Cooperative project for CFD prediction of pedestrian wind environment in the Architectural Institute of Japan

Transcription

1 Journal of Wind Engineering and Industrial Aerodynamics 95 (27) Cooperative project for CFD prediction of pedestrian wind environment in the Architectural Institute of Japan R. Yoshie a,, A. Mochida b, Y. Tominaga c, H. Kataoka d, K. Harimoto e, T. Nozu f, T. Shirasawa g a Department of Architecture, Tokyo Polytechnic University, 583 Iiyama, Atsugi, Kanagawa Japan b Graduate School of Engineering, Tohoku University, Japan c Niigata Institute of Technology, Japan d Technical Research Institute, Obayashi Corp., Japan e Technology Center, Taisei Corp., Japan f Institute of Technology, Shimizu Corp., Japan g The University of Tokushima, Japan Available online 23 April 27 Abstract CFD (computational fluid dynamics) is being increasingly applied to the prediction of the wind environment around actual high-rise buildings. Despite this increasing use, the prediction accuracy and many factors that might affect simulation results are not yet thoroughly understood. In order to clarify ambiguities and make a guideline for CFD prediction of the wind environment, a working group was organized by the Architectural Institute of Japan. This group has carried out various comparative studies as follows. First stage: Flow fields around two types of single high-rise buildings. Second stage: Flow field around a high-rise building located in a city. Last stage: Flow fields around two types of Building Complexes in actual urban areas. This paper describes some of the results of the investigation by the working group, and discusses the influences of various calculation conditions on CFD results, and also on the present status and the problems in CFD prediction of the wind environment. r 27 Elsevier Ltd. All rights reserved. Keywords: CFD; Pedestrian wind environment; RANS model; Guideline Corresponding author. Tel./fax: addresses: yoshie@arch.t-kougei.ac.jp (R. Yoshie), tominaga@abe.niit.ac.jp (Y. Tominaga) /$ - see front matter r 27 Elsevier Ltd. All rights reserved. doi:.6/j.jweia

2 552 R. Yoshie et al. / J. Wind Eng. Ind. Aerodyn. 95 (27) Nomenclature U average wind velocity in stream-wise direction (x direction), m/s V average wind velocity in transverse direction (y direction), m/s W average wind velocity in vertical direction (z direction), m/s k Turbulent kinetic energy, m 2 /s 2 Dissipation rate of turbulent kinetic energy, m 2 /s 3 P k Production of turbulent kinetic energy, m 2 /s 3. Introduction Progress in high-speed processing by personal computer and rapid propagation of software for numerical analysis of fluid dynamics in recent years have enabled prediction of the pedestrian wind environment around high-rise buildings based on CFD (computational fluid dynamics). It is becoming common for calculations to be performed for 6 wind directions of the situation before and after construction of proposed buildings, and for the pedestrian wind environment to be assessed by probability evaluation. However, there have been very few reports on the prediction accuracy of CFD simulations of the pedestrian wind environment around buildings in urban areas. Furthermore, the influences of various calculation conditions (such as size of computational domain, grid resolution, boundary conditions, selection of turbulence model, etc.) on the results of CFD simulation are not yet thoroughly understood. Thus, a working group named Working Group for CFD Prediction of the Wind Environment around a Building has been organized by the Architectural Institute of Japan. The name of this working group has been subsequently changed to Working Group for Preparation of Wind Environment Evaluation Guideline based on CFD. Since its inception, it has been making continuous efforts to prepare guidelines for proper use of CFD for calculation of the wind environment. Comparative and parametric studies have been carried out on several building configurations to elucidate the problems on setting or selecting various calculation conditions and turbulence models for CFD simulation of the pedestrian wind environment in urban areas. Although there have been the recommendations with similar objectives proposed by COST group (Franke et al., 24), those are mainly based on the results published by other authors. On the other hand, we are intended to propose the guidelines based on the results of our own benchmark tests. The present article introduces some of the results achieved by the working group and discusses the influence of calculation conditions and turbulence models on CFD calculation results and also on the present status and problems in CFD prediction of the pedestrian wind environment around buildings. 2. General features of comparative and parametric studies Figs. 6 show models for comparative and parametric studies as investigated by the working group. The results of studies will be introduced here on the flow field around a single square prism of 2:: (height:width:depth) placed in a turbulent boundary layer

3 R. Yoshie et al. / J. Wind Eng. Ind. Aerodyn. 95 (27) b Wind H = 2b b Fig.. Single high-rise building (2:: square prism). wind 4b 4b b Fig. 2. Single high-rise building (4:4: square prism). wind Z Y X.2m.2m.2m.2m Fig. 3. Simple city block. Fig. 4. A high-rise building in city.

4 554 R. Yoshie et al. / J. Wind Eng. Ind. Aerodyn. 95 (27) Fig. 5. Building complexes in actual urban areas (Niigata). Fig. 6. Building complexes in actual urban areas (Shinjuku). flow (Fig. ), the flow field around a high-rise building located in a city (Fig. 4), and flow fields around building complexes in actual Urban areas in Niigata and Shinjuku, Japan (Figs. 5 and 6). In the studies discussed here, the standard k model or modified k models or DSM were used, but LES (large eddy simulation) was not applied except for flow fields around two types of single prisms (Figs. and 2). It is desirable to use LES for highly accurate CFD. However, it is very difficult to use because it requires a lot of time for calculation in practical analysis due to the limited conditions of computer resources currently available. This is because, prediction and evaluation of the wind environment around buildings in practical application requires a wide computational domain including surrounding building groups and a vast number of grids associated with it. In addition, a number of calculation cases (such as multiple wind directions, situations before and after construction of a proposed building, and measures after construction) are required, and time for evaluation is also limited in the practical design stages. Therefore, the guidelines currently under preparation in the working group are also based on the assumption that the analysis is performed using standard k model or modified k models. 3. Flowfield around a single square prism of 2:: 3.. Outline of wind tunnel experiment In the comparative and parametric studies on the flow around a square prism of 2:: as carried out in the first step of the working group s investigation, the experimental results by Meng and Hibi (998) were used to validate the results of the CFD simulation. In this

5 R. Yoshie et al. / J. Wind Eng. Ind. Aerodyn. 95 (27) Fig. 7. Configuration of experimental model. experiment, a detailed measurement was made of the flow field around a 2:: (height:width:depth) shaped square prism placed in a turbulent boundary layer, in which the exponent for the power law of the vertical profile of average wind velocity was approximately.27 (Fig. 7). A split film probe was used to measure wind velocity, and the average wind velocity in each direction of three-dimensional space and the standard deviation of fluctuating wind velocities were determined. The model building was.8 m square (b and d) and.6 m high (h). The turbulence statistics were measured on a vertical cross-section (Fig. 8(a)) and on horizontal planes at 6 ðz=b ¼ :25Þ and 6 ðz=b ¼ :25Þ of the building height (Fig. 8(b)) Calculation conditions for comparative study (standard calculation conditions) In the working group, the conditions shown in Table and Figs. 9 and were given as the standard calculation conditions for the comparative studies (hereinafter referred as standard calculation conditions ). In addition to the standard calculation conditions, we investigated the influence on the calculation results of changing the boundary conditions, the computational domain, the grid resolution, and the turbulence models, etc Results of calculation by standard k model based on standard calculation conditions The CFD results utilizing the standard k model and based on the standard calculation conditions are compared below with experimental results of average wind velocities. () Wind velocity distribution on vertical cross-section: The distribution of average wind velocity U and W on the vertical cross-section at the center of the building is shown in Figs. and 2. In the figures, the longitudinal dotted lines represent the positions of the measuring lines in the experiment. Wind velocities are plotted transversely using this as the origin. (Positive values are plotted on the right side of the measuring line, and negative values on the left side.) The calculated values agree fairly well with the experimental values. However, near the roof surface of measurement line x=b ¼ :25 (the third measuring line from the left), U is negative in the experiment and reverse flow occurs, but this is not

6 556 R. Yoshie et al. / J. Wind Eng. Ind. Aerodyn. 95 (27) Fig. 8. Outline of wind tunnel experiment. (a) Measuring points in vertical cross-section ðy ¼ Þ. (b) Measuring points in horizontal plane (z ¼ :25b and :25b). Table Standard calculation conditions Computational domain 2b ðxþ3:75b ðyþ:25b ðzþ Grid resolution 6 ðxþ45 ðyþ39 ðzþ ¼5; 3 mesh (Fig. 9). The building was discretized into Scheme for advection term Quick scheme for U, V, W, k, Building wall surface Logarithmic law for smooth surface wall Surface of wind tunnel side wall Logarithmic law for smooth surface wall Surface of wind tunnel ceiling Logarithmic law for smooth surface wall Surface of wind tunnel floor Logarithmic law with roughness length z (z ¼ :8 4 m) Outflow boundary condition Zero gradient condition Inflow boundary condition Interpolated values of U and k from the experimental approaching flow. ¼ Cm =2 kþ (Fig. ) C m ¼ :9

7 R. Yoshie et al. / J. Wind Eng. Ind. Aerodyn. 95 (27) y Side wall of wind tunnel Side wall of wind tunnel x z Ceiling of wind tunnel Floor of wind tunnel x Fig. 9. Computational domain and grid arrangement. (a) Horizontal plane. (b) Vertical section. 2 ceiling 2 ceiling 2 ceiling height form floor z/b Inflow B.C. Exp. height form floor z/b Inflow B.C. Exp. height form floor z/b Inflow B.C U (m/s) k (m2/s2) ε (m2/s3) Fig.. Inflow boundary conditions. reproduced in the calculation. On the lower portion of measuring line x=b ¼ 3:25 (the rightmost measuring line), the calculated U is lower than the experimental value. (2) Wind velocity distribution on a horizontal cross-section: The distributions of U and V on the horizontal plane ðz=b ¼ :25Þ near the ground surface are shown in Figs. 3 and 4, respectively. The calculated values and the experimental values agree relatively well

8 558 R. Yoshie et al. / J. Wind Eng. Ind. Aerodyn. 95 (27) Exp CFD U (m/s) 5 m/s 3 z/b 2 x/b Fig.. Distribution of U in vertical section ðy ¼ Þ. 4 Exp CFD W (m/s) 2 m/s 3 z/b 2 x/b Fig. 2. Distribution of W in vertical section ðy ¼ Þ. except that the calculated U is lower than the experimental value in the wake region. The reattachment length behind the building is longer in the calculation. (3) Wind speed increase ratios near the ground surface: Fig. 5 compares the experimental and calculated scalar wind velocity. The scalar wind velocity (wind speed) near the ground

9 R. Yoshie et al. / J. Wind Eng. Ind. Aerodyn. 95 (27) Exp CFD U (m/s) 5 m/s y/b x/b Fig. 3. Distribution of U in horizontal section ðz ¼ :25bÞ. Exp CFD V (m/s) 2 m/s y/b x/b Fig. 4. Distribution of V in horizontal section ðz ¼ :25bÞ. surface ðz ¼ :25bÞ is normalized according to the wind speed at the same height when there is no building (i.e. wind speed increase ratio). The results based on the modified k models (Launder Kato (LK) type k model (Kato and Launder, 993), and Re- Normalization Group (RNG) k model by Yakhot and Orszag (986)) are also given. If it is limited to the region where the wind speed has increased (region where wind speed increase ratio is. or more), which is important in the evaluation of the pedestrian wind environment, it is predicted within an accuracy of approximately %. However, in the weak wind region behind the building, the wind speed ratio is evaluated lower in the calculation than in the experiment. When the modified k models are compared with the standard k model, prediction accuracy is slightly higher in the modified models in the strong wind region, while it is lower in the weak wind region. The reattachment length behind the building is longer in the modified k models than in the standard k model.

10 56 R. Yoshie et al. / J. Wind Eng. Ind. Aerodyn. 95 (27) Calculation Standard k-ε model x/b = -.75 x/b = -.5 x/b = -.25 x/b = x/b =.5 x/b =.75 x/b =.25 x/b = 2 x/b = 3.25 ± % ± % In front of building Side of building Behind building Experiment LK k-ε model RNG k-ε model Calculation.8.6 Calculation Experiment Experiment Fig. 5. Wind speed increase ratio near floor ðz ¼ :25bÞ Influence of various calculation conditions on the calculated results Here, the results obtained by changing the calculation conditions are summarized without diagrams. Furthermore, the findings obtained from the benchmark test on the flow field around a square prism of 4:4: (Fig. 2) are also given. Detailed information on the studies is in the following references; Working Group for CFD Prediction of Wind Environment Around Building (2), Mochida et al. (22); Shirasawa et al. (23) and Tominaga et al. (23, 24). () When the modified k models were used, a reverse flow on the roof was reproduced, and the prediction accuracy in the strong wind region near the separation region closer to the ground surface was improved. However, in the region behind the building, the reattachment length was longer than for the standard k model, and agreement with the experimental results deteriorated.

11 R. Yoshie et al. / J. Wind Eng. Ind. Aerodyn. 95 (27) (2) When LES was performed, the prediction accuracy in the flow field behind the building was dramatically improved. This is mainly because the periodic vortex shedding behind the square prism was well reproduced by LES (Tominaga et al., 23). (3) When the calculation was conducted on a finer grid (by decreasing the grid width of the standard calculation grid to 2 ), the results were very close to those for the standard grid. Thus, the grid arrangement of the standard condition was sufficient. (4) When the standard computational domain ð2b 3:75b :25bÞ was narrowed down to a small region ð3:8b 7:56b 7:75bÞ, there was almost no change in the results. (5) When k and in the inflow were varied, the calculated results varied widely. Thus, it is important to provide appropriate values for inflow k and. (6) When a first-order upwind scheme was used for the advection term, the velocity distribution at the side region of the building, where wind enters the computational grids diagonally, become less steep. This is not desirable. (7) Comparative studies performed by unifying the calculation conditions showed small differences between the calculated results for three commercial codes and two selfmade codes. 4. Flow field around a high rise-building located in a city This chapter describes the results of the study on the flow field near a high-rise building in a typical (regular) urban block. 4.. General features of the wind tunnel experiment The flow field analyzed here is that around a high-rise building in a simple urban area, for which the wind tunnel experiment was carried at the Niigata Institute of Technology. The low-rise urban block was assumed to be 4 m square and m high as shown in Fig. 6 (simulating a condition where low-rise houses are densely packed), with a high-rise building 25 m square and m high (::4) in a block at the center of this area. One urban block is assumed to be enclosed by two roads (each m wide) and roads 2 and 3 m wide. This is one of the typical Japanese situations where a high-rise condominium is built. Wind direction m road 3m road 45 Fig. 6. General features of wind tunnel experiment. m road m road

12 562 R. Yoshie et al. / J. Wind Eng. Ind. Aerodyn. 95 (27) m 45 m Experiment Scale: /4 Measuring height:5 mm 4m 2m 4m 3m 4m Fig. 7. Measuring points. Table 2 Standard calculation conditions Computational domain :8m :8m :8m (the size of the test section of the wind tunnel) Grid resolution 32 ðxþ3 ðyþ76 ðzþ ¼; 34; 6 mesh (Fig. 8) Scheme for advection term Quick scheme for U, V, W, k, Building wall and ground surface Logarithmic law for smooth surface wall Upper and side surface of computational domain Free slip wall condition (symmetric plane) Turbulence model Standard k model Inflow boundary condition Interpolated values of U and k from the experimental approaching flow ¼ Cm =2 kþ, C m ¼ :9 Outflow boundary condition Zero gradient condition The wind velocity measuring points are shown in Fig. 7. The scale of the experimental model was 4 and the measuring height was 5 mm above the floor of the wind tunnel (2 m above ground in real scale). Wind velocity was measured in three wind directions (,22:5 and 45 ) using a thermister anemometer. In addition, for wind direction only, the wind velocity was measured using a split film probe. The inflow wind velocity U H at the height of the central high-rise building H (H ¼ 25 mm in the experiment and m in real scale) was 6:6m=s General outline of calculation The problems with the CFD analysis on the urban area, as described above, are: () How wide should the computational domain be maintained in the horizontal and vertical directions? (2) How fine should the grid resolution be? (3) To what extent should the surrounding urban blocks be reproduced? (4) What model should be used as a turbulence closure? Based on the standard calculation conditions shown in Table 2 and Fig. 8, the prediction accuracy of CFD simulation was examined. By varying the calculation

13 R. Yoshie et al. / J. Wind Eng. Ind. Aerodyn. 95 (27) a.8m.8m y x b High-rise building was divided into 2(x) 2(y) 27(z) Fig. 8. Computational domain and grid resolution for standard calculation condition. (a) Whole computational domain and grid resolution. (b) Macrograph of central area. conditions of () (4), the influences of the calculation conditions on the CFD results were investigated Comparison of CFD results with experimental results based on standard calculation conditions The calculated results based on the standard calculation conditions and the experimental results are compared in Fig. 9(a) (wind direction ) and in Fig. 9(b) (wind direction 45 ). (In these figures, experimental data at measuring points very close to the high-rise building (points and 4 43) were omitted because their reliability was considered to be low.) The wind speed ratio between the scalar wind velocity and U H at the measuring point is represented on the ordinate. At measuring points 35, 38 where the wind velocity was highest for wind directions and 45, the calculated results were about 5% lower

14 564 R. Yoshie et al. / J. Wind Eng. Ind. Aerodyn. 95 (27) Wind Speed Ratio Wind Speed Ratio CFD.8 Exp Measuring Point No. CFD.8 Exp Measuring Point No. Fig. 9. Comparison between CFD based on standard calculation conditions and experiment. (a) Wind direction ¼. (b) Wind direction ¼ 45. Wind Speed Ratio Small(.5m.5m) Standard(.8m.8m) Larqe(3.6m 3,6m) Measuring Point No. Fig. 2. Influence of horizontal computational domain. than the experimental results, while relatively good matching was observed for the other strong wind regions Influence of various calculation conditions on CFD results () Influence of size of horizontal computational domain: The calculation was carried out in experimental scale. To evaluate the influence of the horizontal computational domain, it was expanded from the standard domain of :8m :8 m to one of 3:6m 3:6 m, and contracted to one of :5m :5 m, which is near the rim of the surrounding block. The results are shown in Fig. 2. When the horizontal computational domain was large ð3:6m 3:6mÞ, the wind speed tended to decrease slightly with the decrease of the obstruction ratio in the horizontal direction. On the other hand, when the calculation was

15 R. Yoshie et al. / J. Wind Eng. Ind. Aerodyn. 95 (27) Wind Speed Ratio Low(2H) Middle(3H) Standard(7.2H) Measuring Point No. Fig. 2. Influence of vertical computational domain. Wind Speed Ratio Coarse mesh Standard mesh Fine mesh Measuring Point No. Fig. 22. Influence of grid width. carried out in the smaller domain ð:5m :5mÞ, the wind speed became higher. The horizontal computational domain has a considerable influence on the results. It is somewhat difficult to interpret this result to real situation. We think if the similar urban area is spread outside of the computational domain, side space of the computational domain should be narrow. But if the outside of the urban area is open terrain, the computational domain should be large. (2) Influence of vertical computational domain: Fig. 2 shows the results obtained when the vertical computational domain was lowered from the standard height of 7:2H to 3H and 2H. When the upper computational domain was 2H, the wind speed was just slightly higher. There was almost no difference between the cases of 7:2H and 3H. It appears that no substantial problem occurred even when the vertical computational domain was lowered to about 3H. (3) Influence of grid resolution: Fig. 22 shows the calculated results when a fine grid and a coarse grid were used. For the fine grid (25ðxÞ22ðyÞðzÞ ¼4; 386; 43 grids), the grid width was set to about :5 of the standard grid in all three directions x, y and z. For the coarse grid (74ðxÞ68ðyÞ48ðzÞ ¼24; 536 grids), it was about.5 times the standard grid. The difference between the calculated results for the standard grid and the fine grid was very small. The difference between the calculated results for the coarse grid and the other cases was also small. The standard grid would be satisfactory, i.e. with each side of the high-rise building divided into portions or more. (4) Influence of reproduction range of surrounding urban blocks: Fig. 24 shows the results of calculation with two rows and three rows each deleted from the peripheral region of the

16 566 R. Yoshie et al. / J. Wind Eng. Ind. Aerodyn. 95 (27) Fig. 23. Reproduction range of surrounding urban blocks. (a) Deleting 2 rows. (b) Deleting 3 rows. Wind Speed Ratio.8 Standard Deleting 2 rows.6 Deleting 3 rows Measuring Point No. Fig. 24. Influence of reproduction range of surrounding urban blocks. Wind Speed Ratio.8.6 Standard k-ε RNG k-ε Exp Measuring Point No. Fig. 25. Influences of deferent turbulence models. surrounding urban blocks, as shown in Fig. 23. The difference from the standard case was very small except at measuring points, 2, 3 and 4 on the roads on the windward side. Therefore, the reproduction range of the surrounding urban blocks would be satisfactory for practical application if at least one block was maintained in the surroundings of the region to be evaluated (Figs. 23 and 24). (5) Influence of modification on turbulence modelling: Fig. 25 shows calculated results based on the modified k model (RNG k model). At measuring points 35, 38, etc.,

17 where the wind speed was high, the results from the modified k model was higher than that of the standard k model, and matching with the experimental results was improved. However, in the weak wind regions such as at measuring points 5 25, where the wind speed was low, matching with the experimental results deteriorated Summary R. Yoshie et al. / J. Wind Eng. Ind. Aerodyn. 95 (27) According to the results of the present study, the influence on the calculated results of the computational domain, the grid resolution, and the reproduction range of the surrounding urban block was relatively low. In the calculation for practical applications, the criteria would be as follows: at least one urban block of several tens of meters should be reproduced in the area surrounding the region to be evaluated, and the upper space region should be maintained at 3H or more, and each side of the high-rise building should be divided into portions or more. And when the similar urban block is spread outside of the computational domain, side space of the computational domain should be narrow, but when the outside of the urban area is open terrain, the computational domain should be large. If it is limited to the highest wind region, which is important for evaluation of the pedestrian wind environment around the building, the wind speed difference between CFD and experiment was about 5% at most for the standard k model. For the RNG model, more accurate prediction can be made in the strong wind region. 5. Flow field in an actual urban area (Niigata) Up to this point, we have reported results of bench mark tests for relatively simple shapes such as single buildings and relatively regular model urban blocks. In actual urban areas, however, buildings have complicated shapes and are distributed in an irregular manner. In order to accurately reproduce them in a CFD simulation, a great number of grids are required. In particular, when an orthogonal structured grid system is used, it is difficult to provide a grid configuration that matches well with the building configuration, and special care must be taken to allow for the influence of mismatching on the prediction accuracy. This problem can be solved if an unstructured grid system is used, although there are some difficulties in preparing this kind of grid. Furthermore, for the wind environment in urban areas, prediction is normally made for 6 wind directions on two patterns, i.e. before and after the construction of the proposed building. In some situations, it is necessary to consider protection against wind such as planting trees. This increases the number of cases that need to be considered, and calculation accuracy must be maintained under restrictive conditions for practical application. This chapter describes a study in an actual urban area in Niigata, where low-rise houses are packed closely together. Wind tunnel experiments and CFD simulation were performed to predict the wind speed distribution and to assess the pedestrian wind environment, and the results were compared. Furthermore, the differences are shown between three types of grid system: a single structured grid system (orthogonal mesh), an overlapping structured grid system, and an unstructured grid system (Tominaga et al., 24).

18 568 R. Yoshie et al. / J. Wind Eng. Ind. Aerodyn. 95 (27) Fig. 26. Actual urban model (CAD data) and measuring points. Table 3 Common calculation conditions specified Inflow boundary Computational domain Upper surface of computational domain Outflow boundary condition Ground surface boundary Building surface boundary Interpolated values of U and k from the experimental approaching flow. ¼ CD=2 k du=dz ¼ PkÞ Area about 5 m ðxþ5 m ðyþ3 m ðzþ that includes the whole urban block Free slip wall condition (symmetric plane) Zero gradient condition Logarithmic law with roughness length z ðz ¼ :24 mþ Logarithmic law for smooth wall 5.. Urban area model under study and outline of wind tunnel experiment The study was performed on an urban area model. This model consisted of an actual city block in Niigata city, Niigata prefecture, Japan with low-rise houses packed closely together. A target building 6 m high (Building A) and two target buildings 8 m high (Buildings B and C) were assumed to be constructed 26 (Fig. 26). Wind tunnel experiments at 25 scale were performed on this model in a turbulent boundary layer with a power law exponent of.25. Scalar wind velocities at 8 mm above the wind tunnel floor (2 m above the ground surface in real scale) were measured by multi-point thermister anemometers Outline of numerical calculation The items shown in Table 3 were specified as common calculation conditions in the comparative study. For the configuration of the urban block, input data for each CFD code were prepared from the same CAD data obtained from the drawing of the wind tunnel model. The features of the compared CFD codes and the different calculation

19 R. Yoshie et al. / J. Wind Eng. Ind. Aerodyn. 95 (27) Table 4 Outline of CFD codes and calculation conditions CFD Code Code O (Self-made code) () Computational method and time integral scheme. (2) Turbulence model. (3) Scheme for advection term. () Overlapping structured grid, artifical compressibility. (2) Standard k2. (3) Third-order upwind. Grid arrangements Code M (Commerical code: STREAM for Windows) () Structured gird, SIMPLE, steady state. (2) Standard k2. (3) QUIK. Code T (Commerical code: SCRYU/Tetra for Windows) () Unstructured grid, SIMPLE, steady state. (2) Standard k2. (3) MUSCL(2nd-order). conditions are summarized in Table 4, as well as general features of the grid systems used in each CFD code Results of numerical analysis () Comparison based on wind speed ratio: CFD simulations were conducted for 6 wind directions. Here, predicted results and experimental results of the wind speed ratio are compared for the wind direction NNE, which is the wind direction most frequently occurring in Niigata. The wind speed ratio is the ratio of wind speed (scalar velocity) at each measuring point ðheight ¼ 2mÞ and the wind speed at the same height at the inflow boundary. There was no substantial difference among codes for overall distribution of wind speed ratio. As a representative example, the distribution of wind speed ratio based on CFD code T (Table 4) is shown in Fig. 27. Regions with very high wind speed were

20 57 R. Yoshie et al. / J. Wind Eng. Ind. Aerodyn. 95 (27) Fig. 27. Distribution of wind speed ratio near ground surface ðz ¼ 2mÞ (Code T) (blue line indicates the wake region). found at the corners on the northwest and east sides of Building A. Furthermore, strong wind caused by contraction flow was seen between Buildings B and C. Fig. 28 represents the correlation of wind speed ratio obtained from CFD codes with the results of wind tunnel experiments. Plotting is shown separately inside and outside the wake region of the target buildings (The wake region is indicated in Fig. 27). The predicted results in the CFD codes were almost identical, and there was no clear difference among the codes. In the wake region of the target buildings, there was a tendency to underestimate the wind speed compared with the results of the wind tunnel experiment, while in other regions the matching was relatively satisfactory. The CFD results often tended to underestimate the wind speed in the wake region of the building in benchmark tests on other configurations. Fig. 29 compares the wind speed ratios at each measuring point. For the positions of the measuring points, see Fig. 26. The CFD results generally agree well with the results of the wind tunnel experiment. In particular, the prediction accuracy is high outside the wake region & near building. The difference from experimental results was large at several measuring points, but these were mostly in alleys in the wake region of the target buildings. This may be because a slight difference in the prediction of the separation shear layer due to the difference in reproducibility of the building configuration may have influenced the calculated results, and because there were shape errors in the actual wind tunnel model and CAD data used, and also errors in the positions of the evaluation point. (2) Comparison based on criteria for assessing wind environment: A rank evaluation was performed on the CFD predicted results according to criteria for assessing the wind environment proposed by Murakami et al. (986). This was based on the occurrence frequency of daily maximum gust wind speed using observation data from the Niigata Regional Meteorological Observatory. The results are summarized in Fig. 3. The gust factor GF was assumed to be 2.5 to convert the calculated average wind speeds to the maximum gust wind speeds. Including the results at the measuring points with rank 4 on the northeast and south sides of Building A, the CFD results generally showed good

21 R. Yoshie et al. / J. Wind Eng. Ind. Aerodyn. 95 (27) CFD Experiment CFD Experiment CFD Experiment Fig. 28. Correlation of wind speed ratio between CFD result and experimental result (wind direction NNE): (a) Code O; (b) Code M; (c) Code T. matching with the wind tunnel experiment results. However, as described above, there were differences between some of the experimental results and CFD codes for the measuring points near low-rise houses in alleys or along large avenues.

22 572 R. Yoshie et al. / J. Wind Eng. Ind. Aerodyn. 95 (27) Wind speed ratio Exp. Code O CodeM Code T Fig. 29. Comparison of wind speed ratio at each measuring point (wind direction NNE). Fig. 3. Comparison of rank for criteria for assessing wind environment at each measuring point. 6. Flow field in an actual urban area (Shinjuku) 6.. General description of wind tunnel experiment and field measurement to be compared To validate the CFD predicted results, it is important to compare them not only with the wind tunnel experiments but also with field measurements. However, adequate field measurement results are not always available, and this has almost never been performed in

23 R. Yoshie et al. / J. Wind Eng. Ind. Aerodyn. 95 (27) Fig. 3. Urban area reproduced and measuring points. the past. Thus, a benchmark test was conducted on the Shinjuku Sub-central Area in Tokyo, Japan in its early development stage, for which detailed wind tunnel experiments and field measurements had been carried out in cooperation with many research organizations during construction (Yoshida et al., 978; Asami et al., 978). Using these data, the validity of the prediction accuracy of CFD was assessed. () Wind tunnel experiment: A number of wind tunnel experiments had been performed on the Shinjuku Sub-central Area. Here, the CFD simulations were carried out based on the condition of the 977 Experiment. (2) Field measurements: Field measurements were carried out from December 975 to November 983. The present CFD simulations were performed for conditions in 977 when the situation of field measurements was similar to those of the wind tunnel experiment model. In the field measurements, three-cup anemometers were used. The measurement heights differed according to the measuring point, being 3 9 m above the ground surface. The reproduced urban area and the measuring point distribution are shown in Fig. 3. Measured wind speeds at encircled points (Nos. 6, 7, 3, 5) were compared with the calculated ones Outline of numerical calculations As for the study on the urban area of Niigata City, common calculation items were designated. For the configuration of urban blocks, no wind tunnel experiment models at the time of measurement were remained. Thus, we made CAD data of the urban area configuration and topography based on the old drawings, photographs, and white maps in 977. Because no details on the topographical height difference were available, the

24 574 R. Yoshie et al. / J. Wind Eng. Ind. Aerodyn. 95 (27) Fig. 32. CAD data of urban area. Fig. 33. Grid arrangements of CFD codes: (a) CFD_A (Self made code); (b) CFD_B (SCRYU/Tetra for Windows); (c) CFD_C (Fluent). reproduction was made stepwise every 5 m by referring to the maps in 977. The CAD data thus prepared are shown in Fig. 32. Fig. 33 shows the grid arrangements of the CFD codes. Here, only CFD_A uses a structured grid system, while the others are based on a nonstructured grid. However, CFD_A uses an overlapping grid Comparison of CFD results with wind tunnel experiment results and field measurements Similar to the wind tunnel experiment and the field measurements, the wind speed obtained by CFD was normalized by the wind speed at reference points. The reference points were the top of the Shinjuku Mitsui Building (D in Fig. 3; height of observation

25 R. Yoshie et al. / J. Wind Eng. Ind. Aerodyn. 95 (27) Wind speed ratio..5 No. 6. NE E SE S SW W NW N Wind direction at reference point Wind speed ratio..5 No. 7.. NE E SE S SW W NW N Wind direction at reference point No.3 Wind speed ratio.5. NE E SE S SW W NW N Wind direction at reference point Wind speed ratio..5 No. 5. NE E SE S SW W NW N Wind direction at reference point Fig. 34. Comparison of wind speed ratio for 6 wind directions.

26 576 R. Yoshie et al. / J. Wind Eng. Ind. Aerodyn. 95 (27) Normalized Velocity field meas. (mean) field meas. (+σ) field meas. (-σ) wind tunnel CFD_A CFD_B CFD_C Measuring Point Fig. 35. Comparison of wind speed ratio at each measuring point with wind direction S. 237 m) for the wind directions NE N NW, and the top of the KDD Building (C in Fig. 3; height of observation 87 m) for the other wind directions. The measurement values for reference wind speeds of 5 m/s or more were extracted from the observation data in the year 977, and the average value of normalized wind speed for each wind direction was used as the measurement value. Fig. 34 compares the experimental results and the field measurements for wind speed ratios (normalized wind speeds) at representative measuring points. As a general trend, the wind tunnel experiment results are within the standard deviation of the field measurements, and the CFD results are also generally within the same range. The difference among calculation results is relatively small regardless of the code and the grid system. But the CFD results deviate from the experiment results and field measurements depending on the measuring point and wind direction as follows. () Measuring point 7, wind direction E: Calculated wind speed is higher than the measured one. Measuring point 7 is located in the contracted flow region when the wind direction is E. As mentioned above, the reproduction accuracy of the CAD data was not necessarily sufficient. The small difference in position between CFD and measurement might cause large difference in wind speeds between them in this contracted flow region. (2) Measurment point 3, wind direction SSW WSW: Calculated wind speeds are lower than the measured ones. For these wind direction, the measuring point 3 is located in wake region of the high-rise buildings. This underestimation in wake region is common to that for the case of the 2:: square prism and for the case of the actual urban area in Niigata. (3) Measurment point 5, wind direction NE: The measured wind speeds largely change from wind direction NE to ENE, while the calculated ones largely change from NNE to NE. The small deference in reference wind directions between measurement and CFD might bring large difference in wind speeds. Next, Fig. 35 compares the wind speed ratios at the measuring points with wind direction S. In general, the difference among the results of the three CFD codes is small, and the CFD results showed good matching with the experimental results and field measurements

27 except at nos. 27, 28, 29, 36. These measuring points are located near the boundary of the computational domain. We should have taken the horizontal computational domain larger, and should have reproduced more urban blocks around these measuring points for the better prediction. 7. Conclusion current status and problems in CFD prediction This paper has described some results of a study by the Working Group for Preparation of Wind Environment Evaluation Guideline based on CFD. In general, prediction accuracy for the weak wind regions behind buildings was not satisfactory, while prediction accuracy for the strong wind region was fairly high. In single building models, which are considered to give the highest accuracy in the experiment, the CFD analysis results were consistent with experimental results within an accuracy about % in the strong wind region. However calculated U was lower than the experimental value in the wake region behind the building. The reattachment length behind the building was longer in the calculation. For the urban block model, which is considered to give the next most accurate results in the experiment, the CFD analysis results in strong wind regions showed a prediction accuracy within or so % for the standard k model and better accuracy for the modified k models. In actual urban area models, the CAD data for CFD simulation did not completely match with experiment and field measurements, and it is difficult to quantitatively describe the prediction accuracy. However, relatively good matching was found in the strong wind region. The reason why the CFD analysis results underestimate the wind velocity in the wake regions may be because it is not possible to reproduce the vortex shedding in RANS type models such as the k model. In LES, this can be improved, and the maximum instantaneous wind velocity can also be evaluated in LES. However, in order to use LES in general-purpose applications for predicting the wind environment around buildings, we need a dramatic increase in computer processing speed in the future. For the time being, we must be content with RANS type models currently in use. In RANS type models, it is only possible to evaluate average wind velocity. To evaluate pedestrian wind environment based on maximum gust wind speed we need to convert from average wind speed to maximum gust wind speed based on the assumption of the gust factor. Although there are problems as described above in the CFD analysis using the RANS type turbulence models, it is advantageous that the detailed and overall spatial distribution of wind velocity can be identified in CFD analysis, while only limited information on wind velocity can be obtained from wind tunnel experiments. Strong wind points that may have been missed in wind tunnel experiments may be identified by the CFD simulation. Further, there are some uncertainties inherent in wind tunnel experiments (such as measuring instrument errors, incidental errors, errors in the position where the sensor is installed, etc.), but there are no such uncertainties in CFD. Furthermore, it is very difficult to have an arbitrary approach flow in the wind tunnel, while this can be freely done in CFD. It thus appears possible to reduce the differences in prediction accuracy between wind tunnel experiment and CFD on practical assessment. Acknowledgment R. Yoshie et al. / J. Wind Eng. Ind. Aerodyn. 95 (27) The authors would like to express their gratitude to the members of the Working Group for Preparation of Wind Environment Evaluation Guideline based on CFD. Note:

28 578 R. Yoshie et al. / J. Wind Eng. Ind. Aerodyn. 95 (27) The working group members are: A. Mochida (Chair, Tohoku Univ.), Y. Tominaga (Secretary, Niigata Inst. of Tech.), Y Ishida (Kajima Corp.), T. Ishihara (Univ. of Tokyo), K. Uehara (National Inst. of Environ. Studies), R. Ooka (I.I.S., Univ. of Tokyo), H. Kataoka (Obayashi Corp.), T. Kurabuchi (Tokyo Univ. of Sci.), N. Kobayashi (Tokyo polytechnic Univ.), N. Tuchiya (Takenaka Corp.), Y. Nonomura (Fujita Corp.), T. Nozu (Shimizu Corp.), K. Harimoto (Taisei Corp.), K. Hibi (Shimizu Corp.), S. Murakami (Keio Univ.), R. Yoshie (Tokyo Polytechnic Univ.). References Asami, Y., Iwasa, Y., Fukao, Y., Kawaguchi, A., Yoshida, M., Sanada, S., Fujii, K., and Members of AIJ, 978. Full-scale measurement of environmental wind on New-Shinjuku-Center characteristics of wind in built-up area (part 2). In: Proceeding of Fifth Symposium on Wind Effects on Structures, pp Franke, J., Hirsch, C., Jensen A.G., Kru s, H.W., Schatzmann, M., Westbury, P.S., Miles, S.D., Wisse, J.A., Wright, N.G., 24. Recommendations on the use of CFD in predicting pedestrian wind environment, COST Action C4 Impact of Wind and Storms on City Life and Built Environment. Kato, M., Launder, B.E., 993. The modelling of turbulent flow around stationary and vibrating square cylinders. In: Nineth Symposium on Turbulent Shear Flows, pp. 4. Meng, Y., Hibi, K., 998. Turbulent measurements of the flow field around a high-rise building. J. Wind Eng. Jpn. (76), Mochida, A., Tominaga, Y., Murakami, S., Yoshie, R., Ishihara, T., Ooka, R., 22. Comparison of k models and DSM applied to flow around a high-rise building report on AIJ cooperative project for CFD prediction on wind environment. Wind Struct. 5 (2 4), Murakami, S., Iwasa, Y., Morikawa, Y., 986. Study on acceptable criteria for assessing wind environment at ground level based on residents diaries. J. Wind Eng. Ind. Aerodyn. 24 (), 8. Shirasawa, T., Mochida, A., Tominaga, Y., Yoshie, R., Kataoka, H., Nozu T., Yoshino, H., 23. Development of CFD method for predicting wind environment around a high-rise building, Part 2: the cross comparison of CFD results on the flow field around a 4:4: prism. AIJ J. Technol. Des. (8), Tominaga, Y., Mochida, A., Murakami, S., 23. Large eddy simulation of flowfield around a high-rise building. In: th International Conference on Wind engineering, B-5. Tominaga, Y., Mochida, A., Shirasawa, T., Yoshie, R., Kataoka, H., Harimoto, K., Nozu, T., 24. Cross comparisons of CFD results of wind environment at pedestrian level around a high-rise building and within a building complex. J. Asian Archit. Build. Eng. 3 (), Tominagada, Y., Mochida, A., Harimoto, K., Kataoka, H., Yoshie, R., 24. Development of CFD method for predicting wind environment around a high-rise building, Part 3: the cross comparison of results for wind environment around building complex in actual urban area using different CFD codes. AIJ J. Technol. Des. (9), Working Group for CFD Prediction of Wind Environment Around Building, 2. Development of CFD method for predicting wind environment around a high-rise building, Part : the cross comparison of CFD results using various k models. AIJ J. Technol. Des. (2), Yakhot, V., Orszag, S.A., 986. Renormalization goup analysis of turbulence. J. Sci. Comput. (), 3 5. Yoshida, M., Sanada, S., Fujii, K., Asami, Y., Iwasa, Y., Fukao, Y., Kawaguchi, A., and Members of AIJ, 978. Full-scale measurement of environmental wind on New-Shinjuku-Center characteristics of wind in built-up area (part ). In: Proceeding of Fifth Symposium on Wind Effects on Structures, pp