Journal of Earthquake Engineering

ISSN: 1363-2469 (Print) 1559-808X (Online) Journal homepage: http://www.tandfonline.com/loi/ueqe20

Shake Table Blind Prediction Tests: Contributions

for Improved Fiber-based Frame Modelling

Romain Sousa, João P. Almeida, António A. Correia & Rui Pinho

To cite this article: Romain Sousa, João P. Almeida, António A. Correia & Rui Pinho (2018):
Shake Table Blind Prediction Tests: Contributions for Improved Fiber-based Frame Modelling,
Journal of Earthquake Engineering, DOI: 10.1080/13632469.2018.1466743

To link to this article: https://doi.org/10.1080/13632469.2018.1466743

Published online: 15 May 2018.

Submit your article to this journal

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=ueqe20
JOURNAL OF EARTHQUAKE ENGINEERING
https://doi.org/10.1080/13632469.2018.1466743

Shake Table Blind Prediction Tests: Contributions for Improved

Fiber-based Frame Modelling
a
Romain Sousa , João P. Almeidab, António A. Correiac, and Rui Pinhod
a
UME School, IUSS Pavia, Pavia, Italy; bEarthquake Engineering and Structural Dynamics Laboratory (EESD),
School of Architecture, Civil and Environmental Engineering (ENAC), École Polytechnique Fédérale de
Lausanne, Lausanne, Switzerland; cNational Laboratory for Civil Engineering, Lisbon, Portugal; dDept. of Civil
Engineering and Architecture, University of Pavia, Pavia, Italy

ABSTRACT ARTICLE HISTORY

Distributed plasticity Euler-Bernoulli fiber beams are regularly used by Received 18 May 2017
researchers and practitioners to simulate the nonlinear seismic response Accepted 11 April 2018
of reinforced concrete framed structures. This study presents a sensitiv- KEYWORDS
ity analysis, combining newly proposed and existing goodness-of-fit Numerical Validation; Shake
measures, to identify commonly used engineering modeling options Table Tests; Reinforced
that most critically influence the response accuracy. The dynamic beha- Concrete Frames; Sensitivity
vior of three distinct structures tested in international blind prediction Analysis; Fiber Element
challenges is considered. Sensitivity parameters include equivalent vis- Formulation; Equivalent
cous damping, element formulation, strain penetration effects, and Viscous Damping; Strain
material models. The results are critically discussed in light of theoretical Penetration; Material Models
shortcomings of the model assumptions.

1. Introduction
Modern seismic design and assessment methodologies of structures tend to rely heavily on
numerical simulation tools. Their development has accompanied the growth of computa-
tional power, the advancement of solution algorithms, and the increasing availability of
experimental data for model calibration.
Past blind prediction tests showed that, by making use of specific combinations of
modeling options, the nonlinear dynamic response of reinforced concrete (RC) structures
can be predicted with satisfactory accuracy. Figure 1 compares the maximum experimen-
tally measured displacements and the numerical estimates submitted by the authors at two
different international blind prediction challenges. Although the authors won the compe-
tition for Structures 2 and 3 (among 38 international teams), and obtained an “Award of
Excellence” for Structure 1 (among 41 international teams), it can be observed in Fig. 1
that there is clear space for simulation improvements.
The comparison of other physically measurable quantities, in particular at a more
localized level, or the results of the participation in other competitions where the units
deformed and failed in non-considered modes, would further emphasize this preliminary
remark. The above-mentioned structures, which will be used throughout the present work,
are described along with the associated experimental program in Section 3.1.

CONTACT Romain Sousa [email protected] UME School, IUSS Pavia, Pavia, Italy.
Color versions of one or more of the figures in the article can be found online at www.tandfonline.com/ueqe.
© 2018 Taylor & Francis Group, LLC
2 R. SOUSA ET AL.

Figure 1. Comparison between experimental results and numerical simulation estimates submitted by
the authors at different blind prediction challenges: maximum positive and negative displacements for
ground motions of increasing intensity.

Additionally, the preceding blind-simulation challenges evidenced a significant disper-

sion of predictions between the participating teams, which is more meaningful upon
observation that such scatter takes place between teams using the same structural simula-
tion software as the winning group. This symptom of lack of consensual modeling
principles causes concern and highlights the need to identify the main sources of inaccu-
racy in nonlinear structural analysis so as to progressively reach an agreement on the
modeling options that minimize the gap between experimental and numerical response.
Figure 2, which shows the results achieved by the participants of the blind prediction pier
test corresponding to Structure 1 [Terzic et al., 2015], further supports the comments
above. It is noted that, even for such a simple structure (a single pier with a mass at the
top), the estimates for the response parameters submitted by the different teams vary in
some cases by more than one order of magnitude.
A first engineering choice, which is generally conditioned by the available structural soft-
ware, is related to the finite element to be used. Among the available approaches, frame models
developed in the framework of Euler-Bernoulli beam theory [Bathe, 1996] with distributed
plasticity are often considered to offer the best compromise between the degree of output detail

0.8 14000
Experimental Experimental
0.7 Winner team 1 Winner team 1
12000
Maximum horizontal displacement (m)

Winner team 2 Winner team 2

Maximum bending moment (kNm)

Winner team 3 Winner team 3

0.6
10000

0.5
8000
0.4
6000
0.3

4000
0.2

0.1 2000

0 0
1 2 3 4 5 6 1 2 3 4 5 6
Earthquake number Earthquake number

Figure 2. Comparison between experimental results and numerical simulations for the ‘Concrete
Column Blind Prediction Contest 2010ʹ: maximum horizontal displacement (left) and maximum bending
moment (right) (adapted from Terzic et al., (2015)).
 JOURNAL OF EARTHQUAKE ENGINEERING 3

and computational demand. Lumped plasticity approaches are simpler and computationally
lighter, but they do not allow modeling the spread of inelasticity throughout the member and
require the a priori knowledge of the location and extent of inelasticity, which is only possible
under limiting assumptions [Hines et al., 2004]. Moreover, lumped plasticity models require a
high level of expertise in order to define appropriate constitutive hysteretic relations for the
hinge taking into account the variations of axial force. Hereinafter, the present paper focuses
solely on distributed plasticity models.
Attempts have been made to identify and measure the importance of different modeling
options (e.g. Yazgan and Dazio [2011a; 2011b], Blandon [2012]). The current work intends to
consolidate some of these findings and to further extend them in order to progressively bridge
the gap between established theoretical principles and experimental results from shake table
tests. Based on the conclusions of the previous studies and preliminary investigations by the
authors, it was found pertinent to analyse the following modeling options given their potential
to more critically affect the simulation results and therefore the prediction of engineering
demand parameters (EDPs) on which performance assessment is based:

(1) Equivalent viscous damping models

(2) Element formulation and discretization scheme
(3) Strain penetration effects
(4) Material constitutive models

These variables will be used in the numerical applications of Section 4, wherein the shake
table responses of the three foregoing RC structures serve as benchmark. The goodness-of-
fit of each approach, assessed in terms of lateral displacements, as well as accelerations
when available, is determined in Section 4 based on the error associated with: (i) the peak
values measured during each response history record, and (ii) a new frequency-domain
approach that evaluates the records under comparison in terms of both amplitude and
frequency content. The methodology adopted for the sensitivity analysis is explained in
detail in Section 3. The next section recalls the theoretical framework enveloping the
aforementioned modeling options, along with a brief overview of the main literature
findings and current modeling recommendations.

2. Theoretical Considerations and Literature Review

2.1. Damping
2.1.1. Observed Damping in Structures
Damping in structures is generally associated with the decay of free-vibration motion due
to the energy dissipation mechanisms in structural and non-structural components and in
the foundation system. In real structures, several irreversible thermodynamic processes
concur to such decay, e.g. global damage in the components, internal friction between
materials or at connections, opening and closing of micro-cracks in the materials, and
friction between the structure itself and non-structural elements [Chopra, 1995].
Measuring and comparing damping of different structures subjected to distinct load-
ings is a favored way to understand the energy dissipation mechanisms in structures. The
relation between damping values and structural period of vibration has been recognized
4 R. SOUSA ET AL.

for a long time [Wakabayashi, 1986]. In Section 2.4.3.4. of the report PEER/ATC 72–1
[PEER/ATC, 2010], a summary of damping ratios inferred from decrements in peak-to-
peak response in free vibrations following shake table or pull-release tests shows that the
modal damping values measured in undamaged RC frame (or frame-wall) structures and
steel braced frame systems ranged from about 0.5% to 3.5% of critical. Not surprisingly,
significantly damaged structures exhibit modal damping ratios that can go up to 11%
[PEER/ATC, 2010]. The compilation of such results indicates that the classical assumption
of 5% of critical damping, traditionally adopted in the not-so-distant past as a reference
modal damping ratio for RC buildings, may result in a non-negligible overestimation of
the energy dissipated during elastic dynamic response.
Over the last two decades, a number of authors proposed predictive expressions for
modal damping ratios based on data measured from hundreds of existing buildings.
Among them, Jeary [1986] proposed ð%Þ ¼ f1 , applicable to tall buildings, where ξ is
the damping ratio for the first mode expressed in percentage and f1 represents the
fundamental frequency (in Hertz). Lagomarsino [1993] suggested ð%Þ ¼/ T1 þ β=T1 ,
where T1 stands for the fundamental period (in seconds), with α = 0.72 and β = 0.7 for RC
buildings and α = 0.32 and β = 0.78 for steel-framed buildings. More recently, Satake et al.
[2003] recommended ð%Þ ¼ 1:4f1 for RC buildings with total height between 10 m and
130 m, and ð%Þ ¼ 1:3f1 for steel buildings. From the previous expressions it is possible to
confirm that for RC buildings with 0.4 s < T1 < 2.5 s, using commonly available period-
height relations, such as the one provided in Eurocode 8 – Part 1 [CEN, 2004], the
damping estimations vary essentially between 0.5% and 3.5%.
Based on a large database of steel buildings with more than 20 stories, Kareem and
Gurley [1996] showed that the damping ratios tend to increase for higher modes of
vibration. However, the results present a large dispersion making difficult to establish a
clear trend, as demonstrated in more detail by Morita and Kanda [1996] and Arakawa and
Yamamoto [2004] for two steel buildings with 6 and 23 stories, respectively. More
recently, Bentz and Kijewski-Correa [2008] concluded that damping depends on the
relative participation of the structural system dominant deformation mechanisms: shear
or flexure. The study defends the hypothesis that damping increases with increasing
relative contributions of the shear mechanisms.

2.1.2. Numerical Issues Associated with Equivalent Viscous Damping

Within the scope of nonlinear frame analysis, the majority of the dissipated energy is
explicitly accounted for through hysteretic material models. The use of generalized
damping is however needed to reproduce sources of energy dissipation that, given their
phenomenological complexity, are difficult to be explicitly simulated. Hence, the required
damping for a given nonlinear dynamic analysis will necessarily depend on the level of
detail of the model and the capabilities of the numerical tool used. In view of the
difficulties to identify and mathematically describe each of these dissipating mechanisms
in actual structures, damping is usually represented in a highly idealized manner by a set
of generalized linear viscous dampers. The damping coefficients are defined so that the
dissipated energy is approximately equivalent to the energy associated to all non-modeled
dissipation mechanisms. This idealization, first introduced by Jacobsen [1930], is therefore
called equivalent viscous damping (EVD). For the overwhelming majority of structural
engineering software, the latter takes on the form of Rayleigh damping, which combines
 JOURNAL OF EARTHQUAKE ENGINEERING 5

stiffness- and mass-proportional components. Its computational attractiveness comes

from the fact that it does not require the computation of new matrices, while keeping
the sparsity of the stiffness matrix [Chopra and Mckenna, 2017]. The main problems of
this approach are briefly summarized below.
First off, stiffness-proportional damping (SPD), regardless of being initial stiffness-
proportional damping (ISPD) or tangent stiffness-proportional damping (TSPD), may
introduce artificial and significantly large axial forces in the members, leading to conver-
gence issues and potential instability during the analyses. Moreover, as demonstrated by
Correia et al. [2013], when employing stiffness-proportional damping, the member forces
are not in equilibrium with the support reactions.
According to ISPD, damping forces remain proportional to their elastic stiffness
regardless of the ductility demand. Hence, this solution may result in an overestimation
of the energy dissipated through viscous damping. Analyses performed by Priestley and
Grant [2005] indicate that the energy absorbed by elastic damping in ISPD may approx-
imate the energy dissipated through hysteresis, even for large ductility levels.
Consequently, it is very likely that such damping model produces a significant under-
estimation of peak response displacements during nonlinear dynamic analyses. Moreover,
as in general the structural stiffness decreases during earthquake loading, the damping
forces may assume unrealistically large values when compared to the member (restoring)
forces [Hall, 2006]. Furthermore, Bernal [1994] showed that spurious damping forces are
likely to arise in massless degrees of freedom (or with relatively small inertia); the latter
tend to undergo abrupt variations in velocity when stiffness changes, leading to unrealis-
tically large viscous damping forces [Jehel et al., 2013].
TSPD appears clearly preferable to ISPD [Grant et al., 2005] since it: (i) significantly
reduces spurious damping forces [Charney, 2008], and (ii) models a decrease in the
dissipated energy for increasing ductility demands, which can be considered as intuitively
meaningful. On the other hand, Chopra and Mckenna [2017] criticize its physical and
conceptual basis, arguing that sudden drops in damping force when yielding occurs, and
triangular loops in force-velocity relations, are unacceptable. Moreover, when the stiffness
matrix becomes negative-definite, TSPD leads to an unrealistic energy input in the
structure.
Mass-proportional damping, despite not having a physically justifiable support, does
not exhibit the numerical deficiencies of the previous models. Nonetheless, Hall [2006]
reported several weaknesses when large rigid-body modes occur, leading to excessively
high velocities; such situation is not common in typical structures but may be
important, e.g. when dealing with base isolated buildings. In addition, the damping
ratio in MPD models decreases exponentially with a decrease of the vibration period,
thus possibly leading to an underdamped higher-mode response.
Recently, several studies addressed some of the abovementioned limitations of Rayleigh
damping. They either defend the application of existing alternatives, such as superposition
of modal damping matrices [Chopra and McKenna, 2016], or propose new models
[Puthanpurayil et al., 2016; Lanzi and Luco, 2017]. However, the present study restricts
to the use of Rayleigh damping as it is the only model widely implemented in analysis
software and available for engineering practice.
6 R. SOUSA ET AL.

2.2. Element Formulation and Discretization Scheme

In order to simulate the spread of inelasticity along the member, Euler-Bernoulli dis-
tributed plasticity approaches require sufficient discretization, which involves mesh refine-
ment and number of integration points (IPs). Such issue is discussed in the following
section, wherein the classical displacement-based (DB) and force-based (FB) formulations
are analysed. Similarly to Rayleigh damping, these two approaches were chosen as they are
widely available in structural analysis software. Additionally, under certain conditions a
phenomenon known as strain localization (or simply localization) occurs. In frame
elements, it can affect both lumped plasticity models [Jirásek, 1997] and distributed
plasticity models. The triggering conditions and the description of how it manifests in
this latter category of models are herein recalled in a separate discussion provided in
Section 2.2.2.

2.2.1. Numerical Accuracy

The evaluation of the basic element forces (and corresponding stiffness matrix) for DB
formulations, or of the basic element displacement (and corresponding flexibility matrix)
for FB formulations, requires the evaluation of an integral along the element length
[Neuenhofer and Filippou, 1997]. Hence, if the problem is well-posed, i.e. if we disregard
the possible occurrence of localization, the larger the number of integration points (for FB
formulations) or the number of elements per member (for DB formulations), the more
accurate will be the numerical integration output. However, an adequate simulation of the
spread of inelasticity along the structural member does not require an extremely refined
mesh or countless integration points (IPs) per member to obtain a close-to-exact response.
Determining the level of refinement for which the results are acceptably close to the exact
response is important to keep the model as computationally light as possible. To do so it is
relevant to bear in mind that the response can be assessed at the global or/and at the local
level as well; i.e., respectively in relation to nodal displacements or curvatures at the IPs.
Each level exhibits distinct convergence properties, as discussed hereinafter. It is also
underlined that the aforementioned exact response holds for Euler-Bernoulli beam theory,
which can nonetheless deviate considerably from experimental results. This point will be
discussed later in Section 3.2. Since a sufficiently discretized scheme holds different
meanings depending on whether a DB or a FB approach is adopted, it will be addressed
separately in two following paragraphs.
The cubic interpolation function for the transverse displacement field used in DB
formulations [Bathe, 1996] creates an artificial restraint among the IPs of the same DB
element concerning the development of inelastic curvatures. The insensitivity of DB
meshes to the increase of the number of IPs per element is intimately related to such
feature; therefore, it is not justified to use more than two IPs per element, alongside with a
Gauss-Legendre quadrature rule. It is recalled that the latter leads to the exact integration
of a polynomial up to degree 2n-1, where n is the number of IPs. Capturing the spread of
inelasticity can thus only be achieved at the expense of refining the element mesh. A
balance, which depends on the loading and boundary conditions, should be struck
between the desired numerical accuracy and computational cost. In general, a member
model with three elements provides an unrealistically stiff response at both the global and
local levels. That effect is even more pronounced with two elements—see also Neuenhofer
 JOURNAL OF EARTHQUAKE ENGINEERING 7

and Filippou [1997], whilst a one-element mesh is wholly unable to simulate the member
inelastic behavior. A minimum of four to six elements per structural member are usually
required. However, whilst the previous rule of thumb typically provides close-to-exact
results at the global level, it can simultaneously fall short of yielding acceptable local-level
responses: computed inelastic curvatures may underestimate significantly the exact value.
Such apparently contradictory feature can be attributed to the fact that DB elements only
satisfy equilibrium in an average (integral) sense. If a reliable local-level response is also
sought, a very refined DB mesh—around one order of magnitude larger than the pre-
viously mentioned rule of thumb—may have to be considered [Calabrese et al., 2010].
Unlike the DB approach, force interpolation functions—which strictly verify equilibrium
—used in FB elements do not restrain the expected development of inelastic curvatures
along the IPs of the same element. Consequently, only one element per structural member is
required [Calabrese et al., 2010]. Furthermore, the use of a Gauss-Lobatto integration
scheme is advised as it controls the end sections of the element, which are privileged
locations for the occurrence of seismic inelastic response. Consider now the linear elastic
response of a planar beam-column element with six degrees of freedom (three at each end
node), under geometric linear behavior. It is straightforward to demonstrate that the
corresponding stiffness matrix, which is a classical result of introductory structural theory,
can only be exactly retrieved when three or more Gauss-Lobatto IPs are used. The use of
two Gauss-Lobatto points per FB element should be avoided as under-integration occurs
and hence not even the simple linear elastic response can be simulated [Scott and Fenves,
2006]. Simulating the spread of inelasticity may however claim for more than three IPs. A
good compromise between numerical accuracy and computational time typically calls for
around five IPs to obtain accurate global-level results, while a close-to-exact local-level
output usually requires a marginally larger number of IPs (around six to seven). It is
recalled that the counterpart local-level accuracy requirements for the DB-mesh counterpart
are incomparably more stringent [Calabrese et al., 2010].

2.2.2. Numerical Localization

Formally, localization corresponds to the loss of ellipticity of the governing differential
equations, resulting in the loss of uniqueness of its mathematical solution. The initial
boundary-value problem is thus mathematically ill-posed and the numerical solution becomes
mesh-dependent [Coleman and Spacone, 2001]. From the computational mechanics’ view-
point, localization is intrinsically related to softening or post-peak constitutive behavior. In the
context of distributed plasticity frame models, localization occurs following the attainment of
a post-peak descending branch of the moment-curvature relation at a given IP.
The problem of localization can perhaps be better understood through the lens of the
auxiliary notion of objectivity [Coleman and Spacone, 2001]. Consider several DB and FB
discretization schemes with an increasing degree of refinement: if the corresponding
global and local-level responses converge to the same (exact) value, the response is said
to be objective. If instead divergent results show up, the response is deemed as non-
objective and localization takes place, with inelastic curvatures concentrating at one IP.
As stated above, the distinctive fundamental trait of this concentration of inelastic
curvatures is that it only occurs upon reaching a softening branch of the moment-
curvature relation. This concentration corresponds to the solution of an ill-posed
problem and therefore it cannot be directly related to a corresponding physical
8 R. SOUSA ET AL.

meaning. Additionally, it is accompanied by simultaneous unloading in the neighbor-

ing IPs (or elements, see discussion in the next section). In order to maintain post-
peak objectivity, regularization techniques must be employed, which are briefly
addressed in the following section.
On the other hand, if the sectional response of every IP depicts a fully hardening post-
yield behavior or, equivalently, if the member is not loaded beyond the attainment of a
possible peak of the sectional moment-curvature curves the model response is objective.
Hence, arising inelastic curvatures correspond to a well-posed problem (the solution exists
and is unique) and carry a clear physical meaning. Their accuracy remains nonetheless
dependent on the adoption of a sufficiently discretized scheme, as discussed above in the
previous section. Therefore, it should be clear that the concentration of curvatures in one
element or IP does not necessarily imply localization, in the sense of non-objectivity. On
the other hand, if localization does occur, its adverse effects may not be relevant if the
incursion in the post-peak branch is limited.

2.2.3. Regularization Techniques

The abovementioned features of localization call for ways of maintaining the post-peak
objectivity of the results. These constitute the so-called regularization techniques, which
require a user-defined parameter specifying the physical dimension of the region into
which curvatures localize. In earthquake engineering, the latter usually corresponds to the
expected length of the plastic hinge. Some authors [Almeida et al., 2016] have explicitly
adopted the term real plastic hinge length to emphasize that confusion often arises with
existing well-known formulas for the equivalent plastic hinge length, which is in principle
only applicable to plastic hinge analysis. This point is further discussed in Section 3.2 with
an application example.
Regularization techniques were first applied to DB elements since they correspond to the
original formulation of the Finite Element Method. For nonlinear fiber frame elements,
wherein the interaction between axial force and bending moment is accounted for at each
IP, the average verification of equilibrium in DB formulations poses various difficulties in the
interpretation of the element behavior. In particular, it is impossible to tell a priori if
localization will take place in the most strained IP [Calabrese et al., 2010] or simultaneously
in both IPs of the element [Almeida et al., 2016]. In fact, the distinct axial forces determined in
each of the two member IPs—their value being symmetric with respect to the applied element
axial load—induce different moment-curvature relations and eventually distinct sectional
flexural capacities. The largest moment demand may thus occur at an IP whose flexural
capacity is artificially larger than the one of the neighboring integration point and conse-
quently the softening branch of the moment-curvature curve may be attained first in this latter
IP (and only at larger inelasticity demands in the former IP). In short, it is not possible to know
in advance, i.e. before running the analysis, how to best regularize the response.
The first regularization method for FB elements was suggested by Coleman and
Spacone [2001]. Due to its time-consuming two-stage application, more recent techniques
were proposed [Scott and Fenves, 2006; Addessi and Ciampi, 2007]. They are based on ad
hoc integration schemes and the observation that, unlike DB approaches, the post-peak
accumulation of curvatures in FB elements is bound to happen in the most strained IP;
therefore, such approaches scale the integration weights at the element ends to match the
value of the expected plastic hinge length. The main problem with such approaches is that
 JOURNAL OF EARTHQUAKE ENGINEERING 9

they can simulate an unrealistically flexible response when an objective strain-hardening

constitutive behavior occurs. To circumvent this problem, a new quadrature rule was
advanced by Scott and HamutçuoğLu [2008], which includes two additional IPs at the
extremities. More recently, a versatile adaptive integration scheme that warrants high
numerical accuracy in the pre-peak phase and a regularized response in the post-peak
branch was proposed by Almeida et al. [2012]. Although work on this field continues to
the present day [Feng et al., 2016; Vásquez et al., 2016], explicit regularization methods are
not readily available in most engineering analysis software. Therefore, in the present work,
regularization was indirectly addressed by defining different element lengths and number
of IPs, as discussed in Section 3.2.

2.3. Strain Penetration Effects

RC members subjected to flexural loading often depict localized deformations occurring at
beam-column and/or column-footing joints. Although relatively large anchorage lengths
are provided in modern seismic design, the corresponding deformations should be
accounted for in analysis. This mechanism can become critical in older RC structures
with deficient anchorage detailing and/or smooth rebars with much-reduced lower bond
strength, leading possibly to anchorage failure as observed in a recent blind-test challenge
[JPEE, 2014]. The fixed-end rotations result from the spread of reinforcing strains along
the anchorage region, which causes strain incompatibility between the reinforcement and
the surrounding concrete and potential bond-slip deformations along the anchorage
length. These phenomena are often referred in the literature as strain penetration effects
(SP, herein adopted), bond behavior, anchorage slip or bond-slip mechanisms, among
others. If adequate anchorage length is provided, the deformation results solely from
penetration of tensile strains along a fraction of the anchorage length. However, if
inadequate anchorage is provided, the strain penetration may extend through the entire
anchorage length, leading to total slippage of the anchored rebar. According to Sezen and
Moehle [2004] and Goodnight et al. [2015], the lateral displacement of RC members can
be increased by up to 40% due to SP effects. Unfortunately, SP deformations are still
neglected in most numerical simulations for seismic analysis.
Micro-scale approaches use 3D solid finite element models to explicitly simulate the
interaction between the longitudinal rebars and the surrounding concrete (e.g. [Salem and
Maekawa, 2004] and [Jendele and Cervenka, 2006]), which are essentially based on
fracture mechanics employing advanced material and bond constitutive relations.
Notwithstanding their obvious simulation capabilities, such approaches are computation-
ally demanding and hence generally inapplicable to analyse large structures. Monti and
Spacone [2000], Girard and Bastien [2002], Lee and Filippou [2015a] and Lee and Filippou
[2015b] presented RC beam elements that explicitly account for the slip between the
reinforcement bars and the surrounding concrete at the cross-sectional level. The advan-
tage of these models is that the SP deformations are directly accounted for at the member
formulation, avoiding the need to define additional elements or assume supplementary
modeling considerations. Such models still tend to be computationally heavy given the
extensive discretization required to accurately capture the rebar response along the
embedded length [Zhao and Sritharan, 2007]. Furthermore, they are usually not available
in current analysis software, which prevents their generalized use by structural engineers.
10 R. SOUSA ET AL.

In view of the limitations above, link elements are alternatively used for frame analysis,
although still somewhat restricted to the research community. They are based on the
assumption that SP deformations can be lumped into a single element located at the
member end(s), usually in the form of a zero-length element. Their properties are
generally based on empirical data and can be assigned at the material level (reinforcement
and concrete) [Zhao and Sritharan, 2007] or at the cross-sectional level with suitable
moment-rotation relations [Sezen and Moehle, 2004]. Practicing engineers often look for
more simplified approaches readily applicable to the generality of available structural
analysis software.
One of the simplest approaches often used in practice—which unfortunately misses
clear application guidance—involves the reduction of the Young’s Modulus of the reinfor-
cing steel to simulate the increase in member flexibility due to SP effects. According to a
study carried out by Varum [2003], the ratio between the Young’s Modulus for perfect
bond conditions and the one adjusted to account for slippage varies with the properties of
the rebar surface and amplitude of the imposed deformations, and can reach values larger
than 20 for smooth rebars subjected to large deformation demands.
Alternatively, the member can be extended of an estimated strain penetration length,
which represents another common and simple method that relies on the assumption that
bond deformations spread along the SP length, contributing to the overall member defor-
mation. In the framework of plastic hinge analysis [Fardis, 2009; Priestley et al., 2007],
different expressions have been empirically calibrated to provide an estimation of the
equivalent SP length (LSP) at the onset of flexural failure. The following Eqs. (1) and (2),
by Priestley et al. [2007] and Eurocode 8 – Part 3 [CEN, 2005] respectively, are widely used:


LSP ¼ 0:022fye dbl fye in MPa (1)

fye dbl

LSP ¼ k pffiffiffi fye in MPa (2)
fc

where fye and dbl are the expected yield stress and diameter of the longitudinal rebars, fc is
the concrete compressive strength, and k can assume values of 0.11 or 0.24 depending on
the concrete confinement model considered to estimate the overall member deformation. If
an improved concrete model is employed (k = 0.11), both expressions produce similar
values for a typical concrete strength (fc≈30MPa). Moreover, despite being calibrated to be
applied at ultimate condition, LSP seems to be also valid at flexural yielding [Fardis, 2009]. It
is noted that a recent extensive experimental program on bridge columns, wherein detailed
rebar strain and slip measurements were obtained, showed that a constant equivalent strain
penetration length was suitable to describe the measured fixed-end rotations throughout a
large range of curvature ductility demands [Goodnight et al., 2014]. A setback of this
approach is that the modified member becomes artificially more flexible up to yield, when
the strain penetration mechanism is expected to be fully developed over Lsp. The additional
deformability is spread along the length of the artificially elongated member instead of
corresponding to a localized flexibility at the element end(s). Moreover, given that the shear
span of the element increases for the same sectional flexural capacity, the shear demand will
necessarily be underestimated. However, the magnitude of this error is limited, given that
the strain penetration length is of the order of 5% of the member length. Finally, it is noted
 JOURNAL OF EARTHQUAKE ENGINEERING 11

that this SP modeling option becomes increasingly difficult to apply as the number of bays
and stories increases, due to evident geometrical compatibility issues.
One last SP modeling option herein recalled, which overcomes some of the previous
limitations, considers an elastic rotational spring at the base of the columns, thus keeping
the element geometry unchanged. The elastic stiffness of the spring can be determined such
that the stiffness of the system with the spring is the same than the stiffness-at-yield of a similar
element, fixed at the base, but elongated by the strain penetration length [Correia, 2011].
Equating the system stiffness at yielding reflects the assumption that the strain penetration
length is essentially fully mobilized when the longitudinal rebars yield. A graphical interpreta-
tion is shown in Fig. 3, from which the rotational stiffness can be derived:

L2c MN 3L2c
Kθ ¼ 3EI h
i¼ h
i (3)
Lc þ Lsp
3
L3c ϕy Lc þ Lsp 3 L3c

where LSP can be computed with Eq. (1) or (2), MN is the nominal yield moment, and ϕy is
the yield curvature, which can be determined from a bilinear idealization of a cross-
sectional moment-curvature analysis.

2.4. Material Stress-Strain Models

The uniaxial stress-strain relations adopted for both concrete and steel affect the cross-
sectional response at each integration point of the element and correspond to the primary
source of structural material nonlinearity. Among the available uniaxial material laws, and
despite the numerous models available in literature, Mander’s concrete model [Mander
et al., 1988] and Menegotto-Pinto steel model [Menegotto et al., 1973] prevail as two of
the most commonly adopted for nonlinear analyses of RC structures. Previous studies
addressing the effect of the selected material models on the computed response of RC
structures (e.g. the works by Yazgan and Dazio [2011b] and Blandon [2012]) have shown
that the choice of different material models, in particular for steel, may have a significant
impact on the computed maximum and residual displacements.

Figure 3. Strain penetration modeling approaches: elongated element (left) and base spring (right)
[Correia, 2011].
12 R. SOUSA ET AL.

2.4.1. Concrete and Confinement

The parameter that most affects the maximum concrete strength and corresponding strain
is the confining reinforcement detailing. In the parametric study carried out in the
following sections, four models considering three concrete constitutive relations with
different confinement models were considered. The first two models adopt the uniaxial
nonlinear model proposed by Mander et al. [1988] with the cyclic rules suggested by
Martinez-Rueda and Elnashai [1997]. The confinement effects provided by the transverse
reinforcement are incorporated through the rules advanced by Mander et al. [1988],
whereby a constant confining pressure is assumed throughoutpthe ffiffiffi entire stress-strain
range. The reference model considers a tensile strength of 0:75 fc [MPa], as proposed
by Priestley et al. [1996] for members subjected to flexure, where the tensile stress is
assumed to increase linearly with strain up to the tensile strength followed by a subsequent
abrupt loss of tensile resistance. The second model neglects tensile behavior.
The third model follows the same constitutive relation assumed by Mander et al. [1988]
but with the proposal by Madas and Elnashai [1992], which considers a confinement
factor that varies during the analysis. This coefficient depends on the cross-section
dimensions, the mechanical characteristics and detailing of the transversal reinforcement,
and on the sectional axial load at each step of the analysis.
Following the observation that high dispersion can be found between the different
stress-strain relations beyond the peak stress in compression, Kappos and Konstantinidis
[1999] proposed an alternative constitutive model, which is the fourth model herein
considered. It was calibrated for high-strength concrete and follows the relation proposed
by Nagashima et al. [1992]. It incorporates confinement effects through the modified
Sheikh and Uzumeri [1982] factor, the so-called confinement effectiveness coefficient, and
assumes constant confinement throughout the entire stress-strain range.

2.4.2. Reinforcement
The well-known stress-strain relation proposed by Menegotto et al. [1973] has shown to be
numerically stable and able to simulate the main behavioral features of complex cyclic
loading histories. The Bauschinger effect is accounted for through the definition of a pair of
parameters (A1 and A2) that adjust the shape of the transition curves. Since the mechanical
properties of the rebars are defined according to test results, the next parametric study
addresses different unloading/reloading transition curves. Two models are considered: the
reference one assumes a value of 0.15 for A2, which for the second A2 = 0.075 is chosen.
The latter yields smoother transition curves and facilitates numerical convergence. However,
excessively low values of A2 can underestimate hysteretic energy dissipation.
The third model considered is the simple bilinear uniaxial model, which neglects the
Bauschinger effect and hence overestimates energy dissipation. Finally, the fourth steel
model corresponds to the Menegotto-Pinto stress-strain relation with the buckling rules
proposed by Monti and Nuti [1992]. It accounts for the reinforcing bar slenderness ratio
(s/dbl) in the response of longitudinal rebars undergoing large inelastic deformations,
where s is the specified tie spacing and dbl is the longitudinal bar diameter. In order to
overcome stability issues associated with small strain reversals, the considered model
incorporates the improvements introduced by Fragiadakis et al. [2008].
 JOURNAL OF EARTHQUAKE ENGINEERING 13

3. Methodology for Sensitivity Analysis

3.1. Case Studies
Three different structures, tested in two international blind prediction challenges, were
selected to perform a sensitivity study. The current section presents a brief description of
the selected structures and of the corresponding experimental programs.

3.1.1. Structure 1
The first structure to be analysed is a full-scale RC bridge column used in the “Concrete
Column Blind Prediction Contest 2010,” sponsored by the Pacific Earthquake Engineering
Research Center (PEER) and the Network for Earthquake Engineering Simulation (NEES).
The specimen was tested on the NEES Large High-Performance Outdoor Shake Table
located at UCSD’s Englekirk Structural Engineering Center. Detailed description of the
experimental test can be found in Terzic et al. [2015] and Schoettler et al. [2015].
The structure is a cantilevered column with a height of about 7.3 m from the face of the
foundation up to the center of gravity of the top RC block with 230 tons representing the
superstructure mass, as shown in Fig. 4. The circular cross-section has a diameter of
approximately 1.2 m.
The analytical modal properties of the bridge column identified through eigenvalue
analyses match well with the experimental ones (both with T1 = 0.78 s) and evidence that
the response is essentially governed by the 1st translational mode. Nevertheless, the
concrete block on the top of the column introduces additionally a significant rotational
inertia that governs the second mode of vibration (T2 = 0.14 s) and partially affects the
fundamental mode (the effective modal masses for the 1st mode equal 84% and 15% for
the transversal and rotational component respectively). Numerical simulation results
indicate that, despite the post-yield hardening behavior of the sectional response, the
global force-displacement curve shows a noteworthy softening behavior that reflects the
important influence of second-order effects. The experimental campaign consisted in the

Figure 4. Structure 1: Elevation view (left; units in m) and cross-section details (right).
14 R. SOUSA ET AL.

Figure 5. Series of applied acceleration records to Structure 1.

application of six consecutive ground motions of varying intensity. The test protocol
included historical ground motions such as those of 1989 Loma Prieta and 1995 Kobe
earthquakes. Figure 5 illustrates the set of earthquake records (Eqk) considered in the
experimental test.
During the experimental test, the column was subjected to curvature and displacement
ductility demands of 12.6 and 6.3, respectively, measured during record 5 [Carrea, 2010].
Idealized yielding curvatures and displacements are considered. The bridge pier experienced
damage accumulation, namely with the yielding of longitudinal rebars (record 1), crack
propagation (record 2), spalling of the cover concrete at the base and residual damage
(record 3), and spalling of concrete core (record 5). The test protocol encompassed also the
simulation of aftershock effects (records 4 and 6), where no relevant additional damage was
observed in the structure. Additional details can be found in Carrea [2010].

3.1.2. Structure 2 and Structure 3

The second group of structures considered in this study is two 3D RC frames with
identical geometrical dimensions (Fig. 6) but different reinforcement detailing. Both
structures were tested at LNEC (National Laboratory for Civil Engineering, Portugal)
3D shake table, which can impose simultaneously movements in the three translational
directions whilst passively restraining the three rotations [Campos Costa et al., 1996]. The
experiments were carried out under the initiative “Blind Test Challenge” included in the
15th World Conference on Earthquake Engineering.
As stated before, the two structures differ only with regard to reinforcement details, see
Table 1. Structure 2 was designed without capacity design considerations, while Structure
3 was detailed in order to develop an appropriate seismic behavior with a ductile response
(Fig. 7). In addition to the self-weight of the mock-up, a set of nine additional masses of
around 1200 kg each were placed eccentrically on top of the slab [Costa et al., 2012].
Considering that the geometric properties of both structures are identical, it is not
surprising that the modal (elastic) properties of Structure 3 are similar to those of Structure 2.
The behavior of the two structures is essentially governed by the two first modes of
vibration along the orthogonal horizontal reference axes (T1 = 0.31 s along X and T2 = 0.28 s
along Y); the third mode (T3 = 0.17 s) represents the vertical flexibility of the slab. A fourth
mode of vibration (T4 = 0.15 s), reflecting a rotation around the vertical axis, contributes
significantly to the torsional response of the structure. Pushover analyses along the X and Y
 JOURNAL OF EARTHQUAKE ENGINEERING 15

Figure 6. Structures 2 and 3: General dimensions of the mock-up.

Table 1. Structures 2 and 3: Section and material details (rebar diameters in mm).
Columns Beams
Long.Rebars(ρl) Transv.Rebars(ρv) Long.Rebars(ρl) Transv.Rebars(ρv)
Structure 2 ø[email protected](0.57%) 2ø10 + 3ø12(0.62%) ø[email protected](0.63%)
8ø10(1.57%)
Structure 3 ø[email protected](1.71%) 4ø10(0.39%) ø[email protected](1.26%)
ρl – Longitudinal reinforcement ratio
ρv – volumetric ratio of transverse reinforcement
f’y (Mpa) fu (Mpa) f’c, columns (Mpa) f’c, beams (Mpa)
ø8 560 653
ø10 559 632 30 35.6
ø12 566 631
f’c – Concrete compressive strength.
fy – Yield strength of steel.
fu – Steel ultimate stress.

Figure 7. Beam-column joint reinforcement details of Structure 2 (left) and Structure 3 (right).
16 R. SOUSA ET AL.

Figure 8. Series of applied acceleration records on Structure 2 in the longitudinal (Y) direction.

directions show that the two structures exhibit a softening response at the global level that is
a combination of second-order effects and material softening at the column’s cross-sections.
Both structures were subjected to four ground motions of increasing intensity applied
simultaneously in the two horizontal directions. The input signals were synthetically
generated based on a time segment of two horizontal orthogonal components of a real
strong motion accelerogram registered during the March 11, 2011 Great East Japan
(Tohoku) earthquake. After the first record, the specimens showed small amplitude cracks
in all structural elements. In the subsequent records damage tended to concentrate at the
top and bottom of the columns, including buckling of the reinforcement in some loca-
tions. After the experimental test, the damage in Structure 2 was seen to be more
significant than in Structure 3. Figure 8 represents the input time-history of Structure 2
in the longitudinal (Y) direction.

3.2. Sensitivity Parameters

Following the considerations of Section 2, the present study analyses the following
sensitivity parameters:

● Equivalent Viscous Damping – Four different models are considered: no damping, initial
stiffness-proportional damping (ISPD), tangent stiffness-proportional damping (TSPD),
and mass-proportional damping (MPD). Additionally, varying percentages of critical
damping, ranging from 0% to 5%, were assigned to the fundamental period of vibration.
● Element Discretization – The force-based (FB) approach was used to simulate the
column response, to the detriment of the DB counterpart, since it strictly satisfies the
differential equilibrium equations governing the original formulation of the Euler-
Bernoulli beam problem. The number of elements and the number of IPs per
element were defined according to the considerations in Section 2.2 and estimates
of the plastic hinge length.

It is important to recall that the equivalent plastic hinge length is a fictitious quantity from
which inelastic member deformations can be computed based on the assumption of an inelastic
curvature profile, usually taken as constant. It is distinct from the real plastic hinge, taken as a
physical length wherein damage concentrates [Almeida et al., 2016]. A well-known expression
for the equivalent plastic hinge length (LP) was proposed by Priestley et al. [2007]:
 JOURNAL OF EARTHQUAKE ENGINEERING 17

Table 2. Estimates of plastic hinge lengths for the case studies.

Equivalent plastic hinge
length (m) Real plastic hinge (m)
Priestley et al.[2007] Bae and Bayrak[2008] Pam and Ho[2009] Mortezaei and Ronagh, 2011
Structure 1 0.53 + 0.41 = 0.94 0.3 (without LSP) 1.2 (without LSP) 0.72/0.54 (far/near fault)
Structures 2 & 3 0.04 + 0.12 = 0.16 0.05 (without LSP) 0.21 (without LSP) 0.12/0.09 (far/near fault)

LP ¼ kLC þ LSP  2LSP (4)

 
fu
where k ¼ 0:2 fy 1  0:08 accounts for the strain-hardening steel properties, LC is the
shear span of the element (i.e. the member end moment-to-shear ratio), fy and fu are the
yielding and ultimate strength of the longitudinal rebars, and LSP is the strain penetration
length determined by Eq. (1). Some recent expressions are available in the literature to
estimate the real plastic hinge length, such as those of Bae and Bayrak [2008], Pam and Ho
[2009], and Mortezaei and Ronagh [2011], or for circular bridge piers that of Goodnight et al.
[2016]. The application of the previous expressions is summarized in Table 2 . It is noted that
the expressions proposed by Bae and Bayrak [2008] and Pam and Ho [2009] do not account
for the strain penetration contribution, while the ones from Mortezaei and Ronagh [2011]
vary with the distance to the earthquake fault.

Unfortunately, it can be observed that such proposals output a wide range of results, which expose
the current difficulty in distinguishing estimates for the equivalent and real plastic hinge lengths.
The former should be significantly smaller than the latter. While additional research is not
available to resolve this apparent conflict, the equivalent plastic hinge length as calculated by
Eq. (4) was used to define the considered element discretization schemes.

Table 3 summarizes the three considered combinations for the mesh and number of IPs for the
columns in each structure: (i) a model with a number of IPs such that the weighted length of the
extremity IP corresponds to the computed plastic hinge length (reference model), (ii) a model
where the extremity IP weights half of the assumed plastic hinge length, and (iii) a model
featuring 10 IPs, representative of a highly refined integration scheme. The IP weighted lengths
shown in Table 3 were determined based on the weight factors associated with the Gauss-Lobatto
integration as found in the literature, e.g. Akin [2000], and the length of each element. The
number of elements used to model the beams of Structure 2 and Structure 3 varied depending on
the reinforcement details along the member.

● Strain-Penetration Effects – Three alternative SP engineering modelling approaches

were studied: (1) base spring with constant rotational stiffness, (2) element elongated
by the strain penetration length, and (3) strain penetration effects neglected. For the
first approach, the column rotational stiffnesses were determined with Eq. (3), resulting

Table 3. Force-based discretization scheme adopted for the case studies.

End IP ≈ LP End IP ≈ 0.5 LP 10 IPs
No. No. IP Weighted No. No. IP Weighted No. No. IP Weighted
Elem. IPs Length (m) Elem. IPs Length (m) Elem. IPs Length (m)
Structure 1 2 3+3 1 2 4+3 0.5 1 10 0.08
Structures 2 & 3 1 4 0.23 1 6 0.09 1 10 0.03
18 R. SOUSA ET AL.

Figure 9. Cyclic response of the same concrete fiber (left) and steel fiber (right) in the base section of
Structure 1 using distinct uniaxial models, for the same cyclic history of imposed lateral displacements.

in the following values: Kθ = 2610x103 kNm/rad, Kθ = 7000 kNm/rad and Kθ = 7450

kNm/rad for Structure 1, 2, and 3 respectively. For the elongated element approach, the
strain penetration length was determined based on Eq. (1), which yielded the following
values: Lsp = 0.41 m for Structure 1 and Lsp = 0.12 m for Structures 2 and 3. Despite the
approximate double bending response of columns in Structures 2 and 3 under lateral
loads, the members were elongated of only Lsp (and not of 2× Lsp), corresponding to
the deformations taking place at the base of the columns. It was hence considered that
the strain penetration effects at the beam-column joints are somehow accounted for in
the numerical model by connecting the columns with the beams at the geometric
centre of the joint, i.e. by disregarding the so-called rigid offsets.
● Material models – The numerical analyses feature eight different uniaxial stress-
strain relations, as discussed in Section 2.4 (four for concrete and four for the
rebars). Figure 9 (left) illustrates the response under compression of four con-
crete constitutive models, for the same fiber at the base section of Structure 1,
when subjected to a cyclic history of lateral displacements. It is observed that
distinct confinement models modify significantly the stress-strain response, in
particular for strains beyond the peak stress. Mander’s model follows the less
steep softening branch when compared to the other confined concrete models.
Consequently, it is expected that member response wherein Mander’s model is
employed will in general show less pronounced localization effects. On the other
hand, Fig. 9 (right) shows the hysteretic curves of the four steel models con-
sidered. As expected, the bilinear model exhibits a significantly larger loop area.
Also, and contrarily to the concrete fiber responses, the maximum rebar strain
demands attained at the end of each imposed displacement cycle may diverge
considerably, in particular for larger imposed displacements. In fact, as the
element response enters the nonlinear range, the neutral axis moves toward
the compressed zone of the cross-section and consequently the strains in the
tensile zone become more sensitive to small variations in the cross-sectional
curvature.
 JOURNAL OF EARTHQUAKE ENGINEERING 19

Table 4. Sensitivity parameters considered in the study (reference properties in bold).

Sensitivity parameters Properties
Damping TSPD 0%
ISPD 0.5%
MPD 1%
1.5%
2%
5%
Element discretization scheme 10 IPs
End IPs ≈ 0.5 Lp
End IPs ≈ Lp
Strain penetration No strain penetration
Longer element
Base Spring
Material model Concrete Mander et al.
No tensile strength
Madas and Elnashai
Kappos and Konstantinidis
Steel Menegotto-Pinto (A2 = 0.15)
Menegotto-Pinto (A2 = 0.075)
Monti-Nuti
Bilinear
In bold: options corresponding to the reference model

Table 4 sums up the sensitivity parameters and associated properties considered in the present
study. The properties highlighted in bold were taken as reference (i.e. they were kept constant
while varying those associated to the sensitivity parameter under analysis). The EVD properties
producing the best results from the initial evaluation of Section 4.1 were selected as reference. All
the analyses were carried out with the structural analysis software SeismoStruct [Seismosoft,
2013].
Simple initial estimations of the shear span ratios for all the analysed structural
columns vary between 6 (Structure 1) and 7.5 (Structures 2 and 3), indicating that shear
deformations are negligible [Almeida et al., 2016]. Unsurprisingly, the shear capacity as
computed by Priestley et al. [2007] is approximately five (Structures 2 and 3) to seven
(Structure 1) times larger than the member shear in equilibrium with the bending
capacities at the member extremities (assuming a cantilever response for Structure 1
and double bending for columns in Structures 2 and 3). Therefore, the shear mechanisms
were neglected in the numerical models.

3.3. Goodness-Of-Fit Evaluation

The goodness-of-fit resulting from sensitivity analyses is expressed for EDPs that are judged
representative of the structural response, and it should be computed with methodologies that
compare experimental and numerical results in an efficient and unbiased way.

3.3.1. Response Engineering Demand Parameters

Lateral displacements and accelerations were chosen as the engineering demand parameters
(EDPs) to be assessed in this study since they are internationally accepted indicators of structural
response [Krawinkler, 2006] and are also part of typically available data from experimental
campaigns. Accelerations were only experimentally measured for Structure 1 and hence the
20 R. SOUSA ET AL.

goodness-of-fit evaluation of Structures 2 and 3 comprises only the comparisons with respect to
structural displacements.
Although local response EDPs such as concrete and rebar strains can better and more
directly relate to structural damage [Berry et al., 2008], their experimental monitoring during
the blind prediction tests was not carried out and thus they are not herein considered. It is
noted that most recent experimental campaigns place a very significant emphasis on the use of
advanced instrumentation allowing for the monitoring of local-level EDPs, followed by
corresponding efforts of numerical simulation [Goodnight et al., 2012; Almeida et al., 2017].

3.3.2. Response Error

In order to assess the accuracy of each model with respect to the experimental results, two
distinct relative error measures were used: Cumulative Error and Peak Error. The former
compares the specific EDP over the entire range of the dynamic response, considering
both phase and amplitude changes, and is herein applied for the first time in the form
described below. On the other hand, the latter error measure compares absolute maximum
values at discrete instants, which represents a more commonly used approach.

Cumulative error. The Cumulative Error evaluates the variations in magnitude of a specific
quantity and is defined as an “averaged” error (herein taken as the Root Mean Square Error,
RMSE) normalized by the Root Mean Square (RMSM) of the magnitudes of the measured
signal. This error measure can be computed in time (t) and frequency (f) domains:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2ffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P  2
1 1
P

RMSE n n xM;i ðtÞ xC;i ðtÞ n2 n XM;i ðf Þ XC;i ðf Þ
Cumulative Error ¼ ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ¼ q ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (6)
RMSM 1
P
2 1
P  2
n xM;i ðt Þ n XM;i ð f Þ

n n2

where xM,i (t) and xc,i (t) are the measured (subscript index M) and calculated (sub-
script index C) time series with n discrete time (t) samples, and XM,i (f) and Xc,i (f) are the
Fourier transforms of the corresponding time series, as a function of frequency (f). The
advantage of using Eq. (6) in the frequency domain is that it allows for the determination
of the error associated with the amplitude and phase independently. In this context, the
signals are separated into their real and imaginary components, where the amplitudes are
the lengths of the vectors in the Argand diagram and the phases are the angles made
between those vectors and the real axis at a specific frequency i (Fig. 10). The previous
equation can be rewritten as:
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1
P h
2
2 i
1
P  2 R R þ I I
n XM;i ðf Þ XC;i ðf Þ n2 n M;i C;i M;i C;i
n2
Cumulative Error ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi2 ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (7)
1 2 2
P  P 
1
X M;i ð f Þ n2 n RM;i þ IM;i
2

n n

where RMi, IMi and RCi, ICi are the real and the imaginary components of the measured
and calculated responses at frequency i, while n is the number of frequency samples of the
Fourier transform.
Noting that the frequency resolution (Δf) of the Fourier transform is given by 1/(n×Δt),
(Δt defines the sampling interval), the summation for all frequencies in the Fourier
transform can be replaced by a summation in the interval defined by two limiting
frequencies, f1 and f2, encompassing the most relevant frequency content of the analysed
signals. Thus, in order to account for higher-mode effects and period elongation under
 JOURNAL OF EARTHQUAKE ENGINEERING 21

Figure 10. Graphical representation of the real and imaginary components of the calculated and
measured quantities in the frequency domain.

nonlinear response, f1 and f2 were herein defined as a function of the fundamental period
of the structures: 1/(4T1) and 1/(0.1T1) respectively [Lepage et al., 2008]. Possible spurious
amplitudes at irrelevant frequencies (from the structural viewpoint) are therefore elimi-
nated from the computed error.
The amplitude component of the Cumulative Error can therefore be determined as:
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1
Pf 2 h
2
2 i
n2 i¼f 1 R M;i þ I M;i R c;i þ ICi
Cumulative Amplitude Error ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Pf 2  (8)
1 2 2
n2 i¼f 1 R M;i þ I M;i

while the Cumulative Phase Error may then be determined by subtracting the amplitude
component (Eq. 8) from the total Cumulative Error (Eq. 7).
It is underlined that the error measure determined with Eq. (7) produces values that vary
between 0 and infinity, with 0 indicating a perfect match while. If the two signals are 180° out-
of-phase, an error of 2 is obtained. Increasing values of the Cumulative Error reflect larger
differences between the signals in terms of amplitude and/or phase and hence a poorer match.
Alternative normalization methods, such as the Theil [1961] inequality coefficient, have been
used in other recent studies as goodness-of-fit measures [Babazadeh et al., 2015].

Peak error. The second error measure represents a more commonly used approach,
which determines the maximum relative error of the considered EDP—displacement
(disp.) or acceleration (acc.)—for each record separately. The main disadvantage of this
error measure is that it is sensitive to the magnitude of the experimental data, and in
particular when the latter tends to zero it is of little use. The error is computed for both
positive “+” and negative “-” values in each direction, as illustrated in Eq. (9).



þ max EDPcomputed;j max EDPmeasured;j
Peak Error;j ¼
 (9a)
max EDPmeasured;j



min EDPcomputed;j min EDPmeasured;j
Peak Error;j ¼
 (9b)
min EDPmeasured;j
22 R. SOUSA ET AL.

Figure 11. Outline of the parametric study.

In the previous equation, the index j indicates each individual record within the full set. It
should be noted that the time instant where the maximum (or minimum) of the measured
signal occurs does not necessarily coincide with the time instant corresponding to the
maximum (or minimum) of the computed response.
To avoid unrealistic experimental peaks due to high-frequency noise in the signal
acquisition phase, the experimental results were low-pass filtered. A cut-off frequency of
25 Hz was used for the results of Structure 1 [Terzic et al., 2015], while the ones from
Structures 2 and 3 used cut-off frequencies of 40 Hz for the accelerations and 20 Hz for
the displacements [Costa et al., 2012].

4. Applications
The present section presents the results of the sensitivity study carried out using nonlinear
dynamic analyses, an outline of which is depicted in Fig. 11.
In the plots of the following sub-sections the Peak Error computed for each individual
earthquake record is illustrated with filled and empty markers for the positive and negative
loading directions respectively. In addition, the black circles with the associated error bars
represent the mean value plus/minus the standard deviation computed for the records in
each set. Due to space limitations, and given that the plots follow essentially the same
format, the legend is presented only for the first figure. For the Cumulative Error the
distinction between negative and positive directions is not applicable and an empty
marker is hence used to identify each record. Regarding this latter error measure,
prevalence is given to the amplitude component (which will henceforth be simply referred
as Cumulative Error), as it appears more representative of the EDP variations. At the
right-hand side of each plot, the denominator of the associated error is presented. This
secondary plot, representing the experimental (exp.) component of the error, provides a
reference for the relative importance of the error values presented in the main plots.
Considering the impossibility to present the full set of results referring to the different
case studies, error types and EDPs, only the results that appear more relevant for each
sensitivity parameter were included in this manuscript.

4.1. Equivalent Viscous Damping

The following paragraphs present the results of the sensitivity analyses for the equivalent
viscous damping. Figs. 12–15 prove that the numerical models simulate more accurately the
experimental results for EVD models with low percentages of critical damping (roughly
between 0.5% and 2%, depending on the structure). For such range of values the damping
forces developed are small and therefore it is not straightforward to assess the differences
 JOURNAL OF EARTHQUAKE ENGINEERING 23

Figure 12. Peak displacement (left) and acceleration (right) error for Structure 1 considering different
EVD models.

Figure 13. Peak displacement error for Structures 2 (left) and 3 (right) considering different EVD
models.

Figure 14. Cumulative displacement (left) and acceleration (right) error for Structure 1 considering
different EVD models.

between MPD or SPD models by looking at these response parameters. Yet, it is clear that, as
the percentage of critical damping increases, the error associated with the ISPD grows with
24 R. SOUSA ET AL.

Figure 15. Cumulative displacement error for Structures 2 (left) and 3 (right) considering different EVD models.

Figure 16. Transverse (left) and rotational (right) history of accelerations measured in Structure 1
during Eqk 4 considering 0.5% MPD and 0.5% TSPD.

respect to the TSPD and MPD counterparts. On the other hand, the use of 0% damping
produces an increase of the error, noticeable even in comparison with the case of 0.5% of
critical damping, which is associated with an overestimation of the simulated displacements.
The right-side plot of Fig. 12 shows the relative error associated with the measured
accelerations of Structure 1. It is interesting to observe that the averaged maximum
accelerations are generally better estimated with MPD models independently of the
percentage of critical damping, although the corresponding variance in the results is
also larger. In order to clarify the reasons of such peculiar behavior, the left and right
plots in Fig. 16 depict the variation of the transverse and rotational accelerations at the top
of the pier considering both MPD and TSPD with 0.5% of critical damping.
The left plot shows a significant difference in the transverse accelerations computed
with both damping models around the time instant t = 200 s. After this point, both models
produce similar values, close to the experimental ones. This localized effect appears to be
connected with the way in which SPD and MPD models deal with the contribution of
higher-mode effects (second mode, in this case).
In the right side plot, the high-frequency rotational accelerations—response parameter
associated with the second mode of vibration—exhibit also larger differences at around
t = 200 s. In this case the values computed with MPD are about four times larger than the
ones obtained with the TSPD model (even considering only 0.5% of critical damping).
 JOURNAL OF EARTHQUAKE ENGINEERING 25

Figure 17. Fourier spectrum of transverse acceleration of Structure 1 during Eqk 4 considering 0.5%
MPD and 0.5% TSPD.

This is due to the fact that while damping forces increase for the higher modes using SPD
models, in the MPD case the higher modes tend to become more important as they are
always less damped than the fundamental mode of vibration. For completeness, Fig. 17
shows the Fourier spectrum of the lateral accelerations determined with both damping
models. The results reveal that both models produce similar Fourier amplitudes except
around 4-7 Hz in the model with 0.5% MPD.
For models with more degrees of freedom and more distributed mass, MPD can allow
for a disproportionate contribution of higher-order local modes that in principle are not
important for the overall structural response. In such cases, the consideration of SPD is
recommended in order to reduce the possible contamination of results by these spurious
local modes.
Considering the similarities obtained in terms of displacements given by the TSPD and
MPD for the same percentage of critical damping, it is difficult to select the most accurate
one. However, the accelerations computed with MPD are more prone to show amplified
response accelerations, and thus inertial and member forces, associated with higher
modes. On the other hand, as shown in Correia et al. [2013], SPD models tend to produce
more significant damping forces that strongly influence member and joint equilibrium. In
the following sections, only the damping models showing the best approximation to the
experimental results are considered. Therefore, critical damping values of 0.5% (for
Structures 1 and 2) and 1% (Structures 1 and 3) were considered for MPD and TSPD.

4.2. Element Discretization

In the following, the results of the parametric tests related to the discretization options indicated
in Table 3 are evaluated for the most accurate damping models identified in the previous section.
The moment-curvature curve of Structure 1 (not depicted) shows a fully hardening post-yield
behavior. Such observation implies that localization issues are not expected and thus the most
26 R. SOUSA ET AL.

Figure 18. Peak displacement (left) and acceleration (right) error for Structure 1 considering different
discretization schemes.

Figure 19. Cumulative displacement (left) and acceleration (right) error for Structure 1 considering
different discretization schemes.

refined approach (corresponding to the use of 10 IPs) should minimize the error, see Sections
2.2.1 and 2.2.2. However, Figs. 18 and 19 show a large variability for both the displacement and
acceleration errors computed for this bridge pier, making it difficult to identify one discretiza-
tion scheme that clearly stands out with respect to another.
Regarding Structures 2 and 3, the columns’ moment-curvature curves exhibit a very
slight softening branch after the peak, which is attained at a curvature of around 0.1 m−1.
However, as depicted in Fig. 20, this value is barely reached during the nonlinear dynamic
time history. Once again, the most refined approach should thus minimize the error.
Nevertheless, similarly to Structure 1, Figs. 21 and 22 do not confirm the previous
expectation and instead it appears that the use of a discretization scheme such that the
integration weight of the extreme IP is approximately similar to the plastic hinge length
(i.e. when 4 IPs are employed) produce improved simulations.
The reason behind this discrepancy between the theoretical considerations of Section 2.2.1
and the experimental results are related with the limitations of Euler-Bernoulli theory. In fact,
recent experimental findings [Goodnight et al., 2014] indicate that due to tension-shift effects,
plastic curvatures follow an approximately linear distribution in the plastic hinge region. This is
unlike what an exact solution of the Euler-Bernoulli theory predicts. The use of a reduced
number of integration points in a FB element, for which the extreme IP weight accounts for this
 JOURNAL OF EARTHQUAKE ENGINEERING 27

Figure 20. History of curvatures measured at the column base below Point B of Structure 2 considering
0.5% TSPD and 4 IPs per column.

Figure 21. Peak displacement error for Structures 2 (left) and 3 (Right), considering different discretiza-
tion schemes.

Figure 22. Cumulative displacement error for Structures 2 (left) and 3 (Right), considering different
discretization schemes.

length along which the plastic curvatures are linear, hence seems to serve as a workaround that
allows for a better match against experimental results. In other words, the need to regularize the
response in the pre-peak branch and not just in the post-peak phase of behavior appears to be
required. Recently proposed methods try to incorporate such observations in the context of
Euler-Bernoulli FB approaches [Feng et al., 2014]. A newly developed axially-equilibrated
displacement-based beam element combines the strict verification of equilibrium, which was
28 R. SOUSA ET AL.

Figure 23. Predicted curvatures at base IP for Structure 1 during Eqk 1 and Eqk 3.

exclusive to FB formulations, with the abovementioned linear distribution of plastic curvatures

[Tarquini et al., 2017].
Related to the above, and despite the absence of base curvature measurements during
the experimental tests herein analysed, the current case studies confirm that the numerical
estimation of this local-level EDP is very sensitive to the adopted discretization. Figure 23
shows the variation of the base curvatures in Structure 1 during Eqk 1 (on the left) and
Eqk 3 (on the right). As expected, while the pier responded essentially in the linear range
during Eqk 1 (pushover analyses on this structure indicate that first yielding occurs for
section curvatures slightly above 0.004 m−1), the base curvatures simulated by two
different models (with 6 and 10 IPs) are very similar; however, during large nonlinear
excursions (Eqk 3), the base curvatures predicted with 10 IPs reach values that are more
than two times larger than those obtained with 6IPs.
Such observations show that local-level measures are much more significantly affected
by different discretization schemes than quantities related to the global-level response, as
observed elsewhere for monotonic analyses [Almeida et al., 2016]. Considering the
magnitude of the observed differences, the results emphasize the need to implement
regularization techniques, applicable not only to the post-peak phase but also to the
pre-peak branch as discussed above, in order to improve the simulation of the member
local response.
As a final observation, it is noted that the Cumulative Error is sensitive to small variations in
the amplitude of the signals under comparison. For instance, the history of measured and
computed displacements for Structure 1 assuming MPD with 0.5% of critical damping, 6IPs
and the reference properties for the remaining parameters, for which the calculated error is
approximately 30% for Eqk 3, is presented in Fig. 24. Despite the very good estimation of the
peak displacements, the relatively large Cumulative Error is due to the mismatch in the small
amplitude part of the response following those peaks, which lasts until the end of the signal.

4.3. Strain Penetration

The results illustrated in Figs. 25–27 indicate that, in general, the consideration of a
rotational spring or an artificially elongated element to model strain penetration effects
improve the estimations with respect to the case where the additional SP flexibility is
 JOURNAL OF EARTHQUAKE ENGINEERING 29

Figure 24. Comparison between measured and simulated displacement time-histories in Structure 1,
considering 0.5% MPD and 6IPs, during Eqk 3.

Figure 25. Peak (left) and cumulative (right) displacement error for Structure 1, considering different SP
models.

Figure 26. Peak (left) and cumulative (right) displacement error for Structure 2, considering different SP
models.

neglected. Furthermore, the use of elongated elements appears to provide more accurate
EDPs than the use of a rotational spring at the base of the columns.
30 R. SOUSA ET AL.

Figure 27. Peak (left) and cumulative (right) displacement error for Structure 3, considering different SP
models.

Table 5. Elastic vibration periods of structure 2 for different strain penetration models.
Experimental Numerical
Specimen +
Modes Specimen shake table Strain penetration neglected Elongated element Base spring
T1_X (s) 0.277 0.333 0.311 0.326 0.365
T2_Y (s) 0.256 0.303 0.285 0.300 0.337
T3_RZ (s) 0.129 0.154 0.146 0.154 0.170

It is noted that the simulation improvements observed for larger seismic intensity
demands bring about significant differences in the structural response at lower intensity
excitations. In fact, each simplified SP modeling option affects the elastic dynamic
structural properties through an (unrealistic) increase in member flexibility; it is recalled
that the elongation of the members by the strain penetration length or the consideration
of an elastic spring are calculated from the assumption that the longitudinal rebars have
yielded. This point is illustrated in Table 5, which summarizes the elastic periods of
vibration of Structure 2 with the different SP modeling options.
It is clear that the models of Structure 2 with elongated elements have a period of vibration
that compare very satisfactorily with the experimental values considering the contribution of
both the specimen and the shake table. The increased flexibility of this modeling option turns
out to reproduce more accurately the flexibility of the coupled system (specimen + shake
table) than the use of a fixed-base model with no consideration for SP effects.
To further illustrate the abovementioned differences, Fig. 28 compares the history of
simulated and experimental displacements in Structure 2 at point A (see Fig. 6) along
direction X, for Eqk 1 and Eqk 3. Particular attention is given to the elastic response
during the first record, between t = 5 s and t = 10 s.
A visual assessment of this series of plots reinforces the conclusion that the use of
longer vertical elements produces better estimations when compared with the use of a
rotational spring. Moreover, as it only requires the calculation of a strain penetration
length estimate, it is simpler to implement than an equivalent spring stiffness and further
it seems to more adequately reproduce the elastic modal properties (see Table 5).
On the other hand, for frame structures with multiple stories and/or bays, geometrical
incompatibilities will arise due to likely different strain penetration lengths for parallel
 JOURNAL OF EARTHQUAKE ENGINEERING 31

Figure 28. Time-history of simulated displacements in Structure 2 (point A, direction X) vs displace-

ments computed with different SP models, during Eqk 1 and Eqk 3.

members and lastly the shear demands will be slightly underestimated given the increased
shear span of the elements.
The results obtained with both simplified SP models also exposed important limita-
tions. In view of the previous observation and the restraints associated with the employ-
ment of more advanced alternatives (see Section 2.3), a new link element for fiber-based
sections that explicitly accounts for the bond-slip response along the embedment length of
the rebars was proposed by Sousa [2015]. This model restrictions an encouraging accuracy
when compared with experimental results.

4.4. Material Stress-Strain Models

4.4.1. Concrete
The concrete models considered in this study differ primarily with respect to the approaches
used to simulate the confinement effects on the cross-section core: in particular, the definition
of the post-peak response may differ significantly as illustrated in Figs. 29–32, which show that
for the analysed cases only marginal deviations in the global structural response were
observed. This can be partially attributed to the relatively small concrete compressive strain
demands, as discussed below in more detail. The model proposed by Kappos and
Konstantinidis [1999], which was calibrated for concrete strengths larger than 50 MPa
(hence larger than those of the analysed structures), offered results in terms of displacement
and acceleration predictions which are in line with those of the reference model [Mander
et al., 1988]. On the other hand, despite the encouraging results for Structure 1, the model
32 R. SOUSA ET AL.

Figure 29. Peak displacement error (left) and acceleration (right) error for Structure 1 considering
different concrete models.

Figure 30. Peak displacement error for Structure 2 (left) and 3 (right) considering different concrete
models.

Figure 31. Cumulative displacement error (left) and acceleration (right) error for Structure 1 considering
different concrete models.

proposed by Madas and Elnashai [1992] showed numerical instability issues for Structures 2
and 3 which prevented the completion of the analyses for the full set of records.
In addition, while Figs. 29 and 31 show that neglecting the concrete tensile strength for
Structure 1 does not seem to adversely affect the match, for Structures 2 and 3 this parameter
 JOURNAL OF EARTHQUAKE ENGINEERING 33

Figure 32. Cumulative displacement error for Structure 2 (left) and 3 (right) considering different
concrete models.

Figure 33. Time-history of simulated displacements in Structure 3 (point A, direction x) considering

Mander’s concrete model with and without tensile strength.

plays a role in the response. Considering the low intensity of Eqk 1, both structures 2 and 3
respond essentially in the elastic regime. Hence, it is not surprising that the displacements
obtained neglecting the tensile response are overestimated since the stiffness of the structure is
significantly underestimated (Figs 30 and 32). For larger intensities, however, the numerically
simulated displacements are essentially similar with or without consideration of the concrete
tensile strength. This effect can be appreciated in more detail in Fig. 33, which depicts the
history of displacements of Structure 3 under elastic (left) and nonlinear response (right).
As mentioned above, the structural response was not significantly affected by the
alternative constitutive relations considered, notwithstanding the significant number of
nonlinear cycles imposed by successive records. It could be expected a priori that the
differences in the pre-peak and the highly distinct post-peak branches would influence the
results. However, that did not turn up to be the case, which can be essentially attributed to
the concrete compressive strains seldom exceeding the strain at peak stress, as illustrated
in Fig. 34. It shows the strain history at diametrically opposed fibers in the base cross-
section of Structure 1, corresponding to the most strained fibers of the confined concrete
(the dashed line indicates the strain at peak compressive stress).
It is apparent that when concrete is subjected to large compressive strain demands under
flexure, the opposite side of the cross-section experiences much larger (absolute) tensile
strains producing large equilibrating forces in the reinforcement. Hence, during cyclic
loading, the outmost rebars tend to elongate alternatively on both sides of the cross-section
34 R. SOUSA ET AL.

Figure 34. History of simulated strains in Structure 1 at diametrically opposed fibers of the base section
compressed core, for the entire set of records and considering alternative concrete models.

Figure 35. Simulated stress-strain response in Structure 1 at diametrically opposed fibers of the base
section, for the entire set of records and considering alternative concrete models.

and to accumulate permanent residual tensile strains. This well-known phenomenon pro-
gressively limits the occurrence of large compressive strains in the concrete to shorter
periods of very strong seismic demand. Moreover, based on the previous figure and the
stress-strain response presented in Fig. 35 it is possible to verify that the maximum strains at
the base of the element are not sensitive to the adopted concrete constitutive relation.

4.4.2. Reinforcing Steel

The present section examines the effect of four different reinforcing steel constitutive
models: the well-known Menegotto-Pinto [Menegotto and Pinto, 1973] (using two differ-
ent transition curves), the model proposed by Monti-Nuti [Monti and Nuti, 1992] to
account for buckling of the longitudinal rebars, and the simple bilinear approximation.
 JOURNAL OF EARTHQUAKE ENGINEERING 35

Figure 36. Peak displacement (left) and acceleration (right) error for Structure 1 considering different
steel models.

Figure 37. Peak displacement error for Structure 2 (left) and 3 (right) considering different steel
models.

Figure 38. Cumulative displacement (left) and acceleration (right) error for Structure 1 considering
different steel models.

Figs. 36–39 confirm that, as expected and noted by others [Blandon, 2012], the computed
error measures are significantly more sensitive to the adopted constitutive model for steel
than for concrete.
Regarding the simple bilinear model, one can confirm the expected increase in the
computed mean values errors and standard deviation. Bearing in mind that all steel
36 R. SOUSA ET AL.

Figure 39. Cumulative displacement error for Structure 2 (left) and 3 (right) considering different steel
models.

Figure 40. Stress-strain response at diametrically opposed fibers of the base section (top) and history of
simulated curvatures (bottom) in Structure 1 for the entire set of records considering alternative steel
models.

models assume the same loading and unloading stiffness, it becomes evident that the
hysteresis loop shape, i.e. unloading and reloading curves, impact significantly the overall
response simulation. The results presented in the top plots of Fig. 40 demonstrate that the
strain demands when the bilinear model is employed may differ significantly from both
approaches featuring the Menegotto-Pinto model. Its abrupt pre-to-post yield stiffness
change proved to be numerically more challenging for the analyses than the Menegotto-
Pinto model. Moreover, the history of curvatures presented in the bottom part of the same
figure shows important deviations during Eqks 3, 4 and 6.
 JOURNAL OF EARTHQUAKE ENGINEERING 37

The previous observations emphasize the importance of the Bauschinger’s effect in

obtaining an accurate representation of the reinforcement behavior and therefore a better
estimation of the structural response. Based on the overall set of computed errors and the
stability of the analyses, one may conclude that the Menegotto-Pinto’s steel model exhibits
the better performance. The use of different transition curve parameters (A2 = 0.15 versus
A2 = 0.075), on the other hand, is seen to have a limited influence.

5. Conclusions
Motivated by the successful participation in blind prediction challenges, the authors decided
to undertake a critical review on fiber-based Euler-Bernoulli distributed plasticity beam
models that are commonly used to simulate the nonlinear dynamic response of RC frame
structures. Based on a review of past studies, four modeling options were selected to be
analysed in detail through a sensitivity analysis: equivalent viscous damping, element
discretization, simulation of strain penetration effects, and material constitutive models
(concrete and steel). The three structures tested in the aforementioned competitions were
used as case studies. The goodness-of-fit was evaluated with: (i) a newly developed proposal
for a Cumulative Error measure, and (ii) the more traditional Peak Error measure. They
were applied to the engineering demand parameters that were measured experimentally,
namely displacements and accelerations. Together with a qualitative comparison of the
time-history responses, the following main conclusions were obtained from this study:

● Most of the energy-dissipation mechanisms are explicitly modeled and thus equiva-
lent damping models should be carefully employed. Based on the analyses carried
out, enhanced performance was obtained with values of critical damping ranging
from 0.5% to 2%. Higher levels of EVD tend to underestimate the structural
response, whilst considering no damping results in a small overestimation of the
response parameters. The study also addressed the specific limitations of both mass-
proportional damping and tangent stiffness-proportional damping, highlighting the
difficulty of advocating one model with respect to the other. However, an important
conclusion is that for the recommended small values of critical damping such
differences become limited.
● The definition of a number of integration points per force-based element such that
the weighted length associated with the extremity IP matches the expected plastic
hinge length, produced somewhat more accurate results for the case studies. This was
not an anticipated result from theoretical considerations since for all the analysed
structures no localization took place and hence the most refined model should have
produced the most accurate results. This discrepancy can be attributed to the
presence of tension-shift effects, as demonstrated by other recent experimental
findings, which cannot be simulated by Euler-Bernoulli beam theory. The need to
develop, calibrate and apply appropriate regularization techniques for fully hardening
post-yield moment-curvature relations, where the response is objective, is hence
confirmed from the present study. This line of research is being pursued in recent
studies [Tarquini et al., 2017].
● Two simple modeling approaches were considered to account for strain penetration
effects, and both contributed to reduce the estimated response errors, particularly for
38 R. SOUSA ET AL.

larger inelastic demands. The elongation of the member by a strain penetration

length tends to produce better results than the use of a rotational spring.
Nonetheless, it should be acknowledged that both approaches have drawbacks,
namely: (i) the modification of the elastic dynamic properties of the initially unda-
maged structure, which can be particularly relevant when a base spring is considered;
(ii) the slight underestimation of the member shear demand associated to the use of
an artificially elongated element. The development of more accurate, while compu-
tationally light, strain penetration models should deserve attention from researchers.
● Amongst the limited sample of concrete models analysed, those proposed by Mander
et al. [1988] and Kappos and Konstatinidis [1999] provided similar estimations of the
global-level structural response, despite the differences in the stress-strain constitu-
tive relations at both pre- and post-peak branch. Given the accumulation of residual
tensile strains in the rebars under cyclic loads of large ductility demand, the concrete
fibers only seldom reach large compressive strains where the differences between the
constitutive models become more pronounced. Moreover, it was shown that neglect-
ing the tensile strength of the concrete does not impact the structural response
during its nonlinear phase. However, if one is interested in the linear phase of
behavior, the consideration of the tensile strength is advised despite the increased
potential for numerical convergence issues.
● Regarding the steel uniaxial relations, the present study indicates that the well-known
Menegotto and Pinto [1973] model combines both accuracy and stability.
Furthermore, it was verified that the definition of different transition curves in the
unloading and reloading branches, through parameters A2 = 0.15 and A2 = 0.075,
produced negligible differences in global response error measures. On the other
hand, the more theoretically advanced model proposed by Monti and Nuti [1992]
did not translate in improved results, essentially due to overshooting problems. As
for the bilinear elastoplastic model, disregarding the Bauschinger effect appears to
influence substantially the global structural response as the errors computed with this
model are consistently larger than the ones obtained with both variants of the
Menegotto-Pinto steel model.
● As a closing recommendation for future studies aiming at resolving the many yet
unclear simulation issues, the conclusions above point out to the need of: (i)
extending the sensitivity analyses to include, in the comparison against numerical
results, experimental measurements of local-level quantities such as strains, curva-
tures, or fixed-end rotations; (ii) considering a much-extended sample of case studies;
(iii) accounting for a larger range of modeling options.

References
Addessi, D. and Ciampi, V. [2007] “A regularized force-based beam element with a damage–plastic
section constitutive law”, International Journal for Numerical Methods in Engineering 70(5), 610–
629. doi:10.1002/(ISSN)1097-0207
Akin, J. E. [2000] “Finite Elements for Analysis and Design,” in Computational Mathematics and
Applications, eds J. Whitman, and J. Davenport, (Academic Press, London, United Kingdom).
 JOURNAL OF EARTHQUAKE ENGINEERING 39

Almeida, J. P., Das, S. and Pinho, R. [2012] “Adaptive force-based frame element for regularized
softening response”, Computers & Structures 102-103, 1–13. doi:10.1016/j.compstruc.2012.03.018
Almeida, J. P., Prodan, O., Rosso, A. and Beyer, K. [2017] “Tests on thin reinforced concrete walls
subjected to in-plane and out-of-plane cyclic loading”, Earthquake Spectra 33(1), 323–345.
doi:10.1193/101915EQS154DP
Almeida, J. P., Tarquini, D. and Beyer, K. [2016] “Modelling approaches for inelastic behaviour of
RC walls: multi-level assessment and dependability of results”, Archives of Computational
Methods in Engineering 23, 69–100. doi:10.1007/s11831-014-9131-y
Arakawa, T. and Yamamoto, K. [2004] “Frequencies and damping ratios of a high rise building
based on microtremor measurement,” Proc. of the 13th World Conference on Earthquake
Engineering, Vancouver, B.C., Canada.
Babazadeh, A., Burgueno, R. and Silva, P. F. [2015] “Use of 3D finite-element models for predicting
intermediate damage limit states in RC bridge columns”, Journal of Structural Engineering 141,
10. doi:10.1061/(ASCE)ST.1943-541X.0001253
Bae, S. and Bayrak, O. [2008] “Seismic performance of full-scale reinforced concrete columns”, ACI
Structural Journal 105(2), 123–133.
Bathe, K.-J. [1996] Finite Element Procedures, Prentice Hall, New Jersey, USA.
Bentz, A. and Kijewski-Correa, T. [2008] “Predictive models for damping in buildings: the role of
structural system characteristics,” Proc. of the 18th Analysis and Computation Specialty
Conference Vancouver, Canada.
Bernal, D. [1994] “Viscous damping in inelastic structural response”, Journal of Structural
Engineering 120(4), 1240–1254. doi:10.1061/(ASCE)0733-9445(1994)120:4(1240)
Berry, M., Lehman, D. and Lowes, L. [2008] “Lumped-plasticity models for performance simulation
of bridge columns”, ACI Structural Journal 105(3), 270–279.
Blandon, C. [2012] “Sensibility of non linear time history analysis of a RC column using large scale
shaking table results as benchmark,” Proc. of the 15th World Conference on Earthquake
Engineering, Lisbon, Portugal.
Calabrese, A., Almeida, J. P. and Pinho, R. [2010] “Numerical issues in distributed inelasticity
modeling of rc frame elements for seismic analysis”, Journal of Earthquake Engineering 14(1), 38–
68. doi:10.1080/13632461003651869
Campos Costa, A., Morais, P., Wainwright, B. and Martins, A. [1996] Characterization of the New
LNEC Shaking Table, Laboratório Nacional de Engenharia Civil, Lisbon, Portugal.
Carrea, F. [2010] “Shake-Table Test on a Full-Scale Bridge Reinforced Concrete Column,” MSc
Thesis, University of Bologna, Bologna, Italy.
CEN. [2004] EN 1998-1. Eurocode 8: Design of Structures for Earthquake Resistance - Part 1: General
Rules, Seismic Actions and Rules for Buildings, Comité Européen de Normalisation, Brussels.
CEN. [2005] EN 1998-3. Eurocode 8: Design of Structures for Earthquake Resistance - Part 3:
Assessment and Retrofitting of Buildings, Comité Européen de Normalisation, Brussels.
Charney, F. A. [2008] “Unintended consequences of modeling damping in structures”, Journal of
Structural Engineering 134(4), 581–592. doi:10.1061/(ASCE)0733-9445(2008)134:4(581)
Chopra, A. K. [1995] Dynamic of Structures, Prentice Hall, Englewood Cliffs, New Jersey, USA.
Chopra, A. K. and Mckenna, F. [2017] “Modeling viscous damping in nonlinear response history
analysis of buildings,” 16th World Conference on Earthquake Engineering, Santiago, Chile.
Chopra, A. K. and McKenna, F. [2016] “Modeling viscous damping in nonlinear response history
analysis of buildings for earthquake excitation”, Earthquake Engineering and Structural Dynamics
45(2), 193–211. doi:10.1002/eqe.2622
Coleman, J. and Spacone, E. [2001] “Localization issues in force-based frame elements”, Journal of
Structural Engineering 127(11), 1257–1265. doi:10.1061/(ASCE)0733-9445(2001)127:11(1257)
Correia, A. A. [2011] “A pile-head macro-element approach to seismic design of monoshaft-
supported bridges,” Ph.D. thesis, Istituto Universitario di Studi Superiori, University of Pavia,
Pavia, Italy.
Correia, A. A., Almeida, J. P. and Pinho, R. [2013] “Seismic energy dissipation in inelastic frames:
understanding state-of-the-practice damping models”, Structural Engineering International 23(2),
148–158. doi:10.2749/101686613X13439149157001
40 R. SOUSA ET AL.

Costa, A., Campos Costa, A., Candeias, P., Guerreiro, L. and Mendes, L. [2012] “15WCEE blind test
challenge - design report,” Proc. of the 15th World Conference on Earthquake Engineering,
Lisbon, Portugal.
Fardis, M. [2009] Seismic Design, Assessment and Retrofitting of Concrete Buildings, Springer
Netherlands.
Feng, D., Ren, X. and Li, J. [2016] “Implicit gradient delocalization method for force-based frame
element”, Journal of Structural Engineering 142, 2. doi:10.1061/(ASCE)ST.1943-541X.0001397
Feng, Y., Kowalsky, M. J. and Nau, J. M. [2014] “Fiber-Based modeling of circular reinforced concrete
bridge columns”, Journal of Earthquake Engineering 18(5), 714–734. doi:10.1080/13632469.2014.904254
Fragiadakis, M., Pinho, R. and Antoniou, S. [2008] “Modelling inelastic buckling of reinforcing bars
under earthquake loading”, Computational Structural Dynamics and Earthquake Engineering:
Structures and Infrastructures 2, 363–377.
Girard, C. and Bastien, J. [2002] “Finite-Element bond-slip model for concrete columns under cyclic
loads”, Journal of Structural Engineering 128(12), 1502–1510. doi:10.1061/(ASCE)0733-9445
(2002)128:12(1502)
Goodnight, J. C., Feng, Y., Kowalsky, M. J. and Nau, J. M. [2015] “The effects of load history and
design variables on performance limit states of circular bridge columns – volume 2: experimental
observations,” Technical Report, Alaska Department of Transportation and Public Facilities, USA.
Goodnight, J. C., Kowalsky, M. J. and Nau, J. M. [2012] “Experimental observations on the effect of
load history on performance limit states of circular bridge columns,” Proc. of the 15th World
Conference on Earthquake Engineering, Lisbon, Portugal.
Goodnight, J. C., Kowalsky, M. J. and Nau, J. M. [2014] “A new look at strain limits and plastic
hinge lengths for reinforced concrete bridge columns,” Proc. of the 10th U.S. National
Conference on Earthquake Engineering, Anchorage, USA.
Goodnight, J. C., Kowalsky, M. J. and Nau, J. M. [2016] “Modified plastic-hinge method for circular
RC bridge columns”, Journal of Structural Engineering 143, 11.
Grant, D., Blandon, C. and Priestley, M. J. N. [2005] Modelling Inelastic Response in Direct
Displacement-Based Design, IUSS Press, Pavia, Italy.
Hall, J. [2006] “Problems encountered from the use (or misuse) of Rayleigh damping”, Earthquake
Engineering & Structural Dynamics 35(5), 525–545. doi:10.1002/eqe.541
Hines, E., Restrepo, J. I. and Seible, F. [2004] “Force-displacement characterization of well-confined
bridge piers”, ACI Structural Journal 101(4), 537–548.
Jacobsen, L. S. [1930] “Steady forced vibrations as influenced by damping”, ASME Transactions 52
(1), 169–181.
Jeary, A. [1986] “Damping in tall buildings - a mechanism and a predictor”, Earthquake Engineering
& Structural Dynamics 14(5), 733–750. doi:10.1002/eqe.4290140505
Jehel, P., Léger, P. and Ibrahimbegovic, A. [2013] “Initial versus tangent stiffness-based Rayleigh
damping in inelastic time history seismic analyses”, Earthquake Engineering & Structural
Dynamics 43(3), 467–484. doi:10.1002/eqe.2357
Jendele, L. and Cervenka, J. [2006] “Finite element modelling of reinforcement with bond”,
Computers & Structures 84(28), 1780–1791. doi:10.1016/j.compstruc.2006.04.010
Jirásek, M. [1997] “Analytical and numerical solutions for frames with softening hinges”, Journal of
Engineering Mechanics 123(1), 8–14. doi:10.1061/(ASCE)0733-9399(1997)123:1(8)
JPEE [2014] “Prémio Ricardo Teixeira Duarte,” Proc. of the 5as Jornadas Portuguesas de Engenharia
de Estruturas, LNEC, Lisboa, Portugal (in Portuguese). Available at the website: http://jpee2014.
lnec.pt/concurs_2.html
Kappos, A. J. and Konstantinidis, D. [1999] “Statistical analysis of confined high strength concrete”,
Materials and Structures 32(10), 734–748.
Kareem, A. and Gurley, K. [1996] “Damping in structures: its evaluation and treatment of
uncertainty”, Journal of Wind Engineering and Industrial Aerodynamics 59, 131–157.
doi:10.1016/0167-6105(96)00004-9
Krawinkler, H. [2006] “Importance of good nonlinear analysis”, The Structural Design of Tall and
Special Buildings 15(5), 515–531. doi:10.1002/(ISSN)1541-7808
 JOURNAL OF EARTHQUAKE ENGINEERING 41

Lagomarsino, S. [1993] “Forecast models for damping and vibration periods of buildings”, Journal of
Wind Engineering and Industrial Aerodynamics 48, 221–239. doi:10.1016/0167-6105(93)90138-E
Lanzi, A. and Luco, J. E. [2017] “Influence of viscous damping models on inelastic seismic response
of fixed and base-isolated structures,” Proc. of the 16th World Conference on Earthquake
Engineering, Santiago, Chile.
Lee, C.-L. and Filippou, F. [2015a] “Frame element with mixed formulations for composite and RC
members with bond slip. II: correlation studies”, Journal of Structural Engineering 141, 11.
Lee, C.-L. and Filippou, F. [2015b] “Frame element with mixed formulations for composite and RC
members with bond slip. I: theory and fixed-end rotation”, Journal of Structural Engineering 141, 11.
Lepage, A., Delgado, S. and Dragovich, J. [2008] “Appropriate models for practical nonlinear
dynamic analysis of reinforced concrete frames,”Proc. of the 14th World Conference on
Earthquake Engineering, Beijing, China.
Madas, P. and Elnashai, A. [1992] “A new passive confinement model for the analysis of concrete
structures subjected to cyclic and transient dynamic loading”, Earthquake Engineering &
Structural Dynamics 21(5), 409–431. doi:10.1002/eqe.4290210503
Mander, J., Priestley, N. and Park, R. [1988] “Theoretical stress-strain model for confined concrete”,
Journal of Structural Engineering 114(8), 1804–1826. doi:10.1061/(ASCE)0733-9445(1988)114:8
(1804)
Martinez-Rueda, J. and Elnashai, A. [1997] “Confined concrete model under cyclic load”, Materials
and Structures 30(3), 139–147. doi:10.1007/BF02486385
Menegotto, M. and Pinto, P. [1973] “Method of analysis for cyclically loaded R.C. plane frames
including changes in geometry and non-elastic behavior of elements under combined normal
force and bending,” Proc. of the Symposium Resistance and Ultimate Deformability of Structures
Acted on by Well-Defined Repeated Loads, Lisbon, Portugal.
Monti, G. and Nuti, C. [1992] “Nonlinear cyclic behavior of reinforcing bars including buckling”, Journal
of Structural Engineering 118(12), 3268–3284. doi:10.1061/(ASCE)0733-9445(1992)118:12(3268)
Monti, G. and Spacone, E. [2000] “Reinforced concrete fiber beam element with bond-slip”, Journal
of Structural Engineering 126(6), 654–661. doi:10.1061/(ASCE)0733-9445(2000)126:6(654)
Morita, K. and Kanda, J. [1996] “Experimental evaluation of the amplitude dependent natural
period and damping ratio of a multi-story structure,” Proc. of the 11th World Conference on
Earthquake Engineering, Acapulco, Mexico.
Mortezaei, A. and Ronagh, H. R. [2011] “Plastic hinge length of reinforced concrete columns
subjected to both far-fault and near-fault ground motions having forward directivity”, The
Structural Design of Tall and Special Buildings 22(12), 903–926. doi:10.1002/tal.729
Nagashima, T., Sugano, S., Kimura, H. and Ichikawa, A. [1992] “Monotonic axial compression test
on ultra high strength concrete tied columns,” Proc. of the 10th World Conference on
Earthquake Engineering, Madrid, Spain.
Neuenhofer, A. and Filippou, F. [1997] “Evaluation of nonlinear frame finite-element models”,
Journal of Structural Engineering 123(7), 958–966. doi:10.1061/(ASCE)0733-9445(1997)123:7(958)
Pam, H. J. and Ho, J. C. M. [2009] “Length of critical region for confinement steel in limited
ductility high-strength reinforced concrete columns”, Engineering Structures 31(12), 2896–2908.
doi:10.1016/j.engstruct.2009.07.015
PEER/ATC. [2010] PEER/ATC 72-1 - Modeling and Acceptance Criteria for Seismic Design and Analysis of
Tall Buildings, Pacific Earthquake Engineering Research Center/Applied Technology Council, USA.
Priestley, M. J. N., Calvi, G. M. and Kowalsky, M. J. [2007] Displacement-Based Seismic Design of
Structures, IUSS Press, Pavia, Italy.
Priestley, M. J. N. and Grant, D. [2005] “Viscous damping in seismic design and analysis”, Journal
of Earthquake Engineering 9(2), 229–255. doi:10.1142/S1363246905002365
Priestley, M. J. N., Seible, F. and Calvi, G. M. [1996] Seismic Design and Retrofit of Bridges, John
Wiley & Sons Inc, New Jersey, USA.
Puthanpurayil, A. M., Lavan, O., Carr, A. J. and Dhakal, R. P. [2016] “Elemental damping
formulation: an alternative modelling of inherent damping in nonlinear dynamic analysis”,
Bulletin of Earthquake Engineering 14(8), 2405–2434. doi:10.1007/s10518-016-9904-9
42 R. SOUSA ET AL.

Salem, H. and Maekawa, K. [2004] “Pre- and postyield finite element method simulation of bond of
ribbed reinforcing bars”, Journal of Structural Engineering 130(4), 671–680. doi:10.1061/(ASCE)
0733-9445(2004)130:4(671)
Satake, N., Suda, K.-I., Arakawa, T., Sasaki, A. and Tamura, Y. [2003] “Damping evaluation using
full-scale data of buildings in Japan”, Journal of Structural Engineering 129(4), 470–477.
doi:10.1061/(ASCE)0733-9445(2003)129:4(470)
Schoettler, M. J., Restrepo, J. I., Guerrini, G., Duck, D. E. and Carrea, F. [2015] A Full-Scale, Single-
Column Bridge Bent Tested by Shake-Table Excitation, Pacific Earthquake Engineering Research
Center, University of California, Berkeley, USA.
Scott, M. H. and Fenves, G. [2006] “Plastic hinge integration methods for force-based beam-column
elements”, Journal of Structural Engineering 132(2), 244–252. doi:10.1061/(ASCE)0733-9445
(2006)132:2(244)
Scott, M. H. and HamutçuoğLu, O. M. [2008] “Numerically consistent regularization of force-based
frame elements”, International Journal for Numerical Methods in Engineering 76(10), 1612–1631.
doi:10.1002/nme.v76:10
Seismosoft [2013] “SeismoStruct v.6.5. - a computer program for static and dynamic nonlinear
analysis of framed structures,” available online from http://www.seismosoft.com
Sezen, H. and Moehle, J. [2004] “Strength and deformation capacity of reinforced concrete columns with
limited ductility,” Proc. of the 13th World Conference on Earthquake Engineering, Vancouver, Canada.
Sheikh, S. A. and Uzumeri, S. M. [1982] “Analytical model for concrete confined in tied columns”,
Journal of the Structural Division 108(12), 2703–2722.
Sousa, R. [2015] “Development and Verification of Innovative Modelling Approaches for the
Analysis of Framed Structures Subjected to Earthquake Action,” Ph.D. thesis, UME School,
IUSS Pavia, Pavia, Italy.
Tarquini, D., Almeida, J. P. and Beyer, K. [2017] “Axially Equilibrated Displacement-Based Beam
Element for Simulating the Cyclic Inelastic Behaviour of RC Members”, 46(9), 1471–1492.
Terzic, V., Schoettler, M. J., Restrepo, J. I. and Mahin, S. [2015] “Concrete column blind prediction
contest 2010: outcomes and observations,” PEER Report No. 2015/01, Pacific Earthquake
Engineering Research Center, University of California, Berkeley, USA.
Theil, H. [1961] “Economic Forecasts and Policy,” in Contributions to Economic Analysis, vol. 15,
2nd Ed., rev. ed.,(North-Holland Publishing Company, Amsterdam, Netherlands).
Varum, H. [2003] “Seismic assessment, strengthening and repair of existing buildings,” Ph.D. thesis,
University of Aveiro, Aveiro, Portugal.
Vásquez, J. A., Llera, J. C. and Hube, M. A. [2016] “A regularized fiber element model for reinforced
concrete shear walls”, Earthquake Engineering & Structural Dynamics 45(13), 2063–2083.
doi:10.1002/eqe.2731
Wakabayashi, M. [1986] Design of Earthquake-Resistant Buildings, McGraw-Hill, New York, USA.
Yazgan, U. and Dazio, A. [2011a] “Simulating maximum and residual displacements of RC
structures: I. Accuracy”, Earthquake Spectra 27(4), 1187–1202. doi:10.1193/1.3650479
Yazgan, U. and Dazio, A. [2011b] “Simulating maximum and residual displacements of RC
structures: II. Sensitivity”, Earthquake Spectra 27(4), 1203–1218. doi:10.1193/1.3650478
Zhao, J. and Sritharan, S. [2007] “Modeling of strain penetration effects in fiber-based analysis of
reinforced concrete structures”, ACI Structural Journal 104(2), 133–141.