Engineering Structures 284 (2023) 115942

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: www.elsevier.com/locate/engstruct

Transmission line failure propagation under extreme wind scenarios:

extratropical cyclones
Fabio Alminhana a,b ,∗, Matthew Mason b , Faris Albermani a
a Albermani & Alminhana Consulting, Mt Ommaney, Brisbane, Queensland 4072, Australia
b
School of Civil Engineering, The University of Queensland, Brisbane, Queensland 4072, Australia

ARTICLE INFO ABSTRACT

Keywords: Catastrophic transmission line cascades can occur during severe storm scenarios and pose a serious threat to
Cascades the power industry. This article describes an event-based approach devised to numerically simulate synoptic
Transmission lines wind fields caused by extratropical cyclones impacting a line segment. Correlated time-dependent wind speeds
Steel lattice towers
are produced at selected points along the line section according to the wind field characteristics. The accuracy
Nonlinear analysis
of the wind field produced by the method is verified by comparing the statistical properties of simulated wind
Time-domain dynamic analysis
Wind loads
records with respective theoretical values. The wind velocity time histories are converted into wind loads
Extratropical cyclone and used in a special-purpose nonlinear dynamic analysis technique, which models multi-span line sections
under progressive failure. A vectorial version of the procedure provided by the standard IEC 60826 is used
to generate wind loads based on the simulated speeds. A case study investigating the response of an existing
230 kV single-circuit line section to two extratropical cyclone wind fields with different intensities is presented.

1. Introduction able to find the ‘‘weakest link’’ within a line network, which can be ei-
ther a support tower or a (fatigued) conductor, with the failure of these
Extreme wind events have considerably impacted power grid per- individual components triggering a line cascade. A collapsed tower will
formances worldwide causing an undesirable number of transmission induce a transverse cascade, whereas a severed conductor will lead to
failures [1]. The presence of strong winds is the common scenario for a longitudinal one. It is noted, however, that cascades are considered
triggering line cascades, which are progressive failures that can affect the most common type of accident involving the collapse of multiple
transmission line segments with catastrophic consequences, resulting in transmission line structures, but not all multi-support accidents are
prolonged power outages with costly repercussions. progressive failures. The independent collapse of consecutive supports
Extratropical cyclones, also known as extended pressure systems is known as multi-failure events [5].
(EPS), are synoptic winds with circulatory air movements around cen- The response of conventional transmission towers to extratropical
tres of low pressure, resulting from the interaction of large scale air cyclone winds is a well explored subject, used as the cornerstone for
masses. They are responsible for the majority of the strongest wind wind load calculations by the line design codes [6–8]. In the last
gusts in mid-latitudes, around 40° to 60°, and account for about 4 out two decades, however, the wind engineering community has devoted
of 5 observations of the maximum annual wind speed in meridional their efforts to investigate a much less explored topic: the impact of
Brazil [2], for example. In Western Europe, as well as other parts of localized wind events on power grids. As a result, only a few studies on
the globe, they are regarded as the main cause of transmission line the response of transmission lines to EPS storms have been published
failures [3]. during this period, e.g. [9–11], none of them addressing the problem of
A transmission line cascade is a difficult-to-control failure sequence transmission line cascades triggered during extratropical cyclones. This
of line supports, triggered by the failure of a single or a limited set article describes an event-based approach to generate wind loads for
of line components. They are commonly classified into vertical, trans- the time-domain dynamic structural analysis technique introduced by
verse and longitudinal cascades, depending on the damage pattern [4]. the authors [12] and used for simulating transmission line progressive
Extreme extratropical cyclone events, sometimes combined with ice failures. The model can replicate wind fields of different storm types,
storms, can overload a large extension of a transmission line and are but the present article solely focuses on EPS storms.

∗ Corresponding author.
E-mail addresses: [email protected] (F. Alminhana), [email protected] (F. Albermani).

https://doi.org/10.1016/j.engstruct.2023.115942
Received 20 October 2022; Received in revised form 14 February 2023; Accepted 1 March 2023
Available online 15 March 2023
0141-0296/© 2023 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
F. Alminhana et al. Engineering Structures 284 (2023) 115942

Fig. 1. Simulated 1-min record of the longitudinal turbulence for points 𝐏1 to 𝐏4 .

2. Wind field modelling 2.2. Validation

2.1. Methodology A validation study was conducted to assess the suitability of the pro-
posed model in generating correlated extended pressure systems (EPS)
wind fields by comparing the statistical properties of simulated wind
A wind field can be mathematically modelled as a vector field, in
speed time histories with those input into the ARMA process. It is noted
which a wind velocity is assigned to each point 𝐏 of the field at any
that ARMA processes become unstable for small time increments [19],
time 𝑡. The total wind velocity at any location can be described by
since the orders required to produce a good approximation may vary
superposing two vectors: a slowly varying non-turbulent component 𝐔̄
largely and become difficult to predict. As such, a parametric study was
and a turbulent fluctuation 𝐮, according to: conducted to determine an appropriate time increment 𝛥𝜏 and orders
̄
𝐔(𝐏, 𝑡) = 𝐔(𝐏, 𝑡) + 𝐮(𝐏, 𝑡) (1) 𝑝 and 𝑞 of the ARMA to properly simulate wind speed time series.
A simple wind field with a set of four points (𝐏1 to 𝐏4 ), all at a height
Eq. (1) is general and can be employed to model wind fields of 10 m was employed. They are aligned horizontally and separated
̄
produced by different types of storms by properly setting vectors 𝐔(𝐏, 𝑡) from 𝐏1 by 10 m, 50 m and 150 m for 𝐏2 , 𝐏3 and 𝐏4 , respectively. The
and 𝐮(𝐏, 𝑡). For EPS storms, the slowly-varying component vector 𝐔̄ can terrain roughness length for a suburban terrain was assumed to be
be expressed as the product of the mean-speed wind profile 𝑈̄ (𝑧ℎ ) and 𝑧𝑜 = 0.8 m. A mean wind speed of 𝑈̄ 𝑟𝑒𝑓 = 30 m s−1 at 𝑧𝑟𝑒𝑓 = 10 m
the unit vector 𝐮̂ : was adopted.
Preliminary simulations indicated that it was not feasible to cal-
̄
𝐔(𝐏) = 𝑈̄ (𝑧ℎ )̂𝐮 (2) culate the coefficients of matrices 𝐀𝑖 and 𝐁𝑖 in Eq. (3) for time steps
smaller than around 𝛥𝜏 ≅ 1 × 10−3 s. For time intervals in the
The ‘‘logarithmic law’’ developed by Prandtl was adopted in this in-
range of 1 × 10−3 < 𝛥𝜏 < 5 × 10−3 s, the ARMA method became
vestigation to describe 𝑈̄ (𝑧ℎ ), whose formulation can be found in wind
unstable. Although it was possible to obtain matrices 𝐀𝑖 and 𝐁𝑖 , the
engineering textbooks, e.g. Holmes [13]. The autoregressive moving-
resulting ARMA series was meaningless in most cases. As such, four
average (ARMA) method proposed by Samaras et al. [14] was selected
time increments 𝛥𝜏 ⩾ 5 × 10−3 s were selected to generate wind speed
to generate the wind turbulence in this work. Only longitudinal and
time series at points 𝐏1 to 𝐏4 , namely: 𝛥𝜏 = 5 × 10−3 s, 1 × 10−2 s,
lateral turbulence are accounted for the this work, as the vertical
5 × 10−2 s and 1 × 10−1 s. The frequency content in these series is upper-
turbulence is regarded as having minor influence in the response of civil
bonded by 𝑓 = 100, 50, 10 and 5 Hz, respectively. It is noted that the
structures [15] and has been neglected in similar investigations [16].
upper-bond frequency for turbulence generation must be greater than
The multivariate form of an ARMA(𝑝, 𝑞) process for simulating the wind
2 Hz [20] to properly reproduce the energy content in the fluctuating
turbulent component 𝐮 at discrete time-steps 𝛥𝑡, where 𝑝 and 𝑞 are the wind.
orders of the process, can be expressed as: Wind speed time series were generated combining the selected time
∑
𝑝 ∑
𝑞 increments 𝛥𝜏 with ARMA orders 𝑝 = 𝑞 = 3, 5 and 10. Each series had a
𝐮(𝑥, 𝑦, 𝑧, 𝑡) = 𝐀𝑖 𝐮(𝑥, 𝑦, 𝑧, 𝑡 − 𝑖𝛥𝑡) + 𝐁𝑖 Ψ𝑖 (3) total duration of 2 h. Ensembles of longitudinal and lateral wind speed
𝑖=1 𝑖=0
time history with 60 samples of 2 min each were formed for points 𝐏1 to
In Eq. (3), 𝐀𝑖 and 𝐁𝑖 are 𝑚 × 𝑚 autoregressive and moving average coef- 𝐏4 , according to the different combinations. For illustration purposes,
ficient matrices, respectively, and Ψ(𝑡) is the 𝑚 × 1 random shock vec- Fig. 1 depicts 1-min samples of the longitudinal wind speed time history
tor, being 𝑚 is the number of locations. The coefficients of matrices 𝐀𝑖 of the four points (𝛥𝜏 = 5 × 10−2 s).
and 𝐁𝑖 are determined according to prescribed cross-correlations, based The power spectral density of ensemble samples were calculated
on the selected wind spectra and coherence functions. The Kaimal through Fast Fourier Transform (FFT). Fig. 2 illustrates the averaged
spectra [17] and the coherence function proposed by Davenport [18] longitudinal turbulence autospectra of each ensemble at point 𝐏1 for
were employed in the present investigation. The random shocks Ψ(𝑡) 𝛥𝜏 = 1 × 10−2 and 5 × 10−2 s (𝑝 = 𝑞 = 5) normalized by the
are generated according to a Gaussian probability density function squared frictional velocity 𝑢∗ , along with the normalized target Kaimal
(PDF) [13]. spectrum [17].

2
F. Alminhana et al. Engineering Structures 284 (2023) 115942

Fig. 2. Longitudinal turbulence autospectra at 𝐏1 for different 𝛥𝜏 and ARMA(5,5).

Fig. 3. Longitudinal co-spectrum and cross-correlation decay - 𝛥𝜏 = 5 × 10−2 s.

The comparison between simulated and target longitudinal turbu- series generated with a time increment of 𝛥𝜏 = 5 × 10−2 s and an
lence spectra for all chosen time-increment and ARMA-order combi- ARMA(5,5). The Gaussian probability density functions used to gen-
nations revealed that spectra agreement enhances with the adoption erate the longitudinal 𝑁(30, 10.33) and lateral 𝑁(0, 7.75) wind speeds
of higher ARMA orders. The same trend was observed for lateral are also included in the respective plot. Close agreement between the
turbulence spectra. The usage of larger 𝛥𝜏 also improves the match histogram and the normal PDF is observed for both components of
between the two spectra (Fig. 2). It is noted, however, that higher simulated wind speed.
ARMA orders require additional computational effort to calculate and Based on these findings, time steps of 𝛥𝜏 = 5 × 10−2 s and ARMA or-
store matrices 𝐀𝑖 and 𝐁𝑖 in Eq. (3), and larger time increments reduce ders of 𝑝 = 𝑞 = 5 will be adopted in the wind field simulations presented
the frequency content of the wind speed series. in section 5. This combination reproduces turbulent energy content
Averaged co-spectral density of simulated series were determined properly up to a frequency of 10 Hz, which is significantly beyond the
based on the series ensembles of locations 𝐏1 to 𝐏4 . These adequately minimum upper-bond frequency of 2 Hz discussed in section 2.1.
match the target co-spectrum calculated according to Davenport [18].
Fig. 3(a) depicts the simulated co-spectral density of longitudinal tur-
3. Wind induced loads on line components
bulence between points 𝐏1 and 𝐏2 , along with the respective theoretical
co-spectrum.
The decay of cross-correlation of longitudinal fluctuations with The calculation of the wind loads on transmission line components
horizontal distance 𝛥𝑥 was also assessed. Fig. 3(b) depicts a comparison based on the time history of wind velocities follows the provisions
between correlation coefficients of points 𝐏2 , 𝐏3 and 𝐏4 in relation to of the transmission line design code IEC 60826 [8], which was also
𝐏1 and the target decay curve, determined according to Davenport’s adopted by previous research [11,21]. The formulation in this code is
coherence model [18]. The simulated correlation points adhere closely scalar based, suitable for cases where the flow produced by the wind
to the theoretical curve. field is assumed to be steady to simplify calculations. In order to tackle
Histograms of longitudinal and lateral wind speeds at location 𝐏1 realistic unsteady wind field flows, code equations were re-expressed in
are depicted at Fig. 4. They were determined from wind speed time a vectorial form. This allows easier updating of wind loads and coupling

3
F. Alminhana et al. Engineering Structures 284 (2023) 115942

Fig. 4. Longitudinal and lateral turbulence PDF functions at point 𝐏1 - 𝛥𝜏 = 5 × 10−2 s.

with the global coordinate reference system employed in the dynamic

analysis.

3.1. Free-stream dynamic pressure

The forces caused by the action of wind on any obstacle immersed

within it are proportional to the free-stream dynamic pressure 𝑞, which
is obtained from the wind velocity 𝐕 in the upstream flow, unaffected
by the obstacle, by means of Bernoulli’s equation, resulting in:
1
𝑞= 𝜌 (𝐕 ⋅ 𝐕) (4)
2
where 𝜌 is the air density, which is dependent on local altitude and
temperature. In standard conditions for temperature and pressure (sea
level and at 15 °C), 𝜌 equals approximately 1.225 kg∕m3 .

3.2. Wind load on conductors and shield wires

Wind loads acting on conductors, shield wires and tower guys are
determined assuming no wind load component is applied along the
cable, i.e. only normal loads to the cables are considered. The total
wind load vector 𝐅𝑐 acting orthogonally to a cable element can be
calculated according to:
[ ] (𝐬 × 𝐯) × 𝐬 Fig. 5. Scheme of a generic tower section for wind loading purposes.
𝐅𝑐 = 𝐶𝑑𝑐 𝑞𝑑𝑐 𝐿𝑐 1 − (𝐯 ⋅ 𝐬)2 (5)
‖ (𝐬 × 𝐯) × 𝐬‖
where 𝐶𝑑𝑐 is the cable drag coefficient, 𝑑𝑐 is the cable diameter, 𝐿𝑐 is
the cable length, 𝐯 = 𝐕∕‖𝐕‖ is the wind velocity unit vector, and 𝐬 is 3.4. Wind load on lattice towers
the cable element chord unit vector. The drag coefficient 𝐶𝑑𝑐 depends
on the Reynolds number 𝑅𝑒 and vary significantly over a wide range Wind forces on lattice towers are traditionally computed as the total
in the 𝑅𝑒 [7]. The IEC 60826 [8] and the ASCE Manual 74 [7] suggest force applied to different sections of the tower (Fig. 5). This method
the adoption of a uniform drag coefficient 𝐶𝑑𝑐 = 1 for general stranded produces fairly accurate results for tower sections with rectangular
conductors under usual design wind speeds. A drag coefficient of 𝐶𝑑𝑐 =
cross-sections [22,23]. The panels of each section are defined using
1 is also adopted in the case studies described in section 5.
each of the section faces and are defined by considering the members
3.3. Wind load on insulator strings contained within it. They are oriented spatially by their normal vector
𝐧 (Fig. 5), and both the total member (net) area 𝐴𝑛 and total surface
The resultant force of wind load acting on an insulator string can (solid) area 𝐴𝑠 should be defined to perform wind loading calculations.
be achieved through the following expression:
[ ] (𝐬 × 𝐯) × 𝐬 The resultant wind load on a tower section acting on the wind
𝐅𝑖 = 𝐶𝑑𝑖 𝑞𝐴 1 − (𝐯 ⋅ 𝐬)2 (6) direction 𝐅𝑡 is calculated by computing the contribution of the 𝑁 panels
‖ (𝐬 × 𝐯) × 𝐬‖
forming a particular section, through the vectorial expression:
where 𝐶𝑑𝑖 is the insulator drag coefficient, 𝑞 is the free-stream wind
pressure, 𝐴 is the insulator wind area, 𝐯 is the wind velocity unit vector, ∑
𝑁
[ ( ) ( ) ]
and 𝐬 is the insulator axis unit vector. In the case of insulators with 𝐅𝑡 = 𝐾𝑞𝐴𝑛,𝑖 𝐶𝑑𝑡𝑦,𝑖 𝐧𝑖 ⋅ 𝐣 (𝐯 ⋅ 𝐣)2 + 𝐶𝑑𝑡𝑥,𝑖 𝐧𝑖 ⋅ 𝐢 (𝐯 ⋅ 𝐢)2 𝐯 (7)
𝑖=1
multiple strings, Eq. (6) is applied to each string. A drag coefficient of
𝐶𝑑𝑖 = 1.20 is considered in this investigation, as recommended by the In this equation, 𝐴𝑛 is the area of the support angles in the plane of
IEC 60826. the face. The factor 𝐾 has a double function. It accounts for the panel’s

4
F. Alminhana et al. Engineering Structures 284 (2023) 115942

Fig. 6. Plan view of TL 230 kV Quinta-Pelotas 3.

position relative to the wind based on the dot product of vectors 𝐯 (wind
velocity unit vector) and 𝐧. Windward faces are kept in the summation
whereas leeward ones are removed by assuming 𝐾 = 0. Additionally,
it applies a factor to account for the effect of the angle of attack of
the wind (𝜃) in relation to the 𝑦-axis (transversal axis of the tower),
equivalent to (1 + 0.2 sin2 2𝜃), as indicated in Eq. (8).
{
0, 𝐯⋅𝐧⩾0
𝐾= [ ] (8)
1 + 0.8 (𝐯 ⋅ 𝐣)2 1 − (𝐯 ⋅ 𝐣)2 , 𝐯 ⋅ 𝐧 < 0
𝐶𝑑𝑡𝑥 and 𝐶𝑑𝑡𝑦 are the drag coefficients to wind acting on directions 𝑥
and 𝑦. They are dependent on the solidity ratio (𝜒) of each face, which
is the relation of net and solid area defined in Eq. (9).
{
𝐴𝑛 ∕𝐴𝑠 (𝐧 ⋅ 𝐢) , f𝑜𝑟 𝐶𝑑𝑡𝑥
𝜒= (9)
𝐴𝑛 ∕𝐴𝑠 (𝐧 ⋅ 𝐣) , f𝑜𝑟 𝐶𝑑𝑡𝑦
The values of 𝐶𝑑𝑡𝑥 and 𝐶𝑑𝑡𝑦 as a function of 𝜒 for flat sided tower
members are given by:

𝐶𝑑𝑡 = 4.1727𝜒 2 − 6.1681𝜒 + 4.0088 (10)

4. Nonlinear time-domain dynamic analysis

The compact nonlinear time-domain dynamic analysis technique

developed by the authors [12] to obtain the response of complete
transmission line sections under time-dependent loads is employed in
this investigation. Each line section is modelled as a group of three
basic types of substructures: conductor and shield-wire spans, supports
and insulator sets. In turn, each substructure is formed by an assembly
of basic elements: cable (conductor and shield wire spans), tension-
only (insulators), truss and beam–column (lattice supports) elements.
The technique accounts for both geometric (all elements) and material
(beam–column elements only) nonlinearities.
The dynamic response of structural systems is generally formulated
in terms of the equations of motion. For linear elastic multi-degree-of-
freedom systems with viscous damping these equations can be written
as: Fig. 7. Transverse outline of tower FS22.

𝐌𝐮̈ + 𝐂𝐮̇ + 𝐊𝐮 = 𝐏 (11)

5
F. Alminhana et al. Engineering Structures 284 (2023) 115942

Fig. 8. Profile and plan view of the investigated line segment.

Fig. 9. Wind speed time histories at support 47 for event E1.

Fig. 10. Time history of tip displacement at node 𝐏𝐿 of support 47.

where: 𝐌, 𝐂 and 𝐊 are the global matrices of mass, viscous damping explicit integration method based on a central finite difference scheme
and stiffness of the structure, respectively; 𝐮̈ , 𝐮̇ and 𝐮 are the time- to solve Eq. (11). A lumped-mass approach a is adopted with damping
dependent vectors of acceleration, velocity and displacements; and 𝐏 is proportional to mass, resulting in a diagonalized matrices 𝐌 and 𝐂.
the vector of applied forces, also time dependent. The technique uses an As a consequence, the system is solved independently for each DOF,

6
F. Alminhana et al. Engineering Structures 284 (2023) 115942

Fig. 11. Transverse wind speeds at support 47 for event E2 (𝑧 = 27.6 m).

according to the following recurring formula: to an everyday stress (EDS) condition: average temperature of 20 °C
[ ∑ 𝑗 and no wind; catenary parameter of 𝐶 = 1755 m, equivalent to a
1 (1 + 𝛼)𝛼 (𝑝𝑖 − 𝑁 𝑗=1 𝑓𝑖 )𝛥𝑡
2
𝑢𝑖+1 = + 2(1 + 𝛼)𝑢𝑖 horizontal tension component of 𝐻 = 22.43 kN. The shield wires are
2 + 𝛼 𝑐𝑚 𝛥𝑡 𝑚 3/8′′ high strength (HS) steel cables, sagged in an EDS condition with
] 𝐶 = 1950 m, resulting in an axial tension with horizontal component
× − (2 − 𝑐𝑚 𝛥𝑡)𝛼 𝑢𝑖−1 (12) of 𝐻 = 7.80 kN. The natural frequencies of low-sagged cables can
be determined analytically [27]. Assuming a levelled span of 450 m,
where 𝑢𝑖+1 , 𝑢𝑖 and 𝑢𝑖−1 are displacements at time steps (𝑖 + 1), 𝑖 and the lowest frequency for the in-plane mode shapes is around 0.30 Hz
(𝑖 − 1), of a given DOF; 𝑚 is the nodal mass; 𝑐𝑚 is the proportionality for both conductors and shield wires. All suspension insulators are
constant between damping coefficient and nodal mass; 𝑝𝑖 is the external conventional I-string assemblies consisting of porcelain insulating bells.
∑ 𝑗
load applied on the DOF and 𝑁 𝑗=1 𝑓𝑖 is the sum of the end forces 𝑓𝑖 of
Their total length is 2.6 m. They have a weight per unit length of
the 𝑁 members connected to a node. Eq. (12) eases the introduction of 500 N∕m and an axial rigidity of 𝐸𝐴 = 100 MN.
element nonlinearities, which can be controlled by member end forces All line suspension supports are freestanding delta-type steel lattice
𝑓𝑖 . Convergence and stability criteria for the method are discussed in towers named FS. They vary from 19.4 m (FS15) to 34.4 m (FS30) in
Refs. [12,24]. The coefficient 𝑐𝑚 is selected in order to obtain a damping total height, with a total self-weight ranging from 25.7 to 48.5 kN. A
ratio 𝜁𝑛 specific for the 𝑛th vibration mode of system substructures. schematic representation of FS22 is illustrated in Fig. 7. These towers
It can combine the effects of the different sources of damping, but 𝑐𝑚 are formed by an identical top geometry (sections S4 and S5), a
is used to account for structural damping only. Aerodynamic damping common body (S3), a body extension (S2) of 5 or 10 m, and legs (S1)
forces are incorporated in the analysis as wind drag forces (Section 3.2). with heights from 1 to 5 m. They are square based with side lengths
ranging from 4.2 m to 7.5 m. All members are structural steel angle
5. Case study sections of grades 250 MPa or 345 MPa.
The natural frequencies of towers FS22, FS23 and FS24 were deter-
In the present case study, the wind field modelling techniques mined using an in-house algorithm based on the Lanczos-Ritz method
described previously, combined with the compact dynamic analysis [28]. The fundamental frequencies for these towers (transverse mode)
method of line sections proposed by the authors [12], is employed to are 6.51, 6.22 and 5.94 Hz, respectively.
assess the response of an existing 230 kV TL section to two wind load
events, namely E1 and E2. Both correspond to extratropical cyclones 5.2. Line modelling
acting transverse to the line, but with different wind intensities: E1 is a
design level event, while E2 is an event adjusted to trigger line failure. Nearly half of the line supports on 230 kV Quinta-Pelotas 3 TL are
within a very long tension section of 45 supports. A line segment with
nine spans and eight suspension supports Figs. 6 and 8, numbered
5.1. Line description 43 to 52, was selected for the simulations. This segment includes the
five supports (46 to 50) that collapsed in the accident reported by
The 230 kV Quinta-Pelotas 3 TL is a single-circuit transmission line, [26]. The first and last suspension towers of the segment (supports 43
which connects electrical substations in the cities of Pelotas and Rio and 52) were replaced by rigid supports. Each span was automatically
Grande in south Brazil. The total length of the line is about 45 km, divided into a number of cable elements with a length close to 5 m. The
crossing mostly flat, open terrain. The line has a total of 103 supports line’s I-string insulators were broken down into three pin-connected
with an average span around 466 m long. It was constructed in 1983 tension-only elements each.
and has experienced two accidents caused by wind storms since it was A time-step interval of 𝛥𝑡 = 1.0 × 10−4 s was adopted to integrate the
built [25,26]. equations of motion Eq. (12), which is just less than the critical time-
The transmission system is composed of three conductor phases step required for the convergence of central finite difference method
and two shield wires. The three conductor phases are formed by a [12]. This time increment is automatically reduced to 𝛥𝑡 = 1.0 × 10−5 s
single aluminium conductor steel reinforced (ACSR) type conductor, when plastic hinges start forming in the tower structure, and kept con-
with the commercial name Grosbeak. All phases are sagged according stant for the rest of the analysis. This guards against the occurrence of

7
F. Alminhana et al. Engineering Structures 284 (2023) 115942

Fig. 12. Perspective of the collapsed line section for event E2. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of
this article.)

excessive plastic displacements within a time-step, drifting the solution all supports are fully restrained. Nodes that contact the ground during
away from the actual load–displacement path [12]. the collapse of a support are vertically restrained from that time step
The mass-proportional constants of 𝑐𝑚 = 1.8 and 0.1 s−1 were set onwards.
to towers and cables, respectively, to account for structural damping.
They were adapted from tests with similar structures [29–31], and are 5.3. Event E1
equivalent to a damping ratio of approximately 𝜁 = 4% (towers) and 2%
(cables). Aerodynamic damping is automatically accounted for as wind This event is introduced to determine the line response under re-
drag forces (Section 4). At the beginning of the simulations, the line vised design conditions. A return period of 𝑇 = 100 years was adopted
section components are assumed at rest. The four bottom leg nodes of for calculating the mean reference wind speed, matching the current

8
F. Alminhana et al. Engineering Structures 284 (2023) 115942

Fig. 13. Axial force time history of the first members to fail in supports 45 to 48. (For interpretation of the references to colour in this figure legend, the reader is referred to
the web version of this article.)

design practice (in ultimate limit state) adopted by Brazilian power util- This reveals significant across-wind effects induced in the supports due
ities. Wind speed recordings from the meteorological station at the In- to unequal wind pressures acting on the spans, resulting in unbalanced
ternational Airport of Pelotas, located around 6 km from the first struc- cable tension in the supports. The across-wind effect is completely
ture of the 230 kV Quinta-Pelotas 3 TL (Fig. 6), were used to determine disregarded in conventional line designs, as it is assumed irrelevant
the resulting mean reference wind speed (𝑈̄ 𝑟𝑒𝑓 = 27.8 m s−1 ) [28]. A compared with along-wind response.
roughness length of 𝑧𝑜 = 0.04 m corresponding to grassland terrain was
adopted. A time increment of 𝛥𝜏 = 5 × 10−2 s and ARMA(5,5) were 5.4. Event E2
used to generate wind-speed time histories (Section 2.2).
The wind loads were gradually applied on the line components, in In event E2, the response of the line section is assessed under an
the negative 𝑌 direction (Fig. 8), after application of the dead-weight extreme extratropical cyclone storm. The wind speeds were incremen-
was completed (𝑡 = 1 s). Wind speeds were linearly ramped from 𝑈̄ = 0 tally increased to identify the threshold of the line segment failure. The
at 𝑡 = 1 s until 𝑈̄ 𝑟𝑒𝑓 was reached at 𝑡 = 5 s. Fig. 9 shows the time histories EPS wind field simulated has the same parameters of event E1, except
of the simulated speeds in the direction transverse to the line, at support for the magnitude of the wind speeds. The mean reference wind speed
47, at heights 𝑧ℎ = 2.5 and 27.6 m above the ground, corresponding to was initially ramped to 𝑈̄ 𝑟𝑒𝑓 = 45 m s−1 from time 𝑡 = 1 to 10 s. No
the geometric centre of tower’s sections S1 and S5 (Fig. 7). damage occurred in the supports during this interval. Subsequently,
The nonlinear dynamic analysis showed that the line section re- wind speeds were accelerated at 1 m s−2 until the first plastic hinge
sponded safely under this event, with all supports remaining within was formed. This occurred at support 44 (longest wind span) at 𝑡 ≅ 19 s
the elastic range after the wind loads were fully applied and remained when the reference speed reached 𝑈̄ 𝑟𝑒𝑓 ≅ 54 m s−1 . This speed was held
acting for 25 s. Fig. 10 shows the time histories of displacements 𝛿𝑋 and constant until the end of the simulation (𝑡 = 65 s).
𝛿𝑌 , longitudinal and transverse to the line axis, respectively, at node Fig. 11 depicts the time history of the wind speeds at the geometric
𝐏𝐿 (Fig. 7) for support 47. Relevant across-wind displacements 𝛿𝑋 are centre of section S5 (refer to Fig. 7) of support 47 (first to collapse),
observed, with maximum magnitude about 30% of the respective 𝛿𝑌 . which has an initial height of 𝑧 = 27.6 m above the ground and is

9
F. Alminhana et al. Engineering Structures 284 (2023) 115942

considered attached to the tower. The wind speeds drop considerably In the second event, the line section was subjected to the action of
after support 47 failure (𝑡 ≅ 40.77 s), as the point moves towards the an extratropical cyclone with the same characteristics as in the first
ground. During this period the mean speed component is recalculated event, except for a stronger storm intensity. The wind speeds were
as elevation 𝑧 of S5 is updated. ramped up to the point of the formation of the first plastic hinge among
Collapse of the entire line section occurred during this event. Al- the members of the supports, with the reference mean speed kept
though plastic hinges were first formed in support 44, it survived constant after this time. The collapse of all supports occurred during
temporarily. Support 47 was the first to collapse at 𝑡 ≅ 40.77 s, which the event. The central support was the first to collapse, and a sequential
correspond to the onset of tower collapse, i.e. the moment that a failure was observed for the supports south of it, but not for those the
particular member fails triggering the failure sequence ensuing the north. This suggests the occurrence of a multi-failure event rather than
tower collapse to the ground. Then, supports 46, 45 and 44 failed a transverse cascade.
consecutively at times 1.72, 4.94 and 14.85 s after support 47 (Fig. 12).
For the supports north of 47, however, the supports did not fail con- CRediT authorship contribution statement
secutively. Although, support 48 collapsed after 47 (9.59 s), support 50
(14.27 s) failed prior to 49 (16.36 s), interrupting the failure sequence. Fabio Alminhana: Conceptualization, Methodology, Development,
Support 51 was the last to fail at 20.64 s after the first line collapse. Validation, Investigation, Writing – original draft, Writing – review &
A perspective of the line section at the end of the analysis is depicted editing. Matthew Mason: Conceptualization, Methodology, Validation,
in Fig. 12. The members in red are those where plastic hinges were Writing – reviewing and editing. Faris Albermani: Conceptualization,
formed. For all supports, the main failure mechanism was the buckling Methodology, Validation, Writing – reviewing and editing.
of a common-body leg member in the leeward face. This caused the
towers to bend around their mid-height transversally to the line until Declaration of competing interest
the right cross arm reached the ground, resulting in the failure modes
observed for supports 44 to 46 and 49 to 51. A second failure mode The authors declare that they have no known competing finan-
was observed for supports 47 and 48, where the leeward legs buckled cial interests or personal relationships that could have appeared to
during the falling of the top part. influence the work reported in this paper.
Fig. 13 illustrates the axial force (𝐹 ) in the leg members of supports
45 to 48, which were the first to fail in these towers. In all cases, Data availability
member buckling occurred when the axial force reached a peak of
around −300 kN, and then dramatically plunged to a residual value of Data will be made available on request.
approximately 10 kN. It is seen that after the collapse of an adjacent
support (red lines), a sudden drop in the axial force occurs. As a support Acknowledgements
starts collapsing, the tension in the attached cables decreases sharply,
while the force in the tower’s buckled member is redistributed to the
The financial support of CNPq - Brazilian National Council for
neighbouring ones. This reduces the loads in the adjacent supports until
Scientific and Technological Development - is acknowledged.
the force redistribution is completed.
As the supports did not collapse consecutively, the occurrence of
References
a multi-failure event [4] rather than a transverse cascade (progressive
failure) is speculated. Two failure modes for line section lattice towers [1] CIGRÉ. Technical brochure 344: Big storm events what we have learned. Tech.
were obtained during the simulation: (i) common body alone and (ii) rep, Conseil International des Grands Réseaux Electriques; 2008.
common body combined with main leg collapse. These modes are [2] Riera JD. Consideração dos carregamentos induzidos pela ação do vento no
similar to those reported in the latest accident with the line [22]. projeto de Linhas de Transmissão. In: Seminário internacional : Ventos sobre
linhas de transmissão. 2002.
[3] Peyrot AH. Wind loading: Uncertainties and honesty suggest simplification. In:
6. Conclusions Electrical transmission and substation structures conference - ASCE. 2009, p.
118–31. http://dx.doi.org/10.1061/41077(363)17.
[4] Alminhana F, Albermani F, Mason M. Cascading collapse of transmission lines.
A numerical model devised to incorporate correlated
Proc Inst Civ Eng - Eng Comput Mech 2018;1–36. http://dx.doi.org/10.1680/
time-dependent wind loads into the dynamic analysis technique out- jencm.18.00010.
lined by the authors [12] has been presented. The model is event-based, [5] CIGRÉ. Brochure 515 - Mechanical security of overhead lines: Containing
but only extratropical-cyclone loads are considered in this article. cascading failures and mitigating their effects. Tech. rep, Conseil International
Wind speeds are calculated as a superposition of a mean wind speed des Grands Réseaux Electriques; 2012.
[6] Standards Australia/Standards New Zealand. AS/NZS7000- Overhead line design.
component and a turbulent component. The well-known logarithmic
Sydney; 2016.
law is used to calculate the mean component for extratropical cyclones. [7] ASCE. ASCE manual 74 - Guidelines for electrical transmission line structural
The turbulent components are generated based on stochastic process loading. Reston, VA: American Society of Civil Engineers; 2010.
theory, using the autoregressive moving-average (ARMA) method pro- [8] International Electrotechnical Commission. IEC 60826 - Design criteria of
posed by Samaras et al. [14]. A validation case was conducted to overhead transmission lines. Geneva; 2003.
[9] Loredo-Souza AM, Davenport AG. The influence of the design methodology in
examine the suitability of the model in reproducing actual wind fields. the response of transmission towers to wind loading. J Wind Eng Ind Aerodyn
This verified the generation of the turbulent components by comparing 2003;91(8):995–1005. http://dx.doi.org/10.1016/S0167-6105(03)00048-5.
statistical properties extracted from the simulated time histories and [10] Paluch MJ, Cappellari TTO, Riera JD. Experimental and numerical assessment
those input into the method. A high level of agreement was achieved. of EPS wind action on long span transmission line conductors. J Wind Eng
Ind Aerodyn 2007;95(7):473–92. http://dx.doi.org/10.1016/j.jweia.2006.09.003,
A case study was presented investigating the response of an existing
URL http://www.sciencedirect.com/science/article/pii/S0167610506001449.
line section to two wind events. For the first event, the line section [11] Miguel LcFF, Fadel Miguel LF, Riera JD, Kaminski J, Ramos de Menezes RC.
was exposed to an extratropical cyclone acting transversally to the line Assessment of code recommendations through simulation of EPS wind loads
section. The storm intensity was adjusted according to current design along a segment of a transmission line. Eng Struct 2012;43:1–11. http://dx.doi.
criteria, based on local wind data. It was verified that all line supports org/10.1016/j.engstruct.2012.05.004.
[12] Alminhana F, Mason M, Albermani F. A compact nonlinear dynamic anal-
behaved safely within the elastic range during the analysis. Significant ysis technique for transmission line cascades. Eng Struct 2018;158(March
across-wind response was observed, which is disregarded in design 2018):164–74.
practice and should be an object for future research. [13] Holmes JD. Wind loading of structures. CRC Press; 2001, p. 368.

10
F. Alminhana et al. Engineering Structures 284 (2023) 115942

[14] Samaras E, Shinzuka M, Tsurui A. ARMA representation of random pro- [23] Mara TG. Capacity assessment of a transmission tower under wind loading (Ph.D.
cesses. J Eng Mech 1985;111(3):449–61. http://dx.doi.org/10.1061/(ASCE)0733- thesis), The University of Western Ontario; 2013.
9399(1985)111:3(449). [24] Alminhana F, Albermani F, Mason M. Comparison of responses of guyed and
[15] Borges JF, Castanheta M. Structural safety. 2nd ed.. Lisbon, Portugal: Laboratório freestanding transmission line towers under conductor breakage loading. Int J
Nacional de Engenharia Civil; 1971. Struct Stab Dyn 2015;15(8):1540021–3.
[16] Chay M. Time domain simulation of downburst wind loads on guyed transmission
[25] da Silva VR, Elias RdA. Desempenho estrutural de LTs do sistema CEEE segundo
line structures (Ph.D. thesis), The University of Queensland; 2008.
differentes procedimentos de cálculo de cargas de vento. In: CIGRÉ - Conseil
[17] Kaimal JCJ, Wyngaard JCJ, Izumi Y, Coté OR, Cote OR. Spectral characteristics
International des Grands Réseaux Electriques, editor. XI seminário nacional de
of surface-layer turbulence. Q J R Meteorol Soc 1972;98(417):563–89.
[18] Davenport AG. The dependence of wind loads upon meteorological parameters. produção e transmissão de energia elétrica. Rio de Janeiro; 1991, p. 1–6.
In: Proceedings international research seminar wind effects on buildings and [26] da Silva VR. Relatório da visita de inspeção do acidente com a LT 230 KV
structures. 1968. QUINTA- PELOTAS 3. Tech. rep, Porto Alegre; 2006, p. 1–12.
[19] Chay MT, Albermani F, Wilson R. Numerical and analytical simulation of [27] Irvine H. Cable structures. Cambridge, MA: Dover Publications; 1992.
downburst wind loads. Eng Struct 2006;28(2):240–54. [28] Alminhana F. Cascading collapse of transmission line structures (Ph.D. thesis),
[20] Sparling BF. The dynamic behavior of guys and guyed masts in turbulent winds School of Civil Engineering, The University of Queensland; 2017, p. 262.
(Ph.D. thesis), The University of Western Ontario; 1995. [29] Peyrot A, Kluge R, Lee J. Longitudinal loading tests on a transmission line. Tech.
[21] Ramos de Menezes RC, Riera JD, Fadel Miguel LF, Kaminski Jr J, Fadel rep, Madison, Wisconsin, USA: EPRI - Eletric Power Research Institute; 1978.
Miguel LF, Ferreira da Silva JBG. On modeling the dynamic response of a [30] da Silva VR, Riera JD, Joaquim Blessman, Nani LF, Galindez EE. Determinação
190m- high TL tower for large river crossing in the Brazilian Amazon region
experimental das propriedades dinâmicas básicas de uma torre de transmissão
Ruy Carlos Ramos de. In: CIGRÉ - Conseil International des Grands Réseaux
de 230 kV. In: VII Seminário nacional de produção e transmissão de energia
Electriques, editor. CIGRE conference. Paris: Conseil International des Grands
elétrica. Brasilia; 1984, p. 10.
Réseaux Electriques; 2012, p. 1–10.
[22] Silva AdOe, de Medeiros JCP, Rocha MM, Leandro Inácio Rippel, Carpeggiani EA, [31] Diana G, Falco M, Cigada A, Manenti A. On the measurement of over
Núñez GJZ, Loredo-Souza AM. Wind loads on metallic latticed transmission line head transmission lines conductor self-damping. IEEE Trans Power Deliv
towers. In: CIGRÉ paper B2-202. Paris: Conseil International des Grands Réseaux 2000;15(1):285–92. http://dx.doi.org/10.1109/61.847264.
Electriques; 2006.

11