Downloaded from https://iranpaper.

ir
https://www.tarjomano.com https://www.tarjomano.com

Journal of Earthquake Engineering

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

Numerical Issues in Distributed Inelasticity

Modeling of RC Frame Elements for Seismic
Analysis

Armando Calabrese, João Pacheco Almeida & Rui Pinho

To cite this article: Armando Calabrese, João Pacheco Almeida & Rui Pinho (2010) Numerical
Issues in Distributed Inelasticity Modeling of RC Frame Elements for Seismic Analysis, Journal
of Earthquake Engineering, 14:S1, 38-68, DOI: 10.1080/13632461003651869

To link to this article: https://doi.org/10.1080/13632461003651869

Published online: 24 Mar 2010.

Submit your article to this journal

Article views: 1308

View related articles

Citing articles: 43 View citing articles

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

https://www.tandfonline.com/action/journalInformation?journalCode=ueqe20
Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

Journal of Earthquake Engineering, 14(S1):38–68, 2010

Copyright  A.S. Elnashai & N.N. Ambraseys
ISSN: 1363-2469 print / 1559-808X online
DOI: 10.1080/13632461003651869

Numerical Issues in Distributed Inelasticity

Modeling of RC Frame Elements for
Seismic Analysis

ARMANDO CALABRESE1, JOÃO PACHECO ALMEIDA1,

and RUI PINHO2
1
ROSE School, IUSS Pavia, Pavia, Italy
2
Dipartimento di Meccanica Strutturale, Università di Pavia, Pavia, Italy

The sustained development of computational power continues to promote the use of distributed
inelasticity fiber frame models. The current article presents a comprehensive application and discus-
sion of state-of-the-art formulations for the nonlinear material response of reinforced concrete
structures. The broad character of the study is imparted by the joint analysis of the effects of the
type of formulation (displacement based versus force based), sectional constitutive behavior (hard-
ening versus softening response), and numerical integration parameters (such as quadrature method,
mesh definition, or number of integration points). Global and local responses are assessed, along with
a critical review of existing regularization techniques. An experimentally tested cantilever is used to
conduct the study and illustrate the previous features. The example shows that the convergence of
displacement-based meshes under objective response can be much slower than what preceding studies
indicate, unlike their force-based counterpart. Additionally, the physical interpretation of the local
response under softening behavior supports the proposal of a novel regularization scheme for
displacement-based elements, validated through comparison against experimental results.

Keywords Reinforced Concrete (RC); Finite Elements (FE); Fiber Elements; Displacement-
Based Elements (DB); Force-Based Elements (FB); Localization; Regularization

1. Introduction
Distributed inelasticity elements are becoming widely employed in earthquake engineer-
ing applications, either for research or professional engineering purposes. Their advan-
tages in relation to the simpler lumped-plasticity models, together with a concise
description of their historical evolution and discussion of existing limitations, can be
found in Fragiadakis and Papadrakakis [2008]. In particular, it is noted that distributed
inelasticity elements (1) allow the inelastic behavior to spread throughout the entire
element and (2) do not require any calibration of their parameters against the response
of an actual or ideal frame element under idealized loading conditions.
The present study uses the renowned fiber approach to represent the cross section
behavior, where each fiber is associated with a uniaxial stress–strain relationship. Such
models feature additional assets, which can be summarized as follows: no requirement of
a prior moment-curvature analysis of members; no need to introduce any element
hysteretic response (as it is implicitly defined by the material constitutive models); direct
modeling of axial load-bending moment interaction (both on strength and stiffness);
Received 18 September 2009; accepted 20 January 2010.
Address correspondence to Armando Calabrese, ROSE School, c/o EUCENTRE, Via Ferrata 1, 27100
Pavia, Italy; E-mail: [email protected]

38
Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

Numerical Issues in Distributed Inelasticity 39

straightforward representation of biaxial loading; and interaction between flexural

strength in orthogonal directions.
Distributed inelasticity frame elements can be implemented with two main different
finite elements (FE) formulations: the classical displacement-based (DB) ones [e.g.,
Hellesland and Scordelis, 1981; Mari and Scordelis, 1984] and the more recent force-
based (FB) formulations [e.g., Neuenhofer and Filippou, 1997; Spacone et al., 1996].
Mixed formulations have also been developed [e.g., Alemdar and White, 2005]. In a DB
approach the displacement field is imposed, whereas in an FB element equilibrium is
strictly satisfied and no restraints are placed on the development of inelastic deformations
throughout the member. For this reason, FB formulations are extremely appealing for
earthquake engineering applications, where significant material nonlinearities are expected
to occur. From the theoretical viewpoint, the differences between these approaches are well
known [e.g., Alemdar and White, 2005; Hjelmstad and Taciroglu, 2005].
For a hardening type of sectional behavior, both FB and DB elements produce an
objective response at the global (force-displacement) and local (moment-curvature)
levels, whereas the results are nonobjective in the case of a softening sectional law.
This numerical issue, commonly known as strain localization or simply localization, was
firstly discussed by Zeris and Mahin [1988] for DB elements, and Coleman and Spacone
[2001a] studied it for FB elements. It is recalled that, particularly in the case of reinforced
concrete (RC) frame elements, the increase of the axial load ratio can induce the change
between a fully hardening and a softening type of sectional constitutive behavior.
To restore the response objectivity for DB elements, it is usual to define the length of
the mesh extremity elements as the plastic hinge length [e.g., Coleman and Spacone,
2001b; Légeron and Paultre, 2005; Papaioannou et al., 2005]. On the other hand, the main
regularization procedures for FB elements have been proposed by Coleman and Spacone
[2001a], Scott and Fenves [2006], and Addessi and Ciampi [2007]. More recent devel-
opments in the regularization of FB elements have also been suggested by Scott and
Hamutçuoglu [2008] and Lee and Filippou [2009].
Besides the aforementioned studies, significant works are the one by Neuenhofer and
Filippou [1997] regarding the identification of the distinct convergence characteristics of the
DB and FB formulations for hardening systems and the study by Papaioannou et al. [2005],
which placed a particular emphasis on the analysis of the numerical integration scheme.
In contrast with the preceding clear theoretical setting, and to the authors’ knowl-
edge, there is no published study that jointly tackles, from the application viewpoint
concerning structural engineering practice, all the parameters that influence the perfor-
mance of such models: type of formulation (DB versus FB), sectional constitutive
behavior (hardening versus softening response), numerical integration scheme (Gauss-
Legendre versus Gauss-Lobatto), and mesh characteristics (number of elements per
member and number of integration points per element). Additionally, the purpose of
the present article includes the inspection, from a critical viewpoint based on the physical
interpretation of the localization phenomenon, of the main regularization techniques.
Again, this was thought to be of relevance from the structural engineering application
standpoint. Both the global and local-level responses were taken into account.
The following section briefly addresses the main finite element models commonly
used for inelastic analysis of frame elements, focusing in particular on the added merits of
complementing such formulations with the traditional sectional fiber approach.
An inspiring parallelism between the physical occurrence of localized responses and
the numerical counterpart of such phenomenon is then carried out, illustrating the
importance of establishing a connection between the fracture mechanics framework and
the FE context.
Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

40 A. Calabrese, J. P. Almeida, and R. Pinho

Several analyses on a literature case study are then performed according to the
previously described objective. Besides the expected amount of useful information,
some novel aspects deserving special consideration arise from the present work.
In particular, it is shown to be of interest to deepen the comparison between the
convergence of the objective hardening responses of DB versus FB formulations.
Although the delayed convergence of the former type of models was noticed by
Neuenhofer and Filippou [1997], the current article shows not only that the relevance
of this issue can be larger than initially expected but addresses the corresponding
justification.
Finally, the critical interpretation of the local responses indicates that physically
meaningful numerical results call for the introduction of a new regularization proposal for
DB elements. The comparison against the experimental tests of the present cantilever
validates this method for restoring objectivity.

2. FE Models for Inelastic Analyses of Frame Structures

Two main general approaches are used in order to perform numerical analyses of frame
structures accounting for material nonlinearity: lumped (or concentrated) inelasticity and
distributed inelasticity (which includes fiber models). Chronologically, lumped plasticity
formulations have appeared earlier, being used for earthquake engineering purposes since
the 1960s. Nowadays, especially due to increasing computational power, distributed
inelasticity models are also gradually becoming more popular in standard professional
practice.

2.1. Distributed Inelasticity Models

In this approach the entire member is modeled as an inelastic element, the source of such
inelasticity being defined at the sectional level. The global inelasticity of the frame is then
obtained by integration of the contribution provided by each controlling section (Fig. 1a).
A major advantage of such models is the nonexistence of a predetermined length where
the inelasticity can occur, because all the sections can have excursions in this field of
response. Though this approach is a closer approximation to reality, it also requires more
computational capacity; that is, more analysis time, as well as memory and disk space.

Core concrete (confined)

x z
z
Steel bars

Integration sections
Cover concrete (unconfined)

FIGURE 1 (a) Example of controlling sections along the element and (b) section fiber
discretization.
Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

Numerical Issues in Distributed Inelasticity 41

A very common and useful way of computing the sectional response is via a refined
discretization into relatively small domains (Fig. 1b), which follow a material uniaxial
inelastic behavior.
Such an approach, commonly referred to as a fiber model, has an inarguable asset: no
previous calibration of the moment-curvature hysteretic rule is required; therefore, there
is no need for any semi-empirical decision. On the other hand, it is difficult to reproduce
the cyclic flexural-shear interaction and degradation, which is a current subject of wide
research.
Fiber elements are commonly modeled with two different finite element formula-
tions. The displacement-based formulation, herein referred to as DB, is based on the
element’s stiffness. On the other hand, the force-based formulation, denoted in the
following as FB, is built on the element’s flexibility. In a DB element the displacement
field is imposed and the element forces are found by energy considerations. Conversely,
in an FB element the force field is imposed, with the element displacements being
obtained by work equivalence balance.
In the DB case, displacement shape functions are used, corresponding, for instance,
to a linear variation of curvature along the element. In contrast, in an FB approach, a
linear moment variation is imposed; i.e., the dual of previously referred linear variation of
curvature. For linear elastic material behavior, the two approaches obviously produce the
same results, provided that only nodal forces act on the element. On the contrary, in case
of material inelasticity, imposing a displacement field does not enable capturing the real
deformed shape because the curvature field can be, in a general case, highly nonlinear. In
this situation, with a DB formulation a refined discretization (meshing) of the structural
element is required, in order to accept the assumption of a linear curvature field inside
each of the subdomains.
Instead, an FB formulation is always exact, because it does not depend on the
assumed sectional constitutive behavior. In fact, it does not restrain in any way the
displacement field of the element. In this sense this formulation can be regarded as
always ‘‘exact,’’ with the only approximation being introduced by the discrete number of
the controlling sections along the element that are used for the numerical integration.
Such an interesting feature enables modeling each structural member with a single FE
element, therefore allowing a one-to-one correspondence between structural elements
(beams and columns) and model mesh elements. In other words, no meshing is theore-
tically required within each element, even if the cross section is not constant. This is
because the force field is always exact, regardless of the level of inelasticity.
Further, it is immediate with FB formulations to take into account loads acting along
the member, whereas this is not the case for DB approaches. Indeed, DB elements need to
be subdivided in order to approximate the curvature distribution induced by distributed
loads along the frame element, whereas FB elements do not require discretization even
when subjected to span loads.

3. Localization Issues

3.1. Localization From a Physical Viewpoint

In the research of fracture mechanics of concrete, the concepts of size effect and
localization are very well known and have been studied for several years. For this reason,
there is a large amount of literature focusing on this subject [e.g., Markeset and
Hillerborg, 1995; Weiss et al., 2001]. In this context, the term localization indicates the
experimental evidence that the descending stress–strain curve becomes specimen size
Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

42 A. Calabrese, J. P. Almeida, and R. Pinho

(b)
(a)
Damage

Moment
zone (a)
Damage
zone
(b) (b)
(b) (a)
(a)
Strain Curvature

FIGURE 2 Localization and size dependence in uniaxial and flexural compression.

dependent and, consequently, it cannot be considered as a pure material property. It has also
been established that strain softening is often spread over a finite-size region of the
material. The above findings mean that a concrete specimen tested in compression will
be damaged, and eventually ‘‘collapse,’’ due to a local mechanism, caused by the concen-
tration of strains into a limited (localized) region of the whole body. As a result, the global
response, namely, the stress–strain curve, does not depend uniquely on concrete properties
but on the specimen size (size effect), regardless of the adopted test procedure; see Fig. 2.
As the sample becomes longer, the total post-peak response of the whole specimen
becomes steeper or more brittle, leading to the possible occurrence of severe snapbacks,
particularly in the testing of high-strength concrete. It should be noted, however, that
localization was first observed for tensile tests, where it is more immediate to think about
a localized deformation (a crack), and later the concept has been extended to compres-
sion. Hence, these two physical phenomena, localization and size effect, exist both in
tension and in compression. Several models have been proposed by different authors to
describe this complex behavior [Bažant, 1989; Bažant et al., 1984; Hillerborg, 1990,
among others].

3.2. Localization From a Numerical Viewpoint

In the previous paragraph, the definition of localization in the framework of fracture
mechanics has been presented. The same term is also frequently used in computational
mechanics to indicate an analogous FE pathology that occurs under the same physical
conditions; i.e., softening constitutive behavior. Zeris and Mahin [1988] indicated that a
conventional DB formulation is unable to establish solutions associated with softening
behavior. Coleman and Spacone [2001a] comprehensively addressed the problem in an
FB context. For illustrative purposes, based also on the findings of the preceding works,
Fig. 3 qualitatively illustrates such a drawback for distributed inelasticity frame elements,
rendering clear that the inelasticity concentration is dependent on the particular choices of
the mesh and/or the numerical integration scheme. Note how the numerical results
converge to a stable solution in the case of a hardening response, with the opposite
happening in the case of softening behavior.
The above numerical localization may be therefore regarded as ‘‘fictitious,’’ because
it happens as a consequence of the finite element assumptions. However, the term appears
to be appropriate because in both situations the result is that damage (real damage in the
first case, computed damage in the second) is localized into a limited zone. Logically, it
would be extremely useful to relate both phenomena—the physical one and the analy-
tical—which would require a sort of interface between the findings of fracture mechanics
and the finite element formulation context. Unfortunately, such relation, though very
appealing, is also extremely difficult to find due to (1) the mentioned limitations of the
Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

Numerical Issues in Distributed Inelasticity 43

HARDENING behaviour – OBJECTIVE response:

F Force - Displacement h Curvature Profile

P
Increasing number of IP’s (FB)
Δ, F or
Finer mesh discretization (DB)

Δ χ
h
SOFTENING behaviour – NON-OBJECTIVE response:

F Force - Displacement Curvature Profile

h

Δ χ

FIGURE 3 Nonlinear static analysis of a simple cantilever—objective and nonobjective

response corresponding to hardening and softening sectional behaviors.

former in completely understanding the physical phenomena (or, at least, in efficiently

describing them) and (2) the difficulties of the latter in including such findings.
Displacement-based finite elements that show softening behavior, such as, for
instance, RC columns modeled under a high axial load ratio and strong lateral loading,
undergo localization because the strains tend to concentrate into the mesh element
subjected to the highest bending moment. Considering the example of a cantilever
under a constant high axial load and a lateral (monotonic/cyclic) displacement history
imposed to the top free extremity, curvatures are expected to concentrate in the bottom
(base) element. This holds regardless of the mesh discretization that is adopted. That is,
independently of the number of subdivisions, the strains always localize in the first
element. Additionally, such features make the response nonobjective, because the com-
puted strains will have to increase at such base element when the mesh is refined, in order
to maintain the same value of the total displacement. Considering that the mesh can
always be more refined (at least, from the theoretical point of view), the strains can also
always increase without converging to any particular value. It will be seen later that, in
fact, the localized response occurs not only in the most strained element, as commonly
accepted, but in particular at the most strained controlling section within such element. At
the global level the post-peak response will also be nonobjective; i.e., different lateral
force values are attained for different meshes and the same imposed lateral displacement.
The higher the number of the mesh elements, the steeper the post-peak response.
Force-based elements are afflicted by an equivalent numerical shortcoming; i.e.,
strains concentrate at the integration point under the highest bending moment. If the
above-mentioned cantilever is modeled with an FB element, the deformation will localize
at the first integration point (near the fixed boundary). Recalling that for the FB
approach—and even for an inelastic response—no meshing is required, a one-to-one
correspondence (one finite element per structural member) will always be adopted
throughout the current work. The nonobjective response in FB elements is thus
Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

44 A. Calabrese, J. P. Almeida, and R. Pinho

characterized by an increase of local strains at the base integration point (IP) as the
number of IPs per element also increases. Such local behavior is reflected in the global
response of the member.
From the previous description it should be evident that both FB and DB elements,
when used to model softening behavior, require regularization procedures in order to
produce consistent and physically significant results. It is also apparent that this is
particularly important in performance-based earthquake engineering (PBEE), where
deformations and displacements of structural members have to be accurately predicted.

4. Case Study
In order to illustrate the influence of finite element formulation, mesh discretization, and
numerical quadrature rule on the computed output, an experimental test is used as a
benchmark. The chosen specimen comes from the work of Tanaka and Park [1990],
where it is referred to as column unit No. 7. This is a largely referred RC cantilever,
loaded with an axial load ratio of 30% and cyclically tested in the horizontal direction by
an imposed displacement at the free node. The same test has also been modeled by
Coleman and Spacone [2001a] first and Scott and Fenves [2006] later. The details of the
specimen are reported in Table 1 and in Table 2. The OpenSees software [Opensees,
2009] is used to perform all the analyses of this study.

4.1. Numerical Moment-Curvature Analyses

The transversal section has been discretized in 250 layers. For the concrete, the Kent and
Park [1971] model has been used. It has to be noted that, although more complex
hysteretic rules are available, the analyses herein presented are monotonic. Moreover,
this model can be easily modified in order to simulate a constant fracture energy, as
proposed by Coleman and Spacone [2001a]. Regarding the concrete parameters, the cover
concrete has been modeled with fc0 ¼ 32:1MPa, "0 ¼ 0:0024mm=mm, and
"20 ¼ 0:0248mm=mm. The confinement effect on the cover concrete has been evaluated
with the Mander model [Mander et al., 1988], resulting in a confinement factor of
kc ¼ 1:215 with which the values fcc0 ¼ 39MPa—compression strength of the confined
concrete—and "c max ¼ 0:0052mm=mm—strain corresponding to the maximum concrete
stress—are obtained [these values are the same as those used by Coleman and Spacone,
2001a, and Scott and Fenves, 2006]. For what concerns the reinforcement steel, a bilinear
stress–strain relationship is used, with elastic modulus E = 200.000MPa, yield stress
fy ¼ 510MPa, and 1% strain-hardening ratio.
Figures 4 and 5 show the sectional response for two different axial load ratios: 30 and
3% (corresponding respectively to an axial force of 2900 and 290kN). The former value
is used in the experiment described above, whereas the latter will be applied in the
following sections to highlight the differences caused by a sectional softening behavior in
contrast to a hardening pattern.
From the comparison between the two curves it is immediate to detect the effect of
the axial load in both increasing the carried moment and reducing the available sectional
curvature ductility. In order to estimate the strain limits for moment-curvature analysis,
the proposals and formulas indicated by Priestley et al. [2007] have been adopted. For the
damage control compression strain, the confined concrete compression strain limit is
taken to occur when fracture of the transverse reinforcement confining the core happens.
For the present case this condition results in a value of "c;dc ¼ 0:0289. A steel tension
strain limit of "s;dc ¼ 0:6"su ¼ 0:06 is also considered. These two limits are indicated in
 TABLE 1 Material properties of the specimen tested by Tanaka and Park [1990]
Concrete Longitudinal reinforcement Transverse reinforcement
Compressive strangth Modulus of rupture Yield strength Ultimate strangth Yield strength Ultimate strength
Downloaded from https://iranpaper.ir

f 0c ½MPa fr[MPa] fy[Mpa] fu[Mpa] fy[Mpa] fu[Mpa]

32.1 4.4 511 675 325 429

TABLE 2 Geometry and detailing of the specimen tested by Tanaka and Park [1990]
Dimension of columns Shear span/full section depth Longitudinal ratio Hoops sets in end regions Axial load ratio
Spacing [mm] Volumetric ratio (to concrete core)
550 · 550 3.0 1.25% 90 2.08% 30%
https://www.tarjomano.com

45
https://www.tarjomano.com
Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

46 A. Calabrese, J. P. Almeida, and R. Pinho

1000

800

Moment [kN*m]
600

400

200
Concrete damage control strain limit
Steel damage control strain limit
0
0.00 0.10 0.20 0.30 0.40 0.50
Curvature [m–1]

FIGURE 4 Moment-curvature diagram for 30% axial load ratio with indication of
damage control strain limits.

800

600
Moment [kN*m]

400

200
Steel damage control strain limit
Concrete damage control strain limit
0
0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35
Curvature [m–1]

FIGURE 5 Moment-curvature diagram for 3% axial load ratio with indication of

damage control strain limits.

Fig. 4 and Fig. 5 above, showing that with low axial load ratio the steel limit is reached
first, and failure due to crushing of the concrete core controls for higher axial load ratios.

5. Results Obtained With the FB Formulation

The response computed with an FB approach is presented for both hardening and soft-
ening conditions. The different behaviors of the element in the two cases are therefore
addressed. Different regularization procedures are then analyzed and compared.

5.1. Objective Response for FB Elements—Hardening Behavior

Very frequently, reinforced concrete elements show essentially a hardening beha-
vior. This leads to an objective response, in the sense that the computed output will
always converge to a unique value. In order to show this characteristic, several
analyses have been performed under hardening conditions. Two hundred fifty layers
are again used to discretize the section. P-
effects were not considered in the
Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

Numerical Issues in Distributed Inelasticity 47

500

400

Horizontal load [kN]

300

200
3 Integration points
4 Integration points
100
5 Integration points
6 Integration points
0
0.00 0.01 0.02 0.03 0.04 0.05
Horizontal displacement [m]

FIGURE 6 Objective global response at 2.5% drift for FB formulation (hardening

response).

analyses, in order to provide a fair comparison between the formulations and

between the theory and the experimental results. The structural member has been
modeled with a single FB element (recall the one-to-one correspondence discussed
in section 2.1). The Gauss-Lobatto quadrature rule was used to numerically inte-
grate the FB elements. Figure 6 depicts the nonlinear static curves comparison
between different numbers of IPs.
It is possible to observe that all the lines get close to each other as the number of IPs
is increased, meaning that the response is objective at the global level. In fact, the results
are virtually unaffected by the number of controlling sections, because all the curves have
the same trend. From Fig. 7 the convergence at the local level can also be assessed, as
well as the exact satisfaction of equilibrium provided by the FB element (note that the
moment profile should be a straight line).

1 1
4 Integration points

6 Integration points
0.8 0.8
8 Integration points
Normalized column length [-]

0.6 0.6

0.4 0.4

0.2 0.2

0 0
800 600 400 200 0 0 0.05 0.1 0.15
Moment [kN*m] Curvature [m–1]

FIGURE 7 Moment and curvature profiles at 2.5% drift (hardening response).

Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

48 A. Calabrese, J. P. Almeida, and R. Pinho

600

500

Horizontal load [kN]

400

300

200 4 Integration points

5 Integration points
100
6 Integration points
0
0.00 0.01 0.02 0.03 0.04
Horizontal displacement [m]

FIGURE 8 Nonobjective global response for FB formulation.

5.2. Nonobjective Response for FB Elements—Softening Behavior

A single element of 1.65 m is again used, and only the number of Gauss-Lobatto IPs is
changed for the various analyses. Under this increased value of the axial load ratio, the
global response is highly nonobjective, meaning that the results do not tend to a unique
value as the number of IPs is increased. The term localization is appropriate because all
the inelastic deformations concentrate at the first controlling section as soon as the
sectional response enters in the softening branch (i.e., negative tangent stiffness). As a
matter of fact, the computed curves are strongly affected by the number of IPs and do not
follow a single pattern after the peak global response is overtaken. This is depicted by
Fig. 8, the features of which clearly contrast with those of Fig. 6.
It is noted that such curves qualitatively match the ones obtained by Coleman and
Spacone [2001a] and reproduced later by Scott and Fenves [2006]. It is also important to
realize that this numerical output has no physical meaning, as simply an FE formulation
output. This is even more evident by looking at the sectional responses of Fig. 9, where
the localized curvatures attain unrealistic values for higher number of IPs.
In order to provide an immediate picture of the phenomenon, the moment and
curvature profiles up to eight IPs for a pre-peak and a post-peak condition are reported
respectively in and Fig. 11. It is apparent that the post-peak behavior is characterized by a
dramatic localization of deformation and that the output generated by the FE formulation
at this stage is without physical meaning, both in terms of curvatures and in terms of
moments. In fact, as can be detected comparing the left plots of Figs. 10 and 11, the

1000 1000
4 Integration points
800 5 Integration points 800
Moment [kN*m]
Moment [kN*m]

6 Integration points
4 IPs end of the series
600 600
5IPs end of the series
6IPs end of the series
400 400

200 200
(i) (ii)
0 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0 0.1 0.2 0.3 0.4 0.5 0.6
Curvature [m–1] Curvature [m–1]

FIGURE 9 Cantilever base moment-curvature response for (1) 1.2% and (2) 3% drift.
Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

Numerical Issues in Distributed Inelasticity 49

1 1
4 Integration points
6 Integration points
0.8 0.8 8 Integration points

Normalized column length [-]

0.6 0.6

0.4 0.4

0.2 0.2

0 0
1000 750 500 250 0 0 0.005 0.01 0.015 0.02 0.025
Moment [kN*m] Curvature [m–1]

FIGURE 10 Moment and curvature profiles at 0.6% drift (post-peak response).

1 1
4 Integration points
6 Integration points
0.8 0.8 8 Integration points
Normalized column length [-]

0.6 0.6

0.4 0.4

0.2 0.2

0 0
800 600 400 200 0 0 0.2 0.4 0.6 0.8
Moment [kN*m] Curvature [m–1]

FIGURE 11 Moment and curvature profiles at 2.5% drift (pre-peak response).

moment profile along the length of the column is no more captured after the peak
response is attained. This is another consequence of the loss of objectivity in the force
prediction, already depicted in Fig. 8 and clearly noticeable in Fig. 12. The element
forces, and consequently the moment diagrams, depend on the number of integration
points once the softening branch is reached (Fig. 12), because element compatibility has
to be verified. This results, in turn, in the divergence in the moment profiles of Fig. 11.

5.3. Regularization Techniques for FB Elements

For the case of softening sectional responses, the need is evident for a regularization procedure
leading to consistent results; i.e., simultaneously objective and of physical value. Coleman
Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

50 A. Calabrese, J. P. Almeida, and R. Pinho

600

500

Horizontal load [kN]

400

300

200 4 IPs - Regularized

5 IPs - Regularized
6 IPs - Regularized
100 4 Integration points
5 Integration points
6 Integration points
0
0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08
Horizontal displacement [m]

FIGURE 12 Regularized global response using the constant fracture energy approach.

and Spacone [2001a] proposed a technique where the concrete uniaxial relationship is
modified, in order to maintain a constant fracture energy Gf. Assuming a value of Gf = 180
MPa for the present case study, the inelastic curves reported in Fig. 12 can be obtained.
The procedure then requires post-processing the curvatures in order to obtain the
regularized sectional response, which is done geometrically assuming a plastic hinge
length, herein referred to as Lp. This length can be estimated, for instance, by using the
formula proposed by Paulay and Priestley [1992]:

Lp ¼ 0:08L þ 0:022fy db ðkN; mmÞ ð1Þ

It is highlighted that, although the previous quantity is a conventional parameter, it is

taken in the present study as the actual length over which the real plastic curvatures
extend, also known as length of plastification. Scott and Fenves [2006] proposed a novel
plastic hinge integration method to restore objectivity for softening response. This
procedure is already available in OpenSees (BeamWithHinges element) and was used in
the present work (Figs. 13 and 14) for further comparison with experimental data.

600

500
Horizontal load [kN]

400

300

200 Lp = 0.08L + 0.022fy·db

Lp = 0.5·h
100 Experimental (1st cycle)
Experimental (2nd cycle)
0
0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08
Horizontal displacement [m]

FIGURE 13 Regularized global response up to 5% drift using the BeamWithHinges

element.
Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

Numerical Issues in Distributed Inelasticity 51

1000

800

Moment [kN*m]
600

400
Lp = 0.08L + 0.022fy·db
Lp = 0.5·h

200 Concrete damage control strain limit

Experimental data
Theoretical monotonic prediction by Tanaka and Park [1990]
0
0.000 0.025 0.050 0.075 0.100 0.125 0.150 0.175 0.200
Curvature [m–1]

FIGURE 14 Regularized local response up to 5% drift using the BeamWithHinges element.

If a specific regularization technique is not available, it is worthy to note that a fairly

common procedure in what concerns FB elements is to set the number of integration
points in such a way that the plastic hinge length is approximately equal to the length
associated to the first controlling section. By doing so, the physical aspect of localization
could be reflected in the numerical model. This possibility has the disadvantage that an
error has to be in most cases accepted for short elements, because the distance associated
with the first IP generally does not exactly coincide with the plastic hinge length.
Moreover, in case of long elements, additional nodes can be added to the frame model
to adjust the plastic hinge length with the first integration point’s length but, obviously,
the advantage of the FB one-to-one correspondence is lost. Very short elements, such as
the cantilever example of this study, cannot in general be regularized in this way.
Finally, notice that not only the number of controlling sections can be changed but
the numerical integration rule could be defined to set a correspondence between the
plastic hinge length and the outermost integration point length, in a similar fashion to that
proposed by Scott and Fenves [2006].

6. Results Obtained With the DB Formulation

Analogously to the previous section, the case study is here analyzed by making use of a
DB approach. The same analyses are performed and the consequences of this formulation
will be highlighted.

6.1. Objective Response for DB Elements—Hardening Behavior

As an initial test, the cantilever is modeled with a number of elements ranging from two
to ten. Each element is integrated with two Gauss-Legendre points. All elements have the
same dimensions within each discretization (i.e., 1.65/2, 1.65/3, 1.65/4 m and so on), with
no meshing refinements at the base, because the purpose of the analyses is only to
illustrate the effect of the FE formulation.
Figure 15 shows that the number of elements does not significantly change the global
results in the case of a positive sectional second stiffness, provided that at least four
elements are used. In fact, the curves all have the same trend and, more importantly, converge
Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

52 A. Calabrese, J. P. Almeida, and R. Pinho

450

400

350
Horizontal load [kN] 300
2 DB Elements
250 3 DB Elements
4 DB Elements
200 5 DB Elements
6 DB Elements
150
7 DB Elements
100 8 DB Elements
9 DB Elements
50 10 DB Elements
0
0.00 0.01 0.02 0.03 0.04 0.05
Horizontal displacement [m]

FIGURE 15 Objective global response at 2.5% drift for DB formulation (hardening

response).

to a unique response as the number of elements per member increase—objective response.

Notice that, in the elastic range of behaviors, a single-element mesh would also provide the
same results. The differences can be detected in the nonlinear range of behaviors. Here,
increasing the number of elements generally produces a lower, more realistic, estimation of
the lateral load. This has to be so, because increasing the number of elements yields a more
flexible model. It is important to realize, however, that though the global response can be
satisfactorily predicted with any of the discretizations with more than three DB elements,
none of them would provide suitable estimates of the local quantities. That could only be
achieved with a much finer mesh, as indicated by Figs. 16 and 17.
Figure 16 also shows an interesting feature that does not occur for the FB formulation
(see Fig. 9): the sectional moment-curvature curves show some scatter, although they
refer to (approximately) the same cross section under the same constant value of imposed
axial load on the cantilever. Intuitively, a perfect match between all the curves could be
expected (regardless of the actual moment-curvature position being dependent on the
mesh discretization). Another observation comes from the moment profiles depicted in

700

600

500
Moment [kN*m]

400
2 DB Elements

300 3 DB Elements

5 DB Elements
200
10 DB Elements
100
200 DB Elements
0
0.00 0.03 0.06 0.09 0.12 0.15
Curvature [m–1]

FIGURE 16 Nonobjective local response at 2.5% drift for DB formulation (hardening

response) in the first IP.
Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

Numerical Issues in Distributed Inelasticity 53

1 1
2 DB Elements
3 DB Elements
0.8 0.8 5 DB Elements
10 DB Elements

Normalized column length [-]

200 DB Elements
0.6 0.6

0.4 0.4

0.2 0.2

0 0
800 600 400 200 0 0 0.05 0.1 0.15
Moment [kN*m] Curvature [m–1]

FIGURE 17 DB formulation: moment and curvature profiles at 2.5% drift (hardening

response).

Fig. 17, which are not coincident. The reason for both features lies in the fact that DB
elements satisfy equilibrium only in an average (integral) sense. This aspect will be
extensively addressed later in this article.

6.1.1. Efficiency of the Numerical Integration Scheme. In addition to the influence of the
mesh discretization, the effect of the numerical quadrature rule is investigated. Figure 18
shows the consequence of the integration scheme in the computed global response.
A mesh with three DB elements of length Lcol/3 is used. The use of two Gauss-
Legendre points, or three Gauss-Lobatto points, allows the exact integration of a polynomial
of the third degree. The other two schemes (four Gauss-Legendre IPs and five Gauss-Lobatto

500

400
Horizontal load [kN]

300

200
10 DB Elements (2 Gauss-Legendre IPs)
2 Gauss-Legendre integration points
100 4 Gauss-Legendre integration points
3 Gauss-Lobatto integration points
5 Gauss-Lobatto integration points
0
0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
Horizontal displacement [m]

FIGURE 18 Three DB elements (L/3 each)—effect of the integration scheme in the

computed global response (note that the lines corresponding to two through five IPs are
virtually coincident).
Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

54 A. Calabrese, J. P. Almeida, and R. Pinho

700

600

Horizontal load [kN] 500

400

300
10 DB Elements (2 Gauss-Legendre IPs)
200 2 Gauss-Legendre integration points
3 Gauss-Legendre integration points
100 4 Gauss-Legendre integration points
5 Gauss-Legendre integration points
0
0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
Horizontal displacement [m]

FIGURE 19 One DB element—effect of the different number of integration points.

IPs) provide the exact result for a polynomial of the seventh degree. As is evident from the
plot, no effect of the integration scheme can be detected at this level by looking at the element
response, even when a relatively crude three DB elements mesh is used. To check whether the
influence of the number of IPs was being ‘‘hidden’’ by the meshing of the cantilever, it was
decided to model the member with a single DB element; see Fig. 19.
Obviously, such modeling is not intended to suitably represent the inelastic behavior
of the element (which is apparent from the comparison with the ten DB-elements mesh
results). Although the curves corresponding to different number of IPs can now be
distinguished, the fact that no major differences are recognizable indicates that two
Gauss-Legendre points are sufficient, in any case, to integrate with enough accuracy
the DB-element quantities. Figure 20 shows the results for a two DB-element mesh and
different number of IPs.
It is no longer possible to distinguish the different series, which further supports the
previous observation. However, notice that the mesh chosen is still too crude to provide

500

400
Horizontal load [kN]

300

200
10 DB Elements (2 Gauss-Legendre IPs)
2 Gauss-Legendre integration points
100 3 Gauss-Legendre integration points
4 Gauss-Legendre integration points
5 Gauss-Legendre integration points
0
0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
Horizontal displacement [m]

FIGURE 20 Two DB elements (L/2 each)—effect of the different number of integration

points (note that the lines corresponding to two through five IPs are virtually coincident).
Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

Numerical Issues in Distributed Inelasticity 55

satisfactory results. Based on the above considerations, it can be stated that, for sectional
hardening behavior and displacement-based elements (1) global results are objective;
(2) apparently there is no justification to use more than two Gauss-Legendre IPs per
element; (3) a mesh of at least four elements per member is required to provide
sufficiently accurate results at the global level. Therefore, throughout the following
analyses a constant number of Gauss IPs per element equal to two will be used—this is
already the case of the models used to get an insight into the local behavior of the
element, depicted in Fig. 17.

6.2. Nonobjective Response for DB Elements—Softening Behavior

In section 5.2, the localization phenomenon for FB elements was presented. However,
because the concentration of deformations depends on the softening response of the
sectional constitutive relation, it is expected that a similar occurrence can also be found
with the more classical stiffness formulation. Setting the axial load to the real value used
in the experiment, localization effects do show up at the global level (Fig. 21). Moreover,
a finer discretization yields a steeper descending branch, which is clearly symptomatic of
the concentration of deformations. Again, within the elastic range of behaviors of the
element, there is no influence of the mesh on the calculated response.
Similarly to what already happened for the hardening condition, in this case there is a
marked divergence in the local quantities’ predictions, as can be detected from the
moment-curvature diagrams at the first Gauss section in Fig. 22.
It can be noted that the above moment-curvature curves are not coincident. From a
conceptual point of view, only the point associated with the 2.5% drift along the moment-
curvature curve should change, not the curve itself (recall that the section and the
imposed value of the cantilever axial load are the same in all cases). The reason behind
such discrepancy is again the fact that the DB formulation does not strictly verify
equilibrium. This characteristic implies that the damage control strain limit turns into a
range of values; such a damage control bound for concrete is indicated with a grey shade
in Fig. 22. Finally, by looking at the profiles obtained for the softening response (Fig. 23),

700

600

500
Horizontal load [kN]

400

300
4 DB Elements
200 6 DB Elements
8 DB Elements
100
10 DB Elements
0
0.00 0.01 0.02 0.03 0.04
Horizontal displacement [m]

FIGURE 21 Nonobjective global response at 2.5% drift for DB formulation (softening

response).
Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

56 A. Calabrese, J. P. Almeida, and R. Pinho

1200

1000

Moment [kN*m] 800

Concrete damage control strain limit

600

2 DB Elements
400
4 DB Elements
6 DB Elements
200
8 DB Elements
10 DB Elements
0
0.00 0.05 0.10 0.15 0.20 0.25
Curvature [m–1]

FIGURE 22 Nonobjective local response at 2.5% drift for DB formulation (softening

response) in the first IP.

1 1
4 DB elements
6 DB elements
0.8 0.8 8 DB Elements
10 DB Elements
Normalized column length [-]

0.6 0.6

0.4 0.4

0.2 0.2

0 0
800 600 400 200 0 0 0.1 0.2 0.3
Moment [kN*m] Curvature [m–1]

FIGURE 23 DB formulation: moment and curvature profiles at 2.5% drift (softening

response).

it is possible to appreciate that there is significant variability in both moment and

curvature distributions.

6.3. Regularization Techniques for DB Formulations

Although DB elements are still more widely employed, a limited number of regulariza-
tion procedures are found in the literature. It is usual to regularize them by assuming that
Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

Numerical Issues in Distributed Inelasticity 57

700

600

Horizontal load [kN]

500

400

300 2 DB Elements, Length Bottom El. = Lp

3 DB Elements, Length Bottom El. = Lp
200 4 DB Elements, Length Bottom El. = Lp
5 DB Elements, Length Bottom El. = Lp
100 Experimental (1st cycle)
Experimental (2nd cycle)
0
0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
Horizontal displacement [m]

FIGURE 24 Softening case. DB regularized response up to 5% drift with bottom DB

element length = Lp.

the extremity elements’ lengths are equal to a multiple of the plastic hinge length. It is
commonly believed that the use of displacement interpolation functions force localization
within a single element instead of one integration point. Such an underlying principle
justifies the regularization criterion of defining the length of the most strained element as
equal to the plastic hinge length, as mentioned in the introduction. Applying this criterion
to the case study cantilever yields the ‘‘regularized’’ global response of Fig. 24, which is
compared to the experimental results.
The findings from the present work (see also section 7.2, Fig. 34), however, seem to
indicate that, at least for the current DB model, localization effectively occurs within the
first IP, similarly to FB elements. This behavior was also encountered in one of the first
studies on the topic [Zeris and Mahin, 1988]. Considering the case of two Gauss points
per element, this means that the length of the extremity elements should therefore be
defined as equal to twice Lp and not only Lp as generally accepted. This effectively means
that the distance corresponding to each IP, where inelasticity will concentrate, is the same
as the plastic hinge length. In other words, a physically consistent regularization techni-
que for the DB formulation requires an intuitively paradoxical definition of the most
strained mesh elements’ lengths as being distinct from the plastic hinge length (or, to be
more precise, the length of plastification). In this way, the correspondence between the
analytical and the physical deformations is kept, because the inelastic response will be
integrated along the appropriate distance. In fact, if such extremity mesh elements’
lengths are exactly equal to the plastic hinge length, the curvature will be overestimated,
as will shortly be seen. For the cantilever under analysis, this theoretically derived
principle does not manage to provide a very satisfactory estimate of the global behavior;
see Fig. 25. However, this latter result can be considered as artificially misleading
because the ratio between the base element’s length (=0.71m) and the total column height
(=1.65 m) is considerably larger than the expected equivalent ratio for common-sized RC
members.
The confirmation of such criterion validity emerges through the comparison of the
local response, which is incomparably closer, in terms of 5% drift point, to the experi-
mental results, as indicated by the ending curvatures depicted in Fig. 26.
Therefore, when modeling softening behavior with DB elements, the possibility of
defining the length of the most strained elements (generally, the extremity ones) as 2·Lp
Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

58 A. Calabrese, J. P. Almeida, and R. Pinho

800

700

600

Horizontal load [kN] 500

400

300
2 DB Elements, Length Bottom El. = 2 x Lp
200 2 DB Elements, Length Bottom El. = Lp
Experimental (1st cycle)
100 Experimental (2nd cycle)
0
0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
Horizontal displacement [m]

FIGURE 25 Softening case; comparison between different bottom DB element lengths

(2 Lp and Lp). Global regularized response up to 5% drift.

1000

750
Lbottom elem = 2LP
Moment [kN*m]

Lbottom elem = LP
500

2 DB Elements, Length Bottom El. = 2 x Lp

250
2 DB Elements, Length Bottom El. = Lp
Experimental data
Theoretical monotonic prediction by Tanaka and Park [1990]
0
0.00 0.05 0.10 0.15 0.20 0.25
Curvature [m–1]

FIGURE 26 Softening case; comparison between different bottom DB element lengths

(2 Lp and Lp). Moment-curvature regularized response up to 5% drift.

instead of simply Lp should be critically assessed. Although the previous considerations

assume two Gauss-Legendre points per element, it is noted that the use of a different number
of IPs would imply an adjusted length associated to the most strained elements, according to
the same rationale. It is believed that this alternative regularization procedure for DB
elements should be considered by engineers and researchers that use software packages
where this type of formulation is implemented. However, the proposed approach, though
adequate for softening conditions, could be unsuitable for modeling elements with hardening
sectional behavior. In this case, in fact, a greater accuracy is obtained by reducing the size of
the element subjected to the largest excitation, as will be analyzed in detail in section 7.1. It is
again observed that the novel regularization technique presented in this work was based on
the plastic hinge length in view of the fact that this value has often been used for the same
purpose in similar studies. However, the influence on the accuracy of the results regarding
the use of such conventional quantity, versus the employment of a physical characteristic
length like the so-called length of plastification, should be thoroughly investigated. Other
Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

Numerical Issues in Distributed Inelasticity 59

500

400

Horizontal load [kN] 300

200

100 Displacement formulation

Force formulation
0
0 0.02 0.04 0.06 0.08 0.1
Horizontal displacement [m]

FIGURE 27 Comparison of the DB and FB inelastic curves of the test column.

general considerations on the use of a plastic hinge length as a regularization parameter can
be found in Calabrese [2008].

7. Comparison Between FB and DB Elements

The first example aims to underline the fundamental differences between the two
formulations, recalling the discussion presented in section 2.1. The response of a simple
inelastic analysis considering sectional hardening behavior is depicted in Fig. 27, for
which a two-element mesh with two Gauss-Legendre IPs per element has been used.
As expected, though the overall trend is the same for the two formulations, the DB
model is artificially stronger than the FB one, and also stiffer, as soon as the element goes
into the inelastic range. Similar considerations on the renowned effects caused by the two
formulations were also presented by Coleman and Spacone [2001b].
For what concerns the softening case, it can be observed in Fig. 28 that the
differences in the post-peak branches of the DB elements are reduced with respect to
those of their FB counterparts. The discrepancies in the DB results begin to be more
detectable, increasing greatly the number of subdivisions. Regarding the numerical
integration procedure, it has to be recalled that in these examples the Gauss-Legendre
quadrature rule is used, which does not include any integration point at the base of the
cantilever. On the contrary, when illustrating the flexibility formulation, the Gauss-
Lobatto rule sets an integration point at the base of the column, capturing in this way
the highest moments and the highest deformations. For this reason, the nonobjectivity
obtained with the DB formulation is ‘‘regularized’’ by the element dimension and by the
absence of a controlling point at the more strained section.

7.1. Local Convergence for FB and DB Formulations

Figures 16 and 17 indicate that the curvature estimation provided by the DB formulation,
even under hardening conditions, is affected by localization, because an extremely large
number of subdivisions is required in order to converge to a unique value. It is therefore
useful to assess the curvature at the first integration point as a function of the number of
DB elements used to discretize the cantilever. Figure 29 refers to the curvature at the first
controlling section resulting from a nonlinear static analysis up to 2.5% drift under
hardening conditions.
Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

60 A. Calabrese, J. P. Almeida, and R. Pinho

700

600

Horizontal load [kN]

500

400

300
4 DB Elements
200 6 DB Elements
8 DB Elements
10 DB Elements
100 FB Element - 4 IP s
FB Element - 5 IP s
FB Element - 6 IP s
0
0.00 0.01 0.02 0.03 0.04
Horizontal displacement [m]

FIGURE 28 Nonobjective global responses at 2.5% drift for DB and FB formulation

(softening response).

# IPs, FB
2 3 4 5 6 7 8 9 10
0.16

0.14
Curvature [m–1]

0.12

0.1

0.08

0.06
DB formulation
FB formulation
0.04
1 100
# Elements, DB

FIGURE 29 Hardening case. Curvature at the first IP at 2.5% drift for different mesh
refinements (DB) and numbers of IPs (FB).

The outcomes from FB and DB models are plotted in the same graph. The FB results are
obtained by using a single element and a varying number of Gauss-Lobatto IPs. Therefore,
the two horizontal axes in the figure refer respectively to the number of DB elements (bottom
axis) and to the number of Gauss-Lobatto IPs for the FB element (top axis). Note also that the
axis reporting the number of elements is in logarithmic scale, whreas the other is linear.
It is noted that the curvatures obtained with the FB formulation are at the first Gauss-
Lobatto IP; i.e., they are computed exactly at the bottom section of the cantilever. On the
other hand, the curvatures from the DB analyses are all associated with a different
abscissa, because they correspond to the first Gauss-Legendre section in the bottom
Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

Numerical Issues in Distributed Inelasticity 61

element of the mesh (this difference becomes negligible with increasing number of
elements; therefore, general comments and comparisons are valid).
Several features become immediately noticeable from the two plots of Fig. 29. As
expected, the curvature values obtained with a single FB element are generally larger than
the ones predicted by the DB models. This result confirms that the FB element does not
restrain the development of inelastic deformations, as happens with the DB element due
to the imposed linear curvature distribution. Furthermore, both formulations eventually
converge to a unique curvature value (around 0.14 m1), meaning that from the theore-
tical viewpoint the response is objective for both cases.
Anyway, the very large number of elements required by the DB mesh to approach the
final asymptotic value allows one to consider the issue of ‘‘practical localization’’; i.e., a
nonobjective prediction that will be encountered if a ‘‘common engineering mesh’’ is
used (note that the finest mesh—200 elements—corresponds to 8-mm-length elements).
The current study shows that this phenomenon can be much more pronounced than
indicated in the previous work of Neuenhofer and Filippou [1997].
Summarizing, the FB formulation is definitely more efficient than the DB one, also in
what concerns the prediction of curvatures, because only a few IPs are sufficient to attain a
converged response (in this case, six IPs provide an acceptable result at the local level).
The accurate sensitivity of the FB formulation is also responsible for producing
marked localized results in the case of softening systems. Figure 30 refers to the nonlinear
static analysis up to 2.5% drift on the cantilever loaded to 30% of its axial capacity (note
that in this plot both horizontal axes are in linear scale).
As previously discussed, both FB and DB formulations provide global and local
nonobjective responses in the case of softening sectional behavior. In both cases the
computed curvatures increase more than linearly with the number of IPs per element (FB
case) or the number of elements per member (DB case). Figure 30 shows that the localized
effects are more pronounced for the FB formulation, at least considering the ‘‘engineering
range of variation’’ commonly used in structural earthquake engineering models.

# IPs, FB
2 3 4 5 6 7 8 9 10
0.16

0.14
Curvature [m–1]

0.12

0.1

0.08

0.06 DB formulation
FB formulation
0.04
1 100
# Elements, DB

FIGURE 30 Softening case. Curvature at the first IP at 2.5% drift for different mesh
refinements (DB) and numbers of IPs (FB).
Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

62 A. Calabrese, J. P. Almeida, and R. Pinho

7.2. Other Considerations on the DB Element (Hardening and Softening Behaviors)

In the previous sections it has been noted that the sectional response computed within the
DB formulation did not follow the sectional moment-curvature relationship represented in
Figs. 4 and 5. Additionally, it varied as a function of the mesh subdivision. Apparently
this may seem incorrect, because the section does not change and the axial load imposed
on the cantilever is also the same. In other words, one could think that the number of DB
elements could affect the stage reached along each computed moment-curvature curve
but never the curve itself. Moreover, even with a hardening response and from a practical
point of view, a nonobjective curvature prediction has been individuated along with an
objective global response. These two aspects, which are due to the same cause, are now
analyzed.
A major difference between FB and DB elements is that, for the former, equilibrium
is strictly satisfied (in fact, it is enforced), whereas for the latter equilibrium is verified
only in an average sense. This means that in the FB formulation the sectional forces (axial
forces and moments) are in exact equilibrium with the external values. On the contrary, in
the DB formulation the equilibrium is not verified point-wise, because it is the weighted
integral of the internal forces that has to equilibrate the element forces. Using FE
terminology, equilibrium is satisfied in a strong form for FB elements but only in a
weak form for DB ones.
The simplest way to assess this fundamental difference is by observing at the
sectional axial forces of a beam-column element. For FB elements, these forces are
always equal to the applied axial load on the element, regardless of the material
constitutive relationship. On the contrary, in the DB case the applied axial load is equal
to the weighted integral of the section axial forces. Therefore, as long as the material has
an elastic behavior (and loads are applied at the extremities of the element), the sectional
axial forces are exactly the same as the axial load acting on the column. Instead, for
nonlinear material behavior, the section axial forces will in general not be equal to the
applied axial load, as illustrated in Fig. 31.
This figure refers to inelastic analyses up to 5% drift with an axial load ratio of 3%.
A two-element mesh (0.5 m + 1.15 m) with two Gauss-Legendre IPs per element is used
for the DB formulation and the corresponding graph shows the results of the first IP of the
first element, at the column’s base. It is apparent that for the FB formulation the sectional
axial force at the first controlling section is strictly in equilibrium with the external axial
load throughout the analysis; i.e., also in the nonlinear range. Oppositely, as soon as the

DB formulation FB formulation
450 450

400 400
Force [kN]
Force [kN]

350 Axial load on the element 350 Axial load on the element

Axial force at the first controlling section Axial force at the first controlling section
300 300

250 250
0.00 0.02 0.04 0.06 0.08 0.00 0.02 0.04 0.06 0.08
Top node horizontal displacement [m] Top node horizontal displacement [m]

FIGURE 31 Sectional versus element axial forces for both formulations up to 5% drift
in the first IP.
Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

Numerical Issues in Distributed Inelasticity 63

600

Sectional axial force [kN]

500

400

300

200

100

0
0.00 0.02 0.04 0.06
Bottom DB element, upper IP Curvature [m–1]
Bottom DB element, bottom IP
Bottom DB element, weighted sum
800
Sectional moment [kN*m]

600

400

200

0
0.00 0.02 0.04 0.06
Curvature [m–1]

FIGURE 32 Sectional axial force and moment recorded at the base DB element up to
2.5% drift (hardening response).

elastic range of behaviors is surpassed, the axial force at the first controlling section of the
DB element starts increasing. Figure 32 illustrates the same concept by depicting the
evolution of the axial force at the top section of the same element.
The similar but ‘‘mirrored’’ pattern that can be found is readily justified because the
weighted sum of the top and bottom axial forces has to equal the imposed external
element load. The symmetry of Fig. 33 is a reminder that the integration weight at both
sections is 0.5.
Finally, the above considerations explain the differences in the moment-curvature paths
highlighted when commenting on the results of Figs. 16 and 22. The most important
consequence of such inconsistency is that it is conceptually incorrect to evaluate the moment-
curvature sectional response from a fiber-section DB element (because it does depend on the
element characteristics), even for a hardening behavior. On the other hand, the moment-
curvature sectional response extracted from an FB element is the ‘‘real’’ one; i.e., the same
that would be predicted by a moment-curvature analysis on the isolated section.
Another interesting observation comes from the local response of two integration
sections belonging to the same DB element. Figures 32 and 34 (respectively hardening
Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

64 A. Calabrese, J. P. Almeida, and R. Pinho

450

400

350 Axial force at the second controlling section

Axial force at the first controlling section
Force [kN]

Weighted sum
300

250

200

150
0.00 0.02 0.04 0.06 0.08
Top node horizontal displacement [m]
FIGURE 33 Axial forces at the two controlling sections of the bottom DB element.

and softening conditions) refer to a nonlinear static analysis up to 2.5% drift. In both
cases the cantilever is modeled with two DB elements, each of them integrated with two
Gauss-Legendre sections.
It is immediate to detect how, for both values of the axial load, the moment-curvature
curves within each element are significantly dissimilar. This is physically untenable,
because the section is the same at the two IPs, and the axial force applied to the cantilever
is constant. This consideration holds for hardening and softening conditions although it is
significantly more evident for the latter case. Figure 34 indicates that the section starts to
unload at a curvature of about 0.022 m1, in correspondence with an increase in the
sectional axial force. It is noted that the element with a hardening response does not
experience such numerical unloading. Recall that for the DB element, all the local
patterns (for hardening and softening conditions) are severely conditioned by the finite
element formulation and thus have an obviously reduced physical meaning. For the
softening case, all but the bottom integration sections undergo unloading, which is an
important observation for the construction of regularization procedures for the DB
formulation, recalled next.

8. Summary of Numerical Issues

The observations derived from the analyses performed on the case study are herein
summarized, in order to provide a clear compendium that can be considered by engineers
and researchers for an accurate modeling of reinforced concrete structures.
Firstly, it is of paramount importance to know the type of element formulation (FB or
DB) available in the software package being used. The DB approach unrealistically
predicts higher element stiffness and strength compared to the ‘‘exact’’ FB formulation,
well-known features among FE researchers [e.g., Coleman and Spacone, 2001b;
Neuenhofer and Filippou, 1997].
Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

Numerical Issues in Distributed Inelasticity 65

3000

Sectional axial force [kN]

Sectional axial force at the upper integration section
Moment peak response
Sectional axial force starts to increase
2500

2000
0.00 0.02 0.04 0.06 0.08 0.10
Curvature [m–1]

1200

1000
Sectional moment [kN*m]

800

600 Lower integration section

Upper integration section

400
Moment peak response
200
Sectional axial force starts to increase
0
0.00 0.02 0.04 0.06 0.08 0.10
Curvature [m–1]

FIGURE 34 Sectional axial force and moment recorded at the base DB element up to
2.5% drift (softening response).

As theoretically expected, DB elements only manage to verify equilibrium in an

average sense. This issue is responsible for two other drawbacks of this formulation.
Firstly, because in general a strict equilibrium is not verified, the combination of forces
acting in each section can be considerably far off the exact equilibrium with real loading
conditions, thus causing an unreliable or, at least, a ‘‘slowed-down evolution’’ of local
quantities. Secondly, if the sectional constitutive model is obtained via a fiber model—
which takes into account the interaction between axial force and bending moment—this
effect is even more dramatic because it gives rise to a sectional behavior that, ultimately,
is dependent on the number of elements per frame.
In the presence of a hardening sectional response, a Gauss-Lobatto integration scheme
should be used for FB elements. Although a lower bound of four IPs per element is required
in order to provide a reliable result at the global level and a very satisfactory response at the
local level, a choice of a larger number of, for instance, six IPs can also be justifiable in order
to obtain a completely stabilized prediction of the local response. It can thus be inferred that
the objective response predicted by FB elements rapidly converges as the number of IPs is
increased, as was also found by Neuenhofer and Filippou [1997].
Still referring to hardening conditions, the local response of a DB element is also
objective, yet the rate of convergence (as a function of the number of elements per member)
is much slower than the corresponding one for the FB. In fact, it is so slow that, for the
Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

66 A. Calabrese, J. P. Almeida, and R. Pinho

usual range of number of elements per member commonly used in structural analysis,
strongly localized results are produced: the value of the curvature for four DB elements is
less than 50% of the ‘‘objective’’ value. This is naturally of the greatest concern, from the
viewpoint of engineering practice, because it implies that local measures obtained from a
DB model (such as curvatures or strains) can be heavily underestimated. The present case
study showed in a straightforward way that this delayed convergence can be much more
significant than what Neuenhofer and Filippou [1997] previously pointed out. It is under-
lined that such a conclusion does not apply to the global output, which is expected to be
accurate for a mesh with few displacement-based elements.
The numerical analyses performed in this study confirmed the lack of conver-
gence in case of softening sectional response, in accordance to seminal literature
references [Coleman and Spacone, 2001a; Zeris and Mahin, 1998]. In fact, when the
peak of the sectional behavior is surpassed, the global and local responses of either
DB or FB formulations become nonobjective. As the number of IPs per element (FB
formulation) or the number of elements per member (DB approach) augments (1) the
global post-peak branch becomes steeper and (2) the curvatures concentrate at the
bottom IP.
In order to overcome this issue, several regularization procedures for FB elements
have been proposed in the literature [e.g., Coleman and Spacone, 2001a; Lee and
Filippou, 2009; Scott and Fenves, 2006; Scott and Hamutçuoglu, 2008]. If the use of
one of these procedures for FB elements is not available, an alternative is still possible by
specifying the length associated with the first integration point such that it is close to the
plastic hinge length. This can be done by selecting an appropriate number of integration
points or even by choosing a particular numerical integration rule. Such an approach is
not very practical and, for standard dimensions of beams, columns, and plastic hinge
lengths, it generally requires some coarse approximation. It may also require meshing in
more than one element per member, therefore weakening one of the main advantages of
the FB formulation.
A commonly used regularization technique for DB meshes is usually accomplished
by specifying the size of the frame extremities’ elements as equal to the appropriate
equivalent plastic hinge length. However, although this approach regularizes the global
response, the actual inelastic deformations are still overestimated. This is due to the fact
that, unlike what is generally assumed, localization in the DB mesh occurred in the most
strained integration point and not in the whole element. Based on this consideration, a
novel approach is suggested to regularize the response of DB elements. This requires the
length of the first mesh element to be larger than the plastic hinge one, depending on the
particular integration scheme adopted.

9. Conclusions
The initial main objective of the present work was to collect disperse information into a
single and consistent manuscript tackling the influence of the main parameters affecting
the performance of distributed inelasticity models. From the application viewpoint con-
cerning structural engineering practice, it is the authors’ expectation that the conclusions
and comments carried out throughout the study can be taken into account by engineers
and researchers for an accurate modeling of reinforced concrete structures.
Several numerical analyses, reproducing an experimental test widely referred to in
recent literature, have been performed. The consequences of different modeling assump-
tions, such as element formulation, mesh discretization, number of integration points, and
Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

Numerical Issues in Distributed Inelasticity 67

numerical integration scheme, were investigated and systematized for hardening and
softening sectional behaviors.
From the analyses performed, a larger-than-expected delay in convergence for the
objective local response of DB elements should be highlighted. Additionally, for the
usual engineering range of number of IPs per element (FB case) and number of elements
per member (DB case), the localized response is more apparent in the FB formulation
than in the DB one. In other words, although in both cases the response is nonobjective
and untrustworthy, the FB local response shows a considerably higher scatter of base
curvatures in comparison to the DB output.
Despite its eminently practical nature, this work also questions commonly estab-
lished concepts. In particular, it was shown that, in the case of DB formulation and
softening conditions, the localized response does not actually concentrate at the element
level but at the IP level as in FB elements. This consideration led to the proposal of a new
physically meaningful regularization technique where the size of the frame extremities’
elements is set to be equal to a multiple of the plastic hinge length.
The validity of such rationale was confirmed by comparison with the experimental
local response. The small dimensions of the cantilever under study caused a misleading
disagreement in terms of the global response, but such deviation is certainly reduced for
normal-sized RC members. Thus, if two Gauss-Legendre points per element are used, the
‘‘regularized’’ length of the extremity elements should be twice the value of the plastic
hinge length. By doing so, the physical meaning of the local response under softening
behavior is naturally reflected in the numerical integration of the sectional quantities.

Acknowledgments
The first author gratefully acknowledges the Italian Air Force for the research leave to
join the Rose School–IUSS. The second author is grateful for the financial support
provided by the Fundação para a Ciência e Tecnologia through the funding program
Programa Operacional Ciência e Inovação 2010.

References
Addlssi, 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.
Alemdar, B. N. and White, D. W. [2005] ‘‘Displacement, flexibility, and mixed beam-column finite
element formulations for distributed plasticity analysis,’’ Journal of Structural Engineering
131(12), 1811–1819.
Bažant, Z. P. [1989] ‘‘Identification of strain-softening constitutive relation from uniaxial tests by
series coupling model for localization,’’ Cement and Concrete Research 19, 973–977.
Bažant, Z. P., Belytschko, T. B., and Chang, T. P. [1984] ‘‘Continuum theory for strain softening,’’
Journal of Engineering Mechanics 110(12), 1666–1692.
Calabrese, A. [2008] ‘‘Numerical issues in distributed inelasticity modeling of RC frame elements
for seismic analysis,’’ M.Sc. dissertation, ROSE School–IUSS Istituto Universitario di Studi
Superiori, Pavia, Italy.
Coleman, J. and Spacone, E. [2001a] ‘‘Localization issues in force-based frame elements,’’ Journal
of Structural Engineering 127(11), 1257–1265.
Coleman, J. and Spacone, E. [2001b] ‘‘Localization issues in nonlinear frame elements,’’ in
Modelling of Inelastic Behaviour of RC Structures Under Seismic Loads, ed. P. B. Shing and
T. Tanabe (ASCE, Reston, VA) pp. 403–419.
Downloaded from https://iranpaper.ir
https://www.tarjomano.com https://www.tarjomano.com

68 A. Calabrese, J. P. Almeida, and R. Pinho

Fragiadakis, M. and Papadrakakis, M. [2008] ‘‘Modeling, analysis and reliability of seismically

excited structures: Computational issues,’’ International Journal of Computational Methods
5(4), 483–511.
Hellesland, J. and Scordelis, A. [1981] ‘‘Analysis of RC bridge columns under imposed deforma-
tions,’’ Proceedings of the IABSE Colloquium, Delft, The Netherlands, pp. 545–559.
Hillerborg, A. [1990] ‘‘Fracture mechanics concepts applied to moment capacity and rotational
capacity of reinforced concrete beams,’’ Engineering Fracture Mechanics 35(1), 233–240.
Hjelmstad, K. D. and Taciroglu, E. [2005] ‘‘Variational basis of nonlinear flexibility methods for
structural analysis of frames,’’ Journal of Engineering Mechanics 131(11), 1157–1169.
Kent, D. C. and Park, R. [1971] ‘‘Flexural members with confined concrete,’’ Journal of Structural
Division 97(7), 1964–1990.
Lee, C. L. and Filippou, F.C. [2009] ‘‘Efficient beam-column element with variable inelastic end
zones,’’ Journal of Structural Engineering 135(11), 1310–1819.
Légeron, F. and Paultre, P. [2005] ‘‘Damage mechanics modeling of nonlinear seismic behaviour of
concrete structures,’’ Journal of Structural Engineering 131(6), 946–955.
Mander, J. B., Priestley, M. J. N., and Park, R. [1988] ‘‘Theoretical stress–strain model for confined
concrete,’’ Journal of Structural Engineering 114(8), 1804–1823.
Mari, A. and Scordelis, A. [1984] Nonlinear Geometric Material and Time Dependent Analysis of
Three Dimensional Reinforced and Prestressed Concrete Frames, SESM Report 82-12,
Department of Civil Engineering, University of California, Berkeley.
Markeset, G. and Hillerborg, A. [1995] ‘‘Softening of concrete in compression—Localization and
size effects,’’ Cement and Concrete Research 25(4), 702–708.
OpenSees [2009] Open System for Earthquake Engineering Simulation [online]. Available from
http://opensees.berkeley.edu. Accessed 1 February 2008.
Neuenhofer, A. and Filippou, F. C. [1997] ‘‘Evaluation of nonlinear frame finite-element models,’’
Journal of Structural Engineering 123(7), 958–966.
Papaioannou, I., Fragiadakis, M., and Papadrakakis, M. [2005] ‘‘Inelastic analysis of framed
structures using the fiber approach,’’ Proceedings of the Fifth GRACM International Congress
on Computational Mechanics, Limassol, Cyprus.
Paulay, T. and Priestley, M. J. N. [1992] Seismic Design of Reinforced Concrete and Masonry
Buildings, John Wiley & Sons, New York.
Priestley, M. J. N., Calvi, G. M., and Kowalsky, M. J. [2007] Displacement-Based Seismic Design
of Structures, IUSS Press, Pavia, Italy.
Scott, M. H. and Fenves, G. L. [2006] ‘‘Plastic hinge integration methods for force-based beam-
column elements,’’ Journal of Structural Engineering 132(2), 244–252.
Scott, M. H. and Hamutçuoglu, O. M. [2008] ‘‘Numerically consistent regularization of force-based
frame elements,’’ International Journal for Numerical Methods in Engineering 76, 1612–1613.
Spacone, E., Ciampi, V., and Filippou, F. C. [1996] ‘‘Mixed formulation of nonlinear beam finite
element,’’ Computers & Structures 58(1), 71–83.
Tanaka, H. and Park, R. [1990] Effect of Lateral Confining Reinforcement on the Ductile Behaviour
of Reinforced Concrete Columns, Report 90-2, Department of Civil Engineering, University of
Canterbury, New Zealand.
Weiss, W. J., Güler, K., and Shah, S. P. [2001] ‘‘Localization and size-dependent response of
reinforced concrete beams,’’ ACI Structural Journal 98(2), 686–695.
Zeris, C. A. and Mahin, S. A. [1988] ‘‘Analysis of reinforced concrete beam-columns under
uniaxial excitation,’’ Journal of Structural Engineering 114(4), 804–820.