ARTICLE IN PRESS

Journal of Wind Engineering and Industrial Aerodynamics 93 (2005) 741756 www.elsevier.com/locate/jweia

Experimental and numerical study of wind pressures on irregular-plan shapes

M. Gloria Gomes, A. Moret Rodrigues, Pedro Mendes
cnico, Technical University of Lisbon, Av. Rovisco Pais, DECivil/ICIST, Instituto Superior Te 1049-001 Lisbon, Portugal Received 19 July 2004; received in revised form 21 July 2005; accepted 22 August 2005

Abstract This paper presents the results of a program of wind tunnel model tests on pressure distributions for irregular-plan shapes (L- and U-shaped models). The experiments were carried out in a closedcircuit wind tunnel and a multi-channel pressure measurement system was used to measure mean values of loads on 1:100 scale models. The same tests were carried out on a cube-shaped model as an experimental validation. The effectiveness of the model shape in changing the surface pressure distributions is assessed over an extended range of wind directions. The experimental data for the Land U-shaped models showed different wall pressure distributions from those expected for single rectangular blocks. Furthermore, a Computational Fluid Dynamics (CFD) code was used to illustrate some particular cases and to provide a better understanding of the ow patterns around these irregular-plan models and of the pressure distributions induced on their faces. Computed pressure coefcients have also been compared with wind tunnel results for normal and oblique wind incidence. A general good agreement has been found for normal wind incidence whereas some differences have occurred for other directions. r 2005 Elsevier Ltd. All rights reserved.
Keywords: Wind engineering; Wind pressure coefcients; Wind tunnel testing; Irregular-plan shapes; Computational uid dynamics

Corresponding author. Tel.: +351 218418358; fax: +351 218418359.

E-mail address: mgloria.gomes@civil.ist.utl.pt (M.G. Gomes). 0167-6105/$ - see front matter r 2005 Elsevier Ltd. All rights reserved. doi:10.1016/j.jweia.2005.08.008

ARTICLE IN PRESS
742 M.G. Gomes et al. / J. Wind Eng. Ind. Aerodyn. 93 (2005) 741756

1. Introduction The innovative designs both in building forms and structural systems with the increased use of a boarder range of materials require an accurate description of the wind action and its interaction with the buildings. Therefore, more rational and rened wind design approaches have been adopted by standards in order to improve both design and analytical processes. While there is considerable research on cubical and cylindrical structuresa detailed review of this subject can be found in Meroneys paper [1]only a few research studies analyse irregular shapes [2]. In the present work, pressure distributions on irregular-plan L- and U-shaped models were experimentally determined from wind tunnel tests carried out under uniform upstream ow. A cube-shaped model was also tested for comparison and validation purposes. Although L- and U-shapes are very common building congurations, experimental data for such shapes and different wind directions is very limited. Stathopoulos and Zhou [3] have examined wind loads for L-shaped plan view as well as for L-shaped cross section (stepped-roof building) through a numerical study. They have found a good agreement in the latter case between numerically and experimentally obtained results for normal wind incidence. The aim of the present study is to assess the effects of different model shape on the surface pressure distributions over an extended range of incident wind directions. 2. Experimental program The experiments were carried out in a 1.25 1.00 m2 closed-circuit wind tunnel at Laboratorio Nacional de Engenharia Civil (LNEC). More details about this wind tunnel and its functional characteristics can be found in [4]. A preliminary study was carried out in this wind tunnel for a cube under uniform upstream ow (uniform mean velocity and low turbulence intensityless than 0.5% except in the thin boundary layer near the wind tunnel oor) to validate the experimental process. The velocity in the wind tunnel was approximately 10 m/s. Symmetrical L- and Ushaped models were also tested under the same conditions. The models used for the experiments were made of transparent PVC (3 mm of thickness), with a geometric scale of 1:100, and equipped with 35 taps located on each face tested. These pressure taps have been placed as near as possible to the sidewall and roof edges to attempt to capture the high-pressure gradients (suctions) occurring at points of ow separation. The roof is xed by screws so as to be easily removed and to allow the access to the interior of the models. On the cube, although only the roof and one of the walls were monitored, all the faces were tested by rotation of the model. On the symmetrical L- and U-shaped models only the inner faces were measured, as the others were considered to present surface pressure distributions very similar to those of a rectangular block with the same dimensions. This fact is also supported by a study of Stathopoulos and Zhou [3]. The three models are shown in Fig. 1 while their dimensions along with the location of the pressure taps are shown in Fig. 2. The cube was tested just for the normal incidence, while the other models were tested for several ow directions as shown in Figs. 8 and 9. For pressure measurements two different scanivalve transducer models were used: a multi-point electronic pressure scanner DSA3217 and a mechanical Scanivalve Model J. The latter, which belongs to an older

ARTICLE IN PRESS
M.G. Gomes et al. / J. Wind Eng. Ind. Aerodyn. 93 (2005) 741756 743

Fig. 1. Cube, L- and U-shaped models in the close-circuit wind tunnel.

(cm)

(cm)

(cm)

Fig. 2. Cube, L- and U-shaped models along with the pressure tap distribution.

generation, was intended to save time in increasing the measuring capacity. Even so, the number of the Scanivalves available channels was insufcient to perform the pressure measurements simultaneously on all the models faces. Therefore, the experiments had to be carried out on one face at a time, by sweeping through all incident wind directions. Digitization of pressure signals and analysis of data were performed by a LabVIEW application at a sampling frequency of 10 Hz for 5 min. Only mean pressure coefcients (Cp) are shown in this work (i.e., pressures normalised by the upstream dynamic pressure at roof height). The dynamic pressure of the ow was also measured with Pitot tubes and hot-wire anemometers. Although pressure coefcients are referenced to roof height dynamic pressure, in this case any height could be chosen, since upstream ow was uniform. An important problem associated with wind tunnel tests on the ow around bluff bodies is that of blockage. In fact, for some orientations, the L- and U-shaped models blockage exceeded 10%note that many authors dene blockage ratio limits between 5% and 10% [2,5]which can have repercussions on the pressure distributions (specially on suctions) and hence on the drag. However, some tests made in the closed-circuit wind tunnel with windows closed and open (taken as limit situations of blockage effects, since the closed test section tends to overestimate the drag and the open one has the opposite effect [6]) showed maximum differences in pressure measurements within 57%, which suggested little inuence of blockage on results. 3. Numerical study Numerical simulations were carried out in this study through the Computational Fluid Dynamic (CFD) package PHOENICS [7], based on the control volume method. The RNG

ARTICLE IN PRESS
744 M.G. Gomes et al. / J. Wind Eng. Ind. Aerodyn. 93 (2005) 741756

k2e model [8] was used to simulate the turbulence effects. It uses the standard k2e model equations with the exception of revised model constants and a strain dependent term in the energy dissipation equation (last term in Eq. (4)). The governing equations are the timeaveraged continuity, momentum and transport equations for k and e, as follows: qui 0, qxi ! q q qui 1 qp ui uj n nt , qxj qxj r qxi qxj q q kuj qxj qxj q q euj qxj qxj   ! nt qk qui n e, nt S ij sk qxj qxj   ! nt qe e qui e2 C m Z3 1 Z=Z0 e2 , n ce2 ce1 nt Sij k 1 bZ3 se qxj qxj k k (1)

(2)

(3)

(4)

where nt C m k2 =e (isotropic eddy viscosity), Z S k=e, S 2 2S ij S ij and S ij qui =qxj quj =qxi (mean strain tensor). The turbulence constants are [7]: sk 0:72; se 0:72; C m 0:085; ce1 1:42; ce2 1:68; Z0 4:38; b 0:012. These governing equations associated with the boundary conditions were transformed into the discrete form, i.e., an algebric form, using a staggered 3-D Cartesian system and solved numerically by the control volume nite difference method [9]. The adopted solution method for the velocitypressure elds was the SIMPLEST algorithm [10], which is a variant of SIMPLE [11]. In order to enhance the stability of the solution process, under-relaxation techniques were applied to all the equations. Uniform inlet velocity (U r 10 m=s) and low turbulence 1=2 intensity (I 2k =U r 0:1%) proles and outlet conditions in terms of zero normal 3 gradients for all quantities were adopted. Also the no slip condition and turbulence wall functions [12] were considered for the ground boundary and for all solid obstacle surfaces. Free slip conditions were assumed for the top and side boundaries. In order to assess the reliability of the code and to conclude on the suitability of the numerical approach for the type of problems studied, the ow over a cubic obstacle was investigated and compared with other studies. After some trials to test the inuence on results of number of grids and ratio of physical model to domain size, the nal simulation used a variable-spaced grid of 48 cells long 38 high 43 wide with the domain dimension about 15L (length) 5L (height) 10L (width), where L is the reference size of the cubic model (Fig. 3). A ner mesh is required in the vicinity of walls and ground in order to accurately resolve the high-gradient regions of the ow eld. A similar grid arrangement was employed for the L- and U-shaped models, with a ner resolution around all solid surfaces and near the re-entrant corners to improve the accuracy of the simulations. Vertical proles of longitudinal velocity above and downstream the cube are compared with wind-tunnel measurements (Castro and Robins [13]) and numerical results (Zhang et al. [14]) in Fig. 4. Lateral proles downwind of the cube are compared with results from the same authors in Fig. 5. Some discrepancies can be noticed between numerical and experimental results, being more apparent in the near wake region (Fig. 4) essentially due to the inability of the eddy viscosity concept in dealing with ow separation and local anisotropic turbulence [15]. Nevertheless, the main ow characteristics appear to have been

ARTICLE IN PRESS
M.G. Gomes et al. / J. Wind Eng. Ind. Aerodyn. 93 (2005) 741756 745

Wind 4L

L 0.5L

4.5

4L

10L

Fig. 3. Computational domain and grid system.

3.0 2.5 2.0 y z/L 1.5 1.0 0.5 0.0 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 u/Ur present study Zhang et al. [14] Castro & Robins [13]

Wind
z x

Fig. 4. Vertical prole of velocity at plane x L and y 0.

1.2 1.0 0.8 u/Ur 0.6 0.4 0.2 0.0 -0.2 -3 -2 -1 0 y/L 1 2 3 present study Zhang et al. [14] Castro & Robins [13]

Fig. 5. Lateral prole of velocity at plane x 2L and z 0:5L.

ARTICLE IN PRESS
746 M.G. Gomes et al. / J. Wind Eng. Ind. Aerodyn. 93 (2005) 741756

captured by the numerical modelling, which attests the reasonability of the approach and supports the research of the next section for different model congurations.

4. Results and discussion 4.1. Cube Pressures on models are presented in terms of mean pressure coefcients. Fig. 6 shows the results of the surface pressure measurements for the preliminary study carried out on a cube under uniform ow and for the incident ow direction normal to the upstream face of the cube. As expected, for normal winds, mean pressure coefcients are positive on the windward wall, with their maximum value at the stagnation point (around mid-height since the upstream ow is uniform). Since the ow over the windward face separates from the model surface at the corners, near the windward edges pressure decreases, leaving each side face and roof in a bubble of separated ow and with negative pressures (suctions). As expected, negative pressures on the leeward wall are uniformly distributed in the wake

Fig. 6. Surface pressure coefcient distribution for the cube when normal to the incident ow.

ARTICLE IN PRESS
M.G. Gomes et al. / J. Wind Eng. Ind. Aerodyn. 93 (2005) 741756 747

region. Results are compared with the ones published by Castro and Robins [13], as shown in Fig. 7. Except for discrepancies on the roof and side faces, which may be due the higher susceptibility of these faces to the experiment conditions (model edges, turbulence intensity, blockage effects) [2,13], the results of the experiments are in general agreement with those of Castro and Robins [13]. This evidence has provided an experimental validation of the remaining model tests. 4.2. L-shaped model Fig. 8 shows the pressure coefcient contours on the L-shaped model inner faces over an extended range of wind directions. The general characteristics of the pressure distribution on the inner faces of the L-shaped model can be summarised as follows:

For normal winds (a 01) the highest positive pressures (stagnation points) on the front wall are no longer in the middle of the face as in the cube but are moved to

1.2 1 0.8 0.6 0.4 0.2 0 -0.2 0 -0.4 -0.6 -0.8 -1 -1.2

Cp

Closed WT(windows 1 2 3 4 5 closed) Closed WT(windows open) Castro & Robins [13]

x/L

1.2 1 0.8 0.6 0.4 0.2

Cp

0 -0.2 0 -0.4 -0.6 -0.8 -1 -1.2

A 1

B 2

C 3

D 4

E 5

Closed WT(windows closed) Closed WT(windows open) Castro & Robins [13]

x/L
Fig. 7. Comparison of the surface pressure coefcients for the cube when normal to the incident ow, measured in a closed-circuit wind tunnel (with windows closed and open), with Castro and Robins experiments [13].

748

=0 =60 =135

ARTICLE IN PRESS

=30

=90 =150

=45 =120 Front Right Front

=180 Right Front Right

M.G. Gomes et al. / J. Wind Eng. Ind. Aerodyn. 93 (2005) 741756

1.00 0.90 0.80 0.70 0.60 0.50 0.40 0.30 0.20 0.10 0.00 -0.10 -0.20 -0.30 -0.40 -0.50 -0.60 -0.70 -0.80 -0.90 -1.00 -1.10

Fig. 8. Surface pressure coefcient distribution on the L-shaped model inner faces.

ARTICLE IN PRESS
M.G. Gomes et al. / J. Wind Eng. Ind. Aerodyn. 93 (2005) 741756 749

 

positions near the re-entrant corner. This is due to the effect of ow separation that leaves stagnant ow in that re-entrant corner. This phenomenon also affects the right inner face of the model in the sense that positive pressures arise on almost all its extent. Note that in a single rectangular block, as attested by the cube test, the side faces would be in suction. As the angle of the incident ow increases, pressure values become lower on the front wall and higher on the right one. However, higher positive pressure still exists on the reentrant corner. For orientations between 1201 and 1801, the inner faces of the model are not directly exposed to the ow, being rather under the wake region inuence. As a consequence, the pressure coefcient distribution is negative (suction) and almost uniform. However, for a 1201 positive pressure increases near the upper right edge of the right inner face. This can be attributed to the direct incidence of ow on that small area, after skipping over the opposing wing, with a consequent increase of the pressure values.

4.3. U-shaped model Fig. 9 shows the pressure coefcient contours on the U-shaped model inner faces over an extended range of wind directions. The pressure coefcient distribution on the inner faces of the U-shaped model clearly shows that:

Since the ow is symmetrical when the wind blows normally (a 01), the stagnation point is in the middle of the front wall. As on the L-shaped model, these positive pressures on the windward face (front wall) also increase on the downwind end of the lateral wings (right and left walls), whereas suctions only exist near the windward edge. Note that pressure coefcient contourlines have a denser distributionindicating higher pressure gradientsnear the windward edges of the roof in the case of the Ushaped model than in the L-shaped one. This may be due to a higher ow rate rising over the U than the L-shaped model. As the angle of the incident ow increases, pressure decreases on the front wall. Nevertheless, due to the obstruction of the left wing, the pressure does not increase on the inner face of the right wing. Positive pressures are still present near the right edge of the right wing as long as ao901. For a between 901 and 1801 all inner faces have nearly uniform pressure distributions, typical in recirculation zones, which denotes a tendency of the ow to skip past the U gap, leaving almost stagnant ow in the recess. This effect only occurs when the gap width across the recess is relatively small when compared with the length of the model, otherwise the ow would tend to enter the recess and act directly on the right wing [2].

4.4. Numerical versus experimental results The prole of wind-induced pressure coefcients along the vertical and horizontal centrelines of the front and sidewalls for a 01 is compared with the numerical results in

750

=0

=90

ARTICLE IN PRESS

=30 =120

M.G. Gomes et al. / J. Wind Eng. Ind. Aerodyn. 93 (2005) 741756

=45 =150

=60 Left Front Right

=180 Left Front Right

1.00 0.90 0.80 0.70 0.60 0.50 0.40 0.30 0.20 0.10 0.00 -0.10 -0.20 -0.30 -0.40 -0.50 -0.60 -0.70 -0.80 -0.90 -1.00 -1.10

Fig. 9. Surface pressure coefcient distribution on the U-shaped model inner faces.

ARTICLE IN PRESS
M.G. Gomes et al. / J. Wind Eng. Ind. Aerodyn. 93 (2005) 741756 751

Figs. 1012 for the L- and U-shaped models. The general agreement between experimental data and numerical results is quite good. This is particularly noticed in Figs. 11 and 12, where the numerical simulations predict even the thin boundary-layer effect near the ground. The largest deviations between numerical and experimental results are restricted to the upwind ends of lateral wings, with more emphasis for the L-shaped model (Fig. 10), where the numerical simulation underestimates the negative pressures caused there by the accelerating ow.

1.2 1 0.8 0.6 0.4 0.2 0 -0.2 0 -0.4

H Cp

A 0.5 1 x/L

B 1.5 2

L Num Exp

Wind

1.2 1 0.8 0.6 0.4 0.2 0 0 -0.2 -0.4

C D

Cp

C 0.5 1 x/L

D 1.5 2

L Num Exp

Wind

Fig. 10. Pressure coefcients along the horizontal centreline (H/2) of the L- and U-shaped models for normal incident ow (a 01).

1 A B H

z/H

0.5

z/H

0.5 L L Num Exp Wind

A 0 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 1.2 Cp

B 0 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 1.2 Cp

Fig. 11. Pressure coefcients along the vertical centrelines (L/2) of the L-shaped model for normal incident ow (a 01).

1
C D

z/H

0.5

z/H

0.5
L L Num Exp

Wind

C 0 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 1.2 Cp

0 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 1.2 Cp

Fig. 12. Pressure coefcients along the vertical centrelines (L/2) of the U-shaped model for normal incident ow (a 01).

ARTICLE IN PRESS
752 M.G. Gomes et al. / J. Wind Eng. Ind. Aerodyn. 93 (2005) 741756

The numerical results for a 451 and 1801 have also been compared with the experimental data, along the vertical and horizontal centrelines of the front and sidewalls, as shown in Figs. 1318. A reasonable agreement of the numerical results with experimental data has also been found in these cases, although the global performance has not been so good as noticed for a 01. The numerical simulations generally reproduce the experimental prole trends but some large discrepancies of the numerical results occur in particular situations: an overestimation of pressure in the exposed internal corner of the U-shaped model for a 451 (Fig. 13); a deviation of the vertical proles on the side walls of the U-shaped model for a 451 (Fig. 15); and a staggering of the vertical proles for both L- and U-shaped models for a 1801 (Figs. 17 and 18). The reasons for these

1.2 1 0.8 0.6 0.4 0.2 0 -0.2 0 -0.4

H Cp

Wind

A 0.5 1 x/L

B 1.5 2

L Num Exp

1.2 1 0.8 0.6 0.4 0.2 0 0 -0.2 -0.4

D C

Cp

C 0.5 1

D 1.5 x/L 2

E 2.5 3

L Num Exp

L Wind

Fig. 13. Pressure coefcients along the horizontal centreline (H/2) of the L- and U-shaped models for an incident angle of a 451.

1
A B

z/H

z/H

0.5

0.5 Wind
L L Num

0 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1 1.2 Cp

0 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 Cp

1 1.2

Exp

Fig. 14. Pressure coefcients along the vertical centrelines (L/2) of the L-shaped model for an incident angle of a 451.

1
D E

z/H

z/H

z/H

0.5

0.5

0.5
L

L Num Exp

0 -0.9 -0.7 -0.5 -0.3 -0.1 0.1 0.3 Cp

E 0 0 -0.9 -0.7 -0.5 -0.3 -0.1 0.1 0.3 -0.9 -0.7 -0.5 -0.3 -0.1 0.1 0.3 Cp Cp D

Wind

Fig. 15. Pressure coefcients along the vertical centrelines (L/2) of the U-shaped model for an incident angle of a 451.

ARTICLE IN PRESS
M.G. Gomes et al. / J. Wind Eng. Ind. Aerodyn. 93 (2005) 741756
Wind A 0.5 1 B 1.5 2 A B 0.4 0.2 0 -0.2 0 -0.4 -0.6 -0.8 -1 -1.2 Wind C 0.5 1 D 1.5 2 C D H

753

0.4 0.2 0 -0.2 0 -0.4 -0.6 -0.8 -1 -1.2

H Cp

Cp

L Num Exp x/L

L Num Exp x/L

Fig. 16. Pressure coefcients along the horizontal centreline (H/2) of the L- and U-shaped models for an incident angle of a 1801.

Wind 1 1 A z/H 0.5 z/H 0.5 B

L 0 -1.2 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 Cp A 0 -1.2 -1 -0.8 -0.6 -0.4 -0.2 0 Cp B 0.2 0.4 0.6 Num Exp

Fig. 17. Pressure coefcients along the vertical centrelines (L/2) of the L-shaped model for an incident angle of a 1801.

Wind C

z/H

z/H

0.5

0.5 L L Num Exp

0 -1.2 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 Cp

0 -1.2 -1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 Cp

Fig. 18. Pressure coefcients along the vertical centrelines (L/2) of the U-shaped model for an incident angle of a 1801.

discrepancies might be: (1) insufcient mesh renement in high velocity gradient zones, either in intensity or direction, leading to a loss of accuracy in the results; (2) numerical false diffusion due to the skewness of the ow to the gridlines in the case of a 451 [9]; (3) inability of the turbulent model used (RNG k2e) to deal with high vorticity ow regions (e.g. wake region and recirculation zones at the re-entrant corners). In fact, although the RNG k2e model prevents the overproduction of turbulent kinetic energy at high vorticity regions, it is still an isotropic eddy viscosity model, and so this can have direct inuence on accuracy of results in regions where the Reynolds stresses are highly anisotropic [16].

ARTICLE IN PRESS
754 M.G. Gomes et al. / J. Wind Eng. Ind. Aerodyn. 93 (2005) 741756

4.5. Numerically predicted ow pattern In Figs. 19 and 20 the numerically predicted velocity vector eld, on the horizontal midplane and for wind directions of 01, 451 and 1801, is also shown for the L- and U-shaped models, respectively. For all wind directions the wind ows sharply away at high speed from the front wall near the windward corners (separation and acceleration of ow on the corners) and reverses just behind these corners, giving rise to negative pressures. For normal wind incidence (a 01) two symmetrical vortices appear in the wake region of the U-shaped model, very similar to those of rectangular blocks (Fig. 20). However, for the L-geometry only a large and non-symmetrical vortex forms behind the model (Fig. 19). When the wind blows at a skew angle (a 451) a larger region of stagnant air forms in the re-entrant corner of the L-shaped model. The size of this region is highly dependent on the slenderness of the model: a taller model implies a larger stagnation zone in the reentrance, as the ow tends to contour the sides rather than ow into the cavity [2]. For a 1801, the wind ow eld around the L- and the U-shaped models is very similar to that of rectangular blocks in the upwind region. However, the ow patterns are totally different behind the models whereas, instead of two symmetrical vortices, there is a large vortex in the L- and U-cavity with smaller vortices appearing behind the downward back

Ur=10 m/s

=0

=45

=180

Fig. 19. Velocity vector eld around the L-shaped model on the horizontal mid-plane (H/2) for wind directions of 01, 451 and 1801.

Ur=10 m/s

=0

=45

=180

Fig. 20. Velocity vector eld around the U-shaped model on the horizontal mid-plane (H/2) for wind directions of 01, 451 and 1801.

ARTICLE IN PRESS
M.G. Gomes et al. / J. Wind Eng. Ind. Aerodyn. 93 (2005) 741756 755

wall of the wings of the models. All these phenomena are clearly predicted and supports the conclusions derived from the experimental data. 5. Conclusions The present study has shown that incidence of wind on L- and U-shaped models can induce different pressure distributions on their faces from those of single rectangular blocks. The work was carried out through wind tunnel testsfor a wide range of incidence angleson scale models with those two particular shapes and on a cube to represent the single block case and to validate the experimental process. A region of stagnant air, in which the pressure rises to the stagnant value, has occurred in the re-entrant corner of the L-shaped model for ap901. As the angle of the incident ow increases, the pressure eld turns out to be negative and almost uniformly distributed, which is characteristic of a recirculation area. The inuence of the additional wing transforming the L into the U-shaped model was noticeable on the pressure distribution: suctions on the inner faces occurred earlier for the tested angles and pressure in the recess was almost constant for aX901. This derives from the tendency of the ow to skip past the U gap leaving stagnant ow in the recess. Results of a CFD numerical approach based on the RNG k2e turbulence model showed a general good agreement for normal wind incidence (a 01) whereas some differences have occurred for other directions. Denser mesh arrangements in particular ow regions and an improved turbulent model that accounts for the Reynolds-stress anisotropy could certainly enhance the quality of the numerical results. Nevertheless, the results obtained are within the expectations and, together with the experimental data, can provide useful information about wind pressures on such irregular-plan shapes. Acknowledgements The authors would like to express their appreciation to Professor Jorge Saraiva and Eng. Marques da Silva for their valuable suggestions and to Mila for the precious help she gave with the wind tunnel experiments performed at LNEC. References
[1] R.N. Meroney, Wind-tunnel modeling of the ow about bluff bodies, in: F.R.G. Aachen (Ed.), Advances in Wind Engineering, Proceedings of the Seventh International Congress on Wind Engineering, July 610, 1987, Part 2, Elsevier, Amsterdam, 1988, pp. 203223. [2] N.J. Cook, The Designers Guide to Wind Loading of Building Structures. Part 2: Static Structures, Butterworths Press, London, 1985. [3] T. Stathopoulos, Y. Zhou, Numerical simulation of wind-induced pressures on buildings of various geometries, J. Wind Eng. Ind. Aerodynam. 46 and 47 (1993) 419430. [4] M.G. Gomes, Wind effects on buildings. Experimental evaluation of pressure coefcients for L- and U shaped buildings, M.Sc. Thesis, Instituto Superior Tecnico (IST), Technical University of Lisbon, 2003 (in Portuguese). [5] E. Houghton, N. Carruthers, Wind Forces on Buildings and Structures. An Introduction, Edward Arnold, London, 1976. [6] J.B. Barlow, W.H. Rae Jr., A. Pope, Low-Speed Wind Tunnel Testing, third ed, Wiley, New York, 1999. [7] D. Spalding, J. Ludwig, PHOENICS 3.4 Documentation, CHAM, 2001.

ARTICLE IN PRESS
756 M.G. Gomes et al. / J. Wind Eng. Ind. Aerodyn. 93 (2005) 741756 [8] V. Yakhot, S.A. Orszag, S. Thangam, T.B. Gatski, C.G. Speziale, Development of turbulence models for shear ows by a double expansion technique, Phys. Fluids: A: Fluid Dyn. 4 (7) (1992) 15101520. [9] S.V. Patankar, Numer. Heat Transfer Heat Flow, Hemisphere Publishing Corporation, New York, 1980. [10] D.B. Spalding, Mathematical modelling of uid-mechanics, Heat-transfer and chemical-reaction processes, CFDU Report HTS/80/1, Imperial College, London, 1980. [11] S.V. Patankar, D.B. Spalding, A calculation procedure for heat, mass and momentum transfer in threedimensional parabolic ows, Int. J. Heat Mass Transfer 15 (1972) 17871806. [12] B.E. Launder, D.B. Spalding, The numerical computation of turbulent ows, Comput. Methods Appl. Mech. Eng. 3 (1974) 269289. [13] I.P. Castro, A.G. Robins, The ow around a surface-mounted cube in uniform and turbulent streams, J. Fluid Mech. 79 (2) (1977) 307335. [14] Y.Q. Zhang, A.H. Huber, S.P.S. Arya, W.H. Snyder, Numerical simulation to determine the effects of incident wind shear and turbulence level on the ow around a building, J. Wind Eng. Ind. Aerodyn. 46 and 47 (1993) 129134. [15] S. Murakami, A. Mochida, 3-D numerical simulation of airow around a cubic model by means of the k e model, J. Wind Eng. Ind. Aerodyn. 31 (1988) 283303. [16] N.G. Wright, G.J. Easom, Comparison of several computational turbulence models with full-scale measurements of ow around a building, Wind Struct. 2 (4) (1999) 305323.