Journal of Wind Engineering & Industrial Aerodynamics 197 (2020) 104056

Contents lists available at ScienceDirect

Journal of Wind Engineering & Industrial Aerodynamics

journal homepage: www.elsevier.com/locate/jweia

Hybrid pressure integration and buffeting analysis for multi-row wind

loading in an array of single-axis trackers
Zachary J. Taylor a, *, Matthew T.L. Browne b
a
RWDI, Inc., Bromont, Quebec, Canada
b
RWDI, Inc., Guelph, Ontario, Canada

A R T I C L E I N F O A B S T R A C T

Keywords: Unrestrained single-axis solar trackers are uniquely flexible structures that can withstand more deflections than
Single-axis tracker typical aeroelastic structures such as long-span bridges or aircraft wings. These distinctive features and the
Wind loading aerodynamics of multi-row arrays lead to the requirement of novel techniques for design wind loads. The present
Pressure measurement
study presents a hybrid technique using two separate wind tunnel tests: (i) a pressure model at a relatively small
Sectional model
Buffeting response analysis
scale, and (ii) a sectional model at a larger scale. The pressure model test is used to measure the buffeting forces
acting on each row of the array and the sectional model study is used to extract the variation in aerodynamic
stiffness and damping as a function of wind speed. The results of these separate studies are combined to
numerically simulate the buffeting response of the tracker, which predicts significant inertial and self-excited
forces. Comparisons between the proposed hybrid method and wind loads estimated using a dynamic amplifi-
cation approach are performed. It is shown that at high wind speeds, the self-excited forces become significant
and the peak design moments exceed those predicted using pressure data alone. The proposed methodology is
expected to complement aeroelastic modeling as a convenient and efficient design tool.

1. Introduction array field were reported for an inclination angle of 35
and a model scale
of 1:24. The impact of perimeter fencing and interior access roads was
Single-axis solar tracker arrays offer greater efficiency than stationary also investigated. Miller and Zimmerman (1981) also investigated wind
arrays of solar panels. Even though the ground cover ratio tends to be loads on ground-mounted arrays in the wind tunnel using pressure
lower, maintaining better alignment of the tracker with the sun leads to measurements rather than strain gauges. Force coefficients and pressure
greater power production. The essential components of any single-axis distributions on the solar panel surface were reported. The dynamic
solar tracker structure include the tracker torque tube, a drive mecha- response of the collectors and their mounting system was also analyzed
nism and the solar panels themselves. In order to optimize the design of and presented in the form of magnification factors applied to the
these components, it is important to determine the environmental loads root-mean-square (rms) pressures. For the wind tunnel speed of 15.2 m/s,
acting on the structure with a high degree of accuracy. In general, the the dimensionless pressure and force coefficients were found indepen-
largest environmental load acting on these structures is due to the wind. dent of Reynolds Number based on model scale tests of 1:12 and 1:24.
Many studies on the wind loads of static solar multi-row flat-plate Kopp et al. (2012) studied a ground-mounted solar array of 12 rows at
arrays have shown the potential complexity of the flow. Early commer- a tilt angle of 20
for several wind directions, although the building effect
cial industrial projects aimed at reducing the cost of solar collector arrays on roof-mounted arrays was the primary focus of the research. Peak
included Bechtel National Inc (1980) and Miller and Zimmerman (1981). factors in the range of 5–7 were reported, and an increase in wind loading
Bechtel National Inc (1980) describes boundary layer wind tunnel tests on the last downwind row in the array was clearly shown. Wind tunnel
on arrays of collectors in which the mean forces and moments were tests were also conducted by Warsido et al. (2014) to investigate the
measured using a six-component strain gauge force balance. Aero- effect of lateral and longitudinal spacing between panels on the wind
dynamic coefficients were reported for a stand-alone solar collector at loading of a ground-mounted solar array. The lateral spacing was shown
different model scales, inclination (tilt) angles, porosities, heights above to decrease wind loads in general. Both the force and overturning
ground, and aspect ratios. Coefficients for selected collectors within an moment coefficients increased with increasing longitudinal spacing,

* Corresponding author.
E-mail address: [email protected] (Z.J. Taylor).

https://doi.org/10.1016/j.jweia.2019.104056
Received 13 September 2019; Received in revised form 1 December 2019; Accepted 1 December 2019
Available online 24 December 2019
0167-6105/© 2019 Elsevier Ltd. All rights reserved.
Z.J. Taylor, M.T.L. Browne Journal of Wind Engineering & Industrial Aerodynamics 197 (2020) 104056

although unusual wind loading was observed at a small longitudinal bodies. These aerodynamic derivatives are necessary to account for the
spacing. forces induced due to structural response that are not fully captured by
Wind tunnel studies of stand-alone solar panels have also been the quasi-steady simplification. The coupled fluid-structure system ex-
studied. Pfahl et al. (2011) investigated heliostats and photovoltaic (PV) hibits changes in total stiffness and damping depending on the wind
trackers with aspect ratios ranging between 0.5 and 3.0; Stathopoulos speed and the response of the structure. In many cases, quasi-steady
et al. (2014) studied a panel with an aspect ratio of 4.6 at inclination theory can be used to reasonably predict the change observed in aero-
angles between 20
and 45
; Abiola-Ogedengbe et al. (2015) presented dynamic torsional stiffness; however, aerodynamic derivatives are
pressure distributions on the upper and lower surfaces of a single panel necessary to include the effect of empirically observed torsional aero-
for four different wind directions; Aly and Bitsuamlak (2013) examined dynamic damping. The torsional aerodynamic stiffness and damping are
the sensitivity of wind loads at different geometric scales, both experi- also directly linked to the aerodynamic stability of single-axis solar
mentally and numerically by Computational Fluid Dynamics (CFD). tracker systems; however, a discussion on stability is outside the scope of
Other researchers have also employed CFD to study the mean wind loads the current work. The focus of the current study is on the derivation of
on ground-mounted solar panels (Bitsuamlak et al., 2010; Shademan design wind loads assuming that the system has already been shown to be
et al., 2014; Jubayer and Hangan, 2014 & 2016; Reina and De Stefano, aerodynamically stable over the range of wind conditions of interest.
2017). The sheltering effect of upwind rows and the effect of row spacing The following section presents the proposed methodology in detail
are among the topics covered. followed by descriptions of the experimental methods and sample results
The study of Strobel and Banks (2014) on the wind loading due to from the experiments. These results are then used directly in the nu-
potential resonance with vortex shedding from the panels highlighted the merical simulations. Sections 5 and 6 use the results of the current study
need to consider wind loading on arrays beyond the typical limit of 1 Hz to discuss the effects of the most important parameters on the results and
used for structures exposed to wind. For highly flexible structures, one of the relationship between the proposed methodology and aeroelastic
the best available engineering tools has been the aeroelastic model. model testing.
Aeroelastic models behave dynamically similar to their full-scale coun-
terparts and can be placed in a wind tunnel to examine the interaction 2. Methodology
with the turbulent wind. However, these types of tests are associated
with high costs due to the level of intricacy involved in the model design The proposed methodology employs two separate wind tunnel tests
and building process. There are also well-documented limitations on the that are dependent on structural geometry, but independent of structural
turbulence simulations possible in boundary layer wind tunnels at the dynamic properties such as stiffness, damping and mass distribution.
larger scales (e.g., Mooneghi et al., 2016). Therefore, while the use of Based on the data extracted from these wind tunnel tests, buffeting
aeroelastic models is recommended to aid in the understanding of the response analysis is applied to predict design wind loads for the solar
fluid-structure interactions in arrays of single-axis trackers, there is also a tracker system. An overall view of the proposed methodology is sketched
desire to develop a more practical and cost-effective design tool. The in Fig. 1, where each component of the methodology is described in the
intent of this study is to present an alternative approach that can be used following sections.
for design and can also supplement aeroelastic model testing. The
approach adopted in the present study combines the measurement of
2.1. Buffeting response analysis
pressures on each row of the array from a static model and measuring the
dynamic characteristics of a typical panel using the sectional model wind
Considering the coordinate system shown in Fig. 2(a), the quasi-
tunnel test. The results of these studies are then combined with a theo-
steady forces and moment per unit length can be approximated to a
retical method based on numerical simulations in the time domain
first-order Taylor series expansion as follows,
referred to as buffeting response analysis to deliver the best response


predictions. 1 1 dCx
Fx ¼ ρuðtÞ2 BCx ðαÞ  ρðU þ u’ ðtÞ  xÞ
_ 2 B Cx ð0Þ þ αðtÞ (1)
The basis of buffeting response analysis has a long history of use in 2 2 dα
both aeronautical design and in the design of flexible civil structures
including skyscrapers and long-span bridges (Davenport, 1962; Irwin,

1 1 dCz
1977). Flexible structures are susceptible to the turbulent fluctuations Fz ¼ ρuðtÞ2 BCz ðαÞ  ρðU þ u’ ðtÞ  xÞ
_ 2 B Cz ð0Þ þ αðtÞ (2)
2 2 dα
that increase in intensity closer to the ground, and buffeting response
analysis is a numerical procedure to predict the response of the structure

1 1 dCm
and its interaction with the turbulent wind. Properly performed aero- My ¼ ρuðtÞ2 B2 Cm ðαÞ  ρðU þ u’ ðtÞ  xÞ
_ 2 B2 Cm ð0Þ þ αðtÞ (3)
2 2 dα
elastic studies on many long-span suspension bridges have been shown to
be in a relatively good agreement with the predictions of buffeting where ρ is the air density, B the panel width and dCx;z;m =dα, the rate of
response analysis (e.g., Irwin et al., 2005), and the standard practice for
change of the drag, lift and moment coefficients with the angle of wind
the design of long-span bridges is to use buffeting response analysis to
attack. The instantaneous effective angle of attack (valid for small angles)
derive wind loads followed by aeroelastic models to confirm the
is sketched in Fig. 2(b) and given by
predictions.
The terms “quasi-static” or “quasi-steady” are often used in the wðtÞ  z_ w’  z_
context of pressure measurements where a gust or peak factor is used to αðtÞ ¼ þθ¼ þθ (4)
uðtÞ  x_ U þ u’  x_
account for turbulent fluctuations based on a measured mean pressure
coefficient (e.g., Davenport, 1967; Banks and Meroney, 2001). In this where θ is the instantaneous structural rotation, and uðtÞ ¼ U þ u’ðtÞ is
paper the term “quasi-steady” is used in the aeronautical context where the instantaneous wind speed composed of the mean wind speed U and
the aerodynamic buffeting forces are assumed to be due to the angle of the turbulent fluctuation u’ ðtÞ. Substitution of Eq. (4) into Eqs. (1)–(3)
attack at each instant in time. In this case the moment coefficient is and reduction of higher order terms gives the linear approximation to the
expressed as Cm ðαðtÞ; α_ ðtÞ; α
€ðtÞÞ  Cm ðαðtÞÞ. One important correction to aerodynamic forces. In general, the along-wind and across-wind stiffness
this simplifying assumption was introduced in wind engineering by are significantly higher than the torsional stiffness of single-axis trackers.
Scanlan and his collaborators (e.g., Scanlan and Sabzevari, 1969; Scanlan Therefore, one of the most important load effects due to wind is the
and Tomko, 1971) with the inclusion of aerodynamic derivatives for bluff moment induced on the torque tube

2
Z.J. Taylor, M.T.L. Browne Journal of Wind Engineering & Industrial Aerodynamics 197 (2020) 104056

Fig. 1. Overview of proposed hybrid methodology.

Fig. 2. Coordinate system definition for wind axis and body axis forces (top) and definition of effective wind angle of attack (bottom).




1 u’ dCm w’ 1 u’ dCm w’ 1 Bθ_
My ¼ ρU 2 B2 Cm þ Cm 2 þ My ¼ ρU 2 B2 Cm þ Cm 2 þ þ ρU 2 B2 KA*2 þ K 2 A*3 θ (7)
2 U dα U 2 U dα U 2 U
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl}
buffeting buffeting selfexcited


1 x_ dCm z_ dCm
þ ρU 2 B2  Cm 2  þ θ (5) Should the vertical and/or lateral frequencies be similar to the
2 U dα U dα
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} torsional frequency then these motions cannot be neglected. However,
selfexcited there are many single-axis tracker designs where these frequencies are
Inspection of the self-excited terms reveals their effects on the significantly higher than the fundamental torsional mode and these
coupled fluid-structure system with the velocity terms modifying the degrees-of-freedom can be safely ignored. Considering the range of static
total damping and the deflection term (dCm =dαÞθ modifying the total tilt angles expected for a solar tracker system and the potential influence
stiffness. These terms are often referred to as aerodynamic damping and of the torque tube, it is recommended that the moment coefficients and
aerodynamic stiffness, respectively. aerodynamic derivatives are directly measured in a wind tunnel test as
Potential flow theory (Theodorsen, 1935) and experimental data described in Section 3.2 (e.g., Rainey, 1957). However, considering the
suggest that the purely quasi-steady expression of the aerodynamic forces similarity between a solar panel and a flat plate, the following approach
(Eq. (5)) does not capture all observed variations in the aerodynamic can be considered for tilt angles close to 0
. For a flat plate the aero-
stiffness and damping. Of particular importance to solar trackers is the dynamic term A*2 and the aerodynamic stiffness term A*3 can be derived
torsional aerodynamic damping term, i.e., the term proportional to θ._ In from Theodorsen’s (1935) functions
the wind engineering of bridges, the self-excited terms are generally

represented by the so-called “aerodynamic derivatives” (e.g., Scanlan
π 4
A*2 ðKÞ ¼  1F G ;
and Tomko, 1971) for the terms they replace in Theodorsen’s (1935) 8K K


potential flow theory π K
A*3 ðKÞ ¼ 2 F  G (8)
2K 4


1 z_ Bθ_ x_ z x
Mse ¼ ρU 2 B2 KA*1 þ KA*2 þ KA*5 þ K 2 A*4 þ K 2 A*3 θ þ K 2 A*6 (6)

2 U U U B B K J1 ðJ1 þ Y0 Þ þ Y1 ðY1  J0 Þ
F ¼ ;
2 ðJ1 þ Y0 Þ2 þ ðY1  J0 Þ2
In Eq. (6) the A*i terms are the aerodynamic derivatives which are

each functions of the reduced frequency K ¼ ωB=U. Assuming that K J1 J0 þ Y1 Y0
G ¼ (9)
contributions from lateral and vertical deflections can be neglected then 2 ðJ1 þ Y0 Þ2 þ ðY1  J0 Þ2
the moment acting on the torque shaft of the tracker system is written
where Ji and Yi are Bessel functions of order i of the first and second kind,
respectively (Scanlan and Rosenbaum, 1968).

3
Z.J. Taylor, M.T.L. Browne Journal of Wind Engineering & Industrial Aerodynamics 197 (2020) 104056

The preceding discussion in this section demonstrates the two key averaged distance over which gusts are correlated in a turbulent flow. An
components of the wind load: buffeting forces and self-excited forces. The integral length scale can be computed in all three spatial directions ðx; y;
buffeting forces arise due to the fluctuations in the turbulent wind and zÞ and for each component of the wind velocity such that a 3  3 tensor is
the self-excited forces are due to the response of the structure to the used to describe the integral length scales in a turbulent flow.
buffeting forces. The analysis considers both of these effects to determine In classical boundary layer theory, the degree of turbulence is quan-
the peak dynamic forces acting on a structure. Further details of this tified through an equivalent roughness length z0 . The same concept has
procedure are provided in the following sections beginning with a been extended to atmospheric flows with typical values of the roughness
description of the turbulence properties. length corresponding to different types of upwind terrain – in general the
roughness length decreases as the terrain becomes less obstructed (i.e.,
2.2. Turbulence properties flow over open country or water). The ESDU standards provide a
comprehensive set of tools to estimate the roughness length depending
Atmospheric turbulence is generally characterized by a spectral on upwind terrain and to determine the turbulence properties as a
description since a key feature of the wind is the wide range in size of function of elevation based on the chosen roughness length (ESDU,
turbulent gust structures. The make-up of the gust structure for neutral 1985). Some typical target turbulence properties for open country terrain
atmospheric conditions is generally dependent on the wind speed, are summarized in Table 1.
elevation, surrounding terrain and latitude of the site (ESDU, 1985).
The atmospheric boundary layer (ABL) is inherently turbulent and is 2.3. Simulation of buffeting forces
characterized by many parameters including the shape of the mean ve-
locity profile, the profile of turbulence intensity for each velocity Synthetic wind storms can be generated from the target turbulent
component ðu; v; wÞ, and the variation in the integral length scales with velocity spectra using auto-regressive techniques (Novak et al., 1995).
elevation. The turbulence intensity is defined as the standard deviation of The target turbulence spectra take the form of modified von Karman
each velocity component divided by the mean ðIu ;Iv ;Iw Þ ¼ ðσ u ; σ v ; σ w Þ= U. turbulence spectra based on ESDU documentation (ESDU, 1985). The
Conceptually, the integral length scale is approximately the largest turbulence is simulated for each strip of the lumped aerodynamic nu-
merical model as sketched in Fig. 3. The target spectra for each of the
Table 1 three velocity components are shown in Fig. 4 along with the simulated
Example of target turbulence properties for numerical and wind tunnel spectra for a point at 10 m following the target values in Table 1.
experiments. The simulated turbulence represents the variation of wind speed at a
Parameter Value Unit single point at the center of gravity of each strip. However, the structure
Roughness length, z0 0.02 m will respond to the average of the wind speeds over its width, B. This
Streamwise turbulence intensity at 10 m 0.17 % averaging produces a low-pass filter effect typically referred to as the
Streamwise length scale at 10 m 90 m aerodynamic admittance. Therefore, each simulated turbulence time
3-s gust speed at 10 m 30 m/s series is filtered based on the chord-wise aerodynamic admittance
Mean velocity power law exponent 0.14 –
functions of Irwin (1977).

Fig. 3. Definition of aerodynamic strips in the numerical model. The arrows demonstrate a small portion of the generated turbulent inflow.

Fig. 4. Comparison of target and measured power spectral densities of wind velocity at 10 m elevation. Geometric scale from the measured wind tunnel data is 1:30.

4
Z.J. Taylor, M.T.L. Browne Journal of Wind Engineering & Industrial Aerodynamics 197 (2020) 104056

2 straightforward for structures which do not have complex flow condi-

ððη1  1 þ eη1 Þðη3  1 þ eη3 ÞÞ2
1
χ ðf Þ ¼ (10)
η1 η3 tions upwind such as long-span bridges and the first row of solar arrays.
However, whether in urban environments or downwind of the first
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

2ffi couple rows of a solar array, local turbulence due to the presence of other
B fLxu
η1 ¼ 0:95 1 þ 70:78 ; structures causes deviations in the flow field from the expected targets.
Lxu U
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Therefore, direct simulation of the buffeting forces is not possible for the

2ffi downwind rows in an array without a complete description of the flow
D fLz u
η3 ¼ 0:475 1 þ 70:78 (11) field in all three spatial directions.
Lz u U
The time varying moment coefficient is computed by discretizing the
In Eq. (11), D is the depth of the section (Fig. 2) and the parameters Lu surface of the solar array into small mesh elements and assigning each
and Lu are the integral length scales of the u component of velocity in the element of the mesh with a time-varying pressure coefficient Cp ðtÞ and a
along-wind and vertical across-wind directions, respectively. This aero- normal vector n. The pressure coefficients are assigned to the mesh ele-
dynamic admittance formulation is similar to Davenport’s (1962); ments using tributary areas which corresponds approximately to a
however, the modification for the depth of the section has been shown to nearest neighbour interpolation scheme. The resulting applied buffeting
better match experiments when the section aspect ratio is large, e.g. B= moment per unit length time series is then computed by integrating in the
D ≫ 1. For typical single-axis tracker systems, this formulation of the radial direction out from the torque tube as follows
aerodynamic admittance removes less energy from the higher frequency
Z
components of the flow compared to the Davenport formulation, which 1 B=2  
Mb;direct ðtÞ ¼ ρU 2 r  Cp ðr; tÞ  n dr (14)
retains some conservatism in the derivation of design wind loads. 2 B=2
The filtered turbulent time series can then be used directly in Eq. (7)
to obtain the buffeting force and moment time series. Therefore, either of 2.5. Numerical solution
the following formulae can be employed in the time domain simulation
of the buffeting forces More details on the method employed here for the numerical solution

  can be found in Stoyanoff (2001) with essential details of the method-
1 dCm  w’f ðtÞ
u’f ðtÞ
Mb;linear ðtÞ ¼ ρU 2 B2 Cm ð0Þ þ Cm ð0Þ2 þ  (12) ology provided here for completeness. The first step in the analysis is to
2 U dα α¼0 U create a finite element model to determine the fundamental modes of
vibration of the structure. Due to the combined temporal-frequency
1   domain formulation of Eq. (7) it is most convenient to carry out the so-
Mb;nonlinear ðtÞ ¼ ρU 2 B2 Cm αf ;
2 lution in a generalized modal space. The state matrix for the generalized
w’f ðtÞ response of the structure forms the basis of the time domain simulations
αf ðtÞ ¼ (13)
U þ u’f ðtÞ
C* K * _
qðtÞ F * ðtÞ
1
¼ €q ðtÞqðtÞ
_  (15)
’
where the streamwise velocity fluctuation uf ¼ F ½χ ðf ÞF ½u ðtÞ is I 0 qðtÞ 0
filtered in the frequency domain using Eq. (10), likewise for the vertical
2 3
fluctuations w’f . For some static tilt angles the slope of the moment co- 2 3
2ζ1 ω1 ⋯ 0 ω2 ⋯ 0
6 1 7
efficient is far from constant (as will be shown in Section 3.2), therefore, C* ¼ 4 ⋮ ⋱ ⋮ 5 6
; K ¼4 ⋮
*
⋱ ⋮ 7
5 (16)
the first-order approximation of Eq. (12) is expected to be less exact than 0 ⋯ 2ζn ωn 0 ⋯ ω2n
Eq. (13). Although Eq. (12) is more computationally efficient, Eq. (13) is
adopted to simulate the buffeting forces in the present study. It is
where q is a generalized coordinate, n is the number of modes included in
important to note the distinction between αf defined in Eq. (13), which is
the analysis and ωn ; ζn are the natural frequency and structural damping
independent of structural motion, compared to the formulation in Eq. (4).
of the nth mode in rad/s and ratio to critical, respectively. The aero-
If the structural motions were included in Eq. (13) then the self-excited
dynamic stiffness and damping matrices are determined for each appli-
forces would be double counted in the simulation since these are
cable strip in the numerical model (Fig. 3) and generalized for each mode
accounted for with the aerodynamic derivative terms. This observation
of vibration to account for the effect of the self-excited forces on the total
leads to the possibility of treating the buffeting and self-excited forces
stiffness and damping of the coupled fluid-structure system. The aero-
separately and allows for the direct measurement of buffeting forces on a
dynamic damping and stiffness matrices are included in the left-hand side
static model.
of Eq. (15) as described by Stoyanoff (2001). The applied moment due to
buffeting at each strip of the numerical model is either determined
2.4. Direct measurement of buffeting forces analytically using the buffeting theory described by Eq. (13) or through
direct measurement as discussed in Section 2.4. These distributed forces
A static model instrumented with many pressure measurement loca- are then generalized and applied to the above state space system. The
tions is one of the standard methodologies for deriving wind loads on solution of Eq. (15) is carried out in the time-domain using an exact
arrays of solar panels (e.g., Kopp et al., 2012). These models are intri- numerical integration scheme detailed in Stoyanoff (2001). Numerical
cately built to faithfully represent the geometry of the structure while simulations are typically carried out over a range of wind speeds up to
minimizing the effect on the surrounding wind field of the tubes linking and beyond the project criteria.
the panel openings to the pressure measurement system. However, these
models remain considerably more cost effective than aeroelastic models, 2.6. Loads resisted by the structure
which must also represent the dynamic properties and mass distribution
of the structure. Further details on the pressure measurements performed For the design of flexible structures equivalent static wind loads are
for this study are provided in Section 3.1. typically derived, which include the buffeting, self-excited and inertial
A significant advantage of the direct measurement of buffeting forces forces (induced by the motion of the structure). These equivalent static
in this way allows for a realistic characterization of the forces and mo- wind loads are snapshots or statistical representations of the loads
ments acting on downwind rows due to the wake of upwind rows. In the resisted by the structure. The torque per unit length resisted by a single
previous section a method was described which could generate artificial degree of freedom system with torsional stiffness, kθ , is:
time series based on target velocity spectra. This is relatively

5
Z.J. Taylor, M.T.L. Browne Journal of Wind Engineering & Industrial Aerodynamics 197 (2020) 104056



1 u’ dCm w’ 1 Bθ_
Mresist ¼ kθ θ ¼ ρU 2 B2 Cm þ Cm 2 þ þ ρU 2 B2 KA*2 þ K 2 A*3 θ  |{z} I €θ  cθ θ_ (17)
2 U dα U 2 U inertial |{z}
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} damping
buffeting selfexcited

In Eq. (17) the damping terms are expected to be small since they 3. Experiments
occur out-of-phase with the resisting force; however, these will only be
exactly zero for single degree of freedom systems. For extension to multi- 3.1. Pressure integration study
degree of freedom systems the reader is referred to Stoyanoff (2001) and
Stoyanoff and Dallaire (2013). It is useful to represent the resisting force To obtain the direct buffeting forces, a model of the ground mounted
in a coefficient form, which is defined as solar array was constructed at a 1:30 scale. This model (Fig. 6) was tested
RL in one of RWDI’s boundary layer wind tunnels in Guelph, Ontario, for a
Mresist ds generic array of seven rows. The static tilt angle was adjustable within
Cmresist ¼ 0
: (18)
1
2
ρU 2 B2 L the range of 0
through 60
(from horizontal), with 0
, 5
, 10
, 20
, 30
,
45
, and 60
selected for testing at a center-to-center row spacing of 3B.
Although expressed in this coefficient form, examination of Eq. (17)
Each row consisted of 90 solar modules creating row lengths of approx-
indicates that the reference pressure and area are not sufficient to non-
imately 11.5B, and each row was instrumented at 15 discrete spanwise
dimensionalize the resisting moment. To explore its dependence on
locations (Fig. 7).
stiffness, mass moment of inertia, structural damping, chord length,
Taps that measure fluctuating wind pressure were installed on the top
proximity to the ground or inter-row spacing, case-specific dynamic an-
and bottom surfaces of the panels at various locations. At each of the
alyses must therefore be performed when significantly altering these
spanwise locations (Fig. 7) there were 8 pressure taps with 4 on the top
parameters.
surface and 4 on the bottom. Therefore, the total number of simultaneous
In the case where self-excited forces can be neglected, the total
pressure measurement locations was 840 for the entire array with 120
resisting moment is approximated from pressure integration studies
per row. Attention was given to the installation of the pressure taps so
using the measured static pressures and a Dynamic Amplification Factor
that they did not interfere with the flow around the panel, which was
(DAF) approach. The dynamic amplification is determined in the fre-
achieved by imbedding interior pressure passages in the solid model
quency domain as a function of the natural frequency, fn , and the
using rapid-prototyping technology. These interior passages then termi-
damping ratio of the structure ζ (e.g., Davenport, 1964) and the standard
nated underneath the base of the model, connected to rubber tubes to
deviation of the measured moment, σ M
transmit the pressures into a multiplexed pressure scanning system.
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Pressure time series were measured from each tap with empirically
π fn Sðfn Þ determined transfer functions applied to the data in the frequency
DAF ¼ 1 þ : (19)
4ζ σ 2M domain to correct for distortions induced by the tube (Irwin et al., 1979).
The measured data were converted into pressure coefficients based on
where Sðfn Þ is the power spectral density of the moment. However, while the mean dynamic pressure measured at a height of 1.5 m in the wind
the dynamic amplification approach can reasonably reproduce the dy- tunnel, directly above the center of the test model.
namic inertial loading from a static measurement, it does not address the Beyond the modeled area, the influence of the upwind terrain on the
potential contributions of large self-excited forces. A sketch of the two atmospheric boundary layer was simulated in the testing by appropriate
methods is shown in Fig. 5. As the critical wind speed approaches the roughness on the wind tunnel floor and flow conditioning spires at the
unstable region, the self-excited forces can become quite large and upwind end of the working section for each wind direction. The turbu-
important distinctions between the buffeting response analysis and the lence simulation was targeted to represent generic open terrain condi-
DAF method are typically observed. It is also observed that as the reduced tions for all wind directions since these flow conditions represent the
wind speed increases, the available energy in the power spectrum of typical exposure for the full-scale installations. Fig. 4 shows the target
acceleration decreases and leads to a lower predicted dynamic load using and simulated power spectral densities of turbulence fluctuations at 10 m
the DAF method. full scale elevation.
The entire array model was mounted on a turntable that could be
rotated 360
. In this study, due to symmetry, measurements were ob-
tained for 19 wind directions spaced 10
apart. Although the pressure
model technique allows the direct buffeting forces to be obtained for
different wind directions, it is limited in predicting the self-excited and
inertial forces of a moving structure. As these forces are most dominant
for wind azimuths normal to the tracker, the sectional model technique
described in the next section is the most practical approach to determine
these additional aeroelastic forces.

3.2. Sectional model study

The purpose of the sectional model study is to represent a section of

the single-axis tracker with a rigid model supported by springs. Sectional
model studies are typically performed on aircraft wings and long-span
Fig. 5. Sketch demonstrating the regimes and applicability of predicted peak bridges to obtain aerodynamic information about the typical section.
dynamic moment using buffeting response analysis predictions and dynamic Conceptually, the information obtained in this study should be directly
amplification factor (DAF). applicable to the aerodynamic strips sketched in Fig. 3 to understand

6
Z.J. Taylor, M.T.L. Browne Journal of Wind Engineering & Industrial Aerodynamics 197 (2020) 104056

Fig. 6. Photograph of the pressure model in the wind tunnel.

Fig. 7. Layout of the pressure taps for one row of the single-axis tracker model. Units of measurement are ratios to the width B.

Fig. 8. Photograph of the single-axis tracker sectional model in the wind tunnel.

their effect on the full structure. study is shown in Fig. 8. One challenge of modeling solar trackers using
End effects such as those due to the tip vortex may bring some un- the sectional model technique is to balance the mass and the size of the
certainty in the predicted wind loads when establishing the base aero- model with wind tunnel blockage. If the scale of the wind tunnel model is
dynamic properties solely using sectional model tests. These end effects defined by λ ¼ Bms =Bfs where subscript ms refers to model scale and fs
are, however, well modeled in the pressure model test thus there would refers to full scale, the scaling for mass and mass moment of inertia per
remain only an uncertainty considering the self-excited forces at the unit length are given by dimensional analysis
tracker ends. The installation of the trackers close to the ground however
weakens the tip vortex and limits these aerodynamic end effects. λmass ¼ λ2 ; λmmi ¼ λ4 (20)
For single-axis solar tracker systems, the two main aspects recom- The scale for the current study was chosen to be λ ¼ 1=8 and the wall-
mended for sectional model studies of wind loads are to measure (i) the to-wall length of the model in the wind tunnel was 1.8 m.
static force and moment coefficients, and (ii) the aerodynamic de- The model itself has been confirmed to be sufficiently rigid – typically
rivatives. A wind tunnel photo of the sectional model used for the current it needs to be in the range of at least 3 times the frequency modeled in the

7
Z.J. Taylor, M.T.L. Browne Journal of Wind Engineering & Industrial Aerodynamics 197 (2020) 104056

wind tunnel suspension rig. The frequency is controlled by changing the speeds expressed as the reduced wind speed Ur ¼ U=fB where f is the
point of fixity on a square torsion rod, and the damping is controlled torsional frequency measured in the wind tunnel. The aerodynamic de-
through the use of an eddy current system controlled through increasing rivatives are determined using the free vibration method. Although it is
or decreasing the number of permanent magnets located at a fixed dis- expected that forced vibration experiments would yield similar results, it
tance from an aluminum plate that moves with the model. These controls has been shown that there may be sensitivity to the method chosen in the
of the dynamic properties are located within an area that is sheltered multi-laboratory comparison performed by Sarkar et al. (2009). Addi-
from the wind and they allow for a wide range of adjustment in the base tionally, as pointed out by Sarkar et al. (2009) and observed through
dynamic properties of the system. experience on several projects carried out at RWDI, it is expected that
there is also a potential amplitude dependency of the measured aero-
3.2.1. Static force and moment coefficients dynamic derivatives. Therefore, it is necessary to ensure the range of
The wind tunnel model is attached to a sensitive torque shaft on either expected deflections is considered in the test program. In the current
side of the model. This torque shaft is instrumented to measure the drag, experiments, the model is given an initial rotation of approximately 10

lift and torque. To measure the static force and moment coefficients, the from its static position using synchronized pneumatic solenoids on either
wind tunnel model motion needs to be mechanically restrained. Once the side of the model. Once released, the model is free to vibrate and the
motion has been restrained, a calibration procedure is performed on the displacements are measured using two pairs of highly sensitive laser
torque shafts by application of known weights. The static force and transducers (Micro Epsilon optoNCDT 1302-50) on either side of the
moment coefficients are obtained by measuring the applied wind forces model. The displacements are averaged to measure the displacement and
at various wind speeds in terms of the total drag, FD , lift, FL , and moment, differenced to measure the rotation of the model with accuracies of 50
M acting on the model. These forces and moment are normalized in co- μm and 0.02
, respectively. Multiple releases at each wind speed are
efficient form as follows, performed to obtain a better fit to the observed empirical data and to
gauge the repeatability of the measurement.
FD FL M 1
CD ¼ ; CL ¼ ; Cm ¼ 2 ; q ¼ ρU 2 (21) The analysis of the data is carried out on each individual decay curve
qB qB qB 2 using the Ibrahim method (Ibrahim and Mikulcik, 1977), which has been
To capture the variation with angle of attack, the entire supporting demonstrated as effective on bridges (e.g., Stoyanoff and Larose, 2004).
system can be rotated as shown in Fig. 8. The torque shafts move with the The decay curves assessed in the present study all began at amplitudes of
supporting system therefore the forces are measured in a body-aligned 5
allowing the model to recover from its initial rotation of 10
. Expo-
coordinate system at each angle of wind attack (Fig. 2). The conversion nential decay curves were found to reasonably envelope the observed
from the body-aligned coordinate system to the wind coordinate system decays in this range suggesting a minimal dependence on amplitude over
can be performed using a suitable rotation matrix. At each angle of a 5
range. As shown for static tilts of 0
and 20
in Fig. 10, the measured
attack, the forces are recorded with no wind and subtracted from the data points of the aerodynamic derivatives are fit with cubic splines to
forces measured at two different wind speeds. Two wind speeds are used both interpolate and extrapolate the measured data. It should be noted
to simultaneously assess both the repeatability of the data and the effect that the free vibration technique cannot capture strongly positive values
of wind speed on the measured coefficients. of A*2 since this represents a negatively damped system. Also shown in
The force and moment coefficients have all been measured with the Fig. 10 are the Theodorsen (1935) predicted curves for both A*2 and A*3 at
effect of the ground included physically in the wind tunnel (Fig. 8). The 0
static tilt angle and the quasi-steady predictions based on the
measured force and moment coefficients are shown in Fig. 9 in both measured moment coefficient slopes following
body-aligned and wind coordinate systems. No turbulence generating
devices were included in the wind tunnel for the measurement of force 1 dCm
A*3  (22)
and moment coefficients, and the measured turbulence intensity was less K 2 dα
than 1% in the oncoming flow. One important distinction is noted between the measured A*3 and the
prediction from quasi-steady theory at a static tilt angle of 20
. From the
3.2.2. Aerodynamic derivatives measured moment coefficients in Fig. 9 it is clear that there is minimal
The same model and mounting system utilised to measure the static change in moment coefficient with angle of attack for a static tilt angle of
force and moment coefficients is used to extract the aerodynamic de- 20
(for winds generally normal to the structure). Eq. (22) reveals that
rivatives. However, for the aerodynamic derivatives it is no longer locked the quasi-steady variation of stiffness is dependent on the change in
in place and the model is free to move and interact with the surrounding moment coefficient with angle of attack. Therefore, based solely on static
wind field. As in the case of static force and moment coefficients, the moment coefficient measurements, it might be concluded that this is the
wind tunnel tests for aerodynamic derivatives were carried out with most stable tilt angle since no changes in aerodynamic stiffness are ex-
levels of freestream turbulence below 1%. pected (concerns for torsional instability). However, the aerodynamic
The tests for aerodynamic derivatives cover a wide range of wind

Fig. 9. Measured static force and moment coefficients.

8
Z.J. Taylor, M.T.L. Browne Journal of Wind Engineering & Industrial Aerodynamics 197 (2020) 104056

Fig. 10. Examples of measured aerodynamic derivatives at 0
and 20
static tilt angles from the sectional model tests along with the adopted curve fits. Quasi-steady
and theoretical flat plate (Theodorsen, 1935) estimates are included for comparison to measured values. Legend applies to both plots.

derivatives show that the stiffness (through A*3 ) does not vary as pre- as a ratio to the upwind reference wind speed at the tracker height using a
dicted by quasi-steady theory and that the torsional damping (through pair of high frequency pressure-based velocity TFI Cobra probes to cap-
A*2 ) goes negative at a reduced wind speed of approximately Ur  3 ture the turbulence properties of the flow as well as the mean wind speed.
suggesting that this configuration is less aerodynamically stable than a For tilts greater than approximately 40
the data measured in this way
static tilt angle of 0
. were unreliable due to large zones of recirculating flow. For future
studies it is recommended to use particle image velocimetry (e.g., Taylor
et al., 2010; Pratt and Kopp, 2013) to improve the understanding of how
3.3. Measurement of wind speed between rows speed varies from row to row for these large static tilt angles. The
mode-shape weighted mean wind speeds measured for tilt angles of 0
,
Although the buffeting forces can be measured directly using the 5
, 10
and 20
are shown in Fig. 12 as ratios to the upwind reference
pressure model in the wind tunnel (Section 3.1), the self-excited forces speed where the largest value of the mode shape (for the 1st torsional
must still be accounted for in the simulation of the structural response. mode) occurs at the end of each row. In each case, a decrease is observed
The observed effects on both the aerodynamic stiffness and damping of in the wind speed between the first and second row with speeds then
the self-excited forces implies that they may have a significant effect on increasing for rows 5–7 at the end of the tested array. The increase in
the predicted structural response and the resulting design wind loading. speeds towards the end of the array is consistent with observations in the
The self-excited forces are normalized by the reference wind speed dur- mean moment coefficients. It is hypothesized that the flow is accelerating
ing the sectional model tests and can be scaled to full scale wind speeds due to the low pressure region in the wake of the entire array; however,
with knowledge of the design wind speed. However, while this is a further tests would be necessary to determine how more or fewer rows
straightforward process for the first row in a solar array, the wind speed would affect the measured results in the current study.
experienced by the downwind rows must be determined experimentally.
The reason these must be determined experimentally relates to the sig- 4. Numerical simulations of structural response
nificant change in the spectra of the buffeting forces and moments
observed from static pressure model tests of downwind rows both in the 4.1. Comparison between direct measurement and simulation of buffeting
current data and previously published data (e.g., Strobel and Banks, moments
2014).
The pressure model was used to measure the wind speed between A comparison between the simulated buffeting moments and those
each subsequent row as shown in Fig. 11. The wind speed was measured

Fig. 11. Location of inter-row speed measurement with respect to wind direction.

9
Z.J. Taylor, M.T.L. Browne Journal of Wind Engineering & Industrial Aerodynamics 197 (2020) 104056

Fig. 12. Ratio of inter-row mean wind speed to upwind reference wind speed for four different tilt angles.

Fig. 13. Comparison of moment coefficients resisted by the structure as assessed by numerical simulation of the buffeting force and direct pressure measurement at
þ5
static tilt and Ur ¼ U=fB ¼ 2:5. Top: extraction of full-scale time series, bottom left: probability densities, bottom right: power spectral densities. Legend applies to
all plots.

measured directly is carried out for a static tilt angle of 5
. Time series, extract the filtering effect of the ground are more complex (e.g., Larose,
probability densities and power spectral densities for each method are 1999) and outside the scope of the current study.
shown in coefficient form of the resisting moment in Fig. 13. Since these
are the moments the structure must resist, these should be considered as 4.2. Peak resisting wind loads
the design wind loads for the first row in an array. Similarities and dis-
tinctions are clearly observed between the predictions of the two The results of the time domain simulations are assessed to compute
methods. Comparison of the power spectral densities reveals that the time series of the resisting moment coefficient (Eq. (18)). The wind load
variance of the first two torsional modes of vibration are very similar considered for design should be the peak load. This peak load is repre-
with more energy content predicted in the third mode of vibration for the sented by Cb mresist and the row-to-row variation for the case of 5
static tilt
direct measurements and vice versa for the fourth torsion mode. angle is shown for three reduced wind speeds of Ur ¼ 2:0, Ur ¼ 2:83, and
The peaks of the absolute value for the resisting moment are also Ur ¼ 3:16 in Fig. 14. The probability density functions of the resisting
observed to be similar in Fig. 13 However, there is a significant skewness moment coefficients are estimated using a Gaussian kernel density esti-
observed in the direct pressure measurements compared to the relatively mation and shown for each row in Fig. 14. Also shown in Fig. 14 are the
symmetric simulated moments. It is expected that this observation may mean, standard deviation and peak resisting moment at each row. Peak
be due to additional filtering of the turbulent flow field by the ground. moments from the pressure study alone are obtained through a dynamic
The aerodynamic admittance (Eq. (10)) describes the effect that the solar amplification factor (Eq. (19)) to approximate the inertial loading, and
panels themselves have on the incoming turbulent flow; however, further these estimates are wind speed independent in coefficient form. The
research would be required to investigate the effect of the ground in predicted mean and peak moments direct from the pressure study are
filtering the turbulent energy. It should be noted that the effect of the included for comparison in Fig. 14. The DAF approach in the pressure
ground was included for both the measurement of the static force and study is generally shown to be conservative until a high reduced wind
moment coefficients and the aerodynamic derivatives. However, tests to speed is reached (e.g., Ur ¼ 3:16) and the self-excited and inertial

10
Z.J. Taylor, M.T.L. Browne Journal of Wind Engineering & Industrial Aerodynamics 197 (2020) 104056

Fig. 14. Row-to-row variation of the resisting moment coefficient at three different wind speeds. The probability density function is shown (left) along with statistical
quantities (right) based on numerical simulations of self-excited forces and measured pressures. The static tilt angle is þ5
and (a) Ur ¼ U=fB ¼ 2:0, (b) Ur ¼ 2:83,
and (c) Ur ¼ 3:16.

moments become significantly pronounced. The observed shift in the Section 4.1, it is expected that this discrepancy is related to the filtering
mean with wind speed in the numerical simulations is explained through effect of the ground on the turbulent flow around the panel. However,
the marked change in aerodynamic stiffness of the system. further research would be required to fully answer this question.
The effect of tilt, row position and wind speed are shown in Fig. 15. In general, the first two rows experience much higher design wind
Results from the simulation of the buffeting forces are also included for loads compared to the downwind rows – especially for the case of posi-
the first row, which is where they are expected to be most applicable. The tive tilt angles. Therefore, one potential strategy for accommodating the
agreement in design wind loads between the simulated and measured design wind event for an array is to strengthen the perimeter rows.
buffeting forces is observed to be quite good for tilt angles of 5
, 20
and However, the tools developed in the current study are expected to be
20
. Except in the case of 30
the buffeting simulations predict similar suitable to design any desired stow and/or design strategy.
or higher design loads than by using the direct measurement of the
buffeting forces. However, in the case of 30
, the design loads predicted 5. Effects of wind direction
by simulation of the buffeting force match reasonably well with the re-
sults of the second row. For the case of 0
the buffeting simulations result 5.1. Buffeting forces
in noticeably higher loads than those estimated using the direct mea-
surement of the buffeting forces. As discussed previously at the end of Buffeting forces on a multi-row array will occur from all wind

11
Z.J. Taylor, M.T.L. Browne Journal of Wind Engineering & Industrial Aerodynamics 197 (2020) 104056

Fig. 15. Plots of peak resisting moment with reduced wind speed for various static tilt angles on each of the seven rows. Dashed-dot lines are drawn to large moment
values when the numerical solution diverged.

Fig. 16. Plots of measured buffeting forces and moments with wind direction for various static tilt angles on each of the seven rows. Wind directions of 90
and 270

are normal to the single-axis trackers.

directions. The magnitude of these forces depends on several factors corresponding to tilt angles of 0
(top plot), 10
(middle plot), and 30

including the load effect being examined (e.g., normal force or moment), (bottom plot). To illustrate the relative magnitudes for each tilt angle, the
averaging (or tributary) area considered, location within the array, wind coefficients have been normalized by the maximum value from all three
speed, exposure, and several other site and geometric parameters tilt angles. Similarly, the middle column of plots corresponds to the
including tilt angle. The effect of tilt, row position, averaging area, and moments for an end post, and the right column of plots corresponds to the
wind direction are shown in Fig. 16. The normal forces for an averaging moments acting on the full tracker row. As the tilt angle increases, the
area representing an end post are provided in the left column of plots, magnitude of the normal forces and moments on an end post also

12
Z.J. Taylor, M.T.L. Browne Journal of Wind Engineering & Industrial Aerodynamics 197 (2020) 104056

increase with the maximum occurring for a tilt angle of 30
. The general pressure model allows for a more appropriate simulation of turbulence
trends are observed to be similar to those shown by Bechtel National Inc (i.e., Fig. 4).
(1980). The wind direction causing the peak loads at 30
tilt is shifted
40
–50
from the normal wind direction of 90
. Whereas the maximum 6.1.2. Design tools
moments on the full row occur at 10
tilt and within 20
of normal. The Aeroelastic models bear a significant financial cost that may be
row-to-row trends seem to be consistent with those identified by Kopp difficult to justify in many cases. Even in the case that an aeroelastic
et al. (2012). model test has been performed, justification for a second or third itera-
tion of the physical model as the design evolves could be challenging.
5.2. Inertial and self-excited forces Therefore, in cases of design evolution it is expected that the proposed
method offers several advantages. First, the pressure model wind tunnel
Based on experience with full aeroelastic model testing on multi-row test may remain independent of structural properties providing the
tracker systems, the highest torques on an entire tracker row occurred general shape, spacing between the rows and spacing to the ground
from winds within 30
normal to the array. Beyond this range the peak remain the same. Then any change in other design elements, such as
torques dramatically reduced in magnitude. These observations are torque tube thickness or changes in mass distribution, can be accounted
consistent with what has been found for long-span bridges, where the for in the numerical buffeting response analysis. Furthermore, the addi-
highest loads generally occur due to winds normal to the bridge. tion of auxiliary devices such as viscous dampers can all be investigated
The wind tunnel study of the Lions’ Gate Bridge by Irwin and Schuyler numerically as opposed to the challenging alternative to model these
(1977) found that the critical flutter speed increased as the wind direc- effects at model scale. Likewise, due to the numerical nature of the
tion deviated off normal to the bridge. They found that the increase in the proposed method, more information is typically available for design than
critical onset wind speed for flutter could be approximated by consid- would be possible with physical modeling due to limitations of
ering only the portion of the velocity normal to the bridge, or Ucrit ∝ secβ instrumentation.
where β ¼ 0∘ is winds normal to the span. A similar notion was explored
by Scanlan (1999) in the so-called “skewed winds theory” for aero- 6.2. Recommendations for future work
dynamic stability of long flexible structures. Similar to the observations
of Irwin and Schuyler (1977), Scanlan (1999) proposed that the wind The proposed methodology here is continually evolving as it is a
speed should be reduced by the cosine of the angle, in addition to an relatively new technique. As with any combined empirical/analytical
elongation of the effective width as seen by the wind and suitable approach there remain some open questions. These questions and our
transformation of the forces and moments. preliminary understanding are discussed in this section.
As described in the previous section, the maximum buffeting force
may come from a wide range of angles. However, many empirical ob- 6.2.1. Effects of turbulence on the aerodynamic derivatives
servations reveal that in order to maximize either the inertial forces or The aerodynamic derivatives (as described in Section 3.2.2) were
the self-excited forces the flow is required to be well correlated along the measured in a nominally smooth flow (i.e., turbulence intensity < 1%).
length of the structure. Therefore, it is expected that the most significant However, even in open country exposures, these structures will be sub-
effects due to the inertial and self-excited forces will be within 30
of jected to winds of high turbulence intensity due to their proximity to the
normal to the array. ground. The relatively large scale of the sectional model in comparison to
pressure models would make full modeling of the turbulent flow field
6. Discussion impossible (see discussion in Section 6.1.1). Likewise, there is some
ambiguity introduced in extracting the self-excited forces – and from
6.1. Relationship to aeroelastic model testing there the aerodynamic derivatives – when there is a large buffeting force
present. This ambiguity is introduced because the exact buffeting force is
In the absence of the proposed model the only methodology currently unknown during the wind tunnel test and a full description of the aero-
available to examine the peak wind loads of these flexible tracker arrays dynamic admittance function is generally lacking. However, our expe-
is the aeroelastic model test. Aeroelastic modeling is encouraged by the rience on a wide range of bridge cross-sections has shown that the
authors whenever possible as an accurate methodology to examine both aerodynamic derivatives are generally similar between nominally
the stability and wind loads of tracker arrays, and it should be noted that smooth flow conditions and when turbulence with moderate intensity
the proposed hybrid pressure and buffeting analysis methodology is not (less than 20%) is added to the free stream. It should be noted, however,
intended to replace aeroelastic modeling. In the ideal case the proposed that distinctions have been observed between smooth and turbulent
method would be used to complement aeroelastic model testing; how- flows on occasion and future work in this area would be useful to un-
ever, it is expected that the proposed method is suitable to estimate the derstand the effect of atmospheric turbulence on the self-excited forces.
wind loads of a proposed design when time and/or budget prohibits the
use of aeroelastic model testing. 6.2.2. Effects of gust duration
The force and moment coefficients are normalized by the steady wind
6.1.1. Turbulence simulation speed measured in the wind tunnel during the experiment. This mea-
Aeroelastic model testing is generally considered the most represen- surement and definition of the wind speed is generally straightforward in
tative form of empirical testing when examining problems of fluid- a well performed sectional model test. One of the assumptions of the
structure interaction. However, as with any scale model testing, there quasi-steady theory is that this steady wind speed measured in the tunnel
are limitations in interpreting the wind tunnel results at full scale. Larger U can be evaluated instantaneously U ¼ uðtÞ (Eq. (3)). Therefore, the
model scales (on the order of 1:10) are generally best to capture geo- effect of gusts is inherently accounted for in the buffeting force. However,
metric details on aeroelastic models of solar tracker arrays. These large the assumption in the measurement and use of aerodynamic derivatives
scales are chosen in order to capture the aerodynamics due to section to evaluate the self-excited forces is less straightforward because of the
geometry and to properly model the dynamic response of the structure. normalization by reduced frequency K ¼ ωB=U. Structures average out
However, these scales often suffer from limitations of the length scales variations of the wind speed in space and time related to the geometry
achievable in most atmospheric boundary layer tunnel facilities. One and natural frequencies of the structure. Due to this averaging effect,
method employed to resolve this challenge is the so-called partial tur- large structures like buildings and bridges generally respond to changes
bulence simulation (Mooneghi et al., 2016). These limitations are over- in the wind speed on the order of 10 min to 1 h. However, for smaller
come in the proposed hybrid model because the smaller scale of the structures with higher frequencies it is expected that the structure can

13
Z.J. Taylor, M.T.L. Browne Journal of Wind Engineering & Industrial Aerodynamics 197 (2020) 104056

respond to shorter duration gusts. In the current study the normalization Numerically this change in slope limits the growth of deflections with
of the reduced frequency has been performed based on the 1-h mean increasing wind speed in contrast to the results predicted by aerodynamic
wind speed. The validity of this assumption has been further assessed in derivatives. At these large deflections there are uncertainties regarding
Section 6.2.4 and research is ongoing in this area. the accuracy of each method; however, the good agreement at lower
speeds allows some key initial findings to be drawn.
6.2.3. Amplitude dependence As discussed in Section 6.2.2, the aerodynamic derivatives are a
One of the advantages of the quasi-steady approach is in the departing function of reduced frequency which requires the definition of a
assumption that buffeting forces and moments are amplitude indepen- normalized wind speed. The question of averaging period for this wind
dent. The extension of the quasi-steady buffeting theory through the use speed remains an ongoing investigation; however, the direct quasi-steady
of aerodynamic derivatives raises the question of amplitude dependence. method does not require such a definition as the wind speed is evaluated
Free vibration experiments to extract aerodynamic derivatives capture instantaneously with no averaging performed. Therefore, the relatively
any variation in the mean rotations as a function of reduced wind speed; good agreement shown in Fig. 17 suggests that for the current case the
however, the extracted decays are necessarily fit through a given range of effect of gust averaging on the predicted responses is suitably captured in
amplitudes. In contrast to aircraft wings and bridges, the flexibility of the simulation of aerodynamic derivatives using the assumption of the
single-axis tracker systems allows for significant structural rotation prior mean hourly average. It is recommended that further work be performed
to failure. Therefore, it is recommended that future studies attempt to in this area to address these questions both analytically and
quantify the effect of vibration amplitude on the measured aerodynamic experimentally.
derivatives. In the current study minimal variation was observed in the The other preliminary finding of this comparison suggests that the
decays over a range of dynamic rotations between approximately 5
. aerodynamic derivatives measured in the nominal range of 5
may have
However, greater uncertainty is introduced in the self-excited force little dependence on amplitude up to dynamic rotations approaching
description once structural vibrations exceed this nominal range. 20
. Beyond 20
there is a noticeable difference between the two
simulation methods. At very high angles of attack, the section stalls,
6.2.4. Numerical verifications where the flow becomes completely detached and turbulent leading to
Experimental verification of some of the questions raised in Section aerodynamic loads which are difficult to explain with the current theo-
6.2.1-6.2.3 is outside the scope of the current work. Although experi- retical framework. It is expected that the best tool to answer the question
mental verification is recommended, it is possible to perform some pre- around amplitude dependence is either through sectional model extrac-
liminary verification work numerically. The numerical verification has tion of aerodynamic derivatives over a wide range of amplitudes or
been performed using two different types of buffeting response analysis through direct comparison to full aeroelastic model testing. However, it
using the same synthetic wind storms. The first type of buffeting response should be noted that field measurements on single-axis tracker systems
analysis is as described by Eq. (7) with the effect of self-excited forces have shown an amplitude dependence on both frequency and damping in
captured by aerodynamic derivatives. The second method is the direct still air conditions (due to stick-slip conditions in the bearings, etc.).
quasi-steady simulation given by Eq. (3) where Cm ðαÞ is evaluated at each These observations indicate that while aeroelastic model testing and the
time step based on the measured moment coefficient curve (Fig. 9) and proposed hybrid method are expected to be valuable design tools they
the instantaneous angle of attack. The advantages of the direct quasi- should not be expected to fully replicate the exact conditions of the as-
steady simulation are a frequency independent formulation and inde- built structure.
pendence of response amplitude. However, the disadvantage of the direct
quasi-steady simulation is that the aerodynamic damping term (captured 7. Conclusions
by A*2 in the aerodynamic derivatives) is not accounted for.
The comparison of the numerical simulations by both methods is The focus of this study is a new proposed methodology for evaluating
shown in Fig. 17 for a static tilt of 0
where generally good agreement is the design wind loads of single-axis tracker arrays. The method relies on
observed up to a reduced speed of approximately Ur ¼ 3:0. The good primarily two experimental techniques: a pressure integration study and
agreement between the two methods ends at approximately 20
of peak a sectional model study. The results of these studies are then combined in
dynamic rotation at which point the value of the moment coefficient numerical simulations for each row of an array and peak resisting mo-
curve begins decreasing with increasing angle of attack (Fig. 9). ments can be extracted for the design of structural components. The

Fig. 17. Comparison of peak dynamic rotations for buffeting analysis with aerodynamic derivatives and quasi-steady theory at 0
static tilt angle.

14
Z.J. Taylor, M.T.L. Browne Journal of Wind Engineering & Industrial Aerodynamics 197 (2020) 104056

numerical simulations account for three fundamental components of proposed method is the best available alternative. This hybrid method
wind loading on the single-axis tracker: accurately combines the buffeting forces, inertial forces and self-excited
forces to determine design wind loads for arrays of single-axis solar
1. The measured buffeting forces from the pressure integration study; trackers.
2. The change in aerodynamic stiffness and damping through the self-
excited forces measured in the sectional model study; and Author contributions
3. The dynamic properties of the single-axis tracker including mass
distribution, mode shapes, natural frequencies and damping. Zachary J. Taylor: Development of methodology, data analysis,
manuscript preparation, reviewing and editing.
Comparisons have been performed between the proposed method of Matthew T.L. Browne: Performed experiments, data analysis, manu-
direct pressure measurement with a buffeting response analysis of the script preparation, reviewing and editing.
first row using synthetic wind storms. In general, the agreements be-
tween the two methods are reasonable with the simulated buffeting
moments generally greater (i.e., more conservative) than those measured Declaration of competing interest
directly in the wind tunnel experiment. Furthermore, comparisons to the
peak predicted moments using a dynamic amplification factor (DAF) on The authors declare that they have no known competing financial
the measured pressure data have been performed. At lower wind speeds interests or personal relationships that could have appeared to influence
the peak predicted moments using the DAF to approximate the inertial the work reported in this paper.
forces is shown to be suitably conservative. However, at high reduced
wind speeds where self-excited forces become significant, it is shown that Acknowledgments
the DAF approach under-predicts the peak moments.
The proposed method is expected to complement the use of aero- The authors would like to thank Soltec for their collaborative spirit
elastic modeling, and continued calibration of these techniques is rec- and challenging questions that initiated the development of the method
ommended alongside aeroelastic model testing. However, for rapid presented herein. The authors would also like to thank Dr. Stoyan
design decisions or in cases where aeroelastic modeling cannot be per- Stoyanoff for his helpful insights and comments on a draft version of this
formed due to schedule or budgetary restrictions, it is expected that the manuscript.

List of symbols

A*i Aerodynamic derivatives

B Section width or chord
cθ Torsional damping coefficient
Cx ; Cz ; Cm Lateral force, vertical force, and moment coefficients
C* ; K * Damping and stiffness matrices
f Frequency (Hz)
Fx ; Fz ; My Lateral force, vertical force, and moment in the body-aligned coordinate system
F; G – Real and imaginary parts of Theodorsen’s function
I – Mass moment of inertia
Iu ; Iv ; Iw Turbulence intensities
Ji ; Yi Bessel functions of order i of the first and second kind
K ¼ ωB=U Reduced frequency
L Length of the tracker in the lateral direction
Mb Buffeting moment
Mresist Moment the structure must resist
q Reference pressure
Sðf Þ Power spectral density function
uðtÞ; wðtÞ Instantaneous wind speeds
U Mean wind speed
u’ ðtÞ; w’ ðtÞ Fluctuation wind speeds about the mean
u’f ðtÞ; w’f ðtÞ Filtered fluctuating wind speeds about the mean
Lu , Lu Longitudinal and across-wind integral length scales for the along-wind component
x; x;_ x€ Lateral deflection, velocity, and acceleration
_ €z
z; z; Vertical deflection, velocity and acceleration
α Angle of wind attack
ζ Critical damping ratio
η1 ; η3 Parameters for Irwin’s admittance function
_ €
θ; θ; θ Structural rotation, angular velocity and angular acceleration
λ; λmass ; λmmi Scaling factors for geometry, mass and mass moment of inertia
ρ Air density
σM Standard deviation of the moment
χ Chord-wise aerodynamic admittance function
ω Circular frequency (rad/s)

15
Z.J. Taylor, M.T.L. Browne Journal of Wind Engineering & Industrial Aerodynamics 197 (2020) 104056

References Novak, D., Stoyanoff, S., Herda, H., 1995. Error assessment for wind histories generated
by autoregressive method. Struct. Saf. 17, 79–90.
Pfahl, A., Buselmeier, M., Zaschke, M., 2011. Wind loads on heliostats and photovoltaic
Abiola-Ogedengbe, A., Hangan, H., Siddiqui, K., 2015. Experimental investigation of wind
trackers of various aspect ratios. Sol. Energy 85, 2185–2201.
effects on a standalone photovoltaic (PV) module. Renew. Energy 78, 657–665.
Pratt, R.N., Kopp, G.A., 2013. Velocity measurements around low-profile, tilted, solar
Aly, A., Bitsuamlak, G., 2013. Aerodynamics of ground-mounted solar panels: test model
arrays mounted on large flat-roofs, for wall normal directions. J. Wind Eng. Ind.
scale effects. J. Wind Eng. Ind. Aerodyn. 123, 250–260.
Aerodyn. 123, 226–238.
Banks, D., Meroney, R.N., 2001. The applicability of quasi-steady theory to pressure
Rainey, A.G., 1957. Measurement of Aerodynamic Forces on Various Mean Angles of
statistics beneath roof-top vortices. J. Wind Eng. Ind. Aerodyn. 89, 569–598.
Attack on an Airfoil Oscillating in Pitch and on Two Finite-Span Wings Oscillating in
Bechtel National Inc, 1980. Wind Design of Flat Panel Photovoltaic Array Structures.
Bending with Emphasis on Damping in the Stall. Technical Report 1305. National
Bechtel National Inc., California, USA.
Advisory Committee for Aeronautics.
Bitsuamlak, G.T., Dagnew, A.K., Erwin, J., 2010. Evaluation of wind loads on solar panel
Reina, G.P., De Stefano, G., 2017. Computational evaluation of wind loads on sun-
modules using CFD. In: Proceedings of the Fifth International Symposiumon
tracking ground-mounted photovoltaic panel arrays. J. Wind Eng. Ind. Aerodyn. 170,
Computational Wind Engineering, Chapel Hill, North Carolina, USA, May 23–27.
283–293.
Davenport, A.G., 1962. Response of slender line like structures to a gusty wind. Inst. Civ.
Sarkar, P.P., Caracoglia, L., H Jr., F.L., Sato, H., Murakoshi, J., 2009. Comparative and
Eng. 23, 389–408.
sensitivity study of flutter derivatives of selected bridge deck sections, Part 1: analysis
Davenport, A.G., 1964. The buffeting of large superficial structures by atmospheric
of inter-laboratory experimental data. Eng. Struct. 31, 158–169.
turbulence. Ann. N. Y. Acad. Sci. 116, 135–159.
Shademan, M., Barron, R.M., Balachandar, R., Hangan, H., 2014. Numerical simulation of
Davenport, A.G., 1967. Gust loading factors. J. Struct. Div. 11–34. ASCE 5255.ST3.
wind loading on ground-mounted solar panels at different flow configurations. Can.
ESDU, 1985. Characteristics of Atmospheric Turbulence Near the Ground. Part II: Single
J. Civ. Eng. 41, 728–738.
Point Data for Strong Winds (Neutral Atmosphere). ESDU. Technical report 85020.
Scanlan, R.H., 1999. Estimates of skew wind speeds for bridge flutter. J. Bridge Eng. 4,
Ibrahim, S.R., Mikulcik, E.C., 1977. A method for the direct identification of vibration
95–98.
parameters from the free response. Shock Vib. Bull. 47 (Part 4), 183–189.
Scanlan, R.H., Rosenbaum, R., 1968. Aircraft Vibration and Flutter. Dover Publications,
Irwin, P.A., 1977. Wind Tunnel and Analytical Investigations of the Response of Lions’
Inc., New York, USA.
Gate Bridge to a Turbulent Wind. Technical report LTR-LA-210. National Research
Scanlan, R.H., Sabzevari, A., 1969. Experimental aerodynamic coefficients in the
Council of Canada.
analytical study of suspension bridge flutter. J. Mech. Eng. Sci. 11, 234–242.
Irwin, P.A., Cooper, K.R., Girard, R., 1979. Correction of distortion effects caused by
Scanlan, R.H., Tomko, J.J., 1971. Airfoil and bridge deck flutter derivatives. J. Eng. Mech.
tubing systems in measurements of fluctuating pressures. J. Ind. Aerodyn. 5, 93–107.
Div. ASCE 97, 1717–1737.
Irwin, P.A., Schuyler, G.D., 1977. Experiments on a Full Aeroelastic Model of Lions’ Gate
Stathopoulos, T., Zisis, I., Xypnitou, E., 2014. Local and overall wind pressure and force
Bridge in Smooth and Turbulent Flow. Technical report LTR-LA-206. National
coefficients for solar panels. J. Wind Eng. Ind. Aerodyn. 125, 195–206.
Research Council of Canada.
Stoyanoff, S., 2001. A unified approach for 3D stability and time domain response
Irwin, P.A., Stoyanoff, S., Xie, J., Hunter, M., 2005. Tacoma Narrows 50 years later - wind
analysis with application of quasi-steady theory. J. Wind Eng. Ind. Aerodyn. 89,
engineering investigations for parallel bridges. Bridge Struct. 1, 3–17.
1591–1606.
Jubayer, C.M., Hangan, H., 2014. Numerical simulation of wind effects on a stand-alone
Stoyanoff, S., Larose, G.L., 2004. Identification of aerodynamic derivatives: a parametric
ground mounted photovoltaic (PV) system. J. Wind Eng. Ind. Aerodyn. 134, 56–64.
study. In: 5th International Colloquium Bluff Body Aerodynamics and Applications.
Jubayer, C.M., Hangan, H., 2016. A numerical approach to the investigation of wind
Stoyanoff, S., Dallaire, P.-.O., 2013. A direct method for calculation of wind loads on long-
loading on an array of ground mounted solar photovoltaic (PV) panels. J. Wind Eng.
span bridges. In: 12th Americas Conference on Wind Engineering (12ACWE) Seattle,
Ind. Aerodyn. 153, 60–70.
Washington, USA, June 16-20.
Kopp, G.A., Farquhar, S., Morrison, M.J., 2012. Aerodynamic mechanisms for wind loads
Strobel, K., Banks, D., 2014. Effects of vortex shedding in arrays of long inclined flat
on tilted, roof-mounted solar arrays. J. Wind Eng. Ind. Aerodyn. 111, 40–52.
plates and ramifications for ground-mounted photovoltaic arrays. J. Wind Eng. Ind.
Larose, G.L., 1999. Experimental determination of the aerodynamic admittance of a
Aerodyn. 133, 146–149.
bridge deck segment. J. Fluids Struct. 13, 1029–1040.
Taylor, Z.J., Kopp, G.A., Gurka, R., 2010. Flow measurements regarding the timing of
Miller, R.D., Zimmerman, D.K., 1981. Wind Loads on Flat Plate Photovoltaic Array Fields.
vortices during flutter. J. Wind Eng. Ind. Aerodyn. 98, 864–871.
Boeing Engineering and Construction Company, Seattle, Washington, USA.
Theodorsen, T., 1935. General Theory of Aerodynamic Instability and the Mechanism of
Mooneghi, M.A., Irwin, P., Chowdhury, A.G., 2016. Partial turbulence simulation method
Flutter. Technical Report 496. National Advisory Committee for Aeronautics.
for predicting peak wind loads on small structures and building appurtenances.
Warsido, W.P., Bitsuamlak, G.T., Barata, J., 2014. Influence of spacing parameters on the
J. Wind Eng. Ind. Aerodyn. 157, 47–62.
wind loading of solar array. J. Fluids Struct. 48, 295–315.

16