1 s2.0 S0167610523003288 Main

Journal of Wind Engineering & Industrial Aerodynamics 244 (2024) 105626
Contents lists available at ScienceDirect
Journal of Wind Engineering & Industrial Aerodynamics

journal homepage: www.elsevier.com/locate/jweia
Aeroelastic instability mechanisms of single-axis solar trackers

Zachary J. Taylor *, Mark A. Feero , Matthew T.L. Browne
Rowan Williams Davies & Irwin Inc. (RWDI), Guelph, Ontario, Canada
A R T I C L E I N F O A B S T R A C T
Keywords: Single-axis solar trackers are currently the most commonly used racking system in utility scale solar photovoltaic
Torsional flutter (PV) plants. The last decade has seen significant advances in the understanding of single-axis tracker aero
Torsional divergence dynamics after significant failures have occurred due to aerodynamic instabilities. The current study focuses on
Single-axis solar trackers
the aeroelastic mechanisms causing these divergent instabilities. The analytical background as a function of static
aerodynamic instability
fluid-structure interaction
tilt angle is presented for the different instabilities. The role of aerodynamic stiffness and aerodynamic damping
are examined through rigid sectional model testing and these results are used to perform numerical stability
estimates using multiple theoretical approaches. These results are compared with aeroelastic model testing of a
single-axis solar tracker over a wide range of static tilt angles. These analytical, numerical, and experimental
approaches are used to assess the static tilt angles governed by stiffness-driven and damping-driven aerodynamic
instabilities. The small size and light weight of single-axis trackers makes them more susceptible to turbulent
gusts. This aspect has been addressed in the current study through the concept of the structurally averaged wind
speed and the effective gust velocity factor. Comparison to the experimental results suggests that stiffness-driven
torsional instability can respond quite rapidly to the passage of turbulent gusts.
solar PV trackers.
Much of the work on this important topic has remained relatively
1. Introduction hidden from the research community likely due to the pace of devel
opment and the non-disclosure agreements in place in both design and
With the push for the development and deployment of renewable forensic studies. Rohr et al. (2015) published limited sectional model
energy, utility-scale ground mounted solar arrays are becoming testing with CFD simulations focusing on the aeroelastic instability near
increasingly common. Over approximately the last decade, most utility- flat orientation (0◦ static tilt). Martinez-Garcia et al. (2021)and Garcia
scale arrays have opted to use single-axis trackers. These trackers consist et al. (2021) have presented the results of aeroelastic testing in relatively
of long rows of solar photovoltaic (PV) panels mounted structurally to an smooth flow conditions and proposed a non-dimensional critical wind
axis that rotates about the north-south direction to follow the sun as it speed curve as a function of tilt angle. Cárdenas-Rondón et al. (2023)
moves from east to west during the day. Designers of these tracking have evaluated the dependence on tracker height using measured
systems are consistently examining how to increase the structural effi aerodynamic derivatives. A survey of several aeroelastic model tests is
ciency of the design while ensuring that the service life of the plant presented by Enshaei et al. (2023) which shows a similar “W-shaped”
matches the desires of the stakeholders. One of the key parameters critical wind speed curve to Martinez-Garcia et al. (2021). However, the
driving the structural efficiency relates to the wind-induced vibrations of local maximum in critical wind speed near 0◦ static tilt is significantly
these systems. Most systems are quite flexible in torsion leading to the less pronounced in the results of Enshaei et al. (2023) who also show the
possibility of significant wind-induced responses, and several torsional effect of damping levels well below critical. Based on these studies and
failures have been observed due to aeroelastic instabilities over the last the results of the current study it is likely not possible to define a uni
decade. More recently, the importance of the wind engineering of these versal stability curve.
structures has become better understood by stakeholders and designers. Some novel cable-supported approaches to resolve the aeroelastic
In addition to the increased awareness, the authors have worked closely instabilities have been proposed and examined experimentally by He
with tracker designers over approximately the last decade to better et al. (2020) and Liu et al. (2023). Likewise, a parabolic solar collector
understand the aerodynamics involved. This paper summarizes the was assessed by Andre et al. (2017) through a numerical aeroelastic
primary aeroelastic mechanisms specific to ground-mounted single-axis
* Corresponding author.
E-mail address: [email protected] (Z.J. Taylor).
https://doi.org/10.1016/j.jweia.2023.105626
Received 30 June 2023; Received in revised form 23 November 2023; Accepted 1 December 2023
Available online 8 December 2023
0167-6105/© 2023 Elsevier Ltd. All rights reserved.
Z.J. Taylor et al. Journal of Wind Engineering & Industrial Aerodynamics 244 (2024) 105626
Nomenclature J joint acceptance function

kθ torsional stiffness
α wind angle of attack Lc chord length
γ12 coherence as a function of frequency between points 1 and Ls span length
2 Mse self-excited moment
ζt torsional damping ratio to critical pf peak factor
η1 , η3 non-dimensional parameters in the chordwise admittance s coordinate in the along-structure direction
θ structural rotation (θ̇, θ̈ are the angular velocities and Sii power spectral density in components i = u, v, or w
accelerations) Tref reference period for the peak factor
θs static tilt angle t time
ν cycling rate us structurally averaged wind speed
ρ air density u(t) time series of wind speed
σuav standard deviation of the structurally averaged velocity u′, w′ streamwise and across-wind turbulence fluctuations from
φ mode shape the mean, respectively
χ aerodynamic admittance u′f , w′f fluctuating wind speeds filtered by aerodynamic
χT time averaging filter admittance
A∗2 , A∗3 aerodynamic derivatives U average wind speed at torque tube elevation
cθ torsional damping constant Ur = U/(ft Lc ) reduced wind speed at tracker elevation
Cm moment coefficient xLu , rLu streamwise and across-wind streamwise integral scales,
f frequency (in Hz) respectively
ft torsional natural frequency (in Hz) z0 aerodynamic roughness length
H torque tube elevation
Iθ mass moment of inertia
assessment to demonstrate the mechanism of vortex shedding synchro part of the atmospheric boundary layer. Since the vast majority of wind
nization with torsional vibrations similar to those of single-axis trackers. engineering research is based on buildings or bridges this question may
The structural damage resulting from a divergent torsional response appear trivial. However, the high levels of turbulence near the ground
is often observed to be significant and widespread in contrast to local and the relatively small and light nature of solar PV structures makes
ized failures of clamps or other components. Valentin et al. (2022) this an important question. The authors have observed failures of
assessed the structural damage resulting from what they referred to as single-axis solar trackers, and it is clear from these cases that the
torsional galloping and confirmed the potentially catastrophic result of build-up time required between relatively small motion and divergent
divergent torsional motions. motion leading to failure is much shorter than that for a bridge. When
The goal of this study is to pair analytical formulations with wind using an analytical formulation to estimate the onset of divergent mo
tunnel measurements obtained by the authors over the last few years to tion it is important to understand the relationship between the predicted
describe the mechanisms causing aeroelastic instabilities on single-axis wind speed and the appropriate gust averaging time in the atmosphere.
solar trackers. There are many similarities between single-axis trackers A typical single-axis tracker is shown in cross-section in Fig. 1 along
and long-span bridges. Both have generally horizontal alignments, with the coordinate system used in the current study. Many parameters
relatively open exposures, torsional flexibility, and large chord-to- affect the aerodynamic stability of these structures and a full picture of
thickness ratios of the sections exposed to wind. Therefore, part of the the effect of each parameter remains elusive. The primary factors that
current study leverages some of the research carried out on long-span affect the aerodynamic stability of these structures are the static tilt
bridge aerodynamics such as the use of sectional model testing and angle, θs (i.e., the angle formed with the horizontal with no wind) and
the simulation of aeroelastic response using the modified quasi-steady the structural stiffness. These parameters will be the focus of the present
buffeting analysis (e.g., Stoyanoff, 2001; Diana et al., 2020). The work study. However, experience has shown that height/chord ratio, turbu
of Taylor and Browne (2020) focused on the aeroelastic loading of lence, length-to-chord aspect ratio of the tracker, and structural damp
single-axis trackers using similar approaches but did not address aero ing can also affect the aerodynamic stability of single-axis solar trackers.
dynamic stability. There are significant distinctions for single-axis solar Limited discussion on these effects will be presented; however, there are
trackers compared to long-span bridges: (i) the proximity to the ground, many open questions related to certain key parameters that remain
(ii) the relatively small size and mass of single-axis solar trackers, and outside the scope of the current work.
(iii) the torsional stiffness is typically lower than the vertical (bending) The next section focuses on defining the main mechanisms under
stiffness. stood to be driving the aeroelastic instabilities of single-axis solar
For long-span bridges, the wind tunnel instability speeds for flutter trackers. Section 3 presents a discussion on the critical gust duration for
or the onset speed of vortex-induced oscillations are often interpreted these structures. Due to their much smaller size and mass compared to
assuming a 10-min averaging time. Some of the reasons for this inter long-span bridges, the understanding of critical gust duration is more
pretation are described by Irwin (1998), among others, who highlighted important. Section 4 presents the sectional aerodynamics as measured
the challenges of associating wind tunnel speeds with full-scale gust through rigid sectional model and rigid pressure model testing. Section 5
averaging time when the full spectrum of turbulence cannot be simu overviews numerical and experimental methods for determining aero
lated. For the discussion that follows it is important to note that the dynamic instability. The experimental investigation of aerodynamic
actual observed build-up time for instabilities on bridges is shorter than instability is based on aeroelastic model testing in properly simulated
10 min (e.g., Owen et al., 1996). The main reason that a 10-min average atmospheric turbulence. This latter test represents the current state-of-
is used for this purpose is that it is relating a span-averaged gust speed to the-art method for design of single-axis solar trackers and forms the
a single point measurement. An important question addressed in this basis to compare the other methods.
study is how to interpret gust duration for single-axis solar trackers that
are much lighter, much smaller, and are embedded in the most turbulent
2
Fig. 1. Definition of geometry, the effective wind angle of attack α, and sign conventions.
2. Mechanisms speed at tracker height of the u (t) time series. The effective angle of
attack is sketched in Fig. 1 and is defined (for small angles) as
Even for more established fields (e.g., long-span bridges) there exists
debates on the correct terminology of the various instability phenom w′
α= + θ + θs (2)
ena. Therefore, it is not surprising that many different terms are being U + u′
used both in the industry and published articles to describe the observed
in turbulent flow, and in smooth flow conditions α = θ + θs where θ is
instabilities on single-axis solar trackers. The authors’ experience and
the structural rotation from the static position. In Eq. (2), u′ and w′ are
the results of this study suggest that the aerodynamic instabilities
along wind and across-wind turbulent fluctuations with the other terms
observed on single-axis solar trackers can be divided into two main
as defined above. The critical wind speed where the system loses stiff
types: (i) stiffness-driven, and (ii) damping-driven instabilities. In this
ness can be found in terms of torsional frequency fθ as follows
section, these primary mechanisms will be overviewed, and other ter
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
minology will be discussed that has been encountered both in practice Iθ
and in the literature. No attempt is made in the current study to ascribe Ucrit = 2π fθ 1 2 dCm . (3)
ρLc dα
specific flow features to the observed aeroelastic phenomena. Sketches 2
of vortices and their influence can occasionally be useful; however,

where Iθ is the generalized mass moment of inertia per unit length
occasionally this approach can detract from understanding the actual
measured about the centre of rotation. For aircraft wings or bridges
aeroelastic forces and their causes. Attempts have been made to explain
where relatively small rotations (e.g., < ±5◦ ) could lead to structural
flutter on long-span bridge decks through visualization of vortices (e.g.,
failure, this linear approximation is a reasonable description of the
Kubo et al., 1992; Larsen, 2000; Larsen, 2016); however, flow mea
aerodynamic stiffness-driven instability. Likewise, smooth flow wind
surements have indicated these explanations are not physically materi
tunnel observations of this aerodynamic instability are characterized by
alized in some cases of flutter (Taylor et al., 2010a). Particle image
a unidirectional twist that ultimately leads to failure. However, single-
velocimetry (or similar methods) or high-fidelity numerical simulations
axis solar trackers are subjected to strong turbulent fluctuations and
are recommended for future studies to better understand the flow pat
can tolerate much larger dynamic rotations of approximately ±20◦ . At
terns potentially causing the aeroelastic mechanisms described here.
these rotation magnitudes, the non-linearities become more important
However, these are outside the scope of the current study.
and sustained dynamic oscillations (as opposed to unidirectional twist)
have been repeatedly observed through wind tunnel testing over all
2.1. Stiffness-driven instability susceptible static tilt angles. Non-linearities in the system include both
aerodynamic and structural non-linearities. Therefore, although the
Static torsional divergence is a classical linear aerodynamic insta onset can be predicted based on static aerodynamic theory, the wind
bility with the derivation provided in many textbooks (e.g., Simiu and tunnel observations indicate a significant dynamic response can occur
Scanlan, 1996; Blevins, 1977). The derivation is briefly overviewed here due to a stiffness-driven phenomenon.
for context. The equation of motion is linearized using a quasi-steady The onset of the instability occurs when the structural torsional
approximation along with the assumption of a first-order Taylor series stiffness is overcome by the negative aerodynamic stiffness, and the
expansion wind speed at which this occurs can be reasonably predicted by Eq. (3)
( ) for static tilt angles close to 0◦ . Therefore, for tilt angles close to hori
1 1 dCm
Iθ θ̈ + cθ θ̇ + kθ θ = ρ(u(t))2 Lc2 Cm (α(t)) ≈ ρU 2 L2c Cm (0) + α (1) zontal a stiffness-driven aerodynamic instability has been observed in
2 2 dα
the wind tunnel which can be predicted through quasi-steady theory but
where α is the effective angle of attack, Lc is the section chord, Cm is a has been observed to result in large sustained dynamic motions. Due to
static moment coefficient, ρ is the air density and U is the average wind the stiffness-driven nature of the instability, moderate levels of damping
3
have been shown to have a minimal effect on the critical wind speed. critical wind speeds would be considered as 10-min mean wind speeds
The authors have also carried out many aeroelastic model tests where due to the observed build-up period of aerodynamic instabilities. Like
over-damping (i.e., damping ratio exceeding 100%) has been employed. wise, for sectional model testing of bridges in smooth or low turbulence
These tests have shown that significant levels of damping can control the wind tunnel conditions it is typically assumed that the wind tunnel
dynamic nature of this type of instability. However, as would be antic speed corresponds to a 10-min mean. However, single-axis solar trackers
ipated by quasi-steady theory, significant static rotations are observed in are both significantly smaller and lighter than a bridge. They are also
these cases. The aeroelastic response of single-axis trackers with sig built in the most turbulent part of the atmospheric boundary layer near
nificant levels of damping is outside the scope of the current study. the surface. Therefore, an important consideration is to what critical
Most of the terms used to describe this type of instability include gust duration these analytically determined critical wind speeds corre
divergence or galloping (e.g., cyclical torsional divergence (Enshaei spond. The concept of the structurally averaged peak gust speed is dis
et al., 2023) or torsional galloping (Rohr et al., 2015; Valentin et al., cussed in this section to address this question.
2022). The use of the word galloping is believed to trace back to the fact
that this instability can be predicted using quasi-steady theory which is 3.1. Structurally averaged peak gust speed
analogous to across-wind galloping (e.g., Parkinson and Smith, 1964). A
dynamic instability that cannot be predicted by quasi-steady theory is Peak gust speed predictions at a point in space have been previously
described by Blevins (1977) as “torsional galloping”. Therefore, as in the assessed considering different averaging times and different exposures
world of long-span bridges, it is inevitable that various descriptors get (Durst, 1960; Greenway, 1979; Wood, 1983). The focus in these studies
used interchangeably. For this reason, the term “stiffness-driven” is used has typically been on conversion of meteorological records to different
in this study. gust averaging times for the assessment of peak wind loads. However,
given the question of structural response posed in the current study,
2.2. Damping-driven torsional instability these approaches are now considered in the context of the so-called
structurally averaged wind speed. Similar approaches have previously
The quasi-steady formulation used to derive the critical condition for been carried out to simplify wind loading calculations (e.g., Davenport,
static torsional divergence ignores the effects of aerodynamic and 1962b; Vellozzi and Cohen, 1968); however, the intent here is focused
structural damping. As discussed in Taylor and Browne (2020), the on the interpretation of averaging times for point wind speeds consid
right-hand side of Eq. (1) can be divided into buffeting and self-excited ering aerodynamic stability. The approach described by Greenway
moment terms. The quasi-steady theory description of the self-excited (1979) can be used to compute the relationship between gust speeds of
moments does not account for terms proportional to the angular ve different averaging times provided the target power spectral density is
locity. To include these effects, the conventional linear approximation in known. Wood (1983) simplified this approach with approximate func
the so-called Scanlan convention (Scanlan and Sabzevari, 1969; Scanlan tions that have since been adopted into the ESDU methodology for gust
and Tomko, 1971) is considered for the self-excited moments wind speeds (ESDU, 2002). In this section, the approach provided by
( ) Greenway (1979) is considered based on establishing a power spectral
Mse = ρπ L4c fA∗2 θ̇+2πf 2 A∗3 θ . (4)
density target of structurally averaged turbulent fluctuations.
In the frequency domain, the power spectral density of the struc
where f is the frequency of oscillation. The terms A∗2 and A∗3 are referred
turally averaged wind speed is given by
to as aerodynamic derivatives and are determined experimentally as a
function of reduced wind speed Ur = U/(2πfLc ). The solution of the Sus (f ) = Su (f )⋅χ 2 (f )⋅J 2 (f ) (6)
equation of motion with Eq. (4) reveals the following criterion for the
system to remain stable where us is referred to as the structurally averaged wind speed, or the
( ) wind speed to which the structure will respond in a given mode of vi
1 Lc4
ρ 2πf A∗2 − ζθ 2πfθ < 0. (5) bration. The chordwise aerodynamic admittance χ 2 (f), and the joint
4 Iθ
acceptance function J2 (f) follow well-known approaches in wind engi
Therefore, when the negative aerodynamic damping exceeds the neering and are provided in Appendix A for completeness.
structural damping, the system is in a divergent state. Due to the To arrive at a peak gust velocity factor, the power spectral density of
dependence of A∗2 and A∗3 on the total system frequency, the stability of the structurally averaged wind speed is first filtered for a given aver
the system must be iterated over a range of wind speeds to determine the aging time using the time averaging filter given as
critical wind speed. For extension of this single degree-of-freedom model
sin(f πT)
to a generalized approach the reader is referred to Stoyanoff (2001). χ T (f ) = , (7)
f πT
Through forensic work, discussion with designers, and third-party
structural engineers, the authors are aware of many terms being used where the gust duration is given by the averaging time T. The peak gust
in the industry that may lead to some confusion about the underlying speed is then estimated using the expected value of the peak velocity,
mechanism of the observed damping-driven torsional instability. These which is assumed to take the form of a Gumbel distribution (Davenport,
include flutter (Martínez-García et al., 2021), vortex lock-in (Rohr et al., 1964) and given as follows
2015), or stall flutter (Cárdenas-Rondón et al., 2023). As in the case of
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
stiffness-driven torsional instability, it is not expected that consensus pf =
û
av ( ) 0.5772
= 2 ln νTref + √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ (8)
( )
will be reached in the research and practicing community on what to call σuav 2 ln νTref
this class of instability. However, the term damping-driven torsional
instability has been selected here to emphasize its observed dependence where the average cycling rate ν is computed by
on aerodynamic damping effects. This dependence will be shown
quantitatively in Section 4. ⎡
∫∞
⎤12
⎢ f Su (f )df ⎥
2
⎢ ⎥
3. Critical gust duration
av
⎢ ⎥
ν=⎢ 0
⎢ ∫∞
⎥
⎥. (9)
⎢ ⎥
Equations (3) and (5) can each be evaluated to determine a critical ⎢
⎣
⎥
Suav (f )df ⎦
wind speed. However, it is unclear to what averaging period that critical 0
wind speed corresponds. Traditionally for long-span bridges, these
4
The peak velocity gust factor is then Table 1

( ) Summary of aeroelastic model parameters.
( ) ( ) σ u Tref
gfav Tref = 1 + pf Tref ⋅ av . (10) Parameter Symbol Full-scale value
U
Torque tube elevation H 2.3 m
While Eq. (10) predicts the peak structurally averaged gust speed Chord length Lc 1.95⋅H
occurring in the reference time frame, it should be noted that this peak is Free span length Ls 7.6⋅Lc
inherently a low probability event at any instant within that reference Mass moment of inertia per unit length Iθ 112 kg m2/m
Mode shape ( xπ )
time frame. The term “gust factor” is used extensively and for a variety of φ(x)
sin
2Ls
purposes in wind engineering. The use of the term “velocity gust factor”
Torsional damping ratio to critical ζt 4–5%
in this work follows Davenport’s (1964) definition where it “defines the
ratio of the maximum observed gust speed to the concurrent mean wind
speed during some period”. The number of cycles required for the development of aerodynamic
The use of the average cycling rate for a broad-banded process like instabilities has not received a considerable amount of attention and the
turbulent velocity fluctuations is an approximation consistent with experiments required are challenging to define in terms of the use of
Davenport (1964). The validity of this approximation is evaluated based smooth flow versus turbulent flow conditions as well as steady versus
on the wind tunnel measurements in this study (Fig. 9). In this case a accelerating flows. However, for the across-wind galloping instability,
good match is observed between the empirically measured velocity gust Parkinson and Smith (1964) estimated the number of cycles for devel
factor curve and that calculated using this methodology (Greenway, opment of the instability to be approximately 300. The studies of
1979). Pigolotti et al. (2017) and Taylor et al. (2010b) on the flutter of rect
angular cylinders suggests the required number of cycles to be on the
order of 50–200. A smooth flow examination of the build-up of cycles for
3.2. Build-up period
vortex-induced oscillations of a circular cylinder showed the number
can be in excess of 1000 cycles (Dallaire, 2008) depending on Scruton
The structurally averaged wind speed represents the wind speed felt
number. Although static quasi-steady theory can predict the onset of the
in each of the dynamic modes of the structure. However, there is one
stiffness-driven instability for some tilt angles, experience with
additional consideration necessary to estimate what gust duration
single-axis solar trackers – and the measurements described later in
should be considered for aerodynamic instability. The build-up time of
Sections 4 and 5 – demonstrate that dynamic oscillations can result due
the instability is of fundamental importance. Based on the discussion in
to the stiffness-driven instability. In this case it has been observed that
the preceding section it should be clear that the build-up time for the
only a few cycles are necessary. Therefore, 10 cycles have been assumed
wind speed at a single point in space is not sufficient to define the critical
in Fig. 2 and will be shown to be reasonable for a range of tilt angles in
gust duration. For example, if the instability takes n cycles to develop,
later sections. The concept of the structurally averaged velocity gust
then the critical structurally averaged gust duration is equal to Tgust =
factor can be related back to a single point measurement as sketched in
n⋅ft where ft is the torsional natural frequency. However, the filtering
Fig. 2. This conversion is necessary because it is not practical to measure
effects of the structure imply that gfav (Tgust )∕
= gf(Tgust ) where gf(T) is the
the entire velocity field spanning the length of the tracker. In this case,
gust factor defined for a single point in space.
the time required for the instability to develop corresponds to approxi
To illustrate this point, Fig. 2 shows the peak velocity gust factor
mately 6.7 s. For this averaging time, the velocity gust factor for the
calculated for a single point in space at a typical tracker elevation of 2.3
structurally averaged wind speed is approximately 1.2. The conversion
m and a single point in space at 10 m elevation. These velocity gust
to another reference is completed by determining the averaging time
factor curves are derived based on the discussion above using the target
required to obtain the same velocity gust factor. In this case, the single
power spectral densities from ESDU (1985) with an open country
point equivalent at 10 m elevation would correspond to a 32-s gust
roughness length of z0 = 0.03 m. The structurally averaged velocity gust
duration. Another way of expressing this concept is that it is unlikely –
factor is also included in Fig. 2 where the structural parameters are those
based on spatial correlation – that a 6.7-s gust at a point in space could
from the aeroelastic model test parameters provided in Table 1.
excite the structure. Rather it is expected to take a sustained gust of
approximately 30 s to produce sufficient correlation for enough of the
structure to exceed the critical speed for long enough to allow the
instability to develop. The smaller the structure the closer these two
times become.
As noted, the assumption of 10 cycles appears to reasonably match
the results from later sections and the experience of the authors. How
ever, it is likely that the number of cycles varies with a variety of
structural parameters including torque tube elevation, structural
damping, aspect ratio, etc. A full investigation of these parameters is
outside the scope of the current study. The approach of a structurally
averaged wind speed is expected to help future research place the
concept of build-up time into better context. As discussed in the Intro
duction, build-up time of instabilities for bridges has generally been
shorter than the 10-min reference speed assumed. The corollary for
single-axis solar trackers is that the observed build-up time to divergent
oscillations does not necessarily coincide with the critical gust duration.
These are particularly important concepts when detecting wind speeds
for systems with high wind stow strategies (i.e., when the system rotates
to a more aerodynamically stable position). This type of structurally
Fig. 2. Velocity gust factors calculated for a single point at z= 2.3 m and z= 10 averaged velocity gust factor can be calibrated based on a case-by-case
m using the target power spectra density according to ESDU (1985) using basis when performing the type of aeroelastic model testing described in
z0 = 0.03 m. The structurally averaged velocity gust factor is also shown for the Section 5.
solar tracker considered in this study (see parameters in Table 1).
5
4. Sectional aerodynamics 4.1. Static moment coefficients
Most single-axis solar trackers are similar in geometry to what is The sectional model test setup is instrumented with sensitive strain
shown in Fig. 1. A notable difference among single-axis solar trackers is gauged torque shafts located at either end of the model to directly
in the configuration of the panels mounted above the torque tube. Most measure the moments acting on the panel. The torque shafts are located
single-axis trackers can be divided into two categories: 1 module-in- in-line with the axis of the torque tube. The angle of attack of the model
portrait (1P) and 2 modules-in-portrait (2P). The sketch in Fig. 1 is rotated from − 60◦ through +60◦ where the angle of attack follows the
shows a 2P system with a single solar PV panel mounted on either side of definition in Fig. 1. As shown in Fig. 4, the static aerodynamic moment
the torque tube. There are some notable differences between the ge experiences the greatest changes between approximately − 10◦ and
ometry of 1P and 2P systems. Primary among these differences for the +7.5◦ ; therefore, the resolution of angle of attack is increased through
aerodynamics are the aspect ratio (tracker length to chord) of the this range. For reference, the measured moment per unit length is
structure and the relative size of the torque tube to the chord length. normalized as a moment coefficient as
Most 1P systems have longer lengths than their 2P counterparts which
M
gives them a much larger aspect ratio. Likewise, the ratio between tor Cm = 1 . (11)
ρUH2 L2c
que tube dimension and overall chord is larger for 1P systems. Although 2
the aerodynamics are similar, the aspect ratio shown in Fig. 1 and the
where, UH is the mean wind speed at tracker elevation. As discussed, the
data underlying the current study are more analogous to 2P systems than
sectional model tests were carried out in relatively smooth flow condi
1P systems.
tions with essentially no mean shear. The results of these tests are
The sectional aerodynamics reported in this section have been ob
averaged over multiple tracker geometries and shown in Fig. 4. The rigid
tained primarily from sectional model testing of rigid models. The
pressure model testing of an array (Fig. 3b) has been carried out with
sectional model testing has been carried out with inclusion of a ground
mean shear and turbulence properties matching open country condi
plane as shown in Fig. 3a; however, the wind tunnel was configured for
tions (z0 = 0.03 m). The leading row mean moment coefficients are
relatively smooth flow conditions (Iu < 0.5%) in a relatively uniform
extracted from integration of the pressure and compared to the sectional
profile (apart from the boundary layer forming on the ground plane).
model test results in Fig. 4. Further details on the layout of pressure taps
Additional results are included based on rigid pressure model testing
and integration to obtain mean moments are provided by Browne et al.
where the leading row of an array (Fig. 3b) is expected to have similar
(2020). Similar to the sectional moment coefficients, the rigid pressure
mean loading to the sectional model test. These rigid pressure model
model coefficients are averaged over several similar geometries. The
tests are carried out in representative atmospheric boundary layer
moment coefficient is observed to be nearly identical between the
conditions.
sectional model tests (performed in relatively smooth flow conditions)
and the rigid pressure model tests (performed with a simulated atmo
spheric boundary layer) for tilt angles greater than approximately − 5◦ .
Similarities exist between the two experiments over the entire range of
tilt angles; however, the effect of turbulence and/or mean shear is more
evident for tilt angles less than − 5◦ where greater differences are
observed. A representative average of the results is also presented in
Fig. 4 and this average curve will be used in the numerical analyses
discussed in Section 5.2.
The rapid change in moment between approximately − 10◦ and
+7.5◦ shown in Fig. 4 indicates the strong potential of stiffness-driven
torsional instability based on quasi-steady theory as illustrated by Eq.
(3). Based on this theoretical approach, the shape of the moment coef
ficient curve would suggest that the single-axis solar tracker will be
Fig. 4. Moment coefficients as a function of angle of attack as measured on a

sectional model and on the leading row of a rigid pressure model test. A
Fig. 3. Photographs showing examples of (a) sectional model test, and (b) rigid representative average between rigid pressure and sectional model tests is
pressure model test in RWDI’s Peter Irwin wind tunnel. also included.
6
aerodynamically stable beyond this narrow range of tilt angles. How 5. Aeroelastic modeling
ever, as will be shown in the rest of this study, the use of quasi-steady
theory alone to anticipate aerodynamic stability is overly simplistic 5.1. Aeroelastic model wind tunnel test
and not appropriate at most tilt angles for single-axis solar trackers.
The aeroelastic model wind tunnel test remains the state-of-the-art
4.2. Aerodynamic derivatives verification for the design of single-axis solar trackers. Using light
weight balsa wood to replicate the mass moment of inertia of the panels
Aerodynamic derivatives are used to quantify the self-excited forces and spring steel to represent the torque tube, it is possible to build a
in a linear frequency domain model as originally proposed for long-span model at a scale in the range of approximately 1:20 to 1:30 with
bridges by Sabzevari and Scanlan (1968). Aerodynamic derivatives have matching dynamic similitude and shape to the full-scale structure. As
been measured on the same rigid sectional models used to measure static with most aeroelastic models, there are compromises that must be made.
moment coefficients (Fig. 3). These measurements have been taken over In the case of single-axis solar trackers it is not usually possible to
a wide range of static tilt angles using the free-vibration technique simultaneously model the torsional rigidity (GJ) with either the bending
(Stoyanoff and Larose, 2004; Poulsen et al., 1992). rigidity (EI) or elongation stiffness (EA). However, no aerodynamic in
Aerodynamic derivatives have been measured over a wide range of stabilities involving either bending or elongation have been observed in
static tilt angles and averaged over several similar tracker geometries. either wind tunnel testing or full-scale observations for these structures.
The results of this process are shown as contours of the measured Therefore, this compromise is not expected to affect the effectiveness of
aerodynamic derivatives A∗2 and A∗3 in Fig. 5. The data have been linearly the aeroelastic model for use in design. Likewise, due to the relatively
interpolated between measurement points covering the full range of small motions involved in panel-normal loading, it is expected that a
reduced velocities shown in Fig. 5 and the static tilt angles 0◦ , ±2.5◦ , suitable design methodology derived from rigid pressure models is
appropriate for these load effects (Browne et al., 2020).
±5◦ , ±7.5◦ , ±10◦ , ±15◦ , ±20◦ , ±30◦ , ±45◦ , and ±60◦ . As shown in Eq.
As alluded to above, the model used in this study was built using
(4), A2* is analogous to the aerodynamic damping and carries the balsa wood segments to represent the solar PV panels fixed at single
opposite sign to the aerodynamic damping. Thus, a positive A∗2 indicates locations along a spring steel torque tube. A photograph of the model is
negative aerodynamic damping. As discussed in Section 2, when the shown in Fig. 6 with model details presented in Table 1. Where appro
structural damping is overcome, negative aerodynamic damping leads priate in this study, the results are presented in non-dimensional format.
directly to divergent instability referred to as damping-driven torsional However, the elevation of the torque tube with respect to the atmo
instability. A∗3 represents the change in torsional stiffness and has a spheric boundary layer is important and requires a dimensional value.
quasi-steady prediction corresponding to Likewise, the mass moment of inertia per unit length is provided with
( )2 equivalent full-scale units. Based on experience with a wide range of
dCm Ur
A∗3,QS = ⋅ . (12) systems it is expected that most solar PV panels will have an area density
dα 2π
The change in sign of the slope of the moment around +7.5◦ , and the
negative slopes beyond this point, suggests that A∗3 should be zero or
negative based on quasi-steady theory. These effects would be stabiliz
ing as a negative A∗3 indicates the system has positive aerodynamic
stiffness. However, the results of the sectional model tests reveal that the
quasi-steady theory is a poor predictor of aerodynamic stiffness outside
the range of static tilt angles between − 10◦ and +7.5◦ as shown in Fig. 5
where there are no observations of negative A∗3 . However, the smallest
values at the highest reduced wind speeds have been shown for tilt
angles θs < − 40∘ and θs > + 40∘ . Therefore, for these large positive and
negative tilt angles less stiffness variation is expected with increasing
wind speed. Likewise, the most rapid change in aerodynamic stiffness is
expected close to 0◦ tilt angle.
Fig. 6. Photograph of the half-tracker single row aeroelastic model test in

RWDI’s Glenn Schuyler wind tunnel.
Fig. 5. Summary of aerodynamic derivatives as a function of reduced wind speed and static tilt angle based on free vibration sectional model tests. The plot on the
left shows the damping related A∗2 term and the plot on the right shows the stiffness related A∗3 term.
7
of between 10 and 14 kg/m2. Fig. 10. The spanwise correlation and coherence take their typical def
initions where the spatial correlation is the zero-time-lag spatial
5.1.1. Flow conditions cross-correlation along the tracker span (normal to the flow). The
As previously discussed in Section 3, the flow conditions to which the spanwise coherence is a function of spanwise distance and frequency
single-axis solar tracker is subjected can be highly influential in the with the values obtained at equivalent full-scale frequencies of 0.01 Hz,
observed aerodynamic instabilities. Most single-axis trackers are in open 0.1 Hz and 1 Hz shown in Fig. 10. In general, it is shown that the
country conditions and that is the exposure that has been considered in measured values are in good agreement with the theoretical targets for
the current study. The open country exposure targets in this study are open country turbulence at the elevation of the tracker.
derived based on ESDU (1985) using a roughness length of z0 = 0.03 m.
This roughness length is representative of open country exposure in 5.1.2. Results
building codes and standards such as Exposure C from ASCE 7 The aeroelastic model is instrumented with a sensitive encoder that
(ASCE/SEI, 2022) and Category II from Eurocode (CEN, 2005). measures rotation through a timing belt attached to a toothed gear on
The mean velocity and turbulence intensity profiles achieved in the the torque tube. Likewise, the point of fixity representing the central
wind tunnel are scaled up to full scale for comparison to the target values drive of the single-axis solar tracker is created through the use of a strain
in Fig. 7. The turbulence intensity values in the streamwise direction are gauged beam connected to the torque tube on a moment arm. Each of
observed to be slightly below the target; however, the plots of the power these instruments have been carefully calibrated statically and dynam
spectral densities at tracker elevation (Fig. 8) reveal a good overall ically prior to the testing to ensure accuracy of both the tip rotation and
match to the targets. Likewise, the measured peak gust speeds as a integrated torque measurements.
function of averaging time are summarized through the velocity gust The procedure for the wind tunnel test involves setting the wind
factor which is shown in Fig. 9 in comparison to the targets obtained speed and allowing the wind tunnel speed to stabilize. The data are then
using the approach outlined in Section 3. A fit of the von Kármán sampled for an equivalent duration of several minutes at full scale.
spectrum to the measured streamwise power spectral density is included Capturing approximately 10–20 min at full scale increases the proba
in Fig. 8 based on the measured turbulence intensity and streamwise bility of capturing intermittent behaviour that may occur as the critical
integral length scale. The analytical form of the fit von Kármán spectrum wind speed is approached. For aeroelastic model testing it is not possible
is analyzed using Greenway’s (1979) method discussed in Section 3 with to identify when the system has lost stiffness or damping; therefore, the
the resulting estimate of the peak velocity gust factor included in Fig. 9. critical speed needs to be defined based on a response or load effect. In
The estimated peak velocity gust factor from the wind tunnel mea practice, this limit should be determined based on the capacity of the
surement and the analytical estimate are shown to be in good agree system with a suitable factor of safety. However, it is conventional
ment. The target streamwise integral length scale as a ratio to tracker practice to use a 15◦ dynamic threshold defined as the speed at which
height is x Lu = 8.7H based on ESDU (1985) and the roughness length of the absolute value of the difference between the peak and mean rotation
z0 = 0.03 m. Based on an estimate from the autocorrelation of the exceeds 15◦ (i.e., single amplitude). Therefore, this threshold is used in
measured streamwise turbulence velocity fluctuations the integral scale the current study. An example of the measured aeroelastic responses and
in the wind tunnel near tracker elevation is x Lu = 6.3H. The mismatch moment coefficients for a static tilt angle of +20◦ is shown in Fig. 11. It is
between the length scales is shown in the measured power spectral evident from these tests that there is a rapid increase in response with
density being slightly shifted to the right for the streamwise component increasing wind speed starting at a reduced wind speed of approximately
(Fig. 8). 2.5–3.0. Analysis of these response curves at each tilt angle produces the
The importance of spanwise correlation and coherence characteris critical wind speed plot shown in Fig. 12. The apparent “W” shape of the
tics was introduced through the use of the joint acceptance function in curve has been observed on many different types of trackers and re
Section 3 (and Appendix A). Measurements of these values have been ported in the literature previously (Martínez-García et al., 2021; Enshaei
performed over a wide spanwise distance at tracker elevation. These et al., 2023). One clear distinction from the work of Martinez-Garcia
measurements are compared to target values based on ESDU (1986) in et al. (2021) is the magnitude of the critical wind speed at and about
Fig. 7. Vertical profile measurements in the wind tunnel with the mean flow profile (left) and the turbulence intensities (right). Tracker elevation (z= 2.3 m) is
shown in the profile plots as a dashed line. The solid line in the velocity profile plot represents a power law profile with α = 0.14. The target turbulence intensities are
based on ESDU (1985) with z0 = 0.03 m.
8
Fig. 8. Power spectral densities measured in the wind tunnel at tracker elevation (z= 2.3 m). Comparison is made to targets based on ESDU (1985) with z0 = 0.03 m.
aeroelastic model described in Section 5.1.
5.2.1. Stability analysis

A simplified unit span single-degree-of-freedom representation of a
single-axis tracker using aerodynamic derivatives to describe the self-
excited moments is given by
B4 ( ∗ )
θ̈+4πζt ft θ̇+4π2 ft2 θ = πρ ft A2 θ̇+2πft2 A∗3 θ . (13)
Iθ
A stability analysis of this single-degree-of-freedom model reveals
that the system will be divergent if the condition in Eq. (5) is not
satisfied. For real structures, a multi-degree-of-freedom approach is
more appropriate as described in detail in Stoyanoff (2001). Heeg
(2000) provides a thorough assessment of using frequency tracking
methods such as this simplified single-degree-of-freedom approach. In
that study, a cautionary note is provided that the frequency does not in
fact “disappear” as the linear stability assessment would suggest. Based
on the results of this study and the authors’ experience it is true that
oscillatory motion is still possible to occur for wind speeds beyond the
Fig. 9. Velocity gust factors measured at a single point at tracker elevation point of static divergence. It is hypothesized that the motion of the
(z= 2.3 m). Comparison is made to target values obtained from the target power trackers can become sufficiently large that the aerodynamics become
spectral density based on ESDU (1985) with z0 = 0.03 m. non-linear.
Examples at selected tilt angles of the total frequency and total
0◦ static tilt angle. The study by Martinez-Garcia et al. (2021) reported damping including both structural and aerodynamic contributions are
reduced critical wind speeds close to 5 when normalized in the same shown in Fig. 13 based on the aerodynamic derivatives from Fig. 5. It is
way as the current study. The maximum critical reduced wind speed clear from these results that the aerodynamic stiffness and damping
observed in the current study near 0◦ is 3.5. Further discussion on this behaviour varies with static tilt angle. In these examples, both − 60◦ and
point is provided in Section 6.1 below. +45◦ generally show increasing torsional frequency with less dramatic
changes in the aerodynamic damping. The rapid loss of stiffness at − 7.5◦
and 0◦ is also evident from the examples in Fig. 13. The damping ratio is
5.2. Numerical stability analysis calculated based on the analytically calculated torsional frequency at
each critical wind speed. Therefore, the rapid increase in damping for
Many single-axis trackers are driven by a torque tube with a centrally the tilt angles closer to zero is partially due to the corresponding
located drive. In this study a basic architecture is assumed with the decrease in the torsional frequency.
properties shown in Table 1. The numerical model matches the
Fig. 10. Spanwise correlation and coherence characteristics at tracker elevation (z= 2.3 m). The solid lines are targets based on ESDU (1986) with z0 = 0.03 m and
the markers represent measured values.
9
a) b)
Fig. 11. Example of aeroelastic model results at +20 including (a) rotations and (b) moment coefficients. Note that the static tilt angle is zeroed at the beginning of
◦
the test, so the presented rotations are differences to the static tilt angle of +20◦ . In the legend QS refers to quasi-steady and AD to aerodynamic derivatives.
Fig. 12. Results of numerical stability analyses and aeroelastic model tests without any gust correction (left), and (right) with gust correction applied to the stability
analysis and the buffeting analysis results with aerodynamic derivatives. The aeroelastic test results and the quasi-steady simulation results are the same in both plots
with no corrections applied.
Fig. 13. Total frequency and total damping variation with critical reduced wind speed at selected static tilt angles. Total implies the inclusion of both structural and
aerodynamic components.
Critical wind speeds based on the linear stability theory are deter wind gusts than the stiffness-driven torsional instability occurring be
mined from data such as those shown in Fig. 13 over a range of static tilt tween − 10◦ and +7.5◦ .
angles using the aerodynamic derivatives based on Fig. 5. The results of
this analysis are presented in Fig. 12. In the range of static tilt angles 5.2.2. Quasi-steady buffeting analysis
between − 10◦ and +7.5◦ it is observed that the stability theory The classical quasi-steady approach theoretically captures the self-
marginally overpredicts the results of the aeroelastic testing. Based on excited aerodynamic stiffness contribution but does not have any
the discussion in Section 3, the velocity gust factor from Fig. 2 is applied terms providing aerodynamic damping. For the purely quasi-steady
to the predicted wind speeds as shown on the right-hand plot of Fig. 12. simulations the applied aerodynamic moment is described by
After application of the velocity gust factor, the critical speeds from the
1
linear stability theory are all lower than or matching the results of the M(t) = ρ(u(t))2 L2c Cm (α(t)), (14)
2
aeroelastic test. The assumption of 10 cycles of build-up time is expected
to be conservative based on this comparison. For tilt angles outside of w′f (t)
the range of − 10◦ to +7.5◦ no reduction was applied implying the α(t) = + θ(t) + θs . (15)
U + u′f (t)
aeroelastic phenomena over these static tilt angles responds slower to
10
Eq. (15) is evaluated at each time step of the simulation and the reduced wind speeds between approximately − 45◦ to − 10◦ and +7.5◦ to
subscript “f” on the velocity refers to the filtered velocity using a chord- +45◦ . Therefore, in these regions of static tilt angles, the effects of
wise aerodynamic admittance (see Appendix A.1). The instantaneous negative aerodynamic damping are important parts of the instability
angle of attack (Eq. (15)) is used to interpolate the moment coefficient mechanism.
curve in Fig. 4 at each time step. The variable u(t) is the along-wind time The observation that the quasi-steady theory can predict the onset of
series of wind speed at a location in space as defined for Eq. (1). The the instability around 0◦ is further evidence of the stiffness-driven na
simulation at each wind speed was carried out for 1 h with a time step of ture of the instability over this narrow range of static tilt angles. The
0.025 s. The solution was carried out using a custom implicit (iterative) classical static divergence leads to a loss of stiffness which can lead to an
solver. The numerical structural model was comprised of 16 nodes along abrupt onset of the instability due to the passage of turbulent gusts.
the tracker and a turbulent time series of wind was generated for each Fig. 14 shows the measured moment (in coefficient form) from the
node. The turbulent flow field was generated using a method based on aeroelastic model at static tilt angles of 0◦ and − 20◦ at a wind speed just
the approach of Shinozuka and Jan (1972) with power spectral targets before the critical wind speed. The nature of the dynamic moment is
based on ESDU (1985) and cross-spectral targets based on ESDU (1986). observed to be significantly different at 0◦ compared to − 20◦ . Similar
An example of the mean and peak responses and integrated moments visual observations to the recorded data can be witnessed in the wind
extracted over a range of wind speeds from this buffeting analysis is tunnel test. Fig. 15 shows the same data as Fig. 14 but at the wind speed
shown for a static tilt angle of +20◦ in Fig. 11. At this tilt angle, the just beyond the identification of divergence. Dynamic oscillations can be
purely quasi-steady approach does not appropriately capture the aero observed at 0◦ static tilt when the wind speed is pushed even higher as
elastic forces. This shortcoming is further reinforced by interpolating the shown in Fig. 16.
wind speed at which the response exceeds a dynamic single amplitude Prior to the identification of divergence, the tracker apparently re
response of 15◦ (mean subtracted). This procedure is carried out at each sponds to the passage of gusts (Section 3) without a distinct oscillatory
static tilt angle and plotted in Fig. 12. Where the instability is purely response for a static tilt angle of 0◦ . In contrast, the time series at − 20◦
static divergence the quasi-steady approach is shown to match the ex prior to instability (Fig. 14) shows an occasional oscillatory moment
periments reasonably well. However, for static tilt angles less than − 5◦ near the natural frequency of the tracker due to the dynamic response of
and greater than +2.5◦ the results show a poor match to the experi the tracker modulated by the turbulence present in the free-stream.
ments. Therefore, it is concluded that the response of the structure must Once divergence has been identified (Fig. 15), there remains no strong
be influenced by more than quasi-steady aerodynamic stiffness. frequency near the first natural frequency in the moment measured at
0◦ static tilt even though there are strong intermittent dynamic moments
5.2.3. Buffeting analysis with aerodynamic derivatives exceeding the pre-defined threshold. This observation is in contrast to
The use of aerodynamic derivatives in wind engineering applications the strong dynamic oscillatory moment observed at − 20◦ . These ob
was initially introduced to overcome the shortcomings of the purely servations suggest distinctly different mechanisms as alluded to above.
quasi-steady approach described in the previous section. The aero As the wind speed is pushed higher and dynamic oscillations are
dynamic derivative approach to buffeting analysis is similar to the nu observed (Fig. 16), the intermittent divergent motion is shown to occur
merical stability analysis performed in 5.2.1 in that the wind speed is near the natural frequency of the structure with a broad-banded peak
iterated from a low wind speed up to the speed of interest to assess the observed in the power spectra.
total stiffness (structural + aerodynamic) and the total damping Based on the results of the current study the aeroelastic mechanisms
(structural + aerodynamic). The approach considered in the current are divided into three primary categories.
study is similar to that described by Taylor and Browne (2020).
The same numerical model and flow fields described in the previous • − 10◦ ≤ θs ≤ 7.5◦ : Stiffness-driven torsional instability
section were used in this analysis and the mean and peak responses and • − 45◦ ≤ θs < − 10◦ and 7.5◦ < θs < 45◦ : Damping-driven torsional
integrated moments are shown in Fig. 11. Unlike the case of quasi-steady instability
analysis, it is observed that the responses begin to increase more rapidly • Other tilt angles: aerodynamically stable
as the wind speed approaches the critical wind speed with the latter
indicated by dashed lines. This approach cannot be used beyond the In the current study there is minimal test data between 30◦ and 45◦ .
point at which the system is divergent due to the numerical procedure Future studies would be recommended to refine the boundary between
adopted. The same static tilt angles considered in the quasi-steady buf these locations. The sectional aerodynamics extracted using aero
feting analysis were considered with aerodynamic derivatives and the dynamic derivatives indicate that there is still a strong influence of
same 15◦ dynamic rotation criterion was used to define the critical wind stiffness variation in the range of tilt angles governed by damping-driven
speeds which are added to Fig. 12. The comparison between these re torsional instability. However, there are several key observations in the
sults and the aeroelastic model test critical speeds is much closer than in current study that lead to the demarcation of these ranges. The first
the case of the quasi-steady analysis. A significant point of difference is observation relates to the static moment coefficient curve which governs
observed at − 60◦ ; however, in this case the critical wind speed is classical static divergence and shows the strongest slope between
observed to be identified due to large buffeting-induced rotations rather approximately − 10◦ and +7.5◦ . Likewise, the aerodynamic derivatives
than divergent instability. show marked differences outside of this tilt angle range in both the A∗2
and A∗3 terms. The time series shown in Figs. 14, Figure 15, and Fig. 16
6. Discussion also indicate a significant difference in dynamic behaviour between the
two regimes. Another point of distinction appears to be the way in which
6.1. Tilt angle ranges of stability mechanisms the structure responds to gusts. As discussed in Section 3, a relatively
small structure such as a single-axis tracker can respond quickly to
The results of the aeroelastic wind tunnel test and numerical simu turbulent gusts. The response time is modulated by the filtering effects of
lations help to define the aerodynamic stability mechanisms for each the structure and the build-up time of the aeroelastic phenomenon. The
static tilt angle. The static moment coefficients (Fig. 4) and aerodynamic response to gusts has been observed to be faster in the range of tilts
derivatives (Fig. 5) as a function of static tilt angle demonstrate different governed by stiffness-driven torsional instability. An empirical align
aeroelastic regimes. The region of steepest gradient in the moment co ment between numerical predictions and aeroelastic model results
efficient between approximately − 10◦ and +7.5◦ coincides with a region suggests the build-up time may be on the order of 10 cycles for this
of greater A∗3 (as expected from quasi-steady theory) and negative A∗2 . stiffness-driven regime. The good agreement beyond this narrow range
The contours of A∗2 show a transition to strongly positive values at lower of angles where no gust corrections have been applied suggests more
11
Fig. 14. Moment coefficients for static tilt angles of θs = 0∘ and θs = − 20∘ at a reduced wind speed (indicated in the axis labels) just prior to the critical wind speed.
Horizontal lines indicate the moment coefficient corresponding to 15◦ dynamic rotation at ft .
Fig. 15. Moment coefficients for static tilt angles of θs = 0∘ and θs = − 20∘ at a reduced wind speed (indicated in the axis labels) just after the critical wind speed.
Horizontal lines indicate the moment coefficient corresponding to 15◦ dynamic rotation at ft .
Fig. 16. Moment coefficients for static tilt angles of θs = 0∘ and θs = − 20∘ at a reduced wind speed (indicated in the axis labels) beyond the critical wind speed in
divergent motion. Horizontal lines indicate the moment coefficient corresponding to 15◦ dynamic rotation at ft .
cycles are required in the build-up time for damping-driven torsional 6.2. Influence of parameters outside the current study
instability. The gust response is also a likely explanation for some of the
significant difference observed in reduced critical speed between the The results of the current study should not be taken as universal.
work of Martinez-Garcia et al. (2021) carried out in relatively smooth Based on the authors’ experience, there are many parameters that may
flow and the current study. As observed in Fig. 15, when the aeroelastic affect the results of the current study. The parameter focused on in the
behaviour is governed by stiffness-driven torsional instability, the dy current study is the static tilt angle. However, the variation of the critical
namic moment can respond quickly to the passage of turbulent gusts wind speed with static tilt angle can be altered based on many
without much resistance or oscillation. This type of response is also parameters.
strongly evident visually when observing the experiment. The aspect ratio of single-axis trackers is known to play a significant
The classification of different aeroelastic mechanisms with static tilt role in determining the reduced critical wind speed. In this context, the
angle is expected to apply for systems similar to those described in this aspect ratio is defined as the tracker length (between the point of fixity
study. However, as discussed in the following section, the results of the and the free end) divided by the chord length. It is typical that 1P ar
current study cannot be extended universally to any single-axis solar chitectures have a much larger aspect ratio compared to 2P architec
tracker as there are many factors that influence the aerodynamic sta tures. Although the current study is abstracted from any actual tracker
bility and fall outside the scope of the current study. design, it has an aspect ratio of approximately 8 and therefore most
closely resembles a 2P system. Based on similar testing of 1P systems it is
12
observed that the reduced critical wind speed is higher. This observation damping, and differences in the approaching turbulence to understand
suggests that the normalization by chord length does not fully capture their influence on the observations made in this study.
the aeroelastic behaviour. Therefore, the results in this study and others The concept of a structurally averaged gust speed has been intro
that use the reduced wind speed (e.g., Martínez-García et al., 2021) are duced which is a significant consideration for small and light structures
not necessarily appropriate for systems of different aspect ratios. compared to the historical approaches applied to buildings and long-
This study focused on a system with relatively low structural span bridges. Based on comparisons between analytical and experi
damping. Although in typical wind engineering applications a damping mental results, the build-up time of the instability was shown to be
level of 4–5% is relatively high, this range is on the low end of what is shortest for stiffness-driven torsional instability.
typically observed in the first torsion mode of single-axis trackers based In addition to some of the effects mentioned above, a significant
on field testing. The damping in this study is created primarily through effect not included in the current study is the aerodynamic interaction
friction in the bushings at the top of the support posts which is the same within the array. The upwind rows of the array can cause beneficial
way this level of damping is achieved in full-scale trackers. Many tracker sheltering of downwind rows and can lead to increased turbulent fluc
manufacturers now employ the use of supplementary damping devices tuations in their wake. These effects are complex and depend on even
such as viscous dampers. The results of Enshaei et al. (2023) show the more parameters than those already discussed (e.g., ground cover ratio,
effects of torsional damping up to 30% of critical. However, the results height above grade, etc.). For a complete design of a utility scale solar PV
can vary substantially as critical damping levels are approached and array of single-axis trackers it is necessary to understand the aero
exceeded. Although these effects are complex and outside the scope of dynamic stability of the interior array as this interior portion makes up
the current study, it is possible to suppress both the stiffness- and the largest percentage of the entire array. It is recommended that future
damping-driven aerodynamic instabilities with very high damping studies of a similar nature to the current study be carried out to study
levels (e.g., exceeding critical). these effects.
As discussed in Section 3, these relatively small and light structures
are sensitive to the nature of the turbulence to which they are exposed. CRediT authorship contribution statement
As a result, the aerodynamic stability and wind loading is known to be
affected by surface roughness, topographic effects, skewed winds, and Zachary J. Taylor: Conceptualization, Data curation, Formal anal
the presence of upwind trackers. These latter two points are especially ysis, Methodology, Supervision, Validation, Visualization, Writing –
important when interpreting the results of the current study in the original draft, Writing – review & editing. Mark A. Feero: Conceptu
interior of an array. Although skewed winds do not tend to produce the alization, Data curation, Formal analysis, Methodology, Validation,
lowest critical wind speeds for the windward row of an array, it has been Writing – review & editing. Matthew T.L. Browne: Conceptualization,
shown in many multi-row aeroelastic studies that interior rows may Data curation, Formal analysis, Methodology, Supervision, Writing –
experience their lowest critical wind speeds for skewed winds where the review & editing.
beneficial sheltering effects of upwind rows has been altered.
Declaration of competing interest
7. Conclusions
The authors declare that they have no known competing financial
Single-axis solar PV trackers are now used almost universally in large interests or personal relationships that could have appeared to influence
scale utility deployments of solar PV power generation plants. The in the work reported in this paper.
crease in efficiency from being able to track the sun is worth the extra
expense of additional racking equipment to support the panels and allow Data availability
for the components powering the rotation. However, it is important to
ensure reliable wind design of these structures to warrant this additional Portions of the data may be made available upon reasonable request
investment. The focus of the current study is on the different aeroelastic or in the case of collaboration.
mechanisms driving the observation of divergent instabilities in these
single-axis trackers. Analytical, numerical, and experimental ap Acknowledgments
proaches have been employed to better understand the phenomena
driving these damaging divergent rotational responses. The authors are indebted to the skilled model builders at RWDI led
Based on these various approaches, the range of tilt angles for which by Sean Vyles who built and instrumented the vast majority of the
stiffness-driven torsional instability and damping-driven torsional models used in this study and those for our commercial clients. For the
instability have been defined. As discussed, the non-dimensional critical many wind tunnel tests carried out to obtain the results in this study, the
reduced wind speed obtained in this study should be expected to apply authors are also grateful to the wind tunnel technicians at RWDI. The
for similar single-axis solar trackers. However, it was also discussed that authors have had the privilege to work with many of the leading com
many factors not included in the current study can affect the results so panies and structural engineers designing single-axis tracking systems
the observations should not be considered universal. It is recommended and have together learned a lot along the way. The current study would
that future investigations are carried out on effects such as aspect ratio, not have been possible without these fruitful collaborations.
A. Appendix
1.1Chord-wise aerodynamic admittance
The aerodynamic admittance function can be physically described in several ways. In the aeronautical world it is often described as how a wing
responds to a sudden gust of wind. However, in the world of civil structures where structures in the atmospheric boundary layer (ABL) are constantly
subjected to gusts of different sizes and durations, the aerodynamic admittance function describes how the presence of the body acts to average the
turbulent flow in the streamwise direction. Following from this description, the aerodynamic admittance acts like a moving-window average or low
pass filter. The chordwise (streamwise) aerodynamic admittance used here is given by Irwin (1977).
13
4
χ 2 (f ) = (η1 − 1 + e− η1
)(η3 − 1 + e− η3
) (16)
(η1 η3 )2
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
( )2̅
L f ⋅xLu
η1 = 0.95 c 1 + 70.78 (17)
xLu U
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
( )2̅
D 2f ⋅rLu
η3 = 0.475 1 + 70.78 (18)
rLu U
In Eq. (16), f is the frequency (in Hz), xLu is the integral length scale of the u-component in the x (along-wind) direction, U is the average along-wind
velocity, rLu is the across-wind length scale of the u-component of velocity, Lc is the section width and D is the section depth. It should be noted that
many different forms of the chordwise admittance have been proposed and are all broadly similar to the form shown in Eq. (16). The methods
described in this study are general in nature and the specific use of this form of the aerodynamic admittance does not have an appreciable effect on the
results.
The admittance-filtered time series, uf (t), at any point s on the structure can then be calculated in the frequency domain as
⎛ ⎞
∫∞ ∫∞
uf (s, t) = ⎝χ (f ) ⋅ u(s, t)e− i2π ft
dt⎠ei2πft df . (19)
− ∞ − ∞
While the admittance is generally written in the frequency domain following Davenport (1962a), some research has indicated that a suitable time
domain approach may be derived directly (Chen et al., 2000; Scanlan, 2001). However, currently these time domain methods rely on results obtained
in the wind tunnel based on frequency domain approaches. Moreover, the requirement of a convection velocity when converting between temporal
and spatial domains remains a challenge in applying the aerodynamic admittance to potentially non-stationary winds. In the present study the
chordwise admittance is applied exclusively in the frequency domain.
1.2. Joint acceptance
The joint acceptance function relates the effect of spatial filtering in spanwise directions to the mode shapes of the structure (Davenport, 1962a;
Larose, 1997). Physically, this filter quantifies the observations that wind speeds are not fully correlated when measured over the span of a structure.
Taking the form of the joint acceptance suggested by (Davenport, 1962b) gives
∫ Ls ∫ Ls
1
|J(f )|2 = ⎛ ⎞ 2
γ12 (s1 , s2 , f )φ(s1 )φ(s2 )ds1 ds2 . (20)
∫Ls 0 0
⎝ φ (s)ds⎠
2
Where φk is the mode shape in the kth degree of freedom, s1 and s2 are distances along the structure, L is the length of the structure, f is the frequency
(in Hz) and γ is a spatial coherence function. In Eq. (20) the spatial coherence function γ can be approximated using ESDU (1986) or by simpler
approximations (Chen et al., 2000; Davenport, 1962a; Larose, 1997). In the time domain this filter takes the following form
∫Ls
uf (s, t)φ2 (s)ds
us (t) = 0 (21)
∫Ls
2
φ (s)ds
0
with the spatially distributed wind speed u(s, t) already accounting for the inherent or target spatial coherence. Likewise, the aerodynamic admittance
is applied at each point in space to obtain the filtered spatial wind speed uf (s, t).
References CEN, (Comité Européen de Normalisation), 2005. Eurocode 1: Actions on Structures –

Part 1-4: General Actions – Wind Actions. EN 1991-1-4:2005+A1.
Chen, X., Matsumoto, M., Kareem, A., 2000. Time domain flutter and buffeting response
Andre, M., Pйntek, M., Bletzinger, K.-U., Wüchner, R., 2017. Aeroelastic simulation of
analysis of bridges. J. Eng. Mech. 126 (1), 7–16.
the wind-excited torsional vibration of a parabolic trough solar collector. J. Wind
Dallaire, P.-O., 2008. Étude des vibrations éoliennes instationnaires. Université de
Eng. Ind. Aerod. 165, 67–78.
Sherbrooke. M.A.Sc. thesis.
ASCE/SEI, 2022. Minimum Design Loads and Associated Criteria for Buildings and Other
Davenport, A.G., 1962a. Buffeting of a suspension bridge by storm winds. J. Struct. Div.
Structures. ASCE Standard ASCE/SEI 7-22.
88 (3), 233–268.
Blevins, R.D., 1977. Flow-induced Vibration. Robert E. Krieger Publishing Company,
Davenport, A.G., 1962b. The response of slender, line-like structures to a gusty wind.
Inc., Florida, USA.
Proc. Inst. Civ. Eng. 23 (3), 389–408.
Browne, M.T., Taylor, Z.J., Li, S., Gamble, S., 2020. A wind load design method for
Davenport, A.G., 1964. Note on the distribution of the largest value of a random function
ground-mounted multi-row solar arrays based on a compilation of wind tunnel
with application to gust loading. Proc. Inst. Civ. Eng. 28 (2), 187–196.
experiments. J. Wind Eng. Ind. Aerod. 205 (July), 104294.
Diana, G., Stoyanoff, S., Aas-Jakobsen, K., Allsop, A., Andersen, M., Argentini, T., Cid
Cárdenas-Rondón, J.A., Ogueta-Gutiérrez, M., Franchini, S., Manzanares-Bercial, R.,
Montoya, M., Hernández, S., Jurado, J.Á., Katsuchi, H., Kavrakov, I., Kim, H.K.,
2023. Stability analysis of two-dimensional flat solar trackers using aerodynamic
Larose, G., Larsen, A., Morgenthal, G., Øiseth, O., Omarini, S., Rocchi, D.,
derivatives at different heights above ground. J. Wind Eng. Ind. Aerod. 243, 105606.
Svendsen, M., Wu, T., 2020. IABSE Task Group 3.1 Benchmark results. Part 1:
14
numerical analysis of a two-degree-of-freedom bridge deck section based on Owen, J.S., Vann, A.M., Davies, J.P., Blakeborough, A., 1996. The prototype testing of
analytical aerodynamics. Struct. Eng. Int. 30 (3), 401–410. Kessock Bridge: response to vortex shedding. J. Wind Eng. Ind. Aerod. 60 (1–3),
Durst, C.S., 1960. Wind speeds over short periods of time. Meteorol. Mag. 89 (1056), 91–108.
181–186. Parkinson, G.V., Smith, J.D., 1964. The square prism as an aeroelastic non-linear
Enshaei, P., Chowdhury, J., Sauder, H., Banks, D., 2023. Wind Tunnel Testing of oscillator. Q. J. Mech. Appl. Math. 17 (2), 225–239.
Torsional Instability in Single-axis Solar Trackers: Summary of Methodologies and Pigolotti, L., Mannini, C., Bartoli, G., 2017. Experimental study on the flutter-induced
Results. 21st Australasian Wind Engineering Society Workshop. motion of two-degree-of-freedom plates. J. Fluid Struct. 75, 77–98.
ESDU, 1985. Characteristics of Atmospheric Turbulence Near the Ground. Part II. Single Poulsen, N., Damsgaard, A., Reinhold, T., 1992. Determination of flutter derivatives for
Point Data for Strong Winds (Neutral Atmosphere). ESDU Technical report 85020. the Great belt bridge. J. Wind Eng. Ind. Aerod. 41 (1–3), 153–164.
ESDU, 1986. Characteristics of Atmospheric Turbulence Near the Ground. Part III. Rohr, C., Bourke, P.A., Banks, D., 2015. Torsional Instability of Single-Axis Solar
Variations in Space and Time for Strong Winds (Neutral Atmosphere). ESDU Tracking Systems. In: 14th International Conference on Wind Engineering, pp. 1–7.
Technical report 86010. Sabzevari, A., Scanlan, R.H., 1968. Aerodynamic instability of suspension bridges. J. Eng.
ESDU, 2002. Strong Winds in the Atmospheric Boundary Layer. Part 2: Discrete Gust Mech. Div., ASCE 94, 489–519.
Speeds. ESDU Technical report 83045 With Amendments A to C. Scanlan, R.H., 2001. Reexamination of sectional aerodynamic force functions for bridges.
García, E.M., Marigorta, E.B., Gayo, J.P., Manso, A.N., 2021. Experimental determination J. Wind Eng. Ind. Aerod. 89 (14–15), 1257–1266.
of the resistance of a single - axis solar tracker to torsional galloping. Struct. Eng. Scanlan, R.H., Sabzevari, A., 1969. Experimental aerodynamic coefficients in the
Mech. 78 (5), 519–528. analytical study of suspension bridge flutter. J. Mech. Eng. Sci. 11 (3), 234–242.
Greenway, M.E., 1979. An analytical approach to wind velocity gust factors. J. Wind Eng. Scanlan, R.H., Tomko, J.J., 1971. Airfoil and bridge deck flutter derivatives. ASCE J Eng
Ind. Aerod. 5 (1–2), 61–91. Mech Div 97, 1717–1737.
He, X.H., Ding, H., Jing, H.Q., Zhang, F., Wu, X.P., Weng, X.J., 2020. Wind-induced Shinozuka, M., Jan, C.M., 1972. Digital simulation of random processes and its
vibration and its suppression of photovoltaic modules supported by suspension applications. J. Sound Vib. 25 (1), 111–128.
cables. J. Wind Eng. Ind. Aerod. 206 (June), 104275. Simiu, E., Scanlan, R.H., 1996. Wind Effects on Structures: Fundamentals and
Heeg, J., 2000. Dynamic Investigation of Static Divergence: Analysis and Testing, Applications to Design, third ed. John Wiley & Sons, Inc., New York, USA.
November. NASA Technical Report NASA/TP-2000-210310. Stoyanoff, S., 2001. A unified approach for 3D stability and time domain response
Irwin, H.P.A., 1977. Wind Tunnel and Analytical Investigations of the Response of Lions’ analysis with application of quasi-steady theory. J. Wind Eng. Ind. Aerod. 89
Gate Bridge to a Turbulent Wind. National Research Council of Canada. Technical (14–15), 1591–1606.
Report LTR-LA-210. Stoyanoff, S., Larose, G.L., 2004. Identification of Aerodynamic Derivatives: A
Irwin, P.A., 1998. The role of wind tunnel modelling in the prediction of wind effects on Parametric Study. In: 5th International Colloquium Bluff Body Aerodynamics and
bridges. In: Proc. Of the International Symposium on Advances in Bridge Applications.
Aerodynamics, pp. 99–117. Taylor, Z.J., Browne, M.T., 2020. Hybrid pressure integration and buffeting analysis for
Kubo, Y., Hirata, K., Mikawa, K., 1992. Mechanism of aerodynamic vibrations of shallow multi-row wind loading in an array of single-axis trackers. J. Wind Eng. Ind. Aerod.
bridge girder sections. J. Wind Eng. Ind. Aerod. 42 (1–3), 1297–1308. 197, 104056.
Larose, G.L., 1997. The Dynamic Action of Gusty Winds on Long-Span Bridges. Technical Taylor, Z.J., Kopp, G.A., Gurka, R., 2010a. Flow measurements regarding the timing of
University of Denmark. PhD Thesis. vortices during flutter. J. Wind Eng. Ind. Aerod. 98 (12), 864–871.
Larsen, A., 2000. Aerodynamics of the Tacoma narrows bridge - 60 years later. Struct. Taylor, Z.J., Gurka, R., Kopp, G.A., Liberzon, A., 2010b. Long-duration time-resolved PIV
Eng. Int. 10 (4), 243–248. to study unsteady aerodynamics. IEEE Trans. Instrum. Meas. 59 (12), 3262–3269.
Larsen, A., 2016. A generic model for the A2 instability of bluff bridge decks. In: 8th Valentin, D., Valero, C., Egusquiza, M., Presas, A., 2022. Failure investigation of a solar
Inernational Colloquium on Bluff Body Aerodynamics and Applications. tracker due to wind-induced torsional galloping. Eng. Fail. Anal. 135, 106137.
Liu, J., Li, S., Luo, J., Chen, Z., 2023. Experimental study on critical wind velocity of a Vellozzi, J., Cohen, E., 1968. Gust Response Factors. J. Struct. Div., ASCE 94 (6),
33-meter-span flexible photovoltaic support structure and its mitigation. J. Wind 1295–1313.
Eng. Ind. Aerod. 236, 105355. Wood, C.J., 1983. A simplified calculation method for gust factors. J. Wind Eng. Ind.
Martínez-García, E., Blanco-Marigorta, E., Parrondo Gayo, J., Navarro-Manso, A., 2021. Aerod. 12 (3), 385–387.
Influence of inertia and aspect ratio on the torsional galloping of single-axis solar
trackers. Eng. Struct. 243, 112682.
15

1 s2.0 S0167610523003288 Main

Uploaded by

Copyright:

Available Formats

1 s2.0 S0167610523003288 Main

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

1 s2.0 S0167610523003288 Main

Uploaded by

Copyright:

Available Formats

Journal of Wind Engineering & Industrial Aerodynamics 244 (2024) 105626

Contents lists available at ScienceDirect

Journal of Wind Engineering & Industrial Aerodynamics

Aeroelastic instability mechanisms of single-axis solar trackers

Nomenclature J joint acceptance function

of vortices and their influence can occasionally be useful; however,

The peak velocity gust factor is then Table 1

4. Sectional aerodynamics 4.1. Static moment coefficients

Fig. 4. Moment coefficients as a function of angle of attack as measured on a

Fig. 6. Photograph of the half-tracker single row aeroelastic model test in

aeroelastic model described in Section 5.1.

5.2.1. Stability analysis

1.1Chord-wise aerodynamic admittance

1.2. Joint acceptance

References CEN, (Comité Européen de Normalisation), 2005. Eurocode 1: Actions on Structures –

You might also like