Wind pressures in canopies with parapets

A. Poitevin1, B. Natalini2, L. A. Godoy3

1
Research Assistant, Department of Civil Engineering and Surveying, University of Puerto Rico, poite@msn.com
2
Prof. of Mechanical Engineering, CONICET-UNNE, Resistencia, Argentina, bnatalini_2000@yahoo.com.ar
3
Professor, Department of Civil Engineering and Surveying, University of Puerto Rico, luis.godoy@upr.edu

ABSTRACT
A literature review shows that there is a lack of information on the wind load pressures
acting on canopy structures with a parapet on the roof perimeter. On the other hand, a large
number of those structures suffered catastrophic damages due to hurricanes, for example during
Hurricanes Katrina and Rita in 2005. The recent version of the predominant building code for
wind design in the eastern part of the US, the ASCE 7-05, recently introduced the issue of open
structures. However, the code recommendations do not address open structures with parapets. In
this research, the authors address the problem using two different (but complementary)
approaches: a numerical simulation using Computational Fluid Dynamics (CFD) and wind
tunnel testing. Values of pressure coefficients Cp were obtained by both methods, showing good
agreement between them. Further studies are in progress in order to gather data on Cp values
for canopies with parapets.

INTRODUCTION
Open canopies are frequently used in the construction of civil engineering facilities,
either as components of larger structures or as a self supported structures. An example of the
second type may be found in most gas stations throughout the US, in which the roof covers gas
pumps and vehicles. In small gas stations, canopies are supported by a single row of columns,
and do not include perimeter walls, whereas larger canopies are supported by columns aligned in
two rows.

Figure 1: Schematic view of a canopy used in gas stations.

 The roof structure is formed by a system of beams in two directions to support the roof
panels. In some cases, the system is based on cantilever columns. In hurricane-prone areas, such
as the Caribbean islands and the coastal areas in the US, the most critical structural conditions
occur during high wind loads.
A large number of those structures suffered catastrophic damages due to hurricanes, for
example, during Hurricanes Katrina and Rita in 2005 [1]. Figure 2 illustrates the failure of one
canopy structure which had been recently constructed in Chalmette, close to the city of New
Orleans. The canopy rotated as a rigid body, with localized failure at the supports in the weak
direction of the columns.

Figure 2: Chalmette, New Orleans, Hurricane Katrina (photograph by L. A. Godoy).

The predominant specification for wind design in the eastern part of the US is ASCE 7.
The recent version of this code, the ASCE 7-05 [2] was the first one to address the issue of open
structures; however, the recommendations do not address the issue of open structures with
parapets. With such high wind velocities, it is surprising to find such lack of information on the
wind pressures acting on canopy structures with a parapet on the roof perimeter. Our conclusion
is that open canopy structures which include parapets have not been studied in detail in the
research literature or taken into account in the current design codes used in the US.
Canopy structures without parapets have been recently investigated in a small number of
previous investigations, leading to wind pressure distributions estimates and design coefficients
[3, 4, 5, 6, 7, 8], but, no testing has been reported regarding wind pressures in open canopies with
parapets. Because this may be a controversial topic (due to its engineering significance), it is
desirable to have some methodological redundancy to make sure that adequate pressure
coefficients are proposed. The situation calls for numerical and experimental investigations on
the same topic, in order to validate pressure coefficients using independent approaches. In this
research, the authors address the problem using two different (but complementary) approaches: a
numerical simulation using Computational Fluid Dynamics (CFD) and wind tunnel testing.
In this paper, mean pressure coefficients produced with a numerical model are presented
and compared with results of boundary-layer wind tunnel tests on a canopy roof with parapets,
being these values the first ones ever reported on this kind of structures.

EXPERIMENTAL ARRANGEMENT
 MODELS
Experiments were conducted on one 1:50 scale model of a 7.5 m × 7.5 m square roof
with parapets of 1.2 m high, having an eave height of 3.6 m. First, experiments were carried out
with no parapets in order to compare with other wind tunnel results available in open literature.
Wind load coefficients were measured under wind blowing at angles of 0º and 30º relative to one
of the symmetry axis, since, as demonstrated by Gumley [3, 4, 7] among others, these directions
produce the most severe loads on planar canopy roofs with no parapets.
The roof of the model and the parapets were made with a 2 mm thick aluminium plate,
and the columns with a 2.5 mm diameter steel rod. As the models have two axes of symmetry,
only a quarter of the models' roofs had pressure taps in place, thus reducing the number of tubes
needed. In addition, all the tubes were led towards the farthest corner, where they formed a
bundle that went into a horizontal pipe through which they went away from the model to finally
go under the floor. In this way, the scale distortion in both columns and roof thickness and the
possible interference of the tubes upon the measurements were minimized.
Sixteen pressure taps were spread on the roof and twenty on the parapets. Figure 3 shows
the position of the pressure taps on the model and Figure 4 shows the model placed in the
turntable.

z x

Figure 3: Distribution of pressure taps on the model.

 Figure 4: View the model with parapets on the turntable.

WIND SIMULATION
The wind tunnel testing was performed during 2008/2009 at Universidad Nacional del
Nordeste (UNNE) in Resistencia, Argentina. The facility is an open circuit tunnel with a length
of 22.8m (74.8ft), the testing chamber being a square section of 2.4m (7.87ft) width and 1.8m
(5.9ft) in height and uses a 2.25m (7.38ft) fan with a 92kw motor. The maximum wind velocity
is 25m/s (55.9 mph) when the testing section is empty. Further details regarding this wind
facility are given by Wittwer and Möller [9].
All the models were tested under a wind simulation corresponding to a suburban area.
The simulation hardware consisted of two modified Irwin’s spires [10] and 17.1 m of surface
roughness fetch downstream from the spires. In this way, a part-depth boundary layer simulation
of neutrally stable atmosphere was obtained. Mean velocities, when fitted to a potential law, give
an exponent of 0.24. The length scale factor determined according to Cook’s procedure [11] was
approximately 150. The local turbulence intensity at the level of the roof is 0.25. De Bortoli et al.
[12] provided further details of this turbulence simulation, including size, geometry, and
arrangement of the hardware and design criteria.

PRESSURE MEASUREMENT SYSTEM

Pressures were measured using a differential pressure electronic transducer Micro Switch
Honeywell 163 PC. A sequential switch Scanivale 48 D9-1/2, which was driven by means of a
CTLR2 / S2-S6 solenoid controller connected the pressure taps to the transducer through PVC
tubes of 1.5 mm in internal diameter and 650 mm in length. No resonance problems were
detected for tubes of that length (the gain factor being around one) therefore, restrictors of
section were not used for filtering. The DC transducer output was read with a Keithley 2000
digital multimeter. The integration time operation rate of the A/D converter was set to produce
mean values over 15 seconds of time integration.
 Simultaneously to the pressure measurements being taken on the roof, the reference
dynamic pressure, qref, was measured at the eaves height with a Pitot-Prandtl tube connected to a
Van Essen 2500 Betz differential micromanometer of 1 Pa resolution. The probe stayed beside
the model at a distance of about 0.70 m to avoid mutual interference. The reference static
pressure was obtained from the static pressure tap of the same Pitot-Prandlt tube.

COMPUTIONAL FLUID DYNAMICS (CFD) SIMULATION

The wind flow on an open canopy structure with parapets can be represented and
visualized with the use of CFD. The specific CFD software package used in this research is
named EFD.Lab (Engineering Fluid Dynamics), developed by Flomerics Inc. [13]. EFD.Lab
solves the Navier-Stokes equations, which include the formulations of mass, momentum and
energy conservation laws for fluid flows. The equations are supplemented by fluid state
equations defining the nature of the fluid and by empirical dependencies of fluid density,
viscosity and thermal conductivity on temperature [13].
EFD.Lab solves the governing equations using the finite volume method (FVM) on a
prismatic rectangular computational mesh drawn in a 3D Cartesian coordinate system with the
planes orthogonal to its axes. To improve the results, the mesh is refined locally at the solid/fluid
interface. Turbulence is incorporated by means of turbulent intensity and turbulent length
parameters, they were obtained from the wind tunnel and used in the CFD computations order to
emulate the wind tunnel conditions.
The computational mesh used for the simulation is created in a prismatic domain, as
shown in Figure 5. The surface area on the left is defined as the inlet boundary, in which it is
possible to introduce the values for the wind profile. The specific profile adopted in the first
model investigated in this research is a power law (as defined by ASCE 7) for which an exponent
of 0.24 was used, as obtained from the wind tunnel tests. The transverse section of the flow in
Figure 5, showing different velocities on the left indicate that a wind profile was used as an
input. The canopy induces significant changes to the flow, and the program calculates pressures
and Cp values on its contours.
A second computational model was also built, in which laminar flow was assumed with a
constant velocity profile in height. Although the inlet conditions between a power law and a
uniform velocity profile are very different, in the present case in which the canopy is a thin body
at a given distance from the terrain, the results in terms of pressures acting on the canopy are
remarkably similar. The reasons for this similarity are that the canopy is located away from the
turbulence generated at the ground level, and the pressures are dominated by the pressures and
velocities that occur at the canopy height. Such similarity is only specific of this case and would
not occur in enclosed structures.
The results from the initial computational domain were compared to a smaller domain
without a boundary condition. The number of cells for the initial domain was 303,123. The
smaller domain used was 54,954 cells. The results from the comparison produced similar results
between both simulations, whereas the computational time was reduced dramatically. For this
reason, further analysis was performed with the smaller domain using a uniform velocity profile
at the inlet surface as shown in Figure 6.
The CFD software yields computations of wind pressures on the external (top and
bottom) surfaces of the structure, with a wind velocity similar to those specified by design codes.
CFD simulations are extremely useful to identify and define the geometric characteristics of
canopies that will be investigated prior to the construction of physical models to be tested in a
wind tunnel facility. Preliminary investigations allow determining the high pressure areas for
determination of the wind pressure taps to be used on the scale model.

Figure 5: Computational domain with boundary element and wind profile. The inlet condition is defined in
the plane on the left.

The initial model geometry analyzed consists of a 7.6 m (25 ft) x 7.6m (25ft) with a 1.2m
(4ft) parapet. The roof height is located at 3.6m (12ft). The mesh cells sides are orthogonal to the
specified axes on the coordinate system. The mesh can be automatically created using a mesh
generator, but the mesh in this simulation was specified to be refined on the exposed surfaces.
Further sub-meshing was assigned for refinement on all exposed surfaces to the turbulent flow.

Figure 6: Computational domain and meshing used in the present CFD simulations.
 Wind pressures were obtained from the CFD simulations. To better represent the results
obtained from the wind tunnel, data were obtained at the same places where the pressure taps
were located in the physical model. Specifically, pressure taps were located on the top surface,
bottom surface and on all parapets, inside and outside surfaces. Results of the Cp (pressure
coefficient factor) were obtained from the simulation, and have been represented in Figure 7.

Figure 7: Location of pressure taps and measured wind pressures on the top surface. Wind direction acting
from the right (0 degrees).

Two wind directions were used for the computational model, which were the same wind
directions used in the wind tunnel [3,4,7]: One direction at 0 degrees and the second one at 30
degrees of the long axis of the canopy (see Figure 8).
 Figure 8: Wind direction in CFD model at 0 degrees and 30 degrees.

RESULTS
Data from the wind tunnel and computational models were obtained on the top and
bottom surfaces at 64 pressure points leading to Cp values for each surface. Two wind conditions
were considered: one with incident wind oriented in the long axis direction (0°), whereas a
second case was considered by rotating the wind to 30° with respect to the structural axis.

Figure 9: Cp results from top surface (a) wind tunnel, and (b) CFD model for incident wind at 0 degrees from
the structural axis.

Figure 9 shows the Cp values for the top surface of the canopy. Figure 9a shows the wind
tunnel Cp results and Figure 9b shows the CFD Cp results for the wind at 0°.On both, the wind
tunnel and the CFD simulation the top surface shows an uplift almost until the end of the surface
where a downward pressure is clearly shown on both figures.

Figure 10: Cp results from bottom surface (a) wind tunnel, and (b) CFD model for incident wind at 0 degrees
from the structural axis.

Figure 10 shows the Cp values for the bottom surface of the canopy. Figure 10a shows
the wind tunnel Cp results and Figure 10b shows the CFD Cp results for the wind at 0°. The
bottom surface shows on both, the wind tunnel and the CFD simulation an uniform downward
pressure through all the bottom surface.

Figure 11: Cp results from top surface (a) wind tunnel, and (b) CFD model for incident wind at 30 degrees
from the structural axis.

Figure 11 shows the Cp values for the top surface of the canopy. Figure 11a shows the
wind tunnel Cp results and Figure 11b shows the CFD Cp results for the wind at 30°.On both,
the wind tunnel and the CFD simulation the top surface shows an uplift for the first half of the
top surface and a downward pressure on the other half. This effect is clearly shown on both
figures.
Figure 12 shows the Cp values for the bottom surface of the canopy. Figure 12a shows
the wind tunnel Cp results and Figure 12b shows the CFD Cp results for the wind at 30°. The
bottom surface shows on both, the wind tunnel and the CFD simulation a uniform downward
pressure through all the bottom surface. The Cp values on both figures are very similar.
Net values, denoted by Cn, were calculated as the difference between the Cp values
obtained at the same location and on top and bottom surfaces. Figure 13a and Figure 13b are the
Cn results for the wind at 0° for the wind tunnel and the CFD simulation. Both figures showed an
initial downward pressure through the first 1/3 of the length, and uplift for the next 1/3 of the
length and a downward pressure for the last 1/3.
Cn results for the wind at 30° are shown on Figure 14. Figure 14a and Figure 14b shows
as before the wind tunnel test and the CFD simulation. Both figures showed an uplift area
located in the center of the surface, with a downward pressure around that center area.

Figure 12: Cp results from bottom surface (a) wind tunnel, and (b) CFD model for incident wind at 30
degrees from the structural axis.
 Figure 13: Cn results from (a) wind tunnel, and (b) CFD model for incident wind at 0 degrees from the
structural axis.

Figure 14: Cn results from (a) wind tunnel, and (b) CFD model for 30 degrees from the structural axis.

The extreme values for Cp and Cn along the wind axis are shown on Figure 15. Values of
Cp TP (Cp at the top surface), Cp BP (Cp at the bottom surface), and Cn for the canopy with
wind at 0° are plotted. The Cn values show a Cp of +1.3 maximum behind the parapet. The Cn
values changes along the center of the roof to a Cn of -0.5, through the end of the surface it
changes to a Cn of +0.5. The Cn values changes from + (positive) to – (negative) and to +
(positive) Cn values through the roof geometry.
 Figure 15: Cn results from different parapets heights on CFD simulation.

CONCLUSIONS
There is conclusive evidence that a large number of open canopy structures suffered
catastrophic damages due to hurricanes in the US during the last decade. The most recent version
of the ASCE [2] includes considerations for open structures; however, the recommendations do
not account for cases of open structures with parapets. For the specific geometry used in this
investigation, the calculated Cn values for an open structure without parapets using ASCE 7 are
negative on the complete surface and range from -0.8 to -0.3.
But, based on the evidence presented in this paper, those values cannot be trusted to
perform the design or verification of actual canopy structures used in gas stations. Our Cn results
from physical and computational simulations show that there is a strong influence of the parapets
on the pressure coefficients of a canopy structure; this influence was observed in both, wind
tunnel testing and CFD simulations, and good agreement has been obtained between the two
independent methodologies.
For the case studied here, the Cn values obtained in this investigation vary from +1.3, -
0.5 and +0.5. The mean roof pressures vary from pressure downward to upward pressure
depending on the specific locations considered in the roof along the long axis. This significantly
differs from the recommended code values, which predict that only suction, will occur in the
roof.
We conclude that ASCE recommendations for this type of structures are not sufficient to carry
out safe designs of canopy structures with parapets. It is suggested that current code
recommendations should be modified for open structures and should include new coefficients for
open structures with parapets, such as those presented in this paper. Further investigation [14] is
in progress to include the influence of different parapet heights and different roof geometries on
Cp values.
ACKNOWLEDGEMENTS
The authors wish to acknowledge the help of José A. Iturri in making the models and
preparing the experimental setup. Research in Argentina is part of an area supported by the
Facultad de Ingeniería of the Universidad Nacional del Nordeste . Research in Puerto Rico was
supported in part by NSF-CCLI grant DUE-0736828: “A computer-based simulated environment
to learn on structural failures in engineering” (Program Director: Sheryl Sorby).

REFERENCES
[1] NIST, 2006, Performance of Physical Structures in Hurricane Katrina and Hurricane Rita: A
Reconnaissance Report, NIST Technical Note 1476, National Institute of Standards and Technologies,
Gaithersburg, Maryland.

[2] ASCE 7 Standard, 2005, Minimum Design Loads for Buildings and Other Structures, American
Society of Civil Engineers, Reston, VA.

[3]S. J. Gumley, Panel Loading Mean Pressure Study for Canopy Roofs, University of Oxford Dept. of
Eng. Sc., OUEL report No.1380/81, 1981.

[4]S. J. Gumley, Design Extreme Pressures-A Parametric Study for Canopy Roofs, University of Oxford
Dept. of Eng. Sc., OUEL report No.1394/82, 1982.

[5]R. E. Belcher, C. J. Wood, Further Design Extreme Pressures on Canopy Roofs, University of Oxford
Dept. of Eng. Sc., OUEL report No.1481/83, 1983.

[6]J. D. Ginger, C. W. Letchford, 1994. Wind loads on planar canopy roofs, Part 2: Fluctuating pressure
distribution and correlations, Journal of Wind Engineering and Industrial Aerodynamics. 51, 353–370.

[7] C. W. Letchford, J. D. Ginger, 1992. Wind loads on planar canopy roofs, Part 1: Mean pressure
distributions, Journal of Wind Engineering and Industrial Aerodynamics. 45, 25–45.

[8] D. R. Altman, 2001. Wind uplift forces on roof canopies, M.Sc. Thesis, Clemson University, Clemson,
SC.

[9]A. R. Wittwer, S. V. Möller, 2000. Characteristics of the low-speed wind tunnel of the UNNE. Journal
of Wind Engineering and Industrial Aerodynamics 84, 307-320.

[10]Irwin, H.P.A.H., 1981,The design of spires for wind simulation, Journal of Wind Engineering and
Industrial Aerodynamics 7, 361-366.

[11]De Bortoli, M.E., Natalini, B., Paluch, M.J., Natalini, M.B., 2000, Part-depth wind tunnel
simulations of the atmospheric boundary layer. Journal of Wind Engineering and Industrial
Aerodynamics 90, 281-291.

[12]N. J. Cook, 1977/1978, Determination of the model scale factor in wind-tunnel simulations of the
adiabatic atmospheric boundary layer, Journal of Wind Engineering and Industrial Aerodynamics 2,
311-321.

[13] Flomerics, 2009, EFD.Lab v.9. Marlsborough, MA.

[14] A. Poitevin, 2009, Analysis of canopy structures with parapets under wind, Ph. D. Thesis, University
of Puerto Rico, Mayagüez, PR.