Engineering Structures 151 (2017) 599–612

Contents lists available at ScienceDirect

Engineering Structures
journal homepage: [Link]/locate/engstruct

Predictive equations for shear link modeling toward collapse

Omid Moammer, Kiarash M. Dolatshahi ⇑
Department of Civil Engineering, Sharif University of Technology, Tehran, Iran

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

Article history: In this paper the predictive equations for collapse assessment of shear links used in eccentric braced
Received 27 April 2017 frames are developed. An extensive database including results of over 70 cyclic tests on steel wide flange
Revised 17 August 2017 shear links is collected and the structural parameters governing the hysteresis behavior are calibrated
Accepted 18 August 2017
using a simplified numerical model. The methodology of calibration is to minimize the discrepancy
between the experimental hysteresis loops and the corresponding numerical results using Particle
Swarm Optimization (PSO) algorithm. The objective function of PSO algorithm is minimized by iterating
Keywords:
parameters that govern the hysteresis behavior of the numerical model. Stepwise multivariable regres-
Shear links
Collapse assessment
sion is used to present equations for modelling shear link behavior parameters. The coefficient of deter-
Cyclic deteriorations mination for the derived empirical equations shows that the proposed equations can accurately capture
Empirical equations the pre-capping, post-capping and cyclic deterioration behavior of the links for collapse assessments.
Regression Ó 2017 Elsevier Ltd. All rights reserved.
Particle Swarm Optimization

1. Introduction and Krawinkler [3] and Haselton et al. [4] used a set of experimen-
tal data including over 300 and 255 tests, respectively, to calibrate
Reliable collapse assessment of buildings needs accurate hys- their numerical model with the test data.
teresis behavior for the structural components. Various researchers All the above-mentioned hysteresis models are developed
have developed mathematical hysteresis models for different com- based on a specific numerical hysteresis model. Several hysteresis
ponents of the building. The nonlinear behavior of steel and con- models are developed in the past forty years to capture cyclic
crete components for various structural systems is also presented behavior of structural components including stiffness and strength
in different codes, such as ASCE41-13 [1,2]. The stiffness and deterioration. Clough and Johnston model proposed in 1966 con-
strength detrioration is implicitly defined in this code by utilizing siders stiffness softening based on the maximum displacement in
the backbone of the cyclic hysteresis curve without incorporation the loading history [6]. In 1970 Takada et al. [7] developed a trilin-
of deterioration models. Moreover, for some structural compo- ear model to capture the cyclic behavior of reinforced concrete
nents the post-capping behavior is not well defined that makes it components. In this model stiffness deterioration initiates after
impossible to capture the collapse of the building. For instance, the flexural cracking and yielding of rebar. Bouc in 1967 [8] pre-
for shear links in eccentric brace frames (EBF) the residual strength sented a smooth hysteresis model for the behavior of single degree
is defined to be 80% of the nominal strength of the section. More of freedom systems. A series of modification has been conducted
sophisticated models are developed within the last ten years to on this model over the years [8–11]. The updated model is called
capture the post-capping behavior and also stiffness and strength Bouc-Wen-Baber-Noori model that considers stiffness and
deteriorations. For example, Lignos and Krawinkler [3] developed strength deterioration as a function of cyclic dissipated energy.
empirical equations for evaluation of cyclic behavior of plastic This model also incorporates pinching in the behavior of structural
hinge in steel moment resisting frames. The same was developed components. Rainhorn and Sivaselavan in 2000 [12] proposed a
by Haselton et. al. [4] for the prediction of the flexure cyclic behav- smooth hysteresis model that incorporates stiffness and strength
ior of plastic hinge in concrete columns leading to global collapse deterioration for models with or without pinching. This model is
of reinforced concrete structures. Lignos and Karamanci [5] devel- also based on the firstly presented Bouc [8] model. Song and Pinch-
oped equations for modelling the cyclic buckling and fracture of eira model [13] presented in 2000 benefits post-capping behavior,
steel braces. All of these models are developed using results of ultimate deformation and residual path. Stiffness deterioration and
quasi-static cyclic tests on the components. For instance, Lignos pinching has been incorporated in this model. strength deteriora-
tion in this model, can be incorporated using the approximate
⇑ Corresponding author. methodology introduced by Pincheira et al. [14]. A series of papers
E-mail address: dolatshahi@[Link] (K.M. Dolatshahi).
has been published by Krawinkler from 90th where the basis of all

[Link]
0141-0296/Ó 2017 Elsevier Ltd. All rights reserved.
600 O. Moammer, K.M. Dolatshahi / Engineering Structures 151 (2017) 599–612

models was on the energy concept of plastic hinges. The initial step eters based on the standpoint of various users; thus an automated
was put forward by Ranama and Krawinkler in 1993 [15]. This approach is generally more accurate. Palomino in 2011 [27] used
model was modified for several times in the last twenty years by genetic optimization algorithm for finding the suitable set of gov-
other researchers [15–20]. In the most recent modifications Lignos erning parameters. Herein, Particle Swarm Optimization (PSO)
[19] implemented residual strength, asymmetrical deterioration algorithm is used to calibrate modified IMK model parameters.
and ultimate deformation to the model, named Modified IMK Genetic algorithm (GA) could also be used to calibrate model
model. Modified IMK model is capable of capturing response of dif- parameters that leads to the same set of parameters for each spec-
ferent types of structural components using a bilinear behavior. ified case. However, Particle Swarm Optimization (PSO) algorithm
This model is also capable of capturing deterioration and collapse is used due to the inherent ability to select input parameters from a
of structural components. In order to present predictive equations continuous range which is desirable is in this investigation. To sim-
for model parameters, linear regression analysis has been used in ulate the behavior of structures toward collapse it is necessary to
the previous studies [3–5]. Castaldo et al. [21] used multivariable capture the full response of components including, pre-capping,
non-linear regression analysis to present predictive equations for post-capping and deterioration. Fig. 1(b) shows the similar data
ultimate flexural resistance and rotation capacity of RHS temper of Fig. 1(a) with an alternative x axis replaced with the cumulative
T6 aluminum alloy beams. Güneyisi et al. [22] presented a new for- link rotation. In this figure the initiation and evolution of the post-
mulation for steel beams flexural overtrength factor using artificial capping behavior of the link can be observed clearer. The normal-
neural network. Güneyisi et al. [23] used neural network and ized difference in each cycle of the numerical and the experimental
genetic expression programming to predict rotation capacity of cycles is presented in Fig. 1(c). According to Eq. (1) the objective
cold formed steel beams. Güneyisi et al. [24] also used the same function is to minimize the summation of the square of the area
approach to develop a new formulation for flexural overstrength difference in each cycle between the numerical and experimental
factor of thin-walled circular hollow sections. In the most recent results.
investigation on modelling steel links, Rossi et al. [25] used an
X
n
experimental database to calibrate a simple refined model. This e¼ ai ðY exp  Y num Þ2 ð1Þ
model which is based on Ramadan and Ghobarah [26] approach, i¼1
includes five elements connected in series that can capture pre-
capping part of cyclic response of the link. Rossi et al. [25] model where Y exp and Y num are the experimental and numerical area of
does not consider strength, stiffness deterioration and effects of each cycle, respectively. ai also shows the normalized weight of
axial load; this model is only suitable for seismic assessment of each cycle. For example, if the post-capping behavior is of particular
eccentric braced frames before any important deterioration hap- importance, ai can be set higher in the optimization procedure in
pens [25]. the cycles of post-capping. In this paper ai is set to unity for all
As mentioned, various models are available in the literature for cycles for simplicity. MATLAB software is used to minimize the
modeling different structural components. By the knowledge of the objective function using PSO algorithm by changing values of input
authors no model is so far developed for modeling shear links that parameters in their respective domains. In the following a brief
is capable of capturing post-capping behavior. In this paper empir- review of PSO algorithm and the parameters domain are presented.
ical equations are developed based on the results of all available
tests conducted on links with wide flange rolled section. In the fol- 3.1. Deterioration model
lowing research objective, methodology, test database, and regres-
sion equations are presented. Backbone curve defines the non-deteriorative behavior of com-
ponents by means of bounds of strength and deflection, see Fig. 2.
2. Research objective Strength parameters include V eq or M y (equivalent strength of sec-
tion), V c or M c (capping strength) and post yield strength ratio
The purpose of this study is to develop empirical equations to (V c =V eq ) or ðM c =M y Þ. Deflection parameters include hp (pre-
predict the pre- and post-capping behavior of structural links. This capping rotation capacity), hpc (post-capping rotation capacity),
is achieved by matching the experimental results of a set of test and h u (ultimate rotation capacity). Using these parameters the
data with the corresponding numerical model. The governing skeleton of the response of the components is presented regardless
parameters of the numerical model are calibrated such that the dif- of deterioration. In this model, three different modes of deteriora-
ference between each cycle of the test is minimized compared to tion have been implemented, namely, basic strength, post-capping
that of the numerical model. The process is automated with an strength, unloading and reloading stiffness deterioration. Consider-
optimization procedure to find the best set of numerical parame- ing Rahnama and Krawinkler [15] rules for defining deterioration
ters. Regression is finally used to propose the empirical equations of structural components each component has an inherent dissi-
for the dominant parameters of the links. In the following the pating energy capacity which is shown by Et in equation (2). This
methodology, deterioration model and the optimization procedure inherent capacity is independent of loading history. The parameter
is described. bi presented in equation (3) demonstrates deterioration of each
mode in excursion i.

3. Methodology Et ¼ khp V eq ð2Þ

Fig. 1(a) shows the experimental cyclic curve of a typical shear !c

Ei
link. The numerical result of the calibrated corresponding zero- bi ¼ Pi1 ð3Þ
length link is also presented in this figure under the same loading Et  j¼1 Ej
protocol. The governing parameters of the modified IMK model in
Opensees are calibrated such that the discrepancy between the where khp : reference cumulative rotation capacity, hp : pre-capping
numerical and experimental results is minimized. In a similar rotation capacity, V eq : equivalent strength of cross section, bi : dete-
study on steel moment resisting frames Lignos [3] used engineer- rioration parameter in extrusion i, Ei : dissipated energy in extrusion
P
ing judgments to match the experimental and numerical results. i, Et : reference dissipation energy capacity, Ej : dissipated energy in
Engineering judgment could result in different set of model param- pervious extrusions, c: rate of cyclic deterioration.
 O. Moammer, K.M. Dolatshahi / Engineering Structures 151 (2017) 599–612 601

300 300 1
0.9
200 200 0.8

Normalized error
Shear [kips]
100 100 0.7

Shear [kips]
0.6
0 0 0.5
0.4
-100 -100 0.3
-200 Numerical -200 0.2
0.1
Test
-300 -300 0
-0.1 -0.05 0 0.05 0.1 0 0.5 1 1.5 2 0 5 10 15 20 25
Rotation [rad] Cumulative rotation [rad] Cycle
(a) Curve fitting (b) Unwrapped (c) Normalized error
Fig. 1. Methodology of calibration.

Fig. 2. modified IMK model parameters illustration.

Basic and post-capping strength deterioration is taken into pi in Eq. (8) represents the best position of ith particle from the
account by shifting boundaries of strength toward origin through beginning of the procedure. In order to find the most desirable
Eq. (4). Unloading stiffness deterioration of the model is considered amount for the objective function, an iterative procedure including
through Eq. (5). updating the velocity and position of the particles and evaluating
the objective function is conducted. This procedure is repeated
V i ¼ ð1  bi ÞV i1 ð4Þ
until the values of objective function converge. c1 , c2 , and w2
named self confidence, swarm confidence and inertia factor,
K i ¼ ð1  bi ÞK i1 ð5Þ respectively, are weight factors of PSO algorithm [29]. Values of
c1 , c2 , and w2 are set 2, 2 and 1, respectively, according to Poli
et al. [30].
3.2. Optimization algorithm
xi0 ¼ xmin þ randðxmax  xmin Þ ð6Þ
PSO, heuristic evolutionary search method, was first presented
by Kennedy and Eberhart in 1995 [28]. This methodology was xmin þ randðxmax  xmin Þ
inspired from how animals share information to find food, avoid v i0 ¼ ð7Þ
Dt
predators and adapt to their respective environment. In this
method the position (xi0 ) and velocity (v i0 ) of the initial swarm of ðpi  xik Þ ðp g  xik Þ
particles is generated randomly in the design space through Eqs. v ikþ1 ¼ wv ik þ c1 rand þ c2 rand k ð8Þ
Dt Dt
(6) and (7), respectively. xmin and xmax represent boundaries for
position of the particles and rand is a random variable between 0
xikþ1 ¼ xik þ v ikþ1 Dt ð9Þ
and 1 [29]. In this investigation parameters of the modified IMK
model act as particles. Range of these particles is selected accord-
ing to the nature of each specified modified IMK model parameter. 3.3. Domain of input parameters
The velocity and position of each particle is updated through Eqs.
(8) and (9), respectively. In Eqs. (8) and (9), v ikþ1 and xikþ1 represent The rational domain of modified IMK model parameters is
velocity and position of ith particle at time k þ 1, respectively. Posi- selected before optimization process to facilitate the optimization
tion of each particle is evaluated by calculating the objective func- performance and to capture the most reliable parameters. The
tion of PSO algorithm. In Eq. (8) pkg represents the position of the domain of input parameters is used to generate initial swarm of
particle with the best objective function value in current swarm. particles and update their position and velocity. xmin for the pre-
602 O. Moammer, K.M. Dolatshahi / Engineering Structures 151 (2017) 599–612

capping and post-capping rotation capacity would be zero. The In order to have a consistent and valid criterion, UT test speci-
upper bound for the pre-capping rotation capacity, hp , would be mens database of this paper only includes the shear links tested
the maximum rotation of the last cycle that the specimen has under revised loading protocol developed by Richards and Uang
experienced [27]. The upper bound for the post-capping rotation [36]. It should be noted that the early 80s specimens were tested
capacity, hpc , is considered 1.0 rad. It is expected that all specimens under a more relaxed loading protocol and did not experience
have a post-capping rotation capacity of the same range. The upper low cycle fatigue. Therefore, all these test specimens are included
and lower bound for deterioration parameters were initially in the database of Table 1. Fan et al. [38] introduced a material
selected according to the previous studies. The domain of deterio- called Q345GJ steel that benefits higher ductility and low cycle
ration parameters is then modified based on engineering judgment fatigue capacity. They investigated cyclic response of 12 link spec-
for few link components. According to the previous studies on steel imens made of Q345GJ steel [38]. These specimens are not
beams [3] the lower and upper bound for the deterioration param- included in the test database due to different characteristics of
eters including basic strength deterioration (ks Þ and post-capping the material which could affect response of the test specimens.
strength deterioration (kc Þ are set to be 0.5 and 2.5 respectively. The database includes the shear-rotation curves of the links
The upper bounds for (ks Þ and (kc Þ were then modified to be 10 under cyclic loading protocols. Geometric and detailing properties
and 3.5, respectively, by manually setting the parameters for few of the sections are also presented in Table 1. All shear-rotation
specified link components. Based on the previous studies the pre- curves of the test specimens of this databased were collected in
capping strength deterioration (C s Þ and post-capping strength picture format from the respective researchers reports and digi-
deterioration rate ðC c Þ were initially ignored for simplicity and tized using Get Data Digitizer software [39]. The database includes
are set to be unity. The domain of post-capping strength deteriora- three types of specimens. Testing of the first type of specimens
tion rate ðC c Þ was then modified to be between 0.5 and 3.5, based stopped before initiation of post-capping and the specimens did
on engineering judgment for few link components. This parameter not experience any type of deterioration. None of the related
could not be ignored due to capturing the sharp strength drop of parameters of the modified IMK model could be calibrated for
some specimens because of web rupture. The range for strain hard- these specimens. Some of the specimens experienced deterioration
ening ratio is considered between 0.01 and 0.1 of the initial elastic in pre-capping regime and did not enter post-capping regime. ks
stiffness. The elastic stiffness of the specimens is set according to which is pre-capping deterioration parameter is calibrated for
Eq. (10) following beam theory and ASCE41-13 [1]. these specimens. The third type of specimens were fully tested
and the related load-displacement curve experienced post-
K sK b
Ke ¼ ð10Þ capping. In these specimens hp , hpc , ks , kc and C c are calibrated.
Ks þ Kb

GAw 5. Calibration results

Ks ¼ ð11Þ
e
The calibrated parameters of the modified IMK model for all
12EIb specimens of Table 1 is presented in Table 2. The database was
Kb ¼ ð12Þ
e3 calibrated using three deterioration parameters including ks ; kc
where Aw , e and G represent web area, length of the link and shear and C c , precisely up to collapse. Parameter C s , rate of deterioration
modulus of the material, respectively. in pre-capping, is ignored for implicitly and is set to unity. From
71 specimens presented in Table 1, 17 specimens did not experi-
ence any deterioration or post-capping behavior during the test or
4. Test database
had odd loading sequences. These specimens are excluded from
the results of Table 2. Out of 54 remained specimens, 22 speci-
A set of tests on shear links are collected and presented in
mens, although having deterioration, did not experience post-
Table 1. The database is subdivided into two groups. The compo-
capping behavior. Pre-capping deterioration parameters ks , C s
nents of the first group have been tested in 80s. In this group the
have been presented for these specimens in Table 2. From 32
detailing and spacing of transverse stiffeners of the links do not
specimens that experienced post-capping behavior, in some spec-
meet the recent seismic specifications [31]. These specimens
imen, mostly from UT test series, basic strength deterioration has
mainly failed due to web local buckling. Engelhardt and Popov in
been limited due to stiffener spacing and detailing conformed to
1992 [32] showed that increasing link length varies the mode of
contemporary seismic code provisions. In order to calibrate these
failure from inelastic local web buckling into flange local buckling
specimens with three deterioration parameters, the upper value of
and lateral torsional buckling. The links of the second group of the
ks is considered 10 as a large number that deactivates basic
database have been tested in Austin Texas University called UT
strength deterioration.
tests. Arce [33] tested specimens with stiffeners detailing accord-
ing to the contemporary seismic specification and used a severer
loading protocol called AISC introduced in supplement No. 2 6. Predictor variables
(2000) to the 1997 AISC seismic provisions [34]. This protocol
was genuinely invented for EBF link to column connections assess- In this section the dependency of the modified IMK model
ment. Arce specimens could not meet the code requirement of parameters to the geometric and detailing properties of link is
inelastic rotation capacity and failed due to unprecedented frac- investigated. In order to find the variables, the failure mode of
tures in web of the sections. Ryu [35] tested Arce [33] shear link test specimens is investigated; the selected variables are pre-
specimens under a revised loading protocol developed by Richards ferred to be dimensionless. Seismic codes also indicate important
and Uang [36] specifically for shear links. Ryu [35] specimens variables which also have been used in this investigation. Web
easily met the code requirements for inelastic rotation capacity slenderness and web thickness represent web local buckling
but failed due to similar fractures firstly observed by Arce [33]. and web fracture failure modes. Flange slenderness and unbraced
Galvez [37] investigated the reasons of this type of failure by test- length ratio also represent flange local buckling and lateral tor-
ing several specimens under different loading protocols; neverthe- sional buckling, respectively. Dimensionless length of the link
less, fundamental reasons of these types of failure are still and spacing of stiffeners could affect failure mode of specimens
unknown. and have been frequently used in seismic codes. All other possible
 O. Moammer, K.M. Dolatshahi / Engineering Structures 151 (2017) 599–612 603

Table 1
Geometric properties of test specimens.

No Reference Section Name d tw tf bf e a q mode of failure

(in.) (in.) (in.) (in.) (in.) (in.)
1 Hjelmastad and Popov [40] W18  40 1 17.88 0.314 0.521 5.985 28 17.36 1.26 Shear
2 W18  40 2 17.88 0.314 0.521 5.985 28 14 1.26 Shear
3 W18  40 3 17.88 0.314 0.521 5.985 28 9.33 1.26 Shear
4 W18  40 4 17.88 0.314 0.521 5.985 28 7 1.26 Shear
5 W18  40 5 17.88 0.314 0.521 5.985 28 11 1.26 Shear
6 W18  40 6 17.88 0.314 0.521 5.985 28 11.2 1.26 Shear
7 W18  35 7 17.69 0.324 0.378 6 28 9.33 1.59 Shear
8 W18  60 8 18.28 0.422 0.681 7.555 36 17.6 1.41 Shear
9 W18  40 9 17.88 0.314 0.521 6.015 36 12 1.62 Intermediate
10 W16  26 1 15.7 0.26 0.356 5.5 36 15.34 1.86 Intermediate
11 W18  35 11 17.69 0.324 0.378 5.906 36 17.31 2.06 Intermediate
12 W12  22 12 12.31 0.266 0.415 4.03 36 11.9 2.75 Flexural
13 W16  26 13 15.7 0.263 0.356 5.484 36 12 1.87 Intermediate
14 W18  35 14 17.69 0.324 0.378 5.906 36 12 2.06 Intermediate
15 W12  22 15 12.31 0.266 0.415 4 36 11.9 2.77 Flexural
16 Malley and Popov [41] W18  60 16 18.24 0.415 0.695 7.555 36 36 1.25 Shear
17 W18  40 17 17.9 0.315 0.525 6.015 36 12 1.48 Shear
18 W18  60 18 18.24 0.415 0.695 7.555 36 36 1.25 Shear
19 W18  40 20 17.9 0.315 0.525 6.015 36 12 1.48 Shear
20 W18  40 21 17.9 0.315 0.525 6.015 36 12 1.48 Shear
21 W18  40 22 17.9 0.315 0.525 6.015 36 12 1.48 Shear
22 W18  40 23 17.9 0.315 0.525 6.015 36 12 1.48 Shear
23 W18  40 24 17.9 0.315 0.525 6.015 36 12 1.48 Shear
24 W18  40 25 17.9 0.315 0.525 6.015 36 12 1.48 Shear
25 W18  40 26 17.9 0.315 0.525 6.015 36 9 1.48 Shear
26 W18  40 27 17.9 0.315 0.525 6.015 36 7.2 1.48 Shear
27 W18  40 28 17.9 0.315 0.525 6.015 36 9 1.48 Shear
28 Kasai and Popov [42] W8  10 1 7.97 0.17 0.208 3.94 14.5 3.63 1.41 Shear
29 W8  10 3 7.97 0.17 0.208 3.94 14.5 3.63 1.41 Shear
30 W8  10 4 7.97 0.17 0.208 3.94 14.5 3.63 1.41 Shear
31 W8  10 5 7.97 0.17 0.208 3.94 17.5 3.5 1.7 Intermediate
32 W8  10 6 7.97 0.17 0.208 3.94 17.5 3.5 1.7 Intermediate
33 W8  10 7 7.97 0.17 0.208 3.94 11.5 3.83 1.12 Shear
34 Rides and Popov [43] W12  19 Al 12.06 0.254 0.354 4.005 19 6.33 1.5 Shear
35 W12  19 B1 12.06 0.254 0.354 4.005 19 3.17 1.5 Shear
36 W12  19 C1 12.06 0.254 0.354 4.005 19 6.33 1.5 Shear
37 W12  19 D1 12.06 0.254 0.354 4.005 19 6.33 1.5 Shear
38 W12  19 A2 12.06 0.254 0.354 4.005 19 6.33 1.5 Shear
39 W12  19 B2 12.06 0.254 0.354 4.005 19 6.33 1.5 Shear
40 W12  19 C2 12.06 0.254 0.354 4.005 19 6.33 1.5 Shear
41 W12  19 D2 12.06 0.254 0.354 4.005 19 6.33 1.5 Shear
42 Engelhardt and Popov [44] W12  16 1 11.99 0.22 0.265 3.99 28 5.6 2.36 Intermediate
43 W12  16 2 11.99 0.22 0.265 3.99 28 7 2.36 Intermediate
44 W12  22 3 12.31 0.26 0.425 4.03 33 8.25 2.36 Intermediate
45 W12  22 4 12.31 0.26 0.425 4.03 33 11.89 2.3 Intermediate
46 W12  16 5 11.99 0.22 0.265 3.99 44 11.73 3.71 Flexural
47 W12  16 6 11.99 0.22 0.265 3.99 44 11.73 3.7 Flexural
48 W12  16 7 11.99 0.22 0.265 3.99 29 8 2.45 Intermediate
49 W12  22 8 12.31 0.26 0.425 4.03 36 7.75 2.37 Intermediate
50 W12  22 9 12.31 0.26 0.425 4.03 20 6.67 1.34 Shear
51 W12  16 10 11.99 0.22 0.265 3.99 21 5.25 1.77 Intermediate
52 W12  22 11 12.31 0.26 0.425 4.03 27.5 6.88 1.81 Intermediate
53 W12  22 11R1 12.31 0.26 0.425 4.03 26.75 6.69 1.76 Intermediate
54 W12  22 11R2 12.31 0.26 0.425 4.03 26.75 6.69 1.76 Intermediate
55 W12  22 12 12.31 0.26 0.425 4.03 60 11.89 3.95 Flexural
56 Arce and Engelhardt [33] W10  19 1c 10.304 0.265 0.383 4.002 23 5.75 1.77 Intermediate
57 W10  19 2 10.304 0.265 0.383 4.002 30 6 2.31 Intermediate
58 W10  19 3 10.304 0.265 0.383 4.002 48 6 3.69 Flexural
59 W10  33 5 9.744 0.319 0.437 8.045 36.6 6.125 1.67 Intermediate
60 W10  33 6b 9.744 0.319 0.437 8.045 48 9.625 2.19 Intermediate
61 W10  33 7 9.744 0.319 0.437 8.045 73 12 3.33 Flexural
62 W16  36 9 16.056 0.301 0.487 6.963 48 8 1.82 Intermediate
63 Ryu [35] w10  33 4a-rlp 9.744 0.319 0.437 8.045 23 5.75 1.1 Shear
64 w10  33 4c-rlp 9.744 0.319 0.437 8.045 23 5.75 1.1 Shear
65 w16  36 8-rlp 16.056 0.301 0.487 6.963 36.6 5.22 1.5 Shear
66 w10  68 10-rlp 10.4 0.449 0.773 10.26 48 12 1.65 Flexural
67 w10  68 11-rlp 10.4 0.449 0.773 10.26 48 12 1.65 Flexural
68 w18  40 12-rlp 17.82 0.31 0.5 6.094 23 5.75 1.05 Shear
69 Galvez [37] w10  33 8 9.7 0.302 0.435 8 23 5.75 0.99 Shear
70 w10  33 9 9.7 0.302 0.435 8 23 5.75 0.99 Shear
71 w10  33 10 9.7 0.302 0.435 8 23 7.66 0.99 Shear
604 O. Moammer, K.M. Dolatshahi / Engineering Structures 151 (2017) 599–612

Table 2
Calibration results.
+ -
No Reference Name ks Cs kc Cc hp hpc Veq Veq
1 Hjelmastad and Popov [40] 1 0.872 1 128 137
2 2 2.075 1 145 160
3 5 3.303 1 1.5 1.29 0.101 0.058 160 182
4 6 7.377 1 1.725 2.1 0.109 0.05 185 198
5 7 5.5 1 1.538 1.8 0.09 0.087 185 191
6 8 1.285 1 223 248
7 9 4.047 1 150 168
8 1 0.517 1 117 121
9 11 0.556 1 143 174
10 12 1.148 1 115 127
11 13 1.326 1 120 122
12 14 2.146 1 162 180
13 15 3 1 0.8 1.1 0.05 0.5 125 147
14 Malley and Popov [41] 17 3.948 1 1.508 1.548 0.079 0.049 160 160
15 20 2.673 1 1.536 2.138 0.089 0.093 173 173
16 21 2.373 1 170 170
17 22 2.6 1 1.394 1.62 0.079 0.15 167 185
18 23 3.757 1 1.2 2.25 0.082 0.091 162 170
19 25 2.34 1 163 174
20 26 4.915 1 1.47 2 0.118 0.12 194 188
21 27 5 1 1.33 2 0.07 0.115 202 203
22 Kasai and Popov [42] 4 3.5 1 1.636 1 0.058 0.35 50 50
23 6 1.884 1 1.308 1.17 0.017 0.236 50 45
24 7 2.5 1 61 61
25 Rides and Popov [43] B1 8 1 110 118
26 D1 5.241 1 1.75 1.7 0.088 0.193 110 110
27 B2 5.441 1 118 119
28 D2 4.75 1 113 112
29 Engelhardt and Popov [44] 1 2.5 1 1.01 2.079 0.071 0.218 75 75
30 2 0.619 1 70 70
31 3 5.5 1 1.15 1.458 0.044 0.145 100 105
32 4 1.573 1 0.75 1.25 0.044 0.177 105 108
33 5 0.892 1 0.625 1.099 0.027 0.08 47 48
34 6 0.5 1 57 57
35 7 0.45 1 60 60
36 9 6.5 1 0.95 1.3 0.051 0.099 103 104
37 10 3.5 1 1.5 1.884 0.04 0.129 98 98
38 11R2 6.5 1 1.1 1.3 0.041 0.08 94 99
39 Arce and Engeihardt [33] 1c 10 1 3.05 2.2 0.095 0.099 102 105
40 2 10 1 2.05 1.8 0.066 0.329 93 86
41 3 1.3 1 58 58
42 5 10 1 2.12 2.1 0.086 0.07 118 110
43 6b 2.5 1 98 92
44 7 2.2 1 66 70
45 9 1.8 1 164 158
46 Ryu [35] 4a-rlp 10 1 1.95 2.306 0.097 0.382 125 125
47 4c-rlp 10 1 2.25 2.1 0.16 0.397 130 125
48 8-rlp 10 1 2.05 2.2 0.141 0.345 196 201
49 10-rlp 10 1 2.025 2.2 0.125 0.321 196 193
50 11-rlp 10 1 1.45 2.2 0.132 0.254 186 184
51 12-rlp 6 1 2.1 2.05 0.112 0.331 240 238
52 Galvez [37] 8 6.5 1 2 2 0.118 0.172 114 109
53 9 10 1 1.74 2.5 0.13 0.165 122 118
54 10 10 1 1.87 2.4 0.114 0.13 123 112

geometric, detailing, and material properties have been also the following the dependency of the parameters is thoroughly
incorporated in this investigation; however, some of them have discussed.
been excluded from this section due to statistical insignificancy.
In Figs. 3–7 the vertical and horizontal axis represent the modi- 6.1. Web thickness tw
fied IMK model parameters and geometric or detailing properties
of the links, respectively. Higher coefficient of determination (R2) UT tests series specimens failed due to fractures in K-Area of the
of the trend lines represents more influence of each geometric or web of the section. The early 80s specimens also failed due to
detailing property that should be incorporated in the empirical inelastic web local buckling which led to fracture in the web of
equations. The figures express an insight into the trends between the section. Fig. 3 shows a very strong dependency between the
parameters and variables; however, they cannot be directly used calibrated model parameters (hp , C c and kc ) and tw . It can be con-
to extract relationships between the model parameters and pre- cluded that increasing of t w leads to higher values of pre-capping
dictor variables. This is due to the fact that these figures do not and post-capping rotation capacity and deterioration parameters
take into account the correlation between predictor variables. In of the links.
 O. Moammer, K.M. Dolatshahi / Engineering Structures 151 (2017) 599–612 605

10 3.5 3.5
R 2 =0.109 R 2 =0.329 R 2 =0.329
3 3
8
2.5 2.5
6 2 2

Cc

λc
λs

4 1.5 1.5
1 1
2
0.5 0.5
0 0 0
0.1 0.2 0.3 0.4 0.5 0.6 0.1 0.2 0.3 0.4 0.5 0.6 0.1 0.2 0.3 0.4 0.5 0.6
t [inch] t [inch] t [inch]
w w w

(a) (b) (c)

0.2 0.6
R 2 =0.484 R 2 =0
0.5
0.15
0.4
θpc
θp

0.1 0.3

0.2
0.05
0.1

0 0
0.1 0.2 0.3 0.4 0.5 0.6 0.1 0.2 0.3 0.4 0.5 0.6
t w [inch] t w [inch]

(d) (e)
Hjelmastad Malley Kasai Rides Engelhardt Arce Ryu Galvez

Fig. 3. Dependency of modified IMK model parameters to tw .

10 3.5 3.5
R 2 =0.37 R 2 =0.133 R 2 =0.133
3 3
8
2.5 2.5
6 2 2
Cc

λc
λs

4 1.5 1.5
1 1
2
0.5 0.5
0 0 0
10 20 30 40 50 60 10 20 30 40 50 60 10 20 30 40 50 60
(d-2t f )/t w (d-2t f )/t w (d-2t f )/t w

(a) (b) (c)

0.2 0.6
R 2 =0.161 R 2 =0.139
0.5
0.15
0.4
θpc
θp

0.1 0.3

0.2
0.05
0.1

0 0
10 20 30 40 50 60 10 20 30 40 50 60
(d-2t f )/t w (d-2t f )/t w

(d) (e)
Hjelmastad Malley Kasai Rides Engelhardt Arce Ryu Galvez

Fig. 4. Dependency of modified IMK model parameters to ðd  2t f Þ=t w .

606 O. Moammer, K.M. Dolatshahi / Engineering Structures 151 (2017) 599–612

10 3.5 3.5
R 2 =0.211 R 2 =0.259 R 2 =0.259
3 3
8
2.5 2.5
6 2 2

λc
λs

C
4 1.5 1.5
1 1
2
0.5 0.5
0 0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 0 0.5 1 1.5 2 2.5 3 3.5 4 0 0.5 1 1.5 2 2.5 3 3.5 4
ρ=e/(M /V ) ρ=e/(M /V ) ρ=e/(M /V )
p p p p p p

(a) (b) (c)

0.2 0.6
R 2 =0.385 R 2 =0.002
0.5
0.15
0.4
θpc
θp

0.1 0.3

0.2
0.05
0.1

0 0
0 0.5 1 1.5 2 2.5 3 3.5 4 0 0.5 1 1.5 2 2.5 3 3.5 4
ρ=e/(M /V ) ρ=e/(M p /Vp )
p p

(d) (e)
Hjelmastad Malley Kasai Rides Engelhardt Arce Ryu Galvez

Fig. 5. Dependency of modified IMK model parameters to link (q ¼ e=ðMp =Vp Þ).

10 3.5 3.5
R 2 =0.074 R 2 =0.006 R 2 =0.006
3 3
8
2.5 2.5
6 2 2
c

λc
λs

4 1.5 1.5
1 1
2
0.5 0.5
0 0 0
0 0.3 0.6 0.9 1.2 1.5 0 0.3 0.6 0.9 1.2 1.5 0 0.3 0.6 0.9 1.2 1.5
a/d a/d a/d
(a) (b) (c)
0.2 0.6
R 2 =0 R 2 =0.006
0.5
0.15
0.4
θpc
θp

0.1 0.3

0.2
0.05
0.1

0 0
0 0.3 0.6 0.9 1.2 1.5 0 0.3 0.6 0.9 1.2 1.5
a/d a/d
(d) (e)
Hjelmastad Malley Kasai Rides Engelhardt Arce Ryu Galvez

Fig. 6. Dependency of modified IMK model parameters to (a=d).

 O. Moammer, K.M. Dolatshahi / Engineering Structures 151 (2017) 599–612 607

10 3.5 3.5
R 2 =0.007 R 2 =0.074 R 2 =0.074
3 3
8
2.5 2.5
6 2 2

Cc

λc
λs
4 1.5 1.5
1 1
2
0.5 0.5
0 0 0
0 2 4 6 8 10 0 2 4 6 8 10 0 2 4 6 8 10
bf/2t f bf/2t f bf/2t f

(a) (b) (c)

0.2 0.6
R 2 =0.067 R 2 =0.046
0.5
0.15
θpc 0.4
θp

0.1 0.3

0.2
0.05
0.1

0 0
0 2 4 6 8 10 0 2 4 6 8 10
bf/2t f bf/2t f

(d) (e)
Hjelmastad Malley Kasai Rides Engelhardt Arce Ryu Galvez

Fig. 7. Dependency of modified IMK model parameters to (bf =2tf ).

6.2. Web slenderness ((d  2tf)/tw) where F yf , F yw and Z p represent yield strength of flange steel, yield
strength of web steel and the plastic section modulus.
Inelastic web local buckling would be the dominant failure
mode if spacing and detailing of stiffeners do not meet the require- 6.4. Spacing of stiffeners to depth of link (a/d)
ments of the seismic code provisions [45]. As ðd  2t f Þ=tw increases,
the potential of web local buckling rises. Fig. 4 demonstrates a The ratio between stiffeners spacing to depth of the link is
strong dependency between the calibrated model parameters shown by a=d that represents the potential of development of
and ðd  2tf Þ=t w . As the slenderness ratio of the web increases, post-tension field action. Increasing a=d ratio could disorder the
pre-capping and post-capping rotation capacity and the deteriora- post tension field action in the web of the links. Increasing a=d ratio
tion parameters (ks , kc , C c ) decrease. also increases the potential of web local buckling. Poor post tension
field action decreases the parameters related to the capacity of the
6.3. Dimensionless length of link (q = e/(Mp/Vp)) section, namely hp and hpc , and reduces the deterioration parame-
ters. Fig. 6 shows that the dependency between model parameters
Dimensionless length of the link is an index that describes the and a=d is not strong.
dominant action of the link through Eq. (13), whether shear, flex-
ure, or intermediate [46]. As the dimensionless length of the link
6.5. Beam flange slenderness (bf/2tf)
increases, the dominant action of the link shifts from shear to flex-
ure action. The most efficient action of links is shear. Fig. 5 shows a
The slenderness ratio of the flange of the link cross section is
strong dependency between the calibrated modified IMK model
represented by bf =2tf . Increasing bf =2t f leads to the higher poten-
parameters and dimensionless length of the link. Increasing the
tial of flange local buckling that has been observed in some of the
dimensionless length of the link leads to lower amounts for pre-
specimens [40]. Web local buckling triggers flange local buckling
capping and post-capping rotation capacity and deterioration
and consequently lateral torsional buckling. Fig. 7 shows indepen-
parameters. Rossi et al. [25] calibrated simple refined models for
dency of model parameters to bf =2tf .
short, long, and intermediate links; separately. Using dimension-
Unbraced length ratio of the link is represented through Lb =r y .
less length of the link, model parameters are calibrated here for
Lb and ry represent lateral bracing distance and radius of gyration
short, long, and intermediate links by a smooth transition instead
about the y-axis of the section, respectively. Not many specimens
of using three set of equations.
with lateral torsional buckling were observed in the collected test
q ¼ e=ðMp =V p Þ ð13Þ database mostly due to the short length of the links.
where M p and V p represent full plastic moment and shear strength,
respectively. 6.6. Axial restraints

V p ¼ 0:6F yw t w ðd  2t f Þ ð14Þ According to Landolfo et al. [47] based on experimental test

setup, presence of axial restraints at the end of the links results
M p ¼ F yf Z p ð15Þ in different set of model parameters. McDaniel et al. [48], Dusicka
608 O. Moammer, K.M. Dolatshahi / Engineering Structures 151 (2017) 599–612

et al. [49] and Mazzolani et al. [50] exerted restraint to axial defor- (q), web yield stress ðF yw Þ, and spacing ratio of stiffeners ða=dÞ are
mations of the links. However, the results of these studies have not less important in predicting hp . Other predictor variables have been
been incorporated in this paper because of using built up sections eliminated due to statistical insignificancy. Fig. 8 represents the
and presence of axial restraints. Landolfo et al. [47] modified over- predicted hp versus respective calibrated values.
strength factor of the link in the presence of tensile axial forces  a 0:29
induced as a consequence of axial restraints. Effects of axial hp ¼ 0:094 ðqÞ0:52 ðt w Þ1:74 ðF yw Þ0:51 ð19Þ
restraints on model parameters is not the scope of this investiga- d
tion and should be thoroughly investigated in further studies.
rln ¼ 0:27
7. Regression results R2 ¼ 0:74

Empirical equations between the modified IMK model parame-

ters presented in Table 2 and the geometric, detailing and material 7.3. Post-capping rotation capacity
properties of the link are presented below. These equations are pre-
sented through regression using equation (16), following Lignos and Eq. (20) represents the predictive equation for the post-capping
Krawinkler [3] in a similar study on MRFs. In Eq. (16) X1 to Xn are pre- rotation capacity. This equation is based on 32 test specimens
dictor variables related to the geometric, detailing and material which experienced post-capping behavior during the test. Fig. 9
properties of the link cross sections; a1 to an are also regression coef- shows the predicted values for hpc versus the calibrated values of
ficients. Eq. (17) is the logarithmic form of Eq. (16) to make a linear the respective tests. Eq. (20) coincides the physical expectations;
relation between the parameters. The variables of Eq. (18) are the power of coefficient F yw is positive that means increasing of
selected form the trends discussed in section ‘‘Predictor variables”, F yw has a positive effect on the post-capping capacity. Other
failure mode of the specimens and nature of the specific modified parameters are less important statistically and are eliminated from
IMK model parameters. These parameters are those that can have the predictive equation.
influential effects on the response of the link.  3:56
F yw
hpc ¼ 0:039 ð20Þ
RP ¼ a1 ðX 1 Þa2 ðX 2 Þa3 . . . ðX n Þanþ1 ð16Þ 36

logðRPÞ ¼ logða1 Þ þ a2 logðX 1 Þ þ a3 logðX 2 Þ . . . anþ1 logðX n Þ þ e rln ¼ 0:43

ð17Þ
R2 ¼ 0:53
 a  a6
d  2t f 2 aa3 bf
RP ¼ a1 ðqÞa4 ðt w Þa5 ðF yw Þa7 ð18Þ 0.2
tw d 2tf
R 2 =0.742
where F yw is the yield stress of the web of the link in ksi. d and t w are
the depth and thickness of the web of the link in inches, respectively. 0.15
Predicted θp

7.1. Regression method

0.1
In order to find the most statistically significant variables, a
stepwise regression procedure is conducted. Regression coeffi-
cients are unknown model parameters with statistical uncertainty. 0.05
In a stepwise procedure, the coefficient of variation of the regres-
sion coefficients is determined in each step. ðX i1 Þai term related
to ai regression coefficient that has the largest value of COV is elim- 0
0 0.05 0.1 0.15 0.2
inated from the model. After eliminating a term, the quality of the
model should be checked through prediction versus observation Calibrated θ
p
diagrams. Furthermore, the mean of standard deviation of the
model error should not jump and COV of regression coefficients Fig. 8. Predicted versus calibrated pre-capping rotation capacity.
must be in acceptable range; otherwise, the previous model form
is more desirable. In the proposed predictive equations in this 0.6
paper, regression coefficients are set to their respective mean esti- R 2 =0.528
mation [51]. Basic assumptions of linear regression are not violated 0.5
in this procedure. Statistics of regression coefficients and normality
of residuals is investigated in Appendix A. 0.4
Predicted θpc

7.2. Pre-capping rotation capacity 0.3

Eq. (19) is presented for predicting the pre-capping rotation 0.2

capacity of links according to the link geometry, detailing and
material properties. This equation is presented using 32 test spec- 0.1
imens with post-capping behavior. Other tests did not experience
post-capping and hence are excluded in regression for the calcula- 0
tion of the pre-capping rotation capacity. According to Eq. (19) tw is 0 0.1 0.2 0.3 0.4 0.5 0.6
the most important predictor variable having the highest power Calibrated θpc
value. tw has a positive power that shows increasing of tw increases
the pre-capping rotation capacity. Dimensionless length of the link Fig. 9. Predicted versus calibrated post-capping rotation capacity.
 O. Moammer, K.M. Dolatshahi / Engineering Structures 151 (2017) 599–612 609

7.4. Deterioration parameters

Veq
In order to capture the full shear-rotation curve of the test data-
base, ks, Cs, kc, and Cc are used as deterioration parameters. Eq. (21)
for estimation of ks is presented based on 54 test specimens that
have experienced pre-capping deterioration in the cyclic response. Vn

Shear
Cs is implicitly considered to be equal to unity for all specimens.
Eqs. (22) and (23) for kc and Cc, respectively, are also presented
according to 32 specimens that experienced post-capping behavior
(see Fig. 10).
Multi-Linear
 2:20  1:48  1:06
d  2tf a bf Bilinear
ks ¼ 0:203 ðqÞ0:74 ðt w Þ0:69 ð21Þ
500t w d 2tf
Rotation
rln ¼ 0:46
Fig. 11. Idealized monotonic shear rotation curve.
R2 ¼ 0:70

 0:6  0:67
d  2t f a
kc ¼ 0:720 ðqÞ0:23 ð2000tw Þ0:43 ð22Þ
tw d
300
rln ¼ 0:22 R 2 =0.927
2
R ¼ 0:5 250

 0:28  0:44 200

d  2tf a

Predicted Veq
C c ¼ 0:032 ð1000tw Þ0:85 ð23Þ
tw d
150
rln ¼ 0:17
R2 ¼ 0:54 100

7.5. Equivalent yielding point Veq 50

According to Fig. 11 the initial hardening behavior of the link 0

0 50 100 150 200 250 300
follows a multi-linear curve suggested in previous studies
Calibrated Veq
[27,32,33,42,43]. In this research the multi-linear curve is idealized
to a bilinear curve by updating the yielding point, V eq in Fig. 11. V eq
Fig. 12. Predicted versus calibrated V eq .
takes into account the isotropic hardening that has been incorpo-
rated in the modified IMK model. Based on the regression results
of Fig. 12 this parameter is set to 1.2 times V n , nominal shear
strength of the cross section. Nominal shear strength of the cross
ðV c =V eq Þ and ðhp =hy Þ. For modeling link elements, the median
section for all specimens would be minimum of 0:6F yw Aw and
value for ðV c =V eq Þ in the database is used. The median value
2Mp =e [45].
and standard deviation for ðV c =V eq Þ is 1.09 and 0.05, respectively.
V eq ¼ 1:2V n ð24Þ hy is computed through Eq. (25) following recommendations of
ASCE41-13.
r ¼ 0:13
R2 ¼ 0:92
Vn
hy ¼ ð25Þ
7.6. Strain hardening Kee

Strain hardening is defined by the ratio of the peak shear where V n : strength of the link, K e : elastic stiffness of the link, e:
strength to the equivalent nominal shear strength ratio length of the link.

14 3.5 3.5
R 2 =0.706 R 2 =0.498 R 2 =0.544
12 3 3
10
Predicted C c
c
s

2.5 2.5
Predicted λ

Predicted λ

8
2 2
6
1.5 1.5
4
1 1
2
0.5 0.5
2 4 6 8 10 12 14 0.5 1 1.5 2 2.5 3 3.5 0.5 1 1.5 2 2.5 3 3.5
Calibrated λ Calibrated λ Calibrated Cc
s c
(a) (b) (c)
Fig. 10. Predicted versus calibrated deterioration parameters.
610 O. Moammer, K.M. Dolatshahi / Engineering Structures 151 (2017) 599–612

Table 3
Statistics of hp predictive equation.

Coefficient Mean COV Correlation coefficients

a1 a3 a4 a5 a7
a1 2.36 60.18 1 0.08 0.07 0.02 0.98
a3 0.30 68.93 0.08 1 0.55 0.58 0.13
a4 0.52 39.04 0.07 0.55 1 0.56 0.12
a5 1.74 17.77 0.02 0.58 0.56 1 0.17
a7 0.52 69.22 0.98 0.13 0.12 0.17 1

Table 4 of shear link components. Some tests were stopped before any
Statistics of hpc predictive equation. deterioration. These specimens were excluded from the test data-
Coefficient Mean COV Correlation coefficients base. The hysteresis curves of the components were digitized and
a1 a7
calibrated using a numerical model capable of capturing strength
deterioration and post-capping behavior. The predictive equations
a1 16.03 13.84 1 1.00
are proposed based on the regression analyses of the calibrated
a7 3.56 15.66 1.00 1
results. These equations include pre-capping rotation capacity
ðhp Þ, post-capping rotation capacity ðhpc Þ, pre-capping basic
strength deterioration ðks Þ, post-capping strength deterioration
ðkc Þ and its rate ðC c Þ. Generally, based on the results it can be con-
8. Conclusions cluded that:

In this paper predictive equations for seismic assessment of  Parameters defining the behavior of shear links are mainly
shear links are presented. Most of the codes such as ASCE41 are affected by the potential of the web local buckling and tearing
capable of simulating the nonlinear response of the shear links. of the web of the test specimens. ðd  2t f Þ=tw and tw are predic-
However, the prediction is limited to nondeteriorative pre- tor variables that represent web local buckling and web tearing,
capping behavior and no document is available for assessment of respectively.
the shear links until collapse. A set of test consisting of over 70  a=d represents spacing of the stiffeners to depth of the link. Clo-
components are collected and evaluated in this paper. From this ser spacing of the stiffeners leads to less deterioration. Applica-
set of specimens some of the components were not fully tested tion of stiffeners according to the contemporary seismic codes
and stopped before complete deterioration. These specimens are causes no deterioration in pre-capping region.
only utilized for assessment of pre-capping strength deterioration

Table 5
Statistics of ks predictive equation.

Coefficient Mean COV Correlation coefficients

a1 a2 a3 a4 a5 a6
a1 12.08 10.57 1 0.87 0.10 0.07 0.13 0.69
a2 2.20 12.72 0.87 1 0.14 0.00 0.33 0.40
a3 1.48 16.93 0.10 0.14 1 0.56 0.47 0.14
a4 0.74 32.99 0.07 0.00 0.56 1 0.49 0.26
a5 0.69 61.96 0.13 0.33 0.47 0.49 1 0.39
a6 1.06 30.83 0.69 0.40 0.14 0.26 0.39 1

Table 6
Statistics of kc predictive equation.

Coefficient Mean COV Correlation coefficients

a1 a2 a3 a4 a5
a1 2.95 19.98 1 0.90 0.53 0.42 0.31
a2 0.60 28.49 0.90 1 0.50 0.37 0.08
a3 0.68 28.23 0.52 0.50 1 0.62 0.45
a4 0.24 75.23 0.41 0.37 0.62 1 0.48
a5 0.44 56.81 0.30 0.08 0.45 0.48 1

Table 7
Statistics of C c predictive equation.

Coefficient Mean COV Correlation coefficients

a1 a2 a3 a5
a1 2.46 17.16 1 0.89 0.38 0.13
a2 0.28 44.69 0.89 1 0.37 0.32
a3 0.44 26.74 0.38 0.37 1 0.22
a5 0.86 20.11 0.13 0.32 0.22 1
 O. Moammer, K.M. Dolatshahi / Engineering Structures 151 (2017) 599–612 611

1 1 1

Normal Data Quantiles

Normal Data Quantiles

Normal Data Quantiles

0.5 0.5 0.5

0 0 0

-0.5 -0.5 -0.5

-1 -1 -1
-1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1
Normal Theoretical Quantiles Normal Theoretical Quantiles Normal Theoretical Quantiles

(a) (b) (c)

1 1
Normal Data Quantiles

0.5
Normal Data Quantiles 0.5

0 0

-0.5 -0.5

-1 -1
-1 -0.5 0 0.5 1 -1 -0.5 0 0.5 1
Normal Theoretical Quantiles Normal Theoretical Quantiles

(d) (e)
Fig. 13. Normality plot of residuals.

 a=d also represents the post-tension field action of the web of Acknowledgments
the link. It can be concluded that decreasing a=d leads to more
stable development of post-tension field action and less The authors acknowledge the help provided by Mr. Mehrdad
deterioration. Ghiabi and Mr. Hamid Madani for collecting and digitizing the test
 As spacing of stiffeners increases, flange local buckling and lat- data.
eral torsional buckling lead to the failure of the test specimens.
lb =ry and bf =2tf representing lateral torsional buckling and
flange local buckling of the specimens are mostly excluded from Appendix A
the equations due to statistical insignificancy. This is mainly
because of the short length of the shear links. In Tables 3–7, statistics of regression coefficients is presented.
 As the dimensionless length of the link increases, pre-capping In Fig. 13, normality of residuals for predictive equations is
rotation capacity decreases and deterioration increases. It can investigated.
be concluded that lower dimensionless length of the link leads
to more desirable dominant shear action of the link. References
 Many variables are simultaneously contributing in the predic-
tive equations; thus, it is difficult to predict the failure mode [1] Pekelnicky R, Poland C. ASCE 41-13: seismic evaluation and retrofit of existing
based on the proposed equations. However, according to vari- buildings. In: SEAOC 2012 convention proceedings.
[2] Speicher MS, Harris JL. Collapse prevention seismic performance assessment of
ables in the predictive equations it can be expected that increas- new eccentrically braced frames using ASCE 41. Eng Struct 2016;117:344–57.
ing web slenderness, and stiffener spacing leads to higher [3] Lignos DG, Krawinkler H. Deterioration modeling of steel components in
potential of web local buckling. Increasing flange slenderness, support of collapse prediction of steel moment frames under earthquake
loading. J Struct Eng 2010;137(11):1291–302.
and dimensionless length of the link triggers flange local buck-
[4] Haselton CB, Center PEER. Beam-column element model calibrated for
ling. As web thickness increases, possibility of fracture of the predicting flexural response leading to global collapse of RC frame
web of the section is limited. buildings. Pacific Earthquake Engineering Research Center; 2008.
[5] Karamanci E, Lignos DG. Computational approach for collapse assessment of
 Median amount of pre-capping rotation capacity ðhp Þ and post-
concentrically braced frames in seismic regions. J Struct Eng 2014;140(8):
capping rotation capacity ðhpc Þ are 0.085 and 0.19 rad, respec- A4014019.
tively. Deterioration parameters are in the range of 0.5–10. [6] Clough W, Johnston S. Effect of stiffness degradation on earthquake ductility
requirements. In: Proceedings of the second Japan earthquake engineering
symposium. p. 227–32.
Finally, it should be noted that the empirical equations of this [7] Takeda T, Sozen MA, Nielsen NN. Reinforced concrete response to simulated
paper are developed based on the results of all available tests on earthquakes. J Struct Div 1970;96(12):2557–73.
wide flange rolled link specimens. The proposed equations are [8] Bouc R. Forced vibration of mechanical systems with hysteresis. In:
Proceedings of the fourth conference on non-linear oscillation, Prague,
the best that could be achieved based on the available number of Czechoslovakia.
tests. More tests should be conducted on links that capture post- [9] Wen YK, Baber TT. Random vibration of hysteretic degrading system. J Eng
capping behavior in the future to more reliably assess the devel- Mech Div ASCE 1981;107:1069–87.
[10] Baber TT, Noori MN. Modeling general hysteresis behavior and random
oped equations. vibration application. J Vib Acoust Stress Reliab Des 1986;108(4):411–20.
612 O. Moammer, K.M. Dolatshahi / Engineering Structures 151 (2017) 599–612

[11] Baber TT, Noori MN. Random vibration of degrading, pinching systems. J Eng [32] Engelhardt MD, Popov EP. Experimental performance of long links in
Mech 1985;111(8):1010–26. eccentrically braced frames. J Struct Eng 1992;118(11):3067–88.
[12] Sivaselvan MV, Reinhorn AM. Hysteretic models for deteriorating inelastic [33] Arce G. Impact of higher strength steels on local buckling and overstrength of
structures. J Eng Mech 2000;126(6):633–40. links in eccentrically braced frames M.S. thesis. Austin (TX): Department of
[13] Song J-K, Pincheira J. Spectral displacement demands of stiffness-and strength- Civil, Architectural and Environmental Engineering, The University of Texas at
degrading systems. Earthq Spectra 2000;16(4):817–51. Austin; 2002.
[14] Pincheira J, Dotiwala FS, D’Souza JT. Seismic analysis of older reinforced [34] American Institute of Steel Construction. Seismic provisions for structural steel
concrete columns. Earthq Spectra 1999;15(2):245–72. buildings (1997) supplement no. 2” no. 2; 2001. p. 1–35.
[15] Rahnama M, Krawinkler H. Effects of soft soil and hysteresis model on seismic [35] Ryu HC. Effects of loading history on the behavior of links in seismic-resistant
demands Report no. 108. John A. Blume Earthquake Engineering Center, eccentrically braced frames MS thesis. Austin (TX): Department of Civil,
Department of Civil Engineering, Stanford University; 1993. Architectural and Environmental Engineering, The University of Texas at
[16] Ibarra L, Medina R, Krawinkler H. Collapse assessment of deteriorating SDOF Austin; 2005.
systems. In: Proceedings of the 12th European conference on earthquake [36] Richards P, Uang C-M. Development of testing protocol for short links in
engineering. p. 9–13. eccentrically braced frames Report No. SSRP-2003/08. San Diego: Department
[17] Ibarra LF, Krawinkler H. Global collapse of frame structures under seismic of Structural Engineering, University of California; 2003.
excitations Rep. No. TB 152. Stanford (CA): The John A. Blume Earthquake [37] Galvez P. Investigation of factors affecting web fractures in shear links M.S.
Engineering Center, Stanford Univ.; 2005. thesis. Austin (TX): Department of Civil, Architectural and Environmental
[18] Ibarra LF, Medina RA, Krawinkler H. Hysteretic models that incorporate Engineering, The University of Texas at Austin; 2004.
strength and stiffness deterioration. Earthq Eng Struc. Dyn 2005;34 [38] Liu X-G, Fan J-S, Liu Y-F, Yue Q-R, Nie J-G. Experimental research of replaceable
(12):1489–511. Q345GJ steel shear links considering cyclic buckling and plastic overstrength. J
[19] Lignos D, Krawinkler H. Sidesway collapse of deteriorating structural systems Constr Steel Res 2017;134:160–79.
under seismic excitations Rep. No. TB 172. Stanford, CA: The John A. Blume [39] Fedorov S. GetData graph digitizer; 2008. Available [Link]-graph-
Earthquake Engineering Center, Stanford Univ.; 2009. [Link].
[20] Medina R, Krawinkler H. Seismic demands for non-deteriorating frame [40] Hjelmstad KD, Popov EP. Seismic behavior of active beam links in eccentrically
structures and their dependence on ground motions Report No. 144. John A. braced frames. NASA STI/Recon Tech Rep N, vol. 84; 1983. p. 18480.
Blume Earthquake Engineering Center, Stanford University; 2003. [41] Malley JO, Popov EP. Design considerations for shear links in eccentrically
[21] Castaldo P, Nastri E, Piluso V. Ultimate behaviour of RHS temper T6 aluminium braced frames. NASA STI/Recon Tech Rep N, vol. 84; 1983.
alloy beams subjected to non-uniform bending: parametric analysis. Thin- [42] Kasai K, Popov EP. A study of seismically resistant eccentrically braced steel
Wall Struct 2017;115:129–41. frame systems. Earthquake Engineering Research Center, College of
[22] Güneyisi EM, D’Aniello M, Landolfo R, Mermerdasß K. Prediction of the flexural Engineering, University of California; 1986.
overstrength factor for steel beams using artificial neural network. Steel [43] Ricles JM, Popov EP. Experiments on eccentrically braced frames with
Compos Struct 2014;17(3):215–36. composite floors. Earthquake Engineering Research Center, University of
[23] D’Aniello M, Güneyisi EM, Landolfo R, Mermerdasß K. Analytical prediction of California; 1987.
available rotation capacity of cold-formed rectangular and square hollow [44] Engelhardt MD, Popov EP. Behavior of long links in eccentrically braced frames
section beams. Thin-Wall Struct 2014;77:141–52. (Report UBC/EERC-89/01). Berkeley: CA Earthq. Eng. Res. Center, Univ. Calif;
[24] D’Aniello M, Güneyisi EM, Landolfo R, Mermerdasß K. Predictive models of the 1989.
flexural overstrength factor for steel thin-walled circular hollow section [45] American Institute of Steel Construction. Seismic provisions for structural steel
beams. Thin-Wall Struct 2015;94:67–78. buildings (ANSI/AISC 341-10). Chicago (IL): American Institute of Steel
[25] Bosco M, Marino EM, Rossi PP. Modelling of steel link beams of short, Construction; 2010.
intermediate or long length. Eng Struct 2015;84:406–18. [46] Berman JW, Bruneau M. Experimental and analytical investigation of tubular
[26] Ramadan T, Ghobarah A. Analytical model for shear-link behavior. J Struct Eng links for eccentrically braced frames. Eng Struct 2007;29(8):1929–38.
1995;121(11):1574–80. [47] Della Corte G, D’Aniello M, Landolfo R. Analytical and numerical study of
[27] Martinez Palomino YM. Model fitting of a bilinear material with genetic plastic overstrength of shear links. J Constr Steel Res 2013;82:19–32.
algorithm M.S. thesis. Irvine: University of California; 2011. [48] McDaniel CC, Uang C-M, Seible F. Cyclic testing of built-up steel shear links for
[28] Kennedy J, Eberhart R. Particle swarm optimization. In: Proceedings of the IEEE the new bay bridge. J Struct Eng 2003;129(6):801–9.
international conference on neural networks, Perth, Australia. p. 1942–5. [49] Dusicka P, Itani AM, Buckle IG. Cyclic behavior of shear links of various grades
[29] Hassan R, Cohanim B, De Weck O, Venter G. A comparison of particle swarm of plate steel. J Struct Eng 2009;136(4):370–8.
optimization and the genetic algorithm. In: 46th AIAA/ASME/ASCE/AHS/ASC [50] Mazzolani FM, Della Corte G, D’Aniello M. Experimental analysis of steel
structures, structural dynamics and materials conference. p. 1897. dissipative bracing systems for seismic upgrading. J Civ Eng Manage 2009;15
[30] Poli R, Kennedy J, Blackwell T. Particle swarm optimization. Swarm Intell (1):7–19.
2007;1(1):33–57. [51] Gardoni P, Der Kiureghian A, Mosalam KM. Probabilistic capacity models and
[31] American Institute of Steel Construction. Seismic provisions for structural steel fragility estimates for reinforced concrete columns based on experimental
buildings. Chicago (IL): American Institute of Steel Construction; 2002. observations. J Eng Mech 2002;128(10):1024–38.