roof, but XR is somewhat overestimated in this case. by five types of k-ε models, i.e. the standard k-ε model,
On the other hand, the result of LES with inflow Kato-Launder model, Two-layer k1/2 velocity-scale-based
turbulence (LES2) shows close agreement with the model, Two-layer k1/2 velocity-scale-based model with
experiment. Kato-launder modif ication and Two-layer (v’ 2) 1/2
The evaluated reattachment length behind the velocity-scale-based model, were compared. The all
building, XF, is larger than in the experiment in all cases models compared in ref. [5] including three types of the

Table 2. Computed cases for 4:4:1 shaped building model (Test Case B)

JAABE vol.3 no.1 May. 2004 Yoshihide Tominaga 3

 Fig.3. Horizontal distribution of each component of velocity along the lateral direction (y)
near the ground surface at z=1/16H height
(<u>, <v>, <w> indicate streamwise, lateral and vertical components of mean velocity vector,
respectively. Values are normalized by the velocity at the same height at the inflow boundary)

two-layer models overpredicred the reatchment length case are described in Appendix 1 and Table 4.
behind the obstacle as well as in this study. In the two 1) Reattachment length
layer models, viscous-affected near-wall region is The predicted reattachment lengths behind the
resolved by a one-equation model, while the outer region building, XF, are given for all cases in Table 2. The result
is simulated by the k-ε model. In the one-equation model, of the DNS with a third-order upwind scheme shows
the eddy viscosity is made proportional to a velocity very close agreement with the experiment. The evaluated
scale and a length scale. XF value is larger than the experimental value in all
The size of the recirculation region behind the building computed results based on the standard and revised k-ε
is strongly affected by the momentum transfer models for this test case, as well as in the results for Test
mechanism in the wake region, where vortex shedding Case A presented in 3.1. The results of the revised k-ε
plays an important role. Thus, the reproduction of vortex models except for Durbin’s model predict a larger XF
shedding is signif icantly important for accurately value than the result of the standard k-ε. This tendency
predicting the XF value. However, none of the k-ε models is also similar to the results for Test Case A.
compared here could reproduce vortex shedding. This 2) Lateral distributions of each component of velocity
resulted in underestimation of the mixing effect in the vector near ground surface (z=1/16H)
lateral direction causing too large a recirculation region Fig. 3(a) shows the lateral distributions of scalar
behind the building. velocity and each component of mean velocity vector
2) Lateral distributions of <u> near ground surface near the ground surface in the area affected by the
(z=1/16H) separation at the frontal corner. These values are
Fig.2 shows the lateral distributions of the streamwise normalized by the velocity value at the same height at
mean velocity component, <u>, near the ground surface the inflow boundary. The peak measured scalar velocity
in the area affected by the separation at the front corner distribution appears at y/b 3. The standard k-ε and the
in the selected cases. The peak in the measured velocity revised k-ε models overestimate the velocity around this
distribution appears at y/b=-0.9. The standard k-ε (KE8) point. As shown in Fig. 3(b), in this area, the streamwise
and the modified LK model (LK3) underestimate the component, <u>, of the mean velocity vector decreases
velocity around this point. For the Durbin’s model as the distance from the side-wall decreases in the
(DBN), the position and the peak value in the velocity experimental result. On the other hand, the measured
distribution are well reproduced. In DSM, the evaluated <u> values decrease in the area and increase in the area
velocities are generally larger in the region of y/b<-1.5 4<y/b as distance from the wall increases. The results of
than those with other computations. the standard k-ε model do not reproduce this tendency
3.2 Test Case B (4:4:1 shaped building model) at all, while the result of the revised k-ε models show
Outlines of computed cases are listed in Table 2. Five better agreement with the measured distribution.
groups have submitted a total of twelve datasets of Between the results of the two revised k-ε models
results. The performance of the standard k-ε and six types compared here, the distribution of <u> obtained from
of revised k-ε models was examined. Furthermore, DNS the LK model shows much better agreement with the
with a third-order upwind scheme [15] was also included experiment than do the RNG models.
for comparison. Computational conditions in this test As shown in Fig. 3(d), the peak in the measured <w>

4 JAABE vol.3 no.1 May. 2004 Yoshihide Tominaga

distribution appears at y/b 2.5. For the standard k-ε, wind directions in Niigata City. Since no clear differences
the peak value is hardly reproduced. However, the result were observed between the horizontal distributions of
of the RNG and LK models show generally close scalar velocity near the ground surface (z=2m) predicted
agreement with the experiment. by the three CFD codes, the results from Code T are
3.3 Test Case C shown in Fig. 6. This figure illustrates the horizontal
(building complex in actual urban area) distributions of scalar velocity near the ground surface
Finally, prediction accuracy for wind environment (z=2m). The values in Fig. 6 are normalized by the
within an actual building complex, located in Niigata velocity at the same height at the inflow boundary. It
City, Niigata Prefecture, Japan, is examined. Fig. 6 can be seen that high velocity regions appear in the area
illustrates three-target buildings (A~C). Building A is around the corner of the north and east sides of building
60m high, and buildings B and C are both 18m high. A and strong wind blows into the space between
The surrounding area is mostly covered with low-rise buildings with the wind direction from NNE. On the
residential houses. The wind rose of Niigata Local other hand, a high velocity region is observed in the area
Meteorological Observatory is shown in Fig. 4. around the corner of the south side of building A with
Here, we compare the results predicted with three the wind direction from W. The velocities in the street
different codes: a homemade CFD code and two for the NNE wind direction are smaller than those for
commercial CFD codes. The computational conditions the W wind direction.
are described in Appendix 1 and Table 4. Data from an Fig. 7 shows the correlation between the normalized
identical CAD file is used to reproduce the geometries velocities obtained for each code and those of the wind
of the surrounding building blocks. This CAD file is tunnel experiment. The black circle indicates the
produced from a drawing of the experimental model.
Specifications of the CFD codes are compared in Table
3. Fig. 5 illustrates an enlarged view of the computational
grid around the high-rise building model in all cases.
Although CFD simulations were performed for sixteen
different wind directions, only the wind distributions for
wind directions NNE and W are shown here due to the
limitation of available space. These are the prevailing
Fig.4. Wind Rose of Niigata Local Meteorological Observatory

Fig.5. Grid arrangements

Table 3. Computed cases for building complex in actual urban area (Test Case C)

JAABE vol.3 no.1 May. 2004 Yoshihide Tominaga 5

 (1) Wind direction: NNE (2) Wind direction: W
Fig.6. Distributions of normalized scalar velocity near ground surface (z=2m)(Code T)

Fig.7. The correlation between the normalized velocity predicted by each code and wind tunnel exp.

velocities at the measuring points in the wake region. A Fig. 8 compares the normalized velocities at each
similar tendency is observed for all results in Figs. 7(1) measuring point. It is confirmed that all three CFD codes
and (2). It is found that the scalar velocity predicted by compared here can predict the distribution of scalar
all CFD codes tested here tends to be smaller in the wake velocity in reasonable agreement with the measurements
region compared to the experimental value, as well as in except for the wake region and the region far from the
the results for Test Cases A and B. Except for the target buildings. The prediction error in the far region is
velocities in the wake region, the CFD analyses agree mainly caused by the insufficient grid resolution in this
closely with the experimental results. The difference region, which is obviously not fine enough.
between the scalar velocities in the wake region from 4 Conclusions
the CFD and the experimental results is partly because 1) In the first part of this paper, the flowfields around
the definition of the mean scalar velocity measured by two types of a high-rise building model, i.e. a 2:1:1
the non-directivity thermistor anemometers is different shaped model and a 4:4:1 shaped model placed within
from that of CFD (cf. Appendix 3). This point will be the surface boundary layer, were predicted using the
examined in more detail in the next stage of this project. standard k-e model, the revised k-ε models, DSM, LES

6 JAABE vol.3 no.1 May. 2004 Yoshihide Tominaga

 Fig.8. Comparison of normalized velocity value for each measurement point

and DNS with a 3rd order upwind scheme. Results of Note

these predictions were compared with experimental data. The working group members are: A. Mochida (Chair, Tohoku Univ.),
2) The standard k-ε model could not reproduce the Y. Tominaga (Secretary, Niigata Inst. of Tech.), Y. Ishida (Kajima Corp.),
T. Ishihara (Univ. of Tokyo), K. Uehara (National Inst. of Environ.
reverse flow on the roof in Test Case A. This drawback Studies), R. Ooka (I.I.S., Univ. of Tokyo), H. Kataoka (Obayashi Corp.),
was corrected by all revised k-e models tested here. T. Kurabuchi (Tokyo Univ. of Sci.), N. Kobayashi (Tokyo Inst.
However, the revised k-ε models except for the Durbin’s Polytechnics), N. Tuchiya (Takenaka Corp.), Y. Nonomura (Fujita Corp.),
model overestimated the reattachment length behind the T. Nozu (Shimizu Corp.), K. Harimoto (Taisei Corp.), K. Hibi (Shimizu
Corp.), S. Murakami (Keio Univ.), R. Yoshie (Maeda Corp.)
building in comparison with the standard k-ε model in
Test Cases A and B.
Appendix 1 Outline of computational conditions
3) The LK and RNG models provided more accurate
specified by the organizer
results than did the standard k-ε model in the area around
1) Computational domain:
the side face of the building near the ground surface in The computational domain covers the specified sizes, which corresponds
Test Case B. to the size of the wind tunnel in the experiment. The computational
4) In the latter part, the flowfield within a building domain was divided into a specified number of grids. The size of the
complex in an actual urban area (Test Case C) was computational domain, grid discretization and the minimum grid interval
are summarized in Table 4.
predicted by three different CFD codes based on different
grid systems. Results of these predictions were compared 2) Inflow boundary:
At the inflow boundary, the interpolated values of <u> and k obtained
with experimental data. No clear differences were from the experimental results are imposed. The vertical profile of mean
observed between the CFD results given from these three velocity <u(z)> approximately obeyed the power law expressed as
codes for this test case under the computational <u(z)>∝za in the experiment. The value of ε is obtained from the relation
conditions specified by the organizer. Pk=ε. The α value for each test case is shown in Table 4.
5) The CFD codes compared here can predict the 3) Ground surface boundary [18]:
In these cross comparisons, the wall function based on the logarithmic
distribution of the scalar velocity at pedestrian level law of the form containing the roughness length z0 is employed. This is
within the actual building complex in reasonable mainly because the velocity profile should be maintained in the area
agreement with the measurements except for the wake apart from the building. z0 values for each test case are shown in Table 4.
region and the region far from the target buildings where The friction velocity u* is obtained from the relation using the value of
the grid resolution is obviously not fine enough. k at the closest point to the ground in the experiment. It was confirmed
in the preliminary calculation without the building model that the profile
at inflow was maintained at the outflow boundary with this boundary
Acknowledgements condition. Regarding the boundary condition for the ground surface near
The authors would like to express their gratitude to the buildings, more detailed investigation will be done in the next stage
the members of working group for CFD prediction of of this project.
the wind environment around a building [cf. Note]. 4) Lateral and upper surfaces of computational

JAABE vol.3 no.1 May. 2004 Yoshihide Tominaga 7

 Table 4 Computational conditions

domain: 36
In Test Case A, the wall functions based on a logarithmic law for a 3) Kato, M. and Launder, B.E. (1993), “The modeling of turbulent
smooth wall are used. flow around stationary and vibrating square cylinders”, Prep. of
In Test Cases B and C, the normal velocity components defined at the 9th Symp. on Turbulent shear flow, 10-4-1-6
boundaries and the normal gradients of the tangential velocity 4) [Link], H. Kawai, S. Kawamoto et al(1997), “Numerical
components, k, ε across the boundaries, were set to zero. prediction of wind loading on buildings and structure - AIJ
5) Building surface boundary: cooperative project on CFD”, J. of Wind Eng. and Ind. Aerodyn
67&68, 671-685
The wall functions based on logarithmic law for a smooth wall are used.
5) D. Lakehal, W. Rodi(1997), “Calculation of the flow past a surface-
6) Downstream boundary: mounted cube with two-layer turbulence models”, J. Wind Eng.
Zero gradient condition is used for all velocity components, k and ε. Ind. Aerodyn.,67&68(1997) 65-78
Appendix 2 Grid arrangements employed in Test Case C 6) Ishihara,T. and Hibi,K. (1998), “Turbulent measurements of the
Code M: A structured grid system was employed. The whole flow field around a high-rise building”, J. of Wind Eng., Japan,
computational domain was divided into 150×140×38 grids. The target No.76, 55-64(in Japanese)
buildings were surrounded by 2m×2m grids. 7) Murakami,S., Mochida, A. and Ooka, R. (1993), “Numerical
Code D: An unstructured grid system with prismatic cells over the ground simulation of flowfield over surface-mounted cube with various
and building surface was used. The whole computational domain was second-moment closure models”, 9th Symp. on Turbulent Shear
divided into 800,000 using Tetra, Pyramid and Prism cells. The distance Flow,13-5
from solid surfaces of ground and building to the first interior grid point 8) Kataoka,H. and Mizuno, M. (2002), “Numerical flow computation
was set to about 0.6m. around aeroelastic 3D square cylinder using inflow turbulence”,
Code O: An overlapping structured grid system was employed. The whole Wind and Structures, Vol. 5, No. 2-4, pp.379-392
computational domain was divided into 250,000. The grid interval was 9) Tominaga, Y. , Mochida, A. and Murakami, S.(2003) “Large Eddy
5m in the horizontal directions. The sub-computational domain was Simulation Flowf ield around a High-rise Building”, 11th
divided into 250,000. The grid interval in the horizontal directions was ICWE,B10.5
2m. The distance between the ground surface and the first interior grid 10) Yakhot, V. and Orszag,S.A, (1986), “Renormalization group
point was set to about 0.7m. analysis of turbulence”, J. Sci. Comput. 1, 3
Appendix 3 11) Tsuchiya, M., Murakami, S., Mochida, A., Kondo, K. and Ishida,Y.
The mean scalar velocity measured in the wind tunnel using a non- (1997), “Development of a new k-e model for flow and pressure
directivity thermistor anemometer (Sexp) is regarded as the time averaged fields around bluff body”, J. of Wind Eng. and Ind. Aerodyn. 67/
instantaneous scalar velocity, which can be expressed as: 68, 169-182
Sexp=<(u2+v2+w2)1/2>. 12) Tominaga, Y. and Mochida, A. (1999), “CFD prediction of flowfield
On the other hand, the mean scalar velocity given from k-ε model (Sk-ε) and snowdrift around building complex in snowy region”, J. Wind.
is the calculated from the time averaged velocities vector, namely, Eng. Ind. Aerodyn. 81, 273-282
Sk-ε=(<u>2 +<v>2+<w>2) 1/2. 13) Durbin,P.A. (1996), “On the k-e stagnation point anomaly”, Int. J.
Thus, the output of the thermistor anemometer is larger than that given Heat and Fluid Flow, 17, 89-90
from the k-e model. 14) T.H. Shih, W. W. Liou, A. Shabbir, Z. Yang and J. Zhu(1995), “A
Sexp=<(u2+v2+w2)1/2> New k-e Eddy Viscosity Model for High Reynolds Number
=<{(<u>+u’)2+(<v>+v’)2+(<w>+w’)2}1/2> Turbulent Flows” Computers Fluids Vol. 24 No.3 pp.227-238
=<(<u>2+<v>2+<w>2+<u’2+v’2+w’2>)1/2> 15) T.H. Shih, J. Zhu, J.L. Lumley(1993), “A realizable Reynolds stress
=(<u>2+<v>2+<w>2+2k)1/2 algebraic equation model”, NASA TM-105993
=(Sk-ε2+2k)1/2 16) Nagano, Y. and Hattori, H. , (2003)” A new low-Reynolds number
Here, u,v,w: three components of instantaneous velocity vector, <f>: turbulence model with hybrid time-scale of meanflow and
time-averaged value of f, f ’=f-<f>. turbulence for complex wall flow”, Proc. 4th Int. Symp. On
Turbulence, Heat and Mass Transfer(Eds. K. Hanjalic, Y. Nagano
References and F. Arinc), Antalya, Turkey, October 12-17
1) Murakami, S., Mochida, A. and Hayashi, Y. (1990),”Examining 17) Kataoka,H., (2003) “Large Eddy Simulation of building”,
the k-e model by means of a wind tunnel test and large eddy Summaries of Technical Papers of Annual Meeting, Environ. Engg.
simulation of turbulence structure around a cube”, J. Wind Eng. II, AIJ (in Japanese)
Ind. Aerodyn. 35, 87-100 18) Yoshie,R. (1999), “CFD analysis of flow field around a high-rise
2) Murakami,S. (1993), “Comparison of various turbulence models building”, Summaries of Technical Papers of Annual Meeting,
applied to a bluff body”, J. Wind Eng. Ind. Aerodyn., 46&47, 21- Environ. Engg. II, AIJ (in Japanese)

8 JAABE vol.3 no.1 May. 2004 Yoshihide Tominaga

View publication stats