Computers and Structures 84 (2006) 904–916

www.elsevier.com/locate/compstruc

A fibre model for push-over analysis of underdesigned

reinforced concrete frames
Edoardo Cosenza, Gaetano Manfredi *, Gerardo M. Verderame
Department of Structural Analysis and Design, University of Naples Federico II, Via Claudio, 21, 80125 Napoli, Italy

Received 10 January 2005; accepted 1 February 2006

Abstract

Most of the existing reinforced concrete buildings were designed according to early seismic provisions or, sometimes, without applying
any seismic provision. Some problems of strength and ductility, like insufficient shear strength, pull-out of rebars, local mechanisms, etc.,
could characterize their structural behaviour. The above mentioned topics lead to a number of problems in the evaluation of the seismic
behaviour of reinforced concrete (RC) frames. Therefore the assessment of existing RC structures requires advanced tools. A refined
model and numerical procedure for the non-linear analysis of reinforced concrete frames is presented. The current version of the model
proposed is capable of describing the non-linear behaviour of underdesigned reinforced concrete frames including brittle modes of fail-
ure. Selected results of an experimental–theoretical comparison are presented to show the capabilities of this model. The results show the
capacity of the model of describing both the global behaviour and the local deformation at service and ultimate state.
 2006 Elsevier Ltd. All rights reserved.

Keywords: Non-linear; RC frames; Beam; Column; Bond; Hook

1. Modelling of reinforced concrete frames
Many models for the non-linear analysis of RC frames
are proposed in the literature. They can be classified
depending on the level of discretization [1] in point by point
model, member by member models and global models. The
choice of the most suitable model depends on the goals of
the analysis and by the structural properties. Structures
characterized by brittle mechanisms require a non-linear
analysis and then the use of highly discretized models; on
the other hand, for structures with flexural collapse mech-
anisms, member by member or global models can be used to
obtain reliable predictions. A good balance between com-
putational effort and level of reliability of the results should
be achieved in choosing the model by taking into account
the amount of basic information that each model requires.
In the last few years, fibre models have become more and
*
Corresponding author. Tel./fax: +39 081 7683491.
E-mail address: gamanfre@unina.it (G. Manfredi).
more popular. They still keep the basic hypothesis of sub-
dividing the structure in mono-dimensional elements, even
though they could be defined as a hybrid between point by
point and member by member models. The constitutive
laws of concrete and steel are introduced [2]; in recent ver-
sions the hypothesis of perfect bond is removed [3] and
shear collapse is taken into account [4]. Some authors
introduced a joint element accounts for inelastic shear
deformation and bar bond slip in program DRAIN-2DX
[5,6].
Manfredi and Pecce [7] proposed for beams and col-
umns a fibre element that introduces explicitly the bond
law s–s. Limkatanyu and Spacone [8] showed that the
accurate representation of the bond–slip behaviour is cru-
cial in predicting the response of RC frames subjected to
both static and dynamic loadings.
In the assessment of the seismic capacity of existing
underdesigned RC structures all the brittle failure modes
are potentially active, and this occurrence requires the
development of a reliable numerical model in terms of
behaviour and material properties [9]. In this paper a

0045-7949/$ - see front matter  2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compstruc.2006.02.003
 E. Cosenza et al. / Computers and Structures 84 (2006) 904–916
flexural – shear
T interaction
μ
priestley
bending moment
with axial force
Mi
N interior
Mi+1
manfredi - pecce
strut model
infills
fardis-panagiotakos
Fig. 1. The main mechanisms influencing non-linear behaviour of reinforced concrete frames.
numerical model for reinforced concrete frames is pre-
sented, which is an extension of the model proposed in
Manfredi and Pecce [7]. The model is able to predict the
main mechanisms influencing the non-linear behaviour of
reinforced concrete frames (Fig. 1). In particular, the pro-
posed model considers an explicit introduction of advanced
constitutive bond–slip relationships that allow to describ-
ing the structural behaviour in the large post yielding field
for elements under bending and axial forces and to intro-
duce refined models for beam–column joints [13].
2. Element formulation
The beam–column element is characterized by a spread
of plasticity and distributed cracking: it belongs to the fibre
model family. The mechanical properties of the cross-sec-
tion are evaluated by considering the constitutive laws of
the materials.
The classic hypothesis of perfect bond between concrete
and steel is removed and a stress–slip bond constitutive law
is introduced [10]. Such an aspect allows for a more reliable
assessment of the tension stiffening effect, for both elastic
and plastic field, and avoids the approximations due to
the assumption of the plastic hinge length.
For the column, it is possible to consider the variation of
axial forces due to lateral loads, and the related effects in
terms of overall strength and deformation capacity. Also,
considering the axial deformation allows for a detailed sim-
ulation of the interactions between columns and infill walls.
In the beam–column joints, that plays a significant influ-
ence on the structural response, both in terms of strength
(i.e., shear failure of the panel or pull-out of the rebars)
and deformation (due to the cracking of the concrete and
905
panel zone
beam
fixed-end
rotation
joint hooked rebar pull-out
column
exterior
joint
anchorage
footing
zone
hooked rebar pull-out
slippage of rebars), the rotation at the beam–column inter-
face is computed taking into account either the bond
between concrete and steel or the constitutive law of the
hooks.
The influence of shear forces on the behaviour of the
beams was modelled by Priestley et al. [11]; such a model
is based on a reduction of the shear strength depending
on the local ductility, as expressed in terms of linear varia-
tion of the curvature. The model introduced here repre-
sents an improvement of Priestley’s since it enables the
sectional ductility at any step of the analysis to be directly
determined and then evaluate the shear strength of those
sections located in the plastic regions. Therefore, along
with predicting ductile (i.e., flexural) and brittle (i.e., shear)
failures, this method allows also failures characterized by
low ductility due to the bending–shear interaction to be
determined.
The model for infill walls is based on the shear model by
Fardis and Panagiotakos [12]. It takes into account the
strength reduction due to the cracking of the panels and
the post-strength degradation. It is based on four different
steps: initial shear behaviour of the uncracked panel,
behaviour of the cracked panel as equivalent strut, its insta-
bility after the maximum strength and final stage after com-
plete failure characterized by constant residual strength.
2.1. The flexural model
The column is considered as a mono-dimensional ele-
ment, by introducing a simplified deformation model for
the cross-section, as shown in Fig. 2. As mentioned before
the hypothesis of perfect bond between steel and concrete is
removed. Thus, calculations in the generic cross-section
906 E. Cosenza et al. / Computers and Structures 84 (2006) 904–916
b(y)
εc(x)
dc
yg
H h z where Ac, Act and As are the area of the concrete in com-
As
dt
c ds/dx εct(x)
y
εs(x)
Fig. 2. Deformation model for the cross-section.
comprised between two subsequent cracks are performed
considering a linear strain diagram (i.e., concrete in com-
pression and steel in tension).
Another significant hypothesis is based on pre-defining
the distance, Dl, between cracks (sub-element) which occur
in those sections where the cracking moment value is over-
come. However, this is not a basic hypothesis in the pro-
posed model since the sections where cracks open could
also be determined as the analysis proceeds by computing
where the tensile stress of the concrete reaches the limit ten-
sile strength. Using the strain method, the problem is gov-
erned by subsequent parameters: the maximum
compressive strain in the concrete ec(x, yg), the tensile strain
in the steel es(x), the tensile strain in the concrete ect(x) and
the slip between steel rebar and the surrounding concrete in
tension s(x).
The material constitutive laws, concrete in compression
rc = rc(ec), concrete in tension rct = rct(ect), steel in ten-
sion/compression rs = rs(es) and the steel/concrete bond–
slip relationship s = s(s, x), should be added to the above
parameters.
The set of equations is composed by the force equilib-
rium equation of the cross-section, the moment equilibrium
equation around its geometric axis, the force equilibrium
equation of the rebar and a compatibility equation between
steel rebar and tensile concrete. The problem is formally
solved by the subsequent system of equations:
• force equilibrium of the cross-section:
Z Z
rc ðx; yÞbðyÞ dy  rct ðxÞbðyÞ dy  As rs ðxÞ ¼ N ðxÞ
Ac Act
• moment equilibrium around the geometric axis of the
cross-section:
Z Z
rc ðx; yÞybðyÞdy þ rct ðxÞybðyÞdy þ As rs ðxÞðh  y g Þ ¼ MðxÞ
Ac Act
• translational equilibrium of the steel bar:
drs 4
 sðxÞ ¼ 0
dx U
• compatibility equation for the steel bar and the concrete
in tension:
ds
¼ es ðxÞ  ect ðxÞ ð4Þ
dx
pression, of concrete in tension, of the steel bars and U is
the bar diameter, respectively.
Considering the three functions ec(x), es(x), ect(x) as gov-
erning unknowns, the solution can be achieved by resolving
two non-linear algebraic equations and two differential
equations of the first order, linear and non-linear respec-
tively. Since strains have been adopted as unknowns, the
solution in terms of stress is unique in any case, even
though the constitutive laws show a descending branch.
At any integration interval the boundary conditions have
to be associated to the differential equations system. In par-
ticular for x = x* and for x = x* + Dl, that means in the
section corresponding to the crack formation, the stress
(or strain) is immediately obtainable. In these sections
rct = ect = 0.
Is then possible to separate the first two equations from
the others to obtain the value of ec(x) and es(x) in the end
section of the sub-element.
It is to be noted that Eqs. (1) and (2) are coupled to the
differential equations (3) and (4) by means of the two
parameters rct(x) and ect(x) describing the behaviour of
the concrete in tension. By neglecting the ect(x) term in
Eq. (4), usually smaller if compared to es(x) particularly
when the steel is yielded, the differential equations (3)
and (4) form a system in the es(x) and s(x) terms, separate
from the global equilibrium of the section. In such an
hypothesis it is then necessary to use Eqs. (1) and (2),
related to the global equilibrium of the section at the point
where x = x* and x = x* + Dl, in order to obtain bound-
ary conditions, while inside the element the problem
depends on the differential equations. That simplifies the
computational effort, and gives a solution in terms of
s(x) and es(x) which is very close to the effective value.
The numerical solution of the differential set of equations
is quite difficult due to the non-linear constitutive laws and
the fact that the bond relationship depends on the distance
of the section from the crack. The problem can be solved
only by following a numerical approach; in particular, a
discretization is carried out using the finite difference
ð1Þ method, by dividing the region included by two cracks
into (n  1) parts having lengths equal to Dx, as shown
in Fig. 3.
The mechanical properties of the cross-section are eval-
uated in a detailed way, using a fibre approach for material
properties and introducing the appropriate constitutive
ð2Þ laws. The latter aspect enables the model to account also
for specific issues such as the instability of steel rebars in
compression or confinement effects. The pre-cracking stage
(i.e., moment M < Mcr) is determined based on the hypoth-
ð3Þ
esis of linear elastic behaviour.
 E. Cosenza et al. / Computers and Structures 84 (2006) 904–916 907

M M+ΔM
N x x=x* x=x*+ l
x=x*+Δl N

T T+ΔT
y
εs,1 τi τi+1 εs,n
M1 dc,1 dc,n
Mn
i=1 i i+1 i=n
N N Δll
Δx
x
σs,1 σs,n εs,i εs,i+1

x=x* Δll x=x*+ l

x=x*+Δl εs,m

Fig. 3. Beam/column element and cracked sub-element subdivision.

A reliable evaluation of the average steel strain es,m D/m d/m

lim ¼ ð9Þ
within the sub-element is possible as DM m !0 DM m dM m
Z Dl
1
es;m ¼ es ðxÞ dx ð5Þ
Dl 0 2.2. The axial model
where es(x) is the distribution of steel strain along the entire
sub-element and Dl is the length of the sub-element. During the loading process, the non-linear behaviour of
Generally, for a portion subjected at the ends by the the concrete, and the cracking and the plastic deformations
moments M1 and Mn (see Fig. 3) a relationship average of the section result in a change in the axial deformability.
moment Mm  average curvature /m of the sub-element, The proposed fibre model allows the assessment of the
can be uniquely defined; in particular, the moment Mm actual axial behaviour of the member by defining both
can be defined, assuming its linear variation between two the axial deformability of the section (cracked and
ends, as the average of moments in cracked sections, while uncracked) and the sub-element defined by two consecutive
the average curvature /m can be obtained solving the sub- cracks.
element as In general, in the case of bending combined with axial
load, if the strain at the geometric axis level of the element
es;m
/m ¼ ð6Þ is assumed as a reference, and for symmetric cross-sections,
h  d c;m
the axial strain can be expressed as
where dc,m represents the average neutral axis depth, con-  
H
sidering its linear variation between the values taken at eo ¼ es  /  h  ð10Þ
the ends of the sub-element, es,m is the average steel strain 2
and h is the effective depth of the cross-section, where eo is the axial strain, es is the strain in steel, / repre-
respectively. sent the curvature, and h and H are the effective depth and
Defining M im and /im as the moment and the average geometric height of the cross-section, respectively.
iþ1
curvature at the ith step, while M iþ1
m and /m the corre- Considering two consecutive cracks, it is possible to
sponding quantities at the i + 1th step, the moment Mm compute the axial deformability of cracked sections. As
relative to an average curvature within /im and /iþ1 m can the analysis moves away from the cracked section, the ten-
be determined as their linear interpolation, that is sile stress is progressively transferred from the steel rebar
M iþ1 i to the surrounding concrete and then the steel strain
m  Mm
M m ¼ M im þ iþ1 i
ð/m  /im Þ ð7Þ decreases; this results also in a decreased curvature of the
/m  /m
generic section belonging to each sub-element.
and similarly the curvature /m relative to a given moment The remarks on the tension stiffening effect outlined in
Mm within two known values Mi and Mi+1, is determined the previous section allow to be observed the increased
as stiffness of the sub-element lowers the axial strain of the
/iþ1 i generic section compared to the axial strain in the cracked
m  /m
/m ¼ /im þ ðM m  M im Þ ð8Þ section.
M m  M im
iþ1
If, for the seek of simplicity, the stress of the tensile con-
In particular, the incremental ratio D/m/DMm in the crete, ect is neglected, an approach similar to the above
previous expression can be interpreted as the average flex- discussed allows to define the axial strain of the generic sec-
ural tangential deformability of the sub-element; actually, tion belonging to the sub-element described by two consec-
as DMm tends to 0, the ratio D/m/DMm becomes the deriv- utive cracks. The average axial strain of the sub-element
ative of the curve / = /(M), that is can be obtained by averaging the axial strains of different
908 E. Cosenza et al. / Computers and Structures 84 (2006) 904–916

sections of the sub-element. Supposing to adopt the ajj

K 33 ¼ ð17Þ
approximate approach, the axial strain for the generic sec- ðaii  ajj  a2ij Þ
tion i belonging to the sub-element can be defined as aii
K 66 ¼ ð18Þ
  ðaii  ajj  a2ij Þ
es;i H aij
eo;i ¼ es;i   h ð11Þ K 36 ¼ ð19Þ
h  d c;m 2 ðaii  ajj  a2ij Þ
where dc,m represents the average neutral axis depth.
Via equilibrium the following mixed terms can be deduced:
The sub-element average axial strain can be then
obtained averaging the axial strains of the different sec- ðajj  aij Þ
K 32 ¼ K 35 ¼ ðK 33 þ K 66 Þ=L ¼ ð20Þ
tions, as follows: ½ðaii  ajj  a2ij Þ  L
  ðaii  aij Þ
es;m H K 62 ¼ K 65 ¼ ðK 36 þ K 66 Þ=L ¼ ð21Þ
eo;m ¼ es;m   h ½ðaii  ajj  a2ij Þ  L
h  d c;m 2
 
H Finally, from equilibrium and compatibility, the following
¼ es;m  /m  h  ð12Þ translational stiffness are obtained:
2
K 22 ¼ K 55 ¼ K 25 ¼ ðK 23 þ K 26 Þ=L
Therefore, as already stated for the flexural model, if the
axial load and the axial strain at the ith step are defined as ðaii þ ajj  2aij Þ
¼ ð22Þ
Ni and eio;m , while Ni+1 and eiþ1
o;m are the axial load and the ½ðaii  ajj  a2ij Þ  L2 
axial strain at next step, the average axial tangential defor-
mability of the sub-element can be defined as To these stiffness coefficients, those relative to extensional
terms must be added. In a similar way, applying the prin-
iþ1
eo;m  eio;m ciple of virtual work, the extensional strain coefficients of
fmN ¼ ð13Þ
N iþ1  N i the element are calculated and the stiffness coefficients
afterwards.
2.3. Stiffness matrix element’s
3. Special elements and detailing formulation
In order to evaluate the stiffness matrix of the element,
the flexural and axial tangential deformability of the sub- 3.1. The fixed-end rotation model
element within two subsequent cracks have been defined;
in particular, average flexural and axial deformability (or The proposed joint model is governed by the slippage
stiffness) are the following: between the anchored rebar and concrete, while the stiff-
ness of the panel is assumed as infinite and, therefore, does
d/m dM m not affect the joint deformability. The slip of steel rebars at
fmM ¼ kM
m ¼ ð14Þ
dM m d/m the joint–element (i.e., beam or column) interface is evalu-
de o;m dN m ated by a procedure similar to that above described for the
fmN ¼ k Nm ¼ ð15Þ
dN m deo;m flexural model; a force equilibrium equation (3) for the
rebar and a compatibility equation (4) between steel rebar
Within the considered uncracked element intervals, the
and concrete in tension control the problem. The strain of
average flexural and axial stiffness are defined by the corre-
the element can be obtained if the boundary conditions in
sponding quantities evaluated assuming the concrete gross
terms of steel strains are provided in the two end sections of
section.
rebar.
Each term belonging to the stiffness matrix can be eval-
In the section at the interface between joint and beam
uated starting by the flexibility coefficients of the corre-
element, x = Lh the boundary condition is obtained using
sponding auxiliary beam. The generic flexibility
the equilibrium equations (1) and (2) and considering the
coefficient can be evaluated applying the principle of vir-
section s part of the element; in the section, x = 0
tual work; in particular, for the rotational coefficient it
the boundary conditions depends on the behaviour of the
results:
anchorage detailing (see Fig. 4). From a theoretical point
Z
M iM j of view, in this section, two limit boundary conditions
aij ¼ dx ð16Þ can be identified:
L kM
m

where L is the element’s length, Mi and Mj are the moments
diagrams evaluated on the auxiliary element with ith and
jth unitary action, while k M
m is the actual average flexural
tangential stiffness of the sub-element, as previously stated.
The compatibility conditions provide the following stiff-
ness matrix rotational terms:
• if the anchorage is not present, straight rebar is charac-
terized by a free end, thus a unrestricted slippage occurs,
so that es = 0;
• if anchorage is rigid, slip at the inner end of the rebar is
equal to zero s = 0, and a steel stress develops on the
anchoring device.
 E. Cosenza et al. / Computers and Structures 84 (2006) 904–916
Fig. 4. Fixed-end rotation model.
Commonly, anchorage devices have a stress–slip
response that lies between the above boundaries, so that
both slip and steel strain are not zero and are dependent
upon the response of the end anchorage. In particular,
when the rebar is terminated with an end hook, the steel
element can be treated as composed by the hook plus the
straight portion [13].
The steel–concrete interaction in the straight region is
described by the s–s bond law, while the hook can be mod-
elled as a translational non-linear spring whose behaviour
is governed by the stress–slip, rs,h–sh relationship [14] com-
puted in the common section between the hook and the
straight part, as depicted in Fig. 4.
Therefore, for the generic flexural moment in the inter-
face section, it is possible to calculate the end rotation h
related to slip sj as
sj
h¼ ð23Þ
h  dc
where dc represents the neutral axis depth, sj is the slip of
steel rebars and h is the effective depth of the cross-section,
respectively, in the joint–element interface section.
Once the M–h function is known, the joint deformability
can be computed. This can be done by considering the
beam–column intersection as a rotational non-linear spring
whose behaviour is represented by the M–h relationship.
The model proposed for the joint can also be used for the
interfacial section between columns and footing; in this case,
the longitudinal rebars of the columns are terminated into
the footing similarly to exterior joints of a RC plane frame.
3.2. Shear–flexural interaction model
As regards the behaviour of bending elements carrying
high shear forces, refined models must be introduced to
evaluate the shear strength and take into account the ele-
ments decreasing flexural ductility due to shear interaction.
The shear strength of beam/column elements basically
depends on their ductility level. In plastic areas, decreasing
909
shear strength of the element is noted, due to a decreasing
or absent shear resistant mechanisms in the concrete.
The influence of shear forces on the behaviour of the
beams was modelled by Priestley et al. [11]; such model is
based on a reduction of the shear strength depending on
the local ductility, expressed in terms of curvature varying
with a linear trend. The introduced model represents an
improvement of Priestley’s because it enables the sectional
ductility at any step of the analysis to be directly deter-
mined and then the shear strength of those sections located
in the plastic regions to be evaluated. Therefore, along with
predicting ductile (i.e., flexural) and brittle (i.e., shear)
failures, this method allows to determine also failures char-
acterized by low ductility due to the bending–shear interac-
tion (see Fig. 5).
Usually, the shear capacity of a frame element is given
by three terms regarded as independent: a concrete compo-
nent Vc which is function of the section ductility level, a
compression strength component Vp and finally a Vs
component, whose extent is function of the transverse
reinforcement quantity. So
V ¼VcþVpþVs ð24Þ
The concrete resistance component Vc, valid both for circu-
lar and rectangular elements, decreases as ductility in-
creases, according to following equation:
Flexural Response
Shear Force, V
Flexural Response
Shear Capacity
Flexural Response
μφ=1
Curvature ductility,, μφ
Fig. 5. Bending–shear interaction: failures mode.
910 E. Cosenza et al. / Computers and Structures 84 (2006) 904–916
pffiffiffiffi
Vc ¼c fc0 Ae ð25Þ
where c depends on section curvature ductility level, l/,
following a linear relationship, while Ae is the actual shear
strength area considered as 80% of the section’s geometric
area only.
It is assumed that the presence of the axial load, N deter-
mines a shear strength mechanism due to the formation of
a sloped compressed strut. Therefore, the shear strength Vp
is given by the horizontal component of the diagonal com-
pression strength as
H  dc
Vp ¼N  ð26Þ
2L
where H is the geometric section height, L is the length of
column from the critical section to the point of contraflex-
ure, and dc is the neutral axis depth. Vp does not decrease
with increasing ductility.
The transverse reinforcement, contribution, Asw, on
shear strength is based on a mechanism that does not
decrease with increasing ductility. The shear strength is
then given by
Asw fyh h0
Vs ¼ cot b ð27Þ
s The models of different elements, constituting the beam
0
where h is the distance between the outside edge of the stir-
rups, s is the stirrups spacing and b is the slope of the com-
pressed diagonal strut with respect to the centroid of the
element.
3.3. Infill model
Infill walls can represent in some cases a crucial source
of strength for the structure. Their presence and high stiff-
ness can play an important role on the global response of
the building and on the local performance of RC elements.
The masonry panels are implemented introducing the
model proposed by Fardis and Panagiotakos [12] that
correlate the shear force V on the infill wall, with the hori-
zontal displacement D (see Fig. 6). In the resultant diagram
four different segments can be analyzed which represent
respectively: (a) the initial shear behaviour of the
uncracked panel; (b) the equivalent strut behaviour of the
cracked panel; (c) the instability of the panel over its max-
V
Δ
V Vmax
Vcr
hw
K1
lw
Fig. 6. Infill model: shear force versus horizontal displacement.
imum strength; (d) the final stage of the panel when the
failure is achieved and the residual strength remains con-
stant. The main parameters of the model are
• initial stiffness of the uncracked wall, K1, obtained on
average as Gwtwlw/hw, where Gw is the shear modulus
obtained by the diagonal compressive test;
• secant stiffness, K2, equal to the equivalent strut stiffness
computed with the elastic modulus, Ew, equal to the
modulus of the panel in diagonal direction and with
the strut cross-section dimensions according to Main-
stone [15];
• cracking load, Vcr that can be computed as the product
of the shear strength, fws (obtained by the diagonal com-
pressive test) with the plan dimensions of the panel,
fwstwlw;
• maximum load, Vmax equal to 1.3Vcr.
The model is implemented by introducing a diagonal
compressive strut.
4. Global stiffness matrix
element and the column element, have been previously dis-
cussed one by one; in particular, the flexural and nodal ele-
ments are serially jointed to form the beam or column
element. In this viewpoint, a flexibility matrix of the
beam/column element can be defined as the sum of the
single sub-elements flexibility matrix, i.e.
Felem ¼ Fflex þ Fjnt ð28Þ
where Fflex is the flexibility matrix of the flexural element,
while Fjnt is the flexibility matrix of the nodal element.
It is important to point out that the above stated matrix
varies within every loading step, for the non-linearities
associated with the moment-curvature, or the moment-
rotation, and involves some variations in elements charac-
teristics. Therefore, the matrix represents the tangential
flexibility matrix of the beam/column element.
According to the process shown so far, the stiffness
matrix of each beam or column element can be deduced;
in particular, for what it may concern columns, the geomet-
K2
K3=(0.5%-10%)K1
Vres =(5%-10%)Vmax
 E. Cosenza et al. / Computers and Structures 84 (2006) 904–916
ric stiffness matrix is considered as well. The whole struc-
ture stiffness matrix can be formally drawn with respect
to the stiffness method as follows:
X the element to obtain the residual element deformations
Kstruct ¼ ðKbeam þ Kcol Þ ð29Þ
elem
where Kstruct is the whole structure matrix, Kbeam represents
the beam element matrix while Kcol represents the column
element matrix.
5. Solution algorithm
In a finite element that is based on the stiffness method
of analysis the section deformations are obtained directly
from the element’s end deformations by the deformation
interpolation function; the corresponding section resisting
forces are determined from the section force–deformation
relationship. The integral of the section resisting forces
over the element length yields the element resisting forces.
In a finite element that is based on the flexibility method,
the first step is the determination of the element forces from
the current element deformation using the stiffness matrix.
Thus, the force interpolation function yields the section
forces. The first problem is the determination of the section
deformation from the given section force, since the non-lin-
ear section force–deformation is commonly expressed as an
explicit function of section deformation; another problem
arises from the fact that changes in the section stiffness pro-
duce a new element stiffness matrix which change the ele-
ment forces for the given deformation. These problems
are solved in Taucer et al. [16] by a special non-linear solu-
tion method.
At the ith Newton–Raphson iteration it is necessary to
determine the element resisting forces for the current ele-
ment deformation:
qi ¼ qi1 þ Dqi ð30Þ
To this end an iterative process denoted by j is introduced
inside the Newton–Raphson iteration. With the initial
element tangent stiffness matrix Kj=0 = Ki1 and the given
element deformation increments Dqi the corresponding ele-
ment force increments are
DQj¼1 ¼ Kj¼0
el Dq
j¼1
ð31Þ
j=1
The section force increments DD can be determined
from the force interpolation function. With the section flex-
ibility matrix f j=0 = fi1 the linearization of the section
force–deformation relationship (moment-curvature) yields
the section increments:
Ddj¼1 ¼ f j¼0 DDj¼1 ð32Þ
The section deformations are updated d = d + Ddj.
j j1
According to the section force–deformation relationship
section deformations dj correspond to resisting forces
Dj¼1
R and a new tangent flexibility matrix f j=1 .
The section unbalanced forces Dj¼1 u ¼ Dj¼1  Dj¼1
R are
determined and they are then transformed to residual
section deformations rj¼1 ¼ f j¼1 Dj¼1
u .
911
The residual section deformations are the errors made in
the linearization of the section force–deformation relation-
ship. The residual section deformation are integrated along
sj=1. The residual section deformation and the residual ele-
ment deformation are determined but the corresponding
deformation vectors are not updated. The presence of
residual element deformation violates the compatibility of
the element. In order to restore the element compatibility
corrective forces DQj¼2 ¼ Kj¼1 el s
j¼1
must be applied at
j¼1
the ends of the element, where Kel is the updated element
tangent stiffness matrix. A corresponding section force
increment DDj¼2 ¼ bKj¼1 el s
j¼1
is determined inducing sec-
tion deformation increment—f j¼1 bKj¼1 el s
j¼1
.
Thus, in the next iteration j the state of the element
change as Qj=2 = Qj=1 + DQj=2 and the section forces
Dj=2 = Dj=1 + DDj=2 and deformations dj=2 = dj=1 + Ddj=2
are updated, where Ddj¼2 ¼ rj¼1  f j¼1 bKj¼1 elem s
j¼1
.
Convergence is achieved when the selected convergence
criterion is satisfied. The presented non-linear analysis
method offers several advantages. Equilibrium along the
element is always strictly satisfied, since the section forces
are derived from the element force by the force interpola-
tion functions. While equilibrium and compatibility are
satisfied along the element the section force–deformation
is only satisfied within a specified tolerance when the con-
vergence is achieved. When all elements have converged
the ith Newton–Raphson iteration is complete. The ele-
ment force vectors are assembled to form the updated
structure resisting forces. The new structure stiffness matrix
is formed by assembling the current element stiffness
matrix. The structure resisting forces are compared with
the total applied load. If the differences is not within the
specified tolerance, a new Newton–Raphson iteration
begins.
6. A numerical–experimental comparison
In order to give an example of the capabilities of the
proposed model a numerical–experimental comparison is
presented; the experimental reference is a full-scale RC
building, designed according to criteria and construction
methods used in the last 40 years [17] in a large part of
South Europe and tested in the laboratory of the Joint
Research Center of Ispra.
In particular, pseudo-dynamic tests on two full-scale RC
frames (having the same structure, materials and geometry
but one with infill walls and one without) have been per-
formed (Fig. 7). Numerical–experimental comparisons on
both of them allowed assessing the feasibility of the pro-
posed model with reference to the non-linear behaviour
of RC frames and to the interaction problems between
the RC elements and the masonry panels.
The frame was designed for gravity loads and lateral
forces equal to about 8% of its mass. The structure has four
floors, typical floor height of 2.70 m and 3 spans; two of
them are 5.00 m long, while the third has a length equal
912 E. Cosenza et al. / Computers and Structures 84 (2006) 904–916
Fig. 7. Geometry of the Ispra frame: elevation view.
to 2.50 m. The beam layout is the same at each floor, with
equal cross-section and reinforcement, with a slab thick-
ness of 15 cm. The columns have rectangular cross-sec-
tions, constant dimensions along the height and placed
along their weak axis.
The infill frame has wall panels in each span (Fig. 7). In
particular, on each span the openings are located as
follows:
• left span: windows openings with dimensions of
1.20 · 1.10 m at each of the four levels;
• central span: door opening with dimension of
2.00 · 1.90 m and windows openings at the other levels
with dimensions of 2.00 · 1.00 m;
• right span: no openings.
In order to verify the potentialities of the proposed
model the experimental pseudo-dynamic tests are simu-
lated assuming in approximate way that the push-over
analysis results are similar to the envelope of the cyclic test.
The push-over analysis has been developed using the pro-
posed numerical model and applying a pre-defined distri-
bution of horizontal forces; in particular, the distribution
has been selected in order to model the first linear mode
of vibration.
The gravity loads, applied on the frames during the
tests, have been defined simulating dead and live load. In
Fig. 8 load values [18] are reported in order to perform
the frame analysis; the uniform loads on the beams and
the point loads on the top of the columns represent a
scheme of the loads really applied on the tested frame.
6.1. The bare frame
The comparison with experimental results shows that
the use of a bond slip model significantly improves the
prediction capacity over the use of rigid bond model, but
several analyses [9] highlight that much of this difference
derives from bond–slip in the foundations and in the beam
column joints, respect to beam column elements, as demon-
strated also by others authors [19].
In this direction a comparison between the results
obtained by applying the proposed model and the experi-
mental curves is proposed in Fig. 9, which depicts the story
shear versus the interstorey drift. Two different numerical
analyses have been carried out, graphically shown using
two different color curves: the grey curve represents the
numerical result obtained through a rigid joint hypothesis
not including the fixed-end rotation effect, while the black
one is obtained through a strained joint hypothesis includ-
ing the fixed-end rotation effect.
If the joint is assumed to be stiff, the prediction of both
strength and tangent stiffness overestimate the experimen-
tal results, while when the deformability of the joint is con-
sidered allows a very satisfactory agreement with test is
achieved. The graph shows the numerical outcomes for a
concrete strain in compression up to ec = 0.5%.
This result highlights the large influence of the fixed-end
rotation on the local and global behaviour of the structure.
In particular, for low values of the story shear the two
curves are overlapped; in fact, the low flexural moment at
the end of beam–column elements supplies, in the hypoth-
esis of deformable joint (black curve), a low slip demand in
 E. Cosenza et al. / Computers and Structures 84 (2006) 904–916 913

12.7 kN/m
44.3 kN 76.1 kN 60.2 kN 28.4 kN

2.70 m
15.1 kN/m
56.4 kN 88.2 kN 72.3 kN 40.5 kN

2.70 m

15.1 kN/m
56.4 kN 88.2 kN 72.3 kN 40.5 kN
2.70 m

15.1 kN/m
56.4 kN 88.2 kN 72.3 kN 40.5 kN
2.70 m

5.00 m 5.00 m 2.50 m

Fig. 8. Scheme of the frame gravity loads.

Storey 1 300 Shear, V [kN] Storey 2 300 Shear, V [kN]

200 200

100 100
drift [mm] drift [mm]
0 0
-40 -30 -20 -10 0 10 20 30 40 -40 -30 -20 -10 0 10 20 30 40
-100 -100

-200 -200
with fixed end rotation with fixed end rotation
without fixed end rotation -300 without fixed end rotation
-300

Storey 3 200 Shear, V [kN] Storey 4 150 Shear, V [kN]

150
100
100
50
50
drift [mm] drift [mm]
0 0
-80 -60 -40 -20 0 20 40 60 80 -40 -30 -20 -10 0 10 20 30 40
-50
-50
-100
-100
-150 with fixed end rotation
with fixed end rotation
without fixed end rotation
-200 without fixed end rotation -150

Fig. 9. Numerical simulation of bare frame: effect of the fixed-end rotation.

the tensile reinforcement and therefore a low joint defor-
mation. As the story shear increases, the black curve shows
a lower stiffness, as consequence of an increasing joint
deformability, and it is in agreement with the experiments.
In all cases, the numerical analyses appear to agree with the
lab evidence.
6.2. The infill frame
The interaction between infill walls and RC members
can be seen to behave as something in between a shear
panel and a diagonal strut [20]. The boundary stress are
very different for those two cases: in the latter, the reaction
914 E. Cosenza et al. / Computers and Structures 84 (2006) 904–916

N N Linear model
SHear model M
V SH-L

Axial deformation

N
Non-Linear model
SH-NL
M
N Axial deformation

N N
Linear model
STrut model M
V ST-L

Axial deformation

N Non-Linear model
ST-NL
M
N Axial deformation

Fig. 10. Infill/RC column interaction models.

of the wall panel affects the axial loads of both columns
and beams; in the former, the axial loads do not change.
This means that the choice of an appropriate model repre-
sents a crucial step toward a reliable assessment of the
frame characterized by the interaction between bare struc-
ture and infill walls. In order to validate this opinion, a the-
oretical–experimental comparison was carried out on the
infill frame tested at Ispra. The RC elements of the frame
was studied using the flexural model discussed in Section
2.1 while the axial behaviour is modelled considering two
different hypotheses: using the model described in Section
2.2 (non-linear model), that allows evaluating the non-lin-
earities depending on section cracking and reinforcement
yielding, and using a simple linear model (linear model)
with a constant axial stiffness during the incremental
analysis. The masonry panels were analyzed by a model

Storey 1 800 Shear, V [kN] Storey 2 800 Shear, V [kN]

ST-NL
ST-NL 600 ST-L
600
ST-L
SH-NL SH-NL
400 SH-L 400
SH-L
200 200
drift [mm] drift [mm]
0 0
-4.0 -3.0 -2.0 -1.0 0.0 1.0 2.0 3.0 4.0 -5.0 -4.0 -3.0 -2.0 -1.0 0.0 1.0 2.0 3.0 4.0 5.0
-200 -200

-400 -400

-600 -600

-800 -800

Storey 3 600 Shear, V [kN] Storey 4 400 Shear, V [kN]

300
ST-NL 400 ST-NL
ST-L ST-L
SH-NL SH-NL
200
SH-L 200 SH-L
100
drift [mm] drift [mm]
0 0
-3.0 -2.0 -1.0 0.0 1.0 2.0 3.0 -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0
-100
-200
-200
-400
-300

-600 -400

Fig. 11. Numerical simulation of infilled frame: effect of models on frame drift.
 E. Cosenza et al. / Computers and Structures 84 (2006) 904–916 915

column 1 800 Shear, V [kN] column 4 800 Shear, V [kN]

ST-NL
600 600 ST-L
SH-NL
400 400 SH-L

200 200
axial displacement [mm] axial displacement [mm]
0 0
-1.00 -0.50 0.00 0.50 1.00 -1.00 -0.50 0.00 0.50 1.00
-200 -200

-400 -400
ST-NL
-600 ST-L -600
SH-NL
-800 SH-L -800

column 5 800 Shear, V [kN] column 8 800 Shear, V [kN]

ST-NL
600 600 ST-L
SH-NL
400 400 SH-L

200 200
axial displacement [mm] axial displacement [mm]
0 0
-1.00 -0.50 0.00 0.50 1.00 -1.00 -0.50 0.00 0.50 1.00
-200 -200

-400 -400
ST-NL
-600 ST-L -600
SH-NL
-800 SH-L -800

Fig. 12. Numerical simulation of infilled frame: effect of models on column response.

implementing that proposed by Fardis and Panagiotakos
[12]; in particular, both shear and strut models were used
for the panels. The basic hypotheses are summarized in
Fig. 10.
Fig. 11 shows a comparison between the model out-
comes and the experimental curves, in terms of story shear
versus the interstorey drift. It highlights how different
approaches provided similar global results, while the differ-
ence emerges at high levels, where the strut model gives lar-
ger values than the shear one in terms of deformability.
This is due to different column axial loads in the two mod-
els, which is more evident at high floors where low gravity
loads lower the axial stress.
All this is expressed in Fig. 12 where story shear versus
axial displacements are reported for the exterior columns
of the first two floors. It shows how the predictions of
the shear model do not fit the experimental evidence; the
adoption of a linear axial model rather than a non-linear
one does not make a great difference. On the other hand,
the strut model allow for a good agreement with experi-
mental records; the choice of a non-linear axial model
results in a excellent simulation which is able to reproduce
the non-linear behaviour of the columns due to the interac-
tion with infill walls and to the decrease of axial load under
horizontal actions.
7. Conclusions
The paper describes a fibre model proposed for the non-
linear analysis of RC frames. This model directly intro-
duces the s–s bond relationship; such aspect makes it very
useful in the analysis of existing structures where smooth
bars change the bond mechanism. The axial deformability
of elements influencing the interaction between frames and
infill walls is also taken into account. The joint model
allows considering the contribution of the fixed-end rota-
tion when straight and/or hook ends are present.
Overall the model allows for the assessment of the seis-
mic capacity of underdesigned RC structures, where
smooth bars, non-linearities for low levels of load, low
quality of constructive details and consequent potential
for brittle failure modes require sophisticated tools for
the numerical simulation.
The previous numerical–experimental comparisons show
the model’s potentialities and how it is able to highlight
strains and strengths of frames subjected to extensive
non-linear behaviour of elements. As regards the influence
of joint deformability on global behaviour of structure, the
model superbly simulates the strain increasing due to the
fixed-end rotation, as the ISPRA test points out.
The proposed model can be used in the push-over analy-
sis under seismic loading of reinforced concrete buildings
where the structure is subjected to an incremental static
analysis under a pre-defined pattern of horizontal forces.
References
[1] RC frames under earthquake loading—state of the art report. CEB
Bull 1996;231(May).
[2] Filippou FC, Issa A. Nonlinear analysis of reinforced concrete frames
under cyclic load reversal. Report EERC 88/12. 1988.
[3] Monti G, Spacone E. Reinforced concrete fiber beam element with
bond slip. J Struct Eng 2000;126(6):654–61.
[4] Petrangeli M, Pinto PE, Ciampi V. Fiber element for cyclic bending
and shear of RC structures. I : theory. J Eng Mech 1999;125:9.
916 E. Cosenza et al. / Computers and Structures 84 (2006) 904–916
[5] Ghobarah A, Biddah A. Dynamic analysis of reinforced concrete
frames including joint shear deformation. Eng Struct 1999;21(11):
971–87.
[6] Ziyaeifara M, Noguchib H. A refined model for beam elements and
beam–column joints. Comput Struct 2000;76(4):551–64.
[7] Manfredi G, Pecce M. A refined r.c. beam element including bond–
slip relationship for the analysis of continuous beams. Comput Struct
1998;69.
[8] Limkatanyu S, Spacone E. Reinforced concrete frame element with
bond interface. II: state determinations and numerical validation. J
Struct Eng 2002;128(3):356–64.
[9] Cosenza E, Manfredi G, Verderame GM. Seismic assessment of
gravity load designed r.c. frames: critical issues in structural model-
ing. J Earthquake Eng 2002;6(special issue 1).
[10] Eligehausen R, Popov EP, Bertero VV. Local bond-stress
relationships of deformed bars under generalized excitations. UCB/
EERC 83, 23, University of California, Berkeley, CA, USA.
1983.
[11] Priestley MJN, Verma R, Xiao Y. Seismic shear strength of
reinforced concrete columns. J Struct Eng ASCE 1994;120(8):
2310–29.
[12] Fardis MN, Panagiotakos TB. Seismic design and response of bare
and masonry-infilled reinforced concrete buildings Part II, infilled
structures. J Earthquake Eng 1997;1(3):475–503.
[13] Fabbrocino G, Verderame GM, Manfredi G, Cosenza E. Structural
models of critical regions in old-type r.c. frames with smooth rebars.
Eng Struct 2004;26(14):2137–48.
[14] Fabbrocino G, Verderame GM, Manfredi G. Experimental behav-
iour of anchored smooth bars in old type r.c. buildings. Eng Struct
2005;27(10):1575–85.
[15] Mainstone RJ. Supplementary note on the stiffness and strength of
infilled frames. Current paper CP13/74. Building Research Establish-
ment, London. 1974.
[16] Taucer F, Spacone E, Filippou FC. A fiber beam–column element for
seismic response analysis of reinforced concrete structures. Report
EERC 91/17. 1991.
[17] Carvalho EC. Personal communication to icons—topic participants.
1998.
[18] Pinto A, Varum H. Dynamic nonlinear analyses for the-4 storey RC
frame bare, infilled and retrofitted frames. Joint Research Centre-
Ispra, Safety in Structural Mechanics Unit Institute for Systems,
Informatics and Safety European Commission. 1999.
[19] Limkatanyu S, Spacone E. Effects of reinforcement slippage on the
non-linear response under cyclic loadings of RC frame structures.
Earthquake Eng Struct Dyn 2003;32(15):2407–24.
[20] Kappos AJ, Stylianidis KC, Michailidis CN. Analytical models for
brick masonry infilled r.c. frames under lateral loading. J Earthquake
Eng 1998;2(1):59–87.