Computers & Structures Vol. 65, No. I, pp.

l-16, 1997
0 1997 Elsevier Science Ltd. All rights reserved
Pergamun Printed in Great Britain
caz-7949/97 s17.00 + 0.00
PII: 300457!M9(%)002994

NONLINEAR FE ANALYSIS OF R/C STRUCTURES UNDER

MONOTONIC LOADS
H. G. Kwakt and F. C. Filippuu
tDepartment of Civil Engineering., Korea Advanced Institute of Science and Technology,
Kusung-dong, Yusong-gu, Taejon 805-701, Korea
fUniversity of California at Berkeley, 731 Davis Hall, Berkeley, CA 94720, U.S.A.

(Received 28 July 1996)

Abstract-This paper deals with the finite element analysis of the monotonic behavior of reinforced
concrete (R/C) beams and beamcolumn subassemblages. It is assumed that the behavior of these
members can be described by a plane stress field. Concrete and reinforcing bars are represented by separate
material models which are combined together with a model of the interaction between reinforcing bar
and concrete through bond-slip to describe the behavior of the composite reinforced concrete material.
Using the rotating crack model among the smeared crack model, the structural behavior is simulated and
a relation which can consider the tension stiffening effect in finite element analysis is proposed based on
an improved cracking criterion derived from fracture mechanics principles. A new reinforcing steel model
which is embedded inside a concrete element is developed to cope with the difficulty in modeling of
complex geometry. Correlation studies between analytical and experimental results show the validity
of the proposed models and identify the significance of various effects on the local and global response
of reinforced concrete members. 0 1997 Elsevier Science Ltd

INTRODUCTION concerned with the global response of structures.

The assessment of the strength and stiffness of
existing structures and newly designed critical
structures such as offshore platforms, long-span
bridges and nuclear power plants strongly requires
the development of advanced analytical methods
capable of representing the behavior of the struc-
ture under all possible loading conditions both
monotonic and cyclic, its time-dependent behavior
and, especially, its behavior under overloading.
However enormously a need is increased, the
development of s,uch models for reinforced and
prestressed concrete structures is complicated by
differences in short and long term behavior of the
constituent materials, concrete and reinforcing steel.
Moreover, reinforcing steel and concrete interact in a
complex way through bond-slip and aggregate
interlock.
Since the first introduction of the concept of a
“smeared” crack in the study of the axisymmetric
response of prestressed concrete reactor structures by
Rashid [I 1, the smeared crack approach of modeling
the cracking behavior of concrete is almost
exclusively used by investigators in the nonlinear
analysis of R/C structures because its implementation
in a finite element analysis program is more
straightforward than that of the discrete crack
model. Computer time considerations also favor
the smeared crack. model in analyses which are
With the use of the smeared crack model, however,
such problems as the analytical results of the response
of reinforced concrete structures are greatly influ-
enced by the size of the finite element mesh and by
the amount of tension stiffening of concrete are also
appeared on the accuracy of numerical analyses of
R/C structures [2]. In the context of the smeared
crack model two different representations have
emerged: the fixed crack and the rotating crack
model.
Differently from the fixed crack model, the rotating
crack model proposed by Cope et al. [3] assumes that
the crack direction is not fixed but kept perpendicular
to the direction of principal tensile strain during the
subsequent load history. Several tests by Vecchio and
Collins [4] have shown that the crack orientation
changes with loading history and the response of the
specimen depends on the current rather than the
original crack direction. Even though the rotating
crack model eliminates the need for a cracked shear
modulus, a disadvantage of this approach is the
difficulty of correlating the analytical results with
experimental fracture mechanics research which is at
odds with the rotating crack concept. This model has,
nonetheless, been successfully used in analytical
studies of R/C structures whose purpose is to study
the global structural behavior [5].
While the response of lightly reinforced beams in
bending is sensitive to the effect of tension stiffening
2 H. G. Kwak and F. C. Filippou

of concrete, the response of R/C structures in which from fracture mechanics considerations in order to
shear plays an important role such as over-reinforced reduce the numerical error associated with large size
beams and shear walls is much more affected by the finite elements. The validity of the proposed models
bond-slip of reinforcing steel than the tension is established by comparing the analytical predictions
stiffening of concrete. To account for the bond-slip of with results from experimental and previous analyti-
reinforcing steel two different approaches, that is, the cal studies.
bond-link element by Ngo and Scordelis [6] and the
bond-zone element by de Groot et al. [7] are common MATERIAL MODELS
in the finite element analysis of R/C structures.
Concrete
However, the use of these elements in the finite
element analysis of R/C structures imposes the Under combinations of biaxial loading concrete
restrictions such as the finite element mesh arrange- exhibits strength and stress-strain behavior which is
ment along the edge of a concrete element and a different from that under uniaxial loading conditions.
double node to represent the relative stip. In a Figure 1 shows the biaxial strength envelope of
complex structure, particularly in three-dimensional concrete under proportional loading. Under a
models, these requirements lead to a considerable combination of tension and compression the com-
increase in the number of degrees of freedom and pressive strength decreases almost linearly with
have discouraged researchers from including the increasing principal tensile stress [8].
bond-slip effect in many studies to date. To simulate the stress state of concrete under
Recognizing that many of the previously proposed biaxial loading the orthotropic model is adopted in
models and methods have not been fully verified so this study for its simplicity and computational
far, it is the intent of this paper to address some of efficiency. The behavior of the model depends on the
the model selection issues, in particular, with regard location of the present stress state in the principal
to the effects of tension-stiffening and bond-slip. In stress space. In the biaxial compression region the
this paper, a new discrete, embedded steel model is model remains linear elastic for stress combinations
developed for more efficient modeling of complex inside the initial yield surface. Both the initial yield
structures. The behavior of concrete under biaxial and the ultimate load surface are described by
loading conditions is described by a nonlinear Kupfer’s model (Fig. 1). For stress combinations
orthotropic model in which the axes of orthotropy outside the initial yield surface but inside the ultimate
coincide with the principal strain directions (rotating failure surface the behavior of concrete is described
crack model). The effect of size of the finite element by a nonlinear orthotropic model. This model derives
mesh is discussed in connection with a new smeared the biaxial stress-strain response from equivalent
crack model and an improved criterion is derived uniaxial stress-strain relations in the axes of

/’
/’
1’
/ /’
LINEAR ELASTIC

INITIAL YIELD SURFACE

/’
/’
/’ ORTHOTROPIC /4
ULTIMATE LOAD ,/’ NONLINEAR ELASTIC
SURFACE ,/”

1’
Fig. 1. Strength failure envelope of concrete.
 Nonlinear FE analysis of R/C structures
(lip
0.85Dip
0.6Dip
f
eq
Fig. 2. Stress-strain relation of concrete.
orthotropy based! on the concrete stress-strain
relation proposed in this paper [9-l 11.With reference
to the principal axes of orthotropy the incremental
constitutive relationship can be expressed
v$i%z 0
(1 -t&G
where Eland Ezare the secant moduli of elasticity in
the direction of the axes of orthotropy which are
oriented perpendicular and parallel to the crack
direction, v is Poisson’s ratio and (1 - v*).G =
0.25.(E, + Ez- 2va).
In describing the uniaxial stress-strain behavior of
concrete, the model of Hognestad [IO] is used after
some modifications in this paper (Fig. 2). These
modifications are introduced in order to increase the
computational efficiency of the model and in view of
the fact that the response of typical reinforced
concrete structure is much more affected by the
tensile than by the compressive behavior of concrete.
This stems from the fact that the concrete tensile
strength is generally less than 20% of the compressive
strength. In typical reinforced concrete beams and
slabs which are subjected to bending, the maximum
compressive stress at failure does not reach a small
fraction of the compressive strength at failure. The
behavior of these members is, therefore, dominated
by crack formation and propagation and the yielding
crushing
-&
&ip &iu
of reinforcing steel. The equivalent concrete compres-
sive strength in each axis of orthotropy ai+, is
determined from the biaxial failure surface of
concrete where i is equal to 1 or 2 (Fig. 1). In order
to simplify the concrete material model the
stress-strain relation in compression is assumed
piecewise linear with three branches.
When the biaxial stresses exceed Kupfer’s failure
envelope, concrete enters into the strain softening
range of behavior where an orthotropic model
describes the biaxial behavior [l 11. In this region
failure occurs by crushing of concrete when the
principal compressive strain exceeds a limit value c,,,.
In defining the crushing of concrete under biaxial
compressive strains a strain failure surface in
complete analogy to Kupfer’s stress failure envelope
is used. In the biaxial compression-tension and
tension-tension region the following assumptions are
adopted in this study: (1) failure takes place by
cracking and, therefore, the tensile behavior of
concrete dominates the response; (2) the uniaxial
tensile strength of concrete is reduced to the
equivalent tensile strength &, as shown in Fig. 2 to
account for the effect of the compressive stress; in the
tension-tension region the tensile strength remains
equal to the uniaxial tensile strength; (3) the concrete
stress-strain relation in compression is the same as
under uniaxial loading and does not change with
increasing principal tensile stress. The last assump-
tion holds true in the compressive stress range which
is of practical interest in typical reinforced frame
structures.
The use of the orthotropic constitutive relation in
eqn (I) to represent cracked concrete may not be
totally realistic. In the case of a real crack the crack
4 H. G. Kwak and F. C. Filippou

surface is rough and any sliding parallel to the crack tensile strain. Ultimate failure in the compression-
will generate some local stresses or movement normal tension and the tension-tension region takes place by
to the crack. To properly represent this type of cracking when the principal tensile strain exceeds the
behavior the off-diagonal terms of the material value to. When the principal tensile strain exceeds 60,
matrix which relate shear strain with normal stress the material only loses its tensile strength normal to
should not be zero. The relative magnitude of these the crack while it is assumed to retain its strength
off-diagonal terms decreases as the crack widens. parallel to the crack direction.
However, this effect may not be significant in a study
Tension stlfening model
which focuses attention on overall member behavior
and most researchers have neglected it [ 121. The smeared crack model first used by Rashid [I]
The proposed concrete model accounts for represents cracked concrete as an elastic orthotropic
progressive cracking and changes in the crack material with reduced elastic modulus in the direction
direction by assuming that the crack is always normal normal to the crack plane. In contrast to the discrete
to the principal strain direction (the rotating crack crack concept, the smeared crack concept fits the
model). In contrast to the model used by Hand nature of the finite element displacement method
ef al. [I 31 and Lin and Scordelis [ 121, the material since the continuity of the displacement field remains
axes are not fixed after formation of the initial crack, intact. Although this approach is simple to implement
but their orientation is determined from the direction and is, therefore, widely used, it has nevertheless a
of principal strains at the beginning of each iteration. major drawback which is the dependency of the
In developing a numerical algorithm for the results on the size of the finite element mesh used in
rotating crack concept, Gupta and Akbar [14] the analysis [2]. When large finite elements are used,
obtained the rotating crack material matrix as the each element has a large effect on the structural
sum of the conventional fixed crack material matrix stiffness. When a single element cracks, the stiffness
in eqn (I) and a contribution which reflects the of the entire structure is greatly reduced. Higher-
change in crack direction. This is expressed by the order elements in which the material behavior is
following eqn (2). established at a number of integration points do not
markedly change this situation because, in most
[Dmlc*= Pmlc + [Cl (2) cases, when a crack takes place at one integration
point, the element stiffness is reduced enough so that
where [&,I, is given by eqn (1) and [G] reflects the a crack will occur at all other integration points of the
change in crack direction and is given by element in the next iteration. Thus, a crack at an
integration point does not relieve the rest of the
material in the element since the imposed strain
[G] = e’
continuity increases the strains at all other integration
2JX
points. The overall effect is that the difficulty stems

1
from the fact that a crack represents a strain
sit? 28 - sin’ 28 - sin 28 cos 28
discontinuity which cannot be modeled correctly
x sin* 28 sin 28 cos 28 (3) within a single finite element in which the strain varies
~0~2 28
[ sym continuously. Many research efforts have been
devoted to the solution of this problem based, in
where 0 is the angle between the direction normal to particular, on fracture mechanics concepts [ 16, 171.
the crack and the global x-axis. While eqn (3) is Two widely used models of the strain softening
theoretically correct within the assumptions of the behavior of concrete in tension are those of Bazant
rotating crack model, any suitable incremental and Oh [ 161 and Hillerborg et al. [ 171. Both models
material stiffness matrix can be used in the context of have been extensively used in the analysis of RC
an iterative nonlinear solution algorithm. It is, members and yield very satisfactory results when the
therefore, possible to neglect the rotating crack size of the finite element mesh is relatively small. The
contribution [G] provided that the change in crack analytical results, however, differ significantly from
direction is accounted for in the orientation of the the experimental data when the finite element mesh
material axes and in the transformation from size becomes very large. This happens because both
material to element coordinate axes. Milford and models assume a uniform distribution of microcracks
Schnobrich [ 151 found that neglecting the rotating over a significant portion of a relatively*large finite
crack contribution [G] in eqn (3) only rarely increased element while the actual microcracks are concen-
the number of iterations and did not introduce any trated in a much smaller cracked region of the
numerical instabilities. element.
The proposed model assumes that concrete is linear In order to account for the fact that microcracks
elastic in the compression-tension and the biaxial are concentrated in a fracture process zone which
tension region for tensile stress smaller than f,. may be small compared to the size of the finite
Beyond the equivalent tensile strength the tensile element mesh a distribution function for the
stress decreases linearly with increasing principal microcracks across the element width is introduced in
 Nonlinear FE analysis of R/C structures 5

:
h=l

Fig. 3. Assumed distribution of microcracks in an element.

this study [9]. The Idistribution function is exponential maximum aggregate size [ 161, the proposed distri-
so that it can represent the concentration of bution function reduces to Bazant’s model of the
microcracks near the crack tip when the finite element crack band theory.
mesh size becomes fairly large (Fig. 3). The fracture energy Gr is defined as the product of
the area under the equivalent uniaxial stress-strain
f(x) = u..esv (4) curve gr and the fracture zone. It can, therefore, be
expressed as:
in which u and /I are constants to be determined.
Using the boundary conditions that f(0) = 1.O and
f(b/2) = 3/b into eqn (4) yields the following equation Gr = b.g, = fti,lf;.2. h2f(x)dx (6)
for the distribution function

,r(x) = e-2,h InM3 v

where J; is the tensile strength of concrete, 6” is the
(5)

fracture tensile strain which characterizes the end of

where b is the element width. The condition that the strain softening process when the microcracks
f(b/2) = 3/b ensures that, when the finite element coalesce into a continuous crack and Gr is the fracture
mesh size is eqaal to 3 in, i.e. three times the energy which is dissipated in the formation of a crack

=Y

Fig. 4. Stress-strain relation of steel.

 H. G. Kwak and F. C. Filippou

Fig. 5. Steel element embedded in concrete element.

of unit length per unit thickness and is considered a Reinforcing steel

material property. The experimental study by Welch
and Haismen [181 indicates that for normal strength
concrete the value of Gr/’ is in the range of
0.005~.01 mm. If Gr and 5 are known from
measurements, L,,can be determined after substituting
the function,f(?r) from eqn (5) into eqn (6).
2.Gf.ln(3/b)
La= 1;.(3 - b)
which clearly shows that co depends on the finite
element mesh size. This approach of defining co
renders the analytical solution insensitive to the mesh
size and guarantees the objectivity of the results. At
the same time this approach allows for the realistic
representation of the microcrack concentration near
the tip of the crack in the case of large finite elements.
With this approach large finite elements can be used
in the modeling of R/C structures without loss of
accuracy.
The reinforcing steel is modeled as a linear elastic,
linear strain hardening material with yield stress or as
shown in Fig. 4. The reasons for this approximation
are: (1) the computational convenience of the model;
(2) the behavior of R/C member is greatly affected by
the yielding of reinforcing steel when the structure is
subjected to monotonic bending moments [9].
A new discrete reinforcing steel model is used in the
(7)
analysis of plane stress problems in the modeling of
reinforcing steel. In the proposed discrete model the
reinforcing steel is represented by a one-dimensional
truss element which is embedded in the concrete
element as shown in Fig. 5. The nodes of the steel
element do not need to coincide with the nodes of the
concrete element. In this study the end displacements
of the steel element are assumed to be compatible
with the boundary displacements of the concrete
element so that perfect bond is implied.
Even though the end displacements of the steel
element are compatible with the concrete displace-

b
C

- Nj
steel
j
1 nodes
--w lNk

k L ’ ‘.
-1 0 1
(4 00
Fig. 6. Shape function of isoparametric concrete and embedded steel element.
 Nonlinear FE analysis of R/C structures 7

Fig. 7. Bond-slip relation for plane stress problems.

merits, displacement compatibility within the con- relative to the concrete element boundary as shown
crete element is, generally, not satisfied when the two in Figs 5 and 6. By setting o/l, = r,, the shape
node steel element is embedded in the eight-node functions can be simplified to the following relations
serendipity concrek element. This fact, however, does of Ni = (ri - 1).(2r, - l), N, = 4r,.(l - );) and
not seem to affect the accuracy of results of the global Nk = ri.(2ri - I), respectively, which are used to
behavior of R/C structures. construct the transformation matrix [7’&
Since the end points of the reinforcing bar
element do not generally coincide with the nodes of P,, = N,(&).P, (10)
the concrete element in Fig. 5, the one-dimensional
truss element with constant strain has to undergo P,i = N,(a;).P,.. (11)
transformation before it can be assembled
together with the concrete element stiffness matrix.
If the reinforcing bar element crosses the concrete
This can be formally expressed by the following
element boundary on sides 2 and 4 in Fig. 5,
relation:
nodes i, j and k correspond to node numbers 1, 8
and 7 on side 4 and node numbers 3, 4 and 5 on
[KGL],= [~rzlT~[~,lT~[~L~ll~~~~l~~~~l
(81
side 2, respectively. With the notation of Fig. 5
the transformation matrix [Z] has the following
where
form:

1
COll 0 sin f3
]T’1 = [ 0 0 ~0”s0 sin06 (9) 0
IT’]= AI
0
0 0 B2
0 B,
0 B, 00 A,
0
A2
0
1 (12)

Transformation matrix [T2] can be derived with the

procedure used to establish the consistent nodal where
forces of the finite element method. When the

1
eight-node isoparametric element is used in the
-
two-dimensional mesh representation of the member,
A ’ = [ 2p’ 03p + 1 2p2 - 03p + 1
the shape functions for nodes i,j and k in Fig. 6 are
N, = l/25*(5 - I), N, = (1 - <),(I + r) and Nk =

1
l/2( .( < + I), respectively.
- 0
If a concentrated load acts at a point 5 = S,, the
AI = 0 + 4p
4p= - 4pz + 4p
consistent nodal forces can be expressed as eqns (10)
and (11). To obtain the shape function Ni(6,) in

1
eqns (10) and (1 I), the relationship of 6, = (2c,/l, - I)
is substituted for < in the shape functions N,, N, and (13)
NAwhere ci characterizes the position of the steel node
8 H. G. Kwak and F. C. Filippou

q = cz/Iz and 0 is the 2 x 2 null

B = w - 3q+ ’ 0 p = Clll,,
matrix.
’ [ 0 2q* - 3q +I1 In eqn (12) the position of submatrices A and
- 4q' + 4q 0 B within the transformation matrix [TX] is related
B> = to the side of the concrete element which the
[ 0 - 4q2 + 4q 1
reinforcing bar element crosses. If the reinforcing bar

w 0- 0 element crosses the concrete element boundary on

B 3 = 4
w - 41 (14) sides 1 and 2 in Fig. 5, the transformation matrix [Tz]

Update and assemble the element Tangent

stiffness matrices in the total I stiflhess
stif&ess matrix of the structure

i
Assemble the incremental load vector
1
Solve equilibrium equations.
Add displacement incremente to the
current displacements

Determine the element strains for

the curent displacements
1
Use the material constitutive model
to determine the internal stresses
and the internal element nodal forces
I

Determine the difference between

external and internal nodal forces
I

1yes
I Print disolacements. forces and stresses 1

Fig. 8. Outline of solution algorithm.

 Nonlinear FE analysis of R/C structures 9

takes the form SOLUTION ALGORITHM

Every nonlinear analysis algorithm consists of four

1
A, Az A, 0 basic steps: the formation of the current stiffness
O O O O (15)
IT” = [ 0 0 B, B2 B, 0 0 0 matrix, the solution of the equilibrium equations
for the displacement increments, the state determi-
noting that the submatrices A and B are the same as nation of all elements in the model and the
before but cl is now defined as shown on the convergence check. These steps are presented in
right-hand-side of Fig. 5. The steel stiffness matrix some detail in the flow diagram of Fig. 8. Since
[I&], in eqn (8) can now be assembled together with the global stiffness matrix of the structure depends on
the concrete element stiffness matrix to form the total the displacement increments, the solution of equi-
stiffness of the structure. librium equations is typically accomplished with
To account for the bond-slip effect together with an iterative method through the convergence
the tension stiffening effect, the bond-link element by check. The nonlinear solution scheme selected in this
Ngo and Scordelis [6] is selected because the study uses the tangent stiffness matrix at the
bond-link element provides a reasonable compromise beginning of the load step in combination with a
between accuracy and computational efficiency where constant stiffness matrix during the subsequent
the overall structural behavior is of primary interest. correction phase, that is, the incremental-iterative
After constructing the element bond-link stiffness method.
matrix with only shear force along the reinforcing bar The criterion for measuring the convergence of
axis, it is transformed to global coordinates by a the iterative solution is based on the accuracy
rotation matrix. In this study the simple trilinear of satisfying the global equilibrium equations
bond stress-slip model in Fig. 7 is adopted to describe or on the accuracy of determining the total
the material behavior according to the loading displacements. The accuracy of satisfying the
history. This model is a good approximation of the global equilibrium equations is controlled by the
actual behavior in cases which do not exhibit magnitude of the unbalanced nodal forces. The
significant bond-slip and associated bond damage. accuracy of the node displacements depends on the
Under monotonic loading this holds true in all magnitude of the additional displacement increment
R/C members which do not experience anchorage after each iteration. The latter convergence criterion
failure. is used in this study. This can be expressed as

(4 0.5P
l--i-l
/ I f
18 in
T
&t 20 iu

1
1

0.5 P l NODES
@I
ELEMENTNO.
A

Fig. 9. Description of beam J-4: (a) configuration of beam A-l; (b) finite element idealization of beam
J-4. (I in = 2.54 cm.)
IO H. G. Kwak and F. C. Filippou

0 0.1 0.2 0.2 0.4 0

CENTER DEFLECTION (In) CENTER DEFLECTION (In)

(a) Hillerborg’s Model (b) Crack Band Model

/’

j
f -I
0 0.1 0.2 0.2 0.4 0. .I 0 0.1 0.2 0.2 0.4 0.5
CENTER DEFLECTION (In) CENTER DEFLECTION (In)

(c) Proposed Model (d) no Tension Stiffening

Fig. 10. Mesh size effect of beam J-4 (1 kip = 0.4536 ton, 1in = 2.54 cm.)

1
I,2 strength predictions. In order to avoid such problems
C (Ad;)* after crack initiation the load is increased in steps of

112
[ 2.5-5.0% of the ultimate load of the member.
’
CW*
E.,= < TOLER (161

i 1
[ for convergence. This means that very large strain
where the summation extends over all degrees of
freedomj, di is the displacement of degree of freedom
j, Ad: is the corresponding increment after iteration i
and TOLER is the specified tolerance.
In the nonlinear analysis of R/C structures the load
step size must be small enough so that unrealistic
“numerical cracking” does not take place. These
spurious cracks can artificially alter the load transfer
path within the structure and result in incorrect
modes of failure. Crisfield [19] has shown that such
numerical disturbance of the load transfer path after
initiation of cracking can give rise to alternative
equilibrium states and, hence, lead to false ultimate
The failure load is assumed to occur at a load level
for which a large number of iterations are required
increments take place during this step and that
equilibrium cannot be satisfied under the applied
loads. Obviously, the maximum number of iterations
depends on the problem and the specified tolerance,
but a maximum of 30 iterations seems adequate for
a tolerance of 1%. This is the limit in the number of
iterations selected in this study.
APPLICATIONS
Two simply-supported reinforced concrete beams
have been investigated. These beams are specimen J-4
tested by Burns and Siess [20] and specimen A-l
tested by Bresler [21]. In these case studies the
 Nonlinear FE analysis of R/C structures II

concrete was modeled by eight-node serendipity Table 1. Material properties used in applications

plane stress elements with 3 x 3 Gauss integration Material properties

and the reinforcement was modeled by the two-node E, E, .fi
Examples [ksi] [ksi] [ksi] [ki] $10
truss element. In all studies the bond-slip effect is
taken into account with bond-link element and the Beams J-4 3800 29 500 4.82 44.9 0.99
material properties of the three test specimens are A-l 3367 31 600 3.49 80.5 1.53
Beam-column 3900 28 700 4.97 71.4 1.22
summarized in Ta.ble 1.
joint
Specimen J-4 consists of a simply-supported beam
(I ksi = 70.31 kg cm-‘).
with a span of 12 ft (3.7 m) which was subjected to
a concentrated load at midspan. The geometry and
the cross section of beam J-4 are shown in Fig. 9a.
The finite element mode1 in Fig. 9b represents only from the material properties in Table I based on the
half of the structure taking advantage of the recommendations of Eligehausen et aI. [22]. The
symmetry in geometry and loading. The finite tension stiffening effect is taken into account by
element discretization, the arrangement of the the model proposed in tension stiffening effect. The
reinforcement, the loading and support conditions correlation between the measured load-displacement
are shown in Fig. 9b. The parameters of the bond curve of the beam and the analytical results which
stress-slip relation in Fig. 7 are assumed as follows: include the effects of tension stiffening and bond-slip
d,, = 0.0007854 in (0.002 cm), drJ = 0.00589 in is shown in Fig. IO. The results are also compared
(0.015 cm), 7, =: 286 psi (20.1 kg cm-‘) and with those of an earlier study by Barzegar and
Q = 572 psi (40.2 kg cm-‘). These values are derived Schnobrich [23] who included only the tension
stiffening effect by determining the slope of the strain
softening region according to Hillerborg’s model [ 171.
In order to study the effect of finite element mesh
size on the analytical results three different mesh
configurations with 16, 24 and 48 elements were
investigated. In all configurations the number of
elements through the depth of the member remained
the same as shown in Fig. 9b and only the element
size in the span direction was progressively reduced.
The effect of finite element mesh size is studied for
the three tension stiffening models. The results in
Fig. 10ac indicate that all tension stiffening models
exhibit satisfactory behavior and lead to response
..r....ySIip 11 predictions which are essentially independent from
the finite element mesh size. The size dependence is
0.4 0.5 0.6
more pronounced in Hillerborg’s model while the
CENTER DEFLECTION (In)
crack band theory and the model proposed in this
(a) Tension Stiffening Effect of Beam J-4 study show a very satisfactory objectivity of the
results. When comparing, however. the results of the
crack band theory and Hillerborg’s model with those

r 1
of the tension-stiffening model proposed in this study,
it is concluded that the load-deflection behavior of
the former models is stiffer than that of the present
model. This can be attributed to the assumption of
those models that the microcracks are uniformly
distributed over the entire finite element.
If the tension stiffening effect is not included in the
model, however, there is a marked dependence of the
analytical results on the finite element mesh size
(Fig. IOd). The load-deflection curves exhibit more
flexible response with increasing grid refinement.
Thus finite element models which are based on the
tensile strength of concrete and not on a fracture
energy criterion produce results which strongly
0 0.1 0.2 0.2 0.4 0.6 0.6
depend on the analyst’s choice of mesh configuration
CENTER DEFLECTION (in) and can lead to erroneous estimates of structural
(b) Bond-Slip Effect of Beam J-4 stiffness. This fact corroborates earlier findings by
Fig. 1I. Tension stiffening and bond-slip effect of beam J-4. several investigators [2].
(1 kip = 0.4536 ton, I in = 2.54 cm.) The models of tension-stiffening and bond-slip
12 H. G. Kwak and F. C. Filippou

(4
0.5P
W
2@#4
i

@j-----J& ,;;;:

It-_----- 1828.5mm ----+ k 228mm

co*cRETE
@I . NODES
f----

Fig. 12. Description of beam A-l: (a) configuration of beam A-l; (b) finite element idealization of beam
A-l). (1 in = 2.54 cm.)

reveal that tension-stiffening has the opposite effect
than bond-slip on the load-displacement response of
the finite element model of the structure. One way of
including the effect of tension-stiffening between
6 2 6 a
CENTER DEFLECTION (mm)
Fig. 13. Load-deflection relationships of beam A-l.
cracks is to increase the stiffness of the reinforcing
steel model. On the other hand bond-slip reduces the
stiffness of the reinforcing steel model. To identify the
relative contribution of each effect four different
analyses are performed. In Fig. 1la the effect of
tension-stiffening is included in both analytical
results. From this figure, it is clear from the
comparison of these results with the experimental
data shown by the solid line that the inclusion of both
effects yields a very satisfactory agreement of the
model with reality. Figure 11b shows two analytical
results which exclude the effect of tension-stiffening.
In this case the inclusion of bond-slip (dotted line)
produces a slightly mote flexible response than the
experiment while the exclusion of bond-slip produces
a slighter stiffer response. The comparison of Fig. 1 la
with Fig. 11b illustrates why some investigators have
concluded that neither tension-stiffening nor bond-
slip is important in the analysis, provided that the
right size of finite element mesh is used.
12
The geometry and cross-section dimensions of the
next specimen A-l ate presented in Fig. 12a while
Fig. 12b shows the finite element model of this
 Nonlinear FE analysis of R/C structures 13

the effect of bond-slip in their study, the comparison

of analytical predictions demonstrates the
significance of this effect in the load-displacement
response of a heavily reinforced specimen such as
Beam A-I.
To identify the relative contribution of each effect
four different analyses are also performed for this
specimen in the same way with specimen J-l. From
Fig. I4a and b, it is clear that the inclusion of both
effects (the dotted line in Fig. I4a) yields a very
satisfactory agreement of the model with reality but
on the other hand the inclusion of only bond-slip (the
dotted line in Fig. l4b) produces a slightly stiffer
0 3 s a 12 response. Moreover, Fig. l4a and b show that the
CE:NTER DEFLECTION (mm) contribution of bond-slip to the load-ilisplacement
response of the specimen increases with the load.
(a) Bond Slip Effect with Tension Stiffening Effect Near the ultimate strength of the beam the magnitude
of the bond-slip contribution to the load-displace-
ment response is almost twice that of the tension-
stiffening effect which is of opposite sign. By contrast,
in the lightly reinforced beam specimen J-4 in
360 Fig. Ila and b the magnitude of the bond-slip
4 contribution to the load-displacement response is
s comparable to that of the tension-stiffening effect so
zi 270
that these two effects practically cancel each other
8 (long-dashed line in Fig. 11b).
A 160 In order to assess the ability of the proposed finite
2 element model to simulate the behavior of beam-
Ll column joint subassemblages, specimen BC4, which
was tested by Viwathanatepa et a/. [24], is selected for
further study. The subassemblage consisted of two
0 9 in (22.86 cm) by 16 in (40.64 cm) girders framing
0 3 6 s 12
into a l7in (43.18cm) by l7in (43.18cm) square
CENTER DEFLECTION (mm)
column. The dimensions and the reinforcing arrange-
(b) Bond Slip Effect without Tension Stiffening Effecl ment of the subassemblage are shown in Fig. 15. The
Fig. 14. Tension stiffening and bond-slip effect of beam A- I. column was bolted at the top and bottom to steel
(I kip = 0.4536 ton. I in = 2.54cm.) clevises for mounting the specimen on the testing
frame. The main longitudinal reinforcement of the
beams consisted of 4 # 6 bars at the top and 3 # 5
bars at the bottom of the section with #2 tied
structure. The bond stress-slip relation and any other stirrups spaced at 3.5 in (8.9 cm) as transverse
material properties not specifically mentioned in reinforcement (Fig. I5b). The longitudinal reinforce-
Table I are the same as those used in the analysis of ment of the column consisted of I2 # 6 bars with # 2
beam J-4 since the concrete composition and strength ties spaced I .6 in (4.06 cm) center-to-center along
of the two specimens is not much different. Since column providing the transverse reinforcement
Table I only contains the steel properties of the (Fig. 15~). Seven #2 ties were placed inside the
bottom longitudinal steel of specimen A-I(#9 bars), beam<olumn joint region to satisfy the confinement
the following steel properties are assumed for the and shear resistance requirements of the building
top steel and the transverse reinforcement: E,= code. The bond stress-slip relation and any other
29 200 ksi (2.05 x lOh kg cm-?) and ,f; = 50.1 ksi material properties not specifically mentioned in
(3523 kg cm-‘) for #4 bars as top longitudinal steel Table I are the same as those used in the analysis of
and E, = 27500 ksi (1.9 x lO”kgcmm?) and beam J-4 since the concrete composition and strength
,f; = 47.2 ksi (33 I9 kg cm ml) for # 2 bars as transverse of the two specimens is not much different [24].
reinforcement. The beam-column subassemblage was subjected to
Figure I3 compares the analytical results with a constant axial load P of 470 kips (213 ton) which
the measured load--displacement response of beam simulated the effect of gravity loads and a horizontal
A-l. Very satisfactory agreement between analysis load H at the lower column end which was cycled to
and experiment is observed. The analytical results simulate the effect of lateral loads on the subassem-
by Adeghe and Collins [5] are also shown in Fig. 13. blage. Specimen BC4 was subjected to a very severe,
Since Adeghe and Collins did not account for pulse-type loading with a single load reversal in order
14 H. G. Kwak and F. C. Filippou

I ’ h- 2’-7.5” +I!#

REGAyREa 07.10)
%

COLllEs @ 1.eyT07.13)
ATEAcHslM

(a) k?? (b) (0)

Fig. 15. Geometry and reinforcing details of beam to column subassemblage.

to study the monotonic behavior of the subassem-
blage and establish the load-displacement response
envelope. Other specimens were then cycled several
times with gradually increasing displacement ductility
to simulate the effect of earthquake excitation on
the subassemblage. Since the load-displacement
response of subassemblage BC4 during the first load
cycle which represents a monotonic load test to near
failure was well established during the experiment,
it serves as an ideal case study for the proposed
model.
The finite element representation of the beam-
column subassemblage is shown in Fig. 16. Concrete
was modeled by eight-node isoparametric elements
and the longitudinal and transverse reinforcement
was modeled by two-node truss elements. The
bond-slip effect is included in the analysis with bond
link elements.

V = 470 kipr

Fig. 16. Finite element idealization of beam to column subassemblage. (I kip = 0.4536 ton,
I in = 2.54 cm.)
 Nonlinear FE analysis of R/C structures
30
6 0.1 0.2 0.3 0.4 0.6
CENTER DEFLECTION (in)
(a) LoadOisplacement Relationships
0 0.1 0.2 0.3 0.4 0.6
CIENTERDEFLECTION (in)
(b) Tonsion Stilffonlng and Bond-Slip Effoot
Fig. 17. Numerical results of beam to column subassem-
blage. (1 kip =: 0.4536 ton, 1 in = 2.54 cm).
Figure 17 compares the analytical results with the
measured load-displacement response of the sub-
assemblage. With the effects of tension stiffening and
bond-slip the analysis shows excellent agreement with
the experimental results. The lateral force of 36 kips
in Fig. 17a closely approximates the ultimate load of
the specimen. Figu:re 17b shows that the effect of
bond-slip affects the behavior of the subassemblage
much more than te:nsion stiffening. In this case the
bond-slip of reinforcing bars in the joint contributes
approximately 33% of the total deformation of the
subassemblage near the ultimate load of 36 kips
(2531 kg). The observation of the significance of
bond-slip in the beam-column subassemblage agrees
with the results of the over-reinforced concrete beam
A-l where the interaction of bond and shear plays a
significant role in the response. This interaction also
has an important effect on the stress transfer in
beam-column joints, particularly, when these are
subjected to loading simulating the effect of lateral
I5
loads. Even though the tension stiffening effect plays
a minor role in the response of this beam-column
joint, it should not be concluded that it can be
excluded from the model since it helps prevent
numerical instability problems in connection with
crack formation and propagation.
CONCLUSIONS
A new smeared finite element model is proposed
based on an improved cracking criterion which is
derived from fracture mechanics principles. This
0.6
model retains objectivity of the results for very large
finite elements since it considers cracking to be
concentrated over a small region around the
integration point and not over the entire finite
element as do previous models. Moreover, a new
reinforcing steel model which is embedded inside a
concrete element is developed to cope with the
difficulty in modeling of complex geometry, particu-
larly in three-dimensional finite element models.
The correlation studies between analytical and
experimental results and the parametric studies
associated with them lead to the following con-
clusions: (1) the inclusion of tension-stiffening is
important for the independence of the analytical
results from the size of the finite element mesh and
also for avoiding numerical problems in connection
with crack formation and propagation; (2) tension-
stiffening effect and bond-slip cause opposite effects
on the response of R/C members. While tension-
stiffening which accounts for the concrete tensile
0.6
stresses between cracks increases the stiffness of the
member, bond-slip leads to a stiffness reduction. In
lightly reinforced beams, these effects can cancel each
other at certain load stages, thus, leading to the
erroneous impression that they can be neglected in
the analysis. Since bond-slip increases with loading,
while tension-stiffening does not, consistent results
can only be obtained when both effects are included
in the model. Moreover, the effect of bond-slip clearly
outweighs the contribution of tension-stiffening in
heavily reinforced beams and beamcolumn joint
subassemblages. In these cases the exclusion of the
bond-slip effect can lead to significant overestimation
of the stiffness of the member; and (3) the tensile
strength of concrete has insignificant effect on the
load-displacement response of R/C beams. Crack
formation and propagation is rather influenced by the
fracture energy of concrete.
REFERENCES
Y. R. Rashid, Analysis of prestressed concrete pressure
vessels. Nuclear Engng Des. 7(4) 334344 (1968).
Z. P. Bazant and L. Cedolin, Finite element modeling
of crack band propagation. J. Struct. Engng, AXE
109(STl) 1287-1306 (1980).
R. J. Cope, P. V. Rao, L. A. Clark and R. Norris,
Modeling of Reinforced Concrete Behavior for Finite
Element Analysis of Bridge Slabs, Numerical Methods
16 H. G. Kwak and F. C. Filippou

for Nonlinear Problems (Edited by C. Taylor, E. Hinton 1.5. R. V. Milford and W. C. Schnobrich, Nonlinear
and D. R. J. Owen), pp. 457470. Pineridge Press, behavior of reinforced concrete cooling towers. Civil
SwanSed (1980). Engineering Studies SRS no. 514, University of Illinois
4. F. Vecchio and M. P. Collins, The Response of at Urbana, IL (1984).
Reiyfiwced Concrete to In-Plane Shear and Normal 16. Z. P. Bazant and B. H. Oh, Crack band theory for
S/ress, Publication no. 82-03. Department of Civil fracture of concrete. Mater. Swuct. RILEM, Paris,
Engineering, University of Toronto, Toronto (1982). 16(l) 69-93 (1983).
5. L. N. Adeghe and M. P. Collins, A Finite Elemenf 17. A. Hillerborg, M. Modeer and P. E. Peterson, Analysis
Model .for Studying Reinforced Concrete Detailing of crack formation and growth in concrete by means of
Problems, Publication no. 86-12. Department of Civil fracture mechanics and finite element. Cement Concrete
Engineering, University of Toronto (1986). Res. 6(6) 773-782 (1976).
6. D. Ngo and A. C. Scordelis, Finite element analysis of 18. G. B. Welch and B. Haisman, Fracture toughness
reinforced concrete beams. J. ACI 64(3) 152-l 63 (1967). measurements of concrete, Report no. R42, University
7. A. K. de Groot, G. M. A. Kusters and T. Monnier, of South Wales, Sydney (1969).
Numerical modeling of bond-slip behavior. Heron, 19. M. A. Crisfield, Accelerated solution techniques and
Concrete Mech. 26(1B) (1981). concrete cracking. Comput. Meth. Appl. Mech. Engng
8. H. Kupfer, H. K. Hilsdorf and H. Rusch, Behavior of 33, 585407 (1982).
concrete under biaxial stresses. ACI J. 66(66-62) 20. N. H. Burns and C. P. Siess, Loaddeformation
656666 (1969). characteristics of beam-column connections in re-
9. H. G. Kwak and F. C. Filippou, Finite element analysis inforced concrete. Civil Engineering Studies, SRS
of reinforced concrete structures under monotonic no. 234, University of Illinois at Urbana, IL (1962).
loads, Report no. UCB/SEMM-90/14, University of 21. B. Bresler and A. C. Scordelis, Shear strength
California, Berkeley (1990). of reinforced concrete beams. J. ACI 60(l) 51-72
10. E. Hognestad, A study of combined bending and axial (1963).
load in reinforced concrete members, University of 22. R. Eligehausen, E. P. Popov and V. V. Bertero, Local
Illinois Engineering Experiment Station, Bulletin Series bond stress-slip relationships of deformed bars under
no. 399, Bulletin no. I (1951). generalized excitations. Report no. UCB/EERC 83-23,
I I. D. Darwin and D. A. Pecknold, Analysis of cyclic Earthquake Engineering Research Center, University of
loading of plane R/C structures. Comput. Strucr. 7(l) California, Berkeley, CA (1983).
137-147 (1977). 23. F. Barzegar and W. C. Schnobrich, Nonlinear
12. C. S. Lin and A. C. Scordelis, Nonlinear analysis of RC finite element analysis of reinforced concrete under
shells of general form. J. Slruct. Dir.. ASCE lOl(ST3) ~ , short term monotonic loading. Civil Engineering
523-538 i1975). Studies SRS no. 530, University of Illinois at Urbana,
13. F. R. Hand, D. A. Pecknold and W. C. Schnobrich, IL (1986).
Nonlinear layered analysis of RC plates and shells. J. 24. S. Viwathanatepa, E. P. Popov and V. V. Bertero,
Struct. Dir., ASCE 99(ST7) 1491-1505 (1973). Seismic behavior of reinforced concrete interior
14. A. K. Gupta and H. Akbar, A finite element for the beam-column subassemblages. Report no. UCB/
analysis of reinforced concrete structures. Int. J. Numer. EERC-79/14, Earthquake Engineering Research Cen-
Meth. Engng 19, 1705-1712 (1983). ter, University of California, Berkeley, CA (1979).