You are on page 1of 8

The Fifth International Symposium on Computational Wind Engineering (CWE2010)

Chapel Hill, North Carolina, USA May 23-27, 2010

Wind tunnel and CFD modelling of wind pressures on solar


energy systems on flat roofs
Alexander Bronkhorst 1, 2, Jörg Franke 3, Chris Geurts 1, 2, Carine van Bentum 1,
François Grepinet 4
1
TNO, Delft, the Netherlands, chris.geurts@tno.nl
2
Technical University of Eindhoven, Eindhoven, the Netherlands, a.j.bronkhorst@tue.nl
3
University of Siegen, Siegen, Germany, joerg.franke@uni-siegen.de
4
Wagner & Co Solartechnik GmbH, Cölbe, Germany, francois.grepinet@wagner-solar.com

ABSTRACT: Design of solar energy mounting systems requires more knowledge on the wind
patterns around these systems. To obtain more insight in the flow patterns, which cause the pres-
sure distributions on the solar energy systems, a wind tunnel test and Computational Fluid Dy-
namics analysis have been performed. In this study the average pressure coefficients, determined
in the wind tunnel measurements, are compared with Reynolds Averaged Navier-Stokes calcula-
tions. The comparison, based on the median of all observations over 6 wind directions and all
pressure points, showed that the general pressure distribution is well predicted. Overall differ-
ences were found of 39% for the Renormalization Group k-ε turbulence model, 35% for a differ-
ential Reynolds Stress turbulence model with wall reflection term and 35% for a differential Rey-
nolds Stress turbulence model without wall reflection term. The largest differences are observed
in the wake of systems that have a large spacing, which is due to an incorrect prediction of the
separation zones and therefore the shielding effect of the solar energy systems.

1 INTRODUCTION

The increasing growth of solar energy systems in Europe asks for appropriate design data. Flat
roofs of large buildings are often used for placement of these systems. In this case wind loads are
known to be the main cause for damages. Guidelines for these loads have been derived from a
well-defined wind tunnel research performed at TNO and have been implemented in the NVN
7250 (2007) in the Netherlands, a first in its kind standard for solar energy systems integrated in
roofs and facades. Although these guidelines provide a useful tool for designers, they also limit
the design freedom. Obtaining more insight in the flow patterns which cause the loads on the so-
lar energy systems could allow for better solar energy systems designs. To utilize CFD for this
application, validation is needed.
This study investigates the average wind pressures on solar energy systems on a building with
a flat roof using wind tunnel measurements and CFD calculations. A comparison is made be-
tween the Renormalization Group (RNG) k-ε turbulence model and two versions of a differential
Reynolds Stress turbulence model based on the proposal of Launder et al. (1975), both as imple-
mented in the commercial CFD solver Fluent V6.3 (Fluent, 2006). The comparison provides in-
formation on the influence of these turbulence models on the accuracy of CFD for the calculation
of average wind loads on solar energy systems on top of flat roofs. Furthermore, the application
of CFD provides information on the flow patterns around the solar energy systems, which helps
to understand the origin of the pressures observed in the wind tunnel measurements.
The Fifth International Symposium on Computational Wind Engineering (CWE2010)
Chapel Hill, North Carolina, USA May 23-27, 2010

Figure 1. Illustration of the wind tunnel model and a detail of the closed solar energy systems near the roof corner.

2 WIND TUNNEL EXPERIMENT

Wind tunnel experiments have been carried out in the open circuit atmospheric boundary layer
wind tunnel of TNO Built Environment and Geosciences in the Netherlands. A boundary layer
with a roughness length of 2.4 mm has been applied, giving a full scale value of z0 = 0.12 m. The
roughness elements of the upstream fetch extended onto the turn table, as illustrated in figure 1,
to prevent the existence of an internal boundary layer at the location of the model. The wind
speed in the wind tunnel at model roof height was approximately Uref = 10.7 m/s, varying slightly
for the different approach flow directions. At this location the turbulence intensity was about 15%
in longitudinal direction, and 12% in vertical direction. The wind tunnel profiles of the mean
wind speed and turbulence intensity are provided in figure 2. The air density in the wind tunnel at
the time of the measurements was 1.354 kg/m3.
Measurements in this study were carried out on four different configurations of the building and
solar energy systems. A more detailed description of these experiments is given by Geurts et al.
(2005), including values obtained by analysis of the extreme values of the pressure coefficients.
These values have been applied in NVN 7250 (2007). This paper describes the analysis of one
configuration, illustrated in figure 1, with CFD and the comparison with the wind tunnel meas-
urement for time-averaged pressure coefficients.
The wind pressures on solar energy systems on flat roofs have been studied on a model scaled
1:50 of a building with rectangular plan (see figure 3), with full scale height of H = 10 m, width
of B = 30 m and depth of D = 40 m. To make optimal use of symmetry, the roof of this building
was divided into four quarters. This set up allowed researching various effects, such as the loca-
tion of the systems on the roof, the effects of shielding and the effect of orientation relative to the
prevailing winds.
The solar energy systems have a depth of 1.2 metres full-scale and an inclination angle of 35
degrees. The systems have been equipped with 120 pressure taps with an internal diameter of 1
mm, distributed as illustrated in figure 3. Pressure measurements were carried out at 400 Hz, with
a period of 20.4 seconds for each configuration. Average pressure coefficients were determined
from the measured time series using

p
cp = ,
1
2
ρU ref
2

where p is the time averaged pressure.


The Fifth International Symposium on Computational Wind Engineering (CWE2010)
Chapel Hill, North Carolina, USA May 23-27, 2010

4 4 4
Exp. 0° Exp. 0°, I u
Exp. 45° Exp. 0°, I w 3.5
3.5 3.5
Sim. Exp. 45°, I u
Exp. 45°, I w
3 3 3
Sim., Iu=Iw

2.5 2.5 2.5

z/H
z/H

z/H
2 2 2

1.5 1.5 1.5

1 1 1

0.5 0.5 0.5

0 0 0
0 0.5 1 1.5 10 20 30 0 0.025 0.05
2
U/U ref Iu [%], Iw [%] k/U ref
Figure 2. Comparison of experimental and equilibrium inflow profiles for mean velocity (left), turbulence intensities
(middle) and turbulent kinetic energy (right). Note that the experimental turbulent kinetic energy (k) was computed
from the approximation vrms2 = 0.5(urms2 + wrms2).

61 69-72
17-20 41-44 59 60 105-108 77-80
13-16
58 101-104
85-88 37-40
25-28 57 113-116 93-96 53-56 5-8
29-32
49-52 89-92 45-48
1-4 33-36 97-100 117-120
64 21-24
9-12 81-84 109-112 6263
65-68
73-76

Figure 3. Building model with solar panels and numbering of pressure measurement positions.

3 CFD SIMULATION

The numerical simulations were performed with the double precision version of the commercial
flow solver FLUENT V6.3 (Fluent, 2006). The computational domain used for the simulation of
six wind directions is shown in figure 4. It represents a part of the wind tunnel test section, scaled
to full scale for the simulations. Like in the experiment the different approach flow directions
were obtained by rotating the building around its centre. The position of the building for the 0°
and 45° approach flow case is also shown in figure 4 together with the distances of the domain
boundaries from the building. The height of the domain is 7H = 70 m for all wind directions.
At the inflow boundary equilibrium profiles were prescribed for the velocity and the turbu-
lence (Richards and Hoxey, 1993), see figure 2. They use the experimental z0 and Uref together
with a value of κ = 0.4187 for the von Kármán constant. From these profiles also the constant
values for the velocity and turbulence quantities were derived, which were prescribed at the top
boundary. The lateral boundaries of the computational domain coincide with the wind tunnel
walls and were treated as solid smooth walls. Smooth wall boundary conditions were also applied
at the building and the solar energy systems. At the outlet a constant pressure was prescribed. The
ground floor was treated as a rough wall with equivalent sand grain roughness ks = 0.38 m, de-
duced from the hydrodynamic roughness length z0 with a constant Cs = 3.09 used in Fluent’s im-
plementation for rough wall functions (Franke, 2007).
The Fifth International Symposium on Computational Wind Engineering (CWE2010)
Chapel Hill, North Carolina, USA May 23-27, 2010

Figure 4. Size of computational domains for 0° (left) and 45° (right) approach flow case.

The computational domain was meshed with a block structured grid made entirely of hexa-
hedra. For each wind direction a slightly different mesh was used with identical vertical grid point
distribution and identical grid around the building and solar systems up to a distance of 5 m from
the vertical building walls. To keep the grid expansion ratio for all wind directions below 1.3 the
total number of cells varied between 2.0·106 and 2.2·106. The influence of the grid resolution on
the results was not analysed. In figure 5 the surface mesh on the building and solar systems is
shown, as well as a detail of the surface mesh on the lateral faces of the solar systems. This detail
also contains geometry information and the position of the pressure measurement positions on the
inclined and vertical sides.

Figure 5. Surface mesh of building and solar panels (left) and of cross section of solar panel (right).

The statistically steady flow field was computed by solving the Reynolds Averaged Navier-
Stokes (RANS) equations together with the Renormalization Group (RNG) k-ε turbulence model
and a differential Reynolds Stress turbulence Model (DSM), both using standard wall functions.
With the DSM the Reynolds stresses in the wall adjacent cells are derived from the turbulent ki-
netic energy calculated from an additional transport equation (Fluent, 2006). The DSM is based
on the model of Launder et al. (1975) with linear pressure strain modeling (Rotta, 1951; Fu et al.,
1987). The model was used without (DSM) and with (DSM-WR) wall reflection term (Gibson
and Launder, 1978). Both models were chosen because of their relatively accurate pressure pre-
dictions on cubes (e.g. Wright and Easom, 1999).
The convective terms in all transport equations except the ones for the individual Reynolds
stresses were approximated with the 2nd order upwind scheme and the gradients in the diffusive
terms with central differences. For the Reynolds stresses the 1st order upwind scheme was used.
The influence of the 1st order scheme was tested for the 0° case. No noticeable difference in pres-
sures was found, when comparing with results obtained with the 2nd order upwind method for the
Reynolds stress equations. Simulations were run until a drop of the scaled residuals for all equa-
tions below 10-5 was reached. This could not be achieved for the continuity equation and the ε
The Fifth International Symposium on Computational Wind Engineering (CWE2010)
Chapel Hill, North Carolina, USA May 23-27, 2010

equation for all wind directions. Monitoring of pressures at selected positions revealed however
that the flow and surface pressure field did no longer change.
For the comparison with the experiments the computed pressures were extracted at the center
points of the circular pressure taps. As the diameter of the pressure taps would be 50 mm in full
scale, this procedure causes an error when there is a large variation in pressure over the pressure
tap opening (Otto et al., 2008). The significance of this effect is not considered in this paper.

4 RESULTS AND DISCUSSION

Figure 6 illustrates the wind directions that have been analysed with CFD. The roof has been di-
vided in four zones (see figure 6). The zone division is used in the presentation of the results. The
first case investigates an approach flow of 0° for zone 1 and 3 and 180° for zone 2 and 4, the
second case considers the diagonal approach flow per zone (45° for zone 1, 135° for zone 2, 315°
for zone 3 and 225° for zone 4). The results obtained from the pressure points are discussed in
sets of four (as presented in figure 7 and 8).

Figure 6. Division of the building roof in 4 zones. Using the symmetry of the building and applying different system
configurations in each zone various effects can be studied with a minimum of measurements. The illustration at the
right gives the six wind directions that are used for comparison between experiment and CFD.

Figure 7 gives the determined pressure coefficients per zone for a wind approach angle of 0°
(zone 1 and 3) and 180° (zone 2 and 4). The overall pressure distribution is mostly well repro-
duced by both turbulence models for the two wind directions. Larger differences are observed at
several sets of measurement positions. In zone 1 the solution with the RNG model predicts higher
pressures than the experiment for points 1-4 and 9-12, located at the windward side of the first
system. This cannot be attributed to larger turbulent kinetic energy of the RNG model in the stag-
nation region, as the DSM-WR solution has even a 50% higher turbulent kinetic energy but lower
pressure. The high turbulent kinetic energy results from the inhibited transfer from the Reynolds
stress component normal to the stagnation region to the other Reynolds stress components (Craft
et al., 1993). When turning off the wall reflection model the pressure is reduced for the DSM,
which has about the same levels of turbulent kinetic energy as the RNG model.
Due to the higher turbulent kinetic energy the separating streamline close to points 1-4 in the
DSM-WR solution is stronger deflected around the tip of the first module than the streamline of
the DSM solution. This is responsible for the higher negative pressure coefficients of DSM-WR
at points 5-8, located on the leeward surface of the first module. The experimental suction is best
predicted by the RNG model with the least separation streamline curvature. These differences re-
sult from the differences in the prediction of the vertical Reynolds shear stress component 〈u’w’〉,
The Fifth International Symposium on Computational Wind Engineering (CWE2010)
Chapel Hill, North Carolina, USA May 23-27, 2010

where u’ is the fluctuation in x-direction and w’ the one in z-direction, as also discussed by
Wright and Easom (1999, 2003) for the flow around a cube. The higher levels of 〈u’w’〉 with
DSM and DSM-WR cause the aforementioned stronger deflection of the flow and thus a shorter
recirculation region behind the solar systems with corresponding stronger vortices between sys-
tems one and two and systems two and three. The same reasoning applies to positions 9-12 and
13-16.

Zone 1 (0 degrees)
1

0.5

0
Cp [-]

-0.5

-1

-1.5
1-4 5-8 9-12 13-16 17-20 21-24 25-28 29-32 33-36 37-40 41-44 45-48 49-52 53-56

Zone 2 (180 degrees) Zone 3 (0 degrees)


0.5 1

0 0.5

-0.5 0
Cp [-]

Cp [-]

-1 -0.5

-1.5 -1

-2 -1.5
57-60 61-64 65-68 69-72 73-76 77-80 81-84 85-88

Zone 4 (180 degrees)


0.5
Experimental
0 DSM (with WR model)
RNG
-0.5
DSM (no WR model)
Cp [-]

-1

-1.5

-2
89-92 93-96 97-100 101-104 105-108 109-112 113-116 117-120

Figure 7. Pressure coefficients per zone for an approach angle of 0° (zone 1 and 3) and 180° (zone 2 and 4), deter-
mined in the wind tunnel and with CFD (RNG model, DSM and DSM-WR). Sets of 4 closely grouped pressure
points are plotted on the x axis.

The difference in wake regions behind the systems is also the cause of the large differences at
positions 41-44 and 49-52. In the DSM and DSM-WR solution the separation zone behind the
third system is reduced because of higher turbulent mixing in vertical direction. As the experi-
mental results are closer to the two DSM solutions it can be inferred that the experimental recir-
culation region is even shorter. Therefore the shielding effect provided by the first three systems
is not well captured by the CFD solutions.
From the results in zone 3 it can be seen that the effect of the offset of the systems from the
roof’s edges is well reproduced with nearly all positions predicting suction. The largest differ-
ences between experiment and numerical simulations at points 81-84 are again due to the incor-
rect prediction of the separation zone behind the third system.
For approach flow direction 180° the pressure points 57-61 in zone 2 provide information on
the building separation region without the influence of the solar energy systems. Figure 7 shows
that both DSM versions and the RNG model predict lower under pressures than the experiments,
except for point 61 where RNG underestimates the pressure coefficient. This can again be attrib-
uted to the different streamline curvature at separation.
The Fifth International Symposium on Computational Wind Engineering (CWE2010)
Chapel Hill, North Carolina, USA May 23-27, 2010

Finally, the results from zone 4 show that the influence of the different orientation of the in-
clined surface of the systems to the approach flow is well predicted except at positions 93-96,
where again the shielding effect is predicted poorly.

Zone 1 (45 degrees)

0
Cp [-]

-1

-2

-3
1-4 5-8 9-12 13-16 17-20 21-24 25-28 29-32 33-36 37-40 41-44 45-48 49-52 53-56

Zone 2 (135 degrees) Zone 3 (315 degrees)


0.5 1
0
0
-0.5
-1
Cp [-]

Cp [-]

-1
-1.5 -2
-2
-3
-2.5
57-60 61-64 65-68 69-72 73-76 77-80 81-84 85-88

Zone 4 (225 degrees)

Experimental
0
DSM (with WR model)
RNG
-1 DSM (no WR model)
Cp [-]

-2

-3
89-92 93-96 97-100 101-104 105-108 109-112 113-116 117-120

Figure 8. Pressure coefficients per zone for different wind approach angles (zone 1 = 45°; zone 2 = 135°; zone 3 =
315°; zone 4 = 225°), determined in the wind tunnel and with CFD (RNG, DSM and DSM-WR turbulence model).

Figure 8 provides the pressure coefficients per zone for a wind approach angle of 45° (zone 1),
135° (zone 2), 225° (zone 4) and 315° (zone 3). The qualitative agreement is again good for all
turbulence models with similar differences as for the 0° and 180° approach flow. At the pressure
points which are closest to the corner directed to the approach flow, the influence of the stronger
corner vortices predicted with both DSM versions is visible as larger suction (Wright and Easom,
1999).

Table 1. Absolute relative difference between experimental pressure coefficients and coefficients determined with the
RNG and DSM turbulence models. The median and mean differences are determined over all pressure points.
Turbulence model Wind direction [°]
0 45 135 180 225 315 overall
Wind tunnel 0 0 0 0 0 0 0
RNG median [%] 26 49 26 42 45 44 39
mean [%] 112 74 44 284 73 291 146
DSM-WR median [%] 21 45 31 25 48 41 35
mean [%] 63 60 44 117 61 101 74
DSM median [%] 23 47 30 26 44 38 35
mean [%] 65 69 45 134 60 111 81
The Fifth International Symposium on Computational Wind Engineering (CWE2010)
Chapel Hill, North Carolina, USA May 23-27, 2010

The overall difference between the numerical and the wind tunnel results is shown in Table 1.
The mean and the median of the magnitude of the relative differences in cp (with respect to the
measurements) at all measurement points are listed for each of the presented wind direction and
in total over all wind directions. For the wind directions normal to the building the relative differ-
ences are smaller than the ones for the building skewed at 45°, 135°, 225° and 315°. This is partly
due to the fact that for the latter wind directions more measurements with cp close to zero exist, as
can be seen in figure 8. At these positions extremely large differences appear, which is also the
reason for the substantially differences between mean and median; the median provides a more
balanced treatment of the outliers.

5 CONCLUSION

In this study the average pressure coefficients, determined in wind tunnel measurements, are
compared with RANS calculations, using the RNG k-ε turbulence model and a DSM with and
without wall reflection term. The comparison showed that with all turbulence models the general
pressure distribution is well predicted and the influence of different positions of the solar systems
on the roof and their different orientation towards the approach flow qualitatively reproduced.
The largest differences for all models are observed in the wake of systems that have a large spac-
ing, which is due to an incorrect prediction of the separation zones behind the solar energy sys-
tems. The quantitative comparison showed that the DSM leads to the smallest differences; the
overall influence of the wall reflection model is negligible. The RNG predictions deviate most.

6 REFERENCES

Craft, T.J., Graham, L.J.W., Launder, B.E., 1993. Impinging jet studies for turbulence models assessment – II. An
examination of the performance of four turbulence models. Int. J. Heat Mass Transfer 36, 2685-2697.
Fluent, 2006. FLUENT V6.3 User’s guide. Fluent Inc., Lebanon, New Hampshire.
Franke, J., 2007. Introduction to the prediction of wind loads on buildings by Computational Wind Engineering
(CWE). In C.C Baniotopoulos, T. Stathopoulos (eds.), Wind effects on buildings and design of wind-sensitive
structures. Springer, Berlin.
Fu, S., Launder, B. E., Tselepidakis, D. P., 1987. Accommodating the effects of high strain rates in modelling the
pressure-strain correlation. Thermofluids report TFD/87/5, UMIST, Manchester.
Geurts, C.P.W., van Bentum, C.A., Blackmore, P., 2005. Wind loads on solar energy systems, mounted on flat roofs.
Proceedings of the 4th European-African Conference on Wind Engineering, Prague.
Gibson, M. M., Launder, B. E., 1978. Ground effects on pressure fluctuations in the atmospheric boundary layer. J.
Fluid Mech. 86, 491-511.
Launder, B.E., Reece, G.J., Rodi, W., 1975. Progress in the development of Reynolds-stress closure. J. Fluid Mech.
68, 537-566.
NVN 7250:2007 nl, 2007, Solar energy systems – Integration in roofs and facades – Building aspects, NEN, The
Netherlands
Otto, S., Müller, B., Müller, D., 2008, EU-Project EUR-ACTIVE ROOF-er – CFD-Analysis, WP B and C, Final Re-
port, TU Berlin, Berlin.
Richards, P.J., Hoxey, R.P., 1993. Appropriate boundary conditions for computational wind engineering models us-
ing the k-ε model. J. Wind Eng. Ind. Aerodyn. 46&47, 145-153.
Rotta, J., 1951. Statistische Theorie nichthomogener Turbulenz. 1. Mitteilung. Zeitschrift für Physik 129, 547-572.
Tominaga, Y., Mochida, A., Yoshie, R., Kataoka, H., Nozu, T., Yoshikawa, M., Shirasawa, T., 2008. AIJ guidelines
for practical applications of CFD to pedestrian wind environment around buildings. J. Wind Eng. Ind. Aerodyn.
96, 1749-1761.
Wright, N.G., Easom, G.J., 1999. Comparison of several computational turbulence models with full-scale measure-
ments of flow around a building. Wind and Structures 2, 305-323.
Wright, N.G., Easom, G.J., 2003. Non-linear κ−ε turbulence model results for flow over a building at full scale. Ap-
plied Mathematical Modelling 27, 1013-1033.

You might also like