Professional Documents
Culture Documents
XFEM Abaqus
XFEM Abaqus
Sohngårdsholmsvej 57
www.bsn.aau.dk
Synopsis:
The report's content is freely available, but the publication (with source indications) may only happen by agreement
with the authors.
Preface
This report is a product of Poul Reitzel's and Baldvin Johannsson's project work at the 4th
semester of the master degree in Structural and Civil Engineering at Aalborg University.
The project has been completed within the period of 1st of February, 2011, to the 10th of
June, 2011, under the supervision of Christian Frier. The report is prepared and made in
accordance and compliance with the current curriculum of the 4th semester of the master
programme in Civil Engineering at Aalborg University, Denmark.
The project is based on the theme "Numerical analysis of a Reinforced Concrete Beam
in Abaqus 6.10". The project aims at the increase of knowledge to apply an advanced
computational method for the evaluation of crack formation and propagation in reinforced
concrete.
The project report consists of two parts, the main project and the appendix. A reference
to the appendix can be: Appendix B. The main project examines the use of the extended
nite element method, abbreviated XFEM, in combination with the Concrete Damaged
Plasticity material model, abbreviated CDP, in Abaqus 6.10 to model a three-dimensional
reinforced concrete beam loaded to failure.
The project report uses the Harvard method of bibliography with the name of the source
author and year of publication inserted in brackets after the text, for example: Irwin
[1958]. The list of all references is found in the bibliography at the end of the report. The
les used in Matlab and Abaqus 6.10 can be found on the attached CD. An introduction
to the les and how they work can be found in Appendix B.
The authors expect the reader to have a basic knowledge of the standard nite element
method. Experience with computational engineering within the nite element method
framework will make the interpretation of certain technical terms easier.
iii
Table of contents
Chapter 1 Introduction 3
Chapter 2 Fracture mechanics for concrete 7
2.1 The fracture process in concrete . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.2 Linear elastic fracture mechanics . . . . . . . . . . . . . . . . . . . . . . . . 10
2.3 Non-linear fracture mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . 13
v
Group B122b - Spring 2011 TABLE OF CONTENTS
Chapter 8 Conclusion 89
Chapter 9 Suggestions for future work 93
Appendix A Concrete Damaged Plasticity material model 3
A.1 Concrete Damaged Plasticity . . . . . . . . . . . . . . . . . . . . . . . . . . 3
A.1.1 Stress-strain relations . . . . . . . . . . . . . . . . . . . . . . . . . . 3
A.2 Yield function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
A.2.1 Damage and stiness degradation for unixial loading . . . . . . . . . 7
A.2.2 Flow rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
vi
TABLE OF CONTENTS Master Thesis
Bibliography 13
vii
Resumé
Analyse af konstruktioner med komplekse geometrier vil ofte kræve anvendelse af nu-
meriske værktøjer som fx. nite element metoden. Denne afhandling omhandler kohæsiv
revnevækst i beton inden for rammerne af XFEM. Hovedparten af arbejdet er relateret
til modellering af revnevækst i en tre-dimensionel armeret betonbjælke ud fra et model-
leringsteknisk perspektiv.
1
Introduction
1
In this chapter the motivation for this project is described followed by a
presentation of the problem to be handled. This leads to the problem formulation
of the project, covered in the report. Finally, a list of objectives is stated to answer
the problem formulation.
On the other hand concrete is a relatively brittle material with low tensile strength
compared to the compressive strength. For this reason, concrete elements are reinforced
3
Group B122b - Spring 2011 1. Introduction
with steel bars in areas of the cross-section which is subjected to tension to provide some
tensile resistance. In this way, reinforced concrete acts a bi-material utilizing the high
compressive strength of concrete and the tensile strength of steel. A.H. Nilson, D. Darwin
and Charles W. Dolan [2004]
Within the Finite Element framework it is desirable to represent the actual stress
distribution and the development hereof, e.g. by employing an advanced material model,
to precisely model the section stresses at every load stage. A principle sketch of the stress
development in a cross-section for increasing bending moment is shown in gure 1.2.
Figure 1.2. Principle sketch of the stress development for increasing bending moment. Søren
Madsen [2009]
Cracking of the concrete is unavoidable due to its low tensile strength and low extensibility.
The downside of cracking is twofold. Cracking exposes the reinforcing steel to the
surrounding environment, which can cause corrosion of the steel. Secondly, a cracked
construction loses its aesthetical qualities and appears unsafe to reside in. Studies
performed by Arya and Ofori-Darko [1996] reveal that crack spacing is a governing factor
in the rate of corrosion. They found, that several small cracks are more severe than
one big crack. Other studies performed by Schieÿl and Raupach [1997] and Mohammed
et al. [2001] show, that the crack width inuences the time to initiation of the corrosion.
To avoid this, the engineer must ensure that the crack widths and spacing are within
the allowable limits put forth by the governing code. In addition, cracking of a member
will cause reduction in bending stiness, which inuences the deection of the member.
It is important to accurately assess the inuence of cracks on the deection of concrete
members in the serviceability limit state. Piyasena [2002]
The standard FEM is well established as a robust and reliable numerical technique
for studying the behavior of a wide range of engineering and physical problems. All
physical phenomena encountered in engineering mechanics are modeled by dierential
equations, usually too complicated to be solved by analytical methods. For problems
4
Master Thesis
where the solution variables behave in a continuous manner, the FEM is a highly suited
method for approximating the solution to the dierential equation governing the addressed
problem. However, a large number of models in continuum mechanics involve solutions
that are discontinuous in local parts of the solution domain. For example, displacements
change discontinuously across cracks. In the FEM special care must be taken in the
construction of an appropriate mesh, as element topology must align with the geometry
of the discontinuity, and this is not desirable in applications, where the crack location
is unknown a-priori. In applications, such as crack analysis, where the FEM encounters
problems, other, more appropriate numerical techniques exist. One class of techniques is
the so-called enriched methods, which are advantageous for problems having non-smooth
and non-regular solution characteristics. A popular enriched method is the so-called
extended nite element method, abbreviated XFEM. The XFEM was implemented by
Dassault Systémes Simulia Corp. [2010] in their latest version of Abaqus (6.10), which
puts the engineer in a position of being able to qualitatively estimate crack patterns,
spacing and widths for arbitrary geometries. Abaqus 6.10 also oers advanced material
models to accurately model the behavior of concrete shown in gure 1.2. Thomas-Peter
Fries and Andreas Zilian. [2010]
In order to handle the described problem, the following objectives for the project are set:
5
Group B122b - Spring 2011 1. Introduction
Figure 1.3. Boundary conditions, dimensions and reinforcement arrangement in the examined
reinforced concrete beam.
Note, that shear reinforcement is not incorporated into the model in order to simplify the
modeling work load in Abaqus 6.10.
6
Fracture mechanics for
concrete
2
This chapter presents an introduction to cracking in concrete with focus on the
fracture process in compression and tension. This is followed by a description of
the fundamental concepts of linear elastic fracture mechanics, LEFM. Within the
scope of LEFM the theory in Grith [1921]/Irwin [1958] is presented along with
the cohesive crack model proposed by Dugdale [1960] and Barrenblatt [1959].
Following the description of LEFM, the work done by A. Hillerborg and Peterson
[1976], within the eld of nonlinear fracture mechanics, is presented.
1. It is not sucient to specify how cracking is initiated, e.g. by a stress criterion, but
also how it will propagate. The growth of a crack requires the consumption of a
certain amount of energy, called the fracture energy. Therefore, crack propagation
can only be studied through an energy criterion.
2. The calculations must be objective, i.e. mesh renement, choice of coordinates etc.
must not aect the results. This entails that the energy dissipated through cracking
is constant, which is done by specifying the energy dissipated per unit length of the
crack.
3. Two basic types of structural failure may be stated: brittle and plastic. Plastic
failure occurs in materials with a long yield plateau and the structure develops
plastic hinges. For materials with a lack of yield plateau, the fracture is brittle,
7
Group B122b - Spring 2011 2. Fracture mechanics for concrete
which implies the existence of softening. During softening the failure zone propagates
throughout the structure, so the failure is propagating.
4. The area under the load-displacement curve determines the amount of energy
consumed during failure process. This energy determines the ductility of the
structure, and a limit state analysis cannot give an indication of this, because the
post-peak response is not taken into account.
5. Fracture mechanics may opposite to strength criterions predict the inuence of the
structural size on the failure load and ductility.
The ve arguments stated above motivate towards using fracture mechanics in the
modeling of concrete when cracking is of interest. Thus fracture mechanics may lead
to a physical explanation of cracking in concrete, that the current codes, e.g. EN 1992-1-1
[2004], do not by their present empirical formulas.
Micro cracks occur on the level of the hydrated cement, where cracks form in the cement
paste. Meso cracks form in the bond between aggregates and the cement paste. Finally,
macro cracks form in the mortar between the aggregates.
8
2.1. The fracture process in concrete Master Thesis
Figure 2.1. The compressive stress-strain curve for concrete. The curve is divided into four
regions for dierent states of cracking. J.P. Ulfkjær [1992]
Initial cracks on the micro-level, caused by shrinkage, swelling and bleeding, are observed
in the cement paste prior to loading. For loads of approximately 0 − 30 % of the ultimate
load the stress-stain curve is approximately linear and no growth of the initial cracks is
observed. Between approximately 30 − 50 % of the ultimate load a growth in bonding
cracks between the cement paste and aggregates is observed. The cement paste and the
aggregates have dierent elastic modulii, which increases the non-linearity of the stress-
strain curve. Beyond 50 % of the ultimate load macro-cracks start to slowly form in the
mortar, running between the aggregates parallel with the load direction. At app. 75 %
of the ultimate load a more complex crack formation is established, where the bonding
cracks and the cracks in the mortar coalesce until nally failure occurs. J.P. Ulfkjær [1992]
9
Group B122b - Spring 2011 2. Fracture mechanics for concrete
Figure 2.2. A concrete rod subjected to pure tensile loading. Outside the fracture zone, the
cracks are uniformly distributed. Inside the fracture zone a macro-crack forms
which splits the rod in two.
In the following section the basis of linear elastic fracture mechanics is presented.
Figure 2.3. Arbitrary body with an internal crack of length a subjected to an arbitrary force,
F.
Π = Π e + ΠF + ΠK + Πc (2.1)
10
2.2. Linear elastic fracture mechanics Master Thesis
The fracture potential, Πc , is the energy that dissipates during crack growth. By assuming
that crack growth is only dependent on the crack length, a, the equilibrium equation can
be stated, by requiring that the potential energy of the system equals zero, see 2.2.
∂Π
=0 (2.2)
t · ∂a
Grith [1921] introduced a parameter, the energy release rate, G, and dened a fracture
criteria, see equation 2.3.
∂Πc
G=− =R (2.3)
t · ∂a
∂G ∂R
> =0 (2.4)
∂a ∂a
The method proposed by Grith [1921] was based on energy considerations, but is not
adequate in design situations. For this reason, Irwin [1958] developed the stress intensity
factor, abbreviated SIF, concept. The SIF can be understood as a measure of the strength
of a singularity, understood in the sense that the SIF amplies the magnitude of the stresses
around the singularity. The literature distinguishes between the three dierent fracture
modes shown in gure 2.4.
11
Group B122b - Spring 2011 2. Fracture mechanics for concrete
Figure 2.4. Top: Crack mode I. Middle: Crack mode II. Bottom: Crack mode III. NDT [2011]
Mode I fracture is the condition in which the crack plane is perpendicular to the direction
of the applied load and mode II fracture is the condition in which the crack plane is parallel
to the direction of the applied load. Mode III fracture corresponds to a tearing mode and
is only relevant in three dimensions. Mode I and mode II fracture is also referred to as
an opening and in-plane shear mode, respectively. Irwin [1958] showed that the stress
variation near a crack tip in a linear elastic material is dependent on the distance to the
crack tip, called r. More precisely, the stress is singular at the crack-tip with a square-root
singularity in r, see equation 2.5.
K
σij = √ · fij (θ) + higher order terms (2.5)
2πr
From equation 2.5 it can be seen, that a linear relationship exists between the stress
and the SIF, which reects the linear nature of the theory of elasticity. In practical
calculations, only the rst order term of equation 2.5 is included. This is because, that for
r → 0, the rst order term approaches innity while the higher order terms are constant
or zero. Because the stress tends towards innity when r → 0 a stress criterion as a failure
criterion is not appropriate. For this reason, Irwin derived a relationship between the SIF
and the energy release rate, G, see 2.6.
√
K = G·E (2.6)
12
2.3. Non-linear fracture mechanics Master Thesis
K = Kc (2.7)
It should be noted that the global energy balance criteria by citebib:grith is equivalent
to the local stress criteria by Irwin [1958]. Moreover, Kc is also referred to as the
fracture toughness of the material, and is regarded as a constant in LEFM. The Grith
[1921]/Irwin [1958] theory assumes that the stresses in the vicinity of the crack-tip tend
to innity. This contravenes the principle of linear elasticity, relating small strains to
stresses through Hooke's law. In the fracture process zone, abbreviated FPZ, ahead of the
crack-tip, plastic deformation of the material occurs. Specically for concrete debonding
of aggregate from the cement matrix and microcracking occurs. Moreover, cracks coalesce,
branch and deect in the FPZ. To describe this highly non-linear phenomenon, non-linear
fracture mechanics, abbreviated NLFM, must be adopted.
Figure 2.5. The cohesive zone model by Barrenblatt [1959]. A body with a crack of length 2a
subjected to tension, σ . Cohesive stresses, q(x), act along a cohesive zones of length
c at each crack-tip.
13
Group B122b - Spring 2011 2. Fracture mechanics for concrete
Inspired by the concept of Dugdale [1960] and Barrenblatt [1959], A. Hillerborg and
Peterson [1976] redened the FPZ by introducing a so-called ctitious crack in front of the
real crack-tip. The purpose of introducing a ctitious crack was to improve the description
of the tractions acting in FPZ. The closure stress in the FPZ has a maximum value of ft ,
i.e. the tensile strength of the material, at the boundary of the FPZ and is zero at the tip
of the real crack. The variation in-between is given by a softening law, relating stresses
to the crack opening displacement, w. Similar to elastic materials with a constitutive law
described by e.g. Hooke's law, the tension softening law is the constitutive law in the FPZ.
Thus the tension softening law describes the transition between the continuous state and
the discontinuous state of the material behavior. Figure 2.6 illustrates a typical tensile
load-displacement response of concrete and the related ctitious crack ahead of the real
crack. Note that the FPZ extends only over the length of the tension softening region
BCD, see gure 2.6. Tension softening is the relationship between the cohesive stress and
the crack opening displacement in the FPZ. Note that the relation between the closure
stress and the crack opening displacement is non-linear, and that a degradation of the
Young's modulus occurs gradually inside the FPZ. J.L. Asferg [2006]
Figure 2.6. (a) Typical tensile load-displacement curve of concrete with letters indicating the
crack-state: A: Uncracked, linear-elastic behavior, B: The tensile strength has been
reached and microcracking and tension softening occur, C: Stress bridging D: The
crack becomes traction-free. (b) The FPZ related to the load-displacement curve.
The variation of the cohesive stresses is indicated and the crack is divided into a
traction-free zone, a microcracking/bridging zone and a microcracking zone.
14
2.3. Non-linear fracture mechanics Master Thesis
As previously mentioned energy is absorbed during crack growth in order to form the new
crack surfaces. The amount of absorbed energy is the fracture energy, Gf , which is equal
to the area under the tension-softening curve, see gure 2.7 and equation 2.8.
Z wc
Gf = σ(w) dw (2.8)
0
As a nal remark, the ctitious crack model proposed by A. Hillerborg and Peterson [1976]
assumes, that the FPZ has negligible width. For this reason, the model belongs to the
class of discrete crack models. The ctitious crack model by A. Hillerborg and Peterson
[1976] is the basis for the cohesive segments method used in Abaqus 6.10. The cohesive
segments method is described in chapter 4.
15
Smeared vs. discrete crack
modelling
3
This chapter presents the concepts of smeared and discrete crack models for
concrete. Popular techniques, e.g. the XFEM, available for discrete crack
modeling are discussed. Advantages and drawbacks are identied and pointed
out for a discrete vs. smeared approach to crack modeling. The purpose of this
chapter is to illustrate the motivation for working with the XFEM in this report.
In the late 1960's D. Ngo and A.C. Scordelis [1967] performed a numerical simulation of
discrete cracks in concrete. At the same time, Rashid [1968] successfully applied a smeared
crack model for concrete. Discrete crack simulation aims at the initiation and propagation
of dominating cracks, whereas the smeared crack model is based on the observation,
that the heterogeneity of concrete leads to the formation of many, small cracks which,
only in a later stage, nucleates to form one larger, dominant crack. The smeared crack
model captures the deterioration process by smearing the eect of microcracks, that is,
a reduction in stiness, over a given volume. With respect to the problem formulation
stated in chapter 1, the objective of the numerical model is to
Since the pioneering work by D. Ngo and A.C. Scordelis [1967] and Rashid [1968] work
has been done to improve the initially presented crack concepts. With respect to the ve
criteria presented above for a feasible numerical model, popular techniques available for
numerical modeling of cracks within the smeared and discrete crack concepts are presented
in the following sections.
17
Group B122b - Spring 2011 3. Smeared vs. discrete crack modelling
w
cr = (3.1)
lp
µE 0 0
Ds = 0 E 0 (3.2)
0 0 βG
The smeared crack concept suers from three major drawbacks outlined in the following.
Firstly, the model is based on the concept of a crack band in which the exact location
of the crack inside an element is unknown. In a case where crack width or spacing is
of interest a discrete approach should be preferred over a smeared approach. Secondly,
the smeared crack concept has convergence problems as a mesh-renement will aect the
width of the localization band. Lastly, the strain imposed by a crack inside an element
implies adjacent elements to be strained as well. This is illustrated in gure 3.1 and is
referred to as stress-locking. Rots and Blaauwendraad [1989]
18
3.2. Discrete crack concept Master Thesis
Stress-locking refers to the situation, where tensile stresses in adjacent elements is still in
the elastic regime and refuses to decrease. This results in locked-in stresses at locations,
such as the sides of a crack, where the stress should be zero. This drawback is a
consequence of approximating a strong discontinuity using the assumption of displacement
continuity. For the above mentioned reasons, a smeared crack model does not satisfy the
ve specied criteria for a feasible numerical model in this report. In the following section
the discrete crack concept will be introduced. Rots and Blaauwendraad [1989]
Figure 3.2. Standard three-noded two-dimensional interface element. J.L. Asferg [2006]
19
Group B122b - Spring 2011 3. Smeared vs. discrete crack modelling
The use of interface-elements, however, puts a constraint on the crack propagation path.
This is so, because the crack is constrained to propagate along the inserted interface-
elements. This constrain renders the modeling of a-priori unknown crack paths dicult.
A remedy is re-meshing at every simulation stage, however the mesh must conform to
the crack geometry, which in the case of curved or intersecting cracks is dicult to
obtain. Several attempts have been made to construct eective remeshing algorithms,
e.g. by A.R. Ingraea [1985], that reduced the mesh bias. However, such algorithms
are computationally expensive. The missing possibilities of identifying the location of
crack initialization and prediction of arbitrary crack propagation paths render the method
unpreferable according to the ve specied criteria for a feasible numerical model in this
report.
The mesh-dependence was to a large extent alleviated by the advent of so-called enriched
methods. An example of such enriched methods is the XFEM described in chapter 4.
General for all enriched methods is to enrich the polynomial approximation space such
that non-smooth solutions can be modeled independent of the mesh. This enables the
class of methods to model discontinuities at arbitrary locations inside element interiors,
such as a displacement discontinuity imposed by a crack. Moreover, enriched methods put
no restriction on the number of cracks in the model, or whether the cracks are predened
or initiated by fullling a material fracture criterion. For j cracks in the model, the
polynomial approximation space is expanded to a sum over the j nodal sets describing
the j cracks. Finally, by belonging to the class of discrete modeling, the XFEM is able
to estimate crack width and spacing. According to the ve specied criteria for a feasible
20
3.2. Discrete crack concept Master Thesis
numerical model in this report, the XFEM seems a valid candidate. With respect to the
last point referring to the simulation of complete failure, this is a question of the numerical
solver implemented in Abaqus 6.10. Fries and Belytschko. [2000]
A drawback of the discrete crack concept is that it is intended for the representation
of dominating cracks, thus neglecting the eect of microcracks known to occur in
heterogeneous materials like concrete. However, the XFEM has attractive properties with
respect to crack modeling and will be used further on in this project.
21
The eXtended Finite
Element Method
4
This chapter presents the general formulation of the XFEM. Initially, the
background of the XFEM is presented and an introduction to the concept of
discontinuities is given. This is followed by a description of a formulation of the
approximation space in the XFEM. A convenient method for choosing the set
of so-called enriched nodes is presented and the concept of blending elements is
introduced. Various enrichment types are discussed, followed by a description of
two methods for numerical integration of the weak form. Finally, the XFEM
approximation to the displacement is derived using the variational principle.
Unless stated otherwise the sources used in this chapter are Fries and Belytschko.
[2000] and Thomas-Peter Fries and Andreas Zilian. [2010] .
Discontinuities and singularities in eld quantities are observed in many areas of civil
engineering, e.g. singular stresses and strains in the vicinity of a crack-tip, or a jump
in displacement across a crack. For the numerical approximation of these non-smooth
variables two fundamentally dierent approaches exist. The rst method relies on
polynomial approximation, based on nite element shape functions, and requires the
mesh to conform to the discontinuities. Moreover, a rened mesh is required in areas
where eld quantities exhibit high gradients, and remeshing is required in order to
model the evolution of interfaces, e.g. cracks, boundary layers and phase transition.
However, for complex geometries an eective remeshing algorithm can be dicult to
construct, as the elements must conform to the geometry of the discontinuity or projection
errors are introduced. Moreover, this is computationally expensive and not suited for
evolving interfaces. The second, fundamentally dierent, method is based on enriching
the polynomial approximation space with discontinuous functions, such that non-smooth
solutions can be modeled independent of the mesh. This is a basic principle of the XFEM,
and was developed by Belytschko and Black [1999] and N.Moës and Belytschko [1999],
based on the partition of unity concept pioneered by Melenk and Babuska [1996]. In the
following chapter, concepts behind the XFEM for treating discontinuities will be described.
In order to have a common terminology for the description of discontinuities, the following
section is dedicated to this cause.
23
Group B122b - Spring 2011 4. The eXtended Finite Element Method
Figure 4.1. Example of (a) open interface and (b) closed interface.
The interfaces in gure 4.1 dier by having and not having a free end in the domain.
Figure 4.1a is a so-called open interface, because the interface ends inside the domain.
An example of an open interface is a crack. Figure 4.1b is a so-called closed interface,
because the interface does not have any free ends inside the domain. An example of a
closed interface is a material interface. The topological dierence between an open and a
closed interface is described by the level-set method, described in section 4.2.
A distinction is made between moving and xed interfaces. A xed interface is treated
by a Lagrangian description, meaning the relative position of the interface is unchanged
during deformation of the body. A moving interface is treated by an Eulerian description,
24
4.2. Level-set method Master Thesis
meaning the interface moves through the domain. The initial position of the interface
is given, and the future position is then part of the solution. In this report, the cracks
appearing in the 3-dimensional concrete beam are xed interfaces. However, the crack
propagates and one would assume the interface to be moving. This is not the case, since
the crack propagation speed is unknown and not part of the solution in the displacement
variational principle. For this reason, the propagation of the crack is treated as a quasi-
static process and thus, as a xed interface.
A nal distinction is made between so-called strong and weak discontinuities. Strong
discontinuities refer to a jump in a eld quantity across an interface, whereas a weak
discontinuity refers to a jump in the gradient of the eld quantity across an interface. A
discontinuity in the gradient is also referred to as a kink in the eld variable. An example
of a strong and weak discontinuities is given in gure 4.2.
Figure 4.2. Example of (a) strong discontinuity and (b) weak discontinuity in a eld quantity
represented as a surface. The interface is represented as a bold line.
With the denition of strong and weak discontinuities, the displacement exhibits a strong
discontinuity across a crack and a weak discontinuity across a material interface. An
accurate description of the location of the interfaces, e.g. cracks, is necessary in order to
enrich the solution appropriately. This issue is addressed in the following section.
25
Group B122b - Spring 2011 4. The eXtended Finite Element Method
function and is easy to implement into a code. Examples of eligible level-set functions for
a one-dimensional bar with a discontinuity located at x = 0 are shown in gure 4.3.
Figure 4.3. Eligible level-set functions. The red line is the signed distance function and the
black line is an arbitrary level-set function. The interface is located at the red circle
on a one-dimensional bar discretized with nodes indicated as blue stars.
The signed distance function is given by equation 4.1. As the name suggests, the signed
distance function computes the distance from the discontinuity to a given point and assigns
a sign to the distance.
φ(x) = ± min
∗
kx − x∗ k , ∀x ∈ Ω (4.1)
x ∈ Γ
The sign in equation 4.1 is determined by the sign-equation sign(n · (x − x∗ )), where n
is the normal vector to Φ given by n = k∇Φk
∇Φ
, where ∇ is the dierential operator and
26
4.2. Level-set method Master Thesis
k∇Φk = 1 holds for the signed distance function. By convention, n points from the Φ-
negative subdomain into the Φ-positive subdomain, which is the reason for the existence
of the sign-equation.
X
Φh = Ni (x) · Φ(xi ) (4.2)
i∈I
Φh and Φ The approximated level-set function value and exact level-set function value.
x and xi Nodal coordinate and the coordinates of node i.
i and I Node i in the set of all nodes I .
Ni Standard Finite Element shape function belonging to node i.
As previously mentioned, the topological dierence between an open and a closed interface
is described by the number of level-set functions needed to describe the discontinuity. In
the following two subsections, open and closed interfaces are described using the level-set
method.
Γ12 = {x : φ(x) = 0}
This situation is depicted in gure 4.4, where the signed-distance function has been used
as the level-set function.
27
Group B122b - Spring 2011 4. The eXtended Finite Element Method
Figure 4.4. (a) The domain Ω is decomposed into two subdomains Ω1 and Ω2 . Γ12 describes
the interface. The normal vector, n, points from the φ-negative subdomain into the
φ-positive subdomain. (b) Contour values of the signed-distance function.
For more than 2 subdomains, 1 level-set function is no longer sucient. In general, for
closed interfaces, n level-set functions can separate 2n subdomains.
28
4.3. General formulation of the XFEM Master Thesis
Figure 4.5. (a) The domain Ω partially cut by a crack. (b) The signed-distance function φ
discribing the crack path. (c) The level-set function γ dening the crack tips.
X X X
uh (x) = Ni (x) · ui + Mi1 (x) · a1i + . . . + Mim (x) · am
i (4.3)
i∈I
| {z } i∈ I1∗ i∈ ∗
Im
| {z } | {z }
Strd. F EM approx. Enrichment 1 Enrichment m
29
Group B122b - Spring 2011 4. The eXtended Finite Element Method
Note that the approximation in 4.3 consists of a standard nite element part plus
additional m enrichment terms. This form of enrichment is called extrinsic enrichment,
because special enrichment terms are added to the polynomial approximation space,
resulting in more functions and nodal unknowns to be evaluated. The alternative to
extrinsic enrichment is intrinsic enrichment, where the standard nite element shape
functions are replaced by special shape functions, which are able to capture the non-
smooth solution. This is done by expanding the function basis for the elements cut
by a discontinuity, and results in no additional unknowns. However, the amount of
computational work needed to establish the special shape functions makes intrinsic
enrichment unappealing in comparison to extrinsic enrichment. In Abaqus 6.10 extrinsic
enrichment is used.
Each enrichment consists of a local enrichment function, Mim (x), and additional nodal
unknowns am i describing the character of the m'th discontinuity, e.g. if the m'th
discontinuity is a crack, the local enrichment function reects the inuence of the crack
by introducing a jump in the displacement eld. The local enrichment function is given
by equation 4.4.
The global enrichment function, ψ m (x), incorporates the special knowledge of the solution
properties into the approximation space. The partition of unity functions, Ni∗ (x), only
build a partition of unity in a local part of the domain, that is, in elements whose nodes
are all in the nodal subset I ∗ . In elements that are fully enriched the property of 4.5 holds,
that is, Ni∗ (x) build a partition of unity.
X
Ni∗ (x) = 1, ∀x ∈ Ω∗j , ∀j = 1, . . . , m. (4.5)
i∈Ij∗
The partition of unity concept is a well-known property of the standard nite element
shape functions. For this reason Ni∗ (x) = Ni (x) is an option, but not a necessity. For
30
4.4. Choice of enriched nodes Master Thesis
example, Ni∗ (x) can be chosen quadratic while using linear standard nite element shape
functions, Ni (x). In Abaqus 6.10 Ni∗ (x) = Ni (x).
The approximation on the form shown in 4.3 does not have the Kronecker-δ property.
This is seen by evaluating the approximation at x = xk
X X X
uh (xk ) = Ni (xk ) · uk + Mi1 (xk ) · a1k + . . . + Mim (xk ) · am
k (4.6)
i∈I i∈ I1∗ i∈ ∗
Im
X X
uh (xk ) = uk + Mi1 (xk ) · a1k + . . . + Mim (xk ) · am
k (4.7)
i∈ I1∗ i∈ ∗
Im
X X
uh (x) = Ni (x) · ui + Ni∗ (x) · [ψ(x) − ψ(xi )] · ai (4.8)
i∈I i∈ I1∗
31
Group B122b - Spring 2011 4. The eXtended Finite Element Method
I el is the set of element nodes. In other words, an element is cut if the level-set function
changes sign in the element.
For identifying an element containing a crack-tip the following two criteria of the signed
distance function must be met simultaneously:
The reason for identifying the crack tip element is to properly enrich it so that stress and
strain singularities are accounted for. However, if only one element is enriched sub-optimal
convergence rates are obtained. This is due to the fact, that mesh renement reduces the
area over which the singularity is accounted for. An alternative method that accounts for
this problem is the branch enrichment approach, where nodes within a certain radius from
the crack tip are enriched. In this way the enriched area is kept constant during mesh
renement.
∗
Itip = {i : kxi − xi ∗ k < r}
where kxi − xi ∗ k is the Euclidean distance between a point, xi , and the crack-tip, xi ∗ .
Abaqus 6.10 uses this approach, for static cracks, with an enrichment radius of three
times the characteristic element length. For evolving cracks the crack tip enrichment has
not been implemented in Abaqus 6.10. The mathematical description of the enrichment
functions used in Abaqus 6.10 are described in section 4.5.
Since only a subset of the nodes are enriched three types of elements can be dened:
32
4.5. Global enrichment functions Master Thesis
Figure 4.6. Discretized domains in one- and two dimensions with nodal subset I ∗ . (a) and (c)
show the reproducing-, blending- and standard elements. (b) and (d) show that the
function Ni∗ (x), here linear, only builds a partition of unity in reproducing elements
and varies linearly from one to zero over the blending element.
The fully enriched elements are called reproducing elements, because the approximation
shown in 4.3 is able to reproduce any enrichment function exactly in Ω. The partly
enriched elements are also referred to as blending elements because the enrichment is
blended over the element. This is because the partition of unity functions do not build
a partition of unity, which introduces parasitic terms into the approximation if linear or
higher order global enrichment functions are chosen. In Abaqus 6.10 blending elements
do not exist, because the meshing algorithm is constructed such that element boundaries
conform to interfaces. Moreover, only discontinuous enrichment functions, with constant
variation, are used, which eectively eliminates potential problems caused by blending
elements.
33
Group B122b - Spring 2011 4. The eXtended Finite Element Method
However, this type of enrichment function leads to trouble in blending elements. This is
because the function has a linear variation in the domain Ω. As previously mentioned,
troubles arise because the partition of unity function, N ∗ (x), does not build a partition of
unity in the blending elements. When ψ(x) is multiplied with N ∗ (x), a parasitic term is
introduced into the approximation. However, this is only the case if the order of N (x) and
N ∗ (x) is the same. For example, if both shape functions are chosen linear, the enrichment
term becomes parabolic because of the linear variation of ψ(x) and thus, a linear term is
summed with a parabolic term in the blending elements. In other words, the standard
FEM part cannot compensate for the error introduced by the parasitic term. A remedy for
this problem is to choose N (x) one order higher than N ∗ (x), in which case the standard
FEM part will be able to compensate for the error introduced in the approximation by
the parasitic term. An improvement to the abs-function was introduced by N.Moës and
Belytschko [1999]. The improved function is the so-called modied abs-enrichment, which
has the property of being non-zero only in the fully enriched elements. By being zero in
the blending elements, no parasitic terms are introduced into the approximation and thus
optimal convergence rates can be obtained. The modied abs-enrichemt is shown in 4.11.
X X
ψ(x) = |φi | Ni (x) − φi Ni (x) (4.11)
i∈I i∈I
Referring to the problem at hand in this project, a reinforced concrete beam, the
reinforcement and the beam are meshed as independent parts in Abaqus 6.10. This
entails that no elements contain more than one material property and thus, no weak
discontinuities are present in the model because the material interface is coincident with
the element boundaries. In the following section two global enrichment functions for strong
discontinuities will be presented.
34
4.5. Global enrichment functions Master Thesis
(
0 : φ(x) ≤ 0
ψ(x) = H(φ(x)) = (4.12)
1 : φ(x) > 0
−1 : φ(x) < 0
ψ(x) = sign(φ(x)) = 0 : φ(x) = 0 (4.13)
1 : φ(x) > 0
The gradient of these enrichment functions is zero. Note, that the functions do not cause
trouble in blending elements, because they are constant in Ω. In Abaqus 6.10 the jump-
function in 4.14 is used.
(
−1 : φ(x) < 0
ψ(x) = H(φ(x)) = (4.14)
1 : φ(x) ≥ 0
4.5.3 Singularities
At the crack tip a global enrichment function with a singular derivative is needed.
Moreover, the function must be discontinuous along the crack. In practice the four global
enrichment functions in equations 4.15 to 4.18 are often used.
√ θ
ψ 1 (x) = r sin (4.15)
2
√ θ
2
ψ (x) = r sin sin θ (4.16)
2
√ θ
ψ 3 (x) = r cos (4.17)
2
√ θ
4
ψ (x) = r cos sin θ (4.18)
2
The functions depend on a local polar coordinate system at the crack-tip, see gure 4.7,
where θ = 0 is tangent at the crack-tip.
35
Group B122b - Spring 2011 4. The eXtended Finite Element Method
Figure 4.7. The polar coordinate system around the crack-tip. xtip and ytip are the Cartesian
coordinates of the crack-tip. (r, θ) are the radial and angular coordinates,
respectively, from the pole to a point, (x, y).
The four global enrichment functions in equations 4.15 to 4.18 are a result of linear
elastic fracture mechanics, LEFM. They span the linear asymptotic crack-tip function
of elasto-statics, and 4.15 takes the discontinuity in displacement into account. These are
important in crack modeling, because if only sign-enrichment was used, the crack would be
virtually extended to the boundary of the element in which the crack-tip is present. Using
crack-tip enrichment functions ensures that the crack ends exactly at the location of the
crack-tip. Moreover, ψ 1 (x) to ψ 4 (x) are an analytical result from LEFM to the near tip
behavior, that is, the accuracy of the approximation is increased by including analytical
results in the approximation. Abaqus 6.10 takes advantage of these four functions in
representing the singular stress and strain eld near singularities. However, this is only the
case for stationary cracks, because accurate modeling of the crack-tip singularity requires
constantly keeping track of the crack location, and the degree of the singularity depends
on the location in non-isotropic material, e.g. concrete. Moving cracks are modeled with
the so-called cohesive segments method and phantom nodes. This matter is addressed in
the following.
36
4.7. Phantom-node method Master Thesis
described in the previous sections. The displacement jump across a crack is described
using the Phantom node method in Abaqus 6.10.
Figure 4.8. The principle of the Phantom-node method. On top of the nodes in the cracked
elements phantom nodes are added and the integration is performed over the
hatched area. Solid circles represent real nodes and hollow circles represent phantom
nodes. f (X) is the signed-distance level-set function evaluated at the Cartesian
coordinate X .
37
Group B122b - Spring 2011 4. The eXtended Finite Element Method
In the cracked elements phantom nodes are added on top of the real nodes and thus,
leading to an additional element on top of the cracked element. Each element consists of
a real subdomain and a phantom subdomain, e.g. Ω+ 0 and Ωp , where Ω0 = Ω0 ∪ Ωp with
− + −
reference to gure 4.8. Then the displacement eld in the real domain can be interpolated
using the degrees of freedom in the phantom domain. Initially the real node and the
phantom node are tied together. When cracking occurs, e.g. by fullling a damage
criterion, each phantom node and its corresponding real node are no longer tied together
and can move apart. The magnitude of the separation, i.e. the crack opening, is governed
by a cohesive law until the cohesive strength of the element is zero. This relation is
governed by the cohesive segments method. In Abaqus 6.10 the relation between crack
opening and traction is linear and given as input to the model by prescribing a fracture
energy. This is described in detail in chapter 5.
While the XFEM is suitable for modeling the singular stresses and strains around a crack-
tip, the Phantom-node method is only applicable to cohesive crack modeling. In this
fashion, the crack is extended to the element boundary and the singular eld is replaced
by a cohesive traction. However, the Phantom-node method has been extended to model
the crack-tip inside an element by T. Rabczuk and Wall [2008]. Abaqus 6.10 uses the
former approach, that is, the crack-tip always ends at an element boundary. This is a
simplied approach to crack modeling, as mesh-sensitivity is introduced for crude mesh
densities. For this reason the possibility of a precise evaluation of the crack-tip location
and propagation is limited.
Figure 4.9 illustrates the dierence between the XFEM and the Phantom-node method
for a one-dimensional bar with an inter-element discontinuity.
Figure 4.9. The interpolation basis of the XFEM vs. the Phantom-node method for a one-
dimensional element. T.Rabczuk et al. [2008]
38
4.8. Numerical integration of the weak form Master Thesis
Note that the displacement jump, [[u]], in gure 4.9 is identical for the two methods.
In the Phantom-node method the two elements representing the cracked element do not
share nodes and therefore have independent displacement elds. Both elements are only
partially active, which is represented numerically in the denition of the displacement
eld by introducing the jump function, see equation 4.14, which is active based on the
signed-distance level-set function. The displacement jump over a crack is then dened as
the dierence between the displacement elds of the two elements. The approximation of
the displacement, using the Phantom-node method, is given in equation 4.19.
X X
uh (X) = NI (x) · uI · H(f (X)) + NJ (x) · uJ · H(−f (X))
I ∈ {w0+ , wp− } J ∈ {w0− , wp+ }
(4.19)
where w0+ , w0− , wp+ and wp− are the nodes belonging to Ω+ −
0 , Ω0 , Ωp and Ωp , respectively. It
+ −
has been shown that the Phantom-node method is equivalent to the XFEM. An example
is the similarity between enriched degrees of freedom in the XFEM and the phantom
degrees of freedom in the Phantom-node method. From a computational point of view,
the Phantom-node method is superior to the XFEM, however more simple for previously
mentioned reasons, e.g. crack-tip position. Note, that the Phantom-node method is based
on the XFEM, in the sense that the standard FEM approximation space is enriched.
Moreover, both methods reect the jump in the displacement eld by piecewise integration.
Furthermore, both methods use the level-set method in the topological description of the
discontinuity and in the evaluation of the global enrichment functions. The numerical
integration method used in both methods is described in the following section.
1. Element decomposition.
2. Integrand transformation.
39
Group B122b - Spring 2011 4. The eXtended Finite Element Method
Figure 4.10. Element decomposition of a triangle into two sub-elements: A triangle and a
quadrilateral. Four Gauss points are placed in each sub-element.
Standard Gauss quadrature is then performed over the sub-elements, in which the
integrand is now smooth. Note that the sub-elements each have been assigned a number of
Gauss points, in this case four. The number of Gauss points and the type of sub-elements
used in Abaqus 6.10 are unknown to the authors. The crack is then an internal boundary
of the domain of integration. A mathematical description of the integration is given in
equation 4.20.
Z K Z
f (x)non-smooth dx = f (x)smooth dx
X
(4.20)
Ωe k=1 Ωke
The sub-division is done by non-overlapping elements that must conform to the same
requirement as the continuous problem. Note that element sub-division is not equivalent
to remeshing, as no additional degrees of freedom are introduced. The sub-elements are
only introduced for the purpose of integration. Moreover, the basis function is associated
with the nodes of the parent domain, which implies that no restriction is imposed on the
shape of the sub-elements.
Recall, that the discontinuity is dened as the zero-level of the interpolated level-set
function, see equation 4.2.
X
Φh = Ni (x) · Φ(xi ) = 0
i∈I
40
4.9. Governing equations Master Thesis
Dierent shape functions can be used for the interpolation than for the approximation
of the displacement. Linear shape functions are particularly useful, as the discontinuity
is a straight line in the reference element, and also when mapped in the real element.
This simplies the integration signicantly, as an element sub-division algorithm is more
easily constructed, than for a curved discontinuity. However, linear shape functions only
allows for a piecewise linear representation of discontinuities. Bi-linear shape functions
produce curved discontinuities, and an ecient element sub-division algorithm is dicult
to construct. For this reason, if bi-linear or higher order shape functions are used,
linearization of the discontinuity is often used. The discontinuity is linearized by drawing
a straight line between the points on the element edges that are cut by the discontinuity,
see gure 4.11.
Figure 4.11. (a) A curved interface in a bi-linear element can be (b) linearized by neglecting
the curvature of the interface or (c) the element is decomposed into two triangles
and linear interpolation is assumed.
The nodal values of the level set function are interpolated using the same shape functions as
in the approximation of the displacement. However, it is unknown whether a linearization
of the interface is performed. In either case, when the sub-division has been performed,
standard Gauss quadrature is adopted. The element decomposition approach is favorable
in computational implementation, because existing nite element integration schemes do
not need any modication. In the following section the weak form of the equilibrium
equations are presented. Systèmes [2010]
Z Z Z
T
δ σ dΩ = T
δu b dΩ + δuT t̂ dΓt , ∀δu (4.21)
Ω Ω Γt
41
Group B122b - Spring 2011 4. The eXtended Finite Element Method
Figure 4.12. Domain Ω supported on Γu and loaded on Γt . An internal crack is dened along
the boundary Γc .
σ · n = t̂ on Γt
σ · n = 0 on Γc
u = û on Γu
The two-dimensional nite element discretized form of the weak form is stated in 4.22.
fx tx
Z Z Z
fy ty
B T · Dep · B dΩ · u = dΩ + dΓ ⇔ K · u = f + ft (4.22)
Ω Ω 0 Γ0
0 0
where fx = Nstd · bx and fy = Nstd · by . bx and by are the body forces in the horizontal
and vertical direction, respectively. ft contains the tractions tx and ty in the horizontal
42
4.9. Governing equations Master Thesis
and vertical direction, respectively. Dep is the elasto-plastic constitutive matrix chosen
according to the adopted material model of the concrete, see chapter 5. The strain
distribution matrix B is given by equation 4.23.
T
∂x Nstd T
∂x Nenr
B= T
∂y Nstd T
∂y Nenr (4.23)
T
∂y Nstd T
∂x Nstd T
∂y Nenr T
∂x Nenr
where Nstd
T are standard FEM shape functions, and N T are the local enrichment functions
enr
given by equation 4.4. The strain distribution matrix is of dimension 3 × 2 · (nel + n∗el ), and
nel and n∗el are the numbers of element nodes and enriched element nodes, respectively.
43
Discontinuous modeling in
Abaqus
5
This chapter presents a summary of the methods described in chapters 2-4, used in
Abaqus 6.10 in relation to discontinuous modeling. With regard to the methods of
discontinuous modeling, preliminary choices made in Abaqus 6.10 for this report
regarding the element type used for the numerical discretization and the material
properties of the examined concrete and steel are presented. The choices are
common for the benchmark tests considering concrete and the reinforced concrete
beam considered in the problem formulation of this project.
In Abaqus 6.10 cracking in concrete and steel is modeled in a discrete fashion, that is, a
strong discontinuity is introduced in the displacement eld by the Phantom-node method,
in which the displacement jump is reproduced by introducing the jump function. For
the remainder of this report the Phantom-node method will be referred to as the XFEM
because the methods are equivalent. The crack is represented as an open interface, but
since the crack is always extended to the boundary of the element, in which it is present,
only one level-set function is used for the topological description of a crack. The level-
set function is chosen as the signed-distance function and is interpolated using the same
interpolation functions as the approximation of the displacement. The crack is modeled
by inserting a cohesive segment in the cracked element. The cohesive segments method
is based on the ctitious crack model by A. Hillerborg and Peterson [1976], that is, non-
linear fracture mechanics is used. The adopted crack initiation criterion is the maximum
principal stress criterion, in which a crack is initiated if the maximum principal stress
reaches the tensile strength of the concrete. The crack propagates perpendicular to the
direction of the maximum principal stress. The evolution of the crack is governed by the
fracture energy, which represents the tension-softening behavior of the concrete during
cracking. The relationship between the crack opening displacement and the closing stresses
acting on the crack is linear, in which case the critical crack opening displacement, wc , is
45
Group B122b - Spring 2011 5. Discontinuous modeling in Abaqus
Figure 5.1. Linear relationship between the closing stress and crack opening displacement upon
cracking.
Figure 5.2. The modeling concept used in Abaqus 6.10 to model cracks in concrete.
The constitutive behavior of concrete is modeled using the so-called Concrete Damaged
Plasticity material model, abbreviated CDP. In tension the CDP material model is used
in coordination with the XFEM and the cohesive segments method, that is, CDP is used
until tensile crack initiation is detected, at which point a cohesive segment is inserted and
the XFEM is activated. For compression the CDP material model is used without the
XFEM. The CDP material model is a continuum-damage based constitutive relation. The
stress-strain relation is given by equation 5.2.
46
Master Thesis
The scalar degradation variable has an initial value of zero for intact material and increases
towards one for complete loss of material stiness. Based on the observation that concrete
is anisotropic, the scalar degradation variable is dierent in tension and compression. Since
the scalar degradation variable is not used for tension, it will not be further discussed
in this report. The compressive scalar degradation variable is determined according to
equation 5.3.
dc = 1 − σc /fc (5.3)
Unless stated otherwise the concrete examined in this report has the following material
properties:
√
The ultimate tensile strength is determined by ft = fcm · 0.1.
Abaqus 6.10 requires a specication of the uniaxial relationship between the inelastic
strain and the stress after yielding has occurred. The uniaxial behavior is then extended
to multiaxial directions. The uniaxial tensile and compressive response of the examined
concrete is shown in gures 5.3 and 5.4, respectively.
Figure 5.3. Response of concrete due to uniaxial tension. The blue circles represent the values
used in Abaqus 6.10.
47
Group B122b - Spring 2011 5. Discontinuous modeling in Abaqus
The blue circles on the uniaxial tensile response in gure 5.4 are used as input in Abaqus
6.10 to describe the tensile behavior of the concrete. However, Abaqus 6.10 requires the
cracking strain versus stress. At the onset of cracking, the cracking strain is equal to zero
and the total strain is equal to ft /Ecm = 5.24 · 10−5 . The exact values given as input in
Abaqus 6.10 are given in table 5.1. More details regarding the stress-strain relations of
the CDP material model are shown in appendix A.
Figure 5.4. Reponse of concrete due to uniaxial compression. The blue circles represent the
values used in Abaqus 6.10.
The non-linear compressive reponse shown in gure 5.4 is based on an empirical stress-
strain relation given in EN 1992-1-1 [2004]. The blue circles on the uniaxial compressive
response in gure 5.4 are used as input in Abaqus 6.10 to describe the compressive behavior
of the concrete. Note that the points shown in gure 5.4 are not directly used, as Abaqus
6.10 requires the specication of the inelastic strains. For this reason the elastic strain at
each point has been subtracted from the total strain. The exact values are given in table
5.2.
48
Master Thesis
kη − η 2
σc = · fcm (5.4)
1 + (k − 2)η
η c /c1
k 1.05 · E |fc1c |
c1 = 0.22% is the strain corresponding to the ultimate compressive strength for the
examined concrete in this report according to EN 1992-1-1 [2004]. As previously
mentioned, the uniaxial behavior of concrete is extended to multiaxial directions in
49
Group B122b - Spring 2011 5. Discontinuous modeling in Abaqus
Tensile cracking is initiated for a principal tensile strain corresponding to a principal tensile
stress equal to ft . Similarly, compressive elastic degradation is initiated for a principal
compressive stress of fc . At the stage of the loading process where a tensile crack forms
a discontinuity in the displacement eld of the cracked element is formed. Upon crack
initiation a cohesive segment is inserted and the XFEM is activated in the cracked domain.
In the uncracked domain solely CDP is used. Unless stated otherwise, the steel examined
in this report has the following material properties:
The element type used for the numerical discretization in all benchmark tests and in
the three-dimensional reinforced concrete beam, investigated in this report, is the C3D8
element from the Abaqus 6.10 library. The C3D8-element is a linear, 8-node, isoparametric
three-dimensional hexahedron with full integration. The elements are integrated using
eight integration points. The C3D8-element is shown in gure 5.5.
50
Master Thesis
Figure 5.5. 2 × 2 × 2 integration point scheme in a C3D8-element. The node numbering and
integration point numbering follows the convention in Abaqus 6.10. The integration
points are shown in bold dots. MIT [2011]
The implementation of the XFEM in Abaqus 6.10 will be veried in the following chapter
based on the modeling choices made in this chapter.
51
Verification of the XFEM
and Abaqus
6
This chapter presents four dierent benchmark tests used to evaluate dierent
aspects of the eectiveness of the XFEM tool and the CDP material model
available in Abaqus 6.10. A description of the individual benchmark tests are
given with respect to model setup and the available basis for a comparison, such
as existing analytical, experimental or numerical results for the given problem.
Finally the quality of the obtained results is used to assess the quality of the
XFEM implementation and the CDP material model in Abaqus 6.10.
In this chapter three benchmark tests are evaluated using the XFEM implementation in
Abaqus 6.10, and one benchmark tests is carried out to verify the compressive behavior
of the examined concrete using the CDP material model. Benchmark tests are critical to
the understanding of the capabilities of the XFEM implementation and the CDP material
model in Abaqus 6.10, because they serve as a source for comparison. Moreover, all
examined benchmark tests are simple in geometry and loading. This makes the numerical
model computationally light, which is convenient in order to verify the XFEM and the
CDP material model eciently. Finally the authors believe, that since a learning process
is involved, it is advantageous to start with a simple, well-dened benchmark problem
before resorting to advanced, three-dimensional modeling of a reinforced concrete beam
with more uncertainty in the outcome of the analysis.
1. Crack-hole interaction for studying the inuence of a hole on the crack propagation
path in a steel plate.
2. Verication of the compressive behavior of concrete using the CDP material model.
3. Crack propagation in a concrete beam in three point bending.
4. Crack formation analysis of a reinforced concrete plate.
The benchmark test introduced in point 1 above deal with analyzing a two-dimensional
steel plate with isotropic and linear elastic material behavior. The reason for analyzing
steel is to start out with a simple material model before analyzing more complex materials,
53
Group B122b - Spring 2011 6. Verication of the XFEM and Abaqus
e.g. concrete, that requires a more sophisticated material model. Moreover, a non-linear
material behavior entails that the weak form of the XFEM is described in an incremental
form, putting higher demands on the quality of the implementation of the numerical solver
in Abaqus 6.10. The material behavior of steel is described using a linear elastic perfectly
plastic material model. The material properties of the steel in test 1 and 2 are shown in
table 5.
This is followed by the benchmark tests introduced in point 2, 3 and 4, where concrete
material properties are implemented. The material properties of the concrete in test 2
and 3 are dierent from the properties of the concrete in test 4, and are thus described in
the introduction to the respective benchmarks. The CDP model described in appendix A
is used in the context of describing the non-linear behavior of concrete.
All benchmark tests are discretized using the C3D8 element described in section 5.
The C3D8 element is a three-dimensional element and is used for four reasons: rstly,
Abaqus 6.10 does not oer the XFEM tool for analyzing non-linear material models with
planar elements. Secondly, only linear continuum elements are allowed when the XFEM
is activated. Thirdly, in order to use the same element in all benchmark tests, thus
eliminating a potential variable between the benchmarks on steel and concrete. Finally, the
quality of the C3D8 element is tested as a candidate for modeling of the three-dimensional
reinforced concrete beam. Although three-dimensional elements are used, the thickness of
the specimens in benchmark test 1, 2 and 3 is irrelevant for the results.
54
6.1. Crack-hole interaction Master Thesis
Figure 6.1. Load and boundary conditions applied to the steel plate model.
Figure 6.2 shows the numerical results obtained by G. Ventura and Belytschko [2003].
The crack is clearly aected by the presence of the hole. It propagates towards the hole
and ends on the periphery of the hole. The numerical results are shown in gure 6.3, and
are seen to be similar with the dierence being the smoothness of the crack paths. It
is observed, that the crack path becomes less aected by the hole the further the initial
crack is placed from the hole.
Figure 6.2. Numerical results of crack paths in a steel plate obtained by G. Ventura and
Belytschko [2003].
55
Group B122b - Spring 2011 6. Verication of the XFEM and Abaqus
Figure 6.3. Crack propagation paths for three dierent initial crack locations: A: 75 mm, B:
150 mm and C: 225 mm are the distances from the center of the hole to the crack.
The results show that Abaqus 6.10 is capable of modeling crack propagation with respect
to a change in geometry, such as the hole in the benchmark problem, without the need
for remeshing.
Figure 6.4. Dimensions in mm and boundary conditions for a section of the examined cylinder.
Note that, due to symmetry about the z-axis, only a section of the cylinder is modeled.
The section is supported from moving in the longitudinal and horizontal direction in the
way shown in gure 6.4. The loading is displacement based in order to get the post-peak
response of the stress-strain relation as output. The stress-strain relation is requested in
an arbitrary integration point. The location of the integration point is irrelevant because
56
6.3. Crack propagation in a concrete beam in three point bending Master Thesis
the stress-state is uniform throughout the cylinder. Figure 6.5 shows the stress-strain
relation given as output from Abaqus 6.10. The gure also compares the relationship to
the expected relationship from EN 1992-1-1 [2004].
Figure 6.5. Stress-strain relation for a cylinder loaded in compression calculated in Abaqus 6.10
and according to EN 1992-1-1 [2004].
The compressive behavior is as expected, and for this reason the implementation of the
CDP material model in Abaqus 6.10 is acceptable. The tensile behavior is examined in 7
57
Group B122b - Spring 2011 6. Verication of the XFEM and Abaqus
Figure 6.6. The setup of a notched beam in three point bending. Numerical and experimental
results of the crack propagation path are compared.
Figure 6.7. Experimental results obtained by J.Davies [1996] for a notched beam in three point
bending.
The intention is to obtain numerical results similar to the experimental work performed
by J.Davies [1996]. The crack propagation path observed in the experiment is the focus
of the analysis. The microcracks developing in branches from the main crack are not
possible to model in Abaqus 6.10, because cracks are not allowed to branch in the
current implementation of the XFEM. Furthermore the inhomogeneous distribution of
the aggregates in the concrete cause crack formation such as illustrated in gure 6.8. This
is not to be expected in the numerical analysis.
58
6.3. Crack propagation in a concrete beam in three point bending Master Thesis
The load is applied as a pressure, see gure 6.9. This is preferable to the 'Concentrated
Force' option in Abaqus 6.10, in order to ensure that stress concentrations at the load
surface do not occur. To prevent crushing at the supports, the beam is supported by a
xed and a rolling 15 mm wide steel block, see gure 6.9.
Figure 6.9. The modied loading and boundary conditions for the three point bend beam model
in Abaqus 6.10.
The procedure is to ensure that the applied pressure and the resulting pressure at the
supports do not exceed the prescribed compressive yield stress of the concrete. The global
load-displacement curve has been plotted in gure 6.10 to illustrate, where the numerical
59
Group B122b - Spring 2011 6. Verication of the XFEM and Abaqus
solver fails to obtain equilibrium. The load-displacement curve looks reasonable, as failure
occurs abruptly upon reaching the maximum load carrying capacity of the plain concrete.
Figure 6.10. Load-displacement curve for the three point bend beam.
Unlike the previous models an initial crack is not inserted prior to the loading, since the
purpose is to determine whether Abaqus 6.10 can correctly initiate crack formation. By
correct initiation the authors allude to crack formation at the location of the notch, as
observed in the experiment by J.Davies [1996].
In the following the beam will be modeled with dierent mesh structures with the purpose
of identifying a quality mesh from a poor mesh.
Figure 6.11 shows a poorly constructed mesh, in the sense that the mesh is unstructured
and unsymmetrical around the notch. The crack propagates from one side of the centerline
of the notch to the other. Compared to the experimental results in gure 6.7 the
propagation path is not similar. It can be observed that the initial crack propagation
is skewed to the right. The experimental results indicate that the crack should propagate
straight up from the notch. However, due to the fact that an intersection of two elements
lies in that path, Abaqus 6.10 cannot simulate a straight crack path from the notch, hence
the skewness to the right.
60
6.3. Crack propagation in a concrete beam in three point bending Master Thesis
Figure 6.11. A poorly constructed mesh of a notched beam in three point bending. The gure
shows the area around the notch where a crack is formed. The crack propagation
path is inconsistent with experimental results.
In gure 6.12 the mesh has been made more structured around the notch. The crack
path is straighter but the initial crack path is incorrect due to the asymmetric element
distribution.
Figure 6.12. An average constructed mesh of a notched beam in three point bending. The gure
shows the area around the notch where a crack is formed. The crack propagation
path is not correct despite being improved.
Figure 6.13 shows a structured and symmetrical mesh. The crack path is very similar to
the experimental path in gure 6.7. However the small branches of cracks in gure 6.7
cannot be modeled. Moreover, the crack path does not deviate from a straight line as
seen in the experiment by J.Davies [1996]. This is due to the inhomogeneous distribution
of aggregates in concrete, which cannot be implemented into the numerical model. This
issue was illustrated in gure 6.8.
61
Group B122b - Spring 2011 6. Verication of the XFEM and Abaqus
Figure 6.13. A well constructed mesh of a notched beam in three point bending. The gure
shows the area around the notch where a crack is formed. The crack propagation
path is similar to the results of the experiment conducted by J.Davies [1996].
In conclusion, the importance of the mesh structure depends on the problem being
investigated. If a straight crack path is desired, e.g. in the analysis of pure mode I fracture,
a symmetric, structured mesh is suggested. However, when analyzing crack propagation
in heterogeneous materials like concrete, where cracking is arbitrary, a structured mesh
is not an advantage with respect to realistic crack propagation paths. In other words,
the randomness of the aggregate distribution can be accounted for by using a random
distribution of elements. Furthermore it is apparent that the XFEM tool in Abaqus
6.10 can model crack propagation of geometrically simple concrete specimens in three
dimensions with the desired precision.
62
6.4. Crack formation analysis of a reinforced concrete plate Master Thesis
Figure 6.14. The boundary conditions and applied loads to the reinforced concrete plate model
in Abaqus 6.10.
The concrete and reinforcement used in the analysis have characteristic values shown in
table 6.1. The reinforcement is embedded into the concrete, which is an option in Abaqus
6.10 that implies innite bond strength at the interface between the concrete and the
reinforcement. Despite the fact that these conditions are not optimal, the authors were
not able to successfully model the bond properties at the interface in Abaqus 6.10 any
dierent. The characteristic compressive strength is obtained from S.B Bhide and M.P
Collins [1989]. Information regarding the modulus of elasticity and Poisson's ratio are not
described in S.B Bhide and M.P Collins [1989]. As a result these material properties have
been assigned the values shown in table 6.1. The CDP material model is used for this
benchmark problem. The compressive behavior of the concrete is described in 15 points
along the curve in gure 6.15. The curve is constructed based on EN 1992-1-1 [2004]. The
√
ultimate tensile strength is determined based on ft = fc · 0.1. The stress-strain behavior
in tension is modeled as linear between 0 M P a − 1.52 M P a. The maximum principal
stress damage criterion is used with a value of ft as the maximum principal stress at
cracking.
63
Group B122b - Spring 2011 6. Verication of the XFEM and Abaqus
Figure 6.15. The stress-strain relation used to describe the compressive behavior of the concrete
for this benchmark problem.
The purpose of the benchmark is to compare the crack formation in the concrete plate
at failure to experimental results obtained by S.B Bhide and M.P Collins [1989]. The
formation of cracks perpendicular to the direction of the applied load are expected. These
cracks are so-called dilatational cracks, which occur because of volumetric expansion of
the concrete. Figure 6.16 compares the crack formation in the concrete plate obtained
from Abaqus 6.10 with the experimental results. The crack formation is similar, however
more structured in the Abaqus 6.10 model. This is because Abaqus 6.10 cannot account
for the inhomogeneous distribution of the aggregates in the concrete.
64
6.5. Concluding remarks Master Thesis
Figure 6.16. A comparison of crack formation at failure for a 890 mm × 890 mm × 70 mm plate.
Right: results obtained in Abaqus 6.10. Left: results obtained from an experiment
by S.B Bhide and M.P Collins [1989].
It is observed from gure 6.16, that Abaqus 6.10 is able to represent the dilatational
response.
Abaqus 6.10 is capable of accounting for an advanced geometry. With the XFEM
implementation, interelement discontinuities are possible and cracks can be allowed
to propagate, as indicated in the crack-hole analysis.
Abaqus 6.10 reects the intended stress-strain relation in concrete when loaded in
compression and using the CDP material model.
Abaqus 6.10 is capable of using the XFEM in conjunction with analysis of crack
propagation in non-linear materials. Crack initialization and propagation according
to a failure criterion are working options in Abaqus 6.10. A structured mesh is
not necessarily an advantage when dealing with concrete due to the inhomogeneous
aggregate distribution, as indicated in the three point bend test.
Abaqus 6.10 is capable of modeling composite materials, e.g. reinforced concrete,
by assuming innite bond strength between the reinforcement and concrete.
65
Results and discussion of
3d beam analysis
7
This chapter presents the results obtained from analyzing a three-dimensional
beam loaded in three point bending. Dimensions, boundary conditions and
material properties are specied for the concrete and the reinforcing steel used for
the beam. The adopted material model and the modeling procedure in Abaqus
6.10 are described. This is followed by an introduction to, and the motivation
for, the studies performed on the beam. Finally, the results are presented and
discussed.
67
Group B122b - Spring 2011 7. Results and discussion of 3d beam analysis
The analytical models used for the deection comparison is based on Bernoulli-Euler the-
ory, and therefore the dimensions of the beam are chosen such that the beam primarily
deects due to bending. The material properties of the concrete and the reinforcement
are shown in tables in chapter 5, and are repeated below for convenience.
Note, that the mean values of the concrete strength parameters are used, because the
focus of the report is not to design the beam, but to obtain a realistic estimate of the
crack widths, crack spacing and crack pattern. Moreover, a realistic estimation of the
deection, crack widths, crack spacing etc. is of interest, rather than an estimation which
is inuenced by a safety factor.
The total area of the provided reinforcement falls within the limits of the minimum
and maximum reinforcement areas according to EN 1992-1-1 [2004]. For this reason
the reinforcement ratio is normal, and the assumption, in the ultimate limit state, of
simultaneous tensile yielding in the reinforcement and compressive failure in the concrete
is not violated. This is desired when comparing the cross-sectional stress distribution
determined in Abaqus 6.10 to the expected analytical distribution, which assumes normal
reinforcement ratio in the beam. The calculation of the reinforcement ratio and the
minimum and maximum limits can be found in appendix B.3. In the following section
the numerical model of the beam in Abaqus 6.10 is described with respect to boundary
conditions, input for the material model of the concrete and the steel and modeling of the
reinforcement.
68
7.1. Model setup Master Thesis
Figure 7.2. The condition at one of the supports on the beam. The beam is supported along a
line parallel with the x-axis, and is extended by 100 mm from the support.
The X, Y and Z direction in table 7.1 are shown in gure 7.1. The load is deection
controlled and applied at the top midpoint of the beam over an area of 250 mm×250 mm;
the top midpoint being the center of deection. In this way the simulation can continue
after the ultimate load has been reached. The load could not be applied along a line, as
the supports, because localized crushing of the concrete was observed causing the analysis
to abort prematurely.
69
Group B122b - Spring 2011 7. Results and discussion of 3d beam analysis
As described in chapter 5, the tensile behavior in the model combining the CDP model
with the XFEM is described without any tension stiening because the Phantom-node
Method is activated at the onset of cracking. For the model without the XFEM, that is,
solely CDP, the tension stiening eect could not be modeled. The authors have attempted
to model the tension stiening eect but without success. Since it is not within the scope
of this report to model tension stiening using the CDP model the issue is not pursued
further. The maximum principal stress criterion, MaxPS, for traction separation laws is
used in the model, with the sub option Damage Evolution activated, in Abaqus 6.10. The
used values are given in table 7.2
In the CDP model four parameters control the evolution and the shape of the yield surface
and the ow potential. These are outlined in the following. The dilation angle, ψ , is
included in the ow potential and can be visualized in the p−q plane, under high conning
pressures, as the angle between the direction of the plastic strain increment and vertical,
see gure 7.3. p and q are the hydrostatic and the deviatoric stress tensors, respectively.
A value of ψ = 38◦ is chosen inspired by Jankowiak and Lodygowski [2005].
Secondly, the so-called eccentricity, , denes the rate at which the Drucker-Prager
hyperbolic ow function approaches the asymptote. The ow potential tends to a straight
line as the eccentricity tends to zero. The asymptote corresponds to a linear Drucker-
Prager model, see gure 7.3. The default value is 0.1, which is accepted in this report.
Note that the friction angle is not given as input to the CDP model. The determination
of the friction angle, β , is further described in appendix A.
70
7.1. Model setup Master Thesis
Figure 7.3. The Drucker-Prager hyperbolic function in the p − q plane. The Drucker-Prager
linear function is shown with dashed lines. β is the friction angle and ψ is the
dilation angle.
The third parameter of inuence is the ratio of biaxial compressive yield stress, fcb , to the
uniaxial compressive yield strength, fc0 , that is fcb /fc0 . The ratio is important for the
shape of the yield surface in the principal stress space, as it determines the location of
the point of biaxial compression, see gure A.1 in appendix A. The default value is 1.16,
which is accepted in this report, and corresponds to a ultimate biaxial compressive yield
strength of 38 M P a · 1.16 = 44.08 M P a.
The last parameter of inuence on the yield surface is K , which describes the ratio of
the second invariant on the tensile meridian, to that on the compressive meridian. The
default value is K = 2/3, which is accepted in this report. The inuence of K on the
yield surface in the deviatoric plane is illustrated in gure 7.4, where an additional yield
surface for K = 1 is plotted for comparison.
71
Group B122b - Spring 2011 7. Results and discussion of 3d beam analysis
Figure 7.4. Yield surfaces in the deviatoric plane. (T.M.) refers to tensile meridian and (C.M.)
refers to compressive meridian. Si is the i'th principal stress, i = 1, 2, 3.
The four nal values of the parameters used in Abaqus 6.10 are summarized in table 7.3.
fcb /fc0 ψ K
1.16 38◦ 0.1 2/3
Table 7.3. The values of the four nal values of the parameters used in Abaqus 6.10.
72
7.1. Model setup Master Thesis
Figure 7.5. The employed stress-strain relationship for the reinforcing steel.
The wires only add stiness to the host elements in which they are located. Figure 7.6
shows an embedded wire superposed on a host element.
73
Group B122b - Spring 2011 7. Results and discussion of 3d beam analysis
A study of the load-deection curve of the beam using the CDP material model
with and without the XFEM. This is done in order to investigate the inuence of
the activation of the XFEM. An analytical expression from EN 1992-1-1 [2004] for
the load-deection curve serve as a guideline for the expected result.
A study of the stress distribution along the beam height in a section located at
the midspan of the beam. The stress-distribution is plotted at various load-stages.
The stress-distribution is plotted for two models; with and without the XFEM. The
stress-distributions are compared to the theoretical distribution and cross-reference
is made to the yield surface.
A study of crack width, spacing and formation using the CDP material model and
the XFEM in the serviceability limit state. Crack widths, spacing and formation
are examined for a deection of 1/250'th of the beam span, equal to 20 mm, and
compared to guidelines by EN 1992-1-1 [2004]. A deection of 1/250'th of the
beam span is the maximum allowable deection in the serviceability limit state
suggested by EN 1992-1-1 [2004]. Note, that the ultimate limit state is not of the
same importance as the serviceability limit state when crack width, spacing and
formation are analyzed. Crack widths are of importance for the durability of the
beam and therefore the cracks caused by the service load should be analyzed.
The main purpose of the two rst analyses in the list of studies is to verify the the CDP
material model and the XFEM for a three-dimensional beam. The nal analysis in the list
of studies is performed to answer the problem formulation of the report. In the following
section the results from the above mentioned list of studies will be presented and discussed
in the mentioned order.
74
7.3. Results and discussion Master Thesis
5. A numerical solution from Abaqus 6.10 where the Concrete Damaged Plasticity
model is used in combination with the XFEM.
P · l3
uuncr = (7.1)
48 · Ecm · Ig
P · l3
ucr = (7.2)
48 · Ecm · Icr
where the moment of inertia for a fully cracked section, Icr , is given by Icr = b · x3 /12 +
n · As · (d − x)2 . The equation for determining Icr assumes a fully cracked section and that
the concrete in the compression zone behaves linear elastic until the steel is yielding. The
equation for Icr is established based on gure 7.7.
75
Group B122b - Spring 2011 7. Results and discussion of 3d beam analysis
The load-deection equation for point 3 above, suggested in EN 1992-1-1 [2004], is given
by
2 ! 2
Mcr Mcr
u = ucr · 1− + uuncr · (7.3)
Ma Ma
fr ·Ig
where the cracking moment, Mcr , is given by Mcr = h−x .
Figure 7.8. A comparison of the load-deection curves for two analytical solutions, the beam
assumed uncracked and fully cracked, a solution recommended in EN 1992-1-1
[2004] and the two numerical solutions from the XFEM and CDP models to the
displacement of the beam as a function of the load.
The analytical displacement curves end when the ultimate load is reached. The ultimate
load is calculated by determining the moment of resistance, Mr , see equation 7.4, and
isolating Pult in Mr = Pult ·l
4 . The ultimate load is found to be Pult = 262.8 kN .
Equation 7.4 is established based on equilibrium between the tensile and the equivalent
rectangular compressive stress distribution in a fully cracked cross-section, see gure 7.9.
76
7.3. Results and discussion Master Thesis
The rectangular compressive stress distribution is assumed equivalent with the compressive
stress distribution shown in gure 7.11.
Figure 7.9. Equivalent stress distribution at ultimate load in a fully cracked cross-section.
4 4
Mr = d − x b xfc (7.4)
10 5
Note that equation 7.3 assumes, that the beam will crack but not fully crack, and will thus
behave in a manner intermediate between the uncracked and fully cracked conditions. The
displacement based on equation 7.3 shows a kink at a load of Pcr = 14.78 kN , which is the
analytical value for the load causing crack initiation. The load has been determined by
evaluating the cracking moment and isolating P in the equation for the midspan moment.
The displacement kink at the cracking load is better visualized in gure 7.10. In the
numerical results the cracking load was found to be Pcr = 16.90 kN , which is in close
agreement with the analytical value.
77
Group B122b - Spring 2011 7. Results and discussion of 3d beam analysis
Figure 7.10. Close-up of the load-deection curves from gure 7.8 illustrating the displacement
kink represented by equation 7.3 at a load of 14.78 kN . A displacement kink is
present at a load of 16.90 kN in the numerical models.
Note, that the load-displacement curves with and without the XFEM are nearly coincident
up to a level of the ultimate load. This illustrates, that the activation of the XFEM does
not inuence the load-deection behavior. With this observation it can be concluded that,
if a deection or ultimate load capacity estimation is desired, the XFEM does not need
to be activated. Recall, that the XFEM introduces additional degrees of freedom to the
system, which increases the computational eort.
As seen from gure 7.8, the load-displacement curves from Abaqus 6.10, with and without
the XFEM, falls within the analytical limits for deections smaller than approximately
25 mm. This does not imply that the numerical model is inaccurate in the ultimate
limit state. However, the purpose of this report is to investigate crack widths, patterns
and spacing in the serviceability limit state and compare the results to guidelines from
the EN 1992-1-1 [2004]. For the sake of this purpose, it is convenient to have agreement
between the load-deection curves using the XFEM in Abaqus 6.10 and equation 7.3.
The load-displacement curves shown in gure 7.8 are for converged results. The
convergence test is based on a uniform h -renement, that is, the element size is varied in
the same proportion in the entire mesh. Convergence tests on the ultimate load and the
corresponding displacement are performed in the following section.
78
7.3. Results and discussion Master Thesis
Figure 7.11. Principle sketch of the stress development at increasing bending moment.
Note that the distribution of the compressive stresses is non-linear close to failure. The
purpose of investigating the cross-sectional stress distribution in the beam is to determine
the validity of the CDP material model and the XFEM. The success criterion in this
analysis is to observe a stress development in a cracked section of the beam in Abaqus
6.10 similar to the theoretical development in the distribution shown in gure 7.11. The
numerically obtained cross-sectional stress-distributions, with and without the XFEM in
Abaqus 6.10, are shown in gure 7.12.
79
Group B122b - Spring 2011 7. Results and discussion of 3d beam analysis
Figure 7.12. A comparison of the stress distribution in a cracked section of the beam. The
black line represents the stress-distribution without the XFEM and the blue line
represents the stress-distribution with the XFEM model. The arrow represents
the direction of the tensile force in the 4 steel reinforcement bars. The load stages
are shown in gure 7.13.
Figure 7.13 shows the load stages of gure 7.12 on a load-deection curve.
80
7.3. Results and discussion Master Thesis
Figure 7.13. Stages 1-6 of gure 7.12 illustrated with points on the load-deection curve from
models with and without the XFEM.
Note, that for all load-stages the concrete carries tension if the XFEM is not activated.
Stage 1: Initially, the stress-distribution is linear. The tensile stresses in the extreme
tension bers are close to the tensile strength of the concrete (1.95 M P a).
Stages 2-4: The compressive stress distribution in the concrete is linear and the
tensile stresses are constant and equal to the tensile strength of the concrete without
the XFEM, and zero with the XFEM because the section is cracked.
Stage 5: The stresses in the midspan section have reached the uniaxial compressive
stress limit, causing crushing in the concrete, aecting the stresses in the
neighboring, cracked section. The concrete exhibits compression softening.
Stage 6: Crushing has occurred in the midspan section, resulting in a stress reduction
in the adjacent, cracked cross-section. The eective load carrying area is reduced due
to crushing according to the CDP material model. Crushing has not yet occurred
in the part of the section, where the largest compressive stresses are present.
The stress-distributions presented above are for converged results. The convergence test
is based on a uniform h -renement and was performed to study the convergence of the
ultimate load and the corresponding deection. The results are shown in gures 7.14
and 7.15, for the ultimate load and the corresponding deection, respectively. The stress-
distributions are plotted for the nest mesh using reduced integration.
81
Group B122b - Spring 2011 7. Results and discussion of 3d beam analysis
Figure 7.14. Convergence study on the ultimate load for the numerical models. The bold line
represents the analytically determined ultimate load.
Figure 7.15. Convergence study on the displacement at the ultimate load level for the numerical
models. The bold line represents the analytically determined deection at a level
of the ultimate load.
Recall, that the analytical ultimate load is determined using equation 7.4, and that the
analytical ultimate deection is determined using equation 7.3.
82
7.3. Results and discussion Master Thesis
The convergence tests show, that the results, with and without the XFEM, converge using
approximately 2800 elements. Note that reduced integration when the XFEM is used is not
possible, because the results do not converge. Generally the stiness of the beam in Abaqus
6.10 is underestimated when compared to the analytical estimations of the deection. A
factor contributing to the larger displacements observed using Abaqus 6.10 is that the
displacement contribution from shearing strains is neglected in the analytical equations,
which is based on Bernoulli-Euler beam theory. This contribution has been determined to
constitute 1, 5% of the total deection, see appendix B.5. However, the main contributing
factor to the dierence in displacement is the cross-sectional stress distribution. As
previously mentioned, the analytical equation for the displacement assumes the cross-
sectional stress distribution in gure 7.7, which is an elastic distribution. Moreover, the
analytical equation for the ultimate load assumes the cross-sectional stress distribution
in gure 7.9, which is a plastic distribution. This could explain why the ultimate load
is more accurately approximated than the deection corresponding to the ultimate load,
because the observed stress distribution at the ultimate load level, see stage 5 in gure
7.12, is plastic and matches the plastic stress distribution shown in gure 7.9.
An additional convergence test has been carried out for a load corresponding to an
analytical deection of 20 mm, see gure 7.16. This load has been determined to be
202.23 kN . The convergence test is based on a uniform h -renement and was performed
to study the convergence of the deection in the serviceability limit state.
Figure 7.16. Convergence study on the displacement at the serviceability limit state for the
numerical models. The bold line represents a deection of 20 mm.
The ultimate load converges to the analytically determined resistance of the beam. The
overestimation in the deection of the beam is approximately 10% in the ultimate limit
state. The deection is underestimated by approximately 7% in the serviceability limit
83
Group B122b - Spring 2011 7. Results and discussion of 3d beam analysis
In conclusion the results are satisfying. The XFEM has been successfully applied to the
CDP material model. The load-deection relation is unaected by the activation of the
XFEM. The same conclusion applies for the estimation of the ultimate load capacity and
the corresponding deection. It is dicult to state a conclusion on the cross-sectional
stress-distribution without the activation of the XFEM, because the tension-stiening
eect was not successfully implemented in the CDP material model. The compressive
stress distribution, with and without the XFEM, are similar to the expected, analytical
distribution.
The maximum crack width that will not endanger the corrosion of the steel reinforcement
depends on the environment in which the structure is placed. The EN 1992-1-1 [2004] puts
a limit to the maximum allowable crack width of 0.3 − 0.4 mm depending on the exposure
class, e.g. the environment, and whether the reinforcing bars are bonded or not. The
limit to the maximum allowable crack width put by the EN 1992-1-1 [2004] is regardless
of the nature of the cracking. For exposure classes X0 and XC1, that is, for concrete inside
buildings with very low humidity or permanently submerged in water, respectively, the
maximum allowable crack width is set to guarantee acceptable appearance of the concrete,
as corrosion is not imminent. For the other exposure classes the limit to the maximum
allowable crack width is put to avoid corrosion of the reinforcement.
In this report, the amount of reinforcement in the beam falls within the requirements to
the minimum and maximum reinforcement areas put by the EN 1992-1-1 [2004]. A limit
to the minimum reinforcement area has been put by the EN 1992-1-1 [2004], because a
minimum amount of bonded reinforcement is required to control cracking. This section will
investigate the possibilities within the XFEM in Abaqus 6.10 to successfully model crack
widths and spacing in the serviceability limit state, that is, for a vertical displacement of
20 mm measured at the midspan of the beam. The equation for the crack spacing, sr,max ,
is given in 7.5.
84
7.3. Results and discussion Master Thesis
3 , 2 ,
The eective tension area is given by Ac,e = b·hc,e , where hc,e = min 2.5 · (h − d), h−x h
Equation 7.5 assumes that the bonded reinforcement is xed at closed centres within the
tension zone, which is fullled by satisfying the inequality: spacing ≤ 5(c + φ/2). The
equation for the crack width, wk , is given in 7.6.
σs − kt ρct,e
f
p,e
(1 + αe ρp,e ) σs
wk = sr,max ≥ 0.6 (7.6)
Es Es
The values of the coecients are chosen to t to the current model and all inequalities
above are fullled. The calculations are shown in appendix B.4. The analytical crack width
and spacing for a load corresponding to a displacement of 20 mm is calculated in appendix
B.4 and is based on a cracked cross-sectional analysis, see gure 7.7. The numerically
estimated crack width and spacing is a part of the output from the model in Abaqus
6.10. Figure 7.18 shows an example crack for a mesh consisting of 348 elements. The
crack width is measured as the dierence in the position of the original and the deformed
coordinates, marked with red dots in gure 7.18. The crack spacing is determined in the
85
Group B122b - Spring 2011 7. Results and discussion of 3d beam analysis
same way. Since numerous cracks appear, the largest crack is chosen to represent the
numerically estimated crack width. This is done for all mesh densities.
Figure 7.18. The crack width and spacing is measured as the dierence in the position of the
original and the deformed coordinates, marked with red dots
Table 7.5 shows the relation between the stress in the tension reinforcement for various
mesh densities, the analytically and the numerically determined crack width and spacing.
In average the analytical determined crack width is 58% smaller than the numerically
determined crack width. The dierence between the analytical and the numerically
obtained crack spacing is arbitrary, because the mesh density greatly inuences the
spacing. However, a tendency towards a spacing of approximately 200 mm exist. The
mesh consisting of 348 consists of a single crack, i.e. no crack spacing can be calculated.
Either the semi-empirical equation proposed by EN 1992-1-1 [2004] is inaccurate, or the
assumption behind innite bond, due to the embedment of the reinforcement in the Abaqus
6.10 model, violates the results to an unsatisfactory degree. No conclusion can be made
with regard to the accuracy of the equations proposed by EN 1992-1-1 [2004], as no
experimental work regarding crack width measurement or crack spacing on the exact
beam in this report is available. With regard to the numerical results it is known, that
crack width is greatly inuenced by the bond force acting on the interface between the
reinforcing steel and the concrete Piyasena [2002]. Since the embedding of the reinforcing
steel in the concrete implies that no slip can occur, the deformation of the concrete must
conform to that of the reinforcing bars, which further widens the cracks. In other words,
cracking results in relaxation in the concrete tensile stresses upon cracking due to slip,
86
7.3. Results and discussion Master Thesis
and this phenomenon cannot be modeled when the reinforcement is embedded. However,
the employed method of determining the crack widths does not consider the contribution
from elastic or plastic deformation of the cracked elements, which must be nonzero.
It is observed that the crack spacing depends on the mesh density to some extent. The
crack height is similar, except for the mesh consisting of 348 elements. The reason must
be found in the previously described cohesive segments method used in Abaqus 6.10,
which assumes that a crack ends at an element edge, thus propagating through the entire
87
Group B122b - Spring 2011 7. Results and discussion of 3d beam analysis
element. The crack in the crudest mesh does not propagate further than the rst element,
because equilibrium would not be obtainable.
Generally, the appearance of the exural cracks is as expected, with the maximum crack
width at the tension face and zero width near the neutral axis. No shear cracks were
observed. It is dicult to conclude on the accuracy of the crack pattern, because no
experimental results are at hand for an identical case. However, experimental results for a
beam loaded in four-point bending is performed by R.I. Gilbert and S. Nejadi [2011] and
shown in gure 7.20.
Figure 7.20. Crack pattern for a beam loaded in four-point bending. R.I. Gilbert and S. Nejadi
[2011]
The observed crack pattern is similar in two ways to the numerically obtained results
in this report. Firstly, the beam does not crack directly at the midspan of the beam.
Secondly, the observed crack heights are approximately 60% of the beam height, which
corresponds to the observed height of the cracks in the Abaqus 6.10 model. Note, that
the dimensions, loading and reinforcement arrangement dier from the beam analyzed in
this report, and for this reason a comparison should be taken lightly. Only two cracks
are observed, which is less than the expected number. The number of cracks showed to
be strongly dependent on the rate of loading. A low load rate was chosen for symmetry
in cracking and equilibrium reasons. A high load rate generally caused errors, such as
convergence problems and level-set function errors. Moreover, Abaqus 6.10 requires that
cracks initiate in the center of the elements, which is also observed in gure 7.19. Finally,
the adopted equations in EN 1992-1-1 [2004] does not take loading and geometry into
account, naturally putting a limit to the quality of the comparison between the semi-
empirical results from EN 1992-1-1 [2004] and the numerically obtained results regarding
crack width and spacing.
88
Conclusion
8
The problem statement of the project reads
Problem
In this report a three-dimensional reinforced concrete beam has been analyzed with respect
to load-deection behavior, cross sectional stress distribution, crack width, crack spacing
and crack pattern. The dimensions of the beam and the reinforcement conguration was
presented in gure 7.1.
Analysis
In order to answer the problem statement a non-linear fracture mechanics approach was
used. Abaqus 6.10 oers the cohesive segments method, which is based on the ctitious
crack model proposed by A. Hillerborg and Peterson [1976]. This model was adopted to
model the cracking behavior in the tensile region of the beam given a fracture energy of
Gf = 0.08 N/mm. The maximum principal stress criterion, MaxPS, was adopted as the
crack initiation criterion with a value of 1.95 M P a, equal to the tensile strength of the
examined concrete. The Concrete Damaged Plasticity material model, abbreviated CDP,
was adopted to model the inelastic stress-strain relation in the compressive region. The
compressive behavior was successfully modeled in chapter 6. The CDP material model
was chosen due to its manner of degrading stiness discretely in the Gauss points.
The quality of the implementation of the XFEM was tested for three benchmarks in chap-
ter 6. The rst benchmark considers crack hole interaction in a steel specimen. The
second benchmark considers crack initiation and propagation in a concrete beam, with-
out reinforcement, loaded in three-point bending. The nal benchmark considered crack
formation in a reinforced concrete specimen. The benchmark results were considered sat-
isfactory, although mesh-sensitivity was observed.
89
Group B122b - Spring 2011 8. Conclusion
Results
The purpose of the analysis of the three-dimensional beam was twofold. Firstly, to verify
the modeling technique adopted in this project for a three-dimensional reinforced concrete
beam, and secondly, to analyze the crack widths, spacing and pattern and compare the
results to the results obtained by use of current empirical formulas given in EN 1992-1-1
[2004].
Load-deection curves were obtained in Abaqus 6.10 for numerical models with and
without the XFEM. These were plotted versus analytical load-deection relations for
uncracked, partly cracked and fully cracked sections. The analytical equation assuming
a partly cracked section is provided by EN 1992-1-1 [2004] and was the main basis of
comparison. The load-deection curves were similar with and without the XFEM. Within
the serviceability limit state the numerical results showed a stier response than the
load-deection relation given by EN 1992-1-1 [2004]. In the ultimate limit state the
load-deection relation given by EN 1992-1-1 [2004] showed stier response than the
numerical results. This was concluded to be due to the elastic stress distribution assumed
by the expression given in EN 1992-1-1 [2004]. Moreover, the analytical expressions do
not account for the deection contribution due to shearing strain. However, this was
determined to constitute only 1.5 % of the total deection. It was concluded that the
inclusion of the XFEM does not improve the estimation of the load-deection considerably.
Furthermore the modeling technique adopted in this report proved to have no negative
impact on the load-deection curve.
The cross sectional stress distribution in a cracked cross section of the beam was
analyzed. The distributions from the numerical models, with and without the XFEM,
were compared. The purpose of the analysis was to verify the validity of the CDP material
model, with and without the XFEM. Results showed similarity in the compressive stress
distribution in every load stage when compared to the expected, analytical distribution.
It was not possible to model the tension stiening behavior in the concrete in the CDP
material model, and for this reason the tensile stresses did not degrade to zero. However,
the stresses degraded to zero when the XFEM was used. Convergence tests showed,
that the numerically estimated ultimate load converged to the analytically determined
ultimate load for approximately 2800 elements. The numerically determined deection at
the ultimate load was approximately 10% higher, with and without the XFEM, which was
explained by the fact, that the analytical equation proposed in EN 1992-1-1 [2004] assumes
a linear stress distribution, whereas the numerically obtained stress distribution was
plastic. In the serviceability limit state analysis, the numerically determined deection was
approximately 7% smaller, with and without the XFEM, than the analytically determined
deection. The numerically determined deection in the serviceability limit state showed a
better approximation to the analytically determined deection, because the cross sectional
stress distributions were similar. This behavior also corresponded to the observed load-
deection behavior.
Cracking in the beam was analyzed with respect to crack width, crack spacing and crack
pattern. The analysis was made using the XFEM in order to visualize the cracks, which
is not possible without the XFEM. Comparison was made to the EN 1992-1-1 [2004]
with respect to crack width and crack spacing. The numerical model cracked in two
90
Master Thesis
locations, which was below the expected number of cracks. This was explained to be
strongly dependent on the rate of loading. The crack width was overestimated in the
numerical model in comparison to the suggested equation in EN 1992-1-1 [2004]. This
was explained by the assumption in the model, that the reinforcing steel was embedded in
the concrete beam, thus allowing no bond slip. The crack spacing was found to be mesh
dependent in the sense, that cracks must initiate in the center of an element. This is a
limitation to the implementation of the XFEM in Abaqus 6.10. For three out of ve mesh
densities the results on the crack spacing showed agreement with the analytically estimated
spacing in EN 1992-1-1 [2004]. The observed crack patterns for various mesh densities
showed similarities with an experimental study on a concrete beam subjected to four-point
bending in two ways. Firstly, the crack heights were similar and equal to approximately
60% of the beam height and secondly, no cracks were observed in the midspan of the beam.
It was noted, that geometry, loading and reinforcement arrangement diered between the
experimental study and the study in this report. For that reason, the comparison was
taken lightly.
This report has shown that the XFEM can be successfully combined with the CDP
material model in Abaqus 6.10 for accurate load-deection estimations and cross-sectional
stress-distributions. Regarding cracking, the current implementation of the XFEM in
Abaqus 6.10 contains few limitations, which inuences the quality of the estimation of
crack width, crack spacing and crack pattern. It was found, that the XFEM is not
necessary in order to accurately estimate the load-deection behavior of a reinforced
concrete beam loaded to failure. The cross-sectional compressive stress distribution can be
successfully modeled with and without the XFEM. However, the tensile stress distribution
requires the implementation of the tension-stiening eect in the CDP material model in
order to yield accurate results. With respect to the estimation of crack widths, crack
spacing and crack patterns the implementation of the XFEM in Abaqus 6.10 needs
improvements, generally by improving the cohesive segments method, so that cracks does
not have to end on elements boundaries, by allowing cracks to initiate in arbitrary locations
and not only on the center of element boundaries and lastly, allowing the coalescence of
cracks.
91
Suggestions for future
work
9
The implementation of the XFEM in Abaqus 6.10 has presented itself as an area of
future research because of the aforementioned limitations. The cohesive segments method
contains the limitation that a crack must end on an element boundary. This limitation
could be alleviated, if crack-tip related quantities, such as the stress intensity factor, were
used to calculate the size of the crack propagation increment and the direction of the
crack. If the concrete is modeled as non-linear, this cannot be done using the J-integral,
which assumes linear-elastic behavior of the material. However, since the aim of this
report was to mainly consider cracking the serviceability limit state, a linear material
model could be adopted and the J-integral could be used. This is because the concrete
behaves approximately linear in the serviceability limit state. This has not been done in
this report, because the J-integral is not implemented for evolving cracks in Abaqus 6.10.
Finally, crack-tip enrichment can be incorporated in the XFEM to allow cracks to end
inside elements and to improve the estimation of crack-tip related quantities, such as the
stress intensity factor.
The bond properties used in this report can be improved. A physically realistic description
of the stress transfer at the interface between the steel and the concrete is necessary, as
well as modeling the bond-slip that occurs at the interface. Shear reinforcement should
also be included for a more realistic estimation of cracking. Moreover, a sensitivity study
on the parameters used in this report should be performed.
This report has lacked experimental data for a comparison of the list of studies in this
project. Experimental studies on the inuence of geometry, reinforcement arrangement
and loading conditions on would assist in the evaluation of the quality of the obtained
results and lead to potential improvements to the used equations from EN 1992-1-1 [2004]
in this study.
93
Appendix
1
Concrete Damaged
Plasticity material model
A
This appendix presents the Concrete Damaged Plasticity material model,
abbreviated CDP, used in this report. Unless stated otherwise the sources used
in this chapter are Systèmes [2010], Jankowiak and Lodygowski [2005].
The stress-strain relations are governed by scalar damaged elasticity, see equation A.2.
3
Group B122b - Spring 2011 A. Concrete Damaged Plasticity material model
The scalar degradation variable, d, can attain values between 0 and 1. In a material
point with d = 0 the concrete is undamaged. d = 1 corresponds to a fully damaged,
while intermediate values represent the irrecoverable, plastic deformation endured by the
concrete for pre-peak loading conditions. Failure of concrete, that is, crushing or cracking,
is therefore associated with a degradation of the elastic stiness. Within the context of
scalar-damage theory the stiness degradation is isotropic and a single parameter, d, is
used for the description. The eective stress at a material point is dened in equation
A.3.
The Cauchy stress is related to the eective stress through the scalar degradation variable,
see equation A.4
σ = (1 − d)σ̄ (A.4)
The factor (1−d) represents the eective area of the load-carrying part of the cross-section.
In the absence of damage the eective stress in equation A.3 is equal to the Cauchy stress.
However, for d 6= 0 the eective stress is more representative than the Cauchy stress, as
the eective, that is, the uncracked or uncrushed area, is carrying the external forces.
For this reason the eective stress is used for plasticity analysis. The evolution of the
degradation variable is controlled by the eective stress and a hardening parameter, e pl .
The eigenvalues of the eective stress tensor, that is, the principal eective stresses,
are used in the evolution equations for the hardening variables. Note, that two single
hardening variables are used; one for compression and one for tension, see equations (A.5-
A.6).
˜pl ˆ ˆpl
c = −(1 − r(σ̄ ) · min (A.5)
˜pl
t
ˆ ) · ˆpl
= r(σ̄ max (A.6)
ˆpl
max Maximum eigenvalue of the plastic strain tensor pl .
ˆpl
min Minimum eigenvalue of the plastic strain tensor pl .
ˆi
σ̄ Eigenvalues, or principal stresses, of the eective stress tensor σ̂ .
r(σ̄ˆ) Stress weight factor, see equation A.7.
4
A.2. Yield function Master Thesis
Σ3i=1 σ̄
ˆ
ˆ) = 3 i
r(σ̄ (A.7)
ˆi
Σi=1 σ̄
where h.i is the Macauley bracket and |.| refers to the absolute value. The stress weight
factor attains values between 0 and 1 and is used to assign a weight to the principal strains
in relation to their corresponding principal stress. The plastic strain tensor contains strains
in directions parallel to the directions of the global Cartesian coordinate system. The
values are determined using the specied uniaxial behavior of the concrete.
1
pl ) = pl ) σ̄ pl (A.8)
5
Group B122b - Spring 2011 A. Concrete Damaged Plasticity material model
Figure A.1. Yield surface in plane stress used in the Concrete Damaged Plasticity in Abaqus
6.10. σc0 and σb0 refer to the unixial and equibiaxial compressive yield stress,
respectively.
The denition of the Mises equivalent eective stress, q , and the eective hydrostatic
pressure, p, follows from equations A.9 and A.10, respectively.
r
3
q= S:S (A.9)
2
where S is the deviatoric part of the stress tensor given by S = σ̄ + p · I . I is the identity
matrix.
1
p = − · trace(σ̄) (A.10)
3
Microcracking and crushing are represented by increasing the value of the hardening
parameter epl , which also inuences the value of the degradation parameter, d, as
previously mentioned.
6
A.2. Yield function Master Thesis
Figure A.2. Reponse of concrete due to uniaxial tension. Jankowiak and Lodygowski [2005]
Figure A.3. Reponse of concrete due to uniaxial compression. Jankowiak and Lodygowski [2005]
7
Group B122b - Spring 2011 A. Concrete Damaged Plasticity material model
The stress-strain relations for uniaxial tension and compression are given by equations
A.11 and A.12, respectively.
σt
σ̄t = = E0 ( − ˜pl
t ) (A.13)
1 − dt
σc
σ̄c = = E0 ( − ˜pl
c ) (A.14)
1 − dc
Note, that eective uniaxial cohesion stresses act on the available load-carrying area.
For this reason the nucleation and propagation of cracks increases eective stresses by
increasing the scalar damage parameter.
∂G(σ̄)
˙pl = λ̇ (A.15)
∂ σ̄
where the ow potential, G, is chosen as the Drucker-Prager hyberbolic function, see
equation A.16.
(A.16)
p
G= (σt0 tan(ψ))2 + q̄ 2 − p̄ tan ψ
8
Guide to Appendix CD
B
B.1 DVD 1
Appendix DVD number one containing all .pdf les and output databases from Abaqus
6.10 with the XFEM activated.
9
Group B122b - Spring 2011 B. Guide to Appendix CD
B.7 Seed45XFEM.odb
Output database for the three-dimensional reinforced concrete beam in Abaqus 6.10. The
mesh consists of 6708 elements and the XFEM is activated.
B.8 Seed50XFEM.odb
Output database for the three-dimensional reinforced concrete beam in Abaqus 6.10. The
mesh consists of 4600 elements and the XFEM is activated.
B.9 Seed60XFEM.odb
Output database for the three-dimensional reinforced concrete beam in Abaqus 6.10. The
mesh consists of 2796 elements and the XFEM is activated.
B.10 Seed70XFEM.odb
Output database for the three-dimensional reinforced concrete beam in Abaqus 6.10. The
mesh consists of 2060 elements and the XFEM is activated.
B.11 Seed80XFEM.odb
Output database for the three-dimensional reinforced concrete beam in Abaqus 6.10. The
mesh consists of 1227 elements and the XFEM is activated.
B.12 Seed150XFEM.odb
Output database for the three-dimensional reinforced concrete beam in Abaqus 6.10. The
mesh consists of 348 elements and the XFEM is activated.
10
B.13. DVD 2 Master Thesis
B.13 DVD 2
Appendix DVD number two containing all output databases from Abaqus 6.10 without
XFEM activated.
B.15 Seed45CDPFull.odb
Output database for the three-dimensional reinforced concrete beam in Abaqus 6.10. The
mesh consists of 6708 elements and the XFEM is not activated. The model uses full
integration.
B.16 Seed50CDPFull.odb
Output database for the three-dimensional reinforced concrete beam in Abaqus 6.10. The
mesh consists of 4600 elements and the XFEM is not activated. The model uses full
integration.
B.17 Seed60CDPFull.odb
Output database for the three-dimensional reinforced concrete beam in Abaqus 6.10. The
mesh consists of 2796 elements and the XFEM is not activated. The model uses full
integration.
B.18 Seed70CDPFull.odb
Output database for the three-dimensional reinforced concrete beam in Abaqus 6.10. The
mesh consists of 2060 elements and the XFEM is not activated. The model uses full
integration.
B.19 Seed80CDPFull.odb
Output database for the three-dimensional reinforced concrete beam in Abaqus 6.10. The
mesh consists of 1227 elements and the XFEM is not activated. The model uses full
integration.
11
Group B122b - Spring 2011 B. Guide to Appendix CD
B.20 Seed150CDPFull.odb
Output database for the three-dimensional reinforced concrete beam in Abaqus 6.10. The
mesh consists of 348 elements and the XFEM is not activated. The model uses full
integration.
B.21 Seed50CDPRed.odb
Output database for the three-dimensional reinforced concrete beam in Abaqus 6.10. The
mesh consists of 4600 elements and the XFEM is not activated. The model uses reduced
integration.
B.22 Seed60CDPRed.odb
Output database for the three-dimensional reinforced concrete beam in Abaqus 6.10. The
mesh consists of 2796 elements and the XFEM is not activated. The model uses reduced
integration.
B.23 Seed70CDPRed.odb
Output database for the three-dimensional reinforced concrete beam in Abaqus 6.10. The
mesh consists of 2060 elements and the XFEM is not activated. The model uses reduced
integration.
B.24 Seed80CDPRed.odb
Output database for the three-dimensional reinforced concrete beam in Abaqus 6.10. The
mesh consists of 1227 elements and the XFEM is not activated. The model uses reduced
integration.
B.25 Seed150CDPRed.odb
Output database for the three-dimensional reinforced concrete beam in Abaqus 6.10. The
mesh consists of 348 elements and the XFEM is not activated. The model uses reduced
integration.
12
Bibliography
A.H. Nilson, D. Darwin and Charles W. Dolan, 2004. A.H. Nilson, D. Darwin
and Charles W. Dolan. Design of Concrete Structures, Thirteenth Edition. The
McGraw-Hill Companies, 2004.
Arya and Ofori-Darko, 1996. C. Arya and F.K. Ofori-Darko. Inuence of crack
frequency on reinforcement corrosion in concrete. Cement and Concrete Research, 26,
No. 3, 345353, 1996.
Belytschko and Black, 1999. T. Belytschko and T. Black. Elastic crack growth in
nite elements with minimal remeshing. International Journal for Numerical Methods
in Engineering, 45, 601620, 1999.
D. Ngo and A.C. Scordelis, 1967. D. Ngo and A.C. Scordelis. Finite element
analysis of reinforced concrete beams. Journal of the American Concrete Institute, 64,
152163, 1967.
Dassault Systémes Simulia Corp., 2010. Dassault Systémes Simulia Corp. Abaqus
6.10, 2010.
13
Group B122b - Spring 2011 BIBLIOGRAPHY
Fries and Belytschko., 2000. Thomas-Peter Fries and Ted Belytschko. The
extended/generalized nite element method: An overview of the method and its
applications. Int. J. Numer. Meth. Engng., 2000, 16, 2000.
G. Ventura and Belytschko, 2003. E. Budyn G. Ventura and T. Belytschko. Vector
level sets for description of propagating cracks in nite elements. Int. Journ. for
Numerical Methods in Engineering, 58, 15711592, 2003.
Grith, 1921. A. Grith. The Phenomenon of Rupture and Flow in Solids. 221,
163198, 1921.
Irwin, 1958. G.R. Irwin. Fracture. Handbuch der Physik, VI, 551590, 1958.
Jankowiak and Lodygowski, 2005. T. Jankowiak and T. Lodygowski. Identication
of parameters of concrete damage plasticity constitutive model. 2005.
J.Davies, 1996. J.Davies. Observation of the Fracture Path Development in Mortar
Beam Specimens. 1996.
J.L. Asferg, 2006. J.L. Asferg. Modeling of Concrete Fracture Applying the eXtended
Finite Element Method. Deparment of Civil Engineering Technical University of
Denmark, 2006.
Melenk and Babuska, 1996. J.M. Melenk and I. Babuska. The partition of unity
nite element method: Basic theory and applications. Computer Methods in Applied
Mechanics and Engineering, 39, 289314, 1996.
14
BIBLIOGRAPHY Master Thesis
N. Khomwan, S.J. Foster and S.T. Smith, 2010. N. Khomwan, S.J. Foster and
S.T. Smith.FE modeling of FPR-repaired planar concrete elements subjected to
monotonic and cyclic loading. Journal of Composites for Concstruction, 14, 720729,
2010.
Piyasena, 2002. R. Piyasena. Crack spacing, crack width and tension stiening eect
in reinforced concrete beams and one-way slabs. Faculty of Engineering and
Information Technology, Grith University, 2002.
Rashid, 1968. Y.R. Rashid. Analysis of reinforced concrete pressure vessels. Journal of
Nuclear Engineering and Design, 7, 334344, 1968.
Rots and Blaauwendraad, 1989. J.G. Rots and J. Blaauwendraad. Crack models for
concrete: discrete or smeared? Fixed Multi-directional or rotating? TNO-Institute for
Building Materials and Stuctures, 34, 159, 1989.
S.B Bhide and M.P Collins, 1989. S.B Bhide and M.P Collins. Inuence of axial
tension on the shear capacity of reinforced concrete members. 1989.
Schieÿl and Raupach, 1997. P. Schieÿl and M. Raupach. Laboratoty studies and
calculations on the inuence of crack width of chloride-induced corrosion of steel in
concrete. ACI Materials Journal, 94, 5662, 1997.
Song, Areias, and Belytschko, 2006. J.H. Song, P.M.A. Areias, and T. Belytschko.
A method for dynamic crack and shear band propagation with phantom nodes. Int. J.
Numer. Meth. Engng., 67, 868893, 2006.
15
Group B122b - Spring 2011 BIBLIOGRAPHY
Z.P. Bazant and B. Oh, 1983. Z.P. Bazant and B. Oh. Crack band theory for fracture
of concrete. Journal of Materials and Structures, 16, 155177, 1983.
16