Some aspects on three-dimensional numerical modelling of reinforced

concrete structures using the nite element method

H.M. Gomes, A.M. Awruch
*
Graduate Program in Civil Engineering, Federal University of Rio Grande do Sul, Av. Osvaldo Aranha, 99, 38 andar-90035-190, Porto Alegre, RS, Brazil
Received 4 November 1999; revised 18 August 2000; accepted 23 August 2000
Abstract
This paper deals with some aspects related to three-dimensional numerical modelling of reinforced concrete structures using the Finite
Element Method (FEM). Some subjects such as the solution technique of the non-linear equilibrium equations and the constitutive model for
concrete and reinforcement steel are emphasised and commented. A robust method for the evaluation of the intersecting points of the
embedded reinforcement bars into the three-dimensional nite element mesh is also presented. The main advantages of the Generalised
Displacement Control Method with the Generalised Displacement Parameter to improve the response of the concrete and reinforced concrete
analyses are highlighted. Finally, a series of numerical examples related to the above-mentioned aspects are presented. q 2001 Elsevier
Science Ltd. All rights reserved.
Keywords: Reinforced concrete; Structural analysis; Finite elements
1. Introduction
In the last three decades, a large number of numerical
models for reinforced concrete analysis have emerged.
They met fairly well the needs for which they were
conceived. These models deal with the behaviour of
concrete structures till a certain limit given by the service
load or prior to the ultimate load. Lately, much attention has
been given to some model renements with the aim to eval-
uate more accurately the post-peak behaviour. For this
purpose, some progresses in material constitutive modelling
and in numerical methods to solve the non-linear equili-
brium equations have been obtained. With respect to the
last aspect, several control methods have been employed
to capture the post-peak behaviour of the structure.
However, the generalised displacement control method
[24] (usually applied to geometrically non-linear analysis)
and the quasi-Newton methods, such as the BFGS [10] (due
to BroydenFletcherGoldfarbShanno), seem to be the
best candidates for the complete evaluation of the load
displacements curves for structures with geometrically
and/or material non-linear behaviours.
Use of the inverse mapping technique improved the steel
reinforcement modelling, because using this technique, the
reinforcement positions are independent of the FE mesh and
one may use irregular or distorted meshes. However,
problems like bond between concrete and reinforcement
bars are not completely solved, especially when using the
reinforcement bar embedding model.
1.1. Concrete material modelling
The concrete is a material, which presents a different
behaviour in tension and compression. For a uniaxial
compression curve, at approximately 0.3f
c
, where f
c
is the
compression peak stress, the concrete changes from a linear
elastic behaviour to a non-linear behaviour, originating the
rst microcracks at this stage. This value is usually proposed
as the elastic limit. At approximately 0.75f
c
, the process of
unstable crack propagation starts. After this level, the beha-
viour is quite non-linear, with a progressive collapse and
loss of the load capacity of the concrete to resist stress
increments till the complete specimen rupture. For a uni-
axial tension stresselongation curve test, a brittle material
behaviour is observed. The elastic limit is observed at
approximately 0.8f
t
where f
t
is the tensile peak stress. As
in compression, there is a progressive loss of the load capa-
city, which depends on factors such as material properties
and dimensions of the tested specimen.
For the non-linear concrete behaviour description, an
elasto-viscoplastic formulation with a smeared crack
approach, which is a slight modication of Chen's Model
[10] is used in this work. A description of the model is
Advances in Engineering Software 32 (2001) 257277
0965-9978/01/$ - see front matter q 2001 Elsevier Science Ltd. All rights reserved.
PII: S0965-9978(00)00093-4
www.elsevier.com/locate/advengsoft
* Corresponding author. Tel.: 155-51-316-3587; fax: 155-51-227-1807.
E-mail address: [email protected] (A.M. Awruch).
depicted in Fig. 1. As indicated in this gure, the failure
surface encloses all loading surfaces and it remains
unchanged during loading. The initial yield surface has a
closed form at the compression range, to account for high
conning pressures, which evolves to the open failure
surface in the hardening phase. In the softening branch,
the yield surface shrinks to a shape proportional to the initial
yielding surface. In the tension range, the yield surface
coincides with the failure surface and no viscoplastic defor-
mation is allowed a priori up to the brittle failure. However,
at this range, the smeared crack approach or an extension of
the elasto-viscoplastic model to the tensiontension region
may be activated. In the hydrostatic plane, the concrete
yielding surface may be written as follows:
f r; s
m
; u r 2kr
f
s
m
; u 0 uuu # 608 1
where r

2J
2
p
is the stress component perpendicular to
the hydrostatic axis, J
2
the second invariant of the stress
deviator tensor, u the similarity angle, s
m
is the hydrostatic
stress and r
f
the failure surface, which depends on the model
parameter adopted. In this work, the Ottosen four-parameter
model [19] is used as yielding and failure surfaces, assum-
ing the following form:
r
f

1
2a
2

2
p
l 1

2l
2
28a3bs
m
=f
c
21
_
_ _
2
where the constants presented in Eq. (2) may be estimated as
described in the CEB-FIP 90 [11] model code or in a gener-
alised manner by the compressive uniaxial strength ( f
c
), the
uniaxial tensile strength ( f
t
), the biaxial compressive
strength ( f
bc
) and the triaxial stress state in the compressive
meridian (j/f
c
, r/f
c
). The parameter k kI
1
; 1
vp
presented
in expression (1) is the shape factor parameter that repre-
sents the surface enlargement or contraction. This parameter
is dened in such a way, that during hardening or softening,
H.M. Gomes, A.M. Awruch / Advances in Engineering Software 32 (2001) 257277 258
Fig. 1. Initial (1), failure (4), loading (1,2,3,4) and softening (5,6,7,8,9) surfaces for concrete.
Fig. 2. Equivalent stress-strain diagram for concrete under uniaxial compression.
both the size and the shape of the subsequent yield surfaces
vary continuously from the initial yield shape to the subse-
quent shapes. This parameter may be dened through
experimental uniaxial tests and extrapolated to multiaxial
stress states by means of equivalent stress and strain rela-
tions. The shape factor parameter is a function of the equiva-
lent viscoplastic strain (e
vp
) and the rst invariant of the
stress tensor (I
1
). The following expressions for k are used
in this work, as suggested by Chen [10]:
where j
c
, j
t
, j
k
and

j are constant values in the
hydrostatic stress axis delimiting the tensiontension,
tensioncompression, compressiontension, compression
compression regions, respectively, and k
0
(e
vp
) is a factor
representing the equivalent stressviscoplastic strain
relation.
The equivalent stressviscoplastic strain diagram for a
uniaxial compression curve is approximated by the follow-
ing equations proposed by the CEB-FIP 90 [11] model code
and indicated in Fig. 2.
For u1
c
u # u1
c;lim
u;
s
c
2
E
ci
E
c1
1
c
1
c1
2
1
c
1
c1
_ _
2
_ _
f
cm
= 1 1
E
ci
E
c1
22
_ _
1
c
1
c1
_ _
4
with
1
c;lim
1
c1

1
2
1
2
E
ci
E
c1
11
_ _
1
1
4
1
2
E
ci
E
c1
11
_ _
2
2
1
2
_ _
1=2
5
and for u1
c
u . u1
c;lim
u;
s
c
2
__
1
1
c;lim
=1
c1
j 2
2
1
c;lim
=1
c1

2
__
1
c
1
c1
_
2
1
_
4
1
c;lim
=1
c1
2j
_
1
c
1
c1
_
21
f
cm
6
H.M. Gomes, A.M. Awruch / Advances in Engineering Software 32 (2001) 257277 259
Fig. 3. Strain-softening diagram with secant unloading and reloading.
kI
1
; 1
vp

1 if
I
1
3
$ j
t
1 1
1 2k
0
1
vp
2j
t
22j
c
1j
t
22j
c
I
1
3
1
I
2
1
9

j
c
2j
t

2
if j
t
$
I
1
3
$ j
c
k
0
1
vp
if j
c
$
I
1
3
$ j
k
k
0
1
vp

j 2
I
1
3


j 1
I
1
3
22j
k

j 2j
k

2
if
I
1
3
, j
k
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
3
with
j 4
__
1
c;lim
1
c1
_
2
_
E
ci
E
c1
22
_
12
1
c;lim
1
c1
2
E
ci
E
c1
__
_
1
c;lim
1
c1
_
E
ci
E
c1
22
_
11
_
2
7
where E
ci
is the modulus of elasticity (MPa) at a
concrete age of 28 days, s
c
the equivalent
uniaxial compression stress (MPa), e
c
the equivalent
uniaxial compression strain, e
c1
is the strain at the
compression peak stress (usually set as 20.0022) and
E
c1
f
c
=1
c1
is the secant modulus at the compression
peak stress ( f
c
) as indicated in Fig. 2. These relations
may be inverted, and adopting a limit for an elastic
material behaviour (for example 0.3f
c
) an explicit rela-
tion for equivalent viscoplastic strain versus equivalent
stress can be found. This relation is used to dene
k
0
(e
vp
).
For the tensiontension region, a strain-softening
model associated to a xed crack model is adopted
in the principal stress component space. The post-
peak tension is modelled as indicated in Fig. 3,
where a G
f
21=2E
ci
1
2
t
l
c
=E
ci
1
t
l
c
is the softening
parameter, G
f
the concrete fracture energy and l
c
the equivalent length, which for quadratic isopara-
metric elements with 20 nodes, may be given approxi-
mately by:
l
c

n
j
j1

n
h
h1

n
z
z1
detJw
j
w
h
w
z
_
_
_
_
1=3
8
where j, h, and z are the element natural axis, J the
element Jacobian matrix, w
i
are the corresponding
weights for the numerical integration and n
i
is the corre-
sponding integration order. This formula provides a
good approximation when the mesh is not distorted
and most cracks are aligned with the mesh lines.
Expression (8), however, does not assure objectiveness
with respect to the integration rule and for cracks not
aligned with the mesh lines. A similar methodology to
dene the characteristic length using the differential
volume corresponding to the integration point instead
of the element volume was proposed by Cervera [15].
This methodology is also dependent of the integration
rule. By assuming an idealised behaviour of a singular
band for the cracked elements, the following expression
for the characteristic length, which depends on the mesh
size, crack direction and spatial position was deduced
by Oliver [18]:
l
c
j
j
; h
j
; 6
j

_

nc
i1
_
2N
p
i
j
j
; h
j
; 6
j

2x
l
j
1
2N
p
i
j
j
; h
j
; 6
j

2y
m
j
1
2N
p
i
j
j
; h
j
; 6
j

2z
n
j
_
f
i
_
21
(9)
where j denotes the integration point, j, h and z are the
element natural axis, l, m and n are the direction cosines of
the normal vector to the crack direction with respect to the
global co-ordinates at integration point j, N
p
i
are the shape
functions of the equivalent C
0
element for the discretized
domain (for 20-node elements, the eight-node N
p
i
shape
functions should be used) and the function f
i
assumes the
values 1 or 0, depending on the C
0
element corners nodes
positions relative to the crack band (1, if the corner node is
ahead of the crack and 0, otherwise). For the identication
of these corner nodes, a set of axis at the centre of each
element is xed and oriented with the normal crack
directions.
To assure mesh objectiveness, the equivalent length must
be lower than the characteristic length introduced by Hille-
borg [8] and given by:
l
H
ch

E
ci
G
f
f
2
t
10
and greater than a characteristic length (l
ch
), which is a
material property depending on the material microstructure
(usually three times the aggregate size). If l
c
. l
H
ch
; the
strength limit must be reduced in order to obtain objective
fracture energy, when a sudden stress drop occurs. This is
accounted by the following relation:
f
t

E
ci
G
f
l
c
_
11
Fig. 3 shows the stressstrain relation in the place where
a crack will be formed. In phase (a), the principal stress
remains in the elastic range. At point (b), when the
failure surface is reached, as indicated in Fig. 1, a
crack is formed and the principal stress and strain
components and corresponding orthotropic directions
are retained for the evaluation of the secant orthotropic
constitutive matrix. At position (c), the previously
opened crack closes partially if the current strain in
the local system e
local
is lower than the last retained
strain e
ref
obtained (point (e)). Thus, the local stress is
computed with the secant modulus, given by the phase
(d), as s
local
s
ref
=1
ref
1
local
. Otherwise, as in point (f),
H.M. Gomes, A.M. Awruch / Advances in Engineering Software 32 (2001) 257277 260
the local stress is computed following the exponential
branch described in the same gure.
The adopted smeared crack model allows the possibility
to have three orthogonal cracks in the principal stress
component space. In the orthotropic stress components
space the stressstrain relation is given by:
s
local

s
D
local
1
local
12
where s
local
and e
local
represent the three principal stress and
strain components and
s
D
local
represents the secant
orthotropic constitutive matrix, dened as:
with
n 1 2n
12
n
21
2n
13
n
21
2n
23
n
32
2n
12
n
23
n
31
2n
21
n
13
n
32
14
and
G
12

1
2
E
1
21 1n
12

1
E
2
21 1n
21

_ _
G
23

1
2
E
2
21 1n
23

1
E
3
21 1n
32

_ _
G
13

1
2
E
3
21 1n
31

1
E
1
21 1n
13

_ _
15
where the orthotropic values for E
1
, E
2
and E
3
are
given by the current value of s
ref
/e
ref
in each of the
principal directions, as depicted in Fig. 3. b is a factor
used to reduce the shear modulus [15] and may assume
values in the range of 0 to 1, depending on the smeared
measure of crack's width. Assuming that the crack
orientation is dened by the direction cosines of its
normal vector, and that the principal stress directions
give this orientation, we have:
S
l
1
m
1
n
1
l
2
m
2
n
2
l
3
m
3
n
3

_
_
_

_
_
_

cosx
0
; x cosx
0
; y cosx
0
; z
cosy
0
; x cosy
0
; y cosy
0
; z
cosz
0
; x cosz
0
; y cosz
0
; z

_
_
_

_
_
_
16
where x
0
, y
0
, and z
0
are the co-ordinate axis of the
rotated co-ordinate system. The corresponding transfor-
mation from local to global co-ordinate system of the
secant orthotropic constitutive matrix may be given by:
s
D
global
T
T
1
s
D
local
T
1
17
where the transformation matrix is given by:
T
1

l
2
1
m
2
1
n
2
1
l
1
m
1
m
1
n
1
l
1
n
1
l
2
2
m
2
2
n
2
2
l
2
m
2
m
2
n
2
l
2
n
2
l
2
3
m
2
3
n
2
3
l
3
m
3
m
3
n
3
l
3
n
3
2l
1
l
2
2m
1
m
2
2n
1
n
2
l
1
m
2
1l
2
m
1
m
1
n
2
1m
2
n
1
l
1
n
2
1l
2
n
1
2l
2
l
3
2m
2
m
3
2n
2
n
3
l
2
m
3
1l
3
m
2
m
2
n
3
1m
3
n
2
l
2
n
3
1l
3
n
2
2l
1
l
3
2m
1
m
3
2n
1
n
3
l
1
m
3
1l
3
m
1
m
1
n
3
1m
3
n
1
l
1
n
3
1l
3
n
1

_
_
_
_
_
_
_
_
_
_
_
_
_
_

_
_
_
_
_
_
_
_
_
_
_
_
_
_
18
As an alternative, a completely viscoplastic model for
the tensiontension region was also included and
applied in some of the numerical examples.
1.2. Reinforcement model
There are nowadays three models available for the three-
dimensional numerical simulation of reinforcement in
concrete: the embedded model, the discrete model and the
smeared model. The smeared model is more adequate to
plate or shell structures where the reinforcement mesh
spacing allows to model sets of bars as layers of equivalent
cross-sectional area. For non-uniform spacing reinforce-
ments and very different cross-sectional areas of bars
H.M. Gomes, A.M. Awruch / Advances in Engineering Software 32 (2001) 257277 261
s
D
local

1 2n
23
n
32
E
1
n
n
12
1n
12
n
32
E
2
n
n
13
1n
12
n
23
E
3
n
0 0 0
n
21
1n
23
n
31
E
1
n
1 2n
13
n
31
E
2
n
n
23
1n
13
n
21
E
3
n
0 0 0
n
31
1n
21
n
32
E
1
n
n
32
1n
12
n
31
E
2
n
1 2n
12
n
21
E
3
n
0 0 0
0 0 0 bG
12
0 0
0 0 0 0 bG
23
0
0 0 0 0 0 bG
13

_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_

_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
13
located at the same layer, the discrete and the embedded
methods are more appropriated. In the discrete formulation,
steel bars are often modelled as bar elements located along
concrete element nodes. This formulation is obviously
restricted, especially in three-dimensional applications and
this drawback may lead to a higher computational cost due
to the unnecessary concrete mesh renement or even to
numerical errors caused by very distorted elements or
equivalent arrangement of reinforcement bars.
The embedded model has been used extensively in 2D
numerical analysis of reinforced concrete structures. The
original formulation proposed by Phillips and Zienkiewicz
[20] was modied by Chang et al. [9] to allow straight bar
segments to be embedded in an arbitrary angle relative to
the natural axis of the concrete isoparametric element.
Balakrishnan and Murray [1] have introduced the embedded
formulation with a bond-slip capability. Some improve-
ments have been presented by Elwi and Hrudey [14],
which allows a more general embedded model for curved
reinforcement.
A great number of the proposed models for embedded
reinforcement in 3D elements assume that the global co-
ordinates of the intersecting points of the reinforcement
bars in the concrete element are given. This approach has
led to simplications in the generation of the required input
data. Isenberg and Levine [16] restricted the positions of
embedded bars to the centre of elements. Cervera [15]
used a distributed set of bars at a constant position with
respect to the element natural axis forming different layers
in the three-dimensional concrete element. Zienkiewicz et
al. [26] and Bhatt et al. [6] imposed the reinforcement to be
parallel to a natural axis. However, Barzegar et al. [2] have
proposed a more general model to overcome the shortcom-
ings of the previous models using the inverse mapping tech-
nique associated with a simplied algorithm to detect the
intersecting points. It was assumed that the reinforcement
bars can be modelled as a sequence of straight segments in
global co-ordinates, therefore a rened mesh is necessary to
dene completely curved bars. The proposed algorithm was
based on the assumption that each straight segment can
intersect or skirt only one point in the brick faces because
they are considered as planar faces by Barzegar et al. [2],
although actually they may be concave or convex surfaces
in the physical domain.
These assumptions can lead to some drawbacks. Let us
consider, for example, the distorted quadratic element
depicted in Fig. 4 and the straight bar segment, with the
sequence of nodes i and j in the global co-ordinate system,
that intersects two points at the same distorted element face.
When mapped into the element natural co-ordinate system,
these two nodes will assume their respective natural co-
ordinates referred to the computational domain, as illu-
strated in the same gure. The bar segment will have a
curved shape and the assumption of a straight line dened
by points i and j at the natural axis space is not adequate.
Distorted element faces due to quadratic interpolation func-
tions and bars with curved shapes (a very common case in
pre-stressed concrete members, where pre-stressed tendons
have often a parabolic trajectory) are situations where the
bar may intersect twice the same element face. To overcome
the restrictions of Barzegar's algorithm, it would be neces-
sary to substitute the straight segment i 2j by several
straight segments to obtain a better representation in the
physical domain (or global co-ordinates space) of bars inter-
secting the same element face in more than one point. As a
search of intersecting points at global axis level (physical
domain) is not adequate for some cases, it seems to be
convenient to handle the search at element natural axis
(computational domain), where the element faces are
planar. Then, a modied embedded formulation is proposed
in this work.
It is assumed that the reinforcement bars are completely
dened in the global co-ordinate system by nodes that
adequately represent the degree of the curve formed by
the bars. For straight bars, two nodes are enough. For para-
bolic bars, at least three nodes are necessary. Note that in the
preceding formulation, even for straight reinforcements, it is
necessary to dene multiple points to account for more than
one intersecting point in distorted element faces. In the
present formulation curves are written in a parametric
form using simple Lagrange interpolation functions, such
that the global co-ordinate of a point in this curve can be
H.M. Gomes, A.M. Awruch / Advances in Engineering Software 32 (2001) 257277 262
Fig. 4. Brick and reinforcement bar segment in global and natural axes.
evaluated as:
x
r
y
r
z
r
_
_
_
_
_
_
_
_
_
_
_
_
_
_

nr
i1
N
r
i
zx
r
i

nr
i1
N
r
i
zy
r
i

nr
i1
N
r
i
zz
r
i
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
19
where N
r
i
z is the Lagrange interpolation function, z the
parametric co-ordinate in the interval [21, 11], x
r
i
; y
r
i
; z
r
i
are the global co-ordinates of nodes dening the curve
and nr denes the degree of the curve formed by a specic
bar.
The point belonging to the reinforcement bar that inter-
cepts an element face should satisfy the following condition:
ij
r
i 1 and ih
r
i # 1 and i6
r
i # 1 20
where j
r
, h
r
and z
r
are the natural reinforcement bar
co-ordinates in the element natural co-ordinate system.
The expression (20) should be evaluated for each of the
natural axis. These natural co-ordinates may be evaluated
as:
j
r
h
r
6
r
_
_
_
_
_
_
_
_
_
_
_
_
_
_
Inverse Mapping of

nr
i1
N
r
i
zx
r
i

nr
i1
N
r
i
zy
r
i

nr
i1
N
r
i
zz
r
i
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
21
In the inverse mapping technique, one looks for element
natural co-ordinates which corresponding global co-ordi-
nates match the reinforcement bar global co-ordinates.
Thus, equating the element global co-ordinate to the
H.M. Gomes, A.M. Awruch / Advances in Engineering Software 32 (2001) 257277 263
Fig. 5. Some reinforcement bar patterns in element natural co-ordinate system.
reinforcement bar global co-ordinates, yields:
x
r
y
r
z
r
_
_
_
_
_
_
_
_
_
_
_
_
_
_
2

ne
j1
N
c
i
j; h; 6x
e
i

ne
j1
N
c
i
j; h; 6y
e
i

ne
j1
N
c
i
j; h; 6z
e
i
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
0 22
where N
c
i
j; h; 6 are the brick element shape functions,
x
e
i
; y
e
i
; z
e
i
are the brick element nodal co-ordinates and ne
is the number of nodes of the brick element. To solve Eq.
(22), Powell's hybrid algorithm is used [17]. This algorithm,
which is a variation of Newton's method, takes care to avoid
large step sizes or increasing residuals.
For each element, a search is performed, in the natural
axis, looking for z belonging to any curve that intercepts,
skirts or just touch the element. It is used a simple bisection
routine to nd the roots of Eq. (22) with constraints (20).
These roots can be more than one for the same element face,
as the case presented in Fig. 4. In each evaluation, the
inverse mapping technique is performed to obtain the
natural reinforcement bar co-ordinates in the element
natural axis system.
Once the intersecting points have been dened, new
points must be also dened to enable the required compat-
ibility between the reinforcement segment and the concrete
element (if 20-node brick elements are used, it is necessary
at least three nodes for the reinforcement segment deni-
tion). Some reinforcement bars patterns, successfully tested
are shown in Figs. 5 and 6.
1.2.1. Constitutive model for the embedded reinforcement
For the constitutive material modelling of the reinforce-
ment steel, a simple uniaxial elasto-viscoplastic model is
used as described in Ref. [27]. Three branches form this
model: a linear-elastic, a linear hardening and a parabolic
hardening softening. In Fig. 7, the uniaxial stressstrain
diagram for a reinforced bar and the parameters used to
dene the curve are shown.
1.3. Evaluation of the concrete and embedded
reinforcement stiffness matrix
For the evaluation of the general stiffness matrix, the
embedded formulation is adopted. The element stiffness
H.M. Gomes, A.M. Awruch / Advances in Engineering Software 32 (2001) 257277 264
Fig. 6. Some reinforcement bar patterns in element natural co-ordinate system (contd).
matrix may be expressed as:
K
rc
K
c
1

nrs
i1
K
r;i
23
where rc stands for reinforced concrete, c for concrete, r for
reinforcement and nrs is the number of the embedded rein-
forcement segments embodied in the brick element. For the
concrete stiffness element, we have:
K
c

_
V
B
T
c
s
D
global
c
B
c
dV 24
where
s
D
c
is the concrete secant orthotropic constitutive
matrix and B
c
the well-known concrete straindisplacement
matrix. For the reinforcement, the element stiffness matrix
is given by:
K
s
A
r
E
r
_
S
B
T
r
B
r
dS 25
where A
r
is the reinforcement area, E
r
the Young modulus
and B
r
the straindisplacement matrix for the reinforce-
ment, which relates concrete brick element displacements
to the axial reinforcement strains. In this paper, it is assumed
perfect bond between reinforcement and concrete. The
concrete strain eld is completely transferred to the reinfor-
cement and the reinforcement is assumed to resist only axial
forces. Thus, as proposed by Elwi and Hrudey [14] and
Ranjbaran [21], the concrete strain, tangent to the reinforce-
ment, is given by:
1
r
11
1
c
11
l
2
1
11
c
22
m
2
1
11
c
33
n
2
1
121
c
12
l
1
m
2
1121
c
23
m
2
n
3
121
c
13
m
1
n
3
26
where 1
c
ij
are the concrete strain components at a particular
point inside the brick element (belonging to a particular
reinforcement bar), l
1
, m
1
, and n
1
are the direction cosines
of the axis tangent to the reinforcement (this transformation
is similar to that used for the secant orthotropic constitutive
matrix). Differentiating Eq. (19), the Jacobian vector at a
generic point of the reinforcement can be evaluated as:
J
rT
z
dx
r
=dz
dy
r
=dz
dz
r
=dz
_
_
_
_
_
_
_
_
_
_
_
_
_
_

r
i1
dN
r
i
z
dz
x
r
i

r
i1
dN
r
i
z
dz
y
r
i

r
i 21
dN
r
i
z
dz
z
r
i
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
27
The reinforcement Jacobian vector itself denes the
tangent to the reinforcement bar. To evaluate the direction
cosines, it is necessary only to normalise this vector using
the Euclidean norm. Through Eq. (17), the matrix B
r
can be
found using the following equation:
1
r
T
1
1
c
T
1
B
c
d B
p
r
d 28
where e
r
is the strain tensor with a rigid rotation to a new co-
ordinate system, whose rst axis is tangent to the reinforce-
ment and B
p
r
is the straindisplacement matrix for the rein-
forcement. Because only element 1
r
11
is relevant, the B
r
vector is dened as a subset of the matrix B
p
r
:
2. Numerical procedures
2.1. Standard elasto-viscoplastic algorithm
For the constitutive formulation, a standard incremental
formulation is adopted. It is assumed that the incremental
total strain tensor components (De
t
) can be separated into an
elastic (De
e
) and a viscoplastic (De
vp
) part at a given time
level. Therefore, it may be written that:
D1
t
t
D1
e
t
1D1
vp
t
29
H.M. Gomes, A.M. Awruch / Advances in Engineering Software 32 (2001) 257277 265
Fig. 7. Uniaxial stressstrain diagram for reinforcement bars.
where D1
t
t
BDd
t
is the relationship between increments
of strain components and incremental displacement compo-
nents, B the straindisplacement matrix and Dd
t
the incre-
mental displacement vector.
The vector of incremental total stress tensor components
Ds
t
at a given time t is dened as:
Ds
t
D D1
t
t
2D1
vp
t
_ _
30
where D represents the constitutive matrix. The total stress
tensor is given by:
s
t
s
t2Dt
1Ds
t
31
The viscoplastic ow rule is dened, assuming an asso-
ciated ow rule, as:
_ 1
vp
gkf r; s
m
; ul
2f
2s
32
where g is the uidity coefcient, f(r,s
m
,u) is the yielding
surface and kl represents a function that assumes the value
of f(r,s
m
,u) for positive values and is zero otherwise. The
fully explicit Euler time integration scheme is adopted to
evaluate the viscoplastic strain increments as follows:
D1
vp
t
_ 1
vp
t
Dt 33
and the total viscoplastic strain tensor is dened by:
1
vp
t1Dt
1
vp
t
1D1
vp
t
34
As Euler explicit scheme is conditionally stable, Dt must
be less or equal to a critical time interval. In this work, Dt
0:61
t
t
= _ 1
vp
t
was adopted.
2.2. Equilibrium equation
At any instant, the required equilibrium equation to be
satised is given by:
_
V
B
T
s
t
dV 1l
t
P
t
0 35
where B is the straindisplacement matrix, P the total
applied force vector, s the stresses vector, l
t
is the current
load factor and V the nite element domain. In the elasto-
viscoplastic algorithm it is dened the pseudo-loads DV
t
,
given by:
DV
t

_
V
B
Ts
DD1
vp
t
dV 1Dl
t
P
t
1C
t
36
where the residual force vector (or the out-of-balance force
vector) C
t
results from the fact that Eq. (35) is only approxi-
mately satised.
3. The generalised displacement control method with a
generalised stiffness parameter
The general behaviour of a non-linear system may
assume characteristics like those depicted in Fig. 8. For a
loaddisplacement curve, the system may present stable
and unstable ranges as indicated in the same gure. These
ranges are dened by limit points, like point A and point D,
and by snap-back points, like points B and C. The limit
points can be dened as points in the loaddisplacement
curve where the tangent is null. It is in such points where
methods like NewtonRaphson with stiffness matrix updat-
ing fail. One can choose to work with NewtonRaphson
using a non-updated stiffness matrix, albeit this could be
H.M. Gomes, A.M. Awruch / Advances in Engineering Software 32 (2001) 257277 266
Fig. 8. General characteristics of a non-linear system response.
questionable in terms of additional CPU time required to
attain equilibrium. Structures that may crack, crush or yield
are subjected to this kind of behaviour. Snap-back points are
characterised by points in the loaddisplacement curve
where the tangent is not dened. Due to its intrinsic unstable
behaviour, the system may present different applied loads
for a same displacement. It is in such points where methods
like the Standard Displacement Control Method fail. This
behaviour is common in structures subjected to sudden fail-
ures, softening or high geometric non-linearity.
Yeon-Bin and Ming-Shan Shieh [24] have shown that a
large number of existing methods for non-linear solution
(such as the NewtonRaphson method, in its standard or
modied form [12], the Standard Displacement Control
Method [4,25], the Arc Length Method [13], the Work
Control Method [23] and the Generalised Displacement
Control Method with the Current Stiffness Parameter
GDCM-CSP [5]) present instabilities near limit and snap-
back points. They also veried that these methods present
some inconsistencies regarding the usual increment load
criterion to reect the stiffness changes, and that another
shortcoming is their inability for the self-denition of load
directions. In this way, Yeon-Bin et al. [24] proposed a new
method to avoid the previous drawbacks for pre- and post-
peak structural behaviour, the so called Generalised Displa-
cement Control Method with the Generalised Stiffness Para-
meter (GDCM-GSP), which was successfully applied to
geometrical non-linear analysis. The general trends in the
geometrical non-linear analysis do not differentiate from
those ones found in a material non-linear analysis. In what
follows, the use of the GDCM-GSP is briey explained.
Consider a non-linear problem solved by an incremental
load loop (j) and an iterative inner loop (i) for unbalanced
residual force corrections. The increment load factor Dl
i
j
may be treated at each iterative step as a new unknown
variable in a given incremental step. Thus, assuming propor-
tional loading, the equilibrium equation may be stated as in
the following equation:
K
i21
Dd
i
Dl
i
P 1DV
i21
37
where Dl
i
is the load increment of iteration (i), DV
i21
the
incremental pseudo-load vector of iteration i 21; P the
nal applied load vector, K
i21
the updated tangent (or
approximately the secant one) stiffness matrix in the itera-
tion step i 21 and Dd
i
the displacement increment vector
in iteration (i). The displacement increment vector may be
obtained by a vector sum given by:
Dd
i

2
Dd
i
1Dl
i

1
Dd
i
38
where
1
Dd
i
and
2
Dd
i
are evaluated by the solution of the
following system of equations:
K
i21

2
Dd
i
DV
i21
and K
i21

1
Dd
i
P 39
In the GDCM-GSP method, the Generalised Stiffness
Parameter (GSP) to evaluate the load factor increment of
the rst iteration i 1 and jth increment step, is dened
as:
GSP
k
1
Dd
1
1
l
1
Dd
1
1
k
1
Dd
1
j21
l
1
Dd
1
j
40
where kl indicates a row vector. Thus, for the rst iteration
and increment step (j), the load factor increment is given by:
Dl
1
j
Dl
1
1
GSP j j
1=2
signGSP 41
where Dl
1
1
represents the initial load factor increment (rst
iteration and rst step), usually a fraction of the total load
vector (in this work a value of 0.01 is adopted). For the next
iterations (i . 1) and at the same increment step j, the load
H.M. Gomes, A.M. Awruch / Advances in Engineering Software 32 (2001) 257277 267
Fig. 9. Single hexahedral 20-node element subjected to a uniform pressure applied in an element face.
H.M. Gomes, A.M. Awruch / Advances in Engineering Software 32 (2001) 257277 268
-2.50E-3 -2.00E-3 -1.50E-3 -1.00E-3 -5.00E-4 0.00E+0
Displacements (m)
0.00
0.20
0.40
0.60
0.80
1.00
L
o
a
d
F
a
c
t
o
r
(

)
GDCM-GSP
Newton-Raphson
Last Converged Load Increment
with the Newton-Raphson Method
Fig. 10. Loaddisplacement curve for the compression stress test.
-8.00E-3 -6.00E-3 -4.00E-3 -2.00E-3 0.00E+0

xx
-25.00
-20.00
-15.00
-10.00
-5.00
0.00

x
x
(
M
P
a
)
GDCM-GSP
Newton-Raphson
Last Converged Load Increment
with the Newton-Raphson Method
Fig. 11. Stressstrain curve for the compression stress test.
factor increment is given by:
Dl
i
j
2
k
1
Dd
1
j21
l
2
Dd
i
j
k
1
Dd
1
j21
l
1
Dd
1
j
42
For the iteration step j 1, it is set
1
Dd
1
0

1
Dd
1
1
. As was
established in Eq. (41), the load increment direction is
dened by the GSP sign, which has the property to change
its sign immediately after the critical points, allowing the
identication of these points. The total load factor at any
incremental step is given by:
l
i
j
l
i
j21
1Dl
i
j
43
Thus, the solution is obtained when a discrete number of
incremental steps are reached or, when possible, the nal
load vector is applied.
4. Examples
4.1. Pure tension and compression tests
To evaluate the adopted concrete constitutive model in
pure tension and pure compression, two examples were
analysed. First, a single 20-node hexahedral element, repre-
senting one quarter of a cubic specimen, is subjected to a
state of stress corresponding to pure tension and pure
compression through an uniform pressure applied in an
element face as indicated in Fig. 9. For the pure tension
model 3 3 3; 14 and 15 point numerical integration
rules were used. Afterwards, three different mesh densities
(with one, ve and 10 elements) were compared. In all
cases, the Ottosen four-parameter model was used as fail-
ure/yielding surface. The adopted material properties are
given in Table 1.
The loaddisplacement curve and the stressstrain curve,
in compression stress state, at the top face for a mesh with
only one element are plotted in Figs. 10 and 11, respec-
tively. Figs. 12 and 13 show the loaddisplacement curve
and the stressstrain curve, in a tension stress state, at the
top face, respectively, for some numerical integration rules
and mesh densities. Fig. 14 shows the cracks patterns, in
tension stress state, for the mesh with one element. In Figs.
1013, solutions obtained with the NewtonRaphson
method were compared with those provided by the
GDCM-GSP. In both cases, only the pre-peak behaviour
H.M. Gomes, A.M. Awruch / Advances in Engineering Software 32 (2001) 257277 269
0.00E+0 2.00E-5 4.00E-5 6.00E-5 8.00E-5
Displacements (m)
0.00
0.20
0.40
0.60
0.80
1.00
L
o
a
d
F
a
c
t
o
r
(

)
Last converged load increment for
the Newton-Raphson method
Newton-Raphson
GDCM-GSP 1 element; 3x3x3
GDCM-GSP 5 elements; 3x3x3
GDCM-GSP 10 elements; 3x3x3
GDCM-GSP 1 element; 14 points rule
GDCM-GSP 1 element; 15 point rule; irregular mesh
Fig. 12. Loaddisplacement curve for the tension stress test.
Table 1
Material properties for the pure tension/compression test
f
c
(MPa) 21.3
E
ci
(MPa) 27560.0
v 0.15
f
t
(MPa) 1.7
G
f
(MN/m) 0.5 10
24
is attained by the NewtonRaphson method, while the
GDCM-GSP represents fairly well the post-peak behaviour.
In the pure tension test, it was employed the Oliver's model
[18] for the evaluation of the characteristic length. It should
be observed that the model is independent with respect to
the several meshes and integration rules used in the tests.
4.2. Cylinder splitting test
The tensile strength of concrete is frequently determined
by experimental tests like the cylinder splitting test (also
known as the Brazilian test). This example has been chosen
to analyse the ability of the proposed model to predict failure
mode ina tensioncompression mode. The numerical analysis
of this example is sensitive to the adopted compression
softening model, since monotonically increasing load
H.M. Gomes, A.M. Awruch / Advances in Engineering Software 32 (2001) 257277 270
Fig. 14. Cracks patterns for the tension stress test (1 element; 3 3 3-
integration rule).
Fig. 15. Finite element mesh for the splitting test (point D is also restrained
in z-direction).
0.00E+0 5.00E-5 1.00E-4 1.50E-4 2.00E-4 2.50E-4
Concete strain
xx
0.00
0.40
0.80
1.20
1.60
2.00
C
o
n
c
r
e
t
e
s
t
r
e
s
s

x
x
(
M
P
a
)
Last converged load increment for
the Newton-Raphson Method
Newton-Raphson
GDCM-GSP 1 element; 3x3x3
Fig. 13. Stressstrain curve for the tension stress test.
displacement curves are obtained if no assumptions are made
with respect to the compression softening.
In this test, a cylindrical concrete specimen with 0.30 m
height and a diameter of 0.15 m is supported horizontally
between the loading platens of the test machine and
compressed along two opposite generators, by metallic
strips, till the test specimen splits across the vertical
diametric plane in two parts. The tensile strength derived
from this experiment may be calculated by:
f
exp
t

2P
pDH
44
where P is the ultimate load in the loaddisplacement curve,
D and H are the diameter and the height, respectively, of the
tested specimen. However, this tensile strength is consider-
ably inuenced by the boundary conditions, leading to scat-
ter data. Upper and lower bounds limits for the ultimate load
can be obtained by a limit analysis and variations of 25%
from the mean value may be found.
Due to symmetry, only a quarter of the specimen was
discretized using 30 quadratic hexahedral elements with
20 nodes (two elements are used for the metallic strip
discretization; a zero displacement is applied in z-direc-
tion at the axial point on the surface half way between
ends to prevent rigid body motion) as shown in Fig. 15.
The loading strip was assumed to behave as linear elastic
material with Young's Modulus 10 times the concrete
Young's modulus. The material properties, which have
been used, are also given in Table 1. The ultimate load
(equal to 0.1201 MN) was calculated by Eq. (44) and
applied as a uniform pressure in the metallic loading
strip. A 3 3 3 integration rule was used. The Ottosen
four-parameter model was used as the failure/yielding
surface. Fig. 16 shows the loaddisplacement diagrams
for the point immediately under the loading platen
(point A in Fig. 15) and Fig. 17 shows the stressstrain
curve for the integration point B indicated in Fig. 15. In
Fig. 18, two views of the tested specimen crack pattern
H.M. Gomes, A.M. Awruch / Advances in Engineering Software 32 (2001) 257277 271
Fig. 16. Loaddisplacement diagram for the point A in the splitting test.
Fig. 17. Stressstrain curve for the point B in the splitting test.
Table 3
Material and geometrical properties for the tested beam
f
c
(MPa) f
t,ex
(MPa) f
t
a
v
a
E
ci
a
(MPa) G
f
a
(MN/m) As (cm
2
) f
y
(MPa) b (cm) h (cm) d (cm) L (cm) P (MN)
22.5 3.96 1.79 0.15 28192.1 0.53 10
24
26.32 555.0 31.0 55.6 46.1 362.8 0.333
a
Values derived from the CEB-FIP-90 model code recommendations.
Table 2
Material and geometrical properties for the tested beam
f
c
(MPa) f
t
(MPa) v
a
E
ci
a
(MPa) G
f
a
(MN/m) As (cm
2
) f
y
(MPa) b (cm) h (cm) d (cm) L (cm) P (MN)
31.1 2.15 0.15 30653.0 0.55 10
24
2.35 550.0 15.3 24.6 22.1 300.0 0.0325
a
Values derived from the CEB-FIP-90 model code recommendations.
H.M. Gomes, A.M. Awruch / Advances in Engineering Software 32 (2001) 257277 272
Fig. 19. Geometrical characteristics of the RC-75-1 beam [23].
Fig. 18. Crack pattern for the splitting test: (a) cut planes (arrows indicate the direction in which the body is observed), (b) cross section and (c) diametric plane.
and cut planes corresponding to the last point of the curve
in Fig. 16 are shown.
4.3. Failure of a beam due to reinforcement yielding
This example shows the failure of a single span rein-
forced concrete beam due to reinforcement yielding.
The example was taken from the test series of Shegg and
Decanini [22] (beam RC-75-1). The geometrical character-
istics of the beam are shown in Fig. 19 and the numerical
values for the load, geometrical and material properties are
listed in Table 2. Due to symmetry, only half of the beam
was discretized by 20 isoparametric 20 node elements
(height 4, length 5, width 1). Fig. 20 shows the mid-span
loaddisplacement curve for a point at the bottom of the
cross-section. In this gure, point A represents the ultimate
load for the experimental work. Relative small differences
may be observed between numerical and experimental tests.
H.M. Gomes, A.M. Awruch / Advances in Engineering Software 32 (2001) 257277 273
-4.00E-2 -3.00E-2 -2.00E-2 -1.00E-2 0.00E+0
Mid span Displacement (m)
0.00
0.20
0.40
0.60
0.80
1.00
L
o
a
d
F
a
c
t
o
r
(

)
Experimental results
Numerical results
Point A
Fig. 20. Loaddisplacement curve for a point at the bottom of the cross-section in the mid-span of the RC-75-1 beam.
Fig. 21. Crack pattern for the RC-75-1 beam at: (a) the peak and at (b) the last increment step.
These variations are expected in deterministic analysis of
concrete structures and could be attributed to spatial
stochastic variations in the material properties of the
beam used in the test. The crack pattern at the peak and
at the last increment step is shown in Fig. 21. Fig. 22
shows the stressstrain history for the mid-span bottom
reinforcement.
Figs. 23 and 24 show the stressstrain history for the top
H.M. Gomes, A.M. Awruch / Advances in Engineering Software 32 (2001) 257277 274
0.00E+0 4.00E-3 8.00E-3 1.20E-2 1.60E-2 2.00E-2
Reinforcement Strain
xx
0.00
200.00
400.00
600.00
R
e
i
n
f
o
r
c
e
m
e
n
t
S
t
r
e
s
s

x
x
(
M
P
a
)
Fig. 22. Stressstrain history for the bottom reinforcement of the RC-75-1 beam.
-6.00E-3 -4.00E-3 -2.00E-3 0.00E+0
Concrete Strain
xx
-30.00
-20.00
-10.00
0.00
C
o
n
c
r
e
t
e
S
t
r
e
s
s

x
x
(
M
P
a
)
Fig. 23. Stressstrain history for the concrete top face at the mid-span of the RC-75-1 beam.
and bottom face, respectively, at the mid-span cross-section.
As indicated by the cracks patterns and by the stressstrain
history at the top face, after the reinforcement yields, the
compressed concrete splits at the mid-span cross-section.
4.4. Bresler and Scordelis [7] series tests
The simple supported beam OA-1 from the test series of
Bresler and Scordelis [7] is analysed. Unfortunately, the
H.M. Gomes, A.M. Awruch / Advances in Engineering Software 32 (2001) 257277 275
0.00E+0 4.00E-4 8.00E-4 1.20E-3 1.60E-3 2.00E-3
Concrete Strain
xx
0.00
0.50
1.00
1.50
2.00
2.50
C
o
n
c
r
e
t
e
S
t
r
e
s
s

x
x
(
M
P
a
)
Fig. 24. Stressstrain history for the concrete bottom face at the mid-span of the RC-75-1 beam.
Fig. 25. Loaddisplacement curves for the mid span for the OA-1 beam.
tests were carried out by load increments disallowing the
post-peak behaviour evaluation. The mode of failure
observed in this test was due to diagonal tension failure,
since the beam has no web reinforcement. This mode of
failure is characterised by a sudden failure after the forma-
tion of the critical diagonal tension crack, splitting along the
tensile reinforcement near the support and in the compres-
sion zone close to the applied load. The geometric charac-
teristics, material properties and load value of the tested
beam are shown in Table 3. Some material properties, like
f
t
, E
ci
, G
f
and n, were derived from the CEB-FIP-90 model
code [11], since reported experimental results were not
found for these properties. It was set n 0:15, E
ci

28192:1 MPa; and some combined values for f
t
and G
f
were tested to evaluate the sensitivity of the model to
these parameters. The loaddisplacement curves diverge
pronounced from the experimental results for values of f
t
different from 2.25 MPa and they are not shown here. Fig.
25 shows the mid-span displacement history for the OA-1
beam and in Fig. 26, the stressstrain history for the mid-
span top and bottom faces are presented using two different
meshes with 20 and 100 (height 4, length 25, width 1)
elements, respectively. In Fig. 27, the crack patterns for
the mesh with 100 elements with the smeared crack model
at the last converged point is showed. For the mesh with 100
elements, no differences were observed between the
smeared crack model and the fully viscoplastic model as
depicted in Figs. 25 and 26. The analysis was stopped
both for the fully viscoplastic and smeared crack options
at point A in Fig. 25 due to excessive deformation in the
reinforcement bar near support (1
r
11
0:07 [7]).
5. Conclusions
Some aspects referred to the three-dimensional numerical
modelling of reinforced concrete structures using the Finite
Element Method have been analysed in this work. Some
particular features about the constitutive model for the
concrete and the reinforcement proposed here were dened
and applied to some examples. The advantages in the use of
the GDCM-GSP for material non-linear problems were
highlighted by numerical examples where post-peak situa-
tions have been reproduced. The proposed method for rein-
forcement embedding was successfully implemented,
overcoming the drawbacks found in the previous methods.
Some topics, like the bond-slip between concrete and steel
were not included, although this specic subject will not
result in very different behaviours for the examples
presented here, as stated by Barzegar [3]. Other aspects,
such as the effect of long-term loads and corrosion of steel
bars, will be presented in another paper.
H.M. Gomes, A.M. Awruch / Advances in Engineering Software 32 (2001) 257277 276
Fig. 26. Stressstrain history at the mid-span of OA-1 beam (top and bottom faces).
Fig. 27. Crack pattern for the OA-1 beam corresponding to the last point in
Fig. 25 (Point A) for the mesh with 100 elements.
Acknowledgements
The authors wish to thank the National Research Councils
CNPq and CAPES for the partial nancial support.
References
[1] Balakrishnan S, Murray DW. Finite element prediction of reinforced
concrete behaviour. Struct. Engrg. Rep. No. 138, University of
Alberta, Edmonton, Alberta, Canada, 1986.
[2] Barzegar F, Manddipudi S. Generating reinforcement in FE model-
ling of concrete structures. J Struct Engng, ASCE 1994;120(5):1656
62.
[3] Barzegar F, Maddipudi S. Three-dimensional modelling of concrete
structures. I: Plain concrete, II: reinforced concrete. J Struct Engng
1997;123(10):133956.
[4] Batoz J-L, Dhatt G. Incremental displacement algorithms for non-
linear problems. Int J Num Methods Engng 1979;14(8):12627.
[5] Bergan PG. Solution technique for non-linear nite element
problems. Int J Num Methods Engng 1978;12:167796.
[6] Bhatt P, Barj MS, Elnounu GF, Memon M. Non-linear nite element
analysis of shearwall-oor slab junction. In: Owen DR, Hinton E,
Onate, editors. Proc Second Int Conf on Computational Plasticity,
Swansea, Wales: Pineridge Press, 1989. p. 131930.
[7] Bresler B, Scordelis C. Shear strength of reinforced concrete beams. J
Am Concrete Inst 1963;60(1):5173.
[8] Cervenka J. Discrete crack modelling in concrete structures. PhD
thesis, University of Colorado, Boulder, 1994.
[9] Chang TY, Taniguchi H, Chen WF. Non-linear nite element analysis
of reinforced concrete panels. J Struct Engng (ASCE)
1987;113(1):12240.
[10] Chen WF, Han DJ. Plasticity for structural engineers. New York:
Springer, 1988 (606pp.).
[11] Comite Euro-Internacional du Beton. CEB-FIP Model Code 1990.
Bulletin d'information, No. 213/214, Lausanne, May, 1993, 437pp.
[12] Criseld MA. Accelerated solution techniques and concrete cracking.
Comput Meth Appl Mech Engng 1982;33:585607.
[13] Criseld MA. An arch length method including line searches and
accelerations. Int J Num Meth Engng 1983;19:126989.
[14] Elwi A, Hrudey M. Finite element model for curved embedding rein-
forcement. J Engng Mech (ASCE) 1989;115(4):74057.
[15] Hinton E. Numerical methods and software for dynamic analysis of
plates and shells. Swansea, UK: Pineridge Press Ltd, 1988 (550pp).
[16] Isenberg J, Levine HS. Analysis of reinforced concrete under shock
loading. Finite Element Analysis of Reinforced Concrete Structures,
ASCE Task Committee on Concrete and Mansory Structures, ASCE,
1985. p. 44464.
[17] Jorge M, Garbow B, Hillstrom K. User guide for MINPACK-1.
Argonne National Labs Report ANL-80-74, Argonne, Illinois, 1980.
[18] Oliver J. A consistent characteristic length for smeared crack models.
Int J Num Methods Engng 1989;28:46174.
[19] Ottosen NS. Constitutive model for short-time loading of concrete. J
Engng Mech Div 1979;105(1):12741.
[20] Phillips DV, Zienkiewicz OC. Finite element non-linear analysis of
concrete structures. Proc Inst Civ Engrs, Part 2 1976;61(3):5988.
[21] Ranjbaran A. Mathematical formulation of embedded reinforcements
in 3D brick elements. Commun Num Methods Engng 1996;12:897
903.
[22] Shegg A, Decanini L. On deections of reinforced concrete elements.
In Proceddings of Jornadas Sulamericanas de Engenharia Estrutural
Porto Alegre, Brazil (in Portuguese), vol. 15(2), 1971. p. 1070120.
[23] Yang Y-B, McGuire W. In: Middleton J, Pande GN, editors.
NUMETA 85 International Conference on Numerical Methods
in Engineering: Theory and Applications, Wales, UK: Univ College
Swansea, 1985. p. 91321.
[24] Yang Y-B, Shieh M-S. Solution method for non-linear problems with
multiple critical points. AIAA J 1990;28(12):21106.
[25] Zienkiewickz OC. Incremental displacement in non-linear analysis.
Int J Num Methods Engng 1971;3:5878.
[26] Zienkiewickz OC, Owen DRJ, Phillips DV, Nayak GC. Finite
element methods in the analysis of reactor vessels. Nuclear Engng
Des 1972;20:50741.
[27] Owen DRJ, Hinton E. Finite elements in plasticity: theory and pratice.
Swansea: Pineridge Press, 1980 (p. 95).
H.M. Gomes, A.M. Awruch / Advances in Engineering Software 32 (2001) 257277 277