at SciVerse ScienceDirect

Building and Environment 60 (2013) 137e149

Contents lists available

Building and Environment

journal homepage: www.elsevier .com/locate/bui ldenv

CFD simulation of wind-induced pressure coefficients on buildings with andwithout balconies: Validation and sensitivity analysis

H. Montazeri*, B. BlockenBuilding Physics and Services, Eindhoven University of Technology, P.O. Box 513, 5600 MB Eindhoven, The Netherlands

a r t i c l e i n f o

Article history:Received 10 July 2012Received in revised form11 October 2012Accepted 20 November 2012

Keywords:Computational fluid dynamics (CFD)Building aerodynamicsWind pressure coefficientSensitivity studyVerification and validation

* Corresponding author. Tel.: þ31 (0) 40 247 5790;E-mail address: h.montazeri@tue.nl (H. Montazeri

0360-1323/$ e see front matter � 2012 Elsevier Ltd.http://dx.doi.org/10.1016/j.buildenv.2012.11.012

a b s t r a c t

Knowledge of the pressure distribution on building walls is important for the evaluation of wind loadsand natural ventilation. Wind-induced pressure distributions are influenced by a wide range of factorsincluding approach-flow conditions, urban surroundings and building geometry. Computational FluidDynamics (CFD) can be a valuable tool for determining mean wind pressure coefficients on buildingfacades. However, while many CFD studies of mean wind pressure on buildings have been performed inthe past, the vast majority of these studies focused on simple building geometries without facade detailssuch as balconies. These details however can drastically influence the flow pattern and the overallpressure distribution on the facade. This paper presents a systematic evaluation of 3D steady Reynolds-Averaged NaviereStokes (RANS) CFD for predicting mean wind pressure distributions on windward andleeward surfaces of a medium-rise building with and without balconies. The evaluation is based ona grid-sensitivity analysis and on validation with wind-tunnel measurements. It is shown that buildingbalconies can lead to very strong changes in wind pressure distribution, because they introduce multipleareas of flow separation and recirculation across the facade. The results show that steady RANS, in spiteof its limitations, can accurately reproduce the mean wind pressure distribution across the windwardfacade of the building. The average deviations from the wind-tunnel measurements are 12% and 10% forthe building with and without balconies, respectively. In addition, also the important impact of thereference static pressure and the turbulence model are demonstrated.

� 2012 Elsevier Ltd. All rights reserved.

1. Introduction

Knowledge of the pressure distribution on building walls isessential to evaluate wind-induced natural ventilation and toassess wind loads on building walls and building components (e.g.[1e8]). As an example, Building Energy Simulation (BES) programsrequire pressure coefficient data as input for analysing ventilationand infiltration flow rates [2]. Similarly, design standards need datawith a high accuracy for effective-cost designs and reduction ofwind damage and cost to building components [9,10].

The pressure distribution on building walls is influenced byawide range of factors including approach-flow conditions [11e13],urban surroundings [14], building geometry [1] and wind direction[15]. In particular, building facade details such as balconies andother protrusions can affect the peak and mean surface pressuredistributions on buildings walls and roofs [16e18].

fax: þ31 (0) 40 243 8595.).

All rights reserved.

Pressure coefficients can be determined using full-scale on-sitemeasurements [15,19e26], reduced-scale wind-tunnel measure-ments [27e32] or numerical simulation with Computational FluidDynamics (CFD) [13,16,33e37]. Full-scale measurements offer theadvantage that the real situation is studied and the full complexity ofthe problem is taken into account. However, full-scale measure-ments are usually only performed in a limited number of points inspace. In addition, there is no or only limited control over theboundary conditions [38]. Reduced-scale wind-tunnel measure-ments allowa strong degree of control over the boundary conditions,however at the expense of e sometimes incompatible e similarityrequirements. Furthermore, wind-tunnel measurements are usuallyalso only performed in a limited set of points in space [13]. CFD on theother hand provides whole-flow field data, i.e. data on the relevantparameters in all points of the computational domain [5,39,40].Unlike wind-tunnel testing, CFD does not suffer from potentiallyincompatible similarity requirements because simulations can beconducted at full scale. CFD simulations easily allow parametricstudies to evaluate alternative design configurations, especiallywhen the different configurations are all a priori embedded withinthe same computational domain and grid (see e.g. [41]). CFD is

Fig. 1. Geometry of building model and balconies (dimensions in metres at modelscale).

H. Montazeri, B. Blocken / Building and Environment 60 (2013) 137e149138

increasingly used to study a wide range of atmospheric and envi-ronmental processes. Examples are pedestrian wind comfort andwind safety around buildings [40,42e46], natural ventilation ofbuildings [5,41,47e53], air pollutant dispersion [54e58], convectiveheat transfer [59e61], etc. In some of these studies, CFD was appliedand evaluated in detail, including verification, validation and sensi-tivity analyses. CFD has also been used onmany occasions in the pastto determinemeanwind-induced pressure distributions on buildingfacades. However, the vast majority of these studies focused onrelatively simple building shapes and plane, smooth facades withoutprotrusions or recessions (e.g. [34e36,62,63]). Nevertheless, manyhistorical and contemporary building facades are characterized byprotrusions and recessions. To the best of our knowledge, a detailedevaluation of steady Reynolds-averaged NaviereStokes (RANS) CFDhas not yet been performed for meanwind pressure distributions onsuch building facades.

This paper therefore presents a systematic and detailed evalu-ation of 3D steady RANS CFD for predicting mean wind pressuredistributions on building facades with and without balconies forboth normal and obliquely approach-flow conditions. The evalua-tion is based on a grid-sensitivity analysis and on validation withwind-tunnel measurements by Chand et al. [17]. The impact ofseveral computational parameters is also investigated, includingthe resolution of the computational grid, the reference staticpressure and the turbulence model.

In Section 2, the wind tunnel experiments by Chand et al. [17]are briefly outlined. Section 3 presents the computational settingsand parameters for the reference case, and the validation of the CFDresults with the wind-tunnel measurements. In Section 4, thesensitivity analysis is performed, including the influence of buildingbalconies on the wind pressure distribution. A discussion on thelimitations of the study is given in Section 5. The main conclusionsare presented in Section 6.

2. Description of wind tunnel experiments

Atmospheric boundary layer wind-tunnel measurements ofwind-induced surface pressure on the facades of a medium-risebuilding were conducted by Chand et al. [17]. The open-circuitwind tunnel was 14 m long and had a test section of 2.5 � 1.8 m2.The atmospheric boundary layer was generated by a combinationof three devices: vortex generators, a grid of horizontal rods anda set of roughness elements on the floor of the test section. Theresulting vertical profile of mean wind speed at the location of thebuilding (but without building model present) is represented bya log law with aerodynamic roughness length z0 ¼ 0.008 m (modelscale, corresponding to 0.24 m in full scale) and a friction velocityu*ABL ¼ 0.73 m/s. The measured incident longitudinal turbulenceintensity ranges from 13% near ground level to about 3% at gradientheight. Because these profiles were measured at the (virtual)location of the building, they represent the incident, rather than theapproach-flow conditions. Using the incident-flow conditions inCFD is important for simulation accuracy [64]. The upstream windvelocity, measured at building height, was equal to 7.1 m/s, yieldinga building Reynolds number of 250,000 which is well above thecritical value of 11,000 for Reynolds number independent flow [65].

The building at scale 1:30 had dimensions width � depth �height ¼ 0.60 � 0.25 � 0.50 m3 (reduced scale, see Fig. 1) corre-sponding to full-scale dimensions 18 � 7.5 � 15 m3, resulting ina blockage ratio of about 6.6%. To evaluate the effect of buildingbalconies on the mean pressure coefficient, measurements werecarried out for a building with and without balconies. Threebalconies with width 0.15 m, depth 0.05 m and height 0.03 mwerepositioned at every one of the five floors, except the ground floor(Fig. 1).

Mean surface pressures were measured along three verticallines on the windward and leeward facade. Each measurement linewas positioned in the middle of the balconies and 45 holes weredrilled at equidistant points along it (Fig. 1). In the remainder of thispaper, we will refer to these vertical lines as “edge lines” and“centre line”. The measurements were performed with a scanningvalve and a digital micro-manometer. Upstream static and dynamicpressures were measured with a Pitot tube mounted 0.90 mupstream of the model and at building height. During the surfacepressure measurements, the static tube was connected to thenegative port of the scanning valve. So, the manometer indicatedthe pressure differences of surface pressure and free stream staticpressure. The results of the wind-tunnel measurements will beshown together with the validation in the next sections.

3. CFD simulations: reference case

A reference case is defined as a starting point for the sensitivityanalysis. It includes a fixed choice for the computational geometryand grid, boundary conditions and turbulence model, as outlinedbelow.

3.1. Computational geometry and grid

A computational model was made of the reduced-scale buildingmodel used in the wind-tunnel measurements. The dimensions ofthe computational domain were chosen based on the best practiceguidelines by Franke et al. [66] and Tominaga et al. [67]. Theupstream domain length is 5H¼ 2.5 m. The resulting dimensions ofthe domain were W � D � H ¼ 10.6 � 10.25 � 3 m3, which corre-sponds to 318 � 307.5 � 90 m3 in full scale. The computational gridwas created using the surface-grid extrusion technique presentedby van Hooff and Blocken [41]. The procedure was executed withthe aid of the pre-processor Gambit 2.4.6, resulting in a hybrid gridwith 2,102,250 prismatic and hexahedral cells. The grid is shown inFig. 2aec. 20 and 10 cells are used along the width and depth of thebalconies, respectively, as shown in Fig. 2c. A maximum stretchingratio of 1.2 controls the cells located in the immediate surroundingsof the building model. The grid resolution resulted from a grid-sensitivity analysis that will be outlined in Section 4.1. Theminimum and maximum cell volumes in the domain are approxi-mately 5 � 10�8 m3 and 7.5 � 10�2 m3, respectively. The distancefrom the centre point of the wall adjacent cell to the wall, for thewindward, leeward and ground plane is 0.0017 m, 0.0022 m and0.0025 m, respectively. This corresponds to y* values between 20and 350. As standard wall functions are used in this study, thesevalues ensure that the centre point of the wall-adjacent cell isplaced in the logarithmic layer. The domain shape (Fig. 2a) allows

Fig. 2. Computational grid. (a) Grid at bottom and side faces of computational domain. (b) Grid at building surfaces and ground surface. (c) Detail of grid near balconies.

H. Montazeri, B. Blocken / Building and Environment 60 (2013) 137e149 139

modelling different wind directions (0� and 45�, corresponding tothe experiments).

3.2. Boundary conditions

For wind direction q ¼ 0�, plane 1 in Fig. 2a is the inlet plane,plane 3 the outlet plane and planes 2 and 4 are the side planes. Inthe simulations the inlet boundary conditions (mean velocity U,turbulent kinetic energy k and turbulence dissipation rate 3) werebased on the measured incident vertical profiles of mean windspeed U and longitudinal turbulence intensity Iu (Fig. 3) Theturbulent kinetic energy k was calculated from U and Iu using Eq.(1), where a is a parameter in the range between 0.5 and 1.5 [50]. Inthis study, as recommended by Tominaga et al. [67], a¼ 1 is chosen.The turbulence dissipation rate 3is given by Eq. (2), with k the vonKarman constant (¼0.42).

Fig. 3. (a) Measured profile (dotted line) and fitted log law profile (solid line) of ratio of m(b) turbulent kinetic energy k and (c) turbulence dissipation rate 3.

kðzÞ ¼ aðIuðzÞuðzÞÞ2 (1)

3ðzÞ ¼ u*3ABLkðzþ z0Þ

(2)

For the ground surface, the standard wall functions by Launder andSpalding [68] with roughnessmodification by Cebeci and Bradshaw[69] are used. The values of the roughness parameters, i.e. the sand-grain roughness height ks (m) and the roughness constant Cs, weredetermined using their consistency relationship with the aero-dynamic roughness length z0 derived by Blocken et al. [70]. ForFluent 6.3, this relationship is:

ks ¼ 9:793z0Cs

(3)

ean wind speed U to mean wind speed UH at building height. Inlet vertical profile of

Fig. 4. Simulated vertical profile of static pressure upstream of the building.

H. Montazeri, B. Blocken / Building and Environment 60 (2013) 137e149140

Given the upstream domain length of 5H, the roughnessparameters are taken to reproduce the roughness of the wind-tunnel turntable, which is considered to be smooth and for whichz0 ¼ 0.0018 m is assumed. The selected (reduced-scale) valuesaccording to Eq. (3) are ks ¼ 0.0025 m and Cs ¼ 7.0. Standard wall

Fig. 5. Comparison of pressure coefficient (CP) by CFD simulation results and wind-tunnefacade; (c) edge lines on leeward facade; (d) centre line on leeward facade.

functions are also used at the building surfaces, but with zeroroughness height ks ¼ 0 (Cs ¼ 0.5). Zero static pressure is applied atthe outlet plane. Symmetry conditions, i.e. zero normal velocity andzero normal gradients of all variables, are applied at the top andlateral sides of the domain. As recommended by Blocken et al. [70e72], the absence of unintended streamwise gradients (i.e. hori-zontal inhomogeneity) in the vertical profiles of mean wind speedand the turbulence parameters was confirmed by performingsimulations in an empty domain (i.e. without building present).

3.3. Solver settings

The commercial CFD code Fluent 6.3.26 was used to perform thesimulations. The 3D steady RANS equations were solved incombinationwith the realizable ke 3turbulencemodel by Shih et al.[73]. The SIMPLE algorithm was used for pressureevelocitycoupling, pressure interpolation was second order and second-order discretization schemes were used for both the convectionterms and the viscous terms of the governing equations. Conver-gence was assumed to obtain when all the scaled residuals levelledoff and reached a minimum of 10�6 for x, y momentum, 10�5 for ymomentum and 10�4 for k, 3and continuity.

3.4. Results and comparison with wind tunnel experiments

The CFD results for the reference case are compared with thewind-tunnel measurements by Chand et al. [17]. The pressurecoefficients are computed as CP ¼ (P � P0)/(0.5rUref

2 ) where P is the

l experiments along (a) edge lines on windward facade; (b) centre line on windward

Fig. 6. Computational grids for grid-sensitivity analysis. (a,d) Coarse grid; (b,e) basic grid; (c,f) fine grid.

H. Montazeri, B. Blocken / Building and Environment 60 (2013) 137e149 141

pressure at the surface, P0 the reference static pressure, r¼ 1.225 kg/m3 the air density and Uref is the reference wind speed at buildingheight (Uref ¼ 7.1 m/s at z¼ 0.5 m). Fig. 4 provides the CFD results ofstatic pressure along a vertical line 0.9 m upstream of the building,where the Pitot tube for the reference static pressure was mountedin the experiment. It can be seen that the static pressure at buildingheight is about 4.3 Pa. Note that Chand et al. [17] did not report thestatic pressure that they used to calculate the CP values. In thepresent study, we used the value of 4.3 Pa for this purpose.

Fig. 5 compares the CFD results and the wind-tunnel results ofCP along the vertical measurement lines at the facade (shown inFig. 1) for the case with balconies. The general agreement is quitegood. For the lines at the windward facade (Fig. 5a and b), theaverage absolute deviation between CFD results andmeasurementsis 0.052 and 0.072 for the edge lines and centre line, respectively. Inparticular the vertical CP gradients (increase and decrease of CPalong these vertical lines) due to the presence of the balconies arequite well reproduced. Note that there are some discrepanciesespecially at the lower half of the facade at edge lines, where CFDoverestimates CP, and at the upper half of the facade at the centreline, where CFD provides underestimations. The exact reasons forthis are not clear, but it is possible that the expected acceleration ofthe flow over the smooth turntable of the wind tunnel, directlyupstream of the building model, has had some effect on the

Fig. 7. Results for grid-sensitivity analysis: pressure coefficient (CP) v

measurement values. Overall, in spite of the well-known defi-ciencies of steady RANS CFD to reproduce separating and recircu-lating flow [67,74], the agreement between the CFD and the wind-tunnel results is considered to be good. Also at the leeward facade,the agreement is quite good (Fig. 5c and d). The average absolutedeviation between CFD results and measurements is 0.069 and0.070 for the edge lines and centre line, respectively. Note thatthere is a systematic underestimation of the absolute value of CP byCFD at the leeward facade.

4. CFD simulations: sensitivity analysis

To analyse the sensitivity of the results to various geometricaland computational parameters, systematic changes are made to thereference case that was outlined in the previous section. In everysection, one of the geometrical or computational parameters isvaried, while all others are kept identical to those in the referencecase.

4.1. Impact of computational grid resolution

Performing a grid-sensitivity analysis is important to reduce thediscretization errors and the computational time. In this study,a grid-sensitivity analysis was performed based on two additional

alues along (a) edge line and (b) centre line for the three grids.

Fig. 8. Impact of static reference pressure on CFD simulation results of pressure coefficient (CP) along (a) edge line and (b) centre line.

H. Montazeri, B. Blocken / Building and Environment 60 (2013) 137e149142

grids; a coarser grid and a finer grid. Coarsening and refining wasperformed with an overall linear factor O2. As mentioned before,the basic mesh had 2,102,250 cells. The coarse grid had 720,937cells, while the fine grid had 6,755,370 cells. The three grids areshown in Fig. 6. The results for CP on the three grids are comparedin Fig. 7, indicating only a very limited dependence of the results onthe grid resolution. The average absolute deviation between theCFD results and the measurements for the edge lines is 0.056, 0.052and 0.052 for the coarse, basic and fine grid, respectively. For thecentre line, the average absolute deviations are 0.068, 0.072 and

Fig. 9. Impact of turbulence model on CFD simulation results of pressure

0.073. A small deviation is between the coarse grid and basic gridfor the upper part of the building, and between fine grid and basicgrid for the space between the third and fourth floors. Negligiblegrid sensitivity is found for the other parts. Therefore, the basic gridis retained for further analysis.

4.2. Impact of reference static pressure

As mentioned before, the actual value of the reference staticpressure in the wind tunnel is not known, as it was not reported in

coefficient (CP) along (a and b) edge line and (c and d) centre line.

Table 1The average absolute deviations for the edge line and centre line for each turbulencemodel.

Turbulence model Edge line Centre line

Standard ke 3model 0.044 0.071Realizable ke 3model 0.052 0.072RNG ke 3model 0.067 0.070Standard keu model 0.098 0.087Reynolds stress model 0.064 0.067

H. Montazeri, B. Blocken / Building and Environment 60 (2013) 137e149 143

the paper by Chand et al. [17]. The paper does mention that it wasmeasured by a Pitot tube placed 0.9 m upstream of the building andat building height. In the present paper, the CFD result of the staticpressure at this position of the Pitot tube was used as the referencestatic pressure (see Subsection 3.4). In the wind-tunnel experi-ments, it was assumed that the presence of the model had noinfluence on the measurement of upstream static pressure at thisposition. However, the simulation results (see Fig. 4) showed thatthe model affects the upstream flow causing an increase in staticpressure. To illustrate the importance of a correct reference valuefor this static pressure, Fig. 8 shows the results obtained withPref ¼ 1, 3, 5 and 7 Pa. For the edge lines, the average absolutedeviations between the CFD results and the measurements are0.165, 0.103, 0.052 and 0.051, for 1, 3, 5 and 7 Pa, respectively. Forthe centre line, these deviations are 0.122, 0.076, 0.072 and 0.099.These results show that a small change in the upstream staticpressure can lead to very large deviations in CP. This implies thataccurate values for this reference static pressure are crucial for

Fig. 10. Comparison of pressure coefficient (CP) by CFD simulation results and wind-tunnel ecentre line on windward facade; (c) edge lines on leeward facade; (d) centre line on leewa

a successful CFD validation effort. For CFD validation purposes, it isimportant that the position of Pref is the same as in the experiments.In general, a reference position that is outside the wind-flowpattern disturbed by the building model should be used.

4.3. Impact of turbulence model

3D steady RANS simulations were made in combination withfive turbulence models: (1) the standard ke 3model (Ske 3) [75]; (2)the realizable ke 3 model (Rke 3) [73]; (3) the renormalizationGroup ke 3model (RNG ke 3) [76,77]; (4) the standard keu model(Skeu) [78] and (5) the Reynolds Stress Model (RSM) [79]. For thestandard keu model, the inlet vertical profile for u is determinedfrom that of k and 3using Eq. (4), where Cm is a constant equal to0.09. For the RSM model, the Reynolds stress components are ob-tained from the turbulent kinetic energy k assuming isotropy ofturbulence (see Eq. (5)):

uðzÞ ¼ 3ðzÞCmkðzÞ (4)

u0i2 ¼ 2

3k ði ¼ 1;2;3Þ

u0iu0j ¼ 0:0

(5)

The results are shown in Fig. 9. The average absolute deviationsfor the edge lines and centre line for each turbulence model are

xperiments for building without balconies along (a) edge lines on windward facade; (b)rd facade.

H. Montazeri, B. Blocken / Building and Environment 60 (2013) 137e149144

given in Table 1. The differences between the models are mostpronounced near ground level, where the RNG ke 3model tends tooverestimate the pressure variations. The standard keu modelgenerally provides a slight overestimate of the pressure, while theresults of the RSM are very close to those of the standard andrealizable ke 3model (reference case).

4.4. Impact of building balconies for perpendicular approach flow

The impact of building balconies on CP is investigated bycomparing simulations for buildings with and without balconies.This subsection presents the results for perpendicular approachflow. The results are displayed in Figs. 5 and 10e12. The followingobservations are made:

� Fig.10 compares the simulated andmeasured CP along the edgelines and centre line for the facade without balconies. For boththe windward and the leeward facade, a good to very goodagreement with the measurements is obtained. The maindeviations are again found at the lower half of the facade. Theaverage absolute deviations in Fig. 10aed are 0.045, 0.046,0.039 and 0.055, respectively.

Fig. 11. (a) Pressure coefficient (CP) distribution across windward facade of building with bfield in cross-section (centre plane). (c) (CP) distribution for building without balconies. (d)

� Comparing Figs. 5 and 10 clearly shows the very large impact ofthe building balconies on the CP distribution. The presence ofthe balconies yields a very different CP profile along the heightof the building facade, with a succession of lower and higherpressure zones, caused by the multitude of flow separation andrecirculation areas induced by the balconies.

� Fig. 11 compares the simulated distribution of CP across theentire windward facade for the case with and without balco-nies. This figure confirms the much larger complexity of the CPdistribution across the facade with balconies. In particular:

1) For the second and third row of balconies, the presence ofthese balconies generally leads to an increase in CP. The reasonis the direct impingement of the flow onto these balconies andonto the facade behind them, yielding a large stagnation area.To some extent, the balconies act as compartments witha nearly constant pressure. This effect is most pronouncedfor the third row of balconies, but also present for the secondrow.

2) For the first and fourth row of balconies, flow separation andreattachment can lead to local increases and decreases in CP.Especially for the side balconies, large gradients areobserved.

alconies, for wind direction perpendicular to the windward facade. (b) Velocity vectorSame as (b), but for building without balconies.

Fig. 12. Impact of balconies on the surface-averaged pressure coefficients (CP) for each balcony space, for wind direction perpendicular to the windward facade. The surface-averaged CP for the case (a) with balconies and (b) without balconies. (c) The relative percentage difference of surface-averaged CP for the two cases.

H. Montazeri, B. Blocken / Building and Environment 60 (2013) 137e149 145

� Fig. 12 shows the impact of balconies on the surface-averagedpressure coefficients for each balcony space. The width of thesurfaces is equal to the width of each balcony, i.e. 0.15 m. Theheight of the surfaces is 0.1 m. The surface-averaged CP valuesare presented for the two cases: with balconies (Fig. 12a) andwithout balconies (Fig. 12b). Fig. 12c also shows the relativepercentage difference of surface-averaged CP for the two cases.The results show that the presence of the fourth row ofbalconies leads to a substantial decrease in the surface-averageCP of about 30%. The third row of balconies, however, is situatedcloser to the stagnation area and the change in surface-averaged CP is therefore negligible. For the second and firstrow of balconies, their presence substantially increases thesurface-averaged CP except for the middle balconies.

4.5. Impact of building balconies for oblique approach flow

Simulations andmeasurements were also made for oblique flow(wind direction 45�). The results are provided in Figs. 13e16. Thefollowing observations are made:

� Fig. 13aec compare the simulated and measured CP along thevertical lines in the middle and at the edges of the windwardfacade. A good to very good agreement is obtained between thesimulations and the measurements. The average absolutedeviations in Fig.13aec are 0.028, 0.029 and 0.028, respectively.

� Fig. 13def compare the simulated and measured CP for theleeward facade. Here, as opposed to the casewith perpendicularflow, the agreement is not good. The average absolute deviationsin Fig. 13def are 0.095, 0.155 and 0.068, respectively.

� Fig. 14aec show the results for the windward facade withbalconies. The average absolute deviations in Fig. 14aec are0.035, 0.058 and 0.039, respectively.

� Fig. 14def show the results for the leeward facade withbalconies, with averaged absolute deviations of 0.095, 0.156and 0.068, respectively.

� Overall, a good agreement is obtained, although locally somesignificant discrepancies are noted. Comparing Figs. 13 and 14again clearly shows the large impact of the building balconieson the CP distribution.

� Fig. 15 finally compares the simulated distribution of CP acrossthe entire windward facade for the case with and withoutbalconies. The complexity of the distribution is directlyattributable to multiple areas of flow separation, recirculationand reattachment on the windward building facade.

� Fig. 16 shows the results on the surface-averaged pressurecoefficients. The results show that the impact of balconies forobliqueflow ismuchmore complex than for perpendicularflow.

5. Discussion

It is important to mention the three main limitations of thisstudy.

(1) The study only considered steady RANS CFD simulations, as thepurpose was to investigate how well steady RANS CFD wouldbe able to reproduce wind pressure distributions acrossbuilding facades, especially for facades with balconies. In spiteof the well-known deficiencies of steady RANS, a good agree-ment was obtained between CFD simulations and wind-tunnel

Fig. 13. Comparison of pressure coefficient (CP) by CFD simulation results and wind-tunnel experiments for building without balconies and for oblique wind direction 45� , along: (a)windward upstream edge line, (b) windward centre line, (c) windward downstream edge line, (def) same lines at leeward facade.

Fig. 14. Comparison of pressure coefficient (CP) by CFD simulation results and wind-tunnel experiments for building with balconies and for oblique wind direction 45� , along: (a)windward upstream edge line, (b) windward centre line, (c) windward downstream edge line, (def) same lines at leeward facade.

H. Montazeri, B. Blocken / Building and Environment 60 (2013) 137e149146

Fig. 15. (a) Pressure coefficient (CP) distribution across windward facade of building with balconies, for wind direction at 45� to the windward facade. (b) CP distribution for buildingwithout balconies.

H. Montazeri, B. Blocken / Building and Environment 60 (2013) 137e149 147

measurements, both for the windward facade without balco-nies and for the windward facade with balconies. Also for theleeward facade and for perpendicular wind, a good agreementwas obtained. However, larger deviations were obtained forwind pressures on the leeward facade for oblique wind.Obtaining a better agreement herewould necessitate the use ofLarge Eddy Simulation (LES), which however is much morecomputationally expensive than steady RANS.

Fig. 16. Impact of balconies on the surface-averaged pressure coefficients for each balcony scase (a) with balconies and (b) without balconies. (c) The relative percentage difference of

(2) Only an isolated medium-rise building was considered. Furtherwork should assess the accuracy of steady RANS to reproducethe effect of balconies on pressure distributions on high-risebuildings, and on buildings surrounded by other buildings.

(3) The explicit focus of the study was the assessment of steadyRANS CFD for a given building geometry with given balconygeometries. Future work could focus on optimal balconydesign. In order to optimize the arrangement of balconies,

pace, for wind direction at 45� to the windward facade. The surface-averaged CP for thesurface-averaged CP for the two cases.

H. Montazeri, B. Blocken / Building and Environment 60 (2013) 137e149148

a parametric analysis needs to be performed. This optimizationshould be done by taking the impact of different factors intoaccount such as building geometry, position, dimension andnumber of balconies, etc. Besides the impact of balconies on thesurface pressure distribution of the building, to achieve anoptimum arrangement of balconies, wind comfort and windsafety assessment also need to be studied.

In spite of these limitations, the present study has analysed thepossibilities and limitations of steady RANS for assessing the effectof balconies on wind-induced pressure coefficients. It has alsoinvestigated the effect of the balconies on these coefficients bycomparing simulations for buildings with and without balconies.Finally, also a detailed sensitivity study has been performed,including the most important computational parameters such ascomputational grid, turbulence model, and wind direction. Notethat the focus of this paper was explicitly on the pressure distri-bution across the building facade. Further studies could includeanalysis of the static pressure distribution in the area around thebuilding (i.e. not at the building facade) and of the wind-velocitypattern.

6. Conclusions

This paper has presented a systematic evaluation of 3D steadyRANS CFD for the prediction of the mean wind pressure distribu-tion on windward and leeward surfaces of a medium-rise buildingwith and without balconies. The evaluation is based on a grid-sensitivity analysis and on validation with wind-tunnel measure-ments. The study was motivated by the lack of knowledge on theaccuracy and reliability of CFD for determining mean wind pres-sure coefficients on building facades with balconies. Althoughindeed many CFD studies of mean wind pressure distributions onbuildings have been performed in the past, the vast majority ofthese studies focused on simple building geometries withoutfacade details such as balconies. These details however areimportant because they can drastically change the flow patternand the overall pressure distribution on the facade. In addition,many historical and contemporary building facades are charac-terized by protrusions and recessions.

The present study has shown that 3D steady RANS CFD, in spiteof its limitations, is suitable to predict the wind-induced meanpressures at windward building facades with (and without)balconies. It has also been shown that the presence of buildingbalconies can indeed lead to very strong changes in wind pressuredistribution on these windward facades, because the balconiesintroduce multiple areas of flow separation, recirculation andreattachment. 3D steady RANS CFD has also been shown to provideaccurate predictions of the meanwind pressure at the leeward wallin case of a perpendicular approach flow wind direction. Thishowever is not the case for oblique flow, where large discrepancieswith the wind-tunnel measurements have been found. Finally, alsothe impact of the turbulence model, the reference static pressureand the wind direction have been investigated, and it has beenshown that a careful selection of these parameters is very impor-tant for accurate and reliable results.

References

[1] Uematsu Y, Isyumov N. Wind pressures acting on low-rise buildings. J WindEng Ind Aerodyn 1999;82:1e25.

[2] Cóstola D, Blocken B, Hensen JLM. Overview of pressure coefficient data inbuilding energy simulation and airflow network programs. Build Environ2009;44:2027e36.

[3] Montazeri H, Azizian R. Experimental study on natural ventilation perfor-mance of one-sided wind catcher. Build Environ 2008;43:2193e202.

[4] Montazeri H, Montazeri F, Azizian R, Mostafavi S. Two-sided wind catcherperformance evaluation using experimental, numerical and analyticalmodeling. Renew Energy 2010;35:1424e35.

[5] Chen Q. Ventilation performance prediction for buildings: a method overviewand recent applications. Build Environ 2009;44:848e58.

[6] Linden PF. The fluid mechanics of natural ventilation. Annu Rev Fluid Mech1999;31:201e38.

[7] Hunt GR, Linden PP. The fluid mechanics of natural ventilationddisplacementventilation by buoyancy-driven flows assisted by wind. Build Environ 1999;34:707e20.

[8] Karava P, Stathopoulos T, Athienitis AK. Wind driven flow through openings ea review of discharge coefficients. Int J Vent 2004;3:255e66.

[9] Ginger JD, Letchford CW. Net pressures on a low-rise full-scale building.J Wind Eng Ind Aerodyn 1999;83:239e50.

[10] Chen X, Zhou N. Equivalent static wind loads on low-rise buildings based onfull-scale pressure measurements. Eng Struct 2007;29:2563e75.

[11] Letchford CW, Mehta KC. The distribution and correlation of fluctuatingpressures on the Texas tech building. J Wind Eng Ind Aerodyn 1993;50:225e34.

[12] Wu F, Sarkar PP, Mehta KC, Zhao Z. Influence of incident wind turbulence onpressure fluctuations near flat-roof corners. J Wind Eng Ind Aerodyn 2001;89:403e20.

[13] Stathopoulos T. Computational wind engineering: past achievements andfuture challenges. J Wind Eng Ind Aerodyn 1997;67e68:509e32.

[14] Kim YC, Yoshida A, Tamura Y. Characteristics of surface wind pressures onlow-rise building located among large group of surrounding buildings. EngStruct 2012;35:18e28.

[15] Levitan ML, Mehta KC, Vann WP, Holmes JD. Field measurements of pressureson the Texas tech building. J Wind Eng Ind Aerodyn 1991;38:227e34.

[16] Stathopoulos T, Zhu X. Wind pressures on building with appurtenances.J Wind Eng Ind Aerodyn 1988;31:265e81.

[17] Chand I, Bhargava PK, Krishak NLV. Effect of balconies on ventilation inducingaeromotive force on low-rise buildings. Build Environ 1998;33:385e96.

[18] Maruta E, Kanda M, Sato J. Effects on surface roughness for wind pressure onglass and cladding of buildings. J Wind Eng Ind Aerodyn 1998;74e76:651e63.

[19] Levitan ML, Mehta KC. Texas tech field experiments for wind loads part 1:building and pressure measuring system. J Wind Eng Ind Aerodyn 1992;43:1565e76.

[20] Maruyama T, Taniguchi T, Okazaki M, Taniike Y. Field experiment measuringthe approaching flows and pressures on a 2.4 m cube. J Wind Eng Ind Aerodyn2008;96:1084e91.

[21] Richardson GM, Robertson AP, Hoxey RP, Surry D. Full-scale and modelinvestigations of pressures on an industrial/agricultural building. J Wind EngInd Aerodyn 1990;36(Part 2):1053e62.

[22] Richards PJ, Hoxey RP. Pressures on a cubic buildingdpart 1: full-scale results.J Wind Eng Ind Aerodyn 2012;102:72e86.

[23] Richards PJ, Hoxey RP. Wind loads on the roof of a 6 m cube. J Wind Eng IndAerodyn 2008;96:984e93.

[24] Richards PJ, Hoxey RP, Short LJ. Wind pressures on a 6 m cube. J Wind Eng IndAerodyn 2001;89:1553e64.

[25] Snæbjörnsson JT, Geurts CPW. Modelling surface pressure fluctuations onmedium-rise buildings. J Wind Eng Ind Aerodyn 2006;94:845e58.

[26] Caracoglia L, Jones NP. Analysis of full-scale wind and pressure measurementson a low-rise building. J Wind Eng Ind Aerodyn 2009;97:157e73.

[27] Castro IP, Robins AG. The flow around a surface-mounted cube in uniform andturbulent streams. J Fluid Mech 1977;79:307e35.

[28] Cochran LS, Cermak JE. Full- and model-scale cladding pressures on the TexasTech University experimental building. J Wind Eng Ind Aerodyn 1992;43:1589e600.

[29] Richards PJ, Hoxey RP, Connell BD, Lander DP. Wind-tunnel modelling of theSilsoe cube. J Wind Eng Ind Aerodyn 2007;95:1384e99.

[30] Surry D. Pressure measurements on the Texas tech building: wind-tunnelmeasurements and comparisons with full scale. J Wind Eng Ind Aerodyn1991;38:235e47.

[31] Richardson GM, Surry D. The Silsoe structures building: comparison betweenfull-scale and wind-tunnel data. J Wind Eng Ind Aerodyn 1994;51:157e76.

[32] Stathopoulos T, Saathoff P. Wind pressure on roofs of various geometries.J Wind Eng Ind Aerodyn 1991;38:273e84.

[33] Murakami S, Mochida A, Hibi K. Three-dimensional numerical simulation ofair flow around a cubic model by means of large eddy simulation. J Wind EngInd Aerodyn 1987;25:291e305.

[34] Murakami S, Mochida A. 3-D numerical simulation of airflow around a cubicmodel by means of the ke 3model. J Wind Eng Ind Aerodyn 1988;31:283e303.

[35] Baetke F, Werner H, Wengle H. Numerical simulation of turbulent flow oversurface-mounted obstacles with sharp edges and corners. J Wind Eng IndAerodyn 1990;35:129e47.

[36] Mochida A, Murakami S, Shoji M, Ishida Y. Numerical simulation of flowfieldaround Texas tech building by large Eddy simulation. J Wind Eng Ind Aerodyn1993;46e47:455e60.

[37] Nore K, Blocken B, Thue JV. On CFD simulation of wind-induced airflow innarrow ventilated facade cavities: coupled and decoupled simulations andmodelling limitations. Build Environ 2010;45:1834e46.

[38] Reinhold T. Wind tunnel modeling for civil engineering applications. Cam-bridge: Cambridge University Press; 1982.

H. Montazeri, B. Blocken / Building and Environment 60 (2013) 137e149 149

[39] Blocken B, Gualtieri C. Ten iterative steps for model development and eval-uation applied to computational fluid dynamics for environmental fluidmechanics. Environ Modell Softw 2012;33:1e22.

[40] Blocken B, Janssen WD, van Hooff T. CFD simulation for pedestrian windcomfort and wind safety in urban areas: general decision framework and casestudy for the Eindhoven University campus. Environ Modell Softw 2012;30:15e34.

[41] van Hooff T, Blocken B. Coupled urban wind flow and indoor natural venti-lation modelling on a high-resolution grid: a case study for the AmsterdamArenA stadium. Environ Modell Softw 2010;25:51e65.

[42] Blocken B, Persoon J. Pedestrian wind comfort around a large football stadiumin an urban environment: CFD simulation, validation and application of thenew Dutch wind nuisance standard. J Wind Eng Ind Aerodyn 2009;97:255e70.

[43] He J, Song CCS. Evaluation of pedestrian winds in urban area by numericalapproach. J Wind Eng Ind Aerodyn 1999;81:295e309.

[44] Richards PJ, Mallinson GD, McMillan D, Li YF. Pedestrian level wind speeds indowntown Auckland. Wind Struct 2002;5:151e64.

[45] Hirsch C, Bouffioux V, Wilquem F. CFD simulation of the impact of newbuildings on wind comfort in an urban area. In: Workshop proceedings, costaction C14, impact of wind and storm on city life and built environment.Nantes, France: 2002.

[46] Yoshie R, Mochida A, Tominaga Y, Kataoka H, Harimoto K, Nozu T, et al.Cooperative project for CFD prediction of pedestrian wind environment in theArchitectural Institute of Japan. J Wind Eng Ind Aerodyn 2007;95:1551e78.

[47] Evola G, Popov V. Computational analysis of wind driven natural ventilation inbuildings. Energy Build 2006;38:491e501.

[48] Jiang Y, Chen Q. Effect of fluctuating wind direction on cross natural ventilationin buildings from large eddy simulation. Build Environ 2002;37:379e86.

[49] van Hooff T, Blocken B, Aanen L, Bronsema B. A venturi-shaped roof for wind-induced natural ventilation of buildings: wind tunnel and CFD evaluation ofdifferent design configurations. Build Environ 2011;46:1797e807.

[50] Norton T, Grant J, Fallon R, Sun D-W. Optimising the ventilation configurationof naturally ventilated livestock buildings for improved indoor environmentalhomogeneity. Build Environ 2010;45:983e95.

[51] Ramponi R, Blocken B. CFD simulation of cross-ventilation for a generic iso-lated building: impact of computational parameters. Build Environ 2012;53:34e48.

[52] Kato S, Murakami S, Mochida A, Akabayashi S-i, Tominaga Y. Velocityepres-sure field of cross ventilation with open windows analyzed by wind tunneland numerical simulation. J Wind Eng Ind Aerodyn 1992;44:2575e86.

[53] Heiselberg P, Li Y, Andersen A, Bjerre M, Chen Z. Experimental and CFDevidence of multiple solutions in a naturally ventilated building. Indoor Air2004;14:43e54.

[54] Blocken B, Stathopoulos T, Saathoff P, Wang X. Numerical evaluation ofpollutant dispersion in the built environment: comparisons between modelsand experiments. J Wind Eng Ind Aerodyn 2008;96:1817e31.

[55] Gousseau P, Blocken B, Stathopoulos T, van Heijst GJF. CFD simulation of near-field pollutant dispersion on a high-resolution grid: a case study by LES andRANS for a building group in downtown Montreal. Atmos Environ 2011;45:428e38.

[56] Hanna SR, Brown MJ, Camelli FE, Chan ST, Coirier WJ, Hansen OR, et al.Detailed simulations of atmospheric flow and dispersion in downtownManhattan: an applications of five computational fluid dynamics models. BullAm Meteorol Soc 2006;87:1713e26.

[57] Tominaga Y, Stathopoulos T. CFD modeling of pollution dispersion in a streetcanyon: comparison between LES and RANS. J Wind Eng Ind Aerodyn 2011;99:340e8.

[58] Tominaga Y, Stathopoulos T. Numerical simulation of dispersion around anisolated cubic building: model evaluation of RANS and LES. Build Environ2010;45:2231e9.

[59] Blocken B, Defraeye T, Derome D, Carmeliet J. High-resolution CFD simula-tions for forced convective heat transfer coefficients at the facade of a low-risebuilding. Build Environ 2009;44:2396e412.

[60] Karava P, Jubayer CM, Savory E. Numerical modelling of forced convectiveheat transfer from the inclined windward roof of an isolated low-rise buildingwith application to photovoltaic/thermal systems. Appl Therm Eng 2011;31:1950e63.

[61] Saneinejad S, Moonen P, Defraeye T, Carmeliet J. Analysis of convective heatand mass transfer at the vertical walls of a street canyon. J Wind Eng IndAerodyn 2011;99:424e33.

[62] Delaunay D, Lakehal D, Pierrat D. Numerical approach for wind loads predic-tion on buildings and structures. J Wind Eng Ind Aerodyn 1995;57:307e21.

[63] Panneer Selvam R. Computation of pressures on Texas Tech Universitybuilding using large eddy simulation. J Wind Eng Ind Aerodyn 1997;67e68:647e57.

[64] Blocken B, Stathopoulos T, Carmeliet J. Wind environmental conditions inpassages between two long narrow perpendicular buildings. J Aerospace Enge ASCE 2008;21:280e7.

[65] Snyder WH. Guideline for fluid modeling of atmospheric diffusion. U.S.environmental protection agency report no. EPA-600/8-81-0091981.

[66] Franke J, Hellsten A, Schlünzen H, Carissimo B. Best practice guideline for theCFD simulation of flows in the urban environment. Brussels: COST Office;2007.

[67] Tominaga Y, Mochida A, Yoshie R, Kataoka H, Nozu T, Yoshikawa M, et al. AIJguidelines for practical applications of CFD to pedestrian wind environmentaround buildings. J Wind Eng Ind Aerodyn 2008;96:1749e61.

[68] Launder BE, Spalding DB. The numerical computation of turbulent flows.Comput Methods Appl Mech Eng 1974;3:269e89.

[69] Cebeci T, Bradshaw P. Momentum transfer in boundary layers. New York:Hemisphere Publishing Corp; 1977.

[70] Blocken B, Stathopoulos T, Carmeliet J. CFD simulation of the atmosphericboundary layer: wall function problems. Atmos Environ 2007;41:238e52.

[71] Blocken B, Moonen P, Stathopoulos T, Carmeliet J. Numerical study on theexistence of the venturi effect in passages between perpendicular buildings.J Eng Mech e ASCE 2008;134:1021e8.

[72] Blocken B, Carmeliet J, Stathopoulos T. CFD evaluation of wind speed condi-tions in passages between parallel buildingsdeffect of wall-function rough-ness modifications for the atmospheric boundary layer flow. J Wind Eng IndAerodyn 2007;95:941e62.

[73] Shih T-H, Liou WW, Shabbir A, Yang Z, Zhu J. A new ke 3eddy viscosity modelfor high reynolds number turbulent flows. Comput Fluids 1995;24:227e38.

[74] Murakami S. Comparison of various turbulence models applied to a bluffbody. J Wind Eng Ind Aerodyn 1993;46e47:21e36.

[75] Jones WP, Launder BE. The prediction of laminarization with a two-equationmodel of turbulence. Int J Heat Mass Transfer 1972;15:301e14.

[76] Yakhot V, Orszag SA, Thangam S, Gatski TB, Speziale CG. Development ofturbulence models for shear flows by a double expansion technique. PhysFluids;4:1510e20.

[77] Choudhury D. Introduction to the renormalization group method and turbu-lence modeling: technical memorandum TM-107; 1993.

[78] Wilcox DC. Turbulence modeling for CFD: La Canada. California: DCWIndustries, Inc; 1998.

[79] Launder BE, Reece GJ, Rodi W. Progress in the development of a Reynolds-stress turbulence closure. J Fluid Mech 1975;68:537e66.