Journal of Constructional Steel Research 133 (2017) 241–255

Contents lists available at ScienceDirect

Journal of Constructional Steel Research

Numerical simulation of steel I-shaped beams using a fiber-based

damage accumulation model
Melina Bosco, Lucia Tirca ⁎
Department of Building, Civil and Environmental Engineering, Concordia University Montreal, 1455 De Maisonneuve Blvd. West, H3G1M8, Montreal, Canada

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

Article history: This study proposes a fiber-based hinge damage accumulation model that is able to replicate the nonlinear re-
Received 20 July 2016 sponse of I-shaped beams of steel moment resisting frames. The model is developed in OpenSees and consists
Received in revised form 7 February 2017 of a beam with hinges element with fiber cross-section discretization within the plastic hinge zone. Among vari-
Accepted 14 February 2017
ous plastic hinge integration methods, the modified Gauss-Radau integration scheme was selected. The proposed
Available online xxxx
model incorporates strength and stiffness deterioration caused by flange local buckling of I-shaped beams which
Keywords:
is simulated by assigning a calibrated low-cycle fatigue material model to flange fibers. In this formulation, fa-
Steel beam tigue material uses a modified rainflow cycle counting algorithm to accumulate damage based on Miner's rule.
Damage accumulation model The values of fatigue material coefficients were calibrated against 16 experimental test results selected from
Plastic hinge the literature. An equation able to predict the fatigue ductility coefficient that follows a linear variation along
Cyclic loading the flange width is proposed based on regression analysis. In addition, a global damage index, DIs, defined as
Low-cycle fatigue the ratio between the number of fibers that reach fatigue and the number of fibers within the top and bottom
Damage index flanges of I-shaped cross-section, is developed and a global damage index value associated with the onset of
beam failure, labelled DIs(80%)prop is proposed. An application comprising a single-storey, one-bay steel MRF is car-
ried out in OpenSees, which validates the proposed beam model as computationally effective under cyclic quasi-
static and dynamic loading.
© 2017 Elsevier Ltd. All rights reserved.

1. Introduction
Performance-based seismic design requires evaluating the perfor-
mance of structures at different hazard levels. To assess the system be-
haviour from yielding to collapse, hysteretic models are needed to
replicate the response upon collapse. Cyclic loading triggers damage ac-
cumulation in members of structures designed to behave in the nonlin-
ear range, which may lead to member failure. To simulate this
behaviour, several damage accumulation models have been proposed
[1].
In the case of a steel moment resisting frame (MRF), the model for
beams and columns should be reliable, robust and computationally eco-
nomic during analysis [2]. To obtain accurate results, the selected hys-
teretic models need to incorporate all significant deterioration modes
observed during experimental tests. In the past, the plastic hinge
model for a steel MRF beam was based on an analytical method that in-
volved observing cross-sections of deformed beams within the plastic
hinge zone. More specifically, this method considered a plastic collapse
approach and a yield-line model [3]. To compute the stress distribution
in the buckled zone, the principle of virtual work was employed.
⁎ Corresponding author.
E-mail addresses: mbosco@dica.unict.it (M. Bosco), Lucia.Tirca@concordia.ca (L. Tirca).
However, this method did not provide practical solutions that could
be applied in a global analytical model for MRFs.
In general, plastic hinge models used in nonlinear dynamic analysis
were developed based on their physical and phenomenological charac-
teristics. According to [4], parameters employed to simulate models de-
veloped based on physical characteristics are defined in terms of
material properties, whereas parameters for phenomenological models
are defined based on the component response (e.g. an MRF beam). Con-
tinuum nonlinear models consider physics of assigned materials and do
not require definition of member stiffness, strength and deformation ca-
pacity because these are embedded in the definition of material proper-
ties. Concentrated hinge models and distributed plasticity models with
fiber-based cross-section discretization are defined by phenomenologi-
cal characteristics. Thus, the concentrated hinge model uses the overall
force-deformation response of the beam member while considering the
inelastic deformation rules associated with data obtained from experi-
mental test results. The distributed plasticity model with fiber-based
cross-section discretization embeds some behavioural assumptions
(e.g. Euler-Bernoulli bending theory) in association with explicit model-
ling of uniaxial material. However, these models are limited in their
ability to capture strength degradation. Therefore, there is a need to de-
velop analytical models that incorporate cyclic deterioration. For exam-
ple, local buckling of flanges and/or web, as well as lateral torsional

http://dx.doi.org/10.1016/j.jcsr.2017.02.020
0143-974X/© 2017 Elsevier Ltd. All rights reserved.
242 M. Bosco, L. Tirca / Journal of Constructional Steel Research 133 (2017) 241–255
buckling, lead to strength and stiffness deterioration of MRF beams. In
the past, deterioration models considering deterioration of geometric
and material properties by including different rate of cyclic deteriora-
tion in the two loading directions, as well as residual strength and ulti-
mate rotation, were specifically developed for nonlinear rotational
springs inserted in the model to replicate the behaviour of plastic
hinges. As such, it is the modified Ibarra-Krawinkler model [5–7],
which was based on the plastic hinge model reported in [8].
To account for cyclic deterioration, in the ATC 72-10 provisions [4]
there are four Options for analytical modelling of component (e.g.
beam) behaviour: (1) incorporate explicitly the cyclic deterioration,
(2) use skeleton curve as a modified backbone curve and ignore addi-
tional cyclic deteriorations, (3) use factors for modification of an initial
backbone curve, and (4) neglect strength deterioration. Among these
Options, it is believed that Option (1) is the most realistic, Options (2)
and (3) account for cyclic deterioration implicitly, and Option (4) does
not implicitly consider a deterioration model but instead recommends
a strict deformation limit associated with 80% of the capping strength
of a descending branch (this means a 20% strength loss).
To overcome these drawbacks, this study proposes a refined cyclic
deterioration beam model, able to simulate the strength and stiffness
degradation of I-shaped flanges caused by local buckling, as well as
the fracture mechanism caused by low-cycle fatigue. Although there
are several finite element platforms, in this study it was selected the
OpenSees framework (Open System for Earthquake Engineering Simu-
lation) [9] because it facilitates the simulation of a multi-storey building
response in the nonlinear range accurately and is computationally more
efficient than other computer programs (e.g. ABAQUS), when the struc-
ture is subjected to dynamic loads.
2. Nonlinear numerical models
In general, the existing nonlinear numerical models for MRF beams
are divided into three categories: i) an elastic beam model with attached
nonlinear rotational springs at its ends, ii) a nonlinear beam model with
distributed plasticity across the beam's length, and iii) a nonlinear beam
model with concentrated plasticity within a finite plastic hinge length.
In the first case, behaviour of the beam's plastic hinges is replicated
by nonlinear rotational springs that are generally made of modified
Ibarra-Krawinkler material [6], known as billin material in OpenSees.
The beam's hysteretic response, calibrated against experimental test re-
sults, embeds several modes of cyclic deterioration (basic strength,
post-capping strength, and unloading/reloading stiffness deterioration),
while the rate of cyclic deterioration is measured by the energy dissipat-
ed under cyclic loading. In addition, the strength bounds on the mono-
tonic backbone curve are established. This deterioration model,
reported in [6], complies with Option (1) of ATC 72-10 [4]. However,
the MRF beam model made of an elastic beam element with attached
nonlinear rotational springs at its ends has often showed numerical con-
vergence problems under dynamic loading. In general, these are related
to the high value of stiffness assigned to rotational springs in order to
preserve the elastic stiffness of the beam. Meanwhile, the elastic
deformability of the MRF beam is already taken into account within
the linear elastic response of the beam member. Further, as reported
in [10], the dynamic response is unreasonably sensitive to the assumed
value of the initial stiffness when nonlinear rotational springs are
employed and Rayleigh damping is considered.
In the second case, the MRF beam response is simulated using the
nonlinear force-based beam-column elements with distributed plastic-
ity and fiber-based cross-section discretization. Herein, plastic hinges
form at the location of maximum bending moment when fibers start
responding in the nonlinear range. However, the occurrence of strain
softening behaviour causes localization in the beam element because
the beam's deformation concentrates in a single integration point. Ac-
cording to [11], this phenomenon leads to a loss of objectivity because
the response depends on the number of integration points per element
rather than its length.
To overpass this drawback, the beam model made of a nonlinear
force-based beam-column element with finite-length plastic hinges lo-
cated at its ends, known as the beam with hinges element in OpenSees, is
considered. In this case, the beam is divided into three segments: the
central beam segment characterised by elastic behaviour, and the two
outermost finite length segments characterised by nonlinear behaviour.
Different integration methods were adopted for the beam with hinges el-
ement, which vary with the number and location of the integration
points. Further, the nonlinear behaviour of plastic hinges can be defined
by assigning either moment curvature relationships or fiber-based
cross-sections. To avoid the numerical convergence problems under dy-
namic loading that were encountered by analysing the MRF beam with
concentrated nonlinear rotational springs in OpenSees, Ribeiro et al. [2]
proposed to use the beam with hinges element with an assigned mo-
ment-curvature relationship. In addition, they developed a procedure
to define the above moment-curvature relationship starting from the
moment-rotation constitutive law derived against experimental test re-
sults [6]. This procedure was required because the moment-curvature
assigned to the plastic hinge could not be obtained by direct scaling of
the associated moment-rotation relationship [12].
When plastic hinges of the beam with hinges element are made of
fiber cross-sections, the strength and stiffness degradation is embedded
by assigning various uniaxial materials to fibers. For example, re-
searchers have assigned the Giuffrè-Menegotto-Pinto steel material to
plastic hinge fibers and assigned a negative value to the parameter con-
trolling the isotropic hardening in order to simulate degradation in the
moment-curvature cyclic response [13]. However, after the occurrence
of first yielding, the strength degradation was suddenly initiated,
while the capping bending moment and the associated rotation were
not reached. Recently, Bai et al. [14] proposed a beam with a fiber-
based plastic hinge cross-section model that is able to replicate the
strength and stiffness degradation caused by local buckling. In their
model developed for steel square hollow structural section beam mem-
bers, deterioration was explicitly incorporated in the stress-strain curve
and the cyclic response was validated against experimental test results.
However, this model was not verified for I-shaped beams. A similar ap-
proach was used by Kasai et al. [15] who simulated the strength and stiff-
ness deterioration localized at a column's base by assigning a stress-strain
response to fiber cross-sections inserted in the zero-length element at-
tached to the column end. The finite length of bucking zone and the rota-
tion characteristic of these fiber-based cross-sections were defined to
simulate the experimental buckling behaviour of steel struts. A degrada-
tion rule for the compressive buckling zone was also embedded [15].
Other studies have emphasized the complexity phenomena of struc-
tural collapse under seismic loading. To overpass this drawback, an ac-
curate numerical model should incorporate strength and stiffness
degradation due to local buckling, as well as the fracture mechanism
due to low-cycle fatigue [1]. The analytical model proposed in [16] com-
prises both aforementioned principles but the method developed is dif-
ficult to apply when analysing the entire MRF system.
Meanwhile, the strength and stiffness degradation of cross-sections
can be simulated in OpenSees by considering a fatigue material
wrapped around the uniaxial steel material assigned to the beam ele-
ment. This approach has been mainly used to simulate the brace frac-
ture of concentrically braced frame caused by low-cycle fatigue [17].
Herein, braces were made of elements with distributed plasticity and
fiber cross-section discretization. In addition, the same approach was
used to simulate the cyclic behaviour of long links of eccentrically
braced frames designed to respond in flexure [18]. However, the cyclic
behaviour of I-shaped links, modelled as explained above, has shown
a sudden reduction in strength and stiffness because both top and bot-
tom flanges have simultaneously reached the fatigue life due to the as-
sumption of plane sections and the omission of composite steel deck in
the model.
 M. Bosco, L. Tirca / Journal of Constructional Steel Research 133 (2017) 241–255
3. Proposed fiber-based damage accumulation model
3.1. Model description
To overcome the drawback related to non-objective response caused
by strain-softening behaviour observed in elements with distributed
plasticity [11], the proposed model uses the beam with hinges
element, whereas fiber-based cross-sections are assigned to hinge seg-
ments. Among various plastic hinge integration methods, the modified
Gauss-Radau integration scheme (known as HingeRadau in OpenSees)
was employed. As depicted in Fig. 1a, plasticity is confined to a single
integration point located at each end of the plastic hinge length. As illus-
trated in Fig. 1b, each flange of I-shaped cross-section is divided into nf,s
segments and nf,l layers, while the web is discretised into nw,l layers. The
Giuffrè-Menegotto-Pinto steel material known as Steel02 in OpenSees
was assigned to fibers. This material model is able to account for accu-
mulated plastic deformation at each point of load reversal. Accordingly,
each hysteretic loop follows the previous loading path for a new
reloading curve, while deformations accumulate. As reported in [3],
the I-shaped cross-section, commonly used for the MRF beams, starts
deteriorating at the location of maximum bending moment due to the
initiation of flange local buckling in the compression zone, which may
also induce web deformation. As expected, there is an interaction be-
tween flange buckling and web deformation. However, the formation
of local plastic mechanism at the level of flange in compression is a com-
plex phenomenon and it leads to the general shape of the plastic mech-
anism, while for the web the local plastic mechanism is sometimes
partially formed. To replicate the failure mechanism caused by low-
cycle fatigue under cyclic loading, a developed recalibrated fatigue ma-
terial assigned to flanges fibers is discussed in the following.
Fatigue material developed in OpenSees [9,19] uses accumulative
strain to predict damage in accordance with the Miner's rule:


n
nðεi Þ
DI ¼ ∑
j¼1 N f ðεi Þ j
Fig. 1. Beam model: a) beam with hinges element with modified Gauss Radau integration method and fiber-based hinges; b) fiber-based cross-section and distribution of fatigue ductility
coefficients.
243
where n(εi) is the number of cycles at the strain amplitude εi and Nf (εi)
is the number of cycles that material can sustain at that amplitude. Man-
son and Coffin [20,21] independently proposed a low-cycle fatigue rela-
tionship between the plastic strain amplitude experienced in each cycle,
εi, and the number of fatigue cycles to failure, Nf (εi). This relationship is
linear on the log-log domain of plastic strain amplitude versus the num-
ber of cycles to failure with a slope equal to m, as given by the following
equation:
 m
εi ¼ ε0 Nf ð2Þ
where ε0 and m are the fatigue ductility coefficient and fatigue ductility
exponent, respectively. To monitor when a fiber of the I-shaped beam
cross-section reaches its fatigue life, the modified cyclic counting meth-
od that records only the four most recent peaks of strain reversal at any
given time, was used in OpenSees to account for the accumulated dam-
age [19]. When the damage index of a fiber reaches 1.0, the fracture is
initiated and the stress in the fiber drops to zero.
The value suggested in the literature for the fatigue ductility
exponent, m, varies from −0.30 to −0.50 [17] in the case of HSS braces,
while for the link beam of eccentrically braced frames a value of −0.458
was used [18]. In the proposed OpenSees model, parameter m is set
equal to −0.5. The definition of ε0 is more complex because the
effect of local buckling implies a gradual stiffness and strength degrada-
tion. To simulate this effect, the fatigue ductility coefficient ε0 assigned
to I-shaped flange fibers has a linear distribution (Fig. 1b). More specif-
ically, a minimum value ε0,min was assigned to fibers of flange ends and
a maximum value (ε0,min + Δε0) to fibers located at the web-flange in-
tersection. Herein, ε0,min is related to the initiation of local buckling and
Δε0 controls the rate of degradation. It is noted that a higher value of Δε0
leads to gradual stiffness and strength degradation, while a value of Δε0
equal to zero corresponds to a sudden reduction of cross-section
strength.
The fiber-based damage accumulation model proposed to replicate
ð1Þ
the nonlinear behaviour of I-shaped MRF beams under cyclic loading
244 M. Bosco, L. Tirca / Journal of Constructional Steel Research 133 (2017) 241–255

is discussed further based on the following features: (i) the length of the
plastic hinge, (ii) the selection of beam with hinges element with modi-
fied Gauss-Radau integration method, and (iii) the predicted low-cycle
fatigue material coefficients calibrated against experimental tests se-
lected from the literature. To set up an accurate model that is able to re-
spond efficiently to the computational time criteria, a sensitivity
analysis is carried out.
Although the proposed model was developed for the OpenSees
framework, it can be implemented in other platforms where the beam
with finite length hinge elements, the type of fiber-based cross-section,
and the low-cyclic fatigue material are defined.
3.2. Experimental data selection and modelling
specimens, displacements are measured at the point of the applied force
F. The loading protocols used in the selected experimental tests are
given in Fig. 3.
The response of each specimen was simulated by a beam model
shown in Fig. 2b, which includes: (1) a rotational elastic spring for rep-
licating the flexural deformability of the setup, (2) a vertical elastic
spring that includes the shear deformability of the cantilever beam,
and (3) an elastic element which simulates the portion lc of the beam
in which stiffeners are introduced to connect the beam to the column.
The stiffness kv of the vertical elastic spring was calculated as follows:
GA
kv ¼ ð4Þ
χLv

Herein, 18 experimental tests reported in five different studies were where A is the area of the beam cross-section, G is the shear modulus of
selected and are given in Table 1. Among them, 16 experimental tests elasticity, and χ is the shear coefficient. The latter parameter is calculat-
[22–25] conducted on I-shaped MRF beams (W-shaped in the North ed by equating the internal and external virtual work of a beam subject-
America standards) were selected from the database developed by ed to the constant shear force V. If the shear stresses are calculated as
Lignos and Krawinkler [7], and two experimental tests carried out on proposed by Jourawsky, the shear coefficient [27] is given by:
MRF beams with European cross-sections (H-shaped) were selected " #
5 2
from [26]. It is noted that beam-column specimens with significant plastic b f ðd−t f Þ η2 bf
χ¼ 15 þ 10η þ 5η þ 2ξ ð5Þ
deformation in the panel zone, beams with reduced beam section, or with 240 A ρ4y ξ ðd−t f Þ
2

significant lateral-torsional buckling, or with web local buckling, are not

the objects of this study. The setup used in tests reported in [22–25] con-
sists of a single beam connected to a W-shaped column (Fig. 2a). The
boundaries of the specimens (column ends and beam end) were nomi-
nally free to rotate, and a lateral support was provided to the beam. The
setup adopted in [26] consists of a vertical beam element, whereas a hor-
izontal steel frame was used to restrain the lateral deformation of the
beam specimen in order to prevent the flexural-torsional buckling failure
mechanism. Quasi-static cyclic displacement loading was applied to all
selected specimens. The size of the I-shaped beam cross-section, the
yield stress obtained from the coupon test taken from the flange (Fyf)
and the web (Fyw), as well as the beam's effective length (Lv) are given
in Table 1. The effective length of the beam specimen is calculated as:
dc
Lv ¼ L− −lc
2
where L is the beam length measured to the centerline of the column, dc is
the depth of the column cross-section, and lc is the length of the connec-
tion between the beam's end and the column face. In the case of speci-
mens tested by Engelhardt and Sabol [22], the displacement δ is
measured at 122 in (3.098 m) from the face of the column. For all other
In Eq. (5), d is the depth of the cross-section, bf is the width of the
flange, tf and tw are the thickness of the flange and of the web respec-
tively, while ρy is the radius of gyration of the cross-section with respect
to its strong axis, ξ =tw/bf and η =2 tf/(d − tf).
3.3. Calibration of the plastic hinge length
To define the length of plastic hinges developed in a MRF beam under
cyclic lateral loading, a preliminary beam model made of force-based
nonlinear beam-column elements with distributed plasticity and fiber
cross-section discretization was considered. The beam was made of 16
nonlinear beam-columns elements with 4 integration points per element
as illustrated in Fig. 4. The fiber-based I-shaped beam cross-section com-
prises 30 flange segments (nf,s), 4 flange layers (nf,l), and 30 web layers
ð3Þ (nw,l) as illustrated in Fig. 1b. The Steel02 material was assigned to all fi-
bers. However, this model does not account for strength and stiffness de-
terioration. Referring to Steel02 material, the relation that provides the
normalised stress σ∗ as a function of normalised strain ε∗ is:
ð1−bÞε
σ  ¼ b ε þ 1
ð6Þ
ð1 þ εR ÞR

Table 1
Selected specimens and characteristics.

ID Specimens Cross-section Fyf (MPa) Fyw (MPa) Lv (mm)

#1 Engelhardt-Sabol-AISC-9410-S-5A [22] W36 × 150 301.3 361.3 2807

#2 Engelhardt-Sabol-AISC-9410-S-6A [22] W36 × 150 301.3 362.3 2845
#3 Engelhardt-Sabol-AISC-9410-S-6B [22] W36 × 150 357.1 367.5 2845
#4 Engelhardt-Sabol-AISC-9410-S-7A [22] W36 × 150 301.3 362.3 2985
#5 Engelhardt-Sabol-AISC-9410-S-7B [22] W36 × 150 301.3 362.3 2985
#6 Engelhardt-Sabol-AISC-9410-S-8A [22] W36 × 150 293.7 328.9 2985
#7 Engelhardt-Sabol-AISC-9410-S-8B [22] W36 × 150 293.7 328.9 2985
#8 Kim et al.-PEER-0007-RC01 [23] W30 × 99 368.9 427.5 3023
#9 Kim et al.-PEER-0007-RC02 [23] W30 × 99 368.9 427.5 3023
#10 Kim et al.-PEER-0007-RC03 [23] W30 × 99 368.9 427. 5 3023
#11 Kim et al.-PEER-0007-RC04 [23] W30 × 99 368.9 427. 5 3023
#12 Kim et al.-PEER-0007-RC08 [23] W30 × 99 368.9 427. 5 3023
#13 Tsai-Popov-EERC-88-19-S-10 [24] W18 × 40 350.3 384.7 1438
#14 Tsai-Popov-EERC-88-19-S-12 [24] W21 × 44 317.2 317.2 1604
#15 Uang-Latham-SMRF-95-03-S2 [25] W36 × 150 330.1 379.5 2464
#16 Uang-Latham-SMRF-95-03-S3 [25] W36 × 150 330.1 379.5 2464
#17 D'Aniello et al.-HEA160 [26] HEA160 337.0 338.0 1731
#18 D'Aniello et al.-HEB240 [26] HEB240 273.0 318.0 1633
 M. Bosco, L. Tirca / Journal of Constructional Steel Research 133 (2017) 241–255 245

Fig. 2. Experimental test: a) setup and b) modelling the test setup.

where b is the strain hardening ratio assumed equal to 0.0015. The pa- those characterised by a coordinate yj (local coordinate system) ranging
rameter R computed from Eq. (7) influences the shape of the transition in the following intervals:
curve and is considered to vary as a function of the plastic excursion ξpl
of the previous loading path in order to account for the Bauschinger effect. d −εy εy d
− ≤y j ≤ and ≤y j ≤ ð8Þ
2 χ χ 2
!
cR1 ξpl
R ¼ R0 1− ð7Þ Then, the length of plastic hinge is defined as the distance between
cR2 þ ξpl
the end of the beam and the point at which 10% of the depth of cross-
Herein, R0 is the value of parameter R during the first loading, while section fibers have yielded.
cR1 and cR2 are experimentally computed. In the proposed model, R0 is In the literature, the length of plastic hinge is often expressed as a
set equal to 18 and cR1, cR2 are equal to 0.925 and 0.25, respectively. function of: i) the wave length Lm of beam flange during local buckling,
Young's modulus for steel material, E, is set equal to 210,000 MPa. In ac- ii) the depth of cross-section, or iii) the shear length Lv. For a beam
cordance with experimental data, the initial yield stress given for flange under moment gradient, the buckling wave length of beam flange was
fibers is different than that assigned to web fibers of I-shaped cross-sec- computed by Lay [28] and is given as follows:
tion. Isotropic hardening is considered by assigning the following
Lm ¼ 2βc
values: 0.022, 1.00, 0.022 and 1.00 to the four parameters a1, a2, a3
3=4
1=4
and a4 that are required to define the Steel02 material [9]. tf d ð9Þ
with c ¼ 0:5 b f and β ¼ 0:6
The plastic hinge length of selected specimens was computed nu- tw c
merically. Vertical displacement loading defined for each experimental
However, tests reported in [26] showed that it is more accurate to
test was applied to the beam's end. To define the length of the plastic
express parameter c as a function of flange width and web thickness
hinge, the numerical response of the model was considered up to the
as per Eq. (10).
displacement δc, which corresponds to the maximum force recorded
during the test. An example conducted on the Specimen used in test c ¼ 0:5 ðb f −t w Þ ð10Þ
#4 (Table 1) is provided in Fig. 5. Herein, the solid line shows the hys- The value of plastic hinge length (Lex pl ) obtained from numerical
teresis loops upon maximum force was reached in the experimental model explained above and the analytical value Lm computed as per
test, as well as the loading protocol up to the associated value. With- Eq. (9) with c given in Eq. (10), are summarized in Table 2. In addition,
in this interval of loading protocol, the numerical model response the following dimensionless parameters are given in Table 2: Lpl -
ex
showed in Fig. 5b replicates accurately the experimental hysteresis normalised to both Lm and Lv, as well as, to the depth of the beam
loops depicted in Fig. 5a. However, in the replicated numerical cross-section d. Herein, the value of shear length Lv is equal to that of ef-
model, the magnitude of force increases with each cycle, as well as fective length given in Table 1 because the replicated specimens are can-
the length of the plastic hinge because no degradation was embed- tilever beams. By analysing the magnitude of the ratios Lex ex
pl /Lm, Lpl /Lv,
ex
ded in the model. and Lpl /d computed for each specimen and the associated COV given
The value of curvature χi, computed for the deformed member due in Table 2, it is concluded that the plastic hinge length is strongly corre-
to displacement δc is plotted for each integration point i in Fig. 6. The lated to the shear length Lv and weakly correlated (higher value of COV)
corresponding deformation within the cross-section is determined to the length of plastic hinge and depth of beam's cross-section, especial-
and compared with the yielding strain εy. Hence, the yielded fibers are ly when H-shaped profiles (European profiles) are considered. Based on
246 M. Bosco, L. Tirca / Journal of Constructional Steel Research 133 (2017) 241–255

Fig. 3. Loading protocols (units mm).

the obtained results, a predicted value for the plastic hinge length Lpl
- overlap the hysteresis loops resulting from the experimental tests.
pred
equal to αLv is proposed. By minimizing the sum of the square of
the differences obtained between the actual plastic hinge length and
the corresponding length computed as αLv, the coefficient α is equal
to 0.22, which is consistent with reference [29].
3.4. Calibration of low-cycle fatigue material coefficients
In this study, the demanded low-cycle fatigue ductility coefficients
(m and ε0) are calibrated against 18 experimental tests provided in
Table 1. The numerical model developed for each specimen was ex-
plained in Part 3.1 and the plastic hinge length (Lex
pl ), used herein, is
set as obtained from the numerical model when the tests were repli-
cated. Considering m equal to − 0.5, the calibration of ε0 was made
such that the hysteresis loops of numerical model (expressed in
terms of applied force F versus beam's deflection δ) are able to
This analysis leads to a linear variation of fatigue ductility coefficient,
and the matching values for ε0,min and Δε0 are listed in Table 2. In ad-
dition, the comparison between experimental and numerical hyster-
esis loops of Specimens #4, #12, #14, and #15 is depicted in Fig. 7. A
black line is used to represent the experimental test response, a grey
line shows the response of numerical model, and the red dashed line
depicts the skeleton curve resulting from the experimental test data.
This comparison proves that the proposed OpenSees beam model is
able to simulate accurately the stiffness and strength degradation
of considered specimens.
3.5. Sensitivity analysis
To develop a computational efficient MRF beam model, a sensitivity
analysis is carried out in order to optimize the number of segments

Fig. 4. Beam with distributed plasticity and 4 integration points per element.
 M. Bosco, L. Tirca / Journal of Constructional Steel Research 133 (2017) 241–255
Fig. 5. Response of Specimen #4: a) experimental hysteresis loops; b) numerical hysteresis loops without the consideration of strength and stiffness degradation (solid line shows the
response up to the achievement of the maximum force recorded in the test).
within the flange nf,s, the number of layers within the flange nf,l and the
number of layers within the web nw,l (Fig. 1b). Following an approach
that is commonly used in the literature [30], the number of flange seg-
ments nf,s and the number of flange layers nf,l, as well as the number
of web layers nw,l were selected independently of the cross-section
depth and thickness of the web and flanges. The reference beam
model is made of 30 flange segments (nf,s = 30), 4 layers per flange
(nf,l = 4) and 30 layers for the web (nw,l = 30). In this sensitivity anal-
ysis, nf,s varies from 5 to 30, nf,l from 4 to 7, and nw,l from 10 to 30.
Results from the sensitivity analysis are shown in Fig. 8 by means of
graphs in which the applied force F is related to displacement δ. For ex-
ample, only results obtained by replicating tests #4 [22] and #14 [24]
are shown herein. For Specimen #4, the fatigue ductility coefficients
ε0,min and Δε0 are set equal to 0.03 and 0.154 respectively, while for
Specimen #14 they are set as 0.03 and 0.10 respectively. As depicted
in Fig. 8a, when flanges are discretized in few segments (e.g. nf,s = 5
or nf,s = 10), the strength degradation is slightly larger than that obtain-
ed for nf,s = 20 and nf,s = 30. Instead, there is no difference in the re-
sponse when the number of flange segments is N20. In this regard,
nf,s = 20 is suggested to be the minimum number of flange segments.
Further, a variation of the number of web layers in the range of 10 to
30 does not affect the response (Fig. 8b). However, these results might
be questioned when fatigue material is also assigned to the web. The re-
sults depicted in Fig. 8c show no difference in the response when the
number of flange layers increases from 4 to 7. Thus, the minimum num-
ber of flange layers is proposed as 4.
Fig. 6. Evaluation of plastic hinge length: a) yielded region, b) curvatures.
247
3.6. Advantages of the proposed beam model vs. other beam models
To highlight the differences between the cyclic response from yield-
ing to failure of the proposed beam model (beam with hinges element
and fiber-based plastic hinge cross-sections) and that of a beam made
of 16 force-based beam-column elements with distributed plasticity
depicted in Fig. 4, a first comparison was carried out and the results
are discussed below. The second comparison was performed to show
the differences between the nonlinear response of the proposed beam
model and that of a similar beam with hinges and fiber-based cross-sec-
tions within the plastic hinge zone and two Points Gauss Radau integra-
tion scheme (named as HingeRadauTwo in OpenSees) as depicted in Fig.
9. It should be noted that the two integration points are included within
the plastic hinge length and are located at a distance equal to 1/4Lex
pl and
3/4Lexpl respectively [11], where the distance is measured from the
beam's end. Both aforementioned beam models used for comparison
purpose are able to experience strength and stiffness degradation be-
cause a calibrated fatigue material was assigned to their plastic hinge fi-
bers. The experimental test employed in this analysis corresponds to
Specimen #12 (Table 1) and it was selected because experimental
data in terms of strain was also provided [23].
The nonlinear behaviour of all OpenSees beam models was analysed
in terms of global and local response. For each beam model, the global
response is emphasized by the hysteresis loops expressed in terms of
force F versus drift angle as depicted in the first column of Fig. 10. Here-
in, the drift angle is calculated as the ratio between the specimen
248 M. Bosco, L. Tirca / Journal of Constructional Steel Research 133 (2017) 241–255

Table 2 are depicted up to a drift angle equal to 0.03 because at higher drift an-
Length of the plastic hinge and calibrated fatigue material coefficients. gles the strain gauges were over-ranged.
Specimen ID Lex
pl (mm) Lm (mm) Lex
pl /Lm Lex
pl /Lv Lex
pl /d ε0 Δε0 It can be concluded from analysis of the data shown in Fig. 10 that
#1 693.33 373.18 1.858 0.247 0.760 0.0350 0.140
the proposed beam model is able to accurately predict the strain in
#2 597.13 373.18 1.600 0.210 0.655 0.0300 0.150 the I-shaped flanges. Instead, the other two beam models used for
#3 493.79 373.18 1.323 0.174 0.542 0.0300 0.170 comparison purpose highly overestimate the strain magnitude
#4 617.33 373.18 1.654 0.207 0.677 0.0300 0.154 because of the yielding localization due to softening behaviour. For the
#5 681.24 373.18 1.826 0.228 0.747 0.0300 0.142
proposed beam model and the consideration of the largest drift angle
#6 675.67 373.18 1.811 0.226 0.741 0.0300 0.154
#7 677.53 373.18 1.816 0.227 0.743 0.0300 0.154 (δ/Lv = 0.03), the ratio ε/εy is equal to 6.7. For the same drift angle,
#8 691.63 287.43 2.406 0.229 0.917 0.0250 0.058 the ratio ε/εy is equal to 19 when the beam with hinges with the two
#9 745.32 287.43 2.593 0.247 0.988 0.0305 0.048 points Gauss Radau integration method is selected, and ε/εy = 100
#10 750.70 287.43 2.612 0.248 0.996 0.0250 0.050 when the beam with distributed plasticity model is considered.
#11 608.68 287.43 2.118 0.201 0.807 0.0250 0.050
#12 744.47 287.43 2.590 0.246 0.987 0.0270 0.049
#13 306.44 201.61 1.520 0.213 0.673 0.0290 0.150 4. Prediction and validation of fatigue ductility coefficients
#14 329.34 181.86 1.811 0.205 0.626 0.0300 0.100
#15 487.92 373.18 1.657 0.198 0.535 0.0290 0.160 To predict the fatigue ductility coefficients ε0,min and Δε0 assigned to
#16 441.24 373.18 1.483 0.179 0.484 0.0290 0.160
hinge fibers of the MRF beam model, analytical relations are defined as
#17 400.85 148.45 1.307 0.232 2.637 0.0470 0.150
#18 436.46 246.94 1.182 0.267 1.819 0.0750 0.195 follows. Among the considered experimental tests, the specimens made
COV 0.251 0.113 0.577 of European profiles HEA 160 and HEB 240 have been discharged be-
COVa 0.249 0.108 0.220 cause these cross-sections are typically used for columns. From data re-
a
Calculated without specimens #17 and#18. ported in Table 2, the required ε0,min coefficient varies from 0.025 to
0.035. Due to the low variation of ε0,min, the average value of 0.029 is
adopted for all investigated specimens. Meanwhile, the coefficient Δε0
resulted in the range of 0.048 to 0.17. To define analytically an equation
that is able to predict Δε0, all values resulting from the calibration are
displacement, δ and the shear length, Lv. The global response is used to plotted as a function of geometric properties of the beam component
calibrate the fatigue ductility coefficients (ε0,min and Δε0) assigned to and its cross-section.
flange fibers of each beam model. Then, the local response is analysed Theoretical and experimental studies conducted on the plastic rota-
in terms of strain measured at the first integration point. The second col- tion capacity of beams have shown that I-shaped beams characterised
umn of Fig. 10 shows the maximum value of strain recorded at the up- by smaller width to thickness ratio of flanges (compact cross-section)
permost fiber of the beam's flange versus the corresponding drift angle are able to reach the maximum strength at a larger end rotation and
measured at each load reversal. As depicted, the strain is normalised to show slower strength degradation after the maximum strength is
the yield strain, εy. Crosses, triangles and diamonds symbols pinpoint reached. Thus, Δε0 is expected to be strongly dependent on the slender-
the experimental values recorded at different locations along the top ness of the beam flange, λf, given in Eq. (11).
flange. Black, light grey, and dark grey dots are used to respectively
identify the response predicted by: the proposed beam model, the rffiffiffiffiffiffiffi
beam with hinges with two points Gauss Radau integration method, bf F yf
λf ¼ ð11Þ
and the beam with distributed plasticity model. The obtained results 2t f E

Fig. 7. Comparison between the experimental and numerical response with fatigue material.
 M. Bosco, L. Tirca / Journal of Constructional Steel Research 133 (2017) 241–255 249

Fig. 8. Sensitivity analysis: a) variation of number of segments within the flange, b) variation of number of layers within the web, c) variation of number of layers within the flange.

Further, for a laterally braced beam with a moment gradient, the mo-
ment at the critical section continues to increase because of strain hard-
ening, and the flange local buckling is delayed until the yielded length is
equal to a full local buckling wavelength. Thus, the beam tends to devel-
op a higher bending moment and rotation capacity than that expected
in the case of a beam with a uniform moment. To take into
account the effect of moment gradient, the dependence of Δε0 on the
ratio Lm/Lv is also investigated. The relationship between Δε0 and
other geometric beam characteristics, already considered elsewhere
[6,26], is also investigated. These additional parameters are: bf/Lv, bftf/
dLv, bfd/L2v , Lv/ρy, Af/A and λw. Herein, Af/A is the ratio of the area of
flanges to the gross area of the cross-section and λw is the slenderness
of the beam web, calculated as:
rffiffiffiffiffiffiffiffiffi
h F yw
λw ¼
tw E ent symbols are used to pinpoint parameters referring to experimental
where h is the clear distance between flanges (e.g. h = d − 2tf). Among
these parameters, Lv/ρy is sensitive to lateral-torsional buckling.
Plotting the demanded coefficient Δε0 against the selected I-shaped
beam parameters produces trends that are shown in Fig. 11. A different
symbol is used to pinpoint the associated experimental test values ob-
tained by different investigators. A regression line is included in the in-
dividual plot to illustrate the overall trends. The coefficient of
determination R2 is also reported in the figure. As expected: Δε0 de-
creases as the slenderness of the flange increases; and Δε0 shows low
correlation (lower values of R2) with the slenderness of the web and
the ratio Af/A although it is strongly correlated with ratios Lm/Lv and
bftf/dLv. Based on regression analysis, the following equation is proposed
to predict the Δε0value:
bf t f Lm
Δε0;pred ¼ 0:217−0:770 λ f þ 0:452 þ 0:902 ≥0 ð13Þ
d Lv Lv
A comparison between the calibrated (demanded) and predicted Δε0
value is shown in Fig. 12 for all selected specimens given in Table 1. Differ-
ð12Þ
tests carried out by several investigators. A good match between the cal-
ibrated and predicted Δε0 value is obtained.
To validate the proposed equation, the cyclic response of all selected
specimens is computed for the proposed beam model with a plastic
hinge length of 0.22 Lv and predicted value of fatigue material
250 M. Bosco, L. Tirca / Journal of Constructional Steel Research 133 (2017) 241–255

Fig. 9. Beam with hinges element with two Points Gauss Radau integration scheme.

coefficients: ε0,min = 0.029 and Δε0 = Δε0,pred. Since the predicted
values of ε0,min and Δε0,pred are not exactly the calibrated values, slight
approximations are introduced in the model. Despite this, the compari-
son between experimental and numerical response plotted in Fig. 13 for
Specimens #4, #12, #14, and #15, shows good agreement. This means
that the use of predicted Δε0,pred values in the beam model leads to ac-
curate results. In addition, the accumulated energy computed for each
experimental tests (Ea) is compared to that (Em) resulting from the pro-
posed MRF beam model comprising the calibrated Δε0value (labelled
calibrated model), as well as the predicted Δε0,pred value (labelled pre-
dicted model). Fig. 14 shows the ratios Em/Ea computed for the
calibrated and predicted model of all selected specimens. Hence, for
the calibrated models the Em/Ea ratio is in the range of 0.840 to 1.15,
while for the predicted models it is in the range of 0.840 to 1.187. Fur-
ther, the average of ratios Em/Ea computed for the calibrated model of
all specimens and that computed for the predicted model is equal to
1.00.
Therefore, the proposed deterioration model is able to simulate ac-
curately the near failure state of the I-shaped beam of moment resisting
frame caused by flange local buckling. This model is not recommended
for I-shaped cross-sections characterised by large value of web slender-
ness because the cross-section deterioration due to web local buckling

Fig. 10. Global and local response of simulated Specimen #12: a) beam with hinges with modified Gauss Radau integration scheme (proposed model); b) beam with hinges with two
points Gauss Radau integration scheme; c) beam with distributed plasticity.
 M. Bosco, L. Tirca / Journal of Constructional Steel Research 133 (2017) 241–255 251

Fig. 11. Correlation between Δε0 and geometrical properties of the beams according to the proposed beam model.
252 M. Bosco, L. Tirca / Journal of Constructional Steel Research 133 (2017) 241–255

Fig. 14. Comparison between experimental and numerical dissipated energy computed for
Fig. 12. Comparison between calibrated and predicted Δε0 values. calibrated and predicted model.

was not included. In addition, the equations proposed to predict the

values of ε0,min and Δε0,pred are verified for the following ranges of 5. Global damage index proposed for the MRF beam model
parameters:
To quantify the reserve capacity of a deteriorated MRF beam, a global
● 53.6 ≤ h/tw ≤ 56.4 damage index, DIs is proposed as follows. This global damage index
● 5.75 ≤ bf/2tf ≤ 7.79 (DIs) is calculated as the number of fibers in which the Miner's damage
● 2.70 ≤ Lv/d ≤ 4.01 index is equal to one (i.e. the number of fibers that reach fatigue) over
● 455 mm ≤ d ≤ 911 mm the total number of top and bottom flanges fibers. For DIs = 1.0, flanges
● 293.7 MPa ≤ Fyf ≤ 368.9 MPa are not able to sustain the demanded bending moment. In Fig. 15, the
● 317.2 MPa ≤ Fyw ≤ 427.5 MPa DIs computed for Specimens #4, #12, #14 and #15 is plotted with
grey dots, which correspond to each drift angle at load reversal. In the
same figure, the envelope of bending moment normalised to the maxi-
For cross-sections characterised by values of parameters different mum recorded bending moment is plotted by a solid line against the as-
from those reported above, a linear distribution of fatigue ductility coef- sociated δ/Lv ratio. It is also shown that only in the case of Specimen #12
ficient ε0 in flange fibers is still able to simulate the gradual degradation the value of DIs reaches 1.0; however, even for lower values of DIs, the
of stiffness and strength caused by local buckling, but the values of ε0,min strength of the cross-section is substantially reduced. Although in the
and Δε0 have to be calibrated. proposed model the strength and stiffness deterioration was included

Fig. 13. Comparison between experimental and numerical response with predicted fatigue material coefficients.
 M. Bosco, L. Tirca / Journal of Constructional Steel Research 133 (2017) 241–255 253

Fig. 15. Strength degradation of replicated Specimens #4, #12, #14 and #15 and global damage index.

by means of fatigue material, it is concluded that under earthquake rotations can be reached when the same specimens are subjected to dif-
loading, the collapse damage index could be associated with 80% of ferent time-strain history. In Table 3, it is also reported the ratio ε/εy
the M/Mmax ratio as is recommended in Option 4 of reference [4]. computed for all specimens where ε is the maximum value of strain
Considering the above approach, the value of DIs associated with recorded at the beam's flange that corresponds to DIs(80%)prop =
M/Mmax = 0.8 is labelled DIs(80%) and is given for all specimens in 0.375 and εy is the yield strain. The ratio ε/εy is in the range of 10.5
Table 3 together with the related drift angle (δ/Lv). From computa- to 18.4. It is noted that similar ε/εy values mark the end of descend-
tions, DIs(80%) is in the range of 0.275 to 0.533 with an average value ing branch of the stress–strain curve when the experimental
of 0.373 and a standard deviation of 0.080, while the corresponding
average drift angle is 0.038. From these results, a value of 0.375 is rec-
ommended as collapse indicator for the proposed global damage
index, DIs(80%)prop.
The plastic rotations θpl associated with DIs(80%)prop are in the range
of 0.021 rad to 0.040 rad with an average value of 0.028 rad and are
given for all specimens in Table 3. However, it is noted that θpl is strictly
related to the adopted loading protocol, i.e. different values of plastic

Table 3
Drift angle, DIs(80%), plastic rotations and ε/εy ratio associated to DIs(80%)prop, margin safety
and plastic rotation associated to beam failure.

Specimen ID δ/Lv(80%) DIS(80%) DIS(80%)prop = 0.375 No. cycles Collapse

to collapse
θpl (rad) ε/εy θpl (rad)

#1 0.032 0.359 0.0250 14.0 4 0.0387

#2 0.031 0.349 0.0262 14.4 4 0.0321
#3 0.031 0.383 0.0391 17.9 6 0.0392
#4 0.034 0.440 0.0210 11.3 5 0.0370
#5 0.034 0.451 0.0235 12.5 6 0.0402
#6 0.030 0.282 0.0233 12.7 6 0.0397
#7 0.030 0.295 0.0236 12.9 6 0.0407
#8 0.040 0.355 0.0289 10.5 3 0.0454
#9 0.042 0.326 0.0300 11.3 2 0.0437
#10 0.044 0.388 0.0303 10.9 3 0.0488
#11 0.042 0.359 0.0298 10.8 4 0.0525
#12 0.042 0.366 0.0302 10.9 5 0.0539
#13 0.058 0.525 0.0400 18.4 1 0.0511
#14 0.050 0.533 0.0281 15.0 1 0.0446
#15 0.036 0.274 0.0223 13.1 3 0.0354
#16 0.036 0.275 0.0225 13.2 5 0.0423
Average value 0.038 0.373 0.0277 13.1 0.0428
Fig. 16. Case study: a) single-storey MRF; b) loading protocol; c) moment-plastic rotation
Standard dev. 0.008 0.080 0.006 2.39 0.006
response at the left end of the beam model.
254 M. Bosco, L. Tirca / Journal of Constructional Steel Research 133 (2017) 241–255
buckling behaviour of steel struts is determined [15]. Since model-
ling of component behaviour beyond the onset of significant degra-
dation is an “immature science” [4], the selected DIs(80%)prop value
is conservative. To show the safety margin of the proposed beam
model, the number of cycles that the selected specimens can sustain
beyond DIs(80%) is also reported in Table 3, as well as the values of
plastic rotations computed at the end of applied loading.
6. Case study
To evaluate the response of the single-storey, one-bay, moment
resisting frame depicted in Fig. 16a, the proposed MRF beam model de-
veloped in OpenSees is employed. The cross-sections adopted for the
beam (W21x44) and columns (W14x176) are those used for Specimen
#14 (Table 1). An elastic beam column element is used to model the col-
umns in order to ensure that the columns do not fail before the beam.
The storey height is 3.40 m, the span length is 6.00 m and the funda-
mental period of the MRF frame is 0.40 s. The seismic response of the
MRF is evaluated by cyclic pushover analysis and incremental nonlinear
dynamic analysis. When the latter is employed, two ground motion re-
cords are considered as seismic input. The first is the crustal record
(963–090) selected from the PEER-NGA ground motion database and
corresponds to Northridge earthquake, while the second record
Fig. 17. Dynamic response of beam model of single-storey MRF: a) unscaled accelerograms, b) time-history of damage index DIs, c) time-history series of normalised strain associated with
DIs(80%)prop, d) M-θpl hysteresis loops corresponding to DIs(80%)prop.
(FKS009) was selected from the K-NET databases (http://www.
kyoshin.bosai.go.jp/) and was recorded during the Tohoku subduction
earthquake (March 2011). The total duration of the former record is
t = 45.0 s and that of the latter is t = 300.0 s. The significant duration
td of the two ground motions is calculated as proposed by Trifunac,
and is the time interval between the 5% and the 95% Arias intensity.
The obtained values are td = 9.08 s for the crustal record and td =
74.0 s for the subduction record. These two records were selected to
show that the proposed model is computationally effective and free of
numerical convergence problems.
The cyclic pushover analysis [31] is performed using the sequence of
nonlinear static loading shown in Fig. 16b. The moment versus plastic
rotation response (M-θpl) recorded in the integration point of the left
plastic hinge segment of the proposed MRF beam model is shown in
Fig. 16c. In the aforementioned figure, the M-θpl response is plotted by
a solid line up to the attainment of DIs(80%)prop and by a dashed line up
to where DIs = 1.00 is reached. The plastic rotation θpl associated with
DIs(80%)prop is equal to 0.039 rad, and that corresponding to DIs = 1.00
is equal to 0.067 rad.
Then, using the two selected accelerograms plotted in Fig. 17a, the
nonlinear response of MRF was investigated by incremental dynamic
analysis. Hence, both accelerograms were incremented up to where
DIs(80%)prop = 0.375 was reached. The time-history series of DIs is
 M. Bosco, L. Tirca / Journal of Constructional Steel Research 133 (2017) 241–255
depicted in Fig. 17b, where DIs(80%)prop = 0.375 is associated with a scal-
ing factor of 2.2 applied to #963-090 record and 2.1 applied to the
FKS009 record. In addition, the time-history series of the normalised
strain (ε/εy) corresponding to DIs(80%)prop is shown in Fig. 17c under
both scaled ground motions, while the moment-plastic rotation hyster-
esis loops are plotted in Fig. 17d. When the nonlinear dynamic analysis
is considered, the plastic rotation corresponding to DIs(80%)prop is larger
than that obtained from cyclic pushover analysis. Meanwhile, the
value of maximum rotation is 0.045 rad under the scaled subduction
ground motion and 0.048 rad under the scaled crustal ground motion.
Thus, the plastic rotation at failure predicted by the proposed model de-
pends on the duration of ground motions.
Therefore, the proposed fiber-based hinge damage accumulation
model for I-shaped beams gives practitioners the possibility to assess
the performance of new and existing MRFs from yielding to failure, as
well as to assess the collapse safety of MRF buildings.
7. Conclusions
This study presents a fiber-based hinge damage accumulation model
proposed to replicate the nonlinear response of I-shaped beams of mo-
ment resisting frames. The proposed OpenSees beam model is made of a
beam with hinges element with fiber-based cross-section discretization
within the plastic hinge zone to which the modified Gauss-Radau inte-
gration scheme was assigned. The proposed model incorporates the
strength and stiffness deterioration caused by flange local buckling
which is simulated by assigning a calibrated low-cycle fatigue material
model to flange fibers. In this formulation, fatigue material uses a mod-
ified rainflow cycle counting algorithm to accumulate damage based on
Miner's rule. The values of fatigue material coefficients were calibrated
against 16 experimental test results selected from the literature.
Results of the analysis reveal a strong correlation between fatigue
material coefficients and parameters such as: flange slenderness, the
ratio of buckling wave length to shear length, and ratio of the product's
flange width and thickness to the product's cross-section depth and
shear length. An equation able to predict the fatigue ductility coefficient
that follows a linear variation along the flange width is proposed based
on regression analysis. In addition, a global damage index, DIs, defined
as the ratio between the number of fibers that reach fatigue and the
number of fibers within the top and bottom flanges of an I-shaped
cross-section is defined. A value of the global damage index DIs(80%)prop
associated with the onset of failure is proposed. Although it complies
with Option 4 of ATC 72-10, this approach is conservative. An applica-
tion comprising a single-storey, one-bay MRF is carried out in OpenSees.
The main conclusions drawn from the analyses are:
● The proposed fiber-based hinge damage accumulation model is able
to replicate both the global (e.g. force versus drift angle) and local re-
sponse expressed in terms of normalised strain recorded in the
flanges of I-shaped beam cross-sections (ε/εy) of selected specimens.
● The proposed global damage index for MRF beams (DIs) quantifies
the reduction of the bending moment capacity as the specimen fol-
lows the descending branch.
● A value of global damage index DIs(80%)prop equal to 0.375 is pro-
posed herein as an indicator of a beam at the onset of failure when
nonlinear dynamic analysis is used.
● From the case study, the proposed beam model was validated as
being computationally effective under dynamic loading.
Acknowledgements
Financial support for this study from the Natural Sciences and
Engineering Research Council of Canada (NSERC) is gratefully
acknowledged.
255
References
[1] C.A. Castiglioni, R. Pucinotti, Failure criteria and cumulative damage models for steel
components under cyclic loading, J. Constr. Steel Res. 65 (2009) 751–765.
[2] F.L. Ribeiro, A.R. Barbosa, M.H. Scott, L.C. Neves, Deterioration modeling of steel Mo-
ment-resisting frames using finite-length plastic hinge force based beam columns
elements, J. Struct. Eng. 141 (2) (2015) http://dx.doi.org/10.1061/(ASCE)ST.1943-
541X.0001052.
[3] V. Gioncu, D. Pectu, Available rotation capacity of wide-flange beams and beam-col-
umns. Part 1: theoretical approaches, J. Constr. Steel Res. 43 (1–3) (1997) 161–217.
[4] PEER/ATC, Modeling and Acceptance Criteria for Seismic Design and Analysis of Tall
Buildings. Rep. No. 72-1, ATC, Applied Technology Council, Redwood City, CA, 2010.
[5] D.G. Lignos, H. Krawinkler, in: A. John (Ed.), Sidesway collapse of deteriorating
structural systems under seismic excitations, Report No. TR 172, Blume Earthquake
Engineering Center, Department of Civil Engineering, Stanford University, Stanford,
2009.
[6] D.G. Lignos, H. Krawinkler, Deterioration modelling of steel components in support
of collapse prediction of steel moment frames under earthquake loading, J. Struct.
Eng. 137 (11) (2011) 1291–1302.
[7] D.G. Lignos, H. Krawinkler, Development and utilization of structural component da-
tabases for performance-based earthquake engineering, J. Struct. Eng. 139 (8)
(2013) 1382–1394.
[8] L.F. Ibarra, R.A. Medina, H. Krawinkler, Hysteretic models that incorporate strength
and stiffness deterioration, Earthq. Eng. Struct. Dyn. 34 (12) (2005) 1489–1511.
[9] S. Mazzoni, F. McKenna, M.H. Scott, G.L. Fenves, B. Jeremic, OpenSees Command
Language Manual, Pacific Earthquake Engineering Research Center, University of
California at Berkeley, 2003.
[10] A.K. Chopra, F. McKenna, Modeling viscous damping in nonlinear response history
analysis of buildings for earthquake excitation, Earthq. Eng. Struct. Dyn. 45 (2016)
193–211.
[11] M.H. Scott, G.L. Fenves, Plastic hinge integration methods for force-based beam-col-
umn elements, J. Struct. Eng. 132 (2) (2006) 244–252.
[12] M.H. Scott, K.L. Ryan, Moment-rotation behaviour of force-based plastic hinge ele-
ments, Earthq. Spectra 29 (2) (2013) 597–607.
[13] T. Guo, L.L. Song, G.D. Zhang, Numerical simulation and seismic fragility analysis of a
self-centering steel MRF with web friction devices, J. Earthq. Eng. 19 (5) (2015)
731–751.
[14] Y. Bai, M. Kurata, M. Nakashima, Modelling of post-local buckling behaviour of
beam-columns with hollow steel section, Proceedings of 8th International Confer-
ence on Behaviour of Steel Structures in Seismic Area, Shanghai, China, 2015.
[15] K. Kasai, T.T. Nam, B.F. Maison, Structural collapse correlative analysis using phe-
nomenological fiber hinge elements to simulate two-directional column deteriora-
tions, Earthq. Eng. Struct. Dyn. (2016) http://dx.doi.org/10.1002/eqe.2742.
[16] K. Lee, B. Stojadinovic, A plastic collapse method for evaluating rotation capacity of
full-restrained steel moment connections, Theor. Appl. Mech. 35 (2008) 191–214.
[17] L. Tirca, L. Chen, Numerical simulation of inelastic cyclic response of HHS braces
upon fracture, Adv. Steel Constr. J. 10 (4) (2014) 442–462.
[18] M. Malakoutian, Seismic Response Evaluation of Linked Column Frame
Systems(PhD Thesis) University of Washington, 2012.
[19] P. Uriz, S. Mahin, Toward Earthquake Resistant Design of Concentrically Braced Steel
Frame Structures, PEER Report 2008/08, 2008.
[20] L. Coffin, A study of the effect of cyclic thermal stress on a ductile metal, Trans. ASME
76 (1954) 931–950.
[21] S. Manson, Fatigue: a complex subject – some simple approximations, Exp. Mech. 5
(7) (1965) 193–226.
[22] M.D. Engelhardt, T.A. Sabol, Testing of welded steel moment connections in re-
sponse to the Northridge Earthquake, Progress Report to the AISC Advisory Sub-
committee on Special Moment Resisting Frame Research, 1994.
[23] T. Kim, A.S. Whittaker, A.S.J. Gilani, V.V. Bertero, S.H. Takhirov, Cover-plate and
Flange-plate Reinforced Steel Moment Resisting Connections. PEER Report 2000/
07, Pacific Earthquake Engineering Research Center, University of California, Berke-
ley, 2000.
[24] K.C. Tsai, E.P. Popov, Steel Beam-column Joints in Seismic Moment Resisting Frames.
Report No. UCB/EERC-88/19, Earthquake Engineering Research Center, Berkeley,
1988.
[25] C.M. Uang, C. Latham, Cyclic Test of Full Scale MNH-SMRFTH Dual Strong Axis Mo-
ment Connections. Report No. TR-95/01, University of California, San Diego, USA,
1995.
[26] M. D'Aniello, R. Landolfo, V. Piluso, G. Rizzano, Ultimate behaviour of steel beams
under non-uniform bending, J. Constr. Steel Res. 78 (2012) 144–158.
[27] M. Bosco, E.M. Marino, P.P. Rossi, Modelling of steel link beams of short, intermedi-
ate or long length, Eng. Struct. 84 (2015) 406–418.
[28] M.G. Lay, Flange local buckling in wide-flange shapes, J. Struct. Div. ASCE 91 (6)
(1965) 95–115.
[29] F.J.D. Arbona, Panel Zone Behaviour in Steel Moment Resisting Frames(Master De-
gree Dissertation) Rose school, Pavia, 2007.
[30] S.M. Kostic, F.C. Filippou, Section discretization of fiber beam-column elements for
cyclic inelastic response, J. Struct. Eng. 138 (5) (2012) 592–601.
[31] P. Panyakapo, Cyclic pushover analysis procedure to estimate seismic demands for
buildings, Eng. Struct. 66 (2014) 10–23.