Engineering Structures 143 (2017) 549557

Contents lists available at ScienceDirect

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

Performance-based design of steel towers subject to wind action

Rodolfo K. Tessari , Henrique M. Kroetz, Andr T. Beck
Department of Structural Engineering, University of So Paulo, Av. Trabalhador So-carlense 400, 13566-590 So Carlos, SP, Brazil

a r t i c l e i n f o a b s t r a c t

Article history: Performance-based Wind Engineering (PBWE) is a novel design philosophy that aims to identify and
Received 21 September 2016 quantify the uncertainties involved in structural design, in order to ensure predictable performance levels
Revised 23 March 2017 to engineering structures. Due to the recent proposal of the methodology and formulation complexity,
Accepted 26 March 2017
there are few studies related to PBWE, each presenting different limitations. This paper proposes an
Available online 29 April 2017
application of the Performance-based Wind Engineering methodology to the probabilistic analysis of
steel towers, evaluating different calculation models for the estimation of wind forces on this type of
Keywords:
structure. Uncertainties involved in the characterization of the wind field and the structural strength
Performance-based Engineering
Performance-based Wind Engineering
were investigated, and two procedures of the Brazilian winds standard NBR6123:88 for the estimation
Structural reliability of wind forces on steel towers were analyzed. A case study concerning the reliability estimation of a
Probabilistic analysis telecommunication tower was also conducted. It was found that both studied calculation models lead
Latticed telecommunication tower to similar safety levels, and that the design of towers considering that wind always blows from the worst
direction is too conservative. It is also shown that, in PBWE, minimum cost design can be guided by
assigning same target reliability, but different mean recurrence intervals for different performance levels.
2017 Elsevier Ltd. All rights reserved.

1. Introduction Classically, risk management in structural design problems is

Inherently random and knowledge-based uncertainties are part
of the structural design process. Aleatory uncertainty is often asso-
ciated to environmental hazards (e.g. earthquakes, storms, land-
slides and tornadoes), whose intensities and frequencies are
difficult to predict, while epistemic uncertainty encompasses lim-
ited databases and limited models that cannot assure perfect rep-
resentation of structural responses. Uncertainties can be reduced
through research, but cannot be eliminated, giving rise to struc-
tural risks.
Risk represents a function of the likelihood of a hazardous event
taking place, and its social and economic consequences. It strongly
depends on location and typology of the concerned facility, since
any structure is subjected to a certain range of hazards. Therefore,
selection of the relevant threats to each structure and acknowledg-
ment of the uncertainties involved is important and demands
attention. To neglect or to underestimate risks may compromise
structural safety, and may expose people to avoidable, dangerous
situations; whereas overestimating risks can lead to misallocation
of resources [1].
Corresponding author.
E-mail addresses: tessari.rodolfo@usp.br (R.K. Tessari), henrique.kroetz@usp.br
(H.M. Kroetz), atbeck@sc.usp.br (A.T. Beck).
addressed through deterministic or semi-probabilistic approaches
(encoded in design standards). In the deterministic scenario, load
and resistance parameters are admitted as perfectly known and
invariant. In the latter, factors of safety calibrated according to tra-
ditional engineering practice are applied to the nominal or charac-
teristic values of design variables. Despite the advances in the field
of structural engineering during the past decades, both approaches
still present some flaws [25].
The search for more effective methods for quantifying and mit-
igating risks led to development of the Performance-based Engi-
neering (PBE) approach. The definition of PBE was first outlined
in SEAOCs Vision 2000 report [6] and soon became widespread
in the literature [79]. The PBE design philosophy represents a
paradigm in which the prescriptive approach imposed by struc-
tural standards is replaced by the quantitative assessment of
design alternatives against performance objectives, described in
probabilistic terms [10].
The first publications regarding performance-based procedures
were developed in the United States for seismic design and retrofit
of buildings [6,11]. These implementations were driven by the
1994 Northridge and 1995 Kobe earthquakes, that together caused
a total estimated loss of U$120 billion [12], even though affected
structures complied with seismic codes, based on traditional
design philosophies, prevailing at the time [9].

http://dx.doi.org/10.1016/j.engstruct.2017.03.053
0141-0296/ 2017 Elsevier Ltd. All rights reserved.
550 R.K. Tessari et al. / Engineering Structures 143 (2017) 549557
Naturally, the first-generation PBE procedures presented some
shortcomings [13]. In order to fill these gaps, a more robust
methodology was developed in the Pacific Earthquake Engineering
Research (PEER) Center, that turned out to be the most popular and
currently applied Performance-based Earthquake Engineering
(PBEE) formulation [14].
PEERs PBEE methodology was built over the foundations of
structural reliability theory. This provided a theoretical framework
to the probabilistic treatment of uncertainties and an explicit
system-level performance assessment [9]. Its successful applica-
tions made the technical and scientific community consider PEERs
rational approach ideal for the design of structures subjected to
natural hazards. In recent years, many studies have been carried
out on the performance assessment of structures subject to tsuna-
mis [15] and hurricanes [2,16,17].
Another branch of great interest for further developments deals
with the application of PBE concepts to wind engineering.
Performance-based Wind Engineering (PBWE) was first proposed
in 2004 [18] as the result of the Italian PERBACCO project, and
assumed a concise and general format in 2009, when its methodol-
ogy was established [19].
Recently, studies illustrating the applicability of PBWE method-
ology and proposing incremental improvements have been pub-
lished: Ciampoli, Petrini and Augusti [20] evaluated the
performance of a long span suspension bridge; Ciampoli and Pet-
rini [21] analyzed the structural behavior of a 74 floor building;
and Petrini, Gkoumas and Bontempi [22] performed an extensive
literature review on damage and loss analysis in order to expand
the PBWE procedure. Griffis et al. [23] also briefly proposed an
alternative procedure for PBWE, focusing on nonlinear dynamic
analysis of structures.
The rapid evolution and diffusion of the PBWE concept, over the
last years, is remarkable. However, due to the novelty and com-
plexity of the methodology, there are still few studies related to
the theme, each presenting different limitations. No studies
addressing PBWE design of steel towers were found in the pub-
lished literature. In this paper, the PBWE framework is adapted
to analyze the structural behavior of such structures in probabilis-
tic terms. Special attention is given to the choice of proper param-
eters characterizing the wind field and structural behavior, as well
as to the comparison of different wind action models for steel tow-
ers. The developed methodology is demonstrated in application to
a popular telecommunications steel tower.
2. The PBWE methodology
The present study is based on the Performance-based Wind
Engineering framework proposed in [19,24]. The central objective
of the PBWE procedure is to assess if a structural facility fulfills
specific performance requirements (usually related to safety, func-
tionality and comfort), as specified by end-users, stakeholders or
society. It consists of the following steps [25]:
i. characterizing the wind hazard at the candidate location of
the structure;
ii. probabilistic modeling of the wind-structure interaction
phenomena;
iii. analyzing the structural response within the scope of
stochastic mechanics;
iv. characterizing and evaluating the variables that express the
structural damage and govern the considered performance
measures;
v. defining the decision variables (DVs) that are appropriate to
quantify the intended structural performances (mainly in
terms of consequences of damage);
vi. assessing the structural risk based on the probabilistic
description of the DVs;
vii optimizing design (i.e. minimizing risk or maximizing a util-
ity function) by appropriate decision strategies.
The first step consists in choosing a set of intensity measure
parameters (IMs) that are sufficient to describe the site-specific
hazard (e.g. mean wind velocity, turbulence intensity, direction).
Attention must be paid to the selection of each IM, since they have
direct impact on the output of all the subsequent stages.
Step ii. corresponds to the choice and probabilistic characteriza-
tion of a set of interaction parameters (IPs) able to represent the inter-
action between the environment and the structure. Aerodynamic
coefficients and aeroelastic derivatives are proper examples of IPs.
The process continues (step iii.) with selection of the most sig-
nificant random structural parameters (SPs) and engineering
demand parameters (EDPs) that influence the structural behavior.
Material and geometrical properties are normally chosen as SPs,
whereas EDPs are represented by the acceleration, stress or dis-
placement at selected points. Such parameters are then probabilis-
tically assessed through structural response analyses.
In step iv., damage measures (DMs) and decision variables are
specified in order to quantify the structural damage and the build-
ing performance, respectively. Both are strongly interconnected
and depend on the considered facility type and usage. Typical
DMs are the loss of occupant comfort and the damage to structural
and non-structural components in a building, due to excessive
vibration or displacements. DVs include the number of casualties,
the economic losses or some threshold that represents the collapse
or loss of serviceability condition during windstorms.
DVs which quantify the performance objectives must distin-
guish between low- and high-performance levels [1]: the former
are related to ultimate limit states (ULS) and imply possible conse-
quences on structural integrity and personal safety; the latter are
associated to serviceability limit states (SLS) and affect operability
and comfort.
Regardless of the target performance, in the original PBWE pro-
cedure (as in the PEERs approach), the structural risk is defined as


the probability of a relevant DV exceeding a threshold value dv :



Gdv PDV > dv
RRRRR

Gdv jDM  f DMjEDP  f EDPjIM; SP; IP
f IPjIM; SP  f IM  f SP  dDM  dEDP  dIP  dIM  dSP
1
where f  symbolizes the probability density function, f j the
conditional probability density function, and G the complemen-
tary cumulative distribution function.
Through the multiple integral expressed in Eq. (1), it is possible
to separate the risk assessment into the previously stated elemen-
tary steps:
 site-specific hazard analysis and structural characterization, i.e.
the assessment of f IM and f SP, respectively;
 interaction analysis, that corresponds to the estimation of
f IPjIM; SP;
 structural analysis, aimed at assessing the probability density
function of the structural response f EDPjIM; IP; SP, conditional
on the output of the previous steps;
 damage analysis, that gives the damage probability density
function f DMjEDP conditional on EDP;


 finally, loss analysis, that matches the value of Gdv jDM.
With respect to the PEER approach, the major advance provided
by the PBWE methodology is the decoupled analysis of parameters
that characterize the structural behavior from parameters that
 R.K. Tessari et al. / Engineering Structures 143 (2017) 549557
qualify the wind field. This is possible due to the strict dependency
of the former on the candidate location of the structure, whereas
the latter depends only on the design configuration. Moreover,
unlike earthquakes, winds have instability phenomena related to
fluidstructure interaction [26]. Thus, the introduction of the inter-
action analysis stage is essential to determine the relevant features
of the wind effects. Hence, the current procedure allows to assess
sensitivity of each stochastic variable in the propagation of uncer-
tainties, identifying its contribution to the failure probability of the
structure. It is also possible to check the response variability occa-
sioned by consideration of different aerodynamic and aeroelastic
phenomena.
Given the complexity of the multi-dimensional integrals in Eq.
(1), the complete PBWE procedure is rarely applied to practical
design problems. Studies dealing with decision variables are usu-
ally related to optimization problems. In such cases, DVs are
rewritten in terms of the total expected cost of failure or the finan-
cial loss rate for discrete levels of damage [27,28]. Most researches
adopt assumptions that limit the quantification of risk to the like-
lihood of the structural response exceeding a certain threshold,
defined by particular engineering parameters. Ciampoli and Petrini
[24] argue that these simplifications are generally applied due to
the ethical and practical difficulties in accounting for intangibles
aspects of, for example, the casualties resulting from structural
collapse.
Therefore, if the performance is expressed by the fulfillment of a
limit state quantified in terms of EDP, the whole procedure can be
simplified. The structural risk is thus evaluated as the probability
of exceedance given by:
Z Z Z



Gedp Gedp jIM; SP; IP  f IPjIM; SP  f IM
 f SP  dIP  dIM  dSP
3. Design alternatives for wind actions on towers
International design standards contain different, alternative
procedures for calculating wind loads on structures. They follow
a general framework that usually begins with definition of the risk
category for a given structure. Different mean recurrence intervals
are associated to each risk category, leading to the selection of a
proper basic wind speed according to the site of the facility. Based
on wind speed, wind loads are estimated incorporating factors that
account for wind exposure and topographic features. Once the
wind loads are defined, material codes provide detailed prescrip-
tive requirements which should be complied with.
Although similarities can be identified among wind design
codes, each one follows more or less complex, specific calculation
procedures. As Lungu, Gelder and Trandafir [29] pointed out, in
their comparative study of Eurocode 1, ISO DIS 4354 and ASCE
795, each employs unique definitions of wind field characteris-
tics. The main differences between such documents include:
 the mathematical law used to describe the mean wind speed
profile in the atmospheric boundary layer (logarithmic law or
power law);
 the different averaging time intervals used to determine the ref-
erence or basic wind speed;
 the power spectral density functions of Davenport, Solari and
von Karman for along-wind gustiness, related to wind
turbulence.
Regarding the wind speed, similar conclusions were drawn by
Zhou, Kijewski and Kareem [30]. Comparing five major interna-
tional codes, they noticed that the wind-speed profile is closely
551
connected to the reference averaging time. Shorter averaging times
(e.g. the 3 s gust speed provided by ASCE 798) have flatter distri-
butions than those based on 1 h mean wind velocity (like the Cana-
dian and Australian standards). This leads to a large scatter in the
predicted values of the wind load effects.
Other studies related to the lack of harmonization and align-
ment of wind codes and standards have similar findings [31,32].
However, the coefficients of variation for both along-wind and
cross-wind responses of high-rise building were surprisingly small
(in the range of 1418%), considering the complexity of the calcu-
lations involved. The most plausible cause points to the fact that
many procedures are inter-related and have a common source of
origin.
As reported in [33], several international standards (e.g. NBC
and ASCE 7-95) employ the Gust Response Factor (GRF) method
suggested by Davenport [34], with slight modifications to deter-
mine the dynamic response of slender structures subjected to wind
action. The method uses a statistical approach using influence
lines, in which the total wind response is formed by a mean and
a fluctuating component.
The Brazilian wind code, NBR6123:88 [35], is also based on the
method of Davenport, but has minor differences in the parameters
that define the static wind loads. In addition, the structural vibra-
tion (in its natural modes) takes place around the deformed posi-
tion defined by the pressures caused by the static wind
component (mean speed).
NBR6123:88 provides two alternative design approaches: the
equivalent static wind loads (ESWL) method, which indirectly con-
siders dynamic features; and the discrete dynamic method. Fol-
lowing, both models are briefly presented.
3.1. ESWL method
2
The procedure starts with calculation of the reference wind
speed V 0 , defined as the 3 s gust speed, determined in open terrain
exposure at an elevation of 10 m, with a mean return period of
50 years. The characteristic wind speed V k is then given by:
V k z V 0 S1 S2 z S3 m=s 3
where S1 accounts for topographic features; S2 depends on the wind
exposure of the structure; and S3 considers the safety level required
for the structure during its life-time.
The S2 factor applies only to the ESWL method and describes the
profile of the mean wind speed according to a power law:
S2 z b C g z=10p 4
in which z is the height above ground, b and p are constants that
depend on terrain category and building class; and C g is a gust fac-
tor that equals 1.0 for open terrains and class A buildings.
With V k , one calculates the wind dynamic pressure q:
1
qz q V 2 z N=m2  5
2 a k
with air density qa being taken as 1.225 kg/m3 (at NTP).
Before proceeding, it is necessary to divide the structure of the
tower into panels or modules (usually defined by the region com-
prised between consecutive horizontal bracing members), and
determine the aspect ratio of each module. The aspect ratio / is
the ratio between the total frontal area of all individual members
(effective area), Ae , and the frontal envelope area (contour area), Ac .
The longitudinal drag forces F A on the towers modules are then
determined by the product of the dynamic pressure, the effective
area and the drag coefficient C d , which is a function of the aspect
ratio:
F A Ae  C d /  q N 6
552 R.K. Tessari et al. / Engineering Structures 143 (2017) 549557
Besides the forces acting directly on the tower, the wind load on
additional structures such as ladders, platforms, cables and anten-
nas must be estimated. NBR6123:88 does not specify how these
additional forces should be determined, but the internal codes of
the former Brazilian telecommunications company specify guideli-
nes for determining wind effect on accessories. According to TELE-
BRAS codes [36,37], wind forces acting on accessories are simply
added to those acting on the tower, without considering any
mutual interference. Such common approach has the benefit of
simplicity, but raises some criticism [38,39].
Once the wind loads acting on each module and its accessories
have been calculated, the resulting force must be distributed
according to the panel height and the wind incidence angle (a).
The decomposition process varies according to tower geometry,
with square and triangular cross-sections being the most common
shapes for latticed steel towers. Squared section towers have only
two critical wind incidence directions, while triangular ones are
more sensitive to a. Therefore, the decomposition process is illus-
trated herein for a tetrahedral module. As shown in Fig. 1, F T is the
total drag force acting on the tower module, with an incidence
angle a. Half of F T is then applied to the upper nodes and half to
the lower ones, as recommended by [36]. If the module has a pyra-
midal shape, F T can be decomposed proportionally to the influence
area of each node. F T =2 is further decomposed into forces F X and F Y
acting on each vertex of the lattice section, aligned with x and y
axes. These forces, in turn, depend on f X a and f Y a, which
account, in an approximate way, for shielding caused by the lattice
structure itself [40].
3.2. Discrete dynamic method
An alternative procedure presented by NBR6123:88 is the dis-
crete dynamic method. It is recommended for very flexible slender
(especially weakly damped) structures which can develop signifi-
cant fluctuating responses due to wind gust. In this method, the
mean wind response F i is separated from the fluctuating response
F^i , caused by the turbulent wind. Thus, the total wind force is given
by:
F i F i F^i ; N
and the mean response expressed by:
 2p
2 zi
F i q0 b C Ai Ai N
zref
where: q0 12 qa V 2p is the dynamic pressure, given in [N/m2 ];
V p 0; 69 V 0 S1 S3 is the design wind speed, in [m/s]; C Ai ; Ai and zi
correspond to the drag coefficient, the effective area and the height
above ground of the tower module i, respectively; zref is the refer-
Fig. 1. Resulting drag force on each module vertex according to the panel height
and the wind incidence angle.
ence height, taken as 10 meters; b and p are tabulated coefficients
that depend exclusively on the terrain roughness.
The fluctuating component, in turn, is given by:
X
n  p
C Ai Ai zzi lm
i
ref
F^i q0 b n mi lm
2 i1
N 9
i X
n
mi lm 2
i
i1
where n is the dynamic amplification factor, mi is the lumped mass
of module i and lmi is the modal coordinate i corresponding to the
natural frequency of vibration m.
When more than one mode of vibration is retained in the solu-
tion (m > 1), the superposition of the fluctuating responses
th
obtained for the j mode (j 1; . . . ; m) must be computed using
the root mean square (rms) criteria:
v
uX
u m
F i F i t F^2j 10
j1
4. Numerical application of PBWE
As the need for cellular coverage increases, freestanding latticed
steel towers to support cellular and microwave antennas become
omni-present. A large number of such towers is located in highly
populated urban areas, raising concerns about the possible conse-
quences of structural collapse. Due to the lightweight of these
structures, and the rare occurrence of earthquakes in Brazil, wind
forces are the primary concern in the design. In this context,
telecommunication towers are an ideal case-study for implemen-
tation of the PBWE methodology. Further advantage is the low
interference of other accidental loadings. Hereafter, the procedure
summarized in Section 2 is applied to the risk assessment of a
telecommunication tower, in the framework of PBWE.
7
4.1. Structural model
The numerical case-study example considered herein is a free-
8
standing latticed steel tower, used to support cellular antennas.
The geometry of the structure, the position of the antennas and
other accessories, and the self-weight of the whole system were
supplied by a manufacturing company.
The tower is located in Florianpolis, SC, Brazil. It has a total
height of 50 meters and triangular cross-section. It is formed by
two distinct segments: the bottom part (up to 42 m) forms a pyra-
mid frustum with an opening of 5.70 m at its base and 1.50 m at
the top; the upper part of the tower is 6.0 m high and has parallel
legs. The legs were made with omega shape profiles and the
bracing members with symmetric single angles, both in hot-
rolled ASTM A36 steel. The tower is divided into 26 sections, each
comprehending the region between consecutive horizontal
bracings.
The structural analyses were carried out on a 3D finite element
(FE) model, consisting of 267 nodes and 798 truss elements. In
total, 13 different cross-sections were used to discretize the bars,
in correspondence to the original design. Additional structures
were replaced by nodal loads.
The plan view of the tower with a schematic division of sections
and the layout of the antennas is shown in Fig. 2, along with the
implemented three-dimensional model.
 R.K. Tessari et al. / Engineering Structures 143 (2017) 549557 553

where E and F y are the Young modulus and the yield strength of the
steel; A; Imin e Lef represent the cross-section area, the minimum
moment of inertia and the effective bar length; and Nbar is the nor-
mal force in the bracing member.
In fact, for low slenderness (i.e., short members), plastic buck-
ling can control the behavior of the bar. The interval for which
inelastic buckling shall be considered is defined to be
p
Lef =r 6 p 2E=F y , according to ASCE 10:97 [41]. In the present
work, this corresponds to a limiting value of 127. As almost all
members of the structural model have a higher slenderness, only
elastic buckling was considered.
Finally, the low-performance level is related to global stability
of the tower. It is assumed that global instability occurs if the buck-
ling or tensile stresses in a leg member are exceeded. The limit
state equations are similar to Eqs. (12) and (13), but N bar now rep-
resents the normal force in leg members:
Unfortunately, we could not find proper references to quantify
the costs associated to loss of performance. Hence, our study does
not include the DM integral in Eq. (1). The problem is solved in
terms of the probabilities associated to the three performance
levels described in this section. Eq. (2) is solved numerically by
means of common reliability analysis.

5.2. Uncertainty parameters

5.2.1. Wind field parameters

Fig. 2. Three-dimensional and plan views of the tower.
5. Structural performance analysis
5.1. Performance requirements
The structural performance of the antenna-carrying tower is
divided in discrete levels. Hence, for computational purposes, the
EDP integral in Eq. (2) is replaced by a sum over three levels, as
follows.
The full operability of the antennas is characterized by the
allowable deflection of 1 400 0000 , according to [36]. This high-
performance level characterizes a service failure, and is written
in terms of the lateral displacement at the top of tower, repre-
sented by d:
g 1 X d  dX 1:45 m  dX 0
In the numerical analyses, the deterministic unpredictability of
both the magnitude and incidence direction of mean wind speeds
have been considered, as well as the roughness length of the
terrain.
Wind speeds (V 0 ) associated to a given mean return period are
nominal values. However, the maximum wind speed V max to be
observed in a fixed period (design live) is a random variable, whose
statistics are evaluated in reference to V 0 . In this work, discrete
time periods, varying from 1 to 50 years, were considered. Table 1
shows mean values and coefficients of variation (COV) of the max-
imum wind speeds associated to different return periods. These
statistics were calculated from wind speed records available in
the Wolfram Research database [42], corresponding to the city of
Florianpolis, SC (SBFL Station, 27.58 S 48.56 W). The data com-
prises a period of 42 years, extending from 1973 to 2015. Maxi-
mum wind speeds were found to follow Gumbel (Type I)
extreme value distributions.
The wind incidence angle a with respect to the global x-axis
(counted counterclockwise) is characterized by a uniform distribu-
tion, with lower and upper bounds equal to p=6 and p=2.
The variability of the roughness length z0 is indirectly consid-
ered by the exponent p of the power law that describes the mean
wind speed profile Eqs. (4), (8), and (9). It is modeled by a non-
dimensional Gaussian random variable with unitary mean and
COV equal to 0.047. This random variable becomes dimensional
by multiplication by the nominal value recommended in
NBR6123:88, according to the adopted wind model: 0.125 in the
static case; 0.23 in the dynamic case. More details can be found
11 in [40].

An intermediate performance level is given by yielding or buck-

ling of any diagonal or horizontal bracing members. This is also Table 1
considered a service failure, requiring maintenance with the even- Statistical data of V max according to different mean return periods.
tual replacement of individual failed bars. This performance level is Variable Mean COV
expressed by:
V1 25.86 m/s 0.256
g 2 X Fy A  Nbar yielding or 12 V2 27.75 m/s 0.238
V 10 37.46 m/s 0.177
p2 EImin V 25 42.35 m/s 0.156
g 2 X Nbar buckling; 13 V 50 45.98 m/s 0.144
L2ef
554 R.K. Tessari et al. / Engineering Structures 143 (2017) 549557

5.2.2. Structural characterization

Two mechanical properties of the tower are considered as ran-
dom elements of the SP vector: the Young modulus E, assumed to
be Gaussian with mean lE =197.4 GPa and COV = 0.076; and the
yield stress F y , characterized by a Lognormal distribution, mean
lFy = 277.5 MPa and COV = 0.068 [43].
In addition, a dead load factor, D, was also incorporated in the
analysis to account for possible deviations in estimating the struc-
tural self-weight. According to [44], D is modeled by a Normal dis-
tribution, with biased mean of 1.05 and COV equal to 10%.

5.2.3. Windstructure interaction

The last primary source of uncertainties is related with the
physical mechanisms (aerodynamic and aeroelastic phenomena)
which arise when the structure is immersed in the wind field.
The gust factor C g was chosen as the only random component of
the interaction parameter vector. It is normally distributed, with
nominal mean equal to 0.98 (as given by NBR6123:88) and COV Fig. 3. Reliability indexes for incidence angle of 30.
of 0.125 [45]. It reflects the epistemic uncertainty involved in the
conversion of dynamic wind effects into static equivalents, in the
ESWL method.

6. Numerical results

A set of 40 scenarios were analyzed, combining the two wind

load models described in Section 3, five mean wind speed values,
and four different assumptions w.r.t. wind incidence angle: first,
a was considered random; latter, the deterministic values of 30,
60 and 90 (critical directions for triangular section towers) were
assumed. Table 2 summarizes the considered scenarios.
For each scenario, reliability index and corresponding failure
probability of the structure were estimated using the First Order
Reliability Method (FORM) and crude and importance sampling
Monte Carlo simulation techniques (both combined with Latin
Hypercube stratified sampling). Twenty numerical analyses were
carried out for each load model, each composed of 15104 crude
Monte Carlo and 3104 importance sampling Monte Carlo simula-
tions. Sensitivity analyses were also conducted to assess the most Fig. 4. Reliability indexes for incidence angle of 60.
important variables to engineering problems involving wind action
on steel towers.

6.1. Results

Results of the numerical simulations are reported in Figs. 36.

The reliability index b is plotted against maximum wind speeds,
according to wind incidence angles. In the graphics, SLS1 refers
to the deflection serviceability limit state, SLS2 to the SLS related
to buckling and yielding of bracing members, and ULS refers to
the ultimate limit state. Distinction is also made between the wind
load models analyzed: the static (STA) model is represented by
gray curves, while the dynamic model (DYN) is indicated by black
curves. The output of each reliability method is indicated by the
following symbols: () FORM, (+) Crude Monte Carlo and ()
Importance Monte Carlo.

Table 2
Numerical simulation scenarios.

Load model Wind max. speed Incidence angle Fig. 5. Reliability indexes for incidence angle of 90.

V1
Static V2 a 30
V 10 a 60 Since the towers design was based on a basic wind speed of
Dynamic V 25 a 90 30 m/s for a return period of 50 years (much lower than the nearly
V 50 a = R.V.
46 m/s used in this study), the estimated failure probabilities are
 R.K. Tessari et al. / Engineering Structures 143 (2017) 549557
Fig. 6. Reliability indexes for random incidence angle.
high for a civil structure. However, the obtained data supports the
comparisons drawn.
6.1.1. Discussion of wind load models
Results presented in Figs. 36 show that the static load model is
more conservative, leading to lower reliability indexes, especially
for the high-performance requirement (deflection limit state). This
means that structures designed according to the ESWL method, in
the deflection limit state, will be generally more conservative,
comparing to those designed by the dynamic wind load model.
However, the difference in results reduces for lower performance
levels. Hence, it may be concluded that both NBR6123:88 proce-
dures lead to structures of similar safety levels.
6.1.2. Discussion of wind directionality
From Figs. 36, one observes that the low performance level
was the only one to show large variability depending on the inci-
dent wind direction. As expected, the angle of 90 (Fig. 5) proved
to be the most critical to structural stability, presenting the lowest
reliability levels amongst all cases. However, the assumption that
the wind component with the highest speed blows from the most
unfavorable direction is shown to be quite conservative: from
Fig. 6, which admits a random incidence direction, it can be seen
that the calculated reliability indexes resemble the case where
a 60 (Fig. 4).
The limit state SLS1 remained invariant w.r.t. the wind inci-
dence direction, since calculation of the peak displacement at the
top of the tower was defined as the vector sum of the translations
of each degree of freedom of the top nodes.
For SLS2, the instability of either element types masked the
importance of a, since no distinction was made between buckling
or yielding of horizontal or diagonal bracing bars.
6.1.3. Discussion of largest wind speed
Figs. 36 show that there is an approximately linear relation-
ship between reliability indexes and demand wind speeds. This
implies that b indexes for different wind speeds could be estimated
from a simple (almost linear) regression.
6.1.4. Discussion of performance levels
The last step of the Performance-based Design procedure corre-
sponds to structural design optimization. This can be attained by
minimizing risk or maximizing a utility function tied to the build-
ing performance. For structural engineering systems, (reduced)
cost is a frequently used measure of performance. Hence, maximiz-
555
ing performance may sometimes be equivalent to minimizing costs
and the optimal structure will be the one that minimize all costs
over its life-cycle (costs of construction, operation, inspection,
maintenance, and disposal), including expected costs of failure
[46,47].
The discussion in this session is limited to the balance between
expected failure costs for different performance measures. As fail-
ure costs can vary substantially depending on the location of the
tower (e.g., densely populated urban areas vs. uninhabited rural
areas) and the size of the structure, the proportionality between
ultimate and service failure costs is problem-specific. As men-
tioned above, no specific references to quantify expected costs of
failure for this communications tower were found. Assume, how-
ever, the cost of ultimate failure to be 25 times higher than the cost
for a bracing element service failure (SLS2). If reliability indexes for
these two performance levels were of similar magnitude, then
similar expected failure costs would be obtained for inversely
proportional return periods. Consider, for instance, the mean
return period of two years for a SLS2 service failure, and the mean
50-year return period for the ultimate failure, for a random inci-
dent angle (Fig. 7). By comparing the reliability indexes estimated
for these performance levels, it is evident that the probability of
operational failures is much lower than the probability of
structural collapse (b2yr 50yr
SLS2 2:46 and bULS 1:59). Hence, the ULS
dominates expected costs of failure. This situation is typical of
structures designed in accordance to ultimate limit states, follow-
ing normative prescriptions. Such difference allows concluding
that it could be cheaper to reduce the cross-sectional area of brac-
ing bars, in order to attain a better design (lighter structure with
more balanced expected costs of failure).
6.1.5. Sensitivity analysis
Through the use of FORM, a sensitivity analysis is conducted, in
order to investigate propagation of the uncertainties affecting the
input parameters (IMs, SPs and IP) to the structural response.
Figs. 811 show the sensitivity indexes for the most significant
variables, for each analysis scenario (remaining variables are omit-
ted for clarity). It is clear that the relative importance of the vari-
ables remained almost constant over the various scenarios for
wind incidence direction.
A preponderance of the maximum wind speed over the other
variables can be observed, especially for the dynamical load model.
In the static model, the gust factor appears with growing impor-
tance, as the wind speed increases, becoming the second parame-
ter of greater contribution to failure probabilities. Uncertainties in
Incidence Angle (=RV)
3.3
Reliability index,
2.6
2.46
1.9
1.59
1.2
0.5
V2 V50
24 30 36 42 48
Wind speed, m/s
SLS2 - STA SLS2 - DYN
ULS - STA ULS - DYN
Fig. 7. Comparison between the SLS2 reliability index for a period of 2 years (V 2 )
with the ULS value for a period of 50 years (V 50 ) when aw =RV.
556 R.K. Tessari et al. / Engineering Structures 143 (2017) 549557

Fig. 8. Random parameter sensitivity for the static (left) and dynamic (right) load
models. Wind incidence direction equal to 30.
Fig. 11. Random parameter sensitivity for the static (left) and dynamic (right) load
models. Wind incidence direction taken as aleatory.

100 %
11.7 12.4 18.3 23.0
80 27.3
6.4
8.1
60

40 83.9 82.7
74.3 67.6 61.1
20

0
V1 V2 V10 V25 V50
Speed (V) Angle () Gust (Cg)
(a) ESWL method

Fig. 9. Random parameter sensitivity for the static (left) and dynamic (right) load 100 %
models. Wind incidence direction equal to 60. 7.0 9.4
80

60
95.2 94.7 91.9 89.3
40 86.0

20

0
V1 V2 V10 V25 V50
Speed (V) Angle ()
(b) Discrete dynamic method
Fig. 12. Random parameters sensitivity for the low performance level and wind
incidence direction taken as aleatory.

Fig. 10. Random parameter sensitivity for the static (left) and dynamic (right) load
models. Wind incidence direction equal to 90.
Looking closely at Fig. 11, the findings about the wind direction-
ality become apparent, since only the ultimate limit state is found
to be sensitive to changes in a. Fig. 12 exhibits the sensitivity
indexes based solely on ULS, highlighting the growing importance
of the a angle, as the maximum wind speed increases.

the steel yield strength F y , the dead load factor D and the exponent
p resulted practically irrelevant, allowing the replacement of these
random variables by their means. The Young modulus had a rela-
tively low importance, exhibiting sensitivity indexes in the range
of 1.55.7%.
7. Concluding remarks
In this study, the PBWE methodology was applied to the prob-
abilistic assessment of the performance of steel towers. Two differ-
ent models to determine the wind action on latticed towers were
 R.K. Tessari et al. / Engineering Structures 143 (2017) 549557
considered: a static and a dynamic model. Through a case study
involving a telecommunications tower, it was found that both
models lead to very similar safety levels. The assumption that wind
always blows from the most unfavorable direction proved to be
excessively conservative. Structural parameters SP such as steel
yield strength, dead load factor and exponent p, frequently mod-
eled as stochastic, showed little contribution to failure probabili-
ties, given the dominance of wind field parameters.
The case study presented herein allowed us to illustrate the
potential of the PBWE probabilistic approach for the design of steel
towers. By comparing the reliability index associated with differ-
ent hazard intensities, it was shown how specific design strategies
can be drawn, in order to optimize structural performance. By
associating different performance levels to different mean return
periods, and by requiring similar failure probabilities, one can
avoid dominance of one performance measure over the others, in
terms of consequences of failure.
Acknowledgments
The authors kindly acknowledge financial support of this
research project by Brazilian agencies CNPq (National Council for
Research and Development) and CAPES (National Council for
Post-Graduate Education).
References
[1] Augusti G, Ciampoli M. Performance-based design in risk assessment and
reduction. Probab Eng Mech 2008;23(4):496508. http://dx.doi.org/10.1016/j.
probengmech.2008.01.007. URL <GotoISI>://WOS:000259894400017.
[2] Ellingwood BR. Proability-based codified design: past accomplishments and
future challenges. Struct Saf 1994;13(3):15976. http://dx.doi.org/10.1016/
0167-4730(94)90024-8. URL <GotoISI>://WOS:A1994NE27700003.
[3] Hendawi S, Frangopol DM. System reliability and redundancy in structural
design and evaluation. Struct Saf 1994;16(12):4771. http://dx.doi.org/
10.1016/0167-4730(94)00027-n. URL <GotoISI>://WOS:A1994PQ24900006.
[4] Ellingwood BR, Tekie PB. Wind load statistics for probability-based structural
design. J Struct Eng ASCE 1999;125(4):45363. http://dx.doi.org/10.1061/
(asce)0733-9445(1999)125:4(453). URL <GotoISI>://WOS:000079280500011.
[5] Zhang H, Ellingwood BR, Rasmussen KJR. System reliabilities in steel structural
frame design by inelastic analysis. Eng Struct 2014;81:3418.
[6] Blue Book, Sacramento, CA, 1999.
[7] Ghobarah A. Performance-based design in earthquake engineering:
state of development. Eng Struct 2001;23(8):87884. URL <GotoISI>://
WOS:000169055000001.
[8] Bertero RD, Bertero VV. Performance-based seismic engineering: the need for a
reliable conceptual comprehensive approach. Earthquake Eng Struct Dyn
2002;31(3):62752. http://dx.doi.org/10.1002/eqe.146.
[9] Lee T-H, Mosalam KM, Probabilistic seismic evaluation of reinforced concrete
structural components and systems, PEER Report 2006/04.
[10] Aktan AE, Ellingwood BR, Kehoe B. Performance-based engineering of
constructed systems 1. J Struct Eng 2007;133(3):31123.
[11] Federal Emergency Management Agency, Washington, D.C., FEMA-273. NEHRP
guidelines for the seismic rehabilitation of buildings; 1997.
[12] USGS, Significant earthquakes archives. earthquake hazards program.
apresenta informaes acerca dos terremotos mais significativos ocorridos
at o presente momento. disponvel em: http://earthquake.usgs.gov/
earthquakes/eqarchives/significant/; 2014.
[13] Porter KA. An overview of peers performance-based earthquake engineering
methodology. Appl Stat Probab Civil Eng 2003;1 and 2:97380. URL
<GotoISI>://WOS:000189453300130.
[14] Gunay S, Mosalam KM. Peer performance-based earthquake engineering met-
hodology, revisited. J Earthquake Eng 2013;17(6):82958. http://dx.doi.org/
10.1080/13632469.2013.787377. URL <GotoISI>://WOS:000320571400003.
[15] Robertson I, Riggs HR, Development of performance based tsunami
engineering; 2011. URL https://nees.org/resources/1861.
[16] Barbato M, Petrini F, Ciampoli M. A preliminary proposal for a probabilistic
performance-based hurricane engineering framework. Structures Congress
2011 Proceedings of the 2011 Structures Congress. p. 161829.
[17] Barbato M, Petrini F, Unnikrishnan VU, Ciampoli M. Performance-based
hurricane engineering (pbhe) framework. Struct Saf 2013;45:2435.
[18] Paulotto C, Ciampoli M, Augusti G. Some proposals for a first step towards a
performance based wind engineering. In: Proceedings of the IFED international
forum in engineering decision making; First Forum, December, 59, 2004.
[19] Petrini F. A probabilistic approach to performance-based wind engineering
(pbwe). Dottorato di ricerca: Universit degli Studi di Roma La Sapienza,
Rome, Italy; 2009.
557
[20] Ciampoli M, Petrini F, Augusti G. Performance-based wind engineering:
towards a general procedure. Struct Saf 2011;33(6):36778.
[21] Ciampoli M, Petrini F. Performance-based aeolian risk assessment and
reduction for tall buildings. Probab Eng Mech 2012;28:7584.
[22] Petrini F, Gkoumas K, Bontempi F, Damage and loss evaluation in the
performance-based wind engineering. In: Safety, Reliability, Risk and Life-
Cycle Performance of Structures and Infrastructures Proceedings of the 11th
International Conference on Structural Safety and Reliability, ICOSSAR 2013,
2013. pp. 17911797.
[23] Griffis L, Patel V, Muthukumar S, Baldava S. A framework for performance-
based wind engineering. In: Advances in Hurricane Engineering: Learning
from Our Past - Proceedings of the 2012 ATC and SEI Conference on Advances
in Hurricane Engineering, 2013. pp. 12051216.
[24] Ciampoli M, Petrini F. Performance-based design of structures under aeolian
hazard. In: Proceedings of the 11th International Conference on Applications
of Statistics and Probability in Civil Engineering, 2011. pp. 899906.
[25] Petrini F, Ciampoli M. Performance-based wind design of tall buildings. Struct
Infrastruct Eng 2012;8(10):95466.
[26] Petrini F, Ciampoli M, Augusti G. The role of uncertainties in aeolian risk
assessment. Computational Methods in Stochastic Dynamics, Vol. 22 of
Computational Methods in Applied Sciences. Springer; 2011. p. 187208.
http://dx.doi.org/10.1007/978-90-481-9987-7.
[27] Beck AT, Kougioumtzoglou IA, dos Santos KRM. Optimal performance-based
design of non-linear stochastic dynamical rc structures subject to stationary
wind excitation. Eng Struct 2014;78:14553. http://dx.doi.org/10.1016/j.
engstruct.2014.07.047. URL http://www.sciencedirect.com/science/article/pii/
S0141029614004684.
[28] Li G, Hu H. Risk design optimization using many-objective evolutionary
algorithm with application to performance-based wind engineering of tall
buildings. Struct Saf 2014;48:114.
[29] Lungu D, Gelder PV, Trandafir R. Comparative study of eurocode 1, iso and asce
procedures for calculating wind loads; 1996.
[30] Zhou Y, Kijewski T, Kareem A. Along-wind load effects on tall buildings:
Comparative study of major international codes and standards. J Struct Eng
ASCE 2002;128(6):78896. http://dx.doi.org/10.1061/(asce)0733-9445(2002)
128:6(788). URL <GotoISI>://WOS:000175745800012.
[31] Holmes JD. Developments in codification of wind loads in the asia pacific. In:
7th Asia-Pacific Conference on Wind Engineering, APCWE-VII; 2009.
[32] Holmes J, Tamura Y, Krishna P, Comparison of wind loads calculated by fifteen
different codes and standards, for low, medium and high-rise buildings. In:
11th Americas Conference on Wind Engineering; 2009.
[33] Merce RN, Guimares MJR, Doz GN, Brito JLV. Anlise de torres metlicas
submetidas ao do vento: um estudo comparativo. Revista Sul-Americana
de Engenharia Estrutural 2007;4(1):6181.
[34] Davenport A. The response of slender structures to wind. Wind Climate in
Cities, Vol. 277 of NATO Advanced Study Institute Series. Springer; 1995. p.
20939. http://dx.doi.org/10.1007/978-94-017-3686-2.
[35] Associao Brasileira de Normas Tcnicas, Rio de Janeiro, NBR 6123. Foras
devidas ao vento em edificaes; 1988.
[36] Telecomunicaes Brasileiras S/A, Prtica 240410-600. Procedimentos de
projeto para torres metlicas auto-suportadas, estaiadas e postes metlicos;
1997.
[37] Telecomunicaes Brasileiras S/A, Prtica 240400-702. Especificaes gerais
para adoo de parmetros bsicos e apresentao de memorial de clculo
para torres e postes metlicos; 1997.
[38] Holmes JD, Banks R, Roberts G. Drag and aerodynamic interference on
microwave dish antennas and their supporting towers. J Wind Eng Ind
Aerodyn 1993;50:2639.
[39] Carril-Jr CF, Isyumov N, Brasil R. Experimental study of the wind forces on
rectangular latticed communication towers with antennas. J Wind Eng Ind
Aerodyn 2003;91(8):100722. http://dx.doi.org/10.1016/s0167-6105(03)
00049-7. URL <GotoISI>://WOS:000185717100004.
[40] Tessari RK. Performance-based design of steel towers subjected to wind
action. So Carlos, SP: University of So Paulo; 2016. Master thesis.
[41] American Society of Civil Engineers, Reston, VA, ASCE 10:97. Design of Latticed
Steel Transmission Structures; 1997.
[42] W. R. Inc., Mathematica 9.0 (2012). URL http://www.wolfram.com/.
[43] Hess PE, Bruchman D, Assakkaf IA, Ayyub BM. Uncertainties in material and
geometric strength and load variables. Naval Eng J 2002;114(2):13965.
http://dx.doi.org/10.1111/j.1559-3584.2002.tb00128.x. URL <GotoISI>://
WOS:000179964600009.
[44] Ellingwood B, Galambus TV, MacGregor JG, Cornell CA. Development of a
Probability Based Load Criteria for American National Standard
A58. Cambridge, MA: National Bureau of Standards; 1980.
[45] Joint Committee on Structural Safety, Denmark, JCSS Probabilistic model code;
2001. URL http://www.jcss.ethz.ch/.
[46] Beck AT, de Santana Gomes WJ. A comparison of deterministic, reliability-
based and risk-based structural optimization under uncertainty. Probab Eng
Mech 2012;28:1829.
[47] Beck AT, Gomes WJ, Lopez RH, Miguel LF. A comparison between robust and
risk-based optimization under uncertainty. Struct Multidiscip Optim 2015;52
(3):47992.